1 /****************************************************************************** 2 ******************************************************************************* 3 ** 4 ** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. 5 ** Copyright (C) 2004-2007 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/semaphore.h> 41 #include <asm/uaccess.h> 42 43 #include <linux/dlm.h> 44 #include "config.h" 45 46 #define DLM_LOCKSPACE_LEN 64 47 48 /* Size of the temp buffer midcomms allocates on the stack. 49 We try to make this large enough so most messages fit. 50 FIXME: should sctp make this unnecessary? */ 51 52 #define DLM_INBUF_LEN 148 53 54 struct dlm_ls; 55 struct dlm_lkb; 56 struct dlm_rsb; 57 struct dlm_member; 58 struct dlm_lkbtable; 59 struct dlm_rsbtable; 60 struct dlm_dirtable; 61 struct dlm_direntry; 62 struct dlm_recover; 63 struct dlm_header; 64 struct dlm_message; 65 struct dlm_rcom; 66 struct dlm_mhandle; 67 68 #define log_print(fmt, args...) \ 69 printk(KERN_ERR "dlm: "fmt"\n" , ##args) 70 #define log_error(ls, fmt, args...) \ 71 printk(KERN_ERR "dlm: %s: " fmt "\n", (ls)->ls_name , ##args) 72 73 #define log_debug(ls, fmt, args...) \ 74 do { \ 75 if (dlm_config.ci_log_debug) \ 76 printk(KERN_DEBUG "dlm: %s: " fmt "\n", \ 77 (ls)->ls_name , ##args); \ 78 } while (0) 79 80 #define DLM_ASSERT(x, do) \ 81 { \ 82 if (!(x)) \ 83 { \ 84 printk(KERN_ERR "\nDLM: Assertion failed on line %d of file %s\n" \ 85 "DLM: assertion: \"%s\"\n" \ 86 "DLM: time = %lu\n", \ 87 __LINE__, __FILE__, #x, jiffies); \ 88 {do} \ 89 printk("\n"); \ 90 BUG(); \ 91 panic("DLM: Record message above and reboot.\n"); \ 92 } \ 93 } 94 95 96 struct dlm_direntry { 97 struct list_head list; 98 uint32_t master_nodeid; 99 uint16_t length; 100 char name[1]; 101 }; 102 103 struct dlm_dirtable { 104 struct list_head list; 105 rwlock_t lock; 106 }; 107 108 struct dlm_rsbtable { 109 struct list_head list; 110 struct list_head toss; 111 rwlock_t lock; 112 }; 113 114 struct dlm_lkbtable { 115 struct list_head list; 116 rwlock_t lock; 117 uint16_t counter; 118 }; 119 120 /* 121 * Lockspace member (per node in a ls) 122 */ 123 124 struct dlm_member { 125 struct list_head list; 126 int nodeid; 127 int weight; 128 }; 129 130 /* 131 * Save and manage recovery state for a lockspace. 132 */ 133 134 struct dlm_recover { 135 struct list_head list; 136 int *nodeids; 137 int node_count; 138 uint64_t seq; 139 }; 140 141 /* 142 * Pass input args to second stage locking function. 143 */ 144 145 struct dlm_args { 146 uint32_t flags; 147 void (*astfn) (void *astparam); 148 void *astparam; 149 void (*bastfn) (void *astparam, int mode); 150 int mode; 151 struct dlm_lksb *lksb; 152 unsigned long timeout; 153 }; 154 155 156 /* 157 * Lock block 158 * 159 * A lock can be one of three types: 160 * 161 * local copy lock is mastered locally 162 * (lkb_nodeid is zero and DLM_LKF_MSTCPY is not set) 163 * process copy lock is mastered on a remote node 164 * (lkb_nodeid is non-zero and DLM_LKF_MSTCPY is not set) 165 * master copy master node's copy of a lock owned by remote node 166 * (lkb_nodeid is non-zero and DLM_LKF_MSTCPY is set) 167 * 168 * lkb_exflags: a copy of the most recent flags arg provided to dlm_lock or 169 * dlm_unlock. The dlm does not modify these or use any private flags in 170 * this field; it only contains DLM_LKF_ flags from dlm.h. These flags 171 * are sent as-is to the remote master when the lock is remote. 172 * 173 * lkb_flags: internal dlm flags (DLM_IFL_ prefix) from dlm_internal.h. 174 * Some internal flags are shared between the master and process nodes; 175 * these shared flags are kept in the lower two bytes. One of these 176 * flags set on the master copy will be propagated to the process copy 177 * and v.v. Other internal flags are private to the master or process 178 * node (e.g. DLM_IFL_MSTCPY). These are kept in the high two bytes. 179 * 180 * lkb_sbflags: status block flags. These flags are copied directly into 181 * the caller's lksb.sb_flags prior to the dlm_lock/dlm_unlock completion 182 * ast. All defined in dlm.h with DLM_SBF_ prefix. 183 * 184 * lkb_status: the lock status indicates which rsb queue the lock is 185 * on, grant, convert, or wait. DLM_LKSTS_ WAITING/GRANTED/CONVERT 186 * 187 * lkb_wait_type: the dlm message type (DLM_MSG_ prefix) for which a 188 * reply is needed. Only set when the lkb is on the lockspace waiters 189 * list awaiting a reply from a remote node. 190 * 191 * lkb_nodeid: when the lkb is a local copy, nodeid is 0; when the lkb 192 * is a master copy, nodeid specifies the remote lock holder, when the 193 * lkb is a process copy, the nodeid specifies the lock master. 194 */ 195 196 /* lkb_ast_type */ 197 198 #define AST_COMP 1 199 #define AST_BAST 2 200 201 /* lkb_status */ 202 203 #define DLM_LKSTS_WAITING 1 204 #define DLM_LKSTS_GRANTED 2 205 #define DLM_LKSTS_CONVERT 3 206 207 /* lkb_flags */ 208 209 #define DLM_IFL_MSTCPY 0x00010000 210 #define DLM_IFL_RESEND 0x00020000 211 #define DLM_IFL_DEAD 0x00040000 212 #define DLM_IFL_OVERLAP_UNLOCK 0x00080000 213 #define DLM_IFL_OVERLAP_CANCEL 0x00100000 214 #define DLM_IFL_ENDOFLIFE 0x00200000 215 #define DLM_IFL_WATCH_TIMEWARN 0x00400000 216 #define DLM_IFL_TIMEOUT_CANCEL 0x00800000 217 #define DLM_IFL_DEADLOCK_CANCEL 0x01000000 218 #define DLM_IFL_USER 0x00000001 219 #define DLM_IFL_ORPHAN 0x00000002 220 221 struct dlm_lkb { 222 struct dlm_rsb *lkb_resource; /* the rsb */ 223 struct kref lkb_ref; 224 int lkb_nodeid; /* copied from rsb */ 225 int lkb_ownpid; /* pid of lock owner */ 226 uint32_t lkb_id; /* our lock ID */ 227 uint32_t lkb_remid; /* lock ID on remote partner */ 228 uint32_t lkb_exflags; /* external flags from caller */ 229 uint32_t lkb_sbflags; /* lksb flags */ 230 uint32_t lkb_flags; /* internal flags */ 231 uint32_t lkb_lvbseq; /* lvb sequence number */ 232 233 int8_t lkb_status; /* granted, waiting, convert */ 234 int8_t lkb_rqmode; /* requested lock mode */ 235 int8_t lkb_grmode; /* granted lock mode */ 236 int8_t lkb_bastmode; /* requested mode */ 237 int8_t lkb_highbast; /* highest mode bast sent for */ 238 int8_t lkb_wait_type; /* type of reply waiting for */ 239 int8_t lkb_wait_count; 240 int8_t lkb_ast_type; /* type of ast queued for */ 241 242 struct list_head lkb_idtbl_list; /* lockspace lkbtbl */ 243 struct list_head lkb_statequeue; /* rsb g/c/w list */ 244 struct list_head lkb_rsb_lookup; /* waiting for rsb lookup */ 245 struct list_head lkb_wait_reply; /* waiting for remote reply */ 246 struct list_head lkb_astqueue; /* need ast to be sent */ 247 struct list_head lkb_ownqueue; /* list of locks for a process */ 248 struct list_head lkb_time_list; 249 unsigned long lkb_timestamp; 250 unsigned long lkb_timeout_cs; 251 252 char *lkb_lvbptr; 253 struct dlm_lksb *lkb_lksb; /* caller's status block */ 254 void (*lkb_astfn) (void *astparam); 255 void (*lkb_bastfn) (void *astparam, int mode); 256 union { 257 void *lkb_astparam; /* caller's ast arg */ 258 struct dlm_user_args *lkb_ua; 259 }; 260 }; 261 262 263 struct dlm_rsb { 264 struct dlm_ls *res_ls; /* the lockspace */ 265 struct kref res_ref; 266 struct mutex res_mutex; 267 unsigned long res_flags; 268 int res_length; /* length of rsb name */ 269 int res_nodeid; 270 uint32_t res_lvbseq; 271 uint32_t res_hash; 272 uint32_t res_bucket; /* rsbtbl */ 273 unsigned long res_toss_time; 274 uint32_t res_first_lkid; 275 struct list_head res_lookup; /* lkbs waiting on first */ 276 struct list_head res_hashchain; /* rsbtbl */ 277 struct list_head res_grantqueue; 278 struct list_head res_convertqueue; 279 struct list_head res_waitqueue; 280 281 struct list_head res_root_list; /* used for recovery */ 282 struct list_head res_recover_list; /* used for recovery */ 283 int res_recover_locks_count; 284 285 char *res_lvbptr; 286 char res_name[1]; 287 }; 288 289 /* find_rsb() flags */ 290 291 #define R_MASTER 1 /* only return rsb if it's a master */ 292 #define R_CREATE 2 /* create/add rsb if not found */ 293 294 /* rsb_flags */ 295 296 enum rsb_flags { 297 RSB_MASTER_UNCERTAIN, 298 RSB_VALNOTVALID, 299 RSB_VALNOTVALID_PREV, 300 RSB_NEW_MASTER, 301 RSB_NEW_MASTER2, 302 RSB_RECOVER_CONVERT, 303 RSB_LOCKS_PURGED, 304 }; 305 306 static inline void rsb_set_flag(struct dlm_rsb *r, enum rsb_flags flag) 307 { 308 __set_bit(flag, &r->res_flags); 309 } 310 311 static inline void rsb_clear_flag(struct dlm_rsb *r, enum rsb_flags flag) 312 { 313 __clear_bit(flag, &r->res_flags); 314 } 315 316 static inline int rsb_flag(struct dlm_rsb *r, enum rsb_flags flag) 317 { 318 return test_bit(flag, &r->res_flags); 319 } 320 321 322 /* dlm_header is first element of all structs sent between nodes */ 323 324 #define DLM_HEADER_MAJOR 0x00030000 325 #define DLM_HEADER_MINOR 0x00000000 326 327 #define DLM_MSG 1 328 #define DLM_RCOM 2 329 330 struct dlm_header { 331 uint32_t h_version; 332 uint32_t h_lockspace; 333 uint32_t h_nodeid; /* nodeid of sender */ 334 uint16_t h_length; 335 uint8_t h_cmd; /* DLM_MSG, DLM_RCOM */ 336 uint8_t h_pad; 337 }; 338 339 340 #define DLM_MSG_REQUEST 1 341 #define DLM_MSG_CONVERT 2 342 #define DLM_MSG_UNLOCK 3 343 #define DLM_MSG_CANCEL 4 344 #define DLM_MSG_REQUEST_REPLY 5 345 #define DLM_MSG_CONVERT_REPLY 6 346 #define DLM_MSG_UNLOCK_REPLY 7 347 #define DLM_MSG_CANCEL_REPLY 8 348 #define DLM_MSG_GRANT 9 349 #define DLM_MSG_BAST 10 350 #define DLM_MSG_LOOKUP 11 351 #define DLM_MSG_REMOVE 12 352 #define DLM_MSG_LOOKUP_REPLY 13 353 #define DLM_MSG_PURGE 14 354 355 struct dlm_message { 356 struct dlm_header m_header; 357 uint32_t m_type; /* DLM_MSG_ */ 358 uint32_t m_nodeid; 359 uint32_t m_pid; 360 uint32_t m_lkid; /* lkid on sender */ 361 uint32_t m_remid; /* lkid on receiver */ 362 uint32_t m_parent_lkid; 363 uint32_t m_parent_remid; 364 uint32_t m_exflags; 365 uint32_t m_sbflags; 366 uint32_t m_flags; 367 uint32_t m_lvbseq; 368 uint32_t m_hash; 369 int m_status; 370 int m_grmode; 371 int m_rqmode; 372 int m_bastmode; 373 int m_asts; 374 int m_result; /* 0 or -EXXX */ 375 char m_extra[0]; /* name or lvb */ 376 }; 377 378 379 #define DLM_RS_NODES 0x00000001 380 #define DLM_RS_NODES_ALL 0x00000002 381 #define DLM_RS_DIR 0x00000004 382 #define DLM_RS_DIR_ALL 0x00000008 383 #define DLM_RS_LOCKS 0x00000010 384 #define DLM_RS_LOCKS_ALL 0x00000020 385 #define DLM_RS_DONE 0x00000040 386 #define DLM_RS_DONE_ALL 0x00000080 387 388 #define DLM_RCOM_STATUS 1 389 #define DLM_RCOM_NAMES 2 390 #define DLM_RCOM_LOOKUP 3 391 #define DLM_RCOM_LOCK 4 392 #define DLM_RCOM_STATUS_REPLY 5 393 #define DLM_RCOM_NAMES_REPLY 6 394 #define DLM_RCOM_LOOKUP_REPLY 7 395 #define DLM_RCOM_LOCK_REPLY 8 396 397 struct dlm_rcom { 398 struct dlm_header rc_header; 399 uint32_t rc_type; /* DLM_RCOM_ */ 400 int rc_result; /* multi-purpose */ 401 uint64_t rc_id; /* match reply with request */ 402 uint64_t rc_seq; /* sender's ls_recover_seq */ 403 uint64_t rc_seq_reply; /* remote ls_recover_seq */ 404 char rc_buf[0]; 405 }; 406 407 union dlm_packet { 408 struct dlm_header header; /* common to other two */ 409 struct dlm_message message; 410 struct dlm_rcom rcom; 411 }; 412 413 struct rcom_config { 414 __le32 rf_lvblen; 415 __le32 rf_lsflags; 416 __le64 rf_unused; 417 }; 418 419 struct rcom_lock { 420 __le32 rl_ownpid; 421 __le32 rl_lkid; 422 __le32 rl_remid; 423 __le32 rl_parent_lkid; 424 __le32 rl_parent_remid; 425 __le32 rl_exflags; 426 __le32 rl_flags; 427 __le32 rl_lvbseq; 428 __le32 rl_result; 429 int8_t rl_rqmode; 430 int8_t rl_grmode; 431 int8_t rl_status; 432 int8_t rl_asts; 433 __le16 rl_wait_type; 434 __le16 rl_namelen; 435 char rl_name[DLM_RESNAME_MAXLEN]; 436 char rl_lvb[0]; 437 }; 438 439 struct dlm_ls { 440 struct list_head ls_list; /* list of lockspaces */ 441 dlm_lockspace_t *ls_local_handle; 442 uint32_t ls_global_id; /* global unique lockspace ID */ 443 uint32_t ls_exflags; 444 int ls_lvblen; 445 int ls_count; /* reference count */ 446 unsigned long ls_flags; /* LSFL_ */ 447 struct kobject ls_kobj; 448 449 struct dlm_rsbtable *ls_rsbtbl; 450 uint32_t ls_rsbtbl_size; 451 452 struct dlm_lkbtable *ls_lkbtbl; 453 uint32_t ls_lkbtbl_size; 454 455 struct dlm_dirtable *ls_dirtbl; 456 uint32_t ls_dirtbl_size; 457 458 struct mutex ls_waiters_mutex; 459 struct list_head ls_waiters; /* lkbs needing a reply */ 460 461 struct mutex ls_orphans_mutex; 462 struct list_head ls_orphans; 463 464 struct mutex ls_timeout_mutex; 465 struct list_head ls_timeout; 466 467 struct list_head ls_nodes; /* current nodes in ls */ 468 struct list_head ls_nodes_gone; /* dead node list, recovery */ 469 int ls_num_nodes; /* number of nodes in ls */ 470 int ls_low_nodeid; 471 int ls_total_weight; 472 int *ls_node_array; 473 gfp_t ls_allocation; 474 475 struct dlm_rsb ls_stub_rsb; /* for returning errors */ 476 struct dlm_lkb ls_stub_lkb; /* for returning errors */ 477 struct dlm_message ls_stub_ms; /* for faking a reply */ 478 479 struct dentry *ls_debug_rsb_dentry; /* debugfs */ 480 struct dentry *ls_debug_waiters_dentry; /* debugfs */ 481 struct dentry *ls_debug_locks_dentry; /* debugfs */ 482 483 wait_queue_head_t ls_uevent_wait; /* user part of join/leave */ 484 int ls_uevent_result; 485 struct completion ls_members_done; 486 int ls_members_result; 487 488 struct miscdevice ls_device; 489 490 /* recovery related */ 491 492 struct timer_list ls_timer; 493 struct task_struct *ls_recoverd_task; 494 struct mutex ls_recoverd_active; 495 spinlock_t ls_recover_lock; 496 unsigned long ls_recover_begin; /* jiffies timestamp */ 497 uint32_t ls_recover_status; /* DLM_RS_ */ 498 uint64_t ls_recover_seq; 499 struct dlm_recover *ls_recover_args; 500 struct rw_semaphore ls_in_recovery; /* block local requests */ 501 struct rw_semaphore ls_recv_active; /* block dlm_recv */ 502 struct list_head ls_requestqueue;/* queue remote requests */ 503 struct mutex ls_requestqueue_mutex; 504 struct dlm_rcom *ls_recover_buf; 505 int ls_recover_nodeid; /* for debugging */ 506 uint64_t ls_rcom_seq; 507 spinlock_t ls_rcom_spin; 508 struct list_head ls_recover_list; 509 spinlock_t ls_recover_list_lock; 510 int ls_recover_list_count; 511 wait_queue_head_t ls_wait_general; 512 struct mutex ls_clear_proc_locks; 513 514 struct list_head ls_root_list; /* root resources */ 515 struct rw_semaphore ls_root_sem; /* protect root_list */ 516 517 int ls_namelen; 518 char ls_name[1]; 519 }; 520 521 #define LSFL_WORK 0 522 #define LSFL_RUNNING 1 523 #define LSFL_RECOVERY_STOP 2 524 #define LSFL_RCOM_READY 3 525 #define LSFL_RCOM_WAIT 4 526 #define LSFL_UEVENT_WAIT 5 527 #define LSFL_TIMEWARN 6 528 529 /* much of this is just saving user space pointers associated with the 530 lock that we pass back to the user lib with an ast */ 531 532 struct dlm_user_args { 533 struct dlm_user_proc *proc; /* each process that opens the lockspace 534 device has private data 535 (dlm_user_proc) on the struct file, 536 the process's locks point back to it*/ 537 struct dlm_lksb lksb; 538 int old_mode; 539 int update_user_lvb; 540 struct dlm_lksb __user *user_lksb; 541 void __user *castparam; 542 void __user *castaddr; 543 void __user *bastparam; 544 void __user *bastaddr; 545 uint64_t xid; 546 }; 547 548 #define DLM_PROC_FLAGS_CLOSING 1 549 #define DLM_PROC_FLAGS_COMPAT 2 550 551 /* locks list is kept so we can remove all a process's locks when it 552 exits (or orphan those that are persistent) */ 553 554 struct dlm_user_proc { 555 dlm_lockspace_t *lockspace; 556 unsigned long flags; /* DLM_PROC_FLAGS */ 557 struct list_head asts; 558 spinlock_t asts_spin; 559 struct list_head locks; 560 spinlock_t locks_spin; 561 struct list_head unlocking; 562 wait_queue_head_t wait; 563 }; 564 565 static inline int dlm_locking_stopped(struct dlm_ls *ls) 566 { 567 return !test_bit(LSFL_RUNNING, &ls->ls_flags); 568 } 569 570 static inline int dlm_recovery_stopped(struct dlm_ls *ls) 571 { 572 return test_bit(LSFL_RECOVERY_STOP, &ls->ls_flags); 573 } 574 575 static inline int dlm_no_directory(struct dlm_ls *ls) 576 { 577 return (ls->ls_exflags & DLM_LSFL_NODIR) ? 1 : 0; 578 } 579 580 int dlm_netlink_init(void); 581 void dlm_netlink_exit(void); 582 void dlm_timeout_warn(struct dlm_lkb *lkb); 583 584 #ifdef CONFIG_DLM_DEBUG 585 int dlm_register_debugfs(void); 586 void dlm_unregister_debugfs(void); 587 int dlm_create_debug_file(struct dlm_ls *ls); 588 void dlm_delete_debug_file(struct dlm_ls *ls); 589 #else 590 static inline int dlm_register_debugfs(void) { return 0; } 591 static inline void dlm_unregister_debugfs(void) { } 592 static inline int dlm_create_debug_file(struct dlm_ls *ls) { return 0; } 593 static inline void dlm_delete_debug_file(struct dlm_ls *ls) { } 594 #endif 595 596 #endif /* __DLM_INTERNAL_DOT_H__ */ 597 598