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 "jfs_incore.h"
53 #include "jfs_inode.h"
54 #include "jfs_filsys.h"
55 #include "jfs_metapage.h"
56 #include "jfs_dinode.h"
59 #include "jfs_superblock.h"
60 #include "jfs_debug.h"
63 * transaction management structures
66 int freetid; /* index of a free tid structure */
67 int freelock; /* index first free lock word */
68 wait_queue_head_t freewait; /* eventlist of free tblock */
69 wait_queue_head_t freelockwait; /* eventlist of free tlock */
70 wait_queue_head_t lowlockwait; /* eventlist of ample tlocks */
71 int tlocksInUse; /* Number of tlocks in use */
72 spinlock_t LazyLock; /* synchronize sync_queue & unlock_queue */
73 /* struct tblock *sync_queue; * Transactions waiting for data sync */
74 struct list_head unlock_queue; /* Txns waiting to be released */
75 struct list_head anon_list; /* inodes having anonymous txns */
76 struct list_head anon_list2; /* inodes having anonymous txns
77 that couldn't be sync'ed */
80 int jfs_tlocks_low; /* Indicates low number of available tlocks */
82 #ifdef CONFIG_JFS_STATISTICS
86 uint txBegin_lockslow;
89 uint txBeginAnon_barrier;
90 uint txBeginAnon_lockslow;
92 uint txLockAlloc_freelock;
96 static int nTxBlock = -1; /* number of transaction blocks */
97 module_param(nTxBlock, int, 0);
98 MODULE_PARM_DESC(nTxBlock,
99 "Number of transaction blocks (max:65536)");
101 static int nTxLock = -1; /* number of transaction locks */
102 module_param(nTxLock, int, 0);
103 MODULE_PARM_DESC(nTxLock,
104 "Number of transaction locks (max:65536)");
106 struct tblock *TxBlock; /* transaction block table */
107 static int TxLockLWM; /* Low water mark for number of txLocks used */
108 static int TxLockHWM; /* High water mark for number of txLocks used */
109 static int TxLockVHWM; /* Very High water mark */
110 struct tlock *TxLock; /* transaction lock table */
113 * transaction management lock
115 static DEFINE_SPINLOCK(jfsTxnLock);
117 #define TXN_LOCK() spin_lock(&jfsTxnLock)
118 #define TXN_UNLOCK() spin_unlock(&jfsTxnLock)
120 #define LAZY_LOCK_INIT() spin_lock_init(&TxAnchor.LazyLock);
121 #define LAZY_LOCK(flags) spin_lock_irqsave(&TxAnchor.LazyLock, flags)
122 #define LAZY_UNLOCK(flags) spin_unlock_irqrestore(&TxAnchor.LazyLock, flags)
124 static DECLARE_WAIT_QUEUE_HEAD(jfs_commit_thread_wait);
125 static int jfs_commit_thread_waking;
128 * Retry logic exist outside these macros to protect from spurrious wakeups.
130 static inline void TXN_SLEEP_DROP_LOCK(wait_queue_head_t * event)
132 DECLARE_WAITQUEUE(wait, current);
134 add_wait_queue(event, &wait);
135 set_current_state(TASK_UNINTERRUPTIBLE);
138 __set_current_state(TASK_RUNNING);
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 commiting 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;
1147 struct jfs_inode_info *jfs_ip;
1150 struct super_block *sb;
1152 jfs_info("txCommit, tid = %d, flag = %d", tid, flag);
1153 /* is read-only file system ? */
1154 if (isReadOnly(iplist[0])) {
1159 sb = cd.sb = iplist[0]->i_sb;
1163 tid = txBegin(sb, 0);
1164 tblk = tid_to_tblock(tid);
1167 * initialize commit structure
1169 log = JFS_SBI(sb)->log;
1172 /* initialize log record descriptor in commit */
1174 lrd->logtid = cpu_to_le32(tblk->logtid);
1177 tblk->xflag |= flag;
1179 if ((flag & (COMMIT_FORCE | COMMIT_SYNC)) == 0)
1180 tblk->xflag |= COMMIT_LAZY;
1182 * prepare non-journaled objects for commit
1184 * flush data pages of non-journaled file
1185 * to prevent the file getting non-initialized disk blocks
1193 * acquire transaction lock on (on-disk) inodes
1195 * update on-disk inode from in-memory inode
1196 * acquiring transaction locks for AFTER records
1197 * on the on-disk inode of file object
1199 * sort the inodes array by inode number in descending order
1200 * to prevent deadlock when acquiring transaction lock
1201 * of on-disk inodes on multiple on-disk inode pages by
1202 * multiple concurrent transactions
1204 for (k = 0; k < cd.nip; k++) {
1205 top = (cd.iplist[k])->i_ino;
1206 for (n = k + 1; n < cd.nip; n++) {
1208 if (ip->i_ino > top) {
1210 cd.iplist[n] = cd.iplist[k];
1216 jfs_ip = JFS_IP(ip);
1219 * BUGBUG - This code has temporarily been removed. The
1220 * intent is to ensure that any file data is written before
1221 * the metadata is committed to the journal. This prevents
1222 * uninitialized data from appearing in a file after the
1223 * journal has been replayed. (The uninitialized data
1224 * could be sensitive data removed by another user.)
1226 * The problem now is that we are holding the IWRITELOCK
1227 * on the inode, and calling filemap_fdatawrite on an
1228 * unmapped page will cause a deadlock in jfs_get_block.
1230 * The long term solution is to pare down the use of
1231 * IWRITELOCK. We are currently holding it too long.
1232 * We could also be smarter about which data pages need
1233 * to be written before the transaction is committed and
1234 * when we don't need to worry about it at all.
1236 * if ((!S_ISDIR(ip->i_mode))
1237 * && (tblk->flag & COMMIT_DELETE) == 0)
1238 * filemap_write_and_wait(ip->i_mapping);
1242 * Mark inode as not dirty. It will still be on the dirty
1243 * inode list, but we'll know not to commit it again unless
1244 * it gets marked dirty again
1246 clear_cflag(COMMIT_Dirty, ip);
1248 /* inherit anonymous tlock(s) of inode */
1249 if (jfs_ip->atlhead) {
1250 lid_to_tlock(jfs_ip->atltail)->next = tblk->next;
1251 tblk->next = jfs_ip->atlhead;
1253 tblk->last = jfs_ip->atltail;
1254 jfs_ip->atlhead = jfs_ip->atltail = 0;
1256 list_del_init(&jfs_ip->anon_inode_list);
1261 * acquire transaction lock on on-disk inode page
1262 * (become first tlock of the tblk's tlock list)
1264 if (((rc = diWrite(tid, ip))))
1269 * write log records from transaction locks
1271 * txUpdateMap() resets XAD_NEW in XAD.
1273 if ((rc = txLog(log, tblk, &cd)))
1277 * Ensure that inode isn't reused before
1278 * lazy commit thread finishes processing
1280 if (tblk->xflag & COMMIT_DELETE) {
1281 atomic_inc(&tblk->u.ip->i_count);
1283 * Avoid a rare deadlock
1285 * If the inode is locked, we may be blocked in
1286 * jfs_commit_inode. If so, we don't want the
1287 * lazy_commit thread doing the last iput() on the inode
1288 * since that may block on the locked inode. Instead,
1289 * commit the transaction synchronously, so the last iput
1290 * will be done by the calling thread (or later)
1292 if (tblk->u.ip->i_state & I_LOCK)
1293 tblk->xflag &= ~COMMIT_LAZY;
1296 ASSERT((!(tblk->xflag & COMMIT_DELETE)) ||
1297 ((tblk->u.ip->i_nlink == 0) &&
1298 !test_cflag(COMMIT_Nolink, tblk->u.ip)));
1301 * write COMMIT log record
1303 lrd->type = cpu_to_le16(LOG_COMMIT);
1305 lsn = lmLog(log, tblk, lrd, NULL);
1307 lmGroupCommit(log, tblk);
1310 * - transaction is now committed -
1314 * force pages in careful update
1315 * (imap addressing structure update)
1317 if (flag & COMMIT_FORCE)
1321 * update allocation map.
1323 * update inode allocation map and inode:
1324 * free pager lock on memory object of inode if any.
1325 * update block allocation map.
1327 * txUpdateMap() resets XAD_NEW in XAD.
1329 if (tblk->xflag & COMMIT_FORCE)
1333 * free transaction locks and pageout/free pages
1337 if ((tblk->flag & tblkGC_LAZY) == 0)
1342 * reset in-memory object state
1344 for (k = 0; k < cd.nip; k++) {
1346 jfs_ip = JFS_IP(ip);
1349 * reset in-memory inode state
1360 jfs_info("txCommit: tid = %d, returning %d", tid, rc);
1367 * FUNCTION: Writes AFTER log records for all lines modified
1368 * by tid for segments specified by inodes in comdata.
1369 * Code assumes only WRITELOCKS are recorded in lockwords.
1375 static int txLog(struct jfs_log * log, struct tblock * tblk, struct commit * cd)
1381 struct lrd *lrd = &cd->lrd;
1384 * write log record(s) for each tlock of transaction,
1386 for (lid = tblk->next; lid; lid = tlck->next) {
1387 tlck = lid_to_tlock(lid);
1389 tlck->flag |= tlckLOG;
1391 /* initialize lrd common */
1393 lrd->aggregate = cpu_to_le32(JFS_SBI(ip->i_sb)->aggregate);
1394 lrd->log.redopage.fileset = cpu_to_le32(JFS_IP(ip)->fileset);
1395 lrd->log.redopage.inode = cpu_to_le32(ip->i_ino);
1397 /* write log record of page from the tlock */
1398 switch (tlck->type & tlckTYPE) {
1400 xtLog(log, tblk, lrd, tlck);
1404 dtLog(log, tblk, lrd, tlck);
1408 diLog(log, tblk, lrd, tlck, cd);
1412 mapLog(log, tblk, lrd, tlck);
1416 dataLog(log, tblk, lrd, tlck);
1420 jfs_err("UFO tlock:0x%p", tlck);
1430 * function: log inode tlock and format maplock to update bmap;
1432 static int diLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1433 struct tlock * tlck, struct commit * cd)
1436 struct metapage *mp;
1438 struct pxd_lock *pxdlock;
1442 /* initialize as REDOPAGE record format */
1443 lrd->log.redopage.type = cpu_to_le16(LOG_INODE);
1444 lrd->log.redopage.l2linesize = cpu_to_le16(L2INODESLOTSIZE);
1446 pxd = &lrd->log.redopage.pxd;
1451 if (tlck->type & tlckENTRY) {
1452 /* log after-image for logredo(): */
1453 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1454 PXDaddress(pxd, mp->index);
1456 mp->logical_size >> tblk->sb->s_blocksize_bits);
1457 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1459 /* mark page as homeward bound */
1460 tlck->flag |= tlckWRITEPAGE;
1461 } else if (tlck->type & tlckFREE) {
1465 * (pages of the freed inode extent have been invalidated and
1466 * a maplock for free of the extent has been formatted at
1469 * the tlock had been acquired on the inode allocation map page
1470 * (iag) that specifies the freed extent, even though the map
1471 * page is not itself logged, to prevent pageout of the map
1472 * page before the log;
1475 /* log LOG_NOREDOINOEXT of the freed inode extent for
1476 * logredo() to start NoRedoPage filters, and to update
1477 * imap and bmap for free of the extent;
1479 lrd->type = cpu_to_le16(LOG_NOREDOINOEXT);
1481 * For the LOG_NOREDOINOEXT record, we need
1482 * to pass the IAG number and inode extent
1483 * index (within that IAG) from which the
1484 * the extent being released. These have been
1485 * passed to us in the iplist[1] and iplist[2].
1487 lrd->log.noredoinoext.iagnum =
1488 cpu_to_le32((u32) (size_t) cd->iplist[1]);
1489 lrd->log.noredoinoext.inoext_idx =
1490 cpu_to_le32((u32) (size_t) cd->iplist[2]);
1492 pxdlock = (struct pxd_lock *) & tlck->lock;
1493 *pxd = pxdlock->pxd;
1494 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1497 tlck->flag |= tlckUPDATEMAP;
1499 /* mark page as homeward bound */
1500 tlck->flag |= tlckWRITEPAGE;
1502 jfs_err("diLog: UFO type tlck:0x%p", tlck);
1505 * alloc/free external EA extent
1507 * a maplock for txUpdateMap() to update bPWMAP for alloc/free
1508 * of the extent has been formatted at txLock() time;
1511 assert(tlck->type & tlckEA);
1513 /* log LOG_UPDATEMAP for logredo() to update bmap for
1514 * alloc of new (and free of old) external EA extent;
1516 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1517 pxdlock = (struct pxd_lock *) & tlck->lock;
1518 nlock = pxdlock->index;
1519 for (i = 0; i < nlock; i++, pxdlock++) {
1520 if (pxdlock->flag & mlckALLOCPXD)
1521 lrd->log.updatemap.type =
1522 cpu_to_le16(LOG_ALLOCPXD);
1524 lrd->log.updatemap.type =
1525 cpu_to_le16(LOG_FREEPXD);
1526 lrd->log.updatemap.nxd = cpu_to_le16(1);
1527 lrd->log.updatemap.pxd = pxdlock->pxd;
1529 cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1533 tlck->flag |= tlckUPDATEMAP;
1535 #endif /* _JFS_WIP */
1543 * function: log data tlock
1545 static int dataLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1546 struct tlock * tlck)
1548 struct metapage *mp;
1553 /* initialize as REDOPAGE record format */
1554 lrd->log.redopage.type = cpu_to_le16(LOG_DATA);
1555 lrd->log.redopage.l2linesize = cpu_to_le16(L2DATASLOTSIZE);
1557 pxd = &lrd->log.redopage.pxd;
1559 /* log after-image for logredo(): */
1560 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1562 if (jfs_dirtable_inline(tlck->ip)) {
1564 * The table has been truncated, we've must have deleted
1565 * the last entry, so don't bother logging this
1569 metapage_homeok(mp);
1570 discard_metapage(mp);
1575 PXDaddress(pxd, mp->index);
1576 PXDlength(pxd, mp->logical_size >> tblk->sb->s_blocksize_bits);
1578 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1580 /* mark page as homeward bound */
1581 tlck->flag |= tlckWRITEPAGE;
1589 * function: log dtree tlock and format maplock to update bmap;
1591 static void dtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1592 struct tlock * tlck)
1594 struct metapage *mp;
1595 struct pxd_lock *pxdlock;
1600 /* initialize as REDOPAGE/NOREDOPAGE record format */
1601 lrd->log.redopage.type = cpu_to_le16(LOG_DTREE);
1602 lrd->log.redopage.l2linesize = cpu_to_le16(L2DTSLOTSIZE);
1604 pxd = &lrd->log.redopage.pxd;
1606 if (tlck->type & tlckBTROOT)
1607 lrd->log.redopage.type |= cpu_to_le16(LOG_BTROOT);
1610 * page extension via relocation: entry insertion;
1611 * page extension in-place: entry insertion;
1612 * new right page from page split, reinitialized in-line
1613 * root from root page split: entry insertion;
1615 if (tlck->type & (tlckNEW | tlckEXTEND)) {
1616 /* log after-image of the new page for logredo():
1617 * mark log (LOG_NEW) for logredo() to initialize
1618 * freelist and update bmap for alloc of the new page;
1620 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1621 if (tlck->type & tlckEXTEND)
1622 lrd->log.redopage.type |= cpu_to_le16(LOG_EXTEND);
1624 lrd->log.redopage.type |= cpu_to_le16(LOG_NEW);
1625 PXDaddress(pxd, mp->index);
1627 mp->logical_size >> tblk->sb->s_blocksize_bits);
1628 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1630 /* format a maplock for txUpdateMap() to update bPMAP for
1631 * alloc of the new page;
1633 if (tlck->type & tlckBTROOT)
1635 tlck->flag |= tlckUPDATEMAP;
1636 pxdlock = (struct pxd_lock *) & tlck->lock;
1637 pxdlock->flag = mlckALLOCPXD;
1638 pxdlock->pxd = *pxd;
1642 /* mark page as homeward bound */
1643 tlck->flag |= tlckWRITEPAGE;
1648 * entry insertion/deletion,
1649 * sibling page link update (old right page before split);
1651 if (tlck->type & (tlckENTRY | tlckRELINK)) {
1652 /* log after-image for logredo(): */
1653 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1654 PXDaddress(pxd, mp->index);
1656 mp->logical_size >> tblk->sb->s_blocksize_bits);
1657 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1659 /* mark page as homeward bound */
1660 tlck->flag |= tlckWRITEPAGE;
1665 * page deletion: page has been invalidated
1666 * page relocation: source extent
1668 * a maplock for free of the page has been formatted
1669 * at txLock() time);
1671 if (tlck->type & (tlckFREE | tlckRELOCATE)) {
1672 /* log LOG_NOREDOPAGE of the deleted page for logredo()
1673 * to start NoRedoPage filter and to update bmap for free
1674 * of the deletd page
1676 lrd->type = cpu_to_le16(LOG_NOREDOPAGE);
1677 pxdlock = (struct pxd_lock *) & tlck->lock;
1678 *pxd = pxdlock->pxd;
1679 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1681 /* a maplock for txUpdateMap() for free of the page
1682 * has been formatted at txLock() time;
1684 tlck->flag |= tlckUPDATEMAP;
1692 * function: log xtree tlock and format maplock to update bmap;
1694 static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1695 struct tlock * tlck)
1698 struct metapage *mp;
1700 struct xtlock *xtlck;
1701 struct maplock *maplock;
1702 struct xdlistlock *xadlock;
1703 struct pxd_lock *pxdlock;
1710 /* initialize as REDOPAGE/NOREDOPAGE record format */
1711 lrd->log.redopage.type = cpu_to_le16(LOG_XTREE);
1712 lrd->log.redopage.l2linesize = cpu_to_le16(L2XTSLOTSIZE);
1714 page_pxd = &lrd->log.redopage.pxd;
1716 if (tlck->type & tlckBTROOT) {
1717 lrd->log.redopage.type |= cpu_to_le16(LOG_BTROOT);
1718 p = &JFS_IP(ip)->i_xtroot;
1719 if (S_ISDIR(ip->i_mode))
1720 lrd->log.redopage.type |=
1721 cpu_to_le16(LOG_DIR_XTREE);
1723 p = (xtpage_t *) mp->data;
1724 next = le16_to_cpu(p->header.nextindex);
1726 xtlck = (struct xtlock *) & tlck->lock;
1728 maplock = (struct maplock *) & tlck->lock;
1729 xadlock = (struct xdlistlock *) maplock;
1732 * entry insertion/extension;
1733 * sibling page link update (old right page before split);
1735 if (tlck->type & (tlckNEW | tlckGROW | tlckRELINK)) {
1736 /* log after-image for logredo():
1737 * logredo() will update bmap for alloc of new/extended
1738 * extents (XAD_NEW|XAD_EXTEND) of XAD[lwm:next) from
1739 * after-image of XADlist;
1740 * logredo() resets (XAD_NEW|XAD_EXTEND) flag when
1741 * applying the after-image to the meta-data page.
1743 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1744 PXDaddress(page_pxd, mp->index);
1746 mp->logical_size >> tblk->sb->s_blocksize_bits);
1747 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1749 /* format a maplock for txUpdateMap() to update bPMAP
1750 * for alloc of new/extended extents of XAD[lwm:next)
1751 * from the page itself;
1752 * txUpdateMap() resets (XAD_NEW|XAD_EXTEND) flag.
1754 lwm = xtlck->lwm.offset;
1756 lwm = XTPAGEMAXSLOT;
1761 jfs_err("xtLog: lwm > next\n");
1764 tlck->flag |= tlckUPDATEMAP;
1765 xadlock->flag = mlckALLOCXADLIST;
1766 xadlock->count = next - lwm;
1767 if ((xadlock->count <= 4) && (tblk->xflag & COMMIT_LAZY)) {
1771 * Lazy commit may allow xtree to be modified before
1772 * txUpdateMap runs. Copy xad into linelock to
1773 * preserve correct data.
1775 * We can fit twice as may pxd's as xads in the lock
1777 xadlock->flag = mlckALLOCPXDLIST;
1778 pxd = xadlock->xdlist = &xtlck->pxdlock;
1779 for (i = 0; i < xadlock->count; i++) {
1780 PXDaddress(pxd, addressXAD(&p->xad[lwm + i]));
1781 PXDlength(pxd, lengthXAD(&p->xad[lwm + i]));
1782 p->xad[lwm + i].flag &=
1783 ~(XAD_NEW | XAD_EXTENDED);
1788 * xdlist will point to into inode's xtree, ensure
1789 * that transaction is not committed lazily.
1791 xadlock->flag = mlckALLOCXADLIST;
1792 xadlock->xdlist = &p->xad[lwm];
1793 tblk->xflag &= ~COMMIT_LAZY;
1795 jfs_info("xtLog: alloc ip:0x%p mp:0x%p tlck:0x%p lwm:%d "
1796 "count:%d", tlck->ip, mp, tlck, lwm, xadlock->count);
1801 /* mark page as homeward bound */
1802 tlck->flag |= tlckWRITEPAGE;
1808 * page deletion: file deletion/truncation (ref. xtTruncate())
1810 * (page will be invalidated after log is written and bmap
1811 * is updated from the page);
1813 if (tlck->type & tlckFREE) {
1814 /* LOG_NOREDOPAGE log for NoRedoPage filter:
1815 * if page free from file delete, NoRedoFile filter from
1816 * inode image of zero link count will subsume NoRedoPage
1817 * filters for each page;
1818 * if page free from file truncattion, write NoRedoPage
1821 * upadte of block allocation map for the page itself:
1822 * if page free from deletion and truncation, LOG_UPDATEMAP
1823 * log for the page itself is generated from processing
1824 * its parent page xad entries;
1826 /* if page free from file truncation, log LOG_NOREDOPAGE
1827 * of the deleted page for logredo() to start NoRedoPage
1828 * filter for the page;
1830 if (tblk->xflag & COMMIT_TRUNCATE) {
1831 /* write NOREDOPAGE for the page */
1832 lrd->type = cpu_to_le16(LOG_NOREDOPAGE);
1833 PXDaddress(page_pxd, mp->index);
1835 mp->logical_size >> tblk->sb->
1838 cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1840 if (tlck->type & tlckBTROOT) {
1841 /* Empty xtree must be logged */
1842 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1844 cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1848 /* init LOG_UPDATEMAP of the freed extents
1849 * XAD[XTENTRYSTART:hwm) from the deleted page itself
1850 * for logredo() to update bmap;
1852 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1853 lrd->log.updatemap.type = cpu_to_le16(LOG_FREEXADLIST);
1854 xtlck = (struct xtlock *) & tlck->lock;
1855 hwm = xtlck->hwm.offset;
1856 lrd->log.updatemap.nxd =
1857 cpu_to_le16(hwm - XTENTRYSTART + 1);
1858 /* reformat linelock for lmLog() */
1859 xtlck->header.offset = XTENTRYSTART;
1860 xtlck->header.length = hwm - XTENTRYSTART + 1;
1862 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1864 /* format a maplock for txUpdateMap() to update bmap
1865 * to free extents of XAD[XTENTRYSTART:hwm) from the
1866 * deleted page itself;
1868 tlck->flag |= tlckUPDATEMAP;
1869 xadlock->count = hwm - XTENTRYSTART + 1;
1870 if ((xadlock->count <= 4) && (tblk->xflag & COMMIT_LAZY)) {
1874 * Lazy commit may allow xtree to be modified before
1875 * txUpdateMap runs. Copy xad into linelock to
1876 * preserve correct data.
1878 * We can fit twice as may pxd's as xads in the lock
1880 xadlock->flag = mlckFREEPXDLIST;
1881 pxd = xadlock->xdlist = &xtlck->pxdlock;
1882 for (i = 0; i < xadlock->count; i++) {
1884 addressXAD(&p->xad[XTENTRYSTART + i]));
1886 lengthXAD(&p->xad[XTENTRYSTART + i]));
1891 * xdlist will point to into inode's xtree, ensure
1892 * that transaction is not committed lazily.
1894 xadlock->flag = mlckFREEXADLIST;
1895 xadlock->xdlist = &p->xad[XTENTRYSTART];
1896 tblk->xflag &= ~COMMIT_LAZY;
1898 jfs_info("xtLog: free ip:0x%p mp:0x%p count:%d lwm:2",
1899 tlck->ip, mp, xadlock->count);
1903 /* mark page as invalid */
1904 if (((tblk->xflag & COMMIT_PWMAP) || S_ISDIR(ip->i_mode))
1905 && !(tlck->type & tlckBTROOT))
1906 tlck->flag |= tlckFREEPAGE;
1908 else (tblk->xflag & COMMIT_PMAP)
1915 * page/entry truncation: file truncation (ref. xtTruncate())
1917 * |----------+------+------+---------------|
1919 * | | hwm - hwm before truncation
1920 * | next - truncation point
1921 * lwm - lwm before truncation
1924 if (tlck->type & tlckTRUNCATE) {
1925 /* This odd declaration suppresses a bogus gcc warning */
1926 pxd_t pxd = pxd; /* truncated extent of xad */
1930 * For truncation the entire linelock may be used, so it would
1931 * be difficult to store xad list in linelock itself.
1932 * Therefore, we'll just force transaction to be committed
1933 * synchronously, so that xtree pages won't be changed before
1936 tblk->xflag &= ~COMMIT_LAZY;
1937 lwm = xtlck->lwm.offset;
1939 lwm = XTPAGEMAXSLOT;
1940 hwm = xtlck->hwm.offset;
1941 twm = xtlck->twm.offset;
1946 /* log after-image for logredo():
1948 * logredo() will update bmap for alloc of new/extended
1949 * extents (XAD_NEW|XAD_EXTEND) of XAD[lwm:next) from
1950 * after-image of XADlist;
1951 * logredo() resets (XAD_NEW|XAD_EXTEND) flag when
1952 * applying the after-image to the meta-data page.
1954 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1955 PXDaddress(page_pxd, mp->index);
1957 mp->logical_size >> tblk->sb->s_blocksize_bits);
1958 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1961 * truncate entry XAD[twm == next - 1]:
1963 if (twm == next - 1) {
1964 /* init LOG_UPDATEMAP for logredo() to update bmap for
1965 * free of truncated delta extent of the truncated
1966 * entry XAD[next - 1]:
1967 * (xtlck->pxdlock = truncated delta extent);
1969 pxdlock = (struct pxd_lock *) & xtlck->pxdlock;
1970 /* assert(pxdlock->type & tlckTRUNCATE); */
1971 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1972 lrd->log.updatemap.type = cpu_to_le16(LOG_FREEPXD);
1973 lrd->log.updatemap.nxd = cpu_to_le16(1);
1974 lrd->log.updatemap.pxd = pxdlock->pxd;
1975 pxd = pxdlock->pxd; /* save to format maplock */
1977 cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1981 * free entries XAD[next:hwm]:
1984 /* init LOG_UPDATEMAP of the freed extents
1985 * XAD[next:hwm] from the deleted page itself
1986 * for logredo() to update bmap;
1988 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1989 lrd->log.updatemap.type =
1990 cpu_to_le16(LOG_FREEXADLIST);
1991 xtlck = (struct xtlock *) & tlck->lock;
1992 hwm = xtlck->hwm.offset;
1993 lrd->log.updatemap.nxd =
1994 cpu_to_le16(hwm - next + 1);
1995 /* reformat linelock for lmLog() */
1996 xtlck->header.offset = next;
1997 xtlck->header.length = hwm - next + 1;
2000 cpu_to_le32(lmLog(log, tblk, lrd, tlck));
2004 * format maplock(s) for txUpdateMap() to update bmap
2009 * allocate entries XAD[lwm:next):
2012 /* format a maplock for txUpdateMap() to update bPMAP
2013 * for alloc of new/extended extents of XAD[lwm:next)
2014 * from the page itself;
2015 * txUpdateMap() resets (XAD_NEW|XAD_EXTEND) flag.
2017 tlck->flag |= tlckUPDATEMAP;
2018 xadlock->flag = mlckALLOCXADLIST;
2019 xadlock->count = next - lwm;
2020 xadlock->xdlist = &p->xad[lwm];
2022 jfs_info("xtLog: alloc ip:0x%p mp:0x%p count:%d "
2024 tlck->ip, mp, xadlock->count, lwm, next);
2030 * truncate entry XAD[twm == next - 1]:
2032 if (twm == next - 1) {
2033 /* format a maplock for txUpdateMap() to update bmap
2034 * to free truncated delta extent of the truncated
2035 * entry XAD[next - 1];
2036 * (xtlck->pxdlock = truncated delta extent);
2038 tlck->flag |= tlckUPDATEMAP;
2039 pxdlock = (struct pxd_lock *) xadlock;
2040 pxdlock->flag = mlckFREEPXD;
2044 jfs_info("xtLog: truncate ip:0x%p mp:0x%p count:%d "
2045 "hwm:%d", ip, mp, pxdlock->count, hwm);
2051 * free entries XAD[next:hwm]:
2054 /* format a maplock for txUpdateMap() to update bmap
2055 * to free extents of XAD[next:hwm] from thedeleted
2058 tlck->flag |= tlckUPDATEMAP;
2059 xadlock->flag = mlckFREEXADLIST;
2060 xadlock->count = hwm - next + 1;
2061 xadlock->xdlist = &p->xad[next];
2063 jfs_info("xtLog: free ip:0x%p mp:0x%p count:%d "
2065 tlck->ip, mp, xadlock->count, next, hwm);
2069 /* mark page as homeward bound */
2070 tlck->flag |= tlckWRITEPAGE;
2078 * function: log from maplock of freed data extents;
2080 static void mapLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
2081 struct tlock * tlck)
2083 struct pxd_lock *pxdlock;
2088 * page relocation: free the source page extent
2090 * a maplock for txUpdateMap() for free of the page
2091 * has been formatted at txLock() time saving the src
2092 * relocated page address;
2094 if (tlck->type & tlckRELOCATE) {
2095 /* log LOG_NOREDOPAGE of the old relocated page
2096 * for logredo() to start NoRedoPage filter;
2098 lrd->type = cpu_to_le16(LOG_NOREDOPAGE);
2099 pxdlock = (struct pxd_lock *) & tlck->lock;
2100 pxd = &lrd->log.redopage.pxd;
2101 *pxd = pxdlock->pxd;
2102 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
2104 /* (N.B. currently, logredo() does NOT update bmap
2105 * for free of the page itself for (LOG_XTREE|LOG_NOREDOPAGE);
2106 * if page free from relocation, LOG_UPDATEMAP log is
2107 * specifically generated now for logredo()
2108 * to update bmap for free of src relocated page;
2109 * (new flag LOG_RELOCATE may be introduced which will
2110 * inform logredo() to start NORedoPage filter and also
2111 * update block allocation map at the same time, thus
2112 * avoiding an extra log write);
2114 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
2115 lrd->log.updatemap.type = cpu_to_le16(LOG_FREEPXD);
2116 lrd->log.updatemap.nxd = cpu_to_le16(1);
2117 lrd->log.updatemap.pxd = pxdlock->pxd;
2118 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
2120 /* a maplock for txUpdateMap() for free of the page
2121 * has been formatted at txLock() time;
2123 tlck->flag |= tlckUPDATEMAP;
2128 * Otherwise it's not a relocate request
2132 /* log LOG_UPDATEMAP for logredo() to update bmap for
2133 * free of truncated/relocated delta extent of the data;
2134 * e.g.: external EA extent, relocated/truncated extent
2135 * from xtTailgate();
2137 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
2138 pxdlock = (struct pxd_lock *) & tlck->lock;
2139 nlock = pxdlock->index;
2140 for (i = 0; i < nlock; i++, pxdlock++) {
2141 if (pxdlock->flag & mlckALLOCPXD)
2142 lrd->log.updatemap.type =
2143 cpu_to_le16(LOG_ALLOCPXD);
2145 lrd->log.updatemap.type =
2146 cpu_to_le16(LOG_FREEPXD);
2147 lrd->log.updatemap.nxd = cpu_to_le16(1);
2148 lrd->log.updatemap.pxd = pxdlock->pxd;
2150 cpu_to_le32(lmLog(log, tblk, lrd, NULL));
2151 jfs_info("mapLog: xaddr:0x%lx xlen:0x%x",
2152 (ulong) addressPXD(&pxdlock->pxd),
2153 lengthPXD(&pxdlock->pxd));
2157 tlck->flag |= tlckUPDATEMAP;
2164 * function: acquire maplock for EA/ACL extents or
2165 * set COMMIT_INLINE flag;
2167 void txEA(tid_t tid, struct inode *ip, dxd_t * oldea, dxd_t * newea)
2169 struct tlock *tlck = NULL;
2170 struct pxd_lock *maplock = NULL, *pxdlock = NULL;
2173 * format maplock for alloc of new EA extent
2176 /* Since the newea could be a completely zeroed entry we need to
2177 * check for the two flags which indicate we should actually
2178 * commit new EA data
2180 if (newea->flag & DXD_EXTENT) {
2181 tlck = txMaplock(tid, ip, tlckMAP);
2182 maplock = (struct pxd_lock *) & tlck->lock;
2183 pxdlock = (struct pxd_lock *) maplock;
2184 pxdlock->flag = mlckALLOCPXD;
2185 PXDaddress(&pxdlock->pxd, addressDXD(newea));
2186 PXDlength(&pxdlock->pxd, lengthDXD(newea));
2189 } else if (newea->flag & DXD_INLINE) {
2192 set_cflag(COMMIT_Inlineea, ip);
2197 * format maplock for free of old EA extent
2199 if (!test_cflag(COMMIT_Nolink, ip) && oldea->flag & DXD_EXTENT) {
2201 tlck = txMaplock(tid, ip, tlckMAP);
2202 maplock = (struct pxd_lock *) & tlck->lock;
2203 pxdlock = (struct pxd_lock *) maplock;
2206 pxdlock->flag = mlckFREEPXD;
2207 PXDaddress(&pxdlock->pxd, addressDXD(oldea));
2208 PXDlength(&pxdlock->pxd, lengthDXD(oldea));
2216 * function: synchronously write pages locked by transaction
2217 * after txLog() but before txUpdateMap();
2219 static void txForce(struct tblock * tblk)
2223 struct metapage *mp;
2226 * reverse the order of transaction tlocks in
2227 * careful update order of address index pages
2228 * (right to left, bottom up)
2230 tlck = lid_to_tlock(tblk->next);
2234 tlck = lid_to_tlock(lid);
2236 tlck->next = tblk->next;
2242 * synchronously write the page, and
2243 * hold the page for txUpdateMap();
2245 for (lid = tblk->next; lid; lid = next) {
2246 tlck = lid_to_tlock(lid);
2249 if ((mp = tlck->mp) != NULL &&
2250 (tlck->type & tlckBTROOT) == 0) {
2251 assert(mp->xflag & COMMIT_PAGE);
2253 if (tlck->flag & tlckWRITEPAGE) {
2254 tlck->flag &= ~tlckWRITEPAGE;
2256 /* do not release page to freelist */
2260 * The "right" thing to do here is to
2261 * synchronously write the metadata.
2262 * With the current implementation this
2263 * is hard since write_metapage requires
2264 * us to kunmap & remap the page. If we
2265 * have tlocks pointing into the metadata
2266 * pages, we don't want to do this. I think
2267 * we can get by with synchronously writing
2268 * the pages when they are released.
2270 assert(mp->nohomeok);
2271 set_bit(META_dirty, &mp->flag);
2272 set_bit(META_sync, &mp->flag);
2282 * function: update persistent allocation map (and working map
2287 static void txUpdateMap(struct tblock * tblk)
2290 struct inode *ipimap;
2293 struct maplock *maplock;
2294 struct pxd_lock pxdlock;
2297 struct metapage *mp = NULL;
2299 ipimap = JFS_SBI(tblk->sb)->ipimap;
2301 maptype = (tblk->xflag & COMMIT_PMAP) ? COMMIT_PMAP : COMMIT_PWMAP;
2305 * update block allocation map
2307 * update allocation state in pmap (and wmap) and
2308 * update lsn of the pmap page;
2311 * scan each tlock/page of transaction for block allocation/free:
2313 * for each tlock/page of transaction, update map.
2314 * ? are there tlock for pmap and pwmap at the same time ?
2316 for (lid = tblk->next; lid; lid = tlck->next) {
2317 tlck = lid_to_tlock(lid);
2319 if ((tlck->flag & tlckUPDATEMAP) == 0)
2322 if (tlck->flag & tlckFREEPAGE) {
2324 * Another thread may attempt to reuse freed space
2325 * immediately, so we want to get rid of the metapage
2326 * before anyone else has a chance to get it.
2327 * Lock metapage, update maps, then invalidate
2331 ASSERT(mp->xflag & COMMIT_PAGE);
2337 * . in-line PXD list:
2338 * . out-of-line XAD list:
2340 maplock = (struct maplock *) & tlck->lock;
2341 nlock = maplock->index;
2343 for (k = 0; k < nlock; k++, maplock++) {
2345 * allocate blocks in persistent map:
2347 * blocks have been allocated from wmap at alloc time;
2349 if (maplock->flag & mlckALLOC) {
2350 txAllocPMap(ipimap, maplock, tblk);
2353 * free blocks in persistent and working map:
2354 * blocks will be freed in pmap and then in wmap;
2356 * ? tblock specifies the PMAP/PWMAP based upon
2359 * free blocks in persistent map:
2360 * blocks will be freed from wmap at last reference
2361 * release of the object for regular files;
2363 * Alway free blocks from both persistent & working
2364 * maps for directories
2366 else { /* (maplock->flag & mlckFREE) */
2368 if (tlck->flag & tlckDIRECTORY)
2369 txFreeMap(ipimap, maplock,
2370 tblk, COMMIT_PWMAP);
2372 txFreeMap(ipimap, maplock,
2376 if (tlck->flag & tlckFREEPAGE) {
2377 if (!(tblk->flag & tblkGC_LAZY)) {
2378 /* This is equivalent to txRelease */
2379 ASSERT(mp->lid == lid);
2382 assert(mp->nohomeok == 1);
2383 metapage_homeok(mp);
2384 discard_metapage(mp);
2389 * update inode allocation map
2391 * update allocation state in pmap and
2392 * update lsn of the pmap page;
2393 * update in-memory inode flag/state
2395 * unlock mapper/write lock
2397 if (tblk->xflag & COMMIT_CREATE) {
2398 diUpdatePMap(ipimap, tblk->ino, false, tblk);
2399 /* update persistent block allocation map
2400 * for the allocation of inode extent;
2402 pxdlock.flag = mlckALLOCPXD;
2403 pxdlock.pxd = tblk->u.ixpxd;
2405 txAllocPMap(ipimap, (struct maplock *) & pxdlock, tblk);
2406 } else if (tblk->xflag & COMMIT_DELETE) {
2408 diUpdatePMap(ipimap, ip->i_ino, true, tblk);
2416 * function: allocate from persistent map;
2425 * allocate from persistent map;
2426 * free from persistent map;
2427 * (e.g., tmp file - free from working map at releae
2428 * of last reference);
2429 * free from persistent and working map;
2431 * lsn - log sequence number;
2433 static void txAllocPMap(struct inode *ip, struct maplock * maplock,
2434 struct tblock * tblk)
2436 struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
2437 struct xdlistlock *xadlistlock;
2441 struct pxd_lock *pxdlock;
2442 struct xdlistlock *pxdlistlock;
2447 * allocate from persistent map;
2449 if (maplock->flag & mlckALLOCXADLIST) {
2450 xadlistlock = (struct xdlistlock *) maplock;
2451 xad = xadlistlock->xdlist;
2452 for (n = 0; n < xadlistlock->count; n++, xad++) {
2453 if (xad->flag & (XAD_NEW | XAD_EXTENDED)) {
2454 xaddr = addressXAD(xad);
2455 xlen = lengthXAD(xad);
2456 dbUpdatePMap(ipbmap, false, xaddr,
2458 xad->flag &= ~(XAD_NEW | XAD_EXTENDED);
2459 jfs_info("allocPMap: xaddr:0x%lx xlen:%d",
2460 (ulong) xaddr, xlen);
2463 } else if (maplock->flag & mlckALLOCPXD) {
2464 pxdlock = (struct pxd_lock *) maplock;
2465 xaddr = addressPXD(&pxdlock->pxd);
2466 xlen = lengthPXD(&pxdlock->pxd);
2467 dbUpdatePMap(ipbmap, false, xaddr, (s64) xlen, tblk);
2468 jfs_info("allocPMap: xaddr:0x%lx xlen:%d", (ulong) xaddr, xlen);
2469 } else { /* (maplock->flag & mlckALLOCPXDLIST) */
2471 pxdlistlock = (struct xdlistlock *) maplock;
2472 pxd = pxdlistlock->xdlist;
2473 for (n = 0; n < pxdlistlock->count; n++, pxd++) {
2474 xaddr = addressPXD(pxd);
2475 xlen = lengthPXD(pxd);
2476 dbUpdatePMap(ipbmap, false, xaddr, (s64) xlen,
2478 jfs_info("allocPMap: xaddr:0x%lx xlen:%d",
2479 (ulong) xaddr, xlen);
2487 * function: free from persistent and/or working map;
2489 * todo: optimization
2491 void txFreeMap(struct inode *ip,
2492 struct maplock * maplock, struct tblock * tblk, int maptype)
2494 struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
2495 struct xdlistlock *xadlistlock;
2499 struct pxd_lock *pxdlock;
2500 struct xdlistlock *pxdlistlock;
2504 jfs_info("txFreeMap: tblk:0x%p maplock:0x%p maptype:0x%x",
2505 tblk, maplock, maptype);
2508 * free from persistent map;
2510 if (maptype == COMMIT_PMAP || maptype == COMMIT_PWMAP) {
2511 if (maplock->flag & mlckFREEXADLIST) {
2512 xadlistlock = (struct xdlistlock *) maplock;
2513 xad = xadlistlock->xdlist;
2514 for (n = 0; n < xadlistlock->count; n++, xad++) {
2515 if (!(xad->flag & XAD_NEW)) {
2516 xaddr = addressXAD(xad);
2517 xlen = lengthXAD(xad);
2518 dbUpdatePMap(ipbmap, true, xaddr,
2520 jfs_info("freePMap: xaddr:0x%lx "
2522 (ulong) xaddr, xlen);
2525 } else if (maplock->flag & mlckFREEPXD) {
2526 pxdlock = (struct pxd_lock *) maplock;
2527 xaddr = addressPXD(&pxdlock->pxd);
2528 xlen = lengthPXD(&pxdlock->pxd);
2529 dbUpdatePMap(ipbmap, true, xaddr, (s64) xlen,
2531 jfs_info("freePMap: xaddr:0x%lx xlen:%d",
2532 (ulong) xaddr, xlen);
2533 } else { /* (maplock->flag & mlckALLOCPXDLIST) */
2535 pxdlistlock = (struct xdlistlock *) maplock;
2536 pxd = pxdlistlock->xdlist;
2537 for (n = 0; n < pxdlistlock->count; n++, pxd++) {
2538 xaddr = addressPXD(pxd);
2539 xlen = lengthPXD(pxd);
2540 dbUpdatePMap(ipbmap, true, xaddr,
2542 jfs_info("freePMap: xaddr:0x%lx xlen:%d",
2543 (ulong) xaddr, xlen);
2549 * free from working map;
2551 if (maptype == COMMIT_PWMAP || maptype == COMMIT_WMAP) {
2552 if (maplock->flag & mlckFREEXADLIST) {
2553 xadlistlock = (struct xdlistlock *) maplock;
2554 xad = xadlistlock->xdlist;
2555 for (n = 0; n < xadlistlock->count; n++, xad++) {
2556 xaddr = addressXAD(xad);
2557 xlen = lengthXAD(xad);
2558 dbFree(ip, xaddr, (s64) xlen);
2560 jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
2561 (ulong) xaddr, xlen);
2563 } else if (maplock->flag & mlckFREEPXD) {
2564 pxdlock = (struct pxd_lock *) maplock;
2565 xaddr = addressPXD(&pxdlock->pxd);
2566 xlen = lengthPXD(&pxdlock->pxd);
2567 dbFree(ip, xaddr, (s64) xlen);
2568 jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
2569 (ulong) xaddr, xlen);
2570 } else { /* (maplock->flag & mlckFREEPXDLIST) */
2572 pxdlistlock = (struct xdlistlock *) maplock;
2573 pxd = pxdlistlock->xdlist;
2574 for (n = 0; n < pxdlistlock->count; n++, pxd++) {
2575 xaddr = addressPXD(pxd);
2576 xlen = lengthPXD(pxd);
2577 dbFree(ip, xaddr, (s64) xlen);
2578 jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
2579 (ulong) xaddr, xlen);
2588 * function: remove tlock from inode anonymous locklist
2590 void txFreelock(struct inode *ip)
2592 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
2593 struct tlock *xtlck, *tlck;
2594 lid_t xlid = 0, lid;
2596 if (!jfs_ip->atlhead)
2600 xtlck = (struct tlock *) &jfs_ip->atlhead;
2602 while ((lid = xtlck->next) != 0) {
2603 tlck = lid_to_tlock(lid);
2604 if (tlck->flag & tlckFREELOCK) {
2605 xtlck->next = tlck->next;
2613 if (jfs_ip->atlhead)
2614 jfs_ip->atltail = xlid;
2616 jfs_ip->atltail = 0;
2618 * If inode was on anon_list, remove it
2620 list_del_init(&jfs_ip->anon_inode_list);
2628 * function: abort tx before commit;
2630 * frees line-locks and segment locks for all
2631 * segments in comdata structure.
2632 * Optionally sets state of file-system to FM_DIRTY in super-block.
2633 * log age of page-frames in memory for which caller has
2634 * are reset to 0 (to avoid logwarap).
2636 void txAbort(tid_t tid, int dirty)
2639 struct metapage *mp;
2640 struct tblock *tblk = tid_to_tblock(tid);
2644 * free tlocks of the transaction
2646 for (lid = tblk->next; lid; lid = next) {
2647 tlck = lid_to_tlock(lid);
2650 JFS_IP(tlck->ip)->xtlid = 0;
2656 * reset lsn of page to avoid logwarap:
2658 * (page may have been previously committed by another
2659 * transaction(s) but has not been paged, i.e.,
2660 * it may be on logsync list even though it has not
2661 * been logged for the current tx.)
2663 if (mp->xflag & COMMIT_PAGE && mp->lsn)
2666 /* insert tlock at head of freelist */
2672 /* caller will free the transaction block */
2674 tblk->next = tblk->last = 0;
2677 * mark filesystem dirty
2680 jfs_error(tblk->sb, "txAbort");
2686 * txLazyCommit(void)
2688 * All transactions except those changing ipimap (COMMIT_FORCE) are
2689 * processed by this routine. This insures that the inode and block
2690 * allocation maps are updated in order. For synchronous transactions,
2691 * let the user thread finish processing after txUpdateMap() is called.
2693 static void txLazyCommit(struct tblock * tblk)
2695 struct jfs_log *log;
2697 while (((tblk->flag & tblkGC_READY) == 0) &&
2698 ((tblk->flag & tblkGC_UNLOCKED) == 0)) {
2699 /* We must have gotten ahead of the user thread
2701 jfs_info("jfs_lazycommit: tblk 0x%p not unlocked", tblk);
2705 jfs_info("txLazyCommit: processing tblk 0x%p", tblk);
2709 log = (struct jfs_log *) JFS_SBI(tblk->sb)->log;
2711 spin_lock_irq(&log->gclock); // LOGGC_LOCK
2713 tblk->flag |= tblkGC_COMMITTED;
2715 if (tblk->flag & tblkGC_READY)
2718 wake_up_all(&tblk->gcwait); // LOGGC_WAKEUP
2721 * Can't release log->gclock until we've tested tblk->flag
2723 if (tblk->flag & tblkGC_LAZY) {
2724 spin_unlock_irq(&log->gclock); // LOGGC_UNLOCK
2726 tblk->flag &= ~tblkGC_LAZY;
2727 txEnd(tblk - TxBlock); /* Convert back to tid */
2729 spin_unlock_irq(&log->gclock); // LOGGC_UNLOCK
2731 jfs_info("txLazyCommit: done: tblk = 0x%p", tblk);
2735 * jfs_lazycommit(void)
2737 * To be run as a kernel daemon. If lbmIODone is called in an interrupt
2738 * context, or where blocking is not wanted, this routine will process
2739 * committed transactions from the unlock queue.
2741 int jfs_lazycommit(void *arg)
2744 struct tblock *tblk;
2745 unsigned long flags;
2746 struct jfs_sb_info *sbi;
2750 jfs_commit_thread_waking = 0; /* OK to wake another thread */
2751 while (!list_empty(&TxAnchor.unlock_queue)) {
2753 list_for_each_entry(tblk, &TxAnchor.unlock_queue,
2756 sbi = JFS_SBI(tblk->sb);
2758 * For each volume, the transactions must be
2759 * handled in order. If another commit thread
2760 * is handling a tblk for this superblock,
2763 if (sbi->commit_state & IN_LAZYCOMMIT)
2766 sbi->commit_state |= IN_LAZYCOMMIT;
2770 * Remove transaction from queue
2772 list_del(&tblk->cqueue);
2778 sbi->commit_state &= ~IN_LAZYCOMMIT;
2780 * Don't continue in the for loop. (We can't
2781 * anyway, it's unsafe!) We want to go back to
2782 * the beginning of the list.
2787 /* If there was nothing to do, don't continue */
2791 /* In case a wakeup came while all threads were active */
2792 jfs_commit_thread_waking = 0;
2794 if (freezing(current)) {
2798 DECLARE_WAITQUEUE(wq, current);
2800 add_wait_queue(&jfs_commit_thread_wait, &wq);
2801 set_current_state(TASK_INTERRUPTIBLE);
2804 __set_current_state(TASK_RUNNING);
2805 remove_wait_queue(&jfs_commit_thread_wait, &wq);
2807 } while (!kthread_should_stop());
2809 if (!list_empty(&TxAnchor.unlock_queue))
2810 jfs_err("jfs_lazycommit being killed w/pending transactions!");
2812 jfs_info("jfs_lazycommit being killed\n");
2816 void txLazyUnlock(struct tblock * tblk)
2818 unsigned long flags;
2822 list_add_tail(&tblk->cqueue, &TxAnchor.unlock_queue);
2824 * Don't wake up a commit thread if there is already one servicing
2825 * this superblock, or if the last one we woke up hasn't started yet.
2827 if (!(JFS_SBI(tblk->sb)->commit_state & IN_LAZYCOMMIT) &&
2828 !jfs_commit_thread_waking) {
2829 jfs_commit_thread_waking = 1;
2830 wake_up(&jfs_commit_thread_wait);
2835 static void LogSyncRelease(struct metapage * mp)
2837 struct jfs_log *log = mp->log;
2839 assert(mp->nohomeok);
2841 metapage_homeok(mp);
2847 * Block all new transactions and push anonymous transactions to
2850 * This does almost the same thing as jfs_sync below. We don't
2851 * worry about deadlocking when jfs_tlocks_low is set, since we would
2852 * expect jfs_sync to get us out of that jam.
2854 void txQuiesce(struct super_block *sb)
2857 struct jfs_inode_info *jfs_ip;
2858 struct jfs_log *log = JFS_SBI(sb)->log;
2861 set_bit(log_QUIESCE, &log->flag);
2865 while (!list_empty(&TxAnchor.anon_list)) {
2866 jfs_ip = list_entry(TxAnchor.anon_list.next,
2867 struct jfs_inode_info,
2869 ip = &jfs_ip->vfs_inode;
2872 * inode will be removed from anonymous list
2873 * when it is committed
2876 tid = txBegin(ip->i_sb, COMMIT_INODE | COMMIT_FORCE);
2877 mutex_lock(&jfs_ip->commit_mutex);
2878 txCommit(tid, 1, &ip, 0);
2880 mutex_unlock(&jfs_ip->commit_mutex);
2882 * Just to be safe. I don't know how
2883 * long we can run without blocking
2890 * If jfs_sync is running in parallel, there could be some inodes
2891 * on anon_list2. Let's check.
2893 if (!list_empty(&TxAnchor.anon_list2)) {
2894 list_splice(&TxAnchor.anon_list2, &TxAnchor.anon_list);
2895 INIT_LIST_HEAD(&TxAnchor.anon_list2);
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;
2935 * write each inode on the anonymous inode list
2938 while (jfs_tlocks_low && !list_empty(&TxAnchor.anon_list)) {
2939 jfs_ip = list_entry(TxAnchor.anon_list.next,
2940 struct jfs_inode_info,
2942 ip = &jfs_ip->vfs_inode;
2946 * Inode is being freed
2948 list_del_init(&jfs_ip->anon_inode_list);
2949 } else if (mutex_trylock(&jfs_ip->commit_mutex)) {
2951 * inode will be removed from anonymous list
2952 * when it is committed
2955 tid = txBegin(ip->i_sb, COMMIT_INODE);
2956 rc = txCommit(tid, 1, &ip, 0);
2958 mutex_unlock(&jfs_ip->commit_mutex);
2962 * Just to be safe. I don't know how
2963 * long we can run without blocking
2968 /* We can't get the commit mutex. It may
2969 * be held by a thread waiting for tlock's
2970 * so let's not block here. Save it to
2971 * put back on the anon_list.
2974 /* Take off anon_list */
2975 list_del(&jfs_ip->anon_inode_list);
2977 /* Put on anon_list2 */
2978 list_add(&jfs_ip->anon_inode_list,
2979 &TxAnchor.anon_list2);
2986 /* Add anon_list2 back to anon_list */
2987 list_splice_init(&TxAnchor.anon_list2, &TxAnchor.anon_list);
2989 if (freezing(current)) {
2993 set_current_state(TASK_INTERRUPTIBLE);
2996 __set_current_state(TASK_RUNNING);
2998 } while (!kthread_should_stop());
3000 jfs_info("jfs_sync being killed");
3004 #if defined(CONFIG_PROC_FS) && defined(CONFIG_JFS_DEBUG)
3005 int jfs_txanchor_read(char *buffer, char **start, off_t offset, int length,
3006 int *eof, void *data)
3015 waitqueue_active(&TxAnchor.freewait) ? "active" : "empty";
3017 waitqueue_active(&TxAnchor.freelockwait) ? "active" : "empty";
3019 waitqueue_active(&TxAnchor.lowlockwait) ? "active" : "empty";
3021 len += sprintf(buffer,
3027 "freelockwait = %s\n"
3028 "lowlockwait = %s\n"
3029 "tlocksInUse = %d\n"
3030 "jfs_tlocks_low = %d\n"
3031 "unlock_queue is %sempty\n",
3037 TxAnchor.tlocksInUse,
3039 list_empty(&TxAnchor.unlock_queue) ? "" : "not ");
3042 *start = buffer + begin;
3057 #if defined(CONFIG_PROC_FS) && defined(CONFIG_JFS_STATISTICS)
3058 int jfs_txstats_read(char *buffer, char **start, off_t offset, int length,
3059 int *eof, void *data)
3064 len += sprintf(buffer,
3067 "calls to txBegin = %d\n"
3068 "txBegin blocked by sync barrier = %d\n"
3069 "txBegin blocked by tlocks low = %d\n"
3070 "txBegin blocked by no free tid = %d\n"
3071 "calls to txBeginAnon = %d\n"
3072 "txBeginAnon blocked by sync barrier = %d\n"
3073 "txBeginAnon blocked by tlocks low = %d\n"
3074 "calls to txLockAlloc = %d\n"
3075 "tLockAlloc blocked by no free lock = %d\n",
3077 TxStat.txBegin_barrier,
3078 TxStat.txBegin_lockslow,
3079 TxStat.txBegin_freetid,
3081 TxStat.txBeginAnon_barrier,
3082 TxStat.txBeginAnon_lockslow,
3084 TxStat.txLockAlloc_freelock);
3087 *start = buffer + begin;