| 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 <linux/module.h> |
| 13 |
|
| 14 |
#include "dlm_internal.h" |
| 15 |
#include "lockspace.h" |
| 16 |
#include "member.h" |
| 17 |
#include "recoverd.h" |
| 18 |
#include "dir.h" |
| 19 |
#include "midcomms.h" |
| 20 |
#include "config.h" |
| 21 |
#include "memory.h" |
| 22 |
#include "lock.h" |
| 23 |
#include "recover.h" |
| 24 |
#include "requestqueue.h" |
| 25 |
#include "user.h" |
| 26 |
#include "ast.h" |
| 27 |
|
| 28 |
static int ls_count; |
| 29 |
static struct mutex ls_lock; |
| 30 |
static struct list_head lslist; |
| 31 |
static spinlock_t lslist_lock; |
| 32 |
|
| 33 |
static ssize_t dlm_control_store(struct dlm_ls *ls, const char *buf, size_t len) |
| 34 |
{ |
| 35 |
ssize_t ret = len; |
| 36 |
int n; |
| 37 |
int rc = kstrtoint(buf, 0, &n); |
| 38 |
|
| 39 |
if (rc) |
| 40 |
return rc; |
| 41 |
ls = dlm_find_lockspace_local(ls); |
| 42 |
if (!ls) |
| 43 |
return -EINVAL; |
| 44 |
|
| 45 |
switch (n) { |
| 46 |
case 0: |
| 47 |
dlm_ls_stop(ls); |
| 48 |
break; |
| 49 |
case 1: |
| 50 |
dlm_ls_start(ls); |
| 51 |
break; |
| 52 |
default: |
| 53 |
ret = -EINVAL; |
| 54 |
} |
| 55 |
dlm_put_lockspace(ls); |
| 56 |
return ret; |
| 57 |
} |
| 58 |
|
| 59 |
static ssize_t dlm_event_store(struct dlm_ls *ls, const char *buf, size_t len) |
| 60 |
{ |
| 61 |
int rc = kstrtoint(buf, 0, &ls->ls_uevent_result); |
| 62 |
|
| 63 |
if (rc) |
| 64 |
return rc; |
| 65 |
set_bit(LSFL_UEVENT_WAIT, &ls->ls_flags); |
| 66 |
wake_up(&ls->ls_uevent_wait); |
| 67 |
return len; |
| 68 |
} |
| 69 |
|
| 70 |
static ssize_t dlm_id_show(struct dlm_ls *ls, char *buf) |
| 71 |
{ |
| 72 |
return snprintf(buf, PAGE_SIZE, "%u\n", ls->ls_global_id); |
| 73 |
} |
| 74 |
|
| 75 |
static ssize_t dlm_id_store(struct dlm_ls *ls, const char *buf, size_t len) |
| 76 |
{ |
| 77 |
int rc = kstrtouint(buf, 0, &ls->ls_global_id); |
| 78 |
|
| 79 |
if (rc) |
| 80 |
return rc; |
| 81 |
return len; |
| 82 |
} |
| 83 |
|
| 84 |
static ssize_t dlm_nodir_show(struct dlm_ls *ls, char *buf) |
| 85 |
{ |
| 86 |
return snprintf(buf, PAGE_SIZE, "%u\n", dlm_no_directory(ls)); |
| 87 |
} |
| 88 |
|
| 89 |
static ssize_t dlm_nodir_store(struct dlm_ls *ls, const char *buf, size_t len) |
| 90 |
{ |
| 91 |
int val; |
| 92 |
int rc = kstrtoint(buf, 0, &val); |
| 93 |
|
| 94 |
if (rc) |
| 95 |
return rc; |
| 96 |
if (val == 1) |
| 97 |
set_bit(LSFL_NODIR, &ls->ls_flags); |
| 98 |
return len; |
| 99 |
} |
| 100 |
|
| 101 |
static ssize_t dlm_recover_status_show(struct dlm_ls *ls, char *buf) |
| 102 |
{ |
| 103 |
uint32_t status = dlm_recover_status(ls); |
| 104 |
return snprintf(buf, PAGE_SIZE, "%x\n", status); |
| 105 |
} |
| 106 |
|
| 107 |
static ssize_t dlm_recover_nodeid_show(struct dlm_ls *ls, char *buf) |
| 108 |
{ |
| 109 |
return snprintf(buf, PAGE_SIZE, "%d\n", ls->ls_recover_nodeid); |
| 110 |
} |
| 111 |
|
| 112 |
struct dlm_attr { |
| 113 |
struct attribute attr; |
| 114 |
ssize_t (*show)(struct dlm_ls *, char *); |
| 115 |
ssize_t (*store)(struct dlm_ls *, const char *, size_t); |
| 116 |
}; |
| 117 |
|
| 118 |
static struct dlm_attr dlm_attr_control = { |
| 119 |
.attr = {.name = "control", .mode = S_IWUSR}, |
| 120 |
.store = dlm_control_store |
| 121 |
}; |
| 122 |
|
| 123 |
static struct dlm_attr dlm_attr_event = { |
| 124 |
.attr = {.name = "event_done", .mode = S_IWUSR}, |
| 125 |
.store = dlm_event_store |
| 126 |
}; |
| 127 |
|
| 128 |
static struct dlm_attr dlm_attr_id = { |
| 129 |
.attr = {.name = "id", .mode = S_IRUGO | S_IWUSR}, |
| 130 |
.show = dlm_id_show, |
| 131 |
.store = dlm_id_store |
| 132 |
}; |
| 133 |
|
| 134 |
static struct dlm_attr dlm_attr_nodir = { |
| 135 |
.attr = {.name = "nodir", .mode = S_IRUGO | S_IWUSR}, |
| 136 |
.show = dlm_nodir_show, |
| 137 |
.store = dlm_nodir_store |
| 138 |
}; |
| 139 |
|
| 140 |
static struct dlm_attr dlm_attr_recover_status = { |
| 141 |
.attr = {.name = "recover_status", .mode = S_IRUGO}, |
| 142 |
.show = dlm_recover_status_show |
| 143 |
}; |
| 144 |
|
| 145 |
static struct dlm_attr dlm_attr_recover_nodeid = { |
| 146 |
.attr = {.name = "recover_nodeid", .mode = S_IRUGO}, |
| 147 |
.show = dlm_recover_nodeid_show |
| 148 |
}; |
| 149 |
|
| 150 |
static struct attribute *dlm_attrs[] = { |
| 151 |
&dlm_attr_control.attr, |
| 152 |
&dlm_attr_event.attr, |
| 153 |
&dlm_attr_id.attr, |
| 154 |
&dlm_attr_nodir.attr, |
| 155 |
&dlm_attr_recover_status.attr, |
| 156 |
&dlm_attr_recover_nodeid.attr, |
| 157 |
NULL, |
| 158 |
}; |
| 159 |
ATTRIBUTE_GROUPS(dlm); |
| 160 |
|
| 161 |
static ssize_t dlm_attr_show(struct kobject *kobj, struct attribute *attr, |
| 162 |
char *buf) |
| 163 |
{ |
| 164 |
struct dlm_ls *ls = container_of(kobj, struct dlm_ls, ls_kobj); |
| 165 |
struct dlm_attr *a = container_of(attr, struct dlm_attr, attr); |
| 166 |
return a->show ? a->show(ls, buf) : 0; |
| 167 |
} |
| 168 |
|
| 169 |
static ssize_t dlm_attr_store(struct kobject *kobj, struct attribute *attr, |
| 170 |
const char *buf, size_t len) |
| 171 |
{ |
| 172 |
struct dlm_ls *ls = container_of(kobj, struct dlm_ls, ls_kobj); |
| 173 |
struct dlm_attr *a = container_of(attr, struct dlm_attr, attr); |
| 174 |
return a->store ? a->store(ls, buf, len) : len; |
| 175 |
} |
| 176 |
|
| 177 |
static const struct sysfs_ops dlm_attr_ops = { |
| 178 |
.show = dlm_attr_show, |
| 179 |
.store = dlm_attr_store, |
| 180 |
}; |
| 181 |
|
| 182 |
static struct kobj_type dlm_ktype = { |
| 183 |
.default_groups = dlm_groups, |
| 184 |
.sysfs_ops = &dlm_attr_ops, |
| 185 |
}; |
| 186 |
|
| 187 |
static struct kset *dlm_kset; |
| 188 |
|
| 189 |
static int do_uevent(struct dlm_ls *ls, int in, unsigned int release_recover) |
| 190 |
{ |
| 191 |
char message[512] = {}; |
| 192 |
char *envp[] = { message, NULL }; |
| 193 |
|
| 194 |
if (in) { |
| 195 |
kobject_uevent(&ls->ls_kobj, KOBJ_ONLINE); |
| 196 |
} else { |
| 197 |
snprintf(message, 511, "RELEASE_RECOVER=%u", release_recover); |
| 198 |
kobject_uevent_env(&ls->ls_kobj, KOBJ_OFFLINE, envp); |
| 199 |
} |
| 200 |
|
| 201 |
log_rinfo(ls, "%s the lockspace group...", in ? "joining" : "leaving"); |
| 202 |
|
| 203 |
/* dlm_controld will see the uevent, do the necessary group management |
| 204 |
and then write to sysfs to wake us */ |
| 205 |
|
| 206 |
wait_event(ls->ls_uevent_wait, |
| 207 |
test_and_clear_bit(LSFL_UEVENT_WAIT, &ls->ls_flags)); |
| 208 |
|
| 209 |
log_rinfo(ls, "group event done %d", ls->ls_uevent_result); |
| 210 |
|
| 211 |
return ls->ls_uevent_result; |
| 212 |
} |
| 213 |
|
| 214 |
static int dlm_uevent(const struct kobject *kobj, struct kobj_uevent_env *env) |
| 215 |
{ |
| 216 |
const struct dlm_ls *ls = container_of(kobj, struct dlm_ls, ls_kobj); |
| 217 |
|
| 218 |
add_uevent_var(env, "LOCKSPACE=%s", ls->ls_name); |
| 219 |
return 0; |
| 220 |
} |
| 221 |
|
| 222 |
static const struct kset_uevent_ops dlm_uevent_ops = { |
| 223 |
.uevent = dlm_uevent, |
| 224 |
}; |
| 225 |
|
| 226 |
int __init dlm_lockspace_init(void) |
| 227 |
{ |
| 228 |
ls_count = 0; |
| 229 |
mutex_init(&ls_lock); |
| 230 |
INIT_LIST_HEAD(&lslist); |
| 231 |
spin_lock_init(&lslist_lock); |
| 232 |
|
| 233 |
dlm_kset = kset_create_and_add("dlm", &dlm_uevent_ops, kernel_kobj); |
| 234 |
if (!dlm_kset) { |
| 235 |
printk(KERN_WARNING "%s: can not create kset\n", __func__); |
| 236 |
return -ENOMEM; |
| 237 |
} |
| 238 |
return 0; |
| 239 |
} |
| 240 |
|
| 241 |
void dlm_lockspace_exit(void) |
| 242 |
{ |
| 243 |
kset_unregister(dlm_kset); |
| 244 |
} |
| 245 |
|
| 246 |
struct dlm_ls *dlm_find_lockspace_global(uint32_t id) |
| 247 |
{ |
| 248 |
struct dlm_ls *ls; |
| 249 |
|
| 250 |
spin_lock_bh(&lslist_lock); |
| 251 |
|
| 252 |
list_for_each_entry(ls, &lslist, ls_list) { |
| 253 |
if (ls->ls_global_id == id) { |
| 254 |
atomic_inc(&ls->ls_count); |
| 255 |
goto out; |
| 256 |
} |
| 257 |
} |
| 258 |
ls = NULL; |
| 259 |
out: |
| 260 |
spin_unlock_bh(&lslist_lock); |
| 261 |
return ls; |
| 262 |
} |
| 263 |
|
| 264 |
struct dlm_ls *dlm_find_lockspace_local(dlm_lockspace_t *lockspace) |
| 265 |
{ |
| 266 |
struct dlm_ls *ls = lockspace; |
| 267 |
|
| 268 |
atomic_inc(&ls->ls_count); |
| 269 |
return ls; |
| 270 |
} |
| 271 |
|
| 272 |
struct dlm_ls *dlm_find_lockspace_device(int minor) |
| 273 |
{ |
| 274 |
struct dlm_ls *ls; |
| 275 |
|
| 276 |
spin_lock_bh(&lslist_lock); |
| 277 |
list_for_each_entry(ls, &lslist, ls_list) { |
| 278 |
if (ls->ls_device.minor == minor) { |
| 279 |
atomic_inc(&ls->ls_count); |
| 280 |
goto out; |
| 281 |
} |
| 282 |
} |
| 283 |
ls = NULL; |
| 284 |
out: |
| 285 |
spin_unlock_bh(&lslist_lock); |
| 286 |
return ls; |
| 287 |
} |
| 288 |
|
| 289 |
void dlm_put_lockspace(struct dlm_ls *ls) |
| 290 |
{ |
| 291 |
if (atomic_dec_and_test(&ls->ls_count)) |
| 292 |
wake_up(&ls->ls_count_wait); |
| 293 |
} |
| 294 |
|
| 295 |
static void remove_lockspace(struct dlm_ls *ls) |
| 296 |
{ |
| 297 |
retry: |
| 298 |
wait_event(ls->ls_count_wait, atomic_read(&ls->ls_count) == 0); |
| 299 |
|
| 300 |
spin_lock_bh(&lslist_lock); |
| 301 |
if (atomic_read(&ls->ls_count) != 0) { |
| 302 |
spin_unlock_bh(&lslist_lock); |
| 303 |
goto retry; |
| 304 |
} |
| 305 |
|
| 306 |
WARN_ON(ls->ls_create_count != 0); |
| 307 |
list_del(&ls->ls_list); |
| 308 |
spin_unlock_bh(&lslist_lock); |
| 309 |
} |
| 310 |
|
| 311 |
static int threads_start(void) |
| 312 |
{ |
| 313 |
int error; |
| 314 |
|
| 315 |
/* Thread for sending/receiving messages for all lockspace's */ |
| 316 |
error = dlm_midcomms_start(); |
| 317 |
if (error) |
| 318 |
log_print("cannot start dlm midcomms %d", error); |
| 319 |
|
| 320 |
return error; |
| 321 |
} |
| 322 |
|
| 323 |
static int lkb_idr_free(struct dlm_lkb *lkb) |
| 324 |
{ |
| 325 |
if (lkb->lkb_lvbptr && test_bit(DLM_IFL_MSTCPY_BIT, &lkb->lkb_iflags)) |
| 326 |
dlm_free_lvb(lkb->lkb_lvbptr); |
| 327 |
|
| 328 |
dlm_free_lkb(lkb); |
| 329 |
return 0; |
| 330 |
} |
| 331 |
|
| 332 |
static void rhash_free_rsb(void *ptr, void *arg) |
| 333 |
{ |
| 334 |
struct dlm_rsb *rsb = ptr; |
| 335 |
|
| 336 |
dlm_free_rsb(rsb); |
| 337 |
} |
| 338 |
|
| 339 |
static void free_lockspace(struct work_struct *work) |
| 340 |
{ |
| 341 |
struct dlm_ls *ls = container_of(work, struct dlm_ls, ls_free_work); |
| 342 |
struct dlm_lkb *lkb; |
| 343 |
unsigned long id; |
| 344 |
|
| 345 |
/* |
| 346 |
* Free all lkb's in xa |
| 347 |
*/ |
| 348 |
xa_for_each(&ls->ls_lkbxa, id, lkb) { |
| 349 |
lkb_idr_free(lkb); |
| 350 |
} |
| 351 |
xa_destroy(&ls->ls_lkbxa); |
| 352 |
|
| 353 |
/* |
| 354 |
* Free all rsb's on rsbtbl |
| 355 |
*/ |
| 356 |
rhashtable_free_and_destroy(&ls->ls_rsbtbl, rhash_free_rsb, NULL); |
| 357 |
|
| 358 |
kfree(ls); |
| 359 |
} |
| 360 |
|
| 361 |
static int new_lockspace(const char *name, const char *cluster, |
| 362 |
uint32_t flags, int lvblen, |
| 363 |
const struct dlm_lockspace_ops *ops, void *ops_arg, |
| 364 |
int *ops_result, dlm_lockspace_t **lockspace) |
| 365 |
{ |
| 366 |
struct dlm_ls *ls; |
| 367 |
int namelen = strlen(name); |
| 368 |
int error; |
| 369 |
|
| 370 |
if (namelen > DLM_LOCKSPACE_LEN || namelen == 0) |
| 371 |
return -EINVAL; |
| 372 |
|
| 373 |
if (lvblen % 8) |
| 374 |
return -EINVAL; |
| 375 |
|
| 376 |
if (!try_module_get(THIS_MODULE)) |
| 377 |
return -EINVAL; |
| 378 |
|
| 379 |
if (!dlm_user_daemon_available()) { |
| 380 |
log_print("dlm user daemon not available"); |
| 381 |
error = -EUNATCH; |
| 382 |
goto out; |
| 383 |
} |
| 384 |
|
| 385 |
if (ops && ops_result) { |
| 386 |
if (!dlm_config.ci_recover_callbacks) |
| 387 |
*ops_result = -EOPNOTSUPP; |
| 388 |
else |
| 389 |
*ops_result = 0; |
| 390 |
} |
| 391 |
|
| 392 |
if (!cluster) |
| 393 |
log_print("dlm cluster name '%s' is being used without an application provided cluster name", |
| 394 |
dlm_config.ci_cluster_name); |
| 395 |
|
| 396 |
if (dlm_config.ci_recover_callbacks && cluster && |
| 397 |
strncmp(cluster, dlm_config.ci_cluster_name, DLM_LOCKSPACE_LEN)) { |
| 398 |
log_print("dlm cluster name '%s' does not match " |
| 399 |
"the application cluster name '%s'", |
| 400 |
dlm_config.ci_cluster_name, cluster); |
| 401 |
error = -EBADR; |
| 402 |
goto out; |
| 403 |
} |
| 404 |
|
| 405 |
error = 0; |
| 406 |
|
| 407 |
spin_lock_bh(&lslist_lock); |
| 408 |
list_for_each_entry(ls, &lslist, ls_list) { |
| 409 |
WARN_ON(ls->ls_create_count <= 0); |
| 410 |
if (ls->ls_namelen != namelen) |
| 411 |
continue; |
| 412 |
if (memcmp(ls->ls_name, name, namelen)) |
| 413 |
continue; |
| 414 |
if (flags & DLM_LSFL_NEWEXCL) { |
| 415 |
error = -EEXIST; |
| 416 |
break; |
| 417 |
} |
| 418 |
ls->ls_create_count++; |
| 419 |
*lockspace = ls; |
| 420 |
error = 1; |
| 421 |
break; |
| 422 |
} |
| 423 |
spin_unlock_bh(&lslist_lock); |
| 424 |
|
| 425 |
if (error) |
| 426 |
goto out; |
| 427 |
|
| 428 |
error = -ENOMEM; |
| 429 |
|
| 430 |
ls = kzalloc_obj(*ls, GFP_NOFS); |
| 431 |
if (!ls) |
| 432 |
goto out; |
| 433 |
memcpy(ls->ls_name, name, namelen); |
| 434 |
ls->ls_namelen = namelen; |
| 435 |
ls->ls_lvblen = lvblen; |
| 436 |
atomic_set(&ls->ls_count, 0); |
| 437 |
init_waitqueue_head(&ls->ls_count_wait); |
| 438 |
ls->ls_flags = 0; |
| 439 |
|
| 440 |
if (ops && dlm_config.ci_recover_callbacks) { |
| 441 |
ls->ls_ops = ops; |
| 442 |
ls->ls_ops_arg = ops_arg; |
| 443 |
} |
| 444 |
|
| 445 |
if (flags & DLM_LSFL_SOFTIRQ) |
| 446 |
set_bit(LSFL_SOFTIRQ, &ls->ls_flags); |
| 447 |
|
| 448 |
/* ls_exflags are forced to match among nodes, and we don't |
| 449 |
* need to require all nodes to have some flags set |
| 450 |
*/ |
| 451 |
ls->ls_exflags = (flags & ~(DLM_LSFL_FS | DLM_LSFL_NEWEXCL | |
| 452 |
DLM_LSFL_SOFTIRQ)); |
| 453 |
|
| 454 |
INIT_LIST_HEAD(&ls->ls_slow_inactive); |
| 455 |
INIT_LIST_HEAD(&ls->ls_slow_active); |
| 456 |
rwlock_init(&ls->ls_rsbtbl_lock); |
| 457 |
|
| 458 |
error = rhashtable_init(&ls->ls_rsbtbl, &dlm_rhash_rsb_params); |
| 459 |
if (error) |
| 460 |
goto out_lsfree; |
| 461 |
|
| 462 |
xa_init_flags(&ls->ls_lkbxa, XA_FLAGS_ALLOC | XA_FLAGS_LOCK_BH); |
| 463 |
rwlock_init(&ls->ls_lkbxa_lock); |
| 464 |
|
| 465 |
INIT_LIST_HEAD(&ls->ls_waiters); |
| 466 |
spin_lock_init(&ls->ls_waiters_lock); |
| 467 |
INIT_LIST_HEAD(&ls->ls_orphans); |
| 468 |
spin_lock_init(&ls->ls_orphans_lock); |
| 469 |
|
| 470 |
INIT_LIST_HEAD(&ls->ls_nodes); |
| 471 |
INIT_LIST_HEAD(&ls->ls_nodes_gone); |
| 472 |
ls->ls_num_nodes = 0; |
| 473 |
ls->ls_low_nodeid = 0; |
| 474 |
ls->ls_total_weight = 0; |
| 475 |
ls->ls_node_array = NULL; |
| 476 |
|
| 477 |
memset(&ls->ls_local_rsb, 0, sizeof(struct dlm_rsb)); |
| 478 |
ls->ls_local_rsb.res_ls = ls; |
| 479 |
|
| 480 |
ls->ls_debug_rsb_dentry = NULL; |
| 481 |
ls->ls_debug_waiters_dentry = NULL; |
| 482 |
|
| 483 |
init_waitqueue_head(&ls->ls_uevent_wait); |
| 484 |
ls->ls_uevent_result = 0; |
| 485 |
init_completion(&ls->ls_recovery_done); |
| 486 |
ls->ls_recovery_result = -1; |
| 487 |
|
| 488 |
spin_lock_init(&ls->ls_cb_lock); |
| 489 |
INIT_LIST_HEAD(&ls->ls_cb_delay); |
| 490 |
|
| 491 |
INIT_WORK(&ls->ls_free_work, free_lockspace); |
| 492 |
|
| 493 |
ls->ls_recoverd_task = NULL; |
| 494 |
mutex_init(&ls->ls_recoverd_active); |
| 495 |
spin_lock_init(&ls->ls_recover_lock); |
| 496 |
spin_lock_init(&ls->ls_rcom_spin); |
| 497 |
get_random_bytes(&ls->ls_rcom_seq, sizeof(uint64_t)); |
| 498 |
ls->ls_recover_status = 0; |
| 499 |
ls->ls_recover_seq = get_random_u64(); |
| 500 |
ls->ls_recover_args = NULL; |
| 501 |
init_rwsem(&ls->ls_in_recovery); |
| 502 |
rwlock_init(&ls->ls_recv_active); |
| 503 |
INIT_LIST_HEAD(&ls->ls_requestqueue); |
| 504 |
rwlock_init(&ls->ls_requestqueue_lock); |
| 505 |
spin_lock_init(&ls->ls_clear_proc_locks); |
| 506 |
|
| 507 |
/* Due backwards compatibility with 3.1 we need to use maximum |
| 508 |
* possible dlm message size to be sure the message will fit and |
| 509 |
* not having out of bounds issues. However on sending side 3.2 |
| 510 |
* might send less. |
| 511 |
*/ |
| 512 |
ls->ls_recover_buf = kmalloc(DLM_MAX_SOCKET_BUFSIZE, GFP_NOFS); |
| 513 |
if (!ls->ls_recover_buf) { |
| 514 |
error = -ENOMEM; |
| 515 |
goto out_lkbxa; |
| 516 |
} |
| 517 |
|
| 518 |
ls->ls_slot = 0; |
| 519 |
ls->ls_num_slots = 0; |
| 520 |
ls->ls_slots_size = 0; |
| 521 |
ls->ls_slots = NULL; |
| 522 |
|
| 523 |
INIT_LIST_HEAD(&ls->ls_recover_list); |
| 524 |
spin_lock_init(&ls->ls_recover_list_lock); |
| 525 |
xa_init_flags(&ls->ls_recover_xa, XA_FLAGS_ALLOC | XA_FLAGS_LOCK_BH); |
| 526 |
spin_lock_init(&ls->ls_recover_xa_lock); |
| 527 |
ls->ls_recover_list_count = 0; |
| 528 |
init_waitqueue_head(&ls->ls_wait_general); |
| 529 |
INIT_LIST_HEAD(&ls->ls_masters_list); |
| 530 |
rwlock_init(&ls->ls_masters_lock); |
| 531 |
INIT_LIST_HEAD(&ls->ls_dir_dump_list); |
| 532 |
rwlock_init(&ls->ls_dir_dump_lock); |
| 533 |
|
| 534 |
INIT_LIST_HEAD(&ls->ls_scan_list); |
| 535 |
spin_lock_init(&ls->ls_scan_lock); |
| 536 |
timer_setup(&ls->ls_scan_timer, dlm_rsb_scan, TIMER_DEFERRABLE); |
| 537 |
|
| 538 |
spin_lock_bh(&lslist_lock); |
| 539 |
ls->ls_create_count = 1; |
| 540 |
list_add(&ls->ls_list, &lslist); |
| 541 |
spin_unlock_bh(&lslist_lock); |
| 542 |
|
| 543 |
if (flags & DLM_LSFL_FS) |
| 544 |
set_bit(LSFL_FS, &ls->ls_flags); |
| 545 |
|
| 546 |
error = dlm_callback_start(ls); |
| 547 |
if (error) { |
| 548 |
log_error(ls, "can't start dlm_callback %d", error); |
| 549 |
goto out_delist; |
| 550 |
} |
| 551 |
|
| 552 |
init_waitqueue_head(&ls->ls_recover_lock_wait); |
| 553 |
|
| 554 |
/* |
| 555 |
* Once started, dlm_recoverd first looks for ls in lslist, then |
| 556 |
* initializes ls_in_recovery as locked in "down" mode. We need |
| 557 |
* to wait for the wakeup from dlm_recoverd because in_recovery |
| 558 |
* has to start out in down mode. |
| 559 |
*/ |
| 560 |
|
| 561 |
error = dlm_recoverd_start(ls); |
| 562 |
if (error) { |
| 563 |
log_error(ls, "can't start dlm_recoverd %d", error); |
| 564 |
goto out_callback; |
| 565 |
} |
| 566 |
|
| 567 |
wait_event(ls->ls_recover_lock_wait, |
| 568 |
test_bit(LSFL_RECOVER_LOCK, &ls->ls_flags)); |
| 569 |
|
| 570 |
ls->ls_kobj.kset = dlm_kset; |
| 571 |
error = kobject_init_and_add(&ls->ls_kobj, &dlm_ktype, NULL, |
| 572 |
"%s", ls->ls_name); |
| 573 |
if (error) |
| 574 |
goto out_recoverd; |
| 575 |
kobject_uevent(&ls->ls_kobj, KOBJ_ADD); |
| 576 |
|
| 577 |
/* This uevent triggers dlm_controld in userspace to add us to the |
| 578 |
group of nodes that are members of this lockspace (managed by the |
| 579 |
cluster infrastructure.) Once it's done that, it tells us who the |
| 580 |
current lockspace members are (via configfs) and then tells the |
| 581 |
lockspace to start running (via sysfs) in dlm_ls_start(). */ |
| 582 |
|
| 583 |
error = do_uevent(ls, 1, 0); |
| 584 |
if (error < 0) |
| 585 |
goto out_recoverd; |
| 586 |
|
| 587 |
/* wait until recovery is successful or failed */ |
| 588 |
wait_for_completion(&ls->ls_recovery_done); |
| 589 |
error = ls->ls_recovery_result; |
| 590 |
if (error) |
| 591 |
goto out_members; |
| 592 |
|
| 593 |
dlm_create_debug_file(ls); |
| 594 |
|
| 595 |
log_rinfo(ls, "join complete"); |
| 596 |
*lockspace = ls; |
| 597 |
return 0; |
| 598 |
|
| 599 |
out_members: |
| 600 |
do_uevent(ls, 0, 0); |
| 601 |
dlm_clear_members(ls); |
| 602 |
kfree(ls->ls_node_array); |
| 603 |
out_recoverd: |
| 604 |
dlm_recoverd_stop(ls); |
| 605 |
out_callback: |
| 606 |
dlm_callback_stop(ls); |
| 607 |
out_delist: |
| 608 |
spin_lock_bh(&lslist_lock); |
| 609 |
list_del(&ls->ls_list); |
| 610 |
spin_unlock_bh(&lslist_lock); |
| 611 |
xa_destroy(&ls->ls_recover_xa); |
| 612 |
kfree(ls->ls_recover_buf); |
| 613 |
out_lkbxa: |
| 614 |
xa_destroy(&ls->ls_lkbxa); |
| 615 |
rhashtable_destroy(&ls->ls_rsbtbl); |
| 616 |
out_lsfree: |
| 617 |
kobject_put(&ls->ls_kobj); |
| 618 |
kfree(ls); |
| 619 |
out: |
| 620 |
module_put(THIS_MODULE); |
| 621 |
return error; |
| 622 |
} |
| 623 |
|
| 624 |
static int __dlm_new_lockspace(const char *name, const char *cluster, |
| 625 |
uint32_t flags, int lvblen, |
| 626 |
const struct dlm_lockspace_ops *ops, |
| 627 |
void *ops_arg, int *ops_result, |
| 628 |
dlm_lockspace_t **lockspace) |
| 629 |
{ |
| 630 |
int error = 0; |
| 631 |
|
| 632 |
mutex_lock(&ls_lock); |
| 633 |
if (!ls_count) |
| 634 |
error = threads_start(); |
| 635 |
if (error) |
| 636 |
goto out; |
| 637 |
|
| 638 |
error = new_lockspace(name, cluster, flags, lvblen, ops, ops_arg, |
| 639 |
ops_result, lockspace); |
| 640 |
if (!error) |
| 641 |
ls_count++; |
| 642 |
if (error > 0) |
| 643 |
error = 0; |
| 644 |
if (!ls_count) { |
| 645 |
dlm_midcomms_shutdown(); |
| 646 |
dlm_midcomms_stop(); |
| 647 |
} |
| 648 |
out: |
| 649 |
mutex_unlock(&ls_lock); |
| 650 |
return error; |
| 651 |
} |
| 652 |
|
| 653 |
int dlm_new_lockspace(const char *name, const char *cluster, uint32_t flags, |
| 654 |
int lvblen, const struct dlm_lockspace_ops *ops, |
| 655 |
void *ops_arg, int *ops_result, |
| 656 |
dlm_lockspace_t **lockspace) |
| 657 |
{ |
| 658 |
return __dlm_new_lockspace(name, cluster, flags | DLM_LSFL_FS, lvblen, |
| 659 |
ops, ops_arg, ops_result, lockspace); |
| 660 |
} |
| 661 |
|
| 662 |
int dlm_new_user_lockspace(const char *name, const char *cluster, |
| 663 |
uint32_t flags, int lvblen, |
| 664 |
const struct dlm_lockspace_ops *ops, |
| 665 |
void *ops_arg, int *ops_result, |
| 666 |
dlm_lockspace_t **lockspace) |
| 667 |
{ |
| 668 |
if (flags & DLM_LSFL_SOFTIRQ) |
| 669 |
return -EINVAL; |
| 670 |
|
| 671 |
return __dlm_new_lockspace(name, cluster, flags, lvblen, ops, |
| 672 |
ops_arg, ops_result, lockspace); |
| 673 |
} |
| 674 |
|
| 675 |
/* NOTE: We check the lkbxa here rather than the resource table. |
| 676 |
This is because there may be LKBs queued as ASTs that have been unlinked |
| 677 |
from their RSBs and are pending deletion once the AST has been delivered */ |
| 678 |
|
| 679 |
static int lockspace_busy(struct dlm_ls *ls, unsigned int release_option) |
| 680 |
{ |
| 681 |
struct dlm_lkb *lkb; |
| 682 |
unsigned long id; |
| 683 |
int rv = 0; |
| 684 |
|
| 685 |
read_lock_bh(&ls->ls_lkbxa_lock); |
| 686 |
if (release_option == DLM_RELEASE_NO_LOCKS) { |
| 687 |
xa_for_each(&ls->ls_lkbxa, id, lkb) { |
| 688 |
rv = 1; |
| 689 |
break; |
| 690 |
} |
| 691 |
} else if (release_option == DLM_RELEASE_UNUSED) { |
| 692 |
/* TODO: handle this UNUSED option as NO_LOCKS in later patch */ |
| 693 |
xa_for_each(&ls->ls_lkbxa, id, lkb) { |
| 694 |
if (lkb->lkb_nodeid == 0 && |
| 695 |
lkb->lkb_grmode != DLM_LOCK_IV) { |
| 696 |
rv = 1; |
| 697 |
break; |
| 698 |
} |
| 699 |
} |
| 700 |
} else { |
| 701 |
rv = 0; |
| 702 |
} |
| 703 |
read_unlock_bh(&ls->ls_lkbxa_lock); |
| 704 |
return rv; |
| 705 |
} |
| 706 |
|
| 707 |
static int release_lockspace(struct dlm_ls *ls, unsigned int release_option) |
| 708 |
{ |
| 709 |
int busy, rv; |
| 710 |
|
| 711 |
busy = lockspace_busy(ls, release_option); |
| 712 |
|
| 713 |
spin_lock_bh(&lslist_lock); |
| 714 |
if (ls->ls_create_count == 1) { |
| 715 |
if (busy) { |
| 716 |
rv = -EBUSY; |
| 717 |
} else { |
| 718 |
/* remove_lockspace takes ls off lslist */ |
| 719 |
ls->ls_create_count = 0; |
| 720 |
rv = 0; |
| 721 |
} |
| 722 |
} else if (ls->ls_create_count > 1) { |
| 723 |
rv = --ls->ls_create_count; |
| 724 |
} else { |
| 725 |
rv = -EINVAL; |
| 726 |
} |
| 727 |
spin_unlock_bh(&lslist_lock); |
| 728 |
|
| 729 |
if (rv) { |
| 730 |
log_debug(ls, "release_lockspace no remove %d", rv); |
| 731 |
return rv; |
| 732 |
} |
| 733 |
|
| 734 |
if (ls_count == 1) |
| 735 |
dlm_midcomms_version_wait(); |
| 736 |
|
| 737 |
dlm_device_deregister(ls); |
| 738 |
|
| 739 |
if (release_option != DLM_RELEASE_NO_EVENT && |
| 740 |
dlm_user_daemon_available()) |
| 741 |
do_uevent(ls, 0, (release_option == DLM_RELEASE_RECOVER)); |
| 742 |
|
| 743 |
dlm_recoverd_stop(ls); |
| 744 |
|
| 745 |
/* clear the LSFL_RUNNING flag to fast up |
| 746 |
* time_shutdown_sync(), we don't care anymore |
| 747 |
*/ |
| 748 |
clear_bit(LSFL_RUNNING, &ls->ls_flags); |
| 749 |
timer_shutdown_sync(&ls->ls_scan_timer); |
| 750 |
|
| 751 |
if (ls_count == 1) { |
| 752 |
dlm_clear_members(ls); |
| 753 |
dlm_midcomms_shutdown(); |
| 754 |
} |
| 755 |
|
| 756 |
dlm_callback_stop(ls); |
| 757 |
|
| 758 |
remove_lockspace(ls); |
| 759 |
|
| 760 |
dlm_delete_debug_file(ls); |
| 761 |
|
| 762 |
kobject_put(&ls->ls_kobj); |
| 763 |
|
| 764 |
xa_destroy(&ls->ls_recover_xa); |
| 765 |
kfree(ls->ls_recover_buf); |
| 766 |
|
| 767 |
/* |
| 768 |
* Free structures on any other lists |
| 769 |
*/ |
| 770 |
|
| 771 |
dlm_purge_requestqueue(ls); |
| 772 |
kfree(ls->ls_recover_args); |
| 773 |
dlm_clear_members(ls); |
| 774 |
dlm_clear_members_gone(ls); |
| 775 |
kfree(ls->ls_node_array); |
| 776 |
|
| 777 |
log_rinfo(ls, "%s final free", __func__); |
| 778 |
|
| 779 |
/* delayed free of data structures see free_lockspace() */ |
| 780 |
queue_work(dlm_wq, &ls->ls_free_work); |
| 781 |
module_put(THIS_MODULE); |
| 782 |
return 0; |
| 783 |
} |
| 784 |
|
| 785 |
/* |
| 786 |
* Called when a system has released all its locks and is not going to use the |
| 787 |
* lockspace any longer. We free everything we're managing for this lockspace. |
| 788 |
* Remaining nodes will go through the recovery process as if we'd died. The |
| 789 |
* lockspace must continue to function as usual, participating in recoveries, |
| 790 |
* until this returns. |
| 791 |
* |
| 792 |
* See DLM_RELEASE defines for release_option values and their meaning. |
| 793 |
*/ |
| 794 |
|
| 795 |
int dlm_release_lockspace(void *lockspace, unsigned int release_option) |
| 796 |
{ |
| 797 |
struct dlm_ls *ls; |
| 798 |
int error; |
| 799 |
|
| 800 |
if (release_option > __DLM_RELEASE_MAX) |
| 801 |
return -EINVAL; |
| 802 |
|
| 803 |
ls = dlm_find_lockspace_local(lockspace); |
| 804 |
if (!ls) |
| 805 |
return -EINVAL; |
| 806 |
dlm_put_lockspace(ls); |
| 807 |
|
| 808 |
mutex_lock(&ls_lock); |
| 809 |
error = release_lockspace(ls, release_option); |
| 810 |
if (!error) |
| 811 |
ls_count--; |
| 812 |
if (!ls_count) |
| 813 |
dlm_midcomms_stop(); |
| 814 |
mutex_unlock(&ls_lock); |
| 815 |
|
| 816 |
return error; |
| 817 |
} |
| 818 |
|
| 819 |
void dlm_stop_lockspaces(void) |
| 820 |
{ |
| 821 |
struct dlm_ls *ls; |
| 822 |
int count; |
| 823 |
|
| 824 |
restart: |
| 825 |
count = 0; |
| 826 |
spin_lock_bh(&lslist_lock); |
| 827 |
list_for_each_entry(ls, &lslist, ls_list) { |
| 828 |
if (!test_bit(LSFL_RUNNING, &ls->ls_flags)) { |
| 829 |
count++; |
| 830 |
continue; |
| 831 |
} |
| 832 |
spin_unlock_bh(&lslist_lock); |
| 833 |
log_error(ls, "no userland control daemon, stopping lockspace"); |
| 834 |
dlm_ls_stop(ls); |
| 835 |
goto restart; |
| 836 |
} |
| 837 |
spin_unlock_bh(&lslist_lock); |
| 838 |
|
| 839 |
if (count) |
| 840 |
log_print("dlm user daemon left %d lockspaces", count); |
| 841 |
} |