xref: /openbmc/linux/fs/jfs/jfs_logmgr.c (revision 1da177e4)
1 /*
2  *   Copyright (C) International Business Machines Corp., 2000-2004
3  *   Portions Copyright (C) Christoph Hellwig, 2001-2002
4  *
5  *   This program is free software;  you can redistribute it and/or modify
6  *   it under the terms of the GNU General Public License as published by
7  *   the Free Software Foundation; either version 2 of the License, or
8  *   (at your option) any later version.
9  *
10  *   This program is distributed in the hope that it will be useful,
11  *   but WITHOUT ANY WARRANTY;  without even the implied warranty of
12  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See
13  *   the GNU General Public License for more details.
14  *
15  *   You should have received a copy of the GNU General Public License
16  *   along with this program;  if not, write to the Free Software
17  *   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18  */
19 
20 /*
21  *	jfs_logmgr.c: log manager
22  *
23  * for related information, see transaction manager (jfs_txnmgr.c), and
24  * recovery manager (jfs_logredo.c).
25  *
26  * note: for detail, RTFS.
27  *
28  *	log buffer manager:
29  * special purpose buffer manager supporting log i/o requirements.
30  * per log serial pageout of logpage
31  * queuing i/o requests and redrive i/o at iodone
32  * maintain current logpage buffer
33  * no caching since append only
34  * appropriate jfs buffer cache buffers as needed
35  *
36  *	group commit:
37  * transactions which wrote COMMIT records in the same in-memory
38  * log page during the pageout of previous/current log page(s) are
39  * committed together by the pageout of the page.
40  *
41  *	TBD lazy commit:
42  * transactions are committed asynchronously when the log page
43  * containing it COMMIT is paged out when it becomes full;
44  *
45  *	serialization:
46  * . a per log lock serialize log write.
47  * . a per log lock serialize group commit.
48  * . a per log lock serialize log open/close;
49  *
50  *	TBD log integrity:
51  * careful-write (ping-pong) of last logpage to recover from crash
52  * in overwrite.
53  * detection of split (out-of-order) write of physical sectors
54  * of last logpage via timestamp at end of each sector
55  * with its mirror data array at trailer).
56  *
57  *	alternatives:
58  * lsn - 64-bit monotonically increasing integer vs
59  * 32-bit lspn and page eor.
60  */
61 
62 #include <linux/fs.h>
63 #include <linux/blkdev.h>
64 #include <linux/interrupt.h>
65 #include <linux/smp_lock.h>
66 #include <linux/completion.h>
67 #include <linux/buffer_head.h>		/* for sync_blockdev() */
68 #include <linux/bio.h>
69 #include <linux/suspend.h>
70 #include <linux/delay.h>
71 #include "jfs_incore.h"
72 #include "jfs_filsys.h"
73 #include "jfs_metapage.h"
74 #include "jfs_txnmgr.h"
75 #include "jfs_debug.h"
76 
77 
78 /*
79  * lbuf's ready to be redriven.  Protected by log_redrive_lock (jfsIO thread)
80  */
81 static struct lbuf *log_redrive_list;
82 static DEFINE_SPINLOCK(log_redrive_lock);
83 DECLARE_WAIT_QUEUE_HEAD(jfs_IO_thread_wait);
84 
85 
86 /*
87  *	log read/write serialization (per log)
88  */
89 #define LOG_LOCK_INIT(log)	init_MUTEX(&(log)->loglock)
90 #define LOG_LOCK(log)		down(&((log)->loglock))
91 #define LOG_UNLOCK(log)		up(&((log)->loglock))
92 
93 
94 /*
95  *	log group commit serialization (per log)
96  */
97 
98 #define LOGGC_LOCK_INIT(log)	spin_lock_init(&(log)->gclock)
99 #define LOGGC_LOCK(log)		spin_lock_irq(&(log)->gclock)
100 #define LOGGC_UNLOCK(log)	spin_unlock_irq(&(log)->gclock)
101 #define LOGGC_WAKEUP(tblk)	wake_up_all(&(tblk)->gcwait)
102 
103 /*
104  *	log sync serialization (per log)
105  */
106 #define	LOGSYNC_DELTA(logsize)		min((logsize)/8, 128*LOGPSIZE)
107 #define	LOGSYNC_BARRIER(logsize)	((logsize)/4)
108 /*
109 #define	LOGSYNC_DELTA(logsize)		min((logsize)/4, 256*LOGPSIZE)
110 #define	LOGSYNC_BARRIER(logsize)	((logsize)/2)
111 */
112 
113 
114 /*
115  *	log buffer cache synchronization
116  */
117 static DEFINE_SPINLOCK(jfsLCacheLock);
118 
119 #define	LCACHE_LOCK(flags)	spin_lock_irqsave(&jfsLCacheLock, flags)
120 #define	LCACHE_UNLOCK(flags)	spin_unlock_irqrestore(&jfsLCacheLock, flags)
121 
122 /*
123  * See __SLEEP_COND in jfs_locks.h
124  */
125 #define LCACHE_SLEEP_COND(wq, cond, flags)	\
126 do {						\
127 	if (cond)				\
128 		break;				\
129 	__SLEEP_COND(wq, cond, LCACHE_LOCK(flags), LCACHE_UNLOCK(flags)); \
130 } while (0)
131 
132 #define	LCACHE_WAKEUP(event)	wake_up(event)
133 
134 
135 /*
136  *	lbuf buffer cache (lCache) control
137  */
138 /* log buffer manager pageout control (cumulative, inclusive) */
139 #define	lbmREAD		0x0001
140 #define	lbmWRITE	0x0002	/* enqueue at tail of write queue;
141 				 * init pageout if at head of queue;
142 				 */
143 #define	lbmRELEASE	0x0004	/* remove from write queue
144 				 * at completion of pageout;
145 				 * do not free/recycle it yet:
146 				 * caller will free it;
147 				 */
148 #define	lbmSYNC		0x0008	/* do not return to freelist
149 				 * when removed from write queue;
150 				 */
151 #define lbmFREE		0x0010	/* return to freelist
152 				 * at completion of pageout;
153 				 * the buffer may be recycled;
154 				 */
155 #define	lbmDONE		0x0020
156 #define	lbmERROR	0x0040
157 #define lbmGC		0x0080	/* lbmIODone to perform post-GC processing
158 				 * of log page
159 				 */
160 #define lbmDIRECT	0x0100
161 
162 /*
163  * Global list of active external journals
164  */
165 static LIST_HEAD(jfs_external_logs);
166 static struct jfs_log *dummy_log = NULL;
167 static DECLARE_MUTEX(jfs_log_sem);
168 
169 /*
170  * external references
171  */
172 extern void txLazyUnlock(struct tblock * tblk);
173 extern int jfs_stop_threads;
174 extern struct completion jfsIOwait;
175 extern int jfs_tlocks_low;
176 
177 /*
178  * forward references
179  */
180 static int lmWriteRecord(struct jfs_log * log, struct tblock * tblk,
181 			 struct lrd * lrd, struct tlock * tlck);
182 
183 static int lmNextPage(struct jfs_log * log);
184 static int lmLogFileSystem(struct jfs_log * log, struct jfs_sb_info *sbi,
185 			   int activate);
186 
187 static int open_inline_log(struct super_block *sb);
188 static int open_dummy_log(struct super_block *sb);
189 static int lbmLogInit(struct jfs_log * log);
190 static void lbmLogShutdown(struct jfs_log * log);
191 static struct lbuf *lbmAllocate(struct jfs_log * log, int);
192 static void lbmFree(struct lbuf * bp);
193 static void lbmfree(struct lbuf * bp);
194 static int lbmRead(struct jfs_log * log, int pn, struct lbuf ** bpp);
195 static void lbmWrite(struct jfs_log * log, struct lbuf * bp, int flag, int cant_block);
196 static void lbmDirectWrite(struct jfs_log * log, struct lbuf * bp, int flag);
197 static int lbmIOWait(struct lbuf * bp, int flag);
198 static bio_end_io_t lbmIODone;
199 static void lbmStartIO(struct lbuf * bp);
200 static void lmGCwrite(struct jfs_log * log, int cant_block);
201 static int lmLogSync(struct jfs_log * log, int nosyncwait);
202 
203 
204 
205 /*
206  *	statistics
207  */
208 #ifdef CONFIG_JFS_STATISTICS
209 static struct lmStat {
210 	uint commit;		/* # of commit */
211 	uint pagedone;		/* # of page written */
212 	uint submitted;		/* # of pages submitted */
213 	uint full_page;		/* # of full pages submitted */
214 	uint partial_page;	/* # of partial pages submitted */
215 } lmStat;
216 #endif
217 
218 
219 /*
220  * NAME:	lmLog()
221  *
222  * FUNCTION:	write a log record;
223  *
224  * PARAMETER:
225  *
226  * RETURN:	lsn - offset to the next log record to write (end-of-log);
227  *		-1  - error;
228  *
229  * note: todo: log error handler
230  */
231 int lmLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
232 	  struct tlock * tlck)
233 {
234 	int lsn;
235 	int diffp, difft;
236 	struct metapage *mp = NULL;
237 
238 	jfs_info("lmLog: log:0x%p tblk:0x%p, lrd:0x%p tlck:0x%p",
239 		 log, tblk, lrd, tlck);
240 
241 	LOG_LOCK(log);
242 
243 	/* log by (out-of-transaction) JFS ? */
244 	if (tblk == NULL)
245 		goto writeRecord;
246 
247 	/* log from page ? */
248 	if (tlck == NULL ||
249 	    tlck->type & tlckBTROOT || (mp = tlck->mp) == NULL)
250 		goto writeRecord;
251 
252 	/*
253 	 *      initialize/update page/transaction recovery lsn
254 	 */
255 	lsn = log->lsn;
256 
257 	LOGSYNC_LOCK(log);
258 
259 	/*
260 	 * initialize page lsn if first log write of the page
261 	 */
262 	if (mp->lsn == 0) {
263 		mp->log = log;
264 		mp->lsn = lsn;
265 		log->count++;
266 
267 		/* insert page at tail of logsynclist */
268 		list_add_tail(&mp->synclist, &log->synclist);
269 	}
270 
271 	/*
272 	 *      initialize/update lsn of tblock of the page
273 	 *
274 	 * transaction inherits oldest lsn of pages associated
275 	 * with allocation/deallocation of resources (their
276 	 * log records are used to reconstruct allocation map
277 	 * at recovery time: inode for inode allocation map,
278 	 * B+-tree index of extent descriptors for block
279 	 * allocation map);
280 	 * allocation map pages inherit transaction lsn at
281 	 * commit time to allow forwarding log syncpt past log
282 	 * records associated with allocation/deallocation of
283 	 * resources only after persistent map of these map pages
284 	 * have been updated and propagated to home.
285 	 */
286 	/*
287 	 * initialize transaction lsn:
288 	 */
289 	if (tblk->lsn == 0) {
290 		/* inherit lsn of its first page logged */
291 		tblk->lsn = mp->lsn;
292 		log->count++;
293 
294 		/* insert tblock after the page on logsynclist */
295 		list_add(&tblk->synclist, &mp->synclist);
296 	}
297 	/*
298 	 * update transaction lsn:
299 	 */
300 	else {
301 		/* inherit oldest/smallest lsn of page */
302 		logdiff(diffp, mp->lsn, log);
303 		logdiff(difft, tblk->lsn, log);
304 		if (diffp < difft) {
305 			/* update tblock lsn with page lsn */
306 			tblk->lsn = mp->lsn;
307 
308 			/* move tblock after page on logsynclist */
309 			list_move(&tblk->synclist, &mp->synclist);
310 		}
311 	}
312 
313 	LOGSYNC_UNLOCK(log);
314 
315 	/*
316 	 *      write the log record
317 	 */
318       writeRecord:
319 	lsn = lmWriteRecord(log, tblk, lrd, tlck);
320 
321 	/*
322 	 * forward log syncpt if log reached next syncpt trigger
323 	 */
324 	logdiff(diffp, lsn, log);
325 	if (diffp >= log->nextsync)
326 		lsn = lmLogSync(log, 0);
327 
328 	/* update end-of-log lsn */
329 	log->lsn = lsn;
330 
331 	LOG_UNLOCK(log);
332 
333 	/* return end-of-log address */
334 	return lsn;
335 }
336 
337 
338 /*
339  * NAME:	lmWriteRecord()
340  *
341  * FUNCTION:	move the log record to current log page
342  *
343  * PARAMETER:	cd	- commit descriptor
344  *
345  * RETURN:	end-of-log address
346  *
347  * serialization: LOG_LOCK() held on entry/exit
348  */
349 static int
350 lmWriteRecord(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
351 	      struct tlock * tlck)
352 {
353 	int lsn = 0;		/* end-of-log address */
354 	struct lbuf *bp;	/* dst log page buffer */
355 	struct logpage *lp;	/* dst log page */
356 	caddr_t dst;		/* destination address in log page */
357 	int dstoffset;		/* end-of-log offset in log page */
358 	int freespace;		/* free space in log page */
359 	caddr_t p;		/* src meta-data page */
360 	caddr_t src;
361 	int srclen;
362 	int nbytes;		/* number of bytes to move */
363 	int i;
364 	int len;
365 	struct linelock *linelock;
366 	struct lv *lv;
367 	struct lvd *lvd;
368 	int l2linesize;
369 
370 	len = 0;
371 
372 	/* retrieve destination log page to write */
373 	bp = (struct lbuf *) log->bp;
374 	lp = (struct logpage *) bp->l_ldata;
375 	dstoffset = log->eor;
376 
377 	/* any log data to write ? */
378 	if (tlck == NULL)
379 		goto moveLrd;
380 
381 	/*
382 	 *      move log record data
383 	 */
384 	/* retrieve source meta-data page to log */
385 	if (tlck->flag & tlckPAGELOCK) {
386 		p = (caddr_t) (tlck->mp->data);
387 		linelock = (struct linelock *) & tlck->lock;
388 	}
389 	/* retrieve source in-memory inode to log */
390 	else if (tlck->flag & tlckINODELOCK) {
391 		if (tlck->type & tlckDTREE)
392 			p = (caddr_t) &JFS_IP(tlck->ip)->i_dtroot;
393 		else
394 			p = (caddr_t) &JFS_IP(tlck->ip)->i_xtroot;
395 		linelock = (struct linelock *) & tlck->lock;
396 	}
397 #ifdef	_JFS_WIP
398 	else if (tlck->flag & tlckINLINELOCK) {
399 
400 		inlinelock = (struct inlinelock *) & tlck;
401 		p = (caddr_t) & inlinelock->pxd;
402 		linelock = (struct linelock *) & tlck;
403 	}
404 #endif				/* _JFS_WIP */
405 	else {
406 		jfs_err("lmWriteRecord: UFO tlck:0x%p", tlck);
407 		return 0;	/* Probably should trap */
408 	}
409 	l2linesize = linelock->l2linesize;
410 
411       moveData:
412 	ASSERT(linelock->index <= linelock->maxcnt);
413 
414 	lv = linelock->lv;
415 	for (i = 0; i < linelock->index; i++, lv++) {
416 		if (lv->length == 0)
417 			continue;
418 
419 		/* is page full ? */
420 		if (dstoffset >= LOGPSIZE - LOGPTLRSIZE) {
421 			/* page become full: move on to next page */
422 			lmNextPage(log);
423 
424 			bp = log->bp;
425 			lp = (struct logpage *) bp->l_ldata;
426 			dstoffset = LOGPHDRSIZE;
427 		}
428 
429 		/*
430 		 * move log vector data
431 		 */
432 		src = (u8 *) p + (lv->offset << l2linesize);
433 		srclen = lv->length << l2linesize;
434 		len += srclen;
435 		while (srclen > 0) {
436 			freespace = (LOGPSIZE - LOGPTLRSIZE) - dstoffset;
437 			nbytes = min(freespace, srclen);
438 			dst = (caddr_t) lp + dstoffset;
439 			memcpy(dst, src, nbytes);
440 			dstoffset += nbytes;
441 
442 			/* is page not full ? */
443 			if (dstoffset < LOGPSIZE - LOGPTLRSIZE)
444 				break;
445 
446 			/* page become full: move on to next page */
447 			lmNextPage(log);
448 
449 			bp = (struct lbuf *) log->bp;
450 			lp = (struct logpage *) bp->l_ldata;
451 			dstoffset = LOGPHDRSIZE;
452 
453 			srclen -= nbytes;
454 			src += nbytes;
455 		}
456 
457 		/*
458 		 * move log vector descriptor
459 		 */
460 		len += 4;
461 		lvd = (struct lvd *) ((caddr_t) lp + dstoffset);
462 		lvd->offset = cpu_to_le16(lv->offset);
463 		lvd->length = cpu_to_le16(lv->length);
464 		dstoffset += 4;
465 		jfs_info("lmWriteRecord: lv offset:%d length:%d",
466 			 lv->offset, lv->length);
467 	}
468 
469 	if ((i = linelock->next)) {
470 		linelock = (struct linelock *) lid_to_tlock(i);
471 		goto moveData;
472 	}
473 
474 	/*
475 	 *      move log record descriptor
476 	 */
477       moveLrd:
478 	lrd->length = cpu_to_le16(len);
479 
480 	src = (caddr_t) lrd;
481 	srclen = LOGRDSIZE;
482 
483 	while (srclen > 0) {
484 		freespace = (LOGPSIZE - LOGPTLRSIZE) - dstoffset;
485 		nbytes = min(freespace, srclen);
486 		dst = (caddr_t) lp + dstoffset;
487 		memcpy(dst, src, nbytes);
488 
489 		dstoffset += nbytes;
490 		srclen -= nbytes;
491 
492 		/* are there more to move than freespace of page ? */
493 		if (srclen)
494 			goto pageFull;
495 
496 		/*
497 		 * end of log record descriptor
498 		 */
499 
500 		/* update last log record eor */
501 		log->eor = dstoffset;
502 		bp->l_eor = dstoffset;
503 		lsn = (log->page << L2LOGPSIZE) + dstoffset;
504 
505 		if (lrd->type & cpu_to_le16(LOG_COMMIT)) {
506 			tblk->clsn = lsn;
507 			jfs_info("wr: tclsn:0x%x, beor:0x%x", tblk->clsn,
508 				 bp->l_eor);
509 
510 			INCREMENT(lmStat.commit);	/* # of commit */
511 
512 			/*
513 			 * enqueue tblock for group commit:
514 			 *
515 			 * enqueue tblock of non-trivial/synchronous COMMIT
516 			 * at tail of group commit queue
517 			 * (trivial/asynchronous COMMITs are ignored by
518 			 * group commit.)
519 			 */
520 			LOGGC_LOCK(log);
521 
522 			/* init tblock gc state */
523 			tblk->flag = tblkGC_QUEUE;
524 			tblk->bp = log->bp;
525 			tblk->pn = log->page;
526 			tblk->eor = log->eor;
527 
528 			/* enqueue transaction to commit queue */
529 			list_add_tail(&tblk->cqueue, &log->cqueue);
530 
531 			LOGGC_UNLOCK(log);
532 		}
533 
534 		jfs_info("lmWriteRecord: lrd:0x%04x bp:0x%p pn:%d eor:0x%x",
535 			le16_to_cpu(lrd->type), log->bp, log->page, dstoffset);
536 
537 		/* page not full ? */
538 		if (dstoffset < LOGPSIZE - LOGPTLRSIZE)
539 			return lsn;
540 
541 	      pageFull:
542 		/* page become full: move on to next page */
543 		lmNextPage(log);
544 
545 		bp = (struct lbuf *) log->bp;
546 		lp = (struct logpage *) bp->l_ldata;
547 		dstoffset = LOGPHDRSIZE;
548 		src += nbytes;
549 	}
550 
551 	return lsn;
552 }
553 
554 
555 /*
556  * NAME:	lmNextPage()
557  *
558  * FUNCTION:	write current page and allocate next page.
559  *
560  * PARAMETER:	log
561  *
562  * RETURN:	0
563  *
564  * serialization: LOG_LOCK() held on entry/exit
565  */
566 static int lmNextPage(struct jfs_log * log)
567 {
568 	struct logpage *lp;
569 	int lspn;		/* log sequence page number */
570 	int pn;			/* current page number */
571 	struct lbuf *bp;
572 	struct lbuf *nextbp;
573 	struct tblock *tblk;
574 
575 	/* get current log page number and log sequence page number */
576 	pn = log->page;
577 	bp = log->bp;
578 	lp = (struct logpage *) bp->l_ldata;
579 	lspn = le32_to_cpu(lp->h.page);
580 
581 	LOGGC_LOCK(log);
582 
583 	/*
584 	 *      write or queue the full page at the tail of write queue
585 	 */
586 	/* get the tail tblk on commit queue */
587 	if (list_empty(&log->cqueue))
588 		tblk = NULL;
589 	else
590 		tblk = list_entry(log->cqueue.prev, struct tblock, cqueue);
591 
592 	/* every tblk who has COMMIT record on the current page,
593 	 * and has not been committed, must be on commit queue
594 	 * since tblk is queued at commit queueu at the time
595 	 * of writing its COMMIT record on the page before
596 	 * page becomes full (even though the tblk thread
597 	 * who wrote COMMIT record may have been suspended
598 	 * currently);
599 	 */
600 
601 	/* is page bound with outstanding tail tblk ? */
602 	if (tblk && tblk->pn == pn) {
603 		/* mark tblk for end-of-page */
604 		tblk->flag |= tblkGC_EOP;
605 
606 		if (log->cflag & logGC_PAGEOUT) {
607 			/* if page is not already on write queue,
608 			 * just enqueue (no lbmWRITE to prevent redrive)
609 			 * buffer to wqueue to ensure correct serial order
610 			 * of the pages since log pages will be added
611 			 * continuously
612 			 */
613 			if (bp->l_wqnext == NULL)
614 				lbmWrite(log, bp, 0, 0);
615 		} else {
616 			/*
617 			 * No current GC leader, initiate group commit
618 			 */
619 			log->cflag |= logGC_PAGEOUT;
620 			lmGCwrite(log, 0);
621 		}
622 	}
623 	/* page is not bound with outstanding tblk:
624 	 * init write or mark it to be redriven (lbmWRITE)
625 	 */
626 	else {
627 		/* finalize the page */
628 		bp->l_ceor = bp->l_eor;
629 		lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_ceor);
630 		lbmWrite(log, bp, lbmWRITE | lbmRELEASE | lbmFREE, 0);
631 	}
632 	LOGGC_UNLOCK(log);
633 
634 	/*
635 	 *      allocate/initialize next page
636 	 */
637 	/* if log wraps, the first data page of log is 2
638 	 * (0 never used, 1 is superblock).
639 	 */
640 	log->page = (pn == log->size - 1) ? 2 : pn + 1;
641 	log->eor = LOGPHDRSIZE;	/* ? valid page empty/full at logRedo() */
642 
643 	/* allocate/initialize next log page buffer */
644 	nextbp = lbmAllocate(log, log->page);
645 	nextbp->l_eor = log->eor;
646 	log->bp = nextbp;
647 
648 	/* initialize next log page */
649 	lp = (struct logpage *) nextbp->l_ldata;
650 	lp->h.page = lp->t.page = cpu_to_le32(lspn + 1);
651 	lp->h.eor = lp->t.eor = cpu_to_le16(LOGPHDRSIZE);
652 
653 	return 0;
654 }
655 
656 
657 /*
658  * NAME:	lmGroupCommit()
659  *
660  * FUNCTION:	group commit
661  *	initiate pageout of the pages with COMMIT in the order of
662  *	page number - redrive pageout of the page at the head of
663  *	pageout queue until full page has been written.
664  *
665  * RETURN:
666  *
667  * NOTE:
668  *	LOGGC_LOCK serializes log group commit queue, and
669  *	transaction blocks on the commit queue.
670  *	N.B. LOG_LOCK is NOT held during lmGroupCommit().
671  */
672 int lmGroupCommit(struct jfs_log * log, struct tblock * tblk)
673 {
674 	int rc = 0;
675 
676 	LOGGC_LOCK(log);
677 
678 	/* group committed already ? */
679 	if (tblk->flag & tblkGC_COMMITTED) {
680 		if (tblk->flag & tblkGC_ERROR)
681 			rc = -EIO;
682 
683 		LOGGC_UNLOCK(log);
684 		return rc;
685 	}
686 	jfs_info("lmGroup Commit: tblk = 0x%p, gcrtc = %d", tblk, log->gcrtc);
687 
688 	if (tblk->xflag & COMMIT_LAZY)
689 		tblk->flag |= tblkGC_LAZY;
690 
691 	if ((!(log->cflag & logGC_PAGEOUT)) && (!list_empty(&log->cqueue)) &&
692 	    (!(tblk->xflag & COMMIT_LAZY) || test_bit(log_FLUSH, &log->flag)
693 	     || jfs_tlocks_low)) {
694 		/*
695 		 * No pageout in progress
696 		 *
697 		 * start group commit as its group leader.
698 		 */
699 		log->cflag |= logGC_PAGEOUT;
700 
701 		lmGCwrite(log, 0);
702 	}
703 
704 	if (tblk->xflag & COMMIT_LAZY) {
705 		/*
706 		 * Lazy transactions can leave now
707 		 */
708 		LOGGC_UNLOCK(log);
709 		return 0;
710 	}
711 
712 	/* lmGCwrite gives up LOGGC_LOCK, check again */
713 
714 	if (tblk->flag & tblkGC_COMMITTED) {
715 		if (tblk->flag & tblkGC_ERROR)
716 			rc = -EIO;
717 
718 		LOGGC_UNLOCK(log);
719 		return rc;
720 	}
721 
722 	/* upcount transaction waiting for completion
723 	 */
724 	log->gcrtc++;
725 	tblk->flag |= tblkGC_READY;
726 
727 	__SLEEP_COND(tblk->gcwait, (tblk->flag & tblkGC_COMMITTED),
728 		     LOGGC_LOCK(log), LOGGC_UNLOCK(log));
729 
730 	/* removed from commit queue */
731 	if (tblk->flag & tblkGC_ERROR)
732 		rc = -EIO;
733 
734 	LOGGC_UNLOCK(log);
735 	return rc;
736 }
737 
738 /*
739  * NAME:	lmGCwrite()
740  *
741  * FUNCTION:	group commit write
742  *	initiate write of log page, building a group of all transactions
743  *	with commit records on that page.
744  *
745  * RETURN:	None
746  *
747  * NOTE:
748  *	LOGGC_LOCK must be held by caller.
749  *	N.B. LOG_LOCK is NOT held during lmGroupCommit().
750  */
751 static void lmGCwrite(struct jfs_log * log, int cant_write)
752 {
753 	struct lbuf *bp;
754 	struct logpage *lp;
755 	int gcpn;		/* group commit page number */
756 	struct tblock *tblk;
757 	struct tblock *xtblk = NULL;
758 
759 	/*
760 	 * build the commit group of a log page
761 	 *
762 	 * scan commit queue and make a commit group of all
763 	 * transactions with COMMIT records on the same log page.
764 	 */
765 	/* get the head tblk on the commit queue */
766 	gcpn = list_entry(log->cqueue.next, struct tblock, cqueue)->pn;
767 
768 	list_for_each_entry(tblk, &log->cqueue, cqueue) {
769 		if (tblk->pn != gcpn)
770 			break;
771 
772 		xtblk = tblk;
773 
774 		/* state transition: (QUEUE, READY) -> COMMIT */
775 		tblk->flag |= tblkGC_COMMIT;
776 	}
777 	tblk = xtblk;		/* last tblk of the page */
778 
779 	/*
780 	 * pageout to commit transactions on the log page.
781 	 */
782 	bp = (struct lbuf *) tblk->bp;
783 	lp = (struct logpage *) bp->l_ldata;
784 	/* is page already full ? */
785 	if (tblk->flag & tblkGC_EOP) {
786 		/* mark page to free at end of group commit of the page */
787 		tblk->flag &= ~tblkGC_EOP;
788 		tblk->flag |= tblkGC_FREE;
789 		bp->l_ceor = bp->l_eor;
790 		lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_ceor);
791 		lbmWrite(log, bp, lbmWRITE | lbmRELEASE | lbmGC,
792 			 cant_write);
793 		INCREMENT(lmStat.full_page);
794 	}
795 	/* page is not yet full */
796 	else {
797 		bp->l_ceor = tblk->eor;	/* ? bp->l_ceor = bp->l_eor; */
798 		lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_ceor);
799 		lbmWrite(log, bp, lbmWRITE | lbmGC, cant_write);
800 		INCREMENT(lmStat.partial_page);
801 	}
802 }
803 
804 /*
805  * NAME:	lmPostGC()
806  *
807  * FUNCTION:	group commit post-processing
808  *	Processes transactions after their commit records have been written
809  *	to disk, redriving log I/O if necessary.
810  *
811  * RETURN:	None
812  *
813  * NOTE:
814  *	This routine is called a interrupt time by lbmIODone
815  */
816 static void lmPostGC(struct lbuf * bp)
817 {
818 	unsigned long flags;
819 	struct jfs_log *log = bp->l_log;
820 	struct logpage *lp;
821 	struct tblock *tblk, *temp;
822 
823 	//LOGGC_LOCK(log);
824 	spin_lock_irqsave(&log->gclock, flags);
825 	/*
826 	 * current pageout of group commit completed.
827 	 *
828 	 * remove/wakeup transactions from commit queue who were
829 	 * group committed with the current log page
830 	 */
831 	list_for_each_entry_safe(tblk, temp, &log->cqueue, cqueue) {
832 		if (!(tblk->flag & tblkGC_COMMIT))
833 			break;
834 		/* if transaction was marked GC_COMMIT then
835 		 * it has been shipped in the current pageout
836 		 * and made it to disk - it is committed.
837 		 */
838 
839 		if (bp->l_flag & lbmERROR)
840 			tblk->flag |= tblkGC_ERROR;
841 
842 		/* remove it from the commit queue */
843 		list_del(&tblk->cqueue);
844 		tblk->flag &= ~tblkGC_QUEUE;
845 
846 		if (tblk == log->flush_tblk) {
847 			/* we can stop flushing the log now */
848 			clear_bit(log_FLUSH, &log->flag);
849 			log->flush_tblk = NULL;
850 		}
851 
852 		jfs_info("lmPostGC: tblk = 0x%p, flag = 0x%x", tblk,
853 			 tblk->flag);
854 
855 		if (!(tblk->xflag & COMMIT_FORCE))
856 			/*
857 			 * Hand tblk over to lazy commit thread
858 			 */
859 			txLazyUnlock(tblk);
860 		else {
861 			/* state transition: COMMIT -> COMMITTED */
862 			tblk->flag |= tblkGC_COMMITTED;
863 
864 			if (tblk->flag & tblkGC_READY)
865 				log->gcrtc--;
866 
867 			LOGGC_WAKEUP(tblk);
868 		}
869 
870 		/* was page full before pageout ?
871 		 * (and this is the last tblk bound with the page)
872 		 */
873 		if (tblk->flag & tblkGC_FREE)
874 			lbmFree(bp);
875 		/* did page become full after pageout ?
876 		 * (and this is the last tblk bound with the page)
877 		 */
878 		else if (tblk->flag & tblkGC_EOP) {
879 			/* finalize the page */
880 			lp = (struct logpage *) bp->l_ldata;
881 			bp->l_ceor = bp->l_eor;
882 			lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_eor);
883 			jfs_info("lmPostGC: calling lbmWrite");
884 			lbmWrite(log, bp, lbmWRITE | lbmRELEASE | lbmFREE,
885 				 1);
886 		}
887 
888 	}
889 
890 	/* are there any transactions who have entered lnGroupCommit()
891 	 * (whose COMMITs are after that of the last log page written.
892 	 * They are waiting for new group commit (above at (SLEEP 1))
893 	 * or lazy transactions are on a full (queued) log page,
894 	 * select the latest ready transaction as new group leader and
895 	 * wake her up to lead her group.
896 	 */
897 	if ((!list_empty(&log->cqueue)) &&
898 	    ((log->gcrtc > 0) || (tblk->bp->l_wqnext != NULL) ||
899 	     test_bit(log_FLUSH, &log->flag) || jfs_tlocks_low))
900 		/*
901 		 * Call lmGCwrite with new group leader
902 		 */
903 		lmGCwrite(log, 1);
904 
905 	/* no transaction are ready yet (transactions are only just
906 	 * queued (GC_QUEUE) and not entered for group commit yet).
907 	 * the first transaction entering group commit
908 	 * will elect herself as new group leader.
909 	 */
910 	else
911 		log->cflag &= ~logGC_PAGEOUT;
912 
913 	//LOGGC_UNLOCK(log);
914 	spin_unlock_irqrestore(&log->gclock, flags);
915 	return;
916 }
917 
918 /*
919  * NAME:	lmLogSync()
920  *
921  * FUNCTION:	write log SYNCPT record for specified log
922  *	if new sync address is available
923  *	(normally the case if sync() is executed by back-ground
924  *	process).
925  *	if not, explicitly run jfs_blogsync() to initiate
926  *	getting of new sync address.
927  *	calculate new value of i_nextsync which determines when
928  *	this code is called again.
929  *
930  *	this is called only from lmLog().
931  *
932  * PARAMETER:	ip	- pointer to logs inode.
933  *
934  * RETURN:	0
935  *
936  * serialization: LOG_LOCK() held on entry/exit
937  */
938 static int lmLogSync(struct jfs_log * log, int nosyncwait)
939 {
940 	int logsize;
941 	int written;		/* written since last syncpt */
942 	int free;		/* free space left available */
943 	int delta;		/* additional delta to write normally */
944 	int more;		/* additional write granted */
945 	struct lrd lrd;
946 	int lsn;
947 	struct logsyncblk *lp;
948 
949 	/*
950 	 *      forward syncpt
951 	 */
952 	/* if last sync is same as last syncpt,
953 	 * invoke sync point forward processing to update sync.
954 	 */
955 
956 	if (log->sync == log->syncpt) {
957 		LOGSYNC_LOCK(log);
958 		/* ToDo: push dirty metapages out to disk */
959 //              bmLogSync(log);
960 
961 		if (list_empty(&log->synclist))
962 			log->sync = log->lsn;
963 		else {
964 			lp = list_entry(log->synclist.next,
965 					struct logsyncblk, synclist);
966 			log->sync = lp->lsn;
967 		}
968 		LOGSYNC_UNLOCK(log);
969 
970 	}
971 
972 	/* if sync is different from last syncpt,
973 	 * write a SYNCPT record with syncpt = sync.
974 	 * reset syncpt = sync
975 	 */
976 	if (log->sync != log->syncpt) {
977 		struct jfs_sb_info *sbi;
978 
979 		/*
980 		 * We need to make sure all of the "written" metapages
981 		 * actually make it to disk
982 		 */
983 		list_for_each_entry(sbi, &log->sb_list, log_list) {
984 			if (sbi->flag & JFS_NOINTEGRITY)
985 				continue;
986 			filemap_fdatawrite(sbi->ipbmap->i_mapping);
987 			filemap_fdatawrite(sbi->ipimap->i_mapping);
988 			filemap_fdatawrite(sbi->sb->s_bdev->bd_inode->i_mapping);
989 		}
990 		list_for_each_entry(sbi, &log->sb_list, log_list) {
991 			if (sbi->flag & JFS_NOINTEGRITY)
992 				continue;
993 			filemap_fdatawait(sbi->ipbmap->i_mapping);
994 			filemap_fdatawait(sbi->ipimap->i_mapping);
995 			filemap_fdatawait(sbi->sb->s_bdev->bd_inode->i_mapping);
996 		}
997 
998 		lrd.logtid = 0;
999 		lrd.backchain = 0;
1000 		lrd.type = cpu_to_le16(LOG_SYNCPT);
1001 		lrd.length = 0;
1002 		lrd.log.syncpt.sync = cpu_to_le32(log->sync);
1003 		lsn = lmWriteRecord(log, NULL, &lrd, NULL);
1004 
1005 		log->syncpt = log->sync;
1006 	} else
1007 		lsn = log->lsn;
1008 
1009 	/*
1010 	 *      setup next syncpt trigger (SWAG)
1011 	 */
1012 	logsize = log->logsize;
1013 
1014 	logdiff(written, lsn, log);
1015 	free = logsize - written;
1016 	delta = LOGSYNC_DELTA(logsize);
1017 	more = min(free / 2, delta);
1018 	if (more < 2 * LOGPSIZE) {
1019 		jfs_warn("\n ... Log Wrap ... Log Wrap ... Log Wrap ...\n");
1020 		/*
1021 		 *      log wrapping
1022 		 *
1023 		 * option 1 - panic ? No.!
1024 		 * option 2 - shutdown file systems
1025 		 *            associated with log ?
1026 		 * option 3 - extend log ?
1027 		 */
1028 		/*
1029 		 * option 4 - second chance
1030 		 *
1031 		 * mark log wrapped, and continue.
1032 		 * when all active transactions are completed,
1033 		 * mark log vaild for recovery.
1034 		 * if crashed during invalid state, log state
1035 		 * implies invald log, forcing fsck().
1036 		 */
1037 		/* mark log state log wrap in log superblock */
1038 		/* log->state = LOGWRAP; */
1039 
1040 		/* reset sync point computation */
1041 		log->syncpt = log->sync = lsn;
1042 		log->nextsync = delta;
1043 	} else
1044 		/* next syncpt trigger = written + more */
1045 		log->nextsync = written + more;
1046 
1047 	/* return if lmLogSync() from outside of transaction, e.g., sync() */
1048 	if (nosyncwait)
1049 		return lsn;
1050 
1051 	/* if number of bytes written from last sync point is more
1052 	 * than 1/4 of the log size, stop new transactions from
1053 	 * starting until all current transactions are completed
1054 	 * by setting syncbarrier flag.
1055 	 */
1056 	if (written > LOGSYNC_BARRIER(logsize) && logsize > 32 * LOGPSIZE) {
1057 		set_bit(log_SYNCBARRIER, &log->flag);
1058 		jfs_info("log barrier on: lsn=0x%x syncpt=0x%x", lsn,
1059 			 log->syncpt);
1060 		/*
1061 		 * We may have to initiate group commit
1062 		 */
1063 		jfs_flush_journal(log, 0);
1064 	}
1065 
1066 	return lsn;
1067 }
1068 
1069 
1070 /*
1071  * NAME:	lmLogOpen()
1072  *
1073  * FUNCTION:    open the log on first open;
1074  *	insert filesystem in the active list of the log.
1075  *
1076  * PARAMETER:	ipmnt	- file system mount inode
1077  *		iplog 	- log inode (out)
1078  *
1079  * RETURN:
1080  *
1081  * serialization:
1082  */
1083 int lmLogOpen(struct super_block *sb)
1084 {
1085 	int rc;
1086 	struct block_device *bdev;
1087 	struct jfs_log *log;
1088 	struct jfs_sb_info *sbi = JFS_SBI(sb);
1089 
1090 	if (sbi->flag & JFS_NOINTEGRITY)
1091 		return open_dummy_log(sb);
1092 
1093 	if (sbi->mntflag & JFS_INLINELOG)
1094 		return open_inline_log(sb);
1095 
1096 	down(&jfs_log_sem);
1097 	list_for_each_entry(log, &jfs_external_logs, journal_list) {
1098 		if (log->bdev->bd_dev == sbi->logdev) {
1099 			if (memcmp(log->uuid, sbi->loguuid,
1100 				   sizeof(log->uuid))) {
1101 				jfs_warn("wrong uuid on JFS journal\n");
1102 				up(&jfs_log_sem);
1103 				return -EINVAL;
1104 			}
1105 			/*
1106 			 * add file system to log active file system list
1107 			 */
1108 			if ((rc = lmLogFileSystem(log, sbi, 1))) {
1109 				up(&jfs_log_sem);
1110 				return rc;
1111 			}
1112 			goto journal_found;
1113 		}
1114 	}
1115 
1116 	if (!(log = kmalloc(sizeof(struct jfs_log), GFP_KERNEL))) {
1117 		up(&jfs_log_sem);
1118 		return -ENOMEM;
1119 	}
1120 	memset(log, 0, sizeof(struct jfs_log));
1121 	INIT_LIST_HEAD(&log->sb_list);
1122 	init_waitqueue_head(&log->syncwait);
1123 
1124 	/*
1125 	 *      external log as separate logical volume
1126 	 *
1127 	 * file systems to log may have n-to-1 relationship;
1128 	 */
1129 
1130 	bdev = open_by_devnum(sbi->logdev, FMODE_READ|FMODE_WRITE);
1131 	if (IS_ERR(bdev)) {
1132 		rc = -PTR_ERR(bdev);
1133 		goto free;
1134 	}
1135 
1136 	if ((rc = bd_claim(bdev, log))) {
1137 		goto close;
1138 	}
1139 
1140 	log->bdev = bdev;
1141 	memcpy(log->uuid, sbi->loguuid, sizeof(log->uuid));
1142 
1143 	/*
1144 	 * initialize log:
1145 	 */
1146 	if ((rc = lmLogInit(log)))
1147 		goto unclaim;
1148 
1149 	list_add(&log->journal_list, &jfs_external_logs);
1150 
1151 	/*
1152 	 * add file system to log active file system list
1153 	 */
1154 	if ((rc = lmLogFileSystem(log, sbi, 1)))
1155 		goto shutdown;
1156 
1157 journal_found:
1158 	LOG_LOCK(log);
1159 	list_add(&sbi->log_list, &log->sb_list);
1160 	sbi->log = log;
1161 	LOG_UNLOCK(log);
1162 
1163 	up(&jfs_log_sem);
1164 	return 0;
1165 
1166 	/*
1167 	 *      unwind on error
1168 	 */
1169       shutdown:		/* unwind lbmLogInit() */
1170 	list_del(&log->journal_list);
1171 	lbmLogShutdown(log);
1172 
1173       unclaim:
1174 	bd_release(bdev);
1175 
1176       close:		/* close external log device */
1177 	blkdev_put(bdev);
1178 
1179       free:		/* free log descriptor */
1180 	up(&jfs_log_sem);
1181 	kfree(log);
1182 
1183 	jfs_warn("lmLogOpen: exit(%d)", rc);
1184 	return rc;
1185 }
1186 
1187 static int open_inline_log(struct super_block *sb)
1188 {
1189 	struct jfs_log *log;
1190 	int rc;
1191 
1192 	if (!(log = kmalloc(sizeof(struct jfs_log), GFP_KERNEL)))
1193 		return -ENOMEM;
1194 	memset(log, 0, sizeof(struct jfs_log));
1195 	INIT_LIST_HEAD(&log->sb_list);
1196 	init_waitqueue_head(&log->syncwait);
1197 
1198 	set_bit(log_INLINELOG, &log->flag);
1199 	log->bdev = sb->s_bdev;
1200 	log->base = addressPXD(&JFS_SBI(sb)->logpxd);
1201 	log->size = lengthPXD(&JFS_SBI(sb)->logpxd) >>
1202 	    (L2LOGPSIZE - sb->s_blocksize_bits);
1203 	log->l2bsize = sb->s_blocksize_bits;
1204 	ASSERT(L2LOGPSIZE >= sb->s_blocksize_bits);
1205 
1206 	/*
1207 	 * initialize log.
1208 	 */
1209 	if ((rc = lmLogInit(log))) {
1210 		kfree(log);
1211 		jfs_warn("lmLogOpen: exit(%d)", rc);
1212 		return rc;
1213 	}
1214 
1215 	list_add(&JFS_SBI(sb)->log_list, &log->sb_list);
1216 	JFS_SBI(sb)->log = log;
1217 
1218 	return rc;
1219 }
1220 
1221 static int open_dummy_log(struct super_block *sb)
1222 {
1223 	int rc;
1224 
1225 	down(&jfs_log_sem);
1226 	if (!dummy_log) {
1227 		dummy_log = kmalloc(sizeof(struct jfs_log), GFP_KERNEL);
1228 		if (!dummy_log) {
1229 			up(&jfs_log_sem);
1230 			return -ENOMEM;
1231 		}
1232 		memset(dummy_log, 0, sizeof(struct jfs_log));
1233 		INIT_LIST_HEAD(&dummy_log->sb_list);
1234 		init_waitqueue_head(&dummy_log->syncwait);
1235 		dummy_log->no_integrity = 1;
1236 		/* Make up some stuff */
1237 		dummy_log->base = 0;
1238 		dummy_log->size = 1024;
1239 		rc = lmLogInit(dummy_log);
1240 		if (rc) {
1241 			kfree(dummy_log);
1242 			dummy_log = NULL;
1243 			up(&jfs_log_sem);
1244 			return rc;
1245 		}
1246 	}
1247 
1248 	LOG_LOCK(dummy_log);
1249 	list_add(&JFS_SBI(sb)->log_list, &dummy_log->sb_list);
1250 	JFS_SBI(sb)->log = dummy_log;
1251 	LOG_UNLOCK(dummy_log);
1252 	up(&jfs_log_sem);
1253 
1254 	return 0;
1255 }
1256 
1257 /*
1258  * NAME:	lmLogInit()
1259  *
1260  * FUNCTION:	log initialization at first log open.
1261  *
1262  *	logredo() (or logformat()) should have been run previously.
1263  *	initialize the log from log superblock.
1264  *	set the log state in the superblock to LOGMOUNT and
1265  *	write SYNCPT log record.
1266  *
1267  * PARAMETER:	log	- log structure
1268  *
1269  * RETURN:	0	- if ok
1270  *		-EINVAL	- bad log magic number or superblock dirty
1271  *		error returned from logwait()
1272  *
1273  * serialization: single first open thread
1274  */
1275 int lmLogInit(struct jfs_log * log)
1276 {
1277 	int rc = 0;
1278 	struct lrd lrd;
1279 	struct logsuper *logsuper;
1280 	struct lbuf *bpsuper;
1281 	struct lbuf *bp;
1282 	struct logpage *lp;
1283 	int lsn = 0;
1284 
1285 	jfs_info("lmLogInit: log:0x%p", log);
1286 
1287 	/* initialize the group commit serialization lock */
1288 	LOGGC_LOCK_INIT(log);
1289 
1290 	/* allocate/initialize the log write serialization lock */
1291 	LOG_LOCK_INIT(log);
1292 
1293 	LOGSYNC_LOCK_INIT(log);
1294 
1295 	INIT_LIST_HEAD(&log->synclist);
1296 
1297 	INIT_LIST_HEAD(&log->cqueue);
1298 	log->flush_tblk = NULL;
1299 
1300 	log->count = 0;
1301 
1302 	/*
1303 	 * initialize log i/o
1304 	 */
1305 	if ((rc = lbmLogInit(log)))
1306 		return rc;
1307 
1308 	if (!test_bit(log_INLINELOG, &log->flag))
1309 		log->l2bsize = L2LOGPSIZE;
1310 
1311 	/* check for disabled journaling to disk */
1312 	if (log->no_integrity) {
1313 		/*
1314 		 * Journal pages will still be filled.  When the time comes
1315 		 * to actually do the I/O, the write is not done, and the
1316 		 * endio routine is called directly.
1317 		 */
1318 		bp = lbmAllocate(log , 0);
1319 		log->bp = bp;
1320 		bp->l_pn = bp->l_eor = 0;
1321 	} else {
1322 		/*
1323 		 * validate log superblock
1324 		 */
1325 		if ((rc = lbmRead(log, 1, &bpsuper)))
1326 			goto errout10;
1327 
1328 		logsuper = (struct logsuper *) bpsuper->l_ldata;
1329 
1330 		if (logsuper->magic != cpu_to_le32(LOGMAGIC)) {
1331 			jfs_warn("*** Log Format Error ! ***");
1332 			rc = -EINVAL;
1333 			goto errout20;
1334 		}
1335 
1336 		/* logredo() should have been run successfully. */
1337 		if (logsuper->state != cpu_to_le32(LOGREDONE)) {
1338 			jfs_warn("*** Log Is Dirty ! ***");
1339 			rc = -EINVAL;
1340 			goto errout20;
1341 		}
1342 
1343 		/* initialize log from log superblock */
1344 		if (test_bit(log_INLINELOG,&log->flag)) {
1345 			if (log->size != le32_to_cpu(logsuper->size)) {
1346 				rc = -EINVAL;
1347 				goto errout20;
1348 			}
1349 			jfs_info("lmLogInit: inline log:0x%p base:0x%Lx "
1350 				 "size:0x%x", log,
1351 				 (unsigned long long) log->base, log->size);
1352 		} else {
1353 			if (memcmp(logsuper->uuid, log->uuid, 16)) {
1354 				jfs_warn("wrong uuid on JFS log device");
1355 				goto errout20;
1356 			}
1357 			log->size = le32_to_cpu(logsuper->size);
1358 			log->l2bsize = le32_to_cpu(logsuper->l2bsize);
1359 			jfs_info("lmLogInit: external log:0x%p base:0x%Lx "
1360 				 "size:0x%x", log,
1361 				 (unsigned long long) log->base, log->size);
1362 		}
1363 
1364 		log->page = le32_to_cpu(logsuper->end) / LOGPSIZE;
1365 		log->eor = le32_to_cpu(logsuper->end) - (LOGPSIZE * log->page);
1366 
1367 		/*
1368 		 * initialize for log append write mode
1369 		 */
1370 		/* establish current/end-of-log page/buffer */
1371 		if ((rc = lbmRead(log, log->page, &bp)))
1372 			goto errout20;
1373 
1374 		lp = (struct logpage *) bp->l_ldata;
1375 
1376 		jfs_info("lmLogInit: lsn:0x%x page:%d eor:%d:%d",
1377 			 le32_to_cpu(logsuper->end), log->page, log->eor,
1378 			 le16_to_cpu(lp->h.eor));
1379 
1380 		log->bp = bp;
1381 		bp->l_pn = log->page;
1382 		bp->l_eor = log->eor;
1383 
1384 		/* if current page is full, move on to next page */
1385 		if (log->eor >= LOGPSIZE - LOGPTLRSIZE)
1386 			lmNextPage(log);
1387 
1388 		/*
1389 		 * initialize log syncpoint
1390 		 */
1391 		/*
1392 		 * write the first SYNCPT record with syncpoint = 0
1393 		 * (i.e., log redo up to HERE !);
1394 		 * remove current page from lbm write queue at end of pageout
1395 		 * (to write log superblock update), but do not release to
1396 		 * freelist;
1397 		 */
1398 		lrd.logtid = 0;
1399 		lrd.backchain = 0;
1400 		lrd.type = cpu_to_le16(LOG_SYNCPT);
1401 		lrd.length = 0;
1402 		lrd.log.syncpt.sync = 0;
1403 		lsn = lmWriteRecord(log, NULL, &lrd, NULL);
1404 		bp = log->bp;
1405 		bp->l_ceor = bp->l_eor;
1406 		lp = (struct logpage *) bp->l_ldata;
1407 		lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_eor);
1408 		lbmWrite(log, bp, lbmWRITE | lbmSYNC, 0);
1409 		if ((rc = lbmIOWait(bp, 0)))
1410 			goto errout30;
1411 
1412 		/*
1413 		 * update/write superblock
1414 		 */
1415 		logsuper->state = cpu_to_le32(LOGMOUNT);
1416 		log->serial = le32_to_cpu(logsuper->serial) + 1;
1417 		logsuper->serial = cpu_to_le32(log->serial);
1418 		lbmDirectWrite(log, bpsuper, lbmWRITE | lbmRELEASE | lbmSYNC);
1419 		if ((rc = lbmIOWait(bpsuper, lbmFREE)))
1420 			goto errout30;
1421 	}
1422 
1423 	/* initialize logsync parameters */
1424 	log->logsize = (log->size - 2) << L2LOGPSIZE;
1425 	log->lsn = lsn;
1426 	log->syncpt = lsn;
1427 	log->sync = log->syncpt;
1428 	log->nextsync = LOGSYNC_DELTA(log->logsize);
1429 
1430 	jfs_info("lmLogInit: lsn:0x%x syncpt:0x%x sync:0x%x",
1431 		 log->lsn, log->syncpt, log->sync);
1432 
1433 	/*
1434 	 * initialize for lazy/group commit
1435 	 */
1436 	log->clsn = lsn;
1437 
1438 	return 0;
1439 
1440 	/*
1441 	 *      unwind on error
1442 	 */
1443       errout30:		/* release log page */
1444 	log->wqueue = NULL;
1445 	bp->l_wqnext = NULL;
1446 	lbmFree(bp);
1447 
1448       errout20:		/* release log superblock */
1449 	lbmFree(bpsuper);
1450 
1451       errout10:		/* unwind lbmLogInit() */
1452 	lbmLogShutdown(log);
1453 
1454 	jfs_warn("lmLogInit: exit(%d)", rc);
1455 	return rc;
1456 }
1457 
1458 
1459 /*
1460  * NAME:	lmLogClose()
1461  *
1462  * FUNCTION:	remove file system <ipmnt> from active list of log <iplog>
1463  *		and close it on last close.
1464  *
1465  * PARAMETER:	sb	- superblock
1466  *
1467  * RETURN:	errors from subroutines
1468  *
1469  * serialization:
1470  */
1471 int lmLogClose(struct super_block *sb)
1472 {
1473 	struct jfs_sb_info *sbi = JFS_SBI(sb);
1474 	struct jfs_log *log = sbi->log;
1475 	struct block_device *bdev;
1476 	int rc = 0;
1477 
1478 	jfs_info("lmLogClose: log:0x%p", log);
1479 
1480 	down(&jfs_log_sem);
1481 	LOG_LOCK(log);
1482 	list_del(&sbi->log_list);
1483 	LOG_UNLOCK(log);
1484 	sbi->log = NULL;
1485 
1486 	/*
1487 	 * We need to make sure all of the "written" metapages
1488 	 * actually make it to disk
1489 	 */
1490 	sync_blockdev(sb->s_bdev);
1491 
1492 	if (test_bit(log_INLINELOG, &log->flag)) {
1493 		/*
1494 		 *      in-line log in host file system
1495 		 */
1496 		rc = lmLogShutdown(log);
1497 		kfree(log);
1498 		goto out;
1499 	}
1500 
1501 	if (!log->no_integrity)
1502 		lmLogFileSystem(log, sbi, 0);
1503 
1504 	if (!list_empty(&log->sb_list))
1505 		goto out;
1506 
1507 	/*
1508 	 * TODO: ensure that the dummy_log is in a state to allow
1509 	 * lbmLogShutdown to deallocate all the buffers and call
1510 	 * kfree against dummy_log.  For now, leave dummy_log & its
1511 	 * buffers in memory, and resuse if another no-integrity mount
1512 	 * is requested.
1513 	 */
1514 	if (log->no_integrity)
1515 		goto out;
1516 
1517 	/*
1518 	 *      external log as separate logical volume
1519 	 */
1520 	list_del(&log->journal_list);
1521 	bdev = log->bdev;
1522 	rc = lmLogShutdown(log);
1523 
1524 	bd_release(bdev);
1525 	blkdev_put(bdev);
1526 
1527 	kfree(log);
1528 
1529       out:
1530 	up(&jfs_log_sem);
1531 	jfs_info("lmLogClose: exit(%d)", rc);
1532 	return rc;
1533 }
1534 
1535 
1536 /*
1537  * NAME:	jfs_flush_journal()
1538  *
1539  * FUNCTION:	initiate write of any outstanding transactions to the journal
1540  *		and optionally wait until they are all written to disk
1541  *
1542  *		wait == 0  flush until latest txn is committed, don't wait
1543  *		wait == 1  flush until latest txn is committed, wait
1544  *		wait > 1   flush until all txn's are complete, wait
1545  */
1546 void jfs_flush_journal(struct jfs_log *log, int wait)
1547 {
1548 	int i;
1549 	struct tblock *target = NULL;
1550 
1551 	/* jfs_write_inode may call us during read-only mount */
1552 	if (!log)
1553 		return;
1554 
1555 	jfs_info("jfs_flush_journal: log:0x%p wait=%d", log, wait);
1556 
1557 	LOGGC_LOCK(log);
1558 
1559 	if (!list_empty(&log->cqueue)) {
1560 		/*
1561 		 * This ensures that we will keep writing to the journal as long
1562 		 * as there are unwritten commit records
1563 		 */
1564 		target = list_entry(log->cqueue.prev, struct tblock, cqueue);
1565 
1566 		if (test_bit(log_FLUSH, &log->flag)) {
1567 			/*
1568 			 * We're already flushing.
1569 			 * if flush_tblk is NULL, we are flushing everything,
1570 			 * so leave it that way.  Otherwise, update it to the
1571 			 * latest transaction
1572 			 */
1573 			if (log->flush_tblk)
1574 				log->flush_tblk = target;
1575 		} else {
1576 			/* Only flush until latest transaction is committed */
1577 			log->flush_tblk = target;
1578 			set_bit(log_FLUSH, &log->flag);
1579 
1580 			/*
1581 			 * Initiate I/O on outstanding transactions
1582 			 */
1583 			if (!(log->cflag & logGC_PAGEOUT)) {
1584 				log->cflag |= logGC_PAGEOUT;
1585 				lmGCwrite(log, 0);
1586 			}
1587 		}
1588 	}
1589 	if ((wait > 1) || test_bit(log_SYNCBARRIER, &log->flag)) {
1590 		/* Flush until all activity complete */
1591 		set_bit(log_FLUSH, &log->flag);
1592 		log->flush_tblk = NULL;
1593 	}
1594 
1595 	if (wait && target && !(target->flag & tblkGC_COMMITTED)) {
1596 		DECLARE_WAITQUEUE(__wait, current);
1597 
1598 		add_wait_queue(&target->gcwait, &__wait);
1599 		set_current_state(TASK_UNINTERRUPTIBLE);
1600 		LOGGC_UNLOCK(log);
1601 		schedule();
1602 		current->state = TASK_RUNNING;
1603 		LOGGC_LOCK(log);
1604 		remove_wait_queue(&target->gcwait, &__wait);
1605 	}
1606 	LOGGC_UNLOCK(log);
1607 
1608 	if (wait < 2)
1609 		return;
1610 
1611 	/*
1612 	 * If there was recent activity, we may need to wait
1613 	 * for the lazycommit thread to catch up
1614 	 */
1615 	if ((!list_empty(&log->cqueue)) || !list_empty(&log->synclist)) {
1616 		for (i = 0; i < 800; i++) {	/* Too much? */
1617 			msleep(250);
1618 			if (list_empty(&log->cqueue) &&
1619 			    list_empty(&log->synclist))
1620 				break;
1621 		}
1622 	}
1623 	assert(list_empty(&log->cqueue));
1624 	assert(list_empty(&log->synclist));
1625 	clear_bit(log_FLUSH, &log->flag);
1626 }
1627 
1628 /*
1629  * NAME:	lmLogShutdown()
1630  *
1631  * FUNCTION:	log shutdown at last LogClose().
1632  *
1633  *		write log syncpt record.
1634  *		update super block to set redone flag to 0.
1635  *
1636  * PARAMETER:	log	- log inode
1637  *
1638  * RETURN:	0	- success
1639  *
1640  * serialization: single last close thread
1641  */
1642 int lmLogShutdown(struct jfs_log * log)
1643 {
1644 	int rc;
1645 	struct lrd lrd;
1646 	int lsn;
1647 	struct logsuper *logsuper;
1648 	struct lbuf *bpsuper;
1649 	struct lbuf *bp;
1650 	struct logpage *lp;
1651 
1652 	jfs_info("lmLogShutdown: log:0x%p", log);
1653 
1654 	jfs_flush_journal(log, 2);
1655 
1656 	/*
1657 	 * write the last SYNCPT record with syncpoint = 0
1658 	 * (i.e., log redo up to HERE !)
1659 	 */
1660 	lrd.logtid = 0;
1661 	lrd.backchain = 0;
1662 	lrd.type = cpu_to_le16(LOG_SYNCPT);
1663 	lrd.length = 0;
1664 	lrd.log.syncpt.sync = 0;
1665 
1666 	lsn = lmWriteRecord(log, NULL, &lrd, NULL);
1667 	bp = log->bp;
1668 	lp = (struct logpage *) bp->l_ldata;
1669 	lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_eor);
1670 	lbmWrite(log, log->bp, lbmWRITE | lbmRELEASE | lbmSYNC, 0);
1671 	lbmIOWait(log->bp, lbmFREE);
1672 
1673 	/*
1674 	 * synchronous update log superblock
1675 	 * mark log state as shutdown cleanly
1676 	 * (i.e., Log does not need to be replayed).
1677 	 */
1678 	if ((rc = lbmRead(log, 1, &bpsuper)))
1679 		goto out;
1680 
1681 	logsuper = (struct logsuper *) bpsuper->l_ldata;
1682 	logsuper->state = cpu_to_le32(LOGREDONE);
1683 	logsuper->end = cpu_to_le32(lsn);
1684 	lbmDirectWrite(log, bpsuper, lbmWRITE | lbmRELEASE | lbmSYNC);
1685 	rc = lbmIOWait(bpsuper, lbmFREE);
1686 
1687 	jfs_info("lmLogShutdown: lsn:0x%x page:%d eor:%d",
1688 		 lsn, log->page, log->eor);
1689 
1690       out:
1691 	/*
1692 	 * shutdown per log i/o
1693 	 */
1694 	lbmLogShutdown(log);
1695 
1696 	if (rc) {
1697 		jfs_warn("lmLogShutdown: exit(%d)", rc);
1698 	}
1699 	return rc;
1700 }
1701 
1702 
1703 /*
1704  * NAME:	lmLogFileSystem()
1705  *
1706  * FUNCTION:	insert (<activate> = true)/remove (<activate> = false)
1707  *	file system into/from log active file system list.
1708  *
1709  * PARAMETE:	log	- pointer to logs inode.
1710  *		fsdev	- kdev_t of filesystem.
1711  *		serial  - pointer to returned log serial number
1712  *		activate - insert/remove device from active list.
1713  *
1714  * RETURN:	0	- success
1715  *		errors returned by vms_iowait().
1716  */
1717 static int lmLogFileSystem(struct jfs_log * log, struct jfs_sb_info *sbi,
1718 			   int activate)
1719 {
1720 	int rc = 0;
1721 	int i;
1722 	struct logsuper *logsuper;
1723 	struct lbuf *bpsuper;
1724 	char *uuid = sbi->uuid;
1725 
1726 	/*
1727 	 * insert/remove file system device to log active file system list.
1728 	 */
1729 	if ((rc = lbmRead(log, 1, &bpsuper)))
1730 		return rc;
1731 
1732 	logsuper = (struct logsuper *) bpsuper->l_ldata;
1733 	if (activate) {
1734 		for (i = 0; i < MAX_ACTIVE; i++)
1735 			if (!memcmp(logsuper->active[i].uuid, NULL_UUID, 16)) {
1736 				memcpy(logsuper->active[i].uuid, uuid, 16);
1737 				sbi->aggregate = i;
1738 				break;
1739 			}
1740 		if (i == MAX_ACTIVE) {
1741 			jfs_warn("Too many file systems sharing journal!");
1742 			lbmFree(bpsuper);
1743 			return -EMFILE;	/* Is there a better rc? */
1744 		}
1745 	} else {
1746 		for (i = 0; i < MAX_ACTIVE; i++)
1747 			if (!memcmp(logsuper->active[i].uuid, uuid, 16)) {
1748 				memcpy(logsuper->active[i].uuid, NULL_UUID, 16);
1749 				break;
1750 			}
1751 		if (i == MAX_ACTIVE) {
1752 			jfs_warn("Somebody stomped on the journal!");
1753 			lbmFree(bpsuper);
1754 			return -EIO;
1755 		}
1756 
1757 	}
1758 
1759 	/*
1760 	 * synchronous write log superblock:
1761 	 *
1762 	 * write sidestream bypassing write queue:
1763 	 * at file system mount, log super block is updated for
1764 	 * activation of the file system before any log record
1765 	 * (MOUNT record) of the file system, and at file system
1766 	 * unmount, all meta data for the file system has been
1767 	 * flushed before log super block is updated for deactivation
1768 	 * of the file system.
1769 	 */
1770 	lbmDirectWrite(log, bpsuper, lbmWRITE | lbmRELEASE | lbmSYNC);
1771 	rc = lbmIOWait(bpsuper, lbmFREE);
1772 
1773 	return rc;
1774 }
1775 
1776 /*
1777  *		log buffer manager (lbm)
1778  *		------------------------
1779  *
1780  * special purpose buffer manager supporting log i/o requirements.
1781  *
1782  * per log write queue:
1783  * log pageout occurs in serial order by fifo write queue and
1784  * restricting to a single i/o in pregress at any one time.
1785  * a circular singly-linked list
1786  * (log->wrqueue points to the tail, and buffers are linked via
1787  * bp->wrqueue field), and
1788  * maintains log page in pageout ot waiting for pageout in serial pageout.
1789  */
1790 
1791 /*
1792  *	lbmLogInit()
1793  *
1794  * initialize per log I/O setup at lmLogInit()
1795  */
1796 static int lbmLogInit(struct jfs_log * log)
1797 {				/* log inode */
1798 	int i;
1799 	struct lbuf *lbuf;
1800 
1801 	jfs_info("lbmLogInit: log:0x%p", log);
1802 
1803 	/* initialize current buffer cursor */
1804 	log->bp = NULL;
1805 
1806 	/* initialize log device write queue */
1807 	log->wqueue = NULL;
1808 
1809 	/*
1810 	 * Each log has its own buffer pages allocated to it.  These are
1811 	 * not managed by the page cache.  This ensures that a transaction
1812 	 * writing to the log does not block trying to allocate a page from
1813 	 * the page cache (for the log).  This would be bad, since page
1814 	 * allocation waits on the kswapd thread that may be committing inodes
1815 	 * which would cause log activity.  Was that clear?  I'm trying to
1816 	 * avoid deadlock here.
1817 	 */
1818 	init_waitqueue_head(&log->free_wait);
1819 
1820 	log->lbuf_free = NULL;
1821 
1822 	for (i = 0; i < LOGPAGES; i++) {
1823 		lbuf = kmalloc(sizeof(struct lbuf), GFP_KERNEL);
1824 		if (lbuf == 0)
1825 			goto error;
1826 		lbuf->l_ldata = (char *) get_zeroed_page(GFP_KERNEL);
1827 		if (lbuf->l_ldata == 0) {
1828 			kfree(lbuf);
1829 			goto error;
1830 		}
1831 		lbuf->l_log = log;
1832 		init_waitqueue_head(&lbuf->l_ioevent);
1833 
1834 		lbuf->l_freelist = log->lbuf_free;
1835 		log->lbuf_free = lbuf;
1836 	}
1837 
1838 	return (0);
1839 
1840       error:
1841 	lbmLogShutdown(log);
1842 	return -ENOMEM;
1843 }
1844 
1845 
1846 /*
1847  *	lbmLogShutdown()
1848  *
1849  * finalize per log I/O setup at lmLogShutdown()
1850  */
1851 static void lbmLogShutdown(struct jfs_log * log)
1852 {
1853 	struct lbuf *lbuf;
1854 
1855 	jfs_info("lbmLogShutdown: log:0x%p", log);
1856 
1857 	lbuf = log->lbuf_free;
1858 	while (lbuf) {
1859 		struct lbuf *next = lbuf->l_freelist;
1860 		free_page((unsigned long) lbuf->l_ldata);
1861 		kfree(lbuf);
1862 		lbuf = next;
1863 	}
1864 
1865 	log->bp = NULL;
1866 }
1867 
1868 
1869 /*
1870  *	lbmAllocate()
1871  *
1872  * allocate an empty log buffer
1873  */
1874 static struct lbuf *lbmAllocate(struct jfs_log * log, int pn)
1875 {
1876 	struct lbuf *bp;
1877 	unsigned long flags;
1878 
1879 	/*
1880 	 * recycle from log buffer freelist if any
1881 	 */
1882 	LCACHE_LOCK(flags);
1883 	LCACHE_SLEEP_COND(log->free_wait, (bp = log->lbuf_free), flags);
1884 	log->lbuf_free = bp->l_freelist;
1885 	LCACHE_UNLOCK(flags);
1886 
1887 	bp->l_flag = 0;
1888 
1889 	bp->l_wqnext = NULL;
1890 	bp->l_freelist = NULL;
1891 
1892 	bp->l_pn = pn;
1893 	bp->l_blkno = log->base + (pn << (L2LOGPSIZE - log->l2bsize));
1894 	bp->l_ceor = 0;
1895 
1896 	return bp;
1897 }
1898 
1899 
1900 /*
1901  *	lbmFree()
1902  *
1903  * release a log buffer to freelist
1904  */
1905 static void lbmFree(struct lbuf * bp)
1906 {
1907 	unsigned long flags;
1908 
1909 	LCACHE_LOCK(flags);
1910 
1911 	lbmfree(bp);
1912 
1913 	LCACHE_UNLOCK(flags);
1914 }
1915 
1916 static void lbmfree(struct lbuf * bp)
1917 {
1918 	struct jfs_log *log = bp->l_log;
1919 
1920 	assert(bp->l_wqnext == NULL);
1921 
1922 	/*
1923 	 * return the buffer to head of freelist
1924 	 */
1925 	bp->l_freelist = log->lbuf_free;
1926 	log->lbuf_free = bp;
1927 
1928 	wake_up(&log->free_wait);
1929 	return;
1930 }
1931 
1932 
1933 /*
1934  * NAME:	lbmRedrive
1935  *
1936  * FUNCTION:	add a log buffer to the the log redrive list
1937  *
1938  * PARAMETER:
1939  *     bp	- log buffer
1940  *
1941  * NOTES:
1942  *	Takes log_redrive_lock.
1943  */
1944 static inline void lbmRedrive(struct lbuf *bp)
1945 {
1946 	unsigned long flags;
1947 
1948 	spin_lock_irqsave(&log_redrive_lock, flags);
1949 	bp->l_redrive_next = log_redrive_list;
1950 	log_redrive_list = bp;
1951 	spin_unlock_irqrestore(&log_redrive_lock, flags);
1952 
1953 	wake_up(&jfs_IO_thread_wait);
1954 }
1955 
1956 
1957 /*
1958  *	lbmRead()
1959  */
1960 static int lbmRead(struct jfs_log * log, int pn, struct lbuf ** bpp)
1961 {
1962 	struct bio *bio;
1963 	struct lbuf *bp;
1964 
1965 	/*
1966 	 * allocate a log buffer
1967 	 */
1968 	*bpp = bp = lbmAllocate(log, pn);
1969 	jfs_info("lbmRead: bp:0x%p pn:0x%x", bp, pn);
1970 
1971 	bp->l_flag |= lbmREAD;
1972 
1973 	bio = bio_alloc(GFP_NOFS, 1);
1974 
1975 	bio->bi_sector = bp->l_blkno << (log->l2bsize - 9);
1976 	bio->bi_bdev = log->bdev;
1977 	bio->bi_io_vec[0].bv_page = virt_to_page(bp->l_ldata);
1978 	bio->bi_io_vec[0].bv_len = LOGPSIZE;
1979 	bio->bi_io_vec[0].bv_offset = 0;
1980 
1981 	bio->bi_vcnt = 1;
1982 	bio->bi_idx = 0;
1983 	bio->bi_size = LOGPSIZE;
1984 
1985 	bio->bi_end_io = lbmIODone;
1986 	bio->bi_private = bp;
1987 	submit_bio(READ_SYNC, bio);
1988 
1989 	wait_event(bp->l_ioevent, (bp->l_flag != lbmREAD));
1990 
1991 	return 0;
1992 }
1993 
1994 
1995 /*
1996  *	lbmWrite()
1997  *
1998  * buffer at head of pageout queue stays after completion of
1999  * partial-page pageout and redriven by explicit initiation of
2000  * pageout by caller until full-page pageout is completed and
2001  * released.
2002  *
2003  * device driver i/o done redrives pageout of new buffer at
2004  * head of pageout queue when current buffer at head of pageout
2005  * queue is released at the completion of its full-page pageout.
2006  *
2007  * LOGGC_LOCK() serializes lbmWrite() by lmNextPage() and lmGroupCommit().
2008  * LCACHE_LOCK() serializes xflag between lbmWrite() and lbmIODone()
2009  */
2010 static void lbmWrite(struct jfs_log * log, struct lbuf * bp, int flag,
2011 		     int cant_block)
2012 {
2013 	struct lbuf *tail;
2014 	unsigned long flags;
2015 
2016 	jfs_info("lbmWrite: bp:0x%p flag:0x%x pn:0x%x", bp, flag, bp->l_pn);
2017 
2018 	/* map the logical block address to physical block address */
2019 	bp->l_blkno =
2020 	    log->base + (bp->l_pn << (L2LOGPSIZE - log->l2bsize));
2021 
2022 	LCACHE_LOCK(flags);		/* disable+lock */
2023 
2024 	/*
2025 	 * initialize buffer for device driver
2026 	 */
2027 	bp->l_flag = flag;
2028 
2029 	/*
2030 	 *      insert bp at tail of write queue associated with log
2031 	 *
2032 	 * (request is either for bp already/currently at head of queue
2033 	 * or new bp to be inserted at tail)
2034 	 */
2035 	tail = log->wqueue;
2036 
2037 	/* is buffer not already on write queue ? */
2038 	if (bp->l_wqnext == NULL) {
2039 		/* insert at tail of wqueue */
2040 		if (tail == NULL) {
2041 			log->wqueue = bp;
2042 			bp->l_wqnext = bp;
2043 		} else {
2044 			log->wqueue = bp;
2045 			bp->l_wqnext = tail->l_wqnext;
2046 			tail->l_wqnext = bp;
2047 		}
2048 
2049 		tail = bp;
2050 	}
2051 
2052 	/* is buffer at head of wqueue and for write ? */
2053 	if ((bp != tail->l_wqnext) || !(flag & lbmWRITE)) {
2054 		LCACHE_UNLOCK(flags);	/* unlock+enable */
2055 		return;
2056 	}
2057 
2058 	LCACHE_UNLOCK(flags);	/* unlock+enable */
2059 
2060 	if (cant_block)
2061 		lbmRedrive(bp);
2062 	else if (flag & lbmSYNC)
2063 		lbmStartIO(bp);
2064 	else {
2065 		LOGGC_UNLOCK(log);
2066 		lbmStartIO(bp);
2067 		LOGGC_LOCK(log);
2068 	}
2069 }
2070 
2071 
2072 /*
2073  *	lbmDirectWrite()
2074  *
2075  * initiate pageout bypassing write queue for sidestream
2076  * (e.g., log superblock) write;
2077  */
2078 static void lbmDirectWrite(struct jfs_log * log, struct lbuf * bp, int flag)
2079 {
2080 	jfs_info("lbmDirectWrite: bp:0x%p flag:0x%x pn:0x%x",
2081 		 bp, flag, bp->l_pn);
2082 
2083 	/*
2084 	 * initialize buffer for device driver
2085 	 */
2086 	bp->l_flag = flag | lbmDIRECT;
2087 
2088 	/* map the logical block address to physical block address */
2089 	bp->l_blkno =
2090 	    log->base + (bp->l_pn << (L2LOGPSIZE - log->l2bsize));
2091 
2092 	/*
2093 	 *      initiate pageout of the page
2094 	 */
2095 	lbmStartIO(bp);
2096 }
2097 
2098 
2099 /*
2100  * NAME:	lbmStartIO()
2101  *
2102  * FUNCTION:	Interface to DD strategy routine
2103  *
2104  * RETURN:      none
2105  *
2106  * serialization: LCACHE_LOCK() is NOT held during log i/o;
2107  */
2108 static void lbmStartIO(struct lbuf * bp)
2109 {
2110 	struct bio *bio;
2111 	struct jfs_log *log = bp->l_log;
2112 
2113 	jfs_info("lbmStartIO\n");
2114 
2115 	bio = bio_alloc(GFP_NOFS, 1);
2116 	bio->bi_sector = bp->l_blkno << (log->l2bsize - 9);
2117 	bio->bi_bdev = log->bdev;
2118 	bio->bi_io_vec[0].bv_page = virt_to_page(bp->l_ldata);
2119 	bio->bi_io_vec[0].bv_len = LOGPSIZE;
2120 	bio->bi_io_vec[0].bv_offset = 0;
2121 
2122 	bio->bi_vcnt = 1;
2123 	bio->bi_idx = 0;
2124 	bio->bi_size = LOGPSIZE;
2125 
2126 	bio->bi_end_io = lbmIODone;
2127 	bio->bi_private = bp;
2128 
2129 	/* check if journaling to disk has been disabled */
2130 	if (!log->no_integrity) {
2131 		submit_bio(WRITE_SYNC, bio);
2132 		INCREMENT(lmStat.submitted);
2133 	}
2134 	else {
2135 		bio->bi_size = 0;
2136 		lbmIODone(bio, 0, 0); /* 2nd argument appears to not be used => 0
2137 				       *  3rd argument appears to not be used => 0
2138 				       */
2139 	}
2140 }
2141 
2142 
2143 /*
2144  *	lbmIOWait()
2145  */
2146 static int lbmIOWait(struct lbuf * bp, int flag)
2147 {
2148 	unsigned long flags;
2149 	int rc = 0;
2150 
2151 	jfs_info("lbmIOWait1: bp:0x%p flag:0x%x:0x%x", bp, bp->l_flag, flag);
2152 
2153 	LCACHE_LOCK(flags);		/* disable+lock */
2154 
2155 	LCACHE_SLEEP_COND(bp->l_ioevent, (bp->l_flag & lbmDONE), flags);
2156 
2157 	rc = (bp->l_flag & lbmERROR) ? -EIO : 0;
2158 
2159 	if (flag & lbmFREE)
2160 		lbmfree(bp);
2161 
2162 	LCACHE_UNLOCK(flags);	/* unlock+enable */
2163 
2164 	jfs_info("lbmIOWait2: bp:0x%p flag:0x%x:0x%x", bp, bp->l_flag, flag);
2165 	return rc;
2166 }
2167 
2168 /*
2169  *	lbmIODone()
2170  *
2171  * executed at INTIODONE level
2172  */
2173 static int lbmIODone(struct bio *bio, unsigned int bytes_done, int error)
2174 {
2175 	struct lbuf *bp = bio->bi_private;
2176 	struct lbuf *nextbp, *tail;
2177 	struct jfs_log *log;
2178 	unsigned long flags;
2179 
2180 	if (bio->bi_size)
2181 		return 1;
2182 
2183 	/*
2184 	 * get back jfs buffer bound to the i/o buffer
2185 	 */
2186 	jfs_info("lbmIODone: bp:0x%p flag:0x%x", bp, bp->l_flag);
2187 
2188 	LCACHE_LOCK(flags);		/* disable+lock */
2189 
2190 	bp->l_flag |= lbmDONE;
2191 
2192 	if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
2193 		bp->l_flag |= lbmERROR;
2194 
2195 		jfs_err("lbmIODone: I/O error in JFS log");
2196 	}
2197 
2198 	bio_put(bio);
2199 
2200 	/*
2201 	 *      pagein completion
2202 	 */
2203 	if (bp->l_flag & lbmREAD) {
2204 		bp->l_flag &= ~lbmREAD;
2205 
2206 		LCACHE_UNLOCK(flags);	/* unlock+enable */
2207 
2208 		/* wakeup I/O initiator */
2209 		LCACHE_WAKEUP(&bp->l_ioevent);
2210 
2211 		return 0;
2212 	}
2213 
2214 	/*
2215 	 *      pageout completion
2216 	 *
2217 	 * the bp at the head of write queue has completed pageout.
2218 	 *
2219 	 * if single-commit/full-page pageout, remove the current buffer
2220 	 * from head of pageout queue, and redrive pageout with
2221 	 * the new buffer at head of pageout queue;
2222 	 * otherwise, the partial-page pageout buffer stays at
2223 	 * the head of pageout queue to be redriven for pageout
2224 	 * by lmGroupCommit() until full-page pageout is completed.
2225 	 */
2226 	bp->l_flag &= ~lbmWRITE;
2227 	INCREMENT(lmStat.pagedone);
2228 
2229 	/* update committed lsn */
2230 	log = bp->l_log;
2231 	log->clsn = (bp->l_pn << L2LOGPSIZE) + bp->l_ceor;
2232 
2233 	if (bp->l_flag & lbmDIRECT) {
2234 		LCACHE_WAKEUP(&bp->l_ioevent);
2235 		LCACHE_UNLOCK(flags);
2236 		return 0;
2237 	}
2238 
2239 	tail = log->wqueue;
2240 
2241 	/* single element queue */
2242 	if (bp == tail) {
2243 		/* remove head buffer of full-page pageout
2244 		 * from log device write queue
2245 		 */
2246 		if (bp->l_flag & lbmRELEASE) {
2247 			log->wqueue = NULL;
2248 			bp->l_wqnext = NULL;
2249 		}
2250 	}
2251 	/* multi element queue */
2252 	else {
2253 		/* remove head buffer of full-page pageout
2254 		 * from log device write queue
2255 		 */
2256 		if (bp->l_flag & lbmRELEASE) {
2257 			nextbp = tail->l_wqnext = bp->l_wqnext;
2258 			bp->l_wqnext = NULL;
2259 
2260 			/*
2261 			 * redrive pageout of next page at head of write queue:
2262 			 * redrive next page without any bound tblk
2263 			 * (i.e., page w/o any COMMIT records), or
2264 			 * first page of new group commit which has been
2265 			 * queued after current page (subsequent pageout
2266 			 * is performed synchronously, except page without
2267 			 * any COMMITs) by lmGroupCommit() as indicated
2268 			 * by lbmWRITE flag;
2269 			 */
2270 			if (nextbp->l_flag & lbmWRITE) {
2271 				/*
2272 				 * We can't do the I/O at interrupt time.
2273 				 * The jfsIO thread can do it
2274 				 */
2275 				lbmRedrive(nextbp);
2276 			}
2277 		}
2278 	}
2279 
2280 	/*
2281 	 *      synchronous pageout:
2282 	 *
2283 	 * buffer has not necessarily been removed from write queue
2284 	 * (e.g., synchronous write of partial-page with COMMIT):
2285 	 * leave buffer for i/o initiator to dispose
2286 	 */
2287 	if (bp->l_flag & lbmSYNC) {
2288 		LCACHE_UNLOCK(flags);	/* unlock+enable */
2289 
2290 		/* wakeup I/O initiator */
2291 		LCACHE_WAKEUP(&bp->l_ioevent);
2292 	}
2293 
2294 	/*
2295 	 *      Group Commit pageout:
2296 	 */
2297 	else if (bp->l_flag & lbmGC) {
2298 		LCACHE_UNLOCK(flags);
2299 		lmPostGC(bp);
2300 	}
2301 
2302 	/*
2303 	 *      asynchronous pageout:
2304 	 *
2305 	 * buffer must have been removed from write queue:
2306 	 * insert buffer at head of freelist where it can be recycled
2307 	 */
2308 	else {
2309 		assert(bp->l_flag & lbmRELEASE);
2310 		assert(bp->l_flag & lbmFREE);
2311 		lbmfree(bp);
2312 
2313 		LCACHE_UNLOCK(flags);	/* unlock+enable */
2314 	}
2315 
2316 	return 0;
2317 }
2318 
2319 int jfsIOWait(void *arg)
2320 {
2321 	struct lbuf *bp;
2322 
2323 	daemonize("jfsIO");
2324 
2325 	complete(&jfsIOwait);
2326 
2327 	do {
2328 		DECLARE_WAITQUEUE(wq, current);
2329 
2330 		spin_lock_irq(&log_redrive_lock);
2331 		while ((bp = log_redrive_list) != 0) {
2332 			log_redrive_list = bp->l_redrive_next;
2333 			bp->l_redrive_next = NULL;
2334 			spin_unlock_irq(&log_redrive_lock);
2335 			lbmStartIO(bp);
2336 			spin_lock_irq(&log_redrive_lock);
2337 		}
2338 		if (current->flags & PF_FREEZE) {
2339 			spin_unlock_irq(&log_redrive_lock);
2340 			refrigerator(PF_FREEZE);
2341 		} else {
2342 			add_wait_queue(&jfs_IO_thread_wait, &wq);
2343 			set_current_state(TASK_INTERRUPTIBLE);
2344 			spin_unlock_irq(&log_redrive_lock);
2345 			schedule();
2346 			current->state = TASK_RUNNING;
2347 			remove_wait_queue(&jfs_IO_thread_wait, &wq);
2348 		}
2349 	} while (!jfs_stop_threads);
2350 
2351 	jfs_info("jfsIOWait being killed!");
2352 	complete_and_exit(&jfsIOwait, 0);
2353 }
2354 
2355 /*
2356  * NAME:	lmLogFormat()/jfs_logform()
2357  *
2358  * FUNCTION:	format file system log
2359  *
2360  * PARAMETERS:
2361  *      log	- volume log
2362  *	logAddress - start address of log space in FS block
2363  *	logSize	- length of log space in FS block;
2364  *
2365  * RETURN:	0	- success
2366  *		-EIO	- i/o error
2367  *
2368  * XXX: We're synchronously writing one page at a time.  This needs to
2369  *	be improved by writing multiple pages at once.
2370  */
2371 int lmLogFormat(struct jfs_log *log, s64 logAddress, int logSize)
2372 {
2373 	int rc = -EIO;
2374 	struct jfs_sb_info *sbi;
2375 	struct logsuper *logsuper;
2376 	struct logpage *lp;
2377 	int lspn;		/* log sequence page number */
2378 	struct lrd *lrd_ptr;
2379 	int npages = 0;
2380 	struct lbuf *bp;
2381 
2382 	jfs_info("lmLogFormat: logAddress:%Ld logSize:%d",
2383 		 (long long)logAddress, logSize);
2384 
2385 	sbi = list_entry(log->sb_list.next, struct jfs_sb_info, log_list);
2386 
2387 	/* allocate a log buffer */
2388 	bp = lbmAllocate(log, 1);
2389 
2390 	npages = logSize >> sbi->l2nbperpage;
2391 
2392 	/*
2393 	 *      log space:
2394 	 *
2395 	 * page 0 - reserved;
2396 	 * page 1 - log superblock;
2397 	 * page 2 - log data page: A SYNC log record is written
2398 	 *          into this page at logform time;
2399 	 * pages 3-N - log data page: set to empty log data pages;
2400 	 */
2401 	/*
2402 	 *      init log superblock: log page 1
2403 	 */
2404 	logsuper = (struct logsuper *) bp->l_ldata;
2405 
2406 	logsuper->magic = cpu_to_le32(LOGMAGIC);
2407 	logsuper->version = cpu_to_le32(LOGVERSION);
2408 	logsuper->state = cpu_to_le32(LOGREDONE);
2409 	logsuper->flag = cpu_to_le32(sbi->mntflag);	/* ? */
2410 	logsuper->size = cpu_to_le32(npages);
2411 	logsuper->bsize = cpu_to_le32(sbi->bsize);
2412 	logsuper->l2bsize = cpu_to_le32(sbi->l2bsize);
2413 	logsuper->end = cpu_to_le32(2 * LOGPSIZE + LOGPHDRSIZE + LOGRDSIZE);
2414 
2415 	bp->l_flag = lbmWRITE | lbmSYNC | lbmDIRECT;
2416 	bp->l_blkno = logAddress + sbi->nbperpage;
2417 	lbmStartIO(bp);
2418 	if ((rc = lbmIOWait(bp, 0)))
2419 		goto exit;
2420 
2421 	/*
2422 	 *      init pages 2 to npages-1 as log data pages:
2423 	 *
2424 	 * log page sequence number (lpsn) initialization:
2425 	 *
2426 	 * pn:   0     1     2     3                 n-1
2427 	 *       +-----+-----+=====+=====+===.....===+=====+
2428 	 * lspn:             N-1   0     1           N-2
2429 	 *                   <--- N page circular file ---->
2430 	 *
2431 	 * the N (= npages-2) data pages of the log is maintained as
2432 	 * a circular file for the log records;
2433 	 * lpsn grows by 1 monotonically as each log page is written
2434 	 * to the circular file of the log;
2435 	 * and setLogpage() will not reset the page number even if
2436 	 * the eor is equal to LOGPHDRSIZE. In order for binary search
2437 	 * still work in find log end process, we have to simulate the
2438 	 * log wrap situation at the log format time.
2439 	 * The 1st log page written will have the highest lpsn. Then
2440 	 * the succeeding log pages will have ascending order of
2441 	 * the lspn starting from 0, ... (N-2)
2442 	 */
2443 	lp = (struct logpage *) bp->l_ldata;
2444 	/*
2445 	 * initialize 1st log page to be written: lpsn = N - 1,
2446 	 * write a SYNCPT log record is written to this page
2447 	 */
2448 	lp->h.page = lp->t.page = cpu_to_le32(npages - 3);
2449 	lp->h.eor = lp->t.eor = cpu_to_le16(LOGPHDRSIZE + LOGRDSIZE);
2450 
2451 	lrd_ptr = (struct lrd *) &lp->data;
2452 	lrd_ptr->logtid = 0;
2453 	lrd_ptr->backchain = 0;
2454 	lrd_ptr->type = cpu_to_le16(LOG_SYNCPT);
2455 	lrd_ptr->length = 0;
2456 	lrd_ptr->log.syncpt.sync = 0;
2457 
2458 	bp->l_blkno += sbi->nbperpage;
2459 	bp->l_flag = lbmWRITE | lbmSYNC | lbmDIRECT;
2460 	lbmStartIO(bp);
2461 	if ((rc = lbmIOWait(bp, 0)))
2462 		goto exit;
2463 
2464 	/*
2465 	 *      initialize succeeding log pages: lpsn = 0, 1, ..., (N-2)
2466 	 */
2467 	for (lspn = 0; lspn < npages - 3; lspn++) {
2468 		lp->h.page = lp->t.page = cpu_to_le32(lspn);
2469 		lp->h.eor = lp->t.eor = cpu_to_le16(LOGPHDRSIZE);
2470 
2471 		bp->l_blkno += sbi->nbperpage;
2472 		bp->l_flag = lbmWRITE | lbmSYNC | lbmDIRECT;
2473 		lbmStartIO(bp);
2474 		if ((rc = lbmIOWait(bp, 0)))
2475 			goto exit;
2476 	}
2477 
2478 	rc = 0;
2479 exit:
2480 	/*
2481 	 *      finalize log
2482 	 */
2483 	/* release the buffer */
2484 	lbmFree(bp);
2485 
2486 	return rc;
2487 }
2488 
2489 #ifdef CONFIG_JFS_STATISTICS
2490 int jfs_lmstats_read(char *buffer, char **start, off_t offset, int length,
2491 		      int *eof, void *data)
2492 {
2493 	int len = 0;
2494 	off_t begin;
2495 
2496 	len += sprintf(buffer,
2497 		       "JFS Logmgr stats\n"
2498 		       "================\n"
2499 		       "commits = %d\n"
2500 		       "writes submitted = %d\n"
2501 		       "writes completed = %d\n"
2502 		       "full pages submitted = %d\n"
2503 		       "partial pages submitted = %d\n",
2504 		       lmStat.commit,
2505 		       lmStat.submitted,
2506 		       lmStat.pagedone,
2507 		       lmStat.full_page,
2508 		       lmStat.partial_page);
2509 
2510 	begin = offset;
2511 	*start = buffer + begin;
2512 	len -= begin;
2513 
2514 	if (len > length)
2515 		len = length;
2516 	else
2517 		*eof = 1;
2518 
2519 	if (len < 0)
2520 		len = 0;
2521 
2522 	return len;
2523 }
2524 #endif /* CONFIG_JFS_STATISTICS */
2525