fs/jfs/jfs_logmgr.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *   Copyright (C) International Business Machines Corp., 2000-2004
   4  *   Portions Copyright (C) Christoph Hellwig, 2001-2002
   5  */
   6
   7 /*
   8  *      jfs_logmgr.c: log manager
   9  *
  10  * for related information, see transaction manager (jfs_txnmgr.c), and
  11  * recovery manager (jfs_logredo.c).
  12  *
  13  * note: for detail, RTFS.
  14  *
  15  *      log buffer manager:
  16  * special purpose buffer manager supporting log i/o requirements.
  17  * per log serial pageout of logpage
  18  * queuing i/o requests and redrive i/o at iodone
  19  * maintain current logpage buffer
  20  * no caching since append only
  21  * appropriate jfs buffer cache buffers as needed
  22  *
  23  *      group commit:
  24  * transactions which wrote COMMIT records in the same in-memory
  25  * log page during the pageout of previous/current log page(s) are
  26  * committed together by the pageout of the page.
  27  *
  28  *      TBD lazy commit:
  29  * transactions are committed asynchronously when the log page
  30  * containing it COMMIT is paged out when it becomes full;
  31  *
  32  *      serialization:
  33  * . a per log lock serialize log write.
  34  * . a per log lock serialize group commit.
  35  * . a per log lock serialize log open/close;
  36  *
  37  *      TBD log integrity:
  38  * careful-write (ping-pong) of last logpage to recover from crash
  39  * in overwrite.
  40  * detection of split (out-of-order) write of physical sectors
  41  * of last logpage via timestamp at end of each sector
  42  * with its mirror data array at trailer).
  43  *
  44  *      alternatives:
  45  * lsn - 64-bit monotonically increasing integer vs
  46  * 32-bit lspn and page eor.
  47  */
  48
  49 #include <linux/fs.h>
  50 #include <linux/blkdev.h>
  51 #include <linux/interrupt.h>
  52 #include <linux/completion.h>
  53 #include <linux/kthread.h>
  54 #include <linux/buffer_head.h>          /* for sync_blockdev() */
  55 #include <linux/bio.h>
  56 #include <linux/freezer.h>
  57 #include <linux/export.h>
  58 #include <linux/delay.h>
  59 #include <linux/mutex.h>
  60 #include <linux/seq_file.h>
  61 #include <linux/slab.h>
  62 #include "jfs_incore.h"
  63 #include "jfs_filsys.h"
  64 #include "jfs_metapage.h"
  65 #include "jfs_superblock.h"
  66 #include "jfs_txnmgr.h"
  67 #include "jfs_debug.h"
  68
  69
  70 /*
  71  * lbuf's ready to be redriven.  Protected by log_redrive_lock (jfsIO thread)
  72  */
  73 static struct lbuf *log_redrive_list;
  74 static DEFINE_SPINLOCK(log_redrive_lock);
  75
  76
  77 /*
  78  *      log read/write serialization (per log)
  79  */
  80 #define LOG_LOCK_INIT(log)      mutex_init(&(log)->loglock)
  81 #define LOG_LOCK(log)           mutex_lock(&((log)->loglock))
  82 #define LOG_UNLOCK(log)         mutex_unlock(&((log)->loglock))
  83
  84
  85 /*
  86  *      log group commit serialization (per log)
  87  */
  88
  89 #define LOGGC_LOCK_INIT(log)    spin_lock_init(&(log)->gclock)
  90 #define LOGGC_LOCK(log)         spin_lock_irq(&(log)->gclock)
  91 #define LOGGC_UNLOCK(log)       spin_unlock_irq(&(log)->gclock)
  92 #define LOGGC_WAKEUP(tblk)      wake_up_all(&(tblk)->gcwait)
  93
  94 /*
  95  *      log sync serialization (per log)
  96  */
  97 #define LOGSYNC_DELTA(logsize)          min((logsize)/8, 128*LOGPSIZE)
  98 #define LOGSYNC_BARRIER(logsize)        ((logsize)/4)
  99 /*
 100 #define LOGSYNC_DELTA(logsize)          min((logsize)/4, 256*LOGPSIZE)
 101 #define LOGSYNC_BARRIER(logsize)        ((logsize)/2)
 102 */
 103
 104
 105 /*
 106  *      log buffer cache synchronization
 107  */
 108 static DEFINE_SPINLOCK(jfsLCacheLock);
 109
 110 #define LCACHE_LOCK(flags)      spin_lock_irqsave(&jfsLCacheLock, flags)
 111 #define LCACHE_UNLOCK(flags)    spin_unlock_irqrestore(&jfsLCacheLock, flags)
 112
 113 /*
 114  * See __SLEEP_COND in jfs_locks.h
 115  */
 116 #define LCACHE_SLEEP_COND(wq, cond, flags)      \
 117 do {                                            \
 118         if (cond)                               \
 119                 break;                          \
 120         __SLEEP_COND(wq, cond, LCACHE_LOCK(flags), LCACHE_UNLOCK(flags)); \
 121 } while (0)
 122
 123 #define LCACHE_WAKEUP(event)    wake_up(event)
 124
 125
 126 /*
 127  *      lbuf buffer cache (lCache) control
 128  */
 129 /* log buffer manager pageout control (cumulative, inclusive) */
 130 #define lbmREAD         0x0001
 131 #define lbmWRITE        0x0002  /* enqueue at tail of write queue;
 132                                  * init pageout if at head of queue;
 133                                  */
 134 #define lbmRELEASE      0x0004  /* remove from write queue
 135                                  * at completion of pageout;
 136                                  * do not free/recycle it yet:
 137                                  * caller will free it;
 138                                  */
 139 #define lbmSYNC         0x0008  /* do not return to freelist
 140                                  * when removed from write queue;
 141                                  */
 142 #define lbmFREE         0x0010  /* return to freelist
 143                                  * at completion of pageout;
 144                                  * the buffer may be recycled;
 145                                  */
 146 #define lbmDONE         0x0020
 147 #define lbmERROR        0x0040
 148 #define lbmGC           0x0080  /* lbmIODone to perform post-GC processing
 149                                  * of log page
 150                                  */
 151 #define lbmDIRECT       0x0100
 152
 153 /*
 154  * Global list of active external journals
 155  */
 156 static LIST_HEAD(jfs_external_logs);
 157 static struct jfs_log *dummy_log;
 158 static DEFINE_MUTEX(jfs_log_mutex);
 159
 160 /*
 161  * forward references
 162  */
 163 static int lmWriteRecord(struct jfs_log * log, struct tblock * tblk,
 164                          struct lrd * lrd, struct tlock * tlck);
 165
 166 static int lmNextPage(struct jfs_log * log);
 167 static int lmLogFileSystem(struct jfs_log * log, struct jfs_sb_info *sbi,
 168                            int activate);
 169
 170 static int open_inline_log(struct super_block *sb);
 171 static int open_dummy_log(struct super_block *sb);
 172 static int lbmLogInit(struct jfs_log * log);
 173 static void lbmLogShutdown(struct jfs_log * log);
 174 static struct lbuf *lbmAllocate(struct jfs_log * log, int);
 175 static void lbmFree(struct lbuf * bp);
 176 static void lbmfree(struct lbuf * bp);
 177 static int lbmRead(struct jfs_log * log, int pn, struct lbuf ** bpp);
 178 static void lbmWrite(struct jfs_log * log, struct lbuf * bp, int flag, int cant_block);
 179 static void lbmDirectWrite(struct jfs_log * log, struct lbuf * bp, int flag);
 180 static int lbmIOWait(struct lbuf * bp, int flag);
 181 static bio_end_io_t lbmIODone;
 182 static void lbmStartIO(struct lbuf * bp);
 183 static void lmGCwrite(struct jfs_log * log, int cant_block);
 184 static int lmLogSync(struct jfs_log * log, int hard_sync);
 185
 186
 187
 188 /*
 189  *      statistics
 190  */
 191 #ifdef CONFIG_JFS_STATISTICS
 192 static struct lmStat {
 193         uint commit;            /* # of commit */
 194         uint pagedone;          /* # of page written */
 195         uint submitted;         /* # of pages submitted */
 196         uint full_page;         /* # of full pages submitted */
 197         uint partial_page;      /* # of partial pages submitted */
 198 } lmStat;
 199 #endif
 200
 201 static void write_special_inodes(struct jfs_log *log,
 202                                  int (*writer)(struct address_space *))
 203 {
 204         struct jfs_sb_info *sbi;
 205
 206         list_for_each_entry(sbi, &log->sb_list, log_list) {
 207                 writer(sbi->ipbmap->i_mapping);
 208                 writer(sbi->ipimap->i_mapping);
 209                 writer(sbi->direct_inode->i_mapping);
 210         }
 211 }
 212
 213 /*
 214  * NAME:        lmLog()
 215  *
 216  * FUNCTION:    write a log record;
 217  *
 218  * PARAMETER:
 219  *
 220  * RETURN:      lsn - offset to the next log record to write (end-of-log);
 221  *              -1  - error;
 222  *
 223  * note: todo: log error handler
 224  */
 225 int lmLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
 226           struct tlock * tlck)
 227 {
 228         int lsn;
 229         int diffp, difft;
 230         struct metapage *mp = NULL;
 231         unsigned long flags;
 232
 233         jfs_info("lmLog: log:0x%p tblk:0x%p, lrd:0x%p tlck:0x%p",
 234                  log, tblk, lrd, tlck);
 235
 236         LOG_LOCK(log);
 237
 238         /* log by (out-of-transaction) JFS ? */
 239         if (tblk == NULL)
 240                 goto writeRecord;
 241
 242         /* log from page ? */
 243         if (tlck == NULL ||
 244             tlck->type & tlckBTROOT || (mp = tlck->mp) == NULL)
 245                 goto writeRecord;
 246
 247         /*
 248          *      initialize/update page/transaction recovery lsn
 249          */
 250         lsn = log->lsn;
 251
 252         LOGSYNC_LOCK(log, flags);
 253
 254         /*
 255          * initialize page lsn if first log write of the page
 256          */
 257         if (mp->lsn == 0) {
 258                 mp->log = log;
 259                 mp->lsn = lsn;
 260                 log->count++;
 261
 262                 /* insert page at tail of logsynclist */
 263                 list_add_tail(&mp->synclist, &log->synclist);
 264         }
 265
 266         /*
 267          *      initialize/update lsn of tblock of the page
 268          *
 269          * transaction inherits oldest lsn of pages associated
 270          * with allocation/deallocation of resources (their
 271          * log records are used to reconstruct allocation map
 272          * at recovery time: inode for inode allocation map,
 273          * B+-tree index of extent descriptors for block
 274          * allocation map);
 275          * allocation map pages inherit transaction lsn at
 276          * commit time to allow forwarding log syncpt past log
 277          * records associated with allocation/deallocation of
 278          * resources only after persistent map of these map pages
 279          * have been updated and propagated to home.
 280          */
 281         /*
 282          * initialize transaction lsn:
 283          */
 284         if (tblk->lsn == 0) {
 285                 /* inherit lsn of its first page logged */
 286                 tblk->lsn = mp->lsn;
 287                 log->count++;
 288
 289                 /* insert tblock after the page on logsynclist */
 290                 list_add(&tblk->synclist, &mp->synclist);
 291         }
 292         /*
 293          * update transaction lsn:
 294          */
 295         else {
 296                 /* inherit oldest/smallest lsn of page */
 297                 logdiff(diffp, mp->lsn, log);
 298                 logdiff(difft, tblk->lsn, log);
 299                 if (diffp < difft) {
 300                         /* update tblock lsn with page lsn */
 301                         tblk->lsn = mp->lsn;
 302
 303                         /* move tblock after page on logsynclist */
 304                         list_move(&tblk->synclist, &mp->synclist);
 305                 }
 306         }
 307
 308         LOGSYNC_UNLOCK(log, flags);
 309
 310         /*
 311          *      write the log record
 312          */
 313       writeRecord:
 314         lsn = lmWriteRecord(log, tblk, lrd, tlck);
 315
 316         /*
 317          * forward log syncpt if log reached next syncpt trigger
 318          */
 319         logdiff(diffp, lsn, log);
 320         if (diffp >= log->nextsync)
 321                 lsn = lmLogSync(log, 0);
 322
 323         /* update end-of-log lsn */
 324         log->lsn = lsn;
 325
 326         LOG_UNLOCK(log);
 327
 328         /* return end-of-log address */
 329         return lsn;
 330 }
 331
 332 /*
 333  * NAME:        lmWriteRecord()
 334  *
 335  * FUNCTION:    move the log record to current log page
 336  *
 337  * PARAMETER:   cd      - commit descriptor
 338  *
 339  * RETURN:      end-of-log address
 340  *
 341  * serialization: LOG_LOCK() held on entry/exit
 342  */
 343 static int
 344 lmWriteRecord(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
 345               struct tlock * tlck)
 346 {
 347         int lsn = 0;            /* end-of-log address */
 348         struct lbuf *bp;        /* dst log page buffer */
 349         struct logpage *lp;     /* dst log page */
 350         caddr_t dst;            /* destination address in log page */
 351         int dstoffset;          /* end-of-log offset in log page */
 352         int freespace;          /* free space in log page */
 353         caddr_t p;              /* src meta-data page */
 354         caddr_t src;
 355         int srclen;
 356         int nbytes;             /* number of bytes to move */
 357         int i;
 358         int len;
 359         struct linelock *linelock;
 360         struct lv *lv;
 361         struct lvd *lvd;
 362         int l2linesize;
 363
 364         len = 0;
 365
 366         /* retrieve destination log page to write */
 367         bp = (struct lbuf *) log->bp;
 368         lp = (struct logpage *) bp->l_ldata;
 369         dstoffset = log->eor;
 370
 371         /* any log data to write ? */
 372         if (tlck == NULL)
 373                 goto moveLrd;
 374
 375         /*
 376          *      move log record data
 377          */
 378         /* retrieve source meta-data page to log */
 379         if (tlck->flag & tlckPAGELOCK) {
 380                 p = (caddr_t) (tlck->mp->data);
 381                 linelock = (struct linelock *) & tlck->lock;
 382         }
 383         /* retrieve source in-memory inode to log */
 384         else if (tlck->flag & tlckINODELOCK) {
 385                 if (tlck->type & tlckDTREE)
 386                         p = (caddr_t) &JFS_IP(tlck->ip)->i_dtroot;
 387                 else
 388                         p = (caddr_t) &JFS_IP(tlck->ip)->i_xtroot;
 389                 linelock = (struct linelock *) & tlck->lock;
 390         }
 391         else {
 392                 jfs_err("lmWriteRecord: UFO tlck:0x%p", tlck);
 393                 return 0;       /* Probably should trap */
 394         }
 395         l2linesize = linelock->l2linesize;
 396
 397       moveData:
 398         ASSERT(linelock->index <= linelock->maxcnt);
 399
 400         lv = linelock->lv;
 401         for (i = 0; i < linelock->index; i++, lv++) {
 402                 if (lv->length == 0)
 403                         continue;
 404
 405                 /* is page full ? */
 406                 if (dstoffset >= LOGPSIZE - LOGPTLRSIZE) {
 407                         /* page become full: move on to next page */
 408                         lmNextPage(log);
 409
 410                         bp = log->bp;
 411                         lp = (struct logpage *) bp->l_ldata;
 412                         dstoffset = LOGPHDRSIZE;
 413                 }
 414
 415                 /*
 416                  * move log vector data
 417                  */
 418                 src = (u8 *) p + (lv->offset << l2linesize);
 419                 srclen = lv->length << l2linesize;
 420                 len += srclen;
 421                 while (srclen > 0) {
 422                         freespace = (LOGPSIZE - LOGPTLRSIZE) - dstoffset;
 423                         nbytes = min(freespace, srclen);
 424                         dst = (caddr_t) lp + dstoffset;
 425                         memcpy(dst, src, nbytes);
 426                         dstoffset += nbytes;
 427
 428                         /* is page not full ? */
 429                         if (dstoffset < LOGPSIZE - LOGPTLRSIZE)
 430                                 break;
 431
 432                         /* page become full: move on to next page */
 433                         lmNextPage(log);
 434
 435                         bp = (struct lbuf *) log->bp;
 436                         lp = (struct logpage *) bp->l_ldata;
 437                         dstoffset = LOGPHDRSIZE;
 438
 439                         srclen -= nbytes;
 440                         src += nbytes;
 441                 }
 442
 443                 /*
 444                  * move log vector descriptor
 445                  */
 446                 len += 4;
 447                 lvd = (struct lvd *) ((caddr_t) lp + dstoffset);
 448                 lvd->offset = cpu_to_le16(lv->offset);
 449                 lvd->length = cpu_to_le16(lv->length);
 450                 dstoffset += 4;
 451                 jfs_info("lmWriteRecord: lv offset:%d length:%d",
 452                          lv->offset, lv->length);
 453         }
 454
 455         if ((i = linelock->next)) {
 456                 linelock = (struct linelock *) lid_to_tlock(i);
 457                 goto moveData;
 458         }
 459
 460         /*
 461          *      move log record descriptor
 462          */
 463       moveLrd:
 464         lrd->length = cpu_to_le16(len);
 465
 466         src = (caddr_t) lrd;
 467         srclen = LOGRDSIZE;
 468
 469         while (srclen > 0) {
 470                 freespace = (LOGPSIZE - LOGPTLRSIZE) - dstoffset;
 471                 nbytes = min(freespace, srclen);
 472                 dst = (caddr_t) lp + dstoffset;
 473                 memcpy(dst, src, nbytes);
 474
 475                 dstoffset += nbytes;
 476                 srclen -= nbytes;
 477
 478                 /* are there more to move than freespace of page ? */
 479                 if (srclen)
 480                         goto pageFull;
 481
 482                 /*
 483                  * end of log record descriptor
 484                  */
 485
 486                 /* update last log record eor */
 487                 log->eor = dstoffset;
 488                 bp->l_eor = dstoffset;
 489                 lsn = (log->page << L2LOGPSIZE) + dstoffset;
 490
 491                 if (lrd->type & cpu_to_le16(LOG_COMMIT)) {
 492                         tblk->clsn = lsn;
 493                         jfs_info("wr: tclsn:0x%x, beor:0x%x", tblk->clsn,
 494                                  bp->l_eor);
 495
 496                         INCREMENT(lmStat.commit);       /* # of commit */
 497
 498                         /*
 499                          * enqueue tblock for group commit:
 500                          *
 501                          * enqueue tblock of non-trivial/synchronous COMMIT
 502                          * at tail of group commit queue
 503                          * (trivial/asynchronous COMMITs are ignored by
 504                          * group commit.)
 505                          */
 506                         LOGGC_LOCK(log);
 507
 508                         /* init tblock gc state */
 509                         tblk->flag = tblkGC_QUEUE;
 510                         tblk->bp = log->bp;
 511                         tblk->pn = log->page;
 512                         tblk->eor = log->eor;
 513
 514                         /* enqueue transaction to commit queue */
 515                         list_add_tail(&tblk->cqueue, &log->cqueue);
 516
 517                         LOGGC_UNLOCK(log);
 518                 }
 519
 520                 jfs_info("lmWriteRecord: lrd:0x%04x bp:0x%p pn:%d eor:0x%x",
 521                         le16_to_cpu(lrd->type), log->bp, log->page, dstoffset);
 522
 523                 /* page not full ? */
 524                 if (dstoffset < LOGPSIZE - LOGPTLRSIZE)
 525                         return lsn;
 526
 527               pageFull:
 528                 /* page become full: move on to next page */
 529                 lmNextPage(log);
 530
 531                 bp = (struct lbuf *) log->bp;
 532                 lp = (struct logpage *) bp->l_ldata;
 533                 dstoffset = LOGPHDRSIZE;
 534                 src += nbytes;
 535         }
 536
 537         return lsn;
 538 }
 539
 540
 541 /*
 542  * NAME:        lmNextPage()
 543  *
 544  * FUNCTION:    write current page and allocate next page.
 545  *
 546  * PARAMETER:   log
 547  *
 548  * RETURN:      0
 549  *
 550  * serialization: LOG_LOCK() held on entry/exit
 551  */
 552 static int lmNextPage(struct jfs_log * log)
 553 {
 554         struct logpage *lp;
 555         int lspn;               /* log sequence page number */
 556         int pn;                 /* current page number */
 557         struct lbuf *bp;
 558         struct lbuf *nextbp;
 559         struct tblock *tblk;
 560
 561         /* get current log page number and log sequence page number */
 562         pn = log->page;
 563         bp = log->bp;
 564         lp = (struct logpage *) bp->l_ldata;
 565         lspn = le32_to_cpu(lp->h.page);
 566
 567         LOGGC_LOCK(log);
 568
 569         /*
 570          *      write or queue the full page at the tail of write queue
 571          */
 572         /* get the tail tblk on commit queue */
 573         if (list_empty(&log->cqueue))
 574                 tblk = NULL;
 575         else
 576                 tblk = list_entry(log->cqueue.prev, struct tblock, cqueue);
 577
 578         /* every tblk who has COMMIT record on the current page,
 579          * and has not been committed, must be on commit queue
 580          * since tblk is queued at commit queueu at the time
 581          * of writing its COMMIT record on the page before
 582          * page becomes full (even though the tblk thread
 583          * who wrote COMMIT record may have been suspended
 584          * currently);
 585          */
 586
 587         /* is page bound with outstanding tail tblk ? */
 588         if (tblk && tblk->pn == pn) {
 589                 /* mark tblk for end-of-page */
 590                 tblk->flag |= tblkGC_EOP;
 591
 592                 if (log->cflag & logGC_PAGEOUT) {
 593                         /* if page is not already on write queue,
 594                          * just enqueue (no lbmWRITE to prevent redrive)
 595                          * buffer to wqueue to ensure correct serial order
 596                          * of the pages since log pages will be added
 597                          * continuously
 598                          */
 599                         if (bp->l_wqnext == NULL)
 600                                 lbmWrite(log, bp, 0, 0);
 601                 } else {
 602                         /*
 603                          * No current GC leader, initiate group commit
 604                          */
 605                         log->cflag |= logGC_PAGEOUT;
 606                         lmGCwrite(log, 0);
 607                 }
 608         }
 609         /* page is not bound with outstanding tblk:
 610          * init write or mark it to be redriven (lbmWRITE)
 611          */
 612         else {
 613                 /* finalize the page */
 614                 bp->l_ceor = bp->l_eor;
 615                 lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_ceor);
 616                 lbmWrite(log, bp, lbmWRITE | lbmRELEASE | lbmFREE, 0);
 617         }
 618         LOGGC_UNLOCK(log);
 619
 620         /*
 621          *      allocate/initialize next page
 622          */
 623         /* if log wraps, the first data page of log is 2
 624          * (0 never used, 1 is superblock).
 625          */
 626         log->page = (pn == log->size - 1) ? 2 : pn + 1;
 627         log->eor = LOGPHDRSIZE; /* ? valid page empty/full at logRedo() */
 628
 629         /* allocate/initialize next log page buffer */
 630         nextbp = lbmAllocate(log, log->page);
 631         nextbp->l_eor = log->eor;
 632         log->bp = nextbp;
 633
 634         /* initialize next log page */
 635         lp = (struct logpage *) nextbp->l_ldata;
 636         lp->h.page = lp->t.page = cpu_to_le32(lspn + 1);
 637         lp->h.eor = lp->t.eor = cpu_to_le16(LOGPHDRSIZE);
 638
 639         return 0;
 640 }
 641
 642
 643 /*
 644  * NAME:        lmGroupCommit()
 645  *
 646  * FUNCTION:    group commit
 647  *      initiate pageout of the pages with COMMIT in the order of
 648  *      page number - redrive pageout of the page at the head of
 649  *      pageout queue until full page has been written.
 650  *
 651  * RETURN:
 652  *
 653  * NOTE:
 654  *      LOGGC_LOCK serializes log group commit queue, and
 655  *      transaction blocks on the commit queue.
 656  *      N.B. LOG_LOCK is NOT held during lmGroupCommit().
 657  */
 658 int lmGroupCommit(struct jfs_log * log, struct tblock * tblk)
 659 {
 660         int rc = 0;
 661
 662         LOGGC_LOCK(log);
 663
 664         /* group committed already ? */
 665         if (tblk->flag & tblkGC_COMMITTED) {
 666                 if (tblk->flag & tblkGC_ERROR)
 667                         rc = -EIO;
 668
 669                 LOGGC_UNLOCK(log);
 670                 return rc;
 671         }
 672         jfs_info("lmGroup Commit: tblk = 0x%p, gcrtc = %d", tblk, log->gcrtc);
 673
 674         if (tblk->xflag & COMMIT_LAZY)
 675                 tblk->flag |= tblkGC_LAZY;
 676
 677         if ((!(log->cflag & logGC_PAGEOUT)) && (!list_empty(&log->cqueue)) &&
 678             (!(tblk->xflag & COMMIT_LAZY) || test_bit(log_FLUSH, &log->flag)
 679              || jfs_tlocks_low)) {
 680                 /*
 681                  * No pageout in progress
 682                  *
 683                  * start group commit as its group leader.
 684                  */
 685                 log->cflag |= logGC_PAGEOUT;
 686
 687                 lmGCwrite(log, 0);
 688         }
 689
 690         if (tblk->xflag & COMMIT_LAZY) {
 691                 /*
 692                  * Lazy transactions can leave now
 693                  */
 694                 LOGGC_UNLOCK(log);
 695                 return 0;
 696         }
 697
 698         /* lmGCwrite gives up LOGGC_LOCK, check again */
 699
 700         if (tblk->flag & tblkGC_COMMITTED) {
 701                 if (tblk->flag & tblkGC_ERROR)
 702                         rc = -EIO;
 703
 704                 LOGGC_UNLOCK(log);
 705                 return rc;
 706         }
 707
 708         /* upcount transaction waiting for completion
 709          */
 710         log->gcrtc++;
 711         tblk->flag |= tblkGC_READY;
 712
 713         __SLEEP_COND(tblk->gcwait, (tblk->flag & tblkGC_COMMITTED),
 714                      LOGGC_LOCK(log), LOGGC_UNLOCK(log));
 715
 716         /* removed from commit queue */
 717         if (tblk->flag & tblkGC_ERROR)
 718                 rc = -EIO;
 719
 720         LOGGC_UNLOCK(log);
 721         return rc;
 722 }
 723
 724 /*
 725  * NAME:        lmGCwrite()
 726  *
 727  * FUNCTION:    group commit write
 728  *      initiate write of log page, building a group of all transactions
 729  *      with commit records on that page.
 730  *
 731  * RETURN:      None
 732  *
 733  * NOTE:
 734  *      LOGGC_LOCK must be held by caller.
 735  *      N.B. LOG_LOCK is NOT held during lmGroupCommit().
 736  */
 737 static void lmGCwrite(struct jfs_log * log, int cant_write)
 738 {
 739         struct lbuf *bp;
 740         struct logpage *lp;
 741         int gcpn;               /* group commit page number */
 742         struct tblock *tblk;
 743         struct tblock *xtblk = NULL;
 744
 745         /*
 746          * build the commit group of a log page
 747          *
 748          * scan commit queue and make a commit group of all
 749          * transactions with COMMIT records on the same log page.
 750          */
 751         /* get the head tblk on the commit queue */
 752         gcpn = list_entry(log->cqueue.next, struct tblock, cqueue)->pn;
 753
 754         list_for_each_entry(tblk, &log->cqueue, cqueue) {
 755                 if (tblk->pn != gcpn)
 756                         break;
 757
 758                 xtblk = tblk;
 759
 760                 /* state transition: (QUEUE, READY) -> COMMIT */
 761                 tblk->flag |= tblkGC_COMMIT;
 762         }
 763         tblk = xtblk;           /* last tblk of the page */
 764
 765         /*
 766          * pageout to commit transactions on the log page.
 767          */
 768         bp = (struct lbuf *) tblk->bp;
 769         lp = (struct logpage *) bp->l_ldata;
 770         /* is page already full ? */
 771         if (tblk->flag & tblkGC_EOP) {
 772                 /* mark page to free at end of group commit of the page */
 773                 tblk->flag &= ~tblkGC_EOP;
 774                 tblk->flag |= tblkGC_FREE;
 775                 bp->l_ceor = bp->l_eor;
 776                 lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_ceor);
 777                 lbmWrite(log, bp, lbmWRITE | lbmRELEASE | lbmGC,
 778                          cant_write);
 779                 INCREMENT(lmStat.full_page);
 780         }
 781         /* page is not yet full */
 782         else {
 783                 bp->l_ceor = tblk->eor; /* ? bp->l_ceor = bp->l_eor; */
 784                 lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_ceor);
 785                 lbmWrite(log, bp, lbmWRITE | lbmGC, cant_write);
 786                 INCREMENT(lmStat.partial_page);
 787         }
 788 }
 789
 790 /*
 791  * NAME:        lmPostGC()
 792  *
 793  * FUNCTION:    group commit post-processing
 794  *      Processes transactions after their commit records have been written
 795  *      to disk, redriving log I/O if necessary.
 796  *
 797  * RETURN:      None
 798  *
 799  * NOTE:
 800  *      This routine is called a interrupt time by lbmIODone
 801  */
 802 static void lmPostGC(struct lbuf * bp)
 803 {
 804         unsigned long flags;
 805         struct jfs_log *log = bp->l_log;
 806         struct logpage *lp;
 807         struct tblock *tblk, *temp;
 808
 809         //LOGGC_LOCK(log);
 810         spin_lock_irqsave(&log->gclock, flags);
 811         /*
 812          * current pageout of group commit completed.
 813          *
 814          * remove/wakeup transactions from commit queue who were
 815          * group committed with the current log page
 816          */
 817         list_for_each_entry_safe(tblk, temp, &log->cqueue, cqueue) {
 818                 if (!(tblk->flag & tblkGC_COMMIT))
 819                         break;
 820                 /* if transaction was marked GC_COMMIT then
 821                  * it has been shipped in the current pageout
 822                  * and made it to disk - it is committed.
 823                  */
 824
 825                 if (bp->l_flag & lbmERROR)
 826                         tblk->flag |= tblkGC_ERROR;
 827
 828                 /* remove it from the commit queue */
 829                 list_del(&tblk->cqueue);
 830                 tblk->flag &= ~tblkGC_QUEUE;
 831
 832                 if (tblk == log->flush_tblk) {
 833                         /* we can stop flushing the log now */
 834                         clear_bit(log_FLUSH, &log->flag);
 835                         log->flush_tblk = NULL;
 836                 }
 837
 838                 jfs_info("lmPostGC: tblk = 0x%p, flag = 0x%x", tblk,
 839                          tblk->flag);
 840
 841                 if (!(tblk->xflag & COMMIT_FORCE))
 842                         /*
 843                          * Hand tblk over to lazy commit thread
 844                          */
 845                         txLazyUnlock(tblk);
 846                 else {
 847                         /* state transition: COMMIT -> COMMITTED */
 848                         tblk->flag |= tblkGC_COMMITTED;
 849
 850                         if (tblk->flag & tblkGC_READY)
 851                                 log->gcrtc--;
 852
 853                         LOGGC_WAKEUP(tblk);
 854                 }
 855
 856                 /* was page full before pageout ?
 857                  * (and this is the last tblk bound with the page)
 858                  */
 859                 if (tblk->flag & tblkGC_FREE)
 860                         lbmFree(bp);
 861                 /* did page become full after pageout ?
 862                  * (and this is the last tblk bound with the page)
 863                  */
 864                 else if (tblk->flag & tblkGC_EOP) {
 865                         /* finalize the page */
 866                         lp = (struct logpage *) bp->l_ldata;
 867                         bp->l_ceor = bp->l_eor;
 868                         lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_eor);
 869                         jfs_info("lmPostGC: calling lbmWrite");
 870                         lbmWrite(log, bp, lbmWRITE | lbmRELEASE | lbmFREE,
 871                                  1);
 872                 }
 873
 874         }
 875
 876         /* are there any transactions who have entered lnGroupCommit()
 877          * (whose COMMITs are after that of the last log page written.
 878          * They are waiting for new group commit (above at (SLEEP 1))
 879          * or lazy transactions are on a full (queued) log page,
 880          * select the latest ready transaction as new group leader and
 881          * wake her up to lead her group.
 882          */
 883         if ((!list_empty(&log->cqueue)) &&
 884             ((log->gcrtc > 0) || (tblk->bp->l_wqnext != NULL) ||
 885              test_bit(log_FLUSH, &log->flag) || jfs_tlocks_low))
 886                 /*
 887                  * Call lmGCwrite with new group leader
 888                  */
 889                 lmGCwrite(log, 1);
 890
 891         /* no transaction are ready yet (transactions are only just
 892          * queued (GC_QUEUE) and not entered for group commit yet).
 893          * the first transaction entering group commit
 894          * will elect herself as new group leader.
 895          */
 896         else
 897                 log->cflag &= ~logGC_PAGEOUT;
 898
 899         //LOGGC_UNLOCK(log);
 900         spin_unlock_irqrestore(&log->gclock, flags);
 901         return;
 902 }
 903
 904 /*
 905  * NAME:        lmLogSync()
 906  *
 907  * FUNCTION:    write log SYNCPT record for specified log
 908  *      if new sync address is available
 909  *      (normally the case if sync() is executed by back-ground
 910  *      process).
 911  *      calculate new value of i_nextsync which determines when
 912  *      this code is called again.
 913  *
 914  * PARAMETERS:  log     - log structure
 915  *              hard_sync - 1 to force all metadata to be written
 916  *
 917  * RETURN:      0
 918  *
 919  * serialization: LOG_LOCK() held on entry/exit
 920  */
 921 static int lmLogSync(struct jfs_log * log, int hard_sync)
 922 {
 923         int logsize;
 924         int written;            /* written since last syncpt */
 925         int free;               /* free space left available */
 926         int delta;              /* additional delta to write normally */
 927         int more;               /* additional write granted */
 928         struct lrd lrd;
 929         int lsn;
 930         struct logsyncblk *lp;
 931         unsigned long flags;
 932
 933         /* push dirty metapages out to disk */
 934         if (hard_sync)
 935                 write_special_inodes(log, filemap_fdatawrite);
 936         else
 937                 write_special_inodes(log, filemap_flush);
 938
 939         /*
 940          *      forward syncpt
 941          */
 942         /* if last sync is same as last syncpt,
 943          * invoke sync point forward processing to update sync.
 944          */
 945
 946         if (log->sync == log->syncpt) {
 947                 LOGSYNC_LOCK(log, flags);
 948                 if (list_empty(&log->synclist))
 949                         log->sync = log->lsn;
 950                 else {
 951                         lp = list_entry(log->synclist.next,
 952                                         struct logsyncblk, synclist);
 953                         log->sync = lp->lsn;
 954                 }
 955                 LOGSYNC_UNLOCK(log, flags);
 956
 957         }
 958
 959         /* if sync is different from last syncpt,
 960          * write a SYNCPT record with syncpt = sync.
 961          * reset syncpt = sync
 962          */
 963         if (log->sync != log->syncpt) {
 964                 lrd.logtid = 0;
 965                 lrd.backchain = 0;
 966                 lrd.type = cpu_to_le16(LOG_SYNCPT);
 967                 lrd.length = 0;
 968                 lrd.log.syncpt.sync = cpu_to_le32(log->sync);
 969                 lsn = lmWriteRecord(log, NULL, &lrd, NULL);
 970
 971                 log->syncpt = log->sync;
 972         } else
 973                 lsn = log->lsn;
 974
 975         /*
 976          *      setup next syncpt trigger (SWAG)
 977          */
 978         logsize = log->logsize;
 979
 980         logdiff(written, lsn, log);
 981         free = logsize - written;
 982         delta = LOGSYNC_DELTA(logsize);
 983         more = min(free / 2, delta);
 984         if (more < 2 * LOGPSIZE) {
 985                 jfs_warn("\n ... Log Wrap ... Log Wrap ... Log Wrap ...\n");
 986                 /*
 987                  *      log wrapping
 988                  *
 989                  * option 1 - panic ? No.!
 990                  * option 2 - shutdown file systems
 991                  *            associated with log ?
 992                  * option 3 - extend log ?
 993                  * option 4 - second chance
 994                  *
 995                  * mark log wrapped, and continue.
 996                  * when all active transactions are completed,
 997                  * mark log valid for recovery.
 998                  * if crashed during invalid state, log state
 999                  * implies invalid log, forcing fsck().
1000                  */
1001                 /* mark log state log wrap in log superblock */
1002                 /* log->state = LOGWRAP; */
1003
1004                 /* reset sync point computation */
1005                 log->syncpt = log->sync = lsn;
1006                 log->nextsync = delta;
1007         } else
1008                 /* next syncpt trigger = written + more */
1009                 log->nextsync = written + more;
1010
1011         /* if number of bytes written from last sync point is more
1012          * than 1/4 of the log size, stop new transactions from
1013          * starting until all current transactions are completed
1014          * by setting syncbarrier flag.
1015          */
1016         if (!test_bit(log_SYNCBARRIER, &log->flag) &&
1017             (written > LOGSYNC_BARRIER(logsize)) && log->active) {
1018                 set_bit(log_SYNCBARRIER, &log->flag);
1019                 jfs_info("log barrier on: lsn=0x%x syncpt=0x%x", lsn,
1020                          log->syncpt);
1021                 /*
1022                  * We may have to initiate group commit
1023                  */
1024                 jfs_flush_journal(log, 0);
1025         }
1026
1027         return lsn;
1028 }
1029
1030 /*
1031  * NAME:        jfs_syncpt
1032  *
1033  * FUNCTION:    write log SYNCPT record for specified log
1034  *
1035  * PARAMETERS:  log       - log structure
1036  *              hard_sync - set to 1 to force metadata to be written
1037  */
1038 void jfs_syncpt(struct jfs_log *log, int hard_sync)
1039 {       LOG_LOCK(log);
1040         if (!test_bit(log_QUIESCE, &log->flag))
1041                 lmLogSync(log, hard_sync);
1042         LOG_UNLOCK(log);
1043 }
1044
1045 /*
1046  * NAME:        lmLogOpen()
1047  *
1048  * FUNCTION:    open the log on first open;
1049  *      insert filesystem in the active list of the log.
1050  *
1051  * PARAMETER:   ipmnt   - file system mount inode
1052  *              iplog   - log inode (out)
1053  *
1054  * RETURN:
1055  *
1056  * serialization:
1057  */
1058 int lmLogOpen(struct super_block *sb)
1059 {
1060         int rc;
1061         struct file *bdev_file;
1062         struct jfs_log *log;
1063         struct jfs_sb_info *sbi = JFS_SBI(sb);
1064
1065         if (sbi->flag & JFS_NOINTEGRITY)
1066                 return open_dummy_log(sb);
1067
1068         if (sbi->mntflag & JFS_INLINELOG)
1069                 return open_inline_log(sb);
1070
1071         mutex_lock(&jfs_log_mutex);
1072         list_for_each_entry(log, &jfs_external_logs, journal_list) {
1073                 if (file_bdev(log->bdev_file)->bd_dev == sbi->logdev) {
1074                         if (!uuid_equal(&log->uuid, &sbi->loguuid)) {
1075                                 jfs_warn("wrong uuid on JFS journal");
1076                                 mutex_unlock(&jfs_log_mutex);
1077                                 return -EINVAL;
1078                         }
1079                         /*
1080                          * add file system to log active file system list
1081                          */
1082                         if ((rc = lmLogFileSystem(log, sbi, 1))) {
1083                                 mutex_unlock(&jfs_log_mutex);
1084                                 return rc;
1085                         }
1086                         goto journal_found;
1087                 }
1088         }
1089
1090         if (!(log = kzalloc(sizeof(struct jfs_log), GFP_KERNEL))) {
1091                 mutex_unlock(&jfs_log_mutex);
1092                 return -ENOMEM;
1093         }
1094         INIT_LIST_HEAD(&log->sb_list);
1095         init_waitqueue_head(&log->syncwait);
1096
1097         /*
1098          *      external log as separate logical volume
1099          *
1100          * file systems to log may have n-to-1 relationship;
1101          */
1102
1103         bdev_file = bdev_file_open_by_dev(sbi->logdev,
1104                         BLK_OPEN_READ | BLK_OPEN_WRITE, log, NULL);
1105         if (IS_ERR(bdev_file)) {
1106                 rc = PTR_ERR(bdev_file);
1107                 goto free;
1108         }
1109
1110         log->bdev_file = bdev_file;
1111         uuid_copy(&log->uuid, &sbi->loguuid);
1112
1113         /*
1114          * initialize log:
1115          */
1116         if ((rc = lmLogInit(log)))
1117                 goto close;
1118
1119         list_add(&log->journal_list, &jfs_external_logs);
1120
1121         /*
1122          * add file system to log active file system list
1123          */
1124         if ((rc = lmLogFileSystem(log, sbi, 1)))
1125                 goto shutdown;
1126
1127 journal_found:
1128         LOG_LOCK(log);
1129         list_add(&sbi->log_list, &log->sb_list);
1130         sbi->log = log;
1131         LOG_UNLOCK(log);
1132
1133         mutex_unlock(&jfs_log_mutex);
1134         return 0;
1135
1136         /*
1137          *      unwind on error
1138          */
1139       shutdown:         /* unwind lbmLogInit() */
1140         list_del(&log->journal_list);
1141         lbmLogShutdown(log);
1142
1143       close:            /* close external log device */
1144         fput(bdev_file);
1145
1146       free:             /* free log descriptor */
1147         mutex_unlock(&jfs_log_mutex);
1148         kfree(log);
1149
1150         jfs_warn("lmLogOpen: exit(%d)", rc);
1151         return rc;
1152 }
1153
1154 static int open_inline_log(struct super_block *sb)
1155 {
1156         struct jfs_log *log;
1157         int rc;
1158
1159         if (!(log = kzalloc(sizeof(struct jfs_log), GFP_KERNEL)))
1160                 return -ENOMEM;
1161         INIT_LIST_HEAD(&log->sb_list);
1162         init_waitqueue_head(&log->syncwait);
1163
1164         set_bit(log_INLINELOG, &log->flag);
1165         log->bdev_file = sb->s_bdev_file;
1166         log->base = addressPXD(&JFS_SBI(sb)->logpxd);
1167         log->size = lengthPXD(&JFS_SBI(sb)->logpxd) >>
1168             (L2LOGPSIZE - sb->s_blocksize_bits);
1169         log->l2bsize = sb->s_blocksize_bits;
1170         ASSERT(L2LOGPSIZE >= sb->s_blocksize_bits);
1171
1172         /*
1173          * initialize log.
1174          */
1175         if ((rc = lmLogInit(log))) {
1176                 kfree(log);
1177                 jfs_warn("lmLogOpen: exit(%d)", rc);
1178                 return rc;
1179         }
1180
1181         list_add(&JFS_SBI(sb)->log_list, &log->sb_list);
1182         JFS_SBI(sb)->log = log;
1183
1184         return rc;
1185 }
1186
1187 static int open_dummy_log(struct super_block *sb)
1188 {
1189         int rc;
1190
1191         mutex_lock(&jfs_log_mutex);
1192         if (!dummy_log) {
1193                 dummy_log = kzalloc(sizeof(struct jfs_log), GFP_KERNEL);
1194                 if (!dummy_log) {
1195                         mutex_unlock(&jfs_log_mutex);
1196                         return -ENOMEM;
1197                 }
1198                 INIT_LIST_HEAD(&dummy_log->sb_list);
1199                 init_waitqueue_head(&dummy_log->syncwait);
1200                 dummy_log->no_integrity = 1;
1201                 /* Make up some stuff */
1202                 dummy_log->base = 0;
1203                 dummy_log->size = 1024;
1204                 rc = lmLogInit(dummy_log);
1205                 if (rc) {
1206                         kfree(dummy_log);
1207                         dummy_log = NULL;
1208                         mutex_unlock(&jfs_log_mutex);
1209                         return rc;
1210                 }
1211         }
1212
1213         LOG_LOCK(dummy_log);
1214         list_add(&JFS_SBI(sb)->log_list, &dummy_log->sb_list);
1215         JFS_SBI(sb)->log = dummy_log;
1216         LOG_UNLOCK(dummy_log);
1217         mutex_unlock(&jfs_log_mutex);
1218
1219         return 0;
1220 }
1221
1222 /*
1223  * NAME:        lmLogInit()
1224  *
1225  * FUNCTION:    log initialization at first log open.
1226  *
1227  *      logredo() (or logformat()) should have been run previously.
1228  *      initialize the log from log superblock.
1229  *      set the log state in the superblock to LOGMOUNT and
1230  *      write SYNCPT log record.
1231  *
1232  * PARAMETER:   log     - log structure
1233  *
1234  * RETURN:      0       - if ok
1235  *              -EINVAL - bad log magic number or superblock dirty
1236  *              error returned from logwait()
1237  *
1238  * serialization: single first open thread
1239  */
1240 int lmLogInit(struct jfs_log * log)
1241 {
1242         int rc = 0;
1243         struct lrd lrd;
1244         struct logsuper *logsuper;
1245         struct lbuf *bpsuper;
1246         struct lbuf *bp;
1247         struct logpage *lp;
1248         int lsn = 0;
1249
1250         jfs_info("lmLogInit: log:0x%p", log);
1251
1252         /* initialize the group commit serialization lock */
1253         LOGGC_LOCK_INIT(log);
1254
1255         /* allocate/initialize the log write serialization lock */
1256         LOG_LOCK_INIT(log);
1257
1258         LOGSYNC_LOCK_INIT(log);
1259
1260         INIT_LIST_HEAD(&log->synclist);
1261
1262         INIT_LIST_HEAD(&log->cqueue);
1263         log->flush_tblk = NULL;
1264
1265         log->count = 0;
1266
1267         /*
1268          * initialize log i/o
1269          */
1270         if ((rc = lbmLogInit(log)))
1271                 return rc;
1272
1273         if (!test_bit(log_INLINELOG, &log->flag))
1274                 log->l2bsize = L2LOGPSIZE;
1275
1276         /* check for disabled journaling to disk */
1277         if (log->no_integrity) {
1278                 /*
1279                  * Journal pages will still be filled.  When the time comes
1280                  * to actually do the I/O, the write is not done, and the
1281                  * endio routine is called directly.
1282                  */
1283                 bp = lbmAllocate(log , 0);
1284                 log->bp = bp;
1285                 bp->l_pn = bp->l_eor = 0;
1286         } else {
1287                 /*
1288                  * validate log superblock
1289                  */
1290                 if ((rc = lbmRead(log, 1, &bpsuper)))
1291                         goto errout10;
1292
1293                 logsuper = (struct logsuper *) bpsuper->l_ldata;
1294
1295                 if (logsuper->magic != cpu_to_le32(LOGMAGIC)) {
1296                         jfs_warn("*** Log Format Error ! ***");
1297                         rc = -EINVAL;
1298                         goto errout20;
1299                 }
1300
1301                 /* logredo() should have been run successfully. */
1302                 if (logsuper->state != cpu_to_le32(LOGREDONE)) {
1303                         jfs_warn("*** Log Is Dirty ! ***");
1304                         rc = -EINVAL;
1305                         goto errout20;
1306                 }
1307
1308                 /* initialize log from log superblock */
1309                 if (test_bit(log_INLINELOG,&log->flag)) {
1310                         if (log->size != le32_to_cpu(logsuper->size)) {
1311                                 rc = -EINVAL;
1312                                 goto errout20;
1313                         }
1314                         jfs_info("lmLogInit: inline log:0x%p base:0x%Lx size:0x%x",
1315                                  log, (unsigned long long)log->base, log->size);
1316                 } else {
1317                         if (!uuid_equal(&logsuper->uuid, &log->uuid)) {
1318                                 jfs_warn("wrong uuid on JFS log device");
1319                                 rc = -EINVAL;
1320                                 goto errout20;
1321                         }
1322                         log->size = le32_to_cpu(logsuper->size);
1323                         log->l2bsize = le32_to_cpu(logsuper->l2bsize);
1324                         jfs_info("lmLogInit: external log:0x%p base:0x%Lx size:0x%x",
1325                                  log, (unsigned long long)log->base, log->size);
1326                 }
1327
1328                 log->page = le32_to_cpu(logsuper->end) / LOGPSIZE;
1329                 log->eor = le32_to_cpu(logsuper->end) - (LOGPSIZE * log->page);
1330
1331                 /*
1332                  * initialize for log append write mode
1333                  */
1334                 /* establish current/end-of-log page/buffer */
1335                 if ((rc = lbmRead(log, log->page, &bp)))
1336                         goto errout20;
1337
1338                 lp = (struct logpage *) bp->l_ldata;
1339
1340                 jfs_info("lmLogInit: lsn:0x%x page:%d eor:%d:%d",
1341                          le32_to_cpu(logsuper->end), log->page, log->eor,
1342                          le16_to_cpu(lp->h.eor));
1343
1344                 log->bp = bp;
1345                 bp->l_pn = log->page;
1346                 bp->l_eor = log->eor;
1347
1348                 /* if current page is full, move on to next page */
1349                 if (log->eor >= LOGPSIZE - LOGPTLRSIZE)
1350                         lmNextPage(log);
1351
1352                 /*
1353                  * initialize log syncpoint
1354                  */
1355                 /*
1356                  * write the first SYNCPT record with syncpoint = 0
1357                  * (i.e., log redo up to HERE !);
1358                  * remove current page from lbm write queue at end of pageout
1359                  * (to write log superblock update), but do not release to
1360                  * freelist;
1361                  */
1362                 lrd.logtid = 0;
1363                 lrd.backchain = 0;
1364                 lrd.type = cpu_to_le16(LOG_SYNCPT);
1365                 lrd.length = 0;
1366                 lrd.log.syncpt.sync = 0;
1367                 lsn = lmWriteRecord(log, NULL, &lrd, NULL);
1368                 bp = log->bp;
1369                 bp->l_ceor = bp->l_eor;
1370                 lp = (struct logpage *) bp->l_ldata;
1371                 lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_eor);
1372                 lbmWrite(log, bp, lbmWRITE | lbmSYNC, 0);
1373                 if ((rc = lbmIOWait(bp, 0)))
1374                         goto errout30;
1375
1376                 /*
1377                  * update/write superblock
1378                  */
1379                 logsuper->state = cpu_to_le32(LOGMOUNT);
1380                 log->serial = le32_to_cpu(logsuper->serial) + 1;
1381                 logsuper->serial = cpu_to_le32(log->serial);
1382                 lbmDirectWrite(log, bpsuper, lbmWRITE | lbmRELEASE | lbmSYNC);
1383                 if ((rc = lbmIOWait(bpsuper, lbmFREE)))
1384                         goto errout30;
1385         }
1386
1387         /* initialize logsync parameters */
1388         log->logsize = (log->size - 2) << L2LOGPSIZE;
1389         log->lsn = lsn;
1390         log->syncpt = lsn;
1391         log->sync = log->syncpt;
1392         log->nextsync = LOGSYNC_DELTA(log->logsize);
1393
1394         jfs_info("lmLogInit: lsn:0x%x syncpt:0x%x sync:0x%x",
1395                  log->lsn, log->syncpt, log->sync);
1396
1397         /*
1398          * initialize for lazy/group commit
1399          */
1400         log->clsn = lsn;
1401
1402         return 0;
1403
1404         /*
1405          *      unwind on error
1406          */
1407       errout30:         /* release log page */
1408         log->wqueue = NULL;
1409         bp->l_wqnext = NULL;
1410         lbmFree(bp);
1411
1412       errout20:         /* release log superblock */
1413         lbmFree(bpsuper);
1414
1415       errout10:         /* unwind lbmLogInit() */
1416         lbmLogShutdown(log);
1417
1418         jfs_warn("lmLogInit: exit(%d)", rc);
1419         return rc;
1420 }
1421
1422
1423 /*
1424  * NAME:        lmLogClose()
1425  *
1426  * FUNCTION:    remove file system <ipmnt> from active list of log <iplog>
1427  *              and close it on last close.
1428  *
1429  * PARAMETER:   sb      - superblock
1430  *
1431  * RETURN:      errors from subroutines
1432  *
1433  * serialization:
1434  */
1435 int lmLogClose(struct super_block *sb)
1436 {
1437         struct jfs_sb_info *sbi = JFS_SBI(sb);
1438         struct jfs_log *log = sbi->log;
1439         struct file *bdev_file;
1440         int rc = 0;
1441
1442         jfs_info("lmLogClose: log:0x%p", log);
1443
1444         mutex_lock(&jfs_log_mutex);
1445         LOG_LOCK(log);
1446         list_del(&sbi->log_list);
1447         LOG_UNLOCK(log);
1448         sbi->log = NULL;
1449
1450         /*
1451          * We need to make sure all of the "written" metapages
1452          * actually make it to disk
1453          */
1454         sync_blockdev(sb->s_bdev);
1455
1456         if (test_bit(log_INLINELOG, &log->flag)) {
1457                 /*
1458                  *      in-line log in host file system
1459                  */
1460                 rc = lmLogShutdown(log);
1461                 kfree(log);
1462                 goto out;
1463         }
1464
1465         if (!log->no_integrity)
1466                 lmLogFileSystem(log, sbi, 0);
1467
1468         if (!list_empty(&log->sb_list))
1469                 goto out;
1470
1471         /*
1472          * TODO: ensure that the dummy_log is in a state to allow
1473          * lbmLogShutdown to deallocate all the buffers and call
1474          * kfree against dummy_log.  For now, leave dummy_log & its
1475          * buffers in memory, and resuse if another no-integrity mount
1476          * is requested.
1477          */
1478         if (log->no_integrity)
1479                 goto out;
1480
1481         /*
1482          *      external log as separate logical volume
1483          */
1484         list_del(&log->journal_list);
1485         bdev_file = log->bdev_file;
1486         rc = lmLogShutdown(log);
1487
1488         fput(bdev_file);
1489
1490         kfree(log);
1491
1492       out:
1493         mutex_unlock(&jfs_log_mutex);
1494         jfs_info("lmLogClose: exit(%d)", rc);
1495         return rc;
1496 }
1497
1498
1499 /*
1500  * NAME:        jfs_flush_journal()
1501  *
1502  * FUNCTION:    initiate write of any outstanding transactions to the journal
1503  *              and optionally wait until they are all written to disk
1504  *
1505  *              wait == 0  flush until latest txn is committed, don't wait
1506  *              wait == 1  flush until latest txn is committed, wait
1507  *              wait > 1   flush until all txn's are complete, wait
1508  */
1509 void jfs_flush_journal(struct jfs_log *log, int wait)
1510 {
1511         int i;
1512         struct tblock *target = NULL;
1513
1514         /* jfs_write_inode may call us during read-only mount */
1515         if (!log)
1516                 return;
1517
1518         jfs_info("jfs_flush_journal: log:0x%p wait=%d", log, wait);
1519
1520         LOGGC_LOCK(log);
1521
1522         if (!list_empty(&log->cqueue)) {
1523                 /*
1524                  * This ensures that we will keep writing to the journal as long
1525                  * as there are unwritten commit records
1526                  */
1527                 target = list_entry(log->cqueue.prev, struct tblock, cqueue);
1528
1529                 if (test_bit(log_FLUSH, &log->flag)) {
1530                         /*
1531                          * We're already flushing.
1532                          * if flush_tblk is NULL, we are flushing everything,
1533                          * so leave it that way.  Otherwise, update it to the
1534                          * latest transaction
1535                          */
1536                         if (log->flush_tblk)
1537                                 log->flush_tblk = target;
1538                 } else {
1539                         /* Only flush until latest transaction is committed */
1540                         log->flush_tblk = target;
1541                         set_bit(log_FLUSH, &log->flag);
1542
1543                         /*
1544                          * Initiate I/O on outstanding transactions
1545                          */
1546                         if (!(log->cflag & logGC_PAGEOUT)) {
1547                                 log->cflag |= logGC_PAGEOUT;
1548                                 lmGCwrite(log, 0);
1549                         }
1550                 }
1551         }
1552         if ((wait > 1) || test_bit(log_SYNCBARRIER, &log->flag)) {
1553                 /* Flush until all activity complete */
1554                 set_bit(log_FLUSH, &log->flag);
1555                 log->flush_tblk = NULL;
1556         }
1557
1558         if (wait && target && !(target->flag & tblkGC_COMMITTED)) {
1559                 DECLARE_WAITQUEUE(__wait, current);
1560
1561                 add_wait_queue(&target->gcwait, &__wait);
1562                 set_current_state(TASK_UNINTERRUPTIBLE);
1563                 LOGGC_UNLOCK(log);
1564                 schedule();
1565                 LOGGC_LOCK(log);
1566                 remove_wait_queue(&target->gcwait, &__wait);
1567         }
1568         LOGGC_UNLOCK(log);
1569
1570         if (wait < 2)
1571                 return;
1572
1573         write_special_inodes(log, filemap_fdatawrite);
1574
1575         /*
1576          * If there was recent activity, we may need to wait
1577          * for the lazycommit thread to catch up
1578          */
1579         if ((!list_empty(&log->cqueue)) || !list_empty(&log->synclist)) {
1580                 for (i = 0; i < 200; i++) {     /* Too much? */
1581                         msleep(250);
1582                         write_special_inodes(log, filemap_fdatawrite);
1583                         if (list_empty(&log->cqueue) &&
1584                             list_empty(&log->synclist))
1585                                 break;
1586                 }
1587         }
1588         assert(list_empty(&log->cqueue));
1589
1590 #ifdef CONFIG_JFS_DEBUG
1591         if (!list_empty(&log->synclist)) {
1592                 struct logsyncblk *lp;
1593
1594                 printk(KERN_ERR "jfs_flush_journal: synclist not empty\n");
1595                 list_for_each_entry(lp, &log->synclist, synclist) {
1596                         if (lp->xflag & COMMIT_PAGE) {
1597                                 struct metapage *mp = (struct metapage *)lp;
1598                                 print_hex_dump(KERN_ERR, "metapage: ",
1599                                                DUMP_PREFIX_ADDRESS, 16, 4,
1600                                                mp, sizeof(struct metapage), 0);
1601                                 print_hex_dump(KERN_ERR, "page: ",
1602                                                DUMP_PREFIX_ADDRESS, 16,
1603                                                sizeof(long), mp->page,
1604                                                sizeof(struct page), 0);
1605                         } else
1606                                 print_hex_dump(KERN_ERR, "tblock:",
1607                                                DUMP_PREFIX_ADDRESS, 16, 4,
1608                                                lp, sizeof(struct tblock), 0);
1609                 }
1610         }
1611 #else
1612         WARN_ON(!list_empty(&log->synclist));
1613 #endif
1614         clear_bit(log_FLUSH, &log->flag);
1615 }
1616
1617 /*
1618  * NAME:        lmLogShutdown()
1619  *
1620  * FUNCTION:    log shutdown at last LogClose().
1621  *
1622  *              write log syncpt record.
1623  *              update super block to set redone flag to 0.
1624  *
1625  * PARAMETER:   log     - log inode
1626  *
1627  * RETURN:      0       - success
1628  *
1629  * serialization: single last close thread
1630  */
1631 int lmLogShutdown(struct jfs_log * log)
1632 {
1633         int rc;
1634         struct lrd lrd;
1635         int lsn;
1636         struct logsuper *logsuper;
1637         struct lbuf *bpsuper;
1638         struct lbuf *bp;
1639         struct logpage *lp;
1640
1641         jfs_info("lmLogShutdown: log:0x%p", log);
1642
1643         jfs_flush_journal(log, 2);
1644
1645         /*
1646          * write the last SYNCPT record with syncpoint = 0
1647          * (i.e., log redo up to HERE !)
1648          */
1649         lrd.logtid = 0;
1650         lrd.backchain = 0;
1651         lrd.type = cpu_to_le16(LOG_SYNCPT);
1652         lrd.length = 0;
1653         lrd.log.syncpt.sync = 0;
1654
1655         lsn = lmWriteRecord(log, NULL, &lrd, NULL);
1656         bp = log->bp;
1657         lp = (struct logpage *) bp->l_ldata;
1658         lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_eor);
1659         lbmWrite(log, log->bp, lbmWRITE | lbmRELEASE | lbmSYNC, 0);
1660         lbmIOWait(log->bp, lbmFREE);
1661         log->bp = NULL;
1662
1663         /*
1664          * synchronous update log superblock
1665          * mark log state as shutdown cleanly
1666          * (i.e., Log does not need to be replayed).
1667          */
1668         if ((rc = lbmRead(log, 1, &bpsuper)))
1669                 goto out;
1670
1671         logsuper = (struct logsuper *) bpsuper->l_ldata;
1672         logsuper->state = cpu_to_le32(LOGREDONE);
1673         logsuper->end = cpu_to_le32(lsn);
1674         lbmDirectWrite(log, bpsuper, lbmWRITE | lbmRELEASE | lbmSYNC);
1675         rc = lbmIOWait(bpsuper, lbmFREE);
1676
1677         jfs_info("lmLogShutdown: lsn:0x%x page:%d eor:%d",
1678                  lsn, log->page, log->eor);
1679
1680       out:
1681         /*
1682          * shutdown per log i/o
1683          */
1684         lbmLogShutdown(log);
1685
1686         if (rc) {
1687                 jfs_warn("lmLogShutdown: exit(%d)", rc);
1688         }
1689         return rc;
1690 }
1691
1692
1693 /*
1694  * NAME:        lmLogFileSystem()
1695  *
1696  * FUNCTION:    insert (<activate> = true)/remove (<activate> = false)
1697  *      file system into/from log active file system list.
1698  *
1699  * PARAMETE:    log     - pointer to logs inode.
1700  *              fsdev   - kdev_t of filesystem.
1701  *              serial  - pointer to returned log serial number
1702  *              activate - insert/remove device from active list.
1703  *
1704  * RETURN:      0       - success
1705  *              errors returned by vms_iowait().
1706  */
1707 static int lmLogFileSystem(struct jfs_log * log, struct jfs_sb_info *sbi,
1708                            int activate)
1709 {
1710         int rc = 0;
1711         int i;
1712         struct logsuper *logsuper;
1713         struct lbuf *bpsuper;
1714         uuid_t *uuid = &sbi->uuid;
1715
1716         /*
1717          * insert/remove file system device to log active file system list.
1718          */
1719         if ((rc = lbmRead(log, 1, &bpsuper)))
1720                 return rc;
1721
1722         logsuper = (struct logsuper *) bpsuper->l_ldata;
1723         if (activate) {
1724                 for (i = 0; i < MAX_ACTIVE; i++)
1725                         if (uuid_is_null(&logsuper->active[i].uuid)) {
1726                                 uuid_copy(&logsuper->active[i].uuid, uuid);
1727                                 sbi->aggregate = i;
1728                                 break;
1729                         }
1730                 if (i == MAX_ACTIVE) {
1731                         jfs_warn("Too many file systems sharing journal!");
1732                         lbmFree(bpsuper);
1733                         return -EMFILE; /* Is there a better rc? */
1734                 }
1735         } else {
1736                 for (i = 0; i < MAX_ACTIVE; i++)
1737                         if (uuid_equal(&logsuper->active[i].uuid, uuid)) {
1738                                 uuid_copy(&logsuper->active[i].uuid,
1739                                           &uuid_null);
1740                                 break;
1741                         }
1742                 if (i == MAX_ACTIVE) {
1743                         jfs_warn("Somebody stomped on the journal!");
1744                         lbmFree(bpsuper);
1745                         return -EIO;
1746                 }
1747
1748         }
1749
1750         /*
1751          * synchronous write log superblock:
1752          *
1753          * write sidestream bypassing write queue:
1754          * at file system mount, log super block is updated for
1755          * activation of the file system before any log record
1756          * (MOUNT record) of the file system, and at file system
1757          * unmount, all meta data for the file system has been
1758          * flushed before log super block is updated for deactivation
1759          * of the file system.
1760          */
1761         lbmDirectWrite(log, bpsuper, lbmWRITE | lbmRELEASE | lbmSYNC);
1762         rc = lbmIOWait(bpsuper, lbmFREE);
1763
1764         return rc;
1765 }
1766
1767 /*
1768  *              log buffer manager (lbm)
1769  *              ------------------------
1770  *
1771  * special purpose buffer manager supporting log i/o requirements.
1772  *
1773  * per log write queue:
1774  * log pageout occurs in serial order by fifo write queue and
1775  * restricting to a single i/o in pregress at any one time.
1776  * a circular singly-linked list
1777  * (log->wrqueue points to the tail, and buffers are linked via
1778  * bp->wrqueue field), and
1779  * maintains log page in pageout ot waiting for pageout in serial pageout.
1780  */
1781
1782 /*
1783  *      lbmLogInit()
1784  *
1785  * initialize per log I/O setup at lmLogInit()
1786  */
1787 static int lbmLogInit(struct jfs_log * log)
1788 {                               /* log inode */
1789         int i;
1790         struct lbuf *lbuf;
1791
1792         jfs_info("lbmLogInit: log:0x%p", log);
1793
1794         /* initialize current buffer cursor */
1795         log->bp = NULL;
1796
1797         /* initialize log device write queue */
1798         log->wqueue = NULL;
1799
1800         /*
1801          * Each log has its own buffer pages allocated to it.  These are
1802          * not managed by the page cache.  This ensures that a transaction
1803          * writing to the log does not block trying to allocate a page from
1804          * the page cache (for the log).  This would be bad, since page
1805          * allocation waits on the kswapd thread that may be committing inodes
1806          * which would cause log activity.  Was that clear?  I'm trying to
1807          * avoid deadlock here.
1808          */
1809         init_waitqueue_head(&log->free_wait);
1810
1811         log->lbuf_free = NULL;
1812
1813         for (i = 0; i < LOGPAGES;) {
1814                 char *buffer;
1815                 uint offset;
1816                 struct page *page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1817
1818                 if (!page)
1819                         goto error;
1820                 buffer = page_address(page);
1821                 for (offset = 0; offset < PAGE_SIZE; offset += LOGPSIZE) {
1822                         lbuf = kmalloc(sizeof(struct lbuf), GFP_KERNEL);
1823                         if (lbuf == NULL) {
1824                                 if (offset == 0)
1825                                         __free_page(page);
1826                                 goto error;
1827                         }
1828                         if (offset) /* we already have one reference */
1829                                 get_page(page);
1830                         lbuf->l_offset = offset;
1831                         lbuf->l_ldata = buffer + offset;
1832                         lbuf->l_page = page;
1833                         lbuf->l_log = log;
1834                         init_waitqueue_head(&lbuf->l_ioevent);
1835
1836                         lbuf->l_freelist = log->lbuf_free;
1837                         log->lbuf_free = lbuf;
1838                         i++;
1839                 }
1840         }
1841
1842         return (0);
1843
1844       error:
1845         lbmLogShutdown(log);
1846         return -ENOMEM;
1847 }
1848
1849
1850 /*
1851  *      lbmLogShutdown()
1852  *
1853  * finalize per log I/O setup at lmLogShutdown()
1854  */
1855 static void lbmLogShutdown(struct jfs_log * log)
1856 {
1857         struct lbuf *lbuf;
1858
1859         jfs_info("lbmLogShutdown: log:0x%p", log);
1860
1861         lbuf = log->lbuf_free;
1862         while (lbuf) {
1863                 struct lbuf *next = lbuf->l_freelist;
1864                 __free_page(lbuf->l_page);
1865                 kfree(lbuf);
1866                 lbuf = next;
1867         }
1868 }
1869
1870
1871 /*
1872  *      lbmAllocate()
1873  *
1874  * allocate an empty log buffer
1875  */
1876 static struct lbuf *lbmAllocate(struct jfs_log * log, int pn)
1877 {
1878         struct lbuf *bp;
1879         unsigned long flags;
1880
1881         /*
1882          * recycle from log buffer freelist if any
1883          */
1884         LCACHE_LOCK(flags);
1885         LCACHE_SLEEP_COND(log->free_wait, (bp = log->lbuf_free), flags);
1886         log->lbuf_free = bp->l_freelist;
1887         LCACHE_UNLOCK(flags);
1888
1889         bp->l_flag = 0;
1890
1891         bp->l_wqnext = NULL;
1892         bp->l_freelist = NULL;
1893
1894         bp->l_pn = pn;
1895         bp->l_blkno = log->base + (pn << (L2LOGPSIZE - log->l2bsize));
1896         bp->l_ceor = 0;
1897
1898         return bp;
1899 }
1900
1901
1902 /*
1903  *      lbmFree()
1904  *
1905  * release a log buffer to freelist
1906  */
1907 static void lbmFree(struct lbuf * bp)
1908 {
1909         unsigned long flags;
1910
1911         LCACHE_LOCK(flags);
1912
1913         lbmfree(bp);
1914
1915         LCACHE_UNLOCK(flags);
1916 }
1917
1918 static void lbmfree(struct lbuf * bp)
1919 {
1920         struct jfs_log *log = bp->l_log;
1921
1922         assert(bp->l_wqnext == NULL);
1923
1924         /*
1925          * return the buffer to head of freelist
1926          */
1927         bp->l_freelist = log->lbuf_free;
1928         log->lbuf_free = bp;
1929
1930         wake_up(&log->free_wait);
1931         return;
1932 }
1933
1934
1935 /*
1936  * NAME:        lbmRedrive
1937  *
1938  * FUNCTION:    add a log buffer to the log redrive list
1939  *
1940  * PARAMETER:
1941  *      bp      - log buffer
1942  *
1943  * NOTES:
1944  *      Takes log_redrive_lock.
1945  */
1946 static inline void lbmRedrive(struct lbuf *bp)
1947 {
1948         unsigned long flags;
1949
1950         spin_lock_irqsave(&log_redrive_lock, flags);
1951         bp->l_redrive_next = log_redrive_list;
1952         log_redrive_list = bp;
1953         spin_unlock_irqrestore(&log_redrive_lock, flags);
1954
1955         wake_up_process(jfsIOthread);
1956 }
1957
1958
1959 /*
1960  *      lbmRead()
1961  */
1962 static int lbmRead(struct jfs_log * log, int pn, struct lbuf ** bpp)
1963 {
1964         struct bio *bio;
1965         struct lbuf *bp;
1966
1967         /*
1968          * allocate a log buffer
1969          */
1970         *bpp = bp = lbmAllocate(log, pn);
1971         jfs_info("lbmRead: bp:0x%p pn:0x%x", bp, pn);
1972
1973         bp->l_flag |= lbmREAD;
1974
1975         bio = bio_alloc(file_bdev(log->bdev_file), 1, REQ_OP_READ, GFP_NOFS);
1976         bio->bi_iter.bi_sector = bp->l_blkno << (log->l2bsize - 9);
1977         __bio_add_page(bio, bp->l_page, LOGPSIZE, bp->l_offset);
1978         BUG_ON(bio->bi_iter.bi_size != LOGPSIZE);
1979
1980         bio->bi_end_io = lbmIODone;
1981         bio->bi_private = bp;
1982         /*check if journaling to disk has been disabled*/
1983         if (log->no_integrity) {
1984                 bio->bi_iter.bi_size = 0;
1985                 lbmIODone(bio);
1986         } else {
1987                 submit_bio(bio);
1988         }
1989
1990         wait_event(bp->l_ioevent, (bp->l_flag != lbmREAD));
1991
1992         return 0;
1993 }
1994
1995
1996 /*
1997  *      lbmWrite()
1998  *
1999  * buffer at head of pageout queue stays after completion of
2000  * partial-page pageout and redriven by explicit initiation of
2001  * pageout by caller until full-page pageout is completed and
2002  * released.
2003  *
2004  * device driver i/o done redrives pageout of new buffer at
2005  * head of pageout queue when current buffer at head of pageout
2006  * queue is released at the completion of its full-page pageout.
2007  *
2008  * LOGGC_LOCK() serializes lbmWrite() by lmNextPage() and lmGroupCommit().
2009  * LCACHE_LOCK() serializes xflag between lbmWrite() and lbmIODone()
2010  */
2011 static void lbmWrite(struct jfs_log * log, struct lbuf * bp, int flag,
2012                      int cant_block)
2013 {
2014         struct lbuf *tail;
2015         unsigned long flags;
2016
2017         jfs_info("lbmWrite: bp:0x%p flag:0x%x pn:0x%x", bp, flag, bp->l_pn);
2018
2019         /* map the logical block address to physical block address */
2020         bp->l_blkno =
2021             log->base + (bp->l_pn << (L2LOGPSIZE - log->l2bsize));
2022
2023         LCACHE_LOCK(flags);             /* disable+lock */
2024
2025         /*
2026          * initialize buffer for device driver
2027          */
2028         bp->l_flag = flag;
2029
2030         /*
2031          *      insert bp at tail of write queue associated with log
2032          *
2033          * (request is either for bp already/currently at head of queue
2034          * or new bp to be inserted at tail)
2035          */
2036         tail = log->wqueue;
2037
2038         /* is buffer not already on write queue ? */
2039         if (bp->l_wqnext == NULL) {
2040                 /* insert at tail of wqueue */
2041                 if (tail == NULL) {
2042                         log->wqueue = bp;
2043                         bp->l_wqnext = bp;
2044                 } else {
2045                         log->wqueue = bp;
2046                         bp->l_wqnext = tail->l_wqnext;
2047                         tail->l_wqnext = bp;
2048                 }
2049
2050                 tail = bp;
2051         }
2052
2053         /* is buffer at head of wqueue and for write ? */
2054         if ((bp != tail->l_wqnext) || !(flag & lbmWRITE)) {
2055                 LCACHE_UNLOCK(flags);   /* unlock+enable */
2056                 return;
2057         }
2058
2059         LCACHE_UNLOCK(flags);   /* unlock+enable */
2060
2061         if (cant_block)
2062                 lbmRedrive(bp);
2063         else if (flag & lbmSYNC)
2064                 lbmStartIO(bp);
2065         else {
2066                 LOGGC_UNLOCK(log);
2067                 lbmStartIO(bp);
2068                 LOGGC_LOCK(log);
2069         }
2070 }
2071
2072
2073 /*
2074  *      lbmDirectWrite()
2075  *
2076  * initiate pageout bypassing write queue for sidestream
2077  * (e.g., log superblock) write;
2078  */
2079 static void lbmDirectWrite(struct jfs_log * log, struct lbuf * bp, int flag)
2080 {
2081         jfs_info("lbmDirectWrite: bp:0x%p flag:0x%x pn:0x%x",
2082                  bp, flag, bp->l_pn);
2083
2084         /*
2085          * initialize buffer for device driver
2086          */
2087         bp->l_flag = flag | lbmDIRECT;
2088
2089         /* map the logical block address to physical block address */
2090         bp->l_blkno =
2091             log->base + (bp->l_pn << (L2LOGPSIZE - log->l2bsize));
2092
2093         /*
2094          *      initiate pageout of the page
2095          */
2096         lbmStartIO(bp);
2097 }
2098
2099
2100 /*
2101  * NAME:        lbmStartIO()
2102  *
2103  * FUNCTION:    Interface to DD strategy routine
2104  *
2105  * RETURN:      none
2106  *
2107  * serialization: LCACHE_LOCK() is NOT held during log i/o;
2108  */
2109 static void lbmStartIO(struct lbuf * bp)
2110 {
2111         struct bio *bio;
2112         struct jfs_log *log = bp->l_log;
2113         struct block_device *bdev = NULL;
2114
2115         jfs_info("lbmStartIO");
2116
2117         if (!log->no_integrity)
2118                 bdev = file_bdev(log->bdev_file);
2119
2120         bio = bio_alloc(bdev, 1, REQ_OP_WRITE | REQ_SYNC,
2121                         GFP_NOFS);
2122         bio->bi_iter.bi_sector = bp->l_blkno << (log->l2bsize - 9);
2123         __bio_add_page(bio, bp->l_page, LOGPSIZE, bp->l_offset);
2124         BUG_ON(bio->bi_iter.bi_size != LOGPSIZE);
2125
2126         bio->bi_end_io = lbmIODone;
2127         bio->bi_private = bp;
2128
2129         /* check if journaling to disk has been disabled */
2130         if (log->no_integrity) {
2131                 bio->bi_iter.bi_size = 0;
2132                 lbmIODone(bio);
2133         } else {
2134                 submit_bio(bio);
2135                 INCREMENT(lmStat.submitted);
2136         }
2137 }
2138
2139
2140 /*
2141  *      lbmIOWait()
2142  */
2143 static int lbmIOWait(struct lbuf * bp, int flag)
2144 {
2145         unsigned long flags;
2146         int rc = 0;
2147
2148         jfs_info("lbmIOWait1: bp:0x%p flag:0x%x:0x%x", bp, bp->l_flag, flag);
2149
2150         LCACHE_LOCK(flags);             /* disable+lock */
2151
2152         LCACHE_SLEEP_COND(bp->l_ioevent, (bp->l_flag & lbmDONE), flags);
2153
2154         rc = (bp->l_flag & lbmERROR) ? -EIO : 0;
2155
2156         if (flag & lbmFREE)
2157                 lbmfree(bp);
2158
2159         LCACHE_UNLOCK(flags);   /* unlock+enable */
2160
2161         jfs_info("lbmIOWait2: bp:0x%p flag:0x%x:0x%x", bp, bp->l_flag, flag);
2162         return rc;
2163 }
2164
2165 /*
2166  *      lbmIODone()
2167  *
2168  * executed at INTIODONE level
2169  */
2170 static void lbmIODone(struct bio *bio)
2171 {
2172         struct lbuf *bp = bio->bi_private;
2173         struct lbuf *nextbp, *tail;
2174         struct jfs_log *log;
2175         unsigned long flags;
2176
2177         /*
2178          * get back jfs buffer bound to the i/o buffer
2179          */
2180         jfs_info("lbmIODone: bp:0x%p flag:0x%x", bp, bp->l_flag);
2181
2182         LCACHE_LOCK(flags);             /* disable+lock */
2183
2184         bp->l_flag |= lbmDONE;
2185
2186         if (bio->bi_status) {
2187                 bp->l_flag |= lbmERROR;
2188
2189                 jfs_err("lbmIODone: I/O error in JFS log");
2190         }
2191
2192         bio_put(bio);
2193
2194         /*
2195          *      pagein completion
2196          */
2197         if (bp->l_flag & lbmREAD) {
2198                 bp->l_flag &= ~lbmREAD;
2199
2200                 LCACHE_UNLOCK(flags);   /* unlock+enable */
2201
2202                 /* wakeup I/O initiator */
2203                 LCACHE_WAKEUP(&bp->l_ioevent);
2204
2205                 return;
2206         }
2207
2208         /*
2209          *      pageout completion
2210          *
2211          * the bp at the head of write queue has completed pageout.
2212          *
2213          * if single-commit/full-page pageout, remove the current buffer
2214          * from head of pageout queue, and redrive pageout with
2215          * the new buffer at head of pageout queue;
2216          * otherwise, the partial-page pageout buffer stays at
2217          * the head of pageout queue to be redriven for pageout
2218          * by lmGroupCommit() until full-page pageout is completed.
2219          */
2220         bp->l_flag &= ~lbmWRITE;
2221         INCREMENT(lmStat.pagedone);
2222
2223         /* update committed lsn */
2224         log = bp->l_log;
2225         log->clsn = (bp->l_pn << L2LOGPSIZE) + bp->l_ceor;
2226
2227         if (bp->l_flag & lbmDIRECT) {
2228                 LCACHE_WAKEUP(&bp->l_ioevent);
2229                 LCACHE_UNLOCK(flags);
2230                 return;
2231         }
2232
2233         tail = log->wqueue;
2234
2235         /* single element queue */
2236         if (bp == tail) {
2237                 /* remove head buffer of full-page pageout
2238                  * from log device write queue
2239                  */
2240                 if (bp->l_flag & lbmRELEASE) {
2241                         log->wqueue = NULL;
2242                         bp->l_wqnext = NULL;
2243                 }
2244         }
2245         /* multi element queue */
2246         else {
2247                 /* remove head buffer of full-page pageout
2248                  * from log device write queue
2249                  */
2250                 if (bp->l_flag & lbmRELEASE) {
2251                         nextbp = tail->l_wqnext = bp->l_wqnext;
2252                         bp->l_wqnext = NULL;
2253
2254                         /*
2255                          * redrive pageout of next page at head of write queue:
2256                          * redrive next page without any bound tblk
2257                          * (i.e., page w/o any COMMIT records), or
2258                          * first page of new group commit which has been
2259                          * queued after current page (subsequent pageout
2260                          * is performed synchronously, except page without
2261                          * any COMMITs) by lmGroupCommit() as indicated
2262                          * by lbmWRITE flag;
2263                          */
2264                         if (nextbp->l_flag & lbmWRITE) {
2265                                 /*
2266                                  * We can't do the I/O at interrupt time.
2267                                  * The jfsIO thread can do it
2268                                  */
2269                                 lbmRedrive(nextbp);
2270                         }
2271                 }
2272         }
2273
2274         /*
2275          *      synchronous pageout:
2276          *
2277          * buffer has not necessarily been removed from write queue
2278          * (e.g., synchronous write of partial-page with COMMIT):
2279          * leave buffer for i/o initiator to dispose
2280          */
2281         if (bp->l_flag & lbmSYNC) {
2282                 LCACHE_UNLOCK(flags);   /* unlock+enable */
2283
2284                 /* wakeup I/O initiator */
2285                 LCACHE_WAKEUP(&bp->l_ioevent);
2286         }
2287
2288         /*
2289          *      Group Commit pageout:
2290          */
2291         else if (bp->l_flag & lbmGC) {
2292                 LCACHE_UNLOCK(flags);
2293                 lmPostGC(bp);
2294         }
2295
2296         /*
2297          *      asynchronous pageout:
2298          *
2299          * buffer must have been removed from write queue:
2300          * insert buffer at head of freelist where it can be recycled
2301          */
2302         else {
2303                 assert(bp->l_flag & lbmRELEASE);
2304                 assert(bp->l_flag & lbmFREE);
2305                 lbmfree(bp);
2306
2307                 LCACHE_UNLOCK(flags);   /* unlock+enable */
2308         }
2309 }
2310
2311 int jfsIOWait(void *arg)
2312 {
2313         struct lbuf *bp;
2314
2315         do {
2316                 spin_lock_irq(&log_redrive_lock);
2317                 while ((bp = log_redrive_list)) {
2318                         log_redrive_list = bp->l_redrive_next;
2319                         bp->l_redrive_next = NULL;
2320                         spin_unlock_irq(&log_redrive_lock);
2321                         lbmStartIO(bp);
2322                         spin_lock_irq(&log_redrive_lock);
2323                 }
2324
2325                 if (freezing(current)) {
2326                         spin_unlock_irq(&log_redrive_lock);
2327                         try_to_freeze();
2328                 } else {
2329                         set_current_state(TASK_INTERRUPTIBLE);
2330                         spin_unlock_irq(&log_redrive_lock);
2331                         schedule();
2332                 }
2333         } while (!kthread_should_stop());
2334
2335         jfs_info("jfsIOWait being killed!");
2336         return 0;
2337 }
2338
2339 /*
2340  * NAME:        lmLogFormat()/jfs_logform()
2341  *
2342  * FUNCTION:    format file system log
2343  *
2344  * PARAMETERS:
2345  *      log     - volume log
2346  *      logAddress - start address of log space in FS block
2347  *      logSize - length of log space in FS block;
2348  *
2349  * RETURN:      0       - success
2350  *              -EIO    - i/o error
2351  *
2352  * XXX: We're synchronously writing one page at a time.  This needs to
2353  *      be improved by writing multiple pages at once.
2354  */
2355 int lmLogFormat(struct jfs_log *log, s64 logAddress, int logSize)
2356 {
2357         int rc = -EIO;
2358         struct jfs_sb_info *sbi;
2359         struct logsuper *logsuper;
2360         struct logpage *lp;
2361         int lspn;               /* log sequence page number */
2362         struct lrd *lrd_ptr;
2363         int npages = 0;
2364         struct lbuf *bp;
2365
2366         jfs_info("lmLogFormat: logAddress:%Ld logSize:%d",
2367                  (long long)logAddress, logSize);
2368
2369         sbi = list_entry(log->sb_list.next, struct jfs_sb_info, log_list);
2370
2371         /* allocate a log buffer */
2372         bp = lbmAllocate(log, 1);
2373
2374         npages = logSize >> sbi->l2nbperpage;
2375
2376         /*
2377          *      log space:
2378          *
2379          * page 0 - reserved;
2380          * page 1 - log superblock;
2381          * page 2 - log data page: A SYNC log record is written
2382          *          into this page at logform time;
2383          * pages 3-N - log data page: set to empty log data pages;
2384          */
2385         /*
2386          *      init log superblock: log page 1
2387          */
2388         logsuper = (struct logsuper *) bp->l_ldata;
2389
2390         logsuper->magic = cpu_to_le32(LOGMAGIC);
2391         logsuper->version = cpu_to_le32(LOGVERSION);
2392         logsuper->state = cpu_to_le32(LOGREDONE);
2393         logsuper->flag = cpu_to_le32(sbi->mntflag);     /* ? */
2394         logsuper->size = cpu_to_le32(npages);
2395         logsuper->bsize = cpu_to_le32(sbi->bsize);
2396         logsuper->l2bsize = cpu_to_le32(sbi->l2bsize);
2397         logsuper->end = cpu_to_le32(2 * LOGPSIZE + LOGPHDRSIZE + LOGRDSIZE);
2398
2399         bp->l_flag = lbmWRITE | lbmSYNC | lbmDIRECT;
2400         bp->l_blkno = logAddress + sbi->nbperpage;
2401         lbmStartIO(bp);
2402         if ((rc = lbmIOWait(bp, 0)))
2403                 goto exit;
2404
2405         /*
2406          *      init pages 2 to npages-1 as log data pages:
2407          *
2408          * log page sequence number (lpsn) initialization:
2409          *
2410          * pn:   0     1     2     3                 n-1
2411          *       +-----+-----+=====+=====+===.....===+=====+
2412          * lspn:             N-1   0     1           N-2
2413          *                   <--- N page circular file ---->
2414          *
2415          * the N (= npages-2) data pages of the log is maintained as
2416          * a circular file for the log records;
2417          * lpsn grows by 1 monotonically as each log page is written
2418          * to the circular file of the log;
2419          * and setLogpage() will not reset the page number even if
2420          * the eor is equal to LOGPHDRSIZE. In order for binary search
2421          * still work in find log end process, we have to simulate the
2422          * log wrap situation at the log format time.
2423          * The 1st log page written will have the highest lpsn. Then
2424          * the succeeding log pages will have ascending order of
2425          * the lspn starting from 0, ... (N-2)
2426          */
2427         lp = (struct logpage *) bp->l_ldata;
2428         /*
2429          * initialize 1st log page to be written: lpsn = N - 1,
2430          * write a SYNCPT log record is written to this page
2431          */
2432         lp->h.page = lp->t.page = cpu_to_le32(npages - 3);
2433         lp->h.eor = lp->t.eor = cpu_to_le16(LOGPHDRSIZE + LOGRDSIZE);
2434
2435         lrd_ptr = (struct lrd *) &lp->data;
2436         lrd_ptr->logtid = 0;
2437         lrd_ptr->backchain = 0;
2438         lrd_ptr->type = cpu_to_le16(LOG_SYNCPT);
2439         lrd_ptr->length = 0;
2440         lrd_ptr->log.syncpt.sync = 0;
2441
2442         bp->l_blkno += sbi->nbperpage;
2443         bp->l_flag = lbmWRITE | lbmSYNC | lbmDIRECT;
2444         lbmStartIO(bp);
2445         if ((rc = lbmIOWait(bp, 0)))
2446                 goto exit;
2447
2448         /*
2449          *      initialize succeeding log pages: lpsn = 0, 1, ..., (N-2)
2450          */
2451         for (lspn = 0; lspn < npages - 3; lspn++) {
2452                 lp->h.page = lp->t.page = cpu_to_le32(lspn);
2453                 lp->h.eor = lp->t.eor = cpu_to_le16(LOGPHDRSIZE);
2454
2455                 bp->l_blkno += sbi->nbperpage;
2456                 bp->l_flag = lbmWRITE | lbmSYNC | lbmDIRECT;
2457                 lbmStartIO(bp);
2458                 if ((rc = lbmIOWait(bp, 0)))
2459                         goto exit;
2460         }
2461
2462         rc = 0;
2463 exit:
2464         /*
2465          *      finalize log
2466          */
2467         /* release the buffer */
2468         lbmFree(bp);
2469
2470         return rc;
2471 }
2472
2473 #ifdef CONFIG_JFS_STATISTICS
2474 int jfs_lmstats_proc_show(struct seq_file *m, void *v)
2475 {
2476         seq_printf(m,
2477                        "JFS Logmgr stats\n"
2478                        "================\n"
2479                        "commits = %d\n"
2480                        "writes submitted = %d\n"
2481                        "writes completed = %d\n"
2482                        "full pages submitted = %d\n"
2483                        "partial pages submitted = %d\n",
2484                        lmStat.commit,
2485                        lmStat.submitted,
2486                        lmStat.pagedone,
2487                        lmStat.full_page,
2488                        lmStat.partial_page);
2489         return 0;
2490 }
2491 #endif /* CONFIG_JFS_STATISTICS */