xref: /linux/fs/jfs/jfs_logmgr.c (revision 3f0a50f345f78183f6e9b39c2f45ca5dcaa511ca)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *   Copyright (C) International Business Machines Corp., 2000-2004
4  *   Portions Copyright (C) Christoph Hellwig, 2001-2002
5  */
6 
7 /*
8  *	jfs_logmgr.c: log manager
9  *
10  * for related information, see transaction manager (jfs_txnmgr.c), and
11  * recovery manager (jfs_logredo.c).
12  *
13  * note: for detail, RTFS.
14  *
15  *	log buffer manager:
16  * special purpose buffer manager supporting log i/o requirements.
17  * per log serial pageout of logpage
18  * queuing i/o requests and redrive i/o at iodone
19  * maintain current logpage buffer
20  * no caching since append only
21  * appropriate jfs buffer cache buffers as needed
22  *
23  *	group commit:
24  * transactions which wrote COMMIT records in the same in-memory
25  * log page during the pageout of previous/current log page(s) are
26  * committed together by the pageout of the page.
27  *
28  *	TBD lazy commit:
29  * transactions are committed asynchronously when the log page
30  * containing it COMMIT is paged out when it becomes full;
31  *
32  *	serialization:
33  * . a per log lock serialize log write.
34  * . a per log lock serialize group commit.
35  * . a per log lock serialize log open/close;
36  *
37  *	TBD log integrity:
38  * careful-write (ping-pong) of last logpage to recover from crash
39  * in overwrite.
40  * detection of split (out-of-order) write of physical sectors
41  * of last logpage via timestamp at end of each sector
42  * with its mirror data array at trailer).
43  *
44  *	alternatives:
45  * lsn - 64-bit monotonically increasing integer vs
46  * 32-bit lspn and page eor.
47  */
48 
49 #include <linux/fs.h>
50 #include <linux/blkdev.h>
51 #include <linux/interrupt.h>
52 #include <linux/completion.h>
53 #include <linux/kthread.h>
54 #include <linux/buffer_head.h>		/* for sync_blockdev() */
55 #include <linux/bio.h>
56 #include <linux/freezer.h>
57 #include <linux/export.h>
58 #include <linux/delay.h>
59 #include <linux/mutex.h>
60 #include <linux/seq_file.h>
61 #include <linux/slab.h>
62 #include "jfs_incore.h"
63 #include "jfs_filsys.h"
64 #include "jfs_metapage.h"
65 #include "jfs_superblock.h"
66 #include "jfs_txnmgr.h"
67 #include "jfs_debug.h"
68 
69 
70 /*
71  * lbuf's ready to be redriven.  Protected by log_redrive_lock (jfsIO thread)
72  */
73 static struct lbuf *log_redrive_list;
74 static DEFINE_SPINLOCK(log_redrive_lock);
75 
76 
77 /*
78  *	log read/write serialization (per log)
79  */
80 #define LOG_LOCK_INIT(log)	mutex_init(&(log)->loglock)
81 #define LOG_LOCK(log)		mutex_lock(&((log)->loglock))
82 #define LOG_UNLOCK(log)		mutex_unlock(&((log)->loglock))
83 
84 
85 /*
86  *	log group commit serialization (per log)
87  */
88 
89 #define LOGGC_LOCK_INIT(log)	spin_lock_init(&(log)->gclock)
90 #define LOGGC_LOCK(log)		spin_lock_irq(&(log)->gclock)
91 #define LOGGC_UNLOCK(log)	spin_unlock_irq(&(log)->gclock)
92 #define LOGGC_WAKEUP(tblk)	wake_up_all(&(tblk)->gcwait)
93 
94 /*
95  *	log sync serialization (per log)
96  */
97 #define	LOGSYNC_DELTA(logsize)		min((logsize)/8, 128*LOGPSIZE)
98 #define	LOGSYNC_BARRIER(logsize)	((logsize)/4)
99 /*
100 #define	LOGSYNC_DELTA(logsize)		min((logsize)/4, 256*LOGPSIZE)
101 #define	LOGSYNC_BARRIER(logsize)	((logsize)/2)
102 */
103 
104 
105 /*
106  *	log buffer cache synchronization
107  */
108 static DEFINE_SPINLOCK(jfsLCacheLock);
109 
110 #define	LCACHE_LOCK(flags)	spin_lock_irqsave(&jfsLCacheLock, flags)
111 #define	LCACHE_UNLOCK(flags)	spin_unlock_irqrestore(&jfsLCacheLock, flags)
112 
113 /*
114  * See __SLEEP_COND in jfs_locks.h
115  */
116 #define LCACHE_SLEEP_COND(wq, cond, flags)	\
117 do {						\
118 	if (cond)				\
119 		break;				\
120 	__SLEEP_COND(wq, cond, LCACHE_LOCK(flags), LCACHE_UNLOCK(flags)); \
121 } while (0)
122 
123 #define	LCACHE_WAKEUP(event)	wake_up(event)
124 
125 
126 /*
127  *	lbuf buffer cache (lCache) control
128  */
129 /* log buffer manager pageout control (cumulative, inclusive) */
130 #define	lbmREAD		0x0001
131 #define	lbmWRITE	0x0002	/* enqueue at tail of write queue;
132 				 * init pageout if at head of queue;
133 				 */
134 #define	lbmRELEASE	0x0004	/* remove from write queue
135 				 * at completion of pageout;
136 				 * do not free/recycle it yet:
137 				 * caller will free it;
138 				 */
139 #define	lbmSYNC		0x0008	/* do not return to freelist
140 				 * when removed from write queue;
141 				 */
142 #define lbmFREE		0x0010	/* return to freelist
143 				 * at completion of pageout;
144 				 * the buffer may be recycled;
145 				 */
146 #define	lbmDONE		0x0020
147 #define	lbmERROR	0x0040
148 #define lbmGC		0x0080	/* lbmIODone to perform post-GC processing
149 				 * of log page
150 				 */
151 #define lbmDIRECT	0x0100
152 
153 /*
154  * Global list of active external journals
155  */
156 static LIST_HEAD(jfs_external_logs);
157 static struct jfs_log *dummy_log;
158 static DEFINE_MUTEX(jfs_log_mutex);
159 
160 /*
161  * forward references
162  */
163 static int lmWriteRecord(struct jfs_log * log, struct tblock * tblk,
164 			 struct lrd * lrd, struct tlock * tlck);
165 
166 static int lmNextPage(struct jfs_log * log);
167 static int lmLogFileSystem(struct jfs_log * log, struct jfs_sb_info *sbi,
168 			   int activate);
169 
170 static int open_inline_log(struct super_block *sb);
171 static int open_dummy_log(struct super_block *sb);
172 static int lbmLogInit(struct jfs_log * log);
173 static void lbmLogShutdown(struct jfs_log * log);
174 static struct lbuf *lbmAllocate(struct jfs_log * log, int);
175 static void lbmFree(struct lbuf * bp);
176 static void lbmfree(struct lbuf * bp);
177 static int lbmRead(struct jfs_log * log, int pn, struct lbuf ** bpp);
178 static void lbmWrite(struct jfs_log * log, struct lbuf * bp, int flag, int cant_block);
179 static void lbmDirectWrite(struct jfs_log * log, struct lbuf * bp, int flag);
180 static int lbmIOWait(struct lbuf * bp, int flag);
181 static bio_end_io_t lbmIODone;
182 static void lbmStartIO(struct lbuf * bp);
183 static void lmGCwrite(struct jfs_log * log, int cant_block);
184 static int lmLogSync(struct jfs_log * log, int hard_sync);
185 
186 
187 
188 /*
189  *	statistics
190  */
191 #ifdef CONFIG_JFS_STATISTICS
192 static struct lmStat {
193 	uint commit;		/* # of commit */
194 	uint pagedone;		/* # of page written */
195 	uint submitted;		/* # of pages submitted */
196 	uint full_page;		/* # of full pages submitted */
197 	uint partial_page;	/* # of partial pages submitted */
198 } lmStat;
199 #endif
200 
201 static void write_special_inodes(struct jfs_log *log,
202 				 int (*writer)(struct address_space *))
203 {
204 	struct jfs_sb_info *sbi;
205 
206 	list_for_each_entry(sbi, &log->sb_list, log_list) {
207 		writer(sbi->ipbmap->i_mapping);
208 		writer(sbi->ipimap->i_mapping);
209 		writer(sbi->direct_inode->i_mapping);
210 	}
211 }
212 
213 /*
214  * NAME:	lmLog()
215  *
216  * FUNCTION:	write a log record;
217  *
218  * PARAMETER:
219  *
220  * RETURN:	lsn - offset to the next log record to write (end-of-log);
221  *		-1  - error;
222  *
223  * note: todo: log error handler
224  */
225 int lmLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
226 	  struct tlock * tlck)
227 {
228 	int lsn;
229 	int diffp, difft;
230 	struct metapage *mp = NULL;
231 	unsigned long flags;
232 
233 	jfs_info("lmLog: log:0x%p tblk:0x%p, lrd:0x%p tlck:0x%p",
234 		 log, tblk, lrd, tlck);
235 
236 	LOG_LOCK(log);
237 
238 	/* log by (out-of-transaction) JFS ? */
239 	if (tblk == NULL)
240 		goto writeRecord;
241 
242 	/* log from page ? */
243 	if (tlck == NULL ||
244 	    tlck->type & tlckBTROOT || (mp = tlck->mp) == NULL)
245 		goto writeRecord;
246 
247 	/*
248 	 *	initialize/update page/transaction recovery lsn
249 	 */
250 	lsn = log->lsn;
251 
252 	LOGSYNC_LOCK(log, flags);
253 
254 	/*
255 	 * initialize page lsn if first log write of the page
256 	 */
257 	if (mp->lsn == 0) {
258 		mp->log = log;
259 		mp->lsn = lsn;
260 		log->count++;
261 
262 		/* insert page at tail of logsynclist */
263 		list_add_tail(&mp->synclist, &log->synclist);
264 	}
265 
266 	/*
267 	 *	initialize/update lsn of tblock of the page
268 	 *
269 	 * transaction inherits oldest lsn of pages associated
270 	 * with allocation/deallocation of resources (their
271 	 * log records are used to reconstruct allocation map
272 	 * at recovery time: inode for inode allocation map,
273 	 * B+-tree index of extent descriptors for block
274 	 * allocation map);
275 	 * allocation map pages inherit transaction lsn at
276 	 * commit time to allow forwarding log syncpt past log
277 	 * records associated with allocation/deallocation of
278 	 * resources only after persistent map of these map pages
279 	 * have been updated and propagated to home.
280 	 */
281 	/*
282 	 * initialize transaction lsn:
283 	 */
284 	if (tblk->lsn == 0) {
285 		/* inherit lsn of its first page logged */
286 		tblk->lsn = mp->lsn;
287 		log->count++;
288 
289 		/* insert tblock after the page on logsynclist */
290 		list_add(&tblk->synclist, &mp->synclist);
291 	}
292 	/*
293 	 * update transaction lsn:
294 	 */
295 	else {
296 		/* inherit oldest/smallest lsn of page */
297 		logdiff(diffp, mp->lsn, log);
298 		logdiff(difft, tblk->lsn, log);
299 		if (diffp < difft) {
300 			/* update tblock lsn with page lsn */
301 			tblk->lsn = mp->lsn;
302 
303 			/* move tblock after page on logsynclist */
304 			list_move(&tblk->synclist, &mp->synclist);
305 		}
306 	}
307 
308 	LOGSYNC_UNLOCK(log, flags);
309 
310 	/*
311 	 *	write the log record
312 	 */
313       writeRecord:
314 	lsn = lmWriteRecord(log, tblk, lrd, tlck);
315 
316 	/*
317 	 * forward log syncpt if log reached next syncpt trigger
318 	 */
319 	logdiff(diffp, lsn, log);
320 	if (diffp >= log->nextsync)
321 		lsn = lmLogSync(log, 0);
322 
323 	/* update end-of-log lsn */
324 	log->lsn = lsn;
325 
326 	LOG_UNLOCK(log);
327 
328 	/* return end-of-log address */
329 	return lsn;
330 }
331 
332 /*
333  * NAME:	lmWriteRecord()
334  *
335  * FUNCTION:	move the log record to current log page
336  *
337  * PARAMETER:	cd	- commit descriptor
338  *
339  * RETURN:	end-of-log address
340  *
341  * serialization: LOG_LOCK() held on entry/exit
342  */
343 static int
344 lmWriteRecord(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
345 	      struct tlock * tlck)
346 {
347 	int lsn = 0;		/* end-of-log address */
348 	struct lbuf *bp;	/* dst log page buffer */
349 	struct logpage *lp;	/* dst log page */
350 	caddr_t dst;		/* destination address in log page */
351 	int dstoffset;		/* end-of-log offset in log page */
352 	int freespace;		/* free space in log page */
353 	caddr_t p;		/* src meta-data page */
354 	caddr_t src;
355 	int srclen;
356 	int nbytes;		/* number of bytes to move */
357 	int i;
358 	int len;
359 	struct linelock *linelock;
360 	struct lv *lv;
361 	struct lvd *lvd;
362 	int l2linesize;
363 
364 	len = 0;
365 
366 	/* retrieve destination log page to write */
367 	bp = (struct lbuf *) log->bp;
368 	lp = (struct logpage *) bp->l_ldata;
369 	dstoffset = log->eor;
370 
371 	/* any log data to write ? */
372 	if (tlck == NULL)
373 		goto moveLrd;
374 
375 	/*
376 	 *	move log record data
377 	 */
378 	/* retrieve source meta-data page to log */
379 	if (tlck->flag & tlckPAGELOCK) {
380 		p = (caddr_t) (tlck->mp->data);
381 		linelock = (struct linelock *) & tlck->lock;
382 	}
383 	/* retrieve source in-memory inode to log */
384 	else if (tlck->flag & tlckINODELOCK) {
385 		if (tlck->type & tlckDTREE)
386 			p = (caddr_t) &JFS_IP(tlck->ip)->i_dtroot;
387 		else
388 			p = (caddr_t) &JFS_IP(tlck->ip)->i_xtroot;
389 		linelock = (struct linelock *) & tlck->lock;
390 	}
391 #ifdef	_JFS_WIP
392 	else if (tlck->flag & tlckINLINELOCK) {
393 
394 		inlinelock = (struct inlinelock *) & tlck;
395 		p = (caddr_t) & inlinelock->pxd;
396 		linelock = (struct linelock *) & tlck;
397 	}
398 #endif				/* _JFS_WIP */
399 	else {
400 		jfs_err("lmWriteRecord: UFO tlck:0x%p", tlck);
401 		return 0;	/* Probably should trap */
402 	}
403 	l2linesize = linelock->l2linesize;
404 
405       moveData:
406 	ASSERT(linelock->index <= linelock->maxcnt);
407 
408 	lv = linelock->lv;
409 	for (i = 0; i < linelock->index; i++, lv++) {
410 		if (lv->length == 0)
411 			continue;
412 
413 		/* is page full ? */
414 		if (dstoffset >= LOGPSIZE - LOGPTLRSIZE) {
415 			/* page become full: move on to next page */
416 			lmNextPage(log);
417 
418 			bp = log->bp;
419 			lp = (struct logpage *) bp->l_ldata;
420 			dstoffset = LOGPHDRSIZE;
421 		}
422 
423 		/*
424 		 * move log vector data
425 		 */
426 		src = (u8 *) p + (lv->offset << l2linesize);
427 		srclen = lv->length << l2linesize;
428 		len += srclen;
429 		while (srclen > 0) {
430 			freespace = (LOGPSIZE - LOGPTLRSIZE) - dstoffset;
431 			nbytes = min(freespace, srclen);
432 			dst = (caddr_t) lp + dstoffset;
433 			memcpy(dst, src, nbytes);
434 			dstoffset += nbytes;
435 
436 			/* is page not full ? */
437 			if (dstoffset < LOGPSIZE - LOGPTLRSIZE)
438 				break;
439 
440 			/* page become full: move on to next page */
441 			lmNextPage(log);
442 
443 			bp = (struct lbuf *) log->bp;
444 			lp = (struct logpage *) bp->l_ldata;
445 			dstoffset = LOGPHDRSIZE;
446 
447 			srclen -= nbytes;
448 			src += nbytes;
449 		}
450 
451 		/*
452 		 * move log vector descriptor
453 		 */
454 		len += 4;
455 		lvd = (struct lvd *) ((caddr_t) lp + dstoffset);
456 		lvd->offset = cpu_to_le16(lv->offset);
457 		lvd->length = cpu_to_le16(lv->length);
458 		dstoffset += 4;
459 		jfs_info("lmWriteRecord: lv offset:%d length:%d",
460 			 lv->offset, lv->length);
461 	}
462 
463 	if ((i = linelock->next)) {
464 		linelock = (struct linelock *) lid_to_tlock(i);
465 		goto moveData;
466 	}
467 
468 	/*
469 	 *	move log record descriptor
470 	 */
471       moveLrd:
472 	lrd->length = cpu_to_le16(len);
473 
474 	src = (caddr_t) lrd;
475 	srclen = LOGRDSIZE;
476 
477 	while (srclen > 0) {
478 		freespace = (LOGPSIZE - LOGPTLRSIZE) - dstoffset;
479 		nbytes = min(freespace, srclen);
480 		dst = (caddr_t) lp + dstoffset;
481 		memcpy(dst, src, nbytes);
482 
483 		dstoffset += nbytes;
484 		srclen -= nbytes;
485 
486 		/* are there more to move than freespace of page ? */
487 		if (srclen)
488 			goto pageFull;
489 
490 		/*
491 		 * end of log record descriptor
492 		 */
493 
494 		/* update last log record eor */
495 		log->eor = dstoffset;
496 		bp->l_eor = dstoffset;
497 		lsn = (log->page << L2LOGPSIZE) + dstoffset;
498 
499 		if (lrd->type & cpu_to_le16(LOG_COMMIT)) {
500 			tblk->clsn = lsn;
501 			jfs_info("wr: tclsn:0x%x, beor:0x%x", tblk->clsn,
502 				 bp->l_eor);
503 
504 			INCREMENT(lmStat.commit);	/* # of commit */
505 
506 			/*
507 			 * enqueue tblock for group commit:
508 			 *
509 			 * enqueue tblock of non-trivial/synchronous COMMIT
510 			 * at tail of group commit queue
511 			 * (trivial/asynchronous COMMITs are ignored by
512 			 * group commit.)
513 			 */
514 			LOGGC_LOCK(log);
515 
516 			/* init tblock gc state */
517 			tblk->flag = tblkGC_QUEUE;
518 			tblk->bp = log->bp;
519 			tblk->pn = log->page;
520 			tblk->eor = log->eor;
521 
522 			/* enqueue transaction to commit queue */
523 			list_add_tail(&tblk->cqueue, &log->cqueue);
524 
525 			LOGGC_UNLOCK(log);
526 		}
527 
528 		jfs_info("lmWriteRecord: lrd:0x%04x bp:0x%p pn:%d eor:0x%x",
529 			le16_to_cpu(lrd->type), log->bp, log->page, dstoffset);
530 
531 		/* page not full ? */
532 		if (dstoffset < LOGPSIZE - LOGPTLRSIZE)
533 			return lsn;
534 
535 	      pageFull:
536 		/* page become full: move on to next page */
537 		lmNextPage(log);
538 
539 		bp = (struct lbuf *) log->bp;
540 		lp = (struct logpage *) bp->l_ldata;
541 		dstoffset = LOGPHDRSIZE;
542 		src += nbytes;
543 	}
544 
545 	return lsn;
546 }
547 
548 
549 /*
550  * NAME:	lmNextPage()
551  *
552  * FUNCTION:	write current page and allocate next page.
553  *
554  * PARAMETER:	log
555  *
556  * RETURN:	0
557  *
558  * serialization: LOG_LOCK() held on entry/exit
559  */
560 static int lmNextPage(struct jfs_log * log)
561 {
562 	struct logpage *lp;
563 	int lspn;		/* log sequence page number */
564 	int pn;			/* current page number */
565 	struct lbuf *bp;
566 	struct lbuf *nextbp;
567 	struct tblock *tblk;
568 
569 	/* get current log page number and log sequence page number */
570 	pn = log->page;
571 	bp = log->bp;
572 	lp = (struct logpage *) bp->l_ldata;
573 	lspn = le32_to_cpu(lp->h.page);
574 
575 	LOGGC_LOCK(log);
576 
577 	/*
578 	 *	write or queue the full page at the tail of write queue
579 	 */
580 	/* get the tail tblk on commit queue */
581 	if (list_empty(&log->cqueue))
582 		tblk = NULL;
583 	else
584 		tblk = list_entry(log->cqueue.prev, struct tblock, cqueue);
585 
586 	/* every tblk who has COMMIT record on the current page,
587 	 * and has not been committed, must be on commit queue
588 	 * since tblk is queued at commit queueu at the time
589 	 * of writing its COMMIT record on the page before
590 	 * page becomes full (even though the tblk thread
591 	 * who wrote COMMIT record may have been suspended
592 	 * currently);
593 	 */
594 
595 	/* is page bound with outstanding tail tblk ? */
596 	if (tblk && tblk->pn == pn) {
597 		/* mark tblk for end-of-page */
598 		tblk->flag |= tblkGC_EOP;
599 
600 		if (log->cflag & logGC_PAGEOUT) {
601 			/* if page is not already on write queue,
602 			 * just enqueue (no lbmWRITE to prevent redrive)
603 			 * buffer to wqueue to ensure correct serial order
604 			 * of the pages since log pages will be added
605 			 * continuously
606 			 */
607 			if (bp->l_wqnext == NULL)
608 				lbmWrite(log, bp, 0, 0);
609 		} else {
610 			/*
611 			 * No current GC leader, initiate group commit
612 			 */
613 			log->cflag |= logGC_PAGEOUT;
614 			lmGCwrite(log, 0);
615 		}
616 	}
617 	/* page is not bound with outstanding tblk:
618 	 * init write or mark it to be redriven (lbmWRITE)
619 	 */
620 	else {
621 		/* finalize the page */
622 		bp->l_ceor = bp->l_eor;
623 		lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_ceor);
624 		lbmWrite(log, bp, lbmWRITE | lbmRELEASE | lbmFREE, 0);
625 	}
626 	LOGGC_UNLOCK(log);
627 
628 	/*
629 	 *	allocate/initialize next page
630 	 */
631 	/* if log wraps, the first data page of log is 2
632 	 * (0 never used, 1 is superblock).
633 	 */
634 	log->page = (pn == log->size - 1) ? 2 : pn + 1;
635 	log->eor = LOGPHDRSIZE;	/* ? valid page empty/full at logRedo() */
636 
637 	/* allocate/initialize next log page buffer */
638 	nextbp = lbmAllocate(log, log->page);
639 	nextbp->l_eor = log->eor;
640 	log->bp = nextbp;
641 
642 	/* initialize next log page */
643 	lp = (struct logpage *) nextbp->l_ldata;
644 	lp->h.page = lp->t.page = cpu_to_le32(lspn + 1);
645 	lp->h.eor = lp->t.eor = cpu_to_le16(LOGPHDRSIZE);
646 
647 	return 0;
648 }
649 
650 
651 /*
652  * NAME:	lmGroupCommit()
653  *
654  * FUNCTION:	group commit
655  *	initiate pageout of the pages with COMMIT in the order of
656  *	page number - redrive pageout of the page at the head of
657  *	pageout queue until full page has been written.
658  *
659  * RETURN:
660  *
661  * NOTE:
662  *	LOGGC_LOCK serializes log group commit queue, and
663  *	transaction blocks on the commit queue.
664  *	N.B. LOG_LOCK is NOT held during lmGroupCommit().
665  */
666 int lmGroupCommit(struct jfs_log * log, struct tblock * tblk)
667 {
668 	int rc = 0;
669 
670 	LOGGC_LOCK(log);
671 
672 	/* group committed already ? */
673 	if (tblk->flag & tblkGC_COMMITTED) {
674 		if (tblk->flag & tblkGC_ERROR)
675 			rc = -EIO;
676 
677 		LOGGC_UNLOCK(log);
678 		return rc;
679 	}
680 	jfs_info("lmGroup Commit: tblk = 0x%p, gcrtc = %d", tblk, log->gcrtc);
681 
682 	if (tblk->xflag & COMMIT_LAZY)
683 		tblk->flag |= tblkGC_LAZY;
684 
685 	if ((!(log->cflag & logGC_PAGEOUT)) && (!list_empty(&log->cqueue)) &&
686 	    (!(tblk->xflag & COMMIT_LAZY) || test_bit(log_FLUSH, &log->flag)
687 	     || jfs_tlocks_low)) {
688 		/*
689 		 * No pageout in progress
690 		 *
691 		 * start group commit as its group leader.
692 		 */
693 		log->cflag |= logGC_PAGEOUT;
694 
695 		lmGCwrite(log, 0);
696 	}
697 
698 	if (tblk->xflag & COMMIT_LAZY) {
699 		/*
700 		 * Lazy transactions can leave now
701 		 */
702 		LOGGC_UNLOCK(log);
703 		return 0;
704 	}
705 
706 	/* lmGCwrite gives up LOGGC_LOCK, check again */
707 
708 	if (tblk->flag & tblkGC_COMMITTED) {
709 		if (tblk->flag & tblkGC_ERROR)
710 			rc = -EIO;
711 
712 		LOGGC_UNLOCK(log);
713 		return rc;
714 	}
715 
716 	/* upcount transaction waiting for completion
717 	 */
718 	log->gcrtc++;
719 	tblk->flag |= tblkGC_READY;
720 
721 	__SLEEP_COND(tblk->gcwait, (tblk->flag & tblkGC_COMMITTED),
722 		     LOGGC_LOCK(log), LOGGC_UNLOCK(log));
723 
724 	/* removed from commit queue */
725 	if (tblk->flag & tblkGC_ERROR)
726 		rc = -EIO;
727 
728 	LOGGC_UNLOCK(log);
729 	return rc;
730 }
731 
732 /*
733  * NAME:	lmGCwrite()
734  *
735  * FUNCTION:	group commit write
736  *	initiate write of log page, building a group of all transactions
737  *	with commit records on that page.
738  *
739  * RETURN:	None
740  *
741  * NOTE:
742  *	LOGGC_LOCK must be held by caller.
743  *	N.B. LOG_LOCK is NOT held during lmGroupCommit().
744  */
745 static void lmGCwrite(struct jfs_log * log, int cant_write)
746 {
747 	struct lbuf *bp;
748 	struct logpage *lp;
749 	int gcpn;		/* group commit page number */
750 	struct tblock *tblk;
751 	struct tblock *xtblk = NULL;
752 
753 	/*
754 	 * build the commit group of a log page
755 	 *
756 	 * scan commit queue and make a commit group of all
757 	 * transactions with COMMIT records on the same log page.
758 	 */
759 	/* get the head tblk on the commit queue */
760 	gcpn = list_entry(log->cqueue.next, struct tblock, cqueue)->pn;
761 
762 	list_for_each_entry(tblk, &log->cqueue, cqueue) {
763 		if (tblk->pn != gcpn)
764 			break;
765 
766 		xtblk = tblk;
767 
768 		/* state transition: (QUEUE, READY) -> COMMIT */
769 		tblk->flag |= tblkGC_COMMIT;
770 	}
771 	tblk = xtblk;		/* last tblk of the page */
772 
773 	/*
774 	 * pageout to commit transactions on the log page.
775 	 */
776 	bp = (struct lbuf *) tblk->bp;
777 	lp = (struct logpage *) bp->l_ldata;
778 	/* is page already full ? */
779 	if (tblk->flag & tblkGC_EOP) {
780 		/* mark page to free at end of group commit of the page */
781 		tblk->flag &= ~tblkGC_EOP;
782 		tblk->flag |= tblkGC_FREE;
783 		bp->l_ceor = bp->l_eor;
784 		lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_ceor);
785 		lbmWrite(log, bp, lbmWRITE | lbmRELEASE | lbmGC,
786 			 cant_write);
787 		INCREMENT(lmStat.full_page);
788 	}
789 	/* page is not yet full */
790 	else {
791 		bp->l_ceor = tblk->eor;	/* ? bp->l_ceor = bp->l_eor; */
792 		lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_ceor);
793 		lbmWrite(log, bp, lbmWRITE | lbmGC, cant_write);
794 		INCREMENT(lmStat.partial_page);
795 	}
796 }
797 
798 /*
799  * NAME:	lmPostGC()
800  *
801  * FUNCTION:	group commit post-processing
802  *	Processes transactions after their commit records have been written
803  *	to disk, redriving log I/O if necessary.
804  *
805  * RETURN:	None
806  *
807  * NOTE:
808  *	This routine is called a interrupt time by lbmIODone
809  */
810 static void lmPostGC(struct lbuf * bp)
811 {
812 	unsigned long flags;
813 	struct jfs_log *log = bp->l_log;
814 	struct logpage *lp;
815 	struct tblock *tblk, *temp;
816 
817 	//LOGGC_LOCK(log);
818 	spin_lock_irqsave(&log->gclock, flags);
819 	/*
820 	 * current pageout of group commit completed.
821 	 *
822 	 * remove/wakeup transactions from commit queue who were
823 	 * group committed with the current log page
824 	 */
825 	list_for_each_entry_safe(tblk, temp, &log->cqueue, cqueue) {
826 		if (!(tblk->flag & tblkGC_COMMIT))
827 			break;
828 		/* if transaction was marked GC_COMMIT then
829 		 * it has been shipped in the current pageout
830 		 * and made it to disk - it is committed.
831 		 */
832 
833 		if (bp->l_flag & lbmERROR)
834 			tblk->flag |= tblkGC_ERROR;
835 
836 		/* remove it from the commit queue */
837 		list_del(&tblk->cqueue);
838 		tblk->flag &= ~tblkGC_QUEUE;
839 
840 		if (tblk == log->flush_tblk) {
841 			/* we can stop flushing the log now */
842 			clear_bit(log_FLUSH, &log->flag);
843 			log->flush_tblk = NULL;
844 		}
845 
846 		jfs_info("lmPostGC: tblk = 0x%p, flag = 0x%x", tblk,
847 			 tblk->flag);
848 
849 		if (!(tblk->xflag & COMMIT_FORCE))
850 			/*
851 			 * Hand tblk over to lazy commit thread
852 			 */
853 			txLazyUnlock(tblk);
854 		else {
855 			/* state transition: COMMIT -> COMMITTED */
856 			tblk->flag |= tblkGC_COMMITTED;
857 
858 			if (tblk->flag & tblkGC_READY)
859 				log->gcrtc--;
860 
861 			LOGGC_WAKEUP(tblk);
862 		}
863 
864 		/* was page full before pageout ?
865 		 * (and this is the last tblk bound with the page)
866 		 */
867 		if (tblk->flag & tblkGC_FREE)
868 			lbmFree(bp);
869 		/* did page become full after pageout ?
870 		 * (and this is the last tblk bound with the page)
871 		 */
872 		else if (tblk->flag & tblkGC_EOP) {
873 			/* finalize the page */
874 			lp = (struct logpage *) bp->l_ldata;
875 			bp->l_ceor = bp->l_eor;
876 			lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_eor);
877 			jfs_info("lmPostGC: calling lbmWrite");
878 			lbmWrite(log, bp, lbmWRITE | lbmRELEASE | lbmFREE,
879 				 1);
880 		}
881 
882 	}
883 
884 	/* are there any transactions who have entered lnGroupCommit()
885 	 * (whose COMMITs are after that of the last log page written.
886 	 * They are waiting for new group commit (above at (SLEEP 1))
887 	 * or lazy transactions are on a full (queued) log page,
888 	 * select the latest ready transaction as new group leader and
889 	 * wake her up to lead her group.
890 	 */
891 	if ((!list_empty(&log->cqueue)) &&
892 	    ((log->gcrtc > 0) || (tblk->bp->l_wqnext != NULL) ||
893 	     test_bit(log_FLUSH, &log->flag) || jfs_tlocks_low))
894 		/*
895 		 * Call lmGCwrite with new group leader
896 		 */
897 		lmGCwrite(log, 1);
898 
899 	/* no transaction are ready yet (transactions are only just
900 	 * queued (GC_QUEUE) and not entered for group commit yet).
901 	 * the first transaction entering group commit
902 	 * will elect herself as new group leader.
903 	 */
904 	else
905 		log->cflag &= ~logGC_PAGEOUT;
906 
907 	//LOGGC_UNLOCK(log);
908 	spin_unlock_irqrestore(&log->gclock, flags);
909 	return;
910 }
911 
912 /*
913  * NAME:	lmLogSync()
914  *
915  * FUNCTION:	write log SYNCPT record for specified log
916  *	if new sync address is available
917  *	(normally the case if sync() is executed by back-ground
918  *	process).
919  *	calculate new value of i_nextsync which determines when
920  *	this code is called again.
921  *
922  * PARAMETERS:	log	- log structure
923  *		hard_sync - 1 to force all metadata to be written
924  *
925  * RETURN:	0
926  *
927  * serialization: LOG_LOCK() held on entry/exit
928  */
929 static int lmLogSync(struct jfs_log * log, int hard_sync)
930 {
931 	int logsize;
932 	int written;		/* written since last syncpt */
933 	int free;		/* free space left available */
934 	int delta;		/* additional delta to write normally */
935 	int more;		/* additional write granted */
936 	struct lrd lrd;
937 	int lsn;
938 	struct logsyncblk *lp;
939 	unsigned long flags;
940 
941 	/* push dirty metapages out to disk */
942 	if (hard_sync)
943 		write_special_inodes(log, filemap_fdatawrite);
944 	else
945 		write_special_inodes(log, filemap_flush);
946 
947 	/*
948 	 *	forward syncpt
949 	 */
950 	/* if last sync is same as last syncpt,
951 	 * invoke sync point forward processing to update sync.
952 	 */
953 
954 	if (log->sync == log->syncpt) {
955 		LOGSYNC_LOCK(log, flags);
956 		if (list_empty(&log->synclist))
957 			log->sync = log->lsn;
958 		else {
959 			lp = list_entry(log->synclist.next,
960 					struct logsyncblk, synclist);
961 			log->sync = lp->lsn;
962 		}
963 		LOGSYNC_UNLOCK(log, flags);
964 
965 	}
966 
967 	/* if sync is different from last syncpt,
968 	 * write a SYNCPT record with syncpt = sync.
969 	 * reset syncpt = sync
970 	 */
971 	if (log->sync != log->syncpt) {
972 		lrd.logtid = 0;
973 		lrd.backchain = 0;
974 		lrd.type = cpu_to_le16(LOG_SYNCPT);
975 		lrd.length = 0;
976 		lrd.log.syncpt.sync = cpu_to_le32(log->sync);
977 		lsn = lmWriteRecord(log, NULL, &lrd, NULL);
978 
979 		log->syncpt = log->sync;
980 	} else
981 		lsn = log->lsn;
982 
983 	/*
984 	 *	setup next syncpt trigger (SWAG)
985 	 */
986 	logsize = log->logsize;
987 
988 	logdiff(written, lsn, log);
989 	free = logsize - written;
990 	delta = LOGSYNC_DELTA(logsize);
991 	more = min(free / 2, delta);
992 	if (more < 2 * LOGPSIZE) {
993 		jfs_warn("\n ... Log Wrap ... Log Wrap ... Log Wrap ...\n");
994 		/*
995 		 *	log wrapping
996 		 *
997 		 * option 1 - panic ? No.!
998 		 * option 2 - shutdown file systems
999 		 *	      associated with log ?
1000 		 * option 3 - extend log ?
1001 		 * option 4 - second chance
1002 		 *
1003 		 * mark log wrapped, and continue.
1004 		 * when all active transactions are completed,
1005 		 * mark log valid for recovery.
1006 		 * if crashed during invalid state, log state
1007 		 * implies invalid log, forcing fsck().
1008 		 */
1009 		/* mark log state log wrap in log superblock */
1010 		/* log->state = LOGWRAP; */
1011 
1012 		/* reset sync point computation */
1013 		log->syncpt = log->sync = lsn;
1014 		log->nextsync = delta;
1015 	} else
1016 		/* next syncpt trigger = written + more */
1017 		log->nextsync = written + more;
1018 
1019 	/* if number of bytes written from last sync point is more
1020 	 * than 1/4 of the log size, stop new transactions from
1021 	 * starting until all current transactions are completed
1022 	 * by setting syncbarrier flag.
1023 	 */
1024 	if (!test_bit(log_SYNCBARRIER, &log->flag) &&
1025 	    (written > LOGSYNC_BARRIER(logsize)) && log->active) {
1026 		set_bit(log_SYNCBARRIER, &log->flag);
1027 		jfs_info("log barrier on: lsn=0x%x syncpt=0x%x", lsn,
1028 			 log->syncpt);
1029 		/*
1030 		 * We may have to initiate group commit
1031 		 */
1032 		jfs_flush_journal(log, 0);
1033 	}
1034 
1035 	return lsn;
1036 }
1037 
1038 /*
1039  * NAME:	jfs_syncpt
1040  *
1041  * FUNCTION:	write log SYNCPT record for specified log
1042  *
1043  * PARAMETERS:	log	  - log structure
1044  *		hard_sync - set to 1 to force metadata to be written
1045  */
1046 void jfs_syncpt(struct jfs_log *log, int hard_sync)
1047 {	LOG_LOCK(log);
1048 	if (!test_bit(log_QUIESCE, &log->flag))
1049 		lmLogSync(log, hard_sync);
1050 	LOG_UNLOCK(log);
1051 }
1052 
1053 /*
1054  * NAME:	lmLogOpen()
1055  *
1056  * FUNCTION:	open the log on first open;
1057  *	insert filesystem in the active list of the log.
1058  *
1059  * PARAMETER:	ipmnt	- file system mount inode
1060  *		iplog	- log inode (out)
1061  *
1062  * RETURN:
1063  *
1064  * serialization:
1065  */
1066 int lmLogOpen(struct super_block *sb)
1067 {
1068 	int rc;
1069 	struct block_device *bdev;
1070 	struct jfs_log *log;
1071 	struct jfs_sb_info *sbi = JFS_SBI(sb);
1072 
1073 	if (sbi->flag & JFS_NOINTEGRITY)
1074 		return open_dummy_log(sb);
1075 
1076 	if (sbi->mntflag & JFS_INLINELOG)
1077 		return open_inline_log(sb);
1078 
1079 	mutex_lock(&jfs_log_mutex);
1080 	list_for_each_entry(log, &jfs_external_logs, journal_list) {
1081 		if (log->bdev->bd_dev == sbi->logdev) {
1082 			if (!uuid_equal(&log->uuid, &sbi->loguuid)) {
1083 				jfs_warn("wrong uuid on JFS journal");
1084 				mutex_unlock(&jfs_log_mutex);
1085 				return -EINVAL;
1086 			}
1087 			/*
1088 			 * add file system to log active file system list
1089 			 */
1090 			if ((rc = lmLogFileSystem(log, sbi, 1))) {
1091 				mutex_unlock(&jfs_log_mutex);
1092 				return rc;
1093 			}
1094 			goto journal_found;
1095 		}
1096 	}
1097 
1098 	if (!(log = kzalloc(sizeof(struct jfs_log), GFP_KERNEL))) {
1099 		mutex_unlock(&jfs_log_mutex);
1100 		return -ENOMEM;
1101 	}
1102 	INIT_LIST_HEAD(&log->sb_list);
1103 	init_waitqueue_head(&log->syncwait);
1104 
1105 	/*
1106 	 *	external log as separate logical volume
1107 	 *
1108 	 * file systems to log may have n-to-1 relationship;
1109 	 */
1110 
1111 	bdev = blkdev_get_by_dev(sbi->logdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
1112 				 log);
1113 	if (IS_ERR(bdev)) {
1114 		rc = PTR_ERR(bdev);
1115 		goto free;
1116 	}
1117 
1118 	log->bdev = bdev;
1119 	uuid_copy(&log->uuid, &sbi->loguuid);
1120 
1121 	/*
1122 	 * initialize log:
1123 	 */
1124 	if ((rc = lmLogInit(log)))
1125 		goto close;
1126 
1127 	list_add(&log->journal_list, &jfs_external_logs);
1128 
1129 	/*
1130 	 * add file system to log active file system list
1131 	 */
1132 	if ((rc = lmLogFileSystem(log, sbi, 1)))
1133 		goto shutdown;
1134 
1135 journal_found:
1136 	LOG_LOCK(log);
1137 	list_add(&sbi->log_list, &log->sb_list);
1138 	sbi->log = log;
1139 	LOG_UNLOCK(log);
1140 
1141 	mutex_unlock(&jfs_log_mutex);
1142 	return 0;
1143 
1144 	/*
1145 	 *	unwind on error
1146 	 */
1147       shutdown:		/* unwind lbmLogInit() */
1148 	list_del(&log->journal_list);
1149 	lbmLogShutdown(log);
1150 
1151       close:		/* close external log device */
1152 	blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1153 
1154       free:		/* free log descriptor */
1155 	mutex_unlock(&jfs_log_mutex);
1156 	kfree(log);
1157 
1158 	jfs_warn("lmLogOpen: exit(%d)", rc);
1159 	return rc;
1160 }
1161 
1162 static int open_inline_log(struct super_block *sb)
1163 {
1164 	struct jfs_log *log;
1165 	int rc;
1166 
1167 	if (!(log = kzalloc(sizeof(struct jfs_log), GFP_KERNEL)))
1168 		return -ENOMEM;
1169 	INIT_LIST_HEAD(&log->sb_list);
1170 	init_waitqueue_head(&log->syncwait);
1171 
1172 	set_bit(log_INLINELOG, &log->flag);
1173 	log->bdev = sb->s_bdev;
1174 	log->base = addressPXD(&JFS_SBI(sb)->logpxd);
1175 	log->size = lengthPXD(&JFS_SBI(sb)->logpxd) >>
1176 	    (L2LOGPSIZE - sb->s_blocksize_bits);
1177 	log->l2bsize = sb->s_blocksize_bits;
1178 	ASSERT(L2LOGPSIZE >= sb->s_blocksize_bits);
1179 
1180 	/*
1181 	 * initialize log.
1182 	 */
1183 	if ((rc = lmLogInit(log))) {
1184 		kfree(log);
1185 		jfs_warn("lmLogOpen: exit(%d)", rc);
1186 		return rc;
1187 	}
1188 
1189 	list_add(&JFS_SBI(sb)->log_list, &log->sb_list);
1190 	JFS_SBI(sb)->log = log;
1191 
1192 	return rc;
1193 }
1194 
1195 static int open_dummy_log(struct super_block *sb)
1196 {
1197 	int rc;
1198 
1199 	mutex_lock(&jfs_log_mutex);
1200 	if (!dummy_log) {
1201 		dummy_log = kzalloc(sizeof(struct jfs_log), GFP_KERNEL);
1202 		if (!dummy_log) {
1203 			mutex_unlock(&jfs_log_mutex);
1204 			return -ENOMEM;
1205 		}
1206 		INIT_LIST_HEAD(&dummy_log->sb_list);
1207 		init_waitqueue_head(&dummy_log->syncwait);
1208 		dummy_log->no_integrity = 1;
1209 		/* Make up some stuff */
1210 		dummy_log->base = 0;
1211 		dummy_log->size = 1024;
1212 		rc = lmLogInit(dummy_log);
1213 		if (rc) {
1214 			kfree(dummy_log);
1215 			dummy_log = NULL;
1216 			mutex_unlock(&jfs_log_mutex);
1217 			return rc;
1218 		}
1219 	}
1220 
1221 	LOG_LOCK(dummy_log);
1222 	list_add(&JFS_SBI(sb)->log_list, &dummy_log->sb_list);
1223 	JFS_SBI(sb)->log = dummy_log;
1224 	LOG_UNLOCK(dummy_log);
1225 	mutex_unlock(&jfs_log_mutex);
1226 
1227 	return 0;
1228 }
1229 
1230 /*
1231  * NAME:	lmLogInit()
1232  *
1233  * FUNCTION:	log initialization at first log open.
1234  *
1235  *	logredo() (or logformat()) should have been run previously.
1236  *	initialize the log from log superblock.
1237  *	set the log state in the superblock to LOGMOUNT and
1238  *	write SYNCPT log record.
1239  *
1240  * PARAMETER:	log	- log structure
1241  *
1242  * RETURN:	0	- if ok
1243  *		-EINVAL	- bad log magic number or superblock dirty
1244  *		error returned from logwait()
1245  *
1246  * serialization: single first open thread
1247  */
1248 int lmLogInit(struct jfs_log * log)
1249 {
1250 	int rc = 0;
1251 	struct lrd lrd;
1252 	struct logsuper *logsuper;
1253 	struct lbuf *bpsuper;
1254 	struct lbuf *bp;
1255 	struct logpage *lp;
1256 	int lsn = 0;
1257 
1258 	jfs_info("lmLogInit: log:0x%p", log);
1259 
1260 	/* initialize the group commit serialization lock */
1261 	LOGGC_LOCK_INIT(log);
1262 
1263 	/* allocate/initialize the log write serialization lock */
1264 	LOG_LOCK_INIT(log);
1265 
1266 	LOGSYNC_LOCK_INIT(log);
1267 
1268 	INIT_LIST_HEAD(&log->synclist);
1269 
1270 	INIT_LIST_HEAD(&log->cqueue);
1271 	log->flush_tblk = NULL;
1272 
1273 	log->count = 0;
1274 
1275 	/*
1276 	 * initialize log i/o
1277 	 */
1278 	if ((rc = lbmLogInit(log)))
1279 		return rc;
1280 
1281 	if (!test_bit(log_INLINELOG, &log->flag))
1282 		log->l2bsize = L2LOGPSIZE;
1283 
1284 	/* check for disabled journaling to disk */
1285 	if (log->no_integrity) {
1286 		/*
1287 		 * Journal pages will still be filled.  When the time comes
1288 		 * to actually do the I/O, the write is not done, and the
1289 		 * endio routine is called directly.
1290 		 */
1291 		bp = lbmAllocate(log , 0);
1292 		log->bp = bp;
1293 		bp->l_pn = bp->l_eor = 0;
1294 	} else {
1295 		/*
1296 		 * validate log superblock
1297 		 */
1298 		if ((rc = lbmRead(log, 1, &bpsuper)))
1299 			goto errout10;
1300 
1301 		logsuper = (struct logsuper *) bpsuper->l_ldata;
1302 
1303 		if (logsuper->magic != cpu_to_le32(LOGMAGIC)) {
1304 			jfs_warn("*** Log Format Error ! ***");
1305 			rc = -EINVAL;
1306 			goto errout20;
1307 		}
1308 
1309 		/* logredo() should have been run successfully. */
1310 		if (logsuper->state != cpu_to_le32(LOGREDONE)) {
1311 			jfs_warn("*** Log Is Dirty ! ***");
1312 			rc = -EINVAL;
1313 			goto errout20;
1314 		}
1315 
1316 		/* initialize log from log superblock */
1317 		if (test_bit(log_INLINELOG,&log->flag)) {
1318 			if (log->size != le32_to_cpu(logsuper->size)) {
1319 				rc = -EINVAL;
1320 				goto errout20;
1321 			}
1322 			jfs_info("lmLogInit: inline log:0x%p base:0x%Lx size:0x%x",
1323 				 log, (unsigned long long)log->base, log->size);
1324 		} else {
1325 			if (!uuid_equal(&logsuper->uuid, &log->uuid)) {
1326 				jfs_warn("wrong uuid on JFS log device");
1327 				rc = -EINVAL;
1328 				goto errout20;
1329 			}
1330 			log->size = le32_to_cpu(logsuper->size);
1331 			log->l2bsize = le32_to_cpu(logsuper->l2bsize);
1332 			jfs_info("lmLogInit: external log:0x%p base:0x%Lx size:0x%x",
1333 				 log, (unsigned long long)log->base, log->size);
1334 		}
1335 
1336 		log->page = le32_to_cpu(logsuper->end) / LOGPSIZE;
1337 		log->eor = le32_to_cpu(logsuper->end) - (LOGPSIZE * log->page);
1338 
1339 		/*
1340 		 * initialize for log append write mode
1341 		 */
1342 		/* establish current/end-of-log page/buffer */
1343 		if ((rc = lbmRead(log, log->page, &bp)))
1344 			goto errout20;
1345 
1346 		lp = (struct logpage *) bp->l_ldata;
1347 
1348 		jfs_info("lmLogInit: lsn:0x%x page:%d eor:%d:%d",
1349 			 le32_to_cpu(logsuper->end), log->page, log->eor,
1350 			 le16_to_cpu(lp->h.eor));
1351 
1352 		log->bp = bp;
1353 		bp->l_pn = log->page;
1354 		bp->l_eor = log->eor;
1355 
1356 		/* if current page is full, move on to next page */
1357 		if (log->eor >= LOGPSIZE - LOGPTLRSIZE)
1358 			lmNextPage(log);
1359 
1360 		/*
1361 		 * initialize log syncpoint
1362 		 */
1363 		/*
1364 		 * write the first SYNCPT record with syncpoint = 0
1365 		 * (i.e., log redo up to HERE !);
1366 		 * remove current page from lbm write queue at end of pageout
1367 		 * (to write log superblock update), but do not release to
1368 		 * freelist;
1369 		 */
1370 		lrd.logtid = 0;
1371 		lrd.backchain = 0;
1372 		lrd.type = cpu_to_le16(LOG_SYNCPT);
1373 		lrd.length = 0;
1374 		lrd.log.syncpt.sync = 0;
1375 		lsn = lmWriteRecord(log, NULL, &lrd, NULL);
1376 		bp = log->bp;
1377 		bp->l_ceor = bp->l_eor;
1378 		lp = (struct logpage *) bp->l_ldata;
1379 		lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_eor);
1380 		lbmWrite(log, bp, lbmWRITE | lbmSYNC, 0);
1381 		if ((rc = lbmIOWait(bp, 0)))
1382 			goto errout30;
1383 
1384 		/*
1385 		 * update/write superblock
1386 		 */
1387 		logsuper->state = cpu_to_le32(LOGMOUNT);
1388 		log->serial = le32_to_cpu(logsuper->serial) + 1;
1389 		logsuper->serial = cpu_to_le32(log->serial);
1390 		lbmDirectWrite(log, bpsuper, lbmWRITE | lbmRELEASE | lbmSYNC);
1391 		if ((rc = lbmIOWait(bpsuper, lbmFREE)))
1392 			goto errout30;
1393 	}
1394 
1395 	/* initialize logsync parameters */
1396 	log->logsize = (log->size - 2) << L2LOGPSIZE;
1397 	log->lsn = lsn;
1398 	log->syncpt = lsn;
1399 	log->sync = log->syncpt;
1400 	log->nextsync = LOGSYNC_DELTA(log->logsize);
1401 
1402 	jfs_info("lmLogInit: lsn:0x%x syncpt:0x%x sync:0x%x",
1403 		 log->lsn, log->syncpt, log->sync);
1404 
1405 	/*
1406 	 * initialize for lazy/group commit
1407 	 */
1408 	log->clsn = lsn;
1409 
1410 	return 0;
1411 
1412 	/*
1413 	 *	unwind on error
1414 	 */
1415       errout30:		/* release log page */
1416 	log->wqueue = NULL;
1417 	bp->l_wqnext = NULL;
1418 	lbmFree(bp);
1419 
1420       errout20:		/* release log superblock */
1421 	lbmFree(bpsuper);
1422 
1423       errout10:		/* unwind lbmLogInit() */
1424 	lbmLogShutdown(log);
1425 
1426 	jfs_warn("lmLogInit: exit(%d)", rc);
1427 	return rc;
1428 }
1429 
1430 
1431 /*
1432  * NAME:	lmLogClose()
1433  *
1434  * FUNCTION:	remove file system <ipmnt> from active list of log <iplog>
1435  *		and close it on last close.
1436  *
1437  * PARAMETER:	sb	- superblock
1438  *
1439  * RETURN:	errors from subroutines
1440  *
1441  * serialization:
1442  */
1443 int lmLogClose(struct super_block *sb)
1444 {
1445 	struct jfs_sb_info *sbi = JFS_SBI(sb);
1446 	struct jfs_log *log = sbi->log;
1447 	struct block_device *bdev;
1448 	int rc = 0;
1449 
1450 	jfs_info("lmLogClose: log:0x%p", log);
1451 
1452 	mutex_lock(&jfs_log_mutex);
1453 	LOG_LOCK(log);
1454 	list_del(&sbi->log_list);
1455 	LOG_UNLOCK(log);
1456 	sbi->log = NULL;
1457 
1458 	/*
1459 	 * We need to make sure all of the "written" metapages
1460 	 * actually make it to disk
1461 	 */
1462 	sync_blockdev(sb->s_bdev);
1463 
1464 	if (test_bit(log_INLINELOG, &log->flag)) {
1465 		/*
1466 		 *	in-line log in host file system
1467 		 */
1468 		rc = lmLogShutdown(log);
1469 		kfree(log);
1470 		goto out;
1471 	}
1472 
1473 	if (!log->no_integrity)
1474 		lmLogFileSystem(log, sbi, 0);
1475 
1476 	if (!list_empty(&log->sb_list))
1477 		goto out;
1478 
1479 	/*
1480 	 * TODO: ensure that the dummy_log is in a state to allow
1481 	 * lbmLogShutdown to deallocate all the buffers and call
1482 	 * kfree against dummy_log.  For now, leave dummy_log & its
1483 	 * buffers in memory, and resuse if another no-integrity mount
1484 	 * is requested.
1485 	 */
1486 	if (log->no_integrity)
1487 		goto out;
1488 
1489 	/*
1490 	 *	external log as separate logical volume
1491 	 */
1492 	list_del(&log->journal_list);
1493 	bdev = log->bdev;
1494 	rc = lmLogShutdown(log);
1495 
1496 	blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1497 
1498 	kfree(log);
1499 
1500       out:
1501 	mutex_unlock(&jfs_log_mutex);
1502 	jfs_info("lmLogClose: exit(%d)", rc);
1503 	return rc;
1504 }
1505 
1506 
1507 /*
1508  * NAME:	jfs_flush_journal()
1509  *
1510  * FUNCTION:	initiate write of any outstanding transactions to the journal
1511  *		and optionally wait until they are all written to disk
1512  *
1513  *		wait == 0  flush until latest txn is committed, don't wait
1514  *		wait == 1  flush until latest txn is committed, wait
1515  *		wait > 1   flush until all txn's are complete, wait
1516  */
1517 void jfs_flush_journal(struct jfs_log *log, int wait)
1518 {
1519 	int i;
1520 	struct tblock *target = NULL;
1521 
1522 	/* jfs_write_inode may call us during read-only mount */
1523 	if (!log)
1524 		return;
1525 
1526 	jfs_info("jfs_flush_journal: log:0x%p wait=%d", log, wait);
1527 
1528 	LOGGC_LOCK(log);
1529 
1530 	if (!list_empty(&log->cqueue)) {
1531 		/*
1532 		 * This ensures that we will keep writing to the journal as long
1533 		 * as there are unwritten commit records
1534 		 */
1535 		target = list_entry(log->cqueue.prev, struct tblock, cqueue);
1536 
1537 		if (test_bit(log_FLUSH, &log->flag)) {
1538 			/*
1539 			 * We're already flushing.
1540 			 * if flush_tblk is NULL, we are flushing everything,
1541 			 * so leave it that way.  Otherwise, update it to the
1542 			 * latest transaction
1543 			 */
1544 			if (log->flush_tblk)
1545 				log->flush_tblk = target;
1546 		} else {
1547 			/* Only flush until latest transaction is committed */
1548 			log->flush_tblk = target;
1549 			set_bit(log_FLUSH, &log->flag);
1550 
1551 			/*
1552 			 * Initiate I/O on outstanding transactions
1553 			 */
1554 			if (!(log->cflag & logGC_PAGEOUT)) {
1555 				log->cflag |= logGC_PAGEOUT;
1556 				lmGCwrite(log, 0);
1557 			}
1558 		}
1559 	}
1560 	if ((wait > 1) || test_bit(log_SYNCBARRIER, &log->flag)) {
1561 		/* Flush until all activity complete */
1562 		set_bit(log_FLUSH, &log->flag);
1563 		log->flush_tblk = NULL;
1564 	}
1565 
1566 	if (wait && target && !(target->flag & tblkGC_COMMITTED)) {
1567 		DECLARE_WAITQUEUE(__wait, current);
1568 
1569 		add_wait_queue(&target->gcwait, &__wait);
1570 		set_current_state(TASK_UNINTERRUPTIBLE);
1571 		LOGGC_UNLOCK(log);
1572 		schedule();
1573 		LOGGC_LOCK(log);
1574 		remove_wait_queue(&target->gcwait, &__wait);
1575 	}
1576 	LOGGC_UNLOCK(log);
1577 
1578 	if (wait < 2)
1579 		return;
1580 
1581 	write_special_inodes(log, filemap_fdatawrite);
1582 
1583 	/*
1584 	 * If there was recent activity, we may need to wait
1585 	 * for the lazycommit thread to catch up
1586 	 */
1587 	if ((!list_empty(&log->cqueue)) || !list_empty(&log->synclist)) {
1588 		for (i = 0; i < 200; i++) {	/* Too much? */
1589 			msleep(250);
1590 			write_special_inodes(log, filemap_fdatawrite);
1591 			if (list_empty(&log->cqueue) &&
1592 			    list_empty(&log->synclist))
1593 				break;
1594 		}
1595 	}
1596 	assert(list_empty(&log->cqueue));
1597 
1598 #ifdef CONFIG_JFS_DEBUG
1599 	if (!list_empty(&log->synclist)) {
1600 		struct logsyncblk *lp;
1601 
1602 		printk(KERN_ERR "jfs_flush_journal: synclist not empty\n");
1603 		list_for_each_entry(lp, &log->synclist, synclist) {
1604 			if (lp->xflag & COMMIT_PAGE) {
1605 				struct metapage *mp = (struct metapage *)lp;
1606 				print_hex_dump(KERN_ERR, "metapage: ",
1607 					       DUMP_PREFIX_ADDRESS, 16, 4,
1608 					       mp, sizeof(struct metapage), 0);
1609 				print_hex_dump(KERN_ERR, "page: ",
1610 					       DUMP_PREFIX_ADDRESS, 16,
1611 					       sizeof(long), mp->page,
1612 					       sizeof(struct page), 0);
1613 			} else
1614 				print_hex_dump(KERN_ERR, "tblock:",
1615 					       DUMP_PREFIX_ADDRESS, 16, 4,
1616 					       lp, sizeof(struct tblock), 0);
1617 		}
1618 	}
1619 #else
1620 	WARN_ON(!list_empty(&log->synclist));
1621 #endif
1622 	clear_bit(log_FLUSH, &log->flag);
1623 }
1624 
1625 /*
1626  * NAME:	lmLogShutdown()
1627  *
1628  * FUNCTION:	log shutdown at last LogClose().
1629  *
1630  *		write log syncpt record.
1631  *		update super block to set redone flag to 0.
1632  *
1633  * PARAMETER:	log	- log inode
1634  *
1635  * RETURN:	0	- success
1636  *
1637  * serialization: single last close thread
1638  */
1639 int lmLogShutdown(struct jfs_log * log)
1640 {
1641 	int rc;
1642 	struct lrd lrd;
1643 	int lsn;
1644 	struct logsuper *logsuper;
1645 	struct lbuf *bpsuper;
1646 	struct lbuf *bp;
1647 	struct logpage *lp;
1648 
1649 	jfs_info("lmLogShutdown: log:0x%p", log);
1650 
1651 	jfs_flush_journal(log, 2);
1652 
1653 	/*
1654 	 * write the last SYNCPT record with syncpoint = 0
1655 	 * (i.e., log redo up to HERE !)
1656 	 */
1657 	lrd.logtid = 0;
1658 	lrd.backchain = 0;
1659 	lrd.type = cpu_to_le16(LOG_SYNCPT);
1660 	lrd.length = 0;
1661 	lrd.log.syncpt.sync = 0;
1662 
1663 	lsn = lmWriteRecord(log, NULL, &lrd, NULL);
1664 	bp = log->bp;
1665 	lp = (struct logpage *) bp->l_ldata;
1666 	lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_eor);
1667 	lbmWrite(log, log->bp, lbmWRITE | lbmRELEASE | lbmSYNC, 0);
1668 	lbmIOWait(log->bp, lbmFREE);
1669 	log->bp = NULL;
1670 
1671 	/*
1672 	 * synchronous update log superblock
1673 	 * mark log state as shutdown cleanly
1674 	 * (i.e., Log does not need to be replayed).
1675 	 */
1676 	if ((rc = lbmRead(log, 1, &bpsuper)))
1677 		goto out;
1678 
1679 	logsuper = (struct logsuper *) bpsuper->l_ldata;
1680 	logsuper->state = cpu_to_le32(LOGREDONE);
1681 	logsuper->end = cpu_to_le32(lsn);
1682 	lbmDirectWrite(log, bpsuper, lbmWRITE | lbmRELEASE | lbmSYNC);
1683 	rc = lbmIOWait(bpsuper, lbmFREE);
1684 
1685 	jfs_info("lmLogShutdown: lsn:0x%x page:%d eor:%d",
1686 		 lsn, log->page, log->eor);
1687 
1688       out:
1689 	/*
1690 	 * shutdown per log i/o
1691 	 */
1692 	lbmLogShutdown(log);
1693 
1694 	if (rc) {
1695 		jfs_warn("lmLogShutdown: exit(%d)", rc);
1696 	}
1697 	return rc;
1698 }
1699 
1700 
1701 /*
1702  * NAME:	lmLogFileSystem()
1703  *
1704  * FUNCTION:	insert (<activate> = true)/remove (<activate> = false)
1705  *	file system into/from log active file system list.
1706  *
1707  * PARAMETE:	log	- pointer to logs inode.
1708  *		fsdev	- kdev_t of filesystem.
1709  *		serial	- pointer to returned log serial number
1710  *		activate - insert/remove device from active list.
1711  *
1712  * RETURN:	0	- success
1713  *		errors returned by vms_iowait().
1714  */
1715 static int lmLogFileSystem(struct jfs_log * log, struct jfs_sb_info *sbi,
1716 			   int activate)
1717 {
1718 	int rc = 0;
1719 	int i;
1720 	struct logsuper *logsuper;
1721 	struct lbuf *bpsuper;
1722 	uuid_t *uuid = &sbi->uuid;
1723 
1724 	/*
1725 	 * insert/remove file system device to log active file system list.
1726 	 */
1727 	if ((rc = lbmRead(log, 1, &bpsuper)))
1728 		return rc;
1729 
1730 	logsuper = (struct logsuper *) bpsuper->l_ldata;
1731 	if (activate) {
1732 		for (i = 0; i < MAX_ACTIVE; i++)
1733 			if (uuid_is_null(&logsuper->active[i].uuid)) {
1734 				uuid_copy(&logsuper->active[i].uuid, uuid);
1735 				sbi->aggregate = i;
1736 				break;
1737 			}
1738 		if (i == MAX_ACTIVE) {
1739 			jfs_warn("Too many file systems sharing journal!");
1740 			lbmFree(bpsuper);
1741 			return -EMFILE;	/* Is there a better rc? */
1742 		}
1743 	} else {
1744 		for (i = 0; i < MAX_ACTIVE; i++)
1745 			if (uuid_equal(&logsuper->active[i].uuid, uuid)) {
1746 				uuid_copy(&logsuper->active[i].uuid,
1747 					  &uuid_null);
1748 				break;
1749 			}
1750 		if (i == MAX_ACTIVE) {
1751 			jfs_warn("Somebody stomped on the journal!");
1752 			lbmFree(bpsuper);
1753 			return -EIO;
1754 		}
1755 
1756 	}
1757 
1758 	/*
1759 	 * synchronous write log superblock:
1760 	 *
1761 	 * write sidestream bypassing write queue:
1762 	 * at file system mount, log super block is updated for
1763 	 * activation of the file system before any log record
1764 	 * (MOUNT record) of the file system, and at file system
1765 	 * unmount, all meta data for the file system has been
1766 	 * flushed before log super block is updated for deactivation
1767 	 * of the file system.
1768 	 */
1769 	lbmDirectWrite(log, bpsuper, lbmWRITE | lbmRELEASE | lbmSYNC);
1770 	rc = lbmIOWait(bpsuper, lbmFREE);
1771 
1772 	return rc;
1773 }
1774 
1775 /*
1776  *		log buffer manager (lbm)
1777  *		------------------------
1778  *
1779  * special purpose buffer manager supporting log i/o requirements.
1780  *
1781  * per log write queue:
1782  * log pageout occurs in serial order by fifo write queue and
1783  * restricting to a single i/o in pregress at any one time.
1784  * a circular singly-linked list
1785  * (log->wrqueue points to the tail, and buffers are linked via
1786  * bp->wrqueue field), and
1787  * maintains log page in pageout ot waiting for pageout in serial pageout.
1788  */
1789 
1790 /*
1791  *	lbmLogInit()
1792  *
1793  * initialize per log I/O setup at lmLogInit()
1794  */
1795 static int lbmLogInit(struct jfs_log * log)
1796 {				/* log inode */
1797 	int i;
1798 	struct lbuf *lbuf;
1799 
1800 	jfs_info("lbmLogInit: log:0x%p", log);
1801 
1802 	/* initialize current buffer cursor */
1803 	log->bp = NULL;
1804 
1805 	/* initialize log device write queue */
1806 	log->wqueue = NULL;
1807 
1808 	/*
1809 	 * Each log has its own buffer pages allocated to it.  These are
1810 	 * not managed by the page cache.  This ensures that a transaction
1811 	 * writing to the log does not block trying to allocate a page from
1812 	 * the page cache (for the log).  This would be bad, since page
1813 	 * allocation waits on the kswapd thread that may be committing inodes
1814 	 * which would cause log activity.  Was that clear?  I'm trying to
1815 	 * avoid deadlock here.
1816 	 */
1817 	init_waitqueue_head(&log->free_wait);
1818 
1819 	log->lbuf_free = NULL;
1820 
1821 	for (i = 0; i < LOGPAGES;) {
1822 		char *buffer;
1823 		uint offset;
1824 		struct page *page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1825 
1826 		if (!page)
1827 			goto error;
1828 		buffer = page_address(page);
1829 		for (offset = 0; offset < PAGE_SIZE; offset += LOGPSIZE) {
1830 			lbuf = kmalloc(sizeof(struct lbuf), GFP_KERNEL);
1831 			if (lbuf == NULL) {
1832 				if (offset == 0)
1833 					__free_page(page);
1834 				goto error;
1835 			}
1836 			if (offset) /* we already have one reference */
1837 				get_page(page);
1838 			lbuf->l_offset = offset;
1839 			lbuf->l_ldata = buffer + offset;
1840 			lbuf->l_page = page;
1841 			lbuf->l_log = log;
1842 			init_waitqueue_head(&lbuf->l_ioevent);
1843 
1844 			lbuf->l_freelist = log->lbuf_free;
1845 			log->lbuf_free = lbuf;
1846 			i++;
1847 		}
1848 	}
1849 
1850 	return (0);
1851 
1852       error:
1853 	lbmLogShutdown(log);
1854 	return -ENOMEM;
1855 }
1856 
1857 
1858 /*
1859  *	lbmLogShutdown()
1860  *
1861  * finalize per log I/O setup at lmLogShutdown()
1862  */
1863 static void lbmLogShutdown(struct jfs_log * log)
1864 {
1865 	struct lbuf *lbuf;
1866 
1867 	jfs_info("lbmLogShutdown: log:0x%p", log);
1868 
1869 	lbuf = log->lbuf_free;
1870 	while (lbuf) {
1871 		struct lbuf *next = lbuf->l_freelist;
1872 		__free_page(lbuf->l_page);
1873 		kfree(lbuf);
1874 		lbuf = next;
1875 	}
1876 }
1877 
1878 
1879 /*
1880  *	lbmAllocate()
1881  *
1882  * allocate an empty log buffer
1883  */
1884 static struct lbuf *lbmAllocate(struct jfs_log * log, int pn)
1885 {
1886 	struct lbuf *bp;
1887 	unsigned long flags;
1888 
1889 	/*
1890 	 * recycle from log buffer freelist if any
1891 	 */
1892 	LCACHE_LOCK(flags);
1893 	LCACHE_SLEEP_COND(log->free_wait, (bp = log->lbuf_free), flags);
1894 	log->lbuf_free = bp->l_freelist;
1895 	LCACHE_UNLOCK(flags);
1896 
1897 	bp->l_flag = 0;
1898 
1899 	bp->l_wqnext = NULL;
1900 	bp->l_freelist = NULL;
1901 
1902 	bp->l_pn = pn;
1903 	bp->l_blkno = log->base + (pn << (L2LOGPSIZE - log->l2bsize));
1904 	bp->l_ceor = 0;
1905 
1906 	return bp;
1907 }
1908 
1909 
1910 /*
1911  *	lbmFree()
1912  *
1913  * release a log buffer to freelist
1914  */
1915 static void lbmFree(struct lbuf * bp)
1916 {
1917 	unsigned long flags;
1918 
1919 	LCACHE_LOCK(flags);
1920 
1921 	lbmfree(bp);
1922 
1923 	LCACHE_UNLOCK(flags);
1924 }
1925 
1926 static void lbmfree(struct lbuf * bp)
1927 {
1928 	struct jfs_log *log = bp->l_log;
1929 
1930 	assert(bp->l_wqnext == NULL);
1931 
1932 	/*
1933 	 * return the buffer to head of freelist
1934 	 */
1935 	bp->l_freelist = log->lbuf_free;
1936 	log->lbuf_free = bp;
1937 
1938 	wake_up(&log->free_wait);
1939 	return;
1940 }
1941 
1942 
1943 /*
1944  * NAME:	lbmRedrive
1945  *
1946  * FUNCTION:	add a log buffer to the log redrive list
1947  *
1948  * PARAMETER:
1949  *	bp	- log buffer
1950  *
1951  * NOTES:
1952  *	Takes log_redrive_lock.
1953  */
1954 static inline void lbmRedrive(struct lbuf *bp)
1955 {
1956 	unsigned long flags;
1957 
1958 	spin_lock_irqsave(&log_redrive_lock, flags);
1959 	bp->l_redrive_next = log_redrive_list;
1960 	log_redrive_list = bp;
1961 	spin_unlock_irqrestore(&log_redrive_lock, flags);
1962 
1963 	wake_up_process(jfsIOthread);
1964 }
1965 
1966 
1967 /*
1968  *	lbmRead()
1969  */
1970 static int lbmRead(struct jfs_log * log, int pn, struct lbuf ** bpp)
1971 {
1972 	struct bio *bio;
1973 	struct lbuf *bp;
1974 
1975 	/*
1976 	 * allocate a log buffer
1977 	 */
1978 	*bpp = bp = lbmAllocate(log, pn);
1979 	jfs_info("lbmRead: bp:0x%p pn:0x%x", bp, pn);
1980 
1981 	bp->l_flag |= lbmREAD;
1982 
1983 	bio = bio_alloc(log->bdev, 1, REQ_OP_READ, GFP_NOFS);
1984 	bio->bi_iter.bi_sector = bp->l_blkno << (log->l2bsize - 9);
1985 	bio_add_page(bio, bp->l_page, LOGPSIZE, bp->l_offset);
1986 	BUG_ON(bio->bi_iter.bi_size != LOGPSIZE);
1987 
1988 	bio->bi_end_io = lbmIODone;
1989 	bio->bi_private = bp;
1990 	/*check if journaling to disk has been disabled*/
1991 	if (log->no_integrity) {
1992 		bio->bi_iter.bi_size = 0;
1993 		lbmIODone(bio);
1994 	} else {
1995 		submit_bio(bio);
1996 	}
1997 
1998 	wait_event(bp->l_ioevent, (bp->l_flag != lbmREAD));
1999 
2000 	return 0;
2001 }
2002 
2003 
2004 /*
2005  *	lbmWrite()
2006  *
2007  * buffer at head of pageout queue stays after completion of
2008  * partial-page pageout and redriven by explicit initiation of
2009  * pageout by caller until full-page pageout is completed and
2010  * released.
2011  *
2012  * device driver i/o done redrives pageout of new buffer at
2013  * head of pageout queue when current buffer at head of pageout
2014  * queue is released at the completion of its full-page pageout.
2015  *
2016  * LOGGC_LOCK() serializes lbmWrite() by lmNextPage() and lmGroupCommit().
2017  * LCACHE_LOCK() serializes xflag between lbmWrite() and lbmIODone()
2018  */
2019 static void lbmWrite(struct jfs_log * log, struct lbuf * bp, int flag,
2020 		     int cant_block)
2021 {
2022 	struct lbuf *tail;
2023 	unsigned long flags;
2024 
2025 	jfs_info("lbmWrite: bp:0x%p flag:0x%x pn:0x%x", bp, flag, bp->l_pn);
2026 
2027 	/* map the logical block address to physical block address */
2028 	bp->l_blkno =
2029 	    log->base + (bp->l_pn << (L2LOGPSIZE - log->l2bsize));
2030 
2031 	LCACHE_LOCK(flags);		/* disable+lock */
2032 
2033 	/*
2034 	 * initialize buffer for device driver
2035 	 */
2036 	bp->l_flag = flag;
2037 
2038 	/*
2039 	 *	insert bp at tail of write queue associated with log
2040 	 *
2041 	 * (request is either for bp already/currently at head of queue
2042 	 * or new bp to be inserted at tail)
2043 	 */
2044 	tail = log->wqueue;
2045 
2046 	/* is buffer not already on write queue ? */
2047 	if (bp->l_wqnext == NULL) {
2048 		/* insert at tail of wqueue */
2049 		if (tail == NULL) {
2050 			log->wqueue = bp;
2051 			bp->l_wqnext = bp;
2052 		} else {
2053 			log->wqueue = bp;
2054 			bp->l_wqnext = tail->l_wqnext;
2055 			tail->l_wqnext = bp;
2056 		}
2057 
2058 		tail = bp;
2059 	}
2060 
2061 	/* is buffer at head of wqueue and for write ? */
2062 	if ((bp != tail->l_wqnext) || !(flag & lbmWRITE)) {
2063 		LCACHE_UNLOCK(flags);	/* unlock+enable */
2064 		return;
2065 	}
2066 
2067 	LCACHE_UNLOCK(flags);	/* unlock+enable */
2068 
2069 	if (cant_block)
2070 		lbmRedrive(bp);
2071 	else if (flag & lbmSYNC)
2072 		lbmStartIO(bp);
2073 	else {
2074 		LOGGC_UNLOCK(log);
2075 		lbmStartIO(bp);
2076 		LOGGC_LOCK(log);
2077 	}
2078 }
2079 
2080 
2081 /*
2082  *	lbmDirectWrite()
2083  *
2084  * initiate pageout bypassing write queue for sidestream
2085  * (e.g., log superblock) write;
2086  */
2087 static void lbmDirectWrite(struct jfs_log * log, struct lbuf * bp, int flag)
2088 {
2089 	jfs_info("lbmDirectWrite: bp:0x%p flag:0x%x pn:0x%x",
2090 		 bp, flag, bp->l_pn);
2091 
2092 	/*
2093 	 * initialize buffer for device driver
2094 	 */
2095 	bp->l_flag = flag | lbmDIRECT;
2096 
2097 	/* map the logical block address to physical block address */
2098 	bp->l_blkno =
2099 	    log->base + (bp->l_pn << (L2LOGPSIZE - log->l2bsize));
2100 
2101 	/*
2102 	 *	initiate pageout of the page
2103 	 */
2104 	lbmStartIO(bp);
2105 }
2106 
2107 
2108 /*
2109  * NAME:	lbmStartIO()
2110  *
2111  * FUNCTION:	Interface to DD strategy routine
2112  *
2113  * RETURN:	none
2114  *
2115  * serialization: LCACHE_LOCK() is NOT held during log i/o;
2116  */
2117 static void lbmStartIO(struct lbuf * bp)
2118 {
2119 	struct bio *bio;
2120 	struct jfs_log *log = bp->l_log;
2121 
2122 	jfs_info("lbmStartIO");
2123 
2124 	bio = bio_alloc(log->bdev, 1, REQ_OP_WRITE | REQ_SYNC, GFP_NOFS);
2125 	bio->bi_iter.bi_sector = bp->l_blkno << (log->l2bsize - 9);
2126 	bio_add_page(bio, bp->l_page, LOGPSIZE, bp->l_offset);
2127 	BUG_ON(bio->bi_iter.bi_size != LOGPSIZE);
2128 
2129 	bio->bi_end_io = lbmIODone;
2130 	bio->bi_private = bp;
2131 
2132 	/* check if journaling to disk has been disabled */
2133 	if (log->no_integrity) {
2134 		bio->bi_iter.bi_size = 0;
2135 		lbmIODone(bio);
2136 	} else {
2137 		submit_bio(bio);
2138 		INCREMENT(lmStat.submitted);
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 void lbmIODone(struct bio *bio)
2174 {
2175 	struct lbuf *bp = bio->bi_private;
2176 	struct lbuf *nextbp, *tail;
2177 	struct jfs_log *log;
2178 	unsigned long flags;
2179 
2180 	/*
2181 	 * get back jfs buffer bound to the i/o buffer
2182 	 */
2183 	jfs_info("lbmIODone: bp:0x%p flag:0x%x", bp, bp->l_flag);
2184 
2185 	LCACHE_LOCK(flags);		/* disable+lock */
2186 
2187 	bp->l_flag |= lbmDONE;
2188 
2189 	if (bio->bi_status) {
2190 		bp->l_flag |= lbmERROR;
2191 
2192 		jfs_err("lbmIODone: I/O error in JFS log");
2193 	}
2194 
2195 	bio_put(bio);
2196 
2197 	/*
2198 	 *	pagein completion
2199 	 */
2200 	if (bp->l_flag & lbmREAD) {
2201 		bp->l_flag &= ~lbmREAD;
2202 
2203 		LCACHE_UNLOCK(flags);	/* unlock+enable */
2204 
2205 		/* wakeup I/O initiator */
2206 		LCACHE_WAKEUP(&bp->l_ioevent);
2207 
2208 		return;
2209 	}
2210 
2211 	/*
2212 	 *	pageout completion
2213 	 *
2214 	 * the bp at the head of write queue has completed pageout.
2215 	 *
2216 	 * if single-commit/full-page pageout, remove the current buffer
2217 	 * from head of pageout queue, and redrive pageout with
2218 	 * the new buffer at head of pageout queue;
2219 	 * otherwise, the partial-page pageout buffer stays at
2220 	 * the head of pageout queue to be redriven for pageout
2221 	 * by lmGroupCommit() until full-page pageout is completed.
2222 	 */
2223 	bp->l_flag &= ~lbmWRITE;
2224 	INCREMENT(lmStat.pagedone);
2225 
2226 	/* update committed lsn */
2227 	log = bp->l_log;
2228 	log->clsn = (bp->l_pn << L2LOGPSIZE) + bp->l_ceor;
2229 
2230 	if (bp->l_flag & lbmDIRECT) {
2231 		LCACHE_WAKEUP(&bp->l_ioevent);
2232 		LCACHE_UNLOCK(flags);
2233 		return;
2234 	}
2235 
2236 	tail = log->wqueue;
2237 
2238 	/* single element queue */
2239 	if (bp == tail) {
2240 		/* remove head buffer of full-page pageout
2241 		 * from log device write queue
2242 		 */
2243 		if (bp->l_flag & lbmRELEASE) {
2244 			log->wqueue = NULL;
2245 			bp->l_wqnext = NULL;
2246 		}
2247 	}
2248 	/* multi element queue */
2249 	else {
2250 		/* remove head buffer of full-page pageout
2251 		 * from log device write queue
2252 		 */
2253 		if (bp->l_flag & lbmRELEASE) {
2254 			nextbp = tail->l_wqnext = bp->l_wqnext;
2255 			bp->l_wqnext = NULL;
2256 
2257 			/*
2258 			 * redrive pageout of next page at head of write queue:
2259 			 * redrive next page without any bound tblk
2260 			 * (i.e., page w/o any COMMIT records), or
2261 			 * first page of new group commit which has been
2262 			 * queued after current page (subsequent pageout
2263 			 * is performed synchronously, except page without
2264 			 * any COMMITs) by lmGroupCommit() as indicated
2265 			 * by lbmWRITE flag;
2266 			 */
2267 			if (nextbp->l_flag & lbmWRITE) {
2268 				/*
2269 				 * We can't do the I/O at interrupt time.
2270 				 * The jfsIO thread can do it
2271 				 */
2272 				lbmRedrive(nextbp);
2273 			}
2274 		}
2275 	}
2276 
2277 	/*
2278 	 *	synchronous pageout:
2279 	 *
2280 	 * buffer has not necessarily been removed from write queue
2281 	 * (e.g., synchronous write of partial-page with COMMIT):
2282 	 * leave buffer for i/o initiator to dispose
2283 	 */
2284 	if (bp->l_flag & lbmSYNC) {
2285 		LCACHE_UNLOCK(flags);	/* unlock+enable */
2286 
2287 		/* wakeup I/O initiator */
2288 		LCACHE_WAKEUP(&bp->l_ioevent);
2289 	}
2290 
2291 	/*
2292 	 *	Group Commit pageout:
2293 	 */
2294 	else if (bp->l_flag & lbmGC) {
2295 		LCACHE_UNLOCK(flags);
2296 		lmPostGC(bp);
2297 	}
2298 
2299 	/*
2300 	 *	asynchronous pageout:
2301 	 *
2302 	 * buffer must have been removed from write queue:
2303 	 * insert buffer at head of freelist where it can be recycled
2304 	 */
2305 	else {
2306 		assert(bp->l_flag & lbmRELEASE);
2307 		assert(bp->l_flag & lbmFREE);
2308 		lbmfree(bp);
2309 
2310 		LCACHE_UNLOCK(flags);	/* unlock+enable */
2311 	}
2312 }
2313 
2314 int jfsIOWait(void *arg)
2315 {
2316 	struct lbuf *bp;
2317 
2318 	do {
2319 		spin_lock_irq(&log_redrive_lock);
2320 		while ((bp = log_redrive_list)) {
2321 			log_redrive_list = bp->l_redrive_next;
2322 			bp->l_redrive_next = NULL;
2323 			spin_unlock_irq(&log_redrive_lock);
2324 			lbmStartIO(bp);
2325 			spin_lock_irq(&log_redrive_lock);
2326 		}
2327 
2328 		if (freezing(current)) {
2329 			spin_unlock_irq(&log_redrive_lock);
2330 			try_to_freeze();
2331 		} else {
2332 			set_current_state(TASK_INTERRUPTIBLE);
2333 			spin_unlock_irq(&log_redrive_lock);
2334 			schedule();
2335 		}
2336 	} while (!kthread_should_stop());
2337 
2338 	jfs_info("jfsIOWait being killed!");
2339 	return 0;
2340 }
2341 
2342 /*
2343  * NAME:	lmLogFormat()/jfs_logform()
2344  *
2345  * FUNCTION:	format file system log
2346  *
2347  * PARAMETERS:
2348  *	log	- volume log
2349  *	logAddress - start address of log space in FS block
2350  *	logSize	- length of log space in FS block;
2351  *
2352  * RETURN:	0	- success
2353  *		-EIO	- i/o error
2354  *
2355  * XXX: We're synchronously writing one page at a time.  This needs to
2356  *	be improved by writing multiple pages at once.
2357  */
2358 int lmLogFormat(struct jfs_log *log, s64 logAddress, int logSize)
2359 {
2360 	int rc = -EIO;
2361 	struct jfs_sb_info *sbi;
2362 	struct logsuper *logsuper;
2363 	struct logpage *lp;
2364 	int lspn;		/* log sequence page number */
2365 	struct lrd *lrd_ptr;
2366 	int npages = 0;
2367 	struct lbuf *bp;
2368 
2369 	jfs_info("lmLogFormat: logAddress:%Ld logSize:%d",
2370 		 (long long)logAddress, logSize);
2371 
2372 	sbi = list_entry(log->sb_list.next, struct jfs_sb_info, log_list);
2373 
2374 	/* allocate a log buffer */
2375 	bp = lbmAllocate(log, 1);
2376 
2377 	npages = logSize >> sbi->l2nbperpage;
2378 
2379 	/*
2380 	 *	log space:
2381 	 *
2382 	 * page 0 - reserved;
2383 	 * page 1 - log superblock;
2384 	 * page 2 - log data page: A SYNC log record is written
2385 	 *	    into this page at logform time;
2386 	 * pages 3-N - log data page: set to empty log data pages;
2387 	 */
2388 	/*
2389 	 *	init log superblock: log page 1
2390 	 */
2391 	logsuper = (struct logsuper *) bp->l_ldata;
2392 
2393 	logsuper->magic = cpu_to_le32(LOGMAGIC);
2394 	logsuper->version = cpu_to_le32(LOGVERSION);
2395 	logsuper->state = cpu_to_le32(LOGREDONE);
2396 	logsuper->flag = cpu_to_le32(sbi->mntflag);	/* ? */
2397 	logsuper->size = cpu_to_le32(npages);
2398 	logsuper->bsize = cpu_to_le32(sbi->bsize);
2399 	logsuper->l2bsize = cpu_to_le32(sbi->l2bsize);
2400 	logsuper->end = cpu_to_le32(2 * LOGPSIZE + LOGPHDRSIZE + LOGRDSIZE);
2401 
2402 	bp->l_flag = lbmWRITE | lbmSYNC | lbmDIRECT;
2403 	bp->l_blkno = logAddress + sbi->nbperpage;
2404 	lbmStartIO(bp);
2405 	if ((rc = lbmIOWait(bp, 0)))
2406 		goto exit;
2407 
2408 	/*
2409 	 *	init pages 2 to npages-1 as log data pages:
2410 	 *
2411 	 * log page sequence number (lpsn) initialization:
2412 	 *
2413 	 * pn:   0     1     2     3                 n-1
2414 	 *       +-----+-----+=====+=====+===.....===+=====+
2415 	 * lspn:             N-1   0     1           N-2
2416 	 *                   <--- N page circular file ---->
2417 	 *
2418 	 * the N (= npages-2) data pages of the log is maintained as
2419 	 * a circular file for the log records;
2420 	 * lpsn grows by 1 monotonically as each log page is written
2421 	 * to the circular file of the log;
2422 	 * and setLogpage() will not reset the page number even if
2423 	 * the eor is equal to LOGPHDRSIZE. In order for binary search
2424 	 * still work in find log end process, we have to simulate the
2425 	 * log wrap situation at the log format time.
2426 	 * The 1st log page written will have the highest lpsn. Then
2427 	 * the succeeding log pages will have ascending order of
2428 	 * the lspn starting from 0, ... (N-2)
2429 	 */
2430 	lp = (struct logpage *) bp->l_ldata;
2431 	/*
2432 	 * initialize 1st log page to be written: lpsn = N - 1,
2433 	 * write a SYNCPT log record is written to this page
2434 	 */
2435 	lp->h.page = lp->t.page = cpu_to_le32(npages - 3);
2436 	lp->h.eor = lp->t.eor = cpu_to_le16(LOGPHDRSIZE + LOGRDSIZE);
2437 
2438 	lrd_ptr = (struct lrd *) &lp->data;
2439 	lrd_ptr->logtid = 0;
2440 	lrd_ptr->backchain = 0;
2441 	lrd_ptr->type = cpu_to_le16(LOG_SYNCPT);
2442 	lrd_ptr->length = 0;
2443 	lrd_ptr->log.syncpt.sync = 0;
2444 
2445 	bp->l_blkno += sbi->nbperpage;
2446 	bp->l_flag = lbmWRITE | lbmSYNC | lbmDIRECT;
2447 	lbmStartIO(bp);
2448 	if ((rc = lbmIOWait(bp, 0)))
2449 		goto exit;
2450 
2451 	/*
2452 	 *	initialize succeeding log pages: lpsn = 0, 1, ..., (N-2)
2453 	 */
2454 	for (lspn = 0; lspn < npages - 3; lspn++) {
2455 		lp->h.page = lp->t.page = cpu_to_le32(lspn);
2456 		lp->h.eor = lp->t.eor = cpu_to_le16(LOGPHDRSIZE);
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 	rc = 0;
2466 exit:
2467 	/*
2468 	 *	finalize log
2469 	 */
2470 	/* release the buffer */
2471 	lbmFree(bp);
2472 
2473 	return rc;
2474 }
2475 
2476 #ifdef CONFIG_JFS_STATISTICS
2477 int jfs_lmstats_proc_show(struct seq_file *m, void *v)
2478 {
2479 	seq_printf(m,
2480 		       "JFS Logmgr stats\n"
2481 		       "================\n"
2482 		       "commits = %d\n"
2483 		       "writes submitted = %d\n"
2484 		       "writes completed = %d\n"
2485 		       "full pages submitted = %d\n"
2486 		       "partial pages submitted = %d\n",
2487 		       lmStat.commit,
2488 		       lmStat.submitted,
2489 		       lmStat.pagedone,
2490 		       lmStat.full_page,
2491 		       lmStat.partial_page);
2492 	return 0;
2493 }
2494 #endif /* CONFIG_JFS_STATISTICS */
2495