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 /* Size of the temp buffer midcomms allocates on the stack. 46 We try to make this large enough so most messages fit. 47 FIXME: should sctp make this unnecessary? */ 48 49 #define DLM_INBUF_LEN 148 50 51 struct dlm_ls; 52 struct dlm_lkb; 53 struct dlm_rsb; 54 struct dlm_member; 55 struct dlm_lkbtable; 56 struct dlm_rsbtable; 57 struct dlm_dirtable; 58 struct dlm_direntry; 59 struct dlm_recover; 60 struct dlm_header; 61 struct dlm_message; 62 struct dlm_rcom; 63 struct dlm_mhandle; 64 65 #define log_print(fmt, args...) \ 66 printk(KERN_ERR "dlm: "fmt"\n" , ##args) 67 #define log_error(ls, fmt, args...) \ 68 printk(KERN_ERR "dlm: %s: " fmt "\n", (ls)->ls_name , ##args) 69 70 #define log_debug(ls, fmt, args...) \ 71 do { \ 72 if (dlm_config.ci_log_debug) \ 73 printk(KERN_DEBUG "dlm: %s: " fmt "\n", \ 74 (ls)->ls_name , ##args); \ 75 } while (0) 76 77 #define DLM_ASSERT(x, do) \ 78 { \ 79 if (!(x)) \ 80 { \ 81 printk(KERN_ERR "\nDLM: Assertion failed on line %d of file %s\n" \ 82 "DLM: assertion: \"%s\"\n" \ 83 "DLM: time = %lu\n", \ 84 __LINE__, __FILE__, #x, jiffies); \ 85 {do} \ 86 printk("\n"); \ 87 BUG(); \ 88 panic("DLM: Record message above and reboot.\n"); \ 89 } \ 90 } 91 92 93 struct dlm_direntry { 94 struct list_head list; 95 uint32_t master_nodeid; 96 uint16_t length; 97 char name[1]; 98 }; 99 100 struct dlm_dirtable { 101 struct list_head list; 102 spinlock_t lock; 103 }; 104 105 struct dlm_rsbtable { 106 struct list_head list; 107 struct list_head toss; 108 spinlock_t lock; 109 }; 110 111 struct dlm_lkbtable { 112 struct list_head list; 113 rwlock_t lock; 114 uint16_t counter; 115 }; 116 117 /* 118 * Lockspace member (per node in a ls) 119 */ 120 121 struct dlm_member { 122 struct list_head list; 123 int nodeid; 124 int weight; 125 }; 126 127 /* 128 * Save and manage recovery state for a lockspace. 129 */ 130 131 struct dlm_recover { 132 struct list_head list; 133 int *nodeids; /* nodeids of all members */ 134 int node_count; 135 int *new; /* nodeids of new members */ 136 int new_count; 137 uint64_t seq; 138 }; 139 140 /* 141 * Pass input args to second stage locking function. 142 */ 143 144 struct dlm_args { 145 uint32_t flags; 146 void (*astfn) (void *astparam); 147 void *astparam; 148 void (*bastfn) (void *astparam, int mode); 149 int mode; 150 struct dlm_lksb *lksb; 151 unsigned long timeout; 152 }; 153 154 155 /* 156 * Lock block 157 * 158 * A lock can be one of three types: 159 * 160 * local copy lock is mastered locally 161 * (lkb_nodeid is zero and DLM_LKF_MSTCPY is not set) 162 * process copy lock is mastered on a remote node 163 * (lkb_nodeid is non-zero and DLM_LKF_MSTCPY is not set) 164 * master copy master node's copy of a lock owned by remote node 165 * (lkb_nodeid is non-zero and DLM_LKF_MSTCPY is set) 166 * 167 * lkb_exflags: a copy of the most recent flags arg provided to dlm_lock or 168 * dlm_unlock. The dlm does not modify these or use any private flags in 169 * this field; it only contains DLM_LKF_ flags from dlm.h. These flags 170 * are sent as-is to the remote master when the lock is remote. 171 * 172 * lkb_flags: internal dlm flags (DLM_IFL_ prefix) from dlm_internal.h. 173 * Some internal flags are shared between the master and process nodes; 174 * these shared flags are kept in the lower two bytes. One of these 175 * flags set on the master copy will be propagated to the process copy 176 * and v.v. Other internal flags are private to the master or process 177 * node (e.g. DLM_IFL_MSTCPY). These are kept in the high two bytes. 178 * 179 * lkb_sbflags: status block flags. These flags are copied directly into 180 * the caller's lksb.sb_flags prior to the dlm_lock/dlm_unlock completion 181 * ast. All defined in dlm.h with DLM_SBF_ prefix. 182 * 183 * lkb_status: the lock status indicates which rsb queue the lock is 184 * on, grant, convert, or wait. DLM_LKSTS_ WAITING/GRANTED/CONVERT 185 * 186 * lkb_wait_type: the dlm message type (DLM_MSG_ prefix) for which a 187 * reply is needed. Only set when the lkb is on the lockspace waiters 188 * list awaiting a reply from a remote node. 189 * 190 * lkb_nodeid: when the lkb is a local copy, nodeid is 0; when the lkb 191 * is a master copy, nodeid specifies the remote lock holder, when the 192 * lkb is a process copy, the nodeid specifies the lock master. 193 */ 194 195 /* lkb_status */ 196 197 #define DLM_LKSTS_WAITING 1 198 #define DLM_LKSTS_GRANTED 2 199 #define DLM_LKSTS_CONVERT 3 200 201 /* lkb_flags */ 202 203 #define DLM_IFL_MSTCPY 0x00010000 204 #define DLM_IFL_RESEND 0x00020000 205 #define DLM_IFL_DEAD 0x00040000 206 #define DLM_IFL_OVERLAP_UNLOCK 0x00080000 207 #define DLM_IFL_OVERLAP_CANCEL 0x00100000 208 #define DLM_IFL_ENDOFLIFE 0x00200000 209 #define DLM_IFL_WATCH_TIMEWARN 0x00400000 210 #define DLM_IFL_TIMEOUT_CANCEL 0x00800000 211 #define DLM_IFL_DEADLOCK_CANCEL 0x01000000 212 #define DLM_IFL_USER 0x00000001 213 #define DLM_IFL_ORPHAN 0x00000002 214 215 #define DLM_CALLBACKS_SIZE 6 216 217 #define DLM_CB_CAST 0x00000001 218 #define DLM_CB_BAST 0x00000002 219 #define DLM_CB_SKIP 0x00000004 220 221 struct dlm_callback { 222 uint64_t seq; 223 uint32_t flags; /* DLM_CBF_ */ 224 int sb_status; /* copy to lksb status */ 225 uint8_t sb_flags; /* copy to lksb flags */ 226 int8_t mode; /* rq mode of bast, gr mode of cast */ 227 }; 228 229 struct dlm_lkb { 230 struct dlm_rsb *lkb_resource; /* the rsb */ 231 struct kref lkb_ref; 232 int lkb_nodeid; /* copied from rsb */ 233 int lkb_ownpid; /* pid of lock owner */ 234 uint32_t lkb_id; /* our lock ID */ 235 uint32_t lkb_remid; /* lock ID on remote partner */ 236 uint32_t lkb_exflags; /* external flags from caller */ 237 uint32_t lkb_sbflags; /* lksb flags */ 238 uint32_t lkb_flags; /* internal flags */ 239 uint32_t lkb_lvbseq; /* lvb sequence number */ 240 241 int8_t lkb_status; /* granted, waiting, convert */ 242 int8_t lkb_rqmode; /* requested lock mode */ 243 int8_t lkb_grmode; /* granted lock mode */ 244 int8_t lkb_highbast; /* highest mode bast sent for */ 245 246 int8_t lkb_wait_type; /* type of reply waiting for */ 247 int8_t lkb_wait_count; 248 249 struct list_head lkb_idtbl_list; /* lockspace lkbtbl */ 250 struct list_head lkb_statequeue; /* rsb g/c/w list */ 251 struct list_head lkb_rsb_lookup; /* waiting for rsb lookup */ 252 struct list_head lkb_wait_reply; /* waiting for remote reply */ 253 struct list_head lkb_astqueue; /* need ast to be sent */ 254 struct list_head lkb_ownqueue; /* list of locks for a process */ 255 struct list_head lkb_time_list; 256 ktime_t lkb_timestamp; 257 unsigned long lkb_timeout_cs; 258 259 struct dlm_callback lkb_callbacks[DLM_CALLBACKS_SIZE]; 260 struct dlm_callback lkb_last_cast; 261 struct dlm_callback lkb_last_bast; 262 ktime_t lkb_last_cast_time; /* for debugging */ 263 ktime_t lkb_last_bast_time; /* for debugging */ 264 265 char *lkb_lvbptr; 266 struct dlm_lksb *lkb_lksb; /* caller's status block */ 267 void (*lkb_astfn) (void *astparam); 268 void (*lkb_bastfn) (void *astparam, int mode); 269 union { 270 void *lkb_astparam; /* caller's ast arg */ 271 struct dlm_user_args *lkb_ua; 272 }; 273 }; 274 275 276 struct dlm_rsb { 277 struct dlm_ls *res_ls; /* the lockspace */ 278 struct kref res_ref; 279 struct mutex res_mutex; 280 unsigned long res_flags; 281 int res_length; /* length of rsb name */ 282 int res_nodeid; 283 uint32_t res_lvbseq; 284 uint32_t res_hash; 285 uint32_t res_bucket; /* rsbtbl */ 286 unsigned long res_toss_time; 287 uint32_t res_first_lkid; 288 struct list_head res_lookup; /* lkbs waiting on first */ 289 struct list_head res_hashchain; /* rsbtbl */ 290 struct list_head res_grantqueue; 291 struct list_head res_convertqueue; 292 struct list_head res_waitqueue; 293 294 struct list_head res_root_list; /* used for recovery */ 295 struct list_head res_recover_list; /* used for recovery */ 296 int res_recover_locks_count; 297 298 char *res_lvbptr; 299 char res_name[1]; 300 }; 301 302 /* find_rsb() flags */ 303 304 #define R_MASTER 1 /* only return rsb if it's a master */ 305 #define R_CREATE 2 /* create/add rsb if not found */ 306 307 /* rsb_flags */ 308 309 enum rsb_flags { 310 RSB_MASTER_UNCERTAIN, 311 RSB_VALNOTVALID, 312 RSB_VALNOTVALID_PREV, 313 RSB_NEW_MASTER, 314 RSB_NEW_MASTER2, 315 RSB_RECOVER_CONVERT, 316 RSB_LOCKS_PURGED, 317 }; 318 319 static inline void rsb_set_flag(struct dlm_rsb *r, enum rsb_flags flag) 320 { 321 __set_bit(flag, &r->res_flags); 322 } 323 324 static inline void rsb_clear_flag(struct dlm_rsb *r, enum rsb_flags flag) 325 { 326 __clear_bit(flag, &r->res_flags); 327 } 328 329 static inline int rsb_flag(struct dlm_rsb *r, enum rsb_flags flag) 330 { 331 return test_bit(flag, &r->res_flags); 332 } 333 334 335 /* dlm_header is first element of all structs sent between nodes */ 336 337 #define DLM_HEADER_MAJOR 0x00030000 338 #define DLM_HEADER_MINOR 0x00000000 339 340 #define DLM_MSG 1 341 #define DLM_RCOM 2 342 343 struct dlm_header { 344 uint32_t h_version; 345 uint32_t h_lockspace; 346 uint32_t h_nodeid; /* nodeid of sender */ 347 uint16_t h_length; 348 uint8_t h_cmd; /* DLM_MSG, DLM_RCOM */ 349 uint8_t h_pad; 350 }; 351 352 353 #define DLM_MSG_REQUEST 1 354 #define DLM_MSG_CONVERT 2 355 #define DLM_MSG_UNLOCK 3 356 #define DLM_MSG_CANCEL 4 357 #define DLM_MSG_REQUEST_REPLY 5 358 #define DLM_MSG_CONVERT_REPLY 6 359 #define DLM_MSG_UNLOCK_REPLY 7 360 #define DLM_MSG_CANCEL_REPLY 8 361 #define DLM_MSG_GRANT 9 362 #define DLM_MSG_BAST 10 363 #define DLM_MSG_LOOKUP 11 364 #define DLM_MSG_REMOVE 12 365 #define DLM_MSG_LOOKUP_REPLY 13 366 #define DLM_MSG_PURGE 14 367 368 struct dlm_message { 369 struct dlm_header m_header; 370 uint32_t m_type; /* DLM_MSG_ */ 371 uint32_t m_nodeid; 372 uint32_t m_pid; 373 uint32_t m_lkid; /* lkid on sender */ 374 uint32_t m_remid; /* lkid on receiver */ 375 uint32_t m_parent_lkid; 376 uint32_t m_parent_remid; 377 uint32_t m_exflags; 378 uint32_t m_sbflags; 379 uint32_t m_flags; 380 uint32_t m_lvbseq; 381 uint32_t m_hash; 382 int m_status; 383 int m_grmode; 384 int m_rqmode; 385 int m_bastmode; 386 int m_asts; 387 int m_result; /* 0 or -EXXX */ 388 char m_extra[0]; /* name or lvb */ 389 }; 390 391 392 #define DLM_RS_NODES 0x00000001 393 #define DLM_RS_NODES_ALL 0x00000002 394 #define DLM_RS_DIR 0x00000004 395 #define DLM_RS_DIR_ALL 0x00000008 396 #define DLM_RS_LOCKS 0x00000010 397 #define DLM_RS_LOCKS_ALL 0x00000020 398 #define DLM_RS_DONE 0x00000040 399 #define DLM_RS_DONE_ALL 0x00000080 400 401 #define DLM_RCOM_STATUS 1 402 #define DLM_RCOM_NAMES 2 403 #define DLM_RCOM_LOOKUP 3 404 #define DLM_RCOM_LOCK 4 405 #define DLM_RCOM_STATUS_REPLY 5 406 #define DLM_RCOM_NAMES_REPLY 6 407 #define DLM_RCOM_LOOKUP_REPLY 7 408 #define DLM_RCOM_LOCK_REPLY 8 409 410 struct dlm_rcom { 411 struct dlm_header rc_header; 412 uint32_t rc_type; /* DLM_RCOM_ */ 413 int rc_result; /* multi-purpose */ 414 uint64_t rc_id; /* match reply with request */ 415 uint64_t rc_seq; /* sender's ls_recover_seq */ 416 uint64_t rc_seq_reply; /* remote ls_recover_seq */ 417 char rc_buf[0]; 418 }; 419 420 union dlm_packet { 421 struct dlm_header header; /* common to other two */ 422 struct dlm_message message; 423 struct dlm_rcom rcom; 424 }; 425 426 struct rcom_config { 427 __le32 rf_lvblen; 428 __le32 rf_lsflags; 429 __le64 rf_unused; 430 }; 431 432 struct rcom_lock { 433 __le32 rl_ownpid; 434 __le32 rl_lkid; 435 __le32 rl_remid; 436 __le32 rl_parent_lkid; 437 __le32 rl_parent_remid; 438 __le32 rl_exflags; 439 __le32 rl_flags; 440 __le32 rl_lvbseq; 441 __le32 rl_result; 442 int8_t rl_rqmode; 443 int8_t rl_grmode; 444 int8_t rl_status; 445 int8_t rl_asts; 446 __le16 rl_wait_type; 447 __le16 rl_namelen; 448 char rl_name[DLM_RESNAME_MAXLEN]; 449 char rl_lvb[0]; 450 }; 451 452 struct dlm_ls { 453 struct list_head ls_list; /* list of lockspaces */ 454 dlm_lockspace_t *ls_local_handle; 455 uint32_t ls_global_id; /* global unique lockspace ID */ 456 uint32_t ls_exflags; 457 int ls_lvblen; 458 int ls_count; /* refcount of processes in 459 the dlm using this ls */ 460 int ls_create_count; /* create/release refcount */ 461 unsigned long ls_flags; /* LSFL_ */ 462 unsigned long ls_scan_time; 463 struct kobject ls_kobj; 464 465 struct dlm_rsbtable *ls_rsbtbl; 466 uint32_t ls_rsbtbl_size; 467 468 struct dlm_lkbtable *ls_lkbtbl; 469 uint32_t ls_lkbtbl_size; 470 471 struct dlm_dirtable *ls_dirtbl; 472 uint32_t ls_dirtbl_size; 473 474 struct mutex ls_waiters_mutex; 475 struct list_head ls_waiters; /* lkbs needing a reply */ 476 477 struct mutex ls_orphans_mutex; 478 struct list_head ls_orphans; 479 480 struct mutex ls_timeout_mutex; 481 struct list_head ls_timeout; 482 483 struct list_head ls_nodes; /* current nodes in ls */ 484 struct list_head ls_nodes_gone; /* dead node list, recovery */ 485 int ls_num_nodes; /* number of nodes in ls */ 486 int ls_low_nodeid; 487 int ls_total_weight; 488 int *ls_node_array; 489 490 struct dlm_rsb ls_stub_rsb; /* for returning errors */ 491 struct dlm_lkb ls_stub_lkb; /* for returning errors */ 492 struct dlm_message ls_stub_ms; /* for faking a reply */ 493 494 struct dentry *ls_debug_rsb_dentry; /* debugfs */ 495 struct dentry *ls_debug_waiters_dentry; /* debugfs */ 496 struct dentry *ls_debug_locks_dentry; /* debugfs */ 497 struct dentry *ls_debug_all_dentry; /* debugfs */ 498 499 wait_queue_head_t ls_uevent_wait; /* user part of join/leave */ 500 int ls_uevent_result; 501 struct completion ls_members_done; 502 int ls_members_result; 503 504 struct miscdevice ls_device; 505 506 /* recovery related */ 507 508 struct timer_list ls_timer; 509 struct task_struct *ls_recoverd_task; 510 struct mutex ls_recoverd_active; 511 spinlock_t ls_recover_lock; 512 unsigned long ls_recover_begin; /* jiffies timestamp */ 513 uint32_t ls_recover_status; /* DLM_RS_ */ 514 uint64_t ls_recover_seq; 515 struct dlm_recover *ls_recover_args; 516 struct rw_semaphore ls_in_recovery; /* block local requests */ 517 struct rw_semaphore ls_recv_active; /* block dlm_recv */ 518 struct list_head ls_requestqueue;/* queue remote requests */ 519 struct mutex ls_requestqueue_mutex; 520 struct dlm_rcom *ls_recover_buf; 521 int ls_recover_nodeid; /* for debugging */ 522 uint64_t ls_rcom_seq; 523 spinlock_t ls_rcom_spin; 524 struct list_head ls_recover_list; 525 spinlock_t ls_recover_list_lock; 526 int ls_recover_list_count; 527 wait_queue_head_t ls_wait_general; 528 struct mutex ls_clear_proc_locks; 529 530 struct list_head ls_root_list; /* root resources */ 531 struct rw_semaphore ls_root_sem; /* protect root_list */ 532 533 int ls_namelen; 534 char ls_name[1]; 535 }; 536 537 #define LSFL_WORK 0 538 #define LSFL_RUNNING 1 539 #define LSFL_RECOVERY_STOP 2 540 #define LSFL_RCOM_READY 3 541 #define LSFL_RCOM_WAIT 4 542 #define LSFL_UEVENT_WAIT 5 543 #define LSFL_TIMEWARN 6 544 545 /* much of this is just saving user space pointers associated with the 546 lock that we pass back to the user lib with an ast */ 547 548 struct dlm_user_args { 549 struct dlm_user_proc *proc; /* each process that opens the lockspace 550 device has private data 551 (dlm_user_proc) on the struct file, 552 the process's locks point back to it*/ 553 struct dlm_lksb lksb; 554 struct dlm_lksb __user *user_lksb; 555 void __user *castparam; 556 void __user *castaddr; 557 void __user *bastparam; 558 void __user *bastaddr; 559 uint64_t xid; 560 }; 561 562 #define DLM_PROC_FLAGS_CLOSING 1 563 #define DLM_PROC_FLAGS_COMPAT 2 564 565 /* locks list is kept so we can remove all a process's locks when it 566 exits (or orphan those that are persistent) */ 567 568 struct dlm_user_proc { 569 dlm_lockspace_t *lockspace; 570 unsigned long flags; /* DLM_PROC_FLAGS */ 571 struct list_head asts; 572 spinlock_t asts_spin; 573 struct list_head locks; 574 spinlock_t locks_spin; 575 struct list_head unlocking; 576 wait_queue_head_t wait; 577 }; 578 579 static inline int dlm_locking_stopped(struct dlm_ls *ls) 580 { 581 return !test_bit(LSFL_RUNNING, &ls->ls_flags); 582 } 583 584 static inline int dlm_recovery_stopped(struct dlm_ls *ls) 585 { 586 return test_bit(LSFL_RECOVERY_STOP, &ls->ls_flags); 587 } 588 589 static inline int dlm_no_directory(struct dlm_ls *ls) 590 { 591 return (ls->ls_exflags & DLM_LSFL_NODIR) ? 1 : 0; 592 } 593 594 int dlm_netlink_init(void); 595 void dlm_netlink_exit(void); 596 void dlm_timeout_warn(struct dlm_lkb *lkb); 597 int dlm_plock_init(void); 598 void dlm_plock_exit(void); 599 600 #ifdef CONFIG_DLM_DEBUG 601 int dlm_register_debugfs(void); 602 void dlm_unregister_debugfs(void); 603 int dlm_create_debug_file(struct dlm_ls *ls); 604 void dlm_delete_debug_file(struct dlm_ls *ls); 605 #else 606 static inline int dlm_register_debugfs(void) { return 0; } 607 static inline void dlm_unregister_debugfs(void) { } 608 static inline int dlm_create_debug_file(struct dlm_ls *ls) { return 0; } 609 static inline void dlm_delete_debug_file(struct dlm_ls *ls) { } 610 #endif 611 612 #endif /* __DLM_INTERNAL_DOT_H__ */ 613 614