1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* -*- mode: c; c-basic-offset: 8; -*- 3 * vim: noexpandtab sw=8 ts=8 sts=0: 4 * 5 * dlmcommon.h 6 * 7 * Copyright (C) 2004 Oracle. All rights reserved. 8 */ 9 10 #ifndef DLMCOMMON_H 11 #define DLMCOMMON_H 12 13 #include <linux/kref.h> 14 15 #define DLM_HB_NODE_DOWN_PRI (0xf000000) 16 #define DLM_HB_NODE_UP_PRI (0x8000000) 17 18 #define DLM_LOCKID_NAME_MAX 32 19 20 #define DLM_DOMAIN_NAME_MAX_LEN 255 21 #define DLM_LOCK_RES_OWNER_UNKNOWN O2NM_MAX_NODES 22 #define DLM_THREAD_SHUFFLE_INTERVAL 5 // flush everything every 5 passes 23 #define DLM_THREAD_MS 200 // flush at least every 200 ms 24 25 #define DLM_HASH_SIZE_DEFAULT (1 << 17) 26 #if DLM_HASH_SIZE_DEFAULT < PAGE_SIZE 27 # define DLM_HASH_PAGES 1 28 #else 29 # define DLM_HASH_PAGES (DLM_HASH_SIZE_DEFAULT / PAGE_SIZE) 30 #endif 31 #define DLM_BUCKETS_PER_PAGE (PAGE_SIZE / sizeof(struct hlist_head)) 32 #define DLM_HASH_BUCKETS (DLM_HASH_PAGES * DLM_BUCKETS_PER_PAGE) 33 34 /* Intended to make it easier for us to switch out hash functions */ 35 #define dlm_lockid_hash(_n, _l) full_name_hash(NULL, _n, _l) 36 37 enum dlm_mle_type { 38 DLM_MLE_BLOCK = 0, 39 DLM_MLE_MASTER = 1, 40 DLM_MLE_MIGRATION = 2, 41 DLM_MLE_NUM_TYPES = 3, 42 }; 43 44 struct dlm_master_list_entry { 45 struct hlist_node master_hash_node; 46 struct list_head hb_events; 47 struct dlm_ctxt *dlm; 48 spinlock_t spinlock; 49 wait_queue_head_t wq; 50 atomic_t woken; 51 struct kref mle_refs; 52 int inuse; 53 unsigned long maybe_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; 54 unsigned long vote_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; 55 unsigned long response_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; 56 unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; 57 u8 master; 58 u8 new_master; 59 enum dlm_mle_type type; 60 struct o2hb_callback_func mle_hb_up; 61 struct o2hb_callback_func mle_hb_down; 62 struct dlm_lock_resource *mleres; 63 unsigned char mname[DLM_LOCKID_NAME_MAX]; 64 unsigned int mnamelen; 65 unsigned int mnamehash; 66 }; 67 68 enum dlm_ast_type { 69 DLM_AST = 0, 70 DLM_BAST = 1, 71 DLM_ASTUNLOCK = 2, 72 }; 73 74 75 #define LKM_VALID_FLAGS (LKM_VALBLK | LKM_CONVERT | LKM_UNLOCK | \ 76 LKM_CANCEL | LKM_INVVALBLK | LKM_FORCE | \ 77 LKM_RECOVERY | LKM_LOCAL | LKM_NOQUEUE) 78 79 #define DLM_RECOVERY_LOCK_NAME "$RECOVERY" 80 #define DLM_RECOVERY_LOCK_NAME_LEN 9 81 82 static inline int dlm_is_recovery_lock(const char *lock_name, int name_len) 83 { 84 if (name_len == DLM_RECOVERY_LOCK_NAME_LEN && 85 memcmp(lock_name, DLM_RECOVERY_LOCK_NAME, name_len)==0) 86 return 1; 87 return 0; 88 } 89 90 #define DLM_RECO_STATE_ACTIVE 0x0001 91 #define DLM_RECO_STATE_FINALIZE 0x0002 92 93 struct dlm_recovery_ctxt 94 { 95 struct list_head resources; 96 struct list_head node_data; 97 u8 new_master; 98 u8 dead_node; 99 u16 state; 100 unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; 101 wait_queue_head_t event; 102 }; 103 104 enum dlm_ctxt_state { 105 DLM_CTXT_NEW = 0, 106 DLM_CTXT_JOINED = 1, 107 DLM_CTXT_IN_SHUTDOWN = 2, 108 DLM_CTXT_LEAVING = 3, 109 }; 110 111 struct dlm_ctxt 112 { 113 struct list_head list; 114 struct hlist_head **lockres_hash; 115 struct list_head dirty_list; 116 struct list_head purge_list; 117 struct list_head pending_asts; 118 struct list_head pending_basts; 119 struct list_head tracking_list; 120 unsigned int purge_count; 121 spinlock_t spinlock; 122 spinlock_t ast_lock; 123 spinlock_t track_lock; 124 char *name; 125 u8 node_num; 126 u32 key; 127 u8 joining_node; 128 u8 migrate_done; /* set to 1 means node has migrated all lock resources */ 129 wait_queue_head_t dlm_join_events; 130 unsigned long live_nodes_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; 131 unsigned long domain_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; 132 unsigned long exit_domain_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; 133 unsigned long recovery_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; 134 struct dlm_recovery_ctxt reco; 135 spinlock_t master_lock; 136 struct hlist_head **master_hash; 137 struct list_head mle_hb_events; 138 139 /* these give a really vague idea of the system load */ 140 atomic_t mle_tot_count[DLM_MLE_NUM_TYPES]; 141 atomic_t mle_cur_count[DLM_MLE_NUM_TYPES]; 142 atomic_t res_tot_count; 143 atomic_t res_cur_count; 144 145 struct dlm_debug_ctxt *dlm_debug_ctxt; 146 struct dentry *dlm_debugfs_subroot; 147 148 /* NOTE: Next three are protected by dlm_domain_lock */ 149 struct kref dlm_refs; 150 enum dlm_ctxt_state dlm_state; 151 unsigned int num_joins; 152 153 struct o2hb_callback_func dlm_hb_up; 154 struct o2hb_callback_func dlm_hb_down; 155 struct task_struct *dlm_thread_task; 156 struct task_struct *dlm_reco_thread_task; 157 struct workqueue_struct *dlm_worker; 158 wait_queue_head_t dlm_thread_wq; 159 wait_queue_head_t dlm_reco_thread_wq; 160 wait_queue_head_t ast_wq; 161 wait_queue_head_t migration_wq; 162 163 struct work_struct dispatched_work; 164 struct list_head work_list; 165 spinlock_t work_lock; 166 struct list_head dlm_domain_handlers; 167 struct list_head dlm_eviction_callbacks; 168 169 /* The filesystem specifies this at domain registration. We 170 * cache it here to know what to tell other nodes. */ 171 struct dlm_protocol_version fs_locking_proto; 172 /* This is the inter-dlm communication version */ 173 struct dlm_protocol_version dlm_locking_proto; 174 }; 175 176 static inline struct hlist_head *dlm_lockres_hash(struct dlm_ctxt *dlm, unsigned i) 177 { 178 return dlm->lockres_hash[(i / DLM_BUCKETS_PER_PAGE) % DLM_HASH_PAGES] + (i % DLM_BUCKETS_PER_PAGE); 179 } 180 181 static inline struct hlist_head *dlm_master_hash(struct dlm_ctxt *dlm, 182 unsigned i) 183 { 184 return dlm->master_hash[(i / DLM_BUCKETS_PER_PAGE) % DLM_HASH_PAGES] + 185 (i % DLM_BUCKETS_PER_PAGE); 186 } 187 188 /* these keventd work queue items are for less-frequently 189 * called functions that cannot be directly called from the 190 * net message handlers for some reason, usually because 191 * they need to send net messages of their own. */ 192 void dlm_dispatch_work(struct work_struct *work); 193 194 struct dlm_lock_resource; 195 struct dlm_work_item; 196 197 typedef void (dlm_workfunc_t)(struct dlm_work_item *, void *); 198 199 struct dlm_request_all_locks_priv 200 { 201 u8 reco_master; 202 u8 dead_node; 203 }; 204 205 struct dlm_mig_lockres_priv 206 { 207 struct dlm_lock_resource *lockres; 208 u8 real_master; 209 u8 extra_ref; 210 }; 211 212 struct dlm_assert_master_priv 213 { 214 struct dlm_lock_resource *lockres; 215 u8 request_from; 216 u32 flags; 217 unsigned ignore_higher:1; 218 }; 219 220 struct dlm_deref_lockres_priv 221 { 222 struct dlm_lock_resource *deref_res; 223 u8 deref_node; 224 }; 225 226 struct dlm_work_item 227 { 228 struct list_head list; 229 dlm_workfunc_t *func; 230 struct dlm_ctxt *dlm; 231 void *data; 232 union { 233 struct dlm_request_all_locks_priv ral; 234 struct dlm_mig_lockres_priv ml; 235 struct dlm_assert_master_priv am; 236 struct dlm_deref_lockres_priv dl; 237 } u; 238 }; 239 240 static inline void dlm_init_work_item(struct dlm_ctxt *dlm, 241 struct dlm_work_item *i, 242 dlm_workfunc_t *f, void *data) 243 { 244 memset(i, 0, sizeof(*i)); 245 i->func = f; 246 INIT_LIST_HEAD(&i->list); 247 i->data = data; 248 i->dlm = dlm; /* must have already done a dlm_grab on this! */ 249 } 250 251 252 253 static inline void __dlm_set_joining_node(struct dlm_ctxt *dlm, 254 u8 node) 255 { 256 assert_spin_locked(&dlm->spinlock); 257 258 dlm->joining_node = node; 259 wake_up(&dlm->dlm_join_events); 260 } 261 262 #define DLM_LOCK_RES_UNINITED 0x00000001 263 #define DLM_LOCK_RES_RECOVERING 0x00000002 264 #define DLM_LOCK_RES_READY 0x00000004 265 #define DLM_LOCK_RES_DIRTY 0x00000008 266 #define DLM_LOCK_RES_IN_PROGRESS 0x00000010 267 #define DLM_LOCK_RES_MIGRATING 0x00000020 268 #define DLM_LOCK_RES_DROPPING_REF 0x00000040 269 #define DLM_LOCK_RES_BLOCK_DIRTY 0x00001000 270 #define DLM_LOCK_RES_SETREF_INPROG 0x00002000 271 #define DLM_LOCK_RES_RECOVERY_WAITING 0x00004000 272 273 /* max milliseconds to wait to sync up a network failure with a node death */ 274 #define DLM_NODE_DEATH_WAIT_MAX (5 * 1000) 275 276 #define DLM_PURGE_INTERVAL_MS (8 * 1000) 277 278 struct dlm_lock_resource 279 { 280 /* WARNING: Please see the comment in dlm_init_lockres before 281 * adding fields here. */ 282 struct hlist_node hash_node; 283 struct qstr lockname; 284 struct kref refs; 285 286 /* 287 * Please keep granted, converting, and blocked in this order, 288 * as some funcs want to iterate over all lists. 289 * 290 * All four lists are protected by the hash's reference. 291 */ 292 struct list_head granted; 293 struct list_head converting; 294 struct list_head blocked; 295 struct list_head purge; 296 297 /* 298 * These two lists require you to hold an additional reference 299 * while they are on the list. 300 */ 301 struct list_head dirty; 302 struct list_head recovering; // dlm_recovery_ctxt.resources list 303 304 /* Added during init and removed during release */ 305 struct list_head tracking; /* dlm->tracking_list */ 306 307 /* unused lock resources have their last_used stamped and are 308 * put on a list for the dlm thread to run. */ 309 unsigned long last_used; 310 311 struct dlm_ctxt *dlm; 312 313 unsigned migration_pending:1; 314 atomic_t asts_reserved; 315 spinlock_t spinlock; 316 wait_queue_head_t wq; 317 u8 owner; //node which owns the lock resource, or unknown 318 u16 state; 319 char lvb[DLM_LVB_LEN]; 320 unsigned int inflight_locks; 321 unsigned int inflight_assert_workers; 322 unsigned long refmap[BITS_TO_LONGS(O2NM_MAX_NODES)]; 323 }; 324 325 struct dlm_migratable_lock 326 { 327 __be64 cookie; 328 329 /* these 3 are just padding for the in-memory structure, but 330 * list and flags are actually used when sent over the wire */ 331 __be16 pad1; 332 u8 list; // 0=granted, 1=converting, 2=blocked 333 u8 flags; 334 335 s8 type; 336 s8 convert_type; 337 s8 highest_blocked; 338 u8 node; 339 }; // 16 bytes 340 341 struct dlm_lock 342 { 343 struct dlm_migratable_lock ml; 344 345 struct list_head list; 346 struct list_head ast_list; 347 struct list_head bast_list; 348 struct dlm_lock_resource *lockres; 349 spinlock_t spinlock; 350 struct kref lock_refs; 351 352 // ast and bast must be callable while holding a spinlock! 353 dlm_astlockfunc_t *ast; 354 dlm_bastlockfunc_t *bast; 355 void *astdata; 356 struct dlm_lockstatus *lksb; 357 unsigned ast_pending:1, 358 bast_pending:1, 359 convert_pending:1, 360 lock_pending:1, 361 cancel_pending:1, 362 unlock_pending:1, 363 lksb_kernel_allocated:1; 364 }; 365 366 enum dlm_lockres_list { 367 DLM_GRANTED_LIST = 0, 368 DLM_CONVERTING_LIST = 1, 369 DLM_BLOCKED_LIST = 2, 370 }; 371 372 static inline int dlm_lvb_is_empty(char *lvb) 373 { 374 int i; 375 for (i=0; i<DLM_LVB_LEN; i++) 376 if (lvb[i]) 377 return 0; 378 return 1; 379 } 380 381 static inline char *dlm_list_in_text(enum dlm_lockres_list idx) 382 { 383 if (idx == DLM_GRANTED_LIST) 384 return "granted"; 385 else if (idx == DLM_CONVERTING_LIST) 386 return "converting"; 387 else if (idx == DLM_BLOCKED_LIST) 388 return "blocked"; 389 else 390 return "unknown"; 391 } 392 393 static inline struct list_head * 394 dlm_list_idx_to_ptr(struct dlm_lock_resource *res, enum dlm_lockres_list idx) 395 { 396 struct list_head *ret = NULL; 397 if (idx == DLM_GRANTED_LIST) 398 ret = &res->granted; 399 else if (idx == DLM_CONVERTING_LIST) 400 ret = &res->converting; 401 else if (idx == DLM_BLOCKED_LIST) 402 ret = &res->blocked; 403 else 404 BUG(); 405 return ret; 406 } 407 408 409 410 411 struct dlm_node_iter 412 { 413 unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; 414 int curnode; 415 }; 416 417 418 enum { 419 DLM_MASTER_REQUEST_MSG = 500, 420 DLM_UNUSED_MSG1 = 501, 421 DLM_ASSERT_MASTER_MSG = 502, 422 DLM_CREATE_LOCK_MSG = 503, 423 DLM_CONVERT_LOCK_MSG = 504, 424 DLM_PROXY_AST_MSG = 505, 425 DLM_UNLOCK_LOCK_MSG = 506, 426 DLM_DEREF_LOCKRES_MSG = 507, 427 DLM_MIGRATE_REQUEST_MSG = 508, 428 DLM_MIG_LOCKRES_MSG = 509, 429 DLM_QUERY_JOIN_MSG = 510, 430 DLM_ASSERT_JOINED_MSG = 511, 431 DLM_CANCEL_JOIN_MSG = 512, 432 DLM_EXIT_DOMAIN_MSG = 513, 433 DLM_MASTER_REQUERY_MSG = 514, 434 DLM_LOCK_REQUEST_MSG = 515, 435 DLM_RECO_DATA_DONE_MSG = 516, 436 DLM_BEGIN_RECO_MSG = 517, 437 DLM_FINALIZE_RECO_MSG = 518, 438 DLM_QUERY_REGION = 519, 439 DLM_QUERY_NODEINFO = 520, 440 DLM_BEGIN_EXIT_DOMAIN_MSG = 521, 441 DLM_DEREF_LOCKRES_DONE = 522, 442 }; 443 444 struct dlm_reco_node_data 445 { 446 int state; 447 u8 node_num; 448 struct list_head list; 449 }; 450 451 enum { 452 DLM_RECO_NODE_DATA_DEAD = -1, 453 DLM_RECO_NODE_DATA_INIT = 0, 454 DLM_RECO_NODE_DATA_REQUESTING = 1, 455 DLM_RECO_NODE_DATA_REQUESTED = 2, 456 DLM_RECO_NODE_DATA_RECEIVING = 3, 457 DLM_RECO_NODE_DATA_DONE = 4, 458 DLM_RECO_NODE_DATA_FINALIZE_SENT = 5, 459 }; 460 461 462 enum { 463 DLM_MASTER_RESP_NO = 0, 464 DLM_MASTER_RESP_YES = 1, 465 DLM_MASTER_RESP_MAYBE = 2, 466 DLM_MASTER_RESP_ERROR = 3, 467 }; 468 469 470 struct dlm_master_request 471 { 472 u8 node_idx; 473 u8 namelen; 474 __be16 pad1; 475 __be32 flags; 476 477 u8 name[O2NM_MAX_NAME_LEN]; 478 }; 479 480 #define DLM_ASSERT_RESPONSE_REASSERT 0x00000001 481 #define DLM_ASSERT_RESPONSE_MASTERY_REF 0x00000002 482 483 #define DLM_ASSERT_MASTER_MLE_CLEANUP 0x00000001 484 #define DLM_ASSERT_MASTER_REQUERY 0x00000002 485 #define DLM_ASSERT_MASTER_FINISH_MIGRATION 0x00000004 486 struct dlm_assert_master 487 { 488 u8 node_idx; 489 u8 namelen; 490 __be16 pad1; 491 __be32 flags; 492 493 u8 name[O2NM_MAX_NAME_LEN]; 494 }; 495 496 #define DLM_MIGRATE_RESPONSE_MASTERY_REF 0x00000001 497 498 struct dlm_migrate_request 499 { 500 u8 master; 501 u8 new_master; 502 u8 namelen; 503 u8 pad1; 504 __be32 pad2; 505 u8 name[O2NM_MAX_NAME_LEN]; 506 }; 507 508 struct dlm_master_requery 509 { 510 u8 pad1; 511 u8 pad2; 512 u8 node_idx; 513 u8 namelen; 514 __be32 pad3; 515 u8 name[O2NM_MAX_NAME_LEN]; 516 }; 517 518 #define DLM_MRES_RECOVERY 0x01 519 #define DLM_MRES_MIGRATION 0x02 520 #define DLM_MRES_ALL_DONE 0x04 521 522 /* 523 * We would like to get one whole lockres into a single network 524 * message whenever possible. Generally speaking, there will be 525 * at most one dlm_lock on a lockres for each node in the cluster, 526 * plus (infrequently) any additional locks coming in from userdlm. 527 * 528 * struct _dlm_lockres_page 529 * { 530 * dlm_migratable_lockres mres; 531 * dlm_migratable_lock ml[DLM_MAX_MIGRATABLE_LOCKS]; 532 * u8 pad[DLM_MIG_LOCKRES_RESERVED]; 533 * }; 534 * 535 * from ../cluster/tcp.h 536 * O2NET_MAX_PAYLOAD_BYTES (4096 - sizeof(net_msg)) 537 * (roughly 4080 bytes) 538 * and sizeof(dlm_migratable_lockres) = 112 bytes 539 * and sizeof(dlm_migratable_lock) = 16 bytes 540 * 541 * Choosing DLM_MAX_MIGRATABLE_LOCKS=240 and 542 * DLM_MIG_LOCKRES_RESERVED=128 means we have this: 543 * 544 * (DLM_MAX_MIGRATABLE_LOCKS * sizeof(dlm_migratable_lock)) + 545 * sizeof(dlm_migratable_lockres) + DLM_MIG_LOCKRES_RESERVED = 546 * NET_MAX_PAYLOAD_BYTES 547 * (240 * 16) + 112 + 128 = 4080 548 * 549 * So a lockres would need more than 240 locks before it would 550 * use more than one network packet to recover. Not too bad. 551 */ 552 #define DLM_MAX_MIGRATABLE_LOCKS 240 553 554 struct dlm_migratable_lockres 555 { 556 u8 master; 557 u8 lockname_len; 558 u8 num_locks; // locks sent in this structure 559 u8 flags; 560 __be32 total_locks; // locks to be sent for this migration cookie 561 __be64 mig_cookie; // cookie for this lockres migration 562 // or zero if not needed 563 // 16 bytes 564 u8 lockname[DLM_LOCKID_NAME_MAX]; 565 // 48 bytes 566 u8 lvb[DLM_LVB_LEN]; 567 // 112 bytes 568 struct dlm_migratable_lock ml[0]; // 16 bytes each, begins at byte 112 569 }; 570 #define DLM_MIG_LOCKRES_MAX_LEN \ 571 (sizeof(struct dlm_migratable_lockres) + \ 572 (sizeof(struct dlm_migratable_lock) * \ 573 DLM_MAX_MIGRATABLE_LOCKS) ) 574 575 /* from above, 128 bytes 576 * for some undetermined future use */ 577 #define DLM_MIG_LOCKRES_RESERVED (O2NET_MAX_PAYLOAD_BYTES - \ 578 DLM_MIG_LOCKRES_MAX_LEN) 579 580 struct dlm_create_lock 581 { 582 __be64 cookie; 583 584 __be32 flags; 585 u8 pad1; 586 u8 node_idx; 587 s8 requested_type; 588 u8 namelen; 589 590 u8 name[O2NM_MAX_NAME_LEN]; 591 }; 592 593 struct dlm_convert_lock 594 { 595 __be64 cookie; 596 597 __be32 flags; 598 u8 pad1; 599 u8 node_idx; 600 s8 requested_type; 601 u8 namelen; 602 603 u8 name[O2NM_MAX_NAME_LEN]; 604 605 s8 lvb[0]; 606 }; 607 #define DLM_CONVERT_LOCK_MAX_LEN (sizeof(struct dlm_convert_lock)+DLM_LVB_LEN) 608 609 struct dlm_unlock_lock 610 { 611 __be64 cookie; 612 613 __be32 flags; 614 __be16 pad1; 615 u8 node_idx; 616 u8 namelen; 617 618 u8 name[O2NM_MAX_NAME_LEN]; 619 620 s8 lvb[0]; 621 }; 622 #define DLM_UNLOCK_LOCK_MAX_LEN (sizeof(struct dlm_unlock_lock)+DLM_LVB_LEN) 623 624 struct dlm_proxy_ast 625 { 626 __be64 cookie; 627 628 __be32 flags; 629 u8 node_idx; 630 u8 type; 631 u8 blocked_type; 632 u8 namelen; 633 634 u8 name[O2NM_MAX_NAME_LEN]; 635 636 s8 lvb[0]; 637 }; 638 #define DLM_PROXY_AST_MAX_LEN (sizeof(struct dlm_proxy_ast)+DLM_LVB_LEN) 639 640 #define DLM_MOD_KEY (0x666c6172) 641 enum dlm_query_join_response_code { 642 JOIN_DISALLOW = 0, 643 JOIN_OK = 1, 644 JOIN_OK_NO_MAP = 2, 645 JOIN_PROTOCOL_MISMATCH = 3, 646 }; 647 648 struct dlm_query_join_packet { 649 u8 code; /* Response code. dlm_minor and fs_minor 650 are only valid if this is JOIN_OK */ 651 u8 dlm_minor; /* The minor version of the protocol the 652 dlm is speaking. */ 653 u8 fs_minor; /* The minor version of the protocol the 654 filesystem is speaking. */ 655 u8 reserved; 656 }; 657 658 union dlm_query_join_response { 659 __be32 intval; 660 struct dlm_query_join_packet packet; 661 }; 662 663 struct dlm_lock_request 664 { 665 u8 node_idx; 666 u8 dead_node; 667 __be16 pad1; 668 __be32 pad2; 669 }; 670 671 struct dlm_reco_data_done 672 { 673 u8 node_idx; 674 u8 dead_node; 675 __be16 pad1; 676 __be32 pad2; 677 678 /* unused for now */ 679 /* eventually we can use this to attempt 680 * lvb recovery based on each node's info */ 681 u8 reco_lvb[DLM_LVB_LEN]; 682 }; 683 684 struct dlm_begin_reco 685 { 686 u8 node_idx; 687 u8 dead_node; 688 __be16 pad1; 689 __be32 pad2; 690 }; 691 692 693 #define BITS_PER_BYTE 8 694 #define BITS_TO_BYTES(bits) (((bits)+BITS_PER_BYTE-1)/BITS_PER_BYTE) 695 696 struct dlm_query_join_request 697 { 698 u8 node_idx; 699 u8 pad1[2]; 700 u8 name_len; 701 struct dlm_protocol_version dlm_proto; 702 struct dlm_protocol_version fs_proto; 703 u8 domain[O2NM_MAX_NAME_LEN]; 704 u8 node_map[BITS_TO_BYTES(O2NM_MAX_NODES)]; 705 }; 706 707 struct dlm_assert_joined 708 { 709 u8 node_idx; 710 u8 pad1[2]; 711 u8 name_len; 712 u8 domain[O2NM_MAX_NAME_LEN]; 713 }; 714 715 struct dlm_cancel_join 716 { 717 u8 node_idx; 718 u8 pad1[2]; 719 u8 name_len; 720 u8 domain[O2NM_MAX_NAME_LEN]; 721 }; 722 723 struct dlm_query_region { 724 u8 qr_node; 725 u8 qr_numregions; 726 u8 qr_namelen; 727 u8 pad1; 728 u8 qr_domain[O2NM_MAX_NAME_LEN]; 729 u8 qr_regions[O2HB_MAX_REGION_NAME_LEN * O2NM_MAX_REGIONS]; 730 }; 731 732 struct dlm_node_info { 733 u8 ni_nodenum; 734 u8 pad1; 735 __be16 ni_ipv4_port; 736 __be32 ni_ipv4_address; 737 }; 738 739 struct dlm_query_nodeinfo { 740 u8 qn_nodenum; 741 u8 qn_numnodes; 742 u8 qn_namelen; 743 u8 pad1; 744 u8 qn_domain[O2NM_MAX_NAME_LEN]; 745 struct dlm_node_info qn_nodes[O2NM_MAX_NODES]; 746 }; 747 748 struct dlm_exit_domain 749 { 750 u8 node_idx; 751 u8 pad1[3]; 752 }; 753 754 struct dlm_finalize_reco 755 { 756 u8 node_idx; 757 u8 dead_node; 758 u8 flags; 759 u8 pad1; 760 __be32 pad2; 761 }; 762 763 struct dlm_deref_lockres 764 { 765 u32 pad1; 766 u16 pad2; 767 u8 node_idx; 768 u8 namelen; 769 770 u8 name[O2NM_MAX_NAME_LEN]; 771 }; 772 773 enum { 774 DLM_DEREF_RESPONSE_DONE = 0, 775 DLM_DEREF_RESPONSE_INPROG = 1, 776 }; 777 778 struct dlm_deref_lockres_done { 779 u32 pad1; 780 u16 pad2; 781 u8 node_idx; 782 u8 namelen; 783 784 u8 name[O2NM_MAX_NAME_LEN]; 785 }; 786 787 static inline enum dlm_status 788 __dlm_lockres_state_to_status(struct dlm_lock_resource *res) 789 { 790 enum dlm_status status = DLM_NORMAL; 791 792 assert_spin_locked(&res->spinlock); 793 794 if (res->state & (DLM_LOCK_RES_RECOVERING| 795 DLM_LOCK_RES_RECOVERY_WAITING)) 796 status = DLM_RECOVERING; 797 else if (res->state & DLM_LOCK_RES_MIGRATING) 798 status = DLM_MIGRATING; 799 else if (res->state & DLM_LOCK_RES_IN_PROGRESS) 800 status = DLM_FORWARD; 801 802 return status; 803 } 804 805 static inline u8 dlm_get_lock_cookie_node(u64 cookie) 806 { 807 u8 ret; 808 cookie >>= 56; 809 ret = (u8)(cookie & 0xffULL); 810 return ret; 811 } 812 813 static inline unsigned long long dlm_get_lock_cookie_seq(u64 cookie) 814 { 815 unsigned long long ret; 816 ret = ((unsigned long long)cookie) & 0x00ffffffffffffffULL; 817 return ret; 818 } 819 820 struct dlm_lock * dlm_new_lock(int type, u8 node, u64 cookie, 821 struct dlm_lockstatus *lksb); 822 void dlm_lock_get(struct dlm_lock *lock); 823 void dlm_lock_put(struct dlm_lock *lock); 824 825 void dlm_lock_attach_lockres(struct dlm_lock *lock, 826 struct dlm_lock_resource *res); 827 828 int dlm_create_lock_handler(struct o2net_msg *msg, u32 len, void *data, 829 void **ret_data); 830 int dlm_convert_lock_handler(struct o2net_msg *msg, u32 len, void *data, 831 void **ret_data); 832 int dlm_proxy_ast_handler(struct o2net_msg *msg, u32 len, void *data, 833 void **ret_data); 834 835 void dlm_revert_pending_convert(struct dlm_lock_resource *res, 836 struct dlm_lock *lock); 837 void dlm_revert_pending_lock(struct dlm_lock_resource *res, 838 struct dlm_lock *lock); 839 840 int dlm_unlock_lock_handler(struct o2net_msg *msg, u32 len, void *data, 841 void **ret_data); 842 void dlm_commit_pending_cancel(struct dlm_lock_resource *res, 843 struct dlm_lock *lock); 844 void dlm_commit_pending_unlock(struct dlm_lock_resource *res, 845 struct dlm_lock *lock); 846 847 int dlm_launch_thread(struct dlm_ctxt *dlm); 848 void dlm_complete_thread(struct dlm_ctxt *dlm); 849 int dlm_launch_recovery_thread(struct dlm_ctxt *dlm); 850 void dlm_complete_recovery_thread(struct dlm_ctxt *dlm); 851 void dlm_wait_for_recovery(struct dlm_ctxt *dlm); 852 void dlm_kick_recovery_thread(struct dlm_ctxt *dlm); 853 int dlm_is_node_dead(struct dlm_ctxt *dlm, u8 node); 854 void dlm_wait_for_node_death(struct dlm_ctxt *dlm, u8 node, int timeout); 855 void dlm_wait_for_node_recovery(struct dlm_ctxt *dlm, u8 node, int timeout); 856 857 void dlm_put(struct dlm_ctxt *dlm); 858 struct dlm_ctxt *dlm_grab(struct dlm_ctxt *dlm); 859 int dlm_domain_fully_joined(struct dlm_ctxt *dlm); 860 861 void __dlm_lockres_calc_usage(struct dlm_ctxt *dlm, 862 struct dlm_lock_resource *res); 863 void dlm_lockres_calc_usage(struct dlm_ctxt *dlm, 864 struct dlm_lock_resource *res); 865 static inline void dlm_lockres_get(struct dlm_lock_resource *res) 866 { 867 /* This is called on every lookup, so it might be worth 868 * inlining. */ 869 kref_get(&res->refs); 870 } 871 void dlm_lockres_put(struct dlm_lock_resource *res); 872 void __dlm_unhash_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res); 873 void __dlm_insert_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res); 874 struct dlm_lock_resource * __dlm_lookup_lockres_full(struct dlm_ctxt *dlm, 875 const char *name, 876 unsigned int len, 877 unsigned int hash); 878 struct dlm_lock_resource * __dlm_lookup_lockres(struct dlm_ctxt *dlm, 879 const char *name, 880 unsigned int len, 881 unsigned int hash); 882 struct dlm_lock_resource * dlm_lookup_lockres(struct dlm_ctxt *dlm, 883 const char *name, 884 unsigned int len); 885 886 int dlm_is_host_down(int errno); 887 888 struct dlm_lock_resource * dlm_get_lock_resource(struct dlm_ctxt *dlm, 889 const char *lockid, 890 int namelen, 891 int flags); 892 struct dlm_lock_resource *dlm_new_lockres(struct dlm_ctxt *dlm, 893 const char *name, 894 unsigned int namelen); 895 896 void dlm_lockres_set_refmap_bit(struct dlm_ctxt *dlm, 897 struct dlm_lock_resource *res, int bit); 898 void dlm_lockres_clear_refmap_bit(struct dlm_ctxt *dlm, 899 struct dlm_lock_resource *res, int bit); 900 901 void dlm_lockres_drop_inflight_ref(struct dlm_ctxt *dlm, 902 struct dlm_lock_resource *res); 903 void dlm_lockres_grab_inflight_ref(struct dlm_ctxt *dlm, 904 struct dlm_lock_resource *res); 905 906 void __dlm_lockres_grab_inflight_worker(struct dlm_ctxt *dlm, 907 struct dlm_lock_resource *res); 908 909 void dlm_queue_ast(struct dlm_ctxt *dlm, struct dlm_lock *lock); 910 void dlm_queue_bast(struct dlm_ctxt *dlm, struct dlm_lock *lock); 911 void __dlm_queue_ast(struct dlm_ctxt *dlm, struct dlm_lock *lock); 912 void __dlm_queue_bast(struct dlm_ctxt *dlm, struct dlm_lock *lock); 913 void dlm_do_local_ast(struct dlm_ctxt *dlm, 914 struct dlm_lock_resource *res, 915 struct dlm_lock *lock); 916 int dlm_do_remote_ast(struct dlm_ctxt *dlm, 917 struct dlm_lock_resource *res, 918 struct dlm_lock *lock); 919 void dlm_do_local_bast(struct dlm_ctxt *dlm, 920 struct dlm_lock_resource *res, 921 struct dlm_lock *lock, 922 int blocked_type); 923 int dlm_send_proxy_ast_msg(struct dlm_ctxt *dlm, 924 struct dlm_lock_resource *res, 925 struct dlm_lock *lock, 926 int msg_type, 927 int blocked_type, int flags); 928 static inline int dlm_send_proxy_bast(struct dlm_ctxt *dlm, 929 struct dlm_lock_resource *res, 930 struct dlm_lock *lock, 931 int blocked_type) 932 { 933 return dlm_send_proxy_ast_msg(dlm, res, lock, DLM_BAST, 934 blocked_type, 0); 935 } 936 937 static inline int dlm_send_proxy_ast(struct dlm_ctxt *dlm, 938 struct dlm_lock_resource *res, 939 struct dlm_lock *lock, 940 int flags) 941 { 942 return dlm_send_proxy_ast_msg(dlm, res, lock, DLM_AST, 943 0, flags); 944 } 945 946 void dlm_print_one_lock_resource(struct dlm_lock_resource *res); 947 void __dlm_print_one_lock_resource(struct dlm_lock_resource *res); 948 949 void dlm_kick_thread(struct dlm_ctxt *dlm, struct dlm_lock_resource *res); 950 void __dlm_dirty_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res); 951 952 953 void dlm_hb_node_down_cb(struct o2nm_node *node, int idx, void *data); 954 void dlm_hb_node_up_cb(struct o2nm_node *node, int idx, void *data); 955 956 int dlm_empty_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res); 957 int dlm_finish_migration(struct dlm_ctxt *dlm, 958 struct dlm_lock_resource *res, 959 u8 old_master); 960 void dlm_lockres_release_ast(struct dlm_ctxt *dlm, 961 struct dlm_lock_resource *res); 962 void __dlm_lockres_reserve_ast(struct dlm_lock_resource *res); 963 964 int dlm_master_request_handler(struct o2net_msg *msg, u32 len, void *data, 965 void **ret_data); 966 int dlm_assert_master_handler(struct o2net_msg *msg, u32 len, void *data, 967 void **ret_data); 968 void dlm_assert_master_post_handler(int status, void *data, void *ret_data); 969 int dlm_deref_lockres_handler(struct o2net_msg *msg, u32 len, void *data, 970 void **ret_data); 971 int dlm_deref_lockres_done_handler(struct o2net_msg *msg, u32 len, void *data, 972 void **ret_data); 973 int dlm_migrate_request_handler(struct o2net_msg *msg, u32 len, void *data, 974 void **ret_data); 975 int dlm_mig_lockres_handler(struct o2net_msg *msg, u32 len, void *data, 976 void **ret_data); 977 int dlm_master_requery_handler(struct o2net_msg *msg, u32 len, void *data, 978 void **ret_data); 979 int dlm_request_all_locks_handler(struct o2net_msg *msg, u32 len, void *data, 980 void **ret_data); 981 int dlm_reco_data_done_handler(struct o2net_msg *msg, u32 len, void *data, 982 void **ret_data); 983 int dlm_begin_reco_handler(struct o2net_msg *msg, u32 len, void *data, 984 void **ret_data); 985 int dlm_finalize_reco_handler(struct o2net_msg *msg, u32 len, void *data, 986 void **ret_data); 987 int dlm_do_master_requery(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, 988 u8 nodenum, u8 *real_master); 989 990 void __dlm_do_purge_lockres(struct dlm_ctxt *dlm, 991 struct dlm_lock_resource *res); 992 993 int dlm_dispatch_assert_master(struct dlm_ctxt *dlm, 994 struct dlm_lock_resource *res, 995 int ignore_higher, 996 u8 request_from, 997 u32 flags); 998 999 1000 int dlm_send_one_lockres(struct dlm_ctxt *dlm, 1001 struct dlm_lock_resource *res, 1002 struct dlm_migratable_lockres *mres, 1003 u8 send_to, 1004 u8 flags); 1005 void dlm_move_lockres_to_recovery_list(struct dlm_ctxt *dlm, 1006 struct dlm_lock_resource *res); 1007 1008 /* will exit holding res->spinlock, but may drop in function */ 1009 void __dlm_wait_on_lockres_flags(struct dlm_lock_resource *res, int flags); 1010 1011 /* will exit holding res->spinlock, but may drop in function */ 1012 static inline void __dlm_wait_on_lockres(struct dlm_lock_resource *res) 1013 { 1014 __dlm_wait_on_lockres_flags(res, (DLM_LOCK_RES_IN_PROGRESS| 1015 DLM_LOCK_RES_RECOVERING| 1016 DLM_LOCK_RES_RECOVERY_WAITING| 1017 DLM_LOCK_RES_MIGRATING)); 1018 } 1019 1020 void __dlm_unlink_mle(struct dlm_ctxt *dlm, struct dlm_master_list_entry *mle); 1021 void __dlm_insert_mle(struct dlm_ctxt *dlm, struct dlm_master_list_entry *mle); 1022 1023 /* create/destroy slab caches */ 1024 int dlm_init_master_caches(void); 1025 void dlm_destroy_master_caches(void); 1026 1027 int dlm_init_lock_cache(void); 1028 void dlm_destroy_lock_cache(void); 1029 1030 int dlm_init_mle_cache(void); 1031 void dlm_destroy_mle_cache(void); 1032 1033 void dlm_hb_event_notify_attached(struct dlm_ctxt *dlm, int idx, int node_up); 1034 int dlm_drop_lockres_ref(struct dlm_ctxt *dlm, 1035 struct dlm_lock_resource *res); 1036 void dlm_clean_master_list(struct dlm_ctxt *dlm, 1037 u8 dead_node); 1038 void dlm_force_free_mles(struct dlm_ctxt *dlm); 1039 int dlm_lock_basts_flushed(struct dlm_ctxt *dlm, struct dlm_lock *lock); 1040 int __dlm_lockres_has_locks(struct dlm_lock_resource *res); 1041 int __dlm_lockres_unused(struct dlm_lock_resource *res); 1042 1043 static inline const char * dlm_lock_mode_name(int mode) 1044 { 1045 switch (mode) { 1046 case LKM_EXMODE: 1047 return "EX"; 1048 case LKM_PRMODE: 1049 return "PR"; 1050 case LKM_NLMODE: 1051 return "NL"; 1052 } 1053 return "UNKNOWN"; 1054 } 1055 1056 1057 static inline int dlm_lock_compatible(int existing, int request) 1058 { 1059 /* NO_LOCK compatible with all */ 1060 if (request == LKM_NLMODE || 1061 existing == LKM_NLMODE) 1062 return 1; 1063 1064 /* EX incompatible with all non-NO_LOCK */ 1065 if (request == LKM_EXMODE) 1066 return 0; 1067 1068 /* request must be PR, which is compatible with PR */ 1069 if (existing == LKM_PRMODE) 1070 return 1; 1071 1072 return 0; 1073 } 1074 1075 static inline int dlm_lock_on_list(struct list_head *head, 1076 struct dlm_lock *lock) 1077 { 1078 struct dlm_lock *tmplock; 1079 1080 list_for_each_entry(tmplock, head, list) { 1081 if (tmplock == lock) 1082 return 1; 1083 } 1084 return 0; 1085 } 1086 1087 1088 static inline enum dlm_status dlm_err_to_dlm_status(int err) 1089 { 1090 enum dlm_status ret; 1091 if (err == -ENOMEM) 1092 ret = DLM_SYSERR; 1093 else if (err == -ETIMEDOUT || o2net_link_down(err, NULL)) 1094 ret = DLM_NOLOCKMGR; 1095 else if (err == -EINVAL) 1096 ret = DLM_BADPARAM; 1097 else if (err == -ENAMETOOLONG) 1098 ret = DLM_IVBUFLEN; 1099 else 1100 ret = DLM_BADARGS; 1101 return ret; 1102 } 1103 1104 1105 static inline void dlm_node_iter_init(unsigned long *map, 1106 struct dlm_node_iter *iter) 1107 { 1108 memcpy(iter->node_map, map, sizeof(iter->node_map)); 1109 iter->curnode = -1; 1110 } 1111 1112 static inline int dlm_node_iter_next(struct dlm_node_iter *iter) 1113 { 1114 int bit; 1115 bit = find_next_bit(iter->node_map, O2NM_MAX_NODES, iter->curnode+1); 1116 if (bit >= O2NM_MAX_NODES) { 1117 iter->curnode = O2NM_MAX_NODES; 1118 return -ENOENT; 1119 } 1120 iter->curnode = bit; 1121 return bit; 1122 } 1123 1124 static inline void dlm_set_lockres_owner(struct dlm_ctxt *dlm, 1125 struct dlm_lock_resource *res, 1126 u8 owner) 1127 { 1128 assert_spin_locked(&res->spinlock); 1129 1130 res->owner = owner; 1131 } 1132 1133 static inline void dlm_change_lockres_owner(struct dlm_ctxt *dlm, 1134 struct dlm_lock_resource *res, 1135 u8 owner) 1136 { 1137 assert_spin_locked(&res->spinlock); 1138 1139 if (owner != res->owner) 1140 dlm_set_lockres_owner(dlm, res, owner); 1141 } 1142 1143 #endif /* DLMCOMMON_H */ 1144