blob: 6a1280a013ea1f635fb10e1b90d189895453bbca [file] [log] [blame]
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001/* -*- 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>
Sunil Mushran87d3d3f2008-12-17 14:17:42 -080036#include <linux/debugfs.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090037#include <linux/slab.h>
Mark Fasheha7f6a5f2005-12-15 14:31:23 -080038
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 */
Ingo Molnar34af9462006-06-27 02:53:55 -070059static DEFINE_SPINLOCK(o2hb_live_lock);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -080060static 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
Sunil Mushran536f0742010-10-07 17:03:07 -070065/*
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.
Sunil Mushrane7d656b2010-10-06 17:55:18 -070068 * - o2hb_live_region_bitmap tracks live regions (seen steady iterations).
Sunil Mushran43182d22010-10-06 17:55:16 -070069 * - o2hb_quorum_region_bitmap tracks live regions that have seen all nodes
70 * heartbeat on it.
Sunil Mushranb1c5ebf2010-10-07 17:05:52 -070071 * - o2hb_failed_region_bitmap tracks the regions that have seen io timeouts.
Sunil Mushran536f0742010-10-07 17:03:07 -070072 */
73static unsigned long o2hb_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
Sunil Mushrane7d656b2010-10-06 17:55:18 -070074static unsigned long o2hb_live_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
Sunil Mushran43182d22010-10-06 17:55:16 -070075static unsigned long o2hb_quorum_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
Sunil Mushranb1c5ebf2010-10-07 17:05:52 -070076static unsigned long o2hb_failed_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
Sunil Mushran536f0742010-10-07 17:03:07 -070077
Sunil Mushran8ca8b0bb2010-10-07 17:01:27 -070078#define O2HB_DB_TYPE_LIVENODES 0
Sunil Mushrana6de0132010-10-06 17:55:13 -070079#define O2HB_DB_TYPE_LIVEREGIONS 1
80#define O2HB_DB_TYPE_QUORUMREGIONS 2
81#define O2HB_DB_TYPE_FAILEDREGIONS 3
Sunil Mushran1f285302010-10-06 17:55:12 -070082#define O2HB_DB_TYPE_REGION_LIVENODES 4
83#define O2HB_DB_TYPE_REGION_NUMBER 5
Sunil Mushran43695d02010-10-06 17:55:09 -070084#define O2HB_DB_TYPE_REGION_ELAPSED_TIME 6
Sunil Mushran8ca8b0bb2010-10-07 17:01:27 -070085struct o2hb_debug_buf {
86 int db_type;
87 int db_size;
88 int db_len;
89 void *db_data;
90};
91
92static struct o2hb_debug_buf *o2hb_db_livenodes;
Sunil Mushrana6de0132010-10-06 17:55:13 -070093static struct o2hb_debug_buf *o2hb_db_liveregions;
94static struct o2hb_debug_buf *o2hb_db_quorumregions;
95static struct o2hb_debug_buf *o2hb_db_failedregions;
Sunil Mushran8ca8b0bb2010-10-07 17:01:27 -070096
Sunil Mushran87d3d3f2008-12-17 14:17:42 -080097#define O2HB_DEBUG_DIR "o2hb"
98#define O2HB_DEBUG_LIVENODES "livenodes"
Sunil Mushrana6de0132010-10-06 17:55:13 -070099#define O2HB_DEBUG_LIVEREGIONS "live_regions"
100#define O2HB_DEBUG_QUORUMREGIONS "quorum_regions"
101#define O2HB_DEBUG_FAILEDREGIONS "failed_regions"
Sunil Mushran1f285302010-10-06 17:55:12 -0700102#define O2HB_DEBUG_REGION_NUMBER "num"
Sunil Mushran43695d02010-10-06 17:55:09 -0700103#define O2HB_DEBUG_REGION_ELAPSED_TIME "elapsed_time_in_ms"
Sunil Mushran8ca8b0bb2010-10-07 17:01:27 -0700104
Sunil Mushran87d3d3f2008-12-17 14:17:42 -0800105static struct dentry *o2hb_debug_dir;
106static struct dentry *o2hb_debug_livenodes;
Sunil Mushrana6de0132010-10-06 17:55:13 -0700107static struct dentry *o2hb_debug_liveregions;
108static struct dentry *o2hb_debug_quorumregions;
109static struct dentry *o2hb_debug_failedregions;
Sunil Mushran87d3d3f2008-12-17 14:17:42 -0800110
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800111static LIST_HEAD(o2hb_all_regions);
112
113static struct o2hb_callback {
114 struct list_head list;
115} o2hb_callbacks[O2HB_NUM_CB];
116
117static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type);
118
119#define O2HB_DEFAULT_BLOCK_BITS 9
120
Sunil Mushran54b51872010-10-07 15:26:08 -0700121enum o2hb_heartbeat_modes {
122 O2HB_HEARTBEAT_LOCAL = 0,
123 O2HB_HEARTBEAT_GLOBAL,
124 O2HB_HEARTBEAT_NUM_MODES,
125};
126
127char *o2hb_heartbeat_mode_desc[O2HB_HEARTBEAT_NUM_MODES] = {
128 "local", /* O2HB_HEARTBEAT_LOCAL */
129 "global", /* O2HB_HEARTBEAT_GLOBAL */
130};
131
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800132unsigned int o2hb_dead_threshold = O2HB_DEFAULT_DEAD_THRESHOLD;
Sunil Mushran54b51872010-10-07 15:26:08 -0700133unsigned int o2hb_heartbeat_mode = O2HB_HEARTBEAT_LOCAL;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800134
Sunil Mushran2bd63212010-01-25 16:57:38 -0800135/* Only sets a new threshold if there are no active regions.
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800136 *
137 * No locking or otherwise interesting code is required for reading
138 * o2hb_dead_threshold as it can't change once regions are active and
139 * it's not interesting to anyone until then anyway. */
140static void o2hb_dead_threshold_set(unsigned int threshold)
141{
142 if (threshold > O2HB_MIN_DEAD_THRESHOLD) {
143 spin_lock(&o2hb_live_lock);
144 if (list_empty(&o2hb_all_regions))
145 o2hb_dead_threshold = threshold;
146 spin_unlock(&o2hb_live_lock);
147 }
148}
149
Sunil Mushran54b51872010-10-07 15:26:08 -0700150static int o2hb_global_hearbeat_mode_set(unsigned int hb_mode)
151{
152 int ret = -1;
153
154 if (hb_mode < O2HB_HEARTBEAT_NUM_MODES) {
155 spin_lock(&o2hb_live_lock);
156 if (list_empty(&o2hb_all_regions)) {
157 o2hb_heartbeat_mode = hb_mode;
158 ret = 0;
159 }
160 spin_unlock(&o2hb_live_lock);
161 }
162
163 return ret;
164}
165
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800166struct o2hb_node_event {
167 struct list_head hn_item;
168 enum o2hb_callback_type hn_event_type;
169 struct o2nm_node *hn_node;
170 int hn_node_num;
171};
172
173struct o2hb_disk_slot {
174 struct o2hb_disk_heartbeat_block *ds_raw_block;
175 u8 ds_node_num;
176 u64 ds_last_time;
177 u64 ds_last_generation;
178 u16 ds_equal_samples;
179 u16 ds_changed_samples;
180 struct list_head ds_live_item;
181};
182
183/* each thread owns a region.. when we're asked to tear down the region
184 * we ask the thread to stop, who cleans up the region */
185struct o2hb_region {
186 struct config_item hr_item;
187
188 struct list_head hr_all_item;
189 unsigned hr_unclean_stop:1;
190
191 /* protected by the hr_callback_sem */
192 struct task_struct *hr_task;
193
194 unsigned int hr_blocks;
195 unsigned long long hr_start_block;
196
197 unsigned int hr_block_bits;
198 unsigned int hr_block_bytes;
199
200 unsigned int hr_slots_per_page;
201 unsigned int hr_num_pages;
202
203 struct page **hr_slot_data;
204 struct block_device *hr_bdev;
205 struct o2hb_disk_slot *hr_slots;
206
Sunil Mushran823a6372010-10-06 17:55:21 -0700207 /* live node map of this region */
208 unsigned long hr_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
Sunil Mushran536f0742010-10-07 17:03:07 -0700209 unsigned int hr_region_num;
Sunil Mushran823a6372010-10-06 17:55:21 -0700210
Sunil Mushran1f285302010-10-06 17:55:12 -0700211 struct dentry *hr_debug_dir;
212 struct dentry *hr_debug_livenodes;
213 struct dentry *hr_debug_regnum;
Sunil Mushran43695d02010-10-06 17:55:09 -0700214 struct dentry *hr_debug_elapsed_time;
Sunil Mushran1f285302010-10-06 17:55:12 -0700215 struct o2hb_debug_buf *hr_db_livenodes;
216 struct o2hb_debug_buf *hr_db_regnum;
Sunil Mushran43695d02010-10-06 17:55:09 -0700217 struct o2hb_debug_buf *hr_db_elapsed_time;
Sunil Mushran1f285302010-10-06 17:55:12 -0700218
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800219 /* let the person setting up hb wait for it to return until it
220 * has reached a 'steady' state. This will be fixed when we have
221 * a more complete api that doesn't lead to this sort of fragility. */
222 atomic_t hr_steady_iterations;
223
224 char hr_dev_name[BDEVNAME_SIZE];
225
226 unsigned int hr_timeout_ms;
227
228 /* randomized as the region goes up and down so that a node
229 * recognizes a node going up and down in one iteration */
230 u64 hr_generation;
231
David Howellsc4028952006-11-22 14:57:56 +0000232 struct delayed_work hr_write_timeout_work;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800233 unsigned long hr_last_timeout_start;
234
235 /* Used during o2hb_check_slot to hold a copy of the block
236 * being checked because we temporarily have to zero out the
237 * crc field. */
238 struct o2hb_disk_heartbeat_block *hr_tmp_block;
239};
240
241struct o2hb_bio_wait_ctxt {
242 atomic_t wc_num_reqs;
243 struct completion wc_io_complete;
Mark Fasheha9e2ae32006-03-24 14:20:17 -0800244 int wc_error;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800245};
246
Sunil Mushranb1c5ebf2010-10-07 17:05:52 -0700247static int o2hb_pop_count(void *map, int count)
248{
249 int i = -1, pop = 0;
250
251 while ((i = find_next_bit(map, count, i + 1)) < count)
252 pop++;
253 return pop;
254}
255
David Howellsc4028952006-11-22 14:57:56 +0000256static void o2hb_write_timeout(struct work_struct *work)
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800257{
Sunil Mushranb1c5ebf2010-10-07 17:05:52 -0700258 int failed, quorum;
259 unsigned long flags;
David Howellsc4028952006-11-22 14:57:56 +0000260 struct o2hb_region *reg =
261 container_of(work, struct o2hb_region,
262 hr_write_timeout_work.work);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800263
264 mlog(ML_ERROR, "Heartbeat write timeout to device %s after %u "
265 "milliseconds\n", reg->hr_dev_name,
Sunil Mushran2bd63212010-01-25 16:57:38 -0800266 jiffies_to_msecs(jiffies - reg->hr_last_timeout_start));
Sunil Mushranb1c5ebf2010-10-07 17:05:52 -0700267
268 if (o2hb_global_heartbeat_active()) {
269 spin_lock_irqsave(&o2hb_live_lock, flags);
270 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
271 set_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
272 failed = o2hb_pop_count(&o2hb_failed_region_bitmap,
273 O2NM_MAX_REGIONS);
274 quorum = o2hb_pop_count(&o2hb_quorum_region_bitmap,
275 O2NM_MAX_REGIONS);
276 spin_unlock_irqrestore(&o2hb_live_lock, flags);
277
278 mlog(ML_HEARTBEAT, "Number of regions %d, failed regions %d\n",
279 quorum, failed);
280
281 /*
282 * Fence if the number of failed regions >= half the number
283 * of quorum regions
284 */
285 if ((failed << 1) < quorum)
286 return;
287 }
288
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800289 o2quo_disk_timeout();
290}
291
292static void o2hb_arm_write_timeout(struct o2hb_region *reg)
293{
Tao Mab31d3082009-12-22 10:32:15 +0800294 mlog(ML_HEARTBEAT, "Queue write timeout for %u ms\n",
295 O2HB_MAX_WRITE_TIMEOUT_MS);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800296
Sunil Mushranb1c5ebf2010-10-07 17:05:52 -0700297 if (o2hb_global_heartbeat_active()) {
298 spin_lock(&o2hb_live_lock);
299 clear_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
300 spin_unlock(&o2hb_live_lock);
301 }
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800302 cancel_delayed_work(&reg->hr_write_timeout_work);
303 reg->hr_last_timeout_start = jiffies;
304 schedule_delayed_work(&reg->hr_write_timeout_work,
305 msecs_to_jiffies(O2HB_MAX_WRITE_TIMEOUT_MS));
306}
307
308static void o2hb_disarm_write_timeout(struct o2hb_region *reg)
309{
310 cancel_delayed_work(&reg->hr_write_timeout_work);
311 flush_scheduled_work();
312}
313
Philipp Reisnerb5592922007-01-11 10:58:10 +0100314static inline void o2hb_bio_wait_init(struct o2hb_bio_wait_ctxt *wc)
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800315{
Philipp Reisnerb5592922007-01-11 10:58:10 +0100316 atomic_set(&wc->wc_num_reqs, 1);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800317 init_completion(&wc->wc_io_complete);
Mark Fasheha9e2ae32006-03-24 14:20:17 -0800318 wc->wc_error = 0;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800319}
320
321/* Used in error paths too */
322static inline void o2hb_bio_wait_dec(struct o2hb_bio_wait_ctxt *wc,
323 unsigned int num)
324{
325 /* sadly atomic_sub_and_test() isn't available on all platforms. The
326 * good news is that the fast path only completes one at a time */
327 while(num--) {
328 if (atomic_dec_and_test(&wc->wc_num_reqs)) {
329 BUG_ON(num > 0);
330 complete(&wc->wc_io_complete);
331 }
332 }
333}
334
335static void o2hb_wait_on_io(struct o2hb_region *reg,
336 struct o2hb_bio_wait_ctxt *wc)
337{
338 struct address_space *mapping = reg->hr_bdev->bd_inode->i_mapping;
339
340 blk_run_address_space(mapping);
Philipp Reisnerb5592922007-01-11 10:58:10 +0100341 o2hb_bio_wait_dec(wc, 1);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800342
343 wait_for_completion(&wc->wc_io_complete);
344}
345
Al Viro782e3b32007-10-12 07:17:47 +0100346static void o2hb_bio_end_io(struct bio *bio,
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800347 int error)
348{
349 struct o2hb_bio_wait_ctxt *wc = bio->bi_private;
350
Mark Fasheha9e2ae32006-03-24 14:20:17 -0800351 if (error) {
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800352 mlog(ML_ERROR, "IO Error %d\n", error);
Mark Fasheha9e2ae32006-03-24 14:20:17 -0800353 wc->wc_error = error;
354 }
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800355
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800356 o2hb_bio_wait_dec(wc, 1);
Philipp Reisnerb5592922007-01-11 10:58:10 +0100357 bio_put(bio);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800358}
359
360/* Setup a Bio to cover I/O against num_slots slots starting at
361 * start_slot. */
362static struct bio *o2hb_setup_one_bio(struct o2hb_region *reg,
363 struct o2hb_bio_wait_ctxt *wc,
Philipp Reisnerb5592922007-01-11 10:58:10 +0100364 unsigned int *current_slot,
365 unsigned int max_slots)
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800366{
Philipp Reisnerb5592922007-01-11 10:58:10 +0100367 int len, current_page;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800368 unsigned int vec_len, vec_start;
369 unsigned int bits = reg->hr_block_bits;
370 unsigned int spp = reg->hr_slots_per_page;
Philipp Reisnerb5592922007-01-11 10:58:10 +0100371 unsigned int cs = *current_slot;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800372 struct bio *bio;
373 struct page *page;
374
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800375 /* Testing has shown this allocation to take long enough under
376 * GFP_KERNEL that the local node can get fenced. It would be
377 * nicest if we could pre-allocate these bios and avoid this
378 * all together. */
Philipp Reisnerb5592922007-01-11 10:58:10 +0100379 bio = bio_alloc(GFP_ATOMIC, 16);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800380 if (!bio) {
381 mlog(ML_ERROR, "Could not alloc slots BIO!\n");
382 bio = ERR_PTR(-ENOMEM);
383 goto bail;
384 }
385
386 /* Must put everything in 512 byte sectors for the bio... */
Philipp Reisnerb5592922007-01-11 10:58:10 +0100387 bio->bi_sector = (reg->hr_start_block + cs) << (bits - 9);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800388 bio->bi_bdev = reg->hr_bdev;
389 bio->bi_private = wc;
390 bio->bi_end_io = o2hb_bio_end_io;
391
Philipp Reisnerb5592922007-01-11 10:58:10 +0100392 vec_start = (cs << bits) % PAGE_CACHE_SIZE;
393 while(cs < max_slots) {
394 current_page = cs / spp;
395 page = reg->hr_slot_data[current_page];
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800396
Jan Karabc7e97c2007-10-10 16:25:42 +0200397 vec_len = min(PAGE_CACHE_SIZE - vec_start,
Philipp Reisnerb5592922007-01-11 10:58:10 +0100398 (max_slots-cs) * (PAGE_CACHE_SIZE/spp) );
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800399
400 mlog(ML_HB_BIO, "page %d, vec_len = %u, vec_start = %u\n",
Philipp Reisnerb5592922007-01-11 10:58:10 +0100401 current_page, vec_len, vec_start);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800402
403 len = bio_add_page(bio, page, vec_len, vec_start);
Philipp Reisnerb5592922007-01-11 10:58:10 +0100404 if (len != vec_len) break;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800405
Philipp Reisnerb5592922007-01-11 10:58:10 +0100406 cs += vec_len / (PAGE_CACHE_SIZE/spp);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800407 vec_start = 0;
408 }
409
410bail:
Philipp Reisnerb5592922007-01-11 10:58:10 +0100411 *current_slot = cs;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800412 return bio;
413}
414
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800415static int o2hb_read_slots(struct o2hb_region *reg,
416 unsigned int max_slots)
417{
Philipp Reisnerb5592922007-01-11 10:58:10 +0100418 unsigned int current_slot=0;
419 int status;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800420 struct o2hb_bio_wait_ctxt wc;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800421 struct bio *bio;
422
Philipp Reisnerb5592922007-01-11 10:58:10 +0100423 o2hb_bio_wait_init(&wc);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800424
Philipp Reisnerb5592922007-01-11 10:58:10 +0100425 while(current_slot < max_slots) {
426 bio = o2hb_setup_one_bio(reg, &wc, &current_slot, max_slots);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800427 if (IS_ERR(bio)) {
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800428 status = PTR_ERR(bio);
429 mlog_errno(status);
430 goto bail_and_wait;
431 }
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800432
Philipp Reisnerb5592922007-01-11 10:58:10 +0100433 atomic_inc(&wc.wc_num_reqs);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800434 submit_bio(READ, bio);
435 }
436
437 status = 0;
438
439bail_and_wait:
440 o2hb_wait_on_io(reg, &wc);
Mark Fasheha9e2ae32006-03-24 14:20:17 -0800441 if (wc.wc_error && !status)
442 status = wc.wc_error;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800443
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800444 return status;
445}
446
447static int o2hb_issue_node_write(struct o2hb_region *reg,
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800448 struct o2hb_bio_wait_ctxt *write_wc)
449{
450 int status;
451 unsigned int slot;
452 struct bio *bio;
453
Philipp Reisnerb5592922007-01-11 10:58:10 +0100454 o2hb_bio_wait_init(write_wc);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800455
456 slot = o2nm_this_node();
457
Philipp Reisnerb5592922007-01-11 10:58:10 +0100458 bio = o2hb_setup_one_bio(reg, write_wc, &slot, slot+1);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800459 if (IS_ERR(bio)) {
460 status = PTR_ERR(bio);
461 mlog_errno(status);
462 goto bail;
463 }
464
Philipp Reisnerb5592922007-01-11 10:58:10 +0100465 atomic_inc(&write_wc->wc_num_reqs);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800466 submit_bio(WRITE, bio);
467
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800468 status = 0;
469bail:
470 return status;
471}
472
473static u32 o2hb_compute_block_crc_le(struct o2hb_region *reg,
474 struct o2hb_disk_heartbeat_block *hb_block)
475{
476 __le32 old_cksum;
477 u32 ret;
478
479 /* We want to compute the block crc with a 0 value in the
480 * hb_cksum field. Save it off here and replace after the
481 * crc. */
482 old_cksum = hb_block->hb_cksum;
483 hb_block->hb_cksum = 0;
484
485 ret = crc32_le(0, (unsigned char *) hb_block, reg->hr_block_bytes);
486
487 hb_block->hb_cksum = old_cksum;
488
489 return ret;
490}
491
492static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block *hb_block)
493{
Mark Fasheh70bacbd2006-03-02 11:10:05 -0800494 mlog(ML_ERROR, "Dump slot information: seq = 0x%llx, node = %u, "
495 "cksum = 0x%x, generation 0x%llx\n",
496 (long long)le64_to_cpu(hb_block->hb_seq),
497 hb_block->hb_node, le32_to_cpu(hb_block->hb_cksum),
498 (long long)le64_to_cpu(hb_block->hb_generation));
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800499}
500
501static int o2hb_verify_crc(struct o2hb_region *reg,
502 struct o2hb_disk_heartbeat_block *hb_block)
503{
504 u32 read, computed;
505
506 read = le32_to_cpu(hb_block->hb_cksum);
507 computed = o2hb_compute_block_crc_le(reg, hb_block);
508
509 return read == computed;
510}
511
512/* We want to make sure that nobody is heartbeating on top of us --
513 * this will help detect an invalid configuration. */
514static int o2hb_check_last_timestamp(struct o2hb_region *reg)
515{
516 int node_num, ret;
517 struct o2hb_disk_slot *slot;
518 struct o2hb_disk_heartbeat_block *hb_block;
519
520 node_num = o2nm_this_node();
521
522 ret = 1;
523 slot = &reg->hr_slots[node_num];
524 /* Don't check on our 1st timestamp */
525 if (slot->ds_last_time) {
526 hb_block = slot->ds_raw_block;
527
528 if (le64_to_cpu(hb_block->hb_seq) != slot->ds_last_time)
529 ret = 0;
530 }
531
532 return ret;
533}
534
535static inline void o2hb_prepare_block(struct o2hb_region *reg,
536 u64 generation)
537{
538 int node_num;
539 u64 cputime;
540 struct o2hb_disk_slot *slot;
541 struct o2hb_disk_heartbeat_block *hb_block;
542
543 node_num = o2nm_this_node();
544 slot = &reg->hr_slots[node_num];
545
546 hb_block = (struct o2hb_disk_heartbeat_block *)slot->ds_raw_block;
547 memset(hb_block, 0, reg->hr_block_bytes);
548 /* TODO: time stuff */
549 cputime = CURRENT_TIME.tv_sec;
550 if (!cputime)
551 cputime = 1;
552
553 hb_block->hb_seq = cpu_to_le64(cputime);
554 hb_block->hb_node = node_num;
555 hb_block->hb_generation = cpu_to_le64(generation);
Mark Fasheh0db638f2006-05-09 15:09:35 -0700556 hb_block->hb_dead_ms = cpu_to_le32(o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800557
558 /* This step must always happen last! */
559 hb_block->hb_cksum = cpu_to_le32(o2hb_compute_block_crc_le(reg,
560 hb_block));
561
Mark Fasheh70bacbd2006-03-02 11:10:05 -0800562 mlog(ML_HB_BIO, "our node generation = 0x%llx, cksum = 0x%x\n",
Mark Fasheh5fdf1e62007-04-27 16:50:03 -0700563 (long long)generation,
Mark Fasheh70bacbd2006-03-02 11:10:05 -0800564 le32_to_cpu(hb_block->hb_cksum));
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800565}
566
567static void o2hb_fire_callbacks(struct o2hb_callback *hbcall,
568 struct o2nm_node *node,
569 int idx)
570{
571 struct list_head *iter;
572 struct o2hb_callback_func *f;
573
574 list_for_each(iter, &hbcall->list) {
575 f = list_entry(iter, struct o2hb_callback_func, hc_item);
576 mlog(ML_HEARTBEAT, "calling funcs %p\n", f);
577 (f->hc_func)(node, idx, f->hc_data);
578 }
579}
580
581/* Will run the list in order until we process the passed event */
582static void o2hb_run_event_list(struct o2hb_node_event *queued_event)
583{
584 int empty;
585 struct o2hb_callback *hbcall;
586 struct o2hb_node_event *event;
587
588 spin_lock(&o2hb_live_lock);
589 empty = list_empty(&queued_event->hn_item);
590 spin_unlock(&o2hb_live_lock);
591 if (empty)
592 return;
593
594 /* Holding callback sem assures we don't alter the callback
595 * lists when doing this, and serializes ourselves with other
596 * processes wanting callbacks. */
597 down_write(&o2hb_callback_sem);
598
599 spin_lock(&o2hb_live_lock);
600 while (!list_empty(&o2hb_node_events)
601 && !list_empty(&queued_event->hn_item)) {
602 event = list_entry(o2hb_node_events.next,
603 struct o2hb_node_event,
604 hn_item);
605 list_del_init(&event->hn_item);
606 spin_unlock(&o2hb_live_lock);
607
608 mlog(ML_HEARTBEAT, "Node %s event for %d\n",
609 event->hn_event_type == O2HB_NODE_UP_CB ? "UP" : "DOWN",
610 event->hn_node_num);
611
612 hbcall = hbcall_from_type(event->hn_event_type);
613
614 /* We should *never* have gotten on to the list with a
615 * bad type... This isn't something that we should try
616 * to recover from. */
617 BUG_ON(IS_ERR(hbcall));
618
619 o2hb_fire_callbacks(hbcall, event->hn_node, event->hn_node_num);
620
621 spin_lock(&o2hb_live_lock);
622 }
623 spin_unlock(&o2hb_live_lock);
624
625 up_write(&o2hb_callback_sem);
626}
627
628static void o2hb_queue_node_event(struct o2hb_node_event *event,
629 enum o2hb_callback_type type,
630 struct o2nm_node *node,
631 int node_num)
632{
633 assert_spin_locked(&o2hb_live_lock);
634
Sunil Mushran0e105d32010-10-07 17:00:16 -0700635 BUG_ON((!node) && (type != O2HB_NODE_DOWN_CB));
636
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800637 event->hn_event_type = type;
638 event->hn_node = node;
639 event->hn_node_num = node_num;
640
641 mlog(ML_HEARTBEAT, "Queue node %s event for node %d\n",
642 type == O2HB_NODE_UP_CB ? "UP" : "DOWN", node_num);
643
644 list_add_tail(&event->hn_item, &o2hb_node_events);
645}
646
647static void o2hb_shutdown_slot(struct o2hb_disk_slot *slot)
648{
649 struct o2hb_node_event event =
650 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
651 struct o2nm_node *node;
652
653 node = o2nm_get_node_by_num(slot->ds_node_num);
654 if (!node)
655 return;
656
657 spin_lock(&o2hb_live_lock);
658 if (!list_empty(&slot->ds_live_item)) {
659 mlog(ML_HEARTBEAT, "Shutdown, node %d leaves region\n",
660 slot->ds_node_num);
661
662 list_del_init(&slot->ds_live_item);
663
664 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
665 clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
666
667 o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node,
668 slot->ds_node_num);
669 }
670 }
671 spin_unlock(&o2hb_live_lock);
672
673 o2hb_run_event_list(&event);
674
675 o2nm_node_put(node);
676}
677
Sunil Mushran43182d22010-10-06 17:55:16 -0700678static void o2hb_set_quorum_device(struct o2hb_region *reg,
679 struct o2hb_disk_slot *slot)
680{
681 assert_spin_locked(&o2hb_live_lock);
682
683 if (!o2hb_global_heartbeat_active())
684 return;
685
686 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
687 return;
688
689 /*
690 * A region can be added to the quorum only when it sees all
691 * live nodes heartbeat on it. In other words, the region has been
692 * added to all nodes.
693 */
694 if (memcmp(reg->hr_live_node_bitmap, o2hb_live_node_bitmap,
695 sizeof(o2hb_live_node_bitmap)))
696 return;
697
698 if (slot->ds_changed_samples < O2HB_LIVE_THRESHOLD)
699 return;
700
701 printk(KERN_NOTICE "o2hb: Region %s is now a quorum device\n",
702 config_item_name(&reg->hr_item));
703
704 set_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
705}
706
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800707static int o2hb_check_slot(struct o2hb_region *reg,
708 struct o2hb_disk_slot *slot)
709{
710 int changed = 0, gen_changed = 0;
711 struct o2hb_node_event event =
712 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
713 struct o2nm_node *node;
714 struct o2hb_disk_heartbeat_block *hb_block = reg->hr_tmp_block;
715 u64 cputime;
Mark Fasheh0db638f2006-05-09 15:09:35 -0700716 unsigned int dead_ms = o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS;
717 unsigned int slot_dead_ms;
Sunil Mushran0e105d32010-10-07 17:00:16 -0700718 int tmp;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800719
720 memcpy(hb_block, slot->ds_raw_block, reg->hr_block_bytes);
721
Sunil Mushran0e105d32010-10-07 17:00:16 -0700722 /*
723 * If a node is no longer configured but is still in the livemap, we
724 * may need to clear that bit from the livemap.
725 */
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800726 node = o2nm_get_node_by_num(slot->ds_node_num);
Sunil Mushran0e105d32010-10-07 17:00:16 -0700727 if (!node) {
728 spin_lock(&o2hb_live_lock);
729 tmp = test_bit(slot->ds_node_num, o2hb_live_node_bitmap);
730 spin_unlock(&o2hb_live_lock);
731 if (!tmp)
732 return 0;
733 }
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800734
735 if (!o2hb_verify_crc(reg, hb_block)) {
736 /* all paths from here will drop o2hb_live_lock for
737 * us. */
738 spin_lock(&o2hb_live_lock);
739
740 /* Don't print an error on the console in this case -
741 * a freshly formatted heartbeat area will not have a
742 * crc set on it. */
743 if (list_empty(&slot->ds_live_item))
744 goto out;
745
746 /* The node is live but pushed out a bad crc. We
747 * consider it a transient miss but don't populate any
748 * other values as they may be junk. */
749 mlog(ML_ERROR, "Node %d has written a bad crc to %s\n",
750 slot->ds_node_num, reg->hr_dev_name);
751 o2hb_dump_slot(hb_block);
752
753 slot->ds_equal_samples++;
754 goto fire_callbacks;
755 }
756
757 /* we don't care if these wrap.. the state transitions below
758 * clear at the right places */
759 cputime = le64_to_cpu(hb_block->hb_seq);
760 if (slot->ds_last_time != cputime)
761 slot->ds_changed_samples++;
762 else
763 slot->ds_equal_samples++;
764 slot->ds_last_time = cputime;
765
766 /* The node changed heartbeat generations. We assume this to
767 * mean it dropped off but came back before we timed out. We
768 * want to consider it down for the time being but don't want
769 * to lose any changed_samples state we might build up to
770 * considering it live again. */
771 if (slot->ds_last_generation != le64_to_cpu(hb_block->hb_generation)) {
772 gen_changed = 1;
773 slot->ds_equal_samples = 0;
Mark Fasheh70bacbd2006-03-02 11:10:05 -0800774 mlog(ML_HEARTBEAT, "Node %d changed generation (0x%llx "
775 "to 0x%llx)\n", slot->ds_node_num,
776 (long long)slot->ds_last_generation,
777 (long long)le64_to_cpu(hb_block->hb_generation));
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800778 }
779
780 slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
781
Mark Fasheh70bacbd2006-03-02 11:10:05 -0800782 mlog(ML_HEARTBEAT, "Slot %d gen 0x%llx cksum 0x%x "
783 "seq %llu last %llu changed %u equal %u\n",
784 slot->ds_node_num, (long long)slot->ds_last_generation,
785 le32_to_cpu(hb_block->hb_cksum),
Sunil Mushran2bd63212010-01-25 16:57:38 -0800786 (unsigned long long)le64_to_cpu(hb_block->hb_seq),
Mark Fasheh70bacbd2006-03-02 11:10:05 -0800787 (unsigned long long)slot->ds_last_time, slot->ds_changed_samples,
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800788 slot->ds_equal_samples);
789
790 spin_lock(&o2hb_live_lock);
791
792fire_callbacks:
793 /* dead nodes only come to life after some number of
794 * changes at any time during their dead time */
795 if (list_empty(&slot->ds_live_item) &&
796 slot->ds_changed_samples >= O2HB_LIVE_THRESHOLD) {
Mark Fasheh70bacbd2006-03-02 11:10:05 -0800797 mlog(ML_HEARTBEAT, "Node %d (id 0x%llx) joined my region\n",
798 slot->ds_node_num, (long long)slot->ds_last_generation);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800799
Sunil Mushran823a6372010-10-06 17:55:21 -0700800 set_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
801
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800802 /* first on the list generates a callback */
803 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
Sunil Mushrand6aa1c72010-10-06 18:50:50 -0700804 mlog(ML_HEARTBEAT, "o2hb: Add node %d to live nodes "
805 "bitmap\n", slot->ds_node_num);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800806 set_bit(slot->ds_node_num, o2hb_live_node_bitmap);
807
808 o2hb_queue_node_event(&event, O2HB_NODE_UP_CB, node,
809 slot->ds_node_num);
810
811 changed = 1;
812 }
813
814 list_add_tail(&slot->ds_live_item,
815 &o2hb_live_slots[slot->ds_node_num]);
816
817 slot->ds_equal_samples = 0;
Mark Fasheh0db638f2006-05-09 15:09:35 -0700818
819 /* We want to be sure that all nodes agree on the
820 * number of milliseconds before a node will be
821 * considered dead. The self-fencing timeout is
822 * computed from this value, and a discrepancy might
823 * result in heartbeat calling a node dead when it
824 * hasn't self-fenced yet. */
825 slot_dead_ms = le32_to_cpu(hb_block->hb_dead_ms);
826 if (slot_dead_ms && slot_dead_ms != dead_ms) {
827 /* TODO: Perhaps we can fail the region here. */
828 mlog(ML_ERROR, "Node %d on device %s has a dead count "
829 "of %u ms, but our count is %u ms.\n"
830 "Please double check your configuration values "
831 "for 'O2CB_HEARTBEAT_THRESHOLD'\n",
832 slot->ds_node_num, reg->hr_dev_name, slot_dead_ms,
833 dead_ms);
834 }
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800835 goto out;
836 }
837
838 /* if the list is dead, we're done.. */
839 if (list_empty(&slot->ds_live_item))
840 goto out;
841
842 /* live nodes only go dead after enough consequtive missed
843 * samples.. reset the missed counter whenever we see
844 * activity */
845 if (slot->ds_equal_samples >= o2hb_dead_threshold || gen_changed) {
846 mlog(ML_HEARTBEAT, "Node %d left my region\n",
847 slot->ds_node_num);
848
Sunil Mushran823a6372010-10-06 17:55:21 -0700849 clear_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
850
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800851 /* last off the live_slot generates a callback */
852 list_del_init(&slot->ds_live_item);
853 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
Sunil Mushrand6aa1c72010-10-06 18:50:50 -0700854 mlog(ML_HEARTBEAT, "o2hb: Remove node %d from live "
855 "nodes bitmap\n", slot->ds_node_num);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800856 clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
857
Sunil Mushran0e105d32010-10-07 17:00:16 -0700858 /* node can be null */
859 o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB,
860 node, slot->ds_node_num);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800861
862 changed = 1;
863 }
864
865 /* We don't clear this because the node is still
866 * actually writing new blocks. */
867 if (!gen_changed)
868 slot->ds_changed_samples = 0;
869 goto out;
870 }
871 if (slot->ds_changed_samples) {
872 slot->ds_changed_samples = 0;
873 slot->ds_equal_samples = 0;
874 }
875out:
Sunil Mushran43182d22010-10-06 17:55:16 -0700876 o2hb_set_quorum_device(reg, slot);
877
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800878 spin_unlock(&o2hb_live_lock);
879
880 o2hb_run_event_list(&event);
881
Sunil Mushran0e105d32010-10-07 17:00:16 -0700882 if (node)
883 o2nm_node_put(node);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800884 return changed;
885}
886
887/* This could be faster if we just implmented a find_last_bit, but I
888 * don't think the circumstances warrant it. */
889static int o2hb_highest_node(unsigned long *nodes,
890 int numbits)
891{
892 int highest, node;
893
894 highest = numbits;
895 node = -1;
896 while ((node = find_next_bit(nodes, numbits, node + 1)) != -1) {
897 if (node >= numbits)
898 break;
899
900 highest = node;
901 }
902
903 return highest;
904}
905
Mark Fasheha9e2ae32006-03-24 14:20:17 -0800906static int o2hb_do_disk_heartbeat(struct o2hb_region *reg)
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800907{
908 int i, ret, highest_node, change = 0;
909 unsigned long configured_nodes[BITS_TO_LONGS(O2NM_MAX_NODES)];
Sunil Mushran0e105d32010-10-07 17:00:16 -0700910 unsigned long live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800911 struct o2hb_bio_wait_ctxt write_wc;
912
Mark Fasheha9e2ae32006-03-24 14:20:17 -0800913 ret = o2nm_configured_node_map(configured_nodes,
914 sizeof(configured_nodes));
915 if (ret) {
916 mlog_errno(ret);
917 return ret;
918 }
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800919
Sunil Mushran0e105d32010-10-07 17:00:16 -0700920 /*
921 * If a node is not configured but is in the livemap, we still need
922 * to read the slot so as to be able to remove it from the livemap.
923 */
924 o2hb_fill_node_map(live_node_bitmap, sizeof(live_node_bitmap));
925 i = -1;
926 while ((i = find_next_bit(live_node_bitmap,
927 O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
928 set_bit(i, configured_nodes);
929 }
930
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800931 highest_node = o2hb_highest_node(configured_nodes, O2NM_MAX_NODES);
932 if (highest_node >= O2NM_MAX_NODES) {
933 mlog(ML_NOTICE, "ocfs2_heartbeat: no configured nodes found!\n");
Mark Fasheha9e2ae32006-03-24 14:20:17 -0800934 return -EINVAL;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800935 }
936
937 /* No sense in reading the slots of nodes that don't exist
938 * yet. Of course, if the node definitions have holes in them
939 * then we're reading an empty slot anyway... Consider this
940 * best-effort. */
941 ret = o2hb_read_slots(reg, highest_node + 1);
942 if (ret < 0) {
943 mlog_errno(ret);
Mark Fasheha9e2ae32006-03-24 14:20:17 -0800944 return ret;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800945 }
946
947 /* With an up to date view of the slots, we can check that no
948 * other node has been improperly configured to heartbeat in
949 * our slot. */
950 if (!o2hb_check_last_timestamp(reg))
951 mlog(ML_ERROR, "Device \"%s\": another node is heartbeating "
952 "in our slot!\n", reg->hr_dev_name);
953
954 /* fill in the proper info for our next heartbeat */
955 o2hb_prepare_block(reg, reg->hr_generation);
956
957 /* And fire off the write. Note that we don't wait on this I/O
958 * until later. */
Philipp Reisnerb5592922007-01-11 10:58:10 +0100959 ret = o2hb_issue_node_write(reg, &write_wc);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800960 if (ret < 0) {
961 mlog_errno(ret);
Mark Fasheha9e2ae32006-03-24 14:20:17 -0800962 return ret;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800963 }
964
965 i = -1;
966 while((i = find_next_bit(configured_nodes, O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
967
968 change |= o2hb_check_slot(reg, &reg->hr_slots[i]);
969 }
970
971 /*
972 * We have to be sure we've advertised ourselves on disk
973 * before we can go to steady state. This ensures that
974 * people we find in our steady state have seen us.
975 */
976 o2hb_wait_on_io(reg, &write_wc);
Mark Fasheha9e2ae32006-03-24 14:20:17 -0800977 if (write_wc.wc_error) {
978 /* Do not re-arm the write timeout on I/O error - we
979 * can't be sure that the new block ever made it to
980 * disk */
981 mlog(ML_ERROR, "Write error %d on device \"%s\"\n",
982 write_wc.wc_error, reg->hr_dev_name);
983 return write_wc.wc_error;
984 }
985
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800986 o2hb_arm_write_timeout(reg);
987
988 /* let the person who launched us know when things are steady */
989 if (!change && (atomic_read(&reg->hr_steady_iterations) != 0)) {
990 if (atomic_dec_and_test(&reg->hr_steady_iterations))
991 wake_up(&o2hb_steady_queue);
992 }
Mark Fasheha9e2ae32006-03-24 14:20:17 -0800993
994 return 0;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800995}
996
997/* Subtract b from a, storing the result in a. a *must* have a larger
998 * value than b. */
999static void o2hb_tv_subtract(struct timeval *a,
1000 struct timeval *b)
1001{
1002 /* just return 0 when a is after b */
1003 if (a->tv_sec < b->tv_sec ||
1004 (a->tv_sec == b->tv_sec && a->tv_usec < b->tv_usec)) {
1005 a->tv_sec = 0;
1006 a->tv_usec = 0;
1007 return;
1008 }
1009
1010 a->tv_sec -= b->tv_sec;
1011 a->tv_usec -= b->tv_usec;
1012 while ( a->tv_usec < 0 ) {
1013 a->tv_sec--;
1014 a->tv_usec += 1000000;
1015 }
1016}
1017
1018static unsigned int o2hb_elapsed_msecs(struct timeval *start,
1019 struct timeval *end)
1020{
1021 struct timeval res = *end;
1022
1023 o2hb_tv_subtract(&res, start);
1024
1025 return res.tv_sec * 1000 + res.tv_usec / 1000;
1026}
1027
1028/*
1029 * we ride the region ref that the region dir holds. before the region
1030 * dir is removed and drops it ref it will wait to tear down this
1031 * thread.
1032 */
1033static int o2hb_thread(void *data)
1034{
1035 int i, ret;
1036 struct o2hb_region *reg = data;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001037 struct o2hb_bio_wait_ctxt write_wc;
1038 struct timeval before_hb, after_hb;
1039 unsigned int elapsed_msec;
1040
1041 mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread running\n");
1042
1043 set_user_nice(current, -20);
1044
1045 while (!kthread_should_stop() && !reg->hr_unclean_stop) {
1046 /* We track the time spent inside
Frederik Schwarzer025dfda2008-10-16 19:02:37 +02001047 * o2hb_do_disk_heartbeat so that we avoid more than
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001048 * hr_timeout_ms between disk writes. On busy systems
1049 * this should result in a heartbeat which is less
1050 * likely to time itself out. */
1051 do_gettimeofday(&before_hb);
1052
Mark Fasheha9e2ae32006-03-24 14:20:17 -08001053 i = 0;
1054 do {
1055 ret = o2hb_do_disk_heartbeat(reg);
1056 } while (ret && ++i < 2);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001057
1058 do_gettimeofday(&after_hb);
1059 elapsed_msec = o2hb_elapsed_msecs(&before_hb, &after_hb);
1060
Tao Mab31d3082009-12-22 10:32:15 +08001061 mlog(ML_HEARTBEAT,
1062 "start = %lu.%lu, end = %lu.%lu, msec = %u\n",
Mark Fasheh215c7f92006-02-01 16:42:10 -08001063 before_hb.tv_sec, (unsigned long) before_hb.tv_usec,
1064 after_hb.tv_sec, (unsigned long) after_hb.tv_usec,
1065 elapsed_msec);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001066
1067 if (elapsed_msec < reg->hr_timeout_ms) {
1068 /* the kthread api has blocked signals for us so no
1069 * need to record the return value. */
1070 msleep_interruptible(reg->hr_timeout_ms - elapsed_msec);
1071 }
1072 }
1073
1074 o2hb_disarm_write_timeout(reg);
1075
1076 /* unclean stop is only used in very bad situation */
1077 for(i = 0; !reg->hr_unclean_stop && i < reg->hr_blocks; i++)
1078 o2hb_shutdown_slot(&reg->hr_slots[i]);
1079
1080 /* Explicit down notification - avoid forcing the other nodes
1081 * to timeout on this region when we could just as easily
1082 * write a clear generation - thus indicating to them that
1083 * this node has left this region.
1084 *
1085 * XXX: Should we skip this on unclean_stop? */
1086 o2hb_prepare_block(reg, 0);
Philipp Reisnerb5592922007-01-11 10:58:10 +01001087 ret = o2hb_issue_node_write(reg, &write_wc);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001088 if (ret == 0) {
1089 o2hb_wait_on_io(reg, &write_wc);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001090 } else {
1091 mlog_errno(ret);
1092 }
1093
1094 mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread exiting\n");
1095
1096 return 0;
1097}
1098
Sunil Mushran87d3d3f2008-12-17 14:17:42 -08001099#ifdef CONFIG_DEBUG_FS
1100static int o2hb_debug_open(struct inode *inode, struct file *file)
1101{
Sunil Mushran8ca8b0bb2010-10-07 17:01:27 -07001102 struct o2hb_debug_buf *db = inode->i_private;
Sunil Mushran1f285302010-10-06 17:55:12 -07001103 struct o2hb_region *reg;
Sunil Mushran87d3d3f2008-12-17 14:17:42 -08001104 unsigned long map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1105 char *buf = NULL;
1106 int i = -1;
1107 int out = 0;
1108
Sunil Mushran8ca8b0bb2010-10-07 17:01:27 -07001109 /* max_nodes should be the largest bitmap we pass here */
1110 BUG_ON(sizeof(map) < db->db_size);
1111
Sunil Mushran87d3d3f2008-12-17 14:17:42 -08001112 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1113 if (!buf)
1114 goto bail;
1115
Sunil Mushran8ca8b0bb2010-10-07 17:01:27 -07001116 switch (db->db_type) {
1117 case O2HB_DB_TYPE_LIVENODES:
Sunil Mushrana6de0132010-10-06 17:55:13 -07001118 case O2HB_DB_TYPE_LIVEREGIONS:
1119 case O2HB_DB_TYPE_QUORUMREGIONS:
1120 case O2HB_DB_TYPE_FAILEDREGIONS:
Sunil Mushran8ca8b0bb2010-10-07 17:01:27 -07001121 spin_lock(&o2hb_live_lock);
1122 memcpy(map, db->db_data, db->db_size);
1123 spin_unlock(&o2hb_live_lock);
1124 break;
Sunil Mushran87d3d3f2008-12-17 14:17:42 -08001125
Sunil Mushran1f285302010-10-06 17:55:12 -07001126 case O2HB_DB_TYPE_REGION_LIVENODES:
1127 spin_lock(&o2hb_live_lock);
1128 reg = (struct o2hb_region *)db->db_data;
1129 memcpy(map, reg->hr_live_node_bitmap, db->db_size);
1130 spin_unlock(&o2hb_live_lock);
1131 break;
1132
1133 case O2HB_DB_TYPE_REGION_NUMBER:
1134 reg = (struct o2hb_region *)db->db_data;
1135 out += snprintf(buf + out, PAGE_SIZE - out, "%d\n",
1136 reg->hr_region_num);
1137 goto done;
1138
Sunil Mushran43695d02010-10-06 17:55:09 -07001139 case O2HB_DB_TYPE_REGION_ELAPSED_TIME:
1140 reg = (struct o2hb_region *)db->db_data;
1141 out += snprintf(buf + out, PAGE_SIZE - out, "%u\n",
1142 jiffies_to_msecs(jiffies -
1143 reg->hr_last_timeout_start));
1144 goto done;
1145
Sunil Mushran8ca8b0bb2010-10-07 17:01:27 -07001146 default:
1147 goto done;
1148 }
1149
1150 while ((i = find_next_bit(map, db->db_len, i + 1)) < db->db_len)
Sunil Mushran87d3d3f2008-12-17 14:17:42 -08001151 out += snprintf(buf + out, PAGE_SIZE - out, "%d ", i);
1152 out += snprintf(buf + out, PAGE_SIZE - out, "\n");
1153
Sunil Mushran8ca8b0bb2010-10-07 17:01:27 -07001154done:
Sunil Mushran87d3d3f2008-12-17 14:17:42 -08001155 i_size_write(inode, out);
1156
1157 file->private_data = buf;
1158
1159 return 0;
1160bail:
1161 return -ENOMEM;
1162}
1163
1164static int o2hb_debug_release(struct inode *inode, struct file *file)
1165{
1166 kfree(file->private_data);
1167 return 0;
1168}
1169
1170static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1171 size_t nbytes, loff_t *ppos)
1172{
1173 return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
1174 i_size_read(file->f_mapping->host));
1175}
1176#else
1177static int o2hb_debug_open(struct inode *inode, struct file *file)
1178{
1179 return 0;
1180}
1181static int o2hb_debug_release(struct inode *inode, struct file *file)
1182{
1183 return 0;
1184}
1185static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1186 size_t nbytes, loff_t *ppos)
1187{
1188 return 0;
1189}
1190#endif /* CONFIG_DEBUG_FS */
1191
Alexey Dobriyan828c0952009-10-01 15:43:56 -07001192static const struct file_operations o2hb_debug_fops = {
Sunil Mushran87d3d3f2008-12-17 14:17:42 -08001193 .open = o2hb_debug_open,
1194 .release = o2hb_debug_release,
1195 .read = o2hb_debug_read,
1196 .llseek = generic_file_llseek,
1197};
1198
1199void o2hb_exit(void)
1200{
Sunil Mushran8ca8b0bb2010-10-07 17:01:27 -07001201 kfree(o2hb_db_livenodes);
Sunil Mushrana6de0132010-10-06 17:55:13 -07001202 kfree(o2hb_db_liveregions);
1203 kfree(o2hb_db_quorumregions);
1204 kfree(o2hb_db_failedregions);
1205 debugfs_remove(o2hb_debug_failedregions);
1206 debugfs_remove(o2hb_debug_quorumregions);
1207 debugfs_remove(o2hb_debug_liveregions);
Sunil Mushran8ca8b0bb2010-10-07 17:01:27 -07001208 debugfs_remove(o2hb_debug_livenodes);
1209 debugfs_remove(o2hb_debug_dir);
1210}
1211
1212static struct dentry *o2hb_debug_create(const char *name, struct dentry *dir,
1213 struct o2hb_debug_buf **db, int db_len,
1214 int type, int size, int len, void *data)
1215{
1216 *db = kmalloc(db_len, GFP_KERNEL);
1217 if (!*db)
1218 return NULL;
1219
1220 (*db)->db_type = type;
1221 (*db)->db_size = size;
1222 (*db)->db_len = len;
1223 (*db)->db_data = data;
1224
1225 return debugfs_create_file(name, S_IFREG|S_IRUSR, dir, *db,
1226 &o2hb_debug_fops);
1227}
1228
1229static int o2hb_debug_init(void)
1230{
1231 int ret = -ENOMEM;
1232
1233 o2hb_debug_dir = debugfs_create_dir(O2HB_DEBUG_DIR, NULL);
1234 if (!o2hb_debug_dir) {
1235 mlog_errno(ret);
1236 goto bail;
1237 }
1238
1239 o2hb_debug_livenodes = o2hb_debug_create(O2HB_DEBUG_LIVENODES,
1240 o2hb_debug_dir,
1241 &o2hb_db_livenodes,
1242 sizeof(*o2hb_db_livenodes),
1243 O2HB_DB_TYPE_LIVENODES,
1244 sizeof(o2hb_live_node_bitmap),
1245 O2NM_MAX_NODES,
1246 o2hb_live_node_bitmap);
1247 if (!o2hb_debug_livenodes) {
1248 mlog_errno(ret);
1249 goto bail;
1250 }
Sunil Mushrana6de0132010-10-06 17:55:13 -07001251
1252 o2hb_debug_liveregions = o2hb_debug_create(O2HB_DEBUG_LIVEREGIONS,
1253 o2hb_debug_dir,
1254 &o2hb_db_liveregions,
1255 sizeof(*o2hb_db_liveregions),
1256 O2HB_DB_TYPE_LIVEREGIONS,
1257 sizeof(o2hb_live_region_bitmap),
1258 O2NM_MAX_REGIONS,
1259 o2hb_live_region_bitmap);
1260 if (!o2hb_debug_liveregions) {
1261 mlog_errno(ret);
1262 goto bail;
1263 }
1264
1265 o2hb_debug_quorumregions =
1266 o2hb_debug_create(O2HB_DEBUG_QUORUMREGIONS,
1267 o2hb_debug_dir,
1268 &o2hb_db_quorumregions,
1269 sizeof(*o2hb_db_quorumregions),
1270 O2HB_DB_TYPE_QUORUMREGIONS,
1271 sizeof(o2hb_quorum_region_bitmap),
1272 O2NM_MAX_REGIONS,
1273 o2hb_quorum_region_bitmap);
1274 if (!o2hb_debug_quorumregions) {
1275 mlog_errno(ret);
1276 goto bail;
1277 }
1278
1279 o2hb_debug_failedregions =
1280 o2hb_debug_create(O2HB_DEBUG_FAILEDREGIONS,
1281 o2hb_debug_dir,
1282 &o2hb_db_failedregions,
1283 sizeof(*o2hb_db_failedregions),
1284 O2HB_DB_TYPE_FAILEDREGIONS,
1285 sizeof(o2hb_failed_region_bitmap),
1286 O2NM_MAX_REGIONS,
1287 o2hb_failed_region_bitmap);
1288 if (!o2hb_debug_failedregions) {
1289 mlog_errno(ret);
1290 goto bail;
1291 }
1292
Sunil Mushran8ca8b0bb2010-10-07 17:01:27 -07001293 ret = 0;
1294bail:
1295 if (ret)
1296 o2hb_exit();
1297
1298 return ret;
Sunil Mushran87d3d3f2008-12-17 14:17:42 -08001299}
1300
1301int o2hb_init(void)
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001302{
1303 int i;
1304
1305 for (i = 0; i < ARRAY_SIZE(o2hb_callbacks); i++)
1306 INIT_LIST_HEAD(&o2hb_callbacks[i].list);
1307
1308 for (i = 0; i < ARRAY_SIZE(o2hb_live_slots); i++)
1309 INIT_LIST_HEAD(&o2hb_live_slots[i]);
1310
1311 INIT_LIST_HEAD(&o2hb_node_events);
1312
1313 memset(o2hb_live_node_bitmap, 0, sizeof(o2hb_live_node_bitmap));
Sunil Mushran536f0742010-10-07 17:03:07 -07001314 memset(o2hb_region_bitmap, 0, sizeof(o2hb_region_bitmap));
Sunil Mushrane7d656b2010-10-06 17:55:18 -07001315 memset(o2hb_live_region_bitmap, 0, sizeof(o2hb_live_region_bitmap));
Sunil Mushran43182d22010-10-06 17:55:16 -07001316 memset(o2hb_quorum_region_bitmap, 0, sizeof(o2hb_quorum_region_bitmap));
Sunil Mushranb1c5ebf2010-10-07 17:05:52 -07001317 memset(o2hb_failed_region_bitmap, 0, sizeof(o2hb_failed_region_bitmap));
Sunil Mushran87d3d3f2008-12-17 14:17:42 -08001318
Sunil Mushran8ca8b0bb2010-10-07 17:01:27 -07001319 return o2hb_debug_init();
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001320}
1321
1322/* if we're already in a callback then we're already serialized by the sem */
1323static void o2hb_fill_node_map_from_callback(unsigned long *map,
1324 unsigned bytes)
1325{
1326 BUG_ON(bytes < (BITS_TO_LONGS(O2NM_MAX_NODES) * sizeof(unsigned long)));
1327
1328 memcpy(map, &o2hb_live_node_bitmap, bytes);
1329}
1330
1331/*
1332 * get a map of all nodes that are heartbeating in any regions
1333 */
1334void o2hb_fill_node_map(unsigned long *map, unsigned bytes)
1335{
1336 /* callers want to serialize this map and callbacks so that they
1337 * can trust that they don't miss nodes coming to the party */
1338 down_read(&o2hb_callback_sem);
1339 spin_lock(&o2hb_live_lock);
1340 o2hb_fill_node_map_from_callback(map, bytes);
1341 spin_unlock(&o2hb_live_lock);
1342 up_read(&o2hb_callback_sem);
1343}
1344EXPORT_SYMBOL_GPL(o2hb_fill_node_map);
1345
1346/*
1347 * heartbeat configfs bits. The heartbeat set is a default set under
1348 * the cluster set in nodemanager.c.
1349 */
1350
1351static struct o2hb_region *to_o2hb_region(struct config_item *item)
1352{
1353 return item ? container_of(item, struct o2hb_region, hr_item) : NULL;
1354}
1355
1356/* drop_item only drops its ref after killing the thread, nothing should
1357 * be using the region anymore. this has to clean up any state that
1358 * attributes might have built up. */
1359static void o2hb_region_release(struct config_item *item)
1360{
1361 int i;
1362 struct page *page;
1363 struct o2hb_region *reg = to_o2hb_region(item);
1364
1365 if (reg->hr_tmp_block)
1366 kfree(reg->hr_tmp_block);
1367
1368 if (reg->hr_slot_data) {
1369 for (i = 0; i < reg->hr_num_pages; i++) {
1370 page = reg->hr_slot_data[i];
1371 if (page)
1372 __free_page(page);
1373 }
1374 kfree(reg->hr_slot_data);
1375 }
1376
1377 if (reg->hr_bdev)
Al Viro9a1c3542008-02-22 20:40:24 -05001378 blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001379
1380 if (reg->hr_slots)
1381 kfree(reg->hr_slots);
1382
Sunil Mushran1f285302010-10-06 17:55:12 -07001383 kfree(reg->hr_db_regnum);
1384 kfree(reg->hr_db_livenodes);
1385 debugfs_remove(reg->hr_debug_livenodes);
1386 debugfs_remove(reg->hr_debug_regnum);
1387 debugfs_remove(reg->hr_debug_dir);
1388
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001389 spin_lock(&o2hb_live_lock);
1390 list_del(&reg->hr_all_item);
1391 spin_unlock(&o2hb_live_lock);
1392
1393 kfree(reg);
1394}
1395
1396static int o2hb_read_block_input(struct o2hb_region *reg,
1397 const char *page,
1398 size_t count,
1399 unsigned long *ret_bytes,
1400 unsigned int *ret_bits)
1401{
1402 unsigned long bytes;
1403 char *p = (char *)page;
1404
1405 bytes = simple_strtoul(p, &p, 0);
1406 if (!p || (*p && (*p != '\n')))
1407 return -EINVAL;
1408
1409 /* Heartbeat and fs min / max block sizes are the same. */
1410 if (bytes > 4096 || bytes < 512)
1411 return -ERANGE;
1412 if (hweight16(bytes) != 1)
1413 return -EINVAL;
1414
1415 if (ret_bytes)
1416 *ret_bytes = bytes;
1417 if (ret_bits)
1418 *ret_bits = ffs(bytes) - 1;
1419
1420 return 0;
1421}
1422
1423static ssize_t o2hb_region_block_bytes_read(struct o2hb_region *reg,
1424 char *page)
1425{
1426 return sprintf(page, "%u\n", reg->hr_block_bytes);
1427}
1428
1429static ssize_t o2hb_region_block_bytes_write(struct o2hb_region *reg,
1430 const char *page,
1431 size_t count)
1432{
1433 int status;
1434 unsigned long block_bytes;
1435 unsigned int block_bits;
1436
1437 if (reg->hr_bdev)
1438 return -EINVAL;
1439
1440 status = o2hb_read_block_input(reg, page, count,
1441 &block_bytes, &block_bits);
1442 if (status)
1443 return status;
1444
1445 reg->hr_block_bytes = (unsigned int)block_bytes;
1446 reg->hr_block_bits = block_bits;
1447
1448 return count;
1449}
1450
1451static ssize_t o2hb_region_start_block_read(struct o2hb_region *reg,
1452 char *page)
1453{
1454 return sprintf(page, "%llu\n", reg->hr_start_block);
1455}
1456
1457static ssize_t o2hb_region_start_block_write(struct o2hb_region *reg,
1458 const char *page,
1459 size_t count)
1460{
1461 unsigned long long tmp;
1462 char *p = (char *)page;
1463
1464 if (reg->hr_bdev)
1465 return -EINVAL;
1466
1467 tmp = simple_strtoull(p, &p, 0);
1468 if (!p || (*p && (*p != '\n')))
1469 return -EINVAL;
1470
1471 reg->hr_start_block = tmp;
1472
1473 return count;
1474}
1475
1476static ssize_t o2hb_region_blocks_read(struct o2hb_region *reg,
1477 char *page)
1478{
1479 return sprintf(page, "%d\n", reg->hr_blocks);
1480}
1481
1482static ssize_t o2hb_region_blocks_write(struct o2hb_region *reg,
1483 const char *page,
1484 size_t count)
1485{
1486 unsigned long tmp;
1487 char *p = (char *)page;
1488
1489 if (reg->hr_bdev)
1490 return -EINVAL;
1491
1492 tmp = simple_strtoul(p, &p, 0);
1493 if (!p || (*p && (*p != '\n')))
1494 return -EINVAL;
1495
1496 if (tmp > O2NM_MAX_NODES || tmp == 0)
1497 return -ERANGE;
1498
1499 reg->hr_blocks = (unsigned int)tmp;
1500
1501 return count;
1502}
1503
1504static ssize_t o2hb_region_dev_read(struct o2hb_region *reg,
1505 char *page)
1506{
1507 unsigned int ret = 0;
1508
1509 if (reg->hr_bdev)
1510 ret = sprintf(page, "%s\n", reg->hr_dev_name);
1511
1512 return ret;
1513}
1514
1515static void o2hb_init_region_params(struct o2hb_region *reg)
1516{
1517 reg->hr_slots_per_page = PAGE_CACHE_SIZE >> reg->hr_block_bits;
1518 reg->hr_timeout_ms = O2HB_REGION_TIMEOUT_MS;
1519
1520 mlog(ML_HEARTBEAT, "hr_start_block = %llu, hr_blocks = %u\n",
1521 reg->hr_start_block, reg->hr_blocks);
1522 mlog(ML_HEARTBEAT, "hr_block_bytes = %u, hr_block_bits = %u\n",
1523 reg->hr_block_bytes, reg->hr_block_bits);
1524 mlog(ML_HEARTBEAT, "hr_timeout_ms = %u\n", reg->hr_timeout_ms);
1525 mlog(ML_HEARTBEAT, "dead threshold = %u\n", o2hb_dead_threshold);
1526}
1527
1528static int o2hb_map_slot_data(struct o2hb_region *reg)
1529{
1530 int i, j;
1531 unsigned int last_slot;
1532 unsigned int spp = reg->hr_slots_per_page;
1533 struct page *page;
1534 char *raw;
1535 struct o2hb_disk_slot *slot;
1536
1537 reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL);
1538 if (reg->hr_tmp_block == NULL) {
1539 mlog_errno(-ENOMEM);
1540 return -ENOMEM;
1541 }
1542
1543 reg->hr_slots = kcalloc(reg->hr_blocks,
1544 sizeof(struct o2hb_disk_slot), GFP_KERNEL);
1545 if (reg->hr_slots == NULL) {
1546 mlog_errno(-ENOMEM);
1547 return -ENOMEM;
1548 }
1549
1550 for(i = 0; i < reg->hr_blocks; i++) {
1551 slot = &reg->hr_slots[i];
1552 slot->ds_node_num = i;
1553 INIT_LIST_HEAD(&slot->ds_live_item);
1554 slot->ds_raw_block = NULL;
1555 }
1556
1557 reg->hr_num_pages = (reg->hr_blocks + spp - 1) / spp;
1558 mlog(ML_HEARTBEAT, "Going to require %u pages to cover %u blocks "
1559 "at %u blocks per page\n",
1560 reg->hr_num_pages, reg->hr_blocks, spp);
1561
1562 reg->hr_slot_data = kcalloc(reg->hr_num_pages, sizeof(struct page *),
1563 GFP_KERNEL);
1564 if (!reg->hr_slot_data) {
1565 mlog_errno(-ENOMEM);
1566 return -ENOMEM;
1567 }
1568
1569 for(i = 0; i < reg->hr_num_pages; i++) {
1570 page = alloc_page(GFP_KERNEL);
1571 if (!page) {
1572 mlog_errno(-ENOMEM);
1573 return -ENOMEM;
1574 }
1575
1576 reg->hr_slot_data[i] = page;
1577
1578 last_slot = i * spp;
1579 raw = page_address(page);
1580 for (j = 0;
1581 (j < spp) && ((j + last_slot) < reg->hr_blocks);
1582 j++) {
1583 BUG_ON((j + last_slot) >= reg->hr_blocks);
1584
1585 slot = &reg->hr_slots[j + last_slot];
1586 slot->ds_raw_block =
1587 (struct o2hb_disk_heartbeat_block *) raw;
1588
1589 raw += reg->hr_block_bytes;
1590 }
1591 }
1592
1593 return 0;
1594}
1595
1596/* Read in all the slots available and populate the tracking
1597 * structures so that we can start with a baseline idea of what's
1598 * there. */
1599static int o2hb_populate_slot_data(struct o2hb_region *reg)
1600{
1601 int ret, i;
1602 struct o2hb_disk_slot *slot;
1603 struct o2hb_disk_heartbeat_block *hb_block;
1604
1605 mlog_entry_void();
1606
1607 ret = o2hb_read_slots(reg, reg->hr_blocks);
1608 if (ret) {
1609 mlog_errno(ret);
1610 goto out;
1611 }
1612
1613 /* We only want to get an idea of the values initially in each
1614 * slot, so we do no verification - o2hb_check_slot will
1615 * actually determine if each configured slot is valid and
1616 * whether any values have changed. */
1617 for(i = 0; i < reg->hr_blocks; i++) {
1618 slot = &reg->hr_slots[i];
1619 hb_block = (struct o2hb_disk_heartbeat_block *) slot->ds_raw_block;
1620
1621 /* Only fill the values that o2hb_check_slot uses to
1622 * determine changing slots */
1623 slot->ds_last_time = le64_to_cpu(hb_block->hb_seq);
1624 slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
1625 }
1626
1627out:
1628 mlog_exit(ret);
1629 return ret;
1630}
1631
1632/* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */
1633static ssize_t o2hb_region_dev_write(struct o2hb_region *reg,
1634 const char *page,
1635 size_t count)
1636{
Joel Beckere6c352d2007-02-03 03:04:20 -08001637 struct task_struct *hb_task;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001638 long fd;
1639 int sectsize;
1640 char *p = (char *)page;
1641 struct file *filp = NULL;
1642 struct inode *inode = NULL;
1643 ssize_t ret = -EINVAL;
1644
1645 if (reg->hr_bdev)
1646 goto out;
1647
1648 /* We can't heartbeat without having had our node number
1649 * configured yet. */
1650 if (o2nm_this_node() == O2NM_MAX_NODES)
1651 goto out;
1652
1653 fd = simple_strtol(p, &p, 0);
1654 if (!p || (*p && (*p != '\n')))
1655 goto out;
1656
1657 if (fd < 0 || fd >= INT_MAX)
1658 goto out;
1659
1660 filp = fget(fd);
1661 if (filp == NULL)
1662 goto out;
1663
1664 if (reg->hr_blocks == 0 || reg->hr_start_block == 0 ||
1665 reg->hr_block_bytes == 0)
1666 goto out;
1667
1668 inode = igrab(filp->f_mapping->host);
1669 if (inode == NULL)
1670 goto out;
1671
1672 if (!S_ISBLK(inode->i_mode))
1673 goto out;
1674
1675 reg->hr_bdev = I_BDEV(filp->f_mapping->host);
Al Viro572c4892007-10-08 13:24:05 -04001676 ret = blkdev_get(reg->hr_bdev, FMODE_WRITE | FMODE_READ);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001677 if (ret) {
1678 reg->hr_bdev = NULL;
1679 goto out;
1680 }
1681 inode = NULL;
1682
1683 bdevname(reg->hr_bdev, reg->hr_dev_name);
1684
Martin K. Petersene1defc42009-05-22 17:17:49 -04001685 sectsize = bdev_logical_block_size(reg->hr_bdev);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001686 if (sectsize != reg->hr_block_bytes) {
1687 mlog(ML_ERROR,
1688 "blocksize %u incorrect for device, expected %d",
1689 reg->hr_block_bytes, sectsize);
1690 ret = -EINVAL;
1691 goto out;
1692 }
1693
1694 o2hb_init_region_params(reg);
1695
1696 /* Generation of zero is invalid */
1697 do {
1698 get_random_bytes(&reg->hr_generation,
1699 sizeof(reg->hr_generation));
1700 } while (reg->hr_generation == 0);
1701
1702 ret = o2hb_map_slot_data(reg);
1703 if (ret) {
1704 mlog_errno(ret);
1705 goto out;
1706 }
1707
1708 ret = o2hb_populate_slot_data(reg);
1709 if (ret) {
1710 mlog_errno(ret);
1711 goto out;
1712 }
1713
David Howellsc4028952006-11-22 14:57:56 +00001714 INIT_DELAYED_WORK(&reg->hr_write_timeout_work, o2hb_write_timeout);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001715
1716 /*
1717 * A node is considered live after it has beat LIVE_THRESHOLD
1718 * times. We're not steady until we've given them a chance
1719 * _after_ our first read.
1720 */
1721 atomic_set(&reg->hr_steady_iterations, O2HB_LIVE_THRESHOLD + 1);
1722
Joel Beckere6c352d2007-02-03 03:04:20 -08001723 hb_task = kthread_run(o2hb_thread, reg, "o2hb-%s",
1724 reg->hr_item.ci_name);
1725 if (IS_ERR(hb_task)) {
1726 ret = PTR_ERR(hb_task);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001727 mlog_errno(ret);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001728 goto out;
1729 }
1730
Joel Beckere6c352d2007-02-03 03:04:20 -08001731 spin_lock(&o2hb_live_lock);
1732 reg->hr_task = hb_task;
1733 spin_unlock(&o2hb_live_lock);
1734
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001735 ret = wait_event_interruptible(o2hb_steady_queue,
1736 atomic_read(&reg->hr_steady_iterations) == 0);
1737 if (ret) {
Joel Beckere6df3a62007-02-06 15:45:39 -08001738 /* We got interrupted (hello ptrace!). Clean up */
Joel Beckere6c352d2007-02-03 03:04:20 -08001739 spin_lock(&o2hb_live_lock);
1740 hb_task = reg->hr_task;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001741 reg->hr_task = NULL;
Joel Beckere6c352d2007-02-03 03:04:20 -08001742 spin_unlock(&o2hb_live_lock);
1743
1744 if (hb_task)
1745 kthread_stop(hb_task);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001746 goto out;
1747 }
1748
Joel Beckere6df3a62007-02-06 15:45:39 -08001749 /* Ok, we were woken. Make sure it wasn't by drop_item() */
1750 spin_lock(&o2hb_live_lock);
1751 hb_task = reg->hr_task;
Sunil Mushrane7d656b2010-10-06 17:55:18 -07001752 if (o2hb_global_heartbeat_active())
1753 set_bit(reg->hr_region_num, o2hb_live_region_bitmap);
Joel Beckere6df3a62007-02-06 15:45:39 -08001754 spin_unlock(&o2hb_live_lock);
1755
1756 if (hb_task)
1757 ret = count;
1758 else
1759 ret = -EIO;
1760
Sunil Mushran18c50cb2010-10-06 18:26:59 -07001761 if (hb_task && o2hb_global_heartbeat_active())
1762 printk(KERN_NOTICE "o2hb: Heartbeat started on region %s\n",
1763 config_item_name(&reg->hr_item));
1764
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001765out:
1766 if (filp)
1767 fput(filp);
1768 if (inode)
1769 iput(inode);
1770 if (ret < 0) {
1771 if (reg->hr_bdev) {
Al Viro9a1c3542008-02-22 20:40:24 -05001772 blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001773 reg->hr_bdev = NULL;
1774 }
1775 }
1776 return ret;
1777}
1778
Zhen Wei92efc152006-12-08 00:48:17 -07001779static ssize_t o2hb_region_pid_read(struct o2hb_region *reg,
1780 char *page)
1781{
Joel Beckere6c352d2007-02-03 03:04:20 -08001782 pid_t pid = 0;
1783
1784 spin_lock(&o2hb_live_lock);
1785 if (reg->hr_task)
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001786 pid = task_pid_nr(reg->hr_task);
Joel Beckere6c352d2007-02-03 03:04:20 -08001787 spin_unlock(&o2hb_live_lock);
1788
1789 if (!pid)
Zhen Wei92efc152006-12-08 00:48:17 -07001790 return 0;
1791
Joel Beckere6c352d2007-02-03 03:04:20 -08001792 return sprintf(page, "%u\n", pid);
Zhen Wei92efc152006-12-08 00:48:17 -07001793}
1794
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001795struct o2hb_region_attribute {
1796 struct configfs_attribute attr;
1797 ssize_t (*show)(struct o2hb_region *, char *);
1798 ssize_t (*store)(struct o2hb_region *, const char *, size_t);
1799};
1800
1801static struct o2hb_region_attribute o2hb_region_attr_block_bytes = {
1802 .attr = { .ca_owner = THIS_MODULE,
1803 .ca_name = "block_bytes",
1804 .ca_mode = S_IRUGO | S_IWUSR },
1805 .show = o2hb_region_block_bytes_read,
1806 .store = o2hb_region_block_bytes_write,
1807};
1808
1809static struct o2hb_region_attribute o2hb_region_attr_start_block = {
1810 .attr = { .ca_owner = THIS_MODULE,
1811 .ca_name = "start_block",
1812 .ca_mode = S_IRUGO | S_IWUSR },
1813 .show = o2hb_region_start_block_read,
1814 .store = o2hb_region_start_block_write,
1815};
1816
1817static struct o2hb_region_attribute o2hb_region_attr_blocks = {
1818 .attr = { .ca_owner = THIS_MODULE,
1819 .ca_name = "blocks",
1820 .ca_mode = S_IRUGO | S_IWUSR },
1821 .show = o2hb_region_blocks_read,
1822 .store = o2hb_region_blocks_write,
1823};
1824
1825static struct o2hb_region_attribute o2hb_region_attr_dev = {
1826 .attr = { .ca_owner = THIS_MODULE,
1827 .ca_name = "dev",
1828 .ca_mode = S_IRUGO | S_IWUSR },
1829 .show = o2hb_region_dev_read,
1830 .store = o2hb_region_dev_write,
1831};
1832
Zhen Wei92efc152006-12-08 00:48:17 -07001833static struct o2hb_region_attribute o2hb_region_attr_pid = {
1834 .attr = { .ca_owner = THIS_MODULE,
1835 .ca_name = "pid",
1836 .ca_mode = S_IRUGO | S_IRUSR },
1837 .show = o2hb_region_pid_read,
1838};
1839
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001840static struct configfs_attribute *o2hb_region_attrs[] = {
1841 &o2hb_region_attr_block_bytes.attr,
1842 &o2hb_region_attr_start_block.attr,
1843 &o2hb_region_attr_blocks.attr,
1844 &o2hb_region_attr_dev.attr,
Zhen Wei92efc152006-12-08 00:48:17 -07001845 &o2hb_region_attr_pid.attr,
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001846 NULL,
1847};
1848
1849static ssize_t o2hb_region_show(struct config_item *item,
1850 struct configfs_attribute *attr,
1851 char *page)
1852{
1853 struct o2hb_region *reg = to_o2hb_region(item);
1854 struct o2hb_region_attribute *o2hb_region_attr =
1855 container_of(attr, struct o2hb_region_attribute, attr);
1856 ssize_t ret = 0;
1857
1858 if (o2hb_region_attr->show)
1859 ret = o2hb_region_attr->show(reg, page);
1860 return ret;
1861}
1862
1863static ssize_t o2hb_region_store(struct config_item *item,
1864 struct configfs_attribute *attr,
1865 const char *page, size_t count)
1866{
1867 struct o2hb_region *reg = to_o2hb_region(item);
1868 struct o2hb_region_attribute *o2hb_region_attr =
1869 container_of(attr, struct o2hb_region_attribute, attr);
1870 ssize_t ret = -EINVAL;
1871
1872 if (o2hb_region_attr->store)
1873 ret = o2hb_region_attr->store(reg, page, count);
1874 return ret;
1875}
1876
1877static struct configfs_item_operations o2hb_region_item_ops = {
1878 .release = o2hb_region_release,
1879 .show_attribute = o2hb_region_show,
1880 .store_attribute = o2hb_region_store,
1881};
1882
1883static struct config_item_type o2hb_region_type = {
1884 .ct_item_ops = &o2hb_region_item_ops,
1885 .ct_attrs = o2hb_region_attrs,
1886 .ct_owner = THIS_MODULE,
1887};
1888
1889/* heartbeat set */
1890
1891struct o2hb_heartbeat_group {
1892 struct config_group hs_group;
1893 /* some stuff? */
1894};
1895
1896static struct o2hb_heartbeat_group *to_o2hb_heartbeat_group(struct config_group *group)
1897{
1898 return group ?
1899 container_of(group, struct o2hb_heartbeat_group, hs_group)
1900 : NULL;
1901}
1902
Sunil Mushran1f285302010-10-06 17:55:12 -07001903static int o2hb_debug_region_init(struct o2hb_region *reg, struct dentry *dir)
1904{
1905 int ret = -ENOMEM;
1906
1907 reg->hr_debug_dir =
1908 debugfs_create_dir(config_item_name(&reg->hr_item), dir);
1909 if (!reg->hr_debug_dir) {
1910 mlog_errno(ret);
1911 goto bail;
1912 }
1913
1914 reg->hr_debug_livenodes =
1915 o2hb_debug_create(O2HB_DEBUG_LIVENODES,
1916 reg->hr_debug_dir,
1917 &(reg->hr_db_livenodes),
1918 sizeof(*(reg->hr_db_livenodes)),
1919 O2HB_DB_TYPE_REGION_LIVENODES,
1920 sizeof(reg->hr_live_node_bitmap),
1921 O2NM_MAX_NODES, reg);
1922 if (!reg->hr_debug_livenodes) {
1923 mlog_errno(ret);
1924 goto bail;
1925 }
1926
1927 reg->hr_debug_regnum =
1928 o2hb_debug_create(O2HB_DEBUG_REGION_NUMBER,
1929 reg->hr_debug_dir,
1930 &(reg->hr_db_regnum),
1931 sizeof(*(reg->hr_db_regnum)),
1932 O2HB_DB_TYPE_REGION_NUMBER,
1933 0, O2NM_MAX_NODES, reg);
1934 if (!reg->hr_debug_regnum) {
1935 mlog_errno(ret);
1936 goto bail;
1937 }
1938
Sunil Mushran43695d02010-10-06 17:55:09 -07001939 reg->hr_debug_elapsed_time =
1940 o2hb_debug_create(O2HB_DEBUG_REGION_ELAPSED_TIME,
1941 reg->hr_debug_dir,
1942 &(reg->hr_db_elapsed_time),
1943 sizeof(*(reg->hr_db_elapsed_time)),
1944 O2HB_DB_TYPE_REGION_ELAPSED_TIME,
1945 0, 0, reg);
1946 if (!reg->hr_debug_elapsed_time) {
1947 mlog_errno(ret);
1948 goto bail;
1949 }
1950
Sunil Mushran1f285302010-10-06 17:55:12 -07001951 ret = 0;
1952bail:
1953 return ret;
1954}
1955
Joel Beckerf89ab862008-07-17 14:53:48 -07001956static struct config_item *o2hb_heartbeat_group_make_item(struct config_group *group,
1957 const char *name)
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001958{
1959 struct o2hb_region *reg = NULL;
Sunil Mushran1f285302010-10-06 17:55:12 -07001960 int ret;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001961
Robert P. J. Daycd861282006-12-13 00:34:52 -08001962 reg = kzalloc(sizeof(struct o2hb_region), GFP_KERNEL);
Joel Beckerf89ab862008-07-17 14:53:48 -07001963 if (reg == NULL)
Joel Beckera6795e92008-07-17 15:21:29 -07001964 return ERR_PTR(-ENOMEM);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001965
Sunil Mushranb3c85c42010-10-07 14:31:06 -07001966 if (strlen(name) > O2HB_MAX_REGION_NAME_LEN)
1967 return ERR_PTR(-ENAMETOOLONG);
1968
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001969 spin_lock(&o2hb_live_lock);
Sunil Mushran536f0742010-10-07 17:03:07 -07001970 reg->hr_region_num = 0;
1971 if (o2hb_global_heartbeat_active()) {
1972 reg->hr_region_num = find_first_zero_bit(o2hb_region_bitmap,
1973 O2NM_MAX_REGIONS);
1974 if (reg->hr_region_num >= O2NM_MAX_REGIONS) {
1975 spin_unlock(&o2hb_live_lock);
1976 return ERR_PTR(-EFBIG);
1977 }
1978 set_bit(reg->hr_region_num, o2hb_region_bitmap);
1979 }
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001980 list_add_tail(&reg->hr_all_item, &o2hb_all_regions);
1981 spin_unlock(&o2hb_live_lock);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001982
Sunil Mushran536f0742010-10-07 17:03:07 -07001983 config_item_init_type_name(&reg->hr_item, name, &o2hb_region_type);
1984
Sunil Mushran1f285302010-10-06 17:55:12 -07001985 ret = o2hb_debug_region_init(reg, o2hb_debug_dir);
1986 if (ret) {
1987 config_item_put(&reg->hr_item);
1988 return ERR_PTR(ret);
1989 }
1990
Joel Beckera6795e92008-07-17 15:21:29 -07001991 return &reg->hr_item;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001992}
1993
1994static void o2hb_heartbeat_group_drop_item(struct config_group *group,
1995 struct config_item *item)
1996{
Joel Beckere6c352d2007-02-03 03:04:20 -08001997 struct task_struct *hb_task;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001998 struct o2hb_region *reg = to_o2hb_region(item);
1999
2000 /* stop the thread when the user removes the region dir */
Joel Beckere6c352d2007-02-03 03:04:20 -08002001 spin_lock(&o2hb_live_lock);
Sunil Mushrane7d656b2010-10-06 17:55:18 -07002002 if (o2hb_global_heartbeat_active()) {
Sunil Mushran536f0742010-10-07 17:03:07 -07002003 clear_bit(reg->hr_region_num, o2hb_region_bitmap);
Sunil Mushrane7d656b2010-10-06 17:55:18 -07002004 clear_bit(reg->hr_region_num, o2hb_live_region_bitmap);
2005 }
Joel Beckere6c352d2007-02-03 03:04:20 -08002006 hb_task = reg->hr_task;
2007 reg->hr_task = NULL;
2008 spin_unlock(&o2hb_live_lock);
2009
2010 if (hb_task)
2011 kthread_stop(hb_task);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08002012
Joel Beckere6df3a62007-02-06 15:45:39 -08002013 /*
2014 * If we're racing a dev_write(), we need to wake them. They will
2015 * check reg->hr_task
2016 */
2017 if (atomic_read(&reg->hr_steady_iterations) != 0) {
2018 atomic_set(&reg->hr_steady_iterations, 0);
2019 wake_up(&o2hb_steady_queue);
2020 }
2021
Sunil Mushran18c50cb2010-10-06 18:26:59 -07002022 if (o2hb_global_heartbeat_active())
2023 printk(KERN_NOTICE "o2hb: Heartbeat stopped on region %s\n",
2024 config_item_name(&reg->hr_item));
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08002025 config_item_put(item);
2026}
2027
2028struct o2hb_heartbeat_group_attribute {
2029 struct configfs_attribute attr;
2030 ssize_t (*show)(struct o2hb_heartbeat_group *, char *);
2031 ssize_t (*store)(struct o2hb_heartbeat_group *, const char *, size_t);
2032};
2033
2034static ssize_t o2hb_heartbeat_group_show(struct config_item *item,
2035 struct configfs_attribute *attr,
2036 char *page)
2037{
2038 struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
2039 struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
2040 container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
2041 ssize_t ret = 0;
2042
2043 if (o2hb_heartbeat_group_attr->show)
2044 ret = o2hb_heartbeat_group_attr->show(reg, page);
2045 return ret;
2046}
2047
2048static ssize_t o2hb_heartbeat_group_store(struct config_item *item,
2049 struct configfs_attribute *attr,
2050 const char *page, size_t count)
2051{
2052 struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
2053 struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
2054 container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
2055 ssize_t ret = -EINVAL;
2056
2057 if (o2hb_heartbeat_group_attr->store)
2058 ret = o2hb_heartbeat_group_attr->store(reg, page, count);
2059 return ret;
2060}
2061
2062static ssize_t o2hb_heartbeat_group_threshold_show(struct o2hb_heartbeat_group *group,
2063 char *page)
2064{
2065 return sprintf(page, "%u\n", o2hb_dead_threshold);
2066}
2067
2068static ssize_t o2hb_heartbeat_group_threshold_store(struct o2hb_heartbeat_group *group,
2069 const char *page,
2070 size_t count)
2071{
2072 unsigned long tmp;
2073 char *p = (char *)page;
2074
2075 tmp = simple_strtoul(p, &p, 10);
2076 if (!p || (*p && (*p != '\n')))
2077 return -EINVAL;
2078
2079 /* this will validate ranges for us. */
2080 o2hb_dead_threshold_set((unsigned int) tmp);
2081
2082 return count;
2083}
2084
Sunil Mushran54b51872010-10-07 15:26:08 -07002085static
2086ssize_t o2hb_heartbeat_group_mode_show(struct o2hb_heartbeat_group *group,
2087 char *page)
2088{
2089 return sprintf(page, "%s\n",
2090 o2hb_heartbeat_mode_desc[o2hb_heartbeat_mode]);
2091}
2092
2093static
2094ssize_t o2hb_heartbeat_group_mode_store(struct o2hb_heartbeat_group *group,
2095 const char *page, size_t count)
2096{
2097 unsigned int i;
2098 int ret;
2099 size_t len;
2100
2101 len = (page[count - 1] == '\n') ? count - 1 : count;
2102 if (!len)
2103 return -EINVAL;
2104
2105 for (i = 0; i < O2HB_HEARTBEAT_NUM_MODES; ++i) {
2106 if (strnicmp(page, o2hb_heartbeat_mode_desc[i], len))
2107 continue;
2108
2109 ret = o2hb_global_hearbeat_mode_set(i);
2110 if (!ret)
Sunil Mushran18c50cb2010-10-06 18:26:59 -07002111 printk(KERN_NOTICE "o2hb: Heartbeat mode set to %s\n",
Sunil Mushran54b51872010-10-07 15:26:08 -07002112 o2hb_heartbeat_mode_desc[i]);
2113 return count;
2114 }
2115
2116 return -EINVAL;
2117
2118}
2119
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08002120static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_threshold = {
2121 .attr = { .ca_owner = THIS_MODULE,
2122 .ca_name = "dead_threshold",
2123 .ca_mode = S_IRUGO | S_IWUSR },
2124 .show = o2hb_heartbeat_group_threshold_show,
2125 .store = o2hb_heartbeat_group_threshold_store,
2126};
2127
Sunil Mushran54b51872010-10-07 15:26:08 -07002128static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_mode = {
2129 .attr = { .ca_owner = THIS_MODULE,
2130 .ca_name = "mode",
2131 .ca_mode = S_IRUGO | S_IWUSR },
2132 .show = o2hb_heartbeat_group_mode_show,
2133 .store = o2hb_heartbeat_group_mode_store,
2134};
2135
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08002136static struct configfs_attribute *o2hb_heartbeat_group_attrs[] = {
2137 &o2hb_heartbeat_group_attr_threshold.attr,
Sunil Mushran54b51872010-10-07 15:26:08 -07002138 &o2hb_heartbeat_group_attr_mode.attr,
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08002139 NULL,
2140};
2141
2142static struct configfs_item_operations o2hb_hearbeat_group_item_ops = {
2143 .show_attribute = o2hb_heartbeat_group_show,
2144 .store_attribute = o2hb_heartbeat_group_store,
2145};
2146
2147static struct configfs_group_operations o2hb_heartbeat_group_group_ops = {
2148 .make_item = o2hb_heartbeat_group_make_item,
2149 .drop_item = o2hb_heartbeat_group_drop_item,
2150};
2151
2152static struct config_item_type o2hb_heartbeat_group_type = {
2153 .ct_group_ops = &o2hb_heartbeat_group_group_ops,
2154 .ct_item_ops = &o2hb_hearbeat_group_item_ops,
2155 .ct_attrs = o2hb_heartbeat_group_attrs,
2156 .ct_owner = THIS_MODULE,
2157};
2158
2159/* this is just here to avoid touching group in heartbeat.h which the
2160 * entire damn world #includes */
2161struct config_group *o2hb_alloc_hb_set(void)
2162{
2163 struct o2hb_heartbeat_group *hs = NULL;
2164 struct config_group *ret = NULL;
2165
Robert P. J. Daycd861282006-12-13 00:34:52 -08002166 hs = kzalloc(sizeof(struct o2hb_heartbeat_group), GFP_KERNEL);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08002167 if (hs == NULL)
2168 goto out;
2169
2170 config_group_init_type_name(&hs->hs_group, "heartbeat",
2171 &o2hb_heartbeat_group_type);
2172
2173 ret = &hs->hs_group;
2174out:
2175 if (ret == NULL)
2176 kfree(hs);
2177 return ret;
2178}
2179
2180void o2hb_free_hb_set(struct config_group *group)
2181{
2182 struct o2hb_heartbeat_group *hs = to_o2hb_heartbeat_group(group);
2183 kfree(hs);
2184}
2185
2186/* hb callback registration and issueing */
2187
2188static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type)
2189{
2190 if (type == O2HB_NUM_CB)
2191 return ERR_PTR(-EINVAL);
2192
2193 return &o2hb_callbacks[type];
2194}
2195
2196void o2hb_setup_callback(struct o2hb_callback_func *hc,
2197 enum o2hb_callback_type type,
2198 o2hb_cb_func *func,
2199 void *data,
2200 int priority)
2201{
2202 INIT_LIST_HEAD(&hc->hc_item);
2203 hc->hc_func = func;
2204 hc->hc_data = data;
2205 hc->hc_priority = priority;
2206 hc->hc_type = type;
2207 hc->hc_magic = O2HB_CB_MAGIC;
2208}
2209EXPORT_SYMBOL_GPL(o2hb_setup_callback);
2210
Joel Becker14829422007-06-14 21:40:49 -07002211static struct o2hb_region *o2hb_find_region(const char *region_uuid)
2212{
2213 struct o2hb_region *p, *reg = NULL;
2214
2215 assert_spin_locked(&o2hb_live_lock);
2216
2217 list_for_each_entry(p, &o2hb_all_regions, hr_all_item) {
2218 if (!strcmp(region_uuid, config_item_name(&p->hr_item))) {
2219 reg = p;
2220 break;
2221 }
2222 }
2223
2224 return reg;
2225}
2226
2227static int o2hb_region_get(const char *region_uuid)
2228{
2229 int ret = 0;
2230 struct o2hb_region *reg;
2231
2232 spin_lock(&o2hb_live_lock);
2233
2234 reg = o2hb_find_region(region_uuid);
2235 if (!reg)
2236 ret = -ENOENT;
2237 spin_unlock(&o2hb_live_lock);
2238
Joel Becker16c6a4f2007-06-19 11:34:03 -07002239 if (ret)
2240 goto out;
Joel Becker14829422007-06-14 21:40:49 -07002241
Joel Becker16c6a4f2007-06-19 11:34:03 -07002242 ret = o2nm_depend_this_node();
2243 if (ret)
2244 goto out;
2245
2246 ret = o2nm_depend_item(&reg->hr_item);
2247 if (ret)
2248 o2nm_undepend_this_node();
2249
2250out:
Joel Becker14829422007-06-14 21:40:49 -07002251 return ret;
2252}
2253
2254static void o2hb_region_put(const char *region_uuid)
2255{
2256 struct o2hb_region *reg;
2257
2258 spin_lock(&o2hb_live_lock);
2259
2260 reg = o2hb_find_region(region_uuid);
2261
2262 spin_unlock(&o2hb_live_lock);
2263
Joel Becker16c6a4f2007-06-19 11:34:03 -07002264 if (reg) {
Joel Becker14829422007-06-14 21:40:49 -07002265 o2nm_undepend_item(&reg->hr_item);
Joel Becker16c6a4f2007-06-19 11:34:03 -07002266 o2nm_undepend_this_node();
2267 }
Joel Becker14829422007-06-14 21:40:49 -07002268}
2269
2270int o2hb_register_callback(const char *region_uuid,
2271 struct o2hb_callback_func *hc)
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08002272{
2273 struct o2hb_callback_func *tmp;
2274 struct list_head *iter;
2275 struct o2hb_callback *hbcall;
2276 int ret;
2277
2278 BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2279 BUG_ON(!list_empty(&hc->hc_item));
2280
2281 hbcall = hbcall_from_type(hc->hc_type);
2282 if (IS_ERR(hbcall)) {
2283 ret = PTR_ERR(hbcall);
2284 goto out;
2285 }
2286
Joel Becker14829422007-06-14 21:40:49 -07002287 if (region_uuid) {
2288 ret = o2hb_region_get(region_uuid);
2289 if (ret)
2290 goto out;
2291 }
2292
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08002293 down_write(&o2hb_callback_sem);
2294
2295 list_for_each(iter, &hbcall->list) {
2296 tmp = list_entry(iter, struct o2hb_callback_func, hc_item);
2297 if (hc->hc_priority < tmp->hc_priority) {
2298 list_add_tail(&hc->hc_item, iter);
2299 break;
2300 }
2301 }
2302 if (list_empty(&hc->hc_item))
2303 list_add_tail(&hc->hc_item, &hbcall->list);
2304
2305 up_write(&o2hb_callback_sem);
2306 ret = 0;
2307out:
2308 mlog(ML_HEARTBEAT, "returning %d on behalf of %p for funcs %p\n",
2309 ret, __builtin_return_address(0), hc);
2310 return ret;
2311}
2312EXPORT_SYMBOL_GPL(o2hb_register_callback);
2313
Joel Becker14829422007-06-14 21:40:49 -07002314void o2hb_unregister_callback(const char *region_uuid,
2315 struct o2hb_callback_func *hc)
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08002316{
2317 BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2318
2319 mlog(ML_HEARTBEAT, "on behalf of %p for funcs %p\n",
2320 __builtin_return_address(0), hc);
2321
Joel Becker14829422007-06-14 21:40:49 -07002322 /* XXX Can this happen _with_ a region reference? */
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08002323 if (list_empty(&hc->hc_item))
Joel Beckerc24f72c2007-02-03 03:14:30 -08002324 return;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08002325
Joel Becker14829422007-06-14 21:40:49 -07002326 if (region_uuid)
2327 o2hb_region_put(region_uuid);
2328
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08002329 down_write(&o2hb_callback_sem);
2330
2331 list_del_init(&hc->hc_item);
2332
2333 up_write(&o2hb_callback_sem);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08002334}
2335EXPORT_SYMBOL_GPL(o2hb_unregister_callback);
2336
2337int o2hb_check_node_heartbeating(u8 node_num)
2338{
2339 unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2340
2341 o2hb_fill_node_map(testing_map, sizeof(testing_map));
2342 if (!test_bit(node_num, testing_map)) {
2343 mlog(ML_HEARTBEAT,
2344 "node (%u) does not have heartbeating enabled.\n",
2345 node_num);
2346 return 0;
2347 }
2348
2349 return 1;
2350}
2351EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating);
2352
2353int o2hb_check_node_heartbeating_from_callback(u8 node_num)
2354{
2355 unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2356
2357 o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
2358 if (!test_bit(node_num, testing_map)) {
2359 mlog(ML_HEARTBEAT,
2360 "node (%u) does not have heartbeating enabled.\n",
2361 node_num);
2362 return 0;
2363 }
2364
2365 return 1;
2366}
2367EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback);
2368
2369/* Makes sure our local node is configured with a node number, and is
2370 * heartbeating. */
2371int o2hb_check_local_node_heartbeating(void)
2372{
2373 u8 node_num;
2374
2375 /* if this node was set then we have networking */
2376 node_num = o2nm_this_node();
2377 if (node_num == O2NM_MAX_NODES) {
2378 mlog(ML_HEARTBEAT, "this node has not been configured.\n");
2379 return 0;
2380 }
2381
2382 return o2hb_check_node_heartbeating(node_num);
2383}
2384EXPORT_SYMBOL_GPL(o2hb_check_local_node_heartbeating);
2385
2386/*
2387 * this is just a hack until we get the plumbing which flips file systems
2388 * read only and drops the hb ref instead of killing the node dead.
2389 */
2390void o2hb_stop_all_regions(void)
2391{
2392 struct o2hb_region *reg;
2393
2394 mlog(ML_ERROR, "stopping heartbeat on all active regions.\n");
2395
2396 spin_lock(&o2hb_live_lock);
2397
2398 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item)
2399 reg->hr_unclean_stop = 1;
2400
2401 spin_unlock(&o2hb_live_lock);
2402}
2403EXPORT_SYMBOL_GPL(o2hb_stop_all_regions);
Sunil Mushranb3c85c42010-10-07 14:31:06 -07002404
2405int o2hb_get_all_regions(char *region_uuids, u8 max_regions)
2406{
2407 struct o2hb_region *reg;
2408 int numregs = 0;
2409 char *p;
2410
2411 spin_lock(&o2hb_live_lock);
2412
2413 p = region_uuids;
2414 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2415 mlog(0, "Region: %s\n", config_item_name(&reg->hr_item));
2416 if (numregs < max_regions) {
2417 memcpy(p, config_item_name(&reg->hr_item),
2418 O2HB_MAX_REGION_NAME_LEN);
2419 p += O2HB_MAX_REGION_NAME_LEN;
2420 }
2421 numregs++;
2422 }
2423
2424 spin_unlock(&o2hb_live_lock);
2425
2426 return numregs;
2427}
2428EXPORT_SYMBOL_GPL(o2hb_get_all_regions);
2429
2430int o2hb_global_heartbeat_active(void)
2431{
Sunil Mushran4d94aa12010-10-09 10:27:04 -07002432 return (o2hb_heartbeat_mode == O2HB_HEARTBEAT_GLOBAL);
Sunil Mushranb3c85c42010-10-07 14:31:06 -07002433}
2434EXPORT_SYMBOL(o2hb_global_heartbeat_active);