]>
Commit | Line | Data |
---|---|---|
a7f6a5fb MF |
1 | /* -*- mode: c; c-basic-offset: 8; -*- |
2 | * vim: noexpandtab sw=8 ts=8 sts=0: | |
3 | * | |
4 | * Copyright (C) 2004, 2005 Oracle. All rights reserved. | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or | |
7 | * modify it under the terms of the GNU General Public | |
8 | * License as published by the Free Software Foundation; either | |
9 | * version 2 of the License, or (at your option) any later version. | |
10 | * | |
11 | * This program is distributed in the hope that it will be useful, | |
12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
14 | * General Public License for more details. | |
15 | * | |
16 | * You should have received a copy of the GNU General Public | |
17 | * License along with this program; if not, write to the | |
18 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
19 | * Boston, MA 021110-1307, USA. | |
20 | */ | |
21 | ||
22 | #include <linux/kernel.h> | |
23 | #include <linux/sched.h> | |
24 | #include <linux/jiffies.h> | |
25 | #include <linux/module.h> | |
26 | #include <linux/fs.h> | |
27 | #include <linux/bio.h> | |
28 | #include <linux/blkdev.h> | |
29 | #include <linux/delay.h> | |
30 | #include <linux/file.h> | |
31 | #include <linux/kthread.h> | |
32 | #include <linux/configfs.h> | |
33 | #include <linux/random.h> | |
34 | #include <linux/crc32.h> | |
35 | #include <linux/time.h> | |
87d3d3f3 | 36 | #include <linux/debugfs.h> |
5a0e3ad6 | 37 | #include <linux/slab.h> |
a7f6a5fb MF |
38 | |
39 | #include "heartbeat.h" | |
40 | #include "tcp.h" | |
41 | #include "nodemanager.h" | |
42 | #include "quorum.h" | |
43 | ||
44 | #include "masklog.h" | |
45 | ||
46 | ||
47 | /* | |
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. | |
52 | */ | |
53 | static DECLARE_RWSEM(o2hb_callback_sem); | |
54 | ||
55 | /* | |
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. | |
58 | */ | |
34af946a | 59 | static DEFINE_SPINLOCK(o2hb_live_lock); |
a7f6a5fb MF |
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); | |
64 | ||
8ca8b0bb SM |
65 | #define O2HB_DB_TYPE_LIVENODES 0 |
66 | struct o2hb_debug_buf { | |
67 | int db_type; | |
68 | int db_size; | |
69 | int db_len; | |
70 | void *db_data; | |
71 | }; | |
72 | ||
73 | static struct o2hb_debug_buf *o2hb_db_livenodes; | |
74 | ||
87d3d3f3 SM |
75 | #define O2HB_DEBUG_DIR "o2hb" |
76 | #define O2HB_DEBUG_LIVENODES "livenodes" | |
8ca8b0bb | 77 | |
87d3d3f3 SM |
78 | static struct dentry *o2hb_debug_dir; |
79 | static struct dentry *o2hb_debug_livenodes; | |
80 | ||
a7f6a5fb MF |
81 | static LIST_HEAD(o2hb_all_regions); |
82 | ||
83 | static struct o2hb_callback { | |
84 | struct list_head list; | |
85 | } o2hb_callbacks[O2HB_NUM_CB]; | |
86 | ||
87 | static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type); | |
88 | ||
89 | #define O2HB_DEFAULT_BLOCK_BITS 9 | |
90 | ||
54b5187b SM |
91 | enum o2hb_heartbeat_modes { |
92 | O2HB_HEARTBEAT_LOCAL = 0, | |
93 | O2HB_HEARTBEAT_GLOBAL, | |
94 | O2HB_HEARTBEAT_NUM_MODES, | |
95 | }; | |
96 | ||
97 | char *o2hb_heartbeat_mode_desc[O2HB_HEARTBEAT_NUM_MODES] = { | |
98 | "local", /* O2HB_HEARTBEAT_LOCAL */ | |
99 | "global", /* O2HB_HEARTBEAT_GLOBAL */ | |
100 | }; | |
101 | ||
a7f6a5fb | 102 | unsigned int o2hb_dead_threshold = O2HB_DEFAULT_DEAD_THRESHOLD; |
54b5187b | 103 | unsigned int o2hb_heartbeat_mode = O2HB_HEARTBEAT_LOCAL; |
a7f6a5fb | 104 | |
2bd63216 | 105 | /* Only sets a new threshold if there are no active regions. |
a7f6a5fb MF |
106 | * |
107 | * No locking or otherwise interesting code is required for reading | |
108 | * o2hb_dead_threshold as it can't change once regions are active and | |
109 | * it's not interesting to anyone until then anyway. */ | |
110 | static void o2hb_dead_threshold_set(unsigned int threshold) | |
111 | { | |
112 | if (threshold > O2HB_MIN_DEAD_THRESHOLD) { | |
113 | spin_lock(&o2hb_live_lock); | |
114 | if (list_empty(&o2hb_all_regions)) | |
115 | o2hb_dead_threshold = threshold; | |
116 | spin_unlock(&o2hb_live_lock); | |
117 | } | |
118 | } | |
119 | ||
54b5187b SM |
120 | static int o2hb_global_hearbeat_mode_set(unsigned int hb_mode) |
121 | { | |
122 | int ret = -1; | |
123 | ||
124 | if (hb_mode < O2HB_HEARTBEAT_NUM_MODES) { | |
125 | spin_lock(&o2hb_live_lock); | |
126 | if (list_empty(&o2hb_all_regions)) { | |
127 | o2hb_heartbeat_mode = hb_mode; | |
128 | ret = 0; | |
129 | } | |
130 | spin_unlock(&o2hb_live_lock); | |
131 | } | |
132 | ||
133 | return ret; | |
134 | } | |
135 | ||
a7f6a5fb MF |
136 | struct o2hb_node_event { |
137 | struct list_head hn_item; | |
138 | enum o2hb_callback_type hn_event_type; | |
139 | struct o2nm_node *hn_node; | |
140 | int hn_node_num; | |
141 | }; | |
142 | ||
143 | struct o2hb_disk_slot { | |
144 | struct o2hb_disk_heartbeat_block *ds_raw_block; | |
145 | u8 ds_node_num; | |
146 | u64 ds_last_time; | |
147 | u64 ds_last_generation; | |
148 | u16 ds_equal_samples; | |
149 | u16 ds_changed_samples; | |
150 | struct list_head ds_live_item; | |
151 | }; | |
152 | ||
153 | /* each thread owns a region.. when we're asked to tear down the region | |
154 | * we ask the thread to stop, who cleans up the region */ | |
155 | struct o2hb_region { | |
156 | struct config_item hr_item; | |
157 | ||
158 | struct list_head hr_all_item; | |
159 | unsigned hr_unclean_stop:1; | |
160 | ||
161 | /* protected by the hr_callback_sem */ | |
162 | struct task_struct *hr_task; | |
163 | ||
164 | unsigned int hr_blocks; | |
165 | unsigned long long hr_start_block; | |
166 | ||
167 | unsigned int hr_block_bits; | |
168 | unsigned int hr_block_bytes; | |
169 | ||
170 | unsigned int hr_slots_per_page; | |
171 | unsigned int hr_num_pages; | |
172 | ||
173 | struct page **hr_slot_data; | |
174 | struct block_device *hr_bdev; | |
175 | struct o2hb_disk_slot *hr_slots; | |
176 | ||
177 | /* let the person setting up hb wait for it to return until it | |
178 | * has reached a 'steady' state. This will be fixed when we have | |
179 | * a more complete api that doesn't lead to this sort of fragility. */ | |
180 | atomic_t hr_steady_iterations; | |
181 | ||
182 | char hr_dev_name[BDEVNAME_SIZE]; | |
183 | ||
184 | unsigned int hr_timeout_ms; | |
185 | ||
186 | /* randomized as the region goes up and down so that a node | |
187 | * recognizes a node going up and down in one iteration */ | |
188 | u64 hr_generation; | |
189 | ||
c4028958 | 190 | struct delayed_work hr_write_timeout_work; |
a7f6a5fb MF |
191 | unsigned long hr_last_timeout_start; |
192 | ||
193 | /* Used during o2hb_check_slot to hold a copy of the block | |
194 | * being checked because we temporarily have to zero out the | |
195 | * crc field. */ | |
196 | struct o2hb_disk_heartbeat_block *hr_tmp_block; | |
197 | }; | |
198 | ||
199 | struct o2hb_bio_wait_ctxt { | |
200 | atomic_t wc_num_reqs; | |
201 | struct completion wc_io_complete; | |
a9e2ae39 | 202 | int wc_error; |
a7f6a5fb MF |
203 | }; |
204 | ||
c4028958 | 205 | static void o2hb_write_timeout(struct work_struct *work) |
a7f6a5fb | 206 | { |
c4028958 DH |
207 | struct o2hb_region *reg = |
208 | container_of(work, struct o2hb_region, | |
209 | hr_write_timeout_work.work); | |
a7f6a5fb MF |
210 | |
211 | mlog(ML_ERROR, "Heartbeat write timeout to device %s after %u " | |
212 | "milliseconds\n", reg->hr_dev_name, | |
2bd63216 | 213 | jiffies_to_msecs(jiffies - reg->hr_last_timeout_start)); |
a7f6a5fb MF |
214 | o2quo_disk_timeout(); |
215 | } | |
216 | ||
217 | static void o2hb_arm_write_timeout(struct o2hb_region *reg) | |
218 | { | |
b31d308d TM |
219 | mlog(ML_HEARTBEAT, "Queue write timeout for %u ms\n", |
220 | O2HB_MAX_WRITE_TIMEOUT_MS); | |
a7f6a5fb MF |
221 | |
222 | cancel_delayed_work(®->hr_write_timeout_work); | |
223 | reg->hr_last_timeout_start = jiffies; | |
224 | schedule_delayed_work(®->hr_write_timeout_work, | |
225 | msecs_to_jiffies(O2HB_MAX_WRITE_TIMEOUT_MS)); | |
226 | } | |
227 | ||
228 | static void o2hb_disarm_write_timeout(struct o2hb_region *reg) | |
229 | { | |
230 | cancel_delayed_work(®->hr_write_timeout_work); | |
231 | flush_scheduled_work(); | |
232 | } | |
233 | ||
b559292e | 234 | static inline void o2hb_bio_wait_init(struct o2hb_bio_wait_ctxt *wc) |
a7f6a5fb | 235 | { |
b559292e | 236 | atomic_set(&wc->wc_num_reqs, 1); |
a7f6a5fb | 237 | init_completion(&wc->wc_io_complete); |
a9e2ae39 | 238 | wc->wc_error = 0; |
a7f6a5fb MF |
239 | } |
240 | ||
241 | /* Used in error paths too */ | |
242 | static inline void o2hb_bio_wait_dec(struct o2hb_bio_wait_ctxt *wc, | |
243 | unsigned int num) | |
244 | { | |
245 | /* sadly atomic_sub_and_test() isn't available on all platforms. The | |
246 | * good news is that the fast path only completes one at a time */ | |
247 | while(num--) { | |
248 | if (atomic_dec_and_test(&wc->wc_num_reqs)) { | |
249 | BUG_ON(num > 0); | |
250 | complete(&wc->wc_io_complete); | |
251 | } | |
252 | } | |
253 | } | |
254 | ||
255 | static void o2hb_wait_on_io(struct o2hb_region *reg, | |
256 | struct o2hb_bio_wait_ctxt *wc) | |
257 | { | |
258 | struct address_space *mapping = reg->hr_bdev->bd_inode->i_mapping; | |
259 | ||
260 | blk_run_address_space(mapping); | |
b559292e | 261 | o2hb_bio_wait_dec(wc, 1); |
a7f6a5fb MF |
262 | |
263 | wait_for_completion(&wc->wc_io_complete); | |
264 | } | |
265 | ||
782e3b3b | 266 | static void o2hb_bio_end_io(struct bio *bio, |
a7f6a5fb MF |
267 | int error) |
268 | { | |
269 | struct o2hb_bio_wait_ctxt *wc = bio->bi_private; | |
270 | ||
a9e2ae39 | 271 | if (error) { |
a7f6a5fb | 272 | mlog(ML_ERROR, "IO Error %d\n", error); |
a9e2ae39 MF |
273 | wc->wc_error = error; |
274 | } | |
a7f6a5fb | 275 | |
a7f6a5fb | 276 | o2hb_bio_wait_dec(wc, 1); |
b559292e | 277 | bio_put(bio); |
a7f6a5fb MF |
278 | } |
279 | ||
280 | /* Setup a Bio to cover I/O against num_slots slots starting at | |
281 | * start_slot. */ | |
282 | static struct bio *o2hb_setup_one_bio(struct o2hb_region *reg, | |
283 | struct o2hb_bio_wait_ctxt *wc, | |
b559292e PR |
284 | unsigned int *current_slot, |
285 | unsigned int max_slots) | |
a7f6a5fb | 286 | { |
b559292e | 287 | int len, current_page; |
a7f6a5fb MF |
288 | unsigned int vec_len, vec_start; |
289 | unsigned int bits = reg->hr_block_bits; | |
290 | unsigned int spp = reg->hr_slots_per_page; | |
b559292e | 291 | unsigned int cs = *current_slot; |
a7f6a5fb MF |
292 | struct bio *bio; |
293 | struct page *page; | |
294 | ||
a7f6a5fb MF |
295 | /* Testing has shown this allocation to take long enough under |
296 | * GFP_KERNEL that the local node can get fenced. It would be | |
297 | * nicest if we could pre-allocate these bios and avoid this | |
298 | * all together. */ | |
b559292e | 299 | bio = bio_alloc(GFP_ATOMIC, 16); |
a7f6a5fb MF |
300 | if (!bio) { |
301 | mlog(ML_ERROR, "Could not alloc slots BIO!\n"); | |
302 | bio = ERR_PTR(-ENOMEM); | |
303 | goto bail; | |
304 | } | |
305 | ||
306 | /* Must put everything in 512 byte sectors for the bio... */ | |
b559292e | 307 | bio->bi_sector = (reg->hr_start_block + cs) << (bits - 9); |
a7f6a5fb MF |
308 | bio->bi_bdev = reg->hr_bdev; |
309 | bio->bi_private = wc; | |
310 | bio->bi_end_io = o2hb_bio_end_io; | |
311 | ||
b559292e PR |
312 | vec_start = (cs << bits) % PAGE_CACHE_SIZE; |
313 | while(cs < max_slots) { | |
314 | current_page = cs / spp; | |
315 | page = reg->hr_slot_data[current_page]; | |
a7f6a5fb | 316 | |
bc7e97cb | 317 | vec_len = min(PAGE_CACHE_SIZE - vec_start, |
b559292e | 318 | (max_slots-cs) * (PAGE_CACHE_SIZE/spp) ); |
a7f6a5fb MF |
319 | |
320 | mlog(ML_HB_BIO, "page %d, vec_len = %u, vec_start = %u\n", | |
b559292e | 321 | current_page, vec_len, vec_start); |
a7f6a5fb MF |
322 | |
323 | len = bio_add_page(bio, page, vec_len, vec_start); | |
b559292e | 324 | if (len != vec_len) break; |
a7f6a5fb | 325 | |
b559292e | 326 | cs += vec_len / (PAGE_CACHE_SIZE/spp); |
a7f6a5fb MF |
327 | vec_start = 0; |
328 | } | |
329 | ||
330 | bail: | |
b559292e | 331 | *current_slot = cs; |
a7f6a5fb MF |
332 | return bio; |
333 | } | |
334 | ||
a7f6a5fb MF |
335 | static int o2hb_read_slots(struct o2hb_region *reg, |
336 | unsigned int max_slots) | |
337 | { | |
b559292e PR |
338 | unsigned int current_slot=0; |
339 | int status; | |
a7f6a5fb | 340 | struct o2hb_bio_wait_ctxt wc; |
a7f6a5fb MF |
341 | struct bio *bio; |
342 | ||
b559292e | 343 | o2hb_bio_wait_init(&wc); |
a7f6a5fb | 344 | |
b559292e PR |
345 | while(current_slot < max_slots) { |
346 | bio = o2hb_setup_one_bio(reg, &wc, ¤t_slot, max_slots); | |
a7f6a5fb | 347 | if (IS_ERR(bio)) { |
a7f6a5fb MF |
348 | status = PTR_ERR(bio); |
349 | mlog_errno(status); | |
350 | goto bail_and_wait; | |
351 | } | |
a7f6a5fb | 352 | |
b559292e | 353 | atomic_inc(&wc.wc_num_reqs); |
a7f6a5fb MF |
354 | submit_bio(READ, bio); |
355 | } | |
356 | ||
357 | status = 0; | |
358 | ||
359 | bail_and_wait: | |
360 | o2hb_wait_on_io(reg, &wc); | |
a9e2ae39 MF |
361 | if (wc.wc_error && !status) |
362 | status = wc.wc_error; | |
a7f6a5fb | 363 | |
a7f6a5fb MF |
364 | return status; |
365 | } | |
366 | ||
367 | static int o2hb_issue_node_write(struct o2hb_region *reg, | |
a7f6a5fb MF |
368 | struct o2hb_bio_wait_ctxt *write_wc) |
369 | { | |
370 | int status; | |
371 | unsigned int slot; | |
372 | struct bio *bio; | |
373 | ||
b559292e | 374 | o2hb_bio_wait_init(write_wc); |
a7f6a5fb MF |
375 | |
376 | slot = o2nm_this_node(); | |
377 | ||
b559292e | 378 | bio = o2hb_setup_one_bio(reg, write_wc, &slot, slot+1); |
a7f6a5fb MF |
379 | if (IS_ERR(bio)) { |
380 | status = PTR_ERR(bio); | |
381 | mlog_errno(status); | |
382 | goto bail; | |
383 | } | |
384 | ||
b559292e | 385 | atomic_inc(&write_wc->wc_num_reqs); |
a7f6a5fb MF |
386 | submit_bio(WRITE, bio); |
387 | ||
a7f6a5fb MF |
388 | status = 0; |
389 | bail: | |
390 | return status; | |
391 | } | |
392 | ||
393 | static u32 o2hb_compute_block_crc_le(struct o2hb_region *reg, | |
394 | struct o2hb_disk_heartbeat_block *hb_block) | |
395 | { | |
396 | __le32 old_cksum; | |
397 | u32 ret; | |
398 | ||
399 | /* We want to compute the block crc with a 0 value in the | |
400 | * hb_cksum field. Save it off here and replace after the | |
401 | * crc. */ | |
402 | old_cksum = hb_block->hb_cksum; | |
403 | hb_block->hb_cksum = 0; | |
404 | ||
405 | ret = crc32_le(0, (unsigned char *) hb_block, reg->hr_block_bytes); | |
406 | ||
407 | hb_block->hb_cksum = old_cksum; | |
408 | ||
409 | return ret; | |
410 | } | |
411 | ||
412 | static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block *hb_block) | |
413 | { | |
70bacbdb MF |
414 | mlog(ML_ERROR, "Dump slot information: seq = 0x%llx, node = %u, " |
415 | "cksum = 0x%x, generation 0x%llx\n", | |
416 | (long long)le64_to_cpu(hb_block->hb_seq), | |
417 | hb_block->hb_node, le32_to_cpu(hb_block->hb_cksum), | |
418 | (long long)le64_to_cpu(hb_block->hb_generation)); | |
a7f6a5fb MF |
419 | } |
420 | ||
421 | static int o2hb_verify_crc(struct o2hb_region *reg, | |
422 | struct o2hb_disk_heartbeat_block *hb_block) | |
423 | { | |
424 | u32 read, computed; | |
425 | ||
426 | read = le32_to_cpu(hb_block->hb_cksum); | |
427 | computed = o2hb_compute_block_crc_le(reg, hb_block); | |
428 | ||
429 | return read == computed; | |
430 | } | |
431 | ||
432 | /* We want to make sure that nobody is heartbeating on top of us -- | |
433 | * this will help detect an invalid configuration. */ | |
434 | static int o2hb_check_last_timestamp(struct o2hb_region *reg) | |
435 | { | |
436 | int node_num, ret; | |
437 | struct o2hb_disk_slot *slot; | |
438 | struct o2hb_disk_heartbeat_block *hb_block; | |
439 | ||
440 | node_num = o2nm_this_node(); | |
441 | ||
442 | ret = 1; | |
443 | slot = ®->hr_slots[node_num]; | |
444 | /* Don't check on our 1st timestamp */ | |
445 | if (slot->ds_last_time) { | |
446 | hb_block = slot->ds_raw_block; | |
447 | ||
448 | if (le64_to_cpu(hb_block->hb_seq) != slot->ds_last_time) | |
449 | ret = 0; | |
450 | } | |
451 | ||
452 | return ret; | |
453 | } | |
454 | ||
455 | static inline void o2hb_prepare_block(struct o2hb_region *reg, | |
456 | u64 generation) | |
457 | { | |
458 | int node_num; | |
459 | u64 cputime; | |
460 | struct o2hb_disk_slot *slot; | |
461 | struct o2hb_disk_heartbeat_block *hb_block; | |
462 | ||
463 | node_num = o2nm_this_node(); | |
464 | slot = ®->hr_slots[node_num]; | |
465 | ||
466 | hb_block = (struct o2hb_disk_heartbeat_block *)slot->ds_raw_block; | |
467 | memset(hb_block, 0, reg->hr_block_bytes); | |
468 | /* TODO: time stuff */ | |
469 | cputime = CURRENT_TIME.tv_sec; | |
470 | if (!cputime) | |
471 | cputime = 1; | |
472 | ||
473 | hb_block->hb_seq = cpu_to_le64(cputime); | |
474 | hb_block->hb_node = node_num; | |
475 | hb_block->hb_generation = cpu_to_le64(generation); | |
0db638f4 | 476 | hb_block->hb_dead_ms = cpu_to_le32(o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS); |
a7f6a5fb MF |
477 | |
478 | /* This step must always happen last! */ | |
479 | hb_block->hb_cksum = cpu_to_le32(o2hb_compute_block_crc_le(reg, | |
480 | hb_block)); | |
481 | ||
70bacbdb | 482 | mlog(ML_HB_BIO, "our node generation = 0x%llx, cksum = 0x%x\n", |
5fdf1e67 | 483 | (long long)generation, |
70bacbdb | 484 | le32_to_cpu(hb_block->hb_cksum)); |
a7f6a5fb MF |
485 | } |
486 | ||
487 | static void o2hb_fire_callbacks(struct o2hb_callback *hbcall, | |
488 | struct o2nm_node *node, | |
489 | int idx) | |
490 | { | |
491 | struct list_head *iter; | |
492 | struct o2hb_callback_func *f; | |
493 | ||
494 | list_for_each(iter, &hbcall->list) { | |
495 | f = list_entry(iter, struct o2hb_callback_func, hc_item); | |
496 | mlog(ML_HEARTBEAT, "calling funcs %p\n", f); | |
497 | (f->hc_func)(node, idx, f->hc_data); | |
498 | } | |
499 | } | |
500 | ||
501 | /* Will run the list in order until we process the passed event */ | |
502 | static void o2hb_run_event_list(struct o2hb_node_event *queued_event) | |
503 | { | |
504 | int empty; | |
505 | struct o2hb_callback *hbcall; | |
506 | struct o2hb_node_event *event; | |
507 | ||
508 | spin_lock(&o2hb_live_lock); | |
509 | empty = list_empty(&queued_event->hn_item); | |
510 | spin_unlock(&o2hb_live_lock); | |
511 | if (empty) | |
512 | return; | |
513 | ||
514 | /* Holding callback sem assures we don't alter the callback | |
515 | * lists when doing this, and serializes ourselves with other | |
516 | * processes wanting callbacks. */ | |
517 | down_write(&o2hb_callback_sem); | |
518 | ||
519 | spin_lock(&o2hb_live_lock); | |
520 | while (!list_empty(&o2hb_node_events) | |
521 | && !list_empty(&queued_event->hn_item)) { | |
522 | event = list_entry(o2hb_node_events.next, | |
523 | struct o2hb_node_event, | |
524 | hn_item); | |
525 | list_del_init(&event->hn_item); | |
526 | spin_unlock(&o2hb_live_lock); | |
527 | ||
528 | mlog(ML_HEARTBEAT, "Node %s event for %d\n", | |
529 | event->hn_event_type == O2HB_NODE_UP_CB ? "UP" : "DOWN", | |
530 | event->hn_node_num); | |
531 | ||
532 | hbcall = hbcall_from_type(event->hn_event_type); | |
533 | ||
534 | /* We should *never* have gotten on to the list with a | |
535 | * bad type... This isn't something that we should try | |
536 | * to recover from. */ | |
537 | BUG_ON(IS_ERR(hbcall)); | |
538 | ||
539 | o2hb_fire_callbacks(hbcall, event->hn_node, event->hn_node_num); | |
540 | ||
541 | spin_lock(&o2hb_live_lock); | |
542 | } | |
543 | spin_unlock(&o2hb_live_lock); | |
544 | ||
545 | up_write(&o2hb_callback_sem); | |
546 | } | |
547 | ||
548 | static void o2hb_queue_node_event(struct o2hb_node_event *event, | |
549 | enum o2hb_callback_type type, | |
550 | struct o2nm_node *node, | |
551 | int node_num) | |
552 | { | |
553 | assert_spin_locked(&o2hb_live_lock); | |
554 | ||
0e105d37 SM |
555 | BUG_ON((!node) && (type != O2HB_NODE_DOWN_CB)); |
556 | ||
a7f6a5fb MF |
557 | event->hn_event_type = type; |
558 | event->hn_node = node; | |
559 | event->hn_node_num = node_num; | |
560 | ||
561 | mlog(ML_HEARTBEAT, "Queue node %s event for node %d\n", | |
562 | type == O2HB_NODE_UP_CB ? "UP" : "DOWN", node_num); | |
563 | ||
564 | list_add_tail(&event->hn_item, &o2hb_node_events); | |
565 | } | |
566 | ||
567 | static void o2hb_shutdown_slot(struct o2hb_disk_slot *slot) | |
568 | { | |
569 | struct o2hb_node_event event = | |
570 | { .hn_item = LIST_HEAD_INIT(event.hn_item), }; | |
571 | struct o2nm_node *node; | |
572 | ||
573 | node = o2nm_get_node_by_num(slot->ds_node_num); | |
574 | if (!node) | |
575 | return; | |
576 | ||
577 | spin_lock(&o2hb_live_lock); | |
578 | if (!list_empty(&slot->ds_live_item)) { | |
579 | mlog(ML_HEARTBEAT, "Shutdown, node %d leaves region\n", | |
580 | slot->ds_node_num); | |
581 | ||
582 | list_del_init(&slot->ds_live_item); | |
583 | ||
584 | if (list_empty(&o2hb_live_slots[slot->ds_node_num])) { | |
585 | clear_bit(slot->ds_node_num, o2hb_live_node_bitmap); | |
586 | ||
587 | o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node, | |
588 | slot->ds_node_num); | |
589 | } | |
590 | } | |
591 | spin_unlock(&o2hb_live_lock); | |
592 | ||
593 | o2hb_run_event_list(&event); | |
594 | ||
595 | o2nm_node_put(node); | |
596 | } | |
597 | ||
598 | static int o2hb_check_slot(struct o2hb_region *reg, | |
599 | struct o2hb_disk_slot *slot) | |
600 | { | |
601 | int changed = 0, gen_changed = 0; | |
602 | struct o2hb_node_event event = | |
603 | { .hn_item = LIST_HEAD_INIT(event.hn_item), }; | |
604 | struct o2nm_node *node; | |
605 | struct o2hb_disk_heartbeat_block *hb_block = reg->hr_tmp_block; | |
606 | u64 cputime; | |
0db638f4 MF |
607 | unsigned int dead_ms = o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS; |
608 | unsigned int slot_dead_ms; | |
0e105d37 | 609 | int tmp; |
a7f6a5fb MF |
610 | |
611 | memcpy(hb_block, slot->ds_raw_block, reg->hr_block_bytes); | |
612 | ||
0e105d37 SM |
613 | /* |
614 | * If a node is no longer configured but is still in the livemap, we | |
615 | * may need to clear that bit from the livemap. | |
616 | */ | |
a7f6a5fb | 617 | node = o2nm_get_node_by_num(slot->ds_node_num); |
0e105d37 SM |
618 | if (!node) { |
619 | spin_lock(&o2hb_live_lock); | |
620 | tmp = test_bit(slot->ds_node_num, o2hb_live_node_bitmap); | |
621 | spin_unlock(&o2hb_live_lock); | |
622 | if (!tmp) | |
623 | return 0; | |
624 | } | |
a7f6a5fb MF |
625 | |
626 | if (!o2hb_verify_crc(reg, hb_block)) { | |
627 | /* all paths from here will drop o2hb_live_lock for | |
628 | * us. */ | |
629 | spin_lock(&o2hb_live_lock); | |
630 | ||
631 | /* Don't print an error on the console in this case - | |
632 | * a freshly formatted heartbeat area will not have a | |
633 | * crc set on it. */ | |
634 | if (list_empty(&slot->ds_live_item)) | |
635 | goto out; | |
636 | ||
637 | /* The node is live but pushed out a bad crc. We | |
638 | * consider it a transient miss but don't populate any | |
639 | * other values as they may be junk. */ | |
640 | mlog(ML_ERROR, "Node %d has written a bad crc to %s\n", | |
641 | slot->ds_node_num, reg->hr_dev_name); | |
642 | o2hb_dump_slot(hb_block); | |
643 | ||
644 | slot->ds_equal_samples++; | |
645 | goto fire_callbacks; | |
646 | } | |
647 | ||
648 | /* we don't care if these wrap.. the state transitions below | |
649 | * clear at the right places */ | |
650 | cputime = le64_to_cpu(hb_block->hb_seq); | |
651 | if (slot->ds_last_time != cputime) | |
652 | slot->ds_changed_samples++; | |
653 | else | |
654 | slot->ds_equal_samples++; | |
655 | slot->ds_last_time = cputime; | |
656 | ||
657 | /* The node changed heartbeat generations. We assume this to | |
658 | * mean it dropped off but came back before we timed out. We | |
659 | * want to consider it down for the time being but don't want | |
660 | * to lose any changed_samples state we might build up to | |
661 | * considering it live again. */ | |
662 | if (slot->ds_last_generation != le64_to_cpu(hb_block->hb_generation)) { | |
663 | gen_changed = 1; | |
664 | slot->ds_equal_samples = 0; | |
70bacbdb MF |
665 | mlog(ML_HEARTBEAT, "Node %d changed generation (0x%llx " |
666 | "to 0x%llx)\n", slot->ds_node_num, | |
667 | (long long)slot->ds_last_generation, | |
668 | (long long)le64_to_cpu(hb_block->hb_generation)); | |
a7f6a5fb MF |
669 | } |
670 | ||
671 | slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation); | |
672 | ||
70bacbdb MF |
673 | mlog(ML_HEARTBEAT, "Slot %d gen 0x%llx cksum 0x%x " |
674 | "seq %llu last %llu changed %u equal %u\n", | |
675 | slot->ds_node_num, (long long)slot->ds_last_generation, | |
676 | le32_to_cpu(hb_block->hb_cksum), | |
2bd63216 | 677 | (unsigned long long)le64_to_cpu(hb_block->hb_seq), |
70bacbdb | 678 | (unsigned long long)slot->ds_last_time, slot->ds_changed_samples, |
a7f6a5fb MF |
679 | slot->ds_equal_samples); |
680 | ||
681 | spin_lock(&o2hb_live_lock); | |
682 | ||
683 | fire_callbacks: | |
684 | /* dead nodes only come to life after some number of | |
685 | * changes at any time during their dead time */ | |
686 | if (list_empty(&slot->ds_live_item) && | |
687 | slot->ds_changed_samples >= O2HB_LIVE_THRESHOLD) { | |
70bacbdb MF |
688 | mlog(ML_HEARTBEAT, "Node %d (id 0x%llx) joined my region\n", |
689 | slot->ds_node_num, (long long)slot->ds_last_generation); | |
a7f6a5fb MF |
690 | |
691 | /* first on the list generates a callback */ | |
692 | if (list_empty(&o2hb_live_slots[slot->ds_node_num])) { | |
693 | set_bit(slot->ds_node_num, o2hb_live_node_bitmap); | |
694 | ||
695 | o2hb_queue_node_event(&event, O2HB_NODE_UP_CB, node, | |
696 | slot->ds_node_num); | |
697 | ||
698 | changed = 1; | |
699 | } | |
700 | ||
701 | list_add_tail(&slot->ds_live_item, | |
702 | &o2hb_live_slots[slot->ds_node_num]); | |
703 | ||
704 | slot->ds_equal_samples = 0; | |
0db638f4 MF |
705 | |
706 | /* We want to be sure that all nodes agree on the | |
707 | * number of milliseconds before a node will be | |
708 | * considered dead. The self-fencing timeout is | |
709 | * computed from this value, and a discrepancy might | |
710 | * result in heartbeat calling a node dead when it | |
711 | * hasn't self-fenced yet. */ | |
712 | slot_dead_ms = le32_to_cpu(hb_block->hb_dead_ms); | |
713 | if (slot_dead_ms && slot_dead_ms != dead_ms) { | |
714 | /* TODO: Perhaps we can fail the region here. */ | |
715 | mlog(ML_ERROR, "Node %d on device %s has a dead count " | |
716 | "of %u ms, but our count is %u ms.\n" | |
717 | "Please double check your configuration values " | |
718 | "for 'O2CB_HEARTBEAT_THRESHOLD'\n", | |
719 | slot->ds_node_num, reg->hr_dev_name, slot_dead_ms, | |
720 | dead_ms); | |
721 | } | |
a7f6a5fb MF |
722 | goto out; |
723 | } | |
724 | ||
725 | /* if the list is dead, we're done.. */ | |
726 | if (list_empty(&slot->ds_live_item)) | |
727 | goto out; | |
728 | ||
729 | /* live nodes only go dead after enough consequtive missed | |
730 | * samples.. reset the missed counter whenever we see | |
731 | * activity */ | |
732 | if (slot->ds_equal_samples >= o2hb_dead_threshold || gen_changed) { | |
733 | mlog(ML_HEARTBEAT, "Node %d left my region\n", | |
734 | slot->ds_node_num); | |
735 | ||
736 | /* last off the live_slot generates a callback */ | |
737 | list_del_init(&slot->ds_live_item); | |
738 | if (list_empty(&o2hb_live_slots[slot->ds_node_num])) { | |
739 | clear_bit(slot->ds_node_num, o2hb_live_node_bitmap); | |
740 | ||
0e105d37 SM |
741 | /* node can be null */ |
742 | o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, | |
743 | node, slot->ds_node_num); | |
a7f6a5fb MF |
744 | |
745 | changed = 1; | |
746 | } | |
747 | ||
748 | /* We don't clear this because the node is still | |
749 | * actually writing new blocks. */ | |
750 | if (!gen_changed) | |
751 | slot->ds_changed_samples = 0; | |
752 | goto out; | |
753 | } | |
754 | if (slot->ds_changed_samples) { | |
755 | slot->ds_changed_samples = 0; | |
756 | slot->ds_equal_samples = 0; | |
757 | } | |
758 | out: | |
759 | spin_unlock(&o2hb_live_lock); | |
760 | ||
761 | o2hb_run_event_list(&event); | |
762 | ||
0e105d37 SM |
763 | if (node) |
764 | o2nm_node_put(node); | |
a7f6a5fb MF |
765 | return changed; |
766 | } | |
767 | ||
768 | /* This could be faster if we just implmented a find_last_bit, but I | |
769 | * don't think the circumstances warrant it. */ | |
770 | static int o2hb_highest_node(unsigned long *nodes, | |
771 | int numbits) | |
772 | { | |
773 | int highest, node; | |
774 | ||
775 | highest = numbits; | |
776 | node = -1; | |
777 | while ((node = find_next_bit(nodes, numbits, node + 1)) != -1) { | |
778 | if (node >= numbits) | |
779 | break; | |
780 | ||
781 | highest = node; | |
782 | } | |
783 | ||
784 | return highest; | |
785 | } | |
786 | ||
a9e2ae39 | 787 | static int o2hb_do_disk_heartbeat(struct o2hb_region *reg) |
a7f6a5fb MF |
788 | { |
789 | int i, ret, highest_node, change = 0; | |
790 | unsigned long configured_nodes[BITS_TO_LONGS(O2NM_MAX_NODES)]; | |
0e105d37 | 791 | unsigned long live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)]; |
a7f6a5fb MF |
792 | struct o2hb_bio_wait_ctxt write_wc; |
793 | ||
a9e2ae39 MF |
794 | ret = o2nm_configured_node_map(configured_nodes, |
795 | sizeof(configured_nodes)); | |
796 | if (ret) { | |
797 | mlog_errno(ret); | |
798 | return ret; | |
799 | } | |
a7f6a5fb | 800 | |
0e105d37 SM |
801 | /* |
802 | * If a node is not configured but is in the livemap, we still need | |
803 | * to read the slot so as to be able to remove it from the livemap. | |
804 | */ | |
805 | o2hb_fill_node_map(live_node_bitmap, sizeof(live_node_bitmap)); | |
806 | i = -1; | |
807 | while ((i = find_next_bit(live_node_bitmap, | |
808 | O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) { | |
809 | set_bit(i, configured_nodes); | |
810 | } | |
811 | ||
a7f6a5fb MF |
812 | highest_node = o2hb_highest_node(configured_nodes, O2NM_MAX_NODES); |
813 | if (highest_node >= O2NM_MAX_NODES) { | |
814 | mlog(ML_NOTICE, "ocfs2_heartbeat: no configured nodes found!\n"); | |
a9e2ae39 | 815 | return -EINVAL; |
a7f6a5fb MF |
816 | } |
817 | ||
818 | /* No sense in reading the slots of nodes that don't exist | |
819 | * yet. Of course, if the node definitions have holes in them | |
820 | * then we're reading an empty slot anyway... Consider this | |
821 | * best-effort. */ | |
822 | ret = o2hb_read_slots(reg, highest_node + 1); | |
823 | if (ret < 0) { | |
824 | mlog_errno(ret); | |
a9e2ae39 | 825 | return ret; |
a7f6a5fb MF |
826 | } |
827 | ||
828 | /* With an up to date view of the slots, we can check that no | |
829 | * other node has been improperly configured to heartbeat in | |
830 | * our slot. */ | |
831 | if (!o2hb_check_last_timestamp(reg)) | |
832 | mlog(ML_ERROR, "Device \"%s\": another node is heartbeating " | |
833 | "in our slot!\n", reg->hr_dev_name); | |
834 | ||
835 | /* fill in the proper info for our next heartbeat */ | |
836 | o2hb_prepare_block(reg, reg->hr_generation); | |
837 | ||
838 | /* And fire off the write. Note that we don't wait on this I/O | |
839 | * until later. */ | |
b559292e | 840 | ret = o2hb_issue_node_write(reg, &write_wc); |
a7f6a5fb MF |
841 | if (ret < 0) { |
842 | mlog_errno(ret); | |
a9e2ae39 | 843 | return ret; |
a7f6a5fb MF |
844 | } |
845 | ||
846 | i = -1; | |
847 | while((i = find_next_bit(configured_nodes, O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) { | |
848 | ||
849 | change |= o2hb_check_slot(reg, ®->hr_slots[i]); | |
850 | } | |
851 | ||
852 | /* | |
853 | * We have to be sure we've advertised ourselves on disk | |
854 | * before we can go to steady state. This ensures that | |
855 | * people we find in our steady state have seen us. | |
856 | */ | |
857 | o2hb_wait_on_io(reg, &write_wc); | |
a9e2ae39 MF |
858 | if (write_wc.wc_error) { |
859 | /* Do not re-arm the write timeout on I/O error - we | |
860 | * can't be sure that the new block ever made it to | |
861 | * disk */ | |
862 | mlog(ML_ERROR, "Write error %d on device \"%s\"\n", | |
863 | write_wc.wc_error, reg->hr_dev_name); | |
864 | return write_wc.wc_error; | |
865 | } | |
866 | ||
a7f6a5fb MF |
867 | o2hb_arm_write_timeout(reg); |
868 | ||
869 | /* let the person who launched us know when things are steady */ | |
870 | if (!change && (atomic_read(®->hr_steady_iterations) != 0)) { | |
871 | if (atomic_dec_and_test(®->hr_steady_iterations)) | |
872 | wake_up(&o2hb_steady_queue); | |
873 | } | |
a9e2ae39 MF |
874 | |
875 | return 0; | |
a7f6a5fb MF |
876 | } |
877 | ||
878 | /* Subtract b from a, storing the result in a. a *must* have a larger | |
879 | * value than b. */ | |
880 | static void o2hb_tv_subtract(struct timeval *a, | |
881 | struct timeval *b) | |
882 | { | |
883 | /* just return 0 when a is after b */ | |
884 | if (a->tv_sec < b->tv_sec || | |
885 | (a->tv_sec == b->tv_sec && a->tv_usec < b->tv_usec)) { | |
886 | a->tv_sec = 0; | |
887 | a->tv_usec = 0; | |
888 | return; | |
889 | } | |
890 | ||
891 | a->tv_sec -= b->tv_sec; | |
892 | a->tv_usec -= b->tv_usec; | |
893 | while ( a->tv_usec < 0 ) { | |
894 | a->tv_sec--; | |
895 | a->tv_usec += 1000000; | |
896 | } | |
897 | } | |
898 | ||
899 | static unsigned int o2hb_elapsed_msecs(struct timeval *start, | |
900 | struct timeval *end) | |
901 | { | |
902 | struct timeval res = *end; | |
903 | ||
904 | o2hb_tv_subtract(&res, start); | |
905 | ||
906 | return res.tv_sec * 1000 + res.tv_usec / 1000; | |
907 | } | |
908 | ||
909 | /* | |
910 | * we ride the region ref that the region dir holds. before the region | |
911 | * dir is removed and drops it ref it will wait to tear down this | |
912 | * thread. | |
913 | */ | |
914 | static int o2hb_thread(void *data) | |
915 | { | |
916 | int i, ret; | |
917 | struct o2hb_region *reg = data; | |
a7f6a5fb MF |
918 | struct o2hb_bio_wait_ctxt write_wc; |
919 | struct timeval before_hb, after_hb; | |
920 | unsigned int elapsed_msec; | |
921 | ||
922 | mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread running\n"); | |
923 | ||
924 | set_user_nice(current, -20); | |
925 | ||
926 | while (!kthread_should_stop() && !reg->hr_unclean_stop) { | |
927 | /* We track the time spent inside | |
025dfdaf | 928 | * o2hb_do_disk_heartbeat so that we avoid more than |
a7f6a5fb MF |
929 | * hr_timeout_ms between disk writes. On busy systems |
930 | * this should result in a heartbeat which is less | |
931 | * likely to time itself out. */ | |
932 | do_gettimeofday(&before_hb); | |
933 | ||
a9e2ae39 MF |
934 | i = 0; |
935 | do { | |
936 | ret = o2hb_do_disk_heartbeat(reg); | |
937 | } while (ret && ++i < 2); | |
a7f6a5fb MF |
938 | |
939 | do_gettimeofday(&after_hb); | |
940 | elapsed_msec = o2hb_elapsed_msecs(&before_hb, &after_hb); | |
941 | ||
b31d308d TM |
942 | mlog(ML_HEARTBEAT, |
943 | "start = %lu.%lu, end = %lu.%lu, msec = %u\n", | |
215c7f9f MF |
944 | before_hb.tv_sec, (unsigned long) before_hb.tv_usec, |
945 | after_hb.tv_sec, (unsigned long) after_hb.tv_usec, | |
946 | elapsed_msec); | |
a7f6a5fb MF |
947 | |
948 | if (elapsed_msec < reg->hr_timeout_ms) { | |
949 | /* the kthread api has blocked signals for us so no | |
950 | * need to record the return value. */ | |
951 | msleep_interruptible(reg->hr_timeout_ms - elapsed_msec); | |
952 | } | |
953 | } | |
954 | ||
955 | o2hb_disarm_write_timeout(reg); | |
956 | ||
957 | /* unclean stop is only used in very bad situation */ | |
958 | for(i = 0; !reg->hr_unclean_stop && i < reg->hr_blocks; i++) | |
959 | o2hb_shutdown_slot(®->hr_slots[i]); | |
960 | ||
961 | /* Explicit down notification - avoid forcing the other nodes | |
962 | * to timeout on this region when we could just as easily | |
963 | * write a clear generation - thus indicating to them that | |
964 | * this node has left this region. | |
965 | * | |
966 | * XXX: Should we skip this on unclean_stop? */ | |
967 | o2hb_prepare_block(reg, 0); | |
b559292e | 968 | ret = o2hb_issue_node_write(reg, &write_wc); |
a7f6a5fb MF |
969 | if (ret == 0) { |
970 | o2hb_wait_on_io(reg, &write_wc); | |
a7f6a5fb MF |
971 | } else { |
972 | mlog_errno(ret); | |
973 | } | |
974 | ||
975 | mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread exiting\n"); | |
976 | ||
977 | return 0; | |
978 | } | |
979 | ||
87d3d3f3 SM |
980 | #ifdef CONFIG_DEBUG_FS |
981 | static int o2hb_debug_open(struct inode *inode, struct file *file) | |
982 | { | |
8ca8b0bb | 983 | struct o2hb_debug_buf *db = inode->i_private; |
87d3d3f3 SM |
984 | unsigned long map[BITS_TO_LONGS(O2NM_MAX_NODES)]; |
985 | char *buf = NULL; | |
986 | int i = -1; | |
987 | int out = 0; | |
988 | ||
8ca8b0bb SM |
989 | /* max_nodes should be the largest bitmap we pass here */ |
990 | BUG_ON(sizeof(map) < db->db_size); | |
991 | ||
87d3d3f3 SM |
992 | buf = kmalloc(PAGE_SIZE, GFP_KERNEL); |
993 | if (!buf) | |
994 | goto bail; | |
995 | ||
8ca8b0bb SM |
996 | switch (db->db_type) { |
997 | case O2HB_DB_TYPE_LIVENODES: | |
998 | spin_lock(&o2hb_live_lock); | |
999 | memcpy(map, db->db_data, db->db_size); | |
1000 | spin_unlock(&o2hb_live_lock); | |
1001 | break; | |
87d3d3f3 | 1002 | |
8ca8b0bb SM |
1003 | default: |
1004 | goto done; | |
1005 | } | |
1006 | ||
1007 | while ((i = find_next_bit(map, db->db_len, i + 1)) < db->db_len) | |
87d3d3f3 SM |
1008 | out += snprintf(buf + out, PAGE_SIZE - out, "%d ", i); |
1009 | out += snprintf(buf + out, PAGE_SIZE - out, "\n"); | |
1010 | ||
8ca8b0bb | 1011 | done: |
87d3d3f3 SM |
1012 | i_size_write(inode, out); |
1013 | ||
1014 | file->private_data = buf; | |
1015 | ||
1016 | return 0; | |
1017 | bail: | |
1018 | return -ENOMEM; | |
1019 | } | |
1020 | ||
1021 | static int o2hb_debug_release(struct inode *inode, struct file *file) | |
1022 | { | |
1023 | kfree(file->private_data); | |
1024 | return 0; | |
1025 | } | |
1026 | ||
1027 | static ssize_t o2hb_debug_read(struct file *file, char __user *buf, | |
1028 | size_t nbytes, loff_t *ppos) | |
1029 | { | |
1030 | return simple_read_from_buffer(buf, nbytes, ppos, file->private_data, | |
1031 | i_size_read(file->f_mapping->host)); | |
1032 | } | |
1033 | #else | |
1034 | static int o2hb_debug_open(struct inode *inode, struct file *file) | |
1035 | { | |
1036 | return 0; | |
1037 | } | |
1038 | static int o2hb_debug_release(struct inode *inode, struct file *file) | |
1039 | { | |
1040 | return 0; | |
1041 | } | |
1042 | static ssize_t o2hb_debug_read(struct file *file, char __user *buf, | |
1043 | size_t nbytes, loff_t *ppos) | |
1044 | { | |
1045 | return 0; | |
1046 | } | |
1047 | #endif /* CONFIG_DEBUG_FS */ | |
1048 | ||
828c0950 | 1049 | static const struct file_operations o2hb_debug_fops = { |
87d3d3f3 SM |
1050 | .open = o2hb_debug_open, |
1051 | .release = o2hb_debug_release, | |
1052 | .read = o2hb_debug_read, | |
1053 | .llseek = generic_file_llseek, | |
1054 | }; | |
1055 | ||
1056 | void o2hb_exit(void) | |
1057 | { | |
8ca8b0bb SM |
1058 | kfree(o2hb_db_livenodes); |
1059 | debugfs_remove(o2hb_debug_livenodes); | |
1060 | debugfs_remove(o2hb_debug_dir); | |
1061 | } | |
1062 | ||
1063 | static struct dentry *o2hb_debug_create(const char *name, struct dentry *dir, | |
1064 | struct o2hb_debug_buf **db, int db_len, | |
1065 | int type, int size, int len, void *data) | |
1066 | { | |
1067 | *db = kmalloc(db_len, GFP_KERNEL); | |
1068 | if (!*db) | |
1069 | return NULL; | |
1070 | ||
1071 | (*db)->db_type = type; | |
1072 | (*db)->db_size = size; | |
1073 | (*db)->db_len = len; | |
1074 | (*db)->db_data = data; | |
1075 | ||
1076 | return debugfs_create_file(name, S_IFREG|S_IRUSR, dir, *db, | |
1077 | &o2hb_debug_fops); | |
1078 | } | |
1079 | ||
1080 | static int o2hb_debug_init(void) | |
1081 | { | |
1082 | int ret = -ENOMEM; | |
1083 | ||
1084 | o2hb_debug_dir = debugfs_create_dir(O2HB_DEBUG_DIR, NULL); | |
1085 | if (!o2hb_debug_dir) { | |
1086 | mlog_errno(ret); | |
1087 | goto bail; | |
1088 | } | |
1089 | ||
1090 | o2hb_debug_livenodes = o2hb_debug_create(O2HB_DEBUG_LIVENODES, | |
1091 | o2hb_debug_dir, | |
1092 | &o2hb_db_livenodes, | |
1093 | sizeof(*o2hb_db_livenodes), | |
1094 | O2HB_DB_TYPE_LIVENODES, | |
1095 | sizeof(o2hb_live_node_bitmap), | |
1096 | O2NM_MAX_NODES, | |
1097 | o2hb_live_node_bitmap); | |
1098 | if (!o2hb_debug_livenodes) { | |
1099 | mlog_errno(ret); | |
1100 | goto bail; | |
1101 | } | |
1102 | ret = 0; | |
1103 | bail: | |
1104 | if (ret) | |
1105 | o2hb_exit(); | |
1106 | ||
1107 | return ret; | |
87d3d3f3 SM |
1108 | } |
1109 | ||
1110 | int o2hb_init(void) | |
a7f6a5fb MF |
1111 | { |
1112 | int i; | |
1113 | ||
1114 | for (i = 0; i < ARRAY_SIZE(o2hb_callbacks); i++) | |
1115 | INIT_LIST_HEAD(&o2hb_callbacks[i].list); | |
1116 | ||
1117 | for (i = 0; i < ARRAY_SIZE(o2hb_live_slots); i++) | |
1118 | INIT_LIST_HEAD(&o2hb_live_slots[i]); | |
1119 | ||
1120 | INIT_LIST_HEAD(&o2hb_node_events); | |
1121 | ||
1122 | memset(o2hb_live_node_bitmap, 0, sizeof(o2hb_live_node_bitmap)); | |
87d3d3f3 | 1123 | |
8ca8b0bb | 1124 | return o2hb_debug_init(); |
a7f6a5fb MF |
1125 | } |
1126 | ||
1127 | /* if we're already in a callback then we're already serialized by the sem */ | |
1128 | static void o2hb_fill_node_map_from_callback(unsigned long *map, | |
1129 | unsigned bytes) | |
1130 | { | |
1131 | BUG_ON(bytes < (BITS_TO_LONGS(O2NM_MAX_NODES) * sizeof(unsigned long))); | |
1132 | ||
1133 | memcpy(map, &o2hb_live_node_bitmap, bytes); | |
1134 | } | |
1135 | ||
1136 | /* | |
1137 | * get a map of all nodes that are heartbeating in any regions | |
1138 | */ | |
1139 | void o2hb_fill_node_map(unsigned long *map, unsigned bytes) | |
1140 | { | |
1141 | /* callers want to serialize this map and callbacks so that they | |
1142 | * can trust that they don't miss nodes coming to the party */ | |
1143 | down_read(&o2hb_callback_sem); | |
1144 | spin_lock(&o2hb_live_lock); | |
1145 | o2hb_fill_node_map_from_callback(map, bytes); | |
1146 | spin_unlock(&o2hb_live_lock); | |
1147 | up_read(&o2hb_callback_sem); | |
1148 | } | |
1149 | EXPORT_SYMBOL_GPL(o2hb_fill_node_map); | |
1150 | ||
1151 | /* | |
1152 | * heartbeat configfs bits. The heartbeat set is a default set under | |
1153 | * the cluster set in nodemanager.c. | |
1154 | */ | |
1155 | ||
1156 | static struct o2hb_region *to_o2hb_region(struct config_item *item) | |
1157 | { | |
1158 | return item ? container_of(item, struct o2hb_region, hr_item) : NULL; | |
1159 | } | |
1160 | ||
1161 | /* drop_item only drops its ref after killing the thread, nothing should | |
1162 | * be using the region anymore. this has to clean up any state that | |
1163 | * attributes might have built up. */ | |
1164 | static void o2hb_region_release(struct config_item *item) | |
1165 | { | |
1166 | int i; | |
1167 | struct page *page; | |
1168 | struct o2hb_region *reg = to_o2hb_region(item); | |
1169 | ||
1170 | if (reg->hr_tmp_block) | |
1171 | kfree(reg->hr_tmp_block); | |
1172 | ||
1173 | if (reg->hr_slot_data) { | |
1174 | for (i = 0; i < reg->hr_num_pages; i++) { | |
1175 | page = reg->hr_slot_data[i]; | |
1176 | if (page) | |
1177 | __free_page(page); | |
1178 | } | |
1179 | kfree(reg->hr_slot_data); | |
1180 | } | |
1181 | ||
1182 | if (reg->hr_bdev) | |
9a1c3542 | 1183 | blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE); |
a7f6a5fb MF |
1184 | |
1185 | if (reg->hr_slots) | |
1186 | kfree(reg->hr_slots); | |
1187 | ||
1188 | spin_lock(&o2hb_live_lock); | |
1189 | list_del(®->hr_all_item); | |
1190 | spin_unlock(&o2hb_live_lock); | |
1191 | ||
1192 | kfree(reg); | |
1193 | } | |
1194 | ||
1195 | static int o2hb_read_block_input(struct o2hb_region *reg, | |
1196 | const char *page, | |
1197 | size_t count, | |
1198 | unsigned long *ret_bytes, | |
1199 | unsigned int *ret_bits) | |
1200 | { | |
1201 | unsigned long bytes; | |
1202 | char *p = (char *)page; | |
1203 | ||
1204 | bytes = simple_strtoul(p, &p, 0); | |
1205 | if (!p || (*p && (*p != '\n'))) | |
1206 | return -EINVAL; | |
1207 | ||
1208 | /* Heartbeat and fs min / max block sizes are the same. */ | |
1209 | if (bytes > 4096 || bytes < 512) | |
1210 | return -ERANGE; | |
1211 | if (hweight16(bytes) != 1) | |
1212 | return -EINVAL; | |
1213 | ||
1214 | if (ret_bytes) | |
1215 | *ret_bytes = bytes; | |
1216 | if (ret_bits) | |
1217 | *ret_bits = ffs(bytes) - 1; | |
1218 | ||
1219 | return 0; | |
1220 | } | |
1221 | ||
1222 | static ssize_t o2hb_region_block_bytes_read(struct o2hb_region *reg, | |
1223 | char *page) | |
1224 | { | |
1225 | return sprintf(page, "%u\n", reg->hr_block_bytes); | |
1226 | } | |
1227 | ||
1228 | static ssize_t o2hb_region_block_bytes_write(struct o2hb_region *reg, | |
1229 | const char *page, | |
1230 | size_t count) | |
1231 | { | |
1232 | int status; | |
1233 | unsigned long block_bytes; | |
1234 | unsigned int block_bits; | |
1235 | ||
1236 | if (reg->hr_bdev) | |
1237 | return -EINVAL; | |
1238 | ||
1239 | status = o2hb_read_block_input(reg, page, count, | |
1240 | &block_bytes, &block_bits); | |
1241 | if (status) | |
1242 | return status; | |
1243 | ||
1244 | reg->hr_block_bytes = (unsigned int)block_bytes; | |
1245 | reg->hr_block_bits = block_bits; | |
1246 | ||
1247 | return count; | |
1248 | } | |
1249 | ||
1250 | static ssize_t o2hb_region_start_block_read(struct o2hb_region *reg, | |
1251 | char *page) | |
1252 | { | |
1253 | return sprintf(page, "%llu\n", reg->hr_start_block); | |
1254 | } | |
1255 | ||
1256 | static ssize_t o2hb_region_start_block_write(struct o2hb_region *reg, | |
1257 | const char *page, | |
1258 | size_t count) | |
1259 | { | |
1260 | unsigned long long tmp; | |
1261 | char *p = (char *)page; | |
1262 | ||
1263 | if (reg->hr_bdev) | |
1264 | return -EINVAL; | |
1265 | ||
1266 | tmp = simple_strtoull(p, &p, 0); | |
1267 | if (!p || (*p && (*p != '\n'))) | |
1268 | return -EINVAL; | |
1269 | ||
1270 | reg->hr_start_block = tmp; | |
1271 | ||
1272 | return count; | |
1273 | } | |
1274 | ||
1275 | static ssize_t o2hb_region_blocks_read(struct o2hb_region *reg, | |
1276 | char *page) | |
1277 | { | |
1278 | return sprintf(page, "%d\n", reg->hr_blocks); | |
1279 | } | |
1280 | ||
1281 | static ssize_t o2hb_region_blocks_write(struct o2hb_region *reg, | |
1282 | const char *page, | |
1283 | size_t count) | |
1284 | { | |
1285 | unsigned long tmp; | |
1286 | char *p = (char *)page; | |
1287 | ||
1288 | if (reg->hr_bdev) | |
1289 | return -EINVAL; | |
1290 | ||
1291 | tmp = simple_strtoul(p, &p, 0); | |
1292 | if (!p || (*p && (*p != '\n'))) | |
1293 | return -EINVAL; | |
1294 | ||
1295 | if (tmp > O2NM_MAX_NODES || tmp == 0) | |
1296 | return -ERANGE; | |
1297 | ||
1298 | reg->hr_blocks = (unsigned int)tmp; | |
1299 | ||
1300 | return count; | |
1301 | } | |
1302 | ||
1303 | static ssize_t o2hb_region_dev_read(struct o2hb_region *reg, | |
1304 | char *page) | |
1305 | { | |
1306 | unsigned int ret = 0; | |
1307 | ||
1308 | if (reg->hr_bdev) | |
1309 | ret = sprintf(page, "%s\n", reg->hr_dev_name); | |
1310 | ||
1311 | return ret; | |
1312 | } | |
1313 | ||
1314 | static void o2hb_init_region_params(struct o2hb_region *reg) | |
1315 | { | |
1316 | reg->hr_slots_per_page = PAGE_CACHE_SIZE >> reg->hr_block_bits; | |
1317 | reg->hr_timeout_ms = O2HB_REGION_TIMEOUT_MS; | |
1318 | ||
1319 | mlog(ML_HEARTBEAT, "hr_start_block = %llu, hr_blocks = %u\n", | |
1320 | reg->hr_start_block, reg->hr_blocks); | |
1321 | mlog(ML_HEARTBEAT, "hr_block_bytes = %u, hr_block_bits = %u\n", | |
1322 | reg->hr_block_bytes, reg->hr_block_bits); | |
1323 | mlog(ML_HEARTBEAT, "hr_timeout_ms = %u\n", reg->hr_timeout_ms); | |
1324 | mlog(ML_HEARTBEAT, "dead threshold = %u\n", o2hb_dead_threshold); | |
1325 | } | |
1326 | ||
1327 | static int o2hb_map_slot_data(struct o2hb_region *reg) | |
1328 | { | |
1329 | int i, j; | |
1330 | unsigned int last_slot; | |
1331 | unsigned int spp = reg->hr_slots_per_page; | |
1332 | struct page *page; | |
1333 | char *raw; | |
1334 | struct o2hb_disk_slot *slot; | |
1335 | ||
1336 | reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL); | |
1337 | if (reg->hr_tmp_block == NULL) { | |
1338 | mlog_errno(-ENOMEM); | |
1339 | return -ENOMEM; | |
1340 | } | |
1341 | ||
1342 | reg->hr_slots = kcalloc(reg->hr_blocks, | |
1343 | sizeof(struct o2hb_disk_slot), GFP_KERNEL); | |
1344 | if (reg->hr_slots == NULL) { | |
1345 | mlog_errno(-ENOMEM); | |
1346 | return -ENOMEM; | |
1347 | } | |
1348 | ||
1349 | for(i = 0; i < reg->hr_blocks; i++) { | |
1350 | slot = ®->hr_slots[i]; | |
1351 | slot->ds_node_num = i; | |
1352 | INIT_LIST_HEAD(&slot->ds_live_item); | |
1353 | slot->ds_raw_block = NULL; | |
1354 | } | |
1355 | ||
1356 | reg->hr_num_pages = (reg->hr_blocks + spp - 1) / spp; | |
1357 | mlog(ML_HEARTBEAT, "Going to require %u pages to cover %u blocks " | |
1358 | "at %u blocks per page\n", | |
1359 | reg->hr_num_pages, reg->hr_blocks, spp); | |
1360 | ||
1361 | reg->hr_slot_data = kcalloc(reg->hr_num_pages, sizeof(struct page *), | |
1362 | GFP_KERNEL); | |
1363 | if (!reg->hr_slot_data) { | |
1364 | mlog_errno(-ENOMEM); | |
1365 | return -ENOMEM; | |
1366 | } | |
1367 | ||
1368 | for(i = 0; i < reg->hr_num_pages; i++) { | |
1369 | page = alloc_page(GFP_KERNEL); | |
1370 | if (!page) { | |
1371 | mlog_errno(-ENOMEM); | |
1372 | return -ENOMEM; | |
1373 | } | |
1374 | ||
1375 | reg->hr_slot_data[i] = page; | |
1376 | ||
1377 | last_slot = i * spp; | |
1378 | raw = page_address(page); | |
1379 | for (j = 0; | |
1380 | (j < spp) && ((j + last_slot) < reg->hr_blocks); | |
1381 | j++) { | |
1382 | BUG_ON((j + last_slot) >= reg->hr_blocks); | |
1383 | ||
1384 | slot = ®->hr_slots[j + last_slot]; | |
1385 | slot->ds_raw_block = | |
1386 | (struct o2hb_disk_heartbeat_block *) raw; | |
1387 | ||
1388 | raw += reg->hr_block_bytes; | |
1389 | } | |
1390 | } | |
1391 | ||
1392 | return 0; | |
1393 | } | |
1394 | ||
1395 | /* Read in all the slots available and populate the tracking | |
1396 | * structures so that we can start with a baseline idea of what's | |
1397 | * there. */ | |
1398 | static int o2hb_populate_slot_data(struct o2hb_region *reg) | |
1399 | { | |
1400 | int ret, i; | |
1401 | struct o2hb_disk_slot *slot; | |
1402 | struct o2hb_disk_heartbeat_block *hb_block; | |
1403 | ||
1404 | mlog_entry_void(); | |
1405 | ||
1406 | ret = o2hb_read_slots(reg, reg->hr_blocks); | |
1407 | if (ret) { | |
1408 | mlog_errno(ret); | |
1409 | goto out; | |
1410 | } | |
1411 | ||
1412 | /* We only want to get an idea of the values initially in each | |
1413 | * slot, so we do no verification - o2hb_check_slot will | |
1414 | * actually determine if each configured slot is valid and | |
1415 | * whether any values have changed. */ | |
1416 | for(i = 0; i < reg->hr_blocks; i++) { | |
1417 | slot = ®->hr_slots[i]; | |
1418 | hb_block = (struct o2hb_disk_heartbeat_block *) slot->ds_raw_block; | |
1419 | ||
1420 | /* Only fill the values that o2hb_check_slot uses to | |
1421 | * determine changing slots */ | |
1422 | slot->ds_last_time = le64_to_cpu(hb_block->hb_seq); | |
1423 | slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation); | |
1424 | } | |
1425 | ||
1426 | out: | |
1427 | mlog_exit(ret); | |
1428 | return ret; | |
1429 | } | |
1430 | ||
1431 | /* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */ | |
1432 | static ssize_t o2hb_region_dev_write(struct o2hb_region *reg, | |
1433 | const char *page, | |
1434 | size_t count) | |
1435 | { | |
e6c352db | 1436 | struct task_struct *hb_task; |
a7f6a5fb MF |
1437 | long fd; |
1438 | int sectsize; | |
1439 | char *p = (char *)page; | |
1440 | struct file *filp = NULL; | |
1441 | struct inode *inode = NULL; | |
1442 | ssize_t ret = -EINVAL; | |
1443 | ||
1444 | if (reg->hr_bdev) | |
1445 | goto out; | |
1446 | ||
1447 | /* We can't heartbeat without having had our node number | |
1448 | * configured yet. */ | |
1449 | if (o2nm_this_node() == O2NM_MAX_NODES) | |
1450 | goto out; | |
1451 | ||
1452 | fd = simple_strtol(p, &p, 0); | |
1453 | if (!p || (*p && (*p != '\n'))) | |
1454 | goto out; | |
1455 | ||
1456 | if (fd < 0 || fd >= INT_MAX) | |
1457 | goto out; | |
1458 | ||
1459 | filp = fget(fd); | |
1460 | if (filp == NULL) | |
1461 | goto out; | |
1462 | ||
1463 | if (reg->hr_blocks == 0 || reg->hr_start_block == 0 || | |
1464 | reg->hr_block_bytes == 0) | |
1465 | goto out; | |
1466 | ||
1467 | inode = igrab(filp->f_mapping->host); | |
1468 | if (inode == NULL) | |
1469 | goto out; | |
1470 | ||
1471 | if (!S_ISBLK(inode->i_mode)) | |
1472 | goto out; | |
1473 | ||
1474 | reg->hr_bdev = I_BDEV(filp->f_mapping->host); | |
572c4892 | 1475 | ret = blkdev_get(reg->hr_bdev, FMODE_WRITE | FMODE_READ); |
a7f6a5fb MF |
1476 | if (ret) { |
1477 | reg->hr_bdev = NULL; | |
1478 | goto out; | |
1479 | } | |
1480 | inode = NULL; | |
1481 | ||
1482 | bdevname(reg->hr_bdev, reg->hr_dev_name); | |
1483 | ||
e1defc4f | 1484 | sectsize = bdev_logical_block_size(reg->hr_bdev); |
a7f6a5fb MF |
1485 | if (sectsize != reg->hr_block_bytes) { |
1486 | mlog(ML_ERROR, | |
1487 | "blocksize %u incorrect for device, expected %d", | |
1488 | reg->hr_block_bytes, sectsize); | |
1489 | ret = -EINVAL; | |
1490 | goto out; | |
1491 | } | |
1492 | ||
1493 | o2hb_init_region_params(reg); | |
1494 | ||
1495 | /* Generation of zero is invalid */ | |
1496 | do { | |
1497 | get_random_bytes(®->hr_generation, | |
1498 | sizeof(reg->hr_generation)); | |
1499 | } while (reg->hr_generation == 0); | |
1500 | ||
1501 | ret = o2hb_map_slot_data(reg); | |
1502 | if (ret) { | |
1503 | mlog_errno(ret); | |
1504 | goto out; | |
1505 | } | |
1506 | ||
1507 | ret = o2hb_populate_slot_data(reg); | |
1508 | if (ret) { | |
1509 | mlog_errno(ret); | |
1510 | goto out; | |
1511 | } | |
1512 | ||
c4028958 | 1513 | INIT_DELAYED_WORK(®->hr_write_timeout_work, o2hb_write_timeout); |
a7f6a5fb MF |
1514 | |
1515 | /* | |
1516 | * A node is considered live after it has beat LIVE_THRESHOLD | |
1517 | * times. We're not steady until we've given them a chance | |
1518 | * _after_ our first read. | |
1519 | */ | |
1520 | atomic_set(®->hr_steady_iterations, O2HB_LIVE_THRESHOLD + 1); | |
1521 | ||
e6c352db JB |
1522 | hb_task = kthread_run(o2hb_thread, reg, "o2hb-%s", |
1523 | reg->hr_item.ci_name); | |
1524 | if (IS_ERR(hb_task)) { | |
1525 | ret = PTR_ERR(hb_task); | |
a7f6a5fb | 1526 | mlog_errno(ret); |
a7f6a5fb MF |
1527 | goto out; |
1528 | } | |
1529 | ||
e6c352db JB |
1530 | spin_lock(&o2hb_live_lock); |
1531 | reg->hr_task = hb_task; | |
1532 | spin_unlock(&o2hb_live_lock); | |
1533 | ||
a7f6a5fb MF |
1534 | ret = wait_event_interruptible(o2hb_steady_queue, |
1535 | atomic_read(®->hr_steady_iterations) == 0); | |
1536 | if (ret) { | |
e6df3a66 | 1537 | /* We got interrupted (hello ptrace!). Clean up */ |
e6c352db JB |
1538 | spin_lock(&o2hb_live_lock); |
1539 | hb_task = reg->hr_task; | |
a7f6a5fb | 1540 | reg->hr_task = NULL; |
e6c352db JB |
1541 | spin_unlock(&o2hb_live_lock); |
1542 | ||
1543 | if (hb_task) | |
1544 | kthread_stop(hb_task); | |
a7f6a5fb MF |
1545 | goto out; |
1546 | } | |
1547 | ||
e6df3a66 JB |
1548 | /* Ok, we were woken. Make sure it wasn't by drop_item() */ |
1549 | spin_lock(&o2hb_live_lock); | |
1550 | hb_task = reg->hr_task; | |
1551 | spin_unlock(&o2hb_live_lock); | |
1552 | ||
1553 | if (hb_task) | |
1554 | ret = count; | |
1555 | else | |
1556 | ret = -EIO; | |
1557 | ||
18c50cb0 SM |
1558 | if (hb_task && o2hb_global_heartbeat_active()) |
1559 | printk(KERN_NOTICE "o2hb: Heartbeat started on region %s\n", | |
1560 | config_item_name(®->hr_item)); | |
1561 | ||
a7f6a5fb MF |
1562 | out: |
1563 | if (filp) | |
1564 | fput(filp); | |
1565 | if (inode) | |
1566 | iput(inode); | |
1567 | if (ret < 0) { | |
1568 | if (reg->hr_bdev) { | |
9a1c3542 | 1569 | blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE); |
a7f6a5fb MF |
1570 | reg->hr_bdev = NULL; |
1571 | } | |
1572 | } | |
1573 | return ret; | |
1574 | } | |
1575 | ||
92efc152 ZW |
1576 | static ssize_t o2hb_region_pid_read(struct o2hb_region *reg, |
1577 | char *page) | |
1578 | { | |
e6c352db JB |
1579 | pid_t pid = 0; |
1580 | ||
1581 | spin_lock(&o2hb_live_lock); | |
1582 | if (reg->hr_task) | |
ba25f9dc | 1583 | pid = task_pid_nr(reg->hr_task); |
e6c352db JB |
1584 | spin_unlock(&o2hb_live_lock); |
1585 | ||
1586 | if (!pid) | |
92efc152 ZW |
1587 | return 0; |
1588 | ||
e6c352db | 1589 | return sprintf(page, "%u\n", pid); |
92efc152 ZW |
1590 | } |
1591 | ||
a7f6a5fb MF |
1592 | struct o2hb_region_attribute { |
1593 | struct configfs_attribute attr; | |
1594 | ssize_t (*show)(struct o2hb_region *, char *); | |
1595 | ssize_t (*store)(struct o2hb_region *, const char *, size_t); | |
1596 | }; | |
1597 | ||
1598 | static struct o2hb_region_attribute o2hb_region_attr_block_bytes = { | |
1599 | .attr = { .ca_owner = THIS_MODULE, | |
1600 | .ca_name = "block_bytes", | |
1601 | .ca_mode = S_IRUGO | S_IWUSR }, | |
1602 | .show = o2hb_region_block_bytes_read, | |
1603 | .store = o2hb_region_block_bytes_write, | |
1604 | }; | |
1605 | ||
1606 | static struct o2hb_region_attribute o2hb_region_attr_start_block = { | |
1607 | .attr = { .ca_owner = THIS_MODULE, | |
1608 | .ca_name = "start_block", | |
1609 | .ca_mode = S_IRUGO | S_IWUSR }, | |
1610 | .show = o2hb_region_start_block_read, | |
1611 | .store = o2hb_region_start_block_write, | |
1612 | }; | |
1613 | ||
1614 | static struct o2hb_region_attribute o2hb_region_attr_blocks = { | |
1615 | .attr = { .ca_owner = THIS_MODULE, | |
1616 | .ca_name = "blocks", | |
1617 | .ca_mode = S_IRUGO | S_IWUSR }, | |
1618 | .show = o2hb_region_blocks_read, | |
1619 | .store = o2hb_region_blocks_write, | |
1620 | }; | |
1621 | ||
1622 | static struct o2hb_region_attribute o2hb_region_attr_dev = { | |
1623 | .attr = { .ca_owner = THIS_MODULE, | |
1624 | .ca_name = "dev", | |
1625 | .ca_mode = S_IRUGO | S_IWUSR }, | |
1626 | .show = o2hb_region_dev_read, | |
1627 | .store = o2hb_region_dev_write, | |
1628 | }; | |
1629 | ||
92efc152 ZW |
1630 | static struct o2hb_region_attribute o2hb_region_attr_pid = { |
1631 | .attr = { .ca_owner = THIS_MODULE, | |
1632 | .ca_name = "pid", | |
1633 | .ca_mode = S_IRUGO | S_IRUSR }, | |
1634 | .show = o2hb_region_pid_read, | |
1635 | }; | |
1636 | ||
a7f6a5fb MF |
1637 | static struct configfs_attribute *o2hb_region_attrs[] = { |
1638 | &o2hb_region_attr_block_bytes.attr, | |
1639 | &o2hb_region_attr_start_block.attr, | |
1640 | &o2hb_region_attr_blocks.attr, | |
1641 | &o2hb_region_attr_dev.attr, | |
92efc152 | 1642 | &o2hb_region_attr_pid.attr, |
a7f6a5fb MF |
1643 | NULL, |
1644 | }; | |
1645 | ||
1646 | static ssize_t o2hb_region_show(struct config_item *item, | |
1647 | struct configfs_attribute *attr, | |
1648 | char *page) | |
1649 | { | |
1650 | struct o2hb_region *reg = to_o2hb_region(item); | |
1651 | struct o2hb_region_attribute *o2hb_region_attr = | |
1652 | container_of(attr, struct o2hb_region_attribute, attr); | |
1653 | ssize_t ret = 0; | |
1654 | ||
1655 | if (o2hb_region_attr->show) | |
1656 | ret = o2hb_region_attr->show(reg, page); | |
1657 | return ret; | |
1658 | } | |
1659 | ||
1660 | static ssize_t o2hb_region_store(struct config_item *item, | |
1661 | struct configfs_attribute *attr, | |
1662 | const char *page, size_t count) | |
1663 | { | |
1664 | struct o2hb_region *reg = to_o2hb_region(item); | |
1665 | struct o2hb_region_attribute *o2hb_region_attr = | |
1666 | container_of(attr, struct o2hb_region_attribute, attr); | |
1667 | ssize_t ret = -EINVAL; | |
1668 | ||
1669 | if (o2hb_region_attr->store) | |
1670 | ret = o2hb_region_attr->store(reg, page, count); | |
1671 | return ret; | |
1672 | } | |
1673 | ||
1674 | static struct configfs_item_operations o2hb_region_item_ops = { | |
1675 | .release = o2hb_region_release, | |
1676 | .show_attribute = o2hb_region_show, | |
1677 | .store_attribute = o2hb_region_store, | |
1678 | }; | |
1679 | ||
1680 | static struct config_item_type o2hb_region_type = { | |
1681 | .ct_item_ops = &o2hb_region_item_ops, | |
1682 | .ct_attrs = o2hb_region_attrs, | |
1683 | .ct_owner = THIS_MODULE, | |
1684 | }; | |
1685 | ||
1686 | /* heartbeat set */ | |
1687 | ||
1688 | struct o2hb_heartbeat_group { | |
1689 | struct config_group hs_group; | |
1690 | /* some stuff? */ | |
1691 | }; | |
1692 | ||
1693 | static struct o2hb_heartbeat_group *to_o2hb_heartbeat_group(struct config_group *group) | |
1694 | { | |
1695 | return group ? | |
1696 | container_of(group, struct o2hb_heartbeat_group, hs_group) | |
1697 | : NULL; | |
1698 | } | |
1699 | ||
f89ab861 JB |
1700 | static struct config_item *o2hb_heartbeat_group_make_item(struct config_group *group, |
1701 | const char *name) | |
a7f6a5fb MF |
1702 | { |
1703 | struct o2hb_region *reg = NULL; | |
a7f6a5fb | 1704 | |
cd861280 | 1705 | reg = kzalloc(sizeof(struct o2hb_region), GFP_KERNEL); |
f89ab861 | 1706 | if (reg == NULL) |
a6795e9e | 1707 | return ERR_PTR(-ENOMEM); |
a7f6a5fb | 1708 | |
b3c85c4c SM |
1709 | if (strlen(name) > O2HB_MAX_REGION_NAME_LEN) |
1710 | return ERR_PTR(-ENAMETOOLONG); | |
1711 | ||
a7f6a5fb MF |
1712 | config_item_init_type_name(®->hr_item, name, &o2hb_region_type); |
1713 | ||
a7f6a5fb MF |
1714 | spin_lock(&o2hb_live_lock); |
1715 | list_add_tail(®->hr_all_item, &o2hb_all_regions); | |
1716 | spin_unlock(&o2hb_live_lock); | |
a7f6a5fb | 1717 | |
a6795e9e | 1718 | return ®->hr_item; |
a7f6a5fb MF |
1719 | } |
1720 | ||
1721 | static void o2hb_heartbeat_group_drop_item(struct config_group *group, | |
1722 | struct config_item *item) | |
1723 | { | |
e6c352db | 1724 | struct task_struct *hb_task; |
a7f6a5fb MF |
1725 | struct o2hb_region *reg = to_o2hb_region(item); |
1726 | ||
1727 | /* stop the thread when the user removes the region dir */ | |
e6c352db JB |
1728 | spin_lock(&o2hb_live_lock); |
1729 | hb_task = reg->hr_task; | |
1730 | reg->hr_task = NULL; | |
1731 | spin_unlock(&o2hb_live_lock); | |
1732 | ||
1733 | if (hb_task) | |
1734 | kthread_stop(hb_task); | |
a7f6a5fb | 1735 | |
e6df3a66 JB |
1736 | /* |
1737 | * If we're racing a dev_write(), we need to wake them. They will | |
1738 | * check reg->hr_task | |
1739 | */ | |
1740 | if (atomic_read(®->hr_steady_iterations) != 0) { | |
1741 | atomic_set(®->hr_steady_iterations, 0); | |
1742 | wake_up(&o2hb_steady_queue); | |
1743 | } | |
1744 | ||
18c50cb0 SM |
1745 | if (o2hb_global_heartbeat_active()) |
1746 | printk(KERN_NOTICE "o2hb: Heartbeat stopped on region %s\n", | |
1747 | config_item_name(®->hr_item)); | |
a7f6a5fb MF |
1748 | config_item_put(item); |
1749 | } | |
1750 | ||
1751 | struct o2hb_heartbeat_group_attribute { | |
1752 | struct configfs_attribute attr; | |
1753 | ssize_t (*show)(struct o2hb_heartbeat_group *, char *); | |
1754 | ssize_t (*store)(struct o2hb_heartbeat_group *, const char *, size_t); | |
1755 | }; | |
1756 | ||
1757 | static ssize_t o2hb_heartbeat_group_show(struct config_item *item, | |
1758 | struct configfs_attribute *attr, | |
1759 | char *page) | |
1760 | { | |
1761 | struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item)); | |
1762 | struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr = | |
1763 | container_of(attr, struct o2hb_heartbeat_group_attribute, attr); | |
1764 | ssize_t ret = 0; | |
1765 | ||
1766 | if (o2hb_heartbeat_group_attr->show) | |
1767 | ret = o2hb_heartbeat_group_attr->show(reg, page); | |
1768 | return ret; | |
1769 | } | |
1770 | ||
1771 | static ssize_t o2hb_heartbeat_group_store(struct config_item *item, | |
1772 | struct configfs_attribute *attr, | |
1773 | const char *page, size_t count) | |
1774 | { | |
1775 | struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item)); | |
1776 | struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr = | |
1777 | container_of(attr, struct o2hb_heartbeat_group_attribute, attr); | |
1778 | ssize_t ret = -EINVAL; | |
1779 | ||
1780 | if (o2hb_heartbeat_group_attr->store) | |
1781 | ret = o2hb_heartbeat_group_attr->store(reg, page, count); | |
1782 | return ret; | |
1783 | } | |
1784 | ||
1785 | static ssize_t o2hb_heartbeat_group_threshold_show(struct o2hb_heartbeat_group *group, | |
1786 | char *page) | |
1787 | { | |
1788 | return sprintf(page, "%u\n", o2hb_dead_threshold); | |
1789 | } | |
1790 | ||
1791 | static ssize_t o2hb_heartbeat_group_threshold_store(struct o2hb_heartbeat_group *group, | |
1792 | const char *page, | |
1793 | size_t count) | |
1794 | { | |
1795 | unsigned long tmp; | |
1796 | char *p = (char *)page; | |
1797 | ||
1798 | tmp = simple_strtoul(p, &p, 10); | |
1799 | if (!p || (*p && (*p != '\n'))) | |
1800 | return -EINVAL; | |
1801 | ||
1802 | /* this will validate ranges for us. */ | |
1803 | o2hb_dead_threshold_set((unsigned int) tmp); | |
1804 | ||
1805 | return count; | |
1806 | } | |
1807 | ||
54b5187b SM |
1808 | static |
1809 | ssize_t o2hb_heartbeat_group_mode_show(struct o2hb_heartbeat_group *group, | |
1810 | char *page) | |
1811 | { | |
1812 | return sprintf(page, "%s\n", | |
1813 | o2hb_heartbeat_mode_desc[o2hb_heartbeat_mode]); | |
1814 | } | |
1815 | ||
1816 | static | |
1817 | ssize_t o2hb_heartbeat_group_mode_store(struct o2hb_heartbeat_group *group, | |
1818 | const char *page, size_t count) | |
1819 | { | |
1820 | unsigned int i; | |
1821 | int ret; | |
1822 | size_t len; | |
1823 | ||
1824 | len = (page[count - 1] == '\n') ? count - 1 : count; | |
1825 | if (!len) | |
1826 | return -EINVAL; | |
1827 | ||
1828 | for (i = 0; i < O2HB_HEARTBEAT_NUM_MODES; ++i) { | |
1829 | if (strnicmp(page, o2hb_heartbeat_mode_desc[i], len)) | |
1830 | continue; | |
1831 | ||
1832 | ret = o2hb_global_hearbeat_mode_set(i); | |
1833 | if (!ret) | |
18c50cb0 | 1834 | printk(KERN_NOTICE "o2hb: Heartbeat mode set to %s\n", |
54b5187b SM |
1835 | o2hb_heartbeat_mode_desc[i]); |
1836 | return count; | |
1837 | } | |
1838 | ||
1839 | return -EINVAL; | |
1840 | ||
1841 | } | |
1842 | ||
a7f6a5fb MF |
1843 | static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_threshold = { |
1844 | .attr = { .ca_owner = THIS_MODULE, | |
1845 | .ca_name = "dead_threshold", | |
1846 | .ca_mode = S_IRUGO | S_IWUSR }, | |
1847 | .show = o2hb_heartbeat_group_threshold_show, | |
1848 | .store = o2hb_heartbeat_group_threshold_store, | |
1849 | }; | |
1850 | ||
54b5187b SM |
1851 | static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_mode = { |
1852 | .attr = { .ca_owner = THIS_MODULE, | |
1853 | .ca_name = "mode", | |
1854 | .ca_mode = S_IRUGO | S_IWUSR }, | |
1855 | .show = o2hb_heartbeat_group_mode_show, | |
1856 | .store = o2hb_heartbeat_group_mode_store, | |
1857 | }; | |
1858 | ||
a7f6a5fb MF |
1859 | static struct configfs_attribute *o2hb_heartbeat_group_attrs[] = { |
1860 | &o2hb_heartbeat_group_attr_threshold.attr, | |
54b5187b | 1861 | &o2hb_heartbeat_group_attr_mode.attr, |
a7f6a5fb MF |
1862 | NULL, |
1863 | }; | |
1864 | ||
1865 | static struct configfs_item_operations o2hb_hearbeat_group_item_ops = { | |
1866 | .show_attribute = o2hb_heartbeat_group_show, | |
1867 | .store_attribute = o2hb_heartbeat_group_store, | |
1868 | }; | |
1869 | ||
1870 | static struct configfs_group_operations o2hb_heartbeat_group_group_ops = { | |
1871 | .make_item = o2hb_heartbeat_group_make_item, | |
1872 | .drop_item = o2hb_heartbeat_group_drop_item, | |
1873 | }; | |
1874 | ||
1875 | static struct config_item_type o2hb_heartbeat_group_type = { | |
1876 | .ct_group_ops = &o2hb_heartbeat_group_group_ops, | |
1877 | .ct_item_ops = &o2hb_hearbeat_group_item_ops, | |
1878 | .ct_attrs = o2hb_heartbeat_group_attrs, | |
1879 | .ct_owner = THIS_MODULE, | |
1880 | }; | |
1881 | ||
1882 | /* this is just here to avoid touching group in heartbeat.h which the | |
1883 | * entire damn world #includes */ | |
1884 | struct config_group *o2hb_alloc_hb_set(void) | |
1885 | { | |
1886 | struct o2hb_heartbeat_group *hs = NULL; | |
1887 | struct config_group *ret = NULL; | |
1888 | ||
cd861280 | 1889 | hs = kzalloc(sizeof(struct o2hb_heartbeat_group), GFP_KERNEL); |
a7f6a5fb MF |
1890 | if (hs == NULL) |
1891 | goto out; | |
1892 | ||
1893 | config_group_init_type_name(&hs->hs_group, "heartbeat", | |
1894 | &o2hb_heartbeat_group_type); | |
1895 | ||
1896 | ret = &hs->hs_group; | |
1897 | out: | |
1898 | if (ret == NULL) | |
1899 | kfree(hs); | |
1900 | return ret; | |
1901 | } | |
1902 | ||
1903 | void o2hb_free_hb_set(struct config_group *group) | |
1904 | { | |
1905 | struct o2hb_heartbeat_group *hs = to_o2hb_heartbeat_group(group); | |
1906 | kfree(hs); | |
1907 | } | |
1908 | ||
1909 | /* hb callback registration and issueing */ | |
1910 | ||
1911 | static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type) | |
1912 | { | |
1913 | if (type == O2HB_NUM_CB) | |
1914 | return ERR_PTR(-EINVAL); | |
1915 | ||
1916 | return &o2hb_callbacks[type]; | |
1917 | } | |
1918 | ||
1919 | void o2hb_setup_callback(struct o2hb_callback_func *hc, | |
1920 | enum o2hb_callback_type type, | |
1921 | o2hb_cb_func *func, | |
1922 | void *data, | |
1923 | int priority) | |
1924 | { | |
1925 | INIT_LIST_HEAD(&hc->hc_item); | |
1926 | hc->hc_func = func; | |
1927 | hc->hc_data = data; | |
1928 | hc->hc_priority = priority; | |
1929 | hc->hc_type = type; | |
1930 | hc->hc_magic = O2HB_CB_MAGIC; | |
1931 | } | |
1932 | EXPORT_SYMBOL_GPL(o2hb_setup_callback); | |
1933 | ||
14829422 JB |
1934 | static struct o2hb_region *o2hb_find_region(const char *region_uuid) |
1935 | { | |
1936 | struct o2hb_region *p, *reg = NULL; | |
1937 | ||
1938 | assert_spin_locked(&o2hb_live_lock); | |
1939 | ||
1940 | list_for_each_entry(p, &o2hb_all_regions, hr_all_item) { | |
1941 | if (!strcmp(region_uuid, config_item_name(&p->hr_item))) { | |
1942 | reg = p; | |
1943 | break; | |
1944 | } | |
1945 | } | |
1946 | ||
1947 | return reg; | |
1948 | } | |
1949 | ||
1950 | static int o2hb_region_get(const char *region_uuid) | |
1951 | { | |
1952 | int ret = 0; | |
1953 | struct o2hb_region *reg; | |
1954 | ||
1955 | spin_lock(&o2hb_live_lock); | |
1956 | ||
1957 | reg = o2hb_find_region(region_uuid); | |
1958 | if (!reg) | |
1959 | ret = -ENOENT; | |
1960 | spin_unlock(&o2hb_live_lock); | |
1961 | ||
16c6a4f2 JB |
1962 | if (ret) |
1963 | goto out; | |
1964 | ||
1965 | ret = o2nm_depend_this_node(); | |
1966 | if (ret) | |
1967 | goto out; | |
14829422 | 1968 | |
16c6a4f2 JB |
1969 | ret = o2nm_depend_item(®->hr_item); |
1970 | if (ret) | |
1971 | o2nm_undepend_this_node(); | |
1972 | ||
1973 | out: | |
14829422 JB |
1974 | return ret; |
1975 | } | |
1976 | ||
1977 | static void o2hb_region_put(const char *region_uuid) | |
1978 | { | |
1979 | struct o2hb_region *reg; | |
1980 | ||
1981 | spin_lock(&o2hb_live_lock); | |
1982 | ||
1983 | reg = o2hb_find_region(region_uuid); | |
1984 | ||
1985 | spin_unlock(&o2hb_live_lock); | |
1986 | ||
16c6a4f2 | 1987 | if (reg) { |
14829422 | 1988 | o2nm_undepend_item(®->hr_item); |
16c6a4f2 JB |
1989 | o2nm_undepend_this_node(); |
1990 | } | |
14829422 JB |
1991 | } |
1992 | ||
1993 | int o2hb_register_callback(const char *region_uuid, | |
1994 | struct o2hb_callback_func *hc) | |
a7f6a5fb MF |
1995 | { |
1996 | struct o2hb_callback_func *tmp; | |
1997 | struct list_head *iter; | |
1998 | struct o2hb_callback *hbcall; | |
1999 | int ret; | |
2000 | ||
2001 | BUG_ON(hc->hc_magic != O2HB_CB_MAGIC); | |
2002 | BUG_ON(!list_empty(&hc->hc_item)); | |
2003 | ||
2004 | hbcall = hbcall_from_type(hc->hc_type); | |
2005 | if (IS_ERR(hbcall)) { | |
2006 | ret = PTR_ERR(hbcall); | |
2007 | goto out; | |
2008 | } | |
2009 | ||
14829422 JB |
2010 | if (region_uuid) { |
2011 | ret = o2hb_region_get(region_uuid); | |
2012 | if (ret) | |
2013 | goto out; | |
2014 | } | |
2015 | ||
a7f6a5fb MF |
2016 | down_write(&o2hb_callback_sem); |
2017 | ||
2018 | list_for_each(iter, &hbcall->list) { | |
2019 | tmp = list_entry(iter, struct o2hb_callback_func, hc_item); | |
2020 | if (hc->hc_priority < tmp->hc_priority) { | |
2021 | list_add_tail(&hc->hc_item, iter); | |
2022 | break; | |
2023 | } | |
2024 | } | |
2025 | if (list_empty(&hc->hc_item)) | |
2026 | list_add_tail(&hc->hc_item, &hbcall->list); | |
2027 | ||
2028 | up_write(&o2hb_callback_sem); | |
2029 | ret = 0; | |
2030 | out: | |
2031 | mlog(ML_HEARTBEAT, "returning %d on behalf of %p for funcs %p\n", | |
2032 | ret, __builtin_return_address(0), hc); | |
2033 | return ret; | |
2034 | } | |
2035 | EXPORT_SYMBOL_GPL(o2hb_register_callback); | |
2036 | ||
14829422 JB |
2037 | void o2hb_unregister_callback(const char *region_uuid, |
2038 | struct o2hb_callback_func *hc) | |
a7f6a5fb MF |
2039 | { |
2040 | BUG_ON(hc->hc_magic != O2HB_CB_MAGIC); | |
2041 | ||
2042 | mlog(ML_HEARTBEAT, "on behalf of %p for funcs %p\n", | |
2043 | __builtin_return_address(0), hc); | |
2044 | ||
14829422 | 2045 | /* XXX Can this happen _with_ a region reference? */ |
a7f6a5fb | 2046 | if (list_empty(&hc->hc_item)) |
c24f72cc | 2047 | return; |
a7f6a5fb | 2048 | |
14829422 JB |
2049 | if (region_uuid) |
2050 | o2hb_region_put(region_uuid); | |
2051 | ||
a7f6a5fb MF |
2052 | down_write(&o2hb_callback_sem); |
2053 | ||
2054 | list_del_init(&hc->hc_item); | |
2055 | ||
2056 | up_write(&o2hb_callback_sem); | |
a7f6a5fb MF |
2057 | } |
2058 | EXPORT_SYMBOL_GPL(o2hb_unregister_callback); | |
2059 | ||
2060 | int o2hb_check_node_heartbeating(u8 node_num) | |
2061 | { | |
2062 | unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; | |
2063 | ||
2064 | o2hb_fill_node_map(testing_map, sizeof(testing_map)); | |
2065 | if (!test_bit(node_num, testing_map)) { | |
2066 | mlog(ML_HEARTBEAT, | |
2067 | "node (%u) does not have heartbeating enabled.\n", | |
2068 | node_num); | |
2069 | return 0; | |
2070 | } | |
2071 | ||
2072 | return 1; | |
2073 | } | |
2074 | EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating); | |
2075 | ||
2076 | int o2hb_check_node_heartbeating_from_callback(u8 node_num) | |
2077 | { | |
2078 | unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; | |
2079 | ||
2080 | o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map)); | |
2081 | if (!test_bit(node_num, testing_map)) { | |
2082 | mlog(ML_HEARTBEAT, | |
2083 | "node (%u) does not have heartbeating enabled.\n", | |
2084 | node_num); | |
2085 | return 0; | |
2086 | } | |
2087 | ||
2088 | return 1; | |
2089 | } | |
2090 | EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback); | |
2091 | ||
2092 | /* Makes sure our local node is configured with a node number, and is | |
2093 | * heartbeating. */ | |
2094 | int o2hb_check_local_node_heartbeating(void) | |
2095 | { | |
2096 | u8 node_num; | |
2097 | ||
2098 | /* if this node was set then we have networking */ | |
2099 | node_num = o2nm_this_node(); | |
2100 | if (node_num == O2NM_MAX_NODES) { | |
2101 | mlog(ML_HEARTBEAT, "this node has not been configured.\n"); | |
2102 | return 0; | |
2103 | } | |
2104 | ||
2105 | return o2hb_check_node_heartbeating(node_num); | |
2106 | } | |
2107 | EXPORT_SYMBOL_GPL(o2hb_check_local_node_heartbeating); | |
2108 | ||
2109 | /* | |
2110 | * this is just a hack until we get the plumbing which flips file systems | |
2111 | * read only and drops the hb ref instead of killing the node dead. | |
2112 | */ | |
2113 | void o2hb_stop_all_regions(void) | |
2114 | { | |
2115 | struct o2hb_region *reg; | |
2116 | ||
2117 | mlog(ML_ERROR, "stopping heartbeat on all active regions.\n"); | |
2118 | ||
2119 | spin_lock(&o2hb_live_lock); | |
2120 | ||
2121 | list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) | |
2122 | reg->hr_unclean_stop = 1; | |
2123 | ||
2124 | spin_unlock(&o2hb_live_lock); | |
2125 | } | |
2126 | EXPORT_SYMBOL_GPL(o2hb_stop_all_regions); | |
b3c85c4c SM |
2127 | |
2128 | int o2hb_get_all_regions(char *region_uuids, u8 max_regions) | |
2129 | { | |
2130 | struct o2hb_region *reg; | |
2131 | int numregs = 0; | |
2132 | char *p; | |
2133 | ||
2134 | spin_lock(&o2hb_live_lock); | |
2135 | ||
2136 | p = region_uuids; | |
2137 | list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) { | |
2138 | mlog(0, "Region: %s\n", config_item_name(®->hr_item)); | |
2139 | if (numregs < max_regions) { | |
2140 | memcpy(p, config_item_name(®->hr_item), | |
2141 | O2HB_MAX_REGION_NAME_LEN); | |
2142 | p += O2HB_MAX_REGION_NAME_LEN; | |
2143 | } | |
2144 | numregs++; | |
2145 | } | |
2146 | ||
2147 | spin_unlock(&o2hb_live_lock); | |
2148 | ||
2149 | return numregs; | |
2150 | } | |
2151 | EXPORT_SYMBOL_GPL(o2hb_get_all_regions); | |
2152 | ||
2153 | int o2hb_global_heartbeat_active(void) | |
2154 | { | |
2155 | return 0; | |
2156 | } | |
2157 | EXPORT_SYMBOL(o2hb_global_heartbeat_active); |