[PATCH] sem2mutex: fs/
[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/kthread.h>
68 #include <linux/buffer_head.h>          /* for sync_blockdev() */
69 #include <linux/bio.h>
70 #include <linux/suspend.h>
71 #include <linux/delay.h>
72 #include <linux/mutex.h>
73 #include "jfs_incore.h"
74 #include "jfs_filsys.h"
75 #include "jfs_metapage.h"
76 #include "jfs_superblock.h"
77 #include "jfs_txnmgr.h"
78 #include "jfs_debug.h"
79
80
81 /*
82  * lbuf's ready to be redriven.  Protected by log_redrive_lock (jfsIO thread)
83  */
84 static struct lbuf *log_redrive_list;
85 static DEFINE_SPINLOCK(log_redrive_lock);
86
87
88 /*
89  *      log read/write serialization (per log)
90  */
91 #define LOG_LOCK_INIT(log)      mutex_init(&(log)->loglock)
92 #define LOG_LOCK(log)           mutex_lock(&((log)->loglock))
93 #define LOG_UNLOCK(log)         mutex_unlock(&((log)->loglock))
94
95
96 /*
97  *      log group commit serialization (per log)
98  */
99
100 #define LOGGC_LOCK_INIT(log)    spin_lock_init(&(log)->gclock)
101 #define LOGGC_LOCK(log)         spin_lock_irq(&(log)->gclock)
102 #define LOGGC_UNLOCK(log)       spin_unlock_irq(&(log)->gclock)
103 #define LOGGC_WAKEUP(tblk)      wake_up_all(&(tblk)->gcwait)
104
105 /*
106  *      log sync serialization (per log)
107  */
108 #define LOGSYNC_DELTA(logsize)          min((logsize)/8, 128*LOGPSIZE)
109 #define LOGSYNC_BARRIER(logsize)        ((logsize)/4)
110 /*
111 #define LOGSYNC_DELTA(logsize)          min((logsize)/4, 256*LOGPSIZE)
112 #define LOGSYNC_BARRIER(logsize)        ((logsize)/2)
113 */
114
115
116 /*
117  *      log buffer cache synchronization
118  */
119 static DEFINE_SPINLOCK(jfsLCacheLock);
120
121 #define LCACHE_LOCK(flags)      spin_lock_irqsave(&jfsLCacheLock, flags)
122 #define LCACHE_UNLOCK(flags)    spin_unlock_irqrestore(&jfsLCacheLock, flags)
123
124 /*
125  * See __SLEEP_COND in jfs_locks.h
126  */
127 #define LCACHE_SLEEP_COND(wq, cond, flags)      \
128 do {                                            \
129         if (cond)                               \
130                 break;                          \
131         __SLEEP_COND(wq, cond, LCACHE_LOCK(flags), LCACHE_UNLOCK(flags)); \
132 } while (0)
133
134 #define LCACHE_WAKEUP(event)    wake_up(event)
135
136
137 /*
138  *      lbuf buffer cache (lCache) control
139  */
140 /* log buffer manager pageout control (cumulative, inclusive) */
141 #define lbmREAD         0x0001
142 #define lbmWRITE        0x0002  /* enqueue at tail of write queue;
143                                  * init pageout if at head of queue;
144                                  */
145 #define lbmRELEASE      0x0004  /* remove from write queue
146                                  * at completion of pageout;
147                                  * do not free/recycle it yet:
148                                  * caller will free it;
149                                  */
150 #define lbmSYNC         0x0008  /* do not return to freelist
151                                  * when removed from write queue;
152                                  */
153 #define lbmFREE         0x0010  /* return to freelist
154                                  * at completion of pageout;
155                                  * the buffer may be recycled;
156                                  */
157 #define lbmDONE         0x0020
158 #define lbmERROR        0x0040
159 #define lbmGC           0x0080  /* lbmIODone to perform post-GC processing
160                                  * of log page
161                                  */
162 #define lbmDIRECT       0x0100
163
164 /*
165  * Global list of active external journals
166  */
167 static LIST_HEAD(jfs_external_logs);
168 static struct jfs_log *dummy_log = NULL;
169 static DEFINE_MUTEX(jfs_log_mutex);
170
171 /*
172  * forward references
173  */
174 static int lmWriteRecord(struct jfs_log * log, struct tblock * tblk,
175                          struct lrd * lrd, struct tlock * tlck);
176
177 static int lmNextPage(struct jfs_log * log);
178 static int lmLogFileSystem(struct jfs_log * log, struct jfs_sb_info *sbi,
179                            int activate);
180
181 static int open_inline_log(struct super_block *sb);
182 static int open_dummy_log(struct super_block *sb);
183 static int lbmLogInit(struct jfs_log * log);
184 static void lbmLogShutdown(struct jfs_log * log);
185 static struct lbuf *lbmAllocate(struct jfs_log * log, int);
186 static void lbmFree(struct lbuf * bp);
187 static void lbmfree(struct lbuf * bp);
188 static int lbmRead(struct jfs_log * log, int pn, struct lbuf ** bpp);
189 static void lbmWrite(struct jfs_log * log, struct lbuf * bp, int flag, int cant_block);
190 static void lbmDirectWrite(struct jfs_log * log, struct lbuf * bp, int flag);
191 static int lbmIOWait(struct lbuf * bp, int flag);
192 static bio_end_io_t lbmIODone;
193 static void lbmStartIO(struct lbuf * bp);
194 static void lmGCwrite(struct jfs_log * log, int cant_block);
195 static int lmLogSync(struct jfs_log * log, int hard_sync);
196
197
198
199 /*
200  *      statistics
201  */
202 #ifdef CONFIG_JFS_STATISTICS
203 static struct lmStat {
204         uint commit;            /* # of commit */
205         uint pagedone;          /* # of page written */
206         uint submitted;         /* # of pages submitted */
207         uint full_page;         /* # of full pages submitted */
208         uint partial_page;      /* # of partial pages submitted */
209 } lmStat;
210 #endif
211
212
213 /*
214  * NAME:        lmLog()
215  *
216  * FUNCTION:    write a log record;
217  *
218  * PARAMETER:
219  *
220  * RETURN:      lsn - offset to the next log record to write (end-of-log);
221  *              -1  - error;
222  *
223  * note: todo: log error handler
224  */
225 int lmLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
226           struct tlock * tlck)
227 {
228         int lsn;
229         int diffp, difft;
230         struct metapage *mp = NULL;
231         unsigned long flags;
232
233         jfs_info("lmLog: log:0x%p tblk:0x%p, lrd:0x%p tlck:0x%p",
234                  log, tblk, lrd, tlck);
235
236         LOG_LOCK(log);
237
238         /* log by (out-of-transaction) JFS ? */
239         if (tblk == NULL)
240                 goto writeRecord;
241
242         /* log from page ? */
243         if (tlck == NULL ||
244             tlck->type & tlckBTROOT || (mp = tlck->mp) == NULL)
245                 goto writeRecord;
246
247         /*
248          *      initialize/update page/transaction recovery lsn
249          */
250         lsn = log->lsn;
251
252         LOGSYNC_LOCK(log, flags);
253
254         /*
255          * initialize page lsn if first log write of the page
256          */
257         if (mp->lsn == 0) {
258                 mp->log = log;
259                 mp->lsn = lsn;
260                 log->count++;
261
262                 /* insert page at tail of logsynclist */
263                 list_add_tail(&mp->synclist, &log->synclist);
264         }
265
266         /*
267          *      initialize/update lsn of tblock of the page
268          *
269          * transaction inherits oldest lsn of pages associated
270          * with allocation/deallocation of resources (their
271          * log records are used to reconstruct allocation map
272          * at recovery time: inode for inode allocation map,
273          * B+-tree index of extent descriptors for block
274          * allocation map);
275          * allocation map pages inherit transaction lsn at
276          * commit time to allow forwarding log syncpt past log
277          * records associated with allocation/deallocation of
278          * resources only after persistent map of these map pages
279          * have been updated and propagated to home.
280          */
281         /*
282          * initialize transaction lsn:
283          */
284         if (tblk->lsn == 0) {
285                 /* inherit lsn of its first page logged */
286                 tblk->lsn = mp->lsn;
287                 log->count++;
288
289                 /* insert tblock after the page on logsynclist */
290                 list_add(&tblk->synclist, &mp->synclist);
291         }
292         /*
293          * update transaction lsn:
294          */
295         else {
296                 /* inherit oldest/smallest lsn of page */
297                 logdiff(diffp, mp->lsn, log);
298                 logdiff(difft, tblk->lsn, log);
299                 if (diffp < difft) {
300                         /* update tblock lsn with page lsn */
301                         tblk->lsn = mp->lsn;
302
303                         /* move tblock after page on logsynclist */
304                         list_move(&tblk->synclist, &mp->synclist);
305                 }
306         }
307
308         LOGSYNC_UNLOCK(log, flags);
309
310         /*
311          *      write the log record
312          */
313       writeRecord:
314         lsn = lmWriteRecord(log, tblk, lrd, tlck);
315
316         /*
317          * forward log syncpt if log reached next syncpt trigger
318          */
319         logdiff(diffp, lsn, log);
320         if (diffp >= log->nextsync)
321                 lsn = lmLogSync(log, 0);
322
323         /* update end-of-log lsn */
324         log->lsn = lsn;
325
326         LOG_UNLOCK(log);
327
328         /* return end-of-log address */
329         return lsn;
330 }
331
332 /*
333  * NAME:        lmWriteRecord()
334  *
335  * FUNCTION:    move the log record to current log page
336  *
337  * PARAMETER:   cd      - commit descriptor
338  *
339  * RETURN:      end-of-log address
340  *                      
341  * serialization: LOG_LOCK() held on entry/exit
342  */
343 static int
344 lmWriteRecord(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
345               struct tlock * tlck)
346 {
347         int lsn = 0;            /* end-of-log address */
348         struct lbuf *bp;        /* dst log page buffer */
349         struct logpage *lp;     /* dst log page */
350         caddr_t dst;            /* destination address in log page */
351         int dstoffset;          /* end-of-log offset in log page */
352         int freespace;          /* free space in log page */
353         caddr_t p;              /* src meta-data page */
354         caddr_t src;
355         int srclen;
356         int nbytes;             /* number of bytes to move */
357         int i;
358         int len;
359         struct linelock *linelock;
360         struct lv *lv;
361         struct lvd *lvd;
362         int l2linesize;
363
364         len = 0;
365
366         /* retrieve destination log page to write */
367         bp = (struct lbuf *) log->bp;
368         lp = (struct logpage *) bp->l_ldata;
369         dstoffset = log->eor;
370
371         /* any log data to write ? */
372         if (tlck == NULL)
373                 goto moveLrd;
374
375         /*
376          *      move log record data
377          */
378         /* retrieve source meta-data page to log */
379         if (tlck->flag & tlckPAGELOCK) {
380                 p = (caddr_t) (tlck->mp->data);
381                 linelock = (struct linelock *) & tlck->lock;
382         }
383         /* retrieve source in-memory inode to log */
384         else if (tlck->flag & tlckINODELOCK) {
385                 if (tlck->type & tlckDTREE)
386                         p = (caddr_t) &JFS_IP(tlck->ip)->i_dtroot;
387                 else
388                         p = (caddr_t) &JFS_IP(tlck->ip)->i_xtroot;
389                 linelock = (struct linelock *) & tlck->lock;
390         }
391 #ifdef  _JFS_WIP
392         else if (tlck->flag & tlckINLINELOCK) {
393
394                 inlinelock = (struct inlinelock *) & tlck;
395                 p = (caddr_t) & inlinelock->pxd;
396                 linelock = (struct linelock *) & tlck;
397         }
398 #endif                          /* _JFS_WIP */
399         else {
400                 jfs_err("lmWriteRecord: UFO tlck:0x%p", tlck);
401                 return 0;       /* Probably should trap */
402         }
403         l2linesize = linelock->l2linesize;
404
405       moveData:
406         ASSERT(linelock->index <= linelock->maxcnt);
407
408         lv = linelock->lv;
409         for (i = 0; i < linelock->index; i++, lv++) {
410                 if (lv->length == 0)
411                         continue;
412
413                 /* is page full ? */
414                 if (dstoffset >= LOGPSIZE - LOGPTLRSIZE) {
415                         /* page become full: move on to next page */
416                         lmNextPage(log);
417
418                         bp = log->bp;
419                         lp = (struct logpage *) bp->l_ldata;
420                         dstoffset = LOGPHDRSIZE;
421                 }
422
423                 /*
424                  * move log vector data
425                  */
426                 src = (u8 *) p + (lv->offset << l2linesize);
427                 srclen = lv->length << l2linesize;
428                 len += srclen;
429                 while (srclen > 0) {
430                         freespace = (LOGPSIZE - LOGPTLRSIZE) - dstoffset;
431                         nbytes = min(freespace, srclen);
432                         dst = (caddr_t) lp + dstoffset;
433                         memcpy(dst, src, nbytes);
434                         dstoffset += nbytes;
435
436                         /* is page not full ? */
437                         if (dstoffset < LOGPSIZE - LOGPTLRSIZE)
438                                 break;
439
440                         /* page become full: move on to next page */
441                         lmNextPage(log);
442
443                         bp = (struct lbuf *) log->bp;
444                         lp = (struct logpage *) bp->l_ldata;
445                         dstoffset = LOGPHDRSIZE;
446
447                         srclen -= nbytes;
448                         src += nbytes;
449                 }
450
451                 /*
452                  * move log vector descriptor
453                  */
454                 len += 4;
455                 lvd = (struct lvd *) ((caddr_t) lp + dstoffset);
456                 lvd->offset = cpu_to_le16(lv->offset);
457                 lvd->length = cpu_to_le16(lv->length);
458                 dstoffset += 4;
459                 jfs_info("lmWriteRecord: lv offset:%d length:%d",
460                          lv->offset, lv->length);
461         }
462
463         if ((i = linelock->next)) {
464                 linelock = (struct linelock *) lid_to_tlock(i);
465                 goto moveData;
466         }
467
468         /*
469          *      move log record descriptor
470          */
471       moveLrd:
472         lrd->length = cpu_to_le16(len);
473
474         src = (caddr_t) lrd;
475         srclen = LOGRDSIZE;
476
477         while (srclen > 0) {
478                 freespace = (LOGPSIZE - LOGPTLRSIZE) - dstoffset;
479                 nbytes = min(freespace, srclen);
480                 dst = (caddr_t) lp + dstoffset;
481                 memcpy(dst, src, nbytes);
482
483                 dstoffset += nbytes;
484                 srclen -= nbytes;
485
486                 /* are there more to move than freespace of page ? */
487                 if (srclen)
488                         goto pageFull;
489
490                 /*
491                  * end of log record descriptor
492                  */
493
494                 /* update last log record eor */
495                 log->eor = dstoffset;
496                 bp->l_eor = dstoffset;
497                 lsn = (log->page << L2LOGPSIZE) + dstoffset;
498
499                 if (lrd->type & cpu_to_le16(LOG_COMMIT)) {
500                         tblk->clsn = lsn;
501                         jfs_info("wr: tclsn:0x%x, beor:0x%x", tblk->clsn,
502                                  bp->l_eor);
503
504                         INCREMENT(lmStat.commit);       /* # of commit */
505
506                         /*
507                          * enqueue tblock for group commit:
508                          *
509                          * enqueue tblock of non-trivial/synchronous COMMIT
510                          * at tail of group commit queue
511                          * (trivial/asynchronous COMMITs are ignored by
512                          * group commit.)
513                          */
514                         LOGGC_LOCK(log);
515
516                         /* init tblock gc state */
517                         tblk->flag = tblkGC_QUEUE;
518                         tblk->bp = log->bp;
519                         tblk->pn = log->page;
520                         tblk->eor = log->eor;
521
522                         /* enqueue transaction to commit queue */
523                         list_add_tail(&tblk->cqueue, &log->cqueue);
524
525                         LOGGC_UNLOCK(log);
526                 }
527
528                 jfs_info("lmWriteRecord: lrd:0x%04x bp:0x%p pn:%d eor:0x%x",
529                         le16_to_cpu(lrd->type), log->bp, log->page, dstoffset);
530
531                 /* page not full ? */
532                 if (dstoffset < LOGPSIZE - LOGPTLRSIZE)
533                         return lsn;
534
535               pageFull:
536                 /* page become full: move on to next page */
537                 lmNextPage(log);
538
539                 bp = (struct lbuf *) log->bp;
540                 lp = (struct logpage *) bp->l_ldata;
541                 dstoffset = LOGPHDRSIZE;
542                 src += nbytes;
543         }
544
545         return lsn;
546 }
547
548
549 /*
550  * NAME:        lmNextPage()
551  *
552  * FUNCTION:    write current page and allocate next page.
553  *
554  * PARAMETER:   log
555  *
556  * RETURN:      0
557  *                      
558  * serialization: LOG_LOCK() held on entry/exit
559  */
560 static int lmNextPage(struct jfs_log * log)
561 {
562         struct logpage *lp;
563         int lspn;               /* log sequence page number */
564         int pn;                 /* current page number */
565         struct lbuf *bp;
566         struct lbuf *nextbp;
567         struct tblock *tblk;
568
569         /* get current log page number and log sequence page number */
570         pn = log->page;
571         bp = log->bp;
572         lp = (struct logpage *) bp->l_ldata;
573         lspn = le32_to_cpu(lp->h.page);
574
575         LOGGC_LOCK(log);
576
577         /*
578          *      write or queue the full page at the tail of write queue
579          */
580         /* get the tail tblk on commit queue */
581         if (list_empty(&log->cqueue))
582                 tblk = NULL;
583         else
584                 tblk = list_entry(log->cqueue.prev, struct tblock, cqueue);
585
586         /* every tblk who has COMMIT record on the current page,
587          * and has not been committed, must be on commit queue
588          * since tblk is queued at commit queueu at the time
589          * of writing its COMMIT record on the page before
590          * page becomes full (even though the tblk thread
591          * who wrote COMMIT record may have been suspended
592          * currently);
593          */
594
595         /* is page bound with outstanding tail tblk ? */
596         if (tblk && tblk->pn == pn) {
597                 /* mark tblk for end-of-page */
598                 tblk->flag |= tblkGC_EOP;
599
600                 if (log->cflag & logGC_PAGEOUT) {
601                         /* if page is not already on write queue,
602                          * just enqueue (no lbmWRITE to prevent redrive)
603                          * buffer to wqueue to ensure correct serial order
604                          * of the pages since log pages will be added
605                          * continuously
606                          */
607                         if (bp->l_wqnext == NULL)
608                                 lbmWrite(log, bp, 0, 0);
609                 } else {
610                         /*
611                          * No current GC leader, initiate group commit
612                          */
613                         log->cflag |= logGC_PAGEOUT;
614                         lmGCwrite(log, 0);
615                 }
616         }
617         /* page is not bound with outstanding tblk:
618          * init write or mark it to be redriven (lbmWRITE)
619          */
620         else {
621                 /* finalize the page */
622                 bp->l_ceor = bp->l_eor;
623                 lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_ceor);
624                 lbmWrite(log, bp, lbmWRITE | lbmRELEASE | lbmFREE, 0);
625         }
626         LOGGC_UNLOCK(log);
627
628         /*
629          *      allocate/initialize next page
630          */
631         /* if log wraps, the first data page of log is 2
632          * (0 never used, 1 is superblock).
633          */
634         log->page = (pn == log->size - 1) ? 2 : pn + 1;
635         log->eor = LOGPHDRSIZE; /* ? valid page empty/full at logRedo() */
636
637         /* allocate/initialize next log page buffer */
638         nextbp = lbmAllocate(log, log->page);
639         nextbp->l_eor = log->eor;
640         log->bp = nextbp;
641
642         /* initialize next log page */
643         lp = (struct logpage *) nextbp->l_ldata;
644         lp->h.page = lp->t.page = cpu_to_le32(lspn + 1);
645         lp->h.eor = lp->t.eor = cpu_to_le16(LOGPHDRSIZE);
646
647         return 0;
648 }
649
650
651 /*
652  * NAME:        lmGroupCommit()
653  *
654  * FUNCTION:    group commit
655  *      initiate pageout of the pages with COMMIT in the order of
656  *      page number - redrive pageout of the page at the head of
657  *      pageout queue until full page has been written.
658  *
659  * RETURN:      
660  *
661  * NOTE:
662  *      LOGGC_LOCK serializes log group commit queue, and
663  *      transaction blocks on the commit queue.
664  *      N.B. LOG_LOCK is NOT held during lmGroupCommit().
665  */
666 int lmGroupCommit(struct jfs_log * log, struct tblock * tblk)
667 {
668         int rc = 0;
669
670         LOGGC_LOCK(log);
671
672         /* group committed already ? */
673         if (tblk->flag & tblkGC_COMMITTED) {
674                 if (tblk->flag & tblkGC_ERROR)
675                         rc = -EIO;
676
677                 LOGGC_UNLOCK(log);
678                 return rc;
679         }
680         jfs_info("lmGroup Commit: tblk = 0x%p, gcrtc = %d", tblk, log->gcrtc);
681
682         if (tblk->xflag & COMMIT_LAZY)
683                 tblk->flag |= tblkGC_LAZY;
684
685         if ((!(log->cflag & logGC_PAGEOUT)) && (!list_empty(&log->cqueue)) &&
686             (!(tblk->xflag & COMMIT_LAZY) || test_bit(log_FLUSH, &log->flag)
687              || jfs_tlocks_low)) {
688                 /*
689                  * No pageout in progress
690                  *
691                  * start group commit as its group leader.
692                  */
693                 log->cflag |= logGC_PAGEOUT;
694
695                 lmGCwrite(log, 0);
696         }
697
698         if (tblk->xflag & COMMIT_LAZY) {
699                 /*
700                  * Lazy transactions can leave now
701                  */
702                 LOGGC_UNLOCK(log);
703                 return 0;
704         }
705
706         /* lmGCwrite gives up LOGGC_LOCK, check again */
707
708         if (tblk->flag & tblkGC_COMMITTED) {
709                 if (tblk->flag & tblkGC_ERROR)
710                         rc = -EIO;
711
712                 LOGGC_UNLOCK(log);
713                 return rc;
714         }
715
716         /* upcount transaction waiting for completion
717          */
718         log->gcrtc++;
719         tblk->flag |= tblkGC_READY;
720
721         __SLEEP_COND(tblk->gcwait, (tblk->flag & tblkGC_COMMITTED),
722                      LOGGC_LOCK(log), LOGGC_UNLOCK(log));
723
724         /* removed from commit queue */
725         if (tblk->flag & tblkGC_ERROR)
726                 rc = -EIO;
727
728         LOGGC_UNLOCK(log);
729         return rc;
730 }
731
732 /*
733  * NAME:        lmGCwrite()
734  *
735  * FUNCTION:    group commit write
736  *      initiate write of log page, building a group of all transactions
737  *      with commit records on that page.
738  *
739  * RETURN:      None
740  *
741  * NOTE:
742  *      LOGGC_LOCK must be held by caller.
743  *      N.B. LOG_LOCK is NOT held during lmGroupCommit().
744  */
745 static void lmGCwrite(struct jfs_log * log, int cant_write)
746 {
747         struct lbuf *bp;
748         struct logpage *lp;
749         int gcpn;               /* group commit page number */
750         struct tblock *tblk;
751         struct tblock *xtblk = NULL;
752
753         /*
754          * build the commit group of a log page
755          *
756          * scan commit queue and make a commit group of all
757          * transactions with COMMIT records on the same log page.
758          */
759         /* get the head tblk on the commit queue */
760         gcpn = list_entry(log->cqueue.next, struct tblock, cqueue)->pn;
761
762         list_for_each_entry(tblk, &log->cqueue, cqueue) {
763                 if (tblk->pn != gcpn)
764                         break;
765
766                 xtblk = tblk;
767
768                 /* state transition: (QUEUE, READY) -> COMMIT */
769                 tblk->flag |= tblkGC_COMMIT;
770         }
771         tblk = xtblk;           /* last tblk of the page */
772
773         /*
774          * pageout to commit transactions on the log page.
775          */
776         bp = (struct lbuf *) tblk->bp;
777         lp = (struct logpage *) bp->l_ldata;
778         /* is page already full ? */
779         if (tblk->flag & tblkGC_EOP) {
780                 /* mark page to free at end of group commit of the page */
781                 tblk->flag &= ~tblkGC_EOP;
782                 tblk->flag |= tblkGC_FREE;
783                 bp->l_ceor = bp->l_eor;
784                 lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_ceor);
785                 lbmWrite(log, bp, lbmWRITE | lbmRELEASE | lbmGC,
786                          cant_write);
787                 INCREMENT(lmStat.full_page);
788         }
789         /* page is not yet full */
790         else {
791                 bp->l_ceor = tblk->eor; /* ? bp->l_ceor = bp->l_eor; */
792                 lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_ceor);
793                 lbmWrite(log, bp, lbmWRITE | lbmGC, cant_write);
794                 INCREMENT(lmStat.partial_page);
795         }
796 }
797
798 /*
799  * NAME:        lmPostGC()
800  *
801  * FUNCTION:    group commit post-processing
802  *      Processes transactions after their commit records have been written
803  *      to disk, redriving log I/O if necessary.
804  *
805  * RETURN:      None
806  *
807  * NOTE:
808  *      This routine is called a interrupt time by lbmIODone
809  */
810 static void lmPostGC(struct lbuf * bp)
811 {
812         unsigned long flags;
813         struct jfs_log *log = bp->l_log;
814         struct logpage *lp;
815         struct tblock *tblk, *temp;
816
817         //LOGGC_LOCK(log);
818         spin_lock_irqsave(&log->gclock, flags);
819         /*
820          * current pageout of group commit completed.
821          *
822          * remove/wakeup transactions from commit queue who were
823          * group committed with the current log page
824          */
825         list_for_each_entry_safe(tblk, temp, &log->cqueue, cqueue) {
826                 if (!(tblk->flag & tblkGC_COMMIT))
827                         break;
828                 /* if transaction was marked GC_COMMIT then
829                  * it has been shipped in the current pageout
830                  * and made it to disk - it is committed.
831                  */
832
833                 if (bp->l_flag & lbmERROR)
834                         tblk->flag |= tblkGC_ERROR;
835
836                 /* remove it from the commit queue */
837                 list_del(&tblk->cqueue);
838                 tblk->flag &= ~tblkGC_QUEUE;
839
840                 if (tblk == log->flush_tblk) {
841                         /* we can stop flushing the log now */
842                         clear_bit(log_FLUSH, &log->flag);
843                         log->flush_tblk = NULL;
844                 }
845
846                 jfs_info("lmPostGC: tblk = 0x%p, flag = 0x%x", tblk,
847                          tblk->flag);
848
849                 if (!(tblk->xflag & COMMIT_FORCE))
850                         /*
851                          * Hand tblk over to lazy commit thread
852                          */
853                         txLazyUnlock(tblk);
854                 else {
855                         /* state transition: COMMIT -> COMMITTED */
856                         tblk->flag |= tblkGC_COMMITTED;
857
858                         if (tblk->flag & tblkGC_READY)
859                                 log->gcrtc--;
860
861                         LOGGC_WAKEUP(tblk);
862                 }
863
864                 /* was page full before pageout ?
865                  * (and this is the last tblk bound with the page)
866                  */
867                 if (tblk->flag & tblkGC_FREE)
868                         lbmFree(bp);
869                 /* did page become full after pageout ?
870                  * (and this is the last tblk bound with the page)
871                  */
872                 else if (tblk->flag & tblkGC_EOP) {
873                         /* finalize the page */
874                         lp = (struct logpage *) bp->l_ldata;
875                         bp->l_ceor = bp->l_eor;
876                         lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_eor);
877                         jfs_info("lmPostGC: calling lbmWrite");
878                         lbmWrite(log, bp, lbmWRITE | lbmRELEASE | lbmFREE,
879                                  1);
880                 }
881
882         }
883
884         /* are there any transactions who have entered lnGroupCommit()
885          * (whose COMMITs are after that of the last log page written.
886          * They are waiting for new group commit (above at (SLEEP 1))
887          * or lazy transactions are on a full (queued) log page,
888          * select the latest ready transaction as new group leader and
889          * wake her up to lead her group.
890          */
891         if ((!list_empty(&log->cqueue)) &&
892             ((log->gcrtc > 0) || (tblk->bp->l_wqnext != NULL) ||
893              test_bit(log_FLUSH, &log->flag) || jfs_tlocks_low))
894                 /*
895                  * Call lmGCwrite with new group leader
896                  */
897                 lmGCwrite(log, 1);
898
899         /* no transaction are ready yet (transactions are only just
900          * queued (GC_QUEUE) and not entered for group commit yet).
901          * the first transaction entering group commit
902          * will elect herself as new group leader.
903          */
904         else
905                 log->cflag &= ~logGC_PAGEOUT;
906
907         //LOGGC_UNLOCK(log);
908         spin_unlock_irqrestore(&log->gclock, flags);
909         return;
910 }
911
912 /*
913  * NAME:        lmLogSync()
914  *
915  * FUNCTION:    write log SYNCPT record for specified log
916  *      if new sync address is available
917  *      (normally the case if sync() is executed by back-ground
918  *      process).
919  *      calculate new value of i_nextsync which determines when
920  *      this code is called again.
921  *
922  * PARAMETERS:  log     - log structure
923  *              hard_sync - 1 to force all metadata to be written
924  *
925  * RETURN:      0
926  *                      
927  * serialization: LOG_LOCK() held on entry/exit
928  */
929 static int lmLogSync(struct jfs_log * log, int hard_sync)
930 {
931         int logsize;
932         int written;            /* written since last syncpt */
933         int free;               /* free space left available */
934         int delta;              /* additional delta to write normally */
935         int more;               /* additional write granted */
936         struct lrd lrd;
937         int lsn;
938         struct logsyncblk *lp;
939         struct jfs_sb_info *sbi;
940         unsigned long flags;
941
942         /* push dirty metapages out to disk */
943         if (hard_sync)
944                 list_for_each_entry(sbi, &log->sb_list, log_list) {
945                         filemap_fdatawrite(sbi->ipbmap->i_mapping);
946                         filemap_fdatawrite(sbi->ipimap->i_mapping);
947                         filemap_fdatawrite(sbi->direct_inode->i_mapping);
948                 }
949         else
950                 list_for_each_entry(sbi, &log->sb_list, log_list) {
951                         filemap_flush(sbi->ipbmap->i_mapping);
952                         filemap_flush(sbi->ipimap->i_mapping);
953                         filemap_flush(sbi->direct_inode->i_mapping);
954                 }
955
956         /*
957          *      forward syncpt
958          */
959         /* if last sync is same as last syncpt,
960          * invoke sync point forward processing to update sync.
961          */
962
963         if (log->sync == log->syncpt) {
964                 LOGSYNC_LOCK(log, flags);
965                 if (list_empty(&log->synclist))
966                         log->sync = log->lsn;
967                 else {
968                         lp = list_entry(log->synclist.next,
969                                         struct logsyncblk, synclist);
970                         log->sync = lp->lsn;
971                 }
972                 LOGSYNC_UNLOCK(log, flags);
973
974         }
975
976         /* if sync is different from last syncpt,
977          * write a SYNCPT record with syncpt = sync.
978          * reset syncpt = sync
979          */
980         if (log->sync != log->syncpt) {
981                 lrd.logtid = 0;
982                 lrd.backchain = 0;
983                 lrd.type = cpu_to_le16(LOG_SYNCPT);
984                 lrd.length = 0;
985                 lrd.log.syncpt.sync = cpu_to_le32(log->sync);
986                 lsn = lmWriteRecord(log, NULL, &lrd, NULL);
987
988                 log->syncpt = log->sync;
989         } else
990                 lsn = log->lsn;
991
992         /*
993          *      setup next syncpt trigger (SWAG)
994          */
995         logsize = log->logsize;
996
997         logdiff(written, lsn, log);
998         free = logsize - written;
999         delta = LOGSYNC_DELTA(logsize);
1000         more = min(free / 2, delta);
1001         if (more < 2 * LOGPSIZE) {
1002                 jfs_warn("\n ... Log Wrap ... Log Wrap ... Log Wrap ...\n");
1003                 /*
1004                  *      log wrapping
1005                  *
1006                  * option 1 - panic ? No.!
1007                  * option 2 - shutdown file systems
1008                  *            associated with log ?
1009                  * option 3 - extend log ?
1010                  */
1011                 /*
1012                  * option 4 - second chance
1013                  *
1014                  * mark log wrapped, and continue.
1015                  * when all active transactions are completed,
1016                  * mark log vaild for recovery.
1017                  * if crashed during invalid state, log state
1018                  * implies invald log, forcing fsck().
1019                  */
1020                 /* mark log state log wrap in log superblock */
1021                 /* log->state = LOGWRAP; */
1022
1023                 /* reset sync point computation */
1024                 log->syncpt = log->sync = lsn;
1025                 log->nextsync = delta;
1026         } else
1027                 /* next syncpt trigger = written + more */
1028                 log->nextsync = written + more;
1029
1030         /* if number of bytes written from last sync point is more
1031          * than 1/4 of the log size, stop new transactions from
1032          * starting until all current transactions are completed
1033          * by setting syncbarrier flag.
1034          */
1035         if (!test_bit(log_SYNCBARRIER, &log->flag) &&
1036             (written > LOGSYNC_BARRIER(logsize)) && log->active) {
1037                 set_bit(log_SYNCBARRIER, &log->flag);
1038                 jfs_info("log barrier on: lsn=0x%x syncpt=0x%x", lsn,
1039                          log->syncpt);
1040                 /*
1041                  * We may have to initiate group commit
1042                  */
1043                 jfs_flush_journal(log, 0);
1044         }
1045
1046         return lsn;
1047 }
1048
1049 /*
1050  * NAME:        jfs_syncpt
1051  *
1052  * FUNCTION:    write log SYNCPT record for specified log
1053  *
1054  * PARAMETERS:  log       - log structure
1055  *              hard_sync - set to 1 to force metadata to be written
1056  */
1057 void jfs_syncpt(struct jfs_log *log, int hard_sync)
1058 {       LOG_LOCK(log);
1059         lmLogSync(log, hard_sync);
1060         LOG_UNLOCK(log);
1061 }
1062
1063 /*
1064  * NAME:        lmLogOpen()
1065  *
1066  * FUNCTION:    open the log on first open;
1067  *      insert filesystem in the active list of the log.
1068  *
1069  * PARAMETER:   ipmnt   - file system mount inode
1070  *              iplog   - log inode (out)
1071  *
1072  * RETURN:
1073  *
1074  * serialization:
1075  */
1076 int lmLogOpen(struct super_block *sb)
1077 {
1078         int rc;
1079         struct block_device *bdev;
1080         struct jfs_log *log;
1081         struct jfs_sb_info *sbi = JFS_SBI(sb);
1082
1083         if (sbi->flag & JFS_NOINTEGRITY)
1084                 return open_dummy_log(sb);
1085         
1086         if (sbi->mntflag & JFS_INLINELOG)
1087                 return open_inline_log(sb);
1088
1089         mutex_lock(&jfs_log_mutex);
1090         list_for_each_entry(log, &jfs_external_logs, journal_list) {
1091                 if (log->bdev->bd_dev == sbi->logdev) {
1092                         if (memcmp(log->uuid, sbi->loguuid,
1093                                    sizeof(log->uuid))) {
1094                                 jfs_warn("wrong uuid on JFS journal\n");
1095                                 mutex_unlock(&jfs_log_mutex);
1096                                 return -EINVAL;
1097                         }
1098                         /*
1099                          * add file system to log active file system list
1100                          */
1101                         if ((rc = lmLogFileSystem(log, sbi, 1))) {
1102                                 mutex_unlock(&jfs_log_mutex);
1103                                 return rc;
1104                         }
1105                         goto journal_found;
1106                 }
1107         }
1108
1109         if (!(log = kzalloc(sizeof(struct jfs_log), GFP_KERNEL))) {
1110                 mutex_unlock(&jfs_log_mutex);
1111                 return -ENOMEM;
1112         }
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         mutex_unlock(&jfs_log_mutex);
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         mutex_unlock(&jfs_log_mutex);
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 = kzalloc(sizeof(struct jfs_log), GFP_KERNEL)))
1185                 return -ENOMEM;
1186         INIT_LIST_HEAD(&log->sb_list);
1187         init_waitqueue_head(&log->syncwait);
1188
1189         set_bit(log_INLINELOG, &log->flag);
1190         log->bdev = sb->s_bdev;
1191         log->base = addressPXD(&JFS_SBI(sb)->logpxd);
1192         log->size = lengthPXD(&JFS_SBI(sb)->logpxd) >>
1193             (L2LOGPSIZE - sb->s_blocksize_bits);
1194         log->l2bsize = sb->s_blocksize_bits;
1195         ASSERT(L2LOGPSIZE >= sb->s_blocksize_bits);
1196
1197         /*
1198          * initialize log.
1199          */
1200         if ((rc = lmLogInit(log))) {
1201                 kfree(log);
1202                 jfs_warn("lmLogOpen: exit(%d)", rc);
1203                 return rc;
1204         }
1205
1206         list_add(&JFS_SBI(sb)->log_list, &log->sb_list);
1207         JFS_SBI(sb)->log = log;
1208
1209         return rc;
1210 }
1211
1212 static int open_dummy_log(struct super_block *sb)
1213 {
1214         int rc;
1215
1216         mutex_lock(&jfs_log_mutex);
1217         if (!dummy_log) {
1218                 dummy_log = kzalloc(sizeof(struct jfs_log), GFP_KERNEL);
1219                 if (!dummy_log) {
1220                         mutex_unlock(&jfs_log_mutex);
1221                         return -ENOMEM;
1222                 }
1223                 INIT_LIST_HEAD(&dummy_log->sb_list);
1224                 init_waitqueue_head(&dummy_log->syncwait);
1225                 dummy_log->no_integrity = 1;
1226                 /* Make up some stuff */
1227                 dummy_log->base = 0;
1228                 dummy_log->size = 1024;
1229                 rc = lmLogInit(dummy_log);
1230                 if (rc) {
1231                         kfree(dummy_log);
1232                         dummy_log = NULL;
1233                         mutex_unlock(&jfs_log_mutex);
1234                         return rc;
1235                 }
1236         }
1237
1238         LOG_LOCK(dummy_log);
1239         list_add(&JFS_SBI(sb)->log_list, &dummy_log->sb_list);
1240         JFS_SBI(sb)->log = dummy_log;
1241         LOG_UNLOCK(dummy_log);
1242         mutex_unlock(&jfs_log_mutex);
1243
1244         return 0;
1245 }
1246
1247 /*
1248  * NAME:        lmLogInit()
1249  *
1250  * FUNCTION:    log initialization at first log open.
1251  *
1252  *      logredo() (or logformat()) should have been run previously.
1253  *      initialize the log from log superblock.
1254  *      set the log state in the superblock to LOGMOUNT and
1255  *      write SYNCPT log record.
1256  *              
1257  * PARAMETER:   log     - log structure
1258  *
1259  * RETURN:      0       - if ok
1260  *              -EINVAL - bad log magic number or superblock dirty
1261  *              error returned from logwait()
1262  *                      
1263  * serialization: single first open thread
1264  */
1265 int lmLogInit(struct jfs_log * log)
1266 {
1267         int rc = 0;
1268         struct lrd lrd;
1269         struct logsuper *logsuper;
1270         struct lbuf *bpsuper;
1271         struct lbuf *bp;
1272         struct logpage *lp;
1273         int lsn = 0;
1274
1275         jfs_info("lmLogInit: log:0x%p", log);
1276
1277         /* initialize the group commit serialization lock */
1278         LOGGC_LOCK_INIT(log);
1279
1280         /* allocate/initialize the log write serialization lock */
1281         LOG_LOCK_INIT(log);
1282
1283         LOGSYNC_LOCK_INIT(log);
1284
1285         INIT_LIST_HEAD(&log->synclist);
1286
1287         INIT_LIST_HEAD(&log->cqueue);
1288         log->flush_tblk = NULL;
1289
1290         log->count = 0;
1291
1292         /*
1293          * initialize log i/o
1294          */
1295         if ((rc = lbmLogInit(log)))
1296                 return rc;
1297
1298         if (!test_bit(log_INLINELOG, &log->flag))
1299                 log->l2bsize = L2LOGPSIZE;
1300         
1301         /* check for disabled journaling to disk */
1302         if (log->no_integrity) {
1303                 /*
1304                  * Journal pages will still be filled.  When the time comes
1305                  * to actually do the I/O, the write is not done, and the
1306                  * endio routine is called directly.
1307                  */
1308                 bp = lbmAllocate(log , 0);
1309                 log->bp = bp;
1310                 bp->l_pn = bp->l_eor = 0;
1311         } else {
1312                 /*
1313                  * validate log superblock
1314                  */
1315                 if ((rc = lbmRead(log, 1, &bpsuper)))
1316                         goto errout10;
1317
1318                 logsuper = (struct logsuper *) bpsuper->l_ldata;
1319
1320                 if (logsuper->magic != cpu_to_le32(LOGMAGIC)) {
1321                         jfs_warn("*** Log Format Error ! ***");
1322                         rc = -EINVAL;
1323                         goto errout20;
1324                 }
1325
1326                 /* logredo() should have been run successfully. */
1327                 if (logsuper->state != cpu_to_le32(LOGREDONE)) {
1328                         jfs_warn("*** Log Is Dirty ! ***");
1329                         rc = -EINVAL;
1330                         goto errout20;
1331                 }
1332
1333                 /* initialize log from log superblock */
1334                 if (test_bit(log_INLINELOG,&log->flag)) {
1335                         if (log->size != le32_to_cpu(logsuper->size)) {
1336                                 rc = -EINVAL;
1337                                 goto errout20;
1338                         }
1339                         jfs_info("lmLogInit: inline log:0x%p base:0x%Lx "
1340                                  "size:0x%x", log,
1341                                  (unsigned long long) log->base, log->size);
1342                 } else {
1343                         if (memcmp(logsuper->uuid, log->uuid, 16)) {
1344                                 jfs_warn("wrong uuid on JFS log device");
1345                                 goto errout20;
1346                         }
1347                         log->size = le32_to_cpu(logsuper->size);
1348                         log->l2bsize = le32_to_cpu(logsuper->l2bsize);
1349                         jfs_info("lmLogInit: external log:0x%p base:0x%Lx "
1350                                  "size:0x%x", log,
1351                                  (unsigned long long) log->base, log->size);
1352                 }
1353
1354                 log->page = le32_to_cpu(logsuper->end) / LOGPSIZE;
1355                 log->eor = le32_to_cpu(logsuper->end) - (LOGPSIZE * log->page);
1356
1357                 /*
1358                  * initialize for log append write mode
1359                  */
1360                 /* establish current/end-of-log page/buffer */
1361                 if ((rc = lbmRead(log, log->page, &bp)))
1362                         goto errout20;
1363
1364                 lp = (struct logpage *) bp->l_ldata;
1365
1366                 jfs_info("lmLogInit: lsn:0x%x page:%d eor:%d:%d",
1367                          le32_to_cpu(logsuper->end), log->page, log->eor,
1368                          le16_to_cpu(lp->h.eor));
1369
1370                 log->bp = bp;
1371                 bp->l_pn = log->page;
1372                 bp->l_eor = log->eor;
1373
1374                 /* if current page is full, move on to next page */
1375                 if (log->eor >= LOGPSIZE - LOGPTLRSIZE)
1376                         lmNextPage(log);
1377
1378                 /*
1379                  * initialize log syncpoint
1380                  */
1381                 /*
1382                  * write the first SYNCPT record with syncpoint = 0
1383                  * (i.e., log redo up to HERE !);
1384                  * remove current page from lbm write queue at end of pageout
1385                  * (to write log superblock update), but do not release to
1386                  * freelist;
1387                  */
1388                 lrd.logtid = 0;
1389                 lrd.backchain = 0;
1390                 lrd.type = cpu_to_le16(LOG_SYNCPT);
1391                 lrd.length = 0;
1392                 lrd.log.syncpt.sync = 0;
1393                 lsn = lmWriteRecord(log, NULL, &lrd, NULL);
1394                 bp = log->bp;
1395                 bp->l_ceor = bp->l_eor;
1396                 lp = (struct logpage *) bp->l_ldata;
1397                 lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_eor);
1398                 lbmWrite(log, bp, lbmWRITE | lbmSYNC, 0);
1399                 if ((rc = lbmIOWait(bp, 0)))
1400                         goto errout30;
1401
1402                 /*
1403                  * update/write superblock
1404                  */
1405                 logsuper->state = cpu_to_le32(LOGMOUNT);
1406                 log->serial = le32_to_cpu(logsuper->serial) + 1;
1407                 logsuper->serial = cpu_to_le32(log->serial);
1408                 lbmDirectWrite(log, bpsuper, lbmWRITE | lbmRELEASE | lbmSYNC);
1409                 if ((rc = lbmIOWait(bpsuper, lbmFREE)))
1410                         goto errout30;
1411         }
1412
1413         /* initialize logsync parameters */
1414         log->logsize = (log->size - 2) << L2LOGPSIZE;
1415         log->lsn = lsn;
1416         log->syncpt = lsn;
1417         log->sync = log->syncpt;
1418         log->nextsync = LOGSYNC_DELTA(log->logsize);
1419
1420         jfs_info("lmLogInit: lsn:0x%x syncpt:0x%x sync:0x%x",
1421                  log->lsn, log->syncpt, log->sync);
1422
1423         /*
1424          * initialize for lazy/group commit
1425          */
1426         log->clsn = lsn;
1427
1428         return 0;
1429
1430         /*
1431          *      unwind on error
1432          */
1433       errout30:         /* release log page */
1434         log->wqueue = NULL;
1435         bp->l_wqnext = NULL;
1436         lbmFree(bp);
1437
1438       errout20:         /* release log superblock */
1439         lbmFree(bpsuper);
1440
1441       errout10:         /* unwind lbmLogInit() */
1442         lbmLogShutdown(log);
1443
1444         jfs_warn("lmLogInit: exit(%d)", rc);
1445         return rc;
1446 }
1447
1448
1449 /*
1450  * NAME:        lmLogClose()
1451  *
1452  * FUNCTION:    remove file system <ipmnt> from active list of log <iplog>
1453  *              and close it on last close.
1454  *
1455  * PARAMETER:   sb      - superblock
1456  *
1457  * RETURN:      errors from subroutines
1458  *
1459  * serialization:
1460  */
1461 int lmLogClose(struct super_block *sb)
1462 {
1463         struct jfs_sb_info *sbi = JFS_SBI(sb);
1464         struct jfs_log *log = sbi->log;
1465         struct block_device *bdev;
1466         int rc = 0;
1467
1468         jfs_info("lmLogClose: log:0x%p", log);
1469
1470         mutex_lock(&jfs_log_mutex);
1471         LOG_LOCK(log);
1472         list_del(&sbi->log_list);
1473         LOG_UNLOCK(log);
1474         sbi->log = NULL;
1475
1476         /*
1477          * We need to make sure all of the "written" metapages
1478          * actually make it to disk
1479          */
1480         sync_blockdev(sb->s_bdev);
1481
1482         if (test_bit(log_INLINELOG, &log->flag)) {
1483                 /*
1484                  *      in-line log in host file system
1485                  */
1486                 rc = lmLogShutdown(log);
1487                 kfree(log);
1488                 goto out;
1489         }
1490
1491         if (!log->no_integrity)
1492                 lmLogFileSystem(log, sbi, 0);
1493
1494         if (!list_empty(&log->sb_list))
1495                 goto out;
1496
1497         /*
1498          * TODO: ensure that the dummy_log is in a state to allow
1499          * lbmLogShutdown to deallocate all the buffers and call
1500          * kfree against dummy_log.  For now, leave dummy_log & its
1501          * buffers in memory, and resuse if another no-integrity mount
1502          * is requested.
1503          */
1504         if (log->no_integrity)
1505                 goto out;
1506
1507         /*
1508          *      external log as separate logical volume
1509          */
1510         list_del(&log->journal_list);
1511         bdev = log->bdev;
1512         rc = lmLogShutdown(log);
1513
1514         bd_release(bdev);
1515         blkdev_put(bdev);
1516
1517         kfree(log);
1518
1519       out:
1520         mutex_unlock(&jfs_log_mutex);
1521         jfs_info("lmLogClose: exit(%d)", rc);
1522         return rc;
1523 }
1524
1525
1526 /*
1527  * NAME:        jfs_flush_journal()
1528  *
1529  * FUNCTION:    initiate write of any outstanding transactions to the journal
1530  *              and optionally wait until they are all written to disk
1531  *
1532  *              wait == 0  flush until latest txn is committed, don't wait
1533  *              wait == 1  flush until latest txn is committed, wait
1534  *              wait > 1   flush until all txn's are complete, wait
1535  */
1536 void jfs_flush_journal(struct jfs_log *log, int wait)
1537 {
1538         int i;
1539         struct tblock *target = NULL;
1540         struct jfs_sb_info *sbi;
1541
1542         /* jfs_write_inode may call us during read-only mount */
1543         if (!log)
1544                 return;
1545
1546         jfs_info("jfs_flush_journal: log:0x%p wait=%d", log, wait);
1547
1548         LOGGC_LOCK(log);
1549
1550         if (!list_empty(&log->cqueue)) {
1551                 /*
1552                  * This ensures that we will keep writing to the journal as long
1553                  * as there are unwritten commit records
1554                  */
1555                 target = list_entry(log->cqueue.prev, struct tblock, cqueue);
1556
1557                 if (test_bit(log_FLUSH, &log->flag)) {
1558                         /*
1559                          * We're already flushing.
1560                          * if flush_tblk is NULL, we are flushing everything,
1561                          * so leave it that way.  Otherwise, update it to the
1562                          * latest transaction
1563                          */
1564                         if (log->flush_tblk)
1565                                 log->flush_tblk = target;
1566                 } else {
1567                         /* Only flush until latest transaction is committed */
1568                         log->flush_tblk = target;
1569                         set_bit(log_FLUSH, &log->flag);
1570
1571                         /*
1572                          * Initiate I/O on outstanding transactions
1573                          */
1574                         if (!(log->cflag & logGC_PAGEOUT)) {
1575                                 log->cflag |= logGC_PAGEOUT;
1576                                 lmGCwrite(log, 0);
1577                         }
1578                 }
1579         }
1580         if ((wait > 1) || test_bit(log_SYNCBARRIER, &log->flag)) {
1581                 /* Flush until all activity complete */
1582                 set_bit(log_FLUSH, &log->flag);
1583                 log->flush_tblk = NULL;
1584         }
1585
1586         if (wait && target && !(target->flag & tblkGC_COMMITTED)) {
1587                 DECLARE_WAITQUEUE(__wait, current);
1588
1589                 add_wait_queue(&target->gcwait, &__wait);
1590                 set_current_state(TASK_UNINTERRUPTIBLE);
1591                 LOGGC_UNLOCK(log);
1592                 schedule();
1593                 current->state = TASK_RUNNING;
1594                 LOGGC_LOCK(log);
1595                 remove_wait_queue(&target->gcwait, &__wait);
1596         }
1597         LOGGC_UNLOCK(log);
1598
1599         if (wait < 2)
1600                 return;
1601
1602         list_for_each_entry(sbi, &log->sb_list, log_list) {
1603                 filemap_fdatawrite(sbi->ipbmap->i_mapping);
1604                 filemap_fdatawrite(sbi->ipimap->i_mapping);
1605                 filemap_fdatawrite(sbi->direct_inode->i_mapping);
1606         }
1607
1608         /*
1609          * If there was recent activity, we may need to wait
1610          * for the lazycommit thread to catch up
1611          */
1612         if ((!list_empty(&log->cqueue)) || !list_empty(&log->synclist)) {
1613                 for (i = 0; i < 200; i++) {     /* Too much? */
1614                         msleep(250);
1615                         if (list_empty(&log->cqueue) &&
1616                             list_empty(&log->synclist))
1617                                 break;
1618                 }
1619         }
1620         assert(list_empty(&log->cqueue));
1621
1622 #ifdef CONFIG_JFS_DEBUG
1623         if (!list_empty(&log->synclist)) {
1624                 struct logsyncblk *lp;
1625
1626                 list_for_each_entry(lp, &log->synclist, synclist) {
1627                         if (lp->xflag & COMMIT_PAGE) {
1628                                 struct metapage *mp = (struct metapage *)lp;
1629                                 dump_mem("orphan metapage", lp,
1630                                          sizeof(struct metapage));
1631                                 dump_mem("page", mp->page, sizeof(struct page));
1632                         }
1633                         else
1634                                 dump_mem("orphan tblock", lp,
1635                                          sizeof(struct tblock));
1636                 }
1637         }
1638 #endif
1639         //assert(list_empty(&log->synclist));
1640         clear_bit(log_FLUSH, &log->flag);
1641 }
1642
1643 /*
1644  * NAME:        lmLogShutdown()
1645  *
1646  * FUNCTION:    log shutdown at last LogClose().
1647  *
1648  *              write log syncpt record.
1649  *              update super block to set redone flag to 0.
1650  *
1651  * PARAMETER:   log     - log inode
1652  *
1653  * RETURN:      0       - success
1654  *                      
1655  * serialization: single last close thread
1656  */
1657 int lmLogShutdown(struct jfs_log * log)
1658 {
1659         int rc;
1660         struct lrd lrd;
1661         int lsn;
1662         struct logsuper *logsuper;
1663         struct lbuf *bpsuper;
1664         struct lbuf *bp;
1665         struct logpage *lp;
1666
1667         jfs_info("lmLogShutdown: log:0x%p", log);
1668
1669         jfs_flush_journal(log, 2);
1670
1671         /*
1672          * write the last SYNCPT record with syncpoint = 0
1673          * (i.e., log redo up to HERE !)
1674          */
1675         lrd.logtid = 0;
1676         lrd.backchain = 0;
1677         lrd.type = cpu_to_le16(LOG_SYNCPT);
1678         lrd.length = 0;
1679         lrd.log.syncpt.sync = 0;
1680         
1681         lsn = lmWriteRecord(log, NULL, &lrd, NULL);
1682         bp = log->bp;
1683         lp = (struct logpage *) bp->l_ldata;
1684         lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_eor);
1685         lbmWrite(log, log->bp, lbmWRITE | lbmRELEASE | lbmSYNC, 0);
1686         lbmIOWait(log->bp, lbmFREE);
1687         log->bp = NULL;
1688
1689         /*
1690          * synchronous update log superblock
1691          * mark log state as shutdown cleanly
1692          * (i.e., Log does not need to be replayed).
1693          */
1694         if ((rc = lbmRead(log, 1, &bpsuper)))
1695                 goto out;
1696
1697         logsuper = (struct logsuper *) bpsuper->l_ldata;
1698         logsuper->state = cpu_to_le32(LOGREDONE);
1699         logsuper->end = cpu_to_le32(lsn);
1700         lbmDirectWrite(log, bpsuper, lbmWRITE | lbmRELEASE | lbmSYNC);
1701         rc = lbmIOWait(bpsuper, lbmFREE);
1702
1703         jfs_info("lmLogShutdown: lsn:0x%x page:%d eor:%d",
1704                  lsn, log->page, log->eor);
1705
1706       out:    
1707         /*
1708          * shutdown per log i/o
1709          */
1710         lbmLogShutdown(log);
1711
1712         if (rc) {
1713                 jfs_warn("lmLogShutdown: exit(%d)", rc);
1714         }
1715         return rc;
1716 }
1717
1718
1719 /*
1720  * NAME:        lmLogFileSystem()
1721  *
1722  * FUNCTION:    insert (<activate> = true)/remove (<activate> = false)
1723  *      file system into/from log active file system list.
1724  *
1725  * PARAMETE:    log     - pointer to logs inode.
1726  *              fsdev   - kdev_t of filesystem.
1727  *              serial  - pointer to returned log serial number
1728  *              activate - insert/remove device from active list.
1729  *
1730  * RETURN:      0       - success
1731  *              errors returned by vms_iowait().
1732  */
1733 static int lmLogFileSystem(struct jfs_log * log, struct jfs_sb_info *sbi,
1734                            int activate)
1735 {
1736         int rc = 0;
1737         int i;
1738         struct logsuper *logsuper;
1739         struct lbuf *bpsuper;
1740         char *uuid = sbi->uuid;
1741
1742         /*
1743          * insert/remove file system device to log active file system list.
1744          */
1745         if ((rc = lbmRead(log, 1, &bpsuper)))
1746                 return rc;
1747
1748         logsuper = (struct logsuper *) bpsuper->l_ldata;
1749         if (activate) {
1750                 for (i = 0; i < MAX_ACTIVE; i++)
1751                         if (!memcmp(logsuper->active[i].uuid, NULL_UUID, 16)) {
1752                                 memcpy(logsuper->active[i].uuid, uuid, 16);
1753                                 sbi->aggregate = i;
1754                                 break;
1755                         }
1756                 if (i == MAX_ACTIVE) {
1757                         jfs_warn("Too many file systems sharing journal!");
1758                         lbmFree(bpsuper);
1759                         return -EMFILE; /* Is there a better rc? */
1760                 }
1761         } else {
1762                 for (i = 0; i < MAX_ACTIVE; i++)
1763                         if (!memcmp(logsuper->active[i].uuid, uuid, 16)) {
1764                                 memcpy(logsuper->active[i].uuid, NULL_UUID, 16);
1765                                 break;
1766                         }
1767                 if (i == MAX_ACTIVE) {
1768                         jfs_warn("Somebody stomped on the journal!");
1769                         lbmFree(bpsuper);
1770                         return -EIO;
1771                 }
1772                 
1773         }
1774
1775         /*
1776          * synchronous write log superblock:
1777          *
1778          * write sidestream bypassing write queue:
1779          * at file system mount, log super block is updated for
1780          * activation of the file system before any log record
1781          * (MOUNT record) of the file system, and at file system
1782          * unmount, all meta data for the file system has been
1783          * flushed before log super block is updated for deactivation
1784          * of the file system.
1785          */
1786         lbmDirectWrite(log, bpsuper, lbmWRITE | lbmRELEASE | lbmSYNC);
1787         rc = lbmIOWait(bpsuper, lbmFREE);
1788
1789         return rc;
1790 }
1791
1792 /*
1793  *              log buffer manager (lbm)
1794  *              ------------------------
1795  *
1796  * special purpose buffer manager supporting log i/o requirements.
1797  *
1798  * per log write queue:
1799  * log pageout occurs in serial order by fifo write queue and
1800  * restricting to a single i/o in pregress at any one time.
1801  * a circular singly-linked list
1802  * (log->wrqueue points to the tail, and buffers are linked via
1803  * bp->wrqueue field), and
1804  * maintains log page in pageout ot waiting for pageout in serial pageout.
1805  */
1806
1807 /*
1808  *      lbmLogInit()
1809  *
1810  * initialize per log I/O setup at lmLogInit()
1811  */
1812 static int lbmLogInit(struct jfs_log * log)
1813 {                               /* log inode */
1814         int i;
1815         struct lbuf *lbuf;
1816
1817         jfs_info("lbmLogInit: log:0x%p", log);
1818
1819         /* initialize current buffer cursor */
1820         log->bp = NULL;
1821
1822         /* initialize log device write queue */
1823         log->wqueue = NULL;
1824
1825         /*
1826          * Each log has its own buffer pages allocated to it.  These are
1827          * not managed by the page cache.  This ensures that a transaction
1828          * writing to the log does not block trying to allocate a page from
1829          * the page cache (for the log).  This would be bad, since page
1830          * allocation waits on the kswapd thread that may be committing inodes
1831          * which would cause log activity.  Was that clear?  I'm trying to
1832          * avoid deadlock here.
1833          */
1834         init_waitqueue_head(&log->free_wait);
1835
1836         log->lbuf_free = NULL;
1837
1838         for (i = 0; i < LOGPAGES;) {
1839                 char *buffer;
1840                 uint offset;
1841                 struct page *page;
1842
1843                 buffer = (char *) get_zeroed_page(GFP_KERNEL);
1844                 if (buffer == NULL)
1845                         goto error;
1846                 page = virt_to_page(buffer);
1847                 for (offset = 0; offset < PAGE_SIZE; offset += LOGPSIZE) {
1848                         lbuf = kmalloc(sizeof(struct lbuf), GFP_KERNEL);
1849                         if (lbuf == NULL) {
1850                                 if (offset == 0)
1851                                         free_page((unsigned long) buffer);
1852                                 goto error;
1853                         }
1854                         if (offset) /* we already have one reference */
1855                                 get_page(page);
1856                         lbuf->l_offset = offset;
1857                         lbuf->l_ldata = buffer + offset;
1858                         lbuf->l_page = page;
1859                         lbuf->l_log = log;
1860                         init_waitqueue_head(&lbuf->l_ioevent);
1861
1862                         lbuf->l_freelist = log->lbuf_free;
1863                         log->lbuf_free = lbuf;
1864                         i++;
1865                 }
1866         }
1867
1868         return (0);
1869
1870       error:
1871         lbmLogShutdown(log);
1872         return -ENOMEM;
1873 }
1874
1875
1876 /*
1877  *      lbmLogShutdown()
1878  *
1879  * finalize per log I/O setup at lmLogShutdown()
1880  */
1881 static void lbmLogShutdown(struct jfs_log * log)
1882 {
1883         struct lbuf *lbuf;
1884
1885         jfs_info("lbmLogShutdown: log:0x%p", log);
1886
1887         lbuf = log->lbuf_free;
1888         while (lbuf) {
1889                 struct lbuf *next = lbuf->l_freelist;
1890                 __free_page(lbuf->l_page);
1891                 kfree(lbuf);
1892                 lbuf = next;
1893         }
1894 }
1895
1896
1897 /*
1898  *      lbmAllocate()
1899  *
1900  * allocate an empty log buffer
1901  */
1902 static struct lbuf *lbmAllocate(struct jfs_log * log, int pn)
1903 {
1904         struct lbuf *bp;
1905         unsigned long flags;
1906
1907         /*
1908          * recycle from log buffer freelist if any
1909          */
1910         LCACHE_LOCK(flags);
1911         LCACHE_SLEEP_COND(log->free_wait, (bp = log->lbuf_free), flags);
1912         log->lbuf_free = bp->l_freelist;
1913         LCACHE_UNLOCK(flags);
1914
1915         bp->l_flag = 0;
1916
1917         bp->l_wqnext = NULL;
1918         bp->l_freelist = NULL;
1919
1920         bp->l_pn = pn;
1921         bp->l_blkno = log->base + (pn << (L2LOGPSIZE - log->l2bsize));
1922         bp->l_ceor = 0;
1923
1924         return bp;
1925 }
1926
1927
1928 /*
1929  *      lbmFree()
1930  *
1931  * release a log buffer to freelist
1932  */
1933 static void lbmFree(struct lbuf * bp)
1934 {
1935         unsigned long flags;
1936
1937         LCACHE_LOCK(flags);
1938
1939         lbmfree(bp);
1940
1941         LCACHE_UNLOCK(flags);
1942 }
1943
1944 static void lbmfree(struct lbuf * bp)
1945 {
1946         struct jfs_log *log = bp->l_log;
1947
1948         assert(bp->l_wqnext == NULL);
1949
1950         /*
1951          * return the buffer to head of freelist
1952          */
1953         bp->l_freelist = log->lbuf_free;
1954         log->lbuf_free = bp;
1955
1956         wake_up(&log->free_wait);
1957         return;
1958 }
1959
1960
1961 /*
1962  * NAME:        lbmRedrive
1963  *
1964  * FUNCTION:    add a log buffer to the the log redrive list
1965  *
1966  * PARAMETER:
1967  *     bp       - log buffer
1968  *
1969  * NOTES:
1970  *      Takes log_redrive_lock.
1971  */
1972 static inline void lbmRedrive(struct lbuf *bp)
1973 {
1974         unsigned long flags;
1975
1976         spin_lock_irqsave(&log_redrive_lock, flags);
1977         bp->l_redrive_next = log_redrive_list;
1978         log_redrive_list = bp;
1979         spin_unlock_irqrestore(&log_redrive_lock, flags);
1980
1981         wake_up_process(jfsIOthread);
1982 }
1983
1984
1985 /*
1986  *      lbmRead()
1987  */
1988 static int lbmRead(struct jfs_log * log, int pn, struct lbuf ** bpp)
1989 {
1990         struct bio *bio;
1991         struct lbuf *bp;
1992
1993         /*
1994          * allocate a log buffer
1995          */
1996         *bpp = bp = lbmAllocate(log, pn);
1997         jfs_info("lbmRead: bp:0x%p pn:0x%x", bp, pn);
1998
1999         bp->l_flag |= lbmREAD;
2000
2001         bio = bio_alloc(GFP_NOFS, 1);
2002
2003         bio->bi_sector = bp->l_blkno << (log->l2bsize - 9);
2004         bio->bi_bdev = log->bdev;
2005         bio->bi_io_vec[0].bv_page = bp->l_page;
2006         bio->bi_io_vec[0].bv_len = LOGPSIZE;
2007         bio->bi_io_vec[0].bv_offset = bp->l_offset;
2008
2009         bio->bi_vcnt = 1;
2010         bio->bi_idx = 0;
2011         bio->bi_size = LOGPSIZE;
2012
2013         bio->bi_end_io = lbmIODone;
2014         bio->bi_private = bp;
2015         submit_bio(READ_SYNC, bio);
2016
2017         wait_event(bp->l_ioevent, (bp->l_flag != lbmREAD));
2018
2019         return 0;
2020 }
2021
2022
2023 /*
2024  *      lbmWrite()
2025  *
2026  * buffer at head of pageout queue stays after completion of
2027  * partial-page pageout and redriven by explicit initiation of
2028  * pageout by caller until full-page pageout is completed and
2029  * released.
2030  *
2031  * device driver i/o done redrives pageout of new buffer at
2032  * head of pageout queue when current buffer at head of pageout
2033  * queue is released at the completion of its full-page pageout.
2034  *
2035  * LOGGC_LOCK() serializes lbmWrite() by lmNextPage() and lmGroupCommit().
2036  * LCACHE_LOCK() serializes xflag between lbmWrite() and lbmIODone()
2037  */
2038 static void lbmWrite(struct jfs_log * log, struct lbuf * bp, int flag,
2039                      int cant_block)
2040 {
2041         struct lbuf *tail;
2042         unsigned long flags;
2043
2044         jfs_info("lbmWrite: bp:0x%p flag:0x%x pn:0x%x", bp, flag, bp->l_pn);
2045
2046         /* map the logical block address to physical block address */
2047         bp->l_blkno =
2048             log->base + (bp->l_pn << (L2LOGPSIZE - log->l2bsize));
2049
2050         LCACHE_LOCK(flags);             /* disable+lock */
2051
2052         /*
2053          * initialize buffer for device driver
2054          */
2055         bp->l_flag = flag;
2056
2057         /*
2058          *      insert bp at tail of write queue associated with log
2059          *
2060          * (request is either for bp already/currently at head of queue
2061          * or new bp to be inserted at tail)
2062          */
2063         tail = log->wqueue;
2064
2065         /* is buffer not already on write queue ? */
2066         if (bp->l_wqnext == NULL) {
2067                 /* insert at tail of wqueue */
2068                 if (tail == NULL) {
2069                         log->wqueue = bp;
2070                         bp->l_wqnext = bp;
2071                 } else {
2072                         log->wqueue = bp;
2073                         bp->l_wqnext = tail->l_wqnext;
2074                         tail->l_wqnext = bp;
2075                 }
2076
2077                 tail = bp;
2078         }
2079
2080         /* is buffer at head of wqueue and for write ? */
2081         if ((bp != tail->l_wqnext) || !(flag & lbmWRITE)) {
2082                 LCACHE_UNLOCK(flags);   /* unlock+enable */
2083                 return;
2084         }
2085
2086         LCACHE_UNLOCK(flags);   /* unlock+enable */
2087
2088         if (cant_block)
2089                 lbmRedrive(bp);
2090         else if (flag & lbmSYNC)
2091                 lbmStartIO(bp);
2092         else {
2093                 LOGGC_UNLOCK(log);
2094                 lbmStartIO(bp);
2095                 LOGGC_LOCK(log);
2096         }
2097 }
2098
2099
2100 /*
2101  *      lbmDirectWrite()
2102  *
2103  * initiate pageout bypassing write queue for sidestream
2104  * (e.g., log superblock) write;
2105  */
2106 static void lbmDirectWrite(struct jfs_log * log, struct lbuf * bp, int flag)
2107 {
2108         jfs_info("lbmDirectWrite: bp:0x%p flag:0x%x pn:0x%x",
2109                  bp, flag, bp->l_pn);
2110
2111         /*
2112          * initialize buffer for device driver
2113          */
2114         bp->l_flag = flag | lbmDIRECT;
2115
2116         /* map the logical block address to physical block address */
2117         bp->l_blkno =
2118             log->base + (bp->l_pn << (L2LOGPSIZE - log->l2bsize));
2119
2120         /*
2121          *      initiate pageout of the page
2122          */
2123         lbmStartIO(bp);
2124 }
2125
2126
2127 /*
2128  * NAME:        lbmStartIO()
2129  *
2130  * FUNCTION:    Interface to DD strategy routine
2131  *
2132  * RETURN:      none
2133  *
2134  * serialization: LCACHE_LOCK() is NOT held during log i/o;
2135  */
2136 static void lbmStartIO(struct lbuf * bp)
2137 {
2138         struct bio *bio;
2139         struct jfs_log *log = bp->l_log;
2140
2141         jfs_info("lbmStartIO\n");
2142
2143         bio = bio_alloc(GFP_NOFS, 1);
2144         bio->bi_sector = bp->l_blkno << (log->l2bsize - 9);
2145         bio->bi_bdev = log->bdev;
2146         bio->bi_io_vec[0].bv_page = bp->l_page;
2147         bio->bi_io_vec[0].bv_len = LOGPSIZE;
2148         bio->bi_io_vec[0].bv_offset = bp->l_offset;
2149
2150         bio->bi_vcnt = 1;
2151         bio->bi_idx = 0;
2152         bio->bi_size = LOGPSIZE;
2153
2154         bio->bi_end_io = lbmIODone;
2155         bio->bi_private = bp;
2156
2157         /* check if journaling to disk has been disabled */
2158         if (log->no_integrity) {
2159                 bio->bi_size = 0;
2160                 lbmIODone(bio, 0, 0);
2161         } else {
2162                 submit_bio(WRITE_SYNC, bio);
2163                 INCREMENT(lmStat.submitted);
2164         }
2165 }
2166
2167
2168 /*
2169  *      lbmIOWait()
2170  */
2171 static int lbmIOWait(struct lbuf * bp, int flag)
2172 {
2173         unsigned long flags;
2174         int rc = 0;
2175
2176         jfs_info("lbmIOWait1: bp:0x%p flag:0x%x:0x%x", bp, bp->l_flag, flag);
2177
2178         LCACHE_LOCK(flags);             /* disable+lock */
2179
2180         LCACHE_SLEEP_COND(bp->l_ioevent, (bp->l_flag & lbmDONE), flags);
2181
2182         rc = (bp->l_flag & lbmERROR) ? -EIO : 0;
2183
2184         if (flag & lbmFREE)
2185                 lbmfree(bp);
2186
2187         LCACHE_UNLOCK(flags);   /* unlock+enable */
2188
2189         jfs_info("lbmIOWait2: bp:0x%p flag:0x%x:0x%x", bp, bp->l_flag, flag);
2190         return rc;
2191 }
2192
2193 /*
2194  *      lbmIODone()
2195  *
2196  * executed at INTIODONE level
2197  */
2198 static int lbmIODone(struct bio *bio, unsigned int bytes_done, int error)
2199 {
2200         struct lbuf *bp = bio->bi_private;
2201         struct lbuf *nextbp, *tail;
2202         struct jfs_log *log;
2203         unsigned long flags;
2204
2205         if (bio->bi_size)
2206                 return 1;
2207
2208         /*
2209          * get back jfs buffer bound to the i/o buffer
2210          */
2211         jfs_info("lbmIODone: bp:0x%p flag:0x%x", bp, bp->l_flag);
2212
2213         LCACHE_LOCK(flags);             /* disable+lock */
2214
2215         bp->l_flag |= lbmDONE;
2216
2217         if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
2218                 bp->l_flag |= lbmERROR;
2219
2220                 jfs_err("lbmIODone: I/O error in JFS log");
2221         }
2222
2223         bio_put(bio);
2224
2225         /*
2226          *      pagein completion
2227          */
2228         if (bp->l_flag & lbmREAD) {
2229                 bp->l_flag &= ~lbmREAD;
2230
2231                 LCACHE_UNLOCK(flags);   /* unlock+enable */
2232
2233                 /* wakeup I/O initiator */
2234                 LCACHE_WAKEUP(&bp->l_ioevent);
2235
2236                 return 0;
2237         }
2238
2239         /*
2240          *      pageout completion
2241          *
2242          * the bp at the head of write queue has completed pageout.
2243          *
2244          * if single-commit/full-page pageout, remove the current buffer
2245          * from head of pageout queue, and redrive pageout with
2246          * the new buffer at head of pageout queue;
2247          * otherwise, the partial-page pageout buffer stays at
2248          * the head of pageout queue to be redriven for pageout
2249          * by lmGroupCommit() until full-page pageout is completed.
2250          */
2251         bp->l_flag &= ~lbmWRITE;
2252         INCREMENT(lmStat.pagedone);
2253
2254         /* update committed lsn */
2255         log = bp->l_log;
2256         log->clsn = (bp->l_pn << L2LOGPSIZE) + bp->l_ceor;
2257
2258         if (bp->l_flag & lbmDIRECT) {
2259                 LCACHE_WAKEUP(&bp->l_ioevent);
2260                 LCACHE_UNLOCK(flags);
2261                 return 0;
2262         }
2263
2264         tail = log->wqueue;
2265
2266         /* single element queue */
2267         if (bp == tail) {
2268                 /* remove head buffer of full-page pageout
2269                  * from log device write queue
2270                  */
2271                 if (bp->l_flag & lbmRELEASE) {
2272                         log->wqueue = NULL;
2273                         bp->l_wqnext = NULL;
2274                 }
2275         }
2276         /* multi element queue */
2277         else {
2278                 /* remove head buffer of full-page pageout
2279                  * from log device write queue
2280                  */
2281                 if (bp->l_flag & lbmRELEASE) {
2282                         nextbp = tail->l_wqnext = bp->l_wqnext;
2283                         bp->l_wqnext = NULL;
2284
2285                         /*
2286                          * redrive pageout of next page at head of write queue:
2287                          * redrive next page without any bound tblk
2288                          * (i.e., page w/o any COMMIT records), or
2289                          * first page of new group commit which has been
2290                          * queued after current page (subsequent pageout
2291                          * is performed synchronously, except page without
2292                          * any COMMITs) by lmGroupCommit() as indicated
2293                          * by lbmWRITE flag;
2294                          */
2295                         if (nextbp->l_flag & lbmWRITE) {
2296                                 /*
2297                                  * We can't do the I/O at interrupt time.
2298                                  * The jfsIO thread can do it
2299                                  */
2300                                 lbmRedrive(nextbp);
2301                         }
2302                 }
2303         }
2304
2305         /*
2306          *      synchronous pageout:
2307          *
2308          * buffer has not necessarily been removed from write queue
2309          * (e.g., synchronous write of partial-page with COMMIT):
2310          * leave buffer for i/o initiator to dispose
2311          */
2312         if (bp->l_flag & lbmSYNC) {
2313                 LCACHE_UNLOCK(flags);   /* unlock+enable */
2314
2315                 /* wakeup I/O initiator */
2316                 LCACHE_WAKEUP(&bp->l_ioevent);
2317         }
2318
2319         /*
2320          *      Group Commit pageout:
2321          */
2322         else if (bp->l_flag & lbmGC) {
2323                 LCACHE_UNLOCK(flags);
2324                 lmPostGC(bp);
2325         }
2326
2327         /*
2328          *      asynchronous pageout:
2329          *
2330          * buffer must have been removed from write queue:
2331          * insert buffer at head of freelist where it can be recycled
2332          */
2333         else {
2334                 assert(bp->l_flag & lbmRELEASE);
2335                 assert(bp->l_flag & lbmFREE);
2336                 lbmfree(bp);
2337
2338                 LCACHE_UNLOCK(flags);   /* unlock+enable */
2339         }
2340
2341         return 0;
2342 }
2343
2344 int jfsIOWait(void *arg)
2345 {
2346         struct lbuf *bp;
2347
2348         do {
2349                 spin_lock_irq(&log_redrive_lock);
2350                 while ((bp = log_redrive_list) != 0) {
2351                         log_redrive_list = bp->l_redrive_next;
2352                         bp->l_redrive_next = NULL;
2353                         spin_unlock_irq(&log_redrive_lock);
2354                         lbmStartIO(bp);
2355                         spin_lock_irq(&log_redrive_lock);
2356                 }
2357                 spin_unlock_irq(&log_redrive_lock);
2358
2359                 if (freezing(current)) {
2360                         refrigerator();
2361                 } else {
2362                         set_current_state(TASK_INTERRUPTIBLE);
2363                         schedule();
2364                         current->state = TASK_RUNNING;
2365                 }
2366         } while (!kthread_should_stop());
2367
2368         jfs_info("jfsIOWait being killed!");
2369         return 0;
2370 }
2371
2372 /*
2373  * NAME:        lmLogFormat()/jfs_logform()
2374  *
2375  * FUNCTION:    format file system log
2376  *
2377  * PARAMETERS:
2378  *      log     - volume log
2379  *      logAddress - start address of log space in FS block
2380  *      logSize - length of log space in FS block;
2381  *
2382  * RETURN:      0       - success
2383  *              -EIO    - i/o error
2384  *
2385  * XXX: We're synchronously writing one page at a time.  This needs to
2386  *      be improved by writing multiple pages at once.
2387  */
2388 int lmLogFormat(struct jfs_log *log, s64 logAddress, int logSize)
2389 {
2390         int rc = -EIO;
2391         struct jfs_sb_info *sbi;
2392         struct logsuper *logsuper;
2393         struct logpage *lp;
2394         int lspn;               /* log sequence page number */
2395         struct lrd *lrd_ptr;
2396         int npages = 0;
2397         struct lbuf *bp;
2398
2399         jfs_info("lmLogFormat: logAddress:%Ld logSize:%d",
2400                  (long long)logAddress, logSize);
2401
2402         sbi = list_entry(log->sb_list.next, struct jfs_sb_info, log_list);
2403
2404         /* allocate a log buffer */
2405         bp = lbmAllocate(log, 1);
2406
2407         npages = logSize >> sbi->l2nbperpage;
2408
2409         /*
2410          *      log space:
2411          *
2412          * page 0 - reserved;
2413          * page 1 - log superblock;
2414          * page 2 - log data page: A SYNC log record is written
2415          *          into this page at logform time;
2416          * pages 3-N - log data page: set to empty log data pages;
2417          */
2418         /*
2419          *      init log superblock: log page 1
2420          */
2421         logsuper = (struct logsuper *) bp->l_ldata;
2422
2423         logsuper->magic = cpu_to_le32(LOGMAGIC);
2424         logsuper->version = cpu_to_le32(LOGVERSION);
2425         logsuper->state = cpu_to_le32(LOGREDONE);
2426         logsuper->flag = cpu_to_le32(sbi->mntflag);     /* ? */
2427         logsuper->size = cpu_to_le32(npages);
2428         logsuper->bsize = cpu_to_le32(sbi->bsize);
2429         logsuper->l2bsize = cpu_to_le32(sbi->l2bsize);
2430         logsuper->end = cpu_to_le32(2 * LOGPSIZE + LOGPHDRSIZE + LOGRDSIZE);
2431
2432         bp->l_flag = lbmWRITE | lbmSYNC | lbmDIRECT;
2433         bp->l_blkno = logAddress + sbi->nbperpage;
2434         lbmStartIO(bp);
2435         if ((rc = lbmIOWait(bp, 0)))
2436                 goto exit;
2437
2438         /*
2439          *      init pages 2 to npages-1 as log data pages:
2440          *
2441          * log page sequence number (lpsn) initialization:
2442          *
2443          * pn:   0     1     2     3                 n-1
2444          *       +-----+-----+=====+=====+===.....===+=====+
2445          * lspn:             N-1   0     1           N-2
2446          *                   <--- N page circular file ---->
2447          *
2448          * the N (= npages-2) data pages of the log is maintained as
2449          * a circular file for the log records;
2450          * lpsn grows by 1 monotonically as each log page is written
2451          * to the circular file of the log;
2452          * and setLogpage() will not reset the page number even if
2453          * the eor is equal to LOGPHDRSIZE. In order for binary search
2454          * still work in find log end process, we have to simulate the
2455          * log wrap situation at the log format time.
2456          * The 1st log page written will have the highest lpsn. Then
2457          * the succeeding log pages will have ascending order of
2458          * the lspn starting from 0, ... (N-2)
2459          */
2460         lp = (struct logpage *) bp->l_ldata;
2461         /*
2462          * initialize 1st log page to be written: lpsn = N - 1,
2463          * write a SYNCPT log record is written to this page
2464          */
2465         lp->h.page = lp->t.page = cpu_to_le32(npages - 3);
2466         lp->h.eor = lp->t.eor = cpu_to_le16(LOGPHDRSIZE + LOGRDSIZE);
2467
2468         lrd_ptr = (struct lrd *) &lp->data;
2469         lrd_ptr->logtid = 0;
2470         lrd_ptr->backchain = 0;
2471         lrd_ptr->type = cpu_to_le16(LOG_SYNCPT);
2472         lrd_ptr->length = 0;
2473         lrd_ptr->log.syncpt.sync = 0;
2474
2475         bp->l_blkno += sbi->nbperpage;
2476         bp->l_flag = lbmWRITE | lbmSYNC | lbmDIRECT;
2477         lbmStartIO(bp);
2478         if ((rc = lbmIOWait(bp, 0)))
2479                 goto exit;
2480
2481         /*
2482          *      initialize succeeding log pages: lpsn = 0, 1, ..., (N-2)
2483          */
2484         for (lspn = 0; lspn < npages - 3; lspn++) {
2485                 lp->h.page = lp->t.page = cpu_to_le32(lspn);
2486                 lp->h.eor = lp->t.eor = cpu_to_le16(LOGPHDRSIZE);
2487
2488                 bp->l_blkno += sbi->nbperpage;
2489                 bp->l_flag = lbmWRITE | lbmSYNC | lbmDIRECT;
2490                 lbmStartIO(bp);
2491                 if ((rc = lbmIOWait(bp, 0)))
2492                         goto exit;
2493         }
2494
2495         rc = 0;
2496 exit:
2497         /*
2498          *      finalize log
2499          */
2500         /* release the buffer */
2501         lbmFree(bp);
2502
2503         return rc;
2504 }
2505
2506 #ifdef CONFIG_JFS_STATISTICS
2507 int jfs_lmstats_read(char *buffer, char **start, off_t offset, int length,
2508                       int *eof, void *data)
2509 {
2510         int len = 0;
2511         off_t begin;
2512
2513         len += sprintf(buffer,
2514                        "JFS Logmgr stats\n"
2515                        "================\n"
2516                        "commits = %d\n"
2517                        "writes submitted = %d\n"
2518                        "writes completed = %d\n"
2519                        "full pages submitted = %d\n"
2520                        "partial pages submitted = %d\n",
2521                        lmStat.commit,
2522                        lmStat.submitted,
2523                        lmStat.pagedone,
2524                        lmStat.full_page,
2525                        lmStat.partial_page);
2526
2527         begin = offset;
2528         *start = buffer + begin;
2529         len -= begin;
2530
2531         if (len > length)
2532                 len = length;
2533         else
2534                 *eof = 1;
2535
2536         if (len < 0)
2537                 len = 0;
2538
2539         return len;
2540 }
2541 #endif /* CONFIG_JFS_STATISTICS */