1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
4 * Copyright (C) 2004, 2005 Oracle. All rights reserved.
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.
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.
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.
22 #include <linux/kernel.h>
23 #include <linux/sched.h>
24 #include <linux/jiffies.h>
25 #include <linux/module.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>
40 #include "heartbeat.h"
42 #include "nodemanager.h"
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.
54 static DECLARE_RWSEM(o2hb_callback_sem
);
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.
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
);
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
72 * - o2hb_failed_region_bitmap tracks the regions that have seen io timeouts.
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
)];
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
{
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
;
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"
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
;
114 static LIST_HEAD(o2hb_all_regions
);
116 static struct o2hb_callback
{
117 struct list_head list
;
118 } o2hb_callbacks
[O2HB_NUM_CB
];
120 static struct o2hb_callback
*hbcall_from_type(enum o2hb_callback_type type
);
122 #define O2HB_DEFAULT_BLOCK_BITS 9
124 enum o2hb_heartbeat_modes
{
125 O2HB_HEARTBEAT_LOCAL
= 0,
126 O2HB_HEARTBEAT_GLOBAL
,
127 O2HB_HEARTBEAT_NUM_MODES
,
130 char *o2hb_heartbeat_mode_desc
[O2HB_HEARTBEAT_NUM_MODES
] = {
131 "local", /* O2HB_HEARTBEAT_LOCAL */
132 "global", /* O2HB_HEARTBEAT_GLOBAL */
135 unsigned int o2hb_dead_threshold
= O2HB_DEFAULT_DEAD_THRESHOLD
;
136 unsigned int o2hb_heartbeat_mode
= O2HB_HEARTBEAT_LOCAL
;
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.
144 unsigned int o2hb_dependent_users
;
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.
152 #define O2HB_PIN_CUT_OFF 3
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.
159 * In global heartbeat mode, we pin/unpin all o2hb regions. This solution
160 * works for both file system and userdlm domains.
162 static int o2hb_region_pin(const char *region_uuid
);
163 static void o2hb_region_unpin(const char *region_uuid
);
165 /* Only sets a new threshold if there are no active regions.
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
)
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
);
180 static int o2hb_global_heartbeat_mode_set(unsigned int hb_mode
)
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
;
190 spin_unlock(&o2hb_live_lock
);
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
;
203 struct o2hb_disk_slot
{
204 struct o2hb_disk_heartbeat_block
*ds_raw_block
;
207 u64 ds_last_generation
;
208 u16 ds_equal_samples
;
209 u16 ds_changed_samples
;
210 struct list_head ds_live_item
;
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 */
216 struct config_item hr_item
;
218 struct list_head hr_all_item
;
219 unsigned hr_unclean_stop
:1,
224 /* protected by the hr_callback_sem */
225 struct task_struct
*hr_task
;
227 unsigned int hr_blocks
;
228 unsigned long long hr_start_block
;
230 unsigned int hr_block_bits
;
231 unsigned int hr_block_bytes
;
233 unsigned int hr_slots_per_page
;
234 unsigned int hr_num_pages
;
236 struct page
**hr_slot_data
;
237 struct block_device
*hr_bdev
;
238 struct o2hb_disk_slot
*hr_slots
;
240 /* live node map of this region */
241 unsigned long hr_live_node_bitmap
[BITS_TO_LONGS(O2NM_MAX_NODES
)];
242 unsigned int hr_region_num
;
244 struct dentry
*hr_debug_dir
;
245 struct dentry
*hr_debug_livenodes
;
246 struct dentry
*hr_debug_regnum
;
247 struct dentry
*hr_debug_elapsed_time
;
248 struct dentry
*hr_debug_pinned
;
249 struct o2hb_debug_buf
*hr_db_livenodes
;
250 struct o2hb_debug_buf
*hr_db_regnum
;
251 struct o2hb_debug_buf
*hr_db_elapsed_time
;
252 struct o2hb_debug_buf
*hr_db_pinned
;
254 /* let the person setting up hb wait for it to return until it
255 * has reached a 'steady' state. This will be fixed when we have
256 * a more complete api that doesn't lead to this sort of fragility. */
257 atomic_t hr_steady_iterations
;
259 /* terminate o2hb thread if it does not reach steady state
260 * (hr_steady_iterations == 0) within hr_unsteady_iterations */
261 atomic_t hr_unsteady_iterations
;
263 char hr_dev_name
[BDEVNAME_SIZE
];
265 unsigned int hr_timeout_ms
;
267 /* randomized as the region goes up and down so that a node
268 * recognizes a node going up and down in one iteration */
271 struct delayed_work hr_write_timeout_work
;
272 unsigned long hr_last_timeout_start
;
274 /* Used during o2hb_check_slot to hold a copy of the block
275 * being checked because we temporarily have to zero out the
277 struct o2hb_disk_heartbeat_block
*hr_tmp_block
;
280 struct o2hb_bio_wait_ctxt
{
281 atomic_t wc_num_reqs
;
282 struct completion wc_io_complete
;
286 static void o2hb_write_timeout(struct work_struct
*work
)
290 struct o2hb_region
*reg
=
291 container_of(work
, struct o2hb_region
,
292 hr_write_timeout_work
.work
);
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
));
298 if (o2hb_global_heartbeat_active()) {
299 spin_lock_irqsave(&o2hb_live_lock
, flags
);
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
,
304 quorum
= bitmap_weight(o2hb_quorum_region_bitmap
,
306 spin_unlock_irqrestore(&o2hb_live_lock
, flags
);
308 mlog(ML_HEARTBEAT
, "Number of regions %d, failed regions %d\n",
312 * Fence if the number of failed regions >= half the number
315 if ((failed
<< 1) < quorum
)
319 o2quo_disk_timeout();
322 static void o2hb_arm_write_timeout(struct o2hb_region
*reg
)
324 /* Arm writeout only after thread reaches steady state */
325 if (atomic_read(®
->hr_steady_iterations
) != 0)
328 mlog(ML_HEARTBEAT
, "Queue write timeout for %u ms\n",
329 O2HB_MAX_WRITE_TIMEOUT_MS
);
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
);
336 cancel_delayed_work(®
->hr_write_timeout_work
);
337 reg
->hr_last_timeout_start
= jiffies
;
338 schedule_delayed_work(®
->hr_write_timeout_work
,
339 msecs_to_jiffies(O2HB_MAX_WRITE_TIMEOUT_MS
));
342 static void o2hb_disarm_write_timeout(struct o2hb_region
*reg
)
344 cancel_delayed_work_sync(®
->hr_write_timeout_work
);
347 static inline void o2hb_bio_wait_init(struct o2hb_bio_wait_ctxt
*wc
)
349 atomic_set(&wc
->wc_num_reqs
, 1);
350 init_completion(&wc
->wc_io_complete
);
354 /* Used in error paths too */
355 static inline void o2hb_bio_wait_dec(struct o2hb_bio_wait_ctxt
*wc
,
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 */
361 if (atomic_dec_and_test(&wc
->wc_num_reqs
)) {
363 complete(&wc
->wc_io_complete
);
368 static void o2hb_wait_on_io(struct o2hb_region
*reg
,
369 struct o2hb_bio_wait_ctxt
*wc
)
371 o2hb_bio_wait_dec(wc
, 1);
372 wait_for_completion(&wc
->wc_io_complete
);
375 static void o2hb_bio_end_io(struct bio
*bio
)
377 struct o2hb_bio_wait_ctxt
*wc
= bio
->bi_private
;
380 mlog(ML_ERROR
, "IO Error %d\n", bio
->bi_error
);
381 wc
->wc_error
= bio
->bi_error
;
384 o2hb_bio_wait_dec(wc
, 1);
388 /* Setup a Bio to cover I/O against num_slots slots starting at
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
)
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
;
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
407 bio
= bio_alloc(GFP_ATOMIC
, 16);
409 mlog(ML_ERROR
, "Could not alloc slots BIO!\n");
410 bio
= ERR_PTR(-ENOMEM
);
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
;
420 vec_start
= (cs
<< bits
) % PAGE_CACHE_SIZE
;
421 while(cs
< max_slots
) {
422 current_page
= cs
/ spp
;
423 page
= reg
->hr_slot_data
[current_page
];
425 vec_len
= min(PAGE_CACHE_SIZE
- vec_start
,
426 (max_slots
-cs
) * (PAGE_CACHE_SIZE
/spp
) );
428 mlog(ML_HB_BIO
, "page %d, vec_len = %u, vec_start = %u\n",
429 current_page
, vec_len
, vec_start
);
431 len
= bio_add_page(bio
, page
, vec_len
, vec_start
);
432 if (len
!= vec_len
) break;
434 cs
+= vec_len
/ (PAGE_CACHE_SIZE
/spp
);
443 static int o2hb_read_slots(struct o2hb_region
*reg
,
444 unsigned int max_slots
)
446 unsigned int current_slot
=0;
448 struct o2hb_bio_wait_ctxt wc
;
451 o2hb_bio_wait_init(&wc
);
453 while(current_slot
< max_slots
) {
454 bio
= o2hb_setup_one_bio(reg
, &wc
, ¤t_slot
, max_slots
);
456 status
= PTR_ERR(bio
);
461 atomic_inc(&wc
.wc_num_reqs
);
462 submit_bio(READ
, bio
);
468 o2hb_wait_on_io(reg
, &wc
);
469 if (wc
.wc_error
&& !status
)
470 status
= wc
.wc_error
;
475 static int o2hb_issue_node_write(struct o2hb_region
*reg
,
476 struct o2hb_bio_wait_ctxt
*write_wc
)
482 o2hb_bio_wait_init(write_wc
);
484 slot
= o2nm_this_node();
486 bio
= o2hb_setup_one_bio(reg
, write_wc
, &slot
, slot
+1);
488 status
= PTR_ERR(bio
);
493 atomic_inc(&write_wc
->wc_num_reqs
);
494 submit_bio(WRITE_SYNC
, bio
);
501 static u32
o2hb_compute_block_crc_le(struct o2hb_region
*reg
,
502 struct o2hb_disk_heartbeat_block
*hb_block
)
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
510 old_cksum
= hb_block
->hb_cksum
;
511 hb_block
->hb_cksum
= 0;
513 ret
= crc32_le(0, (unsigned char *) hb_block
, reg
->hr_block_bytes
);
515 hb_block
->hb_cksum
= old_cksum
;
520 static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block
*hb_block
)
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
));
529 static int o2hb_verify_crc(struct o2hb_region
*reg
,
530 struct o2hb_disk_heartbeat_block
*hb_block
)
534 read
= le32_to_cpu(hb_block
->hb_cksum
);
535 computed
= o2hb_compute_block_crc_le(reg
, hb_block
);
537 return read
== computed
;
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.
547 static int o2hb_check_own_slot(struct o2hb_region
*reg
)
549 struct o2hb_disk_slot
*slot
;
550 struct o2hb_disk_heartbeat_block
*hb_block
;
553 slot
= ®
->hr_slots
[o2nm_this_node()];
554 /* Don't check on our 1st timestamp */
555 if (!slot
->ds_last_time
)
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
)
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"
568 if (hb_block
->hb_node
!= slot
->ds_node_num
)
570 else if (le64_to_cpu(hb_block
->hb_generation
) !=
571 slot
->ds_last_generation
)
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
));
586 static inline void o2hb_prepare_block(struct o2hb_region
*reg
,
591 struct o2hb_disk_slot
*slot
;
592 struct o2hb_disk_heartbeat_block
*hb_block
;
594 node_num
= o2nm_this_node();
595 slot
= ®
->hr_slots
[node_num
];
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
;
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
);
609 /* This step must always happen last! */
610 hb_block
->hb_cksum
= cpu_to_le32(o2hb_compute_block_crc_le(reg
,
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
));
618 static void o2hb_fire_callbacks(struct o2hb_callback
*hbcall
,
619 struct o2nm_node
*node
,
622 struct o2hb_callback_func
*f
;
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
);
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
)
633 struct o2hb_callback
*hbcall
;
634 struct o2hb_node_event
*event
;
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
);
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
,
647 list_del_init(&event
->hn_item
);
648 spin_unlock(&o2hb_live_lock
);
650 mlog(ML_HEARTBEAT
, "Node %s event for %d\n",
651 event
->hn_event_type
== O2HB_NODE_UP_CB
? "UP" : "DOWN",
654 hbcall
= hbcall_from_type(event
->hn_event_type
);
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
));
661 o2hb_fire_callbacks(hbcall
, event
->hn_node
, event
->hn_node_num
);
663 spin_lock(&o2hb_live_lock
);
665 spin_unlock(&o2hb_live_lock
);
667 up_write(&o2hb_callback_sem
);
670 static void o2hb_queue_node_event(struct o2hb_node_event
*event
,
671 enum o2hb_callback_type type
,
672 struct o2nm_node
*node
,
675 assert_spin_locked(&o2hb_live_lock
);
677 BUG_ON((!node
) && (type
!= O2HB_NODE_DOWN_CB
));
679 event
->hn_event_type
= type
;
680 event
->hn_node
= node
;
681 event
->hn_node_num
= node_num
;
683 mlog(ML_HEARTBEAT
, "Queue node %s event for node %d\n",
684 type
== O2HB_NODE_UP_CB
? "UP" : "DOWN", node_num
);
686 list_add_tail(&event
->hn_item
, &o2hb_node_events
);
689 static void o2hb_shutdown_slot(struct o2hb_disk_slot
*slot
)
691 struct o2hb_node_event event
=
692 { .hn_item
= LIST_HEAD_INIT(event
.hn_item
), };
693 struct o2nm_node
*node
;
696 node
= o2nm_get_node_by_num(slot
->ds_node_num
);
700 spin_lock(&o2hb_live_lock
);
701 if (!list_empty(&slot
->ds_live_item
)) {
702 mlog(ML_HEARTBEAT
, "Shutdown, node %d leaves region\n",
705 list_del_init(&slot
->ds_live_item
);
707 if (list_empty(&o2hb_live_slots
[slot
->ds_node_num
])) {
708 clear_bit(slot
->ds_node_num
, o2hb_live_node_bitmap
);
710 o2hb_queue_node_event(&event
, O2HB_NODE_DOWN_CB
, node
,
715 spin_unlock(&o2hb_live_lock
);
718 o2hb_run_event_list(&event
);
723 static void o2hb_set_quorum_device(struct o2hb_region
*reg
)
725 if (!o2hb_global_heartbeat_active())
728 /* Prevent race with o2hb_heartbeat_group_drop_item() */
729 if (kthread_should_stop())
732 /* Tag region as quorum only after thread reaches steady state */
733 if (atomic_read(®
->hr_steady_iterations
) != 0)
736 spin_lock(&o2hb_live_lock
);
738 if (test_bit(reg
->hr_region_num
, o2hb_quorum_region_bitmap
))
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.
746 if (memcmp(reg
->hr_live_node_bitmap
, o2hb_live_node_bitmap
,
747 sizeof(o2hb_live_node_bitmap
)))
750 printk(KERN_NOTICE
"o2hb: Region %s (%s) is now a quorum device\n",
751 config_item_name(®
->hr_item
), reg
->hr_dev_name
);
753 set_bit(reg
->hr_region_num
, o2hb_quorum_region_bitmap
);
756 * If global heartbeat active, unpin all regions if the
757 * region count > CUT_OFF
759 if (bitmap_weight(o2hb_quorum_region_bitmap
,
760 O2NM_MAX_REGIONS
) > O2HB_PIN_CUT_OFF
)
761 o2hb_region_unpin(NULL
);
763 spin_unlock(&o2hb_live_lock
);
766 static int o2hb_check_slot(struct o2hb_region
*reg
,
767 struct o2hb_disk_slot
*slot
)
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
;
775 unsigned int dead_ms
= o2hb_dead_threshold
* O2HB_REGION_TIMEOUT_MS
;
776 unsigned int slot_dead_ms
;
780 memcpy(hb_block
, slot
->ds_raw_block
, reg
->hr_block_bytes
);
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.
786 node
= o2nm_get_node_by_num(slot
->ds_node_num
);
788 spin_lock(&o2hb_live_lock
);
789 tmp
= test_bit(slot
->ds_node_num
, o2hb_live_node_bitmap
);
790 spin_unlock(&o2hb_live_lock
);
795 if (!o2hb_verify_crc(reg
, hb_block
)) {
796 /* all paths from here will drop o2hb_live_lock for
798 spin_lock(&o2hb_live_lock
);
800 /* Don't print an error on the console in this case -
801 * a freshly formatted heartbeat area will not have a
803 if (list_empty(&slot
->ds_live_item
))
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
);
813 slot
->ds_equal_samples
++;
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
++;
823 slot
->ds_equal_samples
++;
824 slot
->ds_last_time
= cputime
;
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
)) {
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
));
840 slot
->ds_last_generation
= le64_to_cpu(hb_block
->hb_generation
);
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
);
850 spin_lock(&o2hb_live_lock
);
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
);
860 set_bit(slot
->ds_node_num
, reg
->hr_live_node_bitmap
);
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
);
868 o2hb_queue_node_event(&event
, O2HB_NODE_UP_CB
, node
,
875 list_add_tail(&slot
->ds_live_item
,
876 &o2hb_live_slots
[slot
->ds_node_num
]);
878 slot
->ds_equal_samples
= 0;
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
,
899 /* if the list is dead, we're done.. */
900 if (list_empty(&slot
->ds_live_item
))
903 /* live nodes only go dead after enough consequtive missed
904 * samples.. reset the missed counter whenever we see
906 if (slot
->ds_equal_samples
>= o2hb_dead_threshold
|| gen_changed
) {
907 mlog(ML_HEARTBEAT
, "Node %d left my region\n",
910 clear_bit(slot
->ds_node_num
, reg
->hr_live_node_bitmap
);
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
);
919 /* node can be null */
920 o2hb_queue_node_event(&event
, O2HB_NODE_DOWN_CB
,
921 node
, slot
->ds_node_num
);
927 /* We don't clear this because the node is still
928 * actually writing new blocks. */
930 slot
->ds_changed_samples
= 0;
933 if (slot
->ds_changed_samples
) {
934 slot
->ds_changed_samples
= 0;
935 slot
->ds_equal_samples
= 0;
938 spin_unlock(&o2hb_live_lock
);
941 o2hb_run_event_list(&event
);
948 static int o2hb_highest_node(unsigned long *nodes
, int numbits
)
950 return find_last_bit(nodes
, numbits
);
953 static int o2hb_do_disk_heartbeat(struct o2hb_region
*reg
)
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
;
961 ret
= o2nm_configured_node_map(configured_nodes
,
962 sizeof(configured_nodes
));
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.
972 o2hb_fill_node_map(live_node_bitmap
, sizeof(live_node_bitmap
));
974 while ((i
= find_next_bit(live_node_bitmap
,
975 O2NM_MAX_NODES
, i
+ 1)) < O2NM_MAX_NODES
) {
976 set_bit(i
, configured_nodes
);
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");
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
990 ret
= o2hb_read_slots(reg
, highest_node
+ 1);
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
999 own_slot_ok
= o2hb_check_own_slot(reg
);
1001 /* fill in the proper info for our next heartbeat */
1002 o2hb_prepare_block(reg
, reg
->hr_generation
);
1004 ret
= o2hb_issue_node_write(reg
, &write_wc
);
1011 while((i
= find_next_bit(configured_nodes
,
1012 O2NM_MAX_NODES
, i
+ 1)) < O2NM_MAX_NODES
) {
1013 membership_change
|= o2hb_check_slot(reg
, ®
->hr_slots
[i
]);
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.
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
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
;
1032 /* Skip disarming the timeout if own slot has stale/bad data */
1034 o2hb_set_quorum_device(reg
);
1035 o2hb_arm_write_timeout(reg
);
1039 /* let the person who launched us know when things are steady */
1040 if (atomic_read(®
->hr_steady_iterations
) != 0) {
1041 if (!ret
&& own_slot_ok
&& !membership_change
) {
1042 if (atomic_dec_and_test(®
->hr_steady_iterations
))
1043 wake_up(&o2hb_steady_queue
);
1047 if (atomic_read(®
->hr_steady_iterations
) != 0) {
1048 if (atomic_dec_and_test(®
->hr_unsteady_iterations
)) {
1049 printk(KERN_NOTICE
"o2hb: Unable to stabilize "
1050 "heartbeart on region %s (%s)\n",
1051 config_item_name(®
->hr_item
),
1053 atomic_set(®
->hr_steady_iterations
, 0);
1054 reg
->hr_aborted_start
= 1;
1055 wake_up(&o2hb_steady_queue
);
1063 /* Subtract b from a, storing the result in a. a *must* have a larger
1065 static void o2hb_tv_subtract(struct timeval
*a
,
1068 /* just return 0 when a is after b */
1069 if (a
->tv_sec
< b
->tv_sec
||
1070 (a
->tv_sec
== b
->tv_sec
&& a
->tv_usec
< b
->tv_usec
)) {
1076 a
->tv_sec
-= b
->tv_sec
;
1077 a
->tv_usec
-= b
->tv_usec
;
1078 while ( a
->tv_usec
< 0 ) {
1080 a
->tv_usec
+= 1000000;
1084 static unsigned int o2hb_elapsed_msecs(struct timeval
*start
,
1085 struct timeval
*end
)
1087 struct timeval res
= *end
;
1089 o2hb_tv_subtract(&res
, start
);
1091 return res
.tv_sec
* 1000 + res
.tv_usec
/ 1000;
1095 * we ride the region ref that the region dir holds. before the region
1096 * dir is removed and drops it ref it will wait to tear down this
1099 static int o2hb_thread(void *data
)
1102 struct o2hb_region
*reg
= data
;
1103 struct o2hb_bio_wait_ctxt write_wc
;
1104 struct timeval before_hb
, after_hb
;
1105 unsigned int elapsed_msec
;
1107 mlog(ML_HEARTBEAT
|ML_KTHREAD
, "hb thread running\n");
1109 set_user_nice(current
, MIN_NICE
);
1112 o2nm_depend_this_node();
1114 while (!kthread_should_stop() &&
1115 !reg
->hr_unclean_stop
&& !reg
->hr_aborted_start
) {
1116 /* We track the time spent inside
1117 * o2hb_do_disk_heartbeat so that we avoid more than
1118 * hr_timeout_ms between disk writes. On busy systems
1119 * this should result in a heartbeat which is less
1120 * likely to time itself out. */
1121 do_gettimeofday(&before_hb
);
1123 ret
= o2hb_do_disk_heartbeat(reg
);
1125 do_gettimeofday(&after_hb
);
1126 elapsed_msec
= o2hb_elapsed_msecs(&before_hb
, &after_hb
);
1129 "start = %lu.%lu, end = %lu.%lu, msec = %u, ret = %d\n",
1130 before_hb
.tv_sec
, (unsigned long) before_hb
.tv_usec
,
1131 after_hb
.tv_sec
, (unsigned long) after_hb
.tv_usec
,
1134 if (!kthread_should_stop() &&
1135 elapsed_msec
< reg
->hr_timeout_ms
) {
1136 /* the kthread api has blocked signals for us so no
1137 * need to record the return value. */
1138 msleep_interruptible(reg
->hr_timeout_ms
- elapsed_msec
);
1142 o2hb_disarm_write_timeout(reg
);
1144 /* unclean stop is only used in very bad situation */
1145 for(i
= 0; !reg
->hr_unclean_stop
&& i
< reg
->hr_blocks
; i
++)
1146 o2hb_shutdown_slot(®
->hr_slots
[i
]);
1148 /* Explicit down notification - avoid forcing the other nodes
1149 * to timeout on this region when we could just as easily
1150 * write a clear generation - thus indicating to them that
1151 * this node has left this region.
1153 if (!reg
->hr_unclean_stop
&& !reg
->hr_aborted_start
) {
1154 o2hb_prepare_block(reg
, 0);
1155 ret
= o2hb_issue_node_write(reg
, &write_wc
);
1157 o2hb_wait_on_io(reg
, &write_wc
);
1163 o2nm_undepend_this_node();
1165 mlog(ML_HEARTBEAT
|ML_KTHREAD
, "o2hb thread exiting\n");
1170 #ifdef CONFIG_DEBUG_FS
1171 static int o2hb_debug_open(struct inode
*inode
, struct file
*file
)
1173 struct o2hb_debug_buf
*db
= inode
->i_private
;
1174 struct o2hb_region
*reg
;
1175 unsigned long map
[BITS_TO_LONGS(O2NM_MAX_NODES
)];
1181 /* max_nodes should be the largest bitmap we pass here */
1182 BUG_ON(sizeof(map
) < db
->db_size
);
1184 buf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
1188 switch (db
->db_type
) {
1189 case O2HB_DB_TYPE_LIVENODES
:
1190 case O2HB_DB_TYPE_LIVEREGIONS
:
1191 case O2HB_DB_TYPE_QUORUMREGIONS
:
1192 case O2HB_DB_TYPE_FAILEDREGIONS
:
1193 spin_lock(&o2hb_live_lock
);
1194 memcpy(map
, db
->db_data
, db
->db_size
);
1195 spin_unlock(&o2hb_live_lock
);
1198 case O2HB_DB_TYPE_REGION_LIVENODES
:
1199 spin_lock(&o2hb_live_lock
);
1200 reg
= (struct o2hb_region
*)db
->db_data
;
1201 memcpy(map
, reg
->hr_live_node_bitmap
, db
->db_size
);
1202 spin_unlock(&o2hb_live_lock
);
1205 case O2HB_DB_TYPE_REGION_NUMBER
:
1206 reg
= (struct o2hb_region
*)db
->db_data
;
1207 out
+= snprintf(buf
+ out
, PAGE_SIZE
- out
, "%d\n",
1208 reg
->hr_region_num
);
1211 case O2HB_DB_TYPE_REGION_ELAPSED_TIME
:
1212 reg
= (struct o2hb_region
*)db
->db_data
;
1213 lts
= reg
->hr_last_timeout_start
;
1214 /* If 0, it has never been set before */
1216 lts
= jiffies_to_msecs(jiffies
- lts
);
1217 out
+= snprintf(buf
+ out
, PAGE_SIZE
- out
, "%lu\n", lts
);
1220 case O2HB_DB_TYPE_REGION_PINNED
:
1221 reg
= (struct o2hb_region
*)db
->db_data
;
1222 out
+= snprintf(buf
+ out
, PAGE_SIZE
- out
, "%u\n",
1223 !!reg
->hr_item_pinned
);
1230 while ((i
= find_next_bit(map
, db
->db_len
, i
+ 1)) < db
->db_len
)
1231 out
+= snprintf(buf
+ out
, PAGE_SIZE
- out
, "%d ", i
);
1232 out
+= snprintf(buf
+ out
, PAGE_SIZE
- out
, "\n");
1235 i_size_write(inode
, out
);
1237 file
->private_data
= buf
;
1244 static int o2hb_debug_release(struct inode
*inode
, struct file
*file
)
1246 kfree(file
->private_data
);
1250 static ssize_t
o2hb_debug_read(struct file
*file
, char __user
*buf
,
1251 size_t nbytes
, loff_t
*ppos
)
1253 return simple_read_from_buffer(buf
, nbytes
, ppos
, file
->private_data
,
1254 i_size_read(file
->f_mapping
->host
));
1257 static int o2hb_debug_open(struct inode
*inode
, struct file
*file
)
1261 static int o2hb_debug_release(struct inode
*inode
, struct file
*file
)
1265 static ssize_t
o2hb_debug_read(struct file
*file
, char __user
*buf
,
1266 size_t nbytes
, loff_t
*ppos
)
1270 #endif /* CONFIG_DEBUG_FS */
1272 static const struct file_operations o2hb_debug_fops
= {
1273 .open
= o2hb_debug_open
,
1274 .release
= o2hb_debug_release
,
1275 .read
= o2hb_debug_read
,
1276 .llseek
= generic_file_llseek
,
1279 void o2hb_exit(void)
1281 kfree(o2hb_db_livenodes
);
1282 kfree(o2hb_db_liveregions
);
1283 kfree(o2hb_db_quorumregions
);
1284 kfree(o2hb_db_failedregions
);
1285 debugfs_remove(o2hb_debug_failedregions
);
1286 debugfs_remove(o2hb_debug_quorumregions
);
1287 debugfs_remove(o2hb_debug_liveregions
);
1288 debugfs_remove(o2hb_debug_livenodes
);
1289 debugfs_remove(o2hb_debug_dir
);
1292 static struct dentry
*o2hb_debug_create(const char *name
, struct dentry
*dir
,
1293 struct o2hb_debug_buf
**db
, int db_len
,
1294 int type
, int size
, int len
, void *data
)
1296 *db
= kmalloc(db_len
, GFP_KERNEL
);
1300 (*db
)->db_type
= type
;
1301 (*db
)->db_size
= size
;
1302 (*db
)->db_len
= len
;
1303 (*db
)->db_data
= data
;
1305 return debugfs_create_file(name
, S_IFREG
|S_IRUSR
, dir
, *db
,
1309 static int o2hb_debug_init(void)
1313 o2hb_debug_dir
= debugfs_create_dir(O2HB_DEBUG_DIR
, NULL
);
1314 if (!o2hb_debug_dir
) {
1319 o2hb_debug_livenodes
= o2hb_debug_create(O2HB_DEBUG_LIVENODES
,
1322 sizeof(*o2hb_db_livenodes
),
1323 O2HB_DB_TYPE_LIVENODES
,
1324 sizeof(o2hb_live_node_bitmap
),
1326 o2hb_live_node_bitmap
);
1327 if (!o2hb_debug_livenodes
) {
1332 o2hb_debug_liveregions
= o2hb_debug_create(O2HB_DEBUG_LIVEREGIONS
,
1334 &o2hb_db_liveregions
,
1335 sizeof(*o2hb_db_liveregions
),
1336 O2HB_DB_TYPE_LIVEREGIONS
,
1337 sizeof(o2hb_live_region_bitmap
),
1339 o2hb_live_region_bitmap
);
1340 if (!o2hb_debug_liveregions
) {
1345 o2hb_debug_quorumregions
=
1346 o2hb_debug_create(O2HB_DEBUG_QUORUMREGIONS
,
1348 &o2hb_db_quorumregions
,
1349 sizeof(*o2hb_db_quorumregions
),
1350 O2HB_DB_TYPE_QUORUMREGIONS
,
1351 sizeof(o2hb_quorum_region_bitmap
),
1353 o2hb_quorum_region_bitmap
);
1354 if (!o2hb_debug_quorumregions
) {
1359 o2hb_debug_failedregions
=
1360 o2hb_debug_create(O2HB_DEBUG_FAILEDREGIONS
,
1362 &o2hb_db_failedregions
,
1363 sizeof(*o2hb_db_failedregions
),
1364 O2HB_DB_TYPE_FAILEDREGIONS
,
1365 sizeof(o2hb_failed_region_bitmap
),
1367 o2hb_failed_region_bitmap
);
1368 if (!o2hb_debug_failedregions
) {
1385 for (i
= 0; i
< ARRAY_SIZE(o2hb_callbacks
); i
++)
1386 INIT_LIST_HEAD(&o2hb_callbacks
[i
].list
);
1388 for (i
= 0; i
< ARRAY_SIZE(o2hb_live_slots
); i
++)
1389 INIT_LIST_HEAD(&o2hb_live_slots
[i
]);
1391 INIT_LIST_HEAD(&o2hb_node_events
);
1393 memset(o2hb_live_node_bitmap
, 0, sizeof(o2hb_live_node_bitmap
));
1394 memset(o2hb_region_bitmap
, 0, sizeof(o2hb_region_bitmap
));
1395 memset(o2hb_live_region_bitmap
, 0, sizeof(o2hb_live_region_bitmap
));
1396 memset(o2hb_quorum_region_bitmap
, 0, sizeof(o2hb_quorum_region_bitmap
));
1397 memset(o2hb_failed_region_bitmap
, 0, sizeof(o2hb_failed_region_bitmap
));
1399 o2hb_dependent_users
= 0;
1401 return o2hb_debug_init();
1404 /* if we're already in a callback then we're already serialized by the sem */
1405 static void o2hb_fill_node_map_from_callback(unsigned long *map
,
1408 BUG_ON(bytes
< (BITS_TO_LONGS(O2NM_MAX_NODES
) * sizeof(unsigned long)));
1410 memcpy(map
, &o2hb_live_node_bitmap
, bytes
);
1414 * get a map of all nodes that are heartbeating in any regions
1416 void o2hb_fill_node_map(unsigned long *map
, unsigned bytes
)
1418 /* callers want to serialize this map and callbacks so that they
1419 * can trust that they don't miss nodes coming to the party */
1420 down_read(&o2hb_callback_sem
);
1421 spin_lock(&o2hb_live_lock
);
1422 o2hb_fill_node_map_from_callback(map
, bytes
);
1423 spin_unlock(&o2hb_live_lock
);
1424 up_read(&o2hb_callback_sem
);
1426 EXPORT_SYMBOL_GPL(o2hb_fill_node_map
);
1429 * heartbeat configfs bits. The heartbeat set is a default set under
1430 * the cluster set in nodemanager.c.
1433 static struct o2hb_region
*to_o2hb_region(struct config_item
*item
)
1435 return item
? container_of(item
, struct o2hb_region
, hr_item
) : NULL
;
1438 /* drop_item only drops its ref after killing the thread, nothing should
1439 * be using the region anymore. this has to clean up any state that
1440 * attributes might have built up. */
1441 static void o2hb_region_release(struct config_item
*item
)
1445 struct o2hb_region
*reg
= to_o2hb_region(item
);
1447 mlog(ML_HEARTBEAT
, "hb region release (%s)\n", reg
->hr_dev_name
);
1449 kfree(reg
->hr_tmp_block
);
1451 if (reg
->hr_slot_data
) {
1452 for (i
= 0; i
< reg
->hr_num_pages
; i
++) {
1453 page
= reg
->hr_slot_data
[i
];
1457 kfree(reg
->hr_slot_data
);
1461 blkdev_put(reg
->hr_bdev
, FMODE_READ
|FMODE_WRITE
);
1463 kfree(reg
->hr_slots
);
1465 kfree(reg
->hr_db_regnum
);
1466 kfree(reg
->hr_db_livenodes
);
1467 debugfs_remove(reg
->hr_debug_livenodes
);
1468 debugfs_remove(reg
->hr_debug_regnum
);
1469 debugfs_remove(reg
->hr_debug_elapsed_time
);
1470 debugfs_remove(reg
->hr_debug_pinned
);
1471 debugfs_remove(reg
->hr_debug_dir
);
1473 spin_lock(&o2hb_live_lock
);
1474 list_del(®
->hr_all_item
);
1475 spin_unlock(&o2hb_live_lock
);
1480 static int o2hb_read_block_input(struct o2hb_region
*reg
,
1483 unsigned long *ret_bytes
,
1484 unsigned int *ret_bits
)
1486 unsigned long bytes
;
1487 char *p
= (char *)page
;
1489 bytes
= simple_strtoul(p
, &p
, 0);
1490 if (!p
|| (*p
&& (*p
!= '\n')))
1493 /* Heartbeat and fs min / max block sizes are the same. */
1494 if (bytes
> 4096 || bytes
< 512)
1496 if (hweight16(bytes
) != 1)
1502 *ret_bits
= ffs(bytes
) - 1;
1507 static ssize_t
o2hb_region_block_bytes_read(struct o2hb_region
*reg
,
1510 return sprintf(page
, "%u\n", reg
->hr_block_bytes
);
1513 static ssize_t
o2hb_region_block_bytes_write(struct o2hb_region
*reg
,
1518 unsigned long block_bytes
;
1519 unsigned int block_bits
;
1524 status
= o2hb_read_block_input(reg
, page
, count
,
1525 &block_bytes
, &block_bits
);
1529 reg
->hr_block_bytes
= (unsigned int)block_bytes
;
1530 reg
->hr_block_bits
= block_bits
;
1535 static ssize_t
o2hb_region_start_block_read(struct o2hb_region
*reg
,
1538 return sprintf(page
, "%llu\n", reg
->hr_start_block
);
1541 static ssize_t
o2hb_region_start_block_write(struct o2hb_region
*reg
,
1545 unsigned long long tmp
;
1546 char *p
= (char *)page
;
1551 tmp
= simple_strtoull(p
, &p
, 0);
1552 if (!p
|| (*p
&& (*p
!= '\n')))
1555 reg
->hr_start_block
= tmp
;
1560 static ssize_t
o2hb_region_blocks_read(struct o2hb_region
*reg
,
1563 return sprintf(page
, "%d\n", reg
->hr_blocks
);
1566 static ssize_t
o2hb_region_blocks_write(struct o2hb_region
*reg
,
1571 char *p
= (char *)page
;
1576 tmp
= simple_strtoul(p
, &p
, 0);
1577 if (!p
|| (*p
&& (*p
!= '\n')))
1580 if (tmp
> O2NM_MAX_NODES
|| tmp
== 0)
1583 reg
->hr_blocks
= (unsigned int)tmp
;
1588 static ssize_t
o2hb_region_dev_read(struct o2hb_region
*reg
,
1591 unsigned int ret
= 0;
1594 ret
= sprintf(page
, "%s\n", reg
->hr_dev_name
);
1599 static void o2hb_init_region_params(struct o2hb_region
*reg
)
1601 reg
->hr_slots_per_page
= PAGE_CACHE_SIZE
>> reg
->hr_block_bits
;
1602 reg
->hr_timeout_ms
= O2HB_REGION_TIMEOUT_MS
;
1604 mlog(ML_HEARTBEAT
, "hr_start_block = %llu, hr_blocks = %u\n",
1605 reg
->hr_start_block
, reg
->hr_blocks
);
1606 mlog(ML_HEARTBEAT
, "hr_block_bytes = %u, hr_block_bits = %u\n",
1607 reg
->hr_block_bytes
, reg
->hr_block_bits
);
1608 mlog(ML_HEARTBEAT
, "hr_timeout_ms = %u\n", reg
->hr_timeout_ms
);
1609 mlog(ML_HEARTBEAT
, "dead threshold = %u\n", o2hb_dead_threshold
);
1612 static int o2hb_map_slot_data(struct o2hb_region
*reg
)
1615 unsigned int last_slot
;
1616 unsigned int spp
= reg
->hr_slots_per_page
;
1619 struct o2hb_disk_slot
*slot
;
1621 reg
->hr_tmp_block
= kmalloc(reg
->hr_block_bytes
, GFP_KERNEL
);
1622 if (reg
->hr_tmp_block
== NULL
)
1625 reg
->hr_slots
= kcalloc(reg
->hr_blocks
,
1626 sizeof(struct o2hb_disk_slot
), GFP_KERNEL
);
1627 if (reg
->hr_slots
== NULL
)
1630 for(i
= 0; i
< reg
->hr_blocks
; i
++) {
1631 slot
= ®
->hr_slots
[i
];
1632 slot
->ds_node_num
= i
;
1633 INIT_LIST_HEAD(&slot
->ds_live_item
);
1634 slot
->ds_raw_block
= NULL
;
1637 reg
->hr_num_pages
= (reg
->hr_blocks
+ spp
- 1) / spp
;
1638 mlog(ML_HEARTBEAT
, "Going to require %u pages to cover %u blocks "
1639 "at %u blocks per page\n",
1640 reg
->hr_num_pages
, reg
->hr_blocks
, spp
);
1642 reg
->hr_slot_data
= kcalloc(reg
->hr_num_pages
, sizeof(struct page
*),
1644 if (!reg
->hr_slot_data
)
1647 for(i
= 0; i
< reg
->hr_num_pages
; i
++) {
1648 page
= alloc_page(GFP_KERNEL
);
1652 reg
->hr_slot_data
[i
] = page
;
1654 last_slot
= i
* spp
;
1655 raw
= page_address(page
);
1657 (j
< spp
) && ((j
+ last_slot
) < reg
->hr_blocks
);
1659 BUG_ON((j
+ last_slot
) >= reg
->hr_blocks
);
1661 slot
= ®
->hr_slots
[j
+ last_slot
];
1662 slot
->ds_raw_block
=
1663 (struct o2hb_disk_heartbeat_block
*) raw
;
1665 raw
+= reg
->hr_block_bytes
;
1672 /* Read in all the slots available and populate the tracking
1673 * structures so that we can start with a baseline idea of what's
1675 static int o2hb_populate_slot_data(struct o2hb_region
*reg
)
1678 struct o2hb_disk_slot
*slot
;
1679 struct o2hb_disk_heartbeat_block
*hb_block
;
1681 ret
= o2hb_read_slots(reg
, reg
->hr_blocks
);
1685 /* We only want to get an idea of the values initially in each
1686 * slot, so we do no verification - o2hb_check_slot will
1687 * actually determine if each configured slot is valid and
1688 * whether any values have changed. */
1689 for(i
= 0; i
< reg
->hr_blocks
; i
++) {
1690 slot
= ®
->hr_slots
[i
];
1691 hb_block
= (struct o2hb_disk_heartbeat_block
*) slot
->ds_raw_block
;
1693 /* Only fill the values that o2hb_check_slot uses to
1694 * determine changing slots */
1695 slot
->ds_last_time
= le64_to_cpu(hb_block
->hb_seq
);
1696 slot
->ds_last_generation
= le64_to_cpu(hb_block
->hb_generation
);
1703 /* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */
1704 static ssize_t
o2hb_region_dev_write(struct o2hb_region
*reg
,
1708 struct task_struct
*hb_task
;
1711 char *p
= (char *)page
;
1713 struct inode
*inode
;
1714 ssize_t ret
= -EINVAL
;
1720 /* We can't heartbeat without having had our node number
1721 * configured yet. */
1722 if (o2nm_this_node() == O2NM_MAX_NODES
)
1725 fd
= simple_strtol(p
, &p
, 0);
1726 if (!p
|| (*p
&& (*p
!= '\n')))
1729 if (fd
< 0 || fd
>= INT_MAX
)
1736 if (reg
->hr_blocks
== 0 || reg
->hr_start_block
== 0 ||
1737 reg
->hr_block_bytes
== 0)
1740 inode
= igrab(f
.file
->f_mapping
->host
);
1744 if (!S_ISBLK(inode
->i_mode
))
1747 reg
->hr_bdev
= I_BDEV(f
.file
->f_mapping
->host
);
1748 ret
= blkdev_get(reg
->hr_bdev
, FMODE_WRITE
| FMODE_READ
, NULL
);
1750 reg
->hr_bdev
= NULL
;
1755 bdevname(reg
->hr_bdev
, reg
->hr_dev_name
);
1757 sectsize
= bdev_logical_block_size(reg
->hr_bdev
);
1758 if (sectsize
!= reg
->hr_block_bytes
) {
1760 "blocksize %u incorrect for device, expected %d",
1761 reg
->hr_block_bytes
, sectsize
);
1766 o2hb_init_region_params(reg
);
1768 /* Generation of zero is invalid */
1770 get_random_bytes(®
->hr_generation
,
1771 sizeof(reg
->hr_generation
));
1772 } while (reg
->hr_generation
== 0);
1774 ret
= o2hb_map_slot_data(reg
);
1780 ret
= o2hb_populate_slot_data(reg
);
1786 INIT_DELAYED_WORK(®
->hr_write_timeout_work
, o2hb_write_timeout
);
1789 * A node is considered live after it has beat LIVE_THRESHOLD
1790 * times. We're not steady until we've given them a chance
1791 * _after_ our first read.
1792 * The default threshold is bare minimum so as to limit the delay
1793 * during mounts. For global heartbeat, the threshold doubled for the
1796 live_threshold
= O2HB_LIVE_THRESHOLD
;
1797 if (o2hb_global_heartbeat_active()) {
1798 spin_lock(&o2hb_live_lock
);
1799 if (bitmap_weight(o2hb_region_bitmap
, O2NM_MAX_REGIONS
) == 1)
1800 live_threshold
<<= 1;
1801 spin_unlock(&o2hb_live_lock
);
1804 atomic_set(®
->hr_steady_iterations
, live_threshold
);
1805 /* unsteady_iterations is double the steady_iterations */
1806 atomic_set(®
->hr_unsteady_iterations
, (live_threshold
<< 1));
1808 hb_task
= kthread_run(o2hb_thread
, reg
, "o2hb-%s",
1809 reg
->hr_item
.ci_name
);
1810 if (IS_ERR(hb_task
)) {
1811 ret
= PTR_ERR(hb_task
);
1816 spin_lock(&o2hb_live_lock
);
1817 reg
->hr_task
= hb_task
;
1818 spin_unlock(&o2hb_live_lock
);
1820 ret
= wait_event_interruptible(o2hb_steady_queue
,
1821 atomic_read(®
->hr_steady_iterations
) == 0);
1823 atomic_set(®
->hr_steady_iterations
, 0);
1824 reg
->hr_aborted_start
= 1;
1827 if (reg
->hr_aborted_start
) {
1832 /* Ok, we were woken. Make sure it wasn't by drop_item() */
1833 spin_lock(&o2hb_live_lock
);
1834 hb_task
= reg
->hr_task
;
1835 if (o2hb_global_heartbeat_active())
1836 set_bit(reg
->hr_region_num
, o2hb_live_region_bitmap
);
1837 spin_unlock(&o2hb_live_lock
);
1844 if (hb_task
&& o2hb_global_heartbeat_active())
1845 printk(KERN_NOTICE
"o2hb: Heartbeat started on region %s (%s)\n",
1846 config_item_name(®
->hr_item
), reg
->hr_dev_name
);
1855 blkdev_put(reg
->hr_bdev
, FMODE_READ
|FMODE_WRITE
);
1856 reg
->hr_bdev
= NULL
;
1862 static ssize_t
o2hb_region_pid_read(struct o2hb_region
*reg
,
1867 spin_lock(&o2hb_live_lock
);
1869 pid
= task_pid_nr(reg
->hr_task
);
1870 spin_unlock(&o2hb_live_lock
);
1875 return sprintf(page
, "%u\n", pid
);
1878 struct o2hb_region_attribute
{
1879 struct configfs_attribute attr
;
1880 ssize_t (*show
)(struct o2hb_region
*, char *);
1881 ssize_t (*store
)(struct o2hb_region
*, const char *, size_t);
1884 static struct o2hb_region_attribute o2hb_region_attr_block_bytes
= {
1885 .attr
= { .ca_owner
= THIS_MODULE
,
1886 .ca_name
= "block_bytes",
1887 .ca_mode
= S_IRUGO
| S_IWUSR
},
1888 .show
= o2hb_region_block_bytes_read
,
1889 .store
= o2hb_region_block_bytes_write
,
1892 static struct o2hb_region_attribute o2hb_region_attr_start_block
= {
1893 .attr
= { .ca_owner
= THIS_MODULE
,
1894 .ca_name
= "start_block",
1895 .ca_mode
= S_IRUGO
| S_IWUSR
},
1896 .show
= o2hb_region_start_block_read
,
1897 .store
= o2hb_region_start_block_write
,
1900 static struct o2hb_region_attribute o2hb_region_attr_blocks
= {
1901 .attr
= { .ca_owner
= THIS_MODULE
,
1902 .ca_name
= "blocks",
1903 .ca_mode
= S_IRUGO
| S_IWUSR
},
1904 .show
= o2hb_region_blocks_read
,
1905 .store
= o2hb_region_blocks_write
,
1908 static struct o2hb_region_attribute o2hb_region_attr_dev
= {
1909 .attr
= { .ca_owner
= THIS_MODULE
,
1911 .ca_mode
= S_IRUGO
| S_IWUSR
},
1912 .show
= o2hb_region_dev_read
,
1913 .store
= o2hb_region_dev_write
,
1916 static struct o2hb_region_attribute o2hb_region_attr_pid
= {
1917 .attr
= { .ca_owner
= THIS_MODULE
,
1919 .ca_mode
= S_IRUGO
| S_IRUSR
},
1920 .show
= o2hb_region_pid_read
,
1923 static struct configfs_attribute
*o2hb_region_attrs
[] = {
1924 &o2hb_region_attr_block_bytes
.attr
,
1925 &o2hb_region_attr_start_block
.attr
,
1926 &o2hb_region_attr_blocks
.attr
,
1927 &o2hb_region_attr_dev
.attr
,
1928 &o2hb_region_attr_pid
.attr
,
1932 static ssize_t
o2hb_region_show(struct config_item
*item
,
1933 struct configfs_attribute
*attr
,
1936 struct o2hb_region
*reg
= to_o2hb_region(item
);
1937 struct o2hb_region_attribute
*o2hb_region_attr
=
1938 container_of(attr
, struct o2hb_region_attribute
, attr
);
1941 if (o2hb_region_attr
->show
)
1942 ret
= o2hb_region_attr
->show(reg
, page
);
1946 static ssize_t
o2hb_region_store(struct config_item
*item
,
1947 struct configfs_attribute
*attr
,
1948 const char *page
, size_t count
)
1950 struct o2hb_region
*reg
= to_o2hb_region(item
);
1951 struct o2hb_region_attribute
*o2hb_region_attr
=
1952 container_of(attr
, struct o2hb_region_attribute
, attr
);
1953 ssize_t ret
= -EINVAL
;
1955 if (o2hb_region_attr
->store
)
1956 ret
= o2hb_region_attr
->store(reg
, page
, count
);
1960 static struct configfs_item_operations o2hb_region_item_ops
= {
1961 .release
= o2hb_region_release
,
1962 .show_attribute
= o2hb_region_show
,
1963 .store_attribute
= o2hb_region_store
,
1966 static struct config_item_type o2hb_region_type
= {
1967 .ct_item_ops
= &o2hb_region_item_ops
,
1968 .ct_attrs
= o2hb_region_attrs
,
1969 .ct_owner
= THIS_MODULE
,
1974 struct o2hb_heartbeat_group
{
1975 struct config_group hs_group
;
1979 static struct o2hb_heartbeat_group
*to_o2hb_heartbeat_group(struct config_group
*group
)
1982 container_of(group
, struct o2hb_heartbeat_group
, hs_group
)
1986 static int o2hb_debug_region_init(struct o2hb_region
*reg
, struct dentry
*dir
)
1991 debugfs_create_dir(config_item_name(®
->hr_item
), dir
);
1992 if (!reg
->hr_debug_dir
) {
1997 reg
->hr_debug_livenodes
=
1998 o2hb_debug_create(O2HB_DEBUG_LIVENODES
,
2000 &(reg
->hr_db_livenodes
),
2001 sizeof(*(reg
->hr_db_livenodes
)),
2002 O2HB_DB_TYPE_REGION_LIVENODES
,
2003 sizeof(reg
->hr_live_node_bitmap
),
2004 O2NM_MAX_NODES
, reg
);
2005 if (!reg
->hr_debug_livenodes
) {
2010 reg
->hr_debug_regnum
=
2011 o2hb_debug_create(O2HB_DEBUG_REGION_NUMBER
,
2013 &(reg
->hr_db_regnum
),
2014 sizeof(*(reg
->hr_db_regnum
)),
2015 O2HB_DB_TYPE_REGION_NUMBER
,
2016 0, O2NM_MAX_NODES
, reg
);
2017 if (!reg
->hr_debug_regnum
) {
2022 reg
->hr_debug_elapsed_time
=
2023 o2hb_debug_create(O2HB_DEBUG_REGION_ELAPSED_TIME
,
2025 &(reg
->hr_db_elapsed_time
),
2026 sizeof(*(reg
->hr_db_elapsed_time
)),
2027 O2HB_DB_TYPE_REGION_ELAPSED_TIME
,
2029 if (!reg
->hr_debug_elapsed_time
) {
2034 reg
->hr_debug_pinned
=
2035 o2hb_debug_create(O2HB_DEBUG_REGION_PINNED
,
2037 &(reg
->hr_db_pinned
),
2038 sizeof(*(reg
->hr_db_pinned
)),
2039 O2HB_DB_TYPE_REGION_PINNED
,
2041 if (!reg
->hr_debug_pinned
) {
2051 static struct config_item
*o2hb_heartbeat_group_make_item(struct config_group
*group
,
2054 struct o2hb_region
*reg
= NULL
;
2057 reg
= kzalloc(sizeof(struct o2hb_region
), GFP_KERNEL
);
2059 return ERR_PTR(-ENOMEM
);
2061 if (strlen(name
) > O2HB_MAX_REGION_NAME_LEN
) {
2062 ret
= -ENAMETOOLONG
;
2066 spin_lock(&o2hb_live_lock
);
2067 reg
->hr_region_num
= 0;
2068 if (o2hb_global_heartbeat_active()) {
2069 reg
->hr_region_num
= find_first_zero_bit(o2hb_region_bitmap
,
2071 if (reg
->hr_region_num
>= O2NM_MAX_REGIONS
) {
2072 spin_unlock(&o2hb_live_lock
);
2076 set_bit(reg
->hr_region_num
, o2hb_region_bitmap
);
2078 list_add_tail(®
->hr_all_item
, &o2hb_all_regions
);
2079 spin_unlock(&o2hb_live_lock
);
2081 config_item_init_type_name(®
->hr_item
, name
, &o2hb_region_type
);
2083 ret
= o2hb_debug_region_init(reg
, o2hb_debug_dir
);
2085 config_item_put(®
->hr_item
);
2089 return ®
->hr_item
;
2092 return ERR_PTR(ret
);
2095 static void o2hb_heartbeat_group_drop_item(struct config_group
*group
,
2096 struct config_item
*item
)
2098 struct task_struct
*hb_task
;
2099 struct o2hb_region
*reg
= to_o2hb_region(item
);
2100 int quorum_region
= 0;
2102 /* stop the thread when the user removes the region dir */
2103 spin_lock(&o2hb_live_lock
);
2104 hb_task
= reg
->hr_task
;
2105 reg
->hr_task
= NULL
;
2106 reg
->hr_item_dropped
= 1;
2107 spin_unlock(&o2hb_live_lock
);
2110 kthread_stop(hb_task
);
2112 if (o2hb_global_heartbeat_active()) {
2113 spin_lock(&o2hb_live_lock
);
2114 clear_bit(reg
->hr_region_num
, o2hb_region_bitmap
);
2115 clear_bit(reg
->hr_region_num
, o2hb_live_region_bitmap
);
2116 if (test_bit(reg
->hr_region_num
, o2hb_quorum_region_bitmap
))
2118 clear_bit(reg
->hr_region_num
, o2hb_quorum_region_bitmap
);
2119 spin_unlock(&o2hb_live_lock
);
2120 printk(KERN_NOTICE
"o2hb: Heartbeat %s on region %s (%s)\n",
2121 ((atomic_read(®
->hr_steady_iterations
) == 0) ?
2122 "stopped" : "start aborted"), config_item_name(item
),
2127 * If we're racing a dev_write(), we need to wake them. They will
2128 * check reg->hr_task
2130 if (atomic_read(®
->hr_steady_iterations
) != 0) {
2131 reg
->hr_aborted_start
= 1;
2132 atomic_set(®
->hr_steady_iterations
, 0);
2133 wake_up(&o2hb_steady_queue
);
2136 config_item_put(item
);
2138 if (!o2hb_global_heartbeat_active() || !quorum_region
)
2142 * If global heartbeat active and there are dependent users,
2143 * pin all regions if quorum region count <= CUT_OFF
2145 spin_lock(&o2hb_live_lock
);
2147 if (!o2hb_dependent_users
)
2150 if (bitmap_weight(o2hb_quorum_region_bitmap
,
2151 O2NM_MAX_REGIONS
) <= O2HB_PIN_CUT_OFF
)
2152 o2hb_region_pin(NULL
);
2155 spin_unlock(&o2hb_live_lock
);
2158 struct o2hb_heartbeat_group_attribute
{
2159 struct configfs_attribute attr
;
2160 ssize_t (*show
)(struct o2hb_heartbeat_group
*, char *);
2161 ssize_t (*store
)(struct o2hb_heartbeat_group
*, const char *, size_t);
2164 static ssize_t
o2hb_heartbeat_group_show(struct config_item
*item
,
2165 struct configfs_attribute
*attr
,
2168 struct o2hb_heartbeat_group
*reg
= to_o2hb_heartbeat_group(to_config_group(item
));
2169 struct o2hb_heartbeat_group_attribute
*o2hb_heartbeat_group_attr
=
2170 container_of(attr
, struct o2hb_heartbeat_group_attribute
, attr
);
2173 if (o2hb_heartbeat_group_attr
->show
)
2174 ret
= o2hb_heartbeat_group_attr
->show(reg
, page
);
2178 static ssize_t
o2hb_heartbeat_group_store(struct config_item
*item
,
2179 struct configfs_attribute
*attr
,
2180 const char *page
, size_t count
)
2182 struct o2hb_heartbeat_group
*reg
= to_o2hb_heartbeat_group(to_config_group(item
));
2183 struct o2hb_heartbeat_group_attribute
*o2hb_heartbeat_group_attr
=
2184 container_of(attr
, struct o2hb_heartbeat_group_attribute
, attr
);
2185 ssize_t ret
= -EINVAL
;
2187 if (o2hb_heartbeat_group_attr
->store
)
2188 ret
= o2hb_heartbeat_group_attr
->store(reg
, page
, count
);
2192 static ssize_t
o2hb_heartbeat_group_threshold_show(struct o2hb_heartbeat_group
*group
,
2195 return sprintf(page
, "%u\n", o2hb_dead_threshold
);
2198 static ssize_t
o2hb_heartbeat_group_threshold_store(struct o2hb_heartbeat_group
*group
,
2203 char *p
= (char *)page
;
2205 tmp
= simple_strtoul(p
, &p
, 10);
2206 if (!p
|| (*p
&& (*p
!= '\n')))
2209 /* this will validate ranges for us. */
2210 o2hb_dead_threshold_set((unsigned int) tmp
);
2216 ssize_t
o2hb_heartbeat_group_mode_show(struct o2hb_heartbeat_group
*group
,
2219 return sprintf(page
, "%s\n",
2220 o2hb_heartbeat_mode_desc
[o2hb_heartbeat_mode
]);
2224 ssize_t
o2hb_heartbeat_group_mode_store(struct o2hb_heartbeat_group
*group
,
2225 const char *page
, size_t count
)
2231 len
= (page
[count
- 1] == '\n') ? count
- 1 : count
;
2235 for (i
= 0; i
< O2HB_HEARTBEAT_NUM_MODES
; ++i
) {
2236 if (strncasecmp(page
, o2hb_heartbeat_mode_desc
[i
], len
))
2239 ret
= o2hb_global_heartbeat_mode_set(i
);
2241 printk(KERN_NOTICE
"o2hb: Heartbeat mode set to %s\n",
2242 o2hb_heartbeat_mode_desc
[i
]);
2250 static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_threshold
= {
2251 .attr
= { .ca_owner
= THIS_MODULE
,
2252 .ca_name
= "dead_threshold",
2253 .ca_mode
= S_IRUGO
| S_IWUSR
},
2254 .show
= o2hb_heartbeat_group_threshold_show
,
2255 .store
= o2hb_heartbeat_group_threshold_store
,
2258 static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_mode
= {
2259 .attr
= { .ca_owner
= THIS_MODULE
,
2261 .ca_mode
= S_IRUGO
| S_IWUSR
},
2262 .show
= o2hb_heartbeat_group_mode_show
,
2263 .store
= o2hb_heartbeat_group_mode_store
,
2266 static struct configfs_attribute
*o2hb_heartbeat_group_attrs
[] = {
2267 &o2hb_heartbeat_group_attr_threshold
.attr
,
2268 &o2hb_heartbeat_group_attr_mode
.attr
,
2272 static struct configfs_item_operations o2hb_heartbeat_group_item_ops
= {
2273 .show_attribute
= o2hb_heartbeat_group_show
,
2274 .store_attribute
= o2hb_heartbeat_group_store
,
2277 static struct configfs_group_operations o2hb_heartbeat_group_group_ops
= {
2278 .make_item
= o2hb_heartbeat_group_make_item
,
2279 .drop_item
= o2hb_heartbeat_group_drop_item
,
2282 static struct config_item_type o2hb_heartbeat_group_type
= {
2283 .ct_group_ops
= &o2hb_heartbeat_group_group_ops
,
2284 .ct_item_ops
= &o2hb_heartbeat_group_item_ops
,
2285 .ct_attrs
= o2hb_heartbeat_group_attrs
,
2286 .ct_owner
= THIS_MODULE
,
2289 /* this is just here to avoid touching group in heartbeat.h which the
2290 * entire damn world #includes */
2291 struct config_group
*o2hb_alloc_hb_set(void)
2293 struct o2hb_heartbeat_group
*hs
= NULL
;
2294 struct config_group
*ret
= NULL
;
2296 hs
= kzalloc(sizeof(struct o2hb_heartbeat_group
), GFP_KERNEL
);
2300 config_group_init_type_name(&hs
->hs_group
, "heartbeat",
2301 &o2hb_heartbeat_group_type
);
2303 ret
= &hs
->hs_group
;
2310 void o2hb_free_hb_set(struct config_group
*group
)
2312 struct o2hb_heartbeat_group
*hs
= to_o2hb_heartbeat_group(group
);
2316 /* hb callback registration and issuing */
2318 static struct o2hb_callback
*hbcall_from_type(enum o2hb_callback_type type
)
2320 if (type
== O2HB_NUM_CB
)
2321 return ERR_PTR(-EINVAL
);
2323 return &o2hb_callbacks
[type
];
2326 void o2hb_setup_callback(struct o2hb_callback_func
*hc
,
2327 enum o2hb_callback_type type
,
2332 INIT_LIST_HEAD(&hc
->hc_item
);
2335 hc
->hc_priority
= priority
;
2337 hc
->hc_magic
= O2HB_CB_MAGIC
;
2339 EXPORT_SYMBOL_GPL(o2hb_setup_callback
);
2342 * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2343 * In global heartbeat mode, region_uuid passed is NULL.
2345 * In local, we only pin the matching region. In global we pin all the active
2348 static int o2hb_region_pin(const char *region_uuid
)
2350 int ret
= 0, found
= 0;
2351 struct o2hb_region
*reg
;
2354 assert_spin_locked(&o2hb_live_lock
);
2356 list_for_each_entry(reg
, &o2hb_all_regions
, hr_all_item
) {
2357 if (reg
->hr_item_dropped
)
2360 uuid
= config_item_name(®
->hr_item
);
2362 /* local heartbeat */
2364 if (strcmp(region_uuid
, uuid
))
2369 if (reg
->hr_item_pinned
|| reg
->hr_item_dropped
)
2372 /* Ignore ENOENT only for local hb (userdlm domain) */
2373 ret
= o2nm_depend_item(®
->hr_item
);
2375 mlog(ML_CLUSTER
, "Pin region %s\n", uuid
);
2376 reg
->hr_item_pinned
= 1;
2378 if (ret
== -ENOENT
&& found
)
2381 mlog(ML_ERROR
, "Pin region %s fails with %d\n",
2395 * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2396 * In global heartbeat mode, region_uuid passed is NULL.
2398 * In local, we only unpin the matching region. In global we unpin all the
2401 static void o2hb_region_unpin(const char *region_uuid
)
2403 struct o2hb_region
*reg
;
2407 assert_spin_locked(&o2hb_live_lock
);
2409 list_for_each_entry(reg
, &o2hb_all_regions
, hr_all_item
) {
2410 if (reg
->hr_item_dropped
)
2413 uuid
= config_item_name(®
->hr_item
);
2415 if (strcmp(region_uuid
, uuid
))
2420 if (reg
->hr_item_pinned
) {
2421 mlog(ML_CLUSTER
, "Unpin region %s\n", uuid
);
2422 o2nm_undepend_item(®
->hr_item
);
2423 reg
->hr_item_pinned
= 0;
2430 static int o2hb_region_inc_user(const char *region_uuid
)
2434 spin_lock(&o2hb_live_lock
);
2436 /* local heartbeat */
2437 if (!o2hb_global_heartbeat_active()) {
2438 ret
= o2hb_region_pin(region_uuid
);
2443 * if global heartbeat active and this is the first dependent user,
2444 * pin all regions if quorum region count <= CUT_OFF
2446 o2hb_dependent_users
++;
2447 if (o2hb_dependent_users
> 1)
2450 if (bitmap_weight(o2hb_quorum_region_bitmap
,
2451 O2NM_MAX_REGIONS
) <= O2HB_PIN_CUT_OFF
)
2452 ret
= o2hb_region_pin(NULL
);
2455 spin_unlock(&o2hb_live_lock
);
2459 void o2hb_region_dec_user(const char *region_uuid
)
2461 spin_lock(&o2hb_live_lock
);
2463 /* local heartbeat */
2464 if (!o2hb_global_heartbeat_active()) {
2465 o2hb_region_unpin(region_uuid
);
2470 * if global heartbeat active and there are no dependent users,
2471 * unpin all quorum regions
2473 o2hb_dependent_users
--;
2474 if (!o2hb_dependent_users
)
2475 o2hb_region_unpin(NULL
);
2478 spin_unlock(&o2hb_live_lock
);
2481 int o2hb_register_callback(const char *region_uuid
,
2482 struct o2hb_callback_func
*hc
)
2484 struct o2hb_callback_func
*f
;
2485 struct o2hb_callback
*hbcall
;
2488 BUG_ON(hc
->hc_magic
!= O2HB_CB_MAGIC
);
2489 BUG_ON(!list_empty(&hc
->hc_item
));
2491 hbcall
= hbcall_from_type(hc
->hc_type
);
2492 if (IS_ERR(hbcall
)) {
2493 ret
= PTR_ERR(hbcall
);
2498 ret
= o2hb_region_inc_user(region_uuid
);
2505 down_write(&o2hb_callback_sem
);
2507 list_for_each_entry(f
, &hbcall
->list
, hc_item
) {
2508 if (hc
->hc_priority
< f
->hc_priority
) {
2509 list_add_tail(&hc
->hc_item
, &f
->hc_item
);
2513 if (list_empty(&hc
->hc_item
))
2514 list_add_tail(&hc
->hc_item
, &hbcall
->list
);
2516 up_write(&o2hb_callback_sem
);
2519 mlog(ML_CLUSTER
, "returning %d on behalf of %p for funcs %p\n",
2520 ret
, __builtin_return_address(0), hc
);
2523 EXPORT_SYMBOL_GPL(o2hb_register_callback
);
2525 void o2hb_unregister_callback(const char *region_uuid
,
2526 struct o2hb_callback_func
*hc
)
2528 BUG_ON(hc
->hc_magic
!= O2HB_CB_MAGIC
);
2530 mlog(ML_CLUSTER
, "on behalf of %p for funcs %p\n",
2531 __builtin_return_address(0), hc
);
2533 /* XXX Can this happen _with_ a region reference? */
2534 if (list_empty(&hc
->hc_item
))
2538 o2hb_region_dec_user(region_uuid
);
2540 down_write(&o2hb_callback_sem
);
2542 list_del_init(&hc
->hc_item
);
2544 up_write(&o2hb_callback_sem
);
2546 EXPORT_SYMBOL_GPL(o2hb_unregister_callback
);
2548 int o2hb_check_node_heartbeating(u8 node_num
)
2550 unsigned long testing_map
[BITS_TO_LONGS(O2NM_MAX_NODES
)];
2552 o2hb_fill_node_map(testing_map
, sizeof(testing_map
));
2553 if (!test_bit(node_num
, testing_map
)) {
2555 "node (%u) does not have heartbeating enabled.\n",
2562 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating
);
2564 int o2hb_check_node_heartbeating_no_sem(u8 node_num
)
2566 unsigned long testing_map
[BITS_TO_LONGS(O2NM_MAX_NODES
)];
2567 unsigned long flags
;
2569 spin_lock_irqsave(&o2hb_live_lock
, flags
);
2570 o2hb_fill_node_map_from_callback(testing_map
, sizeof(testing_map
));
2571 spin_unlock_irqrestore(&o2hb_live_lock
, flags
);
2572 if (!test_bit(node_num
, testing_map
)) {
2574 "node (%u) does not have heartbeating enabled.\n",
2581 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_no_sem
);
2583 int o2hb_check_node_heartbeating_from_callback(u8 node_num
)
2585 unsigned long testing_map
[BITS_TO_LONGS(O2NM_MAX_NODES
)];
2587 o2hb_fill_node_map_from_callback(testing_map
, sizeof(testing_map
));
2588 if (!test_bit(node_num
, testing_map
)) {
2590 "node (%u) does not have heartbeating enabled.\n",
2597 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback
);
2599 /* Makes sure our local node is configured with a node number, and is
2601 int o2hb_check_local_node_heartbeating(void)
2605 /* if this node was set then we have networking */
2606 node_num
= o2nm_this_node();
2607 if (node_num
== O2NM_MAX_NODES
) {
2608 mlog(ML_HEARTBEAT
, "this node has not been configured.\n");
2612 return o2hb_check_node_heartbeating(node_num
);
2614 EXPORT_SYMBOL_GPL(o2hb_check_local_node_heartbeating
);
2617 * this is just a hack until we get the plumbing which flips file systems
2618 * read only and drops the hb ref instead of killing the node dead.
2620 void o2hb_stop_all_regions(void)
2622 struct o2hb_region
*reg
;
2624 mlog(ML_ERROR
, "stopping heartbeat on all active regions.\n");
2626 spin_lock(&o2hb_live_lock
);
2628 list_for_each_entry(reg
, &o2hb_all_regions
, hr_all_item
)
2629 reg
->hr_unclean_stop
= 1;
2631 spin_unlock(&o2hb_live_lock
);
2633 EXPORT_SYMBOL_GPL(o2hb_stop_all_regions
);
2635 int o2hb_get_all_regions(char *region_uuids
, u8 max_regions
)
2637 struct o2hb_region
*reg
;
2641 spin_lock(&o2hb_live_lock
);
2644 list_for_each_entry(reg
, &o2hb_all_regions
, hr_all_item
) {
2645 if (reg
->hr_item_dropped
)
2648 mlog(0, "Region: %s\n", config_item_name(®
->hr_item
));
2649 if (numregs
< max_regions
) {
2650 memcpy(p
, config_item_name(®
->hr_item
),
2651 O2HB_MAX_REGION_NAME_LEN
);
2652 p
+= O2HB_MAX_REGION_NAME_LEN
;
2657 spin_unlock(&o2hb_live_lock
);
2661 EXPORT_SYMBOL_GPL(o2hb_get_all_regions
);
2663 int o2hb_global_heartbeat_active(void)
2665 return (o2hb_heartbeat_mode
== O2HB_HEARTBEAT_GLOBAL
);
2667 EXPORT_SYMBOL(o2hb_global_heartbeat_active
);