Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc-2.6
[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/completion.h>
66 #include <linux/kthread.h>
67 #include <linux/buffer_head.h>          /* for sync_blockdev() */
68 #include <linux/bio.h>
69 #include <linux/freezer.h>
70 #include <linux/delay.h>
71 #include <linux/mutex.h>
72 #include <linux/seq_file.h>
73 #include "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 static void write_special_inodes(struct jfs_log *log,
213                                  int (*writer)(struct address_space *))
214 {
215         struct jfs_sb_info *sbi;
216
217         list_for_each_entry(sbi, &log->sb_list, log_list) {
218                 writer(sbi->ipbmap->i_mapping);
219                 writer(sbi->ipimap->i_mapping);
220                 writer(sbi->direct_inode->i_mapping);
221         }
222 }
223
224 /*
225  * NAME:        lmLog()
226  *
227  * FUNCTION:    write a log record;
228  *
229  * PARAMETER:
230  *
231  * RETURN:      lsn - offset to the next log record to write (end-of-log);
232  *              -1  - error;
233  *
234  * note: todo: log error handler
235  */
236 int lmLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
237           struct tlock * tlck)
238 {
239         int lsn;
240         int diffp, difft;
241         struct metapage *mp = NULL;
242         unsigned long flags;
243
244         jfs_info("lmLog: log:0x%p tblk:0x%p, lrd:0x%p tlck:0x%p",
245                  log, tblk, lrd, tlck);
246
247         LOG_LOCK(log);
248
249         /* log by (out-of-transaction) JFS ? */
250         if (tblk == NULL)
251                 goto writeRecord;
252
253         /* log from page ? */
254         if (tlck == NULL ||
255             tlck->type & tlckBTROOT || (mp = tlck->mp) == NULL)
256                 goto writeRecord;
257
258         /*
259          *      initialize/update page/transaction recovery lsn
260          */
261         lsn = log->lsn;
262
263         LOGSYNC_LOCK(log, flags);
264
265         /*
266          * initialize page lsn if first log write of the page
267          */
268         if (mp->lsn == 0) {
269                 mp->log = log;
270                 mp->lsn = lsn;
271                 log->count++;
272
273                 /* insert page at tail of logsynclist */
274                 list_add_tail(&mp->synclist, &log->synclist);
275         }
276
277         /*
278          *      initialize/update lsn of tblock of the page
279          *
280          * transaction inherits oldest lsn of pages associated
281          * with allocation/deallocation of resources (their
282          * log records are used to reconstruct allocation map
283          * at recovery time: inode for inode allocation map,
284          * B+-tree index of extent descriptors for block
285          * allocation map);
286          * allocation map pages inherit transaction lsn at
287          * commit time to allow forwarding log syncpt past log
288          * records associated with allocation/deallocation of
289          * resources only after persistent map of these map pages
290          * have been updated and propagated to home.
291          */
292         /*
293          * initialize transaction lsn:
294          */
295         if (tblk->lsn == 0) {
296                 /* inherit lsn of its first page logged */
297                 tblk->lsn = mp->lsn;
298                 log->count++;
299
300                 /* insert tblock after the page on logsynclist */
301                 list_add(&tblk->synclist, &mp->synclist);
302         }
303         /*
304          * update transaction lsn:
305          */
306         else {
307                 /* inherit oldest/smallest lsn of page */
308                 logdiff(diffp, mp->lsn, log);
309                 logdiff(difft, tblk->lsn, log);
310                 if (diffp < difft) {
311                         /* update tblock lsn with page lsn */
312                         tblk->lsn = mp->lsn;
313
314                         /* move tblock after page on logsynclist */
315                         list_move(&tblk->synclist, &mp->synclist);
316                 }
317         }
318
319         LOGSYNC_UNLOCK(log, flags);
320
321         /*
322          *      write the log record
323          */
324       writeRecord:
325         lsn = lmWriteRecord(log, tblk, lrd, tlck);
326
327         /*
328          * forward log syncpt if log reached next syncpt trigger
329          */
330         logdiff(diffp, lsn, log);
331         if (diffp >= log->nextsync)
332                 lsn = lmLogSync(log, 0);
333
334         /* update end-of-log lsn */
335         log->lsn = lsn;
336
337         LOG_UNLOCK(log);
338
339         /* return end-of-log address */
340         return lsn;
341 }
342
343 /*
344  * NAME:        lmWriteRecord()
345  *
346  * FUNCTION:    move the log record to current log page
347  *
348  * PARAMETER:   cd      - commit descriptor
349  *
350  * RETURN:      end-of-log address
351  *
352  * serialization: LOG_LOCK() held on entry/exit
353  */
354 static int
355 lmWriteRecord(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
356               struct tlock * tlck)
357 {
358         int lsn = 0;            /* end-of-log address */
359         struct lbuf *bp;        /* dst log page buffer */
360         struct logpage *lp;     /* dst log page */
361         caddr_t dst;            /* destination address in log page */
362         int dstoffset;          /* end-of-log offset in log page */
363         int freespace;          /* free space in log page */
364         caddr_t p;              /* src meta-data page */
365         caddr_t src;
366         int srclen;
367         int nbytes;             /* number of bytes to move */
368         int i;
369         int len;
370         struct linelock *linelock;
371         struct lv *lv;
372         struct lvd *lvd;
373         int l2linesize;
374
375         len = 0;
376
377         /* retrieve destination log page to write */
378         bp = (struct lbuf *) log->bp;
379         lp = (struct logpage *) bp->l_ldata;
380         dstoffset = log->eor;
381
382         /* any log data to write ? */
383         if (tlck == NULL)
384                 goto moveLrd;
385
386         /*
387          *      move log record data
388          */
389         /* retrieve source meta-data page to log */
390         if (tlck->flag & tlckPAGELOCK) {
391                 p = (caddr_t) (tlck->mp->data);
392                 linelock = (struct linelock *) & tlck->lock;
393         }
394         /* retrieve source in-memory inode to log */
395         else if (tlck->flag & tlckINODELOCK) {
396                 if (tlck->type & tlckDTREE)
397                         p = (caddr_t) &JFS_IP(tlck->ip)->i_dtroot;
398                 else
399                         p = (caddr_t) &JFS_IP(tlck->ip)->i_xtroot;
400                 linelock = (struct linelock *) & tlck->lock;
401         }
402 #ifdef  _JFS_WIP
403         else if (tlck->flag & tlckINLINELOCK) {
404
405                 inlinelock = (struct inlinelock *) & tlck;
406                 p = (caddr_t) & inlinelock->pxd;
407                 linelock = (struct linelock *) & tlck;
408         }
409 #endif                          /* _JFS_WIP */
410         else {
411                 jfs_err("lmWriteRecord: UFO tlck:0x%p", tlck);
412                 return 0;       /* Probably should trap */
413         }
414         l2linesize = linelock->l2linesize;
415
416       moveData:
417         ASSERT(linelock->index <= linelock->maxcnt);
418
419         lv = linelock->lv;
420         for (i = 0; i < linelock->index; i++, lv++) {
421                 if (lv->length == 0)
422                         continue;
423
424                 /* is page full ? */
425                 if (dstoffset >= LOGPSIZE - LOGPTLRSIZE) {
426                         /* page become full: move on to next page */
427                         lmNextPage(log);
428
429                         bp = log->bp;
430                         lp = (struct logpage *) bp->l_ldata;
431                         dstoffset = LOGPHDRSIZE;
432                 }
433
434                 /*
435                  * move log vector data
436                  */
437                 src = (u8 *) p + (lv->offset << l2linesize);
438                 srclen = lv->length << l2linesize;
439                 len += srclen;
440                 while (srclen > 0) {
441                         freespace = (LOGPSIZE - LOGPTLRSIZE) - dstoffset;
442                         nbytes = min(freespace, srclen);
443                         dst = (caddr_t) lp + dstoffset;
444                         memcpy(dst, src, nbytes);
445                         dstoffset += nbytes;
446
447                         /* is page not full ? */
448                         if (dstoffset < LOGPSIZE - LOGPTLRSIZE)
449                                 break;
450
451                         /* page become full: move on to next page */
452                         lmNextPage(log);
453
454                         bp = (struct lbuf *) log->bp;
455                         lp = (struct logpage *) bp->l_ldata;
456                         dstoffset = LOGPHDRSIZE;
457
458                         srclen -= nbytes;
459                         src += nbytes;
460                 }
461
462                 /*
463                  * move log vector descriptor
464                  */
465                 len += 4;
466                 lvd = (struct lvd *) ((caddr_t) lp + dstoffset);
467                 lvd->offset = cpu_to_le16(lv->offset);
468                 lvd->length = cpu_to_le16(lv->length);
469                 dstoffset += 4;
470                 jfs_info("lmWriteRecord: lv offset:%d length:%d",
471                          lv->offset, lv->length);
472         }
473
474         if ((i = linelock->next)) {
475                 linelock = (struct linelock *) lid_to_tlock(i);
476                 goto moveData;
477         }
478
479         /*
480          *      move log record descriptor
481          */
482       moveLrd:
483         lrd->length = cpu_to_le16(len);
484
485         src = (caddr_t) lrd;
486         srclen = LOGRDSIZE;
487
488         while (srclen > 0) {
489                 freespace = (LOGPSIZE - LOGPTLRSIZE) - dstoffset;
490                 nbytes = min(freespace, srclen);
491                 dst = (caddr_t) lp + dstoffset;
492                 memcpy(dst, src, nbytes);
493
494                 dstoffset += nbytes;
495                 srclen -= nbytes;
496
497                 /* are there more to move than freespace of page ? */
498                 if (srclen)
499                         goto pageFull;
500
501                 /*
502                  * end of log record descriptor
503                  */
504
505                 /* update last log record eor */
506                 log->eor = dstoffset;
507                 bp->l_eor = dstoffset;
508                 lsn = (log->page << L2LOGPSIZE) + dstoffset;
509
510                 if (lrd->type & cpu_to_le16(LOG_COMMIT)) {
511                         tblk->clsn = lsn;
512                         jfs_info("wr: tclsn:0x%x, beor:0x%x", tblk->clsn,
513                                  bp->l_eor);
514
515                         INCREMENT(lmStat.commit);       /* # of commit */
516
517                         /*
518                          * enqueue tblock for group commit:
519                          *
520                          * enqueue tblock of non-trivial/synchronous COMMIT
521                          * at tail of group commit queue
522                          * (trivial/asynchronous COMMITs are ignored by
523                          * group commit.)
524                          */
525                         LOGGC_LOCK(log);
526
527                         /* init tblock gc state */
528                         tblk->flag = tblkGC_QUEUE;
529                         tblk->bp = log->bp;
530                         tblk->pn = log->page;
531                         tblk->eor = log->eor;
532
533                         /* enqueue transaction to commit queue */
534                         list_add_tail(&tblk->cqueue, &log->cqueue);
535
536                         LOGGC_UNLOCK(log);
537                 }
538
539                 jfs_info("lmWriteRecord: lrd:0x%04x bp:0x%p pn:%d eor:0x%x",
540                         le16_to_cpu(lrd->type), log->bp, log->page, dstoffset);
541
542                 /* page not full ? */
543                 if (dstoffset < LOGPSIZE - LOGPTLRSIZE)
544                         return lsn;
545
546               pageFull:
547                 /* page become full: move on to next page */
548                 lmNextPage(log);
549
550                 bp = (struct lbuf *) log->bp;
551                 lp = (struct logpage *) bp->l_ldata;
552                 dstoffset = LOGPHDRSIZE;
553                 src += nbytes;
554         }
555
556         return lsn;
557 }
558
559
560 /*
561  * NAME:        lmNextPage()
562  *
563  * FUNCTION:    write current page and allocate next page.
564  *
565  * PARAMETER:   log
566  *
567  * RETURN:      0
568  *
569  * serialization: LOG_LOCK() held on entry/exit
570  */
571 static int lmNextPage(struct jfs_log * log)
572 {
573         struct logpage *lp;
574         int lspn;               /* log sequence page number */
575         int pn;                 /* current page number */
576         struct lbuf *bp;
577         struct lbuf *nextbp;
578         struct tblock *tblk;
579
580         /* get current log page number and log sequence page number */
581         pn = log->page;
582         bp = log->bp;
583         lp = (struct logpage *) bp->l_ldata;
584         lspn = le32_to_cpu(lp->h.page);
585
586         LOGGC_LOCK(log);
587
588         /*
589          *      write or queue the full page at the tail of write queue
590          */
591         /* get the tail tblk on commit queue */
592         if (list_empty(&log->cqueue))
593                 tblk = NULL;
594         else
595                 tblk = list_entry(log->cqueue.prev, struct tblock, cqueue);
596
597         /* every tblk who has COMMIT record on the current page,
598          * and has not been committed, must be on commit queue
599          * since tblk is queued at commit queueu at the time
600          * of writing its COMMIT record on the page before
601          * page becomes full (even though the tblk thread
602          * who wrote COMMIT record may have been suspended
603          * currently);
604          */
605
606         /* is page bound with outstanding tail tblk ? */
607         if (tblk && tblk->pn == pn) {
608                 /* mark tblk for end-of-page */
609                 tblk->flag |= tblkGC_EOP;
610
611                 if (log->cflag & logGC_PAGEOUT) {
612                         /* if page is not already on write queue,
613                          * just enqueue (no lbmWRITE to prevent redrive)
614                          * buffer to wqueue to ensure correct serial order
615                          * of the pages since log pages will be added
616                          * continuously
617                          */
618                         if (bp->l_wqnext == NULL)
619                                 lbmWrite(log, bp, 0, 0);
620                 } else {
621                         /*
622                          * No current GC leader, initiate group commit
623                          */
624                         log->cflag |= logGC_PAGEOUT;
625                         lmGCwrite(log, 0);
626                 }
627         }
628         /* page is not bound with outstanding tblk:
629          * init write or mark it to be redriven (lbmWRITE)
630          */
631         else {
632                 /* finalize the page */
633                 bp->l_ceor = bp->l_eor;
634                 lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_ceor);
635                 lbmWrite(log, bp, lbmWRITE | lbmRELEASE | lbmFREE, 0);
636         }
637         LOGGC_UNLOCK(log);
638
639         /*
640          *      allocate/initialize next page
641          */
642         /* if log wraps, the first data page of log is 2
643          * (0 never used, 1 is superblock).
644          */
645         log->page = (pn == log->size - 1) ? 2 : pn + 1;
646         log->eor = LOGPHDRSIZE; /* ? valid page empty/full at logRedo() */
647
648         /* allocate/initialize next log page buffer */
649         nextbp = lbmAllocate(log, log->page);
650         nextbp->l_eor = log->eor;
651         log->bp = nextbp;
652
653         /* initialize next log page */
654         lp = (struct logpage *) nextbp->l_ldata;
655         lp->h.page = lp->t.page = cpu_to_le32(lspn + 1);
656         lp->h.eor = lp->t.eor = cpu_to_le16(LOGPHDRSIZE);
657
658         return 0;
659 }
660
661
662 /*
663  * NAME:        lmGroupCommit()
664  *
665  * FUNCTION:    group commit
666  *      initiate pageout of the pages with COMMIT in the order of
667  *      page number - redrive pageout of the page at the head of
668  *      pageout queue until full page has been written.
669  *
670  * RETURN:
671  *
672  * NOTE:
673  *      LOGGC_LOCK serializes log group commit queue, and
674  *      transaction blocks on the commit queue.
675  *      N.B. LOG_LOCK is NOT held during lmGroupCommit().
676  */
677 int lmGroupCommit(struct jfs_log * log, struct tblock * tblk)
678 {
679         int rc = 0;
680
681         LOGGC_LOCK(log);
682
683         /* group committed already ? */
684         if (tblk->flag & tblkGC_COMMITTED) {
685                 if (tblk->flag & tblkGC_ERROR)
686                         rc = -EIO;
687
688                 LOGGC_UNLOCK(log);
689                 return rc;
690         }
691         jfs_info("lmGroup Commit: tblk = 0x%p, gcrtc = %d", tblk, log->gcrtc);
692
693         if (tblk->xflag & COMMIT_LAZY)
694                 tblk->flag |= tblkGC_LAZY;
695
696         if ((!(log->cflag & logGC_PAGEOUT)) && (!list_empty(&log->cqueue)) &&
697             (!(tblk->xflag & COMMIT_LAZY) || test_bit(log_FLUSH, &log->flag)
698              || jfs_tlocks_low)) {
699                 /*
700                  * No pageout in progress
701                  *
702                  * start group commit as its group leader.
703                  */
704                 log->cflag |= logGC_PAGEOUT;
705
706                 lmGCwrite(log, 0);
707         }
708
709         if (tblk->xflag & COMMIT_LAZY) {
710                 /*
711                  * Lazy transactions can leave now
712                  */
713                 LOGGC_UNLOCK(log);
714                 return 0;
715         }
716
717         /* lmGCwrite gives up LOGGC_LOCK, check again */
718
719         if (tblk->flag & tblkGC_COMMITTED) {
720                 if (tblk->flag & tblkGC_ERROR)
721                         rc = -EIO;
722
723                 LOGGC_UNLOCK(log);
724                 return rc;
725         }
726
727         /* upcount transaction waiting for completion
728          */
729         log->gcrtc++;
730         tblk->flag |= tblkGC_READY;
731
732         __SLEEP_COND(tblk->gcwait, (tblk->flag & tblkGC_COMMITTED),
733                      LOGGC_LOCK(log), LOGGC_UNLOCK(log));
734
735         /* removed from commit queue */
736         if (tblk->flag & tblkGC_ERROR)
737                 rc = -EIO;
738
739         LOGGC_UNLOCK(log);
740         return rc;
741 }
742
743 /*
744  * NAME:        lmGCwrite()
745  *
746  * FUNCTION:    group commit write
747  *      initiate write of log page, building a group of all transactions
748  *      with commit records on that page.
749  *
750  * RETURN:      None
751  *
752  * NOTE:
753  *      LOGGC_LOCK must be held by caller.
754  *      N.B. LOG_LOCK is NOT held during lmGroupCommit().
755  */
756 static void lmGCwrite(struct jfs_log * log, int cant_write)
757 {
758         struct lbuf *bp;
759         struct logpage *lp;
760         int gcpn;               /* group commit page number */
761         struct tblock *tblk;
762         struct tblock *xtblk = NULL;
763
764         /*
765          * build the commit group of a log page
766          *
767          * scan commit queue and make a commit group of all
768          * transactions with COMMIT records on the same log page.
769          */
770         /* get the head tblk on the commit queue */
771         gcpn = list_entry(log->cqueue.next, struct tblock, cqueue)->pn;
772
773         list_for_each_entry(tblk, &log->cqueue, cqueue) {
774                 if (tblk->pn != gcpn)
775                         break;
776
777                 xtblk = tblk;
778
779                 /* state transition: (QUEUE, READY) -> COMMIT */
780                 tblk->flag |= tblkGC_COMMIT;
781         }
782         tblk = xtblk;           /* last tblk of the page */
783
784         /*
785          * pageout to commit transactions on the log page.
786          */
787         bp = (struct lbuf *) tblk->bp;
788         lp = (struct logpage *) bp->l_ldata;
789         /* is page already full ? */
790         if (tblk->flag & tblkGC_EOP) {
791                 /* mark page to free at end of group commit of the page */
792                 tblk->flag &= ~tblkGC_EOP;
793                 tblk->flag |= tblkGC_FREE;
794                 bp->l_ceor = bp->l_eor;
795                 lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_ceor);
796                 lbmWrite(log, bp, lbmWRITE | lbmRELEASE | lbmGC,
797                          cant_write);
798                 INCREMENT(lmStat.full_page);
799         }
800         /* page is not yet full */
801         else {
802                 bp->l_ceor = tblk->eor; /* ? bp->l_ceor = bp->l_eor; */
803                 lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_ceor);
804                 lbmWrite(log, bp, lbmWRITE | lbmGC, cant_write);
805                 INCREMENT(lmStat.partial_page);
806         }
807 }
808
809 /*
810  * NAME:        lmPostGC()
811  *
812  * FUNCTION:    group commit post-processing
813  *      Processes transactions after their commit records have been written
814  *      to disk, redriving log I/O if necessary.
815  *
816  * RETURN:      None
817  *
818  * NOTE:
819  *      This routine is called a interrupt time by lbmIODone
820  */
821 static void lmPostGC(struct lbuf * bp)
822 {
823         unsigned long flags;
824         struct jfs_log *log = bp->l_log;
825         struct logpage *lp;
826         struct tblock *tblk, *temp;
827
828         //LOGGC_LOCK(log);
829         spin_lock_irqsave(&log->gclock, flags);
830         /*
831          * current pageout of group commit completed.
832          *
833          * remove/wakeup transactions from commit queue who were
834          * group committed with the current log page
835          */
836         list_for_each_entry_safe(tblk, temp, &log->cqueue, cqueue) {
837                 if (!(tblk->flag & tblkGC_COMMIT))
838                         break;
839                 /* if transaction was marked GC_COMMIT then
840                  * it has been shipped in the current pageout
841                  * and made it to disk - it is committed.
842                  */
843
844                 if (bp->l_flag & lbmERROR)
845                         tblk->flag |= tblkGC_ERROR;
846
847                 /* remove it from the commit queue */
848                 list_del(&tblk->cqueue);
849                 tblk->flag &= ~tblkGC_QUEUE;
850
851                 if (tblk == log->flush_tblk) {
852                         /* we can stop flushing the log now */
853                         clear_bit(log_FLUSH, &log->flag);
854                         log->flush_tblk = NULL;
855                 }
856
857                 jfs_info("lmPostGC: tblk = 0x%p, flag = 0x%x", tblk,
858                          tblk->flag);
859
860                 if (!(tblk->xflag & COMMIT_FORCE))
861                         /*
862                          * Hand tblk over to lazy commit thread
863                          */
864                         txLazyUnlock(tblk);
865                 else {
866                         /* state transition: COMMIT -> COMMITTED */
867                         tblk->flag |= tblkGC_COMMITTED;
868
869                         if (tblk->flag & tblkGC_READY)
870                                 log->gcrtc--;
871
872                         LOGGC_WAKEUP(tblk);
873                 }
874
875                 /* was page full before pageout ?
876                  * (and this is the last tblk bound with the page)
877                  */
878                 if (tblk->flag & tblkGC_FREE)
879                         lbmFree(bp);
880                 /* did page become full after pageout ?
881                  * (and this is the last tblk bound with the page)
882                  */
883                 else if (tblk->flag & tblkGC_EOP) {
884                         /* finalize the page */
885                         lp = (struct logpage *) bp->l_ldata;
886                         bp->l_ceor = bp->l_eor;
887                         lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_eor);
888                         jfs_info("lmPostGC: calling lbmWrite");
889                         lbmWrite(log, bp, lbmWRITE | lbmRELEASE | lbmFREE,
890                                  1);
891                 }
892
893         }
894
895         /* are there any transactions who have entered lnGroupCommit()
896          * (whose COMMITs are after that of the last log page written.
897          * They are waiting for new group commit (above at (SLEEP 1))
898          * or lazy transactions are on a full (queued) log page,
899          * select the latest ready transaction as new group leader and
900          * wake her up to lead her group.
901          */
902         if ((!list_empty(&log->cqueue)) &&
903             ((log->gcrtc > 0) || (tblk->bp->l_wqnext != NULL) ||
904              test_bit(log_FLUSH, &log->flag) || jfs_tlocks_low))
905                 /*
906                  * Call lmGCwrite with new group leader
907                  */
908                 lmGCwrite(log, 1);
909
910         /* no transaction are ready yet (transactions are only just
911          * queued (GC_QUEUE) and not entered for group commit yet).
912          * the first transaction entering group commit
913          * will elect herself as new group leader.
914          */
915         else
916                 log->cflag &= ~logGC_PAGEOUT;
917
918         //LOGGC_UNLOCK(log);
919         spin_unlock_irqrestore(&log->gclock, flags);
920         return;
921 }
922
923 /*
924  * NAME:        lmLogSync()
925  *
926  * FUNCTION:    write log SYNCPT record for specified log
927  *      if new sync address is available
928  *      (normally the case if sync() is executed by back-ground
929  *      process).
930  *      calculate new value of i_nextsync which determines when
931  *      this code is called again.
932  *
933  * PARAMETERS:  log     - log structure
934  *              hard_sync - 1 to force all metadata to be written
935  *
936  * RETURN:      0
937  *
938  * serialization: LOG_LOCK() held on entry/exit
939  */
940 static int lmLogSync(struct jfs_log * log, int hard_sync)
941 {
942         int logsize;
943         int written;            /* written since last syncpt */
944         int free;               /* free space left available */
945         int delta;              /* additional delta to write normally */
946         int more;               /* additional write granted */
947         struct lrd lrd;
948         int lsn;
949         struct logsyncblk *lp;
950         unsigned long flags;
951
952         /* push dirty metapages out to disk */
953         if (hard_sync)
954                 write_special_inodes(log, filemap_fdatawrite);
955         else
956                 write_special_inodes(log, filemap_flush);
957
958         /*
959          *      forward syncpt
960          */
961         /* if last sync is same as last syncpt,
962          * invoke sync point forward processing to update sync.
963          */
964
965         if (log->sync == log->syncpt) {
966                 LOGSYNC_LOCK(log, flags);
967                 if (list_empty(&log->synclist))
968                         log->sync = log->lsn;
969                 else {
970                         lp = list_entry(log->synclist.next,
971                                         struct logsyncblk, synclist);
972                         log->sync = lp->lsn;
973                 }
974                 LOGSYNC_UNLOCK(log, flags);
975
976         }
977
978         /* if sync is different from last syncpt,
979          * write a SYNCPT record with syncpt = sync.
980          * reset syncpt = sync
981          */
982         if (log->sync != log->syncpt) {
983                 lrd.logtid = 0;
984                 lrd.backchain = 0;
985                 lrd.type = cpu_to_le16(LOG_SYNCPT);
986                 lrd.length = 0;
987                 lrd.log.syncpt.sync = cpu_to_le32(log->sync);
988                 lsn = lmWriteRecord(log, NULL, &lrd, NULL);
989
990                 log->syncpt = log->sync;
991         } else
992                 lsn = log->lsn;
993
994         /*
995          *      setup next syncpt trigger (SWAG)
996          */
997         logsize = log->logsize;
998
999         logdiff(written, lsn, log);
1000         free = logsize - written;
1001         delta = LOGSYNC_DELTA(logsize);
1002         more = min(free / 2, delta);
1003         if (more < 2 * LOGPSIZE) {
1004                 jfs_warn("\n ... Log Wrap ... Log Wrap ... Log Wrap ...\n");
1005                 /*
1006                  *      log wrapping
1007                  *
1008                  * option 1 - panic ? No.!
1009                  * option 2 - shutdown file systems
1010                  *            associated with log ?
1011                  * option 3 - extend log ?
1012                  */
1013                 /*
1014                  * option 4 - second chance
1015                  *
1016                  * mark log wrapped, and continue.
1017                  * when all active transactions are completed,
1018                  * mark log vaild for recovery.
1019                  * if crashed during invalid state, log state
1020                  * implies invald log, forcing fsck().
1021                  */
1022                 /* mark log state log wrap in log superblock */
1023                 /* log->state = LOGWRAP; */
1024
1025                 /* reset sync point computation */
1026                 log->syncpt = log->sync = lsn;
1027                 log->nextsync = delta;
1028         } else
1029                 /* next syncpt trigger = written + more */
1030                 log->nextsync = written + more;
1031
1032         /* if number of bytes written from last sync point is more
1033          * than 1/4 of the log size, stop new transactions from
1034          * starting until all current transactions are completed
1035          * by setting syncbarrier flag.
1036          */
1037         if (!test_bit(log_SYNCBARRIER, &log->flag) &&
1038             (written > LOGSYNC_BARRIER(logsize)) && log->active) {
1039                 set_bit(log_SYNCBARRIER, &log->flag);
1040                 jfs_info("log barrier on: lsn=0x%x syncpt=0x%x", lsn,
1041                          log->syncpt);
1042                 /*
1043                  * We may have to initiate group commit
1044                  */
1045                 jfs_flush_journal(log, 0);
1046         }
1047
1048         return lsn;
1049 }
1050
1051 /*
1052  * NAME:        jfs_syncpt
1053  *
1054  * FUNCTION:    write log SYNCPT record for specified log
1055  *
1056  * PARAMETERS:  log       - log structure
1057  *              hard_sync - set to 1 to force metadata to be written
1058  */
1059 void jfs_syncpt(struct jfs_log *log, int hard_sync)
1060 {       LOG_LOCK(log);
1061         lmLogSync(log, hard_sync);
1062         LOG_UNLOCK(log);
1063 }
1064
1065 /*
1066  * NAME:        lmLogOpen()
1067  *
1068  * FUNCTION:    open the log on first open;
1069  *      insert filesystem in the active list of the log.
1070  *
1071  * PARAMETER:   ipmnt   - file system mount inode
1072  *              iplog   - log inode (out)
1073  *
1074  * RETURN:
1075  *
1076  * serialization:
1077  */
1078 int lmLogOpen(struct super_block *sb)
1079 {
1080         int rc;
1081         struct block_device *bdev;
1082         struct jfs_log *log;
1083         struct jfs_sb_info *sbi = JFS_SBI(sb);
1084
1085         if (sbi->flag & JFS_NOINTEGRITY)
1086                 return open_dummy_log(sb);
1087
1088         if (sbi->mntflag & JFS_INLINELOG)
1089                 return open_inline_log(sb);
1090
1091         mutex_lock(&jfs_log_mutex);
1092         list_for_each_entry(log, &jfs_external_logs, journal_list) {
1093                 if (log->bdev->bd_dev == sbi->logdev) {
1094                         if (memcmp(log->uuid, sbi->loguuid,
1095                                    sizeof(log->uuid))) {
1096                                 jfs_warn("wrong uuid on JFS journal\n");
1097                                 mutex_unlock(&jfs_log_mutex);
1098                                 return -EINVAL;
1099                         }
1100                         /*
1101                          * add file system to log active file system list
1102                          */
1103                         if ((rc = lmLogFileSystem(log, sbi, 1))) {
1104                                 mutex_unlock(&jfs_log_mutex);
1105                                 return rc;
1106                         }
1107                         goto journal_found;
1108                 }
1109         }
1110
1111         if (!(log = kzalloc(sizeof(struct jfs_log), GFP_KERNEL))) {
1112                 mutex_unlock(&jfs_log_mutex);
1113                 return -ENOMEM;
1114         }
1115         INIT_LIST_HEAD(&log->sb_list);
1116         init_waitqueue_head(&log->syncwait);
1117
1118         /*
1119          *      external log as separate logical volume
1120          *
1121          * file systems to log may have n-to-1 relationship;
1122          */
1123
1124         bdev = open_by_devnum(sbi->logdev, FMODE_READ|FMODE_WRITE);
1125         if (IS_ERR(bdev)) {
1126                 rc = -PTR_ERR(bdev);
1127                 goto free;
1128         }
1129
1130         if ((rc = bd_claim(bdev, log))) {
1131                 goto close;
1132         }
1133
1134         log->bdev = bdev;
1135         memcpy(log->uuid, sbi->loguuid, sizeof(log->uuid));
1136
1137         /*
1138          * initialize log:
1139          */
1140         if ((rc = lmLogInit(log)))
1141                 goto unclaim;
1142
1143         list_add(&log->journal_list, &jfs_external_logs);
1144
1145         /*
1146          * add file system to log active file system list
1147          */
1148         if ((rc = lmLogFileSystem(log, sbi, 1)))
1149                 goto shutdown;
1150
1151 journal_found:
1152         LOG_LOCK(log);
1153         list_add(&sbi->log_list, &log->sb_list);
1154         sbi->log = log;
1155         LOG_UNLOCK(log);
1156
1157         mutex_unlock(&jfs_log_mutex);
1158         return 0;
1159
1160         /*
1161          *      unwind on error
1162          */
1163       shutdown:         /* unwind lbmLogInit() */
1164         list_del(&log->journal_list);
1165         lbmLogShutdown(log);
1166
1167       unclaim:
1168         bd_release(bdev);
1169
1170       close:            /* close external log device */
1171         blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1172
1173       free:             /* free log descriptor */
1174         mutex_unlock(&jfs_log_mutex);
1175         kfree(log);
1176
1177         jfs_warn("lmLogOpen: exit(%d)", rc);
1178         return rc;
1179 }
1180
1181 static int open_inline_log(struct super_block *sb)
1182 {
1183         struct jfs_log *log;
1184         int rc;
1185
1186         if (!(log = kzalloc(sizeof(struct jfs_log), GFP_KERNEL)))
1187                 return -ENOMEM;
1188         INIT_LIST_HEAD(&log->sb_list);
1189         init_waitqueue_head(&log->syncwait);
1190
1191         set_bit(log_INLINELOG, &log->flag);
1192         log->bdev = sb->s_bdev;
1193         log->base = addressPXD(&JFS_SBI(sb)->logpxd);
1194         log->size = lengthPXD(&JFS_SBI(sb)->logpxd) >>
1195             (L2LOGPSIZE - sb->s_blocksize_bits);
1196         log->l2bsize = sb->s_blocksize_bits;
1197         ASSERT(L2LOGPSIZE >= sb->s_blocksize_bits);
1198
1199         /*
1200          * initialize log.
1201          */
1202         if ((rc = lmLogInit(log))) {
1203                 kfree(log);
1204                 jfs_warn("lmLogOpen: exit(%d)", rc);
1205                 return rc;
1206         }
1207
1208         list_add(&JFS_SBI(sb)->log_list, &log->sb_list);
1209         JFS_SBI(sb)->log = log;
1210
1211         return rc;
1212 }
1213
1214 static int open_dummy_log(struct super_block *sb)
1215 {
1216         int rc;
1217
1218         mutex_lock(&jfs_log_mutex);
1219         if (!dummy_log) {
1220                 dummy_log = kzalloc(sizeof(struct jfs_log), GFP_KERNEL);
1221                 if (!dummy_log) {
1222                         mutex_unlock(&jfs_log_mutex);
1223                         return -ENOMEM;
1224                 }
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                         mutex_unlock(&jfs_log_mutex);
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         mutex_unlock(&jfs_log_mutex);
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         mutex_lock(&jfs_log_mutex);
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, FMODE_READ|FMODE_WRITE);
1518
1519         kfree(log);
1520
1521       out:
1522         mutex_unlock(&jfs_log_mutex);
1523         jfs_info("lmLogClose: exit(%d)", rc);
1524         return rc;
1525 }
1526
1527
1528 /*
1529  * NAME:        jfs_flush_journal()
1530  *
1531  * FUNCTION:    initiate write of any outstanding transactions to the journal
1532  *              and optionally wait until they are all written to disk
1533  *
1534  *              wait == 0  flush until latest txn is committed, don't wait
1535  *              wait == 1  flush until latest txn is committed, wait
1536  *              wait > 1   flush until all txn's are complete, wait
1537  */
1538 void jfs_flush_journal(struct jfs_log *log, int wait)
1539 {
1540         int i;
1541         struct tblock *target = NULL;
1542
1543         /* jfs_write_inode may call us during read-only mount */
1544         if (!log)
1545                 return;
1546
1547         jfs_info("jfs_flush_journal: log:0x%p wait=%d", log, wait);
1548
1549         LOGGC_LOCK(log);
1550
1551         if (!list_empty(&log->cqueue)) {
1552                 /*
1553                  * This ensures that we will keep writing to the journal as long
1554                  * as there are unwritten commit records
1555                  */
1556                 target = list_entry(log->cqueue.prev, struct tblock, cqueue);
1557
1558                 if (test_bit(log_FLUSH, &log->flag)) {
1559                         /*
1560                          * We're already flushing.
1561                          * if flush_tblk is NULL, we are flushing everything,
1562                          * so leave it that way.  Otherwise, update it to the
1563                          * latest transaction
1564                          */
1565                         if (log->flush_tblk)
1566                                 log->flush_tblk = target;
1567                 } else {
1568                         /* Only flush until latest transaction is committed */
1569                         log->flush_tblk = target;
1570                         set_bit(log_FLUSH, &log->flag);
1571
1572                         /*
1573                          * Initiate I/O on outstanding transactions
1574                          */
1575                         if (!(log->cflag & logGC_PAGEOUT)) {
1576                                 log->cflag |= logGC_PAGEOUT;
1577                                 lmGCwrite(log, 0);
1578                         }
1579                 }
1580         }
1581         if ((wait > 1) || test_bit(log_SYNCBARRIER, &log->flag)) {
1582                 /* Flush until all activity complete */
1583                 set_bit(log_FLUSH, &log->flag);
1584                 log->flush_tblk = NULL;
1585         }
1586
1587         if (wait && target && !(target->flag & tblkGC_COMMITTED)) {
1588                 DECLARE_WAITQUEUE(__wait, current);
1589
1590                 add_wait_queue(&target->gcwait, &__wait);
1591                 set_current_state(TASK_UNINTERRUPTIBLE);
1592                 LOGGC_UNLOCK(log);
1593                 schedule();
1594                 __set_current_state(TASK_RUNNING);
1595                 LOGGC_LOCK(log);
1596                 remove_wait_queue(&target->gcwait, &__wait);
1597         }
1598         LOGGC_UNLOCK(log);
1599
1600         if (wait < 2)
1601                 return;
1602
1603         write_special_inodes(log, filemap_fdatawrite);
1604
1605         /*
1606          * If there was recent activity, we may need to wait
1607          * for the lazycommit thread to catch up
1608          */
1609         if ((!list_empty(&log->cqueue)) || !list_empty(&log->synclist)) {
1610                 for (i = 0; i < 200; i++) {     /* Too much? */
1611                         msleep(250);
1612                         write_special_inodes(log, filemap_fdatawrite);
1613                         if (list_empty(&log->cqueue) &&
1614                             list_empty(&log->synclist))
1615                                 break;
1616                 }
1617         }
1618         assert(list_empty(&log->cqueue));
1619
1620 #ifdef CONFIG_JFS_DEBUG
1621         if (!list_empty(&log->synclist)) {
1622                 struct logsyncblk *lp;
1623
1624                 printk(KERN_ERR "jfs_flush_journal: synclist not empty\n");
1625                 list_for_each_entry(lp, &log->synclist, synclist) {
1626                         if (lp->xflag & COMMIT_PAGE) {
1627                                 struct metapage *mp = (struct metapage *)lp;
1628                                 print_hex_dump(KERN_ERR, "metapage: ",
1629                                                DUMP_PREFIX_ADDRESS, 16, 4,
1630                                                mp, sizeof(struct metapage), 0);
1631                                 print_hex_dump(KERN_ERR, "page: ",
1632                                                DUMP_PREFIX_ADDRESS, 16,
1633                                                sizeof(long), mp->page,
1634                                                sizeof(struct page), 0);
1635                         } else
1636                                 print_hex_dump(KERN_ERR, "tblock:",
1637                                                DUMP_PREFIX_ADDRESS, 16, 4,
1638                                                lp, sizeof(struct tblock), 0);
1639                 }
1640         }
1641 #else
1642         WARN_ON(!list_empty(&log->synclist));
1643 #endif
1644         clear_bit(log_FLUSH, &log->flag);
1645 }
1646
1647 /*
1648  * NAME:        lmLogShutdown()
1649  *
1650  * FUNCTION:    log shutdown at last LogClose().
1651  *
1652  *              write log syncpt record.
1653  *              update super block to set redone flag to 0.
1654  *
1655  * PARAMETER:   log     - log inode
1656  *
1657  * RETURN:      0       - success
1658  *
1659  * serialization: single last close thread
1660  */
1661 int lmLogShutdown(struct jfs_log * log)
1662 {
1663         int rc;
1664         struct lrd lrd;
1665         int lsn;
1666         struct logsuper *logsuper;
1667         struct lbuf *bpsuper;
1668         struct lbuf *bp;
1669         struct logpage *lp;
1670
1671         jfs_info("lmLogShutdown: log:0x%p", log);
1672
1673         jfs_flush_journal(log, 2);
1674
1675         /*
1676          * write the last SYNCPT record with syncpoint = 0
1677          * (i.e., log redo up to HERE !)
1678          */
1679         lrd.logtid = 0;
1680         lrd.backchain = 0;
1681         lrd.type = cpu_to_le16(LOG_SYNCPT);
1682         lrd.length = 0;
1683         lrd.log.syncpt.sync = 0;
1684
1685         lsn = lmWriteRecord(log, NULL, &lrd, NULL);
1686         bp = log->bp;
1687         lp = (struct logpage *) bp->l_ldata;
1688         lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_eor);
1689         lbmWrite(log, log->bp, lbmWRITE | lbmRELEASE | lbmSYNC, 0);
1690         lbmIOWait(log->bp, lbmFREE);
1691         log->bp = NULL;
1692
1693         /*
1694          * synchronous update log superblock
1695          * mark log state as shutdown cleanly
1696          * (i.e., Log does not need to be replayed).
1697          */
1698         if ((rc = lbmRead(log, 1, &bpsuper)))
1699                 goto out;
1700
1701         logsuper = (struct logsuper *) bpsuper->l_ldata;
1702         logsuper->state = cpu_to_le32(LOGREDONE);
1703         logsuper->end = cpu_to_le32(lsn);
1704         lbmDirectWrite(log, bpsuper, lbmWRITE | lbmRELEASE | lbmSYNC);
1705         rc = lbmIOWait(bpsuper, lbmFREE);
1706
1707         jfs_info("lmLogShutdown: lsn:0x%x page:%d eor:%d",
1708                  lsn, log->page, log->eor);
1709
1710       out:
1711         /*
1712          * shutdown per log i/o
1713          */
1714         lbmLogShutdown(log);
1715
1716         if (rc) {
1717                 jfs_warn("lmLogShutdown: exit(%d)", rc);
1718         }
1719         return rc;
1720 }
1721
1722
1723 /*
1724  * NAME:        lmLogFileSystem()
1725  *
1726  * FUNCTION:    insert (<activate> = true)/remove (<activate> = false)
1727  *      file system into/from log active file system list.
1728  *
1729  * PARAMETE:    log     - pointer to logs inode.
1730  *              fsdev   - kdev_t of filesystem.
1731  *              serial  - pointer to returned log serial number
1732  *              activate - insert/remove device from active list.
1733  *
1734  * RETURN:      0       - success
1735  *              errors returned by vms_iowait().
1736  */
1737 static int lmLogFileSystem(struct jfs_log * log, struct jfs_sb_info *sbi,
1738                            int activate)
1739 {
1740         int rc = 0;
1741         int i;
1742         struct logsuper *logsuper;
1743         struct lbuf *bpsuper;
1744         char *uuid = sbi->uuid;
1745
1746         /*
1747          * insert/remove file system device to log active file system list.
1748          */
1749         if ((rc = lbmRead(log, 1, &bpsuper)))
1750                 return rc;
1751
1752         logsuper = (struct logsuper *) bpsuper->l_ldata;
1753         if (activate) {
1754                 for (i = 0; i < MAX_ACTIVE; i++)
1755                         if (!memcmp(logsuper->active[i].uuid, NULL_UUID, 16)) {
1756                                 memcpy(logsuper->active[i].uuid, uuid, 16);
1757                                 sbi->aggregate = i;
1758                                 break;
1759                         }
1760                 if (i == MAX_ACTIVE) {
1761                         jfs_warn("Too many file systems sharing journal!");
1762                         lbmFree(bpsuper);
1763                         return -EMFILE; /* Is there a better rc? */
1764                 }
1765         } else {
1766                 for (i = 0; i < MAX_ACTIVE; i++)
1767                         if (!memcmp(logsuper->active[i].uuid, uuid, 16)) {
1768                                 memcpy(logsuper->active[i].uuid, NULL_UUID, 16);
1769                                 break;
1770                         }
1771                 if (i == MAX_ACTIVE) {
1772                         jfs_warn("Somebody stomped on the journal!");
1773                         lbmFree(bpsuper);
1774                         return -EIO;
1775                 }
1776
1777         }
1778
1779         /*
1780          * synchronous write log superblock:
1781          *
1782          * write sidestream bypassing write queue:
1783          * at file system mount, log super block is updated for
1784          * activation of the file system before any log record
1785          * (MOUNT record) of the file system, and at file system
1786          * unmount, all meta data for the file system has been
1787          * flushed before log super block is updated for deactivation
1788          * of the file system.
1789          */
1790         lbmDirectWrite(log, bpsuper, lbmWRITE | lbmRELEASE | lbmSYNC);
1791         rc = lbmIOWait(bpsuper, lbmFREE);
1792
1793         return rc;
1794 }
1795
1796 /*
1797  *              log buffer manager (lbm)
1798  *              ------------------------
1799  *
1800  * special purpose buffer manager supporting log i/o requirements.
1801  *
1802  * per log write queue:
1803  * log pageout occurs in serial order by fifo write queue and
1804  * restricting to a single i/o in pregress at any one time.
1805  * a circular singly-linked list
1806  * (log->wrqueue points to the tail, and buffers are linked via
1807  * bp->wrqueue field), and
1808  * maintains log page in pageout ot waiting for pageout in serial pageout.
1809  */
1810
1811 /*
1812  *      lbmLogInit()
1813  *
1814  * initialize per log I/O setup at lmLogInit()
1815  */
1816 static int lbmLogInit(struct jfs_log * log)
1817 {                               /* log inode */
1818         int i;
1819         struct lbuf *lbuf;
1820
1821         jfs_info("lbmLogInit: log:0x%p", log);
1822
1823         /* initialize current buffer cursor */
1824         log->bp = NULL;
1825
1826         /* initialize log device write queue */
1827         log->wqueue = NULL;
1828
1829         /*
1830          * Each log has its own buffer pages allocated to it.  These are
1831          * not managed by the page cache.  This ensures that a transaction
1832          * writing to the log does not block trying to allocate a page from
1833          * the page cache (for the log).  This would be bad, since page
1834          * allocation waits on the kswapd thread that may be committing inodes
1835          * which would cause log activity.  Was that clear?  I'm trying to
1836          * avoid deadlock here.
1837          */
1838         init_waitqueue_head(&log->free_wait);
1839
1840         log->lbuf_free = NULL;
1841
1842         for (i = 0; i < LOGPAGES;) {
1843                 char *buffer;
1844                 uint offset;
1845                 struct page *page;
1846
1847                 buffer = (char *) get_zeroed_page(GFP_KERNEL);
1848                 if (buffer == NULL)
1849                         goto error;
1850                 page = virt_to_page(buffer);
1851                 for (offset = 0; offset < PAGE_SIZE; offset += LOGPSIZE) {
1852                         lbuf = kmalloc(sizeof(struct lbuf), GFP_KERNEL);
1853                         if (lbuf == NULL) {
1854                                 if (offset == 0)
1855                                         free_page((unsigned long) buffer);
1856                                 goto error;
1857                         }
1858                         if (offset) /* we already have one reference */
1859                                 get_page(page);
1860                         lbuf->l_offset = offset;
1861                         lbuf->l_ldata = buffer + offset;
1862                         lbuf->l_page = page;
1863                         lbuf->l_log = log;
1864                         init_waitqueue_head(&lbuf->l_ioevent);
1865
1866                         lbuf->l_freelist = log->lbuf_free;
1867                         log->lbuf_free = lbuf;
1868                         i++;
1869                 }
1870         }
1871
1872         return (0);
1873
1874       error:
1875         lbmLogShutdown(log);
1876         return -ENOMEM;
1877 }
1878
1879
1880 /*
1881  *      lbmLogShutdown()
1882  *
1883  * finalize per log I/O setup at lmLogShutdown()
1884  */
1885 static void lbmLogShutdown(struct jfs_log * log)
1886 {
1887         struct lbuf *lbuf;
1888
1889         jfs_info("lbmLogShutdown: log:0x%p", log);
1890
1891         lbuf = log->lbuf_free;
1892         while (lbuf) {
1893                 struct lbuf *next = lbuf->l_freelist;
1894                 __free_page(lbuf->l_page);
1895                 kfree(lbuf);
1896                 lbuf = next;
1897         }
1898 }
1899
1900
1901 /*
1902  *      lbmAllocate()
1903  *
1904  * allocate an empty log buffer
1905  */
1906 static struct lbuf *lbmAllocate(struct jfs_log * log, int pn)
1907 {
1908         struct lbuf *bp;
1909         unsigned long flags;
1910
1911         /*
1912          * recycle from log buffer freelist if any
1913          */
1914         LCACHE_LOCK(flags);
1915         LCACHE_SLEEP_COND(log->free_wait, (bp = log->lbuf_free), flags);
1916         log->lbuf_free = bp->l_freelist;
1917         LCACHE_UNLOCK(flags);
1918
1919         bp->l_flag = 0;
1920
1921         bp->l_wqnext = NULL;
1922         bp->l_freelist = NULL;
1923
1924         bp->l_pn = pn;
1925         bp->l_blkno = log->base + (pn << (L2LOGPSIZE - log->l2bsize));
1926         bp->l_ceor = 0;
1927
1928         return bp;
1929 }
1930
1931
1932 /*
1933  *      lbmFree()
1934  *
1935  * release a log buffer to freelist
1936  */
1937 static void lbmFree(struct lbuf * bp)
1938 {
1939         unsigned long flags;
1940
1941         LCACHE_LOCK(flags);
1942
1943         lbmfree(bp);
1944
1945         LCACHE_UNLOCK(flags);
1946 }
1947
1948 static void lbmfree(struct lbuf * bp)
1949 {
1950         struct jfs_log *log = bp->l_log;
1951
1952         assert(bp->l_wqnext == NULL);
1953
1954         /*
1955          * return the buffer to head of freelist
1956          */
1957         bp->l_freelist = log->lbuf_free;
1958         log->lbuf_free = bp;
1959
1960         wake_up(&log->free_wait);
1961         return;
1962 }
1963
1964
1965 /*
1966  * NAME:        lbmRedrive
1967  *
1968  * FUNCTION:    add a log buffer to the log redrive list
1969  *
1970  * PARAMETER:
1971  *      bp      - log buffer
1972  *
1973  * NOTES:
1974  *      Takes log_redrive_lock.
1975  */
1976 static inline void lbmRedrive(struct lbuf *bp)
1977 {
1978         unsigned long flags;
1979
1980         spin_lock_irqsave(&log_redrive_lock, flags);
1981         bp->l_redrive_next = log_redrive_list;
1982         log_redrive_list = bp;
1983         spin_unlock_irqrestore(&log_redrive_lock, flags);
1984
1985         wake_up_process(jfsIOthread);
1986 }
1987
1988
1989 /*
1990  *      lbmRead()
1991  */
1992 static int lbmRead(struct jfs_log * log, int pn, struct lbuf ** bpp)
1993 {
1994         struct bio *bio;
1995         struct lbuf *bp;
1996
1997         /*
1998          * allocate a log buffer
1999          */
2000         *bpp = bp = lbmAllocate(log, pn);
2001         jfs_info("lbmRead: bp:0x%p pn:0x%x", bp, pn);
2002
2003         bp->l_flag |= lbmREAD;
2004
2005         bio = bio_alloc(GFP_NOFS, 1);
2006
2007         bio->bi_sector = bp->l_blkno << (log->l2bsize - 9);
2008         bio->bi_bdev = log->bdev;
2009         bio->bi_io_vec[0].bv_page = bp->l_page;
2010         bio->bi_io_vec[0].bv_len = LOGPSIZE;
2011         bio->bi_io_vec[0].bv_offset = bp->l_offset;
2012
2013         bio->bi_vcnt = 1;
2014         bio->bi_idx = 0;
2015         bio->bi_size = LOGPSIZE;
2016
2017         bio->bi_end_io = lbmIODone;
2018         bio->bi_private = bp;
2019         submit_bio(READ_SYNC, bio);
2020
2021         wait_event(bp->l_ioevent, (bp->l_flag != lbmREAD));
2022
2023         return 0;
2024 }
2025
2026
2027 /*
2028  *      lbmWrite()
2029  *
2030  * buffer at head of pageout queue stays after completion of
2031  * partial-page pageout and redriven by explicit initiation of
2032  * pageout by caller until full-page pageout is completed and
2033  * released.
2034  *
2035  * device driver i/o done redrives pageout of new buffer at
2036  * head of pageout queue when current buffer at head of pageout
2037  * queue is released at the completion of its full-page pageout.
2038  *
2039  * LOGGC_LOCK() serializes lbmWrite() by lmNextPage() and lmGroupCommit().
2040  * LCACHE_LOCK() serializes xflag between lbmWrite() and lbmIODone()
2041  */
2042 static void lbmWrite(struct jfs_log * log, struct lbuf * bp, int flag,
2043                      int cant_block)
2044 {
2045         struct lbuf *tail;
2046         unsigned long flags;
2047
2048         jfs_info("lbmWrite: bp:0x%p flag:0x%x pn:0x%x", bp, flag, bp->l_pn);
2049
2050         /* map the logical block address to physical block address */
2051         bp->l_blkno =
2052             log->base + (bp->l_pn << (L2LOGPSIZE - log->l2bsize));
2053
2054         LCACHE_LOCK(flags);             /* disable+lock */
2055
2056         /*
2057          * initialize buffer for device driver
2058          */
2059         bp->l_flag = flag;
2060
2061         /*
2062          *      insert bp at tail of write queue associated with log
2063          *
2064          * (request is either for bp already/currently at head of queue
2065          * or new bp to be inserted at tail)
2066          */
2067         tail = log->wqueue;
2068
2069         /* is buffer not already on write queue ? */
2070         if (bp->l_wqnext == NULL) {
2071                 /* insert at tail of wqueue */
2072                 if (tail == NULL) {
2073                         log->wqueue = bp;
2074                         bp->l_wqnext = bp;
2075                 } else {
2076                         log->wqueue = bp;
2077                         bp->l_wqnext = tail->l_wqnext;
2078                         tail->l_wqnext = bp;
2079                 }
2080
2081                 tail = bp;
2082         }
2083
2084         /* is buffer at head of wqueue and for write ? */
2085         if ((bp != tail->l_wqnext) || !(flag & lbmWRITE)) {
2086                 LCACHE_UNLOCK(flags);   /* unlock+enable */
2087                 return;
2088         }
2089
2090         LCACHE_UNLOCK(flags);   /* unlock+enable */
2091
2092         if (cant_block)
2093                 lbmRedrive(bp);
2094         else if (flag & lbmSYNC)
2095                 lbmStartIO(bp);
2096         else {
2097                 LOGGC_UNLOCK(log);
2098                 lbmStartIO(bp);
2099                 LOGGC_LOCK(log);
2100         }
2101 }
2102
2103
2104 /*
2105  *      lbmDirectWrite()
2106  *
2107  * initiate pageout bypassing write queue for sidestream
2108  * (e.g., log superblock) write;
2109  */
2110 static void lbmDirectWrite(struct jfs_log * log, struct lbuf * bp, int flag)
2111 {
2112         jfs_info("lbmDirectWrite: bp:0x%p flag:0x%x pn:0x%x",
2113                  bp, flag, bp->l_pn);
2114
2115         /*
2116          * initialize buffer for device driver
2117          */
2118         bp->l_flag = flag | lbmDIRECT;
2119
2120         /* map the logical block address to physical block address */
2121         bp->l_blkno =
2122             log->base + (bp->l_pn << (L2LOGPSIZE - log->l2bsize));
2123
2124         /*
2125          *      initiate pageout of the page
2126          */
2127         lbmStartIO(bp);
2128 }
2129
2130
2131 /*
2132  * NAME:        lbmStartIO()
2133  *
2134  * FUNCTION:    Interface to DD strategy routine
2135  *
2136  * RETURN:      none
2137  *
2138  * serialization: LCACHE_LOCK() is NOT held during log i/o;
2139  */
2140 static void lbmStartIO(struct lbuf * bp)
2141 {
2142         struct bio *bio;
2143         struct jfs_log *log = bp->l_log;
2144
2145         jfs_info("lbmStartIO\n");
2146
2147         bio = bio_alloc(GFP_NOFS, 1);
2148         bio->bi_sector = bp->l_blkno << (log->l2bsize - 9);
2149         bio->bi_bdev = log->bdev;
2150         bio->bi_io_vec[0].bv_page = bp->l_page;
2151         bio->bi_io_vec[0].bv_len = LOGPSIZE;
2152         bio->bi_io_vec[0].bv_offset = bp->l_offset;
2153
2154         bio->bi_vcnt = 1;
2155         bio->bi_idx = 0;
2156         bio->bi_size = LOGPSIZE;
2157
2158         bio->bi_end_io = lbmIODone;
2159         bio->bi_private = bp;
2160
2161         /* check if journaling to disk has been disabled */
2162         if (log->no_integrity) {
2163                 bio->bi_size = 0;
2164                 lbmIODone(bio, 0);
2165         } else {
2166                 submit_bio(WRITE_SYNC, bio);
2167                 INCREMENT(lmStat.submitted);
2168         }
2169 }
2170
2171
2172 /*
2173  *      lbmIOWait()
2174  */
2175 static int lbmIOWait(struct lbuf * bp, int flag)
2176 {
2177         unsigned long flags;
2178         int rc = 0;
2179
2180         jfs_info("lbmIOWait1: bp:0x%p flag:0x%x:0x%x", bp, bp->l_flag, flag);
2181
2182         LCACHE_LOCK(flags);             /* disable+lock */
2183
2184         LCACHE_SLEEP_COND(bp->l_ioevent, (bp->l_flag & lbmDONE), flags);
2185
2186         rc = (bp->l_flag & lbmERROR) ? -EIO : 0;
2187
2188         if (flag & lbmFREE)
2189                 lbmfree(bp);
2190
2191         LCACHE_UNLOCK(flags);   /* unlock+enable */
2192
2193         jfs_info("lbmIOWait2: bp:0x%p flag:0x%x:0x%x", bp, bp->l_flag, flag);
2194         return rc;
2195 }
2196
2197 /*
2198  *      lbmIODone()
2199  *
2200  * executed at INTIODONE level
2201  */
2202 static void lbmIODone(struct bio *bio, int error)
2203 {
2204         struct lbuf *bp = bio->bi_private;
2205         struct lbuf *nextbp, *tail;
2206         struct jfs_log *log;
2207         unsigned long flags;
2208
2209         /*
2210          * get back jfs buffer bound to the i/o buffer
2211          */
2212         jfs_info("lbmIODone: bp:0x%p flag:0x%x", bp, bp->l_flag);
2213
2214         LCACHE_LOCK(flags);             /* disable+lock */
2215
2216         bp->l_flag |= lbmDONE;
2217
2218         if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
2219                 bp->l_flag |= lbmERROR;
2220
2221                 jfs_err("lbmIODone: I/O error in JFS log");
2222         }
2223
2224         bio_put(bio);
2225
2226         /*
2227          *      pagein completion
2228          */
2229         if (bp->l_flag & lbmREAD) {
2230                 bp->l_flag &= ~lbmREAD;
2231
2232                 LCACHE_UNLOCK(flags);   /* unlock+enable */
2233
2234                 /* wakeup I/O initiator */
2235                 LCACHE_WAKEUP(&bp->l_ioevent);
2236
2237                 return;
2238         }
2239
2240         /*
2241          *      pageout completion
2242          *
2243          * the bp at the head of write queue has completed pageout.
2244          *
2245          * if single-commit/full-page pageout, remove the current buffer
2246          * from head of pageout queue, and redrive pageout with
2247          * the new buffer at head of pageout queue;
2248          * otherwise, the partial-page pageout buffer stays at
2249          * the head of pageout queue to be redriven for pageout
2250          * by lmGroupCommit() until full-page pageout is completed.
2251          */
2252         bp->l_flag &= ~lbmWRITE;
2253         INCREMENT(lmStat.pagedone);
2254
2255         /* update committed lsn */
2256         log = bp->l_log;
2257         log->clsn = (bp->l_pn << L2LOGPSIZE) + bp->l_ceor;
2258
2259         if (bp->l_flag & lbmDIRECT) {
2260                 LCACHE_WAKEUP(&bp->l_ioevent);
2261                 LCACHE_UNLOCK(flags);
2262                 return;
2263         }
2264
2265         tail = log->wqueue;
2266
2267         /* single element queue */
2268         if (bp == tail) {
2269                 /* remove head buffer of full-page pageout
2270                  * from log device write queue
2271                  */
2272                 if (bp->l_flag & lbmRELEASE) {
2273                         log->wqueue = NULL;
2274                         bp->l_wqnext = NULL;
2275                 }
2276         }
2277         /* multi element queue */
2278         else {
2279                 /* remove head buffer of full-page pageout
2280                  * from log device write queue
2281                  */
2282                 if (bp->l_flag & lbmRELEASE) {
2283                         nextbp = tail->l_wqnext = bp->l_wqnext;
2284                         bp->l_wqnext = NULL;
2285
2286                         /*
2287                          * redrive pageout of next page at head of write queue:
2288                          * redrive next page without any bound tblk
2289                          * (i.e., page w/o any COMMIT records), or
2290                          * first page of new group commit which has been
2291                          * queued after current page (subsequent pageout
2292                          * is performed synchronously, except page without
2293                          * any COMMITs) by lmGroupCommit() as indicated
2294                          * by lbmWRITE flag;
2295                          */
2296                         if (nextbp->l_flag & lbmWRITE) {
2297                                 /*
2298                                  * We can't do the I/O at interrupt time.
2299                                  * The jfsIO thread can do it
2300                                  */
2301                                 lbmRedrive(nextbp);
2302                         }
2303                 }
2304         }
2305
2306         /*
2307          *      synchronous pageout:
2308          *
2309          * buffer has not necessarily been removed from write queue
2310          * (e.g., synchronous write of partial-page with COMMIT):
2311          * leave buffer for i/o initiator to dispose
2312          */
2313         if (bp->l_flag & lbmSYNC) {
2314                 LCACHE_UNLOCK(flags);   /* unlock+enable */
2315
2316                 /* wakeup I/O initiator */
2317                 LCACHE_WAKEUP(&bp->l_ioevent);
2318         }
2319
2320         /*
2321          *      Group Commit pageout:
2322          */
2323         else if (bp->l_flag & lbmGC) {
2324                 LCACHE_UNLOCK(flags);
2325                 lmPostGC(bp);
2326         }
2327
2328         /*
2329          *      asynchronous pageout:
2330          *
2331          * buffer must have been removed from write queue:
2332          * insert buffer at head of freelist where it can be recycled
2333          */
2334         else {
2335                 assert(bp->l_flag & lbmRELEASE);
2336                 assert(bp->l_flag & lbmFREE);
2337                 lbmfree(bp);
2338
2339                 LCACHE_UNLOCK(flags);   /* unlock+enable */
2340         }
2341 }
2342
2343 int jfsIOWait(void *arg)
2344 {
2345         struct lbuf *bp;
2346
2347         do {
2348                 spin_lock_irq(&log_redrive_lock);
2349                 while ((bp = log_redrive_list)) {
2350                         log_redrive_list = bp->l_redrive_next;
2351                         bp->l_redrive_next = NULL;
2352                         spin_unlock_irq(&log_redrive_lock);
2353                         lbmStartIO(bp);
2354                         spin_lock_irq(&log_redrive_lock);
2355                 }
2356
2357                 if (freezing(current)) {
2358                         spin_unlock_irq(&log_redrive_lock);
2359                         refrigerator();
2360                 } else {
2361                         set_current_state(TASK_INTERRUPTIBLE);
2362                         spin_unlock_irq(&log_redrive_lock);
2363                         schedule();
2364                         __set_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 static int jfs_lmstats_proc_show(struct seq_file *m, void *v)
2508 {
2509         seq_printf(m,
2510                        "JFS Logmgr stats\n"
2511                        "================\n"
2512                        "commits = %d\n"
2513                        "writes submitted = %d\n"
2514                        "writes completed = %d\n"
2515                        "full pages submitted = %d\n"
2516                        "partial pages submitted = %d\n",
2517                        lmStat.commit,
2518                        lmStat.submitted,
2519                        lmStat.pagedone,
2520                        lmStat.full_page,
2521                        lmStat.partial_page);
2522         return 0;
2523 }
2524
2525 static int jfs_lmstats_proc_open(struct inode *inode, struct file *file)
2526 {
2527         return single_open(file, jfs_lmstats_proc_show, NULL);
2528 }
2529
2530 const struct file_operations jfs_lmstats_proc_fops = {
2531         .owner          = THIS_MODULE,
2532         .open           = jfs_lmstats_proc_open,
2533         .read           = seq_read,
2534         .llseek         = seq_lseek,
2535         .release        = single_release,
2536 };
2537 #endif /* CONFIG_JFS_STATISTICS */