fs/jfs/jfs_logmgr.c

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