1 /* 2 * Copyright (C) International Business Machines Corp., 2000-2004 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_logmgr.c: log manager 22 * 23 * for related information, see transaction manager (jfs_txnmgr.c), and 24 * recovery manager (jfs_logredo.c). 25 * 26 * note: for detail, RTFS. 27 * 28 * log buffer manager: 29 * special purpose buffer manager supporting log i/o requirements. 30 * per log serial pageout of logpage 31 * queuing i/o requests and redrive i/o at iodone 32 * maintain current logpage buffer 33 * no caching since append only 34 * appropriate jfs buffer cache buffers as needed 35 * 36 * group commit: 37 * transactions which wrote COMMIT records in the same in-memory 38 * log page during the pageout of previous/current log page(s) are 39 * committed together by the pageout of the page. 40 * 41 * TBD lazy commit: 42 * transactions are committed asynchronously when the log page 43 * containing it COMMIT is paged out when it becomes full; 44 * 45 * serialization: 46 * . a per log lock serialize log write. 47 * . a per log lock serialize group commit. 48 * . a per log lock serialize log open/close; 49 * 50 * TBD log integrity: 51 * careful-write (ping-pong) of last logpage to recover from crash 52 * in overwrite. 53 * detection of split (out-of-order) write of physical sectors 54 * of last logpage via timestamp at end of each sector 55 * with its mirror data array at trailer). 56 * 57 * alternatives: 58 * lsn - 64-bit monotonically increasing integer vs 59 * 32-bit lspn and page eor. 60 */ 61 62 #include <linux/fs.h> 63 #include <linux/blkdev.h> 64 #include <linux/interrupt.h> 65 #include <linux/completion.h> 66 #include <linux/kthread.h> 67 #include <linux/buffer_head.h> /* for sync_blockdev() */ 68 #include <linux/bio.h> 69 #include <linux/freezer.h> 70 #include <linux/delay.h> 71 #include <linux/mutex.h> 72 #include <linux/seq_file.h> 73 #include <linux/slab.h> 74 #include "jfs_incore.h" 75 #include "jfs_filsys.h" 76 #include "jfs_metapage.h" 77 #include "jfs_superblock.h" 78 #include "jfs_txnmgr.h" 79 #include "jfs_debug.h" 80 81 82 /* 83 * lbuf's ready to be redriven. Protected by log_redrive_lock (jfsIO thread) 84 */ 85 static struct lbuf *log_redrive_list; 86 static DEFINE_SPINLOCK(log_redrive_lock); 87 88 89 /* 90 * log read/write serialization (per log) 91 */ 92 #define LOG_LOCK_INIT(log) mutex_init(&(log)->loglock) 93 #define LOG_LOCK(log) mutex_lock(&((log)->loglock)) 94 #define LOG_UNLOCK(log) mutex_unlock(&((log)->loglock)) 95 96 97 /* 98 * log group commit serialization (per log) 99 */ 100 101 #define LOGGC_LOCK_INIT(log) spin_lock_init(&(log)->gclock) 102 #define LOGGC_LOCK(log) spin_lock_irq(&(log)->gclock) 103 #define LOGGC_UNLOCK(log) spin_unlock_irq(&(log)->gclock) 104 #define LOGGC_WAKEUP(tblk) wake_up_all(&(tblk)->gcwait) 105 106 /* 107 * log sync serialization (per log) 108 */ 109 #define LOGSYNC_DELTA(logsize) min((logsize)/8, 128*LOGPSIZE) 110 #define LOGSYNC_BARRIER(logsize) ((logsize)/4) 111 /* 112 #define LOGSYNC_DELTA(logsize) min((logsize)/4, 256*LOGPSIZE) 113 #define LOGSYNC_BARRIER(logsize) ((logsize)/2) 114 */ 115 116 117 /* 118 * log buffer cache synchronization 119 */ 120 static DEFINE_SPINLOCK(jfsLCacheLock); 121 122 #define LCACHE_LOCK(flags) spin_lock_irqsave(&jfsLCacheLock, flags) 123 #define LCACHE_UNLOCK(flags) spin_unlock_irqrestore(&jfsLCacheLock, flags) 124 125 /* 126 * See __SLEEP_COND in jfs_locks.h 127 */ 128 #define LCACHE_SLEEP_COND(wq, cond, flags) \ 129 do { \ 130 if (cond) \ 131 break; \ 132 __SLEEP_COND(wq, cond, LCACHE_LOCK(flags), LCACHE_UNLOCK(flags)); \ 133 } while (0) 134 135 #define LCACHE_WAKEUP(event) wake_up(event) 136 137 138 /* 139 * lbuf buffer cache (lCache) control 140 */ 141 /* log buffer manager pageout control (cumulative, inclusive) */ 142 #define lbmREAD 0x0001 143 #define lbmWRITE 0x0002 /* enqueue at tail of write queue; 144 * init pageout if at head of queue; 145 */ 146 #define lbmRELEASE 0x0004 /* remove from write queue 147 * at completion of pageout; 148 * do not free/recycle it yet: 149 * caller will free it; 150 */ 151 #define lbmSYNC 0x0008 /* do not return to freelist 152 * when removed from write queue; 153 */ 154 #define lbmFREE 0x0010 /* return to freelist 155 * at completion of pageout; 156 * the buffer may be recycled; 157 */ 158 #define lbmDONE 0x0020 159 #define lbmERROR 0x0040 160 #define lbmGC 0x0080 /* lbmIODone to perform post-GC processing 161 * of log page 162 */ 163 #define lbmDIRECT 0x0100 164 165 /* 166 * Global list of active external journals 167 */ 168 static LIST_HEAD(jfs_external_logs); 169 static struct jfs_log *dummy_log = NULL; 170 static DEFINE_MUTEX(jfs_log_mutex); 171 172 /* 173 * forward references 174 */ 175 static int lmWriteRecord(struct jfs_log * log, struct tblock * tblk, 176 struct lrd * lrd, struct tlock * tlck); 177 178 static int lmNextPage(struct jfs_log * log); 179 static int lmLogFileSystem(struct jfs_log * log, struct jfs_sb_info *sbi, 180 int activate); 181 182 static int open_inline_log(struct super_block *sb); 183 static int open_dummy_log(struct super_block *sb); 184 static int lbmLogInit(struct jfs_log * log); 185 static void lbmLogShutdown(struct jfs_log * log); 186 static struct lbuf *lbmAllocate(struct jfs_log * log, int); 187 static void lbmFree(struct lbuf * bp); 188 static void lbmfree(struct lbuf * bp); 189 static int lbmRead(struct jfs_log * log, int pn, struct lbuf ** bpp); 190 static void lbmWrite(struct jfs_log * log, struct lbuf * bp, int flag, int cant_block); 191 static void lbmDirectWrite(struct jfs_log * log, struct lbuf * bp, int flag); 192 static int lbmIOWait(struct lbuf * bp, int flag); 193 static bio_end_io_t lbmIODone; 194 static void lbmStartIO(struct lbuf * bp); 195 static void lmGCwrite(struct jfs_log * log, int cant_block); 196 static int lmLogSync(struct jfs_log * log, int hard_sync); 197 198 199 200 /* 201 * statistics 202 */ 203 #ifdef CONFIG_JFS_STATISTICS 204 static struct lmStat { 205 uint commit; /* # of commit */ 206 uint pagedone; /* # of page written */ 207 uint submitted; /* # of pages submitted */ 208 uint full_page; /* # of full pages submitted */ 209 uint partial_page; /* # of partial pages submitted */ 210 } lmStat; 211 #endif 212 213 static void write_special_inodes(struct jfs_log *log, 214 int (*writer)(struct address_space *)) 215 { 216 struct jfs_sb_info *sbi; 217 218 list_for_each_entry(sbi, &log->sb_list, log_list) { 219 writer(sbi->ipbmap->i_mapping); 220 writer(sbi->ipimap->i_mapping); 221 writer(sbi->direct_inode->i_mapping); 222 } 223 } 224 225 /* 226 * NAME: lmLog() 227 * 228 * FUNCTION: write a log record; 229 * 230 * PARAMETER: 231 * 232 * RETURN: lsn - offset to the next log record to write (end-of-log); 233 * -1 - error; 234 * 235 * note: todo: log error handler 236 */ 237 int lmLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd, 238 struct tlock * tlck) 239 { 240 int lsn; 241 int diffp, difft; 242 struct metapage *mp = NULL; 243 unsigned long flags; 244 245 jfs_info("lmLog: log:0x%p tblk:0x%p, lrd:0x%p tlck:0x%p", 246 log, tblk, lrd, tlck); 247 248 LOG_LOCK(log); 249 250 /* log by (out-of-transaction) JFS ? */ 251 if (tblk == NULL) 252 goto writeRecord; 253 254 /* log from page ? */ 255 if (tlck == NULL || 256 tlck->type & tlckBTROOT || (mp = tlck->mp) == NULL) 257 goto writeRecord; 258 259 /* 260 * initialize/update page/transaction recovery lsn 261 */ 262 lsn = log->lsn; 263 264 LOGSYNC_LOCK(log, flags); 265 266 /* 267 * initialize page lsn if first log write of the page 268 */ 269 if (mp->lsn == 0) { 270 mp->log = log; 271 mp->lsn = lsn; 272 log->count++; 273 274 /* insert page at tail of logsynclist */ 275 list_add_tail(&mp->synclist, &log->synclist); 276 } 277 278 /* 279 * initialize/update lsn of tblock of the page 280 * 281 * transaction inherits oldest lsn of pages associated 282 * with allocation/deallocation of resources (their 283 * log records are used to reconstruct allocation map 284 * at recovery time: inode for inode allocation map, 285 * B+-tree index of extent descriptors for block 286 * allocation map); 287 * allocation map pages inherit transaction lsn at 288 * commit time to allow forwarding log syncpt past log 289 * records associated with allocation/deallocation of 290 * resources only after persistent map of these map pages 291 * have been updated and propagated to home. 292 */ 293 /* 294 * initialize transaction lsn: 295 */ 296 if (tblk->lsn == 0) { 297 /* inherit lsn of its first page logged */ 298 tblk->lsn = mp->lsn; 299 log->count++; 300 301 /* insert tblock after the page on logsynclist */ 302 list_add(&tblk->synclist, &mp->synclist); 303 } 304 /* 305 * update transaction lsn: 306 */ 307 else { 308 /* inherit oldest/smallest lsn of page */ 309 logdiff(diffp, mp->lsn, log); 310 logdiff(difft, tblk->lsn, log); 311 if (diffp < difft) { 312 /* update tblock lsn with page lsn */ 313 tblk->lsn = mp->lsn; 314 315 /* move tblock after page on logsynclist */ 316 list_move(&tblk->synclist, &mp->synclist); 317 } 318 } 319 320 LOGSYNC_UNLOCK(log, flags); 321 322 /* 323 * write the log record 324 */ 325 writeRecord: 326 lsn = lmWriteRecord(log, tblk, lrd, tlck); 327 328 /* 329 * forward log syncpt if log reached next syncpt trigger 330 */ 331 logdiff(diffp, lsn, log); 332 if (diffp >= log->nextsync) 333 lsn = lmLogSync(log, 0); 334 335 /* update end-of-log lsn */ 336 log->lsn = lsn; 337 338 LOG_UNLOCK(log); 339 340 /* return end-of-log address */ 341 return lsn; 342 } 343 344 /* 345 * NAME: lmWriteRecord() 346 * 347 * FUNCTION: move the log record to current log page 348 * 349 * PARAMETER: cd - commit descriptor 350 * 351 * RETURN: end-of-log address 352 * 353 * serialization: LOG_LOCK() held on entry/exit 354 */ 355 static int 356 lmWriteRecord(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd, 357 struct tlock * tlck) 358 { 359 int lsn = 0; /* end-of-log address */ 360 struct lbuf *bp; /* dst log page buffer */ 361 struct logpage *lp; /* dst log page */ 362 caddr_t dst; /* destination address in log page */ 363 int dstoffset; /* end-of-log offset in log page */ 364 int freespace; /* free space in log page */ 365 caddr_t p; /* src meta-data page */ 366 caddr_t src; 367 int srclen; 368 int nbytes; /* number of bytes to move */ 369 int i; 370 int len; 371 struct linelock *linelock; 372 struct lv *lv; 373 struct lvd *lvd; 374 int l2linesize; 375 376 len = 0; 377 378 /* retrieve destination log page to write */ 379 bp = (struct lbuf *) log->bp; 380 lp = (struct logpage *) bp->l_ldata; 381 dstoffset = log->eor; 382 383 /* any log data to write ? */ 384 if (tlck == NULL) 385 goto moveLrd; 386 387 /* 388 * move log record data 389 */ 390 /* retrieve source meta-data page to log */ 391 if (tlck->flag & tlckPAGELOCK) { 392 p = (caddr_t) (tlck->mp->data); 393 linelock = (struct linelock *) & tlck->lock; 394 } 395 /* retrieve source in-memory inode to log */ 396 else if (tlck->flag & tlckINODELOCK) { 397 if (tlck->type & tlckDTREE) 398 p = (caddr_t) &JFS_IP(tlck->ip)->i_dtroot; 399 else 400 p = (caddr_t) &JFS_IP(tlck->ip)->i_xtroot; 401 linelock = (struct linelock *) & tlck->lock; 402 } 403 #ifdef _JFS_WIP 404 else if (tlck->flag & tlckINLINELOCK) { 405 406 inlinelock = (struct inlinelock *) & tlck; 407 p = (caddr_t) & inlinelock->pxd; 408 linelock = (struct linelock *) & tlck; 409 } 410 #endif /* _JFS_WIP */ 411 else { 412 jfs_err("lmWriteRecord: UFO tlck:0x%p", tlck); 413 return 0; /* Probably should trap */ 414 } 415 l2linesize = linelock->l2linesize; 416 417 moveData: 418 ASSERT(linelock->index <= linelock->maxcnt); 419 420 lv = linelock->lv; 421 for (i = 0; i < linelock->index; i++, lv++) { 422 if (lv->length == 0) 423 continue; 424 425 /* is page full ? */ 426 if (dstoffset >= LOGPSIZE - LOGPTLRSIZE) { 427 /* page become full: move on to next page */ 428 lmNextPage(log); 429 430 bp = log->bp; 431 lp = (struct logpage *) bp->l_ldata; 432 dstoffset = LOGPHDRSIZE; 433 } 434 435 /* 436 * move log vector data 437 */ 438 src = (u8 *) p + (lv->offset << l2linesize); 439 srclen = lv->length << l2linesize; 440 len += srclen; 441 while (srclen > 0) { 442 freespace = (LOGPSIZE - LOGPTLRSIZE) - dstoffset; 443 nbytes = min(freespace, srclen); 444 dst = (caddr_t) lp + dstoffset; 445 memcpy(dst, src, nbytes); 446 dstoffset += nbytes; 447 448 /* is page not full ? */ 449 if (dstoffset < LOGPSIZE - LOGPTLRSIZE) 450 break; 451 452 /* page become full: move on to next page */ 453 lmNextPage(log); 454 455 bp = (struct lbuf *) log->bp; 456 lp = (struct logpage *) bp->l_ldata; 457 dstoffset = LOGPHDRSIZE; 458 459 srclen -= nbytes; 460 src += nbytes; 461 } 462 463 /* 464 * move log vector descriptor 465 */ 466 len += 4; 467 lvd = (struct lvd *) ((caddr_t) lp + dstoffset); 468 lvd->offset = cpu_to_le16(lv->offset); 469 lvd->length = cpu_to_le16(lv->length); 470 dstoffset += 4; 471 jfs_info("lmWriteRecord: lv offset:%d length:%d", 472 lv->offset, lv->length); 473 } 474 475 if ((i = linelock->next)) { 476 linelock = (struct linelock *) lid_to_tlock(i); 477 goto moveData; 478 } 479 480 /* 481 * move log record descriptor 482 */ 483 moveLrd: 484 lrd->length = cpu_to_le16(len); 485 486 src = (caddr_t) lrd; 487 srclen = LOGRDSIZE; 488 489 while (srclen > 0) { 490 freespace = (LOGPSIZE - LOGPTLRSIZE) - dstoffset; 491 nbytes = min(freespace, srclen); 492 dst = (caddr_t) lp + dstoffset; 493 memcpy(dst, src, nbytes); 494 495 dstoffset += nbytes; 496 srclen -= nbytes; 497 498 /* are there more to move than freespace of page ? */ 499 if (srclen) 500 goto pageFull; 501 502 /* 503 * end of log record descriptor 504 */ 505 506 /* update last log record eor */ 507 log->eor = dstoffset; 508 bp->l_eor = dstoffset; 509 lsn = (log->page << L2LOGPSIZE) + dstoffset; 510 511 if (lrd->type & cpu_to_le16(LOG_COMMIT)) { 512 tblk->clsn = lsn; 513 jfs_info("wr: tclsn:0x%x, beor:0x%x", tblk->clsn, 514 bp->l_eor); 515 516 INCREMENT(lmStat.commit); /* # of commit */ 517 518 /* 519 * enqueue tblock for group commit: 520 * 521 * enqueue tblock of non-trivial/synchronous COMMIT 522 * at tail of group commit queue 523 * (trivial/asynchronous COMMITs are ignored by 524 * group commit.) 525 */ 526 LOGGC_LOCK(log); 527 528 /* init tblock gc state */ 529 tblk->flag = tblkGC_QUEUE; 530 tblk->bp = log->bp; 531 tblk->pn = log->page; 532 tblk->eor = log->eor; 533 534 /* enqueue transaction to commit queue */ 535 list_add_tail(&tblk->cqueue, &log->cqueue); 536 537 LOGGC_UNLOCK(log); 538 } 539 540 jfs_info("lmWriteRecord: lrd:0x%04x bp:0x%p pn:%d eor:0x%x", 541 le16_to_cpu(lrd->type), log->bp, log->page, dstoffset); 542 543 /* page not full ? */ 544 if (dstoffset < LOGPSIZE - LOGPTLRSIZE) 545 return lsn; 546 547 pageFull: 548 /* page become full: move on to next page */ 549 lmNextPage(log); 550 551 bp = (struct lbuf *) log->bp; 552 lp = (struct logpage *) bp->l_ldata; 553 dstoffset = LOGPHDRSIZE; 554 src += nbytes; 555 } 556 557 return lsn; 558 } 559 560 561 /* 562 * NAME: lmNextPage() 563 * 564 * FUNCTION: write current page and allocate next page. 565 * 566 * PARAMETER: log 567 * 568 * RETURN: 0 569 * 570 * serialization: LOG_LOCK() held on entry/exit 571 */ 572 static int lmNextPage(struct jfs_log * log) 573 { 574 struct logpage *lp; 575 int lspn; /* log sequence page number */ 576 int pn; /* current page number */ 577 struct lbuf *bp; 578 struct lbuf *nextbp; 579 struct tblock *tblk; 580 581 /* get current log page number and log sequence page number */ 582 pn = log->page; 583 bp = log->bp; 584 lp = (struct logpage *) bp->l_ldata; 585 lspn = le32_to_cpu(lp->h.page); 586 587 LOGGC_LOCK(log); 588 589 /* 590 * write or queue the full page at the tail of write queue 591 */ 592 /* get the tail tblk on commit queue */ 593 if (list_empty(&log->cqueue)) 594 tblk = NULL; 595 else 596 tblk = list_entry(log->cqueue.prev, struct tblock, cqueue); 597 598 /* every tblk who has COMMIT record on the current page, 599 * and has not been committed, must be on commit queue 600 * since tblk is queued at commit queueu at the time 601 * of writing its COMMIT record on the page before 602 * page becomes full (even though the tblk thread 603 * who wrote COMMIT record may have been suspended 604 * currently); 605 */ 606 607 /* is page bound with outstanding tail tblk ? */ 608 if (tblk && tblk->pn == pn) { 609 /* mark tblk for end-of-page */ 610 tblk->flag |= tblkGC_EOP; 611 612 if (log->cflag & logGC_PAGEOUT) { 613 /* if page is not already on write queue, 614 * just enqueue (no lbmWRITE to prevent redrive) 615 * buffer to wqueue to ensure correct serial order 616 * of the pages since log pages will be added 617 * continuously 618 */ 619 if (bp->l_wqnext == NULL) 620 lbmWrite(log, bp, 0, 0); 621 } else { 622 /* 623 * No current GC leader, initiate group commit 624 */ 625 log->cflag |= logGC_PAGEOUT; 626 lmGCwrite(log, 0); 627 } 628 } 629 /* page is not bound with outstanding tblk: 630 * init write or mark it to be redriven (lbmWRITE) 631 */ 632 else { 633 /* finalize the page */ 634 bp->l_ceor = bp->l_eor; 635 lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_ceor); 636 lbmWrite(log, bp, lbmWRITE | lbmRELEASE | lbmFREE, 0); 637 } 638 LOGGC_UNLOCK(log); 639 640 /* 641 * allocate/initialize next page 642 */ 643 /* if log wraps, the first data page of log is 2 644 * (0 never used, 1 is superblock). 645 */ 646 log->page = (pn == log->size - 1) ? 2 : pn + 1; 647 log->eor = LOGPHDRSIZE; /* ? valid page empty/full at logRedo() */ 648 649 /* allocate/initialize next log page buffer */ 650 nextbp = lbmAllocate(log, log->page); 651 nextbp->l_eor = log->eor; 652 log->bp = nextbp; 653 654 /* initialize next log page */ 655 lp = (struct logpage *) nextbp->l_ldata; 656 lp->h.page = lp->t.page = cpu_to_le32(lspn + 1); 657 lp->h.eor = lp->t.eor = cpu_to_le16(LOGPHDRSIZE); 658 659 return 0; 660 } 661 662 663 /* 664 * NAME: lmGroupCommit() 665 * 666 * FUNCTION: group commit 667 * initiate pageout of the pages with COMMIT in the order of 668 * page number - redrive pageout of the page at the head of 669 * pageout queue until full page has been written. 670 * 671 * RETURN: 672 * 673 * NOTE: 674 * LOGGC_LOCK serializes log group commit queue, and 675 * transaction blocks on the commit queue. 676 * N.B. LOG_LOCK is NOT held during lmGroupCommit(). 677 */ 678 int lmGroupCommit(struct jfs_log * log, struct tblock * tblk) 679 { 680 int rc = 0; 681 682 LOGGC_LOCK(log); 683 684 /* group committed already ? */ 685 if (tblk->flag & tblkGC_COMMITTED) { 686 if (tblk->flag & tblkGC_ERROR) 687 rc = -EIO; 688 689 LOGGC_UNLOCK(log); 690 return rc; 691 } 692 jfs_info("lmGroup Commit: tblk = 0x%p, gcrtc = %d", tblk, log->gcrtc); 693 694 if (tblk->xflag & COMMIT_LAZY) 695 tblk->flag |= tblkGC_LAZY; 696 697 if ((!(log->cflag & logGC_PAGEOUT)) && (!list_empty(&log->cqueue)) && 698 (!(tblk->xflag & COMMIT_LAZY) || test_bit(log_FLUSH, &log->flag) 699 || jfs_tlocks_low)) { 700 /* 701 * No pageout in progress 702 * 703 * start group commit as its group leader. 704 */ 705 log->cflag |= logGC_PAGEOUT; 706 707 lmGCwrite(log, 0); 708 } 709 710 if (tblk->xflag & COMMIT_LAZY) { 711 /* 712 * Lazy transactions can leave now 713 */ 714 LOGGC_UNLOCK(log); 715 return 0; 716 } 717 718 /* lmGCwrite gives up LOGGC_LOCK, check again */ 719 720 if (tblk->flag & tblkGC_COMMITTED) { 721 if (tblk->flag & tblkGC_ERROR) 722 rc = -EIO; 723 724 LOGGC_UNLOCK(log); 725 return rc; 726 } 727 728 /* upcount transaction waiting for completion 729 */ 730 log->gcrtc++; 731 tblk->flag |= tblkGC_READY; 732 733 __SLEEP_COND(tblk->gcwait, (tblk->flag & tblkGC_COMMITTED), 734 LOGGC_LOCK(log), LOGGC_UNLOCK(log)); 735 736 /* removed from commit queue */ 737 if (tblk->flag & tblkGC_ERROR) 738 rc = -EIO; 739 740 LOGGC_UNLOCK(log); 741 return rc; 742 } 743 744 /* 745 * NAME: lmGCwrite() 746 * 747 * FUNCTION: group commit write 748 * initiate write of log page, building a group of all transactions 749 * with commit records on that page. 750 * 751 * RETURN: None 752 * 753 * NOTE: 754 * LOGGC_LOCK must be held by caller. 755 * N.B. LOG_LOCK is NOT held during lmGroupCommit(). 756 */ 757 static void lmGCwrite(struct jfs_log * log, int cant_write) 758 { 759 struct lbuf *bp; 760 struct logpage *lp; 761 int gcpn; /* group commit page number */ 762 struct tblock *tblk; 763 struct tblock *xtblk = NULL; 764 765 /* 766 * build the commit group of a log page 767 * 768 * scan commit queue and make a commit group of all 769 * transactions with COMMIT records on the same log page. 770 */ 771 /* get the head tblk on the commit queue */ 772 gcpn = list_entry(log->cqueue.next, struct tblock, cqueue)->pn; 773 774 list_for_each_entry(tblk, &log->cqueue, cqueue) { 775 if (tblk->pn != gcpn) 776 break; 777 778 xtblk = tblk; 779 780 /* state transition: (QUEUE, READY) -> COMMIT */ 781 tblk->flag |= tblkGC_COMMIT; 782 } 783 tblk = xtblk; /* last tblk of the page */ 784 785 /* 786 * pageout to commit transactions on the log page. 787 */ 788 bp = (struct lbuf *) tblk->bp; 789 lp = (struct logpage *) bp->l_ldata; 790 /* is page already full ? */ 791 if (tblk->flag & tblkGC_EOP) { 792 /* mark page to free at end of group commit of the page */ 793 tblk->flag &= ~tblkGC_EOP; 794 tblk->flag |= tblkGC_FREE; 795 bp->l_ceor = bp->l_eor; 796 lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_ceor); 797 lbmWrite(log, bp, lbmWRITE | lbmRELEASE | lbmGC, 798 cant_write); 799 INCREMENT(lmStat.full_page); 800 } 801 /* page is not yet full */ 802 else { 803 bp->l_ceor = tblk->eor; /* ? bp->l_ceor = bp->l_eor; */ 804 lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_ceor); 805 lbmWrite(log, bp, lbmWRITE | lbmGC, cant_write); 806 INCREMENT(lmStat.partial_page); 807 } 808 } 809 810 /* 811 * NAME: lmPostGC() 812 * 813 * FUNCTION: group commit post-processing 814 * Processes transactions after their commit records have been written 815 * to disk, redriving log I/O if necessary. 816 * 817 * RETURN: None 818 * 819 * NOTE: 820 * This routine is called a interrupt time by lbmIODone 821 */ 822 static void lmPostGC(struct lbuf * bp) 823 { 824 unsigned long flags; 825 struct jfs_log *log = bp->l_log; 826 struct logpage *lp; 827 struct tblock *tblk, *temp; 828 829 //LOGGC_LOCK(log); 830 spin_lock_irqsave(&log->gclock, flags); 831 /* 832 * current pageout of group commit completed. 833 * 834 * remove/wakeup transactions from commit queue who were 835 * group committed with the current log page 836 */ 837 list_for_each_entry_safe(tblk, temp, &log->cqueue, cqueue) { 838 if (!(tblk->flag & tblkGC_COMMIT)) 839 break; 840 /* if transaction was marked GC_COMMIT then 841 * it has been shipped in the current pageout 842 * and made it to disk - it is committed. 843 */ 844 845 if (bp->l_flag & lbmERROR) 846 tblk->flag |= tblkGC_ERROR; 847 848 /* remove it from the commit queue */ 849 list_del(&tblk->cqueue); 850 tblk->flag &= ~tblkGC_QUEUE; 851 852 if (tblk == log->flush_tblk) { 853 /* we can stop flushing the log now */ 854 clear_bit(log_FLUSH, &log->flag); 855 log->flush_tblk = NULL; 856 } 857 858 jfs_info("lmPostGC: tblk = 0x%p, flag = 0x%x", tblk, 859 tblk->flag); 860 861 if (!(tblk->xflag & COMMIT_FORCE)) 862 /* 863 * Hand tblk over to lazy commit thread 864 */ 865 txLazyUnlock(tblk); 866 else { 867 /* state transition: COMMIT -> COMMITTED */ 868 tblk->flag |= tblkGC_COMMITTED; 869 870 if (tblk->flag & tblkGC_READY) 871 log->gcrtc--; 872 873 LOGGC_WAKEUP(tblk); 874 } 875 876 /* was page full before pageout ? 877 * (and this is the last tblk bound with the page) 878 */ 879 if (tblk->flag & tblkGC_FREE) 880 lbmFree(bp); 881 /* did page become full after pageout ? 882 * (and this is the last tblk bound with the page) 883 */ 884 else if (tblk->flag & tblkGC_EOP) { 885 /* finalize the page */ 886 lp = (struct logpage *) bp->l_ldata; 887 bp->l_ceor = bp->l_eor; 888 lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_eor); 889 jfs_info("lmPostGC: calling lbmWrite"); 890 lbmWrite(log, bp, lbmWRITE | lbmRELEASE | lbmFREE, 891 1); 892 } 893 894 } 895 896 /* are there any transactions who have entered lnGroupCommit() 897 * (whose COMMITs are after that of the last log page written. 898 * They are waiting for new group commit (above at (SLEEP 1)) 899 * or lazy transactions are on a full (queued) log page, 900 * select the latest ready transaction as new group leader and 901 * wake her up to lead her group. 902 */ 903 if ((!list_empty(&log->cqueue)) && 904 ((log->gcrtc > 0) || (tblk->bp->l_wqnext != NULL) || 905 test_bit(log_FLUSH, &log->flag) || jfs_tlocks_low)) 906 /* 907 * Call lmGCwrite with new group leader 908 */ 909 lmGCwrite(log, 1); 910 911 /* no transaction are ready yet (transactions are only just 912 * queued (GC_QUEUE) and not entered for group commit yet). 913 * the first transaction entering group commit 914 * will elect herself as new group leader. 915 */ 916 else 917 log->cflag &= ~logGC_PAGEOUT; 918 919 //LOGGC_UNLOCK(log); 920 spin_unlock_irqrestore(&log->gclock, flags); 921 return; 922 } 923 924 /* 925 * NAME: lmLogSync() 926 * 927 * FUNCTION: write log SYNCPT record for specified log 928 * if new sync address is available 929 * (normally the case if sync() is executed by back-ground 930 * process). 931 * calculate new value of i_nextsync which determines when 932 * this code is called again. 933 * 934 * PARAMETERS: log - log structure 935 * hard_sync - 1 to force all metadata to be written 936 * 937 * RETURN: 0 938 * 939 * serialization: LOG_LOCK() held on entry/exit 940 */ 941 static int lmLogSync(struct jfs_log * log, int hard_sync) 942 { 943 int logsize; 944 int written; /* written since last syncpt */ 945 int free; /* free space left available */ 946 int delta; /* additional delta to write normally */ 947 int more; /* additional write granted */ 948 struct lrd lrd; 949 int lsn; 950 struct logsyncblk *lp; 951 unsigned long flags; 952 953 /* push dirty metapages out to disk */ 954 if (hard_sync) 955 write_special_inodes(log, filemap_fdatawrite); 956 else 957 write_special_inodes(log, filemap_flush); 958 959 /* 960 * forward syncpt 961 */ 962 /* if last sync is same as last syncpt, 963 * invoke sync point forward processing to update sync. 964 */ 965 966 if (log->sync == log->syncpt) { 967 LOGSYNC_LOCK(log, flags); 968 if (list_empty(&log->synclist)) 969 log->sync = log->lsn; 970 else { 971 lp = list_entry(log->synclist.next, 972 struct logsyncblk, synclist); 973 log->sync = lp->lsn; 974 } 975 LOGSYNC_UNLOCK(log, flags); 976 977 } 978 979 /* if sync is different from last syncpt, 980 * write a SYNCPT record with syncpt = sync. 981 * reset syncpt = sync 982 */ 983 if (log->sync != log->syncpt) { 984 lrd.logtid = 0; 985 lrd.backchain = 0; 986 lrd.type = cpu_to_le16(LOG_SYNCPT); 987 lrd.length = 0; 988 lrd.log.syncpt.sync = cpu_to_le32(log->sync); 989 lsn = lmWriteRecord(log, NULL, &lrd, NULL); 990 991 log->syncpt = log->sync; 992 } else 993 lsn = log->lsn; 994 995 /* 996 * setup next syncpt trigger (SWAG) 997 */ 998 logsize = log->logsize; 999 1000 logdiff(written, lsn, log); 1001 free = logsize - written; 1002 delta = LOGSYNC_DELTA(logsize); 1003 more = min(free / 2, delta); 1004 if (more < 2 * LOGPSIZE) { 1005 jfs_warn("\n ... Log Wrap ... Log Wrap ... Log Wrap ...\n"); 1006 /* 1007 * log wrapping 1008 * 1009 * option 1 - panic ? No.! 1010 * option 2 - shutdown file systems 1011 * associated with log ? 1012 * option 3 - extend log ? 1013 */ 1014 /* 1015 * option 4 - second chance 1016 * 1017 * mark log wrapped, and continue. 1018 * when all active transactions are completed, 1019 * mark log vaild for recovery. 1020 * if crashed during invalid state, log state 1021 * implies invald log, forcing fsck(). 1022 */ 1023 /* mark log state log wrap in log superblock */ 1024 /* log->state = LOGWRAP; */ 1025 1026 /* reset sync point computation */ 1027 log->syncpt = log->sync = lsn; 1028 log->nextsync = delta; 1029 } else 1030 /* next syncpt trigger = written + more */ 1031 log->nextsync = written + more; 1032 1033 /* if number of bytes written from last sync point is more 1034 * than 1/4 of the log size, stop new transactions from 1035 * starting until all current transactions are completed 1036 * by setting syncbarrier flag. 1037 */ 1038 if (!test_bit(log_SYNCBARRIER, &log->flag) && 1039 (written > LOGSYNC_BARRIER(logsize)) && log->active) { 1040 set_bit(log_SYNCBARRIER, &log->flag); 1041 jfs_info("log barrier on: lsn=0x%x syncpt=0x%x", lsn, 1042 log->syncpt); 1043 /* 1044 * We may have to initiate group commit 1045 */ 1046 jfs_flush_journal(log, 0); 1047 } 1048 1049 return lsn; 1050 } 1051 1052 /* 1053 * NAME: jfs_syncpt 1054 * 1055 * FUNCTION: write log SYNCPT record for specified log 1056 * 1057 * PARAMETERS: log - log structure 1058 * hard_sync - set to 1 to force metadata to be written 1059 */ 1060 void jfs_syncpt(struct jfs_log *log, int hard_sync) 1061 { LOG_LOCK(log); 1062 lmLogSync(log, hard_sync); 1063 LOG_UNLOCK(log); 1064 } 1065 1066 /* 1067 * NAME: lmLogOpen() 1068 * 1069 * FUNCTION: open the log on first open; 1070 * insert filesystem in the active list of the log. 1071 * 1072 * PARAMETER: ipmnt - file system mount inode 1073 * iplog - log inode (out) 1074 * 1075 * RETURN: 1076 * 1077 * serialization: 1078 */ 1079 int lmLogOpen(struct super_block *sb) 1080 { 1081 int rc; 1082 struct block_device *bdev; 1083 struct jfs_log *log; 1084 struct jfs_sb_info *sbi = JFS_SBI(sb); 1085 1086 if (sbi->flag & JFS_NOINTEGRITY) 1087 return open_dummy_log(sb); 1088 1089 if (sbi->mntflag & JFS_INLINELOG) 1090 return open_inline_log(sb); 1091 1092 mutex_lock(&jfs_log_mutex); 1093 list_for_each_entry(log, &jfs_external_logs, journal_list) { 1094 if (log->bdev->bd_dev == sbi->logdev) { 1095 if (memcmp(log->uuid, sbi->loguuid, 1096 sizeof(log->uuid))) { 1097 jfs_warn("wrong uuid on JFS journal\n"); 1098 mutex_unlock(&jfs_log_mutex); 1099 return -EINVAL; 1100 } 1101 /* 1102 * add file system to log active file system list 1103 */ 1104 if ((rc = lmLogFileSystem(log, sbi, 1))) { 1105 mutex_unlock(&jfs_log_mutex); 1106 return rc; 1107 } 1108 goto journal_found; 1109 } 1110 } 1111 1112 if (!(log = kzalloc(sizeof(struct jfs_log), GFP_KERNEL))) { 1113 mutex_unlock(&jfs_log_mutex); 1114 return -ENOMEM; 1115 } 1116 INIT_LIST_HEAD(&log->sb_list); 1117 init_waitqueue_head(&log->syncwait); 1118 1119 /* 1120 * external log as separate logical volume 1121 * 1122 * file systems to log may have n-to-1 relationship; 1123 */ 1124 1125 bdev = open_by_devnum(sbi->logdev, FMODE_READ|FMODE_WRITE); 1126 if (IS_ERR(bdev)) { 1127 rc = -PTR_ERR(bdev); 1128 goto free; 1129 } 1130 1131 if ((rc = bd_claim(bdev, log))) { 1132 goto close; 1133 } 1134 1135 log->bdev = bdev; 1136 memcpy(log->uuid, sbi->loguuid, sizeof(log->uuid)); 1137 1138 /* 1139 * initialize log: 1140 */ 1141 if ((rc = lmLogInit(log))) 1142 goto unclaim; 1143 1144 list_add(&log->journal_list, &jfs_external_logs); 1145 1146 /* 1147 * add file system to log active file system list 1148 */ 1149 if ((rc = lmLogFileSystem(log, sbi, 1))) 1150 goto shutdown; 1151 1152 journal_found: 1153 LOG_LOCK(log); 1154 list_add(&sbi->log_list, &log->sb_list); 1155 sbi->log = log; 1156 LOG_UNLOCK(log); 1157 1158 mutex_unlock(&jfs_log_mutex); 1159 return 0; 1160 1161 /* 1162 * unwind on error 1163 */ 1164 shutdown: /* unwind lbmLogInit() */ 1165 list_del(&log->journal_list); 1166 lbmLogShutdown(log); 1167 1168 unclaim: 1169 bd_release(bdev); 1170 1171 close: /* close external log device */ 1172 blkdev_put(bdev, FMODE_READ|FMODE_WRITE); 1173 1174 free: /* free log descriptor */ 1175 mutex_unlock(&jfs_log_mutex); 1176 kfree(log); 1177 1178 jfs_warn("lmLogOpen: exit(%d)", rc); 1179 return rc; 1180 } 1181 1182 static int open_inline_log(struct super_block *sb) 1183 { 1184 struct jfs_log *log; 1185 int rc; 1186 1187 if (!(log = kzalloc(sizeof(struct jfs_log), GFP_KERNEL))) 1188 return -ENOMEM; 1189 INIT_LIST_HEAD(&log->sb_list); 1190 init_waitqueue_head(&log->syncwait); 1191 1192 set_bit(log_INLINELOG, &log->flag); 1193 log->bdev = sb->s_bdev; 1194 log->base = addressPXD(&JFS_SBI(sb)->logpxd); 1195 log->size = lengthPXD(&JFS_SBI(sb)->logpxd) >> 1196 (L2LOGPSIZE - sb->s_blocksize_bits); 1197 log->l2bsize = sb->s_blocksize_bits; 1198 ASSERT(L2LOGPSIZE >= sb->s_blocksize_bits); 1199 1200 /* 1201 * initialize log. 1202 */ 1203 if ((rc = lmLogInit(log))) { 1204 kfree(log); 1205 jfs_warn("lmLogOpen: exit(%d)", rc); 1206 return rc; 1207 } 1208 1209 list_add(&JFS_SBI(sb)->log_list, &log->sb_list); 1210 JFS_SBI(sb)->log = log; 1211 1212 return rc; 1213 } 1214 1215 static int open_dummy_log(struct super_block *sb) 1216 { 1217 int rc; 1218 1219 mutex_lock(&jfs_log_mutex); 1220 if (!dummy_log) { 1221 dummy_log = kzalloc(sizeof(struct jfs_log), GFP_KERNEL); 1222 if (!dummy_log) { 1223 mutex_unlock(&jfs_log_mutex); 1224 return -ENOMEM; 1225 } 1226 INIT_LIST_HEAD(&dummy_log->sb_list); 1227 init_waitqueue_head(&dummy_log->syncwait); 1228 dummy_log->no_integrity = 1; 1229 /* Make up some stuff */ 1230 dummy_log->base = 0; 1231 dummy_log->size = 1024; 1232 rc = lmLogInit(dummy_log); 1233 if (rc) { 1234 kfree(dummy_log); 1235 dummy_log = NULL; 1236 mutex_unlock(&jfs_log_mutex); 1237 return rc; 1238 } 1239 } 1240 1241 LOG_LOCK(dummy_log); 1242 list_add(&JFS_SBI(sb)->log_list, &dummy_log->sb_list); 1243 JFS_SBI(sb)->log = dummy_log; 1244 LOG_UNLOCK(dummy_log); 1245 mutex_unlock(&jfs_log_mutex); 1246 1247 return 0; 1248 } 1249 1250 /* 1251 * NAME: lmLogInit() 1252 * 1253 * FUNCTION: log initialization at first log open. 1254 * 1255 * logredo() (or logformat()) should have been run previously. 1256 * initialize the log from log superblock. 1257 * set the log state in the superblock to LOGMOUNT and 1258 * write SYNCPT log record. 1259 * 1260 * PARAMETER: log - log structure 1261 * 1262 * RETURN: 0 - if ok 1263 * -EINVAL - bad log magic number or superblock dirty 1264 * error returned from logwait() 1265 * 1266 * serialization: single first open thread 1267 */ 1268 int lmLogInit(struct jfs_log * log) 1269 { 1270 int rc = 0; 1271 struct lrd lrd; 1272 struct logsuper *logsuper; 1273 struct lbuf *bpsuper; 1274 struct lbuf *bp; 1275 struct logpage *lp; 1276 int lsn = 0; 1277 1278 jfs_info("lmLogInit: log:0x%p", log); 1279 1280 /* initialize the group commit serialization lock */ 1281 LOGGC_LOCK_INIT(log); 1282 1283 /* allocate/initialize the log write serialization lock */ 1284 LOG_LOCK_INIT(log); 1285 1286 LOGSYNC_LOCK_INIT(log); 1287 1288 INIT_LIST_HEAD(&log->synclist); 1289 1290 INIT_LIST_HEAD(&log->cqueue); 1291 log->flush_tblk = NULL; 1292 1293 log->count = 0; 1294 1295 /* 1296 * initialize log i/o 1297 */ 1298 if ((rc = lbmLogInit(log))) 1299 return rc; 1300 1301 if (!test_bit(log_INLINELOG, &log->flag)) 1302 log->l2bsize = L2LOGPSIZE; 1303 1304 /* check for disabled journaling to disk */ 1305 if (log->no_integrity) { 1306 /* 1307 * Journal pages will still be filled. When the time comes 1308 * to actually do the I/O, the write is not done, and the 1309 * endio routine is called directly. 1310 */ 1311 bp = lbmAllocate(log , 0); 1312 log->bp = bp; 1313 bp->l_pn = bp->l_eor = 0; 1314 } else { 1315 /* 1316 * validate log superblock 1317 */ 1318 if ((rc = lbmRead(log, 1, &bpsuper))) 1319 goto errout10; 1320 1321 logsuper = (struct logsuper *) bpsuper->l_ldata; 1322 1323 if (logsuper->magic != cpu_to_le32(LOGMAGIC)) { 1324 jfs_warn("*** Log Format Error ! ***"); 1325 rc = -EINVAL; 1326 goto errout20; 1327 } 1328 1329 /* logredo() should have been run successfully. */ 1330 if (logsuper->state != cpu_to_le32(LOGREDONE)) { 1331 jfs_warn("*** Log Is Dirty ! ***"); 1332 rc = -EINVAL; 1333 goto errout20; 1334 } 1335 1336 /* initialize log from log superblock */ 1337 if (test_bit(log_INLINELOG,&log->flag)) { 1338 if (log->size != le32_to_cpu(logsuper->size)) { 1339 rc = -EINVAL; 1340 goto errout20; 1341 } 1342 jfs_info("lmLogInit: inline log:0x%p base:0x%Lx " 1343 "size:0x%x", log, 1344 (unsigned long long) log->base, log->size); 1345 } else { 1346 if (memcmp(logsuper->uuid, log->uuid, 16)) { 1347 jfs_warn("wrong uuid on JFS log device"); 1348 goto errout20; 1349 } 1350 log->size = le32_to_cpu(logsuper->size); 1351 log->l2bsize = le32_to_cpu(logsuper->l2bsize); 1352 jfs_info("lmLogInit: external log:0x%p base:0x%Lx " 1353 "size:0x%x", log, 1354 (unsigned long long) log->base, log->size); 1355 } 1356 1357 log->page = le32_to_cpu(logsuper->end) / LOGPSIZE; 1358 log->eor = le32_to_cpu(logsuper->end) - (LOGPSIZE * log->page); 1359 1360 /* 1361 * initialize for log append write mode 1362 */ 1363 /* establish current/end-of-log page/buffer */ 1364 if ((rc = lbmRead(log, log->page, &bp))) 1365 goto errout20; 1366 1367 lp = (struct logpage *) bp->l_ldata; 1368 1369 jfs_info("lmLogInit: lsn:0x%x page:%d eor:%d:%d", 1370 le32_to_cpu(logsuper->end), log->page, log->eor, 1371 le16_to_cpu(lp->h.eor)); 1372 1373 log->bp = bp; 1374 bp->l_pn = log->page; 1375 bp->l_eor = log->eor; 1376 1377 /* if current page is full, move on to next page */ 1378 if (log->eor >= LOGPSIZE - LOGPTLRSIZE) 1379 lmNextPage(log); 1380 1381 /* 1382 * initialize log syncpoint 1383 */ 1384 /* 1385 * write the first SYNCPT record with syncpoint = 0 1386 * (i.e., log redo up to HERE !); 1387 * remove current page from lbm write queue at end of pageout 1388 * (to write log superblock update), but do not release to 1389 * freelist; 1390 */ 1391 lrd.logtid = 0; 1392 lrd.backchain = 0; 1393 lrd.type = cpu_to_le16(LOG_SYNCPT); 1394 lrd.length = 0; 1395 lrd.log.syncpt.sync = 0; 1396 lsn = lmWriteRecord(log, NULL, &lrd, NULL); 1397 bp = log->bp; 1398 bp->l_ceor = bp->l_eor; 1399 lp = (struct logpage *) bp->l_ldata; 1400 lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_eor); 1401 lbmWrite(log, bp, lbmWRITE | lbmSYNC, 0); 1402 if ((rc = lbmIOWait(bp, 0))) 1403 goto errout30; 1404 1405 /* 1406 * update/write superblock 1407 */ 1408 logsuper->state = cpu_to_le32(LOGMOUNT); 1409 log->serial = le32_to_cpu(logsuper->serial) + 1; 1410 logsuper->serial = cpu_to_le32(log->serial); 1411 lbmDirectWrite(log, bpsuper, lbmWRITE | lbmRELEASE | lbmSYNC); 1412 if ((rc = lbmIOWait(bpsuper, lbmFREE))) 1413 goto errout30; 1414 } 1415 1416 /* initialize logsync parameters */ 1417 log->logsize = (log->size - 2) << L2LOGPSIZE; 1418 log->lsn = lsn; 1419 log->syncpt = lsn; 1420 log->sync = log->syncpt; 1421 log->nextsync = LOGSYNC_DELTA(log->logsize); 1422 1423 jfs_info("lmLogInit: lsn:0x%x syncpt:0x%x sync:0x%x", 1424 log->lsn, log->syncpt, log->sync); 1425 1426 /* 1427 * initialize for lazy/group commit 1428 */ 1429 log->clsn = lsn; 1430 1431 return 0; 1432 1433 /* 1434 * unwind on error 1435 */ 1436 errout30: /* release log page */ 1437 log->wqueue = NULL; 1438 bp->l_wqnext = NULL; 1439 lbmFree(bp); 1440 1441 errout20: /* release log superblock */ 1442 lbmFree(bpsuper); 1443 1444 errout10: /* unwind lbmLogInit() */ 1445 lbmLogShutdown(log); 1446 1447 jfs_warn("lmLogInit: exit(%d)", rc); 1448 return rc; 1449 } 1450 1451 1452 /* 1453 * NAME: lmLogClose() 1454 * 1455 * FUNCTION: remove file system <ipmnt> from active list of log <iplog> 1456 * and close it on last close. 1457 * 1458 * PARAMETER: sb - superblock 1459 * 1460 * RETURN: errors from subroutines 1461 * 1462 * serialization: 1463 */ 1464 int lmLogClose(struct super_block *sb) 1465 { 1466 struct jfs_sb_info *sbi = JFS_SBI(sb); 1467 struct jfs_log *log = sbi->log; 1468 struct block_device *bdev; 1469 int rc = 0; 1470 1471 jfs_info("lmLogClose: log:0x%p", log); 1472 1473 mutex_lock(&jfs_log_mutex); 1474 LOG_LOCK(log); 1475 list_del(&sbi->log_list); 1476 LOG_UNLOCK(log); 1477 sbi->log = NULL; 1478 1479 /* 1480 * We need to make sure all of the "written" metapages 1481 * actually make it to disk 1482 */ 1483 sync_blockdev(sb->s_bdev); 1484 1485 if (test_bit(log_INLINELOG, &log->flag)) { 1486 /* 1487 * in-line log in host file system 1488 */ 1489 rc = lmLogShutdown(log); 1490 kfree(log); 1491 goto out; 1492 } 1493 1494 if (!log->no_integrity) 1495 lmLogFileSystem(log, sbi, 0); 1496 1497 if (!list_empty(&log->sb_list)) 1498 goto out; 1499 1500 /* 1501 * TODO: ensure that the dummy_log is in a state to allow 1502 * lbmLogShutdown to deallocate all the buffers and call 1503 * kfree against dummy_log. For now, leave dummy_log & its 1504 * buffers in memory, and resuse if another no-integrity mount 1505 * is requested. 1506 */ 1507 if (log->no_integrity) 1508 goto out; 1509 1510 /* 1511 * external log as separate logical volume 1512 */ 1513 list_del(&log->journal_list); 1514 bdev = log->bdev; 1515 rc = lmLogShutdown(log); 1516 1517 bd_release(bdev); 1518 blkdev_put(bdev, FMODE_READ|FMODE_WRITE); 1519 1520 kfree(log); 1521 1522 out: 1523 mutex_unlock(&jfs_log_mutex); 1524 jfs_info("lmLogClose: exit(%d)", rc); 1525 return rc; 1526 } 1527 1528 1529 /* 1530 * NAME: jfs_flush_journal() 1531 * 1532 * FUNCTION: initiate write of any outstanding transactions to the journal 1533 * and optionally wait until they are all written to disk 1534 * 1535 * wait == 0 flush until latest txn is committed, don't wait 1536 * wait == 1 flush until latest txn is committed, wait 1537 * wait > 1 flush until all txn's are complete, wait 1538 */ 1539 void jfs_flush_journal(struct jfs_log *log, int wait) 1540 { 1541 int i; 1542 struct tblock *target = NULL; 1543 1544 /* jfs_write_inode may call us during read-only mount */ 1545 if (!log) 1546 return; 1547 1548 jfs_info("jfs_flush_journal: log:0x%p wait=%d", log, wait); 1549 1550 LOGGC_LOCK(log); 1551 1552 if (!list_empty(&log->cqueue)) { 1553 /* 1554 * This ensures that we will keep writing to the journal as long 1555 * as there are unwritten commit records 1556 */ 1557 target = list_entry(log->cqueue.prev, struct tblock, cqueue); 1558 1559 if (test_bit(log_FLUSH, &log->flag)) { 1560 /* 1561 * We're already flushing. 1562 * if flush_tblk is NULL, we are flushing everything, 1563 * so leave it that way. Otherwise, update it to the 1564 * latest transaction 1565 */ 1566 if (log->flush_tblk) 1567 log->flush_tblk = target; 1568 } else { 1569 /* Only flush until latest transaction is committed */ 1570 log->flush_tblk = target; 1571 set_bit(log_FLUSH, &log->flag); 1572 1573 /* 1574 * Initiate I/O on outstanding transactions 1575 */ 1576 if (!(log->cflag & logGC_PAGEOUT)) { 1577 log->cflag |= logGC_PAGEOUT; 1578 lmGCwrite(log, 0); 1579 } 1580 } 1581 } 1582 if ((wait > 1) || test_bit(log_SYNCBARRIER, &log->flag)) { 1583 /* Flush until all activity complete */ 1584 set_bit(log_FLUSH, &log->flag); 1585 log->flush_tblk = NULL; 1586 } 1587 1588 if (wait && target && !(target->flag & tblkGC_COMMITTED)) { 1589 DECLARE_WAITQUEUE(__wait, current); 1590 1591 add_wait_queue(&target->gcwait, &__wait); 1592 set_current_state(TASK_UNINTERRUPTIBLE); 1593 LOGGC_UNLOCK(log); 1594 schedule(); 1595 __set_current_state(TASK_RUNNING); 1596 LOGGC_LOCK(log); 1597 remove_wait_queue(&target->gcwait, &__wait); 1598 } 1599 LOGGC_UNLOCK(log); 1600 1601 if (wait < 2) 1602 return; 1603 1604 write_special_inodes(log, filemap_fdatawrite); 1605 1606 /* 1607 * If there was recent activity, we may need to wait 1608 * for the lazycommit thread to catch up 1609 */ 1610 if ((!list_empty(&log->cqueue)) || !list_empty(&log->synclist)) { 1611 for (i = 0; i < 200; i++) { /* Too much? */ 1612 msleep(250); 1613 write_special_inodes(log, filemap_fdatawrite); 1614 if (list_empty(&log->cqueue) && 1615 list_empty(&log->synclist)) 1616 break; 1617 } 1618 } 1619 assert(list_empty(&log->cqueue)); 1620 1621 #ifdef CONFIG_JFS_DEBUG 1622 if (!list_empty(&log->synclist)) { 1623 struct logsyncblk *lp; 1624 1625 printk(KERN_ERR "jfs_flush_journal: synclist not empty\n"); 1626 list_for_each_entry(lp, &log->synclist, synclist) { 1627 if (lp->xflag & COMMIT_PAGE) { 1628 struct metapage *mp = (struct metapage *)lp; 1629 print_hex_dump(KERN_ERR, "metapage: ", 1630 DUMP_PREFIX_ADDRESS, 16, 4, 1631 mp, sizeof(struct metapage), 0); 1632 print_hex_dump(KERN_ERR, "page: ", 1633 DUMP_PREFIX_ADDRESS, 16, 1634 sizeof(long), mp->page, 1635 sizeof(struct page), 0); 1636 } else 1637 print_hex_dump(KERN_ERR, "tblock:", 1638 DUMP_PREFIX_ADDRESS, 16, 4, 1639 lp, sizeof(struct tblock), 0); 1640 } 1641 } 1642 #else 1643 WARN_ON(!list_empty(&log->synclist)); 1644 #endif 1645 clear_bit(log_FLUSH, &log->flag); 1646 } 1647 1648 /* 1649 * NAME: lmLogShutdown() 1650 * 1651 * FUNCTION: log shutdown at last LogClose(). 1652 * 1653 * write log syncpt record. 1654 * update super block to set redone flag to 0. 1655 * 1656 * PARAMETER: log - log inode 1657 * 1658 * RETURN: 0 - success 1659 * 1660 * serialization: single last close thread 1661 */ 1662 int lmLogShutdown(struct jfs_log * log) 1663 { 1664 int rc; 1665 struct lrd lrd; 1666 int lsn; 1667 struct logsuper *logsuper; 1668 struct lbuf *bpsuper; 1669 struct lbuf *bp; 1670 struct logpage *lp; 1671 1672 jfs_info("lmLogShutdown: log:0x%p", log); 1673 1674 jfs_flush_journal(log, 2); 1675 1676 /* 1677 * write the last SYNCPT record with syncpoint = 0 1678 * (i.e., log redo up to HERE !) 1679 */ 1680 lrd.logtid = 0; 1681 lrd.backchain = 0; 1682 lrd.type = cpu_to_le16(LOG_SYNCPT); 1683 lrd.length = 0; 1684 lrd.log.syncpt.sync = 0; 1685 1686 lsn = lmWriteRecord(log, NULL, &lrd, NULL); 1687 bp = log->bp; 1688 lp = (struct logpage *) bp->l_ldata; 1689 lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_eor); 1690 lbmWrite(log, log->bp, lbmWRITE | lbmRELEASE | lbmSYNC, 0); 1691 lbmIOWait(log->bp, lbmFREE); 1692 log->bp = NULL; 1693 1694 /* 1695 * synchronous update log superblock 1696 * mark log state as shutdown cleanly 1697 * (i.e., Log does not need to be replayed). 1698 */ 1699 if ((rc = lbmRead(log, 1, &bpsuper))) 1700 goto out; 1701 1702 logsuper = (struct logsuper *) bpsuper->l_ldata; 1703 logsuper->state = cpu_to_le32(LOGREDONE); 1704 logsuper->end = cpu_to_le32(lsn); 1705 lbmDirectWrite(log, bpsuper, lbmWRITE | lbmRELEASE | lbmSYNC); 1706 rc = lbmIOWait(bpsuper, lbmFREE); 1707 1708 jfs_info("lmLogShutdown: lsn:0x%x page:%d eor:%d", 1709 lsn, log->page, log->eor); 1710 1711 out: 1712 /* 1713 * shutdown per log i/o 1714 */ 1715 lbmLogShutdown(log); 1716 1717 if (rc) { 1718 jfs_warn("lmLogShutdown: exit(%d)", rc); 1719 } 1720 return rc; 1721 } 1722 1723 1724 /* 1725 * NAME: lmLogFileSystem() 1726 * 1727 * FUNCTION: insert (<activate> = true)/remove (<activate> = false) 1728 * file system into/from log active file system list. 1729 * 1730 * PARAMETE: log - pointer to logs inode. 1731 * fsdev - kdev_t of filesystem. 1732 * serial - pointer to returned log serial number 1733 * activate - insert/remove device from active list. 1734 * 1735 * RETURN: 0 - success 1736 * errors returned by vms_iowait(). 1737 */ 1738 static int lmLogFileSystem(struct jfs_log * log, struct jfs_sb_info *sbi, 1739 int activate) 1740 { 1741 int rc = 0; 1742 int i; 1743 struct logsuper *logsuper; 1744 struct lbuf *bpsuper; 1745 char *uuid = sbi->uuid; 1746 1747 /* 1748 * insert/remove file system device to log active file system list. 1749 */ 1750 if ((rc = lbmRead(log, 1, &bpsuper))) 1751 return rc; 1752 1753 logsuper = (struct logsuper *) bpsuper->l_ldata; 1754 if (activate) { 1755 for (i = 0; i < MAX_ACTIVE; i++) 1756 if (!memcmp(logsuper->active[i].uuid, NULL_UUID, 16)) { 1757 memcpy(logsuper->active[i].uuid, uuid, 16); 1758 sbi->aggregate = i; 1759 break; 1760 } 1761 if (i == MAX_ACTIVE) { 1762 jfs_warn("Too many file systems sharing journal!"); 1763 lbmFree(bpsuper); 1764 return -EMFILE; /* Is there a better rc? */ 1765 } 1766 } else { 1767 for (i = 0; i < MAX_ACTIVE; i++) 1768 if (!memcmp(logsuper->active[i].uuid, uuid, 16)) { 1769 memcpy(logsuper->active[i].uuid, NULL_UUID, 16); 1770 break; 1771 } 1772 if (i == MAX_ACTIVE) { 1773 jfs_warn("Somebody stomped on the journal!"); 1774 lbmFree(bpsuper); 1775 return -EIO; 1776 } 1777 1778 } 1779 1780 /* 1781 * synchronous write log superblock: 1782 * 1783 * write sidestream bypassing write queue: 1784 * at file system mount, log super block is updated for 1785 * activation of the file system before any log record 1786 * (MOUNT record) of the file system, and at file system 1787 * unmount, all meta data for the file system has been 1788 * flushed before log super block is updated for deactivation 1789 * of the file system. 1790 */ 1791 lbmDirectWrite(log, bpsuper, lbmWRITE | lbmRELEASE | lbmSYNC); 1792 rc = lbmIOWait(bpsuper, lbmFREE); 1793 1794 return rc; 1795 } 1796 1797 /* 1798 * log buffer manager (lbm) 1799 * ------------------------ 1800 * 1801 * special purpose buffer manager supporting log i/o requirements. 1802 * 1803 * per log write queue: 1804 * log pageout occurs in serial order by fifo write queue and 1805 * restricting to a single i/o in pregress at any one time. 1806 * a circular singly-linked list 1807 * (log->wrqueue points to the tail, and buffers are linked via 1808 * bp->wrqueue field), and 1809 * maintains log page in pageout ot waiting for pageout in serial pageout. 1810 */ 1811 1812 /* 1813 * lbmLogInit() 1814 * 1815 * initialize per log I/O setup at lmLogInit() 1816 */ 1817 static int lbmLogInit(struct jfs_log * log) 1818 { /* log inode */ 1819 int i; 1820 struct lbuf *lbuf; 1821 1822 jfs_info("lbmLogInit: log:0x%p", log); 1823 1824 /* initialize current buffer cursor */ 1825 log->bp = NULL; 1826 1827 /* initialize log device write queue */ 1828 log->wqueue = NULL; 1829 1830 /* 1831 * Each log has its own buffer pages allocated to it. These are 1832 * not managed by the page cache. This ensures that a transaction 1833 * writing to the log does not block trying to allocate a page from 1834 * the page cache (for the log). This would be bad, since page 1835 * allocation waits on the kswapd thread that may be committing inodes 1836 * which would cause log activity. Was that clear? I'm trying to 1837 * avoid deadlock here. 1838 */ 1839 init_waitqueue_head(&log->free_wait); 1840 1841 log->lbuf_free = NULL; 1842 1843 for (i = 0; i < LOGPAGES;) { 1844 char *buffer; 1845 uint offset; 1846 struct page *page; 1847 1848 buffer = (char *) get_zeroed_page(GFP_KERNEL); 1849 if (buffer == NULL) 1850 goto error; 1851 page = virt_to_page(buffer); 1852 for (offset = 0; offset < PAGE_SIZE; offset += LOGPSIZE) { 1853 lbuf = kmalloc(sizeof(struct lbuf), GFP_KERNEL); 1854 if (lbuf == NULL) { 1855 if (offset == 0) 1856 free_page((unsigned long) buffer); 1857 goto error; 1858 } 1859 if (offset) /* we already have one reference */ 1860 get_page(page); 1861 lbuf->l_offset = offset; 1862 lbuf->l_ldata = buffer + offset; 1863 lbuf->l_page = page; 1864 lbuf->l_log = log; 1865 init_waitqueue_head(&lbuf->l_ioevent); 1866 1867 lbuf->l_freelist = log->lbuf_free; 1868 log->lbuf_free = lbuf; 1869 i++; 1870 } 1871 } 1872 1873 return (0); 1874 1875 error: 1876 lbmLogShutdown(log); 1877 return -ENOMEM; 1878 } 1879 1880 1881 /* 1882 * lbmLogShutdown() 1883 * 1884 * finalize per log I/O setup at lmLogShutdown() 1885 */ 1886 static void lbmLogShutdown(struct jfs_log * log) 1887 { 1888 struct lbuf *lbuf; 1889 1890 jfs_info("lbmLogShutdown: log:0x%p", log); 1891 1892 lbuf = log->lbuf_free; 1893 while (lbuf) { 1894 struct lbuf *next = lbuf->l_freelist; 1895 __free_page(lbuf->l_page); 1896 kfree(lbuf); 1897 lbuf = next; 1898 } 1899 } 1900 1901 1902 /* 1903 * lbmAllocate() 1904 * 1905 * allocate an empty log buffer 1906 */ 1907 static struct lbuf *lbmAllocate(struct jfs_log * log, int pn) 1908 { 1909 struct lbuf *bp; 1910 unsigned long flags; 1911 1912 /* 1913 * recycle from log buffer freelist if any 1914 */ 1915 LCACHE_LOCK(flags); 1916 LCACHE_SLEEP_COND(log->free_wait, (bp = log->lbuf_free), flags); 1917 log->lbuf_free = bp->l_freelist; 1918 LCACHE_UNLOCK(flags); 1919 1920 bp->l_flag = 0; 1921 1922 bp->l_wqnext = NULL; 1923 bp->l_freelist = NULL; 1924 1925 bp->l_pn = pn; 1926 bp->l_blkno = log->base + (pn << (L2LOGPSIZE - log->l2bsize)); 1927 bp->l_ceor = 0; 1928 1929 return bp; 1930 } 1931 1932 1933 /* 1934 * lbmFree() 1935 * 1936 * release a log buffer to freelist 1937 */ 1938 static void lbmFree(struct lbuf * bp) 1939 { 1940 unsigned long flags; 1941 1942 LCACHE_LOCK(flags); 1943 1944 lbmfree(bp); 1945 1946 LCACHE_UNLOCK(flags); 1947 } 1948 1949 static void lbmfree(struct lbuf * bp) 1950 { 1951 struct jfs_log *log = bp->l_log; 1952 1953 assert(bp->l_wqnext == NULL); 1954 1955 /* 1956 * return the buffer to head of freelist 1957 */ 1958 bp->l_freelist = log->lbuf_free; 1959 log->lbuf_free = bp; 1960 1961 wake_up(&log->free_wait); 1962 return; 1963 } 1964 1965 1966 /* 1967 * NAME: lbmRedrive 1968 * 1969 * FUNCTION: add a log buffer to the log redrive list 1970 * 1971 * PARAMETER: 1972 * bp - log buffer 1973 * 1974 * NOTES: 1975 * Takes log_redrive_lock. 1976 */ 1977 static inline void lbmRedrive(struct lbuf *bp) 1978 { 1979 unsigned long flags; 1980 1981 spin_lock_irqsave(&log_redrive_lock, flags); 1982 bp->l_redrive_next = log_redrive_list; 1983 log_redrive_list = bp; 1984 spin_unlock_irqrestore(&log_redrive_lock, flags); 1985 1986 wake_up_process(jfsIOthread); 1987 } 1988 1989 1990 /* 1991 * lbmRead() 1992 */ 1993 static int lbmRead(struct jfs_log * log, int pn, struct lbuf ** bpp) 1994 { 1995 struct bio *bio; 1996 struct lbuf *bp; 1997 1998 /* 1999 * allocate a log buffer 2000 */ 2001 *bpp = bp = lbmAllocate(log, pn); 2002 jfs_info("lbmRead: bp:0x%p pn:0x%x", bp, pn); 2003 2004 bp->l_flag |= lbmREAD; 2005 2006 bio = bio_alloc(GFP_NOFS, 1); 2007 2008 bio->bi_sector = bp->l_blkno << (log->l2bsize - 9); 2009 bio->bi_bdev = log->bdev; 2010 bio->bi_io_vec[0].bv_page = bp->l_page; 2011 bio->bi_io_vec[0].bv_len = LOGPSIZE; 2012 bio->bi_io_vec[0].bv_offset = bp->l_offset; 2013 2014 bio->bi_vcnt = 1; 2015 bio->bi_idx = 0; 2016 bio->bi_size = LOGPSIZE; 2017 2018 bio->bi_end_io = lbmIODone; 2019 bio->bi_private = bp; 2020 submit_bio(READ_SYNC, bio); 2021 2022 wait_event(bp->l_ioevent, (bp->l_flag != lbmREAD)); 2023 2024 return 0; 2025 } 2026 2027 2028 /* 2029 * lbmWrite() 2030 * 2031 * buffer at head of pageout queue stays after completion of 2032 * partial-page pageout and redriven by explicit initiation of 2033 * pageout by caller until full-page pageout is completed and 2034 * released. 2035 * 2036 * device driver i/o done redrives pageout of new buffer at 2037 * head of pageout queue when current buffer at head of pageout 2038 * queue is released at the completion of its full-page pageout. 2039 * 2040 * LOGGC_LOCK() serializes lbmWrite() by lmNextPage() and lmGroupCommit(). 2041 * LCACHE_LOCK() serializes xflag between lbmWrite() and lbmIODone() 2042 */ 2043 static void lbmWrite(struct jfs_log * log, struct lbuf * bp, int flag, 2044 int cant_block) 2045 { 2046 struct lbuf *tail; 2047 unsigned long flags; 2048 2049 jfs_info("lbmWrite: bp:0x%p flag:0x%x pn:0x%x", bp, flag, bp->l_pn); 2050 2051 /* map the logical block address to physical block address */ 2052 bp->l_blkno = 2053 log->base + (bp->l_pn << (L2LOGPSIZE - log->l2bsize)); 2054 2055 LCACHE_LOCK(flags); /* disable+lock */ 2056 2057 /* 2058 * initialize buffer for device driver 2059 */ 2060 bp->l_flag = flag; 2061 2062 /* 2063 * insert bp at tail of write queue associated with log 2064 * 2065 * (request is either for bp already/currently at head of queue 2066 * or new bp to be inserted at tail) 2067 */ 2068 tail = log->wqueue; 2069 2070 /* is buffer not already on write queue ? */ 2071 if (bp->l_wqnext == NULL) { 2072 /* insert at tail of wqueue */ 2073 if (tail == NULL) { 2074 log->wqueue = bp; 2075 bp->l_wqnext = bp; 2076 } else { 2077 log->wqueue = bp; 2078 bp->l_wqnext = tail->l_wqnext; 2079 tail->l_wqnext = bp; 2080 } 2081 2082 tail = bp; 2083 } 2084 2085 /* is buffer at head of wqueue and for write ? */ 2086 if ((bp != tail->l_wqnext) || !(flag & lbmWRITE)) { 2087 LCACHE_UNLOCK(flags); /* unlock+enable */ 2088 return; 2089 } 2090 2091 LCACHE_UNLOCK(flags); /* unlock+enable */ 2092 2093 if (cant_block) 2094 lbmRedrive(bp); 2095 else if (flag & lbmSYNC) 2096 lbmStartIO(bp); 2097 else { 2098 LOGGC_UNLOCK(log); 2099 lbmStartIO(bp); 2100 LOGGC_LOCK(log); 2101 } 2102 } 2103 2104 2105 /* 2106 * lbmDirectWrite() 2107 * 2108 * initiate pageout bypassing write queue for sidestream 2109 * (e.g., log superblock) write; 2110 */ 2111 static void lbmDirectWrite(struct jfs_log * log, struct lbuf * bp, int flag) 2112 { 2113 jfs_info("lbmDirectWrite: bp:0x%p flag:0x%x pn:0x%x", 2114 bp, flag, bp->l_pn); 2115 2116 /* 2117 * initialize buffer for device driver 2118 */ 2119 bp->l_flag = flag | lbmDIRECT; 2120 2121 /* map the logical block address to physical block address */ 2122 bp->l_blkno = 2123 log->base + (bp->l_pn << (L2LOGPSIZE - log->l2bsize)); 2124 2125 /* 2126 * initiate pageout of the page 2127 */ 2128 lbmStartIO(bp); 2129 } 2130 2131 2132 /* 2133 * NAME: lbmStartIO() 2134 * 2135 * FUNCTION: Interface to DD strategy routine 2136 * 2137 * RETURN: none 2138 * 2139 * serialization: LCACHE_LOCK() is NOT held during log i/o; 2140 */ 2141 static void lbmStartIO(struct lbuf * bp) 2142 { 2143 struct bio *bio; 2144 struct jfs_log *log = bp->l_log; 2145 2146 jfs_info("lbmStartIO\n"); 2147 2148 bio = bio_alloc(GFP_NOFS, 1); 2149 bio->bi_sector = bp->l_blkno << (log->l2bsize - 9); 2150 bio->bi_bdev = log->bdev; 2151 bio->bi_io_vec[0].bv_page = bp->l_page; 2152 bio->bi_io_vec[0].bv_len = LOGPSIZE; 2153 bio->bi_io_vec[0].bv_offset = bp->l_offset; 2154 2155 bio->bi_vcnt = 1; 2156 bio->bi_idx = 0; 2157 bio->bi_size = LOGPSIZE; 2158 2159 bio->bi_end_io = lbmIODone; 2160 bio->bi_private = bp; 2161 2162 /* check if journaling to disk has been disabled */ 2163 if (log->no_integrity) { 2164 bio->bi_size = 0; 2165 lbmIODone(bio, 0); 2166 } else { 2167 submit_bio(WRITE_SYNC, bio); 2168 INCREMENT(lmStat.submitted); 2169 } 2170 } 2171 2172 2173 /* 2174 * lbmIOWait() 2175 */ 2176 static int lbmIOWait(struct lbuf * bp, int flag) 2177 { 2178 unsigned long flags; 2179 int rc = 0; 2180 2181 jfs_info("lbmIOWait1: bp:0x%p flag:0x%x:0x%x", bp, bp->l_flag, flag); 2182 2183 LCACHE_LOCK(flags); /* disable+lock */ 2184 2185 LCACHE_SLEEP_COND(bp->l_ioevent, (bp->l_flag & lbmDONE), flags); 2186 2187 rc = (bp->l_flag & lbmERROR) ? -EIO : 0; 2188 2189 if (flag & lbmFREE) 2190 lbmfree(bp); 2191 2192 LCACHE_UNLOCK(flags); /* unlock+enable */ 2193 2194 jfs_info("lbmIOWait2: bp:0x%p flag:0x%x:0x%x", bp, bp->l_flag, flag); 2195 return rc; 2196 } 2197 2198 /* 2199 * lbmIODone() 2200 * 2201 * executed at INTIODONE level 2202 */ 2203 static void lbmIODone(struct bio *bio, int error) 2204 { 2205 struct lbuf *bp = bio->bi_private; 2206 struct lbuf *nextbp, *tail; 2207 struct jfs_log *log; 2208 unsigned long flags; 2209 2210 /* 2211 * get back jfs buffer bound to the i/o buffer 2212 */ 2213 jfs_info("lbmIODone: bp:0x%p flag:0x%x", bp, bp->l_flag); 2214 2215 LCACHE_LOCK(flags); /* disable+lock */ 2216 2217 bp->l_flag |= lbmDONE; 2218 2219 if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) { 2220 bp->l_flag |= lbmERROR; 2221 2222 jfs_err("lbmIODone: I/O error in JFS log"); 2223 } 2224 2225 bio_put(bio); 2226 2227 /* 2228 * pagein completion 2229 */ 2230 if (bp->l_flag & lbmREAD) { 2231 bp->l_flag &= ~lbmREAD; 2232 2233 LCACHE_UNLOCK(flags); /* unlock+enable */ 2234 2235 /* wakeup I/O initiator */ 2236 LCACHE_WAKEUP(&bp->l_ioevent); 2237 2238 return; 2239 } 2240 2241 /* 2242 * pageout completion 2243 * 2244 * the bp at the head of write queue has completed pageout. 2245 * 2246 * if single-commit/full-page pageout, remove the current buffer 2247 * from head of pageout queue, and redrive pageout with 2248 * the new buffer at head of pageout queue; 2249 * otherwise, the partial-page pageout buffer stays at 2250 * the head of pageout queue to be redriven for pageout 2251 * by lmGroupCommit() until full-page pageout is completed. 2252 */ 2253 bp->l_flag &= ~lbmWRITE; 2254 INCREMENT(lmStat.pagedone); 2255 2256 /* update committed lsn */ 2257 log = bp->l_log; 2258 log->clsn = (bp->l_pn << L2LOGPSIZE) + bp->l_ceor; 2259 2260 if (bp->l_flag & lbmDIRECT) { 2261 LCACHE_WAKEUP(&bp->l_ioevent); 2262 LCACHE_UNLOCK(flags); 2263 return; 2264 } 2265 2266 tail = log->wqueue; 2267 2268 /* single element queue */ 2269 if (bp == tail) { 2270 /* remove head buffer of full-page pageout 2271 * from log device write queue 2272 */ 2273 if (bp->l_flag & lbmRELEASE) { 2274 log->wqueue = NULL; 2275 bp->l_wqnext = NULL; 2276 } 2277 } 2278 /* multi element queue */ 2279 else { 2280 /* remove head buffer of full-page pageout 2281 * from log device write queue 2282 */ 2283 if (bp->l_flag & lbmRELEASE) { 2284 nextbp = tail->l_wqnext = bp->l_wqnext; 2285 bp->l_wqnext = NULL; 2286 2287 /* 2288 * redrive pageout of next page at head of write queue: 2289 * redrive next page without any bound tblk 2290 * (i.e., page w/o any COMMIT records), or 2291 * first page of new group commit which has been 2292 * queued after current page (subsequent pageout 2293 * is performed synchronously, except page without 2294 * any COMMITs) by lmGroupCommit() as indicated 2295 * by lbmWRITE flag; 2296 */ 2297 if (nextbp->l_flag & lbmWRITE) { 2298 /* 2299 * We can't do the I/O at interrupt time. 2300 * The jfsIO thread can do it 2301 */ 2302 lbmRedrive(nextbp); 2303 } 2304 } 2305 } 2306 2307 /* 2308 * synchronous pageout: 2309 * 2310 * buffer has not necessarily been removed from write queue 2311 * (e.g., synchronous write of partial-page with COMMIT): 2312 * leave buffer for i/o initiator to dispose 2313 */ 2314 if (bp->l_flag & lbmSYNC) { 2315 LCACHE_UNLOCK(flags); /* unlock+enable */ 2316 2317 /* wakeup I/O initiator */ 2318 LCACHE_WAKEUP(&bp->l_ioevent); 2319 } 2320 2321 /* 2322 * Group Commit pageout: 2323 */ 2324 else if (bp->l_flag & lbmGC) { 2325 LCACHE_UNLOCK(flags); 2326 lmPostGC(bp); 2327 } 2328 2329 /* 2330 * asynchronous pageout: 2331 * 2332 * buffer must have been removed from write queue: 2333 * insert buffer at head of freelist where it can be recycled 2334 */ 2335 else { 2336 assert(bp->l_flag & lbmRELEASE); 2337 assert(bp->l_flag & lbmFREE); 2338 lbmfree(bp); 2339 2340 LCACHE_UNLOCK(flags); /* unlock+enable */ 2341 } 2342 } 2343 2344 int jfsIOWait(void *arg) 2345 { 2346 struct lbuf *bp; 2347 2348 do { 2349 spin_lock_irq(&log_redrive_lock); 2350 while ((bp = log_redrive_list)) { 2351 log_redrive_list = bp->l_redrive_next; 2352 bp->l_redrive_next = NULL; 2353 spin_unlock_irq(&log_redrive_lock); 2354 lbmStartIO(bp); 2355 spin_lock_irq(&log_redrive_lock); 2356 } 2357 2358 if (freezing(current)) { 2359 spin_unlock_irq(&log_redrive_lock); 2360 refrigerator(); 2361 } else { 2362 set_current_state(TASK_INTERRUPTIBLE); 2363 spin_unlock_irq(&log_redrive_lock); 2364 schedule(); 2365 __set_current_state(TASK_RUNNING); 2366 } 2367 } while (!kthread_should_stop()); 2368 2369 jfs_info("jfsIOWait being killed!"); 2370 return 0; 2371 } 2372 2373 /* 2374 * NAME: lmLogFormat()/jfs_logform() 2375 * 2376 * FUNCTION: format file system log 2377 * 2378 * PARAMETERS: 2379 * log - volume log 2380 * logAddress - start address of log space in FS block 2381 * logSize - length of log space in FS block; 2382 * 2383 * RETURN: 0 - success 2384 * -EIO - i/o error 2385 * 2386 * XXX: We're synchronously writing one page at a time. This needs to 2387 * be improved by writing multiple pages at once. 2388 */ 2389 int lmLogFormat(struct jfs_log *log, s64 logAddress, int logSize) 2390 { 2391 int rc = -EIO; 2392 struct jfs_sb_info *sbi; 2393 struct logsuper *logsuper; 2394 struct logpage *lp; 2395 int lspn; /* log sequence page number */ 2396 struct lrd *lrd_ptr; 2397 int npages = 0; 2398 struct lbuf *bp; 2399 2400 jfs_info("lmLogFormat: logAddress:%Ld logSize:%d", 2401 (long long)logAddress, logSize); 2402 2403 sbi = list_entry(log->sb_list.next, struct jfs_sb_info, log_list); 2404 2405 /* allocate a log buffer */ 2406 bp = lbmAllocate(log, 1); 2407 2408 npages = logSize >> sbi->l2nbperpage; 2409 2410 /* 2411 * log space: 2412 * 2413 * page 0 - reserved; 2414 * page 1 - log superblock; 2415 * page 2 - log data page: A SYNC log record is written 2416 * into this page at logform time; 2417 * pages 3-N - log data page: set to empty log data pages; 2418 */ 2419 /* 2420 * init log superblock: log page 1 2421 */ 2422 logsuper = (struct logsuper *) bp->l_ldata; 2423 2424 logsuper->magic = cpu_to_le32(LOGMAGIC); 2425 logsuper->version = cpu_to_le32(LOGVERSION); 2426 logsuper->state = cpu_to_le32(LOGREDONE); 2427 logsuper->flag = cpu_to_le32(sbi->mntflag); /* ? */ 2428 logsuper->size = cpu_to_le32(npages); 2429 logsuper->bsize = cpu_to_le32(sbi->bsize); 2430 logsuper->l2bsize = cpu_to_le32(sbi->l2bsize); 2431 logsuper->end = cpu_to_le32(2 * LOGPSIZE + LOGPHDRSIZE + LOGRDSIZE); 2432 2433 bp->l_flag = lbmWRITE | lbmSYNC | lbmDIRECT; 2434 bp->l_blkno = logAddress + sbi->nbperpage; 2435 lbmStartIO(bp); 2436 if ((rc = lbmIOWait(bp, 0))) 2437 goto exit; 2438 2439 /* 2440 * init pages 2 to npages-1 as log data pages: 2441 * 2442 * log page sequence number (lpsn) initialization: 2443 * 2444 * pn: 0 1 2 3 n-1 2445 * +-----+-----+=====+=====+===.....===+=====+ 2446 * lspn: N-1 0 1 N-2 2447 * <--- N page circular file ----> 2448 * 2449 * the N (= npages-2) data pages of the log is maintained as 2450 * a circular file for the log records; 2451 * lpsn grows by 1 monotonically as each log page is written 2452 * to the circular file of the log; 2453 * and setLogpage() will not reset the page number even if 2454 * the eor is equal to LOGPHDRSIZE. In order for binary search 2455 * still work in find log end process, we have to simulate the 2456 * log wrap situation at the log format time. 2457 * The 1st log page written will have the highest lpsn. Then 2458 * the succeeding log pages will have ascending order of 2459 * the lspn starting from 0, ... (N-2) 2460 */ 2461 lp = (struct logpage *) bp->l_ldata; 2462 /* 2463 * initialize 1st log page to be written: lpsn = N - 1, 2464 * write a SYNCPT log record is written to this page 2465 */ 2466 lp->h.page = lp->t.page = cpu_to_le32(npages - 3); 2467 lp->h.eor = lp->t.eor = cpu_to_le16(LOGPHDRSIZE + LOGRDSIZE); 2468 2469 lrd_ptr = (struct lrd *) &lp->data; 2470 lrd_ptr->logtid = 0; 2471 lrd_ptr->backchain = 0; 2472 lrd_ptr->type = cpu_to_le16(LOG_SYNCPT); 2473 lrd_ptr->length = 0; 2474 lrd_ptr->log.syncpt.sync = 0; 2475 2476 bp->l_blkno += sbi->nbperpage; 2477 bp->l_flag = lbmWRITE | lbmSYNC | lbmDIRECT; 2478 lbmStartIO(bp); 2479 if ((rc = lbmIOWait(bp, 0))) 2480 goto exit; 2481 2482 /* 2483 * initialize succeeding log pages: lpsn = 0, 1, ..., (N-2) 2484 */ 2485 for (lspn = 0; lspn < npages - 3; lspn++) { 2486 lp->h.page = lp->t.page = cpu_to_le32(lspn); 2487 lp->h.eor = lp->t.eor = cpu_to_le16(LOGPHDRSIZE); 2488 2489 bp->l_blkno += sbi->nbperpage; 2490 bp->l_flag = lbmWRITE | lbmSYNC | lbmDIRECT; 2491 lbmStartIO(bp); 2492 if ((rc = lbmIOWait(bp, 0))) 2493 goto exit; 2494 } 2495 2496 rc = 0; 2497 exit: 2498 /* 2499 * finalize log 2500 */ 2501 /* release the buffer */ 2502 lbmFree(bp); 2503 2504 return rc; 2505 } 2506 2507 #ifdef CONFIG_JFS_STATISTICS 2508 static int jfs_lmstats_proc_show(struct seq_file *m, void *v) 2509 { 2510 seq_printf(m, 2511 "JFS Logmgr stats\n" 2512 "================\n" 2513 "commits = %d\n" 2514 "writes submitted = %d\n" 2515 "writes completed = %d\n" 2516 "full pages submitted = %d\n" 2517 "partial pages submitted = %d\n", 2518 lmStat.commit, 2519 lmStat.submitted, 2520 lmStat.pagedone, 2521 lmStat.full_page, 2522 lmStat.partial_page); 2523 return 0; 2524 } 2525 2526 static int jfs_lmstats_proc_open(struct inode *inode, struct file *file) 2527 { 2528 return single_open(file, jfs_lmstats_proc_show, NULL); 2529 } 2530 2531 const struct file_operations jfs_lmstats_proc_fops = { 2532 .owner = THIS_MODULE, 2533 .open = jfs_lmstats_proc_open, 2534 .read = seq_read, 2535 .llseek = seq_lseek, 2536 .release = single_release, 2537 }; 2538 #endif /* CONFIG_JFS_STATISTICS */ 2539