xref: /openbmc/linux/fs/jfs/jfs_logmgr.c (revision a1e58bbd)
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 "jfs_incore.h"
73 #include "jfs_filsys.h"
74 #include "jfs_metapage.h"
75 #include "jfs_superblock.h"
76 #include "jfs_txnmgr.h"
77 #include "jfs_debug.h"
78 
79 
80 /*
81  * lbuf's ready to be redriven.  Protected by log_redrive_lock (jfsIO thread)
82  */
83 static struct lbuf *log_redrive_list;
84 static DEFINE_SPINLOCK(log_redrive_lock);
85 
86 
87 /*
88  *	log read/write serialization (per log)
89  */
90 #define LOG_LOCK_INIT(log)	mutex_init(&(log)->loglock)
91 #define LOG_LOCK(log)		mutex_lock(&((log)->loglock))
92 #define LOG_UNLOCK(log)		mutex_unlock(&((log)->loglock))
93 
94 
95 /*
96  *	log group commit serialization (per log)
97  */
98 
99 #define LOGGC_LOCK_INIT(log)	spin_lock_init(&(log)->gclock)
100 #define LOGGC_LOCK(log)		spin_lock_irq(&(log)->gclock)
101 #define LOGGC_UNLOCK(log)	spin_unlock_irq(&(log)->gclock)
102 #define LOGGC_WAKEUP(tblk)	wake_up_all(&(tblk)->gcwait)
103 
104 /*
105  *	log sync serialization (per log)
106  */
107 #define	LOGSYNC_DELTA(logsize)		min((logsize)/8, 128*LOGPSIZE)
108 #define	LOGSYNC_BARRIER(logsize)	((logsize)/4)
109 /*
110 #define	LOGSYNC_DELTA(logsize)		min((logsize)/4, 256*LOGPSIZE)
111 #define	LOGSYNC_BARRIER(logsize)	((logsize)/2)
112 */
113 
114 
115 /*
116  *	log buffer cache synchronization
117  */
118 static DEFINE_SPINLOCK(jfsLCacheLock);
119 
120 #define	LCACHE_LOCK(flags)	spin_lock_irqsave(&jfsLCacheLock, flags)
121 #define	LCACHE_UNLOCK(flags)	spin_unlock_irqrestore(&jfsLCacheLock, flags)
122 
123 /*
124  * See __SLEEP_COND in jfs_locks.h
125  */
126 #define LCACHE_SLEEP_COND(wq, cond, flags)	\
127 do {						\
128 	if (cond)				\
129 		break;				\
130 	__SLEEP_COND(wq, cond, LCACHE_LOCK(flags), LCACHE_UNLOCK(flags)); \
131 } while (0)
132 
133 #define	LCACHE_WAKEUP(event)	wake_up(event)
134 
135 
136 /*
137  *	lbuf buffer cache (lCache) control
138  */
139 /* log buffer manager pageout control (cumulative, inclusive) */
140 #define	lbmREAD		0x0001
141 #define	lbmWRITE	0x0002	/* enqueue at tail of write queue;
142 				 * init pageout if at head of queue;
143 				 */
144 #define	lbmRELEASE	0x0004	/* remove from write queue
145 				 * at completion of pageout;
146 				 * do not free/recycle it yet:
147 				 * caller will free it;
148 				 */
149 #define	lbmSYNC		0x0008	/* do not return to freelist
150 				 * when removed from write queue;
151 				 */
152 #define lbmFREE		0x0010	/* return to freelist
153 				 * at completion of pageout;
154 				 * the buffer may be recycled;
155 				 */
156 #define	lbmDONE		0x0020
157 #define	lbmERROR	0x0040
158 #define lbmGC		0x0080	/* lbmIODone to perform post-GC processing
159 				 * of log page
160 				 */
161 #define lbmDIRECT	0x0100
162 
163 /*
164  * Global list of active external journals
165  */
166 static LIST_HEAD(jfs_external_logs);
167 static struct jfs_log *dummy_log = NULL;
168 static DEFINE_MUTEX(jfs_log_mutex);
169 
170 /*
171  * forward references
172  */
173 static int lmWriteRecord(struct jfs_log * log, struct tblock * tblk,
174 			 struct lrd * lrd, struct tlock * tlck);
175 
176 static int lmNextPage(struct jfs_log * log);
177 static int lmLogFileSystem(struct jfs_log * log, struct jfs_sb_info *sbi,
178 			   int activate);
179 
180 static int open_inline_log(struct super_block *sb);
181 static int open_dummy_log(struct super_block *sb);
182 static int lbmLogInit(struct jfs_log * log);
183 static void lbmLogShutdown(struct jfs_log * log);
184 static struct lbuf *lbmAllocate(struct jfs_log * log, int);
185 static void lbmFree(struct lbuf * bp);
186 static void lbmfree(struct lbuf * bp);
187 static int lbmRead(struct jfs_log * log, int pn, struct lbuf ** bpp);
188 static void lbmWrite(struct jfs_log * log, struct lbuf * bp, int flag, int cant_block);
189 static void lbmDirectWrite(struct jfs_log * log, struct lbuf * bp, int flag);
190 static int lbmIOWait(struct lbuf * bp, int flag);
191 static bio_end_io_t lbmIODone;
192 static void lbmStartIO(struct lbuf * bp);
193 static void lmGCwrite(struct jfs_log * log, int cant_block);
194 static int lmLogSync(struct jfs_log * log, int hard_sync);
195 
196 
197 
198 /*
199  *	statistics
200  */
201 #ifdef CONFIG_JFS_STATISTICS
202 static struct lmStat {
203 	uint commit;		/* # of commit */
204 	uint pagedone;		/* # of page written */
205 	uint submitted;		/* # of pages submitted */
206 	uint full_page;		/* # of full pages submitted */
207 	uint partial_page;	/* # of partial pages submitted */
208 } lmStat;
209 #endif
210 
211 static void write_special_inodes(struct jfs_log *log,
212 				 int (*writer)(struct address_space *))
213 {
214 	struct jfs_sb_info *sbi;
215 
216 	list_for_each_entry(sbi, &log->sb_list, log_list) {
217 		writer(sbi->ipbmap->i_mapping);
218 		writer(sbi->ipimap->i_mapping);
219 		writer(sbi->direct_inode->i_mapping);
220 	}
221 }
222 
223 /*
224  * NAME:	lmLog()
225  *
226  * FUNCTION:	write a log record;
227  *
228  * PARAMETER:
229  *
230  * RETURN:	lsn - offset to the next log record to write (end-of-log);
231  *		-1  - error;
232  *
233  * note: todo: log error handler
234  */
235 int lmLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
236 	  struct tlock * tlck)
237 {
238 	int lsn;
239 	int diffp, difft;
240 	struct metapage *mp = NULL;
241 	unsigned long flags;
242 
243 	jfs_info("lmLog: log:0x%p tblk:0x%p, lrd:0x%p tlck:0x%p",
244 		 log, tblk, lrd, tlck);
245 
246 	LOG_LOCK(log);
247 
248 	/* log by (out-of-transaction) JFS ? */
249 	if (tblk == NULL)
250 		goto writeRecord;
251 
252 	/* log from page ? */
253 	if (tlck == NULL ||
254 	    tlck->type & tlckBTROOT || (mp = tlck->mp) == NULL)
255 		goto writeRecord;
256 
257 	/*
258 	 *	initialize/update page/transaction recovery lsn
259 	 */
260 	lsn = log->lsn;
261 
262 	LOGSYNC_LOCK(log, flags);
263 
264 	/*
265 	 * initialize page lsn if first log write of the page
266 	 */
267 	if (mp->lsn == 0) {
268 		mp->log = log;
269 		mp->lsn = lsn;
270 		log->count++;
271 
272 		/* insert page at tail of logsynclist */
273 		list_add_tail(&mp->synclist, &log->synclist);
274 	}
275 
276 	/*
277 	 *	initialize/update lsn of tblock of the page
278 	 *
279 	 * transaction inherits oldest lsn of pages associated
280 	 * with allocation/deallocation of resources (their
281 	 * log records are used to reconstruct allocation map
282 	 * at recovery time: inode for inode allocation map,
283 	 * B+-tree index of extent descriptors for block
284 	 * allocation map);
285 	 * allocation map pages inherit transaction lsn at
286 	 * commit time to allow forwarding log syncpt past log
287 	 * records associated with allocation/deallocation of
288 	 * resources only after persistent map of these map pages
289 	 * have been updated and propagated to home.
290 	 */
291 	/*
292 	 * initialize transaction lsn:
293 	 */
294 	if (tblk->lsn == 0) {
295 		/* inherit lsn of its first page logged */
296 		tblk->lsn = mp->lsn;
297 		log->count++;
298 
299 		/* insert tblock after the page on logsynclist */
300 		list_add(&tblk->synclist, &mp->synclist);
301 	}
302 	/*
303 	 * update transaction lsn:
304 	 */
305 	else {
306 		/* inherit oldest/smallest lsn of page */
307 		logdiff(diffp, mp->lsn, log);
308 		logdiff(difft, tblk->lsn, log);
309 		if (diffp < difft) {
310 			/* update tblock lsn with page lsn */
311 			tblk->lsn = mp->lsn;
312 
313 			/* move tblock after page on logsynclist */
314 			list_move(&tblk->synclist, &mp->synclist);
315 		}
316 	}
317 
318 	LOGSYNC_UNLOCK(log, flags);
319 
320 	/*
321 	 *	write the log record
322 	 */
323       writeRecord:
324 	lsn = lmWriteRecord(log, tblk, lrd, tlck);
325 
326 	/*
327 	 * forward log syncpt if log reached next syncpt trigger
328 	 */
329 	logdiff(diffp, lsn, log);
330 	if (diffp >= log->nextsync)
331 		lsn = lmLogSync(log, 0);
332 
333 	/* update end-of-log lsn */
334 	log->lsn = lsn;
335 
336 	LOG_UNLOCK(log);
337 
338 	/* return end-of-log address */
339 	return lsn;
340 }
341 
342 /*
343  * NAME:	lmWriteRecord()
344  *
345  * FUNCTION:	move the log record to current log page
346  *
347  * PARAMETER:	cd	- commit descriptor
348  *
349  * RETURN:	end-of-log address
350  *
351  * serialization: LOG_LOCK() held on entry/exit
352  */
353 static int
354 lmWriteRecord(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
355 	      struct tlock * tlck)
356 {
357 	int lsn = 0;		/* end-of-log address */
358 	struct lbuf *bp;	/* dst log page buffer */
359 	struct logpage *lp;	/* dst log page */
360 	caddr_t dst;		/* destination address in log page */
361 	int dstoffset;		/* end-of-log offset in log page */
362 	int freespace;		/* free space in log page */
363 	caddr_t p;		/* src meta-data page */
364 	caddr_t src;
365 	int srclen;
366 	int nbytes;		/* number of bytes to move */
367 	int i;
368 	int len;
369 	struct linelock *linelock;
370 	struct lv *lv;
371 	struct lvd *lvd;
372 	int l2linesize;
373 
374 	len = 0;
375 
376 	/* retrieve destination log page to write */
377 	bp = (struct lbuf *) log->bp;
378 	lp = (struct logpage *) bp->l_ldata;
379 	dstoffset = log->eor;
380 
381 	/* any log data to write ? */
382 	if (tlck == NULL)
383 		goto moveLrd;
384 
385 	/*
386 	 *	move log record data
387 	 */
388 	/* retrieve source meta-data page to log */
389 	if (tlck->flag & tlckPAGELOCK) {
390 		p = (caddr_t) (tlck->mp->data);
391 		linelock = (struct linelock *) & tlck->lock;
392 	}
393 	/* retrieve source in-memory inode to log */
394 	else if (tlck->flag & tlckINODELOCK) {
395 		if (tlck->type & tlckDTREE)
396 			p = (caddr_t) &JFS_IP(tlck->ip)->i_dtroot;
397 		else
398 			p = (caddr_t) &JFS_IP(tlck->ip)->i_xtroot;
399 		linelock = (struct linelock *) & tlck->lock;
400 	}
401 #ifdef	_JFS_WIP
402 	else if (tlck->flag & tlckINLINELOCK) {
403 
404 		inlinelock = (struct inlinelock *) & tlck;
405 		p = (caddr_t) & inlinelock->pxd;
406 		linelock = (struct linelock *) & tlck;
407 	}
408 #endif				/* _JFS_WIP */
409 	else {
410 		jfs_err("lmWriteRecord: UFO tlck:0x%p", tlck);
411 		return 0;	/* Probably should trap */
412 	}
413 	l2linesize = linelock->l2linesize;
414 
415       moveData:
416 	ASSERT(linelock->index <= linelock->maxcnt);
417 
418 	lv = linelock->lv;
419 	for (i = 0; i < linelock->index; i++, lv++) {
420 		if (lv->length == 0)
421 			continue;
422 
423 		/* is page full ? */
424 		if (dstoffset >= LOGPSIZE - LOGPTLRSIZE) {
425 			/* page become full: move on to next page */
426 			lmNextPage(log);
427 
428 			bp = log->bp;
429 			lp = (struct logpage *) bp->l_ldata;
430 			dstoffset = LOGPHDRSIZE;
431 		}
432 
433 		/*
434 		 * move log vector data
435 		 */
436 		src = (u8 *) p + (lv->offset << l2linesize);
437 		srclen = lv->length << l2linesize;
438 		len += srclen;
439 		while (srclen > 0) {
440 			freespace = (LOGPSIZE - LOGPTLRSIZE) - dstoffset;
441 			nbytes = min(freespace, srclen);
442 			dst = (caddr_t) lp + dstoffset;
443 			memcpy(dst, src, nbytes);
444 			dstoffset += nbytes;
445 
446 			/* is page not full ? */
447 			if (dstoffset < LOGPSIZE - LOGPTLRSIZE)
448 				break;
449 
450 			/* page become full: move on to next page */
451 			lmNextPage(log);
452 
453 			bp = (struct lbuf *) log->bp;
454 			lp = (struct logpage *) bp->l_ldata;
455 			dstoffset = LOGPHDRSIZE;
456 
457 			srclen -= nbytes;
458 			src += nbytes;
459 		}
460 
461 		/*
462 		 * move log vector descriptor
463 		 */
464 		len += 4;
465 		lvd = (struct lvd *) ((caddr_t) lp + dstoffset);
466 		lvd->offset = cpu_to_le16(lv->offset);
467 		lvd->length = cpu_to_le16(lv->length);
468 		dstoffset += 4;
469 		jfs_info("lmWriteRecord: lv offset:%d length:%d",
470 			 lv->offset, lv->length);
471 	}
472 
473 	if ((i = linelock->next)) {
474 		linelock = (struct linelock *) lid_to_tlock(i);
475 		goto moveData;
476 	}
477 
478 	/*
479 	 *	move log record descriptor
480 	 */
481       moveLrd:
482 	lrd->length = cpu_to_le16(len);
483 
484 	src = (caddr_t) lrd;
485 	srclen = LOGRDSIZE;
486 
487 	while (srclen > 0) {
488 		freespace = (LOGPSIZE - LOGPTLRSIZE) - dstoffset;
489 		nbytes = min(freespace, srclen);
490 		dst = (caddr_t) lp + dstoffset;
491 		memcpy(dst, src, nbytes);
492 
493 		dstoffset += nbytes;
494 		srclen -= nbytes;
495 
496 		/* are there more to move than freespace of page ? */
497 		if (srclen)
498 			goto pageFull;
499 
500 		/*
501 		 * end of log record descriptor
502 		 */
503 
504 		/* update last log record eor */
505 		log->eor = dstoffset;
506 		bp->l_eor = dstoffset;
507 		lsn = (log->page << L2LOGPSIZE) + dstoffset;
508 
509 		if (lrd->type & cpu_to_le16(LOG_COMMIT)) {
510 			tblk->clsn = lsn;
511 			jfs_info("wr: tclsn:0x%x, beor:0x%x", tblk->clsn,
512 				 bp->l_eor);
513 
514 			INCREMENT(lmStat.commit);	/* # of commit */
515 
516 			/*
517 			 * enqueue tblock for group commit:
518 			 *
519 			 * enqueue tblock of non-trivial/synchronous COMMIT
520 			 * at tail of group commit queue
521 			 * (trivial/asynchronous COMMITs are ignored by
522 			 * group commit.)
523 			 */
524 			LOGGC_LOCK(log);
525 
526 			/* init tblock gc state */
527 			tblk->flag = tblkGC_QUEUE;
528 			tblk->bp = log->bp;
529 			tblk->pn = log->page;
530 			tblk->eor = log->eor;
531 
532 			/* enqueue transaction to commit queue */
533 			list_add_tail(&tblk->cqueue, &log->cqueue);
534 
535 			LOGGC_UNLOCK(log);
536 		}
537 
538 		jfs_info("lmWriteRecord: lrd:0x%04x bp:0x%p pn:%d eor:0x%x",
539 			le16_to_cpu(lrd->type), log->bp, log->page, dstoffset);
540 
541 		/* page not full ? */
542 		if (dstoffset < LOGPSIZE - LOGPTLRSIZE)
543 			return lsn;
544 
545 	      pageFull:
546 		/* page become full: move on to next page */
547 		lmNextPage(log);
548 
549 		bp = (struct lbuf *) log->bp;
550 		lp = (struct logpage *) bp->l_ldata;
551 		dstoffset = LOGPHDRSIZE;
552 		src += nbytes;
553 	}
554 
555 	return lsn;
556 }
557 
558 
559 /*
560  * NAME:	lmNextPage()
561  *
562  * FUNCTION:	write current page and allocate next page.
563  *
564  * PARAMETER:	log
565  *
566  * RETURN:	0
567  *
568  * serialization: LOG_LOCK() held on entry/exit
569  */
570 static int lmNextPage(struct jfs_log * log)
571 {
572 	struct logpage *lp;
573 	int lspn;		/* log sequence page number */
574 	int pn;			/* current page number */
575 	struct lbuf *bp;
576 	struct lbuf *nextbp;
577 	struct tblock *tblk;
578 
579 	/* get current log page number and log sequence page number */
580 	pn = log->page;
581 	bp = log->bp;
582 	lp = (struct logpage *) bp->l_ldata;
583 	lspn = le32_to_cpu(lp->h.page);
584 
585 	LOGGC_LOCK(log);
586 
587 	/*
588 	 *	write or queue the full page at the tail of write queue
589 	 */
590 	/* get the tail tblk on commit queue */
591 	if (list_empty(&log->cqueue))
592 		tblk = NULL;
593 	else
594 		tblk = list_entry(log->cqueue.prev, struct tblock, cqueue);
595 
596 	/* every tblk who has COMMIT record on the current page,
597 	 * and has not been committed, must be on commit queue
598 	 * since tblk is queued at commit queueu at the time
599 	 * of writing its COMMIT record on the page before
600 	 * page becomes full (even though the tblk thread
601 	 * who wrote COMMIT record may have been suspended
602 	 * currently);
603 	 */
604 
605 	/* is page bound with outstanding tail tblk ? */
606 	if (tblk && tblk->pn == pn) {
607 		/* mark tblk for end-of-page */
608 		tblk->flag |= tblkGC_EOP;
609 
610 		if (log->cflag & logGC_PAGEOUT) {
611 			/* if page is not already on write queue,
612 			 * just enqueue (no lbmWRITE to prevent redrive)
613 			 * buffer to wqueue to ensure correct serial order
614 			 * of the pages since log pages will be added
615 			 * continuously
616 			 */
617 			if (bp->l_wqnext == NULL)
618 				lbmWrite(log, bp, 0, 0);
619 		} else {
620 			/*
621 			 * No current GC leader, initiate group commit
622 			 */
623 			log->cflag |= logGC_PAGEOUT;
624 			lmGCwrite(log, 0);
625 		}
626 	}
627 	/* page is not bound with outstanding tblk:
628 	 * init write or mark it to be redriven (lbmWRITE)
629 	 */
630 	else {
631 		/* finalize the page */
632 		bp->l_ceor = bp->l_eor;
633 		lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_ceor);
634 		lbmWrite(log, bp, lbmWRITE | lbmRELEASE | lbmFREE, 0);
635 	}
636 	LOGGC_UNLOCK(log);
637 
638 	/*
639 	 *	allocate/initialize next page
640 	 */
641 	/* if log wraps, the first data page of log is 2
642 	 * (0 never used, 1 is superblock).
643 	 */
644 	log->page = (pn == log->size - 1) ? 2 : pn + 1;
645 	log->eor = LOGPHDRSIZE;	/* ? valid page empty/full at logRedo() */
646 
647 	/* allocate/initialize next log page buffer */
648 	nextbp = lbmAllocate(log, log->page);
649 	nextbp->l_eor = log->eor;
650 	log->bp = nextbp;
651 
652 	/* initialize next log page */
653 	lp = (struct logpage *) nextbp->l_ldata;
654 	lp->h.page = lp->t.page = cpu_to_le32(lspn + 1);
655 	lp->h.eor = lp->t.eor = cpu_to_le16(LOGPHDRSIZE);
656 
657 	return 0;
658 }
659 
660 
661 /*
662  * NAME:	lmGroupCommit()
663  *
664  * FUNCTION:	group commit
665  *	initiate pageout of the pages with COMMIT in the order of
666  *	page number - redrive pageout of the page at the head of
667  *	pageout queue until full page has been written.
668  *
669  * RETURN:
670  *
671  * NOTE:
672  *	LOGGC_LOCK serializes log group commit queue, and
673  *	transaction blocks on the commit queue.
674  *	N.B. LOG_LOCK is NOT held during lmGroupCommit().
675  */
676 int lmGroupCommit(struct jfs_log * log, struct tblock * tblk)
677 {
678 	int rc = 0;
679 
680 	LOGGC_LOCK(log);
681 
682 	/* group committed already ? */
683 	if (tblk->flag & tblkGC_COMMITTED) {
684 		if (tblk->flag & tblkGC_ERROR)
685 			rc = -EIO;
686 
687 		LOGGC_UNLOCK(log);
688 		return rc;
689 	}
690 	jfs_info("lmGroup Commit: tblk = 0x%p, gcrtc = %d", tblk, log->gcrtc);
691 
692 	if (tblk->xflag & COMMIT_LAZY)
693 		tblk->flag |= tblkGC_LAZY;
694 
695 	if ((!(log->cflag & logGC_PAGEOUT)) && (!list_empty(&log->cqueue)) &&
696 	    (!(tblk->xflag & COMMIT_LAZY) || test_bit(log_FLUSH, &log->flag)
697 	     || jfs_tlocks_low)) {
698 		/*
699 		 * No pageout in progress
700 		 *
701 		 * start group commit as its group leader.
702 		 */
703 		log->cflag |= logGC_PAGEOUT;
704 
705 		lmGCwrite(log, 0);
706 	}
707 
708 	if (tblk->xflag & COMMIT_LAZY) {
709 		/*
710 		 * Lazy transactions can leave now
711 		 */
712 		LOGGC_UNLOCK(log);
713 		return 0;
714 	}
715 
716 	/* lmGCwrite gives up LOGGC_LOCK, check again */
717 
718 	if (tblk->flag & tblkGC_COMMITTED) {
719 		if (tblk->flag & tblkGC_ERROR)
720 			rc = -EIO;
721 
722 		LOGGC_UNLOCK(log);
723 		return rc;
724 	}
725 
726 	/* upcount transaction waiting for completion
727 	 */
728 	log->gcrtc++;
729 	tblk->flag |= tblkGC_READY;
730 
731 	__SLEEP_COND(tblk->gcwait, (tblk->flag & tblkGC_COMMITTED),
732 		     LOGGC_LOCK(log), LOGGC_UNLOCK(log));
733 
734 	/* removed from commit queue */
735 	if (tblk->flag & tblkGC_ERROR)
736 		rc = -EIO;
737 
738 	LOGGC_UNLOCK(log);
739 	return rc;
740 }
741 
742 /*
743  * NAME:	lmGCwrite()
744  *
745  * FUNCTION:	group commit write
746  *	initiate write of log page, building a group of all transactions
747  *	with commit records on that page.
748  *
749  * RETURN:	None
750  *
751  * NOTE:
752  *	LOGGC_LOCK must be held by caller.
753  *	N.B. LOG_LOCK is NOT held during lmGroupCommit().
754  */
755 static void lmGCwrite(struct jfs_log * log, int cant_write)
756 {
757 	struct lbuf *bp;
758 	struct logpage *lp;
759 	int gcpn;		/* group commit page number */
760 	struct tblock *tblk;
761 	struct tblock *xtblk = NULL;
762 
763 	/*
764 	 * build the commit group of a log page
765 	 *
766 	 * scan commit queue and make a commit group of all
767 	 * transactions with COMMIT records on the same log page.
768 	 */
769 	/* get the head tblk on the commit queue */
770 	gcpn = list_entry(log->cqueue.next, struct tblock, cqueue)->pn;
771 
772 	list_for_each_entry(tblk, &log->cqueue, cqueue) {
773 		if (tblk->pn != gcpn)
774 			break;
775 
776 		xtblk = tblk;
777 
778 		/* state transition: (QUEUE, READY) -> COMMIT */
779 		tblk->flag |= tblkGC_COMMIT;
780 	}
781 	tblk = xtblk;		/* last tblk of the page */
782 
783 	/*
784 	 * pageout to commit transactions on the log page.
785 	 */
786 	bp = (struct lbuf *) tblk->bp;
787 	lp = (struct logpage *) bp->l_ldata;
788 	/* is page already full ? */
789 	if (tblk->flag & tblkGC_EOP) {
790 		/* mark page to free at end of group commit of the page */
791 		tblk->flag &= ~tblkGC_EOP;
792 		tblk->flag |= tblkGC_FREE;
793 		bp->l_ceor = bp->l_eor;
794 		lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_ceor);
795 		lbmWrite(log, bp, lbmWRITE | lbmRELEASE | lbmGC,
796 			 cant_write);
797 		INCREMENT(lmStat.full_page);
798 	}
799 	/* page is not yet full */
800 	else {
801 		bp->l_ceor = tblk->eor;	/* ? bp->l_ceor = bp->l_eor; */
802 		lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_ceor);
803 		lbmWrite(log, bp, lbmWRITE | lbmGC, cant_write);
804 		INCREMENT(lmStat.partial_page);
805 	}
806 }
807 
808 /*
809  * NAME:	lmPostGC()
810  *
811  * FUNCTION:	group commit post-processing
812  *	Processes transactions after their commit records have been written
813  *	to disk, redriving log I/O if necessary.
814  *
815  * RETURN:	None
816  *
817  * NOTE:
818  *	This routine is called a interrupt time by lbmIODone
819  */
820 static void lmPostGC(struct lbuf * bp)
821 {
822 	unsigned long flags;
823 	struct jfs_log *log = bp->l_log;
824 	struct logpage *lp;
825 	struct tblock *tblk, *temp;
826 
827 	//LOGGC_LOCK(log);
828 	spin_lock_irqsave(&log->gclock, flags);
829 	/*
830 	 * current pageout of group commit completed.
831 	 *
832 	 * remove/wakeup transactions from commit queue who were
833 	 * group committed with the current log page
834 	 */
835 	list_for_each_entry_safe(tblk, temp, &log->cqueue, cqueue) {
836 		if (!(tblk->flag & tblkGC_COMMIT))
837 			break;
838 		/* if transaction was marked GC_COMMIT then
839 		 * it has been shipped in the current pageout
840 		 * and made it to disk - it is committed.
841 		 */
842 
843 		if (bp->l_flag & lbmERROR)
844 			tblk->flag |= tblkGC_ERROR;
845 
846 		/* remove it from the commit queue */
847 		list_del(&tblk->cqueue);
848 		tblk->flag &= ~tblkGC_QUEUE;
849 
850 		if (tblk == log->flush_tblk) {
851 			/* we can stop flushing the log now */
852 			clear_bit(log_FLUSH, &log->flag);
853 			log->flush_tblk = NULL;
854 		}
855 
856 		jfs_info("lmPostGC: tblk = 0x%p, flag = 0x%x", tblk,
857 			 tblk->flag);
858 
859 		if (!(tblk->xflag & COMMIT_FORCE))
860 			/*
861 			 * Hand tblk over to lazy commit thread
862 			 */
863 			txLazyUnlock(tblk);
864 		else {
865 			/* state transition: COMMIT -> COMMITTED */
866 			tblk->flag |= tblkGC_COMMITTED;
867 
868 			if (tblk->flag & tblkGC_READY)
869 				log->gcrtc--;
870 
871 			LOGGC_WAKEUP(tblk);
872 		}
873 
874 		/* was page full before pageout ?
875 		 * (and this is the last tblk bound with the page)
876 		 */
877 		if (tblk->flag & tblkGC_FREE)
878 			lbmFree(bp);
879 		/* did page become full after pageout ?
880 		 * (and this is the last tblk bound with the page)
881 		 */
882 		else if (tblk->flag & tblkGC_EOP) {
883 			/* finalize the page */
884 			lp = (struct logpage *) bp->l_ldata;
885 			bp->l_ceor = bp->l_eor;
886 			lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_eor);
887 			jfs_info("lmPostGC: calling lbmWrite");
888 			lbmWrite(log, bp, lbmWRITE | lbmRELEASE | lbmFREE,
889 				 1);
890 		}
891 
892 	}
893 
894 	/* are there any transactions who have entered lnGroupCommit()
895 	 * (whose COMMITs are after that of the last log page written.
896 	 * They are waiting for new group commit (above at (SLEEP 1))
897 	 * or lazy transactions are on a full (queued) log page,
898 	 * select the latest ready transaction as new group leader and
899 	 * wake her up to lead her group.
900 	 */
901 	if ((!list_empty(&log->cqueue)) &&
902 	    ((log->gcrtc > 0) || (tblk->bp->l_wqnext != NULL) ||
903 	     test_bit(log_FLUSH, &log->flag) || jfs_tlocks_low))
904 		/*
905 		 * Call lmGCwrite with new group leader
906 		 */
907 		lmGCwrite(log, 1);
908 
909 	/* no transaction are ready yet (transactions are only just
910 	 * queued (GC_QUEUE) and not entered for group commit yet).
911 	 * the first transaction entering group commit
912 	 * will elect herself as new group leader.
913 	 */
914 	else
915 		log->cflag &= ~logGC_PAGEOUT;
916 
917 	//LOGGC_UNLOCK(log);
918 	spin_unlock_irqrestore(&log->gclock, flags);
919 	return;
920 }
921 
922 /*
923  * NAME:	lmLogSync()
924  *
925  * FUNCTION:	write log SYNCPT record for specified log
926  *	if new sync address is available
927  *	(normally the case if sync() is executed by back-ground
928  *	process).
929  *	calculate new value of i_nextsync which determines when
930  *	this code is called again.
931  *
932  * PARAMETERS:	log	- log structure
933  *		hard_sync - 1 to force all metadata to be written
934  *
935  * RETURN:	0
936  *
937  * serialization: LOG_LOCK() held on entry/exit
938  */
939 static int lmLogSync(struct jfs_log * log, int hard_sync)
940 {
941 	int logsize;
942 	int written;		/* written since last syncpt */
943 	int free;		/* free space left available */
944 	int delta;		/* additional delta to write normally */
945 	int more;		/* additional write granted */
946 	struct lrd lrd;
947 	int lsn;
948 	struct logsyncblk *lp;
949 	unsigned long flags;
950 
951 	/* push dirty metapages out to disk */
952 	if (hard_sync)
953 		write_special_inodes(log, filemap_fdatawrite);
954 	else
955 		write_special_inodes(log, filemap_flush);
956 
957 	/*
958 	 *	forward syncpt
959 	 */
960 	/* if last sync is same as last syncpt,
961 	 * invoke sync point forward processing to update sync.
962 	 */
963 
964 	if (log->sync == log->syncpt) {
965 		LOGSYNC_LOCK(log, flags);
966 		if (list_empty(&log->synclist))
967 			log->sync = log->lsn;
968 		else {
969 			lp = list_entry(log->synclist.next,
970 					struct logsyncblk, synclist);
971 			log->sync = lp->lsn;
972 		}
973 		LOGSYNC_UNLOCK(log, flags);
974 
975 	}
976 
977 	/* if sync is different from last syncpt,
978 	 * write a SYNCPT record with syncpt = sync.
979 	 * reset syncpt = sync
980 	 */
981 	if (log->sync != log->syncpt) {
982 		lrd.logtid = 0;
983 		lrd.backchain = 0;
984 		lrd.type = cpu_to_le16(LOG_SYNCPT);
985 		lrd.length = 0;
986 		lrd.log.syncpt.sync = cpu_to_le32(log->sync);
987 		lsn = lmWriteRecord(log, NULL, &lrd, NULL);
988 
989 		log->syncpt = log->sync;
990 	} else
991 		lsn = log->lsn;
992 
993 	/*
994 	 *	setup next syncpt trigger (SWAG)
995 	 */
996 	logsize = log->logsize;
997 
998 	logdiff(written, lsn, log);
999 	free = logsize - written;
1000 	delta = LOGSYNC_DELTA(logsize);
1001 	more = min(free / 2, delta);
1002 	if (more < 2 * LOGPSIZE) {
1003 		jfs_warn("\n ... Log Wrap ... Log Wrap ... Log Wrap ...\n");
1004 		/*
1005 		 *	log wrapping
1006 		 *
1007 		 * option 1 - panic ? No.!
1008 		 * option 2 - shutdown file systems
1009 		 *	      associated with log ?
1010 		 * option 3 - extend log ?
1011 		 */
1012 		/*
1013 		 * option 4 - second chance
1014 		 *
1015 		 * mark log wrapped, and continue.
1016 		 * when all active transactions are completed,
1017 		 * mark log vaild for recovery.
1018 		 * if crashed during invalid state, log state
1019 		 * implies invald log, forcing fsck().
1020 		 */
1021 		/* mark log state log wrap in log superblock */
1022 		/* log->state = LOGWRAP; */
1023 
1024 		/* reset sync point computation */
1025 		log->syncpt = log->sync = lsn;
1026 		log->nextsync = delta;
1027 	} else
1028 		/* next syncpt trigger = written + more */
1029 		log->nextsync = written + more;
1030 
1031 	/* if number of bytes written from last sync point is more
1032 	 * than 1/4 of the log size, stop new transactions from
1033 	 * starting until all current transactions are completed
1034 	 * by setting syncbarrier flag.
1035 	 */
1036 	if (!test_bit(log_SYNCBARRIER, &log->flag) &&
1037 	    (written > LOGSYNC_BARRIER(logsize)) && log->active) {
1038 		set_bit(log_SYNCBARRIER, &log->flag);
1039 		jfs_info("log barrier on: lsn=0x%x syncpt=0x%x", lsn,
1040 			 log->syncpt);
1041 		/*
1042 		 * We may have to initiate group commit
1043 		 */
1044 		jfs_flush_journal(log, 0);
1045 	}
1046 
1047 	return lsn;
1048 }
1049 
1050 /*
1051  * NAME:	jfs_syncpt
1052  *
1053  * FUNCTION:	write log SYNCPT record for specified log
1054  *
1055  * PARAMETERS:	log	  - log structure
1056  *		hard_sync - set to 1 to force metadata to be written
1057  */
1058 void jfs_syncpt(struct jfs_log *log, int hard_sync)
1059 {	LOG_LOCK(log);
1060 	lmLogSync(log, hard_sync);
1061 	LOG_UNLOCK(log);
1062 }
1063 
1064 /*
1065  * NAME:	lmLogOpen()
1066  *
1067  * FUNCTION:	open the log on first open;
1068  *	insert filesystem in the active list of the log.
1069  *
1070  * PARAMETER:	ipmnt	- file system mount inode
1071  *		iplog	- log inode (out)
1072  *
1073  * RETURN:
1074  *
1075  * serialization:
1076  */
1077 int lmLogOpen(struct super_block *sb)
1078 {
1079 	int rc;
1080 	struct block_device *bdev;
1081 	struct jfs_log *log;
1082 	struct jfs_sb_info *sbi = JFS_SBI(sb);
1083 
1084 	if (sbi->flag & JFS_NOINTEGRITY)
1085 		return open_dummy_log(sb);
1086 
1087 	if (sbi->mntflag & JFS_INLINELOG)
1088 		return open_inline_log(sb);
1089 
1090 	mutex_lock(&jfs_log_mutex);
1091 	list_for_each_entry(log, &jfs_external_logs, journal_list) {
1092 		if (log->bdev->bd_dev == sbi->logdev) {
1093 			if (memcmp(log->uuid, sbi->loguuid,
1094 				   sizeof(log->uuid))) {
1095 				jfs_warn("wrong uuid on JFS journal\n");
1096 				mutex_unlock(&jfs_log_mutex);
1097 				return -EINVAL;
1098 			}
1099 			/*
1100 			 * add file system to log active file system list
1101 			 */
1102 			if ((rc = lmLogFileSystem(log, sbi, 1))) {
1103 				mutex_unlock(&jfs_log_mutex);
1104 				return rc;
1105 			}
1106 			goto journal_found;
1107 		}
1108 	}
1109 
1110 	if (!(log = kzalloc(sizeof(struct jfs_log), GFP_KERNEL))) {
1111 		mutex_unlock(&jfs_log_mutex);
1112 		return -ENOMEM;
1113 	}
1114 	INIT_LIST_HEAD(&log->sb_list);
1115 	init_waitqueue_head(&log->syncwait);
1116 
1117 	/*
1118 	 *	external log as separate logical volume
1119 	 *
1120 	 * file systems to log may have n-to-1 relationship;
1121 	 */
1122 
1123 	bdev = open_by_devnum(sbi->logdev, FMODE_READ|FMODE_WRITE);
1124 	if (IS_ERR(bdev)) {
1125 		rc = -PTR_ERR(bdev);
1126 		goto free;
1127 	}
1128 
1129 	if ((rc = bd_claim(bdev, log))) {
1130 		goto close;
1131 	}
1132 
1133 	log->bdev = bdev;
1134 	memcpy(log->uuid, sbi->loguuid, sizeof(log->uuid));
1135 
1136 	/*
1137 	 * initialize log:
1138 	 */
1139 	if ((rc = lmLogInit(log)))
1140 		goto unclaim;
1141 
1142 	list_add(&log->journal_list, &jfs_external_logs);
1143 
1144 	/*
1145 	 * add file system to log active file system list
1146 	 */
1147 	if ((rc = lmLogFileSystem(log, sbi, 1)))
1148 		goto shutdown;
1149 
1150 journal_found:
1151 	LOG_LOCK(log);
1152 	list_add(&sbi->log_list, &log->sb_list);
1153 	sbi->log = log;
1154 	LOG_UNLOCK(log);
1155 
1156 	mutex_unlock(&jfs_log_mutex);
1157 	return 0;
1158 
1159 	/*
1160 	 *	unwind on error
1161 	 */
1162       shutdown:		/* unwind lbmLogInit() */
1163 	list_del(&log->journal_list);
1164 	lbmLogShutdown(log);
1165 
1166       unclaim:
1167 	bd_release(bdev);
1168 
1169       close:		/* close external log device */
1170 	blkdev_put(bdev);
1171 
1172       free:		/* free log descriptor */
1173 	mutex_unlock(&jfs_log_mutex);
1174 	kfree(log);
1175 
1176 	jfs_warn("lmLogOpen: exit(%d)", rc);
1177 	return rc;
1178 }
1179 
1180 static int open_inline_log(struct super_block *sb)
1181 {
1182 	struct jfs_log *log;
1183 	int rc;
1184 
1185 	if (!(log = kzalloc(sizeof(struct jfs_log), GFP_KERNEL)))
1186 		return -ENOMEM;
1187 	INIT_LIST_HEAD(&log->sb_list);
1188 	init_waitqueue_head(&log->syncwait);
1189 
1190 	set_bit(log_INLINELOG, &log->flag);
1191 	log->bdev = sb->s_bdev;
1192 	log->base = addressPXD(&JFS_SBI(sb)->logpxd);
1193 	log->size = lengthPXD(&JFS_SBI(sb)->logpxd) >>
1194 	    (L2LOGPSIZE - sb->s_blocksize_bits);
1195 	log->l2bsize = sb->s_blocksize_bits;
1196 	ASSERT(L2LOGPSIZE >= sb->s_blocksize_bits);
1197 
1198 	/*
1199 	 * initialize log.
1200 	 */
1201 	if ((rc = lmLogInit(log))) {
1202 		kfree(log);
1203 		jfs_warn("lmLogOpen: exit(%d)", rc);
1204 		return rc;
1205 	}
1206 
1207 	list_add(&JFS_SBI(sb)->log_list, &log->sb_list);
1208 	JFS_SBI(sb)->log = log;
1209 
1210 	return rc;
1211 }
1212 
1213 static int open_dummy_log(struct super_block *sb)
1214 {
1215 	int rc;
1216 
1217 	mutex_lock(&jfs_log_mutex);
1218 	if (!dummy_log) {
1219 		dummy_log = kzalloc(sizeof(struct jfs_log), GFP_KERNEL);
1220 		if (!dummy_log) {
1221 			mutex_unlock(&jfs_log_mutex);
1222 			return -ENOMEM;
1223 		}
1224 		INIT_LIST_HEAD(&dummy_log->sb_list);
1225 		init_waitqueue_head(&dummy_log->syncwait);
1226 		dummy_log->no_integrity = 1;
1227 		/* Make up some stuff */
1228 		dummy_log->base = 0;
1229 		dummy_log->size = 1024;
1230 		rc = lmLogInit(dummy_log);
1231 		if (rc) {
1232 			kfree(dummy_log);
1233 			dummy_log = NULL;
1234 			mutex_unlock(&jfs_log_mutex);
1235 			return rc;
1236 		}
1237 	}
1238 
1239 	LOG_LOCK(dummy_log);
1240 	list_add(&JFS_SBI(sb)->log_list, &dummy_log->sb_list);
1241 	JFS_SBI(sb)->log = dummy_log;
1242 	LOG_UNLOCK(dummy_log);
1243 	mutex_unlock(&jfs_log_mutex);
1244 
1245 	return 0;
1246 }
1247 
1248 /*
1249  * NAME:	lmLogInit()
1250  *
1251  * FUNCTION:	log initialization at first log open.
1252  *
1253  *	logredo() (or logformat()) should have been run previously.
1254  *	initialize the log from log superblock.
1255  *	set the log state in the superblock to LOGMOUNT and
1256  *	write SYNCPT log record.
1257  *
1258  * PARAMETER:	log	- log structure
1259  *
1260  * RETURN:	0	- if ok
1261  *		-EINVAL	- bad log magic number or superblock dirty
1262  *		error returned from logwait()
1263  *
1264  * serialization: single first open thread
1265  */
1266 int lmLogInit(struct jfs_log * log)
1267 {
1268 	int rc = 0;
1269 	struct lrd lrd;
1270 	struct logsuper *logsuper;
1271 	struct lbuf *bpsuper;
1272 	struct lbuf *bp;
1273 	struct logpage *lp;
1274 	int lsn = 0;
1275 
1276 	jfs_info("lmLogInit: log:0x%p", log);
1277 
1278 	/* initialize the group commit serialization lock */
1279 	LOGGC_LOCK_INIT(log);
1280 
1281 	/* allocate/initialize the log write serialization lock */
1282 	LOG_LOCK_INIT(log);
1283 
1284 	LOGSYNC_LOCK_INIT(log);
1285 
1286 	INIT_LIST_HEAD(&log->synclist);
1287 
1288 	INIT_LIST_HEAD(&log->cqueue);
1289 	log->flush_tblk = NULL;
1290 
1291 	log->count = 0;
1292 
1293 	/*
1294 	 * initialize log i/o
1295 	 */
1296 	if ((rc = lbmLogInit(log)))
1297 		return rc;
1298 
1299 	if (!test_bit(log_INLINELOG, &log->flag))
1300 		log->l2bsize = L2LOGPSIZE;
1301 
1302 	/* check for disabled journaling to disk */
1303 	if (log->no_integrity) {
1304 		/*
1305 		 * Journal pages will still be filled.  When the time comes
1306 		 * to actually do the I/O, the write is not done, and the
1307 		 * endio routine is called directly.
1308 		 */
1309 		bp = lbmAllocate(log , 0);
1310 		log->bp = bp;
1311 		bp->l_pn = bp->l_eor = 0;
1312 	} else {
1313 		/*
1314 		 * validate log superblock
1315 		 */
1316 		if ((rc = lbmRead(log, 1, &bpsuper)))
1317 			goto errout10;
1318 
1319 		logsuper = (struct logsuper *) bpsuper->l_ldata;
1320 
1321 		if (logsuper->magic != cpu_to_le32(LOGMAGIC)) {
1322 			jfs_warn("*** Log Format Error ! ***");
1323 			rc = -EINVAL;
1324 			goto errout20;
1325 		}
1326 
1327 		/* logredo() should have been run successfully. */
1328 		if (logsuper->state != cpu_to_le32(LOGREDONE)) {
1329 			jfs_warn("*** Log Is Dirty ! ***");
1330 			rc = -EINVAL;
1331 			goto errout20;
1332 		}
1333 
1334 		/* initialize log from log superblock */
1335 		if (test_bit(log_INLINELOG,&log->flag)) {
1336 			if (log->size != le32_to_cpu(logsuper->size)) {
1337 				rc = -EINVAL;
1338 				goto errout20;
1339 			}
1340 			jfs_info("lmLogInit: inline log:0x%p base:0x%Lx "
1341 				 "size:0x%x", log,
1342 				 (unsigned long long) log->base, log->size);
1343 		} else {
1344 			if (memcmp(logsuper->uuid, log->uuid, 16)) {
1345 				jfs_warn("wrong uuid on JFS log device");
1346 				goto errout20;
1347 			}
1348 			log->size = le32_to_cpu(logsuper->size);
1349 			log->l2bsize = le32_to_cpu(logsuper->l2bsize);
1350 			jfs_info("lmLogInit: external log:0x%p base:0x%Lx "
1351 				 "size:0x%x", log,
1352 				 (unsigned long long) log->base, log->size);
1353 		}
1354 
1355 		log->page = le32_to_cpu(logsuper->end) / LOGPSIZE;
1356 		log->eor = le32_to_cpu(logsuper->end) - (LOGPSIZE * log->page);
1357 
1358 		/*
1359 		 * initialize for log append write mode
1360 		 */
1361 		/* establish current/end-of-log page/buffer */
1362 		if ((rc = lbmRead(log, log->page, &bp)))
1363 			goto errout20;
1364 
1365 		lp = (struct logpage *) bp->l_ldata;
1366 
1367 		jfs_info("lmLogInit: lsn:0x%x page:%d eor:%d:%d",
1368 			 le32_to_cpu(logsuper->end), log->page, log->eor,
1369 			 le16_to_cpu(lp->h.eor));
1370 
1371 		log->bp = bp;
1372 		bp->l_pn = log->page;
1373 		bp->l_eor = log->eor;
1374 
1375 		/* if current page is full, move on to next page */
1376 		if (log->eor >= LOGPSIZE - LOGPTLRSIZE)
1377 			lmNextPage(log);
1378 
1379 		/*
1380 		 * initialize log syncpoint
1381 		 */
1382 		/*
1383 		 * write the first SYNCPT record with syncpoint = 0
1384 		 * (i.e., log redo up to HERE !);
1385 		 * remove current page from lbm write queue at end of pageout
1386 		 * (to write log superblock update), but do not release to
1387 		 * freelist;
1388 		 */
1389 		lrd.logtid = 0;
1390 		lrd.backchain = 0;
1391 		lrd.type = cpu_to_le16(LOG_SYNCPT);
1392 		lrd.length = 0;
1393 		lrd.log.syncpt.sync = 0;
1394 		lsn = lmWriteRecord(log, NULL, &lrd, NULL);
1395 		bp = log->bp;
1396 		bp->l_ceor = bp->l_eor;
1397 		lp = (struct logpage *) bp->l_ldata;
1398 		lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_eor);
1399 		lbmWrite(log, bp, lbmWRITE | lbmSYNC, 0);
1400 		if ((rc = lbmIOWait(bp, 0)))
1401 			goto errout30;
1402 
1403 		/*
1404 		 * update/write superblock
1405 		 */
1406 		logsuper->state = cpu_to_le32(LOGMOUNT);
1407 		log->serial = le32_to_cpu(logsuper->serial) + 1;
1408 		logsuper->serial = cpu_to_le32(log->serial);
1409 		lbmDirectWrite(log, bpsuper, lbmWRITE | lbmRELEASE | lbmSYNC);
1410 		if ((rc = lbmIOWait(bpsuper, lbmFREE)))
1411 			goto errout30;
1412 	}
1413 
1414 	/* initialize logsync parameters */
1415 	log->logsize = (log->size - 2) << L2LOGPSIZE;
1416 	log->lsn = lsn;
1417 	log->syncpt = lsn;
1418 	log->sync = log->syncpt;
1419 	log->nextsync = LOGSYNC_DELTA(log->logsize);
1420 
1421 	jfs_info("lmLogInit: lsn:0x%x syncpt:0x%x sync:0x%x",
1422 		 log->lsn, log->syncpt, log->sync);
1423 
1424 	/*
1425 	 * initialize for lazy/group commit
1426 	 */
1427 	log->clsn = lsn;
1428 
1429 	return 0;
1430 
1431 	/*
1432 	 *	unwind on error
1433 	 */
1434       errout30:		/* release log page */
1435 	log->wqueue = NULL;
1436 	bp->l_wqnext = NULL;
1437 	lbmFree(bp);
1438 
1439       errout20:		/* release log superblock */
1440 	lbmFree(bpsuper);
1441 
1442       errout10:		/* unwind lbmLogInit() */
1443 	lbmLogShutdown(log);
1444 
1445 	jfs_warn("lmLogInit: exit(%d)", rc);
1446 	return rc;
1447 }
1448 
1449 
1450 /*
1451  * NAME:	lmLogClose()
1452  *
1453  * FUNCTION:	remove file system <ipmnt> from active list of log <iplog>
1454  *		and close it on last close.
1455  *
1456  * PARAMETER:	sb	- superblock
1457  *
1458  * RETURN:	errors from subroutines
1459  *
1460  * serialization:
1461  */
1462 int lmLogClose(struct super_block *sb)
1463 {
1464 	struct jfs_sb_info *sbi = JFS_SBI(sb);
1465 	struct jfs_log *log = sbi->log;
1466 	struct block_device *bdev;
1467 	int rc = 0;
1468 
1469 	jfs_info("lmLogClose: log:0x%p", log);
1470 
1471 	mutex_lock(&jfs_log_mutex);
1472 	LOG_LOCK(log);
1473 	list_del(&sbi->log_list);
1474 	LOG_UNLOCK(log);
1475 	sbi->log = NULL;
1476 
1477 	/*
1478 	 * We need to make sure all of the "written" metapages
1479 	 * actually make it to disk
1480 	 */
1481 	sync_blockdev(sb->s_bdev);
1482 
1483 	if (test_bit(log_INLINELOG, &log->flag)) {
1484 		/*
1485 		 *	in-line log in host file system
1486 		 */
1487 		rc = lmLogShutdown(log);
1488 		kfree(log);
1489 		goto out;
1490 	}
1491 
1492 	if (!log->no_integrity)
1493 		lmLogFileSystem(log, sbi, 0);
1494 
1495 	if (!list_empty(&log->sb_list))
1496 		goto out;
1497 
1498 	/*
1499 	 * TODO: ensure that the dummy_log is in a state to allow
1500 	 * lbmLogShutdown to deallocate all the buffers and call
1501 	 * kfree against dummy_log.  For now, leave dummy_log & its
1502 	 * buffers in memory, and resuse if another no-integrity mount
1503 	 * is requested.
1504 	 */
1505 	if (log->no_integrity)
1506 		goto out;
1507 
1508 	/*
1509 	 *	external log as separate logical volume
1510 	 */
1511 	list_del(&log->journal_list);
1512 	bdev = log->bdev;
1513 	rc = lmLogShutdown(log);
1514 
1515 	bd_release(bdev);
1516 	blkdev_put(bdev);
1517 
1518 	kfree(log);
1519 
1520       out:
1521 	mutex_unlock(&jfs_log_mutex);
1522 	jfs_info("lmLogClose: exit(%d)", rc);
1523 	return rc;
1524 }
1525 
1526 
1527 /*
1528  * NAME:	jfs_flush_journal()
1529  *
1530  * FUNCTION:	initiate write of any outstanding transactions to the journal
1531  *		and optionally wait until they are all written to disk
1532  *
1533  *		wait == 0  flush until latest txn is committed, don't wait
1534  *		wait == 1  flush until latest txn is committed, wait
1535  *		wait > 1   flush until all txn's are complete, wait
1536  */
1537 void jfs_flush_journal(struct jfs_log *log, int wait)
1538 {
1539 	int i;
1540 	struct tblock *target = NULL;
1541 
1542 	/* jfs_write_inode may call us during read-only mount */
1543 	if (!log)
1544 		return;
1545 
1546 	jfs_info("jfs_flush_journal: log:0x%p wait=%d", log, wait);
1547 
1548 	LOGGC_LOCK(log);
1549 
1550 	if (!list_empty(&log->cqueue)) {
1551 		/*
1552 		 * This ensures that we will keep writing to the journal as long
1553 		 * as there are unwritten commit records
1554 		 */
1555 		target = list_entry(log->cqueue.prev, struct tblock, cqueue);
1556 
1557 		if (test_bit(log_FLUSH, &log->flag)) {
1558 			/*
1559 			 * We're already flushing.
1560 			 * if flush_tblk is NULL, we are flushing everything,
1561 			 * so leave it that way.  Otherwise, update it to the
1562 			 * latest transaction
1563 			 */
1564 			if (log->flush_tblk)
1565 				log->flush_tblk = target;
1566 		} else {
1567 			/* Only flush until latest transaction is committed */
1568 			log->flush_tblk = target;
1569 			set_bit(log_FLUSH, &log->flag);
1570 
1571 			/*
1572 			 * Initiate I/O on outstanding transactions
1573 			 */
1574 			if (!(log->cflag & logGC_PAGEOUT)) {
1575 				log->cflag |= logGC_PAGEOUT;
1576 				lmGCwrite(log, 0);
1577 			}
1578 		}
1579 	}
1580 	if ((wait > 1) || test_bit(log_SYNCBARRIER, &log->flag)) {
1581 		/* Flush until all activity complete */
1582 		set_bit(log_FLUSH, &log->flag);
1583 		log->flush_tblk = NULL;
1584 	}
1585 
1586 	if (wait && target && !(target->flag & tblkGC_COMMITTED)) {
1587 		DECLARE_WAITQUEUE(__wait, current);
1588 
1589 		add_wait_queue(&target->gcwait, &__wait);
1590 		set_current_state(TASK_UNINTERRUPTIBLE);
1591 		LOGGC_UNLOCK(log);
1592 		schedule();
1593 		__set_current_state(TASK_RUNNING);
1594 		LOGGC_LOCK(log);
1595 		remove_wait_queue(&target->gcwait, &__wait);
1596 	}
1597 	LOGGC_UNLOCK(log);
1598 
1599 	if (wait < 2)
1600 		return;
1601 
1602 	write_special_inodes(log, filemap_fdatawrite);
1603 
1604 	/*
1605 	 * If there was recent activity, we may need to wait
1606 	 * for the lazycommit thread to catch up
1607 	 */
1608 	if ((!list_empty(&log->cqueue)) || !list_empty(&log->synclist)) {
1609 		for (i = 0; i < 200; i++) {	/* Too much? */
1610 			msleep(250);
1611 			write_special_inodes(log, filemap_fdatawrite);
1612 			if (list_empty(&log->cqueue) &&
1613 			    list_empty(&log->synclist))
1614 				break;
1615 		}
1616 	}
1617 	assert(list_empty(&log->cqueue));
1618 
1619 #ifdef CONFIG_JFS_DEBUG
1620 	if (!list_empty(&log->synclist)) {
1621 		struct logsyncblk *lp;
1622 
1623 		printk(KERN_ERR "jfs_flush_journal: synclist not empty\n");
1624 		list_for_each_entry(lp, &log->synclist, synclist) {
1625 			if (lp->xflag & COMMIT_PAGE) {
1626 				struct metapage *mp = (struct metapage *)lp;
1627 				print_hex_dump(KERN_ERR, "metapage: ",
1628 					       DUMP_PREFIX_ADDRESS, 16, 4,
1629 					       mp, sizeof(struct metapage), 0);
1630 				print_hex_dump(KERN_ERR, "page: ",
1631 					       DUMP_PREFIX_ADDRESS, 16,
1632 					       sizeof(long), mp->page,
1633 					       sizeof(struct page), 0);
1634 			} else
1635 				print_hex_dump(KERN_ERR, "tblock:",
1636 					       DUMP_PREFIX_ADDRESS, 16, 4,
1637 					       lp, sizeof(struct tblock), 0);
1638 		}
1639 	}
1640 #else
1641 	WARN_ON(!list_empty(&log->synclist));
1642 #endif
1643 	clear_bit(log_FLUSH, &log->flag);
1644 }
1645 
1646 /*
1647  * NAME:	lmLogShutdown()
1648  *
1649  * FUNCTION:	log shutdown at last LogClose().
1650  *
1651  *		write log syncpt record.
1652  *		update super block to set redone flag to 0.
1653  *
1654  * PARAMETER:	log	- log inode
1655  *
1656  * RETURN:	0	- success
1657  *
1658  * serialization: single last close thread
1659  */
1660 int lmLogShutdown(struct jfs_log * log)
1661 {
1662 	int rc;
1663 	struct lrd lrd;
1664 	int lsn;
1665 	struct logsuper *logsuper;
1666 	struct lbuf *bpsuper;
1667 	struct lbuf *bp;
1668 	struct logpage *lp;
1669 
1670 	jfs_info("lmLogShutdown: log:0x%p", log);
1671 
1672 	jfs_flush_journal(log, 2);
1673 
1674 	/*
1675 	 * write the last SYNCPT record with syncpoint = 0
1676 	 * (i.e., log redo up to HERE !)
1677 	 */
1678 	lrd.logtid = 0;
1679 	lrd.backchain = 0;
1680 	lrd.type = cpu_to_le16(LOG_SYNCPT);
1681 	lrd.length = 0;
1682 	lrd.log.syncpt.sync = 0;
1683 
1684 	lsn = lmWriteRecord(log, NULL, &lrd, NULL);
1685 	bp = log->bp;
1686 	lp = (struct logpage *) bp->l_ldata;
1687 	lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_eor);
1688 	lbmWrite(log, log->bp, lbmWRITE | lbmRELEASE | lbmSYNC, 0);
1689 	lbmIOWait(log->bp, lbmFREE);
1690 	log->bp = NULL;
1691 
1692 	/*
1693 	 * synchronous update log superblock
1694 	 * mark log state as shutdown cleanly
1695 	 * (i.e., Log does not need to be replayed).
1696 	 */
1697 	if ((rc = lbmRead(log, 1, &bpsuper)))
1698 		goto out;
1699 
1700 	logsuper = (struct logsuper *) bpsuper->l_ldata;
1701 	logsuper->state = cpu_to_le32(LOGREDONE);
1702 	logsuper->end = cpu_to_le32(lsn);
1703 	lbmDirectWrite(log, bpsuper, lbmWRITE | lbmRELEASE | lbmSYNC);
1704 	rc = lbmIOWait(bpsuper, lbmFREE);
1705 
1706 	jfs_info("lmLogShutdown: lsn:0x%x page:%d eor:%d",
1707 		 lsn, log->page, log->eor);
1708 
1709       out:
1710 	/*
1711 	 * shutdown per log i/o
1712 	 */
1713 	lbmLogShutdown(log);
1714 
1715 	if (rc) {
1716 		jfs_warn("lmLogShutdown: exit(%d)", rc);
1717 	}
1718 	return rc;
1719 }
1720 
1721 
1722 /*
1723  * NAME:	lmLogFileSystem()
1724  *
1725  * FUNCTION:	insert (<activate> = true)/remove (<activate> = false)
1726  *	file system into/from log active file system list.
1727  *
1728  * PARAMETE:	log	- pointer to logs inode.
1729  *		fsdev	- kdev_t of filesystem.
1730  *		serial	- pointer to returned log serial number
1731  *		activate - insert/remove device from active list.
1732  *
1733  * RETURN:	0	- success
1734  *		errors returned by vms_iowait().
1735  */
1736 static int lmLogFileSystem(struct jfs_log * log, struct jfs_sb_info *sbi,
1737 			   int activate)
1738 {
1739 	int rc = 0;
1740 	int i;
1741 	struct logsuper *logsuper;
1742 	struct lbuf *bpsuper;
1743 	char *uuid = sbi->uuid;
1744 
1745 	/*
1746 	 * insert/remove file system device to log active file system list.
1747 	 */
1748 	if ((rc = lbmRead(log, 1, &bpsuper)))
1749 		return rc;
1750 
1751 	logsuper = (struct logsuper *) bpsuper->l_ldata;
1752 	if (activate) {
1753 		for (i = 0; i < MAX_ACTIVE; i++)
1754 			if (!memcmp(logsuper->active[i].uuid, NULL_UUID, 16)) {
1755 				memcpy(logsuper->active[i].uuid, uuid, 16);
1756 				sbi->aggregate = i;
1757 				break;
1758 			}
1759 		if (i == MAX_ACTIVE) {
1760 			jfs_warn("Too many file systems sharing journal!");
1761 			lbmFree(bpsuper);
1762 			return -EMFILE;	/* Is there a better rc? */
1763 		}
1764 	} else {
1765 		for (i = 0; i < MAX_ACTIVE; i++)
1766 			if (!memcmp(logsuper->active[i].uuid, uuid, 16)) {
1767 				memcpy(logsuper->active[i].uuid, NULL_UUID, 16);
1768 				break;
1769 			}
1770 		if (i == MAX_ACTIVE) {
1771 			jfs_warn("Somebody stomped on the journal!");
1772 			lbmFree(bpsuper);
1773 			return -EIO;
1774 		}
1775 
1776 	}
1777 
1778 	/*
1779 	 * synchronous write log superblock:
1780 	 *
1781 	 * write sidestream bypassing write queue:
1782 	 * at file system mount, log super block is updated for
1783 	 * activation of the file system before any log record
1784 	 * (MOUNT record) of the file system, and at file system
1785 	 * unmount, all meta data for the file system has been
1786 	 * flushed before log super block is updated for deactivation
1787 	 * of the file system.
1788 	 */
1789 	lbmDirectWrite(log, bpsuper, lbmWRITE | lbmRELEASE | lbmSYNC);
1790 	rc = lbmIOWait(bpsuper, lbmFREE);
1791 
1792 	return rc;
1793 }
1794 
1795 /*
1796  *		log buffer manager (lbm)
1797  *		------------------------
1798  *
1799  * special purpose buffer manager supporting log i/o requirements.
1800  *
1801  * per log write queue:
1802  * log pageout occurs in serial order by fifo write queue and
1803  * restricting to a single i/o in pregress at any one time.
1804  * a circular singly-linked list
1805  * (log->wrqueue points to the tail, and buffers are linked via
1806  * bp->wrqueue field), and
1807  * maintains log page in pageout ot waiting for pageout in serial pageout.
1808  */
1809 
1810 /*
1811  *	lbmLogInit()
1812  *
1813  * initialize per log I/O setup at lmLogInit()
1814  */
1815 static int lbmLogInit(struct jfs_log * log)
1816 {				/* log inode */
1817 	int i;
1818 	struct lbuf *lbuf;
1819 
1820 	jfs_info("lbmLogInit: log:0x%p", log);
1821 
1822 	/* initialize current buffer cursor */
1823 	log->bp = NULL;
1824 
1825 	/* initialize log device write queue */
1826 	log->wqueue = NULL;
1827 
1828 	/*
1829 	 * Each log has its own buffer pages allocated to it.  These are
1830 	 * not managed by the page cache.  This ensures that a transaction
1831 	 * writing to the log does not block trying to allocate a page from
1832 	 * the page cache (for the log).  This would be bad, since page
1833 	 * allocation waits on the kswapd thread that may be committing inodes
1834 	 * which would cause log activity.  Was that clear?  I'm trying to
1835 	 * avoid deadlock here.
1836 	 */
1837 	init_waitqueue_head(&log->free_wait);
1838 
1839 	log->lbuf_free = NULL;
1840 
1841 	for (i = 0; i < LOGPAGES;) {
1842 		char *buffer;
1843 		uint offset;
1844 		struct page *page;
1845 
1846 		buffer = (char *) get_zeroed_page(GFP_KERNEL);
1847 		if (buffer == NULL)
1848 			goto error;
1849 		page = virt_to_page(buffer);
1850 		for (offset = 0; offset < PAGE_SIZE; offset += LOGPSIZE) {
1851 			lbuf = kmalloc(sizeof(struct lbuf), GFP_KERNEL);
1852 			if (lbuf == NULL) {
1853 				if (offset == 0)
1854 					free_page((unsigned long) buffer);
1855 				goto error;
1856 			}
1857 			if (offset) /* we already have one reference */
1858 				get_page(page);
1859 			lbuf->l_offset = offset;
1860 			lbuf->l_ldata = buffer + offset;
1861 			lbuf->l_page = page;
1862 			lbuf->l_log = log;
1863 			init_waitqueue_head(&lbuf->l_ioevent);
1864 
1865 			lbuf->l_freelist = log->lbuf_free;
1866 			log->lbuf_free = lbuf;
1867 			i++;
1868 		}
1869 	}
1870 
1871 	return (0);
1872 
1873       error:
1874 	lbmLogShutdown(log);
1875 	return -ENOMEM;
1876 }
1877 
1878 
1879 /*
1880  *	lbmLogShutdown()
1881  *
1882  * finalize per log I/O setup at lmLogShutdown()
1883  */
1884 static void lbmLogShutdown(struct jfs_log * log)
1885 {
1886 	struct lbuf *lbuf;
1887 
1888 	jfs_info("lbmLogShutdown: log:0x%p", log);
1889 
1890 	lbuf = log->lbuf_free;
1891 	while (lbuf) {
1892 		struct lbuf *next = lbuf->l_freelist;
1893 		__free_page(lbuf->l_page);
1894 		kfree(lbuf);
1895 		lbuf = next;
1896 	}
1897 }
1898 
1899 
1900 /*
1901  *	lbmAllocate()
1902  *
1903  * allocate an empty log buffer
1904  */
1905 static struct lbuf *lbmAllocate(struct jfs_log * log, int pn)
1906 {
1907 	struct lbuf *bp;
1908 	unsigned long flags;
1909 
1910 	/*
1911 	 * recycle from log buffer freelist if any
1912 	 */
1913 	LCACHE_LOCK(flags);
1914 	LCACHE_SLEEP_COND(log->free_wait, (bp = log->lbuf_free), flags);
1915 	log->lbuf_free = bp->l_freelist;
1916 	LCACHE_UNLOCK(flags);
1917 
1918 	bp->l_flag = 0;
1919 
1920 	bp->l_wqnext = NULL;
1921 	bp->l_freelist = NULL;
1922 
1923 	bp->l_pn = pn;
1924 	bp->l_blkno = log->base + (pn << (L2LOGPSIZE - log->l2bsize));
1925 	bp->l_ceor = 0;
1926 
1927 	return bp;
1928 }
1929 
1930 
1931 /*
1932  *	lbmFree()
1933  *
1934  * release a log buffer to freelist
1935  */
1936 static void lbmFree(struct lbuf * bp)
1937 {
1938 	unsigned long flags;
1939 
1940 	LCACHE_LOCK(flags);
1941 
1942 	lbmfree(bp);
1943 
1944 	LCACHE_UNLOCK(flags);
1945 }
1946 
1947 static void lbmfree(struct lbuf * bp)
1948 {
1949 	struct jfs_log *log = bp->l_log;
1950 
1951 	assert(bp->l_wqnext == NULL);
1952 
1953 	/*
1954 	 * return the buffer to head of freelist
1955 	 */
1956 	bp->l_freelist = log->lbuf_free;
1957 	log->lbuf_free = bp;
1958 
1959 	wake_up(&log->free_wait);
1960 	return;
1961 }
1962 
1963 
1964 /*
1965  * NAME:	lbmRedrive
1966  *
1967  * FUNCTION:	add a log buffer to the log redrive list
1968  *
1969  * PARAMETER:
1970  *	bp	- log buffer
1971  *
1972  * NOTES:
1973  *	Takes log_redrive_lock.
1974  */
1975 static inline void lbmRedrive(struct lbuf *bp)
1976 {
1977 	unsigned long flags;
1978 
1979 	spin_lock_irqsave(&log_redrive_lock, flags);
1980 	bp->l_redrive_next = log_redrive_list;
1981 	log_redrive_list = bp;
1982 	spin_unlock_irqrestore(&log_redrive_lock, flags);
1983 
1984 	wake_up_process(jfsIOthread);
1985 }
1986 
1987 
1988 /*
1989  *	lbmRead()
1990  */
1991 static int lbmRead(struct jfs_log * log, int pn, struct lbuf ** bpp)
1992 {
1993 	struct bio *bio;
1994 	struct lbuf *bp;
1995 
1996 	/*
1997 	 * allocate a log buffer
1998 	 */
1999 	*bpp = bp = lbmAllocate(log, pn);
2000 	jfs_info("lbmRead: bp:0x%p pn:0x%x", bp, pn);
2001 
2002 	bp->l_flag |= lbmREAD;
2003 
2004 	bio = bio_alloc(GFP_NOFS, 1);
2005 
2006 	bio->bi_sector = bp->l_blkno << (log->l2bsize - 9);
2007 	bio->bi_bdev = log->bdev;
2008 	bio->bi_io_vec[0].bv_page = bp->l_page;
2009 	bio->bi_io_vec[0].bv_len = LOGPSIZE;
2010 	bio->bi_io_vec[0].bv_offset = bp->l_offset;
2011 
2012 	bio->bi_vcnt = 1;
2013 	bio->bi_idx = 0;
2014 	bio->bi_size = LOGPSIZE;
2015 
2016 	bio->bi_end_io = lbmIODone;
2017 	bio->bi_private = bp;
2018 	submit_bio(READ_SYNC, bio);
2019 
2020 	wait_event(bp->l_ioevent, (bp->l_flag != lbmREAD));
2021 
2022 	return 0;
2023 }
2024 
2025 
2026 /*
2027  *	lbmWrite()
2028  *
2029  * buffer at head of pageout queue stays after completion of
2030  * partial-page pageout and redriven by explicit initiation of
2031  * pageout by caller until full-page pageout is completed and
2032  * released.
2033  *
2034  * device driver i/o done redrives pageout of new buffer at
2035  * head of pageout queue when current buffer at head of pageout
2036  * queue is released at the completion of its full-page pageout.
2037  *
2038  * LOGGC_LOCK() serializes lbmWrite() by lmNextPage() and lmGroupCommit().
2039  * LCACHE_LOCK() serializes xflag between lbmWrite() and lbmIODone()
2040  */
2041 static void lbmWrite(struct jfs_log * log, struct lbuf * bp, int flag,
2042 		     int cant_block)
2043 {
2044 	struct lbuf *tail;
2045 	unsigned long flags;
2046 
2047 	jfs_info("lbmWrite: bp:0x%p flag:0x%x pn:0x%x", bp, flag, bp->l_pn);
2048 
2049 	/* map the logical block address to physical block address */
2050 	bp->l_blkno =
2051 	    log->base + (bp->l_pn << (L2LOGPSIZE - log->l2bsize));
2052 
2053 	LCACHE_LOCK(flags);		/* disable+lock */
2054 
2055 	/*
2056 	 * initialize buffer for device driver
2057 	 */
2058 	bp->l_flag = flag;
2059 
2060 	/*
2061 	 *	insert bp at tail of write queue associated with log
2062 	 *
2063 	 * (request is either for bp already/currently at head of queue
2064 	 * or new bp to be inserted at tail)
2065 	 */
2066 	tail = log->wqueue;
2067 
2068 	/* is buffer not already on write queue ? */
2069 	if (bp->l_wqnext == NULL) {
2070 		/* insert at tail of wqueue */
2071 		if (tail == NULL) {
2072 			log->wqueue = bp;
2073 			bp->l_wqnext = bp;
2074 		} else {
2075 			log->wqueue = bp;
2076 			bp->l_wqnext = tail->l_wqnext;
2077 			tail->l_wqnext = bp;
2078 		}
2079 
2080 		tail = bp;
2081 	}
2082 
2083 	/* is buffer at head of wqueue and for write ? */
2084 	if ((bp != tail->l_wqnext) || !(flag & lbmWRITE)) {
2085 		LCACHE_UNLOCK(flags);	/* unlock+enable */
2086 		return;
2087 	}
2088 
2089 	LCACHE_UNLOCK(flags);	/* unlock+enable */
2090 
2091 	if (cant_block)
2092 		lbmRedrive(bp);
2093 	else if (flag & lbmSYNC)
2094 		lbmStartIO(bp);
2095 	else {
2096 		LOGGC_UNLOCK(log);
2097 		lbmStartIO(bp);
2098 		LOGGC_LOCK(log);
2099 	}
2100 }
2101 
2102 
2103 /*
2104  *	lbmDirectWrite()
2105  *
2106  * initiate pageout bypassing write queue for sidestream
2107  * (e.g., log superblock) write;
2108  */
2109 static void lbmDirectWrite(struct jfs_log * log, struct lbuf * bp, int flag)
2110 {
2111 	jfs_info("lbmDirectWrite: bp:0x%p flag:0x%x pn:0x%x",
2112 		 bp, flag, bp->l_pn);
2113 
2114 	/*
2115 	 * initialize buffer for device driver
2116 	 */
2117 	bp->l_flag = flag | lbmDIRECT;
2118 
2119 	/* map the logical block address to physical block address */
2120 	bp->l_blkno =
2121 	    log->base + (bp->l_pn << (L2LOGPSIZE - log->l2bsize));
2122 
2123 	/*
2124 	 *	initiate pageout of the page
2125 	 */
2126 	lbmStartIO(bp);
2127 }
2128 
2129 
2130 /*
2131  * NAME:	lbmStartIO()
2132  *
2133  * FUNCTION:	Interface to DD strategy routine
2134  *
2135  * RETURN:	none
2136  *
2137  * serialization: LCACHE_LOCK() is NOT held during log i/o;
2138  */
2139 static void lbmStartIO(struct lbuf * bp)
2140 {
2141 	struct bio *bio;
2142 	struct jfs_log *log = bp->l_log;
2143 
2144 	jfs_info("lbmStartIO\n");
2145 
2146 	bio = bio_alloc(GFP_NOFS, 1);
2147 	bio->bi_sector = bp->l_blkno << (log->l2bsize - 9);
2148 	bio->bi_bdev = log->bdev;
2149 	bio->bi_io_vec[0].bv_page = bp->l_page;
2150 	bio->bi_io_vec[0].bv_len = LOGPSIZE;
2151 	bio->bi_io_vec[0].bv_offset = bp->l_offset;
2152 
2153 	bio->bi_vcnt = 1;
2154 	bio->bi_idx = 0;
2155 	bio->bi_size = LOGPSIZE;
2156 
2157 	bio->bi_end_io = lbmIODone;
2158 	bio->bi_private = bp;
2159 
2160 	/* check if journaling to disk has been disabled */
2161 	if (log->no_integrity) {
2162 		bio->bi_size = 0;
2163 		lbmIODone(bio, 0);
2164 	} else {
2165 		submit_bio(WRITE_SYNC, bio);
2166 		INCREMENT(lmStat.submitted);
2167 	}
2168 }
2169 
2170 
2171 /*
2172  *	lbmIOWait()
2173  */
2174 static int lbmIOWait(struct lbuf * bp, int flag)
2175 {
2176 	unsigned long flags;
2177 	int rc = 0;
2178 
2179 	jfs_info("lbmIOWait1: bp:0x%p flag:0x%x:0x%x", bp, bp->l_flag, flag);
2180 
2181 	LCACHE_LOCK(flags);		/* disable+lock */
2182 
2183 	LCACHE_SLEEP_COND(bp->l_ioevent, (bp->l_flag & lbmDONE), flags);
2184 
2185 	rc = (bp->l_flag & lbmERROR) ? -EIO : 0;
2186 
2187 	if (flag & lbmFREE)
2188 		lbmfree(bp);
2189 
2190 	LCACHE_UNLOCK(flags);	/* unlock+enable */
2191 
2192 	jfs_info("lbmIOWait2: bp:0x%p flag:0x%x:0x%x", bp, bp->l_flag, flag);
2193 	return rc;
2194 }
2195 
2196 /*
2197  *	lbmIODone()
2198  *
2199  * executed at INTIODONE level
2200  */
2201 static void lbmIODone(struct bio *bio, int error)
2202 {
2203 	struct lbuf *bp = bio->bi_private;
2204 	struct lbuf *nextbp, *tail;
2205 	struct jfs_log *log;
2206 	unsigned long flags;
2207 
2208 	/*
2209 	 * get back jfs buffer bound to the i/o buffer
2210 	 */
2211 	jfs_info("lbmIODone: bp:0x%p flag:0x%x", bp, bp->l_flag);
2212 
2213 	LCACHE_LOCK(flags);		/* disable+lock */
2214 
2215 	bp->l_flag |= lbmDONE;
2216 
2217 	if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
2218 		bp->l_flag |= lbmERROR;
2219 
2220 		jfs_err("lbmIODone: I/O error in JFS log");
2221 	}
2222 
2223 	bio_put(bio);
2224 
2225 	/*
2226 	 *	pagein completion
2227 	 */
2228 	if (bp->l_flag & lbmREAD) {
2229 		bp->l_flag &= ~lbmREAD;
2230 
2231 		LCACHE_UNLOCK(flags);	/* unlock+enable */
2232 
2233 		/* wakeup I/O initiator */
2234 		LCACHE_WAKEUP(&bp->l_ioevent);
2235 
2236 		return;
2237 	}
2238 
2239 	/*
2240 	 *	pageout completion
2241 	 *
2242 	 * the bp at the head of write queue has completed pageout.
2243 	 *
2244 	 * if single-commit/full-page pageout, remove the current buffer
2245 	 * from head of pageout queue, and redrive pageout with
2246 	 * the new buffer at head of pageout queue;
2247 	 * otherwise, the partial-page pageout buffer stays at
2248 	 * the head of pageout queue to be redriven for pageout
2249 	 * by lmGroupCommit() until full-page pageout is completed.
2250 	 */
2251 	bp->l_flag &= ~lbmWRITE;
2252 	INCREMENT(lmStat.pagedone);
2253 
2254 	/* update committed lsn */
2255 	log = bp->l_log;
2256 	log->clsn = (bp->l_pn << L2LOGPSIZE) + bp->l_ceor;
2257 
2258 	if (bp->l_flag & lbmDIRECT) {
2259 		LCACHE_WAKEUP(&bp->l_ioevent);
2260 		LCACHE_UNLOCK(flags);
2261 		return;
2262 	}
2263 
2264 	tail = log->wqueue;
2265 
2266 	/* single element queue */
2267 	if (bp == tail) {
2268 		/* remove head buffer of full-page pageout
2269 		 * from log device write queue
2270 		 */
2271 		if (bp->l_flag & lbmRELEASE) {
2272 			log->wqueue = NULL;
2273 			bp->l_wqnext = NULL;
2274 		}
2275 	}
2276 	/* multi element queue */
2277 	else {
2278 		/* remove head buffer of full-page pageout
2279 		 * from log device write queue
2280 		 */
2281 		if (bp->l_flag & lbmRELEASE) {
2282 			nextbp = tail->l_wqnext = bp->l_wqnext;
2283 			bp->l_wqnext = NULL;
2284 
2285 			/*
2286 			 * redrive pageout of next page at head of write queue:
2287 			 * redrive next page without any bound tblk
2288 			 * (i.e., page w/o any COMMIT records), or
2289 			 * first page of new group commit which has been
2290 			 * queued after current page (subsequent pageout
2291 			 * is performed synchronously, except page without
2292 			 * any COMMITs) by lmGroupCommit() as indicated
2293 			 * by lbmWRITE flag;
2294 			 */
2295 			if (nextbp->l_flag & lbmWRITE) {
2296 				/*
2297 				 * We can't do the I/O at interrupt time.
2298 				 * The jfsIO thread can do it
2299 				 */
2300 				lbmRedrive(nextbp);
2301 			}
2302 		}
2303 	}
2304 
2305 	/*
2306 	 *	synchronous pageout:
2307 	 *
2308 	 * buffer has not necessarily been removed from write queue
2309 	 * (e.g., synchronous write of partial-page with COMMIT):
2310 	 * leave buffer for i/o initiator to dispose
2311 	 */
2312 	if (bp->l_flag & lbmSYNC) {
2313 		LCACHE_UNLOCK(flags);	/* unlock+enable */
2314 
2315 		/* wakeup I/O initiator */
2316 		LCACHE_WAKEUP(&bp->l_ioevent);
2317 	}
2318 
2319 	/*
2320 	 *	Group Commit pageout:
2321 	 */
2322 	else if (bp->l_flag & lbmGC) {
2323 		LCACHE_UNLOCK(flags);
2324 		lmPostGC(bp);
2325 	}
2326 
2327 	/*
2328 	 *	asynchronous pageout:
2329 	 *
2330 	 * buffer must have been removed from write queue:
2331 	 * insert buffer at head of freelist where it can be recycled
2332 	 */
2333 	else {
2334 		assert(bp->l_flag & lbmRELEASE);
2335 		assert(bp->l_flag & lbmFREE);
2336 		lbmfree(bp);
2337 
2338 		LCACHE_UNLOCK(flags);	/* unlock+enable */
2339 	}
2340 }
2341 
2342 int jfsIOWait(void *arg)
2343 {
2344 	struct lbuf *bp;
2345 
2346 	do {
2347 		spin_lock_irq(&log_redrive_lock);
2348 		while ((bp = log_redrive_list)) {
2349 			log_redrive_list = bp->l_redrive_next;
2350 			bp->l_redrive_next = NULL;
2351 			spin_unlock_irq(&log_redrive_lock);
2352 			lbmStartIO(bp);
2353 			spin_lock_irq(&log_redrive_lock);
2354 		}
2355 
2356 		if (freezing(current)) {
2357 			spin_unlock_irq(&log_redrive_lock);
2358 			refrigerator();
2359 		} else {
2360 			set_current_state(TASK_INTERRUPTIBLE);
2361 			spin_unlock_irq(&log_redrive_lock);
2362 			schedule();
2363 			__set_current_state(TASK_RUNNING);
2364 		}
2365 	} while (!kthread_should_stop());
2366 
2367 	jfs_info("jfsIOWait being killed!");
2368 	return 0;
2369 }
2370 
2371 /*
2372  * NAME:	lmLogFormat()/jfs_logform()
2373  *
2374  * FUNCTION:	format file system log
2375  *
2376  * PARAMETERS:
2377  *	log	- volume log
2378  *	logAddress - start address of log space in FS block
2379  *	logSize	- length of log space in FS block;
2380  *
2381  * RETURN:	0	- success
2382  *		-EIO	- i/o error
2383  *
2384  * XXX: We're synchronously writing one page at a time.  This needs to
2385  *	be improved by writing multiple pages at once.
2386  */
2387 int lmLogFormat(struct jfs_log *log, s64 logAddress, int logSize)
2388 {
2389 	int rc = -EIO;
2390 	struct jfs_sb_info *sbi;
2391 	struct logsuper *logsuper;
2392 	struct logpage *lp;
2393 	int lspn;		/* log sequence page number */
2394 	struct lrd *lrd_ptr;
2395 	int npages = 0;
2396 	struct lbuf *bp;
2397 
2398 	jfs_info("lmLogFormat: logAddress:%Ld logSize:%d",
2399 		 (long long)logAddress, logSize);
2400 
2401 	sbi = list_entry(log->sb_list.next, struct jfs_sb_info, log_list);
2402 
2403 	/* allocate a log buffer */
2404 	bp = lbmAllocate(log, 1);
2405 
2406 	npages = logSize >> sbi->l2nbperpage;
2407 
2408 	/*
2409 	 *	log space:
2410 	 *
2411 	 * page 0 - reserved;
2412 	 * page 1 - log superblock;
2413 	 * page 2 - log data page: A SYNC log record is written
2414 	 *	    into this page at logform time;
2415 	 * pages 3-N - log data page: set to empty log data pages;
2416 	 */
2417 	/*
2418 	 *	init log superblock: log page 1
2419 	 */
2420 	logsuper = (struct logsuper *) bp->l_ldata;
2421 
2422 	logsuper->magic = cpu_to_le32(LOGMAGIC);
2423 	logsuper->version = cpu_to_le32(LOGVERSION);
2424 	logsuper->state = cpu_to_le32(LOGREDONE);
2425 	logsuper->flag = cpu_to_le32(sbi->mntflag);	/* ? */
2426 	logsuper->size = cpu_to_le32(npages);
2427 	logsuper->bsize = cpu_to_le32(sbi->bsize);
2428 	logsuper->l2bsize = cpu_to_le32(sbi->l2bsize);
2429 	logsuper->end = cpu_to_le32(2 * LOGPSIZE + LOGPHDRSIZE + LOGRDSIZE);
2430 
2431 	bp->l_flag = lbmWRITE | lbmSYNC | lbmDIRECT;
2432 	bp->l_blkno = logAddress + sbi->nbperpage;
2433 	lbmStartIO(bp);
2434 	if ((rc = lbmIOWait(bp, 0)))
2435 		goto exit;
2436 
2437 	/*
2438 	 *	init pages 2 to npages-1 as log data pages:
2439 	 *
2440 	 * log page sequence number (lpsn) initialization:
2441 	 *
2442 	 * pn:   0     1     2     3                 n-1
2443 	 *       +-----+-----+=====+=====+===.....===+=====+
2444 	 * lspn:             N-1   0     1           N-2
2445 	 *                   <--- N page circular file ---->
2446 	 *
2447 	 * the N (= npages-2) data pages of the log is maintained as
2448 	 * a circular file for the log records;
2449 	 * lpsn grows by 1 monotonically as each log page is written
2450 	 * to the circular file of the log;
2451 	 * and setLogpage() will not reset the page number even if
2452 	 * the eor is equal to LOGPHDRSIZE. In order for binary search
2453 	 * still work in find log end process, we have to simulate the
2454 	 * log wrap situation at the log format time.
2455 	 * The 1st log page written will have the highest lpsn. Then
2456 	 * the succeeding log pages will have ascending order of
2457 	 * the lspn starting from 0, ... (N-2)
2458 	 */
2459 	lp = (struct logpage *) bp->l_ldata;
2460 	/*
2461 	 * initialize 1st log page to be written: lpsn = N - 1,
2462 	 * write a SYNCPT log record is written to this page
2463 	 */
2464 	lp->h.page = lp->t.page = cpu_to_le32(npages - 3);
2465 	lp->h.eor = lp->t.eor = cpu_to_le16(LOGPHDRSIZE + LOGRDSIZE);
2466 
2467 	lrd_ptr = (struct lrd *) &lp->data;
2468 	lrd_ptr->logtid = 0;
2469 	lrd_ptr->backchain = 0;
2470 	lrd_ptr->type = cpu_to_le16(LOG_SYNCPT);
2471 	lrd_ptr->length = 0;
2472 	lrd_ptr->log.syncpt.sync = 0;
2473 
2474 	bp->l_blkno += sbi->nbperpage;
2475 	bp->l_flag = lbmWRITE | lbmSYNC | lbmDIRECT;
2476 	lbmStartIO(bp);
2477 	if ((rc = lbmIOWait(bp, 0)))
2478 		goto exit;
2479 
2480 	/*
2481 	 *	initialize succeeding log pages: lpsn = 0, 1, ..., (N-2)
2482 	 */
2483 	for (lspn = 0; lspn < npages - 3; lspn++) {
2484 		lp->h.page = lp->t.page = cpu_to_le32(lspn);
2485 		lp->h.eor = lp->t.eor = cpu_to_le16(LOGPHDRSIZE);
2486 
2487 		bp->l_blkno += sbi->nbperpage;
2488 		bp->l_flag = lbmWRITE | lbmSYNC | lbmDIRECT;
2489 		lbmStartIO(bp);
2490 		if ((rc = lbmIOWait(bp, 0)))
2491 			goto exit;
2492 	}
2493 
2494 	rc = 0;
2495 exit:
2496 	/*
2497 	 *	finalize log
2498 	 */
2499 	/* release the buffer */
2500 	lbmFree(bp);
2501 
2502 	return rc;
2503 }
2504 
2505 #ifdef CONFIG_JFS_STATISTICS
2506 int jfs_lmstats_read(char *buffer, char **start, off_t offset, int length,
2507 		      int *eof, void *data)
2508 {
2509 	int len = 0;
2510 	off_t begin;
2511 
2512 	len += sprintf(buffer,
2513 		       "JFS Logmgr stats\n"
2514 		       "================\n"
2515 		       "commits = %d\n"
2516 		       "writes submitted = %d\n"
2517 		       "writes completed = %d\n"
2518 		       "full pages submitted = %d\n"
2519 		       "partial pages submitted = %d\n",
2520 		       lmStat.commit,
2521 		       lmStat.submitted,
2522 		       lmStat.pagedone,
2523 		       lmStat.full_page,
2524 		       lmStat.partial_page);
2525 
2526 	begin = offset;
2527 	*start = buffer + begin;
2528 	len -= begin;
2529 
2530 	if (len > length)
2531 		len = length;
2532 	else
2533 		*eof = 1;
2534 
2535 	if (len < 0)
2536 		len = 0;
2537 
2538 	return len;
2539 }
2540 #endif /* CONFIG_JFS_STATISTICS */
2541