| 1 |
// SPDX-License-Identifier: GPL-2.0-only |
| 2 |
/****************************************************************************** |
| 3 |
******************************************************************************* |
| 4 |
** |
| 5 |
** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. |
| 6 |
** Copyright (C) 2004-2011 Red Hat, Inc. All rights reserved. |
| 7 |
** |
| 8 |
** |
| 9 |
******************************************************************************* |
| 10 |
******************************************************************************/ |
| 11 |
|
| 12 |
#include "dlm_internal.h" |
| 13 |
#include "lockspace.h" |
| 14 |
#include "member.h" |
| 15 |
#include "dir.h" |
| 16 |
#include "ast.h" |
| 17 |
#include "recover.h" |
| 18 |
#include "lowcomms.h" |
| 19 |
#include "lock.h" |
| 20 |
#include "requestqueue.h" |
| 21 |
#include "recoverd.h" |
| 22 |
|
| 23 |
static int dlm_create_masters_list(struct dlm_ls *ls) |
| 24 |
{ |
| 25 |
struct dlm_rsb *r; |
| 26 |
int error = 0; |
| 27 |
|
| 28 |
write_lock_bh(&ls->ls_masters_lock); |
| 29 |
if (!list_empty(&ls->ls_masters_list)) { |
| 30 |
log_error(ls, "root list not empty"); |
| 31 |
error = -EINVAL; |
| 32 |
goto out; |
| 33 |
} |
| 34 |
|
| 35 |
read_lock_bh(&ls->ls_rsbtbl_lock); |
| 36 |
list_for_each_entry(r, &ls->ls_slow_active, res_slow_list) { |
| 37 |
if (r->res_nodeid) |
| 38 |
continue; |
| 39 |
|
| 40 |
list_add(&r->res_masters_list, &ls->ls_masters_list); |
| 41 |
dlm_hold_rsb(r); |
| 42 |
} |
| 43 |
read_unlock_bh(&ls->ls_rsbtbl_lock); |
| 44 |
out: |
| 45 |
write_unlock_bh(&ls->ls_masters_lock); |
| 46 |
return error; |
| 47 |
} |
| 48 |
|
| 49 |
static void dlm_release_masters_list(struct dlm_ls *ls) |
| 50 |
{ |
| 51 |
struct dlm_rsb *r, *safe; |
| 52 |
|
| 53 |
write_lock_bh(&ls->ls_masters_lock); |
| 54 |
list_for_each_entry_safe(r, safe, &ls->ls_masters_list, res_masters_list) { |
| 55 |
list_del_init(&r->res_masters_list); |
| 56 |
dlm_put_rsb(r); |
| 57 |
} |
| 58 |
write_unlock_bh(&ls->ls_masters_lock); |
| 59 |
} |
| 60 |
|
| 61 |
static void dlm_create_root_list(struct dlm_ls *ls, struct list_head *root_list) |
| 62 |
{ |
| 63 |
struct dlm_rsb *r; |
| 64 |
|
| 65 |
read_lock_bh(&ls->ls_rsbtbl_lock); |
| 66 |
list_for_each_entry(r, &ls->ls_slow_active, res_slow_list) { |
| 67 |
list_add(&r->res_root_list, root_list); |
| 68 |
dlm_hold_rsb(r); |
| 69 |
} |
| 70 |
|
| 71 |
WARN_ON_ONCE(!list_empty(&ls->ls_slow_inactive)); |
| 72 |
read_unlock_bh(&ls->ls_rsbtbl_lock); |
| 73 |
} |
| 74 |
|
| 75 |
static void dlm_release_root_list(struct list_head *root_list) |
| 76 |
{ |
| 77 |
struct dlm_rsb *r, *safe; |
| 78 |
|
| 79 |
list_for_each_entry_safe(r, safe, root_list, res_root_list) { |
| 80 |
list_del_init(&r->res_root_list); |
| 81 |
dlm_put_rsb(r); |
| 82 |
} |
| 83 |
} |
| 84 |
|
| 85 |
/* If the start for which we're re-enabling locking (seq) has been superseded |
| 86 |
by a newer stop (ls_recover_seq), we need to leave locking disabled. |
| 87 |
|
| 88 |
We suspend dlm_recv threads here to avoid the race where dlm_recv a) sees |
| 89 |
locking stopped and b) adds a message to the requestqueue, but dlm_recoverd |
| 90 |
enables locking and clears the requestqueue between a and b. */ |
| 91 |
|
| 92 |
static int enable_locking(struct dlm_ls *ls, uint64_t seq) |
| 93 |
{ |
| 94 |
int error = -EINTR; |
| 95 |
|
| 96 |
write_lock_bh(&ls->ls_recv_active); |
| 97 |
|
| 98 |
spin_lock_bh(&ls->ls_recover_lock); |
| 99 |
if (ls->ls_recover_seq == seq) { |
| 100 |
set_bit(LSFL_RUNNING, &ls->ls_flags); |
| 101 |
/* Schedule next timer if recovery put something on inactive. |
| 102 |
* |
| 103 |
* The rsbs that was queued while recovery on toss hasn't |
| 104 |
* started yet because LSFL_RUNNING was set everything |
| 105 |
* else recovery hasn't started as well because ls_in_recovery |
| 106 |
* is still hold. So we should not run into the case that |
| 107 |
* resume_scan_timer() queues a timer that can occur in |
| 108 |
* a no op. |
| 109 |
*/ |
| 110 |
resume_scan_timer(ls); |
| 111 |
/* unblocks processes waiting to enter the dlm */ |
| 112 |
up_write(&ls->ls_in_recovery); |
| 113 |
clear_bit(LSFL_RECOVER_LOCK, &ls->ls_flags); |
| 114 |
error = 0; |
| 115 |
} |
| 116 |
spin_unlock_bh(&ls->ls_recover_lock); |
| 117 |
|
| 118 |
write_unlock_bh(&ls->ls_recv_active); |
| 119 |
return error; |
| 120 |
} |
| 121 |
|
| 122 |
static int ls_recover(struct dlm_ls *ls, struct dlm_recover *rv) |
| 123 |
{ |
| 124 |
LIST_HEAD(root_list); |
| 125 |
unsigned long start; |
| 126 |
int error, neg = 0; |
| 127 |
|
| 128 |
log_rinfo(ls, "dlm_recover %llu", (unsigned long long)rv->seq); |
| 129 |
|
| 130 |
mutex_lock(&ls->ls_recoverd_active); |
| 131 |
|
| 132 |
dlm_callback_suspend(ls); |
| 133 |
|
| 134 |
dlm_clear_inactive(ls); |
| 135 |
|
| 136 |
/* |
| 137 |
* This list of root rsb's will be the basis of most of the recovery |
| 138 |
* routines. |
| 139 |
*/ |
| 140 |
|
| 141 |
dlm_create_root_list(ls, &root_list); |
| 142 |
|
| 143 |
/* |
| 144 |
* Add or remove nodes from the lockspace's ls_nodes list. |
| 145 |
* |
| 146 |
* Due to the fact that we must report all membership changes to lsops |
| 147 |
* or midcomms layer, it is not permitted to abort ls_recover() until |
| 148 |
* this is done. |
| 149 |
*/ |
| 150 |
|
| 151 |
error = dlm_recover_members(ls, rv, &neg); |
| 152 |
if (error) { |
| 153 |
log_rinfo(ls, "dlm_recover_members error %d", error); |
| 154 |
goto fail_root_list; |
| 155 |
} |
| 156 |
|
| 157 |
dlm_recover_dir_nodeid(ls, &root_list); |
| 158 |
|
| 159 |
/* Create a snapshot of all active rsbs were we are the master of. |
| 160 |
* During the barrier between dlm_recover_members_wait() and |
| 161 |
* dlm_recover_directory() other nodes can dump their necessary |
| 162 |
* directory dlm_rsb (r->res_dir_nodeid == nodeid) in rcom |
| 163 |
* communication dlm_copy_master_names() handling. |
| 164 |
* |
| 165 |
* TODO We should create a per lockspace list that contains rsbs |
| 166 |
* that we are the master of. Instead of creating this list while |
| 167 |
* recovery we keep track of those rsbs while locking handling and |
| 168 |
* recovery can use it when necessary. |
| 169 |
*/ |
| 170 |
error = dlm_create_masters_list(ls); |
| 171 |
if (error) { |
| 172 |
log_rinfo(ls, "dlm_create_masters_list error %d", error); |
| 173 |
goto fail_root_list; |
| 174 |
} |
| 175 |
|
| 176 |
ls->ls_recover_locks_in = 0; |
| 177 |
|
| 178 |
dlm_set_recover_status(ls, DLM_RS_NODES); |
| 179 |
|
| 180 |
error = dlm_recover_members_wait(ls, rv->seq); |
| 181 |
if (error) { |
| 182 |
log_rinfo(ls, "dlm_recover_members_wait error %d", error); |
| 183 |
dlm_release_masters_list(ls); |
| 184 |
goto fail_root_list; |
| 185 |
} |
| 186 |
|
| 187 |
start = jiffies; |
| 188 |
|
| 189 |
/* |
| 190 |
* Rebuild our own share of the directory by collecting from all other |
| 191 |
* nodes their master rsb names that hash to us. |
| 192 |
*/ |
| 193 |
|
| 194 |
error = dlm_recover_directory(ls, rv->seq); |
| 195 |
if (error) { |
| 196 |
log_rinfo(ls, "dlm_recover_directory error %d", error); |
| 197 |
dlm_release_masters_list(ls); |
| 198 |
goto fail_root_list; |
| 199 |
} |
| 200 |
|
| 201 |
dlm_set_recover_status(ls, DLM_RS_DIR); |
| 202 |
|
| 203 |
error = dlm_recover_directory_wait(ls, rv->seq); |
| 204 |
if (error) { |
| 205 |
log_rinfo(ls, "dlm_recover_directory_wait error %d", error); |
| 206 |
dlm_release_masters_list(ls); |
| 207 |
goto fail_root_list; |
| 208 |
} |
| 209 |
|
| 210 |
dlm_release_masters_list(ls); |
| 211 |
|
| 212 |
/* |
| 213 |
* We may have outstanding operations that are waiting for a reply from |
| 214 |
* a failed node. Mark these to be resent after recovery. Unlock and |
| 215 |
* cancel ops can just be completed. |
| 216 |
*/ |
| 217 |
|
| 218 |
dlm_recover_waiters_pre(ls); |
| 219 |
|
| 220 |
if (dlm_recovery_stopped(ls)) { |
| 221 |
error = -EINTR; |
| 222 |
goto fail_root_list; |
| 223 |
} |
| 224 |
|
| 225 |
if (neg || dlm_no_directory(ls)) { |
| 226 |
/* |
| 227 |
* Clear lkb's for departed nodes. |
| 228 |
*/ |
| 229 |
|
| 230 |
dlm_recover_purge(ls, &root_list); |
| 231 |
|
| 232 |
/* |
| 233 |
* Get new master nodeid's for rsb's that were mastered on |
| 234 |
* departed nodes. |
| 235 |
*/ |
| 236 |
|
| 237 |
error = dlm_recover_masters(ls, rv->seq, &root_list); |
| 238 |
if (error) { |
| 239 |
log_rinfo(ls, "dlm_recover_masters error %d", error); |
| 240 |
goto fail_root_list; |
| 241 |
} |
| 242 |
|
| 243 |
/* |
| 244 |
* Send our locks on remastered rsb's to the new masters. |
| 245 |
*/ |
| 246 |
|
| 247 |
error = dlm_recover_locks(ls, rv->seq, &root_list); |
| 248 |
if (error) { |
| 249 |
log_rinfo(ls, "dlm_recover_locks error %d", error); |
| 250 |
goto fail_root_list; |
| 251 |
} |
| 252 |
|
| 253 |
dlm_set_recover_status(ls, DLM_RS_LOCKS); |
| 254 |
|
| 255 |
error = dlm_recover_locks_wait(ls, rv->seq); |
| 256 |
if (error) { |
| 257 |
log_rinfo(ls, "dlm_recover_locks_wait error %d", error); |
| 258 |
goto fail_root_list; |
| 259 |
} |
| 260 |
|
| 261 |
log_rinfo(ls, "dlm_recover_locks %u in", |
| 262 |
ls->ls_recover_locks_in); |
| 263 |
|
| 264 |
/* |
| 265 |
* Finalize state in master rsb's now that all locks can be |
| 266 |
* checked. This includes conversion resolution and lvb |
| 267 |
* settings. |
| 268 |
*/ |
| 269 |
|
| 270 |
dlm_recover_rsbs(ls, &root_list); |
| 271 |
} else { |
| 272 |
/* |
| 273 |
* Other lockspace members may be going through the "neg" steps |
| 274 |
* while also adding us to the lockspace, in which case they'll |
| 275 |
* be doing the recover_locks (RS_LOCKS) barrier. |
| 276 |
*/ |
| 277 |
dlm_set_recover_status(ls, DLM_RS_LOCKS); |
| 278 |
|
| 279 |
error = dlm_recover_locks_wait(ls, rv->seq); |
| 280 |
if (error) { |
| 281 |
log_rinfo(ls, "dlm_recover_locks_wait error %d", error); |
| 282 |
goto fail_root_list; |
| 283 |
} |
| 284 |
} |
| 285 |
|
| 286 |
dlm_release_root_list(&root_list); |
| 287 |
|
| 288 |
/* |
| 289 |
* Purge directory-related requests that are saved in requestqueue. |
| 290 |
* All dir requests from before recovery are invalid now due to the dir |
| 291 |
* rebuild and will be resent by the requesting nodes. |
| 292 |
*/ |
| 293 |
|
| 294 |
dlm_purge_requestqueue(ls); |
| 295 |
|
| 296 |
dlm_set_recover_status(ls, DLM_RS_DONE); |
| 297 |
|
| 298 |
error = dlm_recover_done_wait(ls, rv->seq); |
| 299 |
if (error) { |
| 300 |
log_rinfo(ls, "dlm_recover_done_wait error %d", error); |
| 301 |
goto fail; |
| 302 |
} |
| 303 |
|
| 304 |
dlm_clear_members_gone(ls); |
| 305 |
|
| 306 |
dlm_callback_resume(ls); |
| 307 |
|
| 308 |
error = enable_locking(ls, rv->seq); |
| 309 |
if (error) { |
| 310 |
log_rinfo(ls, "enable_locking error %d", error); |
| 311 |
goto fail; |
| 312 |
} |
| 313 |
|
| 314 |
error = dlm_process_requestqueue(ls); |
| 315 |
if (error) { |
| 316 |
log_rinfo(ls, "dlm_process_requestqueue error %d", error); |
| 317 |
goto fail; |
| 318 |
} |
| 319 |
|
| 320 |
error = dlm_recover_waiters_post(ls); |
| 321 |
if (error) { |
| 322 |
log_rinfo(ls, "dlm_recover_waiters_post error %d", error); |
| 323 |
goto fail; |
| 324 |
} |
| 325 |
|
| 326 |
dlm_recover_grant(ls); |
| 327 |
|
| 328 |
log_rinfo(ls, "dlm_recover %llu generation %u done: %u ms", |
| 329 |
(unsigned long long)rv->seq, ls->ls_generation, |
| 330 |
jiffies_to_msecs(jiffies - start)); |
| 331 |
mutex_unlock(&ls->ls_recoverd_active); |
| 332 |
|
| 333 |
return 0; |
| 334 |
|
| 335 |
fail_root_list: |
| 336 |
dlm_release_root_list(&root_list); |
| 337 |
fail: |
| 338 |
mutex_unlock(&ls->ls_recoverd_active); |
| 339 |
|
| 340 |
return error; |
| 341 |
} |
| 342 |
|
| 343 |
/* The dlm_ls_start() that created the rv we take here may already have been |
| 344 |
stopped via dlm_ls_stop(); in that case we need to leave the RECOVERY_STOP |
| 345 |
flag set. */ |
| 346 |
|
| 347 |
static void do_ls_recovery(struct dlm_ls *ls) |
| 348 |
{ |
| 349 |
struct dlm_recover *rv = NULL; |
| 350 |
int error; |
| 351 |
|
| 352 |
spin_lock_bh(&ls->ls_recover_lock); |
| 353 |
rv = ls->ls_recover_args; |
| 354 |
ls->ls_recover_args = NULL; |
| 355 |
if (rv && ls->ls_recover_seq == rv->seq) |
| 356 |
clear_bit(LSFL_RECOVER_STOP, &ls->ls_flags); |
| 357 |
spin_unlock_bh(&ls->ls_recover_lock); |
| 358 |
|
| 359 |
if (rv) { |
| 360 |
error = ls_recover(ls, rv); |
| 361 |
switch (error) { |
| 362 |
case 0: |
| 363 |
ls->ls_recovery_result = 0; |
| 364 |
complete(&ls->ls_recovery_done); |
| 365 |
|
| 366 |
dlm_lsop_recover_done(ls); |
| 367 |
break; |
| 368 |
case -EINTR: |
| 369 |
/* if recovery was interrupted -EINTR we wait for the next |
| 370 |
* ls_recover() iteration until it hopefully succeeds. |
| 371 |
*/ |
| 372 |
log_rinfo(ls, "%s %llu interrupted and should be queued to run again", |
| 373 |
__func__, (unsigned long long)rv->seq); |
| 374 |
break; |
| 375 |
default: |
| 376 |
log_rinfo(ls, "%s %llu error %d", __func__, |
| 377 |
(unsigned long long)rv->seq, error); |
| 378 |
|
| 379 |
/* let new_lockspace() get aware of critical error */ |
| 380 |
ls->ls_recovery_result = error; |
| 381 |
complete(&ls->ls_recovery_done); |
| 382 |
break; |
| 383 |
} |
| 384 |
|
| 385 |
kfree(rv->nodes); |
| 386 |
kfree(rv); |
| 387 |
} |
| 388 |
} |
| 389 |
|
| 390 |
static int dlm_recoverd(void *arg) |
| 391 |
{ |
| 392 |
struct dlm_ls *ls; |
| 393 |
|
| 394 |
ls = dlm_find_lockspace_local(arg); |
| 395 |
if (!ls) { |
| 396 |
log_print("dlm_recoverd: no lockspace %p", arg); |
| 397 |
return -1; |
| 398 |
} |
| 399 |
|
| 400 |
down_write(&ls->ls_in_recovery); |
| 401 |
set_bit(LSFL_RECOVER_LOCK, &ls->ls_flags); |
| 402 |
wake_up(&ls->ls_recover_lock_wait); |
| 403 |
|
| 404 |
while (1) { |
| 405 |
/* |
| 406 |
* We call kthread_should_stop() after set_current_state(). |
| 407 |
* This is because it works correctly if kthread_stop() is |
| 408 |
* called just before set_current_state(). |
| 409 |
*/ |
| 410 |
set_current_state(TASK_INTERRUPTIBLE); |
| 411 |
if (kthread_should_stop()) { |
| 412 |
set_current_state(TASK_RUNNING); |
| 413 |
break; |
| 414 |
} |
| 415 |
if (!test_bit(LSFL_RECOVER_WORK, &ls->ls_flags) && |
| 416 |
!test_bit(LSFL_RECOVER_DOWN, &ls->ls_flags)) { |
| 417 |
if (kthread_should_stop()) |
| 418 |
break; |
| 419 |
schedule(); |
| 420 |
} |
| 421 |
set_current_state(TASK_RUNNING); |
| 422 |
|
| 423 |
if (test_and_clear_bit(LSFL_RECOVER_DOWN, &ls->ls_flags)) { |
| 424 |
down_write(&ls->ls_in_recovery); |
| 425 |
set_bit(LSFL_RECOVER_LOCK, &ls->ls_flags); |
| 426 |
wake_up(&ls->ls_recover_lock_wait); |
| 427 |
} |
| 428 |
|
| 429 |
if (test_and_clear_bit(LSFL_RECOVER_WORK, &ls->ls_flags)) |
| 430 |
do_ls_recovery(ls); |
| 431 |
} |
| 432 |
|
| 433 |
if (test_bit(LSFL_RECOVER_LOCK, &ls->ls_flags)) |
| 434 |
up_write(&ls->ls_in_recovery); |
| 435 |
|
| 436 |
dlm_put_lockspace(ls); |
| 437 |
return 0; |
| 438 |
} |
| 439 |
|
| 440 |
int dlm_recoverd_start(struct dlm_ls *ls) |
| 441 |
{ |
| 442 |
struct task_struct *p; |
| 443 |
int error = 0; |
| 444 |
|
| 445 |
p = kthread_run(dlm_recoverd, ls, "dlm_recoverd"); |
| 446 |
if (IS_ERR(p)) |
| 447 |
error = PTR_ERR(p); |
| 448 |
else |
| 449 |
ls->ls_recoverd_task = p; |
| 450 |
return error; |
| 451 |
} |
| 452 |
|
| 453 |
void dlm_recoverd_stop(struct dlm_ls *ls) |
| 454 |
{ |
| 455 |
kthread_stop(ls->ls_recoverd_task); |
| 456 |
} |
| 457 |
|
| 458 |
void dlm_recoverd_suspend(struct dlm_ls *ls) |
| 459 |
{ |
| 460 |
wake_up(&ls->ls_wait_general); |
| 461 |
mutex_lock(&ls->ls_recoverd_active); |
| 462 |
} |
| 463 |
|
| 464 |
void dlm_recoverd_resume(struct dlm_ls *ls) |
| 465 |
{ |
| 466 |
mutex_unlock(&ls->ls_recoverd_active); |
| 467 |
} |
| 468 |
|