4f85eceab376015596adbafae5b9eb774211e2b1
[linux-2.6.git] / fs / ocfs2 / cluster / heartbeat.c
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * Copyright (C) 2004, 2005 Oracle.  All rights reserved.
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public
8  * License as published by the Free Software Foundation; either
9  * version 2 of the License, or (at your option) any later version.
10  *
11  * This program is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
14  * General Public License for more details.
15  *
16  * You should have received a copy of the GNU General Public
17  * License along with this program; if not, write to the
18  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
19  * Boston, MA 021110-1307, USA.
20  */
21
22 #include <linux/kernel.h>
23 #include <linux/sched.h>
24 #include <linux/jiffies.h>
25 #include <linux/module.h>
26 #include <linux/fs.h>
27 #include <linux/bio.h>
28 #include <linux/blkdev.h>
29 #include <linux/delay.h>
30 #include <linux/file.h>
31 #include <linux/kthread.h>
32 #include <linux/configfs.h>
33 #include <linux/random.h>
34 #include <linux/crc32.h>
35 #include <linux/time.h>
36 #include <linux/debugfs.h>
37
38 #include "heartbeat.h"
39 #include "tcp.h"
40 #include "nodemanager.h"
41 #include "quorum.h"
42
43 #include "masklog.h"
44
45
46 /*
47  * The first heartbeat pass had one global thread that would serialize all hb
48  * callback calls.  This global serializing sem should only be removed once
49  * we've made sure that all callees can deal with being called concurrently
50  * from multiple hb region threads.
51  */
52 static DECLARE_RWSEM(o2hb_callback_sem);
53
54 /*
55  * multiple hb threads are watching multiple regions.  A node is live
56  * whenever any of the threads sees activity from the node in its region.
57  */
58 static DEFINE_SPINLOCK(o2hb_live_lock);
59 static struct list_head o2hb_live_slots[O2NM_MAX_NODES];
60 static unsigned long o2hb_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
61 static LIST_HEAD(o2hb_node_events);
62 static DECLARE_WAIT_QUEUE_HEAD(o2hb_steady_queue);
63
64 #define O2HB_DEBUG_DIR                  "o2hb"
65 #define O2HB_DEBUG_LIVENODES            "livenodes"
66 static struct dentry *o2hb_debug_dir;
67 static struct dentry *o2hb_debug_livenodes;
68
69 static LIST_HEAD(o2hb_all_regions);
70
71 static struct o2hb_callback {
72         struct list_head list;
73 } o2hb_callbacks[O2HB_NUM_CB];
74
75 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type);
76
77 #define O2HB_DEFAULT_BLOCK_BITS       9
78
79 unsigned int o2hb_dead_threshold = O2HB_DEFAULT_DEAD_THRESHOLD;
80
81 /* Only sets a new threshold if there are no active regions. 
82  *
83  * No locking or otherwise interesting code is required for reading
84  * o2hb_dead_threshold as it can't change once regions are active and
85  * it's not interesting to anyone until then anyway. */
86 static void o2hb_dead_threshold_set(unsigned int threshold)
87 {
88         if (threshold > O2HB_MIN_DEAD_THRESHOLD) {
89                 spin_lock(&o2hb_live_lock);
90                 if (list_empty(&o2hb_all_regions))
91                         o2hb_dead_threshold = threshold;
92                 spin_unlock(&o2hb_live_lock);
93         }
94 }
95
96 struct o2hb_node_event {
97         struct list_head        hn_item;
98         enum o2hb_callback_type hn_event_type;
99         struct o2nm_node        *hn_node;
100         int                     hn_node_num;
101 };
102
103 struct o2hb_disk_slot {
104         struct o2hb_disk_heartbeat_block *ds_raw_block;
105         u8                      ds_node_num;
106         u64                     ds_last_time;
107         u64                     ds_last_generation;
108         u16                     ds_equal_samples;
109         u16                     ds_changed_samples;
110         struct list_head        ds_live_item;
111 };
112
113 /* each thread owns a region.. when we're asked to tear down the region
114  * we ask the thread to stop, who cleans up the region */
115 struct o2hb_region {
116         struct config_item      hr_item;
117
118         struct list_head        hr_all_item;
119         unsigned                hr_unclean_stop:1;
120
121         /* protected by the hr_callback_sem */
122         struct task_struct      *hr_task;
123
124         unsigned int            hr_blocks;
125         unsigned long long      hr_start_block;
126
127         unsigned int            hr_block_bits;
128         unsigned int            hr_block_bytes;
129
130         unsigned int            hr_slots_per_page;
131         unsigned int            hr_num_pages;
132
133         struct page             **hr_slot_data;
134         struct block_device     *hr_bdev;
135         struct o2hb_disk_slot   *hr_slots;
136
137         /* let the person setting up hb wait for it to return until it
138          * has reached a 'steady' state.  This will be fixed when we have
139          * a more complete api that doesn't lead to this sort of fragility. */
140         atomic_t                hr_steady_iterations;
141
142         char                    hr_dev_name[BDEVNAME_SIZE];
143
144         unsigned int            hr_timeout_ms;
145
146         /* randomized as the region goes up and down so that a node
147          * recognizes a node going up and down in one iteration */
148         u64                     hr_generation;
149
150         struct delayed_work     hr_write_timeout_work;
151         unsigned long           hr_last_timeout_start;
152
153         /* Used during o2hb_check_slot to hold a copy of the block
154          * being checked because we temporarily have to zero out the
155          * crc field. */
156         struct o2hb_disk_heartbeat_block *hr_tmp_block;
157 };
158
159 struct o2hb_bio_wait_ctxt {
160         atomic_t          wc_num_reqs;
161         struct completion wc_io_complete;
162         int               wc_error;
163 };
164
165 static void o2hb_write_timeout(struct work_struct *work)
166 {
167         struct o2hb_region *reg =
168                 container_of(work, struct o2hb_region,
169                              hr_write_timeout_work.work);
170
171         mlog(ML_ERROR, "Heartbeat write timeout to device %s after %u "
172              "milliseconds\n", reg->hr_dev_name,
173              jiffies_to_msecs(jiffies - reg->hr_last_timeout_start)); 
174         o2quo_disk_timeout();
175 }
176
177 static void o2hb_arm_write_timeout(struct o2hb_region *reg)
178 {
179         mlog(0, "Queue write timeout for %u ms\n", O2HB_MAX_WRITE_TIMEOUT_MS);
180
181         cancel_delayed_work(&reg->hr_write_timeout_work);
182         reg->hr_last_timeout_start = jiffies;
183         schedule_delayed_work(&reg->hr_write_timeout_work,
184                               msecs_to_jiffies(O2HB_MAX_WRITE_TIMEOUT_MS));
185 }
186
187 static void o2hb_disarm_write_timeout(struct o2hb_region *reg)
188 {
189         cancel_delayed_work(&reg->hr_write_timeout_work);
190         flush_scheduled_work();
191 }
192
193 static inline void o2hb_bio_wait_init(struct o2hb_bio_wait_ctxt *wc)
194 {
195         atomic_set(&wc->wc_num_reqs, 1);
196         init_completion(&wc->wc_io_complete);
197         wc->wc_error = 0;
198 }
199
200 /* Used in error paths too */
201 static inline void o2hb_bio_wait_dec(struct o2hb_bio_wait_ctxt *wc,
202                                      unsigned int num)
203 {
204         /* sadly atomic_sub_and_test() isn't available on all platforms.  The
205          * good news is that the fast path only completes one at a time */
206         while(num--) {
207                 if (atomic_dec_and_test(&wc->wc_num_reqs)) {
208                         BUG_ON(num > 0);
209                         complete(&wc->wc_io_complete);
210                 }
211         }
212 }
213
214 static void o2hb_wait_on_io(struct o2hb_region *reg,
215                             struct o2hb_bio_wait_ctxt *wc)
216 {
217         struct address_space *mapping = reg->hr_bdev->bd_inode->i_mapping;
218
219         blk_run_address_space(mapping);
220         o2hb_bio_wait_dec(wc, 1);
221
222         wait_for_completion(&wc->wc_io_complete);
223 }
224
225 static void o2hb_bio_end_io(struct bio *bio,
226                            int error)
227 {
228         struct o2hb_bio_wait_ctxt *wc = bio->bi_private;
229
230         if (error) {
231                 mlog(ML_ERROR, "IO Error %d\n", error);
232                 wc->wc_error = error;
233         }
234
235         o2hb_bio_wait_dec(wc, 1);
236         bio_put(bio);
237 }
238
239 /* Setup a Bio to cover I/O against num_slots slots starting at
240  * start_slot. */
241 static struct bio *o2hb_setup_one_bio(struct o2hb_region *reg,
242                                       struct o2hb_bio_wait_ctxt *wc,
243                                       unsigned int *current_slot,
244                                       unsigned int max_slots)
245 {
246         int len, current_page;
247         unsigned int vec_len, vec_start;
248         unsigned int bits = reg->hr_block_bits;
249         unsigned int spp = reg->hr_slots_per_page;
250         unsigned int cs = *current_slot;
251         struct bio *bio;
252         struct page *page;
253
254         /* Testing has shown this allocation to take long enough under
255          * GFP_KERNEL that the local node can get fenced. It would be
256          * nicest if we could pre-allocate these bios and avoid this
257          * all together. */
258         bio = bio_alloc(GFP_ATOMIC, 16);
259         if (!bio) {
260                 mlog(ML_ERROR, "Could not alloc slots BIO!\n");
261                 bio = ERR_PTR(-ENOMEM);
262                 goto bail;
263         }
264
265         /* Must put everything in 512 byte sectors for the bio... */
266         bio->bi_sector = (reg->hr_start_block + cs) << (bits - 9);
267         bio->bi_bdev = reg->hr_bdev;
268         bio->bi_private = wc;
269         bio->bi_end_io = o2hb_bio_end_io;
270
271         vec_start = (cs << bits) % PAGE_CACHE_SIZE;
272         while(cs < max_slots) {
273                 current_page = cs / spp;
274                 page = reg->hr_slot_data[current_page];
275
276                 vec_len = min(PAGE_CACHE_SIZE - vec_start,
277                               (max_slots-cs) * (PAGE_CACHE_SIZE/spp) );
278
279                 mlog(ML_HB_BIO, "page %d, vec_len = %u, vec_start = %u\n",
280                      current_page, vec_len, vec_start);
281
282                 len = bio_add_page(bio, page, vec_len, vec_start);
283                 if (len != vec_len) break;
284
285                 cs += vec_len / (PAGE_CACHE_SIZE/spp);
286                 vec_start = 0;
287         }
288
289 bail:
290         *current_slot = cs;
291         return bio;
292 }
293
294 static int o2hb_read_slots(struct o2hb_region *reg,
295                            unsigned int max_slots)
296 {
297         unsigned int current_slot=0;
298         int status;
299         struct o2hb_bio_wait_ctxt wc;
300         struct bio *bio;
301
302         o2hb_bio_wait_init(&wc);
303
304         while(current_slot < max_slots) {
305                 bio = o2hb_setup_one_bio(reg, &wc, &current_slot, max_slots);
306                 if (IS_ERR(bio)) {
307                         status = PTR_ERR(bio);
308                         mlog_errno(status);
309                         goto bail_and_wait;
310                 }
311
312                 atomic_inc(&wc.wc_num_reqs);
313                 submit_bio(READ, bio);
314         }
315
316         status = 0;
317
318 bail_and_wait:
319         o2hb_wait_on_io(reg, &wc);
320         if (wc.wc_error && !status)
321                 status = wc.wc_error;
322
323         return status;
324 }
325
326 static int o2hb_issue_node_write(struct o2hb_region *reg,
327                                  struct o2hb_bio_wait_ctxt *write_wc)
328 {
329         int status;
330         unsigned int slot;
331         struct bio *bio;
332
333         o2hb_bio_wait_init(write_wc);
334
335         slot = o2nm_this_node();
336
337         bio = o2hb_setup_one_bio(reg, write_wc, &slot, slot+1);
338         if (IS_ERR(bio)) {
339                 status = PTR_ERR(bio);
340                 mlog_errno(status);
341                 goto bail;
342         }
343
344         atomic_inc(&write_wc->wc_num_reqs);
345         submit_bio(WRITE, bio);
346
347         status = 0;
348 bail:
349         return status;
350 }
351
352 static u32 o2hb_compute_block_crc_le(struct o2hb_region *reg,
353                                      struct o2hb_disk_heartbeat_block *hb_block)
354 {
355         __le32 old_cksum;
356         u32 ret;
357
358         /* We want to compute the block crc with a 0 value in the
359          * hb_cksum field. Save it off here and replace after the
360          * crc. */
361         old_cksum = hb_block->hb_cksum;
362         hb_block->hb_cksum = 0;
363
364         ret = crc32_le(0, (unsigned char *) hb_block, reg->hr_block_bytes);
365
366         hb_block->hb_cksum = old_cksum;
367
368         return ret;
369 }
370
371 static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block *hb_block)
372 {
373         mlog(ML_ERROR, "Dump slot information: seq = 0x%llx, node = %u, "
374              "cksum = 0x%x, generation 0x%llx\n",
375              (long long)le64_to_cpu(hb_block->hb_seq),
376              hb_block->hb_node, le32_to_cpu(hb_block->hb_cksum),
377              (long long)le64_to_cpu(hb_block->hb_generation));
378 }
379
380 static int o2hb_verify_crc(struct o2hb_region *reg,
381                            struct o2hb_disk_heartbeat_block *hb_block)
382 {
383         u32 read, computed;
384
385         read = le32_to_cpu(hb_block->hb_cksum);
386         computed = o2hb_compute_block_crc_le(reg, hb_block);
387
388         return read == computed;
389 }
390
391 /* We want to make sure that nobody is heartbeating on top of us --
392  * this will help detect an invalid configuration. */
393 static int o2hb_check_last_timestamp(struct o2hb_region *reg)
394 {
395         int node_num, ret;
396         struct o2hb_disk_slot *slot;
397         struct o2hb_disk_heartbeat_block *hb_block;
398
399         node_num = o2nm_this_node();
400
401         ret = 1;
402         slot = &reg->hr_slots[node_num];
403         /* Don't check on our 1st timestamp */
404         if (slot->ds_last_time) {
405                 hb_block = slot->ds_raw_block;
406
407                 if (le64_to_cpu(hb_block->hb_seq) != slot->ds_last_time)
408                         ret = 0;
409         }
410
411         return ret;
412 }
413
414 static inline void o2hb_prepare_block(struct o2hb_region *reg,
415                                       u64 generation)
416 {
417         int node_num;
418         u64 cputime;
419         struct o2hb_disk_slot *slot;
420         struct o2hb_disk_heartbeat_block *hb_block;
421
422         node_num = o2nm_this_node();
423         slot = &reg->hr_slots[node_num];
424
425         hb_block = (struct o2hb_disk_heartbeat_block *)slot->ds_raw_block;
426         memset(hb_block, 0, reg->hr_block_bytes);
427         /* TODO: time stuff */
428         cputime = CURRENT_TIME.tv_sec;
429         if (!cputime)
430                 cputime = 1;
431
432         hb_block->hb_seq = cpu_to_le64(cputime);
433         hb_block->hb_node = node_num;
434         hb_block->hb_generation = cpu_to_le64(generation);
435         hb_block->hb_dead_ms = cpu_to_le32(o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS);
436
437         /* This step must always happen last! */
438         hb_block->hb_cksum = cpu_to_le32(o2hb_compute_block_crc_le(reg,
439                                                                    hb_block));
440
441         mlog(ML_HB_BIO, "our node generation = 0x%llx, cksum = 0x%x\n",
442              (long long)generation,
443              le32_to_cpu(hb_block->hb_cksum));
444 }
445
446 static void o2hb_fire_callbacks(struct o2hb_callback *hbcall,
447                                 struct o2nm_node *node,
448                                 int idx)
449 {
450         struct list_head *iter;
451         struct o2hb_callback_func *f;
452
453         list_for_each(iter, &hbcall->list) {
454                 f = list_entry(iter, struct o2hb_callback_func, hc_item);
455                 mlog(ML_HEARTBEAT, "calling funcs %p\n", f);
456                 (f->hc_func)(node, idx, f->hc_data);
457         }
458 }
459
460 /* Will run the list in order until we process the passed event */
461 static void o2hb_run_event_list(struct o2hb_node_event *queued_event)
462 {
463         int empty;
464         struct o2hb_callback *hbcall;
465         struct o2hb_node_event *event;
466
467         spin_lock(&o2hb_live_lock);
468         empty = list_empty(&queued_event->hn_item);
469         spin_unlock(&o2hb_live_lock);
470         if (empty)
471                 return;
472
473         /* Holding callback sem assures we don't alter the callback
474          * lists when doing this, and serializes ourselves with other
475          * processes wanting callbacks. */
476         down_write(&o2hb_callback_sem);
477
478         spin_lock(&o2hb_live_lock);
479         while (!list_empty(&o2hb_node_events)
480                && !list_empty(&queued_event->hn_item)) {
481                 event = list_entry(o2hb_node_events.next,
482                                    struct o2hb_node_event,
483                                    hn_item);
484                 list_del_init(&event->hn_item);
485                 spin_unlock(&o2hb_live_lock);
486
487                 mlog(ML_HEARTBEAT, "Node %s event for %d\n",
488                      event->hn_event_type == O2HB_NODE_UP_CB ? "UP" : "DOWN",
489                      event->hn_node_num);
490
491                 hbcall = hbcall_from_type(event->hn_event_type);
492
493                 /* We should *never* have gotten on to the list with a
494                  * bad type... This isn't something that we should try
495                  * to recover from. */
496                 BUG_ON(IS_ERR(hbcall));
497
498                 o2hb_fire_callbacks(hbcall, event->hn_node, event->hn_node_num);
499
500                 spin_lock(&o2hb_live_lock);
501         }
502         spin_unlock(&o2hb_live_lock);
503
504         up_write(&o2hb_callback_sem);
505 }
506
507 static void o2hb_queue_node_event(struct o2hb_node_event *event,
508                                   enum o2hb_callback_type type,
509                                   struct o2nm_node *node,
510                                   int node_num)
511 {
512         assert_spin_locked(&o2hb_live_lock);
513
514         event->hn_event_type = type;
515         event->hn_node = node;
516         event->hn_node_num = node_num;
517
518         mlog(ML_HEARTBEAT, "Queue node %s event for node %d\n",
519              type == O2HB_NODE_UP_CB ? "UP" : "DOWN", node_num);
520
521         list_add_tail(&event->hn_item, &o2hb_node_events);
522 }
523
524 static void o2hb_shutdown_slot(struct o2hb_disk_slot *slot)
525 {
526         struct o2hb_node_event event =
527                 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
528         struct o2nm_node *node;
529
530         node = o2nm_get_node_by_num(slot->ds_node_num);
531         if (!node)
532                 return;
533
534         spin_lock(&o2hb_live_lock);
535         if (!list_empty(&slot->ds_live_item)) {
536                 mlog(ML_HEARTBEAT, "Shutdown, node %d leaves region\n",
537                      slot->ds_node_num);
538
539                 list_del_init(&slot->ds_live_item);
540
541                 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
542                         clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
543
544                         o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node,
545                                               slot->ds_node_num);
546                 }
547         }
548         spin_unlock(&o2hb_live_lock);
549
550         o2hb_run_event_list(&event);
551
552         o2nm_node_put(node);
553 }
554
555 static int o2hb_check_slot(struct o2hb_region *reg,
556                            struct o2hb_disk_slot *slot)
557 {
558         int changed = 0, gen_changed = 0;
559         struct o2hb_node_event event =
560                 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
561         struct o2nm_node *node;
562         struct o2hb_disk_heartbeat_block *hb_block = reg->hr_tmp_block;
563         u64 cputime;
564         unsigned int dead_ms = o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS;
565         unsigned int slot_dead_ms;
566
567         memcpy(hb_block, slot->ds_raw_block, reg->hr_block_bytes);
568
569         /* Is this correct? Do we assume that the node doesn't exist
570          * if we're not configured for him? */
571         node = o2nm_get_node_by_num(slot->ds_node_num);
572         if (!node)
573                 return 0;
574
575         if (!o2hb_verify_crc(reg, hb_block)) {
576                 /* all paths from here will drop o2hb_live_lock for
577                  * us. */
578                 spin_lock(&o2hb_live_lock);
579
580                 /* Don't print an error on the console in this case -
581                  * a freshly formatted heartbeat area will not have a
582                  * crc set on it. */
583                 if (list_empty(&slot->ds_live_item))
584                         goto out;
585
586                 /* The node is live but pushed out a bad crc. We
587                  * consider it a transient miss but don't populate any
588                  * other values as they may be junk. */
589                 mlog(ML_ERROR, "Node %d has written a bad crc to %s\n",
590                      slot->ds_node_num, reg->hr_dev_name);
591                 o2hb_dump_slot(hb_block);
592
593                 slot->ds_equal_samples++;
594                 goto fire_callbacks;
595         }
596
597         /* we don't care if these wrap.. the state transitions below
598          * clear at the right places */
599         cputime = le64_to_cpu(hb_block->hb_seq);
600         if (slot->ds_last_time != cputime)
601                 slot->ds_changed_samples++;
602         else
603                 slot->ds_equal_samples++;
604         slot->ds_last_time = cputime;
605
606         /* The node changed heartbeat generations. We assume this to
607          * mean it dropped off but came back before we timed out. We
608          * want to consider it down for the time being but don't want
609          * to lose any changed_samples state we might build up to
610          * considering it live again. */
611         if (slot->ds_last_generation != le64_to_cpu(hb_block->hb_generation)) {
612                 gen_changed = 1;
613                 slot->ds_equal_samples = 0;
614                 mlog(ML_HEARTBEAT, "Node %d changed generation (0x%llx "
615                      "to 0x%llx)\n", slot->ds_node_num,
616                      (long long)slot->ds_last_generation,
617                      (long long)le64_to_cpu(hb_block->hb_generation));
618         }
619
620         slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
621
622         mlog(ML_HEARTBEAT, "Slot %d gen 0x%llx cksum 0x%x "
623              "seq %llu last %llu changed %u equal %u\n",
624              slot->ds_node_num, (long long)slot->ds_last_generation,
625              le32_to_cpu(hb_block->hb_cksum),
626              (unsigned long long)le64_to_cpu(hb_block->hb_seq), 
627              (unsigned long long)slot->ds_last_time, slot->ds_changed_samples,
628              slot->ds_equal_samples);
629
630         spin_lock(&o2hb_live_lock);
631
632 fire_callbacks:
633         /* dead nodes only come to life after some number of
634          * changes at any time during their dead time */
635         if (list_empty(&slot->ds_live_item) &&
636             slot->ds_changed_samples >= O2HB_LIVE_THRESHOLD) {
637                 mlog(ML_HEARTBEAT, "Node %d (id 0x%llx) joined my region\n",
638                      slot->ds_node_num, (long long)slot->ds_last_generation);
639
640                 /* first on the list generates a callback */
641                 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
642                         set_bit(slot->ds_node_num, o2hb_live_node_bitmap);
643
644                         o2hb_queue_node_event(&event, O2HB_NODE_UP_CB, node,
645                                               slot->ds_node_num);
646
647                         changed = 1;
648                 }
649
650                 list_add_tail(&slot->ds_live_item,
651                               &o2hb_live_slots[slot->ds_node_num]);
652
653                 slot->ds_equal_samples = 0;
654
655                 /* We want to be sure that all nodes agree on the
656                  * number of milliseconds before a node will be
657                  * considered dead. The self-fencing timeout is
658                  * computed from this value, and a discrepancy might
659                  * result in heartbeat calling a node dead when it
660                  * hasn't self-fenced yet. */
661                 slot_dead_ms = le32_to_cpu(hb_block->hb_dead_ms);
662                 if (slot_dead_ms && slot_dead_ms != dead_ms) {
663                         /* TODO: Perhaps we can fail the region here. */
664                         mlog(ML_ERROR, "Node %d on device %s has a dead count "
665                              "of %u ms, but our count is %u ms.\n"
666                              "Please double check your configuration values "
667                              "for 'O2CB_HEARTBEAT_THRESHOLD'\n",
668                              slot->ds_node_num, reg->hr_dev_name, slot_dead_ms,
669                              dead_ms);
670                 }
671                 goto out;
672         }
673
674         /* if the list is dead, we're done.. */
675         if (list_empty(&slot->ds_live_item))
676                 goto out;
677
678         /* live nodes only go dead after enough consequtive missed
679          * samples..  reset the missed counter whenever we see
680          * activity */
681         if (slot->ds_equal_samples >= o2hb_dead_threshold || gen_changed) {
682                 mlog(ML_HEARTBEAT, "Node %d left my region\n",
683                      slot->ds_node_num);
684
685                 /* last off the live_slot generates a callback */
686                 list_del_init(&slot->ds_live_item);
687                 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
688                         clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
689
690                         o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node,
691                                               slot->ds_node_num);
692
693                         changed = 1;
694                 }
695
696                 /* We don't clear this because the node is still
697                  * actually writing new blocks. */
698                 if (!gen_changed)
699                         slot->ds_changed_samples = 0;
700                 goto out;
701         }
702         if (slot->ds_changed_samples) {
703                 slot->ds_changed_samples = 0;
704                 slot->ds_equal_samples = 0;
705         }
706 out:
707         spin_unlock(&o2hb_live_lock);
708
709         o2hb_run_event_list(&event);
710
711         o2nm_node_put(node);
712         return changed;
713 }
714
715 /* This could be faster if we just implmented a find_last_bit, but I
716  * don't think the circumstances warrant it. */
717 static int o2hb_highest_node(unsigned long *nodes,
718                              int numbits)
719 {
720         int highest, node;
721
722         highest = numbits;
723         node = -1;
724         while ((node = find_next_bit(nodes, numbits, node + 1)) != -1) {
725                 if (node >= numbits)
726                         break;
727
728                 highest = node;
729         }
730
731         return highest;
732 }
733
734 static int o2hb_do_disk_heartbeat(struct o2hb_region *reg)
735 {
736         int i, ret, highest_node, change = 0;
737         unsigned long configured_nodes[BITS_TO_LONGS(O2NM_MAX_NODES)];
738         struct o2hb_bio_wait_ctxt write_wc;
739
740         ret = o2nm_configured_node_map(configured_nodes,
741                                        sizeof(configured_nodes));
742         if (ret) {
743                 mlog_errno(ret);
744                 return ret;
745         }
746
747         highest_node = o2hb_highest_node(configured_nodes, O2NM_MAX_NODES);
748         if (highest_node >= O2NM_MAX_NODES) {
749                 mlog(ML_NOTICE, "ocfs2_heartbeat: no configured nodes found!\n");
750                 return -EINVAL;
751         }
752
753         /* No sense in reading the slots of nodes that don't exist
754          * yet. Of course, if the node definitions have holes in them
755          * then we're reading an empty slot anyway... Consider this
756          * best-effort. */
757         ret = o2hb_read_slots(reg, highest_node + 1);
758         if (ret < 0) {
759                 mlog_errno(ret);
760                 return ret;
761         }
762
763         /* With an up to date view of the slots, we can check that no
764          * other node has been improperly configured to heartbeat in
765          * our slot. */
766         if (!o2hb_check_last_timestamp(reg))
767                 mlog(ML_ERROR, "Device \"%s\": another node is heartbeating "
768                      "in our slot!\n", reg->hr_dev_name);
769
770         /* fill in the proper info for our next heartbeat */
771         o2hb_prepare_block(reg, reg->hr_generation);
772
773         /* And fire off the write. Note that we don't wait on this I/O
774          * until later. */
775         ret = o2hb_issue_node_write(reg, &write_wc);
776         if (ret < 0) {
777                 mlog_errno(ret);
778                 return ret;
779         }
780
781         i = -1;
782         while((i = find_next_bit(configured_nodes, O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
783
784                 change |= o2hb_check_slot(reg, &reg->hr_slots[i]);
785         }
786
787         /*
788          * We have to be sure we've advertised ourselves on disk
789          * before we can go to steady state.  This ensures that
790          * people we find in our steady state have seen us.
791          */
792         o2hb_wait_on_io(reg, &write_wc);
793         if (write_wc.wc_error) {
794                 /* Do not re-arm the write timeout on I/O error - we
795                  * can't be sure that the new block ever made it to
796                  * disk */
797                 mlog(ML_ERROR, "Write error %d on device \"%s\"\n",
798                      write_wc.wc_error, reg->hr_dev_name);
799                 return write_wc.wc_error;
800         }
801
802         o2hb_arm_write_timeout(reg);
803
804         /* let the person who launched us know when things are steady */
805         if (!change && (atomic_read(&reg->hr_steady_iterations) != 0)) {
806                 if (atomic_dec_and_test(&reg->hr_steady_iterations))
807                         wake_up(&o2hb_steady_queue);
808         }
809
810         return 0;
811 }
812
813 /* Subtract b from a, storing the result in a. a *must* have a larger
814  * value than b. */
815 static void o2hb_tv_subtract(struct timeval *a,
816                              struct timeval *b)
817 {
818         /* just return 0 when a is after b */
819         if (a->tv_sec < b->tv_sec ||
820             (a->tv_sec == b->tv_sec && a->tv_usec < b->tv_usec)) {
821                 a->tv_sec = 0;
822                 a->tv_usec = 0;
823                 return;
824         }
825
826         a->tv_sec -= b->tv_sec;
827         a->tv_usec -= b->tv_usec;
828         while ( a->tv_usec < 0 ) {
829                 a->tv_sec--;
830                 a->tv_usec += 1000000;
831         }
832 }
833
834 static unsigned int o2hb_elapsed_msecs(struct timeval *start,
835                                        struct timeval *end)
836 {
837         struct timeval res = *end;
838
839         o2hb_tv_subtract(&res, start);
840
841         return res.tv_sec * 1000 + res.tv_usec / 1000;
842 }
843
844 /*
845  * we ride the region ref that the region dir holds.  before the region
846  * dir is removed and drops it ref it will wait to tear down this
847  * thread.
848  */
849 static int o2hb_thread(void *data)
850 {
851         int i, ret;
852         struct o2hb_region *reg = data;
853         struct o2hb_bio_wait_ctxt write_wc;
854         struct timeval before_hb, after_hb;
855         unsigned int elapsed_msec;
856
857         mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread running\n");
858
859         set_user_nice(current, -20);
860
861         while (!kthread_should_stop() && !reg->hr_unclean_stop) {
862                 /* We track the time spent inside
863                  * o2hb_do_disk_heartbeat so that we avoid more than
864                  * hr_timeout_ms between disk writes. On busy systems
865                  * this should result in a heartbeat which is less
866                  * likely to time itself out. */
867                 do_gettimeofday(&before_hb);
868
869                 i = 0;
870                 do {
871                         ret = o2hb_do_disk_heartbeat(reg);
872                 } while (ret && ++i < 2);
873
874                 do_gettimeofday(&after_hb);
875                 elapsed_msec = o2hb_elapsed_msecs(&before_hb, &after_hb);
876
877                 mlog(0, "start = %lu.%lu, end = %lu.%lu, msec = %u\n",
878                      before_hb.tv_sec, (unsigned long) before_hb.tv_usec,
879                      after_hb.tv_sec, (unsigned long) after_hb.tv_usec,
880                      elapsed_msec);
881
882                 if (elapsed_msec < reg->hr_timeout_ms) {
883                         /* the kthread api has blocked signals for us so no
884                          * need to record the return value. */
885                         msleep_interruptible(reg->hr_timeout_ms - elapsed_msec);
886                 }
887         }
888
889         o2hb_disarm_write_timeout(reg);
890
891         /* unclean stop is only used in very bad situation */
892         for(i = 0; !reg->hr_unclean_stop && i < reg->hr_blocks; i++)
893                 o2hb_shutdown_slot(&reg->hr_slots[i]);
894
895         /* Explicit down notification - avoid forcing the other nodes
896          * to timeout on this region when we could just as easily
897          * write a clear generation - thus indicating to them that
898          * this node has left this region.
899          *
900          * XXX: Should we skip this on unclean_stop? */
901         o2hb_prepare_block(reg, 0);
902         ret = o2hb_issue_node_write(reg, &write_wc);
903         if (ret == 0) {
904                 o2hb_wait_on_io(reg, &write_wc);
905         } else {
906                 mlog_errno(ret);
907         }
908
909         mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread exiting\n");
910
911         return 0;
912 }
913
914 #ifdef CONFIG_DEBUG_FS
915 static int o2hb_debug_open(struct inode *inode, struct file *file)
916 {
917         unsigned long map[BITS_TO_LONGS(O2NM_MAX_NODES)];
918         char *buf = NULL;
919         int i = -1;
920         int out = 0;
921
922         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
923         if (!buf)
924                 goto bail;
925
926         o2hb_fill_node_map(map, sizeof(map));
927
928         while ((i = find_next_bit(map, O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES)
929                 out += snprintf(buf + out, PAGE_SIZE - out, "%d ", i);
930         out += snprintf(buf + out, PAGE_SIZE - out, "\n");
931
932         i_size_write(inode, out);
933
934         file->private_data = buf;
935
936         return 0;
937 bail:
938         return -ENOMEM;
939 }
940
941 static int o2hb_debug_release(struct inode *inode, struct file *file)
942 {
943         kfree(file->private_data);
944         return 0;
945 }
946
947 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
948                                  size_t nbytes, loff_t *ppos)
949 {
950         return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
951                                        i_size_read(file->f_mapping->host));
952 }
953 #else
954 static int o2hb_debug_open(struct inode *inode, struct file *file)
955 {
956         return 0;
957 }
958 static int o2hb_debug_release(struct inode *inode, struct file *file)
959 {
960         return 0;
961 }
962 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
963                                size_t nbytes, loff_t *ppos)
964 {
965         return 0;
966 }
967 #endif  /* CONFIG_DEBUG_FS */
968
969 static struct file_operations o2hb_debug_fops = {
970         .open =         o2hb_debug_open,
971         .release =      o2hb_debug_release,
972         .read =         o2hb_debug_read,
973         .llseek =       generic_file_llseek,
974 };
975
976 void o2hb_exit(void)
977 {
978         if (o2hb_debug_livenodes)
979                 debugfs_remove(o2hb_debug_livenodes);
980         if (o2hb_debug_dir)
981                 debugfs_remove(o2hb_debug_dir);
982 }
983
984 int o2hb_init(void)
985 {
986         int i;
987
988         for (i = 0; i < ARRAY_SIZE(o2hb_callbacks); i++)
989                 INIT_LIST_HEAD(&o2hb_callbacks[i].list);
990
991         for (i = 0; i < ARRAY_SIZE(o2hb_live_slots); i++)
992                 INIT_LIST_HEAD(&o2hb_live_slots[i]);
993
994         INIT_LIST_HEAD(&o2hb_node_events);
995
996         memset(o2hb_live_node_bitmap, 0, sizeof(o2hb_live_node_bitmap));
997
998         o2hb_debug_dir = debugfs_create_dir(O2HB_DEBUG_DIR, NULL);
999         if (!o2hb_debug_dir) {
1000                 mlog_errno(-ENOMEM);
1001                 return -ENOMEM;
1002         }
1003
1004         o2hb_debug_livenodes = debugfs_create_file(O2HB_DEBUG_LIVENODES,
1005                                                    S_IFREG|S_IRUSR,
1006                                                    o2hb_debug_dir, NULL,
1007                                                    &o2hb_debug_fops);
1008         if (!o2hb_debug_livenodes) {
1009                 mlog_errno(-ENOMEM);
1010                 debugfs_remove(o2hb_debug_dir);
1011                 return -ENOMEM;
1012         }
1013
1014         return 0;
1015 }
1016
1017 /* if we're already in a callback then we're already serialized by the sem */
1018 static void o2hb_fill_node_map_from_callback(unsigned long *map,
1019                                              unsigned bytes)
1020 {
1021         BUG_ON(bytes < (BITS_TO_LONGS(O2NM_MAX_NODES) * sizeof(unsigned long)));
1022
1023         memcpy(map, &o2hb_live_node_bitmap, bytes);
1024 }
1025
1026 /*
1027  * get a map of all nodes that are heartbeating in any regions
1028  */
1029 void o2hb_fill_node_map(unsigned long *map, unsigned bytes)
1030 {
1031         /* callers want to serialize this map and callbacks so that they
1032          * can trust that they don't miss nodes coming to the party */
1033         down_read(&o2hb_callback_sem);
1034         spin_lock(&o2hb_live_lock);
1035         o2hb_fill_node_map_from_callback(map, bytes);
1036         spin_unlock(&o2hb_live_lock);
1037         up_read(&o2hb_callback_sem);
1038 }
1039 EXPORT_SYMBOL_GPL(o2hb_fill_node_map);
1040
1041 /*
1042  * heartbeat configfs bits.  The heartbeat set is a default set under
1043  * the cluster set in nodemanager.c.
1044  */
1045
1046 static struct o2hb_region *to_o2hb_region(struct config_item *item)
1047 {
1048         return item ? container_of(item, struct o2hb_region, hr_item) : NULL;
1049 }
1050
1051 /* drop_item only drops its ref after killing the thread, nothing should
1052  * be using the region anymore.  this has to clean up any state that
1053  * attributes might have built up. */
1054 static void o2hb_region_release(struct config_item *item)
1055 {
1056         int i;
1057         struct page *page;
1058         struct o2hb_region *reg = to_o2hb_region(item);
1059
1060         if (reg->hr_tmp_block)
1061                 kfree(reg->hr_tmp_block);
1062
1063         if (reg->hr_slot_data) {
1064                 for (i = 0; i < reg->hr_num_pages; i++) {
1065                         page = reg->hr_slot_data[i];
1066                         if (page)
1067                                 __free_page(page);
1068                 }
1069                 kfree(reg->hr_slot_data);
1070         }
1071
1072         if (reg->hr_bdev)
1073                 blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1074
1075         if (reg->hr_slots)
1076                 kfree(reg->hr_slots);
1077
1078         spin_lock(&o2hb_live_lock);
1079         list_del(&reg->hr_all_item);
1080         spin_unlock(&o2hb_live_lock);
1081
1082         kfree(reg);
1083 }
1084
1085 static int o2hb_read_block_input(struct o2hb_region *reg,
1086                                  const char *page,
1087                                  size_t count,
1088                                  unsigned long *ret_bytes,
1089                                  unsigned int *ret_bits)
1090 {
1091         unsigned long bytes;
1092         char *p = (char *)page;
1093
1094         bytes = simple_strtoul(p, &p, 0);
1095         if (!p || (*p && (*p != '\n')))
1096                 return -EINVAL;
1097
1098         /* Heartbeat and fs min / max block sizes are the same. */
1099         if (bytes > 4096 || bytes < 512)
1100                 return -ERANGE;
1101         if (hweight16(bytes) != 1)
1102                 return -EINVAL;
1103
1104         if (ret_bytes)
1105                 *ret_bytes = bytes;
1106         if (ret_bits)
1107                 *ret_bits = ffs(bytes) - 1;
1108
1109         return 0;
1110 }
1111
1112 static ssize_t o2hb_region_block_bytes_read(struct o2hb_region *reg,
1113                                             char *page)
1114 {
1115         return sprintf(page, "%u\n", reg->hr_block_bytes);
1116 }
1117
1118 static ssize_t o2hb_region_block_bytes_write(struct o2hb_region *reg,
1119                                              const char *page,
1120                                              size_t count)
1121 {
1122         int status;
1123         unsigned long block_bytes;
1124         unsigned int block_bits;
1125
1126         if (reg->hr_bdev)
1127                 return -EINVAL;
1128
1129         status = o2hb_read_block_input(reg, page, count,
1130                                        &block_bytes, &block_bits);
1131         if (status)
1132                 return status;
1133
1134         reg->hr_block_bytes = (unsigned int)block_bytes;
1135         reg->hr_block_bits = block_bits;
1136
1137         return count;
1138 }
1139
1140 static ssize_t o2hb_region_start_block_read(struct o2hb_region *reg,
1141                                             char *page)
1142 {
1143         return sprintf(page, "%llu\n", reg->hr_start_block);
1144 }
1145
1146 static ssize_t o2hb_region_start_block_write(struct o2hb_region *reg,
1147                                              const char *page,
1148                                              size_t count)
1149 {
1150         unsigned long long tmp;
1151         char *p = (char *)page;
1152
1153         if (reg->hr_bdev)
1154                 return -EINVAL;
1155
1156         tmp = simple_strtoull(p, &p, 0);
1157         if (!p || (*p && (*p != '\n')))
1158                 return -EINVAL;
1159
1160         reg->hr_start_block = tmp;
1161
1162         return count;
1163 }
1164
1165 static ssize_t o2hb_region_blocks_read(struct o2hb_region *reg,
1166                                        char *page)
1167 {
1168         return sprintf(page, "%d\n", reg->hr_blocks);
1169 }
1170
1171 static ssize_t o2hb_region_blocks_write(struct o2hb_region *reg,
1172                                         const char *page,
1173                                         size_t count)
1174 {
1175         unsigned long tmp;
1176         char *p = (char *)page;
1177
1178         if (reg->hr_bdev)
1179                 return -EINVAL;
1180
1181         tmp = simple_strtoul(p, &p, 0);
1182         if (!p || (*p && (*p != '\n')))
1183                 return -EINVAL;
1184
1185         if (tmp > O2NM_MAX_NODES || tmp == 0)
1186                 return -ERANGE;
1187
1188         reg->hr_blocks = (unsigned int)tmp;
1189
1190         return count;
1191 }
1192
1193 static ssize_t o2hb_region_dev_read(struct o2hb_region *reg,
1194                                     char *page)
1195 {
1196         unsigned int ret = 0;
1197
1198         if (reg->hr_bdev)
1199                 ret = sprintf(page, "%s\n", reg->hr_dev_name);
1200
1201         return ret;
1202 }
1203
1204 static void o2hb_init_region_params(struct o2hb_region *reg)
1205 {
1206         reg->hr_slots_per_page = PAGE_CACHE_SIZE >> reg->hr_block_bits;
1207         reg->hr_timeout_ms = O2HB_REGION_TIMEOUT_MS;
1208
1209         mlog(ML_HEARTBEAT, "hr_start_block = %llu, hr_blocks = %u\n",
1210              reg->hr_start_block, reg->hr_blocks);
1211         mlog(ML_HEARTBEAT, "hr_block_bytes = %u, hr_block_bits = %u\n",
1212              reg->hr_block_bytes, reg->hr_block_bits);
1213         mlog(ML_HEARTBEAT, "hr_timeout_ms = %u\n", reg->hr_timeout_ms);
1214         mlog(ML_HEARTBEAT, "dead threshold = %u\n", o2hb_dead_threshold);
1215 }
1216
1217 static int o2hb_map_slot_data(struct o2hb_region *reg)
1218 {
1219         int i, j;
1220         unsigned int last_slot;
1221         unsigned int spp = reg->hr_slots_per_page;
1222         struct page *page;
1223         char *raw;
1224         struct o2hb_disk_slot *slot;
1225
1226         reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL);
1227         if (reg->hr_tmp_block == NULL) {
1228                 mlog_errno(-ENOMEM);
1229                 return -ENOMEM;
1230         }
1231
1232         reg->hr_slots = kcalloc(reg->hr_blocks,
1233                                 sizeof(struct o2hb_disk_slot), GFP_KERNEL);
1234         if (reg->hr_slots == NULL) {
1235                 mlog_errno(-ENOMEM);
1236                 return -ENOMEM;
1237         }
1238
1239         for(i = 0; i < reg->hr_blocks; i++) {
1240                 slot = &reg->hr_slots[i];
1241                 slot->ds_node_num = i;
1242                 INIT_LIST_HEAD(&slot->ds_live_item);
1243                 slot->ds_raw_block = NULL;
1244         }
1245
1246         reg->hr_num_pages = (reg->hr_blocks + spp - 1) / spp;
1247         mlog(ML_HEARTBEAT, "Going to require %u pages to cover %u blocks "
1248                            "at %u blocks per page\n",
1249              reg->hr_num_pages, reg->hr_blocks, spp);
1250
1251         reg->hr_slot_data = kcalloc(reg->hr_num_pages, sizeof(struct page *),
1252                                     GFP_KERNEL);
1253         if (!reg->hr_slot_data) {
1254                 mlog_errno(-ENOMEM);
1255                 return -ENOMEM;
1256         }
1257
1258         for(i = 0; i < reg->hr_num_pages; i++) {
1259                 page = alloc_page(GFP_KERNEL);
1260                 if (!page) {
1261                         mlog_errno(-ENOMEM);
1262                         return -ENOMEM;
1263                 }
1264
1265                 reg->hr_slot_data[i] = page;
1266
1267                 last_slot = i * spp;
1268                 raw = page_address(page);
1269                 for (j = 0;
1270                      (j < spp) && ((j + last_slot) < reg->hr_blocks);
1271                      j++) {
1272                         BUG_ON((j + last_slot) >= reg->hr_blocks);
1273
1274                         slot = &reg->hr_slots[j + last_slot];
1275                         slot->ds_raw_block =
1276                                 (struct o2hb_disk_heartbeat_block *) raw;
1277
1278                         raw += reg->hr_block_bytes;
1279                 }
1280         }
1281
1282         return 0;
1283 }
1284
1285 /* Read in all the slots available and populate the tracking
1286  * structures so that we can start with a baseline idea of what's
1287  * there. */
1288 static int o2hb_populate_slot_data(struct o2hb_region *reg)
1289 {
1290         int ret, i;
1291         struct o2hb_disk_slot *slot;
1292         struct o2hb_disk_heartbeat_block *hb_block;
1293
1294         mlog_entry_void();
1295
1296         ret = o2hb_read_slots(reg, reg->hr_blocks);
1297         if (ret) {
1298                 mlog_errno(ret);
1299                 goto out;
1300         }
1301
1302         /* We only want to get an idea of the values initially in each
1303          * slot, so we do no verification - o2hb_check_slot will
1304          * actually determine if each configured slot is valid and
1305          * whether any values have changed. */
1306         for(i = 0; i < reg->hr_blocks; i++) {
1307                 slot = &reg->hr_slots[i];
1308                 hb_block = (struct o2hb_disk_heartbeat_block *) slot->ds_raw_block;
1309
1310                 /* Only fill the values that o2hb_check_slot uses to
1311                  * determine changing slots */
1312                 slot->ds_last_time = le64_to_cpu(hb_block->hb_seq);
1313                 slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
1314         }
1315
1316 out:
1317         mlog_exit(ret);
1318         return ret;
1319 }
1320
1321 /* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */
1322 static ssize_t o2hb_region_dev_write(struct o2hb_region *reg,
1323                                      const char *page,
1324                                      size_t count)
1325 {
1326         struct task_struct *hb_task;
1327         long fd;
1328         int sectsize;
1329         char *p = (char *)page;
1330         struct file *filp = NULL;
1331         struct inode *inode = NULL;
1332         ssize_t ret = -EINVAL;
1333
1334         if (reg->hr_bdev)
1335                 goto out;
1336
1337         /* We can't heartbeat without having had our node number
1338          * configured yet. */
1339         if (o2nm_this_node() == O2NM_MAX_NODES)
1340                 goto out;
1341
1342         fd = simple_strtol(p, &p, 0);
1343         if (!p || (*p && (*p != '\n')))
1344                 goto out;
1345
1346         if (fd < 0 || fd >= INT_MAX)
1347                 goto out;
1348
1349         filp = fget(fd);
1350         if (filp == NULL)
1351                 goto out;
1352
1353         if (reg->hr_blocks == 0 || reg->hr_start_block == 0 ||
1354             reg->hr_block_bytes == 0)
1355                 goto out;
1356
1357         inode = igrab(filp->f_mapping->host);
1358         if (inode == NULL)
1359                 goto out;
1360
1361         if (!S_ISBLK(inode->i_mode))
1362                 goto out;
1363
1364         reg->hr_bdev = I_BDEV(filp->f_mapping->host);
1365         ret = blkdev_get(reg->hr_bdev, FMODE_WRITE | FMODE_READ);
1366         if (ret) {
1367                 reg->hr_bdev = NULL;
1368                 goto out;
1369         }
1370         inode = NULL;
1371
1372         bdevname(reg->hr_bdev, reg->hr_dev_name);
1373
1374         sectsize = bdev_hardsect_size(reg->hr_bdev);
1375         if (sectsize != reg->hr_block_bytes) {
1376                 mlog(ML_ERROR,
1377                      "blocksize %u incorrect for device, expected %d",
1378                      reg->hr_block_bytes, sectsize);
1379                 ret = -EINVAL;
1380                 goto out;
1381         }
1382
1383         o2hb_init_region_params(reg);
1384
1385         /* Generation of zero is invalid */
1386         do {
1387                 get_random_bytes(&reg->hr_generation,
1388                                  sizeof(reg->hr_generation));
1389         } while (reg->hr_generation == 0);
1390
1391         ret = o2hb_map_slot_data(reg);
1392         if (ret) {
1393                 mlog_errno(ret);
1394                 goto out;
1395         }
1396
1397         ret = o2hb_populate_slot_data(reg);
1398         if (ret) {
1399                 mlog_errno(ret);
1400                 goto out;
1401         }
1402
1403         INIT_DELAYED_WORK(&reg->hr_write_timeout_work, o2hb_write_timeout);
1404
1405         /*
1406          * A node is considered live after it has beat LIVE_THRESHOLD
1407          * times.  We're not steady until we've given them a chance
1408          * _after_ our first read.
1409          */
1410         atomic_set(&reg->hr_steady_iterations, O2HB_LIVE_THRESHOLD + 1);
1411
1412         hb_task = kthread_run(o2hb_thread, reg, "o2hb-%s",
1413                               reg->hr_item.ci_name);
1414         if (IS_ERR(hb_task)) {
1415                 ret = PTR_ERR(hb_task);
1416                 mlog_errno(ret);
1417                 goto out;
1418         }
1419
1420         spin_lock(&o2hb_live_lock);
1421         reg->hr_task = hb_task;
1422         spin_unlock(&o2hb_live_lock);
1423
1424         ret = wait_event_interruptible(o2hb_steady_queue,
1425                                 atomic_read(&reg->hr_steady_iterations) == 0);
1426         if (ret) {
1427                 /* We got interrupted (hello ptrace!).  Clean up */
1428                 spin_lock(&o2hb_live_lock);
1429                 hb_task = reg->hr_task;
1430                 reg->hr_task = NULL;
1431                 spin_unlock(&o2hb_live_lock);
1432
1433                 if (hb_task)
1434                         kthread_stop(hb_task);
1435                 goto out;
1436         }
1437
1438         /* Ok, we were woken.  Make sure it wasn't by drop_item() */
1439         spin_lock(&o2hb_live_lock);
1440         hb_task = reg->hr_task;
1441         spin_unlock(&o2hb_live_lock);
1442
1443         if (hb_task)
1444                 ret = count;
1445         else
1446                 ret = -EIO;
1447
1448 out:
1449         if (filp)
1450                 fput(filp);
1451         if (inode)
1452                 iput(inode);
1453         if (ret < 0) {
1454                 if (reg->hr_bdev) {
1455                         blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1456                         reg->hr_bdev = NULL;
1457                 }
1458         }
1459         return ret;
1460 }
1461
1462 static ssize_t o2hb_region_pid_read(struct o2hb_region *reg,
1463                                       char *page)
1464 {
1465         pid_t pid = 0;
1466
1467         spin_lock(&o2hb_live_lock);
1468         if (reg->hr_task)
1469                 pid = task_pid_nr(reg->hr_task);
1470         spin_unlock(&o2hb_live_lock);
1471
1472         if (!pid)
1473                 return 0;
1474
1475         return sprintf(page, "%u\n", pid);
1476 }
1477
1478 struct o2hb_region_attribute {
1479         struct configfs_attribute attr;
1480         ssize_t (*show)(struct o2hb_region *, char *);
1481         ssize_t (*store)(struct o2hb_region *, const char *, size_t);
1482 };
1483
1484 static struct o2hb_region_attribute o2hb_region_attr_block_bytes = {
1485         .attr   = { .ca_owner = THIS_MODULE,
1486                     .ca_name = "block_bytes",
1487                     .ca_mode = S_IRUGO | S_IWUSR },
1488         .show   = o2hb_region_block_bytes_read,
1489         .store  = o2hb_region_block_bytes_write,
1490 };
1491
1492 static struct o2hb_region_attribute o2hb_region_attr_start_block = {
1493         .attr   = { .ca_owner = THIS_MODULE,
1494                     .ca_name = "start_block",
1495                     .ca_mode = S_IRUGO | S_IWUSR },
1496         .show   = o2hb_region_start_block_read,
1497         .store  = o2hb_region_start_block_write,
1498 };
1499
1500 static struct o2hb_region_attribute o2hb_region_attr_blocks = {
1501         .attr   = { .ca_owner = THIS_MODULE,
1502                     .ca_name = "blocks",
1503                     .ca_mode = S_IRUGO | S_IWUSR },
1504         .show   = o2hb_region_blocks_read,
1505         .store  = o2hb_region_blocks_write,
1506 };
1507
1508 static struct o2hb_region_attribute o2hb_region_attr_dev = {
1509         .attr   = { .ca_owner = THIS_MODULE,
1510                     .ca_name = "dev",
1511                     .ca_mode = S_IRUGO | S_IWUSR },
1512         .show   = o2hb_region_dev_read,
1513         .store  = o2hb_region_dev_write,
1514 };
1515
1516 static struct o2hb_region_attribute o2hb_region_attr_pid = {
1517        .attr   = { .ca_owner = THIS_MODULE,
1518                    .ca_name = "pid",
1519                    .ca_mode = S_IRUGO | S_IRUSR },
1520        .show   = o2hb_region_pid_read,
1521 };
1522
1523 static struct configfs_attribute *o2hb_region_attrs[] = {
1524         &o2hb_region_attr_block_bytes.attr,
1525         &o2hb_region_attr_start_block.attr,
1526         &o2hb_region_attr_blocks.attr,
1527         &o2hb_region_attr_dev.attr,
1528         &o2hb_region_attr_pid.attr,
1529         NULL,
1530 };
1531
1532 static ssize_t o2hb_region_show(struct config_item *item,
1533                                 struct configfs_attribute *attr,
1534                                 char *page)
1535 {
1536         struct o2hb_region *reg = to_o2hb_region(item);
1537         struct o2hb_region_attribute *o2hb_region_attr =
1538                 container_of(attr, struct o2hb_region_attribute, attr);
1539         ssize_t ret = 0;
1540
1541         if (o2hb_region_attr->show)
1542                 ret = o2hb_region_attr->show(reg, page);
1543         return ret;
1544 }
1545
1546 static ssize_t o2hb_region_store(struct config_item *item,
1547                                  struct configfs_attribute *attr,
1548                                  const char *page, size_t count)
1549 {
1550         struct o2hb_region *reg = to_o2hb_region(item);
1551         struct o2hb_region_attribute *o2hb_region_attr =
1552                 container_of(attr, struct o2hb_region_attribute, attr);
1553         ssize_t ret = -EINVAL;
1554
1555         if (o2hb_region_attr->store)
1556                 ret = o2hb_region_attr->store(reg, page, count);
1557         return ret;
1558 }
1559
1560 static struct configfs_item_operations o2hb_region_item_ops = {
1561         .release                = o2hb_region_release,
1562         .show_attribute         = o2hb_region_show,
1563         .store_attribute        = o2hb_region_store,
1564 };
1565
1566 static struct config_item_type o2hb_region_type = {
1567         .ct_item_ops    = &o2hb_region_item_ops,
1568         .ct_attrs       = o2hb_region_attrs,
1569         .ct_owner       = THIS_MODULE,
1570 };
1571
1572 /* heartbeat set */
1573
1574 struct o2hb_heartbeat_group {
1575         struct config_group hs_group;
1576         /* some stuff? */
1577 };
1578
1579 static struct o2hb_heartbeat_group *to_o2hb_heartbeat_group(struct config_group *group)
1580 {
1581         return group ?
1582                 container_of(group, struct o2hb_heartbeat_group, hs_group)
1583                 : NULL;
1584 }
1585
1586 static struct config_item *o2hb_heartbeat_group_make_item(struct config_group *group,
1587                                                           const char *name)
1588 {
1589         struct o2hb_region *reg = NULL;
1590
1591         reg = kzalloc(sizeof(struct o2hb_region), GFP_KERNEL);
1592         if (reg == NULL)
1593                 return ERR_PTR(-ENOMEM);
1594
1595         config_item_init_type_name(&reg->hr_item, name, &o2hb_region_type);
1596
1597         spin_lock(&o2hb_live_lock);
1598         list_add_tail(&reg->hr_all_item, &o2hb_all_regions);
1599         spin_unlock(&o2hb_live_lock);
1600
1601         return &reg->hr_item;
1602 }
1603
1604 static void o2hb_heartbeat_group_drop_item(struct config_group *group,
1605                                            struct config_item *item)
1606 {
1607         struct task_struct *hb_task;
1608         struct o2hb_region *reg = to_o2hb_region(item);
1609
1610         /* stop the thread when the user removes the region dir */
1611         spin_lock(&o2hb_live_lock);
1612         hb_task = reg->hr_task;
1613         reg->hr_task = NULL;
1614         spin_unlock(&o2hb_live_lock);
1615
1616         if (hb_task)
1617                 kthread_stop(hb_task);
1618
1619         /*
1620          * If we're racing a dev_write(), we need to wake them.  They will
1621          * check reg->hr_task
1622          */
1623         if (atomic_read(&reg->hr_steady_iterations) != 0) {
1624                 atomic_set(&reg->hr_steady_iterations, 0);
1625                 wake_up(&o2hb_steady_queue);
1626         }
1627
1628         config_item_put(item);
1629 }
1630
1631 struct o2hb_heartbeat_group_attribute {
1632         struct configfs_attribute attr;
1633         ssize_t (*show)(struct o2hb_heartbeat_group *, char *);
1634         ssize_t (*store)(struct o2hb_heartbeat_group *, const char *, size_t);
1635 };
1636
1637 static ssize_t o2hb_heartbeat_group_show(struct config_item *item,
1638                                          struct configfs_attribute *attr,
1639                                          char *page)
1640 {
1641         struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
1642         struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
1643                 container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
1644         ssize_t ret = 0;
1645
1646         if (o2hb_heartbeat_group_attr->show)
1647                 ret = o2hb_heartbeat_group_attr->show(reg, page);
1648         return ret;
1649 }
1650
1651 static ssize_t o2hb_heartbeat_group_store(struct config_item *item,
1652                                           struct configfs_attribute *attr,
1653                                           const char *page, size_t count)
1654 {
1655         struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
1656         struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
1657                 container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
1658         ssize_t ret = -EINVAL;
1659
1660         if (o2hb_heartbeat_group_attr->store)
1661                 ret = o2hb_heartbeat_group_attr->store(reg, page, count);
1662         return ret;
1663 }
1664
1665 static ssize_t o2hb_heartbeat_group_threshold_show(struct o2hb_heartbeat_group *group,
1666                                                      char *page)
1667 {
1668         return sprintf(page, "%u\n", o2hb_dead_threshold);
1669 }
1670
1671 static ssize_t o2hb_heartbeat_group_threshold_store(struct o2hb_heartbeat_group *group,
1672                                                     const char *page,
1673                                                     size_t count)
1674 {
1675         unsigned long tmp;
1676         char *p = (char *)page;
1677
1678         tmp = simple_strtoul(p, &p, 10);
1679         if (!p || (*p && (*p != '\n')))
1680                 return -EINVAL;
1681
1682         /* this will validate ranges for us. */
1683         o2hb_dead_threshold_set((unsigned int) tmp);
1684
1685         return count;
1686 }
1687
1688 static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_threshold = {
1689         .attr   = { .ca_owner = THIS_MODULE,
1690                     .ca_name = "dead_threshold",
1691                     .ca_mode = S_IRUGO | S_IWUSR },
1692         .show   = o2hb_heartbeat_group_threshold_show,
1693         .store  = o2hb_heartbeat_group_threshold_store,
1694 };
1695
1696 static struct configfs_attribute *o2hb_heartbeat_group_attrs[] = {
1697         &o2hb_heartbeat_group_attr_threshold.attr,
1698         NULL,
1699 };
1700
1701 static struct configfs_item_operations o2hb_hearbeat_group_item_ops = {
1702         .show_attribute         = o2hb_heartbeat_group_show,
1703         .store_attribute        = o2hb_heartbeat_group_store,
1704 };
1705
1706 static struct configfs_group_operations o2hb_heartbeat_group_group_ops = {
1707         .make_item      = o2hb_heartbeat_group_make_item,
1708         .drop_item      = o2hb_heartbeat_group_drop_item,
1709 };
1710
1711 static struct config_item_type o2hb_heartbeat_group_type = {
1712         .ct_group_ops   = &o2hb_heartbeat_group_group_ops,
1713         .ct_item_ops    = &o2hb_hearbeat_group_item_ops,
1714         .ct_attrs       = o2hb_heartbeat_group_attrs,
1715         .ct_owner       = THIS_MODULE,
1716 };
1717
1718 /* this is just here to avoid touching group in heartbeat.h which the
1719  * entire damn world #includes */
1720 struct config_group *o2hb_alloc_hb_set(void)
1721 {
1722         struct o2hb_heartbeat_group *hs = NULL;
1723         struct config_group *ret = NULL;
1724
1725         hs = kzalloc(sizeof(struct o2hb_heartbeat_group), GFP_KERNEL);
1726         if (hs == NULL)
1727                 goto out;
1728
1729         config_group_init_type_name(&hs->hs_group, "heartbeat",
1730                                     &o2hb_heartbeat_group_type);
1731
1732         ret = &hs->hs_group;
1733 out:
1734         if (ret == NULL)
1735                 kfree(hs);
1736         return ret;
1737 }
1738
1739 void o2hb_free_hb_set(struct config_group *group)
1740 {
1741         struct o2hb_heartbeat_group *hs = to_o2hb_heartbeat_group(group);
1742         kfree(hs);
1743 }
1744
1745 /* hb callback registration and issueing */
1746
1747 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type)
1748 {
1749         if (type == O2HB_NUM_CB)
1750                 return ERR_PTR(-EINVAL);
1751
1752         return &o2hb_callbacks[type];
1753 }
1754
1755 void o2hb_setup_callback(struct o2hb_callback_func *hc,
1756                          enum o2hb_callback_type type,
1757                          o2hb_cb_func *func,
1758                          void *data,
1759                          int priority)
1760 {
1761         INIT_LIST_HEAD(&hc->hc_item);
1762         hc->hc_func = func;
1763         hc->hc_data = data;
1764         hc->hc_priority = priority;
1765         hc->hc_type = type;
1766         hc->hc_magic = O2HB_CB_MAGIC;
1767 }
1768 EXPORT_SYMBOL_GPL(o2hb_setup_callback);
1769
1770 static struct o2hb_region *o2hb_find_region(const char *region_uuid)
1771 {
1772         struct o2hb_region *p, *reg = NULL;
1773
1774         assert_spin_locked(&o2hb_live_lock);
1775
1776         list_for_each_entry(p, &o2hb_all_regions, hr_all_item) {
1777                 if (!strcmp(region_uuid, config_item_name(&p->hr_item))) {
1778                         reg = p;
1779                         break;
1780                 }
1781         }
1782
1783         return reg;
1784 }
1785
1786 static int o2hb_region_get(const char *region_uuid)
1787 {
1788         int ret = 0;
1789         struct o2hb_region *reg;
1790
1791         spin_lock(&o2hb_live_lock);
1792
1793         reg = o2hb_find_region(region_uuid);
1794         if (!reg)
1795                 ret = -ENOENT;
1796         spin_unlock(&o2hb_live_lock);
1797
1798         if (ret)
1799                 goto out;
1800
1801         ret = o2nm_depend_this_node();
1802         if (ret)
1803                 goto out;
1804
1805         ret = o2nm_depend_item(&reg->hr_item);
1806         if (ret)
1807                 o2nm_undepend_this_node();
1808
1809 out:
1810         return ret;
1811 }
1812
1813 static void o2hb_region_put(const char *region_uuid)
1814 {
1815         struct o2hb_region *reg;
1816
1817         spin_lock(&o2hb_live_lock);
1818
1819         reg = o2hb_find_region(region_uuid);
1820
1821         spin_unlock(&o2hb_live_lock);
1822
1823         if (reg) {
1824                 o2nm_undepend_item(&reg->hr_item);
1825                 o2nm_undepend_this_node();
1826         }
1827 }
1828
1829 int o2hb_register_callback(const char *region_uuid,
1830                            struct o2hb_callback_func *hc)
1831 {
1832         struct o2hb_callback_func *tmp;
1833         struct list_head *iter;
1834         struct o2hb_callback *hbcall;
1835         int ret;
1836
1837         BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
1838         BUG_ON(!list_empty(&hc->hc_item));
1839
1840         hbcall = hbcall_from_type(hc->hc_type);
1841         if (IS_ERR(hbcall)) {
1842                 ret = PTR_ERR(hbcall);
1843                 goto out;
1844         }
1845
1846         if (region_uuid) {
1847                 ret = o2hb_region_get(region_uuid);
1848                 if (ret)
1849                         goto out;
1850         }
1851
1852         down_write(&o2hb_callback_sem);
1853
1854         list_for_each(iter, &hbcall->list) {
1855                 tmp = list_entry(iter, struct o2hb_callback_func, hc_item);
1856                 if (hc->hc_priority < tmp->hc_priority) {
1857                         list_add_tail(&hc->hc_item, iter);
1858                         break;
1859                 }
1860         }
1861         if (list_empty(&hc->hc_item))
1862                 list_add_tail(&hc->hc_item, &hbcall->list);
1863
1864         up_write(&o2hb_callback_sem);
1865         ret = 0;
1866 out:
1867         mlog(ML_HEARTBEAT, "returning %d on behalf of %p for funcs %p\n",
1868              ret, __builtin_return_address(0), hc);
1869         return ret;
1870 }
1871 EXPORT_SYMBOL_GPL(o2hb_register_callback);
1872
1873 void o2hb_unregister_callback(const char *region_uuid,
1874                               struct o2hb_callback_func *hc)
1875 {
1876         BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
1877
1878         mlog(ML_HEARTBEAT, "on behalf of %p for funcs %p\n",
1879              __builtin_return_address(0), hc);
1880
1881         /* XXX Can this happen _with_ a region reference? */
1882         if (list_empty(&hc->hc_item))
1883                 return;
1884
1885         if (region_uuid)
1886                 o2hb_region_put(region_uuid);
1887
1888         down_write(&o2hb_callback_sem);
1889
1890         list_del_init(&hc->hc_item);
1891
1892         up_write(&o2hb_callback_sem);
1893 }
1894 EXPORT_SYMBOL_GPL(o2hb_unregister_callback);
1895
1896 int o2hb_check_node_heartbeating(u8 node_num)
1897 {
1898         unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1899
1900         o2hb_fill_node_map(testing_map, sizeof(testing_map));
1901         if (!test_bit(node_num, testing_map)) {
1902                 mlog(ML_HEARTBEAT,
1903                      "node (%u) does not have heartbeating enabled.\n",
1904                      node_num);
1905                 return 0;
1906         }
1907
1908         return 1;
1909 }
1910 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating);
1911
1912 int o2hb_check_node_heartbeating_from_callback(u8 node_num)
1913 {
1914         unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1915
1916         o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
1917         if (!test_bit(node_num, testing_map)) {
1918                 mlog(ML_HEARTBEAT,
1919                      "node (%u) does not have heartbeating enabled.\n",
1920                      node_num);
1921                 return 0;
1922         }
1923
1924         return 1;
1925 }
1926 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback);
1927
1928 /* Makes sure our local node is configured with a node number, and is
1929  * heartbeating. */
1930 int o2hb_check_local_node_heartbeating(void)
1931 {
1932         u8 node_num;
1933
1934         /* if this node was set then we have networking */
1935         node_num = o2nm_this_node();
1936         if (node_num == O2NM_MAX_NODES) {
1937                 mlog(ML_HEARTBEAT, "this node has not been configured.\n");
1938                 return 0;
1939         }
1940
1941         return o2hb_check_node_heartbeating(node_num);
1942 }
1943 EXPORT_SYMBOL_GPL(o2hb_check_local_node_heartbeating);
1944
1945 /*
1946  * this is just a hack until we get the plumbing which flips file systems
1947  * read only and drops the hb ref instead of killing the node dead.
1948  */
1949 void o2hb_stop_all_regions(void)
1950 {
1951         struct o2hb_region *reg;
1952
1953         mlog(ML_ERROR, "stopping heartbeat on all active regions.\n");
1954
1955         spin_lock(&o2hb_live_lock);
1956
1957         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item)
1958                 reg->hr_unclean_stop = 1;
1959
1960         spin_unlock(&o2hb_live_lock);
1961 }
1962 EXPORT_SYMBOL_GPL(o2hb_stop_all_regions);