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>
37 #include "heartbeat.h"
39 #include "nodemanager.h"
46 * The first heartbeat pass had one global thread that would serialize all hb
47 * callback calls. This global serializing sem should only be removed once
48 * we've made sure that all callees can deal with being called concurrently
49 * from multiple hb region threads.
51 static DECLARE_RWSEM(o2hb_callback_sem
);
54 * multiple hb threads are watching multiple regions. A node is live
55 * whenever any of the threads sees activity from the node in its region.
57 static spinlock_t o2hb_live_lock
= SPIN_LOCK_UNLOCKED
;
58 static struct list_head o2hb_live_slots
[O2NM_MAX_NODES
];
59 static unsigned long o2hb_live_node_bitmap
[BITS_TO_LONGS(O2NM_MAX_NODES
)];
60 static LIST_HEAD(o2hb_node_events
);
61 static DECLARE_WAIT_QUEUE_HEAD(o2hb_steady_queue
);
63 static LIST_HEAD(o2hb_all_regions
);
65 static struct o2hb_callback
{
66 struct list_head list
;
67 } o2hb_callbacks
[O2HB_NUM_CB
];
69 static struct o2hb_callback
*hbcall_from_type(enum o2hb_callback_type type
);
71 #define O2HB_DEFAULT_BLOCK_BITS 9
73 unsigned int o2hb_dead_threshold
= O2HB_DEFAULT_DEAD_THRESHOLD
;
75 /* Only sets a new threshold if there are no active regions.
77 * No locking or otherwise interesting code is required for reading
78 * o2hb_dead_threshold as it can't change once regions are active and
79 * it's not interesting to anyone until then anyway. */
80 static void o2hb_dead_threshold_set(unsigned int threshold
)
82 if (threshold
> O2HB_MIN_DEAD_THRESHOLD
) {
83 spin_lock(&o2hb_live_lock
);
84 if (list_empty(&o2hb_all_regions
))
85 o2hb_dead_threshold
= threshold
;
86 spin_unlock(&o2hb_live_lock
);
90 struct o2hb_node_event
{
91 struct list_head hn_item
;
92 enum o2hb_callback_type hn_event_type
;
93 struct o2nm_node
*hn_node
;
97 struct o2hb_disk_slot
{
98 struct o2hb_disk_heartbeat_block
*ds_raw_block
;
101 u64 ds_last_generation
;
102 u16 ds_equal_samples
;
103 u16 ds_changed_samples
;
104 struct list_head ds_live_item
;
107 /* each thread owns a region.. when we're asked to tear down the region
108 * we ask the thread to stop, who cleans up the region */
110 struct config_item hr_item
;
112 struct list_head hr_all_item
;
113 unsigned hr_unclean_stop
:1;
115 /* protected by the hr_callback_sem */
116 struct task_struct
*hr_task
;
118 unsigned int hr_blocks
;
119 unsigned long long hr_start_block
;
121 unsigned int hr_block_bits
;
122 unsigned int hr_block_bytes
;
124 unsigned int hr_slots_per_page
;
125 unsigned int hr_num_pages
;
127 struct page
**hr_slot_data
;
128 struct block_device
*hr_bdev
;
129 struct o2hb_disk_slot
*hr_slots
;
131 /* let the person setting up hb wait for it to return until it
132 * has reached a 'steady' state. This will be fixed when we have
133 * a more complete api that doesn't lead to this sort of fragility. */
134 atomic_t hr_steady_iterations
;
136 char hr_dev_name
[BDEVNAME_SIZE
];
138 unsigned int hr_timeout_ms
;
140 /* randomized as the region goes up and down so that a node
141 * recognizes a node going up and down in one iteration */
144 struct work_struct hr_write_timeout_work
;
145 unsigned long hr_last_timeout_start
;
147 /* Used during o2hb_check_slot to hold a copy of the block
148 * being checked because we temporarily have to zero out the
150 struct o2hb_disk_heartbeat_block
*hr_tmp_block
;
153 struct o2hb_bio_wait_ctxt
{
154 atomic_t wc_num_reqs
;
155 struct completion wc_io_complete
;
158 static void o2hb_write_timeout(void *arg
)
160 struct o2hb_region
*reg
= arg
;
162 mlog(ML_ERROR
, "Heartbeat write timeout to device %s after %u "
163 "milliseconds\n", reg
->hr_dev_name
,
164 jiffies_to_msecs(jiffies
- reg
->hr_last_timeout_start
));
165 o2quo_disk_timeout();
168 static void o2hb_arm_write_timeout(struct o2hb_region
*reg
)
170 mlog(0, "Queue write timeout for %u ms\n", O2HB_MAX_WRITE_TIMEOUT_MS
);
172 cancel_delayed_work(®
->hr_write_timeout_work
);
173 reg
->hr_last_timeout_start
= jiffies
;
174 schedule_delayed_work(®
->hr_write_timeout_work
,
175 msecs_to_jiffies(O2HB_MAX_WRITE_TIMEOUT_MS
));
178 static void o2hb_disarm_write_timeout(struct o2hb_region
*reg
)
180 cancel_delayed_work(®
->hr_write_timeout_work
);
181 flush_scheduled_work();
184 static inline void o2hb_bio_wait_init(struct o2hb_bio_wait_ctxt
*wc
,
185 unsigned int num_ios
)
187 atomic_set(&wc
->wc_num_reqs
, num_ios
);
188 init_completion(&wc
->wc_io_complete
);
191 /* Used in error paths too */
192 static inline void o2hb_bio_wait_dec(struct o2hb_bio_wait_ctxt
*wc
,
195 /* sadly atomic_sub_and_test() isn't available on all platforms. The
196 * good news is that the fast path only completes one at a time */
198 if (atomic_dec_and_test(&wc
->wc_num_reqs
)) {
200 complete(&wc
->wc_io_complete
);
205 static void o2hb_wait_on_io(struct o2hb_region
*reg
,
206 struct o2hb_bio_wait_ctxt
*wc
)
208 struct address_space
*mapping
= reg
->hr_bdev
->bd_inode
->i_mapping
;
210 blk_run_address_space(mapping
);
212 wait_for_completion(&wc
->wc_io_complete
);
215 static int o2hb_bio_end_io(struct bio
*bio
,
216 unsigned int bytes_done
,
219 struct o2hb_bio_wait_ctxt
*wc
= bio
->bi_private
;
222 mlog(ML_ERROR
, "IO Error %d\n", error
);
227 o2hb_bio_wait_dec(wc
, 1);
231 /* Setup a Bio to cover I/O against num_slots slots starting at
233 static struct bio
*o2hb_setup_one_bio(struct o2hb_region
*reg
,
234 struct o2hb_bio_wait_ctxt
*wc
,
235 unsigned int start_slot
,
236 unsigned int num_slots
)
238 int i
, nr_vecs
, len
, first_page
, last_page
;
239 unsigned int vec_len
, vec_start
;
240 unsigned int bits
= reg
->hr_block_bits
;
241 unsigned int spp
= reg
->hr_slots_per_page
;
245 nr_vecs
= (num_slots
+ spp
- 1) / spp
;
247 /* Testing has shown this allocation to take long enough under
248 * GFP_KERNEL that the local node can get fenced. It would be
249 * nicest if we could pre-allocate these bios and avoid this
251 bio
= bio_alloc(GFP_ATOMIC
, nr_vecs
);
253 mlog(ML_ERROR
, "Could not alloc slots BIO!\n");
254 bio
= ERR_PTR(-ENOMEM
);
258 /* Must put everything in 512 byte sectors for the bio... */
259 bio
->bi_sector
= (reg
->hr_start_block
+ start_slot
) << (bits
- 9);
260 bio
->bi_bdev
= reg
->hr_bdev
;
261 bio
->bi_private
= wc
;
262 bio
->bi_end_io
= o2hb_bio_end_io
;
264 first_page
= start_slot
/ spp
;
265 last_page
= first_page
+ nr_vecs
;
266 vec_start
= (start_slot
<< bits
) % PAGE_CACHE_SIZE
;
267 for(i
= first_page
; i
< last_page
; i
++) {
268 page
= reg
->hr_slot_data
[i
];
270 vec_len
= PAGE_CACHE_SIZE
;
271 /* last page might be short */
272 if (((i
+ 1) * spp
) > (start_slot
+ num_slots
))
273 vec_len
= ((num_slots
+ start_slot
) % spp
) << bits
;
274 vec_len
-= vec_start
;
276 mlog(ML_HB_BIO
, "page %d, vec_len = %u, vec_start = %u\n",
277 i
, vec_len
, vec_start
);
279 len
= bio_add_page(bio
, page
, vec_len
, vec_start
);
280 if (len
!= vec_len
) {
284 mlog(ML_ERROR
, "Error adding page to bio i = %d, "
285 "vec_len = %u, len = %d\n, start = %u\n",
286 i
, vec_len
, len
, vec_start
);
298 * Compute the maximum number of sectors the bdev can handle in one bio,
301 * Stolen from oracleasm, thanks Joel!
303 static int compute_max_sectors(struct block_device
*bdev
)
305 int max_pages
, max_sectors
, pow_two_sectors
;
307 struct request_queue
*q
;
309 q
= bdev_get_queue(bdev
);
310 max_pages
= q
->max_sectors
>> (PAGE_SHIFT
- 9);
311 if (max_pages
> BIO_MAX_PAGES
)
312 max_pages
= BIO_MAX_PAGES
;
313 if (max_pages
> q
->max_phys_segments
)
314 max_pages
= q
->max_phys_segments
;
315 if (max_pages
> q
->max_hw_segments
)
316 max_pages
= q
->max_hw_segments
;
317 max_pages
--; /* Handle I/Os that straddle a page */
319 max_sectors
= max_pages
<< (PAGE_SHIFT
- 9);
321 /* Why is fls() 1-based???? */
322 pow_two_sectors
= 1 << (fls(max_sectors
) - 1);
324 return pow_two_sectors
;
327 static inline void o2hb_compute_request_limits(struct o2hb_region
*reg
,
328 unsigned int num_slots
,
329 unsigned int *num_bios
,
330 unsigned int *slots_per_bio
)
332 unsigned int max_sectors
, io_sectors
;
334 max_sectors
= compute_max_sectors(reg
->hr_bdev
);
336 io_sectors
= num_slots
<< (reg
->hr_block_bits
- 9);
338 *num_bios
= (io_sectors
+ max_sectors
- 1) / max_sectors
;
339 *slots_per_bio
= max_sectors
>> (reg
->hr_block_bits
- 9);
341 mlog(ML_HB_BIO
, "My io size is %u sectors for %u slots. This "
342 "device can handle %u sectors of I/O\n", io_sectors
, num_slots
,
344 mlog(ML_HB_BIO
, "Will need %u bios holding %u slots each\n",
345 *num_bios
, *slots_per_bio
);
348 static int o2hb_read_slots(struct o2hb_region
*reg
,
349 unsigned int max_slots
)
351 unsigned int num_bios
, slots_per_bio
, start_slot
, num_slots
;
353 struct o2hb_bio_wait_ctxt wc
;
357 o2hb_compute_request_limits(reg
, max_slots
, &num_bios
, &slots_per_bio
);
359 bios
= kcalloc(num_bios
, sizeof(struct bio
*), GFP_KERNEL
);
366 o2hb_bio_wait_init(&wc
, num_bios
);
368 num_slots
= slots_per_bio
;
369 for(i
= 0; i
< num_bios
; i
++) {
370 start_slot
= i
* slots_per_bio
;
372 /* adjust num_slots at last bio */
373 if (max_slots
< (start_slot
+ num_slots
))
374 num_slots
= max_slots
- start_slot
;
376 bio
= o2hb_setup_one_bio(reg
, &wc
, start_slot
, num_slots
);
378 o2hb_bio_wait_dec(&wc
, num_bios
- i
);
380 status
= PTR_ERR(bio
);
386 submit_bio(READ
, bio
);
392 o2hb_wait_on_io(reg
, &wc
);
395 for(i
= 0; i
< num_bios
; i
++)
404 static int o2hb_issue_node_write(struct o2hb_region
*reg
,
405 struct bio
**write_bio
,
406 struct o2hb_bio_wait_ctxt
*write_wc
)
412 o2hb_bio_wait_init(write_wc
, 1);
414 slot
= o2nm_this_node();
416 bio
= o2hb_setup_one_bio(reg
, write_wc
, slot
, 1);
418 status
= PTR_ERR(bio
);
423 submit_bio(WRITE
, bio
);
431 static u32
o2hb_compute_block_crc_le(struct o2hb_region
*reg
,
432 struct o2hb_disk_heartbeat_block
*hb_block
)
437 /* We want to compute the block crc with a 0 value in the
438 * hb_cksum field. Save it off here and replace after the
440 old_cksum
= hb_block
->hb_cksum
;
441 hb_block
->hb_cksum
= 0;
443 ret
= crc32_le(0, (unsigned char *) hb_block
, reg
->hr_block_bytes
);
445 hb_block
->hb_cksum
= old_cksum
;
450 static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block
*hb_block
)
452 mlog(ML_ERROR
, "Dump slot information: seq = 0x%llx, node = %u, "
453 "cksum = 0x%x, generation 0x%llx\n",
454 (long long)le64_to_cpu(hb_block
->hb_seq
),
455 hb_block
->hb_node
, le32_to_cpu(hb_block
->hb_cksum
),
456 (long long)le64_to_cpu(hb_block
->hb_generation
));
459 static int o2hb_verify_crc(struct o2hb_region
*reg
,
460 struct o2hb_disk_heartbeat_block
*hb_block
)
464 read
= le32_to_cpu(hb_block
->hb_cksum
);
465 computed
= o2hb_compute_block_crc_le(reg
, hb_block
);
467 return read
== computed
;
470 /* We want to make sure that nobody is heartbeating on top of us --
471 * this will help detect an invalid configuration. */
472 static int o2hb_check_last_timestamp(struct o2hb_region
*reg
)
475 struct o2hb_disk_slot
*slot
;
476 struct o2hb_disk_heartbeat_block
*hb_block
;
478 node_num
= o2nm_this_node();
481 slot
= ®
->hr_slots
[node_num
];
482 /* Don't check on our 1st timestamp */
483 if (slot
->ds_last_time
) {
484 hb_block
= slot
->ds_raw_block
;
486 if (le64_to_cpu(hb_block
->hb_seq
) != slot
->ds_last_time
)
493 static inline void o2hb_prepare_block(struct o2hb_region
*reg
,
498 struct o2hb_disk_slot
*slot
;
499 struct o2hb_disk_heartbeat_block
*hb_block
;
501 node_num
= o2nm_this_node();
502 slot
= ®
->hr_slots
[node_num
];
504 hb_block
= (struct o2hb_disk_heartbeat_block
*)slot
->ds_raw_block
;
505 memset(hb_block
, 0, reg
->hr_block_bytes
);
506 /* TODO: time stuff */
507 cputime
= CURRENT_TIME
.tv_sec
;
511 hb_block
->hb_seq
= cpu_to_le64(cputime
);
512 hb_block
->hb_node
= node_num
;
513 hb_block
->hb_generation
= cpu_to_le64(generation
);
515 /* This step must always happen last! */
516 hb_block
->hb_cksum
= cpu_to_le32(o2hb_compute_block_crc_le(reg
,
519 mlog(ML_HB_BIO
, "our node generation = 0x%llx, cksum = 0x%x\n",
520 (long long)cpu_to_le64(generation
),
521 le32_to_cpu(hb_block
->hb_cksum
));
524 static void o2hb_fire_callbacks(struct o2hb_callback
*hbcall
,
525 struct o2nm_node
*node
,
528 struct list_head
*iter
;
529 struct o2hb_callback_func
*f
;
531 list_for_each(iter
, &hbcall
->list
) {
532 f
= list_entry(iter
, struct o2hb_callback_func
, hc_item
);
533 mlog(ML_HEARTBEAT
, "calling funcs %p\n", f
);
534 (f
->hc_func
)(node
, idx
, f
->hc_data
);
538 /* Will run the list in order until we process the passed event */
539 static void o2hb_run_event_list(struct o2hb_node_event
*queued_event
)
542 struct o2hb_callback
*hbcall
;
543 struct o2hb_node_event
*event
;
545 spin_lock(&o2hb_live_lock
);
546 empty
= list_empty(&queued_event
->hn_item
);
547 spin_unlock(&o2hb_live_lock
);
551 /* Holding callback sem assures we don't alter the callback
552 * lists when doing this, and serializes ourselves with other
553 * processes wanting callbacks. */
554 down_write(&o2hb_callback_sem
);
556 spin_lock(&o2hb_live_lock
);
557 while (!list_empty(&o2hb_node_events
)
558 && !list_empty(&queued_event
->hn_item
)) {
559 event
= list_entry(o2hb_node_events
.next
,
560 struct o2hb_node_event
,
562 list_del_init(&event
->hn_item
);
563 spin_unlock(&o2hb_live_lock
);
565 mlog(ML_HEARTBEAT
, "Node %s event for %d\n",
566 event
->hn_event_type
== O2HB_NODE_UP_CB
? "UP" : "DOWN",
569 hbcall
= hbcall_from_type(event
->hn_event_type
);
571 /* We should *never* have gotten on to the list with a
572 * bad type... This isn't something that we should try
573 * to recover from. */
574 BUG_ON(IS_ERR(hbcall
));
576 o2hb_fire_callbacks(hbcall
, event
->hn_node
, event
->hn_node_num
);
578 spin_lock(&o2hb_live_lock
);
580 spin_unlock(&o2hb_live_lock
);
582 up_write(&o2hb_callback_sem
);
585 static void o2hb_queue_node_event(struct o2hb_node_event
*event
,
586 enum o2hb_callback_type type
,
587 struct o2nm_node
*node
,
590 assert_spin_locked(&o2hb_live_lock
);
592 event
->hn_event_type
= type
;
593 event
->hn_node
= node
;
594 event
->hn_node_num
= node_num
;
596 mlog(ML_HEARTBEAT
, "Queue node %s event for node %d\n",
597 type
== O2HB_NODE_UP_CB
? "UP" : "DOWN", node_num
);
599 list_add_tail(&event
->hn_item
, &o2hb_node_events
);
602 static void o2hb_shutdown_slot(struct o2hb_disk_slot
*slot
)
604 struct o2hb_node_event event
=
605 { .hn_item
= LIST_HEAD_INIT(event
.hn_item
), };
606 struct o2nm_node
*node
;
608 node
= o2nm_get_node_by_num(slot
->ds_node_num
);
612 spin_lock(&o2hb_live_lock
);
613 if (!list_empty(&slot
->ds_live_item
)) {
614 mlog(ML_HEARTBEAT
, "Shutdown, node %d leaves region\n",
617 list_del_init(&slot
->ds_live_item
);
619 if (list_empty(&o2hb_live_slots
[slot
->ds_node_num
])) {
620 clear_bit(slot
->ds_node_num
, o2hb_live_node_bitmap
);
622 o2hb_queue_node_event(&event
, O2HB_NODE_DOWN_CB
, node
,
626 spin_unlock(&o2hb_live_lock
);
628 o2hb_run_event_list(&event
);
633 static int o2hb_check_slot(struct o2hb_region
*reg
,
634 struct o2hb_disk_slot
*slot
)
636 int changed
= 0, gen_changed
= 0;
637 struct o2hb_node_event event
=
638 { .hn_item
= LIST_HEAD_INIT(event
.hn_item
), };
639 struct o2nm_node
*node
;
640 struct o2hb_disk_heartbeat_block
*hb_block
= reg
->hr_tmp_block
;
643 memcpy(hb_block
, slot
->ds_raw_block
, reg
->hr_block_bytes
);
645 /* Is this correct? Do we assume that the node doesn't exist
646 * if we're not configured for him? */
647 node
= o2nm_get_node_by_num(slot
->ds_node_num
);
651 if (!o2hb_verify_crc(reg
, hb_block
)) {
652 /* all paths from here will drop o2hb_live_lock for
654 spin_lock(&o2hb_live_lock
);
656 /* Don't print an error on the console in this case -
657 * a freshly formatted heartbeat area will not have a
659 if (list_empty(&slot
->ds_live_item
))
662 /* The node is live but pushed out a bad crc. We
663 * consider it a transient miss but don't populate any
664 * other values as they may be junk. */
665 mlog(ML_ERROR
, "Node %d has written a bad crc to %s\n",
666 slot
->ds_node_num
, reg
->hr_dev_name
);
667 o2hb_dump_slot(hb_block
);
669 slot
->ds_equal_samples
++;
673 /* we don't care if these wrap.. the state transitions below
674 * clear at the right places */
675 cputime
= le64_to_cpu(hb_block
->hb_seq
);
676 if (slot
->ds_last_time
!= cputime
)
677 slot
->ds_changed_samples
++;
679 slot
->ds_equal_samples
++;
680 slot
->ds_last_time
= cputime
;
682 /* The node changed heartbeat generations. We assume this to
683 * mean it dropped off but came back before we timed out. We
684 * want to consider it down for the time being but don't want
685 * to lose any changed_samples state we might build up to
686 * considering it live again. */
687 if (slot
->ds_last_generation
!= le64_to_cpu(hb_block
->hb_generation
)) {
689 slot
->ds_equal_samples
= 0;
690 mlog(ML_HEARTBEAT
, "Node %d changed generation (0x%llx "
691 "to 0x%llx)\n", slot
->ds_node_num
,
692 (long long)slot
->ds_last_generation
,
693 (long long)le64_to_cpu(hb_block
->hb_generation
));
696 slot
->ds_last_generation
= le64_to_cpu(hb_block
->hb_generation
);
698 mlog(ML_HEARTBEAT
, "Slot %d gen 0x%llx cksum 0x%x "
699 "seq %llu last %llu changed %u equal %u\n",
700 slot
->ds_node_num
, (long long)slot
->ds_last_generation
,
701 le32_to_cpu(hb_block
->hb_cksum
),
702 (unsigned long long)le64_to_cpu(hb_block
->hb_seq
),
703 (unsigned long long)slot
->ds_last_time
, slot
->ds_changed_samples
,
704 slot
->ds_equal_samples
);
706 spin_lock(&o2hb_live_lock
);
709 /* dead nodes only come to life after some number of
710 * changes at any time during their dead time */
711 if (list_empty(&slot
->ds_live_item
) &&
712 slot
->ds_changed_samples
>= O2HB_LIVE_THRESHOLD
) {
713 mlog(ML_HEARTBEAT
, "Node %d (id 0x%llx) joined my region\n",
714 slot
->ds_node_num
, (long long)slot
->ds_last_generation
);
716 /* first on the list generates a callback */
717 if (list_empty(&o2hb_live_slots
[slot
->ds_node_num
])) {
718 set_bit(slot
->ds_node_num
, o2hb_live_node_bitmap
);
720 o2hb_queue_node_event(&event
, O2HB_NODE_UP_CB
, node
,
726 list_add_tail(&slot
->ds_live_item
,
727 &o2hb_live_slots
[slot
->ds_node_num
]);
729 slot
->ds_equal_samples
= 0;
733 /* if the list is dead, we're done.. */
734 if (list_empty(&slot
->ds_live_item
))
737 /* live nodes only go dead after enough consequtive missed
738 * samples.. reset the missed counter whenever we see
740 if (slot
->ds_equal_samples
>= o2hb_dead_threshold
|| gen_changed
) {
741 mlog(ML_HEARTBEAT
, "Node %d left my region\n",
744 /* last off the live_slot generates a callback */
745 list_del_init(&slot
->ds_live_item
);
746 if (list_empty(&o2hb_live_slots
[slot
->ds_node_num
])) {
747 clear_bit(slot
->ds_node_num
, o2hb_live_node_bitmap
);
749 o2hb_queue_node_event(&event
, O2HB_NODE_DOWN_CB
, node
,
755 /* We don't clear this because the node is still
756 * actually writing new blocks. */
758 slot
->ds_changed_samples
= 0;
761 if (slot
->ds_changed_samples
) {
762 slot
->ds_changed_samples
= 0;
763 slot
->ds_equal_samples
= 0;
766 spin_unlock(&o2hb_live_lock
);
768 o2hb_run_event_list(&event
);
774 /* This could be faster if we just implmented a find_last_bit, but I
775 * don't think the circumstances warrant it. */
776 static int o2hb_highest_node(unsigned long *nodes
,
783 while ((node
= find_next_bit(nodes
, numbits
, node
+ 1)) != -1) {
793 static void o2hb_do_disk_heartbeat(struct o2hb_region
*reg
)
795 int i
, ret
, highest_node
, change
= 0;
796 unsigned long configured_nodes
[BITS_TO_LONGS(O2NM_MAX_NODES
)];
797 struct bio
*write_bio
;
798 struct o2hb_bio_wait_ctxt write_wc
;
800 if (o2nm_configured_node_map(configured_nodes
, sizeof(configured_nodes
)))
803 highest_node
= o2hb_highest_node(configured_nodes
, O2NM_MAX_NODES
);
804 if (highest_node
>= O2NM_MAX_NODES
) {
805 mlog(ML_NOTICE
, "ocfs2_heartbeat: no configured nodes found!\n");
809 /* No sense in reading the slots of nodes that don't exist
810 * yet. Of course, if the node definitions have holes in them
811 * then we're reading an empty slot anyway... Consider this
813 ret
= o2hb_read_slots(reg
, highest_node
+ 1);
819 /* With an up to date view of the slots, we can check that no
820 * other node has been improperly configured to heartbeat in
822 if (!o2hb_check_last_timestamp(reg
))
823 mlog(ML_ERROR
, "Device \"%s\": another node is heartbeating "
824 "in our slot!\n", reg
->hr_dev_name
);
826 /* fill in the proper info for our next heartbeat */
827 o2hb_prepare_block(reg
, reg
->hr_generation
);
829 /* And fire off the write. Note that we don't wait on this I/O
831 ret
= o2hb_issue_node_write(reg
, &write_bio
, &write_wc
);
838 while((i
= find_next_bit(configured_nodes
, O2NM_MAX_NODES
, i
+ 1)) < O2NM_MAX_NODES
) {
840 change
|= o2hb_check_slot(reg
, ®
->hr_slots
[i
]);
844 * We have to be sure we've advertised ourselves on disk
845 * before we can go to steady state. This ensures that
846 * people we find in our steady state have seen us.
848 o2hb_wait_on_io(reg
, &write_wc
);
850 o2hb_arm_write_timeout(reg
);
852 /* let the person who launched us know when things are steady */
853 if (!change
&& (atomic_read(®
->hr_steady_iterations
) != 0)) {
854 if (atomic_dec_and_test(®
->hr_steady_iterations
))
855 wake_up(&o2hb_steady_queue
);
859 /* Subtract b from a, storing the result in a. a *must* have a larger
861 static void o2hb_tv_subtract(struct timeval
*a
,
864 /* just return 0 when a is after b */
865 if (a
->tv_sec
< b
->tv_sec
||
866 (a
->tv_sec
== b
->tv_sec
&& a
->tv_usec
< b
->tv_usec
)) {
872 a
->tv_sec
-= b
->tv_sec
;
873 a
->tv_usec
-= b
->tv_usec
;
874 while ( a
->tv_usec
< 0 ) {
876 a
->tv_usec
+= 1000000;
880 static unsigned int o2hb_elapsed_msecs(struct timeval
*start
,
883 struct timeval res
= *end
;
885 o2hb_tv_subtract(&res
, start
);
887 return res
.tv_sec
* 1000 + res
.tv_usec
/ 1000;
891 * we ride the region ref that the region dir holds. before the region
892 * dir is removed and drops it ref it will wait to tear down this
895 static int o2hb_thread(void *data
)
898 struct o2hb_region
*reg
= data
;
899 struct bio
*write_bio
;
900 struct o2hb_bio_wait_ctxt write_wc
;
901 struct timeval before_hb
, after_hb
;
902 unsigned int elapsed_msec
;
904 mlog(ML_HEARTBEAT
|ML_KTHREAD
, "hb thread running\n");
906 set_user_nice(current
, -20);
908 while (!kthread_should_stop() && !reg
->hr_unclean_stop
) {
909 /* We track the time spent inside
910 * o2hb_do_disk_heartbeat so that we avoid more then
911 * hr_timeout_ms between disk writes. On busy systems
912 * this should result in a heartbeat which is less
913 * likely to time itself out. */
914 do_gettimeofday(&before_hb
);
916 o2hb_do_disk_heartbeat(reg
);
918 do_gettimeofday(&after_hb
);
919 elapsed_msec
= o2hb_elapsed_msecs(&before_hb
, &after_hb
);
921 mlog(0, "start = %lu.%lu, end = %lu.%lu, msec = %u\n",
922 before_hb
.tv_sec
, (unsigned long) before_hb
.tv_usec
,
923 after_hb
.tv_sec
, (unsigned long) after_hb
.tv_usec
,
926 if (elapsed_msec
< reg
->hr_timeout_ms
) {
927 /* the kthread api has blocked signals for us so no
928 * need to record the return value. */
929 msleep_interruptible(reg
->hr_timeout_ms
- elapsed_msec
);
933 o2hb_disarm_write_timeout(reg
);
935 /* unclean stop is only used in very bad situation */
936 for(i
= 0; !reg
->hr_unclean_stop
&& i
< reg
->hr_blocks
; i
++)
937 o2hb_shutdown_slot(®
->hr_slots
[i
]);
939 /* Explicit down notification - avoid forcing the other nodes
940 * to timeout on this region when we could just as easily
941 * write a clear generation - thus indicating to them that
942 * this node has left this region.
944 * XXX: Should we skip this on unclean_stop? */
945 o2hb_prepare_block(reg
, 0);
946 ret
= o2hb_issue_node_write(reg
, &write_bio
, &write_wc
);
948 o2hb_wait_on_io(reg
, &write_wc
);
954 mlog(ML_HEARTBEAT
|ML_KTHREAD
, "hb thread exiting\n");
963 for (i
= 0; i
< ARRAY_SIZE(o2hb_callbacks
); i
++)
964 INIT_LIST_HEAD(&o2hb_callbacks
[i
].list
);
966 for (i
= 0; i
< ARRAY_SIZE(o2hb_live_slots
); i
++)
967 INIT_LIST_HEAD(&o2hb_live_slots
[i
]);
969 INIT_LIST_HEAD(&o2hb_node_events
);
971 memset(o2hb_live_node_bitmap
, 0, sizeof(o2hb_live_node_bitmap
));
974 /* if we're already in a callback then we're already serialized by the sem */
975 static void o2hb_fill_node_map_from_callback(unsigned long *map
,
978 BUG_ON(bytes
< (BITS_TO_LONGS(O2NM_MAX_NODES
) * sizeof(unsigned long)));
980 memcpy(map
, &o2hb_live_node_bitmap
, bytes
);
984 * get a map of all nodes that are heartbeating in any regions
986 void o2hb_fill_node_map(unsigned long *map
, unsigned bytes
)
988 /* callers want to serialize this map and callbacks so that they
989 * can trust that they don't miss nodes coming to the party */
990 down_read(&o2hb_callback_sem
);
991 spin_lock(&o2hb_live_lock
);
992 o2hb_fill_node_map_from_callback(map
, bytes
);
993 spin_unlock(&o2hb_live_lock
);
994 up_read(&o2hb_callback_sem
);
996 EXPORT_SYMBOL_GPL(o2hb_fill_node_map
);
999 * heartbeat configfs bits. The heartbeat set is a default set under
1000 * the cluster set in nodemanager.c.
1003 static struct o2hb_region
*to_o2hb_region(struct config_item
*item
)
1005 return item
? container_of(item
, struct o2hb_region
, hr_item
) : NULL
;
1008 /* drop_item only drops its ref after killing the thread, nothing should
1009 * be using the region anymore. this has to clean up any state that
1010 * attributes might have built up. */
1011 static void o2hb_region_release(struct config_item
*item
)
1015 struct o2hb_region
*reg
= to_o2hb_region(item
);
1017 if (reg
->hr_tmp_block
)
1018 kfree(reg
->hr_tmp_block
);
1020 if (reg
->hr_slot_data
) {
1021 for (i
= 0; i
< reg
->hr_num_pages
; i
++) {
1022 page
= reg
->hr_slot_data
[i
];
1026 kfree(reg
->hr_slot_data
);
1030 blkdev_put(reg
->hr_bdev
);
1033 kfree(reg
->hr_slots
);
1035 spin_lock(&o2hb_live_lock
);
1036 list_del(®
->hr_all_item
);
1037 spin_unlock(&o2hb_live_lock
);
1042 static int o2hb_read_block_input(struct o2hb_region
*reg
,
1045 unsigned long *ret_bytes
,
1046 unsigned int *ret_bits
)
1048 unsigned long bytes
;
1049 char *p
= (char *)page
;
1051 bytes
= simple_strtoul(p
, &p
, 0);
1052 if (!p
|| (*p
&& (*p
!= '\n')))
1055 /* Heartbeat and fs min / max block sizes are the same. */
1056 if (bytes
> 4096 || bytes
< 512)
1058 if (hweight16(bytes
) != 1)
1064 *ret_bits
= ffs(bytes
) - 1;
1069 static ssize_t
o2hb_region_block_bytes_read(struct o2hb_region
*reg
,
1072 return sprintf(page
, "%u\n", reg
->hr_block_bytes
);
1075 static ssize_t
o2hb_region_block_bytes_write(struct o2hb_region
*reg
,
1080 unsigned long block_bytes
;
1081 unsigned int block_bits
;
1086 status
= o2hb_read_block_input(reg
, page
, count
,
1087 &block_bytes
, &block_bits
);
1091 reg
->hr_block_bytes
= (unsigned int)block_bytes
;
1092 reg
->hr_block_bits
= block_bits
;
1097 static ssize_t
o2hb_region_start_block_read(struct o2hb_region
*reg
,
1100 return sprintf(page
, "%llu\n", reg
->hr_start_block
);
1103 static ssize_t
o2hb_region_start_block_write(struct o2hb_region
*reg
,
1107 unsigned long long tmp
;
1108 char *p
= (char *)page
;
1113 tmp
= simple_strtoull(p
, &p
, 0);
1114 if (!p
|| (*p
&& (*p
!= '\n')))
1117 reg
->hr_start_block
= tmp
;
1122 static ssize_t
o2hb_region_blocks_read(struct o2hb_region
*reg
,
1125 return sprintf(page
, "%d\n", reg
->hr_blocks
);
1128 static ssize_t
o2hb_region_blocks_write(struct o2hb_region
*reg
,
1133 char *p
= (char *)page
;
1138 tmp
= simple_strtoul(p
, &p
, 0);
1139 if (!p
|| (*p
&& (*p
!= '\n')))
1142 if (tmp
> O2NM_MAX_NODES
|| tmp
== 0)
1145 reg
->hr_blocks
= (unsigned int)tmp
;
1150 static ssize_t
o2hb_region_dev_read(struct o2hb_region
*reg
,
1153 unsigned int ret
= 0;
1156 ret
= sprintf(page
, "%s\n", reg
->hr_dev_name
);
1161 static void o2hb_init_region_params(struct o2hb_region
*reg
)
1163 reg
->hr_slots_per_page
= PAGE_CACHE_SIZE
>> reg
->hr_block_bits
;
1164 reg
->hr_timeout_ms
= O2HB_REGION_TIMEOUT_MS
;
1166 mlog(ML_HEARTBEAT
, "hr_start_block = %llu, hr_blocks = %u\n",
1167 reg
->hr_start_block
, reg
->hr_blocks
);
1168 mlog(ML_HEARTBEAT
, "hr_block_bytes = %u, hr_block_bits = %u\n",
1169 reg
->hr_block_bytes
, reg
->hr_block_bits
);
1170 mlog(ML_HEARTBEAT
, "hr_timeout_ms = %u\n", reg
->hr_timeout_ms
);
1171 mlog(ML_HEARTBEAT
, "dead threshold = %u\n", o2hb_dead_threshold
);
1174 static int o2hb_map_slot_data(struct o2hb_region
*reg
)
1177 unsigned int last_slot
;
1178 unsigned int spp
= reg
->hr_slots_per_page
;
1181 struct o2hb_disk_slot
*slot
;
1183 reg
->hr_tmp_block
= kmalloc(reg
->hr_block_bytes
, GFP_KERNEL
);
1184 if (reg
->hr_tmp_block
== NULL
) {
1185 mlog_errno(-ENOMEM
);
1189 reg
->hr_slots
= kcalloc(reg
->hr_blocks
,
1190 sizeof(struct o2hb_disk_slot
), GFP_KERNEL
);
1191 if (reg
->hr_slots
== NULL
) {
1192 mlog_errno(-ENOMEM
);
1196 for(i
= 0; i
< reg
->hr_blocks
; i
++) {
1197 slot
= ®
->hr_slots
[i
];
1198 slot
->ds_node_num
= i
;
1199 INIT_LIST_HEAD(&slot
->ds_live_item
);
1200 slot
->ds_raw_block
= NULL
;
1203 reg
->hr_num_pages
= (reg
->hr_blocks
+ spp
- 1) / spp
;
1204 mlog(ML_HEARTBEAT
, "Going to require %u pages to cover %u blocks "
1205 "at %u blocks per page\n",
1206 reg
->hr_num_pages
, reg
->hr_blocks
, spp
);
1208 reg
->hr_slot_data
= kcalloc(reg
->hr_num_pages
, sizeof(struct page
*),
1210 if (!reg
->hr_slot_data
) {
1211 mlog_errno(-ENOMEM
);
1215 for(i
= 0; i
< reg
->hr_num_pages
; i
++) {
1216 page
= alloc_page(GFP_KERNEL
);
1218 mlog_errno(-ENOMEM
);
1222 reg
->hr_slot_data
[i
] = page
;
1224 last_slot
= i
* spp
;
1225 raw
= page_address(page
);
1227 (j
< spp
) && ((j
+ last_slot
) < reg
->hr_blocks
);
1229 BUG_ON((j
+ last_slot
) >= reg
->hr_blocks
);
1231 slot
= ®
->hr_slots
[j
+ last_slot
];
1232 slot
->ds_raw_block
=
1233 (struct o2hb_disk_heartbeat_block
*) raw
;
1235 raw
+= reg
->hr_block_bytes
;
1242 /* Read in all the slots available and populate the tracking
1243 * structures so that we can start with a baseline idea of what's
1245 static int o2hb_populate_slot_data(struct o2hb_region
*reg
)
1248 struct o2hb_disk_slot
*slot
;
1249 struct o2hb_disk_heartbeat_block
*hb_block
;
1253 ret
= o2hb_read_slots(reg
, reg
->hr_blocks
);
1259 /* We only want to get an idea of the values initially in each
1260 * slot, so we do no verification - o2hb_check_slot will
1261 * actually determine if each configured slot is valid and
1262 * whether any values have changed. */
1263 for(i
= 0; i
< reg
->hr_blocks
; i
++) {
1264 slot
= ®
->hr_slots
[i
];
1265 hb_block
= (struct o2hb_disk_heartbeat_block
*) slot
->ds_raw_block
;
1267 /* Only fill the values that o2hb_check_slot uses to
1268 * determine changing slots */
1269 slot
->ds_last_time
= le64_to_cpu(hb_block
->hb_seq
);
1270 slot
->ds_last_generation
= le64_to_cpu(hb_block
->hb_generation
);
1278 /* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */
1279 static ssize_t
o2hb_region_dev_write(struct o2hb_region
*reg
,
1285 char *p
= (char *)page
;
1286 struct file
*filp
= NULL
;
1287 struct inode
*inode
= NULL
;
1288 ssize_t ret
= -EINVAL
;
1293 /* We can't heartbeat without having had our node number
1294 * configured yet. */
1295 if (o2nm_this_node() == O2NM_MAX_NODES
)
1298 fd
= simple_strtol(p
, &p
, 0);
1299 if (!p
|| (*p
&& (*p
!= '\n')))
1302 if (fd
< 0 || fd
>= INT_MAX
)
1309 if (reg
->hr_blocks
== 0 || reg
->hr_start_block
== 0 ||
1310 reg
->hr_block_bytes
== 0)
1313 inode
= igrab(filp
->f_mapping
->host
);
1317 if (!S_ISBLK(inode
->i_mode
))
1320 reg
->hr_bdev
= I_BDEV(filp
->f_mapping
->host
);
1321 ret
= blkdev_get(reg
->hr_bdev
, FMODE_WRITE
| FMODE_READ
, 0);
1323 reg
->hr_bdev
= NULL
;
1328 bdevname(reg
->hr_bdev
, reg
->hr_dev_name
);
1330 sectsize
= bdev_hardsect_size(reg
->hr_bdev
);
1331 if (sectsize
!= reg
->hr_block_bytes
) {
1333 "blocksize %u incorrect for device, expected %d",
1334 reg
->hr_block_bytes
, sectsize
);
1339 o2hb_init_region_params(reg
);
1341 /* Generation of zero is invalid */
1343 get_random_bytes(®
->hr_generation
,
1344 sizeof(reg
->hr_generation
));
1345 } while (reg
->hr_generation
== 0);
1347 ret
= o2hb_map_slot_data(reg
);
1353 ret
= o2hb_populate_slot_data(reg
);
1359 INIT_WORK(®
->hr_write_timeout_work
, o2hb_write_timeout
, reg
);
1362 * A node is considered live after it has beat LIVE_THRESHOLD
1363 * times. We're not steady until we've given them a chance
1364 * _after_ our first read.
1366 atomic_set(®
->hr_steady_iterations
, O2HB_LIVE_THRESHOLD
+ 1);
1368 reg
->hr_task
= kthread_run(o2hb_thread
, reg
, "o2hb-%s",
1369 reg
->hr_item
.ci_name
);
1370 if (IS_ERR(reg
->hr_task
)) {
1371 ret
= PTR_ERR(reg
->hr_task
);
1373 reg
->hr_task
= NULL
;
1377 ret
= wait_event_interruptible(o2hb_steady_queue
,
1378 atomic_read(®
->hr_steady_iterations
) == 0);
1380 kthread_stop(reg
->hr_task
);
1381 reg
->hr_task
= NULL
;
1393 blkdev_put(reg
->hr_bdev
);
1394 reg
->hr_bdev
= NULL
;
1400 struct o2hb_region_attribute
{
1401 struct configfs_attribute attr
;
1402 ssize_t (*show
)(struct o2hb_region
*, char *);
1403 ssize_t (*store
)(struct o2hb_region
*, const char *, size_t);
1406 static struct o2hb_region_attribute o2hb_region_attr_block_bytes
= {
1407 .attr
= { .ca_owner
= THIS_MODULE
,
1408 .ca_name
= "block_bytes",
1409 .ca_mode
= S_IRUGO
| S_IWUSR
},
1410 .show
= o2hb_region_block_bytes_read
,
1411 .store
= o2hb_region_block_bytes_write
,
1414 static struct o2hb_region_attribute o2hb_region_attr_start_block
= {
1415 .attr
= { .ca_owner
= THIS_MODULE
,
1416 .ca_name
= "start_block",
1417 .ca_mode
= S_IRUGO
| S_IWUSR
},
1418 .show
= o2hb_region_start_block_read
,
1419 .store
= o2hb_region_start_block_write
,
1422 static struct o2hb_region_attribute o2hb_region_attr_blocks
= {
1423 .attr
= { .ca_owner
= THIS_MODULE
,
1424 .ca_name
= "blocks",
1425 .ca_mode
= S_IRUGO
| S_IWUSR
},
1426 .show
= o2hb_region_blocks_read
,
1427 .store
= o2hb_region_blocks_write
,
1430 static struct o2hb_region_attribute o2hb_region_attr_dev
= {
1431 .attr
= { .ca_owner
= THIS_MODULE
,
1433 .ca_mode
= S_IRUGO
| S_IWUSR
},
1434 .show
= o2hb_region_dev_read
,
1435 .store
= o2hb_region_dev_write
,
1438 static struct configfs_attribute
*o2hb_region_attrs
[] = {
1439 &o2hb_region_attr_block_bytes
.attr
,
1440 &o2hb_region_attr_start_block
.attr
,
1441 &o2hb_region_attr_blocks
.attr
,
1442 &o2hb_region_attr_dev
.attr
,
1446 static ssize_t
o2hb_region_show(struct config_item
*item
,
1447 struct configfs_attribute
*attr
,
1450 struct o2hb_region
*reg
= to_o2hb_region(item
);
1451 struct o2hb_region_attribute
*o2hb_region_attr
=
1452 container_of(attr
, struct o2hb_region_attribute
, attr
);
1455 if (o2hb_region_attr
->show
)
1456 ret
= o2hb_region_attr
->show(reg
, page
);
1460 static ssize_t
o2hb_region_store(struct config_item
*item
,
1461 struct configfs_attribute
*attr
,
1462 const char *page
, size_t count
)
1464 struct o2hb_region
*reg
= to_o2hb_region(item
);
1465 struct o2hb_region_attribute
*o2hb_region_attr
=
1466 container_of(attr
, struct o2hb_region_attribute
, attr
);
1467 ssize_t ret
= -EINVAL
;
1469 if (o2hb_region_attr
->store
)
1470 ret
= o2hb_region_attr
->store(reg
, page
, count
);
1474 static struct configfs_item_operations o2hb_region_item_ops
= {
1475 .release
= o2hb_region_release
,
1476 .show_attribute
= o2hb_region_show
,
1477 .store_attribute
= o2hb_region_store
,
1480 static struct config_item_type o2hb_region_type
= {
1481 .ct_item_ops
= &o2hb_region_item_ops
,
1482 .ct_attrs
= o2hb_region_attrs
,
1483 .ct_owner
= THIS_MODULE
,
1488 struct o2hb_heartbeat_group
{
1489 struct config_group hs_group
;
1493 static struct o2hb_heartbeat_group
*to_o2hb_heartbeat_group(struct config_group
*group
)
1496 container_of(group
, struct o2hb_heartbeat_group
, hs_group
)
1500 static struct config_item
*o2hb_heartbeat_group_make_item(struct config_group
*group
,
1503 struct o2hb_region
*reg
= NULL
;
1504 struct config_item
*ret
= NULL
;
1506 reg
= kcalloc(1, sizeof(struct o2hb_region
), GFP_KERNEL
);
1508 goto out
; /* ENOMEM */
1510 config_item_init_type_name(®
->hr_item
, name
, &o2hb_region_type
);
1512 ret
= ®
->hr_item
;
1514 spin_lock(&o2hb_live_lock
);
1515 list_add_tail(®
->hr_all_item
, &o2hb_all_regions
);
1516 spin_unlock(&o2hb_live_lock
);
1524 static void o2hb_heartbeat_group_drop_item(struct config_group
*group
,
1525 struct config_item
*item
)
1527 struct o2hb_region
*reg
= to_o2hb_region(item
);
1529 /* stop the thread when the user removes the region dir */
1531 kthread_stop(reg
->hr_task
);
1532 reg
->hr_task
= NULL
;
1535 config_item_put(item
);
1538 struct o2hb_heartbeat_group_attribute
{
1539 struct configfs_attribute attr
;
1540 ssize_t (*show
)(struct o2hb_heartbeat_group
*, char *);
1541 ssize_t (*store
)(struct o2hb_heartbeat_group
*, const char *, size_t);
1544 static ssize_t
o2hb_heartbeat_group_show(struct config_item
*item
,
1545 struct configfs_attribute
*attr
,
1548 struct o2hb_heartbeat_group
*reg
= to_o2hb_heartbeat_group(to_config_group(item
));
1549 struct o2hb_heartbeat_group_attribute
*o2hb_heartbeat_group_attr
=
1550 container_of(attr
, struct o2hb_heartbeat_group_attribute
, attr
);
1553 if (o2hb_heartbeat_group_attr
->show
)
1554 ret
= o2hb_heartbeat_group_attr
->show(reg
, page
);
1558 static ssize_t
o2hb_heartbeat_group_store(struct config_item
*item
,
1559 struct configfs_attribute
*attr
,
1560 const char *page
, size_t count
)
1562 struct o2hb_heartbeat_group
*reg
= to_o2hb_heartbeat_group(to_config_group(item
));
1563 struct o2hb_heartbeat_group_attribute
*o2hb_heartbeat_group_attr
=
1564 container_of(attr
, struct o2hb_heartbeat_group_attribute
, attr
);
1565 ssize_t ret
= -EINVAL
;
1567 if (o2hb_heartbeat_group_attr
->store
)
1568 ret
= o2hb_heartbeat_group_attr
->store(reg
, page
, count
);
1572 static ssize_t
o2hb_heartbeat_group_threshold_show(struct o2hb_heartbeat_group
*group
,
1575 return sprintf(page
, "%u\n", o2hb_dead_threshold
);
1578 static ssize_t
o2hb_heartbeat_group_threshold_store(struct o2hb_heartbeat_group
*group
,
1583 char *p
= (char *)page
;
1585 tmp
= simple_strtoul(p
, &p
, 10);
1586 if (!p
|| (*p
&& (*p
!= '\n')))
1589 /* this will validate ranges for us. */
1590 o2hb_dead_threshold_set((unsigned int) tmp
);
1595 static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_threshold
= {
1596 .attr
= { .ca_owner
= THIS_MODULE
,
1597 .ca_name
= "dead_threshold",
1598 .ca_mode
= S_IRUGO
| S_IWUSR
},
1599 .show
= o2hb_heartbeat_group_threshold_show
,
1600 .store
= o2hb_heartbeat_group_threshold_store
,
1603 static struct configfs_attribute
*o2hb_heartbeat_group_attrs
[] = {
1604 &o2hb_heartbeat_group_attr_threshold
.attr
,
1608 static struct configfs_item_operations o2hb_hearbeat_group_item_ops
= {
1609 .show_attribute
= o2hb_heartbeat_group_show
,
1610 .store_attribute
= o2hb_heartbeat_group_store
,
1613 static struct configfs_group_operations o2hb_heartbeat_group_group_ops
= {
1614 .make_item
= o2hb_heartbeat_group_make_item
,
1615 .drop_item
= o2hb_heartbeat_group_drop_item
,
1618 static struct config_item_type o2hb_heartbeat_group_type
= {
1619 .ct_group_ops
= &o2hb_heartbeat_group_group_ops
,
1620 .ct_item_ops
= &o2hb_hearbeat_group_item_ops
,
1621 .ct_attrs
= o2hb_heartbeat_group_attrs
,
1622 .ct_owner
= THIS_MODULE
,
1625 /* this is just here to avoid touching group in heartbeat.h which the
1626 * entire damn world #includes */
1627 struct config_group
*o2hb_alloc_hb_set(void)
1629 struct o2hb_heartbeat_group
*hs
= NULL
;
1630 struct config_group
*ret
= NULL
;
1632 hs
= kcalloc(1, sizeof(struct o2hb_heartbeat_group
), GFP_KERNEL
);
1636 config_group_init_type_name(&hs
->hs_group
, "heartbeat",
1637 &o2hb_heartbeat_group_type
);
1639 ret
= &hs
->hs_group
;
1646 void o2hb_free_hb_set(struct config_group
*group
)
1648 struct o2hb_heartbeat_group
*hs
= to_o2hb_heartbeat_group(group
);
1652 /* hb callback registration and issueing */
1654 static struct o2hb_callback
*hbcall_from_type(enum o2hb_callback_type type
)
1656 if (type
== O2HB_NUM_CB
)
1657 return ERR_PTR(-EINVAL
);
1659 return &o2hb_callbacks
[type
];
1662 void o2hb_setup_callback(struct o2hb_callback_func
*hc
,
1663 enum o2hb_callback_type type
,
1668 INIT_LIST_HEAD(&hc
->hc_item
);
1671 hc
->hc_priority
= priority
;
1673 hc
->hc_magic
= O2HB_CB_MAGIC
;
1675 EXPORT_SYMBOL_GPL(o2hb_setup_callback
);
1677 int o2hb_register_callback(struct o2hb_callback_func
*hc
)
1679 struct o2hb_callback_func
*tmp
;
1680 struct list_head
*iter
;
1681 struct o2hb_callback
*hbcall
;
1684 BUG_ON(hc
->hc_magic
!= O2HB_CB_MAGIC
);
1685 BUG_ON(!list_empty(&hc
->hc_item
));
1687 hbcall
= hbcall_from_type(hc
->hc_type
);
1688 if (IS_ERR(hbcall
)) {
1689 ret
= PTR_ERR(hbcall
);
1693 down_write(&o2hb_callback_sem
);
1695 list_for_each(iter
, &hbcall
->list
) {
1696 tmp
= list_entry(iter
, struct o2hb_callback_func
, hc_item
);
1697 if (hc
->hc_priority
< tmp
->hc_priority
) {
1698 list_add_tail(&hc
->hc_item
, iter
);
1702 if (list_empty(&hc
->hc_item
))
1703 list_add_tail(&hc
->hc_item
, &hbcall
->list
);
1705 up_write(&o2hb_callback_sem
);
1708 mlog(ML_HEARTBEAT
, "returning %d on behalf of %p for funcs %p\n",
1709 ret
, __builtin_return_address(0), hc
);
1712 EXPORT_SYMBOL_GPL(o2hb_register_callback
);
1714 int o2hb_unregister_callback(struct o2hb_callback_func
*hc
)
1716 BUG_ON(hc
->hc_magic
!= O2HB_CB_MAGIC
);
1718 mlog(ML_HEARTBEAT
, "on behalf of %p for funcs %p\n",
1719 __builtin_return_address(0), hc
);
1721 if (list_empty(&hc
->hc_item
))
1724 down_write(&o2hb_callback_sem
);
1726 list_del_init(&hc
->hc_item
);
1728 up_write(&o2hb_callback_sem
);
1732 EXPORT_SYMBOL_GPL(o2hb_unregister_callback
);
1734 int o2hb_check_node_heartbeating(u8 node_num
)
1736 unsigned long testing_map
[BITS_TO_LONGS(O2NM_MAX_NODES
)];
1738 o2hb_fill_node_map(testing_map
, sizeof(testing_map
));
1739 if (!test_bit(node_num
, testing_map
)) {
1741 "node (%u) does not have heartbeating enabled.\n",
1748 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating
);
1750 int o2hb_check_node_heartbeating_from_callback(u8 node_num
)
1752 unsigned long testing_map
[BITS_TO_LONGS(O2NM_MAX_NODES
)];
1754 o2hb_fill_node_map_from_callback(testing_map
, sizeof(testing_map
));
1755 if (!test_bit(node_num
, testing_map
)) {
1757 "node (%u) does not have heartbeating enabled.\n",
1764 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback
);
1766 /* Makes sure our local node is configured with a node number, and is
1768 int o2hb_check_local_node_heartbeating(void)
1772 /* if this node was set then we have networking */
1773 node_num
= o2nm_this_node();
1774 if (node_num
== O2NM_MAX_NODES
) {
1775 mlog(ML_HEARTBEAT
, "this node has not been configured.\n");
1779 return o2hb_check_node_heartbeating(node_num
);
1781 EXPORT_SYMBOL_GPL(o2hb_check_local_node_heartbeating
);
1784 * this is just a hack until we get the plumbing which flips file systems
1785 * read only and drops the hb ref instead of killing the node dead.
1787 void o2hb_stop_all_regions(void)
1789 struct o2hb_region
*reg
;
1791 mlog(ML_ERROR
, "stopping heartbeat on all active regions.\n");
1793 spin_lock(&o2hb_live_lock
);
1795 list_for_each_entry(reg
, &o2hb_all_regions
, hr_all_item
)
1796 reg
->hr_unclean_stop
= 1;
1798 spin_unlock(&o2hb_live_lock
);
1800 EXPORT_SYMBOL_GPL(o2hb_stop_all_regions
);