1/****************************************************************************** 2******************************************************************************* 3** 4** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. 5** Copyright (C) 2004-2010 Red Hat, Inc. All rights reserved. 6** 7** This copyrighted material is made available to anyone wishing to use, 8** modify, copy, or redistribute it subject to the terms and conditions 9** of the GNU General Public License v.2. 10** 11******************************************************************************* 12******************************************************************************/ 13 14#ifndef __DLM_INTERNAL_DOT_H__ 15#define __DLM_INTERNAL_DOT_H__ 16 17/* 18 * This is the main header file to be included in each DLM source file. 19 */ 20 21#include <linux/module.h> 22#include <linux/slab.h> 23#include <linux/sched.h> 24#include <linux/types.h> 25#include <linux/ctype.h> 26#include <linux/spinlock.h> 27#include <linux/vmalloc.h> 28#include <linux/list.h> 29#include <linux/errno.h> 30#include <linux/random.h> 31#include <linux/delay.h> 32#include <linux/socket.h> 33#include <linux/kthread.h> 34#include <linux/kobject.h> 35#include <linux/kref.h> 36#include <linux/kernel.h> 37#include <linux/jhash.h> 38#include <linux/miscdevice.h> 39#include <linux/mutex.h> 40#include <asm/uaccess.h> 41 42#include <linux/dlm.h> 43#include "config.h" 44 45 46#define DLM_INBUF_LEN 148 47 48struct dlm_ls; 49struct dlm_lkb; 50struct dlm_rsb; 51struct dlm_member; 52struct dlm_lkbtable; 53struct dlm_rsbtable; 54struct dlm_dirtable; 55struct dlm_direntry; 56struct dlm_recover; 57struct dlm_header; 58struct dlm_message; 59struct dlm_rcom; 60struct dlm_mhandle; 61 62#define log_print(fmt, args...) \ 63 printk(KERN_ERR "dlm: "fmt"\n" , ##args) 64#define log_error(ls, fmt, args...) \ 65 printk(KERN_ERR "dlm: %s: " fmt "\n", (ls)->ls_name , ##args) 66 67#define log_debug(ls, fmt, args...) \ 68do { \ 69 if (dlm_config.ci_log_debug) \ 70 printk(KERN_DEBUG "dlm: %s: " fmt "\n", \ 71 (ls)->ls_name , ##args); \ 72} while (0) 73 74#define DLM_ASSERT(x, do) \ 75{ \ 76 if (!(x)) \ 77 { \ 78 printk(KERN_ERR "\nDLM: Assertion failed on line %d of file %s\n" \ 79 "DLM: assertion: \"%s\"\n" \ 80 "DLM: time = %lu\n", \ 81 __LINE__, __FILE__, #x, jiffies); \ 82 {do} \ 83 printk("\n"); \ 84 BUG(); \ 85 panic("DLM: Record message above and reboot.\n"); \ 86 } \ 87} 88 89 90struct dlm_direntry { 91 struct list_head list; 92 uint32_t master_nodeid; 93 uint16_t length; 94 char name[1]; 95}; 96 97struct dlm_dirtable { 98 struct list_head list; 99 spinlock_t lock; 100}; 101 102struct dlm_rsbtable { 103 struct list_head list; 104 struct list_head toss; 105 spinlock_t lock; 106}; 107 108struct dlm_lkbtable { 109 struct list_head list; 110 rwlock_t lock; 111 uint16_t counter; 112}; 113 114/* 115 * Lockspace member (per node in a ls) 116 */ 117 118struct dlm_member { 119 struct list_head list; 120 int nodeid; 121 int weight; 122}; 123 124/* 125 * Save and manage recovery state for a lockspace. 126 */ 127 128struct dlm_recover { 129 struct list_head list; 130 int *nodeids; /* nodeids of all members */ 131 int node_count; 132 int *new; /* nodeids of new members */ 133 int new_count; 134 uint64_t seq; 135}; 136 137/* 138 * Pass input args to second stage locking function. 139 */ 140 141struct dlm_args { 142 uint32_t flags; 143 void (*astfn) (void *astparam); 144 void *astparam; 145 void (*bastfn) (void *astparam, int mode); 146 int mode; 147 struct dlm_lksb *lksb; 148 unsigned long timeout; 149}; 150 151 152/* 153 * Lock block 154 * 155 * A lock can be one of three types: 156 * 157 * local copy lock is mastered locally 158 * (lkb_nodeid is zero and DLM_LKF_MSTCPY is not set) 159 * process copy lock is mastered on a remote node 160 * (lkb_nodeid is non-zero and DLM_LKF_MSTCPY is not set) 161 * master copy master node's copy of a lock owned by remote node 162 * (lkb_nodeid is non-zero and DLM_LKF_MSTCPY is set) 163 * 164 * lkb_exflags: a copy of the most recent flags arg provided to dlm_lock or 165 * dlm_unlock. The dlm does not modify these or use any private flags in 166 * this field; it only contains DLM_LKF_ flags from dlm.h. These flags 167 * are sent as-is to the remote master when the lock is remote. 168 * 169 * lkb_flags: internal dlm flags (DLM_IFL_ prefix) from dlm_internal.h. 170 * Some internal flags are shared between the master and process nodes; 171 * these shared flags are kept in the lower two bytes. One of these 172 * flags set on the master copy will be propagated to the process copy 173 * and v.v. Other internal flags are private to the master or process 174 * node (e.g. DLM_IFL_MSTCPY). These are kept in the high two bytes. 175 * 176 * lkb_sbflags: status block flags. These flags are copied directly into 177 * the caller's lksb.sb_flags prior to the dlm_lock/dlm_unlock completion 178 * ast. All defined in dlm.h with DLM_SBF_ prefix. 179 * 180 * lkb_status: the lock status indicates which rsb queue the lock is 181 * on, grant, convert, or wait. DLM_LKSTS_ WAITING/GRANTED/CONVERT 182 * 183 * lkb_wait_type: the dlm message type (DLM_MSG_ prefix) for which a 184 * reply is needed. Only set when the lkb is on the lockspace waiters 185 * list awaiting a reply from a remote node. 186 * 187 * lkb_nodeid: when the lkb is a local copy, nodeid is 0; when the lkb 188 * is a master copy, nodeid specifies the remote lock holder, when the 189 * lkb is a process copy, the nodeid specifies the lock master. 190 */ 191 192/* lkb_ast_type */ 193 194#define AST_COMP 1 195#define AST_BAST 2 196 197/* lkb_status */ 198 199#define DLM_LKSTS_WAITING 1 200#define DLM_LKSTS_GRANTED 2 201#define DLM_LKSTS_CONVERT 3 202 203/* lkb_flags */ 204 205#define DLM_IFL_MSTCPY 0x00010000 206#define DLM_IFL_RESEND 0x00020000 207#define DLM_IFL_DEAD 0x00040000 208#define DLM_IFL_OVERLAP_UNLOCK 0x00080000 209#define DLM_IFL_OVERLAP_CANCEL 0x00100000 210#define DLM_IFL_ENDOFLIFE 0x00200000 211#define DLM_IFL_WATCH_TIMEWARN 0x00400000 212#define DLM_IFL_TIMEOUT_CANCEL 0x00800000 213#define DLM_IFL_DEADLOCK_CANCEL 0x01000000 214#define DLM_IFL_USER 0x00000001 215#define DLM_IFL_ORPHAN 0x00000002 216 217struct dlm_lkb { 218 struct dlm_rsb *lkb_resource; /* the rsb */ 219 struct kref lkb_ref; 220 int lkb_nodeid; /* copied from rsb */ 221 int lkb_ownpid; /* pid of lock owner */ 222 uint32_t lkb_id; /* our lock ID */ 223 uint32_t lkb_remid; /* lock ID on remote partner */ 224 uint32_t lkb_exflags; /* external flags from caller */ 225 uint32_t lkb_sbflags; /* lksb flags */ 226 uint32_t lkb_flags; /* internal flags */ 227 uint32_t lkb_lvbseq; /* lvb sequence number */ 228 229 int8_t lkb_status; /* granted, waiting, convert */ 230 int8_t lkb_rqmode; /* requested lock mode */ 231 int8_t lkb_grmode; /* granted lock mode */ 232 int8_t lkb_highbast; /* highest mode bast sent for */ 233 234 int8_t lkb_wait_type; /* type of reply waiting for */ 235 int8_t lkb_wait_count; 236 int8_t lkb_ast_type; /* type of ast queued for */ 237 int8_t lkb_ast_first; /* type of first ast queued */ 238 239 int8_t lkb_bastmode; /* req mode of queued bast */ 240 int8_t lkb_castmode; /* gr mode of queued cast */ 241 int8_t lkb_bastmode_done; /* last delivered bastmode */ 242 int8_t lkb_castmode_done; /* last delivered castmode */ 243 244 struct list_head lkb_idtbl_list; /* lockspace lkbtbl */ 245 struct list_head lkb_statequeue; /* rsb g/c/w list */ 246 struct list_head lkb_rsb_lookup; /* waiting for rsb lookup */ 247 struct list_head lkb_wait_reply; /* waiting for remote reply */ 248 struct list_head lkb_astqueue; /* need ast to be sent */ 249 struct list_head lkb_ownqueue; /* list of locks for a process */ 250 struct list_head lkb_time_list; 251 ktime_t lkb_time_bast; /* for debugging */ 252 ktime_t lkb_timestamp; 253 unsigned long lkb_timeout_cs; 254 255 char *lkb_lvbptr; 256 struct dlm_lksb *lkb_lksb; /* caller's status block */ 257 void (*lkb_astfn) (void *astparam); 258 void (*lkb_bastfn) (void *astparam, int mode); 259 union { 260 void *lkb_astparam; /* caller's ast arg */ 261 struct dlm_user_args *lkb_ua; 262 }; 263}; 264 265 266struct dlm_rsb { 267 struct dlm_ls *res_ls; /* the lockspace */ 268 struct kref res_ref; 269 struct mutex res_mutex; 270 unsigned long res_flags; 271 int res_length; /* length of rsb name */ 272 int res_nodeid; 273 uint32_t res_lvbseq; 274 uint32_t res_hash; 275 uint32_t res_bucket; /* rsbtbl */ 276 unsigned long res_toss_time; 277 uint32_t res_first_lkid; 278 struct list_head res_lookup; /* lkbs waiting on first */ 279 struct list_head res_hashchain; /* rsbtbl */ 280 struct list_head res_grantqueue; 281 struct list_head res_convertqueue; 282 struct list_head res_waitqueue; 283 284 struct list_head res_root_list; /* used for recovery */ 285 struct list_head res_recover_list; /* used for recovery */ 286 int res_recover_locks_count; 287 288 char *res_lvbptr; 289 char res_name[1]; 290}; 291 292/* find_rsb() flags */ 293 294#define R_MASTER 1 /* only return rsb if it's a master */ 295#define R_CREATE 2 /* create/add rsb if not found */ 296 297/* rsb_flags */ 298 299enum rsb_flags { 300 RSB_MASTER_UNCERTAIN, 301 RSB_VALNOTVALID, 302 RSB_VALNOTVALID_PREV, 303 RSB_NEW_MASTER, 304 RSB_NEW_MASTER2, 305 RSB_RECOVER_CONVERT, 306 RSB_LOCKS_PURGED, 307}; 308 309static inline void rsb_set_flag(struct dlm_rsb *r, enum rsb_flags flag) 310{ 311 __set_bit(flag, &r->res_flags); 312} 313 314static inline void rsb_clear_flag(struct dlm_rsb *r, enum rsb_flags flag) 315{ 316 __clear_bit(flag, &r->res_flags); 317} 318 319static inline int rsb_flag(struct dlm_rsb *r, enum rsb_flags flag) 320{ 321 return test_bit(flag, &r->res_flags); 322} 323 324 325/* dlm_header is first element of all structs sent between nodes */ 326 327#define DLM_HEADER_MAJOR 0x00030000 328#define DLM_HEADER_MINOR 0x00000000 329 330#define DLM_MSG 1 331#define DLM_RCOM 2 332 333struct dlm_header { 334 uint32_t h_version; 335 uint32_t h_lockspace; 336 uint32_t h_nodeid; /* nodeid of sender */ 337 uint16_t h_length; 338 uint8_t h_cmd; /* DLM_MSG, DLM_RCOM */ 339 uint8_t h_pad; 340}; 341 342 343#define DLM_MSG_REQUEST 1 344#define DLM_MSG_CONVERT 2 345#define DLM_MSG_UNLOCK 3 346#define DLM_MSG_CANCEL 4 347#define DLM_MSG_REQUEST_REPLY 5 348#define DLM_MSG_CONVERT_REPLY 6 349#define DLM_MSG_UNLOCK_REPLY 7 350#define DLM_MSG_CANCEL_REPLY 8 351#define DLM_MSG_GRANT 9 352#define DLM_MSG_BAST 10 353#define DLM_MSG_LOOKUP 11 354#define DLM_MSG_REMOVE 12 355#define DLM_MSG_LOOKUP_REPLY 13 356#define DLM_MSG_PURGE 14 357 358struct dlm_message { 359 struct dlm_header m_header; 360 uint32_t m_type; /* DLM_MSG_ */ 361 uint32_t m_nodeid; 362 uint32_t m_pid; 363 uint32_t m_lkid; /* lkid on sender */ 364 uint32_t m_remid; /* lkid on receiver */ 365 uint32_t m_parent_lkid; 366 uint32_t m_parent_remid; 367 uint32_t m_exflags; 368 uint32_t m_sbflags; 369 uint32_t m_flags; 370 uint32_t m_lvbseq; 371 uint32_t m_hash; 372 int m_status; 373 int m_grmode; 374 int m_rqmode; 375 int m_bastmode; 376 int m_asts; 377 int m_result; /* 0 or -EXXX */ 378 char m_extra[0]; /* name or lvb */ 379}; 380 381 382#define DLM_RS_NODES 0x00000001 383#define DLM_RS_NODES_ALL 0x00000002 384#define DLM_RS_DIR 0x00000004 385#define DLM_RS_DIR_ALL 0x00000008 386#define DLM_RS_LOCKS 0x00000010 387#define DLM_RS_LOCKS_ALL 0x00000020 388#define DLM_RS_DONE 0x00000040 389#define DLM_RS_DONE_ALL 0x00000080 390 391#define DLM_RCOM_STATUS 1 392#define DLM_RCOM_NAMES 2 393#define DLM_RCOM_LOOKUP 3 394#define DLM_RCOM_LOCK 4 395#define DLM_RCOM_STATUS_REPLY 5 396#define DLM_RCOM_NAMES_REPLY 6 397#define DLM_RCOM_LOOKUP_REPLY 7 398#define DLM_RCOM_LOCK_REPLY 8 399 400struct dlm_rcom { 401 struct dlm_header rc_header; 402 uint32_t rc_type; /* DLM_RCOM_ */ 403 int rc_result; /* multi-purpose */ 404 uint64_t rc_id; /* match reply with request */ 405 uint64_t rc_seq; /* sender's ls_recover_seq */ 406 uint64_t rc_seq_reply; /* remote ls_recover_seq */ 407 char rc_buf[0]; 408}; 409 410union dlm_packet { 411 struct dlm_header header; /* common to other two */ 412 struct dlm_message message; 413 struct dlm_rcom rcom; 414}; 415 416struct rcom_config { 417 __le32 rf_lvblen; 418 __le32 rf_lsflags; 419 __le64 rf_unused; 420}; 421 422struct rcom_lock { 423 __le32 rl_ownpid; 424 __le32 rl_lkid; 425 __le32 rl_remid; 426 __le32 rl_parent_lkid; 427 __le32 rl_parent_remid; 428 __le32 rl_exflags; 429 __le32 rl_flags; 430 __le32 rl_lvbseq; 431 __le32 rl_result; 432 int8_t rl_rqmode; 433 int8_t rl_grmode; 434 int8_t rl_status; 435 int8_t rl_asts; 436 __le16 rl_wait_type; 437 __le16 rl_namelen; 438 char rl_name[DLM_RESNAME_MAXLEN]; 439 char rl_lvb[0]; 440}; 441 442struct dlm_ls { 443 struct list_head ls_list; /* list of lockspaces */ 444 dlm_lockspace_t *ls_local_handle; 445 uint32_t ls_global_id; /* global unique lockspace ID */ 446 uint32_t ls_exflags; 447 int ls_lvblen; 448 int ls_count; /* refcount of processes in 449 the dlm using this ls */ 450 int ls_create_count; /* create/release refcount */ 451 unsigned long ls_flags; /* LSFL_ */ 452 unsigned long ls_scan_time; 453 struct kobject ls_kobj; 454 455 struct dlm_rsbtable *ls_rsbtbl; 456 uint32_t ls_rsbtbl_size; 457 458 struct dlm_lkbtable *ls_lkbtbl; 459 uint32_t ls_lkbtbl_size; 460 461 struct dlm_dirtable *ls_dirtbl; 462 uint32_t ls_dirtbl_size; 463 464 struct mutex ls_waiters_mutex; 465 struct list_head ls_waiters; /* lkbs needing a reply */ 466 467 struct mutex ls_orphans_mutex; 468 struct list_head ls_orphans; 469 470 struct mutex ls_timeout_mutex; 471 struct list_head ls_timeout; 472 473 struct list_head ls_nodes; /* current nodes in ls */ 474 struct list_head ls_nodes_gone; /* dead node list, recovery */ 475 int ls_num_nodes; /* number of nodes in ls */ 476 int ls_low_nodeid; 477 int ls_total_weight; 478 int *ls_node_array; 479 480 struct dlm_rsb ls_stub_rsb; /* for returning errors */ 481 struct dlm_lkb ls_stub_lkb; /* for returning errors */ 482 struct dlm_message ls_stub_ms; /* for faking a reply */ 483 484 struct dentry *ls_debug_rsb_dentry; /* debugfs */ 485 struct dentry *ls_debug_waiters_dentry; /* debugfs */ 486 struct dentry *ls_debug_locks_dentry; /* debugfs */ 487 struct dentry *ls_debug_all_dentry; /* debugfs */ 488 489 wait_queue_head_t ls_uevent_wait; /* user part of join/leave */ 490 int ls_uevent_result; 491 struct completion ls_members_done; 492 int ls_members_result; 493 494 struct miscdevice ls_device; 495 496 /* recovery related */ 497 498 struct timer_list ls_timer; 499 struct task_struct *ls_recoverd_task; 500 struct mutex ls_recoverd_active; 501 spinlock_t ls_recover_lock; 502 unsigned long ls_recover_begin; /* jiffies timestamp */ 503 uint32_t ls_recover_status; /* DLM_RS_ */ 504 uint64_t ls_recover_seq; 505 struct dlm_recover *ls_recover_args; 506 struct rw_semaphore ls_in_recovery; /* block local requests */ 507 struct rw_semaphore ls_recv_active; /* block dlm_recv */ 508 struct list_head ls_requestqueue;/* queue remote requests */ 509 struct mutex ls_requestqueue_mutex; 510 struct dlm_rcom *ls_recover_buf; 511 int ls_recover_nodeid; /* for debugging */ 512 uint64_t ls_rcom_seq; 513 spinlock_t ls_rcom_spin; 514 struct list_head ls_recover_list; 515 spinlock_t ls_recover_list_lock; 516 int ls_recover_list_count; 517 wait_queue_head_t ls_wait_general; 518 struct mutex ls_clear_proc_locks; 519 520 struct list_head ls_root_list; /* root resources */ 521 struct rw_semaphore ls_root_sem; /* protect root_list */ 522 523 int ls_namelen; 524 char ls_name[1]; 525}; 526 527#define LSFL_WORK 0 528#define LSFL_RUNNING 1 529#define LSFL_RECOVERY_STOP 2 530#define LSFL_RCOM_READY 3 531#define LSFL_RCOM_WAIT 4 532#define LSFL_UEVENT_WAIT 5 533#define LSFL_TIMEWARN 6 534 535/* much of this is just saving user space pointers associated with the 536 lock that we pass back to the user lib with an ast */ 537 538struct dlm_user_args { 539 struct dlm_user_proc *proc; /* each process that opens the lockspace 540 device has private data 541 (dlm_user_proc) on the struct file, 542 the process's locks point back to it*/ 543 struct dlm_lksb lksb; 544 int old_mode; 545 int update_user_lvb; 546 struct dlm_lksb __user *user_lksb; 547 void __user *castparam; 548 void __user *castaddr; 549 void __user *bastparam; 550 void __user *bastaddr; 551 uint64_t xid; 552}; 553 554#define DLM_PROC_FLAGS_CLOSING 1 555#define DLM_PROC_FLAGS_COMPAT 2 556 557/* locks list is kept so we can remove all a process's locks when it 558 exits (or orphan those that are persistent) */ 559 560struct dlm_user_proc { 561 dlm_lockspace_t *lockspace; 562 unsigned long flags; /* DLM_PROC_FLAGS */ 563 struct list_head asts; 564 spinlock_t asts_spin; 565 struct list_head locks; 566 spinlock_t locks_spin; 567 struct list_head unlocking; 568 wait_queue_head_t wait; 569}; 570 571static inline int dlm_locking_stopped(struct dlm_ls *ls) 572{ 573 return !test_bit(LSFL_RUNNING, &ls->ls_flags); 574} 575 576static inline int dlm_recovery_stopped(struct dlm_ls *ls) 577{ 578 return test_bit(LSFL_RECOVERY_STOP, &ls->ls_flags); 579} 580 581static inline int dlm_no_directory(struct dlm_ls *ls) 582{ 583 return (ls->ls_exflags & DLM_LSFL_NODIR) ? 1 : 0; 584} 585 586int dlm_netlink_init(void); 587void dlm_netlink_exit(void); 588void dlm_timeout_warn(struct dlm_lkb *lkb); 589int dlm_plock_init(void); 590void dlm_plock_exit(void); 591 592#ifdef CONFIG_DLM_DEBUG 593int dlm_register_debugfs(void); 594void dlm_unregister_debugfs(void); 595int dlm_create_debug_file(struct dlm_ls *ls); 596void dlm_delete_debug_file(struct dlm_ls *ls); 597#else 598static inline int dlm_register_debugfs(void) { return 0; } 599static inline void dlm_unregister_debugfs(void) { } 600static inline int dlm_create_debug_file(struct dlm_ls *ls) { return 0; } 601static inline void dlm_delete_debug_file(struct dlm_ls *ls) { } 602#endif 603 604#endif /* __DLM_INTERNAL_DOT_H__ */ 605