fs/ocfs2/dlm/dlmcommon.h

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