cfg80211: handle failed skb allocation
[cascardo/linux.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 #include <linux/slab.h>
38 #include <linux/bitmap.h>
39 #include <linux/ktime.h>
40 #include "heartbeat.h"
41 #include "tcp.h"
42 #include "nodemanager.h"
43 #include "quorum.h"
44
45 #include "masklog.h"
46
47
48 /*
49  * The first heartbeat pass had one global thread that would serialize all hb
50  * callback calls.  This global serializing sem should only be removed once
51  * we've made sure that all callees can deal with being called concurrently
52  * from multiple hb region threads.
53  */
54 static DECLARE_RWSEM(o2hb_callback_sem);
55
56 /*
57  * multiple hb threads are watching multiple regions.  A node is live
58  * whenever any of the threads sees activity from the node in its region.
59  */
60 static DEFINE_SPINLOCK(o2hb_live_lock);
61 static struct list_head o2hb_live_slots[O2NM_MAX_NODES];
62 static unsigned long o2hb_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
63 static LIST_HEAD(o2hb_node_events);
64 static DECLARE_WAIT_QUEUE_HEAD(o2hb_steady_queue);
65
66 /*
67  * In global heartbeat, we maintain a series of region bitmaps.
68  *      - o2hb_region_bitmap allows us to limit the region number to max region.
69  *      - o2hb_live_region_bitmap tracks live regions (seen steady iterations).
70  *      - o2hb_quorum_region_bitmap tracks live regions that have seen all nodes
71  *              heartbeat on it.
72  *      - o2hb_failed_region_bitmap tracks the regions that have seen io timeouts.
73  */
74 static unsigned long o2hb_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
75 static unsigned long o2hb_live_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
76 static unsigned long o2hb_quorum_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
77 static unsigned long o2hb_failed_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
78
79 #define O2HB_DB_TYPE_LIVENODES          0
80 #define O2HB_DB_TYPE_LIVEREGIONS        1
81 #define O2HB_DB_TYPE_QUORUMREGIONS      2
82 #define O2HB_DB_TYPE_FAILEDREGIONS      3
83 #define O2HB_DB_TYPE_REGION_LIVENODES   4
84 #define O2HB_DB_TYPE_REGION_NUMBER      5
85 #define O2HB_DB_TYPE_REGION_ELAPSED_TIME        6
86 #define O2HB_DB_TYPE_REGION_PINNED      7
87 struct o2hb_debug_buf {
88         int db_type;
89         int db_size;
90         int db_len;
91         void *db_data;
92 };
93
94 static struct o2hb_debug_buf *o2hb_db_livenodes;
95 static struct o2hb_debug_buf *o2hb_db_liveregions;
96 static struct o2hb_debug_buf *o2hb_db_quorumregions;
97 static struct o2hb_debug_buf *o2hb_db_failedregions;
98
99 #define O2HB_DEBUG_DIR                  "o2hb"
100 #define O2HB_DEBUG_LIVENODES            "livenodes"
101 #define O2HB_DEBUG_LIVEREGIONS          "live_regions"
102 #define O2HB_DEBUG_QUORUMREGIONS        "quorum_regions"
103 #define O2HB_DEBUG_FAILEDREGIONS        "failed_regions"
104 #define O2HB_DEBUG_REGION_NUMBER        "num"
105 #define O2HB_DEBUG_REGION_ELAPSED_TIME  "elapsed_time_in_ms"
106 #define O2HB_DEBUG_REGION_PINNED        "pinned"
107
108 static struct dentry *o2hb_debug_dir;
109 static struct dentry *o2hb_debug_livenodes;
110 static struct dentry *o2hb_debug_liveregions;
111 static struct dentry *o2hb_debug_quorumregions;
112 static struct dentry *o2hb_debug_failedregions;
113
114 static LIST_HEAD(o2hb_all_regions);
115
116 static struct o2hb_callback {
117         struct list_head list;
118 } o2hb_callbacks[O2HB_NUM_CB];
119
120 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type);
121
122 #define O2HB_DEFAULT_BLOCK_BITS       9
123
124 enum o2hb_heartbeat_modes {
125         O2HB_HEARTBEAT_LOCAL            = 0,
126         O2HB_HEARTBEAT_GLOBAL,
127         O2HB_HEARTBEAT_NUM_MODES,
128 };
129
130 char *o2hb_heartbeat_mode_desc[O2HB_HEARTBEAT_NUM_MODES] = {
131                 "local",        /* O2HB_HEARTBEAT_LOCAL */
132                 "global",       /* O2HB_HEARTBEAT_GLOBAL */
133 };
134
135 unsigned int o2hb_dead_threshold = O2HB_DEFAULT_DEAD_THRESHOLD;
136 unsigned int o2hb_heartbeat_mode = O2HB_HEARTBEAT_LOCAL;
137
138 /*
139  * o2hb_dependent_users tracks the number of registered callbacks that depend
140  * on heartbeat. o2net and o2dlm are two entities that register this callback.
141  * However only o2dlm depends on the heartbeat. It does not want the heartbeat
142  * to stop while a dlm domain is still active.
143  */
144 unsigned int o2hb_dependent_users;
145
146 /*
147  * In global heartbeat mode, all regions are pinned if there are one or more
148  * dependent users and the quorum region count is <= O2HB_PIN_CUT_OFF. All
149  * regions are unpinned if the region count exceeds the cut off or the number
150  * of dependent users falls to zero.
151  */
152 #define O2HB_PIN_CUT_OFF                3
153
154 /*
155  * In local heartbeat mode, we assume the dlm domain name to be the same as
156  * region uuid. This is true for domains created for the file system but not
157  * necessarily true for userdlm domains. This is a known limitation.
158  *
159  * In global heartbeat mode, we pin/unpin all o2hb regions. This solution
160  * works for both file system and userdlm domains.
161  */
162 static int o2hb_region_pin(const char *region_uuid);
163 static void o2hb_region_unpin(const char *region_uuid);
164
165 /* Only sets a new threshold if there are no active regions.
166  *
167  * No locking or otherwise interesting code is required for reading
168  * o2hb_dead_threshold as it can't change once regions are active and
169  * it's not interesting to anyone until then anyway. */
170 static void o2hb_dead_threshold_set(unsigned int threshold)
171 {
172         if (threshold > O2HB_MIN_DEAD_THRESHOLD) {
173                 spin_lock(&o2hb_live_lock);
174                 if (list_empty(&o2hb_all_regions))
175                         o2hb_dead_threshold = threshold;
176                 spin_unlock(&o2hb_live_lock);
177         }
178 }
179
180 static int o2hb_global_heartbeat_mode_set(unsigned int hb_mode)
181 {
182         int ret = -1;
183
184         if (hb_mode < O2HB_HEARTBEAT_NUM_MODES) {
185                 spin_lock(&o2hb_live_lock);
186                 if (list_empty(&o2hb_all_regions)) {
187                         o2hb_heartbeat_mode = hb_mode;
188                         ret = 0;
189                 }
190                 spin_unlock(&o2hb_live_lock);
191         }
192
193         return ret;
194 }
195
196 struct o2hb_node_event {
197         struct list_head        hn_item;
198         enum o2hb_callback_type hn_event_type;
199         struct o2nm_node        *hn_node;
200         int                     hn_node_num;
201 };
202
203 struct o2hb_disk_slot {
204         struct o2hb_disk_heartbeat_block *ds_raw_block;
205         u8                      ds_node_num;
206         u64                     ds_last_time;
207         u64                     ds_last_generation;
208         u16                     ds_equal_samples;
209         u16                     ds_changed_samples;
210         struct list_head        ds_live_item;
211 };
212
213 /* each thread owns a region.. when we're asked to tear down the region
214  * we ask the thread to stop, who cleans up the region */
215 struct o2hb_region {
216         struct config_item      hr_item;
217
218         struct list_head        hr_all_item;
219         unsigned                hr_unclean_stop:1,
220                                 hr_aborted_start:1,
221                                 hr_item_pinned:1,
222                                 hr_item_dropped:1,
223                                 hr_node_deleted:1;
224
225         /* protected by the hr_callback_sem */
226         struct task_struct      *hr_task;
227
228         unsigned int            hr_blocks;
229         unsigned long long      hr_start_block;
230
231         unsigned int            hr_block_bits;
232         unsigned int            hr_block_bytes;
233
234         unsigned int            hr_slots_per_page;
235         unsigned int            hr_num_pages;
236
237         struct page             **hr_slot_data;
238         struct block_device     *hr_bdev;
239         struct o2hb_disk_slot   *hr_slots;
240
241         /* live node map of this region */
242         unsigned long           hr_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
243         unsigned int            hr_region_num;
244
245         struct dentry           *hr_debug_dir;
246         struct dentry           *hr_debug_livenodes;
247         struct dentry           *hr_debug_regnum;
248         struct dentry           *hr_debug_elapsed_time;
249         struct dentry           *hr_debug_pinned;
250         struct o2hb_debug_buf   *hr_db_livenodes;
251         struct o2hb_debug_buf   *hr_db_regnum;
252         struct o2hb_debug_buf   *hr_db_elapsed_time;
253         struct o2hb_debug_buf   *hr_db_pinned;
254
255         /* let the person setting up hb wait for it to return until it
256          * has reached a 'steady' state.  This will be fixed when we have
257          * a more complete api that doesn't lead to this sort of fragility. */
258         atomic_t                hr_steady_iterations;
259
260         /* terminate o2hb thread if it does not reach steady state
261          * (hr_steady_iterations == 0) within hr_unsteady_iterations */
262         atomic_t                hr_unsteady_iterations;
263
264         char                    hr_dev_name[BDEVNAME_SIZE];
265
266         unsigned int            hr_timeout_ms;
267
268         /* randomized as the region goes up and down so that a node
269          * recognizes a node going up and down in one iteration */
270         u64                     hr_generation;
271
272         struct delayed_work     hr_write_timeout_work;
273         unsigned long           hr_last_timeout_start;
274
275         /* Used during o2hb_check_slot to hold a copy of the block
276          * being checked because we temporarily have to zero out the
277          * crc field. */
278         struct o2hb_disk_heartbeat_block *hr_tmp_block;
279 };
280
281 struct o2hb_bio_wait_ctxt {
282         atomic_t          wc_num_reqs;
283         struct completion wc_io_complete;
284         int               wc_error;
285 };
286
287 static void o2hb_write_timeout(struct work_struct *work)
288 {
289         int failed, quorum;
290         struct o2hb_region *reg =
291                 container_of(work, struct o2hb_region,
292                              hr_write_timeout_work.work);
293
294         mlog(ML_ERROR, "Heartbeat write timeout to device %s after %u "
295              "milliseconds\n", reg->hr_dev_name,
296              jiffies_to_msecs(jiffies - reg->hr_last_timeout_start));
297
298         if (o2hb_global_heartbeat_active()) {
299                 spin_lock(&o2hb_live_lock);
300                 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
301                         set_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
302                 failed = bitmap_weight(o2hb_failed_region_bitmap,
303                                         O2NM_MAX_REGIONS);
304                 quorum = bitmap_weight(o2hb_quorum_region_bitmap,
305                                         O2NM_MAX_REGIONS);
306                 spin_unlock(&o2hb_live_lock);
307
308                 mlog(ML_HEARTBEAT, "Number of regions %d, failed regions %d\n",
309                      quorum, failed);
310
311                 /*
312                  * Fence if the number of failed regions >= half the number
313                  * of  quorum regions
314                  */
315                 if ((failed << 1) < quorum)
316                         return;
317         }
318
319         o2quo_disk_timeout();
320 }
321
322 static void o2hb_arm_write_timeout(struct o2hb_region *reg)
323 {
324         /* Arm writeout only after thread reaches steady state */
325         if (atomic_read(&reg->hr_steady_iterations) != 0)
326                 return;
327
328         mlog(ML_HEARTBEAT, "Queue write timeout for %u ms\n",
329              O2HB_MAX_WRITE_TIMEOUT_MS);
330
331         if (o2hb_global_heartbeat_active()) {
332                 spin_lock(&o2hb_live_lock);
333                 clear_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
334                 spin_unlock(&o2hb_live_lock);
335         }
336         cancel_delayed_work(&reg->hr_write_timeout_work);
337         reg->hr_last_timeout_start = jiffies;
338         schedule_delayed_work(&reg->hr_write_timeout_work,
339                               msecs_to_jiffies(O2HB_MAX_WRITE_TIMEOUT_MS));
340 }
341
342 static void o2hb_disarm_write_timeout(struct o2hb_region *reg)
343 {
344         cancel_delayed_work_sync(&reg->hr_write_timeout_work);
345 }
346
347 static inline void o2hb_bio_wait_init(struct o2hb_bio_wait_ctxt *wc)
348 {
349         atomic_set(&wc->wc_num_reqs, 1);
350         init_completion(&wc->wc_io_complete);
351         wc->wc_error = 0;
352 }
353
354 /* Used in error paths too */
355 static inline void o2hb_bio_wait_dec(struct o2hb_bio_wait_ctxt *wc,
356                                      unsigned int num)
357 {
358         /* sadly atomic_sub_and_test() isn't available on all platforms.  The
359          * good news is that the fast path only completes one at a time */
360         while(num--) {
361                 if (atomic_dec_and_test(&wc->wc_num_reqs)) {
362                         BUG_ON(num > 0);
363                         complete(&wc->wc_io_complete);
364                 }
365         }
366 }
367
368 static void o2hb_wait_on_io(struct o2hb_region *reg,
369                             struct o2hb_bio_wait_ctxt *wc)
370 {
371         o2hb_bio_wait_dec(wc, 1);
372         wait_for_completion(&wc->wc_io_complete);
373 }
374
375 static void o2hb_bio_end_io(struct bio *bio)
376 {
377         struct o2hb_bio_wait_ctxt *wc = bio->bi_private;
378
379         if (bio->bi_error) {
380                 mlog(ML_ERROR, "IO Error %d\n", bio->bi_error);
381                 wc->wc_error = bio->bi_error;
382         }
383
384         o2hb_bio_wait_dec(wc, 1);
385         bio_put(bio);
386 }
387
388 /* Setup a Bio to cover I/O against num_slots slots starting at
389  * start_slot. */
390 static struct bio *o2hb_setup_one_bio(struct o2hb_region *reg,
391                                       struct o2hb_bio_wait_ctxt *wc,
392                                       unsigned int *current_slot,
393                                       unsigned int max_slots)
394 {
395         int len, current_page;
396         unsigned int vec_len, vec_start;
397         unsigned int bits = reg->hr_block_bits;
398         unsigned int spp = reg->hr_slots_per_page;
399         unsigned int cs = *current_slot;
400         struct bio *bio;
401         struct page *page;
402
403         /* Testing has shown this allocation to take long enough under
404          * GFP_KERNEL that the local node can get fenced. It would be
405          * nicest if we could pre-allocate these bios and avoid this
406          * all together. */
407         bio = bio_alloc(GFP_ATOMIC, 16);
408         if (!bio) {
409                 mlog(ML_ERROR, "Could not alloc slots BIO!\n");
410                 bio = ERR_PTR(-ENOMEM);
411                 goto bail;
412         }
413
414         /* Must put everything in 512 byte sectors for the bio... */
415         bio->bi_iter.bi_sector = (reg->hr_start_block + cs) << (bits - 9);
416         bio->bi_bdev = reg->hr_bdev;
417         bio->bi_private = wc;
418         bio->bi_end_io = o2hb_bio_end_io;
419
420         vec_start = (cs << bits) % PAGE_SIZE;
421         while(cs < max_slots) {
422                 current_page = cs / spp;
423                 page = reg->hr_slot_data[current_page];
424
425                 vec_len = min(PAGE_SIZE - vec_start,
426                               (max_slots-cs) * (PAGE_SIZE/spp) );
427
428                 mlog(ML_HB_BIO, "page %d, vec_len = %u, vec_start = %u\n",
429                      current_page, vec_len, vec_start);
430
431                 len = bio_add_page(bio, page, vec_len, vec_start);
432                 if (len != vec_len) break;
433
434                 cs += vec_len / (PAGE_SIZE/spp);
435                 vec_start = 0;
436         }
437
438 bail:
439         *current_slot = cs;
440         return bio;
441 }
442
443 static int o2hb_read_slots(struct o2hb_region *reg,
444                            unsigned int max_slots)
445 {
446         unsigned int current_slot=0;
447         int status;
448         struct o2hb_bio_wait_ctxt wc;
449         struct bio *bio;
450
451         o2hb_bio_wait_init(&wc);
452
453         while(current_slot < max_slots) {
454                 bio = o2hb_setup_one_bio(reg, &wc, &current_slot, max_slots);
455                 if (IS_ERR(bio)) {
456                         status = PTR_ERR(bio);
457                         mlog_errno(status);
458                         goto bail_and_wait;
459                 }
460
461                 atomic_inc(&wc.wc_num_reqs);
462                 submit_bio(READ, bio);
463         }
464
465         status = 0;
466
467 bail_and_wait:
468         o2hb_wait_on_io(reg, &wc);
469         if (wc.wc_error && !status)
470                 status = wc.wc_error;
471
472         return status;
473 }
474
475 static int o2hb_issue_node_write(struct o2hb_region *reg,
476                                  struct o2hb_bio_wait_ctxt *write_wc)
477 {
478         int status;
479         unsigned int slot;
480         struct bio *bio;
481
482         o2hb_bio_wait_init(write_wc);
483
484         slot = o2nm_this_node();
485
486         bio = o2hb_setup_one_bio(reg, write_wc, &slot, slot+1);
487         if (IS_ERR(bio)) {
488                 status = PTR_ERR(bio);
489                 mlog_errno(status);
490                 goto bail;
491         }
492
493         atomic_inc(&write_wc->wc_num_reqs);
494         submit_bio(WRITE_SYNC, bio);
495
496         status = 0;
497 bail:
498         return status;
499 }
500
501 static u32 o2hb_compute_block_crc_le(struct o2hb_region *reg,
502                                      struct o2hb_disk_heartbeat_block *hb_block)
503 {
504         __le32 old_cksum;
505         u32 ret;
506
507         /* We want to compute the block crc with a 0 value in the
508          * hb_cksum field. Save it off here and replace after the
509          * crc. */
510         old_cksum = hb_block->hb_cksum;
511         hb_block->hb_cksum = 0;
512
513         ret = crc32_le(0, (unsigned char *) hb_block, reg->hr_block_bytes);
514
515         hb_block->hb_cksum = old_cksum;
516
517         return ret;
518 }
519
520 static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block *hb_block)
521 {
522         mlog(ML_ERROR, "Dump slot information: seq = 0x%llx, node = %u, "
523              "cksum = 0x%x, generation 0x%llx\n",
524              (long long)le64_to_cpu(hb_block->hb_seq),
525              hb_block->hb_node, le32_to_cpu(hb_block->hb_cksum),
526              (long long)le64_to_cpu(hb_block->hb_generation));
527 }
528
529 static int o2hb_verify_crc(struct o2hb_region *reg,
530                            struct o2hb_disk_heartbeat_block *hb_block)
531 {
532         u32 read, computed;
533
534         read = le32_to_cpu(hb_block->hb_cksum);
535         computed = o2hb_compute_block_crc_le(reg, hb_block);
536
537         return read == computed;
538 }
539
540 /*
541  * Compare the slot data with what we wrote in the last iteration.
542  * If the match fails, print an appropriate error message. This is to
543  * detect errors like... another node hearting on the same slot,
544  * flaky device that is losing writes, etc.
545  * Returns 1 if check succeeds, 0 otherwise.
546  */
547 static int o2hb_check_own_slot(struct o2hb_region *reg)
548 {
549         struct o2hb_disk_slot *slot;
550         struct o2hb_disk_heartbeat_block *hb_block;
551         char *errstr;
552
553         slot = &reg->hr_slots[o2nm_this_node()];
554         /* Don't check on our 1st timestamp */
555         if (!slot->ds_last_time)
556                 return 0;
557
558         hb_block = slot->ds_raw_block;
559         if (le64_to_cpu(hb_block->hb_seq) == slot->ds_last_time &&
560             le64_to_cpu(hb_block->hb_generation) == slot->ds_last_generation &&
561             hb_block->hb_node == slot->ds_node_num)
562                 return 1;
563
564 #define ERRSTR1         "Another node is heartbeating on device"
565 #define ERRSTR2         "Heartbeat generation mismatch on device"
566 #define ERRSTR3         "Heartbeat sequence mismatch on device"
567
568         if (hb_block->hb_node != slot->ds_node_num)
569                 errstr = ERRSTR1;
570         else if (le64_to_cpu(hb_block->hb_generation) !=
571                  slot->ds_last_generation)
572                 errstr = ERRSTR2;
573         else
574                 errstr = ERRSTR3;
575
576         mlog(ML_ERROR, "%s (%s): expected(%u:0x%llx, 0x%llx), "
577              "ondisk(%u:0x%llx, 0x%llx)\n", errstr, reg->hr_dev_name,
578              slot->ds_node_num, (unsigned long long)slot->ds_last_generation,
579              (unsigned long long)slot->ds_last_time, hb_block->hb_node,
580              (unsigned long long)le64_to_cpu(hb_block->hb_generation),
581              (unsigned long long)le64_to_cpu(hb_block->hb_seq));
582
583         return 0;
584 }
585
586 static inline void o2hb_prepare_block(struct o2hb_region *reg,
587                                       u64 generation)
588 {
589         int node_num;
590         u64 cputime;
591         struct o2hb_disk_slot *slot;
592         struct o2hb_disk_heartbeat_block *hb_block;
593
594         node_num = o2nm_this_node();
595         slot = &reg->hr_slots[node_num];
596
597         hb_block = (struct o2hb_disk_heartbeat_block *)slot->ds_raw_block;
598         memset(hb_block, 0, reg->hr_block_bytes);
599         /* TODO: time stuff */
600         cputime = CURRENT_TIME.tv_sec;
601         if (!cputime)
602                 cputime = 1;
603
604         hb_block->hb_seq = cpu_to_le64(cputime);
605         hb_block->hb_node = node_num;
606         hb_block->hb_generation = cpu_to_le64(generation);
607         hb_block->hb_dead_ms = cpu_to_le32(o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS);
608
609         /* This step must always happen last! */
610         hb_block->hb_cksum = cpu_to_le32(o2hb_compute_block_crc_le(reg,
611                                                                    hb_block));
612
613         mlog(ML_HB_BIO, "our node generation = 0x%llx, cksum = 0x%x\n",
614              (long long)generation,
615              le32_to_cpu(hb_block->hb_cksum));
616 }
617
618 static void o2hb_fire_callbacks(struct o2hb_callback *hbcall,
619                                 struct o2nm_node *node,
620                                 int idx)
621 {
622         struct o2hb_callback_func *f;
623
624         list_for_each_entry(f, &hbcall->list, hc_item) {
625                 mlog(ML_HEARTBEAT, "calling funcs %p\n", f);
626                 (f->hc_func)(node, idx, f->hc_data);
627         }
628 }
629
630 /* Will run the list in order until we process the passed event */
631 static void o2hb_run_event_list(struct o2hb_node_event *queued_event)
632 {
633         struct o2hb_callback *hbcall;
634         struct o2hb_node_event *event;
635
636         /* Holding callback sem assures we don't alter the callback
637          * lists when doing this, and serializes ourselves with other
638          * processes wanting callbacks. */
639         down_write(&o2hb_callback_sem);
640
641         spin_lock(&o2hb_live_lock);
642         while (!list_empty(&o2hb_node_events)
643                && !list_empty(&queued_event->hn_item)) {
644                 event = list_entry(o2hb_node_events.next,
645                                    struct o2hb_node_event,
646                                    hn_item);
647                 list_del_init(&event->hn_item);
648                 spin_unlock(&o2hb_live_lock);
649
650                 mlog(ML_HEARTBEAT, "Node %s event for %d\n",
651                      event->hn_event_type == O2HB_NODE_UP_CB ? "UP" : "DOWN",
652                      event->hn_node_num);
653
654                 hbcall = hbcall_from_type(event->hn_event_type);
655
656                 /* We should *never* have gotten on to the list with a
657                  * bad type... This isn't something that we should try
658                  * to recover from. */
659                 BUG_ON(IS_ERR(hbcall));
660
661                 o2hb_fire_callbacks(hbcall, event->hn_node, event->hn_node_num);
662
663                 spin_lock(&o2hb_live_lock);
664         }
665         spin_unlock(&o2hb_live_lock);
666
667         up_write(&o2hb_callback_sem);
668 }
669
670 static void o2hb_queue_node_event(struct o2hb_node_event *event,
671                                   enum o2hb_callback_type type,
672                                   struct o2nm_node *node,
673                                   int node_num)
674 {
675         assert_spin_locked(&o2hb_live_lock);
676
677         BUG_ON((!node) && (type != O2HB_NODE_DOWN_CB));
678
679         event->hn_event_type = type;
680         event->hn_node = node;
681         event->hn_node_num = node_num;
682
683         mlog(ML_HEARTBEAT, "Queue node %s event for node %d\n",
684              type == O2HB_NODE_UP_CB ? "UP" : "DOWN", node_num);
685
686         list_add_tail(&event->hn_item, &o2hb_node_events);
687 }
688
689 static void o2hb_shutdown_slot(struct o2hb_disk_slot *slot)
690 {
691         struct o2hb_node_event event =
692                 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
693         struct o2nm_node *node;
694         int queued = 0;
695
696         node = o2nm_get_node_by_num(slot->ds_node_num);
697         if (!node)
698                 return;
699
700         spin_lock(&o2hb_live_lock);
701         if (!list_empty(&slot->ds_live_item)) {
702                 mlog(ML_HEARTBEAT, "Shutdown, node %d leaves region\n",
703                      slot->ds_node_num);
704
705                 list_del_init(&slot->ds_live_item);
706
707                 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
708                         clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
709
710                         o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node,
711                                               slot->ds_node_num);
712                         queued = 1;
713                 }
714         }
715         spin_unlock(&o2hb_live_lock);
716
717         if (queued)
718                 o2hb_run_event_list(&event);
719
720         o2nm_node_put(node);
721 }
722
723 static void o2hb_set_quorum_device(struct o2hb_region *reg)
724 {
725         if (!o2hb_global_heartbeat_active())
726                 return;
727
728         /* Prevent race with o2hb_heartbeat_group_drop_item() */
729         if (kthread_should_stop())
730                 return;
731
732         /* Tag region as quorum only after thread reaches steady state */
733         if (atomic_read(&reg->hr_steady_iterations) != 0)
734                 return;
735
736         spin_lock(&o2hb_live_lock);
737
738         if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
739                 goto unlock;
740
741         /*
742          * A region can be added to the quorum only when it sees all
743          * live nodes heartbeat on it. In other words, the region has been
744          * added to all nodes.
745          */
746         if (memcmp(reg->hr_live_node_bitmap, o2hb_live_node_bitmap,
747                    sizeof(o2hb_live_node_bitmap)))
748                 goto unlock;
749
750         printk(KERN_NOTICE "o2hb: Region %s (%s) is now a quorum device\n",
751                config_item_name(&reg->hr_item), reg->hr_dev_name);
752
753         set_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
754
755         /*
756          * If global heartbeat active, unpin all regions if the
757          * region count > CUT_OFF
758          */
759         if (bitmap_weight(o2hb_quorum_region_bitmap,
760                            O2NM_MAX_REGIONS) > O2HB_PIN_CUT_OFF)
761                 o2hb_region_unpin(NULL);
762 unlock:
763         spin_unlock(&o2hb_live_lock);
764 }
765
766 static int o2hb_check_slot(struct o2hb_region *reg,
767                            struct o2hb_disk_slot *slot)
768 {
769         int changed = 0, gen_changed = 0;
770         struct o2hb_node_event event =
771                 { .hn_item = LIST_HEAD_INIT(event.hn_item), };
772         struct o2nm_node *node;
773         struct o2hb_disk_heartbeat_block *hb_block = reg->hr_tmp_block;
774         u64 cputime;
775         unsigned int dead_ms = o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS;
776         unsigned int slot_dead_ms;
777         int tmp;
778         int queued = 0;
779
780         memcpy(hb_block, slot->ds_raw_block, reg->hr_block_bytes);
781
782         /*
783          * If a node is no longer configured but is still in the livemap, we
784          * may need to clear that bit from the livemap.
785          */
786         node = o2nm_get_node_by_num(slot->ds_node_num);
787         if (!node) {
788                 spin_lock(&o2hb_live_lock);
789                 tmp = test_bit(slot->ds_node_num, o2hb_live_node_bitmap);
790                 spin_unlock(&o2hb_live_lock);
791                 if (!tmp)
792                         return 0;
793         }
794
795         if (!o2hb_verify_crc(reg, hb_block)) {
796                 /* all paths from here will drop o2hb_live_lock for
797                  * us. */
798                 spin_lock(&o2hb_live_lock);
799
800                 /* Don't print an error on the console in this case -
801                  * a freshly formatted heartbeat area will not have a
802                  * crc set on it. */
803                 if (list_empty(&slot->ds_live_item))
804                         goto out;
805
806                 /* The node is live but pushed out a bad crc. We
807                  * consider it a transient miss but don't populate any
808                  * other values as they may be junk. */
809                 mlog(ML_ERROR, "Node %d has written a bad crc to %s\n",
810                      slot->ds_node_num, reg->hr_dev_name);
811                 o2hb_dump_slot(hb_block);
812
813                 slot->ds_equal_samples++;
814                 goto fire_callbacks;
815         }
816
817         /* we don't care if these wrap.. the state transitions below
818          * clear at the right places */
819         cputime = le64_to_cpu(hb_block->hb_seq);
820         if (slot->ds_last_time != cputime)
821                 slot->ds_changed_samples++;
822         else
823                 slot->ds_equal_samples++;
824         slot->ds_last_time = cputime;
825
826         /* The node changed heartbeat generations. We assume this to
827          * mean it dropped off but came back before we timed out. We
828          * want to consider it down for the time being but don't want
829          * to lose any changed_samples state we might build up to
830          * considering it live again. */
831         if (slot->ds_last_generation != le64_to_cpu(hb_block->hb_generation)) {
832                 gen_changed = 1;
833                 slot->ds_equal_samples = 0;
834                 mlog(ML_HEARTBEAT, "Node %d changed generation (0x%llx "
835                      "to 0x%llx)\n", slot->ds_node_num,
836                      (long long)slot->ds_last_generation,
837                      (long long)le64_to_cpu(hb_block->hb_generation));
838         }
839
840         slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
841
842         mlog(ML_HEARTBEAT, "Slot %d gen 0x%llx cksum 0x%x "
843              "seq %llu last %llu changed %u equal %u\n",
844              slot->ds_node_num, (long long)slot->ds_last_generation,
845              le32_to_cpu(hb_block->hb_cksum),
846              (unsigned long long)le64_to_cpu(hb_block->hb_seq),
847              (unsigned long long)slot->ds_last_time, slot->ds_changed_samples,
848              slot->ds_equal_samples);
849
850         spin_lock(&o2hb_live_lock);
851
852 fire_callbacks:
853         /* dead nodes only come to life after some number of
854          * changes at any time during their dead time */
855         if (list_empty(&slot->ds_live_item) &&
856             slot->ds_changed_samples >= O2HB_LIVE_THRESHOLD) {
857                 mlog(ML_HEARTBEAT, "Node %d (id 0x%llx) joined my region\n",
858                      slot->ds_node_num, (long long)slot->ds_last_generation);
859
860                 set_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
861
862                 /* first on the list generates a callback */
863                 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
864                         mlog(ML_HEARTBEAT, "o2hb: Add node %d to live nodes "
865                              "bitmap\n", slot->ds_node_num);
866                         set_bit(slot->ds_node_num, o2hb_live_node_bitmap);
867
868                         o2hb_queue_node_event(&event, O2HB_NODE_UP_CB, node,
869                                               slot->ds_node_num);
870
871                         changed = 1;
872                         queued = 1;
873                 }
874
875                 list_add_tail(&slot->ds_live_item,
876                               &o2hb_live_slots[slot->ds_node_num]);
877
878                 slot->ds_equal_samples = 0;
879
880                 /* We want to be sure that all nodes agree on the
881                  * number of milliseconds before a node will be
882                  * considered dead. The self-fencing timeout is
883                  * computed from this value, and a discrepancy might
884                  * result in heartbeat calling a node dead when it
885                  * hasn't self-fenced yet. */
886                 slot_dead_ms = le32_to_cpu(hb_block->hb_dead_ms);
887                 if (slot_dead_ms && slot_dead_ms != dead_ms) {
888                         /* TODO: Perhaps we can fail the region here. */
889                         mlog(ML_ERROR, "Node %d on device %s has a dead count "
890                              "of %u ms, but our count is %u ms.\n"
891                              "Please double check your configuration values "
892                              "for 'O2CB_HEARTBEAT_THRESHOLD'\n",
893                              slot->ds_node_num, reg->hr_dev_name, slot_dead_ms,
894                              dead_ms);
895                 }
896                 goto out;
897         }
898
899         /* if the list is dead, we're done.. */
900         if (list_empty(&slot->ds_live_item))
901                 goto out;
902
903         /* live nodes only go dead after enough consequtive missed
904          * samples..  reset the missed counter whenever we see
905          * activity */
906         if (slot->ds_equal_samples >= o2hb_dead_threshold || gen_changed) {
907                 mlog(ML_HEARTBEAT, "Node %d left my region\n",
908                      slot->ds_node_num);
909
910                 clear_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
911
912                 /* last off the live_slot generates a callback */
913                 list_del_init(&slot->ds_live_item);
914                 if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
915                         mlog(ML_HEARTBEAT, "o2hb: Remove node %d from live "
916                              "nodes bitmap\n", slot->ds_node_num);
917                         clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
918
919                         /* node can be null */
920                         o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB,
921                                               node, slot->ds_node_num);
922
923                         changed = 1;
924                         queued = 1;
925                 }
926
927                 /* We don't clear this because the node is still
928                  * actually writing new blocks. */
929                 if (!gen_changed)
930                         slot->ds_changed_samples = 0;
931                 goto out;
932         }
933         if (slot->ds_changed_samples) {
934                 slot->ds_changed_samples = 0;
935                 slot->ds_equal_samples = 0;
936         }
937 out:
938         spin_unlock(&o2hb_live_lock);
939
940         if (queued)
941                 o2hb_run_event_list(&event);
942
943         if (node)
944                 o2nm_node_put(node);
945         return changed;
946 }
947
948 static int o2hb_highest_node(unsigned long *nodes, int numbits)
949 {
950         return find_last_bit(nodes, numbits);
951 }
952
953 static int o2hb_do_disk_heartbeat(struct o2hb_region *reg)
954 {
955         int i, ret, highest_node;
956         int membership_change = 0, own_slot_ok = 0;
957         unsigned long configured_nodes[BITS_TO_LONGS(O2NM_MAX_NODES)];
958         unsigned long live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
959         struct o2hb_bio_wait_ctxt write_wc;
960
961         ret = o2nm_configured_node_map(configured_nodes,
962                                        sizeof(configured_nodes));
963         if (ret) {
964                 mlog_errno(ret);
965                 goto bail;
966         }
967
968         /*
969          * If a node is not configured but is in the livemap, we still need
970          * to read the slot so as to be able to remove it from the livemap.
971          */
972         o2hb_fill_node_map(live_node_bitmap, sizeof(live_node_bitmap));
973         i = -1;
974         while ((i = find_next_bit(live_node_bitmap,
975                                   O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
976                 set_bit(i, configured_nodes);
977         }
978
979         highest_node = o2hb_highest_node(configured_nodes, O2NM_MAX_NODES);
980         if (highest_node >= O2NM_MAX_NODES) {
981                 mlog(ML_NOTICE, "o2hb: No configured nodes found!\n");
982                 ret = -EINVAL;
983                 goto bail;
984         }
985
986         /* No sense in reading the slots of nodes that don't exist
987          * yet. Of course, if the node definitions have holes in them
988          * then we're reading an empty slot anyway... Consider this
989          * best-effort. */
990         ret = o2hb_read_slots(reg, highest_node + 1);
991         if (ret < 0) {
992                 mlog_errno(ret);
993                 goto bail;
994         }
995
996         /* With an up to date view of the slots, we can check that no
997          * other node has been improperly configured to heartbeat in
998          * our slot. */
999         own_slot_ok = o2hb_check_own_slot(reg);
1000
1001         /* fill in the proper info for our next heartbeat */
1002         o2hb_prepare_block(reg, reg->hr_generation);
1003
1004         ret = o2hb_issue_node_write(reg, &write_wc);
1005         if (ret < 0) {
1006                 mlog_errno(ret);
1007                 goto bail;
1008         }
1009
1010         i = -1;
1011         while((i = find_next_bit(configured_nodes,
1012                                  O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
1013                 membership_change |= o2hb_check_slot(reg, &reg->hr_slots[i]);
1014         }
1015
1016         /*
1017          * We have to be sure we've advertised ourselves on disk
1018          * before we can go to steady state.  This ensures that
1019          * people we find in our steady state have seen us.
1020          */
1021         o2hb_wait_on_io(reg, &write_wc);
1022         if (write_wc.wc_error) {
1023                 /* Do not re-arm the write timeout on I/O error - we
1024                  * can't be sure that the new block ever made it to
1025                  * disk */
1026                 mlog(ML_ERROR, "Write error %d on device \"%s\"\n",
1027                      write_wc.wc_error, reg->hr_dev_name);
1028                 ret = write_wc.wc_error;
1029                 goto bail;
1030         }
1031
1032         /* Skip disarming the timeout if own slot has stale/bad data */
1033         if (own_slot_ok) {
1034                 o2hb_set_quorum_device(reg);
1035                 o2hb_arm_write_timeout(reg);
1036         }
1037
1038 bail:
1039         /* let the person who launched us know when things are steady */
1040         if (atomic_read(&reg->hr_steady_iterations) != 0) {
1041                 if (!ret && own_slot_ok && !membership_change) {
1042                         if (atomic_dec_and_test(&reg->hr_steady_iterations))
1043                                 wake_up(&o2hb_steady_queue);
1044                 }
1045         }
1046
1047         if (atomic_read(&reg->hr_steady_iterations) != 0) {
1048                 if (atomic_dec_and_test(&reg->hr_unsteady_iterations)) {
1049                         printk(KERN_NOTICE "o2hb: Unable to stabilize "
1050                                "heartbeart on region %s (%s)\n",
1051                                config_item_name(&reg->hr_item),
1052                                reg->hr_dev_name);
1053                         atomic_set(&reg->hr_steady_iterations, 0);
1054                         reg->hr_aborted_start = 1;
1055                         wake_up(&o2hb_steady_queue);
1056                         ret = -EIO;
1057                 }
1058         }
1059
1060         return ret;
1061 }
1062
1063 /*
1064  * we ride the region ref that the region dir holds.  before the region
1065  * dir is removed and drops it ref it will wait to tear down this
1066  * thread.
1067  */
1068 static int o2hb_thread(void *data)
1069 {
1070         int i, ret;
1071         struct o2hb_region *reg = data;
1072         struct o2hb_bio_wait_ctxt write_wc;
1073         ktime_t before_hb, after_hb;
1074         unsigned int elapsed_msec;
1075
1076         mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread running\n");
1077
1078         set_user_nice(current, MIN_NICE);
1079
1080         /* Pin node */
1081         ret = o2nm_depend_this_node();
1082         if (ret) {
1083                 mlog(ML_ERROR, "Node has been deleted, ret = %d\n", ret);
1084                 reg->hr_node_deleted = 1;
1085                 wake_up(&o2hb_steady_queue);
1086                 return 0;
1087         }
1088
1089         while (!kthread_should_stop() &&
1090                !reg->hr_unclean_stop && !reg->hr_aborted_start) {
1091                 /* We track the time spent inside
1092                  * o2hb_do_disk_heartbeat so that we avoid more than
1093                  * hr_timeout_ms between disk writes. On busy systems
1094                  * this should result in a heartbeat which is less
1095                  * likely to time itself out. */
1096                 before_hb = ktime_get_real();
1097
1098                 ret = o2hb_do_disk_heartbeat(reg);
1099
1100                 after_hb = ktime_get_real();
1101
1102                 elapsed_msec = (unsigned int)
1103                                 ktime_ms_delta(after_hb, before_hb);
1104
1105                 mlog(ML_HEARTBEAT,
1106                      "start = %lld, end = %lld, msec = %u, ret = %d\n",
1107                      before_hb.tv64, after_hb.tv64, elapsed_msec, ret);
1108
1109                 if (!kthread_should_stop() &&
1110                     elapsed_msec < reg->hr_timeout_ms) {
1111                         /* the kthread api has blocked signals for us so no
1112                          * need to record the return value. */
1113                         msleep_interruptible(reg->hr_timeout_ms - elapsed_msec);
1114                 }
1115         }
1116
1117         o2hb_disarm_write_timeout(reg);
1118
1119         /* unclean stop is only used in very bad situation */
1120         for(i = 0; !reg->hr_unclean_stop && i < reg->hr_blocks; i++)
1121                 o2hb_shutdown_slot(&reg->hr_slots[i]);
1122
1123         /* Explicit down notification - avoid forcing the other nodes
1124          * to timeout on this region when we could just as easily
1125          * write a clear generation - thus indicating to them that
1126          * this node has left this region.
1127          */
1128         if (!reg->hr_unclean_stop && !reg->hr_aborted_start) {
1129                 o2hb_prepare_block(reg, 0);
1130                 ret = o2hb_issue_node_write(reg, &write_wc);
1131                 if (ret == 0)
1132                         o2hb_wait_on_io(reg, &write_wc);
1133                 else
1134                         mlog_errno(ret);
1135         }
1136
1137         /* Unpin node */
1138         o2nm_undepend_this_node();
1139
1140         mlog(ML_HEARTBEAT|ML_KTHREAD, "o2hb thread exiting\n");
1141
1142         return 0;
1143 }
1144
1145 #ifdef CONFIG_DEBUG_FS
1146 static int o2hb_debug_open(struct inode *inode, struct file *file)
1147 {
1148         struct o2hb_debug_buf *db = inode->i_private;
1149         struct o2hb_region *reg;
1150         unsigned long map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1151         unsigned long lts;
1152         char *buf = NULL;
1153         int i = -1;
1154         int out = 0;
1155
1156         /* max_nodes should be the largest bitmap we pass here */
1157         BUG_ON(sizeof(map) < db->db_size);
1158
1159         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1160         if (!buf)
1161                 goto bail;
1162
1163         switch (db->db_type) {
1164         case O2HB_DB_TYPE_LIVENODES:
1165         case O2HB_DB_TYPE_LIVEREGIONS:
1166         case O2HB_DB_TYPE_QUORUMREGIONS:
1167         case O2HB_DB_TYPE_FAILEDREGIONS:
1168                 spin_lock(&o2hb_live_lock);
1169                 memcpy(map, db->db_data, db->db_size);
1170                 spin_unlock(&o2hb_live_lock);
1171                 break;
1172
1173         case O2HB_DB_TYPE_REGION_LIVENODES:
1174                 spin_lock(&o2hb_live_lock);
1175                 reg = (struct o2hb_region *)db->db_data;
1176                 memcpy(map, reg->hr_live_node_bitmap, db->db_size);
1177                 spin_unlock(&o2hb_live_lock);
1178                 break;
1179
1180         case O2HB_DB_TYPE_REGION_NUMBER:
1181                 reg = (struct o2hb_region *)db->db_data;
1182                 out += snprintf(buf + out, PAGE_SIZE - out, "%d\n",
1183                                 reg->hr_region_num);
1184                 goto done;
1185
1186         case O2HB_DB_TYPE_REGION_ELAPSED_TIME:
1187                 reg = (struct o2hb_region *)db->db_data;
1188                 lts = reg->hr_last_timeout_start;
1189                 /* If 0, it has never been set before */
1190                 if (lts)
1191                         lts = jiffies_to_msecs(jiffies - lts);
1192                 out += snprintf(buf + out, PAGE_SIZE - out, "%lu\n", lts);
1193                 goto done;
1194
1195         case O2HB_DB_TYPE_REGION_PINNED:
1196                 reg = (struct o2hb_region *)db->db_data;
1197                 out += snprintf(buf + out, PAGE_SIZE - out, "%u\n",
1198                                 !!reg->hr_item_pinned);
1199                 goto done;
1200
1201         default:
1202                 goto done;
1203         }
1204
1205         while ((i = find_next_bit(map, db->db_len, i + 1)) < db->db_len)
1206                 out += snprintf(buf + out, PAGE_SIZE - out, "%d ", i);
1207         out += snprintf(buf + out, PAGE_SIZE - out, "\n");
1208
1209 done:
1210         i_size_write(inode, out);
1211
1212         file->private_data = buf;
1213
1214         return 0;
1215 bail:
1216         return -ENOMEM;
1217 }
1218
1219 static int o2hb_debug_release(struct inode *inode, struct file *file)
1220 {
1221         kfree(file->private_data);
1222         return 0;
1223 }
1224
1225 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1226                                  size_t nbytes, loff_t *ppos)
1227 {
1228         return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
1229                                        i_size_read(file->f_mapping->host));
1230 }
1231 #else
1232 static int o2hb_debug_open(struct inode *inode, struct file *file)
1233 {
1234         return 0;
1235 }
1236 static int o2hb_debug_release(struct inode *inode, struct file *file)
1237 {
1238         return 0;
1239 }
1240 static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
1241                                size_t nbytes, loff_t *ppos)
1242 {
1243         return 0;
1244 }
1245 #endif  /* CONFIG_DEBUG_FS */
1246
1247 static const struct file_operations o2hb_debug_fops = {
1248         .open =         o2hb_debug_open,
1249         .release =      o2hb_debug_release,
1250         .read =         o2hb_debug_read,
1251         .llseek =       generic_file_llseek,
1252 };
1253
1254 void o2hb_exit(void)
1255 {
1256         debugfs_remove(o2hb_debug_failedregions);
1257         debugfs_remove(o2hb_debug_quorumregions);
1258         debugfs_remove(o2hb_debug_liveregions);
1259         debugfs_remove(o2hb_debug_livenodes);
1260         debugfs_remove(o2hb_debug_dir);
1261         kfree(o2hb_db_livenodes);
1262         kfree(o2hb_db_liveregions);
1263         kfree(o2hb_db_quorumregions);
1264         kfree(o2hb_db_failedregions);
1265 }
1266
1267 static struct dentry *o2hb_debug_create(const char *name, struct dentry *dir,
1268                                         struct o2hb_debug_buf **db, int db_len,
1269                                         int type, int size, int len, void *data)
1270 {
1271         *db = kmalloc(db_len, GFP_KERNEL);
1272         if (!*db)
1273                 return NULL;
1274
1275         (*db)->db_type = type;
1276         (*db)->db_size = size;
1277         (*db)->db_len = len;
1278         (*db)->db_data = data;
1279
1280         return debugfs_create_file(name, S_IFREG|S_IRUSR, dir, *db,
1281                                    &o2hb_debug_fops);
1282 }
1283
1284 static int o2hb_debug_init(void)
1285 {
1286         int ret = -ENOMEM;
1287
1288         o2hb_debug_dir = debugfs_create_dir(O2HB_DEBUG_DIR, NULL);
1289         if (!o2hb_debug_dir) {
1290                 mlog_errno(ret);
1291                 goto bail;
1292         }
1293
1294         o2hb_debug_livenodes = o2hb_debug_create(O2HB_DEBUG_LIVENODES,
1295                                                  o2hb_debug_dir,
1296                                                  &o2hb_db_livenodes,
1297                                                  sizeof(*o2hb_db_livenodes),
1298                                                  O2HB_DB_TYPE_LIVENODES,
1299                                                  sizeof(o2hb_live_node_bitmap),
1300                                                  O2NM_MAX_NODES,
1301                                                  o2hb_live_node_bitmap);
1302         if (!o2hb_debug_livenodes) {
1303                 mlog_errno(ret);
1304                 goto bail;
1305         }
1306
1307         o2hb_debug_liveregions = o2hb_debug_create(O2HB_DEBUG_LIVEREGIONS,
1308                                                    o2hb_debug_dir,
1309                                                    &o2hb_db_liveregions,
1310                                                    sizeof(*o2hb_db_liveregions),
1311                                                    O2HB_DB_TYPE_LIVEREGIONS,
1312                                                    sizeof(o2hb_live_region_bitmap),
1313                                                    O2NM_MAX_REGIONS,
1314                                                    o2hb_live_region_bitmap);
1315         if (!o2hb_debug_liveregions) {
1316                 mlog_errno(ret);
1317                 goto bail;
1318         }
1319
1320         o2hb_debug_quorumregions =
1321                         o2hb_debug_create(O2HB_DEBUG_QUORUMREGIONS,
1322                                           o2hb_debug_dir,
1323                                           &o2hb_db_quorumregions,
1324                                           sizeof(*o2hb_db_quorumregions),
1325                                           O2HB_DB_TYPE_QUORUMREGIONS,
1326                                           sizeof(o2hb_quorum_region_bitmap),
1327                                           O2NM_MAX_REGIONS,
1328                                           o2hb_quorum_region_bitmap);
1329         if (!o2hb_debug_quorumregions) {
1330                 mlog_errno(ret);
1331                 goto bail;
1332         }
1333
1334         o2hb_debug_failedregions =
1335                         o2hb_debug_create(O2HB_DEBUG_FAILEDREGIONS,
1336                                           o2hb_debug_dir,
1337                                           &o2hb_db_failedregions,
1338                                           sizeof(*o2hb_db_failedregions),
1339                                           O2HB_DB_TYPE_FAILEDREGIONS,
1340                                           sizeof(o2hb_failed_region_bitmap),
1341                                           O2NM_MAX_REGIONS,
1342                                           o2hb_failed_region_bitmap);
1343         if (!o2hb_debug_failedregions) {
1344                 mlog_errno(ret);
1345                 goto bail;
1346         }
1347
1348         ret = 0;
1349 bail:
1350         if (ret)
1351                 o2hb_exit();
1352
1353         return ret;
1354 }
1355
1356 int o2hb_init(void)
1357 {
1358         int i;
1359
1360         for (i = 0; i < ARRAY_SIZE(o2hb_callbacks); i++)
1361                 INIT_LIST_HEAD(&o2hb_callbacks[i].list);
1362
1363         for (i = 0; i < ARRAY_SIZE(o2hb_live_slots); i++)
1364                 INIT_LIST_HEAD(&o2hb_live_slots[i]);
1365
1366         INIT_LIST_HEAD(&o2hb_node_events);
1367
1368         memset(o2hb_live_node_bitmap, 0, sizeof(o2hb_live_node_bitmap));
1369         memset(o2hb_region_bitmap, 0, sizeof(o2hb_region_bitmap));
1370         memset(o2hb_live_region_bitmap, 0, sizeof(o2hb_live_region_bitmap));
1371         memset(o2hb_quorum_region_bitmap, 0, sizeof(o2hb_quorum_region_bitmap));
1372         memset(o2hb_failed_region_bitmap, 0, sizeof(o2hb_failed_region_bitmap));
1373
1374         o2hb_dependent_users = 0;
1375
1376         return o2hb_debug_init();
1377 }
1378
1379 /* if we're already in a callback then we're already serialized by the sem */
1380 static void o2hb_fill_node_map_from_callback(unsigned long *map,
1381                                              unsigned bytes)
1382 {
1383         BUG_ON(bytes < (BITS_TO_LONGS(O2NM_MAX_NODES) * sizeof(unsigned long)));
1384
1385         memcpy(map, &o2hb_live_node_bitmap, bytes);
1386 }
1387
1388 /*
1389  * get a map of all nodes that are heartbeating in any regions
1390  */
1391 void o2hb_fill_node_map(unsigned long *map, unsigned bytes)
1392 {
1393         /* callers want to serialize this map and callbacks so that they
1394          * can trust that they don't miss nodes coming to the party */
1395         down_read(&o2hb_callback_sem);
1396         spin_lock(&o2hb_live_lock);
1397         o2hb_fill_node_map_from_callback(map, bytes);
1398         spin_unlock(&o2hb_live_lock);
1399         up_read(&o2hb_callback_sem);
1400 }
1401 EXPORT_SYMBOL_GPL(o2hb_fill_node_map);
1402
1403 /*
1404  * heartbeat configfs bits.  The heartbeat set is a default set under
1405  * the cluster set in nodemanager.c.
1406  */
1407
1408 static struct o2hb_region *to_o2hb_region(struct config_item *item)
1409 {
1410         return item ? container_of(item, struct o2hb_region, hr_item) : NULL;
1411 }
1412
1413 /* drop_item only drops its ref after killing the thread, nothing should
1414  * be using the region anymore.  this has to clean up any state that
1415  * attributes might have built up. */
1416 static void o2hb_region_release(struct config_item *item)
1417 {
1418         int i;
1419         struct page *page;
1420         struct o2hb_region *reg = to_o2hb_region(item);
1421
1422         mlog(ML_HEARTBEAT, "hb region release (%s)\n", reg->hr_dev_name);
1423
1424         kfree(reg->hr_tmp_block);
1425
1426         if (reg->hr_slot_data) {
1427                 for (i = 0; i < reg->hr_num_pages; i++) {
1428                         page = reg->hr_slot_data[i];
1429                         if (page)
1430                                 __free_page(page);
1431                 }
1432                 kfree(reg->hr_slot_data);
1433         }
1434
1435         if (reg->hr_bdev)
1436                 blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1437
1438         kfree(reg->hr_slots);
1439
1440         debugfs_remove(reg->hr_debug_livenodes);
1441         debugfs_remove(reg->hr_debug_regnum);
1442         debugfs_remove(reg->hr_debug_elapsed_time);
1443         debugfs_remove(reg->hr_debug_pinned);
1444         debugfs_remove(reg->hr_debug_dir);
1445         kfree(reg->hr_db_livenodes);
1446         kfree(reg->hr_db_regnum);
1447         kfree(reg->hr_db_elapsed_time);
1448         kfree(reg->hr_db_pinned);
1449
1450         spin_lock(&o2hb_live_lock);
1451         list_del(&reg->hr_all_item);
1452         spin_unlock(&o2hb_live_lock);
1453
1454         kfree(reg);
1455 }
1456
1457 static int o2hb_read_block_input(struct o2hb_region *reg,
1458                                  const char *page,
1459                                  unsigned long *ret_bytes,
1460                                  unsigned int *ret_bits)
1461 {
1462         unsigned long bytes;
1463         char *p = (char *)page;
1464
1465         bytes = simple_strtoul(p, &p, 0);
1466         if (!p || (*p && (*p != '\n')))
1467                 return -EINVAL;
1468
1469         /* Heartbeat and fs min / max block sizes are the same. */
1470         if (bytes > 4096 || bytes < 512)
1471                 return -ERANGE;
1472         if (hweight16(bytes) != 1)
1473                 return -EINVAL;
1474
1475         if (ret_bytes)
1476                 *ret_bytes = bytes;
1477         if (ret_bits)
1478                 *ret_bits = ffs(bytes) - 1;
1479
1480         return 0;
1481 }
1482
1483 static ssize_t o2hb_region_block_bytes_show(struct config_item *item,
1484                                             char *page)
1485 {
1486         return sprintf(page, "%u\n", to_o2hb_region(item)->hr_block_bytes);
1487 }
1488
1489 static ssize_t o2hb_region_block_bytes_store(struct config_item *item,
1490                                              const char *page,
1491                                              size_t count)
1492 {
1493         struct o2hb_region *reg = to_o2hb_region(item);
1494         int status;
1495         unsigned long block_bytes;
1496         unsigned int block_bits;
1497
1498         if (reg->hr_bdev)
1499                 return -EINVAL;
1500
1501         status = o2hb_read_block_input(reg, page, &block_bytes,
1502                                        &block_bits);
1503         if (status)
1504                 return status;
1505
1506         reg->hr_block_bytes = (unsigned int)block_bytes;
1507         reg->hr_block_bits = block_bits;
1508
1509         return count;
1510 }
1511
1512 static ssize_t o2hb_region_start_block_show(struct config_item *item,
1513                                             char *page)
1514 {
1515         return sprintf(page, "%llu\n", to_o2hb_region(item)->hr_start_block);
1516 }
1517
1518 static ssize_t o2hb_region_start_block_store(struct config_item *item,
1519                                              const char *page,
1520                                              size_t count)
1521 {
1522         struct o2hb_region *reg = to_o2hb_region(item);
1523         unsigned long long tmp;
1524         char *p = (char *)page;
1525
1526         if (reg->hr_bdev)
1527                 return -EINVAL;
1528
1529         tmp = simple_strtoull(p, &p, 0);
1530         if (!p || (*p && (*p != '\n')))
1531                 return -EINVAL;
1532
1533         reg->hr_start_block = tmp;
1534
1535         return count;
1536 }
1537
1538 static ssize_t o2hb_region_blocks_show(struct config_item *item, char *page)
1539 {
1540         return sprintf(page, "%d\n", to_o2hb_region(item)->hr_blocks);
1541 }
1542
1543 static ssize_t o2hb_region_blocks_store(struct config_item *item,
1544                                         const char *page,
1545                                         size_t count)
1546 {
1547         struct o2hb_region *reg = to_o2hb_region(item);
1548         unsigned long tmp;
1549         char *p = (char *)page;
1550
1551         if (reg->hr_bdev)
1552                 return -EINVAL;
1553
1554         tmp = simple_strtoul(p, &p, 0);
1555         if (!p || (*p && (*p != '\n')))
1556                 return -EINVAL;
1557
1558         if (tmp > O2NM_MAX_NODES || tmp == 0)
1559                 return -ERANGE;
1560
1561         reg->hr_blocks = (unsigned int)tmp;
1562
1563         return count;
1564 }
1565
1566 static ssize_t o2hb_region_dev_show(struct config_item *item, char *page)
1567 {
1568         unsigned int ret = 0;
1569
1570         if (to_o2hb_region(item)->hr_bdev)
1571                 ret = sprintf(page, "%s\n", to_o2hb_region(item)->hr_dev_name);
1572
1573         return ret;
1574 }
1575
1576 static void o2hb_init_region_params(struct o2hb_region *reg)
1577 {
1578         reg->hr_slots_per_page = PAGE_SIZE >> reg->hr_block_bits;
1579         reg->hr_timeout_ms = O2HB_REGION_TIMEOUT_MS;
1580
1581         mlog(ML_HEARTBEAT, "hr_start_block = %llu, hr_blocks = %u\n",
1582              reg->hr_start_block, reg->hr_blocks);
1583         mlog(ML_HEARTBEAT, "hr_block_bytes = %u, hr_block_bits = %u\n",
1584              reg->hr_block_bytes, reg->hr_block_bits);
1585         mlog(ML_HEARTBEAT, "hr_timeout_ms = %u\n", reg->hr_timeout_ms);
1586         mlog(ML_HEARTBEAT, "dead threshold = %u\n", o2hb_dead_threshold);
1587 }
1588
1589 static int o2hb_map_slot_data(struct o2hb_region *reg)
1590 {
1591         int i, j;
1592         unsigned int last_slot;
1593         unsigned int spp = reg->hr_slots_per_page;
1594         struct page *page;
1595         char *raw;
1596         struct o2hb_disk_slot *slot;
1597
1598         reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL);
1599         if (reg->hr_tmp_block == NULL)
1600                 return -ENOMEM;
1601
1602         reg->hr_slots = kcalloc(reg->hr_blocks,
1603                                 sizeof(struct o2hb_disk_slot), GFP_KERNEL);
1604         if (reg->hr_slots == NULL)
1605                 return -ENOMEM;
1606
1607         for(i = 0; i < reg->hr_blocks; i++) {
1608                 slot = &reg->hr_slots[i];
1609                 slot->ds_node_num = i;
1610                 INIT_LIST_HEAD(&slot->ds_live_item);
1611                 slot->ds_raw_block = NULL;
1612         }
1613
1614         reg->hr_num_pages = (reg->hr_blocks + spp - 1) / spp;
1615         mlog(ML_HEARTBEAT, "Going to require %u pages to cover %u blocks "
1616                            "at %u blocks per page\n",
1617              reg->hr_num_pages, reg->hr_blocks, spp);
1618
1619         reg->hr_slot_data = kcalloc(reg->hr_num_pages, sizeof(struct page *),
1620                                     GFP_KERNEL);
1621         if (!reg->hr_slot_data)
1622                 return -ENOMEM;
1623
1624         for(i = 0; i < reg->hr_num_pages; i++) {
1625                 page = alloc_page(GFP_KERNEL);
1626                 if (!page)
1627                         return -ENOMEM;
1628
1629                 reg->hr_slot_data[i] = page;
1630
1631                 last_slot = i * spp;
1632                 raw = page_address(page);
1633                 for (j = 0;
1634                      (j < spp) && ((j + last_slot) < reg->hr_blocks);
1635                      j++) {
1636                         BUG_ON((j + last_slot) >= reg->hr_blocks);
1637
1638                         slot = &reg->hr_slots[j + last_slot];
1639                         slot->ds_raw_block =
1640                                 (struct o2hb_disk_heartbeat_block *) raw;
1641
1642                         raw += reg->hr_block_bytes;
1643                 }
1644         }
1645
1646         return 0;
1647 }
1648
1649 /* Read in all the slots available and populate the tracking
1650  * structures so that we can start with a baseline idea of what's
1651  * there. */
1652 static int o2hb_populate_slot_data(struct o2hb_region *reg)
1653 {
1654         int ret, i;
1655         struct o2hb_disk_slot *slot;
1656         struct o2hb_disk_heartbeat_block *hb_block;
1657
1658         ret = o2hb_read_slots(reg, reg->hr_blocks);
1659         if (ret)
1660                 goto out;
1661
1662         /* We only want to get an idea of the values initially in each
1663          * slot, so we do no verification - o2hb_check_slot will
1664          * actually determine if each configured slot is valid and
1665          * whether any values have changed. */
1666         for(i = 0; i < reg->hr_blocks; i++) {
1667                 slot = &reg->hr_slots[i];
1668                 hb_block = (struct o2hb_disk_heartbeat_block *) slot->ds_raw_block;
1669
1670                 /* Only fill the values that o2hb_check_slot uses to
1671                  * determine changing slots */
1672                 slot->ds_last_time = le64_to_cpu(hb_block->hb_seq);
1673                 slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
1674         }
1675
1676 out:
1677         return ret;
1678 }
1679
1680 /* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */
1681 static ssize_t o2hb_region_dev_store(struct config_item *item,
1682                                      const char *page,
1683                                      size_t count)
1684 {
1685         struct o2hb_region *reg = to_o2hb_region(item);
1686         struct task_struct *hb_task;
1687         long fd;
1688         int sectsize;
1689         char *p = (char *)page;
1690         struct fd f;
1691         struct inode *inode;
1692         ssize_t ret = -EINVAL;
1693         int live_threshold;
1694
1695         if (reg->hr_bdev)
1696                 goto out;
1697
1698         /* We can't heartbeat without having had our node number
1699          * configured yet. */
1700         if (o2nm_this_node() == O2NM_MAX_NODES)
1701                 goto out;
1702
1703         fd = simple_strtol(p, &p, 0);
1704         if (!p || (*p && (*p != '\n')))
1705                 goto out;
1706
1707         if (fd < 0 || fd >= INT_MAX)
1708                 goto out;
1709
1710         f = fdget(fd);
1711         if (f.file == NULL)
1712                 goto out;
1713
1714         if (reg->hr_blocks == 0 || reg->hr_start_block == 0 ||
1715             reg->hr_block_bytes == 0)
1716                 goto out2;
1717
1718         inode = igrab(f.file->f_mapping->host);
1719         if (inode == NULL)
1720                 goto out2;
1721
1722         if (!S_ISBLK(inode->i_mode))
1723                 goto out3;
1724
1725         reg->hr_bdev = I_BDEV(f.file->f_mapping->host);
1726         ret = blkdev_get(reg->hr_bdev, FMODE_WRITE | FMODE_READ, NULL);
1727         if (ret) {
1728                 reg->hr_bdev = NULL;
1729                 goto out3;
1730         }
1731         inode = NULL;
1732
1733         bdevname(reg->hr_bdev, reg->hr_dev_name);
1734
1735         sectsize = bdev_logical_block_size(reg->hr_bdev);
1736         if (sectsize != reg->hr_block_bytes) {
1737                 mlog(ML_ERROR,
1738                      "blocksize %u incorrect for device, expected %d",
1739                      reg->hr_block_bytes, sectsize);
1740                 ret = -EINVAL;
1741                 goto out3;
1742         }
1743
1744         o2hb_init_region_params(reg);
1745
1746         /* Generation of zero is invalid */
1747         do {
1748                 get_random_bytes(&reg->hr_generation,
1749                                  sizeof(reg->hr_generation));
1750         } while (reg->hr_generation == 0);
1751
1752         ret = o2hb_map_slot_data(reg);
1753         if (ret) {
1754                 mlog_errno(ret);
1755                 goto out3;
1756         }
1757
1758         ret = o2hb_populate_slot_data(reg);
1759         if (ret) {
1760                 mlog_errno(ret);
1761                 goto out3;
1762         }
1763
1764         INIT_DELAYED_WORK(&reg->hr_write_timeout_work, o2hb_write_timeout);
1765
1766         /*
1767          * A node is considered live after it has beat LIVE_THRESHOLD
1768          * times.  We're not steady until we've given them a chance
1769          * _after_ our first read.
1770          * The default threshold is bare minimum so as to limit the delay
1771          * during mounts. For global heartbeat, the threshold doubled for the
1772          * first region.
1773          */
1774         live_threshold = O2HB_LIVE_THRESHOLD;
1775         if (o2hb_global_heartbeat_active()) {
1776                 spin_lock(&o2hb_live_lock);
1777                 if (bitmap_weight(o2hb_region_bitmap, O2NM_MAX_REGIONS) == 1)
1778                         live_threshold <<= 1;
1779                 spin_unlock(&o2hb_live_lock);
1780         }
1781         ++live_threshold;
1782         atomic_set(&reg->hr_steady_iterations, live_threshold);
1783         /* unsteady_iterations is triple the steady_iterations */
1784         atomic_set(&reg->hr_unsteady_iterations, (live_threshold * 3));
1785
1786         hb_task = kthread_run(o2hb_thread, reg, "o2hb-%s",
1787                               reg->hr_item.ci_name);
1788         if (IS_ERR(hb_task)) {
1789                 ret = PTR_ERR(hb_task);
1790                 mlog_errno(ret);
1791                 goto out3;
1792         }
1793
1794         spin_lock(&o2hb_live_lock);
1795         reg->hr_task = hb_task;
1796         spin_unlock(&o2hb_live_lock);
1797
1798         ret = wait_event_interruptible(o2hb_steady_queue,
1799                                 atomic_read(&reg->hr_steady_iterations) == 0 ||
1800                                 reg->hr_node_deleted);
1801         if (ret) {
1802                 atomic_set(&reg->hr_steady_iterations, 0);
1803                 reg->hr_aborted_start = 1;
1804         }
1805
1806         if (reg->hr_aborted_start) {
1807                 ret = -EIO;
1808                 goto out3;
1809         }
1810
1811         if (reg->hr_node_deleted) {
1812                 ret = -EINVAL;
1813                 goto out3;
1814         }
1815
1816         /* Ok, we were woken.  Make sure it wasn't by drop_item() */
1817         spin_lock(&o2hb_live_lock);
1818         hb_task = reg->hr_task;
1819         if (o2hb_global_heartbeat_active())
1820                 set_bit(reg->hr_region_num, o2hb_live_region_bitmap);
1821         spin_unlock(&o2hb_live_lock);
1822
1823         if (hb_task)
1824                 ret = count;
1825         else
1826                 ret = -EIO;
1827
1828         if (hb_task && o2hb_global_heartbeat_active())
1829                 printk(KERN_NOTICE "o2hb: Heartbeat started on region %s (%s)\n",
1830                        config_item_name(&reg->hr_item), reg->hr_dev_name);
1831
1832 out3:
1833         iput(inode);
1834 out2:
1835         fdput(f);
1836 out:
1837         if (ret < 0) {
1838                 if (reg->hr_bdev) {
1839                         blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
1840                         reg->hr_bdev = NULL;
1841                 }
1842         }
1843         return ret;
1844 }
1845
1846 static ssize_t o2hb_region_pid_show(struct config_item *item, char *page)
1847 {
1848         struct o2hb_region *reg = to_o2hb_region(item);
1849         pid_t pid = 0;
1850
1851         spin_lock(&o2hb_live_lock);
1852         if (reg->hr_task)
1853                 pid = task_pid_nr(reg->hr_task);
1854         spin_unlock(&o2hb_live_lock);
1855
1856         if (!pid)
1857                 return 0;
1858
1859         return sprintf(page, "%u\n", pid);
1860 }
1861
1862 CONFIGFS_ATTR(o2hb_region_, block_bytes);
1863 CONFIGFS_ATTR(o2hb_region_, start_block);
1864 CONFIGFS_ATTR(o2hb_region_, blocks);
1865 CONFIGFS_ATTR(o2hb_region_, dev);
1866 CONFIGFS_ATTR_RO(o2hb_region_, pid);
1867
1868 static struct configfs_attribute *o2hb_region_attrs[] = {
1869         &o2hb_region_attr_block_bytes,
1870         &o2hb_region_attr_start_block,
1871         &o2hb_region_attr_blocks,
1872         &o2hb_region_attr_dev,
1873         &o2hb_region_attr_pid,
1874         NULL,
1875 };
1876
1877 static struct configfs_item_operations o2hb_region_item_ops = {
1878         .release                = o2hb_region_release,
1879 };
1880
1881 static struct config_item_type o2hb_region_type = {
1882         .ct_item_ops    = &o2hb_region_item_ops,
1883         .ct_attrs       = o2hb_region_attrs,
1884         .ct_owner       = THIS_MODULE,
1885 };
1886
1887 /* heartbeat set */
1888
1889 struct o2hb_heartbeat_group {
1890         struct config_group hs_group;
1891         /* some stuff? */
1892 };
1893
1894 static struct o2hb_heartbeat_group *to_o2hb_heartbeat_group(struct config_group *group)
1895 {
1896         return group ?
1897                 container_of(group, struct o2hb_heartbeat_group, hs_group)
1898                 : NULL;
1899 }
1900
1901 static int o2hb_debug_region_init(struct o2hb_region *reg, struct dentry *dir)
1902 {
1903         int ret = -ENOMEM;
1904
1905         reg->hr_debug_dir =
1906                 debugfs_create_dir(config_item_name(&reg->hr_item), dir);
1907         if (!reg->hr_debug_dir) {
1908                 mlog_errno(ret);
1909                 goto bail;
1910         }
1911
1912         reg->hr_debug_livenodes =
1913                         o2hb_debug_create(O2HB_DEBUG_LIVENODES,
1914                                           reg->hr_debug_dir,
1915                                           &(reg->hr_db_livenodes),
1916                                           sizeof(*(reg->hr_db_livenodes)),
1917                                           O2HB_DB_TYPE_REGION_LIVENODES,
1918                                           sizeof(reg->hr_live_node_bitmap),
1919                                           O2NM_MAX_NODES, reg);
1920         if (!reg->hr_debug_livenodes) {
1921                 mlog_errno(ret);
1922                 goto bail;
1923         }
1924
1925         reg->hr_debug_regnum =
1926                         o2hb_debug_create(O2HB_DEBUG_REGION_NUMBER,
1927                                           reg->hr_debug_dir,
1928                                           &(reg->hr_db_regnum),
1929                                           sizeof(*(reg->hr_db_regnum)),
1930                                           O2HB_DB_TYPE_REGION_NUMBER,
1931                                           0, O2NM_MAX_NODES, reg);
1932         if (!reg->hr_debug_regnum) {
1933                 mlog_errno(ret);
1934                 goto bail;
1935         }
1936
1937         reg->hr_debug_elapsed_time =
1938                         o2hb_debug_create(O2HB_DEBUG_REGION_ELAPSED_TIME,
1939                                           reg->hr_debug_dir,
1940                                           &(reg->hr_db_elapsed_time),
1941                                           sizeof(*(reg->hr_db_elapsed_time)),
1942                                           O2HB_DB_TYPE_REGION_ELAPSED_TIME,
1943                                           0, 0, reg);
1944         if (!reg->hr_debug_elapsed_time) {
1945                 mlog_errno(ret);
1946                 goto bail;
1947         }
1948
1949         reg->hr_debug_pinned =
1950                         o2hb_debug_create(O2HB_DEBUG_REGION_PINNED,
1951                                           reg->hr_debug_dir,
1952                                           &(reg->hr_db_pinned),
1953                                           sizeof(*(reg->hr_db_pinned)),
1954                                           O2HB_DB_TYPE_REGION_PINNED,
1955                                           0, 0, reg);
1956         if (!reg->hr_debug_pinned) {
1957                 mlog_errno(ret);
1958                 goto bail;
1959         }
1960
1961         ret = 0;
1962 bail:
1963         return ret;
1964 }
1965
1966 static struct config_item *o2hb_heartbeat_group_make_item(struct config_group *group,
1967                                                           const char *name)
1968 {
1969         struct o2hb_region *reg = NULL;
1970         int ret;
1971
1972         reg = kzalloc(sizeof(struct o2hb_region), GFP_KERNEL);
1973         if (reg == NULL)
1974                 return ERR_PTR(-ENOMEM);
1975
1976         if (strlen(name) > O2HB_MAX_REGION_NAME_LEN) {
1977                 ret = -ENAMETOOLONG;
1978                 goto free;
1979         }
1980
1981         spin_lock(&o2hb_live_lock);
1982         reg->hr_region_num = 0;
1983         if (o2hb_global_heartbeat_active()) {
1984                 reg->hr_region_num = find_first_zero_bit(o2hb_region_bitmap,
1985                                                          O2NM_MAX_REGIONS);
1986                 if (reg->hr_region_num >= O2NM_MAX_REGIONS) {
1987                         spin_unlock(&o2hb_live_lock);
1988                         ret = -EFBIG;
1989                         goto free;
1990                 }
1991                 set_bit(reg->hr_region_num, o2hb_region_bitmap);
1992         }
1993         list_add_tail(&reg->hr_all_item, &o2hb_all_regions);
1994         spin_unlock(&o2hb_live_lock);
1995
1996         config_item_init_type_name(&reg->hr_item, name, &o2hb_region_type);
1997
1998         ret = o2hb_debug_region_init(reg, o2hb_debug_dir);
1999         if (ret) {
2000                 config_item_put(&reg->hr_item);
2001                 goto free;
2002         }
2003
2004         return &reg->hr_item;
2005 free:
2006         kfree(reg);
2007         return ERR_PTR(ret);
2008 }
2009
2010 static void o2hb_heartbeat_group_drop_item(struct config_group *group,
2011                                            struct config_item *item)
2012 {
2013         struct task_struct *hb_task;
2014         struct o2hb_region *reg = to_o2hb_region(item);
2015         int quorum_region = 0;
2016
2017         /* stop the thread when the user removes the region dir */
2018         spin_lock(&o2hb_live_lock);
2019         hb_task = reg->hr_task;
2020         reg->hr_task = NULL;
2021         reg->hr_item_dropped = 1;
2022         spin_unlock(&o2hb_live_lock);
2023
2024         if (hb_task)
2025                 kthread_stop(hb_task);
2026
2027         if (o2hb_global_heartbeat_active()) {
2028                 spin_lock(&o2hb_live_lock);
2029                 clear_bit(reg->hr_region_num, o2hb_region_bitmap);
2030                 clear_bit(reg->hr_region_num, o2hb_live_region_bitmap);
2031                 if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
2032                         quorum_region = 1;
2033                 clear_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
2034                 spin_unlock(&o2hb_live_lock);
2035                 printk(KERN_NOTICE "o2hb: Heartbeat %s on region %s (%s)\n",
2036                        ((atomic_read(&reg->hr_steady_iterations) == 0) ?
2037                         "stopped" : "start aborted"), config_item_name(item),
2038                        reg->hr_dev_name);
2039         }
2040
2041         /*
2042          * If we're racing a dev_write(), we need to wake them.  They will
2043          * check reg->hr_task
2044          */
2045         if (atomic_read(&reg->hr_steady_iterations) != 0) {
2046                 reg->hr_aborted_start = 1;
2047                 atomic_set(&reg->hr_steady_iterations, 0);
2048                 wake_up(&o2hb_steady_queue);
2049         }
2050
2051         config_item_put(item);
2052
2053         if (!o2hb_global_heartbeat_active() || !quorum_region)
2054                 return;
2055
2056         /*
2057          * If global heartbeat active and there are dependent users,
2058          * pin all regions if quorum region count <= CUT_OFF
2059          */
2060         spin_lock(&o2hb_live_lock);
2061
2062         if (!o2hb_dependent_users)
2063                 goto unlock;
2064
2065         if (bitmap_weight(o2hb_quorum_region_bitmap,
2066                            O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2067                 o2hb_region_pin(NULL);
2068
2069 unlock:
2070         spin_unlock(&o2hb_live_lock);
2071 }
2072
2073 static ssize_t o2hb_heartbeat_group_threshold_show(struct config_item *item,
2074                 char *page)
2075 {
2076         return sprintf(page, "%u\n", o2hb_dead_threshold);
2077 }
2078
2079 static ssize_t o2hb_heartbeat_group_threshold_store(struct config_item *item,
2080                 const char *page, size_t count)
2081 {
2082         unsigned long tmp;
2083         char *p = (char *)page;
2084
2085         tmp = simple_strtoul(p, &p, 10);
2086         if (!p || (*p && (*p != '\n')))
2087                 return -EINVAL;
2088
2089         /* this will validate ranges for us. */
2090         o2hb_dead_threshold_set((unsigned int) tmp);
2091
2092         return count;
2093 }
2094
2095 static ssize_t o2hb_heartbeat_group_mode_show(struct config_item *item,
2096                 char *page)
2097 {
2098         return sprintf(page, "%s\n",
2099                        o2hb_heartbeat_mode_desc[o2hb_heartbeat_mode]);
2100 }
2101
2102 static ssize_t o2hb_heartbeat_group_mode_store(struct config_item *item,
2103                 const char *page, size_t count)
2104 {
2105         unsigned int i;
2106         int ret;
2107         size_t len;
2108
2109         len = (page[count - 1] == '\n') ? count - 1 : count;
2110         if (!len)
2111                 return -EINVAL;
2112
2113         for (i = 0; i < O2HB_HEARTBEAT_NUM_MODES; ++i) {
2114                 if (strncasecmp(page, o2hb_heartbeat_mode_desc[i], len))
2115                         continue;
2116
2117                 ret = o2hb_global_heartbeat_mode_set(i);
2118                 if (!ret)
2119                         printk(KERN_NOTICE "o2hb: Heartbeat mode set to %s\n",
2120                                o2hb_heartbeat_mode_desc[i]);
2121                 return count;
2122         }
2123
2124         return -EINVAL;
2125
2126 }
2127
2128 CONFIGFS_ATTR(o2hb_heartbeat_group_, threshold);
2129 CONFIGFS_ATTR(o2hb_heartbeat_group_, mode);
2130
2131 static struct configfs_attribute *o2hb_heartbeat_group_attrs[] = {
2132         &o2hb_heartbeat_group_attr_threshold,
2133         &o2hb_heartbeat_group_attr_mode,
2134         NULL,
2135 };
2136
2137 static struct configfs_group_operations o2hb_heartbeat_group_group_ops = {
2138         .make_item      = o2hb_heartbeat_group_make_item,
2139         .drop_item      = o2hb_heartbeat_group_drop_item,
2140 };
2141
2142 static struct config_item_type o2hb_heartbeat_group_type = {
2143         .ct_group_ops   = &o2hb_heartbeat_group_group_ops,
2144         .ct_attrs       = o2hb_heartbeat_group_attrs,
2145         .ct_owner       = THIS_MODULE,
2146 };
2147
2148 /* this is just here to avoid touching group in heartbeat.h which the
2149  * entire damn world #includes */
2150 struct config_group *o2hb_alloc_hb_set(void)
2151 {
2152         struct o2hb_heartbeat_group *hs = NULL;
2153         struct config_group *ret = NULL;
2154
2155         hs = kzalloc(sizeof(struct o2hb_heartbeat_group), GFP_KERNEL);
2156         if (hs == NULL)
2157                 goto out;
2158
2159         config_group_init_type_name(&hs->hs_group, "heartbeat",
2160                                     &o2hb_heartbeat_group_type);
2161
2162         ret = &hs->hs_group;
2163 out:
2164         if (ret == NULL)
2165                 kfree(hs);
2166         return ret;
2167 }
2168
2169 void o2hb_free_hb_set(struct config_group *group)
2170 {
2171         struct o2hb_heartbeat_group *hs = to_o2hb_heartbeat_group(group);
2172         kfree(hs);
2173 }
2174
2175 /* hb callback registration and issuing */
2176
2177 static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type)
2178 {
2179         if (type == O2HB_NUM_CB)
2180                 return ERR_PTR(-EINVAL);
2181
2182         return &o2hb_callbacks[type];
2183 }
2184
2185 void o2hb_setup_callback(struct o2hb_callback_func *hc,
2186                          enum o2hb_callback_type type,
2187                          o2hb_cb_func *func,
2188                          void *data,
2189                          int priority)
2190 {
2191         INIT_LIST_HEAD(&hc->hc_item);
2192         hc->hc_func = func;
2193         hc->hc_data = data;
2194         hc->hc_priority = priority;
2195         hc->hc_type = type;
2196         hc->hc_magic = O2HB_CB_MAGIC;
2197 }
2198 EXPORT_SYMBOL_GPL(o2hb_setup_callback);
2199
2200 /*
2201  * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2202  * In global heartbeat mode, region_uuid passed is NULL.
2203  *
2204  * In local, we only pin the matching region. In global we pin all the active
2205  * regions.
2206  */
2207 static int o2hb_region_pin(const char *region_uuid)
2208 {
2209         int ret = 0, found = 0;
2210         struct o2hb_region *reg;
2211         char *uuid;
2212
2213         assert_spin_locked(&o2hb_live_lock);
2214
2215         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2216                 if (reg->hr_item_dropped)
2217                         continue;
2218
2219                 uuid = config_item_name(&reg->hr_item);
2220
2221                 /* local heartbeat */
2222                 if (region_uuid) {
2223                         if (strcmp(region_uuid, uuid))
2224                                 continue;
2225                         found = 1;
2226                 }
2227
2228                 if (reg->hr_item_pinned || reg->hr_item_dropped)
2229                         goto skip_pin;
2230
2231                 /* Ignore ENOENT only for local hb (userdlm domain) */
2232                 ret = o2nm_depend_item(&reg->hr_item);
2233                 if (!ret) {
2234                         mlog(ML_CLUSTER, "Pin region %s\n", uuid);
2235                         reg->hr_item_pinned = 1;
2236                 } else {
2237                         if (ret == -ENOENT && found)
2238                                 ret = 0;
2239                         else {
2240                                 mlog(ML_ERROR, "Pin region %s fails with %d\n",
2241                                      uuid, ret);
2242                                 break;
2243                         }
2244                 }
2245 skip_pin:
2246                 if (found)
2247                         break;
2248         }
2249
2250         return ret;
2251 }
2252
2253 /*
2254  * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2255  * In global heartbeat mode, region_uuid passed is NULL.
2256  *
2257  * In local, we only unpin the matching region. In global we unpin all the
2258  * active regions.
2259  */
2260 static void o2hb_region_unpin(const char *region_uuid)
2261 {
2262         struct o2hb_region *reg;
2263         char *uuid;
2264         int found = 0;
2265
2266         assert_spin_locked(&o2hb_live_lock);
2267
2268         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2269                 if (reg->hr_item_dropped)
2270                         continue;
2271
2272                 uuid = config_item_name(&reg->hr_item);
2273                 if (region_uuid) {
2274                         if (strcmp(region_uuid, uuid))
2275                                 continue;
2276                         found = 1;
2277                 }
2278
2279                 if (reg->hr_item_pinned) {
2280                         mlog(ML_CLUSTER, "Unpin region %s\n", uuid);
2281                         o2nm_undepend_item(&reg->hr_item);
2282                         reg->hr_item_pinned = 0;
2283                 }
2284                 if (found)
2285                         break;
2286         }
2287 }
2288
2289 static int o2hb_region_inc_user(const char *region_uuid)
2290 {
2291         int ret = 0;
2292
2293         spin_lock(&o2hb_live_lock);
2294
2295         /* local heartbeat */
2296         if (!o2hb_global_heartbeat_active()) {
2297             ret = o2hb_region_pin(region_uuid);
2298             goto unlock;
2299         }
2300
2301         /*
2302          * if global heartbeat active and this is the first dependent user,
2303          * pin all regions if quorum region count <= CUT_OFF
2304          */
2305         o2hb_dependent_users++;
2306         if (o2hb_dependent_users > 1)
2307                 goto unlock;
2308
2309         if (bitmap_weight(o2hb_quorum_region_bitmap,
2310                            O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
2311                 ret = o2hb_region_pin(NULL);
2312
2313 unlock:
2314         spin_unlock(&o2hb_live_lock);
2315         return ret;
2316 }
2317
2318 void o2hb_region_dec_user(const char *region_uuid)
2319 {
2320         spin_lock(&o2hb_live_lock);
2321
2322         /* local heartbeat */
2323         if (!o2hb_global_heartbeat_active()) {
2324             o2hb_region_unpin(region_uuid);
2325             goto unlock;
2326         }
2327
2328         /*
2329          * if global heartbeat active and there are no dependent users,
2330          * unpin all quorum regions
2331          */
2332         o2hb_dependent_users--;
2333         if (!o2hb_dependent_users)
2334                 o2hb_region_unpin(NULL);
2335
2336 unlock:
2337         spin_unlock(&o2hb_live_lock);
2338 }
2339
2340 int o2hb_register_callback(const char *region_uuid,
2341                            struct o2hb_callback_func *hc)
2342 {
2343         struct o2hb_callback_func *f;
2344         struct o2hb_callback *hbcall;
2345         int ret;
2346
2347         BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2348         BUG_ON(!list_empty(&hc->hc_item));
2349
2350         hbcall = hbcall_from_type(hc->hc_type);
2351         if (IS_ERR(hbcall)) {
2352                 ret = PTR_ERR(hbcall);
2353                 goto out;
2354         }
2355
2356         if (region_uuid) {
2357                 ret = o2hb_region_inc_user(region_uuid);
2358                 if (ret) {
2359                         mlog_errno(ret);
2360                         goto out;
2361                 }
2362         }
2363
2364         down_write(&o2hb_callback_sem);
2365
2366         list_for_each_entry(f, &hbcall->list, hc_item) {
2367                 if (hc->hc_priority < f->hc_priority) {
2368                         list_add_tail(&hc->hc_item, &f->hc_item);
2369                         break;
2370                 }
2371         }
2372         if (list_empty(&hc->hc_item))
2373                 list_add_tail(&hc->hc_item, &hbcall->list);
2374
2375         up_write(&o2hb_callback_sem);
2376         ret = 0;
2377 out:
2378         mlog(ML_CLUSTER, "returning %d on behalf of %p for funcs %p\n",
2379              ret, __builtin_return_address(0), hc);
2380         return ret;
2381 }
2382 EXPORT_SYMBOL_GPL(o2hb_register_callback);
2383
2384 void o2hb_unregister_callback(const char *region_uuid,
2385                               struct o2hb_callback_func *hc)
2386 {
2387         BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
2388
2389         mlog(ML_CLUSTER, "on behalf of %p for funcs %p\n",
2390              __builtin_return_address(0), hc);
2391
2392         /* XXX Can this happen _with_ a region reference? */
2393         if (list_empty(&hc->hc_item))
2394                 return;
2395
2396         if (region_uuid)
2397                 o2hb_region_dec_user(region_uuid);
2398
2399         down_write(&o2hb_callback_sem);
2400
2401         list_del_init(&hc->hc_item);
2402
2403         up_write(&o2hb_callback_sem);
2404 }
2405 EXPORT_SYMBOL_GPL(o2hb_unregister_callback);
2406
2407 int o2hb_check_node_heartbeating(u8 node_num)
2408 {
2409         unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2410
2411         o2hb_fill_node_map(testing_map, sizeof(testing_map));
2412         if (!test_bit(node_num, testing_map)) {
2413                 mlog(ML_HEARTBEAT,
2414                      "node (%u) does not have heartbeating enabled.\n",
2415                      node_num);
2416                 return 0;
2417         }
2418
2419         return 1;
2420 }
2421 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating);
2422
2423 int o2hb_check_node_heartbeating_no_sem(u8 node_num)
2424 {
2425         unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2426
2427         spin_lock(&o2hb_live_lock);
2428         o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
2429         spin_unlock(&o2hb_live_lock);
2430         if (!test_bit(node_num, testing_map)) {
2431                 mlog(ML_HEARTBEAT,
2432                      "node (%u) does not have heartbeating enabled.\n",
2433                      node_num);
2434                 return 0;
2435         }
2436
2437         return 1;
2438 }
2439 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_no_sem);
2440
2441 int o2hb_check_node_heartbeating_from_callback(u8 node_num)
2442 {
2443         unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
2444
2445         o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
2446         if (!test_bit(node_num, testing_map)) {
2447                 mlog(ML_HEARTBEAT,
2448                      "node (%u) does not have heartbeating enabled.\n",
2449                      node_num);
2450                 return 0;
2451         }
2452
2453         return 1;
2454 }
2455 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback);
2456
2457 /* Makes sure our local node is configured with a node number, and is
2458  * heartbeating. */
2459 int o2hb_check_local_node_heartbeating(void)
2460 {
2461         u8 node_num;
2462
2463         /* if this node was set then we have networking */
2464         node_num = o2nm_this_node();
2465         if (node_num == O2NM_MAX_NODES) {
2466                 mlog(ML_HEARTBEAT, "this node has not been configured.\n");
2467                 return 0;
2468         }
2469
2470         return o2hb_check_node_heartbeating(node_num);
2471 }
2472 EXPORT_SYMBOL_GPL(o2hb_check_local_node_heartbeating);
2473
2474 /*
2475  * this is just a hack until we get the plumbing which flips file systems
2476  * read only and drops the hb ref instead of killing the node dead.
2477  */
2478 void o2hb_stop_all_regions(void)
2479 {
2480         struct o2hb_region *reg;
2481
2482         mlog(ML_ERROR, "stopping heartbeat on all active regions.\n");
2483
2484         spin_lock(&o2hb_live_lock);
2485
2486         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item)
2487                 reg->hr_unclean_stop = 1;
2488
2489         spin_unlock(&o2hb_live_lock);
2490 }
2491 EXPORT_SYMBOL_GPL(o2hb_stop_all_regions);
2492
2493 int o2hb_get_all_regions(char *region_uuids, u8 max_regions)
2494 {
2495         struct o2hb_region *reg;
2496         int numregs = 0;
2497         char *p;
2498
2499         spin_lock(&o2hb_live_lock);
2500
2501         p = region_uuids;
2502         list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
2503                 if (reg->hr_item_dropped)
2504                         continue;
2505
2506                 mlog(0, "Region: %s\n", config_item_name(&reg->hr_item));
2507                 if (numregs < max_regions) {
2508                         memcpy(p, config_item_name(&reg->hr_item),
2509                                O2HB_MAX_REGION_NAME_LEN);
2510                         p += O2HB_MAX_REGION_NAME_LEN;
2511                 }
2512                 numregs++;
2513         }
2514
2515         spin_unlock(&o2hb_live_lock);
2516
2517         return numregs;
2518 }
2519 EXPORT_SYMBOL_GPL(o2hb_get_all_regions);
2520
2521 int o2hb_global_heartbeat_active(void)
2522 {
2523         return (o2hb_heartbeat_mode == O2HB_HEARTBEAT_GLOBAL);
2524 }
2525 EXPORT_SYMBOL(o2hb_global_heartbeat_active);