Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 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 | */ |
Ingo Molnar | 34af946 | 2006-06-27 02:53:55 -0700 | [diff] [blame] | 57 | static DEFINE_SPINLOCK(o2hb_live_lock); |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 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 | |
David Howells | c402895 | 2006-11-22 14:57:56 +0000 | [diff] [blame] | 144 | struct delayed_work hr_write_timeout_work; |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 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; |
Mark Fasheh | a9e2ae3 | 2006-03-24 14:20:17 -0800 | [diff] [blame] | 156 | int wc_error; |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 157 | }; |
| 158 | |
David Howells | c402895 | 2006-11-22 14:57:56 +0000 | [diff] [blame] | 159 | static void o2hb_write_timeout(struct work_struct *work) |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 160 | { |
David Howells | c402895 | 2006-11-22 14:57:56 +0000 | [diff] [blame] | 161 | struct o2hb_region *reg = |
| 162 | container_of(work, struct o2hb_region, |
| 163 | hr_write_timeout_work.work); |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 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 | |
Philipp Reisner | b559292 | 2007-01-11 10:58:10 +0100 | [diff] [blame] | 187 | static inline void o2hb_bio_wait_init(struct o2hb_bio_wait_ctxt *wc) |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 188 | { |
Philipp Reisner | b559292 | 2007-01-11 10:58:10 +0100 | [diff] [blame] | 189 | atomic_set(&wc->wc_num_reqs, 1); |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 190 | init_completion(&wc->wc_io_complete); |
Mark Fasheh | a9e2ae3 | 2006-03-24 14:20:17 -0800 | [diff] [blame] | 191 | wc->wc_error = 0; |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 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); |
Philipp Reisner | b559292 | 2007-01-11 10:58:10 +0100 | [diff] [blame] | 214 | o2hb_bio_wait_dec(wc, 1); |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 215 | |
| 216 | wait_for_completion(&wc->wc_io_complete); |
| 217 | } |
| 218 | |
| 219 | static int o2hb_bio_end_io(struct bio *bio, |
| 220 | unsigned int bytes_done, |
| 221 | int error) |
| 222 | { |
| 223 | struct o2hb_bio_wait_ctxt *wc = bio->bi_private; |
| 224 | |
Mark Fasheh | a9e2ae3 | 2006-03-24 14:20:17 -0800 | [diff] [blame] | 225 | if (error) { |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 226 | mlog(ML_ERROR, "IO Error %d\n", error); |
Mark Fasheh | a9e2ae3 | 2006-03-24 14:20:17 -0800 | [diff] [blame] | 227 | wc->wc_error = error; |
| 228 | } |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 229 | |
| 230 | if (bio->bi_size) |
| 231 | return 1; |
| 232 | |
| 233 | o2hb_bio_wait_dec(wc, 1); |
Philipp Reisner | b559292 | 2007-01-11 10:58:10 +0100 | [diff] [blame] | 234 | bio_put(bio); |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 235 | return 0; |
| 236 | } |
| 237 | |
| 238 | /* Setup a Bio to cover I/O against num_slots slots starting at |
| 239 | * start_slot. */ |
| 240 | static struct bio *o2hb_setup_one_bio(struct o2hb_region *reg, |
| 241 | struct o2hb_bio_wait_ctxt *wc, |
Philipp Reisner | b559292 | 2007-01-11 10:58:10 +0100 | [diff] [blame] | 242 | unsigned int *current_slot, |
| 243 | unsigned int max_slots) |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 244 | { |
Philipp Reisner | b559292 | 2007-01-11 10:58:10 +0100 | [diff] [blame] | 245 | int len, current_page; |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 246 | unsigned int vec_len, vec_start; |
| 247 | unsigned int bits = reg->hr_block_bits; |
| 248 | unsigned int spp = reg->hr_slots_per_page; |
Philipp Reisner | b559292 | 2007-01-11 10:58:10 +0100 | [diff] [blame] | 249 | unsigned int cs = *current_slot; |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 250 | struct bio *bio; |
| 251 | struct page *page; |
| 252 | |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 253 | /* Testing has shown this allocation to take long enough under |
| 254 | * GFP_KERNEL that the local node can get fenced. It would be |
| 255 | * nicest if we could pre-allocate these bios and avoid this |
| 256 | * all together. */ |
Philipp Reisner | b559292 | 2007-01-11 10:58:10 +0100 | [diff] [blame] | 257 | bio = bio_alloc(GFP_ATOMIC, 16); |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 258 | if (!bio) { |
| 259 | mlog(ML_ERROR, "Could not alloc slots BIO!\n"); |
| 260 | bio = ERR_PTR(-ENOMEM); |
| 261 | goto bail; |
| 262 | } |
| 263 | |
| 264 | /* Must put everything in 512 byte sectors for the bio... */ |
Philipp Reisner | b559292 | 2007-01-11 10:58:10 +0100 | [diff] [blame] | 265 | bio->bi_sector = (reg->hr_start_block + cs) << (bits - 9); |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 266 | bio->bi_bdev = reg->hr_bdev; |
| 267 | bio->bi_private = wc; |
| 268 | bio->bi_end_io = o2hb_bio_end_io; |
| 269 | |
Philipp Reisner | b559292 | 2007-01-11 10:58:10 +0100 | [diff] [blame] | 270 | vec_start = (cs << bits) % PAGE_CACHE_SIZE; |
| 271 | while(cs < max_slots) { |
| 272 | current_page = cs / spp; |
| 273 | page = reg->hr_slot_data[current_page]; |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 274 | |
Philipp Reisner | b559292 | 2007-01-11 10:58:10 +0100 | [diff] [blame] | 275 | vec_len = min(PAGE_CACHE_SIZE, |
| 276 | (max_slots-cs) * (PAGE_CACHE_SIZE/spp) ); |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 277 | |
| 278 | mlog(ML_HB_BIO, "page %d, vec_len = %u, vec_start = %u\n", |
Philipp Reisner | b559292 | 2007-01-11 10:58:10 +0100 | [diff] [blame] | 279 | current_page, vec_len, vec_start); |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 280 | |
| 281 | len = bio_add_page(bio, page, vec_len, vec_start); |
Philipp Reisner | b559292 | 2007-01-11 10:58:10 +0100 | [diff] [blame] | 282 | if (len != vec_len) break; |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 283 | |
Philipp Reisner | b559292 | 2007-01-11 10:58:10 +0100 | [diff] [blame] | 284 | cs += vec_len / (PAGE_CACHE_SIZE/spp); |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 285 | vec_start = 0; |
| 286 | } |
| 287 | |
| 288 | bail: |
Philipp Reisner | b559292 | 2007-01-11 10:58:10 +0100 | [diff] [blame] | 289 | *current_slot = cs; |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 290 | return bio; |
| 291 | } |
| 292 | |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 293 | static int o2hb_read_slots(struct o2hb_region *reg, |
| 294 | unsigned int max_slots) |
| 295 | { |
Philipp Reisner | b559292 | 2007-01-11 10:58:10 +0100 | [diff] [blame] | 296 | unsigned int current_slot=0; |
| 297 | int status; |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 298 | struct o2hb_bio_wait_ctxt wc; |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 299 | struct bio *bio; |
| 300 | |
Philipp Reisner | b559292 | 2007-01-11 10:58:10 +0100 | [diff] [blame] | 301 | o2hb_bio_wait_init(&wc); |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 302 | |
Philipp Reisner | b559292 | 2007-01-11 10:58:10 +0100 | [diff] [blame] | 303 | while(current_slot < max_slots) { |
| 304 | bio = o2hb_setup_one_bio(reg, &wc, ¤t_slot, max_slots); |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 305 | if (IS_ERR(bio)) { |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 306 | status = PTR_ERR(bio); |
| 307 | mlog_errno(status); |
| 308 | goto bail_and_wait; |
| 309 | } |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 310 | |
Philipp Reisner | b559292 | 2007-01-11 10:58:10 +0100 | [diff] [blame] | 311 | atomic_inc(&wc.wc_num_reqs); |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 312 | submit_bio(READ, bio); |
| 313 | } |
| 314 | |
| 315 | status = 0; |
| 316 | |
| 317 | bail_and_wait: |
| 318 | o2hb_wait_on_io(reg, &wc); |
Mark Fasheh | a9e2ae3 | 2006-03-24 14:20:17 -0800 | [diff] [blame] | 319 | if (wc.wc_error && !status) |
| 320 | status = wc.wc_error; |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 321 | |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 322 | return status; |
| 323 | } |
| 324 | |
| 325 | static int o2hb_issue_node_write(struct o2hb_region *reg, |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 326 | struct o2hb_bio_wait_ctxt *write_wc) |
| 327 | { |
| 328 | int status; |
| 329 | unsigned int slot; |
| 330 | struct bio *bio; |
| 331 | |
Philipp Reisner | b559292 | 2007-01-11 10:58:10 +0100 | [diff] [blame] | 332 | o2hb_bio_wait_init(write_wc); |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 333 | |
| 334 | slot = o2nm_this_node(); |
| 335 | |
Philipp Reisner | b559292 | 2007-01-11 10:58:10 +0100 | [diff] [blame] | 336 | bio = o2hb_setup_one_bio(reg, write_wc, &slot, slot+1); |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 337 | if (IS_ERR(bio)) { |
| 338 | status = PTR_ERR(bio); |
| 339 | mlog_errno(status); |
| 340 | goto bail; |
| 341 | } |
| 342 | |
Philipp Reisner | b559292 | 2007-01-11 10:58:10 +0100 | [diff] [blame] | 343 | atomic_inc(&write_wc->wc_num_reqs); |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 344 | submit_bio(WRITE, bio); |
| 345 | |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 346 | status = 0; |
| 347 | bail: |
| 348 | return status; |
| 349 | } |
| 350 | |
| 351 | static u32 o2hb_compute_block_crc_le(struct o2hb_region *reg, |
| 352 | struct o2hb_disk_heartbeat_block *hb_block) |
| 353 | { |
| 354 | __le32 old_cksum; |
| 355 | u32 ret; |
| 356 | |
| 357 | /* We want to compute the block crc with a 0 value in the |
| 358 | * hb_cksum field. Save it off here and replace after the |
| 359 | * crc. */ |
| 360 | old_cksum = hb_block->hb_cksum; |
| 361 | hb_block->hb_cksum = 0; |
| 362 | |
| 363 | ret = crc32_le(0, (unsigned char *) hb_block, reg->hr_block_bytes); |
| 364 | |
| 365 | hb_block->hb_cksum = old_cksum; |
| 366 | |
| 367 | return ret; |
| 368 | } |
| 369 | |
| 370 | static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block *hb_block) |
| 371 | { |
Mark Fasheh | 70bacbdb | 2006-03-02 11:10:05 -0800 | [diff] [blame] | 372 | mlog(ML_ERROR, "Dump slot information: seq = 0x%llx, node = %u, " |
| 373 | "cksum = 0x%x, generation 0x%llx\n", |
| 374 | (long long)le64_to_cpu(hb_block->hb_seq), |
| 375 | hb_block->hb_node, le32_to_cpu(hb_block->hb_cksum), |
| 376 | (long long)le64_to_cpu(hb_block->hb_generation)); |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 377 | } |
| 378 | |
| 379 | static int o2hb_verify_crc(struct o2hb_region *reg, |
| 380 | struct o2hb_disk_heartbeat_block *hb_block) |
| 381 | { |
| 382 | u32 read, computed; |
| 383 | |
| 384 | read = le32_to_cpu(hb_block->hb_cksum); |
| 385 | computed = o2hb_compute_block_crc_le(reg, hb_block); |
| 386 | |
| 387 | return read == computed; |
| 388 | } |
| 389 | |
| 390 | /* We want to make sure that nobody is heartbeating on top of us -- |
| 391 | * this will help detect an invalid configuration. */ |
| 392 | static int o2hb_check_last_timestamp(struct o2hb_region *reg) |
| 393 | { |
| 394 | int node_num, ret; |
| 395 | struct o2hb_disk_slot *slot; |
| 396 | struct o2hb_disk_heartbeat_block *hb_block; |
| 397 | |
| 398 | node_num = o2nm_this_node(); |
| 399 | |
| 400 | ret = 1; |
| 401 | slot = ®->hr_slots[node_num]; |
| 402 | /* Don't check on our 1st timestamp */ |
| 403 | if (slot->ds_last_time) { |
| 404 | hb_block = slot->ds_raw_block; |
| 405 | |
| 406 | if (le64_to_cpu(hb_block->hb_seq) != slot->ds_last_time) |
| 407 | ret = 0; |
| 408 | } |
| 409 | |
| 410 | return ret; |
| 411 | } |
| 412 | |
| 413 | static inline void o2hb_prepare_block(struct o2hb_region *reg, |
| 414 | u64 generation) |
| 415 | { |
| 416 | int node_num; |
| 417 | u64 cputime; |
| 418 | struct o2hb_disk_slot *slot; |
| 419 | struct o2hb_disk_heartbeat_block *hb_block; |
| 420 | |
| 421 | node_num = o2nm_this_node(); |
| 422 | slot = ®->hr_slots[node_num]; |
| 423 | |
| 424 | hb_block = (struct o2hb_disk_heartbeat_block *)slot->ds_raw_block; |
| 425 | memset(hb_block, 0, reg->hr_block_bytes); |
| 426 | /* TODO: time stuff */ |
| 427 | cputime = CURRENT_TIME.tv_sec; |
| 428 | if (!cputime) |
| 429 | cputime = 1; |
| 430 | |
| 431 | hb_block->hb_seq = cpu_to_le64(cputime); |
| 432 | hb_block->hb_node = node_num; |
| 433 | hb_block->hb_generation = cpu_to_le64(generation); |
Mark Fasheh | 0db638f | 2006-05-09 15:09:35 -0700 | [diff] [blame] | 434 | hb_block->hb_dead_ms = cpu_to_le32(o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS); |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 435 | |
| 436 | /* This step must always happen last! */ |
| 437 | hb_block->hb_cksum = cpu_to_le32(o2hb_compute_block_crc_le(reg, |
| 438 | hb_block)); |
| 439 | |
Mark Fasheh | 70bacbdb | 2006-03-02 11:10:05 -0800 | [diff] [blame] | 440 | mlog(ML_HB_BIO, "our node generation = 0x%llx, cksum = 0x%x\n", |
Mark Fasheh | 5fdf1e6 | 2007-04-27 16:50:03 -0700 | [diff] [blame^] | 441 | (long long)generation, |
Mark Fasheh | 70bacbdb | 2006-03-02 11:10:05 -0800 | [diff] [blame] | 442 | le32_to_cpu(hb_block->hb_cksum)); |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 443 | } |
| 444 | |
| 445 | static void o2hb_fire_callbacks(struct o2hb_callback *hbcall, |
| 446 | struct o2nm_node *node, |
| 447 | int idx) |
| 448 | { |
| 449 | struct list_head *iter; |
| 450 | struct o2hb_callback_func *f; |
| 451 | |
| 452 | list_for_each(iter, &hbcall->list) { |
| 453 | f = list_entry(iter, struct o2hb_callback_func, hc_item); |
| 454 | mlog(ML_HEARTBEAT, "calling funcs %p\n", f); |
| 455 | (f->hc_func)(node, idx, f->hc_data); |
| 456 | } |
| 457 | } |
| 458 | |
| 459 | /* Will run the list in order until we process the passed event */ |
| 460 | static void o2hb_run_event_list(struct o2hb_node_event *queued_event) |
| 461 | { |
| 462 | int empty; |
| 463 | struct o2hb_callback *hbcall; |
| 464 | struct o2hb_node_event *event; |
| 465 | |
| 466 | spin_lock(&o2hb_live_lock); |
| 467 | empty = list_empty(&queued_event->hn_item); |
| 468 | spin_unlock(&o2hb_live_lock); |
| 469 | if (empty) |
| 470 | return; |
| 471 | |
| 472 | /* Holding callback sem assures we don't alter the callback |
| 473 | * lists when doing this, and serializes ourselves with other |
| 474 | * processes wanting callbacks. */ |
| 475 | down_write(&o2hb_callback_sem); |
| 476 | |
| 477 | spin_lock(&o2hb_live_lock); |
| 478 | while (!list_empty(&o2hb_node_events) |
| 479 | && !list_empty(&queued_event->hn_item)) { |
| 480 | event = list_entry(o2hb_node_events.next, |
| 481 | struct o2hb_node_event, |
| 482 | hn_item); |
| 483 | list_del_init(&event->hn_item); |
| 484 | spin_unlock(&o2hb_live_lock); |
| 485 | |
| 486 | mlog(ML_HEARTBEAT, "Node %s event for %d\n", |
| 487 | event->hn_event_type == O2HB_NODE_UP_CB ? "UP" : "DOWN", |
| 488 | event->hn_node_num); |
| 489 | |
| 490 | hbcall = hbcall_from_type(event->hn_event_type); |
| 491 | |
| 492 | /* We should *never* have gotten on to the list with a |
| 493 | * bad type... This isn't something that we should try |
| 494 | * to recover from. */ |
| 495 | BUG_ON(IS_ERR(hbcall)); |
| 496 | |
| 497 | o2hb_fire_callbacks(hbcall, event->hn_node, event->hn_node_num); |
| 498 | |
| 499 | spin_lock(&o2hb_live_lock); |
| 500 | } |
| 501 | spin_unlock(&o2hb_live_lock); |
| 502 | |
| 503 | up_write(&o2hb_callback_sem); |
| 504 | } |
| 505 | |
| 506 | static void o2hb_queue_node_event(struct o2hb_node_event *event, |
| 507 | enum o2hb_callback_type type, |
| 508 | struct o2nm_node *node, |
| 509 | int node_num) |
| 510 | { |
| 511 | assert_spin_locked(&o2hb_live_lock); |
| 512 | |
| 513 | event->hn_event_type = type; |
| 514 | event->hn_node = node; |
| 515 | event->hn_node_num = node_num; |
| 516 | |
| 517 | mlog(ML_HEARTBEAT, "Queue node %s event for node %d\n", |
| 518 | type == O2HB_NODE_UP_CB ? "UP" : "DOWN", node_num); |
| 519 | |
| 520 | list_add_tail(&event->hn_item, &o2hb_node_events); |
| 521 | } |
| 522 | |
| 523 | static void o2hb_shutdown_slot(struct o2hb_disk_slot *slot) |
| 524 | { |
| 525 | struct o2hb_node_event event = |
| 526 | { .hn_item = LIST_HEAD_INIT(event.hn_item), }; |
| 527 | struct o2nm_node *node; |
| 528 | |
| 529 | node = o2nm_get_node_by_num(slot->ds_node_num); |
| 530 | if (!node) |
| 531 | return; |
| 532 | |
| 533 | spin_lock(&o2hb_live_lock); |
| 534 | if (!list_empty(&slot->ds_live_item)) { |
| 535 | mlog(ML_HEARTBEAT, "Shutdown, node %d leaves region\n", |
| 536 | slot->ds_node_num); |
| 537 | |
| 538 | list_del_init(&slot->ds_live_item); |
| 539 | |
| 540 | if (list_empty(&o2hb_live_slots[slot->ds_node_num])) { |
| 541 | clear_bit(slot->ds_node_num, o2hb_live_node_bitmap); |
| 542 | |
| 543 | o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node, |
| 544 | slot->ds_node_num); |
| 545 | } |
| 546 | } |
| 547 | spin_unlock(&o2hb_live_lock); |
| 548 | |
| 549 | o2hb_run_event_list(&event); |
| 550 | |
| 551 | o2nm_node_put(node); |
| 552 | } |
| 553 | |
| 554 | static int o2hb_check_slot(struct o2hb_region *reg, |
| 555 | struct o2hb_disk_slot *slot) |
| 556 | { |
| 557 | int changed = 0, gen_changed = 0; |
| 558 | struct o2hb_node_event event = |
| 559 | { .hn_item = LIST_HEAD_INIT(event.hn_item), }; |
| 560 | struct o2nm_node *node; |
| 561 | struct o2hb_disk_heartbeat_block *hb_block = reg->hr_tmp_block; |
| 562 | u64 cputime; |
Mark Fasheh | 0db638f | 2006-05-09 15:09:35 -0700 | [diff] [blame] | 563 | unsigned int dead_ms = o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS; |
| 564 | unsigned int slot_dead_ms; |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 565 | |
| 566 | memcpy(hb_block, slot->ds_raw_block, reg->hr_block_bytes); |
| 567 | |
| 568 | /* Is this correct? Do we assume that the node doesn't exist |
| 569 | * if we're not configured for him? */ |
| 570 | node = o2nm_get_node_by_num(slot->ds_node_num); |
| 571 | if (!node) |
| 572 | return 0; |
| 573 | |
| 574 | if (!o2hb_verify_crc(reg, hb_block)) { |
| 575 | /* all paths from here will drop o2hb_live_lock for |
| 576 | * us. */ |
| 577 | spin_lock(&o2hb_live_lock); |
| 578 | |
| 579 | /* Don't print an error on the console in this case - |
| 580 | * a freshly formatted heartbeat area will not have a |
| 581 | * crc set on it. */ |
| 582 | if (list_empty(&slot->ds_live_item)) |
| 583 | goto out; |
| 584 | |
| 585 | /* The node is live but pushed out a bad crc. We |
| 586 | * consider it a transient miss but don't populate any |
| 587 | * other values as they may be junk. */ |
| 588 | mlog(ML_ERROR, "Node %d has written a bad crc to %s\n", |
| 589 | slot->ds_node_num, reg->hr_dev_name); |
| 590 | o2hb_dump_slot(hb_block); |
| 591 | |
| 592 | slot->ds_equal_samples++; |
| 593 | goto fire_callbacks; |
| 594 | } |
| 595 | |
| 596 | /* we don't care if these wrap.. the state transitions below |
| 597 | * clear at the right places */ |
| 598 | cputime = le64_to_cpu(hb_block->hb_seq); |
| 599 | if (slot->ds_last_time != cputime) |
| 600 | slot->ds_changed_samples++; |
| 601 | else |
| 602 | slot->ds_equal_samples++; |
| 603 | slot->ds_last_time = cputime; |
| 604 | |
| 605 | /* The node changed heartbeat generations. We assume this to |
| 606 | * mean it dropped off but came back before we timed out. We |
| 607 | * want to consider it down for the time being but don't want |
| 608 | * to lose any changed_samples state we might build up to |
| 609 | * considering it live again. */ |
| 610 | if (slot->ds_last_generation != le64_to_cpu(hb_block->hb_generation)) { |
| 611 | gen_changed = 1; |
| 612 | slot->ds_equal_samples = 0; |
Mark Fasheh | 70bacbdb | 2006-03-02 11:10:05 -0800 | [diff] [blame] | 613 | mlog(ML_HEARTBEAT, "Node %d changed generation (0x%llx " |
| 614 | "to 0x%llx)\n", slot->ds_node_num, |
| 615 | (long long)slot->ds_last_generation, |
| 616 | (long long)le64_to_cpu(hb_block->hb_generation)); |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 617 | } |
| 618 | |
| 619 | slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation); |
| 620 | |
Mark Fasheh | 70bacbdb | 2006-03-02 11:10:05 -0800 | [diff] [blame] | 621 | mlog(ML_HEARTBEAT, "Slot %d gen 0x%llx cksum 0x%x " |
| 622 | "seq %llu last %llu changed %u equal %u\n", |
| 623 | slot->ds_node_num, (long long)slot->ds_last_generation, |
| 624 | le32_to_cpu(hb_block->hb_cksum), |
| 625 | (unsigned long long)le64_to_cpu(hb_block->hb_seq), |
| 626 | (unsigned long long)slot->ds_last_time, slot->ds_changed_samples, |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 627 | slot->ds_equal_samples); |
| 628 | |
| 629 | spin_lock(&o2hb_live_lock); |
| 630 | |
| 631 | fire_callbacks: |
| 632 | /* dead nodes only come to life after some number of |
| 633 | * changes at any time during their dead time */ |
| 634 | if (list_empty(&slot->ds_live_item) && |
| 635 | slot->ds_changed_samples >= O2HB_LIVE_THRESHOLD) { |
Mark Fasheh | 70bacbdb | 2006-03-02 11:10:05 -0800 | [diff] [blame] | 636 | mlog(ML_HEARTBEAT, "Node %d (id 0x%llx) joined my region\n", |
| 637 | slot->ds_node_num, (long long)slot->ds_last_generation); |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 638 | |
| 639 | /* first on the list generates a callback */ |
| 640 | if (list_empty(&o2hb_live_slots[slot->ds_node_num])) { |
| 641 | set_bit(slot->ds_node_num, o2hb_live_node_bitmap); |
| 642 | |
| 643 | o2hb_queue_node_event(&event, O2HB_NODE_UP_CB, node, |
| 644 | slot->ds_node_num); |
| 645 | |
| 646 | changed = 1; |
| 647 | } |
| 648 | |
| 649 | list_add_tail(&slot->ds_live_item, |
| 650 | &o2hb_live_slots[slot->ds_node_num]); |
| 651 | |
| 652 | slot->ds_equal_samples = 0; |
Mark Fasheh | 0db638f | 2006-05-09 15:09:35 -0700 | [diff] [blame] | 653 | |
| 654 | /* We want to be sure that all nodes agree on the |
| 655 | * number of milliseconds before a node will be |
| 656 | * considered dead. The self-fencing timeout is |
| 657 | * computed from this value, and a discrepancy might |
| 658 | * result in heartbeat calling a node dead when it |
| 659 | * hasn't self-fenced yet. */ |
| 660 | slot_dead_ms = le32_to_cpu(hb_block->hb_dead_ms); |
| 661 | if (slot_dead_ms && slot_dead_ms != dead_ms) { |
| 662 | /* TODO: Perhaps we can fail the region here. */ |
| 663 | mlog(ML_ERROR, "Node %d on device %s has a dead count " |
| 664 | "of %u ms, but our count is %u ms.\n" |
| 665 | "Please double check your configuration values " |
| 666 | "for 'O2CB_HEARTBEAT_THRESHOLD'\n", |
| 667 | slot->ds_node_num, reg->hr_dev_name, slot_dead_ms, |
| 668 | dead_ms); |
| 669 | } |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 670 | goto out; |
| 671 | } |
| 672 | |
| 673 | /* if the list is dead, we're done.. */ |
| 674 | if (list_empty(&slot->ds_live_item)) |
| 675 | goto out; |
| 676 | |
| 677 | /* live nodes only go dead after enough consequtive missed |
| 678 | * samples.. reset the missed counter whenever we see |
| 679 | * activity */ |
| 680 | if (slot->ds_equal_samples >= o2hb_dead_threshold || gen_changed) { |
| 681 | mlog(ML_HEARTBEAT, "Node %d left my region\n", |
| 682 | slot->ds_node_num); |
| 683 | |
| 684 | /* last off the live_slot generates a callback */ |
| 685 | list_del_init(&slot->ds_live_item); |
| 686 | if (list_empty(&o2hb_live_slots[slot->ds_node_num])) { |
| 687 | clear_bit(slot->ds_node_num, o2hb_live_node_bitmap); |
| 688 | |
| 689 | o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node, |
| 690 | slot->ds_node_num); |
| 691 | |
| 692 | changed = 1; |
| 693 | } |
| 694 | |
| 695 | /* We don't clear this because the node is still |
| 696 | * actually writing new blocks. */ |
| 697 | if (!gen_changed) |
| 698 | slot->ds_changed_samples = 0; |
| 699 | goto out; |
| 700 | } |
| 701 | if (slot->ds_changed_samples) { |
| 702 | slot->ds_changed_samples = 0; |
| 703 | slot->ds_equal_samples = 0; |
| 704 | } |
| 705 | out: |
| 706 | spin_unlock(&o2hb_live_lock); |
| 707 | |
| 708 | o2hb_run_event_list(&event); |
| 709 | |
| 710 | o2nm_node_put(node); |
| 711 | return changed; |
| 712 | } |
| 713 | |
| 714 | /* This could be faster if we just implmented a find_last_bit, but I |
| 715 | * don't think the circumstances warrant it. */ |
| 716 | static int o2hb_highest_node(unsigned long *nodes, |
| 717 | int numbits) |
| 718 | { |
| 719 | int highest, node; |
| 720 | |
| 721 | highest = numbits; |
| 722 | node = -1; |
| 723 | while ((node = find_next_bit(nodes, numbits, node + 1)) != -1) { |
| 724 | if (node >= numbits) |
| 725 | break; |
| 726 | |
| 727 | highest = node; |
| 728 | } |
| 729 | |
| 730 | return highest; |
| 731 | } |
| 732 | |
Mark Fasheh | a9e2ae3 | 2006-03-24 14:20:17 -0800 | [diff] [blame] | 733 | static int o2hb_do_disk_heartbeat(struct o2hb_region *reg) |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 734 | { |
| 735 | int i, ret, highest_node, change = 0; |
| 736 | unsigned long configured_nodes[BITS_TO_LONGS(O2NM_MAX_NODES)]; |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 737 | struct o2hb_bio_wait_ctxt write_wc; |
| 738 | |
Mark Fasheh | a9e2ae3 | 2006-03-24 14:20:17 -0800 | [diff] [blame] | 739 | ret = o2nm_configured_node_map(configured_nodes, |
| 740 | sizeof(configured_nodes)); |
| 741 | if (ret) { |
| 742 | mlog_errno(ret); |
| 743 | return ret; |
| 744 | } |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 745 | |
| 746 | highest_node = o2hb_highest_node(configured_nodes, O2NM_MAX_NODES); |
| 747 | if (highest_node >= O2NM_MAX_NODES) { |
| 748 | mlog(ML_NOTICE, "ocfs2_heartbeat: no configured nodes found!\n"); |
Mark Fasheh | a9e2ae3 | 2006-03-24 14:20:17 -0800 | [diff] [blame] | 749 | return -EINVAL; |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 750 | } |
| 751 | |
| 752 | /* No sense in reading the slots of nodes that don't exist |
| 753 | * yet. Of course, if the node definitions have holes in them |
| 754 | * then we're reading an empty slot anyway... Consider this |
| 755 | * best-effort. */ |
| 756 | ret = o2hb_read_slots(reg, highest_node + 1); |
| 757 | if (ret < 0) { |
| 758 | mlog_errno(ret); |
Mark Fasheh | a9e2ae3 | 2006-03-24 14:20:17 -0800 | [diff] [blame] | 759 | return ret; |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 760 | } |
| 761 | |
| 762 | /* With an up to date view of the slots, we can check that no |
| 763 | * other node has been improperly configured to heartbeat in |
| 764 | * our slot. */ |
| 765 | if (!o2hb_check_last_timestamp(reg)) |
| 766 | mlog(ML_ERROR, "Device \"%s\": another node is heartbeating " |
| 767 | "in our slot!\n", reg->hr_dev_name); |
| 768 | |
| 769 | /* fill in the proper info for our next heartbeat */ |
| 770 | o2hb_prepare_block(reg, reg->hr_generation); |
| 771 | |
| 772 | /* And fire off the write. Note that we don't wait on this I/O |
| 773 | * until later. */ |
Philipp Reisner | b559292 | 2007-01-11 10:58:10 +0100 | [diff] [blame] | 774 | ret = o2hb_issue_node_write(reg, &write_wc); |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 775 | if (ret < 0) { |
| 776 | mlog_errno(ret); |
Mark Fasheh | a9e2ae3 | 2006-03-24 14:20:17 -0800 | [diff] [blame] | 777 | return ret; |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 778 | } |
| 779 | |
| 780 | i = -1; |
| 781 | while((i = find_next_bit(configured_nodes, O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) { |
| 782 | |
| 783 | change |= o2hb_check_slot(reg, ®->hr_slots[i]); |
| 784 | } |
| 785 | |
| 786 | /* |
| 787 | * We have to be sure we've advertised ourselves on disk |
| 788 | * before we can go to steady state. This ensures that |
| 789 | * people we find in our steady state have seen us. |
| 790 | */ |
| 791 | o2hb_wait_on_io(reg, &write_wc); |
Mark Fasheh | a9e2ae3 | 2006-03-24 14:20:17 -0800 | [diff] [blame] | 792 | if (write_wc.wc_error) { |
| 793 | /* Do not re-arm the write timeout on I/O error - we |
| 794 | * can't be sure that the new block ever made it to |
| 795 | * disk */ |
| 796 | mlog(ML_ERROR, "Write error %d on device \"%s\"\n", |
| 797 | write_wc.wc_error, reg->hr_dev_name); |
| 798 | return write_wc.wc_error; |
| 799 | } |
| 800 | |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 801 | o2hb_arm_write_timeout(reg); |
| 802 | |
| 803 | /* let the person who launched us know when things are steady */ |
| 804 | if (!change && (atomic_read(®->hr_steady_iterations) != 0)) { |
| 805 | if (atomic_dec_and_test(®->hr_steady_iterations)) |
| 806 | wake_up(&o2hb_steady_queue); |
| 807 | } |
Mark Fasheh | a9e2ae3 | 2006-03-24 14:20:17 -0800 | [diff] [blame] | 808 | |
| 809 | return 0; |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 810 | } |
| 811 | |
| 812 | /* Subtract b from a, storing the result in a. a *must* have a larger |
| 813 | * value than b. */ |
| 814 | static void o2hb_tv_subtract(struct timeval *a, |
| 815 | struct timeval *b) |
| 816 | { |
| 817 | /* just return 0 when a is after b */ |
| 818 | if (a->tv_sec < b->tv_sec || |
| 819 | (a->tv_sec == b->tv_sec && a->tv_usec < b->tv_usec)) { |
| 820 | a->tv_sec = 0; |
| 821 | a->tv_usec = 0; |
| 822 | return; |
| 823 | } |
| 824 | |
| 825 | a->tv_sec -= b->tv_sec; |
| 826 | a->tv_usec -= b->tv_usec; |
| 827 | while ( a->tv_usec < 0 ) { |
| 828 | a->tv_sec--; |
| 829 | a->tv_usec += 1000000; |
| 830 | } |
| 831 | } |
| 832 | |
| 833 | static unsigned int o2hb_elapsed_msecs(struct timeval *start, |
| 834 | struct timeval *end) |
| 835 | { |
| 836 | struct timeval res = *end; |
| 837 | |
| 838 | o2hb_tv_subtract(&res, start); |
| 839 | |
| 840 | return res.tv_sec * 1000 + res.tv_usec / 1000; |
| 841 | } |
| 842 | |
| 843 | /* |
| 844 | * we ride the region ref that the region dir holds. before the region |
| 845 | * dir is removed and drops it ref it will wait to tear down this |
| 846 | * thread. |
| 847 | */ |
| 848 | static int o2hb_thread(void *data) |
| 849 | { |
| 850 | int i, ret; |
| 851 | struct o2hb_region *reg = data; |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 852 | struct o2hb_bio_wait_ctxt write_wc; |
| 853 | struct timeval before_hb, after_hb; |
| 854 | unsigned int elapsed_msec; |
| 855 | |
| 856 | mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread running\n"); |
| 857 | |
| 858 | set_user_nice(current, -20); |
| 859 | |
| 860 | while (!kthread_should_stop() && !reg->hr_unclean_stop) { |
| 861 | /* We track the time spent inside |
| 862 | * o2hb_do_disk_heartbeat so that we avoid more then |
| 863 | * hr_timeout_ms between disk writes. On busy systems |
| 864 | * this should result in a heartbeat which is less |
| 865 | * likely to time itself out. */ |
| 866 | do_gettimeofday(&before_hb); |
| 867 | |
Mark Fasheh | a9e2ae3 | 2006-03-24 14:20:17 -0800 | [diff] [blame] | 868 | i = 0; |
| 869 | do { |
| 870 | ret = o2hb_do_disk_heartbeat(reg); |
| 871 | } while (ret && ++i < 2); |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 872 | |
| 873 | do_gettimeofday(&after_hb); |
| 874 | elapsed_msec = o2hb_elapsed_msecs(&before_hb, &after_hb); |
| 875 | |
| 876 | mlog(0, "start = %lu.%lu, end = %lu.%lu, msec = %u\n", |
Mark Fasheh | 215c7f9 | 2006-02-01 16:42:10 -0800 | [diff] [blame] | 877 | before_hb.tv_sec, (unsigned long) before_hb.tv_usec, |
| 878 | after_hb.tv_sec, (unsigned long) after_hb.tv_usec, |
| 879 | elapsed_msec); |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 880 | |
| 881 | if (elapsed_msec < reg->hr_timeout_ms) { |
| 882 | /* the kthread api has blocked signals for us so no |
| 883 | * need to record the return value. */ |
| 884 | msleep_interruptible(reg->hr_timeout_ms - elapsed_msec); |
| 885 | } |
| 886 | } |
| 887 | |
| 888 | o2hb_disarm_write_timeout(reg); |
| 889 | |
| 890 | /* unclean stop is only used in very bad situation */ |
| 891 | for(i = 0; !reg->hr_unclean_stop && i < reg->hr_blocks; i++) |
| 892 | o2hb_shutdown_slot(®->hr_slots[i]); |
| 893 | |
| 894 | /* Explicit down notification - avoid forcing the other nodes |
| 895 | * to timeout on this region when we could just as easily |
| 896 | * write a clear generation - thus indicating to them that |
| 897 | * this node has left this region. |
| 898 | * |
| 899 | * XXX: Should we skip this on unclean_stop? */ |
| 900 | o2hb_prepare_block(reg, 0); |
Philipp Reisner | b559292 | 2007-01-11 10:58:10 +0100 | [diff] [blame] | 901 | ret = o2hb_issue_node_write(reg, &write_wc); |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 902 | if (ret == 0) { |
| 903 | o2hb_wait_on_io(reg, &write_wc); |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 904 | } else { |
| 905 | mlog_errno(ret); |
| 906 | } |
| 907 | |
| 908 | mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread exiting\n"); |
| 909 | |
| 910 | return 0; |
| 911 | } |
| 912 | |
| 913 | void o2hb_init(void) |
| 914 | { |
| 915 | int i; |
| 916 | |
| 917 | for (i = 0; i < ARRAY_SIZE(o2hb_callbacks); i++) |
| 918 | INIT_LIST_HEAD(&o2hb_callbacks[i].list); |
| 919 | |
| 920 | for (i = 0; i < ARRAY_SIZE(o2hb_live_slots); i++) |
| 921 | INIT_LIST_HEAD(&o2hb_live_slots[i]); |
| 922 | |
| 923 | INIT_LIST_HEAD(&o2hb_node_events); |
| 924 | |
| 925 | memset(o2hb_live_node_bitmap, 0, sizeof(o2hb_live_node_bitmap)); |
| 926 | } |
| 927 | |
| 928 | /* if we're already in a callback then we're already serialized by the sem */ |
| 929 | static void o2hb_fill_node_map_from_callback(unsigned long *map, |
| 930 | unsigned bytes) |
| 931 | { |
| 932 | BUG_ON(bytes < (BITS_TO_LONGS(O2NM_MAX_NODES) * sizeof(unsigned long))); |
| 933 | |
| 934 | memcpy(map, &o2hb_live_node_bitmap, bytes); |
| 935 | } |
| 936 | |
| 937 | /* |
| 938 | * get a map of all nodes that are heartbeating in any regions |
| 939 | */ |
| 940 | void o2hb_fill_node_map(unsigned long *map, unsigned bytes) |
| 941 | { |
| 942 | /* callers want to serialize this map and callbacks so that they |
| 943 | * can trust that they don't miss nodes coming to the party */ |
| 944 | down_read(&o2hb_callback_sem); |
| 945 | spin_lock(&o2hb_live_lock); |
| 946 | o2hb_fill_node_map_from_callback(map, bytes); |
| 947 | spin_unlock(&o2hb_live_lock); |
| 948 | up_read(&o2hb_callback_sem); |
| 949 | } |
| 950 | EXPORT_SYMBOL_GPL(o2hb_fill_node_map); |
| 951 | |
| 952 | /* |
| 953 | * heartbeat configfs bits. The heartbeat set is a default set under |
| 954 | * the cluster set in nodemanager.c. |
| 955 | */ |
| 956 | |
| 957 | static struct o2hb_region *to_o2hb_region(struct config_item *item) |
| 958 | { |
| 959 | return item ? container_of(item, struct o2hb_region, hr_item) : NULL; |
| 960 | } |
| 961 | |
| 962 | /* drop_item only drops its ref after killing the thread, nothing should |
| 963 | * be using the region anymore. this has to clean up any state that |
| 964 | * attributes might have built up. */ |
| 965 | static void o2hb_region_release(struct config_item *item) |
| 966 | { |
| 967 | int i; |
| 968 | struct page *page; |
| 969 | struct o2hb_region *reg = to_o2hb_region(item); |
| 970 | |
| 971 | if (reg->hr_tmp_block) |
| 972 | kfree(reg->hr_tmp_block); |
| 973 | |
| 974 | if (reg->hr_slot_data) { |
| 975 | for (i = 0; i < reg->hr_num_pages; i++) { |
| 976 | page = reg->hr_slot_data[i]; |
| 977 | if (page) |
| 978 | __free_page(page); |
| 979 | } |
| 980 | kfree(reg->hr_slot_data); |
| 981 | } |
| 982 | |
| 983 | if (reg->hr_bdev) |
| 984 | blkdev_put(reg->hr_bdev); |
| 985 | |
| 986 | if (reg->hr_slots) |
| 987 | kfree(reg->hr_slots); |
| 988 | |
| 989 | spin_lock(&o2hb_live_lock); |
| 990 | list_del(®->hr_all_item); |
| 991 | spin_unlock(&o2hb_live_lock); |
| 992 | |
| 993 | kfree(reg); |
| 994 | } |
| 995 | |
| 996 | static int o2hb_read_block_input(struct o2hb_region *reg, |
| 997 | const char *page, |
| 998 | size_t count, |
| 999 | unsigned long *ret_bytes, |
| 1000 | unsigned int *ret_bits) |
| 1001 | { |
| 1002 | unsigned long bytes; |
| 1003 | char *p = (char *)page; |
| 1004 | |
| 1005 | bytes = simple_strtoul(p, &p, 0); |
| 1006 | if (!p || (*p && (*p != '\n'))) |
| 1007 | return -EINVAL; |
| 1008 | |
| 1009 | /* Heartbeat and fs min / max block sizes are the same. */ |
| 1010 | if (bytes > 4096 || bytes < 512) |
| 1011 | return -ERANGE; |
| 1012 | if (hweight16(bytes) != 1) |
| 1013 | return -EINVAL; |
| 1014 | |
| 1015 | if (ret_bytes) |
| 1016 | *ret_bytes = bytes; |
| 1017 | if (ret_bits) |
| 1018 | *ret_bits = ffs(bytes) - 1; |
| 1019 | |
| 1020 | return 0; |
| 1021 | } |
| 1022 | |
| 1023 | static ssize_t o2hb_region_block_bytes_read(struct o2hb_region *reg, |
| 1024 | char *page) |
| 1025 | { |
| 1026 | return sprintf(page, "%u\n", reg->hr_block_bytes); |
| 1027 | } |
| 1028 | |
| 1029 | static ssize_t o2hb_region_block_bytes_write(struct o2hb_region *reg, |
| 1030 | const char *page, |
| 1031 | size_t count) |
| 1032 | { |
| 1033 | int status; |
| 1034 | unsigned long block_bytes; |
| 1035 | unsigned int block_bits; |
| 1036 | |
| 1037 | if (reg->hr_bdev) |
| 1038 | return -EINVAL; |
| 1039 | |
| 1040 | status = o2hb_read_block_input(reg, page, count, |
| 1041 | &block_bytes, &block_bits); |
| 1042 | if (status) |
| 1043 | return status; |
| 1044 | |
| 1045 | reg->hr_block_bytes = (unsigned int)block_bytes; |
| 1046 | reg->hr_block_bits = block_bits; |
| 1047 | |
| 1048 | return count; |
| 1049 | } |
| 1050 | |
| 1051 | static ssize_t o2hb_region_start_block_read(struct o2hb_region *reg, |
| 1052 | char *page) |
| 1053 | { |
| 1054 | return sprintf(page, "%llu\n", reg->hr_start_block); |
| 1055 | } |
| 1056 | |
| 1057 | static ssize_t o2hb_region_start_block_write(struct o2hb_region *reg, |
| 1058 | const char *page, |
| 1059 | size_t count) |
| 1060 | { |
| 1061 | unsigned long long tmp; |
| 1062 | char *p = (char *)page; |
| 1063 | |
| 1064 | if (reg->hr_bdev) |
| 1065 | return -EINVAL; |
| 1066 | |
| 1067 | tmp = simple_strtoull(p, &p, 0); |
| 1068 | if (!p || (*p && (*p != '\n'))) |
| 1069 | return -EINVAL; |
| 1070 | |
| 1071 | reg->hr_start_block = tmp; |
| 1072 | |
| 1073 | return count; |
| 1074 | } |
| 1075 | |
| 1076 | static ssize_t o2hb_region_blocks_read(struct o2hb_region *reg, |
| 1077 | char *page) |
| 1078 | { |
| 1079 | return sprintf(page, "%d\n", reg->hr_blocks); |
| 1080 | } |
| 1081 | |
| 1082 | static ssize_t o2hb_region_blocks_write(struct o2hb_region *reg, |
| 1083 | const char *page, |
| 1084 | size_t count) |
| 1085 | { |
| 1086 | unsigned long tmp; |
| 1087 | char *p = (char *)page; |
| 1088 | |
| 1089 | if (reg->hr_bdev) |
| 1090 | return -EINVAL; |
| 1091 | |
| 1092 | tmp = simple_strtoul(p, &p, 0); |
| 1093 | if (!p || (*p && (*p != '\n'))) |
| 1094 | return -EINVAL; |
| 1095 | |
| 1096 | if (tmp > O2NM_MAX_NODES || tmp == 0) |
| 1097 | return -ERANGE; |
| 1098 | |
| 1099 | reg->hr_blocks = (unsigned int)tmp; |
| 1100 | |
| 1101 | return count; |
| 1102 | } |
| 1103 | |
| 1104 | static ssize_t o2hb_region_dev_read(struct o2hb_region *reg, |
| 1105 | char *page) |
| 1106 | { |
| 1107 | unsigned int ret = 0; |
| 1108 | |
| 1109 | if (reg->hr_bdev) |
| 1110 | ret = sprintf(page, "%s\n", reg->hr_dev_name); |
| 1111 | |
| 1112 | return ret; |
| 1113 | } |
| 1114 | |
| 1115 | static void o2hb_init_region_params(struct o2hb_region *reg) |
| 1116 | { |
| 1117 | reg->hr_slots_per_page = PAGE_CACHE_SIZE >> reg->hr_block_bits; |
| 1118 | reg->hr_timeout_ms = O2HB_REGION_TIMEOUT_MS; |
| 1119 | |
| 1120 | mlog(ML_HEARTBEAT, "hr_start_block = %llu, hr_blocks = %u\n", |
| 1121 | reg->hr_start_block, reg->hr_blocks); |
| 1122 | mlog(ML_HEARTBEAT, "hr_block_bytes = %u, hr_block_bits = %u\n", |
| 1123 | reg->hr_block_bytes, reg->hr_block_bits); |
| 1124 | mlog(ML_HEARTBEAT, "hr_timeout_ms = %u\n", reg->hr_timeout_ms); |
| 1125 | mlog(ML_HEARTBEAT, "dead threshold = %u\n", o2hb_dead_threshold); |
| 1126 | } |
| 1127 | |
| 1128 | static int o2hb_map_slot_data(struct o2hb_region *reg) |
| 1129 | { |
| 1130 | int i, j; |
| 1131 | unsigned int last_slot; |
| 1132 | unsigned int spp = reg->hr_slots_per_page; |
| 1133 | struct page *page; |
| 1134 | char *raw; |
| 1135 | struct o2hb_disk_slot *slot; |
| 1136 | |
| 1137 | reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL); |
| 1138 | if (reg->hr_tmp_block == NULL) { |
| 1139 | mlog_errno(-ENOMEM); |
| 1140 | return -ENOMEM; |
| 1141 | } |
| 1142 | |
| 1143 | reg->hr_slots = kcalloc(reg->hr_blocks, |
| 1144 | sizeof(struct o2hb_disk_slot), GFP_KERNEL); |
| 1145 | if (reg->hr_slots == NULL) { |
| 1146 | mlog_errno(-ENOMEM); |
| 1147 | return -ENOMEM; |
| 1148 | } |
| 1149 | |
| 1150 | for(i = 0; i < reg->hr_blocks; i++) { |
| 1151 | slot = ®->hr_slots[i]; |
| 1152 | slot->ds_node_num = i; |
| 1153 | INIT_LIST_HEAD(&slot->ds_live_item); |
| 1154 | slot->ds_raw_block = NULL; |
| 1155 | } |
| 1156 | |
| 1157 | reg->hr_num_pages = (reg->hr_blocks + spp - 1) / spp; |
| 1158 | mlog(ML_HEARTBEAT, "Going to require %u pages to cover %u blocks " |
| 1159 | "at %u blocks per page\n", |
| 1160 | reg->hr_num_pages, reg->hr_blocks, spp); |
| 1161 | |
| 1162 | reg->hr_slot_data = kcalloc(reg->hr_num_pages, sizeof(struct page *), |
| 1163 | GFP_KERNEL); |
| 1164 | if (!reg->hr_slot_data) { |
| 1165 | mlog_errno(-ENOMEM); |
| 1166 | return -ENOMEM; |
| 1167 | } |
| 1168 | |
| 1169 | for(i = 0; i < reg->hr_num_pages; i++) { |
| 1170 | page = alloc_page(GFP_KERNEL); |
| 1171 | if (!page) { |
| 1172 | mlog_errno(-ENOMEM); |
| 1173 | return -ENOMEM; |
| 1174 | } |
| 1175 | |
| 1176 | reg->hr_slot_data[i] = page; |
| 1177 | |
| 1178 | last_slot = i * spp; |
| 1179 | raw = page_address(page); |
| 1180 | for (j = 0; |
| 1181 | (j < spp) && ((j + last_slot) < reg->hr_blocks); |
| 1182 | j++) { |
| 1183 | BUG_ON((j + last_slot) >= reg->hr_blocks); |
| 1184 | |
| 1185 | slot = ®->hr_slots[j + last_slot]; |
| 1186 | slot->ds_raw_block = |
| 1187 | (struct o2hb_disk_heartbeat_block *) raw; |
| 1188 | |
| 1189 | raw += reg->hr_block_bytes; |
| 1190 | } |
| 1191 | } |
| 1192 | |
| 1193 | return 0; |
| 1194 | } |
| 1195 | |
| 1196 | /* Read in all the slots available and populate the tracking |
| 1197 | * structures so that we can start with a baseline idea of what's |
| 1198 | * there. */ |
| 1199 | static int o2hb_populate_slot_data(struct o2hb_region *reg) |
| 1200 | { |
| 1201 | int ret, i; |
| 1202 | struct o2hb_disk_slot *slot; |
| 1203 | struct o2hb_disk_heartbeat_block *hb_block; |
| 1204 | |
| 1205 | mlog_entry_void(); |
| 1206 | |
| 1207 | ret = o2hb_read_slots(reg, reg->hr_blocks); |
| 1208 | if (ret) { |
| 1209 | mlog_errno(ret); |
| 1210 | goto out; |
| 1211 | } |
| 1212 | |
| 1213 | /* We only want to get an idea of the values initially in each |
| 1214 | * slot, so we do no verification - o2hb_check_slot will |
| 1215 | * actually determine if each configured slot is valid and |
| 1216 | * whether any values have changed. */ |
| 1217 | for(i = 0; i < reg->hr_blocks; i++) { |
| 1218 | slot = ®->hr_slots[i]; |
| 1219 | hb_block = (struct o2hb_disk_heartbeat_block *) slot->ds_raw_block; |
| 1220 | |
| 1221 | /* Only fill the values that o2hb_check_slot uses to |
| 1222 | * determine changing slots */ |
| 1223 | slot->ds_last_time = le64_to_cpu(hb_block->hb_seq); |
| 1224 | slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation); |
| 1225 | } |
| 1226 | |
| 1227 | out: |
| 1228 | mlog_exit(ret); |
| 1229 | return ret; |
| 1230 | } |
| 1231 | |
| 1232 | /* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */ |
| 1233 | static ssize_t o2hb_region_dev_write(struct o2hb_region *reg, |
| 1234 | const char *page, |
| 1235 | size_t count) |
| 1236 | { |
Joel Becker | e6c352d | 2007-02-03 03:04:20 -0800 | [diff] [blame] | 1237 | struct task_struct *hb_task; |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1238 | long fd; |
| 1239 | int sectsize; |
| 1240 | char *p = (char *)page; |
| 1241 | struct file *filp = NULL; |
| 1242 | struct inode *inode = NULL; |
| 1243 | ssize_t ret = -EINVAL; |
| 1244 | |
| 1245 | if (reg->hr_bdev) |
| 1246 | goto out; |
| 1247 | |
| 1248 | /* We can't heartbeat without having had our node number |
| 1249 | * configured yet. */ |
| 1250 | if (o2nm_this_node() == O2NM_MAX_NODES) |
| 1251 | goto out; |
| 1252 | |
| 1253 | fd = simple_strtol(p, &p, 0); |
| 1254 | if (!p || (*p && (*p != '\n'))) |
| 1255 | goto out; |
| 1256 | |
| 1257 | if (fd < 0 || fd >= INT_MAX) |
| 1258 | goto out; |
| 1259 | |
| 1260 | filp = fget(fd); |
| 1261 | if (filp == NULL) |
| 1262 | goto out; |
| 1263 | |
| 1264 | if (reg->hr_blocks == 0 || reg->hr_start_block == 0 || |
| 1265 | reg->hr_block_bytes == 0) |
| 1266 | goto out; |
| 1267 | |
| 1268 | inode = igrab(filp->f_mapping->host); |
| 1269 | if (inode == NULL) |
| 1270 | goto out; |
| 1271 | |
| 1272 | if (!S_ISBLK(inode->i_mode)) |
| 1273 | goto out; |
| 1274 | |
| 1275 | reg->hr_bdev = I_BDEV(filp->f_mapping->host); |
| 1276 | ret = blkdev_get(reg->hr_bdev, FMODE_WRITE | FMODE_READ, 0); |
| 1277 | if (ret) { |
| 1278 | reg->hr_bdev = NULL; |
| 1279 | goto out; |
| 1280 | } |
| 1281 | inode = NULL; |
| 1282 | |
| 1283 | bdevname(reg->hr_bdev, reg->hr_dev_name); |
| 1284 | |
| 1285 | sectsize = bdev_hardsect_size(reg->hr_bdev); |
| 1286 | if (sectsize != reg->hr_block_bytes) { |
| 1287 | mlog(ML_ERROR, |
| 1288 | "blocksize %u incorrect for device, expected %d", |
| 1289 | reg->hr_block_bytes, sectsize); |
| 1290 | ret = -EINVAL; |
| 1291 | goto out; |
| 1292 | } |
| 1293 | |
| 1294 | o2hb_init_region_params(reg); |
| 1295 | |
| 1296 | /* Generation of zero is invalid */ |
| 1297 | do { |
| 1298 | get_random_bytes(®->hr_generation, |
| 1299 | sizeof(reg->hr_generation)); |
| 1300 | } while (reg->hr_generation == 0); |
| 1301 | |
| 1302 | ret = o2hb_map_slot_data(reg); |
| 1303 | if (ret) { |
| 1304 | mlog_errno(ret); |
| 1305 | goto out; |
| 1306 | } |
| 1307 | |
| 1308 | ret = o2hb_populate_slot_data(reg); |
| 1309 | if (ret) { |
| 1310 | mlog_errno(ret); |
| 1311 | goto out; |
| 1312 | } |
| 1313 | |
David Howells | c402895 | 2006-11-22 14:57:56 +0000 | [diff] [blame] | 1314 | INIT_DELAYED_WORK(®->hr_write_timeout_work, o2hb_write_timeout); |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1315 | |
| 1316 | /* |
| 1317 | * A node is considered live after it has beat LIVE_THRESHOLD |
| 1318 | * times. We're not steady until we've given them a chance |
| 1319 | * _after_ our first read. |
| 1320 | */ |
| 1321 | atomic_set(®->hr_steady_iterations, O2HB_LIVE_THRESHOLD + 1); |
| 1322 | |
Joel Becker | e6c352d | 2007-02-03 03:04:20 -0800 | [diff] [blame] | 1323 | hb_task = kthread_run(o2hb_thread, reg, "o2hb-%s", |
| 1324 | reg->hr_item.ci_name); |
| 1325 | if (IS_ERR(hb_task)) { |
| 1326 | ret = PTR_ERR(hb_task); |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1327 | mlog_errno(ret); |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1328 | goto out; |
| 1329 | } |
| 1330 | |
Joel Becker | e6c352d | 2007-02-03 03:04:20 -0800 | [diff] [blame] | 1331 | spin_lock(&o2hb_live_lock); |
| 1332 | reg->hr_task = hb_task; |
| 1333 | spin_unlock(&o2hb_live_lock); |
| 1334 | |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1335 | ret = wait_event_interruptible(o2hb_steady_queue, |
| 1336 | atomic_read(®->hr_steady_iterations) == 0); |
| 1337 | if (ret) { |
Joel Becker | e6c352d | 2007-02-03 03:04:20 -0800 | [diff] [blame] | 1338 | spin_lock(&o2hb_live_lock); |
| 1339 | hb_task = reg->hr_task; |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1340 | reg->hr_task = NULL; |
Joel Becker | e6c352d | 2007-02-03 03:04:20 -0800 | [diff] [blame] | 1341 | spin_unlock(&o2hb_live_lock); |
| 1342 | |
| 1343 | if (hb_task) |
| 1344 | kthread_stop(hb_task); |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1345 | goto out; |
| 1346 | } |
| 1347 | |
| 1348 | ret = count; |
| 1349 | out: |
| 1350 | if (filp) |
| 1351 | fput(filp); |
| 1352 | if (inode) |
| 1353 | iput(inode); |
| 1354 | if (ret < 0) { |
| 1355 | if (reg->hr_bdev) { |
| 1356 | blkdev_put(reg->hr_bdev); |
| 1357 | reg->hr_bdev = NULL; |
| 1358 | } |
| 1359 | } |
| 1360 | return ret; |
| 1361 | } |
| 1362 | |
Zhen Wei | 92efc15 | 2006-12-08 00:48:17 -0700 | [diff] [blame] | 1363 | static ssize_t o2hb_region_pid_read(struct o2hb_region *reg, |
| 1364 | char *page) |
| 1365 | { |
Joel Becker | e6c352d | 2007-02-03 03:04:20 -0800 | [diff] [blame] | 1366 | pid_t pid = 0; |
| 1367 | |
| 1368 | spin_lock(&o2hb_live_lock); |
| 1369 | if (reg->hr_task) |
| 1370 | pid = reg->hr_task->pid; |
| 1371 | spin_unlock(&o2hb_live_lock); |
| 1372 | |
| 1373 | if (!pid) |
Zhen Wei | 92efc15 | 2006-12-08 00:48:17 -0700 | [diff] [blame] | 1374 | return 0; |
| 1375 | |
Joel Becker | e6c352d | 2007-02-03 03:04:20 -0800 | [diff] [blame] | 1376 | return sprintf(page, "%u\n", pid); |
Zhen Wei | 92efc15 | 2006-12-08 00:48:17 -0700 | [diff] [blame] | 1377 | } |
| 1378 | |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1379 | struct o2hb_region_attribute { |
| 1380 | struct configfs_attribute attr; |
| 1381 | ssize_t (*show)(struct o2hb_region *, char *); |
| 1382 | ssize_t (*store)(struct o2hb_region *, const char *, size_t); |
| 1383 | }; |
| 1384 | |
| 1385 | static struct o2hb_region_attribute o2hb_region_attr_block_bytes = { |
| 1386 | .attr = { .ca_owner = THIS_MODULE, |
| 1387 | .ca_name = "block_bytes", |
| 1388 | .ca_mode = S_IRUGO | S_IWUSR }, |
| 1389 | .show = o2hb_region_block_bytes_read, |
| 1390 | .store = o2hb_region_block_bytes_write, |
| 1391 | }; |
| 1392 | |
| 1393 | static struct o2hb_region_attribute o2hb_region_attr_start_block = { |
| 1394 | .attr = { .ca_owner = THIS_MODULE, |
| 1395 | .ca_name = "start_block", |
| 1396 | .ca_mode = S_IRUGO | S_IWUSR }, |
| 1397 | .show = o2hb_region_start_block_read, |
| 1398 | .store = o2hb_region_start_block_write, |
| 1399 | }; |
| 1400 | |
| 1401 | static struct o2hb_region_attribute o2hb_region_attr_blocks = { |
| 1402 | .attr = { .ca_owner = THIS_MODULE, |
| 1403 | .ca_name = "blocks", |
| 1404 | .ca_mode = S_IRUGO | S_IWUSR }, |
| 1405 | .show = o2hb_region_blocks_read, |
| 1406 | .store = o2hb_region_blocks_write, |
| 1407 | }; |
| 1408 | |
| 1409 | static struct o2hb_region_attribute o2hb_region_attr_dev = { |
| 1410 | .attr = { .ca_owner = THIS_MODULE, |
| 1411 | .ca_name = "dev", |
| 1412 | .ca_mode = S_IRUGO | S_IWUSR }, |
| 1413 | .show = o2hb_region_dev_read, |
| 1414 | .store = o2hb_region_dev_write, |
| 1415 | }; |
| 1416 | |
Zhen Wei | 92efc15 | 2006-12-08 00:48:17 -0700 | [diff] [blame] | 1417 | static struct o2hb_region_attribute o2hb_region_attr_pid = { |
| 1418 | .attr = { .ca_owner = THIS_MODULE, |
| 1419 | .ca_name = "pid", |
| 1420 | .ca_mode = S_IRUGO | S_IRUSR }, |
| 1421 | .show = o2hb_region_pid_read, |
| 1422 | }; |
| 1423 | |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1424 | static struct configfs_attribute *o2hb_region_attrs[] = { |
| 1425 | &o2hb_region_attr_block_bytes.attr, |
| 1426 | &o2hb_region_attr_start_block.attr, |
| 1427 | &o2hb_region_attr_blocks.attr, |
| 1428 | &o2hb_region_attr_dev.attr, |
Zhen Wei | 92efc15 | 2006-12-08 00:48:17 -0700 | [diff] [blame] | 1429 | &o2hb_region_attr_pid.attr, |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1430 | NULL, |
| 1431 | }; |
| 1432 | |
| 1433 | static ssize_t o2hb_region_show(struct config_item *item, |
| 1434 | struct configfs_attribute *attr, |
| 1435 | char *page) |
| 1436 | { |
| 1437 | struct o2hb_region *reg = to_o2hb_region(item); |
| 1438 | struct o2hb_region_attribute *o2hb_region_attr = |
| 1439 | container_of(attr, struct o2hb_region_attribute, attr); |
| 1440 | ssize_t ret = 0; |
| 1441 | |
| 1442 | if (o2hb_region_attr->show) |
| 1443 | ret = o2hb_region_attr->show(reg, page); |
| 1444 | return ret; |
| 1445 | } |
| 1446 | |
| 1447 | static ssize_t o2hb_region_store(struct config_item *item, |
| 1448 | struct configfs_attribute *attr, |
| 1449 | const char *page, size_t count) |
| 1450 | { |
| 1451 | struct o2hb_region *reg = to_o2hb_region(item); |
| 1452 | struct o2hb_region_attribute *o2hb_region_attr = |
| 1453 | container_of(attr, struct o2hb_region_attribute, attr); |
| 1454 | ssize_t ret = -EINVAL; |
| 1455 | |
| 1456 | if (o2hb_region_attr->store) |
| 1457 | ret = o2hb_region_attr->store(reg, page, count); |
| 1458 | return ret; |
| 1459 | } |
| 1460 | |
| 1461 | static struct configfs_item_operations o2hb_region_item_ops = { |
| 1462 | .release = o2hb_region_release, |
| 1463 | .show_attribute = o2hb_region_show, |
| 1464 | .store_attribute = o2hb_region_store, |
| 1465 | }; |
| 1466 | |
| 1467 | static struct config_item_type o2hb_region_type = { |
| 1468 | .ct_item_ops = &o2hb_region_item_ops, |
| 1469 | .ct_attrs = o2hb_region_attrs, |
| 1470 | .ct_owner = THIS_MODULE, |
| 1471 | }; |
| 1472 | |
| 1473 | /* heartbeat set */ |
| 1474 | |
| 1475 | struct o2hb_heartbeat_group { |
| 1476 | struct config_group hs_group; |
| 1477 | /* some stuff? */ |
| 1478 | }; |
| 1479 | |
| 1480 | static struct o2hb_heartbeat_group *to_o2hb_heartbeat_group(struct config_group *group) |
| 1481 | { |
| 1482 | return group ? |
| 1483 | container_of(group, struct o2hb_heartbeat_group, hs_group) |
| 1484 | : NULL; |
| 1485 | } |
| 1486 | |
| 1487 | static struct config_item *o2hb_heartbeat_group_make_item(struct config_group *group, |
| 1488 | const char *name) |
| 1489 | { |
| 1490 | struct o2hb_region *reg = NULL; |
| 1491 | struct config_item *ret = NULL; |
| 1492 | |
Robert P. J. Day | cd86128 | 2006-12-13 00:34:52 -0800 | [diff] [blame] | 1493 | reg = kzalloc(sizeof(struct o2hb_region), GFP_KERNEL); |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1494 | if (reg == NULL) |
| 1495 | goto out; /* ENOMEM */ |
| 1496 | |
| 1497 | config_item_init_type_name(®->hr_item, name, &o2hb_region_type); |
| 1498 | |
| 1499 | ret = ®->hr_item; |
| 1500 | |
| 1501 | spin_lock(&o2hb_live_lock); |
| 1502 | list_add_tail(®->hr_all_item, &o2hb_all_regions); |
| 1503 | spin_unlock(&o2hb_live_lock); |
| 1504 | out: |
| 1505 | if (ret == NULL) |
| 1506 | kfree(reg); |
| 1507 | |
| 1508 | return ret; |
| 1509 | } |
| 1510 | |
| 1511 | static void o2hb_heartbeat_group_drop_item(struct config_group *group, |
| 1512 | struct config_item *item) |
| 1513 | { |
Joel Becker | e6c352d | 2007-02-03 03:04:20 -0800 | [diff] [blame] | 1514 | struct task_struct *hb_task; |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1515 | struct o2hb_region *reg = to_o2hb_region(item); |
| 1516 | |
| 1517 | /* stop the thread when the user removes the region dir */ |
Joel Becker | e6c352d | 2007-02-03 03:04:20 -0800 | [diff] [blame] | 1518 | spin_lock(&o2hb_live_lock); |
| 1519 | hb_task = reg->hr_task; |
| 1520 | reg->hr_task = NULL; |
| 1521 | spin_unlock(&o2hb_live_lock); |
| 1522 | |
| 1523 | if (hb_task) |
| 1524 | kthread_stop(hb_task); |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1525 | |
| 1526 | config_item_put(item); |
| 1527 | } |
| 1528 | |
| 1529 | struct o2hb_heartbeat_group_attribute { |
| 1530 | struct configfs_attribute attr; |
| 1531 | ssize_t (*show)(struct o2hb_heartbeat_group *, char *); |
| 1532 | ssize_t (*store)(struct o2hb_heartbeat_group *, const char *, size_t); |
| 1533 | }; |
| 1534 | |
| 1535 | static ssize_t o2hb_heartbeat_group_show(struct config_item *item, |
| 1536 | struct configfs_attribute *attr, |
| 1537 | char *page) |
| 1538 | { |
| 1539 | struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item)); |
| 1540 | struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr = |
| 1541 | container_of(attr, struct o2hb_heartbeat_group_attribute, attr); |
| 1542 | ssize_t ret = 0; |
| 1543 | |
| 1544 | if (o2hb_heartbeat_group_attr->show) |
| 1545 | ret = o2hb_heartbeat_group_attr->show(reg, page); |
| 1546 | return ret; |
| 1547 | } |
| 1548 | |
| 1549 | static ssize_t o2hb_heartbeat_group_store(struct config_item *item, |
| 1550 | struct configfs_attribute *attr, |
| 1551 | const char *page, size_t count) |
| 1552 | { |
| 1553 | struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item)); |
| 1554 | struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr = |
| 1555 | container_of(attr, struct o2hb_heartbeat_group_attribute, attr); |
| 1556 | ssize_t ret = -EINVAL; |
| 1557 | |
| 1558 | if (o2hb_heartbeat_group_attr->store) |
| 1559 | ret = o2hb_heartbeat_group_attr->store(reg, page, count); |
| 1560 | return ret; |
| 1561 | } |
| 1562 | |
| 1563 | static ssize_t o2hb_heartbeat_group_threshold_show(struct o2hb_heartbeat_group *group, |
| 1564 | char *page) |
| 1565 | { |
| 1566 | return sprintf(page, "%u\n", o2hb_dead_threshold); |
| 1567 | } |
| 1568 | |
| 1569 | static ssize_t o2hb_heartbeat_group_threshold_store(struct o2hb_heartbeat_group *group, |
| 1570 | const char *page, |
| 1571 | size_t count) |
| 1572 | { |
| 1573 | unsigned long tmp; |
| 1574 | char *p = (char *)page; |
| 1575 | |
| 1576 | tmp = simple_strtoul(p, &p, 10); |
| 1577 | if (!p || (*p && (*p != '\n'))) |
| 1578 | return -EINVAL; |
| 1579 | |
| 1580 | /* this will validate ranges for us. */ |
| 1581 | o2hb_dead_threshold_set((unsigned int) tmp); |
| 1582 | |
| 1583 | return count; |
| 1584 | } |
| 1585 | |
| 1586 | static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_threshold = { |
| 1587 | .attr = { .ca_owner = THIS_MODULE, |
| 1588 | .ca_name = "dead_threshold", |
| 1589 | .ca_mode = S_IRUGO | S_IWUSR }, |
| 1590 | .show = o2hb_heartbeat_group_threshold_show, |
| 1591 | .store = o2hb_heartbeat_group_threshold_store, |
| 1592 | }; |
| 1593 | |
| 1594 | static struct configfs_attribute *o2hb_heartbeat_group_attrs[] = { |
| 1595 | &o2hb_heartbeat_group_attr_threshold.attr, |
| 1596 | NULL, |
| 1597 | }; |
| 1598 | |
| 1599 | static struct configfs_item_operations o2hb_hearbeat_group_item_ops = { |
| 1600 | .show_attribute = o2hb_heartbeat_group_show, |
| 1601 | .store_attribute = o2hb_heartbeat_group_store, |
| 1602 | }; |
| 1603 | |
| 1604 | static struct configfs_group_operations o2hb_heartbeat_group_group_ops = { |
| 1605 | .make_item = o2hb_heartbeat_group_make_item, |
| 1606 | .drop_item = o2hb_heartbeat_group_drop_item, |
| 1607 | }; |
| 1608 | |
| 1609 | static struct config_item_type o2hb_heartbeat_group_type = { |
| 1610 | .ct_group_ops = &o2hb_heartbeat_group_group_ops, |
| 1611 | .ct_item_ops = &o2hb_hearbeat_group_item_ops, |
| 1612 | .ct_attrs = o2hb_heartbeat_group_attrs, |
| 1613 | .ct_owner = THIS_MODULE, |
| 1614 | }; |
| 1615 | |
| 1616 | /* this is just here to avoid touching group in heartbeat.h which the |
| 1617 | * entire damn world #includes */ |
| 1618 | struct config_group *o2hb_alloc_hb_set(void) |
| 1619 | { |
| 1620 | struct o2hb_heartbeat_group *hs = NULL; |
| 1621 | struct config_group *ret = NULL; |
| 1622 | |
Robert P. J. Day | cd86128 | 2006-12-13 00:34:52 -0800 | [diff] [blame] | 1623 | hs = kzalloc(sizeof(struct o2hb_heartbeat_group), GFP_KERNEL); |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1624 | if (hs == NULL) |
| 1625 | goto out; |
| 1626 | |
| 1627 | config_group_init_type_name(&hs->hs_group, "heartbeat", |
| 1628 | &o2hb_heartbeat_group_type); |
| 1629 | |
| 1630 | ret = &hs->hs_group; |
| 1631 | out: |
| 1632 | if (ret == NULL) |
| 1633 | kfree(hs); |
| 1634 | return ret; |
| 1635 | } |
| 1636 | |
| 1637 | void o2hb_free_hb_set(struct config_group *group) |
| 1638 | { |
| 1639 | struct o2hb_heartbeat_group *hs = to_o2hb_heartbeat_group(group); |
| 1640 | kfree(hs); |
| 1641 | } |
| 1642 | |
| 1643 | /* hb callback registration and issueing */ |
| 1644 | |
| 1645 | static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type) |
| 1646 | { |
| 1647 | if (type == O2HB_NUM_CB) |
| 1648 | return ERR_PTR(-EINVAL); |
| 1649 | |
| 1650 | return &o2hb_callbacks[type]; |
| 1651 | } |
| 1652 | |
| 1653 | void o2hb_setup_callback(struct o2hb_callback_func *hc, |
| 1654 | enum o2hb_callback_type type, |
| 1655 | o2hb_cb_func *func, |
| 1656 | void *data, |
| 1657 | int priority) |
| 1658 | { |
| 1659 | INIT_LIST_HEAD(&hc->hc_item); |
| 1660 | hc->hc_func = func; |
| 1661 | hc->hc_data = data; |
| 1662 | hc->hc_priority = priority; |
| 1663 | hc->hc_type = type; |
| 1664 | hc->hc_magic = O2HB_CB_MAGIC; |
| 1665 | } |
| 1666 | EXPORT_SYMBOL_GPL(o2hb_setup_callback); |
| 1667 | |
| 1668 | int o2hb_register_callback(struct o2hb_callback_func *hc) |
| 1669 | { |
| 1670 | struct o2hb_callback_func *tmp; |
| 1671 | struct list_head *iter; |
| 1672 | struct o2hb_callback *hbcall; |
| 1673 | int ret; |
| 1674 | |
| 1675 | BUG_ON(hc->hc_magic != O2HB_CB_MAGIC); |
| 1676 | BUG_ON(!list_empty(&hc->hc_item)); |
| 1677 | |
| 1678 | hbcall = hbcall_from_type(hc->hc_type); |
| 1679 | if (IS_ERR(hbcall)) { |
| 1680 | ret = PTR_ERR(hbcall); |
| 1681 | goto out; |
| 1682 | } |
| 1683 | |
| 1684 | down_write(&o2hb_callback_sem); |
| 1685 | |
| 1686 | list_for_each(iter, &hbcall->list) { |
| 1687 | tmp = list_entry(iter, struct o2hb_callback_func, hc_item); |
| 1688 | if (hc->hc_priority < tmp->hc_priority) { |
| 1689 | list_add_tail(&hc->hc_item, iter); |
| 1690 | break; |
| 1691 | } |
| 1692 | } |
| 1693 | if (list_empty(&hc->hc_item)) |
| 1694 | list_add_tail(&hc->hc_item, &hbcall->list); |
| 1695 | |
| 1696 | up_write(&o2hb_callback_sem); |
| 1697 | ret = 0; |
| 1698 | out: |
| 1699 | mlog(ML_HEARTBEAT, "returning %d on behalf of %p for funcs %p\n", |
| 1700 | ret, __builtin_return_address(0), hc); |
| 1701 | return ret; |
| 1702 | } |
| 1703 | EXPORT_SYMBOL_GPL(o2hb_register_callback); |
| 1704 | |
Joel Becker | c24f72c | 2007-02-03 03:14:30 -0800 | [diff] [blame] | 1705 | void o2hb_unregister_callback(struct o2hb_callback_func *hc) |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1706 | { |
| 1707 | BUG_ON(hc->hc_magic != O2HB_CB_MAGIC); |
| 1708 | |
| 1709 | mlog(ML_HEARTBEAT, "on behalf of %p for funcs %p\n", |
| 1710 | __builtin_return_address(0), hc); |
| 1711 | |
| 1712 | if (list_empty(&hc->hc_item)) |
Joel Becker | c24f72c | 2007-02-03 03:14:30 -0800 | [diff] [blame] | 1713 | return; |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1714 | |
| 1715 | down_write(&o2hb_callback_sem); |
| 1716 | |
| 1717 | list_del_init(&hc->hc_item); |
| 1718 | |
| 1719 | up_write(&o2hb_callback_sem); |
Mark Fasheh | a7f6a5f | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1720 | } |
| 1721 | EXPORT_SYMBOL_GPL(o2hb_unregister_callback); |
| 1722 | |
| 1723 | int o2hb_check_node_heartbeating(u8 node_num) |
| 1724 | { |
| 1725 | unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; |
| 1726 | |
| 1727 | o2hb_fill_node_map(testing_map, sizeof(testing_map)); |
| 1728 | if (!test_bit(node_num, testing_map)) { |
| 1729 | mlog(ML_HEARTBEAT, |
| 1730 | "node (%u) does not have heartbeating enabled.\n", |
| 1731 | node_num); |
| 1732 | return 0; |
| 1733 | } |
| 1734 | |
| 1735 | return 1; |
| 1736 | } |
| 1737 | EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating); |
| 1738 | |
| 1739 | int o2hb_check_node_heartbeating_from_callback(u8 node_num) |
| 1740 | { |
| 1741 | unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; |
| 1742 | |
| 1743 | o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map)); |
| 1744 | if (!test_bit(node_num, testing_map)) { |
| 1745 | mlog(ML_HEARTBEAT, |
| 1746 | "node (%u) does not have heartbeating enabled.\n", |
| 1747 | node_num); |
| 1748 | return 0; |
| 1749 | } |
| 1750 | |
| 1751 | return 1; |
| 1752 | } |
| 1753 | EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback); |
| 1754 | |
| 1755 | /* Makes sure our local node is configured with a node number, and is |
| 1756 | * heartbeating. */ |
| 1757 | int o2hb_check_local_node_heartbeating(void) |
| 1758 | { |
| 1759 | u8 node_num; |
| 1760 | |
| 1761 | /* if this node was set then we have networking */ |
| 1762 | node_num = o2nm_this_node(); |
| 1763 | if (node_num == O2NM_MAX_NODES) { |
| 1764 | mlog(ML_HEARTBEAT, "this node has not been configured.\n"); |
| 1765 | return 0; |
| 1766 | } |
| 1767 | |
| 1768 | return o2hb_check_node_heartbeating(node_num); |
| 1769 | } |
| 1770 | EXPORT_SYMBOL_GPL(o2hb_check_local_node_heartbeating); |
| 1771 | |
| 1772 | /* |
| 1773 | * this is just a hack until we get the plumbing which flips file systems |
| 1774 | * read only and drops the hb ref instead of killing the node dead. |
| 1775 | */ |
| 1776 | void o2hb_stop_all_regions(void) |
| 1777 | { |
| 1778 | struct o2hb_region *reg; |
| 1779 | |
| 1780 | mlog(ML_ERROR, "stopping heartbeat on all active regions.\n"); |
| 1781 | |
| 1782 | spin_lock(&o2hb_live_lock); |
| 1783 | |
| 1784 | list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) |
| 1785 | reg->hr_unclean_stop = 1; |
| 1786 | |
| 1787 | spin_unlock(&o2hb_live_lock); |
| 1788 | } |
| 1789 | EXPORT_SYMBOL_GPL(o2hb_stop_all_regions); |