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 <linux/idr.h> 41 #include <asm/uaccess.h> 42 43 #include <linux/dlm.h> 44 #include "config.h" 45 46 /* Size of the temp buffer midcomms allocates on the stack. 47 We try to make this large enough so most messages fit. 48 FIXME: should sctp make this unnecessary? */ 49 50 #define DLM_INBUF_LEN 148 51 52 struct dlm_ls; 53 struct dlm_lkb; 54 struct dlm_rsb; 55 struct dlm_member; 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 }; 110 111 112 /* 113 * Lockspace member (per node in a ls) 114 */ 115 116 struct dlm_member { 117 struct list_head list; 118 int nodeid; 119 int weight; 120 }; 121 122 /* 123 * Save and manage recovery state for a lockspace. 124 */ 125 126 struct dlm_recover { 127 struct list_head list; 128 int *nodeids; /* nodeids of all members */ 129 int node_count; 130 int *new; /* nodeids of new members */ 131 int new_count; 132 uint64_t seq; 133 }; 134 135 /* 136 * Pass input args to second stage locking function. 137 */ 138 139 struct dlm_args { 140 uint32_t flags; 141 void (*astfn) (void *astparam); 142 void *astparam; 143 void (*bastfn) (void *astparam, int mode); 144 int mode; 145 struct dlm_lksb *lksb; 146 unsigned long timeout; 147 }; 148 149 150 /* 151 * Lock block 152 * 153 * A lock can be one of three types: 154 * 155 * local copy lock is mastered locally 156 * (lkb_nodeid is zero and DLM_LKF_MSTCPY is not set) 157 * process copy lock is mastered on a remote node 158 * (lkb_nodeid is non-zero and DLM_LKF_MSTCPY is not set) 159 * master copy master node's copy of a lock owned by remote node 160 * (lkb_nodeid is non-zero and DLM_LKF_MSTCPY is set) 161 * 162 * lkb_exflags: a copy of the most recent flags arg provided to dlm_lock or 163 * dlm_unlock. The dlm does not modify these or use any private flags in 164 * this field; it only contains DLM_LKF_ flags from dlm.h. These flags 165 * are sent as-is to the remote master when the lock is remote. 166 * 167 * lkb_flags: internal dlm flags (DLM_IFL_ prefix) from dlm_internal.h. 168 * Some internal flags are shared between the master and process nodes; 169 * these shared flags are kept in the lower two bytes. One of these 170 * flags set on the master copy will be propagated to the process copy 171 * and v.v. Other internal flags are private to the master or process 172 * node (e.g. DLM_IFL_MSTCPY). These are kept in the high two bytes. 173 * 174 * lkb_sbflags: status block flags. These flags are copied directly into 175 * the caller's lksb.sb_flags prior to the dlm_lock/dlm_unlock completion 176 * ast. All defined in dlm.h with DLM_SBF_ prefix. 177 * 178 * lkb_status: the lock status indicates which rsb queue the lock is 179 * on, grant, convert, or wait. DLM_LKSTS_ WAITING/GRANTED/CONVERT 180 * 181 * lkb_wait_type: the dlm message type (DLM_MSG_ prefix) for which a 182 * reply is needed. Only set when the lkb is on the lockspace waiters 183 * list awaiting a reply from a remote node. 184 * 185 * lkb_nodeid: when the lkb is a local copy, nodeid is 0; when the lkb 186 * is a master copy, nodeid specifies the remote lock holder, when the 187 * lkb is a process copy, the nodeid specifies the lock master. 188 */ 189 190 /* lkb_status */ 191 192 #define DLM_LKSTS_WAITING 1 193 #define DLM_LKSTS_GRANTED 2 194 #define DLM_LKSTS_CONVERT 3 195 196 /* lkb_flags */ 197 198 #define DLM_IFL_MSTCPY 0x00010000 199 #define DLM_IFL_RESEND 0x00020000 200 #define DLM_IFL_DEAD 0x00040000 201 #define DLM_IFL_OVERLAP_UNLOCK 0x00080000 202 #define DLM_IFL_OVERLAP_CANCEL 0x00100000 203 #define DLM_IFL_ENDOFLIFE 0x00200000 204 #define DLM_IFL_WATCH_TIMEWARN 0x00400000 205 #define DLM_IFL_TIMEOUT_CANCEL 0x00800000 206 #define DLM_IFL_DEADLOCK_CANCEL 0x01000000 207 #define DLM_IFL_STUB_MS 0x02000000 /* magic number for m_flags */ 208 #define DLM_IFL_USER 0x00000001 209 #define DLM_IFL_ORPHAN 0x00000002 210 211 #define DLM_CALLBACKS_SIZE 6 212 213 #define DLM_CB_CAST 0x00000001 214 #define DLM_CB_BAST 0x00000002 215 #define DLM_CB_SKIP 0x00000004 216 217 struct dlm_callback { 218 uint64_t seq; 219 uint32_t flags; /* DLM_CBF_ */ 220 int sb_status; /* copy to lksb status */ 221 uint8_t sb_flags; /* copy to lksb flags */ 222 int8_t mode; /* rq mode of bast, gr mode of cast */ 223 }; 224 225 struct dlm_lkb { 226 struct dlm_rsb *lkb_resource; /* the rsb */ 227 struct kref lkb_ref; 228 int lkb_nodeid; /* copied from rsb */ 229 int lkb_ownpid; /* pid of lock owner */ 230 uint32_t lkb_id; /* our lock ID */ 231 uint32_t lkb_remid; /* lock ID on remote partner */ 232 uint32_t lkb_exflags; /* external flags from caller */ 233 uint32_t lkb_sbflags; /* lksb flags */ 234 uint32_t lkb_flags; /* internal flags */ 235 uint32_t lkb_lvbseq; /* lvb sequence number */ 236 237 int8_t lkb_status; /* granted, waiting, convert */ 238 int8_t lkb_rqmode; /* requested lock mode */ 239 int8_t lkb_grmode; /* granted lock mode */ 240 int8_t lkb_highbast; /* highest mode bast sent for */ 241 242 int8_t lkb_wait_type; /* type of reply waiting for */ 243 int8_t lkb_wait_count; 244 int lkb_wait_nodeid; /* for debugging */ 245 246 struct list_head lkb_statequeue; /* rsb g/c/w list */ 247 struct list_head lkb_rsb_lookup; /* waiting for rsb lookup */ 248 struct list_head lkb_wait_reply; /* waiting for remote reply */ 249 struct list_head lkb_ownqueue; /* list of locks for a process */ 250 struct list_head lkb_time_list; 251 ktime_t lkb_timestamp; 252 ktime_t lkb_wait_time; 253 unsigned long lkb_timeout_cs; 254 255 struct mutex lkb_cb_mutex; 256 struct work_struct lkb_cb_work; 257 struct list_head lkb_cb_list; /* for ls_cb_delay or proc->asts */ 258 struct dlm_callback lkb_callbacks[DLM_CALLBACKS_SIZE]; 259 struct dlm_callback lkb_last_cast; 260 struct dlm_callback lkb_last_bast; 261 ktime_t lkb_last_cast_time; /* for debugging */ 262 ktime_t lkb_last_bast_time; /* for debugging */ 263 264 char *lkb_lvbptr; 265 struct dlm_lksb *lkb_lksb; /* caller's status block */ 266 void (*lkb_astfn) (void *astparam); 267 void (*lkb_bastfn) (void *astparam, int mode); 268 union { 269 void *lkb_astparam; /* caller's ast arg */ 270 struct dlm_user_args *lkb_ua; 271 }; 272 }; 273 274 275 struct dlm_rsb { 276 struct dlm_ls *res_ls; /* the lockspace */ 277 struct kref res_ref; 278 struct mutex res_mutex; 279 unsigned long res_flags; 280 int res_length; /* length of rsb name */ 281 int res_nodeid; 282 uint32_t res_lvbseq; 283 uint32_t res_hash; 284 uint32_t res_bucket; /* rsbtbl */ 285 unsigned long res_toss_time; 286 uint32_t res_first_lkid; 287 struct list_head res_lookup; /* lkbs waiting on first */ 288 struct list_head res_hashchain; /* rsbtbl */ 289 struct list_head res_grantqueue; 290 struct list_head res_convertqueue; 291 struct list_head res_waitqueue; 292 293 struct list_head res_root_list; /* used for recovery */ 294 struct list_head res_recover_list; /* used for recovery */ 295 int res_recover_locks_count; 296 297 char *res_lvbptr; 298 char res_name[DLM_RESNAME_MAXLEN+1]; 299 }; 300 301 /* find_rsb() flags */ 302 303 #define R_MASTER 1 /* only return rsb if it's a master */ 304 #define R_CREATE 2 /* create/add rsb if not found */ 305 306 /* rsb_flags */ 307 308 enum rsb_flags { 309 RSB_MASTER_UNCERTAIN, 310 RSB_VALNOTVALID, 311 RSB_VALNOTVALID_PREV, 312 RSB_NEW_MASTER, 313 RSB_NEW_MASTER2, 314 RSB_RECOVER_CONVERT, 315 RSB_LOCKS_PURGED, 316 }; 317 318 static inline void rsb_set_flag(struct dlm_rsb *r, enum rsb_flags flag) 319 { 320 __set_bit(flag, &r->res_flags); 321 } 322 323 static inline void rsb_clear_flag(struct dlm_rsb *r, enum rsb_flags flag) 324 { 325 __clear_bit(flag, &r->res_flags); 326 } 327 328 static inline int rsb_flag(struct dlm_rsb *r, enum rsb_flags flag) 329 { 330 return test_bit(flag, &r->res_flags); 331 } 332 333 334 /* dlm_header is first element of all structs sent between nodes */ 335 336 #define DLM_HEADER_MAJOR 0x00030000 337 #define DLM_HEADER_MINOR 0x00000000 338 339 #define DLM_MSG 1 340 #define DLM_RCOM 2 341 342 struct dlm_header { 343 uint32_t h_version; 344 uint32_t h_lockspace; 345 uint32_t h_nodeid; /* nodeid of sender */ 346 uint16_t h_length; 347 uint8_t h_cmd; /* DLM_MSG, DLM_RCOM */ 348 uint8_t h_pad; 349 }; 350 351 352 #define DLM_MSG_REQUEST 1 353 #define DLM_MSG_CONVERT 2 354 #define DLM_MSG_UNLOCK 3 355 #define DLM_MSG_CANCEL 4 356 #define DLM_MSG_REQUEST_REPLY 5 357 #define DLM_MSG_CONVERT_REPLY 6 358 #define DLM_MSG_UNLOCK_REPLY 7 359 #define DLM_MSG_CANCEL_REPLY 8 360 #define DLM_MSG_GRANT 9 361 #define DLM_MSG_BAST 10 362 #define DLM_MSG_LOOKUP 11 363 #define DLM_MSG_REMOVE 12 364 #define DLM_MSG_LOOKUP_REPLY 13 365 #define DLM_MSG_PURGE 14 366 367 struct dlm_message { 368 struct dlm_header m_header; 369 uint32_t m_type; /* DLM_MSG_ */ 370 uint32_t m_nodeid; 371 uint32_t m_pid; 372 uint32_t m_lkid; /* lkid on sender */ 373 uint32_t m_remid; /* lkid on receiver */ 374 uint32_t m_parent_lkid; 375 uint32_t m_parent_remid; 376 uint32_t m_exflags; 377 uint32_t m_sbflags; 378 uint32_t m_flags; 379 uint32_t m_lvbseq; 380 uint32_t m_hash; 381 int m_status; 382 int m_grmode; 383 int m_rqmode; 384 int m_bastmode; 385 int m_asts; 386 int m_result; /* 0 or -EXXX */ 387 char m_extra[0]; /* name or lvb */ 388 }; 389 390 391 #define DLM_RS_NODES 0x00000001 392 #define DLM_RS_NODES_ALL 0x00000002 393 #define DLM_RS_DIR 0x00000004 394 #define DLM_RS_DIR_ALL 0x00000008 395 #define DLM_RS_LOCKS 0x00000010 396 #define DLM_RS_LOCKS_ALL 0x00000020 397 #define DLM_RS_DONE 0x00000040 398 #define DLM_RS_DONE_ALL 0x00000080 399 400 #define DLM_RCOM_STATUS 1 401 #define DLM_RCOM_NAMES 2 402 #define DLM_RCOM_LOOKUP 3 403 #define DLM_RCOM_LOCK 4 404 #define DLM_RCOM_STATUS_REPLY 5 405 #define DLM_RCOM_NAMES_REPLY 6 406 #define DLM_RCOM_LOOKUP_REPLY 7 407 #define DLM_RCOM_LOCK_REPLY 8 408 409 struct dlm_rcom { 410 struct dlm_header rc_header; 411 uint32_t rc_type; /* DLM_RCOM_ */ 412 int rc_result; /* multi-purpose */ 413 uint64_t rc_id; /* match reply with request */ 414 uint64_t rc_seq; /* sender's ls_recover_seq */ 415 uint64_t rc_seq_reply; /* remote ls_recover_seq */ 416 char rc_buf[0]; 417 }; 418 419 union dlm_packet { 420 struct dlm_header header; /* common to other two */ 421 struct dlm_message message; 422 struct dlm_rcom rcom; 423 }; 424 425 struct rcom_config { 426 __le32 rf_lvblen; 427 __le32 rf_lsflags; 428 __le64 rf_unused; 429 }; 430 431 struct rcom_lock { 432 __le32 rl_ownpid; 433 __le32 rl_lkid; 434 __le32 rl_remid; 435 __le32 rl_parent_lkid; 436 __le32 rl_parent_remid; 437 __le32 rl_exflags; 438 __le32 rl_flags; 439 __le32 rl_lvbseq; 440 __le32 rl_result; 441 int8_t rl_rqmode; 442 int8_t rl_grmode; 443 int8_t rl_status; 444 int8_t rl_asts; 445 __le16 rl_wait_type; 446 __le16 rl_namelen; 447 char rl_name[DLM_RESNAME_MAXLEN]; 448 char rl_lvb[0]; 449 }; 450 451 struct dlm_ls { 452 struct list_head ls_list; /* list of lockspaces */ 453 dlm_lockspace_t *ls_local_handle; 454 uint32_t ls_global_id; /* global unique lockspace ID */ 455 uint32_t ls_exflags; 456 int ls_lvblen; 457 int ls_count; /* refcount of processes in 458 the dlm using this ls */ 459 int ls_create_count; /* create/release refcount */ 460 unsigned long ls_flags; /* LSFL_ */ 461 unsigned long ls_scan_time; 462 struct kobject ls_kobj; 463 464 struct idr ls_lkbidr; 465 spinlock_t ls_lkbidr_spin; 466 467 struct dlm_rsbtable *ls_rsbtbl; 468 uint32_t ls_rsbtbl_size; 469 470 struct dlm_dirtable *ls_dirtbl; 471 uint32_t ls_dirtbl_size; 472 473 struct mutex ls_waiters_mutex; 474 struct list_head ls_waiters; /* lkbs needing a reply */ 475 476 struct mutex ls_orphans_mutex; 477 struct list_head ls_orphans; 478 479 struct mutex ls_timeout_mutex; 480 struct list_head ls_timeout; 481 482 spinlock_t ls_new_rsb_spin; 483 int ls_new_rsb_count; 484 struct list_head ls_new_rsb; /* new rsb structs */ 485 486 struct list_head ls_nodes; /* current nodes in ls */ 487 struct list_head ls_nodes_gone; /* dead node list, recovery */ 488 int ls_num_nodes; /* number of nodes in ls */ 489 int ls_low_nodeid; 490 int ls_total_weight; 491 int *ls_node_array; 492 493 struct dlm_rsb ls_stub_rsb; /* for returning errors */ 494 struct dlm_lkb ls_stub_lkb; /* for returning errors */ 495 struct dlm_message ls_stub_ms; /* for faking a reply */ 496 497 struct dentry *ls_debug_rsb_dentry; /* debugfs */ 498 struct dentry *ls_debug_waiters_dentry; /* debugfs */ 499 struct dentry *ls_debug_locks_dentry; /* debugfs */ 500 struct dentry *ls_debug_all_dentry; /* debugfs */ 501 502 wait_queue_head_t ls_uevent_wait; /* user part of join/leave */ 503 int ls_uevent_result; 504 struct completion ls_members_done; 505 int ls_members_result; 506 507 struct miscdevice ls_device; 508 509 struct workqueue_struct *ls_callback_wq; 510 511 /* recovery related */ 512 513 struct mutex ls_cb_mutex; 514 struct list_head ls_cb_delay; /* save for queue_work later */ 515 struct timer_list ls_timer; 516 struct task_struct *ls_recoverd_task; 517 struct mutex ls_recoverd_active; 518 spinlock_t ls_recover_lock; 519 unsigned long ls_recover_begin; /* jiffies timestamp */ 520 uint32_t ls_recover_status; /* DLM_RS_ */ 521 uint64_t ls_recover_seq; 522 struct dlm_recover *ls_recover_args; 523 struct rw_semaphore ls_in_recovery; /* block local requests */ 524 struct rw_semaphore ls_recv_active; /* block dlm_recv */ 525 struct list_head ls_requestqueue;/* queue remote requests */ 526 struct mutex ls_requestqueue_mutex; 527 struct dlm_rcom *ls_recover_buf; 528 int ls_recover_nodeid; /* for debugging */ 529 uint64_t ls_rcom_seq; 530 spinlock_t ls_rcom_spin; 531 struct list_head ls_recover_list; 532 spinlock_t ls_recover_list_lock; 533 int ls_recover_list_count; 534 wait_queue_head_t ls_wait_general; 535 struct mutex ls_clear_proc_locks; 536 537 struct list_head ls_root_list; /* root resources */ 538 struct rw_semaphore ls_root_sem; /* protect root_list */ 539 540 int ls_namelen; 541 char ls_name[1]; 542 }; 543 544 #define LSFL_WORK 0 545 #define LSFL_RUNNING 1 546 #define LSFL_RECOVERY_STOP 2 547 #define LSFL_RCOM_READY 3 548 #define LSFL_RCOM_WAIT 4 549 #define LSFL_UEVENT_WAIT 5 550 #define LSFL_TIMEWARN 6 551 #define LSFL_CB_DELAY 7 552 553 /* much of this is just saving user space pointers associated with the 554 lock that we pass back to the user lib with an ast */ 555 556 struct dlm_user_args { 557 struct dlm_user_proc *proc; /* each process that opens the lockspace 558 device has private data 559 (dlm_user_proc) on the struct file, 560 the process's locks point back to it*/ 561 struct dlm_lksb lksb; 562 struct dlm_lksb __user *user_lksb; 563 void __user *castparam; 564 void __user *castaddr; 565 void __user *bastparam; 566 void __user *bastaddr; 567 uint64_t xid; 568 }; 569 570 #define DLM_PROC_FLAGS_CLOSING 1 571 #define DLM_PROC_FLAGS_COMPAT 2 572 573 /* locks list is kept so we can remove all a process's locks when it 574 exits (or orphan those that are persistent) */ 575 576 struct dlm_user_proc { 577 dlm_lockspace_t *lockspace; 578 unsigned long flags; /* DLM_PROC_FLAGS */ 579 struct list_head asts; 580 spinlock_t asts_spin; 581 struct list_head locks; 582 spinlock_t locks_spin; 583 struct list_head unlocking; 584 wait_queue_head_t wait; 585 }; 586 587 static inline int dlm_locking_stopped(struct dlm_ls *ls) 588 { 589 return !test_bit(LSFL_RUNNING, &ls->ls_flags); 590 } 591 592 static inline int dlm_recovery_stopped(struct dlm_ls *ls) 593 { 594 return test_bit(LSFL_RECOVERY_STOP, &ls->ls_flags); 595 } 596 597 static inline int dlm_no_directory(struct dlm_ls *ls) 598 { 599 return (ls->ls_exflags & DLM_LSFL_NODIR) ? 1 : 0; 600 } 601 602 int dlm_netlink_init(void); 603 void dlm_netlink_exit(void); 604 void dlm_timeout_warn(struct dlm_lkb *lkb); 605 int dlm_plock_init(void); 606 void dlm_plock_exit(void); 607 608 #ifdef CONFIG_DLM_DEBUG 609 int dlm_register_debugfs(void); 610 void dlm_unregister_debugfs(void); 611 int dlm_create_debug_file(struct dlm_ls *ls); 612 void dlm_delete_debug_file(struct dlm_ls *ls); 613 #else 614 static inline int dlm_register_debugfs(void) { return 0; } 615 static inline void dlm_unregister_debugfs(void) { } 616 static inline int dlm_create_debug_file(struct dlm_ls *ls) { return 0; } 617 static inline void dlm_delete_debug_file(struct dlm_ls *ls) { } 618 #endif 619 620 #endif /* __DLM_INTERNAL_DOT_H__ */ 621 622