]>
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 | ||
87d3d3f3 SM |
65 | #define O2HB_DEBUG_DIR "o2hb" |
66 | #define O2HB_DEBUG_LIVENODES "livenodes" | |
67 | static struct dentry *o2hb_debug_dir; | |
68 | static struct dentry *o2hb_debug_livenodes; | |
69 | ||
a7f6a5fb MF |
70 | static LIST_HEAD(o2hb_all_regions); |
71 | ||
72 | static struct o2hb_callback { | |
73 | struct list_head list; | |
74 | } o2hb_callbacks[O2HB_NUM_CB]; | |
75 | ||
76 | static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type); | |
77 | ||
78 | #define O2HB_DEFAULT_BLOCK_BITS 9 | |
79 | ||
54b5187b SM |
80 | enum o2hb_heartbeat_modes { |
81 | O2HB_HEARTBEAT_LOCAL = 0, | |
82 | O2HB_HEARTBEAT_GLOBAL, | |
83 | O2HB_HEARTBEAT_NUM_MODES, | |
84 | }; | |
85 | ||
86 | char *o2hb_heartbeat_mode_desc[O2HB_HEARTBEAT_NUM_MODES] = { | |
87 | "local", /* O2HB_HEARTBEAT_LOCAL */ | |
88 | "global", /* O2HB_HEARTBEAT_GLOBAL */ | |
89 | }; | |
90 | ||
a7f6a5fb | 91 | unsigned int o2hb_dead_threshold = O2HB_DEFAULT_DEAD_THRESHOLD; |
54b5187b | 92 | unsigned int o2hb_heartbeat_mode = O2HB_HEARTBEAT_LOCAL; |
a7f6a5fb | 93 | |
2bd63216 | 94 | /* Only sets a new threshold if there are no active regions. |
a7f6a5fb MF |
95 | * |
96 | * No locking or otherwise interesting code is required for reading | |
97 | * o2hb_dead_threshold as it can't change once regions are active and | |
98 | * it's not interesting to anyone until then anyway. */ | |
99 | static void o2hb_dead_threshold_set(unsigned int threshold) | |
100 | { | |
101 | if (threshold > O2HB_MIN_DEAD_THRESHOLD) { | |
102 | spin_lock(&o2hb_live_lock); | |
103 | if (list_empty(&o2hb_all_regions)) | |
104 | o2hb_dead_threshold = threshold; | |
105 | spin_unlock(&o2hb_live_lock); | |
106 | } | |
107 | } | |
108 | ||
54b5187b SM |
109 | static int o2hb_global_hearbeat_mode_set(unsigned int hb_mode) |
110 | { | |
111 | int ret = -1; | |
112 | ||
113 | if (hb_mode < O2HB_HEARTBEAT_NUM_MODES) { | |
114 | spin_lock(&o2hb_live_lock); | |
115 | if (list_empty(&o2hb_all_regions)) { | |
116 | o2hb_heartbeat_mode = hb_mode; | |
117 | ret = 0; | |
118 | } | |
119 | spin_unlock(&o2hb_live_lock); | |
120 | } | |
121 | ||
122 | return ret; | |
123 | } | |
124 | ||
a7f6a5fb MF |
125 | struct o2hb_node_event { |
126 | struct list_head hn_item; | |
127 | enum o2hb_callback_type hn_event_type; | |
128 | struct o2nm_node *hn_node; | |
129 | int hn_node_num; | |
130 | }; | |
131 | ||
132 | struct o2hb_disk_slot { | |
133 | struct o2hb_disk_heartbeat_block *ds_raw_block; | |
134 | u8 ds_node_num; | |
135 | u64 ds_last_time; | |
136 | u64 ds_last_generation; | |
137 | u16 ds_equal_samples; | |
138 | u16 ds_changed_samples; | |
139 | struct list_head ds_live_item; | |
140 | }; | |
141 | ||
142 | /* each thread owns a region.. when we're asked to tear down the region | |
143 | * we ask the thread to stop, who cleans up the region */ | |
144 | struct o2hb_region { | |
145 | struct config_item hr_item; | |
146 | ||
147 | struct list_head hr_all_item; | |
148 | unsigned hr_unclean_stop:1; | |
149 | ||
150 | /* protected by the hr_callback_sem */ | |
151 | struct task_struct *hr_task; | |
152 | ||
153 | unsigned int hr_blocks; | |
154 | unsigned long long hr_start_block; | |
155 | ||
156 | unsigned int hr_block_bits; | |
157 | unsigned int hr_block_bytes; | |
158 | ||
159 | unsigned int hr_slots_per_page; | |
160 | unsigned int hr_num_pages; | |
161 | ||
162 | struct page **hr_slot_data; | |
163 | struct block_device *hr_bdev; | |
164 | struct o2hb_disk_slot *hr_slots; | |
165 | ||
166 | /* let the person setting up hb wait for it to return until it | |
167 | * has reached a 'steady' state. This will be fixed when we have | |
168 | * a more complete api that doesn't lead to this sort of fragility. */ | |
169 | atomic_t hr_steady_iterations; | |
170 | ||
171 | char hr_dev_name[BDEVNAME_SIZE]; | |
172 | ||
173 | unsigned int hr_timeout_ms; | |
174 | ||
175 | /* randomized as the region goes up and down so that a node | |
176 | * recognizes a node going up and down in one iteration */ | |
177 | u64 hr_generation; | |
178 | ||
c4028958 | 179 | struct delayed_work hr_write_timeout_work; |
a7f6a5fb MF |
180 | unsigned long hr_last_timeout_start; |
181 | ||
182 | /* Used during o2hb_check_slot to hold a copy of the block | |
183 | * being checked because we temporarily have to zero out the | |
184 | * crc field. */ | |
185 | struct o2hb_disk_heartbeat_block *hr_tmp_block; | |
186 | }; | |
187 | ||
188 | struct o2hb_bio_wait_ctxt { | |
189 | atomic_t wc_num_reqs; | |
190 | struct completion wc_io_complete; | |
a9e2ae39 | 191 | int wc_error; |
a7f6a5fb MF |
192 | }; |
193 | ||
c4028958 | 194 | static void o2hb_write_timeout(struct work_struct *work) |
a7f6a5fb | 195 | { |
c4028958 DH |
196 | struct o2hb_region *reg = |
197 | container_of(work, struct o2hb_region, | |
198 | hr_write_timeout_work.work); | |
a7f6a5fb MF |
199 | |
200 | mlog(ML_ERROR, "Heartbeat write timeout to device %s after %u " | |
201 | "milliseconds\n", reg->hr_dev_name, | |
2bd63216 | 202 | jiffies_to_msecs(jiffies - reg->hr_last_timeout_start)); |
a7f6a5fb MF |
203 | o2quo_disk_timeout(); |
204 | } | |
205 | ||
206 | static void o2hb_arm_write_timeout(struct o2hb_region *reg) | |
207 | { | |
b31d308d TM |
208 | mlog(ML_HEARTBEAT, "Queue write timeout for %u ms\n", |
209 | O2HB_MAX_WRITE_TIMEOUT_MS); | |
a7f6a5fb MF |
210 | |
211 | cancel_delayed_work(®->hr_write_timeout_work); | |
212 | reg->hr_last_timeout_start = jiffies; | |
213 | schedule_delayed_work(®->hr_write_timeout_work, | |
214 | msecs_to_jiffies(O2HB_MAX_WRITE_TIMEOUT_MS)); | |
215 | } | |
216 | ||
217 | static void o2hb_disarm_write_timeout(struct o2hb_region *reg) | |
218 | { | |
219 | cancel_delayed_work(®->hr_write_timeout_work); | |
220 | flush_scheduled_work(); | |
221 | } | |
222 | ||
b559292e | 223 | static inline void o2hb_bio_wait_init(struct o2hb_bio_wait_ctxt *wc) |
a7f6a5fb | 224 | { |
b559292e | 225 | atomic_set(&wc->wc_num_reqs, 1); |
a7f6a5fb | 226 | init_completion(&wc->wc_io_complete); |
a9e2ae39 | 227 | wc->wc_error = 0; |
a7f6a5fb MF |
228 | } |
229 | ||
230 | /* Used in error paths too */ | |
231 | static inline void o2hb_bio_wait_dec(struct o2hb_bio_wait_ctxt *wc, | |
232 | unsigned int num) | |
233 | { | |
234 | /* sadly atomic_sub_and_test() isn't available on all platforms. The | |
235 | * good news is that the fast path only completes one at a time */ | |
236 | while(num--) { | |
237 | if (atomic_dec_and_test(&wc->wc_num_reqs)) { | |
238 | BUG_ON(num > 0); | |
239 | complete(&wc->wc_io_complete); | |
240 | } | |
241 | } | |
242 | } | |
243 | ||
244 | static void o2hb_wait_on_io(struct o2hb_region *reg, | |
245 | struct o2hb_bio_wait_ctxt *wc) | |
246 | { | |
247 | struct address_space *mapping = reg->hr_bdev->bd_inode->i_mapping; | |
248 | ||
249 | blk_run_address_space(mapping); | |
b559292e | 250 | o2hb_bio_wait_dec(wc, 1); |
a7f6a5fb MF |
251 | |
252 | wait_for_completion(&wc->wc_io_complete); | |
253 | } | |
254 | ||
782e3b3b | 255 | static void o2hb_bio_end_io(struct bio *bio, |
a7f6a5fb MF |
256 | int error) |
257 | { | |
258 | struct o2hb_bio_wait_ctxt *wc = bio->bi_private; | |
259 | ||
a9e2ae39 | 260 | if (error) { |
a7f6a5fb | 261 | mlog(ML_ERROR, "IO Error %d\n", error); |
a9e2ae39 MF |
262 | wc->wc_error = error; |
263 | } | |
a7f6a5fb | 264 | |
a7f6a5fb | 265 | o2hb_bio_wait_dec(wc, 1); |
b559292e | 266 | bio_put(bio); |
a7f6a5fb MF |
267 | } |
268 | ||
269 | /* Setup a Bio to cover I/O against num_slots slots starting at | |
270 | * start_slot. */ | |
271 | static struct bio *o2hb_setup_one_bio(struct o2hb_region *reg, | |
272 | struct o2hb_bio_wait_ctxt *wc, | |
b559292e PR |
273 | unsigned int *current_slot, |
274 | unsigned int max_slots) | |
a7f6a5fb | 275 | { |
b559292e | 276 | int len, current_page; |
a7f6a5fb MF |
277 | unsigned int vec_len, vec_start; |
278 | unsigned int bits = reg->hr_block_bits; | |
279 | unsigned int spp = reg->hr_slots_per_page; | |
b559292e | 280 | unsigned int cs = *current_slot; |
a7f6a5fb MF |
281 | struct bio *bio; |
282 | struct page *page; | |
283 | ||
a7f6a5fb MF |
284 | /* Testing has shown this allocation to take long enough under |
285 | * GFP_KERNEL that the local node can get fenced. It would be | |
286 | * nicest if we could pre-allocate these bios and avoid this | |
287 | * all together. */ | |
b559292e | 288 | bio = bio_alloc(GFP_ATOMIC, 16); |
a7f6a5fb MF |
289 | if (!bio) { |
290 | mlog(ML_ERROR, "Could not alloc slots BIO!\n"); | |
291 | bio = ERR_PTR(-ENOMEM); | |
292 | goto bail; | |
293 | } | |
294 | ||
295 | /* Must put everything in 512 byte sectors for the bio... */ | |
b559292e | 296 | bio->bi_sector = (reg->hr_start_block + cs) << (bits - 9); |
a7f6a5fb MF |
297 | bio->bi_bdev = reg->hr_bdev; |
298 | bio->bi_private = wc; | |
299 | bio->bi_end_io = o2hb_bio_end_io; | |
300 | ||
b559292e PR |
301 | vec_start = (cs << bits) % PAGE_CACHE_SIZE; |
302 | while(cs < max_slots) { | |
303 | current_page = cs / spp; | |
304 | page = reg->hr_slot_data[current_page]; | |
a7f6a5fb | 305 | |
bc7e97cb | 306 | vec_len = min(PAGE_CACHE_SIZE - vec_start, |
b559292e | 307 | (max_slots-cs) * (PAGE_CACHE_SIZE/spp) ); |
a7f6a5fb MF |
308 | |
309 | mlog(ML_HB_BIO, "page %d, vec_len = %u, vec_start = %u\n", | |
b559292e | 310 | current_page, vec_len, vec_start); |
a7f6a5fb MF |
311 | |
312 | len = bio_add_page(bio, page, vec_len, vec_start); | |
b559292e | 313 | if (len != vec_len) break; |
a7f6a5fb | 314 | |
b559292e | 315 | cs += vec_len / (PAGE_CACHE_SIZE/spp); |
a7f6a5fb MF |
316 | vec_start = 0; |
317 | } | |
318 | ||
319 | bail: | |
b559292e | 320 | *current_slot = cs; |
a7f6a5fb MF |
321 | return bio; |
322 | } | |
323 | ||
a7f6a5fb MF |
324 | static int o2hb_read_slots(struct o2hb_region *reg, |
325 | unsigned int max_slots) | |
326 | { | |
b559292e PR |
327 | unsigned int current_slot=0; |
328 | int status; | |
a7f6a5fb | 329 | struct o2hb_bio_wait_ctxt wc; |
a7f6a5fb MF |
330 | struct bio *bio; |
331 | ||
b559292e | 332 | o2hb_bio_wait_init(&wc); |
a7f6a5fb | 333 | |
b559292e PR |
334 | while(current_slot < max_slots) { |
335 | bio = o2hb_setup_one_bio(reg, &wc, ¤t_slot, max_slots); | |
a7f6a5fb | 336 | if (IS_ERR(bio)) { |
a7f6a5fb MF |
337 | status = PTR_ERR(bio); |
338 | mlog_errno(status); | |
339 | goto bail_and_wait; | |
340 | } | |
a7f6a5fb | 341 | |
b559292e | 342 | atomic_inc(&wc.wc_num_reqs); |
a7f6a5fb MF |
343 | submit_bio(READ, bio); |
344 | } | |
345 | ||
346 | status = 0; | |
347 | ||
348 | bail_and_wait: | |
349 | o2hb_wait_on_io(reg, &wc); | |
a9e2ae39 MF |
350 | if (wc.wc_error && !status) |
351 | status = wc.wc_error; | |
a7f6a5fb | 352 | |
a7f6a5fb MF |
353 | return status; |
354 | } | |
355 | ||
356 | static int o2hb_issue_node_write(struct o2hb_region *reg, | |
a7f6a5fb MF |
357 | struct o2hb_bio_wait_ctxt *write_wc) |
358 | { | |
359 | int status; | |
360 | unsigned int slot; | |
361 | struct bio *bio; | |
362 | ||
b559292e | 363 | o2hb_bio_wait_init(write_wc); |
a7f6a5fb MF |
364 | |
365 | slot = o2nm_this_node(); | |
366 | ||
b559292e | 367 | bio = o2hb_setup_one_bio(reg, write_wc, &slot, slot+1); |
a7f6a5fb MF |
368 | if (IS_ERR(bio)) { |
369 | status = PTR_ERR(bio); | |
370 | mlog_errno(status); | |
371 | goto bail; | |
372 | } | |
373 | ||
b559292e | 374 | atomic_inc(&write_wc->wc_num_reqs); |
a7f6a5fb MF |
375 | submit_bio(WRITE, bio); |
376 | ||
a7f6a5fb MF |
377 | status = 0; |
378 | bail: | |
379 | return status; | |
380 | } | |
381 | ||
382 | static u32 o2hb_compute_block_crc_le(struct o2hb_region *reg, | |
383 | struct o2hb_disk_heartbeat_block *hb_block) | |
384 | { | |
385 | __le32 old_cksum; | |
386 | u32 ret; | |
387 | ||
388 | /* We want to compute the block crc with a 0 value in the | |
389 | * hb_cksum field. Save it off here and replace after the | |
390 | * crc. */ | |
391 | old_cksum = hb_block->hb_cksum; | |
392 | hb_block->hb_cksum = 0; | |
393 | ||
394 | ret = crc32_le(0, (unsigned char *) hb_block, reg->hr_block_bytes); | |
395 | ||
396 | hb_block->hb_cksum = old_cksum; | |
397 | ||
398 | return ret; | |
399 | } | |
400 | ||
401 | static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block *hb_block) | |
402 | { | |
70bacbdb MF |
403 | mlog(ML_ERROR, "Dump slot information: seq = 0x%llx, node = %u, " |
404 | "cksum = 0x%x, generation 0x%llx\n", | |
405 | (long long)le64_to_cpu(hb_block->hb_seq), | |
406 | hb_block->hb_node, le32_to_cpu(hb_block->hb_cksum), | |
407 | (long long)le64_to_cpu(hb_block->hb_generation)); | |
a7f6a5fb MF |
408 | } |
409 | ||
410 | static int o2hb_verify_crc(struct o2hb_region *reg, | |
411 | struct o2hb_disk_heartbeat_block *hb_block) | |
412 | { | |
413 | u32 read, computed; | |
414 | ||
415 | read = le32_to_cpu(hb_block->hb_cksum); | |
416 | computed = o2hb_compute_block_crc_le(reg, hb_block); | |
417 | ||
418 | return read == computed; | |
419 | } | |
420 | ||
421 | /* We want to make sure that nobody is heartbeating on top of us -- | |
422 | * this will help detect an invalid configuration. */ | |
423 | static int o2hb_check_last_timestamp(struct o2hb_region *reg) | |
424 | { | |
425 | int node_num, ret; | |
426 | struct o2hb_disk_slot *slot; | |
427 | struct o2hb_disk_heartbeat_block *hb_block; | |
428 | ||
429 | node_num = o2nm_this_node(); | |
430 | ||
431 | ret = 1; | |
432 | slot = ®->hr_slots[node_num]; | |
433 | /* Don't check on our 1st timestamp */ | |
434 | if (slot->ds_last_time) { | |
435 | hb_block = slot->ds_raw_block; | |
436 | ||
437 | if (le64_to_cpu(hb_block->hb_seq) != slot->ds_last_time) | |
438 | ret = 0; | |
439 | } | |
440 | ||
441 | return ret; | |
442 | } | |
443 | ||
444 | static inline void o2hb_prepare_block(struct o2hb_region *reg, | |
445 | u64 generation) | |
446 | { | |
447 | int node_num; | |
448 | u64 cputime; | |
449 | struct o2hb_disk_slot *slot; | |
450 | struct o2hb_disk_heartbeat_block *hb_block; | |
451 | ||
452 | node_num = o2nm_this_node(); | |
453 | slot = ®->hr_slots[node_num]; | |
454 | ||
455 | hb_block = (struct o2hb_disk_heartbeat_block *)slot->ds_raw_block; | |
456 | memset(hb_block, 0, reg->hr_block_bytes); | |
457 | /* TODO: time stuff */ | |
458 | cputime = CURRENT_TIME.tv_sec; | |
459 | if (!cputime) | |
460 | cputime = 1; | |
461 | ||
462 | hb_block->hb_seq = cpu_to_le64(cputime); | |
463 | hb_block->hb_node = node_num; | |
464 | hb_block->hb_generation = cpu_to_le64(generation); | |
0db638f4 | 465 | hb_block->hb_dead_ms = cpu_to_le32(o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS); |
a7f6a5fb MF |
466 | |
467 | /* This step must always happen last! */ | |
468 | hb_block->hb_cksum = cpu_to_le32(o2hb_compute_block_crc_le(reg, | |
469 | hb_block)); | |
470 | ||
70bacbdb | 471 | mlog(ML_HB_BIO, "our node generation = 0x%llx, cksum = 0x%x\n", |
5fdf1e67 | 472 | (long long)generation, |
70bacbdb | 473 | le32_to_cpu(hb_block->hb_cksum)); |
a7f6a5fb MF |
474 | } |
475 | ||
476 | static void o2hb_fire_callbacks(struct o2hb_callback *hbcall, | |
477 | struct o2nm_node *node, | |
478 | int idx) | |
479 | { | |
480 | struct list_head *iter; | |
481 | struct o2hb_callback_func *f; | |
482 | ||
483 | list_for_each(iter, &hbcall->list) { | |
484 | f = list_entry(iter, struct o2hb_callback_func, hc_item); | |
485 | mlog(ML_HEARTBEAT, "calling funcs %p\n", f); | |
486 | (f->hc_func)(node, idx, f->hc_data); | |
487 | } | |
488 | } | |
489 | ||
490 | /* Will run the list in order until we process the passed event */ | |
491 | static void o2hb_run_event_list(struct o2hb_node_event *queued_event) | |
492 | { | |
493 | int empty; | |
494 | struct o2hb_callback *hbcall; | |
495 | struct o2hb_node_event *event; | |
496 | ||
497 | spin_lock(&o2hb_live_lock); | |
498 | empty = list_empty(&queued_event->hn_item); | |
499 | spin_unlock(&o2hb_live_lock); | |
500 | if (empty) | |
501 | return; | |
502 | ||
503 | /* Holding callback sem assures we don't alter the callback | |
504 | * lists when doing this, and serializes ourselves with other | |
505 | * processes wanting callbacks. */ | |
506 | down_write(&o2hb_callback_sem); | |
507 | ||
508 | spin_lock(&o2hb_live_lock); | |
509 | while (!list_empty(&o2hb_node_events) | |
510 | && !list_empty(&queued_event->hn_item)) { | |
511 | event = list_entry(o2hb_node_events.next, | |
512 | struct o2hb_node_event, | |
513 | hn_item); | |
514 | list_del_init(&event->hn_item); | |
515 | spin_unlock(&o2hb_live_lock); | |
516 | ||
517 | mlog(ML_HEARTBEAT, "Node %s event for %d\n", | |
518 | event->hn_event_type == O2HB_NODE_UP_CB ? "UP" : "DOWN", | |
519 | event->hn_node_num); | |
520 | ||
521 | hbcall = hbcall_from_type(event->hn_event_type); | |
522 | ||
523 | /* We should *never* have gotten on to the list with a | |
524 | * bad type... This isn't something that we should try | |
525 | * to recover from. */ | |
526 | BUG_ON(IS_ERR(hbcall)); | |
527 | ||
528 | o2hb_fire_callbacks(hbcall, event->hn_node, event->hn_node_num); | |
529 | ||
530 | spin_lock(&o2hb_live_lock); | |
531 | } | |
532 | spin_unlock(&o2hb_live_lock); | |
533 | ||
534 | up_write(&o2hb_callback_sem); | |
535 | } | |
536 | ||
537 | static void o2hb_queue_node_event(struct o2hb_node_event *event, | |
538 | enum o2hb_callback_type type, | |
539 | struct o2nm_node *node, | |
540 | int node_num) | |
541 | { | |
542 | assert_spin_locked(&o2hb_live_lock); | |
543 | ||
544 | event->hn_event_type = type; | |
545 | event->hn_node = node; | |
546 | event->hn_node_num = node_num; | |
547 | ||
548 | mlog(ML_HEARTBEAT, "Queue node %s event for node %d\n", | |
549 | type == O2HB_NODE_UP_CB ? "UP" : "DOWN", node_num); | |
550 | ||
551 | list_add_tail(&event->hn_item, &o2hb_node_events); | |
552 | } | |
553 | ||
554 | static void o2hb_shutdown_slot(struct o2hb_disk_slot *slot) | |
555 | { | |
556 | struct o2hb_node_event event = | |
557 | { .hn_item = LIST_HEAD_INIT(event.hn_item), }; | |
558 | struct o2nm_node *node; | |
559 | ||
560 | node = o2nm_get_node_by_num(slot->ds_node_num); | |
561 | if (!node) | |
562 | return; | |
563 | ||
564 | spin_lock(&o2hb_live_lock); | |
565 | if (!list_empty(&slot->ds_live_item)) { | |
566 | mlog(ML_HEARTBEAT, "Shutdown, node %d leaves region\n", | |
567 | slot->ds_node_num); | |
568 | ||
569 | list_del_init(&slot->ds_live_item); | |
570 | ||
571 | if (list_empty(&o2hb_live_slots[slot->ds_node_num])) { | |
572 | clear_bit(slot->ds_node_num, o2hb_live_node_bitmap); | |
573 | ||
574 | o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node, | |
575 | slot->ds_node_num); | |
576 | } | |
577 | } | |
578 | spin_unlock(&o2hb_live_lock); | |
579 | ||
580 | o2hb_run_event_list(&event); | |
581 | ||
582 | o2nm_node_put(node); | |
583 | } | |
584 | ||
585 | static int o2hb_check_slot(struct o2hb_region *reg, | |
586 | struct o2hb_disk_slot *slot) | |
587 | { | |
588 | int changed = 0, gen_changed = 0; | |
589 | struct o2hb_node_event event = | |
590 | { .hn_item = LIST_HEAD_INIT(event.hn_item), }; | |
591 | struct o2nm_node *node; | |
592 | struct o2hb_disk_heartbeat_block *hb_block = reg->hr_tmp_block; | |
593 | u64 cputime; | |
0db638f4 MF |
594 | unsigned int dead_ms = o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS; |
595 | unsigned int slot_dead_ms; | |
a7f6a5fb MF |
596 | |
597 | memcpy(hb_block, slot->ds_raw_block, reg->hr_block_bytes); | |
598 | ||
599 | /* Is this correct? Do we assume that the node doesn't exist | |
600 | * if we're not configured for him? */ | |
601 | node = o2nm_get_node_by_num(slot->ds_node_num); | |
602 | if (!node) | |
603 | return 0; | |
604 | ||
605 | if (!o2hb_verify_crc(reg, hb_block)) { | |
606 | /* all paths from here will drop o2hb_live_lock for | |
607 | * us. */ | |
608 | spin_lock(&o2hb_live_lock); | |
609 | ||
610 | /* Don't print an error on the console in this case - | |
611 | * a freshly formatted heartbeat area will not have a | |
612 | * crc set on it. */ | |
613 | if (list_empty(&slot->ds_live_item)) | |
614 | goto out; | |
615 | ||
616 | /* The node is live but pushed out a bad crc. We | |
617 | * consider it a transient miss but don't populate any | |
618 | * other values as they may be junk. */ | |
619 | mlog(ML_ERROR, "Node %d has written a bad crc to %s\n", | |
620 | slot->ds_node_num, reg->hr_dev_name); | |
621 | o2hb_dump_slot(hb_block); | |
622 | ||
623 | slot->ds_equal_samples++; | |
624 | goto fire_callbacks; | |
625 | } | |
626 | ||
627 | /* we don't care if these wrap.. the state transitions below | |
628 | * clear at the right places */ | |
629 | cputime = le64_to_cpu(hb_block->hb_seq); | |
630 | if (slot->ds_last_time != cputime) | |
631 | slot->ds_changed_samples++; | |
632 | else | |
633 | slot->ds_equal_samples++; | |
634 | slot->ds_last_time = cputime; | |
635 | ||
636 | /* The node changed heartbeat generations. We assume this to | |
637 | * mean it dropped off but came back before we timed out. We | |
638 | * want to consider it down for the time being but don't want | |
639 | * to lose any changed_samples state we might build up to | |
640 | * considering it live again. */ | |
641 | if (slot->ds_last_generation != le64_to_cpu(hb_block->hb_generation)) { | |
642 | gen_changed = 1; | |
643 | slot->ds_equal_samples = 0; | |
70bacbdb MF |
644 | mlog(ML_HEARTBEAT, "Node %d changed generation (0x%llx " |
645 | "to 0x%llx)\n", slot->ds_node_num, | |
646 | (long long)slot->ds_last_generation, | |
647 | (long long)le64_to_cpu(hb_block->hb_generation)); | |
a7f6a5fb MF |
648 | } |
649 | ||
650 | slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation); | |
651 | ||
70bacbdb MF |
652 | mlog(ML_HEARTBEAT, "Slot %d gen 0x%llx cksum 0x%x " |
653 | "seq %llu last %llu changed %u equal %u\n", | |
654 | slot->ds_node_num, (long long)slot->ds_last_generation, | |
655 | le32_to_cpu(hb_block->hb_cksum), | |
2bd63216 | 656 | (unsigned long long)le64_to_cpu(hb_block->hb_seq), |
70bacbdb | 657 | (unsigned long long)slot->ds_last_time, slot->ds_changed_samples, |
a7f6a5fb MF |
658 | slot->ds_equal_samples); |
659 | ||
660 | spin_lock(&o2hb_live_lock); | |
661 | ||
662 | fire_callbacks: | |
663 | /* dead nodes only come to life after some number of | |
664 | * changes at any time during their dead time */ | |
665 | if (list_empty(&slot->ds_live_item) && | |
666 | slot->ds_changed_samples >= O2HB_LIVE_THRESHOLD) { | |
70bacbdb MF |
667 | mlog(ML_HEARTBEAT, "Node %d (id 0x%llx) joined my region\n", |
668 | slot->ds_node_num, (long long)slot->ds_last_generation); | |
a7f6a5fb MF |
669 | |
670 | /* first on the list generates a callback */ | |
671 | if (list_empty(&o2hb_live_slots[slot->ds_node_num])) { | |
672 | set_bit(slot->ds_node_num, o2hb_live_node_bitmap); | |
673 | ||
674 | o2hb_queue_node_event(&event, O2HB_NODE_UP_CB, node, | |
675 | slot->ds_node_num); | |
676 | ||
677 | changed = 1; | |
678 | } | |
679 | ||
680 | list_add_tail(&slot->ds_live_item, | |
681 | &o2hb_live_slots[slot->ds_node_num]); | |
682 | ||
683 | slot->ds_equal_samples = 0; | |
0db638f4 MF |
684 | |
685 | /* We want to be sure that all nodes agree on the | |
686 | * number of milliseconds before a node will be | |
687 | * considered dead. The self-fencing timeout is | |
688 | * computed from this value, and a discrepancy might | |
689 | * result in heartbeat calling a node dead when it | |
690 | * hasn't self-fenced yet. */ | |
691 | slot_dead_ms = le32_to_cpu(hb_block->hb_dead_ms); | |
692 | if (slot_dead_ms && slot_dead_ms != dead_ms) { | |
693 | /* TODO: Perhaps we can fail the region here. */ | |
694 | mlog(ML_ERROR, "Node %d on device %s has a dead count " | |
695 | "of %u ms, but our count is %u ms.\n" | |
696 | "Please double check your configuration values " | |
697 | "for 'O2CB_HEARTBEAT_THRESHOLD'\n", | |
698 | slot->ds_node_num, reg->hr_dev_name, slot_dead_ms, | |
699 | dead_ms); | |
700 | } | |
a7f6a5fb MF |
701 | goto out; |
702 | } | |
703 | ||
704 | /* if the list is dead, we're done.. */ | |
705 | if (list_empty(&slot->ds_live_item)) | |
706 | goto out; | |
707 | ||
708 | /* live nodes only go dead after enough consequtive missed | |
709 | * samples.. reset the missed counter whenever we see | |
710 | * activity */ | |
711 | if (slot->ds_equal_samples >= o2hb_dead_threshold || gen_changed) { | |
712 | mlog(ML_HEARTBEAT, "Node %d left my region\n", | |
713 | slot->ds_node_num); | |
714 | ||
715 | /* last off the live_slot generates a callback */ | |
716 | list_del_init(&slot->ds_live_item); | |
717 | if (list_empty(&o2hb_live_slots[slot->ds_node_num])) { | |
718 | clear_bit(slot->ds_node_num, o2hb_live_node_bitmap); | |
719 | ||
720 | o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node, | |
721 | slot->ds_node_num); | |
722 | ||
723 | changed = 1; | |
724 | } | |
725 | ||
726 | /* We don't clear this because the node is still | |
727 | * actually writing new blocks. */ | |
728 | if (!gen_changed) | |
729 | slot->ds_changed_samples = 0; | |
730 | goto out; | |
731 | } | |
732 | if (slot->ds_changed_samples) { | |
733 | slot->ds_changed_samples = 0; | |
734 | slot->ds_equal_samples = 0; | |
735 | } | |
736 | out: | |
737 | spin_unlock(&o2hb_live_lock); | |
738 | ||
739 | o2hb_run_event_list(&event); | |
740 | ||
741 | o2nm_node_put(node); | |
742 | return changed; | |
743 | } | |
744 | ||
745 | /* This could be faster if we just implmented a find_last_bit, but I | |
746 | * don't think the circumstances warrant it. */ | |
747 | static int o2hb_highest_node(unsigned long *nodes, | |
748 | int numbits) | |
749 | { | |
750 | int highest, node; | |
751 | ||
752 | highest = numbits; | |
753 | node = -1; | |
754 | while ((node = find_next_bit(nodes, numbits, node + 1)) != -1) { | |
755 | if (node >= numbits) | |
756 | break; | |
757 | ||
758 | highest = node; | |
759 | } | |
760 | ||
761 | return highest; | |
762 | } | |
763 | ||
a9e2ae39 | 764 | static int o2hb_do_disk_heartbeat(struct o2hb_region *reg) |
a7f6a5fb MF |
765 | { |
766 | int i, ret, highest_node, change = 0; | |
767 | unsigned long configured_nodes[BITS_TO_LONGS(O2NM_MAX_NODES)]; | |
a7f6a5fb MF |
768 | struct o2hb_bio_wait_ctxt write_wc; |
769 | ||
a9e2ae39 MF |
770 | ret = o2nm_configured_node_map(configured_nodes, |
771 | sizeof(configured_nodes)); | |
772 | if (ret) { | |
773 | mlog_errno(ret); | |
774 | return ret; | |
775 | } | |
a7f6a5fb MF |
776 | |
777 | highest_node = o2hb_highest_node(configured_nodes, O2NM_MAX_NODES); | |
778 | if (highest_node >= O2NM_MAX_NODES) { | |
779 | mlog(ML_NOTICE, "ocfs2_heartbeat: no configured nodes found!\n"); | |
a9e2ae39 | 780 | return -EINVAL; |
a7f6a5fb MF |
781 | } |
782 | ||
783 | /* No sense in reading the slots of nodes that don't exist | |
784 | * yet. Of course, if the node definitions have holes in them | |
785 | * then we're reading an empty slot anyway... Consider this | |
786 | * best-effort. */ | |
787 | ret = o2hb_read_slots(reg, highest_node + 1); | |
788 | if (ret < 0) { | |
789 | mlog_errno(ret); | |
a9e2ae39 | 790 | return ret; |
a7f6a5fb MF |
791 | } |
792 | ||
793 | /* With an up to date view of the slots, we can check that no | |
794 | * other node has been improperly configured to heartbeat in | |
795 | * our slot. */ | |
796 | if (!o2hb_check_last_timestamp(reg)) | |
797 | mlog(ML_ERROR, "Device \"%s\": another node is heartbeating " | |
798 | "in our slot!\n", reg->hr_dev_name); | |
799 | ||
800 | /* fill in the proper info for our next heartbeat */ | |
801 | o2hb_prepare_block(reg, reg->hr_generation); | |
802 | ||
803 | /* And fire off the write. Note that we don't wait on this I/O | |
804 | * until later. */ | |
b559292e | 805 | ret = o2hb_issue_node_write(reg, &write_wc); |
a7f6a5fb MF |
806 | if (ret < 0) { |
807 | mlog_errno(ret); | |
a9e2ae39 | 808 | return ret; |
a7f6a5fb MF |
809 | } |
810 | ||
811 | i = -1; | |
812 | while((i = find_next_bit(configured_nodes, O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) { | |
813 | ||
814 | change |= o2hb_check_slot(reg, ®->hr_slots[i]); | |
815 | } | |
816 | ||
817 | /* | |
818 | * We have to be sure we've advertised ourselves on disk | |
819 | * before we can go to steady state. This ensures that | |
820 | * people we find in our steady state have seen us. | |
821 | */ | |
822 | o2hb_wait_on_io(reg, &write_wc); | |
a9e2ae39 MF |
823 | if (write_wc.wc_error) { |
824 | /* Do not re-arm the write timeout on I/O error - we | |
825 | * can't be sure that the new block ever made it to | |
826 | * disk */ | |
827 | mlog(ML_ERROR, "Write error %d on device \"%s\"\n", | |
828 | write_wc.wc_error, reg->hr_dev_name); | |
829 | return write_wc.wc_error; | |
830 | } | |
831 | ||
a7f6a5fb MF |
832 | o2hb_arm_write_timeout(reg); |
833 | ||
834 | /* let the person who launched us know when things are steady */ | |
835 | if (!change && (atomic_read(®->hr_steady_iterations) != 0)) { | |
836 | if (atomic_dec_and_test(®->hr_steady_iterations)) | |
837 | wake_up(&o2hb_steady_queue); | |
838 | } | |
a9e2ae39 MF |
839 | |
840 | return 0; | |
a7f6a5fb MF |
841 | } |
842 | ||
843 | /* Subtract b from a, storing the result in a. a *must* have a larger | |
844 | * value than b. */ | |
845 | static void o2hb_tv_subtract(struct timeval *a, | |
846 | struct timeval *b) | |
847 | { | |
848 | /* just return 0 when a is after b */ | |
849 | if (a->tv_sec < b->tv_sec || | |
850 | (a->tv_sec == b->tv_sec && a->tv_usec < b->tv_usec)) { | |
851 | a->tv_sec = 0; | |
852 | a->tv_usec = 0; | |
853 | return; | |
854 | } | |
855 | ||
856 | a->tv_sec -= b->tv_sec; | |
857 | a->tv_usec -= b->tv_usec; | |
858 | while ( a->tv_usec < 0 ) { | |
859 | a->tv_sec--; | |
860 | a->tv_usec += 1000000; | |
861 | } | |
862 | } | |
863 | ||
864 | static unsigned int o2hb_elapsed_msecs(struct timeval *start, | |
865 | struct timeval *end) | |
866 | { | |
867 | struct timeval res = *end; | |
868 | ||
869 | o2hb_tv_subtract(&res, start); | |
870 | ||
871 | return res.tv_sec * 1000 + res.tv_usec / 1000; | |
872 | } | |
873 | ||
874 | /* | |
875 | * we ride the region ref that the region dir holds. before the region | |
876 | * dir is removed and drops it ref it will wait to tear down this | |
877 | * thread. | |
878 | */ | |
879 | static int o2hb_thread(void *data) | |
880 | { | |
881 | int i, ret; | |
882 | struct o2hb_region *reg = data; | |
a7f6a5fb MF |
883 | struct o2hb_bio_wait_ctxt write_wc; |
884 | struct timeval before_hb, after_hb; | |
885 | unsigned int elapsed_msec; | |
886 | ||
887 | mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread running\n"); | |
888 | ||
889 | set_user_nice(current, -20); | |
890 | ||
891 | while (!kthread_should_stop() && !reg->hr_unclean_stop) { | |
892 | /* We track the time spent inside | |
025dfdaf | 893 | * o2hb_do_disk_heartbeat so that we avoid more than |
a7f6a5fb MF |
894 | * hr_timeout_ms between disk writes. On busy systems |
895 | * this should result in a heartbeat which is less | |
896 | * likely to time itself out. */ | |
897 | do_gettimeofday(&before_hb); | |
898 | ||
a9e2ae39 MF |
899 | i = 0; |
900 | do { | |
901 | ret = o2hb_do_disk_heartbeat(reg); | |
902 | } while (ret && ++i < 2); | |
a7f6a5fb MF |
903 | |
904 | do_gettimeofday(&after_hb); | |
905 | elapsed_msec = o2hb_elapsed_msecs(&before_hb, &after_hb); | |
906 | ||
b31d308d TM |
907 | mlog(ML_HEARTBEAT, |
908 | "start = %lu.%lu, end = %lu.%lu, msec = %u\n", | |
215c7f9f MF |
909 | before_hb.tv_sec, (unsigned long) before_hb.tv_usec, |
910 | after_hb.tv_sec, (unsigned long) after_hb.tv_usec, | |
911 | elapsed_msec); | |
a7f6a5fb MF |
912 | |
913 | if (elapsed_msec < reg->hr_timeout_ms) { | |
914 | /* the kthread api has blocked signals for us so no | |
915 | * need to record the return value. */ | |
916 | msleep_interruptible(reg->hr_timeout_ms - elapsed_msec); | |
917 | } | |
918 | } | |
919 | ||
920 | o2hb_disarm_write_timeout(reg); | |
921 | ||
922 | /* unclean stop is only used in very bad situation */ | |
923 | for(i = 0; !reg->hr_unclean_stop && i < reg->hr_blocks; i++) | |
924 | o2hb_shutdown_slot(®->hr_slots[i]); | |
925 | ||
926 | /* Explicit down notification - avoid forcing the other nodes | |
927 | * to timeout on this region when we could just as easily | |
928 | * write a clear generation - thus indicating to them that | |
929 | * this node has left this region. | |
930 | * | |
931 | * XXX: Should we skip this on unclean_stop? */ | |
932 | o2hb_prepare_block(reg, 0); | |
b559292e | 933 | ret = o2hb_issue_node_write(reg, &write_wc); |
a7f6a5fb MF |
934 | if (ret == 0) { |
935 | o2hb_wait_on_io(reg, &write_wc); | |
a7f6a5fb MF |
936 | } else { |
937 | mlog_errno(ret); | |
938 | } | |
939 | ||
940 | mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread exiting\n"); | |
941 | ||
942 | return 0; | |
943 | } | |
944 | ||
87d3d3f3 SM |
945 | #ifdef CONFIG_DEBUG_FS |
946 | static int o2hb_debug_open(struct inode *inode, struct file *file) | |
947 | { | |
948 | unsigned long map[BITS_TO_LONGS(O2NM_MAX_NODES)]; | |
949 | char *buf = NULL; | |
950 | int i = -1; | |
951 | int out = 0; | |
952 | ||
953 | buf = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
954 | if (!buf) | |
955 | goto bail; | |
956 | ||
957 | o2hb_fill_node_map(map, sizeof(map)); | |
958 | ||
959 | while ((i = find_next_bit(map, O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) | |
960 | out += snprintf(buf + out, PAGE_SIZE - out, "%d ", i); | |
961 | out += snprintf(buf + out, PAGE_SIZE - out, "\n"); | |
962 | ||
963 | i_size_write(inode, out); | |
964 | ||
965 | file->private_data = buf; | |
966 | ||
967 | return 0; | |
968 | bail: | |
969 | return -ENOMEM; | |
970 | } | |
971 | ||
972 | static int o2hb_debug_release(struct inode *inode, struct file *file) | |
973 | { | |
974 | kfree(file->private_data); | |
975 | return 0; | |
976 | } | |
977 | ||
978 | static ssize_t o2hb_debug_read(struct file *file, char __user *buf, | |
979 | size_t nbytes, loff_t *ppos) | |
980 | { | |
981 | return simple_read_from_buffer(buf, nbytes, ppos, file->private_data, | |
982 | i_size_read(file->f_mapping->host)); | |
983 | } | |
984 | #else | |
985 | static int o2hb_debug_open(struct inode *inode, struct file *file) | |
986 | { | |
987 | return 0; | |
988 | } | |
989 | static int o2hb_debug_release(struct inode *inode, struct file *file) | |
990 | { | |
991 | return 0; | |
992 | } | |
993 | static ssize_t o2hb_debug_read(struct file *file, char __user *buf, | |
994 | size_t nbytes, loff_t *ppos) | |
995 | { | |
996 | return 0; | |
997 | } | |
998 | #endif /* CONFIG_DEBUG_FS */ | |
999 | ||
828c0950 | 1000 | static const struct file_operations o2hb_debug_fops = { |
87d3d3f3 SM |
1001 | .open = o2hb_debug_open, |
1002 | .release = o2hb_debug_release, | |
1003 | .read = o2hb_debug_read, | |
1004 | .llseek = generic_file_llseek, | |
1005 | }; | |
1006 | ||
1007 | void o2hb_exit(void) | |
1008 | { | |
1009 | if (o2hb_debug_livenodes) | |
1010 | debugfs_remove(o2hb_debug_livenodes); | |
1011 | if (o2hb_debug_dir) | |
1012 | debugfs_remove(o2hb_debug_dir); | |
1013 | } | |
1014 | ||
1015 | int o2hb_init(void) | |
a7f6a5fb MF |
1016 | { |
1017 | int i; | |
1018 | ||
1019 | for (i = 0; i < ARRAY_SIZE(o2hb_callbacks); i++) | |
1020 | INIT_LIST_HEAD(&o2hb_callbacks[i].list); | |
1021 | ||
1022 | for (i = 0; i < ARRAY_SIZE(o2hb_live_slots); i++) | |
1023 | INIT_LIST_HEAD(&o2hb_live_slots[i]); | |
1024 | ||
1025 | INIT_LIST_HEAD(&o2hb_node_events); | |
1026 | ||
1027 | memset(o2hb_live_node_bitmap, 0, sizeof(o2hb_live_node_bitmap)); | |
87d3d3f3 SM |
1028 | |
1029 | o2hb_debug_dir = debugfs_create_dir(O2HB_DEBUG_DIR, NULL); | |
1030 | if (!o2hb_debug_dir) { | |
1031 | mlog_errno(-ENOMEM); | |
1032 | return -ENOMEM; | |
1033 | } | |
1034 | ||
1035 | o2hb_debug_livenodes = debugfs_create_file(O2HB_DEBUG_LIVENODES, | |
1036 | S_IFREG|S_IRUSR, | |
1037 | o2hb_debug_dir, NULL, | |
1038 | &o2hb_debug_fops); | |
1039 | if (!o2hb_debug_livenodes) { | |
1040 | mlog_errno(-ENOMEM); | |
1041 | debugfs_remove(o2hb_debug_dir); | |
1042 | return -ENOMEM; | |
1043 | } | |
1044 | ||
1045 | return 0; | |
a7f6a5fb MF |
1046 | } |
1047 | ||
1048 | /* if we're already in a callback then we're already serialized by the sem */ | |
1049 | static void o2hb_fill_node_map_from_callback(unsigned long *map, | |
1050 | unsigned bytes) | |
1051 | { | |
1052 | BUG_ON(bytes < (BITS_TO_LONGS(O2NM_MAX_NODES) * sizeof(unsigned long))); | |
1053 | ||
1054 | memcpy(map, &o2hb_live_node_bitmap, bytes); | |
1055 | } | |
1056 | ||
1057 | /* | |
1058 | * get a map of all nodes that are heartbeating in any regions | |
1059 | */ | |
1060 | void o2hb_fill_node_map(unsigned long *map, unsigned bytes) | |
1061 | { | |
1062 | /* callers want to serialize this map and callbacks so that they | |
1063 | * can trust that they don't miss nodes coming to the party */ | |
1064 | down_read(&o2hb_callback_sem); | |
1065 | spin_lock(&o2hb_live_lock); | |
1066 | o2hb_fill_node_map_from_callback(map, bytes); | |
1067 | spin_unlock(&o2hb_live_lock); | |
1068 | up_read(&o2hb_callback_sem); | |
1069 | } | |
1070 | EXPORT_SYMBOL_GPL(o2hb_fill_node_map); | |
1071 | ||
1072 | /* | |
1073 | * heartbeat configfs bits. The heartbeat set is a default set under | |
1074 | * the cluster set in nodemanager.c. | |
1075 | */ | |
1076 | ||
1077 | static struct o2hb_region *to_o2hb_region(struct config_item *item) | |
1078 | { | |
1079 | return item ? container_of(item, struct o2hb_region, hr_item) : NULL; | |
1080 | } | |
1081 | ||
1082 | /* drop_item only drops its ref after killing the thread, nothing should | |
1083 | * be using the region anymore. this has to clean up any state that | |
1084 | * attributes might have built up. */ | |
1085 | static void o2hb_region_release(struct config_item *item) | |
1086 | { | |
1087 | int i; | |
1088 | struct page *page; | |
1089 | struct o2hb_region *reg = to_o2hb_region(item); | |
1090 | ||
1091 | if (reg->hr_tmp_block) | |
1092 | kfree(reg->hr_tmp_block); | |
1093 | ||
1094 | if (reg->hr_slot_data) { | |
1095 | for (i = 0; i < reg->hr_num_pages; i++) { | |
1096 | page = reg->hr_slot_data[i]; | |
1097 | if (page) | |
1098 | __free_page(page); | |
1099 | } | |
1100 | kfree(reg->hr_slot_data); | |
1101 | } | |
1102 | ||
1103 | if (reg->hr_bdev) | |
9a1c3542 | 1104 | blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE); |
a7f6a5fb MF |
1105 | |
1106 | if (reg->hr_slots) | |
1107 | kfree(reg->hr_slots); | |
1108 | ||
1109 | spin_lock(&o2hb_live_lock); | |
1110 | list_del(®->hr_all_item); | |
1111 | spin_unlock(&o2hb_live_lock); | |
1112 | ||
1113 | kfree(reg); | |
1114 | } | |
1115 | ||
1116 | static int o2hb_read_block_input(struct o2hb_region *reg, | |
1117 | const char *page, | |
1118 | size_t count, | |
1119 | unsigned long *ret_bytes, | |
1120 | unsigned int *ret_bits) | |
1121 | { | |
1122 | unsigned long bytes; | |
1123 | char *p = (char *)page; | |
1124 | ||
1125 | bytes = simple_strtoul(p, &p, 0); | |
1126 | if (!p || (*p && (*p != '\n'))) | |
1127 | return -EINVAL; | |
1128 | ||
1129 | /* Heartbeat and fs min / max block sizes are the same. */ | |
1130 | if (bytes > 4096 || bytes < 512) | |
1131 | return -ERANGE; | |
1132 | if (hweight16(bytes) != 1) | |
1133 | return -EINVAL; | |
1134 | ||
1135 | if (ret_bytes) | |
1136 | *ret_bytes = bytes; | |
1137 | if (ret_bits) | |
1138 | *ret_bits = ffs(bytes) - 1; | |
1139 | ||
1140 | return 0; | |
1141 | } | |
1142 | ||
1143 | static ssize_t o2hb_region_block_bytes_read(struct o2hb_region *reg, | |
1144 | char *page) | |
1145 | { | |
1146 | return sprintf(page, "%u\n", reg->hr_block_bytes); | |
1147 | } | |
1148 | ||
1149 | static ssize_t o2hb_region_block_bytes_write(struct o2hb_region *reg, | |
1150 | const char *page, | |
1151 | size_t count) | |
1152 | { | |
1153 | int status; | |
1154 | unsigned long block_bytes; | |
1155 | unsigned int block_bits; | |
1156 | ||
1157 | if (reg->hr_bdev) | |
1158 | return -EINVAL; | |
1159 | ||
1160 | status = o2hb_read_block_input(reg, page, count, | |
1161 | &block_bytes, &block_bits); | |
1162 | if (status) | |
1163 | return status; | |
1164 | ||
1165 | reg->hr_block_bytes = (unsigned int)block_bytes; | |
1166 | reg->hr_block_bits = block_bits; | |
1167 | ||
1168 | return count; | |
1169 | } | |
1170 | ||
1171 | static ssize_t o2hb_region_start_block_read(struct o2hb_region *reg, | |
1172 | char *page) | |
1173 | { | |
1174 | return sprintf(page, "%llu\n", reg->hr_start_block); | |
1175 | } | |
1176 | ||
1177 | static ssize_t o2hb_region_start_block_write(struct o2hb_region *reg, | |
1178 | const char *page, | |
1179 | size_t count) | |
1180 | { | |
1181 | unsigned long long tmp; | |
1182 | char *p = (char *)page; | |
1183 | ||
1184 | if (reg->hr_bdev) | |
1185 | return -EINVAL; | |
1186 | ||
1187 | tmp = simple_strtoull(p, &p, 0); | |
1188 | if (!p || (*p && (*p != '\n'))) | |
1189 | return -EINVAL; | |
1190 | ||
1191 | reg->hr_start_block = tmp; | |
1192 | ||
1193 | return count; | |
1194 | } | |
1195 | ||
1196 | static ssize_t o2hb_region_blocks_read(struct o2hb_region *reg, | |
1197 | char *page) | |
1198 | { | |
1199 | return sprintf(page, "%d\n", reg->hr_blocks); | |
1200 | } | |
1201 | ||
1202 | static ssize_t o2hb_region_blocks_write(struct o2hb_region *reg, | |
1203 | const char *page, | |
1204 | size_t count) | |
1205 | { | |
1206 | unsigned long tmp; | |
1207 | char *p = (char *)page; | |
1208 | ||
1209 | if (reg->hr_bdev) | |
1210 | return -EINVAL; | |
1211 | ||
1212 | tmp = simple_strtoul(p, &p, 0); | |
1213 | if (!p || (*p && (*p != '\n'))) | |
1214 | return -EINVAL; | |
1215 | ||
1216 | if (tmp > O2NM_MAX_NODES || tmp == 0) | |
1217 | return -ERANGE; | |
1218 | ||
1219 | reg->hr_blocks = (unsigned int)tmp; | |
1220 | ||
1221 | return count; | |
1222 | } | |
1223 | ||
1224 | static ssize_t o2hb_region_dev_read(struct o2hb_region *reg, | |
1225 | char *page) | |
1226 | { | |
1227 | unsigned int ret = 0; | |
1228 | ||
1229 | if (reg->hr_bdev) | |
1230 | ret = sprintf(page, "%s\n", reg->hr_dev_name); | |
1231 | ||
1232 | return ret; | |
1233 | } | |
1234 | ||
1235 | static void o2hb_init_region_params(struct o2hb_region *reg) | |
1236 | { | |
1237 | reg->hr_slots_per_page = PAGE_CACHE_SIZE >> reg->hr_block_bits; | |
1238 | reg->hr_timeout_ms = O2HB_REGION_TIMEOUT_MS; | |
1239 | ||
1240 | mlog(ML_HEARTBEAT, "hr_start_block = %llu, hr_blocks = %u\n", | |
1241 | reg->hr_start_block, reg->hr_blocks); | |
1242 | mlog(ML_HEARTBEAT, "hr_block_bytes = %u, hr_block_bits = %u\n", | |
1243 | reg->hr_block_bytes, reg->hr_block_bits); | |
1244 | mlog(ML_HEARTBEAT, "hr_timeout_ms = %u\n", reg->hr_timeout_ms); | |
1245 | mlog(ML_HEARTBEAT, "dead threshold = %u\n", o2hb_dead_threshold); | |
1246 | } | |
1247 | ||
1248 | static int o2hb_map_slot_data(struct o2hb_region *reg) | |
1249 | { | |
1250 | int i, j; | |
1251 | unsigned int last_slot; | |
1252 | unsigned int spp = reg->hr_slots_per_page; | |
1253 | struct page *page; | |
1254 | char *raw; | |
1255 | struct o2hb_disk_slot *slot; | |
1256 | ||
1257 | reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL); | |
1258 | if (reg->hr_tmp_block == NULL) { | |
1259 | mlog_errno(-ENOMEM); | |
1260 | return -ENOMEM; | |
1261 | } | |
1262 | ||
1263 | reg->hr_slots = kcalloc(reg->hr_blocks, | |
1264 | sizeof(struct o2hb_disk_slot), GFP_KERNEL); | |
1265 | if (reg->hr_slots == NULL) { | |
1266 | mlog_errno(-ENOMEM); | |
1267 | return -ENOMEM; | |
1268 | } | |
1269 | ||
1270 | for(i = 0; i < reg->hr_blocks; i++) { | |
1271 | slot = ®->hr_slots[i]; | |
1272 | slot->ds_node_num = i; | |
1273 | INIT_LIST_HEAD(&slot->ds_live_item); | |
1274 | slot->ds_raw_block = NULL; | |
1275 | } | |
1276 | ||
1277 | reg->hr_num_pages = (reg->hr_blocks + spp - 1) / spp; | |
1278 | mlog(ML_HEARTBEAT, "Going to require %u pages to cover %u blocks " | |
1279 | "at %u blocks per page\n", | |
1280 | reg->hr_num_pages, reg->hr_blocks, spp); | |
1281 | ||
1282 | reg->hr_slot_data = kcalloc(reg->hr_num_pages, sizeof(struct page *), | |
1283 | GFP_KERNEL); | |
1284 | if (!reg->hr_slot_data) { | |
1285 | mlog_errno(-ENOMEM); | |
1286 | return -ENOMEM; | |
1287 | } | |
1288 | ||
1289 | for(i = 0; i < reg->hr_num_pages; i++) { | |
1290 | page = alloc_page(GFP_KERNEL); | |
1291 | if (!page) { | |
1292 | mlog_errno(-ENOMEM); | |
1293 | return -ENOMEM; | |
1294 | } | |
1295 | ||
1296 | reg->hr_slot_data[i] = page; | |
1297 | ||
1298 | last_slot = i * spp; | |
1299 | raw = page_address(page); | |
1300 | for (j = 0; | |
1301 | (j < spp) && ((j + last_slot) < reg->hr_blocks); | |
1302 | j++) { | |
1303 | BUG_ON((j + last_slot) >= reg->hr_blocks); | |
1304 | ||
1305 | slot = ®->hr_slots[j + last_slot]; | |
1306 | slot->ds_raw_block = | |
1307 | (struct o2hb_disk_heartbeat_block *) raw; | |
1308 | ||
1309 | raw += reg->hr_block_bytes; | |
1310 | } | |
1311 | } | |
1312 | ||
1313 | return 0; | |
1314 | } | |
1315 | ||
1316 | /* Read in all the slots available and populate the tracking | |
1317 | * structures so that we can start with a baseline idea of what's | |
1318 | * there. */ | |
1319 | static int o2hb_populate_slot_data(struct o2hb_region *reg) | |
1320 | { | |
1321 | int ret, i; | |
1322 | struct o2hb_disk_slot *slot; | |
1323 | struct o2hb_disk_heartbeat_block *hb_block; | |
1324 | ||
1325 | mlog_entry_void(); | |
1326 | ||
1327 | ret = o2hb_read_slots(reg, reg->hr_blocks); | |
1328 | if (ret) { | |
1329 | mlog_errno(ret); | |
1330 | goto out; | |
1331 | } | |
1332 | ||
1333 | /* We only want to get an idea of the values initially in each | |
1334 | * slot, so we do no verification - o2hb_check_slot will | |
1335 | * actually determine if each configured slot is valid and | |
1336 | * whether any values have changed. */ | |
1337 | for(i = 0; i < reg->hr_blocks; i++) { | |
1338 | slot = ®->hr_slots[i]; | |
1339 | hb_block = (struct o2hb_disk_heartbeat_block *) slot->ds_raw_block; | |
1340 | ||
1341 | /* Only fill the values that o2hb_check_slot uses to | |
1342 | * determine changing slots */ | |
1343 | slot->ds_last_time = le64_to_cpu(hb_block->hb_seq); | |
1344 | slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation); | |
1345 | } | |
1346 | ||
1347 | out: | |
1348 | mlog_exit(ret); | |
1349 | return ret; | |
1350 | } | |
1351 | ||
1352 | /* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */ | |
1353 | static ssize_t o2hb_region_dev_write(struct o2hb_region *reg, | |
1354 | const char *page, | |
1355 | size_t count) | |
1356 | { | |
e6c352db | 1357 | struct task_struct *hb_task; |
a7f6a5fb MF |
1358 | long fd; |
1359 | int sectsize; | |
1360 | char *p = (char *)page; | |
1361 | struct file *filp = NULL; | |
1362 | struct inode *inode = NULL; | |
1363 | ssize_t ret = -EINVAL; | |
1364 | ||
1365 | if (reg->hr_bdev) | |
1366 | goto out; | |
1367 | ||
1368 | /* We can't heartbeat without having had our node number | |
1369 | * configured yet. */ | |
1370 | if (o2nm_this_node() == O2NM_MAX_NODES) | |
1371 | goto out; | |
1372 | ||
1373 | fd = simple_strtol(p, &p, 0); | |
1374 | if (!p || (*p && (*p != '\n'))) | |
1375 | goto out; | |
1376 | ||
1377 | if (fd < 0 || fd >= INT_MAX) | |
1378 | goto out; | |
1379 | ||
1380 | filp = fget(fd); | |
1381 | if (filp == NULL) | |
1382 | goto out; | |
1383 | ||
1384 | if (reg->hr_blocks == 0 || reg->hr_start_block == 0 || | |
1385 | reg->hr_block_bytes == 0) | |
1386 | goto out; | |
1387 | ||
1388 | inode = igrab(filp->f_mapping->host); | |
1389 | if (inode == NULL) | |
1390 | goto out; | |
1391 | ||
1392 | if (!S_ISBLK(inode->i_mode)) | |
1393 | goto out; | |
1394 | ||
1395 | reg->hr_bdev = I_BDEV(filp->f_mapping->host); | |
572c4892 | 1396 | ret = blkdev_get(reg->hr_bdev, FMODE_WRITE | FMODE_READ); |
a7f6a5fb MF |
1397 | if (ret) { |
1398 | reg->hr_bdev = NULL; | |
1399 | goto out; | |
1400 | } | |
1401 | inode = NULL; | |
1402 | ||
1403 | bdevname(reg->hr_bdev, reg->hr_dev_name); | |
1404 | ||
e1defc4f | 1405 | sectsize = bdev_logical_block_size(reg->hr_bdev); |
a7f6a5fb MF |
1406 | if (sectsize != reg->hr_block_bytes) { |
1407 | mlog(ML_ERROR, | |
1408 | "blocksize %u incorrect for device, expected %d", | |
1409 | reg->hr_block_bytes, sectsize); | |
1410 | ret = -EINVAL; | |
1411 | goto out; | |
1412 | } | |
1413 | ||
1414 | o2hb_init_region_params(reg); | |
1415 | ||
1416 | /* Generation of zero is invalid */ | |
1417 | do { | |
1418 | get_random_bytes(®->hr_generation, | |
1419 | sizeof(reg->hr_generation)); | |
1420 | } while (reg->hr_generation == 0); | |
1421 | ||
1422 | ret = o2hb_map_slot_data(reg); | |
1423 | if (ret) { | |
1424 | mlog_errno(ret); | |
1425 | goto out; | |
1426 | } | |
1427 | ||
1428 | ret = o2hb_populate_slot_data(reg); | |
1429 | if (ret) { | |
1430 | mlog_errno(ret); | |
1431 | goto out; | |
1432 | } | |
1433 | ||
c4028958 | 1434 | INIT_DELAYED_WORK(®->hr_write_timeout_work, o2hb_write_timeout); |
a7f6a5fb MF |
1435 | |
1436 | /* | |
1437 | * A node is considered live after it has beat LIVE_THRESHOLD | |
1438 | * times. We're not steady until we've given them a chance | |
1439 | * _after_ our first read. | |
1440 | */ | |
1441 | atomic_set(®->hr_steady_iterations, O2HB_LIVE_THRESHOLD + 1); | |
1442 | ||
e6c352db JB |
1443 | hb_task = kthread_run(o2hb_thread, reg, "o2hb-%s", |
1444 | reg->hr_item.ci_name); | |
1445 | if (IS_ERR(hb_task)) { | |
1446 | ret = PTR_ERR(hb_task); | |
a7f6a5fb | 1447 | mlog_errno(ret); |
a7f6a5fb MF |
1448 | goto out; |
1449 | } | |
1450 | ||
e6c352db JB |
1451 | spin_lock(&o2hb_live_lock); |
1452 | reg->hr_task = hb_task; | |
1453 | spin_unlock(&o2hb_live_lock); | |
1454 | ||
a7f6a5fb MF |
1455 | ret = wait_event_interruptible(o2hb_steady_queue, |
1456 | atomic_read(®->hr_steady_iterations) == 0); | |
1457 | if (ret) { | |
e6df3a66 | 1458 | /* We got interrupted (hello ptrace!). Clean up */ |
e6c352db JB |
1459 | spin_lock(&o2hb_live_lock); |
1460 | hb_task = reg->hr_task; | |
a7f6a5fb | 1461 | reg->hr_task = NULL; |
e6c352db JB |
1462 | spin_unlock(&o2hb_live_lock); |
1463 | ||
1464 | if (hb_task) | |
1465 | kthread_stop(hb_task); | |
a7f6a5fb MF |
1466 | goto out; |
1467 | } | |
1468 | ||
e6df3a66 JB |
1469 | /* Ok, we were woken. Make sure it wasn't by drop_item() */ |
1470 | spin_lock(&o2hb_live_lock); | |
1471 | hb_task = reg->hr_task; | |
1472 | spin_unlock(&o2hb_live_lock); | |
1473 | ||
1474 | if (hb_task) | |
1475 | ret = count; | |
1476 | else | |
1477 | ret = -EIO; | |
1478 | ||
a7f6a5fb MF |
1479 | out: |
1480 | if (filp) | |
1481 | fput(filp); | |
1482 | if (inode) | |
1483 | iput(inode); | |
1484 | if (ret < 0) { | |
1485 | if (reg->hr_bdev) { | |
9a1c3542 | 1486 | blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE); |
a7f6a5fb MF |
1487 | reg->hr_bdev = NULL; |
1488 | } | |
1489 | } | |
1490 | return ret; | |
1491 | } | |
1492 | ||
92efc152 ZW |
1493 | static ssize_t o2hb_region_pid_read(struct o2hb_region *reg, |
1494 | char *page) | |
1495 | { | |
e6c352db JB |
1496 | pid_t pid = 0; |
1497 | ||
1498 | spin_lock(&o2hb_live_lock); | |
1499 | if (reg->hr_task) | |
ba25f9dc | 1500 | pid = task_pid_nr(reg->hr_task); |
e6c352db JB |
1501 | spin_unlock(&o2hb_live_lock); |
1502 | ||
1503 | if (!pid) | |
92efc152 ZW |
1504 | return 0; |
1505 | ||
e6c352db | 1506 | return sprintf(page, "%u\n", pid); |
92efc152 ZW |
1507 | } |
1508 | ||
a7f6a5fb MF |
1509 | struct o2hb_region_attribute { |
1510 | struct configfs_attribute attr; | |
1511 | ssize_t (*show)(struct o2hb_region *, char *); | |
1512 | ssize_t (*store)(struct o2hb_region *, const char *, size_t); | |
1513 | }; | |
1514 | ||
1515 | static struct o2hb_region_attribute o2hb_region_attr_block_bytes = { | |
1516 | .attr = { .ca_owner = THIS_MODULE, | |
1517 | .ca_name = "block_bytes", | |
1518 | .ca_mode = S_IRUGO | S_IWUSR }, | |
1519 | .show = o2hb_region_block_bytes_read, | |
1520 | .store = o2hb_region_block_bytes_write, | |
1521 | }; | |
1522 | ||
1523 | static struct o2hb_region_attribute o2hb_region_attr_start_block = { | |
1524 | .attr = { .ca_owner = THIS_MODULE, | |
1525 | .ca_name = "start_block", | |
1526 | .ca_mode = S_IRUGO | S_IWUSR }, | |
1527 | .show = o2hb_region_start_block_read, | |
1528 | .store = o2hb_region_start_block_write, | |
1529 | }; | |
1530 | ||
1531 | static struct o2hb_region_attribute o2hb_region_attr_blocks = { | |
1532 | .attr = { .ca_owner = THIS_MODULE, | |
1533 | .ca_name = "blocks", | |
1534 | .ca_mode = S_IRUGO | S_IWUSR }, | |
1535 | .show = o2hb_region_blocks_read, | |
1536 | .store = o2hb_region_blocks_write, | |
1537 | }; | |
1538 | ||
1539 | static struct o2hb_region_attribute o2hb_region_attr_dev = { | |
1540 | .attr = { .ca_owner = THIS_MODULE, | |
1541 | .ca_name = "dev", | |
1542 | .ca_mode = S_IRUGO | S_IWUSR }, | |
1543 | .show = o2hb_region_dev_read, | |
1544 | .store = o2hb_region_dev_write, | |
1545 | }; | |
1546 | ||
92efc152 ZW |
1547 | static struct o2hb_region_attribute o2hb_region_attr_pid = { |
1548 | .attr = { .ca_owner = THIS_MODULE, | |
1549 | .ca_name = "pid", | |
1550 | .ca_mode = S_IRUGO | S_IRUSR }, | |
1551 | .show = o2hb_region_pid_read, | |
1552 | }; | |
1553 | ||
a7f6a5fb MF |
1554 | static struct configfs_attribute *o2hb_region_attrs[] = { |
1555 | &o2hb_region_attr_block_bytes.attr, | |
1556 | &o2hb_region_attr_start_block.attr, | |
1557 | &o2hb_region_attr_blocks.attr, | |
1558 | &o2hb_region_attr_dev.attr, | |
92efc152 | 1559 | &o2hb_region_attr_pid.attr, |
a7f6a5fb MF |
1560 | NULL, |
1561 | }; | |
1562 | ||
1563 | static ssize_t o2hb_region_show(struct config_item *item, | |
1564 | struct configfs_attribute *attr, | |
1565 | char *page) | |
1566 | { | |
1567 | struct o2hb_region *reg = to_o2hb_region(item); | |
1568 | struct o2hb_region_attribute *o2hb_region_attr = | |
1569 | container_of(attr, struct o2hb_region_attribute, attr); | |
1570 | ssize_t ret = 0; | |
1571 | ||
1572 | if (o2hb_region_attr->show) | |
1573 | ret = o2hb_region_attr->show(reg, page); | |
1574 | return ret; | |
1575 | } | |
1576 | ||
1577 | static ssize_t o2hb_region_store(struct config_item *item, | |
1578 | struct configfs_attribute *attr, | |
1579 | const char *page, size_t count) | |
1580 | { | |
1581 | struct o2hb_region *reg = to_o2hb_region(item); | |
1582 | struct o2hb_region_attribute *o2hb_region_attr = | |
1583 | container_of(attr, struct o2hb_region_attribute, attr); | |
1584 | ssize_t ret = -EINVAL; | |
1585 | ||
1586 | if (o2hb_region_attr->store) | |
1587 | ret = o2hb_region_attr->store(reg, page, count); | |
1588 | return ret; | |
1589 | } | |
1590 | ||
1591 | static struct configfs_item_operations o2hb_region_item_ops = { | |
1592 | .release = o2hb_region_release, | |
1593 | .show_attribute = o2hb_region_show, | |
1594 | .store_attribute = o2hb_region_store, | |
1595 | }; | |
1596 | ||
1597 | static struct config_item_type o2hb_region_type = { | |
1598 | .ct_item_ops = &o2hb_region_item_ops, | |
1599 | .ct_attrs = o2hb_region_attrs, | |
1600 | .ct_owner = THIS_MODULE, | |
1601 | }; | |
1602 | ||
1603 | /* heartbeat set */ | |
1604 | ||
1605 | struct o2hb_heartbeat_group { | |
1606 | struct config_group hs_group; | |
1607 | /* some stuff? */ | |
1608 | }; | |
1609 | ||
1610 | static struct o2hb_heartbeat_group *to_o2hb_heartbeat_group(struct config_group *group) | |
1611 | { | |
1612 | return group ? | |
1613 | container_of(group, struct o2hb_heartbeat_group, hs_group) | |
1614 | : NULL; | |
1615 | } | |
1616 | ||
f89ab861 JB |
1617 | static struct config_item *o2hb_heartbeat_group_make_item(struct config_group *group, |
1618 | const char *name) | |
a7f6a5fb MF |
1619 | { |
1620 | struct o2hb_region *reg = NULL; | |
a7f6a5fb | 1621 | |
cd861280 | 1622 | reg = kzalloc(sizeof(struct o2hb_region), GFP_KERNEL); |
f89ab861 | 1623 | if (reg == NULL) |
a6795e9e | 1624 | return ERR_PTR(-ENOMEM); |
a7f6a5fb MF |
1625 | |
1626 | config_item_init_type_name(®->hr_item, name, &o2hb_region_type); | |
1627 | ||
a7f6a5fb MF |
1628 | spin_lock(&o2hb_live_lock); |
1629 | list_add_tail(®->hr_all_item, &o2hb_all_regions); | |
1630 | spin_unlock(&o2hb_live_lock); | |
a7f6a5fb | 1631 | |
a6795e9e | 1632 | return ®->hr_item; |
a7f6a5fb MF |
1633 | } |
1634 | ||
1635 | static void o2hb_heartbeat_group_drop_item(struct config_group *group, | |
1636 | struct config_item *item) | |
1637 | { | |
e6c352db | 1638 | struct task_struct *hb_task; |
a7f6a5fb MF |
1639 | struct o2hb_region *reg = to_o2hb_region(item); |
1640 | ||
1641 | /* stop the thread when the user removes the region dir */ | |
e6c352db JB |
1642 | spin_lock(&o2hb_live_lock); |
1643 | hb_task = reg->hr_task; | |
1644 | reg->hr_task = NULL; | |
1645 | spin_unlock(&o2hb_live_lock); | |
1646 | ||
1647 | if (hb_task) | |
1648 | kthread_stop(hb_task); | |
a7f6a5fb | 1649 | |
e6df3a66 JB |
1650 | /* |
1651 | * If we're racing a dev_write(), we need to wake them. They will | |
1652 | * check reg->hr_task | |
1653 | */ | |
1654 | if (atomic_read(®->hr_steady_iterations) != 0) { | |
1655 | atomic_set(®->hr_steady_iterations, 0); | |
1656 | wake_up(&o2hb_steady_queue); | |
1657 | } | |
1658 | ||
a7f6a5fb MF |
1659 | config_item_put(item); |
1660 | } | |
1661 | ||
1662 | struct o2hb_heartbeat_group_attribute { | |
1663 | struct configfs_attribute attr; | |
1664 | ssize_t (*show)(struct o2hb_heartbeat_group *, char *); | |
1665 | ssize_t (*store)(struct o2hb_heartbeat_group *, const char *, size_t); | |
1666 | }; | |
1667 | ||
1668 | static ssize_t o2hb_heartbeat_group_show(struct config_item *item, | |
1669 | struct configfs_attribute *attr, | |
1670 | char *page) | |
1671 | { | |
1672 | struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item)); | |
1673 | struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr = | |
1674 | container_of(attr, struct o2hb_heartbeat_group_attribute, attr); | |
1675 | ssize_t ret = 0; | |
1676 | ||
1677 | if (o2hb_heartbeat_group_attr->show) | |
1678 | ret = o2hb_heartbeat_group_attr->show(reg, page); | |
1679 | return ret; | |
1680 | } | |
1681 | ||
1682 | static ssize_t o2hb_heartbeat_group_store(struct config_item *item, | |
1683 | struct configfs_attribute *attr, | |
1684 | const char *page, size_t count) | |
1685 | { | |
1686 | struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item)); | |
1687 | struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr = | |
1688 | container_of(attr, struct o2hb_heartbeat_group_attribute, attr); | |
1689 | ssize_t ret = -EINVAL; | |
1690 | ||
1691 | if (o2hb_heartbeat_group_attr->store) | |
1692 | ret = o2hb_heartbeat_group_attr->store(reg, page, count); | |
1693 | return ret; | |
1694 | } | |
1695 | ||
1696 | static ssize_t o2hb_heartbeat_group_threshold_show(struct o2hb_heartbeat_group *group, | |
1697 | char *page) | |
1698 | { | |
1699 | return sprintf(page, "%u\n", o2hb_dead_threshold); | |
1700 | } | |
1701 | ||
1702 | static ssize_t o2hb_heartbeat_group_threshold_store(struct o2hb_heartbeat_group *group, | |
1703 | const char *page, | |
1704 | size_t count) | |
1705 | { | |
1706 | unsigned long tmp; | |
1707 | char *p = (char *)page; | |
1708 | ||
1709 | tmp = simple_strtoul(p, &p, 10); | |
1710 | if (!p || (*p && (*p != '\n'))) | |
1711 | return -EINVAL; | |
1712 | ||
1713 | /* this will validate ranges for us. */ | |
1714 | o2hb_dead_threshold_set((unsigned int) tmp); | |
1715 | ||
1716 | return count; | |
1717 | } | |
1718 | ||
54b5187b SM |
1719 | static |
1720 | ssize_t o2hb_heartbeat_group_mode_show(struct o2hb_heartbeat_group *group, | |
1721 | char *page) | |
1722 | { | |
1723 | return sprintf(page, "%s\n", | |
1724 | o2hb_heartbeat_mode_desc[o2hb_heartbeat_mode]); | |
1725 | } | |
1726 | ||
1727 | static | |
1728 | ssize_t o2hb_heartbeat_group_mode_store(struct o2hb_heartbeat_group *group, | |
1729 | const char *page, size_t count) | |
1730 | { | |
1731 | unsigned int i; | |
1732 | int ret; | |
1733 | size_t len; | |
1734 | ||
1735 | len = (page[count - 1] == '\n') ? count - 1 : count; | |
1736 | if (!len) | |
1737 | return -EINVAL; | |
1738 | ||
1739 | for (i = 0; i < O2HB_HEARTBEAT_NUM_MODES; ++i) { | |
1740 | if (strnicmp(page, o2hb_heartbeat_mode_desc[i], len)) | |
1741 | continue; | |
1742 | ||
1743 | ret = o2hb_global_hearbeat_mode_set(i); | |
1744 | if (!ret) | |
1745 | printk(KERN_NOTICE "ocfs2: Heartbeat mode set to %s\n", | |
1746 | o2hb_heartbeat_mode_desc[i]); | |
1747 | return count; | |
1748 | } | |
1749 | ||
1750 | return -EINVAL; | |
1751 | ||
1752 | } | |
1753 | ||
a7f6a5fb MF |
1754 | static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_threshold = { |
1755 | .attr = { .ca_owner = THIS_MODULE, | |
1756 | .ca_name = "dead_threshold", | |
1757 | .ca_mode = S_IRUGO | S_IWUSR }, | |
1758 | .show = o2hb_heartbeat_group_threshold_show, | |
1759 | .store = o2hb_heartbeat_group_threshold_store, | |
1760 | }; | |
1761 | ||
54b5187b SM |
1762 | static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_mode = { |
1763 | .attr = { .ca_owner = THIS_MODULE, | |
1764 | .ca_name = "mode", | |
1765 | .ca_mode = S_IRUGO | S_IWUSR }, | |
1766 | .show = o2hb_heartbeat_group_mode_show, | |
1767 | .store = o2hb_heartbeat_group_mode_store, | |
1768 | }; | |
1769 | ||
a7f6a5fb MF |
1770 | static struct configfs_attribute *o2hb_heartbeat_group_attrs[] = { |
1771 | &o2hb_heartbeat_group_attr_threshold.attr, | |
54b5187b | 1772 | &o2hb_heartbeat_group_attr_mode.attr, |
a7f6a5fb MF |
1773 | NULL, |
1774 | }; | |
1775 | ||
1776 | static struct configfs_item_operations o2hb_hearbeat_group_item_ops = { | |
1777 | .show_attribute = o2hb_heartbeat_group_show, | |
1778 | .store_attribute = o2hb_heartbeat_group_store, | |
1779 | }; | |
1780 | ||
1781 | static struct configfs_group_operations o2hb_heartbeat_group_group_ops = { | |
1782 | .make_item = o2hb_heartbeat_group_make_item, | |
1783 | .drop_item = o2hb_heartbeat_group_drop_item, | |
1784 | }; | |
1785 | ||
1786 | static struct config_item_type o2hb_heartbeat_group_type = { | |
1787 | .ct_group_ops = &o2hb_heartbeat_group_group_ops, | |
1788 | .ct_item_ops = &o2hb_hearbeat_group_item_ops, | |
1789 | .ct_attrs = o2hb_heartbeat_group_attrs, | |
1790 | .ct_owner = THIS_MODULE, | |
1791 | }; | |
1792 | ||
1793 | /* this is just here to avoid touching group in heartbeat.h which the | |
1794 | * entire damn world #includes */ | |
1795 | struct config_group *o2hb_alloc_hb_set(void) | |
1796 | { | |
1797 | struct o2hb_heartbeat_group *hs = NULL; | |
1798 | struct config_group *ret = NULL; | |
1799 | ||
cd861280 | 1800 | hs = kzalloc(sizeof(struct o2hb_heartbeat_group), GFP_KERNEL); |
a7f6a5fb MF |
1801 | if (hs == NULL) |
1802 | goto out; | |
1803 | ||
1804 | config_group_init_type_name(&hs->hs_group, "heartbeat", | |
1805 | &o2hb_heartbeat_group_type); | |
1806 | ||
1807 | ret = &hs->hs_group; | |
1808 | out: | |
1809 | if (ret == NULL) | |
1810 | kfree(hs); | |
1811 | return ret; | |
1812 | } | |
1813 | ||
1814 | void o2hb_free_hb_set(struct config_group *group) | |
1815 | { | |
1816 | struct o2hb_heartbeat_group *hs = to_o2hb_heartbeat_group(group); | |
1817 | kfree(hs); | |
1818 | } | |
1819 | ||
1820 | /* hb callback registration and issueing */ | |
1821 | ||
1822 | static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type) | |
1823 | { | |
1824 | if (type == O2HB_NUM_CB) | |
1825 | return ERR_PTR(-EINVAL); | |
1826 | ||
1827 | return &o2hb_callbacks[type]; | |
1828 | } | |
1829 | ||
1830 | void o2hb_setup_callback(struct o2hb_callback_func *hc, | |
1831 | enum o2hb_callback_type type, | |
1832 | o2hb_cb_func *func, | |
1833 | void *data, | |
1834 | int priority) | |
1835 | { | |
1836 | INIT_LIST_HEAD(&hc->hc_item); | |
1837 | hc->hc_func = func; | |
1838 | hc->hc_data = data; | |
1839 | hc->hc_priority = priority; | |
1840 | hc->hc_type = type; | |
1841 | hc->hc_magic = O2HB_CB_MAGIC; | |
1842 | } | |
1843 | EXPORT_SYMBOL_GPL(o2hb_setup_callback); | |
1844 | ||
14829422 JB |
1845 | static struct o2hb_region *o2hb_find_region(const char *region_uuid) |
1846 | { | |
1847 | struct o2hb_region *p, *reg = NULL; | |
1848 | ||
1849 | assert_spin_locked(&o2hb_live_lock); | |
1850 | ||
1851 | list_for_each_entry(p, &o2hb_all_regions, hr_all_item) { | |
1852 | if (!strcmp(region_uuid, config_item_name(&p->hr_item))) { | |
1853 | reg = p; | |
1854 | break; | |
1855 | } | |
1856 | } | |
1857 | ||
1858 | return reg; | |
1859 | } | |
1860 | ||
1861 | static int o2hb_region_get(const char *region_uuid) | |
1862 | { | |
1863 | int ret = 0; | |
1864 | struct o2hb_region *reg; | |
1865 | ||
1866 | spin_lock(&o2hb_live_lock); | |
1867 | ||
1868 | reg = o2hb_find_region(region_uuid); | |
1869 | if (!reg) | |
1870 | ret = -ENOENT; | |
1871 | spin_unlock(&o2hb_live_lock); | |
1872 | ||
16c6a4f2 JB |
1873 | if (ret) |
1874 | goto out; | |
1875 | ||
1876 | ret = o2nm_depend_this_node(); | |
1877 | if (ret) | |
1878 | goto out; | |
14829422 | 1879 | |
16c6a4f2 JB |
1880 | ret = o2nm_depend_item(®->hr_item); |
1881 | if (ret) | |
1882 | o2nm_undepend_this_node(); | |
1883 | ||
1884 | out: | |
14829422 JB |
1885 | return ret; |
1886 | } | |
1887 | ||
1888 | static void o2hb_region_put(const char *region_uuid) | |
1889 | { | |
1890 | struct o2hb_region *reg; | |
1891 | ||
1892 | spin_lock(&o2hb_live_lock); | |
1893 | ||
1894 | reg = o2hb_find_region(region_uuid); | |
1895 | ||
1896 | spin_unlock(&o2hb_live_lock); | |
1897 | ||
16c6a4f2 | 1898 | if (reg) { |
14829422 | 1899 | o2nm_undepend_item(®->hr_item); |
16c6a4f2 JB |
1900 | o2nm_undepend_this_node(); |
1901 | } | |
14829422 JB |
1902 | } |
1903 | ||
1904 | int o2hb_register_callback(const char *region_uuid, | |
1905 | struct o2hb_callback_func *hc) | |
a7f6a5fb MF |
1906 | { |
1907 | struct o2hb_callback_func *tmp; | |
1908 | struct list_head *iter; | |
1909 | struct o2hb_callback *hbcall; | |
1910 | int ret; | |
1911 | ||
1912 | BUG_ON(hc->hc_magic != O2HB_CB_MAGIC); | |
1913 | BUG_ON(!list_empty(&hc->hc_item)); | |
1914 | ||
1915 | hbcall = hbcall_from_type(hc->hc_type); | |
1916 | if (IS_ERR(hbcall)) { | |
1917 | ret = PTR_ERR(hbcall); | |
1918 | goto out; | |
1919 | } | |
1920 | ||
14829422 JB |
1921 | if (region_uuid) { |
1922 | ret = o2hb_region_get(region_uuid); | |
1923 | if (ret) | |
1924 | goto out; | |
1925 | } | |
1926 | ||
a7f6a5fb MF |
1927 | down_write(&o2hb_callback_sem); |
1928 | ||
1929 | list_for_each(iter, &hbcall->list) { | |
1930 | tmp = list_entry(iter, struct o2hb_callback_func, hc_item); | |
1931 | if (hc->hc_priority < tmp->hc_priority) { | |
1932 | list_add_tail(&hc->hc_item, iter); | |
1933 | break; | |
1934 | } | |
1935 | } | |
1936 | if (list_empty(&hc->hc_item)) | |
1937 | list_add_tail(&hc->hc_item, &hbcall->list); | |
1938 | ||
1939 | up_write(&o2hb_callback_sem); | |
1940 | ret = 0; | |
1941 | out: | |
1942 | mlog(ML_HEARTBEAT, "returning %d on behalf of %p for funcs %p\n", | |
1943 | ret, __builtin_return_address(0), hc); | |
1944 | return ret; | |
1945 | } | |
1946 | EXPORT_SYMBOL_GPL(o2hb_register_callback); | |
1947 | ||
14829422 JB |
1948 | void o2hb_unregister_callback(const char *region_uuid, |
1949 | struct o2hb_callback_func *hc) | |
a7f6a5fb MF |
1950 | { |
1951 | BUG_ON(hc->hc_magic != O2HB_CB_MAGIC); | |
1952 | ||
1953 | mlog(ML_HEARTBEAT, "on behalf of %p for funcs %p\n", | |
1954 | __builtin_return_address(0), hc); | |
1955 | ||
14829422 | 1956 | /* XXX Can this happen _with_ a region reference? */ |
a7f6a5fb | 1957 | if (list_empty(&hc->hc_item)) |
c24f72cc | 1958 | return; |
a7f6a5fb | 1959 | |
14829422 JB |
1960 | if (region_uuid) |
1961 | o2hb_region_put(region_uuid); | |
1962 | ||
a7f6a5fb MF |
1963 | down_write(&o2hb_callback_sem); |
1964 | ||
1965 | list_del_init(&hc->hc_item); | |
1966 | ||
1967 | up_write(&o2hb_callback_sem); | |
a7f6a5fb MF |
1968 | } |
1969 | EXPORT_SYMBOL_GPL(o2hb_unregister_callback); | |
1970 | ||
1971 | int o2hb_check_node_heartbeating(u8 node_num) | |
1972 | { | |
1973 | unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; | |
1974 | ||
1975 | o2hb_fill_node_map(testing_map, sizeof(testing_map)); | |
1976 | if (!test_bit(node_num, testing_map)) { | |
1977 | mlog(ML_HEARTBEAT, | |
1978 | "node (%u) does not have heartbeating enabled.\n", | |
1979 | node_num); | |
1980 | return 0; | |
1981 | } | |
1982 | ||
1983 | return 1; | |
1984 | } | |
1985 | EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating); | |
1986 | ||
1987 | int o2hb_check_node_heartbeating_from_callback(u8 node_num) | |
1988 | { | |
1989 | unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; | |
1990 | ||
1991 | o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map)); | |
1992 | if (!test_bit(node_num, testing_map)) { | |
1993 | mlog(ML_HEARTBEAT, | |
1994 | "node (%u) does not have heartbeating enabled.\n", | |
1995 | node_num); | |
1996 | return 0; | |
1997 | } | |
1998 | ||
1999 | return 1; | |
2000 | } | |
2001 | EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback); | |
2002 | ||
2003 | /* Makes sure our local node is configured with a node number, and is | |
2004 | * heartbeating. */ | |
2005 | int o2hb_check_local_node_heartbeating(void) | |
2006 | { | |
2007 | u8 node_num; | |
2008 | ||
2009 | /* if this node was set then we have networking */ | |
2010 | node_num = o2nm_this_node(); | |
2011 | if (node_num == O2NM_MAX_NODES) { | |
2012 | mlog(ML_HEARTBEAT, "this node has not been configured.\n"); | |
2013 | return 0; | |
2014 | } | |
2015 | ||
2016 | return o2hb_check_node_heartbeating(node_num); | |
2017 | } | |
2018 | EXPORT_SYMBOL_GPL(o2hb_check_local_node_heartbeating); | |
2019 | ||
2020 | /* | |
2021 | * this is just a hack until we get the plumbing which flips file systems | |
2022 | * read only and drops the hb ref instead of killing the node dead. | |
2023 | */ | |
2024 | void o2hb_stop_all_regions(void) | |
2025 | { | |
2026 | struct o2hb_region *reg; | |
2027 | ||
2028 | mlog(ML_ERROR, "stopping heartbeat on all active regions.\n"); | |
2029 | ||
2030 | spin_lock(&o2hb_live_lock); | |
2031 | ||
2032 | list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) | |
2033 | reg->hr_unclean_stop = 1; | |
2034 | ||
2035 | spin_unlock(&o2hb_live_lock); | |
2036 | } | |
2037 | EXPORT_SYMBOL_GPL(o2hb_stop_all_regions); |