• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  *   Copyright (C) International Business Machines Corp., 2000-2005
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_txnmgr.c: transaction manager
22  *
23  * notes:
24  * transaction starts with txBegin() and ends with txCommit()
25  * or txAbort().
26  *
27  * tlock is acquired at the time of update;
28  * (obviate scan at commit time for xtree and dtree)
29  * tlock and mp points to each other;
30  * (no hashlist for mp -> tlock).
31  *
32  * special cases:
33  * tlock on in-memory inode:
34  * in-place tlock in the in-memory inode itself;
35  * converted to page lock by iWrite() at commit time.
36  *
37  * tlock during write()/mmap() under anonymous transaction (tid = 0):
38  * transferred (?) to transaction at commit time.
39  *
40  * use the page itself to update allocation maps
41  * (obviate intermediate replication of allocation/deallocation data)
42  * hold on to mp+lock thru update of maps
43  */
44 
45 #include <linux/fs.h>
46 #include <linux/vmalloc.h>
47 #include <linux/completion.h>
48 #include <linux/freezer.h>
49 #include <linux/module.h>
50 #include <linux/moduleparam.h>
51 #include <linux/kthread.h>
52 #include <linux/seq_file.h>
53 #include "jfs_incore.h"
54 #include "jfs_inode.h"
55 #include "jfs_filsys.h"
56 #include "jfs_metapage.h"
57 #include "jfs_dinode.h"
58 #include "jfs_imap.h"
59 #include "jfs_dmap.h"
60 #include "jfs_superblock.h"
61 #include "jfs_debug.h"
62 
63 /*
64  *	transaction management structures
65  */
66 static struct {
67 	int freetid;		/* index of a free tid structure */
68 	int freelock;		/* index first free lock word */
69 	wait_queue_head_t freewait;	/* eventlist of free tblock */
70 	wait_queue_head_t freelockwait;	/* eventlist of free tlock */
71 	wait_queue_head_t lowlockwait;	/* eventlist of ample tlocks */
72 	int tlocksInUse;	/* Number of tlocks in use */
73 	spinlock_t LazyLock;	/* synchronize sync_queue & unlock_queue */
74 /*	struct tblock *sync_queue; * Transactions waiting for data sync */
75 	struct list_head unlock_queue;	/* Txns waiting to be released */
76 	struct list_head anon_list;	/* inodes having anonymous txns */
77 	struct list_head anon_list2;	/* inodes having anonymous txns
78 					   that couldn't be sync'ed */
79 } TxAnchor;
80 
81 int jfs_tlocks_low;		/* Indicates low number of available tlocks */
82 
83 #ifdef CONFIG_JFS_STATISTICS
84 static struct {
85 	uint txBegin;
86 	uint txBegin_barrier;
87 	uint txBegin_lockslow;
88 	uint txBegin_freetid;
89 	uint txBeginAnon;
90 	uint txBeginAnon_barrier;
91 	uint txBeginAnon_lockslow;
92 	uint txLockAlloc;
93 	uint txLockAlloc_freelock;
94 } TxStat;
95 #endif
96 
97 static int nTxBlock = -1;	/* number of transaction blocks */
98 module_param(nTxBlock, int, 0);
99 MODULE_PARM_DESC(nTxBlock,
100 		 "Number of transaction blocks (max:65536)");
101 
102 static int nTxLock = -1;	/* number of transaction locks */
103 module_param(nTxLock, int, 0);
104 MODULE_PARM_DESC(nTxLock,
105 		 "Number of transaction locks (max:65536)");
106 
107 struct tblock *TxBlock;	/* transaction block table */
108 static int TxLockLWM;	/* Low water mark for number of txLocks used */
109 static int TxLockHWM;	/* High water mark for number of txLocks used */
110 static int TxLockVHWM;	/* Very High water mark */
111 struct tlock *TxLock;	/* transaction lock table */
112 
113 /*
114  *	transaction management lock
115  */
116 static DEFINE_SPINLOCK(jfsTxnLock);
117 
118 #define TXN_LOCK()		spin_lock(&jfsTxnLock)
119 #define TXN_UNLOCK()		spin_unlock(&jfsTxnLock)
120 
121 #define LAZY_LOCK_INIT()	spin_lock_init(&TxAnchor.LazyLock);
122 #define LAZY_LOCK(flags)	spin_lock_irqsave(&TxAnchor.LazyLock, flags)
123 #define LAZY_UNLOCK(flags) spin_unlock_irqrestore(&TxAnchor.LazyLock, flags)
124 
125 static DECLARE_WAIT_QUEUE_HEAD(jfs_commit_thread_wait);
126 static int jfs_commit_thread_waking;
127 
128 /*
129  * Retry logic exist outside these macros to protect from spurrious wakeups.
130  */
TXN_SLEEP_DROP_LOCK(wait_queue_head_t * event)131 static inline void TXN_SLEEP_DROP_LOCK(wait_queue_head_t * event)
132 {
133 	DECLARE_WAITQUEUE(wait, current);
134 
135 	add_wait_queue(event, &wait);
136 	set_current_state(TASK_UNINTERRUPTIBLE);
137 	TXN_UNLOCK();
138 	io_schedule();
139 	remove_wait_queue(event, &wait);
140 }
141 
142 #define TXN_SLEEP(event)\
143 {\
144 	TXN_SLEEP_DROP_LOCK(event);\
145 	TXN_LOCK();\
146 }
147 
148 #define TXN_WAKEUP(event) wake_up_all(event)
149 
150 /*
151  *	statistics
152  */
153 static struct {
154 	tid_t maxtid;		/* 4: biggest tid ever used */
155 	lid_t maxlid;		/* 4: biggest lid ever used */
156 	int ntid;		/* 4: # of transactions performed */
157 	int nlid;		/* 4: # of tlocks acquired */
158 	int waitlock;		/* 4: # of tlock wait */
159 } stattx;
160 
161 /*
162  * forward references
163  */
164 static int diLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
165 		struct tlock * tlck, struct commit * cd);
166 static int dataLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
167 		struct tlock * tlck);
168 static void dtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
169 		struct tlock * tlck);
170 static void mapLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
171 		struct tlock * tlck);
172 static void txAllocPMap(struct inode *ip, struct maplock * maplock,
173 		struct tblock * tblk);
174 static void txForce(struct tblock * tblk);
175 static int txLog(struct jfs_log * log, struct tblock * tblk,
176 		struct commit * cd);
177 static void txUpdateMap(struct tblock * tblk);
178 static void txRelease(struct tblock * tblk);
179 static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
180 	   struct tlock * tlck);
181 static void LogSyncRelease(struct metapage * mp);
182 
183 /*
184  *		transaction block/lock management
185  *		---------------------------------
186  */
187 
188 /*
189  * Get a transaction lock from the free list.  If the number in use is
190  * greater than the high water mark, wake up the sync daemon.  This should
191  * free some anonymous transaction locks.  (TXN_LOCK must be held.)
192  */
txLockAlloc(void)193 static lid_t txLockAlloc(void)
194 {
195 	lid_t lid;
196 
197 	INCREMENT(TxStat.txLockAlloc);
198 	if (!TxAnchor.freelock) {
199 		INCREMENT(TxStat.txLockAlloc_freelock);
200 	}
201 
202 	while (!(lid = TxAnchor.freelock))
203 		TXN_SLEEP(&TxAnchor.freelockwait);
204 	TxAnchor.freelock = TxLock[lid].next;
205 	HIGHWATERMARK(stattx.maxlid, lid);
206 	if ((++TxAnchor.tlocksInUse > TxLockHWM) && (jfs_tlocks_low == 0)) {
207 		jfs_info("txLockAlloc tlocks low");
208 		jfs_tlocks_low = 1;
209 		wake_up_process(jfsSyncThread);
210 	}
211 
212 	return lid;
213 }
214 
txLockFree(lid_t lid)215 static void txLockFree(lid_t lid)
216 {
217 	TxLock[lid].tid = 0;
218 	TxLock[lid].next = TxAnchor.freelock;
219 	TxAnchor.freelock = lid;
220 	TxAnchor.tlocksInUse--;
221 	if (jfs_tlocks_low && (TxAnchor.tlocksInUse < TxLockLWM)) {
222 		jfs_info("txLockFree jfs_tlocks_low no more");
223 		jfs_tlocks_low = 0;
224 		TXN_WAKEUP(&TxAnchor.lowlockwait);
225 	}
226 	TXN_WAKEUP(&TxAnchor.freelockwait);
227 }
228 
229 /*
230  * NAME:	txInit()
231  *
232  * FUNCTION:	initialize transaction management structures
233  *
234  * RETURN:
235  *
236  * serialization: single thread at jfs_init()
237  */
txInit(void)238 int txInit(void)
239 {
240 	int k, size;
241 	struct sysinfo si;
242 
243 	/* Set defaults for nTxLock and nTxBlock if unset */
244 
245 	if (nTxLock == -1) {
246 		if (nTxBlock == -1) {
247 			/* Base default on memory size */
248 			si_meminfo(&si);
249 			if (si.totalram > (256 * 1024)) /* 1 GB */
250 				nTxLock = 64 * 1024;
251 			else
252 				nTxLock = si.totalram >> 2;
253 		} else if (nTxBlock > (8 * 1024))
254 			nTxLock = 64 * 1024;
255 		else
256 			nTxLock = nTxBlock << 3;
257 	}
258 	if (nTxBlock == -1)
259 		nTxBlock = nTxLock >> 3;
260 
261 	/* Verify tunable parameters */
262 	if (nTxBlock < 16)
263 		nTxBlock = 16;	/* No one should set it this low */
264 	if (nTxBlock > 65536)
265 		nTxBlock = 65536;
266 	if (nTxLock < 256)
267 		nTxLock = 256;	/* No one should set it this low */
268 	if (nTxLock > 65536)
269 		nTxLock = 65536;
270 
271 	printk(KERN_INFO "JFS: nTxBlock = %d, nTxLock = %d\n",
272 	       nTxBlock, nTxLock);
273 	/*
274 	 * initialize transaction block (tblock) table
275 	 *
276 	 * transaction id (tid) = tblock index
277 	 * tid = 0 is reserved.
278 	 */
279 	TxLockLWM = (nTxLock * 4) / 10;
280 	TxLockHWM = (nTxLock * 7) / 10;
281 	TxLockVHWM = (nTxLock * 8) / 10;
282 
283 	size = sizeof(struct tblock) * nTxBlock;
284 	TxBlock = vmalloc(size);
285 	if (TxBlock == NULL)
286 		return -ENOMEM;
287 
288 	for (k = 1; k < nTxBlock - 1; k++) {
289 		TxBlock[k].next = k + 1;
290 		init_waitqueue_head(&TxBlock[k].gcwait);
291 		init_waitqueue_head(&TxBlock[k].waitor);
292 	}
293 	TxBlock[k].next = 0;
294 	init_waitqueue_head(&TxBlock[k].gcwait);
295 	init_waitqueue_head(&TxBlock[k].waitor);
296 
297 	TxAnchor.freetid = 1;
298 	init_waitqueue_head(&TxAnchor.freewait);
299 
300 	stattx.maxtid = 1;	/* statistics */
301 
302 	/*
303 	 * initialize transaction lock (tlock) table
304 	 *
305 	 * transaction lock id = tlock index
306 	 * tlock id = 0 is reserved.
307 	 */
308 	size = sizeof(struct tlock) * nTxLock;
309 	TxLock = vmalloc(size);
310 	if (TxLock == NULL) {
311 		vfree(TxBlock);
312 		return -ENOMEM;
313 	}
314 
315 	/* initialize tlock table */
316 	for (k = 1; k < nTxLock - 1; k++)
317 		TxLock[k].next = k + 1;
318 	TxLock[k].next = 0;
319 	init_waitqueue_head(&TxAnchor.freelockwait);
320 	init_waitqueue_head(&TxAnchor.lowlockwait);
321 
322 	TxAnchor.freelock = 1;
323 	TxAnchor.tlocksInUse = 0;
324 	INIT_LIST_HEAD(&TxAnchor.anon_list);
325 	INIT_LIST_HEAD(&TxAnchor.anon_list2);
326 
327 	LAZY_LOCK_INIT();
328 	INIT_LIST_HEAD(&TxAnchor.unlock_queue);
329 
330 	stattx.maxlid = 1;	/* statistics */
331 
332 	return 0;
333 }
334 
335 /*
336  * NAME:	txExit()
337  *
338  * FUNCTION:	clean up when module is unloaded
339  */
txExit(void)340 void txExit(void)
341 {
342 	vfree(TxLock);
343 	TxLock = NULL;
344 	vfree(TxBlock);
345 	TxBlock = NULL;
346 }
347 
348 /*
349  * NAME:	txBegin()
350  *
351  * FUNCTION:	start a transaction.
352  *
353  * PARAMETER:	sb	- superblock
354  *		flag	- force for nested tx;
355  *
356  * RETURN:	tid	- transaction id
357  *
358  * note: flag force allows to start tx for nested tx
359  * to prevent deadlock on logsync barrier;
360  */
txBegin(struct super_block * sb,int flag)361 tid_t txBegin(struct super_block *sb, int flag)
362 {
363 	tid_t t;
364 	struct tblock *tblk;
365 	struct jfs_log *log;
366 
367 	jfs_info("txBegin: flag = 0x%x", flag);
368 	log = JFS_SBI(sb)->log;
369 
370 	TXN_LOCK();
371 
372 	INCREMENT(TxStat.txBegin);
373 
374       retry:
375 	if (!(flag & COMMIT_FORCE)) {
376 		/*
377 		 * synchronize with logsync barrier
378 		 */
379 		if (test_bit(log_SYNCBARRIER, &log->flag) ||
380 		    test_bit(log_QUIESCE, &log->flag)) {
381 			INCREMENT(TxStat.txBegin_barrier);
382 			TXN_SLEEP(&log->syncwait);
383 			goto retry;
384 		}
385 	}
386 	if (flag == 0) {
387 		/*
388 		 * Don't begin transaction if we're getting starved for tlocks
389 		 * unless COMMIT_FORCE or COMMIT_INODE (which may ultimately
390 		 * free tlocks)
391 		 */
392 		if (TxAnchor.tlocksInUse > TxLockVHWM) {
393 			INCREMENT(TxStat.txBegin_lockslow);
394 			TXN_SLEEP(&TxAnchor.lowlockwait);
395 			goto retry;
396 		}
397 	}
398 
399 	/*
400 	 * allocate transaction id/block
401 	 */
402 	if ((t = TxAnchor.freetid) == 0) {
403 		jfs_info("txBegin: waiting for free tid");
404 		INCREMENT(TxStat.txBegin_freetid);
405 		TXN_SLEEP(&TxAnchor.freewait);
406 		goto retry;
407 	}
408 
409 	tblk = tid_to_tblock(t);
410 
411 	if ((tblk->next == 0) && !(flag & COMMIT_FORCE)) {
412 		/* Don't let a non-forced transaction take the last tblk */
413 		jfs_info("txBegin: waiting for free tid");
414 		INCREMENT(TxStat.txBegin_freetid);
415 		TXN_SLEEP(&TxAnchor.freewait);
416 		goto retry;
417 	}
418 
419 	TxAnchor.freetid = tblk->next;
420 
421 	/*
422 	 * initialize transaction
423 	 */
424 
425 	/*
426 	 * We can't zero the whole thing or we screw up another thread being
427 	 * awakened after sleeping on tblk->waitor
428 	 *
429 	 * memset(tblk, 0, sizeof(struct tblock));
430 	 */
431 	tblk->next = tblk->last = tblk->xflag = tblk->flag = tblk->lsn = 0;
432 
433 	tblk->sb = sb;
434 	++log->logtid;
435 	tblk->logtid = log->logtid;
436 
437 	++log->active;
438 
439 	HIGHWATERMARK(stattx.maxtid, t);	/* statistics */
440 	INCREMENT(stattx.ntid);	/* statistics */
441 
442 	TXN_UNLOCK();
443 
444 	jfs_info("txBegin: returning tid = %d", t);
445 
446 	return t;
447 }
448 
449 /*
450  * NAME:	txBeginAnon()
451  *
452  * FUNCTION:	start an anonymous transaction.
453  *		Blocks if logsync or available tlocks are low to prevent
454  *		anonymous tlocks from depleting supply.
455  *
456  * PARAMETER:	sb	- superblock
457  *
458  * RETURN:	none
459  */
txBeginAnon(struct super_block * sb)460 void txBeginAnon(struct super_block *sb)
461 {
462 	struct jfs_log *log;
463 
464 	log = JFS_SBI(sb)->log;
465 
466 	TXN_LOCK();
467 	INCREMENT(TxStat.txBeginAnon);
468 
469       retry:
470 	/*
471 	 * synchronize with logsync barrier
472 	 */
473 	if (test_bit(log_SYNCBARRIER, &log->flag) ||
474 	    test_bit(log_QUIESCE, &log->flag)) {
475 		INCREMENT(TxStat.txBeginAnon_barrier);
476 		TXN_SLEEP(&log->syncwait);
477 		goto retry;
478 	}
479 
480 	/*
481 	 * Don't begin transaction if we're getting starved for tlocks
482 	 */
483 	if (TxAnchor.tlocksInUse > TxLockVHWM) {
484 		INCREMENT(TxStat.txBeginAnon_lockslow);
485 		TXN_SLEEP(&TxAnchor.lowlockwait);
486 		goto retry;
487 	}
488 	TXN_UNLOCK();
489 }
490 
491 /*
492  *	txEnd()
493  *
494  * function: free specified transaction block.
495  *
496  *	logsync barrier processing:
497  *
498  * serialization:
499  */
txEnd(tid_t tid)500 void txEnd(tid_t tid)
501 {
502 	struct tblock *tblk = tid_to_tblock(tid);
503 	struct jfs_log *log;
504 
505 	jfs_info("txEnd: tid = %d", tid);
506 	TXN_LOCK();
507 
508 	/*
509 	 * wakeup transactions waiting on the page locked
510 	 * by the current transaction
511 	 */
512 	TXN_WAKEUP(&tblk->waitor);
513 
514 	log = JFS_SBI(tblk->sb)->log;
515 
516 	/*
517 	 * Lazy commit thread can't free this guy until we mark it UNLOCKED,
518 	 * otherwise, we would be left with a transaction that may have been
519 	 * reused.
520 	 *
521 	 * Lazy commit thread will turn off tblkGC_LAZY before calling this
522 	 * routine.
523 	 */
524 	if (tblk->flag & tblkGC_LAZY) {
525 		jfs_info("txEnd called w/lazy tid: %d, tblk = 0x%p", tid, tblk);
526 		TXN_UNLOCK();
527 
528 		spin_lock_irq(&log->gclock);	// LOGGC_LOCK
529 		tblk->flag |= tblkGC_UNLOCKED;
530 		spin_unlock_irq(&log->gclock);	// LOGGC_UNLOCK
531 		return;
532 	}
533 
534 	jfs_info("txEnd: tid: %d, tblk = 0x%p", tid, tblk);
535 
536 	assert(tblk->next == 0);
537 
538 	/*
539 	 * insert tblock back on freelist
540 	 */
541 	tblk->next = TxAnchor.freetid;
542 	TxAnchor.freetid = tid;
543 
544 	/*
545 	 * mark the tblock not active
546 	 */
547 	if (--log->active == 0) {
548 		clear_bit(log_FLUSH, &log->flag);
549 
550 		/*
551 		 * synchronize with logsync barrier
552 		 */
553 		if (test_bit(log_SYNCBARRIER, &log->flag)) {
554 			TXN_UNLOCK();
555 
556 			/* write dirty metadata & forward log syncpt */
557 			jfs_syncpt(log, 1);
558 
559 			jfs_info("log barrier off: 0x%x", log->lsn);
560 
561 			/* enable new transactions start */
562 			clear_bit(log_SYNCBARRIER, &log->flag);
563 
564 			/* wakeup all waitors for logsync barrier */
565 			TXN_WAKEUP(&log->syncwait);
566 
567 			goto wakeup;
568 		}
569 	}
570 
571 	TXN_UNLOCK();
572 wakeup:
573 	/*
574 	 * wakeup all waitors for a free tblock
575 	 */
576 	TXN_WAKEUP(&TxAnchor.freewait);
577 }
578 
579 /*
580  *	txLock()
581  *
582  * function: acquire a transaction lock on the specified <mp>
583  *
584  * parameter:
585  *
586  * return:	transaction lock id
587  *
588  * serialization:
589  */
txLock(tid_t tid,struct inode * ip,struct metapage * mp,int type)590 struct tlock *txLock(tid_t tid, struct inode *ip, struct metapage * mp,
591 		     int type)
592 {
593 	struct jfs_inode_info *jfs_ip = JFS_IP(ip);
594 	int dir_xtree = 0;
595 	lid_t lid;
596 	tid_t xtid;
597 	struct tlock *tlck;
598 	struct xtlock *xtlck;
599 	struct linelock *linelock;
600 	xtpage_t *p;
601 	struct tblock *tblk;
602 
603 	TXN_LOCK();
604 
605 	if (S_ISDIR(ip->i_mode) && (type & tlckXTREE) &&
606 	    !(mp->xflag & COMMIT_PAGE)) {
607 		/*
608 		 * Directory inode is special.  It can have both an xtree tlock
609 		 * and a dtree tlock associated with it.
610 		 */
611 		dir_xtree = 1;
612 		lid = jfs_ip->xtlid;
613 	} else
614 		lid = mp->lid;
615 
616 	/* is page not locked by a transaction ? */
617 	if (lid == 0)
618 		goto allocateLock;
619 
620 	jfs_info("txLock: tid:%d ip:0x%p mp:0x%p lid:%d", tid, ip, mp, lid);
621 
622 	/* is page locked by the requester transaction ? */
623 	tlck = lid_to_tlock(lid);
624 	if ((xtid = tlck->tid) == tid) {
625 		TXN_UNLOCK();
626 		goto grantLock;
627 	}
628 
629 	/*
630 	 * is page locked by anonymous transaction/lock ?
631 	 *
632 	 * (page update without transaction (i.e., file write) is
633 	 * locked under anonymous transaction tid = 0:
634 	 * anonymous tlocks maintained on anonymous tlock list of
635 	 * the inode of the page and available to all anonymous
636 	 * transactions until txCommit() time at which point
637 	 * they are transferred to the transaction tlock list of
638 	 * the committing transaction of the inode)
639 	 */
640 	if (xtid == 0) {
641 		tlck->tid = tid;
642 		TXN_UNLOCK();
643 		tblk = tid_to_tblock(tid);
644 		/*
645 		 * The order of the tlocks in the transaction is important
646 		 * (during truncate, child xtree pages must be freed before
647 		 * parent's tlocks change the working map).
648 		 * Take tlock off anonymous list and add to tail of
649 		 * transaction list
650 		 *
651 		 * Note:  We really need to get rid of the tid & lid and
652 		 * use list_head's.  This code is getting UGLY!
653 		 */
654 		if (jfs_ip->atlhead == lid) {
655 			if (jfs_ip->atltail == lid) {
656 				/* only anonymous txn.
657 				 * Remove from anon_list
658 				 */
659 				TXN_LOCK();
660 				list_del_init(&jfs_ip->anon_inode_list);
661 				TXN_UNLOCK();
662 			}
663 			jfs_ip->atlhead = tlck->next;
664 		} else {
665 			lid_t last;
666 			for (last = jfs_ip->atlhead;
667 			     lid_to_tlock(last)->next != lid;
668 			     last = lid_to_tlock(last)->next) {
669 				assert(last);
670 			}
671 			lid_to_tlock(last)->next = tlck->next;
672 			if (jfs_ip->atltail == lid)
673 				jfs_ip->atltail = last;
674 		}
675 
676 		/* insert the tlock at tail of transaction tlock list */
677 
678 		if (tblk->next)
679 			lid_to_tlock(tblk->last)->next = lid;
680 		else
681 			tblk->next = lid;
682 		tlck->next = 0;
683 		tblk->last = lid;
684 
685 		goto grantLock;
686 	}
687 
688 	goto waitLock;
689 
690 	/*
691 	 * allocate a tlock
692 	 */
693       allocateLock:
694 	lid = txLockAlloc();
695 	tlck = lid_to_tlock(lid);
696 
697 	/*
698 	 * initialize tlock
699 	 */
700 	tlck->tid = tid;
701 
702 	TXN_UNLOCK();
703 
704 	/* mark tlock for meta-data page */
705 	if (mp->xflag & COMMIT_PAGE) {
706 
707 		tlck->flag = tlckPAGELOCK;
708 
709 		/* mark the page dirty and nohomeok */
710 		metapage_nohomeok(mp);
711 
712 		jfs_info("locking mp = 0x%p, nohomeok = %d tid = %d tlck = 0x%p",
713 			 mp, mp->nohomeok, tid, tlck);
714 
715 		/* if anonymous transaction, and buffer is on the group
716 		 * commit synclist, mark inode to show this.  This will
717 		 * prevent the buffer from being marked nohomeok for too
718 		 * long a time.
719 		 */
720 		if ((tid == 0) && mp->lsn)
721 			set_cflag(COMMIT_Synclist, ip);
722 	}
723 	/* mark tlock for in-memory inode */
724 	else
725 		tlck->flag = tlckINODELOCK;
726 
727 	if (S_ISDIR(ip->i_mode))
728 		tlck->flag |= tlckDIRECTORY;
729 
730 	tlck->type = 0;
731 
732 	/* bind the tlock and the page */
733 	tlck->ip = ip;
734 	tlck->mp = mp;
735 	if (dir_xtree)
736 		jfs_ip->xtlid = lid;
737 	else
738 		mp->lid = lid;
739 
740 	/*
741 	 * enqueue transaction lock to transaction/inode
742 	 */
743 	/* insert the tlock at tail of transaction tlock list */
744 	if (tid) {
745 		tblk = tid_to_tblock(tid);
746 		if (tblk->next)
747 			lid_to_tlock(tblk->last)->next = lid;
748 		else
749 			tblk->next = lid;
750 		tlck->next = 0;
751 		tblk->last = lid;
752 	}
753 	/* anonymous transaction:
754 	 * insert the tlock at head of inode anonymous tlock list
755 	 */
756 	else {
757 		tlck->next = jfs_ip->atlhead;
758 		jfs_ip->atlhead = lid;
759 		if (tlck->next == 0) {
760 			/* This inode's first anonymous transaction */
761 			jfs_ip->atltail = lid;
762 			TXN_LOCK();
763 			list_add_tail(&jfs_ip->anon_inode_list,
764 				      &TxAnchor.anon_list);
765 			TXN_UNLOCK();
766 		}
767 	}
768 
769 	/* initialize type dependent area for linelock */
770 	linelock = (struct linelock *) & tlck->lock;
771 	linelock->next = 0;
772 	linelock->flag = tlckLINELOCK;
773 	linelock->maxcnt = TLOCKSHORT;
774 	linelock->index = 0;
775 
776 	switch (type & tlckTYPE) {
777 	case tlckDTREE:
778 		linelock->l2linesize = L2DTSLOTSIZE;
779 		break;
780 
781 	case tlckXTREE:
782 		linelock->l2linesize = L2XTSLOTSIZE;
783 
784 		xtlck = (struct xtlock *) linelock;
785 		xtlck->header.offset = 0;
786 		xtlck->header.length = 2;
787 
788 		if (type & tlckNEW) {
789 			xtlck->lwm.offset = XTENTRYSTART;
790 		} else {
791 			if (mp->xflag & COMMIT_PAGE)
792 				p = (xtpage_t *) mp->data;
793 			else
794 				p = &jfs_ip->i_xtroot;
795 			xtlck->lwm.offset =
796 			    le16_to_cpu(p->header.nextindex);
797 		}
798 		xtlck->lwm.length = 0;	/* ! */
799 		xtlck->twm.offset = 0;
800 		xtlck->hwm.offset = 0;
801 
802 		xtlck->index = 2;
803 		break;
804 
805 	case tlckINODE:
806 		linelock->l2linesize = L2INODESLOTSIZE;
807 		break;
808 
809 	case tlckDATA:
810 		linelock->l2linesize = L2DATASLOTSIZE;
811 		break;
812 
813 	default:
814 		jfs_err("UFO tlock:0x%p", tlck);
815 	}
816 
817 	/*
818 	 * update tlock vector
819 	 */
820       grantLock:
821 	tlck->type |= type;
822 
823 	return tlck;
824 
825 	/*
826 	 * page is being locked by another transaction:
827 	 */
828       waitLock:
829 	/* Only locks on ipimap or ipaimap should reach here */
830 	/* assert(jfs_ip->fileset == AGGREGATE_I); */
831 	if (jfs_ip->fileset != AGGREGATE_I) {
832 		printk(KERN_ERR "txLock: trying to lock locked page!");
833 		print_hex_dump(KERN_ERR, "ip: ", DUMP_PREFIX_ADDRESS, 16, 4,
834 			       ip, sizeof(*ip), 0);
835 		print_hex_dump(KERN_ERR, "mp: ", DUMP_PREFIX_ADDRESS, 16, 4,
836 			       mp, sizeof(*mp), 0);
837 		print_hex_dump(KERN_ERR, "Locker's tblock: ",
838 			       DUMP_PREFIX_ADDRESS, 16, 4, tid_to_tblock(tid),
839 			       sizeof(struct tblock), 0);
840 		print_hex_dump(KERN_ERR, "Tlock: ", DUMP_PREFIX_ADDRESS, 16, 4,
841 			       tlck, sizeof(*tlck), 0);
842 		BUG();
843 	}
844 	INCREMENT(stattx.waitlock);	/* statistics */
845 	TXN_UNLOCK();
846 	release_metapage(mp);
847 	TXN_LOCK();
848 	xtid = tlck->tid;	/* reacquire after dropping TXN_LOCK */
849 
850 	jfs_info("txLock: in waitLock, tid = %d, xtid = %d, lid = %d",
851 		 tid, xtid, lid);
852 
853 	/* Recheck everything since dropping TXN_LOCK */
854 	if (xtid && (tlck->mp == mp) && (mp->lid == lid))
855 		TXN_SLEEP_DROP_LOCK(&tid_to_tblock(xtid)->waitor);
856 	else
857 		TXN_UNLOCK();
858 	jfs_info("txLock: awakened     tid = %d, lid = %d", tid, lid);
859 
860 	return NULL;
861 }
862 
863 /*
864  * NAME:	txRelease()
865  *
866  * FUNCTION:	Release buffers associated with transaction locks, but don't
867  *		mark homeok yet.  The allows other transactions to modify
868  *		buffers, but won't let them go to disk until commit record
869  *		actually gets written.
870  *
871  * PARAMETER:
872  *		tblk	-
873  *
874  * RETURN:	Errors from subroutines.
875  */
txRelease(struct tblock * tblk)876 static void txRelease(struct tblock * tblk)
877 {
878 	struct metapage *mp;
879 	lid_t lid;
880 	struct tlock *tlck;
881 
882 	TXN_LOCK();
883 
884 	for (lid = tblk->next; lid; lid = tlck->next) {
885 		tlck = lid_to_tlock(lid);
886 		if ((mp = tlck->mp) != NULL &&
887 		    (tlck->type & tlckBTROOT) == 0) {
888 			assert(mp->xflag & COMMIT_PAGE);
889 			mp->lid = 0;
890 		}
891 	}
892 
893 	/*
894 	 * wakeup transactions waiting on a page locked
895 	 * by the current transaction
896 	 */
897 	TXN_WAKEUP(&tblk->waitor);
898 
899 	TXN_UNLOCK();
900 }
901 
902 /*
903  * NAME:	txUnlock()
904  *
905  * FUNCTION:	Initiates pageout of pages modified by tid in journalled
906  *		objects and frees their lockwords.
907  */
txUnlock(struct tblock * tblk)908 static void txUnlock(struct tblock * tblk)
909 {
910 	struct tlock *tlck;
911 	struct linelock *linelock;
912 	lid_t lid, next, llid, k;
913 	struct metapage *mp;
914 	struct jfs_log *log;
915 	int difft, diffp;
916 	unsigned long flags;
917 
918 	jfs_info("txUnlock: tblk = 0x%p", tblk);
919 	log = JFS_SBI(tblk->sb)->log;
920 
921 	/*
922 	 * mark page under tlock homeok (its log has been written):
923 	 */
924 	for (lid = tblk->next; lid; lid = next) {
925 		tlck = lid_to_tlock(lid);
926 		next = tlck->next;
927 
928 		jfs_info("unlocking lid = %d, tlck = 0x%p", lid, tlck);
929 
930 		/* unbind page from tlock */
931 		if ((mp = tlck->mp) != NULL &&
932 		    (tlck->type & tlckBTROOT) == 0) {
933 			assert(mp->xflag & COMMIT_PAGE);
934 
935 			/* hold buffer
936 			 */
937 			hold_metapage(mp);
938 
939 			assert(mp->nohomeok > 0);
940 			_metapage_homeok(mp);
941 
942 			/* inherit younger/larger clsn */
943 			LOGSYNC_LOCK(log, flags);
944 			if (mp->clsn) {
945 				logdiff(difft, tblk->clsn, log);
946 				logdiff(diffp, mp->clsn, log);
947 				if (difft > diffp)
948 					mp->clsn = tblk->clsn;
949 			} else
950 				mp->clsn = tblk->clsn;
951 			LOGSYNC_UNLOCK(log, flags);
952 
953 			assert(!(tlck->flag & tlckFREEPAGE));
954 
955 			put_metapage(mp);
956 		}
957 
958 		/* insert tlock, and linelock(s) of the tlock if any,
959 		 * at head of freelist
960 		 */
961 		TXN_LOCK();
962 
963 		llid = ((struct linelock *) & tlck->lock)->next;
964 		while (llid) {
965 			linelock = (struct linelock *) lid_to_tlock(llid);
966 			k = linelock->next;
967 			txLockFree(llid);
968 			llid = k;
969 		}
970 		txLockFree(lid);
971 
972 		TXN_UNLOCK();
973 	}
974 	tblk->next = tblk->last = 0;
975 
976 	/*
977 	 * remove tblock from logsynclist
978 	 * (allocation map pages inherited lsn of tblk and
979 	 * has been inserted in logsync list at txUpdateMap())
980 	 */
981 	if (tblk->lsn) {
982 		LOGSYNC_LOCK(log, flags);
983 		log->count--;
984 		list_del(&tblk->synclist);
985 		LOGSYNC_UNLOCK(log, flags);
986 	}
987 }
988 
989 /*
990  *	txMaplock()
991  *
992  * function: allocate a transaction lock for freed page/entry;
993  *	for freed page, maplock is used as xtlock/dtlock type;
994  */
txMaplock(tid_t tid,struct inode * ip,int type)995 struct tlock *txMaplock(tid_t tid, struct inode *ip, int type)
996 {
997 	struct jfs_inode_info *jfs_ip = JFS_IP(ip);
998 	lid_t lid;
999 	struct tblock *tblk;
1000 	struct tlock *tlck;
1001 	struct maplock *maplock;
1002 
1003 	TXN_LOCK();
1004 
1005 	/*
1006 	 * allocate a tlock
1007 	 */
1008 	lid = txLockAlloc();
1009 	tlck = lid_to_tlock(lid);
1010 
1011 	/*
1012 	 * initialize tlock
1013 	 */
1014 	tlck->tid = tid;
1015 
1016 	/* bind the tlock and the object */
1017 	tlck->flag = tlckINODELOCK;
1018 	if (S_ISDIR(ip->i_mode))
1019 		tlck->flag |= tlckDIRECTORY;
1020 	tlck->ip = ip;
1021 	tlck->mp = NULL;
1022 
1023 	tlck->type = type;
1024 
1025 	/*
1026 	 * enqueue transaction lock to transaction/inode
1027 	 */
1028 	/* insert the tlock at tail of transaction tlock list */
1029 	if (tid) {
1030 		tblk = tid_to_tblock(tid);
1031 		if (tblk->next)
1032 			lid_to_tlock(tblk->last)->next = lid;
1033 		else
1034 			tblk->next = lid;
1035 		tlck->next = 0;
1036 		tblk->last = lid;
1037 	}
1038 	/* anonymous transaction:
1039 	 * insert the tlock at head of inode anonymous tlock list
1040 	 */
1041 	else {
1042 		tlck->next = jfs_ip->atlhead;
1043 		jfs_ip->atlhead = lid;
1044 		if (tlck->next == 0) {
1045 			/* This inode's first anonymous transaction */
1046 			jfs_ip->atltail = lid;
1047 			list_add_tail(&jfs_ip->anon_inode_list,
1048 				      &TxAnchor.anon_list);
1049 		}
1050 	}
1051 
1052 	TXN_UNLOCK();
1053 
1054 	/* initialize type dependent area for maplock */
1055 	maplock = (struct maplock *) & tlck->lock;
1056 	maplock->next = 0;
1057 	maplock->maxcnt = 0;
1058 	maplock->index = 0;
1059 
1060 	return tlck;
1061 }
1062 
1063 /*
1064  *	txLinelock()
1065  *
1066  * function: allocate a transaction lock for log vector list
1067  */
txLinelock(struct linelock * tlock)1068 struct linelock *txLinelock(struct linelock * tlock)
1069 {
1070 	lid_t lid;
1071 	struct tlock *tlck;
1072 	struct linelock *linelock;
1073 
1074 	TXN_LOCK();
1075 
1076 	/* allocate a TxLock structure */
1077 	lid = txLockAlloc();
1078 	tlck = lid_to_tlock(lid);
1079 
1080 	TXN_UNLOCK();
1081 
1082 	/* initialize linelock */
1083 	linelock = (struct linelock *) tlck;
1084 	linelock->next = 0;
1085 	linelock->flag = tlckLINELOCK;
1086 	linelock->maxcnt = TLOCKLONG;
1087 	linelock->index = 0;
1088 	if (tlck->flag & tlckDIRECTORY)
1089 		linelock->flag |= tlckDIRECTORY;
1090 
1091 	/* append linelock after tlock */
1092 	linelock->next = tlock->next;
1093 	tlock->next = lid;
1094 
1095 	return linelock;
1096 }
1097 
1098 /*
1099  *		transaction commit management
1100  *		-----------------------------
1101  */
1102 
1103 /*
1104  * NAME:	txCommit()
1105  *
1106  * FUNCTION:	commit the changes to the objects specified in
1107  *		clist.  For journalled segments only the
1108  *		changes of the caller are committed, ie by tid.
1109  *		for non-journalled segments the data are flushed to
1110  *		disk and then the change to the disk inode and indirect
1111  *		blocks committed (so blocks newly allocated to the
1112  *		segment will be made a part of the segment atomically).
1113  *
1114  *		all of the segments specified in clist must be in
1115  *		one file system. no more than 6 segments are needed
1116  *		to handle all unix svcs.
1117  *
1118  *		if the i_nlink field (i.e. disk inode link count)
1119  *		is zero, and the type of inode is a regular file or
1120  *		directory, or symbolic link , the inode is truncated
1121  *		to zero length. the truncation is committed but the
1122  *		VM resources are unaffected until it is closed (see
1123  *		iput and iclose).
1124  *
1125  * PARAMETER:
1126  *
1127  * RETURN:
1128  *
1129  * serialization:
1130  *		on entry the inode lock on each segment is assumed
1131  *		to be held.
1132  *
1133  * i/o error:
1134  */
txCommit(tid_t tid,int nip,struct inode ** iplist,int flag)1135 int txCommit(tid_t tid,		/* transaction identifier */
1136 	     int nip,		/* number of inodes to commit */
1137 	     struct inode **iplist,	/* list of inode to commit */
1138 	     int flag)
1139 {
1140 	int rc = 0;
1141 	struct commit cd;
1142 	struct jfs_log *log;
1143 	struct tblock *tblk;
1144 	struct lrd *lrd;
1145 	struct inode *ip;
1146 	struct jfs_inode_info *jfs_ip;
1147 	int k, n;
1148 	ino_t top;
1149 	struct super_block *sb;
1150 
1151 	jfs_info("txCommit, tid = %d, flag = %d", tid, flag);
1152 	/* is read-only file system ? */
1153 	if (isReadOnly(iplist[0])) {
1154 		rc = -EROFS;
1155 		goto TheEnd;
1156 	}
1157 
1158 	sb = cd.sb = iplist[0]->i_sb;
1159 	cd.tid = tid;
1160 
1161 	if (tid == 0)
1162 		tid = txBegin(sb, 0);
1163 	tblk = tid_to_tblock(tid);
1164 
1165 	/*
1166 	 * initialize commit structure
1167 	 */
1168 	log = JFS_SBI(sb)->log;
1169 	cd.log = log;
1170 
1171 	/* initialize log record descriptor in commit */
1172 	lrd = &cd.lrd;
1173 	lrd->logtid = cpu_to_le32(tblk->logtid);
1174 	lrd->backchain = 0;
1175 
1176 	tblk->xflag |= flag;
1177 
1178 	if ((flag & (COMMIT_FORCE | COMMIT_SYNC)) == 0)
1179 		tblk->xflag |= COMMIT_LAZY;
1180 	/*
1181 	 *	prepare non-journaled objects for commit
1182 	 *
1183 	 * flush data pages of non-journaled file
1184 	 * to prevent the file getting non-initialized disk blocks
1185 	 * in case of crash.
1186 	 * (new blocks - )
1187 	 */
1188 	cd.iplist = iplist;
1189 	cd.nip = nip;
1190 
1191 	/*
1192 	 *	acquire transaction lock on (on-disk) inodes
1193 	 *
1194 	 * update on-disk inode from in-memory inode
1195 	 * acquiring transaction locks for AFTER records
1196 	 * on the on-disk inode of file object
1197 	 *
1198 	 * sort the inodes array by inode number in descending order
1199 	 * to prevent deadlock when acquiring transaction lock
1200 	 * of on-disk inodes on multiple on-disk inode pages by
1201 	 * multiple concurrent transactions
1202 	 */
1203 	for (k = 0; k < cd.nip; k++) {
1204 		top = (cd.iplist[k])->i_ino;
1205 		for (n = k + 1; n < cd.nip; n++) {
1206 			ip = cd.iplist[n];
1207 			if (ip->i_ino > top) {
1208 				top = ip->i_ino;
1209 				cd.iplist[n] = cd.iplist[k];
1210 				cd.iplist[k] = ip;
1211 			}
1212 		}
1213 
1214 		ip = cd.iplist[k];
1215 		jfs_ip = JFS_IP(ip);
1216 
1217 		/*
1218 		 * BUGBUG - This code has temporarily been removed.  The
1219 		 * intent is to ensure that any file data is written before
1220 		 * the metadata is committed to the journal.  This prevents
1221 		 * uninitialized data from appearing in a file after the
1222 		 * journal has been replayed.  (The uninitialized data
1223 		 * could be sensitive data removed by another user.)
1224 		 *
1225 		 * The problem now is that we are holding the IWRITELOCK
1226 		 * on the inode, and calling filemap_fdatawrite on an
1227 		 * unmapped page will cause a deadlock in jfs_get_block.
1228 		 *
1229 		 * The long term solution is to pare down the use of
1230 		 * IWRITELOCK.  We are currently holding it too long.
1231 		 * We could also be smarter about which data pages need
1232 		 * to be written before the transaction is committed and
1233 		 * when we don't need to worry about it at all.
1234 		 *
1235 		 * if ((!S_ISDIR(ip->i_mode))
1236 		 *    && (tblk->flag & COMMIT_DELETE) == 0)
1237 		 *	filemap_write_and_wait(ip->i_mapping);
1238 		 */
1239 
1240 		/*
1241 		 * Mark inode as not dirty.  It will still be on the dirty
1242 		 * inode list, but we'll know not to commit it again unless
1243 		 * it gets marked dirty again
1244 		 */
1245 		clear_cflag(COMMIT_Dirty, ip);
1246 
1247 		/* inherit anonymous tlock(s) of inode */
1248 		if (jfs_ip->atlhead) {
1249 			lid_to_tlock(jfs_ip->atltail)->next = tblk->next;
1250 			tblk->next = jfs_ip->atlhead;
1251 			if (!tblk->last)
1252 				tblk->last = jfs_ip->atltail;
1253 			jfs_ip->atlhead = jfs_ip->atltail = 0;
1254 			TXN_LOCK();
1255 			list_del_init(&jfs_ip->anon_inode_list);
1256 			TXN_UNLOCK();
1257 		}
1258 
1259 		/*
1260 		 * acquire transaction lock on on-disk inode page
1261 		 * (become first tlock of the tblk's tlock list)
1262 		 */
1263 		if (((rc = diWrite(tid, ip))))
1264 			goto out;
1265 	}
1266 
1267 	/*
1268 	 *	write log records from transaction locks
1269 	 *
1270 	 * txUpdateMap() resets XAD_NEW in XAD.
1271 	 */
1272 	if ((rc = txLog(log, tblk, &cd)))
1273 		goto TheEnd;
1274 
1275 	/*
1276 	 * Ensure that inode isn't reused before
1277 	 * lazy commit thread finishes processing
1278 	 */
1279 	if (tblk->xflag & COMMIT_DELETE) {
1280 		ihold(tblk->u.ip);
1281 		/*
1282 		 * Avoid a rare deadlock
1283 		 *
1284 		 * If the inode is locked, we may be blocked in
1285 		 * jfs_commit_inode.  If so, we don't want the
1286 		 * lazy_commit thread doing the last iput() on the inode
1287 		 * since that may block on the locked inode.  Instead,
1288 		 * commit the transaction synchronously, so the last iput
1289 		 * will be done by the calling thread (or later)
1290 		 */
1291 		/*
1292 		 * I believe this code is no longer needed.  Splitting I_LOCK
1293 		 * into two bits, I_NEW and I_SYNC should prevent this
1294 		 * deadlock as well.  But since I don't have a JFS testload
1295 		 * to verify this, only a trivial s/I_LOCK/I_SYNC/ was done.
1296 		 * Joern
1297 		 */
1298 		if (tblk->u.ip->i_state & I_SYNC)
1299 			tblk->xflag &= ~COMMIT_LAZY;
1300 	}
1301 
1302 	ASSERT((!(tblk->xflag & COMMIT_DELETE)) ||
1303 	       ((tblk->u.ip->i_nlink == 0) &&
1304 		!test_cflag(COMMIT_Nolink, tblk->u.ip)));
1305 
1306 	/*
1307 	 *	write COMMIT log record
1308 	 */
1309 	lrd->type = cpu_to_le16(LOG_COMMIT);
1310 	lrd->length = 0;
1311 	lmLog(log, tblk, lrd, NULL);
1312 
1313 	lmGroupCommit(log, tblk);
1314 
1315 	/*
1316 	 *	- transaction is now committed -
1317 	 */
1318 
1319 	/*
1320 	 * force pages in careful update
1321 	 * (imap addressing structure update)
1322 	 */
1323 	if (flag & COMMIT_FORCE)
1324 		txForce(tblk);
1325 
1326 	/*
1327 	 *	update allocation map.
1328 	 *
1329 	 * update inode allocation map and inode:
1330 	 * free pager lock on memory object of inode if any.
1331 	 * update block allocation map.
1332 	 *
1333 	 * txUpdateMap() resets XAD_NEW in XAD.
1334 	 */
1335 	if (tblk->xflag & COMMIT_FORCE)
1336 		txUpdateMap(tblk);
1337 
1338 	/*
1339 	 *	free transaction locks and pageout/free pages
1340 	 */
1341 	txRelease(tblk);
1342 
1343 	if ((tblk->flag & tblkGC_LAZY) == 0)
1344 		txUnlock(tblk);
1345 
1346 
1347 	/*
1348 	 *	reset in-memory object state
1349 	 */
1350 	for (k = 0; k < cd.nip; k++) {
1351 		ip = cd.iplist[k];
1352 		jfs_ip = JFS_IP(ip);
1353 
1354 		/*
1355 		 * reset in-memory inode state
1356 		 */
1357 		jfs_ip->bxflag = 0;
1358 		jfs_ip->blid = 0;
1359 	}
1360 
1361       out:
1362 	if (rc != 0)
1363 		txAbort(tid, 1);
1364 
1365       TheEnd:
1366 	jfs_info("txCommit: tid = %d, returning %d", tid, rc);
1367 	return rc;
1368 }
1369 
1370 /*
1371  * NAME:	txLog()
1372  *
1373  * FUNCTION:	Writes AFTER log records for all lines modified
1374  *		by tid for segments specified by inodes in comdata.
1375  *		Code assumes only WRITELOCKS are recorded in lockwords.
1376  *
1377  * PARAMETERS:
1378  *
1379  * RETURN :
1380  */
txLog(struct jfs_log * log,struct tblock * tblk,struct commit * cd)1381 static int txLog(struct jfs_log * log, struct tblock * tblk, struct commit * cd)
1382 {
1383 	int rc = 0;
1384 	struct inode *ip;
1385 	lid_t lid;
1386 	struct tlock *tlck;
1387 	struct lrd *lrd = &cd->lrd;
1388 
1389 	/*
1390 	 * write log record(s) for each tlock of transaction,
1391 	 */
1392 	for (lid = tblk->next; lid; lid = tlck->next) {
1393 		tlck = lid_to_tlock(lid);
1394 
1395 		tlck->flag |= tlckLOG;
1396 
1397 		/* initialize lrd common */
1398 		ip = tlck->ip;
1399 		lrd->aggregate = cpu_to_le32(JFS_SBI(ip->i_sb)->aggregate);
1400 		lrd->log.redopage.fileset = cpu_to_le32(JFS_IP(ip)->fileset);
1401 		lrd->log.redopage.inode = cpu_to_le32(ip->i_ino);
1402 
1403 		/* write log record of page from the tlock */
1404 		switch (tlck->type & tlckTYPE) {
1405 		case tlckXTREE:
1406 			xtLog(log, tblk, lrd, tlck);
1407 			break;
1408 
1409 		case tlckDTREE:
1410 			dtLog(log, tblk, lrd, tlck);
1411 			break;
1412 
1413 		case tlckINODE:
1414 			diLog(log, tblk, lrd, tlck, cd);
1415 			break;
1416 
1417 		case tlckMAP:
1418 			mapLog(log, tblk, lrd, tlck);
1419 			break;
1420 
1421 		case tlckDATA:
1422 			dataLog(log, tblk, lrd, tlck);
1423 			break;
1424 
1425 		default:
1426 			jfs_err("UFO tlock:0x%p", tlck);
1427 		}
1428 	}
1429 
1430 	return rc;
1431 }
1432 
1433 /*
1434  *	diLog()
1435  *
1436  * function:	log inode tlock and format maplock to update bmap;
1437  */
diLog(struct jfs_log * log,struct tblock * tblk,struct lrd * lrd,struct tlock * tlck,struct commit * cd)1438 static int diLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1439 		 struct tlock * tlck, struct commit * cd)
1440 {
1441 	int rc = 0;
1442 	struct metapage *mp;
1443 	pxd_t *pxd;
1444 	struct pxd_lock *pxdlock;
1445 
1446 	mp = tlck->mp;
1447 
1448 	/* initialize as REDOPAGE record format */
1449 	lrd->log.redopage.type = cpu_to_le16(LOG_INODE);
1450 	lrd->log.redopage.l2linesize = cpu_to_le16(L2INODESLOTSIZE);
1451 
1452 	pxd = &lrd->log.redopage.pxd;
1453 
1454 	/*
1455 	 *	inode after image
1456 	 */
1457 	if (tlck->type & tlckENTRY) {
1458 		/* log after-image for logredo(): */
1459 		lrd->type = cpu_to_le16(LOG_REDOPAGE);
1460 		PXDaddress(pxd, mp->index);
1461 		PXDlength(pxd,
1462 			  mp->logical_size >> tblk->sb->s_blocksize_bits);
1463 		lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1464 
1465 		/* mark page as homeward bound */
1466 		tlck->flag |= tlckWRITEPAGE;
1467 	} else if (tlck->type & tlckFREE) {
1468 		/*
1469 		 *	free inode extent
1470 		 *
1471 		 * (pages of the freed inode extent have been invalidated and
1472 		 * a maplock for free of the extent has been formatted at
1473 		 * txLock() time);
1474 		 *
1475 		 * the tlock had been acquired on the inode allocation map page
1476 		 * (iag) that specifies the freed extent, even though the map
1477 		 * page is not itself logged, to prevent pageout of the map
1478 		 * page before the log;
1479 		 */
1480 
1481 		/* log LOG_NOREDOINOEXT of the freed inode extent for
1482 		 * logredo() to start NoRedoPage filters, and to update
1483 		 * imap and bmap for free of the extent;
1484 		 */
1485 		lrd->type = cpu_to_le16(LOG_NOREDOINOEXT);
1486 		/*
1487 		 * For the LOG_NOREDOINOEXT record, we need
1488 		 * to pass the IAG number and inode extent
1489 		 * index (within that IAG) from which the
1490 		 * the extent being released.  These have been
1491 		 * passed to us in the iplist[1] and iplist[2].
1492 		 */
1493 		lrd->log.noredoinoext.iagnum =
1494 		    cpu_to_le32((u32) (size_t) cd->iplist[1]);
1495 		lrd->log.noredoinoext.inoext_idx =
1496 		    cpu_to_le32((u32) (size_t) cd->iplist[2]);
1497 
1498 		pxdlock = (struct pxd_lock *) & tlck->lock;
1499 		*pxd = pxdlock->pxd;
1500 		lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1501 
1502 		/* update bmap */
1503 		tlck->flag |= tlckUPDATEMAP;
1504 
1505 		/* mark page as homeward bound */
1506 		tlck->flag |= tlckWRITEPAGE;
1507 	} else
1508 		jfs_err("diLog: UFO type tlck:0x%p", tlck);
1509 #ifdef  _JFS_WIP
1510 	/*
1511 	 *	alloc/free external EA extent
1512 	 *
1513 	 * a maplock for txUpdateMap() to update bPWMAP for alloc/free
1514 	 * of the extent has been formatted at txLock() time;
1515 	 */
1516 	else {
1517 		assert(tlck->type & tlckEA);
1518 
1519 		/* log LOG_UPDATEMAP for logredo() to update bmap for
1520 		 * alloc of new (and free of old) external EA extent;
1521 		 */
1522 		lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1523 		pxdlock = (struct pxd_lock *) & tlck->lock;
1524 		nlock = pxdlock->index;
1525 		for (i = 0; i < nlock; i++, pxdlock++) {
1526 			if (pxdlock->flag & mlckALLOCPXD)
1527 				lrd->log.updatemap.type =
1528 				    cpu_to_le16(LOG_ALLOCPXD);
1529 			else
1530 				lrd->log.updatemap.type =
1531 				    cpu_to_le16(LOG_FREEPXD);
1532 			lrd->log.updatemap.nxd = cpu_to_le16(1);
1533 			lrd->log.updatemap.pxd = pxdlock->pxd;
1534 			lrd->backchain =
1535 			    cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1536 		}
1537 
1538 		/* update bmap */
1539 		tlck->flag |= tlckUPDATEMAP;
1540 	}
1541 #endif				/* _JFS_WIP */
1542 
1543 	return rc;
1544 }
1545 
1546 /*
1547  *	dataLog()
1548  *
1549  * function:	log data tlock
1550  */
dataLog(struct jfs_log * log,struct tblock * tblk,struct lrd * lrd,struct tlock * tlck)1551 static int dataLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1552 	    struct tlock * tlck)
1553 {
1554 	struct metapage *mp;
1555 	pxd_t *pxd;
1556 
1557 	mp = tlck->mp;
1558 
1559 	/* initialize as REDOPAGE record format */
1560 	lrd->log.redopage.type = cpu_to_le16(LOG_DATA);
1561 	lrd->log.redopage.l2linesize = cpu_to_le16(L2DATASLOTSIZE);
1562 
1563 	pxd = &lrd->log.redopage.pxd;
1564 
1565 	/* log after-image for logredo(): */
1566 	lrd->type = cpu_to_le16(LOG_REDOPAGE);
1567 
1568 	if (jfs_dirtable_inline(tlck->ip)) {
1569 		/*
1570 		 * The table has been truncated, we've must have deleted
1571 		 * the last entry, so don't bother logging this
1572 		 */
1573 		mp->lid = 0;
1574 		grab_metapage(mp);
1575 		metapage_homeok(mp);
1576 		discard_metapage(mp);
1577 		tlck->mp = NULL;
1578 		return 0;
1579 	}
1580 
1581 	PXDaddress(pxd, mp->index);
1582 	PXDlength(pxd, mp->logical_size >> tblk->sb->s_blocksize_bits);
1583 
1584 	lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1585 
1586 	/* mark page as homeward bound */
1587 	tlck->flag |= tlckWRITEPAGE;
1588 
1589 	return 0;
1590 }
1591 
1592 /*
1593  *	dtLog()
1594  *
1595  * function:	log dtree tlock and format maplock to update bmap;
1596  */
dtLog(struct jfs_log * log,struct tblock * tblk,struct lrd * lrd,struct tlock * tlck)1597 static void dtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1598 	   struct tlock * tlck)
1599 {
1600 	struct metapage *mp;
1601 	struct pxd_lock *pxdlock;
1602 	pxd_t *pxd;
1603 
1604 	mp = tlck->mp;
1605 
1606 	/* initialize as REDOPAGE/NOREDOPAGE record format */
1607 	lrd->log.redopage.type = cpu_to_le16(LOG_DTREE);
1608 	lrd->log.redopage.l2linesize = cpu_to_le16(L2DTSLOTSIZE);
1609 
1610 	pxd = &lrd->log.redopage.pxd;
1611 
1612 	if (tlck->type & tlckBTROOT)
1613 		lrd->log.redopage.type |= cpu_to_le16(LOG_BTROOT);
1614 
1615 	/*
1616 	 *	page extension via relocation: entry insertion;
1617 	 *	page extension in-place: entry insertion;
1618 	 *	new right page from page split, reinitialized in-line
1619 	 *	root from root page split: entry insertion;
1620 	 */
1621 	if (tlck->type & (tlckNEW | tlckEXTEND)) {
1622 		/* log after-image of the new page for logredo():
1623 		 * mark log (LOG_NEW) for logredo() to initialize
1624 		 * freelist and update bmap for alloc of the new page;
1625 		 */
1626 		lrd->type = cpu_to_le16(LOG_REDOPAGE);
1627 		if (tlck->type & tlckEXTEND)
1628 			lrd->log.redopage.type |= cpu_to_le16(LOG_EXTEND);
1629 		else
1630 			lrd->log.redopage.type |= cpu_to_le16(LOG_NEW);
1631 		PXDaddress(pxd, mp->index);
1632 		PXDlength(pxd,
1633 			  mp->logical_size >> tblk->sb->s_blocksize_bits);
1634 		lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1635 
1636 		/* format a maplock for txUpdateMap() to update bPMAP for
1637 		 * alloc of the new page;
1638 		 */
1639 		if (tlck->type & tlckBTROOT)
1640 			return;
1641 		tlck->flag |= tlckUPDATEMAP;
1642 		pxdlock = (struct pxd_lock *) & tlck->lock;
1643 		pxdlock->flag = mlckALLOCPXD;
1644 		pxdlock->pxd = *pxd;
1645 
1646 		pxdlock->index = 1;
1647 
1648 		/* mark page as homeward bound */
1649 		tlck->flag |= tlckWRITEPAGE;
1650 		return;
1651 	}
1652 
1653 	/*
1654 	 *	entry insertion/deletion,
1655 	 *	sibling page link update (old right page before split);
1656 	 */
1657 	if (tlck->type & (tlckENTRY | tlckRELINK)) {
1658 		/* log after-image for logredo(): */
1659 		lrd->type = cpu_to_le16(LOG_REDOPAGE);
1660 		PXDaddress(pxd, mp->index);
1661 		PXDlength(pxd,
1662 			  mp->logical_size >> tblk->sb->s_blocksize_bits);
1663 		lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1664 
1665 		/* mark page as homeward bound */
1666 		tlck->flag |= tlckWRITEPAGE;
1667 		return;
1668 	}
1669 
1670 	/*
1671 	 *	page deletion: page has been invalidated
1672 	 *	page relocation: source extent
1673 	 *
1674 	 *	a maplock for free of the page has been formatted
1675 	 *	at txLock() time);
1676 	 */
1677 	if (tlck->type & (tlckFREE | tlckRELOCATE)) {
1678 		/* log LOG_NOREDOPAGE of the deleted page for logredo()
1679 		 * to start NoRedoPage filter and to update bmap for free
1680 		 * of the deletd page
1681 		 */
1682 		lrd->type = cpu_to_le16(LOG_NOREDOPAGE);
1683 		pxdlock = (struct pxd_lock *) & tlck->lock;
1684 		*pxd = pxdlock->pxd;
1685 		lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1686 
1687 		/* a maplock for txUpdateMap() for free of the page
1688 		 * has been formatted at txLock() time;
1689 		 */
1690 		tlck->flag |= tlckUPDATEMAP;
1691 	}
1692 	return;
1693 }
1694 
1695 /*
1696  *	xtLog()
1697  *
1698  * function:	log xtree tlock and format maplock to update bmap;
1699  */
xtLog(struct jfs_log * log,struct tblock * tblk,struct lrd * lrd,struct tlock * tlck)1700 static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1701 	   struct tlock * tlck)
1702 {
1703 	struct inode *ip;
1704 	struct metapage *mp;
1705 	xtpage_t *p;
1706 	struct xtlock *xtlck;
1707 	struct maplock *maplock;
1708 	struct xdlistlock *xadlock;
1709 	struct pxd_lock *pxdlock;
1710 	pxd_t *page_pxd;
1711 	int next, lwm, hwm;
1712 
1713 	ip = tlck->ip;
1714 	mp = tlck->mp;
1715 
1716 	/* initialize as REDOPAGE/NOREDOPAGE record format */
1717 	lrd->log.redopage.type = cpu_to_le16(LOG_XTREE);
1718 	lrd->log.redopage.l2linesize = cpu_to_le16(L2XTSLOTSIZE);
1719 
1720 	page_pxd = &lrd->log.redopage.pxd;
1721 
1722 	if (tlck->type & tlckBTROOT) {
1723 		lrd->log.redopage.type |= cpu_to_le16(LOG_BTROOT);
1724 		p = &JFS_IP(ip)->i_xtroot;
1725 		if (S_ISDIR(ip->i_mode))
1726 			lrd->log.redopage.type |=
1727 			    cpu_to_le16(LOG_DIR_XTREE);
1728 	} else
1729 		p = (xtpage_t *) mp->data;
1730 	next = le16_to_cpu(p->header.nextindex);
1731 
1732 	xtlck = (struct xtlock *) & tlck->lock;
1733 
1734 	maplock = (struct maplock *) & tlck->lock;
1735 	xadlock = (struct xdlistlock *) maplock;
1736 
1737 	/*
1738 	 *	entry insertion/extension;
1739 	 *	sibling page link update (old right page before split);
1740 	 */
1741 	if (tlck->type & (tlckNEW | tlckGROW | tlckRELINK)) {
1742 		/* log after-image for logredo():
1743 		 * logredo() will update bmap for alloc of new/extended
1744 		 * extents (XAD_NEW|XAD_EXTEND) of XAD[lwm:next) from
1745 		 * after-image of XADlist;
1746 		 * logredo() resets (XAD_NEW|XAD_EXTEND) flag when
1747 		 * applying the after-image to the meta-data page.
1748 		 */
1749 		lrd->type = cpu_to_le16(LOG_REDOPAGE);
1750 		PXDaddress(page_pxd, mp->index);
1751 		PXDlength(page_pxd,
1752 			  mp->logical_size >> tblk->sb->s_blocksize_bits);
1753 		lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1754 
1755 		/* format a maplock for txUpdateMap() to update bPMAP
1756 		 * for alloc of new/extended extents of XAD[lwm:next)
1757 		 * from the page itself;
1758 		 * txUpdateMap() resets (XAD_NEW|XAD_EXTEND) flag.
1759 		 */
1760 		lwm = xtlck->lwm.offset;
1761 		if (lwm == 0)
1762 			lwm = XTPAGEMAXSLOT;
1763 
1764 		if (lwm == next)
1765 			goto out;
1766 		if (lwm > next) {
1767 			jfs_err("xtLog: lwm > next\n");
1768 			goto out;
1769 		}
1770 		tlck->flag |= tlckUPDATEMAP;
1771 		xadlock->flag = mlckALLOCXADLIST;
1772 		xadlock->count = next - lwm;
1773 		if ((xadlock->count <= 4) && (tblk->xflag & COMMIT_LAZY)) {
1774 			int i;
1775 			pxd_t *pxd;
1776 			/*
1777 			 * Lazy commit may allow xtree to be modified before
1778 			 * txUpdateMap runs.  Copy xad into linelock to
1779 			 * preserve correct data.
1780 			 *
1781 			 * We can fit twice as may pxd's as xads in the lock
1782 			 */
1783 			xadlock->flag = mlckALLOCPXDLIST;
1784 			pxd = xadlock->xdlist = &xtlck->pxdlock;
1785 			for (i = 0; i < xadlock->count; i++) {
1786 				PXDaddress(pxd, addressXAD(&p->xad[lwm + i]));
1787 				PXDlength(pxd, lengthXAD(&p->xad[lwm + i]));
1788 				p->xad[lwm + i].flag &=
1789 				    ~(XAD_NEW | XAD_EXTENDED);
1790 				pxd++;
1791 			}
1792 		} else {
1793 			/*
1794 			 * xdlist will point to into inode's xtree, ensure
1795 			 * that transaction is not committed lazily.
1796 			 */
1797 			xadlock->flag = mlckALLOCXADLIST;
1798 			xadlock->xdlist = &p->xad[lwm];
1799 			tblk->xflag &= ~COMMIT_LAZY;
1800 		}
1801 		jfs_info("xtLog: alloc ip:0x%p mp:0x%p tlck:0x%p lwm:%d "
1802 			 "count:%d", tlck->ip, mp, tlck, lwm, xadlock->count);
1803 
1804 		maplock->index = 1;
1805 
1806 	      out:
1807 		/* mark page as homeward bound */
1808 		tlck->flag |= tlckWRITEPAGE;
1809 
1810 		return;
1811 	}
1812 
1813 	/*
1814 	 *	page deletion: file deletion/truncation (ref. xtTruncate())
1815 	 *
1816 	 * (page will be invalidated after log is written and bmap
1817 	 * is updated from the page);
1818 	 */
1819 	if (tlck->type & tlckFREE) {
1820 		/* LOG_NOREDOPAGE log for NoRedoPage filter:
1821 		 * if page free from file delete, NoRedoFile filter from
1822 		 * inode image of zero link count will subsume NoRedoPage
1823 		 * filters for each page;
1824 		 * if page free from file truncattion, write NoRedoPage
1825 		 * filter;
1826 		 *
1827 		 * upadte of block allocation map for the page itself:
1828 		 * if page free from deletion and truncation, LOG_UPDATEMAP
1829 		 * log for the page itself is generated from processing
1830 		 * its parent page xad entries;
1831 		 */
1832 		/* if page free from file truncation, log LOG_NOREDOPAGE
1833 		 * of the deleted page for logredo() to start NoRedoPage
1834 		 * filter for the page;
1835 		 */
1836 		if (tblk->xflag & COMMIT_TRUNCATE) {
1837 			/* write NOREDOPAGE for the page */
1838 			lrd->type = cpu_to_le16(LOG_NOREDOPAGE);
1839 			PXDaddress(page_pxd, mp->index);
1840 			PXDlength(page_pxd,
1841 				  mp->logical_size >> tblk->sb->
1842 				  s_blocksize_bits);
1843 			lrd->backchain =
1844 			    cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1845 
1846 			if (tlck->type & tlckBTROOT) {
1847 				/* Empty xtree must be logged */
1848 				lrd->type = cpu_to_le16(LOG_REDOPAGE);
1849 				lrd->backchain =
1850 				    cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1851 			}
1852 		}
1853 
1854 		/* init LOG_UPDATEMAP of the freed extents
1855 		 * XAD[XTENTRYSTART:hwm) from the deleted page itself
1856 		 * for logredo() to update bmap;
1857 		 */
1858 		lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1859 		lrd->log.updatemap.type = cpu_to_le16(LOG_FREEXADLIST);
1860 		xtlck = (struct xtlock *) & tlck->lock;
1861 		hwm = xtlck->hwm.offset;
1862 		lrd->log.updatemap.nxd =
1863 		    cpu_to_le16(hwm - XTENTRYSTART + 1);
1864 		/* reformat linelock for lmLog() */
1865 		xtlck->header.offset = XTENTRYSTART;
1866 		xtlck->header.length = hwm - XTENTRYSTART + 1;
1867 		xtlck->index = 1;
1868 		lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1869 
1870 		/* format a maplock for txUpdateMap() to update bmap
1871 		 * to free extents of XAD[XTENTRYSTART:hwm) from the
1872 		 * deleted page itself;
1873 		 */
1874 		tlck->flag |= tlckUPDATEMAP;
1875 		xadlock->count = hwm - XTENTRYSTART + 1;
1876 		if ((xadlock->count <= 4) && (tblk->xflag & COMMIT_LAZY)) {
1877 			int i;
1878 			pxd_t *pxd;
1879 			/*
1880 			 * Lazy commit may allow xtree to be modified before
1881 			 * txUpdateMap runs.  Copy xad into linelock to
1882 			 * preserve correct data.
1883 			 *
1884 			 * We can fit twice as may pxd's as xads in the lock
1885 			 */
1886 			xadlock->flag = mlckFREEPXDLIST;
1887 			pxd = xadlock->xdlist = &xtlck->pxdlock;
1888 			for (i = 0; i < xadlock->count; i++) {
1889 				PXDaddress(pxd,
1890 					addressXAD(&p->xad[XTENTRYSTART + i]));
1891 				PXDlength(pxd,
1892 					lengthXAD(&p->xad[XTENTRYSTART + i]));
1893 				pxd++;
1894 			}
1895 		} else {
1896 			/*
1897 			 * xdlist will point to into inode's xtree, ensure
1898 			 * that transaction is not committed lazily.
1899 			 */
1900 			xadlock->flag = mlckFREEXADLIST;
1901 			xadlock->xdlist = &p->xad[XTENTRYSTART];
1902 			tblk->xflag &= ~COMMIT_LAZY;
1903 		}
1904 		jfs_info("xtLog: free ip:0x%p mp:0x%p count:%d lwm:2",
1905 			 tlck->ip, mp, xadlock->count);
1906 
1907 		maplock->index = 1;
1908 
1909 		/* mark page as invalid */
1910 		if (((tblk->xflag & COMMIT_PWMAP) || S_ISDIR(ip->i_mode))
1911 		    && !(tlck->type & tlckBTROOT))
1912 			tlck->flag |= tlckFREEPAGE;
1913 		/*
1914 		   else (tblk->xflag & COMMIT_PMAP)
1915 		   ? release the page;
1916 		 */
1917 		return;
1918 	}
1919 
1920 	/*
1921 	 *	page/entry truncation: file truncation (ref. xtTruncate())
1922 	 *
1923 	 *	|----------+------+------+---------------|
1924 	 *		   |      |      |
1925 	 *		   |      |     hwm - hwm before truncation
1926 	 *		   |     next - truncation point
1927 	 *		  lwm - lwm before truncation
1928 	 * header ?
1929 	 */
1930 	if (tlck->type & tlckTRUNCATE) {
1931 		/* This odd declaration suppresses a bogus gcc warning */
1932 		pxd_t pxd = pxd;	/* truncated extent of xad */
1933 		int twm;
1934 
1935 		/*
1936 		 * For truncation the entire linelock may be used, so it would
1937 		 * be difficult to store xad list in linelock itself.
1938 		 * Therefore, we'll just force transaction to be committed
1939 		 * synchronously, so that xtree pages won't be changed before
1940 		 * txUpdateMap runs.
1941 		 */
1942 		tblk->xflag &= ~COMMIT_LAZY;
1943 		lwm = xtlck->lwm.offset;
1944 		if (lwm == 0)
1945 			lwm = XTPAGEMAXSLOT;
1946 		hwm = xtlck->hwm.offset;
1947 		twm = xtlck->twm.offset;
1948 
1949 		/*
1950 		 *	write log records
1951 		 */
1952 		/* log after-image for logredo():
1953 		 *
1954 		 * logredo() will update bmap for alloc of new/extended
1955 		 * extents (XAD_NEW|XAD_EXTEND) of XAD[lwm:next) from
1956 		 * after-image of XADlist;
1957 		 * logredo() resets (XAD_NEW|XAD_EXTEND) flag when
1958 		 * applying the after-image to the meta-data page.
1959 		 */
1960 		lrd->type = cpu_to_le16(LOG_REDOPAGE);
1961 		PXDaddress(page_pxd, mp->index);
1962 		PXDlength(page_pxd,
1963 			  mp->logical_size >> tblk->sb->s_blocksize_bits);
1964 		lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1965 
1966 		/*
1967 		 * truncate entry XAD[twm == next - 1]:
1968 		 */
1969 		if (twm == next - 1) {
1970 			/* init LOG_UPDATEMAP for logredo() to update bmap for
1971 			 * free of truncated delta extent of the truncated
1972 			 * entry XAD[next - 1]:
1973 			 * (xtlck->pxdlock = truncated delta extent);
1974 			 */
1975 			pxdlock = (struct pxd_lock *) & xtlck->pxdlock;
1976 			/* assert(pxdlock->type & tlckTRUNCATE); */
1977 			lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1978 			lrd->log.updatemap.type = cpu_to_le16(LOG_FREEPXD);
1979 			lrd->log.updatemap.nxd = cpu_to_le16(1);
1980 			lrd->log.updatemap.pxd = pxdlock->pxd;
1981 			pxd = pxdlock->pxd;	/* save to format maplock */
1982 			lrd->backchain =
1983 			    cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1984 		}
1985 
1986 		/*
1987 		 * free entries XAD[next:hwm]:
1988 		 */
1989 		if (hwm >= next) {
1990 			/* init LOG_UPDATEMAP of the freed extents
1991 			 * XAD[next:hwm] from the deleted page itself
1992 			 * for logredo() to update bmap;
1993 			 */
1994 			lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1995 			lrd->log.updatemap.type =
1996 			    cpu_to_le16(LOG_FREEXADLIST);
1997 			xtlck = (struct xtlock *) & tlck->lock;
1998 			hwm = xtlck->hwm.offset;
1999 			lrd->log.updatemap.nxd =
2000 			    cpu_to_le16(hwm - next + 1);
2001 			/* reformat linelock for lmLog() */
2002 			xtlck->header.offset = next;
2003 			xtlck->header.length = hwm - next + 1;
2004 			xtlck->index = 1;
2005 			lrd->backchain =
2006 			    cpu_to_le32(lmLog(log, tblk, lrd, tlck));
2007 		}
2008 
2009 		/*
2010 		 *	format maplock(s) for txUpdateMap() to update bmap
2011 		 */
2012 		maplock->index = 0;
2013 
2014 		/*
2015 		 * allocate entries XAD[lwm:next):
2016 		 */
2017 		if (lwm < next) {
2018 			/* format a maplock for txUpdateMap() to update bPMAP
2019 			 * for alloc of new/extended extents of XAD[lwm:next)
2020 			 * from the page itself;
2021 			 * txUpdateMap() resets (XAD_NEW|XAD_EXTEND) flag.
2022 			 */
2023 			tlck->flag |= tlckUPDATEMAP;
2024 			xadlock->flag = mlckALLOCXADLIST;
2025 			xadlock->count = next - lwm;
2026 			xadlock->xdlist = &p->xad[lwm];
2027 
2028 			jfs_info("xtLog: alloc ip:0x%p mp:0x%p count:%d "
2029 				 "lwm:%d next:%d",
2030 				 tlck->ip, mp, xadlock->count, lwm, next);
2031 			maplock->index++;
2032 			xadlock++;
2033 		}
2034 
2035 		/*
2036 		 * truncate entry XAD[twm == next - 1]:
2037 		 */
2038 		if (twm == next - 1) {
2039 			/* format a maplock for txUpdateMap() to update bmap
2040 			 * to free truncated delta extent of the truncated
2041 			 * entry XAD[next - 1];
2042 			 * (xtlck->pxdlock = truncated delta extent);
2043 			 */
2044 			tlck->flag |= tlckUPDATEMAP;
2045 			pxdlock = (struct pxd_lock *) xadlock;
2046 			pxdlock->flag = mlckFREEPXD;
2047 			pxdlock->count = 1;
2048 			pxdlock->pxd = pxd;
2049 
2050 			jfs_info("xtLog: truncate ip:0x%p mp:0x%p count:%d "
2051 				 "hwm:%d", ip, mp, pxdlock->count, hwm);
2052 			maplock->index++;
2053 			xadlock++;
2054 		}
2055 
2056 		/*
2057 		 * free entries XAD[next:hwm]:
2058 		 */
2059 		if (hwm >= next) {
2060 			/* format a maplock for txUpdateMap() to update bmap
2061 			 * to free extents of XAD[next:hwm] from thedeleted
2062 			 * page itself;
2063 			 */
2064 			tlck->flag |= tlckUPDATEMAP;
2065 			xadlock->flag = mlckFREEXADLIST;
2066 			xadlock->count = hwm - next + 1;
2067 			xadlock->xdlist = &p->xad[next];
2068 
2069 			jfs_info("xtLog: free ip:0x%p mp:0x%p count:%d "
2070 				 "next:%d hwm:%d",
2071 				 tlck->ip, mp, xadlock->count, next, hwm);
2072 			maplock->index++;
2073 		}
2074 
2075 		/* mark page as homeward bound */
2076 		tlck->flag |= tlckWRITEPAGE;
2077 	}
2078 	return;
2079 }
2080 
2081 /*
2082  *	mapLog()
2083  *
2084  * function:	log from maplock of freed data extents;
2085  */
mapLog(struct jfs_log * log,struct tblock * tblk,struct lrd * lrd,struct tlock * tlck)2086 static void mapLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
2087 		   struct tlock * tlck)
2088 {
2089 	struct pxd_lock *pxdlock;
2090 	int i, nlock;
2091 	pxd_t *pxd;
2092 
2093 	/*
2094 	 *	page relocation: free the source page extent
2095 	 *
2096 	 * a maplock for txUpdateMap() for free of the page
2097 	 * has been formatted at txLock() time saving the src
2098 	 * relocated page address;
2099 	 */
2100 	if (tlck->type & tlckRELOCATE) {
2101 		/* log LOG_NOREDOPAGE of the old relocated page
2102 		 * for logredo() to start NoRedoPage filter;
2103 		 */
2104 		lrd->type = cpu_to_le16(LOG_NOREDOPAGE);
2105 		pxdlock = (struct pxd_lock *) & tlck->lock;
2106 		pxd = &lrd->log.redopage.pxd;
2107 		*pxd = pxdlock->pxd;
2108 		lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
2109 
2110 		/* (N.B. currently, logredo() does NOT update bmap
2111 		 * for free of the page itself for (LOG_XTREE|LOG_NOREDOPAGE);
2112 		 * if page free from relocation, LOG_UPDATEMAP log is
2113 		 * specifically generated now for logredo()
2114 		 * to update bmap for free of src relocated page;
2115 		 * (new flag LOG_RELOCATE may be introduced which will
2116 		 * inform logredo() to start NORedoPage filter and also
2117 		 * update block allocation map at the same time, thus
2118 		 * avoiding an extra log write);
2119 		 */
2120 		lrd->type = cpu_to_le16(LOG_UPDATEMAP);
2121 		lrd->log.updatemap.type = cpu_to_le16(LOG_FREEPXD);
2122 		lrd->log.updatemap.nxd = cpu_to_le16(1);
2123 		lrd->log.updatemap.pxd = pxdlock->pxd;
2124 		lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
2125 
2126 		/* a maplock for txUpdateMap() for free of the page
2127 		 * has been formatted at txLock() time;
2128 		 */
2129 		tlck->flag |= tlckUPDATEMAP;
2130 		return;
2131 	}
2132 	/*
2133 
2134 	 * Otherwise it's not a relocate request
2135 	 *
2136 	 */
2137 	else {
2138 		/* log LOG_UPDATEMAP for logredo() to update bmap for
2139 		 * free of truncated/relocated delta extent of the data;
2140 		 * e.g.: external EA extent, relocated/truncated extent
2141 		 * from xtTailgate();
2142 		 */
2143 		lrd->type = cpu_to_le16(LOG_UPDATEMAP);
2144 		pxdlock = (struct pxd_lock *) & tlck->lock;
2145 		nlock = pxdlock->index;
2146 		for (i = 0; i < nlock; i++, pxdlock++) {
2147 			if (pxdlock->flag & mlckALLOCPXD)
2148 				lrd->log.updatemap.type =
2149 				    cpu_to_le16(LOG_ALLOCPXD);
2150 			else
2151 				lrd->log.updatemap.type =
2152 				    cpu_to_le16(LOG_FREEPXD);
2153 			lrd->log.updatemap.nxd = cpu_to_le16(1);
2154 			lrd->log.updatemap.pxd = pxdlock->pxd;
2155 			lrd->backchain =
2156 			    cpu_to_le32(lmLog(log, tblk, lrd, NULL));
2157 			jfs_info("mapLog: xaddr:0x%lx xlen:0x%x",
2158 				 (ulong) addressPXD(&pxdlock->pxd),
2159 				 lengthPXD(&pxdlock->pxd));
2160 		}
2161 
2162 		/* update bmap */
2163 		tlck->flag |= tlckUPDATEMAP;
2164 	}
2165 }
2166 
2167 /*
2168  *	txEA()
2169  *
2170  * function:	acquire maplock for EA/ACL extents or
2171  *		set COMMIT_INLINE flag;
2172  */
txEA(tid_t tid,struct inode * ip,dxd_t * oldea,dxd_t * newea)2173 void txEA(tid_t tid, struct inode *ip, dxd_t * oldea, dxd_t * newea)
2174 {
2175 	struct tlock *tlck = NULL;
2176 	struct pxd_lock *maplock = NULL, *pxdlock = NULL;
2177 
2178 	/*
2179 	 * format maplock for alloc of new EA extent
2180 	 */
2181 	if (newea) {
2182 		/* Since the newea could be a completely zeroed entry we need to
2183 		 * check for the two flags which indicate we should actually
2184 		 * commit new EA data
2185 		 */
2186 		if (newea->flag & DXD_EXTENT) {
2187 			tlck = txMaplock(tid, ip, tlckMAP);
2188 			maplock = (struct pxd_lock *) & tlck->lock;
2189 			pxdlock = (struct pxd_lock *) maplock;
2190 			pxdlock->flag = mlckALLOCPXD;
2191 			PXDaddress(&pxdlock->pxd, addressDXD(newea));
2192 			PXDlength(&pxdlock->pxd, lengthDXD(newea));
2193 			pxdlock++;
2194 			maplock->index = 1;
2195 		} else if (newea->flag & DXD_INLINE) {
2196 			tlck = NULL;
2197 
2198 			set_cflag(COMMIT_Inlineea, ip);
2199 		}
2200 	}
2201 
2202 	/*
2203 	 * format maplock for free of old EA extent
2204 	 */
2205 	if (!test_cflag(COMMIT_Nolink, ip) && oldea->flag & DXD_EXTENT) {
2206 		if (tlck == NULL) {
2207 			tlck = txMaplock(tid, ip, tlckMAP);
2208 			maplock = (struct pxd_lock *) & tlck->lock;
2209 			pxdlock = (struct pxd_lock *) maplock;
2210 			maplock->index = 0;
2211 		}
2212 		pxdlock->flag = mlckFREEPXD;
2213 		PXDaddress(&pxdlock->pxd, addressDXD(oldea));
2214 		PXDlength(&pxdlock->pxd, lengthDXD(oldea));
2215 		maplock->index++;
2216 	}
2217 }
2218 
2219 /*
2220  *	txForce()
2221  *
2222  * function: synchronously write pages locked by transaction
2223  *	     after txLog() but before txUpdateMap();
2224  */
txForce(struct tblock * tblk)2225 static void txForce(struct tblock * tblk)
2226 {
2227 	struct tlock *tlck;
2228 	lid_t lid, next;
2229 	struct metapage *mp;
2230 
2231 	/*
2232 	 * reverse the order of transaction tlocks in
2233 	 * careful update order of address index pages
2234 	 * (right to left, bottom up)
2235 	 */
2236 	tlck = lid_to_tlock(tblk->next);
2237 	lid = tlck->next;
2238 	tlck->next = 0;
2239 	while (lid) {
2240 		tlck = lid_to_tlock(lid);
2241 		next = tlck->next;
2242 		tlck->next = tblk->next;
2243 		tblk->next = lid;
2244 		lid = next;
2245 	}
2246 
2247 	/*
2248 	 * synchronously write the page, and
2249 	 * hold the page for txUpdateMap();
2250 	 */
2251 	for (lid = tblk->next; lid; lid = next) {
2252 		tlck = lid_to_tlock(lid);
2253 		next = tlck->next;
2254 
2255 		if ((mp = tlck->mp) != NULL &&
2256 		    (tlck->type & tlckBTROOT) == 0) {
2257 			assert(mp->xflag & COMMIT_PAGE);
2258 
2259 			if (tlck->flag & tlckWRITEPAGE) {
2260 				tlck->flag &= ~tlckWRITEPAGE;
2261 
2262 				/* do not release page to freelist */
2263 				force_metapage(mp);
2264 #if 0
2265 				/*
2266 				 * The "right" thing to do here is to
2267 				 * synchronously write the metadata.
2268 				 * With the current implementation this
2269 				 * is hard since write_metapage requires
2270 				 * us to kunmap & remap the page.  If we
2271 				 * have tlocks pointing into the metadata
2272 				 * pages, we don't want to do this.  I think
2273 				 * we can get by with synchronously writing
2274 				 * the pages when they are released.
2275 				 */
2276 				assert(mp->nohomeok);
2277 				set_bit(META_dirty, &mp->flag);
2278 				set_bit(META_sync, &mp->flag);
2279 #endif
2280 			}
2281 		}
2282 	}
2283 }
2284 
2285 /*
2286  *	txUpdateMap()
2287  *
2288  * function:	update persistent allocation map (and working map
2289  *		if appropriate);
2290  *
2291  * parameter:
2292  */
txUpdateMap(struct tblock * tblk)2293 static void txUpdateMap(struct tblock * tblk)
2294 {
2295 	struct inode *ip;
2296 	struct inode *ipimap;
2297 	lid_t lid;
2298 	struct tlock *tlck;
2299 	struct maplock *maplock;
2300 	struct pxd_lock pxdlock;
2301 	int maptype;
2302 	int k, nlock;
2303 	struct metapage *mp = NULL;
2304 
2305 	ipimap = JFS_SBI(tblk->sb)->ipimap;
2306 
2307 	maptype = (tblk->xflag & COMMIT_PMAP) ? COMMIT_PMAP : COMMIT_PWMAP;
2308 
2309 
2310 	/*
2311 	 *	update block allocation map
2312 	 *
2313 	 * update allocation state in pmap (and wmap) and
2314 	 * update lsn of the pmap page;
2315 	 */
2316 	/*
2317 	 * scan each tlock/page of transaction for block allocation/free:
2318 	 *
2319 	 * for each tlock/page of transaction, update map.
2320 	 *  ? are there tlock for pmap and pwmap at the same time ?
2321 	 */
2322 	for (lid = tblk->next; lid; lid = tlck->next) {
2323 		tlck = lid_to_tlock(lid);
2324 
2325 		if ((tlck->flag & tlckUPDATEMAP) == 0)
2326 			continue;
2327 
2328 		if (tlck->flag & tlckFREEPAGE) {
2329 			/*
2330 			 * Another thread may attempt to reuse freed space
2331 			 * immediately, so we want to get rid of the metapage
2332 			 * before anyone else has a chance to get it.
2333 			 * Lock metapage, update maps, then invalidate
2334 			 * the metapage.
2335 			 */
2336 			mp = tlck->mp;
2337 			ASSERT(mp->xflag & COMMIT_PAGE);
2338 			grab_metapage(mp);
2339 		}
2340 
2341 		/*
2342 		 * extent list:
2343 		 * . in-line PXD list:
2344 		 * . out-of-line XAD list:
2345 		 */
2346 		maplock = (struct maplock *) & tlck->lock;
2347 		nlock = maplock->index;
2348 
2349 		for (k = 0; k < nlock; k++, maplock++) {
2350 			/*
2351 			 * allocate blocks in persistent map:
2352 			 *
2353 			 * blocks have been allocated from wmap at alloc time;
2354 			 */
2355 			if (maplock->flag & mlckALLOC) {
2356 				txAllocPMap(ipimap, maplock, tblk);
2357 			}
2358 			/*
2359 			 * free blocks in persistent and working map:
2360 			 * blocks will be freed in pmap and then in wmap;
2361 			 *
2362 			 * ? tblock specifies the PMAP/PWMAP based upon
2363 			 * transaction
2364 			 *
2365 			 * free blocks in persistent map:
2366 			 * blocks will be freed from wmap at last reference
2367 			 * release of the object for regular files;
2368 			 *
2369 			 * Alway free blocks from both persistent & working
2370 			 * maps for directories
2371 			 */
2372 			else {	/* (maplock->flag & mlckFREE) */
2373 
2374 				if (tlck->flag & tlckDIRECTORY)
2375 					txFreeMap(ipimap, maplock,
2376 						  tblk, COMMIT_PWMAP);
2377 				else
2378 					txFreeMap(ipimap, maplock,
2379 						  tblk, maptype);
2380 			}
2381 		}
2382 		if (tlck->flag & tlckFREEPAGE) {
2383 			if (!(tblk->flag & tblkGC_LAZY)) {
2384 				/* This is equivalent to txRelease */
2385 				ASSERT(mp->lid == lid);
2386 				tlck->mp->lid = 0;
2387 			}
2388 			assert(mp->nohomeok == 1);
2389 			metapage_homeok(mp);
2390 			discard_metapage(mp);
2391 			tlck->mp = NULL;
2392 		}
2393 	}
2394 	/*
2395 	 *	update inode allocation map
2396 	 *
2397 	 * update allocation state in pmap and
2398 	 * update lsn of the pmap page;
2399 	 * update in-memory inode flag/state
2400 	 *
2401 	 * unlock mapper/write lock
2402 	 */
2403 	if (tblk->xflag & COMMIT_CREATE) {
2404 		diUpdatePMap(ipimap, tblk->ino, false, tblk);
2405 		/* update persistent block allocation map
2406 		 * for the allocation of inode extent;
2407 		 */
2408 		pxdlock.flag = mlckALLOCPXD;
2409 		pxdlock.pxd = tblk->u.ixpxd;
2410 		pxdlock.index = 1;
2411 		txAllocPMap(ipimap, (struct maplock *) & pxdlock, tblk);
2412 	} else if (tblk->xflag & COMMIT_DELETE) {
2413 		ip = tblk->u.ip;
2414 		diUpdatePMap(ipimap, ip->i_ino, true, tblk);
2415 		iput(ip);
2416 	}
2417 }
2418 
2419 /*
2420  *	txAllocPMap()
2421  *
2422  * function: allocate from persistent map;
2423  *
2424  * parameter:
2425  *	ipbmap	-
2426  *	malock	-
2427  *		xad list:
2428  *		pxd:
2429  *
2430  *	maptype -
2431  *		allocate from persistent map;
2432  *		free from persistent map;
2433  *		(e.g., tmp file - free from working map at releae
2434  *		 of last reference);
2435  *		free from persistent and working map;
2436  *
2437  *	lsn	- log sequence number;
2438  */
txAllocPMap(struct inode * ip,struct maplock * maplock,struct tblock * tblk)2439 static void txAllocPMap(struct inode *ip, struct maplock * maplock,
2440 			struct tblock * tblk)
2441 {
2442 	struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
2443 	struct xdlistlock *xadlistlock;
2444 	xad_t *xad;
2445 	s64 xaddr;
2446 	int xlen;
2447 	struct pxd_lock *pxdlock;
2448 	struct xdlistlock *pxdlistlock;
2449 	pxd_t *pxd;
2450 	int n;
2451 
2452 	/*
2453 	 * allocate from persistent map;
2454 	 */
2455 	if (maplock->flag & mlckALLOCXADLIST) {
2456 		xadlistlock = (struct xdlistlock *) maplock;
2457 		xad = xadlistlock->xdlist;
2458 		for (n = 0; n < xadlistlock->count; n++, xad++) {
2459 			if (xad->flag & (XAD_NEW | XAD_EXTENDED)) {
2460 				xaddr = addressXAD(xad);
2461 				xlen = lengthXAD(xad);
2462 				dbUpdatePMap(ipbmap, false, xaddr,
2463 					     (s64) xlen, tblk);
2464 				xad->flag &= ~(XAD_NEW | XAD_EXTENDED);
2465 				jfs_info("allocPMap: xaddr:0x%lx xlen:%d",
2466 					 (ulong) xaddr, xlen);
2467 			}
2468 		}
2469 	} else if (maplock->flag & mlckALLOCPXD) {
2470 		pxdlock = (struct pxd_lock *) maplock;
2471 		xaddr = addressPXD(&pxdlock->pxd);
2472 		xlen = lengthPXD(&pxdlock->pxd);
2473 		dbUpdatePMap(ipbmap, false, xaddr, (s64) xlen, tblk);
2474 		jfs_info("allocPMap: xaddr:0x%lx xlen:%d", (ulong) xaddr, xlen);
2475 	} else {		/* (maplock->flag & mlckALLOCPXDLIST) */
2476 
2477 		pxdlistlock = (struct xdlistlock *) maplock;
2478 		pxd = pxdlistlock->xdlist;
2479 		for (n = 0; n < pxdlistlock->count; n++, pxd++) {
2480 			xaddr = addressPXD(pxd);
2481 			xlen = lengthPXD(pxd);
2482 			dbUpdatePMap(ipbmap, false, xaddr, (s64) xlen,
2483 				     tblk);
2484 			jfs_info("allocPMap: xaddr:0x%lx xlen:%d",
2485 				 (ulong) xaddr, xlen);
2486 		}
2487 	}
2488 }
2489 
2490 /*
2491  *	txFreeMap()
2492  *
2493  * function:	free from persistent and/or working map;
2494  *
2495  * todo: optimization
2496  */
txFreeMap(struct inode * ip,struct maplock * maplock,struct tblock * tblk,int maptype)2497 void txFreeMap(struct inode *ip,
2498 	       struct maplock * maplock, struct tblock * tblk, int maptype)
2499 {
2500 	struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
2501 	struct xdlistlock *xadlistlock;
2502 	xad_t *xad;
2503 	s64 xaddr;
2504 	int xlen;
2505 	struct pxd_lock *pxdlock;
2506 	struct xdlistlock *pxdlistlock;
2507 	pxd_t *pxd;
2508 	int n;
2509 
2510 	jfs_info("txFreeMap: tblk:0x%p maplock:0x%p maptype:0x%x",
2511 		 tblk, maplock, maptype);
2512 
2513 	/*
2514 	 * free from persistent map;
2515 	 */
2516 	if (maptype == COMMIT_PMAP || maptype == COMMIT_PWMAP) {
2517 		if (maplock->flag & mlckFREEXADLIST) {
2518 			xadlistlock = (struct xdlistlock *) maplock;
2519 			xad = xadlistlock->xdlist;
2520 			for (n = 0; n < xadlistlock->count; n++, xad++) {
2521 				if (!(xad->flag & XAD_NEW)) {
2522 					xaddr = addressXAD(xad);
2523 					xlen = lengthXAD(xad);
2524 					dbUpdatePMap(ipbmap, true, xaddr,
2525 						     (s64) xlen, tblk);
2526 					jfs_info("freePMap: xaddr:0x%lx "
2527 						 "xlen:%d",
2528 						 (ulong) xaddr, xlen);
2529 				}
2530 			}
2531 		} else if (maplock->flag & mlckFREEPXD) {
2532 			pxdlock = (struct pxd_lock *) maplock;
2533 			xaddr = addressPXD(&pxdlock->pxd);
2534 			xlen = lengthPXD(&pxdlock->pxd);
2535 			dbUpdatePMap(ipbmap, true, xaddr, (s64) xlen,
2536 				     tblk);
2537 			jfs_info("freePMap: xaddr:0x%lx xlen:%d",
2538 				 (ulong) xaddr, xlen);
2539 		} else {	/* (maplock->flag & mlckALLOCPXDLIST) */
2540 
2541 			pxdlistlock = (struct xdlistlock *) maplock;
2542 			pxd = pxdlistlock->xdlist;
2543 			for (n = 0; n < pxdlistlock->count; n++, pxd++) {
2544 				xaddr = addressPXD(pxd);
2545 				xlen = lengthPXD(pxd);
2546 				dbUpdatePMap(ipbmap, true, xaddr,
2547 					     (s64) xlen, tblk);
2548 				jfs_info("freePMap: xaddr:0x%lx xlen:%d",
2549 					 (ulong) xaddr, xlen);
2550 			}
2551 		}
2552 	}
2553 
2554 	/*
2555 	 * free from working map;
2556 	 */
2557 	if (maptype == COMMIT_PWMAP || maptype == COMMIT_WMAP) {
2558 		if (maplock->flag & mlckFREEXADLIST) {
2559 			xadlistlock = (struct xdlistlock *) maplock;
2560 			xad = xadlistlock->xdlist;
2561 			for (n = 0; n < xadlistlock->count; n++, xad++) {
2562 				xaddr = addressXAD(xad);
2563 				xlen = lengthXAD(xad);
2564 				dbFree(ip, xaddr, (s64) xlen);
2565 				xad->flag = 0;
2566 				jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
2567 					 (ulong) xaddr, xlen);
2568 			}
2569 		} else if (maplock->flag & mlckFREEPXD) {
2570 			pxdlock = (struct pxd_lock *) maplock;
2571 			xaddr = addressPXD(&pxdlock->pxd);
2572 			xlen = lengthPXD(&pxdlock->pxd);
2573 			dbFree(ip, xaddr, (s64) xlen);
2574 			jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
2575 				 (ulong) xaddr, xlen);
2576 		} else {	/* (maplock->flag & mlckFREEPXDLIST) */
2577 
2578 			pxdlistlock = (struct xdlistlock *) maplock;
2579 			pxd = pxdlistlock->xdlist;
2580 			for (n = 0; n < pxdlistlock->count; n++, pxd++) {
2581 				xaddr = addressPXD(pxd);
2582 				xlen = lengthPXD(pxd);
2583 				dbFree(ip, xaddr, (s64) xlen);
2584 				jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
2585 					 (ulong) xaddr, xlen);
2586 			}
2587 		}
2588 	}
2589 }
2590 
2591 /*
2592  *	txFreelock()
2593  *
2594  * function:	remove tlock from inode anonymous locklist
2595  */
txFreelock(struct inode * ip)2596 void txFreelock(struct inode *ip)
2597 {
2598 	struct jfs_inode_info *jfs_ip = JFS_IP(ip);
2599 	struct tlock *xtlck, *tlck;
2600 	lid_t xlid = 0, lid;
2601 
2602 	if (!jfs_ip->atlhead)
2603 		return;
2604 
2605 	TXN_LOCK();
2606 	xtlck = (struct tlock *) &jfs_ip->atlhead;
2607 
2608 	while ((lid = xtlck->next) != 0) {
2609 		tlck = lid_to_tlock(lid);
2610 		if (tlck->flag & tlckFREELOCK) {
2611 			xtlck->next = tlck->next;
2612 			txLockFree(lid);
2613 		} else {
2614 			xtlck = tlck;
2615 			xlid = lid;
2616 		}
2617 	}
2618 
2619 	if (jfs_ip->atlhead)
2620 		jfs_ip->atltail = xlid;
2621 	else {
2622 		jfs_ip->atltail = 0;
2623 		/*
2624 		 * If inode was on anon_list, remove it
2625 		 */
2626 		list_del_init(&jfs_ip->anon_inode_list);
2627 	}
2628 	TXN_UNLOCK();
2629 }
2630 
2631 /*
2632  *	txAbort()
2633  *
2634  * function: abort tx before commit;
2635  *
2636  * frees line-locks and segment locks for all
2637  * segments in comdata structure.
2638  * Optionally sets state of file-system to FM_DIRTY in super-block.
2639  * log age of page-frames in memory for which caller has
2640  * are reset to 0 (to avoid logwarap).
2641  */
txAbort(tid_t tid,int dirty)2642 void txAbort(tid_t tid, int dirty)
2643 {
2644 	lid_t lid, next;
2645 	struct metapage *mp;
2646 	struct tblock *tblk = tid_to_tblock(tid);
2647 	struct tlock *tlck;
2648 
2649 	/*
2650 	 * free tlocks of the transaction
2651 	 */
2652 	for (lid = tblk->next; lid; lid = next) {
2653 		tlck = lid_to_tlock(lid);
2654 		next = tlck->next;
2655 		mp = tlck->mp;
2656 		JFS_IP(tlck->ip)->xtlid = 0;
2657 
2658 		if (mp) {
2659 			mp->lid = 0;
2660 
2661 			/*
2662 			 * reset lsn of page to avoid logwarap:
2663 			 *
2664 			 * (page may have been previously committed by another
2665 			 * transaction(s) but has not been paged, i.e.,
2666 			 * it may be on logsync list even though it has not
2667 			 * been logged for the current tx.)
2668 			 */
2669 			if (mp->xflag & COMMIT_PAGE && mp->lsn)
2670 				LogSyncRelease(mp);
2671 		}
2672 		/* insert tlock at head of freelist */
2673 		TXN_LOCK();
2674 		txLockFree(lid);
2675 		TXN_UNLOCK();
2676 	}
2677 
2678 	/* caller will free the transaction block */
2679 
2680 	tblk->next = tblk->last = 0;
2681 
2682 	/*
2683 	 * mark filesystem dirty
2684 	 */
2685 	if (dirty)
2686 		jfs_error(tblk->sb, "\n");
2687 
2688 	return;
2689 }
2690 
2691 /*
2692  *	txLazyCommit(void)
2693  *
2694  *	All transactions except those changing ipimap (COMMIT_FORCE) are
2695  *	processed by this routine.  This insures that the inode and block
2696  *	allocation maps are updated in order.  For synchronous transactions,
2697  *	let the user thread finish processing after txUpdateMap() is called.
2698  */
txLazyCommit(struct tblock * tblk)2699 static void txLazyCommit(struct tblock * tblk)
2700 {
2701 	struct jfs_log *log;
2702 
2703 	while (((tblk->flag & tblkGC_READY) == 0) &&
2704 	       ((tblk->flag & tblkGC_UNLOCKED) == 0)) {
2705 		/* We must have gotten ahead of the user thread
2706 		 */
2707 		jfs_info("jfs_lazycommit: tblk 0x%p not unlocked", tblk);
2708 		yield();
2709 	}
2710 
2711 	jfs_info("txLazyCommit: processing tblk 0x%p", tblk);
2712 
2713 	txUpdateMap(tblk);
2714 
2715 	log = (struct jfs_log *) JFS_SBI(tblk->sb)->log;
2716 
2717 	spin_lock_irq(&log->gclock);	// LOGGC_LOCK
2718 
2719 	tblk->flag |= tblkGC_COMMITTED;
2720 
2721 	if (tblk->flag & tblkGC_READY)
2722 		log->gcrtc--;
2723 
2724 	wake_up_all(&tblk->gcwait);	// LOGGC_WAKEUP
2725 
2726 	/*
2727 	 * Can't release log->gclock until we've tested tblk->flag
2728 	 */
2729 	if (tblk->flag & tblkGC_LAZY) {
2730 		spin_unlock_irq(&log->gclock);	// LOGGC_UNLOCK
2731 		txUnlock(tblk);
2732 		tblk->flag &= ~tblkGC_LAZY;
2733 		txEnd(tblk - TxBlock);	/* Convert back to tid */
2734 	} else
2735 		spin_unlock_irq(&log->gclock);	// LOGGC_UNLOCK
2736 
2737 	jfs_info("txLazyCommit: done: tblk = 0x%p", tblk);
2738 }
2739 
2740 /*
2741  *	jfs_lazycommit(void)
2742  *
2743  *	To be run as a kernel daemon.  If lbmIODone is called in an interrupt
2744  *	context, or where blocking is not wanted, this routine will process
2745  *	committed transactions from the unlock queue.
2746  */
jfs_lazycommit(void * arg)2747 int jfs_lazycommit(void *arg)
2748 {
2749 	int WorkDone;
2750 	struct tblock *tblk;
2751 	unsigned long flags;
2752 	struct jfs_sb_info *sbi;
2753 
2754 	do {
2755 		LAZY_LOCK(flags);
2756 		jfs_commit_thread_waking = 0;	/* OK to wake another thread */
2757 		while (!list_empty(&TxAnchor.unlock_queue)) {
2758 			WorkDone = 0;
2759 			list_for_each_entry(tblk, &TxAnchor.unlock_queue,
2760 					    cqueue) {
2761 
2762 				sbi = JFS_SBI(tblk->sb);
2763 				/*
2764 				 * For each volume, the transactions must be
2765 				 * handled in order.  If another commit thread
2766 				 * is handling a tblk for this superblock,
2767 				 * skip it
2768 				 */
2769 				if (sbi->commit_state & IN_LAZYCOMMIT)
2770 					continue;
2771 
2772 				sbi->commit_state |= IN_LAZYCOMMIT;
2773 				WorkDone = 1;
2774 
2775 				/*
2776 				 * Remove transaction from queue
2777 				 */
2778 				list_del(&tblk->cqueue);
2779 
2780 				LAZY_UNLOCK(flags);
2781 				txLazyCommit(tblk);
2782 				LAZY_LOCK(flags);
2783 
2784 				sbi->commit_state &= ~IN_LAZYCOMMIT;
2785 				/*
2786 				 * Don't continue in the for loop.  (We can't
2787 				 * anyway, it's unsafe!)  We want to go back to
2788 				 * the beginning of the list.
2789 				 */
2790 				break;
2791 			}
2792 
2793 			/* If there was nothing to do, don't continue */
2794 			if (!WorkDone)
2795 				break;
2796 		}
2797 		/* In case a wakeup came while all threads were active */
2798 		jfs_commit_thread_waking = 0;
2799 
2800 		if (freezing(current)) {
2801 			LAZY_UNLOCK(flags);
2802 			try_to_freeze();
2803 		} else {
2804 			DECLARE_WAITQUEUE(wq, current);
2805 
2806 			add_wait_queue(&jfs_commit_thread_wait, &wq);
2807 			set_current_state(TASK_INTERRUPTIBLE);
2808 			LAZY_UNLOCK(flags);
2809 			schedule();
2810 			remove_wait_queue(&jfs_commit_thread_wait, &wq);
2811 		}
2812 	} while (!kthread_should_stop());
2813 
2814 	if (!list_empty(&TxAnchor.unlock_queue))
2815 		jfs_err("jfs_lazycommit being killed w/pending transactions!");
2816 	else
2817 		jfs_info("jfs_lazycommit being killed\n");
2818 	return 0;
2819 }
2820 
txLazyUnlock(struct tblock * tblk)2821 void txLazyUnlock(struct tblock * tblk)
2822 {
2823 	unsigned long flags;
2824 
2825 	LAZY_LOCK(flags);
2826 
2827 	list_add_tail(&tblk->cqueue, &TxAnchor.unlock_queue);
2828 	/*
2829 	 * Don't wake up a commit thread if there is already one servicing
2830 	 * this superblock, or if the last one we woke up hasn't started yet.
2831 	 */
2832 	if (!(JFS_SBI(tblk->sb)->commit_state & IN_LAZYCOMMIT) &&
2833 	    !jfs_commit_thread_waking) {
2834 		jfs_commit_thread_waking = 1;
2835 		wake_up(&jfs_commit_thread_wait);
2836 	}
2837 	LAZY_UNLOCK(flags);
2838 }
2839 
LogSyncRelease(struct metapage * mp)2840 static void LogSyncRelease(struct metapage * mp)
2841 {
2842 	struct jfs_log *log = mp->log;
2843 
2844 	assert(mp->nohomeok);
2845 	assert(log);
2846 	metapage_homeok(mp);
2847 }
2848 
2849 /*
2850  *	txQuiesce
2851  *
2852  *	Block all new transactions and push anonymous transactions to
2853  *	completion
2854  *
2855  *	This does almost the same thing as jfs_sync below.  We don't
2856  *	worry about deadlocking when jfs_tlocks_low is set, since we would
2857  *	expect jfs_sync to get us out of that jam.
2858  */
txQuiesce(struct super_block * sb)2859 void txQuiesce(struct super_block *sb)
2860 {
2861 	struct inode *ip;
2862 	struct jfs_inode_info *jfs_ip;
2863 	struct jfs_log *log = JFS_SBI(sb)->log;
2864 	tid_t tid;
2865 
2866 	set_bit(log_QUIESCE, &log->flag);
2867 
2868 	TXN_LOCK();
2869 restart:
2870 	while (!list_empty(&TxAnchor.anon_list)) {
2871 		jfs_ip = list_entry(TxAnchor.anon_list.next,
2872 				    struct jfs_inode_info,
2873 				    anon_inode_list);
2874 		ip = &jfs_ip->vfs_inode;
2875 
2876 		/*
2877 		 * inode will be removed from anonymous list
2878 		 * when it is committed
2879 		 */
2880 		TXN_UNLOCK();
2881 		tid = txBegin(ip->i_sb, COMMIT_INODE | COMMIT_FORCE);
2882 		mutex_lock(&jfs_ip->commit_mutex);
2883 		txCommit(tid, 1, &ip, 0);
2884 		txEnd(tid);
2885 		mutex_unlock(&jfs_ip->commit_mutex);
2886 		/*
2887 		 * Just to be safe.  I don't know how
2888 		 * long we can run without blocking
2889 		 */
2890 		cond_resched();
2891 		TXN_LOCK();
2892 	}
2893 
2894 	/*
2895 	 * If jfs_sync is running in parallel, there could be some inodes
2896 	 * on anon_list2.  Let's check.
2897 	 */
2898 	if (!list_empty(&TxAnchor.anon_list2)) {
2899 		list_splice(&TxAnchor.anon_list2, &TxAnchor.anon_list);
2900 		INIT_LIST_HEAD(&TxAnchor.anon_list2);
2901 		goto restart;
2902 	}
2903 	TXN_UNLOCK();
2904 
2905 	/*
2906 	 * We may need to kick off the group commit
2907 	 */
2908 	jfs_flush_journal(log, 0);
2909 }
2910 
2911 /*
2912  * txResume()
2913  *
2914  * Allows transactions to start again following txQuiesce
2915  */
txResume(struct super_block * sb)2916 void txResume(struct super_block *sb)
2917 {
2918 	struct jfs_log *log = JFS_SBI(sb)->log;
2919 
2920 	clear_bit(log_QUIESCE, &log->flag);
2921 	TXN_WAKEUP(&log->syncwait);
2922 }
2923 
2924 /*
2925  *	jfs_sync(void)
2926  *
2927  *	To be run as a kernel daemon.  This is awakened when tlocks run low.
2928  *	We write any inodes that have anonymous tlocks so they will become
2929  *	available.
2930  */
jfs_sync(void * arg)2931 int jfs_sync(void *arg)
2932 {
2933 	struct inode *ip;
2934 	struct jfs_inode_info *jfs_ip;
2935 	tid_t tid;
2936 
2937 	do {
2938 		/*
2939 		 * write each inode on the anonymous inode list
2940 		 */
2941 		TXN_LOCK();
2942 		while (jfs_tlocks_low && !list_empty(&TxAnchor.anon_list)) {
2943 			jfs_ip = list_entry(TxAnchor.anon_list.next,
2944 					    struct jfs_inode_info,
2945 					    anon_inode_list);
2946 			ip = &jfs_ip->vfs_inode;
2947 
2948 			if (! igrab(ip)) {
2949 				/*
2950 				 * Inode is being freed
2951 				 */
2952 				list_del_init(&jfs_ip->anon_inode_list);
2953 			} else if (mutex_trylock(&jfs_ip->commit_mutex)) {
2954 				/*
2955 				 * inode will be removed from anonymous list
2956 				 * when it is committed
2957 				 */
2958 				TXN_UNLOCK();
2959 				tid = txBegin(ip->i_sb, COMMIT_INODE);
2960 				txCommit(tid, 1, &ip, 0);
2961 				txEnd(tid);
2962 				mutex_unlock(&jfs_ip->commit_mutex);
2963 
2964 				iput(ip);
2965 				/*
2966 				 * Just to be safe.  I don't know how
2967 				 * long we can run without blocking
2968 				 */
2969 				cond_resched();
2970 				TXN_LOCK();
2971 			} else {
2972 				/* We can't get the commit mutex.  It may
2973 				 * be held by a thread waiting for tlock's
2974 				 * so let's not block here.  Save it to
2975 				 * put back on the anon_list.
2976 				 */
2977 
2978 				/* Move from anon_list to anon_list2 */
2979 				list_move(&jfs_ip->anon_inode_list,
2980 					  &TxAnchor.anon_list2);
2981 
2982 				TXN_UNLOCK();
2983 				iput(ip);
2984 				TXN_LOCK();
2985 			}
2986 		}
2987 		/* Add anon_list2 back to anon_list */
2988 		list_splice_init(&TxAnchor.anon_list2, &TxAnchor.anon_list);
2989 
2990 		if (freezing(current)) {
2991 			TXN_UNLOCK();
2992 			try_to_freeze();
2993 		} else {
2994 			set_current_state(TASK_INTERRUPTIBLE);
2995 			TXN_UNLOCK();
2996 			schedule();
2997 		}
2998 	} while (!kthread_should_stop());
2999 
3000 	jfs_info("jfs_sync being killed");
3001 	return 0;
3002 }
3003 
3004 #if defined(CONFIG_PROC_FS) && defined(CONFIG_JFS_DEBUG)
jfs_txanchor_proc_show(struct seq_file * m,void * v)3005 static int jfs_txanchor_proc_show(struct seq_file *m, void *v)
3006 {
3007 	char *freewait;
3008 	char *freelockwait;
3009 	char *lowlockwait;
3010 
3011 	freewait =
3012 	    waitqueue_active(&TxAnchor.freewait) ? "active" : "empty";
3013 	freelockwait =
3014 	    waitqueue_active(&TxAnchor.freelockwait) ? "active" : "empty";
3015 	lowlockwait =
3016 	    waitqueue_active(&TxAnchor.lowlockwait) ? "active" : "empty";
3017 
3018 	seq_printf(m,
3019 		       "JFS TxAnchor\n"
3020 		       "============\n"
3021 		       "freetid = %d\n"
3022 		       "freewait = %s\n"
3023 		       "freelock = %d\n"
3024 		       "freelockwait = %s\n"
3025 		       "lowlockwait = %s\n"
3026 		       "tlocksInUse = %d\n"
3027 		       "jfs_tlocks_low = %d\n"
3028 		       "unlock_queue is %sempty\n",
3029 		       TxAnchor.freetid,
3030 		       freewait,
3031 		       TxAnchor.freelock,
3032 		       freelockwait,
3033 		       lowlockwait,
3034 		       TxAnchor.tlocksInUse,
3035 		       jfs_tlocks_low,
3036 		       list_empty(&TxAnchor.unlock_queue) ? "" : "not ");
3037 	return 0;
3038 }
3039 
jfs_txanchor_proc_open(struct inode * inode,struct file * file)3040 static int jfs_txanchor_proc_open(struct inode *inode, struct file *file)
3041 {
3042 	return single_open(file, jfs_txanchor_proc_show, NULL);
3043 }
3044 
3045 const struct file_operations jfs_txanchor_proc_fops = {
3046 	.owner		= THIS_MODULE,
3047 	.open		= jfs_txanchor_proc_open,
3048 	.read		= seq_read,
3049 	.llseek		= seq_lseek,
3050 	.release	= single_release,
3051 };
3052 #endif
3053 
3054 #if defined(CONFIG_PROC_FS) && defined(CONFIG_JFS_STATISTICS)
jfs_txstats_proc_show(struct seq_file * m,void * v)3055 static int jfs_txstats_proc_show(struct seq_file *m, void *v)
3056 {
3057 	seq_printf(m,
3058 		       "JFS TxStats\n"
3059 		       "===========\n"
3060 		       "calls to txBegin = %d\n"
3061 		       "txBegin blocked by sync barrier = %d\n"
3062 		       "txBegin blocked by tlocks low = %d\n"
3063 		       "txBegin blocked by no free tid = %d\n"
3064 		       "calls to txBeginAnon = %d\n"
3065 		       "txBeginAnon blocked by sync barrier = %d\n"
3066 		       "txBeginAnon blocked by tlocks low = %d\n"
3067 		       "calls to txLockAlloc = %d\n"
3068 		       "tLockAlloc blocked by no free lock = %d\n",
3069 		       TxStat.txBegin,
3070 		       TxStat.txBegin_barrier,
3071 		       TxStat.txBegin_lockslow,
3072 		       TxStat.txBegin_freetid,
3073 		       TxStat.txBeginAnon,
3074 		       TxStat.txBeginAnon_barrier,
3075 		       TxStat.txBeginAnon_lockslow,
3076 		       TxStat.txLockAlloc,
3077 		       TxStat.txLockAlloc_freelock);
3078 	return 0;
3079 }
3080 
jfs_txstats_proc_open(struct inode * inode,struct file * file)3081 static int jfs_txstats_proc_open(struct inode *inode, struct file *file)
3082 {
3083 	return single_open(file, jfs_txstats_proc_show, NULL);
3084 }
3085 
3086 const struct file_operations jfs_txstats_proc_fops = {
3087 	.owner		= THIS_MODULE,
3088 	.open		= jfs_txstats_proc_open,
3089 	.read		= seq_read,
3090 	.llseek		= seq_lseek,
3091 	.release	= single_release,
3092 };
3093 #endif
3094