2 * Copyright (C) International Business Machines Corp., 2000-2005
3 * Portions Copyright (C) Christoph Hellwig, 2001-2002
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.
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.
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
21 * jfs_txnmgr.c: transaction manager
24 * transaction starts with txBegin() and ends with txCommit()
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).
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.
37 * tlock during write()/mmap() under anonymous transaction (tid = 0):
38 * transferred (?) to transaction at commit time.
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
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>
52 #include "jfs_incore.h"
53 #include "jfs_inode.h"
54 #include "jfs_filsys.h"
55 #include "jfs_metapage.h"
56 #include "jfs_dinode.h"
59 #include "jfs_superblock.h"
60 #include "jfs_debug.h"
63 * transaction management structures
66 int freetid; /* index of a free tid structure */
67 int freelock; /* index first free lock word */
68 wait_queue_head_t freewait; /* eventlist of free tblock */
69 wait_queue_head_t freelockwait; /* eventlist of free tlock */
70 wait_queue_head_t lowlockwait; /* eventlist of ample tlocks */
71 int tlocksInUse; /* Number of tlocks in use */
72 spinlock_t LazyLock; /* synchronize sync_queue & unlock_queue */
73 /* struct tblock *sync_queue; * Transactions waiting for data sync */
74 struct list_head unlock_queue; /* Txns waiting to be released */
75 struct list_head anon_list; /* inodes having anonymous txns */
76 struct list_head anon_list2; /* inodes having anonymous txns
77 that couldn't be sync'ed */
80 int jfs_tlocks_low; /* Indicates low number of available tlocks */
82 #ifdef CONFIG_JFS_STATISTICS
86 uint txBegin_lockslow;
89 uint txBeginAnon_barrier;
90 uint txBeginAnon_lockslow;
92 uint txLockAlloc_freelock;
96 static int nTxBlock = -1; /* number of transaction blocks */
97 module_param(nTxBlock, int, 0);
98 MODULE_PARM_DESC(nTxBlock,
99 "Number of transaction blocks (max:65536)");
101 static int nTxLock = -1; /* number of transaction locks */
102 module_param(nTxLock, int, 0);
103 MODULE_PARM_DESC(nTxLock,
104 "Number of transaction locks (max:65536)");
106 struct tblock *TxBlock; /* transaction block table */
107 static int TxLockLWM; /* Low water mark for number of txLocks used */
108 static int TxLockHWM; /* High water mark for number of txLocks used */
109 static int TxLockVHWM; /* Very High water mark */
110 struct tlock *TxLock; /* transaction lock table */
113 * transaction management lock
115 static DEFINE_SPINLOCK(jfsTxnLock);
117 #define TXN_LOCK() spin_lock(&jfsTxnLock)
118 #define TXN_UNLOCK() spin_unlock(&jfsTxnLock)
120 #define LAZY_LOCK_INIT() spin_lock_init(&TxAnchor.LazyLock);
121 #define LAZY_LOCK(flags) spin_lock_irqsave(&TxAnchor.LazyLock, flags)
122 #define LAZY_UNLOCK(flags) spin_unlock_irqrestore(&TxAnchor.LazyLock, flags)
124 DECLARE_WAIT_QUEUE_HEAD(jfs_sync_thread_wait);
125 DECLARE_WAIT_QUEUE_HEAD(jfs_commit_thread_wait);
126 static int jfs_commit_thread_waking;
129 * Retry logic exist outside these macros to protect from spurrious wakeups.
131 static inline void TXN_SLEEP_DROP_LOCK(wait_queue_head_t * event)
133 DECLARE_WAITQUEUE(wait, current);
135 add_wait_queue(event, &wait);
136 set_current_state(TASK_UNINTERRUPTIBLE);
139 current->state = TASK_RUNNING;
140 remove_wait_queue(event, &wait);
143 #define TXN_SLEEP(event)\
145 TXN_SLEEP_DROP_LOCK(event);\
149 #define TXN_WAKEUP(event) wake_up_all(event)
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 */
165 static int diLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
166 struct tlock * tlck, struct commit * cd);
167 static int dataLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
168 struct tlock * tlck);
169 static void dtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
170 struct tlock * tlck);
171 static void mapLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
172 struct tlock * tlck);
173 static void txAllocPMap(struct inode *ip, struct maplock * maplock,
174 struct tblock * tblk);
175 static void txForce(struct tblock * tblk);
176 static int txLog(struct jfs_log * log, struct tblock * tblk,
178 static void txUpdateMap(struct tblock * tblk);
179 static void txRelease(struct tblock * tblk);
180 static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
181 struct tlock * tlck);
182 static void LogSyncRelease(struct metapage * mp);
185 * transaction block/lock management
186 * ---------------------------------
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.)
194 static lid_t txLockAlloc(void)
198 INCREMENT(TxStat.txLockAlloc);
199 if (!TxAnchor.freelock) {
200 INCREMENT(TxStat.txLockAlloc_freelock);
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");
210 wake_up(&jfs_sync_thread_wait);
216 static void txLockFree(lid_t lid)
219 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");
225 TXN_WAKEUP(&TxAnchor.lowlockwait);
227 TXN_WAKEUP(&TxAnchor.freelockwait);
233 * FUNCTION: initialize transaction management structures
237 * serialization: single thread at jfs_init()
244 /* Set defaults for nTxLock and nTxBlock if unset */
247 if (nTxBlock == -1) {
248 /* Base default on memory size */
250 if (si.totalram > (256 * 1024)) /* 1 GB */
253 nTxLock = si.totalram >> 2;
254 } else if (nTxBlock > (8 * 1024))
257 nTxLock = nTxBlock << 3;
260 nTxBlock = nTxLock >> 3;
262 /* Verify tunable parameters */
264 nTxBlock = 16; /* No one should set it this low */
265 if (nTxBlock > 65536)
268 nTxLock = 256; /* No one should set it this low */
272 printk(KERN_INFO "JFS: nTxBlock = %d, nTxLock = %d\n",
275 * initialize transaction block (tblock) table
277 * transaction id (tid) = tblock index
278 * tid = 0 is reserved.
280 TxLockLWM = (nTxLock * 4) / 10;
281 TxLockHWM = (nTxLock * 7) / 10;
282 TxLockVHWM = (nTxLock * 8) / 10;
284 size = sizeof(struct tblock) * nTxBlock;
285 TxBlock = (struct tblock *) vmalloc(size);
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);
295 init_waitqueue_head(&TxBlock[k].gcwait);
296 init_waitqueue_head(&TxBlock[k].waitor);
298 TxAnchor.freetid = 1;
299 init_waitqueue_head(&TxAnchor.freewait);
301 stattx.maxtid = 1; /* statistics */
304 * initialize transaction lock (tlock) table
306 * transaction lock id = tlock index
307 * tlock id = 0 is reserved.
309 size = sizeof(struct tlock) * nTxLock;
310 TxLock = (struct tlock *) vmalloc(size);
311 if (TxLock == NULL) {
316 /* initialize tlock table */
317 for (k = 1; k < nTxLock - 1; k++)
318 TxLock[k].next = k + 1;
320 init_waitqueue_head(&TxAnchor.freelockwait);
321 init_waitqueue_head(&TxAnchor.lowlockwait);
323 TxAnchor.freelock = 1;
324 TxAnchor.tlocksInUse = 0;
325 INIT_LIST_HEAD(&TxAnchor.anon_list);
326 INIT_LIST_HEAD(&TxAnchor.anon_list2);
329 INIT_LIST_HEAD(&TxAnchor.unlock_queue);
331 stattx.maxlid = 1; /* statistics */
339 * FUNCTION: clean up when module is unloaded
352 * FUNCTION: start a transaction.
354 * PARAMETER: sb - superblock
355 * flag - force for nested tx;
357 * RETURN: tid - transaction id
359 * note: flag force allows to start tx for nested tx
360 * to prevent deadlock on logsync barrier;
362 tid_t txBegin(struct super_block *sb, int flag)
368 jfs_info("txBegin: flag = 0x%x", flag);
369 log = JFS_SBI(sb)->log;
373 INCREMENT(TxStat.txBegin);
376 if (!(flag & COMMIT_FORCE)) {
378 * synchronize with logsync barrier
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);
389 * Don't begin transaction if we're getting starved for tlocks
390 * unless COMMIT_FORCE or COMMIT_INODE (which may ultimately
393 if (TxAnchor.tlocksInUse > TxLockVHWM) {
394 INCREMENT(TxStat.txBegin_lockslow);
395 TXN_SLEEP(&TxAnchor.lowlockwait);
401 * allocate transaction id/block
403 if ((t = TxAnchor.freetid) == 0) {
404 jfs_info("txBegin: waiting for free tid");
405 INCREMENT(TxStat.txBegin_freetid);
406 TXN_SLEEP(&TxAnchor.freewait);
410 tblk = tid_to_tblock(t);
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);
420 TxAnchor.freetid = tblk->next;
423 * initialize transaction
427 * We can't zero the whole thing or we screw up another thread being
428 * awakened after sleeping on tblk->waitor
430 * memset(tblk, 0, sizeof(struct tblock));
432 tblk->next = tblk->last = tblk->xflag = tblk->flag = tblk->lsn = 0;
436 tblk->logtid = log->logtid;
440 HIGHWATERMARK(stattx.maxtid, t); /* statistics */
441 INCREMENT(stattx.ntid); /* statistics */
445 jfs_info("txBegin: returning tid = %d", t);
451 * NAME: txBeginAnon()
453 * FUNCTION: start an anonymous transaction.
454 * Blocks if logsync or available tlocks are low to prevent
455 * anonymous tlocks from depleting supply.
457 * PARAMETER: sb - superblock
461 void txBeginAnon(struct super_block *sb)
465 log = JFS_SBI(sb)->log;
468 INCREMENT(TxStat.txBeginAnon);
472 * synchronize with logsync barrier
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);
482 * Don't begin transaction if we're getting starved for tlocks
484 if (TxAnchor.tlocksInUse > TxLockVHWM) {
485 INCREMENT(TxStat.txBeginAnon_lockslow);
486 TXN_SLEEP(&TxAnchor.lowlockwait);
495 * function: free specified transaction block.
497 * logsync barrier processing:
501 void txEnd(tid_t tid)
503 struct tblock *tblk = tid_to_tblock(tid);
506 jfs_info("txEnd: tid = %d", tid);
510 * wakeup transactions waiting on the page locked
511 * by the current transaction
513 TXN_WAKEUP(&tblk->waitor);
515 log = JFS_SBI(tblk->sb)->log;
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
522 * Lazy commit thread will turn off tblkGC_LAZY before calling this
525 if (tblk->flag & tblkGC_LAZY) {
526 jfs_info("txEnd called w/lazy tid: %d, tblk = 0x%p", tid, tblk);
529 spin_lock_irq(&log->gclock); // LOGGC_LOCK
530 tblk->flag |= tblkGC_UNLOCKED;
531 spin_unlock_irq(&log->gclock); // LOGGC_UNLOCK
535 jfs_info("txEnd: tid: %d, tblk = 0x%p", tid, tblk);
537 assert(tblk->next == 0);
540 * insert tblock back on freelist
542 tblk->next = TxAnchor.freetid;
543 TxAnchor.freetid = tid;
546 * mark the tblock not active
548 if (--log->active == 0) {
549 clear_bit(log_FLUSH, &log->flag);
552 * synchronize with logsync barrier
554 if (test_bit(log_SYNCBARRIER, &log->flag)) {
557 /* write dirty metadata & forward log syncpt */
560 jfs_info("log barrier off: 0x%x", log->lsn);
562 /* enable new transactions start */
563 clear_bit(log_SYNCBARRIER, &log->flag);
565 /* wakeup all waitors for logsync barrier */
566 TXN_WAKEUP(&log->syncwait);
575 * wakeup all waitors for a free tblock
577 TXN_WAKEUP(&TxAnchor.freewait);
583 * function: acquire a transaction lock on the specified <mp>
587 * return: transaction lock id
591 struct tlock *txLock(tid_t tid, struct inode *ip, struct metapage * mp,
594 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
599 struct xtlock *xtlck;
600 struct linelock *linelock;
606 if (S_ISDIR(ip->i_mode) && (type & tlckXTREE) &&
607 !(mp->xflag & COMMIT_PAGE)) {
609 * Directory inode is special. It can have both an xtree tlock
610 * and a dtree tlock associated with it.
617 /* is page not locked by a transaction ? */
621 jfs_info("txLock: tid:%d ip:0x%p mp:0x%p lid:%d", tid, ip, mp, lid);
623 /* is page locked by the requester transaction ? */
624 tlck = lid_to_tlock(lid);
625 if ((xtid = tlck->tid) == tid) {
631 * is page locked by anonymous transaction/lock ?
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)
644 tblk = tid_to_tblock(tid);
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
652 * Note: We really need to get rid of the tid & lid and
653 * use list_head's. This code is getting UGLY!
655 if (jfs_ip->atlhead == lid) {
656 if (jfs_ip->atltail == lid) {
657 /* only anonymous txn.
658 * Remove from anon_list
661 list_del_init(&jfs_ip->anon_inode_list);
664 jfs_ip->atlhead = tlck->next;
667 for (last = jfs_ip->atlhead;
668 lid_to_tlock(last)->next != lid;
669 last = lid_to_tlock(last)->next) {
672 lid_to_tlock(last)->next = tlck->next;
673 if (jfs_ip->atltail == lid)
674 jfs_ip->atltail = last;
677 /* insert the tlock at tail of transaction tlock list */
680 lid_to_tlock(tblk->last)->next = lid;
696 tlck = lid_to_tlock(lid);
705 /* mark tlock for meta-data page */
706 if (mp->xflag & COMMIT_PAGE) {
708 tlck->flag = tlckPAGELOCK;
710 /* mark the page dirty and nohomeok */
711 metapage_nohomeok(mp);
713 jfs_info("locking mp = 0x%p, nohomeok = %d tid = %d tlck = 0x%p",
714 mp, mp->nohomeok, tid, tlck);
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
721 if ((tid == 0) && mp->lsn)
722 set_cflag(COMMIT_Synclist, ip);
724 /* mark tlock for in-memory inode */
726 tlck->flag = tlckINODELOCK;
728 if (S_ISDIR(ip->i_mode))
729 tlck->flag |= tlckDIRECTORY;
733 /* bind the tlock and the page */
742 * enqueue transaction lock to transaction/inode
744 /* insert the tlock at tail of transaction tlock list */
746 tblk = tid_to_tblock(tid);
748 lid_to_tlock(tblk->last)->next = lid;
754 /* anonymous transaction:
755 * insert the tlock at head of inode anonymous tlock list
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;
764 list_add_tail(&jfs_ip->anon_inode_list,
765 &TxAnchor.anon_list);
770 /* initialize type dependent area for linelock */
771 linelock = (struct linelock *) & tlck->lock;
773 linelock->flag = tlckLINELOCK;
774 linelock->maxcnt = TLOCKSHORT;
777 switch (type & tlckTYPE) {
779 linelock->l2linesize = L2DTSLOTSIZE;
783 linelock->l2linesize = L2XTSLOTSIZE;
785 xtlck = (struct xtlock *) linelock;
786 xtlck->header.offset = 0;
787 xtlck->header.length = 2;
789 if (type & tlckNEW) {
790 xtlck->lwm.offset = XTENTRYSTART;
792 if (mp->xflag & COMMIT_PAGE)
793 p = (xtpage_t *) mp->data;
795 p = &jfs_ip->i_xtroot;
797 le16_to_cpu(p->header.nextindex);
799 xtlck->lwm.length = 0; /* ! */
800 xtlck->twm.offset = 0;
801 xtlck->hwm.offset = 0;
807 linelock->l2linesize = L2INODESLOTSIZE;
811 linelock->l2linesize = L2DATASLOTSIZE;
815 jfs_err("UFO tlock:0x%p", tlck);
819 * update tlock vector
827 * page is being locked by another transaction:
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));
841 INCREMENT(stattx.waitlock); /* statistics */
843 release_metapage(mp);
845 xtid = tlck->tid; /* reaquire after dropping TXN_LOCK */
847 jfs_info("txLock: in waitLock, tid = %d, xtid = %d, lid = %d",
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);
855 jfs_info("txLock: awakened tid = %d, lid = %d", tid, lid);
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.
871 * RETURN: Errors from subroutines.
873 static void txRelease(struct tblock * tblk)
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);
891 * wakeup transactions waiting on a page locked
892 * by the current transaction
894 TXN_WAKEUP(&tblk->waitor);
902 * FUNCTION: Initiates pageout of pages modified by tid in journalled
903 * objects and frees their lockwords.
905 static void txUnlock(struct tblock * tblk)
908 struct linelock *linelock;
909 lid_t lid, next, llid, k;
915 jfs_info("txUnlock: tblk = 0x%p", tblk);
916 log = JFS_SBI(tblk->sb)->log;
919 * mark page under tlock homeok (its log has been written):
921 for (lid = tblk->next; lid; lid = next) {
922 tlck = lid_to_tlock(lid);
925 jfs_info("unlocking lid = %d, tlck = 0x%p", lid, tlck);
927 /* unbind page from tlock */
928 if ((mp = tlck->mp) != NULL &&
929 (tlck->type & tlckBTROOT) == 0) {
930 assert(mp->xflag & COMMIT_PAGE);
936 assert(mp->nohomeok > 0);
937 _metapage_homeok(mp);
939 /* inherit younger/larger clsn */
940 LOGSYNC_LOCK(log, flags);
942 logdiff(difft, tblk->clsn, log);
943 logdiff(diffp, mp->clsn, log);
945 mp->clsn = tblk->clsn;
947 mp->clsn = tblk->clsn;
948 LOGSYNC_UNLOCK(log, flags);
950 assert(!(tlck->flag & tlckFREEPAGE));
955 /* insert tlock, and linelock(s) of the tlock if any,
956 * at head of freelist
960 llid = ((struct linelock *) & tlck->lock)->next;
962 linelock = (struct linelock *) lid_to_tlock(llid);
971 tblk->next = tblk->last = 0;
974 * remove tblock from logsynclist
975 * (allocation map pages inherited lsn of tblk and
976 * has been inserted in logsync list at txUpdateMap())
979 LOGSYNC_LOCK(log, flags);
981 list_del(&tblk->synclist);
982 LOGSYNC_UNLOCK(log, flags);
989 * function: allocate a transaction lock for freed page/entry;
990 * for freed page, maplock is used as xtlock/dtlock type;
992 struct tlock *txMaplock(tid_t tid, struct inode *ip, int type)
994 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
998 struct maplock *maplock;
1005 lid = txLockAlloc();
1006 tlck = lid_to_tlock(lid);
1013 /* bind the tlock and the object */
1014 tlck->flag = tlckINODELOCK;
1015 if (S_ISDIR(ip->i_mode))
1016 tlck->flag |= tlckDIRECTORY;
1023 * enqueue transaction lock to transaction/inode
1025 /* insert the tlock at tail of transaction tlock list */
1027 tblk = tid_to_tblock(tid);
1029 lid_to_tlock(tblk->last)->next = lid;
1035 /* anonymous transaction:
1036 * insert the tlock at head of inode anonymous tlock list
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);
1051 /* initialize type dependent area for maplock */
1052 maplock = (struct maplock *) & tlck->lock;
1054 maplock->maxcnt = 0;
1063 * function: allocate a transaction lock for log vector list
1065 struct linelock *txLinelock(struct linelock * tlock)
1069 struct linelock *linelock;
1073 /* allocate a TxLock structure */
1074 lid = txLockAlloc();
1075 tlck = lid_to_tlock(lid);
1079 /* initialize linelock */
1080 linelock = (struct linelock *) tlck;
1082 linelock->flag = tlckLINELOCK;
1083 linelock->maxcnt = TLOCKLONG;
1084 linelock->index = 0;
1085 if (tlck->flag & tlckDIRECTORY)
1086 linelock->flag |= tlckDIRECTORY;
1088 /* append linelock after tlock */
1089 linelock->next = tlock->next;
1096 * transaction commit management
1097 * -----------------------------
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).
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.
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
1127 * on entry the inode lock on each segment is assumed
1132 int txCommit(tid_t tid, /* transaction identifier */
1133 int nip, /* number of inodes to commit */
1134 struct inode **iplist, /* list of inode to commit */
1139 struct jfs_log *log;
1140 struct tblock *tblk;
1144 struct jfs_inode_info *jfs_ip;
1147 struct super_block *sb;
1149 jfs_info("txCommit, tid = %d, flag = %d", tid, flag);
1150 /* is read-only file system ? */
1151 if (isReadOnly(iplist[0])) {
1156 sb = cd.sb = iplist[0]->i_sb;
1160 tid = txBegin(sb, 0);
1161 tblk = tid_to_tblock(tid);
1164 * initialize commit structure
1166 log = JFS_SBI(sb)->log;
1169 /* initialize log record descriptor in commit */
1171 lrd->logtid = cpu_to_le32(tblk->logtid);
1174 tblk->xflag |= flag;
1176 if ((flag & (COMMIT_FORCE | COMMIT_SYNC)) == 0)
1177 tblk->xflag |= COMMIT_LAZY;
1179 * prepare non-journaled objects for commit
1181 * flush data pages of non-journaled file
1182 * to prevent the file getting non-initialized disk blocks
1190 * acquire transaction lock on (on-disk) inodes
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
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
1201 for (k = 0; k < cd.nip; k++) {
1202 top = (cd.iplist[k])->i_ino;
1203 for (n = k + 1; n < cd.nip; n++) {
1205 if (ip->i_ino > top) {
1207 cd.iplist[n] = cd.iplist[k];
1213 jfs_ip = JFS_IP(ip);
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.)
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.
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.
1233 * if ((!S_ISDIR(ip->i_mode))
1234 * && (tblk->flag & COMMIT_DELETE) == 0)
1235 * filemap_write_and_wait(ip->i_mapping);
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
1243 clear_cflag(COMMIT_Dirty, ip);
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;
1250 tblk->last = jfs_ip->atltail;
1251 jfs_ip->atlhead = jfs_ip->atltail = 0;
1253 list_del_init(&jfs_ip->anon_inode_list);
1258 * acquire transaction lock on on-disk inode page
1259 * (become first tlock of the tblk's tlock list)
1261 if (((rc = diWrite(tid, ip))))
1266 * write log records from transaction locks
1268 * txUpdateMap() resets XAD_NEW in XAD.
1270 if ((rc = txLog(log, tblk, &cd)))
1274 * Ensure that inode isn't reused before
1275 * lazy commit thread finishes processing
1277 if (tblk->xflag & COMMIT_DELETE) {
1278 atomic_inc(&tblk->u.ip->i_count);
1280 * Avoid a rare deadlock
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)
1289 if (tblk->u.ip->i_state & I_LOCK)
1290 tblk->xflag &= ~COMMIT_LAZY;
1293 ASSERT((!(tblk->xflag & COMMIT_DELETE)) ||
1294 ((tblk->u.ip->i_nlink == 0) &&
1295 !test_cflag(COMMIT_Nolink, tblk->u.ip)));
1298 * write COMMIT log record
1300 lrd->type = cpu_to_le16(LOG_COMMIT);
1302 lsn = lmLog(log, tblk, lrd, NULL);
1304 lmGroupCommit(log, tblk);
1307 * - transaction is now committed -
1311 * force pages in careful update
1312 * (imap addressing structure update)
1314 if (flag & COMMIT_FORCE)
1318 * update allocation map.
1320 * update inode allocation map and inode:
1321 * free pager lock on memory object of inode if any.
1322 * update block allocation map.
1324 * txUpdateMap() resets XAD_NEW in XAD.
1326 if (tblk->xflag & COMMIT_FORCE)
1330 * free transaction locks and pageout/free pages
1334 if ((tblk->flag & tblkGC_LAZY) == 0)
1339 * reset in-memory object state
1341 for (k = 0; k < cd.nip; k++) {
1343 jfs_ip = JFS_IP(ip);
1346 * reset in-memory inode state
1357 jfs_info("txCommit: tid = %d, returning %d", tid, rc);
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.
1372 static int txLog(struct jfs_log * log, struct tblock * tblk, struct commit * cd)
1378 struct lrd *lrd = &cd->lrd;
1381 * write log record(s) for each tlock of transaction,
1383 for (lid = tblk->next; lid; lid = tlck->next) {
1384 tlck = lid_to_tlock(lid);
1386 tlck->flag |= tlckLOG;
1388 /* initialize lrd common */
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);
1394 /* write log record of page from the tlock */
1395 switch (tlck->type & tlckTYPE) {
1397 xtLog(log, tblk, lrd, tlck);
1401 dtLog(log, tblk, lrd, tlck);
1405 diLog(log, tblk, lrd, tlck, cd);
1409 mapLog(log, tblk, lrd, tlck);
1413 dataLog(log, tblk, lrd, tlck);
1417 jfs_err("UFO tlock:0x%p", tlck);
1427 * function: log inode tlock and format maplock to update bmap;
1429 static int diLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1430 struct tlock * tlck, struct commit * cd)
1433 struct metapage *mp;
1435 struct pxd_lock *pxdlock;
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);
1443 pxd = &lrd->log.redopage.pxd;
1448 if (tlck->type & tlckENTRY) {
1449 /* log after-image for logredo(): */
1450 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1451 PXDaddress(pxd, mp->index);
1453 mp->logical_size >> tblk->sb->s_blocksize_bits);
1454 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1456 /* mark page as homeward bound */
1457 tlck->flag |= tlckWRITEPAGE;
1458 } else if (tlck->type & tlckFREE) {
1462 * (pages of the freed inode extent have been invalidated and
1463 * a maplock for free of the extent has been formatted at
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;
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;
1476 lrd->type = cpu_to_le16(LOG_NOREDOINOEXT);
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].
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]);
1489 pxdlock = (struct pxd_lock *) & tlck->lock;
1490 *pxd = pxdlock->pxd;
1491 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1494 tlck->flag |= tlckUPDATEMAP;
1496 /* mark page as homeward bound */
1497 tlck->flag |= tlckWRITEPAGE;
1499 jfs_err("diLog: UFO type tlck:0x%p", tlck);
1502 * alloc/free external EA extent
1504 * a maplock for txUpdateMap() to update bPWMAP for alloc/free
1505 * of the extent has been formatted at txLock() time;
1508 assert(tlck->type & tlckEA);
1510 /* log LOG_UPDATEMAP for logredo() to update bmap for
1511 * alloc of new (and free of old) external EA extent;
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);
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;
1526 cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1530 tlck->flag |= tlckUPDATEMAP;
1532 #endif /* _JFS_WIP */
1540 * function: log data tlock
1542 static int dataLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1543 struct tlock * tlck)
1545 struct metapage *mp;
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);
1554 pxd = &lrd->log.redopage.pxd;
1556 /* log after-image for logredo(): */
1557 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1559 if (jfs_dirtable_inline(tlck->ip)) {
1561 * The table has been truncated, we've must have deleted
1562 * the last entry, so don't bother logging this
1566 metapage_homeok(mp);
1567 discard_metapage(mp);
1572 PXDaddress(pxd, mp->index);
1573 PXDlength(pxd, mp->logical_size >> tblk->sb->s_blocksize_bits);
1575 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1577 /* mark page as homeward bound */
1578 tlck->flag |= tlckWRITEPAGE;
1586 * function: log dtree tlock and format maplock to update bmap;
1588 static void dtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1589 struct tlock * tlck)
1591 struct metapage *mp;
1592 struct pxd_lock *pxdlock;
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);
1601 pxd = &lrd->log.redopage.pxd;
1603 if (tlck->type & tlckBTROOT)
1604 lrd->log.redopage.type |= cpu_to_le16(LOG_BTROOT);
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;
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;
1617 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1618 if (tlck->type & tlckEXTEND)
1619 lrd->log.redopage.type |= cpu_to_le16(LOG_EXTEND);
1621 lrd->log.redopage.type |= cpu_to_le16(LOG_NEW);
1622 PXDaddress(pxd, mp->index);
1624 mp->logical_size >> tblk->sb->s_blocksize_bits);
1625 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1627 /* format a maplock for txUpdateMap() to update bPMAP for
1628 * alloc of the new page;
1630 if (tlck->type & tlckBTROOT)
1632 tlck->flag |= tlckUPDATEMAP;
1633 pxdlock = (struct pxd_lock *) & tlck->lock;
1634 pxdlock->flag = mlckALLOCPXD;
1635 pxdlock->pxd = *pxd;
1639 /* mark page as homeward bound */
1640 tlck->flag |= tlckWRITEPAGE;
1645 * entry insertion/deletion,
1646 * sibling page link update (old right page before split);
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);
1653 mp->logical_size >> tblk->sb->s_blocksize_bits);
1654 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1656 /* mark page as homeward bound */
1657 tlck->flag |= tlckWRITEPAGE;
1662 * page deletion: page has been invalidated
1663 * page relocation: source extent
1665 * a maplock for free of the page has been formatted
1666 * at txLock() time);
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
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));
1678 /* a maplock for txUpdateMap() for free of the page
1679 * has been formatted at txLock() time;
1681 tlck->flag |= tlckUPDATEMAP;
1689 * function: log xtree tlock and format maplock to update bmap;
1691 static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1692 struct tlock * tlck)
1695 struct metapage *mp;
1697 struct xtlock *xtlck;
1698 struct maplock *maplock;
1699 struct xdlistlock *xadlock;
1700 struct pxd_lock *pxdlock;
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);
1711 page_pxd = &lrd->log.redopage.pxd;
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);
1720 p = (xtpage_t *) mp->data;
1721 next = le16_to_cpu(p->header.nextindex);
1723 xtlck = (struct xtlock *) & tlck->lock;
1725 maplock = (struct maplock *) & tlck->lock;
1726 xadlock = (struct xdlistlock *) maplock;
1729 * entry insertion/extension;
1730 * sibling page link update (old right page before split);
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.
1740 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1741 PXDaddress(page_pxd, mp->index);
1743 mp->logical_size >> tblk->sb->s_blocksize_bits);
1744 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
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.
1751 lwm = xtlck->lwm.offset;
1753 lwm = XTPAGEMAXSLOT;
1758 jfs_err("xtLog: lwm > next\n");
1761 tlck->flag |= tlckUPDATEMAP;
1762 xadlock->flag = mlckALLOCXADLIST;
1763 xadlock->count = next - lwm;
1764 if ((xadlock->count <= 4) && (tblk->xflag & COMMIT_LAZY)) {
1768 * Lazy commit may allow xtree to be modified before
1769 * txUpdateMap runs. Copy xad into linelock to
1770 * preserve correct data.
1772 * We can fit twice as may pxd's as xads in the lock
1774 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]));
1779 p->xad[lwm + i].flag &=
1780 ~(XAD_NEW | XAD_EXTENDED);
1785 * xdlist will point to into inode's xtree, ensure
1786 * that transaction is not committed lazily.
1788 xadlock->flag = mlckALLOCXADLIST;
1789 xadlock->xdlist = &p->xad[lwm];
1790 tblk->xflag &= ~COMMIT_LAZY;
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);
1798 /* mark page as homeward bound */
1799 tlck->flag |= tlckWRITEPAGE;
1805 * page deletion: file deletion/truncation (ref. xtTruncate())
1807 * (page will be invalidated after log is written and bmap
1808 * is updated from the page);
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
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;
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;
1827 if (tblk->xflag & COMMIT_TRUNCATE) {
1828 /* write NOREDOPAGE for the page */
1829 lrd->type = cpu_to_le16(LOG_NOREDOPAGE);
1830 PXDaddress(page_pxd, mp->index);
1832 mp->logical_size >> tblk->sb->
1835 cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1837 if (tlck->type & tlckBTROOT) {
1838 /* Empty xtree must be logged */
1839 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1841 cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1845 /* init LOG_UPDATEMAP of the freed extents
1846 * XAD[XTENTRYSTART:hwm) from the deleted page itself
1847 * for logredo() to update bmap;
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;
1859 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1861 /* format a maplock for txUpdateMap() to update bmap
1862 * to free extents of XAD[XTENTRYSTART:hwm) from the
1863 * deleted page itself;
1865 tlck->flag |= tlckUPDATEMAP;
1866 xadlock->count = hwm - XTENTRYSTART + 1;
1867 if ((xadlock->count <= 4) && (tblk->xflag & COMMIT_LAZY)) {
1871 * Lazy commit may allow xtree to be modified before
1872 * txUpdateMap runs. Copy xad into linelock to
1873 * preserve correct data.
1875 * We can fit twice as may pxd's as xads in the lock
1877 xadlock->flag = mlckFREEPXDLIST;
1878 pxd = xadlock->xdlist = &xtlck->pxdlock;
1879 for (i = 0; i < xadlock->count; i++) {
1881 addressXAD(&p->xad[XTENTRYSTART + i]));
1883 lengthXAD(&p->xad[XTENTRYSTART + i]));
1888 * xdlist will point to into inode's xtree, ensure
1889 * that transaction is not committed lazily.
1891 xadlock->flag = mlckFREEXADLIST;
1892 xadlock->xdlist = &p->xad[XTENTRYSTART];
1893 tblk->xflag &= ~COMMIT_LAZY;
1895 jfs_info("xtLog: free ip:0x%p mp:0x%p count:%d lwm:2",
1896 tlck->ip, mp, xadlock->count);
1900 /* mark page as invalid */
1901 if (((tblk->xflag & COMMIT_PWMAP) || S_ISDIR(ip->i_mode))
1902 && !(tlck->type & tlckBTROOT))
1903 tlck->flag |= tlckFREEPAGE;
1905 else (tblk->xflag & COMMIT_PMAP)
1912 * page/entry truncation: file truncation (ref. xtTruncate())
1914 * |----------+------+------+---------------|
1916 * | | hwm - hwm before truncation
1917 * | next - truncation point
1918 * lwm - lwm before truncation
1921 if (tlck->type & tlckTRUNCATE) {
1922 pxd_t pxd; /* truncated extent of xad */
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
1932 tblk->xflag &= ~COMMIT_LAZY;
1933 lwm = xtlck->lwm.offset;
1935 lwm = XTPAGEMAXSLOT;
1936 hwm = xtlck->hwm.offset;
1937 twm = xtlck->twm.offset;
1942 /* log after-image for logredo():
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.
1950 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1951 PXDaddress(page_pxd, mp->index);
1953 mp->logical_size >> tblk->sb->s_blocksize_bits);
1954 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1957 * truncate entry XAD[twm == next - 1]:
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);
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;
1971 pxd = pxdlock->pxd; /* save to format maplock */
1973 cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1977 * free entries XAD[next:hwm]:
1980 /* init LOG_UPDATEMAP of the freed extents
1981 * XAD[next:hwm] from the deleted page itself
1982 * for logredo() to update bmap;
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;
1996 cpu_to_le32(lmLog(log, tblk, lrd, tlck));
2000 * format maplock(s) for txUpdateMap() to update bmap
2005 * allocate entries XAD[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.
2013 tlck->flag |= tlckUPDATEMAP;
2014 xadlock->flag = mlckALLOCXADLIST;
2015 xadlock->count = next - lwm;
2016 xadlock->xdlist = &p->xad[lwm];
2018 jfs_info("xtLog: alloc ip:0x%p mp:0x%p count:%d "
2020 tlck->ip, mp, xadlock->count, lwm, next);
2026 * truncate entry XAD[twm == next - 1]:
2028 if (twm == next - 1) {
2029 struct pxd_lock *pxdlock;
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);
2036 tlck->flag |= tlckUPDATEMAP;
2037 pxdlock = (struct pxd_lock *) xadlock;
2038 pxdlock->flag = mlckFREEPXD;
2042 jfs_info("xtLog: truncate ip:0x%p mp:0x%p count:%d "
2043 "hwm:%d", ip, mp, pxdlock->count, hwm);
2049 * free entries XAD[next:hwm]:
2052 /* format a maplock for txUpdateMap() to update bmap
2053 * to free extents of XAD[next:hwm] from thedeleted
2056 tlck->flag |= tlckUPDATEMAP;
2057 xadlock->flag = mlckFREEXADLIST;
2058 xadlock->count = hwm - next + 1;
2059 xadlock->xdlist = &p->xad[next];
2061 jfs_info("xtLog: free ip:0x%p mp:0x%p count:%d "
2063 tlck->ip, mp, xadlock->count, next, hwm);
2067 /* mark page as homeward bound */
2068 tlck->flag |= tlckWRITEPAGE;
2076 * function: log from maplock of freed data extents;
2078 static void mapLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
2079 struct tlock * tlck)
2081 struct pxd_lock *pxdlock;
2086 * page relocation: free the source page extent
2088 * a maplock for txUpdateMap() for free of the page
2089 * has been formatted at txLock() time saving the src
2090 * relocated page address;
2092 if (tlck->type & tlckRELOCATE) {
2093 /* log LOG_NOREDOPAGE of the old relocated page
2094 * for logredo() to start NoRedoPage filter;
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));
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);
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));
2118 /* a maplock for txUpdateMap() for free of the page
2119 * has been formatted at txLock() time;
2121 tlck->flag |= tlckUPDATEMAP;
2126 * Otherwise it's not a relocate request
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();
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);
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;
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));
2155 tlck->flag |= tlckUPDATEMAP;
2162 * function: acquire maplock for EA/ACL extents or
2163 * set COMMIT_INLINE flag;
2165 void txEA(tid_t tid, struct inode *ip, dxd_t * oldea, dxd_t * newea)
2167 struct tlock *tlck = NULL;
2168 struct pxd_lock *maplock = NULL, *pxdlock = NULL;
2171 * format maplock for alloc of new EA extent
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
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));
2187 } else if (newea->flag & DXD_INLINE) {
2190 set_cflag(COMMIT_Inlineea, ip);
2195 * format maplock for free of old EA extent
2197 if (!test_cflag(COMMIT_Nolink, ip) && oldea->flag & DXD_EXTENT) {
2199 tlck = txMaplock(tid, ip, tlckMAP);
2200 maplock = (struct pxd_lock *) & tlck->lock;
2201 pxdlock = (struct pxd_lock *) maplock;
2204 pxdlock->flag = mlckFREEPXD;
2205 PXDaddress(&pxdlock->pxd, addressDXD(oldea));
2206 PXDlength(&pxdlock->pxd, lengthDXD(oldea));
2214 * function: synchronously write pages locked by transaction
2215 * after txLog() but before txUpdateMap();
2217 static void txForce(struct tblock * tblk)
2221 struct metapage *mp;
2224 * reverse the order of transaction tlocks in
2225 * careful update order of address index pages
2226 * (right to left, bottom up)
2228 tlck = lid_to_tlock(tblk->next);
2232 tlck = lid_to_tlock(lid);
2234 tlck->next = tblk->next;
2240 * synchronously write the page, and
2241 * hold the page for txUpdateMap();
2243 for (lid = tblk->next; lid; lid = next) {
2244 tlck = lid_to_tlock(lid);
2247 if ((mp = tlck->mp) != NULL &&
2248 (tlck->type & tlckBTROOT) == 0) {
2249 assert(mp->xflag & COMMIT_PAGE);
2251 if (tlck->flag & tlckWRITEPAGE) {
2252 tlck->flag &= ~tlckWRITEPAGE;
2254 /* do not release page to freelist */
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.
2268 assert(mp->nohomeok);
2269 set_bit(META_dirty, &mp->flag);
2270 set_bit(META_sync, &mp->flag);
2280 * function: update persistent allocation map (and working map
2285 static void txUpdateMap(struct tblock * tblk)
2288 struct inode *ipimap;
2291 struct maplock *maplock;
2292 struct pxd_lock pxdlock;
2295 struct metapage *mp = NULL;
2297 ipimap = JFS_SBI(tblk->sb)->ipimap;
2299 maptype = (tblk->xflag & COMMIT_PMAP) ? COMMIT_PMAP : COMMIT_PWMAP;
2303 * update block allocation map
2305 * update allocation state in pmap (and wmap) and
2306 * update lsn of the pmap page;
2309 * scan each tlock/page of transaction for block allocation/free:
2311 * for each tlock/page of transaction, update map.
2312 * ? are there tlock for pmap and pwmap at the same time ?
2314 for (lid = tblk->next; lid; lid = tlck->next) {
2315 tlck = lid_to_tlock(lid);
2317 if ((tlck->flag & tlckUPDATEMAP) == 0)
2320 if (tlck->flag & tlckFREEPAGE) {
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
2329 ASSERT(mp->xflag & COMMIT_PAGE);
2335 * . in-line PXD list:
2336 * . out-of-line XAD list:
2338 maplock = (struct maplock *) & tlck->lock;
2339 nlock = maplock->index;
2341 for (k = 0; k < nlock; k++, maplock++) {
2343 * allocate blocks in persistent map:
2345 * blocks have been allocated from wmap at alloc time;
2347 if (maplock->flag & mlckALLOC) {
2348 txAllocPMap(ipimap, maplock, tblk);
2351 * free blocks in persistent and working map:
2352 * blocks will be freed in pmap and then in wmap;
2354 * ? tblock specifies the PMAP/PWMAP based upon
2357 * free blocks in persistent map:
2358 * blocks will be freed from wmap at last reference
2359 * release of the object for regular files;
2361 * Alway free blocks from both persistent & working
2362 * maps for directories
2364 else { /* (maplock->flag & mlckFREE) */
2366 if (tlck->flag & tlckDIRECTORY)
2367 txFreeMap(ipimap, maplock,
2368 tblk, COMMIT_PWMAP);
2370 txFreeMap(ipimap, maplock,
2374 if (tlck->flag & tlckFREEPAGE) {
2375 if (!(tblk->flag & tblkGC_LAZY)) {
2376 /* This is equivalent to txRelease */
2377 ASSERT(mp->lid == lid);
2380 assert(mp->nohomeok == 1);
2381 metapage_homeok(mp);
2382 discard_metapage(mp);
2387 * update inode allocation map
2389 * update allocation state in pmap and
2390 * update lsn of the pmap page;
2391 * update in-memory inode flag/state
2393 * unlock mapper/write lock
2395 if (tblk->xflag & COMMIT_CREATE) {
2396 diUpdatePMap(ipimap, tblk->ino, FALSE, tblk);
2397 /* update persistent block allocation map
2398 * for the allocation of inode extent;
2400 pxdlock.flag = mlckALLOCPXD;
2401 pxdlock.pxd = tblk->u.ixpxd;
2403 txAllocPMap(ipimap, (struct maplock *) & pxdlock, tblk);
2404 } else if (tblk->xflag & COMMIT_DELETE) {
2406 diUpdatePMap(ipimap, ip->i_ino, TRUE, tblk);
2414 * function: allocate from persistent map;
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;
2429 * lsn - log sequence number;
2431 static void txAllocPMap(struct inode *ip, struct maplock * maplock,
2432 struct tblock * tblk)
2434 struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
2435 struct xdlistlock *xadlistlock;
2439 struct pxd_lock *pxdlock;
2440 struct xdlistlock *pxdlistlock;
2445 * allocate from persistent map;
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,
2456 xad->flag &= ~(XAD_NEW | XAD_EXTENDED);
2457 jfs_info("allocPMap: xaddr:0x%lx xlen:%d",
2458 (ulong) xaddr, xlen);
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) */
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,
2476 jfs_info("allocPMap: xaddr:0x%lx xlen:%d",
2477 (ulong) xaddr, xlen);
2485 * function: free from persistent and/or working map;
2487 * todo: optimization
2489 void txFreeMap(struct inode *ip,
2490 struct maplock * maplock, struct tblock * tblk, int maptype)
2492 struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
2493 struct xdlistlock *xadlistlock;
2497 struct pxd_lock *pxdlock;
2498 struct xdlistlock *pxdlistlock;
2502 jfs_info("txFreeMap: tblk:0x%p maplock:0x%p maptype:0x%x",
2503 tblk, maplock, maptype);
2506 * free from persistent map;
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,
2518 jfs_info("freePMap: xaddr:0x%lx "
2520 (ulong) xaddr, xlen);
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,
2529 jfs_info("freePMap: xaddr:0x%lx xlen:%d",
2530 (ulong) xaddr, xlen);
2531 } else { /* (maplock->flag & mlckALLOCPXDLIST) */
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,
2540 jfs_info("freePMap: xaddr:0x%lx xlen:%d",
2541 (ulong) xaddr, xlen);
2547 * free from working map;
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);
2558 jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
2559 (ulong) xaddr, xlen);
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) */
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);
2586 * function: remove tlock from inode anonymous locklist
2588 void txFreelock(struct inode *ip)
2590 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
2591 struct tlock *xtlck, *tlck;
2592 lid_t xlid = 0, lid;
2594 if (!jfs_ip->atlhead)
2598 xtlck = (struct tlock *) &jfs_ip->atlhead;
2600 while ((lid = xtlck->next) != 0) {
2601 tlck = lid_to_tlock(lid);
2602 if (tlck->flag & tlckFREELOCK) {
2603 xtlck->next = tlck->next;
2611 if (jfs_ip->atlhead)
2612 jfs_ip->atltail = xlid;
2614 jfs_ip->atltail = 0;
2616 * If inode was on anon_list, remove it
2618 list_del_init(&jfs_ip->anon_inode_list);
2626 * function: abort tx before commit;
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).
2634 void txAbort(tid_t tid, int dirty)
2637 struct metapage *mp;
2638 struct tblock *tblk = tid_to_tblock(tid);
2642 * free tlocks of the transaction
2644 for (lid = tblk->next; lid; lid = next) {
2645 tlck = lid_to_tlock(lid);
2648 JFS_IP(tlck->ip)->xtlid = 0;
2654 * reset lsn of page to avoid logwarap:
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.)
2661 if (mp->xflag & COMMIT_PAGE && mp->lsn)
2664 /* insert tlock at head of freelist */
2670 /* caller will free the transaction block */
2672 tblk->next = tblk->last = 0;
2675 * mark filesystem dirty
2678 jfs_error(tblk->sb, "txAbort");
2684 * txLazyCommit(void)
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.
2691 static void txLazyCommit(struct tblock * tblk)
2693 struct jfs_log *log;
2695 while (((tblk->flag & tblkGC_READY) == 0) &&
2696 ((tblk->flag & tblkGC_UNLOCKED) == 0)) {
2697 /* We must have gotten ahead of the user thread
2699 jfs_info("jfs_lazycommit: tblk 0x%p not unlocked", tblk);
2703 jfs_info("txLazyCommit: processing tblk 0x%p", tblk);
2707 log = (struct jfs_log *) JFS_SBI(tblk->sb)->log;
2709 spin_lock_irq(&log->gclock); // LOGGC_LOCK
2711 tblk->flag |= tblkGC_COMMITTED;
2713 if (tblk->flag & tblkGC_READY)
2716 wake_up_all(&tblk->gcwait); // LOGGC_WAKEUP
2719 * Can't release log->gclock until we've tested tblk->flag
2721 if (tblk->flag & tblkGC_LAZY) {
2722 spin_unlock_irq(&log->gclock); // LOGGC_UNLOCK
2724 tblk->flag &= ~tblkGC_LAZY;
2725 txEnd(tblk - TxBlock); /* Convert back to tid */
2727 spin_unlock_irq(&log->gclock); // LOGGC_UNLOCK
2729 jfs_info("txLazyCommit: done: tblk = 0x%p", tblk);
2733 * jfs_lazycommit(void)
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.
2739 int jfs_lazycommit(void *arg)
2742 struct tblock *tblk;
2743 unsigned long flags;
2744 struct jfs_sb_info *sbi;
2746 daemonize("jfsCommit");
2748 complete(&jfsIOwait);
2752 jfs_commit_thread_waking = 0; /* OK to wake another thread */
2753 while (!list_empty(&TxAnchor.unlock_queue)) {
2755 list_for_each_entry(tblk, &TxAnchor.unlock_queue,
2758 sbi = JFS_SBI(tblk->sb);
2760 * For each volume, the transactions must be
2761 * handled in order. If another commit thread
2762 * is handling a tblk for this superblock,
2765 if (sbi->commit_state & IN_LAZYCOMMIT)
2768 sbi->commit_state |= IN_LAZYCOMMIT;
2772 * Remove transaction from queue
2774 list_del(&tblk->cqueue);
2780 sbi->commit_state &= ~IN_LAZYCOMMIT;
2782 * Don't continue in the for loop. (We can't
2783 * anyway, it's unsafe!) We want to go back to
2784 * the beginning of the list.
2789 /* If there was nothing to do, don't continue */
2793 /* In case a wakeup came while all threads were active */
2794 jfs_commit_thread_waking = 0;
2796 if (freezing(current)) {
2800 DECLARE_WAITQUEUE(wq, current);
2802 add_wait_queue(&jfs_commit_thread_wait, &wq);
2803 set_current_state(TASK_INTERRUPTIBLE);
2806 current->state = TASK_RUNNING;
2807 remove_wait_queue(&jfs_commit_thread_wait, &wq);
2809 } while (!jfs_stop_threads);
2811 if (!list_empty(&TxAnchor.unlock_queue))
2812 jfs_err("jfs_lazycommit being killed w/pending transactions!");
2814 jfs_info("jfs_lazycommit being killed\n");
2815 complete_and_exit(&jfsIOwait, 0);
2818 void txLazyUnlock(struct tblock * tblk)
2820 unsigned long flags;
2824 list_add_tail(&tblk->cqueue, &TxAnchor.unlock_queue);
2826 * Don't wake up a commit thread if there is already one servicing
2827 * this superblock, or if the last one we woke up hasn't started yet.
2829 if (!(JFS_SBI(tblk->sb)->commit_state & IN_LAZYCOMMIT) &&
2830 !jfs_commit_thread_waking) {
2831 jfs_commit_thread_waking = 1;
2832 wake_up(&jfs_commit_thread_wait);
2837 static void LogSyncRelease(struct metapage * mp)
2839 struct jfs_log *log = mp->log;
2841 assert(mp->nohomeok);
2843 metapage_homeok(mp);
2849 * Block all new transactions and push anonymous transactions to
2852 * This does almost the same thing as jfs_sync below. We don't
2853 * worry about deadlocking when jfs_tlocks_low is set, since we would
2854 * expect jfs_sync to get us out of that jam.
2856 void txQuiesce(struct super_block *sb)
2859 struct jfs_inode_info *jfs_ip;
2860 struct jfs_log *log = JFS_SBI(sb)->log;
2863 set_bit(log_QUIESCE, &log->flag);
2867 while (!list_empty(&TxAnchor.anon_list)) {
2868 jfs_ip = list_entry(TxAnchor.anon_list.next,
2869 struct jfs_inode_info,
2871 ip = &jfs_ip->vfs_inode;
2874 * inode will be removed from anonymous list
2875 * when it is committed
2878 tid = txBegin(ip->i_sb, COMMIT_INODE | COMMIT_FORCE);
2879 down(&jfs_ip->commit_sem);
2880 txCommit(tid, 1, &ip, 0);
2882 up(&jfs_ip->commit_sem);
2884 * Just to be safe. I don't know how
2885 * long we can run without blocking
2892 * If jfs_sync is running in parallel, there could be some inodes
2893 * on anon_list2. Let's check.
2895 if (!list_empty(&TxAnchor.anon_list2)) {
2896 list_splice(&TxAnchor.anon_list2, &TxAnchor.anon_list);
2897 INIT_LIST_HEAD(&TxAnchor.anon_list2);
2903 * We may need to kick off the group commit
2905 jfs_flush_journal(log, 0);
2911 * Allows transactions to start again following txQuiesce
2913 void txResume(struct super_block *sb)
2915 struct jfs_log *log = JFS_SBI(sb)->log;
2917 clear_bit(log_QUIESCE, &log->flag);
2918 TXN_WAKEUP(&log->syncwait);
2924 * To be run as a kernel daemon. This is awakened when tlocks run low.
2925 * We write any inodes that have anonymous tlocks so they will become
2928 int jfs_sync(void *arg)
2931 struct jfs_inode_info *jfs_ip;
2935 daemonize("jfsSync");
2937 complete(&jfsIOwait);
2941 * write each inode on the anonymous inode list
2944 while (jfs_tlocks_low && !list_empty(&TxAnchor.anon_list)) {
2945 jfs_ip = list_entry(TxAnchor.anon_list.next,
2946 struct jfs_inode_info,
2948 ip = &jfs_ip->vfs_inode;
2952 * Inode is being freed
2954 list_del_init(&jfs_ip->anon_inode_list);
2955 } else if (! down_trylock(&jfs_ip->commit_sem)) {
2957 * inode will be removed from anonymous list
2958 * when it is committed
2961 tid = txBegin(ip->i_sb, COMMIT_INODE);
2962 rc = txCommit(tid, 1, &ip, 0);
2964 up(&jfs_ip->commit_sem);
2968 * Just to be safe. I don't know how
2969 * long we can run without blocking
2974 /* We can't get the commit semaphore. It may
2975 * be held by a thread waiting for tlock's
2976 * so let's not block here. Save it to
2977 * put back on the anon_list.
2980 /* Take off anon_list */
2981 list_del(&jfs_ip->anon_inode_list);
2983 /* Put on anon_list2 */
2984 list_add(&jfs_ip->anon_inode_list,
2985 &TxAnchor.anon_list2);
2992 /* Add anon_list2 back to anon_list */
2993 list_splice_init(&TxAnchor.anon_list2, &TxAnchor.anon_list);
2995 if (freezing(current)) {
2999 DECLARE_WAITQUEUE(wq, current);
3001 add_wait_queue(&jfs_sync_thread_wait, &wq);
3002 set_current_state(TASK_INTERRUPTIBLE);
3005 current->state = TASK_RUNNING;
3006 remove_wait_queue(&jfs_sync_thread_wait, &wq);
3008 } while (!jfs_stop_threads);
3010 jfs_info("jfs_sync being killed");
3011 complete_and_exit(&jfsIOwait, 0);
3014 #if defined(CONFIG_PROC_FS) && defined(CONFIG_JFS_DEBUG)
3015 int jfs_txanchor_read(char *buffer, char **start, off_t offset, int length,
3016 int *eof, void *data)
3025 waitqueue_active(&TxAnchor.freewait) ? "active" : "empty";
3027 waitqueue_active(&TxAnchor.freelockwait) ? "active" : "empty";
3029 waitqueue_active(&TxAnchor.lowlockwait) ? "active" : "empty";
3031 len += sprintf(buffer,
3037 "freelockwait = %s\n"
3038 "lowlockwait = %s\n"
3039 "tlocksInUse = %d\n"
3040 "jfs_tlocks_low = %d\n"
3041 "unlock_queue is %sempty\n",
3047 TxAnchor.tlocksInUse,
3049 list_empty(&TxAnchor.unlock_queue) ? "" : "not ");
3052 *start = buffer + begin;
3067 #if defined(CONFIG_PROC_FS) && defined(CONFIG_JFS_STATISTICS)
3068 int jfs_txstats_read(char *buffer, char **start, off_t offset, int length,
3069 int *eof, void *data)
3074 len += sprintf(buffer,
3077 "calls to txBegin = %d\n"
3078 "txBegin blocked by sync barrier = %d\n"
3079 "txBegin blocked by tlocks low = %d\n"
3080 "txBegin blocked by no free tid = %d\n"
3081 "calls to txBeginAnon = %d\n"
3082 "txBeginAnon blocked by sync barrier = %d\n"
3083 "txBeginAnon blocked by tlocks low = %d\n"
3084 "calls to txLockAlloc = %d\n"
3085 "tLockAlloc blocked by no free lock = %d\n",
3087 TxStat.txBegin_barrier,
3088 TxStat.txBegin_lockslow,
3089 TxStat.txBegin_freetid,
3091 TxStat.txBeginAnon_barrier,
3092 TxStat.txBeginAnon_lockslow,
3094 TxStat.txLockAlloc_freelock);
3097 *start = buffer + begin;