2 * Copyright (C) International Business Machines Corp., 2000-2005
3 * Portions Copyright (C) Christoph Hellwig, 2001-2002
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
13 * the GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
21 * jfs_txnmgr.c: transaction manager
24 * transaction starts with txBegin() and ends with txCommit()
27 * tlock is acquired at the time of update;
28 * (obviate scan at commit time for xtree and dtree)
29 * tlock and mp points to each other;
30 * (no hashlist for mp -> tlock).
33 * tlock on in-memory inode:
34 * in-place tlock in the in-memory inode itself;
35 * converted to page lock by iWrite() at commit time.
37 * tlock during write()/mmap() under anonymous transaction (tid = 0):
38 * transferred (?) to transaction at commit time.
40 * use the page itself to update allocation maps
41 * (obviate intermediate replication of allocation/deallocation data)
42 * hold on to mp+lock thru update of maps
46 #include <linux/vmalloc.h>
47 #include <linux/completion.h>
48 #include <linux/freezer.h>
49 #include <linux/module.h>
50 #include <linux/moduleparam.h>
51 #include <linux/kthread.h>
52 #include <linux/seq_file.h>
53 #include "jfs_incore.h"
54 #include "jfs_inode.h"
55 #include "jfs_filsys.h"
56 #include "jfs_metapage.h"
57 #include "jfs_dinode.h"
60 #include "jfs_superblock.h"
61 #include "jfs_debug.h"
64 * transaction management structures
67 int freetid; /* index of a free tid structure */
68 int freelock; /* index first free lock word */
69 wait_queue_head_t freewait; /* eventlist of free tblock */
70 wait_queue_head_t freelockwait; /* eventlist of free tlock */
71 wait_queue_head_t lowlockwait; /* eventlist of ample tlocks */
72 int tlocksInUse; /* Number of tlocks in use */
73 spinlock_t LazyLock; /* synchronize sync_queue & unlock_queue */
74 /* struct tblock *sync_queue; * Transactions waiting for data sync */
75 struct list_head unlock_queue; /* Txns waiting to be released */
76 struct list_head anon_list; /* inodes having anonymous txns */
77 struct list_head anon_list2; /* inodes having anonymous txns
78 that couldn't be sync'ed */
81 int jfs_tlocks_low; /* Indicates low number of available tlocks */
83 #ifdef CONFIG_JFS_STATISTICS
87 uint txBegin_lockslow;
90 uint txBeginAnon_barrier;
91 uint txBeginAnon_lockslow;
93 uint txLockAlloc_freelock;
97 static int nTxBlock = -1; /* number of transaction blocks */
98 module_param(nTxBlock, int, 0);
99 MODULE_PARM_DESC(nTxBlock,
100 "Number of transaction blocks (max:65536)");
102 static int nTxLock = -1; /* number of transaction locks */
103 module_param(nTxLock, int, 0);
104 MODULE_PARM_DESC(nTxLock,
105 "Number of transaction locks (max:65536)");
107 struct tblock *TxBlock; /* transaction block table */
108 static int TxLockLWM; /* Low water mark for number of txLocks used */
109 static int TxLockHWM; /* High water mark for number of txLocks used */
110 static int TxLockVHWM; /* Very High water mark */
111 struct tlock *TxLock; /* transaction lock table */
114 * transaction management lock
116 static DEFINE_SPINLOCK(jfsTxnLock);
118 #define TXN_LOCK() spin_lock(&jfsTxnLock)
119 #define TXN_UNLOCK() spin_unlock(&jfsTxnLock)
121 #define LAZY_LOCK_INIT() spin_lock_init(&TxAnchor.LazyLock);
122 #define LAZY_LOCK(flags) spin_lock_irqsave(&TxAnchor.LazyLock, flags)
123 #define LAZY_UNLOCK(flags) spin_unlock_irqrestore(&TxAnchor.LazyLock, flags)
125 static DECLARE_WAIT_QUEUE_HEAD(jfs_commit_thread_wait);
126 static int jfs_commit_thread_waking;
129 * Retry logic exist outside these macros to protect from spurrious wakeups.
131 static inline void TXN_SLEEP_DROP_LOCK(wait_queue_head_t * event)
133 DECLARE_WAITQUEUE(wait, current);
135 add_wait_queue(event, &wait);
136 set_current_state(TASK_UNINTERRUPTIBLE);
139 __set_current_state(TASK_RUNNING);
140 remove_wait_queue(event, &wait);
143 #define TXN_SLEEP(event)\
145 TXN_SLEEP_DROP_LOCK(event);\
149 #define TXN_WAKEUP(event) wake_up_all(event)
155 tid_t maxtid; /* 4: biggest tid ever used */
156 lid_t maxlid; /* 4: biggest lid ever used */
157 int ntid; /* 4: # of transactions performed */
158 int nlid; /* 4: # of tlocks acquired */
159 int waitlock; /* 4: # of tlock wait */
165 static int diLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
166 struct tlock * tlck, struct commit * cd);
167 static int dataLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
168 struct tlock * tlck);
169 static void dtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
170 struct tlock * tlck);
171 static void mapLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
172 struct tlock * tlck);
173 static void txAllocPMap(struct inode *ip, struct maplock * maplock,
174 struct tblock * tblk);
175 static void txForce(struct tblock * tblk);
176 static int txLog(struct jfs_log * log, struct tblock * tblk,
178 static void txUpdateMap(struct tblock * tblk);
179 static void txRelease(struct tblock * tblk);
180 static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
181 struct tlock * tlck);
182 static void LogSyncRelease(struct metapage * mp);
185 * transaction block/lock management
186 * ---------------------------------
190 * Get a transaction lock from the free list. If the number in use is
191 * greater than the high water mark, wake up the sync daemon. This should
192 * free some anonymous transaction locks. (TXN_LOCK must be held.)
194 static lid_t txLockAlloc(void)
198 INCREMENT(TxStat.txLockAlloc);
199 if (!TxAnchor.freelock) {
200 INCREMENT(TxStat.txLockAlloc_freelock);
203 while (!(lid = TxAnchor.freelock))
204 TXN_SLEEP(&TxAnchor.freelockwait);
205 TxAnchor.freelock = TxLock[lid].next;
206 HIGHWATERMARK(stattx.maxlid, lid);
207 if ((++TxAnchor.tlocksInUse > TxLockHWM) && (jfs_tlocks_low == 0)) {
208 jfs_info("txLockAlloc tlocks low");
210 wake_up_process(jfsSyncThread);
216 static void txLockFree(lid_t lid)
219 TxLock[lid].next = TxAnchor.freelock;
220 TxAnchor.freelock = lid;
221 TxAnchor.tlocksInUse--;
222 if (jfs_tlocks_low && (TxAnchor.tlocksInUse < TxLockLWM)) {
223 jfs_info("txLockFree jfs_tlocks_low no more");
225 TXN_WAKEUP(&TxAnchor.lowlockwait);
227 TXN_WAKEUP(&TxAnchor.freelockwait);
233 * FUNCTION: initialize transaction management structures
237 * serialization: single thread at jfs_init()
244 /* Set defaults for nTxLock and nTxBlock if unset */
247 if (nTxBlock == -1) {
248 /* Base default on memory size */
250 if (si.totalram > (256 * 1024)) /* 1 GB */
253 nTxLock = si.totalram >> 2;
254 } else if (nTxBlock > (8 * 1024))
257 nTxLock = nTxBlock << 3;
260 nTxBlock = nTxLock >> 3;
262 /* Verify tunable parameters */
264 nTxBlock = 16; /* No one should set it this low */
265 if (nTxBlock > 65536)
268 nTxLock = 256; /* No one should set it this low */
272 printk(KERN_INFO "JFS: nTxBlock = %d, nTxLock = %d\n",
275 * initialize transaction block (tblock) table
277 * transaction id (tid) = tblock index
278 * tid = 0 is reserved.
280 TxLockLWM = (nTxLock * 4) / 10;
281 TxLockHWM = (nTxLock * 7) / 10;
282 TxLockVHWM = (nTxLock * 8) / 10;
284 size = sizeof(struct tblock) * nTxBlock;
285 TxBlock = vmalloc(size);
289 for (k = 1; k < nTxBlock - 1; k++) {
290 TxBlock[k].next = k + 1;
291 init_waitqueue_head(&TxBlock[k].gcwait);
292 init_waitqueue_head(&TxBlock[k].waitor);
295 init_waitqueue_head(&TxBlock[k].gcwait);
296 init_waitqueue_head(&TxBlock[k].waitor);
298 TxAnchor.freetid = 1;
299 init_waitqueue_head(&TxAnchor.freewait);
301 stattx.maxtid = 1; /* statistics */
304 * initialize transaction lock (tlock) table
306 * transaction lock id = tlock index
307 * tlock id = 0 is reserved.
309 size = sizeof(struct tlock) * nTxLock;
310 TxLock = vmalloc(size);
311 if (TxLock == NULL) {
316 /* initialize tlock table */
317 for (k = 1; k < nTxLock - 1; k++)
318 TxLock[k].next = k + 1;
320 init_waitqueue_head(&TxAnchor.freelockwait);
321 init_waitqueue_head(&TxAnchor.lowlockwait);
323 TxAnchor.freelock = 1;
324 TxAnchor.tlocksInUse = 0;
325 INIT_LIST_HEAD(&TxAnchor.anon_list);
326 INIT_LIST_HEAD(&TxAnchor.anon_list2);
329 INIT_LIST_HEAD(&TxAnchor.unlock_queue);
331 stattx.maxlid = 1; /* statistics */
339 * FUNCTION: clean up when module is unloaded
352 * FUNCTION: start a transaction.
354 * PARAMETER: sb - superblock
355 * flag - force for nested tx;
357 * RETURN: tid - transaction id
359 * note: flag force allows to start tx for nested tx
360 * to prevent deadlock on logsync barrier;
362 tid_t txBegin(struct super_block *sb, int flag)
368 jfs_info("txBegin: flag = 0x%x", flag);
369 log = JFS_SBI(sb)->log;
373 INCREMENT(TxStat.txBegin);
376 if (!(flag & COMMIT_FORCE)) {
378 * synchronize with logsync barrier
380 if (test_bit(log_SYNCBARRIER, &log->flag) ||
381 test_bit(log_QUIESCE, &log->flag)) {
382 INCREMENT(TxStat.txBegin_barrier);
383 TXN_SLEEP(&log->syncwait);
389 * Don't begin transaction if we're getting starved for tlocks
390 * unless COMMIT_FORCE or COMMIT_INODE (which may ultimately
393 if (TxAnchor.tlocksInUse > TxLockVHWM) {
394 INCREMENT(TxStat.txBegin_lockslow);
395 TXN_SLEEP(&TxAnchor.lowlockwait);
401 * allocate transaction id/block
403 if ((t = TxAnchor.freetid) == 0) {
404 jfs_info("txBegin: waiting for free tid");
405 INCREMENT(TxStat.txBegin_freetid);
406 TXN_SLEEP(&TxAnchor.freewait);
410 tblk = tid_to_tblock(t);
412 if ((tblk->next == 0) && !(flag & COMMIT_FORCE)) {
413 /* Don't let a non-forced transaction take the last tblk */
414 jfs_info("txBegin: waiting for free tid");
415 INCREMENT(TxStat.txBegin_freetid);
416 TXN_SLEEP(&TxAnchor.freewait);
420 TxAnchor.freetid = tblk->next;
423 * initialize transaction
427 * We can't zero the whole thing or we screw up another thread being
428 * awakened after sleeping on tblk->waitor
430 * memset(tblk, 0, sizeof(struct tblock));
432 tblk->next = tblk->last = tblk->xflag = tblk->flag = tblk->lsn = 0;
436 tblk->logtid = log->logtid;
440 HIGHWATERMARK(stattx.maxtid, t); /* statistics */
441 INCREMENT(stattx.ntid); /* statistics */
445 jfs_info("txBegin: returning tid = %d", t);
451 * NAME: txBeginAnon()
453 * FUNCTION: start an anonymous transaction.
454 * Blocks if logsync or available tlocks are low to prevent
455 * anonymous tlocks from depleting supply.
457 * PARAMETER: sb - superblock
461 void txBeginAnon(struct super_block *sb)
465 log = JFS_SBI(sb)->log;
468 INCREMENT(TxStat.txBeginAnon);
472 * synchronize with logsync barrier
474 if (test_bit(log_SYNCBARRIER, &log->flag) ||
475 test_bit(log_QUIESCE, &log->flag)) {
476 INCREMENT(TxStat.txBeginAnon_barrier);
477 TXN_SLEEP(&log->syncwait);
482 * Don't begin transaction if we're getting starved for tlocks
484 if (TxAnchor.tlocksInUse > TxLockVHWM) {
485 INCREMENT(TxStat.txBeginAnon_lockslow);
486 TXN_SLEEP(&TxAnchor.lowlockwait);
495 * function: free specified transaction block.
497 * logsync barrier processing:
501 void txEnd(tid_t tid)
503 struct tblock *tblk = tid_to_tblock(tid);
506 jfs_info("txEnd: tid = %d", tid);
510 * wakeup transactions waiting on the page locked
511 * by the current transaction
513 TXN_WAKEUP(&tblk->waitor);
515 log = JFS_SBI(tblk->sb)->log;
518 * Lazy commit thread can't free this guy until we mark it UNLOCKED,
519 * otherwise, we would be left with a transaction that may have been
522 * Lazy commit thread will turn off tblkGC_LAZY before calling this
525 if (tblk->flag & tblkGC_LAZY) {
526 jfs_info("txEnd called w/lazy tid: %d, tblk = 0x%p", tid, tblk);
529 spin_lock_irq(&log->gclock); // LOGGC_LOCK
530 tblk->flag |= tblkGC_UNLOCKED;
531 spin_unlock_irq(&log->gclock); // LOGGC_UNLOCK
535 jfs_info("txEnd: tid: %d, tblk = 0x%p", tid, tblk);
537 assert(tblk->next == 0);
540 * insert tblock back on freelist
542 tblk->next = TxAnchor.freetid;
543 TxAnchor.freetid = tid;
546 * mark the tblock not active
548 if (--log->active == 0) {
549 clear_bit(log_FLUSH, &log->flag);
552 * synchronize with logsync barrier
554 if (test_bit(log_SYNCBARRIER, &log->flag)) {
557 /* write dirty metadata & forward log syncpt */
560 jfs_info("log barrier off: 0x%x", log->lsn);
562 /* enable new transactions start */
563 clear_bit(log_SYNCBARRIER, &log->flag);
565 /* wakeup all waitors for logsync barrier */
566 TXN_WAKEUP(&log->syncwait);
575 * wakeup all waitors for a free tblock
577 TXN_WAKEUP(&TxAnchor.freewait);
583 * function: acquire a transaction lock on the specified <mp>
587 * return: transaction lock id
591 struct tlock *txLock(tid_t tid, struct inode *ip, struct metapage * mp,
594 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
599 struct xtlock *xtlck;
600 struct linelock *linelock;
606 if (S_ISDIR(ip->i_mode) && (type & tlckXTREE) &&
607 !(mp->xflag & COMMIT_PAGE)) {
609 * Directory inode is special. It can have both an xtree tlock
610 * and a dtree tlock associated with it.
617 /* is page not locked by a transaction ? */
621 jfs_info("txLock: tid:%d ip:0x%p mp:0x%p lid:%d", tid, ip, mp, lid);
623 /* is page locked by the requester transaction ? */
624 tlck = lid_to_tlock(lid);
625 if ((xtid = tlck->tid) == tid) {
631 * is page locked by anonymous transaction/lock ?
633 * (page update without transaction (i.e., file write) is
634 * locked under anonymous transaction tid = 0:
635 * anonymous tlocks maintained on anonymous tlock list of
636 * the inode of the page and available to all anonymous
637 * transactions until txCommit() time at which point
638 * they are transferred to the transaction tlock list of
639 * the committing transaction of the inode)
644 tblk = tid_to_tblock(tid);
646 * The order of the tlocks in the transaction is important
647 * (during truncate, child xtree pages must be freed before
648 * parent's tlocks change the working map).
649 * Take tlock off anonymous list and add to tail of
652 * Note: We really need to get rid of the tid & lid and
653 * use list_head's. This code is getting UGLY!
655 if (jfs_ip->atlhead == lid) {
656 if (jfs_ip->atltail == lid) {
657 /* only anonymous txn.
658 * Remove from anon_list
661 list_del_init(&jfs_ip->anon_inode_list);
664 jfs_ip->atlhead = tlck->next;
667 for (last = jfs_ip->atlhead;
668 lid_to_tlock(last)->next != lid;
669 last = lid_to_tlock(last)->next) {
672 lid_to_tlock(last)->next = tlck->next;
673 if (jfs_ip->atltail == lid)
674 jfs_ip->atltail = last;
677 /* insert the tlock at tail of transaction tlock list */
680 lid_to_tlock(tblk->last)->next = lid;
696 tlck = lid_to_tlock(lid);
705 /* mark tlock for meta-data page */
706 if (mp->xflag & COMMIT_PAGE) {
708 tlck->flag = tlckPAGELOCK;
710 /* mark the page dirty and nohomeok */
711 metapage_nohomeok(mp);
713 jfs_info("locking mp = 0x%p, nohomeok = %d tid = %d tlck = 0x%p",
714 mp, mp->nohomeok, tid, tlck);
716 /* if anonymous transaction, and buffer is on the group
717 * commit synclist, mark inode to show this. This will
718 * prevent the buffer from being marked nohomeok for too
721 if ((tid == 0) && mp->lsn)
722 set_cflag(COMMIT_Synclist, ip);
724 /* mark tlock for in-memory inode */
726 tlck->flag = tlckINODELOCK;
728 if (S_ISDIR(ip->i_mode))
729 tlck->flag |= tlckDIRECTORY;
733 /* bind the tlock and the page */
742 * enqueue transaction lock to transaction/inode
744 /* insert the tlock at tail of transaction tlock list */
746 tblk = tid_to_tblock(tid);
748 lid_to_tlock(tblk->last)->next = lid;
754 /* anonymous transaction:
755 * insert the tlock at head of inode anonymous tlock list
758 tlck->next = jfs_ip->atlhead;
759 jfs_ip->atlhead = lid;
760 if (tlck->next == 0) {
761 /* This inode's first anonymous transaction */
762 jfs_ip->atltail = lid;
764 list_add_tail(&jfs_ip->anon_inode_list,
765 &TxAnchor.anon_list);
770 /* initialize type dependent area for linelock */
771 linelock = (struct linelock *) & tlck->lock;
773 linelock->flag = tlckLINELOCK;
774 linelock->maxcnt = TLOCKSHORT;
777 switch (type & tlckTYPE) {
779 linelock->l2linesize = L2DTSLOTSIZE;
783 linelock->l2linesize = L2XTSLOTSIZE;
785 xtlck = (struct xtlock *) linelock;
786 xtlck->header.offset = 0;
787 xtlck->header.length = 2;
789 if (type & tlckNEW) {
790 xtlck->lwm.offset = XTENTRYSTART;
792 if (mp->xflag & COMMIT_PAGE)
793 p = (xtpage_t *) mp->data;
795 p = &jfs_ip->i_xtroot;
797 le16_to_cpu(p->header.nextindex);
799 xtlck->lwm.length = 0; /* ! */
800 xtlck->twm.offset = 0;
801 xtlck->hwm.offset = 0;
807 linelock->l2linesize = L2INODESLOTSIZE;
811 linelock->l2linesize = L2DATASLOTSIZE;
815 jfs_err("UFO tlock:0x%p", tlck);
819 * update tlock vector
827 * page is being locked by another transaction:
830 /* Only locks on ipimap or ipaimap should reach here */
831 /* assert(jfs_ip->fileset == AGGREGATE_I); */
832 if (jfs_ip->fileset != AGGREGATE_I) {
833 printk(KERN_ERR "txLock: trying to lock locked page!");
834 print_hex_dump(KERN_ERR, "ip: ", DUMP_PREFIX_ADDRESS, 16, 4,
836 print_hex_dump(KERN_ERR, "mp: ", DUMP_PREFIX_ADDRESS, 16, 4,
838 print_hex_dump(KERN_ERR, "Locker's tblock: ",
839 DUMP_PREFIX_ADDRESS, 16, 4, tid_to_tblock(tid),
840 sizeof(struct tblock), 0);
841 print_hex_dump(KERN_ERR, "Tlock: ", DUMP_PREFIX_ADDRESS, 16, 4,
842 tlck, sizeof(*tlck), 0);
845 INCREMENT(stattx.waitlock); /* statistics */
847 release_metapage(mp);
849 xtid = tlck->tid; /* reacquire after dropping TXN_LOCK */
851 jfs_info("txLock: in waitLock, tid = %d, xtid = %d, lid = %d",
854 /* Recheck everything since dropping TXN_LOCK */
855 if (xtid && (tlck->mp == mp) && (mp->lid == lid))
856 TXN_SLEEP_DROP_LOCK(&tid_to_tblock(xtid)->waitor);
859 jfs_info("txLock: awakened tid = %d, lid = %d", tid, lid);
867 * FUNCTION: Release buffers associated with transaction locks, but don't
868 * mark homeok yet. The allows other transactions to modify
869 * buffers, but won't let them go to disk until commit record
870 * actually gets written.
875 * RETURN: Errors from subroutines.
877 static void txRelease(struct tblock * tblk)
885 for (lid = tblk->next; lid; lid = tlck->next) {
886 tlck = lid_to_tlock(lid);
887 if ((mp = tlck->mp) != NULL &&
888 (tlck->type & tlckBTROOT) == 0) {
889 assert(mp->xflag & COMMIT_PAGE);
895 * wakeup transactions waiting on a page locked
896 * by the current transaction
898 TXN_WAKEUP(&tblk->waitor);
906 * FUNCTION: Initiates pageout of pages modified by tid in journalled
907 * objects and frees their lockwords.
909 static void txUnlock(struct tblock * tblk)
912 struct linelock *linelock;
913 lid_t lid, next, llid, k;
919 jfs_info("txUnlock: tblk = 0x%p", tblk);
920 log = JFS_SBI(tblk->sb)->log;
923 * mark page under tlock homeok (its log has been written):
925 for (lid = tblk->next; lid; lid = next) {
926 tlck = lid_to_tlock(lid);
929 jfs_info("unlocking lid = %d, tlck = 0x%p", lid, tlck);
931 /* unbind page from tlock */
932 if ((mp = tlck->mp) != NULL &&
933 (tlck->type & tlckBTROOT) == 0) {
934 assert(mp->xflag & COMMIT_PAGE);
940 assert(mp->nohomeok > 0);
941 _metapage_homeok(mp);
943 /* inherit younger/larger clsn */
944 LOGSYNC_LOCK(log, flags);
946 logdiff(difft, tblk->clsn, log);
947 logdiff(diffp, mp->clsn, log);
949 mp->clsn = tblk->clsn;
951 mp->clsn = tblk->clsn;
952 LOGSYNC_UNLOCK(log, flags);
954 assert(!(tlck->flag & tlckFREEPAGE));
959 /* insert tlock, and linelock(s) of the tlock if any,
960 * at head of freelist
964 llid = ((struct linelock *) & tlck->lock)->next;
966 linelock = (struct linelock *) lid_to_tlock(llid);
975 tblk->next = tblk->last = 0;
978 * remove tblock from logsynclist
979 * (allocation map pages inherited lsn of tblk and
980 * has been inserted in logsync list at txUpdateMap())
983 LOGSYNC_LOCK(log, flags);
985 list_del(&tblk->synclist);
986 LOGSYNC_UNLOCK(log, flags);
993 * function: allocate a transaction lock for freed page/entry;
994 * for freed page, maplock is used as xtlock/dtlock type;
996 struct tlock *txMaplock(tid_t tid, struct inode *ip, int type)
998 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
1000 struct tblock *tblk;
1002 struct maplock *maplock;
1009 lid = txLockAlloc();
1010 tlck = lid_to_tlock(lid);
1017 /* bind the tlock and the object */
1018 tlck->flag = tlckINODELOCK;
1019 if (S_ISDIR(ip->i_mode))
1020 tlck->flag |= tlckDIRECTORY;
1027 * enqueue transaction lock to transaction/inode
1029 /* insert the tlock at tail of transaction tlock list */
1031 tblk = tid_to_tblock(tid);
1033 lid_to_tlock(tblk->last)->next = lid;
1039 /* anonymous transaction:
1040 * insert the tlock at head of inode anonymous tlock list
1043 tlck->next = jfs_ip->atlhead;
1044 jfs_ip->atlhead = lid;
1045 if (tlck->next == 0) {
1046 /* This inode's first anonymous transaction */
1047 jfs_ip->atltail = lid;
1048 list_add_tail(&jfs_ip->anon_inode_list,
1049 &TxAnchor.anon_list);
1055 /* initialize type dependent area for maplock */
1056 maplock = (struct maplock *) & tlck->lock;
1058 maplock->maxcnt = 0;
1067 * function: allocate a transaction lock for log vector list
1069 struct linelock *txLinelock(struct linelock * tlock)
1073 struct linelock *linelock;
1077 /* allocate a TxLock structure */
1078 lid = txLockAlloc();
1079 tlck = lid_to_tlock(lid);
1083 /* initialize linelock */
1084 linelock = (struct linelock *) tlck;
1086 linelock->flag = tlckLINELOCK;
1087 linelock->maxcnt = TLOCKLONG;
1088 linelock->index = 0;
1089 if (tlck->flag & tlckDIRECTORY)
1090 linelock->flag |= tlckDIRECTORY;
1092 /* append linelock after tlock */
1093 linelock->next = tlock->next;
1100 * transaction commit management
1101 * -----------------------------
1107 * FUNCTION: commit the changes to the objects specified in
1108 * clist. For journalled segments only the
1109 * changes of the caller are committed, ie by tid.
1110 * for non-journalled segments the data are flushed to
1111 * disk and then the change to the disk inode and indirect
1112 * blocks committed (so blocks newly allocated to the
1113 * segment will be made a part of the segment atomically).
1115 * all of the segments specified in clist must be in
1116 * one file system. no more than 6 segments are needed
1117 * to handle all unix svcs.
1119 * if the i_nlink field (i.e. disk inode link count)
1120 * is zero, and the type of inode is a regular file or
1121 * directory, or symbolic link , the inode is truncated
1122 * to zero length. the truncation is committed but the
1123 * VM resources are unaffected until it is closed (see
1131 * on entry the inode lock on each segment is assumed
1136 int txCommit(tid_t tid, /* transaction identifier */
1137 int nip, /* number of inodes to commit */
1138 struct inode **iplist, /* list of inode to commit */
1143 struct jfs_log *log;
1144 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) {
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)
1293 * I believe this code is no longer needed. Splitting I_LOCK
1294 * into two bits, I_NEW and I_SYNC should prevent this
1295 * deadlock as well. But since I don't have a JFS testload
1296 * to verify this, only a trivial s/I_LOCK/I_SYNC/ was done.
1299 if (tblk->u.ip->i_state & I_SYNC)
1300 tblk->xflag &= ~COMMIT_LAZY;
1303 ASSERT((!(tblk->xflag & COMMIT_DELETE)) ||
1304 ((tblk->u.ip->i_nlink == 0) &&
1305 !test_cflag(COMMIT_Nolink, tblk->u.ip)));
1308 * write COMMIT log record
1310 lrd->type = cpu_to_le16(LOG_COMMIT);
1312 lmLog(log, tblk, lrd, NULL);
1314 lmGroupCommit(log, tblk);
1317 * - transaction is now committed -
1321 * force pages in careful update
1322 * (imap addressing structure update)
1324 if (flag & COMMIT_FORCE)
1328 * update allocation map.
1330 * update inode allocation map and inode:
1331 * free pager lock on memory object of inode if any.
1332 * update block allocation map.
1334 * txUpdateMap() resets XAD_NEW in XAD.
1336 if (tblk->xflag & COMMIT_FORCE)
1340 * free transaction locks and pageout/free pages
1344 if ((tblk->flag & tblkGC_LAZY) == 0)
1349 * reset in-memory object state
1351 for (k = 0; k < cd.nip; k++) {
1353 jfs_ip = JFS_IP(ip);
1356 * reset in-memory inode state
1367 jfs_info("txCommit: tid = %d, returning %d", tid, rc);
1374 * FUNCTION: Writes AFTER log records for all lines modified
1375 * by tid for segments specified by inodes in comdata.
1376 * Code assumes only WRITELOCKS are recorded in lockwords.
1382 static int txLog(struct jfs_log * log, struct tblock * tblk, struct commit * cd)
1388 struct lrd *lrd = &cd->lrd;
1391 * write log record(s) for each tlock of transaction,
1393 for (lid = tblk->next; lid; lid = tlck->next) {
1394 tlck = lid_to_tlock(lid);
1396 tlck->flag |= tlckLOG;
1398 /* initialize lrd common */
1400 lrd->aggregate = cpu_to_le32(JFS_SBI(ip->i_sb)->aggregate);
1401 lrd->log.redopage.fileset = cpu_to_le32(JFS_IP(ip)->fileset);
1402 lrd->log.redopage.inode = cpu_to_le32(ip->i_ino);
1404 /* write log record of page from the tlock */
1405 switch (tlck->type & tlckTYPE) {
1407 xtLog(log, tblk, lrd, tlck);
1411 dtLog(log, tblk, lrd, tlck);
1415 diLog(log, tblk, lrd, tlck, cd);
1419 mapLog(log, tblk, lrd, tlck);
1423 dataLog(log, tblk, lrd, tlck);
1427 jfs_err("UFO tlock:0x%p", tlck);
1437 * function: log inode tlock and format maplock to update bmap;
1439 static int diLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1440 struct tlock * tlck, struct commit * cd)
1443 struct metapage *mp;
1445 struct pxd_lock *pxdlock;
1449 /* initialize as REDOPAGE record format */
1450 lrd->log.redopage.type = cpu_to_le16(LOG_INODE);
1451 lrd->log.redopage.l2linesize = cpu_to_le16(L2INODESLOTSIZE);
1453 pxd = &lrd->log.redopage.pxd;
1458 if (tlck->type & tlckENTRY) {
1459 /* log after-image for logredo(): */
1460 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1461 PXDaddress(pxd, mp->index);
1463 mp->logical_size >> tblk->sb->s_blocksize_bits);
1464 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1466 /* mark page as homeward bound */
1467 tlck->flag |= tlckWRITEPAGE;
1468 } else if (tlck->type & tlckFREE) {
1472 * (pages of the freed inode extent have been invalidated and
1473 * a maplock for free of the extent has been formatted at
1476 * the tlock had been acquired on the inode allocation map page
1477 * (iag) that specifies the freed extent, even though the map
1478 * page is not itself logged, to prevent pageout of the map
1479 * page before the log;
1482 /* log LOG_NOREDOINOEXT of the freed inode extent for
1483 * logredo() to start NoRedoPage filters, and to update
1484 * imap and bmap for free of the extent;
1486 lrd->type = cpu_to_le16(LOG_NOREDOINOEXT);
1488 * For the LOG_NOREDOINOEXT record, we need
1489 * to pass the IAG number and inode extent
1490 * index (within that IAG) from which the
1491 * the extent being released. These have been
1492 * passed to us in the iplist[1] and iplist[2].
1494 lrd->log.noredoinoext.iagnum =
1495 cpu_to_le32((u32) (size_t) cd->iplist[1]);
1496 lrd->log.noredoinoext.inoext_idx =
1497 cpu_to_le32((u32) (size_t) cd->iplist[2]);
1499 pxdlock = (struct pxd_lock *) & tlck->lock;
1500 *pxd = pxdlock->pxd;
1501 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1504 tlck->flag |= tlckUPDATEMAP;
1506 /* mark page as homeward bound */
1507 tlck->flag |= tlckWRITEPAGE;
1509 jfs_err("diLog: UFO type tlck:0x%p", tlck);
1512 * alloc/free external EA extent
1514 * a maplock for txUpdateMap() to update bPWMAP for alloc/free
1515 * of the extent has been formatted at txLock() time;
1518 assert(tlck->type & tlckEA);
1520 /* log LOG_UPDATEMAP for logredo() to update bmap for
1521 * alloc of new (and free of old) external EA extent;
1523 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1524 pxdlock = (struct pxd_lock *) & tlck->lock;
1525 nlock = pxdlock->index;
1526 for (i = 0; i < nlock; i++, pxdlock++) {
1527 if (pxdlock->flag & mlckALLOCPXD)
1528 lrd->log.updatemap.type =
1529 cpu_to_le16(LOG_ALLOCPXD);
1531 lrd->log.updatemap.type =
1532 cpu_to_le16(LOG_FREEPXD);
1533 lrd->log.updatemap.nxd = cpu_to_le16(1);
1534 lrd->log.updatemap.pxd = pxdlock->pxd;
1536 cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1540 tlck->flag |= tlckUPDATEMAP;
1542 #endif /* _JFS_WIP */
1550 * function: log data tlock
1552 static int dataLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1553 struct tlock * tlck)
1555 struct metapage *mp;
1560 /* initialize as REDOPAGE record format */
1561 lrd->log.redopage.type = cpu_to_le16(LOG_DATA);
1562 lrd->log.redopage.l2linesize = cpu_to_le16(L2DATASLOTSIZE);
1564 pxd = &lrd->log.redopage.pxd;
1566 /* log after-image for logredo(): */
1567 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1569 if (jfs_dirtable_inline(tlck->ip)) {
1571 * The table has been truncated, we've must have deleted
1572 * the last entry, so don't bother logging this
1576 metapage_homeok(mp);
1577 discard_metapage(mp);
1582 PXDaddress(pxd, mp->index);
1583 PXDlength(pxd, mp->logical_size >> tblk->sb->s_blocksize_bits);
1585 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1587 /* mark page as homeward bound */
1588 tlck->flag |= tlckWRITEPAGE;
1596 * function: log dtree tlock and format maplock to update bmap;
1598 static void dtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1599 struct tlock * tlck)
1601 struct metapage *mp;
1602 struct pxd_lock *pxdlock;
1607 /* initialize as REDOPAGE/NOREDOPAGE record format */
1608 lrd->log.redopage.type = cpu_to_le16(LOG_DTREE);
1609 lrd->log.redopage.l2linesize = cpu_to_le16(L2DTSLOTSIZE);
1611 pxd = &lrd->log.redopage.pxd;
1613 if (tlck->type & tlckBTROOT)
1614 lrd->log.redopage.type |= cpu_to_le16(LOG_BTROOT);
1617 * page extension via relocation: entry insertion;
1618 * page extension in-place: entry insertion;
1619 * new right page from page split, reinitialized in-line
1620 * root from root page split: entry insertion;
1622 if (tlck->type & (tlckNEW | tlckEXTEND)) {
1623 /* log after-image of the new page for logredo():
1624 * mark log (LOG_NEW) for logredo() to initialize
1625 * freelist and update bmap for alloc of the new page;
1627 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1628 if (tlck->type & tlckEXTEND)
1629 lrd->log.redopage.type |= cpu_to_le16(LOG_EXTEND);
1631 lrd->log.redopage.type |= cpu_to_le16(LOG_NEW);
1632 PXDaddress(pxd, mp->index);
1634 mp->logical_size >> tblk->sb->s_blocksize_bits);
1635 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1637 /* format a maplock for txUpdateMap() to update bPMAP for
1638 * alloc of the new page;
1640 if (tlck->type & tlckBTROOT)
1642 tlck->flag |= tlckUPDATEMAP;
1643 pxdlock = (struct pxd_lock *) & tlck->lock;
1644 pxdlock->flag = mlckALLOCPXD;
1645 pxdlock->pxd = *pxd;
1649 /* mark page as homeward bound */
1650 tlck->flag |= tlckWRITEPAGE;
1655 * entry insertion/deletion,
1656 * sibling page link update (old right page before split);
1658 if (tlck->type & (tlckENTRY | tlckRELINK)) {
1659 /* log after-image for logredo(): */
1660 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1661 PXDaddress(pxd, mp->index);
1663 mp->logical_size >> tblk->sb->s_blocksize_bits);
1664 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1666 /* mark page as homeward bound */
1667 tlck->flag |= tlckWRITEPAGE;
1672 * page deletion: page has been invalidated
1673 * page relocation: source extent
1675 * a maplock for free of the page has been formatted
1676 * at txLock() time);
1678 if (tlck->type & (tlckFREE | tlckRELOCATE)) {
1679 /* log LOG_NOREDOPAGE of the deleted page for logredo()
1680 * to start NoRedoPage filter and to update bmap for free
1681 * of the deletd page
1683 lrd->type = cpu_to_le16(LOG_NOREDOPAGE);
1684 pxdlock = (struct pxd_lock *) & tlck->lock;
1685 *pxd = pxdlock->pxd;
1686 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1688 /* a maplock for txUpdateMap() for free of the page
1689 * has been formatted at txLock() time;
1691 tlck->flag |= tlckUPDATEMAP;
1699 * function: log xtree tlock and format maplock to update bmap;
1701 static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
1702 struct tlock * tlck)
1705 struct metapage *mp;
1707 struct xtlock *xtlck;
1708 struct maplock *maplock;
1709 struct xdlistlock *xadlock;
1710 struct pxd_lock *pxdlock;
1717 /* initialize as REDOPAGE/NOREDOPAGE record format */
1718 lrd->log.redopage.type = cpu_to_le16(LOG_XTREE);
1719 lrd->log.redopage.l2linesize = cpu_to_le16(L2XTSLOTSIZE);
1721 page_pxd = &lrd->log.redopage.pxd;
1723 if (tlck->type & tlckBTROOT) {
1724 lrd->log.redopage.type |= cpu_to_le16(LOG_BTROOT);
1725 p = &JFS_IP(ip)->i_xtroot;
1726 if (S_ISDIR(ip->i_mode))
1727 lrd->log.redopage.type |=
1728 cpu_to_le16(LOG_DIR_XTREE);
1730 p = (xtpage_t *) mp->data;
1731 next = le16_to_cpu(p->header.nextindex);
1733 xtlck = (struct xtlock *) & tlck->lock;
1735 maplock = (struct maplock *) & tlck->lock;
1736 xadlock = (struct xdlistlock *) maplock;
1739 * entry insertion/extension;
1740 * sibling page link update (old right page before split);
1742 if (tlck->type & (tlckNEW | tlckGROW | tlckRELINK)) {
1743 /* log after-image for logredo():
1744 * logredo() will update bmap for alloc of new/extended
1745 * extents (XAD_NEW|XAD_EXTEND) of XAD[lwm:next) from
1746 * after-image of XADlist;
1747 * logredo() resets (XAD_NEW|XAD_EXTEND) flag when
1748 * applying the after-image to the meta-data page.
1750 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1751 PXDaddress(page_pxd, mp->index);
1753 mp->logical_size >> tblk->sb->s_blocksize_bits);
1754 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1756 /* format a maplock for txUpdateMap() to update bPMAP
1757 * for alloc of new/extended extents of XAD[lwm:next)
1758 * from the page itself;
1759 * txUpdateMap() resets (XAD_NEW|XAD_EXTEND) flag.
1761 lwm = xtlck->lwm.offset;
1763 lwm = XTPAGEMAXSLOT;
1768 jfs_err("xtLog: lwm > next\n");
1771 tlck->flag |= tlckUPDATEMAP;
1772 xadlock->flag = mlckALLOCXADLIST;
1773 xadlock->count = next - lwm;
1774 if ((xadlock->count <= 4) && (tblk->xflag & COMMIT_LAZY)) {
1778 * Lazy commit may allow xtree to be modified before
1779 * txUpdateMap runs. Copy xad into linelock to
1780 * preserve correct data.
1782 * We can fit twice as may pxd's as xads in the lock
1784 xadlock->flag = mlckALLOCPXDLIST;
1785 pxd = xadlock->xdlist = &xtlck->pxdlock;
1786 for (i = 0; i < xadlock->count; i++) {
1787 PXDaddress(pxd, addressXAD(&p->xad[lwm + i]));
1788 PXDlength(pxd, lengthXAD(&p->xad[lwm + i]));
1789 p->xad[lwm + i].flag &=
1790 ~(XAD_NEW | XAD_EXTENDED);
1795 * xdlist will point to into inode's xtree, ensure
1796 * that transaction is not committed lazily.
1798 xadlock->flag = mlckALLOCXADLIST;
1799 xadlock->xdlist = &p->xad[lwm];
1800 tblk->xflag &= ~COMMIT_LAZY;
1802 jfs_info("xtLog: alloc ip:0x%p mp:0x%p tlck:0x%p lwm:%d "
1803 "count:%d", tlck->ip, mp, tlck, lwm, xadlock->count);
1808 /* mark page as homeward bound */
1809 tlck->flag |= tlckWRITEPAGE;
1815 * page deletion: file deletion/truncation (ref. xtTruncate())
1817 * (page will be invalidated after log is written and bmap
1818 * is updated from the page);
1820 if (tlck->type & tlckFREE) {
1821 /* LOG_NOREDOPAGE log for NoRedoPage filter:
1822 * if page free from file delete, NoRedoFile filter from
1823 * inode image of zero link count will subsume NoRedoPage
1824 * filters for each page;
1825 * if page free from file truncattion, write NoRedoPage
1828 * upadte of block allocation map for the page itself:
1829 * if page free from deletion and truncation, LOG_UPDATEMAP
1830 * log for the page itself is generated from processing
1831 * its parent page xad entries;
1833 /* if page free from file truncation, log LOG_NOREDOPAGE
1834 * of the deleted page for logredo() to start NoRedoPage
1835 * filter for the page;
1837 if (tblk->xflag & COMMIT_TRUNCATE) {
1838 /* write NOREDOPAGE for the page */
1839 lrd->type = cpu_to_le16(LOG_NOREDOPAGE);
1840 PXDaddress(page_pxd, mp->index);
1842 mp->logical_size >> tblk->sb->
1845 cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1847 if (tlck->type & tlckBTROOT) {
1848 /* Empty xtree must be logged */
1849 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1851 cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1855 /* init LOG_UPDATEMAP of the freed extents
1856 * XAD[XTENTRYSTART:hwm) from the deleted page itself
1857 * for logredo() to update bmap;
1859 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1860 lrd->log.updatemap.type = cpu_to_le16(LOG_FREEXADLIST);
1861 xtlck = (struct xtlock *) & tlck->lock;
1862 hwm = xtlck->hwm.offset;
1863 lrd->log.updatemap.nxd =
1864 cpu_to_le16(hwm - XTENTRYSTART + 1);
1865 /* reformat linelock for lmLog() */
1866 xtlck->header.offset = XTENTRYSTART;
1867 xtlck->header.length = hwm - XTENTRYSTART + 1;
1869 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1871 /* format a maplock for txUpdateMap() to update bmap
1872 * to free extents of XAD[XTENTRYSTART:hwm) from the
1873 * deleted page itself;
1875 tlck->flag |= tlckUPDATEMAP;
1876 xadlock->count = hwm - XTENTRYSTART + 1;
1877 if ((xadlock->count <= 4) && (tblk->xflag & COMMIT_LAZY)) {
1881 * Lazy commit may allow xtree to be modified before
1882 * txUpdateMap runs. Copy xad into linelock to
1883 * preserve correct data.
1885 * We can fit twice as may pxd's as xads in the lock
1887 xadlock->flag = mlckFREEPXDLIST;
1888 pxd = xadlock->xdlist = &xtlck->pxdlock;
1889 for (i = 0; i < xadlock->count; i++) {
1891 addressXAD(&p->xad[XTENTRYSTART + i]));
1893 lengthXAD(&p->xad[XTENTRYSTART + i]));
1898 * xdlist will point to into inode's xtree, ensure
1899 * that transaction is not committed lazily.
1901 xadlock->flag = mlckFREEXADLIST;
1902 xadlock->xdlist = &p->xad[XTENTRYSTART];
1903 tblk->xflag &= ~COMMIT_LAZY;
1905 jfs_info("xtLog: free ip:0x%p mp:0x%p count:%d lwm:2",
1906 tlck->ip, mp, xadlock->count);
1910 /* mark page as invalid */
1911 if (((tblk->xflag & COMMIT_PWMAP) || S_ISDIR(ip->i_mode))
1912 && !(tlck->type & tlckBTROOT))
1913 tlck->flag |= tlckFREEPAGE;
1915 else (tblk->xflag & COMMIT_PMAP)
1922 * page/entry truncation: file truncation (ref. xtTruncate())
1924 * |----------+------+------+---------------|
1926 * | | hwm - hwm before truncation
1927 * | next - truncation point
1928 * lwm - lwm before truncation
1931 if (tlck->type & tlckTRUNCATE) {
1932 /* This odd declaration suppresses a bogus gcc warning */
1933 pxd_t pxd = pxd; /* truncated extent of xad */
1937 * For truncation the entire linelock may be used, so it would
1938 * be difficult to store xad list in linelock itself.
1939 * Therefore, we'll just force transaction to be committed
1940 * synchronously, so that xtree pages won't be changed before
1943 tblk->xflag &= ~COMMIT_LAZY;
1944 lwm = xtlck->lwm.offset;
1946 lwm = XTPAGEMAXSLOT;
1947 hwm = xtlck->hwm.offset;
1948 twm = xtlck->twm.offset;
1953 /* log after-image for logredo():
1955 * logredo() will update bmap for alloc of new/extended
1956 * extents (XAD_NEW|XAD_EXTEND) of XAD[lwm:next) from
1957 * after-image of XADlist;
1958 * logredo() resets (XAD_NEW|XAD_EXTEND) flag when
1959 * applying the after-image to the meta-data page.
1961 lrd->type = cpu_to_le16(LOG_REDOPAGE);
1962 PXDaddress(page_pxd, mp->index);
1964 mp->logical_size >> tblk->sb->s_blocksize_bits);
1965 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck));
1968 * truncate entry XAD[twm == next - 1]:
1970 if (twm == next - 1) {
1971 /* init LOG_UPDATEMAP for logredo() to update bmap for
1972 * free of truncated delta extent of the truncated
1973 * entry XAD[next - 1]:
1974 * (xtlck->pxdlock = truncated delta extent);
1976 pxdlock = (struct pxd_lock *) & xtlck->pxdlock;
1977 /* assert(pxdlock->type & tlckTRUNCATE); */
1978 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1979 lrd->log.updatemap.type = cpu_to_le16(LOG_FREEPXD);
1980 lrd->log.updatemap.nxd = cpu_to_le16(1);
1981 lrd->log.updatemap.pxd = pxdlock->pxd;
1982 pxd = pxdlock->pxd; /* save to format maplock */
1984 cpu_to_le32(lmLog(log, tblk, lrd, NULL));
1988 * free entries XAD[next:hwm]:
1991 /* init LOG_UPDATEMAP of the freed extents
1992 * XAD[next:hwm] from the deleted page itself
1993 * for logredo() to update bmap;
1995 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
1996 lrd->log.updatemap.type =
1997 cpu_to_le16(LOG_FREEXADLIST);
1998 xtlck = (struct xtlock *) & tlck->lock;
1999 hwm = xtlck->hwm.offset;
2000 lrd->log.updatemap.nxd =
2001 cpu_to_le16(hwm - next + 1);
2002 /* reformat linelock for lmLog() */
2003 xtlck->header.offset = next;
2004 xtlck->header.length = hwm - next + 1;
2007 cpu_to_le32(lmLog(log, tblk, lrd, tlck));
2011 * format maplock(s) for txUpdateMap() to update bmap
2016 * allocate entries XAD[lwm:next):
2019 /* format a maplock for txUpdateMap() to update bPMAP
2020 * for alloc of new/extended extents of XAD[lwm:next)
2021 * from the page itself;
2022 * txUpdateMap() resets (XAD_NEW|XAD_EXTEND) flag.
2024 tlck->flag |= tlckUPDATEMAP;
2025 xadlock->flag = mlckALLOCXADLIST;
2026 xadlock->count = next - lwm;
2027 xadlock->xdlist = &p->xad[lwm];
2029 jfs_info("xtLog: alloc ip:0x%p mp:0x%p count:%d "
2031 tlck->ip, mp, xadlock->count, lwm, next);
2037 * truncate entry XAD[twm == next - 1]:
2039 if (twm == next - 1) {
2040 /* format a maplock for txUpdateMap() to update bmap
2041 * to free truncated delta extent of the truncated
2042 * entry XAD[next - 1];
2043 * (xtlck->pxdlock = truncated delta extent);
2045 tlck->flag |= tlckUPDATEMAP;
2046 pxdlock = (struct pxd_lock *) xadlock;
2047 pxdlock->flag = mlckFREEPXD;
2051 jfs_info("xtLog: truncate ip:0x%p mp:0x%p count:%d "
2052 "hwm:%d", ip, mp, pxdlock->count, hwm);
2058 * free entries XAD[next:hwm]:
2061 /* format a maplock for txUpdateMap() to update bmap
2062 * to free extents of XAD[next:hwm] from thedeleted
2065 tlck->flag |= tlckUPDATEMAP;
2066 xadlock->flag = mlckFREEXADLIST;
2067 xadlock->count = hwm - next + 1;
2068 xadlock->xdlist = &p->xad[next];
2070 jfs_info("xtLog: free ip:0x%p mp:0x%p count:%d "
2072 tlck->ip, mp, xadlock->count, next, hwm);
2076 /* mark page as homeward bound */
2077 tlck->flag |= tlckWRITEPAGE;
2085 * function: log from maplock of freed data extents;
2087 static void mapLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
2088 struct tlock * tlck)
2090 struct pxd_lock *pxdlock;
2095 * page relocation: free the source page extent
2097 * a maplock for txUpdateMap() for free of the page
2098 * has been formatted at txLock() time saving the src
2099 * relocated page address;
2101 if (tlck->type & tlckRELOCATE) {
2102 /* log LOG_NOREDOPAGE of the old relocated page
2103 * for logredo() to start NoRedoPage filter;
2105 lrd->type = cpu_to_le16(LOG_NOREDOPAGE);
2106 pxdlock = (struct pxd_lock *) & tlck->lock;
2107 pxd = &lrd->log.redopage.pxd;
2108 *pxd = pxdlock->pxd;
2109 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
2111 /* (N.B. currently, logredo() does NOT update bmap
2112 * for free of the page itself for (LOG_XTREE|LOG_NOREDOPAGE);
2113 * if page free from relocation, LOG_UPDATEMAP log is
2114 * specifically generated now for logredo()
2115 * to update bmap for free of src relocated page;
2116 * (new flag LOG_RELOCATE may be introduced which will
2117 * inform logredo() to start NORedoPage filter and also
2118 * update block allocation map at the same time, thus
2119 * avoiding an extra log write);
2121 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
2122 lrd->log.updatemap.type = cpu_to_le16(LOG_FREEPXD);
2123 lrd->log.updatemap.nxd = cpu_to_le16(1);
2124 lrd->log.updatemap.pxd = pxdlock->pxd;
2125 lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));
2127 /* a maplock for txUpdateMap() for free of the page
2128 * has been formatted at txLock() time;
2130 tlck->flag |= tlckUPDATEMAP;
2135 * Otherwise it's not a relocate request
2139 /* log LOG_UPDATEMAP for logredo() to update bmap for
2140 * free of truncated/relocated delta extent of the data;
2141 * e.g.: external EA extent, relocated/truncated extent
2142 * from xtTailgate();
2144 lrd->type = cpu_to_le16(LOG_UPDATEMAP);
2145 pxdlock = (struct pxd_lock *) & tlck->lock;
2146 nlock = pxdlock->index;
2147 for (i = 0; i < nlock; i++, pxdlock++) {
2148 if (pxdlock->flag & mlckALLOCPXD)
2149 lrd->log.updatemap.type =
2150 cpu_to_le16(LOG_ALLOCPXD);
2152 lrd->log.updatemap.type =
2153 cpu_to_le16(LOG_FREEPXD);
2154 lrd->log.updatemap.nxd = cpu_to_le16(1);
2155 lrd->log.updatemap.pxd = pxdlock->pxd;
2157 cpu_to_le32(lmLog(log, tblk, lrd, NULL));
2158 jfs_info("mapLog: xaddr:0x%lx xlen:0x%x",
2159 (ulong) addressPXD(&pxdlock->pxd),
2160 lengthPXD(&pxdlock->pxd));
2164 tlck->flag |= tlckUPDATEMAP;
2171 * function: acquire maplock for EA/ACL extents or
2172 * set COMMIT_INLINE flag;
2174 void txEA(tid_t tid, struct inode *ip, dxd_t * oldea, dxd_t * newea)
2176 struct tlock *tlck = NULL;
2177 struct pxd_lock *maplock = NULL, *pxdlock = NULL;
2180 * format maplock for alloc of new EA extent
2183 /* Since the newea could be a completely zeroed entry we need to
2184 * check for the two flags which indicate we should actually
2185 * commit new EA data
2187 if (newea->flag & DXD_EXTENT) {
2188 tlck = txMaplock(tid, ip, tlckMAP);
2189 maplock = (struct pxd_lock *) & tlck->lock;
2190 pxdlock = (struct pxd_lock *) maplock;
2191 pxdlock->flag = mlckALLOCPXD;
2192 PXDaddress(&pxdlock->pxd, addressDXD(newea));
2193 PXDlength(&pxdlock->pxd, lengthDXD(newea));
2196 } else if (newea->flag & DXD_INLINE) {
2199 set_cflag(COMMIT_Inlineea, ip);
2204 * format maplock for free of old EA extent
2206 if (!test_cflag(COMMIT_Nolink, ip) && oldea->flag & DXD_EXTENT) {
2208 tlck = txMaplock(tid, ip, tlckMAP);
2209 maplock = (struct pxd_lock *) & tlck->lock;
2210 pxdlock = (struct pxd_lock *) maplock;
2213 pxdlock->flag = mlckFREEPXD;
2214 PXDaddress(&pxdlock->pxd, addressDXD(oldea));
2215 PXDlength(&pxdlock->pxd, lengthDXD(oldea));
2223 * function: synchronously write pages locked by transaction
2224 * after txLog() but before txUpdateMap();
2226 static void txForce(struct tblock * tblk)
2230 struct metapage *mp;
2233 * reverse the order of transaction tlocks in
2234 * careful update order of address index pages
2235 * (right to left, bottom up)
2237 tlck = lid_to_tlock(tblk->next);
2241 tlck = lid_to_tlock(lid);
2243 tlck->next = tblk->next;
2249 * synchronously write the page, and
2250 * hold the page for txUpdateMap();
2252 for (lid = tblk->next; lid; lid = next) {
2253 tlck = lid_to_tlock(lid);
2256 if ((mp = tlck->mp) != NULL &&
2257 (tlck->type & tlckBTROOT) == 0) {
2258 assert(mp->xflag & COMMIT_PAGE);
2260 if (tlck->flag & tlckWRITEPAGE) {
2261 tlck->flag &= ~tlckWRITEPAGE;
2263 /* do not release page to freelist */
2267 * The "right" thing to do here is to
2268 * synchronously write the metadata.
2269 * With the current implementation this
2270 * is hard since write_metapage requires
2271 * us to kunmap & remap the page. If we
2272 * have tlocks pointing into the metadata
2273 * pages, we don't want to do this. I think
2274 * we can get by with synchronously writing
2275 * the pages when they are released.
2277 assert(mp->nohomeok);
2278 set_bit(META_dirty, &mp->flag);
2279 set_bit(META_sync, &mp->flag);
2289 * function: update persistent allocation map (and working map
2294 static void txUpdateMap(struct tblock * tblk)
2297 struct inode *ipimap;
2300 struct maplock *maplock;
2301 struct pxd_lock pxdlock;
2304 struct metapage *mp = NULL;
2306 ipimap = JFS_SBI(tblk->sb)->ipimap;
2308 maptype = (tblk->xflag & COMMIT_PMAP) ? COMMIT_PMAP : COMMIT_PWMAP;
2312 * update block allocation map
2314 * update allocation state in pmap (and wmap) and
2315 * update lsn of the pmap page;
2318 * scan each tlock/page of transaction for block allocation/free:
2320 * for each tlock/page of transaction, update map.
2321 * ? are there tlock for pmap and pwmap at the same time ?
2323 for (lid = tblk->next; lid; lid = tlck->next) {
2324 tlck = lid_to_tlock(lid);
2326 if ((tlck->flag & tlckUPDATEMAP) == 0)
2329 if (tlck->flag & tlckFREEPAGE) {
2331 * Another thread may attempt to reuse freed space
2332 * immediately, so we want to get rid of the metapage
2333 * before anyone else has a chance to get it.
2334 * Lock metapage, update maps, then invalidate
2338 ASSERT(mp->xflag & COMMIT_PAGE);
2344 * . in-line PXD list:
2345 * . out-of-line XAD list:
2347 maplock = (struct maplock *) & tlck->lock;
2348 nlock = maplock->index;
2350 for (k = 0; k < nlock; k++, maplock++) {
2352 * allocate blocks in persistent map:
2354 * blocks have been allocated from wmap at alloc time;
2356 if (maplock->flag & mlckALLOC) {
2357 txAllocPMap(ipimap, maplock, tblk);
2360 * free blocks in persistent and working map:
2361 * blocks will be freed in pmap and then in wmap;
2363 * ? tblock specifies the PMAP/PWMAP based upon
2366 * free blocks in persistent map:
2367 * blocks will be freed from wmap at last reference
2368 * release of the object for regular files;
2370 * Alway free blocks from both persistent & working
2371 * maps for directories
2373 else { /* (maplock->flag & mlckFREE) */
2375 if (tlck->flag & tlckDIRECTORY)
2376 txFreeMap(ipimap, maplock,
2377 tblk, COMMIT_PWMAP);
2379 txFreeMap(ipimap, maplock,
2383 if (tlck->flag & tlckFREEPAGE) {
2384 if (!(tblk->flag & tblkGC_LAZY)) {
2385 /* This is equivalent to txRelease */
2386 ASSERT(mp->lid == lid);
2389 assert(mp->nohomeok == 1);
2390 metapage_homeok(mp);
2391 discard_metapage(mp);
2396 * update inode allocation map
2398 * update allocation state in pmap and
2399 * update lsn of the pmap page;
2400 * update in-memory inode flag/state
2402 * unlock mapper/write lock
2404 if (tblk->xflag & COMMIT_CREATE) {
2405 diUpdatePMap(ipimap, tblk->ino, false, tblk);
2406 /* update persistent block allocation map
2407 * for the allocation of inode extent;
2409 pxdlock.flag = mlckALLOCPXD;
2410 pxdlock.pxd = tblk->u.ixpxd;
2412 txAllocPMap(ipimap, (struct maplock *) & pxdlock, tblk);
2413 } else if (tblk->xflag & COMMIT_DELETE) {
2415 diUpdatePMap(ipimap, ip->i_ino, true, tblk);
2423 * function: allocate from persistent map;
2432 * allocate from persistent map;
2433 * free from persistent map;
2434 * (e.g., tmp file - free from working map at releae
2435 * of last reference);
2436 * free from persistent and working map;
2438 * lsn - log sequence number;
2440 static void txAllocPMap(struct inode *ip, struct maplock * maplock,
2441 struct tblock * tblk)
2443 struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
2444 struct xdlistlock *xadlistlock;
2448 struct pxd_lock *pxdlock;
2449 struct xdlistlock *pxdlistlock;
2454 * allocate from persistent map;
2456 if (maplock->flag & mlckALLOCXADLIST) {
2457 xadlistlock = (struct xdlistlock *) maplock;
2458 xad = xadlistlock->xdlist;
2459 for (n = 0; n < xadlistlock->count; n++, xad++) {
2460 if (xad->flag & (XAD_NEW | XAD_EXTENDED)) {
2461 xaddr = addressXAD(xad);
2462 xlen = lengthXAD(xad);
2463 dbUpdatePMap(ipbmap, false, xaddr,
2465 xad->flag &= ~(XAD_NEW | XAD_EXTENDED);
2466 jfs_info("allocPMap: xaddr:0x%lx xlen:%d",
2467 (ulong) xaddr, xlen);
2470 } else if (maplock->flag & mlckALLOCPXD) {
2471 pxdlock = (struct pxd_lock *) maplock;
2472 xaddr = addressPXD(&pxdlock->pxd);
2473 xlen = lengthPXD(&pxdlock->pxd);
2474 dbUpdatePMap(ipbmap, false, xaddr, (s64) xlen, tblk);
2475 jfs_info("allocPMap: xaddr:0x%lx xlen:%d", (ulong) xaddr, xlen);
2476 } else { /* (maplock->flag & mlckALLOCPXDLIST) */
2478 pxdlistlock = (struct xdlistlock *) maplock;
2479 pxd = pxdlistlock->xdlist;
2480 for (n = 0; n < pxdlistlock->count; n++, pxd++) {
2481 xaddr = addressPXD(pxd);
2482 xlen = lengthPXD(pxd);
2483 dbUpdatePMap(ipbmap, false, xaddr, (s64) xlen,
2485 jfs_info("allocPMap: xaddr:0x%lx xlen:%d",
2486 (ulong) xaddr, xlen);
2494 * function: free from persistent and/or working map;
2496 * todo: optimization
2498 void txFreeMap(struct inode *ip,
2499 struct maplock * maplock, struct tblock * tblk, int maptype)
2501 struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
2502 struct xdlistlock *xadlistlock;
2506 struct pxd_lock *pxdlock;
2507 struct xdlistlock *pxdlistlock;
2511 jfs_info("txFreeMap: tblk:0x%p maplock:0x%p maptype:0x%x",
2512 tblk, maplock, maptype);
2515 * free from persistent map;
2517 if (maptype == COMMIT_PMAP || maptype == COMMIT_PWMAP) {
2518 if (maplock->flag & mlckFREEXADLIST) {
2519 xadlistlock = (struct xdlistlock *) maplock;
2520 xad = xadlistlock->xdlist;
2521 for (n = 0; n < xadlistlock->count; n++, xad++) {
2522 if (!(xad->flag & XAD_NEW)) {
2523 xaddr = addressXAD(xad);
2524 xlen = lengthXAD(xad);
2525 dbUpdatePMap(ipbmap, true, xaddr,
2527 jfs_info("freePMap: xaddr:0x%lx "
2529 (ulong) xaddr, xlen);
2532 } else if (maplock->flag & mlckFREEPXD) {
2533 pxdlock = (struct pxd_lock *) maplock;
2534 xaddr = addressPXD(&pxdlock->pxd);
2535 xlen = lengthPXD(&pxdlock->pxd);
2536 dbUpdatePMap(ipbmap, true, xaddr, (s64) xlen,
2538 jfs_info("freePMap: xaddr:0x%lx xlen:%d",
2539 (ulong) xaddr, xlen);
2540 } else { /* (maplock->flag & mlckALLOCPXDLIST) */
2542 pxdlistlock = (struct xdlistlock *) maplock;
2543 pxd = pxdlistlock->xdlist;
2544 for (n = 0; n < pxdlistlock->count; n++, pxd++) {
2545 xaddr = addressPXD(pxd);
2546 xlen = lengthPXD(pxd);
2547 dbUpdatePMap(ipbmap, true, xaddr,
2549 jfs_info("freePMap: xaddr:0x%lx xlen:%d",
2550 (ulong) xaddr, xlen);
2556 * free from working map;
2558 if (maptype == COMMIT_PWMAP || maptype == COMMIT_WMAP) {
2559 if (maplock->flag & mlckFREEXADLIST) {
2560 xadlistlock = (struct xdlistlock *) maplock;
2561 xad = xadlistlock->xdlist;
2562 for (n = 0; n < xadlistlock->count; n++, xad++) {
2563 xaddr = addressXAD(xad);
2564 xlen = lengthXAD(xad);
2565 dbFree(ip, xaddr, (s64) xlen);
2567 jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
2568 (ulong) xaddr, xlen);
2570 } else if (maplock->flag & mlckFREEPXD) {
2571 pxdlock = (struct pxd_lock *) maplock;
2572 xaddr = addressPXD(&pxdlock->pxd);
2573 xlen = lengthPXD(&pxdlock->pxd);
2574 dbFree(ip, xaddr, (s64) xlen);
2575 jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
2576 (ulong) xaddr, xlen);
2577 } else { /* (maplock->flag & mlckFREEPXDLIST) */
2579 pxdlistlock = (struct xdlistlock *) maplock;
2580 pxd = pxdlistlock->xdlist;
2581 for (n = 0; n < pxdlistlock->count; n++, pxd++) {
2582 xaddr = addressPXD(pxd);
2583 xlen = lengthPXD(pxd);
2584 dbFree(ip, xaddr, (s64) xlen);
2585 jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
2586 (ulong) xaddr, xlen);
2595 * function: remove tlock from inode anonymous locklist
2597 void txFreelock(struct inode *ip)
2599 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
2600 struct tlock *xtlck, *tlck;
2601 lid_t xlid = 0, lid;
2603 if (!jfs_ip->atlhead)
2607 xtlck = (struct tlock *) &jfs_ip->atlhead;
2609 while ((lid = xtlck->next) != 0) {
2610 tlck = lid_to_tlock(lid);
2611 if (tlck->flag & tlckFREELOCK) {
2612 xtlck->next = tlck->next;
2620 if (jfs_ip->atlhead)
2621 jfs_ip->atltail = xlid;
2623 jfs_ip->atltail = 0;
2625 * If inode was on anon_list, remove it
2627 list_del_init(&jfs_ip->anon_inode_list);
2635 * function: abort tx before commit;
2637 * frees line-locks and segment locks for all
2638 * segments in comdata structure.
2639 * Optionally sets state of file-system to FM_DIRTY in super-block.
2640 * log age of page-frames in memory for which caller has
2641 * are reset to 0 (to avoid logwarap).
2643 void txAbort(tid_t tid, int dirty)
2646 struct metapage *mp;
2647 struct tblock *tblk = tid_to_tblock(tid);
2651 * free tlocks of the transaction
2653 for (lid = tblk->next; lid; lid = next) {
2654 tlck = lid_to_tlock(lid);
2657 JFS_IP(tlck->ip)->xtlid = 0;
2663 * reset lsn of page to avoid logwarap:
2665 * (page may have been previously committed by another
2666 * transaction(s) but has not been paged, i.e.,
2667 * it may be on logsync list even though it has not
2668 * been logged for the current tx.)
2670 if (mp->xflag & COMMIT_PAGE && mp->lsn)
2673 /* insert tlock at head of freelist */
2679 /* caller will free the transaction block */
2681 tblk->next = tblk->last = 0;
2684 * mark filesystem dirty
2687 jfs_error(tblk->sb, "txAbort");
2693 * txLazyCommit(void)
2695 * All transactions except those changing ipimap (COMMIT_FORCE) are
2696 * processed by this routine. This insures that the inode and block
2697 * allocation maps are updated in order. For synchronous transactions,
2698 * let the user thread finish processing after txUpdateMap() is called.
2700 static void txLazyCommit(struct tblock * tblk)
2702 struct jfs_log *log;
2704 while (((tblk->flag & tblkGC_READY) == 0) &&
2705 ((tblk->flag & tblkGC_UNLOCKED) == 0)) {
2706 /* We must have gotten ahead of the user thread
2708 jfs_info("jfs_lazycommit: tblk 0x%p not unlocked", tblk);
2712 jfs_info("txLazyCommit: processing tblk 0x%p", tblk);
2716 log = (struct jfs_log *) JFS_SBI(tblk->sb)->log;
2718 spin_lock_irq(&log->gclock); // LOGGC_LOCK
2720 tblk->flag |= tblkGC_COMMITTED;
2722 if (tblk->flag & tblkGC_READY)
2725 wake_up_all(&tblk->gcwait); // LOGGC_WAKEUP
2728 * Can't release log->gclock until we've tested tblk->flag
2730 if (tblk->flag & tblkGC_LAZY) {
2731 spin_unlock_irq(&log->gclock); // LOGGC_UNLOCK
2733 tblk->flag &= ~tblkGC_LAZY;
2734 txEnd(tblk - TxBlock); /* Convert back to tid */
2736 spin_unlock_irq(&log->gclock); // LOGGC_UNLOCK
2738 jfs_info("txLazyCommit: done: tblk = 0x%p", tblk);
2742 * jfs_lazycommit(void)
2744 * To be run as a kernel daemon. If lbmIODone is called in an interrupt
2745 * context, or where blocking is not wanted, this routine will process
2746 * committed transactions from the unlock queue.
2748 int jfs_lazycommit(void *arg)
2751 struct tblock *tblk;
2752 unsigned long flags;
2753 struct jfs_sb_info *sbi;
2757 jfs_commit_thread_waking = 0; /* OK to wake another thread */
2758 while (!list_empty(&TxAnchor.unlock_queue)) {
2760 list_for_each_entry(tblk, &TxAnchor.unlock_queue,
2763 sbi = JFS_SBI(tblk->sb);
2765 * For each volume, the transactions must be
2766 * handled in order. If another commit thread
2767 * is handling a tblk for this superblock,
2770 if (sbi->commit_state & IN_LAZYCOMMIT)
2773 sbi->commit_state |= IN_LAZYCOMMIT;
2777 * Remove transaction from queue
2779 list_del(&tblk->cqueue);
2785 sbi->commit_state &= ~IN_LAZYCOMMIT;
2787 * Don't continue in the for loop. (We can't
2788 * anyway, it's unsafe!) We want to go back to
2789 * the beginning of the list.
2794 /* If there was nothing to do, don't continue */
2798 /* In case a wakeup came while all threads were active */
2799 jfs_commit_thread_waking = 0;
2801 if (freezing(current)) {
2805 DECLARE_WAITQUEUE(wq, current);
2807 add_wait_queue(&jfs_commit_thread_wait, &wq);
2808 set_current_state(TASK_INTERRUPTIBLE);
2811 __set_current_state(TASK_RUNNING);
2812 remove_wait_queue(&jfs_commit_thread_wait, &wq);
2814 } while (!kthread_should_stop());
2816 if (!list_empty(&TxAnchor.unlock_queue))
2817 jfs_err("jfs_lazycommit being killed w/pending transactions!");
2819 jfs_info("jfs_lazycommit being killed\n");
2823 void txLazyUnlock(struct tblock * tblk)
2825 unsigned long flags;
2829 list_add_tail(&tblk->cqueue, &TxAnchor.unlock_queue);
2831 * Don't wake up a commit thread if there is already one servicing
2832 * this superblock, or if the last one we woke up hasn't started yet.
2834 if (!(JFS_SBI(tblk->sb)->commit_state & IN_LAZYCOMMIT) &&
2835 !jfs_commit_thread_waking) {
2836 jfs_commit_thread_waking = 1;
2837 wake_up(&jfs_commit_thread_wait);
2842 static void LogSyncRelease(struct metapage * mp)
2844 struct jfs_log *log = mp->log;
2846 assert(mp->nohomeok);
2848 metapage_homeok(mp);
2854 * Block all new transactions and push anonymous transactions to
2857 * This does almost the same thing as jfs_sync below. We don't
2858 * worry about deadlocking when jfs_tlocks_low is set, since we would
2859 * expect jfs_sync to get us out of that jam.
2861 void txQuiesce(struct super_block *sb)
2864 struct jfs_inode_info *jfs_ip;
2865 struct jfs_log *log = JFS_SBI(sb)->log;
2868 set_bit(log_QUIESCE, &log->flag);
2872 while (!list_empty(&TxAnchor.anon_list)) {
2873 jfs_ip = list_entry(TxAnchor.anon_list.next,
2874 struct jfs_inode_info,
2876 ip = &jfs_ip->vfs_inode;
2879 * inode will be removed from anonymous list
2880 * when it is committed
2883 tid = txBegin(ip->i_sb, COMMIT_INODE | COMMIT_FORCE);
2884 mutex_lock(&jfs_ip->commit_mutex);
2885 txCommit(tid, 1, &ip, 0);
2887 mutex_unlock(&jfs_ip->commit_mutex);
2889 * Just to be safe. I don't know how
2890 * long we can run without blocking
2897 * If jfs_sync is running in parallel, there could be some inodes
2898 * on anon_list2. Let's check.
2900 if (!list_empty(&TxAnchor.anon_list2)) {
2901 list_splice(&TxAnchor.anon_list2, &TxAnchor.anon_list);
2902 INIT_LIST_HEAD(&TxAnchor.anon_list2);
2908 * We may need to kick off the group commit
2910 jfs_flush_journal(log, 0);
2916 * Allows transactions to start again following txQuiesce
2918 void txResume(struct super_block *sb)
2920 struct jfs_log *log = JFS_SBI(sb)->log;
2922 clear_bit(log_QUIESCE, &log->flag);
2923 TXN_WAKEUP(&log->syncwait);
2929 * To be run as a kernel daemon. This is awakened when tlocks run low.
2930 * We write any inodes that have anonymous tlocks so they will become
2933 int jfs_sync(void *arg)
2936 struct jfs_inode_info *jfs_ip;
2941 * write each inode on the anonymous inode list
2944 while (jfs_tlocks_low && !list_empty(&TxAnchor.anon_list)) {
2945 jfs_ip = list_entry(TxAnchor.anon_list.next,
2946 struct jfs_inode_info,
2948 ip = &jfs_ip->vfs_inode;
2952 * Inode is being freed
2954 list_del_init(&jfs_ip->anon_inode_list);
2955 } else if (mutex_trylock(&jfs_ip->commit_mutex)) {
2957 * inode will be removed from anonymous list
2958 * when it is committed
2961 tid = txBegin(ip->i_sb, COMMIT_INODE);
2962 txCommit(tid, 1, &ip, 0);
2964 mutex_unlock(&jfs_ip->commit_mutex);
2968 * Just to be safe. I don't know how
2969 * long we can run without blocking
2974 /* We can't get the commit mutex. It may
2975 * be held by a thread waiting for tlock's
2976 * so let's not block here. Save it to
2977 * put back on the anon_list.
2980 /* Take off anon_list */
2981 list_del(&jfs_ip->anon_inode_list);
2983 /* Put on anon_list2 */
2984 list_add(&jfs_ip->anon_inode_list,
2985 &TxAnchor.anon_list2);
2992 /* Add anon_list2 back to anon_list */
2993 list_splice_init(&TxAnchor.anon_list2, &TxAnchor.anon_list);
2995 if (freezing(current)) {
2999 set_current_state(TASK_INTERRUPTIBLE);
3002 __set_current_state(TASK_RUNNING);
3004 } while (!kthread_should_stop());
3006 jfs_info("jfs_sync being killed");
3010 #if defined(CONFIG_PROC_FS) && defined(CONFIG_JFS_DEBUG)
3011 static int jfs_txanchor_proc_show(struct seq_file *m, void *v)
3018 waitqueue_active(&TxAnchor.freewait) ? "active" : "empty";
3020 waitqueue_active(&TxAnchor.freelockwait) ? "active" : "empty";
3022 waitqueue_active(&TxAnchor.lowlockwait) ? "active" : "empty";
3030 "freelockwait = %s\n"
3031 "lowlockwait = %s\n"
3032 "tlocksInUse = %d\n"
3033 "jfs_tlocks_low = %d\n"
3034 "unlock_queue is %sempty\n",
3040 TxAnchor.tlocksInUse,
3042 list_empty(&TxAnchor.unlock_queue) ? "" : "not ");
3046 static int jfs_txanchor_proc_open(struct inode *inode, struct file *file)
3048 return single_open(file, jfs_txanchor_proc_show, NULL);
3051 const struct file_operations jfs_txanchor_proc_fops = {
3052 .owner = THIS_MODULE,
3053 .open = jfs_txanchor_proc_open,
3055 .llseek = seq_lseek,
3056 .release = single_release,
3060 #if defined(CONFIG_PROC_FS) && defined(CONFIG_JFS_STATISTICS)
3061 static int jfs_txstats_proc_show(struct seq_file *m, void *v)
3066 "calls to txBegin = %d\n"
3067 "txBegin blocked by sync barrier = %d\n"
3068 "txBegin blocked by tlocks low = %d\n"
3069 "txBegin blocked by no free tid = %d\n"
3070 "calls to txBeginAnon = %d\n"
3071 "txBeginAnon blocked by sync barrier = %d\n"
3072 "txBeginAnon blocked by tlocks low = %d\n"
3073 "calls to txLockAlloc = %d\n"
3074 "tLockAlloc blocked by no free lock = %d\n",
3076 TxStat.txBegin_barrier,
3077 TxStat.txBegin_lockslow,
3078 TxStat.txBegin_freetid,
3080 TxStat.txBeginAnon_barrier,
3081 TxStat.txBeginAnon_lockslow,
3083 TxStat.txLockAlloc_freelock);
3087 static int jfs_txstats_proc_open(struct inode *inode, struct file *file)
3089 return single_open(file, jfs_txstats_proc_show, NULL);
3092 const struct file_operations jfs_txstats_proc_fops = {
3093 .owner = THIS_MODULE,
3094 .open = jfs_txstats_proc_open,
3096 .llseek = seq_lseek,
3097 .release = single_release,