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