1 /****************************************************************************** 2 ******************************************************************************* 3 ** 4 ** Copyright (C) 2005 Red Hat, Inc. All rights reserved. 5 ** 6 ** This copyrighted material is made available to anyone wishing to use, 7 ** modify, copy, or redistribute it subject to the terms and conditions 8 ** of the GNU General Public License v.2. 9 ** 10 ******************************************************************************* 11 ******************************************************************************/ 12 13 /* Central locking logic has four stages: 14 15 dlm_lock() 16 dlm_unlock() 17 18 request_lock(ls, lkb) 19 convert_lock(ls, lkb) 20 unlock_lock(ls, lkb) 21 cancel_lock(ls, lkb) 22 23 _request_lock(r, lkb) 24 _convert_lock(r, lkb) 25 _unlock_lock(r, lkb) 26 _cancel_lock(r, lkb) 27 28 do_request(r, lkb) 29 do_convert(r, lkb) 30 do_unlock(r, lkb) 31 do_cancel(r, lkb) 32 33 Stage 1 (lock, unlock) is mainly about checking input args and 34 splitting into one of the four main operations: 35 36 dlm_lock = request_lock 37 dlm_lock+CONVERT = convert_lock 38 dlm_unlock = unlock_lock 39 dlm_unlock+CANCEL = cancel_lock 40 41 Stage 2, xxxx_lock(), just finds and locks the relevant rsb which is 42 provided to the next stage. 43 44 Stage 3, _xxxx_lock(), determines if the operation is local or remote. 45 When remote, it calls send_xxxx(), when local it calls do_xxxx(). 46 47 Stage 4, do_xxxx(), is the guts of the operation. It manipulates the 48 given rsb and lkb and queues callbacks. 49 50 For remote operations, send_xxxx() results in the corresponding do_xxxx() 51 function being executed on the remote node. The connecting send/receive 52 calls on local (L) and remote (R) nodes: 53 54 L: send_xxxx() -> R: receive_xxxx() 55 R: do_xxxx() 56 L: receive_xxxx_reply() <- R: send_xxxx_reply() 57 */ 58 #include <linux/types.h> 59 #include "dlm_internal.h" 60 #include <linux/dlm_device.h> 61 #include "memory.h" 62 #include "lowcomms.h" 63 #include "requestqueue.h" 64 #include "util.h" 65 #include "dir.h" 66 #include "member.h" 67 #include "lockspace.h" 68 #include "ast.h" 69 #include "lock.h" 70 #include "rcom.h" 71 #include "recover.h" 72 #include "lvb_table.h" 73 #include "user.h" 74 #include "config.h" 75 76 static int send_request(struct dlm_rsb *r, struct dlm_lkb *lkb); 77 static int send_convert(struct dlm_rsb *r, struct dlm_lkb *lkb); 78 static int send_unlock(struct dlm_rsb *r, struct dlm_lkb *lkb); 79 static int send_cancel(struct dlm_rsb *r, struct dlm_lkb *lkb); 80 static int send_grant(struct dlm_rsb *r, struct dlm_lkb *lkb); 81 static int send_bast(struct dlm_rsb *r, struct dlm_lkb *lkb, int mode); 82 static int send_lookup(struct dlm_rsb *r, struct dlm_lkb *lkb); 83 static int send_remove(struct dlm_rsb *r); 84 static int _request_lock(struct dlm_rsb *r, struct dlm_lkb *lkb); 85 static void __receive_convert_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, 86 struct dlm_message *ms); 87 static int receive_extralen(struct dlm_message *ms); 88 89 /* 90 * Lock compatibilty matrix - thanks Steve 91 * UN = Unlocked state. Not really a state, used as a flag 92 * PD = Padding. Used to make the matrix a nice power of two in size 93 * Other states are the same as the VMS DLM. 94 * Usage: matrix[grmode+1][rqmode+1] (although m[rq+1][gr+1] is the same) 95 */ 96 97 static const int __dlm_compat_matrix[8][8] = { 98 /* UN NL CR CW PR PW EX PD */ 99 {1, 1, 1, 1, 1, 1, 1, 0}, /* UN */ 100 {1, 1, 1, 1, 1, 1, 1, 0}, /* NL */ 101 {1, 1, 1, 1, 1, 1, 0, 0}, /* CR */ 102 {1, 1, 1, 1, 0, 0, 0, 0}, /* CW */ 103 {1, 1, 1, 0, 1, 0, 0, 0}, /* PR */ 104 {1, 1, 1, 0, 0, 0, 0, 0}, /* PW */ 105 {1, 1, 0, 0, 0, 0, 0, 0}, /* EX */ 106 {0, 0, 0, 0, 0, 0, 0, 0} /* PD */ 107 }; 108 109 /* 110 * This defines the direction of transfer of LVB data. 111 * Granted mode is the row; requested mode is the column. 112 * Usage: matrix[grmode+1][rqmode+1] 113 * 1 = LVB is returned to the caller 114 * 0 = LVB is written to the resource 115 * -1 = nothing happens to the LVB 116 */ 117 118 const int dlm_lvb_operations[8][8] = { 119 /* UN NL CR CW PR PW EX PD*/ 120 { -1, 1, 1, 1, 1, 1, 1, -1 }, /* UN */ 121 { -1, 1, 1, 1, 1, 1, 1, 0 }, /* NL */ 122 { -1, -1, 1, 1, 1, 1, 1, 0 }, /* CR */ 123 { -1, -1, -1, 1, 1, 1, 1, 0 }, /* CW */ 124 { -1, -1, -1, -1, 1, 1, 1, 0 }, /* PR */ 125 { -1, 0, 0, 0, 0, 0, 1, 0 }, /* PW */ 126 { -1, 0, 0, 0, 0, 0, 0, 0 }, /* EX */ 127 { -1, 0, 0, 0, 0, 0, 0, 0 } /* PD */ 128 }; 129 130 #define modes_compat(gr, rq) \ 131 __dlm_compat_matrix[(gr)->lkb_grmode + 1][(rq)->lkb_rqmode + 1] 132 133 int dlm_modes_compat(int mode1, int mode2) 134 { 135 return __dlm_compat_matrix[mode1 + 1][mode2 + 1]; 136 } 137 138 /* 139 * Compatibility matrix for conversions with QUECVT set. 140 * Granted mode is the row; requested mode is the column. 141 * Usage: matrix[grmode+1][rqmode+1] 142 */ 143 144 static const int __quecvt_compat_matrix[8][8] = { 145 /* UN NL CR CW PR PW EX PD */ 146 {0, 0, 0, 0, 0, 0, 0, 0}, /* UN */ 147 {0, 0, 1, 1, 1, 1, 1, 0}, /* NL */ 148 {0, 0, 0, 1, 1, 1, 1, 0}, /* CR */ 149 {0, 0, 0, 0, 1, 1, 1, 0}, /* CW */ 150 {0, 0, 0, 1, 0, 1, 1, 0}, /* PR */ 151 {0, 0, 0, 0, 0, 0, 1, 0}, /* PW */ 152 {0, 0, 0, 0, 0, 0, 0, 0}, /* EX */ 153 {0, 0, 0, 0, 0, 0, 0, 0} /* PD */ 154 }; 155 156 void dlm_print_lkb(struct dlm_lkb *lkb) 157 { 158 printk(KERN_ERR "lkb: nodeid %d id %x remid %x exflags %x flags %x\n" 159 " status %d rqmode %d grmode %d wait_type %d ast_type %d\n", 160 lkb->lkb_nodeid, lkb->lkb_id, lkb->lkb_remid, lkb->lkb_exflags, 161 lkb->lkb_flags, lkb->lkb_status, lkb->lkb_rqmode, 162 lkb->lkb_grmode, lkb->lkb_wait_type, lkb->lkb_ast_type); 163 } 164 165 void dlm_print_rsb(struct dlm_rsb *r) 166 { 167 printk(KERN_ERR "rsb: nodeid %d flags %lx first %x rlc %d name %s\n", 168 r->res_nodeid, r->res_flags, r->res_first_lkid, 169 r->res_recover_locks_count, r->res_name); 170 } 171 172 void dlm_dump_rsb(struct dlm_rsb *r) 173 { 174 struct dlm_lkb *lkb; 175 176 dlm_print_rsb(r); 177 178 printk(KERN_ERR "rsb: root_list empty %d recover_list empty %d\n", 179 list_empty(&r->res_root_list), list_empty(&r->res_recover_list)); 180 printk(KERN_ERR "rsb lookup list\n"); 181 list_for_each_entry(lkb, &r->res_lookup, lkb_rsb_lookup) 182 dlm_print_lkb(lkb); 183 printk(KERN_ERR "rsb grant queue:\n"); 184 list_for_each_entry(lkb, &r->res_grantqueue, lkb_statequeue) 185 dlm_print_lkb(lkb); 186 printk(KERN_ERR "rsb convert queue:\n"); 187 list_for_each_entry(lkb, &r->res_convertqueue, lkb_statequeue) 188 dlm_print_lkb(lkb); 189 printk(KERN_ERR "rsb wait queue:\n"); 190 list_for_each_entry(lkb, &r->res_waitqueue, lkb_statequeue) 191 dlm_print_lkb(lkb); 192 } 193 194 /* Threads cannot use the lockspace while it's being recovered */ 195 196 static inline void lock_recovery(struct dlm_ls *ls) 197 { 198 down_read(&ls->ls_in_recovery); 199 } 200 201 static inline void unlock_recovery(struct dlm_ls *ls) 202 { 203 up_read(&ls->ls_in_recovery); 204 } 205 206 static inline int lock_recovery_try(struct dlm_ls *ls) 207 { 208 return down_read_trylock(&ls->ls_in_recovery); 209 } 210 211 static inline int can_be_queued(struct dlm_lkb *lkb) 212 { 213 return !(lkb->lkb_exflags & DLM_LKF_NOQUEUE); 214 } 215 216 static inline int force_blocking_asts(struct dlm_lkb *lkb) 217 { 218 return (lkb->lkb_exflags & DLM_LKF_NOQUEUEBAST); 219 } 220 221 static inline int is_demoted(struct dlm_lkb *lkb) 222 { 223 return (lkb->lkb_sbflags & DLM_SBF_DEMOTED); 224 } 225 226 static inline int is_remote(struct dlm_rsb *r) 227 { 228 DLM_ASSERT(r->res_nodeid >= 0, dlm_print_rsb(r);); 229 return !!r->res_nodeid; 230 } 231 232 static inline int is_process_copy(struct dlm_lkb *lkb) 233 { 234 return (lkb->lkb_nodeid && !(lkb->lkb_flags & DLM_IFL_MSTCPY)); 235 } 236 237 static inline int is_master_copy(struct dlm_lkb *lkb) 238 { 239 if (lkb->lkb_flags & DLM_IFL_MSTCPY) 240 DLM_ASSERT(lkb->lkb_nodeid, dlm_print_lkb(lkb);); 241 return (lkb->lkb_flags & DLM_IFL_MSTCPY) ? 1 : 0; 242 } 243 244 static inline int middle_conversion(struct dlm_lkb *lkb) 245 { 246 if ((lkb->lkb_grmode==DLM_LOCK_PR && lkb->lkb_rqmode==DLM_LOCK_CW) || 247 (lkb->lkb_rqmode==DLM_LOCK_PR && lkb->lkb_grmode==DLM_LOCK_CW)) 248 return 1; 249 return 0; 250 } 251 252 static inline int down_conversion(struct dlm_lkb *lkb) 253 { 254 return (!middle_conversion(lkb) && lkb->lkb_rqmode < lkb->lkb_grmode); 255 } 256 257 static void queue_cast(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv) 258 { 259 if (is_master_copy(lkb)) 260 return; 261 262 DLM_ASSERT(lkb->lkb_lksb, dlm_print_lkb(lkb);); 263 264 lkb->lkb_lksb->sb_status = rv; 265 lkb->lkb_lksb->sb_flags = lkb->lkb_sbflags; 266 267 dlm_add_ast(lkb, AST_COMP); 268 } 269 270 static void queue_bast(struct dlm_rsb *r, struct dlm_lkb *lkb, int rqmode) 271 { 272 if (is_master_copy(lkb)) 273 send_bast(r, lkb, rqmode); 274 else { 275 lkb->lkb_bastmode = rqmode; 276 dlm_add_ast(lkb, AST_BAST); 277 } 278 } 279 280 /* 281 * Basic operations on rsb's and lkb's 282 */ 283 284 static struct dlm_rsb *create_rsb(struct dlm_ls *ls, char *name, int len) 285 { 286 struct dlm_rsb *r; 287 288 r = allocate_rsb(ls, len); 289 if (!r) 290 return NULL; 291 292 r->res_ls = ls; 293 r->res_length = len; 294 memcpy(r->res_name, name, len); 295 mutex_init(&r->res_mutex); 296 297 INIT_LIST_HEAD(&r->res_lookup); 298 INIT_LIST_HEAD(&r->res_grantqueue); 299 INIT_LIST_HEAD(&r->res_convertqueue); 300 INIT_LIST_HEAD(&r->res_waitqueue); 301 INIT_LIST_HEAD(&r->res_root_list); 302 INIT_LIST_HEAD(&r->res_recover_list); 303 304 return r; 305 } 306 307 static int search_rsb_list(struct list_head *head, char *name, int len, 308 unsigned int flags, struct dlm_rsb **r_ret) 309 { 310 struct dlm_rsb *r; 311 int error = 0; 312 313 list_for_each_entry(r, head, res_hashchain) { 314 if (len == r->res_length && !memcmp(name, r->res_name, len)) 315 goto found; 316 } 317 return -EBADR; 318 319 found: 320 if (r->res_nodeid && (flags & R_MASTER)) 321 error = -ENOTBLK; 322 *r_ret = r; 323 return error; 324 } 325 326 static int _search_rsb(struct dlm_ls *ls, char *name, int len, int b, 327 unsigned int flags, struct dlm_rsb **r_ret) 328 { 329 struct dlm_rsb *r; 330 int error; 331 332 error = search_rsb_list(&ls->ls_rsbtbl[b].list, name, len, flags, &r); 333 if (!error) { 334 kref_get(&r->res_ref); 335 goto out; 336 } 337 error = search_rsb_list(&ls->ls_rsbtbl[b].toss, name, len, flags, &r); 338 if (error) 339 goto out; 340 341 list_move(&r->res_hashchain, &ls->ls_rsbtbl[b].list); 342 343 if (dlm_no_directory(ls)) 344 goto out; 345 346 if (r->res_nodeid == -1) { 347 rsb_clear_flag(r, RSB_MASTER_UNCERTAIN); 348 r->res_first_lkid = 0; 349 } else if (r->res_nodeid > 0) { 350 rsb_set_flag(r, RSB_MASTER_UNCERTAIN); 351 r->res_first_lkid = 0; 352 } else { 353 DLM_ASSERT(r->res_nodeid == 0, dlm_print_rsb(r);); 354 DLM_ASSERT(!rsb_flag(r, RSB_MASTER_UNCERTAIN),); 355 } 356 out: 357 *r_ret = r; 358 return error; 359 } 360 361 static int search_rsb(struct dlm_ls *ls, char *name, int len, int b, 362 unsigned int flags, struct dlm_rsb **r_ret) 363 { 364 int error; 365 write_lock(&ls->ls_rsbtbl[b].lock); 366 error = _search_rsb(ls, name, len, b, flags, r_ret); 367 write_unlock(&ls->ls_rsbtbl[b].lock); 368 return error; 369 } 370 371 /* 372 * Find rsb in rsbtbl and potentially create/add one 373 * 374 * Delaying the release of rsb's has a similar benefit to applications keeping 375 * NL locks on an rsb, but without the guarantee that the cached master value 376 * will still be valid when the rsb is reused. Apps aren't always smart enough 377 * to keep NL locks on an rsb that they may lock again shortly; this can lead 378 * to excessive master lookups and removals if we don't delay the release. 379 * 380 * Searching for an rsb means looking through both the normal list and toss 381 * list. When found on the toss list the rsb is moved to the normal list with 382 * ref count of 1; when found on normal list the ref count is incremented. 383 */ 384 385 static int find_rsb(struct dlm_ls *ls, char *name, int namelen, 386 unsigned int flags, struct dlm_rsb **r_ret) 387 { 388 struct dlm_rsb *r, *tmp; 389 uint32_t hash, bucket; 390 int error = 0; 391 392 if (dlm_no_directory(ls)) 393 flags |= R_CREATE; 394 395 hash = jhash(name, namelen, 0); 396 bucket = hash & (ls->ls_rsbtbl_size - 1); 397 398 error = search_rsb(ls, name, namelen, bucket, flags, &r); 399 if (!error) 400 goto out; 401 402 if (error == -EBADR && !(flags & R_CREATE)) 403 goto out; 404 405 /* the rsb was found but wasn't a master copy */ 406 if (error == -ENOTBLK) 407 goto out; 408 409 error = -ENOMEM; 410 r = create_rsb(ls, name, namelen); 411 if (!r) 412 goto out; 413 414 r->res_hash = hash; 415 r->res_bucket = bucket; 416 r->res_nodeid = -1; 417 kref_init(&r->res_ref); 418 419 /* With no directory, the master can be set immediately */ 420 if (dlm_no_directory(ls)) { 421 int nodeid = dlm_dir_nodeid(r); 422 if (nodeid == dlm_our_nodeid()) 423 nodeid = 0; 424 r->res_nodeid = nodeid; 425 } 426 427 write_lock(&ls->ls_rsbtbl[bucket].lock); 428 error = _search_rsb(ls, name, namelen, bucket, 0, &tmp); 429 if (!error) { 430 write_unlock(&ls->ls_rsbtbl[bucket].lock); 431 free_rsb(r); 432 r = tmp; 433 goto out; 434 } 435 list_add(&r->res_hashchain, &ls->ls_rsbtbl[bucket].list); 436 write_unlock(&ls->ls_rsbtbl[bucket].lock); 437 error = 0; 438 out: 439 *r_ret = r; 440 return error; 441 } 442 443 int dlm_find_rsb(struct dlm_ls *ls, char *name, int namelen, 444 unsigned int flags, struct dlm_rsb **r_ret) 445 { 446 return find_rsb(ls, name, namelen, flags, r_ret); 447 } 448 449 /* This is only called to add a reference when the code already holds 450 a valid reference to the rsb, so there's no need for locking. */ 451 452 static inline void hold_rsb(struct dlm_rsb *r) 453 { 454 kref_get(&r->res_ref); 455 } 456 457 void dlm_hold_rsb(struct dlm_rsb *r) 458 { 459 hold_rsb(r); 460 } 461 462 static void toss_rsb(struct kref *kref) 463 { 464 struct dlm_rsb *r = container_of(kref, struct dlm_rsb, res_ref); 465 struct dlm_ls *ls = r->res_ls; 466 467 DLM_ASSERT(list_empty(&r->res_root_list), dlm_print_rsb(r);); 468 kref_init(&r->res_ref); 469 list_move(&r->res_hashchain, &ls->ls_rsbtbl[r->res_bucket].toss); 470 r->res_toss_time = jiffies; 471 if (r->res_lvbptr) { 472 free_lvb(r->res_lvbptr); 473 r->res_lvbptr = NULL; 474 } 475 } 476 477 /* When all references to the rsb are gone it's transfered to 478 the tossed list for later disposal. */ 479 480 static void put_rsb(struct dlm_rsb *r) 481 { 482 struct dlm_ls *ls = r->res_ls; 483 uint32_t bucket = r->res_bucket; 484 485 write_lock(&ls->ls_rsbtbl[bucket].lock); 486 kref_put(&r->res_ref, toss_rsb); 487 write_unlock(&ls->ls_rsbtbl[bucket].lock); 488 } 489 490 void dlm_put_rsb(struct dlm_rsb *r) 491 { 492 put_rsb(r); 493 } 494 495 /* See comment for unhold_lkb */ 496 497 static void unhold_rsb(struct dlm_rsb *r) 498 { 499 int rv; 500 rv = kref_put(&r->res_ref, toss_rsb); 501 DLM_ASSERT(!rv, dlm_dump_rsb(r);); 502 } 503 504 static void kill_rsb(struct kref *kref) 505 { 506 struct dlm_rsb *r = container_of(kref, struct dlm_rsb, res_ref); 507 508 /* All work is done after the return from kref_put() so we 509 can release the write_lock before the remove and free. */ 510 511 DLM_ASSERT(list_empty(&r->res_lookup), dlm_dump_rsb(r);); 512 DLM_ASSERT(list_empty(&r->res_grantqueue), dlm_dump_rsb(r);); 513 DLM_ASSERT(list_empty(&r->res_convertqueue), dlm_dump_rsb(r);); 514 DLM_ASSERT(list_empty(&r->res_waitqueue), dlm_dump_rsb(r);); 515 DLM_ASSERT(list_empty(&r->res_root_list), dlm_dump_rsb(r);); 516 DLM_ASSERT(list_empty(&r->res_recover_list), dlm_dump_rsb(r);); 517 } 518 519 /* Attaching/detaching lkb's from rsb's is for rsb reference counting. 520 The rsb must exist as long as any lkb's for it do. */ 521 522 static void attach_lkb(struct dlm_rsb *r, struct dlm_lkb *lkb) 523 { 524 hold_rsb(r); 525 lkb->lkb_resource = r; 526 } 527 528 static void detach_lkb(struct dlm_lkb *lkb) 529 { 530 if (lkb->lkb_resource) { 531 put_rsb(lkb->lkb_resource); 532 lkb->lkb_resource = NULL; 533 } 534 } 535 536 static int create_lkb(struct dlm_ls *ls, struct dlm_lkb **lkb_ret) 537 { 538 struct dlm_lkb *lkb, *tmp; 539 uint32_t lkid = 0; 540 uint16_t bucket; 541 542 lkb = allocate_lkb(ls); 543 if (!lkb) 544 return -ENOMEM; 545 546 lkb->lkb_nodeid = -1; 547 lkb->lkb_grmode = DLM_LOCK_IV; 548 kref_init(&lkb->lkb_ref); 549 INIT_LIST_HEAD(&lkb->lkb_ownqueue); 550 551 get_random_bytes(&bucket, sizeof(bucket)); 552 bucket &= (ls->ls_lkbtbl_size - 1); 553 554 write_lock(&ls->ls_lkbtbl[bucket].lock); 555 556 /* counter can roll over so we must verify lkid is not in use */ 557 558 while (lkid == 0) { 559 lkid = bucket | (ls->ls_lkbtbl[bucket].counter++ << 16); 560 561 list_for_each_entry(tmp, &ls->ls_lkbtbl[bucket].list, 562 lkb_idtbl_list) { 563 if (tmp->lkb_id != lkid) 564 continue; 565 lkid = 0; 566 break; 567 } 568 } 569 570 lkb->lkb_id = lkid; 571 list_add(&lkb->lkb_idtbl_list, &ls->ls_lkbtbl[bucket].list); 572 write_unlock(&ls->ls_lkbtbl[bucket].lock); 573 574 *lkb_ret = lkb; 575 return 0; 576 } 577 578 static struct dlm_lkb *__find_lkb(struct dlm_ls *ls, uint32_t lkid) 579 { 580 uint16_t bucket = lkid & 0xFFFF; 581 struct dlm_lkb *lkb; 582 583 list_for_each_entry(lkb, &ls->ls_lkbtbl[bucket].list, lkb_idtbl_list) { 584 if (lkb->lkb_id == lkid) 585 return lkb; 586 } 587 return NULL; 588 } 589 590 static int find_lkb(struct dlm_ls *ls, uint32_t lkid, struct dlm_lkb **lkb_ret) 591 { 592 struct dlm_lkb *lkb; 593 uint16_t bucket = lkid & 0xFFFF; 594 595 if (bucket >= ls->ls_lkbtbl_size) 596 return -EBADSLT; 597 598 read_lock(&ls->ls_lkbtbl[bucket].lock); 599 lkb = __find_lkb(ls, lkid); 600 if (lkb) 601 kref_get(&lkb->lkb_ref); 602 read_unlock(&ls->ls_lkbtbl[bucket].lock); 603 604 *lkb_ret = lkb; 605 return lkb ? 0 : -ENOENT; 606 } 607 608 static void kill_lkb(struct kref *kref) 609 { 610 struct dlm_lkb *lkb = container_of(kref, struct dlm_lkb, lkb_ref); 611 612 /* All work is done after the return from kref_put() so we 613 can release the write_lock before the detach_lkb */ 614 615 DLM_ASSERT(!lkb->lkb_status, dlm_print_lkb(lkb);); 616 } 617 618 /* __put_lkb() is used when an lkb may not have an rsb attached to 619 it so we need to provide the lockspace explicitly */ 620 621 static int __put_lkb(struct dlm_ls *ls, struct dlm_lkb *lkb) 622 { 623 uint16_t bucket = lkb->lkb_id & 0xFFFF; 624 625 write_lock(&ls->ls_lkbtbl[bucket].lock); 626 if (kref_put(&lkb->lkb_ref, kill_lkb)) { 627 list_del(&lkb->lkb_idtbl_list); 628 write_unlock(&ls->ls_lkbtbl[bucket].lock); 629 630 detach_lkb(lkb); 631 632 /* for local/process lkbs, lvbptr points to caller's lksb */ 633 if (lkb->lkb_lvbptr && is_master_copy(lkb)) 634 free_lvb(lkb->lkb_lvbptr); 635 free_lkb(lkb); 636 return 1; 637 } else { 638 write_unlock(&ls->ls_lkbtbl[bucket].lock); 639 return 0; 640 } 641 } 642 643 int dlm_put_lkb(struct dlm_lkb *lkb) 644 { 645 struct dlm_ls *ls; 646 647 DLM_ASSERT(lkb->lkb_resource, dlm_print_lkb(lkb);); 648 DLM_ASSERT(lkb->lkb_resource->res_ls, dlm_print_lkb(lkb);); 649 650 ls = lkb->lkb_resource->res_ls; 651 return __put_lkb(ls, lkb); 652 } 653 654 /* This is only called to add a reference when the code already holds 655 a valid reference to the lkb, so there's no need for locking. */ 656 657 static inline void hold_lkb(struct dlm_lkb *lkb) 658 { 659 kref_get(&lkb->lkb_ref); 660 } 661 662 /* This is called when we need to remove a reference and are certain 663 it's not the last ref. e.g. del_lkb is always called between a 664 find_lkb/put_lkb and is always the inverse of a previous add_lkb. 665 put_lkb would work fine, but would involve unnecessary locking */ 666 667 static inline void unhold_lkb(struct dlm_lkb *lkb) 668 { 669 int rv; 670 rv = kref_put(&lkb->lkb_ref, kill_lkb); 671 DLM_ASSERT(!rv, dlm_print_lkb(lkb);); 672 } 673 674 static void lkb_add_ordered(struct list_head *new, struct list_head *head, 675 int mode) 676 { 677 struct dlm_lkb *lkb = NULL; 678 679 list_for_each_entry(lkb, head, lkb_statequeue) 680 if (lkb->lkb_rqmode < mode) 681 break; 682 683 if (!lkb) 684 list_add_tail(new, head); 685 else 686 __list_add(new, lkb->lkb_statequeue.prev, &lkb->lkb_statequeue); 687 } 688 689 /* add/remove lkb to rsb's grant/convert/wait queue */ 690 691 static void add_lkb(struct dlm_rsb *r, struct dlm_lkb *lkb, int status) 692 { 693 kref_get(&lkb->lkb_ref); 694 695 DLM_ASSERT(!lkb->lkb_status, dlm_print_lkb(lkb);); 696 697 lkb->lkb_status = status; 698 699 switch (status) { 700 case DLM_LKSTS_WAITING: 701 if (lkb->lkb_exflags & DLM_LKF_HEADQUE) 702 list_add(&lkb->lkb_statequeue, &r->res_waitqueue); 703 else 704 list_add_tail(&lkb->lkb_statequeue, &r->res_waitqueue); 705 break; 706 case DLM_LKSTS_GRANTED: 707 /* convention says granted locks kept in order of grmode */ 708 lkb_add_ordered(&lkb->lkb_statequeue, &r->res_grantqueue, 709 lkb->lkb_grmode); 710 break; 711 case DLM_LKSTS_CONVERT: 712 if (lkb->lkb_exflags & DLM_LKF_HEADQUE) 713 list_add(&lkb->lkb_statequeue, &r->res_convertqueue); 714 else 715 list_add_tail(&lkb->lkb_statequeue, 716 &r->res_convertqueue); 717 break; 718 default: 719 DLM_ASSERT(0, dlm_print_lkb(lkb); printk("sts=%d\n", status);); 720 } 721 } 722 723 static void del_lkb(struct dlm_rsb *r, struct dlm_lkb *lkb) 724 { 725 lkb->lkb_status = 0; 726 list_del(&lkb->lkb_statequeue); 727 unhold_lkb(lkb); 728 } 729 730 static void move_lkb(struct dlm_rsb *r, struct dlm_lkb *lkb, int sts) 731 { 732 hold_lkb(lkb); 733 del_lkb(r, lkb); 734 add_lkb(r, lkb, sts); 735 unhold_lkb(lkb); 736 } 737 738 /* add/remove lkb from global waiters list of lkb's waiting for 739 a reply from a remote node */ 740 741 static void add_to_waiters(struct dlm_lkb *lkb, int mstype) 742 { 743 struct dlm_ls *ls = lkb->lkb_resource->res_ls; 744 745 mutex_lock(&ls->ls_waiters_mutex); 746 if (lkb->lkb_wait_type) { 747 log_print("add_to_waiters error %d", lkb->lkb_wait_type); 748 goto out; 749 } 750 lkb->lkb_wait_type = mstype; 751 kref_get(&lkb->lkb_ref); 752 list_add(&lkb->lkb_wait_reply, &ls->ls_waiters); 753 out: 754 mutex_unlock(&ls->ls_waiters_mutex); 755 } 756 757 static int _remove_from_waiters(struct dlm_lkb *lkb) 758 { 759 int error = 0; 760 761 if (!lkb->lkb_wait_type) { 762 log_print("remove_from_waiters error"); 763 error = -EINVAL; 764 goto out; 765 } 766 lkb->lkb_wait_type = 0; 767 list_del(&lkb->lkb_wait_reply); 768 unhold_lkb(lkb); 769 out: 770 return error; 771 } 772 773 static int remove_from_waiters(struct dlm_lkb *lkb) 774 { 775 struct dlm_ls *ls = lkb->lkb_resource->res_ls; 776 int error; 777 778 mutex_lock(&ls->ls_waiters_mutex); 779 error = _remove_from_waiters(lkb); 780 mutex_unlock(&ls->ls_waiters_mutex); 781 return error; 782 } 783 784 static void dir_remove(struct dlm_rsb *r) 785 { 786 int to_nodeid; 787 788 if (dlm_no_directory(r->res_ls)) 789 return; 790 791 to_nodeid = dlm_dir_nodeid(r); 792 if (to_nodeid != dlm_our_nodeid()) 793 send_remove(r); 794 else 795 dlm_dir_remove_entry(r->res_ls, to_nodeid, 796 r->res_name, r->res_length); 797 } 798 799 /* FIXME: shouldn't this be able to exit as soon as one non-due rsb is 800 found since they are in order of newest to oldest? */ 801 802 static int shrink_bucket(struct dlm_ls *ls, int b) 803 { 804 struct dlm_rsb *r; 805 int count = 0, found; 806 807 for (;;) { 808 found = 0; 809 write_lock(&ls->ls_rsbtbl[b].lock); 810 list_for_each_entry_reverse(r, &ls->ls_rsbtbl[b].toss, 811 res_hashchain) { 812 if (!time_after_eq(jiffies, r->res_toss_time + 813 dlm_config.toss_secs * HZ)) 814 continue; 815 found = 1; 816 break; 817 } 818 819 if (!found) { 820 write_unlock(&ls->ls_rsbtbl[b].lock); 821 break; 822 } 823 824 if (kref_put(&r->res_ref, kill_rsb)) { 825 list_del(&r->res_hashchain); 826 write_unlock(&ls->ls_rsbtbl[b].lock); 827 828 if (is_master(r)) 829 dir_remove(r); 830 free_rsb(r); 831 count++; 832 } else { 833 write_unlock(&ls->ls_rsbtbl[b].lock); 834 log_error(ls, "tossed rsb in use %s", r->res_name); 835 } 836 } 837 838 return count; 839 } 840 841 void dlm_scan_rsbs(struct dlm_ls *ls) 842 { 843 int i; 844 845 if (dlm_locking_stopped(ls)) 846 return; 847 848 for (i = 0; i < ls->ls_rsbtbl_size; i++) { 849 shrink_bucket(ls, i); 850 cond_resched(); 851 } 852 } 853 854 /* lkb is master or local copy */ 855 856 static void set_lvb_lock(struct dlm_rsb *r, struct dlm_lkb *lkb) 857 { 858 int b, len = r->res_ls->ls_lvblen; 859 860 /* b=1 lvb returned to caller 861 b=0 lvb written to rsb or invalidated 862 b=-1 do nothing */ 863 864 b = dlm_lvb_operations[lkb->lkb_grmode + 1][lkb->lkb_rqmode + 1]; 865 866 if (b == 1) { 867 if (!lkb->lkb_lvbptr) 868 return; 869 870 if (!(lkb->lkb_exflags & DLM_LKF_VALBLK)) 871 return; 872 873 if (!r->res_lvbptr) 874 return; 875 876 memcpy(lkb->lkb_lvbptr, r->res_lvbptr, len); 877 lkb->lkb_lvbseq = r->res_lvbseq; 878 879 } else if (b == 0) { 880 if (lkb->lkb_exflags & DLM_LKF_IVVALBLK) { 881 rsb_set_flag(r, RSB_VALNOTVALID); 882 return; 883 } 884 885 if (!lkb->lkb_lvbptr) 886 return; 887 888 if (!(lkb->lkb_exflags & DLM_LKF_VALBLK)) 889 return; 890 891 if (!r->res_lvbptr) 892 r->res_lvbptr = allocate_lvb(r->res_ls); 893 894 if (!r->res_lvbptr) 895 return; 896 897 memcpy(r->res_lvbptr, lkb->lkb_lvbptr, len); 898 r->res_lvbseq++; 899 lkb->lkb_lvbseq = r->res_lvbseq; 900 rsb_clear_flag(r, RSB_VALNOTVALID); 901 } 902 903 if (rsb_flag(r, RSB_VALNOTVALID)) 904 lkb->lkb_sbflags |= DLM_SBF_VALNOTVALID; 905 } 906 907 static void set_lvb_unlock(struct dlm_rsb *r, struct dlm_lkb *lkb) 908 { 909 if (lkb->lkb_grmode < DLM_LOCK_PW) 910 return; 911 912 if (lkb->lkb_exflags & DLM_LKF_IVVALBLK) { 913 rsb_set_flag(r, RSB_VALNOTVALID); 914 return; 915 } 916 917 if (!lkb->lkb_lvbptr) 918 return; 919 920 if (!(lkb->lkb_exflags & DLM_LKF_VALBLK)) 921 return; 922 923 if (!r->res_lvbptr) 924 r->res_lvbptr = allocate_lvb(r->res_ls); 925 926 if (!r->res_lvbptr) 927 return; 928 929 memcpy(r->res_lvbptr, lkb->lkb_lvbptr, r->res_ls->ls_lvblen); 930 r->res_lvbseq++; 931 rsb_clear_flag(r, RSB_VALNOTVALID); 932 } 933 934 /* lkb is process copy (pc) */ 935 936 static void set_lvb_lock_pc(struct dlm_rsb *r, struct dlm_lkb *lkb, 937 struct dlm_message *ms) 938 { 939 int b; 940 941 if (!lkb->lkb_lvbptr) 942 return; 943 944 if (!(lkb->lkb_exflags & DLM_LKF_VALBLK)) 945 return; 946 947 b = dlm_lvb_operations[lkb->lkb_grmode + 1][lkb->lkb_rqmode + 1]; 948 if (b == 1) { 949 int len = receive_extralen(ms); 950 memcpy(lkb->lkb_lvbptr, ms->m_extra, len); 951 lkb->lkb_lvbseq = ms->m_lvbseq; 952 } 953 } 954 955 /* Manipulate lkb's on rsb's convert/granted/waiting queues 956 remove_lock -- used for unlock, removes lkb from granted 957 revert_lock -- used for cancel, moves lkb from convert to granted 958 grant_lock -- used for request and convert, adds lkb to granted or 959 moves lkb from convert or waiting to granted 960 961 Each of these is used for master or local copy lkb's. There is 962 also a _pc() variation used to make the corresponding change on 963 a process copy (pc) lkb. */ 964 965 static void _remove_lock(struct dlm_rsb *r, struct dlm_lkb *lkb) 966 { 967 del_lkb(r, lkb); 968 lkb->lkb_grmode = DLM_LOCK_IV; 969 /* this unhold undoes the original ref from create_lkb() 970 so this leads to the lkb being freed */ 971 unhold_lkb(lkb); 972 } 973 974 static void remove_lock(struct dlm_rsb *r, struct dlm_lkb *lkb) 975 { 976 set_lvb_unlock(r, lkb); 977 _remove_lock(r, lkb); 978 } 979 980 static void remove_lock_pc(struct dlm_rsb *r, struct dlm_lkb *lkb) 981 { 982 _remove_lock(r, lkb); 983 } 984 985 static void revert_lock(struct dlm_rsb *r, struct dlm_lkb *lkb) 986 { 987 lkb->lkb_rqmode = DLM_LOCK_IV; 988 989 switch (lkb->lkb_status) { 990 case DLM_LKSTS_GRANTED: 991 break; 992 case DLM_LKSTS_CONVERT: 993 move_lkb(r, lkb, DLM_LKSTS_GRANTED); 994 break; 995 case DLM_LKSTS_WAITING: 996 del_lkb(r, lkb); 997 lkb->lkb_grmode = DLM_LOCK_IV; 998 /* this unhold undoes the original ref from create_lkb() 999 so this leads to the lkb being freed */ 1000 unhold_lkb(lkb); 1001 break; 1002 default: 1003 log_print("invalid status for revert %d", lkb->lkb_status); 1004 } 1005 } 1006 1007 static void revert_lock_pc(struct dlm_rsb *r, struct dlm_lkb *lkb) 1008 { 1009 revert_lock(r, lkb); 1010 } 1011 1012 static void _grant_lock(struct dlm_rsb *r, struct dlm_lkb *lkb) 1013 { 1014 if (lkb->lkb_grmode != lkb->lkb_rqmode) { 1015 lkb->lkb_grmode = lkb->lkb_rqmode; 1016 if (lkb->lkb_status) 1017 move_lkb(r, lkb, DLM_LKSTS_GRANTED); 1018 else 1019 add_lkb(r, lkb, DLM_LKSTS_GRANTED); 1020 } 1021 1022 lkb->lkb_rqmode = DLM_LOCK_IV; 1023 } 1024 1025 static void grant_lock(struct dlm_rsb *r, struct dlm_lkb *lkb) 1026 { 1027 set_lvb_lock(r, lkb); 1028 _grant_lock(r, lkb); 1029 lkb->lkb_highbast = 0; 1030 } 1031 1032 static void grant_lock_pc(struct dlm_rsb *r, struct dlm_lkb *lkb, 1033 struct dlm_message *ms) 1034 { 1035 set_lvb_lock_pc(r, lkb, ms); 1036 _grant_lock(r, lkb); 1037 } 1038 1039 /* called by grant_pending_locks() which means an async grant message must 1040 be sent to the requesting node in addition to granting the lock if the 1041 lkb belongs to a remote node. */ 1042 1043 static void grant_lock_pending(struct dlm_rsb *r, struct dlm_lkb *lkb) 1044 { 1045 grant_lock(r, lkb); 1046 if (is_master_copy(lkb)) 1047 send_grant(r, lkb); 1048 else 1049 queue_cast(r, lkb, 0); 1050 } 1051 1052 static inline int first_in_list(struct dlm_lkb *lkb, struct list_head *head) 1053 { 1054 struct dlm_lkb *first = list_entry(head->next, struct dlm_lkb, 1055 lkb_statequeue); 1056 if (lkb->lkb_id == first->lkb_id) 1057 return 1; 1058 1059 return 0; 1060 } 1061 1062 /* Check if the given lkb conflicts with another lkb on the queue. */ 1063 1064 static int queue_conflict(struct list_head *head, struct dlm_lkb *lkb) 1065 { 1066 struct dlm_lkb *this; 1067 1068 list_for_each_entry(this, head, lkb_statequeue) { 1069 if (this == lkb) 1070 continue; 1071 if (!modes_compat(this, lkb)) 1072 return 1; 1073 } 1074 return 0; 1075 } 1076 1077 /* 1078 * "A conversion deadlock arises with a pair of lock requests in the converting 1079 * queue for one resource. The granted mode of each lock blocks the requested 1080 * mode of the other lock." 1081 * 1082 * Part 2: if the granted mode of lkb is preventing the first lkb in the 1083 * convert queue from being granted, then demote lkb (set grmode to NL). 1084 * This second form requires that we check for conv-deadlk even when 1085 * now == 0 in _can_be_granted(). 1086 * 1087 * Example: 1088 * Granted Queue: empty 1089 * Convert Queue: NL->EX (first lock) 1090 * PR->EX (second lock) 1091 * 1092 * The first lock can't be granted because of the granted mode of the second 1093 * lock and the second lock can't be granted because it's not first in the 1094 * list. We demote the granted mode of the second lock (the lkb passed to this 1095 * function). 1096 * 1097 * After the resolution, the "grant pending" function needs to go back and try 1098 * to grant locks on the convert queue again since the first lock can now be 1099 * granted. 1100 */ 1101 1102 static int conversion_deadlock_detect(struct dlm_rsb *rsb, struct dlm_lkb *lkb) 1103 { 1104 struct dlm_lkb *this, *first = NULL, *self = NULL; 1105 1106 list_for_each_entry(this, &rsb->res_convertqueue, lkb_statequeue) { 1107 if (!first) 1108 first = this; 1109 if (this == lkb) { 1110 self = lkb; 1111 continue; 1112 } 1113 1114 if (!modes_compat(this, lkb) && !modes_compat(lkb, this)) 1115 return 1; 1116 } 1117 1118 /* if lkb is on the convert queue and is preventing the first 1119 from being granted, then there's deadlock and we demote lkb. 1120 multiple converting locks may need to do this before the first 1121 converting lock can be granted. */ 1122 1123 if (self && self != first) { 1124 if (!modes_compat(lkb, first) && 1125 !queue_conflict(&rsb->res_grantqueue, first)) 1126 return 1; 1127 } 1128 1129 return 0; 1130 } 1131 1132 /* 1133 * Return 1 if the lock can be granted, 0 otherwise. 1134 * Also detect and resolve conversion deadlocks. 1135 * 1136 * lkb is the lock to be granted 1137 * 1138 * now is 1 if the function is being called in the context of the 1139 * immediate request, it is 0 if called later, after the lock has been 1140 * queued. 1141 * 1142 * References are from chapter 6 of "VAXcluster Principles" by Roy Davis 1143 */ 1144 1145 static int _can_be_granted(struct dlm_rsb *r, struct dlm_lkb *lkb, int now) 1146 { 1147 int8_t conv = (lkb->lkb_grmode != DLM_LOCK_IV); 1148 1149 /* 1150 * 6-10: Version 5.4 introduced an option to address the phenomenon of 1151 * a new request for a NL mode lock being blocked. 1152 * 1153 * 6-11: If the optional EXPEDITE flag is used with the new NL mode 1154 * request, then it would be granted. In essence, the use of this flag 1155 * tells the Lock Manager to expedite theis request by not considering 1156 * what may be in the CONVERTING or WAITING queues... As of this 1157 * writing, the EXPEDITE flag can be used only with new requests for NL 1158 * mode locks. This flag is not valid for conversion requests. 1159 * 1160 * A shortcut. Earlier checks return an error if EXPEDITE is used in a 1161 * conversion or used with a non-NL requested mode. We also know an 1162 * EXPEDITE request is always granted immediately, so now must always 1163 * be 1. The full condition to grant an expedite request: (now && 1164 * !conv && lkb->rqmode == DLM_LOCK_NL && (flags & EXPEDITE)) can 1165 * therefore be shortened to just checking the flag. 1166 */ 1167 1168 if (lkb->lkb_exflags & DLM_LKF_EXPEDITE) 1169 return 1; 1170 1171 /* 1172 * A shortcut. Without this, !queue_conflict(grantqueue, lkb) would be 1173 * added to the remaining conditions. 1174 */ 1175 1176 if (queue_conflict(&r->res_grantqueue, lkb)) 1177 goto out; 1178 1179 /* 1180 * 6-3: By default, a conversion request is immediately granted if the 1181 * requested mode is compatible with the modes of all other granted 1182 * locks 1183 */ 1184 1185 if (queue_conflict(&r->res_convertqueue, lkb)) 1186 goto out; 1187 1188 /* 1189 * 6-5: But the default algorithm for deciding whether to grant or 1190 * queue conversion requests does not by itself guarantee that such 1191 * requests are serviced on a "first come first serve" basis. This, in 1192 * turn, can lead to a phenomenon known as "indefinate postponement". 1193 * 1194 * 6-7: This issue is dealt with by using the optional QUECVT flag with 1195 * the system service employed to request a lock conversion. This flag 1196 * forces certain conversion requests to be queued, even if they are 1197 * compatible with the granted modes of other locks on the same 1198 * resource. Thus, the use of this flag results in conversion requests 1199 * being ordered on a "first come first servce" basis. 1200 * 1201 * DCT: This condition is all about new conversions being able to occur 1202 * "in place" while the lock remains on the granted queue (assuming 1203 * nothing else conflicts.) IOW if QUECVT isn't set, a conversion 1204 * doesn't _have_ to go onto the convert queue where it's processed in 1205 * order. The "now" variable is necessary to distinguish converts 1206 * being received and processed for the first time now, because once a 1207 * convert is moved to the conversion queue the condition below applies 1208 * requiring fifo granting. 1209 */ 1210 1211 if (now && conv && !(lkb->lkb_exflags & DLM_LKF_QUECVT)) 1212 return 1; 1213 1214 /* 1215 * The NOORDER flag is set to avoid the standard vms rules on grant 1216 * order. 1217 */ 1218 1219 if (lkb->lkb_exflags & DLM_LKF_NOORDER) 1220 return 1; 1221 1222 /* 1223 * 6-3: Once in that queue [CONVERTING], a conversion request cannot be 1224 * granted until all other conversion requests ahead of it are granted 1225 * and/or canceled. 1226 */ 1227 1228 if (!now && conv && first_in_list(lkb, &r->res_convertqueue)) 1229 return 1; 1230 1231 /* 1232 * 6-4: By default, a new request is immediately granted only if all 1233 * three of the following conditions are satisfied when the request is 1234 * issued: 1235 * - The queue of ungranted conversion requests for the resource is 1236 * empty. 1237 * - The queue of ungranted new requests for the resource is empty. 1238 * - The mode of the new request is compatible with the most 1239 * restrictive mode of all granted locks on the resource. 1240 */ 1241 1242 if (now && !conv && list_empty(&r->res_convertqueue) && 1243 list_empty(&r->res_waitqueue)) 1244 return 1; 1245 1246 /* 1247 * 6-4: Once a lock request is in the queue of ungranted new requests, 1248 * it cannot be granted until the queue of ungranted conversion 1249 * requests is empty, all ungranted new requests ahead of it are 1250 * granted and/or canceled, and it is compatible with the granted mode 1251 * of the most restrictive lock granted on the resource. 1252 */ 1253 1254 if (!now && !conv && list_empty(&r->res_convertqueue) && 1255 first_in_list(lkb, &r->res_waitqueue)) 1256 return 1; 1257 1258 out: 1259 /* 1260 * The following, enabled by CONVDEADLK, departs from VMS. 1261 */ 1262 1263 if (conv && (lkb->lkb_exflags & DLM_LKF_CONVDEADLK) && 1264 conversion_deadlock_detect(r, lkb)) { 1265 lkb->lkb_grmode = DLM_LOCK_NL; 1266 lkb->lkb_sbflags |= DLM_SBF_DEMOTED; 1267 } 1268 1269 return 0; 1270 } 1271 1272 /* 1273 * The ALTPR and ALTCW flags aren't traditional lock manager flags, but are a 1274 * simple way to provide a big optimization to applications that can use them. 1275 */ 1276 1277 static int can_be_granted(struct dlm_rsb *r, struct dlm_lkb *lkb, int now) 1278 { 1279 uint32_t flags = lkb->lkb_exflags; 1280 int rv; 1281 int8_t alt = 0, rqmode = lkb->lkb_rqmode; 1282 1283 rv = _can_be_granted(r, lkb, now); 1284 if (rv) 1285 goto out; 1286 1287 if (lkb->lkb_sbflags & DLM_SBF_DEMOTED) 1288 goto out; 1289 1290 if (rqmode != DLM_LOCK_PR && flags & DLM_LKF_ALTPR) 1291 alt = DLM_LOCK_PR; 1292 else if (rqmode != DLM_LOCK_CW && flags & DLM_LKF_ALTCW) 1293 alt = DLM_LOCK_CW; 1294 1295 if (alt) { 1296 lkb->lkb_rqmode = alt; 1297 rv = _can_be_granted(r, lkb, now); 1298 if (rv) 1299 lkb->lkb_sbflags |= DLM_SBF_ALTMODE; 1300 else 1301 lkb->lkb_rqmode = rqmode; 1302 } 1303 out: 1304 return rv; 1305 } 1306 1307 static int grant_pending_convert(struct dlm_rsb *r, int high) 1308 { 1309 struct dlm_lkb *lkb, *s; 1310 int hi, demoted, quit, grant_restart, demote_restart; 1311 1312 quit = 0; 1313 restart: 1314 grant_restart = 0; 1315 demote_restart = 0; 1316 hi = DLM_LOCK_IV; 1317 1318 list_for_each_entry_safe(lkb, s, &r->res_convertqueue, lkb_statequeue) { 1319 demoted = is_demoted(lkb); 1320 if (can_be_granted(r, lkb, 0)) { 1321 grant_lock_pending(r, lkb); 1322 grant_restart = 1; 1323 } else { 1324 hi = max_t(int, lkb->lkb_rqmode, hi); 1325 if (!demoted && is_demoted(lkb)) 1326 demote_restart = 1; 1327 } 1328 } 1329 1330 if (grant_restart) 1331 goto restart; 1332 if (demote_restart && !quit) { 1333 quit = 1; 1334 goto restart; 1335 } 1336 1337 return max_t(int, high, hi); 1338 } 1339 1340 static int grant_pending_wait(struct dlm_rsb *r, int high) 1341 { 1342 struct dlm_lkb *lkb, *s; 1343 1344 list_for_each_entry_safe(lkb, s, &r->res_waitqueue, lkb_statequeue) { 1345 if (can_be_granted(r, lkb, 0)) 1346 grant_lock_pending(r, lkb); 1347 else 1348 high = max_t(int, lkb->lkb_rqmode, high); 1349 } 1350 1351 return high; 1352 } 1353 1354 static void grant_pending_locks(struct dlm_rsb *r) 1355 { 1356 struct dlm_lkb *lkb, *s; 1357 int high = DLM_LOCK_IV; 1358 1359 DLM_ASSERT(is_master(r), dlm_dump_rsb(r);); 1360 1361 high = grant_pending_convert(r, high); 1362 high = grant_pending_wait(r, high); 1363 1364 if (high == DLM_LOCK_IV) 1365 return; 1366 1367 /* 1368 * If there are locks left on the wait/convert queue then send blocking 1369 * ASTs to granted locks based on the largest requested mode (high) 1370 * found above. FIXME: highbast < high comparison not valid for PR/CW. 1371 */ 1372 1373 list_for_each_entry_safe(lkb, s, &r->res_grantqueue, lkb_statequeue) { 1374 if (lkb->lkb_bastaddr && (lkb->lkb_highbast < high) && 1375 !__dlm_compat_matrix[lkb->lkb_grmode+1][high+1]) { 1376 queue_bast(r, lkb, high); 1377 lkb->lkb_highbast = high; 1378 } 1379 } 1380 } 1381 1382 static void send_bast_queue(struct dlm_rsb *r, struct list_head *head, 1383 struct dlm_lkb *lkb) 1384 { 1385 struct dlm_lkb *gr; 1386 1387 list_for_each_entry(gr, head, lkb_statequeue) { 1388 if (gr->lkb_bastaddr && 1389 gr->lkb_highbast < lkb->lkb_rqmode && 1390 !modes_compat(gr, lkb)) { 1391 queue_bast(r, gr, lkb->lkb_rqmode); 1392 gr->lkb_highbast = lkb->lkb_rqmode; 1393 } 1394 } 1395 } 1396 1397 static void send_blocking_asts(struct dlm_rsb *r, struct dlm_lkb *lkb) 1398 { 1399 send_bast_queue(r, &r->res_grantqueue, lkb); 1400 } 1401 1402 static void send_blocking_asts_all(struct dlm_rsb *r, struct dlm_lkb *lkb) 1403 { 1404 send_bast_queue(r, &r->res_grantqueue, lkb); 1405 send_bast_queue(r, &r->res_convertqueue, lkb); 1406 } 1407 1408 /* set_master(r, lkb) -- set the master nodeid of a resource 1409 1410 The purpose of this function is to set the nodeid field in the given 1411 lkb using the nodeid field in the given rsb. If the rsb's nodeid is 1412 known, it can just be copied to the lkb and the function will return 1413 0. If the rsb's nodeid is _not_ known, it needs to be looked up 1414 before it can be copied to the lkb. 1415 1416 When the rsb nodeid is being looked up remotely, the initial lkb 1417 causing the lookup is kept on the ls_waiters list waiting for the 1418 lookup reply. Other lkb's waiting for the same rsb lookup are kept 1419 on the rsb's res_lookup list until the master is verified. 1420 1421 Return values: 1422 0: nodeid is set in rsb/lkb and the caller should go ahead and use it 1423 1: the rsb master is not available and the lkb has been placed on 1424 a wait queue 1425 */ 1426 1427 static int set_master(struct dlm_rsb *r, struct dlm_lkb *lkb) 1428 { 1429 struct dlm_ls *ls = r->res_ls; 1430 int error, dir_nodeid, ret_nodeid, our_nodeid = dlm_our_nodeid(); 1431 1432 if (rsb_flag(r, RSB_MASTER_UNCERTAIN)) { 1433 rsb_clear_flag(r, RSB_MASTER_UNCERTAIN); 1434 r->res_first_lkid = lkb->lkb_id; 1435 lkb->lkb_nodeid = r->res_nodeid; 1436 return 0; 1437 } 1438 1439 if (r->res_first_lkid && r->res_first_lkid != lkb->lkb_id) { 1440 list_add_tail(&lkb->lkb_rsb_lookup, &r->res_lookup); 1441 return 1; 1442 } 1443 1444 if (r->res_nodeid == 0) { 1445 lkb->lkb_nodeid = 0; 1446 return 0; 1447 } 1448 1449 if (r->res_nodeid > 0) { 1450 lkb->lkb_nodeid = r->res_nodeid; 1451 return 0; 1452 } 1453 1454 DLM_ASSERT(r->res_nodeid == -1, dlm_dump_rsb(r);); 1455 1456 dir_nodeid = dlm_dir_nodeid(r); 1457 1458 if (dir_nodeid != our_nodeid) { 1459 r->res_first_lkid = lkb->lkb_id; 1460 send_lookup(r, lkb); 1461 return 1; 1462 } 1463 1464 for (;;) { 1465 /* It's possible for dlm_scand to remove an old rsb for 1466 this same resource from the toss list, us to create 1467 a new one, look up the master locally, and find it 1468 already exists just before dlm_scand does the 1469 dir_remove() on the previous rsb. */ 1470 1471 error = dlm_dir_lookup(ls, our_nodeid, r->res_name, 1472 r->res_length, &ret_nodeid); 1473 if (!error) 1474 break; 1475 log_debug(ls, "dir_lookup error %d %s", error, r->res_name); 1476 schedule(); 1477 } 1478 1479 if (ret_nodeid == our_nodeid) { 1480 r->res_first_lkid = 0; 1481 r->res_nodeid = 0; 1482 lkb->lkb_nodeid = 0; 1483 } else { 1484 r->res_first_lkid = lkb->lkb_id; 1485 r->res_nodeid = ret_nodeid; 1486 lkb->lkb_nodeid = ret_nodeid; 1487 } 1488 return 0; 1489 } 1490 1491 static void process_lookup_list(struct dlm_rsb *r) 1492 { 1493 struct dlm_lkb *lkb, *safe; 1494 1495 list_for_each_entry_safe(lkb, safe, &r->res_lookup, lkb_rsb_lookup) { 1496 list_del(&lkb->lkb_rsb_lookup); 1497 _request_lock(r, lkb); 1498 schedule(); 1499 } 1500 } 1501 1502 /* confirm_master -- confirm (or deny) an rsb's master nodeid */ 1503 1504 static void confirm_master(struct dlm_rsb *r, int error) 1505 { 1506 struct dlm_lkb *lkb; 1507 1508 if (!r->res_first_lkid) 1509 return; 1510 1511 switch (error) { 1512 case 0: 1513 case -EINPROGRESS: 1514 r->res_first_lkid = 0; 1515 process_lookup_list(r); 1516 break; 1517 1518 case -EAGAIN: 1519 /* the remote master didn't queue our NOQUEUE request; 1520 make a waiting lkb the first_lkid */ 1521 1522 r->res_first_lkid = 0; 1523 1524 if (!list_empty(&r->res_lookup)) { 1525 lkb = list_entry(r->res_lookup.next, struct dlm_lkb, 1526 lkb_rsb_lookup); 1527 list_del(&lkb->lkb_rsb_lookup); 1528 r->res_first_lkid = lkb->lkb_id; 1529 _request_lock(r, lkb); 1530 } else 1531 r->res_nodeid = -1; 1532 break; 1533 1534 default: 1535 log_error(r->res_ls, "confirm_master unknown error %d", error); 1536 } 1537 } 1538 1539 static int set_lock_args(int mode, struct dlm_lksb *lksb, uint32_t flags, 1540 int namelen, uint32_t parent_lkid, void *ast, 1541 void *astarg, void *bast, struct dlm_args *args) 1542 { 1543 int rv = -EINVAL; 1544 1545 /* check for invalid arg usage */ 1546 1547 if (mode < 0 || mode > DLM_LOCK_EX) 1548 goto out; 1549 1550 if (!(flags & DLM_LKF_CONVERT) && (namelen > DLM_RESNAME_MAXLEN)) 1551 goto out; 1552 1553 if (flags & DLM_LKF_CANCEL) 1554 goto out; 1555 1556 if (flags & DLM_LKF_QUECVT && !(flags & DLM_LKF_CONVERT)) 1557 goto out; 1558 1559 if (flags & DLM_LKF_CONVDEADLK && !(flags & DLM_LKF_CONVERT)) 1560 goto out; 1561 1562 if (flags & DLM_LKF_CONVDEADLK && flags & DLM_LKF_NOQUEUE) 1563 goto out; 1564 1565 if (flags & DLM_LKF_EXPEDITE && flags & DLM_LKF_CONVERT) 1566 goto out; 1567 1568 if (flags & DLM_LKF_EXPEDITE && flags & DLM_LKF_QUECVT) 1569 goto out; 1570 1571 if (flags & DLM_LKF_EXPEDITE && flags & DLM_LKF_NOQUEUE) 1572 goto out; 1573 1574 if (flags & DLM_LKF_EXPEDITE && mode != DLM_LOCK_NL) 1575 goto out; 1576 1577 if (!ast || !lksb) 1578 goto out; 1579 1580 if (flags & DLM_LKF_VALBLK && !lksb->sb_lvbptr) 1581 goto out; 1582 1583 /* parent/child locks not yet supported */ 1584 if (parent_lkid) 1585 goto out; 1586 1587 if (flags & DLM_LKF_CONVERT && !lksb->sb_lkid) 1588 goto out; 1589 1590 /* these args will be copied to the lkb in validate_lock_args, 1591 it cannot be done now because when converting locks, fields in 1592 an active lkb cannot be modified before locking the rsb */ 1593 1594 args->flags = flags; 1595 args->astaddr = ast; 1596 args->astparam = (long) astarg; 1597 args->bastaddr = bast; 1598 args->mode = mode; 1599 args->lksb = lksb; 1600 rv = 0; 1601 out: 1602 return rv; 1603 } 1604 1605 static int set_unlock_args(uint32_t flags, void *astarg, struct dlm_args *args) 1606 { 1607 if (flags & ~(DLM_LKF_CANCEL | DLM_LKF_VALBLK | DLM_LKF_IVVALBLK | 1608 DLM_LKF_FORCEUNLOCK)) 1609 return -EINVAL; 1610 1611 args->flags = flags; 1612 args->astparam = (long) astarg; 1613 return 0; 1614 } 1615 1616 static int validate_lock_args(struct dlm_ls *ls, struct dlm_lkb *lkb, 1617 struct dlm_args *args) 1618 { 1619 int rv = -EINVAL; 1620 1621 if (args->flags & DLM_LKF_CONVERT) { 1622 if (lkb->lkb_flags & DLM_IFL_MSTCPY) 1623 goto out; 1624 1625 if (args->flags & DLM_LKF_QUECVT && 1626 !__quecvt_compat_matrix[lkb->lkb_grmode+1][args->mode+1]) 1627 goto out; 1628 1629 rv = -EBUSY; 1630 if (lkb->lkb_status != DLM_LKSTS_GRANTED) 1631 goto out; 1632 1633 if (lkb->lkb_wait_type) 1634 goto out; 1635 } 1636 1637 lkb->lkb_exflags = args->flags; 1638 lkb->lkb_sbflags = 0; 1639 lkb->lkb_astaddr = args->astaddr; 1640 lkb->lkb_astparam = args->astparam; 1641 lkb->lkb_bastaddr = args->bastaddr; 1642 lkb->lkb_rqmode = args->mode; 1643 lkb->lkb_lksb = args->lksb; 1644 lkb->lkb_lvbptr = args->lksb->sb_lvbptr; 1645 lkb->lkb_ownpid = (int) current->pid; 1646 rv = 0; 1647 out: 1648 return rv; 1649 } 1650 1651 static int validate_unlock_args(struct dlm_lkb *lkb, struct dlm_args *args) 1652 { 1653 int rv = -EINVAL; 1654 1655 if (lkb->lkb_flags & DLM_IFL_MSTCPY) 1656 goto out; 1657 1658 if (args->flags & DLM_LKF_FORCEUNLOCK) 1659 goto out_ok; 1660 1661 if (args->flags & DLM_LKF_CANCEL && 1662 lkb->lkb_status == DLM_LKSTS_GRANTED) 1663 goto out; 1664 1665 if (!(args->flags & DLM_LKF_CANCEL) && 1666 lkb->lkb_status != DLM_LKSTS_GRANTED) 1667 goto out; 1668 1669 rv = -EBUSY; 1670 if (lkb->lkb_wait_type) 1671 goto out; 1672 1673 out_ok: 1674 lkb->lkb_exflags = args->flags; 1675 lkb->lkb_sbflags = 0; 1676 lkb->lkb_astparam = args->astparam; 1677 1678 rv = 0; 1679 out: 1680 return rv; 1681 } 1682 1683 /* 1684 * Four stage 4 varieties: 1685 * do_request(), do_convert(), do_unlock(), do_cancel() 1686 * These are called on the master node for the given lock and 1687 * from the central locking logic. 1688 */ 1689 1690 static int do_request(struct dlm_rsb *r, struct dlm_lkb *lkb) 1691 { 1692 int error = 0; 1693 1694 if (can_be_granted(r, lkb, 1)) { 1695 grant_lock(r, lkb); 1696 queue_cast(r, lkb, 0); 1697 goto out; 1698 } 1699 1700 if (can_be_queued(lkb)) { 1701 error = -EINPROGRESS; 1702 add_lkb(r, lkb, DLM_LKSTS_WAITING); 1703 send_blocking_asts(r, lkb); 1704 goto out; 1705 } 1706 1707 error = -EAGAIN; 1708 if (force_blocking_asts(lkb)) 1709 send_blocking_asts_all(r, lkb); 1710 queue_cast(r, lkb, -EAGAIN); 1711 1712 out: 1713 return error; 1714 } 1715 1716 static int do_convert(struct dlm_rsb *r, struct dlm_lkb *lkb) 1717 { 1718 int error = 0; 1719 1720 /* changing an existing lock may allow others to be granted */ 1721 1722 if (can_be_granted(r, lkb, 1)) { 1723 grant_lock(r, lkb); 1724 queue_cast(r, lkb, 0); 1725 grant_pending_locks(r); 1726 goto out; 1727 } 1728 1729 if (can_be_queued(lkb)) { 1730 if (is_demoted(lkb)) 1731 grant_pending_locks(r); 1732 error = -EINPROGRESS; 1733 del_lkb(r, lkb); 1734 add_lkb(r, lkb, DLM_LKSTS_CONVERT); 1735 send_blocking_asts(r, lkb); 1736 goto out; 1737 } 1738 1739 error = -EAGAIN; 1740 if (force_blocking_asts(lkb)) 1741 send_blocking_asts_all(r, lkb); 1742 queue_cast(r, lkb, -EAGAIN); 1743 1744 out: 1745 return error; 1746 } 1747 1748 static int do_unlock(struct dlm_rsb *r, struct dlm_lkb *lkb) 1749 { 1750 remove_lock(r, lkb); 1751 queue_cast(r, lkb, -DLM_EUNLOCK); 1752 grant_pending_locks(r); 1753 return -DLM_EUNLOCK; 1754 } 1755 1756 /* FIXME: if revert_lock() finds that the lkb is granted, we should 1757 skip the queue_cast(ECANCEL). It indicates that the request/convert 1758 completed (and queued a normal ast) just before the cancel; we don't 1759 want to clobber the sb_result for the normal ast with ECANCEL. */ 1760 1761 static int do_cancel(struct dlm_rsb *r, struct dlm_lkb *lkb) 1762 { 1763 revert_lock(r, lkb); 1764 queue_cast(r, lkb, -DLM_ECANCEL); 1765 grant_pending_locks(r); 1766 return -DLM_ECANCEL; 1767 } 1768 1769 /* 1770 * Four stage 3 varieties: 1771 * _request_lock(), _convert_lock(), _unlock_lock(), _cancel_lock() 1772 */ 1773 1774 /* add a new lkb to a possibly new rsb, called by requesting process */ 1775 1776 static int _request_lock(struct dlm_rsb *r, struct dlm_lkb *lkb) 1777 { 1778 int error; 1779 1780 /* set_master: sets lkb nodeid from r */ 1781 1782 error = set_master(r, lkb); 1783 if (error < 0) 1784 goto out; 1785 if (error) { 1786 error = 0; 1787 goto out; 1788 } 1789 1790 if (is_remote(r)) 1791 /* receive_request() calls do_request() on remote node */ 1792 error = send_request(r, lkb); 1793 else 1794 error = do_request(r, lkb); 1795 out: 1796 return error; 1797 } 1798 1799 /* change some property of an existing lkb, e.g. mode */ 1800 1801 static int _convert_lock(struct dlm_rsb *r, struct dlm_lkb *lkb) 1802 { 1803 int error; 1804 1805 if (is_remote(r)) 1806 /* receive_convert() calls do_convert() on remote node */ 1807 error = send_convert(r, lkb); 1808 else 1809 error = do_convert(r, lkb); 1810 1811 return error; 1812 } 1813 1814 /* remove an existing lkb from the granted queue */ 1815 1816 static int _unlock_lock(struct dlm_rsb *r, struct dlm_lkb *lkb) 1817 { 1818 int error; 1819 1820 if (is_remote(r)) 1821 /* receive_unlock() calls do_unlock() on remote node */ 1822 error = send_unlock(r, lkb); 1823 else 1824 error = do_unlock(r, lkb); 1825 1826 return error; 1827 } 1828 1829 /* remove an existing lkb from the convert or wait queue */ 1830 1831 static int _cancel_lock(struct dlm_rsb *r, struct dlm_lkb *lkb) 1832 { 1833 int error; 1834 1835 if (is_remote(r)) 1836 /* receive_cancel() calls do_cancel() on remote node */ 1837 error = send_cancel(r, lkb); 1838 else 1839 error = do_cancel(r, lkb); 1840 1841 return error; 1842 } 1843 1844 /* 1845 * Four stage 2 varieties: 1846 * request_lock(), convert_lock(), unlock_lock(), cancel_lock() 1847 */ 1848 1849 static int request_lock(struct dlm_ls *ls, struct dlm_lkb *lkb, char *name, 1850 int len, struct dlm_args *args) 1851 { 1852 struct dlm_rsb *r; 1853 int error; 1854 1855 error = validate_lock_args(ls, lkb, args); 1856 if (error) 1857 goto out; 1858 1859 error = find_rsb(ls, name, len, R_CREATE, &r); 1860 if (error) 1861 goto out; 1862 1863 lock_rsb(r); 1864 1865 attach_lkb(r, lkb); 1866 lkb->lkb_lksb->sb_lkid = lkb->lkb_id; 1867 1868 error = _request_lock(r, lkb); 1869 1870 unlock_rsb(r); 1871 put_rsb(r); 1872 1873 out: 1874 return error; 1875 } 1876 1877 static int convert_lock(struct dlm_ls *ls, struct dlm_lkb *lkb, 1878 struct dlm_args *args) 1879 { 1880 struct dlm_rsb *r; 1881 int error; 1882 1883 r = lkb->lkb_resource; 1884 1885 hold_rsb(r); 1886 lock_rsb(r); 1887 1888 error = validate_lock_args(ls, lkb, args); 1889 if (error) 1890 goto out; 1891 1892 error = _convert_lock(r, lkb); 1893 out: 1894 unlock_rsb(r); 1895 put_rsb(r); 1896 return error; 1897 } 1898 1899 static int unlock_lock(struct dlm_ls *ls, struct dlm_lkb *lkb, 1900 struct dlm_args *args) 1901 { 1902 struct dlm_rsb *r; 1903 int error; 1904 1905 r = lkb->lkb_resource; 1906 1907 hold_rsb(r); 1908 lock_rsb(r); 1909 1910 error = validate_unlock_args(lkb, args); 1911 if (error) 1912 goto out; 1913 1914 error = _unlock_lock(r, lkb); 1915 out: 1916 unlock_rsb(r); 1917 put_rsb(r); 1918 return error; 1919 } 1920 1921 static int cancel_lock(struct dlm_ls *ls, struct dlm_lkb *lkb, 1922 struct dlm_args *args) 1923 { 1924 struct dlm_rsb *r; 1925 int error; 1926 1927 r = lkb->lkb_resource; 1928 1929 hold_rsb(r); 1930 lock_rsb(r); 1931 1932 error = validate_unlock_args(lkb, args); 1933 if (error) 1934 goto out; 1935 1936 error = _cancel_lock(r, lkb); 1937 out: 1938 unlock_rsb(r); 1939 put_rsb(r); 1940 return error; 1941 } 1942 1943 /* 1944 * Two stage 1 varieties: dlm_lock() and dlm_unlock() 1945 */ 1946 1947 int dlm_lock(dlm_lockspace_t *lockspace, 1948 int mode, 1949 struct dlm_lksb *lksb, 1950 uint32_t flags, 1951 void *name, 1952 unsigned int namelen, 1953 uint32_t parent_lkid, 1954 void (*ast) (void *astarg), 1955 void *astarg, 1956 void (*bast) (void *astarg, int mode)) 1957 { 1958 struct dlm_ls *ls; 1959 struct dlm_lkb *lkb; 1960 struct dlm_args args; 1961 int error, convert = flags & DLM_LKF_CONVERT; 1962 1963 ls = dlm_find_lockspace_local(lockspace); 1964 if (!ls) 1965 return -EINVAL; 1966 1967 lock_recovery(ls); 1968 1969 if (convert) 1970 error = find_lkb(ls, lksb->sb_lkid, &lkb); 1971 else 1972 error = create_lkb(ls, &lkb); 1973 1974 if (error) 1975 goto out; 1976 1977 error = set_lock_args(mode, lksb, flags, namelen, parent_lkid, ast, 1978 astarg, bast, &args); 1979 if (error) 1980 goto out_put; 1981 1982 if (convert) 1983 error = convert_lock(ls, lkb, &args); 1984 else 1985 error = request_lock(ls, lkb, name, namelen, &args); 1986 1987 if (error == -EINPROGRESS) 1988 error = 0; 1989 out_put: 1990 if (convert || error) 1991 __put_lkb(ls, lkb); 1992 if (error == -EAGAIN) 1993 error = 0; 1994 out: 1995 unlock_recovery(ls); 1996 dlm_put_lockspace(ls); 1997 return error; 1998 } 1999 2000 int dlm_unlock(dlm_lockspace_t *lockspace, 2001 uint32_t lkid, 2002 uint32_t flags, 2003 struct dlm_lksb *lksb, 2004 void *astarg) 2005 { 2006 struct dlm_ls *ls; 2007 struct dlm_lkb *lkb; 2008 struct dlm_args args; 2009 int error; 2010 2011 ls = dlm_find_lockspace_local(lockspace); 2012 if (!ls) 2013 return -EINVAL; 2014 2015 lock_recovery(ls); 2016 2017 error = find_lkb(ls, lkid, &lkb); 2018 if (error) 2019 goto out; 2020 2021 error = set_unlock_args(flags, astarg, &args); 2022 if (error) 2023 goto out_put; 2024 2025 if (flags & DLM_LKF_CANCEL) 2026 error = cancel_lock(ls, lkb, &args); 2027 else 2028 error = unlock_lock(ls, lkb, &args); 2029 2030 if (error == -DLM_EUNLOCK || error == -DLM_ECANCEL) 2031 error = 0; 2032 out_put: 2033 dlm_put_lkb(lkb); 2034 out: 2035 unlock_recovery(ls); 2036 dlm_put_lockspace(ls); 2037 return error; 2038 } 2039 2040 /* 2041 * send/receive routines for remote operations and replies 2042 * 2043 * send_args 2044 * send_common 2045 * send_request receive_request 2046 * send_convert receive_convert 2047 * send_unlock receive_unlock 2048 * send_cancel receive_cancel 2049 * send_grant receive_grant 2050 * send_bast receive_bast 2051 * send_lookup receive_lookup 2052 * send_remove receive_remove 2053 * 2054 * send_common_reply 2055 * receive_request_reply send_request_reply 2056 * receive_convert_reply send_convert_reply 2057 * receive_unlock_reply send_unlock_reply 2058 * receive_cancel_reply send_cancel_reply 2059 * receive_lookup_reply send_lookup_reply 2060 */ 2061 2062 static int create_message(struct dlm_rsb *r, struct dlm_lkb *lkb, 2063 int to_nodeid, int mstype, 2064 struct dlm_message **ms_ret, 2065 struct dlm_mhandle **mh_ret) 2066 { 2067 struct dlm_message *ms; 2068 struct dlm_mhandle *mh; 2069 char *mb; 2070 int mb_len = sizeof(struct dlm_message); 2071 2072 switch (mstype) { 2073 case DLM_MSG_REQUEST: 2074 case DLM_MSG_LOOKUP: 2075 case DLM_MSG_REMOVE: 2076 mb_len += r->res_length; 2077 break; 2078 case DLM_MSG_CONVERT: 2079 case DLM_MSG_UNLOCK: 2080 case DLM_MSG_REQUEST_REPLY: 2081 case DLM_MSG_CONVERT_REPLY: 2082 case DLM_MSG_GRANT: 2083 if (lkb && lkb->lkb_lvbptr) 2084 mb_len += r->res_ls->ls_lvblen; 2085 break; 2086 } 2087 2088 /* get_buffer gives us a message handle (mh) that we need to 2089 pass into lowcomms_commit and a message buffer (mb) that we 2090 write our data into */ 2091 2092 mh = dlm_lowcomms_get_buffer(to_nodeid, mb_len, GFP_KERNEL, &mb); 2093 if (!mh) 2094 return -ENOBUFS; 2095 2096 memset(mb, 0, mb_len); 2097 2098 ms = (struct dlm_message *) mb; 2099 2100 ms->m_header.h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR); 2101 ms->m_header.h_lockspace = r->res_ls->ls_global_id; 2102 ms->m_header.h_nodeid = dlm_our_nodeid(); 2103 ms->m_header.h_length = mb_len; 2104 ms->m_header.h_cmd = DLM_MSG; 2105 2106 ms->m_type = mstype; 2107 2108 *mh_ret = mh; 2109 *ms_ret = ms; 2110 return 0; 2111 } 2112 2113 /* further lowcomms enhancements or alternate implementations may make 2114 the return value from this function useful at some point */ 2115 2116 static int send_message(struct dlm_mhandle *mh, struct dlm_message *ms) 2117 { 2118 dlm_message_out(ms); 2119 dlm_lowcomms_commit_buffer(mh); 2120 return 0; 2121 } 2122 2123 static void send_args(struct dlm_rsb *r, struct dlm_lkb *lkb, 2124 struct dlm_message *ms) 2125 { 2126 ms->m_nodeid = lkb->lkb_nodeid; 2127 ms->m_pid = lkb->lkb_ownpid; 2128 ms->m_lkid = lkb->lkb_id; 2129 ms->m_remid = lkb->lkb_remid; 2130 ms->m_exflags = lkb->lkb_exflags; 2131 ms->m_sbflags = lkb->lkb_sbflags; 2132 ms->m_flags = lkb->lkb_flags; 2133 ms->m_lvbseq = lkb->lkb_lvbseq; 2134 ms->m_status = lkb->lkb_status; 2135 ms->m_grmode = lkb->lkb_grmode; 2136 ms->m_rqmode = lkb->lkb_rqmode; 2137 ms->m_hash = r->res_hash; 2138 2139 /* m_result and m_bastmode are set from function args, 2140 not from lkb fields */ 2141 2142 if (lkb->lkb_bastaddr) 2143 ms->m_asts |= AST_BAST; 2144 if (lkb->lkb_astaddr) 2145 ms->m_asts |= AST_COMP; 2146 2147 if (ms->m_type == DLM_MSG_REQUEST || ms->m_type == DLM_MSG_LOOKUP) 2148 memcpy(ms->m_extra, r->res_name, r->res_length); 2149 2150 else if (lkb->lkb_lvbptr) 2151 memcpy(ms->m_extra, lkb->lkb_lvbptr, r->res_ls->ls_lvblen); 2152 2153 } 2154 2155 static int send_common(struct dlm_rsb *r, struct dlm_lkb *lkb, int mstype) 2156 { 2157 struct dlm_message *ms; 2158 struct dlm_mhandle *mh; 2159 int to_nodeid, error; 2160 2161 add_to_waiters(lkb, mstype); 2162 2163 to_nodeid = r->res_nodeid; 2164 2165 error = create_message(r, lkb, to_nodeid, mstype, &ms, &mh); 2166 if (error) 2167 goto fail; 2168 2169 send_args(r, lkb, ms); 2170 2171 error = send_message(mh, ms); 2172 if (error) 2173 goto fail; 2174 return 0; 2175 2176 fail: 2177 remove_from_waiters(lkb); 2178 return error; 2179 } 2180 2181 static int send_request(struct dlm_rsb *r, struct dlm_lkb *lkb) 2182 { 2183 return send_common(r, lkb, DLM_MSG_REQUEST); 2184 } 2185 2186 static int send_convert(struct dlm_rsb *r, struct dlm_lkb *lkb) 2187 { 2188 int error; 2189 2190 error = send_common(r, lkb, DLM_MSG_CONVERT); 2191 2192 /* down conversions go without a reply from the master */ 2193 if (!error && down_conversion(lkb)) { 2194 remove_from_waiters(lkb); 2195 r->res_ls->ls_stub_ms.m_result = 0; 2196 r->res_ls->ls_stub_ms.m_flags = lkb->lkb_flags; 2197 __receive_convert_reply(r, lkb, &r->res_ls->ls_stub_ms); 2198 } 2199 2200 return error; 2201 } 2202 2203 /* FIXME: if this lkb is the only lock we hold on the rsb, then set 2204 MASTER_UNCERTAIN to force the next request on the rsb to confirm 2205 that the master is still correct. */ 2206 2207 static int send_unlock(struct dlm_rsb *r, struct dlm_lkb *lkb) 2208 { 2209 return send_common(r, lkb, DLM_MSG_UNLOCK); 2210 } 2211 2212 static int send_cancel(struct dlm_rsb *r, struct dlm_lkb *lkb) 2213 { 2214 return send_common(r, lkb, DLM_MSG_CANCEL); 2215 } 2216 2217 static int send_grant(struct dlm_rsb *r, struct dlm_lkb *lkb) 2218 { 2219 struct dlm_message *ms; 2220 struct dlm_mhandle *mh; 2221 int to_nodeid, error; 2222 2223 to_nodeid = lkb->lkb_nodeid; 2224 2225 error = create_message(r, lkb, to_nodeid, DLM_MSG_GRANT, &ms, &mh); 2226 if (error) 2227 goto out; 2228 2229 send_args(r, lkb, ms); 2230 2231 ms->m_result = 0; 2232 2233 error = send_message(mh, ms); 2234 out: 2235 return error; 2236 } 2237 2238 static int send_bast(struct dlm_rsb *r, struct dlm_lkb *lkb, int mode) 2239 { 2240 struct dlm_message *ms; 2241 struct dlm_mhandle *mh; 2242 int to_nodeid, error; 2243 2244 to_nodeid = lkb->lkb_nodeid; 2245 2246 error = create_message(r, NULL, to_nodeid, DLM_MSG_BAST, &ms, &mh); 2247 if (error) 2248 goto out; 2249 2250 send_args(r, lkb, ms); 2251 2252 ms->m_bastmode = mode; 2253 2254 error = send_message(mh, ms); 2255 out: 2256 return error; 2257 } 2258 2259 static int send_lookup(struct dlm_rsb *r, struct dlm_lkb *lkb) 2260 { 2261 struct dlm_message *ms; 2262 struct dlm_mhandle *mh; 2263 int to_nodeid, error; 2264 2265 add_to_waiters(lkb, DLM_MSG_LOOKUP); 2266 2267 to_nodeid = dlm_dir_nodeid(r); 2268 2269 error = create_message(r, NULL, to_nodeid, DLM_MSG_LOOKUP, &ms, &mh); 2270 if (error) 2271 goto fail; 2272 2273 send_args(r, lkb, ms); 2274 2275 error = send_message(mh, ms); 2276 if (error) 2277 goto fail; 2278 return 0; 2279 2280 fail: 2281 remove_from_waiters(lkb); 2282 return error; 2283 } 2284 2285 static int send_remove(struct dlm_rsb *r) 2286 { 2287 struct dlm_message *ms; 2288 struct dlm_mhandle *mh; 2289 int to_nodeid, error; 2290 2291 to_nodeid = dlm_dir_nodeid(r); 2292 2293 error = create_message(r, NULL, to_nodeid, DLM_MSG_REMOVE, &ms, &mh); 2294 if (error) 2295 goto out; 2296 2297 memcpy(ms->m_extra, r->res_name, r->res_length); 2298 ms->m_hash = r->res_hash; 2299 2300 error = send_message(mh, ms); 2301 out: 2302 return error; 2303 } 2304 2305 static int send_common_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, 2306 int mstype, int rv) 2307 { 2308 struct dlm_message *ms; 2309 struct dlm_mhandle *mh; 2310 int to_nodeid, error; 2311 2312 to_nodeid = lkb->lkb_nodeid; 2313 2314 error = create_message(r, lkb, to_nodeid, mstype, &ms, &mh); 2315 if (error) 2316 goto out; 2317 2318 send_args(r, lkb, ms); 2319 2320 ms->m_result = rv; 2321 2322 error = send_message(mh, ms); 2323 out: 2324 return error; 2325 } 2326 2327 static int send_request_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv) 2328 { 2329 return send_common_reply(r, lkb, DLM_MSG_REQUEST_REPLY, rv); 2330 } 2331 2332 static int send_convert_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv) 2333 { 2334 return send_common_reply(r, lkb, DLM_MSG_CONVERT_REPLY, rv); 2335 } 2336 2337 static int send_unlock_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv) 2338 { 2339 return send_common_reply(r, lkb, DLM_MSG_UNLOCK_REPLY, rv); 2340 } 2341 2342 static int send_cancel_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv) 2343 { 2344 return send_common_reply(r, lkb, DLM_MSG_CANCEL_REPLY, rv); 2345 } 2346 2347 static int send_lookup_reply(struct dlm_ls *ls, struct dlm_message *ms_in, 2348 int ret_nodeid, int rv) 2349 { 2350 struct dlm_rsb *r = &ls->ls_stub_rsb; 2351 struct dlm_message *ms; 2352 struct dlm_mhandle *mh; 2353 int error, nodeid = ms_in->m_header.h_nodeid; 2354 2355 error = create_message(r, NULL, nodeid, DLM_MSG_LOOKUP_REPLY, &ms, &mh); 2356 if (error) 2357 goto out; 2358 2359 ms->m_lkid = ms_in->m_lkid; 2360 ms->m_result = rv; 2361 ms->m_nodeid = ret_nodeid; 2362 2363 error = send_message(mh, ms); 2364 out: 2365 return error; 2366 } 2367 2368 /* which args we save from a received message depends heavily on the type 2369 of message, unlike the send side where we can safely send everything about 2370 the lkb for any type of message */ 2371 2372 static void receive_flags(struct dlm_lkb *lkb, struct dlm_message *ms) 2373 { 2374 lkb->lkb_exflags = ms->m_exflags; 2375 lkb->lkb_sbflags = ms->m_sbflags; 2376 lkb->lkb_flags = (lkb->lkb_flags & 0xFFFF0000) | 2377 (ms->m_flags & 0x0000FFFF); 2378 } 2379 2380 static void receive_flags_reply(struct dlm_lkb *lkb, struct dlm_message *ms) 2381 { 2382 lkb->lkb_sbflags = ms->m_sbflags; 2383 lkb->lkb_flags = (lkb->lkb_flags & 0xFFFF0000) | 2384 (ms->m_flags & 0x0000FFFF); 2385 } 2386 2387 static int receive_extralen(struct dlm_message *ms) 2388 { 2389 return (ms->m_header.h_length - sizeof(struct dlm_message)); 2390 } 2391 2392 static int receive_lvb(struct dlm_ls *ls, struct dlm_lkb *lkb, 2393 struct dlm_message *ms) 2394 { 2395 int len; 2396 2397 if (lkb->lkb_exflags & DLM_LKF_VALBLK) { 2398 if (!lkb->lkb_lvbptr) 2399 lkb->lkb_lvbptr = allocate_lvb(ls); 2400 if (!lkb->lkb_lvbptr) 2401 return -ENOMEM; 2402 len = receive_extralen(ms); 2403 memcpy(lkb->lkb_lvbptr, ms->m_extra, len); 2404 } 2405 return 0; 2406 } 2407 2408 static int receive_request_args(struct dlm_ls *ls, struct dlm_lkb *lkb, 2409 struct dlm_message *ms) 2410 { 2411 lkb->lkb_nodeid = ms->m_header.h_nodeid; 2412 lkb->lkb_ownpid = ms->m_pid; 2413 lkb->lkb_remid = ms->m_lkid; 2414 lkb->lkb_grmode = DLM_LOCK_IV; 2415 lkb->lkb_rqmode = ms->m_rqmode; 2416 lkb->lkb_bastaddr = (void *) (long) (ms->m_asts & AST_BAST); 2417 lkb->lkb_astaddr = (void *) (long) (ms->m_asts & AST_COMP); 2418 2419 DLM_ASSERT(is_master_copy(lkb), dlm_print_lkb(lkb);); 2420 2421 if (receive_lvb(ls, lkb, ms)) 2422 return -ENOMEM; 2423 2424 return 0; 2425 } 2426 2427 static int receive_convert_args(struct dlm_ls *ls, struct dlm_lkb *lkb, 2428 struct dlm_message *ms) 2429 { 2430 if (lkb->lkb_nodeid != ms->m_header.h_nodeid) { 2431 log_error(ls, "convert_args nodeid %d %d lkid %x %x", 2432 lkb->lkb_nodeid, ms->m_header.h_nodeid, 2433 lkb->lkb_id, lkb->lkb_remid); 2434 return -EINVAL; 2435 } 2436 2437 if (!is_master_copy(lkb)) 2438 return -EINVAL; 2439 2440 if (lkb->lkb_status != DLM_LKSTS_GRANTED) 2441 return -EBUSY; 2442 2443 if (receive_lvb(ls, lkb, ms)) 2444 return -ENOMEM; 2445 2446 lkb->lkb_rqmode = ms->m_rqmode; 2447 lkb->lkb_lvbseq = ms->m_lvbseq; 2448 2449 return 0; 2450 } 2451 2452 static int receive_unlock_args(struct dlm_ls *ls, struct dlm_lkb *lkb, 2453 struct dlm_message *ms) 2454 { 2455 if (!is_master_copy(lkb)) 2456 return -EINVAL; 2457 if (receive_lvb(ls, lkb, ms)) 2458 return -ENOMEM; 2459 return 0; 2460 } 2461 2462 /* We fill in the stub-lkb fields with the info that send_xxxx_reply() 2463 uses to send a reply and that the remote end uses to process the reply. */ 2464 2465 static void setup_stub_lkb(struct dlm_ls *ls, struct dlm_message *ms) 2466 { 2467 struct dlm_lkb *lkb = &ls->ls_stub_lkb; 2468 lkb->lkb_nodeid = ms->m_header.h_nodeid; 2469 lkb->lkb_remid = ms->m_lkid; 2470 } 2471 2472 static void receive_request(struct dlm_ls *ls, struct dlm_message *ms) 2473 { 2474 struct dlm_lkb *lkb; 2475 struct dlm_rsb *r; 2476 int error, namelen; 2477 2478 error = create_lkb(ls, &lkb); 2479 if (error) 2480 goto fail; 2481 2482 receive_flags(lkb, ms); 2483 lkb->lkb_flags |= DLM_IFL_MSTCPY; 2484 error = receive_request_args(ls, lkb, ms); 2485 if (error) { 2486 __put_lkb(ls, lkb); 2487 goto fail; 2488 } 2489 2490 namelen = receive_extralen(ms); 2491 2492 error = find_rsb(ls, ms->m_extra, namelen, R_MASTER, &r); 2493 if (error) { 2494 __put_lkb(ls, lkb); 2495 goto fail; 2496 } 2497 2498 lock_rsb(r); 2499 2500 attach_lkb(r, lkb); 2501 error = do_request(r, lkb); 2502 send_request_reply(r, lkb, error); 2503 2504 unlock_rsb(r); 2505 put_rsb(r); 2506 2507 if (error == -EINPROGRESS) 2508 error = 0; 2509 if (error) 2510 dlm_put_lkb(lkb); 2511 return; 2512 2513 fail: 2514 setup_stub_lkb(ls, ms); 2515 send_request_reply(&ls->ls_stub_rsb, &ls->ls_stub_lkb, error); 2516 } 2517 2518 static void receive_convert(struct dlm_ls *ls, struct dlm_message *ms) 2519 { 2520 struct dlm_lkb *lkb; 2521 struct dlm_rsb *r; 2522 int error, reply = 1; 2523 2524 error = find_lkb(ls, ms->m_remid, &lkb); 2525 if (error) 2526 goto fail; 2527 2528 r = lkb->lkb_resource; 2529 2530 hold_rsb(r); 2531 lock_rsb(r); 2532 2533 receive_flags(lkb, ms); 2534 error = receive_convert_args(ls, lkb, ms); 2535 if (error) 2536 goto out; 2537 reply = !down_conversion(lkb); 2538 2539 error = do_convert(r, lkb); 2540 out: 2541 if (reply) 2542 send_convert_reply(r, lkb, error); 2543 2544 unlock_rsb(r); 2545 put_rsb(r); 2546 dlm_put_lkb(lkb); 2547 return; 2548 2549 fail: 2550 setup_stub_lkb(ls, ms); 2551 send_convert_reply(&ls->ls_stub_rsb, &ls->ls_stub_lkb, error); 2552 } 2553 2554 static void receive_unlock(struct dlm_ls *ls, struct dlm_message *ms) 2555 { 2556 struct dlm_lkb *lkb; 2557 struct dlm_rsb *r; 2558 int error; 2559 2560 error = find_lkb(ls, ms->m_remid, &lkb); 2561 if (error) 2562 goto fail; 2563 2564 r = lkb->lkb_resource; 2565 2566 hold_rsb(r); 2567 lock_rsb(r); 2568 2569 receive_flags(lkb, ms); 2570 error = receive_unlock_args(ls, lkb, ms); 2571 if (error) 2572 goto out; 2573 2574 error = do_unlock(r, lkb); 2575 out: 2576 send_unlock_reply(r, lkb, error); 2577 2578 unlock_rsb(r); 2579 put_rsb(r); 2580 dlm_put_lkb(lkb); 2581 return; 2582 2583 fail: 2584 setup_stub_lkb(ls, ms); 2585 send_unlock_reply(&ls->ls_stub_rsb, &ls->ls_stub_lkb, error); 2586 } 2587 2588 static void receive_cancel(struct dlm_ls *ls, struct dlm_message *ms) 2589 { 2590 struct dlm_lkb *lkb; 2591 struct dlm_rsb *r; 2592 int error; 2593 2594 error = find_lkb(ls, ms->m_remid, &lkb); 2595 if (error) 2596 goto fail; 2597 2598 receive_flags(lkb, ms); 2599 2600 r = lkb->lkb_resource; 2601 2602 hold_rsb(r); 2603 lock_rsb(r); 2604 2605 error = do_cancel(r, lkb); 2606 send_cancel_reply(r, lkb, error); 2607 2608 unlock_rsb(r); 2609 put_rsb(r); 2610 dlm_put_lkb(lkb); 2611 return; 2612 2613 fail: 2614 setup_stub_lkb(ls, ms); 2615 send_cancel_reply(&ls->ls_stub_rsb, &ls->ls_stub_lkb, error); 2616 } 2617 2618 static void receive_grant(struct dlm_ls *ls, struct dlm_message *ms) 2619 { 2620 struct dlm_lkb *lkb; 2621 struct dlm_rsb *r; 2622 int error; 2623 2624 error = find_lkb(ls, ms->m_remid, &lkb); 2625 if (error) { 2626 log_error(ls, "receive_grant no lkb"); 2627 return; 2628 } 2629 DLM_ASSERT(is_process_copy(lkb), dlm_print_lkb(lkb);); 2630 2631 r = lkb->lkb_resource; 2632 2633 hold_rsb(r); 2634 lock_rsb(r); 2635 2636 receive_flags_reply(lkb, ms); 2637 grant_lock_pc(r, lkb, ms); 2638 queue_cast(r, lkb, 0); 2639 2640 unlock_rsb(r); 2641 put_rsb(r); 2642 dlm_put_lkb(lkb); 2643 } 2644 2645 static void receive_bast(struct dlm_ls *ls, struct dlm_message *ms) 2646 { 2647 struct dlm_lkb *lkb; 2648 struct dlm_rsb *r; 2649 int error; 2650 2651 error = find_lkb(ls, ms->m_remid, &lkb); 2652 if (error) { 2653 log_error(ls, "receive_bast no lkb"); 2654 return; 2655 } 2656 DLM_ASSERT(is_process_copy(lkb), dlm_print_lkb(lkb);); 2657 2658 r = lkb->lkb_resource; 2659 2660 hold_rsb(r); 2661 lock_rsb(r); 2662 2663 queue_bast(r, lkb, ms->m_bastmode); 2664 2665 unlock_rsb(r); 2666 put_rsb(r); 2667 dlm_put_lkb(lkb); 2668 } 2669 2670 static void receive_lookup(struct dlm_ls *ls, struct dlm_message *ms) 2671 { 2672 int len, error, ret_nodeid, dir_nodeid, from_nodeid, our_nodeid; 2673 2674 from_nodeid = ms->m_header.h_nodeid; 2675 our_nodeid = dlm_our_nodeid(); 2676 2677 len = receive_extralen(ms); 2678 2679 dir_nodeid = dlm_hash2nodeid(ls, ms->m_hash); 2680 if (dir_nodeid != our_nodeid) { 2681 log_error(ls, "lookup dir_nodeid %d from %d", 2682 dir_nodeid, from_nodeid); 2683 error = -EINVAL; 2684 ret_nodeid = -1; 2685 goto out; 2686 } 2687 2688 error = dlm_dir_lookup(ls, from_nodeid, ms->m_extra, len, &ret_nodeid); 2689 2690 /* Optimization: we're master so treat lookup as a request */ 2691 if (!error && ret_nodeid == our_nodeid) { 2692 receive_request(ls, ms); 2693 return; 2694 } 2695 out: 2696 send_lookup_reply(ls, ms, ret_nodeid, error); 2697 } 2698 2699 static void receive_remove(struct dlm_ls *ls, struct dlm_message *ms) 2700 { 2701 int len, dir_nodeid, from_nodeid; 2702 2703 from_nodeid = ms->m_header.h_nodeid; 2704 2705 len = receive_extralen(ms); 2706 2707 dir_nodeid = dlm_hash2nodeid(ls, ms->m_hash); 2708 if (dir_nodeid != dlm_our_nodeid()) { 2709 log_error(ls, "remove dir entry dir_nodeid %d from %d", 2710 dir_nodeid, from_nodeid); 2711 return; 2712 } 2713 2714 dlm_dir_remove_entry(ls, from_nodeid, ms->m_extra, len); 2715 } 2716 2717 static void receive_request_reply(struct dlm_ls *ls, struct dlm_message *ms) 2718 { 2719 struct dlm_lkb *lkb; 2720 struct dlm_rsb *r; 2721 int error, mstype; 2722 2723 error = find_lkb(ls, ms->m_remid, &lkb); 2724 if (error) { 2725 log_error(ls, "receive_request_reply no lkb"); 2726 return; 2727 } 2728 DLM_ASSERT(is_process_copy(lkb), dlm_print_lkb(lkb);); 2729 2730 mstype = lkb->lkb_wait_type; 2731 error = remove_from_waiters(lkb); 2732 if (error) { 2733 log_error(ls, "receive_request_reply not on waiters"); 2734 goto out; 2735 } 2736 2737 /* this is the value returned from do_request() on the master */ 2738 error = ms->m_result; 2739 2740 r = lkb->lkb_resource; 2741 hold_rsb(r); 2742 lock_rsb(r); 2743 2744 /* Optimization: the dir node was also the master, so it took our 2745 lookup as a request and sent request reply instead of lookup reply */ 2746 if (mstype == DLM_MSG_LOOKUP) { 2747 r->res_nodeid = ms->m_header.h_nodeid; 2748 lkb->lkb_nodeid = r->res_nodeid; 2749 } 2750 2751 switch (error) { 2752 case -EAGAIN: 2753 /* request would block (be queued) on remote master; 2754 the unhold undoes the original ref from create_lkb() 2755 so it leads to the lkb being freed */ 2756 queue_cast(r, lkb, -EAGAIN); 2757 confirm_master(r, -EAGAIN); 2758 unhold_lkb(lkb); 2759 break; 2760 2761 case -EINPROGRESS: 2762 case 0: 2763 /* request was queued or granted on remote master */ 2764 receive_flags_reply(lkb, ms); 2765 lkb->lkb_remid = ms->m_lkid; 2766 if (error) 2767 add_lkb(r, lkb, DLM_LKSTS_WAITING); 2768 else { 2769 grant_lock_pc(r, lkb, ms); 2770 queue_cast(r, lkb, 0); 2771 } 2772 confirm_master(r, error); 2773 break; 2774 2775 case -EBADR: 2776 case -ENOTBLK: 2777 /* find_rsb failed to find rsb or rsb wasn't master */ 2778 r->res_nodeid = -1; 2779 lkb->lkb_nodeid = -1; 2780 _request_lock(r, lkb); 2781 break; 2782 2783 default: 2784 log_error(ls, "receive_request_reply error %d", error); 2785 } 2786 2787 unlock_rsb(r); 2788 put_rsb(r); 2789 out: 2790 dlm_put_lkb(lkb); 2791 } 2792 2793 static void __receive_convert_reply(struct dlm_rsb *r, struct dlm_lkb *lkb, 2794 struct dlm_message *ms) 2795 { 2796 int error = ms->m_result; 2797 2798 /* this is the value returned from do_convert() on the master */ 2799 2800 switch (error) { 2801 case -EAGAIN: 2802 /* convert would block (be queued) on remote master */ 2803 queue_cast(r, lkb, -EAGAIN); 2804 break; 2805 2806 case -EINPROGRESS: 2807 /* convert was queued on remote master */ 2808 del_lkb(r, lkb); 2809 add_lkb(r, lkb, DLM_LKSTS_CONVERT); 2810 break; 2811 2812 case 0: 2813 /* convert was granted on remote master */ 2814 receive_flags_reply(lkb, ms); 2815 grant_lock_pc(r, lkb, ms); 2816 queue_cast(r, lkb, 0); 2817 break; 2818 2819 default: 2820 log_error(r->res_ls, "receive_convert_reply error %d", error); 2821 } 2822 } 2823 2824 static void _receive_convert_reply(struct dlm_lkb *lkb, struct dlm_message *ms) 2825 { 2826 struct dlm_rsb *r = lkb->lkb_resource; 2827 2828 hold_rsb(r); 2829 lock_rsb(r); 2830 2831 __receive_convert_reply(r, lkb, ms); 2832 2833 unlock_rsb(r); 2834 put_rsb(r); 2835 } 2836 2837 static void receive_convert_reply(struct dlm_ls *ls, struct dlm_message *ms) 2838 { 2839 struct dlm_lkb *lkb; 2840 int error; 2841 2842 error = find_lkb(ls, ms->m_remid, &lkb); 2843 if (error) { 2844 log_error(ls, "receive_convert_reply no lkb"); 2845 return; 2846 } 2847 DLM_ASSERT(is_process_copy(lkb), dlm_print_lkb(lkb);); 2848 2849 error = remove_from_waiters(lkb); 2850 if (error) { 2851 log_error(ls, "receive_convert_reply not on waiters"); 2852 goto out; 2853 } 2854 2855 _receive_convert_reply(lkb, ms); 2856 out: 2857 dlm_put_lkb(lkb); 2858 } 2859 2860 static void _receive_unlock_reply(struct dlm_lkb *lkb, struct dlm_message *ms) 2861 { 2862 struct dlm_rsb *r = lkb->lkb_resource; 2863 int error = ms->m_result; 2864 2865 hold_rsb(r); 2866 lock_rsb(r); 2867 2868 /* this is the value returned from do_unlock() on the master */ 2869 2870 switch (error) { 2871 case -DLM_EUNLOCK: 2872 receive_flags_reply(lkb, ms); 2873 remove_lock_pc(r, lkb); 2874 queue_cast(r, lkb, -DLM_EUNLOCK); 2875 break; 2876 default: 2877 log_error(r->res_ls, "receive_unlock_reply error %d", error); 2878 } 2879 2880 unlock_rsb(r); 2881 put_rsb(r); 2882 } 2883 2884 static void receive_unlock_reply(struct dlm_ls *ls, struct dlm_message *ms) 2885 { 2886 struct dlm_lkb *lkb; 2887 int error; 2888 2889 error = find_lkb(ls, ms->m_remid, &lkb); 2890 if (error) { 2891 log_error(ls, "receive_unlock_reply no lkb"); 2892 return; 2893 } 2894 DLM_ASSERT(is_process_copy(lkb), dlm_print_lkb(lkb);); 2895 2896 error = remove_from_waiters(lkb); 2897 if (error) { 2898 log_error(ls, "receive_unlock_reply not on waiters"); 2899 goto out; 2900 } 2901 2902 _receive_unlock_reply(lkb, ms); 2903 out: 2904 dlm_put_lkb(lkb); 2905 } 2906 2907 static void _receive_cancel_reply(struct dlm_lkb *lkb, struct dlm_message *ms) 2908 { 2909 struct dlm_rsb *r = lkb->lkb_resource; 2910 int error = ms->m_result; 2911 2912 hold_rsb(r); 2913 lock_rsb(r); 2914 2915 /* this is the value returned from do_cancel() on the master */ 2916 2917 switch (error) { 2918 case -DLM_ECANCEL: 2919 receive_flags_reply(lkb, ms); 2920 revert_lock_pc(r, lkb); 2921 queue_cast(r, lkb, -DLM_ECANCEL); 2922 break; 2923 default: 2924 log_error(r->res_ls, "receive_cancel_reply error %d", error); 2925 } 2926 2927 unlock_rsb(r); 2928 put_rsb(r); 2929 } 2930 2931 static void receive_cancel_reply(struct dlm_ls *ls, struct dlm_message *ms) 2932 { 2933 struct dlm_lkb *lkb; 2934 int error; 2935 2936 error = find_lkb(ls, ms->m_remid, &lkb); 2937 if (error) { 2938 log_error(ls, "receive_cancel_reply no lkb"); 2939 return; 2940 } 2941 DLM_ASSERT(is_process_copy(lkb), dlm_print_lkb(lkb);); 2942 2943 error = remove_from_waiters(lkb); 2944 if (error) { 2945 log_error(ls, "receive_cancel_reply not on waiters"); 2946 goto out; 2947 } 2948 2949 _receive_cancel_reply(lkb, ms); 2950 out: 2951 dlm_put_lkb(lkb); 2952 } 2953 2954 static void receive_lookup_reply(struct dlm_ls *ls, struct dlm_message *ms) 2955 { 2956 struct dlm_lkb *lkb; 2957 struct dlm_rsb *r; 2958 int error, ret_nodeid; 2959 2960 error = find_lkb(ls, ms->m_lkid, &lkb); 2961 if (error) { 2962 log_error(ls, "receive_lookup_reply no lkb"); 2963 return; 2964 } 2965 2966 error = remove_from_waiters(lkb); 2967 if (error) { 2968 log_error(ls, "receive_lookup_reply not on waiters"); 2969 goto out; 2970 } 2971 2972 /* this is the value returned by dlm_dir_lookup on dir node 2973 FIXME: will a non-zero error ever be returned? */ 2974 error = ms->m_result; 2975 2976 r = lkb->lkb_resource; 2977 hold_rsb(r); 2978 lock_rsb(r); 2979 2980 ret_nodeid = ms->m_nodeid; 2981 if (ret_nodeid == dlm_our_nodeid()) { 2982 r->res_nodeid = 0; 2983 ret_nodeid = 0; 2984 r->res_first_lkid = 0; 2985 } else { 2986 /* set_master() will copy res_nodeid to lkb_nodeid */ 2987 r->res_nodeid = ret_nodeid; 2988 } 2989 2990 _request_lock(r, lkb); 2991 2992 if (!ret_nodeid) 2993 process_lookup_list(r); 2994 2995 unlock_rsb(r); 2996 put_rsb(r); 2997 out: 2998 dlm_put_lkb(lkb); 2999 } 3000 3001 int dlm_receive_message(struct dlm_header *hd, int nodeid, int recovery) 3002 { 3003 struct dlm_message *ms = (struct dlm_message *) hd; 3004 struct dlm_ls *ls; 3005 int error; 3006 3007 if (!recovery) 3008 dlm_message_in(ms); 3009 3010 ls = dlm_find_lockspace_global(hd->h_lockspace); 3011 if (!ls) { 3012 log_print("drop message %d from %d for unknown lockspace %d", 3013 ms->m_type, nodeid, hd->h_lockspace); 3014 return -EINVAL; 3015 } 3016 3017 /* recovery may have just ended leaving a bunch of backed-up requests 3018 in the requestqueue; wait while dlm_recoverd clears them */ 3019 3020 if (!recovery) 3021 dlm_wait_requestqueue(ls); 3022 3023 /* recovery may have just started while there were a bunch of 3024 in-flight requests -- save them in requestqueue to be processed 3025 after recovery. we can't let dlm_recvd block on the recovery 3026 lock. if dlm_recoverd is calling this function to clear the 3027 requestqueue, it needs to be interrupted (-EINTR) if another 3028 recovery operation is starting. */ 3029 3030 while (1) { 3031 if (dlm_locking_stopped(ls)) { 3032 if (recovery) { 3033 error = -EINTR; 3034 goto out; 3035 } 3036 error = dlm_add_requestqueue(ls, nodeid, hd); 3037 if (error == -EAGAIN) 3038 continue; 3039 else { 3040 error = -EINTR; 3041 goto out; 3042 } 3043 } 3044 3045 if (lock_recovery_try(ls)) 3046 break; 3047 schedule(); 3048 } 3049 3050 switch (ms->m_type) { 3051 3052 /* messages sent to a master node */ 3053 3054 case DLM_MSG_REQUEST: 3055 receive_request(ls, ms); 3056 break; 3057 3058 case DLM_MSG_CONVERT: 3059 receive_convert(ls, ms); 3060 break; 3061 3062 case DLM_MSG_UNLOCK: 3063 receive_unlock(ls, ms); 3064 break; 3065 3066 case DLM_MSG_CANCEL: 3067 receive_cancel(ls, ms); 3068 break; 3069 3070 /* messages sent from a master node (replies to above) */ 3071 3072 case DLM_MSG_REQUEST_REPLY: 3073 receive_request_reply(ls, ms); 3074 break; 3075 3076 case DLM_MSG_CONVERT_REPLY: 3077 receive_convert_reply(ls, ms); 3078 break; 3079 3080 case DLM_MSG_UNLOCK_REPLY: 3081 receive_unlock_reply(ls, ms); 3082 break; 3083 3084 case DLM_MSG_CANCEL_REPLY: 3085 receive_cancel_reply(ls, ms); 3086 break; 3087 3088 /* messages sent from a master node (only two types of async msg) */ 3089 3090 case DLM_MSG_GRANT: 3091 receive_grant(ls, ms); 3092 break; 3093 3094 case DLM_MSG_BAST: 3095 receive_bast(ls, ms); 3096 break; 3097 3098 /* messages sent to a dir node */ 3099 3100 case DLM_MSG_LOOKUP: 3101 receive_lookup(ls, ms); 3102 break; 3103 3104 case DLM_MSG_REMOVE: 3105 receive_remove(ls, ms); 3106 break; 3107 3108 /* messages sent from a dir node (remove has no reply) */ 3109 3110 case DLM_MSG_LOOKUP_REPLY: 3111 receive_lookup_reply(ls, ms); 3112 break; 3113 3114 default: 3115 log_error(ls, "unknown message type %d", ms->m_type); 3116 } 3117 3118 unlock_recovery(ls); 3119 out: 3120 dlm_put_lockspace(ls); 3121 dlm_astd_wake(); 3122 return 0; 3123 } 3124 3125 3126 /* 3127 * Recovery related 3128 */ 3129 3130 static void recover_convert_waiter(struct dlm_ls *ls, struct dlm_lkb *lkb) 3131 { 3132 if (middle_conversion(lkb)) { 3133 hold_lkb(lkb); 3134 ls->ls_stub_ms.m_result = -EINPROGRESS; 3135 _remove_from_waiters(lkb); 3136 _receive_convert_reply(lkb, &ls->ls_stub_ms); 3137 3138 /* Same special case as in receive_rcom_lock_args() */ 3139 lkb->lkb_grmode = DLM_LOCK_IV; 3140 rsb_set_flag(lkb->lkb_resource, RSB_RECOVER_CONVERT); 3141 unhold_lkb(lkb); 3142 3143 } else if (lkb->lkb_rqmode >= lkb->lkb_grmode) { 3144 lkb->lkb_flags |= DLM_IFL_RESEND; 3145 } 3146 3147 /* lkb->lkb_rqmode < lkb->lkb_grmode shouldn't happen since down 3148 conversions are async; there's no reply from the remote master */ 3149 } 3150 3151 /* A waiting lkb needs recovery if the master node has failed, or 3152 the master node is changing (only when no directory is used) */ 3153 3154 static int waiter_needs_recovery(struct dlm_ls *ls, struct dlm_lkb *lkb) 3155 { 3156 if (dlm_is_removed(ls, lkb->lkb_nodeid)) 3157 return 1; 3158 3159 if (!dlm_no_directory(ls)) 3160 return 0; 3161 3162 if (dlm_dir_nodeid(lkb->lkb_resource) != lkb->lkb_nodeid) 3163 return 1; 3164 3165 return 0; 3166 } 3167 3168 /* Recovery for locks that are waiting for replies from nodes that are now 3169 gone. We can just complete unlocks and cancels by faking a reply from the 3170 dead node. Requests and up-conversions we flag to be resent after 3171 recovery. Down-conversions can just be completed with a fake reply like 3172 unlocks. Conversions between PR and CW need special attention. */ 3173 3174 void dlm_recover_waiters_pre(struct dlm_ls *ls) 3175 { 3176 struct dlm_lkb *lkb, *safe; 3177 3178 mutex_lock(&ls->ls_waiters_mutex); 3179 3180 list_for_each_entry_safe(lkb, safe, &ls->ls_waiters, lkb_wait_reply) { 3181 log_debug(ls, "pre recover waiter lkid %x type %d flags %x", 3182 lkb->lkb_id, lkb->lkb_wait_type, lkb->lkb_flags); 3183 3184 /* all outstanding lookups, regardless of destination will be 3185 resent after recovery is done */ 3186 3187 if (lkb->lkb_wait_type == DLM_MSG_LOOKUP) { 3188 lkb->lkb_flags |= DLM_IFL_RESEND; 3189 continue; 3190 } 3191 3192 if (!waiter_needs_recovery(ls, lkb)) 3193 continue; 3194 3195 switch (lkb->lkb_wait_type) { 3196 3197 case DLM_MSG_REQUEST: 3198 lkb->lkb_flags |= DLM_IFL_RESEND; 3199 break; 3200 3201 case DLM_MSG_CONVERT: 3202 recover_convert_waiter(ls, lkb); 3203 break; 3204 3205 case DLM_MSG_UNLOCK: 3206 hold_lkb(lkb); 3207 ls->ls_stub_ms.m_result = -DLM_EUNLOCK; 3208 _remove_from_waiters(lkb); 3209 _receive_unlock_reply(lkb, &ls->ls_stub_ms); 3210 dlm_put_lkb(lkb); 3211 break; 3212 3213 case DLM_MSG_CANCEL: 3214 hold_lkb(lkb); 3215 ls->ls_stub_ms.m_result = -DLM_ECANCEL; 3216 _remove_from_waiters(lkb); 3217 _receive_cancel_reply(lkb, &ls->ls_stub_ms); 3218 dlm_put_lkb(lkb); 3219 break; 3220 3221 default: 3222 log_error(ls, "invalid lkb wait_type %d", 3223 lkb->lkb_wait_type); 3224 } 3225 schedule(); 3226 } 3227 mutex_unlock(&ls->ls_waiters_mutex); 3228 } 3229 3230 static int remove_resend_waiter(struct dlm_ls *ls, struct dlm_lkb **lkb_ret) 3231 { 3232 struct dlm_lkb *lkb; 3233 int rv = 0; 3234 3235 mutex_lock(&ls->ls_waiters_mutex); 3236 list_for_each_entry(lkb, &ls->ls_waiters, lkb_wait_reply) { 3237 if (lkb->lkb_flags & DLM_IFL_RESEND) { 3238 rv = lkb->lkb_wait_type; 3239 _remove_from_waiters(lkb); 3240 lkb->lkb_flags &= ~DLM_IFL_RESEND; 3241 break; 3242 } 3243 } 3244 mutex_unlock(&ls->ls_waiters_mutex); 3245 3246 if (!rv) 3247 lkb = NULL; 3248 *lkb_ret = lkb; 3249 return rv; 3250 } 3251 3252 /* Deal with lookups and lkb's marked RESEND from _pre. We may now be the 3253 master or dir-node for r. Processing the lkb may result in it being placed 3254 back on waiters. */ 3255 3256 int dlm_recover_waiters_post(struct dlm_ls *ls) 3257 { 3258 struct dlm_lkb *lkb; 3259 struct dlm_rsb *r; 3260 int error = 0, mstype; 3261 3262 while (1) { 3263 if (dlm_locking_stopped(ls)) { 3264 log_debug(ls, "recover_waiters_post aborted"); 3265 error = -EINTR; 3266 break; 3267 } 3268 3269 mstype = remove_resend_waiter(ls, &lkb); 3270 if (!mstype) 3271 break; 3272 3273 r = lkb->lkb_resource; 3274 3275 log_debug(ls, "recover_waiters_post %x type %d flags %x %s", 3276 lkb->lkb_id, mstype, lkb->lkb_flags, r->res_name); 3277 3278 switch (mstype) { 3279 3280 case DLM_MSG_LOOKUP: 3281 hold_rsb(r); 3282 lock_rsb(r); 3283 _request_lock(r, lkb); 3284 if (is_master(r)) 3285 confirm_master(r, 0); 3286 unlock_rsb(r); 3287 put_rsb(r); 3288 break; 3289 3290 case DLM_MSG_REQUEST: 3291 hold_rsb(r); 3292 lock_rsb(r); 3293 _request_lock(r, lkb); 3294 if (is_master(r)) 3295 confirm_master(r, 0); 3296 unlock_rsb(r); 3297 put_rsb(r); 3298 break; 3299 3300 case DLM_MSG_CONVERT: 3301 hold_rsb(r); 3302 lock_rsb(r); 3303 _convert_lock(r, lkb); 3304 unlock_rsb(r); 3305 put_rsb(r); 3306 break; 3307 3308 default: 3309 log_error(ls, "recover_waiters_post type %d", mstype); 3310 } 3311 } 3312 3313 return error; 3314 } 3315 3316 static void purge_queue(struct dlm_rsb *r, struct list_head *queue, 3317 int (*test)(struct dlm_ls *ls, struct dlm_lkb *lkb)) 3318 { 3319 struct dlm_ls *ls = r->res_ls; 3320 struct dlm_lkb *lkb, *safe; 3321 3322 list_for_each_entry_safe(lkb, safe, queue, lkb_statequeue) { 3323 if (test(ls, lkb)) { 3324 rsb_set_flag(r, RSB_LOCKS_PURGED); 3325 del_lkb(r, lkb); 3326 /* this put should free the lkb */ 3327 if (!dlm_put_lkb(lkb)) 3328 log_error(ls, "purged lkb not released"); 3329 } 3330 } 3331 } 3332 3333 static int purge_dead_test(struct dlm_ls *ls, struct dlm_lkb *lkb) 3334 { 3335 return (is_master_copy(lkb) && dlm_is_removed(ls, lkb->lkb_nodeid)); 3336 } 3337 3338 static int purge_mstcpy_test(struct dlm_ls *ls, struct dlm_lkb *lkb) 3339 { 3340 return is_master_copy(lkb); 3341 } 3342 3343 static void purge_dead_locks(struct dlm_rsb *r) 3344 { 3345 purge_queue(r, &r->res_grantqueue, &purge_dead_test); 3346 purge_queue(r, &r->res_convertqueue, &purge_dead_test); 3347 purge_queue(r, &r->res_waitqueue, &purge_dead_test); 3348 } 3349 3350 void dlm_purge_mstcpy_locks(struct dlm_rsb *r) 3351 { 3352 purge_queue(r, &r->res_grantqueue, &purge_mstcpy_test); 3353 purge_queue(r, &r->res_convertqueue, &purge_mstcpy_test); 3354 purge_queue(r, &r->res_waitqueue, &purge_mstcpy_test); 3355 } 3356 3357 /* Get rid of locks held by nodes that are gone. */ 3358 3359 int dlm_purge_locks(struct dlm_ls *ls) 3360 { 3361 struct dlm_rsb *r; 3362 3363 log_debug(ls, "dlm_purge_locks"); 3364 3365 down_write(&ls->ls_root_sem); 3366 list_for_each_entry(r, &ls->ls_root_list, res_root_list) { 3367 hold_rsb(r); 3368 lock_rsb(r); 3369 if (is_master(r)) 3370 purge_dead_locks(r); 3371 unlock_rsb(r); 3372 unhold_rsb(r); 3373 3374 schedule(); 3375 } 3376 up_write(&ls->ls_root_sem); 3377 3378 return 0; 3379 } 3380 3381 static struct dlm_rsb *find_purged_rsb(struct dlm_ls *ls, int bucket) 3382 { 3383 struct dlm_rsb *r, *r_ret = NULL; 3384 3385 read_lock(&ls->ls_rsbtbl[bucket].lock); 3386 list_for_each_entry(r, &ls->ls_rsbtbl[bucket].list, res_hashchain) { 3387 if (!rsb_flag(r, RSB_LOCKS_PURGED)) 3388 continue; 3389 hold_rsb(r); 3390 rsb_clear_flag(r, RSB_LOCKS_PURGED); 3391 r_ret = r; 3392 break; 3393 } 3394 read_unlock(&ls->ls_rsbtbl[bucket].lock); 3395 return r_ret; 3396 } 3397 3398 void dlm_grant_after_purge(struct dlm_ls *ls) 3399 { 3400 struct dlm_rsb *r; 3401 int bucket = 0; 3402 3403 while (1) { 3404 r = find_purged_rsb(ls, bucket); 3405 if (!r) { 3406 if (bucket == ls->ls_rsbtbl_size - 1) 3407 break; 3408 bucket++; 3409 continue; 3410 } 3411 lock_rsb(r); 3412 if (is_master(r)) { 3413 grant_pending_locks(r); 3414 confirm_master(r, 0); 3415 } 3416 unlock_rsb(r); 3417 put_rsb(r); 3418 schedule(); 3419 } 3420 } 3421 3422 static struct dlm_lkb *search_remid_list(struct list_head *head, int nodeid, 3423 uint32_t remid) 3424 { 3425 struct dlm_lkb *lkb; 3426 3427 list_for_each_entry(lkb, head, lkb_statequeue) { 3428 if (lkb->lkb_nodeid == nodeid && lkb->lkb_remid == remid) 3429 return lkb; 3430 } 3431 return NULL; 3432 } 3433 3434 static struct dlm_lkb *search_remid(struct dlm_rsb *r, int nodeid, 3435 uint32_t remid) 3436 { 3437 struct dlm_lkb *lkb; 3438 3439 lkb = search_remid_list(&r->res_grantqueue, nodeid, remid); 3440 if (lkb) 3441 return lkb; 3442 lkb = search_remid_list(&r->res_convertqueue, nodeid, remid); 3443 if (lkb) 3444 return lkb; 3445 lkb = search_remid_list(&r->res_waitqueue, nodeid, remid); 3446 if (lkb) 3447 return lkb; 3448 return NULL; 3449 } 3450 3451 static int receive_rcom_lock_args(struct dlm_ls *ls, struct dlm_lkb *lkb, 3452 struct dlm_rsb *r, struct dlm_rcom *rc) 3453 { 3454 struct rcom_lock *rl = (struct rcom_lock *) rc->rc_buf; 3455 int lvblen; 3456 3457 lkb->lkb_nodeid = rc->rc_header.h_nodeid; 3458 lkb->lkb_ownpid = rl->rl_ownpid; 3459 lkb->lkb_remid = rl->rl_lkid; 3460 lkb->lkb_exflags = rl->rl_exflags; 3461 lkb->lkb_flags = rl->rl_flags & 0x0000FFFF; 3462 lkb->lkb_flags |= DLM_IFL_MSTCPY; 3463 lkb->lkb_lvbseq = rl->rl_lvbseq; 3464 lkb->lkb_rqmode = rl->rl_rqmode; 3465 lkb->lkb_grmode = rl->rl_grmode; 3466 /* don't set lkb_status because add_lkb wants to itself */ 3467 3468 lkb->lkb_bastaddr = (void *) (long) (rl->rl_asts & AST_BAST); 3469 lkb->lkb_astaddr = (void *) (long) (rl->rl_asts & AST_COMP); 3470 3471 if (lkb->lkb_exflags & DLM_LKF_VALBLK) { 3472 lkb->lkb_lvbptr = allocate_lvb(ls); 3473 if (!lkb->lkb_lvbptr) 3474 return -ENOMEM; 3475 lvblen = rc->rc_header.h_length - sizeof(struct dlm_rcom) - 3476 sizeof(struct rcom_lock); 3477 memcpy(lkb->lkb_lvbptr, rl->rl_lvb, lvblen); 3478 } 3479 3480 /* Conversions between PR and CW (middle modes) need special handling. 3481 The real granted mode of these converting locks cannot be determined 3482 until all locks have been rebuilt on the rsb (recover_conversion) */ 3483 3484 if (rl->rl_wait_type == DLM_MSG_CONVERT && middle_conversion(lkb)) { 3485 rl->rl_status = DLM_LKSTS_CONVERT; 3486 lkb->lkb_grmode = DLM_LOCK_IV; 3487 rsb_set_flag(r, RSB_RECOVER_CONVERT); 3488 } 3489 3490 return 0; 3491 } 3492 3493 /* This lkb may have been recovered in a previous aborted recovery so we need 3494 to check if the rsb already has an lkb with the given remote nodeid/lkid. 3495 If so we just send back a standard reply. If not, we create a new lkb with 3496 the given values and send back our lkid. We send back our lkid by sending 3497 back the rcom_lock struct we got but with the remid field filled in. */ 3498 3499 int dlm_recover_master_copy(struct dlm_ls *ls, struct dlm_rcom *rc) 3500 { 3501 struct rcom_lock *rl = (struct rcom_lock *) rc->rc_buf; 3502 struct dlm_rsb *r; 3503 struct dlm_lkb *lkb; 3504 int error; 3505 3506 if (rl->rl_parent_lkid) { 3507 error = -EOPNOTSUPP; 3508 goto out; 3509 } 3510 3511 error = find_rsb(ls, rl->rl_name, rl->rl_namelen, R_MASTER, &r); 3512 if (error) 3513 goto out; 3514 3515 lock_rsb(r); 3516 3517 lkb = search_remid(r, rc->rc_header.h_nodeid, rl->rl_lkid); 3518 if (lkb) { 3519 error = -EEXIST; 3520 goto out_remid; 3521 } 3522 3523 error = create_lkb(ls, &lkb); 3524 if (error) 3525 goto out_unlock; 3526 3527 error = receive_rcom_lock_args(ls, lkb, r, rc); 3528 if (error) { 3529 __put_lkb(ls, lkb); 3530 goto out_unlock; 3531 } 3532 3533 attach_lkb(r, lkb); 3534 add_lkb(r, lkb, rl->rl_status); 3535 error = 0; 3536 3537 out_remid: 3538 /* this is the new value returned to the lock holder for 3539 saving in its process-copy lkb */ 3540 rl->rl_remid = lkb->lkb_id; 3541 3542 out_unlock: 3543 unlock_rsb(r); 3544 put_rsb(r); 3545 out: 3546 if (error) 3547 log_print("recover_master_copy %d %x", error, rl->rl_lkid); 3548 rl->rl_result = error; 3549 return error; 3550 } 3551 3552 int dlm_recover_process_copy(struct dlm_ls *ls, struct dlm_rcom *rc) 3553 { 3554 struct rcom_lock *rl = (struct rcom_lock *) rc->rc_buf; 3555 struct dlm_rsb *r; 3556 struct dlm_lkb *lkb; 3557 int error; 3558 3559 error = find_lkb(ls, rl->rl_lkid, &lkb); 3560 if (error) { 3561 log_error(ls, "recover_process_copy no lkid %x", rl->rl_lkid); 3562 return error; 3563 } 3564 3565 DLM_ASSERT(is_process_copy(lkb), dlm_print_lkb(lkb);); 3566 3567 error = rl->rl_result; 3568 3569 r = lkb->lkb_resource; 3570 hold_rsb(r); 3571 lock_rsb(r); 3572 3573 switch (error) { 3574 case -EEXIST: 3575 log_debug(ls, "master copy exists %x", lkb->lkb_id); 3576 /* fall through */ 3577 case 0: 3578 lkb->lkb_remid = rl->rl_remid; 3579 break; 3580 default: 3581 log_error(ls, "dlm_recover_process_copy unknown error %d %x", 3582 error, lkb->lkb_id); 3583 } 3584 3585 /* an ack for dlm_recover_locks() which waits for replies from 3586 all the locks it sends to new masters */ 3587 dlm_recovered_lock(r); 3588 3589 unlock_rsb(r); 3590 put_rsb(r); 3591 dlm_put_lkb(lkb); 3592 3593 return 0; 3594 } 3595 3596 int dlm_user_request(struct dlm_ls *ls, struct dlm_user_args *ua, 3597 int mode, uint32_t flags, void *name, unsigned int namelen, 3598 uint32_t parent_lkid) 3599 { 3600 struct dlm_lkb *lkb; 3601 struct dlm_args args; 3602 int error; 3603 3604 lock_recovery(ls); 3605 3606 error = create_lkb(ls, &lkb); 3607 if (error) { 3608 kfree(ua); 3609 goto out; 3610 } 3611 3612 if (flags & DLM_LKF_VALBLK) { 3613 ua->lksb.sb_lvbptr = kmalloc(DLM_USER_LVB_LEN, GFP_KERNEL); 3614 if (!ua->lksb.sb_lvbptr) { 3615 kfree(ua); 3616 __put_lkb(ls, lkb); 3617 error = -ENOMEM; 3618 goto out; 3619 } 3620 } 3621 3622 /* After ua is attached to lkb it will be freed by free_lkb(). 3623 When DLM_IFL_USER is set, the dlm knows that this is a userspace 3624 lock and that lkb_astparam is the dlm_user_args structure. */ 3625 3626 error = set_lock_args(mode, &ua->lksb, flags, namelen, parent_lkid, 3627 DLM_FAKE_USER_AST, ua, DLM_FAKE_USER_AST, &args); 3628 lkb->lkb_flags |= DLM_IFL_USER; 3629 ua->old_mode = DLM_LOCK_IV; 3630 3631 if (error) { 3632 __put_lkb(ls, lkb); 3633 goto out; 3634 } 3635 3636 error = request_lock(ls, lkb, name, namelen, &args); 3637 3638 switch (error) { 3639 case 0: 3640 break; 3641 case -EINPROGRESS: 3642 error = 0; 3643 break; 3644 case -EAGAIN: 3645 error = 0; 3646 /* fall through */ 3647 default: 3648 __put_lkb(ls, lkb); 3649 goto out; 3650 } 3651 3652 /* add this new lkb to the per-process list of locks */ 3653 spin_lock(&ua->proc->locks_spin); 3654 kref_get(&lkb->lkb_ref); 3655 list_add_tail(&lkb->lkb_ownqueue, &ua->proc->locks); 3656 spin_unlock(&ua->proc->locks_spin); 3657 out: 3658 unlock_recovery(ls); 3659 return error; 3660 } 3661 3662 int dlm_user_convert(struct dlm_ls *ls, struct dlm_user_args *ua_tmp, 3663 int mode, uint32_t flags, uint32_t lkid, char *lvb_in) 3664 { 3665 struct dlm_lkb *lkb; 3666 struct dlm_args args; 3667 struct dlm_user_args *ua; 3668 int error; 3669 3670 lock_recovery(ls); 3671 3672 error = find_lkb(ls, lkid, &lkb); 3673 if (error) 3674 goto out; 3675 3676 /* user can change the params on its lock when it converts it, or 3677 add an lvb that didn't exist before */ 3678 3679 ua = (struct dlm_user_args *)lkb->lkb_astparam; 3680 3681 if (flags & DLM_LKF_VALBLK && !ua->lksb.sb_lvbptr) { 3682 ua->lksb.sb_lvbptr = kmalloc(DLM_USER_LVB_LEN, GFP_KERNEL); 3683 if (!ua->lksb.sb_lvbptr) { 3684 error = -ENOMEM; 3685 goto out_put; 3686 } 3687 } 3688 if (lvb_in && ua->lksb.sb_lvbptr) 3689 memcpy(ua->lksb.sb_lvbptr, lvb_in, DLM_USER_LVB_LEN); 3690 3691 ua->castparam = ua_tmp->castparam; 3692 ua->castaddr = ua_tmp->castaddr; 3693 ua->bastparam = ua_tmp->bastparam; 3694 ua->bastaddr = ua_tmp->bastaddr; 3695 ua->user_lksb = ua_tmp->user_lksb; 3696 ua->old_mode = lkb->lkb_grmode; 3697 3698 error = set_lock_args(mode, &ua->lksb, flags, 0, 0, DLM_FAKE_USER_AST, 3699 ua, DLM_FAKE_USER_AST, &args); 3700 if (error) 3701 goto out_put; 3702 3703 error = convert_lock(ls, lkb, &args); 3704 3705 if (error == -EINPROGRESS || error == -EAGAIN) 3706 error = 0; 3707 out_put: 3708 dlm_put_lkb(lkb); 3709 out: 3710 unlock_recovery(ls); 3711 kfree(ua_tmp); 3712 return error; 3713 } 3714 3715 int dlm_user_unlock(struct dlm_ls *ls, struct dlm_user_args *ua_tmp, 3716 uint32_t flags, uint32_t lkid, char *lvb_in) 3717 { 3718 struct dlm_lkb *lkb; 3719 struct dlm_args args; 3720 struct dlm_user_args *ua; 3721 int error; 3722 3723 lock_recovery(ls); 3724 3725 error = find_lkb(ls, lkid, &lkb); 3726 if (error) 3727 goto out; 3728 3729 ua = (struct dlm_user_args *)lkb->lkb_astparam; 3730 3731 if (lvb_in && ua->lksb.sb_lvbptr) 3732 memcpy(ua->lksb.sb_lvbptr, lvb_in, DLM_USER_LVB_LEN); 3733 ua->castparam = ua_tmp->castparam; 3734 ua->user_lksb = ua_tmp->user_lksb; 3735 3736 error = set_unlock_args(flags, ua, &args); 3737 if (error) 3738 goto out_put; 3739 3740 error = unlock_lock(ls, lkb, &args); 3741 3742 if (error == -DLM_EUNLOCK) 3743 error = 0; 3744 if (error) 3745 goto out_put; 3746 3747 spin_lock(&ua->proc->locks_spin); 3748 list_del_init(&lkb->lkb_ownqueue); 3749 spin_unlock(&ua->proc->locks_spin); 3750 3751 /* this removes the reference for the proc->locks list added by 3752 dlm_user_request */ 3753 unhold_lkb(lkb); 3754 out_put: 3755 dlm_put_lkb(lkb); 3756 out: 3757 unlock_recovery(ls); 3758 return error; 3759 } 3760 3761 int dlm_user_cancel(struct dlm_ls *ls, struct dlm_user_args *ua_tmp, 3762 uint32_t flags, uint32_t lkid) 3763 { 3764 struct dlm_lkb *lkb; 3765 struct dlm_args args; 3766 struct dlm_user_args *ua; 3767 int error; 3768 3769 lock_recovery(ls); 3770 3771 error = find_lkb(ls, lkid, &lkb); 3772 if (error) 3773 goto out; 3774 3775 ua = (struct dlm_user_args *)lkb->lkb_astparam; 3776 ua->castparam = ua_tmp->castparam; 3777 ua->user_lksb = ua_tmp->user_lksb; 3778 3779 error = set_unlock_args(flags, ua, &args); 3780 if (error) 3781 goto out_put; 3782 3783 error = cancel_lock(ls, lkb, &args); 3784 3785 if (error == -DLM_ECANCEL) 3786 error = 0; 3787 if (error) 3788 goto out_put; 3789 3790 /* this lkb was removed from the WAITING queue */ 3791 if (lkb->lkb_grmode == DLM_LOCK_IV) { 3792 spin_lock(&ua->proc->locks_spin); 3793 list_del_init(&lkb->lkb_ownqueue); 3794 spin_unlock(&ua->proc->locks_spin); 3795 unhold_lkb(lkb); 3796 } 3797 out_put: 3798 dlm_put_lkb(lkb); 3799 out: 3800 unlock_recovery(ls); 3801 return error; 3802 } 3803 3804 static int orphan_proc_lock(struct dlm_ls *ls, struct dlm_lkb *lkb) 3805 { 3806 struct dlm_user_args *ua = (struct dlm_user_args *)lkb->lkb_astparam; 3807 3808 if (ua->lksb.sb_lvbptr) 3809 kfree(ua->lksb.sb_lvbptr); 3810 kfree(ua); 3811 lkb->lkb_astparam = (long)NULL; 3812 3813 /* TODO: propogate to master if needed */ 3814 return 0; 3815 } 3816 3817 /* The force flag allows the unlock to go ahead even if the lkb isn't granted. 3818 Regardless of what rsb queue the lock is on, it's removed and freed. */ 3819 3820 static int unlock_proc_lock(struct dlm_ls *ls, struct dlm_lkb *lkb) 3821 { 3822 struct dlm_user_args *ua = (struct dlm_user_args *)lkb->lkb_astparam; 3823 struct dlm_args args; 3824 int error; 3825 3826 /* FIXME: we need to handle the case where the lkb is in limbo 3827 while the rsb is being looked up, currently we assert in 3828 _unlock_lock/is_remote because rsb nodeid is -1. */ 3829 3830 set_unlock_args(DLM_LKF_FORCEUNLOCK, ua, &args); 3831 3832 error = unlock_lock(ls, lkb, &args); 3833 if (error == -DLM_EUNLOCK) 3834 error = 0; 3835 return error; 3836 } 3837 3838 /* The ls_clear_proc_locks mutex protects against dlm_user_add_asts() which 3839 1) references lkb->ua which we free here and 2) adds lkbs to proc->asts, 3840 which we clear here. */ 3841 3842 /* proc CLOSING flag is set so no more device_reads should look at proc->asts 3843 list, and no more device_writes should add lkb's to proc->locks list; so we 3844 shouldn't need to take asts_spin or locks_spin here. this assumes that 3845 device reads/writes/closes are serialized -- FIXME: we may need to serialize 3846 them ourself. */ 3847 3848 void dlm_clear_proc_locks(struct dlm_ls *ls, struct dlm_user_proc *proc) 3849 { 3850 struct dlm_lkb *lkb, *safe; 3851 3852 lock_recovery(ls); 3853 mutex_lock(&ls->ls_clear_proc_locks); 3854 3855 list_for_each_entry_safe(lkb, safe, &proc->locks, lkb_ownqueue) { 3856 if (lkb->lkb_ast_type) { 3857 list_del(&lkb->lkb_astqueue); 3858 unhold_lkb(lkb); 3859 } 3860 3861 list_del_init(&lkb->lkb_ownqueue); 3862 3863 if (lkb->lkb_exflags & DLM_LKF_PERSISTENT) { 3864 lkb->lkb_flags |= DLM_IFL_ORPHAN; 3865 orphan_proc_lock(ls, lkb); 3866 } else { 3867 lkb->lkb_flags |= DLM_IFL_DEAD; 3868 unlock_proc_lock(ls, lkb); 3869 } 3870 3871 /* this removes the reference for the proc->locks list 3872 added by dlm_user_request, it may result in the lkb 3873 being freed */ 3874 3875 dlm_put_lkb(lkb); 3876 } 3877 mutex_unlock(&ls->ls_clear_proc_locks); 3878 unlock_recovery(ls); 3879 } 3880