1/* -*- mode: c; c-basic-offset: 8; -*- 2 * vim: noexpandtab sw=8 ts=8 sts=0: 3 * 4 * dlmcommon.h 5 * 6 * Copyright (C) 2004 Oracle. All rights reserved. 7 * 8 * This program is free software; you can redistribute it and/or 9 * modify it under the terms of the GNU General Public 10 * License as published by the Free Software Foundation; either 11 * version 2 of the License, or (at your option) any later version. 12 * 13 * This program is distributed in the hope that it will be useful, 14 * but WITHOUT ANY WARRANTY; without even the implied warranty of 15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 16 * General Public License for more details. 17 * 18 * You should have received a copy of the GNU General Public 19 * License along with this program; if not, write to the 20 * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 21 * Boston, MA 021110-1307, USA. 22 * 23 */ 24 25#ifndef DLMCOMMON_H 26#define DLMCOMMON_H 27 28#include <linux/kref.h> 29 30#define DLM_HB_NODE_DOWN_PRI (0xf000000) 31#define DLM_HB_NODE_UP_PRI (0x8000000) 32 33#define DLM_LOCKID_NAME_MAX 32 34 35#define DLM_DOMAIN_NAME_MAX_LEN 255 36#define DLM_LOCK_RES_OWNER_UNKNOWN O2NM_MAX_NODES 37#define DLM_THREAD_SHUFFLE_INTERVAL 5 // flush everything every 5 passes 38#define DLM_THREAD_MS 200 // flush at least every 200 ms 39 40#define DLM_HASH_SIZE_DEFAULT (1 << 17) 41#if DLM_HASH_SIZE_DEFAULT < PAGE_SIZE 42# define DLM_HASH_PAGES 1 43#else 44# define DLM_HASH_PAGES (DLM_HASH_SIZE_DEFAULT / PAGE_SIZE) 45#endif 46#define DLM_BUCKETS_PER_PAGE (PAGE_SIZE / sizeof(struct hlist_head)) 47#define DLM_HASH_BUCKETS (DLM_HASH_PAGES * DLM_BUCKETS_PER_PAGE) 48 49/* Intended to make it easier for us to switch out hash functions */ 50#define dlm_lockid_hash(_n, _l) full_name_hash(_n, _l) 51 52enum dlm_mle_type { 53 DLM_MLE_BLOCK, 54 DLM_MLE_MASTER, 55 DLM_MLE_MIGRATION, 56 DLM_MLE_NUM_TYPES 57}; 58 59struct dlm_master_list_entry { 60 struct hlist_node master_hash_node; 61 struct list_head hb_events; 62 struct dlm_ctxt *dlm; 63 spinlock_t spinlock; 64 wait_queue_head_t wq; 65 atomic_t woken; 66 struct kref mle_refs; 67 int inuse; 68 unsigned long maybe_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; 69 unsigned long vote_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; 70 unsigned long response_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; 71 unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; 72 u8 master; 73 u8 new_master; 74 enum dlm_mle_type type; 75 struct o2hb_callback_func mle_hb_up; 76 struct o2hb_callback_func mle_hb_down; 77 struct dlm_lock_resource *mleres; 78 unsigned char mname[DLM_LOCKID_NAME_MAX]; 79 unsigned int mnamelen; 80 unsigned int mnamehash; 81}; 82 83enum dlm_ast_type { 84 DLM_AST = 0, 85 DLM_BAST, 86 DLM_ASTUNLOCK 87}; 88 89 90#define LKM_VALID_FLAGS (LKM_VALBLK | LKM_CONVERT | LKM_UNLOCK | \ 91 LKM_CANCEL | LKM_INVVALBLK | LKM_FORCE | \ 92 LKM_RECOVERY | LKM_LOCAL | LKM_NOQUEUE) 93 94#define DLM_RECOVERY_LOCK_NAME "$RECOVERY" 95#define DLM_RECOVERY_LOCK_NAME_LEN 9 96 97static inline int dlm_is_recovery_lock(const char *lock_name, int name_len) 98{ 99 if (name_len == DLM_RECOVERY_LOCK_NAME_LEN && 100 memcmp(lock_name, DLM_RECOVERY_LOCK_NAME, name_len)==0) 101 return 1; 102 return 0; 103} 104 105#define DLM_RECO_STATE_ACTIVE 0x0001 106#define DLM_RECO_STATE_FINALIZE 0x0002 107 108struct dlm_recovery_ctxt 109{ 110 struct list_head resources; 111 struct list_head received; 112 struct list_head node_data; 113 u8 new_master; 114 u8 dead_node; 115 u16 state; 116 unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; 117 wait_queue_head_t event; 118}; 119 120enum dlm_ctxt_state { 121 DLM_CTXT_NEW = 0, 122 DLM_CTXT_JOINED, 123 DLM_CTXT_IN_SHUTDOWN, 124 DLM_CTXT_LEAVING, 125}; 126 127struct dlm_ctxt 128{ 129 struct list_head list; 130 struct hlist_head **lockres_hash; 131 struct list_head dirty_list; 132 struct list_head purge_list; 133 struct list_head pending_asts; 134 struct list_head pending_basts; 135 struct list_head tracking_list; 136 unsigned int purge_count; 137 spinlock_t spinlock; 138 spinlock_t ast_lock; 139 spinlock_t track_lock; 140 char *name; 141 u8 node_num; 142 u32 key; 143 u8 joining_node; 144 wait_queue_head_t dlm_join_events; 145 unsigned long live_nodes_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; 146 unsigned long domain_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; 147 unsigned long recovery_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; 148 struct dlm_recovery_ctxt reco; 149 spinlock_t master_lock; 150 struct hlist_head **master_hash; 151 struct list_head mle_hb_events; 152 153 /* these give a really vague idea of the system load */ 154 atomic_t mle_tot_count[DLM_MLE_NUM_TYPES]; 155 atomic_t mle_cur_count[DLM_MLE_NUM_TYPES]; 156 atomic_t res_tot_count; 157 atomic_t res_cur_count; 158 159 struct dlm_debug_ctxt *dlm_debug_ctxt; 160 struct dentry *dlm_debugfs_subroot; 161 162 /* NOTE: Next three are protected by dlm_domain_lock */ 163 struct kref dlm_refs; 164 enum dlm_ctxt_state dlm_state; 165 unsigned int num_joins; 166 167 struct o2hb_callback_func dlm_hb_up; 168 struct o2hb_callback_func dlm_hb_down; 169 struct task_struct *dlm_thread_task; 170 struct task_struct *dlm_reco_thread_task; 171 struct workqueue_struct *dlm_worker; 172 wait_queue_head_t dlm_thread_wq; 173 wait_queue_head_t dlm_reco_thread_wq; 174 wait_queue_head_t ast_wq; 175 wait_queue_head_t migration_wq; 176 177 struct work_struct dispatched_work; 178 struct list_head work_list; 179 spinlock_t work_lock; 180 struct list_head dlm_domain_handlers; 181 struct list_head dlm_eviction_callbacks; 182 183 /* The filesystem specifies this at domain registration. We 184 * cache it here to know what to tell other nodes. */ 185 struct dlm_protocol_version fs_locking_proto; 186 /* This is the inter-dlm communication version */ 187 struct dlm_protocol_version dlm_locking_proto; 188}; 189 190static inline struct hlist_head *dlm_lockres_hash(struct dlm_ctxt *dlm, unsigned i) 191{ 192 return dlm->lockres_hash[(i / DLM_BUCKETS_PER_PAGE) % DLM_HASH_PAGES] + (i % DLM_BUCKETS_PER_PAGE); 193} 194 195static inline struct hlist_head *dlm_master_hash(struct dlm_ctxt *dlm, 196 unsigned i) 197{ 198 return dlm->master_hash[(i / DLM_BUCKETS_PER_PAGE) % DLM_HASH_PAGES] + 199 (i % DLM_BUCKETS_PER_PAGE); 200} 201 202/* these keventd work queue items are for less-frequently 203 * called functions that cannot be directly called from the 204 * net message handlers for some reason, usually because 205 * they need to send net messages of their own. */ 206void dlm_dispatch_work(struct work_struct *work); 207 208struct dlm_lock_resource; 209struct dlm_work_item; 210 211typedef void (dlm_workfunc_t)(struct dlm_work_item *, void *); 212 213struct dlm_request_all_locks_priv 214{ 215 u8 reco_master; 216 u8 dead_node; 217}; 218 219struct dlm_mig_lockres_priv 220{ 221 struct dlm_lock_resource *lockres; 222 u8 real_master; 223 u8 extra_ref; 224}; 225 226struct dlm_assert_master_priv 227{ 228 struct dlm_lock_resource *lockres; 229 u8 request_from; 230 u32 flags; 231 unsigned ignore_higher:1; 232}; 233 234struct dlm_deref_lockres_priv 235{ 236 struct dlm_lock_resource *deref_res; 237 u8 deref_node; 238}; 239 240struct dlm_work_item 241{ 242 struct list_head list; 243 dlm_workfunc_t *func; 244 struct dlm_ctxt *dlm; 245 void *data; 246 union { 247 struct dlm_request_all_locks_priv ral; 248 struct dlm_mig_lockres_priv ml; 249 struct dlm_assert_master_priv am; 250 struct dlm_deref_lockres_priv dl; 251 } u; 252}; 253 254static inline void dlm_init_work_item(struct dlm_ctxt *dlm, 255 struct dlm_work_item *i, 256 dlm_workfunc_t *f, void *data) 257{ 258 memset(i, 0, sizeof(*i)); 259 i->func = f; 260 INIT_LIST_HEAD(&i->list); 261 i->data = data; 262 i->dlm = dlm; /* must have already done a dlm_grab on this! */ 263} 264 265 266 267static inline void __dlm_set_joining_node(struct dlm_ctxt *dlm, 268 u8 node) 269{ 270 assert_spin_locked(&dlm->spinlock); 271 272 dlm->joining_node = node; 273 wake_up(&dlm->dlm_join_events); 274} 275 276#define DLM_LOCK_RES_UNINITED 0x00000001 277#define DLM_LOCK_RES_RECOVERING 0x00000002 278#define DLM_LOCK_RES_READY 0x00000004 279#define DLM_LOCK_RES_DIRTY 0x00000008 280#define DLM_LOCK_RES_IN_PROGRESS 0x00000010 281#define DLM_LOCK_RES_MIGRATING 0x00000020 282#define DLM_LOCK_RES_DROPPING_REF 0x00000040 283#define DLM_LOCK_RES_BLOCK_DIRTY 0x00001000 284#define DLM_LOCK_RES_SETREF_INPROG 0x00002000 285 286/* max milliseconds to wait to sync up a network failure with a node death */ 287#define DLM_NODE_DEATH_WAIT_MAX (5 * 1000) 288 289#define DLM_PURGE_INTERVAL_MS (8 * 1000) 290 291struct dlm_lock_resource 292{ 293 /* WARNING: Please see the comment in dlm_init_lockres before 294 * adding fields here. */ 295 struct hlist_node hash_node; 296 struct qstr lockname; 297 struct kref refs; 298 299 /* 300 * Please keep granted, converting, and blocked in this order, 301 * as some funcs want to iterate over all lists. 302 * 303 * All four lists are protected by the hash's reference. 304 */ 305 struct list_head granted; 306 struct list_head converting; 307 struct list_head blocked; 308 struct list_head purge; 309 310 /* 311 * These two lists require you to hold an additional reference 312 * while they are on the list. 313 */ 314 struct list_head dirty; 315 struct list_head recovering; // dlm_recovery_ctxt.resources list 316 317 /* Added during init and removed during release */ 318 struct list_head tracking; /* dlm->tracking_list */ 319 320 /* unused lock resources have their last_used stamped and are 321 * put on a list for the dlm thread to run. */ 322 unsigned long last_used; 323 324 struct dlm_ctxt *dlm; 325 326 unsigned migration_pending:1; 327 atomic_t asts_reserved; 328 spinlock_t spinlock; 329 wait_queue_head_t wq; 330 u8 owner; //node which owns the lock resource, or unknown 331 u16 state; 332 char lvb[DLM_LVB_LEN]; 333 unsigned int inflight_locks; 334 unsigned long refmap[BITS_TO_LONGS(O2NM_MAX_NODES)]; 335}; 336 337struct dlm_migratable_lock 338{ 339 __be64 cookie; 340 341 /* these 3 are just padding for the in-memory structure, but 342 * list and flags are actually used when sent over the wire */ 343 __be16 pad1; 344 u8 list; // 0=granted, 1=converting, 2=blocked 345 u8 flags; 346 347 s8 type; 348 s8 convert_type; 349 s8 highest_blocked; 350 u8 node; 351}; // 16 bytes 352 353struct dlm_lock 354{ 355 struct dlm_migratable_lock ml; 356 357 struct list_head list; 358 struct list_head ast_list; 359 struct list_head bast_list; 360 struct dlm_lock_resource *lockres; 361 spinlock_t spinlock; 362 struct kref lock_refs; 363 364 // ast and bast must be callable while holding a spinlock! 365 dlm_astlockfunc_t *ast; 366 dlm_bastlockfunc_t *bast; 367 void *astdata; 368 struct dlm_lockstatus *lksb; 369 unsigned ast_pending:1, 370 bast_pending:1, 371 convert_pending:1, 372 lock_pending:1, 373 cancel_pending:1, 374 unlock_pending:1, 375 lksb_kernel_allocated:1; 376}; 377 378 379#define DLM_LKSB_UNUSED1 0x01 380#define DLM_LKSB_PUT_LVB 0x02 381#define DLM_LKSB_GET_LVB 0x04 382#define DLM_LKSB_UNUSED2 0x08 383#define DLM_LKSB_UNUSED3 0x10 384#define DLM_LKSB_UNUSED4 0x20 385#define DLM_LKSB_UNUSED5 0x40 386#define DLM_LKSB_UNUSED6 0x80 387 388 389enum dlm_lockres_list { 390 DLM_GRANTED_LIST = 0, 391 DLM_CONVERTING_LIST, 392 DLM_BLOCKED_LIST 393}; 394 395static inline int dlm_lvb_is_empty(char *lvb) 396{ 397 int i; 398 for (i=0; i<DLM_LVB_LEN; i++) 399 if (lvb[i]) 400 return 0; 401 return 1; 402} 403 404static inline struct list_head * 405dlm_list_idx_to_ptr(struct dlm_lock_resource *res, enum dlm_lockres_list idx) 406{ 407 struct list_head *ret = NULL; 408 if (idx == DLM_GRANTED_LIST) 409 ret = &res->granted; 410 else if (idx == DLM_CONVERTING_LIST) 411 ret = &res->converting; 412 else if (idx == DLM_BLOCKED_LIST) 413 ret = &res->blocked; 414 else 415 BUG(); 416 return ret; 417} 418 419 420 421 422struct dlm_node_iter 423{ 424 unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; 425 int curnode; 426}; 427 428 429enum { 430 DLM_MASTER_REQUEST_MSG = 500, 431 DLM_UNUSED_MSG1, /* 501 */ 432 DLM_ASSERT_MASTER_MSG, /* 502 */ 433 DLM_CREATE_LOCK_MSG, /* 503 */ 434 DLM_CONVERT_LOCK_MSG, /* 504 */ 435 DLM_PROXY_AST_MSG, /* 505 */ 436 DLM_UNLOCK_LOCK_MSG, /* 506 */ 437 DLM_DEREF_LOCKRES_MSG, /* 507 */ 438 DLM_MIGRATE_REQUEST_MSG, /* 508 */ 439 DLM_MIG_LOCKRES_MSG, /* 509 */ 440 DLM_QUERY_JOIN_MSG, /* 510 */ 441 DLM_ASSERT_JOINED_MSG, /* 511 */ 442 DLM_CANCEL_JOIN_MSG, /* 512 */ 443 DLM_EXIT_DOMAIN_MSG, /* 513 */ 444 DLM_MASTER_REQUERY_MSG, /* 514 */ 445 DLM_LOCK_REQUEST_MSG, /* 515 */ 446 DLM_RECO_DATA_DONE_MSG, /* 516 */ 447 DLM_BEGIN_RECO_MSG, /* 517 */ 448 DLM_FINALIZE_RECO_MSG /* 518 */ 449}; 450 451struct dlm_reco_node_data 452{ 453 int state; 454 u8 node_num; 455 struct list_head list; 456}; 457 458enum { 459 DLM_RECO_NODE_DATA_DEAD = -1, 460 DLM_RECO_NODE_DATA_INIT = 0, 461 DLM_RECO_NODE_DATA_REQUESTING, 462 DLM_RECO_NODE_DATA_REQUESTED, 463 DLM_RECO_NODE_DATA_RECEIVING, 464 DLM_RECO_NODE_DATA_DONE, 465 DLM_RECO_NODE_DATA_FINALIZE_SENT, 466}; 467 468 469enum { 470 DLM_MASTER_RESP_NO = 0, 471 DLM_MASTER_RESP_YES, 472 DLM_MASTER_RESP_MAYBE, 473 DLM_MASTER_RESP_ERROR 474}; 475 476 477struct dlm_master_request 478{ 479 u8 node_idx; 480 u8 namelen; 481 __be16 pad1; 482 __be32 flags; 483 484 u8 name[O2NM_MAX_NAME_LEN]; 485}; 486 487#define DLM_ASSERT_RESPONSE_REASSERT 0x00000001 488#define DLM_ASSERT_RESPONSE_MASTERY_REF 0x00000002 489 490#define DLM_ASSERT_MASTER_MLE_CLEANUP 0x00000001 491#define DLM_ASSERT_MASTER_REQUERY 0x00000002 492#define DLM_ASSERT_MASTER_FINISH_MIGRATION 0x00000004 493struct dlm_assert_master 494{ 495 u8 node_idx; 496 u8 namelen; 497 __be16 pad1; 498 __be32 flags; 499 500 u8 name[O2NM_MAX_NAME_LEN]; 501}; 502 503#define DLM_MIGRATE_RESPONSE_MASTERY_REF 0x00000001 504 505struct dlm_migrate_request 506{ 507 u8 master; 508 u8 new_master; 509 u8 namelen; 510 u8 pad1; 511 __be32 pad2; 512 u8 name[O2NM_MAX_NAME_LEN]; 513}; 514 515struct dlm_master_requery 516{ 517 u8 pad1; 518 u8 pad2; 519 u8 node_idx; 520 u8 namelen; 521 __be32 pad3; 522 u8 name[O2NM_MAX_NAME_LEN]; 523}; 524 525#define DLM_MRES_RECOVERY 0x01 526#define DLM_MRES_MIGRATION 0x02 527#define DLM_MRES_ALL_DONE 0x04 528 529/* 530 * We would like to get one whole lockres into a single network 531 * message whenever possible. Generally speaking, there will be 532 * at most one dlm_lock on a lockres for each node in the cluster, 533 * plus (infrequently) any additional locks coming in from userdlm. 534 * 535 * struct _dlm_lockres_page 536 * { 537 * dlm_migratable_lockres mres; 538 * dlm_migratable_lock ml[DLM_MAX_MIGRATABLE_LOCKS]; 539 * u8 pad[DLM_MIG_LOCKRES_RESERVED]; 540 * }; 541 * 542 * from ../cluster/tcp.h 543 * NET_MAX_PAYLOAD_BYTES (4096 - sizeof(net_msg)) 544 * (roughly 4080 bytes) 545 * and sizeof(dlm_migratable_lockres) = 112 bytes 546 * and sizeof(dlm_migratable_lock) = 16 bytes 547 * 548 * Choosing DLM_MAX_MIGRATABLE_LOCKS=240 and 549 * DLM_MIG_LOCKRES_RESERVED=128 means we have this: 550 * 551 * (DLM_MAX_MIGRATABLE_LOCKS * sizeof(dlm_migratable_lock)) + 552 * sizeof(dlm_migratable_lockres) + DLM_MIG_LOCKRES_RESERVED = 553 * NET_MAX_PAYLOAD_BYTES 554 * (240 * 16) + 112 + 128 = 4080 555 * 556 * So a lockres would need more than 240 locks before it would 557 * use more than one network packet to recover. Not too bad. 558 */ 559#define DLM_MAX_MIGRATABLE_LOCKS 240 560 561struct dlm_migratable_lockres 562{ 563 u8 master; 564 u8 lockname_len; 565 u8 num_locks; // locks sent in this structure 566 u8 flags; 567 __be32 total_locks; // locks to be sent for this migration cookie 568 __be64 mig_cookie; // cookie for this lockres migration 569 // or zero if not needed 570 // 16 bytes 571 u8 lockname[DLM_LOCKID_NAME_MAX]; 572 // 48 bytes 573 u8 lvb[DLM_LVB_LEN]; 574 // 112 bytes 575 struct dlm_migratable_lock ml[0]; // 16 bytes each, begins at byte 112 576}; 577#define DLM_MIG_LOCKRES_MAX_LEN \ 578 (sizeof(struct dlm_migratable_lockres) + \ 579 (sizeof(struct dlm_migratable_lock) * \ 580 DLM_MAX_MIGRATABLE_LOCKS) ) 581 582/* from above, 128 bytes 583 * for some undetermined future use */ 584#define DLM_MIG_LOCKRES_RESERVED (NET_MAX_PAYLOAD_BYTES - \ 585 DLM_MIG_LOCKRES_MAX_LEN) 586 587struct dlm_create_lock 588{ 589 __be64 cookie; 590 591 __be32 flags; 592 u8 pad1; 593 u8 node_idx; 594 s8 requested_type; 595 u8 namelen; 596 597 u8 name[O2NM_MAX_NAME_LEN]; 598}; 599 600struct dlm_convert_lock 601{ 602 __be64 cookie; 603 604 __be32 flags; 605 u8 pad1; 606 u8 node_idx; 607 s8 requested_type; 608 u8 namelen; 609 610 u8 name[O2NM_MAX_NAME_LEN]; 611 612 s8 lvb[0]; 613}; 614#define DLM_CONVERT_LOCK_MAX_LEN (sizeof(struct dlm_convert_lock)+DLM_LVB_LEN) 615 616struct dlm_unlock_lock 617{ 618 __be64 cookie; 619 620 __be32 flags; 621 __be16 pad1; 622 u8 node_idx; 623 u8 namelen; 624 625 u8 name[O2NM_MAX_NAME_LEN]; 626 627 s8 lvb[0]; 628}; 629#define DLM_UNLOCK_LOCK_MAX_LEN (sizeof(struct dlm_unlock_lock)+DLM_LVB_LEN) 630 631struct dlm_proxy_ast 632{ 633 __be64 cookie; 634 635 __be32 flags; 636 u8 node_idx; 637 u8 type; 638 u8 blocked_type; 639 u8 namelen; 640 641 u8 name[O2NM_MAX_NAME_LEN]; 642 643 s8 lvb[0]; 644}; 645#define DLM_PROXY_AST_MAX_LEN (sizeof(struct dlm_proxy_ast)+DLM_LVB_LEN) 646 647#define DLM_MOD_KEY (0x666c6172) 648enum dlm_query_join_response_code { 649 JOIN_DISALLOW = 0, 650 JOIN_OK, 651 JOIN_OK_NO_MAP, 652 JOIN_PROTOCOL_MISMATCH, 653}; 654 655struct dlm_query_join_packet { 656 u8 code; /* Response code. dlm_minor and fs_minor 657 are only valid if this is JOIN_OK */ 658 u8 dlm_minor; /* The minor version of the protocol the 659 dlm is speaking. */ 660 u8 fs_minor; /* The minor version of the protocol the 661 filesystem is speaking. */ 662 u8 reserved; 663}; 664 665union dlm_query_join_response { 666 u32 intval; 667 struct dlm_query_join_packet packet; 668}; 669 670struct dlm_lock_request 671{ 672 u8 node_idx; 673 u8 dead_node; 674 __be16 pad1; 675 __be32 pad2; 676}; 677 678struct dlm_reco_data_done 679{ 680 u8 node_idx; 681 u8 dead_node; 682 __be16 pad1; 683 __be32 pad2; 684 685 /* unused for now */ 686 /* eventually we can use this to attempt 687 * lvb recovery based on each node's info */ 688 u8 reco_lvb[DLM_LVB_LEN]; 689}; 690 691struct dlm_begin_reco 692{ 693 u8 node_idx; 694 u8 dead_node; 695 __be16 pad1; 696 __be32 pad2; 697}; 698 699 700#define BITS_PER_BYTE 8 701#define BITS_TO_BYTES(bits) (((bits)+BITS_PER_BYTE-1)/BITS_PER_BYTE) 702 703struct dlm_query_join_request 704{ 705 u8 node_idx; 706 u8 pad1[2]; 707 u8 name_len; 708 struct dlm_protocol_version dlm_proto; 709 struct dlm_protocol_version fs_proto; 710 u8 domain[O2NM_MAX_NAME_LEN]; 711 u8 node_map[BITS_TO_BYTES(O2NM_MAX_NODES)]; 712}; 713 714struct dlm_assert_joined 715{ 716 u8 node_idx; 717 u8 pad1[2]; 718 u8 name_len; 719 u8 domain[O2NM_MAX_NAME_LEN]; 720}; 721 722struct dlm_cancel_join 723{ 724 u8 node_idx; 725 u8 pad1[2]; 726 u8 name_len; 727 u8 domain[O2NM_MAX_NAME_LEN]; 728}; 729 730struct dlm_exit_domain 731{ 732 u8 node_idx; 733 u8 pad1[3]; 734}; 735 736struct dlm_finalize_reco 737{ 738 u8 node_idx; 739 u8 dead_node; 740 u8 flags; 741 u8 pad1; 742 __be32 pad2; 743}; 744 745struct dlm_deref_lockres 746{ 747 u32 pad1; 748 u16 pad2; 749 u8 node_idx; 750 u8 namelen; 751 752 u8 name[O2NM_MAX_NAME_LEN]; 753}; 754 755static inline enum dlm_status 756__dlm_lockres_state_to_status(struct dlm_lock_resource *res) 757{ 758 enum dlm_status status = DLM_NORMAL; 759 760 assert_spin_locked(&res->spinlock); 761 762 if (res->state & DLM_LOCK_RES_RECOVERING) 763 status = DLM_RECOVERING; 764 else if (res->state & DLM_LOCK_RES_MIGRATING) 765 status = DLM_MIGRATING; 766 else if (res->state & DLM_LOCK_RES_IN_PROGRESS) 767 status = DLM_FORWARD; 768 769 return status; 770} 771 772static inline u8 dlm_get_lock_cookie_node(u64 cookie) 773{ 774 u8 ret; 775 cookie >>= 56; 776 ret = (u8)(cookie & 0xffULL); 777 return ret; 778} 779 780static inline unsigned long long dlm_get_lock_cookie_seq(u64 cookie) 781{ 782 unsigned long long ret; 783 ret = ((unsigned long long)cookie) & 0x00ffffffffffffffULL; 784 return ret; 785} 786 787struct dlm_lock * dlm_new_lock(int type, u8 node, u64 cookie, 788 struct dlm_lockstatus *lksb); 789void dlm_lock_get(struct dlm_lock *lock); 790void dlm_lock_put(struct dlm_lock *lock); 791 792void dlm_lock_attach_lockres(struct dlm_lock *lock, 793 struct dlm_lock_resource *res); 794 795int dlm_create_lock_handler(struct o2net_msg *msg, u32 len, void *data, 796 void **ret_data); 797int dlm_convert_lock_handler(struct o2net_msg *msg, u32 len, void *data, 798 void **ret_data); 799int dlm_proxy_ast_handler(struct o2net_msg *msg, u32 len, void *data, 800 void **ret_data); 801 802void dlm_revert_pending_convert(struct dlm_lock_resource *res, 803 struct dlm_lock *lock); 804void dlm_revert_pending_lock(struct dlm_lock_resource *res, 805 struct dlm_lock *lock); 806 807int dlm_unlock_lock_handler(struct o2net_msg *msg, u32 len, void *data, 808 void **ret_data); 809void dlm_commit_pending_cancel(struct dlm_lock_resource *res, 810 struct dlm_lock *lock); 811void dlm_commit_pending_unlock(struct dlm_lock_resource *res, 812 struct dlm_lock *lock); 813 814int dlm_launch_thread(struct dlm_ctxt *dlm); 815void dlm_complete_thread(struct dlm_ctxt *dlm); 816int dlm_launch_recovery_thread(struct dlm_ctxt *dlm); 817void dlm_complete_recovery_thread(struct dlm_ctxt *dlm); 818void dlm_wait_for_recovery(struct dlm_ctxt *dlm); 819void dlm_kick_recovery_thread(struct dlm_ctxt *dlm); 820int dlm_is_node_dead(struct dlm_ctxt *dlm, u8 node); 821int dlm_wait_for_node_death(struct dlm_ctxt *dlm, u8 node, int timeout); 822int dlm_wait_for_node_recovery(struct dlm_ctxt *dlm, u8 node, int timeout); 823 824void dlm_put(struct dlm_ctxt *dlm); 825struct dlm_ctxt *dlm_grab(struct dlm_ctxt *dlm); 826int dlm_domain_fully_joined(struct dlm_ctxt *dlm); 827 828void __dlm_lockres_calc_usage(struct dlm_ctxt *dlm, 829 struct dlm_lock_resource *res); 830void dlm_lockres_calc_usage(struct dlm_ctxt *dlm, 831 struct dlm_lock_resource *res); 832static inline void dlm_lockres_get(struct dlm_lock_resource *res) 833{ 834 /* This is called on every lookup, so it might be worth 835 * inlining. */ 836 kref_get(&res->refs); 837} 838void dlm_lockres_put(struct dlm_lock_resource *res); 839void __dlm_unhash_lockres(struct dlm_lock_resource *res); 840void __dlm_insert_lockres(struct dlm_ctxt *dlm, 841 struct dlm_lock_resource *res); 842struct dlm_lock_resource * __dlm_lookup_lockres_full(struct dlm_ctxt *dlm, 843 const char *name, 844 unsigned int len, 845 unsigned int hash); 846struct dlm_lock_resource * __dlm_lookup_lockres(struct dlm_ctxt *dlm, 847 const char *name, 848 unsigned int len, 849 unsigned int hash); 850struct dlm_lock_resource * dlm_lookup_lockres(struct dlm_ctxt *dlm, 851 const char *name, 852 unsigned int len); 853 854int dlm_is_host_down(int errno); 855 856struct dlm_lock_resource * dlm_get_lock_resource(struct dlm_ctxt *dlm, 857 const char *lockid, 858 int namelen, 859 int flags); 860struct dlm_lock_resource *dlm_new_lockres(struct dlm_ctxt *dlm, 861 const char *name, 862 unsigned int namelen); 863 864#define dlm_lockres_set_refmap_bit(bit,res) \ 865 __dlm_lockres_set_refmap_bit(bit,res,__FILE__,__LINE__) 866#define dlm_lockres_clear_refmap_bit(bit,res) \ 867 __dlm_lockres_clear_refmap_bit(bit,res,__FILE__,__LINE__) 868 869static inline void __dlm_lockres_set_refmap_bit(int bit, 870 struct dlm_lock_resource *res, 871 const char *file, 872 int line) 873{ 874 //printk("%s:%d:%.*s: setting bit %d\n", file, line, 875 // res->lockname.len, res->lockname.name, bit); 876 set_bit(bit, res->refmap); 877} 878 879static inline void __dlm_lockres_clear_refmap_bit(int bit, 880 struct dlm_lock_resource *res, 881 const char *file, 882 int line) 883{ 884 //printk("%s:%d:%.*s: clearing bit %d\n", file, line, 885 // res->lockname.len, res->lockname.name, bit); 886 clear_bit(bit, res->refmap); 887} 888 889void __dlm_lockres_drop_inflight_ref(struct dlm_ctxt *dlm, 890 struct dlm_lock_resource *res, 891 const char *file, 892 int line); 893void __dlm_lockres_grab_inflight_ref(struct dlm_ctxt *dlm, 894 struct dlm_lock_resource *res, 895 int new_lockres, 896 const char *file, 897 int line); 898#define dlm_lockres_drop_inflight_ref(d,r) \ 899 __dlm_lockres_drop_inflight_ref(d,r,__FILE__,__LINE__) 900#define dlm_lockres_grab_inflight_ref(d,r) \ 901 __dlm_lockres_grab_inflight_ref(d,r,0,__FILE__,__LINE__) 902#define dlm_lockres_grab_inflight_ref_new(d,r) \ 903 __dlm_lockres_grab_inflight_ref(d,r,1,__FILE__,__LINE__) 904 905void dlm_queue_ast(struct dlm_ctxt *dlm, struct dlm_lock *lock); 906void dlm_queue_bast(struct dlm_ctxt *dlm, struct dlm_lock *lock); 907void __dlm_queue_ast(struct dlm_ctxt *dlm, struct dlm_lock *lock); 908void __dlm_queue_bast(struct dlm_ctxt *dlm, struct dlm_lock *lock); 909void dlm_do_local_ast(struct dlm_ctxt *dlm, 910 struct dlm_lock_resource *res, 911 struct dlm_lock *lock); 912int dlm_do_remote_ast(struct dlm_ctxt *dlm, 913 struct dlm_lock_resource *res, 914 struct dlm_lock *lock); 915void dlm_do_local_bast(struct dlm_ctxt *dlm, 916 struct dlm_lock_resource *res, 917 struct dlm_lock *lock, 918 int blocked_type); 919int dlm_send_proxy_ast_msg(struct dlm_ctxt *dlm, 920 struct dlm_lock_resource *res, 921 struct dlm_lock *lock, 922 int msg_type, 923 int blocked_type, int flags); 924static inline int dlm_send_proxy_bast(struct dlm_ctxt *dlm, 925 struct dlm_lock_resource *res, 926 struct dlm_lock *lock, 927 int blocked_type) 928{ 929 return dlm_send_proxy_ast_msg(dlm, res, lock, DLM_BAST, 930 blocked_type, 0); 931} 932 933static inline int dlm_send_proxy_ast(struct dlm_ctxt *dlm, 934 struct dlm_lock_resource *res, 935 struct dlm_lock *lock, 936 int flags) 937{ 938 return dlm_send_proxy_ast_msg(dlm, res, lock, DLM_AST, 939 0, flags); 940} 941 942void dlm_print_one_lock_resource(struct dlm_lock_resource *res); 943void __dlm_print_one_lock_resource(struct dlm_lock_resource *res); 944 945u8 dlm_nm_this_node(struct dlm_ctxt *dlm); 946void dlm_kick_thread(struct dlm_ctxt *dlm, struct dlm_lock_resource *res); 947void __dlm_dirty_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res); 948 949 950int dlm_nm_init(struct dlm_ctxt *dlm); 951int dlm_heartbeat_init(struct dlm_ctxt *dlm); 952void dlm_hb_node_down_cb(struct o2nm_node *node, int idx, void *data); 953void dlm_hb_node_up_cb(struct o2nm_node *node, int idx, void *data); 954 955int dlm_empty_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res); 956int dlm_finish_migration(struct dlm_ctxt *dlm, 957 struct dlm_lock_resource *res, 958 u8 old_master); 959void dlm_lockres_release_ast(struct dlm_ctxt *dlm, 960 struct dlm_lock_resource *res); 961void __dlm_lockres_reserve_ast(struct dlm_lock_resource *res); 962 963int dlm_master_request_handler(struct o2net_msg *msg, u32 len, void *data, 964 void **ret_data); 965int dlm_assert_master_handler(struct o2net_msg *msg, u32 len, void *data, 966 void **ret_data); 967void dlm_assert_master_post_handler(int status, void *data, void *ret_data); 968int dlm_deref_lockres_handler(struct o2net_msg *msg, u32 len, void *data, 969 void **ret_data); 970int dlm_migrate_request_handler(struct o2net_msg *msg, u32 len, void *data, 971 void **ret_data); 972int dlm_mig_lockres_handler(struct o2net_msg *msg, u32 len, void *data, 973 void **ret_data); 974int dlm_master_requery_handler(struct o2net_msg *msg, u32 len, void *data, 975 void **ret_data); 976int dlm_request_all_locks_handler(struct o2net_msg *msg, u32 len, void *data, 977 void **ret_data); 978int dlm_reco_data_done_handler(struct o2net_msg *msg, u32 len, void *data, 979 void **ret_data); 980int dlm_begin_reco_handler(struct o2net_msg *msg, u32 len, void *data, 981 void **ret_data); 982int dlm_finalize_reco_handler(struct o2net_msg *msg, u32 len, void *data, 983 void **ret_data); 984int dlm_do_master_requery(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, 985 u8 nodenum, u8 *real_master); 986 987 988int dlm_dispatch_assert_master(struct dlm_ctxt *dlm, 989 struct dlm_lock_resource *res, 990 int ignore_higher, 991 u8 request_from, 992 u32 flags); 993 994 995int dlm_send_one_lockres(struct dlm_ctxt *dlm, 996 struct dlm_lock_resource *res, 997 struct dlm_migratable_lockres *mres, 998 u8 send_to, 999 u8 flags); 1000void dlm_move_lockres_to_recovery_list(struct dlm_ctxt *dlm, 1001 struct dlm_lock_resource *res); 1002 1003/* will exit holding res->spinlock, but may drop in function */ 1004void __dlm_wait_on_lockres_flags(struct dlm_lock_resource *res, int flags); 1005void __dlm_wait_on_lockres_flags_set(struct dlm_lock_resource *res, int flags); 1006 1007/* will exit holding res->spinlock, but may drop in function */ 1008static inline void __dlm_wait_on_lockres(struct dlm_lock_resource *res) 1009{ 1010 __dlm_wait_on_lockres_flags(res, (DLM_LOCK_RES_IN_PROGRESS| 1011 DLM_LOCK_RES_RECOVERING| 1012 DLM_LOCK_RES_MIGRATING)); 1013} 1014 1015void __dlm_unlink_mle(struct dlm_ctxt *dlm, struct dlm_master_list_entry *mle); 1016void __dlm_insert_mle(struct dlm_ctxt *dlm, struct dlm_master_list_entry *mle); 1017 1018/* create/destroy slab caches */ 1019int dlm_init_master_caches(void); 1020void dlm_destroy_master_caches(void); 1021 1022int dlm_init_lock_cache(void); 1023void dlm_destroy_lock_cache(void); 1024 1025int dlm_init_mle_cache(void); 1026void dlm_destroy_mle_cache(void); 1027 1028void dlm_hb_event_notify_attached(struct dlm_ctxt *dlm, int idx, int node_up); 1029int dlm_drop_lockres_ref(struct dlm_ctxt *dlm, 1030 struct dlm_lock_resource *res); 1031void dlm_clean_master_list(struct dlm_ctxt *dlm, 1032 u8 dead_node); 1033void dlm_force_free_mles(struct dlm_ctxt *dlm); 1034int dlm_lock_basts_flushed(struct dlm_ctxt *dlm, struct dlm_lock *lock); 1035int __dlm_lockres_has_locks(struct dlm_lock_resource *res); 1036int __dlm_lockres_unused(struct dlm_lock_resource *res); 1037 1038static inline const char * dlm_lock_mode_name(int mode) 1039{ 1040 switch (mode) { 1041 case LKM_EXMODE: 1042 return "EX"; 1043 case LKM_PRMODE: 1044 return "PR"; 1045 case LKM_NLMODE: 1046 return "NL"; 1047 } 1048 return "UNKNOWN"; 1049} 1050 1051 1052static inline int dlm_lock_compatible(int existing, int request) 1053{ 1054 /* NO_LOCK compatible with all */ 1055 if (request == LKM_NLMODE || 1056 existing == LKM_NLMODE) 1057 return 1; 1058 1059 /* EX incompatible with all non-NO_LOCK */ 1060 if (request == LKM_EXMODE) 1061 return 0; 1062 1063 /* request must be PR, which is compatible with PR */ 1064 if (existing == LKM_PRMODE) 1065 return 1; 1066 1067 return 0; 1068} 1069 1070static inline int dlm_lock_on_list(struct list_head *head, 1071 struct dlm_lock *lock) 1072{ 1073 struct list_head *iter; 1074 struct dlm_lock *tmplock; 1075 1076 list_for_each(iter, head) { 1077 tmplock = list_entry(iter, struct dlm_lock, list); 1078 if (tmplock == lock) 1079 return 1; 1080 } 1081 return 0; 1082} 1083 1084 1085static inline enum dlm_status dlm_err_to_dlm_status(int err) 1086{ 1087 enum dlm_status ret; 1088 if (err == -ENOMEM) 1089 ret = DLM_SYSERR; 1090 else if (err == -ETIMEDOUT || o2net_link_down(err, NULL)) 1091 ret = DLM_NOLOCKMGR; 1092 else if (err == -EINVAL) 1093 ret = DLM_BADPARAM; 1094 else if (err == -ENAMETOOLONG) 1095 ret = DLM_IVBUFLEN; 1096 else 1097 ret = DLM_BADARGS; 1098 return ret; 1099} 1100 1101 1102static inline void dlm_node_iter_init(unsigned long *map, 1103 struct dlm_node_iter *iter) 1104{ 1105 memcpy(iter->node_map, map, sizeof(iter->node_map)); 1106 iter->curnode = -1; 1107} 1108 1109static inline int dlm_node_iter_next(struct dlm_node_iter *iter) 1110{ 1111 int bit; 1112 bit = find_next_bit(iter->node_map, O2NM_MAX_NODES, iter->curnode+1); 1113 if (bit >= O2NM_MAX_NODES) { 1114 iter->curnode = O2NM_MAX_NODES; 1115 return -ENOENT; 1116 } 1117 iter->curnode = bit; 1118 return bit; 1119} 1120 1121static inline void dlm_set_lockres_owner(struct dlm_ctxt *dlm, 1122 struct dlm_lock_resource *res, 1123 u8 owner) 1124{ 1125 assert_spin_locked(&res->spinlock); 1126 1127 res->owner = owner; 1128} 1129 1130static inline void dlm_change_lockres_owner(struct dlm_ctxt *dlm, 1131 struct dlm_lock_resource *res, 1132 u8 owner) 1133{ 1134 assert_spin_locked(&res->spinlock); 1135 1136 if (owner != res->owner) 1137 dlm_set_lockres_owner(dlm, res, owner); 1138} 1139 1140#endif /* DLMCOMMON_H */ 1141