blob: 41d5f1f92d56f60a45f6066d875598bda7caddf2 [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 Mushran87d3d3f2008-12-17 14:17:42 -080065#define O2HB_DEBUG_DIR "o2hb"
66#define O2HB_DEBUG_LIVENODES "livenodes"
67static struct dentry *o2hb_debug_dir;
68static struct dentry *o2hb_debug_livenodes;
69
Mark Fasheha7f6a5f2005-12-15 14:31:23 -080070static LIST_HEAD(o2hb_all_regions);
71
72static struct o2hb_callback {
73 struct list_head list;
74} o2hb_callbacks[O2HB_NUM_CB];
75
76static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type);
77
78#define O2HB_DEFAULT_BLOCK_BITS 9
79
80unsigned int o2hb_dead_threshold = O2HB_DEFAULT_DEAD_THRESHOLD;
81
Sunil Mushran2bd63212010-01-25 16:57:38 -080082/* Only sets a new threshold if there are no active regions.
Mark Fasheha7f6a5f2005-12-15 14:31:23 -080083 *
84 * No locking or otherwise interesting code is required for reading
85 * o2hb_dead_threshold as it can't change once regions are active and
86 * it's not interesting to anyone until then anyway. */
87static void o2hb_dead_threshold_set(unsigned int threshold)
88{
89 if (threshold > O2HB_MIN_DEAD_THRESHOLD) {
90 spin_lock(&o2hb_live_lock);
91 if (list_empty(&o2hb_all_regions))
92 o2hb_dead_threshold = threshold;
93 spin_unlock(&o2hb_live_lock);
94 }
95}
96
97struct o2hb_node_event {
98 struct list_head hn_item;
99 enum o2hb_callback_type hn_event_type;
100 struct o2nm_node *hn_node;
101 int hn_node_num;
102};
103
104struct o2hb_disk_slot {
105 struct o2hb_disk_heartbeat_block *ds_raw_block;
106 u8 ds_node_num;
107 u64 ds_last_time;
108 u64 ds_last_generation;
109 u16 ds_equal_samples;
110 u16 ds_changed_samples;
111 struct list_head ds_live_item;
112};
113
114/* each thread owns a region.. when we're asked to tear down the region
115 * we ask the thread to stop, who cleans up the region */
116struct o2hb_region {
117 struct config_item hr_item;
118
119 struct list_head hr_all_item;
120 unsigned hr_unclean_stop:1;
121
122 /* protected by the hr_callback_sem */
123 struct task_struct *hr_task;
124
125 unsigned int hr_blocks;
126 unsigned long long hr_start_block;
127
128 unsigned int hr_block_bits;
129 unsigned int hr_block_bytes;
130
131 unsigned int hr_slots_per_page;
132 unsigned int hr_num_pages;
133
134 struct page **hr_slot_data;
135 struct block_device *hr_bdev;
136 struct o2hb_disk_slot *hr_slots;
137
138 /* let the person setting up hb wait for it to return until it
139 * has reached a 'steady' state. This will be fixed when we have
140 * a more complete api that doesn't lead to this sort of fragility. */
141 atomic_t hr_steady_iterations;
142
143 char hr_dev_name[BDEVNAME_SIZE];
144
145 unsigned int hr_timeout_ms;
146
147 /* randomized as the region goes up and down so that a node
148 * recognizes a node going up and down in one iteration */
149 u64 hr_generation;
150
David Howellsc4028952006-11-22 14:57:56 +0000151 struct delayed_work hr_write_timeout_work;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800152 unsigned long hr_last_timeout_start;
153
154 /* Used during o2hb_check_slot to hold a copy of the block
155 * being checked because we temporarily have to zero out the
156 * crc field. */
157 struct o2hb_disk_heartbeat_block *hr_tmp_block;
158};
159
160struct o2hb_bio_wait_ctxt {
161 atomic_t wc_num_reqs;
162 struct completion wc_io_complete;
Mark Fasheha9e2ae32006-03-24 14:20:17 -0800163 int wc_error;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800164};
165
David Howellsc4028952006-11-22 14:57:56 +0000166static void o2hb_write_timeout(struct work_struct *work)
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800167{
David Howellsc4028952006-11-22 14:57:56 +0000168 struct o2hb_region *reg =
169 container_of(work, struct o2hb_region,
170 hr_write_timeout_work.work);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800171
172 mlog(ML_ERROR, "Heartbeat write timeout to device %s after %u "
173 "milliseconds\n", reg->hr_dev_name,
Sunil Mushran2bd63212010-01-25 16:57:38 -0800174 jiffies_to_msecs(jiffies - reg->hr_last_timeout_start));
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800175 o2quo_disk_timeout();
176}
177
178static void o2hb_arm_write_timeout(struct o2hb_region *reg)
179{
Tao Mab31d3082009-12-22 10:32:15 +0800180 mlog(ML_HEARTBEAT, "Queue write timeout for %u ms\n",
181 O2HB_MAX_WRITE_TIMEOUT_MS);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800182
183 cancel_delayed_work(&reg->hr_write_timeout_work);
184 reg->hr_last_timeout_start = jiffies;
185 schedule_delayed_work(&reg->hr_write_timeout_work,
186 msecs_to_jiffies(O2HB_MAX_WRITE_TIMEOUT_MS));
187}
188
189static void o2hb_disarm_write_timeout(struct o2hb_region *reg)
190{
191 cancel_delayed_work(&reg->hr_write_timeout_work);
192 flush_scheduled_work();
193}
194
Philipp Reisnerb5592922007-01-11 10:58:10 +0100195static inline void o2hb_bio_wait_init(struct o2hb_bio_wait_ctxt *wc)
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800196{
Philipp Reisnerb5592922007-01-11 10:58:10 +0100197 atomic_set(&wc->wc_num_reqs, 1);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800198 init_completion(&wc->wc_io_complete);
Mark Fasheha9e2ae32006-03-24 14:20:17 -0800199 wc->wc_error = 0;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800200}
201
202/* Used in error paths too */
203static inline void o2hb_bio_wait_dec(struct o2hb_bio_wait_ctxt *wc,
204 unsigned int num)
205{
206 /* sadly atomic_sub_and_test() isn't available on all platforms. The
207 * good news is that the fast path only completes one at a time */
208 while(num--) {
209 if (atomic_dec_and_test(&wc->wc_num_reqs)) {
210 BUG_ON(num > 0);
211 complete(&wc->wc_io_complete);
212 }
213 }
214}
215
216static void o2hb_wait_on_io(struct o2hb_region *reg,
217 struct o2hb_bio_wait_ctxt *wc)
218{
219 struct address_space *mapping = reg->hr_bdev->bd_inode->i_mapping;
220
221 blk_run_address_space(mapping);
Philipp Reisnerb5592922007-01-11 10:58:10 +0100222 o2hb_bio_wait_dec(wc, 1);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800223
224 wait_for_completion(&wc->wc_io_complete);
225}
226
Al Viro782e3b32007-10-12 07:17:47 +0100227static void o2hb_bio_end_io(struct bio *bio,
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800228 int error)
229{
230 struct o2hb_bio_wait_ctxt *wc = bio->bi_private;
231
Mark Fasheha9e2ae32006-03-24 14:20:17 -0800232 if (error) {
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800233 mlog(ML_ERROR, "IO Error %d\n", error);
Mark Fasheha9e2ae32006-03-24 14:20:17 -0800234 wc->wc_error = error;
235 }
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800236
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800237 o2hb_bio_wait_dec(wc, 1);
Philipp Reisnerb5592922007-01-11 10:58:10 +0100238 bio_put(bio);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800239}
240
241/* Setup a Bio to cover I/O against num_slots slots starting at
242 * start_slot. */
243static struct bio *o2hb_setup_one_bio(struct o2hb_region *reg,
244 struct o2hb_bio_wait_ctxt *wc,
Philipp Reisnerb5592922007-01-11 10:58:10 +0100245 unsigned int *current_slot,
246 unsigned int max_slots)
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800247{
Philipp Reisnerb5592922007-01-11 10:58:10 +0100248 int len, current_page;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800249 unsigned int vec_len, vec_start;
250 unsigned int bits = reg->hr_block_bits;
251 unsigned int spp = reg->hr_slots_per_page;
Philipp Reisnerb5592922007-01-11 10:58:10 +0100252 unsigned int cs = *current_slot;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800253 struct bio *bio;
254 struct page *page;
255
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800256 /* Testing has shown this allocation to take long enough under
257 * GFP_KERNEL that the local node can get fenced. It would be
258 * nicest if we could pre-allocate these bios and avoid this
259 * all together. */
Philipp Reisnerb5592922007-01-11 10:58:10 +0100260 bio = bio_alloc(GFP_ATOMIC, 16);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800261 if (!bio) {
262 mlog(ML_ERROR, "Could not alloc slots BIO!\n");
263 bio = ERR_PTR(-ENOMEM);
264 goto bail;
265 }
266
267 /* Must put everything in 512 byte sectors for the bio... */
Philipp Reisnerb5592922007-01-11 10:58:10 +0100268 bio->bi_sector = (reg->hr_start_block + cs) << (bits - 9);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800269 bio->bi_bdev = reg->hr_bdev;
270 bio->bi_private = wc;
271 bio->bi_end_io = o2hb_bio_end_io;
272
Philipp Reisnerb5592922007-01-11 10:58:10 +0100273 vec_start = (cs << bits) % PAGE_CACHE_SIZE;
274 while(cs < max_slots) {
275 current_page = cs / spp;
276 page = reg->hr_slot_data[current_page];
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800277
Jan Karabc7e97c2007-10-10 16:25:42 +0200278 vec_len = min(PAGE_CACHE_SIZE - vec_start,
Philipp Reisnerb5592922007-01-11 10:58:10 +0100279 (max_slots-cs) * (PAGE_CACHE_SIZE/spp) );
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800280
281 mlog(ML_HB_BIO, "page %d, vec_len = %u, vec_start = %u\n",
Philipp Reisnerb5592922007-01-11 10:58:10 +0100282 current_page, vec_len, vec_start);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800283
284 len = bio_add_page(bio, page, vec_len, vec_start);
Philipp Reisnerb5592922007-01-11 10:58:10 +0100285 if (len != vec_len) break;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800286
Philipp Reisnerb5592922007-01-11 10:58:10 +0100287 cs += vec_len / (PAGE_CACHE_SIZE/spp);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800288 vec_start = 0;
289 }
290
291bail:
Philipp Reisnerb5592922007-01-11 10:58:10 +0100292 *current_slot = cs;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800293 return bio;
294}
295
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800296static int o2hb_read_slots(struct o2hb_region *reg,
297 unsigned int max_slots)
298{
Philipp Reisnerb5592922007-01-11 10:58:10 +0100299 unsigned int current_slot=0;
300 int status;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800301 struct o2hb_bio_wait_ctxt wc;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800302 struct bio *bio;
303
Philipp Reisnerb5592922007-01-11 10:58:10 +0100304 o2hb_bio_wait_init(&wc);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800305
Philipp Reisnerb5592922007-01-11 10:58:10 +0100306 while(current_slot < max_slots) {
307 bio = o2hb_setup_one_bio(reg, &wc, &current_slot, max_slots);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800308 if (IS_ERR(bio)) {
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800309 status = PTR_ERR(bio);
310 mlog_errno(status);
311 goto bail_and_wait;
312 }
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800313
Philipp Reisnerb5592922007-01-11 10:58:10 +0100314 atomic_inc(&wc.wc_num_reqs);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800315 submit_bio(READ, bio);
316 }
317
318 status = 0;
319
320bail_and_wait:
321 o2hb_wait_on_io(reg, &wc);
Mark Fasheha9e2ae32006-03-24 14:20:17 -0800322 if (wc.wc_error && !status)
323 status = wc.wc_error;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800324
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800325 return status;
326}
327
328static int o2hb_issue_node_write(struct o2hb_region *reg,
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800329 struct o2hb_bio_wait_ctxt *write_wc)
330{
331 int status;
332 unsigned int slot;
333 struct bio *bio;
334
Philipp Reisnerb5592922007-01-11 10:58:10 +0100335 o2hb_bio_wait_init(write_wc);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800336
337 slot = o2nm_this_node();
338
Philipp Reisnerb5592922007-01-11 10:58:10 +0100339 bio = o2hb_setup_one_bio(reg, write_wc, &slot, slot+1);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800340 if (IS_ERR(bio)) {
341 status = PTR_ERR(bio);
342 mlog_errno(status);
343 goto bail;
344 }
345
Philipp Reisnerb5592922007-01-11 10:58:10 +0100346 atomic_inc(&write_wc->wc_num_reqs);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800347 submit_bio(WRITE, bio);
348
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800349 status = 0;
350bail:
351 return status;
352}
353
354static u32 o2hb_compute_block_crc_le(struct o2hb_region *reg,
355 struct o2hb_disk_heartbeat_block *hb_block)
356{
357 __le32 old_cksum;
358 u32 ret;
359
360 /* We want to compute the block crc with a 0 value in the
361 * hb_cksum field. Save it off here and replace after the
362 * crc. */
363 old_cksum = hb_block->hb_cksum;
364 hb_block->hb_cksum = 0;
365
366 ret = crc32_le(0, (unsigned char *) hb_block, reg->hr_block_bytes);
367
368 hb_block->hb_cksum = old_cksum;
369
370 return ret;
371}
372
373static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block *hb_block)
374{
Mark Fasheh70bacbd2006-03-02 11:10:05 -0800375 mlog(ML_ERROR, "Dump slot information: seq = 0x%llx, node = %u, "
376 "cksum = 0x%x, generation 0x%llx\n",
377 (long long)le64_to_cpu(hb_block->hb_seq),
378 hb_block->hb_node, le32_to_cpu(hb_block->hb_cksum),
379 (long long)le64_to_cpu(hb_block->hb_generation));
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800380}
381
382static int o2hb_verify_crc(struct o2hb_region *reg,
383 struct o2hb_disk_heartbeat_block *hb_block)
384{
385 u32 read, computed;
386
387 read = le32_to_cpu(hb_block->hb_cksum);
388 computed = o2hb_compute_block_crc_le(reg, hb_block);
389
390 return read == computed;
391}
392
393/* We want to make sure that nobody is heartbeating on top of us --
394 * this will help detect an invalid configuration. */
395static int o2hb_check_last_timestamp(struct o2hb_region *reg)
396{
397 int node_num, ret;
398 struct o2hb_disk_slot *slot;
399 struct o2hb_disk_heartbeat_block *hb_block;
400
401 node_num = o2nm_this_node();
402
403 ret = 1;
404 slot = &reg->hr_slots[node_num];
405 /* Don't check on our 1st timestamp */
406 if (slot->ds_last_time) {
407 hb_block = slot->ds_raw_block;
408
409 if (le64_to_cpu(hb_block->hb_seq) != slot->ds_last_time)
410 ret = 0;
411 }
412
413 return ret;
414}
415
416static inline void o2hb_prepare_block(struct o2hb_region *reg,
417 u64 generation)
418{
419 int node_num;
420 u64 cputime;
421 struct o2hb_disk_slot *slot;
422 struct o2hb_disk_heartbeat_block *hb_block;
423
424 node_num = o2nm_this_node();
425 slot = &reg->hr_slots[node_num];
426
427 hb_block = (struct o2hb_disk_heartbeat_block *)slot->ds_raw_block;
428 memset(hb_block, 0, reg->hr_block_bytes);
429 /* TODO: time stuff */
430 cputime = CURRENT_TIME.tv_sec;
431 if (!cputime)
432 cputime = 1;
433
434 hb_block->hb_seq = cpu_to_le64(cputime);
435 hb_block->hb_node = node_num;
436 hb_block->hb_generation = cpu_to_le64(generation);
Mark Fasheh0db638f2006-05-09 15:09:35 -0700437 hb_block->hb_dead_ms = cpu_to_le32(o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800438
439 /* This step must always happen last! */
440 hb_block->hb_cksum = cpu_to_le32(o2hb_compute_block_crc_le(reg,
441 hb_block));
442
Mark Fasheh70bacbd2006-03-02 11:10:05 -0800443 mlog(ML_HB_BIO, "our node generation = 0x%llx, cksum = 0x%x\n",
Mark Fasheh5fdf1e62007-04-27 16:50:03 -0700444 (long long)generation,
Mark Fasheh70bacbd2006-03-02 11:10:05 -0800445 le32_to_cpu(hb_block->hb_cksum));
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800446}
447
448static void o2hb_fire_callbacks(struct o2hb_callback *hbcall,
449 struct o2nm_node *node,
450 int idx)
451{
452 struct list_head *iter;
453 struct o2hb_callback_func *f;
454
455 list_for_each(iter, &hbcall->list) {
456 f = list_entry(iter, struct o2hb_callback_func, hc_item);
457 mlog(ML_HEARTBEAT, "calling funcs %p\n", f);
458 (f->hc_func)(node, idx, f->hc_data);
459 }
460}
461
462/* Will run the list in order until we process the passed event */
463static void o2hb_run_event_list(struct o2hb_node_event *queued_event)
464{
465 int empty;
466 struct o2hb_callback *hbcall;
467 struct o2hb_node_event *event;
468
469 spin_lock(&o2hb_live_lock);
470 empty = list_empty(&queued_event->hn_item);
471 spin_unlock(&o2hb_live_lock);
472 if (empty)
473 return;
474
475 /* Holding callback sem assures we don't alter the callback
476 * lists when doing this, and serializes ourselves with other
477 * processes wanting callbacks. */
478 down_write(&o2hb_callback_sem);
479
480 spin_lock(&o2hb_live_lock);
481 while (!list_empty(&o2hb_node_events)
482 && !list_empty(&queued_event->hn_item)) {
483 event = list_entry(o2hb_node_events.next,
484 struct o2hb_node_event,
485 hn_item);
486 list_del_init(&event->hn_item);
487 spin_unlock(&o2hb_live_lock);
488
489 mlog(ML_HEARTBEAT, "Node %s event for %d\n",
490 event->hn_event_type == O2HB_NODE_UP_CB ? "UP" : "DOWN",
491 event->hn_node_num);
492
493 hbcall = hbcall_from_type(event->hn_event_type);
494
495 /* We should *never* have gotten on to the list with a
496 * bad type... This isn't something that we should try
497 * to recover from. */
498 BUG_ON(IS_ERR(hbcall));
499
500 o2hb_fire_callbacks(hbcall, event->hn_node, event->hn_node_num);
501
502 spin_lock(&o2hb_live_lock);
503 }
504 spin_unlock(&o2hb_live_lock);
505
506 up_write(&o2hb_callback_sem);
507}
508
509static void o2hb_queue_node_event(struct o2hb_node_event *event,
510 enum o2hb_callback_type type,
511 struct o2nm_node *node,
512 int node_num)
513{
514 assert_spin_locked(&o2hb_live_lock);
515
516 event->hn_event_type = type;
517 event->hn_node = node;
518 event->hn_node_num = node_num;
519
520 mlog(ML_HEARTBEAT, "Queue node %s event for node %d\n",
521 type == O2HB_NODE_UP_CB ? "UP" : "DOWN", node_num);
522
523 list_add_tail(&event->hn_item, &o2hb_node_events);
524}
525
526static void o2hb_shutdown_slot(struct o2hb_disk_slot *slot)
527{
528 struct o2hb_node_event event =
529 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
530 struct o2nm_node *node;
531
532 node = o2nm_get_node_by_num(slot->ds_node_num);
533 if (!node)
534 return;
535
536 spin_lock(&o2hb_live_lock);
537 if (!list_empty(&slot->ds_live_item)) {
538 mlog(ML_HEARTBEAT, "Shutdown, node %d leaves region\n",
539 slot->ds_node_num);
540
541 list_del_init(&slot->ds_live_item);
542
543 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
544 clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
545
546 o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node,
547 slot->ds_node_num);
548 }
549 }
550 spin_unlock(&o2hb_live_lock);
551
552 o2hb_run_event_list(&event);
553
554 o2nm_node_put(node);
555}
556
557static int o2hb_check_slot(struct o2hb_region *reg,
558 struct o2hb_disk_slot *slot)
559{
560 int changed = 0, gen_changed = 0;
561 struct o2hb_node_event event =
562 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
563 struct o2nm_node *node;
564 struct o2hb_disk_heartbeat_block *hb_block = reg->hr_tmp_block;
565 u64 cputime;
Mark Fasheh0db638f2006-05-09 15:09:35 -0700566 unsigned int dead_ms = o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS;
567 unsigned int slot_dead_ms;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800568
569 memcpy(hb_block, slot->ds_raw_block, reg->hr_block_bytes);
570
571 /* Is this correct? Do we assume that the node doesn't exist
572 * if we're not configured for him? */
573 node = o2nm_get_node_by_num(slot->ds_node_num);
574 if (!node)
575 return 0;
576
577 if (!o2hb_verify_crc(reg, hb_block)) {
578 /* all paths from here will drop o2hb_live_lock for
579 * us. */
580 spin_lock(&o2hb_live_lock);
581
582 /* Don't print an error on the console in this case -
583 * a freshly formatted heartbeat area will not have a
584 * crc set on it. */
585 if (list_empty(&slot->ds_live_item))
586 goto out;
587
588 /* The node is live but pushed out a bad crc. We
589 * consider it a transient miss but don't populate any
590 * other values as they may be junk. */
591 mlog(ML_ERROR, "Node %d has written a bad crc to %s\n",
592 slot->ds_node_num, reg->hr_dev_name);
593 o2hb_dump_slot(hb_block);
594
595 slot->ds_equal_samples++;
596 goto fire_callbacks;
597 }
598
599 /* we don't care if these wrap.. the state transitions below
600 * clear at the right places */
601 cputime = le64_to_cpu(hb_block->hb_seq);
602 if (slot->ds_last_time != cputime)
603 slot->ds_changed_samples++;
604 else
605 slot->ds_equal_samples++;
606 slot->ds_last_time = cputime;
607
608 /* The node changed heartbeat generations. We assume this to
609 * mean it dropped off but came back before we timed out. We
610 * want to consider it down for the time being but don't want
611 * to lose any changed_samples state we might build up to
612 * considering it live again. */
613 if (slot->ds_last_generation != le64_to_cpu(hb_block->hb_generation)) {
614 gen_changed = 1;
615 slot->ds_equal_samples = 0;
Mark Fasheh70bacbd2006-03-02 11:10:05 -0800616 mlog(ML_HEARTBEAT, "Node %d changed generation (0x%llx "
617 "to 0x%llx)\n", slot->ds_node_num,
618 (long long)slot->ds_last_generation,
619 (long long)le64_to_cpu(hb_block->hb_generation));
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800620 }
621
622 slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
623
Mark Fasheh70bacbd2006-03-02 11:10:05 -0800624 mlog(ML_HEARTBEAT, "Slot %d gen 0x%llx cksum 0x%x "
625 "seq %llu last %llu changed %u equal %u\n",
626 slot->ds_node_num, (long long)slot->ds_last_generation,
627 le32_to_cpu(hb_block->hb_cksum),
Sunil Mushran2bd63212010-01-25 16:57:38 -0800628 (unsigned long long)le64_to_cpu(hb_block->hb_seq),
Mark Fasheh70bacbd2006-03-02 11:10:05 -0800629 (unsigned long long)slot->ds_last_time, slot->ds_changed_samples,
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800630 slot->ds_equal_samples);
631
632 spin_lock(&o2hb_live_lock);
633
634fire_callbacks:
635 /* dead nodes only come to life after some number of
636 * changes at any time during their dead time */
637 if (list_empty(&slot->ds_live_item) &&
638 slot->ds_changed_samples >= O2HB_LIVE_THRESHOLD) {
Mark Fasheh70bacbd2006-03-02 11:10:05 -0800639 mlog(ML_HEARTBEAT, "Node %d (id 0x%llx) joined my region\n",
640 slot->ds_node_num, (long long)slot->ds_last_generation);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800641
642 /* first on the list generates a callback */
643 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
644 set_bit(slot->ds_node_num, o2hb_live_node_bitmap);
645
646 o2hb_queue_node_event(&event, O2HB_NODE_UP_CB, node,
647 slot->ds_node_num);
648
649 changed = 1;
650 }
651
652 list_add_tail(&slot->ds_live_item,
653 &o2hb_live_slots[slot->ds_node_num]);
654
655 slot->ds_equal_samples = 0;
Mark Fasheh0db638f2006-05-09 15:09:35 -0700656
657 /* We want to be sure that all nodes agree on the
658 * number of milliseconds before a node will be
659 * considered dead. The self-fencing timeout is
660 * computed from this value, and a discrepancy might
661 * result in heartbeat calling a node dead when it
662 * hasn't self-fenced yet. */
663 slot_dead_ms = le32_to_cpu(hb_block->hb_dead_ms);
664 if (slot_dead_ms && slot_dead_ms != dead_ms) {
665 /* TODO: Perhaps we can fail the region here. */
666 mlog(ML_ERROR, "Node %d on device %s has a dead count "
667 "of %u ms, but our count is %u ms.\n"
668 "Please double check your configuration values "
669 "for 'O2CB_HEARTBEAT_THRESHOLD'\n",
670 slot->ds_node_num, reg->hr_dev_name, slot_dead_ms,
671 dead_ms);
672 }
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800673 goto out;
674 }
675
676 /* if the list is dead, we're done.. */
677 if (list_empty(&slot->ds_live_item))
678 goto out;
679
680 /* live nodes only go dead after enough consequtive missed
681 * samples.. reset the missed counter whenever we see
682 * activity */
683 if (slot->ds_equal_samples >= o2hb_dead_threshold || gen_changed) {
684 mlog(ML_HEARTBEAT, "Node %d left my region\n",
685 slot->ds_node_num);
686
687 /* last off the live_slot generates a callback */
688 list_del_init(&slot->ds_live_item);
689 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
690 clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
691
692 o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node,
693 slot->ds_node_num);
694
695 changed = 1;
696 }
697
698 /* We don't clear this because the node is still
699 * actually writing new blocks. */
700 if (!gen_changed)
701 slot->ds_changed_samples = 0;
702 goto out;
703 }
704 if (slot->ds_changed_samples) {
705 slot->ds_changed_samples = 0;
706 slot->ds_equal_samples = 0;
707 }
708out:
709 spin_unlock(&o2hb_live_lock);
710
711 o2hb_run_event_list(&event);
712
713 o2nm_node_put(node);
714 return changed;
715}
716
717/* This could be faster if we just implmented a find_last_bit, but I
718 * don't think the circumstances warrant it. */
719static int o2hb_highest_node(unsigned long *nodes,
720 int numbits)
721{
722 int highest, node;
723
724 highest = numbits;
725 node = -1;
726 while ((node = find_next_bit(nodes, numbits, node + 1)) != -1) {
727 if (node >= numbits)
728 break;
729
730 highest = node;
731 }
732
733 return highest;
734}
735
Mark Fasheha9e2ae32006-03-24 14:20:17 -0800736static int o2hb_do_disk_heartbeat(struct o2hb_region *reg)
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800737{
738 int i, ret, highest_node, change = 0;
739 unsigned long configured_nodes[BITS_TO_LONGS(O2NM_MAX_NODES)];
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800740 struct o2hb_bio_wait_ctxt write_wc;
741
Mark Fasheha9e2ae32006-03-24 14:20:17 -0800742 ret = o2nm_configured_node_map(configured_nodes,
743 sizeof(configured_nodes));
744 if (ret) {
745 mlog_errno(ret);
746 return ret;
747 }
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800748
749 highest_node = o2hb_highest_node(configured_nodes, O2NM_MAX_NODES);
750 if (highest_node >= O2NM_MAX_NODES) {
751 mlog(ML_NOTICE, "ocfs2_heartbeat: no configured nodes found!\n");
Mark Fasheha9e2ae32006-03-24 14:20:17 -0800752 return -EINVAL;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800753 }
754
755 /* No sense in reading the slots of nodes that don't exist
756 * yet. Of course, if the node definitions have holes in them
757 * then we're reading an empty slot anyway... Consider this
758 * best-effort. */
759 ret = o2hb_read_slots(reg, highest_node + 1);
760 if (ret < 0) {
761 mlog_errno(ret);
Mark Fasheha9e2ae32006-03-24 14:20:17 -0800762 return ret;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800763 }
764
765 /* With an up to date view of the slots, we can check that no
766 * other node has been improperly configured to heartbeat in
767 * our slot. */
768 if (!o2hb_check_last_timestamp(reg))
769 mlog(ML_ERROR, "Device \"%s\": another node is heartbeating "
770 "in our slot!\n", reg->hr_dev_name);
771
772 /* fill in the proper info for our next heartbeat */
773 o2hb_prepare_block(reg, reg->hr_generation);
774
775 /* And fire off the write. Note that we don't wait on this I/O
776 * until later. */
Philipp Reisnerb5592922007-01-11 10:58:10 +0100777 ret = o2hb_issue_node_write(reg, &write_wc);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800778 if (ret < 0) {
779 mlog_errno(ret);
Mark Fasheha9e2ae32006-03-24 14:20:17 -0800780 return ret;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800781 }
782
783 i = -1;
784 while((i = find_next_bit(configured_nodes, O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
785
786 change |= o2hb_check_slot(reg, &reg->hr_slots[i]);
787 }
788
789 /*
790 * We have to be sure we've advertised ourselves on disk
791 * before we can go to steady state. This ensures that
792 * people we find in our steady state have seen us.
793 */
794 o2hb_wait_on_io(reg, &write_wc);
Mark Fasheha9e2ae32006-03-24 14:20:17 -0800795 if (write_wc.wc_error) {
796 /* Do not re-arm the write timeout on I/O error - we
797 * can't be sure that the new block ever made it to
798 * disk */
799 mlog(ML_ERROR, "Write error %d on device \"%s\"\n",
800 write_wc.wc_error, reg->hr_dev_name);
801 return write_wc.wc_error;
802 }
803
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800804 o2hb_arm_write_timeout(reg);
805
806 /* let the person who launched us know when things are steady */
807 if (!change && (atomic_read(&reg->hr_steady_iterations) != 0)) {
808 if (atomic_dec_and_test(&reg->hr_steady_iterations))
809 wake_up(&o2hb_steady_queue);
810 }
Mark Fasheha9e2ae32006-03-24 14:20:17 -0800811
812 return 0;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800813}
814
815/* Subtract b from a, storing the result in a. a *must* have a larger
816 * value than b. */
817static void o2hb_tv_subtract(struct timeval *a,
818 struct timeval *b)
819{
820 /* just return 0 when a is after b */
821 if (a->tv_sec < b->tv_sec ||
822 (a->tv_sec == b->tv_sec && a->tv_usec < b->tv_usec)) {
823 a->tv_sec = 0;
824 a->tv_usec = 0;
825 return;
826 }
827
828 a->tv_sec -= b->tv_sec;
829 a->tv_usec -= b->tv_usec;
830 while ( a->tv_usec < 0 ) {
831 a->tv_sec--;
832 a->tv_usec += 1000000;
833 }
834}
835
836static unsigned int o2hb_elapsed_msecs(struct timeval *start,
837 struct timeval *end)
838{
839 struct timeval res = *end;
840
841 o2hb_tv_subtract(&res, start);
842
843 return res.tv_sec * 1000 + res.tv_usec / 1000;
844}
845
846/*
847 * we ride the region ref that the region dir holds. before the region
848 * dir is removed and drops it ref it will wait to tear down this
849 * thread.
850 */
851static int o2hb_thread(void *data)
852{
853 int i, ret;
854 struct o2hb_region *reg = data;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800855 struct o2hb_bio_wait_ctxt write_wc;
856 struct timeval before_hb, after_hb;
857 unsigned int elapsed_msec;
858
859 mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread running\n");
860
861 set_user_nice(current, -20);
862
863 while (!kthread_should_stop() && !reg->hr_unclean_stop) {
864 /* We track the time spent inside
Frederik Schwarzer025dfda2008-10-16 19:02:37 +0200865 * o2hb_do_disk_heartbeat so that we avoid more than
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800866 * hr_timeout_ms between disk writes. On busy systems
867 * this should result in a heartbeat which is less
868 * likely to time itself out. */
869 do_gettimeofday(&before_hb);
870
Mark Fasheha9e2ae32006-03-24 14:20:17 -0800871 i = 0;
872 do {
873 ret = o2hb_do_disk_heartbeat(reg);
874 } while (ret && ++i < 2);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800875
876 do_gettimeofday(&after_hb);
877 elapsed_msec = o2hb_elapsed_msecs(&before_hb, &after_hb);
878
Tao Mab31d3082009-12-22 10:32:15 +0800879 mlog(ML_HEARTBEAT,
880 "start = %lu.%lu, end = %lu.%lu, msec = %u\n",
Mark Fasheh215c7f92006-02-01 16:42:10 -0800881 before_hb.tv_sec, (unsigned long) before_hb.tv_usec,
882 after_hb.tv_sec, (unsigned long) after_hb.tv_usec,
883 elapsed_msec);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800884
885 if (elapsed_msec < reg->hr_timeout_ms) {
886 /* the kthread api has blocked signals for us so no
887 * need to record the return value. */
888 msleep_interruptible(reg->hr_timeout_ms - elapsed_msec);
889 }
890 }
891
892 o2hb_disarm_write_timeout(reg);
893
894 /* unclean stop is only used in very bad situation */
895 for(i = 0; !reg->hr_unclean_stop && i < reg->hr_blocks; i++)
896 o2hb_shutdown_slot(&reg->hr_slots[i]);
897
898 /* Explicit down notification - avoid forcing the other nodes
899 * to timeout on this region when we could just as easily
900 * write a clear generation - thus indicating to them that
901 * this node has left this region.
902 *
903 * XXX: Should we skip this on unclean_stop? */
904 o2hb_prepare_block(reg, 0);
Philipp Reisnerb5592922007-01-11 10:58:10 +0100905 ret = o2hb_issue_node_write(reg, &write_wc);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800906 if (ret == 0) {
907 o2hb_wait_on_io(reg, &write_wc);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800908 } else {
909 mlog_errno(ret);
910 }
911
912 mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread exiting\n");
913
914 return 0;
915}
916
Sunil Mushran87d3d3f2008-12-17 14:17:42 -0800917#ifdef CONFIG_DEBUG_FS
918static int o2hb_debug_open(struct inode *inode, struct file *file)
919{
920 unsigned long map[BITS_TO_LONGS(O2NM_MAX_NODES)];
921 char *buf = NULL;
922 int i = -1;
923 int out = 0;
924
925 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
926 if (!buf)
927 goto bail;
928
929 o2hb_fill_node_map(map, sizeof(map));
930
931 while ((i = find_next_bit(map, O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES)
932 out += snprintf(buf + out, PAGE_SIZE - out, "%d ", i);
933 out += snprintf(buf + out, PAGE_SIZE - out, "\n");
934
935 i_size_write(inode, out);
936
937 file->private_data = buf;
938
939 return 0;
940bail:
941 return -ENOMEM;
942}
943
944static int o2hb_debug_release(struct inode *inode, struct file *file)
945{
946 kfree(file->private_data);
947 return 0;
948}
949
950static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
951 size_t nbytes, loff_t *ppos)
952{
953 return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
954 i_size_read(file->f_mapping->host));
955}
956#else
957static int o2hb_debug_open(struct inode *inode, struct file *file)
958{
959 return 0;
960}
961static int o2hb_debug_release(struct inode *inode, struct file *file)
962{
963 return 0;
964}
965static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
966 size_t nbytes, loff_t *ppos)
967{
968 return 0;
969}
970#endif /* CONFIG_DEBUG_FS */
971
Alexey Dobriyan828c0952009-10-01 15:43:56 -0700972static const struct file_operations o2hb_debug_fops = {
Sunil Mushran87d3d3f2008-12-17 14:17:42 -0800973 .open = o2hb_debug_open,
974 .release = o2hb_debug_release,
975 .read = o2hb_debug_read,
976 .llseek = generic_file_llseek,
977};
978
979void o2hb_exit(void)
980{
981 if (o2hb_debug_livenodes)
982 debugfs_remove(o2hb_debug_livenodes);
983 if (o2hb_debug_dir)
984 debugfs_remove(o2hb_debug_dir);
985}
986
987int o2hb_init(void)
Mark Fasheha7f6a5f2005-12-15 14:31:23 -0800988{
989 int i;
990
991 for (i = 0; i < ARRAY_SIZE(o2hb_callbacks); i++)
992 INIT_LIST_HEAD(&o2hb_callbacks[i].list);
993
994 for (i = 0; i < ARRAY_SIZE(o2hb_live_slots); i++)
995 INIT_LIST_HEAD(&o2hb_live_slots[i]);
996
997 INIT_LIST_HEAD(&o2hb_node_events);
998
999 memset(o2hb_live_node_bitmap, 0, sizeof(o2hb_live_node_bitmap));
Sunil Mushran87d3d3f2008-12-17 14:17:42 -08001000
1001 o2hb_debug_dir = debugfs_create_dir(O2HB_DEBUG_DIR, NULL);
1002 if (!o2hb_debug_dir) {
1003 mlog_errno(-ENOMEM);
1004 return -ENOMEM;
1005 }
1006
1007 o2hb_debug_livenodes = debugfs_create_file(O2HB_DEBUG_LIVENODES,
1008 S_IFREG|S_IRUSR,
1009 o2hb_debug_dir, NULL,
1010 &o2hb_debug_fops);
1011 if (!o2hb_debug_livenodes) {
1012 mlog_errno(-ENOMEM);
1013 debugfs_remove(o2hb_debug_dir);
1014 return -ENOMEM;
1015 }
1016
1017 return 0;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001018}
1019
1020/* if we're already in a callback then we're already serialized by the sem */
1021static void o2hb_fill_node_map_from_callback(unsigned long *map,
1022 unsigned bytes)
1023{
1024 BUG_ON(bytes < (BITS_TO_LONGS(O2NM_MAX_NODES) * sizeof(unsigned long)));
1025
1026 memcpy(map, &o2hb_live_node_bitmap, bytes);
1027}
1028
1029/*
1030 * get a map of all nodes that are heartbeating in any regions
1031 */
1032void o2hb_fill_node_map(unsigned long *map, unsigned bytes)
1033{
1034 /* callers want to serialize this map and callbacks so that they
1035 * can trust that they don't miss nodes coming to the party */
1036 down_read(&o2hb_callback_sem);
1037 spin_lock(&o2hb_live_lock);
1038 o2hb_fill_node_map_from_callback(map, bytes);
1039 spin_unlock(&o2hb_live_lock);
1040 up_read(&o2hb_callback_sem);
1041}
1042EXPORT_SYMBOL_GPL(o2hb_fill_node_map);
1043
1044/*
1045 * heartbeat configfs bits. The heartbeat set is a default set under
1046 * the cluster set in nodemanager.c.
1047 */
1048
1049static struct o2hb_region *to_o2hb_region(struct config_item *item)
1050{
1051 return item ? container_of(item, struct o2hb_region, hr_item) : NULL;
1052}
1053
1054/* drop_item only drops its ref after killing the thread, nothing should
1055 * be using the region anymore. this has to clean up any state that
1056 * attributes might have built up. */
1057static void o2hb_region_release(struct config_item *item)
1058{
1059 int i;
1060 struct page *page;
1061 struct o2hb_region *reg = to_o2hb_region(item);
1062
1063 if (reg->hr_tmp_block)
1064 kfree(reg->hr_tmp_block);
1065
1066 if (reg->hr_slot_data) {
1067 for (i = 0; i < reg->hr_num_pages; i++) {
1068 page = reg->hr_slot_data[i];
1069 if (page)
1070 __free_page(page);
1071 }
1072 kfree(reg->hr_slot_data);
1073 }
1074
1075 if (reg->hr_bdev)
Al Viro9a1c3542008-02-22 20:40:24 -05001076 blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001077
1078 if (reg->hr_slots)
1079 kfree(reg->hr_slots);
1080
1081 spin_lock(&o2hb_live_lock);
1082 list_del(&reg->hr_all_item);
1083 spin_unlock(&o2hb_live_lock);
1084
1085 kfree(reg);
1086}
1087
1088static int o2hb_read_block_input(struct o2hb_region *reg,
1089 const char *page,
1090 size_t count,
1091 unsigned long *ret_bytes,
1092 unsigned int *ret_bits)
1093{
1094 unsigned long bytes;
1095 char *p = (char *)page;
1096
1097 bytes = simple_strtoul(p, &p, 0);
1098 if (!p || (*p && (*p != '\n')))
1099 return -EINVAL;
1100
1101 /* Heartbeat and fs min / max block sizes are the same. */
1102 if (bytes > 4096 || bytes < 512)
1103 return -ERANGE;
1104 if (hweight16(bytes) != 1)
1105 return -EINVAL;
1106
1107 if (ret_bytes)
1108 *ret_bytes = bytes;
1109 if (ret_bits)
1110 *ret_bits = ffs(bytes) - 1;
1111
1112 return 0;
1113}
1114
1115static ssize_t o2hb_region_block_bytes_read(struct o2hb_region *reg,
1116 char *page)
1117{
1118 return sprintf(page, "%u\n", reg->hr_block_bytes);
1119}
1120
1121static ssize_t o2hb_region_block_bytes_write(struct o2hb_region *reg,
1122 const char *page,
1123 size_t count)
1124{
1125 int status;
1126 unsigned long block_bytes;
1127 unsigned int block_bits;
1128
1129 if (reg->hr_bdev)
1130 return -EINVAL;
1131
1132 status = o2hb_read_block_input(reg, page, count,
1133 &block_bytes, &block_bits);
1134 if (status)
1135 return status;
1136
1137 reg->hr_block_bytes = (unsigned int)block_bytes;
1138 reg->hr_block_bits = block_bits;
1139
1140 return count;
1141}
1142
1143static ssize_t o2hb_region_start_block_read(struct o2hb_region *reg,
1144 char *page)
1145{
1146 return sprintf(page, "%llu\n", reg->hr_start_block);
1147}
1148
1149static ssize_t o2hb_region_start_block_write(struct o2hb_region *reg,
1150 const char *page,
1151 size_t count)
1152{
1153 unsigned long long tmp;
1154 char *p = (char *)page;
1155
1156 if (reg->hr_bdev)
1157 return -EINVAL;
1158
1159 tmp = simple_strtoull(p, &p, 0);
1160 if (!p || (*p && (*p != '\n')))
1161 return -EINVAL;
1162
1163 reg->hr_start_block = tmp;
1164
1165 return count;
1166}
1167
1168static ssize_t o2hb_region_blocks_read(struct o2hb_region *reg,
1169 char *page)
1170{
1171 return sprintf(page, "%d\n", reg->hr_blocks);
1172}
1173
1174static ssize_t o2hb_region_blocks_write(struct o2hb_region *reg,
1175 const char *page,
1176 size_t count)
1177{
1178 unsigned long tmp;
1179 char *p = (char *)page;
1180
1181 if (reg->hr_bdev)
1182 return -EINVAL;
1183
1184 tmp = simple_strtoul(p, &p, 0);
1185 if (!p || (*p && (*p != '\n')))
1186 return -EINVAL;
1187
1188 if (tmp > O2NM_MAX_NODES || tmp == 0)
1189 return -ERANGE;
1190
1191 reg->hr_blocks = (unsigned int)tmp;
1192
1193 return count;
1194}
1195
1196static ssize_t o2hb_region_dev_read(struct o2hb_region *reg,
1197 char *page)
1198{
1199 unsigned int ret = 0;
1200
1201 if (reg->hr_bdev)
1202 ret = sprintf(page, "%s\n", reg->hr_dev_name);
1203
1204 return ret;
1205}
1206
1207static void o2hb_init_region_params(struct o2hb_region *reg)
1208{
1209 reg->hr_slots_per_page = PAGE_CACHE_SIZE >> reg->hr_block_bits;
1210 reg->hr_timeout_ms = O2HB_REGION_TIMEOUT_MS;
1211
1212 mlog(ML_HEARTBEAT, "hr_start_block = %llu, hr_blocks = %u\n",
1213 reg->hr_start_block, reg->hr_blocks);
1214 mlog(ML_HEARTBEAT, "hr_block_bytes = %u, hr_block_bits = %u\n",
1215 reg->hr_block_bytes, reg->hr_block_bits);
1216 mlog(ML_HEARTBEAT, "hr_timeout_ms = %u\n", reg->hr_timeout_ms);
1217 mlog(ML_HEARTBEAT, "dead threshold = %u\n", o2hb_dead_threshold);
1218}
1219
1220static int o2hb_map_slot_data(struct o2hb_region *reg)
1221{
1222 int i, j;
1223 unsigned int last_slot;
1224 unsigned int spp = reg->hr_slots_per_page;
1225 struct page *page;
1226 char *raw;
1227 struct o2hb_disk_slot *slot;
1228
1229 reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL);
1230 if (reg->hr_tmp_block == NULL) {
1231 mlog_errno(-ENOMEM);
1232 return -ENOMEM;
1233 }
1234
1235 reg->hr_slots = kcalloc(reg->hr_blocks,
1236 sizeof(struct o2hb_disk_slot), GFP_KERNEL);
1237 if (reg->hr_slots == NULL) {
1238 mlog_errno(-ENOMEM);
1239 return -ENOMEM;
1240 }
1241
1242 for(i = 0; i < reg->hr_blocks; i++) {
1243 slot = &reg->hr_slots[i];
1244 slot->ds_node_num = i;
1245 INIT_LIST_HEAD(&slot->ds_live_item);
1246 slot->ds_raw_block = NULL;
1247 }
1248
1249 reg->hr_num_pages = (reg->hr_blocks + spp - 1) / spp;
1250 mlog(ML_HEARTBEAT, "Going to require %u pages to cover %u blocks "
1251 "at %u blocks per page\n",
1252 reg->hr_num_pages, reg->hr_blocks, spp);
1253
1254 reg->hr_slot_data = kcalloc(reg->hr_num_pages, sizeof(struct page *),
1255 GFP_KERNEL);
1256 if (!reg->hr_slot_data) {
1257 mlog_errno(-ENOMEM);
1258 return -ENOMEM;
1259 }
1260
1261 for(i = 0; i < reg->hr_num_pages; i++) {
1262 page = alloc_page(GFP_KERNEL);
1263 if (!page) {
1264 mlog_errno(-ENOMEM);
1265 return -ENOMEM;
1266 }
1267
1268 reg->hr_slot_data[i] = page;
1269
1270 last_slot = i * spp;
1271 raw = page_address(page);
1272 for (j = 0;
1273 (j < spp) && ((j + last_slot) < reg->hr_blocks);
1274 j++) {
1275 BUG_ON((j + last_slot) >= reg->hr_blocks);
1276
1277 slot = &reg->hr_slots[j + last_slot];
1278 slot->ds_raw_block =
1279 (struct o2hb_disk_heartbeat_block *) raw;
1280
1281 raw += reg->hr_block_bytes;
1282 }
1283 }
1284
1285 return 0;
1286}
1287
1288/* Read in all the slots available and populate the tracking
1289 * structures so that we can start with a baseline idea of what's
1290 * there. */
1291static int o2hb_populate_slot_data(struct o2hb_region *reg)
1292{
1293 int ret, i;
1294 struct o2hb_disk_slot *slot;
1295 struct o2hb_disk_heartbeat_block *hb_block;
1296
1297 mlog_entry_void();
1298
1299 ret = o2hb_read_slots(reg, reg->hr_blocks);
1300 if (ret) {
1301 mlog_errno(ret);
1302 goto out;
1303 }
1304
1305 /* We only want to get an idea of the values initially in each
1306 * slot, so we do no verification - o2hb_check_slot will
1307 * actually determine if each configured slot is valid and
1308 * whether any values have changed. */
1309 for(i = 0; i < reg->hr_blocks; i++) {
1310 slot = &reg->hr_slots[i];
1311 hb_block = (struct o2hb_disk_heartbeat_block *) slot->ds_raw_block;
1312
1313 /* Only fill the values that o2hb_check_slot uses to
1314 * determine changing slots */
1315 slot->ds_last_time = le64_to_cpu(hb_block->hb_seq);
1316 slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
1317 }
1318
1319out:
1320 mlog_exit(ret);
1321 return ret;
1322}
1323
1324/* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */
1325static ssize_t o2hb_region_dev_write(struct o2hb_region *reg,
1326 const char *page,
1327 size_t count)
1328{
Joel Beckere6c352d2007-02-03 03:04:20 -08001329 struct task_struct *hb_task;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001330 long fd;
1331 int sectsize;
1332 char *p = (char *)page;
1333 struct file *filp = NULL;
1334 struct inode *inode = NULL;
1335 ssize_t ret = -EINVAL;
1336
1337 if (reg->hr_bdev)
1338 goto out;
1339
1340 /* We can't heartbeat without having had our node number
1341 * configured yet. */
1342 if (o2nm_this_node() == O2NM_MAX_NODES)
1343 goto out;
1344
1345 fd = simple_strtol(p, &p, 0);
1346 if (!p || (*p && (*p != '\n')))
1347 goto out;
1348
1349 if (fd < 0 || fd >= INT_MAX)
1350 goto out;
1351
1352 filp = fget(fd);
1353 if (filp == NULL)
1354 goto out;
1355
1356 if (reg->hr_blocks == 0 || reg->hr_start_block == 0 ||
1357 reg->hr_block_bytes == 0)
1358 goto out;
1359
1360 inode = igrab(filp->f_mapping->host);
1361 if (inode == NULL)
1362 goto out;
1363
1364 if (!S_ISBLK(inode->i_mode))
1365 goto out;
1366
1367 reg->hr_bdev = I_BDEV(filp->f_mapping->host);
Al Viro572c4892007-10-08 13:24:05 -04001368 ret = blkdev_get(reg->hr_bdev, FMODE_WRITE | FMODE_READ);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001369 if (ret) {
1370 reg->hr_bdev = NULL;
1371 goto out;
1372 }
1373 inode = NULL;
1374
1375 bdevname(reg->hr_bdev, reg->hr_dev_name);
1376
Martin K. Petersene1defc42009-05-22 17:17:49 -04001377 sectsize = bdev_logical_block_size(reg->hr_bdev);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001378 if (sectsize != reg->hr_block_bytes) {
1379 mlog(ML_ERROR,
1380 "blocksize %u incorrect for device, expected %d",
1381 reg->hr_block_bytes, sectsize);
1382 ret = -EINVAL;
1383 goto out;
1384 }
1385
1386 o2hb_init_region_params(reg);
1387
1388 /* Generation of zero is invalid */
1389 do {
1390 get_random_bytes(&reg->hr_generation,
1391 sizeof(reg->hr_generation));
1392 } while (reg->hr_generation == 0);
1393
1394 ret = o2hb_map_slot_data(reg);
1395 if (ret) {
1396 mlog_errno(ret);
1397 goto out;
1398 }
1399
1400 ret = o2hb_populate_slot_data(reg);
1401 if (ret) {
1402 mlog_errno(ret);
1403 goto out;
1404 }
1405
David Howellsc4028952006-11-22 14:57:56 +00001406 INIT_DELAYED_WORK(&reg->hr_write_timeout_work, o2hb_write_timeout);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001407
1408 /*
1409 * A node is considered live after it has beat LIVE_THRESHOLD
1410 * times. We're not steady until we've given them a chance
1411 * _after_ our first read.
1412 */
1413 atomic_set(&reg->hr_steady_iterations, O2HB_LIVE_THRESHOLD + 1);
1414
Joel Beckere6c352d2007-02-03 03:04:20 -08001415 hb_task = kthread_run(o2hb_thread, reg, "o2hb-%s",
1416 reg->hr_item.ci_name);
1417 if (IS_ERR(hb_task)) {
1418 ret = PTR_ERR(hb_task);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001419 mlog_errno(ret);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001420 goto out;
1421 }
1422
Joel Beckere6c352d2007-02-03 03:04:20 -08001423 spin_lock(&o2hb_live_lock);
1424 reg->hr_task = hb_task;
1425 spin_unlock(&o2hb_live_lock);
1426
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001427 ret = wait_event_interruptible(o2hb_steady_queue,
1428 atomic_read(&reg->hr_steady_iterations) == 0);
1429 if (ret) {
Joel Beckere6df3a62007-02-06 15:45:39 -08001430 /* We got interrupted (hello ptrace!). Clean up */
Joel Beckere6c352d2007-02-03 03:04:20 -08001431 spin_lock(&o2hb_live_lock);
1432 hb_task = reg->hr_task;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001433 reg->hr_task = NULL;
Joel Beckere6c352d2007-02-03 03:04:20 -08001434 spin_unlock(&o2hb_live_lock);
1435
1436 if (hb_task)
1437 kthread_stop(hb_task);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001438 goto out;
1439 }
1440
Joel Beckere6df3a62007-02-06 15:45:39 -08001441 /* Ok, we were woken. Make sure it wasn't by drop_item() */
1442 spin_lock(&o2hb_live_lock);
1443 hb_task = reg->hr_task;
1444 spin_unlock(&o2hb_live_lock);
1445
1446 if (hb_task)
1447 ret = count;
1448 else
1449 ret = -EIO;
1450
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001451out:
1452 if (filp)
1453 fput(filp);
1454 if (inode)
1455 iput(inode);
1456 if (ret < 0) {
1457 if (reg->hr_bdev) {
Al Viro9a1c3542008-02-22 20:40:24 -05001458 blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001459 reg->hr_bdev = NULL;
1460 }
1461 }
1462 return ret;
1463}
1464
Zhen Wei92efc152006-12-08 00:48:17 -07001465static ssize_t o2hb_region_pid_read(struct o2hb_region *reg,
1466 char *page)
1467{
Joel Beckere6c352d2007-02-03 03:04:20 -08001468 pid_t pid = 0;
1469
1470 spin_lock(&o2hb_live_lock);
1471 if (reg->hr_task)
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001472 pid = task_pid_nr(reg->hr_task);
Joel Beckere6c352d2007-02-03 03:04:20 -08001473 spin_unlock(&o2hb_live_lock);
1474
1475 if (!pid)
Zhen Wei92efc152006-12-08 00:48:17 -07001476 return 0;
1477
Joel Beckere6c352d2007-02-03 03:04:20 -08001478 return sprintf(page, "%u\n", pid);
Zhen Wei92efc152006-12-08 00:48:17 -07001479}
1480
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001481struct o2hb_region_attribute {
1482 struct configfs_attribute attr;
1483 ssize_t (*show)(struct o2hb_region *, char *);
1484 ssize_t (*store)(struct o2hb_region *, const char *, size_t);
1485};
1486
1487static struct o2hb_region_attribute o2hb_region_attr_block_bytes = {
1488 .attr = { .ca_owner = THIS_MODULE,
1489 .ca_name = "block_bytes",
1490 .ca_mode = S_IRUGO | S_IWUSR },
1491 .show = o2hb_region_block_bytes_read,
1492 .store = o2hb_region_block_bytes_write,
1493};
1494
1495static struct o2hb_region_attribute o2hb_region_attr_start_block = {
1496 .attr = { .ca_owner = THIS_MODULE,
1497 .ca_name = "start_block",
1498 .ca_mode = S_IRUGO | S_IWUSR },
1499 .show = o2hb_region_start_block_read,
1500 .store = o2hb_region_start_block_write,
1501};
1502
1503static struct o2hb_region_attribute o2hb_region_attr_blocks = {
1504 .attr = { .ca_owner = THIS_MODULE,
1505 .ca_name = "blocks",
1506 .ca_mode = S_IRUGO | S_IWUSR },
1507 .show = o2hb_region_blocks_read,
1508 .store = o2hb_region_blocks_write,
1509};
1510
1511static struct o2hb_region_attribute o2hb_region_attr_dev = {
1512 .attr = { .ca_owner = THIS_MODULE,
1513 .ca_name = "dev",
1514 .ca_mode = S_IRUGO | S_IWUSR },
1515 .show = o2hb_region_dev_read,
1516 .store = o2hb_region_dev_write,
1517};
1518
Zhen Wei92efc152006-12-08 00:48:17 -07001519static struct o2hb_region_attribute o2hb_region_attr_pid = {
1520 .attr = { .ca_owner = THIS_MODULE,
1521 .ca_name = "pid",
1522 .ca_mode = S_IRUGO | S_IRUSR },
1523 .show = o2hb_region_pid_read,
1524};
1525
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001526static struct configfs_attribute *o2hb_region_attrs[] = {
1527 &o2hb_region_attr_block_bytes.attr,
1528 &o2hb_region_attr_start_block.attr,
1529 &o2hb_region_attr_blocks.attr,
1530 &o2hb_region_attr_dev.attr,
Zhen Wei92efc152006-12-08 00:48:17 -07001531 &o2hb_region_attr_pid.attr,
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001532 NULL,
1533};
1534
1535static ssize_t o2hb_region_show(struct config_item *item,
1536 struct configfs_attribute *attr,
1537 char *page)
1538{
1539 struct o2hb_region *reg = to_o2hb_region(item);
1540 struct o2hb_region_attribute *o2hb_region_attr =
1541 container_of(attr, struct o2hb_region_attribute, attr);
1542 ssize_t ret = 0;
1543
1544 if (o2hb_region_attr->show)
1545 ret = o2hb_region_attr->show(reg, page);
1546 return ret;
1547}
1548
1549static ssize_t o2hb_region_store(struct config_item *item,
1550 struct configfs_attribute *attr,
1551 const char *page, size_t count)
1552{
1553 struct o2hb_region *reg = to_o2hb_region(item);
1554 struct o2hb_region_attribute *o2hb_region_attr =
1555 container_of(attr, struct o2hb_region_attribute, attr);
1556 ssize_t ret = -EINVAL;
1557
1558 if (o2hb_region_attr->store)
1559 ret = o2hb_region_attr->store(reg, page, count);
1560 return ret;
1561}
1562
1563static struct configfs_item_operations o2hb_region_item_ops = {
1564 .release = o2hb_region_release,
1565 .show_attribute = o2hb_region_show,
1566 .store_attribute = o2hb_region_store,
1567};
1568
1569static struct config_item_type o2hb_region_type = {
1570 .ct_item_ops = &o2hb_region_item_ops,
1571 .ct_attrs = o2hb_region_attrs,
1572 .ct_owner = THIS_MODULE,
1573};
1574
1575/* heartbeat set */
1576
1577struct o2hb_heartbeat_group {
1578 struct config_group hs_group;
1579 /* some stuff? */
1580};
1581
1582static struct o2hb_heartbeat_group *to_o2hb_heartbeat_group(struct config_group *group)
1583{
1584 return group ?
1585 container_of(group, struct o2hb_heartbeat_group, hs_group)
1586 : NULL;
1587}
1588
Joel Beckerf89ab862008-07-17 14:53:48 -07001589static struct config_item *o2hb_heartbeat_group_make_item(struct config_group *group,
1590 const char *name)
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001591{
1592 struct o2hb_region *reg = NULL;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001593
Robert P. J. Daycd861282006-12-13 00:34:52 -08001594 reg = kzalloc(sizeof(struct o2hb_region), GFP_KERNEL);
Joel Beckerf89ab862008-07-17 14:53:48 -07001595 if (reg == NULL)
Joel Beckera6795e92008-07-17 15:21:29 -07001596 return ERR_PTR(-ENOMEM);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001597
1598 config_item_init_type_name(&reg->hr_item, name, &o2hb_region_type);
1599
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001600 spin_lock(&o2hb_live_lock);
1601 list_add_tail(&reg->hr_all_item, &o2hb_all_regions);
1602 spin_unlock(&o2hb_live_lock);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001603
Joel Beckera6795e92008-07-17 15:21:29 -07001604 return &reg->hr_item;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001605}
1606
1607static void o2hb_heartbeat_group_drop_item(struct config_group *group,
1608 struct config_item *item)
1609{
Joel Beckere6c352d2007-02-03 03:04:20 -08001610 struct task_struct *hb_task;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001611 struct o2hb_region *reg = to_o2hb_region(item);
1612
1613 /* stop the thread when the user removes the region dir */
Joel Beckere6c352d2007-02-03 03:04:20 -08001614 spin_lock(&o2hb_live_lock);
1615 hb_task = reg->hr_task;
1616 reg->hr_task = NULL;
1617 spin_unlock(&o2hb_live_lock);
1618
1619 if (hb_task)
1620 kthread_stop(hb_task);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001621
Joel Beckere6df3a62007-02-06 15:45:39 -08001622 /*
1623 * If we're racing a dev_write(), we need to wake them. They will
1624 * check reg->hr_task
1625 */
1626 if (atomic_read(&reg->hr_steady_iterations) != 0) {
1627 atomic_set(&reg->hr_steady_iterations, 0);
1628 wake_up(&o2hb_steady_queue);
1629 }
1630
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001631 config_item_put(item);
1632}
1633
1634struct o2hb_heartbeat_group_attribute {
1635 struct configfs_attribute attr;
1636 ssize_t (*show)(struct o2hb_heartbeat_group *, char *);
1637 ssize_t (*store)(struct o2hb_heartbeat_group *, const char *, size_t);
1638};
1639
1640static ssize_t o2hb_heartbeat_group_show(struct config_item *item,
1641 struct configfs_attribute *attr,
1642 char *page)
1643{
1644 struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
1645 struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
1646 container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
1647 ssize_t ret = 0;
1648
1649 if (o2hb_heartbeat_group_attr->show)
1650 ret = o2hb_heartbeat_group_attr->show(reg, page);
1651 return ret;
1652}
1653
1654static ssize_t o2hb_heartbeat_group_store(struct config_item *item,
1655 struct configfs_attribute *attr,
1656 const char *page, size_t count)
1657{
1658 struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
1659 struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
1660 container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
1661 ssize_t ret = -EINVAL;
1662
1663 if (o2hb_heartbeat_group_attr->store)
1664 ret = o2hb_heartbeat_group_attr->store(reg, page, count);
1665 return ret;
1666}
1667
1668static ssize_t o2hb_heartbeat_group_threshold_show(struct o2hb_heartbeat_group *group,
1669 char *page)
1670{
1671 return sprintf(page, "%u\n", o2hb_dead_threshold);
1672}
1673
1674static ssize_t o2hb_heartbeat_group_threshold_store(struct o2hb_heartbeat_group *group,
1675 const char *page,
1676 size_t count)
1677{
1678 unsigned long tmp;
1679 char *p = (char *)page;
1680
1681 tmp = simple_strtoul(p, &p, 10);
1682 if (!p || (*p && (*p != '\n')))
1683 return -EINVAL;
1684
1685 /* this will validate ranges for us. */
1686 o2hb_dead_threshold_set((unsigned int) tmp);
1687
1688 return count;
1689}
1690
1691static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_threshold = {
1692 .attr = { .ca_owner = THIS_MODULE,
1693 .ca_name = "dead_threshold",
1694 .ca_mode = S_IRUGO | S_IWUSR },
1695 .show = o2hb_heartbeat_group_threshold_show,
1696 .store = o2hb_heartbeat_group_threshold_store,
1697};
1698
1699static struct configfs_attribute *o2hb_heartbeat_group_attrs[] = {
1700 &o2hb_heartbeat_group_attr_threshold.attr,
1701 NULL,
1702};
1703
1704static struct configfs_item_operations o2hb_hearbeat_group_item_ops = {
1705 .show_attribute = o2hb_heartbeat_group_show,
1706 .store_attribute = o2hb_heartbeat_group_store,
1707};
1708
1709static struct configfs_group_operations o2hb_heartbeat_group_group_ops = {
1710 .make_item = o2hb_heartbeat_group_make_item,
1711 .drop_item = o2hb_heartbeat_group_drop_item,
1712};
1713
1714static struct config_item_type o2hb_heartbeat_group_type = {
1715 .ct_group_ops = &o2hb_heartbeat_group_group_ops,
1716 .ct_item_ops = &o2hb_hearbeat_group_item_ops,
1717 .ct_attrs = o2hb_heartbeat_group_attrs,
1718 .ct_owner = THIS_MODULE,
1719};
1720
1721/* this is just here to avoid touching group in heartbeat.h which the
1722 * entire damn world #includes */
1723struct config_group *o2hb_alloc_hb_set(void)
1724{
1725 struct o2hb_heartbeat_group *hs = NULL;
1726 struct config_group *ret = NULL;
1727
Robert P. J. Daycd861282006-12-13 00:34:52 -08001728 hs = kzalloc(sizeof(struct o2hb_heartbeat_group), GFP_KERNEL);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001729 if (hs == NULL)
1730 goto out;
1731
1732 config_group_init_type_name(&hs->hs_group, "heartbeat",
1733 &o2hb_heartbeat_group_type);
1734
1735 ret = &hs->hs_group;
1736out:
1737 if (ret == NULL)
1738 kfree(hs);
1739 return ret;
1740}
1741
1742void o2hb_free_hb_set(struct config_group *group)
1743{
1744 struct o2hb_heartbeat_group *hs = to_o2hb_heartbeat_group(group);
1745 kfree(hs);
1746}
1747
1748/* hb callback registration and issueing */
1749
1750static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type)
1751{
1752 if (type == O2HB_NUM_CB)
1753 return ERR_PTR(-EINVAL);
1754
1755 return &o2hb_callbacks[type];
1756}
1757
1758void o2hb_setup_callback(struct o2hb_callback_func *hc,
1759 enum o2hb_callback_type type,
1760 o2hb_cb_func *func,
1761 void *data,
1762 int priority)
1763{
1764 INIT_LIST_HEAD(&hc->hc_item);
1765 hc->hc_func = func;
1766 hc->hc_data = data;
1767 hc->hc_priority = priority;
1768 hc->hc_type = type;
1769 hc->hc_magic = O2HB_CB_MAGIC;
1770}
1771EXPORT_SYMBOL_GPL(o2hb_setup_callback);
1772
Joel Becker14829422007-06-14 21:40:49 -07001773static struct o2hb_region *o2hb_find_region(const char *region_uuid)
1774{
1775 struct o2hb_region *p, *reg = NULL;
1776
1777 assert_spin_locked(&o2hb_live_lock);
1778
1779 list_for_each_entry(p, &o2hb_all_regions, hr_all_item) {
1780 if (!strcmp(region_uuid, config_item_name(&p->hr_item))) {
1781 reg = p;
1782 break;
1783 }
1784 }
1785
1786 return reg;
1787}
1788
1789static int o2hb_region_get(const char *region_uuid)
1790{
1791 int ret = 0;
1792 struct o2hb_region *reg;
1793
1794 spin_lock(&o2hb_live_lock);
1795
1796 reg = o2hb_find_region(region_uuid);
1797 if (!reg)
1798 ret = -ENOENT;
1799 spin_unlock(&o2hb_live_lock);
1800
Joel Becker16c6a4f2007-06-19 11:34:03 -07001801 if (ret)
1802 goto out;
Joel Becker14829422007-06-14 21:40:49 -07001803
Joel Becker16c6a4f2007-06-19 11:34:03 -07001804 ret = o2nm_depend_this_node();
1805 if (ret)
1806 goto out;
1807
1808 ret = o2nm_depend_item(&reg->hr_item);
1809 if (ret)
1810 o2nm_undepend_this_node();
1811
1812out:
Joel Becker14829422007-06-14 21:40:49 -07001813 return ret;
1814}
1815
1816static void o2hb_region_put(const char *region_uuid)
1817{
1818 struct o2hb_region *reg;
1819
1820 spin_lock(&o2hb_live_lock);
1821
1822 reg = o2hb_find_region(region_uuid);
1823
1824 spin_unlock(&o2hb_live_lock);
1825
Joel Becker16c6a4f2007-06-19 11:34:03 -07001826 if (reg) {
Joel Becker14829422007-06-14 21:40:49 -07001827 o2nm_undepend_item(&reg->hr_item);
Joel Becker16c6a4f2007-06-19 11:34:03 -07001828 o2nm_undepend_this_node();
1829 }
Joel Becker14829422007-06-14 21:40:49 -07001830}
1831
1832int o2hb_register_callback(const char *region_uuid,
1833 struct o2hb_callback_func *hc)
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001834{
1835 struct o2hb_callback_func *tmp;
1836 struct list_head *iter;
1837 struct o2hb_callback *hbcall;
1838 int ret;
1839
1840 BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
1841 BUG_ON(!list_empty(&hc->hc_item));
1842
1843 hbcall = hbcall_from_type(hc->hc_type);
1844 if (IS_ERR(hbcall)) {
1845 ret = PTR_ERR(hbcall);
1846 goto out;
1847 }
1848
Joel Becker14829422007-06-14 21:40:49 -07001849 if (region_uuid) {
1850 ret = o2hb_region_get(region_uuid);
1851 if (ret)
1852 goto out;
1853 }
1854
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001855 down_write(&o2hb_callback_sem);
1856
1857 list_for_each(iter, &hbcall->list) {
1858 tmp = list_entry(iter, struct o2hb_callback_func, hc_item);
1859 if (hc->hc_priority < tmp->hc_priority) {
1860 list_add_tail(&hc->hc_item, iter);
1861 break;
1862 }
1863 }
1864 if (list_empty(&hc->hc_item))
1865 list_add_tail(&hc->hc_item, &hbcall->list);
1866
1867 up_write(&o2hb_callback_sem);
1868 ret = 0;
1869out:
1870 mlog(ML_HEARTBEAT, "returning %d on behalf of %p for funcs %p\n",
1871 ret, __builtin_return_address(0), hc);
1872 return ret;
1873}
1874EXPORT_SYMBOL_GPL(o2hb_register_callback);
1875
Joel Becker14829422007-06-14 21:40:49 -07001876void o2hb_unregister_callback(const char *region_uuid,
1877 struct o2hb_callback_func *hc)
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001878{
1879 BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
1880
1881 mlog(ML_HEARTBEAT, "on behalf of %p for funcs %p\n",
1882 __builtin_return_address(0), hc);
1883
Joel Becker14829422007-06-14 21:40:49 -07001884 /* XXX Can this happen _with_ a region reference? */
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001885 if (list_empty(&hc->hc_item))
Joel Beckerc24f72c2007-02-03 03:14:30 -08001886 return;
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001887
Joel Becker14829422007-06-14 21:40:49 -07001888 if (region_uuid)
1889 o2hb_region_put(region_uuid);
1890
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001891 down_write(&o2hb_callback_sem);
1892
1893 list_del_init(&hc->hc_item);
1894
1895 up_write(&o2hb_callback_sem);
Mark Fasheha7f6a5f2005-12-15 14:31:23 -08001896}
1897EXPORT_SYMBOL_GPL(o2hb_unregister_callback);
1898
1899int o2hb_check_node_heartbeating(u8 node_num)
1900{
1901 unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1902
1903 o2hb_fill_node_map(testing_map, sizeof(testing_map));
1904 if (!test_bit(node_num, testing_map)) {
1905 mlog(ML_HEARTBEAT,
1906 "node (%u) does not have heartbeating enabled.\n",
1907 node_num);
1908 return 0;
1909 }
1910
1911 return 1;
1912}
1913EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating);
1914
1915int o2hb_check_node_heartbeating_from_callback(u8 node_num)
1916{
1917 unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1918
1919 o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
1920 if (!test_bit(node_num, testing_map)) {
1921 mlog(ML_HEARTBEAT,
1922 "node (%u) does not have heartbeating enabled.\n",
1923 node_num);
1924 return 0;
1925 }
1926
1927 return 1;
1928}
1929EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback);
1930
1931/* Makes sure our local node is configured with a node number, and is
1932 * heartbeating. */
1933int o2hb_check_local_node_heartbeating(void)
1934{
1935 u8 node_num;
1936
1937 /* if this node was set then we have networking */
1938 node_num = o2nm_this_node();
1939 if (node_num == O2NM_MAX_NODES) {
1940 mlog(ML_HEARTBEAT, "this node has not been configured.\n");
1941 return 0;
1942 }
1943
1944 return o2hb_check_node_heartbeating(node_num);
1945}
1946EXPORT_SYMBOL_GPL(o2hb_check_local_node_heartbeating);
1947
1948/*
1949 * this is just a hack until we get the plumbing which flips file systems
1950 * read only and drops the hb ref instead of killing the node dead.
1951 */
1952void o2hb_stop_all_regions(void)
1953{
1954 struct o2hb_region *reg;
1955
1956 mlog(ML_ERROR, "stopping heartbeat on all active regions.\n");
1957
1958 spin_lock(&o2hb_live_lock);
1959
1960 list_for_each_entry(reg, &o2hb_all_regions, hr_all_item)
1961 reg->hr_unclean_stop = 1;
1962
1963 spin_unlock(&o2hb_live_lock);
1964}
1965EXPORT_SYMBOL_GPL(o2hb_stop_all_regions);