0001
0002
0003
0004
0005
0006
0007
0008
0009
0010
0011 #include <linux/module.h>
0012 #include <linux/fs.h>
0013 #include <linux/types.h>
0014 #include <linux/slab.h>
0015 #include <linux/highmem.h>
0016 #include <linux/init.h>
0017 #include <linux/sysctl.h>
0018 #include <linux/random.h>
0019 #include <linux/blkdev.h>
0020 #include <linux/socket.h>
0021 #include <linux/inet.h>
0022 #include <linux/spinlock.h>
0023 #include <linux/delay.h>
0024
0025
0026 #include "../cluster/heartbeat.h"
0027 #include "../cluster/nodemanager.h"
0028 #include "../cluster/tcp.h"
0029
0030 #include "dlmapi.h"
0031 #include "dlmcommon.h"
0032
0033 #include "dlmconvert.h"
0034
0035 #define MLOG_MASK_PREFIX ML_DLM
0036 #include "../cluster/masklog.h"
0037
0038 static struct kmem_cache *dlm_lock_cache;
0039
0040 static DEFINE_SPINLOCK(dlm_cookie_lock);
0041 static u64 dlm_next_cookie = 1;
0042
0043 static enum dlm_status dlm_send_remote_lock_request(struct dlm_ctxt *dlm,
0044 struct dlm_lock_resource *res,
0045 struct dlm_lock *lock, int flags);
0046 static void dlm_init_lock(struct dlm_lock *newlock, int type,
0047 u8 node, u64 cookie);
0048 static void dlm_lock_release(struct kref *kref);
0049 static void dlm_lock_detach_lockres(struct dlm_lock *lock);
0050
0051 int dlm_init_lock_cache(void)
0052 {
0053 dlm_lock_cache = kmem_cache_create("o2dlm_lock",
0054 sizeof(struct dlm_lock),
0055 0, SLAB_HWCACHE_ALIGN, NULL);
0056 if (dlm_lock_cache == NULL)
0057 return -ENOMEM;
0058 return 0;
0059 }
0060
0061 void dlm_destroy_lock_cache(void)
0062 {
0063 kmem_cache_destroy(dlm_lock_cache);
0064 }
0065
0066
0067
0068
0069
0070
0071
0072
0073 static int dlm_can_grant_new_lock(struct dlm_lock_resource *res,
0074 struct dlm_lock *lock)
0075 {
0076 struct dlm_lock *tmplock;
0077
0078 list_for_each_entry(tmplock, &res->granted, list) {
0079 if (!dlm_lock_compatible(tmplock->ml.type, lock->ml.type))
0080 return 0;
0081 }
0082
0083 list_for_each_entry(tmplock, &res->converting, list) {
0084 if (!dlm_lock_compatible(tmplock->ml.type, lock->ml.type))
0085 return 0;
0086 if (!dlm_lock_compatible(tmplock->ml.convert_type,
0087 lock->ml.type))
0088 return 0;
0089 }
0090
0091 return 1;
0092 }
0093
0094
0095
0096
0097
0098
0099
0100
0101 static enum dlm_status dlmlock_master(struct dlm_ctxt *dlm,
0102 struct dlm_lock_resource *res,
0103 struct dlm_lock *lock, int flags)
0104 {
0105 int call_ast = 0, kick_thread = 0;
0106 enum dlm_status status = DLM_NORMAL;
0107
0108 mlog(0, "type=%d\n", lock->ml.type);
0109
0110 spin_lock(&res->spinlock);
0111
0112
0113 status = __dlm_lockres_state_to_status(res);
0114 if (status != DLM_NORMAL &&
0115 lock->ml.node != dlm->node_num) {
0116
0117 spin_unlock(&res->spinlock);
0118 dlm_error(status);
0119 return status;
0120 }
0121 __dlm_wait_on_lockres(res);
0122 __dlm_lockres_reserve_ast(res);
0123
0124 if (dlm_can_grant_new_lock(res, lock)) {
0125 mlog(0, "I can grant this lock right away\n");
0126
0127 lock->lksb->status = DLM_NORMAL;
0128 status = DLM_NORMAL;
0129 dlm_lock_get(lock);
0130 list_add_tail(&lock->list, &res->granted);
0131
0132
0133
0134
0135
0136
0137 if (!dlm_is_recovery_lock(res->lockname.name,
0138 res->lockname.len)) {
0139 kick_thread = 1;
0140 call_ast = 1;
0141 } else {
0142 mlog(0, "%s: returning DLM_NORMAL to "
0143 "node %u for reco lock\n", dlm->name,
0144 lock->ml.node);
0145 }
0146 } else {
0147
0148
0149 if (flags & LKM_NOQUEUE) {
0150 status = DLM_NOTQUEUED;
0151 if (dlm_is_recovery_lock(res->lockname.name,
0152 res->lockname.len)) {
0153 mlog(0, "%s: returning NOTQUEUED to "
0154 "node %u for reco lock\n", dlm->name,
0155 lock->ml.node);
0156 }
0157 } else {
0158 status = DLM_NORMAL;
0159 dlm_lock_get(lock);
0160 list_add_tail(&lock->list, &res->blocked);
0161 kick_thread = 1;
0162 }
0163 }
0164
0165 spin_unlock(&res->spinlock);
0166 wake_up(&res->wq);
0167
0168
0169 if (call_ast)
0170 dlm_queue_ast(dlm, lock);
0171 else
0172 dlm_lockres_release_ast(dlm, res);
0173
0174 dlm_lockres_calc_usage(dlm, res);
0175 if (kick_thread)
0176 dlm_kick_thread(dlm, res);
0177
0178 return status;
0179 }
0180
0181 void dlm_revert_pending_lock(struct dlm_lock_resource *res,
0182 struct dlm_lock *lock)
0183 {
0184
0185 list_del_init(&lock->list);
0186 lock->lksb->flags &= ~DLM_LKSB_GET_LVB;
0187 }
0188
0189
0190
0191
0192
0193
0194
0195
0196
0197 static enum dlm_status dlmlock_remote(struct dlm_ctxt *dlm,
0198 struct dlm_lock_resource *res,
0199 struct dlm_lock *lock, int flags)
0200 {
0201 enum dlm_status status = DLM_DENIED;
0202 int lockres_changed = 1;
0203
0204 mlog(0, "type=%d, lockres %.*s, flags = 0x%x\n",
0205 lock->ml.type, res->lockname.len,
0206 res->lockname.name, flags);
0207
0208
0209
0210
0211
0212 spin_lock(&res->spinlock);
0213 __dlm_wait_on_lockres(res);
0214 if (res->owner == dlm->node_num) {
0215 spin_unlock(&res->spinlock);
0216 return DLM_RECOVERING;
0217 }
0218 res->state |= DLM_LOCK_RES_IN_PROGRESS;
0219
0220
0221 dlm_lock_get(lock);
0222 list_add_tail(&lock->list, &res->blocked);
0223 lock->lock_pending = 1;
0224 spin_unlock(&res->spinlock);
0225
0226
0227
0228 status = dlm_send_remote_lock_request(dlm, res, lock, flags);
0229
0230 spin_lock(&res->spinlock);
0231 res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
0232 lock->lock_pending = 0;
0233 if (status != DLM_NORMAL) {
0234 if (status == DLM_RECOVERING &&
0235 dlm_is_recovery_lock(res->lockname.name,
0236 res->lockname.len)) {
0237
0238
0239
0240 mlog(0, "%s: recovery lock was owned by "
0241 "dead node %u, remaster it now.\n",
0242 dlm->name, res->owner);
0243 } else if (status != DLM_NOTQUEUED) {
0244
0245
0246
0247
0248
0249
0250 lockres_changed = 0;
0251 dlm_error(status);
0252 }
0253 dlm_revert_pending_lock(res, lock);
0254 dlm_lock_put(lock);
0255 } else if (dlm_is_recovery_lock(res->lockname.name,
0256 res->lockname.len)) {
0257
0258
0259
0260
0261 mlog(0, "%s: $RECOVERY lock for this node (%u) is "
0262 "mastered by %u; got lock, manually granting (no ast)\n",
0263 dlm->name, dlm->node_num, res->owner);
0264 list_move_tail(&lock->list, &res->granted);
0265 }
0266 spin_unlock(&res->spinlock);
0267
0268 if (lockres_changed)
0269 dlm_lockres_calc_usage(dlm, res);
0270
0271 wake_up(&res->wq);
0272 return status;
0273 }
0274
0275
0276
0277
0278
0279
0280
0281
0282
0283 static enum dlm_status dlm_send_remote_lock_request(struct dlm_ctxt *dlm,
0284 struct dlm_lock_resource *res,
0285 struct dlm_lock *lock, int flags)
0286 {
0287 struct dlm_create_lock create;
0288 int tmpret, status = 0;
0289 enum dlm_status ret;
0290
0291 memset(&create, 0, sizeof(create));
0292 create.node_idx = dlm->node_num;
0293 create.requested_type = lock->ml.type;
0294 create.cookie = lock->ml.cookie;
0295 create.namelen = res->lockname.len;
0296 create.flags = cpu_to_be32(flags);
0297 memcpy(create.name, res->lockname.name, create.namelen);
0298
0299 tmpret = o2net_send_message(DLM_CREATE_LOCK_MSG, dlm->key, &create,
0300 sizeof(create), res->owner, &status);
0301 if (tmpret >= 0) {
0302 ret = status;
0303 if (ret == DLM_REJECTED) {
0304 mlog(ML_ERROR, "%s: res %.*s, Stale lockres no longer "
0305 "owned by node %u. That node is coming back up "
0306 "currently.\n", dlm->name, create.namelen,
0307 create.name, res->owner);
0308 dlm_print_one_lock_resource(res);
0309 BUG();
0310 }
0311 } else {
0312 mlog(ML_ERROR, "%s: res %.*s, Error %d send CREATE LOCK to "
0313 "node %u\n", dlm->name, create.namelen, create.name,
0314 tmpret, res->owner);
0315 if (dlm_is_host_down(tmpret))
0316 ret = DLM_RECOVERING;
0317 else
0318 ret = dlm_err_to_dlm_status(tmpret);
0319 }
0320
0321 return ret;
0322 }
0323
0324 void dlm_lock_get(struct dlm_lock *lock)
0325 {
0326 kref_get(&lock->lock_refs);
0327 }
0328
0329 void dlm_lock_put(struct dlm_lock *lock)
0330 {
0331 kref_put(&lock->lock_refs, dlm_lock_release);
0332 }
0333
0334 static void dlm_lock_release(struct kref *kref)
0335 {
0336 struct dlm_lock *lock;
0337
0338 lock = container_of(kref, struct dlm_lock, lock_refs);
0339
0340 BUG_ON(!list_empty(&lock->list));
0341 BUG_ON(!list_empty(&lock->ast_list));
0342 BUG_ON(!list_empty(&lock->bast_list));
0343 BUG_ON(lock->ast_pending);
0344 BUG_ON(lock->bast_pending);
0345
0346 dlm_lock_detach_lockres(lock);
0347
0348 if (lock->lksb_kernel_allocated) {
0349 mlog(0, "freeing kernel-allocated lksb\n");
0350 kfree(lock->lksb);
0351 }
0352 kmem_cache_free(dlm_lock_cache, lock);
0353 }
0354
0355
0356 void dlm_lock_attach_lockres(struct dlm_lock *lock,
0357 struct dlm_lock_resource *res)
0358 {
0359 dlm_lockres_get(res);
0360 lock->lockres = res;
0361 }
0362
0363
0364 static void dlm_lock_detach_lockres(struct dlm_lock *lock)
0365 {
0366 struct dlm_lock_resource *res;
0367
0368 res = lock->lockres;
0369 if (res) {
0370 lock->lockres = NULL;
0371 mlog(0, "removing lock's lockres reference\n");
0372 dlm_lockres_put(res);
0373 }
0374 }
0375
0376 static void dlm_init_lock(struct dlm_lock *newlock, int type,
0377 u8 node, u64 cookie)
0378 {
0379 INIT_LIST_HEAD(&newlock->list);
0380 INIT_LIST_HEAD(&newlock->ast_list);
0381 INIT_LIST_HEAD(&newlock->bast_list);
0382 spin_lock_init(&newlock->spinlock);
0383 newlock->ml.type = type;
0384 newlock->ml.convert_type = LKM_IVMODE;
0385 newlock->ml.highest_blocked = LKM_IVMODE;
0386 newlock->ml.node = node;
0387 newlock->ml.pad1 = 0;
0388 newlock->ml.list = 0;
0389 newlock->ml.flags = 0;
0390 newlock->ast = NULL;
0391 newlock->bast = NULL;
0392 newlock->astdata = NULL;
0393 newlock->ml.cookie = cpu_to_be64(cookie);
0394 newlock->ast_pending = 0;
0395 newlock->bast_pending = 0;
0396 newlock->convert_pending = 0;
0397 newlock->lock_pending = 0;
0398 newlock->unlock_pending = 0;
0399 newlock->cancel_pending = 0;
0400 newlock->lksb_kernel_allocated = 0;
0401
0402 kref_init(&newlock->lock_refs);
0403 }
0404
0405 struct dlm_lock * dlm_new_lock(int type, u8 node, u64 cookie,
0406 struct dlm_lockstatus *lksb)
0407 {
0408 struct dlm_lock *lock;
0409 int kernel_allocated = 0;
0410
0411 lock = kmem_cache_zalloc(dlm_lock_cache, GFP_NOFS);
0412 if (!lock)
0413 return NULL;
0414
0415 if (!lksb) {
0416
0417 lksb = kzalloc(sizeof(*lksb), GFP_NOFS);
0418 if (!lksb) {
0419 kmem_cache_free(dlm_lock_cache, lock);
0420 return NULL;
0421 }
0422 kernel_allocated = 1;
0423 }
0424
0425 dlm_init_lock(lock, type, node, cookie);
0426 if (kernel_allocated)
0427 lock->lksb_kernel_allocated = 1;
0428 lock->lksb = lksb;
0429 lksb->lockid = lock;
0430 return lock;
0431 }
0432
0433
0434
0435
0436
0437
0438
0439
0440 int dlm_create_lock_handler(struct o2net_msg *msg, u32 len, void *data,
0441 void **ret_data)
0442 {
0443 struct dlm_ctxt *dlm = data;
0444 struct dlm_create_lock *create = (struct dlm_create_lock *)msg->buf;
0445 struct dlm_lock_resource *res = NULL;
0446 struct dlm_lock *newlock = NULL;
0447 struct dlm_lockstatus *lksb = NULL;
0448 enum dlm_status status = DLM_NORMAL;
0449 char *name;
0450 unsigned int namelen;
0451
0452 BUG_ON(!dlm);
0453
0454 if (!dlm_grab(dlm))
0455 return DLM_REJECTED;
0456
0457 name = create->name;
0458 namelen = create->namelen;
0459 status = DLM_REJECTED;
0460 if (!dlm_domain_fully_joined(dlm)) {
0461 mlog(ML_ERROR, "Domain %s not fully joined, but node %u is "
0462 "sending a create_lock message for lock %.*s!\n",
0463 dlm->name, create->node_idx, namelen, name);
0464 dlm_error(status);
0465 goto leave;
0466 }
0467
0468 status = DLM_IVBUFLEN;
0469 if (namelen > DLM_LOCKID_NAME_MAX) {
0470 dlm_error(status);
0471 goto leave;
0472 }
0473
0474 status = DLM_SYSERR;
0475 newlock = dlm_new_lock(create->requested_type,
0476 create->node_idx,
0477 be64_to_cpu(create->cookie), NULL);
0478 if (!newlock) {
0479 dlm_error(status);
0480 goto leave;
0481 }
0482
0483 lksb = newlock->lksb;
0484
0485 if (be32_to_cpu(create->flags) & LKM_GET_LVB) {
0486 lksb->flags |= DLM_LKSB_GET_LVB;
0487 mlog(0, "set DLM_LKSB_GET_LVB flag\n");
0488 }
0489
0490 status = DLM_IVLOCKID;
0491 res = dlm_lookup_lockres(dlm, name, namelen);
0492 if (!res) {
0493 dlm_error(status);
0494 goto leave;
0495 }
0496
0497 spin_lock(&res->spinlock);
0498 status = __dlm_lockres_state_to_status(res);
0499 spin_unlock(&res->spinlock);
0500
0501 if (status != DLM_NORMAL) {
0502 mlog(0, "lockres recovering/migrating/in-progress\n");
0503 goto leave;
0504 }
0505
0506 dlm_lock_attach_lockres(newlock, res);
0507
0508 status = dlmlock_master(dlm, res, newlock, be32_to_cpu(create->flags));
0509 leave:
0510 if (status != DLM_NORMAL)
0511 if (newlock)
0512 dlm_lock_put(newlock);
0513
0514 if (res)
0515 dlm_lockres_put(res);
0516
0517 dlm_put(dlm);
0518
0519 return status;
0520 }
0521
0522
0523
0524 static inline void dlm_get_next_cookie(u8 node_num, u64 *cookie)
0525 {
0526 u64 tmpnode = node_num;
0527
0528
0529 tmpnode <<= 56;
0530
0531 spin_lock(&dlm_cookie_lock);
0532 *cookie = (dlm_next_cookie | tmpnode);
0533 if (++dlm_next_cookie & 0xff00000000000000ull) {
0534 mlog(0, "This node's cookie will now wrap!\n");
0535 dlm_next_cookie = 1;
0536 }
0537 spin_unlock(&dlm_cookie_lock);
0538 }
0539
0540 enum dlm_status dlmlock(struct dlm_ctxt *dlm, int mode,
0541 struct dlm_lockstatus *lksb, int flags,
0542 const char *name, int namelen, dlm_astlockfunc_t *ast,
0543 void *data, dlm_bastlockfunc_t *bast)
0544 {
0545 enum dlm_status status;
0546 struct dlm_lock_resource *res = NULL;
0547 struct dlm_lock *lock = NULL;
0548 int convert = 0, recovery = 0;
0549
0550
0551
0552
0553 if (!lksb) {
0554 dlm_error(DLM_BADARGS);
0555 return DLM_BADARGS;
0556 }
0557
0558 status = DLM_BADPARAM;
0559 if (mode != LKM_EXMODE && mode != LKM_PRMODE && mode != LKM_NLMODE) {
0560 dlm_error(status);
0561 goto error;
0562 }
0563
0564 if (flags & ~LKM_VALID_FLAGS) {
0565 dlm_error(status);
0566 goto error;
0567 }
0568
0569 convert = (flags & LKM_CONVERT);
0570 recovery = (flags & LKM_RECOVERY);
0571
0572 if (recovery &&
0573 (!dlm_is_recovery_lock(name, namelen) || convert) ) {
0574 dlm_error(status);
0575 goto error;
0576 }
0577 if (convert && (flags & LKM_LOCAL)) {
0578 mlog(ML_ERROR, "strange LOCAL convert request!\n");
0579 goto error;
0580 }
0581
0582 if (convert) {
0583
0584
0585
0586 lock = lksb->lockid;
0587 if (!lock) {
0588 mlog(ML_ERROR, "NULL lock pointer in convert "
0589 "request\n");
0590 goto error;
0591 }
0592
0593 res = lock->lockres;
0594 if (!res) {
0595 mlog(ML_ERROR, "NULL lockres pointer in convert "
0596 "request\n");
0597 goto error;
0598 }
0599 dlm_lockres_get(res);
0600
0601
0602
0603
0604
0605
0606 if (lock->lksb != lksb || lock->ast != ast ||
0607 lock->bast != bast || lock->astdata != data) {
0608 status = DLM_BADARGS;
0609 mlog(ML_ERROR, "new args: lksb=%p, ast=%p, bast=%p, "
0610 "astdata=%p\n", lksb, ast, bast, data);
0611 mlog(ML_ERROR, "orig args: lksb=%p, ast=%p, bast=%p, "
0612 "astdata=%p\n", lock->lksb, lock->ast,
0613 lock->bast, lock->astdata);
0614 goto error;
0615 }
0616 retry_convert:
0617 dlm_wait_for_recovery(dlm);
0618
0619 if (res->owner == dlm->node_num)
0620 status = dlmconvert_master(dlm, res, lock, flags, mode);
0621 else
0622 status = dlmconvert_remote(dlm, res, lock, flags, mode);
0623 if (status == DLM_RECOVERING || status == DLM_MIGRATING ||
0624 status == DLM_FORWARD) {
0625
0626
0627
0628
0629 mlog(0, "retrying convert with migration/recovery/"
0630 "in-progress\n");
0631 msleep(100);
0632 goto retry_convert;
0633 }
0634 } else {
0635 u64 tmpcookie;
0636
0637
0638 status = DLM_BADARGS;
0639 if (!name) {
0640 dlm_error(status);
0641 goto error;
0642 }
0643
0644 status = DLM_IVBUFLEN;
0645 if (namelen > DLM_LOCKID_NAME_MAX || namelen < 1) {
0646 dlm_error(status);
0647 goto error;
0648 }
0649
0650 dlm_get_next_cookie(dlm->node_num, &tmpcookie);
0651 lock = dlm_new_lock(mode, dlm->node_num, tmpcookie, lksb);
0652 if (!lock) {
0653 dlm_error(status);
0654 goto error;
0655 }
0656
0657 if (!recovery)
0658 dlm_wait_for_recovery(dlm);
0659
0660
0661 res = dlm_get_lock_resource(dlm, name, namelen, flags);
0662 if (!res) {
0663 status = DLM_IVLOCKID;
0664 dlm_error(status);
0665 goto error;
0666 }
0667
0668 mlog(0, "type=%d, flags = 0x%x\n", mode, flags);
0669 mlog(0, "creating lock: lock=%p res=%p\n", lock, res);
0670
0671 dlm_lock_attach_lockres(lock, res);
0672 lock->ast = ast;
0673 lock->bast = bast;
0674 lock->astdata = data;
0675
0676 retry_lock:
0677 if (flags & LKM_VALBLK) {
0678 mlog(0, "LKM_VALBLK passed by caller\n");
0679
0680
0681
0682 if (mode < LKM_PRMODE)
0683 flags &= ~LKM_VALBLK;
0684 else {
0685 flags |= LKM_GET_LVB;
0686 lock->lksb->flags |= DLM_LKSB_GET_LVB;
0687 }
0688 }
0689
0690 if (res->owner == dlm->node_num)
0691 status = dlmlock_master(dlm, res, lock, flags);
0692 else
0693 status = dlmlock_remote(dlm, res, lock, flags);
0694
0695 if (status == DLM_RECOVERING || status == DLM_MIGRATING ||
0696 status == DLM_FORWARD) {
0697 msleep(100);
0698 if (recovery) {
0699 if (status != DLM_RECOVERING)
0700 goto retry_lock;
0701
0702
0703
0704 dlm_wait_for_node_death(dlm, res->owner,
0705 DLM_NODE_DEATH_WAIT_MAX);
0706 } else {
0707 dlm_wait_for_recovery(dlm);
0708 goto retry_lock;
0709 }
0710 }
0711
0712
0713 spin_lock(&res->spinlock);
0714 dlm_lockres_drop_inflight_ref(dlm, res);
0715 spin_unlock(&res->spinlock);
0716
0717 dlm_lockres_calc_usage(dlm, res);
0718 dlm_kick_thread(dlm, res);
0719
0720 if (status != DLM_NORMAL) {
0721 lock->lksb->flags &= ~DLM_LKSB_GET_LVB;
0722 if (status != DLM_NOTQUEUED)
0723 dlm_error(status);
0724 goto error;
0725 }
0726 }
0727
0728 error:
0729 if (status != DLM_NORMAL) {
0730 if (lock && !convert)
0731 dlm_lock_put(lock);
0732
0733 lksb->status = status;
0734 }
0735
0736
0737
0738 if (res)
0739 dlm_lockres_put(res);
0740
0741 return status;
0742 }
0743 EXPORT_SYMBOL_GPL(dlmlock);