]>
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 | ||
536f0741 SM |
65 | /* |
66 | * In global heartbeat, we maintain a series of region bitmaps. | |
67 | * - o2hb_region_bitmap allows us to limit the region number to max region. | |
e7d656ba | 68 | * - o2hb_live_region_bitmap tracks live regions (seen steady iterations). |
43182d2a SM |
69 | * - o2hb_quorum_region_bitmap tracks live regions that have seen all nodes |
70 | * heartbeat on it. | |
536f0741 SM |
71 | */ |
72 | static unsigned long o2hb_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)]; | |
e7d656ba | 73 | static unsigned long o2hb_live_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)]; |
43182d2a | 74 | static unsigned long o2hb_quorum_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)]; |
536f0741 | 75 | |
8ca8b0bb SM |
76 | #define O2HB_DB_TYPE_LIVENODES 0 |
77 | struct o2hb_debug_buf { | |
78 | int db_type; | |
79 | int db_size; | |
80 | int db_len; | |
81 | void *db_data; | |
82 | }; | |
83 | ||
84 | static struct o2hb_debug_buf *o2hb_db_livenodes; | |
85 | ||
87d3d3f3 SM |
86 | #define O2HB_DEBUG_DIR "o2hb" |
87 | #define O2HB_DEBUG_LIVENODES "livenodes" | |
8ca8b0bb | 88 | |
87d3d3f3 SM |
89 | static struct dentry *o2hb_debug_dir; |
90 | static struct dentry *o2hb_debug_livenodes; | |
91 | ||
a7f6a5fb MF |
92 | static LIST_HEAD(o2hb_all_regions); |
93 | ||
94 | static struct o2hb_callback { | |
95 | struct list_head list; | |
96 | } o2hb_callbacks[O2HB_NUM_CB]; | |
97 | ||
98 | static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type); | |
99 | ||
100 | #define O2HB_DEFAULT_BLOCK_BITS 9 | |
101 | ||
54b5187b SM |
102 | enum o2hb_heartbeat_modes { |
103 | O2HB_HEARTBEAT_LOCAL = 0, | |
104 | O2HB_HEARTBEAT_GLOBAL, | |
105 | O2HB_HEARTBEAT_NUM_MODES, | |
106 | }; | |
107 | ||
108 | char *o2hb_heartbeat_mode_desc[O2HB_HEARTBEAT_NUM_MODES] = { | |
109 | "local", /* O2HB_HEARTBEAT_LOCAL */ | |
110 | "global", /* O2HB_HEARTBEAT_GLOBAL */ | |
111 | }; | |
112 | ||
a7f6a5fb | 113 | unsigned int o2hb_dead_threshold = O2HB_DEFAULT_DEAD_THRESHOLD; |
54b5187b | 114 | unsigned int o2hb_heartbeat_mode = O2HB_HEARTBEAT_LOCAL; |
a7f6a5fb | 115 | |
2bd63216 | 116 | /* Only sets a new threshold if there are no active regions. |
a7f6a5fb MF |
117 | * |
118 | * No locking or otherwise interesting code is required for reading | |
119 | * o2hb_dead_threshold as it can't change once regions are active and | |
120 | * it's not interesting to anyone until then anyway. */ | |
121 | static void o2hb_dead_threshold_set(unsigned int threshold) | |
122 | { | |
123 | if (threshold > O2HB_MIN_DEAD_THRESHOLD) { | |
124 | spin_lock(&o2hb_live_lock); | |
125 | if (list_empty(&o2hb_all_regions)) | |
126 | o2hb_dead_threshold = threshold; | |
127 | spin_unlock(&o2hb_live_lock); | |
128 | } | |
129 | } | |
130 | ||
54b5187b SM |
131 | static int o2hb_global_hearbeat_mode_set(unsigned int hb_mode) |
132 | { | |
133 | int ret = -1; | |
134 | ||
135 | if (hb_mode < O2HB_HEARTBEAT_NUM_MODES) { | |
136 | spin_lock(&o2hb_live_lock); | |
137 | if (list_empty(&o2hb_all_regions)) { | |
138 | o2hb_heartbeat_mode = hb_mode; | |
139 | ret = 0; | |
140 | } | |
141 | spin_unlock(&o2hb_live_lock); | |
142 | } | |
143 | ||
144 | return ret; | |
145 | } | |
146 | ||
a7f6a5fb MF |
147 | struct o2hb_node_event { |
148 | struct list_head hn_item; | |
149 | enum o2hb_callback_type hn_event_type; | |
150 | struct o2nm_node *hn_node; | |
151 | int hn_node_num; | |
152 | }; | |
153 | ||
154 | struct o2hb_disk_slot { | |
155 | struct o2hb_disk_heartbeat_block *ds_raw_block; | |
156 | u8 ds_node_num; | |
157 | u64 ds_last_time; | |
158 | u64 ds_last_generation; | |
159 | u16 ds_equal_samples; | |
160 | u16 ds_changed_samples; | |
161 | struct list_head ds_live_item; | |
162 | }; | |
163 | ||
164 | /* each thread owns a region.. when we're asked to tear down the region | |
165 | * we ask the thread to stop, who cleans up the region */ | |
166 | struct o2hb_region { | |
167 | struct config_item hr_item; | |
168 | ||
169 | struct list_head hr_all_item; | |
170 | unsigned hr_unclean_stop:1; | |
171 | ||
172 | /* protected by the hr_callback_sem */ | |
173 | struct task_struct *hr_task; | |
174 | ||
175 | unsigned int hr_blocks; | |
176 | unsigned long long hr_start_block; | |
177 | ||
178 | unsigned int hr_block_bits; | |
179 | unsigned int hr_block_bytes; | |
180 | ||
181 | unsigned int hr_slots_per_page; | |
182 | unsigned int hr_num_pages; | |
183 | ||
184 | struct page **hr_slot_data; | |
185 | struct block_device *hr_bdev; | |
186 | struct o2hb_disk_slot *hr_slots; | |
187 | ||
823a637a SM |
188 | /* live node map of this region */ |
189 | unsigned long hr_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)]; | |
536f0741 | 190 | unsigned int hr_region_num; |
823a637a | 191 | |
a7f6a5fb MF |
192 | /* let the person setting up hb wait for it to return until it |
193 | * has reached a 'steady' state. This will be fixed when we have | |
194 | * a more complete api that doesn't lead to this sort of fragility. */ | |
195 | atomic_t hr_steady_iterations; | |
196 | ||
197 | char hr_dev_name[BDEVNAME_SIZE]; | |
198 | ||
199 | unsigned int hr_timeout_ms; | |
200 | ||
201 | /* randomized as the region goes up and down so that a node | |
202 | * recognizes a node going up and down in one iteration */ | |
203 | u64 hr_generation; | |
204 | ||
c4028958 | 205 | struct delayed_work hr_write_timeout_work; |
a7f6a5fb MF |
206 | unsigned long hr_last_timeout_start; |
207 | ||
208 | /* Used during o2hb_check_slot to hold a copy of the block | |
209 | * being checked because we temporarily have to zero out the | |
210 | * crc field. */ | |
211 | struct o2hb_disk_heartbeat_block *hr_tmp_block; | |
212 | }; | |
213 | ||
214 | struct o2hb_bio_wait_ctxt { | |
215 | atomic_t wc_num_reqs; | |
216 | struct completion wc_io_complete; | |
a9e2ae39 | 217 | int wc_error; |
a7f6a5fb MF |
218 | }; |
219 | ||
c4028958 | 220 | static void o2hb_write_timeout(struct work_struct *work) |
a7f6a5fb | 221 | { |
c4028958 DH |
222 | struct o2hb_region *reg = |
223 | container_of(work, struct o2hb_region, | |
224 | hr_write_timeout_work.work); | |
a7f6a5fb MF |
225 | |
226 | mlog(ML_ERROR, "Heartbeat write timeout to device %s after %u " | |
227 | "milliseconds\n", reg->hr_dev_name, | |
2bd63216 | 228 | jiffies_to_msecs(jiffies - reg->hr_last_timeout_start)); |
a7f6a5fb MF |
229 | o2quo_disk_timeout(); |
230 | } | |
231 | ||
232 | static void o2hb_arm_write_timeout(struct o2hb_region *reg) | |
233 | { | |
b31d308d TM |
234 | mlog(ML_HEARTBEAT, "Queue write timeout for %u ms\n", |
235 | O2HB_MAX_WRITE_TIMEOUT_MS); | |
a7f6a5fb MF |
236 | |
237 | cancel_delayed_work(®->hr_write_timeout_work); | |
238 | reg->hr_last_timeout_start = jiffies; | |
239 | schedule_delayed_work(®->hr_write_timeout_work, | |
240 | msecs_to_jiffies(O2HB_MAX_WRITE_TIMEOUT_MS)); | |
241 | } | |
242 | ||
243 | static void o2hb_disarm_write_timeout(struct o2hb_region *reg) | |
244 | { | |
245 | cancel_delayed_work(®->hr_write_timeout_work); | |
246 | flush_scheduled_work(); | |
247 | } | |
248 | ||
b559292e | 249 | static inline void o2hb_bio_wait_init(struct o2hb_bio_wait_ctxt *wc) |
a7f6a5fb | 250 | { |
b559292e | 251 | atomic_set(&wc->wc_num_reqs, 1); |
a7f6a5fb | 252 | init_completion(&wc->wc_io_complete); |
a9e2ae39 | 253 | wc->wc_error = 0; |
a7f6a5fb MF |
254 | } |
255 | ||
256 | /* Used in error paths too */ | |
257 | static inline void o2hb_bio_wait_dec(struct o2hb_bio_wait_ctxt *wc, | |
258 | unsigned int num) | |
259 | { | |
260 | /* sadly atomic_sub_and_test() isn't available on all platforms. The | |
261 | * good news is that the fast path only completes one at a time */ | |
262 | while(num--) { | |
263 | if (atomic_dec_and_test(&wc->wc_num_reqs)) { | |
264 | BUG_ON(num > 0); | |
265 | complete(&wc->wc_io_complete); | |
266 | } | |
267 | } | |
268 | } | |
269 | ||
270 | static void o2hb_wait_on_io(struct o2hb_region *reg, | |
271 | struct o2hb_bio_wait_ctxt *wc) | |
272 | { | |
273 | struct address_space *mapping = reg->hr_bdev->bd_inode->i_mapping; | |
274 | ||
275 | blk_run_address_space(mapping); | |
b559292e | 276 | o2hb_bio_wait_dec(wc, 1); |
a7f6a5fb MF |
277 | |
278 | wait_for_completion(&wc->wc_io_complete); | |
279 | } | |
280 | ||
782e3b3b | 281 | static void o2hb_bio_end_io(struct bio *bio, |
a7f6a5fb MF |
282 | int error) |
283 | { | |
284 | struct o2hb_bio_wait_ctxt *wc = bio->bi_private; | |
285 | ||
a9e2ae39 | 286 | if (error) { |
a7f6a5fb | 287 | mlog(ML_ERROR, "IO Error %d\n", error); |
a9e2ae39 MF |
288 | wc->wc_error = error; |
289 | } | |
a7f6a5fb | 290 | |
a7f6a5fb | 291 | o2hb_bio_wait_dec(wc, 1); |
b559292e | 292 | bio_put(bio); |
a7f6a5fb MF |
293 | } |
294 | ||
295 | /* Setup a Bio to cover I/O against num_slots slots starting at | |
296 | * start_slot. */ | |
297 | static struct bio *o2hb_setup_one_bio(struct o2hb_region *reg, | |
298 | struct o2hb_bio_wait_ctxt *wc, | |
b559292e PR |
299 | unsigned int *current_slot, |
300 | unsigned int max_slots) | |
a7f6a5fb | 301 | { |
b559292e | 302 | int len, current_page; |
a7f6a5fb MF |
303 | unsigned int vec_len, vec_start; |
304 | unsigned int bits = reg->hr_block_bits; | |
305 | unsigned int spp = reg->hr_slots_per_page; | |
b559292e | 306 | unsigned int cs = *current_slot; |
a7f6a5fb MF |
307 | struct bio *bio; |
308 | struct page *page; | |
309 | ||
a7f6a5fb MF |
310 | /* Testing has shown this allocation to take long enough under |
311 | * GFP_KERNEL that the local node can get fenced. It would be | |
312 | * nicest if we could pre-allocate these bios and avoid this | |
313 | * all together. */ | |
b559292e | 314 | bio = bio_alloc(GFP_ATOMIC, 16); |
a7f6a5fb MF |
315 | if (!bio) { |
316 | mlog(ML_ERROR, "Could not alloc slots BIO!\n"); | |
317 | bio = ERR_PTR(-ENOMEM); | |
318 | goto bail; | |
319 | } | |
320 | ||
321 | /* Must put everything in 512 byte sectors for the bio... */ | |
b559292e | 322 | bio->bi_sector = (reg->hr_start_block + cs) << (bits - 9); |
a7f6a5fb MF |
323 | bio->bi_bdev = reg->hr_bdev; |
324 | bio->bi_private = wc; | |
325 | bio->bi_end_io = o2hb_bio_end_io; | |
326 | ||
b559292e PR |
327 | vec_start = (cs << bits) % PAGE_CACHE_SIZE; |
328 | while(cs < max_slots) { | |
329 | current_page = cs / spp; | |
330 | page = reg->hr_slot_data[current_page]; | |
a7f6a5fb | 331 | |
bc7e97cb | 332 | vec_len = min(PAGE_CACHE_SIZE - vec_start, |
b559292e | 333 | (max_slots-cs) * (PAGE_CACHE_SIZE/spp) ); |
a7f6a5fb MF |
334 | |
335 | mlog(ML_HB_BIO, "page %d, vec_len = %u, vec_start = %u\n", | |
b559292e | 336 | current_page, vec_len, vec_start); |
a7f6a5fb MF |
337 | |
338 | len = bio_add_page(bio, page, vec_len, vec_start); | |
b559292e | 339 | if (len != vec_len) break; |
a7f6a5fb | 340 | |
b559292e | 341 | cs += vec_len / (PAGE_CACHE_SIZE/spp); |
a7f6a5fb MF |
342 | vec_start = 0; |
343 | } | |
344 | ||
345 | bail: | |
b559292e | 346 | *current_slot = cs; |
a7f6a5fb MF |
347 | return bio; |
348 | } | |
349 | ||
a7f6a5fb MF |
350 | static int o2hb_read_slots(struct o2hb_region *reg, |
351 | unsigned int max_slots) | |
352 | { | |
b559292e PR |
353 | unsigned int current_slot=0; |
354 | int status; | |
a7f6a5fb | 355 | struct o2hb_bio_wait_ctxt wc; |
a7f6a5fb MF |
356 | struct bio *bio; |
357 | ||
b559292e | 358 | o2hb_bio_wait_init(&wc); |
a7f6a5fb | 359 | |
b559292e PR |
360 | while(current_slot < max_slots) { |
361 | bio = o2hb_setup_one_bio(reg, &wc, ¤t_slot, max_slots); | |
a7f6a5fb | 362 | if (IS_ERR(bio)) { |
a7f6a5fb MF |
363 | status = PTR_ERR(bio); |
364 | mlog_errno(status); | |
365 | goto bail_and_wait; | |
366 | } | |
a7f6a5fb | 367 | |
b559292e | 368 | atomic_inc(&wc.wc_num_reqs); |
a7f6a5fb MF |
369 | submit_bio(READ, bio); |
370 | } | |
371 | ||
372 | status = 0; | |
373 | ||
374 | bail_and_wait: | |
375 | o2hb_wait_on_io(reg, &wc); | |
a9e2ae39 MF |
376 | if (wc.wc_error && !status) |
377 | status = wc.wc_error; | |
a7f6a5fb | 378 | |
a7f6a5fb MF |
379 | return status; |
380 | } | |
381 | ||
382 | static int o2hb_issue_node_write(struct o2hb_region *reg, | |
a7f6a5fb MF |
383 | struct o2hb_bio_wait_ctxt *write_wc) |
384 | { | |
385 | int status; | |
386 | unsigned int slot; | |
387 | struct bio *bio; | |
388 | ||
b559292e | 389 | o2hb_bio_wait_init(write_wc); |
a7f6a5fb MF |
390 | |
391 | slot = o2nm_this_node(); | |
392 | ||
b559292e | 393 | bio = o2hb_setup_one_bio(reg, write_wc, &slot, slot+1); |
a7f6a5fb MF |
394 | if (IS_ERR(bio)) { |
395 | status = PTR_ERR(bio); | |
396 | mlog_errno(status); | |
397 | goto bail; | |
398 | } | |
399 | ||
b559292e | 400 | atomic_inc(&write_wc->wc_num_reqs); |
a7f6a5fb MF |
401 | submit_bio(WRITE, bio); |
402 | ||
a7f6a5fb MF |
403 | status = 0; |
404 | bail: | |
405 | return status; | |
406 | } | |
407 | ||
408 | static u32 o2hb_compute_block_crc_le(struct o2hb_region *reg, | |
409 | struct o2hb_disk_heartbeat_block *hb_block) | |
410 | { | |
411 | __le32 old_cksum; | |
412 | u32 ret; | |
413 | ||
414 | /* We want to compute the block crc with a 0 value in the | |
415 | * hb_cksum field. Save it off here and replace after the | |
416 | * crc. */ | |
417 | old_cksum = hb_block->hb_cksum; | |
418 | hb_block->hb_cksum = 0; | |
419 | ||
420 | ret = crc32_le(0, (unsigned char *) hb_block, reg->hr_block_bytes); | |
421 | ||
422 | hb_block->hb_cksum = old_cksum; | |
423 | ||
424 | return ret; | |
425 | } | |
426 | ||
427 | static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block *hb_block) | |
428 | { | |
70bacbdb MF |
429 | mlog(ML_ERROR, "Dump slot information: seq = 0x%llx, node = %u, " |
430 | "cksum = 0x%x, generation 0x%llx\n", | |
431 | (long long)le64_to_cpu(hb_block->hb_seq), | |
432 | hb_block->hb_node, le32_to_cpu(hb_block->hb_cksum), | |
433 | (long long)le64_to_cpu(hb_block->hb_generation)); | |
a7f6a5fb MF |
434 | } |
435 | ||
436 | static int o2hb_verify_crc(struct o2hb_region *reg, | |
437 | struct o2hb_disk_heartbeat_block *hb_block) | |
438 | { | |
439 | u32 read, computed; | |
440 | ||
441 | read = le32_to_cpu(hb_block->hb_cksum); | |
442 | computed = o2hb_compute_block_crc_le(reg, hb_block); | |
443 | ||
444 | return read == computed; | |
445 | } | |
446 | ||
447 | /* We want to make sure that nobody is heartbeating on top of us -- | |
448 | * this will help detect an invalid configuration. */ | |
449 | static int o2hb_check_last_timestamp(struct o2hb_region *reg) | |
450 | { | |
451 | int node_num, ret; | |
452 | struct o2hb_disk_slot *slot; | |
453 | struct o2hb_disk_heartbeat_block *hb_block; | |
454 | ||
455 | node_num = o2nm_this_node(); | |
456 | ||
457 | ret = 1; | |
458 | slot = ®->hr_slots[node_num]; | |
459 | /* Don't check on our 1st timestamp */ | |
460 | if (slot->ds_last_time) { | |
461 | hb_block = slot->ds_raw_block; | |
462 | ||
463 | if (le64_to_cpu(hb_block->hb_seq) != slot->ds_last_time) | |
464 | ret = 0; | |
465 | } | |
466 | ||
467 | return ret; | |
468 | } | |
469 | ||
470 | static inline void o2hb_prepare_block(struct o2hb_region *reg, | |
471 | u64 generation) | |
472 | { | |
473 | int node_num; | |
474 | u64 cputime; | |
475 | struct o2hb_disk_slot *slot; | |
476 | struct o2hb_disk_heartbeat_block *hb_block; | |
477 | ||
478 | node_num = o2nm_this_node(); | |
479 | slot = ®->hr_slots[node_num]; | |
480 | ||
481 | hb_block = (struct o2hb_disk_heartbeat_block *)slot->ds_raw_block; | |
482 | memset(hb_block, 0, reg->hr_block_bytes); | |
483 | /* TODO: time stuff */ | |
484 | cputime = CURRENT_TIME.tv_sec; | |
485 | if (!cputime) | |
486 | cputime = 1; | |
487 | ||
488 | hb_block->hb_seq = cpu_to_le64(cputime); | |
489 | hb_block->hb_node = node_num; | |
490 | hb_block->hb_generation = cpu_to_le64(generation); | |
0db638f4 | 491 | hb_block->hb_dead_ms = cpu_to_le32(o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS); |
a7f6a5fb MF |
492 | |
493 | /* This step must always happen last! */ | |
494 | hb_block->hb_cksum = cpu_to_le32(o2hb_compute_block_crc_le(reg, | |
495 | hb_block)); | |
496 | ||
70bacbdb | 497 | mlog(ML_HB_BIO, "our node generation = 0x%llx, cksum = 0x%x\n", |
5fdf1e67 | 498 | (long long)generation, |
70bacbdb | 499 | le32_to_cpu(hb_block->hb_cksum)); |
a7f6a5fb MF |
500 | } |
501 | ||
502 | static void o2hb_fire_callbacks(struct o2hb_callback *hbcall, | |
503 | struct o2nm_node *node, | |
504 | int idx) | |
505 | { | |
506 | struct list_head *iter; | |
507 | struct o2hb_callback_func *f; | |
508 | ||
509 | list_for_each(iter, &hbcall->list) { | |
510 | f = list_entry(iter, struct o2hb_callback_func, hc_item); | |
511 | mlog(ML_HEARTBEAT, "calling funcs %p\n", f); | |
512 | (f->hc_func)(node, idx, f->hc_data); | |
513 | } | |
514 | } | |
515 | ||
516 | /* Will run the list in order until we process the passed event */ | |
517 | static void o2hb_run_event_list(struct o2hb_node_event *queued_event) | |
518 | { | |
519 | int empty; | |
520 | struct o2hb_callback *hbcall; | |
521 | struct o2hb_node_event *event; | |
522 | ||
523 | spin_lock(&o2hb_live_lock); | |
524 | empty = list_empty(&queued_event->hn_item); | |
525 | spin_unlock(&o2hb_live_lock); | |
526 | if (empty) | |
527 | return; | |
528 | ||
529 | /* Holding callback sem assures we don't alter the callback | |
530 | * lists when doing this, and serializes ourselves with other | |
531 | * processes wanting callbacks. */ | |
532 | down_write(&o2hb_callback_sem); | |
533 | ||
534 | spin_lock(&o2hb_live_lock); | |
535 | while (!list_empty(&o2hb_node_events) | |
536 | && !list_empty(&queued_event->hn_item)) { | |
537 | event = list_entry(o2hb_node_events.next, | |
538 | struct o2hb_node_event, | |
539 | hn_item); | |
540 | list_del_init(&event->hn_item); | |
541 | spin_unlock(&o2hb_live_lock); | |
542 | ||
543 | mlog(ML_HEARTBEAT, "Node %s event for %d\n", | |
544 | event->hn_event_type == O2HB_NODE_UP_CB ? "UP" : "DOWN", | |
545 | event->hn_node_num); | |
546 | ||
547 | hbcall = hbcall_from_type(event->hn_event_type); | |
548 | ||
549 | /* We should *never* have gotten on to the list with a | |
550 | * bad type... This isn't something that we should try | |
551 | * to recover from. */ | |
552 | BUG_ON(IS_ERR(hbcall)); | |
553 | ||
554 | o2hb_fire_callbacks(hbcall, event->hn_node, event->hn_node_num); | |
555 | ||
556 | spin_lock(&o2hb_live_lock); | |
557 | } | |
558 | spin_unlock(&o2hb_live_lock); | |
559 | ||
560 | up_write(&o2hb_callback_sem); | |
561 | } | |
562 | ||
563 | static void o2hb_queue_node_event(struct o2hb_node_event *event, | |
564 | enum o2hb_callback_type type, | |
565 | struct o2nm_node *node, | |
566 | int node_num) | |
567 | { | |
568 | assert_spin_locked(&o2hb_live_lock); | |
569 | ||
0e105d37 SM |
570 | BUG_ON((!node) && (type != O2HB_NODE_DOWN_CB)); |
571 | ||
a7f6a5fb MF |
572 | event->hn_event_type = type; |
573 | event->hn_node = node; | |
574 | event->hn_node_num = node_num; | |
575 | ||
576 | mlog(ML_HEARTBEAT, "Queue node %s event for node %d\n", | |
577 | type == O2HB_NODE_UP_CB ? "UP" : "DOWN", node_num); | |
578 | ||
579 | list_add_tail(&event->hn_item, &o2hb_node_events); | |
580 | } | |
581 | ||
582 | static void o2hb_shutdown_slot(struct o2hb_disk_slot *slot) | |
583 | { | |
584 | struct o2hb_node_event event = | |
585 | { .hn_item = LIST_HEAD_INIT(event.hn_item), }; | |
586 | struct o2nm_node *node; | |
587 | ||
588 | node = o2nm_get_node_by_num(slot->ds_node_num); | |
589 | if (!node) | |
590 | return; | |
591 | ||
592 | spin_lock(&o2hb_live_lock); | |
593 | if (!list_empty(&slot->ds_live_item)) { | |
594 | mlog(ML_HEARTBEAT, "Shutdown, node %d leaves region\n", | |
595 | slot->ds_node_num); | |
596 | ||
597 | list_del_init(&slot->ds_live_item); | |
598 | ||
599 | if (list_empty(&o2hb_live_slots[slot->ds_node_num])) { | |
600 | clear_bit(slot->ds_node_num, o2hb_live_node_bitmap); | |
601 | ||
602 | o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node, | |
603 | slot->ds_node_num); | |
604 | } | |
605 | } | |
606 | spin_unlock(&o2hb_live_lock); | |
607 | ||
608 | o2hb_run_event_list(&event); | |
609 | ||
610 | o2nm_node_put(node); | |
611 | } | |
612 | ||
43182d2a SM |
613 | static void o2hb_set_quorum_device(struct o2hb_region *reg, |
614 | struct o2hb_disk_slot *slot) | |
615 | { | |
616 | assert_spin_locked(&o2hb_live_lock); | |
617 | ||
618 | if (!o2hb_global_heartbeat_active()) | |
619 | return; | |
620 | ||
621 | if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap)) | |
622 | return; | |
623 | ||
624 | /* | |
625 | * A region can be added to the quorum only when it sees all | |
626 | * live nodes heartbeat on it. In other words, the region has been | |
627 | * added to all nodes. | |
628 | */ | |
629 | if (memcmp(reg->hr_live_node_bitmap, o2hb_live_node_bitmap, | |
630 | sizeof(o2hb_live_node_bitmap))) | |
631 | return; | |
632 | ||
633 | if (slot->ds_changed_samples < O2HB_LIVE_THRESHOLD) | |
634 | return; | |
635 | ||
636 | printk(KERN_NOTICE "o2hb: Region %s is now a quorum device\n", | |
637 | config_item_name(®->hr_item)); | |
638 | ||
639 | set_bit(reg->hr_region_num, o2hb_quorum_region_bitmap); | |
640 | } | |
641 | ||
a7f6a5fb MF |
642 | static int o2hb_check_slot(struct o2hb_region *reg, |
643 | struct o2hb_disk_slot *slot) | |
644 | { | |
645 | int changed = 0, gen_changed = 0; | |
646 | struct o2hb_node_event event = | |
647 | { .hn_item = LIST_HEAD_INIT(event.hn_item), }; | |
648 | struct o2nm_node *node; | |
649 | struct o2hb_disk_heartbeat_block *hb_block = reg->hr_tmp_block; | |
650 | u64 cputime; | |
0db638f4 MF |
651 | unsigned int dead_ms = o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS; |
652 | unsigned int slot_dead_ms; | |
0e105d37 | 653 | int tmp; |
a7f6a5fb MF |
654 | |
655 | memcpy(hb_block, slot->ds_raw_block, reg->hr_block_bytes); | |
656 | ||
0e105d37 SM |
657 | /* |
658 | * If a node is no longer configured but is still in the livemap, we | |
659 | * may need to clear that bit from the livemap. | |
660 | */ | |
a7f6a5fb | 661 | node = o2nm_get_node_by_num(slot->ds_node_num); |
0e105d37 SM |
662 | if (!node) { |
663 | spin_lock(&o2hb_live_lock); | |
664 | tmp = test_bit(slot->ds_node_num, o2hb_live_node_bitmap); | |
665 | spin_unlock(&o2hb_live_lock); | |
666 | if (!tmp) | |
667 | return 0; | |
668 | } | |
a7f6a5fb MF |
669 | |
670 | if (!o2hb_verify_crc(reg, hb_block)) { | |
671 | /* all paths from here will drop o2hb_live_lock for | |
672 | * us. */ | |
673 | spin_lock(&o2hb_live_lock); | |
674 | ||
675 | /* Don't print an error on the console in this case - | |
676 | * a freshly formatted heartbeat area will not have a | |
677 | * crc set on it. */ | |
678 | if (list_empty(&slot->ds_live_item)) | |
679 | goto out; | |
680 | ||
681 | /* The node is live but pushed out a bad crc. We | |
682 | * consider it a transient miss but don't populate any | |
683 | * other values as they may be junk. */ | |
684 | mlog(ML_ERROR, "Node %d has written a bad crc to %s\n", | |
685 | slot->ds_node_num, reg->hr_dev_name); | |
686 | o2hb_dump_slot(hb_block); | |
687 | ||
688 | slot->ds_equal_samples++; | |
689 | goto fire_callbacks; | |
690 | } | |
691 | ||
692 | /* we don't care if these wrap.. the state transitions below | |
693 | * clear at the right places */ | |
694 | cputime = le64_to_cpu(hb_block->hb_seq); | |
695 | if (slot->ds_last_time != cputime) | |
696 | slot->ds_changed_samples++; | |
697 | else | |
698 | slot->ds_equal_samples++; | |
699 | slot->ds_last_time = cputime; | |
700 | ||
701 | /* The node changed heartbeat generations. We assume this to | |
702 | * mean it dropped off but came back before we timed out. We | |
703 | * want to consider it down for the time being but don't want | |
704 | * to lose any changed_samples state we might build up to | |
705 | * considering it live again. */ | |
706 | if (slot->ds_last_generation != le64_to_cpu(hb_block->hb_generation)) { | |
707 | gen_changed = 1; | |
708 | slot->ds_equal_samples = 0; | |
70bacbdb MF |
709 | mlog(ML_HEARTBEAT, "Node %d changed generation (0x%llx " |
710 | "to 0x%llx)\n", slot->ds_node_num, | |
711 | (long long)slot->ds_last_generation, | |
712 | (long long)le64_to_cpu(hb_block->hb_generation)); | |
a7f6a5fb MF |
713 | } |
714 | ||
715 | slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation); | |
716 | ||
70bacbdb MF |
717 | mlog(ML_HEARTBEAT, "Slot %d gen 0x%llx cksum 0x%x " |
718 | "seq %llu last %llu changed %u equal %u\n", | |
719 | slot->ds_node_num, (long long)slot->ds_last_generation, | |
720 | le32_to_cpu(hb_block->hb_cksum), | |
2bd63216 | 721 | (unsigned long long)le64_to_cpu(hb_block->hb_seq), |
70bacbdb | 722 | (unsigned long long)slot->ds_last_time, slot->ds_changed_samples, |
a7f6a5fb MF |
723 | slot->ds_equal_samples); |
724 | ||
725 | spin_lock(&o2hb_live_lock); | |
726 | ||
727 | fire_callbacks: | |
728 | /* dead nodes only come to life after some number of | |
729 | * changes at any time during their dead time */ | |
730 | if (list_empty(&slot->ds_live_item) && | |
731 | slot->ds_changed_samples >= O2HB_LIVE_THRESHOLD) { | |
70bacbdb MF |
732 | mlog(ML_HEARTBEAT, "Node %d (id 0x%llx) joined my region\n", |
733 | slot->ds_node_num, (long long)slot->ds_last_generation); | |
a7f6a5fb | 734 | |
823a637a SM |
735 | set_bit(slot->ds_node_num, reg->hr_live_node_bitmap); |
736 | ||
a7f6a5fb MF |
737 | /* first on the list generates a callback */ |
738 | if (list_empty(&o2hb_live_slots[slot->ds_node_num])) { | |
739 | set_bit(slot->ds_node_num, o2hb_live_node_bitmap); | |
740 | ||
741 | o2hb_queue_node_event(&event, O2HB_NODE_UP_CB, node, | |
742 | slot->ds_node_num); | |
743 | ||
744 | changed = 1; | |
745 | } | |
746 | ||
747 | list_add_tail(&slot->ds_live_item, | |
748 | &o2hb_live_slots[slot->ds_node_num]); | |
749 | ||
750 | slot->ds_equal_samples = 0; | |
0db638f4 MF |
751 | |
752 | /* We want to be sure that all nodes agree on the | |
753 | * number of milliseconds before a node will be | |
754 | * considered dead. The self-fencing timeout is | |
755 | * computed from this value, and a discrepancy might | |
756 | * result in heartbeat calling a node dead when it | |
757 | * hasn't self-fenced yet. */ | |
758 | slot_dead_ms = le32_to_cpu(hb_block->hb_dead_ms); | |
759 | if (slot_dead_ms && slot_dead_ms != dead_ms) { | |
760 | /* TODO: Perhaps we can fail the region here. */ | |
761 | mlog(ML_ERROR, "Node %d on device %s has a dead count " | |
762 | "of %u ms, but our count is %u ms.\n" | |
763 | "Please double check your configuration values " | |
764 | "for 'O2CB_HEARTBEAT_THRESHOLD'\n", | |
765 | slot->ds_node_num, reg->hr_dev_name, slot_dead_ms, | |
766 | dead_ms); | |
767 | } | |
a7f6a5fb MF |
768 | goto out; |
769 | } | |
770 | ||
771 | /* if the list is dead, we're done.. */ | |
772 | if (list_empty(&slot->ds_live_item)) | |
773 | goto out; | |
774 | ||
775 | /* live nodes only go dead after enough consequtive missed | |
776 | * samples.. reset the missed counter whenever we see | |
777 | * activity */ | |
778 | if (slot->ds_equal_samples >= o2hb_dead_threshold || gen_changed) { | |
779 | mlog(ML_HEARTBEAT, "Node %d left my region\n", | |
780 | slot->ds_node_num); | |
781 | ||
823a637a SM |
782 | clear_bit(slot->ds_node_num, reg->hr_live_node_bitmap); |
783 | ||
a7f6a5fb MF |
784 | /* last off the live_slot generates a callback */ |
785 | list_del_init(&slot->ds_live_item); | |
786 | if (list_empty(&o2hb_live_slots[slot->ds_node_num])) { | |
787 | clear_bit(slot->ds_node_num, o2hb_live_node_bitmap); | |
788 | ||
0e105d37 SM |
789 | /* node can be null */ |
790 | o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, | |
791 | node, slot->ds_node_num); | |
a7f6a5fb MF |
792 | |
793 | changed = 1; | |
794 | } | |
795 | ||
796 | /* We don't clear this because the node is still | |
797 | * actually writing new blocks. */ | |
798 | if (!gen_changed) | |
799 | slot->ds_changed_samples = 0; | |
800 | goto out; | |
801 | } | |
802 | if (slot->ds_changed_samples) { | |
803 | slot->ds_changed_samples = 0; | |
804 | slot->ds_equal_samples = 0; | |
805 | } | |
806 | out: | |
43182d2a SM |
807 | o2hb_set_quorum_device(reg, slot); |
808 | ||
a7f6a5fb MF |
809 | spin_unlock(&o2hb_live_lock); |
810 | ||
811 | o2hb_run_event_list(&event); | |
812 | ||
0e105d37 SM |
813 | if (node) |
814 | o2nm_node_put(node); | |
a7f6a5fb MF |
815 | return changed; |
816 | } | |
817 | ||
818 | /* This could be faster if we just implmented a find_last_bit, but I | |
819 | * don't think the circumstances warrant it. */ | |
820 | static int o2hb_highest_node(unsigned long *nodes, | |
821 | int numbits) | |
822 | { | |
823 | int highest, node; | |
824 | ||
825 | highest = numbits; | |
826 | node = -1; | |
827 | while ((node = find_next_bit(nodes, numbits, node + 1)) != -1) { | |
828 | if (node >= numbits) | |
829 | break; | |
830 | ||
831 | highest = node; | |
832 | } | |
833 | ||
834 | return highest; | |
835 | } | |
836 | ||
a9e2ae39 | 837 | static int o2hb_do_disk_heartbeat(struct o2hb_region *reg) |
a7f6a5fb MF |
838 | { |
839 | int i, ret, highest_node, change = 0; | |
840 | unsigned long configured_nodes[BITS_TO_LONGS(O2NM_MAX_NODES)]; | |
0e105d37 | 841 | unsigned long live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)]; |
a7f6a5fb MF |
842 | struct o2hb_bio_wait_ctxt write_wc; |
843 | ||
a9e2ae39 MF |
844 | ret = o2nm_configured_node_map(configured_nodes, |
845 | sizeof(configured_nodes)); | |
846 | if (ret) { | |
847 | mlog_errno(ret); | |
848 | return ret; | |
849 | } | |
a7f6a5fb | 850 | |
0e105d37 SM |
851 | /* |
852 | * If a node is not configured but is in the livemap, we still need | |
853 | * to read the slot so as to be able to remove it from the livemap. | |
854 | */ | |
855 | o2hb_fill_node_map(live_node_bitmap, sizeof(live_node_bitmap)); | |
856 | i = -1; | |
857 | while ((i = find_next_bit(live_node_bitmap, | |
858 | O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) { | |
859 | set_bit(i, configured_nodes); | |
860 | } | |
861 | ||
a7f6a5fb MF |
862 | highest_node = o2hb_highest_node(configured_nodes, O2NM_MAX_NODES); |
863 | if (highest_node >= O2NM_MAX_NODES) { | |
864 | mlog(ML_NOTICE, "ocfs2_heartbeat: no configured nodes found!\n"); | |
a9e2ae39 | 865 | return -EINVAL; |
a7f6a5fb MF |
866 | } |
867 | ||
868 | /* No sense in reading the slots of nodes that don't exist | |
869 | * yet. Of course, if the node definitions have holes in them | |
870 | * then we're reading an empty slot anyway... Consider this | |
871 | * best-effort. */ | |
872 | ret = o2hb_read_slots(reg, highest_node + 1); | |
873 | if (ret < 0) { | |
874 | mlog_errno(ret); | |
a9e2ae39 | 875 | return ret; |
a7f6a5fb MF |
876 | } |
877 | ||
878 | /* With an up to date view of the slots, we can check that no | |
879 | * other node has been improperly configured to heartbeat in | |
880 | * our slot. */ | |
881 | if (!o2hb_check_last_timestamp(reg)) | |
882 | mlog(ML_ERROR, "Device \"%s\": another node is heartbeating " | |
883 | "in our slot!\n", reg->hr_dev_name); | |
884 | ||
885 | /* fill in the proper info for our next heartbeat */ | |
886 | o2hb_prepare_block(reg, reg->hr_generation); | |
887 | ||
888 | /* And fire off the write. Note that we don't wait on this I/O | |
889 | * until later. */ | |
b559292e | 890 | ret = o2hb_issue_node_write(reg, &write_wc); |
a7f6a5fb MF |
891 | if (ret < 0) { |
892 | mlog_errno(ret); | |
a9e2ae39 | 893 | return ret; |
a7f6a5fb MF |
894 | } |
895 | ||
896 | i = -1; | |
897 | while((i = find_next_bit(configured_nodes, O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) { | |
898 | ||
899 | change |= o2hb_check_slot(reg, ®->hr_slots[i]); | |
900 | } | |
901 | ||
902 | /* | |
903 | * We have to be sure we've advertised ourselves on disk | |
904 | * before we can go to steady state. This ensures that | |
905 | * people we find in our steady state have seen us. | |
906 | */ | |
907 | o2hb_wait_on_io(reg, &write_wc); | |
a9e2ae39 MF |
908 | if (write_wc.wc_error) { |
909 | /* Do not re-arm the write timeout on I/O error - we | |
910 | * can't be sure that the new block ever made it to | |
911 | * disk */ | |
912 | mlog(ML_ERROR, "Write error %d on device \"%s\"\n", | |
913 | write_wc.wc_error, reg->hr_dev_name); | |
914 | return write_wc.wc_error; | |
915 | } | |
916 | ||
a7f6a5fb MF |
917 | o2hb_arm_write_timeout(reg); |
918 | ||
919 | /* let the person who launched us know when things are steady */ | |
920 | if (!change && (atomic_read(®->hr_steady_iterations) != 0)) { | |
921 | if (atomic_dec_and_test(®->hr_steady_iterations)) | |
922 | wake_up(&o2hb_steady_queue); | |
923 | } | |
a9e2ae39 MF |
924 | |
925 | return 0; | |
a7f6a5fb MF |
926 | } |
927 | ||
928 | /* Subtract b from a, storing the result in a. a *must* have a larger | |
929 | * value than b. */ | |
930 | static void o2hb_tv_subtract(struct timeval *a, | |
931 | struct timeval *b) | |
932 | { | |
933 | /* just return 0 when a is after b */ | |
934 | if (a->tv_sec < b->tv_sec || | |
935 | (a->tv_sec == b->tv_sec && a->tv_usec < b->tv_usec)) { | |
936 | a->tv_sec = 0; | |
937 | a->tv_usec = 0; | |
938 | return; | |
939 | } | |
940 | ||
941 | a->tv_sec -= b->tv_sec; | |
942 | a->tv_usec -= b->tv_usec; | |
943 | while ( a->tv_usec < 0 ) { | |
944 | a->tv_sec--; | |
945 | a->tv_usec += 1000000; | |
946 | } | |
947 | } | |
948 | ||
949 | static unsigned int o2hb_elapsed_msecs(struct timeval *start, | |
950 | struct timeval *end) | |
951 | { | |
952 | struct timeval res = *end; | |
953 | ||
954 | o2hb_tv_subtract(&res, start); | |
955 | ||
956 | return res.tv_sec * 1000 + res.tv_usec / 1000; | |
957 | } | |
958 | ||
959 | /* | |
960 | * we ride the region ref that the region dir holds. before the region | |
961 | * dir is removed and drops it ref it will wait to tear down this | |
962 | * thread. | |
963 | */ | |
964 | static int o2hb_thread(void *data) | |
965 | { | |
966 | int i, ret; | |
967 | struct o2hb_region *reg = data; | |
a7f6a5fb MF |
968 | struct o2hb_bio_wait_ctxt write_wc; |
969 | struct timeval before_hb, after_hb; | |
970 | unsigned int elapsed_msec; | |
971 | ||
972 | mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread running\n"); | |
973 | ||
974 | set_user_nice(current, -20); | |
975 | ||
976 | while (!kthread_should_stop() && !reg->hr_unclean_stop) { | |
977 | /* We track the time spent inside | |
025dfdaf | 978 | * o2hb_do_disk_heartbeat so that we avoid more than |
a7f6a5fb MF |
979 | * hr_timeout_ms between disk writes. On busy systems |
980 | * this should result in a heartbeat which is less | |
981 | * likely to time itself out. */ | |
982 | do_gettimeofday(&before_hb); | |
983 | ||
a9e2ae39 MF |
984 | i = 0; |
985 | do { | |
986 | ret = o2hb_do_disk_heartbeat(reg); | |
987 | } while (ret && ++i < 2); | |
a7f6a5fb MF |
988 | |
989 | do_gettimeofday(&after_hb); | |
990 | elapsed_msec = o2hb_elapsed_msecs(&before_hb, &after_hb); | |
991 | ||
b31d308d TM |
992 | mlog(ML_HEARTBEAT, |
993 | "start = %lu.%lu, end = %lu.%lu, msec = %u\n", | |
215c7f9f MF |
994 | before_hb.tv_sec, (unsigned long) before_hb.tv_usec, |
995 | after_hb.tv_sec, (unsigned long) after_hb.tv_usec, | |
996 | elapsed_msec); | |
a7f6a5fb MF |
997 | |
998 | if (elapsed_msec < reg->hr_timeout_ms) { | |
999 | /* the kthread api has blocked signals for us so no | |
1000 | * need to record the return value. */ | |
1001 | msleep_interruptible(reg->hr_timeout_ms - elapsed_msec); | |
1002 | } | |
1003 | } | |
1004 | ||
1005 | o2hb_disarm_write_timeout(reg); | |
1006 | ||
1007 | /* unclean stop is only used in very bad situation */ | |
1008 | for(i = 0; !reg->hr_unclean_stop && i < reg->hr_blocks; i++) | |
1009 | o2hb_shutdown_slot(®->hr_slots[i]); | |
1010 | ||
1011 | /* Explicit down notification - avoid forcing the other nodes | |
1012 | * to timeout on this region when we could just as easily | |
1013 | * write a clear generation - thus indicating to them that | |
1014 | * this node has left this region. | |
1015 | * | |
1016 | * XXX: Should we skip this on unclean_stop? */ | |
1017 | o2hb_prepare_block(reg, 0); | |
b559292e | 1018 | ret = o2hb_issue_node_write(reg, &write_wc); |
a7f6a5fb MF |
1019 | if (ret == 0) { |
1020 | o2hb_wait_on_io(reg, &write_wc); | |
a7f6a5fb MF |
1021 | } else { |
1022 | mlog_errno(ret); | |
1023 | } | |
1024 | ||
1025 | mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread exiting\n"); | |
1026 | ||
1027 | return 0; | |
1028 | } | |
1029 | ||
87d3d3f3 SM |
1030 | #ifdef CONFIG_DEBUG_FS |
1031 | static int o2hb_debug_open(struct inode *inode, struct file *file) | |
1032 | { | |
8ca8b0bb | 1033 | struct o2hb_debug_buf *db = inode->i_private; |
87d3d3f3 SM |
1034 | unsigned long map[BITS_TO_LONGS(O2NM_MAX_NODES)]; |
1035 | char *buf = NULL; | |
1036 | int i = -1; | |
1037 | int out = 0; | |
1038 | ||
8ca8b0bb SM |
1039 | /* max_nodes should be the largest bitmap we pass here */ |
1040 | BUG_ON(sizeof(map) < db->db_size); | |
1041 | ||
87d3d3f3 SM |
1042 | buf = kmalloc(PAGE_SIZE, GFP_KERNEL); |
1043 | if (!buf) | |
1044 | goto bail; | |
1045 | ||
8ca8b0bb SM |
1046 | switch (db->db_type) { |
1047 | case O2HB_DB_TYPE_LIVENODES: | |
1048 | spin_lock(&o2hb_live_lock); | |
1049 | memcpy(map, db->db_data, db->db_size); | |
1050 | spin_unlock(&o2hb_live_lock); | |
1051 | break; | |
87d3d3f3 | 1052 | |
8ca8b0bb SM |
1053 | default: |
1054 | goto done; | |
1055 | } | |
1056 | ||
1057 | while ((i = find_next_bit(map, db->db_len, i + 1)) < db->db_len) | |
87d3d3f3 SM |
1058 | out += snprintf(buf + out, PAGE_SIZE - out, "%d ", i); |
1059 | out += snprintf(buf + out, PAGE_SIZE - out, "\n"); | |
1060 | ||
8ca8b0bb | 1061 | done: |
87d3d3f3 SM |
1062 | i_size_write(inode, out); |
1063 | ||
1064 | file->private_data = buf; | |
1065 | ||
1066 | return 0; | |
1067 | bail: | |
1068 | return -ENOMEM; | |
1069 | } | |
1070 | ||
1071 | static int o2hb_debug_release(struct inode *inode, struct file *file) | |
1072 | { | |
1073 | kfree(file->private_data); | |
1074 | return 0; | |
1075 | } | |
1076 | ||
1077 | static ssize_t o2hb_debug_read(struct file *file, char __user *buf, | |
1078 | size_t nbytes, loff_t *ppos) | |
1079 | { | |
1080 | return simple_read_from_buffer(buf, nbytes, ppos, file->private_data, | |
1081 | i_size_read(file->f_mapping->host)); | |
1082 | } | |
1083 | #else | |
1084 | static int o2hb_debug_open(struct inode *inode, struct file *file) | |
1085 | { | |
1086 | return 0; | |
1087 | } | |
1088 | static int o2hb_debug_release(struct inode *inode, struct file *file) | |
1089 | { | |
1090 | return 0; | |
1091 | } | |
1092 | static ssize_t o2hb_debug_read(struct file *file, char __user *buf, | |
1093 | size_t nbytes, loff_t *ppos) | |
1094 | { | |
1095 | return 0; | |
1096 | } | |
1097 | #endif /* CONFIG_DEBUG_FS */ | |
1098 | ||
828c0950 | 1099 | static const struct file_operations o2hb_debug_fops = { |
87d3d3f3 SM |
1100 | .open = o2hb_debug_open, |
1101 | .release = o2hb_debug_release, | |
1102 | .read = o2hb_debug_read, | |
1103 | .llseek = generic_file_llseek, | |
1104 | }; | |
1105 | ||
1106 | void o2hb_exit(void) | |
1107 | { | |
8ca8b0bb SM |
1108 | kfree(o2hb_db_livenodes); |
1109 | debugfs_remove(o2hb_debug_livenodes); | |
1110 | debugfs_remove(o2hb_debug_dir); | |
1111 | } | |
1112 | ||
1113 | static struct dentry *o2hb_debug_create(const char *name, struct dentry *dir, | |
1114 | struct o2hb_debug_buf **db, int db_len, | |
1115 | int type, int size, int len, void *data) | |
1116 | { | |
1117 | *db = kmalloc(db_len, GFP_KERNEL); | |
1118 | if (!*db) | |
1119 | return NULL; | |
1120 | ||
1121 | (*db)->db_type = type; | |
1122 | (*db)->db_size = size; | |
1123 | (*db)->db_len = len; | |
1124 | (*db)->db_data = data; | |
1125 | ||
1126 | return debugfs_create_file(name, S_IFREG|S_IRUSR, dir, *db, | |
1127 | &o2hb_debug_fops); | |
1128 | } | |
1129 | ||
1130 | static int o2hb_debug_init(void) | |
1131 | { | |
1132 | int ret = -ENOMEM; | |
1133 | ||
1134 | o2hb_debug_dir = debugfs_create_dir(O2HB_DEBUG_DIR, NULL); | |
1135 | if (!o2hb_debug_dir) { | |
1136 | mlog_errno(ret); | |
1137 | goto bail; | |
1138 | } | |
1139 | ||
1140 | o2hb_debug_livenodes = o2hb_debug_create(O2HB_DEBUG_LIVENODES, | |
1141 | o2hb_debug_dir, | |
1142 | &o2hb_db_livenodes, | |
1143 | sizeof(*o2hb_db_livenodes), | |
1144 | O2HB_DB_TYPE_LIVENODES, | |
1145 | sizeof(o2hb_live_node_bitmap), | |
1146 | O2NM_MAX_NODES, | |
1147 | o2hb_live_node_bitmap); | |
1148 | if (!o2hb_debug_livenodes) { | |
1149 | mlog_errno(ret); | |
1150 | goto bail; | |
1151 | } | |
1152 | ret = 0; | |
1153 | bail: | |
1154 | if (ret) | |
1155 | o2hb_exit(); | |
1156 | ||
1157 | return ret; | |
87d3d3f3 SM |
1158 | } |
1159 | ||
1160 | int o2hb_init(void) | |
a7f6a5fb MF |
1161 | { |
1162 | int i; | |
1163 | ||
1164 | for (i = 0; i < ARRAY_SIZE(o2hb_callbacks); i++) | |
1165 | INIT_LIST_HEAD(&o2hb_callbacks[i].list); | |
1166 | ||
1167 | for (i = 0; i < ARRAY_SIZE(o2hb_live_slots); i++) | |
1168 | INIT_LIST_HEAD(&o2hb_live_slots[i]); | |
1169 | ||
1170 | INIT_LIST_HEAD(&o2hb_node_events); | |
1171 | ||
1172 | memset(o2hb_live_node_bitmap, 0, sizeof(o2hb_live_node_bitmap)); | |
536f0741 | 1173 | memset(o2hb_region_bitmap, 0, sizeof(o2hb_region_bitmap)); |
e7d656ba | 1174 | memset(o2hb_live_region_bitmap, 0, sizeof(o2hb_live_region_bitmap)); |
43182d2a | 1175 | memset(o2hb_quorum_region_bitmap, 0, sizeof(o2hb_quorum_region_bitmap)); |
87d3d3f3 | 1176 | |
8ca8b0bb | 1177 | return o2hb_debug_init(); |
a7f6a5fb MF |
1178 | } |
1179 | ||
1180 | /* if we're already in a callback then we're already serialized by the sem */ | |
1181 | static void o2hb_fill_node_map_from_callback(unsigned long *map, | |
1182 | unsigned bytes) | |
1183 | { | |
1184 | BUG_ON(bytes < (BITS_TO_LONGS(O2NM_MAX_NODES) * sizeof(unsigned long))); | |
1185 | ||
1186 | memcpy(map, &o2hb_live_node_bitmap, bytes); | |
1187 | } | |
1188 | ||
1189 | /* | |
1190 | * get a map of all nodes that are heartbeating in any regions | |
1191 | */ | |
1192 | void o2hb_fill_node_map(unsigned long *map, unsigned bytes) | |
1193 | { | |
1194 | /* callers want to serialize this map and callbacks so that they | |
1195 | * can trust that they don't miss nodes coming to the party */ | |
1196 | down_read(&o2hb_callback_sem); | |
1197 | spin_lock(&o2hb_live_lock); | |
1198 | o2hb_fill_node_map_from_callback(map, bytes); | |
1199 | spin_unlock(&o2hb_live_lock); | |
1200 | up_read(&o2hb_callback_sem); | |
1201 | } | |
1202 | EXPORT_SYMBOL_GPL(o2hb_fill_node_map); | |
1203 | ||
1204 | /* | |
1205 | * heartbeat configfs bits. The heartbeat set is a default set under | |
1206 | * the cluster set in nodemanager.c. | |
1207 | */ | |
1208 | ||
1209 | static struct o2hb_region *to_o2hb_region(struct config_item *item) | |
1210 | { | |
1211 | return item ? container_of(item, struct o2hb_region, hr_item) : NULL; | |
1212 | } | |
1213 | ||
1214 | /* drop_item only drops its ref after killing the thread, nothing should | |
1215 | * be using the region anymore. this has to clean up any state that | |
1216 | * attributes might have built up. */ | |
1217 | static void o2hb_region_release(struct config_item *item) | |
1218 | { | |
1219 | int i; | |
1220 | struct page *page; | |
1221 | struct o2hb_region *reg = to_o2hb_region(item); | |
1222 | ||
1223 | if (reg->hr_tmp_block) | |
1224 | kfree(reg->hr_tmp_block); | |
1225 | ||
1226 | if (reg->hr_slot_data) { | |
1227 | for (i = 0; i < reg->hr_num_pages; i++) { | |
1228 | page = reg->hr_slot_data[i]; | |
1229 | if (page) | |
1230 | __free_page(page); | |
1231 | } | |
1232 | kfree(reg->hr_slot_data); | |
1233 | } | |
1234 | ||
1235 | if (reg->hr_bdev) | |
9a1c3542 | 1236 | blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE); |
a7f6a5fb MF |
1237 | |
1238 | if (reg->hr_slots) | |
1239 | kfree(reg->hr_slots); | |
1240 | ||
1241 | spin_lock(&o2hb_live_lock); | |
1242 | list_del(®->hr_all_item); | |
1243 | spin_unlock(&o2hb_live_lock); | |
1244 | ||
1245 | kfree(reg); | |
1246 | } | |
1247 | ||
1248 | static int o2hb_read_block_input(struct o2hb_region *reg, | |
1249 | const char *page, | |
1250 | size_t count, | |
1251 | unsigned long *ret_bytes, | |
1252 | unsigned int *ret_bits) | |
1253 | { | |
1254 | unsigned long bytes; | |
1255 | char *p = (char *)page; | |
1256 | ||
1257 | bytes = simple_strtoul(p, &p, 0); | |
1258 | if (!p || (*p && (*p != '\n'))) | |
1259 | return -EINVAL; | |
1260 | ||
1261 | /* Heartbeat and fs min / max block sizes are the same. */ | |
1262 | if (bytes > 4096 || bytes < 512) | |
1263 | return -ERANGE; | |
1264 | if (hweight16(bytes) != 1) | |
1265 | return -EINVAL; | |
1266 | ||
1267 | if (ret_bytes) | |
1268 | *ret_bytes = bytes; | |
1269 | if (ret_bits) | |
1270 | *ret_bits = ffs(bytes) - 1; | |
1271 | ||
1272 | return 0; | |
1273 | } | |
1274 | ||
1275 | static ssize_t o2hb_region_block_bytes_read(struct o2hb_region *reg, | |
1276 | char *page) | |
1277 | { | |
1278 | return sprintf(page, "%u\n", reg->hr_block_bytes); | |
1279 | } | |
1280 | ||
1281 | static ssize_t o2hb_region_block_bytes_write(struct o2hb_region *reg, | |
1282 | const char *page, | |
1283 | size_t count) | |
1284 | { | |
1285 | int status; | |
1286 | unsigned long block_bytes; | |
1287 | unsigned int block_bits; | |
1288 | ||
1289 | if (reg->hr_bdev) | |
1290 | return -EINVAL; | |
1291 | ||
1292 | status = o2hb_read_block_input(reg, page, count, | |
1293 | &block_bytes, &block_bits); | |
1294 | if (status) | |
1295 | return status; | |
1296 | ||
1297 | reg->hr_block_bytes = (unsigned int)block_bytes; | |
1298 | reg->hr_block_bits = block_bits; | |
1299 | ||
1300 | return count; | |
1301 | } | |
1302 | ||
1303 | static ssize_t o2hb_region_start_block_read(struct o2hb_region *reg, | |
1304 | char *page) | |
1305 | { | |
1306 | return sprintf(page, "%llu\n", reg->hr_start_block); | |
1307 | } | |
1308 | ||
1309 | static ssize_t o2hb_region_start_block_write(struct o2hb_region *reg, | |
1310 | const char *page, | |
1311 | size_t count) | |
1312 | { | |
1313 | unsigned long long tmp; | |
1314 | char *p = (char *)page; | |
1315 | ||
1316 | if (reg->hr_bdev) | |
1317 | return -EINVAL; | |
1318 | ||
1319 | tmp = simple_strtoull(p, &p, 0); | |
1320 | if (!p || (*p && (*p != '\n'))) | |
1321 | return -EINVAL; | |
1322 | ||
1323 | reg->hr_start_block = tmp; | |
1324 | ||
1325 | return count; | |
1326 | } | |
1327 | ||
1328 | static ssize_t o2hb_region_blocks_read(struct o2hb_region *reg, | |
1329 | char *page) | |
1330 | { | |
1331 | return sprintf(page, "%d\n", reg->hr_blocks); | |
1332 | } | |
1333 | ||
1334 | static ssize_t o2hb_region_blocks_write(struct o2hb_region *reg, | |
1335 | const char *page, | |
1336 | size_t count) | |
1337 | { | |
1338 | unsigned long tmp; | |
1339 | char *p = (char *)page; | |
1340 | ||
1341 | if (reg->hr_bdev) | |
1342 | return -EINVAL; | |
1343 | ||
1344 | tmp = simple_strtoul(p, &p, 0); | |
1345 | if (!p || (*p && (*p != '\n'))) | |
1346 | return -EINVAL; | |
1347 | ||
1348 | if (tmp > O2NM_MAX_NODES || tmp == 0) | |
1349 | return -ERANGE; | |
1350 | ||
1351 | reg->hr_blocks = (unsigned int)tmp; | |
1352 | ||
1353 | return count; | |
1354 | } | |
1355 | ||
1356 | static ssize_t o2hb_region_dev_read(struct o2hb_region *reg, | |
1357 | char *page) | |
1358 | { | |
1359 | unsigned int ret = 0; | |
1360 | ||
1361 | if (reg->hr_bdev) | |
1362 | ret = sprintf(page, "%s\n", reg->hr_dev_name); | |
1363 | ||
1364 | return ret; | |
1365 | } | |
1366 | ||
1367 | static void o2hb_init_region_params(struct o2hb_region *reg) | |
1368 | { | |
1369 | reg->hr_slots_per_page = PAGE_CACHE_SIZE >> reg->hr_block_bits; | |
1370 | reg->hr_timeout_ms = O2HB_REGION_TIMEOUT_MS; | |
1371 | ||
1372 | mlog(ML_HEARTBEAT, "hr_start_block = %llu, hr_blocks = %u\n", | |
1373 | reg->hr_start_block, reg->hr_blocks); | |
1374 | mlog(ML_HEARTBEAT, "hr_block_bytes = %u, hr_block_bits = %u\n", | |
1375 | reg->hr_block_bytes, reg->hr_block_bits); | |
1376 | mlog(ML_HEARTBEAT, "hr_timeout_ms = %u\n", reg->hr_timeout_ms); | |
1377 | mlog(ML_HEARTBEAT, "dead threshold = %u\n", o2hb_dead_threshold); | |
1378 | } | |
1379 | ||
1380 | static int o2hb_map_slot_data(struct o2hb_region *reg) | |
1381 | { | |
1382 | int i, j; | |
1383 | unsigned int last_slot; | |
1384 | unsigned int spp = reg->hr_slots_per_page; | |
1385 | struct page *page; | |
1386 | char *raw; | |
1387 | struct o2hb_disk_slot *slot; | |
1388 | ||
1389 | reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL); | |
1390 | if (reg->hr_tmp_block == NULL) { | |
1391 | mlog_errno(-ENOMEM); | |
1392 | return -ENOMEM; | |
1393 | } | |
1394 | ||
1395 | reg->hr_slots = kcalloc(reg->hr_blocks, | |
1396 | sizeof(struct o2hb_disk_slot), GFP_KERNEL); | |
1397 | if (reg->hr_slots == NULL) { | |
1398 | mlog_errno(-ENOMEM); | |
1399 | return -ENOMEM; | |
1400 | } | |
1401 | ||
1402 | for(i = 0; i < reg->hr_blocks; i++) { | |
1403 | slot = ®->hr_slots[i]; | |
1404 | slot->ds_node_num = i; | |
1405 | INIT_LIST_HEAD(&slot->ds_live_item); | |
1406 | slot->ds_raw_block = NULL; | |
1407 | } | |
1408 | ||
1409 | reg->hr_num_pages = (reg->hr_blocks + spp - 1) / spp; | |
1410 | mlog(ML_HEARTBEAT, "Going to require %u pages to cover %u blocks " | |
1411 | "at %u blocks per page\n", | |
1412 | reg->hr_num_pages, reg->hr_blocks, spp); | |
1413 | ||
1414 | reg->hr_slot_data = kcalloc(reg->hr_num_pages, sizeof(struct page *), | |
1415 | GFP_KERNEL); | |
1416 | if (!reg->hr_slot_data) { | |
1417 | mlog_errno(-ENOMEM); | |
1418 | return -ENOMEM; | |
1419 | } | |
1420 | ||
1421 | for(i = 0; i < reg->hr_num_pages; i++) { | |
1422 | page = alloc_page(GFP_KERNEL); | |
1423 | if (!page) { | |
1424 | mlog_errno(-ENOMEM); | |
1425 | return -ENOMEM; | |
1426 | } | |
1427 | ||
1428 | reg->hr_slot_data[i] = page; | |
1429 | ||
1430 | last_slot = i * spp; | |
1431 | raw = page_address(page); | |
1432 | for (j = 0; | |
1433 | (j < spp) && ((j + last_slot) < reg->hr_blocks); | |
1434 | j++) { | |
1435 | BUG_ON((j + last_slot) >= reg->hr_blocks); | |
1436 | ||
1437 | slot = ®->hr_slots[j + last_slot]; | |
1438 | slot->ds_raw_block = | |
1439 | (struct o2hb_disk_heartbeat_block *) raw; | |
1440 | ||
1441 | raw += reg->hr_block_bytes; | |
1442 | } | |
1443 | } | |
1444 | ||
1445 | return 0; | |
1446 | } | |
1447 | ||
1448 | /* Read in all the slots available and populate the tracking | |
1449 | * structures so that we can start with a baseline idea of what's | |
1450 | * there. */ | |
1451 | static int o2hb_populate_slot_data(struct o2hb_region *reg) | |
1452 | { | |
1453 | int ret, i; | |
1454 | struct o2hb_disk_slot *slot; | |
1455 | struct o2hb_disk_heartbeat_block *hb_block; | |
1456 | ||
1457 | mlog_entry_void(); | |
1458 | ||
1459 | ret = o2hb_read_slots(reg, reg->hr_blocks); | |
1460 | if (ret) { | |
1461 | mlog_errno(ret); | |
1462 | goto out; | |
1463 | } | |
1464 | ||
1465 | /* We only want to get an idea of the values initially in each | |
1466 | * slot, so we do no verification - o2hb_check_slot will | |
1467 | * actually determine if each configured slot is valid and | |
1468 | * whether any values have changed. */ | |
1469 | for(i = 0; i < reg->hr_blocks; i++) { | |
1470 | slot = ®->hr_slots[i]; | |
1471 | hb_block = (struct o2hb_disk_heartbeat_block *) slot->ds_raw_block; | |
1472 | ||
1473 | /* Only fill the values that o2hb_check_slot uses to | |
1474 | * determine changing slots */ | |
1475 | slot->ds_last_time = le64_to_cpu(hb_block->hb_seq); | |
1476 | slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation); | |
1477 | } | |
1478 | ||
1479 | out: | |
1480 | mlog_exit(ret); | |
1481 | return ret; | |
1482 | } | |
1483 | ||
1484 | /* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */ | |
1485 | static ssize_t o2hb_region_dev_write(struct o2hb_region *reg, | |
1486 | const char *page, | |
1487 | size_t count) | |
1488 | { | |
e6c352db | 1489 | struct task_struct *hb_task; |
a7f6a5fb MF |
1490 | long fd; |
1491 | int sectsize; | |
1492 | char *p = (char *)page; | |
1493 | struct file *filp = NULL; | |
1494 | struct inode *inode = NULL; | |
1495 | ssize_t ret = -EINVAL; | |
1496 | ||
1497 | if (reg->hr_bdev) | |
1498 | goto out; | |
1499 | ||
1500 | /* We can't heartbeat without having had our node number | |
1501 | * configured yet. */ | |
1502 | if (o2nm_this_node() == O2NM_MAX_NODES) | |
1503 | goto out; | |
1504 | ||
1505 | fd = simple_strtol(p, &p, 0); | |
1506 | if (!p || (*p && (*p != '\n'))) | |
1507 | goto out; | |
1508 | ||
1509 | if (fd < 0 || fd >= INT_MAX) | |
1510 | goto out; | |
1511 | ||
1512 | filp = fget(fd); | |
1513 | if (filp == NULL) | |
1514 | goto out; | |
1515 | ||
1516 | if (reg->hr_blocks == 0 || reg->hr_start_block == 0 || | |
1517 | reg->hr_block_bytes == 0) | |
1518 | goto out; | |
1519 | ||
1520 | inode = igrab(filp->f_mapping->host); | |
1521 | if (inode == NULL) | |
1522 | goto out; | |
1523 | ||
1524 | if (!S_ISBLK(inode->i_mode)) | |
1525 | goto out; | |
1526 | ||
1527 | reg->hr_bdev = I_BDEV(filp->f_mapping->host); | |
572c4892 | 1528 | ret = blkdev_get(reg->hr_bdev, FMODE_WRITE | FMODE_READ); |
a7f6a5fb MF |
1529 | if (ret) { |
1530 | reg->hr_bdev = NULL; | |
1531 | goto out; | |
1532 | } | |
1533 | inode = NULL; | |
1534 | ||
1535 | bdevname(reg->hr_bdev, reg->hr_dev_name); | |
1536 | ||
e1defc4f | 1537 | sectsize = bdev_logical_block_size(reg->hr_bdev); |
a7f6a5fb MF |
1538 | if (sectsize != reg->hr_block_bytes) { |
1539 | mlog(ML_ERROR, | |
1540 | "blocksize %u incorrect for device, expected %d", | |
1541 | reg->hr_block_bytes, sectsize); | |
1542 | ret = -EINVAL; | |
1543 | goto out; | |
1544 | } | |
1545 | ||
1546 | o2hb_init_region_params(reg); | |
1547 | ||
1548 | /* Generation of zero is invalid */ | |
1549 | do { | |
1550 | get_random_bytes(®->hr_generation, | |
1551 | sizeof(reg->hr_generation)); | |
1552 | } while (reg->hr_generation == 0); | |
1553 | ||
1554 | ret = o2hb_map_slot_data(reg); | |
1555 | if (ret) { | |
1556 | mlog_errno(ret); | |
1557 | goto out; | |
1558 | } | |
1559 | ||
1560 | ret = o2hb_populate_slot_data(reg); | |
1561 | if (ret) { | |
1562 | mlog_errno(ret); | |
1563 | goto out; | |
1564 | } | |
1565 | ||
c4028958 | 1566 | INIT_DELAYED_WORK(®->hr_write_timeout_work, o2hb_write_timeout); |
a7f6a5fb MF |
1567 | |
1568 | /* | |
1569 | * A node is considered live after it has beat LIVE_THRESHOLD | |
1570 | * times. We're not steady until we've given them a chance | |
1571 | * _after_ our first read. | |
1572 | */ | |
1573 | atomic_set(®->hr_steady_iterations, O2HB_LIVE_THRESHOLD + 1); | |
1574 | ||
e6c352db JB |
1575 | hb_task = kthread_run(o2hb_thread, reg, "o2hb-%s", |
1576 | reg->hr_item.ci_name); | |
1577 | if (IS_ERR(hb_task)) { | |
1578 | ret = PTR_ERR(hb_task); | |
a7f6a5fb | 1579 | mlog_errno(ret); |
a7f6a5fb MF |
1580 | goto out; |
1581 | } | |
1582 | ||
e6c352db JB |
1583 | spin_lock(&o2hb_live_lock); |
1584 | reg->hr_task = hb_task; | |
1585 | spin_unlock(&o2hb_live_lock); | |
1586 | ||
a7f6a5fb MF |
1587 | ret = wait_event_interruptible(o2hb_steady_queue, |
1588 | atomic_read(®->hr_steady_iterations) == 0); | |
1589 | if (ret) { | |
e6df3a66 | 1590 | /* We got interrupted (hello ptrace!). Clean up */ |
e6c352db JB |
1591 | spin_lock(&o2hb_live_lock); |
1592 | hb_task = reg->hr_task; | |
a7f6a5fb | 1593 | reg->hr_task = NULL; |
e6c352db JB |
1594 | spin_unlock(&o2hb_live_lock); |
1595 | ||
1596 | if (hb_task) | |
1597 | kthread_stop(hb_task); | |
a7f6a5fb MF |
1598 | goto out; |
1599 | } | |
1600 | ||
e6df3a66 JB |
1601 | /* Ok, we were woken. Make sure it wasn't by drop_item() */ |
1602 | spin_lock(&o2hb_live_lock); | |
1603 | hb_task = reg->hr_task; | |
e7d656ba SM |
1604 | if (o2hb_global_heartbeat_active()) |
1605 | set_bit(reg->hr_region_num, o2hb_live_region_bitmap); | |
e6df3a66 JB |
1606 | spin_unlock(&o2hb_live_lock); |
1607 | ||
1608 | if (hb_task) | |
1609 | ret = count; | |
1610 | else | |
1611 | ret = -EIO; | |
1612 | ||
18c50cb0 SM |
1613 | if (hb_task && o2hb_global_heartbeat_active()) |
1614 | printk(KERN_NOTICE "o2hb: Heartbeat started on region %s\n", | |
1615 | config_item_name(®->hr_item)); | |
1616 | ||
a7f6a5fb MF |
1617 | out: |
1618 | if (filp) | |
1619 | fput(filp); | |
1620 | if (inode) | |
1621 | iput(inode); | |
1622 | if (ret < 0) { | |
1623 | if (reg->hr_bdev) { | |
9a1c3542 | 1624 | blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE); |
a7f6a5fb MF |
1625 | reg->hr_bdev = NULL; |
1626 | } | |
1627 | } | |
1628 | return ret; | |
1629 | } | |
1630 | ||
92efc152 ZW |
1631 | static ssize_t o2hb_region_pid_read(struct o2hb_region *reg, |
1632 | char *page) | |
1633 | { | |
e6c352db JB |
1634 | pid_t pid = 0; |
1635 | ||
1636 | spin_lock(&o2hb_live_lock); | |
1637 | if (reg->hr_task) | |
ba25f9dc | 1638 | pid = task_pid_nr(reg->hr_task); |
e6c352db JB |
1639 | spin_unlock(&o2hb_live_lock); |
1640 | ||
1641 | if (!pid) | |
92efc152 ZW |
1642 | return 0; |
1643 | ||
e6c352db | 1644 | return sprintf(page, "%u\n", pid); |
92efc152 ZW |
1645 | } |
1646 | ||
a7f6a5fb MF |
1647 | struct o2hb_region_attribute { |
1648 | struct configfs_attribute attr; | |
1649 | ssize_t (*show)(struct o2hb_region *, char *); | |
1650 | ssize_t (*store)(struct o2hb_region *, const char *, size_t); | |
1651 | }; | |
1652 | ||
1653 | static struct o2hb_region_attribute o2hb_region_attr_block_bytes = { | |
1654 | .attr = { .ca_owner = THIS_MODULE, | |
1655 | .ca_name = "block_bytes", | |
1656 | .ca_mode = S_IRUGO | S_IWUSR }, | |
1657 | .show = o2hb_region_block_bytes_read, | |
1658 | .store = o2hb_region_block_bytes_write, | |
1659 | }; | |
1660 | ||
1661 | static struct o2hb_region_attribute o2hb_region_attr_start_block = { | |
1662 | .attr = { .ca_owner = THIS_MODULE, | |
1663 | .ca_name = "start_block", | |
1664 | .ca_mode = S_IRUGO | S_IWUSR }, | |
1665 | .show = o2hb_region_start_block_read, | |
1666 | .store = o2hb_region_start_block_write, | |
1667 | }; | |
1668 | ||
1669 | static struct o2hb_region_attribute o2hb_region_attr_blocks = { | |
1670 | .attr = { .ca_owner = THIS_MODULE, | |
1671 | .ca_name = "blocks", | |
1672 | .ca_mode = S_IRUGO | S_IWUSR }, | |
1673 | .show = o2hb_region_blocks_read, | |
1674 | .store = o2hb_region_blocks_write, | |
1675 | }; | |
1676 | ||
1677 | static struct o2hb_region_attribute o2hb_region_attr_dev = { | |
1678 | .attr = { .ca_owner = THIS_MODULE, | |
1679 | .ca_name = "dev", | |
1680 | .ca_mode = S_IRUGO | S_IWUSR }, | |
1681 | .show = o2hb_region_dev_read, | |
1682 | .store = o2hb_region_dev_write, | |
1683 | }; | |
1684 | ||
92efc152 ZW |
1685 | static struct o2hb_region_attribute o2hb_region_attr_pid = { |
1686 | .attr = { .ca_owner = THIS_MODULE, | |
1687 | .ca_name = "pid", | |
1688 | .ca_mode = S_IRUGO | S_IRUSR }, | |
1689 | .show = o2hb_region_pid_read, | |
1690 | }; | |
1691 | ||
a7f6a5fb MF |
1692 | static struct configfs_attribute *o2hb_region_attrs[] = { |
1693 | &o2hb_region_attr_block_bytes.attr, | |
1694 | &o2hb_region_attr_start_block.attr, | |
1695 | &o2hb_region_attr_blocks.attr, | |
1696 | &o2hb_region_attr_dev.attr, | |
92efc152 | 1697 | &o2hb_region_attr_pid.attr, |
a7f6a5fb MF |
1698 | NULL, |
1699 | }; | |
1700 | ||
1701 | static ssize_t o2hb_region_show(struct config_item *item, | |
1702 | struct configfs_attribute *attr, | |
1703 | char *page) | |
1704 | { | |
1705 | struct o2hb_region *reg = to_o2hb_region(item); | |
1706 | struct o2hb_region_attribute *o2hb_region_attr = | |
1707 | container_of(attr, struct o2hb_region_attribute, attr); | |
1708 | ssize_t ret = 0; | |
1709 | ||
1710 | if (o2hb_region_attr->show) | |
1711 | ret = o2hb_region_attr->show(reg, page); | |
1712 | return ret; | |
1713 | } | |
1714 | ||
1715 | static ssize_t o2hb_region_store(struct config_item *item, | |
1716 | struct configfs_attribute *attr, | |
1717 | const char *page, size_t count) | |
1718 | { | |
1719 | struct o2hb_region *reg = to_o2hb_region(item); | |
1720 | struct o2hb_region_attribute *o2hb_region_attr = | |
1721 | container_of(attr, struct o2hb_region_attribute, attr); | |
1722 | ssize_t ret = -EINVAL; | |
1723 | ||
1724 | if (o2hb_region_attr->store) | |
1725 | ret = o2hb_region_attr->store(reg, page, count); | |
1726 | return ret; | |
1727 | } | |
1728 | ||
1729 | static struct configfs_item_operations o2hb_region_item_ops = { | |
1730 | .release = o2hb_region_release, | |
1731 | .show_attribute = o2hb_region_show, | |
1732 | .store_attribute = o2hb_region_store, | |
1733 | }; | |
1734 | ||
1735 | static struct config_item_type o2hb_region_type = { | |
1736 | .ct_item_ops = &o2hb_region_item_ops, | |
1737 | .ct_attrs = o2hb_region_attrs, | |
1738 | .ct_owner = THIS_MODULE, | |
1739 | }; | |
1740 | ||
1741 | /* heartbeat set */ | |
1742 | ||
1743 | struct o2hb_heartbeat_group { | |
1744 | struct config_group hs_group; | |
1745 | /* some stuff? */ | |
1746 | }; | |
1747 | ||
1748 | static struct o2hb_heartbeat_group *to_o2hb_heartbeat_group(struct config_group *group) | |
1749 | { | |
1750 | return group ? | |
1751 | container_of(group, struct o2hb_heartbeat_group, hs_group) | |
1752 | : NULL; | |
1753 | } | |
1754 | ||
f89ab861 JB |
1755 | static struct config_item *o2hb_heartbeat_group_make_item(struct config_group *group, |
1756 | const char *name) | |
a7f6a5fb MF |
1757 | { |
1758 | struct o2hb_region *reg = NULL; | |
a7f6a5fb | 1759 | |
cd861280 | 1760 | reg = kzalloc(sizeof(struct o2hb_region), GFP_KERNEL); |
f89ab861 | 1761 | if (reg == NULL) |
a6795e9e | 1762 | return ERR_PTR(-ENOMEM); |
a7f6a5fb | 1763 | |
b3c85c4c SM |
1764 | if (strlen(name) > O2HB_MAX_REGION_NAME_LEN) |
1765 | return ERR_PTR(-ENAMETOOLONG); | |
1766 | ||
a7f6a5fb | 1767 | spin_lock(&o2hb_live_lock); |
536f0741 SM |
1768 | reg->hr_region_num = 0; |
1769 | if (o2hb_global_heartbeat_active()) { | |
1770 | reg->hr_region_num = find_first_zero_bit(o2hb_region_bitmap, | |
1771 | O2NM_MAX_REGIONS); | |
1772 | if (reg->hr_region_num >= O2NM_MAX_REGIONS) { | |
1773 | spin_unlock(&o2hb_live_lock); | |
1774 | return ERR_PTR(-EFBIG); | |
1775 | } | |
1776 | set_bit(reg->hr_region_num, o2hb_region_bitmap); | |
1777 | } | |
a7f6a5fb MF |
1778 | list_add_tail(®->hr_all_item, &o2hb_all_regions); |
1779 | spin_unlock(&o2hb_live_lock); | |
a7f6a5fb | 1780 | |
536f0741 SM |
1781 | config_item_init_type_name(®->hr_item, name, &o2hb_region_type); |
1782 | ||
a6795e9e | 1783 | return ®->hr_item; |
a7f6a5fb MF |
1784 | } |
1785 | ||
1786 | static void o2hb_heartbeat_group_drop_item(struct config_group *group, | |
1787 | struct config_item *item) | |
1788 | { | |
e6c352db | 1789 | struct task_struct *hb_task; |
a7f6a5fb MF |
1790 | struct o2hb_region *reg = to_o2hb_region(item); |
1791 | ||
1792 | /* stop the thread when the user removes the region dir */ | |
e6c352db | 1793 | spin_lock(&o2hb_live_lock); |
e7d656ba | 1794 | if (o2hb_global_heartbeat_active()) { |
536f0741 | 1795 | clear_bit(reg->hr_region_num, o2hb_region_bitmap); |
e7d656ba SM |
1796 | clear_bit(reg->hr_region_num, o2hb_live_region_bitmap); |
1797 | } | |
e6c352db JB |
1798 | hb_task = reg->hr_task; |
1799 | reg->hr_task = NULL; | |
1800 | spin_unlock(&o2hb_live_lock); | |
1801 | ||
1802 | if (hb_task) | |
1803 | kthread_stop(hb_task); | |
a7f6a5fb | 1804 | |
e6df3a66 JB |
1805 | /* |
1806 | * If we're racing a dev_write(), we need to wake them. They will | |
1807 | * check reg->hr_task | |
1808 | */ | |
1809 | if (atomic_read(®->hr_steady_iterations) != 0) { | |
1810 | atomic_set(®->hr_steady_iterations, 0); | |
1811 | wake_up(&o2hb_steady_queue); | |
1812 | } | |
1813 | ||
18c50cb0 SM |
1814 | if (o2hb_global_heartbeat_active()) |
1815 | printk(KERN_NOTICE "o2hb: Heartbeat stopped on region %s\n", | |
1816 | config_item_name(®->hr_item)); | |
a7f6a5fb MF |
1817 | config_item_put(item); |
1818 | } | |
1819 | ||
1820 | struct o2hb_heartbeat_group_attribute { | |
1821 | struct configfs_attribute attr; | |
1822 | ssize_t (*show)(struct o2hb_heartbeat_group *, char *); | |
1823 | ssize_t (*store)(struct o2hb_heartbeat_group *, const char *, size_t); | |
1824 | }; | |
1825 | ||
1826 | static ssize_t o2hb_heartbeat_group_show(struct config_item *item, | |
1827 | struct configfs_attribute *attr, | |
1828 | char *page) | |
1829 | { | |
1830 | struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item)); | |
1831 | struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr = | |
1832 | container_of(attr, struct o2hb_heartbeat_group_attribute, attr); | |
1833 | ssize_t ret = 0; | |
1834 | ||
1835 | if (o2hb_heartbeat_group_attr->show) | |
1836 | ret = o2hb_heartbeat_group_attr->show(reg, page); | |
1837 | return ret; | |
1838 | } | |
1839 | ||
1840 | static ssize_t o2hb_heartbeat_group_store(struct config_item *item, | |
1841 | struct configfs_attribute *attr, | |
1842 | const char *page, size_t count) | |
1843 | { | |
1844 | struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item)); | |
1845 | struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr = | |
1846 | container_of(attr, struct o2hb_heartbeat_group_attribute, attr); | |
1847 | ssize_t ret = -EINVAL; | |
1848 | ||
1849 | if (o2hb_heartbeat_group_attr->store) | |
1850 | ret = o2hb_heartbeat_group_attr->store(reg, page, count); | |
1851 | return ret; | |
1852 | } | |
1853 | ||
1854 | static ssize_t o2hb_heartbeat_group_threshold_show(struct o2hb_heartbeat_group *group, | |
1855 | char *page) | |
1856 | { | |
1857 | return sprintf(page, "%u\n", o2hb_dead_threshold); | |
1858 | } | |
1859 | ||
1860 | static ssize_t o2hb_heartbeat_group_threshold_store(struct o2hb_heartbeat_group *group, | |
1861 | const char *page, | |
1862 | size_t count) | |
1863 | { | |
1864 | unsigned long tmp; | |
1865 | char *p = (char *)page; | |
1866 | ||
1867 | tmp = simple_strtoul(p, &p, 10); | |
1868 | if (!p || (*p && (*p != '\n'))) | |
1869 | return -EINVAL; | |
1870 | ||
1871 | /* this will validate ranges for us. */ | |
1872 | o2hb_dead_threshold_set((unsigned int) tmp); | |
1873 | ||
1874 | return count; | |
1875 | } | |
1876 | ||
54b5187b SM |
1877 | static |
1878 | ssize_t o2hb_heartbeat_group_mode_show(struct o2hb_heartbeat_group *group, | |
1879 | char *page) | |
1880 | { | |
1881 | return sprintf(page, "%s\n", | |
1882 | o2hb_heartbeat_mode_desc[o2hb_heartbeat_mode]); | |
1883 | } | |
1884 | ||
1885 | static | |
1886 | ssize_t o2hb_heartbeat_group_mode_store(struct o2hb_heartbeat_group *group, | |
1887 | const char *page, size_t count) | |
1888 | { | |
1889 | unsigned int i; | |
1890 | int ret; | |
1891 | size_t len; | |
1892 | ||
1893 | len = (page[count - 1] == '\n') ? count - 1 : count; | |
1894 | if (!len) | |
1895 | return -EINVAL; | |
1896 | ||
1897 | for (i = 0; i < O2HB_HEARTBEAT_NUM_MODES; ++i) { | |
1898 | if (strnicmp(page, o2hb_heartbeat_mode_desc[i], len)) | |
1899 | continue; | |
1900 | ||
1901 | ret = o2hb_global_hearbeat_mode_set(i); | |
1902 | if (!ret) | |
18c50cb0 | 1903 | printk(KERN_NOTICE "o2hb: Heartbeat mode set to %s\n", |
54b5187b SM |
1904 | o2hb_heartbeat_mode_desc[i]); |
1905 | return count; | |
1906 | } | |
1907 | ||
1908 | return -EINVAL; | |
1909 | ||
1910 | } | |
1911 | ||
a7f6a5fb MF |
1912 | static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_threshold = { |
1913 | .attr = { .ca_owner = THIS_MODULE, | |
1914 | .ca_name = "dead_threshold", | |
1915 | .ca_mode = S_IRUGO | S_IWUSR }, | |
1916 | .show = o2hb_heartbeat_group_threshold_show, | |
1917 | .store = o2hb_heartbeat_group_threshold_store, | |
1918 | }; | |
1919 | ||
54b5187b SM |
1920 | static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_mode = { |
1921 | .attr = { .ca_owner = THIS_MODULE, | |
1922 | .ca_name = "mode", | |
1923 | .ca_mode = S_IRUGO | S_IWUSR }, | |
1924 | .show = o2hb_heartbeat_group_mode_show, | |
1925 | .store = o2hb_heartbeat_group_mode_store, | |
1926 | }; | |
1927 | ||
a7f6a5fb MF |
1928 | static struct configfs_attribute *o2hb_heartbeat_group_attrs[] = { |
1929 | &o2hb_heartbeat_group_attr_threshold.attr, | |
54b5187b | 1930 | &o2hb_heartbeat_group_attr_mode.attr, |
a7f6a5fb MF |
1931 | NULL, |
1932 | }; | |
1933 | ||
1934 | static struct configfs_item_operations o2hb_hearbeat_group_item_ops = { | |
1935 | .show_attribute = o2hb_heartbeat_group_show, | |
1936 | .store_attribute = o2hb_heartbeat_group_store, | |
1937 | }; | |
1938 | ||
1939 | static struct configfs_group_operations o2hb_heartbeat_group_group_ops = { | |
1940 | .make_item = o2hb_heartbeat_group_make_item, | |
1941 | .drop_item = o2hb_heartbeat_group_drop_item, | |
1942 | }; | |
1943 | ||
1944 | static struct config_item_type o2hb_heartbeat_group_type = { | |
1945 | .ct_group_ops = &o2hb_heartbeat_group_group_ops, | |
1946 | .ct_item_ops = &o2hb_hearbeat_group_item_ops, | |
1947 | .ct_attrs = o2hb_heartbeat_group_attrs, | |
1948 | .ct_owner = THIS_MODULE, | |
1949 | }; | |
1950 | ||
1951 | /* this is just here to avoid touching group in heartbeat.h which the | |
1952 | * entire damn world #includes */ | |
1953 | struct config_group *o2hb_alloc_hb_set(void) | |
1954 | { | |
1955 | struct o2hb_heartbeat_group *hs = NULL; | |
1956 | struct config_group *ret = NULL; | |
1957 | ||
cd861280 | 1958 | hs = kzalloc(sizeof(struct o2hb_heartbeat_group), GFP_KERNEL); |
a7f6a5fb MF |
1959 | if (hs == NULL) |
1960 | goto out; | |
1961 | ||
1962 | config_group_init_type_name(&hs->hs_group, "heartbeat", | |
1963 | &o2hb_heartbeat_group_type); | |
1964 | ||
1965 | ret = &hs->hs_group; | |
1966 | out: | |
1967 | if (ret == NULL) | |
1968 | kfree(hs); | |
1969 | return ret; | |
1970 | } | |
1971 | ||
1972 | void o2hb_free_hb_set(struct config_group *group) | |
1973 | { | |
1974 | struct o2hb_heartbeat_group *hs = to_o2hb_heartbeat_group(group); | |
1975 | kfree(hs); | |
1976 | } | |
1977 | ||
1978 | /* hb callback registration and issueing */ | |
1979 | ||
1980 | static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type) | |
1981 | { | |
1982 | if (type == O2HB_NUM_CB) | |
1983 | return ERR_PTR(-EINVAL); | |
1984 | ||
1985 | return &o2hb_callbacks[type]; | |
1986 | } | |
1987 | ||
1988 | void o2hb_setup_callback(struct o2hb_callback_func *hc, | |
1989 | enum o2hb_callback_type type, | |
1990 | o2hb_cb_func *func, | |
1991 | void *data, | |
1992 | int priority) | |
1993 | { | |
1994 | INIT_LIST_HEAD(&hc->hc_item); | |
1995 | hc->hc_func = func; | |
1996 | hc->hc_data = data; | |
1997 | hc->hc_priority = priority; | |
1998 | hc->hc_type = type; | |
1999 | hc->hc_magic = O2HB_CB_MAGIC; | |
2000 | } | |
2001 | EXPORT_SYMBOL_GPL(o2hb_setup_callback); | |
2002 | ||
14829422 JB |
2003 | static struct o2hb_region *o2hb_find_region(const char *region_uuid) |
2004 | { | |
2005 | struct o2hb_region *p, *reg = NULL; | |
2006 | ||
2007 | assert_spin_locked(&o2hb_live_lock); | |
2008 | ||
2009 | list_for_each_entry(p, &o2hb_all_regions, hr_all_item) { | |
2010 | if (!strcmp(region_uuid, config_item_name(&p->hr_item))) { | |
2011 | reg = p; | |
2012 | break; | |
2013 | } | |
2014 | } | |
2015 | ||
2016 | return reg; | |
2017 | } | |
2018 | ||
2019 | static int o2hb_region_get(const char *region_uuid) | |
2020 | { | |
2021 | int ret = 0; | |
2022 | struct o2hb_region *reg; | |
2023 | ||
2024 | spin_lock(&o2hb_live_lock); | |
2025 | ||
2026 | reg = o2hb_find_region(region_uuid); | |
2027 | if (!reg) | |
2028 | ret = -ENOENT; | |
2029 | spin_unlock(&o2hb_live_lock); | |
2030 | ||
16c6a4f2 JB |
2031 | if (ret) |
2032 | goto out; | |
2033 | ||
2034 | ret = o2nm_depend_this_node(); | |
2035 | if (ret) | |
2036 | goto out; | |
14829422 | 2037 | |
16c6a4f2 JB |
2038 | ret = o2nm_depend_item(®->hr_item); |
2039 | if (ret) | |
2040 | o2nm_undepend_this_node(); | |
2041 | ||
2042 | out: | |
14829422 JB |
2043 | return ret; |
2044 | } | |
2045 | ||
2046 | static void o2hb_region_put(const char *region_uuid) | |
2047 | { | |
2048 | struct o2hb_region *reg; | |
2049 | ||
2050 | spin_lock(&o2hb_live_lock); | |
2051 | ||
2052 | reg = o2hb_find_region(region_uuid); | |
2053 | ||
2054 | spin_unlock(&o2hb_live_lock); | |
2055 | ||
16c6a4f2 | 2056 | if (reg) { |
14829422 | 2057 | o2nm_undepend_item(®->hr_item); |
16c6a4f2 JB |
2058 | o2nm_undepend_this_node(); |
2059 | } | |
14829422 JB |
2060 | } |
2061 | ||
2062 | int o2hb_register_callback(const char *region_uuid, | |
2063 | struct o2hb_callback_func *hc) | |
a7f6a5fb MF |
2064 | { |
2065 | struct o2hb_callback_func *tmp; | |
2066 | struct list_head *iter; | |
2067 | struct o2hb_callback *hbcall; | |
2068 | int ret; | |
2069 | ||
2070 | BUG_ON(hc->hc_magic != O2HB_CB_MAGIC); | |
2071 | BUG_ON(!list_empty(&hc->hc_item)); | |
2072 | ||
2073 | hbcall = hbcall_from_type(hc->hc_type); | |
2074 | if (IS_ERR(hbcall)) { | |
2075 | ret = PTR_ERR(hbcall); | |
2076 | goto out; | |
2077 | } | |
2078 | ||
14829422 JB |
2079 | if (region_uuid) { |
2080 | ret = o2hb_region_get(region_uuid); | |
2081 | if (ret) | |
2082 | goto out; | |
2083 | } | |
2084 | ||
a7f6a5fb MF |
2085 | down_write(&o2hb_callback_sem); |
2086 | ||
2087 | list_for_each(iter, &hbcall->list) { | |
2088 | tmp = list_entry(iter, struct o2hb_callback_func, hc_item); | |
2089 | if (hc->hc_priority < tmp->hc_priority) { | |
2090 | list_add_tail(&hc->hc_item, iter); | |
2091 | break; | |
2092 | } | |
2093 | } | |
2094 | if (list_empty(&hc->hc_item)) | |
2095 | list_add_tail(&hc->hc_item, &hbcall->list); | |
2096 | ||
2097 | up_write(&o2hb_callback_sem); | |
2098 | ret = 0; | |
2099 | out: | |
2100 | mlog(ML_HEARTBEAT, "returning %d on behalf of %p for funcs %p\n", | |
2101 | ret, __builtin_return_address(0), hc); | |
2102 | return ret; | |
2103 | } | |
2104 | EXPORT_SYMBOL_GPL(o2hb_register_callback); | |
2105 | ||
14829422 JB |
2106 | void o2hb_unregister_callback(const char *region_uuid, |
2107 | struct o2hb_callback_func *hc) | |
a7f6a5fb MF |
2108 | { |
2109 | BUG_ON(hc->hc_magic != O2HB_CB_MAGIC); | |
2110 | ||
2111 | mlog(ML_HEARTBEAT, "on behalf of %p for funcs %p\n", | |
2112 | __builtin_return_address(0), hc); | |
2113 | ||
14829422 | 2114 | /* XXX Can this happen _with_ a region reference? */ |
a7f6a5fb | 2115 | if (list_empty(&hc->hc_item)) |
c24f72cc | 2116 | return; |
a7f6a5fb | 2117 | |
14829422 JB |
2118 | if (region_uuid) |
2119 | o2hb_region_put(region_uuid); | |
2120 | ||
a7f6a5fb MF |
2121 | down_write(&o2hb_callback_sem); |
2122 | ||
2123 | list_del_init(&hc->hc_item); | |
2124 | ||
2125 | up_write(&o2hb_callback_sem); | |
a7f6a5fb MF |
2126 | } |
2127 | EXPORT_SYMBOL_GPL(o2hb_unregister_callback); | |
2128 | ||
2129 | int o2hb_check_node_heartbeating(u8 node_num) | |
2130 | { | |
2131 | unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; | |
2132 | ||
2133 | o2hb_fill_node_map(testing_map, sizeof(testing_map)); | |
2134 | if (!test_bit(node_num, testing_map)) { | |
2135 | mlog(ML_HEARTBEAT, | |
2136 | "node (%u) does not have heartbeating enabled.\n", | |
2137 | node_num); | |
2138 | return 0; | |
2139 | } | |
2140 | ||
2141 | return 1; | |
2142 | } | |
2143 | EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating); | |
2144 | ||
2145 | int o2hb_check_node_heartbeating_from_callback(u8 node_num) | |
2146 | { | |
2147 | unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; | |
2148 | ||
2149 | o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map)); | |
2150 | if (!test_bit(node_num, testing_map)) { | |
2151 | mlog(ML_HEARTBEAT, | |
2152 | "node (%u) does not have heartbeating enabled.\n", | |
2153 | node_num); | |
2154 | return 0; | |
2155 | } | |
2156 | ||
2157 | return 1; | |
2158 | } | |
2159 | EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback); | |
2160 | ||
2161 | /* Makes sure our local node is configured with a node number, and is | |
2162 | * heartbeating. */ | |
2163 | int o2hb_check_local_node_heartbeating(void) | |
2164 | { | |
2165 | u8 node_num; | |
2166 | ||
2167 | /* if this node was set then we have networking */ | |
2168 | node_num = o2nm_this_node(); | |
2169 | if (node_num == O2NM_MAX_NODES) { | |
2170 | mlog(ML_HEARTBEAT, "this node has not been configured.\n"); | |
2171 | return 0; | |
2172 | } | |
2173 | ||
2174 | return o2hb_check_node_heartbeating(node_num); | |
2175 | } | |
2176 | EXPORT_SYMBOL_GPL(o2hb_check_local_node_heartbeating); | |
2177 | ||
2178 | /* | |
2179 | * this is just a hack until we get the plumbing which flips file systems | |
2180 | * read only and drops the hb ref instead of killing the node dead. | |
2181 | */ | |
2182 | void o2hb_stop_all_regions(void) | |
2183 | { | |
2184 | struct o2hb_region *reg; | |
2185 | ||
2186 | mlog(ML_ERROR, "stopping heartbeat on all active regions.\n"); | |
2187 | ||
2188 | spin_lock(&o2hb_live_lock); | |
2189 | ||
2190 | list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) | |
2191 | reg->hr_unclean_stop = 1; | |
2192 | ||
2193 | spin_unlock(&o2hb_live_lock); | |
2194 | } | |
2195 | EXPORT_SYMBOL_GPL(o2hb_stop_all_regions); | |
b3c85c4c SM |
2196 | |
2197 | int o2hb_get_all_regions(char *region_uuids, u8 max_regions) | |
2198 | { | |
2199 | struct o2hb_region *reg; | |
2200 | int numregs = 0; | |
2201 | char *p; | |
2202 | ||
2203 | spin_lock(&o2hb_live_lock); | |
2204 | ||
2205 | p = region_uuids; | |
2206 | list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) { | |
2207 | mlog(0, "Region: %s\n", config_item_name(®->hr_item)); | |
2208 | if (numregs < max_regions) { | |
2209 | memcpy(p, config_item_name(®->hr_item), | |
2210 | O2HB_MAX_REGION_NAME_LEN); | |
2211 | p += O2HB_MAX_REGION_NAME_LEN; | |
2212 | } | |
2213 | numregs++; | |
2214 | } | |
2215 | ||
2216 | spin_unlock(&o2hb_live_lock); | |
2217 | ||
2218 | return numregs; | |
2219 | } | |
2220 | EXPORT_SYMBOL_GPL(o2hb_get_all_regions); | |
2221 | ||
2222 | int o2hb_global_heartbeat_active(void) | |
2223 | { | |
2224 | return 0; | |
2225 | } | |
2226 | EXPORT_SYMBOL(o2hb_global_heartbeat_active); |