JFS: semaphore to mutex conversion.
[linux-2.6.git] / 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 <linux/delay.h>
71 #include "jfs_incore.h"
72 #include "jfs_filsys.h"
73 #include "jfs_metapage.h"
74 #include "jfs_superblock.h"
75 #include "jfs_txnmgr.h"
76 #include "jfs_debug.h"
77
78
79 /*
80  * lbuf's ready to be redriven.  Protected by log_redrive_lock (jfsIO thread)
81  */
82 static struct lbuf *log_redrive_list;
83 static DEFINE_SPINLOCK(log_redrive_lock);
84 DECLARE_WAIT_QUEUE_HEAD(jfs_IO_thread_wait);
85
86
87 /*
88  *      log read/write serialization (per log)
89  */
90 #define LOG_LOCK_INIT(log)      mutex_init(&(log)->loglock)
91 #define LOG_LOCK(log)           mutex_lock(&((log)->loglock))
92 #define LOG_UNLOCK(log)         mutex_unlock(&((log)->loglock))
93
94
95 /*
96  *      log group commit serialization (per log)
97  */
98
99 #define LOGGC_LOCK_INIT(log)    spin_lock_init(&(log)->gclock)
100 #define LOGGC_LOCK(log)         spin_lock_irq(&(log)->gclock)
101 #define LOGGC_UNLOCK(log)       spin_unlock_irq(&(log)->gclock)
102 #define LOGGC_WAKEUP(tblk)      wake_up_all(&(tblk)->gcwait)
103
104 /*
105  *      log sync serialization (per log)
106  */
107 #define LOGSYNC_DELTA(logsize)          min((logsize)/8, 128*LOGPSIZE)
108 #define LOGSYNC_BARRIER(logsize)        ((logsize)/4)
109 /*
110 #define LOGSYNC_DELTA(logsize)          min((logsize)/4, 256*LOGPSIZE)
111 #define LOGSYNC_BARRIER(logsize)        ((logsize)/2)
112 */
113
114
115 /*
116  *      log buffer cache synchronization
117  */
118 static DEFINE_SPINLOCK(jfsLCacheLock);
119
120 #define LCACHE_LOCK(flags)      spin_lock_irqsave(&jfsLCacheLock, flags)
121 #define LCACHE_UNLOCK(flags)    spin_unlock_irqrestore(&jfsLCacheLock, flags)
122
123 /*
124  * See __SLEEP_COND in jfs_locks.h
125  */
126 #define LCACHE_SLEEP_COND(wq, cond, flags)      \
127 do {                                            \
128         if (cond)                               \
129                 break;                          \
130         __SLEEP_COND(wq, cond, LCACHE_LOCK(flags), LCACHE_UNLOCK(flags)); \
131 } while (0)
132
133 #define LCACHE_WAKEUP(event)    wake_up(event)
134
135
136 /*
137  *      lbuf buffer cache (lCache) control
138  */
139 /* log buffer manager pageout control (cumulative, inclusive) */
140 #define lbmREAD         0x0001
141 #define lbmWRITE        0x0002  /* enqueue at tail of write queue;
142                                  * init pageout if at head of queue;
143                                  */
144 #define lbmRELEASE      0x0004  /* remove from write queue
145                                  * at completion of pageout;
146                                  * do not free/recycle it yet:
147                                  * caller will free it;
148                                  */
149 #define lbmSYNC         0x0008  /* do not return to freelist
150                                  * when removed from write queue;
151                                  */
152 #define lbmFREE         0x0010  /* return to freelist
153                                  * at completion of pageout;
154                                  * the buffer may be recycled;
155                                  */
156 #define lbmDONE         0x0020
157 #define lbmERROR        0x0040
158 #define lbmGC           0x0080  /* lbmIODone to perform post-GC processing
159                                  * of log page
160                                  */
161 #define lbmDIRECT       0x0100
162
163 /*
164  * Global list of active external journals
165  */
166 static LIST_HEAD(jfs_external_logs);
167 static struct jfs_log *dummy_log = NULL;
168 static DECLARE_MUTEX(jfs_log_sem);
169
170 /*
171  * forward references
172  */
173 static int lmWriteRecord(struct jfs_log * log, struct tblock * tblk,
174                          struct lrd * lrd, struct tlock * tlck);
175
176 static int lmNextPage(struct jfs_log * log);
177 static int lmLogFileSystem(struct jfs_log * log, struct jfs_sb_info *sbi,
178                            int activate);
179
180 static int open_inline_log(struct super_block *sb);
181 static int open_dummy_log(struct super_block *sb);
182 static int lbmLogInit(struct jfs_log * log);
183 static void lbmLogShutdown(struct jfs_log * log);
184 static struct lbuf *lbmAllocate(struct jfs_log * log, int);
185 static void lbmFree(struct lbuf * bp);
186 static void lbmfree(struct lbuf * bp);
187 static int lbmRead(struct jfs_log * log, int pn, struct lbuf ** bpp);
188 static void lbmWrite(struct jfs_log * log, struct lbuf * bp, int flag, int cant_block);
189 static void lbmDirectWrite(struct jfs_log * log, struct lbuf * bp, int flag);
190 static int lbmIOWait(struct lbuf * bp, int flag);
191 static bio_end_io_t lbmIODone;
192 static void lbmStartIO(struct lbuf * bp);
193 static void lmGCwrite(struct jfs_log * log, int cant_block);
194 static int lmLogSync(struct jfs_log * log, int hard_sync);
195
196
197
198 /*
199  *      statistics
200  */
201 #ifdef CONFIG_JFS_STATISTICS
202 static struct lmStat {
203         uint commit;            /* # of commit */
204         uint pagedone;          /* # of page written */
205         uint submitted;         /* # of pages submitted */
206         uint full_page;         /* # of full pages submitted */
207         uint partial_page;      /* # of partial pages submitted */
208 } lmStat;
209 #endif
210
211
212 /*
213  * NAME:        lmLog()
214  *
215  * FUNCTION:    write a log record;
216  *
217  * PARAMETER:
218  *
219  * RETURN:      lsn - offset to the next log record to write (end-of-log);
220  *              -1  - error;
221  *
222  * note: todo: log error handler
223  */
224 int lmLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
225           struct tlock * tlck)
226 {
227         int lsn;
228         int diffp, difft;
229         struct metapage *mp = NULL;
230         unsigned long flags;
231
232         jfs_info("lmLog: log:0x%p tblk:0x%p, lrd:0x%p tlck:0x%p",
233                  log, tblk, lrd, tlck);
234
235         LOG_LOCK(log);
236
237         /* log by (out-of-transaction) JFS ? */
238         if (tblk == NULL)
239                 goto writeRecord;
240
241         /* log from page ? */
242         if (tlck == NULL ||
243             tlck->type & tlckBTROOT || (mp = tlck->mp) == NULL)
244                 goto writeRecord;
245
246         /*
247          *      initialize/update page/transaction recovery lsn
248          */
249         lsn = log->lsn;
250
251         LOGSYNC_LOCK(log, flags);
252
253         /*
254          * initialize page lsn if first log write of the page
255          */
256         if (mp->lsn == 0) {
257                 mp->log = log;
258                 mp->lsn = lsn;
259                 log->count++;
260
261                 /* insert page at tail of logsynclist */
262                 list_add_tail(&mp->synclist, &log->synclist);
263         }
264
265         /*
266          *      initialize/update lsn of tblock of the page
267          *
268          * transaction inherits oldest lsn of pages associated
269          * with allocation/deallocation of resources (their
270          * log records are used to reconstruct allocation map
271          * at recovery time: inode for inode allocation map,
272          * B+-tree index of extent descriptors for block
273          * allocation map);
274          * allocation map pages inherit transaction lsn at
275          * commit time to allow forwarding log syncpt past log
276          * records associated with allocation/deallocation of
277          * resources only after persistent map of these map pages
278          * have been updated and propagated to home.
279          */
280         /*
281          * initialize transaction lsn:
282          */
283         if (tblk->lsn == 0) {
284                 /* inherit lsn of its first page logged */
285                 tblk->lsn = mp->lsn;
286                 log->count++;
287
288                 /* insert tblock after the page on logsynclist */
289                 list_add(&tblk->synclist, &mp->synclist);
290         }
291         /*
292          * update transaction lsn:
293          */
294         else {
295                 /* inherit oldest/smallest lsn of page */
296                 logdiff(diffp, mp->lsn, log);
297                 logdiff(difft, tblk->lsn, log);
298                 if (diffp < difft) {
299                         /* update tblock lsn with page lsn */
300                         tblk->lsn = mp->lsn;
301
302                         /* move tblock after page on logsynclist */
303                         list_move(&tblk->synclist, &mp->synclist);
304                 }
305         }
306
307         LOGSYNC_UNLOCK(log, flags);
308
309         /*
310          *      write the log record
311          */
312       writeRecord:
313         lsn = lmWriteRecord(log, tblk, lrd, tlck);
314
315         /*
316          * forward log syncpt if log reached next syncpt trigger
317          */
318         logdiff(diffp, lsn, log);
319         if (diffp >= log->nextsync)
320                 lsn = lmLogSync(log, 0);
321
322         /* update end-of-log lsn */
323         log->lsn = lsn;
324
325         LOG_UNLOCK(log);
326
327         /* return end-of-log address */
328         return lsn;
329 }
330
331 /*
332  * NAME:        lmWriteRecord()
333  *
334  * FUNCTION:    move the log record to current log page
335  *
336  * PARAMETER:   cd      - commit descriptor
337  *
338  * RETURN:      end-of-log address
339  *                      
340  * serialization: LOG_LOCK() held on entry/exit
341  */
342 static int
343 lmWriteRecord(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
344               struct tlock * tlck)
345 {
346         int lsn = 0;            /* end-of-log address */
347         struct lbuf *bp;        /* dst log page buffer */
348         struct logpage *lp;     /* dst log page */
349         caddr_t dst;            /* destination address in log page */
350         int dstoffset;          /* end-of-log offset in log page */
351         int freespace;          /* free space in log page */
352         caddr_t p;              /* src meta-data page */
353         caddr_t src;
354         int srclen;
355         int nbytes;             /* number of bytes to move */
356         int i;
357         int len;
358         struct linelock *linelock;
359         struct lv *lv;
360         struct lvd *lvd;
361         int l2linesize;
362
363         len = 0;
364
365         /* retrieve destination log page to write */
366         bp = (struct lbuf *) log->bp;
367         lp = (struct logpage *) bp->l_ldata;
368         dstoffset = log->eor;
369
370         /* any log data to write ? */
371         if (tlck == NULL)
372                 goto moveLrd;
373
374         /*
375          *      move log record data
376          */
377         /* retrieve source meta-data page to log */
378         if (tlck->flag & tlckPAGELOCK) {
379                 p = (caddr_t) (tlck->mp->data);
380                 linelock = (struct linelock *) & tlck->lock;
381         }
382         /* retrieve source in-memory inode to log */
383         else if (tlck->flag & tlckINODELOCK) {
384                 if (tlck->type & tlckDTREE)
385                         p = (caddr_t) &JFS_IP(tlck->ip)->i_dtroot;
386                 else
387                         p = (caddr_t) &JFS_IP(tlck->ip)->i_xtroot;
388                 linelock = (struct linelock *) & tlck->lock;
389         }
390 #ifdef  _JFS_WIP
391         else if (tlck->flag & tlckINLINELOCK) {
392
393                 inlinelock = (struct inlinelock *) & tlck;
394                 p = (caddr_t) & inlinelock->pxd;
395                 linelock = (struct linelock *) & tlck;
396         }
397 #endif                          /* _JFS_WIP */
398         else {
399                 jfs_err("lmWriteRecord: UFO tlck:0x%p", tlck);
400                 return 0;       /* Probably should trap */
401         }
402         l2linesize = linelock->l2linesize;
403
404       moveData:
405         ASSERT(linelock->index <= linelock->maxcnt);
406
407         lv = linelock->lv;
408         for (i = 0; i < linelock->index; i++, lv++) {
409                 if (lv->length == 0)
410                         continue;
411
412                 /* is page full ? */
413                 if (dstoffset >= LOGPSIZE - LOGPTLRSIZE) {
414                         /* page become full: move on to next page */
415                         lmNextPage(log);
416
417                         bp = log->bp;
418                         lp = (struct logpage *) bp->l_ldata;
419                         dstoffset = LOGPHDRSIZE;
420                 }
421
422                 /*
423                  * move log vector data
424                  */
425                 src = (u8 *) p + (lv->offset << l2linesize);
426                 srclen = lv->length << l2linesize;
427                 len += srclen;
428                 while (srclen > 0) {
429                         freespace = (LOGPSIZE - LOGPTLRSIZE) - dstoffset;
430                         nbytes = min(freespace, srclen);
431                         dst = (caddr_t) lp + dstoffset;
432                         memcpy(dst, src, nbytes);
433                         dstoffset += nbytes;
434
435                         /* is page not full ? */
436                         if (dstoffset < LOGPSIZE - LOGPTLRSIZE)
437                                 break;
438
439                         /* page become full: move on to next page */
440                         lmNextPage(log);
441
442                         bp = (struct lbuf *) log->bp;
443                         lp = (struct logpage *) bp->l_ldata;
444                         dstoffset = LOGPHDRSIZE;
445
446                         srclen -= nbytes;
447                         src += nbytes;
448                 }
449
450                 /*
451                  * move log vector descriptor
452                  */
453                 len += 4;
454                 lvd = (struct lvd *) ((caddr_t) lp + dstoffset);
455                 lvd->offset = cpu_to_le16(lv->offset);
456                 lvd->length = cpu_to_le16(lv->length);
457                 dstoffset += 4;
458                 jfs_info("lmWriteRecord: lv offset:%d length:%d",
459                          lv->offset, lv->length);
460         }
461
462         if ((i = linelock->next)) {
463                 linelock = (struct linelock *) lid_to_tlock(i);
464                 goto moveData;
465         }
466
467         /*
468          *      move log record descriptor
469          */
470       moveLrd:
471         lrd->length = cpu_to_le16(len);
472
473         src = (caddr_t) lrd;
474         srclen = LOGRDSIZE;
475
476         while (srclen > 0) {
477                 freespace = (LOGPSIZE - LOGPTLRSIZE) - dstoffset;
478                 nbytes = min(freespace, srclen);
479                 dst = (caddr_t) lp + dstoffset;
480                 memcpy(dst, src, nbytes);
481
482                 dstoffset += nbytes;
483                 srclen -= nbytes;
484
485                 /* are there more to move than freespace of page ? */
486                 if (srclen)
487                         goto pageFull;
488
489                 /*
490                  * end of log record descriptor
491                  */
492
493                 /* update last log record eor */
494                 log->eor = dstoffset;
495                 bp->l_eor = dstoffset;
496                 lsn = (log->page << L2LOGPSIZE) + dstoffset;
497
498                 if (lrd->type & cpu_to_le16(LOG_COMMIT)) {
499                         tblk->clsn = lsn;
500                         jfs_info("wr: tclsn:0x%x, beor:0x%x", tblk->clsn,
501                                  bp->l_eor);
502
503                         INCREMENT(lmStat.commit);       /* # of commit */
504
505                         /*
506                          * enqueue tblock for group commit:
507                          *
508                          * enqueue tblock of non-trivial/synchronous COMMIT
509                          * at tail of group commit queue
510                          * (trivial/asynchronous COMMITs are ignored by
511                          * group commit.)
512                          */
513                         LOGGC_LOCK(log);
514
515                         /* init tblock gc state */
516                         tblk->flag = tblkGC_QUEUE;
517                         tblk->bp = log->bp;
518                         tblk->pn = log->page;
519                         tblk->eor = log->eor;
520
521                         /* enqueue transaction to commit queue */
522                         list_add_tail(&tblk->cqueue, &log->cqueue);
523
524                         LOGGC_UNLOCK(log);
525                 }
526
527                 jfs_info("lmWriteRecord: lrd:0x%04x bp:0x%p pn:%d eor:0x%x",
528                         le16_to_cpu(lrd->type), log->bp, log->page, dstoffset);
529
530                 /* page not full ? */
531                 if (dstoffset < LOGPSIZE - LOGPTLRSIZE)
532                         return lsn;
533
534               pageFull:
535                 /* page become full: move on to next page */
536                 lmNextPage(log);
537
538                 bp = (struct lbuf *) log->bp;
539                 lp = (struct logpage *) bp->l_ldata;
540                 dstoffset = LOGPHDRSIZE;
541                 src += nbytes;
542         }
543
544         return lsn;
545 }
546
547
548 /*
549  * NAME:        lmNextPage()
550  *
551  * FUNCTION:    write current page and allocate next page.
552  *
553  * PARAMETER:   log
554  *
555  * RETURN:      0
556  *                      
557  * serialization: LOG_LOCK() held on entry/exit
558  */
559 static int lmNextPage(struct jfs_log * log)
560 {
561         struct logpage *lp;
562         int lspn;               /* log sequence page number */
563         int pn;                 /* current page number */
564         struct lbuf *bp;
565         struct lbuf *nextbp;
566         struct tblock *tblk;
567
568         /* get current log page number and log sequence page number */
569         pn = log->page;
570         bp = log->bp;
571         lp = (struct logpage *) bp->l_ldata;
572         lspn = le32_to_cpu(lp->h.page);
573
574         LOGGC_LOCK(log);
575
576         /*
577          *      write or queue the full page at the tail of write queue
578          */
579         /* get the tail tblk on commit queue */
580         if (list_empty(&log->cqueue))
581                 tblk = NULL;
582         else
583                 tblk = list_entry(log->cqueue.prev, struct tblock, cqueue);
584
585         /* every tblk who has COMMIT record on the current page,
586          * and has not been committed, must be on commit queue
587          * since tblk is queued at commit queueu at the time
588          * of writing its COMMIT record on the page before
589          * page becomes full (even though the tblk thread
590          * who wrote COMMIT record may have been suspended
591          * currently);
592          */
593
594         /* is page bound with outstanding tail tblk ? */
595         if (tblk && tblk->pn == pn) {
596                 /* mark tblk for end-of-page */
597                 tblk->flag |= tblkGC_EOP;
598
599                 if (log->cflag & logGC_PAGEOUT) {
600                         /* if page is not already on write queue,
601                          * just enqueue (no lbmWRITE to prevent redrive)
602                          * buffer to wqueue to ensure correct serial order
603                          * of the pages since log pages will be added
604                          * continuously
605                          */
606                         if (bp->l_wqnext == NULL)
607                                 lbmWrite(log, bp, 0, 0);
608                 } else {
609                         /*
610                          * No current GC leader, initiate group commit
611                          */
612                         log->cflag |= logGC_PAGEOUT;
613                         lmGCwrite(log, 0);
614                 }
615         }
616         /* page is not bound with outstanding tblk:
617          * init write or mark it to be redriven (lbmWRITE)
618          */
619         else {
620                 /* finalize the page */
621                 bp->l_ceor = bp->l_eor;
622                 lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_ceor);
623                 lbmWrite(log, bp, lbmWRITE | lbmRELEASE | lbmFREE, 0);
624         }
625         LOGGC_UNLOCK(log);
626
627         /*
628          *      allocate/initialize next page
629          */
630         /* if log wraps, the first data page of log is 2
631          * (0 never used, 1 is superblock).
632          */
633         log->page = (pn == log->size - 1) ? 2 : pn + 1;
634         log->eor = LOGPHDRSIZE; /* ? valid page empty/full at logRedo() */
635
636         /* allocate/initialize next log page buffer */
637         nextbp = lbmAllocate(log, log->page);
638         nextbp->l_eor = log->eor;
639         log->bp = nextbp;
640
641         /* initialize next log page */
642         lp = (struct logpage *) nextbp->l_ldata;
643         lp->h.page = lp->t.page = cpu_to_le32(lspn + 1);
644         lp->h.eor = lp->t.eor = cpu_to_le16(LOGPHDRSIZE);
645
646         return 0;
647 }
648
649
650 /*
651  * NAME:        lmGroupCommit()
652  *
653  * FUNCTION:    group commit
654  *      initiate pageout of the pages with COMMIT in the order of
655  *      page number - redrive pageout of the page at the head of
656  *      pageout queue until full page has been written.
657  *
658  * RETURN:      
659  *
660  * NOTE:
661  *      LOGGC_LOCK serializes log group commit queue, and
662  *      transaction blocks on the commit queue.
663  *      N.B. LOG_LOCK is NOT held during lmGroupCommit().
664  */
665 int lmGroupCommit(struct jfs_log * log, struct tblock * tblk)
666 {
667         int rc = 0;
668
669         LOGGC_LOCK(log);
670
671         /* group committed already ? */
672         if (tblk->flag & tblkGC_COMMITTED) {
673                 if (tblk->flag & tblkGC_ERROR)
674                         rc = -EIO;
675
676                 LOGGC_UNLOCK(log);
677                 return rc;
678         }
679         jfs_info("lmGroup Commit: tblk = 0x%p, gcrtc = %d", tblk, log->gcrtc);
680
681         if (tblk->xflag & COMMIT_LAZY)
682                 tblk->flag |= tblkGC_LAZY;
683
684         if ((!(log->cflag & logGC_PAGEOUT)) && (!list_empty(&log->cqueue)) &&
685             (!(tblk->xflag & COMMIT_LAZY) || test_bit(log_FLUSH, &log->flag)
686              || jfs_tlocks_low)) {
687                 /*
688                  * No pageout in progress
689                  *
690                  * start group commit as its group leader.
691                  */
692                 log->cflag |= logGC_PAGEOUT;
693
694                 lmGCwrite(log, 0);
695         }
696
697         if (tblk->xflag & COMMIT_LAZY) {
698                 /*
699                  * Lazy transactions can leave now
700                  */
701                 LOGGC_UNLOCK(log);
702                 return 0;
703         }
704
705         /* lmGCwrite gives up LOGGC_LOCK, check again */
706
707         if (tblk->flag & tblkGC_COMMITTED) {
708                 if (tblk->flag & tblkGC_ERROR)
709                         rc = -EIO;
710
711                 LOGGC_UNLOCK(log);
712                 return rc;
713         }
714
715         /* upcount transaction waiting for completion
716          */
717         log->gcrtc++;
718         tblk->flag |= tblkGC_READY;
719
720         __SLEEP_COND(tblk->gcwait, (tblk->flag & tblkGC_COMMITTED),
721                      LOGGC_LOCK(log), LOGGC_UNLOCK(log));
722
723         /* removed from commit queue */
724         if (tblk->flag & tblkGC_ERROR)
725                 rc = -EIO;
726
727         LOGGC_UNLOCK(log);
728         return rc;
729 }
730
731 /*
732  * NAME:        lmGCwrite()
733  *
734  * FUNCTION:    group commit write
735  *      initiate write of log page, building a group of all transactions
736  *      with commit records on that page.
737  *
738  * RETURN:      None
739  *
740  * NOTE:
741  *      LOGGC_LOCK must be held by caller.
742  *      N.B. LOG_LOCK is NOT held during lmGroupCommit().
743  */
744 static void lmGCwrite(struct jfs_log * log, int cant_write)
745 {
746         struct lbuf *bp;
747         struct logpage *lp;
748         int gcpn;               /* group commit page number */
749         struct tblock *tblk;
750         struct tblock *xtblk = NULL;
751
752         /*
753          * build the commit group of a log page
754          *
755          * scan commit queue and make a commit group of all
756          * transactions with COMMIT records on the same log page.
757          */
758         /* get the head tblk on the commit queue */
759         gcpn = list_entry(log->cqueue.next, struct tblock, cqueue)->pn;
760
761         list_for_each_entry(tblk, &log->cqueue, cqueue) {
762                 if (tblk->pn != gcpn)
763                         break;
764
765                 xtblk = tblk;
766
767                 /* state transition: (QUEUE, READY) -> COMMIT */
768                 tblk->flag |= tblkGC_COMMIT;
769         }
770         tblk = xtblk;           /* last tblk of the page */
771
772         /*
773          * pageout to commit transactions on the log page.
774          */
775         bp = (struct lbuf *) tblk->bp;
776         lp = (struct logpage *) bp->l_ldata;
777         /* is page already full ? */
778         if (tblk->flag & tblkGC_EOP) {
779                 /* mark page to free at end of group commit of the page */
780                 tblk->flag &= ~tblkGC_EOP;
781                 tblk->flag |= tblkGC_FREE;
782                 bp->l_ceor = bp->l_eor;
783                 lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_ceor);
784                 lbmWrite(log, bp, lbmWRITE | lbmRELEASE | lbmGC,
785                          cant_write);
786                 INCREMENT(lmStat.full_page);
787         }
788         /* page is not yet full */
789         else {
790                 bp->l_ceor = tblk->eor; /* ? bp->l_ceor = bp->l_eor; */
791                 lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_ceor);
792                 lbmWrite(log, bp, lbmWRITE | lbmGC, cant_write);
793                 INCREMENT(lmStat.partial_page);
794         }
795 }
796
797 /*
798  * NAME:        lmPostGC()
799  *
800  * FUNCTION:    group commit post-processing
801  *      Processes transactions after their commit records have been written
802  *      to disk, redriving log I/O if necessary.
803  *
804  * RETURN:      None
805  *
806  * NOTE:
807  *      This routine is called a interrupt time by lbmIODone
808  */
809 static void lmPostGC(struct lbuf * bp)
810 {
811         unsigned long flags;
812         struct jfs_log *log = bp->l_log;
813         struct logpage *lp;
814         struct tblock *tblk, *temp;
815
816         //LOGGC_LOCK(log);
817         spin_lock_irqsave(&log->gclock, flags);
818         /*
819          * current pageout of group commit completed.
820          *
821          * remove/wakeup transactions from commit queue who were
822          * group committed with the current log page
823          */
824         list_for_each_entry_safe(tblk, temp, &log->cqueue, cqueue) {
825                 if (!(tblk->flag & tblkGC_COMMIT))
826                         break;
827                 /* if transaction was marked GC_COMMIT then
828                  * it has been shipped in the current pageout
829                  * and made it to disk - it is committed.
830                  */
831
832                 if (bp->l_flag & lbmERROR)
833                         tblk->flag |= tblkGC_ERROR;
834
835                 /* remove it from the commit queue */
836                 list_del(&tblk->cqueue);
837                 tblk->flag &= ~tblkGC_QUEUE;
838
839                 if (tblk == log->flush_tblk) {
840                         /* we can stop flushing the log now */
841                         clear_bit(log_FLUSH, &log->flag);
842                         log->flush_tblk = NULL;
843                 }
844
845                 jfs_info("lmPostGC: tblk = 0x%p, flag = 0x%x", tblk,
846                          tblk->flag);
847
848                 if (!(tblk->xflag & COMMIT_FORCE))
849                         /*
850                          * Hand tblk over to lazy commit thread
851                          */
852                         txLazyUnlock(tblk);
853                 else {
854                         /* state transition: COMMIT -> COMMITTED */
855                         tblk->flag |= tblkGC_COMMITTED;
856
857                         if (tblk->flag & tblkGC_READY)
858                                 log->gcrtc--;
859
860                         LOGGC_WAKEUP(tblk);
861                 }
862
863                 /* was page full before pageout ?
864                  * (and this is the last tblk bound with the page)
865                  */
866                 if (tblk->flag & tblkGC_FREE)
867                         lbmFree(bp);
868                 /* did page become full after pageout ?
869                  * (and this is the last tblk bound with the page)
870                  */
871                 else if (tblk->flag & tblkGC_EOP) {
872                         /* finalize the page */
873                         lp = (struct logpage *) bp->l_ldata;
874                         bp->l_ceor = bp->l_eor;
875                         lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_eor);
876                         jfs_info("lmPostGC: calling lbmWrite");
877                         lbmWrite(log, bp, lbmWRITE | lbmRELEASE | lbmFREE,
878                                  1);
879                 }
880
881         }
882
883         /* are there any transactions who have entered lnGroupCommit()
884          * (whose COMMITs are after that of the last log page written.
885          * They are waiting for new group commit (above at (SLEEP 1))
886          * or lazy transactions are on a full (queued) log page,
887          * select the latest ready transaction as new group leader and
888          * wake her up to lead her group.
889          */
890         if ((!list_empty(&log->cqueue)) &&
891             ((log->gcrtc > 0) || (tblk->bp->l_wqnext != NULL) ||
892              test_bit(log_FLUSH, &log->flag) || jfs_tlocks_low))
893                 /*
894                  * Call lmGCwrite with new group leader
895                  */
896                 lmGCwrite(log, 1);
897
898         /* no transaction are ready yet (transactions are only just
899          * queued (GC_QUEUE) and not entered for group commit yet).
900          * the first transaction entering group commit
901          * will elect herself as new group leader.
902          */
903         else
904                 log->cflag &= ~logGC_PAGEOUT;
905
906         //LOGGC_UNLOCK(log);
907         spin_unlock_irqrestore(&log->gclock, flags);
908         return;
909 }
910
911 /*
912  * NAME:        lmLogSync()
913  *
914  * FUNCTION:    write log SYNCPT record for specified log
915  *      if new sync address is available
916  *      (normally the case if sync() is executed by back-ground
917  *      process).
918  *      calculate new value of i_nextsync which determines when
919  *      this code is called again.
920  *
921  * PARAMETERS:  log     - log structure
922  *              hard_sync - 1 to force all metadata to be written
923  *
924  * RETURN:      0
925  *                      
926  * serialization: LOG_LOCK() held on entry/exit
927  */
928 static int lmLogSync(struct jfs_log * log, int hard_sync)
929 {
930         int logsize;
931         int written;            /* written since last syncpt */
932         int free;               /* free space left available */
933         int delta;              /* additional delta to write normally */
934         int more;               /* additional write granted */
935         struct lrd lrd;
936         int lsn;
937         struct logsyncblk *lp;
938         struct jfs_sb_info *sbi;
939         unsigned long flags;
940
941         /* push dirty metapages out to disk */
942         if (hard_sync)
943                 list_for_each_entry(sbi, &log->sb_list, log_list) {
944                         filemap_fdatawrite(sbi->ipbmap->i_mapping);
945                         filemap_fdatawrite(sbi->ipimap->i_mapping);
946                         filemap_fdatawrite(sbi->direct_inode->i_mapping);
947                 }
948         else
949                 list_for_each_entry(sbi, &log->sb_list, log_list) {
950                         filemap_flush(sbi->ipbmap->i_mapping);
951                         filemap_flush(sbi->ipimap->i_mapping);
952                         filemap_flush(sbi->direct_inode->i_mapping);
953                 }
954
955         /*
956          *      forward syncpt
957          */
958         /* if last sync is same as last syncpt,
959          * invoke sync point forward processing to update sync.
960          */
961
962         if (log->sync == log->syncpt) {
963                 LOGSYNC_LOCK(log, flags);
964                 if (list_empty(&log->synclist))
965                         log->sync = log->lsn;
966                 else {
967                         lp = list_entry(log->synclist.next,
968                                         struct logsyncblk, synclist);
969                         log->sync = lp->lsn;
970                 }
971                 LOGSYNC_UNLOCK(log, flags);
972
973         }
974
975         /* if sync is different from last syncpt,
976          * write a SYNCPT record with syncpt = sync.
977          * reset syncpt = sync
978          */
979         if (log->sync != log->syncpt) {
980                 lrd.logtid = 0;
981                 lrd.backchain = 0;
982                 lrd.type = cpu_to_le16(LOG_SYNCPT);
983                 lrd.length = 0;
984                 lrd.log.syncpt.sync = cpu_to_le32(log->sync);
985                 lsn = lmWriteRecord(log, NULL, &lrd, NULL);
986
987                 log->syncpt = log->sync;
988         } else
989                 lsn = log->lsn;
990
991         /*
992          *      setup next syncpt trigger (SWAG)
993          */
994         logsize = log->logsize;
995
996         logdiff(written, lsn, log);
997         free = logsize - written;
998         delta = LOGSYNC_DELTA(logsize);
999         more = min(free / 2, delta);
1000         if (more < 2 * LOGPSIZE) {
1001                 jfs_warn("\n ... Log Wrap ... Log Wrap ... Log Wrap ...\n");
1002                 /*
1003                  *      log wrapping
1004                  *
1005                  * option 1 - panic ? No.!
1006                  * option 2 - shutdown file systems
1007                  *            associated with log ?
1008                  * option 3 - extend log ?
1009                  */
1010                 /*
1011                  * option 4 - second chance
1012                  *
1013                  * mark log wrapped, and continue.
1014                  * when all active transactions are completed,
1015                  * mark log vaild for recovery.
1016                  * if crashed during invalid state, log state
1017                  * implies invald log, forcing fsck().
1018                  */
1019                 /* mark log state log wrap in log superblock */
1020                 /* log->state = LOGWRAP; */
1021
1022                 /* reset sync point computation */
1023                 log->syncpt = log->sync = lsn;
1024                 log->nextsync = delta;
1025         } else
1026                 /* next syncpt trigger = written + more */
1027                 log->nextsync = written + more;
1028
1029         /* if number of bytes written from last sync point is more
1030          * than 1/4 of the log size, stop new transactions from
1031          * starting until all current transactions are completed
1032          * by setting syncbarrier flag.
1033          */
1034         if (!test_bit(log_SYNCBARRIER, &log->flag) &&
1035             (written > LOGSYNC_BARRIER(logsize)) && log->active) {
1036                 set_bit(log_SYNCBARRIER, &log->flag);
1037                 jfs_info("log barrier on: lsn=0x%x syncpt=0x%x", lsn,
1038                          log->syncpt);
1039                 /*
1040                  * We may have to initiate group commit
1041                  */
1042                 jfs_flush_journal(log, 0);
1043         }
1044
1045         return lsn;
1046 }
1047
1048 /*
1049  * NAME:        jfs_syncpt
1050  *
1051  * FUNCTION:    write log SYNCPT record for specified log
1052  *
1053  * PARAMETERS:  log       - log structure
1054  *              hard_sync - set to 1 to force metadata to be written
1055  */
1056 void jfs_syncpt(struct jfs_log *log, int hard_sync)
1057 {       LOG_LOCK(log);
1058         lmLogSync(log, hard_sync);
1059         LOG_UNLOCK(log);
1060 }
1061
1062 /*
1063  * NAME:        lmLogOpen()
1064  *
1065  * FUNCTION:    open the log on first open;
1066  *      insert filesystem in the active list of the log.
1067  *
1068  * PARAMETER:   ipmnt   - file system mount inode
1069  *              iplog   - log inode (out)
1070  *
1071  * RETURN:
1072  *
1073  * serialization:
1074  */
1075 int lmLogOpen(struct super_block *sb)
1076 {
1077         int rc;
1078         struct block_device *bdev;
1079         struct jfs_log *log;
1080         struct jfs_sb_info *sbi = JFS_SBI(sb);
1081
1082         if (sbi->flag & JFS_NOINTEGRITY)
1083                 return open_dummy_log(sb);
1084         
1085         if (sbi->mntflag & JFS_INLINELOG)
1086                 return open_inline_log(sb);
1087
1088         down(&jfs_log_sem);
1089         list_for_each_entry(log, &jfs_external_logs, journal_list) {
1090                 if (log->bdev->bd_dev == sbi->logdev) {
1091                         if (memcmp(log->uuid, sbi->loguuid,
1092                                    sizeof(log->uuid))) {
1093                                 jfs_warn("wrong uuid on JFS journal\n");
1094                                 up(&jfs_log_sem);
1095                                 return -EINVAL;
1096                         }
1097                         /*
1098                          * add file system to log active file system list
1099                          */
1100                         if ((rc = lmLogFileSystem(log, sbi, 1))) {
1101                                 up(&jfs_log_sem);
1102                                 return rc;
1103                         }
1104                         goto journal_found;
1105                 }
1106         }
1107
1108         if (!(log = kmalloc(sizeof(struct jfs_log), GFP_KERNEL))) {
1109                 up(&jfs_log_sem);
1110                 return -ENOMEM;
1111         }
1112         memset(log, 0, sizeof(struct jfs_log));
1113         INIT_LIST_HEAD(&log->sb_list);
1114         init_waitqueue_head(&log->syncwait);
1115
1116         /*
1117          *      external log as separate logical volume
1118          *
1119          * file systems to log may have n-to-1 relationship;
1120          */
1121
1122         bdev = open_by_devnum(sbi->logdev, FMODE_READ|FMODE_WRITE);
1123         if (IS_ERR(bdev)) {
1124                 rc = -PTR_ERR(bdev);
1125                 goto free;
1126         }
1127
1128         if ((rc = bd_claim(bdev, log))) {
1129                 goto close;
1130         }
1131
1132         log->bdev = bdev;
1133         memcpy(log->uuid, sbi->loguuid, sizeof(log->uuid));
1134         
1135         /*
1136          * initialize log:
1137          */
1138         if ((rc = lmLogInit(log)))
1139                 goto unclaim;
1140
1141         list_add(&log->journal_list, &jfs_external_logs);
1142
1143         /*
1144          * add file system to log active file system list
1145          */
1146         if ((rc = lmLogFileSystem(log, sbi, 1)))
1147                 goto shutdown;
1148
1149 journal_found:
1150         LOG_LOCK(log);
1151         list_add(&sbi->log_list, &log->sb_list);
1152         sbi->log = log;
1153         LOG_UNLOCK(log);
1154
1155         up(&jfs_log_sem);
1156         return 0;
1157
1158         /*
1159          *      unwind on error
1160          */
1161       shutdown:         /* unwind lbmLogInit() */
1162         list_del(&log->journal_list);
1163         lbmLogShutdown(log);
1164
1165       unclaim:
1166         bd_release(bdev);
1167
1168       close:            /* close external log device */
1169         blkdev_put(bdev);
1170
1171       free:             /* free log descriptor */
1172         up(&jfs_log_sem);
1173         kfree(log);
1174
1175         jfs_warn("lmLogOpen: exit(%d)", rc);
1176         return rc;
1177 }
1178
1179 static int open_inline_log(struct super_block *sb)
1180 {
1181         struct jfs_log *log;
1182         int rc;
1183
1184         if (!(log = kmalloc(sizeof(struct jfs_log), GFP_KERNEL)))
1185                 return -ENOMEM;
1186         memset(log, 0, sizeof(struct jfs_log));
1187         INIT_LIST_HEAD(&log->sb_list);
1188         init_waitqueue_head(&log->syncwait);
1189
1190         set_bit(log_INLINELOG, &log->flag);
1191         log->bdev = sb->s_bdev;
1192         log->base = addressPXD(&JFS_SBI(sb)->logpxd);
1193         log->size = lengthPXD(&JFS_SBI(sb)->logpxd) >>
1194             (L2LOGPSIZE - sb->s_blocksize_bits);
1195         log->l2bsize = sb->s_blocksize_bits;
1196         ASSERT(L2LOGPSIZE >= sb->s_blocksize_bits);
1197
1198         /*
1199          * initialize log.
1200          */
1201         if ((rc = lmLogInit(log))) {
1202                 kfree(log);
1203                 jfs_warn("lmLogOpen: exit(%d)", rc);
1204                 return rc;
1205         }
1206
1207         list_add(&JFS_SBI(sb)->log_list, &log->sb_list);
1208         JFS_SBI(sb)->log = log;
1209
1210         return rc;
1211 }
1212
1213 static int open_dummy_log(struct super_block *sb)
1214 {
1215         int rc;
1216
1217         down(&jfs_log_sem);
1218         if (!dummy_log) {
1219                 dummy_log = kmalloc(sizeof(struct jfs_log), GFP_KERNEL);
1220                 if (!dummy_log) {
1221                         up(&jfs_log_sem);
1222                         return -ENOMEM;
1223                 }
1224                 memset(dummy_log, 0, sizeof(struct jfs_log));
1225                 INIT_LIST_HEAD(&dummy_log->sb_list);
1226                 init_waitqueue_head(&dummy_log->syncwait);
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_LIST_HEAD(&log->cqueue);
1290         log->flush_tblk = NULL;
1291
1292         log->count = 0;
1293
1294         /*
1295          * initialize log i/o
1296          */
1297         if ((rc = lbmLogInit(log)))
1298                 return rc;
1299
1300         if (!test_bit(log_INLINELOG, &log->flag))
1301                 log->l2bsize = L2LOGPSIZE;
1302         
1303         /* check for disabled journaling to disk */
1304         if (log->no_integrity) {
1305                 /*
1306                  * Journal pages will still be filled.  When the time comes
1307                  * to actually do the I/O, the write is not done, and the
1308                  * endio routine is called directly.
1309                  */
1310                 bp = lbmAllocate(log , 0);
1311                 log->bp = bp;
1312                 bp->l_pn = bp->l_eor = 0;
1313         } else {
1314                 /*
1315                  * validate log superblock
1316                  */
1317                 if ((rc = lbmRead(log, 1, &bpsuper)))
1318                         goto errout10;
1319
1320                 logsuper = (struct logsuper *) bpsuper->l_ldata;
1321
1322                 if (logsuper->magic != cpu_to_le32(LOGMAGIC)) {
1323                         jfs_warn("*** Log Format Error ! ***");
1324                         rc = -EINVAL;
1325                         goto errout20;
1326                 }
1327
1328                 /* logredo() should have been run successfully. */
1329                 if (logsuper->state != cpu_to_le32(LOGREDONE)) {
1330                         jfs_warn("*** Log Is Dirty ! ***");
1331                         rc = -EINVAL;
1332                         goto errout20;
1333                 }
1334
1335                 /* initialize log from log superblock */
1336                 if (test_bit(log_INLINELOG,&log->flag)) {
1337                         if (log->size != le32_to_cpu(logsuper->size)) {
1338                                 rc = -EINVAL;
1339                                 goto errout20;
1340                         }
1341                         jfs_info("lmLogInit: inline log:0x%p base:0x%Lx "
1342                                  "size:0x%x", log,
1343                                  (unsigned long long) log->base, log->size);
1344                 } else {
1345                         if (memcmp(logsuper->uuid, log->uuid, 16)) {
1346                                 jfs_warn("wrong uuid on JFS log device");
1347                                 goto errout20;
1348                         }
1349                         log->size = le32_to_cpu(logsuper->size);
1350                         log->l2bsize = le32_to_cpu(logsuper->l2bsize);
1351                         jfs_info("lmLogInit: external log:0x%p base:0x%Lx "
1352                                  "size:0x%x", log,
1353                                  (unsigned long long) log->base, log->size);
1354                 }
1355
1356                 log->page = le32_to_cpu(logsuper->end) / LOGPSIZE;
1357                 log->eor = le32_to_cpu(logsuper->end) - (LOGPSIZE * log->page);
1358
1359                 /*
1360                  * initialize for log append write mode
1361                  */
1362                 /* establish current/end-of-log page/buffer */
1363                 if ((rc = lbmRead(log, log->page, &bp)))
1364                         goto errout20;
1365
1366                 lp = (struct logpage *) bp->l_ldata;
1367
1368                 jfs_info("lmLogInit: lsn:0x%x page:%d eor:%d:%d",
1369                          le32_to_cpu(logsuper->end), log->page, log->eor,
1370                          le16_to_cpu(lp->h.eor));
1371
1372                 log->bp = bp;
1373                 bp->l_pn = log->page;
1374                 bp->l_eor = log->eor;
1375
1376                 /* if current page is full, move on to next page */
1377                 if (log->eor >= LOGPSIZE - LOGPTLRSIZE)
1378                         lmNextPage(log);
1379
1380                 /*
1381                  * initialize log syncpoint
1382                  */
1383                 /*
1384                  * write the first SYNCPT record with syncpoint = 0
1385                  * (i.e., log redo up to HERE !);
1386                  * remove current page from lbm write queue at end of pageout
1387                  * (to write log superblock update), but do not release to
1388                  * freelist;
1389                  */
1390                 lrd.logtid = 0;
1391                 lrd.backchain = 0;
1392                 lrd.type = cpu_to_le16(LOG_SYNCPT);
1393                 lrd.length = 0;
1394                 lrd.log.syncpt.sync = 0;
1395                 lsn = lmWriteRecord(log, NULL, &lrd, NULL);
1396                 bp = log->bp;
1397                 bp->l_ceor = bp->l_eor;
1398                 lp = (struct logpage *) bp->l_ldata;
1399                 lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_eor);
1400                 lbmWrite(log, bp, lbmWRITE | lbmSYNC, 0);
1401                 if ((rc = lbmIOWait(bp, 0)))
1402                         goto errout30;
1403
1404                 /*
1405                  * update/write superblock
1406                  */
1407                 logsuper->state = cpu_to_le32(LOGMOUNT);
1408                 log->serial = le32_to_cpu(logsuper->serial) + 1;
1409                 logsuper->serial = cpu_to_le32(log->serial);
1410                 lbmDirectWrite(log, bpsuper, lbmWRITE | lbmRELEASE | lbmSYNC);
1411                 if ((rc = lbmIOWait(bpsuper, lbmFREE)))
1412                         goto errout30;
1413         }
1414
1415         /* initialize logsync parameters */
1416         log->logsize = (log->size - 2) << L2LOGPSIZE;
1417         log->lsn = lsn;
1418         log->syncpt = lsn;
1419         log->sync = log->syncpt;
1420         log->nextsync = LOGSYNC_DELTA(log->logsize);
1421
1422         jfs_info("lmLogInit: lsn:0x%x syncpt:0x%x sync:0x%x",
1423                  log->lsn, log->syncpt, log->sync);
1424
1425         /*
1426          * initialize for lazy/group commit
1427          */
1428         log->clsn = lsn;
1429
1430         return 0;
1431
1432         /*
1433          *      unwind on error
1434          */
1435       errout30:         /* release log page */
1436         log->wqueue = NULL;
1437         bp->l_wqnext = NULL;
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         struct jfs_sb_info *sbi;
1543
1544         /* jfs_write_inode may call us during read-only mount */
1545         if (!log)
1546                 return;
1547
1548         jfs_info("jfs_flush_journal: log:0x%p wait=%d", log, wait);
1549
1550         LOGGC_LOCK(log);
1551
1552         if (!list_empty(&log->cqueue)) {
1553                 /*
1554                  * This ensures that we will keep writing to the journal as long
1555                  * as there are unwritten commit records
1556                  */
1557                 target = list_entry(log->cqueue.prev, struct tblock, cqueue);
1558
1559                 if (test_bit(log_FLUSH, &log->flag)) {
1560                         /*
1561                          * We're already flushing.
1562                          * if flush_tblk is NULL, we are flushing everything,
1563                          * so leave it that way.  Otherwise, update it to the
1564                          * latest transaction
1565                          */
1566                         if (log->flush_tblk)
1567                                 log->flush_tblk = target;
1568                 } else {
1569                         /* Only flush until latest transaction is committed */
1570                         log->flush_tblk = target;
1571                         set_bit(log_FLUSH, &log->flag);
1572
1573                         /*
1574                          * Initiate I/O on outstanding transactions
1575                          */
1576                         if (!(log->cflag & logGC_PAGEOUT)) {
1577                                 log->cflag |= logGC_PAGEOUT;
1578                                 lmGCwrite(log, 0);
1579                         }
1580                 }
1581         }
1582         if ((wait > 1) || test_bit(log_SYNCBARRIER, &log->flag)) {
1583                 /* Flush until all activity complete */
1584                 set_bit(log_FLUSH, &log->flag);
1585                 log->flush_tblk = NULL;
1586         }
1587
1588         if (wait && target && !(target->flag & tblkGC_COMMITTED)) {
1589                 DECLARE_WAITQUEUE(__wait, current);
1590
1591                 add_wait_queue(&target->gcwait, &__wait);
1592                 set_current_state(TASK_UNINTERRUPTIBLE);
1593                 LOGGC_UNLOCK(log);
1594                 schedule();
1595                 current->state = TASK_RUNNING;
1596                 LOGGC_LOCK(log);
1597                 remove_wait_queue(&target->gcwait, &__wait);
1598         }
1599         LOGGC_UNLOCK(log);
1600
1601         if (wait < 2)
1602                 return;
1603
1604         list_for_each_entry(sbi, &log->sb_list, log_list) {
1605                 filemap_fdatawrite(sbi->ipbmap->i_mapping);
1606                 filemap_fdatawrite(sbi->ipimap->i_mapping);
1607                 filemap_fdatawrite(sbi->direct_inode->i_mapping);
1608         }
1609
1610         /*
1611          * If there was recent activity, we may need to wait
1612          * for the lazycommit thread to catch up
1613          */
1614         if ((!list_empty(&log->cqueue)) || !list_empty(&log->synclist)) {
1615                 for (i = 0; i < 200; i++) {     /* Too much? */
1616                         msleep(250);
1617                         if (list_empty(&log->cqueue) &&
1618                             list_empty(&log->synclist))
1619                                 break;
1620                 }
1621         }
1622         assert(list_empty(&log->cqueue));
1623
1624 #ifdef CONFIG_JFS_DEBUG
1625         if (!list_empty(&log->synclist)) {
1626                 struct logsyncblk *lp;
1627
1628                 list_for_each_entry(lp, &log->synclist, synclist) {
1629                         if (lp->xflag & COMMIT_PAGE) {
1630                                 struct metapage *mp = (struct metapage *)lp;
1631                                 dump_mem("orphan metapage", lp,
1632                                          sizeof(struct metapage));
1633                                 dump_mem("page", mp->page, sizeof(struct page));
1634                         }
1635                         else
1636                                 dump_mem("orphan tblock", lp,
1637                                          sizeof(struct tblock));
1638                 }
1639         }
1640 #endif
1641         //assert(list_empty(&log->synclist));
1642         clear_bit(log_FLUSH, &log->flag);
1643 }
1644
1645 /*
1646  * NAME:        lmLogShutdown()
1647  *
1648  * FUNCTION:    log shutdown at last LogClose().
1649  *
1650  *              write log syncpt record.
1651  *              update super block to set redone flag to 0.
1652  *
1653  * PARAMETER:   log     - log inode
1654  *
1655  * RETURN:      0       - success
1656  *                      
1657  * serialization: single last close thread
1658  */
1659 int lmLogShutdown(struct jfs_log * log)
1660 {
1661         int rc;
1662         struct lrd lrd;
1663         int lsn;
1664         struct logsuper *logsuper;
1665         struct lbuf *bpsuper;
1666         struct lbuf *bp;
1667         struct logpage *lp;
1668
1669         jfs_info("lmLogShutdown: log:0x%p", log);
1670
1671         jfs_flush_journal(log, 2);
1672
1673         /*
1674          * write the last SYNCPT record with syncpoint = 0
1675          * (i.e., log redo up to HERE !)
1676          */
1677         lrd.logtid = 0;
1678         lrd.backchain = 0;
1679         lrd.type = cpu_to_le16(LOG_SYNCPT);
1680         lrd.length = 0;
1681         lrd.log.syncpt.sync = 0;
1682         
1683         lsn = lmWriteRecord(log, NULL, &lrd, NULL);
1684         bp = log->bp;
1685         lp = (struct logpage *) bp->l_ldata;
1686         lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_eor);
1687         lbmWrite(log, log->bp, lbmWRITE | lbmRELEASE | lbmSYNC, 0);
1688         lbmIOWait(log->bp, lbmFREE);
1689         log->bp = NULL;
1690
1691         /*
1692          * synchronous update log superblock
1693          * mark log state as shutdown cleanly
1694          * (i.e., Log does not need to be replayed).
1695          */
1696         if ((rc = lbmRead(log, 1, &bpsuper)))
1697                 goto out;
1698
1699         logsuper = (struct logsuper *) bpsuper->l_ldata;
1700         logsuper->state = cpu_to_le32(LOGREDONE);
1701         logsuper->end = cpu_to_le32(lsn);
1702         lbmDirectWrite(log, bpsuper, lbmWRITE | lbmRELEASE | lbmSYNC);
1703         rc = lbmIOWait(bpsuper, lbmFREE);
1704
1705         jfs_info("lmLogShutdown: lsn:0x%x page:%d eor:%d",
1706                  lsn, log->page, log->eor);
1707
1708       out:    
1709         /*
1710          * shutdown per log i/o
1711          */
1712         lbmLogShutdown(log);
1713
1714         if (rc) {
1715                 jfs_warn("lmLogShutdown: exit(%d)", rc);
1716         }
1717         return rc;
1718 }
1719
1720
1721 /*
1722  * NAME:        lmLogFileSystem()
1723  *
1724  * FUNCTION:    insert (<activate> = true)/remove (<activate> = false)
1725  *      file system into/from log active file system list.
1726  *
1727  * PARAMETE:    log     - pointer to logs inode.
1728  *              fsdev   - kdev_t of filesystem.
1729  *              serial  - pointer to returned log serial number
1730  *              activate - insert/remove device from active list.
1731  *
1732  * RETURN:      0       - success
1733  *              errors returned by vms_iowait().
1734  */
1735 static int lmLogFileSystem(struct jfs_log * log, struct jfs_sb_info *sbi,
1736                            int activate)
1737 {
1738         int rc = 0;
1739         int i;
1740         struct logsuper *logsuper;
1741         struct lbuf *bpsuper;
1742         char *uuid = sbi->uuid;
1743
1744         /*
1745          * insert/remove file system device to log active file system list.
1746          */
1747         if ((rc = lbmRead(log, 1, &bpsuper)))
1748                 return rc;
1749
1750         logsuper = (struct logsuper *) bpsuper->l_ldata;
1751         if (activate) {
1752                 for (i = 0; i < MAX_ACTIVE; i++)
1753                         if (!memcmp(logsuper->active[i].uuid, NULL_UUID, 16)) {
1754                                 memcpy(logsuper->active[i].uuid, uuid, 16);
1755                                 sbi->aggregate = i;
1756                                 break;
1757                         }
1758                 if (i == MAX_ACTIVE) {
1759                         jfs_warn("Too many file systems sharing journal!");
1760                         lbmFree(bpsuper);
1761                         return -EMFILE; /* Is there a better rc? */
1762                 }
1763         } else {
1764                 for (i = 0; i < MAX_ACTIVE; i++)
1765                         if (!memcmp(logsuper->active[i].uuid, uuid, 16)) {
1766                                 memcpy(logsuper->active[i].uuid, NULL_UUID, 16);
1767                                 break;
1768                         }
1769                 if (i == MAX_ACTIVE) {
1770                         jfs_warn("Somebody stomped on the journal!");
1771                         lbmFree(bpsuper);
1772                         return -EIO;
1773                 }
1774                 
1775         }
1776
1777         /*
1778          * synchronous write log superblock:
1779          *
1780          * write sidestream bypassing write queue:
1781          * at file system mount, log super block is updated for
1782          * activation of the file system before any log record
1783          * (MOUNT record) of the file system, and at file system
1784          * unmount, all meta data for the file system has been
1785          * flushed before log super block is updated for deactivation
1786          * of the file system.
1787          */
1788         lbmDirectWrite(log, bpsuper, lbmWRITE | lbmRELEASE | lbmSYNC);
1789         rc = lbmIOWait(bpsuper, lbmFREE);
1790
1791         return rc;
1792 }
1793
1794 /*
1795  *              log buffer manager (lbm)
1796  *              ------------------------
1797  *
1798  * special purpose buffer manager supporting log i/o requirements.
1799  *
1800  * per log write queue:
1801  * log pageout occurs in serial order by fifo write queue and
1802  * restricting to a single i/o in pregress at any one time.
1803  * a circular singly-linked list
1804  * (log->wrqueue points to the tail, and buffers are linked via
1805  * bp->wrqueue field), and
1806  * maintains log page in pageout ot waiting for pageout in serial pageout.
1807  */
1808
1809 /*
1810  *      lbmLogInit()
1811  *
1812  * initialize per log I/O setup at lmLogInit()
1813  */
1814 static int lbmLogInit(struct jfs_log * log)
1815 {                               /* log inode */
1816         int i;
1817         struct lbuf *lbuf;
1818
1819         jfs_info("lbmLogInit: log:0x%p", log);
1820
1821         /* initialize current buffer cursor */
1822         log->bp = NULL;
1823
1824         /* initialize log device write queue */
1825         log->wqueue = NULL;
1826
1827         /*
1828          * Each log has its own buffer pages allocated to it.  These are
1829          * not managed by the page cache.  This ensures that a transaction
1830          * writing to the log does not block trying to allocate a page from
1831          * the page cache (for the log).  This would be bad, since page
1832          * allocation waits on the kswapd thread that may be committing inodes
1833          * which would cause log activity.  Was that clear?  I'm trying to
1834          * avoid deadlock here.
1835          */
1836         init_waitqueue_head(&log->free_wait);
1837
1838         log->lbuf_free = NULL;
1839
1840         for (i = 0; i < LOGPAGES;) {
1841                 char *buffer;
1842                 uint offset;
1843                 struct page *page;
1844
1845                 buffer = (char *) get_zeroed_page(GFP_KERNEL);
1846                 if (buffer == NULL)
1847                         goto error;
1848                 page = virt_to_page(buffer);
1849                 for (offset = 0; offset < PAGE_SIZE; offset += LOGPSIZE) {
1850                         lbuf = kmalloc(sizeof(struct lbuf), GFP_KERNEL);
1851                         if (lbuf == NULL) {
1852                                 if (offset == 0)
1853                                         free_page((unsigned long) buffer);
1854                                 goto error;
1855                         }
1856                         if (offset) /* we already have one reference */
1857                                 get_page(page);
1858                         lbuf->l_offset = offset;
1859                         lbuf->l_ldata = buffer + offset;
1860                         lbuf->l_page = page;
1861                         lbuf->l_log = log;
1862                         init_waitqueue_head(&lbuf->l_ioevent);
1863
1864                         lbuf->l_freelist = log->lbuf_free;
1865                         log->lbuf_free = lbuf;
1866                         i++;
1867                 }
1868         }
1869
1870         return (0);
1871
1872       error:
1873         lbmLogShutdown(log);
1874         return -ENOMEM;
1875 }
1876
1877
1878 /*
1879  *      lbmLogShutdown()
1880  *
1881  * finalize per log I/O setup at lmLogShutdown()
1882  */
1883 static void lbmLogShutdown(struct jfs_log * log)
1884 {
1885         struct lbuf *lbuf;
1886
1887         jfs_info("lbmLogShutdown: log:0x%p", log);
1888
1889         lbuf = log->lbuf_free;
1890         while (lbuf) {
1891                 struct lbuf *next = lbuf->l_freelist;
1892                 __free_page(lbuf->l_page);
1893                 kfree(lbuf);
1894                 lbuf = next;
1895         }
1896 }
1897
1898
1899 /*
1900  *      lbmAllocate()
1901  *
1902  * allocate an empty log buffer
1903  */
1904 static struct lbuf *lbmAllocate(struct jfs_log * log, int pn)
1905 {
1906         struct lbuf *bp;
1907         unsigned long flags;
1908
1909         /*
1910          * recycle from log buffer freelist if any
1911          */
1912         LCACHE_LOCK(flags);
1913         LCACHE_SLEEP_COND(log->free_wait, (bp = log->lbuf_free), flags);
1914         log->lbuf_free = bp->l_freelist;
1915         LCACHE_UNLOCK(flags);
1916
1917         bp->l_flag = 0;
1918
1919         bp->l_wqnext = NULL;
1920         bp->l_freelist = NULL;
1921
1922         bp->l_pn = pn;
1923         bp->l_blkno = log->base + (pn << (L2LOGPSIZE - log->l2bsize));
1924         bp->l_ceor = 0;
1925
1926         return bp;
1927 }
1928
1929
1930 /*
1931  *      lbmFree()
1932  *
1933  * release a log buffer to freelist
1934  */
1935 static void lbmFree(struct lbuf * bp)
1936 {
1937         unsigned long flags;
1938
1939         LCACHE_LOCK(flags);
1940
1941         lbmfree(bp);
1942
1943         LCACHE_UNLOCK(flags);
1944 }
1945
1946 static void lbmfree(struct lbuf * bp)
1947 {
1948         struct jfs_log *log = bp->l_log;
1949
1950         assert(bp->l_wqnext == NULL);
1951
1952         /*
1953          * return the buffer to head of freelist
1954          */
1955         bp->l_freelist = log->lbuf_free;
1956         log->lbuf_free = bp;
1957
1958         wake_up(&log->free_wait);
1959         return;
1960 }
1961
1962
1963 /*
1964  * NAME:        lbmRedrive
1965  *
1966  * FUNCTION:    add a log buffer to the the log redrive list
1967  *
1968  * PARAMETER:
1969  *     bp       - log buffer
1970  *
1971  * NOTES:
1972  *      Takes log_redrive_lock.
1973  */
1974 static inline void lbmRedrive(struct lbuf *bp)
1975 {
1976         unsigned long flags;
1977
1978         spin_lock_irqsave(&log_redrive_lock, flags);
1979         bp->l_redrive_next = log_redrive_list;
1980         log_redrive_list = bp;
1981         spin_unlock_irqrestore(&log_redrive_lock, flags);
1982
1983         wake_up(&jfs_IO_thread_wait);
1984 }
1985
1986
1987 /*
1988  *      lbmRead()
1989  */
1990 static int lbmRead(struct jfs_log * log, int pn, struct lbuf ** bpp)
1991 {
1992         struct bio *bio;
1993         struct lbuf *bp;
1994
1995         /*
1996          * allocate a log buffer
1997          */
1998         *bpp = bp = lbmAllocate(log, pn);
1999         jfs_info("lbmRead: bp:0x%p pn:0x%x", bp, pn);
2000
2001         bp->l_flag |= lbmREAD;
2002
2003         bio = bio_alloc(GFP_NOFS, 1);
2004
2005         bio->bi_sector = bp->l_blkno << (log->l2bsize - 9);
2006         bio->bi_bdev = log->bdev;
2007         bio->bi_io_vec[0].bv_page = bp->l_page;
2008         bio->bi_io_vec[0].bv_len = LOGPSIZE;
2009         bio->bi_io_vec[0].bv_offset = bp->l_offset;
2010
2011         bio->bi_vcnt = 1;
2012         bio->bi_idx = 0;
2013         bio->bi_size = LOGPSIZE;
2014
2015         bio->bi_end_io = lbmIODone;
2016         bio->bi_private = bp;
2017         submit_bio(READ_SYNC, bio);
2018
2019         wait_event(bp->l_ioevent, (bp->l_flag != lbmREAD));
2020
2021         return 0;
2022 }
2023
2024
2025 /*
2026  *      lbmWrite()
2027  *
2028  * buffer at head of pageout queue stays after completion of
2029  * partial-page pageout and redriven by explicit initiation of
2030  * pageout by caller until full-page pageout is completed and
2031  * released.
2032  *
2033  * device driver i/o done redrives pageout of new buffer at
2034  * head of pageout queue when current buffer at head of pageout
2035  * queue is released at the completion of its full-page pageout.
2036  *
2037  * LOGGC_LOCK() serializes lbmWrite() by lmNextPage() and lmGroupCommit().
2038  * LCACHE_LOCK() serializes xflag between lbmWrite() and lbmIODone()
2039  */
2040 static void lbmWrite(struct jfs_log * log, struct lbuf * bp, int flag,
2041                      int cant_block)
2042 {
2043         struct lbuf *tail;
2044         unsigned long flags;
2045
2046         jfs_info("lbmWrite: bp:0x%p flag:0x%x pn:0x%x", bp, flag, bp->l_pn);
2047
2048         /* map the logical block address to physical block address */
2049         bp->l_blkno =
2050             log->base + (bp->l_pn << (L2LOGPSIZE - log->l2bsize));
2051
2052         LCACHE_LOCK(flags);             /* disable+lock */
2053
2054         /*
2055          * initialize buffer for device driver
2056          */
2057         bp->l_flag = flag;
2058
2059         /*
2060          *      insert bp at tail of write queue associated with log
2061          *
2062          * (request is either for bp already/currently at head of queue
2063          * or new bp to be inserted at tail)
2064          */
2065         tail = log->wqueue;
2066
2067         /* is buffer not already on write queue ? */
2068         if (bp->l_wqnext == NULL) {
2069                 /* insert at tail of wqueue */
2070                 if (tail == NULL) {
2071                         log->wqueue = bp;
2072                         bp->l_wqnext = bp;
2073                 } else {
2074                         log->wqueue = bp;
2075                         bp->l_wqnext = tail->l_wqnext;
2076                         tail->l_wqnext = bp;
2077                 }
2078
2079                 tail = bp;
2080         }
2081
2082         /* is buffer at head of wqueue and for write ? */
2083         if ((bp != tail->l_wqnext) || !(flag & lbmWRITE)) {
2084                 LCACHE_UNLOCK(flags);   /* unlock+enable */
2085                 return;
2086         }
2087
2088         LCACHE_UNLOCK(flags);   /* unlock+enable */
2089
2090         if (cant_block)
2091                 lbmRedrive(bp);
2092         else if (flag & lbmSYNC)
2093                 lbmStartIO(bp);
2094         else {
2095                 LOGGC_UNLOCK(log);
2096                 lbmStartIO(bp);
2097                 LOGGC_LOCK(log);
2098         }
2099 }
2100
2101
2102 /*
2103  *      lbmDirectWrite()
2104  *
2105  * initiate pageout bypassing write queue for sidestream
2106  * (e.g., log superblock) write;
2107  */
2108 static void lbmDirectWrite(struct jfs_log * log, struct lbuf * bp, int flag)
2109 {
2110         jfs_info("lbmDirectWrite: bp:0x%p flag:0x%x pn:0x%x",
2111                  bp, flag, bp->l_pn);
2112
2113         /*
2114          * initialize buffer for device driver
2115          */
2116         bp->l_flag = flag | lbmDIRECT;
2117
2118         /* map the logical block address to physical block address */
2119         bp->l_blkno =
2120             log->base + (bp->l_pn << (L2LOGPSIZE - log->l2bsize));
2121
2122         /*
2123          *      initiate pageout of the page
2124          */
2125         lbmStartIO(bp);
2126 }
2127
2128
2129 /*
2130  * NAME:        lbmStartIO()
2131  *
2132  * FUNCTION:    Interface to DD strategy routine
2133  *
2134  * RETURN:      none
2135  *
2136  * serialization: LCACHE_LOCK() is NOT held during log i/o;
2137  */
2138 static void lbmStartIO(struct lbuf * bp)
2139 {
2140         struct bio *bio;
2141         struct jfs_log *log = bp->l_log;
2142
2143         jfs_info("lbmStartIO\n");
2144
2145         bio = bio_alloc(GFP_NOFS, 1);
2146         bio->bi_sector = bp->l_blkno << (log->l2bsize - 9);
2147         bio->bi_bdev = log->bdev;
2148         bio->bi_io_vec[0].bv_page = bp->l_page;
2149         bio->bi_io_vec[0].bv_len = LOGPSIZE;
2150         bio->bi_io_vec[0].bv_offset = bp->l_offset;
2151
2152         bio->bi_vcnt = 1;
2153         bio->bi_idx = 0;
2154         bio->bi_size = LOGPSIZE;
2155
2156         bio->bi_end_io = lbmIODone;
2157         bio->bi_private = bp;
2158
2159         /* check if journaling to disk has been disabled */
2160         if (log->no_integrity) {
2161                 bio->bi_size = 0;
2162                 lbmIODone(bio, 0, 0);
2163         } else {
2164                 submit_bio(WRITE_SYNC, bio);
2165                 INCREMENT(lmStat.submitted);
2166         }
2167 }
2168
2169
2170 /*
2171  *      lbmIOWait()
2172  */
2173 static int lbmIOWait(struct lbuf * bp, int flag)
2174 {
2175         unsigned long flags;
2176         int rc = 0;
2177
2178         jfs_info("lbmIOWait1: bp:0x%p flag:0x%x:0x%x", bp, bp->l_flag, flag);
2179
2180         LCACHE_LOCK(flags);             /* disable+lock */
2181
2182         LCACHE_SLEEP_COND(bp->l_ioevent, (bp->l_flag & lbmDONE), flags);
2183
2184         rc = (bp->l_flag & lbmERROR) ? -EIO : 0;
2185
2186         if (flag & lbmFREE)
2187                 lbmfree(bp);
2188
2189         LCACHE_UNLOCK(flags);   /* unlock+enable */
2190
2191         jfs_info("lbmIOWait2: bp:0x%p flag:0x%x:0x%x", bp, bp->l_flag, flag);
2192         return rc;
2193 }
2194
2195 /*
2196  *      lbmIODone()
2197  *
2198  * executed at INTIODONE level
2199  */
2200 static int lbmIODone(struct bio *bio, unsigned int bytes_done, int error)
2201 {
2202         struct lbuf *bp = bio->bi_private;
2203         struct lbuf *nextbp, *tail;
2204         struct jfs_log *log;
2205         unsigned long flags;
2206
2207         if (bio->bi_size)
2208                 return 1;
2209
2210         /*
2211          * get back jfs buffer bound to the i/o buffer
2212          */
2213         jfs_info("lbmIODone: bp:0x%p flag:0x%x", bp, bp->l_flag);
2214
2215         LCACHE_LOCK(flags);             /* disable+lock */
2216
2217         bp->l_flag |= lbmDONE;
2218
2219         if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
2220                 bp->l_flag |= lbmERROR;
2221
2222                 jfs_err("lbmIODone: I/O error in JFS log");
2223         }
2224
2225         bio_put(bio);
2226
2227         /*
2228          *      pagein completion
2229          */
2230         if (bp->l_flag & lbmREAD) {
2231                 bp->l_flag &= ~lbmREAD;
2232
2233                 LCACHE_UNLOCK(flags);   /* unlock+enable */
2234
2235                 /* wakeup I/O initiator */
2236                 LCACHE_WAKEUP(&bp->l_ioevent);
2237
2238                 return 0;
2239         }
2240
2241         /*
2242          *      pageout completion
2243          *
2244          * the bp at the head of write queue has completed pageout.
2245          *
2246          * if single-commit/full-page pageout, remove the current buffer
2247          * from head of pageout queue, and redrive pageout with
2248          * the new buffer at head of pageout queue;
2249          * otherwise, the partial-page pageout buffer stays at
2250          * the head of pageout queue to be redriven for pageout
2251          * by lmGroupCommit() until full-page pageout is completed.
2252          */
2253         bp->l_flag &= ~lbmWRITE;
2254         INCREMENT(lmStat.pagedone);
2255
2256         /* update committed lsn */
2257         log = bp->l_log;
2258         log->clsn = (bp->l_pn << L2LOGPSIZE) + bp->l_ceor;
2259
2260         if (bp->l_flag & lbmDIRECT) {
2261                 LCACHE_WAKEUP(&bp->l_ioevent);
2262                 LCACHE_UNLOCK(flags);
2263                 return 0;
2264         }
2265
2266         tail = log->wqueue;
2267
2268         /* single element queue */
2269         if (bp == tail) {
2270                 /* remove head buffer of full-page pageout
2271                  * from log device write queue
2272                  */
2273                 if (bp->l_flag & lbmRELEASE) {
2274                         log->wqueue = NULL;
2275                         bp->l_wqnext = NULL;
2276                 }
2277         }
2278         /* multi element queue */
2279         else {
2280                 /* remove head buffer of full-page pageout
2281                  * from log device write queue
2282                  */
2283                 if (bp->l_flag & lbmRELEASE) {
2284                         nextbp = tail->l_wqnext = bp->l_wqnext;
2285                         bp->l_wqnext = NULL;
2286
2287                         /*
2288                          * redrive pageout of next page at head of write queue:
2289                          * redrive next page without any bound tblk
2290                          * (i.e., page w/o any COMMIT records), or
2291                          * first page of new group commit which has been
2292                          * queued after current page (subsequent pageout
2293                          * is performed synchronously, except page without
2294                          * any COMMITs) by lmGroupCommit() as indicated
2295                          * by lbmWRITE flag;
2296                          */
2297                         if (nextbp->l_flag & lbmWRITE) {
2298                                 /*
2299                                  * We can't do the I/O at interrupt time.
2300                                  * The jfsIO thread can do it
2301                                  */
2302                                 lbmRedrive(nextbp);
2303                         }
2304                 }
2305         }
2306
2307         /*
2308          *      synchronous pageout:
2309          *
2310          * buffer has not necessarily been removed from write queue
2311          * (e.g., synchronous write of partial-page with COMMIT):
2312          * leave buffer for i/o initiator to dispose
2313          */
2314         if (bp->l_flag & lbmSYNC) {
2315                 LCACHE_UNLOCK(flags);   /* unlock+enable */
2316
2317                 /* wakeup I/O initiator */
2318                 LCACHE_WAKEUP(&bp->l_ioevent);
2319         }
2320
2321         /*
2322          *      Group Commit pageout:
2323          */
2324         else if (bp->l_flag & lbmGC) {
2325                 LCACHE_UNLOCK(flags);
2326                 lmPostGC(bp);
2327         }
2328
2329         /*
2330          *      asynchronous pageout:
2331          *
2332          * buffer must have been removed from write queue:
2333          * insert buffer at head of freelist where it can be recycled
2334          */
2335         else {
2336                 assert(bp->l_flag & lbmRELEASE);
2337                 assert(bp->l_flag & lbmFREE);
2338                 lbmfree(bp);
2339
2340                 LCACHE_UNLOCK(flags);   /* unlock+enable */
2341         }
2342
2343         return 0;
2344 }
2345
2346 int jfsIOWait(void *arg)
2347 {
2348         struct lbuf *bp;
2349
2350         daemonize("jfsIO");
2351
2352         complete(&jfsIOwait);
2353
2354         do {
2355                 DECLARE_WAITQUEUE(wq, current);
2356
2357                 spin_lock_irq(&log_redrive_lock);
2358                 while ((bp = log_redrive_list) != 0) {
2359                         log_redrive_list = bp->l_redrive_next;
2360                         bp->l_redrive_next = NULL;
2361                         spin_unlock_irq(&log_redrive_lock);
2362                         lbmStartIO(bp);
2363                         spin_lock_irq(&log_redrive_lock);
2364                 }
2365                 if (freezing(current)) {
2366                         spin_unlock_irq(&log_redrive_lock);
2367                         refrigerator();
2368                 } else {
2369                         add_wait_queue(&jfs_IO_thread_wait, &wq);
2370                         set_current_state(TASK_INTERRUPTIBLE);
2371                         spin_unlock_irq(&log_redrive_lock);
2372                         schedule();
2373                         current->state = TASK_RUNNING;
2374                         remove_wait_queue(&jfs_IO_thread_wait, &wq);
2375                 }
2376         } while (!jfs_stop_threads);
2377
2378         jfs_info("jfsIOWait being killed!");
2379         complete_and_exit(&jfsIOwait, 0);
2380 }
2381
2382 /*
2383  * NAME:        lmLogFormat()/jfs_logform()
2384  *
2385  * FUNCTION:    format file system log
2386  *
2387  * PARAMETERS:
2388  *      log     - volume log
2389  *      logAddress - start address of log space in FS block
2390  *      logSize - length of log space in FS block;
2391  *
2392  * RETURN:      0       - success
2393  *              -EIO    - i/o error
2394  *
2395  * XXX: We're synchronously writing one page at a time.  This needs to
2396  *      be improved by writing multiple pages at once.
2397  */
2398 int lmLogFormat(struct jfs_log *log, s64 logAddress, int logSize)
2399 {
2400         int rc = -EIO;
2401         struct jfs_sb_info *sbi;
2402         struct logsuper *logsuper;
2403         struct logpage *lp;
2404         int lspn;               /* log sequence page number */
2405         struct lrd *lrd_ptr;
2406         int npages = 0;
2407         struct lbuf *bp;
2408
2409         jfs_info("lmLogFormat: logAddress:%Ld logSize:%d",
2410                  (long long)logAddress, logSize);
2411
2412         sbi = list_entry(log->sb_list.next, struct jfs_sb_info, log_list);
2413
2414         /* allocate a log buffer */
2415         bp = lbmAllocate(log, 1);
2416
2417         npages = logSize >> sbi->l2nbperpage;
2418
2419         /*
2420          *      log space:
2421          *
2422          * page 0 - reserved;
2423          * page 1 - log superblock;
2424          * page 2 - log data page: A SYNC log record is written
2425          *          into this page at logform time;
2426          * pages 3-N - log data page: set to empty log data pages;
2427          */
2428         /*
2429          *      init log superblock: log page 1
2430          */
2431         logsuper = (struct logsuper *) bp->l_ldata;
2432
2433         logsuper->magic = cpu_to_le32(LOGMAGIC);
2434         logsuper->version = cpu_to_le32(LOGVERSION);
2435         logsuper->state = cpu_to_le32(LOGREDONE);
2436         logsuper->flag = cpu_to_le32(sbi->mntflag);     /* ? */
2437         logsuper->size = cpu_to_le32(npages);
2438         logsuper->bsize = cpu_to_le32(sbi->bsize);
2439         logsuper->l2bsize = cpu_to_le32(sbi->l2bsize);
2440         logsuper->end = cpu_to_le32(2 * LOGPSIZE + LOGPHDRSIZE + LOGRDSIZE);
2441
2442         bp->l_flag = lbmWRITE | lbmSYNC | lbmDIRECT;
2443         bp->l_blkno = logAddress + sbi->nbperpage;
2444         lbmStartIO(bp);
2445         if ((rc = lbmIOWait(bp, 0)))
2446                 goto exit;
2447
2448         /*
2449          *      init pages 2 to npages-1 as log data pages:
2450          *
2451          * log page sequence number (lpsn) initialization:
2452          *
2453          * pn:   0     1     2     3                 n-1
2454          *       +-----+-----+=====+=====+===.....===+=====+
2455          * lspn:             N-1   0     1           N-2
2456          *                   <--- N page circular file ---->
2457          *
2458          * the N (= npages-2) data pages of the log is maintained as
2459          * a circular file for the log records;
2460          * lpsn grows by 1 monotonically as each log page is written
2461          * to the circular file of the log;
2462          * and setLogpage() will not reset the page number even if
2463          * the eor is equal to LOGPHDRSIZE. In order for binary search
2464          * still work in find log end process, we have to simulate the
2465          * log wrap situation at the log format time.
2466          * The 1st log page written will have the highest lpsn. Then
2467          * the succeeding log pages will have ascending order of
2468          * the lspn starting from 0, ... (N-2)
2469          */
2470         lp = (struct logpage *) bp->l_ldata;
2471         /*
2472          * initialize 1st log page to be written: lpsn = N - 1,
2473          * write a SYNCPT log record is written to this page
2474          */
2475         lp->h.page = lp->t.page = cpu_to_le32(npages - 3);
2476         lp->h.eor = lp->t.eor = cpu_to_le16(LOGPHDRSIZE + LOGRDSIZE);
2477
2478         lrd_ptr = (struct lrd *) &lp->data;
2479         lrd_ptr->logtid = 0;
2480         lrd_ptr->backchain = 0;
2481         lrd_ptr->type = cpu_to_le16(LOG_SYNCPT);
2482         lrd_ptr->length = 0;
2483         lrd_ptr->log.syncpt.sync = 0;
2484
2485         bp->l_blkno += sbi->nbperpage;
2486         bp->l_flag = lbmWRITE | lbmSYNC | lbmDIRECT;
2487         lbmStartIO(bp);
2488         if ((rc = lbmIOWait(bp, 0)))
2489                 goto exit;
2490
2491         /*
2492          *      initialize succeeding log pages: lpsn = 0, 1, ..., (N-2)
2493          */
2494         for (lspn = 0; lspn < npages - 3; lspn++) {
2495                 lp->h.page = lp->t.page = cpu_to_le32(lspn);
2496                 lp->h.eor = lp->t.eor = cpu_to_le16(LOGPHDRSIZE);
2497
2498                 bp->l_blkno += sbi->nbperpage;
2499                 bp->l_flag = lbmWRITE | lbmSYNC | lbmDIRECT;
2500                 lbmStartIO(bp);
2501                 if ((rc = lbmIOWait(bp, 0)))
2502                         goto exit;
2503         }
2504
2505         rc = 0;
2506 exit:
2507         /*
2508          *      finalize log
2509          */
2510         /* release the buffer */
2511         lbmFree(bp);
2512
2513         return rc;
2514 }
2515
2516 #ifdef CONFIG_JFS_STATISTICS
2517 int jfs_lmstats_read(char *buffer, char **start, off_t offset, int length,
2518                       int *eof, void *data)
2519 {
2520         int len = 0;
2521         off_t begin;
2522
2523         len += sprintf(buffer,
2524                        "JFS Logmgr stats\n"
2525                        "================\n"
2526                        "commits = %d\n"
2527                        "writes submitted = %d\n"
2528                        "writes completed = %d\n"
2529                        "full pages submitted = %d\n"
2530                        "partial pages submitted = %d\n",
2531                        lmStat.commit,
2532                        lmStat.submitted,
2533                        lmStat.pagedone,
2534                        lmStat.full_page,
2535                        lmStat.partial_page);
2536
2537         begin = offset;
2538         *start = buffer + begin;
2539         len -= begin;
2540
2541         if (len > length)
2542                 len = length;
2543         else
2544                 *eof = 1;
2545
2546         if (len < 0)
2547                 len = 0;
2548
2549         return len;
2550 }
2551 #endif /* CONFIG_JFS_STATISTICS */