1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
4 * Copyright (C) 2004, 2005 Oracle. All rights reserved.
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public
8 * License as published by the Free Software Foundation; either
9 * version 2 of the License, or (at your option) any later version.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * General Public License for more details.
16 * You should have received a copy of the GNU General Public
17 * License along with this program; if not, write to the
18 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
19 * Boston, MA 021110-1307, USA.
22 #include <linux/kernel.h>
23 #include <linux/sched.h>
24 #include <linux/jiffies.h>
25 #include <linux/module.h>
27 #include <linux/bio.h>
28 #include <linux/blkdev.h>
29 #include <linux/delay.h>
30 #include <linux/file.h>
31 #include <linux/kthread.h>
32 #include <linux/configfs.h>
33 #include <linux/random.h>
34 #include <linux/crc32.h>
35 #include <linux/time.h>
36 #include <linux/debugfs.h>
37 #include <linux/slab.h>
38 #include <linux/bitmap.h>
39 #include <linux/ktime.h>
40 #include "heartbeat.h"
42 #include "nodemanager.h"
49 * The first heartbeat pass had one global thread that would serialize all hb
50 * callback calls. This global serializing sem should only be removed once
51 * we've made sure that all callees can deal with being called concurrently
52 * from multiple hb region threads.
54 static DECLARE_RWSEM(o2hb_callback_sem
);
57 * multiple hb threads are watching multiple regions. A node is live
58 * whenever any of the threads sees activity from the node in its region.
60 static DEFINE_SPINLOCK(o2hb_live_lock
);
61 static struct list_head o2hb_live_slots
[O2NM_MAX_NODES
];
62 static unsigned long o2hb_live_node_bitmap
[BITS_TO_LONGS(O2NM_MAX_NODES
)];
63 static LIST_HEAD(o2hb_node_events
);
64 static DECLARE_WAIT_QUEUE_HEAD(o2hb_steady_queue
);
67 * In global heartbeat, we maintain a series of region bitmaps.
68 * - o2hb_region_bitmap allows us to limit the region number to max region.
69 * - o2hb_live_region_bitmap tracks live regions (seen steady iterations).
70 * - o2hb_quorum_region_bitmap tracks live regions that have seen all nodes
72 * - o2hb_failed_region_bitmap tracks the regions that have seen io timeouts.
74 static unsigned long o2hb_region_bitmap
[BITS_TO_LONGS(O2NM_MAX_REGIONS
)];
75 static unsigned long o2hb_live_region_bitmap
[BITS_TO_LONGS(O2NM_MAX_REGIONS
)];
76 static unsigned long o2hb_quorum_region_bitmap
[BITS_TO_LONGS(O2NM_MAX_REGIONS
)];
77 static unsigned long o2hb_failed_region_bitmap
[BITS_TO_LONGS(O2NM_MAX_REGIONS
)];
79 #define O2HB_DB_TYPE_LIVENODES 0
80 #define O2HB_DB_TYPE_LIVEREGIONS 1
81 #define O2HB_DB_TYPE_QUORUMREGIONS 2
82 #define O2HB_DB_TYPE_FAILEDREGIONS 3
83 #define O2HB_DB_TYPE_REGION_LIVENODES 4
84 #define O2HB_DB_TYPE_REGION_NUMBER 5
85 #define O2HB_DB_TYPE_REGION_ELAPSED_TIME 6
86 #define O2HB_DB_TYPE_REGION_PINNED 7
87 struct o2hb_debug_buf
{
94 static struct o2hb_debug_buf
*o2hb_db_livenodes
;
95 static struct o2hb_debug_buf
*o2hb_db_liveregions
;
96 static struct o2hb_debug_buf
*o2hb_db_quorumregions
;
97 static struct o2hb_debug_buf
*o2hb_db_failedregions
;
99 #define O2HB_DEBUG_DIR "o2hb"
100 #define O2HB_DEBUG_LIVENODES "livenodes"
101 #define O2HB_DEBUG_LIVEREGIONS "live_regions"
102 #define O2HB_DEBUG_QUORUMREGIONS "quorum_regions"
103 #define O2HB_DEBUG_FAILEDREGIONS "failed_regions"
104 #define O2HB_DEBUG_REGION_NUMBER "num"
105 #define O2HB_DEBUG_REGION_ELAPSED_TIME "elapsed_time_in_ms"
106 #define O2HB_DEBUG_REGION_PINNED "pinned"
108 static struct dentry
*o2hb_debug_dir
;
109 static struct dentry
*o2hb_debug_livenodes
;
110 static struct dentry
*o2hb_debug_liveregions
;
111 static struct dentry
*o2hb_debug_quorumregions
;
112 static struct dentry
*o2hb_debug_failedregions
;
114 static LIST_HEAD(o2hb_all_regions
);
116 static struct o2hb_callback
{
117 struct list_head list
;
118 } o2hb_callbacks
[O2HB_NUM_CB
];
120 static struct o2hb_callback
*hbcall_from_type(enum o2hb_callback_type type
);
122 #define O2HB_DEFAULT_BLOCK_BITS 9
124 enum o2hb_heartbeat_modes
{
125 O2HB_HEARTBEAT_LOCAL
= 0,
126 O2HB_HEARTBEAT_GLOBAL
,
127 O2HB_HEARTBEAT_NUM_MODES
,
130 char *o2hb_heartbeat_mode_desc
[O2HB_HEARTBEAT_NUM_MODES
] = {
131 "local", /* O2HB_HEARTBEAT_LOCAL */
132 "global", /* O2HB_HEARTBEAT_GLOBAL */
135 unsigned int o2hb_dead_threshold
= O2HB_DEFAULT_DEAD_THRESHOLD
;
136 unsigned int o2hb_heartbeat_mode
= O2HB_HEARTBEAT_LOCAL
;
139 * o2hb_dependent_users tracks the number of registered callbacks that depend
140 * on heartbeat. o2net and o2dlm are two entities that register this callback.
141 * However only o2dlm depends on the heartbeat. It does not want the heartbeat
142 * to stop while a dlm domain is still active.
144 unsigned int o2hb_dependent_users
;
147 * In global heartbeat mode, all regions are pinned if there are one or more
148 * dependent users and the quorum region count is <= O2HB_PIN_CUT_OFF. All
149 * regions are unpinned if the region count exceeds the cut off or the number
150 * of dependent users falls to zero.
152 #define O2HB_PIN_CUT_OFF 3
155 * In local heartbeat mode, we assume the dlm domain name to be the same as
156 * region uuid. This is true for domains created for the file system but not
157 * necessarily true for userdlm domains. This is a known limitation.
159 * In global heartbeat mode, we pin/unpin all o2hb regions. This solution
160 * works for both file system and userdlm domains.
162 static int o2hb_region_pin(const char *region_uuid
);
163 static void o2hb_region_unpin(const char *region_uuid
);
165 /* Only sets a new threshold if there are no active regions.
167 * No locking or otherwise interesting code is required for reading
168 * o2hb_dead_threshold as it can't change once regions are active and
169 * it's not interesting to anyone until then anyway. */
170 static void o2hb_dead_threshold_set(unsigned int threshold
)
172 if (threshold
> O2HB_MIN_DEAD_THRESHOLD
) {
173 spin_lock(&o2hb_live_lock
);
174 if (list_empty(&o2hb_all_regions
))
175 o2hb_dead_threshold
= threshold
;
176 spin_unlock(&o2hb_live_lock
);
180 static int o2hb_global_heartbeat_mode_set(unsigned int hb_mode
)
184 if (hb_mode
< O2HB_HEARTBEAT_NUM_MODES
) {
185 spin_lock(&o2hb_live_lock
);
186 if (list_empty(&o2hb_all_regions
)) {
187 o2hb_heartbeat_mode
= hb_mode
;
190 spin_unlock(&o2hb_live_lock
);
196 struct o2hb_node_event
{
197 struct list_head hn_item
;
198 enum o2hb_callback_type hn_event_type
;
199 struct o2nm_node
*hn_node
;
203 struct o2hb_disk_slot
{
204 struct o2hb_disk_heartbeat_block
*ds_raw_block
;
207 u64 ds_last_generation
;
208 u16 ds_equal_samples
;
209 u16 ds_changed_samples
;
210 struct list_head ds_live_item
;
213 /* each thread owns a region.. when we're asked to tear down the region
214 * we ask the thread to stop, who cleans up the region */
216 struct config_item hr_item
;
218 struct list_head hr_all_item
;
219 unsigned hr_unclean_stop
:1,
225 /* protected by the hr_callback_sem */
226 struct task_struct
*hr_task
;
228 unsigned int hr_blocks
;
229 unsigned long long hr_start_block
;
231 unsigned int hr_block_bits
;
232 unsigned int hr_block_bytes
;
234 unsigned int hr_slots_per_page
;
235 unsigned int hr_num_pages
;
237 struct page
**hr_slot_data
;
238 struct block_device
*hr_bdev
;
239 struct o2hb_disk_slot
*hr_slots
;
241 /* live node map of this region */
242 unsigned long hr_live_node_bitmap
[BITS_TO_LONGS(O2NM_MAX_NODES
)];
243 unsigned int hr_region_num
;
245 struct dentry
*hr_debug_dir
;
246 struct dentry
*hr_debug_livenodes
;
247 struct dentry
*hr_debug_regnum
;
248 struct dentry
*hr_debug_elapsed_time
;
249 struct dentry
*hr_debug_pinned
;
250 struct o2hb_debug_buf
*hr_db_livenodes
;
251 struct o2hb_debug_buf
*hr_db_regnum
;
252 struct o2hb_debug_buf
*hr_db_elapsed_time
;
253 struct o2hb_debug_buf
*hr_db_pinned
;
255 /* let the person setting up hb wait for it to return until it
256 * has reached a 'steady' state. This will be fixed when we have
257 * a more complete api that doesn't lead to this sort of fragility. */
258 atomic_t hr_steady_iterations
;
260 /* terminate o2hb thread if it does not reach steady state
261 * (hr_steady_iterations == 0) within hr_unsteady_iterations */
262 atomic_t hr_unsteady_iterations
;
264 char hr_dev_name
[BDEVNAME_SIZE
];
266 unsigned int hr_timeout_ms
;
268 /* randomized as the region goes up and down so that a node
269 * recognizes a node going up and down in one iteration */
272 struct delayed_work hr_write_timeout_work
;
273 unsigned long hr_last_timeout_start
;
275 /* negotiate timer, used to negotiate extending hb timeout. */
276 struct delayed_work hr_nego_timeout_work
;
277 unsigned long hr_nego_node_bitmap
[BITS_TO_LONGS(O2NM_MAX_NODES
)];
279 /* Used during o2hb_check_slot to hold a copy of the block
280 * being checked because we temporarily have to zero out the
282 struct o2hb_disk_heartbeat_block
*hr_tmp_block
;
284 /* Message key for negotiate timeout message. */
286 struct list_head hr_handler_list
;
288 /* last hb status, 0 for success, other value for error. */
289 int hr_last_hb_status
;
292 struct o2hb_bio_wait_ctxt
{
293 atomic_t wc_num_reqs
;
294 struct completion wc_io_complete
;
298 #define O2HB_NEGO_TIMEOUT_MS (O2HB_MAX_WRITE_TIMEOUT_MS/2)
301 O2HB_NEGO_TIMEOUT_MSG
= 1,
302 O2HB_NEGO_APPROVE_MSG
= 2,
305 struct o2hb_nego_msg
{
309 static void o2hb_write_timeout(struct work_struct
*work
)
312 struct o2hb_region
*reg
=
313 container_of(work
, struct o2hb_region
,
314 hr_write_timeout_work
.work
);
316 mlog(ML_ERROR
, "Heartbeat write timeout to device %s after %u "
317 "milliseconds\n", reg
->hr_dev_name
,
318 jiffies_to_msecs(jiffies
- reg
->hr_last_timeout_start
));
320 if (o2hb_global_heartbeat_active()) {
321 spin_lock(&o2hb_live_lock
);
322 if (test_bit(reg
->hr_region_num
, o2hb_quorum_region_bitmap
))
323 set_bit(reg
->hr_region_num
, o2hb_failed_region_bitmap
);
324 failed
= bitmap_weight(o2hb_failed_region_bitmap
,
326 quorum
= bitmap_weight(o2hb_quorum_region_bitmap
,
328 spin_unlock(&o2hb_live_lock
);
330 mlog(ML_HEARTBEAT
, "Number of regions %d, failed regions %d\n",
334 * Fence if the number of failed regions >= half the number
337 if ((failed
<< 1) < quorum
)
341 o2quo_disk_timeout();
344 static void o2hb_arm_timeout(struct o2hb_region
*reg
)
346 /* Arm writeout only after thread reaches steady state */
347 if (atomic_read(®
->hr_steady_iterations
) != 0)
350 mlog(ML_HEARTBEAT
, "Queue write timeout for %u ms\n",
351 O2HB_MAX_WRITE_TIMEOUT_MS
);
353 if (o2hb_global_heartbeat_active()) {
354 spin_lock(&o2hb_live_lock
);
355 clear_bit(reg
->hr_region_num
, o2hb_failed_region_bitmap
);
356 spin_unlock(&o2hb_live_lock
);
358 cancel_delayed_work(®
->hr_write_timeout_work
);
359 schedule_delayed_work(®
->hr_write_timeout_work
,
360 msecs_to_jiffies(O2HB_MAX_WRITE_TIMEOUT_MS
));
362 cancel_delayed_work(®
->hr_nego_timeout_work
);
363 /* negotiate timeout must be less than write timeout. */
364 schedule_delayed_work(®
->hr_nego_timeout_work
,
365 msecs_to_jiffies(O2HB_NEGO_TIMEOUT_MS
));
366 memset(reg
->hr_nego_node_bitmap
, 0, sizeof(reg
->hr_nego_node_bitmap
));
369 static void o2hb_disarm_timeout(struct o2hb_region
*reg
)
371 cancel_delayed_work_sync(®
->hr_write_timeout_work
);
372 cancel_delayed_work_sync(®
->hr_nego_timeout_work
);
375 static int o2hb_send_nego_msg(int key
, int type
, u8 target
)
377 struct o2hb_nego_msg msg
;
380 msg
.node_num
= o2nm_this_node();
382 ret
= o2net_send_message(type
, key
, &msg
, sizeof(msg
),
385 if (ret
== -EAGAIN
|| ret
== -ENOMEM
) {
393 static void o2hb_nego_timeout(struct work_struct
*work
)
395 unsigned long live_node_bitmap
[BITS_TO_LONGS(O2NM_MAX_NODES
)];
396 int master_node
, i
, ret
;
397 struct o2hb_region
*reg
;
399 reg
= container_of(work
, struct o2hb_region
, hr_nego_timeout_work
.work
);
400 /* don't negotiate timeout if last hb failed since it is very
401 * possible io failed. Should let write timeout fence self.
403 if (reg
->hr_last_hb_status
)
406 o2hb_fill_node_map(live_node_bitmap
, sizeof(live_node_bitmap
));
407 /* lowest node as master node to make negotiate decision. */
408 master_node
= find_next_bit(live_node_bitmap
, O2NM_MAX_NODES
, 0);
410 if (master_node
== o2nm_this_node()) {
411 if (!test_bit(master_node
, reg
->hr_nego_node_bitmap
)) {
412 printk(KERN_NOTICE
"o2hb: node %d hb write hung for %ds on region %s (%s).\n",
413 o2nm_this_node(), O2HB_NEGO_TIMEOUT_MS
/1000,
414 config_item_name(®
->hr_item
), reg
->hr_dev_name
);
415 set_bit(master_node
, reg
->hr_nego_node_bitmap
);
417 if (memcmp(reg
->hr_nego_node_bitmap
, live_node_bitmap
,
418 sizeof(reg
->hr_nego_node_bitmap
))) {
419 /* check negotiate bitmap every second to do timeout
422 schedule_delayed_work(®
->hr_nego_timeout_work
,
423 msecs_to_jiffies(1000));
428 printk(KERN_NOTICE
"o2hb: all nodes hb write hung, maybe region %s (%s) is down.\n",
429 config_item_name(®
->hr_item
), reg
->hr_dev_name
);
430 /* approve negotiate timeout request. */
431 o2hb_arm_timeout(reg
);
434 while ((i
= find_next_bit(live_node_bitmap
,
435 O2NM_MAX_NODES
, i
+ 1)) < O2NM_MAX_NODES
) {
436 if (i
== master_node
)
439 mlog(ML_HEARTBEAT
, "send NEGO_APPROVE msg to node %d\n", i
);
440 ret
= o2hb_send_nego_msg(reg
->hr_key
,
441 O2HB_NEGO_APPROVE_MSG
, i
);
443 mlog(ML_ERROR
, "send NEGO_APPROVE msg to node %d fail %d\n",
447 /* negotiate timeout with master node. */
448 printk(KERN_NOTICE
"o2hb: node %d hb write hung for %ds on region %s (%s), negotiate timeout with node %d.\n",
449 o2nm_this_node(), O2HB_NEGO_TIMEOUT_MS
/1000, config_item_name(®
->hr_item
),
450 reg
->hr_dev_name
, master_node
);
451 ret
= o2hb_send_nego_msg(reg
->hr_key
, O2HB_NEGO_TIMEOUT_MSG
,
454 mlog(ML_ERROR
, "send NEGO_TIMEOUT msg to node %d fail %d\n",
459 static int o2hb_nego_timeout_handler(struct o2net_msg
*msg
, u32 len
, void *data
,
462 struct o2hb_region
*reg
= data
;
463 struct o2hb_nego_msg
*nego_msg
;
465 nego_msg
= (struct o2hb_nego_msg
*)msg
->buf
;
466 printk(KERN_NOTICE
"o2hb: receive negotiate timeout message from node %d on region %s (%s).\n",
467 nego_msg
->node_num
, config_item_name(®
->hr_item
), reg
->hr_dev_name
);
468 if (nego_msg
->node_num
< O2NM_MAX_NODES
)
469 set_bit(nego_msg
->node_num
, reg
->hr_nego_node_bitmap
);
471 mlog(ML_ERROR
, "got nego timeout message from bad node.\n");
476 static int o2hb_nego_approve_handler(struct o2net_msg
*msg
, u32 len
, void *data
,
479 struct o2hb_region
*reg
= data
;
481 printk(KERN_NOTICE
"o2hb: negotiate timeout approved by master node on region %s (%s).\n",
482 config_item_name(®
->hr_item
), reg
->hr_dev_name
);
483 o2hb_arm_timeout(reg
);
487 static inline void o2hb_bio_wait_init(struct o2hb_bio_wait_ctxt
*wc
)
489 atomic_set(&wc
->wc_num_reqs
, 1);
490 init_completion(&wc
->wc_io_complete
);
494 /* Used in error paths too */
495 static inline void o2hb_bio_wait_dec(struct o2hb_bio_wait_ctxt
*wc
,
498 /* sadly atomic_sub_and_test() isn't available on all platforms. The
499 * good news is that the fast path only completes one at a time */
501 if (atomic_dec_and_test(&wc
->wc_num_reqs
)) {
503 complete(&wc
->wc_io_complete
);
508 static void o2hb_wait_on_io(struct o2hb_region
*reg
,
509 struct o2hb_bio_wait_ctxt
*wc
)
511 o2hb_bio_wait_dec(wc
, 1);
512 wait_for_completion(&wc
->wc_io_complete
);
515 static void o2hb_bio_end_io(struct bio
*bio
)
517 struct o2hb_bio_wait_ctxt
*wc
= bio
->bi_private
;
520 mlog(ML_ERROR
, "IO Error %d\n", bio
->bi_error
);
521 wc
->wc_error
= bio
->bi_error
;
524 o2hb_bio_wait_dec(wc
, 1);
528 /* Setup a Bio to cover I/O against num_slots slots starting at
530 static struct bio
*o2hb_setup_one_bio(struct o2hb_region
*reg
,
531 struct o2hb_bio_wait_ctxt
*wc
,
532 unsigned int *current_slot
,
533 unsigned int max_slots
, int rw
)
535 int len
, current_page
;
536 unsigned int vec_len
, vec_start
;
537 unsigned int bits
= reg
->hr_block_bits
;
538 unsigned int spp
= reg
->hr_slots_per_page
;
539 unsigned int cs
= *current_slot
;
543 /* Testing has shown this allocation to take long enough under
544 * GFP_KERNEL that the local node can get fenced. It would be
545 * nicest if we could pre-allocate these bios and avoid this
547 bio
= bio_alloc(GFP_ATOMIC
, 16);
549 mlog(ML_ERROR
, "Could not alloc slots BIO!\n");
550 bio
= ERR_PTR(-ENOMEM
);
554 /* Must put everything in 512 byte sectors for the bio... */
555 bio
->bi_iter
.bi_sector
= (reg
->hr_start_block
+ cs
) << (bits
- 9);
556 bio
->bi_bdev
= reg
->hr_bdev
;
557 bio
->bi_private
= wc
;
558 bio
->bi_end_io
= o2hb_bio_end_io
;
561 vec_start
= (cs
<< bits
) % PAGE_SIZE
;
562 while(cs
< max_slots
) {
563 current_page
= cs
/ spp
;
564 page
= reg
->hr_slot_data
[current_page
];
566 vec_len
= min(PAGE_SIZE
- vec_start
,
567 (max_slots
-cs
) * (PAGE_SIZE
/spp
) );
569 mlog(ML_HB_BIO
, "page %d, vec_len = %u, vec_start = %u\n",
570 current_page
, vec_len
, vec_start
);
572 len
= bio_add_page(bio
, page
, vec_len
, vec_start
);
573 if (len
!= vec_len
) break;
575 cs
+= vec_len
/ (PAGE_SIZE
/spp
);
584 static int o2hb_read_slots(struct o2hb_region
*reg
,
585 unsigned int max_slots
)
587 unsigned int current_slot
=0;
589 struct o2hb_bio_wait_ctxt wc
;
592 o2hb_bio_wait_init(&wc
);
594 while(current_slot
< max_slots
) {
595 bio
= o2hb_setup_one_bio(reg
, &wc
, ¤t_slot
, max_slots
,
598 status
= PTR_ERR(bio
);
603 atomic_inc(&wc
.wc_num_reqs
);
610 o2hb_wait_on_io(reg
, &wc
);
611 if (wc
.wc_error
&& !status
)
612 status
= wc
.wc_error
;
617 static int o2hb_issue_node_write(struct o2hb_region
*reg
,
618 struct o2hb_bio_wait_ctxt
*write_wc
)
624 o2hb_bio_wait_init(write_wc
);
626 slot
= o2nm_this_node();
628 bio
= o2hb_setup_one_bio(reg
, write_wc
, &slot
, slot
+1, WRITE_SYNC
);
630 status
= PTR_ERR(bio
);
635 atomic_inc(&write_wc
->wc_num_reqs
);
643 static u32
o2hb_compute_block_crc_le(struct o2hb_region
*reg
,
644 struct o2hb_disk_heartbeat_block
*hb_block
)
649 /* We want to compute the block crc with a 0 value in the
650 * hb_cksum field. Save it off here and replace after the
652 old_cksum
= hb_block
->hb_cksum
;
653 hb_block
->hb_cksum
= 0;
655 ret
= crc32_le(0, (unsigned char *) hb_block
, reg
->hr_block_bytes
);
657 hb_block
->hb_cksum
= old_cksum
;
662 static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block
*hb_block
)
664 mlog(ML_ERROR
, "Dump slot information: seq = 0x%llx, node = %u, "
665 "cksum = 0x%x, generation 0x%llx\n",
666 (long long)le64_to_cpu(hb_block
->hb_seq
),
667 hb_block
->hb_node
, le32_to_cpu(hb_block
->hb_cksum
),
668 (long long)le64_to_cpu(hb_block
->hb_generation
));
671 static int o2hb_verify_crc(struct o2hb_region
*reg
,
672 struct o2hb_disk_heartbeat_block
*hb_block
)
676 read
= le32_to_cpu(hb_block
->hb_cksum
);
677 computed
= o2hb_compute_block_crc_le(reg
, hb_block
);
679 return read
== computed
;
683 * Compare the slot data with what we wrote in the last iteration.
684 * If the match fails, print an appropriate error message. This is to
685 * detect errors like... another node hearting on the same slot,
686 * flaky device that is losing writes, etc.
687 * Returns 1 if check succeeds, 0 otherwise.
689 static int o2hb_check_own_slot(struct o2hb_region
*reg
)
691 struct o2hb_disk_slot
*slot
;
692 struct o2hb_disk_heartbeat_block
*hb_block
;
695 slot
= ®
->hr_slots
[o2nm_this_node()];
696 /* Don't check on our 1st timestamp */
697 if (!slot
->ds_last_time
)
700 hb_block
= slot
->ds_raw_block
;
701 if (le64_to_cpu(hb_block
->hb_seq
) == slot
->ds_last_time
&&
702 le64_to_cpu(hb_block
->hb_generation
) == slot
->ds_last_generation
&&
703 hb_block
->hb_node
== slot
->ds_node_num
)
706 #define ERRSTR1 "Another node is heartbeating on device"
707 #define ERRSTR2 "Heartbeat generation mismatch on device"
708 #define ERRSTR3 "Heartbeat sequence mismatch on device"
710 if (hb_block
->hb_node
!= slot
->ds_node_num
)
712 else if (le64_to_cpu(hb_block
->hb_generation
) !=
713 slot
->ds_last_generation
)
718 mlog(ML_ERROR
, "%s (%s): expected(%u:0x%llx, 0x%llx), "
719 "ondisk(%u:0x%llx, 0x%llx)\n", errstr
, reg
->hr_dev_name
,
720 slot
->ds_node_num
, (unsigned long long)slot
->ds_last_generation
,
721 (unsigned long long)slot
->ds_last_time
, hb_block
->hb_node
,
722 (unsigned long long)le64_to_cpu(hb_block
->hb_generation
),
723 (unsigned long long)le64_to_cpu(hb_block
->hb_seq
));
728 static inline void o2hb_prepare_block(struct o2hb_region
*reg
,
733 struct o2hb_disk_slot
*slot
;
734 struct o2hb_disk_heartbeat_block
*hb_block
;
736 node_num
= o2nm_this_node();
737 slot
= ®
->hr_slots
[node_num
];
739 hb_block
= (struct o2hb_disk_heartbeat_block
*)slot
->ds_raw_block
;
740 memset(hb_block
, 0, reg
->hr_block_bytes
);
741 /* TODO: time stuff */
742 cputime
= CURRENT_TIME
.tv_sec
;
746 hb_block
->hb_seq
= cpu_to_le64(cputime
);
747 hb_block
->hb_node
= node_num
;
748 hb_block
->hb_generation
= cpu_to_le64(generation
);
749 hb_block
->hb_dead_ms
= cpu_to_le32(o2hb_dead_threshold
* O2HB_REGION_TIMEOUT_MS
);
751 /* This step must always happen last! */
752 hb_block
->hb_cksum
= cpu_to_le32(o2hb_compute_block_crc_le(reg
,
755 mlog(ML_HB_BIO
, "our node generation = 0x%llx, cksum = 0x%x\n",
756 (long long)generation
,
757 le32_to_cpu(hb_block
->hb_cksum
));
760 static void o2hb_fire_callbacks(struct o2hb_callback
*hbcall
,
761 struct o2nm_node
*node
,
764 struct o2hb_callback_func
*f
;
766 list_for_each_entry(f
, &hbcall
->list
, hc_item
) {
767 mlog(ML_HEARTBEAT
, "calling funcs %p\n", f
);
768 (f
->hc_func
)(node
, idx
, f
->hc_data
);
772 /* Will run the list in order until we process the passed event */
773 static void o2hb_run_event_list(struct o2hb_node_event
*queued_event
)
775 struct o2hb_callback
*hbcall
;
776 struct o2hb_node_event
*event
;
778 /* Holding callback sem assures we don't alter the callback
779 * lists when doing this, and serializes ourselves with other
780 * processes wanting callbacks. */
781 down_write(&o2hb_callback_sem
);
783 spin_lock(&o2hb_live_lock
);
784 while (!list_empty(&o2hb_node_events
)
785 && !list_empty(&queued_event
->hn_item
)) {
786 event
= list_entry(o2hb_node_events
.next
,
787 struct o2hb_node_event
,
789 list_del_init(&event
->hn_item
);
790 spin_unlock(&o2hb_live_lock
);
792 mlog(ML_HEARTBEAT
, "Node %s event for %d\n",
793 event
->hn_event_type
== O2HB_NODE_UP_CB
? "UP" : "DOWN",
796 hbcall
= hbcall_from_type(event
->hn_event_type
);
798 /* We should *never* have gotten on to the list with a
799 * bad type... This isn't something that we should try
800 * to recover from. */
801 BUG_ON(IS_ERR(hbcall
));
803 o2hb_fire_callbacks(hbcall
, event
->hn_node
, event
->hn_node_num
);
805 spin_lock(&o2hb_live_lock
);
807 spin_unlock(&o2hb_live_lock
);
809 up_write(&o2hb_callback_sem
);
812 static void o2hb_queue_node_event(struct o2hb_node_event
*event
,
813 enum o2hb_callback_type type
,
814 struct o2nm_node
*node
,
817 assert_spin_locked(&o2hb_live_lock
);
819 BUG_ON((!node
) && (type
!= O2HB_NODE_DOWN_CB
));
821 event
->hn_event_type
= type
;
822 event
->hn_node
= node
;
823 event
->hn_node_num
= node_num
;
825 mlog(ML_HEARTBEAT
, "Queue node %s event for node %d\n",
826 type
== O2HB_NODE_UP_CB
? "UP" : "DOWN", node_num
);
828 list_add_tail(&event
->hn_item
, &o2hb_node_events
);
831 static void o2hb_shutdown_slot(struct o2hb_disk_slot
*slot
)
833 struct o2hb_node_event event
=
834 { .hn_item
= LIST_HEAD_INIT(event
.hn_item
), };
835 struct o2nm_node
*node
;
838 node
= o2nm_get_node_by_num(slot
->ds_node_num
);
842 spin_lock(&o2hb_live_lock
);
843 if (!list_empty(&slot
->ds_live_item
)) {
844 mlog(ML_HEARTBEAT
, "Shutdown, node %d leaves region\n",
847 list_del_init(&slot
->ds_live_item
);
849 if (list_empty(&o2hb_live_slots
[slot
->ds_node_num
])) {
850 clear_bit(slot
->ds_node_num
, o2hb_live_node_bitmap
);
852 o2hb_queue_node_event(&event
, O2HB_NODE_DOWN_CB
, node
,
857 spin_unlock(&o2hb_live_lock
);
860 o2hb_run_event_list(&event
);
865 static void o2hb_set_quorum_device(struct o2hb_region
*reg
)
867 if (!o2hb_global_heartbeat_active())
870 /* Prevent race with o2hb_heartbeat_group_drop_item() */
871 if (kthread_should_stop())
874 /* Tag region as quorum only after thread reaches steady state */
875 if (atomic_read(®
->hr_steady_iterations
) != 0)
878 spin_lock(&o2hb_live_lock
);
880 if (test_bit(reg
->hr_region_num
, o2hb_quorum_region_bitmap
))
884 * A region can be added to the quorum only when it sees all
885 * live nodes heartbeat on it. In other words, the region has been
886 * added to all nodes.
888 if (memcmp(reg
->hr_live_node_bitmap
, o2hb_live_node_bitmap
,
889 sizeof(o2hb_live_node_bitmap
)))
892 printk(KERN_NOTICE
"o2hb: Region %s (%s) is now a quorum device\n",
893 config_item_name(®
->hr_item
), reg
->hr_dev_name
);
895 set_bit(reg
->hr_region_num
, o2hb_quorum_region_bitmap
);
898 * If global heartbeat active, unpin all regions if the
899 * region count > CUT_OFF
901 if (bitmap_weight(o2hb_quorum_region_bitmap
,
902 O2NM_MAX_REGIONS
) > O2HB_PIN_CUT_OFF
)
903 o2hb_region_unpin(NULL
);
905 spin_unlock(&o2hb_live_lock
);
908 static int o2hb_check_slot(struct o2hb_region
*reg
,
909 struct o2hb_disk_slot
*slot
)
911 int changed
= 0, gen_changed
= 0;
912 struct o2hb_node_event event
=
913 { .hn_item
= LIST_HEAD_INIT(event
.hn_item
), };
914 struct o2nm_node
*node
;
915 struct o2hb_disk_heartbeat_block
*hb_block
= reg
->hr_tmp_block
;
917 unsigned int dead_ms
= o2hb_dead_threshold
* O2HB_REGION_TIMEOUT_MS
;
918 unsigned int slot_dead_ms
;
922 memcpy(hb_block
, slot
->ds_raw_block
, reg
->hr_block_bytes
);
925 * If a node is no longer configured but is still in the livemap, we
926 * may need to clear that bit from the livemap.
928 node
= o2nm_get_node_by_num(slot
->ds_node_num
);
930 spin_lock(&o2hb_live_lock
);
931 tmp
= test_bit(slot
->ds_node_num
, o2hb_live_node_bitmap
);
932 spin_unlock(&o2hb_live_lock
);
937 if (!o2hb_verify_crc(reg
, hb_block
)) {
938 /* all paths from here will drop o2hb_live_lock for
940 spin_lock(&o2hb_live_lock
);
942 /* Don't print an error on the console in this case -
943 * a freshly formatted heartbeat area will not have a
945 if (list_empty(&slot
->ds_live_item
))
948 /* The node is live but pushed out a bad crc. We
949 * consider it a transient miss but don't populate any
950 * other values as they may be junk. */
951 mlog(ML_ERROR
, "Node %d has written a bad crc to %s\n",
952 slot
->ds_node_num
, reg
->hr_dev_name
);
953 o2hb_dump_slot(hb_block
);
955 slot
->ds_equal_samples
++;
959 /* we don't care if these wrap.. the state transitions below
960 * clear at the right places */
961 cputime
= le64_to_cpu(hb_block
->hb_seq
);
962 if (slot
->ds_last_time
!= cputime
)
963 slot
->ds_changed_samples
++;
965 slot
->ds_equal_samples
++;
966 slot
->ds_last_time
= cputime
;
968 /* The node changed heartbeat generations. We assume this to
969 * mean it dropped off but came back before we timed out. We
970 * want to consider it down for the time being but don't want
971 * to lose any changed_samples state we might build up to
972 * considering it live again. */
973 if (slot
->ds_last_generation
!= le64_to_cpu(hb_block
->hb_generation
)) {
975 slot
->ds_equal_samples
= 0;
976 mlog(ML_HEARTBEAT
, "Node %d changed generation (0x%llx "
977 "to 0x%llx)\n", slot
->ds_node_num
,
978 (long long)slot
->ds_last_generation
,
979 (long long)le64_to_cpu(hb_block
->hb_generation
));
982 slot
->ds_last_generation
= le64_to_cpu(hb_block
->hb_generation
);
984 mlog(ML_HEARTBEAT
, "Slot %d gen 0x%llx cksum 0x%x "
985 "seq %llu last %llu changed %u equal %u\n",
986 slot
->ds_node_num
, (long long)slot
->ds_last_generation
,
987 le32_to_cpu(hb_block
->hb_cksum
),
988 (unsigned long long)le64_to_cpu(hb_block
->hb_seq
),
989 (unsigned long long)slot
->ds_last_time
, slot
->ds_changed_samples
,
990 slot
->ds_equal_samples
);
992 spin_lock(&o2hb_live_lock
);
995 /* dead nodes only come to life after some number of
996 * changes at any time during their dead time */
997 if (list_empty(&slot
->ds_live_item
) &&
998 slot
->ds_changed_samples
>= O2HB_LIVE_THRESHOLD
) {
999 mlog(ML_HEARTBEAT
, "Node %d (id 0x%llx) joined my region\n",
1000 slot
->ds_node_num
, (long long)slot
->ds_last_generation
);
1002 set_bit(slot
->ds_node_num
, reg
->hr_live_node_bitmap
);
1004 /* first on the list generates a callback */
1005 if (list_empty(&o2hb_live_slots
[slot
->ds_node_num
])) {
1006 mlog(ML_HEARTBEAT
, "o2hb: Add node %d to live nodes "
1007 "bitmap\n", slot
->ds_node_num
);
1008 set_bit(slot
->ds_node_num
, o2hb_live_node_bitmap
);
1010 o2hb_queue_node_event(&event
, O2HB_NODE_UP_CB
, node
,
1017 list_add_tail(&slot
->ds_live_item
,
1018 &o2hb_live_slots
[slot
->ds_node_num
]);
1020 slot
->ds_equal_samples
= 0;
1022 /* We want to be sure that all nodes agree on the
1023 * number of milliseconds before a node will be
1024 * considered dead. The self-fencing timeout is
1025 * computed from this value, and a discrepancy might
1026 * result in heartbeat calling a node dead when it
1027 * hasn't self-fenced yet. */
1028 slot_dead_ms
= le32_to_cpu(hb_block
->hb_dead_ms
);
1029 if (slot_dead_ms
&& slot_dead_ms
!= dead_ms
) {
1030 /* TODO: Perhaps we can fail the region here. */
1031 mlog(ML_ERROR
, "Node %d on device %s has a dead count "
1032 "of %u ms, but our count is %u ms.\n"
1033 "Please double check your configuration values "
1034 "for 'O2CB_HEARTBEAT_THRESHOLD'\n",
1035 slot
->ds_node_num
, reg
->hr_dev_name
, slot_dead_ms
,
1041 /* if the list is dead, we're done.. */
1042 if (list_empty(&slot
->ds_live_item
))
1045 /* live nodes only go dead after enough consequtive missed
1046 * samples.. reset the missed counter whenever we see
1048 if (slot
->ds_equal_samples
>= o2hb_dead_threshold
|| gen_changed
) {
1049 mlog(ML_HEARTBEAT
, "Node %d left my region\n",
1052 clear_bit(slot
->ds_node_num
, reg
->hr_live_node_bitmap
);
1054 /* last off the live_slot generates a callback */
1055 list_del_init(&slot
->ds_live_item
);
1056 if (list_empty(&o2hb_live_slots
[slot
->ds_node_num
])) {
1057 mlog(ML_HEARTBEAT
, "o2hb: Remove node %d from live "
1058 "nodes bitmap\n", slot
->ds_node_num
);
1059 clear_bit(slot
->ds_node_num
, o2hb_live_node_bitmap
);
1061 /* node can be null */
1062 o2hb_queue_node_event(&event
, O2HB_NODE_DOWN_CB
,
1063 node
, slot
->ds_node_num
);
1069 /* We don't clear this because the node is still
1070 * actually writing new blocks. */
1072 slot
->ds_changed_samples
= 0;
1075 if (slot
->ds_changed_samples
) {
1076 slot
->ds_changed_samples
= 0;
1077 slot
->ds_equal_samples
= 0;
1080 spin_unlock(&o2hb_live_lock
);
1083 o2hb_run_event_list(&event
);
1086 o2nm_node_put(node
);
1090 static int o2hb_highest_node(unsigned long *nodes
, int numbits
)
1092 return find_last_bit(nodes
, numbits
);
1095 static int o2hb_do_disk_heartbeat(struct o2hb_region
*reg
)
1097 int i
, ret
, highest_node
;
1098 int membership_change
= 0, own_slot_ok
= 0;
1099 unsigned long configured_nodes
[BITS_TO_LONGS(O2NM_MAX_NODES
)];
1100 unsigned long live_node_bitmap
[BITS_TO_LONGS(O2NM_MAX_NODES
)];
1101 struct o2hb_bio_wait_ctxt write_wc
;
1103 ret
= o2nm_configured_node_map(configured_nodes
,
1104 sizeof(configured_nodes
));
1111 * If a node is not configured but is in the livemap, we still need
1112 * to read the slot so as to be able to remove it from the livemap.
1114 o2hb_fill_node_map(live_node_bitmap
, sizeof(live_node_bitmap
));
1116 while ((i
= find_next_bit(live_node_bitmap
,
1117 O2NM_MAX_NODES
, i
+ 1)) < O2NM_MAX_NODES
) {
1118 set_bit(i
, configured_nodes
);
1121 highest_node
= o2hb_highest_node(configured_nodes
, O2NM_MAX_NODES
);
1122 if (highest_node
>= O2NM_MAX_NODES
) {
1123 mlog(ML_NOTICE
, "o2hb: No configured nodes found!\n");
1128 /* No sense in reading the slots of nodes that don't exist
1129 * yet. Of course, if the node definitions have holes in them
1130 * then we're reading an empty slot anyway... Consider this
1132 ret
= o2hb_read_slots(reg
, highest_node
+ 1);
1138 /* With an up to date view of the slots, we can check that no
1139 * other node has been improperly configured to heartbeat in
1141 own_slot_ok
= o2hb_check_own_slot(reg
);
1143 /* fill in the proper info for our next heartbeat */
1144 o2hb_prepare_block(reg
, reg
->hr_generation
);
1146 ret
= o2hb_issue_node_write(reg
, &write_wc
);
1153 while((i
= find_next_bit(configured_nodes
,
1154 O2NM_MAX_NODES
, i
+ 1)) < O2NM_MAX_NODES
) {
1155 membership_change
|= o2hb_check_slot(reg
, ®
->hr_slots
[i
]);
1159 * We have to be sure we've advertised ourselves on disk
1160 * before we can go to steady state. This ensures that
1161 * people we find in our steady state have seen us.
1163 o2hb_wait_on_io(reg
, &write_wc
);
1164 if (write_wc
.wc_error
) {
1165 /* Do not re-arm the write timeout on I/O error - we
1166 * can't be sure that the new block ever made it to
1168 mlog(ML_ERROR
, "Write error %d on device \"%s\"\n",
1169 write_wc
.wc_error
, reg
->hr_dev_name
);
1170 ret
= write_wc
.wc_error
;
1174 /* Skip disarming the timeout if own slot has stale/bad data */
1176 o2hb_set_quorum_device(reg
);
1177 o2hb_arm_timeout(reg
);
1178 reg
->hr_last_timeout_start
= jiffies
;
1182 /* let the person who launched us know when things are steady */
1183 if (atomic_read(®
->hr_steady_iterations
) != 0) {
1184 if (!ret
&& own_slot_ok
&& !membership_change
) {
1185 if (atomic_dec_and_test(®
->hr_steady_iterations
))
1186 wake_up(&o2hb_steady_queue
);
1190 if (atomic_read(®
->hr_steady_iterations
) != 0) {
1191 if (atomic_dec_and_test(®
->hr_unsteady_iterations
)) {
1192 printk(KERN_NOTICE
"o2hb: Unable to stabilize "
1193 "heartbeart on region %s (%s)\n",
1194 config_item_name(®
->hr_item
),
1196 atomic_set(®
->hr_steady_iterations
, 0);
1197 reg
->hr_aborted_start
= 1;
1198 wake_up(&o2hb_steady_queue
);
1207 * we ride the region ref that the region dir holds. before the region
1208 * dir is removed and drops it ref it will wait to tear down this
1211 static int o2hb_thread(void *data
)
1214 struct o2hb_region
*reg
= data
;
1215 struct o2hb_bio_wait_ctxt write_wc
;
1216 ktime_t before_hb
, after_hb
;
1217 unsigned int elapsed_msec
;
1219 mlog(ML_HEARTBEAT
|ML_KTHREAD
, "hb thread running\n");
1221 set_user_nice(current
, MIN_NICE
);
1224 ret
= o2nm_depend_this_node();
1226 mlog(ML_ERROR
, "Node has been deleted, ret = %d\n", ret
);
1227 reg
->hr_node_deleted
= 1;
1228 wake_up(&o2hb_steady_queue
);
1232 while (!kthread_should_stop() &&
1233 !reg
->hr_unclean_stop
&& !reg
->hr_aborted_start
) {
1234 /* We track the time spent inside
1235 * o2hb_do_disk_heartbeat so that we avoid more than
1236 * hr_timeout_ms between disk writes. On busy systems
1237 * this should result in a heartbeat which is less
1238 * likely to time itself out. */
1239 before_hb
= ktime_get_real();
1241 ret
= o2hb_do_disk_heartbeat(reg
);
1242 reg
->hr_last_hb_status
= ret
;
1244 after_hb
= ktime_get_real();
1246 elapsed_msec
= (unsigned int)
1247 ktime_ms_delta(after_hb
, before_hb
);
1250 "start = %lld, end = %lld, msec = %u, ret = %d\n",
1251 before_hb
.tv64
, after_hb
.tv64
, elapsed_msec
, ret
);
1253 if (!kthread_should_stop() &&
1254 elapsed_msec
< reg
->hr_timeout_ms
) {
1255 /* the kthread api has blocked signals for us so no
1256 * need to record the return value. */
1257 msleep_interruptible(reg
->hr_timeout_ms
- elapsed_msec
);
1261 o2hb_disarm_timeout(reg
);
1263 /* unclean stop is only used in very bad situation */
1264 for(i
= 0; !reg
->hr_unclean_stop
&& i
< reg
->hr_blocks
; i
++)
1265 o2hb_shutdown_slot(®
->hr_slots
[i
]);
1267 /* Explicit down notification - avoid forcing the other nodes
1268 * to timeout on this region when we could just as easily
1269 * write a clear generation - thus indicating to them that
1270 * this node has left this region.
1272 if (!reg
->hr_unclean_stop
&& !reg
->hr_aborted_start
) {
1273 o2hb_prepare_block(reg
, 0);
1274 ret
= o2hb_issue_node_write(reg
, &write_wc
);
1276 o2hb_wait_on_io(reg
, &write_wc
);
1282 o2nm_undepend_this_node();
1284 mlog(ML_HEARTBEAT
|ML_KTHREAD
, "o2hb thread exiting\n");
1289 #ifdef CONFIG_DEBUG_FS
1290 static int o2hb_debug_open(struct inode
*inode
, struct file
*file
)
1292 struct o2hb_debug_buf
*db
= inode
->i_private
;
1293 struct o2hb_region
*reg
;
1294 unsigned long map
[BITS_TO_LONGS(O2NM_MAX_NODES
)];
1300 /* max_nodes should be the largest bitmap we pass here */
1301 BUG_ON(sizeof(map
) < db
->db_size
);
1303 buf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
1307 switch (db
->db_type
) {
1308 case O2HB_DB_TYPE_LIVENODES
:
1309 case O2HB_DB_TYPE_LIVEREGIONS
:
1310 case O2HB_DB_TYPE_QUORUMREGIONS
:
1311 case O2HB_DB_TYPE_FAILEDREGIONS
:
1312 spin_lock(&o2hb_live_lock
);
1313 memcpy(map
, db
->db_data
, db
->db_size
);
1314 spin_unlock(&o2hb_live_lock
);
1317 case O2HB_DB_TYPE_REGION_LIVENODES
:
1318 spin_lock(&o2hb_live_lock
);
1319 reg
= (struct o2hb_region
*)db
->db_data
;
1320 memcpy(map
, reg
->hr_live_node_bitmap
, db
->db_size
);
1321 spin_unlock(&o2hb_live_lock
);
1324 case O2HB_DB_TYPE_REGION_NUMBER
:
1325 reg
= (struct o2hb_region
*)db
->db_data
;
1326 out
+= snprintf(buf
+ out
, PAGE_SIZE
- out
, "%d\n",
1327 reg
->hr_region_num
);
1330 case O2HB_DB_TYPE_REGION_ELAPSED_TIME
:
1331 reg
= (struct o2hb_region
*)db
->db_data
;
1332 lts
= reg
->hr_last_timeout_start
;
1333 /* If 0, it has never been set before */
1335 lts
= jiffies_to_msecs(jiffies
- lts
);
1336 out
+= snprintf(buf
+ out
, PAGE_SIZE
- out
, "%lu\n", lts
);
1339 case O2HB_DB_TYPE_REGION_PINNED
:
1340 reg
= (struct o2hb_region
*)db
->db_data
;
1341 out
+= snprintf(buf
+ out
, PAGE_SIZE
- out
, "%u\n",
1342 !!reg
->hr_item_pinned
);
1349 while ((i
= find_next_bit(map
, db
->db_len
, i
+ 1)) < db
->db_len
)
1350 out
+= snprintf(buf
+ out
, PAGE_SIZE
- out
, "%d ", i
);
1351 out
+= snprintf(buf
+ out
, PAGE_SIZE
- out
, "\n");
1354 i_size_write(inode
, out
);
1356 file
->private_data
= buf
;
1363 static int o2hb_debug_release(struct inode
*inode
, struct file
*file
)
1365 kfree(file
->private_data
);
1369 static ssize_t
o2hb_debug_read(struct file
*file
, char __user
*buf
,
1370 size_t nbytes
, loff_t
*ppos
)
1372 return simple_read_from_buffer(buf
, nbytes
, ppos
, file
->private_data
,
1373 i_size_read(file
->f_mapping
->host
));
1376 static int o2hb_debug_open(struct inode
*inode
, struct file
*file
)
1380 static int o2hb_debug_release(struct inode
*inode
, struct file
*file
)
1384 static ssize_t
o2hb_debug_read(struct file
*file
, char __user
*buf
,
1385 size_t nbytes
, loff_t
*ppos
)
1389 #endif /* CONFIG_DEBUG_FS */
1391 static const struct file_operations o2hb_debug_fops
= {
1392 .open
= o2hb_debug_open
,
1393 .release
= o2hb_debug_release
,
1394 .read
= o2hb_debug_read
,
1395 .llseek
= generic_file_llseek
,
1398 void o2hb_exit(void)
1400 debugfs_remove(o2hb_debug_failedregions
);
1401 debugfs_remove(o2hb_debug_quorumregions
);
1402 debugfs_remove(o2hb_debug_liveregions
);
1403 debugfs_remove(o2hb_debug_livenodes
);
1404 debugfs_remove(o2hb_debug_dir
);
1405 kfree(o2hb_db_livenodes
);
1406 kfree(o2hb_db_liveregions
);
1407 kfree(o2hb_db_quorumregions
);
1408 kfree(o2hb_db_failedregions
);
1411 static struct dentry
*o2hb_debug_create(const char *name
, struct dentry
*dir
,
1412 struct o2hb_debug_buf
**db
, int db_len
,
1413 int type
, int size
, int len
, void *data
)
1415 *db
= kmalloc(db_len
, GFP_KERNEL
);
1419 (*db
)->db_type
= type
;
1420 (*db
)->db_size
= size
;
1421 (*db
)->db_len
= len
;
1422 (*db
)->db_data
= data
;
1424 return debugfs_create_file(name
, S_IFREG
|S_IRUSR
, dir
, *db
,
1428 static int o2hb_debug_init(void)
1432 o2hb_debug_dir
= debugfs_create_dir(O2HB_DEBUG_DIR
, NULL
);
1433 if (!o2hb_debug_dir
) {
1438 o2hb_debug_livenodes
= o2hb_debug_create(O2HB_DEBUG_LIVENODES
,
1441 sizeof(*o2hb_db_livenodes
),
1442 O2HB_DB_TYPE_LIVENODES
,
1443 sizeof(o2hb_live_node_bitmap
),
1445 o2hb_live_node_bitmap
);
1446 if (!o2hb_debug_livenodes
) {
1451 o2hb_debug_liveregions
= o2hb_debug_create(O2HB_DEBUG_LIVEREGIONS
,
1453 &o2hb_db_liveregions
,
1454 sizeof(*o2hb_db_liveregions
),
1455 O2HB_DB_TYPE_LIVEREGIONS
,
1456 sizeof(o2hb_live_region_bitmap
),
1458 o2hb_live_region_bitmap
);
1459 if (!o2hb_debug_liveregions
) {
1464 o2hb_debug_quorumregions
=
1465 o2hb_debug_create(O2HB_DEBUG_QUORUMREGIONS
,
1467 &o2hb_db_quorumregions
,
1468 sizeof(*o2hb_db_quorumregions
),
1469 O2HB_DB_TYPE_QUORUMREGIONS
,
1470 sizeof(o2hb_quorum_region_bitmap
),
1472 o2hb_quorum_region_bitmap
);
1473 if (!o2hb_debug_quorumregions
) {
1478 o2hb_debug_failedregions
=
1479 o2hb_debug_create(O2HB_DEBUG_FAILEDREGIONS
,
1481 &o2hb_db_failedregions
,
1482 sizeof(*o2hb_db_failedregions
),
1483 O2HB_DB_TYPE_FAILEDREGIONS
,
1484 sizeof(o2hb_failed_region_bitmap
),
1486 o2hb_failed_region_bitmap
);
1487 if (!o2hb_debug_failedregions
) {
1504 for (i
= 0; i
< ARRAY_SIZE(o2hb_callbacks
); i
++)
1505 INIT_LIST_HEAD(&o2hb_callbacks
[i
].list
);
1507 for (i
= 0; i
< ARRAY_SIZE(o2hb_live_slots
); i
++)
1508 INIT_LIST_HEAD(&o2hb_live_slots
[i
]);
1510 INIT_LIST_HEAD(&o2hb_node_events
);
1512 memset(o2hb_live_node_bitmap
, 0, sizeof(o2hb_live_node_bitmap
));
1513 memset(o2hb_region_bitmap
, 0, sizeof(o2hb_region_bitmap
));
1514 memset(o2hb_live_region_bitmap
, 0, sizeof(o2hb_live_region_bitmap
));
1515 memset(o2hb_quorum_region_bitmap
, 0, sizeof(o2hb_quorum_region_bitmap
));
1516 memset(o2hb_failed_region_bitmap
, 0, sizeof(o2hb_failed_region_bitmap
));
1518 o2hb_dependent_users
= 0;
1520 return o2hb_debug_init();
1523 /* if we're already in a callback then we're already serialized by the sem */
1524 static void o2hb_fill_node_map_from_callback(unsigned long *map
,
1527 BUG_ON(bytes
< (BITS_TO_LONGS(O2NM_MAX_NODES
) * sizeof(unsigned long)));
1529 memcpy(map
, &o2hb_live_node_bitmap
, bytes
);
1533 * get a map of all nodes that are heartbeating in any regions
1535 void o2hb_fill_node_map(unsigned long *map
, unsigned bytes
)
1537 /* callers want to serialize this map and callbacks so that they
1538 * can trust that they don't miss nodes coming to the party */
1539 down_read(&o2hb_callback_sem
);
1540 spin_lock(&o2hb_live_lock
);
1541 o2hb_fill_node_map_from_callback(map
, bytes
);
1542 spin_unlock(&o2hb_live_lock
);
1543 up_read(&o2hb_callback_sem
);
1545 EXPORT_SYMBOL_GPL(o2hb_fill_node_map
);
1548 * heartbeat configfs bits. The heartbeat set is a default set under
1549 * the cluster set in nodemanager.c.
1552 static struct o2hb_region
*to_o2hb_region(struct config_item
*item
)
1554 return item
? container_of(item
, struct o2hb_region
, hr_item
) : NULL
;
1557 /* drop_item only drops its ref after killing the thread, nothing should
1558 * be using the region anymore. this has to clean up any state that
1559 * attributes might have built up. */
1560 static void o2hb_region_release(struct config_item
*item
)
1564 struct o2hb_region
*reg
= to_o2hb_region(item
);
1566 mlog(ML_HEARTBEAT
, "hb region release (%s)\n", reg
->hr_dev_name
);
1568 kfree(reg
->hr_tmp_block
);
1570 if (reg
->hr_slot_data
) {
1571 for (i
= 0; i
< reg
->hr_num_pages
; i
++) {
1572 page
= reg
->hr_slot_data
[i
];
1576 kfree(reg
->hr_slot_data
);
1580 blkdev_put(reg
->hr_bdev
, FMODE_READ
|FMODE_WRITE
);
1582 kfree(reg
->hr_slots
);
1584 debugfs_remove(reg
->hr_debug_livenodes
);
1585 debugfs_remove(reg
->hr_debug_regnum
);
1586 debugfs_remove(reg
->hr_debug_elapsed_time
);
1587 debugfs_remove(reg
->hr_debug_pinned
);
1588 debugfs_remove(reg
->hr_debug_dir
);
1589 kfree(reg
->hr_db_livenodes
);
1590 kfree(reg
->hr_db_regnum
);
1591 kfree(reg
->hr_db_elapsed_time
);
1592 kfree(reg
->hr_db_pinned
);
1594 spin_lock(&o2hb_live_lock
);
1595 list_del(®
->hr_all_item
);
1596 spin_unlock(&o2hb_live_lock
);
1598 o2net_unregister_handler_list(®
->hr_handler_list
);
1602 static int o2hb_read_block_input(struct o2hb_region
*reg
,
1604 unsigned long *ret_bytes
,
1605 unsigned int *ret_bits
)
1607 unsigned long bytes
;
1608 char *p
= (char *)page
;
1610 bytes
= simple_strtoul(p
, &p
, 0);
1611 if (!p
|| (*p
&& (*p
!= '\n')))
1614 /* Heartbeat and fs min / max block sizes are the same. */
1615 if (bytes
> 4096 || bytes
< 512)
1617 if (hweight16(bytes
) != 1)
1623 *ret_bits
= ffs(bytes
) - 1;
1628 static ssize_t
o2hb_region_block_bytes_show(struct config_item
*item
,
1631 return sprintf(page
, "%u\n", to_o2hb_region(item
)->hr_block_bytes
);
1634 static ssize_t
o2hb_region_block_bytes_store(struct config_item
*item
,
1638 struct o2hb_region
*reg
= to_o2hb_region(item
);
1640 unsigned long block_bytes
;
1641 unsigned int block_bits
;
1646 status
= o2hb_read_block_input(reg
, page
, &block_bytes
,
1651 reg
->hr_block_bytes
= (unsigned int)block_bytes
;
1652 reg
->hr_block_bits
= block_bits
;
1657 static ssize_t
o2hb_region_start_block_show(struct config_item
*item
,
1660 return sprintf(page
, "%llu\n", to_o2hb_region(item
)->hr_start_block
);
1663 static ssize_t
o2hb_region_start_block_store(struct config_item
*item
,
1667 struct o2hb_region
*reg
= to_o2hb_region(item
);
1668 unsigned long long tmp
;
1669 char *p
= (char *)page
;
1674 tmp
= simple_strtoull(p
, &p
, 0);
1675 if (!p
|| (*p
&& (*p
!= '\n')))
1678 reg
->hr_start_block
= tmp
;
1683 static ssize_t
o2hb_region_blocks_show(struct config_item
*item
, char *page
)
1685 return sprintf(page
, "%d\n", to_o2hb_region(item
)->hr_blocks
);
1688 static ssize_t
o2hb_region_blocks_store(struct config_item
*item
,
1692 struct o2hb_region
*reg
= to_o2hb_region(item
);
1694 char *p
= (char *)page
;
1699 tmp
= simple_strtoul(p
, &p
, 0);
1700 if (!p
|| (*p
&& (*p
!= '\n')))
1703 if (tmp
> O2NM_MAX_NODES
|| tmp
== 0)
1706 reg
->hr_blocks
= (unsigned int)tmp
;
1711 static ssize_t
o2hb_region_dev_show(struct config_item
*item
, char *page
)
1713 unsigned int ret
= 0;
1715 if (to_o2hb_region(item
)->hr_bdev
)
1716 ret
= sprintf(page
, "%s\n", to_o2hb_region(item
)->hr_dev_name
);
1721 static void o2hb_init_region_params(struct o2hb_region
*reg
)
1723 reg
->hr_slots_per_page
= PAGE_SIZE
>> reg
->hr_block_bits
;
1724 reg
->hr_timeout_ms
= O2HB_REGION_TIMEOUT_MS
;
1726 mlog(ML_HEARTBEAT
, "hr_start_block = %llu, hr_blocks = %u\n",
1727 reg
->hr_start_block
, reg
->hr_blocks
);
1728 mlog(ML_HEARTBEAT
, "hr_block_bytes = %u, hr_block_bits = %u\n",
1729 reg
->hr_block_bytes
, reg
->hr_block_bits
);
1730 mlog(ML_HEARTBEAT
, "hr_timeout_ms = %u\n", reg
->hr_timeout_ms
);
1731 mlog(ML_HEARTBEAT
, "dead threshold = %u\n", o2hb_dead_threshold
);
1734 static int o2hb_map_slot_data(struct o2hb_region
*reg
)
1737 unsigned int last_slot
;
1738 unsigned int spp
= reg
->hr_slots_per_page
;
1741 struct o2hb_disk_slot
*slot
;
1743 reg
->hr_tmp_block
= kmalloc(reg
->hr_block_bytes
, GFP_KERNEL
);
1744 if (reg
->hr_tmp_block
== NULL
)
1747 reg
->hr_slots
= kcalloc(reg
->hr_blocks
,
1748 sizeof(struct o2hb_disk_slot
), GFP_KERNEL
);
1749 if (reg
->hr_slots
== NULL
)
1752 for(i
= 0; i
< reg
->hr_blocks
; i
++) {
1753 slot
= ®
->hr_slots
[i
];
1754 slot
->ds_node_num
= i
;
1755 INIT_LIST_HEAD(&slot
->ds_live_item
);
1756 slot
->ds_raw_block
= NULL
;
1759 reg
->hr_num_pages
= (reg
->hr_blocks
+ spp
- 1) / spp
;
1760 mlog(ML_HEARTBEAT
, "Going to require %u pages to cover %u blocks "
1761 "at %u blocks per page\n",
1762 reg
->hr_num_pages
, reg
->hr_blocks
, spp
);
1764 reg
->hr_slot_data
= kcalloc(reg
->hr_num_pages
, sizeof(struct page
*),
1766 if (!reg
->hr_slot_data
)
1769 for(i
= 0; i
< reg
->hr_num_pages
; i
++) {
1770 page
= alloc_page(GFP_KERNEL
);
1774 reg
->hr_slot_data
[i
] = page
;
1776 last_slot
= i
* spp
;
1777 raw
= page_address(page
);
1779 (j
< spp
) && ((j
+ last_slot
) < reg
->hr_blocks
);
1781 BUG_ON((j
+ last_slot
) >= reg
->hr_blocks
);
1783 slot
= ®
->hr_slots
[j
+ last_slot
];
1784 slot
->ds_raw_block
=
1785 (struct o2hb_disk_heartbeat_block
*) raw
;
1787 raw
+= reg
->hr_block_bytes
;
1794 /* Read in all the slots available and populate the tracking
1795 * structures so that we can start with a baseline idea of what's
1797 static int o2hb_populate_slot_data(struct o2hb_region
*reg
)
1800 struct o2hb_disk_slot
*slot
;
1801 struct o2hb_disk_heartbeat_block
*hb_block
;
1803 ret
= o2hb_read_slots(reg
, reg
->hr_blocks
);
1807 /* We only want to get an idea of the values initially in each
1808 * slot, so we do no verification - o2hb_check_slot will
1809 * actually determine if each configured slot is valid and
1810 * whether any values have changed. */
1811 for(i
= 0; i
< reg
->hr_blocks
; i
++) {
1812 slot
= ®
->hr_slots
[i
];
1813 hb_block
= (struct o2hb_disk_heartbeat_block
*) slot
->ds_raw_block
;
1815 /* Only fill the values that o2hb_check_slot uses to
1816 * determine changing slots */
1817 slot
->ds_last_time
= le64_to_cpu(hb_block
->hb_seq
);
1818 slot
->ds_last_generation
= le64_to_cpu(hb_block
->hb_generation
);
1825 /* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */
1826 static ssize_t
o2hb_region_dev_store(struct config_item
*item
,
1830 struct o2hb_region
*reg
= to_o2hb_region(item
);
1831 struct task_struct
*hb_task
;
1834 char *p
= (char *)page
;
1836 struct inode
*inode
;
1837 ssize_t ret
= -EINVAL
;
1843 /* We can't heartbeat without having had our node number
1844 * configured yet. */
1845 if (o2nm_this_node() == O2NM_MAX_NODES
)
1848 fd
= simple_strtol(p
, &p
, 0);
1849 if (!p
|| (*p
&& (*p
!= '\n')))
1852 if (fd
< 0 || fd
>= INT_MAX
)
1859 if (reg
->hr_blocks
== 0 || reg
->hr_start_block
== 0 ||
1860 reg
->hr_block_bytes
== 0)
1863 inode
= igrab(f
.file
->f_mapping
->host
);
1867 if (!S_ISBLK(inode
->i_mode
))
1870 reg
->hr_bdev
= I_BDEV(f
.file
->f_mapping
->host
);
1871 ret
= blkdev_get(reg
->hr_bdev
, FMODE_WRITE
| FMODE_READ
, NULL
);
1873 reg
->hr_bdev
= NULL
;
1878 bdevname(reg
->hr_bdev
, reg
->hr_dev_name
);
1880 sectsize
= bdev_logical_block_size(reg
->hr_bdev
);
1881 if (sectsize
!= reg
->hr_block_bytes
) {
1883 "blocksize %u incorrect for device, expected %d",
1884 reg
->hr_block_bytes
, sectsize
);
1889 o2hb_init_region_params(reg
);
1891 /* Generation of zero is invalid */
1893 get_random_bytes(®
->hr_generation
,
1894 sizeof(reg
->hr_generation
));
1895 } while (reg
->hr_generation
== 0);
1897 ret
= o2hb_map_slot_data(reg
);
1903 ret
= o2hb_populate_slot_data(reg
);
1909 INIT_DELAYED_WORK(®
->hr_write_timeout_work
, o2hb_write_timeout
);
1910 INIT_DELAYED_WORK(®
->hr_nego_timeout_work
, o2hb_nego_timeout
);
1913 * A node is considered live after it has beat LIVE_THRESHOLD
1914 * times. We're not steady until we've given them a chance
1915 * _after_ our first read.
1916 * The default threshold is bare minimum so as to limit the delay
1917 * during mounts. For global heartbeat, the threshold doubled for the
1920 live_threshold
= O2HB_LIVE_THRESHOLD
;
1921 if (o2hb_global_heartbeat_active()) {
1922 spin_lock(&o2hb_live_lock
);
1923 if (bitmap_weight(o2hb_region_bitmap
, O2NM_MAX_REGIONS
) == 1)
1924 live_threshold
<<= 1;
1925 spin_unlock(&o2hb_live_lock
);
1928 atomic_set(®
->hr_steady_iterations
, live_threshold
);
1929 /* unsteady_iterations is triple the steady_iterations */
1930 atomic_set(®
->hr_unsteady_iterations
, (live_threshold
* 3));
1932 hb_task
= kthread_run(o2hb_thread
, reg
, "o2hb-%s",
1933 reg
->hr_item
.ci_name
);
1934 if (IS_ERR(hb_task
)) {
1935 ret
= PTR_ERR(hb_task
);
1940 spin_lock(&o2hb_live_lock
);
1941 reg
->hr_task
= hb_task
;
1942 spin_unlock(&o2hb_live_lock
);
1944 ret
= wait_event_interruptible(o2hb_steady_queue
,
1945 atomic_read(®
->hr_steady_iterations
) == 0 ||
1946 reg
->hr_node_deleted
);
1948 atomic_set(®
->hr_steady_iterations
, 0);
1949 reg
->hr_aborted_start
= 1;
1952 if (reg
->hr_aborted_start
) {
1957 if (reg
->hr_node_deleted
) {
1962 /* Ok, we were woken. Make sure it wasn't by drop_item() */
1963 spin_lock(&o2hb_live_lock
);
1964 hb_task
= reg
->hr_task
;
1965 if (o2hb_global_heartbeat_active())
1966 set_bit(reg
->hr_region_num
, o2hb_live_region_bitmap
);
1967 spin_unlock(&o2hb_live_lock
);
1974 if (hb_task
&& o2hb_global_heartbeat_active())
1975 printk(KERN_NOTICE
"o2hb: Heartbeat started on region %s (%s)\n",
1976 config_item_name(®
->hr_item
), reg
->hr_dev_name
);
1985 blkdev_put(reg
->hr_bdev
, FMODE_READ
|FMODE_WRITE
);
1986 reg
->hr_bdev
= NULL
;
1992 static ssize_t
o2hb_region_pid_show(struct config_item
*item
, char *page
)
1994 struct o2hb_region
*reg
= to_o2hb_region(item
);
1997 spin_lock(&o2hb_live_lock
);
1999 pid
= task_pid_nr(reg
->hr_task
);
2000 spin_unlock(&o2hb_live_lock
);
2005 return sprintf(page
, "%u\n", pid
);
2008 CONFIGFS_ATTR(o2hb_region_
, block_bytes
);
2009 CONFIGFS_ATTR(o2hb_region_
, start_block
);
2010 CONFIGFS_ATTR(o2hb_region_
, blocks
);
2011 CONFIGFS_ATTR(o2hb_region_
, dev
);
2012 CONFIGFS_ATTR_RO(o2hb_region_
, pid
);
2014 static struct configfs_attribute
*o2hb_region_attrs
[] = {
2015 &o2hb_region_attr_block_bytes
,
2016 &o2hb_region_attr_start_block
,
2017 &o2hb_region_attr_blocks
,
2018 &o2hb_region_attr_dev
,
2019 &o2hb_region_attr_pid
,
2023 static struct configfs_item_operations o2hb_region_item_ops
= {
2024 .release
= o2hb_region_release
,
2027 static struct config_item_type o2hb_region_type
= {
2028 .ct_item_ops
= &o2hb_region_item_ops
,
2029 .ct_attrs
= o2hb_region_attrs
,
2030 .ct_owner
= THIS_MODULE
,
2035 struct o2hb_heartbeat_group
{
2036 struct config_group hs_group
;
2040 static struct o2hb_heartbeat_group
*to_o2hb_heartbeat_group(struct config_group
*group
)
2043 container_of(group
, struct o2hb_heartbeat_group
, hs_group
)
2047 static int o2hb_debug_region_init(struct o2hb_region
*reg
, struct dentry
*dir
)
2052 debugfs_create_dir(config_item_name(®
->hr_item
), dir
);
2053 if (!reg
->hr_debug_dir
) {
2058 reg
->hr_debug_livenodes
=
2059 o2hb_debug_create(O2HB_DEBUG_LIVENODES
,
2061 &(reg
->hr_db_livenodes
),
2062 sizeof(*(reg
->hr_db_livenodes
)),
2063 O2HB_DB_TYPE_REGION_LIVENODES
,
2064 sizeof(reg
->hr_live_node_bitmap
),
2065 O2NM_MAX_NODES
, reg
);
2066 if (!reg
->hr_debug_livenodes
) {
2071 reg
->hr_debug_regnum
=
2072 o2hb_debug_create(O2HB_DEBUG_REGION_NUMBER
,
2074 &(reg
->hr_db_regnum
),
2075 sizeof(*(reg
->hr_db_regnum
)),
2076 O2HB_DB_TYPE_REGION_NUMBER
,
2077 0, O2NM_MAX_NODES
, reg
);
2078 if (!reg
->hr_debug_regnum
) {
2083 reg
->hr_debug_elapsed_time
=
2084 o2hb_debug_create(O2HB_DEBUG_REGION_ELAPSED_TIME
,
2086 &(reg
->hr_db_elapsed_time
),
2087 sizeof(*(reg
->hr_db_elapsed_time
)),
2088 O2HB_DB_TYPE_REGION_ELAPSED_TIME
,
2090 if (!reg
->hr_debug_elapsed_time
) {
2095 reg
->hr_debug_pinned
=
2096 o2hb_debug_create(O2HB_DEBUG_REGION_PINNED
,
2098 &(reg
->hr_db_pinned
),
2099 sizeof(*(reg
->hr_db_pinned
)),
2100 O2HB_DB_TYPE_REGION_PINNED
,
2102 if (!reg
->hr_debug_pinned
) {
2112 static struct config_item
*o2hb_heartbeat_group_make_item(struct config_group
*group
,
2115 struct o2hb_region
*reg
= NULL
;
2118 reg
= kzalloc(sizeof(struct o2hb_region
), GFP_KERNEL
);
2120 return ERR_PTR(-ENOMEM
);
2122 if (strlen(name
) > O2HB_MAX_REGION_NAME_LEN
) {
2123 ret
= -ENAMETOOLONG
;
2127 spin_lock(&o2hb_live_lock
);
2128 reg
->hr_region_num
= 0;
2129 if (o2hb_global_heartbeat_active()) {
2130 reg
->hr_region_num
= find_first_zero_bit(o2hb_region_bitmap
,
2132 if (reg
->hr_region_num
>= O2NM_MAX_REGIONS
) {
2133 spin_unlock(&o2hb_live_lock
);
2137 set_bit(reg
->hr_region_num
, o2hb_region_bitmap
);
2139 list_add_tail(®
->hr_all_item
, &o2hb_all_regions
);
2140 spin_unlock(&o2hb_live_lock
);
2142 config_item_init_type_name(®
->hr_item
, name
, &o2hb_region_type
);
2144 /* this is the same way to generate msg key as dlm, for local heartbeat,
2145 * name is also the same, so make initial crc value different to avoid
2146 * message key conflict.
2148 reg
->hr_key
= crc32_le(reg
->hr_region_num
+ O2NM_MAX_REGIONS
,
2149 name
, strlen(name
));
2150 INIT_LIST_HEAD(®
->hr_handler_list
);
2151 ret
= o2net_register_handler(O2HB_NEGO_TIMEOUT_MSG
, reg
->hr_key
,
2152 sizeof(struct o2hb_nego_msg
),
2153 o2hb_nego_timeout_handler
,
2154 reg
, NULL
, ®
->hr_handler_list
);
2158 ret
= o2net_register_handler(O2HB_NEGO_APPROVE_MSG
, reg
->hr_key
,
2159 sizeof(struct o2hb_nego_msg
),
2160 o2hb_nego_approve_handler
,
2161 reg
, NULL
, ®
->hr_handler_list
);
2163 goto unregister_handler
;
2165 ret
= o2hb_debug_region_init(reg
, o2hb_debug_dir
);
2167 config_item_put(®
->hr_item
);
2168 goto unregister_handler
;
2171 return ®
->hr_item
;
2174 o2net_unregister_handler_list(®
->hr_handler_list
);
2177 return ERR_PTR(ret
);
2180 static void o2hb_heartbeat_group_drop_item(struct config_group
*group
,
2181 struct config_item
*item
)
2183 struct task_struct
*hb_task
;
2184 struct o2hb_region
*reg
= to_o2hb_region(item
);
2185 int quorum_region
= 0;
2187 /* stop the thread when the user removes the region dir */
2188 spin_lock(&o2hb_live_lock
);
2189 hb_task
= reg
->hr_task
;
2190 reg
->hr_task
= NULL
;
2191 reg
->hr_item_dropped
= 1;
2192 spin_unlock(&o2hb_live_lock
);
2195 kthread_stop(hb_task
);
2197 if (o2hb_global_heartbeat_active()) {
2198 spin_lock(&o2hb_live_lock
);
2199 clear_bit(reg
->hr_region_num
, o2hb_region_bitmap
);
2200 clear_bit(reg
->hr_region_num
, o2hb_live_region_bitmap
);
2201 if (test_bit(reg
->hr_region_num
, o2hb_quorum_region_bitmap
))
2203 clear_bit(reg
->hr_region_num
, o2hb_quorum_region_bitmap
);
2204 spin_unlock(&o2hb_live_lock
);
2205 printk(KERN_NOTICE
"o2hb: Heartbeat %s on region %s (%s)\n",
2206 ((atomic_read(®
->hr_steady_iterations
) == 0) ?
2207 "stopped" : "start aborted"), config_item_name(item
),
2212 * If we're racing a dev_write(), we need to wake them. They will
2213 * check reg->hr_task
2215 if (atomic_read(®
->hr_steady_iterations
) != 0) {
2216 reg
->hr_aborted_start
= 1;
2217 atomic_set(®
->hr_steady_iterations
, 0);
2218 wake_up(&o2hb_steady_queue
);
2221 config_item_put(item
);
2223 if (!o2hb_global_heartbeat_active() || !quorum_region
)
2227 * If global heartbeat active and there are dependent users,
2228 * pin all regions if quorum region count <= CUT_OFF
2230 spin_lock(&o2hb_live_lock
);
2232 if (!o2hb_dependent_users
)
2235 if (bitmap_weight(o2hb_quorum_region_bitmap
,
2236 O2NM_MAX_REGIONS
) <= O2HB_PIN_CUT_OFF
)
2237 o2hb_region_pin(NULL
);
2240 spin_unlock(&o2hb_live_lock
);
2243 static ssize_t
o2hb_heartbeat_group_threshold_show(struct config_item
*item
,
2246 return sprintf(page
, "%u\n", o2hb_dead_threshold
);
2249 static ssize_t
o2hb_heartbeat_group_threshold_store(struct config_item
*item
,
2250 const char *page
, size_t count
)
2253 char *p
= (char *)page
;
2255 tmp
= simple_strtoul(p
, &p
, 10);
2256 if (!p
|| (*p
&& (*p
!= '\n')))
2259 /* this will validate ranges for us. */
2260 o2hb_dead_threshold_set((unsigned int) tmp
);
2265 static ssize_t
o2hb_heartbeat_group_mode_show(struct config_item
*item
,
2268 return sprintf(page
, "%s\n",
2269 o2hb_heartbeat_mode_desc
[o2hb_heartbeat_mode
]);
2272 static ssize_t
o2hb_heartbeat_group_mode_store(struct config_item
*item
,
2273 const char *page
, size_t count
)
2279 len
= (page
[count
- 1] == '\n') ? count
- 1 : count
;
2283 for (i
= 0; i
< O2HB_HEARTBEAT_NUM_MODES
; ++i
) {
2284 if (strncasecmp(page
, o2hb_heartbeat_mode_desc
[i
], len
))
2287 ret
= o2hb_global_heartbeat_mode_set(i
);
2289 printk(KERN_NOTICE
"o2hb: Heartbeat mode set to %s\n",
2290 o2hb_heartbeat_mode_desc
[i
]);
2298 CONFIGFS_ATTR(o2hb_heartbeat_group_
, threshold
);
2299 CONFIGFS_ATTR(o2hb_heartbeat_group_
, mode
);
2301 static struct configfs_attribute
*o2hb_heartbeat_group_attrs
[] = {
2302 &o2hb_heartbeat_group_attr_threshold
,
2303 &o2hb_heartbeat_group_attr_mode
,
2307 static struct configfs_group_operations o2hb_heartbeat_group_group_ops
= {
2308 .make_item
= o2hb_heartbeat_group_make_item
,
2309 .drop_item
= o2hb_heartbeat_group_drop_item
,
2312 static struct config_item_type o2hb_heartbeat_group_type
= {
2313 .ct_group_ops
= &o2hb_heartbeat_group_group_ops
,
2314 .ct_attrs
= o2hb_heartbeat_group_attrs
,
2315 .ct_owner
= THIS_MODULE
,
2318 /* this is just here to avoid touching group in heartbeat.h which the
2319 * entire damn world #includes */
2320 struct config_group
*o2hb_alloc_hb_set(void)
2322 struct o2hb_heartbeat_group
*hs
= NULL
;
2323 struct config_group
*ret
= NULL
;
2325 hs
= kzalloc(sizeof(struct o2hb_heartbeat_group
), GFP_KERNEL
);
2329 config_group_init_type_name(&hs
->hs_group
, "heartbeat",
2330 &o2hb_heartbeat_group_type
);
2332 ret
= &hs
->hs_group
;
2339 void o2hb_free_hb_set(struct config_group
*group
)
2341 struct o2hb_heartbeat_group
*hs
= to_o2hb_heartbeat_group(group
);
2345 /* hb callback registration and issuing */
2347 static struct o2hb_callback
*hbcall_from_type(enum o2hb_callback_type type
)
2349 if (type
== O2HB_NUM_CB
)
2350 return ERR_PTR(-EINVAL
);
2352 return &o2hb_callbacks
[type
];
2355 void o2hb_setup_callback(struct o2hb_callback_func
*hc
,
2356 enum o2hb_callback_type type
,
2361 INIT_LIST_HEAD(&hc
->hc_item
);
2364 hc
->hc_priority
= priority
;
2366 hc
->hc_magic
= O2HB_CB_MAGIC
;
2368 EXPORT_SYMBOL_GPL(o2hb_setup_callback
);
2371 * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2372 * In global heartbeat mode, region_uuid passed is NULL.
2374 * In local, we only pin the matching region. In global we pin all the active
2377 static int o2hb_region_pin(const char *region_uuid
)
2379 int ret
= 0, found
= 0;
2380 struct o2hb_region
*reg
;
2383 assert_spin_locked(&o2hb_live_lock
);
2385 list_for_each_entry(reg
, &o2hb_all_regions
, hr_all_item
) {
2386 if (reg
->hr_item_dropped
)
2389 uuid
= config_item_name(®
->hr_item
);
2391 /* local heartbeat */
2393 if (strcmp(region_uuid
, uuid
))
2398 if (reg
->hr_item_pinned
|| reg
->hr_item_dropped
)
2401 /* Ignore ENOENT only for local hb (userdlm domain) */
2402 ret
= o2nm_depend_item(®
->hr_item
);
2404 mlog(ML_CLUSTER
, "Pin region %s\n", uuid
);
2405 reg
->hr_item_pinned
= 1;
2407 if (ret
== -ENOENT
&& found
)
2410 mlog(ML_ERROR
, "Pin region %s fails with %d\n",
2424 * In local heartbeat mode, region_uuid passed matches the dlm domain name.
2425 * In global heartbeat mode, region_uuid passed is NULL.
2427 * In local, we only unpin the matching region. In global we unpin all the
2430 static void o2hb_region_unpin(const char *region_uuid
)
2432 struct o2hb_region
*reg
;
2436 assert_spin_locked(&o2hb_live_lock
);
2438 list_for_each_entry(reg
, &o2hb_all_regions
, hr_all_item
) {
2439 if (reg
->hr_item_dropped
)
2442 uuid
= config_item_name(®
->hr_item
);
2444 if (strcmp(region_uuid
, uuid
))
2449 if (reg
->hr_item_pinned
) {
2450 mlog(ML_CLUSTER
, "Unpin region %s\n", uuid
);
2451 o2nm_undepend_item(®
->hr_item
);
2452 reg
->hr_item_pinned
= 0;
2459 static int o2hb_region_inc_user(const char *region_uuid
)
2463 spin_lock(&o2hb_live_lock
);
2465 /* local heartbeat */
2466 if (!o2hb_global_heartbeat_active()) {
2467 ret
= o2hb_region_pin(region_uuid
);
2472 * if global heartbeat active and this is the first dependent user,
2473 * pin all regions if quorum region count <= CUT_OFF
2475 o2hb_dependent_users
++;
2476 if (o2hb_dependent_users
> 1)
2479 if (bitmap_weight(o2hb_quorum_region_bitmap
,
2480 O2NM_MAX_REGIONS
) <= O2HB_PIN_CUT_OFF
)
2481 ret
= o2hb_region_pin(NULL
);
2484 spin_unlock(&o2hb_live_lock
);
2488 void o2hb_region_dec_user(const char *region_uuid
)
2490 spin_lock(&o2hb_live_lock
);
2492 /* local heartbeat */
2493 if (!o2hb_global_heartbeat_active()) {
2494 o2hb_region_unpin(region_uuid
);
2499 * if global heartbeat active and there are no dependent users,
2500 * unpin all quorum regions
2502 o2hb_dependent_users
--;
2503 if (!o2hb_dependent_users
)
2504 o2hb_region_unpin(NULL
);
2507 spin_unlock(&o2hb_live_lock
);
2510 int o2hb_register_callback(const char *region_uuid
,
2511 struct o2hb_callback_func
*hc
)
2513 struct o2hb_callback_func
*f
;
2514 struct o2hb_callback
*hbcall
;
2517 BUG_ON(hc
->hc_magic
!= O2HB_CB_MAGIC
);
2518 BUG_ON(!list_empty(&hc
->hc_item
));
2520 hbcall
= hbcall_from_type(hc
->hc_type
);
2521 if (IS_ERR(hbcall
)) {
2522 ret
= PTR_ERR(hbcall
);
2527 ret
= o2hb_region_inc_user(region_uuid
);
2534 down_write(&o2hb_callback_sem
);
2536 list_for_each_entry(f
, &hbcall
->list
, hc_item
) {
2537 if (hc
->hc_priority
< f
->hc_priority
) {
2538 list_add_tail(&hc
->hc_item
, &f
->hc_item
);
2542 if (list_empty(&hc
->hc_item
))
2543 list_add_tail(&hc
->hc_item
, &hbcall
->list
);
2545 up_write(&o2hb_callback_sem
);
2548 mlog(ML_CLUSTER
, "returning %d on behalf of %p for funcs %p\n",
2549 ret
, __builtin_return_address(0), hc
);
2552 EXPORT_SYMBOL_GPL(o2hb_register_callback
);
2554 void o2hb_unregister_callback(const char *region_uuid
,
2555 struct o2hb_callback_func
*hc
)
2557 BUG_ON(hc
->hc_magic
!= O2HB_CB_MAGIC
);
2559 mlog(ML_CLUSTER
, "on behalf of %p for funcs %p\n",
2560 __builtin_return_address(0), hc
);
2562 /* XXX Can this happen _with_ a region reference? */
2563 if (list_empty(&hc
->hc_item
))
2567 o2hb_region_dec_user(region_uuid
);
2569 down_write(&o2hb_callback_sem
);
2571 list_del_init(&hc
->hc_item
);
2573 up_write(&o2hb_callback_sem
);
2575 EXPORT_SYMBOL_GPL(o2hb_unregister_callback
);
2577 int o2hb_check_node_heartbeating(u8 node_num
)
2579 unsigned long testing_map
[BITS_TO_LONGS(O2NM_MAX_NODES
)];
2581 o2hb_fill_node_map(testing_map
, sizeof(testing_map
));
2582 if (!test_bit(node_num
, testing_map
)) {
2584 "node (%u) does not have heartbeating enabled.\n",
2591 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating
);
2593 int o2hb_check_node_heartbeating_no_sem(u8 node_num
)
2595 unsigned long testing_map
[BITS_TO_LONGS(O2NM_MAX_NODES
)];
2597 spin_lock(&o2hb_live_lock
);
2598 o2hb_fill_node_map_from_callback(testing_map
, sizeof(testing_map
));
2599 spin_unlock(&o2hb_live_lock
);
2600 if (!test_bit(node_num
, testing_map
)) {
2602 "node (%u) does not have heartbeating enabled.\n",
2609 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_no_sem
);
2611 int o2hb_check_node_heartbeating_from_callback(u8 node_num
)
2613 unsigned long testing_map
[BITS_TO_LONGS(O2NM_MAX_NODES
)];
2615 o2hb_fill_node_map_from_callback(testing_map
, sizeof(testing_map
));
2616 if (!test_bit(node_num
, testing_map
)) {
2618 "node (%u) does not have heartbeating enabled.\n",
2625 EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback
);
2627 /* Makes sure our local node is configured with a node number, and is
2629 int o2hb_check_local_node_heartbeating(void)
2633 /* if this node was set then we have networking */
2634 node_num
= o2nm_this_node();
2635 if (node_num
== O2NM_MAX_NODES
) {
2636 mlog(ML_HEARTBEAT
, "this node has not been configured.\n");
2640 return o2hb_check_node_heartbeating(node_num
);
2642 EXPORT_SYMBOL_GPL(o2hb_check_local_node_heartbeating
);
2645 * this is just a hack until we get the plumbing which flips file systems
2646 * read only and drops the hb ref instead of killing the node dead.
2648 void o2hb_stop_all_regions(void)
2650 struct o2hb_region
*reg
;
2652 mlog(ML_ERROR
, "stopping heartbeat on all active regions.\n");
2654 spin_lock(&o2hb_live_lock
);
2656 list_for_each_entry(reg
, &o2hb_all_regions
, hr_all_item
)
2657 reg
->hr_unclean_stop
= 1;
2659 spin_unlock(&o2hb_live_lock
);
2661 EXPORT_SYMBOL_GPL(o2hb_stop_all_regions
);
2663 int o2hb_get_all_regions(char *region_uuids
, u8 max_regions
)
2665 struct o2hb_region
*reg
;
2669 spin_lock(&o2hb_live_lock
);
2672 list_for_each_entry(reg
, &o2hb_all_regions
, hr_all_item
) {
2673 if (reg
->hr_item_dropped
)
2676 mlog(0, "Region: %s\n", config_item_name(®
->hr_item
));
2677 if (numregs
< max_regions
) {
2678 memcpy(p
, config_item_name(®
->hr_item
),
2679 O2HB_MAX_REGION_NAME_LEN
);
2680 p
+= O2HB_MAX_REGION_NAME_LEN
;
2685 spin_unlock(&o2hb_live_lock
);
2689 EXPORT_SYMBOL_GPL(o2hb_get_all_regions
);
2691 int o2hb_global_heartbeat_active(void)
2693 return (o2hb_heartbeat_mode
== O2HB_HEARTBEAT_GLOBAL
);
2695 EXPORT_SYMBOL(o2hb_global_heartbeat_active
);