1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Copyright (C) International Business Machines Corp., 2000-2005
4 * Portions Copyright (C) Christoph Hellwig, 2001-2002
5 */
6
7 /*
8 * jfs_txnmgr.c: transaction manager
9 *
10 * notes:
11 * transaction starts with txBegin() and ends with txCommit()
12 * or txAbort().
13 *
14 * tlock is acquired at the time of update;
15 * (obviate scan at commit time for xtree and dtree)
16 * tlock and mp points to each other;
17 * (no hashlist for mp -> tlock).
18 *
19 * special cases:
20 * tlock on in-memory inode:
21 * in-place tlock in the in-memory inode itself;
22 * converted to page lock by iWrite() at commit time.
23 *
24 * tlock during write()/mmap() under anonymous transaction (tid = 0):
25 * transferred (?) to transaction at commit time.
26 *
27 * use the page itself to update allocation maps
28 * (obviate intermediate replication of allocation/deallocation data)
29 * hold on to mp+lock thru update of maps
30 */
31
32 #include <linux/fs.h>
33 #include <linux/vmalloc.h>
34 #include <linux/completion.h>
35 #include <linux/freezer.h>
36 #include <linux/module.h>
37 #include <linux/moduleparam.h>
38 #include <linux/kthread.h>
39 #include <linux/seq_file.h>
40 #include "jfs_incore.h"
41 #include "jfs_inode.h"
42 #include "jfs_filsys.h"
43 #include "jfs_metapage.h"
44 #include "jfs_dinode.h"
45 #include "jfs_imap.h"
46 #include "jfs_dmap.h"
47 #include "jfs_superblock.h"
48 #include "jfs_debug.h"
49
50 /*
51 * transaction management structures
52 */
53 static struct {
54 int freetid; /* index of a free tid structure */
55 int freelock; /* index first free lock word */
56 wait_queue_head_t freewait; /* eventlist of free tblock */
57 wait_queue_head_t freelockwait; /* eventlist of free tlock */
58 wait_queue_head_t lowlockwait; /* eventlist of ample tlocks */
59 int tlocksInUse; /* Number of tlocks in use */
60 spinlock_t LazyLock; /* synchronize sync_queue & unlock_queue */
61 /* struct tblock *sync_queue; * Transactions waiting for data sync */
62 struct list_head unlock_queue; /* Txns waiting to be released */
63 struct list_head anon_list; /* inodes having anonymous txns */
64 struct list_head anon_list2; /* inodes having anonymous txns
65 that couldn't be sync'ed */
66 } TxAnchor;
67
68 int jfs_tlocks_low; /* Indicates low number of available tlocks */
69
70 #ifdef CONFIG_JFS_STATISTICS
71 static struct {
72 uint txBegin;
73 uint txBegin_barrier;
74 uint txBegin_lockslow;
75 uint txBegin_freetid;
76 uint txBeginAnon;
77 uint txBeginAnon_barrier;
78 uint txBeginAnon_lockslow;
79 uint txLockAlloc;
80 uint txLockAlloc_freelock;
81 } TxStat;
82 #endif
83
84 static int nTxBlock = -1; /* number of transaction blocks */
85 module_param(nTxBlock, int, 0);
86 MODULE_PARM_DESC(nTxBlock,
87 "Number of transaction blocks (max:65536)");
88
89 static int nTxLock = -1; /* number of transaction locks */
90 module_param(nTxLock, int, 0);
91 MODULE_PARM_DESC(nTxLock,
92 "Number of transaction locks (max:65536)");
93
94 struct tblock *TxBlock; /* transaction block table */
95 static int TxLockLWM; /* Low water mark for number of txLocks used */
96 static int TxLockHWM; /* High water mark for number of txLocks used */
97 static int TxLockVHWM; /* Very High water mark */
98 struct tlock *TxLock; /* transaction lock table */
99
100 /*
101 * transaction management lock
102 */
103 static DEFINE_SPINLOCK(jfsTxnLock);
104
105 #define TXN_LOCK() spin_lock(&jfsTxnLock)
106 #define TXN_UNLOCK() spin_unlock(&jfsTxnLock)
107
108 #define LAZY_LOCK_INIT() spin_lock_init(&TxAnchor.LazyLock);
109 #define LAZY_LOCK(flags) spin_lock_irqsave(&TxAnchor.LazyLock, flags)
110 #define LAZY_UNLOCK(flags) spin_unlock_irqrestore(&TxAnchor.LazyLock, flags)
111
112 static DECLARE_WAIT_QUEUE_HEAD(jfs_commit_thread_wait);
113 static int jfs_commit_thread_waking;
114
115 /*
116 * Retry logic exist outside these macros to protect from spurrious wakeups.
117 */
TXN_SLEEP_DROP_LOCK(wait_queue_head_t * event)118 static inline void TXN_SLEEP_DROP_LOCK(wait_queue_head_t * event)
119 {
120 DECLARE_WAITQUEUE(wait, current);
121
122 add_wait_queue(event, &wait);
123 set_current_state(TASK_UNINTERRUPTIBLE);
124 TXN_UNLOCK();
125 io_schedule();
126 remove_wait_queue(event, &wait);
127 }
128
129 #define TXN_SLEEP(event)\
130 {\
131 TXN_SLEEP_DROP_LOCK(event);\
132 TXN_LOCK();\
133 }
134
135 #define TXN_WAKEUP(event) wake_up_all(event)
136
137 /*
138 * statistics
139 */
140 static struct {
141 tid_t maxtid; /* 4: biggest tid ever used */
142 lid_t maxlid; /* 4: biggest lid ever used */
143 int ntid; /* 4: # of transactions performed */
144 int nlid; /* 4: # of tlocks acquired */
145 int waitlock; /* 4: # of tlock wait */
146 } stattx;
147
148 /*
149 * forward references
150 */
151 static int diLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
152 struct tlock * tlck, struct commit * cd);
153 static int dataLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
154 struct tlock * tlck);
155 static void dtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
156 struct tlock * tlck);
157 static void mapLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
158 struct tlock * tlck);
159 static void txAllocPMap(struct inode *ip, struct maplock * maplock,
160 struct tblock * tblk);
161 static void txForce(struct tblock * tblk);
162 static int txLog(struct jfs_log * log, struct tblock * tblk,
163 struct commit * cd);
164 static void txUpdateMap(struct tblock * tblk);
165 static void txRelease(struct tblock * tblk);
166 static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
167 struct tlock * tlck);
168 static void LogSyncRelease(struct metapage * mp);
169
170 /*
171 * transaction block/lock management
172 * ---------------------------------
173 */
174
175 /*
176 * Get a transaction lock from the free list. If the number in use is
177 * greater than the high water mark, wake up the sync daemon. This should
178 * free some anonymous transaction locks. (TXN_LOCK must be held.)
179 */
txLockAlloc(void)180 static lid_t txLockAlloc(void)
181 {
182 lid_t lid;
183
184 INCREMENT(TxStat.txLockAlloc);
185 if (!TxAnchor.freelock) {
186 INCREMENT(TxStat.txLockAlloc_freelock);
187 }
188
189 while (!(lid = TxAnchor.freelock))
190 TXN_SLEEP(&TxAnchor.freelockwait);
191 TxAnchor.freelock = TxLock[lid].next;
192 HIGHWATERMARK(stattx.maxlid, lid);
193 if ((++TxAnchor.tlocksInUse > TxLockHWM) && (jfs_tlocks_low == 0)) {
194 jfs_info("txLockAlloc tlocks low");
195 jfs_tlocks_low = 1;
196 wake_up_process(jfsSyncThread);
197 }
198
199 return lid;
200 }
201
txLockFree(lid_t lid)202 static void txLockFree(lid_t lid)
203 {
204 TxLock[lid].tid = 0;
205 TxLock[lid].next = TxAnchor.freelock;
206 TxAnchor.freelock = lid;
207 TxAnchor.tlocksInUse--;
208 if (jfs_tlocks_low && (TxAnchor.tlocksInUse < TxLockLWM)) {
209 jfs_info("txLockFree jfs_tlocks_low no more");
210 jfs_tlocks_low = 0;
211 TXN_WAKEUP(&TxAnchor.lowlockwait);
212 }
213 TXN_WAKEUP(&TxAnchor.freelockwait);
214 }
215
216 /*
217 * NAME: txInit()
218 *
219 * FUNCTION: initialize transaction management structures
220 *
221 * RETURN:
222 *
223 * serialization: single thread at jfs_init()
224 */
txInit(void)225 int txInit(void)
226 {
227 int k, size;
228 struct sysinfo si;
229
230 /* Set defaults for nTxLock and nTxBlock if unset */
231
232 if (nTxLock == -1) {
233 if (nTxBlock == -1) {
234 /* Base default on memory size */
235 si_meminfo(&si);
236 if (si.totalram > (256 * 1024)) /* 1 GB */
237 nTxLock = 64 * 1024;
238 else
239 nTxLock = si.totalram >> 2;
240 } else if (nTxBlock > (8 * 1024))
241 nTxLock = 64 * 1024;
242 else
243 nTxLock = nTxBlock << 3;
244 }
245 if (nTxBlock == -1)
246 nTxBlock = nTxLock >> 3;
247
248 /* Verify tunable parameters */
249 if (nTxBlock < 16)
250 nTxBlock = 16; /* No one should set it this low */
251 if (nTxBlock > 65536)
252 nTxBlock = 65536;
253 if (nTxLock < 256)
254 nTxLock = 256; /* No one should set it this low */
255 if (nTxLock > 65536)
256 nTxLock = 65536;
257
258 printk(KERN_INFO "JFS: nTxBlock = %d, nTxLock = %d\n",
259 nTxBlock, nTxLock);
260 /*
261 * initialize transaction block (tblock) table
262 *
263 * transaction id (tid) = tblock index
264 * tid = 0 is reserved.
265 */
266 TxLockLWM = (nTxLock * 4) / 10;
267 TxLockHWM = (nTxLock * 7) / 10;
268 TxLockVHWM = (nTxLock * 8) / 10;
269
270 size = sizeof(struct tblock) * nTxBlock;
271 TxBlock = vmalloc(size);
272 if (TxBlock == NULL)
273 return -ENOMEM;
274
275 for (k = 1; k < nTxBlock - 1; k++) {
276 TxBlock[k].next = k + 1;
277 init_waitqueue_head(&TxBlock[k].gcwait);
278 init_waitqueue_head(&TxBlock[k].waitor);
279 }
280 TxBlock[k].next = 0;
281 init_waitqueue_head(&TxBlock[k].gcwait);
282 init_waitqueue_head(&TxBlock[k].waitor);
283
284 TxAnchor.freetid = 1;
285 init_waitqueue_head(&TxAnchor.freewait);
286
287 stattx.maxtid = 1; /* statistics */
288
289 /*
290 * initialize transaction lock (tlock) table
291 *
292 * transaction lock id = tlock index
293 * tlock id = 0 is reserved.
294 */
295 size = sizeof(struct tlock) * nTxLock;
296 TxLock = vmalloc(size);
297 if (TxLock == NULL) {
298 vfree(TxBlock);
299 return -ENOMEM;
300 }
301
302 /* initialize tlock table */
303 for (k = 1; k < nTxLock - 1; k++)
304 TxLock[k].next = k + 1;
305 TxLock[k].next = 0;
306 init_waitqueue_head(&TxAnchor.freelockwait);
307 init_waitqueue_head(&TxAnchor.lowlockwait);
308
309 TxAnchor.freelock = 1;
310 TxAnchor.tlocksInUse = 0;
311 INIT_LIST_HEAD(&TxAnchor.anon_list);
312 INIT_LIST_HEAD(&TxAnchor.anon_list2);
313
314 LAZY_LOCK_INIT();
315 INIT_LIST_HEAD(&TxAnchor.unlock_queue);
316
317 stattx.maxlid = 1; /* statistics */
318
319 return 0;
320 }
321
322 /*
323 * NAME: txExit()
324 *
325 * FUNCTION: clean up when module is unloaded
326 */
txExit(void)327 void txExit(void)
328 {
329 vfree(TxLock);
330 TxLock = NULL;
331 vfree(TxBlock);
332 TxBlock = NULL;
333 }
334
335 /*
336 * NAME: txBegin()
337 *
338 * FUNCTION: start a transaction.
339 *
340 * PARAMETER: sb - superblock
341 * flag - force for nested tx;
342 *
343 * RETURN: tid - transaction id
344 *
345 * note: flag force allows to start tx for nested tx
346 * to prevent deadlock on logsync barrier;
347 */
txBegin(struct super_block * sb,int flag)348 tid_t txBegin(struct super_block *sb, int flag)
349 {
350 tid_t t;
351 struct tblock *tblk;
352 struct jfs_log *log;
353
354 jfs_info("txBegin: flag = 0x%x", flag);
355 log = JFS_SBI(sb)->log;
356
357 if (!log) {
358 jfs_error(sb, "read-only filesystem\n");
359 return 0;
360 }
361
362 TXN_LOCK();
363
364 INCREMENT(TxStat.txBegin);
365
366 retry:
367 if (!(flag & COMMIT_FORCE)) {
368 /*
369 * synchronize with logsync barrier
370 */
371 if (test_bit(log_SYNCBARRIER, &log->flag) ||
372 test_bit(log_QUIESCE, &log->flag)) {
373 INCREMENT(TxStat.txBegin_barrier);
374 TXN_SLEEP(&log->syncwait);
375 goto retry;
376 }
377 }
378 if (flag == 0) {
379 /*
380 * Don't begin transaction if we're getting starved for tlocks
381 * unless COMMIT_FORCE or COMMIT_INODE (which may ultimately
382 * free tlocks)
383 */
384 if (TxAnchor.tlocksInUse > TxLockVHWM) {
385 INCREMENT(TxStat.txBegin_lockslow);
386 TXN_SLEEP(&TxAnchor.lowlockwait);
387 goto retry;
388 }
389 }
390
391 /*
392 * allocate transaction id/block
393 */
394 if ((t = TxAnchor.freetid) == 0) {
395 jfs_info("txBegin: waiting for free tid");
396 INCREMENT(TxStat.txBegin_freetid);
397 TXN_SLEEP(&TxAnchor.freewait);
398 goto retry;
399 }
400
401 tblk = tid_to_tblock(t);
402
403 if ((tblk->next == 0) && !(flag & COMMIT_FORCE)) {
404 /* Don't let a non-forced transaction take the last tblk */
405 jfs_info("txBegin: waiting for free tid");
406 INCREMENT(TxStat.txBegin_freetid);
407 TXN_SLEEP(&TxAnchor.freewait);
408 goto retry;
409 }
410
411 TxAnchor.freetid = tblk->next;
412
413 /*
414 * initialize transaction
415 */
416
417 /*
418 * We can't zero the whole thing or we screw up another thread being
419 * awakened after sleeping on tblk->waitor
420 *
421 * memset(tblk, 0, sizeof(struct tblock));
422 */
423 tblk->next = tblk->last = tblk->xflag = tblk->flag = tblk->lsn = 0;
424
425 tblk->sb = sb;
426 ++log->logtid;
427 tblk->logtid = log->logtid;
428
429 ++log->active;
430
431 HIGHWATERMARK(stattx.maxtid, t); /* statistics */
432 INCREMENT(stattx.ntid); /* statistics */
433
434 TXN_UNLOCK();
435
436 jfs_info("txBegin: returning tid = %d", t);
437
438 return t;
439 }
440
441 /*
442 * NAME: txBeginAnon()
443 *
444 * FUNCTION: start an anonymous transaction.
445 * Blocks if logsync or available tlocks are low to prevent
446 * anonymous tlocks from depleting supply.
447 *
448 * PARAMETER: sb - superblock
449 *
450 * RETURN: none
451 */
txBeginAnon(struct super_block * sb)452 void txBeginAnon(struct super_block *sb)
453 {
454 struct jfs_log *log;
455
456 log = JFS_SBI(sb)->log;
457
458 TXN_LOCK();
459 INCREMENT(TxStat.txBeginAnon);
460
461 retry:
462 /*
463 * synchronize with logsync barrier
464 */
465 if (test_bit(log_SYNCBARRIER, &log->flag) ||
466 test_bit(log_QUIESCE, &log->flag)) {
467 INCREMENT(TxStat.txBeginAnon_barrier);
468 TXN_SLEEP(&log->syncwait);
469 goto retry;
470 }
471
472 /*
473 * Don't begin transaction if we're getting starved for tlocks
474 */
475 if (TxAnchor.tlocksInUse > TxLockVHWM) {
476 INCREMENT(TxStat.txBeginAnon_lockslow);
477 TXN_SLEEP(&TxAnchor.lowlockwait);
478 goto retry;
479 }
480 TXN_UNLOCK();
481 }
482
483 /*
484 * txEnd()
485 *
486 * function: free specified transaction block.
487 *
488 * logsync barrier processing:
489 *
490 * serialization:
491 */
txEnd(tid_t tid)492 void txEnd(tid_t tid)
493 {
494 struct tblock *tblk = tid_to_tblock(tid);
495 struct jfs_log *log;
496
497 jfs_info("txEnd: tid = %d", tid);
498 TXN_LOCK();
499
500 /*
501 * wakeup transactions waiting on the page locked
502 * by the current transaction
503 */
504 TXN_WAKEUP(&tblk->waitor);
505
506 log = JFS_SBI(tblk->sb)->log;
507
508 /*
509 * Lazy commit thread can't free this guy until we mark it UNLOCKED,
510 * otherwise, we would be left with a transaction that may have been
511 * reused.
512 *
513 * Lazy commit thread will turn off tblkGC_LAZY before calling this
514 * routine.
515 */
516 if (tblk->flag & tblkGC_LAZY) {
517 jfs_info("txEnd called w/lazy tid: %d, tblk = 0x%p", tid, tblk);
518 TXN_UNLOCK();
519
520 spin_lock_irq(&log->gclock); // LOGGC_LOCK
521 tblk->flag |= tblkGC_UNLOCKED;
522 spin_unlock_irq(&log->gclock); // LOGGC_UNLOCK
523 return;
524 }
525
526 jfs_info("txEnd: tid: %d, tblk = 0x%p", tid, tblk);
527
528 assert(tblk->next == 0);
529
530 /*
531 * insert tblock back on freelist
532 */
533 tblk->next = TxAnchor.freetid;
534 TxAnchor.freetid = tid;
535
536 /*
537 * mark the tblock not active
538 */
539 if (--log->active == 0) {
540 clear_bit(log_FLUSH, &log->flag);
541
542 /*
543 * synchronize with logsync barrier
544 */
545 if (test_bit(log_SYNCBARRIER, &log->flag)) {
546 TXN_UNLOCK();
547
548 /* write dirty metadata & forward log syncpt */
549 jfs_syncpt(log, 1);
550
551 jfs_info("log barrier off: 0x%x", log->lsn);
552
553 /* enable new transactions start */
554 clear_bit(log_SYNCBARRIER, &log->flag);
555
556 /* wakeup all waitors for logsync barrier */
557 TXN_WAKEUP(&log->syncwait);
558
559 goto wakeup;
560 }
561 }
562
563 TXN_UNLOCK();
564 wakeup:
565 /*
566 * wakeup all waitors for a free tblock
567 */
568 TXN_WAKEUP(&TxAnchor.freewait);
569 }
570
571 /*
572 * txLock()
573 *
574 * function: acquire a transaction lock on the specified <mp>
575 *
576 * parameter:
577 *
578 * return: transaction lock id
579 *
580 * serialization:
581 */
txLock(tid_t tid,struct inode * ip,struct metapage * mp,int type)582 struct tlock *txLock(tid_t tid, struct inode *ip, struct metapage * mp,
583 int type)
584 {
585 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
586 int dir_xtree = 0;
587 lid_t lid;
588 tid_t xtid;
589 struct tlock *tlck;
590 struct xtlock *xtlck;
591 struct linelock *linelock;
592 xtpage_t *p;
593 struct tblock *tblk;
594
595 TXN_LOCK();
596
597 if (S_ISDIR(ip->i_mode) && (type & tlckXTREE) &&
598 !(mp->xflag & COMMIT_PAGE)) {
599 /*
600 * Directory inode is special. It can have both an xtree tlock
601 * and a dtree tlock associated with it.
602 */
603 dir_xtree = 1;
604 lid = jfs_ip->xtlid;
605 } else
606 lid = mp->lid;
607
608 /* is page not locked by a transaction ? */
609 if (lid == 0)
610 goto allocateLock;
611
612 jfs_info("txLock: tid:%d ip:0x%p mp:0x%p lid:%d", tid, ip, mp, lid);
613
614 /* is page locked by the requester transaction ? */
615 tlck = lid_to_tlock(lid);
616 if ((xtid = tlck->tid) == tid) {
617 TXN_UNLOCK();
618 goto grantLock;
619 }
620
621 /*
622 * is page locked by anonymous transaction/lock ?
623 *
624 * (page update without transaction (i.e., file write) is
625 * locked under anonymous transaction tid = 0:
626 * anonymous tlocks maintained on anonymous tlock list of
627 * the inode of the page and available to all anonymous
628 * transactions until txCommit() time at which point
629 * they are transferred to the transaction tlock list of
630 * the committing transaction of the inode)
631 */
632 if (xtid == 0) {
633 tlck->tid = tid;
634 TXN_UNLOCK();
635 tblk = tid_to_tblock(tid);
636 /*
637 * The order of the tlocks in the transaction is important
638 * (during truncate, child xtree pages must be freed before
639 * parent's tlocks change the working map).
640 * Take tlock off anonymous list and add to tail of
641 * transaction list
642 *
643 * Note: We really need to get rid of the tid & lid and
644 * use list_head's. This code is getting UGLY!
645 */
646 if (jfs_ip->atlhead == lid) {
647 if (jfs_ip->atltail == lid) {
648 /* only anonymous txn.
649 * Remove from anon_list
650 */
651 TXN_LOCK();
652 list_del_init(&jfs_ip->anon_inode_list);
653 TXN_UNLOCK();
654 }
655 jfs_ip->atlhead = tlck->next;
656 } else {
657 lid_t last;
658 for (last = jfs_ip->atlhead;
659 lid_to_tlock(last)->next != lid;
660 last = lid_to_tlock(last)->next) {
661 assert(last);
662 }
663 lid_to_tlock(last)->next = tlck->next;
664 if (jfs_ip->atltail == lid)
665 jfs_ip->atltail = last;
666 }
667
668 /* insert the tlock at tail of transaction tlock list */
669
670 if (tblk->next)
671 lid_to_tlock(tblk->last)->next = lid;
672 else
673 tblk->next = lid;
674 tlck->next = 0;
675 tblk->last = lid;
676
677 goto grantLock;
678 }
679
680 goto waitLock;
681
682 /*
683 * allocate a tlock
684 */
685 allocateLock:
686 lid = txLockAlloc();
687 tlck = lid_to_tlock(lid);
688
689 /*
690 * initialize tlock
691 */
692 tlck->tid = tid;
693
694 TXN_UNLOCK();
695
696 /* mark tlock for meta-data page */
697 if (mp->xflag & COMMIT_PAGE) {
698
699 tlck->flag = tlckPAGELOCK;
700
701 /* mark the page dirty and nohomeok */
702 metapage_nohomeok(mp);
703
704 jfs_info("locking mp = 0x%p, nohomeok = %d tid = %d tlck = 0x%p",
705 mp, mp->nohomeok, tid, tlck);
706
707 /* if anonymous transaction, and buffer is on the group
708 * commit synclist, mark inode to show this. This will
709 * prevent the buffer from being marked nohomeok for too
710 * long a time.
711 */
712 if ((tid == 0) && mp->lsn)
713 set_cflag(COMMIT_Synclist, ip);
714 }
715 /* mark tlock for in-memory inode */
716 else
717 tlck->flag = tlckINODELOCK;
718
719 if (S_ISDIR(ip->i_mode))
720 tlck->flag |= tlckDIRECTORY;
721
722 tlck->type = 0;
723
724 /* bind the tlock and the page */
725 tlck->ip = ip;
726 tlck->mp = mp;
727 if (dir_xtree)
728 jfs_ip->xtlid = lid;
729 else
730 mp->lid = lid;
731
732 /*
733 * enqueue transaction lock to transaction/inode
734 */
735 /* insert the tlock at tail of transaction tlock list */
736 if (tid) {
737 tblk = tid_to_tblock(tid);
738 if (tblk->next)
739 lid_to_tlock(tblk->last)->next = lid;
740 else
741 tblk->next = lid;
742 tlck->next = 0;
743 tblk->last = lid;
744 }
745 /* anonymous transaction:
746 * insert the tlock at head of inode anonymous tlock list
747 */
748 else {
749 tlck->next = jfs_ip->atlhead;
750 jfs_ip->atlhead = lid;
751 if (tlck->next == 0) {
752 /* This inode's first anonymous transaction */
753 jfs_ip->atltail = lid;
754 TXN_LOCK();
755 list_add_tail(&jfs_ip->anon_inode_list,
756 &TxAnchor.anon_list);
757 TXN_UNLOCK();
758 }
759 }
760
761 /* initialize type dependent area for linelock */
762 linelock = (struct linelock *) & tlck->lock;
763 linelock->next = 0;
764 linelock->flag = tlckLINELOCK;
765 linelock->maxcnt = TLOCKSHORT;
766 linelock->index = 0;
767
768 switch (type & tlckTYPE) {
769 case tlckDTREE:
770 linelock->l2linesize = L2DTSLOTSIZE;
771 break;
772
773 case tlckXTREE:
774 linelock->l2linesize = L2XTSLOTSIZE;
775
776 xtlck = (struct xtlock *) linelock;
777 xtlck->header.offset = 0;
778 xtlck->header.length = 2;
779
780 if (type & tlckNEW) {
781 xtlck->lwm.offset = XTENTRYSTART;
782 } else {
783 if (mp->xflag & COMMIT_PAGE)
784 p = (xtpage_t *) mp->data;
785 else
786 p = &jfs_ip->i_xtroot;
787 xtlck->lwm.offset =
788 le16_to_cpu(p->header.nextindex);
789 }
790 xtlck->lwm.length = 0; /* ! */
791 xtlck->twm.offset = 0;
792 xtlck->hwm.offset = 0;
793
794 xtlck->index = 2;
795 break;
796
797 case tlckINODE:
798 linelock->l2linesize = L2INODESLOTSIZE;
799 break;
800
801 case tlckDATA:
802 linelock->l2linesize = L2DATASLOTSIZE;
803 break;
804
805 default:
806 jfs_err("UFO tlock:0x%p", tlck);
807 }
808
809 /*
810 * update tlock vector
811 */
812 grantLock:
813 tlck->type |= type;
814
815 return tlck;
816
817 /*
818 * page is being locked by another transaction:
819 */
820 waitLock:
821 /* Only locks on ipimap or ipaimap should reach here */
822 /* assert(jfs_ip->fileset == AGGREGATE_I); */
823 if (jfs_ip->fileset != AGGREGATE_I) {
824 printk(KERN_ERR "txLock: trying to lock locked page!");
825 print_hex_dump(KERN_ERR, "ip: ", DUMP_PREFIX_ADDRESS, 16, 4,
826 ip, sizeof(*ip), 0);
827 print_hex_dump(KERN_ERR, "mp: ", DUMP_PREFIX_ADDRESS, 16, 4,
828 mp, sizeof(*mp), 0);
829 print_hex_dump(KERN_ERR, "Locker's tblock: ",
830 DUMP_PREFIX_ADDRESS, 16, 4, tid_to_tblock(tid),
831 sizeof(struct tblock), 0);
832 print_hex_dump(KERN_ERR, "Tlock: ", DUMP_PREFIX_ADDRESS, 16, 4,
833 tlck, sizeof(*tlck), 0);
834 BUG();
835 }
836 INCREMENT(stattx.waitlock); /* statistics */
837 TXN_UNLOCK();
838 release_metapage(mp);
839 TXN_LOCK();
840 xtid = tlck->tid; /* reacquire after dropping TXN_LOCK */
841
842 jfs_info("txLock: in waitLock, tid = %d, xtid = %d, lid = %d",
843 tid, xtid, lid);
844
845 /* Recheck everything since dropping TXN_LOCK */
846 if (xtid && (tlck->mp == mp) && (mp->lid == lid))
847 TXN_SLEEP_DROP_LOCK(&tid_to_tblock(xtid)->waitor);
848 else
849 TXN_UNLOCK();
850 jfs_info("txLock: awakened tid = %d, lid = %d", tid, lid);
851
852 return NULL;
853 }
854
855 /*
856 * NAME: txRelease()
857 *
858 * FUNCTION: Release buffers associated with transaction locks, but don't
859 * mark homeok yet. The allows other transactions to modify
860 * buffers, but won't let them go to disk until commit record
861 * actually gets written.
862 *
863 * PARAMETER:
864 * tblk -
865 *
866 * RETURN: Errors from subroutines.
867 */
txRelease(struct tblock * tblk)868 static void txRelease(struct tblock * tblk)
869 {
870 struct metapage *mp;
871 lid_t lid;
872 struct tlock *tlck;
873
874 TXN_LOCK();
875
876 for (lid = tblk->next; lid; lid = tlck->next) {
877 tlck = lid_to_tlock(lid);
878 if ((mp = tlck->mp) != NULL &&
879 (tlck->type & tlckBTROOT) == 0) {
880 assert(mp->xflag & COMMIT_PAGE);
881 mp->lid = 0;
882 }
883 }
884
885 /*
886 * wakeup transactions waiting on a page locked
887 * by the current transaction
888 */
889 TXN_WAKEUP(&tblk->waitor);
890
891 TXN_UNLOCK();
892 }
893
894 /*
895 * NAME: txUnlock()
896 *
897 * FUNCTION: Initiates pageout of pages modified by tid in journalled
898 * objects and frees their lockwords.
899 */
txUnlock(struct tblock * tblk)900 static void txUnlock(struct tblock * tblk)
901 {
902 struct tlock *tlck;
903 struct linelock *linelock;
904 lid_t lid, next, llid, k;
905 struct metapage *mp;
906 struct jfs_log *log;
907 int difft, diffp;
908 unsigned long flags;
909
910 jfs_info("txUnlock: tblk = 0x%p", tblk);
911 log = JFS_SBI(tblk->sb)->log;
912
913 /*
914 * mark page under tlock homeok (its log has been written):
915 */
916 for (lid = tblk->next; lid; lid = next) {
917 tlck = lid_to_tlock(lid);
918 next = tlck->next;
919
920 jfs_info("unlocking lid = %d, tlck = 0x%p", lid, tlck);
921
922 /* unbind page from tlock */
923 if ((mp = tlck->mp) != NULL &&
924 (tlck->type & tlckBTROOT) == 0) {
925 assert(mp->xflag & COMMIT_PAGE);
926
927 /* hold buffer
928 */
929 hold_metapage(mp);
930
931 assert(mp->nohomeok > 0);
932 _metapage_homeok(mp);
933
934 /* inherit younger/larger clsn */
935 LOGSYNC_LOCK(log, flags);
936 if (mp->clsn) {
937 logdiff(difft, tblk->clsn, log);
938 logdiff(diffp, mp->clsn, log);
939 if (difft > diffp)
940 mp->clsn = tblk->clsn;
941 } else
942 mp->clsn = tblk->clsn;
943 LOGSYNC_UNLOCK(log, flags);
944
945 assert(!(tlck->flag & tlckFREEPAGE));
946
947 put_metapage(mp);
948 }
949
950 /* insert tlock, and linelock(s) of the tlock if any,
951 * at head of freelist
952 */
953 TXN_LOCK();
954
955 llid = ((struct linelock *) & tlck->lock)->next;
956 while (llid) {
957 linelock = (struct linelock *) lid_to_tlock(llid);
958 k = linelock->next;
959 txLockFree(llid);
960 llid = k;
961 }
962 txLockFree(lid);
963
964 TXN_UNLOCK();
965 }
966 tblk->next = tblk->last = 0;
967
968 /*
969 * remove tblock from logsynclist
970 * (allocation map pages inherited lsn of tblk and
971 * has been inserted in logsync list at txUpdateMap())
972 */
973 if (tblk->lsn) {
974 LOGSYNC_LOCK(log, flags);
975 log->count--;
976 list_del(&tblk->synclist);
977 LOGSYNC_UNLOCK(log, flags);
978 }
979 }
980
981 /*
982 * txMaplock()
983 *
984 * function: allocate a transaction lock for freed page/entry;
985 * for freed page, maplock is used as xtlock/dtlock type;
986 */
txMaplock(tid_t tid,struct inode * ip,int type)987 struct tlock *txMaplock(tid_t tid, struct inode *ip, int type)
988 {
989 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
990 lid_t lid;
991 struct tblock *tblk;
992 struct tlock *tlck;
993 struct maplock *maplock;
994
995 TXN_LOCK();
996
997 /*
998 * allocate a tlock
999 */
1000 lid = txLockAlloc();
1001 tlck = lid_to_tlock(lid);
1002
1003 /*
1004 * initialize tlock
1005 */
1006 tlck->tid = tid;
1007
1008 /* bind the tlock and the object */
1009 tlck->flag = tlckINODELOCK;
1010 if (S_ISDIR(ip->i_mode))
1011 tlck->flag |= tlckDIRECTORY;
1012 tlck->ip = ip;
1013 tlck->mp = NULL;
1014
1015 tlck->type = type;
1016
1017 /*
1018 * enqueue transaction lock to transaction/inode
1019 */
1020 /* insert the tlock at tail of transaction tlock list */
1021 if (tid) {
1022 tblk = tid_to_tblock(tid);
1023 if (tblk->next)
1024 lid_to_tlock(tblk->last)->next = lid;
1025 else
1026 tblk->next = lid;
1027 tlck->next = 0;
1028 tblk->last = lid;
1029 }
1030 /* anonymous transaction:
1031 * insert the tlock at head of inode anonymous tlock list
1032 */
1033 else {
1034 tlck->next = jfs_ip->atlhead;
1035 jfs_ip->atlhead = lid;
1036 if (tlck->next == 0) {
1037 /* This inode's first anonymous transaction */
1038 jfs_ip->atltail = lid;
1039 list_add_tail(&jfs_ip->anon_inode_list,
1040 &TxAnchor.anon_list);
1041 }
1042 }
1043
1044 TXN_UNLOCK();
1045
1046 /* initialize type dependent area for maplock */
1047 maplock = (struct maplock *) & tlck->lock;
1048 maplock->next = 0;
1049 maplock->maxcnt = 0;
1050 maplock->index = 0;
1051
1052 return tlck;
1053 }
1054
1055 /*
1056 * txLinelock()
1057 *
1058 * function: allocate a transaction lock for log vector list
1059 */
txLinelock(struct linelock * tlock)1060 struct linelock *txLinelock(struct linelock * tlock)
1061 {
1062 lid_t lid;
1063 struct tlock *tlck;
1064 struct linelock *linelock;
1065
1066 TXN_LOCK();
1067
1068 /* allocate a TxLock structure */
1069 lid = txLockAlloc();
1070 tlck = lid_to_tlock(lid);
1071
1072 TXN_UNLOCK();
1073
1074 /* initialize linelock */
1075 linelock = (struct linelock *) tlck;
1076 linelock->next = 0;
1077 linelock->flag = tlckLINELOCK;
1078 linelock->maxcnt = TLOCKLONG;
1079 linelock->index = 0;
1080 if (tlck->flag & tlckDIRECTORY)
1081 linelock->flag |= tlckDIRECTORY;
1082
1083 /* append linelock after tlock */
1084 linelock->next = tlock->next;
1085 tlock->next = lid;
1086
1087 return linelock;
1088 }
1089
1090 /*
1091 * transaction commit management
1092 * -----------------------------
1093 */
1094
1095 /*
1096 * NAME: txCommit()
1097 *
1098 * FUNCTION: commit the changes to the objects specified in
1099 * clist. For journalled segments only the
1100 * changes of the caller are committed, ie by tid.
1101 * for non-journalled segments the data are flushed to
1102 * disk and then the change to the disk inode and indirect
1103 * blocks committed (so blocks newly allocated to the
1104 * segment will be made a part of the segment atomically).
1105 *
1106 * all of the segments specified in clist must be in
1107 * one file system. no more than 6 segments are needed
1108 * to handle all unix svcs.
1109 *
1110 * if the i_nlink field (i.e. disk inode link count)
1111 * is zero, and the type of inode is a regular file or
1112 * directory, or symbolic link , the inode is truncated
1113 * to zero length. the truncation is committed but the
1114 * VM resources are unaffected until it is closed (see
1115 * iput and iclose).
1116 *
1117 * PARAMETER:
1118 *
1119 * RETURN:
1120 *
1121 * serialization:
1122 * on entry the inode lock on each segment is assumed
1123 * to be held.
1124 *
1125 * i/o error:
1126 */
txCommit(tid_t tid,int nip,struct inode ** iplist,int flag)1127 int txCommit(tid_t tid, /* transaction identifier */
1128 int nip, /* number of inodes to commit */
1129 struct inode **iplist, /* list of inode to commit */
1130 int flag)
1131 {
1132 int rc = 0;
1133 struct commit cd;
1134 struct jfs_log *log;
1135 struct tblock *tblk;
1136 struct lrd *lrd;
1137 struct inode *ip;
1138 struct jfs_inode_info *jfs_ip;
1139 int k, n;
1140 ino_t top;
1141 struct super_block *sb;
1142
1143 jfs_info("txCommit, tid = %d, flag = %d", tid, flag);
1144 /* is read-only file system ? */
1145 if (isReadOnly(iplist[0])) {
1146 rc = -EROFS;
1147 goto TheEnd;
1148 }
1149
1150 sb = cd.sb = iplist[0]->i_sb;
1151 cd.tid = tid;
1152
1153 if (tid == 0)
1154 tid = txBegin(sb, 0);
1155 tblk = tid_to_tblock(tid);
1156
1157 /*
1158 * initialize commit structure
1159 */
1160 log = JFS_SBI(sb)->log;
1161 cd.log = log;
1162
1163 /* initialize log record descriptor in commit */
1164 lrd = &cd.lrd;
1165 lrd->logtid = cpu_to_le32(tblk->logtid);
1166 lrd->backchain = 0;
1167
1168 tblk->xflag |= flag;
1169
1170 if ((flag & (COMMIT_FORCE | COMMIT_SYNC)) == 0)
1171 tblk->xflag |= COMMIT_LAZY;
1172 /*
1173 * prepare non-journaled objects for commit
1174 *
1175 * flush data pages of non-journaled file
1176 * to prevent the file getting non-initialized disk blocks
1177 * in case of crash.
1178 * (new blocks - )
1179 */
1180 cd.iplist = iplist;
1181 cd.nip = nip;
1182
1183 /*
1184 * acquire transaction lock on (on-disk) inodes
1185 *
1186 * update on-disk inode from in-memory inode
1187 * acquiring transaction locks for AFTER records
1188 * on the on-disk inode of file object
1189 *
1190 * sort the inodes array by inode number in descending order
1191 * to prevent deadlock when acquiring transaction lock
1192 * of on-disk inodes on multiple on-disk inode pages by
1193 * multiple concurrent transactions
1194 */
1195 for (k = 0; k < cd.nip; k++) {
1196 top = (cd.iplist[k])->i_ino;
1197 for (n = k + 1; n < cd.nip; n++) {
1198 ip = cd.iplist[n];
1199 if (ip->i_ino > top) {
1200 top = ip->i_ino;
1201 cd.iplist[n] = cd.iplist[k];
1202 cd.iplist[k] = ip;
1203 }
1204 }
1205
1206 ip = cd.iplist[k];
1207 jfs_ip = JFS_IP(ip);
1208
1209 /*
1210 * BUGBUG - This code has temporarily been removed. The
1211 * intent is to ensure that any file data is written before
1212 * the metadata is committed to the journal. This prevents
1213 * uninitialized data from appearing in a file after the
1214 * journal has been replayed. (The uninitialized data
1215 * could be sensitive data removed by another user.)
1216 *
1217 * The problem now is that we are holding the IWRITELOCK
1218 * on the inode, and calling filemap_fdatawrite on an
1219 * unmapped page will cause a deadlock in jfs_get_block.
1220 *
1221 * The long term solution is to pare down the use of
1222 * IWRITELOCK. We are currently holding it too long.
1223 * We could also be smarter about which data pages need
1224 * to be written before the transaction is committed and
1225 * when we don't need to worry about it at all.
1226 *
1227 * if ((!S_ISDIR(ip->i_mode))
1228 * && (tblk->flag & COMMIT_DELETE) == 0)
1229 * filemap_write_and_wait(ip->i_mapping);
1230 */
1231
1232 /*
1233 * Mark inode as not dirty. It will still be on the dirty
1234 * inode list, but we'll know not to commit it again unless
1235 * it gets marked dirty again
1236 */
1237 clear_cflag(COMMIT_Dirty, ip);
1238
1239 /* inherit anonymous tlock(s) of inode */
1240 if (jfs_ip->atlhead) {
1241 lid_to_tlock(jfs_ip->atltail)->next = tblk->next;
1242 tblk->next = jfs_ip->atlhead;
1243 if (!tblk->last)
1244 tblk->last = jfs_ip->atltail;
1245 jfs_ip->atlhead = jfs_ip->atltail = 0;
1246 TXN_LOCK();
1247 list_del_init(&jfs_ip->anon_inode_list);
1248 TXN_UNLOCK();
1249 }
1250
1251 /*
1252 * acquire transaction lock on on-disk inode page
1253 * (become first tlock of the tblk's tlock list)
1254 */
1255 if (((rc = diWrite(tid, ip))))
1256 goto out;
1257 }
1258
1259 /*
1260 * write log records from transaction locks
1261 *
1262 * txUpdateMap() resets XAD_NEW in XAD.
1263 */
1264 if ((rc = txLog(log, tblk, &cd)))
1265 goto TheEnd;
1266
1267 /*
1268 * Ensure that inode isn't reused before
1269 * lazy commit thread finishes processing
1270 */
1271 if (tblk->xflag & COMMIT_DELETE) {
1272 ihold(tblk->u.ip);
1273 /*
1274 * Avoid a rare deadlock
1275 *
1276 * If the inode is locked, we may be blocked in
1277 * jfs_commit_inode. If so, we don't want the
1278 * lazy_commit thread doing the last iput() on the inode
1279 * since that may block on the locked inode. Instead,
1280 * commit the transaction synchronously, so the last iput
1281 * will be done by the calling thread (or later)
1282 */
1283 /*
1284 * I believe this code is no longer needed. Splitting I_LOCK
1285 * into two bits, I_NEW and I_SYNC should prevent this
1286 * deadlock as well. But since I don't have a JFS testload
1287 * to verify this, only a trivial s/I_LOCK/I_SYNC/ was done.
1288 * Joern
1289 */
1290 if (tblk->u.ip->i_state & I_SYNC)
1291 tblk->xflag &= ~COMMIT_LAZY;
1292 }
1293
1294 ASSERT((!(tblk->xflag & COMMIT_DELETE)) ||
1295 ((tblk->u.ip->i_nlink == 0) &&
1296 !test_cflag(COMMIT_Nolink, tblk->u.ip)));
1297
1298 /*
1299 * write COMMIT log record
1300 */
1301 lrd->type = cpu_to_le16(LOG_COMMIT);
1302 lrd->length = 0;
1303 lmLog(log, tblk, lrd, NULL);
1304
1305 lmGroupCommit(log, tblk);
1306
1307 /*
1308 * - transaction is now committed -
1309 */
1310
1311 /*
1312 * force pages in careful update
1313 * (imap addressing structure update)
1314 */
1315 if (flag & COMMIT_FORCE)
1316 txForce(tblk);
1317
1318 /*
1319 * update allocation map.
1320 *
1321 * update inode allocation map and inode:
1322 * free pager lock on memory object of inode if any.
1323 * update block allocation map.
1324 *
1325 * txUpdateMap() resets XAD_NEW in XAD.
1326 */
1327 if (tblk->xflag & COMMIT_FORCE)
1328 txUpdateMap(tblk);
1329
1330 /*
1331 * free transaction locks and pageout/free pages
1332 */
1333 txRelease(tblk);
1334
1335 if ((tblk->flag & tblkGC_LAZY) == 0)
1336 txUnlock(tblk);
1337
1338
1339 /*
1340 * reset in-memory object state
1341 */
1342 for (k = 0; k < cd.nip; k++) {
1343 ip = cd.iplist[k];
1344 jfs_ip = JFS_IP(ip);
1345
1346 /*
1347 * reset in-memory inode state
1348 */
1349 jfs_ip->bxflag = 0;
1350 jfs_ip->blid = 0;
1351 }
1352
1353 out:
1354 if (rc != 0)
1355 txAbort(tid, 1);
1356
1357 TheEnd:
1358 jfs_info("txCommit: tid = %d, returning %d", tid, rc);
1359 return rc;
1360 }
1361
1362 /*
1363 * NAME: txLog()
1364 *
1365 * FUNCTION: Writes AFTER log records for all lines modified
1366 * by tid for segments specified by inodes in comdata.
1367 * Code assumes only WRITELOCKS are recorded in lockwords.
1368 *
1369 * PARAMETERS:
1370 *
1371 * RETURN :
1372 */
txLog(struct jfs_log * log,struct tblock * tblk,struct commit * cd)1373 static int txLog(struct jfs_log * log, struct tblock * tblk, struct commit * cd)
1374 {
1375 int rc = 0;
1376 struct inode *ip;
1377 lid_t lid;
1378 struct tlock *tlck;
1379 struct lrd *lrd = &cd->lrd;
1380
1381 /*
1382 * write log record(s) for each tlock of transaction,
1383 */
1384 for (lid = tblk->next; lid; lid = tlck->next) {
1385 tlck = lid_to_tlock(lid);
1386
1387 tlck->flag |= tlckLOG;
1388
1389 /* initialize lrd common */
1390 ip = tlck->ip;
1391 lrd->aggregate = cpu_to_le32(JFS_SBI(ip->i_sb)->aggregate);
1392 lrd->log.redopage.fileset = cpu_to_le32(JFS_IP(ip)->fileset);
1393 lrd->log.redopage.inode = cpu_to_le32(ip->i_ino);
1394
1395 /* write log record of page from the tlock */
1396 switch (tlck->type & tlckTYPE) {
1397 case tlckXTREE:
1398 xtLog(log, tblk, lrd, tlck);
1399 break;
1400
1401 case tlckDTREE:
1402 dtLog(log, tblk, lrd, tlck);
1403 break;
1404
1405 case tlckINODE:
1406 diLog(log, tblk, lrd, tlck, cd);
1407 break;
1408
1409 case tlckMAP:
1410 mapLog(log, tblk, lrd, tlck);
1411 break;
1412
1413 case tlckDATA:
1414 dataLog(log, tblk, lrd, tlck);
1415 break;
1416
1417 default:
1418 jfs_err("UFO tlock:0x%p", tlck);
1419 }
1420 }
1421
1422 return rc;
1423 }
1424
1425 /*
1426 * diLog()
1427 *
1428 * function: log inode tlock and format maplock to update bmap;
1429 */
diLog(struct jfs_log * log,struct tblock * tblk,struct lrd * lrd,struct tlock * tlck,struct commit * cd)1430 static int diLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1431 struct tlock * tlck, struct commit * cd)
1432 {
1433 int rc = 0;
1434 struct metapage *mp;
1435 pxd_t *pxd;
1436 struct pxd_lock *pxdlock;
1437
1438 mp = tlck->mp;
1439
1440 /* initialize as REDOPAGE record format */
1441 lrd->log.redopage.type = cpu_to_le16(LOG_INODE);
1442 lrd->log.redopage.l2linesize = cpu_to_le16(L2INODESLOTSIZE);
1443
1444 pxd = &lrd->log.redopage.pxd;
1445
1446 /*
1447 * inode after image
1448 */
1449 if (tlck->type & tlckENTRY) {
1450 /* log after-image for logredo(): */
1451 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1452 PXDaddress(pxd, mp->index);
1453 PXDlength(pxd,
1454 mp->logical_size >> tblk->sb->s_blocksize_bits);
1455 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1456
1457 /* mark page as homeward bound */
1458 tlck->flag |= tlckWRITEPAGE;
1459 } else if (tlck->type & tlckFREE) {
1460 /*
1461 * free inode extent
1462 *
1463 * (pages of the freed inode extent have been invalidated and
1464 * a maplock for free of the extent has been formatted at
1465 * txLock() time);
1466 *
1467 * the tlock had been acquired on the inode allocation map page
1468 * (iag) that specifies the freed extent, even though the map
1469 * page is not itself logged, to prevent pageout of the map
1470 * page before the log;
1471 */
1472
1473 /* log LOG_NOREDOINOEXT of the freed inode extent for
1474 * logredo() to start NoRedoPage filters, and to update
1475 * imap and bmap for free of the extent;
1476 */
1477 lrd->type = cpu_to_le16(LOG_NOREDOINOEXT);
1478 /*
1479 * For the LOG_NOREDOINOEXT record, we need
1480 * to pass the IAG number and inode extent
1481 * index (within that IAG) from which the
1482 * the extent being released. These have been
1483 * passed to us in the iplist[1] and iplist[2].
1484 */
1485 lrd->log.noredoinoext.iagnum =
1486 cpu_to_le32((u32) (size_t) cd->iplist[1]);
1487 lrd->log.noredoinoext.inoext_idx =
1488 cpu_to_le32((u32) (size_t) cd->iplist[2]);
1489
1490 pxdlock = (struct pxd_lock *) & tlck->lock;
1491 *pxd = pxdlock->pxd;
1492 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1493
1494 /* update bmap */
1495 tlck->flag |= tlckUPDATEMAP;
1496
1497 /* mark page as homeward bound */
1498 tlck->flag |= tlckWRITEPAGE;
1499 } else
1500 jfs_err("diLog: UFO type tlck:0x%p", tlck);
1501 #ifdef _JFS_WIP
1502 /*
1503 * alloc/free external EA extent
1504 *
1505 * a maplock for txUpdateMap() to update bPWMAP for alloc/free
1506 * of the extent has been formatted at txLock() time;
1507 */
1508 else {
1509 assert(tlck->type & tlckEA);
1510
1511 /* log LOG_UPDATEMAP for logredo() to update bmap for
1512 * alloc of new (and free of old) external EA extent;
1513 */
1514 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1515 pxdlock = (struct pxd_lock *) & tlck->lock;
1516 nlock = pxdlock->index;
1517 for (i = 0; i < nlock; i++, pxdlock++) {
1518 if (pxdlock->flag & mlckALLOCPXD)
1519 lrd->log.updatemap.type =
1520 cpu_to_le16(LOG_ALLOCPXD);
1521 else
1522 lrd->log.updatemap.type =
1523 cpu_to_le16(LOG_FREEPXD);
1524 lrd->log.updatemap.nxd = cpu_to_le16(1);
1525 lrd->log.updatemap.pxd = pxdlock->pxd;
1526 lrd->backchain =
1527 cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1528 }
1529
1530 /* update bmap */
1531 tlck->flag |= tlckUPDATEMAP;
1532 }
1533 #endif /* _JFS_WIP */
1534
1535 return rc;
1536 }
1537
1538 /*
1539 * dataLog()
1540 *
1541 * function: log data tlock
1542 */
dataLog(struct jfs_log * log,struct tblock * tblk,struct lrd * lrd,struct tlock * tlck)1543 static int dataLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1544 struct tlock * tlck)
1545 {
1546 struct metapage *mp;
1547 pxd_t *pxd;
1548
1549 mp = tlck->mp;
1550
1551 /* initialize as REDOPAGE record format */
1552 lrd->log.redopage.type = cpu_to_le16(LOG_DATA);
1553 lrd->log.redopage.l2linesize = cpu_to_le16(L2DATASLOTSIZE);
1554
1555 pxd = &lrd->log.redopage.pxd;
1556
1557 /* log after-image for logredo(): */
1558 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1559
1560 if (jfs_dirtable_inline(tlck->ip)) {
1561 /*
1562 * The table has been truncated, we've must have deleted
1563 * the last entry, so don't bother logging this
1564 */
1565 mp->lid = 0;
1566 grab_metapage(mp);
1567 metapage_homeok(mp);
1568 discard_metapage(mp);
1569 tlck->mp = NULL;
1570 return 0;
1571 }
1572
1573 PXDaddress(pxd, mp->index);
1574 PXDlength(pxd, mp->logical_size >> tblk->sb->s_blocksize_bits);
1575
1576 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1577
1578 /* mark page as homeward bound */
1579 tlck->flag |= tlckWRITEPAGE;
1580
1581 return 0;
1582 }
1583
1584 /*
1585 * dtLog()
1586 *
1587 * function: log dtree tlock and format maplock to update bmap;
1588 */
dtLog(struct jfs_log * log,struct tblock * tblk,struct lrd * lrd,struct tlock * tlck)1589 static void dtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1590 struct tlock * tlck)
1591 {
1592 struct metapage *mp;
1593 struct pxd_lock *pxdlock;
1594 pxd_t *pxd;
1595
1596 mp = tlck->mp;
1597
1598 /* initialize as REDOPAGE/NOREDOPAGE record format */
1599 lrd->log.redopage.type = cpu_to_le16(LOG_DTREE);
1600 lrd->log.redopage.l2linesize = cpu_to_le16(L2DTSLOTSIZE);
1601
1602 pxd = &lrd->log.redopage.pxd;
1603
1604 if (tlck->type & tlckBTROOT)
1605 lrd->log.redopage.type |= cpu_to_le16(LOG_BTROOT);
1606
1607 /*
1608 * page extension via relocation: entry insertion;
1609 * page extension in-place: entry insertion;
1610 * new right page from page split, reinitialized in-line
1611 * root from root page split: entry insertion;
1612 */
1613 if (tlck->type & (tlckNEW | tlckEXTEND)) {
1614 /* log after-image of the new page for logredo():
1615 * mark log (LOG_NEW) for logredo() to initialize
1616 * freelist and update bmap for alloc of the new page;
1617 */
1618 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1619 if (tlck->type & tlckEXTEND)
1620 lrd->log.redopage.type |= cpu_to_le16(LOG_EXTEND);
1621 else
1622 lrd->log.redopage.type |= cpu_to_le16(LOG_NEW);
1623 PXDaddress(pxd, mp->index);
1624 PXDlength(pxd,
1625 mp->logical_size >> tblk->sb->s_blocksize_bits);
1626 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1627
1628 /* format a maplock for txUpdateMap() to update bPMAP for
1629 * alloc of the new page;
1630 */
1631 if (tlck->type & tlckBTROOT)
1632 return;
1633 tlck->flag |= tlckUPDATEMAP;
1634 pxdlock = (struct pxd_lock *) & tlck->lock;
1635 pxdlock->flag = mlckALLOCPXD;
1636 pxdlock->pxd = *pxd;
1637
1638 pxdlock->index = 1;
1639
1640 /* mark page as homeward bound */
1641 tlck->flag |= tlckWRITEPAGE;
1642 return;
1643 }
1644
1645 /*
1646 * entry insertion/deletion,
1647 * sibling page link update (old right page before split);
1648 */
1649 if (tlck->type & (tlckENTRY | tlckRELINK)) {
1650 /* log after-image for logredo(): */
1651 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1652 PXDaddress(pxd, mp->index);
1653 PXDlength(pxd,
1654 mp->logical_size >> tblk->sb->s_blocksize_bits);
1655 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1656
1657 /* mark page as homeward bound */
1658 tlck->flag |= tlckWRITEPAGE;
1659 return;
1660 }
1661
1662 /*
1663 * page deletion: page has been invalidated
1664 * page relocation: source extent
1665 *
1666 * a maplock for free of the page has been formatted
1667 * at txLock() time);
1668 */
1669 if (tlck->type & (tlckFREE | tlckRELOCATE)) {
1670 /* log LOG_NOREDOPAGE of the deleted page for logredo()
1671 * to start NoRedoPage filter and to update bmap for free
1672 * of the deletd page
1673 */
1674 lrd->type = cpu_to_le16(LOG_NOREDOPAGE);
1675 pxdlock = (struct pxd_lock *) & tlck->lock;
1676 *pxd = pxdlock->pxd;
1677 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1678
1679 /* a maplock for txUpdateMap() for free of the page
1680 * has been formatted at txLock() time;
1681 */
1682 tlck->flag |= tlckUPDATEMAP;
1683 }
1684 return;
1685 }
1686
1687 /*
1688 * xtLog()
1689 *
1690 * function: log xtree tlock and format maplock to update bmap;
1691 */
xtLog(struct jfs_log * log,struct tblock * tblk,struct lrd * lrd,struct tlock * tlck)1692 static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1693 struct tlock * tlck)
1694 {
1695 struct inode *ip;
1696 struct metapage *mp;
1697 xtpage_t *p;
1698 struct xtlock *xtlck;
1699 struct maplock *maplock;
1700 struct xdlistlock *xadlock;
1701 struct pxd_lock *pxdlock;
1702 pxd_t *page_pxd;
1703 int next, lwm, hwm;
1704
1705 ip = tlck->ip;
1706 mp = tlck->mp;
1707
1708 /* initialize as REDOPAGE/NOREDOPAGE record format */
1709 lrd->log.redopage.type = cpu_to_le16(LOG_XTREE);
1710 lrd->log.redopage.l2linesize = cpu_to_le16(L2XTSLOTSIZE);
1711
1712 page_pxd = &lrd->log.redopage.pxd;
1713
1714 if (tlck->type & tlckBTROOT) {
1715 lrd->log.redopage.type |= cpu_to_le16(LOG_BTROOT);
1716 p = &JFS_IP(ip)->i_xtroot;
1717 if (S_ISDIR(ip->i_mode))
1718 lrd->log.redopage.type |=
1719 cpu_to_le16(LOG_DIR_XTREE);
1720 } else
1721 p = (xtpage_t *) mp->data;
1722 next = le16_to_cpu(p->header.nextindex);
1723
1724 xtlck = (struct xtlock *) & tlck->lock;
1725
1726 maplock = (struct maplock *) & tlck->lock;
1727 xadlock = (struct xdlistlock *) maplock;
1728
1729 /*
1730 * entry insertion/extension;
1731 * sibling page link update (old right page before split);
1732 */
1733 if (tlck->type & (tlckNEW | tlckGROW | tlckRELINK)) {
1734 /* log after-image for logredo():
1735 * logredo() will update bmap for alloc of new/extended
1736 * extents (XAD_NEW|XAD_EXTEND) of XAD[lwm:next) from
1737 * after-image of XADlist;
1738 * logredo() resets (XAD_NEW|XAD_EXTEND) flag when
1739 * applying the after-image to the meta-data page.
1740 */
1741 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1742 PXDaddress(page_pxd, mp->index);
1743 PXDlength(page_pxd,
1744 mp->logical_size >> tblk->sb->s_blocksize_bits);
1745 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1746
1747 /* format a maplock for txUpdateMap() to update bPMAP
1748 * for alloc of new/extended extents of XAD[lwm:next)
1749 * from the page itself;
1750 * txUpdateMap() resets (XAD_NEW|XAD_EXTEND) flag.
1751 */
1752 lwm = xtlck->lwm.offset;
1753 if (lwm == 0)
1754 lwm = XTPAGEMAXSLOT;
1755
1756 if (lwm == next)
1757 goto out;
1758 if (lwm > next) {
1759 jfs_err("xtLog: lwm > next");
1760 goto out;
1761 }
1762 tlck->flag |= tlckUPDATEMAP;
1763 xadlock->flag = mlckALLOCXADLIST;
1764 xadlock->count = next - lwm;
1765 if ((xadlock->count <= 4) && (tblk->xflag & COMMIT_LAZY)) {
1766 int i;
1767 pxd_t *pxd;
1768 /*
1769 * Lazy commit may allow xtree to be modified before
1770 * txUpdateMap runs. Copy xad into linelock to
1771 * preserve correct data.
1772 *
1773 * We can fit twice as may pxd's as xads in the lock
1774 */
1775 xadlock->flag = mlckALLOCPXDLIST;
1776 pxd = xadlock->xdlist = &xtlck->pxdlock;
1777 for (i = 0; i < xadlock->count; i++) {
1778 PXDaddress(pxd, addressXAD(&p->xad[lwm + i]));
1779 PXDlength(pxd, lengthXAD(&p->xad[lwm + i]));
1780 p->xad[lwm + i].flag &=
1781 ~(XAD_NEW | XAD_EXTENDED);
1782 pxd++;
1783 }
1784 } else {
1785 /*
1786 * xdlist will point to into inode's xtree, ensure
1787 * that transaction is not committed lazily.
1788 */
1789 xadlock->flag = mlckALLOCXADLIST;
1790 xadlock->xdlist = &p->xad[lwm];
1791 tblk->xflag &= ~COMMIT_LAZY;
1792 }
1793 jfs_info("xtLog: alloc ip:0x%p mp:0x%p tlck:0x%p lwm:%d count:%d",
1794 tlck->ip, mp, tlck, lwm, xadlock->count);
1795
1796 maplock->index = 1;
1797
1798 out:
1799 /* mark page as homeward bound */
1800 tlck->flag |= tlckWRITEPAGE;
1801
1802 return;
1803 }
1804
1805 /*
1806 * page deletion: file deletion/truncation (ref. xtTruncate())
1807 *
1808 * (page will be invalidated after log is written and bmap
1809 * is updated from the page);
1810 */
1811 if (tlck->type & tlckFREE) {
1812 /* LOG_NOREDOPAGE log for NoRedoPage filter:
1813 * if page free from file delete, NoRedoFile filter from
1814 * inode image of zero link count will subsume NoRedoPage
1815 * filters for each page;
1816 * if page free from file truncattion, write NoRedoPage
1817 * filter;
1818 *
1819 * upadte of block allocation map for the page itself:
1820 * if page free from deletion and truncation, LOG_UPDATEMAP
1821 * log for the page itself is generated from processing
1822 * its parent page xad entries;
1823 */
1824 /* if page free from file truncation, log LOG_NOREDOPAGE
1825 * of the deleted page for logredo() to start NoRedoPage
1826 * filter for the page;
1827 */
1828 if (tblk->xflag & COMMIT_TRUNCATE) {
1829 /* write NOREDOPAGE for the page */
1830 lrd->type = cpu_to_le16(LOG_NOREDOPAGE);
1831 PXDaddress(page_pxd, mp->index);
1832 PXDlength(page_pxd,
1833 mp->logical_size >> tblk->sb->
1834 s_blocksize_bits);
1835 lrd->backchain =
1836 cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1837
1838 if (tlck->type & tlckBTROOT) {
1839 /* Empty xtree must be logged */
1840 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1841 lrd->backchain =
1842 cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1843 }
1844 }
1845
1846 /* init LOG_UPDATEMAP of the freed extents
1847 * XAD[XTENTRYSTART:hwm) from the deleted page itself
1848 * for logredo() to update bmap;
1849 */
1850 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1851 lrd->log.updatemap.type = cpu_to_le16(LOG_FREEXADLIST);
1852 xtlck = (struct xtlock *) & tlck->lock;
1853 hwm = xtlck->hwm.offset;
1854 lrd->log.updatemap.nxd =
1855 cpu_to_le16(hwm - XTENTRYSTART + 1);
1856 /* reformat linelock for lmLog() */
1857 xtlck->header.offset = XTENTRYSTART;
1858 xtlck->header.length = hwm - XTENTRYSTART + 1;
1859 xtlck->index = 1;
1860 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1861
1862 /* format a maplock for txUpdateMap() to update bmap
1863 * to free extents of XAD[XTENTRYSTART:hwm) from the
1864 * deleted page itself;
1865 */
1866 tlck->flag |= tlckUPDATEMAP;
1867 xadlock->count = hwm - XTENTRYSTART + 1;
1868 if ((xadlock->count <= 4) && (tblk->xflag & COMMIT_LAZY)) {
1869 int i;
1870 pxd_t *pxd;
1871 /*
1872 * Lazy commit may allow xtree to be modified before
1873 * txUpdateMap runs. Copy xad into linelock to
1874 * preserve correct data.
1875 *
1876 * We can fit twice as may pxd's as xads in the lock
1877 */
1878 xadlock->flag = mlckFREEPXDLIST;
1879 pxd = xadlock->xdlist = &xtlck->pxdlock;
1880 for (i = 0; i < xadlock->count; i++) {
1881 PXDaddress(pxd,
1882 addressXAD(&p->xad[XTENTRYSTART + i]));
1883 PXDlength(pxd,
1884 lengthXAD(&p->xad[XTENTRYSTART + i]));
1885 pxd++;
1886 }
1887 } else {
1888 /*
1889 * xdlist will point to into inode's xtree, ensure
1890 * that transaction is not committed lazily.
1891 */
1892 xadlock->flag = mlckFREEXADLIST;
1893 xadlock->xdlist = &p->xad[XTENTRYSTART];
1894 tblk->xflag &= ~COMMIT_LAZY;
1895 }
1896 jfs_info("xtLog: free ip:0x%p mp:0x%p count:%d lwm:2",
1897 tlck->ip, mp, xadlock->count);
1898
1899 maplock->index = 1;
1900
1901 /* mark page as invalid */
1902 if (((tblk->xflag & COMMIT_PWMAP) || S_ISDIR(ip->i_mode))
1903 && !(tlck->type & tlckBTROOT))
1904 tlck->flag |= tlckFREEPAGE;
1905 /*
1906 else (tblk->xflag & COMMIT_PMAP)
1907 ? release the page;
1908 */
1909 return;
1910 }
1911
1912 /*
1913 * page/entry truncation: file truncation (ref. xtTruncate())
1914 *
1915 * |----------+------+------+---------------|
1916 * | | |
1917 * | | hwm - hwm before truncation
1918 * | next - truncation point
1919 * lwm - lwm before truncation
1920 * header ?
1921 */
1922 if (tlck->type & tlckTRUNCATE) {
1923 pxd_t pxd; /* truncated extent of xad */
1924 int twm;
1925
1926 /*
1927 * For truncation the entire linelock may be used, so it would
1928 * be difficult to store xad list in linelock itself.
1929 * Therefore, we'll just force transaction to be committed
1930 * synchronously, so that xtree pages won't be changed before
1931 * txUpdateMap runs.
1932 */
1933 tblk->xflag &= ~COMMIT_LAZY;
1934 lwm = xtlck->lwm.offset;
1935 if (lwm == 0)
1936 lwm = XTPAGEMAXSLOT;
1937 hwm = xtlck->hwm.offset;
1938 twm = xtlck->twm.offset;
1939
1940 /*
1941 * write log records
1942 */
1943 /* log after-image for logredo():
1944 *
1945 * logredo() will update bmap for alloc of new/extended
1946 * extents (XAD_NEW|XAD_EXTEND) of XAD[lwm:next) from
1947 * after-image of XADlist;
1948 * logredo() resets (XAD_NEW|XAD_EXTEND) flag when
1949 * applying the after-image to the meta-data page.
1950 */
1951 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1952 PXDaddress(page_pxd, mp->index);
1953 PXDlength(page_pxd,
1954 mp->logical_size >> tblk->sb->s_blocksize_bits);
1955 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1956
1957 /*
1958 * truncate entry XAD[twm == next - 1]:
1959 */
1960 if (twm == next - 1) {
1961 /* init LOG_UPDATEMAP for logredo() to update bmap for
1962 * free of truncated delta extent of the truncated
1963 * entry XAD[next - 1]:
1964 * (xtlck->pxdlock = truncated delta extent);
1965 */
1966 pxdlock = (struct pxd_lock *) & xtlck->pxdlock;
1967 /* assert(pxdlock->type & tlckTRUNCATE); */
1968 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1969 lrd->log.updatemap.type = cpu_to_le16(LOG_FREEPXD);
1970 lrd->log.updatemap.nxd = cpu_to_le16(1);
1971 lrd->log.updatemap.pxd = pxdlock->pxd;
1972 pxd = pxdlock->pxd; /* save to format maplock */
1973 lrd->backchain =
1974 cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1975 }
1976
1977 /*
1978 * free entries XAD[next:hwm]:
1979 */
1980 if (hwm >= next) {
1981 /* init LOG_UPDATEMAP of the freed extents
1982 * XAD[next:hwm] from the deleted page itself
1983 * for logredo() to update bmap;
1984 */
1985 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1986 lrd->log.updatemap.type =
1987 cpu_to_le16(LOG_FREEXADLIST);
1988 xtlck = (struct xtlock *) & tlck->lock;
1989 hwm = xtlck->hwm.offset;
1990 lrd->log.updatemap.nxd =
1991 cpu_to_le16(hwm - next + 1);
1992 /* reformat linelock for lmLog() */
1993 xtlck->header.offset = next;
1994 xtlck->header.length = hwm - next + 1;
1995 xtlck->index = 1;
1996 lrd->backchain =
1997 cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1998 }
1999
2000 /*
2001 * format maplock(s) for txUpdateMap() to update bmap
2002 */
2003 maplock->index = 0;
2004
2005 /*
2006 * allocate entries XAD[lwm:next):
2007 */
2008 if (lwm < next) {
2009 /* format a maplock for txUpdateMap() to update bPMAP
2010 * for alloc of new/extended extents of XAD[lwm:next)
2011 * from the page itself;
2012 * txUpdateMap() resets (XAD_NEW|XAD_EXTEND) flag.
2013 */
2014 tlck->flag |= tlckUPDATEMAP;
2015 xadlock->flag = mlckALLOCXADLIST;
2016 xadlock->count = next - lwm;
2017 xadlock->xdlist = &p->xad[lwm];
2018
2019 jfs_info("xtLog: alloc ip:0x%p mp:0x%p count:%d lwm:%d next:%d",
2020 tlck->ip, mp, xadlock->count, lwm, next);
2021 maplock->index++;
2022 xadlock++;
2023 }
2024
2025 /*
2026 * truncate entry XAD[twm == next - 1]:
2027 */
2028 if (twm == next - 1) {
2029 /* format a maplock for txUpdateMap() to update bmap
2030 * to free truncated delta extent of the truncated
2031 * entry XAD[next - 1];
2032 * (xtlck->pxdlock = truncated delta extent);
2033 */
2034 tlck->flag |= tlckUPDATEMAP;
2035 pxdlock = (struct pxd_lock *) xadlock;
2036 pxdlock->flag = mlckFREEPXD;
2037 pxdlock->count = 1;
2038 pxdlock->pxd = pxd;
2039
2040 jfs_info("xtLog: truncate ip:0x%p mp:0x%p count:%d hwm:%d",
2041 ip, mp, pxdlock->count, hwm);
2042 maplock->index++;
2043 xadlock++;
2044 }
2045
2046 /*
2047 * free entries XAD[next:hwm]:
2048 */
2049 if (hwm >= next) {
2050 /* format a maplock for txUpdateMap() to update bmap
2051 * to free extents of XAD[next:hwm] from thedeleted
2052 * page itself;
2053 */
2054 tlck->flag |= tlckUPDATEMAP;
2055 xadlock->flag = mlckFREEXADLIST;
2056 xadlock->count = hwm - next + 1;
2057 xadlock->xdlist = &p->xad[next];
2058
2059 jfs_info("xtLog: free ip:0x%p mp:0x%p count:%d next:%d hwm:%d",
2060 tlck->ip, mp, xadlock->count, next, hwm);
2061 maplock->index++;
2062 }
2063
2064 /* mark page as homeward bound */
2065 tlck->flag |= tlckWRITEPAGE;
2066 }
2067 return;
2068 }
2069
2070 /*
2071 * mapLog()
2072 *
2073 * function: log from maplock of freed data extents;
2074 */
mapLog(struct jfs_log * log,struct tblock * tblk,struct lrd * lrd,struct tlock * tlck)2075 static void mapLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
2076 struct tlock * tlck)
2077 {
2078 struct pxd_lock *pxdlock;
2079 int i, nlock;
2080 pxd_t *pxd;
2081
2082 /*
2083 * page relocation: free the source page extent
2084 *
2085 * a maplock for txUpdateMap() for free of the page
2086 * has been formatted at txLock() time saving the src
2087 * relocated page address;
2088 */
2089 if (tlck->type & tlckRELOCATE) {
2090 /* log LOG_NOREDOPAGE of the old relocated page
2091 * for logredo() to start NoRedoPage filter;
2092 */
2093 lrd->type = cpu_to_le16(LOG_NOREDOPAGE);
2094 pxdlock = (struct pxd_lock *) & tlck->lock;
2095 pxd = &lrd->log.redopage.pxd;
2096 *pxd = pxdlock->pxd;
2097 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
2098
2099 /* (N.B. currently, logredo() does NOT update bmap
2100 * for free of the page itself for (LOG_XTREE|LOG_NOREDOPAGE);
2101 * if page free from relocation, LOG_UPDATEMAP log is
2102 * specifically generated now for logredo()
2103 * to update bmap for free of src relocated page;
2104 * (new flag LOG_RELOCATE may be introduced which will
2105 * inform logredo() to start NORedoPage filter and also
2106 * update block allocation map at the same time, thus
2107 * avoiding an extra log write);
2108 */
2109 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
2110 lrd->log.updatemap.type = cpu_to_le16(LOG_FREEPXD);
2111 lrd->log.updatemap.nxd = cpu_to_le16(1);
2112 lrd->log.updatemap.pxd = pxdlock->pxd;
2113 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
2114
2115 /* a maplock for txUpdateMap() for free of the page
2116 * has been formatted at txLock() time;
2117 */
2118 tlck->flag |= tlckUPDATEMAP;
2119 return;
2120 }
2121 /*
2122
2123 * Otherwise it's not a relocate request
2124 *
2125 */
2126 else {
2127 /* log LOG_UPDATEMAP for logredo() to update bmap for
2128 * free of truncated/relocated delta extent of the data;
2129 * e.g.: external EA extent, relocated/truncated extent
2130 * from xtTailgate();
2131 */
2132 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
2133 pxdlock = (struct pxd_lock *) & tlck->lock;
2134 nlock = pxdlock->index;
2135 for (i = 0; i < nlock; i++, pxdlock++) {
2136 if (pxdlock->flag & mlckALLOCPXD)
2137 lrd->log.updatemap.type =
2138 cpu_to_le16(LOG_ALLOCPXD);
2139 else
2140 lrd->log.updatemap.type =
2141 cpu_to_le16(LOG_FREEPXD);
2142 lrd->log.updatemap.nxd = cpu_to_le16(1);
2143 lrd->log.updatemap.pxd = pxdlock->pxd;
2144 lrd->backchain =
2145 cpu_to_le32(lmLog(log, tblk, lrd, NULL));
2146 jfs_info("mapLog: xaddr:0x%lx xlen:0x%x",
2147 (ulong) addressPXD(&pxdlock->pxd),
2148 lengthPXD(&pxdlock->pxd));
2149 }
2150
2151 /* update bmap */
2152 tlck->flag |= tlckUPDATEMAP;
2153 }
2154 }
2155
2156 /*
2157 * txEA()
2158 *
2159 * function: acquire maplock for EA/ACL extents or
2160 * set COMMIT_INLINE flag;
2161 */
txEA(tid_t tid,struct inode * ip,dxd_t * oldea,dxd_t * newea)2162 void txEA(tid_t tid, struct inode *ip, dxd_t * oldea, dxd_t * newea)
2163 {
2164 struct tlock *tlck = NULL;
2165 struct pxd_lock *maplock = NULL, *pxdlock = NULL;
2166
2167 /*
2168 * format maplock for alloc of new EA extent
2169 */
2170 if (newea) {
2171 /* Since the newea could be a completely zeroed entry we need to
2172 * check for the two flags which indicate we should actually
2173 * commit new EA data
2174 */
2175 if (newea->flag & DXD_EXTENT) {
2176 tlck = txMaplock(tid, ip, tlckMAP);
2177 maplock = (struct pxd_lock *) & tlck->lock;
2178 pxdlock = (struct pxd_lock *) maplock;
2179 pxdlock->flag = mlckALLOCPXD;
2180 PXDaddress(&pxdlock->pxd, addressDXD(newea));
2181 PXDlength(&pxdlock->pxd, lengthDXD(newea));
2182 pxdlock++;
2183 maplock->index = 1;
2184 } else if (newea->flag & DXD_INLINE) {
2185 tlck = NULL;
2186
2187 set_cflag(COMMIT_Inlineea, ip);
2188 }
2189 }
2190
2191 /*
2192 * format maplock for free of old EA extent
2193 */
2194 if (!test_cflag(COMMIT_Nolink, ip) && oldea->flag & DXD_EXTENT) {
2195 if (tlck == NULL) {
2196 tlck = txMaplock(tid, ip, tlckMAP);
2197 maplock = (struct pxd_lock *) & tlck->lock;
2198 pxdlock = (struct pxd_lock *) maplock;
2199 maplock->index = 0;
2200 }
2201 pxdlock->flag = mlckFREEPXD;
2202 PXDaddress(&pxdlock->pxd, addressDXD(oldea));
2203 PXDlength(&pxdlock->pxd, lengthDXD(oldea));
2204 maplock->index++;
2205 }
2206 }
2207
2208 /*
2209 * txForce()
2210 *
2211 * function: synchronously write pages locked by transaction
2212 * after txLog() but before txUpdateMap();
2213 */
txForce(struct tblock * tblk)2214 static void txForce(struct tblock * tblk)
2215 {
2216 struct tlock *tlck;
2217 lid_t lid, next;
2218 struct metapage *mp;
2219
2220 /*
2221 * reverse the order of transaction tlocks in
2222 * careful update order of address index pages
2223 * (right to left, bottom up)
2224 */
2225 tlck = lid_to_tlock(tblk->next);
2226 lid = tlck->next;
2227 tlck->next = 0;
2228 while (lid) {
2229 tlck = lid_to_tlock(lid);
2230 next = tlck->next;
2231 tlck->next = tblk->next;
2232 tblk->next = lid;
2233 lid = next;
2234 }
2235
2236 /*
2237 * synchronously write the page, and
2238 * hold the page for txUpdateMap();
2239 */
2240 for (lid = tblk->next; lid; lid = next) {
2241 tlck = lid_to_tlock(lid);
2242 next = tlck->next;
2243
2244 if ((mp = tlck->mp) != NULL &&
2245 (tlck->type & tlckBTROOT) == 0) {
2246 assert(mp->xflag & COMMIT_PAGE);
2247
2248 if (tlck->flag & tlckWRITEPAGE) {
2249 tlck->flag &= ~tlckWRITEPAGE;
2250
2251 /* do not release page to freelist */
2252 force_metapage(mp);
2253 #if 0
2254 /*
2255 * The "right" thing to do here is to
2256 * synchronously write the metadata.
2257 * With the current implementation this
2258 * is hard since write_metapage requires
2259 * us to kunmap & remap the page. If we
2260 * have tlocks pointing into the metadata
2261 * pages, we don't want to do this. I think
2262 * we can get by with synchronously writing
2263 * the pages when they are released.
2264 */
2265 assert(mp->nohomeok);
2266 set_bit(META_dirty, &mp->flag);
2267 set_bit(META_sync, &mp->flag);
2268 #endif
2269 }
2270 }
2271 }
2272 }
2273
2274 /*
2275 * txUpdateMap()
2276 *
2277 * function: update persistent allocation map (and working map
2278 * if appropriate);
2279 *
2280 * parameter:
2281 */
txUpdateMap(struct tblock * tblk)2282 static void txUpdateMap(struct tblock * tblk)
2283 {
2284 struct inode *ip;
2285 struct inode *ipimap;
2286 lid_t lid;
2287 struct tlock *tlck;
2288 struct maplock *maplock;
2289 struct pxd_lock pxdlock;
2290 int maptype;
2291 int k, nlock;
2292 struct metapage *mp = NULL;
2293
2294 ipimap = JFS_SBI(tblk->sb)->ipimap;
2295
2296 maptype = (tblk->xflag & COMMIT_PMAP) ? COMMIT_PMAP : COMMIT_PWMAP;
2297
2298
2299 /*
2300 * update block allocation map
2301 *
2302 * update allocation state in pmap (and wmap) and
2303 * update lsn of the pmap page;
2304 */
2305 /*
2306 * scan each tlock/page of transaction for block allocation/free:
2307 *
2308 * for each tlock/page of transaction, update map.
2309 * ? are there tlock for pmap and pwmap at the same time ?
2310 */
2311 for (lid = tblk->next; lid; lid = tlck->next) {
2312 tlck = lid_to_tlock(lid);
2313
2314 if ((tlck->flag & tlckUPDATEMAP) == 0)
2315 continue;
2316
2317 if (tlck->flag & tlckFREEPAGE) {
2318 /*
2319 * Another thread may attempt to reuse freed space
2320 * immediately, so we want to get rid of the metapage
2321 * before anyone else has a chance to get it.
2322 * Lock metapage, update maps, then invalidate
2323 * the metapage.
2324 */
2325 mp = tlck->mp;
2326 ASSERT(mp->xflag & COMMIT_PAGE);
2327 grab_metapage(mp);
2328 }
2329
2330 /*
2331 * extent list:
2332 * . in-line PXD list:
2333 * . out-of-line XAD list:
2334 */
2335 maplock = (struct maplock *) & tlck->lock;
2336 nlock = maplock->index;
2337
2338 for (k = 0; k < nlock; k++, maplock++) {
2339 /*
2340 * allocate blocks in persistent map:
2341 *
2342 * blocks have been allocated from wmap at alloc time;
2343 */
2344 if (maplock->flag & mlckALLOC) {
2345 txAllocPMap(ipimap, maplock, tblk);
2346 }
2347 /*
2348 * free blocks in persistent and working map:
2349 * blocks will be freed in pmap and then in wmap;
2350 *
2351 * ? tblock specifies the PMAP/PWMAP based upon
2352 * transaction
2353 *
2354 * free blocks in persistent map:
2355 * blocks will be freed from wmap at last reference
2356 * release of the object for regular files;
2357 *
2358 * Alway free blocks from both persistent & working
2359 * maps for directories
2360 */
2361 else { /* (maplock->flag & mlckFREE) */
2362
2363 if (tlck->flag & tlckDIRECTORY)
2364 txFreeMap(ipimap, maplock,
2365 tblk, COMMIT_PWMAP);
2366 else
2367 txFreeMap(ipimap, maplock,
2368 tblk, maptype);
2369 }
2370 }
2371 if (tlck->flag & tlckFREEPAGE) {
2372 if (!(tblk->flag & tblkGC_LAZY)) {
2373 /* This is equivalent to txRelease */
2374 ASSERT(mp->lid == lid);
2375 tlck->mp->lid = 0;
2376 }
2377 assert(mp->nohomeok == 1);
2378 metapage_homeok(mp);
2379 discard_metapage(mp);
2380 tlck->mp = NULL;
2381 }
2382 }
2383 /*
2384 * update inode allocation map
2385 *
2386 * update allocation state in pmap and
2387 * update lsn of the pmap page;
2388 * update in-memory inode flag/state
2389 *
2390 * unlock mapper/write lock
2391 */
2392 if (tblk->xflag & COMMIT_CREATE) {
2393 diUpdatePMap(ipimap, tblk->ino, false, tblk);
2394 /* update persistent block allocation map
2395 * for the allocation of inode extent;
2396 */
2397 pxdlock.flag = mlckALLOCPXD;
2398 pxdlock.pxd = tblk->u.ixpxd;
2399 pxdlock.index = 1;
2400 txAllocPMap(ipimap, (struct maplock *) & pxdlock, tblk);
2401 } else if (tblk->xflag & COMMIT_DELETE) {
2402 ip = tblk->u.ip;
2403 diUpdatePMap(ipimap, ip->i_ino, true, tblk);
2404 iput(ip);
2405 }
2406 }
2407
2408 /*
2409 * txAllocPMap()
2410 *
2411 * function: allocate from persistent map;
2412 *
2413 * parameter:
2414 * ipbmap -
2415 * malock -
2416 * xad list:
2417 * pxd:
2418 *
2419 * maptype -
2420 * allocate from persistent map;
2421 * free from persistent map;
2422 * (e.g., tmp file - free from working map at releae
2423 * of last reference);
2424 * free from persistent and working map;
2425 *
2426 * lsn - log sequence number;
2427 */
txAllocPMap(struct inode * ip,struct maplock * maplock,struct tblock * tblk)2428 static void txAllocPMap(struct inode *ip, struct maplock * maplock,
2429 struct tblock * tblk)
2430 {
2431 struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
2432 struct xdlistlock *xadlistlock;
2433 xad_t *xad;
2434 s64 xaddr;
2435 int xlen;
2436 struct pxd_lock *pxdlock;
2437 struct xdlistlock *pxdlistlock;
2438 pxd_t *pxd;
2439 int n;
2440
2441 /*
2442 * allocate from persistent map;
2443 */
2444 if (maplock->flag & mlckALLOCXADLIST) {
2445 xadlistlock = (struct xdlistlock *) maplock;
2446 xad = xadlistlock->xdlist;
2447 for (n = 0; n < xadlistlock->count; n++, xad++) {
2448 if (xad->flag & (XAD_NEW | XAD_EXTENDED)) {
2449 xaddr = addressXAD(xad);
2450 xlen = lengthXAD(xad);
2451 dbUpdatePMap(ipbmap, false, xaddr,
2452 (s64) xlen, tblk);
2453 xad->flag &= ~(XAD_NEW | XAD_EXTENDED);
2454 jfs_info("allocPMap: xaddr:0x%lx xlen:%d",
2455 (ulong) xaddr, xlen);
2456 }
2457 }
2458 } else if (maplock->flag & mlckALLOCPXD) {
2459 pxdlock = (struct pxd_lock *) maplock;
2460 xaddr = addressPXD(&pxdlock->pxd);
2461 xlen = lengthPXD(&pxdlock->pxd);
2462 dbUpdatePMap(ipbmap, false, xaddr, (s64) xlen, tblk);
2463 jfs_info("allocPMap: xaddr:0x%lx xlen:%d", (ulong) xaddr, xlen);
2464 } else { /* (maplock->flag & mlckALLOCPXDLIST) */
2465
2466 pxdlistlock = (struct xdlistlock *) maplock;
2467 pxd = pxdlistlock->xdlist;
2468 for (n = 0; n < pxdlistlock->count; n++, pxd++) {
2469 xaddr = addressPXD(pxd);
2470 xlen = lengthPXD(pxd);
2471 dbUpdatePMap(ipbmap, false, xaddr, (s64) xlen,
2472 tblk);
2473 jfs_info("allocPMap: xaddr:0x%lx xlen:%d",
2474 (ulong) xaddr, xlen);
2475 }
2476 }
2477 }
2478
2479 /*
2480 * txFreeMap()
2481 *
2482 * function: free from persistent and/or working map;
2483 *
2484 * todo: optimization
2485 */
txFreeMap(struct inode * ip,struct maplock * maplock,struct tblock * tblk,int maptype)2486 void txFreeMap(struct inode *ip,
2487 struct maplock * maplock, struct tblock * tblk, int maptype)
2488 {
2489 struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
2490 struct xdlistlock *xadlistlock;
2491 xad_t *xad;
2492 s64 xaddr;
2493 int xlen;
2494 struct pxd_lock *pxdlock;
2495 struct xdlistlock *pxdlistlock;
2496 pxd_t *pxd;
2497 int n;
2498
2499 jfs_info("txFreeMap: tblk:0x%p maplock:0x%p maptype:0x%x",
2500 tblk, maplock, maptype);
2501
2502 /*
2503 * free from persistent map;
2504 */
2505 if (maptype == COMMIT_PMAP || maptype == COMMIT_PWMAP) {
2506 if (maplock->flag & mlckFREEXADLIST) {
2507 xadlistlock = (struct xdlistlock *) maplock;
2508 xad = xadlistlock->xdlist;
2509 for (n = 0; n < xadlistlock->count; n++, xad++) {
2510 if (!(xad->flag & XAD_NEW)) {
2511 xaddr = addressXAD(xad);
2512 xlen = lengthXAD(xad);
2513 dbUpdatePMap(ipbmap, true, xaddr,
2514 (s64) xlen, tblk);
2515 jfs_info("freePMap: xaddr:0x%lx xlen:%d",
2516 (ulong) xaddr, xlen);
2517 }
2518 }
2519 } else if (maplock->flag & mlckFREEPXD) {
2520 pxdlock = (struct pxd_lock *) maplock;
2521 xaddr = addressPXD(&pxdlock->pxd);
2522 xlen = lengthPXD(&pxdlock->pxd);
2523 dbUpdatePMap(ipbmap, true, xaddr, (s64) xlen,
2524 tblk);
2525 jfs_info("freePMap: xaddr:0x%lx xlen:%d",
2526 (ulong) xaddr, xlen);
2527 } else { /* (maplock->flag & mlckALLOCPXDLIST) */
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 dbUpdatePMap(ipbmap, true, xaddr,
2535 (s64) xlen, tblk);
2536 jfs_info("freePMap: xaddr:0x%lx xlen:%d",
2537 (ulong) xaddr, xlen);
2538 }
2539 }
2540 }
2541
2542 /*
2543 * free from working map;
2544 */
2545 if (maptype == COMMIT_PWMAP || maptype == COMMIT_WMAP) {
2546 if (maplock->flag & mlckFREEXADLIST) {
2547 xadlistlock = (struct xdlistlock *) maplock;
2548 xad = xadlistlock->xdlist;
2549 for (n = 0; n < xadlistlock->count; n++, xad++) {
2550 xaddr = addressXAD(xad);
2551 xlen = lengthXAD(xad);
2552 dbFree(ip, xaddr, (s64) xlen);
2553 xad->flag = 0;
2554 jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
2555 (ulong) xaddr, xlen);
2556 }
2557 } else if (maplock->flag & mlckFREEPXD) {
2558 pxdlock = (struct pxd_lock *) maplock;
2559 xaddr = addressPXD(&pxdlock->pxd);
2560 xlen = lengthPXD(&pxdlock->pxd);
2561 dbFree(ip, xaddr, (s64) xlen);
2562 jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
2563 (ulong) xaddr, xlen);
2564 } else { /* (maplock->flag & mlckFREEPXDLIST) */
2565
2566 pxdlistlock = (struct xdlistlock *) maplock;
2567 pxd = pxdlistlock->xdlist;
2568 for (n = 0; n < pxdlistlock->count; n++, pxd++) {
2569 xaddr = addressPXD(pxd);
2570 xlen = lengthPXD(pxd);
2571 dbFree(ip, xaddr, (s64) xlen);
2572 jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
2573 (ulong) xaddr, xlen);
2574 }
2575 }
2576 }
2577 }
2578
2579 /*
2580 * txFreelock()
2581 *
2582 * function: remove tlock from inode anonymous locklist
2583 */
txFreelock(struct inode * ip)2584 void txFreelock(struct inode *ip)
2585 {
2586 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
2587 struct tlock *xtlck, *tlck;
2588 lid_t xlid = 0, lid;
2589
2590 if (!jfs_ip->atlhead)
2591 return;
2592
2593 TXN_LOCK();
2594 xtlck = (struct tlock *) &jfs_ip->atlhead;
2595
2596 while ((lid = xtlck->next) != 0) {
2597 tlck = lid_to_tlock(lid);
2598 if (tlck->flag & tlckFREELOCK) {
2599 xtlck->next = tlck->next;
2600 txLockFree(lid);
2601 } else {
2602 xtlck = tlck;
2603 xlid = lid;
2604 }
2605 }
2606
2607 if (jfs_ip->atlhead)
2608 jfs_ip->atltail = xlid;
2609 else {
2610 jfs_ip->atltail = 0;
2611 /*
2612 * If inode was on anon_list, remove it
2613 */
2614 list_del_init(&jfs_ip->anon_inode_list);
2615 }
2616 TXN_UNLOCK();
2617 }
2618
2619 /*
2620 * txAbort()
2621 *
2622 * function: abort tx before commit;
2623 *
2624 * frees line-locks and segment locks for all
2625 * segments in comdata structure.
2626 * Optionally sets state of file-system to FM_DIRTY in super-block.
2627 * log age of page-frames in memory for which caller has
2628 * are reset to 0 (to avoid logwarap).
2629 */
txAbort(tid_t tid,int dirty)2630 void txAbort(tid_t tid, int dirty)
2631 {
2632 lid_t lid, next;
2633 struct metapage *mp;
2634 struct tblock *tblk = tid_to_tblock(tid);
2635 struct tlock *tlck;
2636
2637 /*
2638 * free tlocks of the transaction
2639 */
2640 for (lid = tblk->next; lid; lid = next) {
2641 tlck = lid_to_tlock(lid);
2642 next = tlck->next;
2643 mp = tlck->mp;
2644 JFS_IP(tlck->ip)->xtlid = 0;
2645
2646 if (mp) {
2647 mp->lid = 0;
2648
2649 /*
2650 * reset lsn of page to avoid logwarap:
2651 *
2652 * (page may have been previously committed by another
2653 * transaction(s) but has not been paged, i.e.,
2654 * it may be on logsync list even though it has not
2655 * been logged for the current tx.)
2656 */
2657 if (mp->xflag & COMMIT_PAGE && mp->lsn)
2658 LogSyncRelease(mp);
2659 }
2660 /* insert tlock at head of freelist */
2661 TXN_LOCK();
2662 txLockFree(lid);
2663 TXN_UNLOCK();
2664 }
2665
2666 /* caller will free the transaction block */
2667
2668 tblk->next = tblk->last = 0;
2669
2670 /*
2671 * mark filesystem dirty
2672 */
2673 if (dirty)
2674 jfs_error(tblk->sb, "\n");
2675
2676 return;
2677 }
2678
2679 /*
2680 * txLazyCommit(void)
2681 *
2682 * All transactions except those changing ipimap (COMMIT_FORCE) are
2683 * processed by this routine. This insures that the inode and block
2684 * allocation maps are updated in order. For synchronous transactions,
2685 * let the user thread finish processing after txUpdateMap() is called.
2686 */
txLazyCommit(struct tblock * tblk)2687 static void txLazyCommit(struct tblock * tblk)
2688 {
2689 struct jfs_log *log;
2690
2691 while (((tblk->flag & tblkGC_READY) == 0) &&
2692 ((tblk->flag & tblkGC_UNLOCKED) == 0)) {
2693 /* We must have gotten ahead of the user thread
2694 */
2695 jfs_info("jfs_lazycommit: tblk 0x%p not unlocked", tblk);
2696 yield();
2697 }
2698
2699 jfs_info("txLazyCommit: processing tblk 0x%p", tblk);
2700
2701 txUpdateMap(tblk);
2702
2703 log = (struct jfs_log *) JFS_SBI(tblk->sb)->log;
2704
2705 spin_lock_irq(&log->gclock); // LOGGC_LOCK
2706
2707 tblk->flag |= tblkGC_COMMITTED;
2708
2709 if (tblk->flag & tblkGC_READY)
2710 log->gcrtc--;
2711
2712 wake_up_all(&tblk->gcwait); // LOGGC_WAKEUP
2713
2714 /*
2715 * Can't release log->gclock until we've tested tblk->flag
2716 */
2717 if (tblk->flag & tblkGC_LAZY) {
2718 spin_unlock_irq(&log->gclock); // LOGGC_UNLOCK
2719 txUnlock(tblk);
2720 tblk->flag &= ~tblkGC_LAZY;
2721 txEnd(tblk - TxBlock); /* Convert back to tid */
2722 } else
2723 spin_unlock_irq(&log->gclock); // LOGGC_UNLOCK
2724
2725 jfs_info("txLazyCommit: done: tblk = 0x%p", tblk);
2726 }
2727
2728 /*
2729 * jfs_lazycommit(void)
2730 *
2731 * To be run as a kernel daemon. If lbmIODone is called in an interrupt
2732 * context, or where blocking is not wanted, this routine will process
2733 * committed transactions from the unlock queue.
2734 */
jfs_lazycommit(void * arg)2735 int jfs_lazycommit(void *arg)
2736 {
2737 int WorkDone;
2738 struct tblock *tblk;
2739 unsigned long flags;
2740 struct jfs_sb_info *sbi;
2741
2742 do {
2743 LAZY_LOCK(flags);
2744 jfs_commit_thread_waking = 0; /* OK to wake another thread */
2745 while (!list_empty(&TxAnchor.unlock_queue)) {
2746 WorkDone = 0;
2747 list_for_each_entry(tblk, &TxAnchor.unlock_queue,
2748 cqueue) {
2749
2750 sbi = JFS_SBI(tblk->sb);
2751 /*
2752 * For each volume, the transactions must be
2753 * handled in order. If another commit thread
2754 * is handling a tblk for this superblock,
2755 * skip it
2756 */
2757 if (sbi->commit_state & IN_LAZYCOMMIT)
2758 continue;
2759
2760 sbi->commit_state |= IN_LAZYCOMMIT;
2761 WorkDone = 1;
2762
2763 /*
2764 * Remove transaction from queue
2765 */
2766 list_del(&tblk->cqueue);
2767
2768 LAZY_UNLOCK(flags);
2769 txLazyCommit(tblk);
2770 LAZY_LOCK(flags);
2771
2772 sbi->commit_state &= ~IN_LAZYCOMMIT;
2773 /*
2774 * Don't continue in the for loop. (We can't
2775 * anyway, it's unsafe!) We want to go back to
2776 * the beginning of the list.
2777 */
2778 break;
2779 }
2780
2781 /* If there was nothing to do, don't continue */
2782 if (!WorkDone)
2783 break;
2784 }
2785 /* In case a wakeup came while all threads were active */
2786 jfs_commit_thread_waking = 0;
2787
2788 if (freezing(current)) {
2789 LAZY_UNLOCK(flags);
2790 try_to_freeze();
2791 } else {
2792 DECLARE_WAITQUEUE(wq, current);
2793
2794 add_wait_queue(&jfs_commit_thread_wait, &wq);
2795 set_current_state(TASK_INTERRUPTIBLE);
2796 LAZY_UNLOCK(flags);
2797 schedule();
2798 remove_wait_queue(&jfs_commit_thread_wait, &wq);
2799 }
2800 } while (!kthread_should_stop());
2801
2802 if (!list_empty(&TxAnchor.unlock_queue))
2803 jfs_err("jfs_lazycommit being killed w/pending transactions!");
2804 else
2805 jfs_info("jfs_lazycommit being killed");
2806 return 0;
2807 }
2808
txLazyUnlock(struct tblock * tblk)2809 void txLazyUnlock(struct tblock * tblk)
2810 {
2811 unsigned long flags;
2812
2813 LAZY_LOCK(flags);
2814
2815 list_add_tail(&tblk->cqueue, &TxAnchor.unlock_queue);
2816 /*
2817 * Don't wake up a commit thread if there is already one servicing
2818 * this superblock, or if the last one we woke up hasn't started yet.
2819 */
2820 if (!(JFS_SBI(tblk->sb)->commit_state & IN_LAZYCOMMIT) &&
2821 !jfs_commit_thread_waking) {
2822 jfs_commit_thread_waking = 1;
2823 wake_up(&jfs_commit_thread_wait);
2824 }
2825 LAZY_UNLOCK(flags);
2826 }
2827
LogSyncRelease(struct metapage * mp)2828 static void LogSyncRelease(struct metapage * mp)
2829 {
2830 struct jfs_log *log = mp->log;
2831
2832 assert(mp->nohomeok);
2833 assert(log);
2834 metapage_homeok(mp);
2835 }
2836
2837 /*
2838 * txQuiesce
2839 *
2840 * Block all new transactions and push anonymous transactions to
2841 * completion
2842 *
2843 * This does almost the same thing as jfs_sync below. We don't
2844 * worry about deadlocking when jfs_tlocks_low is set, since we would
2845 * expect jfs_sync to get us out of that jam.
2846 */
txQuiesce(struct super_block * sb)2847 void txQuiesce(struct super_block *sb)
2848 {
2849 struct inode *ip;
2850 struct jfs_inode_info *jfs_ip;
2851 struct jfs_log *log = JFS_SBI(sb)->log;
2852 tid_t tid;
2853
2854 set_bit(log_QUIESCE, &log->flag);
2855
2856 TXN_LOCK();
2857 restart:
2858 while (!list_empty(&TxAnchor.anon_list)) {
2859 jfs_ip = list_entry(TxAnchor.anon_list.next,
2860 struct jfs_inode_info,
2861 anon_inode_list);
2862 ip = &jfs_ip->vfs_inode;
2863
2864 /*
2865 * inode will be removed from anonymous list
2866 * when it is committed
2867 */
2868 TXN_UNLOCK();
2869 tid = txBegin(ip->i_sb, COMMIT_INODE | COMMIT_FORCE);
2870 mutex_lock(&jfs_ip->commit_mutex);
2871 txCommit(tid, 1, &ip, 0);
2872 txEnd(tid);
2873 mutex_unlock(&jfs_ip->commit_mutex);
2874 /*
2875 * Just to be safe. I don't know how
2876 * long we can run without blocking
2877 */
2878 cond_resched();
2879 TXN_LOCK();
2880 }
2881
2882 /*
2883 * If jfs_sync is running in parallel, there could be some inodes
2884 * on anon_list2. Let's check.
2885 */
2886 if (!list_empty(&TxAnchor.anon_list2)) {
2887 list_splice_init(&TxAnchor.anon_list2, &TxAnchor.anon_list);
2888 goto restart;
2889 }
2890 TXN_UNLOCK();
2891
2892 /*
2893 * We may need to kick off the group commit
2894 */
2895 jfs_flush_journal(log, 0);
2896 }
2897
2898 /*
2899 * txResume()
2900 *
2901 * Allows transactions to start again following txQuiesce
2902 */
txResume(struct super_block * sb)2903 void txResume(struct super_block *sb)
2904 {
2905 struct jfs_log *log = JFS_SBI(sb)->log;
2906
2907 clear_bit(log_QUIESCE, &log->flag);
2908 TXN_WAKEUP(&log->syncwait);
2909 }
2910
2911 /*
2912 * jfs_sync(void)
2913 *
2914 * To be run as a kernel daemon. This is awakened when tlocks run low.
2915 * We write any inodes that have anonymous tlocks so they will become
2916 * available.
2917 */
jfs_sync(void * arg)2918 int jfs_sync(void *arg)
2919 {
2920 struct inode *ip;
2921 struct jfs_inode_info *jfs_ip;
2922 tid_t tid;
2923
2924 do {
2925 /*
2926 * write each inode on the anonymous inode list
2927 */
2928 TXN_LOCK();
2929 while (jfs_tlocks_low && !list_empty(&TxAnchor.anon_list)) {
2930 jfs_ip = list_entry(TxAnchor.anon_list.next,
2931 struct jfs_inode_info,
2932 anon_inode_list);
2933 ip = &jfs_ip->vfs_inode;
2934
2935 if (! igrab(ip)) {
2936 /*
2937 * Inode is being freed
2938 */
2939 list_del_init(&jfs_ip->anon_inode_list);
2940 } else if (mutex_trylock(&jfs_ip->commit_mutex)) {
2941 /*
2942 * inode will be removed from anonymous list
2943 * when it is committed
2944 */
2945 TXN_UNLOCK();
2946 tid = txBegin(ip->i_sb, COMMIT_INODE);
2947 txCommit(tid, 1, &ip, 0);
2948 txEnd(tid);
2949 mutex_unlock(&jfs_ip->commit_mutex);
2950
2951 iput(ip);
2952 /*
2953 * Just to be safe. I don't know how
2954 * long we can run without blocking
2955 */
2956 cond_resched();
2957 TXN_LOCK();
2958 } else {
2959 /* We can't get the commit mutex. It may
2960 * be held by a thread waiting for tlock's
2961 * so let's not block here. Save it to
2962 * put back on the anon_list.
2963 */
2964
2965 /* Move from anon_list to anon_list2 */
2966 list_move(&jfs_ip->anon_inode_list,
2967 &TxAnchor.anon_list2);
2968
2969 TXN_UNLOCK();
2970 iput(ip);
2971 TXN_LOCK();
2972 }
2973 }
2974 /* Add anon_list2 back to anon_list */
2975 list_splice_init(&TxAnchor.anon_list2, &TxAnchor.anon_list);
2976
2977 if (freezing(current)) {
2978 TXN_UNLOCK();
2979 try_to_freeze();
2980 } else {
2981 set_current_state(TASK_INTERRUPTIBLE);
2982 TXN_UNLOCK();
2983 schedule();
2984 }
2985 } while (!kthread_should_stop());
2986
2987 jfs_info("jfs_sync being killed");
2988 return 0;
2989 }
2990
2991 #if defined(CONFIG_PROC_FS) && defined(CONFIG_JFS_DEBUG)
jfs_txanchor_proc_show(struct seq_file * m,void * v)2992 int jfs_txanchor_proc_show(struct seq_file *m, void *v)
2993 {
2994 char *freewait;
2995 char *freelockwait;
2996 char *lowlockwait;
2997
2998 freewait =
2999 waitqueue_active(&TxAnchor.freewait) ? "active" : "empty";
3000 freelockwait =
3001 waitqueue_active(&TxAnchor.freelockwait) ? "active" : "empty";
3002 lowlockwait =
3003 waitqueue_active(&TxAnchor.lowlockwait) ? "active" : "empty";
3004
3005 seq_printf(m,
3006 "JFS TxAnchor\n"
3007 "============\n"
3008 "freetid = %d\n"
3009 "freewait = %s\n"
3010 "freelock = %d\n"
3011 "freelockwait = %s\n"
3012 "lowlockwait = %s\n"
3013 "tlocksInUse = %d\n"
3014 "jfs_tlocks_low = %d\n"
3015 "unlock_queue is %sempty\n",
3016 TxAnchor.freetid,
3017 freewait,
3018 TxAnchor.freelock,
3019 freelockwait,
3020 lowlockwait,
3021 TxAnchor.tlocksInUse,
3022 jfs_tlocks_low,
3023 list_empty(&TxAnchor.unlock_queue) ? "" : "not ");
3024 return 0;
3025 }
3026 #endif
3027
3028 #if defined(CONFIG_PROC_FS) && defined(CONFIG_JFS_STATISTICS)
jfs_txstats_proc_show(struct seq_file * m,void * v)3029 int jfs_txstats_proc_show(struct seq_file *m, void *v)
3030 {
3031 seq_printf(m,
3032 "JFS TxStats\n"
3033 "===========\n"
3034 "calls to txBegin = %d\n"
3035 "txBegin blocked by sync barrier = %d\n"
3036 "txBegin blocked by tlocks low = %d\n"
3037 "txBegin blocked by no free tid = %d\n"
3038 "calls to txBeginAnon = %d\n"
3039 "txBeginAnon blocked by sync barrier = %d\n"
3040 "txBeginAnon blocked by tlocks low = %d\n"
3041 "calls to txLockAlloc = %d\n"
3042 "tLockAlloc blocked by no free lock = %d\n",
3043 TxStat.txBegin,
3044 TxStat.txBegin_barrier,
3045 TxStat.txBegin_lockslow,
3046 TxStat.txBegin_freetid,
3047 TxStat.txBeginAnon,
3048 TxStat.txBeginAnon_barrier,
3049 TxStat.txBeginAnon_lockslow,
3050 TxStat.txLockAlloc,
3051 TxStat.txLockAlloc_freelock);
3052 return 0;
3053 }
3054 #endif
3055