1/* 2 * Copyright (c) International Business Machines Corp., 2000-2002 3 * Portions Copyright (c) Christoph Hellwig, 2001-2002 4 * 5 * This program is free software; you can redistribute it and/or modify 6 * it under the terms of the GNU General Public License as published by 7 * the Free Software Foundation; either version 2 of the License, or 8 * (at your option) any later version. 9 * 10 * This program is distributed in the hope that it will be useful, 11 * but WITHOUT ANY WARRANTY; without even the implied warranty of 12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See 13 * the GNU General Public License for more details. 14 * 15 * You should have received a copy of the GNU General Public License 16 * along with this program; if not, write to the Free Software 17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA 18 */ 19 20/* 21 * jfs_txnmgr.c: transaction manager 22 * 23 * notes: 24 * transaction starts with txBegin() and ends with txCommit() 25 * or txAbort(). 26 * 27 * tlock is acquired at the time of update; 28 * (obviate scan at commit time for xtree and dtree) 29 * tlock and mp points to each other; 30 * (no hashlist for mp -> tlock). 31 * 32 * special cases: 33 * tlock on in-memory inode: 34 * in-place tlock in the in-memory inode itself; 35 * converted to page lock by iWrite() at commit time. 36 * 37 * tlock during write()/mmap() under anonymous transaction (tid = 0): 38 * transferred (?) to transaction at commit time. 39 * 40 * use the page itself to update allocation maps 41 * (obviate intermediate replication of allocation/deallocation data) 42 * hold on to mp+lock thru update of maps 43 */ 44 45 46#include <linux/fs.h> 47#include <linux/vmalloc.h> 48#include <linux/smp_lock.h> 49#include <linux/completion.h> 50#include "jfs_incore.h" 51#include "jfs_filsys.h" 52#include "jfs_metapage.h" 53#include "jfs_dinode.h" 54#include "jfs_imap.h" 55#include "jfs_dmap.h" 56#include "jfs_superblock.h" 57#include "jfs_debug.h" 58 59/* 60 * transaction management structures 61 */ 62static struct { 63 /* tblock */ 64 int freetid; /* index of a free tid structure */ 65 wait_queue_head_t freewait; /* eventlist of free tblock */ 66 67 /* tlock */ 68 int freelock; /* index first free lock word */ 69 wait_queue_head_t freelockwait; /* eventlist of free tlock */ 70 wait_queue_head_t lowlockwait; /* eventlist of ample tlocks */ 71 int tlocksInUse; /* Number of tlocks in use */ 72 spinlock_t LazyLock; /* synchronize sync_queue & unlock_queue */ 73/* struct tblock *sync_queue; * Transactions waiting for data sync */ 74 struct tblock *unlock_queue; /* Txns waiting to be released */ 75 struct tblock *unlock_tail; /* Tail of unlock_queue */ 76 struct list_head anon_list; /* inodes having anonymous txns */ 77 struct list_head anon_list2; /* inodes having anonymous txns 78 that couldn't be sync'ed */ 79} TxAnchor; 80 81static int nTxBlock = 512; /* number of transaction blocks */ 82struct tblock *TxBlock; /* transaction block table */ 83 84static int nTxLock = 4096; /* number of transaction locks */ 85static int TxLockLWM = 4096*.4; /* Low water mark for number of txLocks used */ 86static int TxLockHWM = 4096*.8; /* High water mark for number of txLocks used */ 87struct tlock *TxLock; /* transaction lock table */ 88static int TlocksLow = 0; /* Indicates low number of available tlocks */ 89 90 91/* 92 * transaction management lock 93 */ 94static spinlock_t jfsTxnLock = SPIN_LOCK_UNLOCKED; 95 96#define TXN_LOCK() spin_lock(&jfsTxnLock) 97#define TXN_UNLOCK() spin_unlock(&jfsTxnLock) 98 99#define LAZY_LOCK_INIT() spin_lock_init(&TxAnchor.LazyLock); 100#define LAZY_LOCK(flags) spin_lock_irqsave(&TxAnchor.LazyLock, flags) 101#define LAZY_UNLOCK(flags) spin_unlock_irqrestore(&TxAnchor.LazyLock, flags) 102 103DECLARE_WAIT_QUEUE_HEAD(jfs_sync_thread_wait); 104DECLARE_WAIT_QUEUE_HEAD(jfs_commit_thread_wait); 105 106/* 107 * Retry logic exist outside these macros to protect from spurrious wakeups. 108 */ 109static inline void TXN_SLEEP_DROP_LOCK(wait_queue_head_t * event) 110{ 111 DECLARE_WAITQUEUE(wait, current); 112 113 add_wait_queue(event, &wait); 114 set_current_state(TASK_UNINTERRUPTIBLE); 115 TXN_UNLOCK(); 116 schedule(); 117 current->state = TASK_RUNNING; 118 remove_wait_queue(event, &wait); 119} 120 121#define TXN_SLEEP(event)\ 122{\ 123 TXN_SLEEP_DROP_LOCK(event);\ 124 TXN_LOCK();\ 125} 126 127#define TXN_WAKEUP(event) wake_up_all(event) 128 129 130/* 131 * statistics 132 */ 133struct { 134 tid_t maxtid; /* 4: biggest tid ever used */ 135 lid_t maxlid; /* 4: biggest lid ever used */ 136 int ntid; /* 4: # of transactions performed */ 137 int nlid; /* 4: # of tlocks acquired */ 138 int waitlock; /* 4: # of tlock wait */ 139} stattx; 140 141 142/* 143 * external references 144 */ 145extern int lmGroupCommit(struct jfs_log * log, struct tblock * tblk); 146extern void lmSync(struct jfs_log *); 147extern int jfs_commit_inode(struct inode *, int); 148extern int jfs_stop_threads; 149 150struct task_struct *jfsCommitTask; 151extern struct completion jfsIOwait; 152 153/* 154 * forward references 155 */ 156int diLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd, 157 struct tlock * tlck, struct commit * cd); 158int dataLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd, 159 struct tlock * tlck); 160void dtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd, 161 struct tlock * tlck); 162void inlineLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd, 163 struct tlock * tlck); 164void mapLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd, 165 struct tlock * tlck); 166void txAbortCommit(struct commit * cd, int exval); 167static void txAllocPMap(struct inode *ip, struct maplock * maplock, 168 struct tblock * tblk); 169void txForce(struct tblock * tblk); 170static int txLog(struct jfs_log * log, struct tblock * tblk, struct commit * cd); 171int txMoreLock(void); 172static void txUpdateMap(struct tblock * tblk); 173static void txRelease(struct tblock * tblk); 174void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd, 175 struct tlock * tlck); 176static void LogSyncRelease(struct metapage * mp); 177 178/* 179 * transaction block/lock management 180 * --------------------------------- 181 */ 182 183/* 184 * Get a transaction lock from the free list. If the number in use is 185 * greater than the high water mark, wake up the sync daemon. This should 186 * free some anonymous transaction locks. (TXN_LOCK must be held.) 187 */ 188static lid_t txLockAlloc(void) 189{ 190 lid_t lid; 191 192 while (!(lid = TxAnchor.freelock)) 193 TXN_SLEEP(&TxAnchor.freelockwait); 194 TxAnchor.freelock = TxLock[lid].next; 195 HIGHWATERMARK(stattx.maxlid, lid); 196 if ((++TxAnchor.tlocksInUse > TxLockHWM) && (TlocksLow == 0)) { 197 jEVENT(0,("txLockAlloc TlocksLow\n")); 198 TlocksLow = 1; 199 wake_up(&jfs_sync_thread_wait); 200 } 201 202 return lid; 203} 204 205static void txLockFree(lid_t lid) 206{ 207 TxLock[lid].next = TxAnchor.freelock; 208 TxAnchor.freelock = lid; 209 TxAnchor.tlocksInUse--; 210 if (TlocksLow && (TxAnchor.tlocksInUse < TxLockLWM)) { 211 jEVENT(0,("txLockFree TlocksLow no more\n")); 212 TlocksLow = 0; 213 TXN_WAKEUP(&TxAnchor.lowlockwait); 214 } 215 TXN_WAKEUP(&TxAnchor.freelockwait); 216} 217 218/* 219 * NAME: txInit() 220 * 221 * FUNCTION: initialize transaction management structures 222 * 223 * RETURN: 224 * 225 * serialization: single thread at jfs_init() 226 */ 227int txInit(void) 228{ 229 int k, size; 230 231 /* 232 * initialize transaction block (tblock) table 233 * 234 * transaction id (tid) = tblock index 235 * tid = 0 is reserved. 236 */ 237 size = sizeof(struct tblock) * nTxBlock; 238 TxBlock = (struct tblock *) vmalloc(size); 239 if (TxBlock == NULL) 240 return ENOMEM; 241 242 for (k = 1; k < nTxBlock - 1; k++) { 243 TxBlock[k].next = k + 1; 244 init_waitqueue_head(&TxBlock[k].gcwait); 245 init_waitqueue_head(&TxBlock[k].waitor); 246 } 247 TxBlock[k].next = 0; 248 init_waitqueue_head(&TxBlock[k].gcwait); 249 init_waitqueue_head(&TxBlock[k].waitor); 250 251 TxAnchor.freetid = 1; 252 init_waitqueue_head(&TxAnchor.freewait); 253 254 stattx.maxtid = 1; /* statistics */ 255 256 /* 257 * initialize transaction lock (tlock) table 258 * 259 * transaction lock id = tlock index 260 * tlock id = 0 is reserved. 261 */ 262 size = sizeof(struct tlock) * nTxLock; 263 TxLock = (struct tlock *) vmalloc(size); 264 if (TxLock == NULL) { 265 vfree(TxBlock); 266 return ENOMEM; 267 } 268 269 /* initialize tlock table */ 270 for (k = 1; k < nTxLock - 1; k++) 271 TxLock[k].next = k + 1; 272 TxLock[k].next = 0; 273 init_waitqueue_head(&TxAnchor.freelockwait); 274 init_waitqueue_head(&TxAnchor.lowlockwait); 275 276 TxAnchor.freelock = 1; 277 TxAnchor.tlocksInUse = 0; 278 INIT_LIST_HEAD(&TxAnchor.anon_list); 279 INIT_LIST_HEAD(&TxAnchor.anon_list2); 280 281 stattx.maxlid = 1; /* statistics */ 282 283 return 0; 284} 285 286/* 287 * NAME: txExit() 288 * 289 * FUNCTION: clean up when module is unloaded 290 */ 291void txExit(void) 292{ 293 vfree(TxLock); 294 TxLock = 0; 295 vfree(TxBlock); 296 TxBlock = 0; 297} 298 299 300/* 301 * NAME: txBegin() 302 * 303 * FUNCTION: start a transaction. 304 * 305 * PARAMETER: sb - superblock 306 * flag - force for nested tx; 307 * 308 * RETURN: tid - transaction id 309 * 310 * note: flag force allows to start tx for nested tx 311 * to prevent deadlock on logsync barrier; 312 */ 313tid_t txBegin(struct super_block *sb, int flag) 314{ 315 tid_t t; 316 struct tblock *tblk; 317 struct jfs_log *log; 318 319 jFYI(1, ("txBegin: flag = 0x%x\n", flag)); 320 log = JFS_SBI(sb)->log; 321 322 TXN_LOCK(); 323 324 retry: 325 if (!(flag & COMMIT_FORCE)) { 326 /* 327 * synchronize with logsync barrier 328 */ 329 if (test_bit(log_SYNCBARRIER, &log->flag) || 330 test_bit(log_QUIESCE, &log->flag)) { 331 TXN_SLEEP(&log->syncwait); 332 goto retry; 333 } 334 } 335 if (flag == 0) { 336 /* 337 * Don't begin transaction if we're getting starved for tlocks 338 * unless COMMIT_FORCE or COMMIT_INODE (which may ultimately 339 * free tlocks) 340 */ 341 if (TlocksLow) { 342 TXN_SLEEP(&TxAnchor.lowlockwait); 343 goto retry; 344 } 345 } 346 347 /* 348 * allocate transaction id/block 349 */ 350 if ((t = TxAnchor.freetid) == 0) { 351 jFYI(1, ("txBegin: waiting for free tid\n")); 352 TXN_SLEEP(&TxAnchor.freewait); 353 goto retry; 354 } 355 356 tblk = tid_to_tblock(t); 357 358 if ((tblk->next == 0) && (current != jfsCommitTask)) { 359 /* Save one tblk for jfsCommit thread */ 360 jFYI(1, ("txBegin: waiting for free tid\n")); 361 TXN_SLEEP(&TxAnchor.freewait); 362 goto retry; 363 } 364 365 TxAnchor.freetid = tblk->next; 366 367 /* 368 * initialize transaction 369 */ 370 371 /* 372 * We can't zero the whole thing or we screw up another thread being 373 * awakened after sleeping on tblk->waitor 374 * 375 * memset(tblk, 0, sizeof(struct tblock)); 376 */ 377 tblk->next = tblk->last = tblk->xflag = tblk->flag = tblk->lsn = 0; 378 379 tblk->sb = sb; 380 ++log->logtid; 381 tblk->logtid = log->logtid; 382 383 ++log->active; 384 385 HIGHWATERMARK(stattx.maxtid, t); /* statistics */ 386 INCREMENT(stattx.ntid); /* statistics */ 387 388 TXN_UNLOCK(); 389 390 jFYI(1, ("txBegin: returning tid = %d\n", t)); 391 392 return t; 393} 394 395 396/* 397 * NAME: txBeginAnon() 398 * 399 * FUNCTION: start an anonymous transaction. 400 * Blocks if logsync or available tlocks are low to prevent 401 * anonymous tlocks from depleting supply. 402 * 403 * PARAMETER: sb - superblock 404 * 405 * RETURN: none 406 */ 407void txBeginAnon(struct super_block *sb) 408{ 409 struct jfs_log *log; 410 411 log = JFS_SBI(sb)->log; 412 413 TXN_LOCK(); 414 415 retry: 416 /* 417 * synchronize with logsync barrier 418 */ 419 if (test_bit(log_SYNCBARRIER, &log->flag) || 420 test_bit(log_QUIESCE, &log->flag)) { 421 TXN_SLEEP(&log->syncwait); 422 goto retry; 423 } 424 425 /* 426 * Don't begin transaction if we're getting starved for tlocks 427 */ 428 if (TlocksLow) { 429 TXN_SLEEP(&TxAnchor.lowlockwait); 430 goto retry; 431 } 432 TXN_UNLOCK(); 433} 434 435 436/* 437 * txEnd() 438 * 439 * function: free specified transaction block. 440 * 441 * logsync barrier processing: 442 * 443 * serialization: 444 */ 445void txEnd(tid_t tid) 446{ 447 struct tblock *tblk = tid_to_tblock(tid); 448 struct jfs_log *log; 449 450 jFYI(1, ("txEnd: tid = %d\n", tid)); 451 TXN_LOCK(); 452 453 /* 454 * wakeup transactions waiting on the page locked 455 * by the current transaction 456 */ 457 TXN_WAKEUP(&tblk->waitor); 458 459 log = JFS_SBI(tblk->sb)->log; 460 461 /* 462 * Lazy commit thread can't free this guy until we mark it UNLOCKED, 463 * otherwise, we would be left with a transaction that may have been 464 * reused. 465 * 466 * Lazy commit thread will turn off tblkGC_LAZY before calling this 467 * routine. 468 */ 469 if (tblk->flag & tblkGC_LAZY) { 470 jFYI(1, 471 ("txEnd called w/lazy tid: %d, tblk = 0x%p\n", 472 tid, tblk)); 473 TXN_UNLOCK(); 474 475 spin_lock_irq(&log->gclock); // LOGGC_LOCK 476 tblk->flag |= tblkGC_UNLOCKED; 477 spin_unlock_irq(&log->gclock); // LOGGC_UNLOCK 478 return; 479 } 480 481 jFYI(1, ("txEnd: tid: %d, tblk = 0x%p\n", tid, tblk)); 482 483 assert(tblk->next == 0); 484 485 /* 486 * insert tblock back on freelist 487 */ 488 tblk->next = TxAnchor.freetid; 489 TxAnchor.freetid = tid; 490 491 /* 492 * mark the tblock not active 493 */ 494 --log->active; 495 496 /* 497 * synchronize with logsync barrier 498 */ 499 if (test_bit(log_SYNCBARRIER, &log->flag) && log->active == 0) { 500 /* forward log syncpt */ 501 /* lmSync(log); */ 502 503 jFYI(1, (" log barrier off: 0x%x\n", log->lsn)); 504 505 /* enable new transactions start */ 506 clear_bit(log_SYNCBARRIER, &log->flag); 507 508 /* wakeup all waitors for logsync barrier */ 509 TXN_WAKEUP(&log->syncwait); 510 } 511 512 /* 513 * wakeup all waitors for a free tblock 514 */ 515 TXN_WAKEUP(&TxAnchor.freewait); 516 517 TXN_UNLOCK(); 518 jFYI(1, ("txEnd: exitting\n")); 519} 520 521 522/* 523 * txLock() 524 * 525 * function: acquire a transaction lock on the specified <mp> 526 * 527 * parameter: 528 * 529 * return: transaction lock id 530 * 531 * serialization: 532 */ 533struct tlock *txLock(tid_t tid, struct inode *ip, struct metapage * mp, 534 int type) 535{ 536 struct jfs_inode_info *jfs_ip = JFS_IP(ip); 537 int dir_xtree = 0; 538 lid_t lid; 539 tid_t xtid; 540 struct tlock *tlck; 541 struct xtlock *xtlck; 542 struct linelock *linelock; 543 xtpage_t *p; 544 struct tblock *tblk; 545 546 assert(!test_cflag(COMMIT_Nolink, ip)); 547 548 TXN_LOCK(); 549 550 if (S_ISDIR(ip->i_mode) && (type & tlckXTREE) && 551 !(mp->xflag & COMMIT_PAGE)) { 552 /* 553 * Directory inode is special. It can have both an xtree tlock 554 * and a dtree tlock associated with it. 555 */ 556 dir_xtree = 1; 557 lid = jfs_ip->xtlid; 558 } else 559 lid = mp->lid; 560 561 /* is page not locked by a transaction ? */ 562 if (lid == 0) 563 goto allocateLock; 564 565 jFYI(1, ("txLock: tid:%d ip:0x%p mp:0x%p lid:%d\n", 566 tid, ip, mp, lid)); 567 568 /* is page locked by the requester transaction ? */ 569 tlck = lid_to_tlock(lid); 570 if ((xtid = tlck->tid) == tid) 571 goto grantLock; 572 573 /* 574 * is page locked by anonymous transaction/lock ? 575 * 576 * (page update without transaction (i.e., file write) is 577 * locked under anonymous transaction tid = 0: 578 * anonymous tlocks maintained on anonymous tlock list of 579 * the inode of the page and available to all anonymous 580 * transactions until txCommit() time at which point 581 * they are transferred to the transaction tlock list of 582 * the commiting transaction of the inode) 583 */ 584 if (xtid == 0) { 585 tlck->tid = tid; 586 tblk = tid_to_tblock(tid); 587 /* 588 * The order of the tlocks in the transaction is important 589 * (during truncate, child xtree pages must be freed before 590 * parent's tlocks change the working map). 591 * Take tlock off anonymous list and add to tail of 592 * transaction list 593 * 594 * Note: We really need to get rid of the tid & lid and 595 * use list_head's. This code is getting UGLY! 596 */ 597 if (jfs_ip->atlhead == lid) { 598 if (jfs_ip->atltail == lid) { 599 /* only anonymous txn. 600 * Remove from anon_list 601 */ 602 list_del_init(&jfs_ip->anon_inode_list); 603 } 604 jfs_ip->atlhead = tlck->next; 605 } else { 606 lid_t last; 607 for (last = jfs_ip->atlhead; 608 lid_to_tlock(last)->next != lid; 609 last = lid_to_tlock(last)->next) { 610 assert(last); 611 } 612 lid_to_tlock(last)->next = tlck->next; 613 if (jfs_ip->atltail == lid) 614 jfs_ip->atltail = last; 615 } 616 617 /* insert the tlock at tail of transaction tlock list */ 618 619 if (tblk->next) 620 lid_to_tlock(tblk->last)->next = lid; 621 else 622 tblk->next = lid; 623 tlck->next = 0; 624 tblk->last = lid; 625 626 goto grantLock; 627 } 628 629 goto waitLock; 630 631 /* 632 * allocate a tlock 633 */ 634 allocateLock: 635 lid = txLockAlloc(); 636 tlck = lid_to_tlock(lid); 637 638 /* 639 * initialize tlock 640 */ 641 tlck->tid = tid; 642 643 /* mark tlock for meta-data page */ 644 if (mp->xflag & COMMIT_PAGE) { 645 646 tlck->flag = tlckPAGELOCK; 647 648 /* mark the page dirty and nohomeok */ 649 mark_metapage_dirty(mp); 650 atomic_inc(&mp->nohomeok); 651 652 jFYI(1, 653 ("locking mp = 0x%p, nohomeok = %d tid = %d tlck = 0x%p\n", 654 mp, atomic_read(&mp->nohomeok), tid, tlck)); 655 656 /* if anonymous transaction, and buffer is on the group 657 * commit synclist, mark inode to show this. This will 658 * prevent the buffer from being marked nohomeok for too 659 * long a time. 660 */ 661 if ((tid == 0) && mp->lsn) 662 set_cflag(COMMIT_Synclist, ip); 663 } 664 /* mark tlock for in-memory inode */ 665 else 666 tlck->flag = tlckINODELOCK; 667 668 tlck->type = 0; 669 670 /* bind the tlock and the page */ 671 tlck->ip = ip; 672 tlck->mp = mp; 673 if (dir_xtree) 674 jfs_ip->xtlid = lid; 675 else 676 mp->lid = lid; 677 678 /* 679 * enqueue transaction lock to transaction/inode 680 */ 681 /* insert the tlock at tail of transaction tlock list */ 682 if (tid) { 683 tblk = tid_to_tblock(tid); 684 if (tblk->next) 685 lid_to_tlock(tblk->last)->next = lid; 686 else 687 tblk->next = lid; 688 tlck->next = 0; 689 tblk->last = lid; 690 } 691 /* anonymous transaction: 692 * insert the tlock at head of inode anonymous tlock list 693 */ 694 else { 695 tlck->next = jfs_ip->atlhead; 696 jfs_ip->atlhead = lid; 697 if (tlck->next == 0) { 698 /* This inode's first anonymous transaction */ 699 jfs_ip->atltail = lid; 700 list_add_tail(&jfs_ip->anon_inode_list, 701 &TxAnchor.anon_list); 702 } 703 } 704 705 /* initialize type dependent area for linelock */ 706 linelock = (struct linelock *) & tlck->lock; 707 linelock->next = 0; 708 linelock->flag = tlckLINELOCK; 709 linelock->maxcnt = TLOCKSHORT; 710 linelock->index = 0; 711 712 switch (type & tlckTYPE) { 713 case tlckDTREE: 714 linelock->l2linesize = L2DTSLOTSIZE; 715 break; 716 717 case tlckXTREE: 718 linelock->l2linesize = L2XTSLOTSIZE; 719 720 xtlck = (struct xtlock *) linelock; 721 xtlck->header.offset = 0; 722 xtlck->header.length = 2; 723 724 if (type & tlckNEW) { 725 xtlck->lwm.offset = XTENTRYSTART; 726 } else { 727 if (mp->xflag & COMMIT_PAGE) 728 p = (xtpage_t *) mp->data; 729 else 730 p = &jfs_ip->i_xtroot; 731 xtlck->lwm.offset = 732 le16_to_cpu(p->header.nextindex); 733 } 734 xtlck->lwm.length = 0; /* ! */ 735 xtlck->twm.offset = 0; 736 xtlck->hwm.offset = 0; 737 738 xtlck->index = 2; 739 break; 740 741 case tlckINODE: 742 linelock->l2linesize = L2INODESLOTSIZE; 743 break; 744 745 case tlckDATA: 746 linelock->l2linesize = L2DATASLOTSIZE; 747 break; 748 749 default: 750 jERROR(1, ("UFO tlock:0x%p\n", tlck)); 751 } 752 753 /* 754 * update tlock vector 755 */ 756 grantLock: 757 tlck->type |= type; 758 759 TXN_UNLOCK(); 760 761 return tlck; 762 763 /* 764 * page is being locked by another transaction: 765 */ 766 waitLock: 767 /* Only locks on ipimap or ipaimap should reach here */ 768 /* assert(jfs_ip->fileset == AGGREGATE_I); */ 769 if (jfs_ip->fileset != AGGREGATE_I) { 770 jERROR(1, ("txLock: trying to lock locked page!\n")); 771 dump_mem("ip", ip, sizeof(struct inode)); 772 dump_mem("mp", mp, sizeof(struct metapage)); 773 dump_mem("Locker's tblk", tid_to_tblock(tid), 774 sizeof(struct tblock)); 775 dump_mem("Tlock", tlck, sizeof(struct tlock)); 776 BUG(); 777 } 778 INCREMENT(stattx.waitlock); /* statistics */ 779 release_metapage(mp); 780 781 jEVENT(0, ("txLock: in waitLock, tid = %d, xtid = %d, lid = %d\n", 782 tid, xtid, lid)); 783 TXN_SLEEP_DROP_LOCK(&tid_to_tblock(xtid)->waitor); 784 jEVENT(0, ("txLock: awakened tid = %d, lid = %d\n", tid, lid)); 785 786 return NULL; 787} 788 789 790/* 791 * NAME: txRelease() 792 * 793 * FUNCTION: Release buffers associated with transaction locks, but don't 794 * mark homeok yet. The allows other transactions to modify 795 * buffers, but won't let them go to disk until commit record 796 * actually gets written. 797 * 798 * PARAMETER: 799 * tblk - 800 * 801 * RETURN: Errors from subroutines. 802 */ 803static void txRelease(struct tblock * tblk) 804{ 805 struct metapage *mp; 806 lid_t lid; 807 struct tlock *tlck; 808 809 TXN_LOCK(); 810 811 for (lid = tblk->next; lid; lid = tlck->next) { 812 tlck = lid_to_tlock(lid); 813 if ((mp = tlck->mp) != NULL && 814 (tlck->type & tlckBTROOT) == 0) { 815 assert(mp->xflag & COMMIT_PAGE); 816 mp->lid = 0; 817 } 818 } 819 820 /* 821 * wakeup transactions waiting on a page locked 822 * by the current transaction 823 */ 824 TXN_WAKEUP(&tblk->waitor); 825 826 TXN_UNLOCK(); 827} 828 829 830/* 831 * NAME: txUnlock() 832 * 833 * FUNCTION: Initiates pageout of pages modified by tid in journalled 834 * objects and frees their lockwords. 835 */ 836static void txUnlock(struct tblock * tblk) 837{ 838 struct tlock *tlck; 839 struct linelock *linelock; 840 lid_t lid, next, llid, k; 841 struct metapage *mp; 842 struct jfs_log *log; 843 int difft, diffp; 844 845 jFYI(1, ("txUnlock: tblk = 0x%p\n", tblk)); 846 log = JFS_SBI(tblk->sb)->log; 847 848 /* 849 * mark page under tlock homeok (its log has been written): 850 */ 851 for (lid = tblk->next; lid; lid = next) { 852 tlck = lid_to_tlock(lid); 853 next = tlck->next; 854 855 jFYI(1, ("unlocking lid = %d, tlck = 0x%p\n", lid, tlck)); 856 857 /* unbind page from tlock */ 858 if ((mp = tlck->mp) != NULL && 859 (tlck->type & tlckBTROOT) == 0) { 860 assert(mp->xflag & COMMIT_PAGE); 861 862 /* hold buffer 863 * 864 * It's possible that someone else has the metapage. 865 * The only things were changing are nohomeok, which 866 * is handled atomically, and clsn which is protected 867 * by the LOGSYNC_LOCK. 868 */ 869 hold_metapage(mp, 1); 870 871 assert(atomic_read(&mp->nohomeok) > 0); 872 atomic_dec(&mp->nohomeok); 873 874 /* inherit younger/larger clsn */ 875 LOGSYNC_LOCK(log); 876 if (mp->clsn) { 877 logdiff(difft, tblk->clsn, log); 878 logdiff(diffp, mp->clsn, log); 879 if (difft > diffp) 880 mp->clsn = tblk->clsn; 881 } else 882 mp->clsn = tblk->clsn; 883 LOGSYNC_UNLOCK(log); 884 885 assert(!(tlck->flag & tlckFREEPAGE)); 886 887 if (tlck->flag & tlckWRITEPAGE) { 888 write_metapage(mp); 889 } else { 890 /* release page which has been forced */ 891 release_metapage(mp); 892 } 893 } 894 895 /* insert tlock, and linelock(s) of the tlock if any, 896 * at head of freelist 897 */ 898 TXN_LOCK(); 899 900 llid = ((struct linelock *) & tlck->lock)->next; 901 while (llid) { 902 linelock = (struct linelock *) lid_to_tlock(llid); 903 k = linelock->next; 904 txLockFree(llid); 905 llid = k; 906 } 907 txLockFree(lid); 908 909 TXN_UNLOCK(); 910 } 911 tblk->next = tblk->last = 0; 912 913 /* 914 * remove tblock from logsynclist 915 * (allocation map pages inherited lsn of tblk and 916 * has been inserted in logsync list at txUpdateMap()) 917 */ 918 if (tblk->lsn) { 919 LOGSYNC_LOCK(log); 920 log->count--; 921 list_del(&tblk->synclist); 922 LOGSYNC_UNLOCK(log); 923 } 924} 925 926 927/* 928 * txMaplock() 929 * 930 * function: allocate a transaction lock for freed page/entry; 931 * for freed page, maplock is used as xtlock/dtlock type; 932 */ 933struct tlock *txMaplock(tid_t tid, struct inode *ip, int type) 934{ 935 struct jfs_inode_info *jfs_ip = JFS_IP(ip); 936 lid_t lid; 937 struct tblock *tblk; 938 struct tlock *tlck; 939 struct maplock *maplock; 940 941 TXN_LOCK(); 942 943 /* 944 * allocate a tlock 945 */ 946 lid = txLockAlloc(); 947 tlck = lid_to_tlock(lid); 948 949 /* 950 * initialize tlock 951 */ 952 tlck->tid = tid; 953 954 /* bind the tlock and the object */ 955 tlck->flag = tlckINODELOCK; 956 tlck->ip = ip; 957 tlck->mp = NULL; 958 959 tlck->type = type; 960 961 /* 962 * enqueue transaction lock to transaction/inode 963 */ 964 /* insert the tlock at tail of transaction tlock list */ 965 if (tid) { 966 tblk = tid_to_tblock(tid); 967 if (tblk->next) 968 lid_to_tlock(tblk->last)->next = lid; 969 else 970 tblk->next = lid; 971 tlck->next = 0; 972 tblk->last = lid; 973 } 974 /* anonymous transaction: 975 * insert the tlock at head of inode anonymous tlock list 976 */ 977 else { 978 tlck->next = jfs_ip->atlhead; 979 jfs_ip->atlhead = lid; 980 if (tlck->next == 0) { 981 /* This inode's first anonymous transaction */ 982 jfs_ip->atltail = lid; 983 list_add_tail(&jfs_ip->anon_inode_list, 984 &TxAnchor.anon_list); 985 } 986 } 987 988 TXN_UNLOCK(); 989 990 /* initialize type dependent area for maplock */ 991 maplock = (struct maplock *) & tlck->lock; 992 maplock->next = 0; 993 maplock->maxcnt = 0; 994 maplock->index = 0; 995 996 return tlck; 997} 998 999 1000/* 1001 * txLinelock() 1002 * 1003 * function: allocate a transaction lock for log vector list 1004 */ 1005struct linelock *txLinelock(struct linelock * tlock) 1006{ 1007 lid_t lid; 1008 struct tlock *tlck; 1009 struct linelock *linelock; 1010 1011 TXN_LOCK(); 1012 1013 /* allocate a TxLock structure */ 1014 lid = txLockAlloc(); 1015 tlck = lid_to_tlock(lid); 1016 1017 TXN_UNLOCK(); 1018 1019 /* initialize linelock */ 1020 linelock = (struct linelock *) tlck; 1021 linelock->next = 0; 1022 linelock->flag = tlckLINELOCK; 1023 linelock->maxcnt = TLOCKLONG; 1024 linelock->index = 0; 1025 1026 /* append linelock after tlock */ 1027 linelock->next = tlock->next; 1028 tlock->next = lid; 1029 1030 return linelock; 1031} 1032 1033 1034 1035/* 1036 * transaction commit management 1037 * ----------------------------- 1038 */ 1039 1040/* 1041 * NAME: txCommit() 1042 * 1043 * FUNCTION: commit the changes to the objects specified in 1044 * clist. For journalled segments only the 1045 * changes of the caller are committed, ie by tid. 1046 * for non-journalled segments the data are flushed to 1047 * disk and then the change to the disk inode and indirect 1048 * blocks committed (so blocks newly allocated to the 1049 * segment will be made a part of the segment atomically). 1050 * 1051 * all of the segments specified in clist must be in 1052 * one file system. no more than 6 segments are needed 1053 * to handle all unix svcs. 1054 * 1055 * if the i_nlink field (i.e. disk inode link count) 1056 * is zero, and the type of inode is a regular file or 1057 * directory, or symbolic link , the inode is truncated 1058 * to zero length. the truncation is committed but the 1059 * VM resources are unaffected until it is closed (see 1060 * iput and iclose). 1061 * 1062 * PARAMETER: 1063 * 1064 * RETURN: 1065 * 1066 * serialization: 1067 * on entry the inode lock on each segment is assumed 1068 * to be held. 1069 * 1070 * i/o error: 1071 */ 1072int txCommit(tid_t tid, /* transaction identifier */ 1073 int nip, /* number of inodes to commit */ 1074 struct inode **iplist, /* list of inode to commit */ 1075 int flag) 1076{ 1077 int rc = 0, rc1 = 0; 1078 struct commit cd; 1079 struct jfs_log *log; 1080 struct tblock *tblk; 1081 struct lrd *lrd; 1082 int lsn; 1083 struct inode *ip; 1084 struct jfs_inode_info *jfs_ip; 1085 int k, n; 1086 ino_t top; 1087 struct super_block *sb; 1088 1089 jFYI(1, ("txCommit, tid = %d, flag = %d\n", tid, flag)); 1090 /* is read-only file system ? */ 1091 if (isReadOnly(iplist[0])) { 1092 rc = EROFS; 1093 goto TheEnd; 1094 } 1095 1096 sb = cd.sb = iplist[0]->i_sb; 1097 cd.tid = tid; 1098 1099 if (tid == 0) 1100 tid = txBegin(sb, 0); 1101 tblk = tid_to_tblock(tid); 1102 1103 /* 1104 * initialize commit structure 1105 */ 1106 log = JFS_SBI(sb)->log; 1107 cd.log = log; 1108 1109 /* initialize log record descriptor in commit */ 1110 lrd = &cd.lrd; 1111 lrd->logtid = cpu_to_le32(tblk->logtid); 1112 lrd->backchain = 0; 1113 1114 tblk->xflag |= flag; 1115 1116 if ((flag & (COMMIT_FORCE | COMMIT_SYNC)) == 0) 1117 tblk->xflag |= COMMIT_LAZY; 1118 /* 1119 * prepare non-journaled objects for commit 1120 * 1121 * flush data pages of non-journaled file 1122 * to prevent the file getting non-initialized disk blocks 1123 * in case of crash. 1124 * (new blocks - ) 1125 */ 1126 cd.iplist = iplist; 1127 cd.nip = nip; 1128 1129 /* 1130 * acquire transaction lock on (on-disk) inodes 1131 * 1132 * update on-disk inode from in-memory inode 1133 * acquiring transaction locks for AFTER records 1134 * on the on-disk inode of file object 1135 * 1136 * sort the inodes array by inode number in descending order 1137 * to prevent deadlock when acquiring transaction lock 1138 * of on-disk inodes on multiple on-disk inode pages by 1139 * multiple concurrent transactions 1140 */ 1141 for (k = 0; k < cd.nip; k++) { 1142 top = (cd.iplist[k])->i_ino; 1143 for (n = k + 1; n < cd.nip; n++) { 1144 ip = cd.iplist[n]; 1145 if (ip->i_ino > top) { 1146 top = ip->i_ino; 1147 cd.iplist[n] = cd.iplist[k]; 1148 cd.iplist[k] = ip; 1149 } 1150 } 1151 1152 ip = cd.iplist[k]; 1153 jfs_ip = JFS_IP(ip); 1154 1155 /* 1156 * BUGBUG - Should we call filemap_fdatasync here instead 1157 * of fsync_inode_data? 1158 * If we do, we have a deadlock condition since we may end 1159 * up recursively calling jfs_get_block with the IWRITELOCK 1160 * held. We may be able to do away with IWRITELOCK while 1161 * committing transactions and use i_sem instead. 1162 */ 1163 if ((!S_ISDIR(ip->i_mode)) 1164 && (tblk->flag & COMMIT_DELETE) == 0) 1165 fsync_inode_data_buffers(ip); 1166 1167 /* 1168 * Mark inode as not dirty. It will still be on the dirty 1169 * inode list, but we'll know not to commit it again unless 1170 * it gets marked dirty again 1171 */ 1172 clear_cflag(COMMIT_Dirty, ip); 1173 1174 /* inherit anonymous tlock(s) of inode */ 1175 if (jfs_ip->atlhead) { 1176 lid_to_tlock(jfs_ip->atltail)->next = tblk->next; 1177 tblk->next = jfs_ip->atlhead; 1178 if (!tblk->last) 1179 tblk->last = jfs_ip->atltail; 1180 jfs_ip->atlhead = jfs_ip->atltail = 0; 1181 TXN_LOCK(); 1182 list_del_init(&jfs_ip->anon_inode_list); 1183 TXN_UNLOCK(); 1184 } 1185 1186 /* 1187 * acquire transaction lock on on-disk inode page 1188 * (become first tlock of the tblk's tlock list) 1189 */ 1190 if (((rc = diWrite(tid, ip)))) 1191 goto out; 1192 } 1193 1194 /* 1195 * write log records from transaction locks 1196 * 1197 * txUpdateMap() resets XAD_NEW in XAD. 1198 */ 1199 if ((rc = txLog(log, tblk, &cd))) 1200 goto TheEnd; 1201 1202 /* 1203 * Ensure that inode isn't reused before 1204 * lazy commit thread finishes processing 1205 */ 1206 if (tblk->xflag & (COMMIT_CREATE | COMMIT_DELETE)) 1207 atomic_inc(&tblk->ip->i_count); 1208 1209 ASSERT((!(tblk->xflag & COMMIT_DELETE)) || 1210 ((tblk->ip->i_nlink == 0) && 1211 !test_cflag(COMMIT_Nolink, tblk->ip))); 1212 1213 /* 1214 * write COMMIT log record 1215 */ 1216 lrd->type = cpu_to_le16(LOG_COMMIT); 1217 lrd->length = 0; 1218 lsn = lmLog(log, tblk, lrd, NULL); 1219 1220 lmGroupCommit(log, tblk); 1221 1222 /* 1223 * - transaction is now committed - 1224 */ 1225 1226 /* 1227 * force pages in careful update 1228 * (imap addressing structure update) 1229 */ 1230 if (flag & COMMIT_FORCE) 1231 txForce(tblk); 1232 1233 /* 1234 * update allocation map. 1235 * 1236 * update inode allocation map and inode: 1237 * free pager lock on memory object of inode if any. 1238 * update block allocation map. 1239 * 1240 * txUpdateMap() resets XAD_NEW in XAD. 1241 */ 1242 if (tblk->xflag & COMMIT_FORCE) 1243 txUpdateMap(tblk); 1244 1245 /* 1246 * free transaction locks and pageout/free pages 1247 */ 1248 txRelease(tblk); 1249 1250 if ((tblk->flag & tblkGC_LAZY) == 0) 1251 txUnlock(tblk); 1252 1253 1254 /* 1255 * reset in-memory object state 1256 */ 1257 for (k = 0; k < cd.nip; k++) { 1258 ip = cd.iplist[k]; 1259 jfs_ip = JFS_IP(ip); 1260 1261 /* 1262 * reset in-memory inode state 1263 */ 1264 jfs_ip->bxflag = 0; 1265 jfs_ip->blid = 0; 1266 } 1267 1268 out: 1269 if (rc != 0) 1270 txAbortCommit(&cd, rc); 1271 else 1272 rc = rc1; 1273 1274 TheEnd: 1275 jFYI(1, ("txCommit: tid = %d, returning %d\n", tid, rc)); 1276 return rc; 1277} 1278 1279 1280/* 1281 * NAME: txLog() 1282 * 1283 * FUNCTION: Writes AFTER log records for all lines modified 1284 * by tid for segments specified by inodes in comdata. 1285 * Code assumes only WRITELOCKS are recorded in lockwords. 1286 * 1287 * PARAMETERS: 1288 * 1289 * RETURN : 1290 */ 1291static int txLog(struct jfs_log * log, struct tblock * tblk, struct commit * cd) 1292{ 1293 int rc = 0; 1294 struct inode *ip; 1295 lid_t lid; 1296 struct tlock *tlck; 1297 struct lrd *lrd = &cd->lrd; 1298 1299 /* 1300 * write log record(s) for each tlock of transaction, 1301 */ 1302 for (lid = tblk->next; lid; lid = tlck->next) { 1303 tlck = lid_to_tlock(lid); 1304 1305 tlck->flag |= tlckLOG; 1306 1307 /* initialize lrd common */ 1308 ip = tlck->ip; 1309 lrd->aggregate = cpu_to_le32(kdev_t_to_nr(ip->i_dev)); 1310 lrd->log.redopage.fileset = cpu_to_le32(JFS_IP(ip)->fileset); 1311 lrd->log.redopage.inode = cpu_to_le32(ip->i_ino); 1312 1313 if (tlck->mp) 1314 hold_metapage(tlck->mp, 0); 1315 1316 /* write log record of page from the tlock */ 1317 switch (tlck->type & tlckTYPE) { 1318 case tlckXTREE: 1319 xtLog(log, tblk, lrd, tlck); 1320 break; 1321 1322 case tlckDTREE: 1323 dtLog(log, tblk, lrd, tlck); 1324 break; 1325 1326 case tlckINODE: 1327 diLog(log, tblk, lrd, tlck, cd); 1328 break; 1329 1330 case tlckMAP: 1331 mapLog(log, tblk, lrd, tlck); 1332 break; 1333 1334 case tlckDATA: 1335 dataLog(log, tblk, lrd, tlck); 1336 break; 1337 1338 default: 1339 jERROR(1, ("UFO tlock:0x%p\n", tlck)); 1340 } 1341 if (tlck->mp) 1342 release_metapage(tlck->mp); 1343 } 1344 1345 return rc; 1346} 1347 1348 1349/* 1350 * diLog() 1351 * 1352 * function: log inode tlock and format maplock to update bmap; 1353 */ 1354int diLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd, 1355 struct tlock * tlck, struct commit * cd) 1356{ 1357 int rc = 0; 1358 struct metapage *mp; 1359 pxd_t *pxd; 1360 struct pxd_lock *pxdlock; 1361 1362 mp = tlck->mp; 1363 1364 /* initialize as REDOPAGE record format */ 1365 lrd->log.redopage.type = cpu_to_le16(LOG_INODE); 1366 lrd->log.redopage.l2linesize = cpu_to_le16(L2INODESLOTSIZE); 1367 1368 pxd = &lrd->log.redopage.pxd; 1369 1370 /* 1371 * inode after image 1372 */ 1373 if (tlck->type & tlckENTRY) { 1374 /* log after-image for logredo(): */ 1375 lrd->type = cpu_to_le16(LOG_REDOPAGE); 1376// *pxd = mp->cm_pxd; 1377 PXDaddress(pxd, mp->index); 1378 PXDlength(pxd, 1379 mp->logical_size >> tblk->sb->s_blocksize_bits); 1380 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck)); 1381 1382 /* mark page as homeward bound */ 1383 tlck->flag |= tlckWRITEPAGE; 1384 } else if (tlck->type & tlckFREE) { 1385 /* 1386 * free inode extent 1387 * 1388 * (pages of the freed inode extent have been invalidated and 1389 * a maplock for free of the extent has been formatted at 1390 * txLock() time); 1391 * 1392 * the tlock had been acquired on the inode allocation map page 1393 * (iag) that specifies the freed extent, even though the map 1394 * page is not itself logged, to prevent pageout of the map 1395 * page before the log; 1396 */ 1397 assert(tlck->type & tlckFREE); 1398 1399 /* log LOG_NOREDOINOEXT of the freed inode extent for 1400 * logredo() to start NoRedoPage filters, and to update 1401 * imap and bmap for free of the extent; 1402 */ 1403 lrd->type = cpu_to_le16(LOG_NOREDOINOEXT); 1404 /* 1405 * For the LOG_NOREDOINOEXT record, we need 1406 * to pass the IAG number and inode extent 1407 * index (within that IAG) from which the 1408 * the extent being released. These have been 1409 * passed to us in the iplist[1] and iplist[2]. 1410 */ 1411 lrd->log.noredoinoext.iagnum = 1412 cpu_to_le32((u32) (size_t) cd->iplist[1]); 1413 lrd->log.noredoinoext.inoext_idx = 1414 cpu_to_le32((u32) (size_t) cd->iplist[2]); 1415 1416 pxdlock = (struct pxd_lock *) & tlck->lock; 1417 *pxd = pxdlock->pxd; 1418 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL)); 1419 1420 /* update bmap */ 1421 tlck->flag |= tlckUPDATEMAP; 1422 1423 /* mark page as homeward bound */ 1424 tlck->flag |= tlckWRITEPAGE; 1425 } else { 1426 jERROR(2, ("diLog: UFO type tlck:0x%p\n", tlck)); 1427 } 1428#ifdef _JFS_WIP 1429 /* 1430 * alloc/free external EA extent 1431 * 1432 * a maplock for txUpdateMap() to update bPWMAP for alloc/free 1433 * of the extent has been formatted at txLock() time; 1434 */ 1435 else { 1436 assert(tlck->type & tlckEA); 1437 1438 /* log LOG_UPDATEMAP for logredo() to update bmap for 1439 * alloc of new (and free of old) external EA extent; 1440 */ 1441 lrd->type = cpu_to_le16(LOG_UPDATEMAP); 1442 pxdlock = (struct pxd_lock *) & tlck->lock; 1443 nlock = pxdlock->index; 1444 for (i = 0; i < nlock; i++, pxdlock++) { 1445 if (pxdlock->flag & mlckALLOCPXD) 1446 lrd->log.updatemap.type = 1447 cpu_to_le16(LOG_ALLOCPXD); 1448 else 1449 lrd->log.updatemap.type = 1450 cpu_to_le16(LOG_FREEPXD); 1451 lrd->log.updatemap.nxd = cpu_to_le16(1); 1452 lrd->log.updatemap.pxd = pxdlock->pxd; 1453 lrd->backchain = 1454 cpu_to_le32(lmLog(log, tblk, lrd, NULL)); 1455 } 1456 1457 /* update bmap */ 1458 tlck->flag |= tlckUPDATEMAP; 1459 } 1460#endif /* _JFS_WIP */ 1461 1462 return rc; 1463} 1464 1465 1466/* 1467 * dataLog() 1468 * 1469 * function: log data tlock 1470 */ 1471int dataLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd, 1472 struct tlock * tlck) 1473{ 1474 struct metapage *mp; 1475 pxd_t *pxd; 1476 int rc; 1477 s64 xaddr; 1478 int xflag; 1479 s32 xlen; 1480 1481 mp = tlck->mp; 1482 1483 /* initialize as REDOPAGE record format */ 1484 lrd->log.redopage.type = cpu_to_le16(LOG_DATA); 1485 lrd->log.redopage.l2linesize = cpu_to_le16(L2DATASLOTSIZE); 1486 1487 pxd = &lrd->log.redopage.pxd; 1488 1489 /* log after-image for logredo(): */ 1490 lrd->type = cpu_to_le16(LOG_REDOPAGE); 1491 1492 if (JFS_IP(tlck->ip)->next_index < MAX_INLINE_DIRTABLE_ENTRY) { 1493 /* 1494 * The table has been truncated, we've must have deleted 1495 * the last entry, so don't bother logging this 1496 */ 1497 mp->lid = 0; 1498 atomic_dec(&mp->nohomeok); 1499 discard_metapage(mp); 1500 tlck->mp = 0; 1501 return 0; 1502 } 1503 1504 rc = xtLookup(tlck->ip, mp->index, 1, &xflag, &xaddr, &xlen, 1); 1505 if (rc || (xlen == 0)) { 1506 jERROR(1, ("dataLog: can't find physical address\n")); 1507 return 0; 1508 } 1509 1510 PXDaddress(pxd, xaddr); 1511 PXDlength(pxd, mp->logical_size >> tblk->sb->s_blocksize_bits); 1512 1513 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck)); 1514 1515 /* mark page as homeward bound */ 1516 tlck->flag |= tlckWRITEPAGE; 1517 1518 return 0; 1519} 1520 1521 1522/* 1523 * dtLog() 1524 * 1525 * function: log dtree tlock and format maplock to update bmap; 1526 */ 1527void dtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd, 1528 struct tlock * tlck) 1529{ 1530 struct inode *ip; 1531 struct metapage *mp; 1532 struct pxd_lock *pxdlock; 1533 pxd_t *pxd; 1534 1535 ip = tlck->ip; 1536 mp = tlck->mp; 1537 1538 /* initialize as REDOPAGE/NOREDOPAGE record format */ 1539 lrd->log.redopage.type = cpu_to_le16(LOG_DTREE); 1540 lrd->log.redopage.l2linesize = cpu_to_le16(L2DTSLOTSIZE); 1541 1542 pxd = &lrd->log.redopage.pxd; 1543 1544 if (tlck->type & tlckBTROOT) 1545 lrd->log.redopage.type |= cpu_to_le16(LOG_BTROOT); 1546 1547 /* 1548 * page extension via relocation: entry insertion; 1549 * page extension in-place: entry insertion; 1550 * new right page from page split, reinitialized in-line 1551 * root from root page split: entry insertion; 1552 */ 1553 if (tlck->type & (tlckNEW | tlckEXTEND)) { 1554 /* log after-image of the new page for logredo(): 1555 * mark log (LOG_NEW) for logredo() to initialize 1556 * freelist and update bmap for alloc of the new page; 1557 */ 1558 lrd->type = cpu_to_le16(LOG_REDOPAGE); 1559 if (tlck->type & tlckEXTEND) 1560 lrd->log.redopage.type |= cpu_to_le16(LOG_EXTEND); 1561 else 1562 lrd->log.redopage.type |= cpu_to_le16(LOG_NEW); 1563// *pxd = mp->cm_pxd; 1564 PXDaddress(pxd, mp->index); 1565 PXDlength(pxd, 1566 mp->logical_size >> tblk->sb->s_blocksize_bits); 1567 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck)); 1568 1569 /* format a maplock for txUpdateMap() to update bPMAP for 1570 * alloc of the new page; 1571 */ 1572 if (tlck->type & tlckBTROOT) 1573 return; 1574 tlck->flag |= tlckUPDATEMAP; 1575 pxdlock = (struct pxd_lock *) & tlck->lock; 1576 pxdlock->flag = mlckALLOCPXD; 1577 pxdlock->pxd = *pxd; 1578 1579 pxdlock->index = 1; 1580 1581 /* mark page as homeward bound */ 1582 tlck->flag |= tlckWRITEPAGE; 1583 return; 1584 } 1585 1586 /* 1587 * entry insertion/deletion, 1588 * sibling page link update (old right page before split); 1589 */ 1590 if (tlck->type & (tlckENTRY | tlckRELINK)) { 1591 /* log after-image for logredo(): */ 1592 lrd->type = cpu_to_le16(LOG_REDOPAGE); 1593 PXDaddress(pxd, mp->index); 1594 PXDlength(pxd, 1595 mp->logical_size >> tblk->sb->s_blocksize_bits); 1596 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck)); 1597 1598 /* mark page as homeward bound */ 1599 tlck->flag |= tlckWRITEPAGE; 1600 return; 1601 } 1602 1603 /* 1604 * page deletion: page has been invalidated 1605 * page relocation: source extent 1606 * 1607 * a maplock for free of the page has been formatted 1608 * at txLock() time); 1609 */ 1610 if (tlck->type & (tlckFREE | tlckRELOCATE)) { 1611 /* log LOG_NOREDOPAGE of the deleted page for logredo() 1612 * to start NoRedoPage filter and to update bmap for free 1613 * of the deletd page 1614 */ 1615 lrd->type = cpu_to_le16(LOG_NOREDOPAGE); 1616 pxdlock = (struct pxd_lock *) & tlck->lock; 1617 *pxd = pxdlock->pxd; 1618 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL)); 1619 1620 /* a maplock for txUpdateMap() for free of the page 1621 * has been formatted at txLock() time; 1622 */ 1623 tlck->flag |= tlckUPDATEMAP; 1624 } 1625 return; 1626} 1627 1628 1629/* 1630 * xtLog() 1631 * 1632 * function: log xtree tlock and format maplock to update bmap; 1633 */ 1634void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd, 1635 struct tlock * tlck) 1636{ 1637 struct inode *ip; 1638 struct metapage *mp; 1639 xtpage_t *p; 1640 struct xtlock *xtlck; 1641 struct maplock *maplock; 1642 struct xdlistlock *xadlock; 1643 struct pxd_lock *pxdlock; 1644 pxd_t *pxd; 1645 int next, lwm, hwm; 1646 1647 ip = tlck->ip; 1648 mp = tlck->mp; 1649 1650 /* initialize as REDOPAGE/NOREDOPAGE record format */ 1651 lrd->log.redopage.type = cpu_to_le16(LOG_XTREE); 1652 lrd->log.redopage.l2linesize = cpu_to_le16(L2XTSLOTSIZE); 1653 1654 pxd = &lrd->log.redopage.pxd; 1655 1656 if (tlck->type & tlckBTROOT) { 1657 lrd->log.redopage.type |= cpu_to_le16(LOG_BTROOT); 1658 p = &JFS_IP(ip)->i_xtroot; 1659 if (S_ISDIR(ip->i_mode)) 1660 lrd->log.redopage.type |= 1661 cpu_to_le16(LOG_DIR_XTREE); 1662 } else 1663 p = (xtpage_t *) mp->data; 1664 next = le16_to_cpu(p->header.nextindex); 1665 1666 xtlck = (struct xtlock *) & tlck->lock; 1667 1668 maplock = (struct maplock *) & tlck->lock; 1669 xadlock = (struct xdlistlock *) maplock; 1670 1671 /* 1672 * entry insertion/extension; 1673 * sibling page link update (old right page before split); 1674 */ 1675 if (tlck->type & (tlckNEW | tlckGROW | tlckRELINK)) { 1676 /* log after-image for logredo(): 1677 * logredo() will update bmap for alloc of new/extended 1678 * extents (XAD_NEW|XAD_EXTEND) of XAD[lwm:next) from 1679 * after-image of XADlist; 1680 * logredo() resets (XAD_NEW|XAD_EXTEND) flag when 1681 * applying the after-image to the meta-data page. 1682 */ 1683 lrd->type = cpu_to_le16(LOG_REDOPAGE); 1684// *pxd = mp->cm_pxd; 1685 PXDaddress(pxd, mp->index); 1686 PXDlength(pxd, 1687 mp->logical_size >> tblk->sb->s_blocksize_bits); 1688 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck)); 1689 1690 /* format a maplock for txUpdateMap() to update bPMAP 1691 * for alloc of new/extended extents of XAD[lwm:next) 1692 * from the page itself; 1693 * txUpdateMap() resets (XAD_NEW|XAD_EXTEND) flag. 1694 */ 1695 lwm = xtlck->lwm.offset; 1696 if (lwm == 0) 1697 lwm = XTPAGEMAXSLOT; 1698 1699 if (lwm == next) 1700 goto out; 1701 assert(lwm < next); 1702 tlck->flag |= tlckUPDATEMAP; 1703 xadlock->flag = mlckALLOCXADLIST; 1704 xadlock->count = next - lwm; 1705 if ((xadlock->count <= 2) && (tblk->xflag & COMMIT_LAZY)) { 1706 int i; 1707 /* 1708 * Lazy commit may allow xtree to be modified before 1709 * txUpdateMap runs. Copy xad into linelock to 1710 * preserve correct data. 1711 */ 1712 xadlock->xdlist = &xtlck->pxdlock; 1713 memcpy(xadlock->xdlist, &p->xad[lwm], 1714 sizeof(xad_t) * xadlock->count); 1715 1716 for (i = 0; i < xadlock->count; i++) 1717 p->xad[lwm + i].flag &= 1718 ~(XAD_NEW | XAD_EXTENDED); 1719 } else { 1720 /* 1721 * xdlist will point to into inode's xtree, ensure 1722 * that transaction is not committed lazily. 1723 */ 1724 xadlock->xdlist = &p->xad[lwm]; 1725 tblk->xflag &= ~COMMIT_LAZY; 1726 } 1727 jFYI(1, 1728 ("xtLog: alloc ip:0x%p mp:0x%p tlck:0x%p lwm:%d count:%d\n", 1729 tlck->ip, mp, tlck, lwm, xadlock->count)); 1730 1731 maplock->index = 1; 1732 1733 out: 1734 /* mark page as homeward bound */ 1735 tlck->flag |= tlckWRITEPAGE; 1736 1737 return; 1738 } 1739 1740 /* 1741 * page deletion: file deletion/truncation (ref. xtTruncate()) 1742 * 1743 * (page will be invalidated after log is written and bmap 1744 * is updated from the page); 1745 */ 1746 if (tlck->type & tlckFREE) { 1747 /* LOG_NOREDOPAGE log for NoRedoPage filter: 1748 * if page free from file delete, NoRedoFile filter from 1749 * inode image of zero link count will subsume NoRedoPage 1750 * filters for each page; 1751 * if page free from file truncattion, write NoRedoPage 1752 * filter; 1753 * 1754 * upadte of block allocation map for the page itself: 1755 * if page free from deletion and truncation, LOG_UPDATEMAP 1756 * log for the page itself is generated from processing 1757 * its parent page xad entries; 1758 */ 1759 /* if page free from file truncation, log LOG_NOREDOPAGE 1760 * of the deleted page for logredo() to start NoRedoPage 1761 * filter for the page; 1762 */ 1763 if (tblk->xflag & COMMIT_TRUNCATE) { 1764 /* write NOREDOPAGE for the page */ 1765 lrd->type = cpu_to_le16(LOG_NOREDOPAGE); 1766 PXDaddress(pxd, mp->index); 1767 PXDlength(pxd, 1768 mp->logical_size >> tblk->sb-> 1769 s_blocksize_bits); 1770 lrd->backchain = 1771 cpu_to_le32(lmLog(log, tblk, lrd, NULL)); 1772 1773 if (tlck->type & tlckBTROOT) { 1774 /* Empty xtree must be logged */ 1775 lrd->type = cpu_to_le16(LOG_REDOPAGE); 1776 lrd->backchain = 1777 cpu_to_le32(lmLog(log, tblk, lrd, tlck)); 1778 } 1779 } 1780 1781 /* init LOG_UPDATEMAP of the freed extents 1782 * XAD[XTENTRYSTART:hwm) from the deleted page itself 1783 * for logredo() to update bmap; 1784 */ 1785 lrd->type = cpu_to_le16(LOG_UPDATEMAP); 1786 lrd->log.updatemap.type = cpu_to_le16(LOG_FREEXADLIST); 1787 xtlck = (struct xtlock *) & tlck->lock; 1788 hwm = xtlck->hwm.offset; 1789 lrd->log.updatemap.nxd = 1790 cpu_to_le16(hwm - XTENTRYSTART + 1); 1791 /* reformat linelock for lmLog() */ 1792 xtlck->header.offset = XTENTRYSTART; 1793 xtlck->header.length = hwm - XTENTRYSTART + 1; 1794 xtlck->index = 1; 1795 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck)); 1796 1797 /* format a maplock for txUpdateMap() to update bmap 1798 * to free extents of XAD[XTENTRYSTART:hwm) from the 1799 * deleted page itself; 1800 */ 1801 tlck->flag |= tlckUPDATEMAP; 1802 xadlock->flag = mlckFREEXADLIST; 1803 xadlock->count = hwm - XTENTRYSTART + 1; 1804 if ((xadlock->count <= 2) && (tblk->xflag & COMMIT_LAZY)) { 1805 /* 1806 * Lazy commit may allow xtree to be modified before 1807 * txUpdateMap runs. Copy xad into linelock to 1808 * preserve correct data. 1809 */ 1810 xadlock->xdlist = &xtlck->pxdlock; 1811 memcpy(xadlock->xdlist, &p->xad[XTENTRYSTART], 1812 sizeof(xad_t) * xadlock->count); 1813 } else { 1814 /* 1815 * xdlist will point to into inode's xtree, ensure 1816 * that transaction is not committed lazily unless 1817 * we're deleting the inode (unlink). In that case 1818 * we have special logic for the inode to be 1819 * unlocked by the lazy commit thread. 1820 */ 1821 xadlock->xdlist = &p->xad[XTENTRYSTART]; 1822 if ((tblk->xflag & COMMIT_LAZY) && 1823 (tblk->xflag & COMMIT_DELETE) && 1824 (tblk->ip == ip)) 1825 set_cflag(COMMIT_Holdlock, ip); 1826 else 1827 tblk->xflag &= ~COMMIT_LAZY; 1828 } 1829 jFYI(1, 1830 ("xtLog: free ip:0x%p mp:0x%p count:%d lwm:2\n", 1831 tlck->ip, mp, xadlock->count)); 1832 1833 maplock->index = 1; 1834 1835 /* mark page as invalid */ 1836 if (((tblk->xflag & COMMIT_PWMAP) || S_ISDIR(ip->i_mode)) 1837 && !(tlck->type & tlckBTROOT)) 1838 tlck->flag |= tlckFREEPAGE; 1839 /* 1840 else (tblk->xflag & COMMIT_PMAP) 1841 ? release the page; 1842 */ 1843 return; 1844 } 1845 1846 /* 1847 * page/entry truncation: file truncation (ref. xtTruncate()) 1848 * 1849 * |----------+------+------+---------------| 1850 * | | | 1851 * | | hwm - hwm before truncation 1852 * | next - truncation point 1853 * lwm - lwm before truncation 1854 * header ? 1855 */ 1856 if (tlck->type & tlckTRUNCATE) { 1857 pxd_t tpxd; /* truncated extent of xad */ 1858 int twm; 1859 1860 /* 1861 * For truncation the entire linelock may be used, so it would 1862 * be difficult to store xad list in linelock itself. 1863 * Therefore, we'll just force transaction to be committed 1864 * synchronously, so that xtree pages won't be changed before 1865 * txUpdateMap runs. 1866 */ 1867 tblk->xflag &= ~COMMIT_LAZY; 1868 lwm = xtlck->lwm.offset; 1869 if (lwm == 0) 1870 lwm = XTPAGEMAXSLOT; 1871 hwm = xtlck->hwm.offset; 1872 twm = xtlck->twm.offset; 1873 1874 /* 1875 * write log records 1876 */ 1877 /* 1878 * allocate entries XAD[lwm:next]: 1879 */ 1880 if (lwm < next) { 1881 /* log after-image for logredo(): 1882 * logredo() will update bmap for alloc of new/extended 1883 * extents (XAD_NEW|XAD_EXTEND) of XAD[lwm:next) from 1884 * after-image of XADlist; 1885 * logredo() resets (XAD_NEW|XAD_EXTEND) flag when 1886 * applying the after-image to the meta-data page. 1887 */ 1888 lrd->type = cpu_to_le16(LOG_REDOPAGE); 1889 PXDaddress(pxd, mp->index); 1890 PXDlength(pxd, 1891 mp->logical_size >> tblk->sb-> 1892 s_blocksize_bits); 1893 lrd->backchain = 1894 cpu_to_le32(lmLog(log, tblk, lrd, tlck)); 1895 } 1896 1897 /* 1898 * truncate entry XAD[twm == next - 1]: 1899 */ 1900 if (twm == next - 1) { 1901 /* init LOG_UPDATEMAP for logredo() to update bmap for 1902 * free of truncated delta extent of the truncated 1903 * entry XAD[next - 1]: 1904 * (xtlck->pxdlock = truncated delta extent); 1905 */ 1906 pxdlock = (struct pxd_lock *) & xtlck->pxdlock; 1907 /* assert(pxdlock->type & tlckTRUNCATE); */ 1908 lrd->type = cpu_to_le16(LOG_UPDATEMAP); 1909 lrd->log.updatemap.type = cpu_to_le16(LOG_FREEPXD); 1910 lrd->log.updatemap.nxd = cpu_to_le16(1); 1911 lrd->log.updatemap.pxd = pxdlock->pxd; 1912 tpxd = pxdlock->pxd; /* save to format maplock */ 1913 lrd->backchain = 1914 cpu_to_le32(lmLog(log, tblk, lrd, NULL)); 1915 } 1916 1917 /* 1918 * free entries XAD[next:hwm]: 1919 */ 1920 if (hwm >= next) { 1921 /* init LOG_UPDATEMAP of the freed extents 1922 * XAD[next:hwm] from the deleted page itself 1923 * for logredo() to update bmap; 1924 */ 1925 lrd->type = cpu_to_le16(LOG_UPDATEMAP); 1926 lrd->log.updatemap.type = 1927 cpu_to_le16(LOG_FREEXADLIST); 1928 xtlck = (struct xtlock *) & tlck->lock; 1929 hwm = xtlck->hwm.offset; 1930 lrd->log.updatemap.nxd = 1931 cpu_to_le16(hwm - next + 1); 1932 /* reformat linelock for lmLog() */ 1933 xtlck->header.offset = next; 1934 xtlck->header.length = hwm - next + 1; 1935 xtlck->index = 1; 1936 lrd->backchain = 1937 cpu_to_le32(lmLog(log, tblk, lrd, tlck)); 1938 } 1939 1940 /* 1941 * format maplock(s) for txUpdateMap() to update bmap 1942 */ 1943 maplock->index = 0; 1944 1945 /* 1946 * allocate entries XAD[lwm:next): 1947 */ 1948 if (lwm < next) { 1949 /* format a maplock for txUpdateMap() to update bPMAP 1950 * for alloc of new/extended extents of XAD[lwm:next) 1951 * from the page itself; 1952 * txUpdateMap() resets (XAD_NEW|XAD_EXTEND) flag. 1953 */ 1954 tlck->flag |= tlckUPDATEMAP; 1955 xadlock->flag = mlckALLOCXADLIST; 1956 xadlock->count = next - lwm; 1957 xadlock->xdlist = &p->xad[lwm]; 1958 1959 jFYI(1, 1960 ("xtLog: alloc ip:0x%p mp:0x%p count:%d lwm:%d next:%d\n", 1961 tlck->ip, mp, xadlock->count, lwm, next)); 1962 maplock->index++; 1963 xadlock++; 1964 } 1965 1966 /* 1967 * truncate entry XAD[twm == next - 1]: 1968 */ 1969 if (twm == next - 1) { 1970 struct pxd_lock *pxdlock; 1971 1972 /* format a maplock for txUpdateMap() to update bmap 1973 * to free truncated delta extent of the truncated 1974 * entry XAD[next - 1]; 1975 * (xtlck->pxdlock = truncated delta extent); 1976 */ 1977 tlck->flag |= tlckUPDATEMAP; 1978 pxdlock = (struct pxd_lock *) xadlock; 1979 pxdlock->flag = mlckFREEPXD; 1980 pxdlock->count = 1; 1981 pxdlock->pxd = tpxd; 1982 1983 jFYI(1, 1984 ("xtLog: truncate ip:0x%p mp:0x%p count:%d hwm:%d\n", 1985 ip, mp, pxdlock->count, hwm)); 1986 maplock->index++; 1987 xadlock++; 1988 } 1989 1990 /* 1991 * free entries XAD[next:hwm]: 1992 */ 1993 if (hwm >= next) { 1994 /* format a maplock for txUpdateMap() to update bmap 1995 * to free extents of XAD[next:hwm] from thedeleted 1996 * page itself; 1997 */ 1998 tlck->flag |= tlckUPDATEMAP; 1999 xadlock->flag = mlckFREEXADLIST; 2000 xadlock->count = hwm - next + 1; 2001 xadlock->xdlist = &p->xad[next]; 2002 2003 jFYI(1, 2004 ("xtLog: free ip:0x%p mp:0x%p count:%d next:%d hwm:%d\n", 2005 tlck->ip, mp, xadlock->count, next, hwm)); 2006 maplock->index++; 2007 } 2008 2009 /* mark page as homeward bound */ 2010 tlck->flag |= tlckWRITEPAGE; 2011 } 2012 return; 2013} 2014 2015 2016/* 2017 * mapLog() 2018 * 2019 * function: log from maplock of freed data extents; 2020 */ 2021void mapLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd, 2022 struct tlock * tlck) 2023{ 2024 struct pxd_lock *pxdlock; 2025 int i, nlock; 2026 pxd_t *pxd; 2027 2028 /* 2029 * page relocation: free the source page extent 2030 * 2031 * a maplock for txUpdateMap() for free of the page 2032 * has been formatted at txLock() time saving the src 2033 * relocated page address; 2034 */ 2035 if (tlck->type & tlckRELOCATE) { 2036 /* log LOG_NOREDOPAGE of the old relocated page 2037 * for logredo() to start NoRedoPage filter; 2038 */ 2039 lrd->type = cpu_to_le16(LOG_NOREDOPAGE); 2040 pxdlock = (struct pxd_lock *) & tlck->lock; 2041 pxd = &lrd->log.redopage.pxd; 2042 *pxd = pxdlock->pxd; 2043 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL)); 2044 2045 /* (N.B. currently, logredo() does NOT update bmap 2046 * for free of the page itself for (LOG_XTREE|LOG_NOREDOPAGE); 2047 * if page free from relocation, LOG_UPDATEMAP log is 2048 * specifically generated now for logredo() 2049 * to update bmap for free of src relocated page; 2050 * (new flag LOG_RELOCATE may be introduced which will 2051 * inform logredo() to start NORedoPage filter and also 2052 * update block allocation map at the same time, thus 2053 * avoiding an extra log write); 2054 */ 2055 lrd->type = cpu_to_le16(LOG_UPDATEMAP); 2056 lrd->log.updatemap.type = cpu_to_le16(LOG_FREEPXD); 2057 lrd->log.updatemap.nxd = cpu_to_le16(1); 2058 lrd->log.updatemap.pxd = pxdlock->pxd; 2059 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL)); 2060 2061 /* a maplock for txUpdateMap() for free of the page 2062 * has been formatted at txLock() time; 2063 */ 2064 tlck->flag |= tlckUPDATEMAP; 2065 return; 2066 } 2067 /* 2068 2069 * Otherwise it's not a relocate request 2070 * 2071 */ 2072 else { 2073 /* log LOG_UPDATEMAP for logredo() to update bmap for 2074 * free of truncated/relocated delta extent of the data; 2075 * e.g.: external EA extent, relocated/truncated extent 2076 * from xtTailgate(); 2077 */ 2078 lrd->type = cpu_to_le16(LOG_UPDATEMAP); 2079 pxdlock = (struct pxd_lock *) & tlck->lock; 2080 nlock = pxdlock->index; 2081 for (i = 0; i < nlock; i++, pxdlock++) { 2082 if (pxdlock->flag & mlckALLOCPXD) 2083 lrd->log.updatemap.type = 2084 cpu_to_le16(LOG_ALLOCPXD); 2085 else 2086 lrd->log.updatemap.type = 2087 cpu_to_le16(LOG_FREEPXD); 2088 lrd->log.updatemap.nxd = cpu_to_le16(1); 2089 lrd->log.updatemap.pxd = pxdlock->pxd; 2090 lrd->backchain = 2091 cpu_to_le32(lmLog(log, tblk, lrd, NULL)); 2092 jFYI(1, ("mapLog: xaddr:0x%lx xlen:0x%x\n", 2093 (ulong) addressPXD(&pxdlock->pxd), 2094 lengthPXD(&pxdlock->pxd))); 2095 } 2096 2097 /* update bmap */ 2098 tlck->flag |= tlckUPDATEMAP; 2099 } 2100} 2101 2102 2103/* 2104 * txEA() 2105 * 2106 * function: acquire maplock for EA/ACL extents or 2107 * set COMMIT_INLINE flag; 2108 */ 2109void txEA(tid_t tid, struct inode *ip, dxd_t * oldea, dxd_t * newea) 2110{ 2111 struct tlock *tlck = NULL; 2112 struct pxd_lock *maplock = NULL, *pxdlock = NULL; 2113 2114 /* 2115 * format maplock for alloc of new EA extent 2116 */ 2117 if (newea) { 2118 /* Since the newea could be a completely zeroed entry we need to 2119 * check for the two flags which indicate we should actually 2120 * commit new EA data 2121 */ 2122 if (newea->flag & DXD_EXTENT) { 2123 tlck = txMaplock(tid, ip, tlckMAP); 2124 maplock = (struct pxd_lock *) & tlck->lock; 2125 pxdlock = (struct pxd_lock *) maplock; 2126 pxdlock->flag = mlckALLOCPXD; 2127 PXDaddress(&pxdlock->pxd, addressDXD(newea)); 2128 PXDlength(&pxdlock->pxd, lengthDXD(newea)); 2129 pxdlock++; 2130 maplock->index = 1; 2131 } else if (newea->flag & DXD_INLINE) { 2132 tlck = NULL; 2133 2134 set_cflag(COMMIT_Inlineea, ip); 2135 } 2136 } 2137 2138 /* 2139 * format maplock for free of old EA extent 2140 */ 2141 if (!test_cflag(COMMIT_Nolink, ip) && oldea->flag & DXD_EXTENT) { 2142 if (tlck == NULL) { 2143 tlck = txMaplock(tid, ip, tlckMAP); 2144 maplock = (struct pxd_lock *) & tlck->lock; 2145 pxdlock = (struct pxd_lock *) maplock; 2146 maplock->index = 0; 2147 } 2148 pxdlock->flag = mlckFREEPXD; 2149 PXDaddress(&pxdlock->pxd, addressDXD(oldea)); 2150 PXDlength(&pxdlock->pxd, lengthDXD(oldea)); 2151 maplock->index++; 2152 } 2153} 2154 2155 2156/* 2157 * txForce() 2158 * 2159 * function: synchronously write pages locked by transaction 2160 * after txLog() but before txUpdateMap(); 2161 */ 2162void txForce(struct tblock * tblk) 2163{ 2164 struct tlock *tlck; 2165 lid_t lid, next; 2166 struct metapage *mp; 2167 2168 /* 2169 * reverse the order of transaction tlocks in 2170 * careful update order of address index pages 2171 * (right to left, bottom up) 2172 */ 2173 tlck = lid_to_tlock(tblk->next); 2174 lid = tlck->next; 2175 tlck->next = 0; 2176 while (lid) { 2177 tlck = lid_to_tlock(lid); 2178 next = tlck->next; 2179 tlck->next = tblk->next; 2180 tblk->next = lid; 2181 lid = next; 2182 } 2183 2184 /* 2185 * synchronously write the page, and 2186 * hold the page for txUpdateMap(); 2187 */ 2188 for (lid = tblk->next; lid; lid = next) { 2189 tlck = lid_to_tlock(lid); 2190 next = tlck->next; 2191 2192 if ((mp = tlck->mp) != NULL && 2193 (tlck->type & tlckBTROOT) == 0) { 2194 assert(mp->xflag & COMMIT_PAGE); 2195 2196 if (tlck->flag & tlckWRITEPAGE) { 2197 tlck->flag &= ~tlckWRITEPAGE; 2198 2199 /* do not release page to freelist */ 2200 assert(atomic_read(&mp->nohomeok)); 2201 hold_metapage(mp, 0); 2202 write_metapage(mp); 2203 } 2204 } 2205 } 2206} 2207 2208 2209/* 2210 * txUpdateMap() 2211 * 2212 * function: update persistent allocation map (and working map 2213 * if appropriate); 2214 * 2215 * parameter: 2216 */ 2217static void txUpdateMap(struct tblock * tblk) 2218{ 2219 struct inode *ip; 2220 struct inode *ipimap; 2221 lid_t lid; 2222 struct tlock *tlck; 2223 struct maplock *maplock; 2224 struct pxd_lock pxdlock; 2225 int maptype; 2226 int k, nlock; 2227 struct metapage *mp = 0; 2228 2229 ipimap = JFS_SBI(tblk->sb)->ipimap; 2230 2231 maptype = (tblk->xflag & COMMIT_PMAP) ? COMMIT_PMAP : COMMIT_PWMAP; 2232 2233 2234 /* 2235 * update block allocation map 2236 * 2237 * update allocation state in pmap (and wmap) and 2238 * update lsn of the pmap page; 2239 */ 2240 /* 2241 * scan each tlock/page of transaction for block allocation/free: 2242 * 2243 * for each tlock/page of transaction, update map. 2244 * ? are there tlock for pmap and pwmap at the same time ? 2245 */ 2246 for (lid = tblk->next; lid; lid = tlck->next) { 2247 tlck = lid_to_tlock(lid); 2248 2249 if ((tlck->flag & tlckUPDATEMAP) == 0) 2250 continue; 2251 2252 if (tlck->flag & tlckFREEPAGE) { 2253 /* 2254 * Another thread may attempt to reuse freed space 2255 * immediately, so we want to get rid of the metapage 2256 * before anyone else has a chance to get it. 2257 * Lock metapage, update maps, then invalidate 2258 * the metapage. 2259 */ 2260 mp = tlck->mp; 2261 ASSERT(mp->xflag & COMMIT_PAGE); 2262 hold_metapage(mp, 0); 2263 } 2264 2265 /* 2266 * extent list: 2267 * . in-line PXD list: 2268 * . out-of-line XAD list: 2269 */ 2270 maplock = (struct maplock *) & tlck->lock; 2271 nlock = maplock->index; 2272 2273 for (k = 0; k < nlock; k++, maplock++) { 2274 /* 2275 * allocate blocks in persistent map: 2276 * 2277 * blocks have been allocated from wmap at alloc time; 2278 */ 2279 if (maplock->flag & mlckALLOC) { 2280 txAllocPMap(ipimap, maplock, tblk); 2281 } 2282 /* 2283 * free blocks in persistent and working map: 2284 * blocks will be freed in pmap and then in wmap; 2285 * 2286 * ? tblock specifies the PMAP/PWMAP based upon 2287 * transaction 2288 * 2289 * free blocks in persistent map: 2290 * blocks will be freed from wmap at last reference 2291 * release of the object for regular files; 2292 * 2293 * Alway free blocks from both persistent & working 2294 * maps for directories 2295 */ 2296 else { /* (maplock->flag & mlckFREE) */ 2297 2298 if (S_ISDIR(tlck->ip->i_mode)) 2299 txFreeMap(ipimap, maplock, 2300 tblk, COMMIT_PWMAP); 2301 else 2302 txFreeMap(ipimap, maplock, 2303 tblk, maptype); 2304 } 2305 } 2306 if (tlck->flag & tlckFREEPAGE) { 2307 if (!(tblk->flag & tblkGC_LAZY)) { 2308 /* This is equivalent to txRelease */ 2309 ASSERT(mp->lid == lid); 2310 tlck->mp->lid = 0; 2311 } 2312 assert(atomic_read(&mp->nohomeok) == 1); 2313 atomic_dec(&mp->nohomeok); 2314 discard_metapage(mp); 2315 tlck->mp = 0; 2316 } 2317 } 2318 /* 2319 * update inode allocation map 2320 * 2321 * update allocation state in pmap and 2322 * update lsn of the pmap page; 2323 * update in-memory inode flag/state 2324 * 2325 * unlock mapper/write lock 2326 */ 2327 if (tblk->xflag & COMMIT_CREATE) { 2328 ip = tblk->ip; 2329 2330 ASSERT(test_cflag(COMMIT_New, ip)); 2331 clear_cflag(COMMIT_New, ip); 2332 2333 diUpdatePMap(ipimap, ip->i_ino, FALSE, tblk); 2334 ipimap->i_state |= I_DIRTY; 2335 /* update persistent block allocation map 2336 * for the allocation of inode extent; 2337 */ 2338 pxdlock.flag = mlckALLOCPXD; 2339 pxdlock.pxd = JFS_IP(ip)->ixpxd; 2340 pxdlock.index = 1; 2341 txAllocPMap(ip, (struct maplock *) & pxdlock, tblk); 2342 iput(ip); 2343 } else if (tblk->xflag & COMMIT_DELETE) { 2344 ip = tblk->ip; 2345 diUpdatePMap(ipimap, ip->i_ino, TRUE, tblk); 2346 ipimap->i_state |= I_DIRTY; 2347 if (test_and_clear_cflag(COMMIT_Holdlock, ip)) { 2348 if (tblk->flag & tblkGC_LAZY) 2349 IWRITE_UNLOCK(ip); 2350 } 2351 iput(ip); 2352 } 2353} 2354 2355 2356/* 2357 * txAllocPMap() 2358 * 2359 * function: allocate from persistent map; 2360 * 2361 * parameter: 2362 * ipbmap - 2363 * malock - 2364 * xad list: 2365 * pxd: 2366 * 2367 * maptype - 2368 * allocate from persistent map; 2369 * free from persistent map; 2370 * (e.g., tmp file - free from working map at releae 2371 * of last reference); 2372 * free from persistent and working map; 2373 * 2374 * lsn - log sequence number; 2375 */ 2376static void txAllocPMap(struct inode *ip, struct maplock * maplock, 2377 struct tblock * tblk) 2378{ 2379 struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap; 2380 struct xdlistlock *xadlistlock; 2381 xad_t *xad; 2382 s64 xaddr; 2383 int xlen; 2384 struct pxd_lock *pxdlock; 2385 struct xdlistlock *pxdlistlock; 2386 pxd_t *pxd; 2387 int n; 2388 2389 /* 2390 * allocate from persistent map; 2391 */ 2392 if (maplock->flag & mlckALLOCXADLIST) { 2393 xadlistlock = (struct xdlistlock *) maplock; 2394 xad = xadlistlock->xdlist; 2395 for (n = 0; n < xadlistlock->count; n++, xad++) { 2396 if (xad->flag & (XAD_NEW | XAD_EXTENDED)) { 2397 xaddr = addressXAD(xad); 2398 xlen = lengthXAD(xad); 2399 dbUpdatePMap(ipbmap, FALSE, xaddr, 2400 (s64) xlen, tblk); 2401 xad->flag &= ~(XAD_NEW | XAD_EXTENDED); 2402 jFYI(1, 2403 ("allocPMap: xaddr:0x%lx xlen:%d\n", 2404 (ulong) xaddr, xlen)); 2405 } 2406 } 2407 } else if (maplock->flag & mlckALLOCPXD) { 2408 pxdlock = (struct pxd_lock *) maplock; 2409 xaddr = addressPXD(&pxdlock->pxd); 2410 xlen = lengthPXD(&pxdlock->pxd); 2411 dbUpdatePMap(ipbmap, FALSE, xaddr, (s64) xlen, tblk); 2412 jFYI(1, 2413 ("allocPMap: xaddr:0x%lx xlen:%d\n", (ulong) xaddr, 2414 xlen)); 2415 } else { /* (maplock->flag & mlckALLOCPXDLIST) */ 2416 2417 pxdlistlock = (struct xdlistlock *) maplock; 2418 pxd = pxdlistlock->xdlist; 2419 for (n = 0; n < pxdlistlock->count; n++, pxd++) { 2420 xaddr = addressPXD(pxd); 2421 xlen = lengthPXD(pxd); 2422 dbUpdatePMap(ipbmap, FALSE, xaddr, (s64) xlen, 2423 tblk); 2424 jFYI(1, 2425 ("allocPMap: xaddr:0x%lx xlen:%d\n", 2426 (ulong) xaddr, xlen)); 2427 } 2428 } 2429} 2430 2431 2432/* 2433 * txFreeMap() 2434 * 2435 * function: free from persistent and/or working map; 2436 * 2437 * todo: optimization 2438 */ 2439void txFreeMap(struct inode *ip, 2440 struct maplock * maplock, struct tblock * tblk, int maptype) 2441{ 2442 struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap; 2443 struct xdlistlock *xadlistlock; 2444 xad_t *xad; 2445 s64 xaddr; 2446 int xlen; 2447 struct pxd_lock *pxdlock; 2448 struct xdlistlock *pxdlistlock; 2449 pxd_t *pxd; 2450 int n; 2451 2452 jFYI(1, 2453 ("txFreeMap: tblk:0x%p maplock:0x%p maptype:0x%x\n", 2454 tblk, maplock, maptype)); 2455 2456 /* 2457 * free from persistent map; 2458 */ 2459 if (maptype == COMMIT_PMAP || maptype == COMMIT_PWMAP) { 2460 if (maplock->flag & mlckFREEXADLIST) { 2461 xadlistlock = (struct xdlistlock *) maplock; 2462 xad = xadlistlock->xdlist; 2463 for (n = 0; n < xadlistlock->count; n++, xad++) { 2464 if (!(xad->flag & XAD_NEW)) { 2465 xaddr = addressXAD(xad); 2466 xlen = lengthXAD(xad); 2467 dbUpdatePMap(ipbmap, TRUE, xaddr, 2468 (s64) xlen, tblk); 2469 jFYI(1, 2470 ("freePMap: xaddr:0x%lx xlen:%d\n", 2471 (ulong) xaddr, xlen)); 2472 } 2473 } 2474 } else if (maplock->flag & mlckFREEPXD) { 2475 pxdlock = (struct pxd_lock *) maplock; 2476 xaddr = addressPXD(&pxdlock->pxd); 2477 xlen = lengthPXD(&pxdlock->pxd); 2478 dbUpdatePMap(ipbmap, TRUE, xaddr, (s64) xlen, 2479 tblk); 2480 jFYI(1, 2481 ("freePMap: xaddr:0x%lx xlen:%d\n", 2482 (ulong) xaddr, xlen)); 2483 } else { /* (maplock->flag & mlckALLOCPXDLIST) */ 2484 2485 pxdlistlock = (struct xdlistlock *) maplock; 2486 pxd = pxdlistlock->xdlist; 2487 for (n = 0; n < pxdlistlock->count; n++, pxd++) { 2488 xaddr = addressPXD(pxd); 2489 xlen = lengthPXD(pxd); 2490 dbUpdatePMap(ipbmap, TRUE, xaddr, 2491 (s64) xlen, tblk); 2492 jFYI(1, 2493 ("freePMap: xaddr:0x%lx xlen:%d\n", 2494 (ulong) xaddr, xlen)); 2495 } 2496 } 2497 } 2498 2499 /* 2500 * free from working map; 2501 */ 2502 if (maptype == COMMIT_PWMAP || maptype == COMMIT_WMAP) { 2503 if (maplock->flag & mlckFREEXADLIST) { 2504 xadlistlock = (struct xdlistlock *) maplock; 2505 xad = xadlistlock->xdlist; 2506 for (n = 0; n < xadlistlock->count; n++, xad++) { 2507 xaddr = addressXAD(xad); 2508 xlen = lengthXAD(xad); 2509 dbFree(ip, xaddr, (s64) xlen); 2510 xad->flag = 0; 2511 jFYI(1, 2512 ("freeWMap: xaddr:0x%lx xlen:%d\n", 2513 (ulong) xaddr, xlen)); 2514 } 2515 } else if (maplock->flag & mlckFREEPXD) { 2516 pxdlock = (struct pxd_lock *) maplock; 2517 xaddr = addressPXD(&pxdlock->pxd); 2518 xlen = lengthPXD(&pxdlock->pxd); 2519 dbFree(ip, xaddr, (s64) xlen); 2520 jFYI(1, 2521 ("freeWMap: xaddr:0x%lx xlen:%d\n", 2522 (ulong) xaddr, xlen)); 2523 } else { /* (maplock->flag & mlckFREEPXDLIST) */ 2524 2525 pxdlistlock = (struct xdlistlock *) maplock; 2526 pxd = pxdlistlock->xdlist; 2527 for (n = 0; n < pxdlistlock->count; n++, pxd++) { 2528 xaddr = addressPXD(pxd); 2529 xlen = lengthPXD(pxd); 2530 dbFree(ip, xaddr, (s64) xlen); 2531 jFYI(1, 2532 ("freeWMap: xaddr:0x%lx xlen:%d\n", 2533 (ulong) xaddr, xlen)); 2534 } 2535 } 2536 } 2537} 2538 2539 2540/* 2541 * txFreelock() 2542 * 2543 * function: remove tlock from inode anonymous locklist 2544 */ 2545void txFreelock(struct inode *ip) 2546{ 2547 struct jfs_inode_info *jfs_ip = JFS_IP(ip); 2548 struct tlock *xtlck, *tlck; 2549 lid_t xlid = 0, lid; 2550 2551 if (!jfs_ip->atlhead) 2552 return; 2553 2554 xtlck = (struct tlock *) &jfs_ip->atlhead; 2555 2556 while ((lid = xtlck->next)) { 2557 tlck = lid_to_tlock(lid); 2558 if (tlck->flag & tlckFREELOCK) { 2559 xtlck->next = tlck->next; 2560 txLockFree(lid); 2561 } else { 2562 xtlck = tlck; 2563 xlid = lid; 2564 } 2565 } 2566 2567 if (jfs_ip->atlhead) 2568 jfs_ip->atltail = xlid; 2569 else { 2570 jfs_ip->atltail = 0; 2571 /* 2572 * If inode was on anon_list, remove it 2573 */ 2574 TXN_LOCK(); 2575 list_del_init(&jfs_ip->anon_inode_list); 2576 TXN_UNLOCK(); 2577 } 2578} 2579 2580 2581/* 2582 * txAbort() 2583 * 2584 * function: abort tx before commit; 2585 * 2586 * frees line-locks and segment locks for all 2587 * segments in comdata structure. 2588 * Optionally sets state of file-system to FM_DIRTY in super-block. 2589 * log age of page-frames in memory for which caller has 2590 * are reset to 0 (to avoid logwarap). 2591 */ 2592void txAbort(tid_t tid, int dirty) 2593{ 2594 lid_t lid, next; 2595 struct metapage *mp; 2596 struct tblock *tblk = tid_to_tblock(tid); 2597 2598 jEVENT(1, ("txAbort: tid:%d dirty:0x%x\n", tid, dirty)); 2599 2600 /* 2601 * free tlocks of the transaction 2602 */ 2603 for (lid = tblk->next; lid; lid = next) { 2604 next = lid_to_tlock(lid)->next; 2605 2606 mp = lid_to_tlock(lid)->mp; 2607 2608 if (mp) { 2609 mp->lid = 0; 2610 2611 /* 2612 * reset lsn of page to avoid logwarap: 2613 * 2614 * (page may have been previously committed by another 2615 * transaction(s) but has not been paged, i.e., 2616 * it may be on logsync list even though it has not 2617 * been logged for the current tx.) 2618 */ 2619 if (mp->xflag & COMMIT_PAGE && mp->lsn) 2620 LogSyncRelease(mp); 2621 } 2622 /* insert tlock at head of freelist */ 2623 TXN_LOCK(); 2624 txLockFree(lid); 2625 TXN_UNLOCK(); 2626 } 2627 2628 /* caller will free the transaction block */ 2629 2630 tblk->next = tblk->last = 0; 2631 2632 /* 2633 * mark filesystem dirty 2634 */ 2635 if (dirty) 2636 updateSuper(tblk->sb, FM_DIRTY); 2637 2638 return; 2639} 2640 2641 2642/* 2643 * txAbortCommit() 2644 * 2645 * function: abort commit. 2646 * 2647 * frees tlocks of transaction; line-locks and segment locks for all 2648 * segments in comdata structure. frees malloc storage 2649 * sets state of file-system to FM_MDIRTY in super-block. 2650 * log age of page-frames in memory for which caller has 2651 * are reset to 0 (to avoid logwarap). 2652 */ 2653void txAbortCommit(struct commit * cd, int exval) 2654{ 2655 struct tblock *tblk; 2656 tid_t tid; 2657 lid_t lid, next; 2658 struct metapage *mp; 2659 2660 assert(exval == EIO || exval == ENOMEM); 2661 jEVENT(1, ("txAbortCommit: cd:0x%p\n", cd)); 2662 2663 /* 2664 * free tlocks of the transaction 2665 */ 2666 tid = cd->tid; 2667 tblk = tid_to_tblock(tid); 2668 for (lid = tblk->next; lid; lid = next) { 2669 next = lid_to_tlock(lid)->next; 2670 2671 mp = lid_to_tlock(lid)->mp; 2672 if (mp) { 2673 mp->lid = 0; 2674 2675 /* 2676 * reset lsn of page to avoid logwarap; 2677 */ 2678 if (mp->xflag & COMMIT_PAGE) 2679 LogSyncRelease(mp); 2680 } 2681 2682 /* insert tlock at head of freelist */ 2683 TXN_LOCK(); 2684 txLockFree(lid); 2685 TXN_UNLOCK(); 2686 } 2687 2688 tblk->next = tblk->last = 0; 2689 2690 /* free the transaction block */ 2691 txEnd(tid); 2692 2693 /* 2694 * mark filesystem dirty 2695 */ 2696 updateSuper(cd->sb, FM_DIRTY); 2697} 2698 2699 2700/* 2701 * txLazyCommit(void) 2702 * 2703 * All transactions except those changing ipimap (COMMIT_FORCE) are 2704 * processed by this routine. This insures that the inode and block 2705 * allocation maps are updated in order. For synchronous transactions, 2706 * let the user thread finish processing after txUpdateMap() is called. 2707 */ 2708void txLazyCommit(struct tblock * tblk) 2709{ 2710 struct jfs_log *log; 2711 2712 while (((tblk->flag & tblkGC_READY) == 0) && 2713 ((tblk->flag & tblkGC_UNLOCKED) == 0)) { 2714 /* We must have gotten ahead of the user thread 2715 */ 2716 jFYI(1, ("txLazyCommit: tblk 0x%p not unlocked\n", tblk)); 2717 schedule(); 2718 } 2719 2720 jFYI(1, ("txLazyCommit: processing tblk 0x%p\n", tblk)); 2721 2722 txUpdateMap(tblk); 2723 2724 log = (struct jfs_log *) JFS_SBI(tblk->sb)->log; 2725 2726 spin_lock_irq(&log->gclock); // LOGGC_LOCK 2727 2728 tblk->flag |= tblkGC_COMMITTED; 2729 2730 if ((tblk->flag & tblkGC_READY) || (tblk->flag & tblkGC_LAZY)) 2731 log->gcrtc--; 2732 2733 if (tblk->flag & tblkGC_READY) 2734 wake_up(&tblk->gcwait); // LOGGC_WAKEUP 2735 2736 /* 2737 * Can't release log->gclock until we've tested tblk->flag 2738 */ 2739 if (tblk->flag & tblkGC_LAZY) { 2740 spin_unlock_irq(&log->gclock); // LOGGC_UNLOCK 2741 txUnlock(tblk); 2742 tblk->flag &= ~tblkGC_LAZY; 2743 txEnd(tblk - TxBlock); /* Convert back to tid */ 2744 } else 2745 spin_unlock_irq(&log->gclock); // LOGGC_UNLOCK 2746 2747 jFYI(1, ("txLazyCommit: done: tblk = 0x%p\n", tblk)); 2748} 2749 2750/* 2751 * jfs_lazycommit(void) 2752 * 2753 * To be run as a kernel daemon. If lbmIODone is called in an interrupt 2754 * context, or where blocking is not wanted, this routine will process 2755 * committed transactions from the unlock queue. 2756 */ 2757int jfs_lazycommit(void *arg) 2758{ 2759 int WorkDone; 2760 struct tblock *tblk; 2761 unsigned long flags; 2762 2763 lock_kernel(); 2764 2765 daemonize(); 2766 current->tty = NULL; 2767 strcpy(current->comm, "jfsCommit"); 2768 2769 unlock_kernel(); 2770 2771 jfsCommitTask = current; 2772 2773 spin_lock_irq(¤t->sigmask_lock); 2774 sigfillset(¤t->blocked); 2775 recalc_sigpending(current); 2776 spin_unlock_irq(¤t->sigmask_lock); 2777 2778 LAZY_LOCK_INIT(); 2779 TxAnchor.unlock_queue = TxAnchor.unlock_tail = 0; 2780 2781 complete(&jfsIOwait); 2782 2783 do { 2784 DECLARE_WAITQUEUE(wq, current); 2785 2786 LAZY_LOCK(flags); 2787restart: 2788 WorkDone = 0; 2789 while ((tblk = TxAnchor.unlock_queue)) { 2790 /* 2791 * We can't get ahead of user thread. Spinning is 2792 * simpler than blocking/waking. We shouldn't spin 2793 * very long, since user thread shouldn't be blocking 2794 * between lmGroupCommit & txEnd. 2795 */ 2796 WorkDone = 1; 2797 2798 /* 2799 * Remove first transaction from queue 2800 */ 2801 TxAnchor.unlock_queue = tblk->cqnext; 2802 tblk->cqnext = 0; 2803 if (TxAnchor.unlock_tail == tblk) 2804 TxAnchor.unlock_tail = 0; 2805 2806 LAZY_UNLOCK(flags); 2807 txLazyCommit(tblk); 2808 2809 /* 2810 * We can be running indefinately if other processors 2811 * are adding transactions to this list 2812 */ 2813 cond_resched(); 2814 LAZY_LOCK(flags); 2815 } 2816 2817 if (WorkDone) 2818 goto restart; 2819 2820 add_wait_queue(&jfs_commit_thread_wait, &wq); 2821 set_current_state(TASK_INTERRUPTIBLE); 2822 LAZY_UNLOCK(flags); 2823 schedule(); 2824 current->state = TASK_RUNNING; 2825 remove_wait_queue(&jfs_commit_thread_wait, &wq); 2826 } while (!jfs_stop_threads); 2827 2828 if (TxAnchor.unlock_queue) 2829 jERROR(1, ("jfs_lazycommit being killed with pending transactions!\n")); 2830 else 2831 jFYI(1, ("jfs_lazycommit being killed\n")); 2832 complete(&jfsIOwait); 2833 return 0; 2834} 2835 2836void txLazyUnlock(struct tblock * tblk) 2837{ 2838 unsigned long flags; 2839 2840 LAZY_LOCK(flags); 2841 2842 if (TxAnchor.unlock_tail) 2843 TxAnchor.unlock_tail->cqnext = tblk; 2844 else 2845 TxAnchor.unlock_queue = tblk; 2846 TxAnchor.unlock_tail = tblk; 2847 tblk->cqnext = 0; 2848 LAZY_UNLOCK(flags); 2849 wake_up(&jfs_commit_thread_wait); 2850} 2851 2852static void LogSyncRelease(struct metapage * mp) 2853{ 2854 struct jfs_log *log = mp->log; 2855 2856 assert(atomic_read(&mp->nohomeok)); 2857 assert(log); 2858 atomic_dec(&mp->nohomeok); 2859 2860 if (atomic_read(&mp->nohomeok)) 2861 return; 2862 2863 hold_metapage(mp, 0); 2864 2865 LOGSYNC_LOCK(log); 2866 mp->log = NULL; 2867 mp->lsn = 0; 2868 mp->clsn = 0; 2869 log->count--; 2870 list_del_init(&mp->synclist); 2871 LOGSYNC_UNLOCK(log); 2872 2873 release_metapage(mp); 2874} 2875 2876/* 2877 * txQuiesce 2878 * 2879 * Block all new transactions and push anonymous transactions to 2880 * completion 2881 * 2882 * This does almost the same thing as jfs_sync below. We don't 2883 * worry about deadlocking when TlocksLow is set, since we would 2884 * expect jfs_sync to get us out of that jam. 2885 */ 2886void txQuiesce(struct super_block *sb) 2887{ 2888 struct inode *ip; 2889 struct jfs_inode_info *jfs_ip; 2890 struct jfs_log *log = JFS_SBI(sb)->log; 2891 int rc; 2892 tid_t tid; 2893 2894 set_bit(log_QUIESCE, &log->flag); 2895 2896 TXN_LOCK(); 2897restart: 2898 while (!list_empty(&TxAnchor.anon_list)) { 2899 jfs_ip = list_entry(TxAnchor.anon_list.next, 2900 struct jfs_inode_info, 2901 anon_inode_list); 2902 ip = jfs_ip->inode; 2903 2904 /* 2905 * inode will be removed from anonymous list 2906 * when it is committed 2907 */ 2908 TXN_UNLOCK(); 2909 tid = txBegin(ip->i_sb, COMMIT_INODE | COMMIT_FORCE); 2910 down(&jfs_ip->commit_sem); 2911 rc = txCommit(tid, 1, &ip, 0); 2912 txEnd(tid); 2913 up(&jfs_ip->commit_sem); 2914 /* 2915 * Just to be safe. I don't know how 2916 * long we can run without blocking 2917 */ 2918 cond_resched(); 2919 TXN_LOCK(); 2920 } 2921 2922 /* 2923 * If jfs_sync is running in parallel, there could be some inodes 2924 * on anon_list2. Let's check. 2925 */ 2926 if (!list_empty(&TxAnchor.anon_list2)) { 2927 list_splice(&TxAnchor.anon_list2, &TxAnchor.anon_list); 2928 INIT_LIST_HEAD(&TxAnchor.anon_list2); 2929 goto restart; 2930 } 2931 TXN_UNLOCK(); 2932} 2933 2934/* 2935 * txResume() 2936 * 2937 * Allows transactions to start again following txQuiesce 2938 */ 2939void txResume(struct super_block *sb) 2940{ 2941 struct jfs_log *log = JFS_SBI(sb)->log; 2942 2943 clear_bit(log_QUIESCE, &log->flag); 2944 TXN_WAKEUP(&log->syncwait); 2945} 2946 2947/* 2948 * jfs_sync(void) 2949 * 2950 * To be run as a kernel daemon. This is awakened when tlocks run low. 2951 * We write any inodes that have anonymous tlocks so they will become 2952 * available. 2953 */ 2954int jfs_sync(void *arg) 2955{ 2956 struct inode *ip; 2957 struct jfs_inode_info *jfs_ip; 2958 int rc; 2959 tid_t tid; 2960 2961 lock_kernel(); 2962 2963 daemonize(); 2964 current->tty = NULL; 2965 strcpy(current->comm, "jfsSync"); 2966 2967 unlock_kernel(); 2968 2969 spin_lock_irq(¤t->sigmask_lock); 2970 sigfillset(¤t->blocked); 2971 recalc_sigpending(current); 2972 spin_unlock_irq(¤t->sigmask_lock); 2973 2974 complete(&jfsIOwait); 2975 2976 do { 2977 DECLARE_WAITQUEUE(wq, current); 2978 /* 2979 * write each inode on the anonymous inode list 2980 */ 2981 TXN_LOCK(); 2982 while (TlocksLow && !list_empty(&TxAnchor.anon_list)) { 2983 jfs_ip = list_entry(TxAnchor.anon_list.next, 2984 struct jfs_inode_info, 2985 anon_inode_list); 2986 ip = jfs_ip->inode; 2987 2988 /* 2989 * down_trylock returns 0 on success. This is 2990 * inconsistent with spin_trylock. 2991 */ 2992 if (! down_trylock(&jfs_ip->commit_sem)) { 2993 /* 2994 * inode will be removed from anonymous list 2995 * when it is committed 2996 */ 2997 TXN_UNLOCK(); 2998 tid = txBegin(ip->i_sb, 2999 COMMIT_INODE | COMMIT_FORCE); 3000 rc = txCommit(tid, 1, &ip, 0); 3001 txEnd(tid); 3002 up(&jfs_ip->commit_sem); 3003 /* 3004 * Just to be safe. I don't know how 3005 * long we can run without blocking 3006 */ 3007 cond_resched(); 3008 TXN_LOCK(); 3009 } else { 3010 /* We can't get the commit semaphore. It may 3011 * be held by a thread waiting for tlock's 3012 * so let's not block here. Save it to 3013 * put back on the anon_list. 3014 */ 3015 3016 /* Take off anon_list */ 3017 list_del(&jfs_ip->anon_inode_list); 3018 3019 /* Put on anon_list2 */ 3020 list_add(&jfs_ip->anon_inode_list, 3021 &TxAnchor.anon_list2); 3022 } 3023 } 3024 /* Add anon_list2 back to anon_list */ 3025 if (!list_empty(&TxAnchor.anon_list2)) { 3026 list_splice(&TxAnchor.anon_list2, &TxAnchor.anon_list); 3027 INIT_LIST_HEAD(&TxAnchor.anon_list2); 3028 } 3029 add_wait_queue(&jfs_sync_thread_wait, &wq); 3030 set_current_state(TASK_INTERRUPTIBLE); 3031 TXN_UNLOCK(); 3032 schedule(); 3033 current->state = TASK_RUNNING; 3034 remove_wait_queue(&jfs_sync_thread_wait, &wq); 3035 } while (!jfs_stop_threads); 3036 3037 jFYI(1, ("jfs_sync being killed\n")); 3038 complete(&jfsIOwait); 3039 return 0; 3040} 3041 3042#if defined(CONFIG_PROC_FS) && defined(CONFIG_JFS_DEBUG) 3043int jfs_txanchor_read(char *buffer, char **start, off_t offset, int length, 3044 int *eof, void *data) 3045{ 3046 int len = 0; 3047 off_t begin; 3048 char *freewait; 3049 char *freelockwait; 3050 char *lowlockwait; 3051 3052 freewait = 3053 waitqueue_active(&TxAnchor.freewait) ? "active" : "empty"; 3054 freelockwait = 3055 waitqueue_active(&TxAnchor.freelockwait) ? "active" : "empty"; 3056 lowlockwait = 3057 waitqueue_active(&TxAnchor.lowlockwait) ? "active" : "empty"; 3058 3059 len += sprintf(buffer, 3060 "JFS TxAnchor\n" 3061 "============\n" 3062 "freetid = %d\n" 3063 "freewait = %s\n" 3064 "freelock = %d\n" 3065 "freelockwait = %s\n" 3066 "lowlockwait = %s\n" 3067 "tlocksInUse = %d\n" 3068 "unlock_queue = 0x%p\n" 3069 "unlock_tail = 0x%p\n", 3070 TxAnchor.freetid, 3071 freewait, 3072 TxAnchor.freelock, 3073 freelockwait, 3074 lowlockwait, 3075 TxAnchor.tlocksInUse, 3076 TxAnchor.unlock_queue, 3077 TxAnchor.unlock_tail); 3078 3079 begin = offset; 3080 *start = buffer + begin; 3081 len -= begin; 3082 3083 if (len > length) 3084 len = length; 3085 else 3086 *eof = 1; 3087 3088 if (len < 0) 3089 len = 0; 3090 3091 return len; 3092} 3093#endif 3094