Loading...
1/*
2 * Copyright (C) International Business Machines Corp., 2000-2005
3 * Portions Copyright (C) Christoph Hellwig, 2001-2002
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
13 * the GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18 */
19
20/*
21 * jfs_txnmgr.c: transaction manager
22 *
23 * notes:
24 * transaction starts with txBegin() and ends with txCommit()
25 * or txAbort().
26 *
27 * tlock is acquired at the time of update;
28 * (obviate scan at commit time for xtree and dtree)
29 * tlock and mp points to each other;
30 * (no hashlist for mp -> tlock).
31 *
32 * special cases:
33 * tlock on in-memory inode:
34 * in-place tlock in the in-memory inode itself;
35 * converted to page lock by iWrite() at commit time.
36 *
37 * tlock during write()/mmap() under anonymous transaction (tid = 0):
38 * transferred (?) to transaction at commit time.
39 *
40 * use the page itself to update allocation maps
41 * (obviate intermediate replication of allocation/deallocation data)
42 * hold on to mp+lock thru update of maps
43 */
44
45#include <linux/fs.h>
46#include <linux/vmalloc.h>
47#include <linux/completion.h>
48#include <linux/freezer.h>
49#include <linux/module.h>
50#include <linux/moduleparam.h>
51#include <linux/kthread.h>
52#include <linux/seq_file.h>
53#include "jfs_incore.h"
54#include "jfs_inode.h"
55#include "jfs_filsys.h"
56#include "jfs_metapage.h"
57#include "jfs_dinode.h"
58#include "jfs_imap.h"
59#include "jfs_dmap.h"
60#include "jfs_superblock.h"
61#include "jfs_debug.h"
62
63/*
64 * transaction management structures
65 */
66static struct {
67 int freetid; /* index of a free tid structure */
68 int freelock; /* index first free lock word */
69 wait_queue_head_t freewait; /* eventlist of free tblock */
70 wait_queue_head_t freelockwait; /* eventlist of free tlock */
71 wait_queue_head_t lowlockwait; /* eventlist of ample tlocks */
72 int tlocksInUse; /* Number of tlocks in use */
73 spinlock_t LazyLock; /* synchronize sync_queue & unlock_queue */
74/* struct tblock *sync_queue; * Transactions waiting for data sync */
75 struct list_head unlock_queue; /* Txns waiting to be released */
76 struct list_head anon_list; /* inodes having anonymous txns */
77 struct list_head anon_list2; /* inodes having anonymous txns
78 that couldn't be sync'ed */
79} TxAnchor;
80
81int jfs_tlocks_low; /* Indicates low number of available tlocks */
82
83#ifdef CONFIG_JFS_STATISTICS
84static struct {
85 uint txBegin;
86 uint txBegin_barrier;
87 uint txBegin_lockslow;
88 uint txBegin_freetid;
89 uint txBeginAnon;
90 uint txBeginAnon_barrier;
91 uint txBeginAnon_lockslow;
92 uint txLockAlloc;
93 uint txLockAlloc_freelock;
94} TxStat;
95#endif
96
97static int nTxBlock = -1; /* number of transaction blocks */
98module_param(nTxBlock, int, 0);
99MODULE_PARM_DESC(nTxBlock,
100 "Number of transaction blocks (max:65536)");
101
102static int nTxLock = -1; /* number of transaction locks */
103module_param(nTxLock, int, 0);
104MODULE_PARM_DESC(nTxLock,
105 "Number of transaction locks (max:65536)");
106
107struct tblock *TxBlock; /* transaction block table */
108static int TxLockLWM; /* Low water mark for number of txLocks used */
109static int TxLockHWM; /* High water mark for number of txLocks used */
110static int TxLockVHWM; /* Very High water mark */
111struct tlock *TxLock; /* transaction lock table */
112
113/*
114 * transaction management lock
115 */
116static DEFINE_SPINLOCK(jfsTxnLock);
117
118#define TXN_LOCK() spin_lock(&jfsTxnLock)
119#define TXN_UNLOCK() spin_unlock(&jfsTxnLock)
120
121#define LAZY_LOCK_INIT() spin_lock_init(&TxAnchor.LazyLock);
122#define LAZY_LOCK(flags) spin_lock_irqsave(&TxAnchor.LazyLock, flags)
123#define LAZY_UNLOCK(flags) spin_unlock_irqrestore(&TxAnchor.LazyLock, flags)
124
125static DECLARE_WAIT_QUEUE_HEAD(jfs_commit_thread_wait);
126static int jfs_commit_thread_waking;
127
128/*
129 * Retry logic exist outside these macros to protect from spurrious wakeups.
130 */
131static inline void TXN_SLEEP_DROP_LOCK(wait_queue_head_t * event)
132{
133 DECLARE_WAITQUEUE(wait, current);
134
135 add_wait_queue(event, &wait);
136 set_current_state(TASK_UNINTERRUPTIBLE);
137 TXN_UNLOCK();
138 io_schedule();
139 remove_wait_queue(event, &wait);
140}
141
142#define TXN_SLEEP(event)\
143{\
144 TXN_SLEEP_DROP_LOCK(event);\
145 TXN_LOCK();\
146}
147
148#define TXN_WAKEUP(event) wake_up_all(event)
149
150/*
151 * statistics
152 */
153static struct {
154 tid_t maxtid; /* 4: biggest tid ever used */
155 lid_t maxlid; /* 4: biggest lid ever used */
156 int ntid; /* 4: # of transactions performed */
157 int nlid; /* 4: # of tlocks acquired */
158 int waitlock; /* 4: # of tlock wait */
159} stattx;
160
161/*
162 * forward references
163 */
164static int diLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
165 struct tlock * tlck, struct commit * cd);
166static int dataLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
167 struct tlock * tlck);
168static void dtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
169 struct tlock * tlck);
170static void mapLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
171 struct tlock * tlck);
172static void txAllocPMap(struct inode *ip, struct maplock * maplock,
173 struct tblock * tblk);
174static void txForce(struct tblock * tblk);
175static int txLog(struct jfs_log * log, struct tblock * tblk,
176 struct commit * cd);
177static void txUpdateMap(struct tblock * tblk);
178static void txRelease(struct tblock * tblk);
179static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
180 struct tlock * tlck);
181static void LogSyncRelease(struct metapage * mp);
182
183/*
184 * transaction block/lock management
185 * ---------------------------------
186 */
187
188/*
189 * Get a transaction lock from the free list. If the number in use is
190 * greater than the high water mark, wake up the sync daemon. This should
191 * free some anonymous transaction locks. (TXN_LOCK must be held.)
192 */
193static lid_t txLockAlloc(void)
194{
195 lid_t lid;
196
197 INCREMENT(TxStat.txLockAlloc);
198 if (!TxAnchor.freelock) {
199 INCREMENT(TxStat.txLockAlloc_freelock);
200 }
201
202 while (!(lid = TxAnchor.freelock))
203 TXN_SLEEP(&TxAnchor.freelockwait);
204 TxAnchor.freelock = TxLock[lid].next;
205 HIGHWATERMARK(stattx.maxlid, lid);
206 if ((++TxAnchor.tlocksInUse > TxLockHWM) && (jfs_tlocks_low == 0)) {
207 jfs_info("txLockAlloc tlocks low");
208 jfs_tlocks_low = 1;
209 wake_up_process(jfsSyncThread);
210 }
211
212 return lid;
213}
214
215static void txLockFree(lid_t lid)
216{
217 TxLock[lid].tid = 0;
218 TxLock[lid].next = TxAnchor.freelock;
219 TxAnchor.freelock = lid;
220 TxAnchor.tlocksInUse--;
221 if (jfs_tlocks_low && (TxAnchor.tlocksInUse < TxLockLWM)) {
222 jfs_info("txLockFree jfs_tlocks_low no more");
223 jfs_tlocks_low = 0;
224 TXN_WAKEUP(&TxAnchor.lowlockwait);
225 }
226 TXN_WAKEUP(&TxAnchor.freelockwait);
227}
228
229/*
230 * NAME: txInit()
231 *
232 * FUNCTION: initialize transaction management structures
233 *
234 * RETURN:
235 *
236 * serialization: single thread at jfs_init()
237 */
238int txInit(void)
239{
240 int k, size;
241 struct sysinfo si;
242
243 /* Set defaults for nTxLock and nTxBlock if unset */
244
245 if (nTxLock == -1) {
246 if (nTxBlock == -1) {
247 /* Base default on memory size */
248 si_meminfo(&si);
249 if (si.totalram > (256 * 1024)) /* 1 GB */
250 nTxLock = 64 * 1024;
251 else
252 nTxLock = si.totalram >> 2;
253 } else if (nTxBlock > (8 * 1024))
254 nTxLock = 64 * 1024;
255 else
256 nTxLock = nTxBlock << 3;
257 }
258 if (nTxBlock == -1)
259 nTxBlock = nTxLock >> 3;
260
261 /* Verify tunable parameters */
262 if (nTxBlock < 16)
263 nTxBlock = 16; /* No one should set it this low */
264 if (nTxBlock > 65536)
265 nTxBlock = 65536;
266 if (nTxLock < 256)
267 nTxLock = 256; /* No one should set it this low */
268 if (nTxLock > 65536)
269 nTxLock = 65536;
270
271 printk(KERN_INFO "JFS: nTxBlock = %d, nTxLock = %d\n",
272 nTxBlock, nTxLock);
273 /*
274 * initialize transaction block (tblock) table
275 *
276 * transaction id (tid) = tblock index
277 * tid = 0 is reserved.
278 */
279 TxLockLWM = (nTxLock * 4) / 10;
280 TxLockHWM = (nTxLock * 7) / 10;
281 TxLockVHWM = (nTxLock * 8) / 10;
282
283 size = sizeof(struct tblock) * nTxBlock;
284 TxBlock = vmalloc(size);
285 if (TxBlock == NULL)
286 return -ENOMEM;
287
288 for (k = 1; k < nTxBlock - 1; k++) {
289 TxBlock[k].next = k + 1;
290 init_waitqueue_head(&TxBlock[k].gcwait);
291 init_waitqueue_head(&TxBlock[k].waitor);
292 }
293 TxBlock[k].next = 0;
294 init_waitqueue_head(&TxBlock[k].gcwait);
295 init_waitqueue_head(&TxBlock[k].waitor);
296
297 TxAnchor.freetid = 1;
298 init_waitqueue_head(&TxAnchor.freewait);
299
300 stattx.maxtid = 1; /* statistics */
301
302 /*
303 * initialize transaction lock (tlock) table
304 *
305 * transaction lock id = tlock index
306 * tlock id = 0 is reserved.
307 */
308 size = sizeof(struct tlock) * nTxLock;
309 TxLock = vmalloc(size);
310 if (TxLock == NULL) {
311 vfree(TxBlock);
312 return -ENOMEM;
313 }
314
315 /* initialize tlock table */
316 for (k = 1; k < nTxLock - 1; k++)
317 TxLock[k].next = k + 1;
318 TxLock[k].next = 0;
319 init_waitqueue_head(&TxAnchor.freelockwait);
320 init_waitqueue_head(&TxAnchor.lowlockwait);
321
322 TxAnchor.freelock = 1;
323 TxAnchor.tlocksInUse = 0;
324 INIT_LIST_HEAD(&TxAnchor.anon_list);
325 INIT_LIST_HEAD(&TxAnchor.anon_list2);
326
327 LAZY_LOCK_INIT();
328 INIT_LIST_HEAD(&TxAnchor.unlock_queue);
329
330 stattx.maxlid = 1; /* statistics */
331
332 return 0;
333}
334
335/*
336 * NAME: txExit()
337 *
338 * FUNCTION: clean up when module is unloaded
339 */
340void txExit(void)
341{
342 vfree(TxLock);
343 TxLock = NULL;
344 vfree(TxBlock);
345 TxBlock = NULL;
346}
347
348/*
349 * NAME: txBegin()
350 *
351 * FUNCTION: start a transaction.
352 *
353 * PARAMETER: sb - superblock
354 * flag - force for nested tx;
355 *
356 * RETURN: tid - transaction id
357 *
358 * note: flag force allows to start tx for nested tx
359 * to prevent deadlock on logsync barrier;
360 */
361tid_t txBegin(struct super_block *sb, int flag)
362{
363 tid_t t;
364 struct tblock *tblk;
365 struct jfs_log *log;
366
367 jfs_info("txBegin: flag = 0x%x", flag);
368 log = JFS_SBI(sb)->log;
369
370 TXN_LOCK();
371
372 INCREMENT(TxStat.txBegin);
373
374 retry:
375 if (!(flag & COMMIT_FORCE)) {
376 /*
377 * synchronize with logsync barrier
378 */
379 if (test_bit(log_SYNCBARRIER, &log->flag) ||
380 test_bit(log_QUIESCE, &log->flag)) {
381 INCREMENT(TxStat.txBegin_barrier);
382 TXN_SLEEP(&log->syncwait);
383 goto retry;
384 }
385 }
386 if (flag == 0) {
387 /*
388 * Don't begin transaction if we're getting starved for tlocks
389 * unless COMMIT_FORCE or COMMIT_INODE (which may ultimately
390 * free tlocks)
391 */
392 if (TxAnchor.tlocksInUse > TxLockVHWM) {
393 INCREMENT(TxStat.txBegin_lockslow);
394 TXN_SLEEP(&TxAnchor.lowlockwait);
395 goto retry;
396 }
397 }
398
399 /*
400 * allocate transaction id/block
401 */
402 if ((t = TxAnchor.freetid) == 0) {
403 jfs_info("txBegin: waiting for free tid");
404 INCREMENT(TxStat.txBegin_freetid);
405 TXN_SLEEP(&TxAnchor.freewait);
406 goto retry;
407 }
408
409 tblk = tid_to_tblock(t);
410
411 if ((tblk->next == 0) && !(flag & COMMIT_FORCE)) {
412 /* Don't let a non-forced transaction take the last tblk */
413 jfs_info("txBegin: waiting for free tid");
414 INCREMENT(TxStat.txBegin_freetid);
415 TXN_SLEEP(&TxAnchor.freewait);
416 goto retry;
417 }
418
419 TxAnchor.freetid = tblk->next;
420
421 /*
422 * initialize transaction
423 */
424
425 /*
426 * We can't zero the whole thing or we screw up another thread being
427 * awakened after sleeping on tblk->waitor
428 *
429 * memset(tblk, 0, sizeof(struct tblock));
430 */
431 tblk->next = tblk->last = tblk->xflag = tblk->flag = tblk->lsn = 0;
432
433 tblk->sb = sb;
434 ++log->logtid;
435 tblk->logtid = log->logtid;
436
437 ++log->active;
438
439 HIGHWATERMARK(stattx.maxtid, t); /* statistics */
440 INCREMENT(stattx.ntid); /* statistics */
441
442 TXN_UNLOCK();
443
444 jfs_info("txBegin: returning tid = %d", t);
445
446 return t;
447}
448
449/*
450 * NAME: txBeginAnon()
451 *
452 * FUNCTION: start an anonymous transaction.
453 * Blocks if logsync or available tlocks are low to prevent
454 * anonymous tlocks from depleting supply.
455 *
456 * PARAMETER: sb - superblock
457 *
458 * RETURN: none
459 */
460void txBeginAnon(struct super_block *sb)
461{
462 struct jfs_log *log;
463
464 log = JFS_SBI(sb)->log;
465
466 TXN_LOCK();
467 INCREMENT(TxStat.txBeginAnon);
468
469 retry:
470 /*
471 * synchronize with logsync barrier
472 */
473 if (test_bit(log_SYNCBARRIER, &log->flag) ||
474 test_bit(log_QUIESCE, &log->flag)) {
475 INCREMENT(TxStat.txBeginAnon_barrier);
476 TXN_SLEEP(&log->syncwait);
477 goto retry;
478 }
479
480 /*
481 * Don't begin transaction if we're getting starved for tlocks
482 */
483 if (TxAnchor.tlocksInUse > TxLockVHWM) {
484 INCREMENT(TxStat.txBeginAnon_lockslow);
485 TXN_SLEEP(&TxAnchor.lowlockwait);
486 goto retry;
487 }
488 TXN_UNLOCK();
489}
490
491/*
492 * txEnd()
493 *
494 * function: free specified transaction block.
495 *
496 * logsync barrier processing:
497 *
498 * serialization:
499 */
500void txEnd(tid_t tid)
501{
502 struct tblock *tblk = tid_to_tblock(tid);
503 struct jfs_log *log;
504
505 jfs_info("txEnd: tid = %d", tid);
506 TXN_LOCK();
507
508 /*
509 * wakeup transactions waiting on the page locked
510 * by the current transaction
511 */
512 TXN_WAKEUP(&tblk->waitor);
513
514 log = JFS_SBI(tblk->sb)->log;
515
516 /*
517 * Lazy commit thread can't free this guy until we mark it UNLOCKED,
518 * otherwise, we would be left with a transaction that may have been
519 * reused.
520 *
521 * Lazy commit thread will turn off tblkGC_LAZY before calling this
522 * routine.
523 */
524 if (tblk->flag & tblkGC_LAZY) {
525 jfs_info("txEnd called w/lazy tid: %d, tblk = 0x%p", tid, tblk);
526 TXN_UNLOCK();
527
528 spin_lock_irq(&log->gclock); // LOGGC_LOCK
529 tblk->flag |= tblkGC_UNLOCKED;
530 spin_unlock_irq(&log->gclock); // LOGGC_UNLOCK
531 return;
532 }
533
534 jfs_info("txEnd: tid: %d, tblk = 0x%p", tid, tblk);
535
536 assert(tblk->next == 0);
537
538 /*
539 * insert tblock back on freelist
540 */
541 tblk->next = TxAnchor.freetid;
542 TxAnchor.freetid = tid;
543
544 /*
545 * mark the tblock not active
546 */
547 if (--log->active == 0) {
548 clear_bit(log_FLUSH, &log->flag);
549
550 /*
551 * synchronize with logsync barrier
552 */
553 if (test_bit(log_SYNCBARRIER, &log->flag)) {
554 TXN_UNLOCK();
555
556 /* write dirty metadata & forward log syncpt */
557 jfs_syncpt(log, 1);
558
559 jfs_info("log barrier off: 0x%x", log->lsn);
560
561 /* enable new transactions start */
562 clear_bit(log_SYNCBARRIER, &log->flag);
563
564 /* wakeup all waitors for logsync barrier */
565 TXN_WAKEUP(&log->syncwait);
566
567 goto wakeup;
568 }
569 }
570
571 TXN_UNLOCK();
572wakeup:
573 /*
574 * wakeup all waitors for a free tblock
575 */
576 TXN_WAKEUP(&TxAnchor.freewait);
577}
578
579/*
580 * txLock()
581 *
582 * function: acquire a transaction lock on the specified <mp>
583 *
584 * parameter:
585 *
586 * return: transaction lock id
587 *
588 * serialization:
589 */
590struct tlock *txLock(tid_t tid, struct inode *ip, struct metapage * mp,
591 int type)
592{
593 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
594 int dir_xtree = 0;
595 lid_t lid;
596 tid_t xtid;
597 struct tlock *tlck;
598 struct xtlock *xtlck;
599 struct linelock *linelock;
600 xtpage_t *p;
601 struct tblock *tblk;
602
603 TXN_LOCK();
604
605 if (S_ISDIR(ip->i_mode) && (type & tlckXTREE) &&
606 !(mp->xflag & COMMIT_PAGE)) {
607 /*
608 * Directory inode is special. It can have both an xtree tlock
609 * and a dtree tlock associated with it.
610 */
611 dir_xtree = 1;
612 lid = jfs_ip->xtlid;
613 } else
614 lid = mp->lid;
615
616 /* is page not locked by a transaction ? */
617 if (lid == 0)
618 goto allocateLock;
619
620 jfs_info("txLock: tid:%d ip:0x%p mp:0x%p lid:%d", tid, ip, mp, lid);
621
622 /* is page locked by the requester transaction ? */
623 tlck = lid_to_tlock(lid);
624 if ((xtid = tlck->tid) == tid) {
625 TXN_UNLOCK();
626 goto grantLock;
627 }
628
629 /*
630 * is page locked by anonymous transaction/lock ?
631 *
632 * (page update without transaction (i.e., file write) is
633 * locked under anonymous transaction tid = 0:
634 * anonymous tlocks maintained on anonymous tlock list of
635 * the inode of the page and available to all anonymous
636 * transactions until txCommit() time at which point
637 * they are transferred to the transaction tlock list of
638 * the committing transaction of the inode)
639 */
640 if (xtid == 0) {
641 tlck->tid = tid;
642 TXN_UNLOCK();
643 tblk = tid_to_tblock(tid);
644 /*
645 * The order of the tlocks in the transaction is important
646 * (during truncate, child xtree pages must be freed before
647 * parent's tlocks change the working map).
648 * Take tlock off anonymous list and add to tail of
649 * transaction list
650 *
651 * Note: We really need to get rid of the tid & lid and
652 * use list_head's. This code is getting UGLY!
653 */
654 if (jfs_ip->atlhead == lid) {
655 if (jfs_ip->atltail == lid) {
656 /* only anonymous txn.
657 * Remove from anon_list
658 */
659 TXN_LOCK();
660 list_del_init(&jfs_ip->anon_inode_list);
661 TXN_UNLOCK();
662 }
663 jfs_ip->atlhead = tlck->next;
664 } else {
665 lid_t last;
666 for (last = jfs_ip->atlhead;
667 lid_to_tlock(last)->next != lid;
668 last = lid_to_tlock(last)->next) {
669 assert(last);
670 }
671 lid_to_tlock(last)->next = tlck->next;
672 if (jfs_ip->atltail == lid)
673 jfs_ip->atltail = last;
674 }
675
676 /* insert the tlock at tail of transaction tlock list */
677
678 if (tblk->next)
679 lid_to_tlock(tblk->last)->next = lid;
680 else
681 tblk->next = lid;
682 tlck->next = 0;
683 tblk->last = lid;
684
685 goto grantLock;
686 }
687
688 goto waitLock;
689
690 /*
691 * allocate a tlock
692 */
693 allocateLock:
694 lid = txLockAlloc();
695 tlck = lid_to_tlock(lid);
696
697 /*
698 * initialize tlock
699 */
700 tlck->tid = tid;
701
702 TXN_UNLOCK();
703
704 /* mark tlock for meta-data page */
705 if (mp->xflag & COMMIT_PAGE) {
706
707 tlck->flag = tlckPAGELOCK;
708
709 /* mark the page dirty and nohomeok */
710 metapage_nohomeok(mp);
711
712 jfs_info("locking mp = 0x%p, nohomeok = %d tid = %d tlck = 0x%p",
713 mp, mp->nohomeok, tid, tlck);
714
715 /* if anonymous transaction, and buffer is on the group
716 * commit synclist, mark inode to show this. This will
717 * prevent the buffer from being marked nohomeok for too
718 * long a time.
719 */
720 if ((tid == 0) && mp->lsn)
721 set_cflag(COMMIT_Synclist, ip);
722 }
723 /* mark tlock for in-memory inode */
724 else
725 tlck->flag = tlckINODELOCK;
726
727 if (S_ISDIR(ip->i_mode))
728 tlck->flag |= tlckDIRECTORY;
729
730 tlck->type = 0;
731
732 /* bind the tlock and the page */
733 tlck->ip = ip;
734 tlck->mp = mp;
735 if (dir_xtree)
736 jfs_ip->xtlid = lid;
737 else
738 mp->lid = lid;
739
740 /*
741 * enqueue transaction lock to transaction/inode
742 */
743 /* insert the tlock at tail of transaction tlock list */
744 if (tid) {
745 tblk = tid_to_tblock(tid);
746 if (tblk->next)
747 lid_to_tlock(tblk->last)->next = lid;
748 else
749 tblk->next = lid;
750 tlck->next = 0;
751 tblk->last = lid;
752 }
753 /* anonymous transaction:
754 * insert the tlock at head of inode anonymous tlock list
755 */
756 else {
757 tlck->next = jfs_ip->atlhead;
758 jfs_ip->atlhead = lid;
759 if (tlck->next == 0) {
760 /* This inode's first anonymous transaction */
761 jfs_ip->atltail = lid;
762 TXN_LOCK();
763 list_add_tail(&jfs_ip->anon_inode_list,
764 &TxAnchor.anon_list);
765 TXN_UNLOCK();
766 }
767 }
768
769 /* initialize type dependent area for linelock */
770 linelock = (struct linelock *) & tlck->lock;
771 linelock->next = 0;
772 linelock->flag = tlckLINELOCK;
773 linelock->maxcnt = TLOCKSHORT;
774 linelock->index = 0;
775
776 switch (type & tlckTYPE) {
777 case tlckDTREE:
778 linelock->l2linesize = L2DTSLOTSIZE;
779 break;
780
781 case tlckXTREE:
782 linelock->l2linesize = L2XTSLOTSIZE;
783
784 xtlck = (struct xtlock *) linelock;
785 xtlck->header.offset = 0;
786 xtlck->header.length = 2;
787
788 if (type & tlckNEW) {
789 xtlck->lwm.offset = XTENTRYSTART;
790 } else {
791 if (mp->xflag & COMMIT_PAGE)
792 p = (xtpage_t *) mp->data;
793 else
794 p = &jfs_ip->i_xtroot;
795 xtlck->lwm.offset =
796 le16_to_cpu(p->header.nextindex);
797 }
798 xtlck->lwm.length = 0; /* ! */
799 xtlck->twm.offset = 0;
800 xtlck->hwm.offset = 0;
801
802 xtlck->index = 2;
803 break;
804
805 case tlckINODE:
806 linelock->l2linesize = L2INODESLOTSIZE;
807 break;
808
809 case tlckDATA:
810 linelock->l2linesize = L2DATASLOTSIZE;
811 break;
812
813 default:
814 jfs_err("UFO tlock:0x%p", tlck);
815 }
816
817 /*
818 * update tlock vector
819 */
820 grantLock:
821 tlck->type |= type;
822
823 return tlck;
824
825 /*
826 * page is being locked by another transaction:
827 */
828 waitLock:
829 /* Only locks on ipimap or ipaimap should reach here */
830 /* assert(jfs_ip->fileset == AGGREGATE_I); */
831 if (jfs_ip->fileset != AGGREGATE_I) {
832 printk(KERN_ERR "txLock: trying to lock locked page!");
833 print_hex_dump(KERN_ERR, "ip: ", DUMP_PREFIX_ADDRESS, 16, 4,
834 ip, sizeof(*ip), 0);
835 print_hex_dump(KERN_ERR, "mp: ", DUMP_PREFIX_ADDRESS, 16, 4,
836 mp, sizeof(*mp), 0);
837 print_hex_dump(KERN_ERR, "Locker's tblock: ",
838 DUMP_PREFIX_ADDRESS, 16, 4, tid_to_tblock(tid),
839 sizeof(struct tblock), 0);
840 print_hex_dump(KERN_ERR, "Tlock: ", DUMP_PREFIX_ADDRESS, 16, 4,
841 tlck, sizeof(*tlck), 0);
842 BUG();
843 }
844 INCREMENT(stattx.waitlock); /* statistics */
845 TXN_UNLOCK();
846 release_metapage(mp);
847 TXN_LOCK();
848 xtid = tlck->tid; /* reacquire after dropping TXN_LOCK */
849
850 jfs_info("txLock: in waitLock, tid = %d, xtid = %d, lid = %d",
851 tid, xtid, lid);
852
853 /* Recheck everything since dropping TXN_LOCK */
854 if (xtid && (tlck->mp == mp) && (mp->lid == lid))
855 TXN_SLEEP_DROP_LOCK(&tid_to_tblock(xtid)->waitor);
856 else
857 TXN_UNLOCK();
858 jfs_info("txLock: awakened tid = %d, lid = %d", tid, lid);
859
860 return NULL;
861}
862
863/*
864 * NAME: txRelease()
865 *
866 * FUNCTION: Release buffers associated with transaction locks, but don't
867 * mark homeok yet. The allows other transactions to modify
868 * buffers, but won't let them go to disk until commit record
869 * actually gets written.
870 *
871 * PARAMETER:
872 * tblk -
873 *
874 * RETURN: Errors from subroutines.
875 */
876static void txRelease(struct tblock * tblk)
877{
878 struct metapage *mp;
879 lid_t lid;
880 struct tlock *tlck;
881
882 TXN_LOCK();
883
884 for (lid = tblk->next; lid; lid = tlck->next) {
885 tlck = lid_to_tlock(lid);
886 if ((mp = tlck->mp) != NULL &&
887 (tlck->type & tlckBTROOT) == 0) {
888 assert(mp->xflag & COMMIT_PAGE);
889 mp->lid = 0;
890 }
891 }
892
893 /*
894 * wakeup transactions waiting on a page locked
895 * by the current transaction
896 */
897 TXN_WAKEUP(&tblk->waitor);
898
899 TXN_UNLOCK();
900}
901
902/*
903 * NAME: txUnlock()
904 *
905 * FUNCTION: Initiates pageout of pages modified by tid in journalled
906 * objects and frees their lockwords.
907 */
908static void txUnlock(struct tblock * tblk)
909{
910 struct tlock *tlck;
911 struct linelock *linelock;
912 lid_t lid, next, llid, k;
913 struct metapage *mp;
914 struct jfs_log *log;
915 int difft, diffp;
916 unsigned long flags;
917
918 jfs_info("txUnlock: tblk = 0x%p", tblk);
919 log = JFS_SBI(tblk->sb)->log;
920
921 /*
922 * mark page under tlock homeok (its log has been written):
923 */
924 for (lid = tblk->next; lid; lid = next) {
925 tlck = lid_to_tlock(lid);
926 next = tlck->next;
927
928 jfs_info("unlocking lid = %d, tlck = 0x%p", lid, tlck);
929
930 /* unbind page from tlock */
931 if ((mp = tlck->mp) != NULL &&
932 (tlck->type & tlckBTROOT) == 0) {
933 assert(mp->xflag & COMMIT_PAGE);
934
935 /* hold buffer
936 */
937 hold_metapage(mp);
938
939 assert(mp->nohomeok > 0);
940 _metapage_homeok(mp);
941
942 /* inherit younger/larger clsn */
943 LOGSYNC_LOCK(log, flags);
944 if (mp->clsn) {
945 logdiff(difft, tblk->clsn, log);
946 logdiff(diffp, mp->clsn, log);
947 if (difft > diffp)
948 mp->clsn = tblk->clsn;
949 } else
950 mp->clsn = tblk->clsn;
951 LOGSYNC_UNLOCK(log, flags);
952
953 assert(!(tlck->flag & tlckFREEPAGE));
954
955 put_metapage(mp);
956 }
957
958 /* insert tlock, and linelock(s) of the tlock if any,
959 * at head of freelist
960 */
961 TXN_LOCK();
962
963 llid = ((struct linelock *) & tlck->lock)->next;
964 while (llid) {
965 linelock = (struct linelock *) lid_to_tlock(llid);
966 k = linelock->next;
967 txLockFree(llid);
968 llid = k;
969 }
970 txLockFree(lid);
971
972 TXN_UNLOCK();
973 }
974 tblk->next = tblk->last = 0;
975
976 /*
977 * remove tblock from logsynclist
978 * (allocation map pages inherited lsn of tblk and
979 * has been inserted in logsync list at txUpdateMap())
980 */
981 if (tblk->lsn) {
982 LOGSYNC_LOCK(log, flags);
983 log->count--;
984 list_del(&tblk->synclist);
985 LOGSYNC_UNLOCK(log, flags);
986 }
987}
988
989/*
990 * txMaplock()
991 *
992 * function: allocate a transaction lock for freed page/entry;
993 * for freed page, maplock is used as xtlock/dtlock type;
994 */
995struct tlock *txMaplock(tid_t tid, struct inode *ip, int type)
996{
997 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
998 lid_t lid;
999 struct tblock *tblk;
1000 struct tlock *tlck;
1001 struct maplock *maplock;
1002
1003 TXN_LOCK();
1004
1005 /*
1006 * allocate a tlock
1007 */
1008 lid = txLockAlloc();
1009 tlck = lid_to_tlock(lid);
1010
1011 /*
1012 * initialize tlock
1013 */
1014 tlck->tid = tid;
1015
1016 /* bind the tlock and the object */
1017 tlck->flag = tlckINODELOCK;
1018 if (S_ISDIR(ip->i_mode))
1019 tlck->flag |= tlckDIRECTORY;
1020 tlck->ip = ip;
1021 tlck->mp = NULL;
1022
1023 tlck->type = type;
1024
1025 /*
1026 * enqueue transaction lock to transaction/inode
1027 */
1028 /* insert the tlock at tail of transaction tlock list */
1029 if (tid) {
1030 tblk = tid_to_tblock(tid);
1031 if (tblk->next)
1032 lid_to_tlock(tblk->last)->next = lid;
1033 else
1034 tblk->next = lid;
1035 tlck->next = 0;
1036 tblk->last = lid;
1037 }
1038 /* anonymous transaction:
1039 * insert the tlock at head of inode anonymous tlock list
1040 */
1041 else {
1042 tlck->next = jfs_ip->atlhead;
1043 jfs_ip->atlhead = lid;
1044 if (tlck->next == 0) {
1045 /* This inode's first anonymous transaction */
1046 jfs_ip->atltail = lid;
1047 list_add_tail(&jfs_ip->anon_inode_list,
1048 &TxAnchor.anon_list);
1049 }
1050 }
1051
1052 TXN_UNLOCK();
1053
1054 /* initialize type dependent area for maplock */
1055 maplock = (struct maplock *) & tlck->lock;
1056 maplock->next = 0;
1057 maplock->maxcnt = 0;
1058 maplock->index = 0;
1059
1060 return tlck;
1061}
1062
1063/*
1064 * txLinelock()
1065 *
1066 * function: allocate a transaction lock for log vector list
1067 */
1068struct linelock *txLinelock(struct linelock * tlock)
1069{
1070 lid_t lid;
1071 struct tlock *tlck;
1072 struct linelock *linelock;
1073
1074 TXN_LOCK();
1075
1076 /* allocate a TxLock structure */
1077 lid = txLockAlloc();
1078 tlck = lid_to_tlock(lid);
1079
1080 TXN_UNLOCK();
1081
1082 /* initialize linelock */
1083 linelock = (struct linelock *) tlck;
1084 linelock->next = 0;
1085 linelock->flag = tlckLINELOCK;
1086 linelock->maxcnt = TLOCKLONG;
1087 linelock->index = 0;
1088 if (tlck->flag & tlckDIRECTORY)
1089 linelock->flag |= tlckDIRECTORY;
1090
1091 /* append linelock after tlock */
1092 linelock->next = tlock->next;
1093 tlock->next = lid;
1094
1095 return linelock;
1096}
1097
1098/*
1099 * transaction commit management
1100 * -----------------------------
1101 */
1102
1103/*
1104 * NAME: txCommit()
1105 *
1106 * FUNCTION: commit the changes to the objects specified in
1107 * clist. For journalled segments only the
1108 * changes of the caller are committed, ie by tid.
1109 * for non-journalled segments the data are flushed to
1110 * disk and then the change to the disk inode and indirect
1111 * blocks committed (so blocks newly allocated to the
1112 * segment will be made a part of the segment atomically).
1113 *
1114 * all of the segments specified in clist must be in
1115 * one file system. no more than 6 segments are needed
1116 * to handle all unix svcs.
1117 *
1118 * if the i_nlink field (i.e. disk inode link count)
1119 * is zero, and the type of inode is a regular file or
1120 * directory, or symbolic link , the inode is truncated
1121 * to zero length. the truncation is committed but the
1122 * VM resources are unaffected until it is closed (see
1123 * iput and iclose).
1124 *
1125 * PARAMETER:
1126 *
1127 * RETURN:
1128 *
1129 * serialization:
1130 * on entry the inode lock on each segment is assumed
1131 * to be held.
1132 *
1133 * i/o error:
1134 */
1135int txCommit(tid_t tid, /* transaction identifier */
1136 int nip, /* number of inodes to commit */
1137 struct inode **iplist, /* list of inode to commit */
1138 int flag)
1139{
1140 int rc = 0;
1141 struct commit cd;
1142 struct jfs_log *log;
1143 struct tblock *tblk;
1144 struct lrd *lrd;
1145 struct inode *ip;
1146 struct jfs_inode_info *jfs_ip;
1147 int k, n;
1148 ino_t top;
1149 struct super_block *sb;
1150
1151 jfs_info("txCommit, tid = %d, flag = %d", tid, flag);
1152 /* is read-only file system ? */
1153 if (isReadOnly(iplist[0])) {
1154 rc = -EROFS;
1155 goto TheEnd;
1156 }
1157
1158 sb = cd.sb = iplist[0]->i_sb;
1159 cd.tid = tid;
1160
1161 if (tid == 0)
1162 tid = txBegin(sb, 0);
1163 tblk = tid_to_tblock(tid);
1164
1165 /*
1166 * initialize commit structure
1167 */
1168 log = JFS_SBI(sb)->log;
1169 cd.log = log;
1170
1171 /* initialize log record descriptor in commit */
1172 lrd = &cd.lrd;
1173 lrd->logtid = cpu_to_le32(tblk->logtid);
1174 lrd->backchain = 0;
1175
1176 tblk->xflag |= flag;
1177
1178 if ((flag & (COMMIT_FORCE | COMMIT_SYNC)) == 0)
1179 tblk->xflag |= COMMIT_LAZY;
1180 /*
1181 * prepare non-journaled objects for commit
1182 *
1183 * flush data pages of non-journaled file
1184 * to prevent the file getting non-initialized disk blocks
1185 * in case of crash.
1186 * (new blocks - )
1187 */
1188 cd.iplist = iplist;
1189 cd.nip = nip;
1190
1191 /*
1192 * acquire transaction lock on (on-disk) inodes
1193 *
1194 * update on-disk inode from in-memory inode
1195 * acquiring transaction locks for AFTER records
1196 * on the on-disk inode of file object
1197 *
1198 * sort the inodes array by inode number in descending order
1199 * to prevent deadlock when acquiring transaction lock
1200 * of on-disk inodes on multiple on-disk inode pages by
1201 * multiple concurrent transactions
1202 */
1203 for (k = 0; k < cd.nip; k++) {
1204 top = (cd.iplist[k])->i_ino;
1205 for (n = k + 1; n < cd.nip; n++) {
1206 ip = cd.iplist[n];
1207 if (ip->i_ino > top) {
1208 top = ip->i_ino;
1209 cd.iplist[n] = cd.iplist[k];
1210 cd.iplist[k] = ip;
1211 }
1212 }
1213
1214 ip = cd.iplist[k];
1215 jfs_ip = JFS_IP(ip);
1216
1217 /*
1218 * BUGBUG - This code has temporarily been removed. The
1219 * intent is to ensure that any file data is written before
1220 * the metadata is committed to the journal. This prevents
1221 * uninitialized data from appearing in a file after the
1222 * journal has been replayed. (The uninitialized data
1223 * could be sensitive data removed by another user.)
1224 *
1225 * The problem now is that we are holding the IWRITELOCK
1226 * on the inode, and calling filemap_fdatawrite on an
1227 * unmapped page will cause a deadlock in jfs_get_block.
1228 *
1229 * The long term solution is to pare down the use of
1230 * IWRITELOCK. We are currently holding it too long.
1231 * We could also be smarter about which data pages need
1232 * to be written before the transaction is committed and
1233 * when we don't need to worry about it at all.
1234 *
1235 * if ((!S_ISDIR(ip->i_mode))
1236 * && (tblk->flag & COMMIT_DELETE) == 0)
1237 * filemap_write_and_wait(ip->i_mapping);
1238 */
1239
1240 /*
1241 * Mark inode as not dirty. It will still be on the dirty
1242 * inode list, but we'll know not to commit it again unless
1243 * it gets marked dirty again
1244 */
1245 clear_cflag(COMMIT_Dirty, ip);
1246
1247 /* inherit anonymous tlock(s) of inode */
1248 if (jfs_ip->atlhead) {
1249 lid_to_tlock(jfs_ip->atltail)->next = tblk->next;
1250 tblk->next = jfs_ip->atlhead;
1251 if (!tblk->last)
1252 tblk->last = jfs_ip->atltail;
1253 jfs_ip->atlhead = jfs_ip->atltail = 0;
1254 TXN_LOCK();
1255 list_del_init(&jfs_ip->anon_inode_list);
1256 TXN_UNLOCK();
1257 }
1258
1259 /*
1260 * acquire transaction lock on on-disk inode page
1261 * (become first tlock of the tblk's tlock list)
1262 */
1263 if (((rc = diWrite(tid, ip))))
1264 goto out;
1265 }
1266
1267 /*
1268 * write log records from transaction locks
1269 *
1270 * txUpdateMap() resets XAD_NEW in XAD.
1271 */
1272 if ((rc = txLog(log, tblk, &cd)))
1273 goto TheEnd;
1274
1275 /*
1276 * Ensure that inode isn't reused before
1277 * lazy commit thread finishes processing
1278 */
1279 if (tblk->xflag & COMMIT_DELETE) {
1280 ihold(tblk->u.ip);
1281 /*
1282 * Avoid a rare deadlock
1283 *
1284 * If the inode is locked, we may be blocked in
1285 * jfs_commit_inode. If so, we don't want the
1286 * lazy_commit thread doing the last iput() on the inode
1287 * since that may block on the locked inode. Instead,
1288 * commit the transaction synchronously, so the last iput
1289 * will be done by the calling thread (or later)
1290 */
1291 /*
1292 * I believe this code is no longer needed. Splitting I_LOCK
1293 * into two bits, I_NEW and I_SYNC should prevent this
1294 * deadlock as well. But since I don't have a JFS testload
1295 * to verify this, only a trivial s/I_LOCK/I_SYNC/ was done.
1296 * Joern
1297 */
1298 if (tblk->u.ip->i_state & I_SYNC)
1299 tblk->xflag &= ~COMMIT_LAZY;
1300 }
1301
1302 ASSERT((!(tblk->xflag & COMMIT_DELETE)) ||
1303 ((tblk->u.ip->i_nlink == 0) &&
1304 !test_cflag(COMMIT_Nolink, tblk->u.ip)));
1305
1306 /*
1307 * write COMMIT log record
1308 */
1309 lrd->type = cpu_to_le16(LOG_COMMIT);
1310 lrd->length = 0;
1311 lmLog(log, tblk, lrd, NULL);
1312
1313 lmGroupCommit(log, tblk);
1314
1315 /*
1316 * - transaction is now committed -
1317 */
1318
1319 /*
1320 * force pages in careful update
1321 * (imap addressing structure update)
1322 */
1323 if (flag & COMMIT_FORCE)
1324 txForce(tblk);
1325
1326 /*
1327 * update allocation map.
1328 *
1329 * update inode allocation map and inode:
1330 * free pager lock on memory object of inode if any.
1331 * update block allocation map.
1332 *
1333 * txUpdateMap() resets XAD_NEW in XAD.
1334 */
1335 if (tblk->xflag & COMMIT_FORCE)
1336 txUpdateMap(tblk);
1337
1338 /*
1339 * free transaction locks and pageout/free pages
1340 */
1341 txRelease(tblk);
1342
1343 if ((tblk->flag & tblkGC_LAZY) == 0)
1344 txUnlock(tblk);
1345
1346
1347 /*
1348 * reset in-memory object state
1349 */
1350 for (k = 0; k < cd.nip; k++) {
1351 ip = cd.iplist[k];
1352 jfs_ip = JFS_IP(ip);
1353
1354 /*
1355 * reset in-memory inode state
1356 */
1357 jfs_ip->bxflag = 0;
1358 jfs_ip->blid = 0;
1359 }
1360
1361 out:
1362 if (rc != 0)
1363 txAbort(tid, 1);
1364
1365 TheEnd:
1366 jfs_info("txCommit: tid = %d, returning %d", tid, rc);
1367 return rc;
1368}
1369
1370/*
1371 * NAME: txLog()
1372 *
1373 * FUNCTION: Writes AFTER log records for all lines modified
1374 * by tid for segments specified by inodes in comdata.
1375 * Code assumes only WRITELOCKS are recorded in lockwords.
1376 *
1377 * PARAMETERS:
1378 *
1379 * RETURN :
1380 */
1381static int txLog(struct jfs_log * log, struct tblock * tblk, struct commit * cd)
1382{
1383 int rc = 0;
1384 struct inode *ip;
1385 lid_t lid;
1386 struct tlock *tlck;
1387 struct lrd *lrd = &cd->lrd;
1388
1389 /*
1390 * write log record(s) for each tlock of transaction,
1391 */
1392 for (lid = tblk->next; lid; lid = tlck->next) {
1393 tlck = lid_to_tlock(lid);
1394
1395 tlck->flag |= tlckLOG;
1396
1397 /* initialize lrd common */
1398 ip = tlck->ip;
1399 lrd->aggregate = cpu_to_le32(JFS_SBI(ip->i_sb)->aggregate);
1400 lrd->log.redopage.fileset = cpu_to_le32(JFS_IP(ip)->fileset);
1401 lrd->log.redopage.inode = cpu_to_le32(ip->i_ino);
1402
1403 /* write log record of page from the tlock */
1404 switch (tlck->type & tlckTYPE) {
1405 case tlckXTREE:
1406 xtLog(log, tblk, lrd, tlck);
1407 break;
1408
1409 case tlckDTREE:
1410 dtLog(log, tblk, lrd, tlck);
1411 break;
1412
1413 case tlckINODE:
1414 diLog(log, tblk, lrd, tlck, cd);
1415 break;
1416
1417 case tlckMAP:
1418 mapLog(log, tblk, lrd, tlck);
1419 break;
1420
1421 case tlckDATA:
1422 dataLog(log, tblk, lrd, tlck);
1423 break;
1424
1425 default:
1426 jfs_err("UFO tlock:0x%p", tlck);
1427 }
1428 }
1429
1430 return rc;
1431}
1432
1433/*
1434 * diLog()
1435 *
1436 * function: log inode tlock and format maplock to update bmap;
1437 */
1438static int diLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1439 struct tlock * tlck, struct commit * cd)
1440{
1441 int rc = 0;
1442 struct metapage *mp;
1443 pxd_t *pxd;
1444 struct pxd_lock *pxdlock;
1445
1446 mp = tlck->mp;
1447
1448 /* initialize as REDOPAGE record format */
1449 lrd->log.redopage.type = cpu_to_le16(LOG_INODE);
1450 lrd->log.redopage.l2linesize = cpu_to_le16(L2INODESLOTSIZE);
1451
1452 pxd = &lrd->log.redopage.pxd;
1453
1454 /*
1455 * inode after image
1456 */
1457 if (tlck->type & tlckENTRY) {
1458 /* log after-image for logredo(): */
1459 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1460 PXDaddress(pxd, mp->index);
1461 PXDlength(pxd,
1462 mp->logical_size >> tblk->sb->s_blocksize_bits);
1463 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1464
1465 /* mark page as homeward bound */
1466 tlck->flag |= tlckWRITEPAGE;
1467 } else if (tlck->type & tlckFREE) {
1468 /*
1469 * free inode extent
1470 *
1471 * (pages of the freed inode extent have been invalidated and
1472 * a maplock for free of the extent has been formatted at
1473 * txLock() time);
1474 *
1475 * the tlock had been acquired on the inode allocation map page
1476 * (iag) that specifies the freed extent, even though the map
1477 * page is not itself logged, to prevent pageout of the map
1478 * page before the log;
1479 */
1480
1481 /* log LOG_NOREDOINOEXT of the freed inode extent for
1482 * logredo() to start NoRedoPage filters, and to update
1483 * imap and bmap for free of the extent;
1484 */
1485 lrd->type = cpu_to_le16(LOG_NOREDOINOEXT);
1486 /*
1487 * For the LOG_NOREDOINOEXT record, we need
1488 * to pass the IAG number and inode extent
1489 * index (within that IAG) from which the
1490 * the extent being released. These have been
1491 * passed to us in the iplist[1] and iplist[2].
1492 */
1493 lrd->log.noredoinoext.iagnum =
1494 cpu_to_le32((u32) (size_t) cd->iplist[1]);
1495 lrd->log.noredoinoext.inoext_idx =
1496 cpu_to_le32((u32) (size_t) cd->iplist[2]);
1497
1498 pxdlock = (struct pxd_lock *) & tlck->lock;
1499 *pxd = pxdlock->pxd;
1500 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1501
1502 /* update bmap */
1503 tlck->flag |= tlckUPDATEMAP;
1504
1505 /* mark page as homeward bound */
1506 tlck->flag |= tlckWRITEPAGE;
1507 } else
1508 jfs_err("diLog: UFO type tlck:0x%p", tlck);
1509#ifdef _JFS_WIP
1510 /*
1511 * alloc/free external EA extent
1512 *
1513 * a maplock for txUpdateMap() to update bPWMAP for alloc/free
1514 * of the extent has been formatted at txLock() time;
1515 */
1516 else {
1517 assert(tlck->type & tlckEA);
1518
1519 /* log LOG_UPDATEMAP for logredo() to update bmap for
1520 * alloc of new (and free of old) external EA extent;
1521 */
1522 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1523 pxdlock = (struct pxd_lock *) & tlck->lock;
1524 nlock = pxdlock->index;
1525 for (i = 0; i < nlock; i++, pxdlock++) {
1526 if (pxdlock->flag & mlckALLOCPXD)
1527 lrd->log.updatemap.type =
1528 cpu_to_le16(LOG_ALLOCPXD);
1529 else
1530 lrd->log.updatemap.type =
1531 cpu_to_le16(LOG_FREEPXD);
1532 lrd->log.updatemap.nxd = cpu_to_le16(1);
1533 lrd->log.updatemap.pxd = pxdlock->pxd;
1534 lrd->backchain =
1535 cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1536 }
1537
1538 /* update bmap */
1539 tlck->flag |= tlckUPDATEMAP;
1540 }
1541#endif /* _JFS_WIP */
1542
1543 return rc;
1544}
1545
1546/*
1547 * dataLog()
1548 *
1549 * function: log data tlock
1550 */
1551static int dataLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1552 struct tlock * tlck)
1553{
1554 struct metapage *mp;
1555 pxd_t *pxd;
1556
1557 mp = tlck->mp;
1558
1559 /* initialize as REDOPAGE record format */
1560 lrd->log.redopage.type = cpu_to_le16(LOG_DATA);
1561 lrd->log.redopage.l2linesize = cpu_to_le16(L2DATASLOTSIZE);
1562
1563 pxd = &lrd->log.redopage.pxd;
1564
1565 /* log after-image for logredo(): */
1566 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1567
1568 if (jfs_dirtable_inline(tlck->ip)) {
1569 /*
1570 * The table has been truncated, we've must have deleted
1571 * the last entry, so don't bother logging this
1572 */
1573 mp->lid = 0;
1574 grab_metapage(mp);
1575 metapage_homeok(mp);
1576 discard_metapage(mp);
1577 tlck->mp = NULL;
1578 return 0;
1579 }
1580
1581 PXDaddress(pxd, mp->index);
1582 PXDlength(pxd, mp->logical_size >> tblk->sb->s_blocksize_bits);
1583
1584 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1585
1586 /* mark page as homeward bound */
1587 tlck->flag |= tlckWRITEPAGE;
1588
1589 return 0;
1590}
1591
1592/*
1593 * dtLog()
1594 *
1595 * function: log dtree tlock and format maplock to update bmap;
1596 */
1597static void dtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1598 struct tlock * tlck)
1599{
1600 struct metapage *mp;
1601 struct pxd_lock *pxdlock;
1602 pxd_t *pxd;
1603
1604 mp = tlck->mp;
1605
1606 /* initialize as REDOPAGE/NOREDOPAGE record format */
1607 lrd->log.redopage.type = cpu_to_le16(LOG_DTREE);
1608 lrd->log.redopage.l2linesize = cpu_to_le16(L2DTSLOTSIZE);
1609
1610 pxd = &lrd->log.redopage.pxd;
1611
1612 if (tlck->type & tlckBTROOT)
1613 lrd->log.redopage.type |= cpu_to_le16(LOG_BTROOT);
1614
1615 /*
1616 * page extension via relocation: entry insertion;
1617 * page extension in-place: entry insertion;
1618 * new right page from page split, reinitialized in-line
1619 * root from root page split: entry insertion;
1620 */
1621 if (tlck->type & (tlckNEW | tlckEXTEND)) {
1622 /* log after-image of the new page for logredo():
1623 * mark log (LOG_NEW) for logredo() to initialize
1624 * freelist and update bmap for alloc of the new page;
1625 */
1626 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1627 if (tlck->type & tlckEXTEND)
1628 lrd->log.redopage.type |= cpu_to_le16(LOG_EXTEND);
1629 else
1630 lrd->log.redopage.type |= cpu_to_le16(LOG_NEW);
1631 PXDaddress(pxd, mp->index);
1632 PXDlength(pxd,
1633 mp->logical_size >> tblk->sb->s_blocksize_bits);
1634 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1635
1636 /* format a maplock for txUpdateMap() to update bPMAP for
1637 * alloc of the new page;
1638 */
1639 if (tlck->type & tlckBTROOT)
1640 return;
1641 tlck->flag |= tlckUPDATEMAP;
1642 pxdlock = (struct pxd_lock *) & tlck->lock;
1643 pxdlock->flag = mlckALLOCPXD;
1644 pxdlock->pxd = *pxd;
1645
1646 pxdlock->index = 1;
1647
1648 /* mark page as homeward bound */
1649 tlck->flag |= tlckWRITEPAGE;
1650 return;
1651 }
1652
1653 /*
1654 * entry insertion/deletion,
1655 * sibling page link update (old right page before split);
1656 */
1657 if (tlck->type & (tlckENTRY | tlckRELINK)) {
1658 /* log after-image for logredo(): */
1659 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1660 PXDaddress(pxd, mp->index);
1661 PXDlength(pxd,
1662 mp->logical_size >> tblk->sb->s_blocksize_bits);
1663 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1664
1665 /* mark page as homeward bound */
1666 tlck->flag |= tlckWRITEPAGE;
1667 return;
1668 }
1669
1670 /*
1671 * page deletion: page has been invalidated
1672 * page relocation: source extent
1673 *
1674 * a maplock for free of the page has been formatted
1675 * at txLock() time);
1676 */
1677 if (tlck->type & (tlckFREE | tlckRELOCATE)) {
1678 /* log LOG_NOREDOPAGE of the deleted page for logredo()
1679 * to start NoRedoPage filter and to update bmap for free
1680 * of the deletd page
1681 */
1682 lrd->type = cpu_to_le16(LOG_NOREDOPAGE);
1683 pxdlock = (struct pxd_lock *) & tlck->lock;
1684 *pxd = pxdlock->pxd;
1685 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1686
1687 /* a maplock for txUpdateMap() for free of the page
1688 * has been formatted at txLock() time;
1689 */
1690 tlck->flag |= tlckUPDATEMAP;
1691 }
1692 return;
1693}
1694
1695/*
1696 * xtLog()
1697 *
1698 * function: log xtree tlock and format maplock to update bmap;
1699 */
1700static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1701 struct tlock * tlck)
1702{
1703 struct inode *ip;
1704 struct metapage *mp;
1705 xtpage_t *p;
1706 struct xtlock *xtlck;
1707 struct maplock *maplock;
1708 struct xdlistlock *xadlock;
1709 struct pxd_lock *pxdlock;
1710 pxd_t *page_pxd;
1711 int next, lwm, hwm;
1712
1713 ip = tlck->ip;
1714 mp = tlck->mp;
1715
1716 /* initialize as REDOPAGE/NOREDOPAGE record format */
1717 lrd->log.redopage.type = cpu_to_le16(LOG_XTREE);
1718 lrd->log.redopage.l2linesize = cpu_to_le16(L2XTSLOTSIZE);
1719
1720 page_pxd = &lrd->log.redopage.pxd;
1721
1722 if (tlck->type & tlckBTROOT) {
1723 lrd->log.redopage.type |= cpu_to_le16(LOG_BTROOT);
1724 p = &JFS_IP(ip)->i_xtroot;
1725 if (S_ISDIR(ip->i_mode))
1726 lrd->log.redopage.type |=
1727 cpu_to_le16(LOG_DIR_XTREE);
1728 } else
1729 p = (xtpage_t *) mp->data;
1730 next = le16_to_cpu(p->header.nextindex);
1731
1732 xtlck = (struct xtlock *) & tlck->lock;
1733
1734 maplock = (struct maplock *) & tlck->lock;
1735 xadlock = (struct xdlistlock *) maplock;
1736
1737 /*
1738 * entry insertion/extension;
1739 * sibling page link update (old right page before split);
1740 */
1741 if (tlck->type & (tlckNEW | tlckGROW | tlckRELINK)) {
1742 /* log after-image for logredo():
1743 * logredo() will update bmap for alloc of new/extended
1744 * extents (XAD_NEW|XAD_EXTEND) of XAD[lwm:next) from
1745 * after-image of XADlist;
1746 * logredo() resets (XAD_NEW|XAD_EXTEND) flag when
1747 * applying the after-image to the meta-data page.
1748 */
1749 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1750 PXDaddress(page_pxd, mp->index);
1751 PXDlength(page_pxd,
1752 mp->logical_size >> tblk->sb->s_blocksize_bits);
1753 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1754
1755 /* format a maplock for txUpdateMap() to update bPMAP
1756 * for alloc of new/extended extents of XAD[lwm:next)
1757 * from the page itself;
1758 * txUpdateMap() resets (XAD_NEW|XAD_EXTEND) flag.
1759 */
1760 lwm = xtlck->lwm.offset;
1761 if (lwm == 0)
1762 lwm = XTPAGEMAXSLOT;
1763
1764 if (lwm == next)
1765 goto out;
1766 if (lwm > next) {
1767 jfs_err("xtLog: lwm > next");
1768 goto out;
1769 }
1770 tlck->flag |= tlckUPDATEMAP;
1771 xadlock->flag = mlckALLOCXADLIST;
1772 xadlock->count = next - lwm;
1773 if ((xadlock->count <= 4) && (tblk->xflag & COMMIT_LAZY)) {
1774 int i;
1775 pxd_t *pxd;
1776 /*
1777 * Lazy commit may allow xtree to be modified before
1778 * txUpdateMap runs. Copy xad into linelock to
1779 * preserve correct data.
1780 *
1781 * We can fit twice as may pxd's as xads in the lock
1782 */
1783 xadlock->flag = mlckALLOCPXDLIST;
1784 pxd = xadlock->xdlist = &xtlck->pxdlock;
1785 for (i = 0; i < xadlock->count; i++) {
1786 PXDaddress(pxd, addressXAD(&p->xad[lwm + i]));
1787 PXDlength(pxd, lengthXAD(&p->xad[lwm + i]));
1788 p->xad[lwm + i].flag &=
1789 ~(XAD_NEW | XAD_EXTENDED);
1790 pxd++;
1791 }
1792 } else {
1793 /*
1794 * xdlist will point to into inode's xtree, ensure
1795 * that transaction is not committed lazily.
1796 */
1797 xadlock->flag = mlckALLOCXADLIST;
1798 xadlock->xdlist = &p->xad[lwm];
1799 tblk->xflag &= ~COMMIT_LAZY;
1800 }
1801 jfs_info("xtLog: alloc ip:0x%p mp:0x%p tlck:0x%p lwm:%d count:%d",
1802 tlck->ip, mp, tlck, lwm, xadlock->count);
1803
1804 maplock->index = 1;
1805
1806 out:
1807 /* mark page as homeward bound */
1808 tlck->flag |= tlckWRITEPAGE;
1809
1810 return;
1811 }
1812
1813 /*
1814 * page deletion: file deletion/truncation (ref. xtTruncate())
1815 *
1816 * (page will be invalidated after log is written and bmap
1817 * is updated from the page);
1818 */
1819 if (tlck->type & tlckFREE) {
1820 /* LOG_NOREDOPAGE log for NoRedoPage filter:
1821 * if page free from file delete, NoRedoFile filter from
1822 * inode image of zero link count will subsume NoRedoPage
1823 * filters for each page;
1824 * if page free from file truncattion, write NoRedoPage
1825 * filter;
1826 *
1827 * upadte of block allocation map for the page itself:
1828 * if page free from deletion and truncation, LOG_UPDATEMAP
1829 * log for the page itself is generated from processing
1830 * its parent page xad entries;
1831 */
1832 /* if page free from file truncation, log LOG_NOREDOPAGE
1833 * of the deleted page for logredo() to start NoRedoPage
1834 * filter for the page;
1835 */
1836 if (tblk->xflag & COMMIT_TRUNCATE) {
1837 /* write NOREDOPAGE for the page */
1838 lrd->type = cpu_to_le16(LOG_NOREDOPAGE);
1839 PXDaddress(page_pxd, mp->index);
1840 PXDlength(page_pxd,
1841 mp->logical_size >> tblk->sb->
1842 s_blocksize_bits);
1843 lrd->backchain =
1844 cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1845
1846 if (tlck->type & tlckBTROOT) {
1847 /* Empty xtree must be logged */
1848 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1849 lrd->backchain =
1850 cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1851 }
1852 }
1853
1854 /* init LOG_UPDATEMAP of the freed extents
1855 * XAD[XTENTRYSTART:hwm) from the deleted page itself
1856 * for logredo() to update bmap;
1857 */
1858 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1859 lrd->log.updatemap.type = cpu_to_le16(LOG_FREEXADLIST);
1860 xtlck = (struct xtlock *) & tlck->lock;
1861 hwm = xtlck->hwm.offset;
1862 lrd->log.updatemap.nxd =
1863 cpu_to_le16(hwm - XTENTRYSTART + 1);
1864 /* reformat linelock for lmLog() */
1865 xtlck->header.offset = XTENTRYSTART;
1866 xtlck->header.length = hwm - XTENTRYSTART + 1;
1867 xtlck->index = 1;
1868 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1869
1870 /* format a maplock for txUpdateMap() to update bmap
1871 * to free extents of XAD[XTENTRYSTART:hwm) from the
1872 * deleted page itself;
1873 */
1874 tlck->flag |= tlckUPDATEMAP;
1875 xadlock->count = hwm - XTENTRYSTART + 1;
1876 if ((xadlock->count <= 4) && (tblk->xflag & COMMIT_LAZY)) {
1877 int i;
1878 pxd_t *pxd;
1879 /*
1880 * Lazy commit may allow xtree to be modified before
1881 * txUpdateMap runs. Copy xad into linelock to
1882 * preserve correct data.
1883 *
1884 * We can fit twice as may pxd's as xads in the lock
1885 */
1886 xadlock->flag = mlckFREEPXDLIST;
1887 pxd = xadlock->xdlist = &xtlck->pxdlock;
1888 for (i = 0; i < xadlock->count; i++) {
1889 PXDaddress(pxd,
1890 addressXAD(&p->xad[XTENTRYSTART + i]));
1891 PXDlength(pxd,
1892 lengthXAD(&p->xad[XTENTRYSTART + i]));
1893 pxd++;
1894 }
1895 } else {
1896 /*
1897 * xdlist will point to into inode's xtree, ensure
1898 * that transaction is not committed lazily.
1899 */
1900 xadlock->flag = mlckFREEXADLIST;
1901 xadlock->xdlist = &p->xad[XTENTRYSTART];
1902 tblk->xflag &= ~COMMIT_LAZY;
1903 }
1904 jfs_info("xtLog: free ip:0x%p mp:0x%p count:%d lwm:2",
1905 tlck->ip, mp, xadlock->count);
1906
1907 maplock->index = 1;
1908
1909 /* mark page as invalid */
1910 if (((tblk->xflag & COMMIT_PWMAP) || S_ISDIR(ip->i_mode))
1911 && !(tlck->type & tlckBTROOT))
1912 tlck->flag |= tlckFREEPAGE;
1913 /*
1914 else (tblk->xflag & COMMIT_PMAP)
1915 ? release the page;
1916 */
1917 return;
1918 }
1919
1920 /*
1921 * page/entry truncation: file truncation (ref. xtTruncate())
1922 *
1923 * |----------+------+------+---------------|
1924 * | | |
1925 * | | hwm - hwm before truncation
1926 * | next - truncation point
1927 * lwm - lwm before truncation
1928 * header ?
1929 */
1930 if (tlck->type & tlckTRUNCATE) {
1931 /* This odd declaration suppresses a bogus gcc warning */
1932 pxd_t pxd = pxd; /* truncated extent of xad */
1933 int twm;
1934
1935 /*
1936 * For truncation the entire linelock may be used, so it would
1937 * be difficult to store xad list in linelock itself.
1938 * Therefore, we'll just force transaction to be committed
1939 * synchronously, so that xtree pages won't be changed before
1940 * txUpdateMap runs.
1941 */
1942 tblk->xflag &= ~COMMIT_LAZY;
1943 lwm = xtlck->lwm.offset;
1944 if (lwm == 0)
1945 lwm = XTPAGEMAXSLOT;
1946 hwm = xtlck->hwm.offset;
1947 twm = xtlck->twm.offset;
1948
1949 /*
1950 * write log records
1951 */
1952 /* log after-image for logredo():
1953 *
1954 * logredo() will update bmap for alloc of new/extended
1955 * extents (XAD_NEW|XAD_EXTEND) of XAD[lwm:next) from
1956 * after-image of XADlist;
1957 * logredo() resets (XAD_NEW|XAD_EXTEND) flag when
1958 * applying the after-image to the meta-data page.
1959 */
1960 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1961 PXDaddress(page_pxd, mp->index);
1962 PXDlength(page_pxd,
1963 mp->logical_size >> tblk->sb->s_blocksize_bits);
1964 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1965
1966 /*
1967 * truncate entry XAD[twm == next - 1]:
1968 */
1969 if (twm == next - 1) {
1970 /* init LOG_UPDATEMAP for logredo() to update bmap for
1971 * free of truncated delta extent of the truncated
1972 * entry XAD[next - 1]:
1973 * (xtlck->pxdlock = truncated delta extent);
1974 */
1975 pxdlock = (struct pxd_lock *) & xtlck->pxdlock;
1976 /* assert(pxdlock->type & tlckTRUNCATE); */
1977 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1978 lrd->log.updatemap.type = cpu_to_le16(LOG_FREEPXD);
1979 lrd->log.updatemap.nxd = cpu_to_le16(1);
1980 lrd->log.updatemap.pxd = pxdlock->pxd;
1981 pxd = pxdlock->pxd; /* save to format maplock */
1982 lrd->backchain =
1983 cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1984 }
1985
1986 /*
1987 * free entries XAD[next:hwm]:
1988 */
1989 if (hwm >= next) {
1990 /* init LOG_UPDATEMAP of the freed extents
1991 * XAD[next:hwm] from the deleted page itself
1992 * for logredo() to update bmap;
1993 */
1994 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1995 lrd->log.updatemap.type =
1996 cpu_to_le16(LOG_FREEXADLIST);
1997 xtlck = (struct xtlock *) & tlck->lock;
1998 hwm = xtlck->hwm.offset;
1999 lrd->log.updatemap.nxd =
2000 cpu_to_le16(hwm - next + 1);
2001 /* reformat linelock for lmLog() */
2002 xtlck->header.offset = next;
2003 xtlck->header.length = hwm - next + 1;
2004 xtlck->index = 1;
2005 lrd->backchain =
2006 cpu_to_le32(lmLog(log, tblk, lrd, tlck));
2007 }
2008
2009 /*
2010 * format maplock(s) for txUpdateMap() to update bmap
2011 */
2012 maplock->index = 0;
2013
2014 /*
2015 * allocate entries XAD[lwm:next):
2016 */
2017 if (lwm < next) {
2018 /* format a maplock for txUpdateMap() to update bPMAP
2019 * for alloc of new/extended extents of XAD[lwm:next)
2020 * from the page itself;
2021 * txUpdateMap() resets (XAD_NEW|XAD_EXTEND) flag.
2022 */
2023 tlck->flag |= tlckUPDATEMAP;
2024 xadlock->flag = mlckALLOCXADLIST;
2025 xadlock->count = next - lwm;
2026 xadlock->xdlist = &p->xad[lwm];
2027
2028 jfs_info("xtLog: alloc ip:0x%p mp:0x%p count:%d lwm:%d next:%d",
2029 tlck->ip, mp, xadlock->count, lwm, next);
2030 maplock->index++;
2031 xadlock++;
2032 }
2033
2034 /*
2035 * truncate entry XAD[twm == next - 1]:
2036 */
2037 if (twm == next - 1) {
2038 /* format a maplock for txUpdateMap() to update bmap
2039 * to free truncated delta extent of the truncated
2040 * entry XAD[next - 1];
2041 * (xtlck->pxdlock = truncated delta extent);
2042 */
2043 tlck->flag |= tlckUPDATEMAP;
2044 pxdlock = (struct pxd_lock *) xadlock;
2045 pxdlock->flag = mlckFREEPXD;
2046 pxdlock->count = 1;
2047 pxdlock->pxd = pxd;
2048
2049 jfs_info("xtLog: truncate ip:0x%p mp:0x%p count:%d hwm:%d",
2050 ip, mp, pxdlock->count, hwm);
2051 maplock->index++;
2052 xadlock++;
2053 }
2054
2055 /*
2056 * free entries XAD[next:hwm]:
2057 */
2058 if (hwm >= next) {
2059 /* format a maplock for txUpdateMap() to update bmap
2060 * to free extents of XAD[next:hwm] from thedeleted
2061 * page itself;
2062 */
2063 tlck->flag |= tlckUPDATEMAP;
2064 xadlock->flag = mlckFREEXADLIST;
2065 xadlock->count = hwm - next + 1;
2066 xadlock->xdlist = &p->xad[next];
2067
2068 jfs_info("xtLog: free ip:0x%p mp:0x%p count:%d next:%d hwm:%d",
2069 tlck->ip, mp, xadlock->count, next, hwm);
2070 maplock->index++;
2071 }
2072
2073 /* mark page as homeward bound */
2074 tlck->flag |= tlckWRITEPAGE;
2075 }
2076 return;
2077}
2078
2079/*
2080 * mapLog()
2081 *
2082 * function: log from maplock of freed data extents;
2083 */
2084static void mapLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
2085 struct tlock * tlck)
2086{
2087 struct pxd_lock *pxdlock;
2088 int i, nlock;
2089 pxd_t *pxd;
2090
2091 /*
2092 * page relocation: free the source page extent
2093 *
2094 * a maplock for txUpdateMap() for free of the page
2095 * has been formatted at txLock() time saving the src
2096 * relocated page address;
2097 */
2098 if (tlck->type & tlckRELOCATE) {
2099 /* log LOG_NOREDOPAGE of the old relocated page
2100 * for logredo() to start NoRedoPage filter;
2101 */
2102 lrd->type = cpu_to_le16(LOG_NOREDOPAGE);
2103 pxdlock = (struct pxd_lock *) & tlck->lock;
2104 pxd = &lrd->log.redopage.pxd;
2105 *pxd = pxdlock->pxd;
2106 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
2107
2108 /* (N.B. currently, logredo() does NOT update bmap
2109 * for free of the page itself for (LOG_XTREE|LOG_NOREDOPAGE);
2110 * if page free from relocation, LOG_UPDATEMAP log is
2111 * specifically generated now for logredo()
2112 * to update bmap for free of src relocated page;
2113 * (new flag LOG_RELOCATE may be introduced which will
2114 * inform logredo() to start NORedoPage filter and also
2115 * update block allocation map at the same time, thus
2116 * avoiding an extra log write);
2117 */
2118 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
2119 lrd->log.updatemap.type = cpu_to_le16(LOG_FREEPXD);
2120 lrd->log.updatemap.nxd = cpu_to_le16(1);
2121 lrd->log.updatemap.pxd = pxdlock->pxd;
2122 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
2123
2124 /* a maplock for txUpdateMap() for free of the page
2125 * has been formatted at txLock() time;
2126 */
2127 tlck->flag |= tlckUPDATEMAP;
2128 return;
2129 }
2130 /*
2131
2132 * Otherwise it's not a relocate request
2133 *
2134 */
2135 else {
2136 /* log LOG_UPDATEMAP for logredo() to update bmap for
2137 * free of truncated/relocated delta extent of the data;
2138 * e.g.: external EA extent, relocated/truncated extent
2139 * from xtTailgate();
2140 */
2141 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
2142 pxdlock = (struct pxd_lock *) & tlck->lock;
2143 nlock = pxdlock->index;
2144 for (i = 0; i < nlock; i++, pxdlock++) {
2145 if (pxdlock->flag & mlckALLOCPXD)
2146 lrd->log.updatemap.type =
2147 cpu_to_le16(LOG_ALLOCPXD);
2148 else
2149 lrd->log.updatemap.type =
2150 cpu_to_le16(LOG_FREEPXD);
2151 lrd->log.updatemap.nxd = cpu_to_le16(1);
2152 lrd->log.updatemap.pxd = pxdlock->pxd;
2153 lrd->backchain =
2154 cpu_to_le32(lmLog(log, tblk, lrd, NULL));
2155 jfs_info("mapLog: xaddr:0x%lx xlen:0x%x",
2156 (ulong) addressPXD(&pxdlock->pxd),
2157 lengthPXD(&pxdlock->pxd));
2158 }
2159
2160 /* update bmap */
2161 tlck->flag |= tlckUPDATEMAP;
2162 }
2163}
2164
2165/*
2166 * txEA()
2167 *
2168 * function: acquire maplock for EA/ACL extents or
2169 * set COMMIT_INLINE flag;
2170 */
2171void txEA(tid_t tid, struct inode *ip, dxd_t * oldea, dxd_t * newea)
2172{
2173 struct tlock *tlck = NULL;
2174 struct pxd_lock *maplock = NULL, *pxdlock = NULL;
2175
2176 /*
2177 * format maplock for alloc of new EA extent
2178 */
2179 if (newea) {
2180 /* Since the newea could be a completely zeroed entry we need to
2181 * check for the two flags which indicate we should actually
2182 * commit new EA data
2183 */
2184 if (newea->flag & DXD_EXTENT) {
2185 tlck = txMaplock(tid, ip, tlckMAP);
2186 maplock = (struct pxd_lock *) & tlck->lock;
2187 pxdlock = (struct pxd_lock *) maplock;
2188 pxdlock->flag = mlckALLOCPXD;
2189 PXDaddress(&pxdlock->pxd, addressDXD(newea));
2190 PXDlength(&pxdlock->pxd, lengthDXD(newea));
2191 pxdlock++;
2192 maplock->index = 1;
2193 } else if (newea->flag & DXD_INLINE) {
2194 tlck = NULL;
2195
2196 set_cflag(COMMIT_Inlineea, ip);
2197 }
2198 }
2199
2200 /*
2201 * format maplock for free of old EA extent
2202 */
2203 if (!test_cflag(COMMIT_Nolink, ip) && oldea->flag & DXD_EXTENT) {
2204 if (tlck == NULL) {
2205 tlck = txMaplock(tid, ip, tlckMAP);
2206 maplock = (struct pxd_lock *) & tlck->lock;
2207 pxdlock = (struct pxd_lock *) maplock;
2208 maplock->index = 0;
2209 }
2210 pxdlock->flag = mlckFREEPXD;
2211 PXDaddress(&pxdlock->pxd, addressDXD(oldea));
2212 PXDlength(&pxdlock->pxd, lengthDXD(oldea));
2213 maplock->index++;
2214 }
2215}
2216
2217/*
2218 * txForce()
2219 *
2220 * function: synchronously write pages locked by transaction
2221 * after txLog() but before txUpdateMap();
2222 */
2223static void txForce(struct tblock * tblk)
2224{
2225 struct tlock *tlck;
2226 lid_t lid, next;
2227 struct metapage *mp;
2228
2229 /*
2230 * reverse the order of transaction tlocks in
2231 * careful update order of address index pages
2232 * (right to left, bottom up)
2233 */
2234 tlck = lid_to_tlock(tblk->next);
2235 lid = tlck->next;
2236 tlck->next = 0;
2237 while (lid) {
2238 tlck = lid_to_tlock(lid);
2239 next = tlck->next;
2240 tlck->next = tblk->next;
2241 tblk->next = lid;
2242 lid = next;
2243 }
2244
2245 /*
2246 * synchronously write the page, and
2247 * hold the page for txUpdateMap();
2248 */
2249 for (lid = tblk->next; lid; lid = next) {
2250 tlck = lid_to_tlock(lid);
2251 next = tlck->next;
2252
2253 if ((mp = tlck->mp) != NULL &&
2254 (tlck->type & tlckBTROOT) == 0) {
2255 assert(mp->xflag & COMMIT_PAGE);
2256
2257 if (tlck->flag & tlckWRITEPAGE) {
2258 tlck->flag &= ~tlckWRITEPAGE;
2259
2260 /* do not release page to freelist */
2261 force_metapage(mp);
2262#if 0
2263 /*
2264 * The "right" thing to do here is to
2265 * synchronously write the metadata.
2266 * With the current implementation this
2267 * is hard since write_metapage requires
2268 * us to kunmap & remap the page. If we
2269 * have tlocks pointing into the metadata
2270 * pages, we don't want to do this. I think
2271 * we can get by with synchronously writing
2272 * the pages when they are released.
2273 */
2274 assert(mp->nohomeok);
2275 set_bit(META_dirty, &mp->flag);
2276 set_bit(META_sync, &mp->flag);
2277#endif
2278 }
2279 }
2280 }
2281}
2282
2283/*
2284 * txUpdateMap()
2285 *
2286 * function: update persistent allocation map (and working map
2287 * if appropriate);
2288 *
2289 * parameter:
2290 */
2291static void txUpdateMap(struct tblock * tblk)
2292{
2293 struct inode *ip;
2294 struct inode *ipimap;
2295 lid_t lid;
2296 struct tlock *tlck;
2297 struct maplock *maplock;
2298 struct pxd_lock pxdlock;
2299 int maptype;
2300 int k, nlock;
2301 struct metapage *mp = NULL;
2302
2303 ipimap = JFS_SBI(tblk->sb)->ipimap;
2304
2305 maptype = (tblk->xflag & COMMIT_PMAP) ? COMMIT_PMAP : COMMIT_PWMAP;
2306
2307
2308 /*
2309 * update block allocation map
2310 *
2311 * update allocation state in pmap (and wmap) and
2312 * update lsn of the pmap page;
2313 */
2314 /*
2315 * scan each tlock/page of transaction for block allocation/free:
2316 *
2317 * for each tlock/page of transaction, update map.
2318 * ? are there tlock for pmap and pwmap at the same time ?
2319 */
2320 for (lid = tblk->next; lid; lid = tlck->next) {
2321 tlck = lid_to_tlock(lid);
2322
2323 if ((tlck->flag & tlckUPDATEMAP) == 0)
2324 continue;
2325
2326 if (tlck->flag & tlckFREEPAGE) {
2327 /*
2328 * Another thread may attempt to reuse freed space
2329 * immediately, so we want to get rid of the metapage
2330 * before anyone else has a chance to get it.
2331 * Lock metapage, update maps, then invalidate
2332 * the metapage.
2333 */
2334 mp = tlck->mp;
2335 ASSERT(mp->xflag & COMMIT_PAGE);
2336 grab_metapage(mp);
2337 }
2338
2339 /*
2340 * extent list:
2341 * . in-line PXD list:
2342 * . out-of-line XAD list:
2343 */
2344 maplock = (struct maplock *) & tlck->lock;
2345 nlock = maplock->index;
2346
2347 for (k = 0; k < nlock; k++, maplock++) {
2348 /*
2349 * allocate blocks in persistent map:
2350 *
2351 * blocks have been allocated from wmap at alloc time;
2352 */
2353 if (maplock->flag & mlckALLOC) {
2354 txAllocPMap(ipimap, maplock, tblk);
2355 }
2356 /*
2357 * free blocks in persistent and working map:
2358 * blocks will be freed in pmap and then in wmap;
2359 *
2360 * ? tblock specifies the PMAP/PWMAP based upon
2361 * transaction
2362 *
2363 * free blocks in persistent map:
2364 * blocks will be freed from wmap at last reference
2365 * release of the object for regular files;
2366 *
2367 * Alway free blocks from both persistent & working
2368 * maps for directories
2369 */
2370 else { /* (maplock->flag & mlckFREE) */
2371
2372 if (tlck->flag & tlckDIRECTORY)
2373 txFreeMap(ipimap, maplock,
2374 tblk, COMMIT_PWMAP);
2375 else
2376 txFreeMap(ipimap, maplock,
2377 tblk, maptype);
2378 }
2379 }
2380 if (tlck->flag & tlckFREEPAGE) {
2381 if (!(tblk->flag & tblkGC_LAZY)) {
2382 /* This is equivalent to txRelease */
2383 ASSERT(mp->lid == lid);
2384 tlck->mp->lid = 0;
2385 }
2386 assert(mp->nohomeok == 1);
2387 metapage_homeok(mp);
2388 discard_metapage(mp);
2389 tlck->mp = NULL;
2390 }
2391 }
2392 /*
2393 * update inode allocation map
2394 *
2395 * update allocation state in pmap and
2396 * update lsn of the pmap page;
2397 * update in-memory inode flag/state
2398 *
2399 * unlock mapper/write lock
2400 */
2401 if (tblk->xflag & COMMIT_CREATE) {
2402 diUpdatePMap(ipimap, tblk->ino, false, tblk);
2403 /* update persistent block allocation map
2404 * for the allocation of inode extent;
2405 */
2406 pxdlock.flag = mlckALLOCPXD;
2407 pxdlock.pxd = tblk->u.ixpxd;
2408 pxdlock.index = 1;
2409 txAllocPMap(ipimap, (struct maplock *) & pxdlock, tblk);
2410 } else if (tblk->xflag & COMMIT_DELETE) {
2411 ip = tblk->u.ip;
2412 diUpdatePMap(ipimap, ip->i_ino, true, tblk);
2413 iput(ip);
2414 }
2415}
2416
2417/*
2418 * txAllocPMap()
2419 *
2420 * function: allocate from persistent map;
2421 *
2422 * parameter:
2423 * ipbmap -
2424 * malock -
2425 * xad list:
2426 * pxd:
2427 *
2428 * maptype -
2429 * allocate from persistent map;
2430 * free from persistent map;
2431 * (e.g., tmp file - free from working map at releae
2432 * of last reference);
2433 * free from persistent and working map;
2434 *
2435 * lsn - log sequence number;
2436 */
2437static void txAllocPMap(struct inode *ip, struct maplock * maplock,
2438 struct tblock * tblk)
2439{
2440 struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
2441 struct xdlistlock *xadlistlock;
2442 xad_t *xad;
2443 s64 xaddr;
2444 int xlen;
2445 struct pxd_lock *pxdlock;
2446 struct xdlistlock *pxdlistlock;
2447 pxd_t *pxd;
2448 int n;
2449
2450 /*
2451 * allocate from persistent map;
2452 */
2453 if (maplock->flag & mlckALLOCXADLIST) {
2454 xadlistlock = (struct xdlistlock *) maplock;
2455 xad = xadlistlock->xdlist;
2456 for (n = 0; n < xadlistlock->count; n++, xad++) {
2457 if (xad->flag & (XAD_NEW | XAD_EXTENDED)) {
2458 xaddr = addressXAD(xad);
2459 xlen = lengthXAD(xad);
2460 dbUpdatePMap(ipbmap, false, xaddr,
2461 (s64) xlen, tblk);
2462 xad->flag &= ~(XAD_NEW | XAD_EXTENDED);
2463 jfs_info("allocPMap: xaddr:0x%lx xlen:%d",
2464 (ulong) xaddr, xlen);
2465 }
2466 }
2467 } else if (maplock->flag & mlckALLOCPXD) {
2468 pxdlock = (struct pxd_lock *) maplock;
2469 xaddr = addressPXD(&pxdlock->pxd);
2470 xlen = lengthPXD(&pxdlock->pxd);
2471 dbUpdatePMap(ipbmap, false, xaddr, (s64) xlen, tblk);
2472 jfs_info("allocPMap: xaddr:0x%lx xlen:%d", (ulong) xaddr, xlen);
2473 } else { /* (maplock->flag & mlckALLOCPXDLIST) */
2474
2475 pxdlistlock = (struct xdlistlock *) maplock;
2476 pxd = pxdlistlock->xdlist;
2477 for (n = 0; n < pxdlistlock->count; n++, pxd++) {
2478 xaddr = addressPXD(pxd);
2479 xlen = lengthPXD(pxd);
2480 dbUpdatePMap(ipbmap, false, xaddr, (s64) xlen,
2481 tblk);
2482 jfs_info("allocPMap: xaddr:0x%lx xlen:%d",
2483 (ulong) xaddr, xlen);
2484 }
2485 }
2486}
2487
2488/*
2489 * txFreeMap()
2490 *
2491 * function: free from persistent and/or working map;
2492 *
2493 * todo: optimization
2494 */
2495void txFreeMap(struct inode *ip,
2496 struct maplock * maplock, struct tblock * tblk, int maptype)
2497{
2498 struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
2499 struct xdlistlock *xadlistlock;
2500 xad_t *xad;
2501 s64 xaddr;
2502 int xlen;
2503 struct pxd_lock *pxdlock;
2504 struct xdlistlock *pxdlistlock;
2505 pxd_t *pxd;
2506 int n;
2507
2508 jfs_info("txFreeMap: tblk:0x%p maplock:0x%p maptype:0x%x",
2509 tblk, maplock, maptype);
2510
2511 /*
2512 * free from persistent map;
2513 */
2514 if (maptype == COMMIT_PMAP || maptype == COMMIT_PWMAP) {
2515 if (maplock->flag & mlckFREEXADLIST) {
2516 xadlistlock = (struct xdlistlock *) maplock;
2517 xad = xadlistlock->xdlist;
2518 for (n = 0; n < xadlistlock->count; n++, xad++) {
2519 if (!(xad->flag & XAD_NEW)) {
2520 xaddr = addressXAD(xad);
2521 xlen = lengthXAD(xad);
2522 dbUpdatePMap(ipbmap, true, xaddr,
2523 (s64) xlen, tblk);
2524 jfs_info("freePMap: xaddr:0x%lx xlen:%d",
2525 (ulong) xaddr, xlen);
2526 }
2527 }
2528 } else if (maplock->flag & mlckFREEPXD) {
2529 pxdlock = (struct pxd_lock *) maplock;
2530 xaddr = addressPXD(&pxdlock->pxd);
2531 xlen = lengthPXD(&pxdlock->pxd);
2532 dbUpdatePMap(ipbmap, true, xaddr, (s64) xlen,
2533 tblk);
2534 jfs_info("freePMap: xaddr:0x%lx xlen:%d",
2535 (ulong) xaddr, xlen);
2536 } else { /* (maplock->flag & mlckALLOCPXDLIST) */
2537
2538 pxdlistlock = (struct xdlistlock *) maplock;
2539 pxd = pxdlistlock->xdlist;
2540 for (n = 0; n < pxdlistlock->count; n++, pxd++) {
2541 xaddr = addressPXD(pxd);
2542 xlen = lengthPXD(pxd);
2543 dbUpdatePMap(ipbmap, true, xaddr,
2544 (s64) xlen, tblk);
2545 jfs_info("freePMap: xaddr:0x%lx xlen:%d",
2546 (ulong) xaddr, xlen);
2547 }
2548 }
2549 }
2550
2551 /*
2552 * free from working map;
2553 */
2554 if (maptype == COMMIT_PWMAP || maptype == COMMIT_WMAP) {
2555 if (maplock->flag & mlckFREEXADLIST) {
2556 xadlistlock = (struct xdlistlock *) maplock;
2557 xad = xadlistlock->xdlist;
2558 for (n = 0; n < xadlistlock->count; n++, xad++) {
2559 xaddr = addressXAD(xad);
2560 xlen = lengthXAD(xad);
2561 dbFree(ip, xaddr, (s64) xlen);
2562 xad->flag = 0;
2563 jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
2564 (ulong) xaddr, xlen);
2565 }
2566 } else if (maplock->flag & mlckFREEPXD) {
2567 pxdlock = (struct pxd_lock *) maplock;
2568 xaddr = addressPXD(&pxdlock->pxd);
2569 xlen = lengthPXD(&pxdlock->pxd);
2570 dbFree(ip, xaddr, (s64) xlen);
2571 jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
2572 (ulong) xaddr, xlen);
2573 } else { /* (maplock->flag & mlckFREEPXDLIST) */
2574
2575 pxdlistlock = (struct xdlistlock *) maplock;
2576 pxd = pxdlistlock->xdlist;
2577 for (n = 0; n < pxdlistlock->count; n++, pxd++) {
2578 xaddr = addressPXD(pxd);
2579 xlen = lengthPXD(pxd);
2580 dbFree(ip, xaddr, (s64) xlen);
2581 jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
2582 (ulong) xaddr, xlen);
2583 }
2584 }
2585 }
2586}
2587
2588/*
2589 * txFreelock()
2590 *
2591 * function: remove tlock from inode anonymous locklist
2592 */
2593void txFreelock(struct inode *ip)
2594{
2595 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
2596 struct tlock *xtlck, *tlck;
2597 lid_t xlid = 0, lid;
2598
2599 if (!jfs_ip->atlhead)
2600 return;
2601
2602 TXN_LOCK();
2603 xtlck = (struct tlock *) &jfs_ip->atlhead;
2604
2605 while ((lid = xtlck->next) != 0) {
2606 tlck = lid_to_tlock(lid);
2607 if (tlck->flag & tlckFREELOCK) {
2608 xtlck->next = tlck->next;
2609 txLockFree(lid);
2610 } else {
2611 xtlck = tlck;
2612 xlid = lid;
2613 }
2614 }
2615
2616 if (jfs_ip->atlhead)
2617 jfs_ip->atltail = xlid;
2618 else {
2619 jfs_ip->atltail = 0;
2620 /*
2621 * If inode was on anon_list, remove it
2622 */
2623 list_del_init(&jfs_ip->anon_inode_list);
2624 }
2625 TXN_UNLOCK();
2626}
2627
2628/*
2629 * txAbort()
2630 *
2631 * function: abort tx before commit;
2632 *
2633 * frees line-locks and segment locks for all
2634 * segments in comdata structure.
2635 * Optionally sets state of file-system to FM_DIRTY in super-block.
2636 * log age of page-frames in memory for which caller has
2637 * are reset to 0 (to avoid logwarap).
2638 */
2639void txAbort(tid_t tid, int dirty)
2640{
2641 lid_t lid, next;
2642 struct metapage *mp;
2643 struct tblock *tblk = tid_to_tblock(tid);
2644 struct tlock *tlck;
2645
2646 /*
2647 * free tlocks of the transaction
2648 */
2649 for (lid = tblk->next; lid; lid = next) {
2650 tlck = lid_to_tlock(lid);
2651 next = tlck->next;
2652 mp = tlck->mp;
2653 JFS_IP(tlck->ip)->xtlid = 0;
2654
2655 if (mp) {
2656 mp->lid = 0;
2657
2658 /*
2659 * reset lsn of page to avoid logwarap:
2660 *
2661 * (page may have been previously committed by another
2662 * transaction(s) but has not been paged, i.e.,
2663 * it may be on logsync list even though it has not
2664 * been logged for the current tx.)
2665 */
2666 if (mp->xflag & COMMIT_PAGE && mp->lsn)
2667 LogSyncRelease(mp);
2668 }
2669 /* insert tlock at head of freelist */
2670 TXN_LOCK();
2671 txLockFree(lid);
2672 TXN_UNLOCK();
2673 }
2674
2675 /* caller will free the transaction block */
2676
2677 tblk->next = tblk->last = 0;
2678
2679 /*
2680 * mark filesystem dirty
2681 */
2682 if (dirty)
2683 jfs_error(tblk->sb, "\n");
2684
2685 return;
2686}
2687
2688/*
2689 * txLazyCommit(void)
2690 *
2691 * All transactions except those changing ipimap (COMMIT_FORCE) are
2692 * processed by this routine. This insures that the inode and block
2693 * allocation maps are updated in order. For synchronous transactions,
2694 * let the user thread finish processing after txUpdateMap() is called.
2695 */
2696static void txLazyCommit(struct tblock * tblk)
2697{
2698 struct jfs_log *log;
2699
2700 while (((tblk->flag & tblkGC_READY) == 0) &&
2701 ((tblk->flag & tblkGC_UNLOCKED) == 0)) {
2702 /* We must have gotten ahead of the user thread
2703 */
2704 jfs_info("jfs_lazycommit: tblk 0x%p not unlocked", tblk);
2705 yield();
2706 }
2707
2708 jfs_info("txLazyCommit: processing tblk 0x%p", tblk);
2709
2710 txUpdateMap(tblk);
2711
2712 log = (struct jfs_log *) JFS_SBI(tblk->sb)->log;
2713
2714 spin_lock_irq(&log->gclock); // LOGGC_LOCK
2715
2716 tblk->flag |= tblkGC_COMMITTED;
2717
2718 if (tblk->flag & tblkGC_READY)
2719 log->gcrtc--;
2720
2721 wake_up_all(&tblk->gcwait); // LOGGC_WAKEUP
2722
2723 /*
2724 * Can't release log->gclock until we've tested tblk->flag
2725 */
2726 if (tblk->flag & tblkGC_LAZY) {
2727 spin_unlock_irq(&log->gclock); // LOGGC_UNLOCK
2728 txUnlock(tblk);
2729 tblk->flag &= ~tblkGC_LAZY;
2730 txEnd(tblk - TxBlock); /* Convert back to tid */
2731 } else
2732 spin_unlock_irq(&log->gclock); // LOGGC_UNLOCK
2733
2734 jfs_info("txLazyCommit: done: tblk = 0x%p", tblk);
2735}
2736
2737/*
2738 * jfs_lazycommit(void)
2739 *
2740 * To be run as a kernel daemon. If lbmIODone is called in an interrupt
2741 * context, or where blocking is not wanted, this routine will process
2742 * committed transactions from the unlock queue.
2743 */
2744int jfs_lazycommit(void *arg)
2745{
2746 int WorkDone;
2747 struct tblock *tblk;
2748 unsigned long flags;
2749 struct jfs_sb_info *sbi;
2750
2751 do {
2752 LAZY_LOCK(flags);
2753 jfs_commit_thread_waking = 0; /* OK to wake another thread */
2754 while (!list_empty(&TxAnchor.unlock_queue)) {
2755 WorkDone = 0;
2756 list_for_each_entry(tblk, &TxAnchor.unlock_queue,
2757 cqueue) {
2758
2759 sbi = JFS_SBI(tblk->sb);
2760 /*
2761 * For each volume, the transactions must be
2762 * handled in order. If another commit thread
2763 * is handling a tblk for this superblock,
2764 * skip it
2765 */
2766 if (sbi->commit_state & IN_LAZYCOMMIT)
2767 continue;
2768
2769 sbi->commit_state |= IN_LAZYCOMMIT;
2770 WorkDone = 1;
2771
2772 /*
2773 * Remove transaction from queue
2774 */
2775 list_del(&tblk->cqueue);
2776
2777 LAZY_UNLOCK(flags);
2778 txLazyCommit(tblk);
2779 LAZY_LOCK(flags);
2780
2781 sbi->commit_state &= ~IN_LAZYCOMMIT;
2782 /*
2783 * Don't continue in the for loop. (We can't
2784 * anyway, it's unsafe!) We want to go back to
2785 * the beginning of the list.
2786 */
2787 break;
2788 }
2789
2790 /* If there was nothing to do, don't continue */
2791 if (!WorkDone)
2792 break;
2793 }
2794 /* In case a wakeup came while all threads were active */
2795 jfs_commit_thread_waking = 0;
2796
2797 if (freezing(current)) {
2798 LAZY_UNLOCK(flags);
2799 try_to_freeze();
2800 } else {
2801 DECLARE_WAITQUEUE(wq, current);
2802
2803 add_wait_queue(&jfs_commit_thread_wait, &wq);
2804 set_current_state(TASK_INTERRUPTIBLE);
2805 LAZY_UNLOCK(flags);
2806 schedule();
2807 remove_wait_queue(&jfs_commit_thread_wait, &wq);
2808 }
2809 } while (!kthread_should_stop());
2810
2811 if (!list_empty(&TxAnchor.unlock_queue))
2812 jfs_err("jfs_lazycommit being killed w/pending transactions!");
2813 else
2814 jfs_info("jfs_lazycommit being killed");
2815 return 0;
2816}
2817
2818void txLazyUnlock(struct tblock * tblk)
2819{
2820 unsigned long flags;
2821
2822 LAZY_LOCK(flags);
2823
2824 list_add_tail(&tblk->cqueue, &TxAnchor.unlock_queue);
2825 /*
2826 * Don't wake up a commit thread if there is already one servicing
2827 * this superblock, or if the last one we woke up hasn't started yet.
2828 */
2829 if (!(JFS_SBI(tblk->sb)->commit_state & IN_LAZYCOMMIT) &&
2830 !jfs_commit_thread_waking) {
2831 jfs_commit_thread_waking = 1;
2832 wake_up(&jfs_commit_thread_wait);
2833 }
2834 LAZY_UNLOCK(flags);
2835}
2836
2837static void LogSyncRelease(struct metapage * mp)
2838{
2839 struct jfs_log *log = mp->log;
2840
2841 assert(mp->nohomeok);
2842 assert(log);
2843 metapage_homeok(mp);
2844}
2845
2846/*
2847 * txQuiesce
2848 *
2849 * Block all new transactions and push anonymous transactions to
2850 * completion
2851 *
2852 * This does almost the same thing as jfs_sync below. We don't
2853 * worry about deadlocking when jfs_tlocks_low is set, since we would
2854 * expect jfs_sync to get us out of that jam.
2855 */
2856void txQuiesce(struct super_block *sb)
2857{
2858 struct inode *ip;
2859 struct jfs_inode_info *jfs_ip;
2860 struct jfs_log *log = JFS_SBI(sb)->log;
2861 tid_t tid;
2862
2863 set_bit(log_QUIESCE, &log->flag);
2864
2865 TXN_LOCK();
2866restart:
2867 while (!list_empty(&TxAnchor.anon_list)) {
2868 jfs_ip = list_entry(TxAnchor.anon_list.next,
2869 struct jfs_inode_info,
2870 anon_inode_list);
2871 ip = &jfs_ip->vfs_inode;
2872
2873 /*
2874 * inode will be removed from anonymous list
2875 * when it is committed
2876 */
2877 TXN_UNLOCK();
2878 tid = txBegin(ip->i_sb, COMMIT_INODE | COMMIT_FORCE);
2879 mutex_lock(&jfs_ip->commit_mutex);
2880 txCommit(tid, 1, &ip, 0);
2881 txEnd(tid);
2882 mutex_unlock(&jfs_ip->commit_mutex);
2883 /*
2884 * Just to be safe. I don't know how
2885 * long we can run without blocking
2886 */
2887 cond_resched();
2888 TXN_LOCK();
2889 }
2890
2891 /*
2892 * If jfs_sync is running in parallel, there could be some inodes
2893 * on anon_list2. Let's check.
2894 */
2895 if (!list_empty(&TxAnchor.anon_list2)) {
2896 list_splice_init(&TxAnchor.anon_list2, &TxAnchor.anon_list);
2897 goto restart;
2898 }
2899 TXN_UNLOCK();
2900
2901 /*
2902 * We may need to kick off the group commit
2903 */
2904 jfs_flush_journal(log, 0);
2905}
2906
2907/*
2908 * txResume()
2909 *
2910 * Allows transactions to start again following txQuiesce
2911 */
2912void txResume(struct super_block *sb)
2913{
2914 struct jfs_log *log = JFS_SBI(sb)->log;
2915
2916 clear_bit(log_QUIESCE, &log->flag);
2917 TXN_WAKEUP(&log->syncwait);
2918}
2919
2920/*
2921 * jfs_sync(void)
2922 *
2923 * To be run as a kernel daemon. This is awakened when tlocks run low.
2924 * We write any inodes that have anonymous tlocks so they will become
2925 * available.
2926 */
2927int jfs_sync(void *arg)
2928{
2929 struct inode *ip;
2930 struct jfs_inode_info *jfs_ip;
2931 tid_t tid;
2932
2933 do {
2934 /*
2935 * write each inode on the anonymous inode list
2936 */
2937 TXN_LOCK();
2938 while (jfs_tlocks_low && !list_empty(&TxAnchor.anon_list)) {
2939 jfs_ip = list_entry(TxAnchor.anon_list.next,
2940 struct jfs_inode_info,
2941 anon_inode_list);
2942 ip = &jfs_ip->vfs_inode;
2943
2944 if (! igrab(ip)) {
2945 /*
2946 * Inode is being freed
2947 */
2948 list_del_init(&jfs_ip->anon_inode_list);
2949 } else if (mutex_trylock(&jfs_ip->commit_mutex)) {
2950 /*
2951 * inode will be removed from anonymous list
2952 * when it is committed
2953 */
2954 TXN_UNLOCK();
2955 tid = txBegin(ip->i_sb, COMMIT_INODE);
2956 txCommit(tid, 1, &ip, 0);
2957 txEnd(tid);
2958 mutex_unlock(&jfs_ip->commit_mutex);
2959
2960 iput(ip);
2961 /*
2962 * Just to be safe. I don't know how
2963 * long we can run without blocking
2964 */
2965 cond_resched();
2966 TXN_LOCK();
2967 } else {
2968 /* We can't get the commit mutex. It may
2969 * be held by a thread waiting for tlock's
2970 * so let's not block here. Save it to
2971 * put back on the anon_list.
2972 */
2973
2974 /* Move from anon_list to anon_list2 */
2975 list_move(&jfs_ip->anon_inode_list,
2976 &TxAnchor.anon_list2);
2977
2978 TXN_UNLOCK();
2979 iput(ip);
2980 TXN_LOCK();
2981 }
2982 }
2983 /* Add anon_list2 back to anon_list */
2984 list_splice_init(&TxAnchor.anon_list2, &TxAnchor.anon_list);
2985
2986 if (freezing(current)) {
2987 TXN_UNLOCK();
2988 try_to_freeze();
2989 } else {
2990 set_current_state(TASK_INTERRUPTIBLE);
2991 TXN_UNLOCK();
2992 schedule();
2993 }
2994 } while (!kthread_should_stop());
2995
2996 jfs_info("jfs_sync being killed");
2997 return 0;
2998}
2999
3000#if defined(CONFIG_PROC_FS) && defined(CONFIG_JFS_DEBUG)
3001static int jfs_txanchor_proc_show(struct seq_file *m, void *v)
3002{
3003 char *freewait;
3004 char *freelockwait;
3005 char *lowlockwait;
3006
3007 freewait =
3008 waitqueue_active(&TxAnchor.freewait) ? "active" : "empty";
3009 freelockwait =
3010 waitqueue_active(&TxAnchor.freelockwait) ? "active" : "empty";
3011 lowlockwait =
3012 waitqueue_active(&TxAnchor.lowlockwait) ? "active" : "empty";
3013
3014 seq_printf(m,
3015 "JFS TxAnchor\n"
3016 "============\n"
3017 "freetid = %d\n"
3018 "freewait = %s\n"
3019 "freelock = %d\n"
3020 "freelockwait = %s\n"
3021 "lowlockwait = %s\n"
3022 "tlocksInUse = %d\n"
3023 "jfs_tlocks_low = %d\n"
3024 "unlock_queue is %sempty\n",
3025 TxAnchor.freetid,
3026 freewait,
3027 TxAnchor.freelock,
3028 freelockwait,
3029 lowlockwait,
3030 TxAnchor.tlocksInUse,
3031 jfs_tlocks_low,
3032 list_empty(&TxAnchor.unlock_queue) ? "" : "not ");
3033 return 0;
3034}
3035
3036static int jfs_txanchor_proc_open(struct inode *inode, struct file *file)
3037{
3038 return single_open(file, jfs_txanchor_proc_show, NULL);
3039}
3040
3041const struct file_operations jfs_txanchor_proc_fops = {
3042 .open = jfs_txanchor_proc_open,
3043 .read = seq_read,
3044 .llseek = seq_lseek,
3045 .release = single_release,
3046};
3047#endif
3048
3049#if defined(CONFIG_PROC_FS) && defined(CONFIG_JFS_STATISTICS)
3050static int jfs_txstats_proc_show(struct seq_file *m, void *v)
3051{
3052 seq_printf(m,
3053 "JFS TxStats\n"
3054 "===========\n"
3055 "calls to txBegin = %d\n"
3056 "txBegin blocked by sync barrier = %d\n"
3057 "txBegin blocked by tlocks low = %d\n"
3058 "txBegin blocked by no free tid = %d\n"
3059 "calls to txBeginAnon = %d\n"
3060 "txBeginAnon blocked by sync barrier = %d\n"
3061 "txBeginAnon blocked by tlocks low = %d\n"
3062 "calls to txLockAlloc = %d\n"
3063 "tLockAlloc blocked by no free lock = %d\n",
3064 TxStat.txBegin,
3065 TxStat.txBegin_barrier,
3066 TxStat.txBegin_lockslow,
3067 TxStat.txBegin_freetid,
3068 TxStat.txBeginAnon,
3069 TxStat.txBeginAnon_barrier,
3070 TxStat.txBeginAnon_lockslow,
3071 TxStat.txLockAlloc,
3072 TxStat.txLockAlloc_freelock);
3073 return 0;
3074}
3075
3076static int jfs_txstats_proc_open(struct inode *inode, struct file *file)
3077{
3078 return single_open(file, jfs_txstats_proc_show, NULL);
3079}
3080
3081const struct file_operations jfs_txstats_proc_fops = {
3082 .open = jfs_txstats_proc_open,
3083 .read = seq_read,
3084 .llseek = seq_lseek,
3085 .release = single_release,
3086};
3087#endif
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 */
53static 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
68int jfs_tlocks_low; /* Indicates low number of available tlocks */
69
70#ifdef CONFIG_JFS_STATISTICS
71static 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
84static int nTxBlock = -1; /* number of transaction blocks */
85module_param(nTxBlock, int, 0);
86MODULE_PARM_DESC(nTxBlock,
87 "Number of transaction blocks (max:65536)");
88
89static int nTxLock = -1; /* number of transaction locks */
90module_param(nTxLock, int, 0);
91MODULE_PARM_DESC(nTxLock,
92 "Number of transaction locks (max:65536)");
93
94struct tblock *TxBlock; /* transaction block table */
95static int TxLockLWM; /* Low water mark for number of txLocks used */
96static int TxLockHWM; /* High water mark for number of txLocks used */
97static int TxLockVHWM; /* Very High water mark */
98struct tlock *TxLock; /* transaction lock table */
99
100/*
101 * transaction management lock
102 */
103static 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
112static DECLARE_WAIT_QUEUE_HEAD(jfs_commit_thread_wait);
113static int jfs_commit_thread_waking;
114
115/*
116 * Retry logic exist outside these macros to protect from spurrious wakeups.
117 */
118static 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 */
140static 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 */
151static void diLog(struct jfs_log *log, struct tblock *tblk, struct lrd *lrd,
152 struct tlock *tlck, struct commit *cd);
153static void dataLog(struct jfs_log *log, struct tblock *tblk, struct lrd *lrd,
154 struct tlock *tlck);
155static void dtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
156 struct tlock * tlck);
157static void mapLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
158 struct tlock * tlck);
159static void txAllocPMap(struct inode *ip, struct maplock * maplock,
160 struct tblock * tblk);
161static void txForce(struct tblock * tblk);
162static void txLog(struct jfs_log *log, struct tblock *tblk,
163 struct commit *cd);
164static void txUpdateMap(struct tblock * tblk);
165static void txRelease(struct tblock * tblk);
166static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
167 struct tlock * tlck);
168static 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 */
180static 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
202static 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 */
225int 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 */
327void 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 */
348tid_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 */
452void 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 */
492void 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();
564wakeup:
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 */
582struct 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 */
868static 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 */
900static 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 */
987struct 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 */
1060struct 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 */
1127int 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 */
1372static 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 */
1428static 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 */
1506static 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 */
1552static 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 */
1655static 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 */
2038static 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 */
2125void 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 */
2177static 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 */
2245static 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 */
2391static 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 */
2449void 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 */
2547void 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 */
2593void 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 */
2650static 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 */
2698int jfs_lazycommit(void *arg)
2699{
2700 int WorkDone;
2701 struct tblock *tblk;
2702 unsigned long flags;
2703 struct jfs_sb_info *sbi;
2704
2705 set_freezable();
2706 do {
2707 LAZY_LOCK(flags);
2708 jfs_commit_thread_waking = 0; /* OK to wake another thread */
2709 while (!list_empty(&TxAnchor.unlock_queue)) {
2710 WorkDone = 0;
2711 list_for_each_entry(tblk, &TxAnchor.unlock_queue,
2712 cqueue) {
2713
2714 sbi = JFS_SBI(tblk->sb);
2715 /*
2716 * For each volume, the transactions must be
2717 * handled in order. If another commit thread
2718 * is handling a tblk for this superblock,
2719 * skip it
2720 */
2721 if (sbi->commit_state & IN_LAZYCOMMIT)
2722 continue;
2723
2724 sbi->commit_state |= IN_LAZYCOMMIT;
2725 WorkDone = 1;
2726
2727 /*
2728 * Remove transaction from queue
2729 */
2730 list_del(&tblk->cqueue);
2731
2732 LAZY_UNLOCK(flags);
2733 txLazyCommit(tblk);
2734 LAZY_LOCK(flags);
2735
2736 sbi->commit_state &= ~IN_LAZYCOMMIT;
2737 /*
2738 * Don't continue in the for loop. (We can't
2739 * anyway, it's unsafe!) We want to go back to
2740 * the beginning of the list.
2741 */
2742 break;
2743 }
2744
2745 /* If there was nothing to do, don't continue */
2746 if (!WorkDone)
2747 break;
2748 }
2749 /* In case a wakeup came while all threads were active */
2750 jfs_commit_thread_waking = 0;
2751
2752 if (freezing(current)) {
2753 LAZY_UNLOCK(flags);
2754 try_to_freeze();
2755 } else {
2756 DECLARE_WAITQUEUE(wq, current);
2757
2758 add_wait_queue(&jfs_commit_thread_wait, &wq);
2759 set_current_state(TASK_INTERRUPTIBLE);
2760 LAZY_UNLOCK(flags);
2761 schedule();
2762 remove_wait_queue(&jfs_commit_thread_wait, &wq);
2763 }
2764 } while (!kthread_should_stop());
2765
2766 if (!list_empty(&TxAnchor.unlock_queue))
2767 jfs_err("jfs_lazycommit being killed w/pending transactions!");
2768 else
2769 jfs_info("jfs_lazycommit being killed");
2770 return 0;
2771}
2772
2773void txLazyUnlock(struct tblock * tblk)
2774{
2775 unsigned long flags;
2776
2777 LAZY_LOCK(flags);
2778
2779 list_add_tail(&tblk->cqueue, &TxAnchor.unlock_queue);
2780 /*
2781 * Don't wake up a commit thread if there is already one servicing
2782 * this superblock, or if the last one we woke up hasn't started yet.
2783 */
2784 if (!(JFS_SBI(tblk->sb)->commit_state & IN_LAZYCOMMIT) &&
2785 !jfs_commit_thread_waking) {
2786 jfs_commit_thread_waking = 1;
2787 wake_up(&jfs_commit_thread_wait);
2788 }
2789 LAZY_UNLOCK(flags);
2790}
2791
2792static void LogSyncRelease(struct metapage * mp)
2793{
2794 struct jfs_log *log = mp->log;
2795
2796 assert(mp->nohomeok);
2797 assert(log);
2798 metapage_homeok(mp);
2799}
2800
2801/*
2802 * txQuiesce
2803 *
2804 * Block all new transactions and push anonymous transactions to
2805 * completion
2806 *
2807 * This does almost the same thing as jfs_sync below. We don't
2808 * worry about deadlocking when jfs_tlocks_low is set, since we would
2809 * expect jfs_sync to get us out of that jam.
2810 */
2811void txQuiesce(struct super_block *sb)
2812{
2813 struct inode *ip;
2814 struct jfs_inode_info *jfs_ip;
2815 struct jfs_log *log = JFS_SBI(sb)->log;
2816 tid_t tid;
2817
2818 set_bit(log_QUIESCE, &log->flag);
2819
2820 TXN_LOCK();
2821restart:
2822 while (!list_empty(&TxAnchor.anon_list)) {
2823 jfs_ip = list_entry(TxAnchor.anon_list.next,
2824 struct jfs_inode_info,
2825 anon_inode_list);
2826 ip = &jfs_ip->vfs_inode;
2827
2828 /*
2829 * inode will be removed from anonymous list
2830 * when it is committed
2831 */
2832 TXN_UNLOCK();
2833 tid = txBegin(ip->i_sb, COMMIT_INODE | COMMIT_FORCE);
2834 mutex_lock(&jfs_ip->commit_mutex);
2835 txCommit(tid, 1, &ip, 0);
2836 txEnd(tid);
2837 mutex_unlock(&jfs_ip->commit_mutex);
2838 /*
2839 * Just to be safe. I don't know how
2840 * long we can run without blocking
2841 */
2842 cond_resched();
2843 TXN_LOCK();
2844 }
2845
2846 /*
2847 * If jfs_sync is running in parallel, there could be some inodes
2848 * on anon_list2. Let's check.
2849 */
2850 if (!list_empty(&TxAnchor.anon_list2)) {
2851 list_splice_init(&TxAnchor.anon_list2, &TxAnchor.anon_list);
2852 goto restart;
2853 }
2854 TXN_UNLOCK();
2855
2856 /*
2857 * We may need to kick off the group commit
2858 */
2859 jfs_flush_journal(log, 0);
2860}
2861
2862/*
2863 * txResume()
2864 *
2865 * Allows transactions to start again following txQuiesce
2866 */
2867void txResume(struct super_block *sb)
2868{
2869 struct jfs_log *log = JFS_SBI(sb)->log;
2870
2871 clear_bit(log_QUIESCE, &log->flag);
2872 TXN_WAKEUP(&log->syncwait);
2873}
2874
2875/*
2876 * jfs_sync(void)
2877 *
2878 * To be run as a kernel daemon. This is awakened when tlocks run low.
2879 * We write any inodes that have anonymous tlocks so they will become
2880 * available.
2881 */
2882int jfs_sync(void *arg)
2883{
2884 struct inode *ip;
2885 struct jfs_inode_info *jfs_ip;
2886 tid_t tid;
2887
2888 set_freezable();
2889 do {
2890 /*
2891 * write each inode on the anonymous inode list
2892 */
2893 TXN_LOCK();
2894 while (jfs_tlocks_low && !list_empty(&TxAnchor.anon_list)) {
2895 jfs_ip = list_entry(TxAnchor.anon_list.next,
2896 struct jfs_inode_info,
2897 anon_inode_list);
2898 ip = &jfs_ip->vfs_inode;
2899
2900 if (! igrab(ip)) {
2901 /*
2902 * Inode is being freed
2903 */
2904 list_del_init(&jfs_ip->anon_inode_list);
2905 } else if (mutex_trylock(&jfs_ip->commit_mutex)) {
2906 /*
2907 * inode will be removed from anonymous list
2908 * when it is committed
2909 */
2910 TXN_UNLOCK();
2911 tid = txBegin(ip->i_sb, COMMIT_INODE);
2912 txCommit(tid, 1, &ip, 0);
2913 txEnd(tid);
2914 mutex_unlock(&jfs_ip->commit_mutex);
2915
2916 iput(ip);
2917 /*
2918 * Just to be safe. I don't know how
2919 * long we can run without blocking
2920 */
2921 cond_resched();
2922 TXN_LOCK();
2923 } else {
2924 /* We can't get the commit mutex. It may
2925 * be held by a thread waiting for tlock's
2926 * so let's not block here. Save it to
2927 * put back on the anon_list.
2928 */
2929
2930 /* Move from anon_list to anon_list2 */
2931 list_move(&jfs_ip->anon_inode_list,
2932 &TxAnchor.anon_list2);
2933
2934 TXN_UNLOCK();
2935 iput(ip);
2936 TXN_LOCK();
2937 }
2938 }
2939 /* Add anon_list2 back to anon_list */
2940 list_splice_init(&TxAnchor.anon_list2, &TxAnchor.anon_list);
2941
2942 if (freezing(current)) {
2943 TXN_UNLOCK();
2944 try_to_freeze();
2945 } else {
2946 set_current_state(TASK_INTERRUPTIBLE);
2947 TXN_UNLOCK();
2948 schedule();
2949 }
2950 } while (!kthread_should_stop());
2951
2952 jfs_info("jfs_sync being killed");
2953 return 0;
2954}
2955
2956#if defined(CONFIG_PROC_FS) && defined(CONFIG_JFS_DEBUG)
2957int jfs_txanchor_proc_show(struct seq_file *m, void *v)
2958{
2959 char *freewait;
2960 char *freelockwait;
2961 char *lowlockwait;
2962
2963 freewait =
2964 waitqueue_active(&TxAnchor.freewait) ? "active" : "empty";
2965 freelockwait =
2966 waitqueue_active(&TxAnchor.freelockwait) ? "active" : "empty";
2967 lowlockwait =
2968 waitqueue_active(&TxAnchor.lowlockwait) ? "active" : "empty";
2969
2970 seq_printf(m,
2971 "JFS TxAnchor\n"
2972 "============\n"
2973 "freetid = %d\n"
2974 "freewait = %s\n"
2975 "freelock = %d\n"
2976 "freelockwait = %s\n"
2977 "lowlockwait = %s\n"
2978 "tlocksInUse = %d\n"
2979 "jfs_tlocks_low = %d\n"
2980 "unlock_queue is %sempty\n",
2981 TxAnchor.freetid,
2982 freewait,
2983 TxAnchor.freelock,
2984 freelockwait,
2985 lowlockwait,
2986 TxAnchor.tlocksInUse,
2987 jfs_tlocks_low,
2988 list_empty(&TxAnchor.unlock_queue) ? "" : "not ");
2989 return 0;
2990}
2991#endif
2992
2993#if defined(CONFIG_PROC_FS) && defined(CONFIG_JFS_STATISTICS)
2994int jfs_txstats_proc_show(struct seq_file *m, void *v)
2995{
2996 seq_printf(m,
2997 "JFS TxStats\n"
2998 "===========\n"
2999 "calls to txBegin = %d\n"
3000 "txBegin blocked by sync barrier = %d\n"
3001 "txBegin blocked by tlocks low = %d\n"
3002 "txBegin blocked by no free tid = %d\n"
3003 "calls to txBeginAnon = %d\n"
3004 "txBeginAnon blocked by sync barrier = %d\n"
3005 "txBeginAnon blocked by tlocks low = %d\n"
3006 "calls to txLockAlloc = %d\n"
3007 "tLockAlloc blocked by no free lock = %d\n",
3008 TxStat.txBegin,
3009 TxStat.txBegin_barrier,
3010 TxStat.txBegin_lockslow,
3011 TxStat.txBegin_freetid,
3012 TxStat.txBeginAnon,
3013 TxStat.txBeginAnon_barrier,
3014 TxStat.txBeginAnon_lockslow,
3015 TxStat.txLockAlloc,
3016 TxStat.txLockAlloc_freelock);
3017 return 0;
3018}
3019#endif