• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *   Copyright (C) International Business Machines Corp., 2000-2005
4  *   Portions Copyright (C) Christoph Hellwig, 2001-2002
5  */
6 
7 /*
8  *	jfs_txnmgr.c: transaction manager
9  *
10  * notes:
11  * transaction starts with txBegin() and ends with txCommit()
12  * or txAbort().
13  *
14  * tlock is acquired at the time of update;
15  * (obviate scan at commit time for xtree and dtree)
16  * tlock and mp points to each other;
17  * (no hashlist for mp -> tlock).
18  *
19  * special cases:
20  * tlock on in-memory inode:
21  * in-place tlock in the in-memory inode itself;
22  * converted to page lock by iWrite() at commit time.
23  *
24  * tlock during write()/mmap() under anonymous transaction (tid = 0):
25  * transferred (?) to transaction at commit time.
26  *
27  * use the page itself to update allocation maps
28  * (obviate intermediate replication of allocation/deallocation data)
29  * hold on to mp+lock thru update of maps
30  */
31 
32 #include <linux/fs.h>
33 #include <linux/vmalloc.h>
34 #include <linux/completion.h>
35 #include <linux/freezer.h>
36 #include <linux/module.h>
37 #include <linux/moduleparam.h>
38 #include <linux/kthread.h>
39 #include <linux/seq_file.h>
40 #include "jfs_incore.h"
41 #include "jfs_inode.h"
42 #include "jfs_filsys.h"
43 #include "jfs_metapage.h"
44 #include "jfs_dinode.h"
45 #include "jfs_imap.h"
46 #include "jfs_dmap.h"
47 #include "jfs_superblock.h"
48 #include "jfs_debug.h"
49 
50 /*
51  *	transaction management structures
52  */
53 static struct {
54 	int freetid;		/* index of a free tid structure */
55 	int freelock;		/* index first free lock word */
56 	wait_queue_head_t freewait;	/* eventlist of free tblock */
57 	wait_queue_head_t freelockwait;	/* eventlist of free tlock */
58 	wait_queue_head_t lowlockwait;	/* eventlist of ample tlocks */
59 	int tlocksInUse;	/* Number of tlocks in use */
60 	spinlock_t LazyLock;	/* synchronize sync_queue & unlock_queue */
61 /*	struct tblock *sync_queue; * Transactions waiting for data sync */
62 	struct list_head unlock_queue;	/* Txns waiting to be released */
63 	struct list_head anon_list;	/* inodes having anonymous txns */
64 	struct list_head anon_list2;	/* inodes having anonymous txns
65 					   that couldn't be sync'ed */
66 } TxAnchor;
67 
68 int jfs_tlocks_low;		/* Indicates low number of available tlocks */
69 
70 #ifdef CONFIG_JFS_STATISTICS
71 static struct {
72 	uint txBegin;
73 	uint txBegin_barrier;
74 	uint txBegin_lockslow;
75 	uint txBegin_freetid;
76 	uint txBeginAnon;
77 	uint txBeginAnon_barrier;
78 	uint txBeginAnon_lockslow;
79 	uint txLockAlloc;
80 	uint txLockAlloc_freelock;
81 } TxStat;
82 #endif
83 
84 static int nTxBlock = -1;	/* number of transaction blocks */
85 module_param(nTxBlock, int, 0);
86 MODULE_PARM_DESC(nTxBlock,
87 		 "Number of transaction blocks (max:65536)");
88 
89 static int nTxLock = -1;	/* number of transaction locks */
90 module_param(nTxLock, int, 0);
91 MODULE_PARM_DESC(nTxLock,
92 		 "Number of transaction locks (max:65536)");
93 
94 struct tblock *TxBlock;	/* transaction block table */
95 static int TxLockLWM;	/* Low water mark for number of txLocks used */
96 static int TxLockHWM;	/* High water mark for number of txLocks used */
97 static int TxLockVHWM;	/* Very High water mark */
98 struct tlock *TxLock;	/* transaction lock table */
99 
100 /*
101  *	transaction management lock
102  */
103 static DEFINE_SPINLOCK(jfsTxnLock);
104 
105 #define TXN_LOCK()		spin_lock(&jfsTxnLock)
106 #define TXN_UNLOCK()		spin_unlock(&jfsTxnLock)
107 
108 #define LAZY_LOCK_INIT()	spin_lock_init(&TxAnchor.LazyLock)
109 #define LAZY_LOCK(flags)	spin_lock_irqsave(&TxAnchor.LazyLock, flags)
110 #define LAZY_UNLOCK(flags) spin_unlock_irqrestore(&TxAnchor.LazyLock, flags)
111 
112 static DECLARE_WAIT_QUEUE_HEAD(jfs_commit_thread_wait);
113 static int jfs_commit_thread_waking;
114 
115 /*
116  * Retry logic exist outside these macros to protect from spurrious wakeups.
117  */
TXN_SLEEP_DROP_LOCK(wait_queue_head_t * event)118 static inline void TXN_SLEEP_DROP_LOCK(wait_queue_head_t * event)
119 {
120 	DECLARE_WAITQUEUE(wait, current);
121 
122 	add_wait_queue(event, &wait);
123 	set_current_state(TASK_UNINTERRUPTIBLE);
124 	TXN_UNLOCK();
125 	io_schedule();
126 	remove_wait_queue(event, &wait);
127 }
128 
129 #define TXN_SLEEP(event)\
130 {\
131 	TXN_SLEEP_DROP_LOCK(event);\
132 	TXN_LOCK();\
133 }
134 
135 #define TXN_WAKEUP(event) wake_up_all(event)
136 
137 /*
138  *	statistics
139  */
140 static struct {
141 	tid_t maxtid;		/* 4: biggest tid ever used */
142 	lid_t maxlid;		/* 4: biggest lid ever used */
143 	int ntid;		/* 4: # of transactions performed */
144 	int nlid;		/* 4: # of tlocks acquired */
145 	int waitlock;		/* 4: # of tlock wait */
146 } stattx;
147 
148 /*
149  * forward references
150  */
151 static void diLog(struct jfs_log *log, struct tblock *tblk, struct lrd *lrd,
152 		struct tlock *tlck, struct commit *cd);
153 static void dataLog(struct jfs_log *log, struct tblock *tblk, struct lrd *lrd,
154 		struct tlock *tlck);
155 static void dtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
156 		struct tlock * tlck);
157 static void mapLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
158 		struct tlock * tlck);
159 static void txAllocPMap(struct inode *ip, struct maplock * maplock,
160 		struct tblock * tblk);
161 static void txForce(struct tblock * tblk);
162 static void txLog(struct jfs_log *log, struct tblock *tblk,
163 		struct commit *cd);
164 static void txUpdateMap(struct tblock * tblk);
165 static void txRelease(struct tblock * tblk);
166 static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
167 	   struct tlock * tlck);
168 static void LogSyncRelease(struct metapage * mp);
169 
170 /*
171  *		transaction block/lock management
172  *		---------------------------------
173  */
174 
175 /*
176  * Get a transaction lock from the free list.  If the number in use is
177  * greater than the high water mark, wake up the sync daemon.  This should
178  * free some anonymous transaction locks.  (TXN_LOCK must be held.)
179  */
txLockAlloc(void)180 static lid_t txLockAlloc(void)
181 {
182 	lid_t lid;
183 
184 	INCREMENT(TxStat.txLockAlloc);
185 	if (!TxAnchor.freelock) {
186 		INCREMENT(TxStat.txLockAlloc_freelock);
187 	}
188 
189 	while (!(lid = TxAnchor.freelock))
190 		TXN_SLEEP(&TxAnchor.freelockwait);
191 	TxAnchor.freelock = TxLock[lid].next;
192 	HIGHWATERMARK(stattx.maxlid, lid);
193 	if ((++TxAnchor.tlocksInUse > TxLockHWM) && (jfs_tlocks_low == 0)) {
194 		jfs_info("txLockAlloc tlocks low");
195 		jfs_tlocks_low = 1;
196 		wake_up_process(jfsSyncThread);
197 	}
198 
199 	return lid;
200 }
201 
txLockFree(lid_t lid)202 static void txLockFree(lid_t lid)
203 {
204 	TxLock[lid].tid = 0;
205 	TxLock[lid].next = TxAnchor.freelock;
206 	TxAnchor.freelock = lid;
207 	TxAnchor.tlocksInUse--;
208 	if (jfs_tlocks_low && (TxAnchor.tlocksInUse < TxLockLWM)) {
209 		jfs_info("txLockFree jfs_tlocks_low no more");
210 		jfs_tlocks_low = 0;
211 		TXN_WAKEUP(&TxAnchor.lowlockwait);
212 	}
213 	TXN_WAKEUP(&TxAnchor.freelockwait);
214 }
215 
216 /*
217  * NAME:	txInit()
218  *
219  * FUNCTION:	initialize transaction management structures
220  *
221  * RETURN:
222  *
223  * serialization: single thread at jfs_init()
224  */
txInit(void)225 int txInit(void)
226 {
227 	int k, size;
228 	struct sysinfo si;
229 
230 	/* Set defaults for nTxLock and nTxBlock if unset */
231 
232 	if (nTxLock == -1) {
233 		if (nTxBlock == -1) {
234 			/* Base default on memory size */
235 			si_meminfo(&si);
236 			if (si.totalram > (256 * 1024)) /* 1 GB */
237 				nTxLock = 64 * 1024;
238 			else
239 				nTxLock = si.totalram >> 2;
240 		} else if (nTxBlock > (8 * 1024))
241 			nTxLock = 64 * 1024;
242 		else
243 			nTxLock = nTxBlock << 3;
244 	}
245 	if (nTxBlock == -1)
246 		nTxBlock = nTxLock >> 3;
247 
248 	/* Verify tunable parameters */
249 	if (nTxBlock < 16)
250 		nTxBlock = 16;	/* No one should set it this low */
251 	if (nTxBlock > 65536)
252 		nTxBlock = 65536;
253 	if (nTxLock < 256)
254 		nTxLock = 256;	/* No one should set it this low */
255 	if (nTxLock > 65536)
256 		nTxLock = 65536;
257 
258 	printk(KERN_INFO "JFS: nTxBlock = %d, nTxLock = %d\n",
259 	       nTxBlock, nTxLock);
260 	/*
261 	 * initialize transaction block (tblock) table
262 	 *
263 	 * transaction id (tid) = tblock index
264 	 * tid = 0 is reserved.
265 	 */
266 	TxLockLWM = (nTxLock * 4) / 10;
267 	TxLockHWM = (nTxLock * 7) / 10;
268 	TxLockVHWM = (nTxLock * 8) / 10;
269 
270 	size = sizeof(struct tblock) * nTxBlock;
271 	TxBlock = vmalloc(size);
272 	if (TxBlock == NULL)
273 		return -ENOMEM;
274 
275 	for (k = 1; k < nTxBlock - 1; k++) {
276 		TxBlock[k].next = k + 1;
277 		init_waitqueue_head(&TxBlock[k].gcwait);
278 		init_waitqueue_head(&TxBlock[k].waitor);
279 	}
280 	TxBlock[k].next = 0;
281 	init_waitqueue_head(&TxBlock[k].gcwait);
282 	init_waitqueue_head(&TxBlock[k].waitor);
283 
284 	TxAnchor.freetid = 1;
285 	init_waitqueue_head(&TxAnchor.freewait);
286 
287 	stattx.maxtid = 1;	/* statistics */
288 
289 	/*
290 	 * initialize transaction lock (tlock) table
291 	 *
292 	 * transaction lock id = tlock index
293 	 * tlock id = 0 is reserved.
294 	 */
295 	size = sizeof(struct tlock) * nTxLock;
296 	TxLock = vmalloc(size);
297 	if (TxLock == NULL) {
298 		vfree(TxBlock);
299 		return -ENOMEM;
300 	}
301 
302 	/* initialize tlock table */
303 	for (k = 1; k < nTxLock - 1; k++)
304 		TxLock[k].next = k + 1;
305 	TxLock[k].next = 0;
306 	init_waitqueue_head(&TxAnchor.freelockwait);
307 	init_waitqueue_head(&TxAnchor.lowlockwait);
308 
309 	TxAnchor.freelock = 1;
310 	TxAnchor.tlocksInUse = 0;
311 	INIT_LIST_HEAD(&TxAnchor.anon_list);
312 	INIT_LIST_HEAD(&TxAnchor.anon_list2);
313 
314 	LAZY_LOCK_INIT();
315 	INIT_LIST_HEAD(&TxAnchor.unlock_queue);
316 
317 	stattx.maxlid = 1;	/* statistics */
318 
319 	return 0;
320 }
321 
322 /*
323  * NAME:	txExit()
324  *
325  * FUNCTION:	clean up when module is unloaded
326  */
txExit(void)327 void txExit(void)
328 {
329 	vfree(TxLock);
330 	TxLock = NULL;
331 	vfree(TxBlock);
332 	TxBlock = NULL;
333 }
334 
335 /*
336  * NAME:	txBegin()
337  *
338  * FUNCTION:	start a transaction.
339  *
340  * PARAMETER:	sb	- superblock
341  *		flag	- force for nested tx;
342  *
343  * RETURN:	tid	- transaction id
344  *
345  * note: flag force allows to start tx for nested tx
346  * to prevent deadlock on logsync barrier;
347  */
txBegin(struct super_block * sb,int flag)348 tid_t txBegin(struct super_block *sb, int flag)
349 {
350 	tid_t t;
351 	struct tblock *tblk;
352 	struct jfs_log *log;
353 
354 	jfs_info("txBegin: flag = 0x%x", flag);
355 	log = JFS_SBI(sb)->log;
356 
357 	if (!log) {
358 		jfs_error(sb, "read-only filesystem\n");
359 		return 0;
360 	}
361 
362 	TXN_LOCK();
363 
364 	INCREMENT(TxStat.txBegin);
365 
366       retry:
367 	if (!(flag & COMMIT_FORCE)) {
368 		/*
369 		 * synchronize with logsync barrier
370 		 */
371 		if (test_bit(log_SYNCBARRIER, &log->flag) ||
372 		    test_bit(log_QUIESCE, &log->flag)) {
373 			INCREMENT(TxStat.txBegin_barrier);
374 			TXN_SLEEP(&log->syncwait);
375 			goto retry;
376 		}
377 	}
378 	if (flag == 0) {
379 		/*
380 		 * Don't begin transaction if we're getting starved for tlocks
381 		 * unless COMMIT_FORCE or COMMIT_INODE (which may ultimately
382 		 * free tlocks)
383 		 */
384 		if (TxAnchor.tlocksInUse > TxLockVHWM) {
385 			INCREMENT(TxStat.txBegin_lockslow);
386 			TXN_SLEEP(&TxAnchor.lowlockwait);
387 			goto retry;
388 		}
389 	}
390 
391 	/*
392 	 * allocate transaction id/block
393 	 */
394 	if ((t = TxAnchor.freetid) == 0) {
395 		jfs_info("txBegin: waiting for free tid");
396 		INCREMENT(TxStat.txBegin_freetid);
397 		TXN_SLEEP(&TxAnchor.freewait);
398 		goto retry;
399 	}
400 
401 	tblk = tid_to_tblock(t);
402 
403 	if ((tblk->next == 0) && !(flag & COMMIT_FORCE)) {
404 		/* Don't let a non-forced transaction take the last tblk */
405 		jfs_info("txBegin: waiting for free tid");
406 		INCREMENT(TxStat.txBegin_freetid);
407 		TXN_SLEEP(&TxAnchor.freewait);
408 		goto retry;
409 	}
410 
411 	TxAnchor.freetid = tblk->next;
412 
413 	/*
414 	 * initialize transaction
415 	 */
416 
417 	/*
418 	 * We can't zero the whole thing or we screw up another thread being
419 	 * awakened after sleeping on tblk->waitor
420 	 *
421 	 * memset(tblk, 0, sizeof(struct tblock));
422 	 */
423 	tblk->next = tblk->last = tblk->xflag = tblk->flag = tblk->lsn = 0;
424 
425 	tblk->sb = sb;
426 	++log->logtid;
427 	tblk->logtid = log->logtid;
428 
429 	++log->active;
430 
431 	HIGHWATERMARK(stattx.maxtid, t);	/* statistics */
432 	INCREMENT(stattx.ntid);	/* statistics */
433 
434 	TXN_UNLOCK();
435 
436 	jfs_info("txBegin: returning tid = %d", t);
437 
438 	return t;
439 }
440 
441 /*
442  * NAME:	txBeginAnon()
443  *
444  * FUNCTION:	start an anonymous transaction.
445  *		Blocks if logsync or available tlocks are low to prevent
446  *		anonymous tlocks from depleting supply.
447  *
448  * PARAMETER:	sb	- superblock
449  *
450  * RETURN:	none
451  */
txBeginAnon(struct super_block * sb)452 void txBeginAnon(struct super_block *sb)
453 {
454 	struct jfs_log *log;
455 
456 	log = JFS_SBI(sb)->log;
457 
458 	TXN_LOCK();
459 	INCREMENT(TxStat.txBeginAnon);
460 
461       retry:
462 	/*
463 	 * synchronize with logsync barrier
464 	 */
465 	if (test_bit(log_SYNCBARRIER, &log->flag) ||
466 	    test_bit(log_QUIESCE, &log->flag)) {
467 		INCREMENT(TxStat.txBeginAnon_barrier);
468 		TXN_SLEEP(&log->syncwait);
469 		goto retry;
470 	}
471 
472 	/*
473 	 * Don't begin transaction if we're getting starved for tlocks
474 	 */
475 	if (TxAnchor.tlocksInUse > TxLockVHWM) {
476 		INCREMENT(TxStat.txBeginAnon_lockslow);
477 		TXN_SLEEP(&TxAnchor.lowlockwait);
478 		goto retry;
479 	}
480 	TXN_UNLOCK();
481 }
482 
483 /*
484  *	txEnd()
485  *
486  * function: free specified transaction block.
487  *
488  *	logsync barrier processing:
489  *
490  * serialization:
491  */
txEnd(tid_t tid)492 void txEnd(tid_t tid)
493 {
494 	struct tblock *tblk = tid_to_tblock(tid);
495 	struct jfs_log *log;
496 
497 	jfs_info("txEnd: tid = %d", tid);
498 	TXN_LOCK();
499 
500 	/*
501 	 * wakeup transactions waiting on the page locked
502 	 * by the current transaction
503 	 */
504 	TXN_WAKEUP(&tblk->waitor);
505 
506 	log = JFS_SBI(tblk->sb)->log;
507 
508 	/*
509 	 * Lazy commit thread can't free this guy until we mark it UNLOCKED,
510 	 * otherwise, we would be left with a transaction that may have been
511 	 * reused.
512 	 *
513 	 * Lazy commit thread will turn off tblkGC_LAZY before calling this
514 	 * routine.
515 	 */
516 	if (tblk->flag & tblkGC_LAZY) {
517 		jfs_info("txEnd called w/lazy tid: %d, tblk = 0x%p", tid, tblk);
518 		TXN_UNLOCK();
519 
520 		spin_lock_irq(&log->gclock);	// LOGGC_LOCK
521 		tblk->flag |= tblkGC_UNLOCKED;
522 		spin_unlock_irq(&log->gclock);	// LOGGC_UNLOCK
523 		return;
524 	}
525 
526 	jfs_info("txEnd: tid: %d, tblk = 0x%p", tid, tblk);
527 
528 	assert(tblk->next == 0);
529 
530 	/*
531 	 * insert tblock back on freelist
532 	 */
533 	tblk->next = TxAnchor.freetid;
534 	TxAnchor.freetid = tid;
535 
536 	/*
537 	 * mark the tblock not active
538 	 */
539 	if (--log->active == 0) {
540 		clear_bit(log_FLUSH, &log->flag);
541 
542 		/*
543 		 * synchronize with logsync barrier
544 		 */
545 		if (test_bit(log_SYNCBARRIER, &log->flag)) {
546 			TXN_UNLOCK();
547 
548 			/* write dirty metadata & forward log syncpt */
549 			jfs_syncpt(log, 1);
550 
551 			jfs_info("log barrier off: 0x%x", log->lsn);
552 
553 			/* enable new transactions start */
554 			clear_bit(log_SYNCBARRIER, &log->flag);
555 
556 			/* wakeup all waitors for logsync barrier */
557 			TXN_WAKEUP(&log->syncwait);
558 
559 			goto wakeup;
560 		}
561 	}
562 
563 	TXN_UNLOCK();
564 wakeup:
565 	/*
566 	 * wakeup all waitors for a free tblock
567 	 */
568 	TXN_WAKEUP(&TxAnchor.freewait);
569 }
570 
571 /*
572  *	txLock()
573  *
574  * function: acquire a transaction lock on the specified <mp>
575  *
576  * parameter:
577  *
578  * return:	transaction lock id
579  *
580  * serialization:
581  */
txLock(tid_t tid,struct inode * ip,struct metapage * mp,int type)582 struct tlock *txLock(tid_t tid, struct inode *ip, struct metapage * mp,
583 		     int type)
584 {
585 	struct jfs_inode_info *jfs_ip = JFS_IP(ip);
586 	int dir_xtree = 0;
587 	lid_t lid;
588 	tid_t xtid;
589 	struct tlock *tlck;
590 	struct xtlock *xtlck;
591 	struct linelock *linelock;
592 	xtpage_t *p;
593 	struct tblock *tblk;
594 
595 	TXN_LOCK();
596 
597 	if (S_ISDIR(ip->i_mode) && (type & tlckXTREE) &&
598 	    !(mp->xflag & COMMIT_PAGE)) {
599 		/*
600 		 * Directory inode is special.  It can have both an xtree tlock
601 		 * and a dtree tlock associated with it.
602 		 */
603 		dir_xtree = 1;
604 		lid = jfs_ip->xtlid;
605 	} else
606 		lid = mp->lid;
607 
608 	/* is page not locked by a transaction ? */
609 	if (lid == 0)
610 		goto allocateLock;
611 
612 	jfs_info("txLock: tid:%d ip:0x%p mp:0x%p lid:%d", tid, ip, mp, lid);
613 
614 	/* is page locked by the requester transaction ? */
615 	tlck = lid_to_tlock(lid);
616 	if ((xtid = tlck->tid) == tid) {
617 		TXN_UNLOCK();
618 		goto grantLock;
619 	}
620 
621 	/*
622 	 * is page locked by anonymous transaction/lock ?
623 	 *
624 	 * (page update without transaction (i.e., file write) is
625 	 * locked under anonymous transaction tid = 0:
626 	 * anonymous tlocks maintained on anonymous tlock list of
627 	 * the inode of the page and available to all anonymous
628 	 * transactions until txCommit() time at which point
629 	 * they are transferred to the transaction tlock list of
630 	 * the committing transaction of the inode)
631 	 */
632 	if (xtid == 0) {
633 		tlck->tid = tid;
634 		TXN_UNLOCK();
635 		tblk = tid_to_tblock(tid);
636 		/*
637 		 * The order of the tlocks in the transaction is important
638 		 * (during truncate, child xtree pages must be freed before
639 		 * parent's tlocks change the working map).
640 		 * Take tlock off anonymous list and add to tail of
641 		 * transaction list
642 		 *
643 		 * Note:  We really need to get rid of the tid & lid and
644 		 * use list_head's.  This code is getting UGLY!
645 		 */
646 		if (jfs_ip->atlhead == lid) {
647 			if (jfs_ip->atltail == lid) {
648 				/* only anonymous txn.
649 				 * Remove from anon_list
650 				 */
651 				TXN_LOCK();
652 				list_del_init(&jfs_ip->anon_inode_list);
653 				TXN_UNLOCK();
654 			}
655 			jfs_ip->atlhead = tlck->next;
656 		} else {
657 			lid_t last;
658 			for (last = jfs_ip->atlhead;
659 			     lid_to_tlock(last)->next != lid;
660 			     last = lid_to_tlock(last)->next) {
661 				assert(last);
662 			}
663 			lid_to_tlock(last)->next = tlck->next;
664 			if (jfs_ip->atltail == lid)
665 				jfs_ip->atltail = last;
666 		}
667 
668 		/* insert the tlock at tail of transaction tlock list */
669 
670 		if (tblk->next)
671 			lid_to_tlock(tblk->last)->next = lid;
672 		else
673 			tblk->next = lid;
674 		tlck->next = 0;
675 		tblk->last = lid;
676 
677 		goto grantLock;
678 	}
679 
680 	goto waitLock;
681 
682 	/*
683 	 * allocate a tlock
684 	 */
685       allocateLock:
686 	lid = txLockAlloc();
687 	tlck = lid_to_tlock(lid);
688 
689 	/*
690 	 * initialize tlock
691 	 */
692 	tlck->tid = tid;
693 
694 	TXN_UNLOCK();
695 
696 	/* mark tlock for meta-data page */
697 	if (mp->xflag & COMMIT_PAGE) {
698 
699 		tlck->flag = tlckPAGELOCK;
700 
701 		/* mark the page dirty and nohomeok */
702 		metapage_nohomeok(mp);
703 
704 		jfs_info("locking mp = 0x%p, nohomeok = %d tid = %d tlck = 0x%p",
705 			 mp, mp->nohomeok, tid, tlck);
706 
707 		/* if anonymous transaction, and buffer is on the group
708 		 * commit synclist, mark inode to show this.  This will
709 		 * prevent the buffer from being marked nohomeok for too
710 		 * long a time.
711 		 */
712 		if ((tid == 0) && mp->lsn)
713 			set_cflag(COMMIT_Synclist, ip);
714 	}
715 	/* mark tlock for in-memory inode */
716 	else
717 		tlck->flag = tlckINODELOCK;
718 
719 	if (S_ISDIR(ip->i_mode))
720 		tlck->flag |= tlckDIRECTORY;
721 
722 	tlck->type = 0;
723 
724 	/* bind the tlock and the page */
725 	tlck->ip = ip;
726 	tlck->mp = mp;
727 	if (dir_xtree)
728 		jfs_ip->xtlid = lid;
729 	else
730 		mp->lid = lid;
731 
732 	/*
733 	 * enqueue transaction lock to transaction/inode
734 	 */
735 	/* insert the tlock at tail of transaction tlock list */
736 	if (tid) {
737 		tblk = tid_to_tblock(tid);
738 		if (tblk->next)
739 			lid_to_tlock(tblk->last)->next = lid;
740 		else
741 			tblk->next = lid;
742 		tlck->next = 0;
743 		tblk->last = lid;
744 	}
745 	/* anonymous transaction:
746 	 * insert the tlock at head of inode anonymous tlock list
747 	 */
748 	else {
749 		tlck->next = jfs_ip->atlhead;
750 		jfs_ip->atlhead = lid;
751 		if (tlck->next == 0) {
752 			/* This inode's first anonymous transaction */
753 			jfs_ip->atltail = lid;
754 			TXN_LOCK();
755 			list_add_tail(&jfs_ip->anon_inode_list,
756 				      &TxAnchor.anon_list);
757 			TXN_UNLOCK();
758 		}
759 	}
760 
761 	/* initialize type dependent area for linelock */
762 	linelock = (struct linelock *) & tlck->lock;
763 	linelock->next = 0;
764 	linelock->flag = tlckLINELOCK;
765 	linelock->maxcnt = TLOCKSHORT;
766 	linelock->index = 0;
767 
768 	switch (type & tlckTYPE) {
769 	case tlckDTREE:
770 		linelock->l2linesize = L2DTSLOTSIZE;
771 		break;
772 
773 	case tlckXTREE:
774 		linelock->l2linesize = L2XTSLOTSIZE;
775 
776 		xtlck = (struct xtlock *) linelock;
777 		xtlck->header.offset = 0;
778 		xtlck->header.length = 2;
779 
780 		if (type & tlckNEW) {
781 			xtlck->lwm.offset = XTENTRYSTART;
782 		} else {
783 			if (mp->xflag & COMMIT_PAGE)
784 				p = (xtpage_t *) mp->data;
785 			else
786 				p = &jfs_ip->i_xtroot;
787 			xtlck->lwm.offset =
788 			    le16_to_cpu(p->header.nextindex);
789 		}
790 		xtlck->lwm.length = 0;	/* ! */
791 		xtlck->twm.offset = 0;
792 		xtlck->hwm.offset = 0;
793 
794 		xtlck->index = 2;
795 		break;
796 
797 	case tlckINODE:
798 		linelock->l2linesize = L2INODESLOTSIZE;
799 		break;
800 
801 	case tlckDATA:
802 		linelock->l2linesize = L2DATASLOTSIZE;
803 		break;
804 
805 	default:
806 		jfs_err("UFO tlock:0x%p", tlck);
807 	}
808 
809 	/*
810 	 * update tlock vector
811 	 */
812       grantLock:
813 	tlck->type |= type;
814 
815 	return tlck;
816 
817 	/*
818 	 * page is being locked by another transaction:
819 	 */
820       waitLock:
821 	/* Only locks on ipimap or ipaimap should reach here */
822 	/* assert(jfs_ip->fileset == AGGREGATE_I); */
823 	if (jfs_ip->fileset != AGGREGATE_I) {
824 		printk(KERN_ERR "txLock: trying to lock locked page!");
825 		print_hex_dump(KERN_ERR, "ip: ", DUMP_PREFIX_ADDRESS, 16, 4,
826 			       ip, sizeof(*ip), 0);
827 		print_hex_dump(KERN_ERR, "mp: ", DUMP_PREFIX_ADDRESS, 16, 4,
828 			       mp, sizeof(*mp), 0);
829 		print_hex_dump(KERN_ERR, "Locker's tblock: ",
830 			       DUMP_PREFIX_ADDRESS, 16, 4, tid_to_tblock(tid),
831 			       sizeof(struct tblock), 0);
832 		print_hex_dump(KERN_ERR, "Tlock: ", DUMP_PREFIX_ADDRESS, 16, 4,
833 			       tlck, sizeof(*tlck), 0);
834 		BUG();
835 	}
836 	INCREMENT(stattx.waitlock);	/* statistics */
837 	TXN_UNLOCK();
838 	release_metapage(mp);
839 	TXN_LOCK();
840 	xtid = tlck->tid;	/* reacquire after dropping TXN_LOCK */
841 
842 	jfs_info("txLock: in waitLock, tid = %d, xtid = %d, lid = %d",
843 		 tid, xtid, lid);
844 
845 	/* Recheck everything since dropping TXN_LOCK */
846 	if (xtid && (tlck->mp == mp) && (mp->lid == lid))
847 		TXN_SLEEP_DROP_LOCK(&tid_to_tblock(xtid)->waitor);
848 	else
849 		TXN_UNLOCK();
850 	jfs_info("txLock: awakened     tid = %d, lid = %d", tid, lid);
851 
852 	return NULL;
853 }
854 
855 /*
856  * NAME:	txRelease()
857  *
858  * FUNCTION:	Release buffers associated with transaction locks, but don't
859  *		mark homeok yet.  The allows other transactions to modify
860  *		buffers, but won't let them go to disk until commit record
861  *		actually gets written.
862  *
863  * PARAMETER:
864  *		tblk	-
865  *
866  * RETURN:	Errors from subroutines.
867  */
txRelease(struct tblock * tblk)868 static void txRelease(struct tblock * tblk)
869 {
870 	struct metapage *mp;
871 	lid_t lid;
872 	struct tlock *tlck;
873 
874 	TXN_LOCK();
875 
876 	for (lid = tblk->next; lid; lid = tlck->next) {
877 		tlck = lid_to_tlock(lid);
878 		if ((mp = tlck->mp) != NULL &&
879 		    (tlck->type & tlckBTROOT) == 0) {
880 			assert(mp->xflag & COMMIT_PAGE);
881 			mp->lid = 0;
882 		}
883 	}
884 
885 	/*
886 	 * wakeup transactions waiting on a page locked
887 	 * by the current transaction
888 	 */
889 	TXN_WAKEUP(&tblk->waitor);
890 
891 	TXN_UNLOCK();
892 }
893 
894 /*
895  * NAME:	txUnlock()
896  *
897  * FUNCTION:	Initiates pageout of pages modified by tid in journalled
898  *		objects and frees their lockwords.
899  */
txUnlock(struct tblock * tblk)900 static void txUnlock(struct tblock * tblk)
901 {
902 	struct tlock *tlck;
903 	struct linelock *linelock;
904 	lid_t lid, next, llid, k;
905 	struct metapage *mp;
906 	struct jfs_log *log;
907 	int difft, diffp;
908 	unsigned long flags;
909 
910 	jfs_info("txUnlock: tblk = 0x%p", tblk);
911 	log = JFS_SBI(tblk->sb)->log;
912 
913 	/*
914 	 * mark page under tlock homeok (its log has been written):
915 	 */
916 	for (lid = tblk->next; lid; lid = next) {
917 		tlck = lid_to_tlock(lid);
918 		next = tlck->next;
919 
920 		jfs_info("unlocking lid = %d, tlck = 0x%p", lid, tlck);
921 
922 		/* unbind page from tlock */
923 		if ((mp = tlck->mp) != NULL &&
924 		    (tlck->type & tlckBTROOT) == 0) {
925 			assert(mp->xflag & COMMIT_PAGE);
926 
927 			/* hold buffer
928 			 */
929 			hold_metapage(mp);
930 
931 			assert(mp->nohomeok > 0);
932 			_metapage_homeok(mp);
933 
934 			/* inherit younger/larger clsn */
935 			LOGSYNC_LOCK(log, flags);
936 			if (mp->clsn) {
937 				logdiff(difft, tblk->clsn, log);
938 				logdiff(diffp, mp->clsn, log);
939 				if (difft > diffp)
940 					mp->clsn = tblk->clsn;
941 			} else
942 				mp->clsn = tblk->clsn;
943 			LOGSYNC_UNLOCK(log, flags);
944 
945 			assert(!(tlck->flag & tlckFREEPAGE));
946 
947 			put_metapage(mp);
948 		}
949 
950 		/* insert tlock, and linelock(s) of the tlock if any,
951 		 * at head of freelist
952 		 */
953 		TXN_LOCK();
954 
955 		llid = ((struct linelock *) & tlck->lock)->next;
956 		while (llid) {
957 			linelock = (struct linelock *) lid_to_tlock(llid);
958 			k = linelock->next;
959 			txLockFree(llid);
960 			llid = k;
961 		}
962 		txLockFree(lid);
963 
964 		TXN_UNLOCK();
965 	}
966 	tblk->next = tblk->last = 0;
967 
968 	/*
969 	 * remove tblock from logsynclist
970 	 * (allocation map pages inherited lsn of tblk and
971 	 * has been inserted in logsync list at txUpdateMap())
972 	 */
973 	if (tblk->lsn) {
974 		LOGSYNC_LOCK(log, flags);
975 		log->count--;
976 		list_del(&tblk->synclist);
977 		LOGSYNC_UNLOCK(log, flags);
978 	}
979 }
980 
981 /*
982  *	txMaplock()
983  *
984  * function: allocate a transaction lock for freed page/entry;
985  *	for freed page, maplock is used as xtlock/dtlock type;
986  */
txMaplock(tid_t tid,struct inode * ip,int type)987 struct tlock *txMaplock(tid_t tid, struct inode *ip, int type)
988 {
989 	struct jfs_inode_info *jfs_ip = JFS_IP(ip);
990 	lid_t lid;
991 	struct tblock *tblk;
992 	struct tlock *tlck;
993 	struct maplock *maplock;
994 
995 	TXN_LOCK();
996 
997 	/*
998 	 * allocate a tlock
999 	 */
1000 	lid = txLockAlloc();
1001 	tlck = lid_to_tlock(lid);
1002 
1003 	/*
1004 	 * initialize tlock
1005 	 */
1006 	tlck->tid = tid;
1007 
1008 	/* bind the tlock and the object */
1009 	tlck->flag = tlckINODELOCK;
1010 	if (S_ISDIR(ip->i_mode))
1011 		tlck->flag |= tlckDIRECTORY;
1012 	tlck->ip = ip;
1013 	tlck->mp = NULL;
1014 
1015 	tlck->type = type;
1016 
1017 	/*
1018 	 * enqueue transaction lock to transaction/inode
1019 	 */
1020 	/* insert the tlock at tail of transaction tlock list */
1021 	if (tid) {
1022 		tblk = tid_to_tblock(tid);
1023 		if (tblk->next)
1024 			lid_to_tlock(tblk->last)->next = lid;
1025 		else
1026 			tblk->next = lid;
1027 		tlck->next = 0;
1028 		tblk->last = lid;
1029 	}
1030 	/* anonymous transaction:
1031 	 * insert the tlock at head of inode anonymous tlock list
1032 	 */
1033 	else {
1034 		tlck->next = jfs_ip->atlhead;
1035 		jfs_ip->atlhead = lid;
1036 		if (tlck->next == 0) {
1037 			/* This inode's first anonymous transaction */
1038 			jfs_ip->atltail = lid;
1039 			list_add_tail(&jfs_ip->anon_inode_list,
1040 				      &TxAnchor.anon_list);
1041 		}
1042 	}
1043 
1044 	TXN_UNLOCK();
1045 
1046 	/* initialize type dependent area for maplock */
1047 	maplock = (struct maplock *) & tlck->lock;
1048 	maplock->next = 0;
1049 	maplock->maxcnt = 0;
1050 	maplock->index = 0;
1051 
1052 	return tlck;
1053 }
1054 
1055 /*
1056  *	txLinelock()
1057  *
1058  * function: allocate a transaction lock for log vector list
1059  */
txLinelock(struct linelock * tlock)1060 struct linelock *txLinelock(struct linelock * tlock)
1061 {
1062 	lid_t lid;
1063 	struct tlock *tlck;
1064 	struct linelock *linelock;
1065 
1066 	TXN_LOCK();
1067 
1068 	/* allocate a TxLock structure */
1069 	lid = txLockAlloc();
1070 	tlck = lid_to_tlock(lid);
1071 
1072 	TXN_UNLOCK();
1073 
1074 	/* initialize linelock */
1075 	linelock = (struct linelock *) tlck;
1076 	linelock->next = 0;
1077 	linelock->flag = tlckLINELOCK;
1078 	linelock->maxcnt = TLOCKLONG;
1079 	linelock->index = 0;
1080 	if (tlck->flag & tlckDIRECTORY)
1081 		linelock->flag |= tlckDIRECTORY;
1082 
1083 	/* append linelock after tlock */
1084 	linelock->next = tlock->next;
1085 	tlock->next = lid;
1086 
1087 	return linelock;
1088 }
1089 
1090 /*
1091  *		transaction commit management
1092  *		-----------------------------
1093  */
1094 
1095 /*
1096  * NAME:	txCommit()
1097  *
1098  * FUNCTION:	commit the changes to the objects specified in
1099  *		clist.  For journalled segments only the
1100  *		changes of the caller are committed, ie by tid.
1101  *		for non-journalled segments the data are flushed to
1102  *		disk and then the change to the disk inode and indirect
1103  *		blocks committed (so blocks newly allocated to the
1104  *		segment will be made a part of the segment atomically).
1105  *
1106  *		all of the segments specified in clist must be in
1107  *		one file system. no more than 6 segments are needed
1108  *		to handle all unix svcs.
1109  *
1110  *		if the i_nlink field (i.e. disk inode link count)
1111  *		is zero, and the type of inode is a regular file or
1112  *		directory, or symbolic link , the inode is truncated
1113  *		to zero length. the truncation is committed but the
1114  *		VM resources are unaffected until it is closed (see
1115  *		iput and iclose).
1116  *
1117  * PARAMETER:
1118  *
1119  * RETURN:
1120  *
1121  * serialization:
1122  *		on entry the inode lock on each segment is assumed
1123  *		to be held.
1124  *
1125  * i/o error:
1126  */
txCommit(tid_t tid,int nip,struct inode ** iplist,int flag)1127 int txCommit(tid_t tid,		/* transaction identifier */
1128 	     int nip,		/* number of inodes to commit */
1129 	     struct inode **iplist,	/* list of inode to commit */
1130 	     int flag)
1131 {
1132 	int rc = 0;
1133 	struct commit cd;
1134 	struct jfs_log *log;
1135 	struct tblock *tblk;
1136 	struct lrd *lrd;
1137 	struct inode *ip;
1138 	struct jfs_inode_info *jfs_ip;
1139 	int k, n;
1140 	ino_t top;
1141 	struct super_block *sb;
1142 
1143 	jfs_info("txCommit, tid = %d, flag = %d", tid, flag);
1144 	/* is read-only file system ? */
1145 	if (isReadOnly(iplist[0])) {
1146 		rc = -EROFS;
1147 		goto TheEnd;
1148 	}
1149 
1150 	sb = cd.sb = iplist[0]->i_sb;
1151 	cd.tid = tid;
1152 
1153 	if (tid == 0)
1154 		tid = txBegin(sb, 0);
1155 	tblk = tid_to_tblock(tid);
1156 
1157 	/*
1158 	 * initialize commit structure
1159 	 */
1160 	log = JFS_SBI(sb)->log;
1161 	cd.log = log;
1162 
1163 	/* initialize log record descriptor in commit */
1164 	lrd = &cd.lrd;
1165 	lrd->logtid = cpu_to_le32(tblk->logtid);
1166 	lrd->backchain = 0;
1167 
1168 	tblk->xflag |= flag;
1169 
1170 	if ((flag & (COMMIT_FORCE | COMMIT_SYNC)) == 0)
1171 		tblk->xflag |= COMMIT_LAZY;
1172 	/*
1173 	 *	prepare non-journaled objects for commit
1174 	 *
1175 	 * flush data pages of non-journaled file
1176 	 * to prevent the file getting non-initialized disk blocks
1177 	 * in case of crash.
1178 	 * (new blocks - )
1179 	 */
1180 	cd.iplist = iplist;
1181 	cd.nip = nip;
1182 
1183 	/*
1184 	 *	acquire transaction lock on (on-disk) inodes
1185 	 *
1186 	 * update on-disk inode from in-memory inode
1187 	 * acquiring transaction locks for AFTER records
1188 	 * on the on-disk inode of file object
1189 	 *
1190 	 * sort the inodes array by inode number in descending order
1191 	 * to prevent deadlock when acquiring transaction lock
1192 	 * of on-disk inodes on multiple on-disk inode pages by
1193 	 * multiple concurrent transactions
1194 	 */
1195 	for (k = 0; k < cd.nip; k++) {
1196 		top = (cd.iplist[k])->i_ino;
1197 		for (n = k + 1; n < cd.nip; n++) {
1198 			ip = cd.iplist[n];
1199 			if (ip->i_ino > top) {
1200 				top = ip->i_ino;
1201 				cd.iplist[n] = cd.iplist[k];
1202 				cd.iplist[k] = ip;
1203 			}
1204 		}
1205 
1206 		ip = cd.iplist[k];
1207 		jfs_ip = JFS_IP(ip);
1208 
1209 		/*
1210 		 * BUGBUG - This code has temporarily been removed.  The
1211 		 * intent is to ensure that any file data is written before
1212 		 * the metadata is committed to the journal.  This prevents
1213 		 * uninitialized data from appearing in a file after the
1214 		 * journal has been replayed.  (The uninitialized data
1215 		 * could be sensitive data removed by another user.)
1216 		 *
1217 		 * The problem now is that we are holding the IWRITELOCK
1218 		 * on the inode, and calling filemap_fdatawrite on an
1219 		 * unmapped page will cause a deadlock in jfs_get_block.
1220 		 *
1221 		 * The long term solution is to pare down the use of
1222 		 * IWRITELOCK.  We are currently holding it too long.
1223 		 * We could also be smarter about which data pages need
1224 		 * to be written before the transaction is committed and
1225 		 * when we don't need to worry about it at all.
1226 		 *
1227 		 * if ((!S_ISDIR(ip->i_mode))
1228 		 *    && (tblk->flag & COMMIT_DELETE) == 0)
1229 		 *	filemap_write_and_wait(ip->i_mapping);
1230 		 */
1231 
1232 		/*
1233 		 * Mark inode as not dirty.  It will still be on the dirty
1234 		 * inode list, but we'll know not to commit it again unless
1235 		 * it gets marked dirty again
1236 		 */
1237 		clear_cflag(COMMIT_Dirty, ip);
1238 
1239 		/* inherit anonymous tlock(s) of inode */
1240 		if (jfs_ip->atlhead) {
1241 			lid_to_tlock(jfs_ip->atltail)->next = tblk->next;
1242 			tblk->next = jfs_ip->atlhead;
1243 			if (!tblk->last)
1244 				tblk->last = jfs_ip->atltail;
1245 			jfs_ip->atlhead = jfs_ip->atltail = 0;
1246 			TXN_LOCK();
1247 			list_del_init(&jfs_ip->anon_inode_list);
1248 			TXN_UNLOCK();
1249 		}
1250 
1251 		/*
1252 		 * acquire transaction lock on on-disk inode page
1253 		 * (become first tlock of the tblk's tlock list)
1254 		 */
1255 		if (((rc = diWrite(tid, ip))))
1256 			goto out;
1257 	}
1258 
1259 	/*
1260 	 *	write log records from transaction locks
1261 	 *
1262 	 * txUpdateMap() resets XAD_NEW in XAD.
1263 	 */
1264 	txLog(log, tblk, &cd);
1265 
1266 	/*
1267 	 * Ensure that inode isn't reused before
1268 	 * lazy commit thread finishes processing
1269 	 */
1270 	if (tblk->xflag & COMMIT_DELETE) {
1271 		ihold(tblk->u.ip);
1272 		/*
1273 		 * Avoid a rare deadlock
1274 		 *
1275 		 * If the inode is locked, we may be blocked in
1276 		 * jfs_commit_inode.  If so, we don't want the
1277 		 * lazy_commit thread doing the last iput() on the inode
1278 		 * since that may block on the locked inode.  Instead,
1279 		 * commit the transaction synchronously, so the last iput
1280 		 * will be done by the calling thread (or later)
1281 		 */
1282 		/*
1283 		 * I believe this code is no longer needed.  Splitting I_LOCK
1284 		 * into two bits, I_NEW and I_SYNC should prevent this
1285 		 * deadlock as well.  But since I don't have a JFS testload
1286 		 * to verify this, only a trivial s/I_LOCK/I_SYNC/ was done.
1287 		 * Joern
1288 		 */
1289 		if (tblk->u.ip->i_state & I_SYNC)
1290 			tblk->xflag &= ~COMMIT_LAZY;
1291 	}
1292 
1293 	ASSERT((!(tblk->xflag & COMMIT_DELETE)) ||
1294 	       ((tblk->u.ip->i_nlink == 0) &&
1295 		!test_cflag(COMMIT_Nolink, tblk->u.ip)));
1296 
1297 	/*
1298 	 *	write COMMIT log record
1299 	 */
1300 	lrd->type = cpu_to_le16(LOG_COMMIT);
1301 	lrd->length = 0;
1302 	lmLog(log, tblk, lrd, NULL);
1303 
1304 	lmGroupCommit(log, tblk);
1305 
1306 	/*
1307 	 *	- transaction is now committed -
1308 	 */
1309 
1310 	/*
1311 	 * force pages in careful update
1312 	 * (imap addressing structure update)
1313 	 */
1314 	if (flag & COMMIT_FORCE)
1315 		txForce(tblk);
1316 
1317 	/*
1318 	 *	update allocation map.
1319 	 *
1320 	 * update inode allocation map and inode:
1321 	 * free pager lock on memory object of inode if any.
1322 	 * update block allocation map.
1323 	 *
1324 	 * txUpdateMap() resets XAD_NEW in XAD.
1325 	 */
1326 	if (tblk->xflag & COMMIT_FORCE)
1327 		txUpdateMap(tblk);
1328 
1329 	/*
1330 	 *	free transaction locks and pageout/free pages
1331 	 */
1332 	txRelease(tblk);
1333 
1334 	if ((tblk->flag & tblkGC_LAZY) == 0)
1335 		txUnlock(tblk);
1336 
1337 
1338 	/*
1339 	 *	reset in-memory object state
1340 	 */
1341 	for (k = 0; k < cd.nip; k++) {
1342 		ip = cd.iplist[k];
1343 		jfs_ip = JFS_IP(ip);
1344 
1345 		/*
1346 		 * reset in-memory inode state
1347 		 */
1348 		jfs_ip->bxflag = 0;
1349 		jfs_ip->blid = 0;
1350 	}
1351 
1352       out:
1353 	if (rc != 0)
1354 		txAbort(tid, 1);
1355 
1356       TheEnd:
1357 	jfs_info("txCommit: tid = %d, returning %d", tid, rc);
1358 	return rc;
1359 }
1360 
1361 /*
1362  * NAME:	txLog()
1363  *
1364  * FUNCTION:	Writes AFTER log records for all lines modified
1365  *		by tid for segments specified by inodes in comdata.
1366  *		Code assumes only WRITELOCKS are recorded in lockwords.
1367  *
1368  * PARAMETERS:
1369  *
1370  * RETURN :
1371  */
txLog(struct jfs_log * log,struct tblock * tblk,struct commit * cd)1372 static void txLog(struct jfs_log *log, struct tblock *tblk, struct commit *cd)
1373 {
1374 	struct inode *ip;
1375 	lid_t lid;
1376 	struct tlock *tlck;
1377 	struct lrd *lrd = &cd->lrd;
1378 
1379 	/*
1380 	 * write log record(s) for each tlock of transaction,
1381 	 */
1382 	for (lid = tblk->next; lid; lid = tlck->next) {
1383 		tlck = lid_to_tlock(lid);
1384 
1385 		tlck->flag |= tlckLOG;
1386 
1387 		/* initialize lrd common */
1388 		ip = tlck->ip;
1389 		lrd->aggregate = cpu_to_le32(JFS_SBI(ip->i_sb)->aggregate);
1390 		lrd->log.redopage.fileset = cpu_to_le32(JFS_IP(ip)->fileset);
1391 		lrd->log.redopage.inode = cpu_to_le32(ip->i_ino);
1392 
1393 		/* write log record of page from the tlock */
1394 		switch (tlck->type & tlckTYPE) {
1395 		case tlckXTREE:
1396 			xtLog(log, tblk, lrd, tlck);
1397 			break;
1398 
1399 		case tlckDTREE:
1400 			dtLog(log, tblk, lrd, tlck);
1401 			break;
1402 
1403 		case tlckINODE:
1404 			diLog(log, tblk, lrd, tlck, cd);
1405 			break;
1406 
1407 		case tlckMAP:
1408 			mapLog(log, tblk, lrd, tlck);
1409 			break;
1410 
1411 		case tlckDATA:
1412 			dataLog(log, tblk, lrd, tlck);
1413 			break;
1414 
1415 		default:
1416 			jfs_err("UFO tlock:0x%p", tlck);
1417 		}
1418 	}
1419 
1420 	return;
1421 }
1422 
1423 /*
1424  *	diLog()
1425  *
1426  * function:	log inode tlock and format maplock to update bmap;
1427  */
diLog(struct jfs_log * log,struct tblock * tblk,struct lrd * lrd,struct tlock * tlck,struct commit * cd)1428 static void diLog(struct jfs_log *log, struct tblock *tblk, struct lrd *lrd,
1429 		 struct tlock *tlck, struct commit *cd)
1430 {
1431 	struct metapage *mp;
1432 	pxd_t *pxd;
1433 	struct pxd_lock *pxdlock;
1434 
1435 	mp = tlck->mp;
1436 
1437 	/* initialize as REDOPAGE record format */
1438 	lrd->log.redopage.type = cpu_to_le16(LOG_INODE);
1439 	lrd->log.redopage.l2linesize = cpu_to_le16(L2INODESLOTSIZE);
1440 
1441 	pxd = &lrd->log.redopage.pxd;
1442 
1443 	/*
1444 	 *	inode after image
1445 	 */
1446 	if (tlck->type & tlckENTRY) {
1447 		/* log after-image for logredo(): */
1448 		lrd->type = cpu_to_le16(LOG_REDOPAGE);
1449 		PXDaddress(pxd, mp->index);
1450 		PXDlength(pxd,
1451 			  mp->logical_size >> tblk->sb->s_blocksize_bits);
1452 		lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1453 
1454 		/* mark page as homeward bound */
1455 		tlck->flag |= tlckWRITEPAGE;
1456 	} else if (tlck->type & tlckFREE) {
1457 		/*
1458 		 *	free inode extent
1459 		 *
1460 		 * (pages of the freed inode extent have been invalidated and
1461 		 * a maplock for free of the extent has been formatted at
1462 		 * txLock() time);
1463 		 *
1464 		 * the tlock had been acquired on the inode allocation map page
1465 		 * (iag) that specifies the freed extent, even though the map
1466 		 * page is not itself logged, to prevent pageout of the map
1467 		 * page before the log;
1468 		 */
1469 
1470 		/* log LOG_NOREDOINOEXT of the freed inode extent for
1471 		 * logredo() to start NoRedoPage filters, and to update
1472 		 * imap and bmap for free of the extent;
1473 		 */
1474 		lrd->type = cpu_to_le16(LOG_NOREDOINOEXT);
1475 		/*
1476 		 * For the LOG_NOREDOINOEXT record, we need
1477 		 * to pass the IAG number and inode extent
1478 		 * index (within that IAG) from which the
1479 		 * extent is being released.  These have been
1480 		 * passed to us in the iplist[1] and iplist[2].
1481 		 */
1482 		lrd->log.noredoinoext.iagnum =
1483 		    cpu_to_le32((u32) (size_t) cd->iplist[1]);
1484 		lrd->log.noredoinoext.inoext_idx =
1485 		    cpu_to_le32((u32) (size_t) cd->iplist[2]);
1486 
1487 		pxdlock = (struct pxd_lock *) & tlck->lock;
1488 		*pxd = pxdlock->pxd;
1489 		lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1490 
1491 		/* update bmap */
1492 		tlck->flag |= tlckUPDATEMAP;
1493 
1494 		/* mark page as homeward bound */
1495 		tlck->flag |= tlckWRITEPAGE;
1496 	} else
1497 		jfs_err("diLog: UFO type tlck:0x%p", tlck);
1498 #ifdef  _JFS_WIP
1499 	/*
1500 	 *	alloc/free external EA extent
1501 	 *
1502 	 * a maplock for txUpdateMap() to update bPWMAP for alloc/free
1503 	 * of the extent has been formatted at txLock() time;
1504 	 */
1505 	else {
1506 		assert(tlck->type & tlckEA);
1507 
1508 		/* log LOG_UPDATEMAP for logredo() to update bmap for
1509 		 * alloc of new (and free of old) external EA extent;
1510 		 */
1511 		lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1512 		pxdlock = (struct pxd_lock *) & tlck->lock;
1513 		nlock = pxdlock->index;
1514 		for (i = 0; i < nlock; i++, pxdlock++) {
1515 			if (pxdlock->flag & mlckALLOCPXD)
1516 				lrd->log.updatemap.type =
1517 				    cpu_to_le16(LOG_ALLOCPXD);
1518 			else
1519 				lrd->log.updatemap.type =
1520 				    cpu_to_le16(LOG_FREEPXD);
1521 			lrd->log.updatemap.nxd = cpu_to_le16(1);
1522 			lrd->log.updatemap.pxd = pxdlock->pxd;
1523 			lrd->backchain =
1524 			    cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1525 		}
1526 
1527 		/* update bmap */
1528 		tlck->flag |= tlckUPDATEMAP;
1529 	}
1530 #endif				/* _JFS_WIP */
1531 
1532 	return;
1533 }
1534 
1535 /*
1536  *	dataLog()
1537  *
1538  * function:	log data tlock
1539  */
dataLog(struct jfs_log * log,struct tblock * tblk,struct lrd * lrd,struct tlock * tlck)1540 static void dataLog(struct jfs_log *log, struct tblock *tblk, struct lrd *lrd,
1541 	    struct tlock *tlck)
1542 {
1543 	struct metapage *mp;
1544 	pxd_t *pxd;
1545 
1546 	mp = tlck->mp;
1547 
1548 	/* initialize as REDOPAGE record format */
1549 	lrd->log.redopage.type = cpu_to_le16(LOG_DATA);
1550 	lrd->log.redopage.l2linesize = cpu_to_le16(L2DATASLOTSIZE);
1551 
1552 	pxd = &lrd->log.redopage.pxd;
1553 
1554 	/* log after-image for logredo(): */
1555 	lrd->type = cpu_to_le16(LOG_REDOPAGE);
1556 
1557 	if (jfs_dirtable_inline(tlck->ip)) {
1558 		/*
1559 		 * The table has been truncated, we've must have deleted
1560 		 * the last entry, so don't bother logging this
1561 		 */
1562 		mp->lid = 0;
1563 		grab_metapage(mp);
1564 		metapage_homeok(mp);
1565 		discard_metapage(mp);
1566 		tlck->mp = NULL;
1567 		return;
1568 	}
1569 
1570 	PXDaddress(pxd, mp->index);
1571 	PXDlength(pxd, mp->logical_size >> tblk->sb->s_blocksize_bits);
1572 
1573 	lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1574 
1575 	/* mark page as homeward bound */
1576 	tlck->flag |= tlckWRITEPAGE;
1577 
1578 	return;
1579 }
1580 
1581 /*
1582  *	dtLog()
1583  *
1584  * function:	log dtree tlock and format maplock to update bmap;
1585  */
dtLog(struct jfs_log * log,struct tblock * tblk,struct lrd * lrd,struct tlock * tlck)1586 static void dtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1587 	   struct tlock * tlck)
1588 {
1589 	struct metapage *mp;
1590 	struct pxd_lock *pxdlock;
1591 	pxd_t *pxd;
1592 
1593 	mp = tlck->mp;
1594 
1595 	/* initialize as REDOPAGE/NOREDOPAGE record format */
1596 	lrd->log.redopage.type = cpu_to_le16(LOG_DTREE);
1597 	lrd->log.redopage.l2linesize = cpu_to_le16(L2DTSLOTSIZE);
1598 
1599 	pxd = &lrd->log.redopage.pxd;
1600 
1601 	if (tlck->type & tlckBTROOT)
1602 		lrd->log.redopage.type |= cpu_to_le16(LOG_BTROOT);
1603 
1604 	/*
1605 	 *	page extension via relocation: entry insertion;
1606 	 *	page extension in-place: entry insertion;
1607 	 *	new right page from page split, reinitialized in-line
1608 	 *	root from root page split: entry insertion;
1609 	 */
1610 	if (tlck->type & (tlckNEW | tlckEXTEND)) {
1611 		/* log after-image of the new page for logredo():
1612 		 * mark log (LOG_NEW) for logredo() to initialize
1613 		 * freelist and update bmap for alloc of the new page;
1614 		 */
1615 		lrd->type = cpu_to_le16(LOG_REDOPAGE);
1616 		if (tlck->type & tlckEXTEND)
1617 			lrd->log.redopage.type |= cpu_to_le16(LOG_EXTEND);
1618 		else
1619 			lrd->log.redopage.type |= cpu_to_le16(LOG_NEW);
1620 		PXDaddress(pxd, mp->index);
1621 		PXDlength(pxd,
1622 			  mp->logical_size >> tblk->sb->s_blocksize_bits);
1623 		lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1624 
1625 		/* format a maplock for txUpdateMap() to update bPMAP for
1626 		 * alloc of the new page;
1627 		 */
1628 		if (tlck->type & tlckBTROOT)
1629 			return;
1630 		tlck->flag |= tlckUPDATEMAP;
1631 		pxdlock = (struct pxd_lock *) & tlck->lock;
1632 		pxdlock->flag = mlckALLOCPXD;
1633 		pxdlock->pxd = *pxd;
1634 
1635 		pxdlock->index = 1;
1636 
1637 		/* mark page as homeward bound */
1638 		tlck->flag |= tlckWRITEPAGE;
1639 		return;
1640 	}
1641 
1642 	/*
1643 	 *	entry insertion/deletion,
1644 	 *	sibling page link update (old right page before split);
1645 	 */
1646 	if (tlck->type & (tlckENTRY | tlckRELINK)) {
1647 		/* log after-image for logredo(): */
1648 		lrd->type = cpu_to_le16(LOG_REDOPAGE);
1649 		PXDaddress(pxd, mp->index);
1650 		PXDlength(pxd,
1651 			  mp->logical_size >> tblk->sb->s_blocksize_bits);
1652 		lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1653 
1654 		/* mark page as homeward bound */
1655 		tlck->flag |= tlckWRITEPAGE;
1656 		return;
1657 	}
1658 
1659 	/*
1660 	 *	page deletion: page has been invalidated
1661 	 *	page relocation: source extent
1662 	 *
1663 	 *	a maplock for free of the page has been formatted
1664 	 *	at txLock() time);
1665 	 */
1666 	if (tlck->type & (tlckFREE | tlckRELOCATE)) {
1667 		/* log LOG_NOREDOPAGE of the deleted page for logredo()
1668 		 * to start NoRedoPage filter and to update bmap for free
1669 		 * of the deletd page
1670 		 */
1671 		lrd->type = cpu_to_le16(LOG_NOREDOPAGE);
1672 		pxdlock = (struct pxd_lock *) & tlck->lock;
1673 		*pxd = pxdlock->pxd;
1674 		lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1675 
1676 		/* a maplock for txUpdateMap() for free of the page
1677 		 * has been formatted at txLock() time;
1678 		 */
1679 		tlck->flag |= tlckUPDATEMAP;
1680 	}
1681 	return;
1682 }
1683 
1684 /*
1685  *	xtLog()
1686  *
1687  * function:	log xtree tlock and format maplock to update bmap;
1688  */
xtLog(struct jfs_log * log,struct tblock * tblk,struct lrd * lrd,struct tlock * tlck)1689 static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1690 	   struct tlock * tlck)
1691 {
1692 	struct inode *ip;
1693 	struct metapage *mp;
1694 	xtpage_t *p;
1695 	struct xtlock *xtlck;
1696 	struct maplock *maplock;
1697 	struct xdlistlock *xadlock;
1698 	struct pxd_lock *pxdlock;
1699 	pxd_t *page_pxd;
1700 	int next, lwm, hwm;
1701 
1702 	ip = tlck->ip;
1703 	mp = tlck->mp;
1704 
1705 	/* initialize as REDOPAGE/NOREDOPAGE record format */
1706 	lrd->log.redopage.type = cpu_to_le16(LOG_XTREE);
1707 	lrd->log.redopage.l2linesize = cpu_to_le16(L2XTSLOTSIZE);
1708 
1709 	page_pxd = &lrd->log.redopage.pxd;
1710 
1711 	if (tlck->type & tlckBTROOT) {
1712 		lrd->log.redopage.type |= cpu_to_le16(LOG_BTROOT);
1713 		p = &JFS_IP(ip)->i_xtroot;
1714 		if (S_ISDIR(ip->i_mode))
1715 			lrd->log.redopage.type |=
1716 			    cpu_to_le16(LOG_DIR_XTREE);
1717 	} else
1718 		p = (xtpage_t *) mp->data;
1719 	next = le16_to_cpu(p->header.nextindex);
1720 
1721 	xtlck = (struct xtlock *) & tlck->lock;
1722 
1723 	maplock = (struct maplock *) & tlck->lock;
1724 	xadlock = (struct xdlistlock *) maplock;
1725 
1726 	/*
1727 	 *	entry insertion/extension;
1728 	 *	sibling page link update (old right page before split);
1729 	 */
1730 	if (tlck->type & (tlckNEW | tlckGROW | tlckRELINK)) {
1731 		/* log after-image for logredo():
1732 		 * logredo() will update bmap for alloc of new/extended
1733 		 * extents (XAD_NEW|XAD_EXTEND) of XAD[lwm:next) from
1734 		 * after-image of XADlist;
1735 		 * logredo() resets (XAD_NEW|XAD_EXTEND) flag when
1736 		 * applying the after-image to the meta-data page.
1737 		 */
1738 		lrd->type = cpu_to_le16(LOG_REDOPAGE);
1739 		PXDaddress(page_pxd, mp->index);
1740 		PXDlength(page_pxd,
1741 			  mp->logical_size >> tblk->sb->s_blocksize_bits);
1742 		lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1743 
1744 		/* format a maplock for txUpdateMap() to update bPMAP
1745 		 * for alloc of new/extended extents of XAD[lwm:next)
1746 		 * from the page itself;
1747 		 * txUpdateMap() resets (XAD_NEW|XAD_EXTEND) flag.
1748 		 */
1749 		lwm = xtlck->lwm.offset;
1750 		if (lwm == 0)
1751 			lwm = XTPAGEMAXSLOT;
1752 
1753 		if (lwm == next)
1754 			goto out;
1755 		if (lwm > next) {
1756 			jfs_err("xtLog: lwm > next");
1757 			goto out;
1758 		}
1759 		tlck->flag |= tlckUPDATEMAP;
1760 		xadlock->flag = mlckALLOCXADLIST;
1761 		xadlock->count = next - lwm;
1762 		if ((xadlock->count <= 4) && (tblk->xflag & COMMIT_LAZY)) {
1763 			int i;
1764 			pxd_t *pxd;
1765 			/*
1766 			 * Lazy commit may allow xtree to be modified before
1767 			 * txUpdateMap runs.  Copy xad into linelock to
1768 			 * preserve correct data.
1769 			 *
1770 			 * We can fit twice as may pxd's as xads in the lock
1771 			 */
1772 			xadlock->flag = mlckALLOCPXDLIST;
1773 			pxd = xadlock->xdlist = &xtlck->pxdlock;
1774 			for (i = 0; i < xadlock->count; i++) {
1775 				PXDaddress(pxd, addressXAD(&p->xad[lwm + i]));
1776 				PXDlength(pxd, lengthXAD(&p->xad[lwm + i]));
1777 				p->xad[lwm + i].flag &=
1778 				    ~(XAD_NEW | XAD_EXTENDED);
1779 				pxd++;
1780 			}
1781 		} else {
1782 			/*
1783 			 * xdlist will point to into inode's xtree, ensure
1784 			 * that transaction is not committed lazily.
1785 			 */
1786 			xadlock->flag = mlckALLOCXADLIST;
1787 			xadlock->xdlist = &p->xad[lwm];
1788 			tblk->xflag &= ~COMMIT_LAZY;
1789 		}
1790 		jfs_info("xtLog: alloc ip:0x%p mp:0x%p tlck:0x%p lwm:%d count:%d",
1791 			 tlck->ip, mp, tlck, lwm, xadlock->count);
1792 
1793 		maplock->index = 1;
1794 
1795 	      out:
1796 		/* mark page as homeward bound */
1797 		tlck->flag |= tlckWRITEPAGE;
1798 
1799 		return;
1800 	}
1801 
1802 	/*
1803 	 *	page deletion: file deletion/truncation (ref. xtTruncate())
1804 	 *
1805 	 * (page will be invalidated after log is written and bmap
1806 	 * is updated from the page);
1807 	 */
1808 	if (tlck->type & tlckFREE) {
1809 		/* LOG_NOREDOPAGE log for NoRedoPage filter:
1810 		 * if page free from file delete, NoRedoFile filter from
1811 		 * inode image of zero link count will subsume NoRedoPage
1812 		 * filters for each page;
1813 		 * if page free from file truncattion, write NoRedoPage
1814 		 * filter;
1815 		 *
1816 		 * upadte of block allocation map for the page itself:
1817 		 * if page free from deletion and truncation, LOG_UPDATEMAP
1818 		 * log for the page itself is generated from processing
1819 		 * its parent page xad entries;
1820 		 */
1821 		/* if page free from file truncation, log LOG_NOREDOPAGE
1822 		 * of the deleted page for logredo() to start NoRedoPage
1823 		 * filter for the page;
1824 		 */
1825 		if (tblk->xflag & COMMIT_TRUNCATE) {
1826 			/* write NOREDOPAGE for the page */
1827 			lrd->type = cpu_to_le16(LOG_NOREDOPAGE);
1828 			PXDaddress(page_pxd, mp->index);
1829 			PXDlength(page_pxd,
1830 				  mp->logical_size >> tblk->sb->
1831 				  s_blocksize_bits);
1832 			lrd->backchain =
1833 			    cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1834 
1835 			if (tlck->type & tlckBTROOT) {
1836 				/* Empty xtree must be logged */
1837 				lrd->type = cpu_to_le16(LOG_REDOPAGE);
1838 				lrd->backchain =
1839 				    cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1840 			}
1841 		}
1842 
1843 		/* init LOG_UPDATEMAP of the freed extents
1844 		 * XAD[XTENTRYSTART:hwm) from the deleted page itself
1845 		 * for logredo() to update bmap;
1846 		 */
1847 		lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1848 		lrd->log.updatemap.type = cpu_to_le16(LOG_FREEXADLIST);
1849 		xtlck = (struct xtlock *) & tlck->lock;
1850 		hwm = xtlck->hwm.offset;
1851 		lrd->log.updatemap.nxd =
1852 		    cpu_to_le16(hwm - XTENTRYSTART + 1);
1853 		/* reformat linelock for lmLog() */
1854 		xtlck->header.offset = XTENTRYSTART;
1855 		xtlck->header.length = hwm - XTENTRYSTART + 1;
1856 		xtlck->index = 1;
1857 		lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1858 
1859 		/* format a maplock for txUpdateMap() to update bmap
1860 		 * to free extents of XAD[XTENTRYSTART:hwm) from the
1861 		 * deleted page itself;
1862 		 */
1863 		tlck->flag |= tlckUPDATEMAP;
1864 		xadlock->count = hwm - XTENTRYSTART + 1;
1865 		if ((xadlock->count <= 4) && (tblk->xflag & COMMIT_LAZY)) {
1866 			int i;
1867 			pxd_t *pxd;
1868 			/*
1869 			 * Lazy commit may allow xtree to be modified before
1870 			 * txUpdateMap runs.  Copy xad into linelock to
1871 			 * preserve correct data.
1872 			 *
1873 			 * We can fit twice as may pxd's as xads in the lock
1874 			 */
1875 			xadlock->flag = mlckFREEPXDLIST;
1876 			pxd = xadlock->xdlist = &xtlck->pxdlock;
1877 			for (i = 0; i < xadlock->count; i++) {
1878 				PXDaddress(pxd,
1879 					addressXAD(&p->xad[XTENTRYSTART + i]));
1880 				PXDlength(pxd,
1881 					lengthXAD(&p->xad[XTENTRYSTART + i]));
1882 				pxd++;
1883 			}
1884 		} else {
1885 			/*
1886 			 * xdlist will point to into inode's xtree, ensure
1887 			 * that transaction is not committed lazily.
1888 			 */
1889 			xadlock->flag = mlckFREEXADLIST;
1890 			xadlock->xdlist = &p->xad[XTENTRYSTART];
1891 			tblk->xflag &= ~COMMIT_LAZY;
1892 		}
1893 		jfs_info("xtLog: free ip:0x%p mp:0x%p count:%d lwm:2",
1894 			 tlck->ip, mp, xadlock->count);
1895 
1896 		maplock->index = 1;
1897 
1898 		/* mark page as invalid */
1899 		if (((tblk->xflag & COMMIT_PWMAP) || S_ISDIR(ip->i_mode))
1900 		    && !(tlck->type & tlckBTROOT))
1901 			tlck->flag |= tlckFREEPAGE;
1902 		/*
1903 		   else (tblk->xflag & COMMIT_PMAP)
1904 		   ? release the page;
1905 		 */
1906 		return;
1907 	}
1908 
1909 	/*
1910 	 *	page/entry truncation: file truncation (ref. xtTruncate())
1911 	 *
1912 	 *	|----------+------+------+---------------|
1913 	 *		   |      |      |
1914 	 *		   |      |     hwm - hwm before truncation
1915 	 *		   |     next - truncation point
1916 	 *		  lwm - lwm before truncation
1917 	 * header ?
1918 	 */
1919 	if (tlck->type & tlckTRUNCATE) {
1920 		pxd_t pxd;	/* truncated extent of xad */
1921 		int twm;
1922 
1923 		/*
1924 		 * For truncation the entire linelock may be used, so it would
1925 		 * be difficult to store xad list in linelock itself.
1926 		 * Therefore, we'll just force transaction to be committed
1927 		 * synchronously, so that xtree pages won't be changed before
1928 		 * txUpdateMap runs.
1929 		 */
1930 		tblk->xflag &= ~COMMIT_LAZY;
1931 		lwm = xtlck->lwm.offset;
1932 		if (lwm == 0)
1933 			lwm = XTPAGEMAXSLOT;
1934 		hwm = xtlck->hwm.offset;
1935 		twm = xtlck->twm.offset;
1936 
1937 		/*
1938 		 *	write log records
1939 		 */
1940 		/* log after-image for logredo():
1941 		 *
1942 		 * logredo() will update bmap for alloc of new/extended
1943 		 * extents (XAD_NEW|XAD_EXTEND) of XAD[lwm:next) from
1944 		 * after-image of XADlist;
1945 		 * logredo() resets (XAD_NEW|XAD_EXTEND) flag when
1946 		 * applying the after-image to the meta-data page.
1947 		 */
1948 		lrd->type = cpu_to_le16(LOG_REDOPAGE);
1949 		PXDaddress(page_pxd, mp->index);
1950 		PXDlength(page_pxd,
1951 			  mp->logical_size >> tblk->sb->s_blocksize_bits);
1952 		lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1953 
1954 		/*
1955 		 * truncate entry XAD[twm == next - 1]:
1956 		 */
1957 		if (twm == next - 1) {
1958 			/* init LOG_UPDATEMAP for logredo() to update bmap for
1959 			 * free of truncated delta extent of the truncated
1960 			 * entry XAD[next - 1]:
1961 			 * (xtlck->pxdlock = truncated delta extent);
1962 			 */
1963 			pxdlock = (struct pxd_lock *) & xtlck->pxdlock;
1964 			/* assert(pxdlock->type & tlckTRUNCATE); */
1965 			lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1966 			lrd->log.updatemap.type = cpu_to_le16(LOG_FREEPXD);
1967 			lrd->log.updatemap.nxd = cpu_to_le16(1);
1968 			lrd->log.updatemap.pxd = pxdlock->pxd;
1969 			pxd = pxdlock->pxd;	/* save to format maplock */
1970 			lrd->backchain =
1971 			    cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1972 		}
1973 
1974 		/*
1975 		 * free entries XAD[next:hwm]:
1976 		 */
1977 		if (hwm >= next) {
1978 			/* init LOG_UPDATEMAP of the freed extents
1979 			 * XAD[next:hwm] from the deleted page itself
1980 			 * for logredo() to update bmap;
1981 			 */
1982 			lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1983 			lrd->log.updatemap.type =
1984 			    cpu_to_le16(LOG_FREEXADLIST);
1985 			xtlck = (struct xtlock *) & tlck->lock;
1986 			hwm = xtlck->hwm.offset;
1987 			lrd->log.updatemap.nxd =
1988 			    cpu_to_le16(hwm - next + 1);
1989 			/* reformat linelock for lmLog() */
1990 			xtlck->header.offset = next;
1991 			xtlck->header.length = hwm - next + 1;
1992 			xtlck->index = 1;
1993 			lrd->backchain =
1994 			    cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1995 		}
1996 
1997 		/*
1998 		 *	format maplock(s) for txUpdateMap() to update bmap
1999 		 */
2000 		maplock->index = 0;
2001 
2002 		/*
2003 		 * allocate entries XAD[lwm:next):
2004 		 */
2005 		if (lwm < next) {
2006 			/* format a maplock for txUpdateMap() to update bPMAP
2007 			 * for alloc of new/extended extents of XAD[lwm:next)
2008 			 * from the page itself;
2009 			 * txUpdateMap() resets (XAD_NEW|XAD_EXTEND) flag.
2010 			 */
2011 			tlck->flag |= tlckUPDATEMAP;
2012 			xadlock->flag = mlckALLOCXADLIST;
2013 			xadlock->count = next - lwm;
2014 			xadlock->xdlist = &p->xad[lwm];
2015 
2016 			jfs_info("xtLog: alloc ip:0x%p mp:0x%p count:%d lwm:%d next:%d",
2017 				 tlck->ip, mp, xadlock->count, lwm, next);
2018 			maplock->index++;
2019 			xadlock++;
2020 		}
2021 
2022 		/*
2023 		 * truncate entry XAD[twm == next - 1]:
2024 		 */
2025 		if (twm == next - 1) {
2026 			/* format a maplock for txUpdateMap() to update bmap
2027 			 * to free truncated delta extent of the truncated
2028 			 * entry XAD[next - 1];
2029 			 * (xtlck->pxdlock = truncated delta extent);
2030 			 */
2031 			tlck->flag |= tlckUPDATEMAP;
2032 			pxdlock = (struct pxd_lock *) xadlock;
2033 			pxdlock->flag = mlckFREEPXD;
2034 			pxdlock->count = 1;
2035 			pxdlock->pxd = pxd;
2036 
2037 			jfs_info("xtLog: truncate ip:0x%p mp:0x%p count:%d hwm:%d",
2038 				 ip, mp, pxdlock->count, hwm);
2039 			maplock->index++;
2040 			xadlock++;
2041 		}
2042 
2043 		/*
2044 		 * free entries XAD[next:hwm]:
2045 		 */
2046 		if (hwm >= next) {
2047 			/* format a maplock for txUpdateMap() to update bmap
2048 			 * to free extents of XAD[next:hwm] from thedeleted
2049 			 * page itself;
2050 			 */
2051 			tlck->flag |= tlckUPDATEMAP;
2052 			xadlock->flag = mlckFREEXADLIST;
2053 			xadlock->count = hwm - next + 1;
2054 			xadlock->xdlist = &p->xad[next];
2055 
2056 			jfs_info("xtLog: free ip:0x%p mp:0x%p count:%d next:%d hwm:%d",
2057 				 tlck->ip, mp, xadlock->count, next, hwm);
2058 			maplock->index++;
2059 		}
2060 
2061 		/* mark page as homeward bound */
2062 		tlck->flag |= tlckWRITEPAGE;
2063 	}
2064 	return;
2065 }
2066 
2067 /*
2068  *	mapLog()
2069  *
2070  * function:	log from maplock of freed data extents;
2071  */
mapLog(struct jfs_log * log,struct tblock * tblk,struct lrd * lrd,struct tlock * tlck)2072 static void mapLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
2073 		   struct tlock * tlck)
2074 {
2075 	struct pxd_lock *pxdlock;
2076 	int i, nlock;
2077 	pxd_t *pxd;
2078 
2079 	/*
2080 	 *	page relocation: free the source page extent
2081 	 *
2082 	 * a maplock for txUpdateMap() for free of the page
2083 	 * has been formatted at txLock() time saving the src
2084 	 * relocated page address;
2085 	 */
2086 	if (tlck->type & tlckRELOCATE) {
2087 		/* log LOG_NOREDOPAGE of the old relocated page
2088 		 * for logredo() to start NoRedoPage filter;
2089 		 */
2090 		lrd->type = cpu_to_le16(LOG_NOREDOPAGE);
2091 		pxdlock = (struct pxd_lock *) & tlck->lock;
2092 		pxd = &lrd->log.redopage.pxd;
2093 		*pxd = pxdlock->pxd;
2094 		lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
2095 
2096 		/* (N.B. currently, logredo() does NOT update bmap
2097 		 * for free of the page itself for (LOG_XTREE|LOG_NOREDOPAGE);
2098 		 * if page free from relocation, LOG_UPDATEMAP log is
2099 		 * specifically generated now for logredo()
2100 		 * to update bmap for free of src relocated page;
2101 		 * (new flag LOG_RELOCATE may be introduced which will
2102 		 * inform logredo() to start NORedoPage filter and also
2103 		 * update block allocation map at the same time, thus
2104 		 * avoiding an extra log write);
2105 		 */
2106 		lrd->type = cpu_to_le16(LOG_UPDATEMAP);
2107 		lrd->log.updatemap.type = cpu_to_le16(LOG_FREEPXD);
2108 		lrd->log.updatemap.nxd = cpu_to_le16(1);
2109 		lrd->log.updatemap.pxd = pxdlock->pxd;
2110 		lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
2111 
2112 		/* a maplock for txUpdateMap() for free of the page
2113 		 * has been formatted at txLock() time;
2114 		 */
2115 		tlck->flag |= tlckUPDATEMAP;
2116 		return;
2117 	}
2118 	/*
2119 
2120 	 * Otherwise it's not a relocate request
2121 	 *
2122 	 */
2123 	else {
2124 		/* log LOG_UPDATEMAP for logredo() to update bmap for
2125 		 * free of truncated/relocated delta extent of the data;
2126 		 * e.g.: external EA extent, relocated/truncated extent
2127 		 * from xtTailgate();
2128 		 */
2129 		lrd->type = cpu_to_le16(LOG_UPDATEMAP);
2130 		pxdlock = (struct pxd_lock *) & tlck->lock;
2131 		nlock = pxdlock->index;
2132 		for (i = 0; i < nlock; i++, pxdlock++) {
2133 			if (pxdlock->flag & mlckALLOCPXD)
2134 				lrd->log.updatemap.type =
2135 				    cpu_to_le16(LOG_ALLOCPXD);
2136 			else
2137 				lrd->log.updatemap.type =
2138 				    cpu_to_le16(LOG_FREEPXD);
2139 			lrd->log.updatemap.nxd = cpu_to_le16(1);
2140 			lrd->log.updatemap.pxd = pxdlock->pxd;
2141 			lrd->backchain =
2142 			    cpu_to_le32(lmLog(log, tblk, lrd, NULL));
2143 			jfs_info("mapLog: xaddr:0x%lx xlen:0x%x",
2144 				 (ulong) addressPXD(&pxdlock->pxd),
2145 				 lengthPXD(&pxdlock->pxd));
2146 		}
2147 
2148 		/* update bmap */
2149 		tlck->flag |= tlckUPDATEMAP;
2150 	}
2151 }
2152 
2153 /*
2154  *	txEA()
2155  *
2156  * function:	acquire maplock for EA/ACL extents or
2157  *		set COMMIT_INLINE flag;
2158  */
txEA(tid_t tid,struct inode * ip,dxd_t * oldea,dxd_t * newea)2159 void txEA(tid_t tid, struct inode *ip, dxd_t * oldea, dxd_t * newea)
2160 {
2161 	struct tlock *tlck = NULL;
2162 	struct pxd_lock *maplock = NULL, *pxdlock = NULL;
2163 
2164 	/*
2165 	 * format maplock for alloc of new EA extent
2166 	 */
2167 	if (newea) {
2168 		/* Since the newea could be a completely zeroed entry we need to
2169 		 * check for the two flags which indicate we should actually
2170 		 * commit new EA data
2171 		 */
2172 		if (newea->flag & DXD_EXTENT) {
2173 			tlck = txMaplock(tid, ip, tlckMAP);
2174 			maplock = (struct pxd_lock *) & tlck->lock;
2175 			pxdlock = (struct pxd_lock *) maplock;
2176 			pxdlock->flag = mlckALLOCPXD;
2177 			PXDaddress(&pxdlock->pxd, addressDXD(newea));
2178 			PXDlength(&pxdlock->pxd, lengthDXD(newea));
2179 			pxdlock++;
2180 			maplock->index = 1;
2181 		} else if (newea->flag & DXD_INLINE) {
2182 			tlck = NULL;
2183 
2184 			set_cflag(COMMIT_Inlineea, ip);
2185 		}
2186 	}
2187 
2188 	/*
2189 	 * format maplock for free of old EA extent
2190 	 */
2191 	if (!test_cflag(COMMIT_Nolink, ip) && oldea->flag & DXD_EXTENT) {
2192 		if (tlck == NULL) {
2193 			tlck = txMaplock(tid, ip, tlckMAP);
2194 			maplock = (struct pxd_lock *) & tlck->lock;
2195 			pxdlock = (struct pxd_lock *) maplock;
2196 			maplock->index = 0;
2197 		}
2198 		pxdlock->flag = mlckFREEPXD;
2199 		PXDaddress(&pxdlock->pxd, addressDXD(oldea));
2200 		PXDlength(&pxdlock->pxd, lengthDXD(oldea));
2201 		maplock->index++;
2202 	}
2203 }
2204 
2205 /*
2206  *	txForce()
2207  *
2208  * function: synchronously write pages locked by transaction
2209  *	     after txLog() but before txUpdateMap();
2210  */
txForce(struct tblock * tblk)2211 static void txForce(struct tblock * tblk)
2212 {
2213 	struct tlock *tlck;
2214 	lid_t lid, next;
2215 	struct metapage *mp;
2216 
2217 	/*
2218 	 * reverse the order of transaction tlocks in
2219 	 * careful update order of address index pages
2220 	 * (right to left, bottom up)
2221 	 */
2222 	tlck = lid_to_tlock(tblk->next);
2223 	lid = tlck->next;
2224 	tlck->next = 0;
2225 	while (lid) {
2226 		tlck = lid_to_tlock(lid);
2227 		next = tlck->next;
2228 		tlck->next = tblk->next;
2229 		tblk->next = lid;
2230 		lid = next;
2231 	}
2232 
2233 	/*
2234 	 * synchronously write the page, and
2235 	 * hold the page for txUpdateMap();
2236 	 */
2237 	for (lid = tblk->next; lid; lid = next) {
2238 		tlck = lid_to_tlock(lid);
2239 		next = tlck->next;
2240 
2241 		if ((mp = tlck->mp) != NULL &&
2242 		    (tlck->type & tlckBTROOT) == 0) {
2243 			assert(mp->xflag & COMMIT_PAGE);
2244 
2245 			if (tlck->flag & tlckWRITEPAGE) {
2246 				tlck->flag &= ~tlckWRITEPAGE;
2247 
2248 				/* do not release page to freelist */
2249 				force_metapage(mp);
2250 #if 0
2251 				/*
2252 				 * The "right" thing to do here is to
2253 				 * synchronously write the metadata.
2254 				 * With the current implementation this
2255 				 * is hard since write_metapage requires
2256 				 * us to kunmap & remap the page.  If we
2257 				 * have tlocks pointing into the metadata
2258 				 * pages, we don't want to do this.  I think
2259 				 * we can get by with synchronously writing
2260 				 * the pages when they are released.
2261 				 */
2262 				assert(mp->nohomeok);
2263 				set_bit(META_dirty, &mp->flag);
2264 				set_bit(META_sync, &mp->flag);
2265 #endif
2266 			}
2267 		}
2268 	}
2269 }
2270 
2271 /*
2272  *	txUpdateMap()
2273  *
2274  * function:	update persistent allocation map (and working map
2275  *		if appropriate);
2276  *
2277  * parameter:
2278  */
txUpdateMap(struct tblock * tblk)2279 static void txUpdateMap(struct tblock * tblk)
2280 {
2281 	struct inode *ip;
2282 	struct inode *ipimap;
2283 	lid_t lid;
2284 	struct tlock *tlck;
2285 	struct maplock *maplock;
2286 	struct pxd_lock pxdlock;
2287 	int maptype;
2288 	int k, nlock;
2289 	struct metapage *mp = NULL;
2290 
2291 	ipimap = JFS_SBI(tblk->sb)->ipimap;
2292 
2293 	maptype = (tblk->xflag & COMMIT_PMAP) ? COMMIT_PMAP : COMMIT_PWMAP;
2294 
2295 
2296 	/*
2297 	 *	update block allocation map
2298 	 *
2299 	 * update allocation state in pmap (and wmap) and
2300 	 * update lsn of the pmap page;
2301 	 */
2302 	/*
2303 	 * scan each tlock/page of transaction for block allocation/free:
2304 	 *
2305 	 * for each tlock/page of transaction, update map.
2306 	 *  ? are there tlock for pmap and pwmap at the same time ?
2307 	 */
2308 	for (lid = tblk->next; lid; lid = tlck->next) {
2309 		tlck = lid_to_tlock(lid);
2310 
2311 		if ((tlck->flag & tlckUPDATEMAP) == 0)
2312 			continue;
2313 
2314 		if (tlck->flag & tlckFREEPAGE) {
2315 			/*
2316 			 * Another thread may attempt to reuse freed space
2317 			 * immediately, so we want to get rid of the metapage
2318 			 * before anyone else has a chance to get it.
2319 			 * Lock metapage, update maps, then invalidate
2320 			 * the metapage.
2321 			 */
2322 			mp = tlck->mp;
2323 			ASSERT(mp->xflag & COMMIT_PAGE);
2324 			grab_metapage(mp);
2325 		}
2326 
2327 		/*
2328 		 * extent list:
2329 		 * . in-line PXD list:
2330 		 * . out-of-line XAD list:
2331 		 */
2332 		maplock = (struct maplock *) & tlck->lock;
2333 		nlock = maplock->index;
2334 
2335 		for (k = 0; k < nlock; k++, maplock++) {
2336 			/*
2337 			 * allocate blocks in persistent map:
2338 			 *
2339 			 * blocks have been allocated from wmap at alloc time;
2340 			 */
2341 			if (maplock->flag & mlckALLOC) {
2342 				txAllocPMap(ipimap, maplock, tblk);
2343 			}
2344 			/*
2345 			 * free blocks in persistent and working map:
2346 			 * blocks will be freed in pmap and then in wmap;
2347 			 *
2348 			 * ? tblock specifies the PMAP/PWMAP based upon
2349 			 * transaction
2350 			 *
2351 			 * free blocks in persistent map:
2352 			 * blocks will be freed from wmap at last reference
2353 			 * release of the object for regular files;
2354 			 *
2355 			 * Alway free blocks from both persistent & working
2356 			 * maps for directories
2357 			 */
2358 			else {	/* (maplock->flag & mlckFREE) */
2359 
2360 				if (tlck->flag & tlckDIRECTORY)
2361 					txFreeMap(ipimap, maplock,
2362 						  tblk, COMMIT_PWMAP);
2363 				else
2364 					txFreeMap(ipimap, maplock,
2365 						  tblk, maptype);
2366 			}
2367 		}
2368 		if (tlck->flag & tlckFREEPAGE) {
2369 			if (!(tblk->flag & tblkGC_LAZY)) {
2370 				/* This is equivalent to txRelease */
2371 				ASSERT(mp->lid == lid);
2372 				tlck->mp->lid = 0;
2373 			}
2374 			assert(mp->nohomeok == 1);
2375 			metapage_homeok(mp);
2376 			discard_metapage(mp);
2377 			tlck->mp = NULL;
2378 		}
2379 	}
2380 	/*
2381 	 *	update inode allocation map
2382 	 *
2383 	 * update allocation state in pmap and
2384 	 * update lsn of the pmap page;
2385 	 * update in-memory inode flag/state
2386 	 *
2387 	 * unlock mapper/write lock
2388 	 */
2389 	if (tblk->xflag & COMMIT_CREATE) {
2390 		diUpdatePMap(ipimap, tblk->ino, false, tblk);
2391 		/* update persistent block allocation map
2392 		 * for the allocation of inode extent;
2393 		 */
2394 		pxdlock.flag = mlckALLOCPXD;
2395 		pxdlock.pxd = tblk->u.ixpxd;
2396 		pxdlock.index = 1;
2397 		txAllocPMap(ipimap, (struct maplock *) & pxdlock, tblk);
2398 	} else if (tblk->xflag & COMMIT_DELETE) {
2399 		ip = tblk->u.ip;
2400 		diUpdatePMap(ipimap, ip->i_ino, true, tblk);
2401 		iput(ip);
2402 	}
2403 }
2404 
2405 /*
2406  *	txAllocPMap()
2407  *
2408  * function: allocate from persistent map;
2409  *
2410  * parameter:
2411  *	ipbmap	-
2412  *	malock	-
2413  *		xad list:
2414  *		pxd:
2415  *
2416  *	maptype -
2417  *		allocate from persistent map;
2418  *		free from persistent map;
2419  *		(e.g., tmp file - free from working map at releae
2420  *		 of last reference);
2421  *		free from persistent and working map;
2422  *
2423  *	lsn	- log sequence number;
2424  */
txAllocPMap(struct inode * ip,struct maplock * maplock,struct tblock * tblk)2425 static void txAllocPMap(struct inode *ip, struct maplock * maplock,
2426 			struct tblock * tblk)
2427 {
2428 	struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
2429 	struct xdlistlock *xadlistlock;
2430 	xad_t *xad;
2431 	s64 xaddr;
2432 	int xlen;
2433 	struct pxd_lock *pxdlock;
2434 	struct xdlistlock *pxdlistlock;
2435 	pxd_t *pxd;
2436 	int n;
2437 
2438 	/*
2439 	 * allocate from persistent map;
2440 	 */
2441 	if (maplock->flag & mlckALLOCXADLIST) {
2442 		xadlistlock = (struct xdlistlock *) maplock;
2443 		xad = xadlistlock->xdlist;
2444 		for (n = 0; n < xadlistlock->count; n++, xad++) {
2445 			if (xad->flag & (XAD_NEW | XAD_EXTENDED)) {
2446 				xaddr = addressXAD(xad);
2447 				xlen = lengthXAD(xad);
2448 				dbUpdatePMap(ipbmap, false, xaddr,
2449 					     (s64) xlen, tblk);
2450 				xad->flag &= ~(XAD_NEW | XAD_EXTENDED);
2451 				jfs_info("allocPMap: xaddr:0x%lx xlen:%d",
2452 					 (ulong) xaddr, xlen);
2453 			}
2454 		}
2455 	} else if (maplock->flag & mlckALLOCPXD) {
2456 		pxdlock = (struct pxd_lock *) maplock;
2457 		xaddr = addressPXD(&pxdlock->pxd);
2458 		xlen = lengthPXD(&pxdlock->pxd);
2459 		dbUpdatePMap(ipbmap, false, xaddr, (s64) xlen, tblk);
2460 		jfs_info("allocPMap: xaddr:0x%lx xlen:%d", (ulong) xaddr, xlen);
2461 	} else {		/* (maplock->flag & mlckALLOCPXDLIST) */
2462 
2463 		pxdlistlock = (struct xdlistlock *) maplock;
2464 		pxd = pxdlistlock->xdlist;
2465 		for (n = 0; n < pxdlistlock->count; n++, pxd++) {
2466 			xaddr = addressPXD(pxd);
2467 			xlen = lengthPXD(pxd);
2468 			dbUpdatePMap(ipbmap, false, xaddr, (s64) xlen,
2469 				     tblk);
2470 			jfs_info("allocPMap: xaddr:0x%lx xlen:%d",
2471 				 (ulong) xaddr, xlen);
2472 		}
2473 	}
2474 }
2475 
2476 /*
2477  *	txFreeMap()
2478  *
2479  * function:	free from persistent and/or working map;
2480  *
2481  * todo: optimization
2482  */
txFreeMap(struct inode * ip,struct maplock * maplock,struct tblock * tblk,int maptype)2483 void txFreeMap(struct inode *ip,
2484 	       struct maplock * maplock, struct tblock * tblk, int maptype)
2485 {
2486 	struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
2487 	struct xdlistlock *xadlistlock;
2488 	xad_t *xad;
2489 	s64 xaddr;
2490 	int xlen;
2491 	struct pxd_lock *pxdlock;
2492 	struct xdlistlock *pxdlistlock;
2493 	pxd_t *pxd;
2494 	int n;
2495 
2496 	jfs_info("txFreeMap: tblk:0x%p maplock:0x%p maptype:0x%x",
2497 		 tblk, maplock, maptype);
2498 
2499 	/*
2500 	 * free from persistent map;
2501 	 */
2502 	if (maptype == COMMIT_PMAP || maptype == COMMIT_PWMAP) {
2503 		if (maplock->flag & mlckFREEXADLIST) {
2504 			xadlistlock = (struct xdlistlock *) maplock;
2505 			xad = xadlistlock->xdlist;
2506 			for (n = 0; n < xadlistlock->count; n++, xad++) {
2507 				if (!(xad->flag & XAD_NEW)) {
2508 					xaddr = addressXAD(xad);
2509 					xlen = lengthXAD(xad);
2510 					dbUpdatePMap(ipbmap, true, xaddr,
2511 						     (s64) xlen, tblk);
2512 					jfs_info("freePMap: xaddr:0x%lx xlen:%d",
2513 						 (ulong) xaddr, xlen);
2514 				}
2515 			}
2516 		} else if (maplock->flag & mlckFREEPXD) {
2517 			pxdlock = (struct pxd_lock *) maplock;
2518 			xaddr = addressPXD(&pxdlock->pxd);
2519 			xlen = lengthPXD(&pxdlock->pxd);
2520 			dbUpdatePMap(ipbmap, true, xaddr, (s64) xlen,
2521 				     tblk);
2522 			jfs_info("freePMap: xaddr:0x%lx xlen:%d",
2523 				 (ulong) xaddr, xlen);
2524 		} else {	/* (maplock->flag & mlckALLOCPXDLIST) */
2525 
2526 			pxdlistlock = (struct xdlistlock *) maplock;
2527 			pxd = pxdlistlock->xdlist;
2528 			for (n = 0; n < pxdlistlock->count; n++, pxd++) {
2529 				xaddr = addressPXD(pxd);
2530 				xlen = lengthPXD(pxd);
2531 				dbUpdatePMap(ipbmap, true, xaddr,
2532 					     (s64) xlen, tblk);
2533 				jfs_info("freePMap: xaddr:0x%lx xlen:%d",
2534 					 (ulong) xaddr, xlen);
2535 			}
2536 		}
2537 	}
2538 
2539 	/*
2540 	 * free from working map;
2541 	 */
2542 	if (maptype == COMMIT_PWMAP || maptype == COMMIT_WMAP) {
2543 		if (maplock->flag & mlckFREEXADLIST) {
2544 			xadlistlock = (struct xdlistlock *) maplock;
2545 			xad = xadlistlock->xdlist;
2546 			for (n = 0; n < xadlistlock->count; n++, xad++) {
2547 				xaddr = addressXAD(xad);
2548 				xlen = lengthXAD(xad);
2549 				dbFree(ip, xaddr, (s64) xlen);
2550 				xad->flag = 0;
2551 				jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
2552 					 (ulong) xaddr, xlen);
2553 			}
2554 		} else if (maplock->flag & mlckFREEPXD) {
2555 			pxdlock = (struct pxd_lock *) maplock;
2556 			xaddr = addressPXD(&pxdlock->pxd);
2557 			xlen = lengthPXD(&pxdlock->pxd);
2558 			dbFree(ip, xaddr, (s64) xlen);
2559 			jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
2560 				 (ulong) xaddr, xlen);
2561 		} else {	/* (maplock->flag & mlckFREEPXDLIST) */
2562 
2563 			pxdlistlock = (struct xdlistlock *) maplock;
2564 			pxd = pxdlistlock->xdlist;
2565 			for (n = 0; n < pxdlistlock->count; n++, pxd++) {
2566 				xaddr = addressPXD(pxd);
2567 				xlen = lengthPXD(pxd);
2568 				dbFree(ip, xaddr, (s64) xlen);
2569 				jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
2570 					 (ulong) xaddr, xlen);
2571 			}
2572 		}
2573 	}
2574 }
2575 
2576 /*
2577  *	txFreelock()
2578  *
2579  * function:	remove tlock from inode anonymous locklist
2580  */
txFreelock(struct inode * ip)2581 void txFreelock(struct inode *ip)
2582 {
2583 	struct jfs_inode_info *jfs_ip = JFS_IP(ip);
2584 	struct tlock *xtlck, *tlck;
2585 	lid_t xlid = 0, lid;
2586 
2587 	if (!jfs_ip->atlhead)
2588 		return;
2589 
2590 	TXN_LOCK();
2591 	xtlck = (struct tlock *) &jfs_ip->atlhead;
2592 
2593 	while ((lid = xtlck->next) != 0) {
2594 		tlck = lid_to_tlock(lid);
2595 		if (tlck->flag & tlckFREELOCK) {
2596 			xtlck->next = tlck->next;
2597 			txLockFree(lid);
2598 		} else {
2599 			xtlck = tlck;
2600 			xlid = lid;
2601 		}
2602 	}
2603 
2604 	if (jfs_ip->atlhead)
2605 		jfs_ip->atltail = xlid;
2606 	else {
2607 		jfs_ip->atltail = 0;
2608 		/*
2609 		 * If inode was on anon_list, remove it
2610 		 */
2611 		list_del_init(&jfs_ip->anon_inode_list);
2612 	}
2613 	TXN_UNLOCK();
2614 }
2615 
2616 /*
2617  *	txAbort()
2618  *
2619  * function: abort tx before commit;
2620  *
2621  * frees line-locks and segment locks for all
2622  * segments in comdata structure.
2623  * Optionally sets state of file-system to FM_DIRTY in super-block.
2624  * log age of page-frames in memory for which caller has
2625  * are reset to 0 (to avoid logwarap).
2626  */
txAbort(tid_t tid,int dirty)2627 void txAbort(tid_t tid, int dirty)
2628 {
2629 	lid_t lid, next;
2630 	struct metapage *mp;
2631 	struct tblock *tblk = tid_to_tblock(tid);
2632 	struct tlock *tlck;
2633 
2634 	/*
2635 	 * free tlocks of the transaction
2636 	 */
2637 	for (lid = tblk->next; lid; lid = next) {
2638 		tlck = lid_to_tlock(lid);
2639 		next = tlck->next;
2640 		mp = tlck->mp;
2641 		JFS_IP(tlck->ip)->xtlid = 0;
2642 
2643 		if (mp) {
2644 			mp->lid = 0;
2645 
2646 			/*
2647 			 * reset lsn of page to avoid logwarap:
2648 			 *
2649 			 * (page may have been previously committed by another
2650 			 * transaction(s) but has not been paged, i.e.,
2651 			 * it may be on logsync list even though it has not
2652 			 * been logged for the current tx.)
2653 			 */
2654 			if (mp->xflag & COMMIT_PAGE && mp->lsn)
2655 				LogSyncRelease(mp);
2656 		}
2657 		/* insert tlock at head of freelist */
2658 		TXN_LOCK();
2659 		txLockFree(lid);
2660 		TXN_UNLOCK();
2661 	}
2662 
2663 	/* caller will free the transaction block */
2664 
2665 	tblk->next = tblk->last = 0;
2666 
2667 	/*
2668 	 * mark filesystem dirty
2669 	 */
2670 	if (dirty)
2671 		jfs_error(tblk->sb, "\n");
2672 
2673 	return;
2674 }
2675 
2676 /*
2677  *	txLazyCommit(void)
2678  *
2679  *	All transactions except those changing ipimap (COMMIT_FORCE) are
2680  *	processed by this routine.  This insures that the inode and block
2681  *	allocation maps are updated in order.  For synchronous transactions,
2682  *	let the user thread finish processing after txUpdateMap() is called.
2683  */
txLazyCommit(struct tblock * tblk)2684 static void txLazyCommit(struct tblock * tblk)
2685 {
2686 	struct jfs_log *log;
2687 
2688 	while (((tblk->flag & tblkGC_READY) == 0) &&
2689 	       ((tblk->flag & tblkGC_UNLOCKED) == 0)) {
2690 		/* We must have gotten ahead of the user thread
2691 		 */
2692 		jfs_info("jfs_lazycommit: tblk 0x%p not unlocked", tblk);
2693 		yield();
2694 	}
2695 
2696 	jfs_info("txLazyCommit: processing tblk 0x%p", tblk);
2697 
2698 	txUpdateMap(tblk);
2699 
2700 	log = (struct jfs_log *) JFS_SBI(tblk->sb)->log;
2701 
2702 	spin_lock_irq(&log->gclock);	// LOGGC_LOCK
2703 
2704 	tblk->flag |= tblkGC_COMMITTED;
2705 
2706 	if (tblk->flag & tblkGC_READY)
2707 		log->gcrtc--;
2708 
2709 	wake_up_all(&tblk->gcwait);	// LOGGC_WAKEUP
2710 
2711 	/*
2712 	 * Can't release log->gclock until we've tested tblk->flag
2713 	 */
2714 	if (tblk->flag & tblkGC_LAZY) {
2715 		spin_unlock_irq(&log->gclock);	// LOGGC_UNLOCK
2716 		txUnlock(tblk);
2717 		tblk->flag &= ~tblkGC_LAZY;
2718 		txEnd(tblk - TxBlock);	/* Convert back to tid */
2719 	} else
2720 		spin_unlock_irq(&log->gclock);	// LOGGC_UNLOCK
2721 
2722 	jfs_info("txLazyCommit: done: tblk = 0x%p", tblk);
2723 }
2724 
2725 /*
2726  *	jfs_lazycommit(void)
2727  *
2728  *	To be run as a kernel daemon.  If lbmIODone is called in an interrupt
2729  *	context, or where blocking is not wanted, this routine will process
2730  *	committed transactions from the unlock queue.
2731  */
jfs_lazycommit(void * arg)2732 int jfs_lazycommit(void *arg)
2733 {
2734 	int WorkDone;
2735 	struct tblock *tblk;
2736 	unsigned long flags;
2737 	struct jfs_sb_info *sbi;
2738 
2739 	do {
2740 		LAZY_LOCK(flags);
2741 		jfs_commit_thread_waking = 0;	/* OK to wake another thread */
2742 		while (!list_empty(&TxAnchor.unlock_queue)) {
2743 			WorkDone = 0;
2744 			list_for_each_entry(tblk, &TxAnchor.unlock_queue,
2745 					    cqueue) {
2746 
2747 				sbi = JFS_SBI(tblk->sb);
2748 				/*
2749 				 * For each volume, the transactions must be
2750 				 * handled in order.  If another commit thread
2751 				 * is handling a tblk for this superblock,
2752 				 * skip it
2753 				 */
2754 				if (sbi->commit_state & IN_LAZYCOMMIT)
2755 					continue;
2756 
2757 				sbi->commit_state |= IN_LAZYCOMMIT;
2758 				WorkDone = 1;
2759 
2760 				/*
2761 				 * Remove transaction from queue
2762 				 */
2763 				list_del(&tblk->cqueue);
2764 
2765 				LAZY_UNLOCK(flags);
2766 				txLazyCommit(tblk);
2767 				LAZY_LOCK(flags);
2768 
2769 				sbi->commit_state &= ~IN_LAZYCOMMIT;
2770 				/*
2771 				 * Don't continue in the for loop.  (We can't
2772 				 * anyway, it's unsafe!)  We want to go back to
2773 				 * the beginning of the list.
2774 				 */
2775 				break;
2776 			}
2777 
2778 			/* If there was nothing to do, don't continue */
2779 			if (!WorkDone)
2780 				break;
2781 		}
2782 		/* In case a wakeup came while all threads were active */
2783 		jfs_commit_thread_waking = 0;
2784 
2785 		if (freezing(current)) {
2786 			LAZY_UNLOCK(flags);
2787 			try_to_freeze();
2788 		} else {
2789 			DECLARE_WAITQUEUE(wq, current);
2790 
2791 			add_wait_queue(&jfs_commit_thread_wait, &wq);
2792 			set_current_state(TASK_INTERRUPTIBLE);
2793 			LAZY_UNLOCK(flags);
2794 			schedule();
2795 			remove_wait_queue(&jfs_commit_thread_wait, &wq);
2796 		}
2797 	} while (!kthread_should_stop());
2798 
2799 	if (!list_empty(&TxAnchor.unlock_queue))
2800 		jfs_err("jfs_lazycommit being killed w/pending transactions!");
2801 	else
2802 		jfs_info("jfs_lazycommit being killed");
2803 	return 0;
2804 }
2805 
txLazyUnlock(struct tblock * tblk)2806 void txLazyUnlock(struct tblock * tblk)
2807 {
2808 	unsigned long flags;
2809 
2810 	LAZY_LOCK(flags);
2811 
2812 	list_add_tail(&tblk->cqueue, &TxAnchor.unlock_queue);
2813 	/*
2814 	 * Don't wake up a commit thread if there is already one servicing
2815 	 * this superblock, or if the last one we woke up hasn't started yet.
2816 	 */
2817 	if (!(JFS_SBI(tblk->sb)->commit_state & IN_LAZYCOMMIT) &&
2818 	    !jfs_commit_thread_waking) {
2819 		jfs_commit_thread_waking = 1;
2820 		wake_up(&jfs_commit_thread_wait);
2821 	}
2822 	LAZY_UNLOCK(flags);
2823 }
2824 
LogSyncRelease(struct metapage * mp)2825 static void LogSyncRelease(struct metapage * mp)
2826 {
2827 	struct jfs_log *log = mp->log;
2828 
2829 	assert(mp->nohomeok);
2830 	assert(log);
2831 	metapage_homeok(mp);
2832 }
2833 
2834 /*
2835  *	txQuiesce
2836  *
2837  *	Block all new transactions and push anonymous transactions to
2838  *	completion
2839  *
2840  *	This does almost the same thing as jfs_sync below.  We don't
2841  *	worry about deadlocking when jfs_tlocks_low is set, since we would
2842  *	expect jfs_sync to get us out of that jam.
2843  */
txQuiesce(struct super_block * sb)2844 void txQuiesce(struct super_block *sb)
2845 {
2846 	struct inode *ip;
2847 	struct jfs_inode_info *jfs_ip;
2848 	struct jfs_log *log = JFS_SBI(sb)->log;
2849 	tid_t tid;
2850 
2851 	set_bit(log_QUIESCE, &log->flag);
2852 
2853 	TXN_LOCK();
2854 restart:
2855 	while (!list_empty(&TxAnchor.anon_list)) {
2856 		jfs_ip = list_entry(TxAnchor.anon_list.next,
2857 				    struct jfs_inode_info,
2858 				    anon_inode_list);
2859 		ip = &jfs_ip->vfs_inode;
2860 
2861 		/*
2862 		 * inode will be removed from anonymous list
2863 		 * when it is committed
2864 		 */
2865 		TXN_UNLOCK();
2866 		tid = txBegin(ip->i_sb, COMMIT_INODE | COMMIT_FORCE);
2867 		mutex_lock(&jfs_ip->commit_mutex);
2868 		txCommit(tid, 1, &ip, 0);
2869 		txEnd(tid);
2870 		mutex_unlock(&jfs_ip->commit_mutex);
2871 		/*
2872 		 * Just to be safe.  I don't know how
2873 		 * long we can run without blocking
2874 		 */
2875 		cond_resched();
2876 		TXN_LOCK();
2877 	}
2878 
2879 	/*
2880 	 * If jfs_sync is running in parallel, there could be some inodes
2881 	 * on anon_list2.  Let's check.
2882 	 */
2883 	if (!list_empty(&TxAnchor.anon_list2)) {
2884 		list_splice_init(&TxAnchor.anon_list2, &TxAnchor.anon_list);
2885 		goto restart;
2886 	}
2887 	TXN_UNLOCK();
2888 
2889 	/*
2890 	 * We may need to kick off the group commit
2891 	 */
2892 	jfs_flush_journal(log, 0);
2893 }
2894 
2895 /*
2896  * txResume()
2897  *
2898  * Allows transactions to start again following txQuiesce
2899  */
txResume(struct super_block * sb)2900 void txResume(struct super_block *sb)
2901 {
2902 	struct jfs_log *log = JFS_SBI(sb)->log;
2903 
2904 	clear_bit(log_QUIESCE, &log->flag);
2905 	TXN_WAKEUP(&log->syncwait);
2906 }
2907 
2908 /*
2909  *	jfs_sync(void)
2910  *
2911  *	To be run as a kernel daemon.  This is awakened when tlocks run low.
2912  *	We write any inodes that have anonymous tlocks so they will become
2913  *	available.
2914  */
jfs_sync(void * arg)2915 int jfs_sync(void *arg)
2916 {
2917 	struct inode *ip;
2918 	struct jfs_inode_info *jfs_ip;
2919 	tid_t tid;
2920 
2921 	do {
2922 		/*
2923 		 * write each inode on the anonymous inode list
2924 		 */
2925 		TXN_LOCK();
2926 		while (jfs_tlocks_low && !list_empty(&TxAnchor.anon_list)) {
2927 			jfs_ip = list_entry(TxAnchor.anon_list.next,
2928 					    struct jfs_inode_info,
2929 					    anon_inode_list);
2930 			ip = &jfs_ip->vfs_inode;
2931 
2932 			if (! igrab(ip)) {
2933 				/*
2934 				 * Inode is being freed
2935 				 */
2936 				list_del_init(&jfs_ip->anon_inode_list);
2937 			} else if (mutex_trylock(&jfs_ip->commit_mutex)) {
2938 				/*
2939 				 * inode will be removed from anonymous list
2940 				 * when it is committed
2941 				 */
2942 				TXN_UNLOCK();
2943 				tid = txBegin(ip->i_sb, COMMIT_INODE);
2944 				txCommit(tid, 1, &ip, 0);
2945 				txEnd(tid);
2946 				mutex_unlock(&jfs_ip->commit_mutex);
2947 
2948 				iput(ip);
2949 				/*
2950 				 * Just to be safe.  I don't know how
2951 				 * long we can run without blocking
2952 				 */
2953 				cond_resched();
2954 				TXN_LOCK();
2955 			} else {
2956 				/* We can't get the commit mutex.  It may
2957 				 * be held by a thread waiting for tlock's
2958 				 * so let's not block here.  Save it to
2959 				 * put back on the anon_list.
2960 				 */
2961 
2962 				/* Move from anon_list to anon_list2 */
2963 				list_move(&jfs_ip->anon_inode_list,
2964 					  &TxAnchor.anon_list2);
2965 
2966 				TXN_UNLOCK();
2967 				iput(ip);
2968 				TXN_LOCK();
2969 			}
2970 		}
2971 		/* Add anon_list2 back to anon_list */
2972 		list_splice_init(&TxAnchor.anon_list2, &TxAnchor.anon_list);
2973 
2974 		if (freezing(current)) {
2975 			TXN_UNLOCK();
2976 			try_to_freeze();
2977 		} else {
2978 			set_current_state(TASK_INTERRUPTIBLE);
2979 			TXN_UNLOCK();
2980 			schedule();
2981 		}
2982 	} while (!kthread_should_stop());
2983 
2984 	jfs_info("jfs_sync being killed");
2985 	return 0;
2986 }
2987 
2988 #if defined(CONFIG_PROC_FS) && defined(CONFIG_JFS_DEBUG)
jfs_txanchor_proc_show(struct seq_file * m,void * v)2989 int jfs_txanchor_proc_show(struct seq_file *m, void *v)
2990 {
2991 	char *freewait;
2992 	char *freelockwait;
2993 	char *lowlockwait;
2994 
2995 	freewait =
2996 	    waitqueue_active(&TxAnchor.freewait) ? "active" : "empty";
2997 	freelockwait =
2998 	    waitqueue_active(&TxAnchor.freelockwait) ? "active" : "empty";
2999 	lowlockwait =
3000 	    waitqueue_active(&TxAnchor.lowlockwait) ? "active" : "empty";
3001 
3002 	seq_printf(m,
3003 		       "JFS TxAnchor\n"
3004 		       "============\n"
3005 		       "freetid = %d\n"
3006 		       "freewait = %s\n"
3007 		       "freelock = %d\n"
3008 		       "freelockwait = %s\n"
3009 		       "lowlockwait = %s\n"
3010 		       "tlocksInUse = %d\n"
3011 		       "jfs_tlocks_low = %d\n"
3012 		       "unlock_queue is %sempty\n",
3013 		       TxAnchor.freetid,
3014 		       freewait,
3015 		       TxAnchor.freelock,
3016 		       freelockwait,
3017 		       lowlockwait,
3018 		       TxAnchor.tlocksInUse,
3019 		       jfs_tlocks_low,
3020 		       list_empty(&TxAnchor.unlock_queue) ? "" : "not ");
3021 	return 0;
3022 }
3023 #endif
3024 
3025 #if defined(CONFIG_PROC_FS) && defined(CONFIG_JFS_STATISTICS)
jfs_txstats_proc_show(struct seq_file * m,void * v)3026 int jfs_txstats_proc_show(struct seq_file *m, void *v)
3027 {
3028 	seq_printf(m,
3029 		       "JFS TxStats\n"
3030 		       "===========\n"
3031 		       "calls to txBegin = %d\n"
3032 		       "txBegin blocked by sync barrier = %d\n"
3033 		       "txBegin blocked by tlocks low = %d\n"
3034 		       "txBegin blocked by no free tid = %d\n"
3035 		       "calls to txBeginAnon = %d\n"
3036 		       "txBeginAnon blocked by sync barrier = %d\n"
3037 		       "txBeginAnon blocked by tlocks low = %d\n"
3038 		       "calls to txLockAlloc = %d\n"
3039 		       "tLockAlloc blocked by no free lock = %d\n",
3040 		       TxStat.txBegin,
3041 		       TxStat.txBegin_barrier,
3042 		       TxStat.txBegin_lockslow,
3043 		       TxStat.txBegin_freetid,
3044 		       TxStat.txBeginAnon,
3045 		       TxStat.txBeginAnon_barrier,
3046 		       TxStat.txBeginAnon_lockslow,
3047 		       TxStat.txLockAlloc,
3048 		       TxStat.txLockAlloc_freelock);
3049 	return 0;
3050 }
3051 #endif
3052