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>
39 #include "heartbeat.h"
41 #include "nodemanager.h"
48 * The first heartbeat pass had one global thread that would serialize all hb
49 * callback calls. This global serializing sem should only be removed once
50 * we've made sure that all callees can deal with being called concurrently
51 * from multiple hb region threads.
53 static DECLARE_RWSEM(o2hb_callback_sem
);
56 * multiple hb threads are watching multiple regions. A node is live
57 * whenever any of the threads sees activity from the node in its region.
59 static DEFINE_SPINLOCK(o2hb_live_lock
);
60 static struct list_head o2hb_live_slots
[O2NM_MAX_NODES
];
61 static unsigned long o2hb_live_node_bitmap
[BITS_TO_LONGS(O2NM_MAX_NODES
)];
62 static LIST_HEAD(o2hb_node_events
);
63 static DECLARE_WAIT_QUEUE_HEAD(o2hb_steady_queue
);
66 * In global heartbeat, we maintain a series of region bitmaps.
67 * - o2hb_region_bitmap allows us to limit the region number to max region.
68 * - o2hb_live_region_bitmap tracks live regions (seen steady iterations).
69 * - o2hb_quorum_region_bitmap tracks live regions that have seen all nodes
71 * - o2hb_failed_region_bitmap tracks the regions that have seen io timeouts.
73 static unsigned long o2hb_region_bitmap
[BITS_TO_LONGS(O2NM_MAX_REGIONS
)];
74 static unsigned long o2hb_live_region_bitmap
[BITS_TO_LONGS(O2NM_MAX_REGIONS
)];
75 static unsigned long o2hb_quorum_region_bitmap
[BITS_TO_LONGS(O2NM_MAX_REGIONS
)];
76 static unsigned long o2hb_failed_region_bitmap
[BITS_TO_LONGS(O2NM_MAX_REGIONS
)];
78 #define O2HB_DB_TYPE_LIVENODES 0
79 #define O2HB_DB_TYPE_LIVEREGIONS 1
80 #define O2HB_DB_TYPE_QUORUMREGIONS 2
81 #define O2HB_DB_TYPE_FAILEDREGIONS 3
82 #define O2HB_DB_TYPE_REGION_LIVENODES 4
83 #define O2HB_DB_TYPE_REGION_NUMBER 5
84 #define O2HB_DB_TYPE_REGION_ELAPSED_TIME 6
85 #define O2HB_DB_TYPE_REGION_PINNED 7
86 struct o2hb_debug_buf
{
93 static struct o2hb_debug_buf
*o2hb_db_livenodes
;
94 static struct o2hb_debug_buf
*o2hb_db_liveregions
;
95 static struct o2hb_debug_buf
*o2hb_db_quorumregions
;
96 static struct o2hb_debug_buf
*o2hb_db_failedregions
;
98 #define O2HB_DEBUG_DIR "o2hb"
99 #define O2HB_DEBUG_LIVENODES "livenodes"
100 #define O2HB_DEBUG_LIVEREGIONS "live_regions"
101 #define O2HB_DEBUG_QUORUMREGIONS "quorum_regions"
102 #define O2HB_DEBUG_FAILEDREGIONS "failed_regions"
103 #define O2HB_DEBUG_REGION_NUMBER "num"
104 #define O2HB_DEBUG_REGION_ELAPSED_TIME "elapsed_time_in_ms"
105 #define O2HB_DEBUG_REGION_PINNED "pinned"
107 static struct dentry
*o2hb_debug_dir
;
108 static struct dentry
*o2hb_debug_livenodes
;
109 static struct dentry
*o2hb_debug_liveregions
;
110 static struct dentry
*o2hb_debug_quorumregions
;
111 static struct dentry
*o2hb_debug_failedregions
;
113 static LIST_HEAD(o2hb_all_regions
);
115 static struct o2hb_callback
{
116 struct list_head list
;
117 } o2hb_callbacks
[O2HB_NUM_CB
];
119 static struct o2hb_callback
*hbcall_from_type(enum o2hb_callback_type type
);
121 #define O2HB_DEFAULT_BLOCK_BITS 9
123 enum o2hb_heartbeat_modes
{
124 O2HB_HEARTBEAT_LOCAL
= 0,
125 O2HB_HEARTBEAT_GLOBAL
,
126 O2HB_HEARTBEAT_NUM_MODES
,
129 char *o2hb_heartbeat_mode_desc
[O2HB_HEARTBEAT_NUM_MODES
] = {
130 "local", /* O2HB_HEARTBEAT_LOCAL */
131 "global", /* O2HB_HEARTBEAT_GLOBAL */
134 unsigned int o2hb_dead_threshold
= O2HB_DEFAULT_DEAD_THRESHOLD
;
135 unsigned int o2hb_heartbeat_mode
= O2HB_HEARTBEAT_LOCAL
;
138 * o2hb_dependent_users tracks the number of registered callbacks that depend
139 * on heartbeat. o2net and o2dlm are two entities that register this callback.
140 * However only o2dlm depends on the heartbeat. It does not want the heartbeat
141 * to stop while a dlm domain is still active.
143 unsigned int o2hb_dependent_users
;
146 * In global heartbeat mode, all regions are pinned if there are one or more
147 * dependent users and the quorum region count is <= O2HB_PIN_CUT_OFF. All
148 * regions are unpinned if the region count exceeds the cut off or the number
149 * of dependent users falls to zero.
151 #define O2HB_PIN_CUT_OFF 3
154 * In local heartbeat mode, we assume the dlm domain name to be the same as
155 * region uuid. This is true for domains created for the file system but not
156 * necessarily true for userdlm domains. This is a known limitation.
158 * In global heartbeat mode, we pin/unpin all o2hb regions. This solution
159 * works for both file system and userdlm domains.
161 static int o2hb_region_pin(const char *region_uuid
);
162 static void o2hb_region_unpin(const char *region_uuid
);
164 /* Only sets a new threshold if there are no active regions.
166 * No locking or otherwise interesting code is required for reading
167 * o2hb_dead_threshold as it can't change once regions are active and
168 * it's not interesting to anyone until then anyway. */
169 static void o2hb_dead_threshold_set(unsigned int threshold
)
171 if (threshold
> O2HB_MIN_DEAD_THRESHOLD
) {
172 spin_lock(&o2hb_live_lock
);
173 if (list_empty(&o2hb_all_regions
))
174 o2hb_dead_threshold
= threshold
;
175 spin_unlock(&o2hb_live_lock
);
179 static int o2hb_global_hearbeat_mode_set(unsigned int hb_mode
)
183 if (hb_mode
< O2HB_HEARTBEAT_NUM_MODES
) {
184 spin_lock(&o2hb_live_lock
);
185 if (list_empty(&o2hb_all_regions
)) {
186 o2hb_heartbeat_mode
= hb_mode
;
189 spin_unlock(&o2hb_live_lock
);
195 struct o2hb_node_event
{
196 struct list_head hn_item
;
197 enum o2hb_callback_type hn_event_type
;
198 struct o2nm_node
*hn_node
;
202 struct o2hb_disk_slot
{
203 struct o2hb_disk_heartbeat_block
*ds_raw_block
;
206 u64 ds_last_generation
;
207 u16 ds_equal_samples
;
208 u16 ds_changed_samples
;
209 struct list_head ds_live_item
;
212 /* each thread owns a region.. when we're asked to tear down the region
213 * we ask the thread to stop, who cleans up the region */
215 struct config_item hr_item
;
217 struct list_head hr_all_item
;
218 unsigned hr_unclean_stop
:1,
222 /* protected by the hr_callback_sem */
223 struct task_struct
*hr_task
;
225 unsigned int hr_blocks
;
226 unsigned long long hr_start_block
;
228 unsigned int hr_block_bits
;
229 unsigned int hr_block_bytes
;
231 unsigned int hr_slots_per_page
;
232 unsigned int hr_num_pages
;
234 struct page
**hr_slot_data
;
235 struct block_device
*hr_bdev
;
236 struct o2hb_disk_slot
*hr_slots
;
238 /* live node map of this region */
239 unsigned long hr_live_node_bitmap
[BITS_TO_LONGS(O2NM_MAX_NODES
)];
240 unsigned int hr_region_num
;
242 struct dentry
*hr_debug_dir
;
243 struct dentry
*hr_debug_livenodes
;
244 struct dentry
*hr_debug_regnum
;
245 struct dentry
*hr_debug_elapsed_time
;
246 struct dentry
*hr_debug_pinned
;
247 struct o2hb_debug_buf
*hr_db_livenodes
;
248 struct o2hb_debug_buf
*hr_db_regnum
;
249 struct o2hb_debug_buf
*hr_db_elapsed_time
;
250 struct o2hb_debug_buf
*hr_db_pinned
;
252 /* let the person setting up hb wait for it to return until it
253 * has reached a 'steady' state. This will be fixed when we have
254 * a more complete api that doesn't lead to this sort of fragility. */
255 atomic_t hr_steady_iterations
;
257 char hr_dev_name
[BDEVNAME_SIZE
];
259 unsigned int hr_timeout_ms
;
261 /* randomized as the region goes up and down so that a node
262 * recognizes a node going up and down in one iteration */
265 struct delayed_work hr_write_timeout_work
;
266 unsigned long hr_last_timeout_start
;
268 /* Used during o2hb_check_slot to hold a copy of the block
269 * being checked because we temporarily have to zero out the
271 struct o2hb_disk_heartbeat_block
*hr_tmp_block
;
274 struct o2hb_bio_wait_ctxt
{
275 atomic_t wc_num_reqs
;
276 struct completion wc_io_complete
;
280 static int o2hb_pop_count(void *map
, int count
)
284 while ((i
= find_next_bit(map
, count
, i
+ 1)) < count
)
289 static void o2hb_write_timeout(struct work_struct
*work
)
293 struct o2hb_region
*reg
=
294 container_of(work
, struct o2hb_region
,
295 hr_write_timeout_work
.work
);
297 mlog(ML_ERROR
, "Heartbeat write timeout to device %s after %u "
298 "milliseconds\n", reg
->hr_dev_name
,
299 jiffies_to_msecs(jiffies
- reg
->hr_last_timeout_start
));
301 if (o2hb_global_heartbeat_active()) {
302 spin_lock_irqsave(&o2hb_live_lock
, flags
);
303 if (test_bit(reg
->hr_region_num
, o2hb_quorum_region_bitmap
))
304 set_bit(reg
->hr_region_num
, o2hb_failed_region_bitmap
);
305 failed
= o2hb_pop_count(&o2hb_failed_region_bitmap
,
307 quorum
= o2hb_pop_count(&o2hb_quorum_region_bitmap
,
309 spin_unlock_irqrestore(&o2hb_live_lock
, flags
);
311 mlog(ML_HEARTBEAT
, "Number of regions %d, failed regions %d\n",
315 * Fence if the number of failed regions >= half the number
318 if ((failed
<< 1) < quorum
)
322 o2quo_disk_timeout();
325 static void o2hb_arm_write_timeout(struct o2hb_region
*reg
)
327 mlog(ML_HEARTBEAT
, "Queue write timeout for %u ms\n",
328 O2HB_MAX_WRITE_TIMEOUT_MS
);
330 if (o2hb_global_heartbeat_active()) {
331 spin_lock(&o2hb_live_lock
);
332 clear_bit(reg
->hr_region_num
, o2hb_failed_region_bitmap
);
333 spin_unlock(&o2hb_live_lock
);
335 cancel_delayed_work(®
->hr_write_timeout_work
);
336 reg
->hr_last_timeout_start
= jiffies
;
337 schedule_delayed_work(®
->hr_write_timeout_work
,
338 msecs_to_jiffies(O2HB_MAX_WRITE_TIMEOUT_MS
));
341 static void o2hb_disarm_write_timeout(struct o2hb_region
*reg
)
343 cancel_delayed_work_sync(®
->hr_write_timeout_work
);
346 static inline void o2hb_bio_wait_init(struct o2hb_bio_wait_ctxt
*wc
)
348 atomic_set(&wc
->wc_num_reqs
, 1);
349 init_completion(&wc
->wc_io_complete
);
353 /* Used in error paths too */
354 static inline void o2hb_bio_wait_dec(struct o2hb_bio_wait_ctxt
*wc
,
357 /* sadly atomic_sub_and_test() isn't available on all platforms. The
358 * good news is that the fast path only completes one at a time */
360 if (atomic_dec_and_test(&wc
->wc_num_reqs
)) {
362 complete(&wc
->wc_io_complete
);
367 static void o2hb_wait_on_io(struct o2hb_region
*reg
,
368 struct o2hb_bio_wait_ctxt
*wc
)
370 struct address_space
*mapping
= reg
->hr_bdev
->bd_inode
->i_mapping
;
372 blk_run_address_space(mapping
);
373 o2hb_bio_wait_dec(wc
, 1);
375 wait_for_completion(&wc
->wc_io_complete
);
378 static void o2hb_bio_end_io(struct bio
*bio
,
381 struct o2hb_bio_wait_ctxt
*wc
= bio
->bi_private
;
384 mlog(ML_ERROR
, "IO Error %d\n", error
);
385 wc
->wc_error
= error
;
388 o2hb_bio_wait_dec(wc
, 1);
392 /* Setup a Bio to cover I/O against num_slots slots starting at
394 static struct bio
*o2hb_setup_one_bio(struct o2hb_region
*reg
,
395 struct o2hb_bio_wait_ctxt
*wc
,
396 unsigned int *current_slot
,
397 unsigned int max_slots
)
399 int len
, current_page
;
400 unsigned int vec_len
, vec_start
;
401 unsigned int bits
= reg
->hr_block_bits
;
402 unsigned int spp
= reg
->hr_slots_per_page
;
403 unsigned int cs
= *current_slot
;
407 /* Testing has shown this allocation to take long enough under
408 * GFP_KERNEL that the local node can get fenced. It would be
409 * nicest if we could pre-allocate these bios and avoid this
411 bio
= bio_alloc(GFP_ATOMIC
, 16);
413 mlog(ML_ERROR
, "Could not alloc slots BIO!\n");
414 bio
= ERR_PTR(-ENOMEM
);
418 /* Must put everything in 512 byte sectors for the bio... */
419 bio
->bi_sector
= (reg
->hr_start_block
+ cs
) << (bits
- 9);
420 bio
->bi_bdev
= reg
->hr_bdev
;
421 bio
->bi_private
= wc
;
422 bio
->bi_end_io
= o2hb_bio_end_io
;
424 vec_start
= (cs
<< bits
) % PAGE_CACHE_SIZE
;
425 while(cs
< max_slots
) {
426 current_page
= cs
/ spp
;
427 page
= reg
->hr_slot_data
[current_page
];
429 vec_len
= min(PAGE_CACHE_SIZE
- vec_start
,
430 (max_slots
-cs
) * (PAGE_CACHE_SIZE
/spp
) );
432 mlog(ML_HB_BIO
, "page %d, vec_len = %u, vec_start = %u\n",
433 current_page
, vec_len
, vec_start
);
435 len
= bio_add_page(bio
, page
, vec_len
, vec_start
);
436 if (len
!= vec_len
) break;
438 cs
+= vec_len
/ (PAGE_CACHE_SIZE
/spp
);
447 static int o2hb_read_slots(struct o2hb_region
*reg
,
448 unsigned int max_slots
)
450 unsigned int current_slot
=0;
452 struct o2hb_bio_wait_ctxt wc
;
455 o2hb_bio_wait_init(&wc
);
457 while(current_slot
< max_slots
) {
458 bio
= o2hb_setup_one_bio(reg
, &wc
, ¤t_slot
, max_slots
);
460 status
= PTR_ERR(bio
);
465 atomic_inc(&wc
.wc_num_reqs
);
466 submit_bio(READ
, bio
);
472 o2hb_wait_on_io(reg
, &wc
);
473 if (wc
.wc_error
&& !status
)
474 status
= wc
.wc_error
;
479 static int o2hb_issue_node_write(struct o2hb_region
*reg
,
480 struct o2hb_bio_wait_ctxt
*write_wc
)
486 o2hb_bio_wait_init(write_wc
);
488 slot
= o2nm_this_node();
490 bio
= o2hb_setup_one_bio(reg
, write_wc
, &slot
, slot
+1);
492 status
= PTR_ERR(bio
);
497 atomic_inc(&write_wc
->wc_num_reqs
);
498 submit_bio(WRITE
, bio
);
505 static u32
o2hb_compute_block_crc_le(struct o2hb_region
*reg
,
506 struct o2hb_disk_heartbeat_block
*hb_block
)
511 /* We want to compute the block crc with a 0 value in the
512 * hb_cksum field. Save it off here and replace after the
514 old_cksum
= hb_block
->hb_cksum
;
515 hb_block
->hb_cksum
= 0;
517 ret
= crc32_le(0, (unsigned char *) hb_block
, reg
->hr_block_bytes
);
519 hb_block
->hb_cksum
= old_cksum
;
524 static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block
*hb_block
)
526 mlog(ML_ERROR
, "Dump slot information: seq = 0x%llx, node = %u, "
527 "cksum = 0x%x, generation 0x%llx\n",
528 (long long)le64_to_cpu(hb_block
->hb_seq
),
529 hb_block
->hb_node
, le32_to_cpu(hb_block
->hb_cksum
),
530 (long long)le64_to_cpu(hb_block
->hb_generation
));
533 static int o2hb_verify_crc(struct o2hb_region
*reg
,
534 struct o2hb_disk_heartbeat_block
*hb_block
)
538 read
= le32_to_cpu(hb_block
->hb_cksum
);
539 computed
= o2hb_compute_block_crc_le(reg
, hb_block
);
541 return read
== computed
;
544 /* We want to make sure that nobody is heartbeating on top of us --
545 * this will help detect an invalid configuration. */
546 static int o2hb_check_last_timestamp(struct o2hb_region
*reg
)
549 struct o2hb_disk_slot
*slot
;
550 struct o2hb_disk_heartbeat_block
*hb_block
;
552 node_num
= o2nm_this_node();
555 slot
= ®
->hr_slots
[node_num
];
556 /* Don't check on our 1st timestamp */
557 if (slot
->ds_last_time
) {
558 hb_block
= slot
->ds_raw_block
;
560 if (le64_to_cpu(hb_block
->hb_seq
) != slot
->ds_last_time
)
567 static inline void o2hb_prepare_block(struct o2hb_region
*reg
,
572 struct o2hb_disk_slot
*slot
;
573 struct o2hb_disk_heartbeat_block
*hb_block
;
575 node_num
= o2nm_this_node();
576 slot
= ®
->hr_slots
[node_num
];
578 hb_block
= (struct o2hb_disk_heartbeat_block
*)slot
->ds_raw_block
;
579 memset(hb_block
, 0, reg
->hr_block_bytes
);
580 /* TODO: time stuff */
581 cputime
= CURRENT_TIME
.tv_sec
;
585 hb_block
->hb_seq
= cpu_to_le64(cputime
);
586 hb_block
->hb_node
= node_num
;
587 hb_block
->hb_generation
= cpu_to_le64(generation
);
588 hb_block
->hb_dead_ms
= cpu_to_le32(o2hb_dead_threshold
* O2HB_REGION_TIMEOUT_MS
);
590 /* This step must always happen last! */
591 hb_block
->hb_cksum
= cpu_to_le32(o2hb_compute_block_crc_le(reg
,
594 mlog(ML_HB_BIO
, "our node generation = 0x%llx, cksum = 0x%x\n",
595 (long long)generation
,
596 le32_to_cpu(hb_block
->hb_cksum
));
599 static void o2hb_fire_callbacks(struct o2hb_callback
*hbcall
,
600 struct o2nm_node
*node
,
603 struct list_head
*iter
;
604 struct o2hb_callback_func
*f
;
606 list_for_each(iter
, &hbcall
->list
) {
607 f
= list_entry(iter
, struct o2hb_callback_func
, hc_item
);
608 mlog(ML_HEARTBEAT
, "calling funcs %p\n", f
);
609 (f
->hc_func
)(node
, idx
, f
->hc_data
);
613 /* Will run the list in order until we process the passed event */
614 static void o2hb_run_event_list(struct o2hb_node_event
*queued_event
)
617 struct o2hb_callback
*hbcall
;
618 struct o2hb_node_event
*event
;
620 spin_lock(&o2hb_live_lock
);
621 empty
= list_empty(&queued_event
->hn_item
);
622 spin_unlock(&o2hb_live_lock
);
626 /* Holding callback sem assures we don't alter the callback
627 * lists when doing this, and serializes ourselves with other
628 * processes wanting callbacks. */
629 down_write(&o2hb_callback_sem
);
631 spin_lock(&o2hb_live_lock
);
632 while (!list_empty(&o2hb_node_events
)
633 && !list_empty(&queued_event
->hn_item
)) {
634 event
= list_entry(o2hb_node_events
.next
,
635 struct o2hb_node_event
,
637 list_del_init(&event
->hn_item
);
638 spin_unlock(&o2hb_live_lock
);
640 mlog(ML_HEARTBEAT
, "Node %s event for %d\n",
641 event
->hn_event_type
== O2HB_NODE_UP_CB
? "UP" : "DOWN",
644 hbcall
= hbcall_from_type(event
->hn_event_type
);
646 /* We should *never* have gotten on to the list with a
647 * bad type... This isn't something that we should try
648 * to recover from. */
649 BUG_ON(IS_ERR(hbcall
));
651 o2hb_fire_callbacks(hbcall
, event
->hn_node
, event
->hn_node_num
);
653 spin_lock(&o2hb_live_lock
);
655 spin_unlock(&o2hb_live_lock
);
657 up_write(&o2hb_callback_sem
);
660 static void o2hb_queue_node_event(struct o2hb_node_event
*event
,
661 enum o2hb_callback_type type
,
662 struct o2nm_node
*node
,
665 assert_spin_locked(&o2hb_live_lock
);
667 BUG_ON((!node
) && (type
!= O2HB_NODE_DOWN_CB
));
669 event
->hn_event_type
= type
;
670 event
->hn_node
= node
;
671 event
->hn_node_num
= node_num
;
673 mlog(ML_HEARTBEAT
, "Queue node %s event for node %d\n",
674 type
== O2HB_NODE_UP_CB
? "UP" : "DOWN", node_num
);
676 list_add_tail(&event
->hn_item
, &o2hb_node_events
);
679 static void o2hb_shutdown_slot(struct o2hb_disk_slot
*slot
)
681 struct o2hb_node_event event
=
682 { .hn_item
= LIST_HEAD_INIT(event
.hn_item
), };
683 struct o2nm_node
*node
;
685 node
= o2nm_get_node_by_num(slot
->ds_node_num
);
689 spin_lock(&o2hb_live_lock
);
690 if (!list_empty(&slot
->ds_live_item
)) {
691 mlog(ML_HEARTBEAT
, "Shutdown, node %d leaves region\n",
694 list_del_init(&slot
->ds_live_item
);
696 if (list_empty(&o2hb_live_slots
[slot
->ds_node_num
])) {
697 clear_bit(slot
->ds_node_num
, o2hb_live_node_bitmap
);
699 o2hb_queue_node_event(&event
, O2HB_NODE_DOWN_CB
, node
,
703 spin_unlock(&o2hb_live_lock
);
705 o2hb_run_event_list(&event
);
710 static void o2hb_set_quorum_device(struct o2hb_region
*reg
,
711 struct o2hb_disk_slot
*slot
)
713 assert_spin_locked(&o2hb_live_lock
);
715 if (!o2hb_global_heartbeat_active())
718 if (test_bit(reg
->hr_region_num
, o2hb_quorum_region_bitmap
))
722 * A region can be added to the quorum only when it sees all
723 * live nodes heartbeat on it. In other words, the region has been
724 * added to all nodes.
726 if (memcmp(reg
->hr_live_node_bitmap
, o2hb_live_node_bitmap
,
727 sizeof(o2hb_live_node_bitmap
)))
730 if (slot
->ds_changed_samples
< O2HB_LIVE_THRESHOLD
)
733 printk(KERN_NOTICE
"o2hb: Region %s is now a quorum device\n",
734 config_item_name(®
->hr_item
));
736 set_bit(reg
->hr_region_num
, o2hb_quorum_region_bitmap
);
739 * If global heartbeat active, unpin all regions if the
740 * region count > CUT_OFF
742 if (o2hb_pop_count(&o2hb_quorum_region_bitmap
,
743 O2NM_MAX_REGIONS
) > O2HB_PIN_CUT_OFF
)
744 o2hb_region_unpin(NULL
);
747 static int o2hb_check_slot(struct o2hb_region
*reg
,
748 struct o2hb_disk_slot
*slot
)
750 int changed
= 0, gen_changed
= 0;
751 struct o2hb_node_event event
=
752 { .hn_item
= LIST_HEAD_INIT(event
.hn_item
), };
753 struct o2nm_node
*node
;
754 struct o2hb_disk_heartbeat_block
*hb_block
= reg
->hr_tmp_block
;
756 unsigned int dead_ms
= o2hb_dead_threshold
* O2HB_REGION_TIMEOUT_MS
;
757 unsigned int slot_dead_ms
;
760 memcpy(hb_block
, slot
->ds_raw_block
, reg
->hr_block_bytes
);
763 * If a node is no longer configured but is still in the livemap, we
764 * may need to clear that bit from the livemap.
766 node
= o2nm_get_node_by_num(slot
->ds_node_num
);
768 spin_lock(&o2hb_live_lock
);
769 tmp
= test_bit(slot
->ds_node_num
, o2hb_live_node_bitmap
);
770 spin_unlock(&o2hb_live_lock
);
775 if (!o2hb_verify_crc(reg
, hb_block
)) {
776 /* all paths from here will drop o2hb_live_lock for
778 spin_lock(&o2hb_live_lock
);
780 /* Don't print an error on the console in this case -
781 * a freshly formatted heartbeat area will not have a
783 if (list_empty(&slot
->ds_live_item
))
786 /* The node is live but pushed out a bad crc. We
787 * consider it a transient miss but don't populate any
788 * other values as they may be junk. */
789 mlog(ML_ERROR
, "Node %d has written a bad crc to %s\n",
790 slot
->ds_node_num
, reg
->hr_dev_name
);
791 o2hb_dump_slot(hb_block
);
793 slot
->ds_equal_samples
++;
797 /* we don't care if these wrap.. the state transitions below
798 * clear at the right places */
799 cputime
= le64_to_cpu(hb_block
->hb_seq
);
800 if (slot
->ds_last_time
!= cputime
)
801 slot
->ds_changed_samples
++;
803 slot
->ds_equal_samples
++;
804 slot
->ds_last_time
= cputime
;
806 /* The node changed heartbeat generations. We assume this to
807 * mean it dropped off but came back before we timed out. We
808 * want to consider it down for the time being but don't want
809 * to lose any changed_samples state we might build up to
810 * considering it live again. */
811 if (slot
->ds_last_generation
!= le64_to_cpu(hb_block
->hb_generation
)) {
813 slot
->ds_equal_samples
= 0;
814 mlog(ML_HEARTBEAT
, "Node %d changed generation (0x%llx "
815 "to 0x%llx)\n", slot
->ds_node_num
,
816 (long long)slot
->ds_last_generation
,
817 (long long)le64_to_cpu(hb_block
->hb_generation
));
820 slot
->ds_last_generation
= le64_to_cpu(hb_block
->hb_generation
);
822 mlog(ML_HEARTBEAT
, "Slot %d gen 0x%llx cksum 0x%x "
823 "seq %llu last %llu changed %u equal %u\n",
824 slot
->ds_node_num
, (long long)slot
->ds_last_generation
,
825 le32_to_cpu(hb_block
->hb_cksum
),
826 (unsigned long long)le64_to_cpu(hb_block
->hb_seq
),
827 (unsigned long long)slot
->ds_last_time
, slot
->ds_changed_samples
,
828 slot
->ds_equal_samples
);
830 spin_lock(&o2hb_live_lock
);
833 /* dead nodes only come to life after some number of
834 * changes at any time during their dead time */
835 if (list_empty(&slot
->ds_live_item
) &&
836 slot
->ds_changed_samples
>= O2HB_LIVE_THRESHOLD
) {
837 mlog(ML_HEARTBEAT
, "Node %d (id 0x%llx) joined my region\n",
838 slot
->ds_node_num
, (long long)slot
->ds_last_generation
);
840 set_bit(slot
->ds_node_num
, reg
->hr_live_node_bitmap
);
842 /* first on the list generates a callback */
843 if (list_empty(&o2hb_live_slots
[slot
->ds_node_num
])) {
844 mlog(ML_HEARTBEAT
, "o2hb: Add node %d to live nodes "
845 "bitmap\n", slot
->ds_node_num
);
846 set_bit(slot
->ds_node_num
, o2hb_live_node_bitmap
);
848 o2hb_queue_node_event(&event
, O2HB_NODE_UP_CB
, node
,
854 list_add_tail(&slot
->ds_live_item
,
855 &o2hb_live_slots
[slot
->ds_node_num
]);
857 slot
->ds_equal_samples
= 0;
859 /* We want to be sure that all nodes agree on the
860 * number of milliseconds before a node will be
861 * considered dead. The self-fencing timeout is
862 * computed from this value, and a discrepancy might
863 * result in heartbeat calling a node dead when it
864 * hasn't self-fenced yet. */
865 slot_dead_ms
= le32_to_cpu(hb_block
->hb_dead_ms
);
866 if (slot_dead_ms
&& slot_dead_ms
!= dead_ms
) {
867 /* TODO: Perhaps we can fail the region here. */
868 mlog(ML_ERROR
, "Node %d on device %s has a dead count "
869 "of %u ms, but our count is %u ms.\n"
870 "Please double check your configuration values "
871 "for 'O2CB_HEARTBEAT_THRESHOLD'\n",
872 slot
->ds_node_num
, reg
->hr_dev_name
, slot_dead_ms
,
878 /* if the list is dead, we're done.. */
879 if (list_empty(&slot
->ds_live_item
))
882 /* live nodes only go dead after enough consequtive missed
883 * samples.. reset the missed counter whenever we see
885 if (slot
->ds_equal_samples
>= o2hb_dead_threshold
|| gen_changed
) {
886 mlog(ML_HEARTBEAT
, "Node %d left my region\n",
889 clear_bit(slot
->ds_node_num
, reg
->hr_live_node_bitmap
);
891 /* last off the live_slot generates a callback */
892 list_del_init(&slot
->ds_live_item
);
893 if (list_empty(&o2hb_live_slots
[slot
->ds_node_num
])) {
894 mlog(ML_HEARTBEAT
, "o2hb: Remove node %d from live "
895 "nodes bitmap\n", slot
->ds_node_num
);
896 clear_bit(slot
->ds_node_num
, o2hb_live_node_bitmap
);
898 /* node can be null */
899 o2hb_queue_node_event(&event
, O2HB_NODE_DOWN_CB
,
900 node
, slot
->ds_node_num
);
905 /* We don't clear this because the node is still
906 * actually writing new blocks. */
908 slot
->ds_changed_samples
= 0;
911 if (slot
->ds_changed_samples
) {
912 slot
->ds_changed_samples
= 0;
913 slot
->ds_equal_samples
= 0;
916 o2hb_set_quorum_device(reg
, slot
);
918 spin_unlock(&o2hb_live_lock
);
920 o2hb_run_event_list(&event
);
927 /* This could be faster if we just implmented a find_last_bit, but I
928 * don't think the circumstances warrant it. */
929 static int o2hb_highest_node(unsigned long *nodes
,
936 while ((node
= find_next_bit(nodes
, numbits
, node
+ 1)) != -1) {
946 static int o2hb_do_disk_heartbeat(struct o2hb_region
*reg
)
948 int i
, ret
, highest_node
, change
= 0;
949 unsigned long configured_nodes
[BITS_TO_LONGS(O2NM_MAX_NODES
)];
950 unsigned long live_node_bitmap
[BITS_TO_LONGS(O2NM_MAX_NODES
)];
951 struct o2hb_bio_wait_ctxt write_wc
;
953 ret
= o2nm_configured_node_map(configured_nodes
,
954 sizeof(configured_nodes
));
961 * If a node is not configured but is in the livemap, we still need
962 * to read the slot so as to be able to remove it from the livemap.
964 o2hb_fill_node_map(live_node_bitmap
, sizeof(live_node_bitmap
));
966 while ((i
= find_next_bit(live_node_bitmap
,
967 O2NM_MAX_NODES
, i
+ 1)) < O2NM_MAX_NODES
) {
968 set_bit(i
, configured_nodes
);
971 highest_node
= o2hb_highest_node(configured_nodes
, O2NM_MAX_NODES
);
972 if (highest_node
>= O2NM_MAX_NODES
) {
973 mlog(ML_NOTICE
, "ocfs2_heartbeat: no configured nodes found!\n");
977 /* No sense in reading the slots of nodes that don't exist
978 * yet. Of course, if the node definitions have holes in them
979 * then we're reading an empty slot anyway... Consider this
981 ret
= o2hb_read_slots(reg
, highest_node
+ 1);
987 /* With an up to date view of the slots, we can check that no
988 * other node has been improperly configured to heartbeat in
990 if (!o2hb_check_last_timestamp(reg
))
991 mlog(ML_ERROR
, "Device \"%s\": another node is heartbeating "
992 "in our slot!\n", reg
->hr_dev_name
);
994 /* fill in the proper info for our next heartbeat */
995 o2hb_prepare_block(reg
, reg
->hr_generation
);
997 /* And fire off the write. Note that we don't wait on this I/O
999 ret
= o2hb_issue_node_write(reg
, &write_wc
);
1006 while((i
= find_next_bit(configured_nodes
, O2NM_MAX_NODES
, i
+ 1)) < O2NM_MAX_NODES
) {
1008 change
|= o2hb_check_slot(reg
, ®
->hr_slots
[i
]);
1012 * We have to be sure we've advertised ourselves on disk
1013 * before we can go to steady state. This ensures that
1014 * people we find in our steady state have seen us.
1016 o2hb_wait_on_io(reg
, &write_wc
);
1017 if (write_wc
.wc_error
) {
1018 /* Do not re-arm the write timeout on I/O error - we
1019 * can't be sure that the new block ever made it to
1021 mlog(ML_ERROR
, "Write error %d on device \"%s\"\n",
1022 write_wc
.wc_error
, reg
->hr_dev_name
);
1023 return write_wc
.wc_error
;
1026 o2hb_arm_write_timeout(reg
);
1028 /* let the person who launched us know when things are steady */
1029 if (!change
&& (atomic_read(®
->hr_steady_iterations
) != 0)) {
1030 if (atomic_dec_and_test(®
->hr_steady_iterations
))
1031 wake_up(&o2hb_steady_queue
);
1037 /* Subtract b from a, storing the result in a. a *must* have a larger
1039 static void o2hb_tv_subtract(struct timeval
*a
,
1042 /* just return 0 when a is after b */
1043 if (a
->tv_sec
< b
->tv_sec
||
1044 (a
->tv_sec
== b
->tv_sec
&& a
->tv_usec
< b
->tv_usec
)) {
1050 a
->tv_sec
-= b
->tv_sec
;
1051 a
->tv_usec
-= b
->tv_usec
;
1052 while ( a
->tv_usec
< 0 ) {
1054 a
->tv_usec
+= 1000000;
1058 static unsigned int o2hb_elapsed_msecs(struct timeval
*start
,
1059 struct timeval
*end
)
1061 struct timeval res
= *end
;
1063 o2hb_tv_subtract(&res
, start
);
1065 return res
.tv_sec
* 1000 + res
.tv_usec
/ 1000;
1069 * we ride the region ref that the region dir holds. before the region
1070 * dir is removed and drops it ref it will wait to tear down this
1073 static int o2hb_thread(void *data
)
1076 struct o2hb_region
*reg
= data
;
1077 struct o2hb_bio_wait_ctxt write_wc
;
1078 struct timeval before_hb
, after_hb
;
1079 unsigned int elapsed_msec
;
1081 mlog(ML_HEARTBEAT
|ML_KTHREAD
, "hb thread running\n");
1083 set_user_nice(current
, -20);
1086 o2nm_depend_this_node();
1088 while (!kthread_should_stop() && !reg
->hr_unclean_stop
) {
1089 /* We track the time spent inside
1090 * o2hb_do_disk_heartbeat so that we avoid more than
1091 * hr_timeout_ms between disk writes. On busy systems
1092 * this should result in a heartbeat which is less
1093 * likely to time itself out. */
1094 do_gettimeofday(&before_hb
);
1098 ret
= o2hb_do_disk_heartbeat(reg
);
1099 } while (ret
&& ++i
< 2);
1101 do_gettimeofday(&after_hb
);
1102 elapsed_msec
= o2hb_elapsed_msecs(&before_hb
, &after_hb
);
1105 "start = %lu.%lu, end = %lu.%lu, msec = %u\n",
1106 before_hb
.tv_sec
, (unsigned long) before_hb
.tv_usec
,
1107 after_hb
.tv_sec
, (unsigned long) after_hb
.tv_usec
,
1110 if (elapsed_msec
< reg
->hr_timeout_ms
) {
1111 /* the kthread api has blocked signals for us so no
1112 * need to record the return value. */
1113 msleep_interruptible(reg
->hr_timeout_ms
- elapsed_msec
);
1117 o2hb_disarm_write_timeout(reg
);
1119 /* unclean stop is only used in very bad situation */
1120 for(i
= 0; !reg
->hr_unclean_stop
&& i
< reg
->hr_blocks
; i
++)
1121 o2hb_shutdown_slot(®
->hr_slots
[i
]);
1123 /* Explicit down notification - avoid forcing the other nodes
1124 * to timeout on this region when we could just as easily
1125 * write a clear generation - thus indicating to them that
1126 * this node has left this region.
1128 * XXX: Should we skip this on unclean_stop? */
1129 o2hb_prepare_block(reg
, 0);
1130 ret
= o2hb_issue_node_write(reg
, &write_wc
);
1132 o2hb_wait_on_io(reg
, &write_wc
);
1138 o2nm_undepend_this_node();
1140 mlog(ML_HEARTBEAT
|ML_KTHREAD
, "hb thread exiting\n");
1145 #ifdef CONFIG_DEBUG_FS
1146 static int o2hb_debug_open(struct inode
*inode
, struct file
*file
)
1148 struct o2hb_debug_buf
*db
= inode
->i_private
;
1149 struct o2hb_region
*reg
;
1150 unsigned long map
[BITS_TO_LONGS(O2NM_MAX_NODES
)];
1155 /* max_nodes should be the largest bitmap we pass here */
1156 BUG_ON(sizeof(map
) < db
->db_size
);
1158 buf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
1162 switch (db
->db_type
) {
1163 case O2HB_DB_TYPE_LIVENODES
:
1164 case O2HB_DB_TYPE_LIVEREGIONS
:
1165 case O2HB_DB_TYPE_QUORUMREGIONS
:
1166 case O2HB_DB_TYPE_FAILEDREGIONS
:
1167 spin_lock(&o2hb_live_lock
);
1168 memcpy(map
, db
->db_data
, db
->db_size
);
1169 spin_unlock(&o2hb_live_lock
);
1172 case O2HB_DB_TYPE_REGION_LIVENODES
:
1173 spin_lock(&o2hb_live_lock
);
1174 reg
= (struct o2hb_region
*)db
->db_data
;
1175 memcpy(map
, reg
->hr_live_node_bitmap
, db
->db_size
);
1176 spin_unlock(&o2hb_live_lock
);
1179 case O2HB_DB_TYPE_REGION_NUMBER
:
1180 reg
= (struct o2hb_region
*)db
->db_data
;
1181 out
+= snprintf(buf
+ out
, PAGE_SIZE
- out
, "%d\n",
1182 reg
->hr_region_num
);
1185 case O2HB_DB_TYPE_REGION_ELAPSED_TIME
:
1186 reg
= (struct o2hb_region
*)db
->db_data
;
1187 out
+= snprintf(buf
+ out
, PAGE_SIZE
- out
, "%u\n",
1188 jiffies_to_msecs(jiffies
-
1189 reg
->hr_last_timeout_start
));
1192 case O2HB_DB_TYPE_REGION_PINNED
:
1193 reg
= (struct o2hb_region
*)db
->db_data
;
1194 out
+= snprintf(buf
+ out
, PAGE_SIZE
- out
, "%u\n",
1195 !!reg
->hr_item_pinned
);
1202 while ((i
= find_next_bit(map
, db
->db_len
, i
+ 1)) < db
->db_len
)
1203 out
+= snprintf(buf
+ out
, PAGE_SIZE
- out
, "%d ", i
);
1204 out
+= snprintf(buf
+ out
, PAGE_SIZE
- out
, "\n");
1207 i_size_write(inode
, out
);
1209 file
->private_data
= buf
;
1216 static int o2hb_debug_release(struct inode
*inode
, struct file
*file
)
1218 kfree(file
->private_data
);
1222 static ssize_t
o2hb_debug_read(struct file
*file
, char __user
*buf
,
1223 size_t nbytes
, loff_t
*ppos
)
1225 return simple_read_from_buffer(buf
, nbytes
, ppos
, file
->private_data
,
1226 i_size_read(file
->f_mapping
->host
));
1229 static int o2hb_debug_open(struct inode
*inode
, struct file
*file
)
1233 static int o2hb_debug_release(struct inode
*inode
, struct file
*file
)
1237 static ssize_t
o2hb_debug_read(struct file
*file
, char __user
*buf
,
1238 size_t nbytes
, loff_t
*ppos
)
1242 #endif /* CONFIG_DEBUG_FS */
1244 static const struct file_operations o2hb_debug_fops
= {
1245 .open
= o2hb_debug_open
,
1246 .release
= o2hb_debug_release
,
1247 .read
= o2hb_debug_read
,
1248 .llseek
= generic_file_llseek
,
1251 void o2hb_exit(void)
1253 kfree(o2hb_db_livenodes
);
1254 kfree(o2hb_db_liveregions
);
1255 kfree(o2hb_db_quorumregions
);
1256 kfree(o2hb_db_failedregions
);
1257 debugfs_remove(o2hb_debug_failedregions
);
1258 debugfs_remove(o2hb_debug_quorumregions
);
1259 debugfs_remove(o2hb_debug_liveregions
);
1260 debugfs_remove(o2hb_debug_livenodes
);
1261 debugfs_remove(o2hb_debug_dir
);
1264 static struct dentry
*o2hb_debug_create(const char *name
, struct dentry
*dir
,
1265 struct o2hb_debug_buf
**db
, int db_len
,
1266 int type
, int size
, int len
, void *data
)
1268 *db
= kmalloc(db_len
, GFP_KERNEL
);
1272 (*db
)->db_type
= type
;
1273 (*db
)->db_size
= size
;
1274 (*db
)->db_len
= len
;
1275 (*db
)->db_data
= data
;
1277 return debugfs_create_file(name
, S_IFREG
|S_IRUSR
, dir
, *db
,
1281 static int o2hb_debug_init(void)
1285 o2hb_debug_dir
= debugfs_create_dir(O2HB_DEBUG_DIR
, NULL
);
1286 if (!o2hb_debug_dir
) {
1291 o2hb_debug_livenodes
= o2hb_debug_create(O2HB_DEBUG_LIVENODES
,
1294 sizeof(*o2hb_db_livenodes
),
1295 O2HB_DB_TYPE_LIVENODES
,
1296 sizeof(o2hb_live_node_bitmap
),
1298 o2hb_live_node_bitmap
);
1299 if (!o2hb_debug_livenodes
) {
1304 o2hb_debug_liveregions
= o2hb_debug_create(O2HB_DEBUG_LIVEREGIONS
,
1306 &o2hb_db_liveregions
,
1307 sizeof(*o2hb_db_liveregions
),
1308 O2HB_DB_TYPE_LIVEREGIONS
,
1309 sizeof(o2hb_live_region_bitmap
),
1311 o2hb_live_region_bitmap
);
1312 if (!o2hb_debug_liveregions
) {
1317 o2hb_debug_quorumregions
=
1318 o2hb_debug_create(O2HB_DEBUG_QUORUMREGIONS
,
1320 &o2hb_db_quorumregions
,
1321 sizeof(*o2hb_db_quorumregions
),
1322 O2HB_DB_TYPE_QUORUMREGIONS
,
1323 sizeof(o2hb_quorum_region_bitmap
),
1325 o2hb_quorum_region_bitmap
);
1326 if (!o2hb_debug_quorumregions
) {
1331 o2hb_debug_failedregions
=
1332 o2hb_debug_create(O2HB_DEBUG_FAILEDREGIONS
,
1334 &o2hb_db_failedregions
,
1335 sizeof(*o2hb_db_failedregions
),
1336 O2HB_DB_TYPE_FAILEDREGIONS
,
1337 sizeof(o2hb_failed_region_bitmap
),
1339 o2hb_failed_region_bitmap
);
1340 if (!o2hb_debug_failedregions
) {
1357 for (i
= 0; i
< ARRAY_SIZE(o2hb_callbacks
); i
++)
1358 INIT_LIST_HEAD(&o2hb_callbacks
[i
].list
);
1360 for (i
= 0; i
< ARRAY_SIZE(o2hb_live_slots
); i
++)
1361 INIT_LIST_HEAD(&o2hb_live_slots
[i
]);
1363 INIT_LIST_HEAD(&o2hb_node_events
);
1365 memset(o2hb_live_node_bitmap
, 0, sizeof(o2hb_live_node_bitmap
));
1366 memset(o2hb_region_bitmap
, 0, sizeof(o2hb_region_bitmap
));
1367 memset(o2hb_live_region_bitmap
, 0, sizeof(o2hb_live_region_bitmap
));
1368 memset(o2hb_quorum_region_bitmap
, 0, sizeof(o2hb_quorum_region_bitmap
));
1369 memset(o2hb_failed_region_bitmap
, 0, sizeof(o2hb_failed_region_bitmap
));
1371 o2hb_dependent_users
= 0;
1373 return o2hb_debug_init();
1376 /* if we're already in a callback then we're already serialized by the sem */
1377 static void o2hb_fill_node_map_from_callback(unsigned long *map
,
1380 BUG_ON(bytes
< (BITS_TO_LONGS(O2NM_MAX_NODES
) * sizeof(unsigned long)));
1382 memcpy(map
, &o2hb_live_node_bitmap
, bytes
);
1386 * get a map of all nodes that are heartbeating in any regions
1388 void o2hb_fill_node_map(unsigned long *map
, unsigned bytes
)
1390 /* callers want to serialize this map and callbacks so that they
1391 * can trust that they don't miss nodes coming to the party */
1392 down_read(&o2hb_callback_sem
);
1393 spin_lock(&o2hb_live_lock
);
1394 o2hb_fill_node_map_from_callback(map
, bytes
);
1395 spin_unlock(&o2hb_live_lock
);
1396 up_read(&o2hb_callback_sem
);
1398 EXPORT_SYMBOL_GPL(o2hb_fill_node_map
);
1401 * heartbeat configfs bits. The heartbeat set is a default set under
1402 * the cluster set in nodemanager.c.
1405 static struct o2hb_region
*to_o2hb_region(struct config_item
*item
)
1407 return item
? container_of(item
, struct o2hb_region
, hr_item
) : NULL
;
1410 /* drop_item only drops its ref after killing the thread, nothing should
1411 * be using the region anymore. this has to clean up any state that
1412 * attributes might have built up. */
1413 static void o2hb_region_release(struct config_item
*item
)
1417 struct o2hb_region
*reg
= to_o2hb_region(item
);
1419 if (reg
->hr_tmp_block
)
1420 kfree(reg
->hr_tmp_block
);
1422 if (reg
->hr_slot_data
) {
1423 for (i
= 0; i
< reg
->hr_num_pages
; i
++) {
1424 page
= reg
->hr_slot_data
[i
];
1428 kfree(reg
->hr_slot_data
);
1432 blkdev_put(reg
->hr_bdev
, FMODE_READ
|FMODE_WRITE
);
1435 kfree(reg
->hr_slots
);
1437 kfree(reg
->hr_db_regnum
);
1438 kfree(reg
->hr_db_livenodes
);
1439 debugfs_remove(reg
->hr_debug_livenodes
);
1440 debugfs_remove(reg
->hr_debug_regnum
);
1441 debugfs_remove(reg
->hr_debug_elapsed_time
);
1442 debugfs_remove(reg
->hr_debug_pinned
);
1443 debugfs_remove(reg
->hr_debug_dir
);
1445 spin_lock(&o2hb_live_lock
);
1446 list_del(®
->hr_all_item
);
1447 spin_unlock(&o2hb_live_lock
);
1452 static int o2hb_read_block_input(struct o2hb_region
*reg
,
1455 unsigned long *ret_bytes
,
1456 unsigned int *ret_bits
)
1458 unsigned long bytes
;
1459 char *p
= (char *)page
;
1461 bytes
= simple_strtoul(p
, &p
, 0);
1462 if (!p
|| (*p
&& (*p
!= '\n')))
1465 /* Heartbeat and fs min / max block sizes are the same. */
1466 if (bytes
> 4096 || bytes
< 512)
1468 if (hweight16(bytes
) != 1)
1474 *ret_bits
= ffs(bytes
) - 1;
1479 static ssize_t
o2hb_region_block_bytes_read(struct o2hb_region
*reg
,
1482 return sprintf(page
, "%u\n", reg
->hr_block_bytes
);
1485 static ssize_t
o2hb_region_block_bytes_write(struct o2hb_region
*reg
,
1490 unsigned long block_bytes
;
1491 unsigned int block_bits
;
1496 status
= o2hb_read_block_input(reg
, page
, count
,
1497 &block_bytes
, &block_bits
);
1501 reg
->hr_block_bytes
= (unsigned int)block_bytes
;
1502 reg
->hr_block_bits
= block_bits
;
1507 static ssize_t
o2hb_region_start_block_read(struct o2hb_region
*reg
,
1510 return sprintf(page
, "%llu\n", reg
->hr_start_block
);
1513 static ssize_t
o2hb_region_start_block_write(struct o2hb_region
*reg
,
1517 unsigned long long tmp
;
1518 char *p
= (char *)page
;
1523 tmp
= simple_strtoull(p
, &p
, 0);
1524 if (!p
|| (*p
&& (*p
!= '\n')))
1527 reg
->hr_start_block
= tmp
;
1532 static ssize_t
o2hb_region_blocks_read(struct o2hb_region
*reg
,
1535 return sprintf(page
, "%d\n", reg
->hr_blocks
);
1538 static ssize_t
o2hb_region_blocks_write(struct o2hb_region
*reg
,
1543 char *p
= (char *)page
;
1548 tmp
= simple_strtoul(p
, &p
, 0);
1549 if (!p
|| (*p
&& (*p
!= '\n')))
1552 if (tmp
> O2NM_MAX_NODES
|| tmp
== 0)
1555 reg
->hr_blocks
= (unsigned int)tmp
;
1560 static ssize_t
o2hb_region_dev_read(struct o2hb_region
*reg
,
1563 unsigned int ret
= 0;
1566 ret
= sprintf(page
, "%s\n", reg
->hr_dev_name
);
1571 static void o2hb_init_region_params(struct o2hb_region
*reg
)
1573 reg
->hr_slots_per_page
= PAGE_CACHE_SIZE
>> reg
->hr_block_bits
;
1574 reg
->hr_timeout_ms
= O2HB_REGION_TIMEOUT_MS
;
1576 mlog(ML_HEARTBEAT
, "hr_start_block = %llu, hr_blocks = %u\n",
1577 reg
->hr_start_block
, reg
->hr_blocks
);
1578 mlog(ML_HEARTBEAT
, "hr_block_bytes = %u, hr_block_bits = %u\n",
1579 reg
->hr_block_bytes
, reg
->hr_block_bits
);
1580 mlog(ML_HEARTBEAT
, "hr_timeout_ms = %u\n", reg
->hr_timeout_ms
);
1581 mlog(ML_HEARTBEAT
, "dead threshold = %u\n", o2hb_dead_threshold
);
1584 static int o2hb_map_slot_data(struct o2hb_region
*reg
)
1587 unsigned int last_slot
;
1588 unsigned int spp
= reg
->hr_slots_per_page
;
1591 struct o2hb_disk_slot
*slot
;
1593 reg
->hr_tmp_block
= kmalloc(reg
->hr_block_bytes
, GFP_KERNEL
);
1594 if (reg
->hr_tmp_block
== NULL
) {
1595 mlog_errno(-ENOMEM
);
1599 reg
->hr_slots
= kcalloc(reg
->hr_blocks
,
1600 sizeof(struct o2hb_disk_slot
), GFP_KERNEL
);
1601 if (reg
->hr_slots
== NULL
) {
1602 mlog_errno(-ENOMEM
);
1606 for(i
= 0; i
< reg
->hr_blocks
; i
++) {
1607 slot
= ®
->hr_slots
[i
];
1608 slot
->ds_node_num
= i
;
1609 INIT_LIST_HEAD(&slot
->ds_live_item
);
1610 slot
->ds_raw_block
= NULL
;
1613 reg
->hr_num_pages
= (reg
->hr_blocks
+ spp
- 1) / spp
;
1614 mlog(ML_HEARTBEAT
, "Going to require %u pages to cover %u blocks "
1615 "at %u blocks per page\n",
1616 reg
->hr_num_pages
, reg
->hr_blocks
, spp
);
1618 reg
->hr_slot_data
= kcalloc(reg
->hr_num_pages
, sizeof(struct page
*),
1620 if (!reg
->hr_slot_data
) {
1621 mlog_errno(-ENOMEM
);
1625 for(i
= 0; i
< reg
->hr_num_pages
; i
++) {
1626 page
= alloc_page(GFP_KERNEL
);
1628 mlog_errno(-ENOMEM
);
1632 reg
->hr_slot_data
[i
] = page
;
1634 last_slot
= i
* spp
;
1635 raw
= page_address(page
);
1637 (j
< spp
) && ((j
+ last_slot
) < reg
->hr_blocks
);
1639 BUG_ON((j
+ last_slot
) >= reg
->hr_blocks
);
1641 slot
= ®
->hr_slots
[j
+ last_slot
];
1642 slot
->ds_raw_block
=
1643 (struct o2hb_disk_heartbeat_block
*) raw
;
1645 raw
+= reg
->hr_block_bytes
;
1652 /* Read in all the slots available and populate the tracking
1653 * structures so that we can start with a baseline idea of what's
1655 static int o2hb_populate_slot_data(struct o2hb_region
*reg
)
1658 struct o2hb_disk_slot
*slot
;
1659 struct o2hb_disk_heartbeat_block
*hb_block
;
1663 ret
= o2hb_read_slots(reg
, reg
->hr_blocks
);
1669 /* We only want to get an idea of the values initially in each
1670 * slot, so we do no verification - o2hb_check_slot will
1671 * actually determine if each configured slot is valid and
1672 * whether any values have changed. */
1673 for(i
= 0; i
< reg
->hr_blocks
; i
++) {
1674 slot
= ®
->hr_slots
[i
];
1675 hb_block
= (struct o2hb_disk_heartbeat_block
*) slot
->ds_raw_block
;
1677 /* Only fill the values that o2hb_check_slot uses to
1678 * determine changing slots */
1679 slot
->ds_last_time
= le64_to_cpu(hb_block
->hb_seq
);
1680 slot
->ds_last_generation
= le64_to_cpu(hb_block
->hb_generation
);
1688 /* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */
1689 static ssize_t
o2hb_region_dev_write(struct o2hb_region
*reg
,
1693 struct task_struct
*hb_task
;
1696 char *p
= (char *)page
;
1697 struct file
*filp
= NULL
;
1698 struct inode
*inode
= NULL
;
1699 ssize_t ret
= -EINVAL
;
1704 /* We can't heartbeat without having had our node number
1705 * configured yet. */
1706 if (o2nm_this_node() == O2NM_MAX_NODES
)
1709 fd
= simple_strtol(p
, &p
, 0);
1710 if (!p
|| (*p
&& (*p
!= '\n')))
1713 if (fd
< 0 || fd
>= INT_MAX
)
1720 if (reg
->hr_blocks
== 0 || reg
->hr_start_block
== 0 ||
1721 reg
->hr_block_bytes
== 0)
1724 inode
= igrab(filp
->f_mapping
->host
);
1728 if (!S_ISBLK(inode
->i_mode
))
1731 reg
->hr_bdev
= I_BDEV(filp
->f_mapping
->host
);
1732 ret
= blkdev_get(reg
->hr_bdev
, FMODE_WRITE
| FMODE_READ
, NULL
);
1734 reg
->hr_bdev
= NULL
;
1739 bdevname(reg
->hr_bdev
, reg
->hr_dev_name
);
1741 sectsize
= bdev_logical_block_size(reg
->hr_bdev
);
1742 if (sectsize
!= reg
->hr_block_bytes
) {
1744 "blocksize %u incorrect for device, expected %d",
1745 reg
->hr_block_bytes
, sectsize
);
1750 o2hb_init_region_params(reg
);
1752 /* Generation of zero is invalid */
1754 get_random_bytes(®
->hr_generation
,
1755 sizeof(reg
->hr_generation
));
1756 } while (reg
->hr_generation
== 0);
1758 ret
= o2hb_map_slot_data(reg
);
1764 ret
= o2hb_populate_slot_data(reg
);
1770 INIT_DELAYED_WORK(®
->hr_write_timeout_work
, o2hb_write_timeout
);
1773 * A node is considered live after it has beat LIVE_THRESHOLD
1774 * times. We're not steady until we've given them a chance
1775 * _after_ our first read.
1777 atomic_set(®
->hr_steady_iterations
, O2HB_LIVE_THRESHOLD
+ 1);
1779 hb_task
= kthread_run(o2hb_thread
, reg
, "o2hb-%s",
1780 reg
->hr_item
.ci_name
);
1781 if (IS_ERR(hb_task
)) {
1782 ret
= PTR_ERR(hb_task
);
1787 spin_lock(&o2hb_live_lock
);
1788 reg
->hr_task
= hb_task
;
1789 spin_unlock(&o2hb_live_lock
);
1791 ret
= wait_event_interruptible(o2hb_steady_queue
,
1792 atomic_read(®
->hr_steady_iterations
) == 0);
1794 /* We got interrupted (hello ptrace!). Clean up */
1795 spin_lock(&o2hb_live_lock
);
1796 hb_task
= reg
->hr_task
;
1797 reg
->hr_task
= NULL
;
1798 spin_unlock(&o2hb_live_lock
);
1801 kthread_stop(hb_task
);
1805 /* Ok, we were woken. Make sure it wasn't by drop_item() */
1806 spin_lock(&o2hb_live_lock
);
1807 hb_task
= reg
->hr_task
;
1808 if (o2hb_global_heartbeat_active())
1809 set_bit(reg
->hr_region_num
, o2hb_live_region_bitmap
);
1810 spin_unlock(&o2hb_live_lock
);
1817 if (hb_task
&& o2hb_global_heartbeat_active())
1818 printk(KERN_NOTICE
"o2hb: Heartbeat started on region %s\n",
1819 config_item_name(®
->hr_item
));
1828 blkdev_put(reg
->hr_bdev
, FMODE_READ
|FMODE_WRITE
);
1829 reg
->hr_bdev
= NULL
;
1835 static ssize_t
o2hb_region_pid_read(struct o2hb_region
*reg
,
1840 spin_lock(&o2hb_live_lock
);
1842 pid
= task_pid_nr(reg
->hr_task
);
1843 spin_unlock(&o2hb_live_lock
);
1848 return sprintf(page
, "%u\n", pid
);
1851 struct o2hb_region_attribute
{
1852 struct configfs_attribute attr
;
1853 ssize_t (*show
)(struct o2hb_region
*, char *);
1854 ssize_t (*store
)(struct o2hb_region
*, const char *, size_t);
1857 static struct o2hb_region_attribute o2hb_region_attr_block_bytes
= {
1858 .attr
= { .ca_owner
= THIS_MODULE
,
1859 .ca_name
= "block_bytes",
1860 .ca_mode
= S_IRUGO
| S_IWUSR
},
1861 .show
= o2hb_region_block_bytes_read
,
1862 .store
= o2hb_region_block_bytes_write
,
1865 static struct o2hb_region_attribute o2hb_region_attr_start_block
= {
1866 .attr
= { .ca_owner
= THIS_MODULE
,
1867 .ca_name
= "start_block",
1868 .ca_mode
= S_IRUGO
| S_IWUSR
},
1869 .show
= o2hb_region_start_block_read
,
1870 .store
= o2hb_region_start_block_write
,
1873 static struct o2hb_region_attribute o2hb_region_attr_blocks
= {
1874 .attr
= { .ca_owner
= THIS_MODULE
,
1875 .ca_name
= "blocks",
1876 .ca_mode
= S_IRUGO
| S_IWUSR
},
1877 .show
= o2hb_region_blocks_read
,
1878 .store
= o2hb_region_blocks_write
,
1881 static struct o2hb_region_attribute o2hb_region_attr_dev
= {
1882 .attr
= { .ca_owner
= THIS_MODULE
,
1884 .ca_mode
= S_IRUGO
| S_IWUSR
},
1885 .show
= o2hb_region_dev_read
,
1886 .store
= o2hb_region_dev_write
,
1889 static struct o2hb_region_attribute o2hb_region_attr_pid
= {
1890 .attr
= { .ca_owner
= THIS_MODULE
,
1892 .ca_mode
= S_IRUGO
| S_IRUSR
},
1893 .show
= o2hb_region_pid_read
,
1896 static struct configfs_attribute
*o2hb_region_attrs
[] = {
1897 &o2hb_region_attr_block_bytes
.attr
,
1898 &o2hb_region_attr_start_block
.attr
,
1899 &o2hb_region_attr_blocks
.attr
,
1900 &o2hb_region_attr_dev
.attr
,
1901 &o2hb_region_attr_pid
.attr
,
1905 static ssize_t
o2hb_region_show(struct config_item
*item
,
1906 struct configfs_attribute
*attr
,
1909 struct o2hb_region
*reg
= to_o2hb_region(item
);
1910 struct o2hb_region_attribute
*o2hb_region_attr
=
1911 container_of(attr
, struct o2hb_region_attribute
, attr
);
1914 if (o2hb_region_attr
->show
)
1915 ret
= o2hb_region_attr
->show(reg
, page
);
1919 static ssize_t
o2hb_region_store(struct config_item
*item
,
1920 struct configfs_attribute
*attr
,
1921 const char *page
, size_t count
)
1923 struct o2hb_region
*reg
= to_o2hb_region(item
);
1924 struct o2hb_region_attribute
*o2hb_region_attr
=
1925 container_of(attr
, struct o2hb_region_attribute
, attr
);
1926 ssize_t ret
= -EINVAL
;
1928 if (o2hb_region_attr
->store
)
1929 ret
= o2hb_region_attr
->store(reg
, page
, count
);
1933 static struct configfs_item_operations o2hb_region_item_ops
= {
1934 .release
= o2hb_region_release
,
1935 .show_attribute
= o2hb_region_show
,
1936 .store_attribute
= o2hb_region_store
,
1939 static struct config_item_type o2hb_region_type
= {
1940 .ct_item_ops
= &o2hb_region_item_ops
,
1941 .ct_attrs
= o2hb_region_attrs
,
1942 .ct_owner
= THIS_MODULE
,
1947 struct o2hb_heartbeat_group
{
1948 struct config_group hs_group
;
1952 static struct o2hb_heartbeat_group
*to_o2hb_heartbeat_group(struct config_group
*group
)
1955 container_of(group
, struct o2hb_heartbeat_group
, hs_group
)
1959 static int o2hb_debug_region_init(struct o2hb_region
*reg
, struct dentry
*dir
)
1964 debugfs_create_dir(config_item_name(®
->hr_item
), dir
);
1965 if (!reg
->hr_debug_dir
) {
1970 reg
->hr_debug_livenodes
=
1971 o2hb_debug_create(O2HB_DEBUG_LIVENODES
,
1973 &(reg
->hr_db_livenodes
),
1974 sizeof(*(reg
->hr_db_livenodes
)),
1975 O2HB_DB_TYPE_REGION_LIVENODES
,
1976 sizeof(reg
->hr_live_node_bitmap
),
1977 O2NM_MAX_NODES
, reg
);
1978 if (!reg
->hr_debug_livenodes
) {
1983 reg
->hr_debug_regnum
=
1984 o2hb_debug_create(O2HB_DEBUG_REGION_NUMBER
,
1986 &(reg
->hr_db_regnum
),
1987 sizeof(*(reg
->hr_db_regnum
)),
1988 O2HB_DB_TYPE_REGION_NUMBER
,
1989 0, O2NM_MAX_NODES
, reg
);
1990 if (!reg
->hr_debug_regnum
) {
1995 reg
->hr_debug_elapsed_time
=
1996 o2hb_debug_create(O2HB_DEBUG_REGION_ELAPSED_TIME
,
1998 &(reg
->hr_db_elapsed_time
),
1999 sizeof(*(reg
->hr_db_elapsed_time
)),
2000 O2HB_DB_TYPE_REGION_ELAPSED_TIME
,
2002 if (!reg
->hr_debug_elapsed_time
) {
2007 reg
->hr_debug_pinned
=
2008 o2hb_debug_create(O2HB_DEBUG_REGION_PINNED
,
2010 &(reg
->hr_db_pinned
),
2011 sizeof(*(reg
->hr_db_pinned
)),
2012 O2HB_DB_TYPE_REGION_PINNED
,
2014 if (!reg
->hr_debug_pinned
) {
2024 static struct config_item
*o2hb_heartbeat_group_make_item(struct config_group
*group
,
2027 struct o2hb_region
*reg
= NULL
;
2030 reg
= kzalloc(sizeof(struct o2hb_region
), GFP_KERNEL
);
2032 return ERR_PTR(-ENOMEM
);
2034 if (strlen(name
) > O2HB_MAX_REGION_NAME_LEN
) {
2035 ret
= -ENAMETOOLONG
;
2039 spin_lock(&o2hb_live_lock
);
2040 reg
->hr_region_num
= 0;
2041 if (o2hb_global_heartbeat_active()) {
2042 reg
->hr_region_num
= find_first_zero_bit(o2hb_region_bitmap
,
2044 if (reg
->hr_region_num
>= O2NM_MAX_REGIONS
) {
2045 spin_unlock(&o2hb_live_lock
);
2049 set_bit(reg
->hr_region_num
, o2hb_region_bitmap
);
2051 list_add_tail(®
->hr_all_item
, &o2hb_all_regions
);
2052 spin_unlock(&o2hb_live_lock
);
2054 config_item_init_type_name(®
->hr_item
, name
, &o2hb_region_type
);
2056 ret
= o2hb_debug_region_init(reg
, o2hb_debug_dir
);
2058 config_item_put(®
->hr_item
);
2062 return ®
->hr_item
;
2065 return ERR_PTR(ret
);
2068 static void o2hb_heartbeat_group_drop_item(struct config_group
*group
,
2069 struct config_item
*item
)
2071 struct task_struct
*hb_task
;
2072 struct o2hb_region
*reg
= to_o2hb_region(item
);
2073 int quorum_region
= 0;
2075 /* stop the thread when the user removes the region dir */
2076 spin_lock(&o2hb_live_lock
);
2077 if (o2hb_global_heartbeat_active()) {
2078 clear_bit(reg
->hr_region_num
, o2hb_region_bitmap
);
2079 clear_bit(reg
->hr_region_num
, o2hb_live_region_bitmap
);
2080 if (test_bit(reg
->hr_region_num
, o2hb_quorum_region_bitmap
))
2082 clear_bit(reg
->hr_region_num
, o2hb_quorum_region_bitmap
);
2084 hb_task
= reg
->hr_task
;
2085 reg
->hr_task
= NULL
;
2086 reg
->hr_item_dropped
= 1;
2087 spin_unlock(&o2hb_live_lock
);
2090 kthread_stop(hb_task
);
2093 * If we're racing a dev_write(), we need to wake them. They will
2094 * check reg->hr_task
2096 if (atomic_read(®
->hr_steady_iterations
) != 0) {
2097 atomic_set(®
->hr_steady_iterations
, 0);
2098 wake_up(&o2hb_steady_queue
);
2101 if (o2hb_global_heartbeat_active())
2102 printk(KERN_NOTICE
"o2hb: Heartbeat stopped on region %s\n",
2103 config_item_name(®
->hr_item
));
2105 config_item_put(item
);
2107 if (!o2hb_global_heartbeat_active() || !quorum_region
)
2111 * If global heartbeat active and there are dependent users,
2112 * pin all regions if quorum region count <= CUT_OFF
2114 spin_lock(&o2hb_live_lock
);
2116 if (!o2hb_dependent_users
)
2119 if (o2hb_pop_count(&o2hb_quorum_region_bitmap
,
2120 O2NM_MAX_REGIONS
) <= O2HB_PIN_CUT_OFF
)
2121 o2hb_region_pin(NULL
);
2124 spin_unlock(&o2hb_live_lock
);
2127 struct o2hb_heartbeat_group_attribute
{
2128 struct configfs_attribute attr
;
2129 ssize_t (*show
)(struct o2hb_heartbeat_group
*, char *);
2130 ssize_t (*store
)(struct o2hb_heartbeat_group
*, const char *, size_t);
2133 static ssize_t
o2hb_heartbeat_group_show(struct config_item
*item
,
2134 struct configfs_attribute
*attr
,
2137 struct o2hb_heartbeat_group
*reg
= to_o2hb_heartbeat_group(to_config_group(item
));
2138 struct o2hb_heartbeat_group_attribute
*o2hb_heartbeat_group_attr
=
2139 container_of(attr
, struct o2hb_heartbeat_group_attribute
, attr
);
2142 if (o2hb_heartbeat_group_attr
->show
)
2143 ret
= o2hb_heartbeat_group_attr
->show(reg
, page
);
2147 static ssize_t
o2hb_heartbeat_group_store(struct config_item
*item
,
2148 struct configfs_attribute
*attr
,
2149 const char *page
, size_t count
)
2151 struct o2hb_heartbeat_group
*reg
= to_o2hb_heartbeat_group(to_config_group(item
));
2152 struct o2hb_heartbeat_group_attribute
*o2hb_heartbeat_group_attr
=
2153 container_of(attr
, struct o2hb_heartbeat_group_attribute
, attr
);
2154 ssize_t ret
= -EINVAL
;
2156 if (o2hb_heartbeat_group_attr
->store
)
2157 ret
= o2hb_heartbeat_group_attr
->store(reg
, page
, count
);
2161 static ssize_t
o2hb_heartbeat_group_threshold_show(struct o2hb_heartbeat_group
*group
,
2164 return sprintf(page
, "%u\n", o2hb_dead_threshold
);
2167 static ssize_t
o2hb_heartbeat_group_threshold_store(struct o2hb_heartbeat_group
*group
,
2172 char *p
= (char *)page
;
2174 tmp
= simple_strtoul(p
, &p
, 10);
2175 if (!p
|| (*p
&& (*p
!= '\n')))
2178 /* this will validate ranges for us. */
2179 o2hb_dead_threshold_set((unsigned int) tmp
);
2185 ssize_t
o2hb_heartbeat_group_mode_show(struct o2hb_heartbeat_group
*group
,
2188 return sprintf(page
, "%s\n",
2189 o2hb_heartbeat_mode_desc
[o2hb_heartbeat_mode
]);
2193 ssize_t
o2hb_heartbeat_group_mode_store(struct o2hb_heartbeat_group
*group
,
2194 const char *page
, size_t count
)
2200 len
= (page
[count
- 1] == '\n') ? count
- 1 : count
;
2204 for (i
= 0; i
< O2HB_HEARTBEAT_NUM_MODES
; ++i
) {
2205 if (strnicmp(page
, o2hb_heartbeat_mode_desc
[i
], len
))
2208 ret
= o2hb_global_hearbeat_mode_set(i
);
2210 printk(KERN_NOTICE
"o2hb: Heartbeat mode set to %s\n",
2211 o2hb_heartbeat_mode_desc
[i
]);
2219 static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_threshold
= {
2220 .attr
= { .ca_owner
= THIS_MODULE
,
2221 .ca_name
= "dead_threshold",
2222 .ca_mode
= S_IRUGO
| S_IWUSR
},
2223 .show
= o2hb_heartbeat_group_threshold_show
,
2224 .store
= o2hb_heartbeat_group_threshold_store
,
2227 static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_mode
= {
2228 .attr
= { .ca_owner
= THIS_MODULE
,
2230 .ca_mode
= S_IRUGO
| S_IWUSR
},
2231 .show
= o2hb_heartbeat_group_mode_show
,
2232 .store
= o2hb_heartbeat_group_mode_store
,
2235 static struct configfs_attribute
*o2hb_heartbeat_group_attrs
[] = {
2236 &o2hb_heartbeat_group_attr_threshold
.attr
,
2237 &o2hb_heartbeat_group_attr_mode
.attr
,
2241 static struct configfs_item_operations o2hb_hearbeat_group_item_ops
= {
2242 .show_attribute
= o2hb_heartbeat_group_show
,
2243 .store_attribute
= o2hb_heartbeat_group_store
,
2246 static struct configfs_group_operations o2hb_heartbeat_group_group_ops
= {
2247 .make_item
= o2hb_heartbeat_group_make_item
,
2248 .drop_item
= o2hb_heartbeat_group_drop_item
,
2251 static struct config_item_type o2hb_heartbeat_group_type
= {
2252 .ct_group_ops
= &o2hb_heartbeat_group_group_ops
,
2253 .ct_item_ops
= &o2hb_hearbeat_group_item_ops
,
2254 .ct_attrs
= o2hb_heartbeat_group_attrs
,
2255 .ct_owner
= THIS_MODULE
,
2258 /* this is just here to avoid touching group in heartbeat.h which the
2259 * entire damn world #includes */
2260 struct config_group
*o2hb_alloc_hb_set(void)
2262 struct o2hb_heartbeat_group
*hs
= NULL
;
2263 struct config_group
*ret
= NULL
;
2265 hs
= kzalloc(sizeof(struct o2hb_heartbeat_group
), GFP_KERNEL
);
2269 config_group_init_type_name(&hs
->hs_group
, "heartbeat",
2270 &o2hb_heartbeat_group_type
);
2272 ret
= &hs
->hs_group
;
2279 void o2hb_free_hb_set(struct config_group
*group
)
2281 struct o2hb_heartbeat_group
*hs
= to_o2hb_heartbeat_group(group
);
2285 /* hb callback registration and issueing */
2287 static struct o2hb_callback
*hbcall_from_type(enum o2hb_callback_type type
)
2289 if (type
== O2HB_NUM_CB
)
2290 return ERR_PTR(-EINVAL
);
2292 return &o2hb_callbacks
[type
];
2295 void o2hb_setup_callback(struct o2hb_callback_func
*hc
,
2296 enum o2hb_callback_type type
,
2301 INIT_LIST_HEAD(&hc
->hc_item
);
2304 hc
->hc_priority
= priority
;
2306 hc
->hc_magic
= O2HB_CB_MAGIC
;
2308 EXPORT_SYMBOL_GPL(o2hb_setup_callback
);
2311 * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2312 * In global heartbeat mode, region_uuid passed is NULL.
2314 * In local, we only pin the matching region. In global we pin all the active
2317 static int o2hb_region_pin(const char *region_uuid
)
2319 int ret
= 0, found
= 0;
2320 struct o2hb_region
*reg
;
2323 assert_spin_locked(&o2hb_live_lock
);
2325 list_for_each_entry(reg
, &o2hb_all_regions
, hr_all_item
) {
2326 uuid
= config_item_name(®
->hr_item
);
2328 /* local heartbeat */
2330 if (strcmp(region_uuid
, uuid
))
2335 if (reg
->hr_item_pinned
|| reg
->hr_item_dropped
)
2338 /* Ignore ENOENT only for local hb (userdlm domain) */
2339 ret
= o2nm_depend_item(®
->hr_item
);
2341 mlog(ML_CLUSTER
, "Pin region %s\n", uuid
);
2342 reg
->hr_item_pinned
= 1;
2344 if (ret
== -ENOENT
&& found
)
2347 mlog(ML_ERROR
, "Pin region %s fails with %d\n",
2361 * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2362 * In global heartbeat mode, region_uuid passed is NULL.
2364 * In local, we only unpin the matching region. In global we unpin all the
2367 static void o2hb_region_unpin(const char *region_uuid
)
2369 struct o2hb_region
*reg
;
2373 assert_spin_locked(&o2hb_live_lock
);
2375 list_for_each_entry(reg
, &o2hb_all_regions
, hr_all_item
) {
2376 uuid
= config_item_name(®
->hr_item
);
2378 if (strcmp(region_uuid
, uuid
))
2383 if (reg
->hr_item_pinned
) {
2384 mlog(ML_CLUSTER
, "Unpin region %s\n", uuid
);
2385 o2nm_undepend_item(®
->hr_item
);
2386 reg
->hr_item_pinned
= 0;
2393 static int o2hb_region_inc_user(const char *region_uuid
)
2397 spin_lock(&o2hb_live_lock
);
2399 /* local heartbeat */
2400 if (!o2hb_global_heartbeat_active()) {
2401 ret
= o2hb_region_pin(region_uuid
);
2406 * if global heartbeat active and this is the first dependent user,
2407 * pin all regions if quorum region count <= CUT_OFF
2409 o2hb_dependent_users
++;
2410 if (o2hb_dependent_users
> 1)
2413 if (o2hb_pop_count(&o2hb_quorum_region_bitmap
,
2414 O2NM_MAX_REGIONS
) <= O2HB_PIN_CUT_OFF
)
2415 ret
= o2hb_region_pin(NULL
);
2418 spin_unlock(&o2hb_live_lock
);
2422 void o2hb_region_dec_user(const char *region_uuid
)
2424 spin_lock(&o2hb_live_lock
);
2426 /* local heartbeat */
2427 if (!o2hb_global_heartbeat_active()) {
2428 o2hb_region_unpin(region_uuid
);
2433 * if global heartbeat active and there are no dependent users,
2434 * unpin all quorum regions
2436 o2hb_dependent_users
--;
2437 if (!o2hb_dependent_users
)
2438 o2hb_region_unpin(NULL
);
2441 spin_unlock(&o2hb_live_lock
);
2444 int o2hb_register_callback(const char *region_uuid
,
2445 struct o2hb_callback_func
*hc
)
2447 struct o2hb_callback_func
*tmp
;
2448 struct list_head
*iter
;
2449 struct o2hb_callback
*hbcall
;
2452 BUG_ON(hc
->hc_magic
!= O2HB_CB_MAGIC
);
2453 BUG_ON(!list_empty(&hc
->hc_item
));
2455 hbcall
= hbcall_from_type(hc
->hc_type
);
2456 if (IS_ERR(hbcall
)) {
2457 ret
= PTR_ERR(hbcall
);
2462 ret
= o2hb_region_inc_user(region_uuid
);
2469 down_write(&o2hb_callback_sem
);
2471 list_for_each(iter
, &hbcall
->list
) {
2472 tmp
= list_entry(iter
, struct o2hb_callback_func
, hc_item
);
2473 if (hc
->hc_priority
< tmp
->hc_priority
) {
2474 list_add_tail(&hc
->hc_item
, iter
);
2478 if (list_empty(&hc
->hc_item
))
2479 list_add_tail(&hc
->hc_item
, &hbcall
->list
);
2481 up_write(&o2hb_callback_sem
);
2484 mlog(ML_CLUSTER
, "returning %d on behalf of %p for funcs %p\n",
2485 ret
, __builtin_return_address(0), hc
);
2488 EXPORT_SYMBOL_GPL(o2hb_register_callback
);
2490 void o2hb_unregister_callback(const char *region_uuid
,
2491 struct o2hb_callback_func
*hc
)
2493 BUG_ON(hc
->hc_magic
!= O2HB_CB_MAGIC
);
2495 mlog(ML_CLUSTER
, "on behalf of %p for funcs %p\n",
2496 __builtin_return_address(0), hc
);
2498 /* XXX Can this happen _with_ a region reference? */
2499 if (list_empty(&hc
->hc_item
))
2503 o2hb_region_dec_user(region_uuid
);
2505 down_write(&o2hb_callback_sem
);
2507 list_del_init(&hc
->hc_item
);
2509 up_write(&o2hb_callback_sem
);
2511 EXPORT_SYMBOL_GPL(o2hb_unregister_callback
);
2513 int o2hb_check_node_heartbeating(u8 node_num
)
2515 unsigned long testing_map
[BITS_TO_LONGS(O2NM_MAX_NODES
)];
2517 o2hb_fill_node_map(testing_map
, sizeof(testing_map
));
2518 if (!test_bit(node_num
, testing_map
)) {
2520 "node (%u) does not have heartbeating enabled.\n",
2527 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating
);
2529 int o2hb_check_node_heartbeating_from_callback(u8 node_num
)
2531 unsigned long testing_map
[BITS_TO_LONGS(O2NM_MAX_NODES
)];
2533 o2hb_fill_node_map_from_callback(testing_map
, sizeof(testing_map
));
2534 if (!test_bit(node_num
, testing_map
)) {
2536 "node (%u) does not have heartbeating enabled.\n",
2543 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback
);
2545 /* Makes sure our local node is configured with a node number, and is
2547 int o2hb_check_local_node_heartbeating(void)
2551 /* if this node was set then we have networking */
2552 node_num
= o2nm_this_node();
2553 if (node_num
== O2NM_MAX_NODES
) {
2554 mlog(ML_HEARTBEAT
, "this node has not been configured.\n");
2558 return o2hb_check_node_heartbeating(node_num
);
2560 EXPORT_SYMBOL_GPL(o2hb_check_local_node_heartbeating
);
2563 * this is just a hack until we get the plumbing which flips file systems
2564 * read only and drops the hb ref instead of killing the node dead.
2566 void o2hb_stop_all_regions(void)
2568 struct o2hb_region
*reg
;
2570 mlog(ML_ERROR
, "stopping heartbeat on all active regions.\n");
2572 spin_lock(&o2hb_live_lock
);
2574 list_for_each_entry(reg
, &o2hb_all_regions
, hr_all_item
)
2575 reg
->hr_unclean_stop
= 1;
2577 spin_unlock(&o2hb_live_lock
);
2579 EXPORT_SYMBOL_GPL(o2hb_stop_all_regions
);
2581 int o2hb_get_all_regions(char *region_uuids
, u8 max_regions
)
2583 struct o2hb_region
*reg
;
2587 spin_lock(&o2hb_live_lock
);
2590 list_for_each_entry(reg
, &o2hb_all_regions
, hr_all_item
) {
2591 mlog(0, "Region: %s\n", config_item_name(®
->hr_item
));
2592 if (numregs
< max_regions
) {
2593 memcpy(p
, config_item_name(®
->hr_item
),
2594 O2HB_MAX_REGION_NAME_LEN
);
2595 p
+= O2HB_MAX_REGION_NAME_LEN
;
2600 spin_unlock(&o2hb_live_lock
);
2604 EXPORT_SYMBOL_GPL(o2hb_get_all_regions
);
2606 int o2hb_global_heartbeat_active(void)
2608 return (o2hb_heartbeat_mode
== O2HB_HEARTBEAT_GLOBAL
);
2610 EXPORT_SYMBOL(o2hb_global_heartbeat_active
);