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/completion.h>
48 #include <linux/freezer.h>
49 #include <linux/module.h>
50 #include <linux/moduleparam.h>
51 #include <linux/kthread.h>
52 #include <linux/seq_file.h>
53 #include "jfs_incore.h"
54 #include "jfs_inode.h"
55 #include "jfs_filsys.h"
56 #include "jfs_metapage.h"
57 #include "jfs_dinode.h"
60 #include "jfs_superblock.h"
61 #include "jfs_debug.h"
64 * transaction management structures
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 */
81 int jfs_tlocks_low; /* Indicates low number of available tlocks */
83 #ifdef CONFIG_JFS_STATISTICS
87 uint txBegin_lockslow;
90 uint txBeginAnon_barrier;
91 uint txBeginAnon_lockslow;
93 uint txLockAlloc_freelock;
97 static int nTxBlock = -1; /* number of transaction blocks */
98 module_param(nTxBlock, int, 0);
99 MODULE_PARM_DESC(nTxBlock,
100 "Number of transaction blocks (max:65536)");
102 static int nTxLock = -1; /* number of transaction locks */
103 module_param(nTxLock, int, 0);
104 MODULE_PARM_DESC(nTxLock,
105 "Number of transaction locks (max:65536)");
107 struct tblock *TxBlock; /* transaction block table */
108 static int TxLockLWM; /* Low water mark for number of txLocks used */
109 static int TxLockHWM; /* High water mark for number of txLocks used */
110 static int TxLockVHWM; /* Very High water mark */
111 struct tlock *TxLock; /* transaction lock table */
114 * transaction management lock
116 static DEFINE_SPINLOCK(jfsTxnLock);
118 #define TXN_LOCK() spin_lock(&jfsTxnLock)
119 #define TXN_UNLOCK() spin_unlock(&jfsTxnLock)
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)
125 static 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 remove_wait_queue(event, &wait);
142 #define TXN_SLEEP(event)\
144 TXN_SLEEP_DROP_LOCK(event);\
148 #define TXN_WAKEUP(event) wake_up_all(event)
154 tid_t maxtid; /* 4: biggest tid ever used */
155 lid_t maxlid; /* 4: biggest lid ever used */
156 int ntid; /* 4: # of transactions performed */
157 int nlid; /* 4: # of tlocks acquired */
158 int waitlock; /* 4: # of tlock wait */
164 static int diLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
165 struct tlock * tlck, struct commit * cd);
166 static int dataLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
167 struct tlock * tlck);
168 static void dtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
169 struct tlock * tlck);
170 static void mapLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
171 struct tlock * tlck);
172 static void txAllocPMap(struct inode *ip, struct maplock * maplock,
173 struct tblock * tblk);
174 static void txForce(struct tblock * tblk);
175 static int txLog(struct jfs_log * log, struct tblock * tblk,
177 static void txUpdateMap(struct tblock * tblk);
178 static void txRelease(struct tblock * tblk);
179 static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
180 struct tlock * tlck);
181 static void LogSyncRelease(struct metapage * mp);
184 * transaction block/lock management
185 * ---------------------------------
189 * Get a transaction lock from the free list. If the number in use is
190 * greater than the high water mark, wake up the sync daemon. This should
191 * free some anonymous transaction locks. (TXN_LOCK must be held.)
193 static lid_t txLockAlloc(void)
197 INCREMENT(TxStat.txLockAlloc);
198 if (!TxAnchor.freelock) {
199 INCREMENT(TxStat.txLockAlloc_freelock);
202 while (!(lid = TxAnchor.freelock))
203 TXN_SLEEP(&TxAnchor.freelockwait);
204 TxAnchor.freelock = TxLock[lid].next;
205 HIGHWATERMARK(stattx.maxlid, lid);
206 if ((++TxAnchor.tlocksInUse > TxLockHWM) && (jfs_tlocks_low == 0)) {
207 jfs_info("txLockAlloc tlocks low");
209 wake_up_process(jfsSyncThread);
215 static void txLockFree(lid_t lid)
218 TxLock[lid].next = TxAnchor.freelock;
219 TxAnchor.freelock = lid;
220 TxAnchor.tlocksInUse--;
221 if (jfs_tlocks_low && (TxAnchor.tlocksInUse < TxLockLWM)) {
222 jfs_info("txLockFree jfs_tlocks_low no more");
224 TXN_WAKEUP(&TxAnchor.lowlockwait);
226 TXN_WAKEUP(&TxAnchor.freelockwait);
232 * FUNCTION: initialize transaction management structures
236 * serialization: single thread at jfs_init()
243 /* Set defaults for nTxLock and nTxBlock if unset */
246 if (nTxBlock == -1) {
247 /* Base default on memory size */
249 if (si.totalram > (256 * 1024)) /* 1 GB */
252 nTxLock = si.totalram >> 2;
253 } else if (nTxBlock > (8 * 1024))
256 nTxLock = nTxBlock << 3;
259 nTxBlock = nTxLock >> 3;
261 /* Verify tunable parameters */
263 nTxBlock = 16; /* No one should set it this low */
264 if (nTxBlock > 65536)
267 nTxLock = 256; /* No one should set it this low */
271 printk(KERN_INFO "JFS: nTxBlock = %d, nTxLock = %d\n",
274 * initialize transaction block (tblock) table
276 * transaction id (tid) = tblock index
277 * tid = 0 is reserved.
279 TxLockLWM = (nTxLock * 4) / 10;
280 TxLockHWM = (nTxLock * 7) / 10;
281 TxLockVHWM = (nTxLock * 8) / 10;
283 size = sizeof(struct tblock) * nTxBlock;
284 TxBlock = vmalloc(size);
288 for (k = 1; k < nTxBlock - 1; k++) {
289 TxBlock[k].next = k + 1;
290 init_waitqueue_head(&TxBlock[k].gcwait);
291 init_waitqueue_head(&TxBlock[k].waitor);
294 init_waitqueue_head(&TxBlock[k].gcwait);
295 init_waitqueue_head(&TxBlock[k].waitor);
297 TxAnchor.freetid = 1;
298 init_waitqueue_head(&TxAnchor.freewait);
300 stattx.maxtid = 1; /* statistics */
303 * initialize transaction lock (tlock) table
305 * transaction lock id = tlock index
306 * tlock id = 0 is reserved.
308 size = sizeof(struct tlock) * nTxLock;
309 TxLock = vmalloc(size);
310 if (TxLock == NULL) {
315 /* initialize tlock table */
316 for (k = 1; k < nTxLock - 1; k++)
317 TxLock[k].next = k + 1;
319 init_waitqueue_head(&TxAnchor.freelockwait);
320 init_waitqueue_head(&TxAnchor.lowlockwait);
322 TxAnchor.freelock = 1;
323 TxAnchor.tlocksInUse = 0;
324 INIT_LIST_HEAD(&TxAnchor.anon_list);
325 INIT_LIST_HEAD(&TxAnchor.anon_list2);
328 INIT_LIST_HEAD(&TxAnchor.unlock_queue);
330 stattx.maxlid = 1; /* statistics */
338 * FUNCTION: clean up when module is unloaded
351 * FUNCTION: start a transaction.
353 * PARAMETER: sb - superblock
354 * flag - force for nested tx;
356 * RETURN: tid - transaction id
358 * note: flag force allows to start tx for nested tx
359 * to prevent deadlock on logsync barrier;
361 tid_t txBegin(struct super_block *sb, int flag)
367 jfs_info("txBegin: flag = 0x%x", flag);
368 log = JFS_SBI(sb)->log;
372 INCREMENT(TxStat.txBegin);
375 if (!(flag & COMMIT_FORCE)) {
377 * synchronize with logsync barrier
379 if (test_bit(log_SYNCBARRIER, &log->flag) ||
380 test_bit(log_QUIESCE, &log->flag)) {
381 INCREMENT(TxStat.txBegin_barrier);
382 TXN_SLEEP(&log->syncwait);
388 * Don't begin transaction if we're getting starved for tlocks
389 * unless COMMIT_FORCE or COMMIT_INODE (which may ultimately
392 if (TxAnchor.tlocksInUse > TxLockVHWM) {
393 INCREMENT(TxStat.txBegin_lockslow);
394 TXN_SLEEP(&TxAnchor.lowlockwait);
400 * allocate transaction id/block
402 if ((t = TxAnchor.freetid) == 0) {
403 jfs_info("txBegin: waiting for free tid");
404 INCREMENT(TxStat.txBegin_freetid);
405 TXN_SLEEP(&TxAnchor.freewait);
409 tblk = tid_to_tblock(t);
411 if ((tblk->next == 0) && !(flag & COMMIT_FORCE)) {
412 /* Don't let a non-forced transaction take the last tblk */
413 jfs_info("txBegin: waiting for free tid");
414 INCREMENT(TxStat.txBegin_freetid);
415 TXN_SLEEP(&TxAnchor.freewait);
419 TxAnchor.freetid = tblk->next;
422 * initialize transaction
426 * We can't zero the whole thing or we screw up another thread being
427 * awakened after sleeping on tblk->waitor
429 * memset(tblk, 0, sizeof(struct tblock));
431 tblk->next = tblk->last = tblk->xflag = tblk->flag = tblk->lsn = 0;
435 tblk->logtid = log->logtid;
439 HIGHWATERMARK(stattx.maxtid, t); /* statistics */
440 INCREMENT(stattx.ntid); /* statistics */
444 jfs_info("txBegin: returning tid = %d", t);
450 * NAME: txBeginAnon()
452 * FUNCTION: start an anonymous transaction.
453 * Blocks if logsync or available tlocks are low to prevent
454 * anonymous tlocks from depleting supply.
456 * PARAMETER: sb - superblock
460 void txBeginAnon(struct super_block *sb)
464 log = JFS_SBI(sb)->log;
467 INCREMENT(TxStat.txBeginAnon);
471 * synchronize with logsync barrier
473 if (test_bit(log_SYNCBARRIER, &log->flag) ||
474 test_bit(log_QUIESCE, &log->flag)) {
475 INCREMENT(TxStat.txBeginAnon_barrier);
476 TXN_SLEEP(&log->syncwait);
481 * Don't begin transaction if we're getting starved for tlocks
483 if (TxAnchor.tlocksInUse > TxLockVHWM) {
484 INCREMENT(TxStat.txBeginAnon_lockslow);
485 TXN_SLEEP(&TxAnchor.lowlockwait);
494 * function: free specified transaction block.
496 * logsync barrier processing:
500 void txEnd(tid_t tid)
502 struct tblock *tblk = tid_to_tblock(tid);
505 jfs_info("txEnd: tid = %d", tid);
509 * wakeup transactions waiting on the page locked
510 * by the current transaction
512 TXN_WAKEUP(&tblk->waitor);
514 log = JFS_SBI(tblk->sb)->log;
517 * Lazy commit thread can't free this guy until we mark it UNLOCKED,
518 * otherwise, we would be left with a transaction that may have been
521 * Lazy commit thread will turn off tblkGC_LAZY before calling this
524 if (tblk->flag & tblkGC_LAZY) {
525 jfs_info("txEnd called w/lazy tid: %d, tblk = 0x%p", tid, tblk);
528 spin_lock_irq(&log->gclock); // LOGGC_LOCK
529 tblk->flag |= tblkGC_UNLOCKED;
530 spin_unlock_irq(&log->gclock); // LOGGC_UNLOCK
534 jfs_info("txEnd: tid: %d, tblk = 0x%p", tid, tblk);
536 assert(tblk->next == 0);
539 * insert tblock back on freelist
541 tblk->next = TxAnchor.freetid;
542 TxAnchor.freetid = tid;
545 * mark the tblock not active
547 if (--log->active == 0) {
548 clear_bit(log_FLUSH, &log->flag);
551 * synchronize with logsync barrier
553 if (test_bit(log_SYNCBARRIER, &log->flag)) {
556 /* write dirty metadata & forward log syncpt */
559 jfs_info("log barrier off: 0x%x", log->lsn);
561 /* enable new transactions start */
562 clear_bit(log_SYNCBARRIER, &log->flag);
564 /* wakeup all waitors for logsync barrier */
565 TXN_WAKEUP(&log->syncwait);
574 * wakeup all waitors for a free tblock
576 TXN_WAKEUP(&TxAnchor.freewait);
582 * function: acquire a transaction lock on the specified <mp>
586 * return: transaction lock id
590 struct tlock *txLock(tid_t tid, struct inode *ip, struct metapage * mp,
593 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
598 struct xtlock *xtlck;
599 struct linelock *linelock;
605 if (S_ISDIR(ip->i_mode) && (type & tlckXTREE) &&
606 !(mp->xflag & COMMIT_PAGE)) {
608 * Directory inode is special. It can have both an xtree tlock
609 * and a dtree tlock associated with it.
616 /* is page not locked by a transaction ? */
620 jfs_info("txLock: tid:%d ip:0x%p mp:0x%p lid:%d", tid, ip, mp, lid);
622 /* is page locked by the requester transaction ? */
623 tlck = lid_to_tlock(lid);
624 if ((xtid = tlck->tid) == tid) {
630 * is page locked by anonymous transaction/lock ?
632 * (page update without transaction (i.e., file write) is
633 * locked under anonymous transaction tid = 0:
634 * anonymous tlocks maintained on anonymous tlock list of
635 * the inode of the page and available to all anonymous
636 * transactions until txCommit() time at which point
637 * they are transferred to the transaction tlock list of
638 * the committing transaction of the inode)
643 tblk = tid_to_tblock(tid);
645 * The order of the tlocks in the transaction is important
646 * (during truncate, child xtree pages must be freed before
647 * parent's tlocks change the working map).
648 * Take tlock off anonymous list and add to tail of
651 * Note: We really need to get rid of the tid & lid and
652 * use list_head's. This code is getting UGLY!
654 if (jfs_ip->atlhead == lid) {
655 if (jfs_ip->atltail == lid) {
656 /* only anonymous txn.
657 * Remove from anon_list
660 list_del_init(&jfs_ip->anon_inode_list);
663 jfs_ip->atlhead = tlck->next;
666 for (last = jfs_ip->atlhead;
667 lid_to_tlock(last)->next != lid;
668 last = lid_to_tlock(last)->next) {
671 lid_to_tlock(last)->next = tlck->next;
672 if (jfs_ip->atltail == lid)
673 jfs_ip->atltail = last;
676 /* insert the tlock at tail of transaction tlock list */
679 lid_to_tlock(tblk->last)->next = lid;
695 tlck = lid_to_tlock(lid);
704 /* mark tlock for meta-data page */
705 if (mp->xflag & COMMIT_PAGE) {
707 tlck->flag = tlckPAGELOCK;
709 /* mark the page dirty and nohomeok */
710 metapage_nohomeok(mp);
712 jfs_info("locking mp = 0x%p, nohomeok = %d tid = %d tlck = 0x%p",
713 mp, mp->nohomeok, tid, tlck);
715 /* if anonymous transaction, and buffer is on the group
716 * commit synclist, mark inode to show this. This will
717 * prevent the buffer from being marked nohomeok for too
720 if ((tid == 0) && mp->lsn)
721 set_cflag(COMMIT_Synclist, ip);
723 /* mark tlock for in-memory inode */
725 tlck->flag = tlckINODELOCK;
727 if (S_ISDIR(ip->i_mode))
728 tlck->flag |= tlckDIRECTORY;
732 /* bind the tlock and the page */
741 * enqueue transaction lock to transaction/inode
743 /* insert the tlock at tail of transaction tlock list */
745 tblk = tid_to_tblock(tid);
747 lid_to_tlock(tblk->last)->next = lid;
753 /* anonymous transaction:
754 * insert the tlock at head of inode anonymous tlock list
757 tlck->next = jfs_ip->atlhead;
758 jfs_ip->atlhead = lid;
759 if (tlck->next == 0) {
760 /* This inode's first anonymous transaction */
761 jfs_ip->atltail = lid;
763 list_add_tail(&jfs_ip->anon_inode_list,
764 &TxAnchor.anon_list);
769 /* initialize type dependent area for linelock */
770 linelock = (struct linelock *) & tlck->lock;
772 linelock->flag = tlckLINELOCK;
773 linelock->maxcnt = TLOCKSHORT;
776 switch (type & tlckTYPE) {
778 linelock->l2linesize = L2DTSLOTSIZE;
782 linelock->l2linesize = L2XTSLOTSIZE;
784 xtlck = (struct xtlock *) linelock;
785 xtlck->header.offset = 0;
786 xtlck->header.length = 2;
788 if (type & tlckNEW) {
789 xtlck->lwm.offset = XTENTRYSTART;
791 if (mp->xflag & COMMIT_PAGE)
792 p = (xtpage_t *) mp->data;
794 p = &jfs_ip->i_xtroot;
796 le16_to_cpu(p->header.nextindex);
798 xtlck->lwm.length = 0; /* ! */
799 xtlck->twm.offset = 0;
800 xtlck->hwm.offset = 0;
806 linelock->l2linesize = L2INODESLOTSIZE;
810 linelock->l2linesize = L2DATASLOTSIZE;
814 jfs_err("UFO tlock:0x%p", tlck);
818 * update tlock vector
826 * page is being locked by another transaction:
829 /* Only locks on ipimap or ipaimap should reach here */
830 /* assert(jfs_ip->fileset == AGGREGATE_I); */
831 if (jfs_ip->fileset != AGGREGATE_I) {
832 printk(KERN_ERR "txLock: trying to lock locked page!");
833 print_hex_dump(KERN_ERR, "ip: ", DUMP_PREFIX_ADDRESS, 16, 4,
835 print_hex_dump(KERN_ERR, "mp: ", DUMP_PREFIX_ADDRESS, 16, 4,
837 print_hex_dump(KERN_ERR, "Locker's tblock: ",
838 DUMP_PREFIX_ADDRESS, 16, 4, tid_to_tblock(tid),
839 sizeof(struct tblock), 0);
840 print_hex_dump(KERN_ERR, "Tlock: ", DUMP_PREFIX_ADDRESS, 16, 4,
841 tlck, sizeof(*tlck), 0);
844 INCREMENT(stattx.waitlock); /* statistics */
846 release_metapage(mp);
848 xtid = tlck->tid; /* reacquire after dropping TXN_LOCK */
850 jfs_info("txLock: in waitLock, tid = %d, xtid = %d, lid = %d",
853 /* Recheck everything since dropping TXN_LOCK */
854 if (xtid && (tlck->mp == mp) && (mp->lid == lid))
855 TXN_SLEEP_DROP_LOCK(&tid_to_tblock(xtid)->waitor);
858 jfs_info("txLock: awakened tid = %d, lid = %d", tid, lid);
866 * FUNCTION: Release buffers associated with transaction locks, but don't
867 * mark homeok yet. The allows other transactions to modify
868 * buffers, but won't let them go to disk until commit record
869 * actually gets written.
874 * RETURN: Errors from subroutines.
876 static void txRelease(struct tblock * tblk)
884 for (lid = tblk->next; lid; lid = tlck->next) {
885 tlck = lid_to_tlock(lid);
886 if ((mp = tlck->mp) != NULL &&
887 (tlck->type & tlckBTROOT) == 0) {
888 assert(mp->xflag & COMMIT_PAGE);
894 * wakeup transactions waiting on a page locked
895 * by the current transaction
897 TXN_WAKEUP(&tblk->waitor);
905 * FUNCTION: Initiates pageout of pages modified by tid in journalled
906 * objects and frees their lockwords.
908 static void txUnlock(struct tblock * tblk)
911 struct linelock *linelock;
912 lid_t lid, next, llid, k;
918 jfs_info("txUnlock: tblk = 0x%p", tblk);
919 log = JFS_SBI(tblk->sb)->log;
922 * mark page under tlock homeok (its log has been written):
924 for (lid = tblk->next; lid; lid = next) {
925 tlck = lid_to_tlock(lid);
928 jfs_info("unlocking lid = %d, tlck = 0x%p", lid, tlck);
930 /* unbind page from tlock */
931 if ((mp = tlck->mp) != NULL &&
932 (tlck->type & tlckBTROOT) == 0) {
933 assert(mp->xflag & COMMIT_PAGE);
939 assert(mp->nohomeok > 0);
940 _metapage_homeok(mp);
942 /* inherit younger/larger clsn */
943 LOGSYNC_LOCK(log, flags);
945 logdiff(difft, tblk->clsn, log);
946 logdiff(diffp, mp->clsn, log);
948 mp->clsn = tblk->clsn;
950 mp->clsn = tblk->clsn;
951 LOGSYNC_UNLOCK(log, flags);
953 assert(!(tlck->flag & tlckFREEPAGE));
958 /* insert tlock, and linelock(s) of the tlock if any,
959 * at head of freelist
963 llid = ((struct linelock *) & tlck->lock)->next;
965 linelock = (struct linelock *) lid_to_tlock(llid);
974 tblk->next = tblk->last = 0;
977 * remove tblock from logsynclist
978 * (allocation map pages inherited lsn of tblk and
979 * has been inserted in logsync list at txUpdateMap())
982 LOGSYNC_LOCK(log, flags);
984 list_del(&tblk->synclist);
985 LOGSYNC_UNLOCK(log, flags);
992 * function: allocate a transaction lock for freed page/entry;
993 * for freed page, maplock is used as xtlock/dtlock type;
995 struct tlock *txMaplock(tid_t tid, struct inode *ip, int type)
997 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
1001 struct maplock *maplock;
1008 lid = txLockAlloc();
1009 tlck = lid_to_tlock(lid);
1016 /* bind the tlock and the object */
1017 tlck->flag = tlckINODELOCK;
1018 if (S_ISDIR(ip->i_mode))
1019 tlck->flag |= tlckDIRECTORY;
1026 * enqueue transaction lock to transaction/inode
1028 /* insert the tlock at tail of transaction tlock list */
1030 tblk = tid_to_tblock(tid);
1032 lid_to_tlock(tblk->last)->next = lid;
1038 /* anonymous transaction:
1039 * insert the tlock at head of inode anonymous tlock list
1042 tlck->next = jfs_ip->atlhead;
1043 jfs_ip->atlhead = lid;
1044 if (tlck->next == 0) {
1045 /* This inode's first anonymous transaction */
1046 jfs_ip->atltail = lid;
1047 list_add_tail(&jfs_ip->anon_inode_list,
1048 &TxAnchor.anon_list);
1054 /* initialize type dependent area for maplock */
1055 maplock = (struct maplock *) & tlck->lock;
1057 maplock->maxcnt = 0;
1066 * function: allocate a transaction lock for log vector list
1068 struct linelock *txLinelock(struct linelock * tlock)
1072 struct linelock *linelock;
1076 /* allocate a TxLock structure */
1077 lid = txLockAlloc();
1078 tlck = lid_to_tlock(lid);
1082 /* initialize linelock */
1083 linelock = (struct linelock *) tlck;
1085 linelock->flag = tlckLINELOCK;
1086 linelock->maxcnt = TLOCKLONG;
1087 linelock->index = 0;
1088 if (tlck->flag & tlckDIRECTORY)
1089 linelock->flag |= tlckDIRECTORY;
1091 /* append linelock after tlock */
1092 linelock->next = tlock->next;
1099 * transaction commit management
1100 * -----------------------------
1106 * FUNCTION: commit the changes to the objects specified in
1107 * clist. For journalled segments only the
1108 * changes of the caller are committed, ie by tid.
1109 * for non-journalled segments the data are flushed to
1110 * disk and then the change to the disk inode and indirect
1111 * blocks committed (so blocks newly allocated to the
1112 * segment will be made a part of the segment atomically).
1114 * all of the segments specified in clist must be in
1115 * one file system. no more than 6 segments are needed
1116 * to handle all unix svcs.
1118 * if the i_nlink field (i.e. disk inode link count)
1119 * is zero, and the type of inode is a regular file or
1120 * directory, or symbolic link , the inode is truncated
1121 * to zero length. the truncation is committed but the
1122 * VM resources are unaffected until it is closed (see
1130 * on entry the inode lock on each segment is assumed
1135 int txCommit(tid_t tid, /* transaction identifier */
1136 int nip, /* number of inodes to commit */
1137 struct inode **iplist, /* list of inode to commit */
1142 struct jfs_log *log;
1143 struct tblock *tblk;
1146 struct jfs_inode_info *jfs_ip;
1149 struct super_block *sb;
1151 jfs_info("txCommit, tid = %d, flag = %d", tid, flag);
1152 /* is read-only file system ? */
1153 if (isReadOnly(iplist[0])) {
1158 sb = cd.sb = iplist[0]->i_sb;
1162 tid = txBegin(sb, 0);
1163 tblk = tid_to_tblock(tid);
1166 * initialize commit structure
1168 log = JFS_SBI(sb)->log;
1171 /* initialize log record descriptor in commit */
1173 lrd->logtid = cpu_to_le32(tblk->logtid);
1176 tblk->xflag |= flag;
1178 if ((flag & (COMMIT_FORCE | COMMIT_SYNC)) == 0)
1179 tblk->xflag |= COMMIT_LAZY;
1181 * prepare non-journaled objects for commit
1183 * flush data pages of non-journaled file
1184 * to prevent the file getting non-initialized disk blocks
1192 * acquire transaction lock on (on-disk) inodes
1194 * update on-disk inode from in-memory inode
1195 * acquiring transaction locks for AFTER records
1196 * on the on-disk inode of file object
1198 * sort the inodes array by inode number in descending order
1199 * to prevent deadlock when acquiring transaction lock
1200 * of on-disk inodes on multiple on-disk inode pages by
1201 * multiple concurrent transactions
1203 for (k = 0; k < cd.nip; k++) {
1204 top = (cd.iplist[k])->i_ino;
1205 for (n = k + 1; n < cd.nip; n++) {
1207 if (ip->i_ino > top) {
1209 cd.iplist[n] = cd.iplist[k];
1215 jfs_ip = JFS_IP(ip);
1218 * BUGBUG - This code has temporarily been removed. The
1219 * intent is to ensure that any file data is written before
1220 * the metadata is committed to the journal. This prevents
1221 * uninitialized data from appearing in a file after the
1222 * journal has been replayed. (The uninitialized data
1223 * could be sensitive data removed by another user.)
1225 * The problem now is that we are holding the IWRITELOCK
1226 * on the inode, and calling filemap_fdatawrite on an
1227 * unmapped page will cause a deadlock in jfs_get_block.
1229 * The long term solution is to pare down the use of
1230 * IWRITELOCK. We are currently holding it too long.
1231 * We could also be smarter about which data pages need
1232 * to be written before the transaction is committed and
1233 * when we don't need to worry about it at all.
1235 * if ((!S_ISDIR(ip->i_mode))
1236 * && (tblk->flag & COMMIT_DELETE) == 0)
1237 * filemap_write_and_wait(ip->i_mapping);
1241 * Mark inode as not dirty. It will still be on the dirty
1242 * inode list, but we'll know not to commit it again unless
1243 * it gets marked dirty again
1245 clear_cflag(COMMIT_Dirty, ip);
1247 /* inherit anonymous tlock(s) of inode */
1248 if (jfs_ip->atlhead) {
1249 lid_to_tlock(jfs_ip->atltail)->next = tblk->next;
1250 tblk->next = jfs_ip->atlhead;
1252 tblk->last = jfs_ip->atltail;
1253 jfs_ip->atlhead = jfs_ip->atltail = 0;
1255 list_del_init(&jfs_ip->anon_inode_list);
1260 * acquire transaction lock on on-disk inode page
1261 * (become first tlock of the tblk's tlock list)
1263 if (((rc = diWrite(tid, ip))))
1268 * write log records from transaction locks
1270 * txUpdateMap() resets XAD_NEW in XAD.
1272 if ((rc = txLog(log, tblk, &cd)))
1276 * Ensure that inode isn't reused before
1277 * lazy commit thread finishes processing
1279 if (tblk->xflag & COMMIT_DELETE) {
1282 * Avoid a rare deadlock
1284 * If the inode is locked, we may be blocked in
1285 * jfs_commit_inode. If so, we don't want the
1286 * lazy_commit thread doing the last iput() on the inode
1287 * since that may block on the locked inode. Instead,
1288 * commit the transaction synchronously, so the last iput
1289 * will be done by the calling thread (or later)
1292 * I believe this code is no longer needed. Splitting I_LOCK
1293 * into two bits, I_NEW and I_SYNC should prevent this
1294 * deadlock as well. But since I don't have a JFS testload
1295 * to verify this, only a trivial s/I_LOCK/I_SYNC/ was done.
1298 if (tblk->u.ip->i_state & I_SYNC)
1299 tblk->xflag &= ~COMMIT_LAZY;
1302 ASSERT((!(tblk->xflag & COMMIT_DELETE)) ||
1303 ((tblk->u.ip->i_nlink == 0) &&
1304 !test_cflag(COMMIT_Nolink, tblk->u.ip)));
1307 * write COMMIT log record
1309 lrd->type = cpu_to_le16(LOG_COMMIT);
1311 lmLog(log, tblk, lrd, NULL);
1313 lmGroupCommit(log, tblk);
1316 * - transaction is now committed -
1320 * force pages in careful update
1321 * (imap addressing structure update)
1323 if (flag & COMMIT_FORCE)
1327 * update allocation map.
1329 * update inode allocation map and inode:
1330 * free pager lock on memory object of inode if any.
1331 * update block allocation map.
1333 * txUpdateMap() resets XAD_NEW in XAD.
1335 if (tblk->xflag & COMMIT_FORCE)
1339 * free transaction locks and pageout/free pages
1343 if ((tblk->flag & tblkGC_LAZY) == 0)
1348 * reset in-memory object state
1350 for (k = 0; k < cd.nip; k++) {
1352 jfs_ip = JFS_IP(ip);
1355 * reset in-memory inode state
1366 jfs_info("txCommit: tid = %d, returning %d", tid, rc);
1373 * FUNCTION: Writes AFTER log records for all lines modified
1374 * by tid for segments specified by inodes in comdata.
1375 * Code assumes only WRITELOCKS are recorded in lockwords.
1381 static int txLog(struct jfs_log * log, struct tblock * tblk, struct commit * cd)
1387 struct lrd *lrd = &cd->lrd;
1390 * write log record(s) for each tlock of transaction,
1392 for (lid = tblk->next; lid; lid = tlck->next) {
1393 tlck = lid_to_tlock(lid);
1395 tlck->flag |= tlckLOG;
1397 /* initialize lrd common */
1399 lrd->aggregate = cpu_to_le32(JFS_SBI(ip->i_sb)->aggregate);
1400 lrd->log.redopage.fileset = cpu_to_le32(JFS_IP(ip)->fileset);
1401 lrd->log.redopage.inode = cpu_to_le32(ip->i_ino);
1403 /* write log record of page from the tlock */
1404 switch (tlck->type & tlckTYPE) {
1406 xtLog(log, tblk, lrd, tlck);
1410 dtLog(log, tblk, lrd, tlck);
1414 diLog(log, tblk, lrd, tlck, cd);
1418 mapLog(log, tblk, lrd, tlck);
1422 dataLog(log, tblk, lrd, tlck);
1426 jfs_err("UFO tlock:0x%p", tlck);
1436 * function: log inode tlock and format maplock to update bmap;
1438 static int diLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1439 struct tlock * tlck, struct commit * cd)
1442 struct metapage *mp;
1444 struct pxd_lock *pxdlock;
1448 /* initialize as REDOPAGE record format */
1449 lrd->log.redopage.type = cpu_to_le16(LOG_INODE);
1450 lrd->log.redopage.l2linesize = cpu_to_le16(L2INODESLOTSIZE);
1452 pxd = &lrd->log.redopage.pxd;
1457 if (tlck->type & tlckENTRY) {
1458 /* log after-image for logredo(): */
1459 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1460 PXDaddress(pxd, mp->index);
1462 mp->logical_size >> tblk->sb->s_blocksize_bits);
1463 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1465 /* mark page as homeward bound */
1466 tlck->flag |= tlckWRITEPAGE;
1467 } else if (tlck->type & tlckFREE) {
1471 * (pages of the freed inode extent have been invalidated and
1472 * a maplock for free of the extent has been formatted at
1475 * the tlock had been acquired on the inode allocation map page
1476 * (iag) that specifies the freed extent, even though the map
1477 * page is not itself logged, to prevent pageout of the map
1478 * page before the log;
1481 /* log LOG_NOREDOINOEXT of the freed inode extent for
1482 * logredo() to start NoRedoPage filters, and to update
1483 * imap and bmap for free of the extent;
1485 lrd->type = cpu_to_le16(LOG_NOREDOINOEXT);
1487 * For the LOG_NOREDOINOEXT record, we need
1488 * to pass the IAG number and inode extent
1489 * index (within that IAG) from which the
1490 * the extent being released. These have been
1491 * passed to us in the iplist[1] and iplist[2].
1493 lrd->log.noredoinoext.iagnum =
1494 cpu_to_le32((u32) (size_t) cd->iplist[1]);
1495 lrd->log.noredoinoext.inoext_idx =
1496 cpu_to_le32((u32) (size_t) cd->iplist[2]);
1498 pxdlock = (struct pxd_lock *) & tlck->lock;
1499 *pxd = pxdlock->pxd;
1500 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1503 tlck->flag |= tlckUPDATEMAP;
1505 /* mark page as homeward bound */
1506 tlck->flag |= tlckWRITEPAGE;
1508 jfs_err("diLog: UFO type tlck:0x%p", tlck);
1511 * alloc/free external EA extent
1513 * a maplock for txUpdateMap() to update bPWMAP for alloc/free
1514 * of the extent has been formatted at txLock() time;
1517 assert(tlck->type & tlckEA);
1519 /* log LOG_UPDATEMAP for logredo() to update bmap for
1520 * alloc of new (and free of old) external EA extent;
1522 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1523 pxdlock = (struct pxd_lock *) & tlck->lock;
1524 nlock = pxdlock->index;
1525 for (i = 0; i < nlock; i++, pxdlock++) {
1526 if (pxdlock->flag & mlckALLOCPXD)
1527 lrd->log.updatemap.type =
1528 cpu_to_le16(LOG_ALLOCPXD);
1530 lrd->log.updatemap.type =
1531 cpu_to_le16(LOG_FREEPXD);
1532 lrd->log.updatemap.nxd = cpu_to_le16(1);
1533 lrd->log.updatemap.pxd = pxdlock->pxd;
1535 cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1539 tlck->flag |= tlckUPDATEMAP;
1541 #endif /* _JFS_WIP */
1549 * function: log data tlock
1551 static int dataLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1552 struct tlock * tlck)
1554 struct metapage *mp;
1559 /* initialize as REDOPAGE record format */
1560 lrd->log.redopage.type = cpu_to_le16(LOG_DATA);
1561 lrd->log.redopage.l2linesize = cpu_to_le16(L2DATASLOTSIZE);
1563 pxd = &lrd->log.redopage.pxd;
1565 /* log after-image for logredo(): */
1566 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1568 if (jfs_dirtable_inline(tlck->ip)) {
1570 * The table has been truncated, we've must have deleted
1571 * the last entry, so don't bother logging this
1575 metapage_homeok(mp);
1576 discard_metapage(mp);
1581 PXDaddress(pxd, mp->index);
1582 PXDlength(pxd, mp->logical_size >> tblk->sb->s_blocksize_bits);
1584 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1586 /* mark page as homeward bound */
1587 tlck->flag |= tlckWRITEPAGE;
1595 * function: log dtree tlock and format maplock to update bmap;
1597 static void dtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1598 struct tlock * tlck)
1600 struct metapage *mp;
1601 struct pxd_lock *pxdlock;
1606 /* initialize as REDOPAGE/NOREDOPAGE record format */
1607 lrd->log.redopage.type = cpu_to_le16(LOG_DTREE);
1608 lrd->log.redopage.l2linesize = cpu_to_le16(L2DTSLOTSIZE);
1610 pxd = &lrd->log.redopage.pxd;
1612 if (tlck->type & tlckBTROOT)
1613 lrd->log.redopage.type |= cpu_to_le16(LOG_BTROOT);
1616 * page extension via relocation: entry insertion;
1617 * page extension in-place: entry insertion;
1618 * new right page from page split, reinitialized in-line
1619 * root from root page split: entry insertion;
1621 if (tlck->type & (tlckNEW | tlckEXTEND)) {
1622 /* log after-image of the new page for logredo():
1623 * mark log (LOG_NEW) for logredo() to initialize
1624 * freelist and update bmap for alloc of the new page;
1626 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1627 if (tlck->type & tlckEXTEND)
1628 lrd->log.redopage.type |= cpu_to_le16(LOG_EXTEND);
1630 lrd->log.redopage.type |= cpu_to_le16(LOG_NEW);
1631 PXDaddress(pxd, mp->index);
1633 mp->logical_size >> tblk->sb->s_blocksize_bits);
1634 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1636 /* format a maplock for txUpdateMap() to update bPMAP for
1637 * alloc of the new page;
1639 if (tlck->type & tlckBTROOT)
1641 tlck->flag |= tlckUPDATEMAP;
1642 pxdlock = (struct pxd_lock *) & tlck->lock;
1643 pxdlock->flag = mlckALLOCPXD;
1644 pxdlock->pxd = *pxd;
1648 /* mark page as homeward bound */
1649 tlck->flag |= tlckWRITEPAGE;
1654 * entry insertion/deletion,
1655 * sibling page link update (old right page before split);
1657 if (tlck->type & (tlckENTRY | tlckRELINK)) {
1658 /* log after-image for logredo(): */
1659 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1660 PXDaddress(pxd, mp->index);
1662 mp->logical_size >> tblk->sb->s_blocksize_bits);
1663 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1665 /* mark page as homeward bound */
1666 tlck->flag |= tlckWRITEPAGE;
1671 * page deletion: page has been invalidated
1672 * page relocation: source extent
1674 * a maplock for free of the page has been formatted
1675 * at txLock() time);
1677 if (tlck->type & (tlckFREE | tlckRELOCATE)) {
1678 /* log LOG_NOREDOPAGE of the deleted page for logredo()
1679 * to start NoRedoPage filter and to update bmap for free
1680 * of the deletd page
1682 lrd->type = cpu_to_le16(LOG_NOREDOPAGE);
1683 pxdlock = (struct pxd_lock *) & tlck->lock;
1684 *pxd = pxdlock->pxd;
1685 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1687 /* a maplock for txUpdateMap() for free of the page
1688 * has been formatted at txLock() time;
1690 tlck->flag |= tlckUPDATEMAP;
1698 * function: log xtree tlock and format maplock to update bmap;
1700 static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1701 struct tlock * tlck)
1704 struct metapage *mp;
1706 struct xtlock *xtlck;
1707 struct maplock *maplock;
1708 struct xdlistlock *xadlock;
1709 struct pxd_lock *pxdlock;
1716 /* initialize as REDOPAGE/NOREDOPAGE record format */
1717 lrd->log.redopage.type = cpu_to_le16(LOG_XTREE);
1718 lrd->log.redopage.l2linesize = cpu_to_le16(L2XTSLOTSIZE);
1720 page_pxd = &lrd->log.redopage.pxd;
1722 if (tlck->type & tlckBTROOT) {
1723 lrd->log.redopage.type |= cpu_to_le16(LOG_BTROOT);
1724 p = &JFS_IP(ip)->i_xtroot;
1725 if (S_ISDIR(ip->i_mode))
1726 lrd->log.redopage.type |=
1727 cpu_to_le16(LOG_DIR_XTREE);
1729 p = (xtpage_t *) mp->data;
1730 next = le16_to_cpu(p->header.nextindex);
1732 xtlck = (struct xtlock *) & tlck->lock;
1734 maplock = (struct maplock *) & tlck->lock;
1735 xadlock = (struct xdlistlock *) maplock;
1738 * entry insertion/extension;
1739 * sibling page link update (old right page before split);
1741 if (tlck->type & (tlckNEW | tlckGROW | tlckRELINK)) {
1742 /* log after-image for logredo():
1743 * logredo() will update bmap for alloc of new/extended
1744 * extents (XAD_NEW|XAD_EXTEND) of XAD[lwm:next) from
1745 * after-image of XADlist;
1746 * logredo() resets (XAD_NEW|XAD_EXTEND) flag when
1747 * applying the after-image to the meta-data page.
1749 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1750 PXDaddress(page_pxd, mp->index);
1752 mp->logical_size >> tblk->sb->s_blocksize_bits);
1753 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1755 /* format a maplock for txUpdateMap() to update bPMAP
1756 * for alloc of new/extended extents of XAD[lwm:next)
1757 * from the page itself;
1758 * txUpdateMap() resets (XAD_NEW|XAD_EXTEND) flag.
1760 lwm = xtlck->lwm.offset;
1762 lwm = XTPAGEMAXSLOT;
1767 jfs_err("xtLog: lwm > next");
1770 tlck->flag |= tlckUPDATEMAP;
1771 xadlock->flag = mlckALLOCXADLIST;
1772 xadlock->count = next - lwm;
1773 if ((xadlock->count <= 4) && (tblk->xflag & COMMIT_LAZY)) {
1777 * Lazy commit may allow xtree to be modified before
1778 * txUpdateMap runs. Copy xad into linelock to
1779 * preserve correct data.
1781 * We can fit twice as may pxd's as xads in the lock
1783 xadlock->flag = mlckALLOCPXDLIST;
1784 pxd = xadlock->xdlist = &xtlck->pxdlock;
1785 for (i = 0; i < xadlock->count; i++) {
1786 PXDaddress(pxd, addressXAD(&p->xad[lwm + i]));
1787 PXDlength(pxd, lengthXAD(&p->xad[lwm + i]));
1788 p->xad[lwm + i].flag &=
1789 ~(XAD_NEW | XAD_EXTENDED);
1794 * xdlist will point to into inode's xtree, ensure
1795 * that transaction is not committed lazily.
1797 xadlock->flag = mlckALLOCXADLIST;
1798 xadlock->xdlist = &p->xad[lwm];
1799 tblk->xflag &= ~COMMIT_LAZY;
1801 jfs_info("xtLog: alloc ip:0x%p mp:0x%p tlck:0x%p lwm:%d count:%d",
1802 tlck->ip, mp, tlck, lwm, xadlock->count);
1807 /* mark page as homeward bound */
1808 tlck->flag |= tlckWRITEPAGE;
1814 * page deletion: file deletion/truncation (ref. xtTruncate())
1816 * (page will be invalidated after log is written and bmap
1817 * is updated from the page);
1819 if (tlck->type & tlckFREE) {
1820 /* LOG_NOREDOPAGE log for NoRedoPage filter:
1821 * if page free from file delete, NoRedoFile filter from
1822 * inode image of zero link count will subsume NoRedoPage
1823 * filters for each page;
1824 * if page free from file truncattion, write NoRedoPage
1827 * upadte of block allocation map for the page itself:
1828 * if page free from deletion and truncation, LOG_UPDATEMAP
1829 * log for the page itself is generated from processing
1830 * its parent page xad entries;
1832 /* if page free from file truncation, log LOG_NOREDOPAGE
1833 * of the deleted page for logredo() to start NoRedoPage
1834 * filter for the page;
1836 if (tblk->xflag & COMMIT_TRUNCATE) {
1837 /* write NOREDOPAGE for the page */
1838 lrd->type = cpu_to_le16(LOG_NOREDOPAGE);
1839 PXDaddress(page_pxd, mp->index);
1841 mp->logical_size >> tblk->sb->
1844 cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1846 if (tlck->type & tlckBTROOT) {
1847 /* Empty xtree must be logged */
1848 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1850 cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1854 /* init LOG_UPDATEMAP of the freed extents
1855 * XAD[XTENTRYSTART:hwm) from the deleted page itself
1856 * for logredo() to update bmap;
1858 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1859 lrd->log.updatemap.type = cpu_to_le16(LOG_FREEXADLIST);
1860 xtlck = (struct xtlock *) & tlck->lock;
1861 hwm = xtlck->hwm.offset;
1862 lrd->log.updatemap.nxd =
1863 cpu_to_le16(hwm - XTENTRYSTART + 1);
1864 /* reformat linelock for lmLog() */
1865 xtlck->header.offset = XTENTRYSTART;
1866 xtlck->header.length = hwm - XTENTRYSTART + 1;
1868 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1870 /* format a maplock for txUpdateMap() to update bmap
1871 * to free extents of XAD[XTENTRYSTART:hwm) from the
1872 * deleted page itself;
1874 tlck->flag |= tlckUPDATEMAP;
1875 xadlock->count = hwm - XTENTRYSTART + 1;
1876 if ((xadlock->count <= 4) && (tblk->xflag & COMMIT_LAZY)) {
1880 * Lazy commit may allow xtree to be modified before
1881 * txUpdateMap runs. Copy xad into linelock to
1882 * preserve correct data.
1884 * We can fit twice as may pxd's as xads in the lock
1886 xadlock->flag = mlckFREEPXDLIST;
1887 pxd = xadlock->xdlist = &xtlck->pxdlock;
1888 for (i = 0; i < xadlock->count; i++) {
1890 addressXAD(&p->xad[XTENTRYSTART + i]));
1892 lengthXAD(&p->xad[XTENTRYSTART + i]));
1897 * xdlist will point to into inode's xtree, ensure
1898 * that transaction is not committed lazily.
1900 xadlock->flag = mlckFREEXADLIST;
1901 xadlock->xdlist = &p->xad[XTENTRYSTART];
1902 tblk->xflag &= ~COMMIT_LAZY;
1904 jfs_info("xtLog: free ip:0x%p mp:0x%p count:%d lwm:2",
1905 tlck->ip, mp, xadlock->count);
1909 /* mark page as invalid */
1910 if (((tblk->xflag & COMMIT_PWMAP) || S_ISDIR(ip->i_mode))
1911 && !(tlck->type & tlckBTROOT))
1912 tlck->flag |= tlckFREEPAGE;
1914 else (tblk->xflag & COMMIT_PMAP)
1921 * page/entry truncation: file truncation (ref. xtTruncate())
1923 * |----------+------+------+---------------|
1925 * | | hwm - hwm before truncation
1926 * | next - truncation point
1927 * lwm - lwm before truncation
1930 if (tlck->type & tlckTRUNCATE) {
1931 pxd_t pxd; /* truncated extent of xad */
1935 * For truncation the entire linelock may be used, so it would
1936 * be difficult to store xad list in linelock itself.
1937 * Therefore, we'll just force transaction to be committed
1938 * synchronously, so that xtree pages won't be changed before
1941 tblk->xflag &= ~COMMIT_LAZY;
1942 lwm = xtlck->lwm.offset;
1944 lwm = XTPAGEMAXSLOT;
1945 hwm = xtlck->hwm.offset;
1946 twm = xtlck->twm.offset;
1951 /* log after-image for logredo():
1953 * logredo() will update bmap for alloc of new/extended
1954 * extents (XAD_NEW|XAD_EXTEND) of XAD[lwm:next) from
1955 * after-image of XADlist;
1956 * logredo() resets (XAD_NEW|XAD_EXTEND) flag when
1957 * applying the after-image to the meta-data page.
1959 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1960 PXDaddress(page_pxd, mp->index);
1962 mp->logical_size >> tblk->sb->s_blocksize_bits);
1963 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1966 * truncate entry XAD[twm == next - 1]:
1968 if (twm == next - 1) {
1969 /* init LOG_UPDATEMAP for logredo() to update bmap for
1970 * free of truncated delta extent of the truncated
1971 * entry XAD[next - 1]:
1972 * (xtlck->pxdlock = truncated delta extent);
1974 pxdlock = (struct pxd_lock *) & xtlck->pxdlock;
1975 /* assert(pxdlock->type & tlckTRUNCATE); */
1976 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1977 lrd->log.updatemap.type = cpu_to_le16(LOG_FREEPXD);
1978 lrd->log.updatemap.nxd = cpu_to_le16(1);
1979 lrd->log.updatemap.pxd = pxdlock->pxd;
1980 pxd = pxdlock->pxd; /* save to format maplock */
1982 cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1986 * free entries XAD[next:hwm]:
1989 /* init LOG_UPDATEMAP of the freed extents
1990 * XAD[next:hwm] from the deleted page itself
1991 * for logredo() to update bmap;
1993 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1994 lrd->log.updatemap.type =
1995 cpu_to_le16(LOG_FREEXADLIST);
1996 xtlck = (struct xtlock *) & tlck->lock;
1997 hwm = xtlck->hwm.offset;
1998 lrd->log.updatemap.nxd =
1999 cpu_to_le16(hwm - next + 1);
2000 /* reformat linelock for lmLog() */
2001 xtlck->header.offset = next;
2002 xtlck->header.length = hwm - next + 1;
2005 cpu_to_le32(lmLog(log, tblk, lrd, tlck));
2009 * format maplock(s) for txUpdateMap() to update bmap
2014 * allocate entries XAD[lwm:next):
2017 /* format a maplock for txUpdateMap() to update bPMAP
2018 * for alloc of new/extended extents of XAD[lwm:next)
2019 * from the page itself;
2020 * txUpdateMap() resets (XAD_NEW|XAD_EXTEND) flag.
2022 tlck->flag |= tlckUPDATEMAP;
2023 xadlock->flag = mlckALLOCXADLIST;
2024 xadlock->count = next - lwm;
2025 xadlock->xdlist = &p->xad[lwm];
2027 jfs_info("xtLog: alloc ip:0x%p mp:0x%p count:%d lwm:%d next:%d",
2028 tlck->ip, mp, xadlock->count, lwm, next);
2034 * truncate entry XAD[twm == next - 1]:
2036 if (twm == next - 1) {
2037 /* format a maplock for txUpdateMap() to update bmap
2038 * to free truncated delta extent of the truncated
2039 * entry XAD[next - 1];
2040 * (xtlck->pxdlock = truncated delta extent);
2042 tlck->flag |= tlckUPDATEMAP;
2043 pxdlock = (struct pxd_lock *) xadlock;
2044 pxdlock->flag = mlckFREEPXD;
2048 jfs_info("xtLog: truncate ip:0x%p mp:0x%p count:%d hwm:%d",
2049 ip, mp, pxdlock->count, hwm);
2055 * free entries XAD[next:hwm]:
2058 /* format a maplock for txUpdateMap() to update bmap
2059 * to free extents of XAD[next:hwm] from thedeleted
2062 tlck->flag |= tlckUPDATEMAP;
2063 xadlock->flag = mlckFREEXADLIST;
2064 xadlock->count = hwm - next + 1;
2065 xadlock->xdlist = &p->xad[next];
2067 jfs_info("xtLog: free ip:0x%p mp:0x%p count:%d next:%d hwm:%d",
2068 tlck->ip, mp, xadlock->count, next, hwm);
2072 /* mark page as homeward bound */
2073 tlck->flag |= tlckWRITEPAGE;
2081 * function: log from maplock of freed data extents;
2083 static void mapLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
2084 struct tlock * tlck)
2086 struct pxd_lock *pxdlock;
2091 * page relocation: free the source page extent
2093 * a maplock for txUpdateMap() for free of the page
2094 * has been formatted at txLock() time saving the src
2095 * relocated page address;
2097 if (tlck->type & tlckRELOCATE) {
2098 /* log LOG_NOREDOPAGE of the old relocated page
2099 * for logredo() to start NoRedoPage filter;
2101 lrd->type = cpu_to_le16(LOG_NOREDOPAGE);
2102 pxdlock = (struct pxd_lock *) & tlck->lock;
2103 pxd = &lrd->log.redopage.pxd;
2104 *pxd = pxdlock->pxd;
2105 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
2107 /* (N.B. currently, logredo() does NOT update bmap
2108 * for free of the page itself for (LOG_XTREE|LOG_NOREDOPAGE);
2109 * if page free from relocation, LOG_UPDATEMAP log is
2110 * specifically generated now for logredo()
2111 * to update bmap for free of src relocated page;
2112 * (new flag LOG_RELOCATE may be introduced which will
2113 * inform logredo() to start NORedoPage filter and also
2114 * update block allocation map at the same time, thus
2115 * avoiding an extra log write);
2117 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
2118 lrd->log.updatemap.type = cpu_to_le16(LOG_FREEPXD);
2119 lrd->log.updatemap.nxd = cpu_to_le16(1);
2120 lrd->log.updatemap.pxd = pxdlock->pxd;
2121 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
2123 /* a maplock for txUpdateMap() for free of the page
2124 * has been formatted at txLock() time;
2126 tlck->flag |= tlckUPDATEMAP;
2131 * Otherwise it's not a relocate request
2135 /* log LOG_UPDATEMAP for logredo() to update bmap for
2136 * free of truncated/relocated delta extent of the data;
2137 * e.g.: external EA extent, relocated/truncated extent
2138 * from xtTailgate();
2140 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
2141 pxdlock = (struct pxd_lock *) & tlck->lock;
2142 nlock = pxdlock->index;
2143 for (i = 0; i < nlock; i++, pxdlock++) {
2144 if (pxdlock->flag & mlckALLOCPXD)
2145 lrd->log.updatemap.type =
2146 cpu_to_le16(LOG_ALLOCPXD);
2148 lrd->log.updatemap.type =
2149 cpu_to_le16(LOG_FREEPXD);
2150 lrd->log.updatemap.nxd = cpu_to_le16(1);
2151 lrd->log.updatemap.pxd = pxdlock->pxd;
2153 cpu_to_le32(lmLog(log, tblk, lrd, NULL));
2154 jfs_info("mapLog: xaddr:0x%lx xlen:0x%x",
2155 (ulong) addressPXD(&pxdlock->pxd),
2156 lengthPXD(&pxdlock->pxd));
2160 tlck->flag |= tlckUPDATEMAP;
2167 * function: acquire maplock for EA/ACL extents or
2168 * set COMMIT_INLINE flag;
2170 void txEA(tid_t tid, struct inode *ip, dxd_t * oldea, dxd_t * newea)
2172 struct tlock *tlck = NULL;
2173 struct pxd_lock *maplock = NULL, *pxdlock = NULL;
2176 * format maplock for alloc of new EA extent
2179 /* Since the newea could be a completely zeroed entry we need to
2180 * check for the two flags which indicate we should actually
2181 * commit new EA data
2183 if (newea->flag & DXD_EXTENT) {
2184 tlck = txMaplock(tid, ip, tlckMAP);
2185 maplock = (struct pxd_lock *) & tlck->lock;
2186 pxdlock = (struct pxd_lock *) maplock;
2187 pxdlock->flag = mlckALLOCPXD;
2188 PXDaddress(&pxdlock->pxd, addressDXD(newea));
2189 PXDlength(&pxdlock->pxd, lengthDXD(newea));
2192 } else if (newea->flag & DXD_INLINE) {
2195 set_cflag(COMMIT_Inlineea, ip);
2200 * format maplock for free of old EA extent
2202 if (!test_cflag(COMMIT_Nolink, ip) && oldea->flag & DXD_EXTENT) {
2204 tlck = txMaplock(tid, ip, tlckMAP);
2205 maplock = (struct pxd_lock *) & tlck->lock;
2206 pxdlock = (struct pxd_lock *) maplock;
2209 pxdlock->flag = mlckFREEPXD;
2210 PXDaddress(&pxdlock->pxd, addressDXD(oldea));
2211 PXDlength(&pxdlock->pxd, lengthDXD(oldea));
2219 * function: synchronously write pages locked by transaction
2220 * after txLog() but before txUpdateMap();
2222 static void txForce(struct tblock * tblk)
2226 struct metapage *mp;
2229 * reverse the order of transaction tlocks in
2230 * careful update order of address index pages
2231 * (right to left, bottom up)
2233 tlck = lid_to_tlock(tblk->next);
2237 tlck = lid_to_tlock(lid);
2239 tlck->next = tblk->next;
2245 * synchronously write the page, and
2246 * hold the page for txUpdateMap();
2248 for (lid = tblk->next; lid; lid = next) {
2249 tlck = lid_to_tlock(lid);
2252 if ((mp = tlck->mp) != NULL &&
2253 (tlck->type & tlckBTROOT) == 0) {
2254 assert(mp->xflag & COMMIT_PAGE);
2256 if (tlck->flag & tlckWRITEPAGE) {
2257 tlck->flag &= ~tlckWRITEPAGE;
2259 /* do not release page to freelist */
2263 * The "right" thing to do here is to
2264 * synchronously write the metadata.
2265 * With the current implementation this
2266 * is hard since write_metapage requires
2267 * us to kunmap & remap the page. If we
2268 * have tlocks pointing into the metadata
2269 * pages, we don't want to do this. I think
2270 * we can get by with synchronously writing
2271 * the pages when they are released.
2273 assert(mp->nohomeok);
2274 set_bit(META_dirty, &mp->flag);
2275 set_bit(META_sync, &mp->flag);
2285 * function: update persistent allocation map (and working map
2290 static void txUpdateMap(struct tblock * tblk)
2293 struct inode *ipimap;
2296 struct maplock *maplock;
2297 struct pxd_lock pxdlock;
2300 struct metapage *mp = NULL;
2302 ipimap = JFS_SBI(tblk->sb)->ipimap;
2304 maptype = (tblk->xflag & COMMIT_PMAP) ? COMMIT_PMAP : COMMIT_PWMAP;
2308 * update block allocation map
2310 * update allocation state in pmap (and wmap) and
2311 * update lsn of the pmap page;
2314 * scan each tlock/page of transaction for block allocation/free:
2316 * for each tlock/page of transaction, update map.
2317 * ? are there tlock for pmap and pwmap at the same time ?
2319 for (lid = tblk->next; lid; lid = tlck->next) {
2320 tlck = lid_to_tlock(lid);
2322 if ((tlck->flag & tlckUPDATEMAP) == 0)
2325 if (tlck->flag & tlckFREEPAGE) {
2327 * Another thread may attempt to reuse freed space
2328 * immediately, so we want to get rid of the metapage
2329 * before anyone else has a chance to get it.
2330 * Lock metapage, update maps, then invalidate
2334 ASSERT(mp->xflag & COMMIT_PAGE);
2340 * . in-line PXD list:
2341 * . out-of-line XAD list:
2343 maplock = (struct maplock *) & tlck->lock;
2344 nlock = maplock->index;
2346 for (k = 0; k < nlock; k++, maplock++) {
2348 * allocate blocks in persistent map:
2350 * blocks have been allocated from wmap at alloc time;
2352 if (maplock->flag & mlckALLOC) {
2353 txAllocPMap(ipimap, maplock, tblk);
2356 * free blocks in persistent and working map:
2357 * blocks will be freed in pmap and then in wmap;
2359 * ? tblock specifies the PMAP/PWMAP based upon
2362 * free blocks in persistent map:
2363 * blocks will be freed from wmap at last reference
2364 * release of the object for regular files;
2366 * Alway free blocks from both persistent & working
2367 * maps for directories
2369 else { /* (maplock->flag & mlckFREE) */
2371 if (tlck->flag & tlckDIRECTORY)
2372 txFreeMap(ipimap, maplock,
2373 tblk, COMMIT_PWMAP);
2375 txFreeMap(ipimap, maplock,
2379 if (tlck->flag & tlckFREEPAGE) {
2380 if (!(tblk->flag & tblkGC_LAZY)) {
2381 /* This is equivalent to txRelease */
2382 ASSERT(mp->lid == lid);
2385 assert(mp->nohomeok == 1);
2386 metapage_homeok(mp);
2387 discard_metapage(mp);
2392 * update inode allocation map
2394 * update allocation state in pmap and
2395 * update lsn of the pmap page;
2396 * update in-memory inode flag/state
2398 * unlock mapper/write lock
2400 if (tblk->xflag & COMMIT_CREATE) {
2401 diUpdatePMap(ipimap, tblk->ino, false, tblk);
2402 /* update persistent block allocation map
2403 * for the allocation of inode extent;
2405 pxdlock.flag = mlckALLOCPXD;
2406 pxdlock.pxd = tblk->u.ixpxd;
2408 txAllocPMap(ipimap, (struct maplock *) & pxdlock, tblk);
2409 } else if (tblk->xflag & COMMIT_DELETE) {
2411 diUpdatePMap(ipimap, ip->i_ino, true, tblk);
2419 * function: allocate from persistent map;
2428 * allocate from persistent map;
2429 * free from persistent map;
2430 * (e.g., tmp file - free from working map at releae
2431 * of last reference);
2432 * free from persistent and working map;
2434 * lsn - log sequence number;
2436 static void txAllocPMap(struct inode *ip, struct maplock * maplock,
2437 struct tblock * tblk)
2439 struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
2440 struct xdlistlock *xadlistlock;
2444 struct pxd_lock *pxdlock;
2445 struct xdlistlock *pxdlistlock;
2450 * allocate from persistent map;
2452 if (maplock->flag & mlckALLOCXADLIST) {
2453 xadlistlock = (struct xdlistlock *) maplock;
2454 xad = xadlistlock->xdlist;
2455 for (n = 0; n < xadlistlock->count; n++, xad++) {
2456 if (xad->flag & (XAD_NEW | XAD_EXTENDED)) {
2457 xaddr = addressXAD(xad);
2458 xlen = lengthXAD(xad);
2459 dbUpdatePMap(ipbmap, false, xaddr,
2461 xad->flag &= ~(XAD_NEW | XAD_EXTENDED);
2462 jfs_info("allocPMap: xaddr:0x%lx xlen:%d",
2463 (ulong) xaddr, xlen);
2466 } else if (maplock->flag & mlckALLOCPXD) {
2467 pxdlock = (struct pxd_lock *) maplock;
2468 xaddr = addressPXD(&pxdlock->pxd);
2469 xlen = lengthPXD(&pxdlock->pxd);
2470 dbUpdatePMap(ipbmap, false, xaddr, (s64) xlen, tblk);
2471 jfs_info("allocPMap: xaddr:0x%lx xlen:%d", (ulong) xaddr, xlen);
2472 } else { /* (maplock->flag & mlckALLOCPXDLIST) */
2474 pxdlistlock = (struct xdlistlock *) maplock;
2475 pxd = pxdlistlock->xdlist;
2476 for (n = 0; n < pxdlistlock->count; n++, pxd++) {
2477 xaddr = addressPXD(pxd);
2478 xlen = lengthPXD(pxd);
2479 dbUpdatePMap(ipbmap, false, xaddr, (s64) xlen,
2481 jfs_info("allocPMap: xaddr:0x%lx xlen:%d",
2482 (ulong) xaddr, xlen);
2490 * function: free from persistent and/or working map;
2492 * todo: optimization
2494 void txFreeMap(struct inode *ip,
2495 struct maplock * maplock, struct tblock * tblk, int maptype)
2497 struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
2498 struct xdlistlock *xadlistlock;
2502 struct pxd_lock *pxdlock;
2503 struct xdlistlock *pxdlistlock;
2507 jfs_info("txFreeMap: tblk:0x%p maplock:0x%p maptype:0x%x",
2508 tblk, maplock, maptype);
2511 * free from persistent map;
2513 if (maptype == COMMIT_PMAP || maptype == COMMIT_PWMAP) {
2514 if (maplock->flag & mlckFREEXADLIST) {
2515 xadlistlock = (struct xdlistlock *) maplock;
2516 xad = xadlistlock->xdlist;
2517 for (n = 0; n < xadlistlock->count; n++, xad++) {
2518 if (!(xad->flag & XAD_NEW)) {
2519 xaddr = addressXAD(xad);
2520 xlen = lengthXAD(xad);
2521 dbUpdatePMap(ipbmap, true, xaddr,
2523 jfs_info("freePMap: xaddr:0x%lx xlen:%d",
2524 (ulong) xaddr, xlen);
2527 } else if (maplock->flag & mlckFREEPXD) {
2528 pxdlock = (struct pxd_lock *) maplock;
2529 xaddr = addressPXD(&pxdlock->pxd);
2530 xlen = lengthPXD(&pxdlock->pxd);
2531 dbUpdatePMap(ipbmap, true, xaddr, (s64) xlen,
2533 jfs_info("freePMap: xaddr:0x%lx xlen:%d",
2534 (ulong) xaddr, xlen);
2535 } else { /* (maplock->flag & mlckALLOCPXDLIST) */
2537 pxdlistlock = (struct xdlistlock *) maplock;
2538 pxd = pxdlistlock->xdlist;
2539 for (n = 0; n < pxdlistlock->count; n++, pxd++) {
2540 xaddr = addressPXD(pxd);
2541 xlen = lengthPXD(pxd);
2542 dbUpdatePMap(ipbmap, true, xaddr,
2544 jfs_info("freePMap: xaddr:0x%lx xlen:%d",
2545 (ulong) xaddr, xlen);
2551 * free from working map;
2553 if (maptype == COMMIT_PWMAP || maptype == COMMIT_WMAP) {
2554 if (maplock->flag & mlckFREEXADLIST) {
2555 xadlistlock = (struct xdlistlock *) maplock;
2556 xad = xadlistlock->xdlist;
2557 for (n = 0; n < xadlistlock->count; n++, xad++) {
2558 xaddr = addressXAD(xad);
2559 xlen = lengthXAD(xad);
2560 dbFree(ip, xaddr, (s64) xlen);
2562 jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
2563 (ulong) xaddr, xlen);
2565 } else if (maplock->flag & mlckFREEPXD) {
2566 pxdlock = (struct pxd_lock *) maplock;
2567 xaddr = addressPXD(&pxdlock->pxd);
2568 xlen = lengthPXD(&pxdlock->pxd);
2569 dbFree(ip, xaddr, (s64) xlen);
2570 jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
2571 (ulong) xaddr, xlen);
2572 } else { /* (maplock->flag & mlckFREEPXDLIST) */
2574 pxdlistlock = (struct xdlistlock *) maplock;
2575 pxd = pxdlistlock->xdlist;
2576 for (n = 0; n < pxdlistlock->count; n++, pxd++) {
2577 xaddr = addressPXD(pxd);
2578 xlen = lengthPXD(pxd);
2579 dbFree(ip, xaddr, (s64) xlen);
2580 jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
2581 (ulong) xaddr, xlen);
2590 * function: remove tlock from inode anonymous locklist
2592 void txFreelock(struct inode *ip)
2594 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
2595 struct tlock *xtlck, *tlck;
2596 lid_t xlid = 0, lid;
2598 if (!jfs_ip->atlhead)
2602 xtlck = (struct tlock *) &jfs_ip->atlhead;
2604 while ((lid = xtlck->next) != 0) {
2605 tlck = lid_to_tlock(lid);
2606 if (tlck->flag & tlckFREELOCK) {
2607 xtlck->next = tlck->next;
2615 if (jfs_ip->atlhead)
2616 jfs_ip->atltail = xlid;
2618 jfs_ip->atltail = 0;
2620 * If inode was on anon_list, remove it
2622 list_del_init(&jfs_ip->anon_inode_list);
2630 * function: abort tx before commit;
2632 * frees line-locks and segment locks for all
2633 * segments in comdata structure.
2634 * Optionally sets state of file-system to FM_DIRTY in super-block.
2635 * log age of page-frames in memory for which caller has
2636 * are reset to 0 (to avoid logwarap).
2638 void txAbort(tid_t tid, int dirty)
2641 struct metapage *mp;
2642 struct tblock *tblk = tid_to_tblock(tid);
2646 * free tlocks of the transaction
2648 for (lid = tblk->next; lid; lid = next) {
2649 tlck = lid_to_tlock(lid);
2652 JFS_IP(tlck->ip)->xtlid = 0;
2658 * reset lsn of page to avoid logwarap:
2660 * (page may have been previously committed by another
2661 * transaction(s) but has not been paged, i.e.,
2662 * it may be on logsync list even though it has not
2663 * been logged for the current tx.)
2665 if (mp->xflag & COMMIT_PAGE && mp->lsn)
2668 /* insert tlock at head of freelist */
2674 /* caller will free the transaction block */
2676 tblk->next = tblk->last = 0;
2679 * mark filesystem dirty
2682 jfs_error(tblk->sb, "\n");
2688 * txLazyCommit(void)
2690 * All transactions except those changing ipimap (COMMIT_FORCE) are
2691 * processed by this routine. This insures that the inode and block
2692 * allocation maps are updated in order. For synchronous transactions,
2693 * let the user thread finish processing after txUpdateMap() is called.
2695 static void txLazyCommit(struct tblock * tblk)
2697 struct jfs_log *log;
2699 while (((tblk->flag & tblkGC_READY) == 0) &&
2700 ((tblk->flag & tblkGC_UNLOCKED) == 0)) {
2701 /* We must have gotten ahead of the user thread
2703 jfs_info("jfs_lazycommit: tblk 0x%p not unlocked", tblk);
2707 jfs_info("txLazyCommit: processing tblk 0x%p", tblk);
2711 log = (struct jfs_log *) JFS_SBI(tblk->sb)->log;
2713 spin_lock_irq(&log->gclock); // LOGGC_LOCK
2715 tblk->flag |= tblkGC_COMMITTED;
2717 if (tblk->flag & tblkGC_READY)
2720 wake_up_all(&tblk->gcwait); // LOGGC_WAKEUP
2723 * Can't release log->gclock until we've tested tblk->flag
2725 if (tblk->flag & tblkGC_LAZY) {
2726 spin_unlock_irq(&log->gclock); // LOGGC_UNLOCK
2728 tblk->flag &= ~tblkGC_LAZY;
2729 txEnd(tblk - TxBlock); /* Convert back to tid */
2731 spin_unlock_irq(&log->gclock); // LOGGC_UNLOCK
2733 jfs_info("txLazyCommit: done: tblk = 0x%p", tblk);
2737 * jfs_lazycommit(void)
2739 * To be run as a kernel daemon. If lbmIODone is called in an interrupt
2740 * context, or where blocking is not wanted, this routine will process
2741 * committed transactions from the unlock queue.
2743 int jfs_lazycommit(void *arg)
2746 struct tblock *tblk;
2747 unsigned long flags;
2748 struct jfs_sb_info *sbi;
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 remove_wait_queue(&jfs_commit_thread_wait, &wq);
2808 } while (!kthread_should_stop());
2810 if (!list_empty(&TxAnchor.unlock_queue))
2811 jfs_err("jfs_lazycommit being killed w/pending transactions!");
2813 jfs_info("jfs_lazycommit being killed");
2817 void txLazyUnlock(struct tblock * tblk)
2819 unsigned long flags;
2823 list_add_tail(&tblk->cqueue, &TxAnchor.unlock_queue);
2825 * Don't wake up a commit thread if there is already one servicing
2826 * this superblock, or if the last one we woke up hasn't started yet.
2828 if (!(JFS_SBI(tblk->sb)->commit_state & IN_LAZYCOMMIT) &&
2829 !jfs_commit_thread_waking) {
2830 jfs_commit_thread_waking = 1;
2831 wake_up(&jfs_commit_thread_wait);
2836 static void LogSyncRelease(struct metapage * mp)
2838 struct jfs_log *log = mp->log;
2840 assert(mp->nohomeok);
2842 metapage_homeok(mp);
2848 * Block all new transactions and push anonymous transactions to
2851 * This does almost the same thing as jfs_sync below. We don't
2852 * worry about deadlocking when jfs_tlocks_low is set, since we would
2853 * expect jfs_sync to get us out of that jam.
2855 void txQuiesce(struct super_block *sb)
2858 struct jfs_inode_info *jfs_ip;
2859 struct jfs_log *log = JFS_SBI(sb)->log;
2862 set_bit(log_QUIESCE, &log->flag);
2866 while (!list_empty(&TxAnchor.anon_list)) {
2867 jfs_ip = list_entry(TxAnchor.anon_list.next,
2868 struct jfs_inode_info,
2870 ip = &jfs_ip->vfs_inode;
2873 * inode will be removed from anonymous list
2874 * when it is committed
2877 tid = txBegin(ip->i_sb, COMMIT_INODE | COMMIT_FORCE);
2878 mutex_lock(&jfs_ip->commit_mutex);
2879 txCommit(tid, 1, &ip, 0);
2881 mutex_unlock(&jfs_ip->commit_mutex);
2883 * Just to be safe. I don't know how
2884 * long we can run without blocking
2891 * If jfs_sync is running in parallel, there could be some inodes
2892 * on anon_list2. Let's check.
2894 if (!list_empty(&TxAnchor.anon_list2)) {
2895 list_splice_init(&TxAnchor.anon_list2, &TxAnchor.anon_list);
2901 * We may need to kick off the group commit
2903 jfs_flush_journal(log, 0);
2909 * Allows transactions to start again following txQuiesce
2911 void txResume(struct super_block *sb)
2913 struct jfs_log *log = JFS_SBI(sb)->log;
2915 clear_bit(log_QUIESCE, &log->flag);
2916 TXN_WAKEUP(&log->syncwait);
2922 * To be run as a kernel daemon. This is awakened when tlocks run low.
2923 * We write any inodes that have anonymous tlocks so they will become
2926 int jfs_sync(void *arg)
2929 struct jfs_inode_info *jfs_ip;
2934 * write each inode on the anonymous inode list
2937 while (jfs_tlocks_low && !list_empty(&TxAnchor.anon_list)) {
2938 jfs_ip = list_entry(TxAnchor.anon_list.next,
2939 struct jfs_inode_info,
2941 ip = &jfs_ip->vfs_inode;
2945 * Inode is being freed
2947 list_del_init(&jfs_ip->anon_inode_list);
2948 } else if (mutex_trylock(&jfs_ip->commit_mutex)) {
2950 * inode will be removed from anonymous list
2951 * when it is committed
2954 tid = txBegin(ip->i_sb, COMMIT_INODE);
2955 txCommit(tid, 1, &ip, 0);
2957 mutex_unlock(&jfs_ip->commit_mutex);
2961 * Just to be safe. I don't know how
2962 * long we can run without blocking
2967 /* We can't get the commit mutex. It may
2968 * be held by a thread waiting for tlock's
2969 * so let's not block here. Save it to
2970 * put back on the anon_list.
2973 /* Move from anon_list to anon_list2 */
2974 list_move(&jfs_ip->anon_inode_list,
2975 &TxAnchor.anon_list2);
2982 /* Add anon_list2 back to anon_list */
2983 list_splice_init(&TxAnchor.anon_list2, &TxAnchor.anon_list);
2985 if (freezing(current)) {
2989 set_current_state(TASK_INTERRUPTIBLE);
2993 } while (!kthread_should_stop());
2995 jfs_info("jfs_sync being killed");
2999 #if defined(CONFIG_PROC_FS) && defined(CONFIG_JFS_DEBUG)
3000 int jfs_txanchor_proc_show(struct seq_file *m, void *v)
3007 waitqueue_active(&TxAnchor.freewait) ? "active" : "empty";
3009 waitqueue_active(&TxAnchor.freelockwait) ? "active" : "empty";
3011 waitqueue_active(&TxAnchor.lowlockwait) ? "active" : "empty";
3019 "freelockwait = %s\n"
3020 "lowlockwait = %s\n"
3021 "tlocksInUse = %d\n"
3022 "jfs_tlocks_low = %d\n"
3023 "unlock_queue is %sempty\n",
3029 TxAnchor.tlocksInUse,
3031 list_empty(&TxAnchor.unlock_queue) ? "" : "not ");
3036 #if defined(CONFIG_PROC_FS) && defined(CONFIG_JFS_STATISTICS)
3037 int jfs_txstats_proc_show(struct seq_file *m, void *v)
3042 "calls to txBegin = %d\n"
3043 "txBegin blocked by sync barrier = %d\n"
3044 "txBegin blocked by tlocks low = %d\n"
3045 "txBegin blocked by no free tid = %d\n"
3046 "calls to txBeginAnon = %d\n"
3047 "txBeginAnon blocked by sync barrier = %d\n"
3048 "txBeginAnon blocked by tlocks low = %d\n"
3049 "calls to txLockAlloc = %d\n"
3050 "tLockAlloc blocked by no free lock = %d\n",
3052 TxStat.txBegin_barrier,
3053 TxStat.txBegin_lockslow,
3054 TxStat.txBegin_freetid,
3056 TxStat.txBeginAnon_barrier,
3057 TxStat.txBeginAnon_lockslow,
3059 TxStat.txLockAlloc_freelock);