blob: ac3d66948e8ceae440e77d944bd2667f9a9e8e76 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
Linus Torvalds1da177e2005-04-16 15:20:36 -070045#include <linux/fs.h>
46#include <linux/vmalloc.h>
47#include <linux/smp_lock.h>
48#include <linux/completion.h>
49#include <linux/suspend.h>
50#include <linux/module.h>
51#include <linux/moduleparam.h>
Christoph Hellwig91dbb4d2006-02-15 12:49:04 -060052#include <linux/kthread.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070053#include "jfs_incore.h"
Dave Kleikamp1868f4a2005-05-04 15:29:35 -050054#include "jfs_inode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070055#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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113/*
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
Christoph Hellwig91dbb4d2006-02-15 12:49:04 -0600125static DECLARE_WAIT_QUEUE_HEAD(jfs_commit_thread_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126static 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 schedule();
139 current->state = TASK_RUNNING;
140 remove_wait_queue(event, &wait);
141}
142
143#define TXN_SLEEP(event)\
144{\
145 TXN_SLEEP_DROP_LOCK(event);\
146 TXN_LOCK();\
147}
148
149#define TXN_WAKEUP(event) wake_up_all(event)
150
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151/*
152 * statistics
153 */
154static struct {
155 tid_t maxtid; /* 4: biggest tid ever used */
156 lid_t maxlid; /* 4: biggest lid ever used */
157 int ntid; /* 4: # of transactions performed */
158 int nlid; /* 4: # of tlocks acquired */
159 int waitlock; /* 4: # of tlock wait */
160} stattx;
161
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162/*
163 * forward references
164 */
165static int diLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
166 struct tlock * tlck, struct commit * cd);
167static int dataLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
168 struct tlock * tlck);
169static void dtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
170 struct tlock * tlck);
171static void mapLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
172 struct tlock * tlck);
173static void txAllocPMap(struct inode *ip, struct maplock * maplock,
174 struct tblock * tblk);
175static void txForce(struct tblock * tblk);
176static int txLog(struct jfs_log * log, struct tblock * tblk,
177 struct commit * cd);
178static void txUpdateMap(struct tblock * tblk);
179static void txRelease(struct tblock * tblk);
180static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
181 struct tlock * tlck);
182static void LogSyncRelease(struct metapage * mp);
183
184/*
185 * transaction block/lock management
186 * ---------------------------------
187 */
188
189/*
190 * Get a transaction lock from the free list. If the number in use is
191 * greater than the high water mark, wake up the sync daemon. This should
192 * free some anonymous transaction locks. (TXN_LOCK must be held.)
193 */
194static lid_t txLockAlloc(void)
195{
196 lid_t lid;
197
198 INCREMENT(TxStat.txLockAlloc);
199 if (!TxAnchor.freelock) {
200 INCREMENT(TxStat.txLockAlloc_freelock);
201 }
202
203 while (!(lid = TxAnchor.freelock))
204 TXN_SLEEP(&TxAnchor.freelockwait);
205 TxAnchor.freelock = TxLock[lid].next;
206 HIGHWATERMARK(stattx.maxlid, lid);
207 if ((++TxAnchor.tlocksInUse > TxLockHWM) && (jfs_tlocks_low == 0)) {
208 jfs_info("txLockAlloc tlocks low");
209 jfs_tlocks_low = 1;
Christoph Hellwig91dbb4d2006-02-15 12:49:04 -0600210 wake_up_process(jfsSyncThread);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 }
212
213 return lid;
214}
215
216static void txLockFree(lid_t lid)
217{
Dave Kleikamp7fab4792005-05-02 12:25:02 -0600218 TxLock[lid].tid = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 TxLock[lid].next = TxAnchor.freelock;
220 TxAnchor.freelock = lid;
221 TxAnchor.tlocksInUse--;
222 if (jfs_tlocks_low && (TxAnchor.tlocksInUse < TxLockLWM)) {
223 jfs_info("txLockFree jfs_tlocks_low no more");
224 jfs_tlocks_low = 0;
225 TXN_WAKEUP(&TxAnchor.lowlockwait);
226 }
227 TXN_WAKEUP(&TxAnchor.freelockwait);
228}
229
230/*
231 * NAME: txInit()
232 *
233 * FUNCTION: initialize transaction management structures
234 *
235 * RETURN:
236 *
237 * serialization: single thread at jfs_init()
238 */
239int txInit(void)
240{
241 int k, size;
242 struct sysinfo si;
243
244 /* Set defaults for nTxLock and nTxBlock if unset */
245
246 if (nTxLock == -1) {
247 if (nTxBlock == -1) {
248 /* Base default on memory size */
249 si_meminfo(&si);
250 if (si.totalram > (256 * 1024)) /* 1 GB */
251 nTxLock = 64 * 1024;
252 else
253 nTxLock = si.totalram >> 2;
254 } else if (nTxBlock > (8 * 1024))
255 nTxLock = 64 * 1024;
256 else
257 nTxLock = nTxBlock << 3;
258 }
259 if (nTxBlock == -1)
260 nTxBlock = nTxLock >> 3;
261
262 /* Verify tunable parameters */
263 if (nTxBlock < 16)
264 nTxBlock = 16; /* No one should set it this low */
265 if (nTxBlock > 65536)
266 nTxBlock = 65536;
267 if (nTxLock < 256)
268 nTxLock = 256; /* No one should set it this low */
269 if (nTxLock > 65536)
270 nTxLock = 65536;
271
272 printk(KERN_INFO "JFS: nTxBlock = %d, nTxLock = %d\n",
273 nTxBlock, nTxLock);
274 /*
275 * initialize transaction block (tblock) table
276 *
277 * transaction id (tid) = tblock index
278 * tid = 0 is reserved.
279 */
280 TxLockLWM = (nTxLock * 4) / 10;
281 TxLockHWM = (nTxLock * 7) / 10;
282 TxLockVHWM = (nTxLock * 8) / 10;
283
284 size = sizeof(struct tblock) * nTxBlock;
285 TxBlock = (struct tblock *) vmalloc(size);
286 if (TxBlock == NULL)
287 return -ENOMEM;
288
289 for (k = 1; k < nTxBlock - 1; k++) {
290 TxBlock[k].next = k + 1;
291 init_waitqueue_head(&TxBlock[k].gcwait);
292 init_waitqueue_head(&TxBlock[k].waitor);
293 }
294 TxBlock[k].next = 0;
295 init_waitqueue_head(&TxBlock[k].gcwait);
296 init_waitqueue_head(&TxBlock[k].waitor);
297
298 TxAnchor.freetid = 1;
299 init_waitqueue_head(&TxAnchor.freewait);
300
301 stattx.maxtid = 1; /* statistics */
302
303 /*
304 * initialize transaction lock (tlock) table
305 *
306 * transaction lock id = tlock index
307 * tlock id = 0 is reserved.
308 */
309 size = sizeof(struct tlock) * nTxLock;
310 TxLock = (struct tlock *) vmalloc(size);
311 if (TxLock == NULL) {
312 vfree(TxBlock);
313 return -ENOMEM;
314 }
315
316 /* initialize tlock table */
317 for (k = 1; k < nTxLock - 1; k++)
318 TxLock[k].next = k + 1;
319 TxLock[k].next = 0;
320 init_waitqueue_head(&TxAnchor.freelockwait);
321 init_waitqueue_head(&TxAnchor.lowlockwait);
322
323 TxAnchor.freelock = 1;
324 TxAnchor.tlocksInUse = 0;
325 INIT_LIST_HEAD(&TxAnchor.anon_list);
326 INIT_LIST_HEAD(&TxAnchor.anon_list2);
327
328 LAZY_LOCK_INIT();
329 INIT_LIST_HEAD(&TxAnchor.unlock_queue);
330
331 stattx.maxlid = 1; /* statistics */
332
333 return 0;
334}
335
336/*
337 * NAME: txExit()
338 *
339 * FUNCTION: clean up when module is unloaded
340 */
341void txExit(void)
342{
343 vfree(TxLock);
344 TxLock = NULL;
345 vfree(TxBlock);
346 TxBlock = NULL;
347}
348
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349/*
350 * NAME: txBegin()
351 *
352 * FUNCTION: start a transaction.
353 *
354 * PARAMETER: sb - superblock
355 * flag - force for nested tx;
356 *
357 * RETURN: tid - transaction id
358 *
359 * note: flag force allows to start tx for nested tx
360 * to prevent deadlock on logsync barrier;
361 */
362tid_t txBegin(struct super_block *sb, int flag)
363{
364 tid_t t;
365 struct tblock *tblk;
366 struct jfs_log *log;
367
368 jfs_info("txBegin: flag = 0x%x", flag);
369 log = JFS_SBI(sb)->log;
370
371 TXN_LOCK();
372
373 INCREMENT(TxStat.txBegin);
374
375 retry:
376 if (!(flag & COMMIT_FORCE)) {
377 /*
378 * synchronize with logsync barrier
379 */
380 if (test_bit(log_SYNCBARRIER, &log->flag) ||
381 test_bit(log_QUIESCE, &log->flag)) {
382 INCREMENT(TxStat.txBegin_barrier);
383 TXN_SLEEP(&log->syncwait);
384 goto retry;
385 }
386 }
387 if (flag == 0) {
388 /*
389 * Don't begin transaction if we're getting starved for tlocks
390 * unless COMMIT_FORCE or COMMIT_INODE (which may ultimately
391 * free tlocks)
392 */
393 if (TxAnchor.tlocksInUse > TxLockVHWM) {
394 INCREMENT(TxStat.txBegin_lockslow);
395 TXN_SLEEP(&TxAnchor.lowlockwait);
396 goto retry;
397 }
398 }
399
400 /*
401 * allocate transaction id/block
402 */
403 if ((t = TxAnchor.freetid) == 0) {
404 jfs_info("txBegin: waiting for free tid");
405 INCREMENT(TxStat.txBegin_freetid);
406 TXN_SLEEP(&TxAnchor.freewait);
407 goto retry;
408 }
409
410 tblk = tid_to_tblock(t);
411
412 if ((tblk->next == 0) && !(flag & COMMIT_FORCE)) {
413 /* Don't let a non-forced transaction take the last tblk */
414 jfs_info("txBegin: waiting for free tid");
415 INCREMENT(TxStat.txBegin_freetid);
416 TXN_SLEEP(&TxAnchor.freewait);
417 goto retry;
418 }
419
420 TxAnchor.freetid = tblk->next;
421
422 /*
423 * initialize transaction
424 */
425
426 /*
427 * We can't zero the whole thing or we screw up another thread being
428 * awakened after sleeping on tblk->waitor
429 *
430 * memset(tblk, 0, sizeof(struct tblock));
431 */
432 tblk->next = tblk->last = tblk->xflag = tblk->flag = tblk->lsn = 0;
433
434 tblk->sb = sb;
435 ++log->logtid;
436 tblk->logtid = log->logtid;
437
438 ++log->active;
439
440 HIGHWATERMARK(stattx.maxtid, t); /* statistics */
441 INCREMENT(stattx.ntid); /* statistics */
442
443 TXN_UNLOCK();
444
445 jfs_info("txBegin: returning tid = %d", t);
446
447 return t;
448}
449
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450/*
451 * NAME: txBeginAnon()
452 *
453 * FUNCTION: start an anonymous transaction.
454 * Blocks if logsync or available tlocks are low to prevent
455 * anonymous tlocks from depleting supply.
456 *
457 * PARAMETER: sb - superblock
458 *
459 * RETURN: none
460 */
461void txBeginAnon(struct super_block *sb)
462{
463 struct jfs_log *log;
464
465 log = JFS_SBI(sb)->log;
466
467 TXN_LOCK();
468 INCREMENT(TxStat.txBeginAnon);
469
470 retry:
471 /*
472 * synchronize with logsync barrier
473 */
474 if (test_bit(log_SYNCBARRIER, &log->flag) ||
475 test_bit(log_QUIESCE, &log->flag)) {
476 INCREMENT(TxStat.txBeginAnon_barrier);
477 TXN_SLEEP(&log->syncwait);
478 goto retry;
479 }
480
481 /*
482 * Don't begin transaction if we're getting starved for tlocks
483 */
484 if (TxAnchor.tlocksInUse > TxLockVHWM) {
485 INCREMENT(TxStat.txBeginAnon_lockslow);
486 TXN_SLEEP(&TxAnchor.lowlockwait);
487 goto retry;
488 }
489 TXN_UNLOCK();
490}
491
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492/*
493 * txEnd()
494 *
495 * function: free specified transaction block.
496 *
497 * logsync barrier processing:
498 *
499 * serialization:
500 */
501void txEnd(tid_t tid)
502{
503 struct tblock *tblk = tid_to_tblock(tid);
504 struct jfs_log *log;
505
506 jfs_info("txEnd: tid = %d", tid);
507 TXN_LOCK();
508
509 /*
510 * wakeup transactions waiting on the page locked
511 * by the current transaction
512 */
513 TXN_WAKEUP(&tblk->waitor);
514
515 log = JFS_SBI(tblk->sb)->log;
516
517 /*
518 * Lazy commit thread can't free this guy until we mark it UNLOCKED,
519 * otherwise, we would be left with a transaction that may have been
520 * reused.
521 *
522 * Lazy commit thread will turn off tblkGC_LAZY before calling this
523 * routine.
524 */
525 if (tblk->flag & tblkGC_LAZY) {
526 jfs_info("txEnd called w/lazy tid: %d, tblk = 0x%p", tid, tblk);
527 TXN_UNLOCK();
528
529 spin_lock_irq(&log->gclock); // LOGGC_LOCK
530 tblk->flag |= tblkGC_UNLOCKED;
531 spin_unlock_irq(&log->gclock); // LOGGC_UNLOCK
532 return;
533 }
534
535 jfs_info("txEnd: tid: %d, tblk = 0x%p", tid, tblk);
536
537 assert(tblk->next == 0);
538
539 /*
540 * insert tblock back on freelist
541 */
542 tblk->next = TxAnchor.freetid;
543 TxAnchor.freetid = tid;
544
545 /*
546 * mark the tblock not active
547 */
548 if (--log->active == 0) {
549 clear_bit(log_FLUSH, &log->flag);
550
551 /*
552 * synchronize with logsync barrier
553 */
554 if (test_bit(log_SYNCBARRIER, &log->flag)) {
Dave Kleikampcbc3d652005-07-27 09:17:57 -0500555 TXN_UNLOCK();
556
557 /* write dirty metadata & forward log syncpt */
558 jfs_syncpt(log, 1);
559
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 jfs_info("log barrier off: 0x%x", log->lsn);
561
562 /* enable new transactions start */
563 clear_bit(log_SYNCBARRIER, &log->flag);
564
565 /* wakeup all waitors for logsync barrier */
566 TXN_WAKEUP(&log->syncwait);
Dave Kleikamp1c627822005-05-02 12:25:08 -0600567
Dave Kleikamp1c627822005-05-02 12:25:08 -0600568 goto wakeup;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 }
570 }
571
Dave Kleikamp1c627822005-05-02 12:25:08 -0600572 TXN_UNLOCK();
573wakeup:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 /*
575 * wakeup all waitors for a free tblock
576 */
577 TXN_WAKEUP(&TxAnchor.freewait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578}
579
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580/*
581 * txLock()
582 *
583 * function: acquire a transaction lock on the specified <mp>
584 *
585 * parameter:
586 *
587 * return: transaction lock id
588 *
589 * serialization:
590 */
591struct tlock *txLock(tid_t tid, struct inode *ip, struct metapage * mp,
592 int type)
593{
594 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
595 int dir_xtree = 0;
596 lid_t lid;
597 tid_t xtid;
598 struct tlock *tlck;
599 struct xtlock *xtlck;
600 struct linelock *linelock;
601 xtpage_t *p;
602 struct tblock *tblk;
603
604 TXN_LOCK();
605
606 if (S_ISDIR(ip->i_mode) && (type & tlckXTREE) &&
607 !(mp->xflag & COMMIT_PAGE)) {
608 /*
609 * Directory inode is special. It can have both an xtree tlock
610 * and a dtree tlock associated with it.
611 */
612 dir_xtree = 1;
613 lid = jfs_ip->xtlid;
614 } else
615 lid = mp->lid;
616
617 /* is page not locked by a transaction ? */
618 if (lid == 0)
619 goto allocateLock;
620
621 jfs_info("txLock: tid:%d ip:0x%p mp:0x%p lid:%d", tid, ip, mp, lid);
622
623 /* is page locked by the requester transaction ? */
624 tlck = lid_to_tlock(lid);
Dave Kleikamp7fab4792005-05-02 12:25:02 -0600625 if ((xtid = tlck->tid) == tid) {
626 TXN_UNLOCK();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 goto grantLock;
Dave Kleikamp7fab4792005-05-02 12:25:02 -0600628 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629
630 /*
631 * is page locked by anonymous transaction/lock ?
632 *
633 * (page update without transaction (i.e., file write) is
634 * locked under anonymous transaction tid = 0:
635 * anonymous tlocks maintained on anonymous tlock list of
636 * the inode of the page and available to all anonymous
637 * transactions until txCommit() time at which point
638 * they are transferred to the transaction tlock list of
639 * the commiting transaction of the inode)
640 */
641 if (xtid == 0) {
642 tlck->tid = tid;
Dave Kleikamp7fab4792005-05-02 12:25:02 -0600643 TXN_UNLOCK();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 tblk = tid_to_tblock(tid);
645 /*
646 * The order of the tlocks in the transaction is important
647 * (during truncate, child xtree pages must be freed before
648 * parent's tlocks change the working map).
649 * Take tlock off anonymous list and add to tail of
650 * transaction list
651 *
652 * Note: We really need to get rid of the tid & lid and
653 * use list_head's. This code is getting UGLY!
654 */
655 if (jfs_ip->atlhead == lid) {
656 if (jfs_ip->atltail == lid) {
657 /* only anonymous txn.
658 * Remove from anon_list
659 */
Dave Kleikamp8a9cd6d2005-08-10 11:14:39 -0500660 TXN_LOCK();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 list_del_init(&jfs_ip->anon_inode_list);
Dave Kleikamp8a9cd6d2005-08-10 11:14:39 -0500662 TXN_UNLOCK();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 }
664 jfs_ip->atlhead = tlck->next;
665 } else {
666 lid_t last;
667 for (last = jfs_ip->atlhead;
668 lid_to_tlock(last)->next != lid;
669 last = lid_to_tlock(last)->next) {
670 assert(last);
671 }
672 lid_to_tlock(last)->next = tlck->next;
673 if (jfs_ip->atltail == lid)
674 jfs_ip->atltail = last;
675 }
676
677 /* insert the tlock at tail of transaction tlock list */
678
679 if (tblk->next)
680 lid_to_tlock(tblk->last)->next = lid;
681 else
682 tblk->next = lid;
683 tlck->next = 0;
684 tblk->last = lid;
685
686 goto grantLock;
687 }
688
689 goto waitLock;
690
691 /*
692 * allocate a tlock
693 */
694 allocateLock:
695 lid = txLockAlloc();
696 tlck = lid_to_tlock(lid);
697
698 /*
699 * initialize tlock
700 */
701 tlck->tid = tid;
702
Dave Kleikamp7fab4792005-05-02 12:25:02 -0600703 TXN_UNLOCK();
704
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 /* mark tlock for meta-data page */
706 if (mp->xflag & COMMIT_PAGE) {
707
708 tlck->flag = tlckPAGELOCK;
709
710 /* mark the page dirty and nohomeok */
Dave Kleikamp7fab4792005-05-02 12:25:02 -0600711 metapage_nohomeok(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712
713 jfs_info("locking mp = 0x%p, nohomeok = %d tid = %d tlck = 0x%p",
Dave Kleikamp7fab4792005-05-02 12:25:02 -0600714 mp, mp->nohomeok, tid, tlck);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715
716 /* if anonymous transaction, and buffer is on the group
717 * commit synclist, mark inode to show this. This will
718 * prevent the buffer from being marked nohomeok for too
719 * long a time.
720 */
721 if ((tid == 0) && mp->lsn)
722 set_cflag(COMMIT_Synclist, ip);
723 }
724 /* mark tlock for in-memory inode */
725 else
726 tlck->flag = tlckINODELOCK;
727
Dave Kleikamp438282d2005-09-20 14:58:11 -0500728 if (S_ISDIR(ip->i_mode))
729 tlck->flag |= tlckDIRECTORY;
730
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 tlck->type = 0;
732
733 /* bind the tlock and the page */
734 tlck->ip = ip;
735 tlck->mp = mp;
736 if (dir_xtree)
737 jfs_ip->xtlid = lid;
738 else
739 mp->lid = lid;
740
741 /*
742 * enqueue transaction lock to transaction/inode
743 */
744 /* insert the tlock at tail of transaction tlock list */
745 if (tid) {
746 tblk = tid_to_tblock(tid);
747 if (tblk->next)
748 lid_to_tlock(tblk->last)->next = lid;
749 else
750 tblk->next = lid;
751 tlck->next = 0;
752 tblk->last = lid;
753 }
754 /* anonymous transaction:
755 * insert the tlock at head of inode anonymous tlock list
756 */
757 else {
758 tlck->next = jfs_ip->atlhead;
759 jfs_ip->atlhead = lid;
760 if (tlck->next == 0) {
761 /* This inode's first anonymous transaction */
762 jfs_ip->atltail = lid;
Dave Kleikamp7fab4792005-05-02 12:25:02 -0600763 TXN_LOCK();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 list_add_tail(&jfs_ip->anon_inode_list,
765 &TxAnchor.anon_list);
Dave Kleikamp7fab4792005-05-02 12:25:02 -0600766 TXN_UNLOCK();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 }
768 }
769
770 /* initialize type dependent area for linelock */
771 linelock = (struct linelock *) & tlck->lock;
772 linelock->next = 0;
773 linelock->flag = tlckLINELOCK;
774 linelock->maxcnt = TLOCKSHORT;
775 linelock->index = 0;
776
777 switch (type & tlckTYPE) {
778 case tlckDTREE:
779 linelock->l2linesize = L2DTSLOTSIZE;
780 break;
781
782 case tlckXTREE:
783 linelock->l2linesize = L2XTSLOTSIZE;
784
785 xtlck = (struct xtlock *) linelock;
786 xtlck->header.offset = 0;
787 xtlck->header.length = 2;
788
789 if (type & tlckNEW) {
790 xtlck->lwm.offset = XTENTRYSTART;
791 } else {
792 if (mp->xflag & COMMIT_PAGE)
793 p = (xtpage_t *) mp->data;
794 else
795 p = &jfs_ip->i_xtroot;
796 xtlck->lwm.offset =
797 le16_to_cpu(p->header.nextindex);
798 }
799 xtlck->lwm.length = 0; /* ! */
800 xtlck->twm.offset = 0;
801 xtlck->hwm.offset = 0;
802
803 xtlck->index = 2;
804 break;
805
806 case tlckINODE:
807 linelock->l2linesize = L2INODESLOTSIZE;
808 break;
809
810 case tlckDATA:
811 linelock->l2linesize = L2DATASLOTSIZE;
812 break;
813
814 default:
815 jfs_err("UFO tlock:0x%p", tlck);
816 }
817
818 /*
819 * update tlock vector
820 */
821 grantLock:
822 tlck->type |= type;
823
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 return tlck;
825
826 /*
827 * page is being locked by another transaction:
828 */
829 waitLock:
830 /* Only locks on ipimap or ipaimap should reach here */
831 /* assert(jfs_ip->fileset == AGGREGATE_I); */
832 if (jfs_ip->fileset != AGGREGATE_I) {
833 jfs_err("txLock: trying to lock locked page!");
834 dump_mem("ip", ip, sizeof(struct inode));
835 dump_mem("mp", mp, sizeof(struct metapage));
836 dump_mem("Locker's tblk", tid_to_tblock(tid),
837 sizeof(struct tblock));
838 dump_mem("Tlock", tlck, sizeof(struct tlock));
839 BUG();
840 }
841 INCREMENT(stattx.waitlock); /* statistics */
Dave Kleikamp7fab4792005-05-02 12:25:02 -0600842 TXN_UNLOCK();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 release_metapage(mp);
Dave Kleikamp7fab4792005-05-02 12:25:02 -0600844 TXN_LOCK();
845 xtid = tlck->tid; /* reaquire after dropping TXN_LOCK */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846
847 jfs_info("txLock: in waitLock, tid = %d, xtid = %d, lid = %d",
848 tid, xtid, lid);
Dave Kleikamp7fab4792005-05-02 12:25:02 -0600849
850 /* Recheck everything since dropping TXN_LOCK */
851 if (xtid && (tlck->mp == mp) && (mp->lid == lid))
852 TXN_SLEEP_DROP_LOCK(&tid_to_tblock(xtid)->waitor);
853 else
854 TXN_UNLOCK();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 jfs_info("txLock: awakened tid = %d, lid = %d", tid, lid);
856
857 return NULL;
858}
859
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860/*
861 * NAME: txRelease()
862 *
863 * FUNCTION: Release buffers associated with transaction locks, but don't
864 * mark homeok yet. The allows other transactions to modify
865 * buffers, but won't let them go to disk until commit record
866 * actually gets written.
867 *
868 * PARAMETER:
869 * tblk -
870 *
871 * RETURN: Errors from subroutines.
872 */
873static void txRelease(struct tblock * tblk)
874{
875 struct metapage *mp;
876 lid_t lid;
877 struct tlock *tlck;
878
879 TXN_LOCK();
880
881 for (lid = tblk->next; lid; lid = tlck->next) {
882 tlck = lid_to_tlock(lid);
883 if ((mp = tlck->mp) != NULL &&
884 (tlck->type & tlckBTROOT) == 0) {
885 assert(mp->xflag & COMMIT_PAGE);
886 mp->lid = 0;
887 }
888 }
889
890 /*
891 * wakeup transactions waiting on a page locked
892 * by the current transaction
893 */
894 TXN_WAKEUP(&tblk->waitor);
895
896 TXN_UNLOCK();
897}
898
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899/*
900 * NAME: txUnlock()
901 *
902 * FUNCTION: Initiates pageout of pages modified by tid in journalled
903 * objects and frees their lockwords.
904 */
905static void txUnlock(struct tblock * tblk)
906{
907 struct tlock *tlck;
908 struct linelock *linelock;
909 lid_t lid, next, llid, k;
910 struct metapage *mp;
911 struct jfs_log *log;
912 int difft, diffp;
Dave Kleikamp7fab4792005-05-02 12:25:02 -0600913 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914
915 jfs_info("txUnlock: tblk = 0x%p", tblk);
916 log = JFS_SBI(tblk->sb)->log;
917
918 /*
919 * mark page under tlock homeok (its log has been written):
920 */
921 for (lid = tblk->next; lid; lid = next) {
922 tlck = lid_to_tlock(lid);
923 next = tlck->next;
924
925 jfs_info("unlocking lid = %d, tlck = 0x%p", lid, tlck);
926
927 /* unbind page from tlock */
928 if ((mp = tlck->mp) != NULL &&
929 (tlck->type & tlckBTROOT) == 0) {
930 assert(mp->xflag & COMMIT_PAGE);
931
932 /* hold buffer
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 */
Dave Kleikamp7fab4792005-05-02 12:25:02 -0600934 hold_metapage(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935
Dave Kleikamp7fab4792005-05-02 12:25:02 -0600936 assert(mp->nohomeok > 0);
937 _metapage_homeok(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938
939 /* inherit younger/larger clsn */
Dave Kleikamp7fab4792005-05-02 12:25:02 -0600940 LOGSYNC_LOCK(log, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 if (mp->clsn) {
942 logdiff(difft, tblk->clsn, log);
943 logdiff(diffp, mp->clsn, log);
944 if (difft > diffp)
945 mp->clsn = tblk->clsn;
946 } else
947 mp->clsn = tblk->clsn;
Dave Kleikamp7fab4792005-05-02 12:25:02 -0600948 LOGSYNC_UNLOCK(log, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949
950 assert(!(tlck->flag & tlckFREEPAGE));
951
Dave Kleikamp7fab4792005-05-02 12:25:02 -0600952 put_metapage(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 }
954
955 /* insert tlock, and linelock(s) of the tlock if any,
956 * at head of freelist
957 */
958 TXN_LOCK();
959
960 llid = ((struct linelock *) & tlck->lock)->next;
961 while (llid) {
962 linelock = (struct linelock *) lid_to_tlock(llid);
963 k = linelock->next;
964 txLockFree(llid);
965 llid = k;
966 }
967 txLockFree(lid);
968
969 TXN_UNLOCK();
970 }
971 tblk->next = tblk->last = 0;
972
973 /*
974 * remove tblock from logsynclist
975 * (allocation map pages inherited lsn of tblk and
976 * has been inserted in logsync list at txUpdateMap())
977 */
978 if (tblk->lsn) {
Dave Kleikamp7fab4792005-05-02 12:25:02 -0600979 LOGSYNC_LOCK(log, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 log->count--;
981 list_del(&tblk->synclist);
Dave Kleikamp7fab4792005-05-02 12:25:02 -0600982 LOGSYNC_UNLOCK(log, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 }
984}
985
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986/*
987 * txMaplock()
988 *
989 * function: allocate a transaction lock for freed page/entry;
990 * for freed page, maplock is used as xtlock/dtlock type;
991 */
992struct tlock *txMaplock(tid_t tid, struct inode *ip, int type)
993{
994 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
995 lid_t lid;
996 struct tblock *tblk;
997 struct tlock *tlck;
998 struct maplock *maplock;
999
1000 TXN_LOCK();
1001
1002 /*
1003 * allocate a tlock
1004 */
1005 lid = txLockAlloc();
1006 tlck = lid_to_tlock(lid);
1007
1008 /*
1009 * initialize tlock
1010 */
1011 tlck->tid = tid;
1012
1013 /* bind the tlock and the object */
1014 tlck->flag = tlckINODELOCK;
Dave Kleikamp438282d2005-09-20 14:58:11 -05001015 if (S_ISDIR(ip->i_mode))
1016 tlck->flag |= tlckDIRECTORY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 tlck->ip = ip;
1018 tlck->mp = NULL;
1019
1020 tlck->type = type;
1021
1022 /*
1023 * enqueue transaction lock to transaction/inode
1024 */
1025 /* insert the tlock at tail of transaction tlock list */
1026 if (tid) {
1027 tblk = tid_to_tblock(tid);
1028 if (tblk->next)
1029 lid_to_tlock(tblk->last)->next = lid;
1030 else
1031 tblk->next = lid;
1032 tlck->next = 0;
1033 tblk->last = lid;
1034 }
1035 /* anonymous transaction:
1036 * insert the tlock at head of inode anonymous tlock list
1037 */
1038 else {
1039 tlck->next = jfs_ip->atlhead;
1040 jfs_ip->atlhead = lid;
1041 if (tlck->next == 0) {
1042 /* This inode's first anonymous transaction */
1043 jfs_ip->atltail = lid;
1044 list_add_tail(&jfs_ip->anon_inode_list,
1045 &TxAnchor.anon_list);
1046 }
1047 }
1048
1049 TXN_UNLOCK();
1050
1051 /* initialize type dependent area for maplock */
1052 maplock = (struct maplock *) & tlck->lock;
1053 maplock->next = 0;
1054 maplock->maxcnt = 0;
1055 maplock->index = 0;
1056
1057 return tlck;
1058}
1059
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060/*
1061 * txLinelock()
1062 *
1063 * function: allocate a transaction lock for log vector list
1064 */
1065struct linelock *txLinelock(struct linelock * tlock)
1066{
1067 lid_t lid;
1068 struct tlock *tlck;
1069 struct linelock *linelock;
1070
1071 TXN_LOCK();
1072
1073 /* allocate a TxLock structure */
1074 lid = txLockAlloc();
1075 tlck = lid_to_tlock(lid);
1076
1077 TXN_UNLOCK();
1078
1079 /* initialize linelock */
1080 linelock = (struct linelock *) tlck;
1081 linelock->next = 0;
1082 linelock->flag = tlckLINELOCK;
1083 linelock->maxcnt = TLOCKLONG;
1084 linelock->index = 0;
Dave Kleikamp438282d2005-09-20 14:58:11 -05001085 if (tlck->flag & tlckDIRECTORY)
1086 linelock->flag |= tlckDIRECTORY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087
1088 /* append linelock after tlock */
1089 linelock->next = tlock->next;
1090 tlock->next = lid;
1091
1092 return linelock;
1093}
1094
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095/*
1096 * transaction commit management
1097 * -----------------------------
1098 */
1099
1100/*
1101 * NAME: txCommit()
1102 *
1103 * FUNCTION: commit the changes to the objects specified in
1104 * clist. For journalled segments only the
1105 * changes of the caller are committed, ie by tid.
1106 * for non-journalled segments the data are flushed to
1107 * disk and then the change to the disk inode and indirect
1108 * blocks committed (so blocks newly allocated to the
1109 * segment will be made a part of the segment atomically).
1110 *
1111 * all of the segments specified in clist must be in
1112 * one file system. no more than 6 segments are needed
1113 * to handle all unix svcs.
1114 *
1115 * if the i_nlink field (i.e. disk inode link count)
1116 * is zero, and the type of inode is a regular file or
1117 * directory, or symbolic link , the inode is truncated
1118 * to zero length. the truncation is committed but the
1119 * VM resources are unaffected until it is closed (see
1120 * iput and iclose).
1121 *
1122 * PARAMETER:
1123 *
1124 * RETURN:
1125 *
1126 * serialization:
1127 * on entry the inode lock on each segment is assumed
1128 * to be held.
1129 *
1130 * i/o error:
1131 */
1132int txCommit(tid_t tid, /* transaction identifier */
1133 int nip, /* number of inodes to commit */
1134 struct inode **iplist, /* list of inode to commit */
1135 int flag)
1136{
1137 int rc = 0;
1138 struct commit cd;
1139 struct jfs_log *log;
1140 struct tblock *tblk;
1141 struct lrd *lrd;
1142 int lsn;
1143 struct inode *ip;
1144 struct jfs_inode_info *jfs_ip;
1145 int k, n;
1146 ino_t top;
1147 struct super_block *sb;
1148
1149 jfs_info("txCommit, tid = %d, flag = %d", tid, flag);
1150 /* is read-only file system ? */
1151 if (isReadOnly(iplist[0])) {
1152 rc = -EROFS;
1153 goto TheEnd;
1154 }
1155
1156 sb = cd.sb = iplist[0]->i_sb;
1157 cd.tid = tid;
1158
1159 if (tid == 0)
1160 tid = txBegin(sb, 0);
1161 tblk = tid_to_tblock(tid);
1162
1163 /*
1164 * initialize commit structure
1165 */
1166 log = JFS_SBI(sb)->log;
1167 cd.log = log;
1168
1169 /* initialize log record descriptor in commit */
1170 lrd = &cd.lrd;
1171 lrd->logtid = cpu_to_le32(tblk->logtid);
1172 lrd->backchain = 0;
1173
1174 tblk->xflag |= flag;
1175
1176 if ((flag & (COMMIT_FORCE | COMMIT_SYNC)) == 0)
1177 tblk->xflag |= COMMIT_LAZY;
1178 /*
1179 * prepare non-journaled objects for commit
1180 *
1181 * flush data pages of non-journaled file
1182 * to prevent the file getting non-initialized disk blocks
1183 * in case of crash.
1184 * (new blocks - )
1185 */
1186 cd.iplist = iplist;
1187 cd.nip = nip;
1188
1189 /*
1190 * acquire transaction lock on (on-disk) inodes
1191 *
1192 * update on-disk inode from in-memory inode
1193 * acquiring transaction locks for AFTER records
1194 * on the on-disk inode of file object
1195 *
1196 * sort the inodes array by inode number in descending order
1197 * to prevent deadlock when acquiring transaction lock
1198 * of on-disk inodes on multiple on-disk inode pages by
1199 * multiple concurrent transactions
1200 */
1201 for (k = 0; k < cd.nip; k++) {
1202 top = (cd.iplist[k])->i_ino;
1203 for (n = k + 1; n < cd.nip; n++) {
1204 ip = cd.iplist[n];
1205 if (ip->i_ino > top) {
1206 top = ip->i_ino;
1207 cd.iplist[n] = cd.iplist[k];
1208 cd.iplist[k] = ip;
1209 }
1210 }
1211
1212 ip = cd.iplist[k];
1213 jfs_ip = JFS_IP(ip);
1214
1215 /*
1216 * BUGBUG - This code has temporarily been removed. The
1217 * intent is to ensure that any file data is written before
1218 * the metadata is committed to the journal. This prevents
1219 * uninitialized data from appearing in a file after the
1220 * journal has been replayed. (The uninitialized data
1221 * could be sensitive data removed by another user.)
1222 *
1223 * The problem now is that we are holding the IWRITELOCK
1224 * on the inode, and calling filemap_fdatawrite on an
1225 * unmapped page will cause a deadlock in jfs_get_block.
1226 *
1227 * The long term solution is to pare down the use of
1228 * IWRITELOCK. We are currently holding it too long.
1229 * We could also be smarter about which data pages need
1230 * to be written before the transaction is committed and
1231 * when we don't need to worry about it at all.
1232 *
1233 * if ((!S_ISDIR(ip->i_mode))
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -08001234 * && (tblk->flag & COMMIT_DELETE) == 0)
1235 * filemap_write_and_wait(ip->i_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 */
1237
1238 /*
1239 * Mark inode as not dirty. It will still be on the dirty
1240 * inode list, but we'll know not to commit it again unless
1241 * it gets marked dirty again
1242 */
1243 clear_cflag(COMMIT_Dirty, ip);
1244
1245 /* inherit anonymous tlock(s) of inode */
1246 if (jfs_ip->atlhead) {
1247 lid_to_tlock(jfs_ip->atltail)->next = tblk->next;
1248 tblk->next = jfs_ip->atlhead;
1249 if (!tblk->last)
1250 tblk->last = jfs_ip->atltail;
1251 jfs_ip->atlhead = jfs_ip->atltail = 0;
1252 TXN_LOCK();
1253 list_del_init(&jfs_ip->anon_inode_list);
1254 TXN_UNLOCK();
1255 }
1256
1257 /*
1258 * acquire transaction lock on on-disk inode page
1259 * (become first tlock of the tblk's tlock list)
1260 */
1261 if (((rc = diWrite(tid, ip))))
1262 goto out;
1263 }
1264
1265 /*
1266 * write log records from transaction locks
1267 *
1268 * txUpdateMap() resets XAD_NEW in XAD.
1269 */
1270 if ((rc = txLog(log, tblk, &cd)))
1271 goto TheEnd;
1272
1273 /*
1274 * Ensure that inode isn't reused before
1275 * lazy commit thread finishes processing
1276 */
1277 if (tblk->xflag & COMMIT_DELETE) {
1278 atomic_inc(&tblk->u.ip->i_count);
1279 /*
1280 * Avoid a rare deadlock
1281 *
1282 * If the inode is locked, we may be blocked in
1283 * jfs_commit_inode. If so, we don't want the
1284 * lazy_commit thread doing the last iput() on the inode
1285 * since that may block on the locked inode. Instead,
1286 * commit the transaction synchronously, so the last iput
1287 * will be done by the calling thread (or later)
1288 */
1289 if (tblk->u.ip->i_state & I_LOCK)
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 lsn = 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
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361/*
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 int txLog(struct jfs_log * log, struct tblock * tblk, struct commit * cd)
1373{
1374 int rc = 0;
1375 struct inode *ip;
1376 lid_t lid;
1377 struct tlock *tlck;
1378 struct lrd *lrd = &cd->lrd;
1379
1380 /*
1381 * write log record(s) for each tlock of transaction,
1382 */
1383 for (lid = tblk->next; lid; lid = tlck->next) {
1384 tlck = lid_to_tlock(lid);
1385
1386 tlck->flag |= tlckLOG;
1387
1388 /* initialize lrd common */
1389 ip = tlck->ip;
1390 lrd->aggregate = cpu_to_le32(JFS_SBI(ip->i_sb)->aggregate);
1391 lrd->log.redopage.fileset = cpu_to_le32(JFS_IP(ip)->fileset);
1392 lrd->log.redopage.inode = cpu_to_le32(ip->i_ino);
1393
1394 /* write log record of page from the tlock */
1395 switch (tlck->type & tlckTYPE) {
1396 case tlckXTREE:
1397 xtLog(log, tblk, lrd, tlck);
1398 break;
1399
1400 case tlckDTREE:
1401 dtLog(log, tblk, lrd, tlck);
1402 break;
1403
1404 case tlckINODE:
1405 diLog(log, tblk, lrd, tlck, cd);
1406 break;
1407
1408 case tlckMAP:
1409 mapLog(log, tblk, lrd, tlck);
1410 break;
1411
1412 case tlckDATA:
1413 dataLog(log, tblk, lrd, tlck);
1414 break;
1415
1416 default:
1417 jfs_err("UFO tlock:0x%p", tlck);
1418 }
1419 }
1420
1421 return rc;
1422}
1423
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424/*
1425 * diLog()
1426 *
1427 * function: log inode tlock and format maplock to update bmap;
1428 */
1429static int diLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1430 struct tlock * tlck, struct commit * cd)
1431{
1432 int rc = 0;
1433 struct metapage *mp;
1434 pxd_t *pxd;
1435 struct pxd_lock *pxdlock;
1436
1437 mp = tlck->mp;
1438
1439 /* initialize as REDOPAGE record format */
1440 lrd->log.redopage.type = cpu_to_le16(LOG_INODE);
1441 lrd->log.redopage.l2linesize = cpu_to_le16(L2INODESLOTSIZE);
1442
1443 pxd = &lrd->log.redopage.pxd;
1444
1445 /*
1446 * inode after image
1447 */
1448 if (tlck->type & tlckENTRY) {
1449 /* log after-image for logredo(): */
1450 lrd->type = cpu_to_le16(LOG_REDOPAGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 PXDaddress(pxd, mp->index);
1452 PXDlength(pxd,
1453 mp->logical_size >> tblk->sb->s_blocksize_bits);
1454 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1455
1456 /* mark page as homeward bound */
1457 tlck->flag |= tlckWRITEPAGE;
1458 } else if (tlck->type & tlckFREE) {
1459 /*
1460 * free inode extent
1461 *
1462 * (pages of the freed inode extent have been invalidated and
1463 * a maplock for free of the extent has been formatted at
1464 * txLock() time);
1465 *
1466 * the tlock had been acquired on the inode allocation map page
1467 * (iag) that specifies the freed extent, even though the map
1468 * page is not itself logged, to prevent pageout of the map
1469 * page before the log;
1470 */
1471
1472 /* log LOG_NOREDOINOEXT of the freed inode extent for
1473 * logredo() to start NoRedoPage filters, and to update
1474 * imap and bmap for free of the extent;
1475 */
1476 lrd->type = cpu_to_le16(LOG_NOREDOINOEXT);
1477 /*
1478 * For the LOG_NOREDOINOEXT record, we need
1479 * to pass the IAG number and inode extent
1480 * index (within that IAG) from which the
1481 * the extent being released. These have been
1482 * passed to us in the iplist[1] and iplist[2].
1483 */
1484 lrd->log.noredoinoext.iagnum =
1485 cpu_to_le32((u32) (size_t) cd->iplist[1]);
1486 lrd->log.noredoinoext.inoext_idx =
1487 cpu_to_le32((u32) (size_t) cd->iplist[2]);
1488
1489 pxdlock = (struct pxd_lock *) & tlck->lock;
1490 *pxd = pxdlock->pxd;
1491 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1492
1493 /* update bmap */
1494 tlck->flag |= tlckUPDATEMAP;
1495
1496 /* mark page as homeward bound */
1497 tlck->flag |= tlckWRITEPAGE;
1498 } else
1499 jfs_err("diLog: UFO type tlck:0x%p", tlck);
1500#ifdef _JFS_WIP
1501 /*
1502 * alloc/free external EA extent
1503 *
1504 * a maplock for txUpdateMap() to update bPWMAP for alloc/free
1505 * of the extent has been formatted at txLock() time;
1506 */
1507 else {
1508 assert(tlck->type & tlckEA);
1509
1510 /* log LOG_UPDATEMAP for logredo() to update bmap for
1511 * alloc of new (and free of old) external EA extent;
1512 */
1513 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1514 pxdlock = (struct pxd_lock *) & tlck->lock;
1515 nlock = pxdlock->index;
1516 for (i = 0; i < nlock; i++, pxdlock++) {
1517 if (pxdlock->flag & mlckALLOCPXD)
1518 lrd->log.updatemap.type =
1519 cpu_to_le16(LOG_ALLOCPXD);
1520 else
1521 lrd->log.updatemap.type =
1522 cpu_to_le16(LOG_FREEPXD);
1523 lrd->log.updatemap.nxd = cpu_to_le16(1);
1524 lrd->log.updatemap.pxd = pxdlock->pxd;
1525 lrd->backchain =
1526 cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1527 }
1528
1529 /* update bmap */
1530 tlck->flag |= tlckUPDATEMAP;
1531 }
1532#endif /* _JFS_WIP */
1533
1534 return rc;
1535}
1536
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537/*
1538 * dataLog()
1539 *
1540 * function: log data tlock
1541 */
1542static int dataLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1543 struct tlock * tlck)
1544{
1545 struct metapage *mp;
1546 pxd_t *pxd;
1547
1548 mp = tlck->mp;
1549
1550 /* initialize as REDOPAGE record format */
1551 lrd->log.redopage.type = cpu_to_le16(LOG_DATA);
1552 lrd->log.redopage.l2linesize = cpu_to_le16(L2DATASLOTSIZE);
1553
1554 pxd = &lrd->log.redopage.pxd;
1555
1556 /* log after-image for logredo(): */
1557 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1558
1559 if (jfs_dirtable_inline(tlck->ip)) {
1560 /*
1561 * The table has been truncated, we've must have deleted
1562 * the last entry, so don't bother logging this
1563 */
1564 mp->lid = 0;
Dave Kleikamp7fab4792005-05-02 12:25:02 -06001565 grab_metapage(mp);
1566 metapage_homeok(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 discard_metapage(mp);
1568 tlck->mp = NULL;
1569 return 0;
1570 }
1571
1572 PXDaddress(pxd, mp->index);
1573 PXDlength(pxd, mp->logical_size >> tblk->sb->s_blocksize_bits);
1574
1575 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1576
1577 /* mark page as homeward bound */
1578 tlck->flag |= tlckWRITEPAGE;
1579
1580 return 0;
1581}
1582
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583/*
1584 * dtLog()
1585 *
1586 * function: log dtree tlock and format maplock to update bmap;
1587 */
1588static void dtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1589 struct tlock * tlck)
1590{
1591 struct metapage *mp;
1592 struct pxd_lock *pxdlock;
1593 pxd_t *pxd;
1594
1595 mp = tlck->mp;
1596
1597 /* initialize as REDOPAGE/NOREDOPAGE record format */
1598 lrd->log.redopage.type = cpu_to_le16(LOG_DTREE);
1599 lrd->log.redopage.l2linesize = cpu_to_le16(L2DTSLOTSIZE);
1600
1601 pxd = &lrd->log.redopage.pxd;
1602
1603 if (tlck->type & tlckBTROOT)
1604 lrd->log.redopage.type |= cpu_to_le16(LOG_BTROOT);
1605
1606 /*
1607 * page extension via relocation: entry insertion;
1608 * page extension in-place: entry insertion;
1609 * new right page from page split, reinitialized in-line
1610 * root from root page split: entry insertion;
1611 */
1612 if (tlck->type & (tlckNEW | tlckEXTEND)) {
1613 /* log after-image of the new page for logredo():
1614 * mark log (LOG_NEW) for logredo() to initialize
1615 * freelist and update bmap for alloc of the new page;
1616 */
1617 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1618 if (tlck->type & tlckEXTEND)
1619 lrd->log.redopage.type |= cpu_to_le16(LOG_EXTEND);
1620 else
1621 lrd->log.redopage.type |= cpu_to_le16(LOG_NEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622 PXDaddress(pxd, mp->index);
1623 PXDlength(pxd,
1624 mp->logical_size >> tblk->sb->s_blocksize_bits);
1625 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1626
1627 /* format a maplock for txUpdateMap() to update bPMAP for
1628 * alloc of the new page;
1629 */
1630 if (tlck->type & tlckBTROOT)
1631 return;
1632 tlck->flag |= tlckUPDATEMAP;
1633 pxdlock = (struct pxd_lock *) & tlck->lock;
1634 pxdlock->flag = mlckALLOCPXD;
1635 pxdlock->pxd = *pxd;
1636
1637 pxdlock->index = 1;
1638
1639 /* mark page as homeward bound */
1640 tlck->flag |= tlckWRITEPAGE;
1641 return;
1642 }
1643
1644 /*
1645 * entry insertion/deletion,
1646 * sibling page link update (old right page before split);
1647 */
1648 if (tlck->type & (tlckENTRY | tlckRELINK)) {
1649 /* log after-image for logredo(): */
1650 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1651 PXDaddress(pxd, mp->index);
1652 PXDlength(pxd,
1653 mp->logical_size >> tblk->sb->s_blocksize_bits);
1654 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1655
1656 /* mark page as homeward bound */
1657 tlck->flag |= tlckWRITEPAGE;
1658 return;
1659 }
1660
1661 /*
1662 * page deletion: page has been invalidated
1663 * page relocation: source extent
1664 *
1665 * a maplock for free of the page has been formatted
1666 * at txLock() time);
1667 */
1668 if (tlck->type & (tlckFREE | tlckRELOCATE)) {
1669 /* log LOG_NOREDOPAGE of the deleted page for logredo()
1670 * to start NoRedoPage filter and to update bmap for free
1671 * of the deletd page
1672 */
1673 lrd->type = cpu_to_le16(LOG_NOREDOPAGE);
1674 pxdlock = (struct pxd_lock *) & tlck->lock;
1675 *pxd = pxdlock->pxd;
1676 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1677
1678 /* a maplock for txUpdateMap() for free of the page
1679 * has been formatted at txLock() time;
1680 */
1681 tlck->flag |= tlckUPDATEMAP;
1682 }
1683 return;
1684}
1685
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686/*
1687 * xtLog()
1688 *
1689 * function: log xtree tlock and format maplock to update bmap;
1690 */
1691static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1692 struct tlock * tlck)
1693{
1694 struct inode *ip;
1695 struct metapage *mp;
1696 xtpage_t *p;
1697 struct xtlock *xtlck;
1698 struct maplock *maplock;
1699 struct xdlistlock *xadlock;
1700 struct pxd_lock *pxdlock;
Dave Kleikamp66f31312005-05-02 12:24:46 -06001701 pxd_t *page_pxd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 int next, lwm, hwm;
1703
1704 ip = tlck->ip;
1705 mp = tlck->mp;
1706
1707 /* initialize as REDOPAGE/NOREDOPAGE record format */
1708 lrd->log.redopage.type = cpu_to_le16(LOG_XTREE);
1709 lrd->log.redopage.l2linesize = cpu_to_le16(L2XTSLOTSIZE);
1710
Dave Kleikamp66f31312005-05-02 12:24:46 -06001711 page_pxd = &lrd->log.redopage.pxd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712
1713 if (tlck->type & tlckBTROOT) {
1714 lrd->log.redopage.type |= cpu_to_le16(LOG_BTROOT);
1715 p = &JFS_IP(ip)->i_xtroot;
1716 if (S_ISDIR(ip->i_mode))
1717 lrd->log.redopage.type |=
1718 cpu_to_le16(LOG_DIR_XTREE);
1719 } else
1720 p = (xtpage_t *) mp->data;
1721 next = le16_to_cpu(p->header.nextindex);
1722
1723 xtlck = (struct xtlock *) & tlck->lock;
1724
1725 maplock = (struct maplock *) & tlck->lock;
1726 xadlock = (struct xdlistlock *) maplock;
1727
1728 /*
1729 * entry insertion/extension;
1730 * sibling page link update (old right page before split);
1731 */
1732 if (tlck->type & (tlckNEW | tlckGROW | tlckRELINK)) {
1733 /* log after-image for logredo():
1734 * logredo() will update bmap for alloc of new/extended
1735 * extents (XAD_NEW|XAD_EXTEND) of XAD[lwm:next) from
1736 * after-image of XADlist;
1737 * logredo() resets (XAD_NEW|XAD_EXTEND) flag when
1738 * applying the after-image to the meta-data page.
1739 */
1740 lrd->type = cpu_to_le16(LOG_REDOPAGE);
Dave Kleikamp66f31312005-05-02 12:24:46 -06001741 PXDaddress(page_pxd, mp->index);
1742 PXDlength(page_pxd,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 mp->logical_size >> tblk->sb->s_blocksize_bits);
1744 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1745
1746 /* format a maplock for txUpdateMap() to update bPMAP
1747 * for alloc of new/extended extents of XAD[lwm:next)
1748 * from the page itself;
1749 * txUpdateMap() resets (XAD_NEW|XAD_EXTEND) flag.
1750 */
1751 lwm = xtlck->lwm.offset;
1752 if (lwm == 0)
1753 lwm = XTPAGEMAXSLOT;
1754
1755 if (lwm == next)
1756 goto out;
1757 if (lwm > next) {
1758 jfs_err("xtLog: lwm > next\n");
1759 goto out;
1760 }
1761 tlck->flag |= tlckUPDATEMAP;
1762 xadlock->flag = mlckALLOCXADLIST;
1763 xadlock->count = next - lwm;
Dave Kleikamp66f31312005-05-02 12:24:46 -06001764 if ((xadlock->count <= 4) && (tblk->xflag & COMMIT_LAZY)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 int i;
Dave Kleikamp66f31312005-05-02 12:24:46 -06001766 pxd_t *pxd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767 /*
1768 * Lazy commit may allow xtree to be modified before
1769 * txUpdateMap runs. Copy xad into linelock to
1770 * preserve correct data.
Dave Kleikamp66f31312005-05-02 12:24:46 -06001771 *
1772 * We can fit twice as may pxd's as xads in the lock
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773 */
Dave Kleikamp66f31312005-05-02 12:24:46 -06001774 xadlock->flag = mlckALLOCPXDLIST;
1775 pxd = xadlock->xdlist = &xtlck->pxdlock;
1776 for (i = 0; i < xadlock->count; i++) {
1777 PXDaddress(pxd, addressXAD(&p->xad[lwm + i]));
1778 PXDlength(pxd, lengthXAD(&p->xad[lwm + i]));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 p->xad[lwm + i].flag &=
1780 ~(XAD_NEW | XAD_EXTENDED);
Dave Kleikamp66f31312005-05-02 12:24:46 -06001781 pxd++;
1782 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 } else {
1784 /*
1785 * xdlist will point to into inode's xtree, ensure
1786 * that transaction is not committed lazily.
1787 */
Dave Kleikamp66f31312005-05-02 12:24:46 -06001788 xadlock->flag = mlckALLOCXADLIST;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789 xadlock->xdlist = &p->xad[lwm];
1790 tblk->xflag &= ~COMMIT_LAZY;
1791 }
1792 jfs_info("xtLog: alloc ip:0x%p mp:0x%p tlck:0x%p lwm:%d "
1793 "count:%d", tlck->ip, mp, tlck, lwm, xadlock->count);
1794
1795 maplock->index = 1;
1796
1797 out:
1798 /* mark page as homeward bound */
1799 tlck->flag |= tlckWRITEPAGE;
1800
1801 return;
1802 }
1803
1804 /*
1805 * page deletion: file deletion/truncation (ref. xtTruncate())
1806 *
1807 * (page will be invalidated after log is written and bmap
1808 * is updated from the page);
1809 */
1810 if (tlck->type & tlckFREE) {
1811 /* LOG_NOREDOPAGE log for NoRedoPage filter:
1812 * if page free from file delete, NoRedoFile filter from
1813 * inode image of zero link count will subsume NoRedoPage
1814 * filters for each page;
1815 * if page free from file truncattion, write NoRedoPage
1816 * filter;
1817 *
1818 * upadte of block allocation map for the page itself:
1819 * if page free from deletion and truncation, LOG_UPDATEMAP
1820 * log for the page itself is generated from processing
1821 * its parent page xad entries;
1822 */
1823 /* if page free from file truncation, log LOG_NOREDOPAGE
1824 * of the deleted page for logredo() to start NoRedoPage
1825 * filter for the page;
1826 */
1827 if (tblk->xflag & COMMIT_TRUNCATE) {
1828 /* write NOREDOPAGE for the page */
1829 lrd->type = cpu_to_le16(LOG_NOREDOPAGE);
Dave Kleikamp66f31312005-05-02 12:24:46 -06001830 PXDaddress(page_pxd, mp->index);
1831 PXDlength(page_pxd,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 mp->logical_size >> tblk->sb->
1833 s_blocksize_bits);
1834 lrd->backchain =
1835 cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1836
1837 if (tlck->type & tlckBTROOT) {
1838 /* Empty xtree must be logged */
1839 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1840 lrd->backchain =
1841 cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1842 }
1843 }
1844
1845 /* init LOG_UPDATEMAP of the freed extents
1846 * XAD[XTENTRYSTART:hwm) from the deleted page itself
1847 * for logredo() to update bmap;
1848 */
1849 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1850 lrd->log.updatemap.type = cpu_to_le16(LOG_FREEXADLIST);
1851 xtlck = (struct xtlock *) & tlck->lock;
1852 hwm = xtlck->hwm.offset;
1853 lrd->log.updatemap.nxd =
1854 cpu_to_le16(hwm - XTENTRYSTART + 1);
1855 /* reformat linelock for lmLog() */
1856 xtlck->header.offset = XTENTRYSTART;
1857 xtlck->header.length = hwm - XTENTRYSTART + 1;
1858 xtlck->index = 1;
1859 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1860
1861 /* format a maplock for txUpdateMap() to update bmap
1862 * to free extents of XAD[XTENTRYSTART:hwm) from the
1863 * deleted page itself;
1864 */
1865 tlck->flag |= tlckUPDATEMAP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866 xadlock->count = hwm - XTENTRYSTART + 1;
Dave Kleikamp66f31312005-05-02 12:24:46 -06001867 if ((xadlock->count <= 4) && (tblk->xflag & COMMIT_LAZY)) {
1868 int i;
1869 pxd_t *pxd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870 /*
1871 * Lazy commit may allow xtree to be modified before
1872 * txUpdateMap runs. Copy xad into linelock to
1873 * preserve correct data.
Dave Kleikamp66f31312005-05-02 12:24:46 -06001874 *
1875 * We can fit twice as may pxd's as xads in the lock
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876 */
Dave Kleikamp66f31312005-05-02 12:24:46 -06001877 xadlock->flag = mlckFREEPXDLIST;
1878 pxd = xadlock->xdlist = &xtlck->pxdlock;
1879 for (i = 0; i < xadlock->count; i++) {
1880 PXDaddress(pxd,
1881 addressXAD(&p->xad[XTENTRYSTART + i]));
1882 PXDlength(pxd,
1883 lengthXAD(&p->xad[XTENTRYSTART + i]));
1884 pxd++;
1885 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 } else {
1887 /*
1888 * xdlist will point to into inode's xtree, ensure
1889 * that transaction is not committed lazily.
1890 */
Dave Kleikamp66f31312005-05-02 12:24:46 -06001891 xadlock->flag = mlckFREEXADLIST;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 xadlock->xdlist = &p->xad[XTENTRYSTART];
1893 tblk->xflag &= ~COMMIT_LAZY;
1894 }
1895 jfs_info("xtLog: free ip:0x%p mp:0x%p count:%d lwm:2",
1896 tlck->ip, mp, xadlock->count);
1897
1898 maplock->index = 1;
1899
1900 /* mark page as invalid */
1901 if (((tblk->xflag & COMMIT_PWMAP) || S_ISDIR(ip->i_mode))
1902 && !(tlck->type & tlckBTROOT))
1903 tlck->flag |= tlckFREEPAGE;
1904 /*
1905 else (tblk->xflag & COMMIT_PMAP)
1906 ? release the page;
1907 */
1908 return;
1909 }
1910
1911 /*
1912 * page/entry truncation: file truncation (ref. xtTruncate())
1913 *
1914 * |----------+------+------+---------------|
1915 * | | |
1916 * | | hwm - hwm before truncation
1917 * | next - truncation point
1918 * lwm - lwm before truncation
1919 * header ?
1920 */
1921 if (tlck->type & tlckTRUNCATE) {
Dave Kleikamp66f31312005-05-02 12:24:46 -06001922 pxd_t pxd; /* truncated extent of xad */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923 int twm;
1924
1925 /*
1926 * For truncation the entire linelock may be used, so it would
1927 * be difficult to store xad list in linelock itself.
1928 * Therefore, we'll just force transaction to be committed
1929 * synchronously, so that xtree pages won't be changed before
1930 * txUpdateMap runs.
1931 */
1932 tblk->xflag &= ~COMMIT_LAZY;
1933 lwm = xtlck->lwm.offset;
1934 if (lwm == 0)
1935 lwm = XTPAGEMAXSLOT;
1936 hwm = xtlck->hwm.offset;
1937 twm = xtlck->twm.offset;
1938
1939 /*
1940 * write log records
1941 */
1942 /* log after-image for logredo():
1943 *
1944 * logredo() will update bmap for alloc of new/extended
1945 * extents (XAD_NEW|XAD_EXTEND) of XAD[lwm:next) from
1946 * after-image of XADlist;
1947 * logredo() resets (XAD_NEW|XAD_EXTEND) flag when
1948 * applying the after-image to the meta-data page.
1949 */
1950 lrd->type = cpu_to_le16(LOG_REDOPAGE);
Dave Kleikamp66f31312005-05-02 12:24:46 -06001951 PXDaddress(page_pxd, mp->index);
1952 PXDlength(page_pxd,
1953 mp->logical_size >> tblk->sb->s_blocksize_bits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1955
1956 /*
1957 * truncate entry XAD[twm == next - 1]:
1958 */
1959 if (twm == next - 1) {
1960 /* init LOG_UPDATEMAP for logredo() to update bmap for
1961 * free of truncated delta extent of the truncated
1962 * entry XAD[next - 1]:
1963 * (xtlck->pxdlock = truncated delta extent);
1964 */
1965 pxdlock = (struct pxd_lock *) & xtlck->pxdlock;
1966 /* assert(pxdlock->type & tlckTRUNCATE); */
1967 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1968 lrd->log.updatemap.type = cpu_to_le16(LOG_FREEPXD);
1969 lrd->log.updatemap.nxd = cpu_to_le16(1);
1970 lrd->log.updatemap.pxd = pxdlock->pxd;
Dave Kleikamp66f31312005-05-02 12:24:46 -06001971 pxd = pxdlock->pxd; /* save to format maplock */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 lrd->backchain =
1973 cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1974 }
1975
1976 /*
1977 * free entries XAD[next:hwm]:
1978 */
1979 if (hwm >= next) {
1980 /* init LOG_UPDATEMAP of the freed extents
1981 * XAD[next:hwm] from the deleted page itself
1982 * for logredo() to update bmap;
1983 */
1984 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1985 lrd->log.updatemap.type =
1986 cpu_to_le16(LOG_FREEXADLIST);
1987 xtlck = (struct xtlock *) & tlck->lock;
1988 hwm = xtlck->hwm.offset;
1989 lrd->log.updatemap.nxd =
1990 cpu_to_le16(hwm - next + 1);
1991 /* reformat linelock for lmLog() */
1992 xtlck->header.offset = next;
1993 xtlck->header.length = hwm - next + 1;
1994 xtlck->index = 1;
1995 lrd->backchain =
1996 cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1997 }
1998
1999 /*
2000 * format maplock(s) for txUpdateMap() to update bmap
2001 */
2002 maplock->index = 0;
2003
2004 /*
2005 * allocate entries XAD[lwm:next):
2006 */
2007 if (lwm < next) {
2008 /* format a maplock for txUpdateMap() to update bPMAP
2009 * for alloc of new/extended extents of XAD[lwm:next)
2010 * from the page itself;
2011 * txUpdateMap() resets (XAD_NEW|XAD_EXTEND) flag.
2012 */
2013 tlck->flag |= tlckUPDATEMAP;
2014 xadlock->flag = mlckALLOCXADLIST;
2015 xadlock->count = next - lwm;
2016 xadlock->xdlist = &p->xad[lwm];
2017
2018 jfs_info("xtLog: alloc ip:0x%p mp:0x%p count:%d "
2019 "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 struct pxd_lock *pxdlock;
2030
2031 /* format a maplock for txUpdateMap() to update bmap
2032 * to free truncated delta extent of the truncated
2033 * entry XAD[next - 1];
2034 * (xtlck->pxdlock = truncated delta extent);
2035 */
2036 tlck->flag |= tlckUPDATEMAP;
2037 pxdlock = (struct pxd_lock *) xadlock;
2038 pxdlock->flag = mlckFREEPXD;
2039 pxdlock->count = 1;
Dave Kleikamp66f31312005-05-02 12:24:46 -06002040 pxdlock->pxd = pxd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041
2042 jfs_info("xtLog: truncate ip:0x%p mp:0x%p count:%d "
2043 "hwm:%d", ip, mp, pxdlock->count, hwm);
2044 maplock->index++;
2045 xadlock++;
2046 }
2047
2048 /*
2049 * free entries XAD[next:hwm]:
2050 */
2051 if (hwm >= next) {
2052 /* format a maplock for txUpdateMap() to update bmap
2053 * to free extents of XAD[next:hwm] from thedeleted
2054 * page itself;
2055 */
2056 tlck->flag |= tlckUPDATEMAP;
2057 xadlock->flag = mlckFREEXADLIST;
2058 xadlock->count = hwm - next + 1;
2059 xadlock->xdlist = &p->xad[next];
2060
2061 jfs_info("xtLog: free ip:0x%p mp:0x%p count:%d "
2062 "next:%d hwm:%d",
2063 tlck->ip, mp, xadlock->count, next, hwm);
2064 maplock->index++;
2065 }
2066
2067 /* mark page as homeward bound */
2068 tlck->flag |= tlckWRITEPAGE;
2069 }
2070 return;
2071}
2072
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073/*
2074 * mapLog()
2075 *
2076 * function: log from maplock of freed data extents;
2077 */
Dave Kleikamp6cb12692005-09-15 23:25:41 -05002078static void mapLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
2079 struct tlock * tlck)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080{
2081 struct pxd_lock *pxdlock;
2082 int i, nlock;
2083 pxd_t *pxd;
2084
2085 /*
2086 * page relocation: free the source page extent
2087 *
2088 * a maplock for txUpdateMap() for free of the page
2089 * has been formatted at txLock() time saving the src
2090 * relocated page address;
2091 */
2092 if (tlck->type & tlckRELOCATE) {
2093 /* log LOG_NOREDOPAGE of the old relocated page
2094 * for logredo() to start NoRedoPage filter;
2095 */
2096 lrd->type = cpu_to_le16(LOG_NOREDOPAGE);
2097 pxdlock = (struct pxd_lock *) & tlck->lock;
2098 pxd = &lrd->log.redopage.pxd;
2099 *pxd = pxdlock->pxd;
2100 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
2101
2102 /* (N.B. currently, logredo() does NOT update bmap
2103 * for free of the page itself for (LOG_XTREE|LOG_NOREDOPAGE);
2104 * if page free from relocation, LOG_UPDATEMAP log is
2105 * specifically generated now for logredo()
2106 * to update bmap for free of src relocated page;
2107 * (new flag LOG_RELOCATE may be introduced which will
2108 * inform logredo() to start NORedoPage filter and also
2109 * update block allocation map at the same time, thus
2110 * avoiding an extra log write);
2111 */
2112 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
2113 lrd->log.updatemap.type = cpu_to_le16(LOG_FREEPXD);
2114 lrd->log.updatemap.nxd = cpu_to_le16(1);
2115 lrd->log.updatemap.pxd = pxdlock->pxd;
2116 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
2117
2118 /* a maplock for txUpdateMap() for free of the page
2119 * has been formatted at txLock() time;
2120 */
2121 tlck->flag |= tlckUPDATEMAP;
2122 return;
2123 }
2124 /*
2125
2126 * Otherwise it's not a relocate request
2127 *
2128 */
2129 else {
2130 /* log LOG_UPDATEMAP for logredo() to update bmap for
2131 * free of truncated/relocated delta extent of the data;
2132 * e.g.: external EA extent, relocated/truncated extent
2133 * from xtTailgate();
2134 */
2135 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
2136 pxdlock = (struct pxd_lock *) & tlck->lock;
2137 nlock = pxdlock->index;
2138 for (i = 0; i < nlock; i++, pxdlock++) {
2139 if (pxdlock->flag & mlckALLOCPXD)
2140 lrd->log.updatemap.type =
2141 cpu_to_le16(LOG_ALLOCPXD);
2142 else
2143 lrd->log.updatemap.type =
2144 cpu_to_le16(LOG_FREEPXD);
2145 lrd->log.updatemap.nxd = cpu_to_le16(1);
2146 lrd->log.updatemap.pxd = pxdlock->pxd;
2147 lrd->backchain =
2148 cpu_to_le32(lmLog(log, tblk, lrd, NULL));
2149 jfs_info("mapLog: xaddr:0x%lx xlen:0x%x",
2150 (ulong) addressPXD(&pxdlock->pxd),
2151 lengthPXD(&pxdlock->pxd));
2152 }
2153
2154 /* update bmap */
2155 tlck->flag |= tlckUPDATEMAP;
2156 }
2157}
2158
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159/*
2160 * txEA()
2161 *
2162 * function: acquire maplock for EA/ACL extents or
2163 * set COMMIT_INLINE flag;
2164 */
2165void txEA(tid_t tid, struct inode *ip, dxd_t * oldea, dxd_t * newea)
2166{
2167 struct tlock *tlck = NULL;
2168 struct pxd_lock *maplock = NULL, *pxdlock = NULL;
2169
2170 /*
2171 * format maplock for alloc of new EA extent
2172 */
2173 if (newea) {
2174 /* Since the newea could be a completely zeroed entry we need to
2175 * check for the two flags which indicate we should actually
2176 * commit new EA data
2177 */
2178 if (newea->flag & DXD_EXTENT) {
2179 tlck = txMaplock(tid, ip, tlckMAP);
2180 maplock = (struct pxd_lock *) & tlck->lock;
2181 pxdlock = (struct pxd_lock *) maplock;
2182 pxdlock->flag = mlckALLOCPXD;
2183 PXDaddress(&pxdlock->pxd, addressDXD(newea));
2184 PXDlength(&pxdlock->pxd, lengthDXD(newea));
2185 pxdlock++;
2186 maplock->index = 1;
2187 } else if (newea->flag & DXD_INLINE) {
2188 tlck = NULL;
2189
2190 set_cflag(COMMIT_Inlineea, ip);
2191 }
2192 }
2193
2194 /*
2195 * format maplock for free of old EA extent
2196 */
2197 if (!test_cflag(COMMIT_Nolink, ip) && oldea->flag & DXD_EXTENT) {
2198 if (tlck == NULL) {
2199 tlck = txMaplock(tid, ip, tlckMAP);
2200 maplock = (struct pxd_lock *) & tlck->lock;
2201 pxdlock = (struct pxd_lock *) maplock;
2202 maplock->index = 0;
2203 }
2204 pxdlock->flag = mlckFREEPXD;
2205 PXDaddress(&pxdlock->pxd, addressDXD(oldea));
2206 PXDlength(&pxdlock->pxd, lengthDXD(oldea));
2207 maplock->index++;
2208 }
2209}
2210
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211/*
2212 * txForce()
2213 *
2214 * function: synchronously write pages locked by transaction
2215 * after txLog() but before txUpdateMap();
2216 */
Dave Kleikamp6cb12692005-09-15 23:25:41 -05002217static void txForce(struct tblock * tblk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218{
2219 struct tlock *tlck;
2220 lid_t lid, next;
2221 struct metapage *mp;
2222
2223 /*
2224 * reverse the order of transaction tlocks in
2225 * careful update order of address index pages
2226 * (right to left, bottom up)
2227 */
2228 tlck = lid_to_tlock(tblk->next);
2229 lid = tlck->next;
2230 tlck->next = 0;
2231 while (lid) {
2232 tlck = lid_to_tlock(lid);
2233 next = tlck->next;
2234 tlck->next = tblk->next;
2235 tblk->next = lid;
2236 lid = next;
2237 }
2238
2239 /*
2240 * synchronously write the page, and
2241 * hold the page for txUpdateMap();
2242 */
2243 for (lid = tblk->next; lid; lid = next) {
2244 tlck = lid_to_tlock(lid);
2245 next = tlck->next;
2246
2247 if ((mp = tlck->mp) != NULL &&
2248 (tlck->type & tlckBTROOT) == 0) {
2249 assert(mp->xflag & COMMIT_PAGE);
2250
2251 if (tlck->flag & tlckWRITEPAGE) {
2252 tlck->flag &= ~tlckWRITEPAGE;
2253
2254 /* do not release page to freelist */
Dave Kleikamp7fab4792005-05-02 12:25:02 -06002255 force_metapage(mp);
2256#if 0
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257 /*
2258 * The "right" thing to do here is to
2259 * synchronously write the metadata.
2260 * With the current implementation this
2261 * is hard since write_metapage requires
2262 * us to kunmap & remap the page. If we
2263 * have tlocks pointing into the metadata
2264 * pages, we don't want to do this. I think
2265 * we can get by with synchronously writing
2266 * the pages when they are released.
2267 */
Dave Kleikamp7fab4792005-05-02 12:25:02 -06002268 assert(mp->nohomeok);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269 set_bit(META_dirty, &mp->flag);
2270 set_bit(META_sync, &mp->flag);
Dave Kleikamp7fab4792005-05-02 12:25:02 -06002271#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 }
2273 }
2274 }
2275}
2276
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277/*
2278 * txUpdateMap()
2279 *
2280 * function: update persistent allocation map (and working map
2281 * if appropriate);
2282 *
2283 * parameter:
2284 */
2285static void txUpdateMap(struct tblock * tblk)
2286{
2287 struct inode *ip;
2288 struct inode *ipimap;
2289 lid_t lid;
2290 struct tlock *tlck;
2291 struct maplock *maplock;
2292 struct pxd_lock pxdlock;
2293 int maptype;
2294 int k, nlock;
2295 struct metapage *mp = NULL;
2296
2297 ipimap = JFS_SBI(tblk->sb)->ipimap;
2298
2299 maptype = (tblk->xflag & COMMIT_PMAP) ? COMMIT_PMAP : COMMIT_PWMAP;
2300
2301
2302 /*
2303 * update block allocation map
2304 *
2305 * update allocation state in pmap (and wmap) and
2306 * update lsn of the pmap page;
2307 */
2308 /*
2309 * scan each tlock/page of transaction for block allocation/free:
2310 *
2311 * for each tlock/page of transaction, update map.
2312 * ? are there tlock for pmap and pwmap at the same time ?
2313 */
2314 for (lid = tblk->next; lid; lid = tlck->next) {
2315 tlck = lid_to_tlock(lid);
2316
2317 if ((tlck->flag & tlckUPDATEMAP) == 0)
2318 continue;
2319
2320 if (tlck->flag & tlckFREEPAGE) {
2321 /*
2322 * Another thread may attempt to reuse freed space
2323 * immediately, so we want to get rid of the metapage
2324 * before anyone else has a chance to get it.
2325 * Lock metapage, update maps, then invalidate
2326 * the metapage.
2327 */
2328 mp = tlck->mp;
2329 ASSERT(mp->xflag & COMMIT_PAGE);
Dave Kleikamp7fab4792005-05-02 12:25:02 -06002330 grab_metapage(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331 }
2332
2333 /*
2334 * extent list:
2335 * . in-line PXD list:
2336 * . out-of-line XAD list:
2337 */
2338 maplock = (struct maplock *) & tlck->lock;
2339 nlock = maplock->index;
2340
2341 for (k = 0; k < nlock; k++, maplock++) {
2342 /*
2343 * allocate blocks in persistent map:
2344 *
2345 * blocks have been allocated from wmap at alloc time;
2346 */
2347 if (maplock->flag & mlckALLOC) {
2348 txAllocPMap(ipimap, maplock, tblk);
2349 }
2350 /*
2351 * free blocks in persistent and working map:
2352 * blocks will be freed in pmap and then in wmap;
2353 *
2354 * ? tblock specifies the PMAP/PWMAP based upon
2355 * transaction
2356 *
2357 * free blocks in persistent map:
2358 * blocks will be freed from wmap at last reference
2359 * release of the object for regular files;
2360 *
2361 * Alway free blocks from both persistent & working
2362 * maps for directories
2363 */
2364 else { /* (maplock->flag & mlckFREE) */
2365
Dave Kleikamp438282d2005-09-20 14:58:11 -05002366 if (tlck->flag & tlckDIRECTORY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367 txFreeMap(ipimap, maplock,
2368 tblk, COMMIT_PWMAP);
2369 else
2370 txFreeMap(ipimap, maplock,
2371 tblk, maptype);
2372 }
2373 }
2374 if (tlck->flag & tlckFREEPAGE) {
2375 if (!(tblk->flag & tblkGC_LAZY)) {
2376 /* This is equivalent to txRelease */
2377 ASSERT(mp->lid == lid);
2378 tlck->mp->lid = 0;
2379 }
Dave Kleikamp7fab4792005-05-02 12:25:02 -06002380 assert(mp->nohomeok == 1);
2381 metapage_homeok(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 discard_metapage(mp);
2383 tlck->mp = NULL;
2384 }
2385 }
2386 /*
2387 * update inode allocation map
2388 *
2389 * update allocation state in pmap and
2390 * update lsn of the pmap page;
2391 * update in-memory inode flag/state
2392 *
2393 * unlock mapper/write lock
2394 */
2395 if (tblk->xflag & COMMIT_CREATE) {
2396 diUpdatePMap(ipimap, tblk->ino, FALSE, tblk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002397 /* update persistent block allocation map
2398 * for the allocation of inode extent;
2399 */
2400 pxdlock.flag = mlckALLOCPXD;
2401 pxdlock.pxd = tblk->u.ixpxd;
2402 pxdlock.index = 1;
2403 txAllocPMap(ipimap, (struct maplock *) & pxdlock, tblk);
2404 } else if (tblk->xflag & COMMIT_DELETE) {
2405 ip = tblk->u.ip;
2406 diUpdatePMap(ipimap, ip->i_ino, TRUE, tblk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407 iput(ip);
2408 }
2409}
2410
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411/*
2412 * txAllocPMap()
2413 *
2414 * function: allocate from persistent map;
2415 *
2416 * parameter:
2417 * ipbmap -
2418 * malock -
2419 * xad list:
2420 * pxd:
2421 *
2422 * maptype -
2423 * allocate from persistent map;
2424 * free from persistent map;
2425 * (e.g., tmp file - free from working map at releae
2426 * of last reference);
2427 * free from persistent and working map;
2428 *
2429 * lsn - log sequence number;
2430 */
2431static void txAllocPMap(struct inode *ip, struct maplock * maplock,
2432 struct tblock * tblk)
2433{
2434 struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
2435 struct xdlistlock *xadlistlock;
2436 xad_t *xad;
2437 s64 xaddr;
2438 int xlen;
2439 struct pxd_lock *pxdlock;
2440 struct xdlistlock *pxdlistlock;
2441 pxd_t *pxd;
2442 int n;
2443
2444 /*
2445 * allocate from persistent map;
2446 */
2447 if (maplock->flag & mlckALLOCXADLIST) {
2448 xadlistlock = (struct xdlistlock *) maplock;
2449 xad = xadlistlock->xdlist;
2450 for (n = 0; n < xadlistlock->count; n++, xad++) {
2451 if (xad->flag & (XAD_NEW | XAD_EXTENDED)) {
2452 xaddr = addressXAD(xad);
2453 xlen = lengthXAD(xad);
2454 dbUpdatePMap(ipbmap, FALSE, xaddr,
2455 (s64) xlen, tblk);
2456 xad->flag &= ~(XAD_NEW | XAD_EXTENDED);
2457 jfs_info("allocPMap: xaddr:0x%lx xlen:%d",
2458 (ulong) xaddr, xlen);
2459 }
2460 }
2461 } else if (maplock->flag & mlckALLOCPXD) {
2462 pxdlock = (struct pxd_lock *) maplock;
2463 xaddr = addressPXD(&pxdlock->pxd);
2464 xlen = lengthPXD(&pxdlock->pxd);
2465 dbUpdatePMap(ipbmap, FALSE, xaddr, (s64) xlen, tblk);
2466 jfs_info("allocPMap: xaddr:0x%lx xlen:%d", (ulong) xaddr, xlen);
2467 } else { /* (maplock->flag & mlckALLOCPXDLIST) */
2468
2469 pxdlistlock = (struct xdlistlock *) maplock;
2470 pxd = pxdlistlock->xdlist;
2471 for (n = 0; n < pxdlistlock->count; n++, pxd++) {
2472 xaddr = addressPXD(pxd);
2473 xlen = lengthPXD(pxd);
2474 dbUpdatePMap(ipbmap, FALSE, xaddr, (s64) xlen,
2475 tblk);
2476 jfs_info("allocPMap: xaddr:0x%lx xlen:%d",
2477 (ulong) xaddr, xlen);
2478 }
2479 }
2480}
2481
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482/*
2483 * txFreeMap()
2484 *
2485 * function: free from persistent and/or working map;
2486 *
2487 * todo: optimization
2488 */
2489void txFreeMap(struct inode *ip,
2490 struct maplock * maplock, struct tblock * tblk, int maptype)
2491{
2492 struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
2493 struct xdlistlock *xadlistlock;
2494 xad_t *xad;
2495 s64 xaddr;
2496 int xlen;
2497 struct pxd_lock *pxdlock;
2498 struct xdlistlock *pxdlistlock;
2499 pxd_t *pxd;
2500 int n;
2501
2502 jfs_info("txFreeMap: tblk:0x%p maplock:0x%p maptype:0x%x",
2503 tblk, maplock, maptype);
2504
2505 /*
2506 * free from persistent map;
2507 */
2508 if (maptype == COMMIT_PMAP || maptype == COMMIT_PWMAP) {
2509 if (maplock->flag & mlckFREEXADLIST) {
2510 xadlistlock = (struct xdlistlock *) maplock;
2511 xad = xadlistlock->xdlist;
2512 for (n = 0; n < xadlistlock->count; n++, xad++) {
2513 if (!(xad->flag & XAD_NEW)) {
2514 xaddr = addressXAD(xad);
2515 xlen = lengthXAD(xad);
2516 dbUpdatePMap(ipbmap, TRUE, xaddr,
2517 (s64) xlen, tblk);
2518 jfs_info("freePMap: xaddr:0x%lx "
2519 "xlen:%d",
2520 (ulong) xaddr, xlen);
2521 }
2522 }
2523 } else if (maplock->flag & mlckFREEPXD) {
2524 pxdlock = (struct pxd_lock *) maplock;
2525 xaddr = addressPXD(&pxdlock->pxd);
2526 xlen = lengthPXD(&pxdlock->pxd);
2527 dbUpdatePMap(ipbmap, TRUE, xaddr, (s64) xlen,
2528 tblk);
2529 jfs_info("freePMap: xaddr:0x%lx xlen:%d",
2530 (ulong) xaddr, xlen);
2531 } else { /* (maplock->flag & mlckALLOCPXDLIST) */
2532
2533 pxdlistlock = (struct xdlistlock *) maplock;
2534 pxd = pxdlistlock->xdlist;
2535 for (n = 0; n < pxdlistlock->count; n++, pxd++) {
2536 xaddr = addressPXD(pxd);
2537 xlen = lengthPXD(pxd);
2538 dbUpdatePMap(ipbmap, TRUE, xaddr,
2539 (s64) xlen, tblk);
2540 jfs_info("freePMap: xaddr:0x%lx xlen:%d",
2541 (ulong) xaddr, xlen);
2542 }
2543 }
2544 }
2545
2546 /*
2547 * free from working map;
2548 */
2549 if (maptype == COMMIT_PWMAP || maptype == COMMIT_WMAP) {
2550 if (maplock->flag & mlckFREEXADLIST) {
2551 xadlistlock = (struct xdlistlock *) maplock;
2552 xad = xadlistlock->xdlist;
2553 for (n = 0; n < xadlistlock->count; n++, xad++) {
2554 xaddr = addressXAD(xad);
2555 xlen = lengthXAD(xad);
2556 dbFree(ip, xaddr, (s64) xlen);
2557 xad->flag = 0;
2558 jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
2559 (ulong) xaddr, xlen);
2560 }
2561 } else if (maplock->flag & mlckFREEPXD) {
2562 pxdlock = (struct pxd_lock *) maplock;
2563 xaddr = addressPXD(&pxdlock->pxd);
2564 xlen = lengthPXD(&pxdlock->pxd);
2565 dbFree(ip, xaddr, (s64) xlen);
2566 jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
2567 (ulong) xaddr, xlen);
2568 } else { /* (maplock->flag & mlckFREEPXDLIST) */
2569
2570 pxdlistlock = (struct xdlistlock *) maplock;
2571 pxd = pxdlistlock->xdlist;
2572 for (n = 0; n < pxdlistlock->count; n++, pxd++) {
2573 xaddr = addressPXD(pxd);
2574 xlen = lengthPXD(pxd);
2575 dbFree(ip, xaddr, (s64) xlen);
2576 jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
2577 (ulong) xaddr, xlen);
2578 }
2579 }
2580 }
2581}
2582
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583/*
2584 * txFreelock()
2585 *
2586 * function: remove tlock from inode anonymous locklist
2587 */
2588void txFreelock(struct inode *ip)
2589{
2590 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
2591 struct tlock *xtlck, *tlck;
2592 lid_t xlid = 0, lid;
2593
2594 if (!jfs_ip->atlhead)
2595 return;
2596
2597 TXN_LOCK();
2598 xtlck = (struct tlock *) &jfs_ip->atlhead;
2599
2600 while ((lid = xtlck->next) != 0) {
2601 tlck = lid_to_tlock(lid);
2602 if (tlck->flag & tlckFREELOCK) {
2603 xtlck->next = tlck->next;
2604 txLockFree(lid);
2605 } else {
2606 xtlck = tlck;
2607 xlid = lid;
2608 }
2609 }
2610
2611 if (jfs_ip->atlhead)
2612 jfs_ip->atltail = xlid;
2613 else {
2614 jfs_ip->atltail = 0;
2615 /*
2616 * If inode was on anon_list, remove it
2617 */
2618 list_del_init(&jfs_ip->anon_inode_list);
2619 }
2620 TXN_UNLOCK();
2621}
2622
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623/*
2624 * txAbort()
2625 *
2626 * function: abort tx before commit;
2627 *
2628 * frees line-locks and segment locks for all
2629 * segments in comdata structure.
2630 * Optionally sets state of file-system to FM_DIRTY in super-block.
2631 * log age of page-frames in memory for which caller has
2632 * are reset to 0 (to avoid logwarap).
2633 */
2634void txAbort(tid_t tid, int dirty)
2635{
2636 lid_t lid, next;
2637 struct metapage *mp;
2638 struct tblock *tblk = tid_to_tblock(tid);
2639 struct tlock *tlck;
2640
2641 /*
2642 * free tlocks of the transaction
2643 */
2644 for (lid = tblk->next; lid; lid = next) {
2645 tlck = lid_to_tlock(lid);
2646 next = tlck->next;
2647 mp = tlck->mp;
2648 JFS_IP(tlck->ip)->xtlid = 0;
2649
2650 if (mp) {
2651 mp->lid = 0;
2652
2653 /*
2654 * reset lsn of page to avoid logwarap:
2655 *
2656 * (page may have been previously committed by another
2657 * transaction(s) but has not been paged, i.e.,
2658 * it may be on logsync list even though it has not
2659 * been logged for the current tx.)
2660 */
2661 if (mp->xflag & COMMIT_PAGE && mp->lsn)
2662 LogSyncRelease(mp);
2663 }
2664 /* insert tlock at head of freelist */
2665 TXN_LOCK();
2666 txLockFree(lid);
2667 TXN_UNLOCK();
2668 }
2669
2670 /* caller will free the transaction block */
2671
2672 tblk->next = tblk->last = 0;
2673
2674 /*
2675 * mark filesystem dirty
2676 */
2677 if (dirty)
2678 jfs_error(tblk->sb, "txAbort");
2679
2680 return;
2681}
2682
2683/*
2684 * txLazyCommit(void)
2685 *
2686 * All transactions except those changing ipimap (COMMIT_FORCE) are
2687 * processed by this routine. This insures that the inode and block
2688 * allocation maps are updated in order. For synchronous transactions,
2689 * let the user thread finish processing after txUpdateMap() is called.
2690 */
2691static void txLazyCommit(struct tblock * tblk)
2692{
2693 struct jfs_log *log;
2694
2695 while (((tblk->flag & tblkGC_READY) == 0) &&
2696 ((tblk->flag & tblkGC_UNLOCKED) == 0)) {
2697 /* We must have gotten ahead of the user thread
2698 */
2699 jfs_info("jfs_lazycommit: tblk 0x%p not unlocked", tblk);
2700 yield();
2701 }
2702
2703 jfs_info("txLazyCommit: processing tblk 0x%p", tblk);
2704
2705 txUpdateMap(tblk);
2706
2707 log = (struct jfs_log *) JFS_SBI(tblk->sb)->log;
2708
2709 spin_lock_irq(&log->gclock); // LOGGC_LOCK
2710
2711 tblk->flag |= tblkGC_COMMITTED;
2712
2713 if (tblk->flag & tblkGC_READY)
2714 log->gcrtc--;
2715
2716 wake_up_all(&tblk->gcwait); // LOGGC_WAKEUP
2717
2718 /*
2719 * Can't release log->gclock until we've tested tblk->flag
2720 */
2721 if (tblk->flag & tblkGC_LAZY) {
2722 spin_unlock_irq(&log->gclock); // LOGGC_UNLOCK
2723 txUnlock(tblk);
2724 tblk->flag &= ~tblkGC_LAZY;
2725 txEnd(tblk - TxBlock); /* Convert back to tid */
2726 } else
2727 spin_unlock_irq(&log->gclock); // LOGGC_UNLOCK
2728
2729 jfs_info("txLazyCommit: done: tblk = 0x%p", tblk);
2730}
2731
2732/*
2733 * jfs_lazycommit(void)
2734 *
2735 * To be run as a kernel daemon. If lbmIODone is called in an interrupt
2736 * context, or where blocking is not wanted, this routine will process
2737 * committed transactions from the unlock queue.
2738 */
2739int jfs_lazycommit(void *arg)
2740{
2741 int WorkDone;
2742 struct tblock *tblk;
2743 unsigned long flags;
2744 struct jfs_sb_info *sbi;
2745
Linus Torvalds1da177e2005-04-16 15:20:36 -07002746 do {
2747 LAZY_LOCK(flags);
2748 jfs_commit_thread_waking = 0; /* OK to wake another thread */
2749 while (!list_empty(&TxAnchor.unlock_queue)) {
2750 WorkDone = 0;
2751 list_for_each_entry(tblk, &TxAnchor.unlock_queue,
2752 cqueue) {
2753
2754 sbi = JFS_SBI(tblk->sb);
2755 /*
2756 * For each volume, the transactions must be
2757 * handled in order. If another commit thread
2758 * is handling a tblk for this superblock,
2759 * skip it
2760 */
2761 if (sbi->commit_state & IN_LAZYCOMMIT)
2762 continue;
2763
2764 sbi->commit_state |= IN_LAZYCOMMIT;
2765 WorkDone = 1;
2766
2767 /*
2768 * Remove transaction from queue
2769 */
2770 list_del(&tblk->cqueue);
2771
2772 LAZY_UNLOCK(flags);
2773 txLazyCommit(tblk);
2774 LAZY_LOCK(flags);
2775
2776 sbi->commit_state &= ~IN_LAZYCOMMIT;
2777 /*
2778 * Don't continue in the for loop. (We can't
2779 * anyway, it's unsafe!) We want to go back to
2780 * the beginning of the list.
2781 */
2782 break;
2783 }
2784
2785 /* If there was nothing to do, don't continue */
2786 if (!WorkDone)
2787 break;
2788 }
2789 /* In case a wakeup came while all threads were active */
2790 jfs_commit_thread_waking = 0;
2791
Christoph Lameter3e1d1d22005-06-24 23:13:50 -07002792 if (freezing(current)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002793 LAZY_UNLOCK(flags);
Christoph Lameter3e1d1d22005-06-24 23:13:50 -07002794 refrigerator();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002795 } else {
2796 DECLARE_WAITQUEUE(wq, current);
2797
2798 add_wait_queue(&jfs_commit_thread_wait, &wq);
2799 set_current_state(TASK_INTERRUPTIBLE);
2800 LAZY_UNLOCK(flags);
2801 schedule();
2802 current->state = TASK_RUNNING;
2803 remove_wait_queue(&jfs_commit_thread_wait, &wq);
2804 }
Christoph Hellwig91dbb4d2006-02-15 12:49:04 -06002805 } while (!kthread_should_stop());
Linus Torvalds1da177e2005-04-16 15:20:36 -07002806
2807 if (!list_empty(&TxAnchor.unlock_queue))
2808 jfs_err("jfs_lazycommit being killed w/pending transactions!");
2809 else
2810 jfs_info("jfs_lazycommit being killed\n");
Christoph Hellwig91dbb4d2006-02-15 12:49:04 -06002811 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812}
2813
2814void txLazyUnlock(struct tblock * tblk)
2815{
2816 unsigned long flags;
2817
2818 LAZY_LOCK(flags);
2819
2820 list_add_tail(&tblk->cqueue, &TxAnchor.unlock_queue);
2821 /*
2822 * Don't wake up a commit thread if there is already one servicing
2823 * this superblock, or if the last one we woke up hasn't started yet.
2824 */
2825 if (!(JFS_SBI(tblk->sb)->commit_state & IN_LAZYCOMMIT) &&
2826 !jfs_commit_thread_waking) {
2827 jfs_commit_thread_waking = 1;
2828 wake_up(&jfs_commit_thread_wait);
2829 }
2830 LAZY_UNLOCK(flags);
2831}
2832
2833static void LogSyncRelease(struct metapage * mp)
2834{
2835 struct jfs_log *log = mp->log;
2836
Dave Kleikamp7fab4792005-05-02 12:25:02 -06002837 assert(mp->nohomeok);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002838 assert(log);
Dave Kleikamp7fab4792005-05-02 12:25:02 -06002839 metapage_homeok(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002840}
2841
2842/*
2843 * txQuiesce
2844 *
2845 * Block all new transactions and push anonymous transactions to
2846 * completion
2847 *
2848 * This does almost the same thing as jfs_sync below. We don't
2849 * worry about deadlocking when jfs_tlocks_low is set, since we would
2850 * expect jfs_sync to get us out of that jam.
2851 */
2852void txQuiesce(struct super_block *sb)
2853{
2854 struct inode *ip;
2855 struct jfs_inode_info *jfs_ip;
2856 struct jfs_log *log = JFS_SBI(sb)->log;
2857 tid_t tid;
2858
2859 set_bit(log_QUIESCE, &log->flag);
2860
2861 TXN_LOCK();
2862restart:
2863 while (!list_empty(&TxAnchor.anon_list)) {
2864 jfs_ip = list_entry(TxAnchor.anon_list.next,
2865 struct jfs_inode_info,
2866 anon_inode_list);
2867 ip = &jfs_ip->vfs_inode;
2868
2869 /*
2870 * inode will be removed from anonymous list
2871 * when it is committed
2872 */
2873 TXN_UNLOCK();
2874 tid = txBegin(ip->i_sb, COMMIT_INODE | COMMIT_FORCE);
Ingo Molnar1de87442006-01-24 15:22:50 -06002875 mutex_lock(&jfs_ip->commit_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002876 txCommit(tid, 1, &ip, 0);
2877 txEnd(tid);
Ingo Molnar1de87442006-01-24 15:22:50 -06002878 mutex_unlock(&jfs_ip->commit_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002879 /*
2880 * Just to be safe. I don't know how
2881 * long we can run without blocking
2882 */
2883 cond_resched();
2884 TXN_LOCK();
2885 }
2886
2887 /*
2888 * If jfs_sync is running in parallel, there could be some inodes
2889 * on anon_list2. Let's check.
2890 */
2891 if (!list_empty(&TxAnchor.anon_list2)) {
2892 list_splice(&TxAnchor.anon_list2, &TxAnchor.anon_list);
2893 INIT_LIST_HEAD(&TxAnchor.anon_list2);
2894 goto restart;
2895 }
2896 TXN_UNLOCK();
2897
2898 /*
2899 * We may need to kick off the group commit
2900 */
2901 jfs_flush_journal(log, 0);
2902}
2903
2904/*
2905 * txResume()
2906 *
2907 * Allows transactions to start again following txQuiesce
2908 */
2909void txResume(struct super_block *sb)
2910{
2911 struct jfs_log *log = JFS_SBI(sb)->log;
2912
2913 clear_bit(log_QUIESCE, &log->flag);
2914 TXN_WAKEUP(&log->syncwait);
2915}
2916
2917/*
2918 * jfs_sync(void)
2919 *
2920 * To be run as a kernel daemon. This is awakened when tlocks run low.
2921 * We write any inodes that have anonymous tlocks so they will become
2922 * available.
2923 */
2924int jfs_sync(void *arg)
2925{
2926 struct inode *ip;
2927 struct jfs_inode_info *jfs_ip;
2928 int rc;
2929 tid_t tid;
2930
Linus Torvalds1da177e2005-04-16 15:20:36 -07002931 do {
2932 /*
2933 * write each inode on the anonymous inode list
2934 */
2935 TXN_LOCK();
2936 while (jfs_tlocks_low && !list_empty(&TxAnchor.anon_list)) {
2937 jfs_ip = list_entry(TxAnchor.anon_list.next,
2938 struct jfs_inode_info,
2939 anon_inode_list);
2940 ip = &jfs_ip->vfs_inode;
2941
2942 if (! igrab(ip)) {
2943 /*
2944 * Inode is being freed
2945 */
2946 list_del_init(&jfs_ip->anon_inode_list);
Ingo Molnar1de87442006-01-24 15:22:50 -06002947 } else if (! !mutex_trylock(&jfs_ip->commit_mutex)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002948 /*
2949 * inode will be removed from anonymous list
2950 * when it is committed
2951 */
2952 TXN_UNLOCK();
2953 tid = txBegin(ip->i_sb, COMMIT_INODE);
2954 rc = txCommit(tid, 1, &ip, 0);
2955 txEnd(tid);
Ingo Molnar1de87442006-01-24 15:22:50 -06002956 mutex_unlock(&jfs_ip->commit_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002957
2958 iput(ip);
2959 /*
2960 * Just to be safe. I don't know how
2961 * long we can run without blocking
2962 */
2963 cond_resched();
2964 TXN_LOCK();
2965 } else {
Ingo Molnar1de87442006-01-24 15:22:50 -06002966 /* We can't get the commit mutex. It may
Linus Torvalds1da177e2005-04-16 15:20:36 -07002967 * be held by a thread waiting for tlock's
2968 * so let's not block here. Save it to
2969 * put back on the anon_list.
2970 */
2971
2972 /* Take off anon_list */
2973 list_del(&jfs_ip->anon_inode_list);
2974
2975 /* Put on anon_list2 */
2976 list_add(&jfs_ip->anon_inode_list,
2977 &TxAnchor.anon_list2);
2978
2979 TXN_UNLOCK();
2980 iput(ip);
2981 TXN_LOCK();
2982 }
2983 }
2984 /* Add anon_list2 back to anon_list */
2985 list_splice_init(&TxAnchor.anon_list2, &TxAnchor.anon_list);
2986
Christoph Lameter3e1d1d22005-06-24 23:13:50 -07002987 if (freezing(current)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002988 TXN_UNLOCK();
Christoph Lameter3e1d1d22005-06-24 23:13:50 -07002989 refrigerator();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002990 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002991 set_current_state(TASK_INTERRUPTIBLE);
2992 TXN_UNLOCK();
2993 schedule();
2994 current->state = TASK_RUNNING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002995 }
Christoph Hellwig91dbb4d2006-02-15 12:49:04 -06002996 } while (!kthread_should_stop());
Linus Torvalds1da177e2005-04-16 15:20:36 -07002997
2998 jfs_info("jfs_sync being killed");
Christoph Hellwig91dbb4d2006-02-15 12:49:04 -06002999 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003000}
3001
3002#if defined(CONFIG_PROC_FS) && defined(CONFIG_JFS_DEBUG)
3003int jfs_txanchor_read(char *buffer, char **start, off_t offset, int length,
3004 int *eof, void *data)
3005{
3006 int len = 0;
3007 off_t begin;
3008 char *freewait;
3009 char *freelockwait;
3010 char *lowlockwait;
3011
3012 freewait =
3013 waitqueue_active(&TxAnchor.freewait) ? "active" : "empty";
3014 freelockwait =
3015 waitqueue_active(&TxAnchor.freelockwait) ? "active" : "empty";
3016 lowlockwait =
3017 waitqueue_active(&TxAnchor.lowlockwait) ? "active" : "empty";
3018
3019 len += sprintf(buffer,
3020 "JFS TxAnchor\n"
3021 "============\n"
3022 "freetid = %d\n"
3023 "freewait = %s\n"
3024 "freelock = %d\n"
3025 "freelockwait = %s\n"
3026 "lowlockwait = %s\n"
3027 "tlocksInUse = %d\n"
3028 "jfs_tlocks_low = %d\n"
3029 "unlock_queue is %sempty\n",
3030 TxAnchor.freetid,
3031 freewait,
3032 TxAnchor.freelock,
3033 freelockwait,
3034 lowlockwait,
3035 TxAnchor.tlocksInUse,
3036 jfs_tlocks_low,
3037 list_empty(&TxAnchor.unlock_queue) ? "" : "not ");
3038
3039 begin = offset;
3040 *start = buffer + begin;
3041 len -= begin;
3042
3043 if (len > length)
3044 len = length;
3045 else
3046 *eof = 1;
3047
3048 if (len < 0)
3049 len = 0;
3050
3051 return len;
3052}
3053#endif
3054
3055#if defined(CONFIG_PROC_FS) && defined(CONFIG_JFS_STATISTICS)
3056int jfs_txstats_read(char *buffer, char **start, off_t offset, int length,
3057 int *eof, void *data)
3058{
3059 int len = 0;
3060 off_t begin;
3061
3062 len += sprintf(buffer,
3063 "JFS TxStats\n"
3064 "===========\n"
3065 "calls to txBegin = %d\n"
3066 "txBegin blocked by sync barrier = %d\n"
3067 "txBegin blocked by tlocks low = %d\n"
3068 "txBegin blocked by no free tid = %d\n"
3069 "calls to txBeginAnon = %d\n"
3070 "txBeginAnon blocked by sync barrier = %d\n"
3071 "txBeginAnon blocked by tlocks low = %d\n"
3072 "calls to txLockAlloc = %d\n"
3073 "tLockAlloc blocked by no free lock = %d\n",
3074 TxStat.txBegin,
3075 TxStat.txBegin_barrier,
3076 TxStat.txBegin_lockslow,
3077 TxStat.txBegin_freetid,
3078 TxStat.txBeginAnon,
3079 TxStat.txBeginAnon_barrier,
3080 TxStat.txBeginAnon_lockslow,
3081 TxStat.txLockAlloc,
3082 TxStat.txLockAlloc_freelock);
3083
3084 begin = offset;
3085 *start = buffer + begin;
3086 len -= begin;
3087
3088 if (len > length)
3089 len = length;
3090 else
3091 *eof = 1;
3092
3093 if (len < 0)
3094 len = 0;
3095
3096 return len;
3097}
3098#endif