fs/ocfs2/dlm/dlmcommon.h

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