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 << 14) 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 52 enum dlm_mle_type { 53 DLM_MLE_BLOCK, 54 DLM_MLE_MASTER, 55 DLM_MLE_MIGRATION 56 }; 57 58 struct dlm_lock_name { 59 u8 len; 60 u8 name[DLM_LOCKID_NAME_MAX]; 61 }; 62 63 struct dlm_master_list_entry { 64 struct list_head list; 65 struct list_head hb_events; 66 struct dlm_ctxt *dlm; 67 spinlock_t spinlock; 68 wait_queue_head_t wq; 69 atomic_t woken; 70 struct kref mle_refs; 71 int inuse; 72 unsigned long maybe_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; 73 unsigned long vote_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; 74 unsigned long response_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; 75 unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; 76 u8 master; 77 u8 new_master; 78 enum dlm_mle_type type; 79 struct o2hb_callback_func mle_hb_up; 80 struct o2hb_callback_func mle_hb_down; 81 union { 82 struct dlm_lock_resource *res; 83 struct dlm_lock_name name; 84 } u; 85 }; 86 87 enum dlm_ast_type { 88 DLM_AST = 0, 89 DLM_BAST, 90 DLM_ASTUNLOCK 91 }; 92 93 94 #define LKM_VALID_FLAGS (LKM_VALBLK | LKM_CONVERT | LKM_UNLOCK | \ 95 LKM_CANCEL | LKM_INVVALBLK | LKM_FORCE | \ 96 LKM_RECOVERY | LKM_LOCAL | LKM_NOQUEUE) 97 98 #define DLM_RECOVERY_LOCK_NAME "$RECOVERY" 99 #define DLM_RECOVERY_LOCK_NAME_LEN 9 100 101 static inline int dlm_is_recovery_lock(const char *lock_name, int name_len) 102 { 103 if (name_len == DLM_RECOVERY_LOCK_NAME_LEN && 104 memcmp(lock_name, DLM_RECOVERY_LOCK_NAME, name_len)==0) 105 return 1; 106 return 0; 107 } 108 109 #define DLM_RECO_STATE_ACTIVE 0x0001 110 #define DLM_RECO_STATE_FINALIZE 0x0002 111 112 struct dlm_recovery_ctxt 113 { 114 struct list_head resources; 115 struct list_head received; 116 struct list_head node_data; 117 u8 new_master; 118 u8 dead_node; 119 u16 state; 120 unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; 121 wait_queue_head_t event; 122 }; 123 124 enum dlm_ctxt_state { 125 DLM_CTXT_NEW = 0, 126 DLM_CTXT_JOINED, 127 DLM_CTXT_IN_SHUTDOWN, 128 DLM_CTXT_LEAVING, 129 }; 130 131 struct dlm_ctxt 132 { 133 struct list_head list; 134 struct hlist_head **lockres_hash; 135 struct list_head dirty_list; 136 struct list_head purge_list; 137 struct list_head pending_asts; 138 struct list_head pending_basts; 139 struct list_head tracking_list; 140 unsigned int purge_count; 141 spinlock_t spinlock; 142 spinlock_t ast_lock; 143 char *name; 144 u8 node_num; 145 u32 key; 146 u8 joining_node; 147 wait_queue_head_t dlm_join_events; 148 unsigned long live_nodes_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; 149 unsigned long domain_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; 150 unsigned long recovery_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; 151 struct dlm_recovery_ctxt reco; 152 spinlock_t master_lock; 153 struct list_head master_list; 154 struct list_head mle_hb_events; 155 156 /* these give a really vague idea of the system load */ 157 atomic_t local_resources; 158 atomic_t remote_resources; 159 atomic_t unknown_resources; 160 161 struct dlm_debug_ctxt *dlm_debug_ctxt; 162 struct dentry *dlm_debugfs_subroot; 163 164 /* NOTE: Next three are protected by dlm_domain_lock */ 165 struct kref dlm_refs; 166 enum dlm_ctxt_state dlm_state; 167 unsigned int num_joins; 168 169 struct o2hb_callback_func dlm_hb_up; 170 struct o2hb_callback_func dlm_hb_down; 171 struct task_struct *dlm_thread_task; 172 struct task_struct *dlm_reco_thread_task; 173 struct workqueue_struct *dlm_worker; 174 wait_queue_head_t dlm_thread_wq; 175 wait_queue_head_t dlm_reco_thread_wq; 176 wait_queue_head_t ast_wq; 177 wait_queue_head_t migration_wq; 178 179 struct work_struct dispatched_work; 180 struct list_head work_list; 181 spinlock_t work_lock; 182 struct list_head dlm_domain_handlers; 183 struct list_head dlm_eviction_callbacks; 184 185 /* The filesystem specifies this at domain registration. We 186 * cache it here to know what to tell other nodes. */ 187 struct dlm_protocol_version fs_locking_proto; 188 /* This is the inter-dlm communication version */ 189 struct dlm_protocol_version dlm_locking_proto; 190 }; 191 192 static inline struct hlist_head *dlm_lockres_hash(struct dlm_ctxt *dlm, unsigned i) 193 { 194 return dlm->lockres_hash[(i / DLM_BUCKETS_PER_PAGE) % DLM_HASH_PAGES] + (i % DLM_BUCKETS_PER_PAGE); 195 } 196 197 /* these keventd work queue items are for less-frequently 198 * called functions that cannot be directly called from the 199 * net message handlers for some reason, usually because 200 * they need to send net messages of their own. */ 201 void dlm_dispatch_work(struct work_struct *work); 202 203 struct dlm_lock_resource; 204 struct dlm_work_item; 205 206 typedef void (dlm_workfunc_t)(struct dlm_work_item *, void *); 207 208 struct dlm_request_all_locks_priv 209 { 210 u8 reco_master; 211 u8 dead_node; 212 }; 213 214 struct dlm_mig_lockres_priv 215 { 216 struct dlm_lock_resource *lockres; 217 u8 real_master; 218 u8 extra_ref; 219 }; 220 221 struct dlm_assert_master_priv 222 { 223 struct dlm_lock_resource *lockres; 224 u8 request_from; 225 u32 flags; 226 unsigned ignore_higher:1; 227 }; 228 229 struct dlm_deref_lockres_priv 230 { 231 struct dlm_lock_resource *deref_res; 232 u8 deref_node; 233 }; 234 235 struct dlm_work_item 236 { 237 struct list_head list; 238 dlm_workfunc_t *func; 239 struct dlm_ctxt *dlm; 240 void *data; 241 union { 242 struct dlm_request_all_locks_priv ral; 243 struct dlm_mig_lockres_priv ml; 244 struct dlm_assert_master_priv am; 245 struct dlm_deref_lockres_priv dl; 246 } u; 247 }; 248 249 static inline void dlm_init_work_item(struct dlm_ctxt *dlm, 250 struct dlm_work_item *i, 251 dlm_workfunc_t *f, void *data) 252 { 253 memset(i, 0, sizeof(*i)); 254 i->func = f; 255 INIT_LIST_HEAD(&i->list); 256 i->data = data; 257 i->dlm = dlm; /* must have already done a dlm_grab on this! */ 258 } 259 260 261 262 static inline void __dlm_set_joining_node(struct dlm_ctxt *dlm, 263 u8 node) 264 { 265 assert_spin_locked(&dlm->spinlock); 266 267 dlm->joining_node = node; 268 wake_up(&dlm->dlm_join_events); 269 } 270 271 #define DLM_LOCK_RES_UNINITED 0x00000001 272 #define DLM_LOCK_RES_RECOVERING 0x00000002 273 #define DLM_LOCK_RES_READY 0x00000004 274 #define DLM_LOCK_RES_DIRTY 0x00000008 275 #define DLM_LOCK_RES_IN_PROGRESS 0x00000010 276 #define DLM_LOCK_RES_MIGRATING 0x00000020 277 #define DLM_LOCK_RES_DROPPING_REF 0x00000040 278 #define DLM_LOCK_RES_BLOCK_DIRTY 0x00001000 279 #define DLM_LOCK_RES_SETREF_INPROG 0x00002000 280 281 /* max milliseconds to wait to sync up a network failure with a node death */ 282 #define DLM_NODE_DEATH_WAIT_MAX (5 * 1000) 283 284 #define DLM_PURGE_INTERVAL_MS (8 * 1000) 285 286 struct dlm_lock_resource 287 { 288 /* WARNING: Please see the comment in dlm_init_lockres before 289 * adding fields here. */ 290 struct hlist_node hash_node; 291 struct qstr lockname; 292 struct kref refs; 293 294 /* 295 * Please keep granted, converting, and blocked in this order, 296 * as some funcs want to iterate over all lists. 297 * 298 * All four lists are protected by the hash's reference. 299 */ 300 struct list_head granted; 301 struct list_head converting; 302 struct list_head blocked; 303 struct list_head purge; 304 305 /* 306 * These two lists require you to hold an additional reference 307 * while they are on the list. 308 */ 309 struct list_head dirty; 310 struct list_head recovering; // dlm_recovery_ctxt.resources list 311 312 /* Added during init and removed during release */ 313 struct list_head tracking; /* dlm->tracking_list */ 314 315 /* unused lock resources have their last_used stamped and are 316 * put on a list for the dlm thread to run. */ 317 unsigned long last_used; 318 319 unsigned migration_pending:1; 320 atomic_t asts_reserved; 321 spinlock_t spinlock; 322 wait_queue_head_t wq; 323 u8 owner; //node which owns the lock resource, or unknown 324 u16 state; 325 char lvb[DLM_LVB_LEN]; 326 unsigned int inflight_locks; 327 unsigned long refmap[BITS_TO_LONGS(O2NM_MAX_NODES)]; 328 }; 329 330 struct dlm_migratable_lock 331 { 332 __be64 cookie; 333 334 /* these 3 are just padding for the in-memory structure, but 335 * list and flags are actually used when sent over the wire */ 336 __be16 pad1; 337 u8 list; // 0=granted, 1=converting, 2=blocked 338 u8 flags; 339 340 s8 type; 341 s8 convert_type; 342 s8 highest_blocked; 343 u8 node; 344 }; // 16 bytes 345 346 struct dlm_lock 347 { 348 struct dlm_migratable_lock ml; 349 350 struct list_head list; 351 struct list_head ast_list; 352 struct list_head bast_list; 353 struct dlm_lock_resource *lockres; 354 spinlock_t spinlock; 355 struct kref lock_refs; 356 357 // ast and bast must be callable while holding a spinlock! 358 dlm_astlockfunc_t *ast; 359 dlm_bastlockfunc_t *bast; 360 void *astdata; 361 struct dlm_lockstatus *lksb; 362 unsigned ast_pending:1, 363 bast_pending:1, 364 convert_pending:1, 365 lock_pending:1, 366 cancel_pending:1, 367 unlock_pending:1, 368 lksb_kernel_allocated:1; 369 }; 370 371 372 #define DLM_LKSB_UNUSED1 0x01 373 #define DLM_LKSB_PUT_LVB 0x02 374 #define DLM_LKSB_GET_LVB 0x04 375 #define DLM_LKSB_UNUSED2 0x08 376 #define DLM_LKSB_UNUSED3 0x10 377 #define DLM_LKSB_UNUSED4 0x20 378 #define DLM_LKSB_UNUSED5 0x40 379 #define DLM_LKSB_UNUSED6 0x80 380 381 382 enum dlm_lockres_list { 383 DLM_GRANTED_LIST = 0, 384 DLM_CONVERTING_LIST, 385 DLM_BLOCKED_LIST 386 }; 387 388 static inline int dlm_lvb_is_empty(char *lvb) 389 { 390 int i; 391 for (i=0; i<DLM_LVB_LEN; i++) 392 if (lvb[i]) 393 return 0; 394 return 1; 395 } 396 397 static inline struct list_head * 398 dlm_list_idx_to_ptr(struct dlm_lock_resource *res, enum dlm_lockres_list idx) 399 { 400 struct list_head *ret = NULL; 401 if (idx == DLM_GRANTED_LIST) 402 ret = &res->granted; 403 else if (idx == DLM_CONVERTING_LIST) 404 ret = &res->converting; 405 else if (idx == DLM_BLOCKED_LIST) 406 ret = &res->blocked; 407 else 408 BUG(); 409 return ret; 410 } 411 412 413 414 415 struct dlm_node_iter 416 { 417 unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; 418 int curnode; 419 }; 420 421 422 enum { 423 DLM_MASTER_REQUEST_MSG = 500, 424 DLM_UNUSED_MSG1, /* 501 */ 425 DLM_ASSERT_MASTER_MSG, /* 502 */ 426 DLM_CREATE_LOCK_MSG, /* 503 */ 427 DLM_CONVERT_LOCK_MSG, /* 504 */ 428 DLM_PROXY_AST_MSG, /* 505 */ 429 DLM_UNLOCK_LOCK_MSG, /* 506 */ 430 DLM_DEREF_LOCKRES_MSG, /* 507 */ 431 DLM_MIGRATE_REQUEST_MSG, /* 508 */ 432 DLM_MIG_LOCKRES_MSG, /* 509 */ 433 DLM_QUERY_JOIN_MSG, /* 510 */ 434 DLM_ASSERT_JOINED_MSG, /* 511 */ 435 DLM_CANCEL_JOIN_MSG, /* 512 */ 436 DLM_EXIT_DOMAIN_MSG, /* 513 */ 437 DLM_MASTER_REQUERY_MSG, /* 514 */ 438 DLM_LOCK_REQUEST_MSG, /* 515 */ 439 DLM_RECO_DATA_DONE_MSG, /* 516 */ 440 DLM_BEGIN_RECO_MSG, /* 517 */ 441 DLM_FINALIZE_RECO_MSG /* 518 */ 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, 455 DLM_RECO_NODE_DATA_REQUESTED, 456 DLM_RECO_NODE_DATA_RECEIVING, 457 DLM_RECO_NODE_DATA_DONE, 458 DLM_RECO_NODE_DATA_FINALIZE_SENT, 459 }; 460 461 462 enum { 463 DLM_MASTER_RESP_NO = 0, 464 DLM_MASTER_RESP_YES, 465 DLM_MASTER_RESP_MAYBE, 466 DLM_MASTER_RESP_ERROR 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 * NET_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 (NET_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, 644 JOIN_OK_NO_MAP, 645 JOIN_PROTOCOL_MISMATCH, 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 u32 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_exit_domain 724 { 725 u8 node_idx; 726 u8 pad1[3]; 727 }; 728 729 struct dlm_finalize_reco 730 { 731 u8 node_idx; 732 u8 dead_node; 733 u8 flags; 734 u8 pad1; 735 __be32 pad2; 736 }; 737 738 struct dlm_deref_lockres 739 { 740 u32 pad1; 741 u16 pad2; 742 u8 node_idx; 743 u8 namelen; 744 745 u8 name[O2NM_MAX_NAME_LEN]; 746 }; 747 748 static inline enum dlm_status 749 __dlm_lockres_state_to_status(struct dlm_lock_resource *res) 750 { 751 enum dlm_status status = DLM_NORMAL; 752 753 assert_spin_locked(&res->spinlock); 754 755 if (res->state & DLM_LOCK_RES_RECOVERING) 756 status = DLM_RECOVERING; 757 else if (res->state & DLM_LOCK_RES_MIGRATING) 758 status = DLM_MIGRATING; 759 else if (res->state & DLM_LOCK_RES_IN_PROGRESS) 760 status = DLM_FORWARD; 761 762 return status; 763 } 764 765 static inline u8 dlm_get_lock_cookie_node(u64 cookie) 766 { 767 u8 ret; 768 cookie >>= 56; 769 ret = (u8)(cookie & 0xffULL); 770 return ret; 771 } 772 773 static inline unsigned long long dlm_get_lock_cookie_seq(u64 cookie) 774 { 775 unsigned long long ret; 776 ret = ((unsigned long long)cookie) & 0x00ffffffffffffffULL; 777 return ret; 778 } 779 780 struct dlm_lock * dlm_new_lock(int type, u8 node, u64 cookie, 781 struct dlm_lockstatus *lksb); 782 void dlm_lock_get(struct dlm_lock *lock); 783 void dlm_lock_put(struct dlm_lock *lock); 784 785 void dlm_lock_attach_lockres(struct dlm_lock *lock, 786 struct dlm_lock_resource *res); 787 788 int dlm_create_lock_handler(struct o2net_msg *msg, u32 len, void *data, 789 void **ret_data); 790 int dlm_convert_lock_handler(struct o2net_msg *msg, u32 len, void *data, 791 void **ret_data); 792 int dlm_proxy_ast_handler(struct o2net_msg *msg, u32 len, void *data, 793 void **ret_data); 794 795 void dlm_revert_pending_convert(struct dlm_lock_resource *res, 796 struct dlm_lock *lock); 797 void dlm_revert_pending_lock(struct dlm_lock_resource *res, 798 struct dlm_lock *lock); 799 800 int dlm_unlock_lock_handler(struct o2net_msg *msg, u32 len, void *data, 801 void **ret_data); 802 void dlm_commit_pending_cancel(struct dlm_lock_resource *res, 803 struct dlm_lock *lock); 804 void dlm_commit_pending_unlock(struct dlm_lock_resource *res, 805 struct dlm_lock *lock); 806 807 int dlm_launch_thread(struct dlm_ctxt *dlm); 808 void dlm_complete_thread(struct dlm_ctxt *dlm); 809 int dlm_launch_recovery_thread(struct dlm_ctxt *dlm); 810 void dlm_complete_recovery_thread(struct dlm_ctxt *dlm); 811 void dlm_wait_for_recovery(struct dlm_ctxt *dlm); 812 void dlm_kick_recovery_thread(struct dlm_ctxt *dlm); 813 int dlm_is_node_dead(struct dlm_ctxt *dlm, u8 node); 814 int dlm_wait_for_node_death(struct dlm_ctxt *dlm, u8 node, int timeout); 815 int dlm_wait_for_node_recovery(struct dlm_ctxt *dlm, u8 node, int timeout); 816 817 void dlm_put(struct dlm_ctxt *dlm); 818 struct dlm_ctxt *dlm_grab(struct dlm_ctxt *dlm); 819 int dlm_domain_fully_joined(struct dlm_ctxt *dlm); 820 821 void __dlm_lockres_calc_usage(struct dlm_ctxt *dlm, 822 struct dlm_lock_resource *res); 823 void dlm_lockres_calc_usage(struct dlm_ctxt *dlm, 824 struct dlm_lock_resource *res); 825 static inline void dlm_lockres_get(struct dlm_lock_resource *res) 826 { 827 /* This is called on every lookup, so it might be worth 828 * inlining. */ 829 kref_get(&res->refs); 830 } 831 void dlm_lockres_put(struct dlm_lock_resource *res); 832 void __dlm_unhash_lockres(struct dlm_lock_resource *res); 833 void __dlm_insert_lockres(struct dlm_ctxt *dlm, 834 struct dlm_lock_resource *res); 835 struct dlm_lock_resource * __dlm_lookup_lockres_full(struct dlm_ctxt *dlm, 836 const char *name, 837 unsigned int len, 838 unsigned int hash); 839 struct dlm_lock_resource * __dlm_lookup_lockres(struct dlm_ctxt *dlm, 840 const char *name, 841 unsigned int len, 842 unsigned int hash); 843 struct dlm_lock_resource * dlm_lookup_lockres(struct dlm_ctxt *dlm, 844 const char *name, 845 unsigned int len); 846 847 int dlm_is_host_down(int errno); 848 void dlm_change_lockres_owner(struct dlm_ctxt *dlm, 849 struct dlm_lock_resource *res, 850 u8 owner); 851 struct dlm_lock_resource * dlm_get_lock_resource(struct dlm_ctxt *dlm, 852 const char *lockid, 853 int namelen, 854 int flags); 855 struct dlm_lock_resource *dlm_new_lockres(struct dlm_ctxt *dlm, 856 const char *name, 857 unsigned int namelen); 858 859 #define dlm_lockres_set_refmap_bit(bit,res) \ 860 __dlm_lockres_set_refmap_bit(bit,res,__FILE__,__LINE__) 861 #define dlm_lockres_clear_refmap_bit(bit,res) \ 862 __dlm_lockres_clear_refmap_bit(bit,res,__FILE__,__LINE__) 863 864 static inline void __dlm_lockres_set_refmap_bit(int bit, 865 struct dlm_lock_resource *res, 866 const char *file, 867 int line) 868 { 869 //printk("%s:%d:%.*s: setting bit %d\n", file, line, 870 // res->lockname.len, res->lockname.name, bit); 871 set_bit(bit, res->refmap); 872 } 873 874 static inline void __dlm_lockres_clear_refmap_bit(int bit, 875 struct dlm_lock_resource *res, 876 const char *file, 877 int line) 878 { 879 //printk("%s:%d:%.*s: clearing bit %d\n", file, line, 880 // res->lockname.len, res->lockname.name, bit); 881 clear_bit(bit, res->refmap); 882 } 883 884 void __dlm_lockres_drop_inflight_ref(struct dlm_ctxt *dlm, 885 struct dlm_lock_resource *res, 886 const char *file, 887 int line); 888 void __dlm_lockres_grab_inflight_ref(struct dlm_ctxt *dlm, 889 struct dlm_lock_resource *res, 890 int new_lockres, 891 const char *file, 892 int line); 893 #define dlm_lockres_drop_inflight_ref(d,r) \ 894 __dlm_lockres_drop_inflight_ref(d,r,__FILE__,__LINE__) 895 #define dlm_lockres_grab_inflight_ref(d,r) \ 896 __dlm_lockres_grab_inflight_ref(d,r,0,__FILE__,__LINE__) 897 #define dlm_lockres_grab_inflight_ref_new(d,r) \ 898 __dlm_lockres_grab_inflight_ref(d,r,1,__FILE__,__LINE__) 899 900 void dlm_queue_ast(struct dlm_ctxt *dlm, struct dlm_lock *lock); 901 void dlm_queue_bast(struct dlm_ctxt *dlm, struct dlm_lock *lock); 902 void dlm_do_local_ast(struct dlm_ctxt *dlm, 903 struct dlm_lock_resource *res, 904 struct dlm_lock *lock); 905 int dlm_do_remote_ast(struct dlm_ctxt *dlm, 906 struct dlm_lock_resource *res, 907 struct dlm_lock *lock); 908 void dlm_do_local_bast(struct dlm_ctxt *dlm, 909 struct dlm_lock_resource *res, 910 struct dlm_lock *lock, 911 int blocked_type); 912 int dlm_send_proxy_ast_msg(struct dlm_ctxt *dlm, 913 struct dlm_lock_resource *res, 914 struct dlm_lock *lock, 915 int msg_type, 916 int blocked_type, int flags); 917 static inline int dlm_send_proxy_bast(struct dlm_ctxt *dlm, 918 struct dlm_lock_resource *res, 919 struct dlm_lock *lock, 920 int blocked_type) 921 { 922 return dlm_send_proxy_ast_msg(dlm, res, lock, DLM_BAST, 923 blocked_type, 0); 924 } 925 926 static inline int dlm_send_proxy_ast(struct dlm_ctxt *dlm, 927 struct dlm_lock_resource *res, 928 struct dlm_lock *lock, 929 int flags) 930 { 931 return dlm_send_proxy_ast_msg(dlm, res, lock, DLM_AST, 932 0, flags); 933 } 934 935 void dlm_print_one_lock_resource(struct dlm_lock_resource *res); 936 void __dlm_print_one_lock_resource(struct dlm_lock_resource *res); 937 938 u8 dlm_nm_this_node(struct dlm_ctxt *dlm); 939 void dlm_kick_thread(struct dlm_ctxt *dlm, struct dlm_lock_resource *res); 940 void __dlm_dirty_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res); 941 942 943 int dlm_nm_init(struct dlm_ctxt *dlm); 944 int dlm_heartbeat_init(struct dlm_ctxt *dlm); 945 void dlm_hb_node_down_cb(struct o2nm_node *node, int idx, void *data); 946 void dlm_hb_node_up_cb(struct o2nm_node *node, int idx, void *data); 947 948 int dlm_empty_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res); 949 int dlm_finish_migration(struct dlm_ctxt *dlm, 950 struct dlm_lock_resource *res, 951 u8 old_master); 952 void dlm_lockres_release_ast(struct dlm_ctxt *dlm, 953 struct dlm_lock_resource *res); 954 void __dlm_lockres_reserve_ast(struct dlm_lock_resource *res); 955 956 int dlm_master_request_handler(struct o2net_msg *msg, u32 len, void *data, 957 void **ret_data); 958 int dlm_assert_master_handler(struct o2net_msg *msg, u32 len, void *data, 959 void **ret_data); 960 void dlm_assert_master_post_handler(int status, void *data, void *ret_data); 961 int dlm_deref_lockres_handler(struct o2net_msg *msg, u32 len, void *data, 962 void **ret_data); 963 int dlm_migrate_request_handler(struct o2net_msg *msg, u32 len, void *data, 964 void **ret_data); 965 int dlm_mig_lockres_handler(struct o2net_msg *msg, u32 len, void *data, 966 void **ret_data); 967 int dlm_master_requery_handler(struct o2net_msg *msg, u32 len, void *data, 968 void **ret_data); 969 int dlm_request_all_locks_handler(struct o2net_msg *msg, u32 len, void *data, 970 void **ret_data); 971 int dlm_reco_data_done_handler(struct o2net_msg *msg, u32 len, void *data, 972 void **ret_data); 973 int dlm_begin_reco_handler(struct o2net_msg *msg, u32 len, void *data, 974 void **ret_data); 975 int dlm_finalize_reco_handler(struct o2net_msg *msg, u32 len, void *data, 976 void **ret_data); 977 int dlm_do_master_requery(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, 978 u8 nodenum, u8 *real_master); 979 980 981 int dlm_dispatch_assert_master(struct dlm_ctxt *dlm, 982 struct dlm_lock_resource *res, 983 int ignore_higher, 984 u8 request_from, 985 u32 flags); 986 987 988 int dlm_send_one_lockres(struct dlm_ctxt *dlm, 989 struct dlm_lock_resource *res, 990 struct dlm_migratable_lockres *mres, 991 u8 send_to, 992 u8 flags); 993 void dlm_move_lockres_to_recovery_list(struct dlm_ctxt *dlm, 994 struct dlm_lock_resource *res); 995 996 /* will exit holding res->spinlock, but may drop in function */ 997 void __dlm_wait_on_lockres_flags(struct dlm_lock_resource *res, int flags); 998 void __dlm_wait_on_lockres_flags_set(struct dlm_lock_resource *res, int flags); 999 1000 /* will exit holding res->spinlock, but may drop in function */ 1001 static inline void __dlm_wait_on_lockres(struct dlm_lock_resource *res) 1002 { 1003 __dlm_wait_on_lockres_flags(res, (DLM_LOCK_RES_IN_PROGRESS| 1004 DLM_LOCK_RES_RECOVERING| 1005 DLM_LOCK_RES_MIGRATING)); 1006 } 1007 1008 /* create/destroy slab caches */ 1009 int dlm_init_master_caches(void); 1010 void dlm_destroy_master_caches(void); 1011 1012 int dlm_init_lock_cache(void); 1013 void dlm_destroy_lock_cache(void); 1014 1015 int dlm_init_mle_cache(void); 1016 void dlm_destroy_mle_cache(void); 1017 1018 void dlm_hb_event_notify_attached(struct dlm_ctxt *dlm, int idx, int node_up); 1019 int dlm_drop_lockres_ref(struct dlm_ctxt *dlm, 1020 struct dlm_lock_resource *res); 1021 void dlm_clean_master_list(struct dlm_ctxt *dlm, 1022 u8 dead_node); 1023 int dlm_lock_basts_flushed(struct dlm_ctxt *dlm, struct dlm_lock *lock); 1024 int __dlm_lockres_has_locks(struct dlm_lock_resource *res); 1025 int __dlm_lockres_unused(struct dlm_lock_resource *res); 1026 1027 static inline const char * dlm_lock_mode_name(int mode) 1028 { 1029 switch (mode) { 1030 case LKM_EXMODE: 1031 return "EX"; 1032 case LKM_PRMODE: 1033 return "PR"; 1034 case LKM_NLMODE: 1035 return "NL"; 1036 } 1037 return "UNKNOWN"; 1038 } 1039 1040 1041 static inline int dlm_lock_compatible(int existing, int request) 1042 { 1043 /* NO_LOCK compatible with all */ 1044 if (request == LKM_NLMODE || 1045 existing == LKM_NLMODE) 1046 return 1; 1047 1048 /* EX incompatible with all non-NO_LOCK */ 1049 if (request == LKM_EXMODE) 1050 return 0; 1051 1052 /* request must be PR, which is compatible with PR */ 1053 if (existing == LKM_PRMODE) 1054 return 1; 1055 1056 return 0; 1057 } 1058 1059 static inline int dlm_lock_on_list(struct list_head *head, 1060 struct dlm_lock *lock) 1061 { 1062 struct list_head *iter; 1063 struct dlm_lock *tmplock; 1064 1065 list_for_each(iter, head) { 1066 tmplock = list_entry(iter, struct dlm_lock, list); 1067 if (tmplock == lock) 1068 return 1; 1069 } 1070 return 0; 1071 } 1072 1073 1074 static inline enum dlm_status dlm_err_to_dlm_status(int err) 1075 { 1076 enum dlm_status ret; 1077 if (err == -ENOMEM) 1078 ret = DLM_SYSERR; 1079 else if (err == -ETIMEDOUT || o2net_link_down(err, NULL)) 1080 ret = DLM_NOLOCKMGR; 1081 else if (err == -EINVAL) 1082 ret = DLM_BADPARAM; 1083 else if (err == -ENAMETOOLONG) 1084 ret = DLM_IVBUFLEN; 1085 else 1086 ret = DLM_BADARGS; 1087 return ret; 1088 } 1089 1090 1091 static inline void dlm_node_iter_init(unsigned long *map, 1092 struct dlm_node_iter *iter) 1093 { 1094 memcpy(iter->node_map, map, sizeof(iter->node_map)); 1095 iter->curnode = -1; 1096 } 1097 1098 static inline int dlm_node_iter_next(struct dlm_node_iter *iter) 1099 { 1100 int bit; 1101 bit = find_next_bit(iter->node_map, O2NM_MAX_NODES, iter->curnode+1); 1102 if (bit >= O2NM_MAX_NODES) { 1103 iter->curnode = O2NM_MAX_NODES; 1104 return -ENOENT; 1105 } 1106 iter->curnode = bit; 1107 return bit; 1108 } 1109 1110 1111 1112 #endif /* DLMCOMMON_H */ 1113