xref: /openbmc/linux/fs/jfs/jfs_txnmgr.c (revision 2ff51719)
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 = (xtpage_t *) &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 	return;
1499 }
1500 
1501 /*
1502  *	dataLog()
1503  *
1504  * function:	log data tlock
1505  */
dataLog(struct jfs_log * log,struct tblock * tblk,struct lrd * lrd,struct tlock * tlck)1506 static void dataLog(struct jfs_log *log, struct tblock *tblk, struct lrd *lrd,
1507 	    struct tlock *tlck)
1508 {
1509 	struct metapage *mp;
1510 	pxd_t *pxd;
1511 
1512 	mp = tlck->mp;
1513 
1514 	/* initialize as REDOPAGE record format */
1515 	lrd->log.redopage.type = cpu_to_le16(LOG_DATA);
1516 	lrd->log.redopage.l2linesize = cpu_to_le16(L2DATASLOTSIZE);
1517 
1518 	pxd = &lrd->log.redopage.pxd;
1519 
1520 	/* log after-image for logredo(): */
1521 	lrd->type = cpu_to_le16(LOG_REDOPAGE);
1522 
1523 	if (jfs_dirtable_inline(tlck->ip)) {
1524 		/*
1525 		 * The table has been truncated, we've must have deleted
1526 		 * the last entry, so don't bother logging this
1527 		 */
1528 		mp->lid = 0;
1529 		grab_metapage(mp);
1530 		metapage_homeok(mp);
1531 		discard_metapage(mp);
1532 		tlck->mp = NULL;
1533 		return;
1534 	}
1535 
1536 	PXDaddress(pxd, mp->index);
1537 	PXDlength(pxd, mp->logical_size >> tblk->sb->s_blocksize_bits);
1538 
1539 	lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1540 
1541 	/* mark page as homeward bound */
1542 	tlck->flag |= tlckWRITEPAGE;
1543 
1544 	return;
1545 }
1546 
1547 /*
1548  *	dtLog()
1549  *
1550  * function:	log dtree tlock and format maplock to update bmap;
1551  */
dtLog(struct jfs_log * log,struct tblock * tblk,struct lrd * lrd,struct tlock * tlck)1552 static void dtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1553 	   struct tlock * tlck)
1554 {
1555 	struct metapage *mp;
1556 	struct pxd_lock *pxdlock;
1557 	pxd_t *pxd;
1558 
1559 	mp = tlck->mp;
1560 
1561 	/* initialize as REDOPAGE/NOREDOPAGE record format */
1562 	lrd->log.redopage.type = cpu_to_le16(LOG_DTREE);
1563 	lrd->log.redopage.l2linesize = cpu_to_le16(L2DTSLOTSIZE);
1564 
1565 	pxd = &lrd->log.redopage.pxd;
1566 
1567 	if (tlck->type & tlckBTROOT)
1568 		lrd->log.redopage.type |= cpu_to_le16(LOG_BTROOT);
1569 
1570 	/*
1571 	 *	page extension via relocation: entry insertion;
1572 	 *	page extension in-place: entry insertion;
1573 	 *	new right page from page split, reinitialized in-line
1574 	 *	root from root page split: entry insertion;
1575 	 */
1576 	if (tlck->type & (tlckNEW | tlckEXTEND)) {
1577 		/* log after-image of the new page for logredo():
1578 		 * mark log (LOG_NEW) for logredo() to initialize
1579 		 * freelist and update bmap for alloc of the new page;
1580 		 */
1581 		lrd->type = cpu_to_le16(LOG_REDOPAGE);
1582 		if (tlck->type & tlckEXTEND)
1583 			lrd->log.redopage.type |= cpu_to_le16(LOG_EXTEND);
1584 		else
1585 			lrd->log.redopage.type |= cpu_to_le16(LOG_NEW);
1586 		PXDaddress(pxd, mp->index);
1587 		PXDlength(pxd,
1588 			  mp->logical_size >> tblk->sb->s_blocksize_bits);
1589 		lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1590 
1591 		/* format a maplock for txUpdateMap() to update bPMAP for
1592 		 * alloc of the new page;
1593 		 */
1594 		if (tlck->type & tlckBTROOT)
1595 			return;
1596 		tlck->flag |= tlckUPDATEMAP;
1597 		pxdlock = (struct pxd_lock *) & tlck->lock;
1598 		pxdlock->flag = mlckALLOCPXD;
1599 		pxdlock->pxd = *pxd;
1600 
1601 		pxdlock->index = 1;
1602 
1603 		/* mark page as homeward bound */
1604 		tlck->flag |= tlckWRITEPAGE;
1605 		return;
1606 	}
1607 
1608 	/*
1609 	 *	entry insertion/deletion,
1610 	 *	sibling page link update (old right page before split);
1611 	 */
1612 	if (tlck->type & (tlckENTRY | tlckRELINK)) {
1613 		/* log after-image for logredo(): */
1614 		lrd->type = cpu_to_le16(LOG_REDOPAGE);
1615 		PXDaddress(pxd, mp->index);
1616 		PXDlength(pxd,
1617 			  mp->logical_size >> tblk->sb->s_blocksize_bits);
1618 		lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1619 
1620 		/* mark page as homeward bound */
1621 		tlck->flag |= tlckWRITEPAGE;
1622 		return;
1623 	}
1624 
1625 	/*
1626 	 *	page deletion: page has been invalidated
1627 	 *	page relocation: source extent
1628 	 *
1629 	 *	a maplock for free of the page has been formatted
1630 	 *	at txLock() time);
1631 	 */
1632 	if (tlck->type & (tlckFREE | tlckRELOCATE)) {
1633 		/* log LOG_NOREDOPAGE of the deleted page for logredo()
1634 		 * to start NoRedoPage filter and to update bmap for free
1635 		 * of the deletd page
1636 		 */
1637 		lrd->type = cpu_to_le16(LOG_NOREDOPAGE);
1638 		pxdlock = (struct pxd_lock *) & tlck->lock;
1639 		*pxd = pxdlock->pxd;
1640 		lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1641 
1642 		/* a maplock for txUpdateMap() for free of the page
1643 		 * has been formatted at txLock() time;
1644 		 */
1645 		tlck->flag |= tlckUPDATEMAP;
1646 	}
1647 	return;
1648 }
1649 
1650 /*
1651  *	xtLog()
1652  *
1653  * function:	log xtree tlock and format maplock to update bmap;
1654  */
xtLog(struct jfs_log * log,struct tblock * tblk,struct lrd * lrd,struct tlock * tlck)1655 static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1656 	   struct tlock * tlck)
1657 {
1658 	struct inode *ip;
1659 	struct metapage *mp;
1660 	xtpage_t *p;
1661 	struct xtlock *xtlck;
1662 	struct maplock *maplock;
1663 	struct xdlistlock *xadlock;
1664 	struct pxd_lock *pxdlock;
1665 	pxd_t *page_pxd;
1666 	int next, lwm, hwm;
1667 
1668 	ip = tlck->ip;
1669 	mp = tlck->mp;
1670 
1671 	/* initialize as REDOPAGE/NOREDOPAGE record format */
1672 	lrd->log.redopage.type = cpu_to_le16(LOG_XTREE);
1673 	lrd->log.redopage.l2linesize = cpu_to_le16(L2XTSLOTSIZE);
1674 
1675 	page_pxd = &lrd->log.redopage.pxd;
1676 
1677 	if (tlck->type & tlckBTROOT) {
1678 		lrd->log.redopage.type |= cpu_to_le16(LOG_BTROOT);
1679 		p = (xtpage_t *) &JFS_IP(ip)->i_xtroot;
1680 		if (S_ISDIR(ip->i_mode))
1681 			lrd->log.redopage.type |=
1682 			    cpu_to_le16(LOG_DIR_XTREE);
1683 	} else
1684 		p = (xtpage_t *) mp->data;
1685 	next = le16_to_cpu(p->header.nextindex);
1686 
1687 	xtlck = (struct xtlock *) & tlck->lock;
1688 
1689 	maplock = (struct maplock *) & tlck->lock;
1690 	xadlock = (struct xdlistlock *) maplock;
1691 
1692 	/*
1693 	 *	entry insertion/extension;
1694 	 *	sibling page link update (old right page before split);
1695 	 */
1696 	if (tlck->type & (tlckNEW | tlckGROW | tlckRELINK)) {
1697 		/* log after-image for logredo():
1698 		 * logredo() will update bmap for alloc of new/extended
1699 		 * extents (XAD_NEW|XAD_EXTEND) of XAD[lwm:next) from
1700 		 * after-image of XADlist;
1701 		 * logredo() resets (XAD_NEW|XAD_EXTEND) flag when
1702 		 * applying the after-image to the meta-data page.
1703 		 */
1704 		lrd->type = cpu_to_le16(LOG_REDOPAGE);
1705 		PXDaddress(page_pxd, mp->index);
1706 		PXDlength(page_pxd,
1707 			  mp->logical_size >> tblk->sb->s_blocksize_bits);
1708 		lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1709 
1710 		/* format a maplock for txUpdateMap() to update bPMAP
1711 		 * for alloc of new/extended extents of XAD[lwm:next)
1712 		 * from the page itself;
1713 		 * txUpdateMap() resets (XAD_NEW|XAD_EXTEND) flag.
1714 		 */
1715 		lwm = xtlck->lwm.offset;
1716 		if (lwm == 0)
1717 			lwm = XTPAGEMAXSLOT;
1718 
1719 		if (lwm == next)
1720 			goto out;
1721 		if (lwm > next) {
1722 			jfs_err("xtLog: lwm > next");
1723 			goto out;
1724 		}
1725 		tlck->flag |= tlckUPDATEMAP;
1726 		xadlock->flag = mlckALLOCXADLIST;
1727 		xadlock->count = next - lwm;
1728 		if ((xadlock->count <= 4) && (tblk->xflag & COMMIT_LAZY)) {
1729 			int i;
1730 			pxd_t *pxd;
1731 			/*
1732 			 * Lazy commit may allow xtree to be modified before
1733 			 * txUpdateMap runs.  Copy xad into linelock to
1734 			 * preserve correct data.
1735 			 *
1736 			 * We can fit twice as may pxd's as xads in the lock
1737 			 */
1738 			xadlock->flag = mlckALLOCPXDLIST;
1739 			pxd = xadlock->xdlist = &xtlck->pxdlock;
1740 			for (i = 0; i < xadlock->count; i++) {
1741 				PXDaddress(pxd, addressXAD(&p->xad[lwm + i]));
1742 				PXDlength(pxd, lengthXAD(&p->xad[lwm + i]));
1743 				p->xad[lwm + i].flag &=
1744 				    ~(XAD_NEW | XAD_EXTENDED);
1745 				pxd++;
1746 			}
1747 		} else {
1748 			/*
1749 			 * xdlist will point to into inode's xtree, ensure
1750 			 * that transaction is not committed lazily.
1751 			 */
1752 			xadlock->flag = mlckALLOCXADLIST;
1753 			xadlock->xdlist = &p->xad[lwm];
1754 			tblk->xflag &= ~COMMIT_LAZY;
1755 		}
1756 		jfs_info("xtLog: alloc ip:0x%p mp:0x%p tlck:0x%p lwm:%d count:%d",
1757 			 tlck->ip, mp, tlck, lwm, xadlock->count);
1758 
1759 		maplock->index = 1;
1760 
1761 	      out:
1762 		/* mark page as homeward bound */
1763 		tlck->flag |= tlckWRITEPAGE;
1764 
1765 		return;
1766 	}
1767 
1768 	/*
1769 	 *	page deletion: file deletion/truncation (ref. xtTruncate())
1770 	 *
1771 	 * (page will be invalidated after log is written and bmap
1772 	 * is updated from the page);
1773 	 */
1774 	if (tlck->type & tlckFREE) {
1775 		/* LOG_NOREDOPAGE log for NoRedoPage filter:
1776 		 * if page free from file delete, NoRedoFile filter from
1777 		 * inode image of zero link count will subsume NoRedoPage
1778 		 * filters for each page;
1779 		 * if page free from file truncattion, write NoRedoPage
1780 		 * filter;
1781 		 *
1782 		 * upadte of block allocation map for the page itself:
1783 		 * if page free from deletion and truncation, LOG_UPDATEMAP
1784 		 * log for the page itself is generated from processing
1785 		 * its parent page xad entries;
1786 		 */
1787 		/* if page free from file truncation, log LOG_NOREDOPAGE
1788 		 * of the deleted page for logredo() to start NoRedoPage
1789 		 * filter for the page;
1790 		 */
1791 		if (tblk->xflag & COMMIT_TRUNCATE) {
1792 			/* write NOREDOPAGE for the page */
1793 			lrd->type = cpu_to_le16(LOG_NOREDOPAGE);
1794 			PXDaddress(page_pxd, mp->index);
1795 			PXDlength(page_pxd,
1796 				  mp->logical_size >> tblk->sb->
1797 				  s_blocksize_bits);
1798 			lrd->backchain =
1799 			    cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1800 
1801 			if (tlck->type & tlckBTROOT) {
1802 				/* Empty xtree must be logged */
1803 				lrd->type = cpu_to_le16(LOG_REDOPAGE);
1804 				lrd->backchain =
1805 				    cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1806 			}
1807 		}
1808 
1809 		/* init LOG_UPDATEMAP of the freed extents
1810 		 * XAD[XTENTRYSTART:hwm) from the deleted page itself
1811 		 * for logredo() to update bmap;
1812 		 */
1813 		lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1814 		lrd->log.updatemap.type = cpu_to_le16(LOG_FREEXADLIST);
1815 		xtlck = (struct xtlock *) & tlck->lock;
1816 		hwm = xtlck->hwm.offset;
1817 		lrd->log.updatemap.nxd =
1818 		    cpu_to_le16(hwm - XTENTRYSTART + 1);
1819 		/* reformat linelock for lmLog() */
1820 		xtlck->header.offset = XTENTRYSTART;
1821 		xtlck->header.length = hwm - XTENTRYSTART + 1;
1822 		xtlck->index = 1;
1823 		lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1824 
1825 		/* format a maplock for txUpdateMap() to update bmap
1826 		 * to free extents of XAD[XTENTRYSTART:hwm) from the
1827 		 * deleted page itself;
1828 		 */
1829 		tlck->flag |= tlckUPDATEMAP;
1830 		xadlock->count = hwm - XTENTRYSTART + 1;
1831 		if ((xadlock->count <= 4) && (tblk->xflag & COMMIT_LAZY)) {
1832 			int i;
1833 			pxd_t *pxd;
1834 			/*
1835 			 * Lazy commit may allow xtree to be modified before
1836 			 * txUpdateMap runs.  Copy xad into linelock to
1837 			 * preserve correct data.
1838 			 *
1839 			 * We can fit twice as may pxd's as xads in the lock
1840 			 */
1841 			xadlock->flag = mlckFREEPXDLIST;
1842 			pxd = xadlock->xdlist = &xtlck->pxdlock;
1843 			for (i = 0; i < xadlock->count; i++) {
1844 				PXDaddress(pxd,
1845 					addressXAD(&p->xad[XTENTRYSTART + i]));
1846 				PXDlength(pxd,
1847 					lengthXAD(&p->xad[XTENTRYSTART + i]));
1848 				pxd++;
1849 			}
1850 		} else {
1851 			/*
1852 			 * xdlist will point to into inode's xtree, ensure
1853 			 * that transaction is not committed lazily.
1854 			 */
1855 			xadlock->flag = mlckFREEXADLIST;
1856 			xadlock->xdlist = &p->xad[XTENTRYSTART];
1857 			tblk->xflag &= ~COMMIT_LAZY;
1858 		}
1859 		jfs_info("xtLog: free ip:0x%p mp:0x%p count:%d lwm:2",
1860 			 tlck->ip, mp, xadlock->count);
1861 
1862 		maplock->index = 1;
1863 
1864 		/* mark page as invalid */
1865 		if (((tblk->xflag & COMMIT_PWMAP) || S_ISDIR(ip->i_mode))
1866 		    && !(tlck->type & tlckBTROOT))
1867 			tlck->flag |= tlckFREEPAGE;
1868 		/*
1869 		   else (tblk->xflag & COMMIT_PMAP)
1870 		   ? release the page;
1871 		 */
1872 		return;
1873 	}
1874 
1875 	/*
1876 	 *	page/entry truncation: file truncation (ref. xtTruncate())
1877 	 *
1878 	 *	|----------+------+------+---------------|
1879 	 *		   |      |      |
1880 	 *		   |      |     hwm - hwm before truncation
1881 	 *		   |     next - truncation point
1882 	 *		  lwm - lwm before truncation
1883 	 * header ?
1884 	 */
1885 	if (tlck->type & tlckTRUNCATE) {
1886 		pxd_t pxd;	/* truncated extent of xad */
1887 		int twm;
1888 
1889 		/*
1890 		 * For truncation the entire linelock may be used, so it would
1891 		 * be difficult to store xad list in linelock itself.
1892 		 * Therefore, we'll just force transaction to be committed
1893 		 * synchronously, so that xtree pages won't be changed before
1894 		 * txUpdateMap runs.
1895 		 */
1896 		tblk->xflag &= ~COMMIT_LAZY;
1897 		lwm = xtlck->lwm.offset;
1898 		if (lwm == 0)
1899 			lwm = XTPAGEMAXSLOT;
1900 		hwm = xtlck->hwm.offset;
1901 		twm = xtlck->twm.offset;
1902 
1903 		/*
1904 		 *	write log records
1905 		 */
1906 		/* log after-image for logredo():
1907 		 *
1908 		 * logredo() will update bmap for alloc of new/extended
1909 		 * extents (XAD_NEW|XAD_EXTEND) of XAD[lwm:next) from
1910 		 * after-image of XADlist;
1911 		 * logredo() resets (XAD_NEW|XAD_EXTEND) flag when
1912 		 * applying the after-image to the meta-data page.
1913 		 */
1914 		lrd->type = cpu_to_le16(LOG_REDOPAGE);
1915 		PXDaddress(page_pxd, mp->index);
1916 		PXDlength(page_pxd,
1917 			  mp->logical_size >> tblk->sb->s_blocksize_bits);
1918 		lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1919 
1920 		/*
1921 		 * truncate entry XAD[twm == next - 1]:
1922 		 */
1923 		if (twm == next - 1) {
1924 			/* init LOG_UPDATEMAP for logredo() to update bmap for
1925 			 * free of truncated delta extent of the truncated
1926 			 * entry XAD[next - 1]:
1927 			 * (xtlck->pxdlock = truncated delta extent);
1928 			 */
1929 			pxdlock = (struct pxd_lock *) & xtlck->pxdlock;
1930 			/* assert(pxdlock->type & tlckTRUNCATE); */
1931 			lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1932 			lrd->log.updatemap.type = cpu_to_le16(LOG_FREEPXD);
1933 			lrd->log.updatemap.nxd = cpu_to_le16(1);
1934 			lrd->log.updatemap.pxd = pxdlock->pxd;
1935 			pxd = pxdlock->pxd;	/* save to format maplock */
1936 			lrd->backchain =
1937 			    cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1938 		}
1939 
1940 		/*
1941 		 * free entries XAD[next:hwm]:
1942 		 */
1943 		if (hwm >= next) {
1944 			/* init LOG_UPDATEMAP of the freed extents
1945 			 * XAD[next:hwm] from the deleted page itself
1946 			 * for logredo() to update bmap;
1947 			 */
1948 			lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1949 			lrd->log.updatemap.type =
1950 			    cpu_to_le16(LOG_FREEXADLIST);
1951 			xtlck = (struct xtlock *) & tlck->lock;
1952 			hwm = xtlck->hwm.offset;
1953 			lrd->log.updatemap.nxd =
1954 			    cpu_to_le16(hwm - next + 1);
1955 			/* reformat linelock for lmLog() */
1956 			xtlck->header.offset = next;
1957 			xtlck->header.length = hwm - next + 1;
1958 			xtlck->index = 1;
1959 			lrd->backchain =
1960 			    cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1961 		}
1962 
1963 		/*
1964 		 *	format maplock(s) for txUpdateMap() to update bmap
1965 		 */
1966 		maplock->index = 0;
1967 
1968 		/*
1969 		 * allocate entries XAD[lwm:next):
1970 		 */
1971 		if (lwm < next) {
1972 			/* format a maplock for txUpdateMap() to update bPMAP
1973 			 * for alloc of new/extended extents of XAD[lwm:next)
1974 			 * from the page itself;
1975 			 * txUpdateMap() resets (XAD_NEW|XAD_EXTEND) flag.
1976 			 */
1977 			tlck->flag |= tlckUPDATEMAP;
1978 			xadlock->flag = mlckALLOCXADLIST;
1979 			xadlock->count = next - lwm;
1980 			xadlock->xdlist = &p->xad[lwm];
1981 
1982 			jfs_info("xtLog: alloc ip:0x%p mp:0x%p count:%d lwm:%d next:%d",
1983 				 tlck->ip, mp, xadlock->count, lwm, next);
1984 			maplock->index++;
1985 			xadlock++;
1986 		}
1987 
1988 		/*
1989 		 * truncate entry XAD[twm == next - 1]:
1990 		 */
1991 		if (twm == next - 1) {
1992 			/* format a maplock for txUpdateMap() to update bmap
1993 			 * to free truncated delta extent of the truncated
1994 			 * entry XAD[next - 1];
1995 			 * (xtlck->pxdlock = truncated delta extent);
1996 			 */
1997 			tlck->flag |= tlckUPDATEMAP;
1998 			pxdlock = (struct pxd_lock *) xadlock;
1999 			pxdlock->flag = mlckFREEPXD;
2000 			pxdlock->count = 1;
2001 			pxdlock->pxd = pxd;
2002 
2003 			jfs_info("xtLog: truncate ip:0x%p mp:0x%p count:%d hwm:%d",
2004 				 ip, mp, pxdlock->count, hwm);
2005 			maplock->index++;
2006 			xadlock++;
2007 		}
2008 
2009 		/*
2010 		 * free entries XAD[next:hwm]:
2011 		 */
2012 		if (hwm >= next) {
2013 			/* format a maplock for txUpdateMap() to update bmap
2014 			 * to free extents of XAD[next:hwm] from thedeleted
2015 			 * page itself;
2016 			 */
2017 			tlck->flag |= tlckUPDATEMAP;
2018 			xadlock->flag = mlckFREEXADLIST;
2019 			xadlock->count = hwm - next + 1;
2020 			xadlock->xdlist = &p->xad[next];
2021 
2022 			jfs_info("xtLog: free ip:0x%p mp:0x%p count:%d next:%d hwm:%d",
2023 				 tlck->ip, mp, xadlock->count, next, hwm);
2024 			maplock->index++;
2025 		}
2026 
2027 		/* mark page as homeward bound */
2028 		tlck->flag |= tlckWRITEPAGE;
2029 	}
2030 	return;
2031 }
2032 
2033 /*
2034  *	mapLog()
2035  *
2036  * function:	log from maplock of freed data extents;
2037  */
mapLog(struct jfs_log * log,struct tblock * tblk,struct lrd * lrd,struct tlock * tlck)2038 static void mapLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
2039 		   struct tlock * tlck)
2040 {
2041 	struct pxd_lock *pxdlock;
2042 	int i, nlock;
2043 	pxd_t *pxd;
2044 
2045 	/*
2046 	 *	page relocation: free the source page extent
2047 	 *
2048 	 * a maplock for txUpdateMap() for free of the page
2049 	 * has been formatted at txLock() time saving the src
2050 	 * relocated page address;
2051 	 */
2052 	if (tlck->type & tlckRELOCATE) {
2053 		/* log LOG_NOREDOPAGE of the old relocated page
2054 		 * for logredo() to start NoRedoPage filter;
2055 		 */
2056 		lrd->type = cpu_to_le16(LOG_NOREDOPAGE);
2057 		pxdlock = (struct pxd_lock *) & tlck->lock;
2058 		pxd = &lrd->log.redopage.pxd;
2059 		*pxd = pxdlock->pxd;
2060 		lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
2061 
2062 		/* (N.B. currently, logredo() does NOT update bmap
2063 		 * for free of the page itself for (LOG_XTREE|LOG_NOREDOPAGE);
2064 		 * if page free from relocation, LOG_UPDATEMAP log is
2065 		 * specifically generated now for logredo()
2066 		 * to update bmap for free of src relocated page;
2067 		 * (new flag LOG_RELOCATE may be introduced which will
2068 		 * inform logredo() to start NORedoPage filter and also
2069 		 * update block allocation map at the same time, thus
2070 		 * avoiding an extra log write);
2071 		 */
2072 		lrd->type = cpu_to_le16(LOG_UPDATEMAP);
2073 		lrd->log.updatemap.type = cpu_to_le16(LOG_FREEPXD);
2074 		lrd->log.updatemap.nxd = cpu_to_le16(1);
2075 		lrd->log.updatemap.pxd = pxdlock->pxd;
2076 		lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
2077 
2078 		/* a maplock for txUpdateMap() for free of the page
2079 		 * has been formatted at txLock() time;
2080 		 */
2081 		tlck->flag |= tlckUPDATEMAP;
2082 		return;
2083 	}
2084 	/*
2085 
2086 	 * Otherwise it's not a relocate request
2087 	 *
2088 	 */
2089 	else {
2090 		/* log LOG_UPDATEMAP for logredo() to update bmap for
2091 		 * free of truncated/relocated delta extent of the data;
2092 		 * e.g.: external EA extent, relocated/truncated extent
2093 		 * from xtTailgate();
2094 		 */
2095 		lrd->type = cpu_to_le16(LOG_UPDATEMAP);
2096 		pxdlock = (struct pxd_lock *) & tlck->lock;
2097 		nlock = pxdlock->index;
2098 		for (i = 0; i < nlock; i++, pxdlock++) {
2099 			if (pxdlock->flag & mlckALLOCPXD)
2100 				lrd->log.updatemap.type =
2101 				    cpu_to_le16(LOG_ALLOCPXD);
2102 			else
2103 				lrd->log.updatemap.type =
2104 				    cpu_to_le16(LOG_FREEPXD);
2105 			lrd->log.updatemap.nxd = cpu_to_le16(1);
2106 			lrd->log.updatemap.pxd = pxdlock->pxd;
2107 			lrd->backchain =
2108 			    cpu_to_le32(lmLog(log, tblk, lrd, NULL));
2109 			jfs_info("mapLog: xaddr:0x%lx xlen:0x%x",
2110 				 (ulong) addressPXD(&pxdlock->pxd),
2111 				 lengthPXD(&pxdlock->pxd));
2112 		}
2113 
2114 		/* update bmap */
2115 		tlck->flag |= tlckUPDATEMAP;
2116 	}
2117 }
2118 
2119 /*
2120  *	txEA()
2121  *
2122  * function:	acquire maplock for EA/ACL extents or
2123  *		set COMMIT_INLINE flag;
2124  */
txEA(tid_t tid,struct inode * ip,dxd_t * oldea,dxd_t * newea)2125 void txEA(tid_t tid, struct inode *ip, dxd_t * oldea, dxd_t * newea)
2126 {
2127 	struct tlock *tlck = NULL;
2128 	struct pxd_lock *maplock = NULL, *pxdlock = NULL;
2129 
2130 	/*
2131 	 * format maplock for alloc of new EA extent
2132 	 */
2133 	if (newea) {
2134 		/* Since the newea could be a completely zeroed entry we need to
2135 		 * check for the two flags which indicate we should actually
2136 		 * commit new EA data
2137 		 */
2138 		if (newea->flag & DXD_EXTENT) {
2139 			tlck = txMaplock(tid, ip, tlckMAP);
2140 			maplock = (struct pxd_lock *) & tlck->lock;
2141 			pxdlock = (struct pxd_lock *) maplock;
2142 			pxdlock->flag = mlckALLOCPXD;
2143 			PXDaddress(&pxdlock->pxd, addressDXD(newea));
2144 			PXDlength(&pxdlock->pxd, lengthDXD(newea));
2145 			pxdlock++;
2146 			maplock->index = 1;
2147 		} else if (newea->flag & DXD_INLINE) {
2148 			tlck = NULL;
2149 
2150 			set_cflag(COMMIT_Inlineea, ip);
2151 		}
2152 	}
2153 
2154 	/*
2155 	 * format maplock for free of old EA extent
2156 	 */
2157 	if (!test_cflag(COMMIT_Nolink, ip) && oldea->flag & DXD_EXTENT) {
2158 		if (tlck == NULL) {
2159 			tlck = txMaplock(tid, ip, tlckMAP);
2160 			maplock = (struct pxd_lock *) & tlck->lock;
2161 			pxdlock = (struct pxd_lock *) maplock;
2162 			maplock->index = 0;
2163 		}
2164 		pxdlock->flag = mlckFREEPXD;
2165 		PXDaddress(&pxdlock->pxd, addressDXD(oldea));
2166 		PXDlength(&pxdlock->pxd, lengthDXD(oldea));
2167 		maplock->index++;
2168 	}
2169 }
2170 
2171 /*
2172  *	txForce()
2173  *
2174  * function: synchronously write pages locked by transaction
2175  *	     after txLog() but before txUpdateMap();
2176  */
txForce(struct tblock * tblk)2177 static void txForce(struct tblock * tblk)
2178 {
2179 	struct tlock *tlck;
2180 	lid_t lid, next;
2181 	struct metapage *mp;
2182 
2183 	/*
2184 	 * reverse the order of transaction tlocks in
2185 	 * careful update order of address index pages
2186 	 * (right to left, bottom up)
2187 	 */
2188 	tlck = lid_to_tlock(tblk->next);
2189 	lid = tlck->next;
2190 	tlck->next = 0;
2191 	while (lid) {
2192 		tlck = lid_to_tlock(lid);
2193 		next = tlck->next;
2194 		tlck->next = tblk->next;
2195 		tblk->next = lid;
2196 		lid = next;
2197 	}
2198 
2199 	/*
2200 	 * synchronously write the page, and
2201 	 * hold the page for txUpdateMap();
2202 	 */
2203 	for (lid = tblk->next; lid; lid = next) {
2204 		tlck = lid_to_tlock(lid);
2205 		next = tlck->next;
2206 
2207 		if ((mp = tlck->mp) != NULL &&
2208 		    (tlck->type & tlckBTROOT) == 0) {
2209 			assert(mp->xflag & COMMIT_PAGE);
2210 
2211 			if (tlck->flag & tlckWRITEPAGE) {
2212 				tlck->flag &= ~tlckWRITEPAGE;
2213 
2214 				/* do not release page to freelist */
2215 				force_metapage(mp);
2216 #if 0
2217 				/*
2218 				 * The "right" thing to do here is to
2219 				 * synchronously write the metadata.
2220 				 * With the current implementation this
2221 				 * is hard since write_metapage requires
2222 				 * us to kunmap & remap the page.  If we
2223 				 * have tlocks pointing into the metadata
2224 				 * pages, we don't want to do this.  I think
2225 				 * we can get by with synchronously writing
2226 				 * the pages when they are released.
2227 				 */
2228 				assert(mp->nohomeok);
2229 				set_bit(META_dirty, &mp->flag);
2230 				set_bit(META_sync, &mp->flag);
2231 #endif
2232 			}
2233 		}
2234 	}
2235 }
2236 
2237 /*
2238  *	txUpdateMap()
2239  *
2240  * function:	update persistent allocation map (and working map
2241  *		if appropriate);
2242  *
2243  * parameter:
2244  */
txUpdateMap(struct tblock * tblk)2245 static void txUpdateMap(struct tblock * tblk)
2246 {
2247 	struct inode *ip;
2248 	struct inode *ipimap;
2249 	lid_t lid;
2250 	struct tlock *tlck;
2251 	struct maplock *maplock;
2252 	struct pxd_lock pxdlock;
2253 	int maptype;
2254 	int k, nlock;
2255 	struct metapage *mp = NULL;
2256 
2257 	ipimap = JFS_SBI(tblk->sb)->ipimap;
2258 
2259 	maptype = (tblk->xflag & COMMIT_PMAP) ? COMMIT_PMAP : COMMIT_PWMAP;
2260 
2261 
2262 	/*
2263 	 *	update block allocation map
2264 	 *
2265 	 * update allocation state in pmap (and wmap) and
2266 	 * update lsn of the pmap page;
2267 	 */
2268 	/*
2269 	 * scan each tlock/page of transaction for block allocation/free:
2270 	 *
2271 	 * for each tlock/page of transaction, update map.
2272 	 *  ? are there tlock for pmap and pwmap at the same time ?
2273 	 */
2274 	for (lid = tblk->next; lid; lid = tlck->next) {
2275 		tlck = lid_to_tlock(lid);
2276 
2277 		if ((tlck->flag & tlckUPDATEMAP) == 0)
2278 			continue;
2279 
2280 		if (tlck->flag & tlckFREEPAGE) {
2281 			/*
2282 			 * Another thread may attempt to reuse freed space
2283 			 * immediately, so we want to get rid of the metapage
2284 			 * before anyone else has a chance to get it.
2285 			 * Lock metapage, update maps, then invalidate
2286 			 * the metapage.
2287 			 */
2288 			mp = tlck->mp;
2289 			ASSERT(mp->xflag & COMMIT_PAGE);
2290 			grab_metapage(mp);
2291 		}
2292 
2293 		/*
2294 		 * extent list:
2295 		 * . in-line PXD list:
2296 		 * . out-of-line XAD list:
2297 		 */
2298 		maplock = (struct maplock *) & tlck->lock;
2299 		nlock = maplock->index;
2300 
2301 		for (k = 0; k < nlock; k++, maplock++) {
2302 			/*
2303 			 * allocate blocks in persistent map:
2304 			 *
2305 			 * blocks have been allocated from wmap at alloc time;
2306 			 */
2307 			if (maplock->flag & mlckALLOC) {
2308 				txAllocPMap(ipimap, maplock, tblk);
2309 			}
2310 			/*
2311 			 * free blocks in persistent and working map:
2312 			 * blocks will be freed in pmap and then in wmap;
2313 			 *
2314 			 * ? tblock specifies the PMAP/PWMAP based upon
2315 			 * transaction
2316 			 *
2317 			 * free blocks in persistent map:
2318 			 * blocks will be freed from wmap at last reference
2319 			 * release of the object for regular files;
2320 			 *
2321 			 * Alway free blocks from both persistent & working
2322 			 * maps for directories
2323 			 */
2324 			else {	/* (maplock->flag & mlckFREE) */
2325 
2326 				if (tlck->flag & tlckDIRECTORY)
2327 					txFreeMap(ipimap, maplock,
2328 						  tblk, COMMIT_PWMAP);
2329 				else
2330 					txFreeMap(ipimap, maplock,
2331 						  tblk, maptype);
2332 			}
2333 		}
2334 		if (tlck->flag & tlckFREEPAGE) {
2335 			if (!(tblk->flag & tblkGC_LAZY)) {
2336 				/* This is equivalent to txRelease */
2337 				ASSERT(mp->lid == lid);
2338 				tlck->mp->lid = 0;
2339 			}
2340 			assert(mp->nohomeok == 1);
2341 			metapage_homeok(mp);
2342 			discard_metapage(mp);
2343 			tlck->mp = NULL;
2344 		}
2345 	}
2346 	/*
2347 	 *	update inode allocation map
2348 	 *
2349 	 * update allocation state in pmap and
2350 	 * update lsn of the pmap page;
2351 	 * update in-memory inode flag/state
2352 	 *
2353 	 * unlock mapper/write lock
2354 	 */
2355 	if (tblk->xflag & COMMIT_CREATE) {
2356 		diUpdatePMap(ipimap, tblk->ino, false, tblk);
2357 		/* update persistent block allocation map
2358 		 * for the allocation of inode extent;
2359 		 */
2360 		pxdlock.flag = mlckALLOCPXD;
2361 		pxdlock.pxd = tblk->u.ixpxd;
2362 		pxdlock.index = 1;
2363 		txAllocPMap(ipimap, (struct maplock *) & pxdlock, tblk);
2364 	} else if (tblk->xflag & COMMIT_DELETE) {
2365 		ip = tblk->u.ip;
2366 		diUpdatePMap(ipimap, ip->i_ino, true, tblk);
2367 		iput(ip);
2368 	}
2369 }
2370 
2371 /*
2372  *	txAllocPMap()
2373  *
2374  * function: allocate from persistent map;
2375  *
2376  * parameter:
2377  *	ipbmap	-
2378  *	malock	-
2379  *		xad list:
2380  *		pxd:
2381  *
2382  *	maptype -
2383  *		allocate from persistent map;
2384  *		free from persistent map;
2385  *		(e.g., tmp file - free from working map at releae
2386  *		 of last reference);
2387  *		free from persistent and working map;
2388  *
2389  *	lsn	- log sequence number;
2390  */
txAllocPMap(struct inode * ip,struct maplock * maplock,struct tblock * tblk)2391 static void txAllocPMap(struct inode *ip, struct maplock * maplock,
2392 			struct tblock * tblk)
2393 {
2394 	struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
2395 	struct xdlistlock *xadlistlock;
2396 	xad_t *xad;
2397 	s64 xaddr;
2398 	int xlen;
2399 	struct pxd_lock *pxdlock;
2400 	struct xdlistlock *pxdlistlock;
2401 	pxd_t *pxd;
2402 	int n;
2403 
2404 	/*
2405 	 * allocate from persistent map;
2406 	 */
2407 	if (maplock->flag & mlckALLOCXADLIST) {
2408 		xadlistlock = (struct xdlistlock *) maplock;
2409 		xad = xadlistlock->xdlist;
2410 		for (n = 0; n < xadlistlock->count; n++, xad++) {
2411 			if (xad->flag & (XAD_NEW | XAD_EXTENDED)) {
2412 				xaddr = addressXAD(xad);
2413 				xlen = lengthXAD(xad);
2414 				dbUpdatePMap(ipbmap, false, xaddr,
2415 					     (s64) xlen, tblk);
2416 				xad->flag &= ~(XAD_NEW | XAD_EXTENDED);
2417 				jfs_info("allocPMap: xaddr:0x%lx xlen:%d",
2418 					 (ulong) xaddr, xlen);
2419 			}
2420 		}
2421 	} else if (maplock->flag & mlckALLOCPXD) {
2422 		pxdlock = (struct pxd_lock *) maplock;
2423 		xaddr = addressPXD(&pxdlock->pxd);
2424 		xlen = lengthPXD(&pxdlock->pxd);
2425 		dbUpdatePMap(ipbmap, false, xaddr, (s64) xlen, tblk);
2426 		jfs_info("allocPMap: xaddr:0x%lx xlen:%d", (ulong) xaddr, xlen);
2427 	} else {		/* (maplock->flag & mlckALLOCPXDLIST) */
2428 
2429 		pxdlistlock = (struct xdlistlock *) maplock;
2430 		pxd = pxdlistlock->xdlist;
2431 		for (n = 0; n < pxdlistlock->count; n++, pxd++) {
2432 			xaddr = addressPXD(pxd);
2433 			xlen = lengthPXD(pxd);
2434 			dbUpdatePMap(ipbmap, false, xaddr, (s64) xlen,
2435 				     tblk);
2436 			jfs_info("allocPMap: xaddr:0x%lx xlen:%d",
2437 				 (ulong) xaddr, xlen);
2438 		}
2439 	}
2440 }
2441 
2442 /*
2443  *	txFreeMap()
2444  *
2445  * function:	free from persistent and/or working map;
2446  *
2447  * todo: optimization
2448  */
txFreeMap(struct inode * ip,struct maplock * maplock,struct tblock * tblk,int maptype)2449 void txFreeMap(struct inode *ip,
2450 	       struct maplock * maplock, struct tblock * tblk, int maptype)
2451 {
2452 	struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
2453 	struct xdlistlock *xadlistlock;
2454 	xad_t *xad;
2455 	s64 xaddr;
2456 	int xlen;
2457 	struct pxd_lock *pxdlock;
2458 	struct xdlistlock *pxdlistlock;
2459 	pxd_t *pxd;
2460 	int n;
2461 
2462 	jfs_info("txFreeMap: tblk:0x%p maplock:0x%p maptype:0x%x",
2463 		 tblk, maplock, maptype);
2464 
2465 	/*
2466 	 * free from persistent map;
2467 	 */
2468 	if (maptype == COMMIT_PMAP || maptype == COMMIT_PWMAP) {
2469 		if (maplock->flag & mlckFREEXADLIST) {
2470 			xadlistlock = (struct xdlistlock *) maplock;
2471 			xad = xadlistlock->xdlist;
2472 			for (n = 0; n < xadlistlock->count; n++, xad++) {
2473 				if (!(xad->flag & XAD_NEW)) {
2474 					xaddr = addressXAD(xad);
2475 					xlen = lengthXAD(xad);
2476 					dbUpdatePMap(ipbmap, true, xaddr,
2477 						     (s64) xlen, tblk);
2478 					jfs_info("freePMap: xaddr:0x%lx xlen:%d",
2479 						 (ulong) xaddr, xlen);
2480 				}
2481 			}
2482 		} else if (maplock->flag & mlckFREEPXD) {
2483 			pxdlock = (struct pxd_lock *) maplock;
2484 			xaddr = addressPXD(&pxdlock->pxd);
2485 			xlen = lengthPXD(&pxdlock->pxd);
2486 			dbUpdatePMap(ipbmap, true, xaddr, (s64) xlen,
2487 				     tblk);
2488 			jfs_info("freePMap: xaddr:0x%lx xlen:%d",
2489 				 (ulong) xaddr, xlen);
2490 		} else {	/* (maplock->flag & mlckALLOCPXDLIST) */
2491 
2492 			pxdlistlock = (struct xdlistlock *) maplock;
2493 			pxd = pxdlistlock->xdlist;
2494 			for (n = 0; n < pxdlistlock->count; n++, pxd++) {
2495 				xaddr = addressPXD(pxd);
2496 				xlen = lengthPXD(pxd);
2497 				dbUpdatePMap(ipbmap, true, xaddr,
2498 					     (s64) xlen, tblk);
2499 				jfs_info("freePMap: xaddr:0x%lx xlen:%d",
2500 					 (ulong) xaddr, xlen);
2501 			}
2502 		}
2503 	}
2504 
2505 	/*
2506 	 * free from working map;
2507 	 */
2508 	if (maptype == COMMIT_PWMAP || maptype == COMMIT_WMAP) {
2509 		if (maplock->flag & mlckFREEXADLIST) {
2510 			xadlistlock = (struct xdlistlock *) maplock;
2511 			xad = xadlistlock->xdlist;
2512 			for (n = 0; n < xadlistlock->count; n++, xad++) {
2513 				xaddr = addressXAD(xad);
2514 				xlen = lengthXAD(xad);
2515 				dbFree(ip, xaddr, (s64) xlen);
2516 				xad->flag = 0;
2517 				jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
2518 					 (ulong) xaddr, xlen);
2519 			}
2520 		} else if (maplock->flag & mlckFREEPXD) {
2521 			pxdlock = (struct pxd_lock *) maplock;
2522 			xaddr = addressPXD(&pxdlock->pxd);
2523 			xlen = lengthPXD(&pxdlock->pxd);
2524 			dbFree(ip, xaddr, (s64) xlen);
2525 			jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
2526 				 (ulong) xaddr, xlen);
2527 		} else {	/* (maplock->flag & mlckFREEPXDLIST) */
2528 
2529 			pxdlistlock = (struct xdlistlock *) maplock;
2530 			pxd = pxdlistlock->xdlist;
2531 			for (n = 0; n < pxdlistlock->count; n++, pxd++) {
2532 				xaddr = addressPXD(pxd);
2533 				xlen = lengthPXD(pxd);
2534 				dbFree(ip, xaddr, (s64) xlen);
2535 				jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
2536 					 (ulong) xaddr, xlen);
2537 			}
2538 		}
2539 	}
2540 }
2541 
2542 /*
2543  *	txFreelock()
2544  *
2545  * function:	remove tlock from inode anonymous locklist
2546  */
txFreelock(struct inode * ip)2547 void txFreelock(struct inode *ip)
2548 {
2549 	struct jfs_inode_info *jfs_ip = JFS_IP(ip);
2550 	struct tlock *xtlck, *tlck;
2551 	lid_t xlid = 0, lid;
2552 
2553 	if (!jfs_ip->atlhead)
2554 		return;
2555 
2556 	TXN_LOCK();
2557 	xtlck = (struct tlock *) &jfs_ip->atlhead;
2558 
2559 	while ((lid = xtlck->next) != 0) {
2560 		tlck = lid_to_tlock(lid);
2561 		if (tlck->flag & tlckFREELOCK) {
2562 			xtlck->next = tlck->next;
2563 			txLockFree(lid);
2564 		} else {
2565 			xtlck = tlck;
2566 			xlid = lid;
2567 		}
2568 	}
2569 
2570 	if (jfs_ip->atlhead)
2571 		jfs_ip->atltail = xlid;
2572 	else {
2573 		jfs_ip->atltail = 0;
2574 		/*
2575 		 * If inode was on anon_list, remove it
2576 		 */
2577 		list_del_init(&jfs_ip->anon_inode_list);
2578 	}
2579 	TXN_UNLOCK();
2580 }
2581 
2582 /*
2583  *	txAbort()
2584  *
2585  * function: abort tx before commit;
2586  *
2587  * frees line-locks and segment locks for all
2588  * segments in comdata structure.
2589  * Optionally sets state of file-system to FM_DIRTY in super-block.
2590  * log age of page-frames in memory for which caller has
2591  * are reset to 0 (to avoid logwarap).
2592  */
txAbort(tid_t tid,int dirty)2593 void txAbort(tid_t tid, int dirty)
2594 {
2595 	lid_t lid, next;
2596 	struct metapage *mp;
2597 	struct tblock *tblk = tid_to_tblock(tid);
2598 	struct tlock *tlck;
2599 
2600 	/*
2601 	 * free tlocks of the transaction
2602 	 */
2603 	for (lid = tblk->next; lid; lid = next) {
2604 		tlck = lid_to_tlock(lid);
2605 		next = tlck->next;
2606 		mp = tlck->mp;
2607 		JFS_IP(tlck->ip)->xtlid = 0;
2608 
2609 		if (mp) {
2610 			mp->lid = 0;
2611 
2612 			/*
2613 			 * reset lsn of page to avoid logwarap:
2614 			 *
2615 			 * (page may have been previously committed by another
2616 			 * transaction(s) but has not been paged, i.e.,
2617 			 * it may be on logsync list even though it has not
2618 			 * been logged for the current tx.)
2619 			 */
2620 			if (mp->xflag & COMMIT_PAGE && mp->lsn)
2621 				LogSyncRelease(mp);
2622 		}
2623 		/* insert tlock at head of freelist */
2624 		TXN_LOCK();
2625 		txLockFree(lid);
2626 		TXN_UNLOCK();
2627 	}
2628 
2629 	/* caller will free the transaction block */
2630 
2631 	tblk->next = tblk->last = 0;
2632 
2633 	/*
2634 	 * mark filesystem dirty
2635 	 */
2636 	if (dirty)
2637 		jfs_error(tblk->sb, "\n");
2638 
2639 	return;
2640 }
2641 
2642 /*
2643  *	txLazyCommit(void)
2644  *
2645  *	All transactions except those changing ipimap (COMMIT_FORCE) are
2646  *	processed by this routine.  This insures that the inode and block
2647  *	allocation maps are updated in order.  For synchronous transactions,
2648  *	let the user thread finish processing after txUpdateMap() is called.
2649  */
txLazyCommit(struct tblock * tblk)2650 static void txLazyCommit(struct tblock * tblk)
2651 {
2652 	struct jfs_log *log;
2653 
2654 	while (((tblk->flag & tblkGC_READY) == 0) &&
2655 	       ((tblk->flag & tblkGC_UNLOCKED) == 0)) {
2656 		/* We must have gotten ahead of the user thread
2657 		 */
2658 		jfs_info("jfs_lazycommit: tblk 0x%p not unlocked", tblk);
2659 		yield();
2660 	}
2661 
2662 	jfs_info("txLazyCommit: processing tblk 0x%p", tblk);
2663 
2664 	txUpdateMap(tblk);
2665 
2666 	log = (struct jfs_log *) JFS_SBI(tblk->sb)->log;
2667 
2668 	spin_lock_irq(&log->gclock);	// LOGGC_LOCK
2669 
2670 	tblk->flag |= tblkGC_COMMITTED;
2671 
2672 	if (tblk->flag & tblkGC_READY)
2673 		log->gcrtc--;
2674 
2675 	wake_up_all(&tblk->gcwait);	// LOGGC_WAKEUP
2676 
2677 	/*
2678 	 * Can't release log->gclock until we've tested tblk->flag
2679 	 */
2680 	if (tblk->flag & tblkGC_LAZY) {
2681 		spin_unlock_irq(&log->gclock);	// LOGGC_UNLOCK
2682 		txUnlock(tblk);
2683 		tblk->flag &= ~tblkGC_LAZY;
2684 		txEnd(tblk - TxBlock);	/* Convert back to tid */
2685 	} else
2686 		spin_unlock_irq(&log->gclock);	// LOGGC_UNLOCK
2687 
2688 	jfs_info("txLazyCommit: done: tblk = 0x%p", tblk);
2689 }
2690 
2691 /*
2692  *	jfs_lazycommit(void)
2693  *
2694  *	To be run as a kernel daemon.  If lbmIODone is called in an interrupt
2695  *	context, or where blocking is not wanted, this routine will process
2696  *	committed transactions from the unlock queue.
2697  */
jfs_lazycommit(void * arg)2698 int jfs_lazycommit(void *arg)
2699 {
2700 	int WorkDone;
2701 	struct tblock *tblk;
2702 	unsigned long flags;
2703 	struct jfs_sb_info *sbi;
2704 
2705 	do {
2706 		LAZY_LOCK(flags);
2707 		jfs_commit_thread_waking = 0;	/* OK to wake another thread */
2708 		while (!list_empty(&TxAnchor.unlock_queue)) {
2709 			WorkDone = 0;
2710 			list_for_each_entry(tblk, &TxAnchor.unlock_queue,
2711 					    cqueue) {
2712 
2713 				sbi = JFS_SBI(tblk->sb);
2714 				/*
2715 				 * For each volume, the transactions must be
2716 				 * handled in order.  If another commit thread
2717 				 * is handling a tblk for this superblock,
2718 				 * skip it
2719 				 */
2720 				if (sbi->commit_state & IN_LAZYCOMMIT)
2721 					continue;
2722 
2723 				sbi->commit_state |= IN_LAZYCOMMIT;
2724 				WorkDone = 1;
2725 
2726 				/*
2727 				 * Remove transaction from queue
2728 				 */
2729 				list_del(&tblk->cqueue);
2730 
2731 				LAZY_UNLOCK(flags);
2732 				txLazyCommit(tblk);
2733 				LAZY_LOCK(flags);
2734 
2735 				sbi->commit_state &= ~IN_LAZYCOMMIT;
2736 				/*
2737 				 * Don't continue in the for loop.  (We can't
2738 				 * anyway, it's unsafe!)  We want to go back to
2739 				 * the beginning of the list.
2740 				 */
2741 				break;
2742 			}
2743 
2744 			/* If there was nothing to do, don't continue */
2745 			if (!WorkDone)
2746 				break;
2747 		}
2748 		/* In case a wakeup came while all threads were active */
2749 		jfs_commit_thread_waking = 0;
2750 
2751 		if (freezing(current)) {
2752 			LAZY_UNLOCK(flags);
2753 			try_to_freeze();
2754 		} else {
2755 			DECLARE_WAITQUEUE(wq, current);
2756 
2757 			add_wait_queue(&jfs_commit_thread_wait, &wq);
2758 			set_current_state(TASK_INTERRUPTIBLE);
2759 			LAZY_UNLOCK(flags);
2760 			schedule();
2761 			remove_wait_queue(&jfs_commit_thread_wait, &wq);
2762 		}
2763 	} while (!kthread_should_stop());
2764 
2765 	if (!list_empty(&TxAnchor.unlock_queue))
2766 		jfs_err("jfs_lazycommit being killed w/pending transactions!");
2767 	else
2768 		jfs_info("jfs_lazycommit being killed");
2769 	return 0;
2770 }
2771 
txLazyUnlock(struct tblock * tblk)2772 void txLazyUnlock(struct tblock * tblk)
2773 {
2774 	unsigned long flags;
2775 
2776 	LAZY_LOCK(flags);
2777 
2778 	list_add_tail(&tblk->cqueue, &TxAnchor.unlock_queue);
2779 	/*
2780 	 * Don't wake up a commit thread if there is already one servicing
2781 	 * this superblock, or if the last one we woke up hasn't started yet.
2782 	 */
2783 	if (!(JFS_SBI(tblk->sb)->commit_state & IN_LAZYCOMMIT) &&
2784 	    !jfs_commit_thread_waking) {
2785 		jfs_commit_thread_waking = 1;
2786 		wake_up(&jfs_commit_thread_wait);
2787 	}
2788 	LAZY_UNLOCK(flags);
2789 }
2790 
LogSyncRelease(struct metapage * mp)2791 static void LogSyncRelease(struct metapage * mp)
2792 {
2793 	struct jfs_log *log = mp->log;
2794 
2795 	assert(mp->nohomeok);
2796 	assert(log);
2797 	metapage_homeok(mp);
2798 }
2799 
2800 /*
2801  *	txQuiesce
2802  *
2803  *	Block all new transactions and push anonymous transactions to
2804  *	completion
2805  *
2806  *	This does almost the same thing as jfs_sync below.  We don't
2807  *	worry about deadlocking when jfs_tlocks_low is set, since we would
2808  *	expect jfs_sync to get us out of that jam.
2809  */
txQuiesce(struct super_block * sb)2810 void txQuiesce(struct super_block *sb)
2811 {
2812 	struct inode *ip;
2813 	struct jfs_inode_info *jfs_ip;
2814 	struct jfs_log *log = JFS_SBI(sb)->log;
2815 	tid_t tid;
2816 
2817 	set_bit(log_QUIESCE, &log->flag);
2818 
2819 	TXN_LOCK();
2820 restart:
2821 	while (!list_empty(&TxAnchor.anon_list)) {
2822 		jfs_ip = list_entry(TxAnchor.anon_list.next,
2823 				    struct jfs_inode_info,
2824 				    anon_inode_list);
2825 		ip = &jfs_ip->vfs_inode;
2826 
2827 		/*
2828 		 * inode will be removed from anonymous list
2829 		 * when it is committed
2830 		 */
2831 		TXN_UNLOCK();
2832 		tid = txBegin(ip->i_sb, COMMIT_INODE | COMMIT_FORCE);
2833 		mutex_lock(&jfs_ip->commit_mutex);
2834 		txCommit(tid, 1, &ip, 0);
2835 		txEnd(tid);
2836 		mutex_unlock(&jfs_ip->commit_mutex);
2837 		/*
2838 		 * Just to be safe.  I don't know how
2839 		 * long we can run without blocking
2840 		 */
2841 		cond_resched();
2842 		TXN_LOCK();
2843 	}
2844 
2845 	/*
2846 	 * If jfs_sync is running in parallel, there could be some inodes
2847 	 * on anon_list2.  Let's check.
2848 	 */
2849 	if (!list_empty(&TxAnchor.anon_list2)) {
2850 		list_splice_init(&TxAnchor.anon_list2, &TxAnchor.anon_list);
2851 		goto restart;
2852 	}
2853 	TXN_UNLOCK();
2854 
2855 	/*
2856 	 * We may need to kick off the group commit
2857 	 */
2858 	jfs_flush_journal(log, 0);
2859 }
2860 
2861 /*
2862  * txResume()
2863  *
2864  * Allows transactions to start again following txQuiesce
2865  */
txResume(struct super_block * sb)2866 void txResume(struct super_block *sb)
2867 {
2868 	struct jfs_log *log = JFS_SBI(sb)->log;
2869 
2870 	clear_bit(log_QUIESCE, &log->flag);
2871 	TXN_WAKEUP(&log->syncwait);
2872 }
2873 
2874 /*
2875  *	jfs_sync(void)
2876  *
2877  *	To be run as a kernel daemon.  This is awakened when tlocks run low.
2878  *	We write any inodes that have anonymous tlocks so they will become
2879  *	available.
2880  */
jfs_sync(void * arg)2881 int jfs_sync(void *arg)
2882 {
2883 	struct inode *ip;
2884 	struct jfs_inode_info *jfs_ip;
2885 	tid_t tid;
2886 
2887 	do {
2888 		/*
2889 		 * write each inode on the anonymous inode list
2890 		 */
2891 		TXN_LOCK();
2892 		while (jfs_tlocks_low && !list_empty(&TxAnchor.anon_list)) {
2893 			jfs_ip = list_entry(TxAnchor.anon_list.next,
2894 					    struct jfs_inode_info,
2895 					    anon_inode_list);
2896 			ip = &jfs_ip->vfs_inode;
2897 
2898 			if (! igrab(ip)) {
2899 				/*
2900 				 * Inode is being freed
2901 				 */
2902 				list_del_init(&jfs_ip->anon_inode_list);
2903 			} else if (mutex_trylock(&jfs_ip->commit_mutex)) {
2904 				/*
2905 				 * inode will be removed from anonymous list
2906 				 * when it is committed
2907 				 */
2908 				TXN_UNLOCK();
2909 				tid = txBegin(ip->i_sb, COMMIT_INODE);
2910 				txCommit(tid, 1, &ip, 0);
2911 				txEnd(tid);
2912 				mutex_unlock(&jfs_ip->commit_mutex);
2913 
2914 				iput(ip);
2915 				/*
2916 				 * Just to be safe.  I don't know how
2917 				 * long we can run without blocking
2918 				 */
2919 				cond_resched();
2920 				TXN_LOCK();
2921 			} else {
2922 				/* We can't get the commit mutex.  It may
2923 				 * be held by a thread waiting for tlock's
2924 				 * so let's not block here.  Save it to
2925 				 * put back on the anon_list.
2926 				 */
2927 
2928 				/* Move from anon_list to anon_list2 */
2929 				list_move(&jfs_ip->anon_inode_list,
2930 					  &TxAnchor.anon_list2);
2931 
2932 				TXN_UNLOCK();
2933 				iput(ip);
2934 				TXN_LOCK();
2935 			}
2936 		}
2937 		/* Add anon_list2 back to anon_list */
2938 		list_splice_init(&TxAnchor.anon_list2, &TxAnchor.anon_list);
2939 
2940 		if (freezing(current)) {
2941 			TXN_UNLOCK();
2942 			try_to_freeze();
2943 		} else {
2944 			set_current_state(TASK_INTERRUPTIBLE);
2945 			TXN_UNLOCK();
2946 			schedule();
2947 		}
2948 	} while (!kthread_should_stop());
2949 
2950 	jfs_info("jfs_sync being killed");
2951 	return 0;
2952 }
2953 
2954 #if defined(CONFIG_PROC_FS) && defined(CONFIG_JFS_DEBUG)
jfs_txanchor_proc_show(struct seq_file * m,void * v)2955 int jfs_txanchor_proc_show(struct seq_file *m, void *v)
2956 {
2957 	char *freewait;
2958 	char *freelockwait;
2959 	char *lowlockwait;
2960 
2961 	freewait =
2962 	    waitqueue_active(&TxAnchor.freewait) ? "active" : "empty";
2963 	freelockwait =
2964 	    waitqueue_active(&TxAnchor.freelockwait) ? "active" : "empty";
2965 	lowlockwait =
2966 	    waitqueue_active(&TxAnchor.lowlockwait) ? "active" : "empty";
2967 
2968 	seq_printf(m,
2969 		       "JFS TxAnchor\n"
2970 		       "============\n"
2971 		       "freetid = %d\n"
2972 		       "freewait = %s\n"
2973 		       "freelock = %d\n"
2974 		       "freelockwait = %s\n"
2975 		       "lowlockwait = %s\n"
2976 		       "tlocksInUse = %d\n"
2977 		       "jfs_tlocks_low = %d\n"
2978 		       "unlock_queue is %sempty\n",
2979 		       TxAnchor.freetid,
2980 		       freewait,
2981 		       TxAnchor.freelock,
2982 		       freelockwait,
2983 		       lowlockwait,
2984 		       TxAnchor.tlocksInUse,
2985 		       jfs_tlocks_low,
2986 		       list_empty(&TxAnchor.unlock_queue) ? "" : "not ");
2987 	return 0;
2988 }
2989 #endif
2990 
2991 #if defined(CONFIG_PROC_FS) && defined(CONFIG_JFS_STATISTICS)
jfs_txstats_proc_show(struct seq_file * m,void * v)2992 int jfs_txstats_proc_show(struct seq_file *m, void *v)
2993 {
2994 	seq_printf(m,
2995 		       "JFS TxStats\n"
2996 		       "===========\n"
2997 		       "calls to txBegin = %d\n"
2998 		       "txBegin blocked by sync barrier = %d\n"
2999 		       "txBegin blocked by tlocks low = %d\n"
3000 		       "txBegin blocked by no free tid = %d\n"
3001 		       "calls to txBeginAnon = %d\n"
3002 		       "txBeginAnon blocked by sync barrier = %d\n"
3003 		       "txBeginAnon blocked by tlocks low = %d\n"
3004 		       "calls to txLockAlloc = %d\n"
3005 		       "tLockAlloc blocked by no free lock = %d\n",
3006 		       TxStat.txBegin,
3007 		       TxStat.txBegin_barrier,
3008 		       TxStat.txBegin_lockslow,
3009 		       TxStat.txBegin_freetid,
3010 		       TxStat.txBeginAnon,
3011 		       TxStat.txBeginAnon_barrier,
3012 		       TxStat.txBeginAnon_lockslow,
3013 		       TxStat.txLockAlloc,
3014 		       TxStat.txLockAlloc_freelock);
3015 	return 0;
3016 }
3017 #endif
3018