2 * Copyright (C) International Business Machines Corp., 2000-2005
3 * Portions Copyright (C) Christoph Hellwig, 2001-2002
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
13 * the GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
21 * jfs_txnmgr.c: transaction manager
24 * transaction starts with txBegin() and ends with txCommit()
27 * tlock is acquired at the time of update;
28 * (obviate scan at commit time for xtree and dtree)
29 * tlock and mp points to each other;
30 * (no hashlist for mp -> tlock).
33 * tlock on in-memory inode:
34 * in-place tlock in the in-memory inode itself;
35 * converted to page lock by iWrite() at commit time.
37 * tlock during write()/mmap() under anonymous transaction (tid = 0):
38 * transferred (?) to transaction at commit time.
40 * use the page itself to update allocation maps
41 * (obviate intermediate replication of allocation/deallocation data)
42 * hold on to mp+lock thru update of maps
46 #include <linux/vmalloc.h>
47 #include <linux/smp_lock.h>
48 #include <linux/completion.h>
49 #include <linux/suspend.h>
50 #include <linux/module.h>
51 #include <linux/moduleparam.h>
52 #include "jfs_incore.h"
53 #include "jfs_inode.h"
54 #include "jfs_filsys.h"
55 #include "jfs_metapage.h"
56 #include "jfs_dinode.h"
59 #include "jfs_superblock.h"
60 #include "jfs_debug.h"
63 * transaction management structures
66 int freetid
; /* index of a free tid structure */
67 int freelock
; /* index first free lock word */
68 wait_queue_head_t freewait
; /* eventlist of free tblock */
69 wait_queue_head_t freelockwait
; /* eventlist of free tlock */
70 wait_queue_head_t lowlockwait
; /* eventlist of ample tlocks */
71 int tlocksInUse
; /* Number of tlocks in use */
72 spinlock_t LazyLock
; /* synchronize sync_queue & unlock_queue */
73 /* struct tblock *sync_queue; * Transactions waiting for data sync */
74 struct list_head unlock_queue
; /* Txns waiting to be released */
75 struct list_head anon_list
; /* inodes having anonymous txns */
76 struct list_head anon_list2
; /* inodes having anonymous txns
77 that couldn't be sync'ed */
80 int jfs_tlocks_low
; /* Indicates low number of available tlocks */
82 #ifdef CONFIG_JFS_STATISTICS
86 uint txBegin_lockslow
;
89 uint txBeginAnon_barrier
;
90 uint txBeginAnon_lockslow
;
92 uint txLockAlloc_freelock
;
96 static int nTxBlock
= -1; /* number of transaction blocks */
97 module_param(nTxBlock
, int, 0);
98 MODULE_PARM_DESC(nTxBlock
,
99 "Number of transaction blocks (max:65536)");
101 static int nTxLock
= -1; /* number of transaction locks */
102 module_param(nTxLock
, int, 0);
103 MODULE_PARM_DESC(nTxLock
,
104 "Number of transaction locks (max:65536)");
106 struct tblock
*TxBlock
; /* transaction block table */
107 static int TxLockLWM
; /* Low water mark for number of txLocks used */
108 static int TxLockHWM
; /* High water mark for number of txLocks used */
109 static int TxLockVHWM
; /* Very High water mark */
110 struct tlock
*TxLock
; /* transaction lock table */
113 * transaction management lock
115 static DEFINE_SPINLOCK(jfsTxnLock
);
117 #define TXN_LOCK() spin_lock(&jfsTxnLock)
118 #define TXN_UNLOCK() spin_unlock(&jfsTxnLock)
120 #define LAZY_LOCK_INIT() spin_lock_init(&TxAnchor.LazyLock);
121 #define LAZY_LOCK(flags) spin_lock_irqsave(&TxAnchor.LazyLock, flags)
122 #define LAZY_UNLOCK(flags) spin_unlock_irqrestore(&TxAnchor.LazyLock, flags)
124 DECLARE_WAIT_QUEUE_HEAD(jfs_sync_thread_wait
);
125 DECLARE_WAIT_QUEUE_HEAD(jfs_commit_thread_wait
);
126 static int jfs_commit_thread_waking
;
129 * Retry logic exist outside these macros to protect from spurrious wakeups.
131 static inline void TXN_SLEEP_DROP_LOCK(wait_queue_head_t
* event
)
133 DECLARE_WAITQUEUE(wait
, current
);
135 add_wait_queue(event
, &wait
);
136 set_current_state(TASK_UNINTERRUPTIBLE
);
139 current
->state
= TASK_RUNNING
;
140 remove_wait_queue(event
, &wait
);
143 #define TXN_SLEEP(event)\
145 TXN_SLEEP_DROP_LOCK(event);\
149 #define TXN_WAKEUP(event) wake_up_all(event)
155 tid_t maxtid
; /* 4: biggest tid ever used */
156 lid_t maxlid
; /* 4: biggest lid ever used */
157 int ntid
; /* 4: # of transactions performed */
158 int nlid
; /* 4: # of tlocks acquired */
159 int waitlock
; /* 4: # of tlock wait */
165 static int diLog(struct jfs_log
* log
, struct tblock
* tblk
, struct lrd
* lrd
,
166 struct tlock
* tlck
, struct commit
* cd
);
167 static int dataLog(struct jfs_log
* log
, struct tblock
* tblk
, struct lrd
* lrd
,
168 struct tlock
* tlck
);
169 static void dtLog(struct jfs_log
* log
, struct tblock
* tblk
, struct lrd
* lrd
,
170 struct tlock
* tlck
);
171 static void mapLog(struct jfs_log
* log
, struct tblock
* tblk
, struct lrd
* lrd
,
172 struct tlock
* tlck
);
173 static void txAllocPMap(struct inode
*ip
, struct maplock
* maplock
,
174 struct tblock
* tblk
);
175 static void txForce(struct tblock
* tblk
);
176 static int txLog(struct jfs_log
* log
, struct tblock
* tblk
,
178 static void txUpdateMap(struct tblock
* tblk
);
179 static void txRelease(struct tblock
* tblk
);
180 static void xtLog(struct jfs_log
* log
, struct tblock
* tblk
, struct lrd
* lrd
,
181 struct tlock
* tlck
);
182 static void LogSyncRelease(struct metapage
* mp
);
185 * transaction block/lock management
186 * ---------------------------------
190 * Get a transaction lock from the free list. If the number in use is
191 * greater than the high water mark, wake up the sync daemon. This should
192 * free some anonymous transaction locks. (TXN_LOCK must be held.)
194 static lid_t
txLockAlloc(void)
198 INCREMENT(TxStat
.txLockAlloc
);
199 if (!TxAnchor
.freelock
) {
200 INCREMENT(TxStat
.txLockAlloc_freelock
);
203 while (!(lid
= TxAnchor
.freelock
))
204 TXN_SLEEP(&TxAnchor
.freelockwait
);
205 TxAnchor
.freelock
= TxLock
[lid
].next
;
206 HIGHWATERMARK(stattx
.maxlid
, lid
);
207 if ((++TxAnchor
.tlocksInUse
> TxLockHWM
) && (jfs_tlocks_low
== 0)) {
208 jfs_info("txLockAlloc tlocks low");
210 wake_up(&jfs_sync_thread_wait
);
216 static void txLockFree(lid_t lid
)
219 TxLock
[lid
].next
= TxAnchor
.freelock
;
220 TxAnchor
.freelock
= lid
;
221 TxAnchor
.tlocksInUse
--;
222 if (jfs_tlocks_low
&& (TxAnchor
.tlocksInUse
< TxLockLWM
)) {
223 jfs_info("txLockFree jfs_tlocks_low no more");
225 TXN_WAKEUP(&TxAnchor
.lowlockwait
);
227 TXN_WAKEUP(&TxAnchor
.freelockwait
);
233 * FUNCTION: initialize transaction management structures
237 * serialization: single thread at jfs_init()
244 /* Set defaults for nTxLock and nTxBlock if unset */
247 if (nTxBlock
== -1) {
248 /* Base default on memory size */
250 if (si
.totalram
> (256 * 1024)) /* 1 GB */
253 nTxLock
= si
.totalram
>> 2;
254 } else if (nTxBlock
> (8 * 1024))
257 nTxLock
= nTxBlock
<< 3;
260 nTxBlock
= nTxLock
>> 3;
262 /* Verify tunable parameters */
264 nTxBlock
= 16; /* No one should set it this low */
265 if (nTxBlock
> 65536)
268 nTxLock
= 256; /* No one should set it this low */
272 printk(KERN_INFO
"JFS: nTxBlock = %d, nTxLock = %d\n",
275 * initialize transaction block (tblock) table
277 * transaction id (tid) = tblock index
278 * tid = 0 is reserved.
280 TxLockLWM
= (nTxLock
* 4) / 10;
281 TxLockHWM
= (nTxLock
* 7) / 10;
282 TxLockVHWM
= (nTxLock
* 8) / 10;
284 size
= sizeof(struct tblock
) * nTxBlock
;
285 TxBlock
= (struct tblock
*) vmalloc(size
);
289 for (k
= 1; k
< nTxBlock
- 1; k
++) {
290 TxBlock
[k
].next
= k
+ 1;
291 init_waitqueue_head(&TxBlock
[k
].gcwait
);
292 init_waitqueue_head(&TxBlock
[k
].waitor
);
295 init_waitqueue_head(&TxBlock
[k
].gcwait
);
296 init_waitqueue_head(&TxBlock
[k
].waitor
);
298 TxAnchor
.freetid
= 1;
299 init_waitqueue_head(&TxAnchor
.freewait
);
301 stattx
.maxtid
= 1; /* statistics */
304 * initialize transaction lock (tlock) table
306 * transaction lock id = tlock index
307 * tlock id = 0 is reserved.
309 size
= sizeof(struct tlock
) * nTxLock
;
310 TxLock
= (struct tlock
*) vmalloc(size
);
311 if (TxLock
== NULL
) {
316 /* initialize tlock table */
317 for (k
= 1; k
< nTxLock
- 1; k
++)
318 TxLock
[k
].next
= k
+ 1;
320 init_waitqueue_head(&TxAnchor
.freelockwait
);
321 init_waitqueue_head(&TxAnchor
.lowlockwait
);
323 TxAnchor
.freelock
= 1;
324 TxAnchor
.tlocksInUse
= 0;
325 INIT_LIST_HEAD(&TxAnchor
.anon_list
);
326 INIT_LIST_HEAD(&TxAnchor
.anon_list2
);
329 INIT_LIST_HEAD(&TxAnchor
.unlock_queue
);
331 stattx
.maxlid
= 1; /* statistics */
339 * FUNCTION: clean up when module is unloaded
352 * FUNCTION: start a transaction.
354 * PARAMETER: sb - superblock
355 * flag - force for nested tx;
357 * RETURN: tid - transaction id
359 * note: flag force allows to start tx for nested tx
360 * to prevent deadlock on logsync barrier;
362 tid_t
txBegin(struct super_block
*sb
, int flag
)
368 jfs_info("txBegin: flag = 0x%x", flag
);
369 log
= JFS_SBI(sb
)->log
;
373 INCREMENT(TxStat
.txBegin
);
376 if (!(flag
& COMMIT_FORCE
)) {
378 * synchronize with logsync barrier
380 if (test_bit(log_SYNCBARRIER
, &log
->flag
) ||
381 test_bit(log_QUIESCE
, &log
->flag
)) {
382 INCREMENT(TxStat
.txBegin_barrier
);
383 TXN_SLEEP(&log
->syncwait
);
389 * Don't begin transaction if we're getting starved for tlocks
390 * unless COMMIT_FORCE or COMMIT_INODE (which may ultimately
393 if (TxAnchor
.tlocksInUse
> TxLockVHWM
) {
394 INCREMENT(TxStat
.txBegin_lockslow
);
395 TXN_SLEEP(&TxAnchor
.lowlockwait
);
401 * allocate transaction id/block
403 if ((t
= TxAnchor
.freetid
) == 0) {
404 jfs_info("txBegin: waiting for free tid");
405 INCREMENT(TxStat
.txBegin_freetid
);
406 TXN_SLEEP(&TxAnchor
.freewait
);
410 tblk
= tid_to_tblock(t
);
412 if ((tblk
->next
== 0) && !(flag
& COMMIT_FORCE
)) {
413 /* Don't let a non-forced transaction take the last tblk */
414 jfs_info("txBegin: waiting for free tid");
415 INCREMENT(TxStat
.txBegin_freetid
);
416 TXN_SLEEP(&TxAnchor
.freewait
);
420 TxAnchor
.freetid
= tblk
->next
;
423 * initialize transaction
427 * We can't zero the whole thing or we screw up another thread being
428 * awakened after sleeping on tblk->waitor
430 * memset(tblk, 0, sizeof(struct tblock));
432 tblk
->next
= tblk
->last
= tblk
->xflag
= tblk
->flag
= tblk
->lsn
= 0;
436 tblk
->logtid
= log
->logtid
;
440 HIGHWATERMARK(stattx
.maxtid
, t
); /* statistics */
441 INCREMENT(stattx
.ntid
); /* statistics */
445 jfs_info("txBegin: returning tid = %d", t
);
451 * NAME: txBeginAnon()
453 * FUNCTION: start an anonymous transaction.
454 * Blocks if logsync or available tlocks are low to prevent
455 * anonymous tlocks from depleting supply.
457 * PARAMETER: sb - superblock
461 void txBeginAnon(struct super_block
*sb
)
465 log
= JFS_SBI(sb
)->log
;
468 INCREMENT(TxStat
.txBeginAnon
);
472 * synchronize with logsync barrier
474 if (test_bit(log_SYNCBARRIER
, &log
->flag
) ||
475 test_bit(log_QUIESCE
, &log
->flag
)) {
476 INCREMENT(TxStat
.txBeginAnon_barrier
);
477 TXN_SLEEP(&log
->syncwait
);
482 * Don't begin transaction if we're getting starved for tlocks
484 if (TxAnchor
.tlocksInUse
> TxLockVHWM
) {
485 INCREMENT(TxStat
.txBeginAnon_lockslow
);
486 TXN_SLEEP(&TxAnchor
.lowlockwait
);
495 * function: free specified transaction block.
497 * logsync barrier processing:
501 void txEnd(tid_t tid
)
503 struct tblock
*tblk
= tid_to_tblock(tid
);
506 jfs_info("txEnd: tid = %d", tid
);
510 * wakeup transactions waiting on the page locked
511 * by the current transaction
513 TXN_WAKEUP(&tblk
->waitor
);
515 log
= JFS_SBI(tblk
->sb
)->log
;
518 * Lazy commit thread can't free this guy until we mark it UNLOCKED,
519 * otherwise, we would be left with a transaction that may have been
522 * Lazy commit thread will turn off tblkGC_LAZY before calling this
525 if (tblk
->flag
& tblkGC_LAZY
) {
526 jfs_info("txEnd called w/lazy tid: %d, tblk = 0x%p", tid
, tblk
);
529 spin_lock_irq(&log
->gclock
); // LOGGC_LOCK
530 tblk
->flag
|= tblkGC_UNLOCKED
;
531 spin_unlock_irq(&log
->gclock
); // LOGGC_UNLOCK
535 jfs_info("txEnd: tid: %d, tblk = 0x%p", tid
, tblk
);
537 assert(tblk
->next
== 0);
540 * insert tblock back on freelist
542 tblk
->next
= TxAnchor
.freetid
;
543 TxAnchor
.freetid
= tid
;
546 * mark the tblock not active
548 if (--log
->active
== 0) {
549 clear_bit(log_FLUSH
, &log
->flag
);
552 * synchronize with logsync barrier
554 if (test_bit(log_SYNCBARRIER
, &log
->flag
)) {
557 /* write dirty metadata & forward log syncpt */
560 jfs_info("log barrier off: 0x%x", log
->lsn
);
562 /* enable new transactions start */
563 clear_bit(log_SYNCBARRIER
, &log
->flag
);
565 /* wakeup all waitors for logsync barrier */
566 TXN_WAKEUP(&log
->syncwait
);
575 * wakeup all waitors for a free tblock
577 TXN_WAKEUP(&TxAnchor
.freewait
);
583 * function: acquire a transaction lock on the specified <mp>
587 * return: transaction lock id
591 struct tlock
*txLock(tid_t tid
, struct inode
*ip
, struct metapage
* mp
,
594 struct jfs_inode_info
*jfs_ip
= JFS_IP(ip
);
599 struct xtlock
*xtlck
;
600 struct linelock
*linelock
;
606 if (S_ISDIR(ip
->i_mode
) && (type
& tlckXTREE
) &&
607 !(mp
->xflag
& COMMIT_PAGE
)) {
609 * Directory inode is special. It can have both an xtree tlock
610 * and a dtree tlock associated with it.
617 /* is page not locked by a transaction ? */
621 jfs_info("txLock: tid:%d ip:0x%p mp:0x%p lid:%d", tid
, ip
, mp
, lid
);
623 /* is page locked by the requester transaction ? */
624 tlck
= lid_to_tlock(lid
);
625 if ((xtid
= tlck
->tid
) == tid
) {
631 * is page locked by anonymous transaction/lock ?
633 * (page update without transaction (i.e., file write) is
634 * locked under anonymous transaction tid = 0:
635 * anonymous tlocks maintained on anonymous tlock list of
636 * the inode of the page and available to all anonymous
637 * transactions until txCommit() time at which point
638 * they are transferred to the transaction tlock list of
639 * the commiting transaction of the inode)
644 tblk
= tid_to_tblock(tid
);
646 * The order of the tlocks in the transaction is important
647 * (during truncate, child xtree pages must be freed before
648 * parent's tlocks change the working map).
649 * Take tlock off anonymous list and add to tail of
652 * Note: We really need to get rid of the tid & lid and
653 * use list_head's. This code is getting UGLY!
655 if (jfs_ip
->atlhead
== lid
) {
656 if (jfs_ip
->atltail
== lid
) {
657 /* only anonymous txn.
658 * Remove from anon_list
661 list_del_init(&jfs_ip
->anon_inode_list
);
664 jfs_ip
->atlhead
= tlck
->next
;
667 for (last
= jfs_ip
->atlhead
;
668 lid_to_tlock(last
)->next
!= lid
;
669 last
= lid_to_tlock(last
)->next
) {
672 lid_to_tlock(last
)->next
= tlck
->next
;
673 if (jfs_ip
->atltail
== lid
)
674 jfs_ip
->atltail
= last
;
677 /* insert the tlock at tail of transaction tlock list */
680 lid_to_tlock(tblk
->last
)->next
= lid
;
696 tlck
= lid_to_tlock(lid
);
705 /* mark tlock for meta-data page */
706 if (mp
->xflag
& COMMIT_PAGE
) {
708 tlck
->flag
= tlckPAGELOCK
;
710 /* mark the page dirty and nohomeok */
711 metapage_nohomeok(mp
);
713 jfs_info("locking mp = 0x%p, nohomeok = %d tid = %d tlck = 0x%p",
714 mp
, mp
->nohomeok
, tid
, tlck
);
716 /* if anonymous transaction, and buffer is on the group
717 * commit synclist, mark inode to show this. This will
718 * prevent the buffer from being marked nohomeok for too
721 if ((tid
== 0) && mp
->lsn
)
722 set_cflag(COMMIT_Synclist
, ip
);
724 /* mark tlock for in-memory inode */
726 tlck
->flag
= tlckINODELOCK
;
728 if (S_ISDIR(ip
->i_mode
))
729 tlck
->flag
|= tlckDIRECTORY
;
733 /* bind the tlock and the page */
742 * enqueue transaction lock to transaction/inode
744 /* insert the tlock at tail of transaction tlock list */
746 tblk
= tid_to_tblock(tid
);
748 lid_to_tlock(tblk
->last
)->next
= lid
;
754 /* anonymous transaction:
755 * insert the tlock at head of inode anonymous tlock list
758 tlck
->next
= jfs_ip
->atlhead
;
759 jfs_ip
->atlhead
= lid
;
760 if (tlck
->next
== 0) {
761 /* This inode's first anonymous transaction */
762 jfs_ip
->atltail
= lid
;
764 list_add_tail(&jfs_ip
->anon_inode_list
,
765 &TxAnchor
.anon_list
);
770 /* initialize type dependent area for linelock */
771 linelock
= (struct linelock
*) & tlck
->lock
;
773 linelock
->flag
= tlckLINELOCK
;
774 linelock
->maxcnt
= TLOCKSHORT
;
777 switch (type
& tlckTYPE
) {
779 linelock
->l2linesize
= L2DTSLOTSIZE
;
783 linelock
->l2linesize
= L2XTSLOTSIZE
;
785 xtlck
= (struct xtlock
*) linelock
;
786 xtlck
->header
.offset
= 0;
787 xtlck
->header
.length
= 2;
789 if (type
& tlckNEW
) {
790 xtlck
->lwm
.offset
= XTENTRYSTART
;
792 if (mp
->xflag
& COMMIT_PAGE
)
793 p
= (xtpage_t
*) mp
->data
;
795 p
= &jfs_ip
->i_xtroot
;
797 le16_to_cpu(p
->header
.nextindex
);
799 xtlck
->lwm
.length
= 0; /* ! */
800 xtlck
->twm
.offset
= 0;
801 xtlck
->hwm
.offset
= 0;
807 linelock
->l2linesize
= L2INODESLOTSIZE
;
811 linelock
->l2linesize
= L2DATASLOTSIZE
;
815 jfs_err("UFO tlock:0x%p", tlck
);
819 * update tlock vector
827 * page is being locked by another transaction:
830 /* Only locks on ipimap or ipaimap should reach here */
831 /* assert(jfs_ip->fileset == AGGREGATE_I); */
832 if (jfs_ip
->fileset
!= AGGREGATE_I
) {
833 jfs_err("txLock: trying to lock locked page!");
834 dump_mem("ip", ip
, sizeof(struct inode
));
835 dump_mem("mp", mp
, sizeof(struct metapage
));
836 dump_mem("Locker's tblk", tid_to_tblock(tid
),
837 sizeof(struct tblock
));
838 dump_mem("Tlock", tlck
, sizeof(struct tlock
));
841 INCREMENT(stattx
.waitlock
); /* statistics */
843 release_metapage(mp
);
845 xtid
= tlck
->tid
; /* reaquire after dropping TXN_LOCK */
847 jfs_info("txLock: in waitLock, tid = %d, xtid = %d, lid = %d",
850 /* Recheck everything since dropping TXN_LOCK */
851 if (xtid
&& (tlck
->mp
== mp
) && (mp
->lid
== lid
))
852 TXN_SLEEP_DROP_LOCK(&tid_to_tblock(xtid
)->waitor
);
855 jfs_info("txLock: awakened tid = %d, lid = %d", tid
, lid
);
863 * FUNCTION: Release buffers associated with transaction locks, but don't
864 * mark homeok yet. The allows other transactions to modify
865 * buffers, but won't let them go to disk until commit record
866 * actually gets written.
871 * RETURN: Errors from subroutines.
873 static void txRelease(struct tblock
* tblk
)
881 for (lid
= tblk
->next
; lid
; lid
= tlck
->next
) {
882 tlck
= lid_to_tlock(lid
);
883 if ((mp
= tlck
->mp
) != NULL
&&
884 (tlck
->type
& tlckBTROOT
) == 0) {
885 assert(mp
->xflag
& COMMIT_PAGE
);
891 * wakeup transactions waiting on a page locked
892 * by the current transaction
894 TXN_WAKEUP(&tblk
->waitor
);
902 * FUNCTION: Initiates pageout of pages modified by tid in journalled
903 * objects and frees their lockwords.
905 static void txUnlock(struct tblock
* tblk
)
908 struct linelock
*linelock
;
909 lid_t lid
, next
, llid
, k
;
915 jfs_info("txUnlock: tblk = 0x%p", tblk
);
916 log
= JFS_SBI(tblk
->sb
)->log
;
919 * mark page under tlock homeok (its log has been written):
921 for (lid
= tblk
->next
; lid
; lid
= next
) {
922 tlck
= lid_to_tlock(lid
);
925 jfs_info("unlocking lid = %d, tlck = 0x%p", lid
, tlck
);
927 /* unbind page from tlock */
928 if ((mp
= tlck
->mp
) != NULL
&&
929 (tlck
->type
& tlckBTROOT
) == 0) {
930 assert(mp
->xflag
& COMMIT_PAGE
);
936 assert(mp
->nohomeok
> 0);
937 _metapage_homeok(mp
);
939 /* inherit younger/larger clsn */
940 LOGSYNC_LOCK(log
, flags
);
942 logdiff(difft
, tblk
->clsn
, log
);
943 logdiff(diffp
, mp
->clsn
, log
);
945 mp
->clsn
= tblk
->clsn
;
947 mp
->clsn
= tblk
->clsn
;
948 LOGSYNC_UNLOCK(log
, flags
);
950 assert(!(tlck
->flag
& tlckFREEPAGE
));
955 /* insert tlock, and linelock(s) of the tlock if any,
956 * at head of freelist
960 llid
= ((struct linelock
*) & tlck
->lock
)->next
;
962 linelock
= (struct linelock
*) lid_to_tlock(llid
);
971 tblk
->next
= tblk
->last
= 0;
974 * remove tblock from logsynclist
975 * (allocation map pages inherited lsn of tblk and
976 * has been inserted in logsync list at txUpdateMap())
979 LOGSYNC_LOCK(log
, flags
);
981 list_del(&tblk
->synclist
);
982 LOGSYNC_UNLOCK(log
, flags
);
989 * function: allocate a transaction lock for freed page/entry;
990 * for freed page, maplock is used as xtlock/dtlock type;
992 struct tlock
*txMaplock(tid_t tid
, struct inode
*ip
, int type
)
994 struct jfs_inode_info
*jfs_ip
= JFS_IP(ip
);
998 struct maplock
*maplock
;
1005 lid
= txLockAlloc();
1006 tlck
= lid_to_tlock(lid
);
1013 /* bind the tlock and the object */
1014 tlck
->flag
= tlckINODELOCK
;
1015 if (S_ISDIR(ip
->i_mode
))
1016 tlck
->flag
|= tlckDIRECTORY
;
1023 * enqueue transaction lock to transaction/inode
1025 /* insert the tlock at tail of transaction tlock list */
1027 tblk
= tid_to_tblock(tid
);
1029 lid_to_tlock(tblk
->last
)->next
= lid
;
1035 /* anonymous transaction:
1036 * insert the tlock at head of inode anonymous tlock list
1039 tlck
->next
= jfs_ip
->atlhead
;
1040 jfs_ip
->atlhead
= lid
;
1041 if (tlck
->next
== 0) {
1042 /* This inode's first anonymous transaction */
1043 jfs_ip
->atltail
= lid
;
1044 list_add_tail(&jfs_ip
->anon_inode_list
,
1045 &TxAnchor
.anon_list
);
1051 /* initialize type dependent area for maplock */
1052 maplock
= (struct maplock
*) & tlck
->lock
;
1054 maplock
->maxcnt
= 0;
1063 * function: allocate a transaction lock for log vector list
1065 struct linelock
*txLinelock(struct linelock
* tlock
)
1069 struct linelock
*linelock
;
1073 /* allocate a TxLock structure */
1074 lid
= txLockAlloc();
1075 tlck
= lid_to_tlock(lid
);
1079 /* initialize linelock */
1080 linelock
= (struct linelock
*) tlck
;
1082 linelock
->flag
= tlckLINELOCK
;
1083 linelock
->maxcnt
= TLOCKLONG
;
1084 linelock
->index
= 0;
1085 if (tlck
->flag
& tlckDIRECTORY
)
1086 linelock
->flag
|= tlckDIRECTORY
;
1088 /* append linelock after tlock */
1089 linelock
->next
= tlock
->next
;
1096 * transaction commit management
1097 * -----------------------------
1103 * FUNCTION: commit the changes to the objects specified in
1104 * clist. For journalled segments only the
1105 * changes of the caller are committed, ie by tid.
1106 * for non-journalled segments the data are flushed to
1107 * disk and then the change to the disk inode and indirect
1108 * blocks committed (so blocks newly allocated to the
1109 * segment will be made a part of the segment atomically).
1111 * all of the segments specified in clist must be in
1112 * one file system. no more than 6 segments are needed
1113 * to handle all unix svcs.
1115 * if the i_nlink field (i.e. disk inode link count)
1116 * is zero, and the type of inode is a regular file or
1117 * directory, or symbolic link , the inode is truncated
1118 * to zero length. the truncation is committed but the
1119 * VM resources are unaffected until it is closed (see
1127 * on entry the inode lock on each segment is assumed
1132 int txCommit(tid_t tid
, /* transaction identifier */
1133 int nip
, /* number of inodes to commit */
1134 struct inode
**iplist
, /* list of inode to commit */
1139 struct jfs_log
*log
;
1140 struct tblock
*tblk
;
1144 struct jfs_inode_info
*jfs_ip
;
1147 struct super_block
*sb
;
1149 jfs_info("txCommit, tid = %d, flag = %d", tid
, flag
);
1150 /* is read-only file system ? */
1151 if (isReadOnly(iplist
[0])) {
1156 sb
= cd
.sb
= iplist
[0]->i_sb
;
1160 tid
= txBegin(sb
, 0);
1161 tblk
= tid_to_tblock(tid
);
1164 * initialize commit structure
1166 log
= JFS_SBI(sb
)->log
;
1169 /* initialize log record descriptor in commit */
1171 lrd
->logtid
= cpu_to_le32(tblk
->logtid
);
1174 tblk
->xflag
|= flag
;
1176 if ((flag
& (COMMIT_FORCE
| COMMIT_SYNC
)) == 0)
1177 tblk
->xflag
|= COMMIT_LAZY
;
1179 * prepare non-journaled objects for commit
1181 * flush data pages of non-journaled file
1182 * to prevent the file getting non-initialized disk blocks
1190 * acquire transaction lock on (on-disk) inodes
1192 * update on-disk inode from in-memory inode
1193 * acquiring transaction locks for AFTER records
1194 * on the on-disk inode of file object
1196 * sort the inodes array by inode number in descending order
1197 * to prevent deadlock when acquiring transaction lock
1198 * of on-disk inodes on multiple on-disk inode pages by
1199 * multiple concurrent transactions
1201 for (k
= 0; k
< cd
.nip
; k
++) {
1202 top
= (cd
.iplist
[k
])->i_ino
;
1203 for (n
= k
+ 1; n
< cd
.nip
; n
++) {
1205 if (ip
->i_ino
> top
) {
1207 cd
.iplist
[n
] = cd
.iplist
[k
];
1213 jfs_ip
= JFS_IP(ip
);
1216 * BUGBUG - This code has temporarily been removed. The
1217 * intent is to ensure that any file data is written before
1218 * the metadata is committed to the journal. This prevents
1219 * uninitialized data from appearing in a file after the
1220 * journal has been replayed. (The uninitialized data
1221 * could be sensitive data removed by another user.)
1223 * The problem now is that we are holding the IWRITELOCK
1224 * on the inode, and calling filemap_fdatawrite on an
1225 * unmapped page will cause a deadlock in jfs_get_block.
1227 * The long term solution is to pare down the use of
1228 * IWRITELOCK. We are currently holding it too long.
1229 * We could also be smarter about which data pages need
1230 * to be written before the transaction is committed and
1231 * when we don't need to worry about it at all.
1233 * if ((!S_ISDIR(ip->i_mode))
1234 * && (tblk->flag & COMMIT_DELETE) == 0) {
1235 * filemap_fdatawrite(ip->i_mapping);
1236 * filemap_fdatawait(ip->i_mapping);
1241 * Mark inode as not dirty. It will still be on the dirty
1242 * inode list, but we'll know not to commit it again unless
1243 * it gets marked dirty again
1245 clear_cflag(COMMIT_Dirty
, ip
);
1247 /* inherit anonymous tlock(s) of inode */
1248 if (jfs_ip
->atlhead
) {
1249 lid_to_tlock(jfs_ip
->atltail
)->next
= tblk
->next
;
1250 tblk
->next
= jfs_ip
->atlhead
;
1252 tblk
->last
= jfs_ip
->atltail
;
1253 jfs_ip
->atlhead
= jfs_ip
->atltail
= 0;
1255 list_del_init(&jfs_ip
->anon_inode_list
);
1260 * acquire transaction lock on on-disk inode page
1261 * (become first tlock of the tblk's tlock list)
1263 if (((rc
= diWrite(tid
, ip
))))
1268 * write log records from transaction locks
1270 * txUpdateMap() resets XAD_NEW in XAD.
1272 if ((rc
= txLog(log
, tblk
, &cd
)))
1276 * Ensure that inode isn't reused before
1277 * lazy commit thread finishes processing
1279 if (tblk
->xflag
& COMMIT_DELETE
) {
1280 atomic_inc(&tblk
->u
.ip
->i_count
);
1282 * Avoid a rare deadlock
1284 * If the inode is locked, we may be blocked in
1285 * jfs_commit_inode. If so, we don't want the
1286 * lazy_commit thread doing the last iput() on the inode
1287 * since that may block on the locked inode. Instead,
1288 * commit the transaction synchronously, so the last iput
1289 * will be done by the calling thread (or later)
1291 if (tblk
->u
.ip
->i_state
& I_LOCK
)
1292 tblk
->xflag
&= ~COMMIT_LAZY
;
1295 ASSERT((!(tblk
->xflag
& COMMIT_DELETE
)) ||
1296 ((tblk
->u
.ip
->i_nlink
== 0) &&
1297 !test_cflag(COMMIT_Nolink
, tblk
->u
.ip
)));
1300 * write COMMIT log record
1302 lrd
->type
= cpu_to_le16(LOG_COMMIT
);
1304 lsn
= lmLog(log
, tblk
, lrd
, NULL
);
1306 lmGroupCommit(log
, tblk
);
1309 * - transaction is now committed -
1313 * force pages in careful update
1314 * (imap addressing structure update)
1316 if (flag
& COMMIT_FORCE
)
1320 * update allocation map.
1322 * update inode allocation map and inode:
1323 * free pager lock on memory object of inode if any.
1324 * update block allocation map.
1326 * txUpdateMap() resets XAD_NEW in XAD.
1328 if (tblk
->xflag
& COMMIT_FORCE
)
1332 * free transaction locks and pageout/free pages
1336 if ((tblk
->flag
& tblkGC_LAZY
) == 0)
1341 * reset in-memory object state
1343 for (k
= 0; k
< cd
.nip
; k
++) {
1345 jfs_ip
= JFS_IP(ip
);
1348 * reset in-memory inode state
1359 jfs_info("txCommit: tid = %d, returning %d", tid
, rc
);
1366 * FUNCTION: Writes AFTER log records for all lines modified
1367 * by tid for segments specified by inodes in comdata.
1368 * Code assumes only WRITELOCKS are recorded in lockwords.
1374 static int txLog(struct jfs_log
* log
, struct tblock
* tblk
, struct commit
* cd
)
1380 struct lrd
*lrd
= &cd
->lrd
;
1383 * write log record(s) for each tlock of transaction,
1385 for (lid
= tblk
->next
; lid
; lid
= tlck
->next
) {
1386 tlck
= lid_to_tlock(lid
);
1388 tlck
->flag
|= tlckLOG
;
1390 /* initialize lrd common */
1392 lrd
->aggregate
= cpu_to_le32(JFS_SBI(ip
->i_sb
)->aggregate
);
1393 lrd
->log
.redopage
.fileset
= cpu_to_le32(JFS_IP(ip
)->fileset
);
1394 lrd
->log
.redopage
.inode
= cpu_to_le32(ip
->i_ino
);
1396 /* write log record of page from the tlock */
1397 switch (tlck
->type
& tlckTYPE
) {
1399 xtLog(log
, tblk
, lrd
, tlck
);
1403 dtLog(log
, tblk
, lrd
, tlck
);
1407 diLog(log
, tblk
, lrd
, tlck
, cd
);
1411 mapLog(log
, tblk
, lrd
, tlck
);
1415 dataLog(log
, tblk
, lrd
, tlck
);
1419 jfs_err("UFO tlock:0x%p", tlck
);
1429 * function: log inode tlock and format maplock to update bmap;
1431 static int diLog(struct jfs_log
* log
, struct tblock
* tblk
, struct lrd
* lrd
,
1432 struct tlock
* tlck
, struct commit
* cd
)
1435 struct metapage
*mp
;
1437 struct pxd_lock
*pxdlock
;
1441 /* initialize as REDOPAGE record format */
1442 lrd
->log
.redopage
.type
= cpu_to_le16(LOG_INODE
);
1443 lrd
->log
.redopage
.l2linesize
= cpu_to_le16(L2INODESLOTSIZE
);
1445 pxd
= &lrd
->log
.redopage
.pxd
;
1450 if (tlck
->type
& tlckENTRY
) {
1451 /* log after-image for logredo(): */
1452 lrd
->type
= cpu_to_le16(LOG_REDOPAGE
);
1453 PXDaddress(pxd
, mp
->index
);
1455 mp
->logical_size
>> tblk
->sb
->s_blocksize_bits
);
1456 lrd
->backchain
= cpu_to_le32(lmLog(log
, tblk
, lrd
, tlck
));
1458 /* mark page as homeward bound */
1459 tlck
->flag
|= tlckWRITEPAGE
;
1460 } else if (tlck
->type
& tlckFREE
) {
1464 * (pages of the freed inode extent have been invalidated and
1465 * a maplock for free of the extent has been formatted at
1468 * the tlock had been acquired on the inode allocation map page
1469 * (iag) that specifies the freed extent, even though the map
1470 * page is not itself logged, to prevent pageout of the map
1471 * page before the log;
1474 /* log LOG_NOREDOINOEXT of the freed inode extent for
1475 * logredo() to start NoRedoPage filters, and to update
1476 * imap and bmap for free of the extent;
1478 lrd
->type
= cpu_to_le16(LOG_NOREDOINOEXT
);
1480 * For the LOG_NOREDOINOEXT record, we need
1481 * to pass the IAG number and inode extent
1482 * index (within that IAG) from which the
1483 * the extent being released. These have been
1484 * passed to us in the iplist[1] and iplist[2].
1486 lrd
->log
.noredoinoext
.iagnum
=
1487 cpu_to_le32((u32
) (size_t) cd
->iplist
[1]);
1488 lrd
->log
.noredoinoext
.inoext_idx
=
1489 cpu_to_le32((u32
) (size_t) cd
->iplist
[2]);
1491 pxdlock
= (struct pxd_lock
*) & tlck
->lock
;
1492 *pxd
= pxdlock
->pxd
;
1493 lrd
->backchain
= cpu_to_le32(lmLog(log
, tblk
, lrd
, NULL
));
1496 tlck
->flag
|= tlckUPDATEMAP
;
1498 /* mark page as homeward bound */
1499 tlck
->flag
|= tlckWRITEPAGE
;
1501 jfs_err("diLog: UFO type tlck:0x%p", tlck
);
1504 * alloc/free external EA extent
1506 * a maplock for txUpdateMap() to update bPWMAP for alloc/free
1507 * of the extent has been formatted at txLock() time;
1510 assert(tlck
->type
& tlckEA
);
1512 /* log LOG_UPDATEMAP for logredo() to update bmap for
1513 * alloc of new (and free of old) external EA extent;
1515 lrd
->type
= cpu_to_le16(LOG_UPDATEMAP
);
1516 pxdlock
= (struct pxd_lock
*) & tlck
->lock
;
1517 nlock
= pxdlock
->index
;
1518 for (i
= 0; i
< nlock
; i
++, pxdlock
++) {
1519 if (pxdlock
->flag
& mlckALLOCPXD
)
1520 lrd
->log
.updatemap
.type
=
1521 cpu_to_le16(LOG_ALLOCPXD
);
1523 lrd
->log
.updatemap
.type
=
1524 cpu_to_le16(LOG_FREEPXD
);
1525 lrd
->log
.updatemap
.nxd
= cpu_to_le16(1);
1526 lrd
->log
.updatemap
.pxd
= pxdlock
->pxd
;
1528 cpu_to_le32(lmLog(log
, tblk
, lrd
, NULL
));
1532 tlck
->flag
|= tlckUPDATEMAP
;
1534 #endif /* _JFS_WIP */
1542 * function: log data tlock
1544 static int dataLog(struct jfs_log
* log
, struct tblock
* tblk
, struct lrd
* lrd
,
1545 struct tlock
* tlck
)
1547 struct metapage
*mp
;
1552 /* initialize as REDOPAGE record format */
1553 lrd
->log
.redopage
.type
= cpu_to_le16(LOG_DATA
);
1554 lrd
->log
.redopage
.l2linesize
= cpu_to_le16(L2DATASLOTSIZE
);
1556 pxd
= &lrd
->log
.redopage
.pxd
;
1558 /* log after-image for logredo(): */
1559 lrd
->type
= cpu_to_le16(LOG_REDOPAGE
);
1561 if (jfs_dirtable_inline(tlck
->ip
)) {
1563 * The table has been truncated, we've must have deleted
1564 * the last entry, so don't bother logging this
1568 metapage_homeok(mp
);
1569 discard_metapage(mp
);
1574 PXDaddress(pxd
, mp
->index
);
1575 PXDlength(pxd
, mp
->logical_size
>> tblk
->sb
->s_blocksize_bits
);
1577 lrd
->backchain
= cpu_to_le32(lmLog(log
, tblk
, lrd
, tlck
));
1579 /* mark page as homeward bound */
1580 tlck
->flag
|= tlckWRITEPAGE
;
1588 * function: log dtree tlock and format maplock to update bmap;
1590 static void dtLog(struct jfs_log
* log
, struct tblock
* tblk
, struct lrd
* lrd
,
1591 struct tlock
* tlck
)
1593 struct metapage
*mp
;
1594 struct pxd_lock
*pxdlock
;
1599 /* initialize as REDOPAGE/NOREDOPAGE record format */
1600 lrd
->log
.redopage
.type
= cpu_to_le16(LOG_DTREE
);
1601 lrd
->log
.redopage
.l2linesize
= cpu_to_le16(L2DTSLOTSIZE
);
1603 pxd
= &lrd
->log
.redopage
.pxd
;
1605 if (tlck
->type
& tlckBTROOT
)
1606 lrd
->log
.redopage
.type
|= cpu_to_le16(LOG_BTROOT
);
1609 * page extension via relocation: entry insertion;
1610 * page extension in-place: entry insertion;
1611 * new right page from page split, reinitialized in-line
1612 * root from root page split: entry insertion;
1614 if (tlck
->type
& (tlckNEW
| tlckEXTEND
)) {
1615 /* log after-image of the new page for logredo():
1616 * mark log (LOG_NEW) for logredo() to initialize
1617 * freelist and update bmap for alloc of the new page;
1619 lrd
->type
= cpu_to_le16(LOG_REDOPAGE
);
1620 if (tlck
->type
& tlckEXTEND
)
1621 lrd
->log
.redopage
.type
|= cpu_to_le16(LOG_EXTEND
);
1623 lrd
->log
.redopage
.type
|= cpu_to_le16(LOG_NEW
);
1624 PXDaddress(pxd
, mp
->index
);
1626 mp
->logical_size
>> tblk
->sb
->s_blocksize_bits
);
1627 lrd
->backchain
= cpu_to_le32(lmLog(log
, tblk
, lrd
, tlck
));
1629 /* format a maplock for txUpdateMap() to update bPMAP for
1630 * alloc of the new page;
1632 if (tlck
->type
& tlckBTROOT
)
1634 tlck
->flag
|= tlckUPDATEMAP
;
1635 pxdlock
= (struct pxd_lock
*) & tlck
->lock
;
1636 pxdlock
->flag
= mlckALLOCPXD
;
1637 pxdlock
->pxd
= *pxd
;
1641 /* mark page as homeward bound */
1642 tlck
->flag
|= tlckWRITEPAGE
;
1647 * entry insertion/deletion,
1648 * sibling page link update (old right page before split);
1650 if (tlck
->type
& (tlckENTRY
| tlckRELINK
)) {
1651 /* log after-image for logredo(): */
1652 lrd
->type
= cpu_to_le16(LOG_REDOPAGE
);
1653 PXDaddress(pxd
, mp
->index
);
1655 mp
->logical_size
>> tblk
->sb
->s_blocksize_bits
);
1656 lrd
->backchain
= cpu_to_le32(lmLog(log
, tblk
, lrd
, tlck
));
1658 /* mark page as homeward bound */
1659 tlck
->flag
|= tlckWRITEPAGE
;
1664 * page deletion: page has been invalidated
1665 * page relocation: source extent
1667 * a maplock for free of the page has been formatted
1668 * at txLock() time);
1670 if (tlck
->type
& (tlckFREE
| tlckRELOCATE
)) {
1671 /* log LOG_NOREDOPAGE of the deleted page for logredo()
1672 * to start NoRedoPage filter and to update bmap for free
1673 * of the deletd page
1675 lrd
->type
= cpu_to_le16(LOG_NOREDOPAGE
);
1676 pxdlock
= (struct pxd_lock
*) & tlck
->lock
;
1677 *pxd
= pxdlock
->pxd
;
1678 lrd
->backchain
= cpu_to_le32(lmLog(log
, tblk
, lrd
, NULL
));
1680 /* a maplock for txUpdateMap() for free of the page
1681 * has been formatted at txLock() time;
1683 tlck
->flag
|= tlckUPDATEMAP
;
1691 * function: log xtree tlock and format maplock to update bmap;
1693 static void xtLog(struct jfs_log
* log
, struct tblock
* tblk
, struct lrd
* lrd
,
1694 struct tlock
* tlck
)
1697 struct metapage
*mp
;
1699 struct xtlock
*xtlck
;
1700 struct maplock
*maplock
;
1701 struct xdlistlock
*xadlock
;
1702 struct pxd_lock
*pxdlock
;
1709 /* initialize as REDOPAGE/NOREDOPAGE record format */
1710 lrd
->log
.redopage
.type
= cpu_to_le16(LOG_XTREE
);
1711 lrd
->log
.redopage
.l2linesize
= cpu_to_le16(L2XTSLOTSIZE
);
1713 page_pxd
= &lrd
->log
.redopage
.pxd
;
1715 if (tlck
->type
& tlckBTROOT
) {
1716 lrd
->log
.redopage
.type
|= cpu_to_le16(LOG_BTROOT
);
1717 p
= &JFS_IP(ip
)->i_xtroot
;
1718 if (S_ISDIR(ip
->i_mode
))
1719 lrd
->log
.redopage
.type
|=
1720 cpu_to_le16(LOG_DIR_XTREE
);
1722 p
= (xtpage_t
*) mp
->data
;
1723 next
= le16_to_cpu(p
->header
.nextindex
);
1725 xtlck
= (struct xtlock
*) & tlck
->lock
;
1727 maplock
= (struct maplock
*) & tlck
->lock
;
1728 xadlock
= (struct xdlistlock
*) maplock
;
1731 * entry insertion/extension;
1732 * sibling page link update (old right page before split);
1734 if (tlck
->type
& (tlckNEW
| tlckGROW
| tlckRELINK
)) {
1735 /* log after-image for logredo():
1736 * logredo() will update bmap for alloc of new/extended
1737 * extents (XAD_NEW|XAD_EXTEND) of XAD[lwm:next) from
1738 * after-image of XADlist;
1739 * logredo() resets (XAD_NEW|XAD_EXTEND) flag when
1740 * applying the after-image to the meta-data page.
1742 lrd
->type
= cpu_to_le16(LOG_REDOPAGE
);
1743 PXDaddress(page_pxd
, mp
->index
);
1745 mp
->logical_size
>> tblk
->sb
->s_blocksize_bits
);
1746 lrd
->backchain
= cpu_to_le32(lmLog(log
, tblk
, lrd
, tlck
));
1748 /* format a maplock for txUpdateMap() to update bPMAP
1749 * for alloc of new/extended extents of XAD[lwm:next)
1750 * from the page itself;
1751 * txUpdateMap() resets (XAD_NEW|XAD_EXTEND) flag.
1753 lwm
= xtlck
->lwm
.offset
;
1755 lwm
= XTPAGEMAXSLOT
;
1760 jfs_err("xtLog: lwm > next\n");
1763 tlck
->flag
|= tlckUPDATEMAP
;
1764 xadlock
->flag
= mlckALLOCXADLIST
;
1765 xadlock
->count
= next
- lwm
;
1766 if ((xadlock
->count
<= 4) && (tblk
->xflag
& COMMIT_LAZY
)) {
1770 * Lazy commit may allow xtree to be modified before
1771 * txUpdateMap runs. Copy xad into linelock to
1772 * preserve correct data.
1774 * We can fit twice as may pxd's as xads in the lock
1776 xadlock
->flag
= mlckALLOCPXDLIST
;
1777 pxd
= xadlock
->xdlist
= &xtlck
->pxdlock
;
1778 for (i
= 0; i
< xadlock
->count
; i
++) {
1779 PXDaddress(pxd
, addressXAD(&p
->xad
[lwm
+ i
]));
1780 PXDlength(pxd
, lengthXAD(&p
->xad
[lwm
+ i
]));
1781 p
->xad
[lwm
+ i
].flag
&=
1782 ~(XAD_NEW
| XAD_EXTENDED
);
1787 * xdlist will point to into inode's xtree, ensure
1788 * that transaction is not committed lazily.
1790 xadlock
->flag
= mlckALLOCXADLIST
;
1791 xadlock
->xdlist
= &p
->xad
[lwm
];
1792 tblk
->xflag
&= ~COMMIT_LAZY
;
1794 jfs_info("xtLog: alloc ip:0x%p mp:0x%p tlck:0x%p lwm:%d "
1795 "count:%d", tlck
->ip
, mp
, tlck
, lwm
, xadlock
->count
);
1800 /* mark page as homeward bound */
1801 tlck
->flag
|= tlckWRITEPAGE
;
1807 * page deletion: file deletion/truncation (ref. xtTruncate())
1809 * (page will be invalidated after log is written and bmap
1810 * is updated from the page);
1812 if (tlck
->type
& tlckFREE
) {
1813 /* LOG_NOREDOPAGE log for NoRedoPage filter:
1814 * if page free from file delete, NoRedoFile filter from
1815 * inode image of zero link count will subsume NoRedoPage
1816 * filters for each page;
1817 * if page free from file truncattion, write NoRedoPage
1820 * upadte of block allocation map for the page itself:
1821 * if page free from deletion and truncation, LOG_UPDATEMAP
1822 * log for the page itself is generated from processing
1823 * its parent page xad entries;
1825 /* if page free from file truncation, log LOG_NOREDOPAGE
1826 * of the deleted page for logredo() to start NoRedoPage
1827 * filter for the page;
1829 if (tblk
->xflag
& COMMIT_TRUNCATE
) {
1830 /* write NOREDOPAGE for the page */
1831 lrd
->type
= cpu_to_le16(LOG_NOREDOPAGE
);
1832 PXDaddress(page_pxd
, mp
->index
);
1834 mp
->logical_size
>> tblk
->sb
->
1837 cpu_to_le32(lmLog(log
, tblk
, lrd
, NULL
));
1839 if (tlck
->type
& tlckBTROOT
) {
1840 /* Empty xtree must be logged */
1841 lrd
->type
= cpu_to_le16(LOG_REDOPAGE
);
1843 cpu_to_le32(lmLog(log
, tblk
, lrd
, tlck
));
1847 /* init LOG_UPDATEMAP of the freed extents
1848 * XAD[XTENTRYSTART:hwm) from the deleted page itself
1849 * for logredo() to update bmap;
1851 lrd
->type
= cpu_to_le16(LOG_UPDATEMAP
);
1852 lrd
->log
.updatemap
.type
= cpu_to_le16(LOG_FREEXADLIST
);
1853 xtlck
= (struct xtlock
*) & tlck
->lock
;
1854 hwm
= xtlck
->hwm
.offset
;
1855 lrd
->log
.updatemap
.nxd
=
1856 cpu_to_le16(hwm
- XTENTRYSTART
+ 1);
1857 /* reformat linelock for lmLog() */
1858 xtlck
->header
.offset
= XTENTRYSTART
;
1859 xtlck
->header
.length
= hwm
- XTENTRYSTART
+ 1;
1861 lrd
->backchain
= cpu_to_le32(lmLog(log
, tblk
, lrd
, tlck
));
1863 /* format a maplock for txUpdateMap() to update bmap
1864 * to free extents of XAD[XTENTRYSTART:hwm) from the
1865 * deleted page itself;
1867 tlck
->flag
|= tlckUPDATEMAP
;
1868 xadlock
->count
= hwm
- XTENTRYSTART
+ 1;
1869 if ((xadlock
->count
<= 4) && (tblk
->xflag
& COMMIT_LAZY
)) {
1873 * Lazy commit may allow xtree to be modified before
1874 * txUpdateMap runs. Copy xad into linelock to
1875 * preserve correct data.
1877 * We can fit twice as may pxd's as xads in the lock
1879 xadlock
->flag
= mlckFREEPXDLIST
;
1880 pxd
= xadlock
->xdlist
= &xtlck
->pxdlock
;
1881 for (i
= 0; i
< xadlock
->count
; i
++) {
1883 addressXAD(&p
->xad
[XTENTRYSTART
+ i
]));
1885 lengthXAD(&p
->xad
[XTENTRYSTART
+ i
]));
1890 * xdlist will point to into inode's xtree, ensure
1891 * that transaction is not committed lazily.
1893 xadlock
->flag
= mlckFREEXADLIST
;
1894 xadlock
->xdlist
= &p
->xad
[XTENTRYSTART
];
1895 tblk
->xflag
&= ~COMMIT_LAZY
;
1897 jfs_info("xtLog: free ip:0x%p mp:0x%p count:%d lwm:2",
1898 tlck
->ip
, mp
, xadlock
->count
);
1902 /* mark page as invalid */
1903 if (((tblk
->xflag
& COMMIT_PWMAP
) || S_ISDIR(ip
->i_mode
))
1904 && !(tlck
->type
& tlckBTROOT
))
1905 tlck
->flag
|= tlckFREEPAGE
;
1907 else (tblk->xflag & COMMIT_PMAP)
1914 * page/entry truncation: file truncation (ref. xtTruncate())
1916 * |----------+------+------+---------------|
1918 * | | hwm - hwm before truncation
1919 * | next - truncation point
1920 * lwm - lwm before truncation
1923 if (tlck
->type
& tlckTRUNCATE
) {
1924 pxd_t pxd
; /* truncated extent of xad */
1928 * For truncation the entire linelock may be used, so it would
1929 * be difficult to store xad list in linelock itself.
1930 * Therefore, we'll just force transaction to be committed
1931 * synchronously, so that xtree pages won't be changed before
1934 tblk
->xflag
&= ~COMMIT_LAZY
;
1935 lwm
= xtlck
->lwm
.offset
;
1937 lwm
= XTPAGEMAXSLOT
;
1938 hwm
= xtlck
->hwm
.offset
;
1939 twm
= xtlck
->twm
.offset
;
1944 /* log after-image for logredo():
1946 * logredo() will update bmap for alloc of new/extended
1947 * extents (XAD_NEW|XAD_EXTEND) of XAD[lwm:next) from
1948 * after-image of XADlist;
1949 * logredo() resets (XAD_NEW|XAD_EXTEND) flag when
1950 * applying the after-image to the meta-data page.
1952 lrd
->type
= cpu_to_le16(LOG_REDOPAGE
);
1953 PXDaddress(page_pxd
, mp
->index
);
1955 mp
->logical_size
>> tblk
->sb
->s_blocksize_bits
);
1956 lrd
->backchain
= cpu_to_le32(lmLog(log
, tblk
, lrd
, tlck
));
1959 * truncate entry XAD[twm == next - 1]:
1961 if (twm
== next
- 1) {
1962 /* init LOG_UPDATEMAP for logredo() to update bmap for
1963 * free of truncated delta extent of the truncated
1964 * entry XAD[next - 1]:
1965 * (xtlck->pxdlock = truncated delta extent);
1967 pxdlock
= (struct pxd_lock
*) & xtlck
->pxdlock
;
1968 /* assert(pxdlock->type & tlckTRUNCATE); */
1969 lrd
->type
= cpu_to_le16(LOG_UPDATEMAP
);
1970 lrd
->log
.updatemap
.type
= cpu_to_le16(LOG_FREEPXD
);
1971 lrd
->log
.updatemap
.nxd
= cpu_to_le16(1);
1972 lrd
->log
.updatemap
.pxd
= pxdlock
->pxd
;
1973 pxd
= pxdlock
->pxd
; /* save to format maplock */
1975 cpu_to_le32(lmLog(log
, tblk
, lrd
, NULL
));
1979 * free entries XAD[next:hwm]:
1982 /* init LOG_UPDATEMAP of the freed extents
1983 * XAD[next:hwm] from the deleted page itself
1984 * for logredo() to update bmap;
1986 lrd
->type
= cpu_to_le16(LOG_UPDATEMAP
);
1987 lrd
->log
.updatemap
.type
=
1988 cpu_to_le16(LOG_FREEXADLIST
);
1989 xtlck
= (struct xtlock
*) & tlck
->lock
;
1990 hwm
= xtlck
->hwm
.offset
;
1991 lrd
->log
.updatemap
.nxd
=
1992 cpu_to_le16(hwm
- next
+ 1);
1993 /* reformat linelock for lmLog() */
1994 xtlck
->header
.offset
= next
;
1995 xtlck
->header
.length
= hwm
- next
+ 1;
1998 cpu_to_le32(lmLog(log
, tblk
, lrd
, tlck
));
2002 * format maplock(s) for txUpdateMap() to update bmap
2007 * allocate entries XAD[lwm:next):
2010 /* format a maplock for txUpdateMap() to update bPMAP
2011 * for alloc of new/extended extents of XAD[lwm:next)
2012 * from the page itself;
2013 * txUpdateMap() resets (XAD_NEW|XAD_EXTEND) flag.
2015 tlck
->flag
|= tlckUPDATEMAP
;
2016 xadlock
->flag
= mlckALLOCXADLIST
;
2017 xadlock
->count
= next
- lwm
;
2018 xadlock
->xdlist
= &p
->xad
[lwm
];
2020 jfs_info("xtLog: alloc ip:0x%p mp:0x%p count:%d "
2022 tlck
->ip
, mp
, xadlock
->count
, lwm
, next
);
2028 * truncate entry XAD[twm == next - 1]:
2030 if (twm
== next
- 1) {
2031 struct pxd_lock
*pxdlock
;
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
;
2748 daemonize("jfsCommit");
2750 complete(&jfsIOwait
);
2754 jfs_commit_thread_waking
= 0; /* OK to wake another thread */
2755 while (!list_empty(&TxAnchor
.unlock_queue
)) {
2757 list_for_each_entry(tblk
, &TxAnchor
.unlock_queue
,
2760 sbi
= JFS_SBI(tblk
->sb
);
2762 * For each volume, the transactions must be
2763 * handled in order. If another commit thread
2764 * is handling a tblk for this superblock,
2767 if (sbi
->commit_state
& IN_LAZYCOMMIT
)
2770 sbi
->commit_state
|= IN_LAZYCOMMIT
;
2774 * Remove transaction from queue
2776 list_del(&tblk
->cqueue
);
2782 sbi
->commit_state
&= ~IN_LAZYCOMMIT
;
2784 * Don't continue in the for loop. (We can't
2785 * anyway, it's unsafe!) We want to go back to
2786 * the beginning of the list.
2791 /* If there was nothing to do, don't continue */
2795 /* In case a wakeup came while all threads were active */
2796 jfs_commit_thread_waking
= 0;
2798 if (freezing(current
)) {
2802 DECLARE_WAITQUEUE(wq
, current
);
2804 add_wait_queue(&jfs_commit_thread_wait
, &wq
);
2805 set_current_state(TASK_INTERRUPTIBLE
);
2808 current
->state
= TASK_RUNNING
;
2809 remove_wait_queue(&jfs_commit_thread_wait
, &wq
);
2811 } while (!jfs_stop_threads
);
2813 if (!list_empty(&TxAnchor
.unlock_queue
))
2814 jfs_err("jfs_lazycommit being killed w/pending transactions!");
2816 jfs_info("jfs_lazycommit being killed\n");
2817 complete_and_exit(&jfsIOwait
, 0);
2820 void txLazyUnlock(struct tblock
* tblk
)
2822 unsigned long flags
;
2826 list_add_tail(&tblk
->cqueue
, &TxAnchor
.unlock_queue
);
2828 * Don't wake up a commit thread if there is already one servicing
2829 * this superblock, or if the last one we woke up hasn't started yet.
2831 if (!(JFS_SBI(tblk
->sb
)->commit_state
& IN_LAZYCOMMIT
) &&
2832 !jfs_commit_thread_waking
) {
2833 jfs_commit_thread_waking
= 1;
2834 wake_up(&jfs_commit_thread_wait
);
2839 static void LogSyncRelease(struct metapage
* mp
)
2841 struct jfs_log
*log
= mp
->log
;
2843 assert(mp
->nohomeok
);
2845 metapage_homeok(mp
);
2851 * Block all new transactions and push anonymous transactions to
2854 * This does almost the same thing as jfs_sync below. We don't
2855 * worry about deadlocking when jfs_tlocks_low is set, since we would
2856 * expect jfs_sync to get us out of that jam.
2858 void txQuiesce(struct super_block
*sb
)
2861 struct jfs_inode_info
*jfs_ip
;
2862 struct jfs_log
*log
= JFS_SBI(sb
)->log
;
2865 set_bit(log_QUIESCE
, &log
->flag
);
2869 while (!list_empty(&TxAnchor
.anon_list
)) {
2870 jfs_ip
= list_entry(TxAnchor
.anon_list
.next
,
2871 struct jfs_inode_info
,
2873 ip
= &jfs_ip
->vfs_inode
;
2876 * inode will be removed from anonymous list
2877 * when it is committed
2880 tid
= txBegin(ip
->i_sb
, COMMIT_INODE
| COMMIT_FORCE
);
2881 down(&jfs_ip
->commit_sem
);
2882 txCommit(tid
, 1, &ip
, 0);
2884 up(&jfs_ip
->commit_sem
);
2886 * Just to be safe. I don't know how
2887 * long we can run without blocking
2894 * If jfs_sync is running in parallel, there could be some inodes
2895 * on anon_list2. Let's check.
2897 if (!list_empty(&TxAnchor
.anon_list2
)) {
2898 list_splice(&TxAnchor
.anon_list2
, &TxAnchor
.anon_list
);
2899 INIT_LIST_HEAD(&TxAnchor
.anon_list2
);
2905 * We may need to kick off the group commit
2907 jfs_flush_journal(log
, 0);
2913 * Allows transactions to start again following txQuiesce
2915 void txResume(struct super_block
*sb
)
2917 struct jfs_log
*log
= JFS_SBI(sb
)->log
;
2919 clear_bit(log_QUIESCE
, &log
->flag
);
2920 TXN_WAKEUP(&log
->syncwait
);
2926 * To be run as a kernel daemon. This is awakened when tlocks run low.
2927 * We write any inodes that have anonymous tlocks so they will become
2930 int jfs_sync(void *arg
)
2933 struct jfs_inode_info
*jfs_ip
;
2937 daemonize("jfsSync");
2939 complete(&jfsIOwait
);
2943 * write each inode on the anonymous inode list
2946 while (jfs_tlocks_low
&& !list_empty(&TxAnchor
.anon_list
)) {
2947 jfs_ip
= list_entry(TxAnchor
.anon_list
.next
,
2948 struct jfs_inode_info
,
2950 ip
= &jfs_ip
->vfs_inode
;
2954 * Inode is being freed
2956 list_del_init(&jfs_ip
->anon_inode_list
);
2957 } else if (! down_trylock(&jfs_ip
->commit_sem
)) {
2959 * inode will be removed from anonymous list
2960 * when it is committed
2963 tid
= txBegin(ip
->i_sb
, COMMIT_INODE
);
2964 rc
= txCommit(tid
, 1, &ip
, 0);
2966 up(&jfs_ip
->commit_sem
);
2970 * Just to be safe. I don't know how
2971 * long we can run without blocking
2976 /* We can't get the commit semaphore. It may
2977 * be held by a thread waiting for tlock's
2978 * so let's not block here. Save it to
2979 * put back on the anon_list.
2982 /* Take off anon_list */
2983 list_del(&jfs_ip
->anon_inode_list
);
2985 /* Put on anon_list2 */
2986 list_add(&jfs_ip
->anon_inode_list
,
2987 &TxAnchor
.anon_list2
);
2994 /* Add anon_list2 back to anon_list */
2995 list_splice_init(&TxAnchor
.anon_list2
, &TxAnchor
.anon_list
);
2997 if (freezing(current
)) {
3001 DECLARE_WAITQUEUE(wq
, current
);
3003 add_wait_queue(&jfs_sync_thread_wait
, &wq
);
3004 set_current_state(TASK_INTERRUPTIBLE
);
3007 current
->state
= TASK_RUNNING
;
3008 remove_wait_queue(&jfs_sync_thread_wait
, &wq
);
3010 } while (!jfs_stop_threads
);
3012 jfs_info("jfs_sync being killed");
3013 complete_and_exit(&jfsIOwait
, 0);
3016 #if defined(CONFIG_PROC_FS) && defined(CONFIG_JFS_DEBUG)
3017 int jfs_txanchor_read(char *buffer
, char **start
, off_t offset
, int length
,
3018 int *eof
, void *data
)
3027 waitqueue_active(&TxAnchor
.freewait
) ? "active" : "empty";
3029 waitqueue_active(&TxAnchor
.freelockwait
) ? "active" : "empty";
3031 waitqueue_active(&TxAnchor
.lowlockwait
) ? "active" : "empty";
3033 len
+= sprintf(buffer
,
3039 "freelockwait = %s\n"
3040 "lowlockwait = %s\n"
3041 "tlocksInUse = %d\n"
3042 "jfs_tlocks_low = %d\n"
3043 "unlock_queue is %sempty\n",
3049 TxAnchor
.tlocksInUse
,
3051 list_empty(&TxAnchor
.unlock_queue
) ? "" : "not ");
3054 *start
= buffer
+ begin
;
3069 #if defined(CONFIG_PROC_FS) && defined(CONFIG_JFS_STATISTICS)
3070 int jfs_txstats_read(char *buffer
, char **start
, off_t offset
, int length
,
3071 int *eof
, void *data
)
3076 len
+= sprintf(buffer
,
3079 "calls to txBegin = %d\n"
3080 "txBegin blocked by sync barrier = %d\n"
3081 "txBegin blocked by tlocks low = %d\n"
3082 "txBegin blocked by no free tid = %d\n"
3083 "calls to txBeginAnon = %d\n"
3084 "txBeginAnon blocked by sync barrier = %d\n"
3085 "txBeginAnon blocked by tlocks low = %d\n"
3086 "calls to txLockAlloc = %d\n"
3087 "tLockAlloc blocked by no free lock = %d\n",
3089 TxStat
.txBegin_barrier
,
3090 TxStat
.txBegin_lockslow
,
3091 TxStat
.txBegin_freetid
,
3093 TxStat
.txBeginAnon_barrier
,
3094 TxStat
.txBeginAnon_lockslow
,
3096 TxStat
.txLockAlloc_freelock
);
3099 *start
= buffer
+ begin
;