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> | |
36 | ||
37 | #include "heartbeat.h" | |
38 | #include "tcp.h" | |
39 | #include "nodemanager.h" | |
40 | #include "quorum.h" | |
41 | ||
42 | #include "masklog.h" | |
43 | ||
44 | ||
45 | /* | |
46 | * The first heartbeat pass had one global thread that would serialize all hb | |
47 | * callback calls. This global serializing sem should only be removed once | |
48 | * we've made sure that all callees can deal with being called concurrently | |
49 | * from multiple hb region threads. | |
50 | */ | |
51 | static DECLARE_RWSEM(o2hb_callback_sem); | |
52 | ||
53 | /* | |
54 | * multiple hb threads are watching multiple regions. A node is live | |
55 | * whenever any of the threads sees activity from the node in its region. | |
56 | */ | |
34af946a | 57 | static DEFINE_SPINLOCK(o2hb_live_lock); |
a7f6a5fb MF |
58 | static struct list_head o2hb_live_slots[O2NM_MAX_NODES]; |
59 | static unsigned long o2hb_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)]; | |
60 | static LIST_HEAD(o2hb_node_events); | |
61 | static DECLARE_WAIT_QUEUE_HEAD(o2hb_steady_queue); | |
62 | ||
63 | static LIST_HEAD(o2hb_all_regions); | |
64 | ||
65 | static struct o2hb_callback { | |
66 | struct list_head list; | |
67 | } o2hb_callbacks[O2HB_NUM_CB]; | |
68 | ||
69 | static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type); | |
70 | ||
71 | #define O2HB_DEFAULT_BLOCK_BITS 9 | |
72 | ||
73 | unsigned int o2hb_dead_threshold = O2HB_DEFAULT_DEAD_THRESHOLD; | |
74 | ||
75 | /* Only sets a new threshold if there are no active regions. | |
76 | * | |
77 | * No locking or otherwise interesting code is required for reading | |
78 | * o2hb_dead_threshold as it can't change once regions are active and | |
79 | * it's not interesting to anyone until then anyway. */ | |
80 | static void o2hb_dead_threshold_set(unsigned int threshold) | |
81 | { | |
82 | if (threshold > O2HB_MIN_DEAD_THRESHOLD) { | |
83 | spin_lock(&o2hb_live_lock); | |
84 | if (list_empty(&o2hb_all_regions)) | |
85 | o2hb_dead_threshold = threshold; | |
86 | spin_unlock(&o2hb_live_lock); | |
87 | } | |
88 | } | |
89 | ||
90 | struct o2hb_node_event { | |
91 | struct list_head hn_item; | |
92 | enum o2hb_callback_type hn_event_type; | |
93 | struct o2nm_node *hn_node; | |
94 | int hn_node_num; | |
95 | }; | |
96 | ||
97 | struct o2hb_disk_slot { | |
98 | struct o2hb_disk_heartbeat_block *ds_raw_block; | |
99 | u8 ds_node_num; | |
100 | u64 ds_last_time; | |
101 | u64 ds_last_generation; | |
102 | u16 ds_equal_samples; | |
103 | u16 ds_changed_samples; | |
104 | struct list_head ds_live_item; | |
105 | }; | |
106 | ||
107 | /* each thread owns a region.. when we're asked to tear down the region | |
108 | * we ask the thread to stop, who cleans up the region */ | |
109 | struct o2hb_region { | |
110 | struct config_item hr_item; | |
111 | ||
112 | struct list_head hr_all_item; | |
113 | unsigned hr_unclean_stop:1; | |
114 | ||
115 | /* protected by the hr_callback_sem */ | |
116 | struct task_struct *hr_task; | |
117 | ||
118 | unsigned int hr_blocks; | |
119 | unsigned long long hr_start_block; | |
120 | ||
121 | unsigned int hr_block_bits; | |
122 | unsigned int hr_block_bytes; | |
123 | ||
124 | unsigned int hr_slots_per_page; | |
125 | unsigned int hr_num_pages; | |
126 | ||
127 | struct page **hr_slot_data; | |
128 | struct block_device *hr_bdev; | |
129 | struct o2hb_disk_slot *hr_slots; | |
130 | ||
131 | /* let the person setting up hb wait for it to return until it | |
132 | * has reached a 'steady' state. This will be fixed when we have | |
133 | * a more complete api that doesn't lead to this sort of fragility. */ | |
134 | atomic_t hr_steady_iterations; | |
135 | ||
136 | char hr_dev_name[BDEVNAME_SIZE]; | |
137 | ||
138 | unsigned int hr_timeout_ms; | |
139 | ||
140 | /* randomized as the region goes up and down so that a node | |
141 | * recognizes a node going up and down in one iteration */ | |
142 | u64 hr_generation; | |
143 | ||
c4028958 | 144 | struct delayed_work hr_write_timeout_work; |
a7f6a5fb MF |
145 | unsigned long hr_last_timeout_start; |
146 | ||
147 | /* Used during o2hb_check_slot to hold a copy of the block | |
148 | * being checked because we temporarily have to zero out the | |
149 | * crc field. */ | |
150 | struct o2hb_disk_heartbeat_block *hr_tmp_block; | |
151 | }; | |
152 | ||
153 | struct o2hb_bio_wait_ctxt { | |
154 | atomic_t wc_num_reqs; | |
155 | struct completion wc_io_complete; | |
a9e2ae39 | 156 | int wc_error; |
a7f6a5fb MF |
157 | }; |
158 | ||
c4028958 | 159 | static void o2hb_write_timeout(struct work_struct *work) |
a7f6a5fb | 160 | { |
c4028958 DH |
161 | struct o2hb_region *reg = |
162 | container_of(work, struct o2hb_region, | |
163 | hr_write_timeout_work.work); | |
a7f6a5fb MF |
164 | |
165 | mlog(ML_ERROR, "Heartbeat write timeout to device %s after %u " | |
166 | "milliseconds\n", reg->hr_dev_name, | |
167 | jiffies_to_msecs(jiffies - reg->hr_last_timeout_start)); | |
168 | o2quo_disk_timeout(); | |
169 | } | |
170 | ||
171 | static void o2hb_arm_write_timeout(struct o2hb_region *reg) | |
172 | { | |
173 | mlog(0, "Queue write timeout for %u ms\n", O2HB_MAX_WRITE_TIMEOUT_MS); | |
174 | ||
175 | cancel_delayed_work(®->hr_write_timeout_work); | |
176 | reg->hr_last_timeout_start = jiffies; | |
177 | schedule_delayed_work(®->hr_write_timeout_work, | |
178 | msecs_to_jiffies(O2HB_MAX_WRITE_TIMEOUT_MS)); | |
179 | } | |
180 | ||
181 | static void o2hb_disarm_write_timeout(struct o2hb_region *reg) | |
182 | { | |
183 | cancel_delayed_work(®->hr_write_timeout_work); | |
184 | flush_scheduled_work(); | |
185 | } | |
186 | ||
b559292e | 187 | static inline void o2hb_bio_wait_init(struct o2hb_bio_wait_ctxt *wc) |
a7f6a5fb | 188 | { |
b559292e | 189 | atomic_set(&wc->wc_num_reqs, 1); |
a7f6a5fb | 190 | init_completion(&wc->wc_io_complete); |
a9e2ae39 | 191 | wc->wc_error = 0; |
a7f6a5fb MF |
192 | } |
193 | ||
194 | /* Used in error paths too */ | |
195 | static inline void o2hb_bio_wait_dec(struct o2hb_bio_wait_ctxt *wc, | |
196 | unsigned int num) | |
197 | { | |
198 | /* sadly atomic_sub_and_test() isn't available on all platforms. The | |
199 | * good news is that the fast path only completes one at a time */ | |
200 | while(num--) { | |
201 | if (atomic_dec_and_test(&wc->wc_num_reqs)) { | |
202 | BUG_ON(num > 0); | |
203 | complete(&wc->wc_io_complete); | |
204 | } | |
205 | } | |
206 | } | |
207 | ||
208 | static void o2hb_wait_on_io(struct o2hb_region *reg, | |
209 | struct o2hb_bio_wait_ctxt *wc) | |
210 | { | |
211 | struct address_space *mapping = reg->hr_bdev->bd_inode->i_mapping; | |
212 | ||
213 | blk_run_address_space(mapping); | |
b559292e | 214 | o2hb_bio_wait_dec(wc, 1); |
a7f6a5fb MF |
215 | |
216 | wait_for_completion(&wc->wc_io_complete); | |
217 | } | |
218 | ||
219 | static int o2hb_bio_end_io(struct bio *bio, | |
a7f6a5fb MF |
220 | int error) |
221 | { | |
222 | struct o2hb_bio_wait_ctxt *wc = bio->bi_private; | |
223 | ||
a9e2ae39 | 224 | if (error) { |
a7f6a5fb | 225 | mlog(ML_ERROR, "IO Error %d\n", error); |
a9e2ae39 MF |
226 | wc->wc_error = error; |
227 | } | |
a7f6a5fb | 228 | |
a7f6a5fb | 229 | o2hb_bio_wait_dec(wc, 1); |
b559292e | 230 | bio_put(bio); |
a7f6a5fb MF |
231 | return 0; |
232 | } | |
233 | ||
234 | /* Setup a Bio to cover I/O against num_slots slots starting at | |
235 | * start_slot. */ | |
236 | static struct bio *o2hb_setup_one_bio(struct o2hb_region *reg, | |
237 | struct o2hb_bio_wait_ctxt *wc, | |
b559292e PR |
238 | unsigned int *current_slot, |
239 | unsigned int max_slots) | |
a7f6a5fb | 240 | { |
b559292e | 241 | int len, current_page; |
a7f6a5fb MF |
242 | unsigned int vec_len, vec_start; |
243 | unsigned int bits = reg->hr_block_bits; | |
244 | unsigned int spp = reg->hr_slots_per_page; | |
b559292e | 245 | unsigned int cs = *current_slot; |
a7f6a5fb MF |
246 | struct bio *bio; |
247 | struct page *page; | |
248 | ||
a7f6a5fb MF |
249 | /* Testing has shown this allocation to take long enough under |
250 | * GFP_KERNEL that the local node can get fenced. It would be | |
251 | * nicest if we could pre-allocate these bios and avoid this | |
252 | * all together. */ | |
b559292e | 253 | bio = bio_alloc(GFP_ATOMIC, 16); |
a7f6a5fb MF |
254 | if (!bio) { |
255 | mlog(ML_ERROR, "Could not alloc slots BIO!\n"); | |
256 | bio = ERR_PTR(-ENOMEM); | |
257 | goto bail; | |
258 | } | |
259 | ||
260 | /* Must put everything in 512 byte sectors for the bio... */ | |
b559292e | 261 | bio->bi_sector = (reg->hr_start_block + cs) << (bits - 9); |
a7f6a5fb MF |
262 | bio->bi_bdev = reg->hr_bdev; |
263 | bio->bi_private = wc; | |
264 | bio->bi_end_io = o2hb_bio_end_io; | |
265 | ||
b559292e PR |
266 | vec_start = (cs << bits) % PAGE_CACHE_SIZE; |
267 | while(cs < max_slots) { | |
268 | current_page = cs / spp; | |
269 | page = reg->hr_slot_data[current_page]; | |
a7f6a5fb | 270 | |
b559292e PR |
271 | vec_len = min(PAGE_CACHE_SIZE, |
272 | (max_slots-cs) * (PAGE_CACHE_SIZE/spp) ); | |
a7f6a5fb MF |
273 | |
274 | mlog(ML_HB_BIO, "page %d, vec_len = %u, vec_start = %u\n", | |
b559292e | 275 | current_page, vec_len, vec_start); |
a7f6a5fb MF |
276 | |
277 | len = bio_add_page(bio, page, vec_len, vec_start); | |
b559292e | 278 | if (len != vec_len) break; |
a7f6a5fb | 279 | |
b559292e | 280 | cs += vec_len / (PAGE_CACHE_SIZE/spp); |
a7f6a5fb MF |
281 | vec_start = 0; |
282 | } | |
283 | ||
284 | bail: | |
b559292e | 285 | *current_slot = cs; |
a7f6a5fb MF |
286 | return bio; |
287 | } | |
288 | ||
a7f6a5fb MF |
289 | static int o2hb_read_slots(struct o2hb_region *reg, |
290 | unsigned int max_slots) | |
291 | { | |
b559292e PR |
292 | unsigned int current_slot=0; |
293 | int status; | |
a7f6a5fb | 294 | struct o2hb_bio_wait_ctxt wc; |
a7f6a5fb MF |
295 | struct bio *bio; |
296 | ||
b559292e | 297 | o2hb_bio_wait_init(&wc); |
a7f6a5fb | 298 | |
b559292e PR |
299 | while(current_slot < max_slots) { |
300 | bio = o2hb_setup_one_bio(reg, &wc, ¤t_slot, max_slots); | |
a7f6a5fb | 301 | if (IS_ERR(bio)) { |
a7f6a5fb MF |
302 | status = PTR_ERR(bio); |
303 | mlog_errno(status); | |
304 | goto bail_and_wait; | |
305 | } | |
a7f6a5fb | 306 | |
b559292e | 307 | atomic_inc(&wc.wc_num_reqs); |
a7f6a5fb MF |
308 | submit_bio(READ, bio); |
309 | } | |
310 | ||
311 | status = 0; | |
312 | ||
313 | bail_and_wait: | |
314 | o2hb_wait_on_io(reg, &wc); | |
a9e2ae39 MF |
315 | if (wc.wc_error && !status) |
316 | status = wc.wc_error; | |
a7f6a5fb | 317 | |
a7f6a5fb MF |
318 | return status; |
319 | } | |
320 | ||
321 | static int o2hb_issue_node_write(struct o2hb_region *reg, | |
a7f6a5fb MF |
322 | struct o2hb_bio_wait_ctxt *write_wc) |
323 | { | |
324 | int status; | |
325 | unsigned int slot; | |
326 | struct bio *bio; | |
327 | ||
b559292e | 328 | o2hb_bio_wait_init(write_wc); |
a7f6a5fb MF |
329 | |
330 | slot = o2nm_this_node(); | |
331 | ||
b559292e | 332 | bio = o2hb_setup_one_bio(reg, write_wc, &slot, slot+1); |
a7f6a5fb MF |
333 | if (IS_ERR(bio)) { |
334 | status = PTR_ERR(bio); | |
335 | mlog_errno(status); | |
336 | goto bail; | |
337 | } | |
338 | ||
b559292e | 339 | atomic_inc(&write_wc->wc_num_reqs); |
a7f6a5fb MF |
340 | submit_bio(WRITE, bio); |
341 | ||
a7f6a5fb MF |
342 | status = 0; |
343 | bail: | |
344 | return status; | |
345 | } | |
346 | ||
347 | static u32 o2hb_compute_block_crc_le(struct o2hb_region *reg, | |
348 | struct o2hb_disk_heartbeat_block *hb_block) | |
349 | { | |
350 | __le32 old_cksum; | |
351 | u32 ret; | |
352 | ||
353 | /* We want to compute the block crc with a 0 value in the | |
354 | * hb_cksum field. Save it off here and replace after the | |
355 | * crc. */ | |
356 | old_cksum = hb_block->hb_cksum; | |
357 | hb_block->hb_cksum = 0; | |
358 | ||
359 | ret = crc32_le(0, (unsigned char *) hb_block, reg->hr_block_bytes); | |
360 | ||
361 | hb_block->hb_cksum = old_cksum; | |
362 | ||
363 | return ret; | |
364 | } | |
365 | ||
366 | static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block *hb_block) | |
367 | { | |
70bacbdb MF |
368 | mlog(ML_ERROR, "Dump slot information: seq = 0x%llx, node = %u, " |
369 | "cksum = 0x%x, generation 0x%llx\n", | |
370 | (long long)le64_to_cpu(hb_block->hb_seq), | |
371 | hb_block->hb_node, le32_to_cpu(hb_block->hb_cksum), | |
372 | (long long)le64_to_cpu(hb_block->hb_generation)); | |
a7f6a5fb MF |
373 | } |
374 | ||
375 | static int o2hb_verify_crc(struct o2hb_region *reg, | |
376 | struct o2hb_disk_heartbeat_block *hb_block) | |
377 | { | |
378 | u32 read, computed; | |
379 | ||
380 | read = le32_to_cpu(hb_block->hb_cksum); | |
381 | computed = o2hb_compute_block_crc_le(reg, hb_block); | |
382 | ||
383 | return read == computed; | |
384 | } | |
385 | ||
386 | /* We want to make sure that nobody is heartbeating on top of us -- | |
387 | * this will help detect an invalid configuration. */ | |
388 | static int o2hb_check_last_timestamp(struct o2hb_region *reg) | |
389 | { | |
390 | int node_num, ret; | |
391 | struct o2hb_disk_slot *slot; | |
392 | struct o2hb_disk_heartbeat_block *hb_block; | |
393 | ||
394 | node_num = o2nm_this_node(); | |
395 | ||
396 | ret = 1; | |
397 | slot = ®->hr_slots[node_num]; | |
398 | /* Don't check on our 1st timestamp */ | |
399 | if (slot->ds_last_time) { | |
400 | hb_block = slot->ds_raw_block; | |
401 | ||
402 | if (le64_to_cpu(hb_block->hb_seq) != slot->ds_last_time) | |
403 | ret = 0; | |
404 | } | |
405 | ||
406 | return ret; | |
407 | } | |
408 | ||
409 | static inline void o2hb_prepare_block(struct o2hb_region *reg, | |
410 | u64 generation) | |
411 | { | |
412 | int node_num; | |
413 | u64 cputime; | |
414 | struct o2hb_disk_slot *slot; | |
415 | struct o2hb_disk_heartbeat_block *hb_block; | |
416 | ||
417 | node_num = o2nm_this_node(); | |
418 | slot = ®->hr_slots[node_num]; | |
419 | ||
420 | hb_block = (struct o2hb_disk_heartbeat_block *)slot->ds_raw_block; | |
421 | memset(hb_block, 0, reg->hr_block_bytes); | |
422 | /* TODO: time stuff */ | |
423 | cputime = CURRENT_TIME.tv_sec; | |
424 | if (!cputime) | |
425 | cputime = 1; | |
426 | ||
427 | hb_block->hb_seq = cpu_to_le64(cputime); | |
428 | hb_block->hb_node = node_num; | |
429 | hb_block->hb_generation = cpu_to_le64(generation); | |
0db638f4 | 430 | hb_block->hb_dead_ms = cpu_to_le32(o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS); |
a7f6a5fb MF |
431 | |
432 | /* This step must always happen last! */ | |
433 | hb_block->hb_cksum = cpu_to_le32(o2hb_compute_block_crc_le(reg, | |
434 | hb_block)); | |
435 | ||
70bacbdb | 436 | mlog(ML_HB_BIO, "our node generation = 0x%llx, cksum = 0x%x\n", |
5fdf1e67 | 437 | (long long)generation, |
70bacbdb | 438 | le32_to_cpu(hb_block->hb_cksum)); |
a7f6a5fb MF |
439 | } |
440 | ||
441 | static void o2hb_fire_callbacks(struct o2hb_callback *hbcall, | |
442 | struct o2nm_node *node, | |
443 | int idx) | |
444 | { | |
445 | struct list_head *iter; | |
446 | struct o2hb_callback_func *f; | |
447 | ||
448 | list_for_each(iter, &hbcall->list) { | |
449 | f = list_entry(iter, struct o2hb_callback_func, hc_item); | |
450 | mlog(ML_HEARTBEAT, "calling funcs %p\n", f); | |
451 | (f->hc_func)(node, idx, f->hc_data); | |
452 | } | |
453 | } | |
454 | ||
455 | /* Will run the list in order until we process the passed event */ | |
456 | static void o2hb_run_event_list(struct o2hb_node_event *queued_event) | |
457 | { | |
458 | int empty; | |
459 | struct o2hb_callback *hbcall; | |
460 | struct o2hb_node_event *event; | |
461 | ||
462 | spin_lock(&o2hb_live_lock); | |
463 | empty = list_empty(&queued_event->hn_item); | |
464 | spin_unlock(&o2hb_live_lock); | |
465 | if (empty) | |
466 | return; | |
467 | ||
468 | /* Holding callback sem assures we don't alter the callback | |
469 | * lists when doing this, and serializes ourselves with other | |
470 | * processes wanting callbacks. */ | |
471 | down_write(&o2hb_callback_sem); | |
472 | ||
473 | spin_lock(&o2hb_live_lock); | |
474 | while (!list_empty(&o2hb_node_events) | |
475 | && !list_empty(&queued_event->hn_item)) { | |
476 | event = list_entry(o2hb_node_events.next, | |
477 | struct o2hb_node_event, | |
478 | hn_item); | |
479 | list_del_init(&event->hn_item); | |
480 | spin_unlock(&o2hb_live_lock); | |
481 | ||
482 | mlog(ML_HEARTBEAT, "Node %s event for %d\n", | |
483 | event->hn_event_type == O2HB_NODE_UP_CB ? "UP" : "DOWN", | |
484 | event->hn_node_num); | |
485 | ||
486 | hbcall = hbcall_from_type(event->hn_event_type); | |
487 | ||
488 | /* We should *never* have gotten on to the list with a | |
489 | * bad type... This isn't something that we should try | |
490 | * to recover from. */ | |
491 | BUG_ON(IS_ERR(hbcall)); | |
492 | ||
493 | o2hb_fire_callbacks(hbcall, event->hn_node, event->hn_node_num); | |
494 | ||
495 | spin_lock(&o2hb_live_lock); | |
496 | } | |
497 | spin_unlock(&o2hb_live_lock); | |
498 | ||
499 | up_write(&o2hb_callback_sem); | |
500 | } | |
501 | ||
502 | static void o2hb_queue_node_event(struct o2hb_node_event *event, | |
503 | enum o2hb_callback_type type, | |
504 | struct o2nm_node *node, | |
505 | int node_num) | |
506 | { | |
507 | assert_spin_locked(&o2hb_live_lock); | |
508 | ||
509 | event->hn_event_type = type; | |
510 | event->hn_node = node; | |
511 | event->hn_node_num = node_num; | |
512 | ||
513 | mlog(ML_HEARTBEAT, "Queue node %s event for node %d\n", | |
514 | type == O2HB_NODE_UP_CB ? "UP" : "DOWN", node_num); | |
515 | ||
516 | list_add_tail(&event->hn_item, &o2hb_node_events); | |
517 | } | |
518 | ||
519 | static void o2hb_shutdown_slot(struct o2hb_disk_slot *slot) | |
520 | { | |
521 | struct o2hb_node_event event = | |
522 | { .hn_item = LIST_HEAD_INIT(event.hn_item), }; | |
523 | struct o2nm_node *node; | |
524 | ||
525 | node = o2nm_get_node_by_num(slot->ds_node_num); | |
526 | if (!node) | |
527 | return; | |
528 | ||
529 | spin_lock(&o2hb_live_lock); | |
530 | if (!list_empty(&slot->ds_live_item)) { | |
531 | mlog(ML_HEARTBEAT, "Shutdown, node %d leaves region\n", | |
532 | slot->ds_node_num); | |
533 | ||
534 | list_del_init(&slot->ds_live_item); | |
535 | ||
536 | if (list_empty(&o2hb_live_slots[slot->ds_node_num])) { | |
537 | clear_bit(slot->ds_node_num, o2hb_live_node_bitmap); | |
538 | ||
539 | o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node, | |
540 | slot->ds_node_num); | |
541 | } | |
542 | } | |
543 | spin_unlock(&o2hb_live_lock); | |
544 | ||
545 | o2hb_run_event_list(&event); | |
546 | ||
547 | o2nm_node_put(node); | |
548 | } | |
549 | ||
550 | static int o2hb_check_slot(struct o2hb_region *reg, | |
551 | struct o2hb_disk_slot *slot) | |
552 | { | |
553 | int changed = 0, gen_changed = 0; | |
554 | struct o2hb_node_event event = | |
555 | { .hn_item = LIST_HEAD_INIT(event.hn_item), }; | |
556 | struct o2nm_node *node; | |
557 | struct o2hb_disk_heartbeat_block *hb_block = reg->hr_tmp_block; | |
558 | u64 cputime; | |
0db638f4 MF |
559 | unsigned int dead_ms = o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS; |
560 | unsigned int slot_dead_ms; | |
a7f6a5fb MF |
561 | |
562 | memcpy(hb_block, slot->ds_raw_block, reg->hr_block_bytes); | |
563 | ||
564 | /* Is this correct? Do we assume that the node doesn't exist | |
565 | * if we're not configured for him? */ | |
566 | node = o2nm_get_node_by_num(slot->ds_node_num); | |
567 | if (!node) | |
568 | return 0; | |
569 | ||
570 | if (!o2hb_verify_crc(reg, hb_block)) { | |
571 | /* all paths from here will drop o2hb_live_lock for | |
572 | * us. */ | |
573 | spin_lock(&o2hb_live_lock); | |
574 | ||
575 | /* Don't print an error on the console in this case - | |
576 | * a freshly formatted heartbeat area will not have a | |
577 | * crc set on it. */ | |
578 | if (list_empty(&slot->ds_live_item)) | |
579 | goto out; | |
580 | ||
581 | /* The node is live but pushed out a bad crc. We | |
582 | * consider it a transient miss but don't populate any | |
583 | * other values as they may be junk. */ | |
584 | mlog(ML_ERROR, "Node %d has written a bad crc to %s\n", | |
585 | slot->ds_node_num, reg->hr_dev_name); | |
586 | o2hb_dump_slot(hb_block); | |
587 | ||
588 | slot->ds_equal_samples++; | |
589 | goto fire_callbacks; | |
590 | } | |
591 | ||
592 | /* we don't care if these wrap.. the state transitions below | |
593 | * clear at the right places */ | |
594 | cputime = le64_to_cpu(hb_block->hb_seq); | |
595 | if (slot->ds_last_time != cputime) | |
596 | slot->ds_changed_samples++; | |
597 | else | |
598 | slot->ds_equal_samples++; | |
599 | slot->ds_last_time = cputime; | |
600 | ||
601 | /* The node changed heartbeat generations. We assume this to | |
602 | * mean it dropped off but came back before we timed out. We | |
603 | * want to consider it down for the time being but don't want | |
604 | * to lose any changed_samples state we might build up to | |
605 | * considering it live again. */ | |
606 | if (slot->ds_last_generation != le64_to_cpu(hb_block->hb_generation)) { | |
607 | gen_changed = 1; | |
608 | slot->ds_equal_samples = 0; | |
70bacbdb MF |
609 | mlog(ML_HEARTBEAT, "Node %d changed generation (0x%llx " |
610 | "to 0x%llx)\n", slot->ds_node_num, | |
611 | (long long)slot->ds_last_generation, | |
612 | (long long)le64_to_cpu(hb_block->hb_generation)); | |
a7f6a5fb MF |
613 | } |
614 | ||
615 | slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation); | |
616 | ||
70bacbdb MF |
617 | mlog(ML_HEARTBEAT, "Slot %d gen 0x%llx cksum 0x%x " |
618 | "seq %llu last %llu changed %u equal %u\n", | |
619 | slot->ds_node_num, (long long)slot->ds_last_generation, | |
620 | le32_to_cpu(hb_block->hb_cksum), | |
621 | (unsigned long long)le64_to_cpu(hb_block->hb_seq), | |
622 | (unsigned long long)slot->ds_last_time, slot->ds_changed_samples, | |
a7f6a5fb MF |
623 | slot->ds_equal_samples); |
624 | ||
625 | spin_lock(&o2hb_live_lock); | |
626 | ||
627 | fire_callbacks: | |
628 | /* dead nodes only come to life after some number of | |
629 | * changes at any time during their dead time */ | |
630 | if (list_empty(&slot->ds_live_item) && | |
631 | slot->ds_changed_samples >= O2HB_LIVE_THRESHOLD) { | |
70bacbdb MF |
632 | mlog(ML_HEARTBEAT, "Node %d (id 0x%llx) joined my region\n", |
633 | slot->ds_node_num, (long long)slot->ds_last_generation); | |
a7f6a5fb MF |
634 | |
635 | /* first on the list generates a callback */ | |
636 | if (list_empty(&o2hb_live_slots[slot->ds_node_num])) { | |
637 | set_bit(slot->ds_node_num, o2hb_live_node_bitmap); | |
638 | ||
639 | o2hb_queue_node_event(&event, O2HB_NODE_UP_CB, node, | |
640 | slot->ds_node_num); | |
641 | ||
642 | changed = 1; | |
643 | } | |
644 | ||
645 | list_add_tail(&slot->ds_live_item, | |
646 | &o2hb_live_slots[slot->ds_node_num]); | |
647 | ||
648 | slot->ds_equal_samples = 0; | |
0db638f4 MF |
649 | |
650 | /* We want to be sure that all nodes agree on the | |
651 | * number of milliseconds before a node will be | |
652 | * considered dead. The self-fencing timeout is | |
653 | * computed from this value, and a discrepancy might | |
654 | * result in heartbeat calling a node dead when it | |
655 | * hasn't self-fenced yet. */ | |
656 | slot_dead_ms = le32_to_cpu(hb_block->hb_dead_ms); | |
657 | if (slot_dead_ms && slot_dead_ms != dead_ms) { | |
658 | /* TODO: Perhaps we can fail the region here. */ | |
659 | mlog(ML_ERROR, "Node %d on device %s has a dead count " | |
660 | "of %u ms, but our count is %u ms.\n" | |
661 | "Please double check your configuration values " | |
662 | "for 'O2CB_HEARTBEAT_THRESHOLD'\n", | |
663 | slot->ds_node_num, reg->hr_dev_name, slot_dead_ms, | |
664 | dead_ms); | |
665 | } | |
a7f6a5fb MF |
666 | goto out; |
667 | } | |
668 | ||
669 | /* if the list is dead, we're done.. */ | |
670 | if (list_empty(&slot->ds_live_item)) | |
671 | goto out; | |
672 | ||
673 | /* live nodes only go dead after enough consequtive missed | |
674 | * samples.. reset the missed counter whenever we see | |
675 | * activity */ | |
676 | if (slot->ds_equal_samples >= o2hb_dead_threshold || gen_changed) { | |
677 | mlog(ML_HEARTBEAT, "Node %d left my region\n", | |
678 | slot->ds_node_num); | |
679 | ||
680 | /* last off the live_slot generates a callback */ | |
681 | list_del_init(&slot->ds_live_item); | |
682 | if (list_empty(&o2hb_live_slots[slot->ds_node_num])) { | |
683 | clear_bit(slot->ds_node_num, o2hb_live_node_bitmap); | |
684 | ||
685 | o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node, | |
686 | slot->ds_node_num); | |
687 | ||
688 | changed = 1; | |
689 | } | |
690 | ||
691 | /* We don't clear this because the node is still | |
692 | * actually writing new blocks. */ | |
693 | if (!gen_changed) | |
694 | slot->ds_changed_samples = 0; | |
695 | goto out; | |
696 | } | |
697 | if (slot->ds_changed_samples) { | |
698 | slot->ds_changed_samples = 0; | |
699 | slot->ds_equal_samples = 0; | |
700 | } | |
701 | out: | |
702 | spin_unlock(&o2hb_live_lock); | |
703 | ||
704 | o2hb_run_event_list(&event); | |
705 | ||
706 | o2nm_node_put(node); | |
707 | return changed; | |
708 | } | |
709 | ||
710 | /* This could be faster if we just implmented a find_last_bit, but I | |
711 | * don't think the circumstances warrant it. */ | |
712 | static int o2hb_highest_node(unsigned long *nodes, | |
713 | int numbits) | |
714 | { | |
715 | int highest, node; | |
716 | ||
717 | highest = numbits; | |
718 | node = -1; | |
719 | while ((node = find_next_bit(nodes, numbits, node + 1)) != -1) { | |
720 | if (node >= numbits) | |
721 | break; | |
722 | ||
723 | highest = node; | |
724 | } | |
725 | ||
726 | return highest; | |
727 | } | |
728 | ||
a9e2ae39 | 729 | static int o2hb_do_disk_heartbeat(struct o2hb_region *reg) |
a7f6a5fb MF |
730 | { |
731 | int i, ret, highest_node, change = 0; | |
732 | unsigned long configured_nodes[BITS_TO_LONGS(O2NM_MAX_NODES)]; | |
a7f6a5fb MF |
733 | struct o2hb_bio_wait_ctxt write_wc; |
734 | ||
a9e2ae39 MF |
735 | ret = o2nm_configured_node_map(configured_nodes, |
736 | sizeof(configured_nodes)); | |
737 | if (ret) { | |
738 | mlog_errno(ret); | |
739 | return ret; | |
740 | } | |
a7f6a5fb MF |
741 | |
742 | highest_node = o2hb_highest_node(configured_nodes, O2NM_MAX_NODES); | |
743 | if (highest_node >= O2NM_MAX_NODES) { | |
744 | mlog(ML_NOTICE, "ocfs2_heartbeat: no configured nodes found!\n"); | |
a9e2ae39 | 745 | return -EINVAL; |
a7f6a5fb MF |
746 | } |
747 | ||
748 | /* No sense in reading the slots of nodes that don't exist | |
749 | * yet. Of course, if the node definitions have holes in them | |
750 | * then we're reading an empty slot anyway... Consider this | |
751 | * best-effort. */ | |
752 | ret = o2hb_read_slots(reg, highest_node + 1); | |
753 | if (ret < 0) { | |
754 | mlog_errno(ret); | |
a9e2ae39 | 755 | return ret; |
a7f6a5fb MF |
756 | } |
757 | ||
758 | /* With an up to date view of the slots, we can check that no | |
759 | * other node has been improperly configured to heartbeat in | |
760 | * our slot. */ | |
761 | if (!o2hb_check_last_timestamp(reg)) | |
762 | mlog(ML_ERROR, "Device \"%s\": another node is heartbeating " | |
763 | "in our slot!\n", reg->hr_dev_name); | |
764 | ||
765 | /* fill in the proper info for our next heartbeat */ | |
766 | o2hb_prepare_block(reg, reg->hr_generation); | |
767 | ||
768 | /* And fire off the write. Note that we don't wait on this I/O | |
769 | * until later. */ | |
b559292e | 770 | ret = o2hb_issue_node_write(reg, &write_wc); |
a7f6a5fb MF |
771 | if (ret < 0) { |
772 | mlog_errno(ret); | |
a9e2ae39 | 773 | return ret; |
a7f6a5fb MF |
774 | } |
775 | ||
776 | i = -1; | |
777 | while((i = find_next_bit(configured_nodes, O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) { | |
778 | ||
779 | change |= o2hb_check_slot(reg, ®->hr_slots[i]); | |
780 | } | |
781 | ||
782 | /* | |
783 | * We have to be sure we've advertised ourselves on disk | |
784 | * before we can go to steady state. This ensures that | |
785 | * people we find in our steady state have seen us. | |
786 | */ | |
787 | o2hb_wait_on_io(reg, &write_wc); | |
a9e2ae39 MF |
788 | if (write_wc.wc_error) { |
789 | /* Do not re-arm the write timeout on I/O error - we | |
790 | * can't be sure that the new block ever made it to | |
791 | * disk */ | |
792 | mlog(ML_ERROR, "Write error %d on device \"%s\"\n", | |
793 | write_wc.wc_error, reg->hr_dev_name); | |
794 | return write_wc.wc_error; | |
795 | } | |
796 | ||
a7f6a5fb MF |
797 | o2hb_arm_write_timeout(reg); |
798 | ||
799 | /* let the person who launched us know when things are steady */ | |
800 | if (!change && (atomic_read(®->hr_steady_iterations) != 0)) { | |
801 | if (atomic_dec_and_test(®->hr_steady_iterations)) | |
802 | wake_up(&o2hb_steady_queue); | |
803 | } | |
a9e2ae39 MF |
804 | |
805 | return 0; | |
a7f6a5fb MF |
806 | } |
807 | ||
808 | /* Subtract b from a, storing the result in a. a *must* have a larger | |
809 | * value than b. */ | |
810 | static void o2hb_tv_subtract(struct timeval *a, | |
811 | struct timeval *b) | |
812 | { | |
813 | /* just return 0 when a is after b */ | |
814 | if (a->tv_sec < b->tv_sec || | |
815 | (a->tv_sec == b->tv_sec && a->tv_usec < b->tv_usec)) { | |
816 | a->tv_sec = 0; | |
817 | a->tv_usec = 0; | |
818 | return; | |
819 | } | |
820 | ||
821 | a->tv_sec -= b->tv_sec; | |
822 | a->tv_usec -= b->tv_usec; | |
823 | while ( a->tv_usec < 0 ) { | |
824 | a->tv_sec--; | |
825 | a->tv_usec += 1000000; | |
826 | } | |
827 | } | |
828 | ||
829 | static unsigned int o2hb_elapsed_msecs(struct timeval *start, | |
830 | struct timeval *end) | |
831 | { | |
832 | struct timeval res = *end; | |
833 | ||
834 | o2hb_tv_subtract(&res, start); | |
835 | ||
836 | return res.tv_sec * 1000 + res.tv_usec / 1000; | |
837 | } | |
838 | ||
839 | /* | |
840 | * we ride the region ref that the region dir holds. before the region | |
841 | * dir is removed and drops it ref it will wait to tear down this | |
842 | * thread. | |
843 | */ | |
844 | static int o2hb_thread(void *data) | |
845 | { | |
846 | int i, ret; | |
847 | struct o2hb_region *reg = data; | |
a7f6a5fb MF |
848 | struct o2hb_bio_wait_ctxt write_wc; |
849 | struct timeval before_hb, after_hb; | |
850 | unsigned int elapsed_msec; | |
851 | ||
852 | mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread running\n"); | |
853 | ||
854 | set_user_nice(current, -20); | |
855 | ||
856 | while (!kthread_should_stop() && !reg->hr_unclean_stop) { | |
857 | /* We track the time spent inside | |
858 | * o2hb_do_disk_heartbeat so that we avoid more then | |
859 | * hr_timeout_ms between disk writes. On busy systems | |
860 | * this should result in a heartbeat which is less | |
861 | * likely to time itself out. */ | |
862 | do_gettimeofday(&before_hb); | |
863 | ||
a9e2ae39 MF |
864 | i = 0; |
865 | do { | |
866 | ret = o2hb_do_disk_heartbeat(reg); | |
867 | } while (ret && ++i < 2); | |
a7f6a5fb MF |
868 | |
869 | do_gettimeofday(&after_hb); | |
870 | elapsed_msec = o2hb_elapsed_msecs(&before_hb, &after_hb); | |
871 | ||
872 | mlog(0, "start = %lu.%lu, end = %lu.%lu, msec = %u\n", | |
215c7f9f MF |
873 | before_hb.tv_sec, (unsigned long) before_hb.tv_usec, |
874 | after_hb.tv_sec, (unsigned long) after_hb.tv_usec, | |
875 | elapsed_msec); | |
a7f6a5fb MF |
876 | |
877 | if (elapsed_msec < reg->hr_timeout_ms) { | |
878 | /* the kthread api has blocked signals for us so no | |
879 | * need to record the return value. */ | |
880 | msleep_interruptible(reg->hr_timeout_ms - elapsed_msec); | |
881 | } | |
882 | } | |
883 | ||
884 | o2hb_disarm_write_timeout(reg); | |
885 | ||
886 | /* unclean stop is only used in very bad situation */ | |
887 | for(i = 0; !reg->hr_unclean_stop && i < reg->hr_blocks; i++) | |
888 | o2hb_shutdown_slot(®->hr_slots[i]); | |
889 | ||
890 | /* Explicit down notification - avoid forcing the other nodes | |
891 | * to timeout on this region when we could just as easily | |
892 | * write a clear generation - thus indicating to them that | |
893 | * this node has left this region. | |
894 | * | |
895 | * XXX: Should we skip this on unclean_stop? */ | |
896 | o2hb_prepare_block(reg, 0); | |
b559292e | 897 | ret = o2hb_issue_node_write(reg, &write_wc); |
a7f6a5fb MF |
898 | if (ret == 0) { |
899 | o2hb_wait_on_io(reg, &write_wc); | |
a7f6a5fb MF |
900 | } else { |
901 | mlog_errno(ret); | |
902 | } | |
903 | ||
904 | mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread exiting\n"); | |
905 | ||
906 | return 0; | |
907 | } | |
908 | ||
909 | void o2hb_init(void) | |
910 | { | |
911 | int i; | |
912 | ||
913 | for (i = 0; i < ARRAY_SIZE(o2hb_callbacks); i++) | |
914 | INIT_LIST_HEAD(&o2hb_callbacks[i].list); | |
915 | ||
916 | for (i = 0; i < ARRAY_SIZE(o2hb_live_slots); i++) | |
917 | INIT_LIST_HEAD(&o2hb_live_slots[i]); | |
918 | ||
919 | INIT_LIST_HEAD(&o2hb_node_events); | |
920 | ||
921 | memset(o2hb_live_node_bitmap, 0, sizeof(o2hb_live_node_bitmap)); | |
922 | } | |
923 | ||
924 | /* if we're already in a callback then we're already serialized by the sem */ | |
925 | static void o2hb_fill_node_map_from_callback(unsigned long *map, | |
926 | unsigned bytes) | |
927 | { | |
928 | BUG_ON(bytes < (BITS_TO_LONGS(O2NM_MAX_NODES) * sizeof(unsigned long))); | |
929 | ||
930 | memcpy(map, &o2hb_live_node_bitmap, bytes); | |
931 | } | |
932 | ||
933 | /* | |
934 | * get a map of all nodes that are heartbeating in any regions | |
935 | */ | |
936 | void o2hb_fill_node_map(unsigned long *map, unsigned bytes) | |
937 | { | |
938 | /* callers want to serialize this map and callbacks so that they | |
939 | * can trust that they don't miss nodes coming to the party */ | |
940 | down_read(&o2hb_callback_sem); | |
941 | spin_lock(&o2hb_live_lock); | |
942 | o2hb_fill_node_map_from_callback(map, bytes); | |
943 | spin_unlock(&o2hb_live_lock); | |
944 | up_read(&o2hb_callback_sem); | |
945 | } | |
946 | EXPORT_SYMBOL_GPL(o2hb_fill_node_map); | |
947 | ||
948 | /* | |
949 | * heartbeat configfs bits. The heartbeat set is a default set under | |
950 | * the cluster set in nodemanager.c. | |
951 | */ | |
952 | ||
953 | static struct o2hb_region *to_o2hb_region(struct config_item *item) | |
954 | { | |
955 | return item ? container_of(item, struct o2hb_region, hr_item) : NULL; | |
956 | } | |
957 | ||
958 | /* drop_item only drops its ref after killing the thread, nothing should | |
959 | * be using the region anymore. this has to clean up any state that | |
960 | * attributes might have built up. */ | |
961 | static void o2hb_region_release(struct config_item *item) | |
962 | { | |
963 | int i; | |
964 | struct page *page; | |
965 | struct o2hb_region *reg = to_o2hb_region(item); | |
966 | ||
967 | if (reg->hr_tmp_block) | |
968 | kfree(reg->hr_tmp_block); | |
969 | ||
970 | if (reg->hr_slot_data) { | |
971 | for (i = 0; i < reg->hr_num_pages; i++) { | |
972 | page = reg->hr_slot_data[i]; | |
973 | if (page) | |
974 | __free_page(page); | |
975 | } | |
976 | kfree(reg->hr_slot_data); | |
977 | } | |
978 | ||
979 | if (reg->hr_bdev) | |
980 | blkdev_put(reg->hr_bdev); | |
981 | ||
982 | if (reg->hr_slots) | |
983 | kfree(reg->hr_slots); | |
984 | ||
985 | spin_lock(&o2hb_live_lock); | |
986 | list_del(®->hr_all_item); | |
987 | spin_unlock(&o2hb_live_lock); | |
988 | ||
989 | kfree(reg); | |
990 | } | |
991 | ||
992 | static int o2hb_read_block_input(struct o2hb_region *reg, | |
993 | const char *page, | |
994 | size_t count, | |
995 | unsigned long *ret_bytes, | |
996 | unsigned int *ret_bits) | |
997 | { | |
998 | unsigned long bytes; | |
999 | char *p = (char *)page; | |
1000 | ||
1001 | bytes = simple_strtoul(p, &p, 0); | |
1002 | if (!p || (*p && (*p != '\n'))) | |
1003 | return -EINVAL; | |
1004 | ||
1005 | /* Heartbeat and fs min / max block sizes are the same. */ | |
1006 | if (bytes > 4096 || bytes < 512) | |
1007 | return -ERANGE; | |
1008 | if (hweight16(bytes) != 1) | |
1009 | return -EINVAL; | |
1010 | ||
1011 | if (ret_bytes) | |
1012 | *ret_bytes = bytes; | |
1013 | if (ret_bits) | |
1014 | *ret_bits = ffs(bytes) - 1; | |
1015 | ||
1016 | return 0; | |
1017 | } | |
1018 | ||
1019 | static ssize_t o2hb_region_block_bytes_read(struct o2hb_region *reg, | |
1020 | char *page) | |
1021 | { | |
1022 | return sprintf(page, "%u\n", reg->hr_block_bytes); | |
1023 | } | |
1024 | ||
1025 | static ssize_t o2hb_region_block_bytes_write(struct o2hb_region *reg, | |
1026 | const char *page, | |
1027 | size_t count) | |
1028 | { | |
1029 | int status; | |
1030 | unsigned long block_bytes; | |
1031 | unsigned int block_bits; | |
1032 | ||
1033 | if (reg->hr_bdev) | |
1034 | return -EINVAL; | |
1035 | ||
1036 | status = o2hb_read_block_input(reg, page, count, | |
1037 | &block_bytes, &block_bits); | |
1038 | if (status) | |
1039 | return status; | |
1040 | ||
1041 | reg->hr_block_bytes = (unsigned int)block_bytes; | |
1042 | reg->hr_block_bits = block_bits; | |
1043 | ||
1044 | return count; | |
1045 | } | |
1046 | ||
1047 | static ssize_t o2hb_region_start_block_read(struct o2hb_region *reg, | |
1048 | char *page) | |
1049 | { | |
1050 | return sprintf(page, "%llu\n", reg->hr_start_block); | |
1051 | } | |
1052 | ||
1053 | static ssize_t o2hb_region_start_block_write(struct o2hb_region *reg, | |
1054 | const char *page, | |
1055 | size_t count) | |
1056 | { | |
1057 | unsigned long long tmp; | |
1058 | char *p = (char *)page; | |
1059 | ||
1060 | if (reg->hr_bdev) | |
1061 | return -EINVAL; | |
1062 | ||
1063 | tmp = simple_strtoull(p, &p, 0); | |
1064 | if (!p || (*p && (*p != '\n'))) | |
1065 | return -EINVAL; | |
1066 | ||
1067 | reg->hr_start_block = tmp; | |
1068 | ||
1069 | return count; | |
1070 | } | |
1071 | ||
1072 | static ssize_t o2hb_region_blocks_read(struct o2hb_region *reg, | |
1073 | char *page) | |
1074 | { | |
1075 | return sprintf(page, "%d\n", reg->hr_blocks); | |
1076 | } | |
1077 | ||
1078 | static ssize_t o2hb_region_blocks_write(struct o2hb_region *reg, | |
1079 | const char *page, | |
1080 | size_t count) | |
1081 | { | |
1082 | unsigned long tmp; | |
1083 | char *p = (char *)page; | |
1084 | ||
1085 | if (reg->hr_bdev) | |
1086 | return -EINVAL; | |
1087 | ||
1088 | tmp = simple_strtoul(p, &p, 0); | |
1089 | if (!p || (*p && (*p != '\n'))) | |
1090 | return -EINVAL; | |
1091 | ||
1092 | if (tmp > O2NM_MAX_NODES || tmp == 0) | |
1093 | return -ERANGE; | |
1094 | ||
1095 | reg->hr_blocks = (unsigned int)tmp; | |
1096 | ||
1097 | return count; | |
1098 | } | |
1099 | ||
1100 | static ssize_t o2hb_region_dev_read(struct o2hb_region *reg, | |
1101 | char *page) | |
1102 | { | |
1103 | unsigned int ret = 0; | |
1104 | ||
1105 | if (reg->hr_bdev) | |
1106 | ret = sprintf(page, "%s\n", reg->hr_dev_name); | |
1107 | ||
1108 | return ret; | |
1109 | } | |
1110 | ||
1111 | static void o2hb_init_region_params(struct o2hb_region *reg) | |
1112 | { | |
1113 | reg->hr_slots_per_page = PAGE_CACHE_SIZE >> reg->hr_block_bits; | |
1114 | reg->hr_timeout_ms = O2HB_REGION_TIMEOUT_MS; | |
1115 | ||
1116 | mlog(ML_HEARTBEAT, "hr_start_block = %llu, hr_blocks = %u\n", | |
1117 | reg->hr_start_block, reg->hr_blocks); | |
1118 | mlog(ML_HEARTBEAT, "hr_block_bytes = %u, hr_block_bits = %u\n", | |
1119 | reg->hr_block_bytes, reg->hr_block_bits); | |
1120 | mlog(ML_HEARTBEAT, "hr_timeout_ms = %u\n", reg->hr_timeout_ms); | |
1121 | mlog(ML_HEARTBEAT, "dead threshold = %u\n", o2hb_dead_threshold); | |
1122 | } | |
1123 | ||
1124 | static int o2hb_map_slot_data(struct o2hb_region *reg) | |
1125 | { | |
1126 | int i, j; | |
1127 | unsigned int last_slot; | |
1128 | unsigned int spp = reg->hr_slots_per_page; | |
1129 | struct page *page; | |
1130 | char *raw; | |
1131 | struct o2hb_disk_slot *slot; | |
1132 | ||
1133 | reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL); | |
1134 | if (reg->hr_tmp_block == NULL) { | |
1135 | mlog_errno(-ENOMEM); | |
1136 | return -ENOMEM; | |
1137 | } | |
1138 | ||
1139 | reg->hr_slots = kcalloc(reg->hr_blocks, | |
1140 | sizeof(struct o2hb_disk_slot), GFP_KERNEL); | |
1141 | if (reg->hr_slots == NULL) { | |
1142 | mlog_errno(-ENOMEM); | |
1143 | return -ENOMEM; | |
1144 | } | |
1145 | ||
1146 | for(i = 0; i < reg->hr_blocks; i++) { | |
1147 | slot = ®->hr_slots[i]; | |
1148 | slot->ds_node_num = i; | |
1149 | INIT_LIST_HEAD(&slot->ds_live_item); | |
1150 | slot->ds_raw_block = NULL; | |
1151 | } | |
1152 | ||
1153 | reg->hr_num_pages = (reg->hr_blocks + spp - 1) / spp; | |
1154 | mlog(ML_HEARTBEAT, "Going to require %u pages to cover %u blocks " | |
1155 | "at %u blocks per page\n", | |
1156 | reg->hr_num_pages, reg->hr_blocks, spp); | |
1157 | ||
1158 | reg->hr_slot_data = kcalloc(reg->hr_num_pages, sizeof(struct page *), | |
1159 | GFP_KERNEL); | |
1160 | if (!reg->hr_slot_data) { | |
1161 | mlog_errno(-ENOMEM); | |
1162 | return -ENOMEM; | |
1163 | } | |
1164 | ||
1165 | for(i = 0; i < reg->hr_num_pages; i++) { | |
1166 | page = alloc_page(GFP_KERNEL); | |
1167 | if (!page) { | |
1168 | mlog_errno(-ENOMEM); | |
1169 | return -ENOMEM; | |
1170 | } | |
1171 | ||
1172 | reg->hr_slot_data[i] = page; | |
1173 | ||
1174 | last_slot = i * spp; | |
1175 | raw = page_address(page); | |
1176 | for (j = 0; | |
1177 | (j < spp) && ((j + last_slot) < reg->hr_blocks); | |
1178 | j++) { | |
1179 | BUG_ON((j + last_slot) >= reg->hr_blocks); | |
1180 | ||
1181 | slot = ®->hr_slots[j + last_slot]; | |
1182 | slot->ds_raw_block = | |
1183 | (struct o2hb_disk_heartbeat_block *) raw; | |
1184 | ||
1185 | raw += reg->hr_block_bytes; | |
1186 | } | |
1187 | } | |
1188 | ||
1189 | return 0; | |
1190 | } | |
1191 | ||
1192 | /* Read in all the slots available and populate the tracking | |
1193 | * structures so that we can start with a baseline idea of what's | |
1194 | * there. */ | |
1195 | static int o2hb_populate_slot_data(struct o2hb_region *reg) | |
1196 | { | |
1197 | int ret, i; | |
1198 | struct o2hb_disk_slot *slot; | |
1199 | struct o2hb_disk_heartbeat_block *hb_block; | |
1200 | ||
1201 | mlog_entry_void(); | |
1202 | ||
1203 | ret = o2hb_read_slots(reg, reg->hr_blocks); | |
1204 | if (ret) { | |
1205 | mlog_errno(ret); | |
1206 | goto out; | |
1207 | } | |
1208 | ||
1209 | /* We only want to get an idea of the values initially in each | |
1210 | * slot, so we do no verification - o2hb_check_slot will | |
1211 | * actually determine if each configured slot is valid and | |
1212 | * whether any values have changed. */ | |
1213 | for(i = 0; i < reg->hr_blocks; i++) { | |
1214 | slot = ®->hr_slots[i]; | |
1215 | hb_block = (struct o2hb_disk_heartbeat_block *) slot->ds_raw_block; | |
1216 | ||
1217 | /* Only fill the values that o2hb_check_slot uses to | |
1218 | * determine changing slots */ | |
1219 | slot->ds_last_time = le64_to_cpu(hb_block->hb_seq); | |
1220 | slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation); | |
1221 | } | |
1222 | ||
1223 | out: | |
1224 | mlog_exit(ret); | |
1225 | return ret; | |
1226 | } | |
1227 | ||
1228 | /* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */ | |
1229 | static ssize_t o2hb_region_dev_write(struct o2hb_region *reg, | |
1230 | const char *page, | |
1231 | size_t count) | |
1232 | { | |
e6c352db | 1233 | struct task_struct *hb_task; |
a7f6a5fb MF |
1234 | long fd; |
1235 | int sectsize; | |
1236 | char *p = (char *)page; | |
1237 | struct file *filp = NULL; | |
1238 | struct inode *inode = NULL; | |
1239 | ssize_t ret = -EINVAL; | |
1240 | ||
1241 | if (reg->hr_bdev) | |
1242 | goto out; | |
1243 | ||
1244 | /* We can't heartbeat without having had our node number | |
1245 | * configured yet. */ | |
1246 | if (o2nm_this_node() == O2NM_MAX_NODES) | |
1247 | goto out; | |
1248 | ||
1249 | fd = simple_strtol(p, &p, 0); | |
1250 | if (!p || (*p && (*p != '\n'))) | |
1251 | goto out; | |
1252 | ||
1253 | if (fd < 0 || fd >= INT_MAX) | |
1254 | goto out; | |
1255 | ||
1256 | filp = fget(fd); | |
1257 | if (filp == NULL) | |
1258 | goto out; | |
1259 | ||
1260 | if (reg->hr_blocks == 0 || reg->hr_start_block == 0 || | |
1261 | reg->hr_block_bytes == 0) | |
1262 | goto out; | |
1263 | ||
1264 | inode = igrab(filp->f_mapping->host); | |
1265 | if (inode == NULL) | |
1266 | goto out; | |
1267 | ||
1268 | if (!S_ISBLK(inode->i_mode)) | |
1269 | goto out; | |
1270 | ||
1271 | reg->hr_bdev = I_BDEV(filp->f_mapping->host); | |
1272 | ret = blkdev_get(reg->hr_bdev, FMODE_WRITE | FMODE_READ, 0); | |
1273 | if (ret) { | |
1274 | reg->hr_bdev = NULL; | |
1275 | goto out; | |
1276 | } | |
1277 | inode = NULL; | |
1278 | ||
1279 | bdevname(reg->hr_bdev, reg->hr_dev_name); | |
1280 | ||
1281 | sectsize = bdev_hardsect_size(reg->hr_bdev); | |
1282 | if (sectsize != reg->hr_block_bytes) { | |
1283 | mlog(ML_ERROR, | |
1284 | "blocksize %u incorrect for device, expected %d", | |
1285 | reg->hr_block_bytes, sectsize); | |
1286 | ret = -EINVAL; | |
1287 | goto out; | |
1288 | } | |
1289 | ||
1290 | o2hb_init_region_params(reg); | |
1291 | ||
1292 | /* Generation of zero is invalid */ | |
1293 | do { | |
1294 | get_random_bytes(®->hr_generation, | |
1295 | sizeof(reg->hr_generation)); | |
1296 | } while (reg->hr_generation == 0); | |
1297 | ||
1298 | ret = o2hb_map_slot_data(reg); | |
1299 | if (ret) { | |
1300 | mlog_errno(ret); | |
1301 | goto out; | |
1302 | } | |
1303 | ||
1304 | ret = o2hb_populate_slot_data(reg); | |
1305 | if (ret) { | |
1306 | mlog_errno(ret); | |
1307 | goto out; | |
1308 | } | |
1309 | ||
c4028958 | 1310 | INIT_DELAYED_WORK(®->hr_write_timeout_work, o2hb_write_timeout); |
a7f6a5fb MF |
1311 | |
1312 | /* | |
1313 | * A node is considered live after it has beat LIVE_THRESHOLD | |
1314 | * times. We're not steady until we've given them a chance | |
1315 | * _after_ our first read. | |
1316 | */ | |
1317 | atomic_set(®->hr_steady_iterations, O2HB_LIVE_THRESHOLD + 1); | |
1318 | ||
e6c352db JB |
1319 | hb_task = kthread_run(o2hb_thread, reg, "o2hb-%s", |
1320 | reg->hr_item.ci_name); | |
1321 | if (IS_ERR(hb_task)) { | |
1322 | ret = PTR_ERR(hb_task); | |
a7f6a5fb | 1323 | mlog_errno(ret); |
a7f6a5fb MF |
1324 | goto out; |
1325 | } | |
1326 | ||
e6c352db JB |
1327 | spin_lock(&o2hb_live_lock); |
1328 | reg->hr_task = hb_task; | |
1329 | spin_unlock(&o2hb_live_lock); | |
1330 | ||
a7f6a5fb MF |
1331 | ret = wait_event_interruptible(o2hb_steady_queue, |
1332 | atomic_read(®->hr_steady_iterations) == 0); | |
1333 | if (ret) { | |
e6df3a66 | 1334 | /* We got interrupted (hello ptrace!). Clean up */ |
e6c352db JB |
1335 | spin_lock(&o2hb_live_lock); |
1336 | hb_task = reg->hr_task; | |
a7f6a5fb | 1337 | reg->hr_task = NULL; |
e6c352db JB |
1338 | spin_unlock(&o2hb_live_lock); |
1339 | ||
1340 | if (hb_task) | |
1341 | kthread_stop(hb_task); | |
a7f6a5fb MF |
1342 | goto out; |
1343 | } | |
1344 | ||
e6df3a66 JB |
1345 | /* Ok, we were woken. Make sure it wasn't by drop_item() */ |
1346 | spin_lock(&o2hb_live_lock); | |
1347 | hb_task = reg->hr_task; | |
1348 | spin_unlock(&o2hb_live_lock); | |
1349 | ||
1350 | if (hb_task) | |
1351 | ret = count; | |
1352 | else | |
1353 | ret = -EIO; | |
1354 | ||
a7f6a5fb MF |
1355 | out: |
1356 | if (filp) | |
1357 | fput(filp); | |
1358 | if (inode) | |
1359 | iput(inode); | |
1360 | if (ret < 0) { | |
1361 | if (reg->hr_bdev) { | |
1362 | blkdev_put(reg->hr_bdev); | |
1363 | reg->hr_bdev = NULL; | |
1364 | } | |
1365 | } | |
1366 | return ret; | |
1367 | } | |
1368 | ||
92efc152 ZW |
1369 | static ssize_t o2hb_region_pid_read(struct o2hb_region *reg, |
1370 | char *page) | |
1371 | { | |
e6c352db JB |
1372 | pid_t pid = 0; |
1373 | ||
1374 | spin_lock(&o2hb_live_lock); | |
1375 | if (reg->hr_task) | |
1376 | pid = reg->hr_task->pid; | |
1377 | spin_unlock(&o2hb_live_lock); | |
1378 | ||
1379 | if (!pid) | |
92efc152 ZW |
1380 | return 0; |
1381 | ||
e6c352db | 1382 | return sprintf(page, "%u\n", pid); |
92efc152 ZW |
1383 | } |
1384 | ||
a7f6a5fb MF |
1385 | struct o2hb_region_attribute { |
1386 | struct configfs_attribute attr; | |
1387 | ssize_t (*show)(struct o2hb_region *, char *); | |
1388 | ssize_t (*store)(struct o2hb_region *, const char *, size_t); | |
1389 | }; | |
1390 | ||
1391 | static struct o2hb_region_attribute o2hb_region_attr_block_bytes = { | |
1392 | .attr = { .ca_owner = THIS_MODULE, | |
1393 | .ca_name = "block_bytes", | |
1394 | .ca_mode = S_IRUGO | S_IWUSR }, | |
1395 | .show = o2hb_region_block_bytes_read, | |
1396 | .store = o2hb_region_block_bytes_write, | |
1397 | }; | |
1398 | ||
1399 | static struct o2hb_region_attribute o2hb_region_attr_start_block = { | |
1400 | .attr = { .ca_owner = THIS_MODULE, | |
1401 | .ca_name = "start_block", | |
1402 | .ca_mode = S_IRUGO | S_IWUSR }, | |
1403 | .show = o2hb_region_start_block_read, | |
1404 | .store = o2hb_region_start_block_write, | |
1405 | }; | |
1406 | ||
1407 | static struct o2hb_region_attribute o2hb_region_attr_blocks = { | |
1408 | .attr = { .ca_owner = THIS_MODULE, | |
1409 | .ca_name = "blocks", | |
1410 | .ca_mode = S_IRUGO | S_IWUSR }, | |
1411 | .show = o2hb_region_blocks_read, | |
1412 | .store = o2hb_region_blocks_write, | |
1413 | }; | |
1414 | ||
1415 | static struct o2hb_region_attribute o2hb_region_attr_dev = { | |
1416 | .attr = { .ca_owner = THIS_MODULE, | |
1417 | .ca_name = "dev", | |
1418 | .ca_mode = S_IRUGO | S_IWUSR }, | |
1419 | .show = o2hb_region_dev_read, | |
1420 | .store = o2hb_region_dev_write, | |
1421 | }; | |
1422 | ||
92efc152 ZW |
1423 | static struct o2hb_region_attribute o2hb_region_attr_pid = { |
1424 | .attr = { .ca_owner = THIS_MODULE, | |
1425 | .ca_name = "pid", | |
1426 | .ca_mode = S_IRUGO | S_IRUSR }, | |
1427 | .show = o2hb_region_pid_read, | |
1428 | }; | |
1429 | ||
a7f6a5fb MF |
1430 | static struct configfs_attribute *o2hb_region_attrs[] = { |
1431 | &o2hb_region_attr_block_bytes.attr, | |
1432 | &o2hb_region_attr_start_block.attr, | |
1433 | &o2hb_region_attr_blocks.attr, | |
1434 | &o2hb_region_attr_dev.attr, | |
92efc152 | 1435 | &o2hb_region_attr_pid.attr, |
a7f6a5fb MF |
1436 | NULL, |
1437 | }; | |
1438 | ||
1439 | static ssize_t o2hb_region_show(struct config_item *item, | |
1440 | struct configfs_attribute *attr, | |
1441 | char *page) | |
1442 | { | |
1443 | struct o2hb_region *reg = to_o2hb_region(item); | |
1444 | struct o2hb_region_attribute *o2hb_region_attr = | |
1445 | container_of(attr, struct o2hb_region_attribute, attr); | |
1446 | ssize_t ret = 0; | |
1447 | ||
1448 | if (o2hb_region_attr->show) | |
1449 | ret = o2hb_region_attr->show(reg, page); | |
1450 | return ret; | |
1451 | } | |
1452 | ||
1453 | static ssize_t o2hb_region_store(struct config_item *item, | |
1454 | struct configfs_attribute *attr, | |
1455 | const char *page, size_t count) | |
1456 | { | |
1457 | struct o2hb_region *reg = to_o2hb_region(item); | |
1458 | struct o2hb_region_attribute *o2hb_region_attr = | |
1459 | container_of(attr, struct o2hb_region_attribute, attr); | |
1460 | ssize_t ret = -EINVAL; | |
1461 | ||
1462 | if (o2hb_region_attr->store) | |
1463 | ret = o2hb_region_attr->store(reg, page, count); | |
1464 | return ret; | |
1465 | } | |
1466 | ||
1467 | static struct configfs_item_operations o2hb_region_item_ops = { | |
1468 | .release = o2hb_region_release, | |
1469 | .show_attribute = o2hb_region_show, | |
1470 | .store_attribute = o2hb_region_store, | |
1471 | }; | |
1472 | ||
1473 | static struct config_item_type o2hb_region_type = { | |
1474 | .ct_item_ops = &o2hb_region_item_ops, | |
1475 | .ct_attrs = o2hb_region_attrs, | |
1476 | .ct_owner = THIS_MODULE, | |
1477 | }; | |
1478 | ||
1479 | /* heartbeat set */ | |
1480 | ||
1481 | struct o2hb_heartbeat_group { | |
1482 | struct config_group hs_group; | |
1483 | /* some stuff? */ | |
1484 | }; | |
1485 | ||
1486 | static struct o2hb_heartbeat_group *to_o2hb_heartbeat_group(struct config_group *group) | |
1487 | { | |
1488 | return group ? | |
1489 | container_of(group, struct o2hb_heartbeat_group, hs_group) | |
1490 | : NULL; | |
1491 | } | |
1492 | ||
1493 | static struct config_item *o2hb_heartbeat_group_make_item(struct config_group *group, | |
1494 | const char *name) | |
1495 | { | |
1496 | struct o2hb_region *reg = NULL; | |
1497 | struct config_item *ret = NULL; | |
1498 | ||
cd861280 | 1499 | reg = kzalloc(sizeof(struct o2hb_region), GFP_KERNEL); |
a7f6a5fb MF |
1500 | if (reg == NULL) |
1501 | goto out; /* ENOMEM */ | |
1502 | ||
1503 | config_item_init_type_name(®->hr_item, name, &o2hb_region_type); | |
1504 | ||
1505 | ret = ®->hr_item; | |
1506 | ||
1507 | spin_lock(&o2hb_live_lock); | |
1508 | list_add_tail(®->hr_all_item, &o2hb_all_regions); | |
1509 | spin_unlock(&o2hb_live_lock); | |
1510 | out: | |
1511 | if (ret == NULL) | |
1512 | kfree(reg); | |
1513 | ||
1514 | return ret; | |
1515 | } | |
1516 | ||
1517 | static void o2hb_heartbeat_group_drop_item(struct config_group *group, | |
1518 | struct config_item *item) | |
1519 | { | |
e6c352db | 1520 | struct task_struct *hb_task; |
a7f6a5fb MF |
1521 | struct o2hb_region *reg = to_o2hb_region(item); |
1522 | ||
1523 | /* stop the thread when the user removes the region dir */ | |
e6c352db JB |
1524 | spin_lock(&o2hb_live_lock); |
1525 | hb_task = reg->hr_task; | |
1526 | reg->hr_task = NULL; | |
1527 | spin_unlock(&o2hb_live_lock); | |
1528 | ||
1529 | if (hb_task) | |
1530 | kthread_stop(hb_task); | |
a7f6a5fb | 1531 | |
e6df3a66 JB |
1532 | /* |
1533 | * If we're racing a dev_write(), we need to wake them. They will | |
1534 | * check reg->hr_task | |
1535 | */ | |
1536 | if (atomic_read(®->hr_steady_iterations) != 0) { | |
1537 | atomic_set(®->hr_steady_iterations, 0); | |
1538 | wake_up(&o2hb_steady_queue); | |
1539 | } | |
1540 | ||
a7f6a5fb MF |
1541 | config_item_put(item); |
1542 | } | |
1543 | ||
1544 | struct o2hb_heartbeat_group_attribute { | |
1545 | struct configfs_attribute attr; | |
1546 | ssize_t (*show)(struct o2hb_heartbeat_group *, char *); | |
1547 | ssize_t (*store)(struct o2hb_heartbeat_group *, const char *, size_t); | |
1548 | }; | |
1549 | ||
1550 | static ssize_t o2hb_heartbeat_group_show(struct config_item *item, | |
1551 | struct configfs_attribute *attr, | |
1552 | char *page) | |
1553 | { | |
1554 | struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item)); | |
1555 | struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr = | |
1556 | container_of(attr, struct o2hb_heartbeat_group_attribute, attr); | |
1557 | ssize_t ret = 0; | |
1558 | ||
1559 | if (o2hb_heartbeat_group_attr->show) | |
1560 | ret = o2hb_heartbeat_group_attr->show(reg, page); | |
1561 | return ret; | |
1562 | } | |
1563 | ||
1564 | static ssize_t o2hb_heartbeat_group_store(struct config_item *item, | |
1565 | struct configfs_attribute *attr, | |
1566 | const char *page, size_t count) | |
1567 | { | |
1568 | struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item)); | |
1569 | struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr = | |
1570 | container_of(attr, struct o2hb_heartbeat_group_attribute, attr); | |
1571 | ssize_t ret = -EINVAL; | |
1572 | ||
1573 | if (o2hb_heartbeat_group_attr->store) | |
1574 | ret = o2hb_heartbeat_group_attr->store(reg, page, count); | |
1575 | return ret; | |
1576 | } | |
1577 | ||
1578 | static ssize_t o2hb_heartbeat_group_threshold_show(struct o2hb_heartbeat_group *group, | |
1579 | char *page) | |
1580 | { | |
1581 | return sprintf(page, "%u\n", o2hb_dead_threshold); | |
1582 | } | |
1583 | ||
1584 | static ssize_t o2hb_heartbeat_group_threshold_store(struct o2hb_heartbeat_group *group, | |
1585 | const char *page, | |
1586 | size_t count) | |
1587 | { | |
1588 | unsigned long tmp; | |
1589 | char *p = (char *)page; | |
1590 | ||
1591 | tmp = simple_strtoul(p, &p, 10); | |
1592 | if (!p || (*p && (*p != '\n'))) | |
1593 | return -EINVAL; | |
1594 | ||
1595 | /* this will validate ranges for us. */ | |
1596 | o2hb_dead_threshold_set((unsigned int) tmp); | |
1597 | ||
1598 | return count; | |
1599 | } | |
1600 | ||
1601 | static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_threshold = { | |
1602 | .attr = { .ca_owner = THIS_MODULE, | |
1603 | .ca_name = "dead_threshold", | |
1604 | .ca_mode = S_IRUGO | S_IWUSR }, | |
1605 | .show = o2hb_heartbeat_group_threshold_show, | |
1606 | .store = o2hb_heartbeat_group_threshold_store, | |
1607 | }; | |
1608 | ||
1609 | static struct configfs_attribute *o2hb_heartbeat_group_attrs[] = { | |
1610 | &o2hb_heartbeat_group_attr_threshold.attr, | |
1611 | NULL, | |
1612 | }; | |
1613 | ||
1614 | static struct configfs_item_operations o2hb_hearbeat_group_item_ops = { | |
1615 | .show_attribute = o2hb_heartbeat_group_show, | |
1616 | .store_attribute = o2hb_heartbeat_group_store, | |
1617 | }; | |
1618 | ||
1619 | static struct configfs_group_operations o2hb_heartbeat_group_group_ops = { | |
1620 | .make_item = o2hb_heartbeat_group_make_item, | |
1621 | .drop_item = o2hb_heartbeat_group_drop_item, | |
1622 | }; | |
1623 | ||
1624 | static struct config_item_type o2hb_heartbeat_group_type = { | |
1625 | .ct_group_ops = &o2hb_heartbeat_group_group_ops, | |
1626 | .ct_item_ops = &o2hb_hearbeat_group_item_ops, | |
1627 | .ct_attrs = o2hb_heartbeat_group_attrs, | |
1628 | .ct_owner = THIS_MODULE, | |
1629 | }; | |
1630 | ||
1631 | /* this is just here to avoid touching group in heartbeat.h which the | |
1632 | * entire damn world #includes */ | |
1633 | struct config_group *o2hb_alloc_hb_set(void) | |
1634 | { | |
1635 | struct o2hb_heartbeat_group *hs = NULL; | |
1636 | struct config_group *ret = NULL; | |
1637 | ||
cd861280 | 1638 | hs = kzalloc(sizeof(struct o2hb_heartbeat_group), GFP_KERNEL); |
a7f6a5fb MF |
1639 | if (hs == NULL) |
1640 | goto out; | |
1641 | ||
1642 | config_group_init_type_name(&hs->hs_group, "heartbeat", | |
1643 | &o2hb_heartbeat_group_type); | |
1644 | ||
1645 | ret = &hs->hs_group; | |
1646 | out: | |
1647 | if (ret == NULL) | |
1648 | kfree(hs); | |
1649 | return ret; | |
1650 | } | |
1651 | ||
1652 | void o2hb_free_hb_set(struct config_group *group) | |
1653 | { | |
1654 | struct o2hb_heartbeat_group *hs = to_o2hb_heartbeat_group(group); | |
1655 | kfree(hs); | |
1656 | } | |
1657 | ||
1658 | /* hb callback registration and issueing */ | |
1659 | ||
1660 | static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type) | |
1661 | { | |
1662 | if (type == O2HB_NUM_CB) | |
1663 | return ERR_PTR(-EINVAL); | |
1664 | ||
1665 | return &o2hb_callbacks[type]; | |
1666 | } | |
1667 | ||
1668 | void o2hb_setup_callback(struct o2hb_callback_func *hc, | |
1669 | enum o2hb_callback_type type, | |
1670 | o2hb_cb_func *func, | |
1671 | void *data, | |
1672 | int priority) | |
1673 | { | |
1674 | INIT_LIST_HEAD(&hc->hc_item); | |
1675 | hc->hc_func = func; | |
1676 | hc->hc_data = data; | |
1677 | hc->hc_priority = priority; | |
1678 | hc->hc_type = type; | |
1679 | hc->hc_magic = O2HB_CB_MAGIC; | |
1680 | } | |
1681 | EXPORT_SYMBOL_GPL(o2hb_setup_callback); | |
1682 | ||
14829422 JB |
1683 | static struct o2hb_region *o2hb_find_region(const char *region_uuid) |
1684 | { | |
1685 | struct o2hb_region *p, *reg = NULL; | |
1686 | ||
1687 | assert_spin_locked(&o2hb_live_lock); | |
1688 | ||
1689 | list_for_each_entry(p, &o2hb_all_regions, hr_all_item) { | |
1690 | if (!strcmp(region_uuid, config_item_name(&p->hr_item))) { | |
1691 | reg = p; | |
1692 | break; | |
1693 | } | |
1694 | } | |
1695 | ||
1696 | return reg; | |
1697 | } | |
1698 | ||
1699 | static int o2hb_region_get(const char *region_uuid) | |
1700 | { | |
1701 | int ret = 0; | |
1702 | struct o2hb_region *reg; | |
1703 | ||
1704 | spin_lock(&o2hb_live_lock); | |
1705 | ||
1706 | reg = o2hb_find_region(region_uuid); | |
1707 | if (!reg) | |
1708 | ret = -ENOENT; | |
1709 | spin_unlock(&o2hb_live_lock); | |
1710 | ||
16c6a4f2 JB |
1711 | if (ret) |
1712 | goto out; | |
1713 | ||
1714 | ret = o2nm_depend_this_node(); | |
1715 | if (ret) | |
1716 | goto out; | |
14829422 | 1717 | |
16c6a4f2 JB |
1718 | ret = o2nm_depend_item(®->hr_item); |
1719 | if (ret) | |
1720 | o2nm_undepend_this_node(); | |
1721 | ||
1722 | out: | |
14829422 JB |
1723 | return ret; |
1724 | } | |
1725 | ||
1726 | static void o2hb_region_put(const char *region_uuid) | |
1727 | { | |
1728 | struct o2hb_region *reg; | |
1729 | ||
1730 | spin_lock(&o2hb_live_lock); | |
1731 | ||
1732 | reg = o2hb_find_region(region_uuid); | |
1733 | ||
1734 | spin_unlock(&o2hb_live_lock); | |
1735 | ||
16c6a4f2 | 1736 | if (reg) { |
14829422 | 1737 | o2nm_undepend_item(®->hr_item); |
16c6a4f2 JB |
1738 | o2nm_undepend_this_node(); |
1739 | } | |
14829422 JB |
1740 | } |
1741 | ||
1742 | int o2hb_register_callback(const char *region_uuid, | |
1743 | struct o2hb_callback_func *hc) | |
a7f6a5fb MF |
1744 | { |
1745 | struct o2hb_callback_func *tmp; | |
1746 | struct list_head *iter; | |
1747 | struct o2hb_callback *hbcall; | |
1748 | int ret; | |
1749 | ||
1750 | BUG_ON(hc->hc_magic != O2HB_CB_MAGIC); | |
1751 | BUG_ON(!list_empty(&hc->hc_item)); | |
1752 | ||
1753 | hbcall = hbcall_from_type(hc->hc_type); | |
1754 | if (IS_ERR(hbcall)) { | |
1755 | ret = PTR_ERR(hbcall); | |
1756 | goto out; | |
1757 | } | |
1758 | ||
14829422 JB |
1759 | if (region_uuid) { |
1760 | ret = o2hb_region_get(region_uuid); | |
1761 | if (ret) | |
1762 | goto out; | |
1763 | } | |
1764 | ||
a7f6a5fb MF |
1765 | down_write(&o2hb_callback_sem); |
1766 | ||
1767 | list_for_each(iter, &hbcall->list) { | |
1768 | tmp = list_entry(iter, struct o2hb_callback_func, hc_item); | |
1769 | if (hc->hc_priority < tmp->hc_priority) { | |
1770 | list_add_tail(&hc->hc_item, iter); | |
1771 | break; | |
1772 | } | |
1773 | } | |
1774 | if (list_empty(&hc->hc_item)) | |
1775 | list_add_tail(&hc->hc_item, &hbcall->list); | |
1776 | ||
1777 | up_write(&o2hb_callback_sem); | |
1778 | ret = 0; | |
1779 | out: | |
1780 | mlog(ML_HEARTBEAT, "returning %d on behalf of %p for funcs %p\n", | |
1781 | ret, __builtin_return_address(0), hc); | |
1782 | return ret; | |
1783 | } | |
1784 | EXPORT_SYMBOL_GPL(o2hb_register_callback); | |
1785 | ||
14829422 JB |
1786 | void o2hb_unregister_callback(const char *region_uuid, |
1787 | struct o2hb_callback_func *hc) | |
a7f6a5fb MF |
1788 | { |
1789 | BUG_ON(hc->hc_magic != O2HB_CB_MAGIC); | |
1790 | ||
1791 | mlog(ML_HEARTBEAT, "on behalf of %p for funcs %p\n", | |
1792 | __builtin_return_address(0), hc); | |
1793 | ||
14829422 | 1794 | /* XXX Can this happen _with_ a region reference? */ |
a7f6a5fb | 1795 | if (list_empty(&hc->hc_item)) |
c24f72cc | 1796 | return; |
a7f6a5fb | 1797 | |
14829422 JB |
1798 | if (region_uuid) |
1799 | o2hb_region_put(region_uuid); | |
1800 | ||
a7f6a5fb MF |
1801 | down_write(&o2hb_callback_sem); |
1802 | ||
1803 | list_del_init(&hc->hc_item); | |
1804 | ||
1805 | up_write(&o2hb_callback_sem); | |
a7f6a5fb MF |
1806 | } |
1807 | EXPORT_SYMBOL_GPL(o2hb_unregister_callback); | |
1808 | ||
1809 | int o2hb_check_node_heartbeating(u8 node_num) | |
1810 | { | |
1811 | unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; | |
1812 | ||
1813 | o2hb_fill_node_map(testing_map, sizeof(testing_map)); | |
1814 | if (!test_bit(node_num, testing_map)) { | |
1815 | mlog(ML_HEARTBEAT, | |
1816 | "node (%u) does not have heartbeating enabled.\n", | |
1817 | node_num); | |
1818 | return 0; | |
1819 | } | |
1820 | ||
1821 | return 1; | |
1822 | } | |
1823 | EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating); | |
1824 | ||
1825 | int o2hb_check_node_heartbeating_from_callback(u8 node_num) | |
1826 | { | |
1827 | unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; | |
1828 | ||
1829 | o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map)); | |
1830 | if (!test_bit(node_num, testing_map)) { | |
1831 | mlog(ML_HEARTBEAT, | |
1832 | "node (%u) does not have heartbeating enabled.\n", | |
1833 | node_num); | |
1834 | return 0; | |
1835 | } | |
1836 | ||
1837 | return 1; | |
1838 | } | |
1839 | EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback); | |
1840 | ||
1841 | /* Makes sure our local node is configured with a node number, and is | |
1842 | * heartbeating. */ | |
1843 | int o2hb_check_local_node_heartbeating(void) | |
1844 | { | |
1845 | u8 node_num; | |
1846 | ||
1847 | /* if this node was set then we have networking */ | |
1848 | node_num = o2nm_this_node(); | |
1849 | if (node_num == O2NM_MAX_NODES) { | |
1850 | mlog(ML_HEARTBEAT, "this node has not been configured.\n"); | |
1851 | return 0; | |
1852 | } | |
1853 | ||
1854 | return o2hb_check_node_heartbeating(node_num); | |
1855 | } | |
1856 | EXPORT_SYMBOL_GPL(o2hb_check_local_node_heartbeating); | |
1857 | ||
1858 | /* | |
1859 | * this is just a hack until we get the plumbing which flips file systems | |
1860 | * read only and drops the hb ref instead of killing the node dead. | |
1861 | */ | |
1862 | void o2hb_stop_all_regions(void) | |
1863 | { | |
1864 | struct o2hb_region *reg; | |
1865 | ||
1866 | mlog(ML_ERROR, "stopping heartbeat on all active regions.\n"); | |
1867 | ||
1868 | spin_lock(&o2hb_live_lock); | |
1869 | ||
1870 | list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) | |
1871 | reg->hr_unclean_stop = 1; | |
1872 | ||
1873 | spin_unlock(&o2hb_live_lock); | |
1874 | } | |
1875 | EXPORT_SYMBOL_GPL(o2hb_stop_all_regions); |