1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Copyright (C) International Business Machines Corp., 2000-2005
4 * Portions Copyright (C) Christoph Hellwig, 2001-2002
8 * jfs_txnmgr.c: transaction manager
11 * transaction starts with txBegin() and ends with txCommit()
14 * tlock is acquired at the time of update;
15 * (obviate scan at commit time for xtree and dtree)
16 * tlock and mp points to each other;
17 * (no hashlist for mp -> tlock).
20 * tlock on in-memory inode:
21 * in-place tlock in the in-memory inode itself;
22 * converted to page lock by iWrite() at commit time.
24 * tlock during write()/mmap() under anonymous transaction (tid = 0):
25 * transferred (?) to transaction at commit time.
27 * use the page itself to update allocation maps
28 * (obviate intermediate replication of allocation/deallocation data)
29 * hold on to mp+lock thru update of maps
33 #include <linux/vmalloc.h>
34 #include <linux/completion.h>
35 #include <linux/freezer.h>
36 #include <linux/module.h>
37 #include <linux/moduleparam.h>
38 #include <linux/kthread.h>
39 #include <linux/seq_file.h>
40 #include "jfs_incore.h"
41 #include "jfs_inode.h"
42 #include "jfs_filsys.h"
43 #include "jfs_metapage.h"
44 #include "jfs_dinode.h"
47 #include "jfs_superblock.h"
48 #include "jfs_debug.h"
51 * transaction management structures
54 int freetid; /* index of a free tid structure */
55 int freelock; /* index first free lock word */
56 wait_queue_head_t freewait; /* eventlist of free tblock */
57 wait_queue_head_t freelockwait; /* eventlist of free tlock */
58 wait_queue_head_t lowlockwait; /* eventlist of ample tlocks */
59 int tlocksInUse; /* Number of tlocks in use */
60 spinlock_t LazyLock; /* synchronize sync_queue & unlock_queue */
61 /* struct tblock *sync_queue; * Transactions waiting for data sync */
62 struct list_head unlock_queue; /* Txns waiting to be released */
63 struct list_head anon_list; /* inodes having anonymous txns */
64 struct list_head anon_list2; /* inodes having anonymous txns
65 that couldn't be sync'ed */
68 int jfs_tlocks_low; /* Indicates low number of available tlocks */
70 #ifdef CONFIG_JFS_STATISTICS
74 uint txBegin_lockslow;
77 uint txBeginAnon_barrier;
78 uint txBeginAnon_lockslow;
80 uint txLockAlloc_freelock;
84 static int nTxBlock = -1; /* number of transaction blocks */
85 module_param(nTxBlock, int, 0);
86 MODULE_PARM_DESC(nTxBlock,
87 "Number of transaction blocks (max:65536)");
89 static int nTxLock = -1; /* number of transaction locks */
90 module_param(nTxLock, int, 0);
91 MODULE_PARM_DESC(nTxLock,
92 "Number of transaction locks (max:65536)");
94 struct tblock *TxBlock; /* transaction block table */
95 static int TxLockLWM; /* Low water mark for number of txLocks used */
96 static int TxLockHWM; /* High water mark for number of txLocks used */
97 static int TxLockVHWM; /* Very High water mark */
98 struct tlock *TxLock; /* transaction lock table */
101 * transaction management lock
103 static DEFINE_SPINLOCK(jfsTxnLock);
105 #define TXN_LOCK() spin_lock(&jfsTxnLock)
106 #define TXN_UNLOCK() spin_unlock(&jfsTxnLock)
108 #define LAZY_LOCK_INIT() spin_lock_init(&TxAnchor.LazyLock)
109 #define LAZY_LOCK(flags) spin_lock_irqsave(&TxAnchor.LazyLock, flags)
110 #define LAZY_UNLOCK(flags) spin_unlock_irqrestore(&TxAnchor.LazyLock, flags)
112 static DECLARE_WAIT_QUEUE_HEAD(jfs_commit_thread_wait);
113 static int jfs_commit_thread_waking;
116 * Retry logic exist outside these macros to protect from spurrious wakeups.
118 static inline void TXN_SLEEP_DROP_LOCK(wait_queue_head_t * event)
120 DECLARE_WAITQUEUE(wait, current);
122 add_wait_queue(event, &wait);
123 set_current_state(TASK_UNINTERRUPTIBLE);
126 remove_wait_queue(event, &wait);
129 #define TXN_SLEEP(event)\
131 TXN_SLEEP_DROP_LOCK(event);\
135 #define TXN_WAKEUP(event) wake_up_all(event)
141 tid_t maxtid; /* 4: biggest tid ever used */
142 lid_t maxlid; /* 4: biggest lid ever used */
143 int ntid; /* 4: # of transactions performed */
144 int nlid; /* 4: # of tlocks acquired */
145 int waitlock; /* 4: # of tlock wait */
151 static void diLog(struct jfs_log *log, struct tblock *tblk, struct lrd *lrd,
152 struct tlock *tlck, struct commit *cd);
153 static void dataLog(struct jfs_log *log, struct tblock *tblk, struct lrd *lrd,
155 static void dtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
156 struct tlock * tlck);
157 static void mapLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
158 struct tlock * tlck);
159 static void txAllocPMap(struct inode *ip, struct maplock * maplock,
160 struct tblock * tblk);
161 static void txForce(struct tblock * tblk);
162 static void txLog(struct jfs_log *log, struct tblock *tblk,
164 static void txUpdateMap(struct tblock * tblk);
165 static void txRelease(struct tblock * tblk);
166 static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
167 struct tlock * tlck);
168 static void LogSyncRelease(struct metapage * mp);
171 * transaction block/lock management
172 * ---------------------------------
176 * Get a transaction lock from the free list. If the number in use is
177 * greater than the high water mark, wake up the sync daemon. This should
178 * free some anonymous transaction locks. (TXN_LOCK must be held.)
180 static lid_t txLockAlloc(void)
184 INCREMENT(TxStat.txLockAlloc);
185 if (!TxAnchor.freelock) {
186 INCREMENT(TxStat.txLockAlloc_freelock);
189 while (!(lid = TxAnchor.freelock))
190 TXN_SLEEP(&TxAnchor.freelockwait);
191 TxAnchor.freelock = TxLock[lid].next;
192 HIGHWATERMARK(stattx.maxlid, lid);
193 if ((++TxAnchor.tlocksInUse > TxLockHWM) && (jfs_tlocks_low == 0)) {
194 jfs_info("txLockAlloc tlocks low");
196 wake_up_process(jfsSyncThread);
202 static void txLockFree(lid_t lid)
205 TxLock[lid].next = TxAnchor.freelock;
206 TxAnchor.freelock = lid;
207 TxAnchor.tlocksInUse--;
208 if (jfs_tlocks_low && (TxAnchor.tlocksInUse < TxLockLWM)) {
209 jfs_info("txLockFree jfs_tlocks_low no more");
211 TXN_WAKEUP(&TxAnchor.lowlockwait);
213 TXN_WAKEUP(&TxAnchor.freelockwait);
219 * FUNCTION: initialize transaction management structures
223 * serialization: single thread at jfs_init()
230 /* Set defaults for nTxLock and nTxBlock if unset */
233 if (nTxBlock == -1) {
234 /* Base default on memory size */
236 if (si.totalram > (256 * 1024)) /* 1 GB */
239 nTxLock = si.totalram >> 2;
240 } else if (nTxBlock > (8 * 1024))
243 nTxLock = nTxBlock << 3;
246 nTxBlock = nTxLock >> 3;
248 /* Verify tunable parameters */
250 nTxBlock = 16; /* No one should set it this low */
251 if (nTxBlock > 65536)
254 nTxLock = 256; /* No one should set it this low */
258 printk(KERN_INFO "JFS: nTxBlock = %d, nTxLock = %d\n",
261 * initialize transaction block (tblock) table
263 * transaction id (tid) = tblock index
264 * tid = 0 is reserved.
266 TxLockLWM = (nTxLock * 4) / 10;
267 TxLockHWM = (nTxLock * 7) / 10;
268 TxLockVHWM = (nTxLock * 8) / 10;
270 size = sizeof(struct tblock) * nTxBlock;
271 TxBlock = vmalloc(size);
275 for (k = 1; k < nTxBlock - 1; k++) {
276 TxBlock[k].next = k + 1;
277 init_waitqueue_head(&TxBlock[k].gcwait);
278 init_waitqueue_head(&TxBlock[k].waitor);
281 init_waitqueue_head(&TxBlock[k].gcwait);
282 init_waitqueue_head(&TxBlock[k].waitor);
284 TxAnchor.freetid = 1;
285 init_waitqueue_head(&TxAnchor.freewait);
287 stattx.maxtid = 1; /* statistics */
290 * initialize transaction lock (tlock) table
292 * transaction lock id = tlock index
293 * tlock id = 0 is reserved.
295 size = sizeof(struct tlock) * nTxLock;
296 TxLock = vmalloc(size);
297 if (TxLock == NULL) {
302 /* initialize tlock table */
303 for (k = 1; k < nTxLock - 1; k++)
304 TxLock[k].next = k + 1;
306 init_waitqueue_head(&TxAnchor.freelockwait);
307 init_waitqueue_head(&TxAnchor.lowlockwait);
309 TxAnchor.freelock = 1;
310 TxAnchor.tlocksInUse = 0;
311 INIT_LIST_HEAD(&TxAnchor.anon_list);
312 INIT_LIST_HEAD(&TxAnchor.anon_list2);
315 INIT_LIST_HEAD(&TxAnchor.unlock_queue);
317 stattx.maxlid = 1; /* statistics */
325 * FUNCTION: clean up when module is unloaded
338 * FUNCTION: start a transaction.
340 * PARAMETER: sb - superblock
341 * flag - force for nested tx;
343 * RETURN: tid - transaction id
345 * note: flag force allows to start tx for nested tx
346 * to prevent deadlock on logsync barrier;
348 tid_t txBegin(struct super_block *sb, int flag)
354 jfs_info("txBegin: flag = 0x%x", flag);
355 log = JFS_SBI(sb)->log;
359 INCREMENT(TxStat.txBegin);
362 if (!(flag & COMMIT_FORCE)) {
364 * synchronize with logsync barrier
366 if (test_bit(log_SYNCBARRIER, &log->flag) ||
367 test_bit(log_QUIESCE, &log->flag)) {
368 INCREMENT(TxStat.txBegin_barrier);
369 TXN_SLEEP(&log->syncwait);
375 * Don't begin transaction if we're getting starved for tlocks
376 * unless COMMIT_FORCE or COMMIT_INODE (which may ultimately
379 if (TxAnchor.tlocksInUse > TxLockVHWM) {
380 INCREMENT(TxStat.txBegin_lockslow);
381 TXN_SLEEP(&TxAnchor.lowlockwait);
387 * allocate transaction id/block
389 if ((t = TxAnchor.freetid) == 0) {
390 jfs_info("txBegin: waiting for free tid");
391 INCREMENT(TxStat.txBegin_freetid);
392 TXN_SLEEP(&TxAnchor.freewait);
396 tblk = tid_to_tblock(t);
398 if ((tblk->next == 0) && !(flag & COMMIT_FORCE)) {
399 /* Don't let a non-forced transaction take the last tblk */
400 jfs_info("txBegin: waiting for free tid");
401 INCREMENT(TxStat.txBegin_freetid);
402 TXN_SLEEP(&TxAnchor.freewait);
406 TxAnchor.freetid = tblk->next;
409 * initialize transaction
413 * We can't zero the whole thing or we screw up another thread being
414 * awakened after sleeping on tblk->waitor
416 * memset(tblk, 0, sizeof(struct tblock));
418 tblk->next = tblk->last = tblk->xflag = tblk->flag = tblk->lsn = 0;
422 tblk->logtid = log->logtid;
426 HIGHWATERMARK(stattx.maxtid, t); /* statistics */
427 INCREMENT(stattx.ntid); /* statistics */
431 jfs_info("txBegin: returning tid = %d", t);
437 * NAME: txBeginAnon()
439 * FUNCTION: start an anonymous transaction.
440 * Blocks if logsync or available tlocks are low to prevent
441 * anonymous tlocks from depleting supply.
443 * PARAMETER: sb - superblock
447 void txBeginAnon(struct super_block *sb)
451 log = JFS_SBI(sb)->log;
454 INCREMENT(TxStat.txBeginAnon);
458 * synchronize with logsync barrier
460 if (test_bit(log_SYNCBARRIER, &log->flag) ||
461 test_bit(log_QUIESCE, &log->flag)) {
462 INCREMENT(TxStat.txBeginAnon_barrier);
463 TXN_SLEEP(&log->syncwait);
468 * Don't begin transaction if we're getting starved for tlocks
470 if (TxAnchor.tlocksInUse > TxLockVHWM) {
471 INCREMENT(TxStat.txBeginAnon_lockslow);
472 TXN_SLEEP(&TxAnchor.lowlockwait);
481 * function: free specified transaction block.
483 * logsync barrier processing:
487 void txEnd(tid_t tid)
489 struct tblock *tblk = tid_to_tblock(tid);
492 jfs_info("txEnd: tid = %d", tid);
496 * wakeup transactions waiting on the page locked
497 * by the current transaction
499 TXN_WAKEUP(&tblk->waitor);
501 log = JFS_SBI(tblk->sb)->log;
504 * Lazy commit thread can't free this guy until we mark it UNLOCKED,
505 * otherwise, we would be left with a transaction that may have been
508 * Lazy commit thread will turn off tblkGC_LAZY before calling this
511 if (tblk->flag & tblkGC_LAZY) {
512 jfs_info("txEnd called w/lazy tid: %d, tblk = 0x%p", tid, tblk);
515 spin_lock_irq(&log->gclock); // LOGGC_LOCK
516 tblk->flag |= tblkGC_UNLOCKED;
517 spin_unlock_irq(&log->gclock); // LOGGC_UNLOCK
521 jfs_info("txEnd: tid: %d, tblk = 0x%p", tid, tblk);
523 assert(tblk->next == 0);
526 * insert tblock back on freelist
528 tblk->next = TxAnchor.freetid;
529 TxAnchor.freetid = tid;
532 * mark the tblock not active
534 if (--log->active == 0) {
535 clear_bit(log_FLUSH, &log->flag);
538 * synchronize with logsync barrier
540 if (test_bit(log_SYNCBARRIER, &log->flag)) {
543 /* write dirty metadata & forward log syncpt */
546 jfs_info("log barrier off: 0x%x", log->lsn);
548 /* enable new transactions start */
549 clear_bit(log_SYNCBARRIER, &log->flag);
551 /* wakeup all waitors for logsync barrier */
552 TXN_WAKEUP(&log->syncwait);
561 * wakeup all waitors for a free tblock
563 TXN_WAKEUP(&TxAnchor.freewait);
569 * function: acquire a transaction lock on the specified <mp>
573 * return: transaction lock id
577 struct tlock *txLock(tid_t tid, struct inode *ip, struct metapage * mp,
580 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
585 struct xtlock *xtlck;
586 struct linelock *linelock;
592 if (S_ISDIR(ip->i_mode) && (type & tlckXTREE) &&
593 !(mp->xflag & COMMIT_PAGE)) {
595 * Directory inode is special. It can have both an xtree tlock
596 * and a dtree tlock associated with it.
603 /* is page not locked by a transaction ? */
607 jfs_info("txLock: tid:%d ip:0x%p mp:0x%p lid:%d", tid, ip, mp, lid);
609 /* is page locked by the requester transaction ? */
610 tlck = lid_to_tlock(lid);
611 if ((xtid = tlck->tid) == tid) {
617 * is page locked by anonymous transaction/lock ?
619 * (page update without transaction (i.e., file write) is
620 * locked under anonymous transaction tid = 0:
621 * anonymous tlocks maintained on anonymous tlock list of
622 * the inode of the page and available to all anonymous
623 * transactions until txCommit() time at which point
624 * they are transferred to the transaction tlock list of
625 * the committing transaction of the inode)
630 tblk = tid_to_tblock(tid);
632 * The order of the tlocks in the transaction is important
633 * (during truncate, child xtree pages must be freed before
634 * parent's tlocks change the working map).
635 * Take tlock off anonymous list and add to tail of
638 * Note: We really need to get rid of the tid & lid and
639 * use list_head's. This code is getting UGLY!
641 if (jfs_ip->atlhead == lid) {
642 if (jfs_ip->atltail == lid) {
643 /* only anonymous txn.
644 * Remove from anon_list
647 list_del_init(&jfs_ip->anon_inode_list);
650 jfs_ip->atlhead = tlck->next;
653 for (last = jfs_ip->atlhead;
654 lid_to_tlock(last)->next != lid;
655 last = lid_to_tlock(last)->next) {
658 lid_to_tlock(last)->next = tlck->next;
659 if (jfs_ip->atltail == lid)
660 jfs_ip->atltail = last;
663 /* insert the tlock at tail of transaction tlock list */
666 lid_to_tlock(tblk->last)->next = lid;
682 tlck = lid_to_tlock(lid);
691 /* mark tlock for meta-data page */
692 if (mp->xflag & COMMIT_PAGE) {
694 tlck->flag = tlckPAGELOCK;
696 /* mark the page dirty and nohomeok */
697 metapage_nohomeok(mp);
699 jfs_info("locking mp = 0x%p, nohomeok = %d tid = %d tlck = 0x%p",
700 mp, mp->nohomeok, tid, tlck);
702 /* if anonymous transaction, and buffer is on the group
703 * commit synclist, mark inode to show this. This will
704 * prevent the buffer from being marked nohomeok for too
707 if ((tid == 0) && mp->lsn)
708 set_cflag(COMMIT_Synclist, ip);
710 /* mark tlock for in-memory inode */
712 tlck->flag = tlckINODELOCK;
714 if (S_ISDIR(ip->i_mode))
715 tlck->flag |= tlckDIRECTORY;
719 /* bind the tlock and the page */
728 * enqueue transaction lock to transaction/inode
730 /* insert the tlock at tail of transaction tlock list */
732 tblk = tid_to_tblock(tid);
734 lid_to_tlock(tblk->last)->next = lid;
740 /* anonymous transaction:
741 * insert the tlock at head of inode anonymous tlock list
744 tlck->next = jfs_ip->atlhead;
745 jfs_ip->atlhead = lid;
746 if (tlck->next == 0) {
747 /* This inode's first anonymous transaction */
748 jfs_ip->atltail = lid;
750 list_add_tail(&jfs_ip->anon_inode_list,
751 &TxAnchor.anon_list);
756 /* initialize type dependent area for linelock */
757 linelock = (struct linelock *) & tlck->lock;
759 linelock->flag = tlckLINELOCK;
760 linelock->maxcnt = TLOCKSHORT;
763 switch (type & tlckTYPE) {
765 linelock->l2linesize = L2DTSLOTSIZE;
769 linelock->l2linesize = L2XTSLOTSIZE;
771 xtlck = (struct xtlock *) linelock;
772 xtlck->header.offset = 0;
773 xtlck->header.length = 2;
775 if (type & tlckNEW) {
776 xtlck->lwm.offset = XTENTRYSTART;
778 if (mp->xflag & COMMIT_PAGE)
779 p = (xtpage_t *) mp->data;
781 p = &jfs_ip->i_xtroot;
783 le16_to_cpu(p->header.nextindex);
785 xtlck->lwm.length = 0; /* ! */
786 xtlck->twm.offset = 0;
787 xtlck->hwm.offset = 0;
793 linelock->l2linesize = L2INODESLOTSIZE;
797 linelock->l2linesize = L2DATASLOTSIZE;
801 jfs_err("UFO tlock:0x%p", tlck);
805 * update tlock vector
813 * page is being locked by another transaction:
816 /* Only locks on ipimap or ipaimap should reach here */
817 /* assert(jfs_ip->fileset == AGGREGATE_I); */
818 if (jfs_ip->fileset != AGGREGATE_I) {
819 printk(KERN_ERR "txLock: trying to lock locked page!");
820 print_hex_dump(KERN_ERR, "ip: ", DUMP_PREFIX_ADDRESS, 16, 4,
822 print_hex_dump(KERN_ERR, "mp: ", DUMP_PREFIX_ADDRESS, 16, 4,
824 print_hex_dump(KERN_ERR, "Locker's tblock: ",
825 DUMP_PREFIX_ADDRESS, 16, 4, tid_to_tblock(tid),
826 sizeof(struct tblock), 0);
827 print_hex_dump(KERN_ERR, "Tlock: ", DUMP_PREFIX_ADDRESS, 16, 4,
828 tlck, sizeof(*tlck), 0);
831 INCREMENT(stattx.waitlock); /* statistics */
833 release_metapage(mp);
835 xtid = tlck->tid; /* reacquire after dropping TXN_LOCK */
837 jfs_info("txLock: in waitLock, tid = %d, xtid = %d, lid = %d",
840 /* Recheck everything since dropping TXN_LOCK */
841 if (xtid && (tlck->mp == mp) && (mp->lid == lid))
842 TXN_SLEEP_DROP_LOCK(&tid_to_tblock(xtid)->waitor);
845 jfs_info("txLock: awakened tid = %d, lid = %d", tid, lid);
853 * FUNCTION: Release buffers associated with transaction locks, but don't
854 * mark homeok yet. The allows other transactions to modify
855 * buffers, but won't let them go to disk until commit record
856 * actually gets written.
861 * RETURN: Errors from subroutines.
863 static void txRelease(struct tblock * tblk)
871 for (lid = tblk->next; lid; lid = tlck->next) {
872 tlck = lid_to_tlock(lid);
873 if ((mp = tlck->mp) != NULL &&
874 (tlck->type & tlckBTROOT) == 0) {
875 assert(mp->xflag & COMMIT_PAGE);
881 * wakeup transactions waiting on a page locked
882 * by the current transaction
884 TXN_WAKEUP(&tblk->waitor);
892 * FUNCTION: Initiates pageout of pages modified by tid in journalled
893 * objects and frees their lockwords.
895 static void txUnlock(struct tblock * tblk)
898 struct linelock *linelock;
899 lid_t lid, next, llid, k;
905 jfs_info("txUnlock: tblk = 0x%p", tblk);
906 log = JFS_SBI(tblk->sb)->log;
909 * mark page under tlock homeok (its log has been written):
911 for (lid = tblk->next; lid; lid = next) {
912 tlck = lid_to_tlock(lid);
915 jfs_info("unlocking lid = %d, tlck = 0x%p", lid, tlck);
917 /* unbind page from tlock */
918 if ((mp = tlck->mp) != NULL &&
919 (tlck->type & tlckBTROOT) == 0) {
920 assert(mp->xflag & COMMIT_PAGE);
926 assert(mp->nohomeok > 0);
927 _metapage_homeok(mp);
929 /* inherit younger/larger clsn */
930 LOGSYNC_LOCK(log, flags);
932 logdiff(difft, tblk->clsn, log);
933 logdiff(diffp, mp->clsn, log);
935 mp->clsn = tblk->clsn;
937 mp->clsn = tblk->clsn;
938 LOGSYNC_UNLOCK(log, flags);
940 assert(!(tlck->flag & tlckFREEPAGE));
945 /* insert tlock, and linelock(s) of the tlock if any,
946 * at head of freelist
950 llid = ((struct linelock *) & tlck->lock)->next;
952 linelock = (struct linelock *) lid_to_tlock(llid);
961 tblk->next = tblk->last = 0;
964 * remove tblock from logsynclist
965 * (allocation map pages inherited lsn of tblk and
966 * has been inserted in logsync list at txUpdateMap())
969 LOGSYNC_LOCK(log, flags);
971 list_del(&tblk->synclist);
972 LOGSYNC_UNLOCK(log, flags);
979 * function: allocate a transaction lock for freed page/entry;
980 * for freed page, maplock is used as xtlock/dtlock type;
982 struct tlock *txMaplock(tid_t tid, struct inode *ip, int type)
984 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
988 struct maplock *maplock;
996 tlck = lid_to_tlock(lid);
1003 /* bind the tlock and the object */
1004 tlck->flag = tlckINODELOCK;
1005 if (S_ISDIR(ip->i_mode))
1006 tlck->flag |= tlckDIRECTORY;
1013 * enqueue transaction lock to transaction/inode
1015 /* insert the tlock at tail of transaction tlock list */
1017 tblk = tid_to_tblock(tid);
1019 lid_to_tlock(tblk->last)->next = lid;
1025 /* anonymous transaction:
1026 * insert the tlock at head of inode anonymous tlock list
1029 tlck->next = jfs_ip->atlhead;
1030 jfs_ip->atlhead = lid;
1031 if (tlck->next == 0) {
1032 /* This inode's first anonymous transaction */
1033 jfs_ip->atltail = lid;
1034 list_add_tail(&jfs_ip->anon_inode_list,
1035 &TxAnchor.anon_list);
1041 /* initialize type dependent area for maplock */
1042 maplock = (struct maplock *) & tlck->lock;
1044 maplock->maxcnt = 0;
1053 * function: allocate a transaction lock for log vector list
1055 struct linelock *txLinelock(struct linelock * tlock)
1059 struct linelock *linelock;
1063 /* allocate a TxLock structure */
1064 lid = txLockAlloc();
1065 tlck = lid_to_tlock(lid);
1069 /* initialize linelock */
1070 linelock = (struct linelock *) tlck;
1072 linelock->flag = tlckLINELOCK;
1073 linelock->maxcnt = TLOCKLONG;
1074 linelock->index = 0;
1075 if (tlck->flag & tlckDIRECTORY)
1076 linelock->flag |= tlckDIRECTORY;
1078 /* append linelock after tlock */
1079 linelock->next = tlock->next;
1086 * transaction commit management
1087 * -----------------------------
1093 * FUNCTION: commit the changes to the objects specified in
1094 * clist. For journalled segments only the
1095 * changes of the caller are committed, ie by tid.
1096 * for non-journalled segments the data are flushed to
1097 * disk and then the change to the disk inode and indirect
1098 * blocks committed (so blocks newly allocated to the
1099 * segment will be made a part of the segment atomically).
1101 * all of the segments specified in clist must be in
1102 * one file system. no more than 6 segments are needed
1103 * to handle all unix svcs.
1105 * if the i_nlink field (i.e. disk inode link count)
1106 * is zero, and the type of inode is a regular file or
1107 * directory, or symbolic link , the inode is truncated
1108 * to zero length. the truncation is committed but the
1109 * VM resources are unaffected until it is closed (see
1117 * on entry the inode lock on each segment is assumed
1122 int txCommit(tid_t tid, /* transaction identifier */
1123 int nip, /* number of inodes to commit */
1124 struct inode **iplist, /* list of inode to commit */
1129 struct jfs_log *log;
1130 struct tblock *tblk;
1133 struct jfs_inode_info *jfs_ip;
1136 struct super_block *sb;
1138 jfs_info("txCommit, tid = %d, flag = %d", tid, flag);
1139 /* is read-only file system ? */
1140 if (isReadOnly(iplist[0])) {
1145 sb = cd.sb = iplist[0]->i_sb;
1149 tid = txBegin(sb, 0);
1150 tblk = tid_to_tblock(tid);
1153 * initialize commit structure
1155 log = JFS_SBI(sb)->log;
1158 /* initialize log record descriptor in commit */
1160 lrd->logtid = cpu_to_le32(tblk->logtid);
1163 tblk->xflag |= flag;
1165 if ((flag & (COMMIT_FORCE | COMMIT_SYNC)) == 0)
1166 tblk->xflag |= COMMIT_LAZY;
1168 * prepare non-journaled objects for commit
1170 * flush data pages of non-journaled file
1171 * to prevent the file getting non-initialized disk blocks
1179 * acquire transaction lock on (on-disk) inodes
1181 * update on-disk inode from in-memory inode
1182 * acquiring transaction locks for AFTER records
1183 * on the on-disk inode of file object
1185 * sort the inodes array by inode number in descending order
1186 * to prevent deadlock when acquiring transaction lock
1187 * of on-disk inodes on multiple on-disk inode pages by
1188 * multiple concurrent transactions
1190 for (k = 0; k < cd.nip; k++) {
1191 top = (cd.iplist[k])->i_ino;
1192 for (n = k + 1; n < cd.nip; n++) {
1194 if (ip->i_ino > top) {
1196 cd.iplist[n] = cd.iplist[k];
1202 jfs_ip = JFS_IP(ip);
1205 * BUGBUG - This code has temporarily been removed. The
1206 * intent is to ensure that any file data is written before
1207 * the metadata is committed to the journal. This prevents
1208 * uninitialized data from appearing in a file after the
1209 * journal has been replayed. (The uninitialized data
1210 * could be sensitive data removed by another user.)
1212 * The problem now is that we are holding the IWRITELOCK
1213 * on the inode, and calling filemap_fdatawrite on an
1214 * unmapped page will cause a deadlock in jfs_get_block.
1216 * The long term solution is to pare down the use of
1217 * IWRITELOCK. We are currently holding it too long.
1218 * We could also be smarter about which data pages need
1219 * to be written before the transaction is committed and
1220 * when we don't need to worry about it at all.
1222 * if ((!S_ISDIR(ip->i_mode))
1223 * && (tblk->flag & COMMIT_DELETE) == 0)
1224 * filemap_write_and_wait(ip->i_mapping);
1228 * Mark inode as not dirty. It will still be on the dirty
1229 * inode list, but we'll know not to commit it again unless
1230 * it gets marked dirty again
1232 clear_cflag(COMMIT_Dirty, ip);
1234 /* inherit anonymous tlock(s) of inode */
1235 if (jfs_ip->atlhead) {
1236 lid_to_tlock(jfs_ip->atltail)->next = tblk->next;
1237 tblk->next = jfs_ip->atlhead;
1239 tblk->last = jfs_ip->atltail;
1240 jfs_ip->atlhead = jfs_ip->atltail = 0;
1242 list_del_init(&jfs_ip->anon_inode_list);
1247 * acquire transaction lock on on-disk inode page
1248 * (become first tlock of the tblk's tlock list)
1250 if (((rc = diWrite(tid, ip))))
1255 * write log records from transaction locks
1257 * txUpdateMap() resets XAD_NEW in XAD.
1259 txLog(log, tblk, &cd);
1262 * Ensure that inode isn't reused before
1263 * lazy commit thread finishes processing
1265 if (tblk->xflag & COMMIT_DELETE) {
1268 * Avoid a rare deadlock
1270 * If the inode is locked, we may be blocked in
1271 * jfs_commit_inode. If so, we don't want the
1272 * lazy_commit thread doing the last iput() on the inode
1273 * since that may block on the locked inode. Instead,
1274 * commit the transaction synchronously, so the last iput
1275 * will be done by the calling thread (or later)
1278 * I believe this code is no longer needed. Splitting I_LOCK
1279 * into two bits, I_NEW and I_SYNC should prevent this
1280 * deadlock as well. But since I don't have a JFS testload
1281 * to verify this, only a trivial s/I_LOCK/I_SYNC/ was done.
1284 if (tblk->u.ip->i_state & I_SYNC)
1285 tblk->xflag &= ~COMMIT_LAZY;
1288 ASSERT((!(tblk->xflag & COMMIT_DELETE)) ||
1289 ((tblk->u.ip->i_nlink == 0) &&
1290 !test_cflag(COMMIT_Nolink, tblk->u.ip)));
1293 * write COMMIT log record
1295 lrd->type = cpu_to_le16(LOG_COMMIT);
1297 lmLog(log, tblk, lrd, NULL);
1299 lmGroupCommit(log, tblk);
1302 * - transaction is now committed -
1306 * force pages in careful update
1307 * (imap addressing structure update)
1309 if (flag & COMMIT_FORCE)
1313 * update allocation map.
1315 * update inode allocation map and inode:
1316 * free pager lock on memory object of inode if any.
1317 * update block allocation map.
1319 * txUpdateMap() resets XAD_NEW in XAD.
1321 if (tblk->xflag & COMMIT_FORCE)
1325 * free transaction locks and pageout/free pages
1329 if ((tblk->flag & tblkGC_LAZY) == 0)
1334 * reset in-memory object state
1336 for (k = 0; k < cd.nip; k++) {
1338 jfs_ip = JFS_IP(ip);
1341 * reset in-memory inode state
1352 jfs_info("txCommit: tid = %d, returning %d", tid, rc);
1359 * FUNCTION: Writes AFTER log records for all lines modified
1360 * by tid for segments specified by inodes in comdata.
1361 * Code assumes only WRITELOCKS are recorded in lockwords.
1367 static void txLog(struct jfs_log *log, struct tblock *tblk, struct commit *cd)
1372 struct lrd *lrd = &cd->lrd;
1375 * write log record(s) for each tlock of transaction,
1377 for (lid = tblk->next; lid; lid = tlck->next) {
1378 tlck = lid_to_tlock(lid);
1380 tlck->flag |= tlckLOG;
1382 /* initialize lrd common */
1384 lrd->aggregate = cpu_to_le32(JFS_SBI(ip->i_sb)->aggregate);
1385 lrd->log.redopage.fileset = cpu_to_le32(JFS_IP(ip)->fileset);
1386 lrd->log.redopage.inode = cpu_to_le32(ip->i_ino);
1388 /* write log record of page from the tlock */
1389 switch (tlck->type & tlckTYPE) {
1391 xtLog(log, tblk, lrd, tlck);
1395 dtLog(log, tblk, lrd, tlck);
1399 diLog(log, tblk, lrd, tlck, cd);
1403 mapLog(log, tblk, lrd, tlck);
1407 dataLog(log, tblk, lrd, tlck);
1411 jfs_err("UFO tlock:0x%p", tlck);
1421 * function: log inode tlock and format maplock to update bmap;
1423 static void diLog(struct jfs_log *log, struct tblock *tblk, struct lrd *lrd,
1424 struct tlock *tlck, struct commit *cd)
1426 struct metapage *mp;
1428 struct pxd_lock *pxdlock;
1432 /* initialize as REDOPAGE record format */
1433 lrd->log.redopage.type = cpu_to_le16(LOG_INODE);
1434 lrd->log.redopage.l2linesize = cpu_to_le16(L2INODESLOTSIZE);
1436 pxd = &lrd->log.redopage.pxd;
1441 if (tlck->type & tlckENTRY) {
1442 /* log after-image for logredo(): */
1443 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1444 PXDaddress(pxd, mp->index);
1446 mp->logical_size >> tblk->sb->s_blocksize_bits);
1447 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1449 /* mark page as homeward bound */
1450 tlck->flag |= tlckWRITEPAGE;
1451 } else if (tlck->type & tlckFREE) {
1455 * (pages of the freed inode extent have been invalidated and
1456 * a maplock for free of the extent has been formatted at
1459 * the tlock had been acquired on the inode allocation map page
1460 * (iag) that specifies the freed extent, even though the map
1461 * page is not itself logged, to prevent pageout of the map
1462 * page before the log;
1465 /* log LOG_NOREDOINOEXT of the freed inode extent for
1466 * logredo() to start NoRedoPage filters, and to update
1467 * imap and bmap for free of the extent;
1469 lrd->type = cpu_to_le16(LOG_NOREDOINOEXT);
1471 * For the LOG_NOREDOINOEXT record, we need
1472 * to pass the IAG number and inode extent
1473 * index (within that IAG) from which the
1474 * extent is being released. These have been
1475 * passed to us in the iplist[1] and iplist[2].
1477 lrd->log.noredoinoext.iagnum =
1478 cpu_to_le32((u32) (size_t) cd->iplist[1]);
1479 lrd->log.noredoinoext.inoext_idx =
1480 cpu_to_le32((u32) (size_t) cd->iplist[2]);
1482 pxdlock = (struct pxd_lock *) & tlck->lock;
1483 *pxd = pxdlock->pxd;
1484 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1487 tlck->flag |= tlckUPDATEMAP;
1489 /* mark page as homeward bound */
1490 tlck->flag |= tlckWRITEPAGE;
1492 jfs_err("diLog: UFO type tlck:0x%p", tlck);
1499 * function: log data tlock
1501 static void dataLog(struct jfs_log *log, struct tblock *tblk, struct lrd *lrd,
1504 struct metapage *mp;
1509 /* initialize as REDOPAGE record format */
1510 lrd->log.redopage.type = cpu_to_le16(LOG_DATA);
1511 lrd->log.redopage.l2linesize = cpu_to_le16(L2DATASLOTSIZE);
1513 pxd = &lrd->log.redopage.pxd;
1515 /* log after-image for logredo(): */
1516 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1518 if (jfs_dirtable_inline(tlck->ip)) {
1520 * The table has been truncated, we've must have deleted
1521 * the last entry, so don't bother logging this
1525 metapage_homeok(mp);
1526 discard_metapage(mp);
1531 PXDaddress(pxd, mp->index);
1532 PXDlength(pxd, mp->logical_size >> tblk->sb->s_blocksize_bits);
1534 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1536 /* mark page as homeward bound */
1537 tlck->flag |= tlckWRITEPAGE;
1545 * function: log dtree tlock and format maplock to update bmap;
1547 static void dtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1548 struct tlock * tlck)
1550 struct metapage *mp;
1551 struct pxd_lock *pxdlock;
1556 /* initialize as REDOPAGE/NOREDOPAGE record format */
1557 lrd->log.redopage.type = cpu_to_le16(LOG_DTREE);
1558 lrd->log.redopage.l2linesize = cpu_to_le16(L2DTSLOTSIZE);
1560 pxd = &lrd->log.redopage.pxd;
1562 if (tlck->type & tlckBTROOT)
1563 lrd->log.redopage.type |= cpu_to_le16(LOG_BTROOT);
1566 * page extension via relocation: entry insertion;
1567 * page extension in-place: entry insertion;
1568 * new right page from page split, reinitialized in-line
1569 * root from root page split: entry insertion;
1571 if (tlck->type & (tlckNEW | tlckEXTEND)) {
1572 /* log after-image of the new page for logredo():
1573 * mark log (LOG_NEW) for logredo() to initialize
1574 * freelist and update bmap for alloc of the new page;
1576 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1577 if (tlck->type & tlckEXTEND)
1578 lrd->log.redopage.type |= cpu_to_le16(LOG_EXTEND);
1580 lrd->log.redopage.type |= cpu_to_le16(LOG_NEW);
1581 PXDaddress(pxd, mp->index);
1583 mp->logical_size >> tblk->sb->s_blocksize_bits);
1584 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1586 /* format a maplock for txUpdateMap() to update bPMAP for
1587 * alloc of the new page;
1589 if (tlck->type & tlckBTROOT)
1591 tlck->flag |= tlckUPDATEMAP;
1592 pxdlock = (struct pxd_lock *) & tlck->lock;
1593 pxdlock->flag = mlckALLOCPXD;
1594 pxdlock->pxd = *pxd;
1598 /* mark page as homeward bound */
1599 tlck->flag |= tlckWRITEPAGE;
1604 * entry insertion/deletion,
1605 * sibling page link update (old right page before split);
1607 if (tlck->type & (tlckENTRY | tlckRELINK)) {
1608 /* log after-image for logredo(): */
1609 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1610 PXDaddress(pxd, mp->index);
1612 mp->logical_size >> tblk->sb->s_blocksize_bits);
1613 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1615 /* mark page as homeward bound */
1616 tlck->flag |= tlckWRITEPAGE;
1621 * page deletion: page has been invalidated
1622 * page relocation: source extent
1624 * a maplock for free of the page has been formatted
1625 * at txLock() time);
1627 if (tlck->type & (tlckFREE | tlckRELOCATE)) {
1628 /* log LOG_NOREDOPAGE of the deleted page for logredo()
1629 * to start NoRedoPage filter and to update bmap for free
1630 * of the deletd page
1632 lrd->type = cpu_to_le16(LOG_NOREDOPAGE);
1633 pxdlock = (struct pxd_lock *) & tlck->lock;
1634 *pxd = pxdlock->pxd;
1635 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1637 /* a maplock for txUpdateMap() for free of the page
1638 * has been formatted at txLock() time;
1640 tlck->flag |= tlckUPDATEMAP;
1648 * function: log xtree tlock and format maplock to update bmap;
1650 static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1651 struct tlock * tlck)
1654 struct metapage *mp;
1656 struct xtlock *xtlck;
1657 struct maplock *maplock;
1658 struct xdlistlock *xadlock;
1659 struct pxd_lock *pxdlock;
1666 /* initialize as REDOPAGE/NOREDOPAGE record format */
1667 lrd->log.redopage.type = cpu_to_le16(LOG_XTREE);
1668 lrd->log.redopage.l2linesize = cpu_to_le16(L2XTSLOTSIZE);
1670 page_pxd = &lrd->log.redopage.pxd;
1672 if (tlck->type & tlckBTROOT) {
1673 lrd->log.redopage.type |= cpu_to_le16(LOG_BTROOT);
1674 p = &JFS_IP(ip)->i_xtroot;
1675 if (S_ISDIR(ip->i_mode))
1676 lrd->log.redopage.type |=
1677 cpu_to_le16(LOG_DIR_XTREE);
1679 p = (xtpage_t *) mp->data;
1680 next = le16_to_cpu(p->header.nextindex);
1682 xtlck = (struct xtlock *) & tlck->lock;
1684 maplock = (struct maplock *) & tlck->lock;
1685 xadlock = (struct xdlistlock *) maplock;
1688 * entry insertion/extension;
1689 * sibling page link update (old right page before split);
1691 if (tlck->type & (tlckNEW | tlckGROW | tlckRELINK)) {
1692 /* log after-image for logredo():
1693 * logredo() will update bmap for alloc of new/extended
1694 * extents (XAD_NEW|XAD_EXTEND) of XAD[lwm:next) from
1695 * after-image of XADlist;
1696 * logredo() resets (XAD_NEW|XAD_EXTEND) flag when
1697 * applying the after-image to the meta-data page.
1699 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1700 PXDaddress(page_pxd, mp->index);
1702 mp->logical_size >> tblk->sb->s_blocksize_bits);
1703 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1705 /* format a maplock for txUpdateMap() to update bPMAP
1706 * for alloc of new/extended extents of XAD[lwm:next)
1707 * from the page itself;
1708 * txUpdateMap() resets (XAD_NEW|XAD_EXTEND) flag.
1710 lwm = xtlck->lwm.offset;
1712 lwm = XTPAGEMAXSLOT;
1717 jfs_err("xtLog: lwm > next");
1720 tlck->flag |= tlckUPDATEMAP;
1721 xadlock->flag = mlckALLOCXADLIST;
1722 xadlock->count = next - lwm;
1723 if ((xadlock->count <= 4) && (tblk->xflag & COMMIT_LAZY)) {
1727 * Lazy commit may allow xtree to be modified before
1728 * txUpdateMap runs. Copy xad into linelock to
1729 * preserve correct data.
1731 * We can fit twice as may pxd's as xads in the lock
1733 xadlock->flag = mlckALLOCPXDLIST;
1734 pxd = xadlock->xdlist = &xtlck->pxdlock;
1735 for (i = 0; i < xadlock->count; i++) {
1736 PXDaddress(pxd, addressXAD(&p->xad[lwm + i]));
1737 PXDlength(pxd, lengthXAD(&p->xad[lwm + i]));
1738 p->xad[lwm + i].flag &=
1739 ~(XAD_NEW | XAD_EXTENDED);
1744 * xdlist will point to into inode's xtree, ensure
1745 * that transaction is not committed lazily.
1747 xadlock->flag = mlckALLOCXADLIST;
1748 xadlock->xdlist = &p->xad[lwm];
1749 tblk->xflag &= ~COMMIT_LAZY;
1751 jfs_info("xtLog: alloc ip:0x%p mp:0x%p tlck:0x%p lwm:%d count:%d",
1752 tlck->ip, mp, tlck, lwm, xadlock->count);
1757 /* mark page as homeward bound */
1758 tlck->flag |= tlckWRITEPAGE;
1764 * page deletion: file deletion/truncation (ref. xtTruncate())
1766 * (page will be invalidated after log is written and bmap
1767 * is updated from the page);
1769 if (tlck->type & tlckFREE) {
1770 /* LOG_NOREDOPAGE log for NoRedoPage filter:
1771 * if page free from file delete, NoRedoFile filter from
1772 * inode image of zero link count will subsume NoRedoPage
1773 * filters for each page;
1774 * if page free from file truncattion, write NoRedoPage
1777 * upadte of block allocation map for the page itself:
1778 * if page free from deletion and truncation, LOG_UPDATEMAP
1779 * log for the page itself is generated from processing
1780 * its parent page xad entries;
1782 /* if page free from file truncation, log LOG_NOREDOPAGE
1783 * of the deleted page for logredo() to start NoRedoPage
1784 * filter for the page;
1786 if (tblk->xflag & COMMIT_TRUNCATE) {
1787 /* write NOREDOPAGE for the page */
1788 lrd->type = cpu_to_le16(LOG_NOREDOPAGE);
1789 PXDaddress(page_pxd, mp->index);
1791 mp->logical_size >> tblk->sb->
1794 cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1796 if (tlck->type & tlckBTROOT) {
1797 /* Empty xtree must be logged */
1798 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1800 cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1804 /* init LOG_UPDATEMAP of the freed extents
1805 * XAD[XTENTRYSTART:hwm) from the deleted page itself
1806 * for logredo() to update bmap;
1808 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1809 lrd->log.updatemap.type = cpu_to_le16(LOG_FREEXADLIST);
1810 xtlck = (struct xtlock *) & tlck->lock;
1811 hwm = xtlck->hwm.offset;
1812 lrd->log.updatemap.nxd =
1813 cpu_to_le16(hwm - XTENTRYSTART + 1);
1814 /* reformat linelock for lmLog() */
1815 xtlck->header.offset = XTENTRYSTART;
1816 xtlck->header.length = hwm - XTENTRYSTART + 1;
1818 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1820 /* format a maplock for txUpdateMap() to update bmap
1821 * to free extents of XAD[XTENTRYSTART:hwm) from the
1822 * deleted page itself;
1824 tlck->flag |= tlckUPDATEMAP;
1825 xadlock->count = hwm - XTENTRYSTART + 1;
1826 if ((xadlock->count <= 4) && (tblk->xflag & COMMIT_LAZY)) {
1830 * Lazy commit may allow xtree to be modified before
1831 * txUpdateMap runs. Copy xad into linelock to
1832 * preserve correct data.
1834 * We can fit twice as may pxd's as xads in the lock
1836 xadlock->flag = mlckFREEPXDLIST;
1837 pxd = xadlock->xdlist = &xtlck->pxdlock;
1838 for (i = 0; i < xadlock->count; i++) {
1840 addressXAD(&p->xad[XTENTRYSTART + i]));
1842 lengthXAD(&p->xad[XTENTRYSTART + i]));
1847 * xdlist will point to into inode's xtree, ensure
1848 * that transaction is not committed lazily.
1850 xadlock->flag = mlckFREEXADLIST;
1851 xadlock->xdlist = &p->xad[XTENTRYSTART];
1852 tblk->xflag &= ~COMMIT_LAZY;
1854 jfs_info("xtLog: free ip:0x%p mp:0x%p count:%d lwm:2",
1855 tlck->ip, mp, xadlock->count);
1859 /* mark page as invalid */
1860 if (((tblk->xflag & COMMIT_PWMAP) || S_ISDIR(ip->i_mode))
1861 && !(tlck->type & tlckBTROOT))
1862 tlck->flag |= tlckFREEPAGE;
1864 else (tblk->xflag & COMMIT_PMAP)
1871 * page/entry truncation: file truncation (ref. xtTruncate())
1873 * |----------+------+------+---------------|
1875 * | | hwm - hwm before truncation
1876 * | next - truncation point
1877 * lwm - lwm before truncation
1880 if (tlck->type & tlckTRUNCATE) {
1881 pxd_t pxd; /* truncated extent of xad */
1885 * For truncation the entire linelock may be used, so it would
1886 * be difficult to store xad list in linelock itself.
1887 * Therefore, we'll just force transaction to be committed
1888 * synchronously, so that xtree pages won't be changed before
1891 tblk->xflag &= ~COMMIT_LAZY;
1892 lwm = xtlck->lwm.offset;
1894 lwm = XTPAGEMAXSLOT;
1895 hwm = xtlck->hwm.offset;
1896 twm = xtlck->twm.offset;
1901 /* log after-image for logredo():
1903 * logredo() will update bmap for alloc of new/extended
1904 * extents (XAD_NEW|XAD_EXTEND) of XAD[lwm:next) from
1905 * after-image of XADlist;
1906 * logredo() resets (XAD_NEW|XAD_EXTEND) flag when
1907 * applying the after-image to the meta-data page.
1909 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1910 PXDaddress(page_pxd, mp->index);
1912 mp->logical_size >> tblk->sb->s_blocksize_bits);
1913 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1916 * truncate entry XAD[twm == next - 1]:
1918 if (twm == next - 1) {
1919 /* init LOG_UPDATEMAP for logredo() to update bmap for
1920 * free of truncated delta extent of the truncated
1921 * entry XAD[next - 1]:
1922 * (xtlck->pxdlock = truncated delta extent);
1924 pxdlock = (struct pxd_lock *) & xtlck->pxdlock;
1925 /* assert(pxdlock->type & tlckTRUNCATE); */
1926 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1927 lrd->log.updatemap.type = cpu_to_le16(LOG_FREEPXD);
1928 lrd->log.updatemap.nxd = cpu_to_le16(1);
1929 lrd->log.updatemap.pxd = pxdlock->pxd;
1930 pxd = pxdlock->pxd; /* save to format maplock */
1932 cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1936 * free entries XAD[next:hwm]:
1939 /* init LOG_UPDATEMAP of the freed extents
1940 * XAD[next:hwm] from the deleted page itself
1941 * for logredo() to update bmap;
1943 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1944 lrd->log.updatemap.type =
1945 cpu_to_le16(LOG_FREEXADLIST);
1946 xtlck = (struct xtlock *) & tlck->lock;
1947 hwm = xtlck->hwm.offset;
1948 lrd->log.updatemap.nxd =
1949 cpu_to_le16(hwm - next + 1);
1950 /* reformat linelock for lmLog() */
1951 xtlck->header.offset = next;
1952 xtlck->header.length = hwm - next + 1;
1955 cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1959 * format maplock(s) for txUpdateMap() to update bmap
1964 * allocate entries XAD[lwm:next):
1967 /* format a maplock for txUpdateMap() to update bPMAP
1968 * for alloc of new/extended extents of XAD[lwm:next)
1969 * from the page itself;
1970 * txUpdateMap() resets (XAD_NEW|XAD_EXTEND) flag.
1972 tlck->flag |= tlckUPDATEMAP;
1973 xadlock->flag = mlckALLOCXADLIST;
1974 xadlock->count = next - lwm;
1975 xadlock->xdlist = &p->xad[lwm];
1977 jfs_info("xtLog: alloc ip:0x%p mp:0x%p count:%d lwm:%d next:%d",
1978 tlck->ip, mp, xadlock->count, lwm, next);
1984 * truncate entry XAD[twm == next - 1]:
1986 if (twm == next - 1) {
1987 /* format a maplock for txUpdateMap() to update bmap
1988 * to free truncated delta extent of the truncated
1989 * entry XAD[next - 1];
1990 * (xtlck->pxdlock = truncated delta extent);
1992 tlck->flag |= tlckUPDATEMAP;
1993 pxdlock = (struct pxd_lock *) xadlock;
1994 pxdlock->flag = mlckFREEPXD;
1998 jfs_info("xtLog: truncate ip:0x%p mp:0x%p count:%d hwm:%d",
1999 ip, mp, pxdlock->count, hwm);
2005 * free entries XAD[next:hwm]:
2008 /* format a maplock for txUpdateMap() to update bmap
2009 * to free extents of XAD[next:hwm] from thedeleted
2012 tlck->flag |= tlckUPDATEMAP;
2013 xadlock->flag = mlckFREEXADLIST;
2014 xadlock->count = hwm - next + 1;
2015 xadlock->xdlist = &p->xad[next];
2017 jfs_info("xtLog: free ip:0x%p mp:0x%p count:%d next:%d hwm:%d",
2018 tlck->ip, mp, xadlock->count, next, hwm);
2022 /* mark page as homeward bound */
2023 tlck->flag |= tlckWRITEPAGE;
2031 * function: log from maplock of freed data extents;
2033 static void mapLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
2034 struct tlock * tlck)
2036 struct pxd_lock *pxdlock;
2041 * page relocation: free the source page extent
2043 * a maplock for txUpdateMap() for free of the page
2044 * has been formatted at txLock() time saving the src
2045 * relocated page address;
2047 if (tlck->type & tlckRELOCATE) {
2048 /* log LOG_NOREDOPAGE of the old relocated page
2049 * for logredo() to start NoRedoPage filter;
2051 lrd->type = cpu_to_le16(LOG_NOREDOPAGE);
2052 pxdlock = (struct pxd_lock *) & tlck->lock;
2053 pxd = &lrd->log.redopage.pxd;
2054 *pxd = pxdlock->pxd;
2055 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
2057 /* (N.B. currently, logredo() does NOT update bmap
2058 * for free of the page itself for (LOG_XTREE|LOG_NOREDOPAGE);
2059 * if page free from relocation, LOG_UPDATEMAP log is
2060 * specifically generated now for logredo()
2061 * to update bmap for free of src relocated page;
2062 * (new flag LOG_RELOCATE may be introduced which will
2063 * inform logredo() to start NORedoPage filter and also
2064 * update block allocation map at the same time, thus
2065 * avoiding an extra log write);
2067 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
2068 lrd->log.updatemap.type = cpu_to_le16(LOG_FREEPXD);
2069 lrd->log.updatemap.nxd = cpu_to_le16(1);
2070 lrd->log.updatemap.pxd = pxdlock->pxd;
2071 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
2073 /* a maplock for txUpdateMap() for free of the page
2074 * has been formatted at txLock() time;
2076 tlck->flag |= tlckUPDATEMAP;
2081 * Otherwise it's not a relocate request
2085 /* log LOG_UPDATEMAP for logredo() to update bmap for
2086 * free of truncated/relocated delta extent of the data;
2087 * e.g.: external EA extent, relocated/truncated extent
2088 * from xtTailgate();
2090 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
2091 pxdlock = (struct pxd_lock *) & tlck->lock;
2092 nlock = pxdlock->index;
2093 for (i = 0; i < nlock; i++, pxdlock++) {
2094 if (pxdlock->flag & mlckALLOCPXD)
2095 lrd->log.updatemap.type =
2096 cpu_to_le16(LOG_ALLOCPXD);
2098 lrd->log.updatemap.type =
2099 cpu_to_le16(LOG_FREEPXD);
2100 lrd->log.updatemap.nxd = cpu_to_le16(1);
2101 lrd->log.updatemap.pxd = pxdlock->pxd;
2103 cpu_to_le32(lmLog(log, tblk, lrd, NULL));
2104 jfs_info("mapLog: xaddr:0x%lx xlen:0x%x",
2105 (ulong) addressPXD(&pxdlock->pxd),
2106 lengthPXD(&pxdlock->pxd));
2110 tlck->flag |= tlckUPDATEMAP;
2117 * function: acquire maplock for EA/ACL extents or
2118 * set COMMIT_INLINE flag;
2120 void txEA(tid_t tid, struct inode *ip, dxd_t * oldea, dxd_t * newea)
2122 struct tlock *tlck = NULL;
2123 struct pxd_lock *maplock = NULL, *pxdlock = NULL;
2126 * format maplock for alloc of new EA extent
2129 /* Since the newea could be a completely zeroed entry we need to
2130 * check for the two flags which indicate we should actually
2131 * commit new EA data
2133 if (newea->flag & DXD_EXTENT) {
2134 tlck = txMaplock(tid, ip, tlckMAP);
2135 maplock = (struct pxd_lock *) & tlck->lock;
2136 pxdlock = (struct pxd_lock *) maplock;
2137 pxdlock->flag = mlckALLOCPXD;
2138 PXDaddress(&pxdlock->pxd, addressDXD(newea));
2139 PXDlength(&pxdlock->pxd, lengthDXD(newea));
2142 } else if (newea->flag & DXD_INLINE) {
2145 set_cflag(COMMIT_Inlineea, ip);
2150 * format maplock for free of old EA extent
2152 if (!test_cflag(COMMIT_Nolink, ip) && oldea->flag & DXD_EXTENT) {
2154 tlck = txMaplock(tid, ip, tlckMAP);
2155 maplock = (struct pxd_lock *) & tlck->lock;
2156 pxdlock = (struct pxd_lock *) maplock;
2159 pxdlock->flag = mlckFREEPXD;
2160 PXDaddress(&pxdlock->pxd, addressDXD(oldea));
2161 PXDlength(&pxdlock->pxd, lengthDXD(oldea));
2169 * function: synchronously write pages locked by transaction
2170 * after txLog() but before txUpdateMap();
2172 static void txForce(struct tblock * tblk)
2176 struct metapage *mp;
2179 * reverse the order of transaction tlocks in
2180 * careful update order of address index pages
2181 * (right to left, bottom up)
2183 tlck = lid_to_tlock(tblk->next);
2187 tlck = lid_to_tlock(lid);
2189 tlck->next = tblk->next;
2195 * synchronously write the page, and
2196 * hold the page for txUpdateMap();
2198 for (lid = tblk->next; lid; lid = next) {
2199 tlck = lid_to_tlock(lid);
2202 if ((mp = tlck->mp) != NULL &&
2203 (tlck->type & tlckBTROOT) == 0) {
2204 assert(mp->xflag & COMMIT_PAGE);
2206 if (tlck->flag & tlckWRITEPAGE) {
2207 tlck->flag &= ~tlckWRITEPAGE;
2209 /* do not release page to freelist */
2213 * The "right" thing to do here is to
2214 * synchronously write the metadata.
2215 * With the current implementation this
2216 * is hard since write_metapage requires
2217 * us to kunmap & remap the page. If we
2218 * have tlocks pointing into the metadata
2219 * pages, we don't want to do this. I think
2220 * we can get by with synchronously writing
2221 * the pages when they are released.
2223 assert(mp->nohomeok);
2224 set_bit(META_dirty, &mp->flag);
2225 set_bit(META_sync, &mp->flag);
2235 * function: update persistent allocation map (and working map
2240 static void txUpdateMap(struct tblock * tblk)
2243 struct inode *ipimap;
2246 struct maplock *maplock;
2247 struct pxd_lock pxdlock;
2250 struct metapage *mp = NULL;
2252 ipimap = JFS_SBI(tblk->sb)->ipimap;
2254 maptype = (tblk->xflag & COMMIT_PMAP) ? COMMIT_PMAP : COMMIT_PWMAP;
2258 * update block allocation map
2260 * update allocation state in pmap (and wmap) and
2261 * update lsn of the pmap page;
2264 * scan each tlock/page of transaction for block allocation/free:
2266 * for each tlock/page of transaction, update map.
2267 * ? are there tlock for pmap and pwmap at the same time ?
2269 for (lid = tblk->next; lid; lid = tlck->next) {
2270 tlck = lid_to_tlock(lid);
2272 if ((tlck->flag & tlckUPDATEMAP) == 0)
2275 if (tlck->flag & tlckFREEPAGE) {
2277 * Another thread may attempt to reuse freed space
2278 * immediately, so we want to get rid of the metapage
2279 * before anyone else has a chance to get it.
2280 * Lock metapage, update maps, then invalidate
2284 ASSERT(mp->xflag & COMMIT_PAGE);
2290 * . in-line PXD list:
2291 * . out-of-line XAD list:
2293 maplock = (struct maplock *) & tlck->lock;
2294 nlock = maplock->index;
2296 for (k = 0; k < nlock; k++, maplock++) {
2298 * allocate blocks in persistent map:
2300 * blocks have been allocated from wmap at alloc time;
2302 if (maplock->flag & mlckALLOC) {
2303 txAllocPMap(ipimap, maplock, tblk);
2306 * free blocks in persistent and working map:
2307 * blocks will be freed in pmap and then in wmap;
2309 * ? tblock specifies the PMAP/PWMAP based upon
2312 * free blocks in persistent map:
2313 * blocks will be freed from wmap at last reference
2314 * release of the object for regular files;
2316 * Alway free blocks from both persistent & working
2317 * maps for directories
2319 else { /* (maplock->flag & mlckFREE) */
2321 if (tlck->flag & tlckDIRECTORY)
2322 txFreeMap(ipimap, maplock,
2323 tblk, COMMIT_PWMAP);
2325 txFreeMap(ipimap, maplock,
2329 if (tlck->flag & tlckFREEPAGE) {
2330 if (!(tblk->flag & tblkGC_LAZY)) {
2331 /* This is equivalent to txRelease */
2332 ASSERT(mp->lid == lid);
2335 assert(mp->nohomeok == 1);
2336 metapage_homeok(mp);
2337 discard_metapage(mp);
2342 * update inode allocation map
2344 * update allocation state in pmap and
2345 * update lsn of the pmap page;
2346 * update in-memory inode flag/state
2348 * unlock mapper/write lock
2350 if (tblk->xflag & COMMIT_CREATE) {
2351 diUpdatePMap(ipimap, tblk->ino, false, tblk);
2352 /* update persistent block allocation map
2353 * for the allocation of inode extent;
2355 pxdlock.flag = mlckALLOCPXD;
2356 pxdlock.pxd = tblk->u.ixpxd;
2358 txAllocPMap(ipimap, (struct maplock *) & pxdlock, tblk);
2359 } else if (tblk->xflag & COMMIT_DELETE) {
2361 diUpdatePMap(ipimap, ip->i_ino, true, tblk);
2369 * function: allocate from persistent map;
2378 * allocate from persistent map;
2379 * free from persistent map;
2380 * (e.g., tmp file - free from working map at releae
2381 * of last reference);
2382 * free from persistent and working map;
2384 * lsn - log sequence number;
2386 static void txAllocPMap(struct inode *ip, struct maplock * maplock,
2387 struct tblock * tblk)
2389 struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
2390 struct xdlistlock *xadlistlock;
2394 struct pxd_lock *pxdlock;
2395 struct xdlistlock *pxdlistlock;
2400 * allocate from persistent map;
2402 if (maplock->flag & mlckALLOCXADLIST) {
2403 xadlistlock = (struct xdlistlock *) maplock;
2404 xad = xadlistlock->xdlist;
2405 for (n = 0; n < xadlistlock->count; n++, xad++) {
2406 if (xad->flag & (XAD_NEW | XAD_EXTENDED)) {
2407 xaddr = addressXAD(xad);
2408 xlen = lengthXAD(xad);
2409 dbUpdatePMap(ipbmap, false, xaddr,
2411 xad->flag &= ~(XAD_NEW | XAD_EXTENDED);
2412 jfs_info("allocPMap: xaddr:0x%lx xlen:%d",
2413 (ulong) xaddr, xlen);
2416 } else if (maplock->flag & mlckALLOCPXD) {
2417 pxdlock = (struct pxd_lock *) maplock;
2418 xaddr = addressPXD(&pxdlock->pxd);
2419 xlen = lengthPXD(&pxdlock->pxd);
2420 dbUpdatePMap(ipbmap, false, xaddr, (s64) xlen, tblk);
2421 jfs_info("allocPMap: xaddr:0x%lx xlen:%d", (ulong) xaddr, xlen);
2422 } else { /* (maplock->flag & mlckALLOCPXDLIST) */
2424 pxdlistlock = (struct xdlistlock *) maplock;
2425 pxd = pxdlistlock->xdlist;
2426 for (n = 0; n < pxdlistlock->count; n++, pxd++) {
2427 xaddr = addressPXD(pxd);
2428 xlen = lengthPXD(pxd);
2429 dbUpdatePMap(ipbmap, false, xaddr, (s64) xlen,
2431 jfs_info("allocPMap: xaddr:0x%lx xlen:%d",
2432 (ulong) xaddr, xlen);
2440 * function: free from persistent and/or working map;
2442 * todo: optimization
2444 void txFreeMap(struct inode *ip,
2445 struct maplock * maplock, struct tblock * tblk, int maptype)
2447 struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
2448 struct xdlistlock *xadlistlock;
2452 struct pxd_lock *pxdlock;
2453 struct xdlistlock *pxdlistlock;
2457 jfs_info("txFreeMap: tblk:0x%p maplock:0x%p maptype:0x%x",
2458 tblk, maplock, maptype);
2461 * free from persistent map;
2463 if (maptype == COMMIT_PMAP || maptype == COMMIT_PWMAP) {
2464 if (maplock->flag & mlckFREEXADLIST) {
2465 xadlistlock = (struct xdlistlock *) maplock;
2466 xad = xadlistlock->xdlist;
2467 for (n = 0; n < xadlistlock->count; n++, xad++) {
2468 if (!(xad->flag & XAD_NEW)) {
2469 xaddr = addressXAD(xad);
2470 xlen = lengthXAD(xad);
2471 dbUpdatePMap(ipbmap, true, xaddr,
2473 jfs_info("freePMap: xaddr:0x%lx xlen:%d",
2474 (ulong) xaddr, xlen);
2477 } else if (maplock->flag & mlckFREEPXD) {
2478 pxdlock = (struct pxd_lock *) maplock;
2479 xaddr = addressPXD(&pxdlock->pxd);
2480 xlen = lengthPXD(&pxdlock->pxd);
2481 dbUpdatePMap(ipbmap, true, xaddr, (s64) xlen,
2483 jfs_info("freePMap: xaddr:0x%lx xlen:%d",
2484 (ulong) xaddr, xlen);
2485 } else { /* (maplock->flag & mlckALLOCPXDLIST) */
2487 pxdlistlock = (struct xdlistlock *) maplock;
2488 pxd = pxdlistlock->xdlist;
2489 for (n = 0; n < pxdlistlock->count; n++, pxd++) {
2490 xaddr = addressPXD(pxd);
2491 xlen = lengthPXD(pxd);
2492 dbUpdatePMap(ipbmap, true, xaddr,
2494 jfs_info("freePMap: xaddr:0x%lx xlen:%d",
2495 (ulong) xaddr, xlen);
2501 * free from working map;
2503 if (maptype == COMMIT_PWMAP || maptype == COMMIT_WMAP) {
2504 if (maplock->flag & mlckFREEXADLIST) {
2505 xadlistlock = (struct xdlistlock *) maplock;
2506 xad = xadlistlock->xdlist;
2507 for (n = 0; n < xadlistlock->count; n++, xad++) {
2508 xaddr = addressXAD(xad);
2509 xlen = lengthXAD(xad);
2510 dbFree(ip, xaddr, (s64) xlen);
2512 jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
2513 (ulong) xaddr, xlen);
2515 } else if (maplock->flag & mlckFREEPXD) {
2516 pxdlock = (struct pxd_lock *) maplock;
2517 xaddr = addressPXD(&pxdlock->pxd);
2518 xlen = lengthPXD(&pxdlock->pxd);
2519 dbFree(ip, xaddr, (s64) xlen);
2520 jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
2521 (ulong) xaddr, xlen);
2522 } else { /* (maplock->flag & mlckFREEPXDLIST) */
2524 pxdlistlock = (struct xdlistlock *) maplock;
2525 pxd = pxdlistlock->xdlist;
2526 for (n = 0; n < pxdlistlock->count; n++, pxd++) {
2527 xaddr = addressPXD(pxd);
2528 xlen = lengthPXD(pxd);
2529 dbFree(ip, xaddr, (s64) xlen);
2530 jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
2531 (ulong) xaddr, xlen);
2540 * function: remove tlock from inode anonymous locklist
2542 void txFreelock(struct inode *ip)
2544 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
2545 struct tlock *xtlck, *tlck;
2546 lid_t xlid = 0, lid;
2548 if (!jfs_ip->atlhead)
2552 xtlck = (struct tlock *) &jfs_ip->atlhead;
2554 while ((lid = xtlck->next) != 0) {
2555 tlck = lid_to_tlock(lid);
2556 if (tlck->flag & tlckFREELOCK) {
2557 xtlck->next = tlck->next;
2565 if (jfs_ip->atlhead)
2566 jfs_ip->atltail = xlid;
2568 jfs_ip->atltail = 0;
2570 * If inode was on anon_list, remove it
2572 list_del_init(&jfs_ip->anon_inode_list);
2580 * function: abort tx before commit;
2582 * frees line-locks and segment locks for all
2583 * segments in comdata structure.
2584 * Optionally sets state of file-system to FM_DIRTY in super-block.
2585 * log age of page-frames in memory for which caller has
2586 * are reset to 0 (to avoid logwarap).
2588 void txAbort(tid_t tid, int dirty)
2591 struct metapage *mp;
2592 struct tblock *tblk = tid_to_tblock(tid);
2596 * free tlocks of the transaction
2598 for (lid = tblk->next; lid; lid = next) {
2599 tlck = lid_to_tlock(lid);
2602 JFS_IP(tlck->ip)->xtlid = 0;
2608 * reset lsn of page to avoid logwarap:
2610 * (page may have been previously committed by another
2611 * transaction(s) but has not been paged, i.e.,
2612 * it may be on logsync list even though it has not
2613 * been logged for the current tx.)
2615 if (mp->xflag & COMMIT_PAGE && mp->lsn)
2618 /* insert tlock at head of freelist */
2624 /* caller will free the transaction block */
2626 tblk->next = tblk->last = 0;
2629 * mark filesystem dirty
2632 jfs_error(tblk->sb, "\n");
2638 * txLazyCommit(void)
2640 * All transactions except those changing ipimap (COMMIT_FORCE) are
2641 * processed by this routine. This insures that the inode and block
2642 * allocation maps are updated in order. For synchronous transactions,
2643 * let the user thread finish processing after txUpdateMap() is called.
2645 static void txLazyCommit(struct tblock * tblk)
2647 struct jfs_log *log;
2649 while (((tblk->flag & tblkGC_READY) == 0) &&
2650 ((tblk->flag & tblkGC_UNLOCKED) == 0)) {
2651 /* We must have gotten ahead of the user thread
2653 jfs_info("jfs_lazycommit: tblk 0x%p not unlocked", tblk);
2657 jfs_info("txLazyCommit: processing tblk 0x%p", tblk);
2661 log = (struct jfs_log *) JFS_SBI(tblk->sb)->log;
2663 spin_lock_irq(&log->gclock); // LOGGC_LOCK
2665 tblk->flag |= tblkGC_COMMITTED;
2667 if (tblk->flag & tblkGC_READY)
2670 wake_up_all(&tblk->gcwait); // LOGGC_WAKEUP
2673 * Can't release log->gclock until we've tested tblk->flag
2675 if (tblk->flag & tblkGC_LAZY) {
2676 spin_unlock_irq(&log->gclock); // LOGGC_UNLOCK
2678 tblk->flag &= ~tblkGC_LAZY;
2679 txEnd(tblk - TxBlock); /* Convert back to tid */
2681 spin_unlock_irq(&log->gclock); // LOGGC_UNLOCK
2683 jfs_info("txLazyCommit: done: tblk = 0x%p", tblk);
2687 * jfs_lazycommit(void)
2689 * To be run as a kernel daemon. If lbmIODone is called in an interrupt
2690 * context, or where blocking is not wanted, this routine will process
2691 * committed transactions from the unlock queue.
2693 int jfs_lazycommit(void *arg)
2696 struct tblock *tblk;
2697 unsigned long flags;
2698 struct jfs_sb_info *sbi;
2702 jfs_commit_thread_waking = 0; /* OK to wake another thread */
2703 while (!list_empty(&TxAnchor.unlock_queue)) {
2705 list_for_each_entry(tblk, &TxAnchor.unlock_queue,
2708 sbi = JFS_SBI(tblk->sb);
2710 * For each volume, the transactions must be
2711 * handled in order. If another commit thread
2712 * is handling a tblk for this superblock,
2715 if (sbi->commit_state & IN_LAZYCOMMIT)
2718 sbi->commit_state |= IN_LAZYCOMMIT;
2722 * Remove transaction from queue
2724 list_del(&tblk->cqueue);
2730 sbi->commit_state &= ~IN_LAZYCOMMIT;
2732 * Don't continue in the for loop. (We can't
2733 * anyway, it's unsafe!) We want to go back to
2734 * the beginning of the list.
2739 /* If there was nothing to do, don't continue */
2743 /* In case a wakeup came while all threads were active */
2744 jfs_commit_thread_waking = 0;
2746 if (freezing(current)) {
2750 DECLARE_WAITQUEUE(wq, current);
2752 add_wait_queue(&jfs_commit_thread_wait, &wq);
2753 set_current_state(TASK_INTERRUPTIBLE);
2756 remove_wait_queue(&jfs_commit_thread_wait, &wq);
2758 } while (!kthread_should_stop());
2760 if (!list_empty(&TxAnchor.unlock_queue))
2761 jfs_err("jfs_lazycommit being killed w/pending transactions!");
2763 jfs_info("jfs_lazycommit being killed");
2767 void txLazyUnlock(struct tblock * tblk)
2769 unsigned long flags;
2773 list_add_tail(&tblk->cqueue, &TxAnchor.unlock_queue);
2775 * Don't wake up a commit thread if there is already one servicing
2776 * this superblock, or if the last one we woke up hasn't started yet.
2778 if (!(JFS_SBI(tblk->sb)->commit_state & IN_LAZYCOMMIT) &&
2779 !jfs_commit_thread_waking) {
2780 jfs_commit_thread_waking = 1;
2781 wake_up(&jfs_commit_thread_wait);
2786 static void LogSyncRelease(struct metapage * mp)
2788 struct jfs_log *log = mp->log;
2790 assert(mp->nohomeok);
2792 metapage_homeok(mp);
2798 * Block all new transactions and push anonymous transactions to
2801 * This does almost the same thing as jfs_sync below. We don't
2802 * worry about deadlocking when jfs_tlocks_low is set, since we would
2803 * expect jfs_sync to get us out of that jam.
2805 void txQuiesce(struct super_block *sb)
2808 struct jfs_inode_info *jfs_ip;
2809 struct jfs_log *log = JFS_SBI(sb)->log;
2812 set_bit(log_QUIESCE, &log->flag);
2816 while (!list_empty(&TxAnchor.anon_list)) {
2817 jfs_ip = list_entry(TxAnchor.anon_list.next,
2818 struct jfs_inode_info,
2820 ip = &jfs_ip->vfs_inode;
2823 * inode will be removed from anonymous list
2824 * when it is committed
2827 tid = txBegin(ip->i_sb, COMMIT_INODE | COMMIT_FORCE);
2828 mutex_lock(&jfs_ip->commit_mutex);
2829 txCommit(tid, 1, &ip, 0);
2831 mutex_unlock(&jfs_ip->commit_mutex);
2833 * Just to be safe. I don't know how
2834 * long we can run without blocking
2841 * If jfs_sync is running in parallel, there could be some inodes
2842 * on anon_list2. Let's check.
2844 if (!list_empty(&TxAnchor.anon_list2)) {
2845 list_splice_init(&TxAnchor.anon_list2, &TxAnchor.anon_list);
2851 * We may need to kick off the group commit
2853 jfs_flush_journal(log, 0);
2859 * Allows transactions to start again following txQuiesce
2861 void txResume(struct super_block *sb)
2863 struct jfs_log *log = JFS_SBI(sb)->log;
2865 clear_bit(log_QUIESCE, &log->flag);
2866 TXN_WAKEUP(&log->syncwait);
2872 * To be run as a kernel daemon. This is awakened when tlocks run low.
2873 * We write any inodes that have anonymous tlocks so they will become
2876 int jfs_sync(void *arg)
2879 struct jfs_inode_info *jfs_ip;
2884 * write each inode on the anonymous inode list
2887 while (jfs_tlocks_low && !list_empty(&TxAnchor.anon_list)) {
2888 jfs_ip = list_entry(TxAnchor.anon_list.next,
2889 struct jfs_inode_info,
2891 ip = &jfs_ip->vfs_inode;
2895 * Inode is being freed
2897 list_del_init(&jfs_ip->anon_inode_list);
2898 } else if (mutex_trylock(&jfs_ip->commit_mutex)) {
2900 * inode will be removed from anonymous list
2901 * when it is committed
2904 tid = txBegin(ip->i_sb, COMMIT_INODE);
2905 txCommit(tid, 1, &ip, 0);
2907 mutex_unlock(&jfs_ip->commit_mutex);
2911 * Just to be safe. I don't know how
2912 * long we can run without blocking
2917 /* We can't get the commit mutex. It may
2918 * be held by a thread waiting for tlock's
2919 * so let's not block here. Save it to
2920 * put back on the anon_list.
2923 /* Move from anon_list to anon_list2 */
2924 list_move(&jfs_ip->anon_inode_list,
2925 &TxAnchor.anon_list2);
2932 /* Add anon_list2 back to anon_list */
2933 list_splice_init(&TxAnchor.anon_list2, &TxAnchor.anon_list);
2935 if (freezing(current)) {
2939 set_current_state(TASK_INTERRUPTIBLE);
2943 } while (!kthread_should_stop());
2945 jfs_info("jfs_sync being killed");
2949 #if defined(CONFIG_PROC_FS) && defined(CONFIG_JFS_DEBUG)
2950 int jfs_txanchor_proc_show(struct seq_file *m, void *v)
2957 waitqueue_active(&TxAnchor.freewait) ? "active" : "empty";
2959 waitqueue_active(&TxAnchor.freelockwait) ? "active" : "empty";
2961 waitqueue_active(&TxAnchor.lowlockwait) ? "active" : "empty";
2969 "freelockwait = %s\n"
2970 "lowlockwait = %s\n"
2971 "tlocksInUse = %d\n"
2972 "jfs_tlocks_low = %d\n"
2973 "unlock_queue is %sempty\n",
2979 TxAnchor.tlocksInUse,
2981 list_empty(&TxAnchor.unlock_queue) ? "" : "not ");
2986 #if defined(CONFIG_PROC_FS) && defined(CONFIG_JFS_STATISTICS)
2987 int jfs_txstats_proc_show(struct seq_file *m, void *v)
2992 "calls to txBegin = %d\n"
2993 "txBegin blocked by sync barrier = %d\n"
2994 "txBegin blocked by tlocks low = %d\n"
2995 "txBegin blocked by no free tid = %d\n"
2996 "calls to txBeginAnon = %d\n"
2997 "txBeginAnon blocked by sync barrier = %d\n"
2998 "txBeginAnon blocked by tlocks low = %d\n"
2999 "calls to txLockAlloc = %d\n"
3000 "tLockAlloc blocked by no free lock = %d\n",
3002 TxStat.txBegin_barrier,
3003 TxStat.txBegin_lockslow,
3004 TxStat.txBegin_freetid,
3006 TxStat.txBeginAnon_barrier,
3007 TxStat.txBeginAnon_lockslow,
3009 TxStat.txLockAlloc_freelock);