2 * Copyright (C) International Business Machines Corp., 2000-2005
3 * Portions Copyright (C) Christoph Hellwig, 2001-2002
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
13 * the GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
21 * jfs_txnmgr.c: transaction manager
24 * transaction starts with txBegin() and ends with txCommit()
27 * tlock is acquired at the time of update;
28 * (obviate scan at commit time for xtree and dtree)
29 * tlock and mp points to each other;
30 * (no hashlist for mp -> tlock).
33 * tlock on in-memory inode:
34 * in-place tlock in the in-memory inode itself;
35 * converted to page lock by iWrite() at commit time.
37 * tlock during write()/mmap() under anonymous transaction (tid = 0):
38 * transferred (?) to transaction at commit time.
40 * use the page itself to update allocation maps
41 * (obviate intermediate replication of allocation/deallocation data)
42 * hold on to mp+lock thru update of maps
46 #include <linux/vmalloc.h>
47 #include <linux/completion.h>
48 #include <linux/freezer.h>
49 #include <linux/module.h>
50 #include <linux/moduleparam.h>
51 #include <linux/kthread.h>
52 #include "jfs_incore.h"
53 #include "jfs_inode.h"
54 #include "jfs_filsys.h"
55 #include "jfs_metapage.h"
56 #include "jfs_dinode.h"
59 #include "jfs_superblock.h"
60 #include "jfs_debug.h"
63 * transaction management structures
66 int freetid
; /* index of a free tid structure */
67 int freelock
; /* index first free lock word */
68 wait_queue_head_t freewait
; /* eventlist of free tblock */
69 wait_queue_head_t freelockwait
; /* eventlist of free tlock */
70 wait_queue_head_t lowlockwait
; /* eventlist of ample tlocks */
71 int tlocksInUse
; /* Number of tlocks in use */
72 spinlock_t LazyLock
; /* synchronize sync_queue & unlock_queue */
73 /* struct tblock *sync_queue; * Transactions waiting for data sync */
74 struct list_head unlock_queue
; /* Txns waiting to be released */
75 struct list_head anon_list
; /* inodes having anonymous txns */
76 struct list_head anon_list2
; /* inodes having anonymous txns
77 that couldn't be sync'ed */
80 int jfs_tlocks_low
; /* Indicates low number of available tlocks */
82 #ifdef CONFIG_JFS_STATISTICS
86 uint txBegin_lockslow
;
89 uint txBeginAnon_barrier
;
90 uint txBeginAnon_lockslow
;
92 uint txLockAlloc_freelock
;
96 static int nTxBlock
= -1; /* number of transaction blocks */
97 module_param(nTxBlock
, int, 0);
98 MODULE_PARM_DESC(nTxBlock
,
99 "Number of transaction blocks (max:65536)");
101 static int nTxLock
= -1; /* number of transaction locks */
102 module_param(nTxLock
, int, 0);
103 MODULE_PARM_DESC(nTxLock
,
104 "Number of transaction locks (max:65536)");
106 struct tblock
*TxBlock
; /* transaction block table */
107 static int TxLockLWM
; /* Low water mark for number of txLocks used */
108 static int TxLockHWM
; /* High water mark for number of txLocks used */
109 static int TxLockVHWM
; /* Very High water mark */
110 struct tlock
*TxLock
; /* transaction lock table */
113 * transaction management lock
115 static DEFINE_SPINLOCK(jfsTxnLock
);
117 #define TXN_LOCK() spin_lock(&jfsTxnLock)
118 #define TXN_UNLOCK() spin_unlock(&jfsTxnLock)
120 #define LAZY_LOCK_INIT() spin_lock_init(&TxAnchor.LazyLock);
121 #define LAZY_LOCK(flags) spin_lock_irqsave(&TxAnchor.LazyLock, flags)
122 #define LAZY_UNLOCK(flags) spin_unlock_irqrestore(&TxAnchor.LazyLock, flags)
124 static DECLARE_WAIT_QUEUE_HEAD(jfs_commit_thread_wait
);
125 static int jfs_commit_thread_waking
;
128 * Retry logic exist outside these macros to protect from spurrious wakeups.
130 static inline void TXN_SLEEP_DROP_LOCK(wait_queue_head_t
* event
)
132 DECLARE_WAITQUEUE(wait
, current
);
134 add_wait_queue(event
, &wait
);
135 set_current_state(TASK_UNINTERRUPTIBLE
);
138 __set_current_state(TASK_RUNNING
);
139 remove_wait_queue(event
, &wait
);
142 #define TXN_SLEEP(event)\
144 TXN_SLEEP_DROP_LOCK(event);\
148 #define TXN_WAKEUP(event) wake_up_all(event)
154 tid_t maxtid
; /* 4: biggest tid ever used */
155 lid_t maxlid
; /* 4: biggest lid ever used */
156 int ntid
; /* 4: # of transactions performed */
157 int nlid
; /* 4: # of tlocks acquired */
158 int waitlock
; /* 4: # of tlock wait */
164 static int diLog(struct jfs_log
* log
, struct tblock
* tblk
, struct lrd
* lrd
,
165 struct tlock
* tlck
, struct commit
* cd
);
166 static int dataLog(struct jfs_log
* log
, struct tblock
* tblk
, struct lrd
* lrd
,
167 struct tlock
* tlck
);
168 static void dtLog(struct jfs_log
* log
, struct tblock
* tblk
, struct lrd
* lrd
,
169 struct tlock
* tlck
);
170 static void mapLog(struct jfs_log
* log
, struct tblock
* tblk
, struct lrd
* lrd
,
171 struct tlock
* tlck
);
172 static void txAllocPMap(struct inode
*ip
, struct maplock
* maplock
,
173 struct tblock
* tblk
);
174 static void txForce(struct tblock
* tblk
);
175 static int txLog(struct jfs_log
* log
, struct tblock
* tblk
,
177 static void txUpdateMap(struct tblock
* tblk
);
178 static void txRelease(struct tblock
* tblk
);
179 static void xtLog(struct jfs_log
* log
, struct tblock
* tblk
, struct lrd
* lrd
,
180 struct tlock
* tlck
);
181 static void LogSyncRelease(struct metapage
* mp
);
184 * transaction block/lock management
185 * ---------------------------------
189 * Get a transaction lock from the free list. If the number in use is
190 * greater than the high water mark, wake up the sync daemon. This should
191 * free some anonymous transaction locks. (TXN_LOCK must be held.)
193 static lid_t
txLockAlloc(void)
197 INCREMENT(TxStat
.txLockAlloc
);
198 if (!TxAnchor
.freelock
) {
199 INCREMENT(TxStat
.txLockAlloc_freelock
);
202 while (!(lid
= TxAnchor
.freelock
))
203 TXN_SLEEP(&TxAnchor
.freelockwait
);
204 TxAnchor
.freelock
= TxLock
[lid
].next
;
205 HIGHWATERMARK(stattx
.maxlid
, lid
);
206 if ((++TxAnchor
.tlocksInUse
> TxLockHWM
) && (jfs_tlocks_low
== 0)) {
207 jfs_info("txLockAlloc tlocks low");
209 wake_up_process(jfsSyncThread
);
215 static void txLockFree(lid_t lid
)
218 TxLock
[lid
].next
= TxAnchor
.freelock
;
219 TxAnchor
.freelock
= lid
;
220 TxAnchor
.tlocksInUse
--;
221 if (jfs_tlocks_low
&& (TxAnchor
.tlocksInUse
< TxLockLWM
)) {
222 jfs_info("txLockFree jfs_tlocks_low no more");
224 TXN_WAKEUP(&TxAnchor
.lowlockwait
);
226 TXN_WAKEUP(&TxAnchor
.freelockwait
);
232 * FUNCTION: initialize transaction management structures
236 * serialization: single thread at jfs_init()
243 /* Set defaults for nTxLock and nTxBlock if unset */
246 if (nTxBlock
== -1) {
247 /* Base default on memory size */
249 if (si
.totalram
> (256 * 1024)) /* 1 GB */
252 nTxLock
= si
.totalram
>> 2;
253 } else if (nTxBlock
> (8 * 1024))
256 nTxLock
= nTxBlock
<< 3;
259 nTxBlock
= nTxLock
>> 3;
261 /* Verify tunable parameters */
263 nTxBlock
= 16; /* No one should set it this low */
264 if (nTxBlock
> 65536)
267 nTxLock
= 256; /* No one should set it this low */
271 printk(KERN_INFO
"JFS: nTxBlock = %d, nTxLock = %d\n",
274 * initialize transaction block (tblock) table
276 * transaction id (tid) = tblock index
277 * tid = 0 is reserved.
279 TxLockLWM
= (nTxLock
* 4) / 10;
280 TxLockHWM
= (nTxLock
* 7) / 10;
281 TxLockVHWM
= (nTxLock
* 8) / 10;
283 size
= sizeof(struct tblock
) * nTxBlock
;
284 TxBlock
= vmalloc(size
);
288 for (k
= 1; k
< nTxBlock
- 1; k
++) {
289 TxBlock
[k
].next
= k
+ 1;
290 init_waitqueue_head(&TxBlock
[k
].gcwait
);
291 init_waitqueue_head(&TxBlock
[k
].waitor
);
294 init_waitqueue_head(&TxBlock
[k
].gcwait
);
295 init_waitqueue_head(&TxBlock
[k
].waitor
);
297 TxAnchor
.freetid
= 1;
298 init_waitqueue_head(&TxAnchor
.freewait
);
300 stattx
.maxtid
= 1; /* statistics */
303 * initialize transaction lock (tlock) table
305 * transaction lock id = tlock index
306 * tlock id = 0 is reserved.
308 size
= sizeof(struct tlock
) * nTxLock
;
309 TxLock
= vmalloc(size
);
310 if (TxLock
== NULL
) {
315 /* initialize tlock table */
316 for (k
= 1; k
< nTxLock
- 1; k
++)
317 TxLock
[k
].next
= k
+ 1;
319 init_waitqueue_head(&TxAnchor
.freelockwait
);
320 init_waitqueue_head(&TxAnchor
.lowlockwait
);
322 TxAnchor
.freelock
= 1;
323 TxAnchor
.tlocksInUse
= 0;
324 INIT_LIST_HEAD(&TxAnchor
.anon_list
);
325 INIT_LIST_HEAD(&TxAnchor
.anon_list2
);
328 INIT_LIST_HEAD(&TxAnchor
.unlock_queue
);
330 stattx
.maxlid
= 1; /* statistics */
338 * FUNCTION: clean up when module is unloaded
351 * FUNCTION: start a transaction.
353 * PARAMETER: sb - superblock
354 * flag - force for nested tx;
356 * RETURN: tid - transaction id
358 * note: flag force allows to start tx for nested tx
359 * to prevent deadlock on logsync barrier;
361 tid_t
txBegin(struct super_block
*sb
, int flag
)
367 jfs_info("txBegin: flag = 0x%x", flag
);
368 log
= JFS_SBI(sb
)->log
;
372 INCREMENT(TxStat
.txBegin
);
375 if (!(flag
& COMMIT_FORCE
)) {
377 * synchronize with logsync barrier
379 if (test_bit(log_SYNCBARRIER
, &log
->flag
) ||
380 test_bit(log_QUIESCE
, &log
->flag
)) {
381 INCREMENT(TxStat
.txBegin_barrier
);
382 TXN_SLEEP(&log
->syncwait
);
388 * Don't begin transaction if we're getting starved for tlocks
389 * unless COMMIT_FORCE or COMMIT_INODE (which may ultimately
392 if (TxAnchor
.tlocksInUse
> TxLockVHWM
) {
393 INCREMENT(TxStat
.txBegin_lockslow
);
394 TXN_SLEEP(&TxAnchor
.lowlockwait
);
400 * allocate transaction id/block
402 if ((t
= TxAnchor
.freetid
) == 0) {
403 jfs_info("txBegin: waiting for free tid");
404 INCREMENT(TxStat
.txBegin_freetid
);
405 TXN_SLEEP(&TxAnchor
.freewait
);
409 tblk
= tid_to_tblock(t
);
411 if ((tblk
->next
== 0) && !(flag
& COMMIT_FORCE
)) {
412 /* Don't let a non-forced transaction take the last tblk */
413 jfs_info("txBegin: waiting for free tid");
414 INCREMENT(TxStat
.txBegin_freetid
);
415 TXN_SLEEP(&TxAnchor
.freewait
);
419 TxAnchor
.freetid
= tblk
->next
;
422 * initialize transaction
426 * We can't zero the whole thing or we screw up another thread being
427 * awakened after sleeping on tblk->waitor
429 * memset(tblk, 0, sizeof(struct tblock));
431 tblk
->next
= tblk
->last
= tblk
->xflag
= tblk
->flag
= tblk
->lsn
= 0;
435 tblk
->logtid
= log
->logtid
;
439 HIGHWATERMARK(stattx
.maxtid
, t
); /* statistics */
440 INCREMENT(stattx
.ntid
); /* statistics */
444 jfs_info("txBegin: returning tid = %d", t
);
450 * NAME: txBeginAnon()
452 * FUNCTION: start an anonymous transaction.
453 * Blocks if logsync or available tlocks are low to prevent
454 * anonymous tlocks from depleting supply.
456 * PARAMETER: sb - superblock
460 void txBeginAnon(struct super_block
*sb
)
464 log
= JFS_SBI(sb
)->log
;
467 INCREMENT(TxStat
.txBeginAnon
);
471 * synchronize with logsync barrier
473 if (test_bit(log_SYNCBARRIER
, &log
->flag
) ||
474 test_bit(log_QUIESCE
, &log
->flag
)) {
475 INCREMENT(TxStat
.txBeginAnon_barrier
);
476 TXN_SLEEP(&log
->syncwait
);
481 * Don't begin transaction if we're getting starved for tlocks
483 if (TxAnchor
.tlocksInUse
> TxLockVHWM
) {
484 INCREMENT(TxStat
.txBeginAnon_lockslow
);
485 TXN_SLEEP(&TxAnchor
.lowlockwait
);
494 * function: free specified transaction block.
496 * logsync barrier processing:
500 void txEnd(tid_t tid
)
502 struct tblock
*tblk
= tid_to_tblock(tid
);
505 jfs_info("txEnd: tid = %d", tid
);
509 * wakeup transactions waiting on the page locked
510 * by the current transaction
512 TXN_WAKEUP(&tblk
->waitor
);
514 log
= JFS_SBI(tblk
->sb
)->log
;
517 * Lazy commit thread can't free this guy until we mark it UNLOCKED,
518 * otherwise, we would be left with a transaction that may have been
521 * Lazy commit thread will turn off tblkGC_LAZY before calling this
524 if (tblk
->flag
& tblkGC_LAZY
) {
525 jfs_info("txEnd called w/lazy tid: %d, tblk = 0x%p", tid
, tblk
);
528 spin_lock_irq(&log
->gclock
); // LOGGC_LOCK
529 tblk
->flag
|= tblkGC_UNLOCKED
;
530 spin_unlock_irq(&log
->gclock
); // LOGGC_UNLOCK
534 jfs_info("txEnd: tid: %d, tblk = 0x%p", tid
, tblk
);
536 assert(tblk
->next
== 0);
539 * insert tblock back on freelist
541 tblk
->next
= TxAnchor
.freetid
;
542 TxAnchor
.freetid
= tid
;
545 * mark the tblock not active
547 if (--log
->active
== 0) {
548 clear_bit(log_FLUSH
, &log
->flag
);
551 * synchronize with logsync barrier
553 if (test_bit(log_SYNCBARRIER
, &log
->flag
)) {
556 /* write dirty metadata & forward log syncpt */
559 jfs_info("log barrier off: 0x%x", log
->lsn
);
561 /* enable new transactions start */
562 clear_bit(log_SYNCBARRIER
, &log
->flag
);
564 /* wakeup all waitors for logsync barrier */
565 TXN_WAKEUP(&log
->syncwait
);
574 * wakeup all waitors for a free tblock
576 TXN_WAKEUP(&TxAnchor
.freewait
);
582 * function: acquire a transaction lock on the specified <mp>
586 * return: transaction lock id
590 struct tlock
*txLock(tid_t tid
, struct inode
*ip
, struct metapage
* mp
,
593 struct jfs_inode_info
*jfs_ip
= JFS_IP(ip
);
598 struct xtlock
*xtlck
;
599 struct linelock
*linelock
;
605 if (S_ISDIR(ip
->i_mode
) && (type
& tlckXTREE
) &&
606 !(mp
->xflag
& COMMIT_PAGE
)) {
608 * Directory inode is special. It can have both an xtree tlock
609 * and a dtree tlock associated with it.
616 /* is page not locked by a transaction ? */
620 jfs_info("txLock: tid:%d ip:0x%p mp:0x%p lid:%d", tid
, ip
, mp
, lid
);
622 /* is page locked by the requester transaction ? */
623 tlck
= lid_to_tlock(lid
);
624 if ((xtid
= tlck
->tid
) == tid
) {
630 * is page locked by anonymous transaction/lock ?
632 * (page update without transaction (i.e., file write) is
633 * locked under anonymous transaction tid = 0:
634 * anonymous tlocks maintained on anonymous tlock list of
635 * the inode of the page and available to all anonymous
636 * transactions until txCommit() time at which point
637 * they are transferred to the transaction tlock list of
638 * the commiting transaction of the inode)
643 tblk
= tid_to_tblock(tid
);
645 * The order of the tlocks in the transaction is important
646 * (during truncate, child xtree pages must be freed before
647 * parent's tlocks change the working map).
648 * Take tlock off anonymous list and add to tail of
651 * Note: We really need to get rid of the tid & lid and
652 * use list_head's. This code is getting UGLY!
654 if (jfs_ip
->atlhead
== lid
) {
655 if (jfs_ip
->atltail
== lid
) {
656 /* only anonymous txn.
657 * Remove from anon_list
660 list_del_init(&jfs_ip
->anon_inode_list
);
663 jfs_ip
->atlhead
= tlck
->next
;
666 for (last
= jfs_ip
->atlhead
;
667 lid_to_tlock(last
)->next
!= lid
;
668 last
= lid_to_tlock(last
)->next
) {
671 lid_to_tlock(last
)->next
= tlck
->next
;
672 if (jfs_ip
->atltail
== lid
)
673 jfs_ip
->atltail
= last
;
676 /* insert the tlock at tail of transaction tlock list */
679 lid_to_tlock(tblk
->last
)->next
= lid
;
695 tlck
= lid_to_tlock(lid
);
704 /* mark tlock for meta-data page */
705 if (mp
->xflag
& COMMIT_PAGE
) {
707 tlck
->flag
= tlckPAGELOCK
;
709 /* mark the page dirty and nohomeok */
710 metapage_nohomeok(mp
);
712 jfs_info("locking mp = 0x%p, nohomeok = %d tid = %d tlck = 0x%p",
713 mp
, mp
->nohomeok
, tid
, tlck
);
715 /* if anonymous transaction, and buffer is on the group
716 * commit synclist, mark inode to show this. This will
717 * prevent the buffer from being marked nohomeok for too
720 if ((tid
== 0) && mp
->lsn
)
721 set_cflag(COMMIT_Synclist
, ip
);
723 /* mark tlock for in-memory inode */
725 tlck
->flag
= tlckINODELOCK
;
727 if (S_ISDIR(ip
->i_mode
))
728 tlck
->flag
|= tlckDIRECTORY
;
732 /* bind the tlock and the page */
741 * enqueue transaction lock to transaction/inode
743 /* insert the tlock at tail of transaction tlock list */
745 tblk
= tid_to_tblock(tid
);
747 lid_to_tlock(tblk
->last
)->next
= lid
;
753 /* anonymous transaction:
754 * insert the tlock at head of inode anonymous tlock list
757 tlck
->next
= jfs_ip
->atlhead
;
758 jfs_ip
->atlhead
= lid
;
759 if (tlck
->next
== 0) {
760 /* This inode's first anonymous transaction */
761 jfs_ip
->atltail
= lid
;
763 list_add_tail(&jfs_ip
->anon_inode_list
,
764 &TxAnchor
.anon_list
);
769 /* initialize type dependent area for linelock */
770 linelock
= (struct linelock
*) & tlck
->lock
;
772 linelock
->flag
= tlckLINELOCK
;
773 linelock
->maxcnt
= TLOCKSHORT
;
776 switch (type
& tlckTYPE
) {
778 linelock
->l2linesize
= L2DTSLOTSIZE
;
782 linelock
->l2linesize
= L2XTSLOTSIZE
;
784 xtlck
= (struct xtlock
*) linelock
;
785 xtlck
->header
.offset
= 0;
786 xtlck
->header
.length
= 2;
788 if (type
& tlckNEW
) {
789 xtlck
->lwm
.offset
= XTENTRYSTART
;
791 if (mp
->xflag
& COMMIT_PAGE
)
792 p
= (xtpage_t
*) mp
->data
;
794 p
= &jfs_ip
->i_xtroot
;
796 le16_to_cpu(p
->header
.nextindex
);
798 xtlck
->lwm
.length
= 0; /* ! */
799 xtlck
->twm
.offset
= 0;
800 xtlck
->hwm
.offset
= 0;
806 linelock
->l2linesize
= L2INODESLOTSIZE
;
810 linelock
->l2linesize
= L2DATASLOTSIZE
;
814 jfs_err("UFO tlock:0x%p", tlck
);
818 * update tlock vector
826 * page is being locked by another transaction:
829 /* Only locks on ipimap or ipaimap should reach here */
830 /* assert(jfs_ip->fileset == AGGREGATE_I); */
831 if (jfs_ip
->fileset
!= AGGREGATE_I
) {
832 jfs_err("txLock: trying to lock locked page!");
833 dump_mem("ip", ip
, sizeof(struct inode
));
834 dump_mem("mp", mp
, sizeof(struct metapage
));
835 dump_mem("Locker's tblk", tid_to_tblock(tid
),
836 sizeof(struct tblock
));
837 dump_mem("Tlock", tlck
, sizeof(struct tlock
));
840 INCREMENT(stattx
.waitlock
); /* statistics */
842 release_metapage(mp
);
844 xtid
= tlck
->tid
; /* reacquire after dropping TXN_LOCK */
846 jfs_info("txLock: in waitLock, tid = %d, xtid = %d, lid = %d",
849 /* Recheck everything since dropping TXN_LOCK */
850 if (xtid
&& (tlck
->mp
== mp
) && (mp
->lid
== lid
))
851 TXN_SLEEP_DROP_LOCK(&tid_to_tblock(xtid
)->waitor
);
854 jfs_info("txLock: awakened tid = %d, lid = %d", tid
, lid
);
862 * FUNCTION: Release buffers associated with transaction locks, but don't
863 * mark homeok yet. The allows other transactions to modify
864 * buffers, but won't let them go to disk until commit record
865 * actually gets written.
870 * RETURN: Errors from subroutines.
872 static void txRelease(struct tblock
* tblk
)
880 for (lid
= tblk
->next
; lid
; lid
= tlck
->next
) {
881 tlck
= lid_to_tlock(lid
);
882 if ((mp
= tlck
->mp
) != NULL
&&
883 (tlck
->type
& tlckBTROOT
) == 0) {
884 assert(mp
->xflag
& COMMIT_PAGE
);
890 * wakeup transactions waiting on a page locked
891 * by the current transaction
893 TXN_WAKEUP(&tblk
->waitor
);
901 * FUNCTION: Initiates pageout of pages modified by tid in journalled
902 * objects and frees their lockwords.
904 static void txUnlock(struct tblock
* tblk
)
907 struct linelock
*linelock
;
908 lid_t lid
, next
, llid
, k
;
914 jfs_info("txUnlock: tblk = 0x%p", tblk
);
915 log
= JFS_SBI(tblk
->sb
)->log
;
918 * mark page under tlock homeok (its log has been written):
920 for (lid
= tblk
->next
; lid
; lid
= next
) {
921 tlck
= lid_to_tlock(lid
);
924 jfs_info("unlocking lid = %d, tlck = 0x%p", lid
, tlck
);
926 /* unbind page from tlock */
927 if ((mp
= tlck
->mp
) != NULL
&&
928 (tlck
->type
& tlckBTROOT
) == 0) {
929 assert(mp
->xflag
& COMMIT_PAGE
);
935 assert(mp
->nohomeok
> 0);
936 _metapage_homeok(mp
);
938 /* inherit younger/larger clsn */
939 LOGSYNC_LOCK(log
, flags
);
941 logdiff(difft
, tblk
->clsn
, log
);
942 logdiff(diffp
, mp
->clsn
, log
);
944 mp
->clsn
= tblk
->clsn
;
946 mp
->clsn
= tblk
->clsn
;
947 LOGSYNC_UNLOCK(log
, flags
);
949 assert(!(tlck
->flag
& tlckFREEPAGE
));
954 /* insert tlock, and linelock(s) of the tlock if any,
955 * at head of freelist
959 llid
= ((struct linelock
*) & tlck
->lock
)->next
;
961 linelock
= (struct linelock
*) lid_to_tlock(llid
);
970 tblk
->next
= tblk
->last
= 0;
973 * remove tblock from logsynclist
974 * (allocation map pages inherited lsn of tblk and
975 * has been inserted in logsync list at txUpdateMap())
978 LOGSYNC_LOCK(log
, flags
);
980 list_del(&tblk
->synclist
);
981 LOGSYNC_UNLOCK(log
, flags
);
988 * function: allocate a transaction lock for freed page/entry;
989 * for freed page, maplock is used as xtlock/dtlock type;
991 struct tlock
*txMaplock(tid_t tid
, struct inode
*ip
, int type
)
993 struct jfs_inode_info
*jfs_ip
= JFS_IP(ip
);
997 struct maplock
*maplock
;
1004 lid
= txLockAlloc();
1005 tlck
= lid_to_tlock(lid
);
1012 /* bind the tlock and the object */
1013 tlck
->flag
= tlckINODELOCK
;
1014 if (S_ISDIR(ip
->i_mode
))
1015 tlck
->flag
|= tlckDIRECTORY
;
1022 * enqueue transaction lock to transaction/inode
1024 /* insert the tlock at tail of transaction tlock list */
1026 tblk
= tid_to_tblock(tid
);
1028 lid_to_tlock(tblk
->last
)->next
= lid
;
1034 /* anonymous transaction:
1035 * insert the tlock at head of inode anonymous tlock list
1038 tlck
->next
= jfs_ip
->atlhead
;
1039 jfs_ip
->atlhead
= lid
;
1040 if (tlck
->next
== 0) {
1041 /* This inode's first anonymous transaction */
1042 jfs_ip
->atltail
= lid
;
1043 list_add_tail(&jfs_ip
->anon_inode_list
,
1044 &TxAnchor
.anon_list
);
1050 /* initialize type dependent area for maplock */
1051 maplock
= (struct maplock
*) & tlck
->lock
;
1053 maplock
->maxcnt
= 0;
1062 * function: allocate a transaction lock for log vector list
1064 struct linelock
*txLinelock(struct linelock
* tlock
)
1068 struct linelock
*linelock
;
1072 /* allocate a TxLock structure */
1073 lid
= txLockAlloc();
1074 tlck
= lid_to_tlock(lid
);
1078 /* initialize linelock */
1079 linelock
= (struct linelock
*) tlck
;
1081 linelock
->flag
= tlckLINELOCK
;
1082 linelock
->maxcnt
= TLOCKLONG
;
1083 linelock
->index
= 0;
1084 if (tlck
->flag
& tlckDIRECTORY
)
1085 linelock
->flag
|= tlckDIRECTORY
;
1087 /* append linelock after tlock */
1088 linelock
->next
= tlock
->next
;
1095 * transaction commit management
1096 * -----------------------------
1102 * FUNCTION: commit the changes to the objects specified in
1103 * clist. For journalled segments only the
1104 * changes of the caller are committed, ie by tid.
1105 * for non-journalled segments the data are flushed to
1106 * disk and then the change to the disk inode and indirect
1107 * blocks committed (so blocks newly allocated to the
1108 * segment will be made a part of the segment atomically).
1110 * all of the segments specified in clist must be in
1111 * one file system. no more than 6 segments are needed
1112 * to handle all unix svcs.
1114 * if the i_nlink field (i.e. disk inode link count)
1115 * is zero, and the type of inode is a regular file or
1116 * directory, or symbolic link , the inode is truncated
1117 * to zero length. the truncation is committed but the
1118 * VM resources are unaffected until it is closed (see
1126 * on entry the inode lock on each segment is assumed
1131 int txCommit(tid_t tid
, /* transaction identifier */
1132 int nip
, /* number of inodes to commit */
1133 struct inode
**iplist
, /* list of inode to commit */
1138 struct jfs_log
*log
;
1139 struct tblock
*tblk
;
1143 struct jfs_inode_info
*jfs_ip
;
1146 struct super_block
*sb
;
1148 jfs_info("txCommit, tid = %d, flag = %d", tid
, flag
);
1149 /* is read-only file system ? */
1150 if (isReadOnly(iplist
[0])) {
1155 sb
= cd
.sb
= iplist
[0]->i_sb
;
1159 tid
= txBegin(sb
, 0);
1160 tblk
= tid_to_tblock(tid
);
1163 * initialize commit structure
1165 log
= JFS_SBI(sb
)->log
;
1168 /* initialize log record descriptor in commit */
1170 lrd
->logtid
= cpu_to_le32(tblk
->logtid
);
1173 tblk
->xflag
|= flag
;
1175 if ((flag
& (COMMIT_FORCE
| COMMIT_SYNC
)) == 0)
1176 tblk
->xflag
|= COMMIT_LAZY
;
1178 * prepare non-journaled objects for commit
1180 * flush data pages of non-journaled file
1181 * to prevent the file getting non-initialized disk blocks
1189 * acquire transaction lock on (on-disk) inodes
1191 * update on-disk inode from in-memory inode
1192 * acquiring transaction locks for AFTER records
1193 * on the on-disk inode of file object
1195 * sort the inodes array by inode number in descending order
1196 * to prevent deadlock when acquiring transaction lock
1197 * of on-disk inodes on multiple on-disk inode pages by
1198 * multiple concurrent transactions
1200 for (k
= 0; k
< cd
.nip
; k
++) {
1201 top
= (cd
.iplist
[k
])->i_ino
;
1202 for (n
= k
+ 1; n
< cd
.nip
; n
++) {
1204 if (ip
->i_ino
> top
) {
1206 cd
.iplist
[n
] = cd
.iplist
[k
];
1212 jfs_ip
= JFS_IP(ip
);
1215 * BUGBUG - This code has temporarily been removed. The
1216 * intent is to ensure that any file data is written before
1217 * the metadata is committed to the journal. This prevents
1218 * uninitialized data from appearing in a file after the
1219 * journal has been replayed. (The uninitialized data
1220 * could be sensitive data removed by another user.)
1222 * The problem now is that we are holding the IWRITELOCK
1223 * on the inode, and calling filemap_fdatawrite on an
1224 * unmapped page will cause a deadlock in jfs_get_block.
1226 * The long term solution is to pare down the use of
1227 * IWRITELOCK. We are currently holding it too long.
1228 * We could also be smarter about which data pages need
1229 * to be written before the transaction is committed and
1230 * when we don't need to worry about it at all.
1232 * if ((!S_ISDIR(ip->i_mode))
1233 * && (tblk->flag & COMMIT_DELETE) == 0)
1234 * filemap_write_and_wait(ip->i_mapping);
1238 * Mark inode as not dirty. It will still be on the dirty
1239 * inode list, but we'll know not to commit it again unless
1240 * it gets marked dirty again
1242 clear_cflag(COMMIT_Dirty
, ip
);
1244 /* inherit anonymous tlock(s) of inode */
1245 if (jfs_ip
->atlhead
) {
1246 lid_to_tlock(jfs_ip
->atltail
)->next
= tblk
->next
;
1247 tblk
->next
= jfs_ip
->atlhead
;
1249 tblk
->last
= jfs_ip
->atltail
;
1250 jfs_ip
->atlhead
= jfs_ip
->atltail
= 0;
1252 list_del_init(&jfs_ip
->anon_inode_list
);
1257 * acquire transaction lock on on-disk inode page
1258 * (become first tlock of the tblk's tlock list)
1260 if (((rc
= diWrite(tid
, ip
))))
1265 * write log records from transaction locks
1267 * txUpdateMap() resets XAD_NEW in XAD.
1269 if ((rc
= txLog(log
, tblk
, &cd
)))
1273 * Ensure that inode isn't reused before
1274 * lazy commit thread finishes processing
1276 if (tblk
->xflag
& COMMIT_DELETE
) {
1277 atomic_inc(&tblk
->u
.ip
->i_count
);
1279 * Avoid a rare deadlock
1281 * If the inode is locked, we may be blocked in
1282 * jfs_commit_inode. If so, we don't want the
1283 * lazy_commit thread doing the last iput() on the inode
1284 * since that may block on the locked inode. Instead,
1285 * commit the transaction synchronously, so the last iput
1286 * will be done by the calling thread (or later)
1288 if (tblk
->u
.ip
->i_state
& I_LOCK
)
1289 tblk
->xflag
&= ~COMMIT_LAZY
;
1292 ASSERT((!(tblk
->xflag
& COMMIT_DELETE
)) ||
1293 ((tblk
->u
.ip
->i_nlink
== 0) &&
1294 !test_cflag(COMMIT_Nolink
, tblk
->u
.ip
)));
1297 * write COMMIT log record
1299 lrd
->type
= cpu_to_le16(LOG_COMMIT
);
1301 lsn
= lmLog(log
, tblk
, lrd
, NULL
);
1303 lmGroupCommit(log
, tblk
);
1306 * - transaction is now committed -
1310 * force pages in careful update
1311 * (imap addressing structure update)
1313 if (flag
& COMMIT_FORCE
)
1317 * update allocation map.
1319 * update inode allocation map and inode:
1320 * free pager lock on memory object of inode if any.
1321 * update block allocation map.
1323 * txUpdateMap() resets XAD_NEW in XAD.
1325 if (tblk
->xflag
& COMMIT_FORCE
)
1329 * free transaction locks and pageout/free pages
1333 if ((tblk
->flag
& tblkGC_LAZY
) == 0)
1338 * reset in-memory object state
1340 for (k
= 0; k
< cd
.nip
; k
++) {
1342 jfs_ip
= JFS_IP(ip
);
1345 * reset in-memory inode state
1356 jfs_info("txCommit: tid = %d, returning %d", tid
, rc
);
1363 * FUNCTION: Writes AFTER log records for all lines modified
1364 * by tid for segments specified by inodes in comdata.
1365 * Code assumes only WRITELOCKS are recorded in lockwords.
1371 static int txLog(struct jfs_log
* log
, struct tblock
* tblk
, struct commit
* cd
)
1377 struct lrd
*lrd
= &cd
->lrd
;
1380 * write log record(s) for each tlock of transaction,
1382 for (lid
= tblk
->next
; lid
; lid
= tlck
->next
) {
1383 tlck
= lid_to_tlock(lid
);
1385 tlck
->flag
|= tlckLOG
;
1387 /* initialize lrd common */
1389 lrd
->aggregate
= cpu_to_le32(JFS_SBI(ip
->i_sb
)->aggregate
);
1390 lrd
->log
.redopage
.fileset
= cpu_to_le32(JFS_IP(ip
)->fileset
);
1391 lrd
->log
.redopage
.inode
= cpu_to_le32(ip
->i_ino
);
1393 /* write log record of page from the tlock */
1394 switch (tlck
->type
& tlckTYPE
) {
1396 xtLog(log
, tblk
, lrd
, tlck
);
1400 dtLog(log
, tblk
, lrd
, tlck
);
1404 diLog(log
, tblk
, lrd
, tlck
, cd
);
1408 mapLog(log
, tblk
, lrd
, tlck
);
1412 dataLog(log
, tblk
, lrd
, tlck
);
1416 jfs_err("UFO tlock:0x%p", tlck
);
1426 * function: log inode tlock and format maplock to update bmap;
1428 static int diLog(struct jfs_log
* log
, struct tblock
* tblk
, struct lrd
* lrd
,
1429 struct tlock
* tlck
, struct commit
* cd
)
1432 struct metapage
*mp
;
1434 struct pxd_lock
*pxdlock
;
1438 /* initialize as REDOPAGE record format */
1439 lrd
->log
.redopage
.type
= cpu_to_le16(LOG_INODE
);
1440 lrd
->log
.redopage
.l2linesize
= cpu_to_le16(L2INODESLOTSIZE
);
1442 pxd
= &lrd
->log
.redopage
.pxd
;
1447 if (tlck
->type
& tlckENTRY
) {
1448 /* log after-image for logredo(): */
1449 lrd
->type
= cpu_to_le16(LOG_REDOPAGE
);
1450 PXDaddress(pxd
, mp
->index
);
1452 mp
->logical_size
>> tblk
->sb
->s_blocksize_bits
);
1453 lrd
->backchain
= cpu_to_le32(lmLog(log
, tblk
, lrd
, tlck
));
1455 /* mark page as homeward bound */
1456 tlck
->flag
|= tlckWRITEPAGE
;
1457 } else if (tlck
->type
& tlckFREE
) {
1461 * (pages of the freed inode extent have been invalidated and
1462 * a maplock for free of the extent has been formatted at
1465 * the tlock had been acquired on the inode allocation map page
1466 * (iag) that specifies the freed extent, even though the map
1467 * page is not itself logged, to prevent pageout of the map
1468 * page before the log;
1471 /* log LOG_NOREDOINOEXT of the freed inode extent for
1472 * logredo() to start NoRedoPage filters, and to update
1473 * imap and bmap for free of the extent;
1475 lrd
->type
= cpu_to_le16(LOG_NOREDOINOEXT
);
1477 * For the LOG_NOREDOINOEXT record, we need
1478 * to pass the IAG number and inode extent
1479 * index (within that IAG) from which the
1480 * the extent being released. These have been
1481 * passed to us in the iplist[1] and iplist[2].
1483 lrd
->log
.noredoinoext
.iagnum
=
1484 cpu_to_le32((u32
) (size_t) cd
->iplist
[1]);
1485 lrd
->log
.noredoinoext
.inoext_idx
=
1486 cpu_to_le32((u32
) (size_t) cd
->iplist
[2]);
1488 pxdlock
= (struct pxd_lock
*) & tlck
->lock
;
1489 *pxd
= pxdlock
->pxd
;
1490 lrd
->backchain
= cpu_to_le32(lmLog(log
, tblk
, lrd
, NULL
));
1493 tlck
->flag
|= tlckUPDATEMAP
;
1495 /* mark page as homeward bound */
1496 tlck
->flag
|= tlckWRITEPAGE
;
1498 jfs_err("diLog: UFO type tlck:0x%p", tlck
);
1501 * alloc/free external EA extent
1503 * a maplock for txUpdateMap() to update bPWMAP for alloc/free
1504 * of the extent has been formatted at txLock() time;
1507 assert(tlck
->type
& tlckEA
);
1509 /* log LOG_UPDATEMAP for logredo() to update bmap for
1510 * alloc of new (and free of old) external EA extent;
1512 lrd
->type
= cpu_to_le16(LOG_UPDATEMAP
);
1513 pxdlock
= (struct pxd_lock
*) & tlck
->lock
;
1514 nlock
= pxdlock
->index
;
1515 for (i
= 0; i
< nlock
; i
++, pxdlock
++) {
1516 if (pxdlock
->flag
& mlckALLOCPXD
)
1517 lrd
->log
.updatemap
.type
=
1518 cpu_to_le16(LOG_ALLOCPXD
);
1520 lrd
->log
.updatemap
.type
=
1521 cpu_to_le16(LOG_FREEPXD
);
1522 lrd
->log
.updatemap
.nxd
= cpu_to_le16(1);
1523 lrd
->log
.updatemap
.pxd
= pxdlock
->pxd
;
1525 cpu_to_le32(lmLog(log
, tblk
, lrd
, NULL
));
1529 tlck
->flag
|= tlckUPDATEMAP
;
1531 #endif /* _JFS_WIP */
1539 * function: log data tlock
1541 static int dataLog(struct jfs_log
* log
, struct tblock
* tblk
, struct lrd
* lrd
,
1542 struct tlock
* tlck
)
1544 struct metapage
*mp
;
1549 /* initialize as REDOPAGE record format */
1550 lrd
->log
.redopage
.type
= cpu_to_le16(LOG_DATA
);
1551 lrd
->log
.redopage
.l2linesize
= cpu_to_le16(L2DATASLOTSIZE
);
1553 pxd
= &lrd
->log
.redopage
.pxd
;
1555 /* log after-image for logredo(): */
1556 lrd
->type
= cpu_to_le16(LOG_REDOPAGE
);
1558 if (jfs_dirtable_inline(tlck
->ip
)) {
1560 * The table has been truncated, we've must have deleted
1561 * the last entry, so don't bother logging this
1565 metapage_homeok(mp
);
1566 discard_metapage(mp
);
1571 PXDaddress(pxd
, mp
->index
);
1572 PXDlength(pxd
, mp
->logical_size
>> tblk
->sb
->s_blocksize_bits
);
1574 lrd
->backchain
= cpu_to_le32(lmLog(log
, tblk
, lrd
, tlck
));
1576 /* mark page as homeward bound */
1577 tlck
->flag
|= tlckWRITEPAGE
;
1585 * function: log dtree tlock and format maplock to update bmap;
1587 static void dtLog(struct jfs_log
* log
, struct tblock
* tblk
, struct lrd
* lrd
,
1588 struct tlock
* tlck
)
1590 struct metapage
*mp
;
1591 struct pxd_lock
*pxdlock
;
1596 /* initialize as REDOPAGE/NOREDOPAGE record format */
1597 lrd
->log
.redopage
.type
= cpu_to_le16(LOG_DTREE
);
1598 lrd
->log
.redopage
.l2linesize
= cpu_to_le16(L2DTSLOTSIZE
);
1600 pxd
= &lrd
->log
.redopage
.pxd
;
1602 if (tlck
->type
& tlckBTROOT
)
1603 lrd
->log
.redopage
.type
|= cpu_to_le16(LOG_BTROOT
);
1606 * page extension via relocation: entry insertion;
1607 * page extension in-place: entry insertion;
1608 * new right page from page split, reinitialized in-line
1609 * root from root page split: entry insertion;
1611 if (tlck
->type
& (tlckNEW
| tlckEXTEND
)) {
1612 /* log after-image of the new page for logredo():
1613 * mark log (LOG_NEW) for logredo() to initialize
1614 * freelist and update bmap for alloc of the new page;
1616 lrd
->type
= cpu_to_le16(LOG_REDOPAGE
);
1617 if (tlck
->type
& tlckEXTEND
)
1618 lrd
->log
.redopage
.type
|= cpu_to_le16(LOG_EXTEND
);
1620 lrd
->log
.redopage
.type
|= cpu_to_le16(LOG_NEW
);
1621 PXDaddress(pxd
, mp
->index
);
1623 mp
->logical_size
>> tblk
->sb
->s_blocksize_bits
);
1624 lrd
->backchain
= cpu_to_le32(lmLog(log
, tblk
, lrd
, tlck
));
1626 /* format a maplock for txUpdateMap() to update bPMAP for
1627 * alloc of the new page;
1629 if (tlck
->type
& tlckBTROOT
)
1631 tlck
->flag
|= tlckUPDATEMAP
;
1632 pxdlock
= (struct pxd_lock
*) & tlck
->lock
;
1633 pxdlock
->flag
= mlckALLOCPXD
;
1634 pxdlock
->pxd
= *pxd
;
1638 /* mark page as homeward bound */
1639 tlck
->flag
|= tlckWRITEPAGE
;
1644 * entry insertion/deletion,
1645 * sibling page link update (old right page before split);
1647 if (tlck
->type
& (tlckENTRY
| tlckRELINK
)) {
1648 /* log after-image for logredo(): */
1649 lrd
->type
= cpu_to_le16(LOG_REDOPAGE
);
1650 PXDaddress(pxd
, mp
->index
);
1652 mp
->logical_size
>> tblk
->sb
->s_blocksize_bits
);
1653 lrd
->backchain
= cpu_to_le32(lmLog(log
, tblk
, lrd
, tlck
));
1655 /* mark page as homeward bound */
1656 tlck
->flag
|= tlckWRITEPAGE
;
1661 * page deletion: page has been invalidated
1662 * page relocation: source extent
1664 * a maplock for free of the page has been formatted
1665 * at txLock() time);
1667 if (tlck
->type
& (tlckFREE
| tlckRELOCATE
)) {
1668 /* log LOG_NOREDOPAGE of the deleted page for logredo()
1669 * to start NoRedoPage filter and to update bmap for free
1670 * of the deletd page
1672 lrd
->type
= cpu_to_le16(LOG_NOREDOPAGE
);
1673 pxdlock
= (struct pxd_lock
*) & tlck
->lock
;
1674 *pxd
= pxdlock
->pxd
;
1675 lrd
->backchain
= cpu_to_le32(lmLog(log
, tblk
, lrd
, NULL
));
1677 /* a maplock for txUpdateMap() for free of the page
1678 * has been formatted at txLock() time;
1680 tlck
->flag
|= tlckUPDATEMAP
;
1688 * function: log xtree tlock and format maplock to update bmap;
1690 static void xtLog(struct jfs_log
* log
, struct tblock
* tblk
, struct lrd
* lrd
,
1691 struct tlock
* tlck
)
1694 struct metapage
*mp
;
1696 struct xtlock
*xtlck
;
1697 struct maplock
*maplock
;
1698 struct xdlistlock
*xadlock
;
1699 struct pxd_lock
*pxdlock
;
1706 /* initialize as REDOPAGE/NOREDOPAGE record format */
1707 lrd
->log
.redopage
.type
= cpu_to_le16(LOG_XTREE
);
1708 lrd
->log
.redopage
.l2linesize
= cpu_to_le16(L2XTSLOTSIZE
);
1710 page_pxd
= &lrd
->log
.redopage
.pxd
;
1712 if (tlck
->type
& tlckBTROOT
) {
1713 lrd
->log
.redopage
.type
|= cpu_to_le16(LOG_BTROOT
);
1714 p
= &JFS_IP(ip
)->i_xtroot
;
1715 if (S_ISDIR(ip
->i_mode
))
1716 lrd
->log
.redopage
.type
|=
1717 cpu_to_le16(LOG_DIR_XTREE
);
1719 p
= (xtpage_t
*) mp
->data
;
1720 next
= le16_to_cpu(p
->header
.nextindex
);
1722 xtlck
= (struct xtlock
*) & tlck
->lock
;
1724 maplock
= (struct maplock
*) & tlck
->lock
;
1725 xadlock
= (struct xdlistlock
*) maplock
;
1728 * entry insertion/extension;
1729 * sibling page link update (old right page before split);
1731 if (tlck
->type
& (tlckNEW
| tlckGROW
| tlckRELINK
)) {
1732 /* log after-image for logredo():
1733 * logredo() will update bmap for alloc of new/extended
1734 * extents (XAD_NEW|XAD_EXTEND) of XAD[lwm:next) from
1735 * after-image of XADlist;
1736 * logredo() resets (XAD_NEW|XAD_EXTEND) flag when
1737 * applying the after-image to the meta-data page.
1739 lrd
->type
= cpu_to_le16(LOG_REDOPAGE
);
1740 PXDaddress(page_pxd
, mp
->index
);
1742 mp
->logical_size
>> tblk
->sb
->s_blocksize_bits
);
1743 lrd
->backchain
= cpu_to_le32(lmLog(log
, tblk
, lrd
, tlck
));
1745 /* format a maplock for txUpdateMap() to update bPMAP
1746 * for alloc of new/extended extents of XAD[lwm:next)
1747 * from the page itself;
1748 * txUpdateMap() resets (XAD_NEW|XAD_EXTEND) flag.
1750 lwm
= xtlck
->lwm
.offset
;
1752 lwm
= XTPAGEMAXSLOT
;
1757 jfs_err("xtLog: lwm > next\n");
1760 tlck
->flag
|= tlckUPDATEMAP
;
1761 xadlock
->flag
= mlckALLOCXADLIST
;
1762 xadlock
->count
= next
- lwm
;
1763 if ((xadlock
->count
<= 4) && (tblk
->xflag
& COMMIT_LAZY
)) {
1767 * Lazy commit may allow xtree to be modified before
1768 * txUpdateMap runs. Copy xad into linelock to
1769 * preserve correct data.
1771 * We can fit twice as may pxd's as xads in the lock
1773 xadlock
->flag
= mlckALLOCPXDLIST
;
1774 pxd
= xadlock
->xdlist
= &xtlck
->pxdlock
;
1775 for (i
= 0; i
< xadlock
->count
; i
++) {
1776 PXDaddress(pxd
, addressXAD(&p
->xad
[lwm
+ i
]));
1777 PXDlength(pxd
, lengthXAD(&p
->xad
[lwm
+ i
]));
1778 p
->xad
[lwm
+ i
].flag
&=
1779 ~(XAD_NEW
| XAD_EXTENDED
);
1784 * xdlist will point to into inode's xtree, ensure
1785 * that transaction is not committed lazily.
1787 xadlock
->flag
= mlckALLOCXADLIST
;
1788 xadlock
->xdlist
= &p
->xad
[lwm
];
1789 tblk
->xflag
&= ~COMMIT_LAZY
;
1791 jfs_info("xtLog: alloc ip:0x%p mp:0x%p tlck:0x%p lwm:%d "
1792 "count:%d", tlck
->ip
, mp
, tlck
, lwm
, xadlock
->count
);
1797 /* mark page as homeward bound */
1798 tlck
->flag
|= tlckWRITEPAGE
;
1804 * page deletion: file deletion/truncation (ref. xtTruncate())
1806 * (page will be invalidated after log is written and bmap
1807 * is updated from the page);
1809 if (tlck
->type
& tlckFREE
) {
1810 /* LOG_NOREDOPAGE log for NoRedoPage filter:
1811 * if page free from file delete, NoRedoFile filter from
1812 * inode image of zero link count will subsume NoRedoPage
1813 * filters for each page;
1814 * if page free from file truncattion, write NoRedoPage
1817 * upadte of block allocation map for the page itself:
1818 * if page free from deletion and truncation, LOG_UPDATEMAP
1819 * log for the page itself is generated from processing
1820 * its parent page xad entries;
1822 /* if page free from file truncation, log LOG_NOREDOPAGE
1823 * of the deleted page for logredo() to start NoRedoPage
1824 * filter for the page;
1826 if (tblk
->xflag
& COMMIT_TRUNCATE
) {
1827 /* write NOREDOPAGE for the page */
1828 lrd
->type
= cpu_to_le16(LOG_NOREDOPAGE
);
1829 PXDaddress(page_pxd
, mp
->index
);
1831 mp
->logical_size
>> tblk
->sb
->
1834 cpu_to_le32(lmLog(log
, tblk
, lrd
, NULL
));
1836 if (tlck
->type
& tlckBTROOT
) {
1837 /* Empty xtree must be logged */
1838 lrd
->type
= cpu_to_le16(LOG_REDOPAGE
);
1840 cpu_to_le32(lmLog(log
, tblk
, lrd
, tlck
));
1844 /* init LOG_UPDATEMAP of the freed extents
1845 * XAD[XTENTRYSTART:hwm) from the deleted page itself
1846 * for logredo() to update bmap;
1848 lrd
->type
= cpu_to_le16(LOG_UPDATEMAP
);
1849 lrd
->log
.updatemap
.type
= cpu_to_le16(LOG_FREEXADLIST
);
1850 xtlck
= (struct xtlock
*) & tlck
->lock
;
1851 hwm
= xtlck
->hwm
.offset
;
1852 lrd
->log
.updatemap
.nxd
=
1853 cpu_to_le16(hwm
- XTENTRYSTART
+ 1);
1854 /* reformat linelock for lmLog() */
1855 xtlck
->header
.offset
= XTENTRYSTART
;
1856 xtlck
->header
.length
= hwm
- XTENTRYSTART
+ 1;
1858 lrd
->backchain
= cpu_to_le32(lmLog(log
, tblk
, lrd
, tlck
));
1860 /* format a maplock for txUpdateMap() to update bmap
1861 * to free extents of XAD[XTENTRYSTART:hwm) from the
1862 * deleted page itself;
1864 tlck
->flag
|= tlckUPDATEMAP
;
1865 xadlock
->count
= hwm
- XTENTRYSTART
+ 1;
1866 if ((xadlock
->count
<= 4) && (tblk
->xflag
& COMMIT_LAZY
)) {
1870 * Lazy commit may allow xtree to be modified before
1871 * txUpdateMap runs. Copy xad into linelock to
1872 * preserve correct data.
1874 * We can fit twice as may pxd's as xads in the lock
1876 xadlock
->flag
= mlckFREEPXDLIST
;
1877 pxd
= xadlock
->xdlist
= &xtlck
->pxdlock
;
1878 for (i
= 0; i
< xadlock
->count
; i
++) {
1880 addressXAD(&p
->xad
[XTENTRYSTART
+ i
]));
1882 lengthXAD(&p
->xad
[XTENTRYSTART
+ i
]));
1887 * xdlist will point to into inode's xtree, ensure
1888 * that transaction is not committed lazily.
1890 xadlock
->flag
= mlckFREEXADLIST
;
1891 xadlock
->xdlist
= &p
->xad
[XTENTRYSTART
];
1892 tblk
->xflag
&= ~COMMIT_LAZY
;
1894 jfs_info("xtLog: free ip:0x%p mp:0x%p count:%d lwm:2",
1895 tlck
->ip
, mp
, xadlock
->count
);
1899 /* mark page as invalid */
1900 if (((tblk
->xflag
& COMMIT_PWMAP
) || S_ISDIR(ip
->i_mode
))
1901 && !(tlck
->type
& tlckBTROOT
))
1902 tlck
->flag
|= tlckFREEPAGE
;
1904 else (tblk->xflag & COMMIT_PMAP)
1911 * page/entry truncation: file truncation (ref. xtTruncate())
1913 * |----------+------+------+---------------|
1915 * | | hwm - hwm before truncation
1916 * | next - truncation point
1917 * lwm - lwm before truncation
1920 if (tlck
->type
& tlckTRUNCATE
) {
1921 /* This odd declaration suppresses a bogus gcc warning */
1922 pxd_t pxd
= pxd
; /* truncated extent of xad */
1926 * For truncation the entire linelock may be used, so it would
1927 * be difficult to store xad list in linelock itself.
1928 * Therefore, we'll just force transaction to be committed
1929 * synchronously, so that xtree pages won't be changed before
1932 tblk
->xflag
&= ~COMMIT_LAZY
;
1933 lwm
= xtlck
->lwm
.offset
;
1935 lwm
= XTPAGEMAXSLOT
;
1936 hwm
= xtlck
->hwm
.offset
;
1937 twm
= xtlck
->twm
.offset
;
1942 /* log after-image for logredo():
1944 * logredo() will update bmap for alloc of new/extended
1945 * extents (XAD_NEW|XAD_EXTEND) of XAD[lwm:next) from
1946 * after-image of XADlist;
1947 * logredo() resets (XAD_NEW|XAD_EXTEND) flag when
1948 * applying the after-image to the meta-data page.
1950 lrd
->type
= cpu_to_le16(LOG_REDOPAGE
);
1951 PXDaddress(page_pxd
, mp
->index
);
1953 mp
->logical_size
>> tblk
->sb
->s_blocksize_bits
);
1954 lrd
->backchain
= cpu_to_le32(lmLog(log
, tblk
, lrd
, tlck
));
1957 * truncate entry XAD[twm == next - 1]:
1959 if (twm
== next
- 1) {
1960 /* init LOG_UPDATEMAP for logredo() to update bmap for
1961 * free of truncated delta extent of the truncated
1962 * entry XAD[next - 1]:
1963 * (xtlck->pxdlock = truncated delta extent);
1965 pxdlock
= (struct pxd_lock
*) & xtlck
->pxdlock
;
1966 /* assert(pxdlock->type & tlckTRUNCATE); */
1967 lrd
->type
= cpu_to_le16(LOG_UPDATEMAP
);
1968 lrd
->log
.updatemap
.type
= cpu_to_le16(LOG_FREEPXD
);
1969 lrd
->log
.updatemap
.nxd
= cpu_to_le16(1);
1970 lrd
->log
.updatemap
.pxd
= pxdlock
->pxd
;
1971 pxd
= pxdlock
->pxd
; /* save to format maplock */
1973 cpu_to_le32(lmLog(log
, tblk
, lrd
, NULL
));
1977 * free entries XAD[next:hwm]:
1980 /* init LOG_UPDATEMAP of the freed extents
1981 * XAD[next:hwm] from the deleted page itself
1982 * for logredo() to update bmap;
1984 lrd
->type
= cpu_to_le16(LOG_UPDATEMAP
);
1985 lrd
->log
.updatemap
.type
=
1986 cpu_to_le16(LOG_FREEXADLIST
);
1987 xtlck
= (struct xtlock
*) & tlck
->lock
;
1988 hwm
= xtlck
->hwm
.offset
;
1989 lrd
->log
.updatemap
.nxd
=
1990 cpu_to_le16(hwm
- next
+ 1);
1991 /* reformat linelock for lmLog() */
1992 xtlck
->header
.offset
= next
;
1993 xtlck
->header
.length
= hwm
- next
+ 1;
1996 cpu_to_le32(lmLog(log
, tblk
, lrd
, tlck
));
2000 * format maplock(s) for txUpdateMap() to update bmap
2005 * allocate entries XAD[lwm:next):
2008 /* format a maplock for txUpdateMap() to update bPMAP
2009 * for alloc of new/extended extents of XAD[lwm:next)
2010 * from the page itself;
2011 * txUpdateMap() resets (XAD_NEW|XAD_EXTEND) flag.
2013 tlck
->flag
|= tlckUPDATEMAP
;
2014 xadlock
->flag
= mlckALLOCXADLIST
;
2015 xadlock
->count
= next
- lwm
;
2016 xadlock
->xdlist
= &p
->xad
[lwm
];
2018 jfs_info("xtLog: alloc ip:0x%p mp:0x%p count:%d "
2020 tlck
->ip
, mp
, xadlock
->count
, lwm
, next
);
2026 * truncate entry XAD[twm == next - 1]:
2028 if (twm
== next
- 1) {
2029 /* format a maplock for txUpdateMap() to update bmap
2030 * to free truncated delta extent of the truncated
2031 * entry XAD[next - 1];
2032 * (xtlck->pxdlock = truncated delta extent);
2034 tlck
->flag
|= tlckUPDATEMAP
;
2035 pxdlock
= (struct pxd_lock
*) xadlock
;
2036 pxdlock
->flag
= mlckFREEPXD
;
2040 jfs_info("xtLog: truncate ip:0x%p mp:0x%p count:%d "
2041 "hwm:%d", ip
, mp
, pxdlock
->count
, hwm
);
2047 * free entries XAD[next:hwm]:
2050 /* format a maplock for txUpdateMap() to update bmap
2051 * to free extents of XAD[next:hwm] from thedeleted
2054 tlck
->flag
|= tlckUPDATEMAP
;
2055 xadlock
->flag
= mlckFREEXADLIST
;
2056 xadlock
->count
= hwm
- next
+ 1;
2057 xadlock
->xdlist
= &p
->xad
[next
];
2059 jfs_info("xtLog: free ip:0x%p mp:0x%p count:%d "
2061 tlck
->ip
, mp
, xadlock
->count
, next
, hwm
);
2065 /* mark page as homeward bound */
2066 tlck
->flag
|= tlckWRITEPAGE
;
2074 * function: log from maplock of freed data extents;
2076 static void mapLog(struct jfs_log
* log
, struct tblock
* tblk
, struct lrd
* lrd
,
2077 struct tlock
* tlck
)
2079 struct pxd_lock
*pxdlock
;
2084 * page relocation: free the source page extent
2086 * a maplock for txUpdateMap() for free of the page
2087 * has been formatted at txLock() time saving the src
2088 * relocated page address;
2090 if (tlck
->type
& tlckRELOCATE
) {
2091 /* log LOG_NOREDOPAGE of the old relocated page
2092 * for logredo() to start NoRedoPage filter;
2094 lrd
->type
= cpu_to_le16(LOG_NOREDOPAGE
);
2095 pxdlock
= (struct pxd_lock
*) & tlck
->lock
;
2096 pxd
= &lrd
->log
.redopage
.pxd
;
2097 *pxd
= pxdlock
->pxd
;
2098 lrd
->backchain
= cpu_to_le32(lmLog(log
, tblk
, lrd
, NULL
));
2100 /* (N.B. currently, logredo() does NOT update bmap
2101 * for free of the page itself for (LOG_XTREE|LOG_NOREDOPAGE);
2102 * if page free from relocation, LOG_UPDATEMAP log is
2103 * specifically generated now for logredo()
2104 * to update bmap for free of src relocated page;
2105 * (new flag LOG_RELOCATE may be introduced which will
2106 * inform logredo() to start NORedoPage filter and also
2107 * update block allocation map at the same time, thus
2108 * avoiding an extra log write);
2110 lrd
->type
= cpu_to_le16(LOG_UPDATEMAP
);
2111 lrd
->log
.updatemap
.type
= cpu_to_le16(LOG_FREEPXD
);
2112 lrd
->log
.updatemap
.nxd
= cpu_to_le16(1);
2113 lrd
->log
.updatemap
.pxd
= pxdlock
->pxd
;
2114 lrd
->backchain
= cpu_to_le32(lmLog(log
, tblk
, lrd
, NULL
));
2116 /* a maplock for txUpdateMap() for free of the page
2117 * has been formatted at txLock() time;
2119 tlck
->flag
|= tlckUPDATEMAP
;
2124 * Otherwise it's not a relocate request
2128 /* log LOG_UPDATEMAP for logredo() to update bmap for
2129 * free of truncated/relocated delta extent of the data;
2130 * e.g.: external EA extent, relocated/truncated extent
2131 * from xtTailgate();
2133 lrd
->type
= cpu_to_le16(LOG_UPDATEMAP
);
2134 pxdlock
= (struct pxd_lock
*) & tlck
->lock
;
2135 nlock
= pxdlock
->index
;
2136 for (i
= 0; i
< nlock
; i
++, pxdlock
++) {
2137 if (pxdlock
->flag
& mlckALLOCPXD
)
2138 lrd
->log
.updatemap
.type
=
2139 cpu_to_le16(LOG_ALLOCPXD
);
2141 lrd
->log
.updatemap
.type
=
2142 cpu_to_le16(LOG_FREEPXD
);
2143 lrd
->log
.updatemap
.nxd
= cpu_to_le16(1);
2144 lrd
->log
.updatemap
.pxd
= pxdlock
->pxd
;
2146 cpu_to_le32(lmLog(log
, tblk
, lrd
, NULL
));
2147 jfs_info("mapLog: xaddr:0x%lx xlen:0x%x",
2148 (ulong
) addressPXD(&pxdlock
->pxd
),
2149 lengthPXD(&pxdlock
->pxd
));
2153 tlck
->flag
|= tlckUPDATEMAP
;
2160 * function: acquire maplock for EA/ACL extents or
2161 * set COMMIT_INLINE flag;
2163 void txEA(tid_t tid
, struct inode
*ip
, dxd_t
* oldea
, dxd_t
* newea
)
2165 struct tlock
*tlck
= NULL
;
2166 struct pxd_lock
*maplock
= NULL
, *pxdlock
= NULL
;
2169 * format maplock for alloc of new EA extent
2172 /* Since the newea could be a completely zeroed entry we need to
2173 * check for the two flags which indicate we should actually
2174 * commit new EA data
2176 if (newea
->flag
& DXD_EXTENT
) {
2177 tlck
= txMaplock(tid
, ip
, tlckMAP
);
2178 maplock
= (struct pxd_lock
*) & tlck
->lock
;
2179 pxdlock
= (struct pxd_lock
*) maplock
;
2180 pxdlock
->flag
= mlckALLOCPXD
;
2181 PXDaddress(&pxdlock
->pxd
, addressDXD(newea
));
2182 PXDlength(&pxdlock
->pxd
, lengthDXD(newea
));
2185 } else if (newea
->flag
& DXD_INLINE
) {
2188 set_cflag(COMMIT_Inlineea
, ip
);
2193 * format maplock for free of old EA extent
2195 if (!test_cflag(COMMIT_Nolink
, ip
) && oldea
->flag
& DXD_EXTENT
) {
2197 tlck
= txMaplock(tid
, ip
, tlckMAP
);
2198 maplock
= (struct pxd_lock
*) & tlck
->lock
;
2199 pxdlock
= (struct pxd_lock
*) maplock
;
2202 pxdlock
->flag
= mlckFREEPXD
;
2203 PXDaddress(&pxdlock
->pxd
, addressDXD(oldea
));
2204 PXDlength(&pxdlock
->pxd
, lengthDXD(oldea
));
2212 * function: synchronously write pages locked by transaction
2213 * after txLog() but before txUpdateMap();
2215 static void txForce(struct tblock
* tblk
)
2219 struct metapage
*mp
;
2222 * reverse the order of transaction tlocks in
2223 * careful update order of address index pages
2224 * (right to left, bottom up)
2226 tlck
= lid_to_tlock(tblk
->next
);
2230 tlck
= lid_to_tlock(lid
);
2232 tlck
->next
= tblk
->next
;
2238 * synchronously write the page, and
2239 * hold the page for txUpdateMap();
2241 for (lid
= tblk
->next
; lid
; lid
= next
) {
2242 tlck
= lid_to_tlock(lid
);
2245 if ((mp
= tlck
->mp
) != NULL
&&
2246 (tlck
->type
& tlckBTROOT
) == 0) {
2247 assert(mp
->xflag
& COMMIT_PAGE
);
2249 if (tlck
->flag
& tlckWRITEPAGE
) {
2250 tlck
->flag
&= ~tlckWRITEPAGE
;
2252 /* do not release page to freelist */
2256 * The "right" thing to do here is to
2257 * synchronously write the metadata.
2258 * With the current implementation this
2259 * is hard since write_metapage requires
2260 * us to kunmap & remap the page. If we
2261 * have tlocks pointing into the metadata
2262 * pages, we don't want to do this. I think
2263 * we can get by with synchronously writing
2264 * the pages when they are released.
2266 assert(mp
->nohomeok
);
2267 set_bit(META_dirty
, &mp
->flag
);
2268 set_bit(META_sync
, &mp
->flag
);
2278 * function: update persistent allocation map (and working map
2283 static void txUpdateMap(struct tblock
* tblk
)
2286 struct inode
*ipimap
;
2289 struct maplock
*maplock
;
2290 struct pxd_lock pxdlock
;
2293 struct metapage
*mp
= NULL
;
2295 ipimap
= JFS_SBI(tblk
->sb
)->ipimap
;
2297 maptype
= (tblk
->xflag
& COMMIT_PMAP
) ? COMMIT_PMAP
: COMMIT_PWMAP
;
2301 * update block allocation map
2303 * update allocation state in pmap (and wmap) and
2304 * update lsn of the pmap page;
2307 * scan each tlock/page of transaction for block allocation/free:
2309 * for each tlock/page of transaction, update map.
2310 * ? are there tlock for pmap and pwmap at the same time ?
2312 for (lid
= tblk
->next
; lid
; lid
= tlck
->next
) {
2313 tlck
= lid_to_tlock(lid
);
2315 if ((tlck
->flag
& tlckUPDATEMAP
) == 0)
2318 if (tlck
->flag
& tlckFREEPAGE
) {
2320 * Another thread may attempt to reuse freed space
2321 * immediately, so we want to get rid of the metapage
2322 * before anyone else has a chance to get it.
2323 * Lock metapage, update maps, then invalidate
2327 ASSERT(mp
->xflag
& COMMIT_PAGE
);
2333 * . in-line PXD list:
2334 * . out-of-line XAD list:
2336 maplock
= (struct maplock
*) & tlck
->lock
;
2337 nlock
= maplock
->index
;
2339 for (k
= 0; k
< nlock
; k
++, maplock
++) {
2341 * allocate blocks in persistent map:
2343 * blocks have been allocated from wmap at alloc time;
2345 if (maplock
->flag
& mlckALLOC
) {
2346 txAllocPMap(ipimap
, maplock
, tblk
);
2349 * free blocks in persistent and working map:
2350 * blocks will be freed in pmap and then in wmap;
2352 * ? tblock specifies the PMAP/PWMAP based upon
2355 * free blocks in persistent map:
2356 * blocks will be freed from wmap at last reference
2357 * release of the object for regular files;
2359 * Alway free blocks from both persistent & working
2360 * maps for directories
2362 else { /* (maplock->flag & mlckFREE) */
2364 if (tlck
->flag
& tlckDIRECTORY
)
2365 txFreeMap(ipimap
, maplock
,
2366 tblk
, COMMIT_PWMAP
);
2368 txFreeMap(ipimap
, maplock
,
2372 if (tlck
->flag
& tlckFREEPAGE
) {
2373 if (!(tblk
->flag
& tblkGC_LAZY
)) {
2374 /* This is equivalent to txRelease */
2375 ASSERT(mp
->lid
== lid
);
2378 assert(mp
->nohomeok
== 1);
2379 metapage_homeok(mp
);
2380 discard_metapage(mp
);
2385 * update inode allocation map
2387 * update allocation state in pmap and
2388 * update lsn of the pmap page;
2389 * update in-memory inode flag/state
2391 * unlock mapper/write lock
2393 if (tblk
->xflag
& COMMIT_CREATE
) {
2394 diUpdatePMap(ipimap
, tblk
->ino
, false, tblk
);
2395 /* update persistent block allocation map
2396 * for the allocation of inode extent;
2398 pxdlock
.flag
= mlckALLOCPXD
;
2399 pxdlock
.pxd
= tblk
->u
.ixpxd
;
2401 txAllocPMap(ipimap
, (struct maplock
*) & pxdlock
, tblk
);
2402 } else if (tblk
->xflag
& COMMIT_DELETE
) {
2404 diUpdatePMap(ipimap
, ip
->i_ino
, true, tblk
);
2412 * function: allocate from persistent map;
2421 * allocate from persistent map;
2422 * free from persistent map;
2423 * (e.g., tmp file - free from working map at releae
2424 * of last reference);
2425 * free from persistent and working map;
2427 * lsn - log sequence number;
2429 static void txAllocPMap(struct inode
*ip
, struct maplock
* maplock
,
2430 struct tblock
* tblk
)
2432 struct inode
*ipbmap
= JFS_SBI(ip
->i_sb
)->ipbmap
;
2433 struct xdlistlock
*xadlistlock
;
2437 struct pxd_lock
*pxdlock
;
2438 struct xdlistlock
*pxdlistlock
;
2443 * allocate from persistent map;
2445 if (maplock
->flag
& mlckALLOCXADLIST
) {
2446 xadlistlock
= (struct xdlistlock
*) maplock
;
2447 xad
= xadlistlock
->xdlist
;
2448 for (n
= 0; n
< xadlistlock
->count
; n
++, xad
++) {
2449 if (xad
->flag
& (XAD_NEW
| XAD_EXTENDED
)) {
2450 xaddr
= addressXAD(xad
);
2451 xlen
= lengthXAD(xad
);
2452 dbUpdatePMap(ipbmap
, false, xaddr
,
2454 xad
->flag
&= ~(XAD_NEW
| XAD_EXTENDED
);
2455 jfs_info("allocPMap: xaddr:0x%lx xlen:%d",
2456 (ulong
) xaddr
, xlen
);
2459 } else if (maplock
->flag
& mlckALLOCPXD
) {
2460 pxdlock
= (struct pxd_lock
*) maplock
;
2461 xaddr
= addressPXD(&pxdlock
->pxd
);
2462 xlen
= lengthPXD(&pxdlock
->pxd
);
2463 dbUpdatePMap(ipbmap
, false, xaddr
, (s64
) xlen
, tblk
);
2464 jfs_info("allocPMap: xaddr:0x%lx xlen:%d", (ulong
) xaddr
, xlen
);
2465 } else { /* (maplock->flag & mlckALLOCPXDLIST) */
2467 pxdlistlock
= (struct xdlistlock
*) maplock
;
2468 pxd
= pxdlistlock
->xdlist
;
2469 for (n
= 0; n
< pxdlistlock
->count
; n
++, pxd
++) {
2470 xaddr
= addressPXD(pxd
);
2471 xlen
= lengthPXD(pxd
);
2472 dbUpdatePMap(ipbmap
, false, xaddr
, (s64
) xlen
,
2474 jfs_info("allocPMap: xaddr:0x%lx xlen:%d",
2475 (ulong
) xaddr
, xlen
);
2483 * function: free from persistent and/or working map;
2485 * todo: optimization
2487 void txFreeMap(struct inode
*ip
,
2488 struct maplock
* maplock
, struct tblock
* tblk
, int maptype
)
2490 struct inode
*ipbmap
= JFS_SBI(ip
->i_sb
)->ipbmap
;
2491 struct xdlistlock
*xadlistlock
;
2495 struct pxd_lock
*pxdlock
;
2496 struct xdlistlock
*pxdlistlock
;
2500 jfs_info("txFreeMap: tblk:0x%p maplock:0x%p maptype:0x%x",
2501 tblk
, maplock
, maptype
);
2504 * free from persistent map;
2506 if (maptype
== COMMIT_PMAP
|| maptype
== COMMIT_PWMAP
) {
2507 if (maplock
->flag
& mlckFREEXADLIST
) {
2508 xadlistlock
= (struct xdlistlock
*) maplock
;
2509 xad
= xadlistlock
->xdlist
;
2510 for (n
= 0; n
< xadlistlock
->count
; n
++, xad
++) {
2511 if (!(xad
->flag
& XAD_NEW
)) {
2512 xaddr
= addressXAD(xad
);
2513 xlen
= lengthXAD(xad
);
2514 dbUpdatePMap(ipbmap
, true, xaddr
,
2516 jfs_info("freePMap: xaddr:0x%lx "
2518 (ulong
) xaddr
, xlen
);
2521 } else if (maplock
->flag
& mlckFREEPXD
) {
2522 pxdlock
= (struct pxd_lock
*) maplock
;
2523 xaddr
= addressPXD(&pxdlock
->pxd
);
2524 xlen
= lengthPXD(&pxdlock
->pxd
);
2525 dbUpdatePMap(ipbmap
, true, xaddr
, (s64
) xlen
,
2527 jfs_info("freePMap: xaddr:0x%lx xlen:%d",
2528 (ulong
) xaddr
, xlen
);
2529 } else { /* (maplock->flag & mlckALLOCPXDLIST) */
2531 pxdlistlock
= (struct xdlistlock
*) maplock
;
2532 pxd
= pxdlistlock
->xdlist
;
2533 for (n
= 0; n
< pxdlistlock
->count
; n
++, pxd
++) {
2534 xaddr
= addressPXD(pxd
);
2535 xlen
= lengthPXD(pxd
);
2536 dbUpdatePMap(ipbmap
, true, xaddr
,
2538 jfs_info("freePMap: xaddr:0x%lx xlen:%d",
2539 (ulong
) xaddr
, xlen
);
2545 * free from working map;
2547 if (maptype
== COMMIT_PWMAP
|| maptype
== COMMIT_WMAP
) {
2548 if (maplock
->flag
& mlckFREEXADLIST
) {
2549 xadlistlock
= (struct xdlistlock
*) maplock
;
2550 xad
= xadlistlock
->xdlist
;
2551 for (n
= 0; n
< xadlistlock
->count
; n
++, xad
++) {
2552 xaddr
= addressXAD(xad
);
2553 xlen
= lengthXAD(xad
);
2554 dbFree(ip
, xaddr
, (s64
) xlen
);
2556 jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
2557 (ulong
) xaddr
, xlen
);
2559 } else if (maplock
->flag
& mlckFREEPXD
) {
2560 pxdlock
= (struct pxd_lock
*) maplock
;
2561 xaddr
= addressPXD(&pxdlock
->pxd
);
2562 xlen
= lengthPXD(&pxdlock
->pxd
);
2563 dbFree(ip
, xaddr
, (s64
) xlen
);
2564 jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
2565 (ulong
) xaddr
, xlen
);
2566 } else { /* (maplock->flag & mlckFREEPXDLIST) */
2568 pxdlistlock
= (struct xdlistlock
*) maplock
;
2569 pxd
= pxdlistlock
->xdlist
;
2570 for (n
= 0; n
< pxdlistlock
->count
; n
++, pxd
++) {
2571 xaddr
= addressPXD(pxd
);
2572 xlen
= lengthPXD(pxd
);
2573 dbFree(ip
, xaddr
, (s64
) xlen
);
2574 jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
2575 (ulong
) xaddr
, xlen
);
2584 * function: remove tlock from inode anonymous locklist
2586 void txFreelock(struct inode
*ip
)
2588 struct jfs_inode_info
*jfs_ip
= JFS_IP(ip
);
2589 struct tlock
*xtlck
, *tlck
;
2590 lid_t xlid
= 0, lid
;
2592 if (!jfs_ip
->atlhead
)
2596 xtlck
= (struct tlock
*) &jfs_ip
->atlhead
;
2598 while ((lid
= xtlck
->next
) != 0) {
2599 tlck
= lid_to_tlock(lid
);
2600 if (tlck
->flag
& tlckFREELOCK
) {
2601 xtlck
->next
= tlck
->next
;
2609 if (jfs_ip
->atlhead
)
2610 jfs_ip
->atltail
= xlid
;
2612 jfs_ip
->atltail
= 0;
2614 * If inode was on anon_list, remove it
2616 list_del_init(&jfs_ip
->anon_inode_list
);
2624 * function: abort tx before commit;
2626 * frees line-locks and segment locks for all
2627 * segments in comdata structure.
2628 * Optionally sets state of file-system to FM_DIRTY in super-block.
2629 * log age of page-frames in memory for which caller has
2630 * are reset to 0 (to avoid logwarap).
2632 void txAbort(tid_t tid
, int dirty
)
2635 struct metapage
*mp
;
2636 struct tblock
*tblk
= tid_to_tblock(tid
);
2640 * free tlocks of the transaction
2642 for (lid
= tblk
->next
; lid
; lid
= next
) {
2643 tlck
= lid_to_tlock(lid
);
2646 JFS_IP(tlck
->ip
)->xtlid
= 0;
2652 * reset lsn of page to avoid logwarap:
2654 * (page may have been previously committed by another
2655 * transaction(s) but has not been paged, i.e.,
2656 * it may be on logsync list even though it has not
2657 * been logged for the current tx.)
2659 if (mp
->xflag
& COMMIT_PAGE
&& mp
->lsn
)
2662 /* insert tlock at head of freelist */
2668 /* caller will free the transaction block */
2670 tblk
->next
= tblk
->last
= 0;
2673 * mark filesystem dirty
2676 jfs_error(tblk
->sb
, "txAbort");
2682 * txLazyCommit(void)
2684 * All transactions except those changing ipimap (COMMIT_FORCE) are
2685 * processed by this routine. This insures that the inode and block
2686 * allocation maps are updated in order. For synchronous transactions,
2687 * let the user thread finish processing after txUpdateMap() is called.
2689 static void txLazyCommit(struct tblock
* tblk
)
2691 struct jfs_log
*log
;
2693 while (((tblk
->flag
& tblkGC_READY
) == 0) &&
2694 ((tblk
->flag
& tblkGC_UNLOCKED
) == 0)) {
2695 /* We must have gotten ahead of the user thread
2697 jfs_info("jfs_lazycommit: tblk 0x%p not unlocked", tblk
);
2701 jfs_info("txLazyCommit: processing tblk 0x%p", tblk
);
2705 log
= (struct jfs_log
*) JFS_SBI(tblk
->sb
)->log
;
2707 spin_lock_irq(&log
->gclock
); // LOGGC_LOCK
2709 tblk
->flag
|= tblkGC_COMMITTED
;
2711 if (tblk
->flag
& tblkGC_READY
)
2714 wake_up_all(&tblk
->gcwait
); // LOGGC_WAKEUP
2717 * Can't release log->gclock until we've tested tblk->flag
2719 if (tblk
->flag
& tblkGC_LAZY
) {
2720 spin_unlock_irq(&log
->gclock
); // LOGGC_UNLOCK
2722 tblk
->flag
&= ~tblkGC_LAZY
;
2723 txEnd(tblk
- TxBlock
); /* Convert back to tid */
2725 spin_unlock_irq(&log
->gclock
); // LOGGC_UNLOCK
2727 jfs_info("txLazyCommit: done: tblk = 0x%p", tblk
);
2731 * jfs_lazycommit(void)
2733 * To be run as a kernel daemon. If lbmIODone is called in an interrupt
2734 * context, or where blocking is not wanted, this routine will process
2735 * committed transactions from the unlock queue.
2737 int jfs_lazycommit(void *arg
)
2740 struct tblock
*tblk
;
2741 unsigned long flags
;
2742 struct jfs_sb_info
*sbi
;
2746 jfs_commit_thread_waking
= 0; /* OK to wake another thread */
2747 while (!list_empty(&TxAnchor
.unlock_queue
)) {
2749 list_for_each_entry(tblk
, &TxAnchor
.unlock_queue
,
2752 sbi
= JFS_SBI(tblk
->sb
);
2754 * For each volume, the transactions must be
2755 * handled in order. If another commit thread
2756 * is handling a tblk for this superblock,
2759 if (sbi
->commit_state
& IN_LAZYCOMMIT
)
2762 sbi
->commit_state
|= IN_LAZYCOMMIT
;
2766 * Remove transaction from queue
2768 list_del(&tblk
->cqueue
);
2774 sbi
->commit_state
&= ~IN_LAZYCOMMIT
;
2776 * Don't continue in the for loop. (We can't
2777 * anyway, it's unsafe!) We want to go back to
2778 * the beginning of the list.
2783 /* If there was nothing to do, don't continue */
2787 /* In case a wakeup came while all threads were active */
2788 jfs_commit_thread_waking
= 0;
2790 if (freezing(current
)) {
2794 DECLARE_WAITQUEUE(wq
, current
);
2796 add_wait_queue(&jfs_commit_thread_wait
, &wq
);
2797 set_current_state(TASK_INTERRUPTIBLE
);
2800 __set_current_state(TASK_RUNNING
);
2801 remove_wait_queue(&jfs_commit_thread_wait
, &wq
);
2803 } while (!kthread_should_stop());
2805 if (!list_empty(&TxAnchor
.unlock_queue
))
2806 jfs_err("jfs_lazycommit being killed w/pending transactions!");
2808 jfs_info("jfs_lazycommit being killed\n");
2812 void txLazyUnlock(struct tblock
* tblk
)
2814 unsigned long flags
;
2818 list_add_tail(&tblk
->cqueue
, &TxAnchor
.unlock_queue
);
2820 * Don't wake up a commit thread if there is already one servicing
2821 * this superblock, or if the last one we woke up hasn't started yet.
2823 if (!(JFS_SBI(tblk
->sb
)->commit_state
& IN_LAZYCOMMIT
) &&
2824 !jfs_commit_thread_waking
) {
2825 jfs_commit_thread_waking
= 1;
2826 wake_up(&jfs_commit_thread_wait
);
2831 static void LogSyncRelease(struct metapage
* mp
)
2833 struct jfs_log
*log
= mp
->log
;
2835 assert(mp
->nohomeok
);
2837 metapage_homeok(mp
);
2843 * Block all new transactions and push anonymous transactions to
2846 * This does almost the same thing as jfs_sync below. We don't
2847 * worry about deadlocking when jfs_tlocks_low is set, since we would
2848 * expect jfs_sync to get us out of that jam.
2850 void txQuiesce(struct super_block
*sb
)
2853 struct jfs_inode_info
*jfs_ip
;
2854 struct jfs_log
*log
= JFS_SBI(sb
)->log
;
2857 set_bit(log_QUIESCE
, &log
->flag
);
2861 while (!list_empty(&TxAnchor
.anon_list
)) {
2862 jfs_ip
= list_entry(TxAnchor
.anon_list
.next
,
2863 struct jfs_inode_info
,
2865 ip
= &jfs_ip
->vfs_inode
;
2868 * inode will be removed from anonymous list
2869 * when it is committed
2872 tid
= txBegin(ip
->i_sb
, COMMIT_INODE
| COMMIT_FORCE
);
2873 mutex_lock(&jfs_ip
->commit_mutex
);
2874 txCommit(tid
, 1, &ip
, 0);
2876 mutex_unlock(&jfs_ip
->commit_mutex
);
2878 * Just to be safe. I don't know how
2879 * long we can run without blocking
2886 * If jfs_sync is running in parallel, there could be some inodes
2887 * on anon_list2. Let's check.
2889 if (!list_empty(&TxAnchor
.anon_list2
)) {
2890 list_splice(&TxAnchor
.anon_list2
, &TxAnchor
.anon_list
);
2891 INIT_LIST_HEAD(&TxAnchor
.anon_list2
);
2897 * We may need to kick off the group commit
2899 jfs_flush_journal(log
, 0);
2905 * Allows transactions to start again following txQuiesce
2907 void txResume(struct super_block
*sb
)
2909 struct jfs_log
*log
= JFS_SBI(sb
)->log
;
2911 clear_bit(log_QUIESCE
, &log
->flag
);
2912 TXN_WAKEUP(&log
->syncwait
);
2918 * To be run as a kernel daemon. This is awakened when tlocks run low.
2919 * We write any inodes that have anonymous tlocks so they will become
2922 int jfs_sync(void *arg
)
2925 struct jfs_inode_info
*jfs_ip
;
2931 * write each inode on the anonymous inode list
2934 while (jfs_tlocks_low
&& !list_empty(&TxAnchor
.anon_list
)) {
2935 jfs_ip
= list_entry(TxAnchor
.anon_list
.next
,
2936 struct jfs_inode_info
,
2938 ip
= &jfs_ip
->vfs_inode
;
2942 * Inode is being freed
2944 list_del_init(&jfs_ip
->anon_inode_list
);
2945 } else if (mutex_trylock(&jfs_ip
->commit_mutex
)) {
2947 * inode will be removed from anonymous list
2948 * when it is committed
2951 tid
= txBegin(ip
->i_sb
, COMMIT_INODE
);
2952 rc
= txCommit(tid
, 1, &ip
, 0);
2954 mutex_unlock(&jfs_ip
->commit_mutex
);
2958 * Just to be safe. I don't know how
2959 * long we can run without blocking
2964 /* We can't get the commit mutex. It may
2965 * be held by a thread waiting for tlock's
2966 * so let's not block here. Save it to
2967 * put back on the anon_list.
2970 /* Take off anon_list */
2971 list_del(&jfs_ip
->anon_inode_list
);
2973 /* Put on anon_list2 */
2974 list_add(&jfs_ip
->anon_inode_list
,
2975 &TxAnchor
.anon_list2
);
2982 /* Add anon_list2 back to anon_list */
2983 list_splice_init(&TxAnchor
.anon_list2
, &TxAnchor
.anon_list
);
2985 if (freezing(current
)) {
2989 set_current_state(TASK_INTERRUPTIBLE
);
2992 __set_current_state(TASK_RUNNING
);
2994 } while (!kthread_should_stop());
2996 jfs_info("jfs_sync being killed");
3000 #if defined(CONFIG_PROC_FS) && defined(CONFIG_JFS_DEBUG)
3001 int jfs_txanchor_read(char *buffer
, char **start
, off_t offset
, int length
,
3002 int *eof
, void *data
)
3011 waitqueue_active(&TxAnchor
.freewait
) ? "active" : "empty";
3013 waitqueue_active(&TxAnchor
.freelockwait
) ? "active" : "empty";
3015 waitqueue_active(&TxAnchor
.lowlockwait
) ? "active" : "empty";
3017 len
+= sprintf(buffer
,
3023 "freelockwait = %s\n"
3024 "lowlockwait = %s\n"
3025 "tlocksInUse = %d\n"
3026 "jfs_tlocks_low = %d\n"
3027 "unlock_queue is %sempty\n",
3033 TxAnchor
.tlocksInUse
,
3035 list_empty(&TxAnchor
.unlock_queue
) ? "" : "not ");
3038 *start
= buffer
+ begin
;
3053 #if defined(CONFIG_PROC_FS) && defined(CONFIG_JFS_STATISTICS)
3054 int jfs_txstats_read(char *buffer
, char **start
, off_t offset
, int length
,
3055 int *eof
, void *data
)
3060 len
+= sprintf(buffer
,
3063 "calls to txBegin = %d\n"
3064 "txBegin blocked by sync barrier = %d\n"
3065 "txBegin blocked by tlocks low = %d\n"
3066 "txBegin blocked by no free tid = %d\n"
3067 "calls to txBeginAnon = %d\n"
3068 "txBeginAnon blocked by sync barrier = %d\n"
3069 "txBeginAnon blocked by tlocks low = %d\n"
3070 "calls to txLockAlloc = %d\n"
3071 "tLockAlloc blocked by no free lock = %d\n",
3073 TxStat
.txBegin_barrier
,
3074 TxStat
.txBegin_lockslow
,
3075 TxStat
.txBegin_freetid
,
3077 TxStat
.txBeginAnon_barrier
,
3078 TxStat
.txBeginAnon_lockslow
,
3080 TxStat
.txLockAlloc_freelock
);
3083 *start
= buffer
+ begin
;