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 52 enum dlm_mle_type { 53 DLM_MLE_BLOCK = 0, 54 DLM_MLE_MASTER = 1, 55 DLM_MLE_MIGRATION = 2, 56 DLM_MLE_NUM_TYPES = 3, 57 }; 58 59 struct 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 83 enum dlm_ast_type { 84 DLM_AST = 0, 85 DLM_BAST = 1, 86 DLM_ASTUNLOCK = 2, 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 97 static 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 107 108 struct dlm_recovery_ctxt 109 { 110 struct list_head resources; 111 struct list_head node_data; 112 u8 new_master; 113 u8 dead_node; 114 u16 state; 115 unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; 116 wait_queue_head_t event; 117 }; 118 119 enum dlm_ctxt_state { 120 DLM_CTXT_NEW = 0, 121 DLM_CTXT_JOINED = 1, 122 DLM_CTXT_IN_SHUTDOWN = 2, 123 DLM_CTXT_LEAVING = 3, 124 }; 125 126 struct dlm_ctxt 127 { 128 struct list_head list; 129 struct hlist_head **lockres_hash; 130 struct list_head dirty_list; 131 struct list_head purge_list; 132 struct list_head pending_asts; 133 struct list_head pending_basts; 134 struct list_head tracking_list; 135 unsigned int purge_count; 136 spinlock_t spinlock; 137 spinlock_t ast_lock; 138 spinlock_t track_lock; 139 char *name; 140 u8 node_num; 141 u32 key; 142 u8 joining_node; 143 wait_queue_head_t dlm_join_events; 144 unsigned long live_nodes_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; 145 unsigned long domain_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; 146 unsigned long exit_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 }; 189 190 static 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 } 194 195 static 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. */ 206 void dlm_dispatch_work(struct work_struct *work); 207 208 struct dlm_lock_resource; 209 struct dlm_work_item; 210 211 typedef void (dlm_workfunc_t)(struct dlm_work_item *, void *); 212 213 struct dlm_request_all_locks_priv 214 { 215 u8 reco_master; 216 u8 dead_node; 217 }; 218 219 struct dlm_mig_lockres_priv 220 { 221 struct dlm_lock_resource *lockres; 222 u8 real_master; 223 u8 extra_ref; 224 }; 225 226 struct dlm_assert_master_priv 227 { 228 struct dlm_lock_resource *lockres; 229 u8 request_from; 230 u32 flags; 231 unsigned ignore_higher:1; 232 }; 233 234 struct dlm_deref_lockres_priv 235 { 236 struct dlm_lock_resource *deref_res; 237 u8 deref_node; 238 }; 239 240 struct 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 }; 253 254 static 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 } 264 265 266 267 static 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) 290 291 struct 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 }; 336 337 struct 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 352 353 struct 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 387 388 389 enum dlm_lockres_list { 390 DLM_GRANTED_LIST = 0, 391 DLM_CONVERTING_LIST = 1, 392 DLM_BLOCKED_LIST = 2, 393 }; 394 395 static 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 } 403 404 static inline char *dlm_list_in_text(enum dlm_lockres_list idx) 405 { 406 if (idx == DLM_GRANTED_LIST) 407 return "granted"; 408 else if (idx == DLM_CONVERTING_LIST) 409 return "converting"; 410 else if (idx == DLM_BLOCKED_LIST) 411 return "blocked"; 412 else 413 return "unknown"; 414 } 415 416 static inline struct list_head * 417 dlm_list_idx_to_ptr(struct dlm_lock_resource *res, enum dlm_lockres_list idx) 418 { 419 struct list_head *ret = NULL; 420 if (idx == DLM_GRANTED_LIST) 421 ret = &res->granted; 422 else if (idx == DLM_CONVERTING_LIST) 423 ret = &res->converting; 424 else if (idx == DLM_BLOCKED_LIST) 425 ret = &res->blocked; 426 else 427 BUG(); 428 return ret; 429 } 430 431 432 433 434 struct dlm_node_iter 435 { 436 unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; 437 int curnode; 438 }; 439 440 441 enum { 442 DLM_MASTER_REQUEST_MSG = 500, 443 DLM_UNUSED_MSG1 = 501, 444 DLM_ASSERT_MASTER_MSG = 502, 445 DLM_CREATE_LOCK_MSG = 503, 446 DLM_CONVERT_LOCK_MSG = 504, 447 DLM_PROXY_AST_MSG = 505, 448 DLM_UNLOCK_LOCK_MSG = 506, 449 DLM_DEREF_LOCKRES_MSG = 507, 450 DLM_MIGRATE_REQUEST_MSG = 508, 451 DLM_MIG_LOCKRES_MSG = 509, 452 DLM_QUERY_JOIN_MSG = 510, 453 DLM_ASSERT_JOINED_MSG = 511, 454 DLM_CANCEL_JOIN_MSG = 512, 455 DLM_EXIT_DOMAIN_MSG = 513, 456 DLM_MASTER_REQUERY_MSG = 514, 457 DLM_LOCK_REQUEST_MSG = 515, 458 DLM_RECO_DATA_DONE_MSG = 516, 459 DLM_BEGIN_RECO_MSG = 517, 460 DLM_FINALIZE_RECO_MSG = 518, 461 DLM_QUERY_REGION = 519, 462 DLM_QUERY_NODEINFO = 520, 463 DLM_BEGIN_EXIT_DOMAIN_MSG = 521, 464 }; 465 466 struct dlm_reco_node_data 467 { 468 int state; 469 u8 node_num; 470 struct list_head list; 471 }; 472 473 enum { 474 DLM_RECO_NODE_DATA_DEAD = -1, 475 DLM_RECO_NODE_DATA_INIT = 0, 476 DLM_RECO_NODE_DATA_REQUESTING = 1, 477 DLM_RECO_NODE_DATA_REQUESTED = 2, 478 DLM_RECO_NODE_DATA_RECEIVING = 3, 479 DLM_RECO_NODE_DATA_DONE = 4, 480 DLM_RECO_NODE_DATA_FINALIZE_SENT = 5, 481 }; 482 483 484 enum { 485 DLM_MASTER_RESP_NO = 0, 486 DLM_MASTER_RESP_YES = 1, 487 DLM_MASTER_RESP_MAYBE = 2, 488 DLM_MASTER_RESP_ERROR = 3, 489 }; 490 491 492 struct dlm_master_request 493 { 494 u8 node_idx; 495 u8 namelen; 496 __be16 pad1; 497 __be32 flags; 498 499 u8 name[O2NM_MAX_NAME_LEN]; 500 }; 501 502 #define DLM_ASSERT_RESPONSE_REASSERT 0x00000001 503 #define DLM_ASSERT_RESPONSE_MASTERY_REF 0x00000002 504 505 #define DLM_ASSERT_MASTER_MLE_CLEANUP 0x00000001 506 #define DLM_ASSERT_MASTER_REQUERY 0x00000002 507 #define DLM_ASSERT_MASTER_FINISH_MIGRATION 0x00000004 508 struct dlm_assert_master 509 { 510 u8 node_idx; 511 u8 namelen; 512 __be16 pad1; 513 __be32 flags; 514 515 u8 name[O2NM_MAX_NAME_LEN]; 516 }; 517 518 #define DLM_MIGRATE_RESPONSE_MASTERY_REF 0x00000001 519 520 struct dlm_migrate_request 521 { 522 u8 master; 523 u8 new_master; 524 u8 namelen; 525 u8 pad1; 526 __be32 pad2; 527 u8 name[O2NM_MAX_NAME_LEN]; 528 }; 529 530 struct dlm_master_requery 531 { 532 u8 pad1; 533 u8 pad2; 534 u8 node_idx; 535 u8 namelen; 536 __be32 pad3; 537 u8 name[O2NM_MAX_NAME_LEN]; 538 }; 539 540 #define DLM_MRES_RECOVERY 0x01 541 #define DLM_MRES_MIGRATION 0x02 542 #define DLM_MRES_ALL_DONE 0x04 543 544 /* 545 * We would like to get one whole lockres into a single network 546 * message whenever possible. Generally speaking, there will be 547 * at most one dlm_lock on a lockres for each node in the cluster, 548 * plus (infrequently) any additional locks coming in from userdlm. 549 * 550 * struct _dlm_lockres_page 551 * { 552 * dlm_migratable_lockres mres; 553 * dlm_migratable_lock ml[DLM_MAX_MIGRATABLE_LOCKS]; 554 * u8 pad[DLM_MIG_LOCKRES_RESERVED]; 555 * }; 556 * 557 * from ../cluster/tcp.h 558 * NET_MAX_PAYLOAD_BYTES (4096 - sizeof(net_msg)) 559 * (roughly 4080 bytes) 560 * and sizeof(dlm_migratable_lockres) = 112 bytes 561 * and sizeof(dlm_migratable_lock) = 16 bytes 562 * 563 * Choosing DLM_MAX_MIGRATABLE_LOCKS=240 and 564 * DLM_MIG_LOCKRES_RESERVED=128 means we have this: 565 * 566 * (DLM_MAX_MIGRATABLE_LOCKS * sizeof(dlm_migratable_lock)) + 567 * sizeof(dlm_migratable_lockres) + DLM_MIG_LOCKRES_RESERVED = 568 * NET_MAX_PAYLOAD_BYTES 569 * (240 * 16) + 112 + 128 = 4080 570 * 571 * So a lockres would need more than 240 locks before it would 572 * use more than one network packet to recover. Not too bad. 573 */ 574 #define DLM_MAX_MIGRATABLE_LOCKS 240 575 576 struct dlm_migratable_lockres 577 { 578 u8 master; 579 u8 lockname_len; 580 u8 num_locks; // locks sent in this structure 581 u8 flags; 582 __be32 total_locks; // locks to be sent for this migration cookie 583 __be64 mig_cookie; // cookie for this lockres migration 584 // or zero if not needed 585 // 16 bytes 586 u8 lockname[DLM_LOCKID_NAME_MAX]; 587 // 48 bytes 588 u8 lvb[DLM_LVB_LEN]; 589 // 112 bytes 590 struct dlm_migratable_lock ml[0]; // 16 bytes each, begins at byte 112 591 }; 592 #define DLM_MIG_LOCKRES_MAX_LEN \ 593 (sizeof(struct dlm_migratable_lockres) + \ 594 (sizeof(struct dlm_migratable_lock) * \ 595 DLM_MAX_MIGRATABLE_LOCKS) ) 596 597 /* from above, 128 bytes 598 * for some undetermined future use */ 599 #define DLM_MIG_LOCKRES_RESERVED (NET_MAX_PAYLOAD_BYTES - \ 600 DLM_MIG_LOCKRES_MAX_LEN) 601 602 struct dlm_create_lock 603 { 604 __be64 cookie; 605 606 __be32 flags; 607 u8 pad1; 608 u8 node_idx; 609 s8 requested_type; 610 u8 namelen; 611 612 u8 name[O2NM_MAX_NAME_LEN]; 613 }; 614 615 struct dlm_convert_lock 616 { 617 __be64 cookie; 618 619 __be32 flags; 620 u8 pad1; 621 u8 node_idx; 622 s8 requested_type; 623 u8 namelen; 624 625 u8 name[O2NM_MAX_NAME_LEN]; 626 627 s8 lvb[0]; 628 }; 629 #define DLM_CONVERT_LOCK_MAX_LEN (sizeof(struct dlm_convert_lock)+DLM_LVB_LEN) 630 631 struct dlm_unlock_lock 632 { 633 __be64 cookie; 634 635 __be32 flags; 636 __be16 pad1; 637 u8 node_idx; 638 u8 namelen; 639 640 u8 name[O2NM_MAX_NAME_LEN]; 641 642 s8 lvb[0]; 643 }; 644 #define DLM_UNLOCK_LOCK_MAX_LEN (sizeof(struct dlm_unlock_lock)+DLM_LVB_LEN) 645 646 struct dlm_proxy_ast 647 { 648 __be64 cookie; 649 650 __be32 flags; 651 u8 node_idx; 652 u8 type; 653 u8 blocked_type; 654 u8 namelen; 655 656 u8 name[O2NM_MAX_NAME_LEN]; 657 658 s8 lvb[0]; 659 }; 660 #define DLM_PROXY_AST_MAX_LEN (sizeof(struct dlm_proxy_ast)+DLM_LVB_LEN) 661 662 #define DLM_MOD_KEY (0x666c6172) 663 enum dlm_query_join_response_code { 664 JOIN_DISALLOW = 0, 665 JOIN_OK = 1, 666 JOIN_OK_NO_MAP = 2, 667 JOIN_PROTOCOL_MISMATCH = 3, 668 }; 669 670 struct dlm_query_join_packet { 671 u8 code; /* Response code. dlm_minor and fs_minor 672 are only valid if this is JOIN_OK */ 673 u8 dlm_minor; /* The minor version of the protocol the 674 dlm is speaking. */ 675 u8 fs_minor; /* The minor version of the protocol the 676 filesystem is speaking. */ 677 u8 reserved; 678 }; 679 680 union dlm_query_join_response { 681 __be32 intval; 682 struct dlm_query_join_packet packet; 683 }; 684 685 struct dlm_lock_request 686 { 687 u8 node_idx; 688 u8 dead_node; 689 __be16 pad1; 690 __be32 pad2; 691 }; 692 693 struct dlm_reco_data_done 694 { 695 u8 node_idx; 696 u8 dead_node; 697 __be16 pad1; 698 __be32 pad2; 699 700 /* unused for now */ 701 /* eventually we can use this to attempt 702 * lvb recovery based on each node's info */ 703 u8 reco_lvb[DLM_LVB_LEN]; 704 }; 705 706 struct dlm_begin_reco 707 { 708 u8 node_idx; 709 u8 dead_node; 710 __be16 pad1; 711 __be32 pad2; 712 }; 713 714 715 #define BITS_PER_BYTE 8 716 #define BITS_TO_BYTES(bits) (((bits)+BITS_PER_BYTE-1)/BITS_PER_BYTE) 717 718 struct dlm_query_join_request 719 { 720 u8 node_idx; 721 u8 pad1[2]; 722 u8 name_len; 723 struct dlm_protocol_version dlm_proto; 724 struct dlm_protocol_version fs_proto; 725 u8 domain[O2NM_MAX_NAME_LEN]; 726 u8 node_map[BITS_TO_BYTES(O2NM_MAX_NODES)]; 727 }; 728 729 struct dlm_assert_joined 730 { 731 u8 node_idx; 732 u8 pad1[2]; 733 u8 name_len; 734 u8 domain[O2NM_MAX_NAME_LEN]; 735 }; 736 737 struct dlm_cancel_join 738 { 739 u8 node_idx; 740 u8 pad1[2]; 741 u8 name_len; 742 u8 domain[O2NM_MAX_NAME_LEN]; 743 }; 744 745 struct dlm_query_region { 746 u8 qr_node; 747 u8 qr_numregions; 748 u8 qr_namelen; 749 u8 pad1; 750 u8 qr_domain[O2NM_MAX_NAME_LEN]; 751 u8 qr_regions[O2HB_MAX_REGION_NAME_LEN * O2NM_MAX_REGIONS]; 752 }; 753 754 struct dlm_node_info { 755 u8 ni_nodenum; 756 u8 pad1; 757 __be16 ni_ipv4_port; 758 __be32 ni_ipv4_address; 759 }; 760 761 struct dlm_query_nodeinfo { 762 u8 qn_nodenum; 763 u8 qn_numnodes; 764 u8 qn_namelen; 765 u8 pad1; 766 u8 qn_domain[O2NM_MAX_NAME_LEN]; 767 struct dlm_node_info qn_nodes[O2NM_MAX_NODES]; 768 }; 769 770 struct dlm_exit_domain 771 { 772 u8 node_idx; 773 u8 pad1[3]; 774 }; 775 776 struct dlm_finalize_reco 777 { 778 u8 node_idx; 779 u8 dead_node; 780 u8 flags; 781 u8 pad1; 782 __be32 pad2; 783 }; 784 785 struct dlm_deref_lockres 786 { 787 u32 pad1; 788 u16 pad2; 789 u8 node_idx; 790 u8 namelen; 791 792 u8 name[O2NM_MAX_NAME_LEN]; 793 }; 794 795 static inline enum dlm_status 796 __dlm_lockres_state_to_status(struct dlm_lock_resource *res) 797 { 798 enum dlm_status status = DLM_NORMAL; 799 800 assert_spin_locked(&res->spinlock); 801 802 if (res->state & DLM_LOCK_RES_RECOVERING) 803 status = DLM_RECOVERING; 804 else if (res->state & DLM_LOCK_RES_MIGRATING) 805 status = DLM_MIGRATING; 806 else if (res->state & DLM_LOCK_RES_IN_PROGRESS) 807 status = DLM_FORWARD; 808 809 return status; 810 } 811 812 static inline u8 dlm_get_lock_cookie_node(u64 cookie) 813 { 814 u8 ret; 815 cookie >>= 56; 816 ret = (u8)(cookie & 0xffULL); 817 return ret; 818 } 819 820 static inline unsigned long long dlm_get_lock_cookie_seq(u64 cookie) 821 { 822 unsigned long long ret; 823 ret = ((unsigned long long)cookie) & 0x00ffffffffffffffULL; 824 return ret; 825 } 826 827 struct dlm_lock * dlm_new_lock(int type, u8 node, u64 cookie, 828 struct dlm_lockstatus *lksb); 829 void dlm_lock_get(struct dlm_lock *lock); 830 void dlm_lock_put(struct dlm_lock *lock); 831 832 void dlm_lock_attach_lockres(struct dlm_lock *lock, 833 struct dlm_lock_resource *res); 834 835 int dlm_create_lock_handler(struct o2net_msg *msg, u32 len, void *data, 836 void **ret_data); 837 int dlm_convert_lock_handler(struct o2net_msg *msg, u32 len, void *data, 838 void **ret_data); 839 int dlm_proxy_ast_handler(struct o2net_msg *msg, u32 len, void *data, 840 void **ret_data); 841 842 void dlm_revert_pending_convert(struct dlm_lock_resource *res, 843 struct dlm_lock *lock); 844 void dlm_revert_pending_lock(struct dlm_lock_resource *res, 845 struct dlm_lock *lock); 846 847 int dlm_unlock_lock_handler(struct o2net_msg *msg, u32 len, void *data, 848 void **ret_data); 849 void dlm_commit_pending_cancel(struct dlm_lock_resource *res, 850 struct dlm_lock *lock); 851 void dlm_commit_pending_unlock(struct dlm_lock_resource *res, 852 struct dlm_lock *lock); 853 854 int dlm_launch_thread(struct dlm_ctxt *dlm); 855 void dlm_complete_thread(struct dlm_ctxt *dlm); 856 int dlm_launch_recovery_thread(struct dlm_ctxt *dlm); 857 void dlm_complete_recovery_thread(struct dlm_ctxt *dlm); 858 void dlm_wait_for_recovery(struct dlm_ctxt *dlm); 859 void dlm_kick_recovery_thread(struct dlm_ctxt *dlm); 860 int dlm_is_node_dead(struct dlm_ctxt *dlm, u8 node); 861 void dlm_wait_for_node_death(struct dlm_ctxt *dlm, u8 node, int timeout); 862 void dlm_wait_for_node_recovery(struct dlm_ctxt *dlm, u8 node, int timeout); 863 864 void dlm_put(struct dlm_ctxt *dlm); 865 struct dlm_ctxt *dlm_grab(struct dlm_ctxt *dlm); 866 int dlm_domain_fully_joined(struct dlm_ctxt *dlm); 867 868 void __dlm_lockres_calc_usage(struct dlm_ctxt *dlm, 869 struct dlm_lock_resource *res); 870 void dlm_lockres_calc_usage(struct dlm_ctxt *dlm, 871 struct dlm_lock_resource *res); 872 static inline void dlm_lockres_get(struct dlm_lock_resource *res) 873 { 874 /* This is called on every lookup, so it might be worth 875 * inlining. */ 876 kref_get(&res->refs); 877 } 878 void dlm_lockres_put(struct dlm_lock_resource *res); 879 void __dlm_unhash_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res); 880 void __dlm_insert_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res); 881 struct dlm_lock_resource * __dlm_lookup_lockres_full(struct dlm_ctxt *dlm, 882 const char *name, 883 unsigned int len, 884 unsigned int hash); 885 struct dlm_lock_resource * __dlm_lookup_lockres(struct dlm_ctxt *dlm, 886 const char *name, 887 unsigned int len, 888 unsigned int hash); 889 struct dlm_lock_resource * dlm_lookup_lockres(struct dlm_ctxt *dlm, 890 const char *name, 891 unsigned int len); 892 893 int dlm_is_host_down(int errno); 894 895 struct dlm_lock_resource * dlm_get_lock_resource(struct dlm_ctxt *dlm, 896 const char *lockid, 897 int namelen, 898 int flags); 899 struct dlm_lock_resource *dlm_new_lockres(struct dlm_ctxt *dlm, 900 const char *name, 901 unsigned int namelen); 902 903 void dlm_lockres_set_refmap_bit(struct dlm_ctxt *dlm, 904 struct dlm_lock_resource *res, int bit); 905 void dlm_lockres_clear_refmap_bit(struct dlm_ctxt *dlm, 906 struct dlm_lock_resource *res, int bit); 907 908 void dlm_lockres_drop_inflight_ref(struct dlm_ctxt *dlm, 909 struct dlm_lock_resource *res); 910 void dlm_lockres_grab_inflight_ref(struct dlm_ctxt *dlm, 911 struct dlm_lock_resource *res); 912 913 void dlm_queue_ast(struct dlm_ctxt *dlm, struct dlm_lock *lock); 914 void dlm_queue_bast(struct dlm_ctxt *dlm, struct dlm_lock *lock); 915 void __dlm_queue_ast(struct dlm_ctxt *dlm, struct dlm_lock *lock); 916 void __dlm_queue_bast(struct dlm_ctxt *dlm, struct dlm_lock *lock); 917 void dlm_do_local_ast(struct dlm_ctxt *dlm, 918 struct dlm_lock_resource *res, 919 struct dlm_lock *lock); 920 int dlm_do_remote_ast(struct dlm_ctxt *dlm, 921 struct dlm_lock_resource *res, 922 struct dlm_lock *lock); 923 void dlm_do_local_bast(struct dlm_ctxt *dlm, 924 struct dlm_lock_resource *res, 925 struct dlm_lock *lock, 926 int blocked_type); 927 int dlm_send_proxy_ast_msg(struct dlm_ctxt *dlm, 928 struct dlm_lock_resource *res, 929 struct dlm_lock *lock, 930 int msg_type, 931 int blocked_type, int flags); 932 static inline int dlm_send_proxy_bast(struct dlm_ctxt *dlm, 933 struct dlm_lock_resource *res, 934 struct dlm_lock *lock, 935 int blocked_type) 936 { 937 return dlm_send_proxy_ast_msg(dlm, res, lock, DLM_BAST, 938 blocked_type, 0); 939 } 940 941 static inline int dlm_send_proxy_ast(struct dlm_ctxt *dlm, 942 struct dlm_lock_resource *res, 943 struct dlm_lock *lock, 944 int flags) 945 { 946 return dlm_send_proxy_ast_msg(dlm, res, lock, DLM_AST, 947 0, flags); 948 } 949 950 void dlm_print_one_lock_resource(struct dlm_lock_resource *res); 951 void __dlm_print_one_lock_resource(struct dlm_lock_resource *res); 952 953 u8 dlm_nm_this_node(struct dlm_ctxt *dlm); 954 void dlm_kick_thread(struct dlm_ctxt *dlm, struct dlm_lock_resource *res); 955 void __dlm_dirty_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res); 956 957 958 int dlm_nm_init(struct dlm_ctxt *dlm); 959 int dlm_heartbeat_init(struct dlm_ctxt *dlm); 960 void dlm_hb_node_down_cb(struct o2nm_node *node, int idx, void *data); 961 void dlm_hb_node_up_cb(struct o2nm_node *node, int idx, void *data); 962 963 int dlm_empty_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res); 964 int dlm_finish_migration(struct dlm_ctxt *dlm, 965 struct dlm_lock_resource *res, 966 u8 old_master); 967 void dlm_lockres_release_ast(struct dlm_ctxt *dlm, 968 struct dlm_lock_resource *res); 969 void __dlm_lockres_reserve_ast(struct dlm_lock_resource *res); 970 971 int dlm_master_request_handler(struct o2net_msg *msg, u32 len, void *data, 972 void **ret_data); 973 int dlm_assert_master_handler(struct o2net_msg *msg, u32 len, void *data, 974 void **ret_data); 975 void dlm_assert_master_post_handler(int status, void *data, void *ret_data); 976 int dlm_deref_lockres_handler(struct o2net_msg *msg, u32 len, void *data, 977 void **ret_data); 978 int dlm_migrate_request_handler(struct o2net_msg *msg, u32 len, void *data, 979 void **ret_data); 980 int dlm_mig_lockres_handler(struct o2net_msg *msg, u32 len, void *data, 981 void **ret_data); 982 int dlm_master_requery_handler(struct o2net_msg *msg, u32 len, void *data, 983 void **ret_data); 984 int dlm_request_all_locks_handler(struct o2net_msg *msg, u32 len, void *data, 985 void **ret_data); 986 int dlm_reco_data_done_handler(struct o2net_msg *msg, u32 len, void *data, 987 void **ret_data); 988 int dlm_begin_reco_handler(struct o2net_msg *msg, u32 len, void *data, 989 void **ret_data); 990 int dlm_finalize_reco_handler(struct o2net_msg *msg, u32 len, void *data, 991 void **ret_data); 992 int dlm_do_master_requery(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, 993 u8 nodenum, u8 *real_master); 994 995 996 int dlm_dispatch_assert_master(struct dlm_ctxt *dlm, 997 struct dlm_lock_resource *res, 998 int ignore_higher, 999 u8 request_from, 1000 u32 flags); 1001 1002 1003 int dlm_send_one_lockres(struct dlm_ctxt *dlm, 1004 struct dlm_lock_resource *res, 1005 struct dlm_migratable_lockres *mres, 1006 u8 send_to, 1007 u8 flags); 1008 void dlm_move_lockres_to_recovery_list(struct dlm_ctxt *dlm, 1009 struct dlm_lock_resource *res); 1010 1011 /* will exit holding res->spinlock, but may drop in function */ 1012 void __dlm_wait_on_lockres_flags(struct dlm_lock_resource *res, int flags); 1013 void __dlm_wait_on_lockres_flags_set(struct dlm_lock_resource *res, int flags); 1014 1015 /* will exit holding res->spinlock, but may drop in function */ 1016 static inline void __dlm_wait_on_lockres(struct dlm_lock_resource *res) 1017 { 1018 __dlm_wait_on_lockres_flags(res, (DLM_LOCK_RES_IN_PROGRESS| 1019 DLM_LOCK_RES_RECOVERING| 1020 DLM_LOCK_RES_MIGRATING)); 1021 } 1022 1023 void __dlm_unlink_mle(struct dlm_ctxt *dlm, struct dlm_master_list_entry *mle); 1024 void __dlm_insert_mle(struct dlm_ctxt *dlm, struct dlm_master_list_entry *mle); 1025 1026 /* create/destroy slab caches */ 1027 int dlm_init_master_caches(void); 1028 void dlm_destroy_master_caches(void); 1029 1030 int dlm_init_lock_cache(void); 1031 void dlm_destroy_lock_cache(void); 1032 1033 int dlm_init_mle_cache(void); 1034 void dlm_destroy_mle_cache(void); 1035 1036 void dlm_hb_event_notify_attached(struct dlm_ctxt *dlm, int idx, int node_up); 1037 int dlm_drop_lockres_ref(struct dlm_ctxt *dlm, 1038 struct dlm_lock_resource *res); 1039 void dlm_clean_master_list(struct dlm_ctxt *dlm, 1040 u8 dead_node); 1041 void dlm_force_free_mles(struct dlm_ctxt *dlm); 1042 int dlm_lock_basts_flushed(struct dlm_ctxt *dlm, struct dlm_lock *lock); 1043 int __dlm_lockres_has_locks(struct dlm_lock_resource *res); 1044 int __dlm_lockres_unused(struct dlm_lock_resource *res); 1045 1046 static inline const char * dlm_lock_mode_name(int mode) 1047 { 1048 switch (mode) { 1049 case LKM_EXMODE: 1050 return "EX"; 1051 case LKM_PRMODE: 1052 return "PR"; 1053 case LKM_NLMODE: 1054 return "NL"; 1055 } 1056 return "UNKNOWN"; 1057 } 1058 1059 1060 static inline int dlm_lock_compatible(int existing, int request) 1061 { 1062 /* NO_LOCK compatible with all */ 1063 if (request == LKM_NLMODE || 1064 existing == LKM_NLMODE) 1065 return 1; 1066 1067 /* EX incompatible with all non-NO_LOCK */ 1068 if (request == LKM_EXMODE) 1069 return 0; 1070 1071 /* request must be PR, which is compatible with PR */ 1072 if (existing == LKM_PRMODE) 1073 return 1; 1074 1075 return 0; 1076 } 1077 1078 static inline int dlm_lock_on_list(struct list_head *head, 1079 struct dlm_lock *lock) 1080 { 1081 struct dlm_lock *tmplock; 1082 1083 list_for_each_entry(tmplock, head, list) { 1084 if (tmplock == lock) 1085 return 1; 1086 } 1087 return 0; 1088 } 1089 1090 1091 static inline enum dlm_status dlm_err_to_dlm_status(int err) 1092 { 1093 enum dlm_status ret; 1094 if (err == -ENOMEM) 1095 ret = DLM_SYSERR; 1096 else if (err == -ETIMEDOUT || o2net_link_down(err, NULL)) 1097 ret = DLM_NOLOCKMGR; 1098 else if (err == -EINVAL) 1099 ret = DLM_BADPARAM; 1100 else if (err == -ENAMETOOLONG) 1101 ret = DLM_IVBUFLEN; 1102 else 1103 ret = DLM_BADARGS; 1104 return ret; 1105 } 1106 1107 1108 static inline void dlm_node_iter_init(unsigned long *map, 1109 struct dlm_node_iter *iter) 1110 { 1111 memcpy(iter->node_map, map, sizeof(iter->node_map)); 1112 iter->curnode = -1; 1113 } 1114 1115 static inline int dlm_node_iter_next(struct dlm_node_iter *iter) 1116 { 1117 int bit; 1118 bit = find_next_bit(iter->node_map, O2NM_MAX_NODES, iter->curnode+1); 1119 if (bit >= O2NM_MAX_NODES) { 1120 iter->curnode = O2NM_MAX_NODES; 1121 return -ENOENT; 1122 } 1123 iter->curnode = bit; 1124 return bit; 1125 } 1126 1127 static inline void dlm_set_lockres_owner(struct dlm_ctxt *dlm, 1128 struct dlm_lock_resource *res, 1129 u8 owner) 1130 { 1131 assert_spin_locked(&res->spinlock); 1132 1133 res->owner = owner; 1134 } 1135 1136 static inline void dlm_change_lockres_owner(struct dlm_ctxt *dlm, 1137 struct dlm_lock_resource *res, 1138 u8 owner) 1139 { 1140 assert_spin_locked(&res->spinlock); 1141 1142 if (owner != res->owner) 1143 dlm_set_lockres_owner(dlm, res, owner); 1144 } 1145 1146 #endif /* DLMCOMMON_H */ 1147