2 * Copyright (C) International Business Machines Corp., 2000-2004
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_logmgr.c: log manager
23 * for related information, see transaction manager (jfs_txnmgr.c), and
24 * recovery manager (jfs_logredo.c).
26 * note: for detail, RTFS.
29 * special purpose buffer manager supporting log i/o requirements.
30 * per log serial pageout of logpage
31 * queuing i/o requests and redrive i/o at iodone
32 * maintain current logpage buffer
33 * no caching since append only
34 * appropriate jfs buffer cache buffers as needed
37 * transactions which wrote COMMIT records in the same in-memory
38 * log page during the pageout of previous/current log page(s) are
39 * committed together by the pageout of the page.
42 * transactions are committed asynchronously when the log page
43 * containing it COMMIT is paged out when it becomes full;
46 * . a per log lock serialize log write.
47 * . a per log lock serialize group commit.
48 * . a per log lock serialize log open/close;
51 * careful-write (ping-pong) of last logpage to recover from crash
53 * detection of split (out-of-order) write of physical sectors
54 * of last logpage via timestamp at end of each sector
55 * with its mirror data array at trailer).
58 * lsn - 64-bit monotonically increasing integer vs
59 * 32-bit lspn and page eor.
63 #include <linux/blkdev.h>
64 #include <linux/interrupt.h>
65 #include <linux/completion.h>
66 #include <linux/kthread.h>
67 #include <linux/buffer_head.h> /* for sync_blockdev() */
68 #include <linux/bio.h>
69 #include <linux/freezer.h>
70 #include <linux/delay.h>
71 #include <linux/mutex.h>
72 #include "jfs_incore.h"
73 #include "jfs_filsys.h"
74 #include "jfs_metapage.h"
75 #include "jfs_superblock.h"
76 #include "jfs_txnmgr.h"
77 #include "jfs_debug.h"
81 * lbuf's ready to be redriven. Protected by log_redrive_lock (jfsIO thread)
83 static struct lbuf
*log_redrive_list
;
84 static DEFINE_SPINLOCK(log_redrive_lock
);
88 * log read/write serialization (per log)
90 #define LOG_LOCK_INIT(log) mutex_init(&(log)->loglock)
91 #define LOG_LOCK(log) mutex_lock(&((log)->loglock))
92 #define LOG_UNLOCK(log) mutex_unlock(&((log)->loglock))
96 * log group commit serialization (per log)
99 #define LOGGC_LOCK_INIT(log) spin_lock_init(&(log)->gclock)
100 #define LOGGC_LOCK(log) spin_lock_irq(&(log)->gclock)
101 #define LOGGC_UNLOCK(log) spin_unlock_irq(&(log)->gclock)
102 #define LOGGC_WAKEUP(tblk) wake_up_all(&(tblk)->gcwait)
105 * log sync serialization (per log)
107 #define LOGSYNC_DELTA(logsize) min((logsize)/8, 128*LOGPSIZE)
108 #define LOGSYNC_BARRIER(logsize) ((logsize)/4)
110 #define LOGSYNC_DELTA(logsize) min((logsize)/4, 256*LOGPSIZE)
111 #define LOGSYNC_BARRIER(logsize) ((logsize)/2)
116 * log buffer cache synchronization
118 static DEFINE_SPINLOCK(jfsLCacheLock
);
120 #define LCACHE_LOCK(flags) spin_lock_irqsave(&jfsLCacheLock, flags)
121 #define LCACHE_UNLOCK(flags) spin_unlock_irqrestore(&jfsLCacheLock, flags)
124 * See __SLEEP_COND in jfs_locks.h
126 #define LCACHE_SLEEP_COND(wq, cond, flags) \
130 __SLEEP_COND(wq, cond, LCACHE_LOCK(flags), LCACHE_UNLOCK(flags)); \
133 #define LCACHE_WAKEUP(event) wake_up(event)
137 * lbuf buffer cache (lCache) control
139 /* log buffer manager pageout control (cumulative, inclusive) */
140 #define lbmREAD 0x0001
141 #define lbmWRITE 0x0002 /* enqueue at tail of write queue;
142 * init pageout if at head of queue;
144 #define lbmRELEASE 0x0004 /* remove from write queue
145 * at completion of pageout;
146 * do not free/recycle it yet:
147 * caller will free it;
149 #define lbmSYNC 0x0008 /* do not return to freelist
150 * when removed from write queue;
152 #define lbmFREE 0x0010 /* return to freelist
153 * at completion of pageout;
154 * the buffer may be recycled;
156 #define lbmDONE 0x0020
157 #define lbmERROR 0x0040
158 #define lbmGC 0x0080 /* lbmIODone to perform post-GC processing
161 #define lbmDIRECT 0x0100
164 * Global list of active external journals
166 static LIST_HEAD(jfs_external_logs
);
167 static struct jfs_log
*dummy_log
= NULL
;
168 static DEFINE_MUTEX(jfs_log_mutex
);
173 static int lmWriteRecord(struct jfs_log
* log
, struct tblock
* tblk
,
174 struct lrd
* lrd
, struct tlock
* tlck
);
176 static int lmNextPage(struct jfs_log
* log
);
177 static int lmLogFileSystem(struct jfs_log
* log
, struct jfs_sb_info
*sbi
,
180 static int open_inline_log(struct super_block
*sb
);
181 static int open_dummy_log(struct super_block
*sb
);
182 static int lbmLogInit(struct jfs_log
* log
);
183 static void lbmLogShutdown(struct jfs_log
* log
);
184 static struct lbuf
*lbmAllocate(struct jfs_log
* log
, int);
185 static void lbmFree(struct lbuf
* bp
);
186 static void lbmfree(struct lbuf
* bp
);
187 static int lbmRead(struct jfs_log
* log
, int pn
, struct lbuf
** bpp
);
188 static void lbmWrite(struct jfs_log
* log
, struct lbuf
* bp
, int flag
, int cant_block
);
189 static void lbmDirectWrite(struct jfs_log
* log
, struct lbuf
* bp
, int flag
);
190 static int lbmIOWait(struct lbuf
* bp
, int flag
);
191 static bio_end_io_t lbmIODone
;
192 static void lbmStartIO(struct lbuf
* bp
);
193 static void lmGCwrite(struct jfs_log
* log
, int cant_block
);
194 static int lmLogSync(struct jfs_log
* log
, int hard_sync
);
201 #ifdef CONFIG_JFS_STATISTICS
202 static struct lmStat
{
203 uint commit
; /* # of commit */
204 uint pagedone
; /* # of page written */
205 uint submitted
; /* # of pages submitted */
206 uint full_page
; /* # of full pages submitted */
207 uint partial_page
; /* # of partial pages submitted */
211 static void write_special_inodes(struct jfs_log
*log
,
212 int (*writer
)(struct address_space
*))
214 struct jfs_sb_info
*sbi
;
216 list_for_each_entry(sbi
, &log
->sb_list
, log_list
) {
217 writer(sbi
->ipbmap
->i_mapping
);
218 writer(sbi
->ipimap
->i_mapping
);
219 writer(sbi
->direct_inode
->i_mapping
);
226 * FUNCTION: write a log record;
230 * RETURN: lsn - offset to the next log record to write (end-of-log);
233 * note: todo: log error handler
235 int lmLog(struct jfs_log
* log
, struct tblock
* tblk
, struct lrd
* lrd
,
240 struct metapage
*mp
= NULL
;
243 jfs_info("lmLog: log:0x%p tblk:0x%p, lrd:0x%p tlck:0x%p",
244 log
, tblk
, lrd
, tlck
);
248 /* log by (out-of-transaction) JFS ? */
252 /* log from page ? */
254 tlck
->type
& tlckBTROOT
|| (mp
= tlck
->mp
) == NULL
)
258 * initialize/update page/transaction recovery lsn
262 LOGSYNC_LOCK(log
, flags
);
265 * initialize page lsn if first log write of the page
272 /* insert page at tail of logsynclist */
273 list_add_tail(&mp
->synclist
, &log
->synclist
);
277 * initialize/update lsn of tblock of the page
279 * transaction inherits oldest lsn of pages associated
280 * with allocation/deallocation of resources (their
281 * log records are used to reconstruct allocation map
282 * at recovery time: inode for inode allocation map,
283 * B+-tree index of extent descriptors for block
285 * allocation map pages inherit transaction lsn at
286 * commit time to allow forwarding log syncpt past log
287 * records associated with allocation/deallocation of
288 * resources only after persistent map of these map pages
289 * have been updated and propagated to home.
292 * initialize transaction lsn:
294 if (tblk
->lsn
== 0) {
295 /* inherit lsn of its first page logged */
299 /* insert tblock after the page on logsynclist */
300 list_add(&tblk
->synclist
, &mp
->synclist
);
303 * update transaction lsn:
306 /* inherit oldest/smallest lsn of page */
307 logdiff(diffp
, mp
->lsn
, log
);
308 logdiff(difft
, tblk
->lsn
, log
);
310 /* update tblock lsn with page lsn */
313 /* move tblock after page on logsynclist */
314 list_move(&tblk
->synclist
, &mp
->synclist
);
318 LOGSYNC_UNLOCK(log
, flags
);
321 * write the log record
324 lsn
= lmWriteRecord(log
, tblk
, lrd
, tlck
);
327 * forward log syncpt if log reached next syncpt trigger
329 logdiff(diffp
, lsn
, log
);
330 if (diffp
>= log
->nextsync
)
331 lsn
= lmLogSync(log
, 0);
333 /* update end-of-log lsn */
338 /* return end-of-log address */
343 * NAME: lmWriteRecord()
345 * FUNCTION: move the log record to current log page
347 * PARAMETER: cd - commit descriptor
349 * RETURN: end-of-log address
351 * serialization: LOG_LOCK() held on entry/exit
354 lmWriteRecord(struct jfs_log
* log
, struct tblock
* tblk
, struct lrd
* lrd
,
357 int lsn
= 0; /* end-of-log address */
358 struct lbuf
*bp
; /* dst log page buffer */
359 struct logpage
*lp
; /* dst log page */
360 caddr_t dst
; /* destination address in log page */
361 int dstoffset
; /* end-of-log offset in log page */
362 int freespace
; /* free space in log page */
363 caddr_t p
; /* src meta-data page */
366 int nbytes
; /* number of bytes to move */
369 struct linelock
*linelock
;
376 /* retrieve destination log page to write */
377 bp
= (struct lbuf
*) log
->bp
;
378 lp
= (struct logpage
*) bp
->l_ldata
;
379 dstoffset
= log
->eor
;
381 /* any log data to write ? */
386 * move log record data
388 /* retrieve source meta-data page to log */
389 if (tlck
->flag
& tlckPAGELOCK
) {
390 p
= (caddr_t
) (tlck
->mp
->data
);
391 linelock
= (struct linelock
*) & tlck
->lock
;
393 /* retrieve source in-memory inode to log */
394 else if (tlck
->flag
& tlckINODELOCK
) {
395 if (tlck
->type
& tlckDTREE
)
396 p
= (caddr_t
) &JFS_IP(tlck
->ip
)->i_dtroot
;
398 p
= (caddr_t
) &JFS_IP(tlck
->ip
)->i_xtroot
;
399 linelock
= (struct linelock
*) & tlck
->lock
;
402 else if (tlck
->flag
& tlckINLINELOCK
) {
404 inlinelock
= (struct inlinelock
*) & tlck
;
405 p
= (caddr_t
) & inlinelock
->pxd
;
406 linelock
= (struct linelock
*) & tlck
;
408 #endif /* _JFS_WIP */
410 jfs_err("lmWriteRecord: UFO tlck:0x%p", tlck
);
411 return 0; /* Probably should trap */
413 l2linesize
= linelock
->l2linesize
;
416 ASSERT(linelock
->index
<= linelock
->maxcnt
);
419 for (i
= 0; i
< linelock
->index
; i
++, lv
++) {
424 if (dstoffset
>= LOGPSIZE
- LOGPTLRSIZE
) {
425 /* page become full: move on to next page */
429 lp
= (struct logpage
*) bp
->l_ldata
;
430 dstoffset
= LOGPHDRSIZE
;
434 * move log vector data
436 src
= (u8
*) p
+ (lv
->offset
<< l2linesize
);
437 srclen
= lv
->length
<< l2linesize
;
440 freespace
= (LOGPSIZE
- LOGPTLRSIZE
) - dstoffset
;
441 nbytes
= min(freespace
, srclen
);
442 dst
= (caddr_t
) lp
+ dstoffset
;
443 memcpy(dst
, src
, nbytes
);
446 /* is page not full ? */
447 if (dstoffset
< LOGPSIZE
- LOGPTLRSIZE
)
450 /* page become full: move on to next page */
453 bp
= (struct lbuf
*) log
->bp
;
454 lp
= (struct logpage
*) bp
->l_ldata
;
455 dstoffset
= LOGPHDRSIZE
;
462 * move log vector descriptor
465 lvd
= (struct lvd
*) ((caddr_t
) lp
+ dstoffset
);
466 lvd
->offset
= cpu_to_le16(lv
->offset
);
467 lvd
->length
= cpu_to_le16(lv
->length
);
469 jfs_info("lmWriteRecord: lv offset:%d length:%d",
470 lv
->offset
, lv
->length
);
473 if ((i
= linelock
->next
)) {
474 linelock
= (struct linelock
*) lid_to_tlock(i
);
479 * move log record descriptor
482 lrd
->length
= cpu_to_le16(len
);
488 freespace
= (LOGPSIZE
- LOGPTLRSIZE
) - dstoffset
;
489 nbytes
= min(freespace
, srclen
);
490 dst
= (caddr_t
) lp
+ dstoffset
;
491 memcpy(dst
, src
, nbytes
);
496 /* are there more to move than freespace of page ? */
501 * end of log record descriptor
504 /* update last log record eor */
505 log
->eor
= dstoffset
;
506 bp
->l_eor
= dstoffset
;
507 lsn
= (log
->page
<< L2LOGPSIZE
) + dstoffset
;
509 if (lrd
->type
& cpu_to_le16(LOG_COMMIT
)) {
511 jfs_info("wr: tclsn:0x%x, beor:0x%x", tblk
->clsn
,
514 INCREMENT(lmStat
.commit
); /* # of commit */
517 * enqueue tblock for group commit:
519 * enqueue tblock of non-trivial/synchronous COMMIT
520 * at tail of group commit queue
521 * (trivial/asynchronous COMMITs are ignored by
526 /* init tblock gc state */
527 tblk
->flag
= tblkGC_QUEUE
;
529 tblk
->pn
= log
->page
;
530 tblk
->eor
= log
->eor
;
532 /* enqueue transaction to commit queue */
533 list_add_tail(&tblk
->cqueue
, &log
->cqueue
);
538 jfs_info("lmWriteRecord: lrd:0x%04x bp:0x%p pn:%d eor:0x%x",
539 le16_to_cpu(lrd
->type
), log
->bp
, log
->page
, dstoffset
);
541 /* page not full ? */
542 if (dstoffset
< LOGPSIZE
- LOGPTLRSIZE
)
546 /* page become full: move on to next page */
549 bp
= (struct lbuf
*) log
->bp
;
550 lp
= (struct logpage
*) bp
->l_ldata
;
551 dstoffset
= LOGPHDRSIZE
;
562 * FUNCTION: write current page and allocate next page.
568 * serialization: LOG_LOCK() held on entry/exit
570 static int lmNextPage(struct jfs_log
* log
)
573 int lspn
; /* log sequence page number */
574 int pn
; /* current page number */
579 /* get current log page number and log sequence page number */
582 lp
= (struct logpage
*) bp
->l_ldata
;
583 lspn
= le32_to_cpu(lp
->h
.page
);
588 * write or queue the full page at the tail of write queue
590 /* get the tail tblk on commit queue */
591 if (list_empty(&log
->cqueue
))
594 tblk
= list_entry(log
->cqueue
.prev
, struct tblock
, cqueue
);
596 /* every tblk who has COMMIT record on the current page,
597 * and has not been committed, must be on commit queue
598 * since tblk is queued at commit queueu at the time
599 * of writing its COMMIT record on the page before
600 * page becomes full (even though the tblk thread
601 * who wrote COMMIT record may have been suspended
605 /* is page bound with outstanding tail tblk ? */
606 if (tblk
&& tblk
->pn
== pn
) {
607 /* mark tblk for end-of-page */
608 tblk
->flag
|= tblkGC_EOP
;
610 if (log
->cflag
& logGC_PAGEOUT
) {
611 /* if page is not already on write queue,
612 * just enqueue (no lbmWRITE to prevent redrive)
613 * buffer to wqueue to ensure correct serial order
614 * of the pages since log pages will be added
617 if (bp
->l_wqnext
== NULL
)
618 lbmWrite(log
, bp
, 0, 0);
621 * No current GC leader, initiate group commit
623 log
->cflag
|= logGC_PAGEOUT
;
627 /* page is not bound with outstanding tblk:
628 * init write or mark it to be redriven (lbmWRITE)
631 /* finalize the page */
632 bp
->l_ceor
= bp
->l_eor
;
633 lp
->h
.eor
= lp
->t
.eor
= cpu_to_le16(bp
->l_ceor
);
634 lbmWrite(log
, bp
, lbmWRITE
| lbmRELEASE
| lbmFREE
, 0);
639 * allocate/initialize next page
641 /* if log wraps, the first data page of log is 2
642 * (0 never used, 1 is superblock).
644 log
->page
= (pn
== log
->size
- 1) ? 2 : pn
+ 1;
645 log
->eor
= LOGPHDRSIZE
; /* ? valid page empty/full at logRedo() */
647 /* allocate/initialize next log page buffer */
648 nextbp
= lbmAllocate(log
, log
->page
);
649 nextbp
->l_eor
= log
->eor
;
652 /* initialize next log page */
653 lp
= (struct logpage
*) nextbp
->l_ldata
;
654 lp
->h
.page
= lp
->t
.page
= cpu_to_le32(lspn
+ 1);
655 lp
->h
.eor
= lp
->t
.eor
= cpu_to_le16(LOGPHDRSIZE
);
662 * NAME: lmGroupCommit()
664 * FUNCTION: group commit
665 * initiate pageout of the pages with COMMIT in the order of
666 * page number - redrive pageout of the page at the head of
667 * pageout queue until full page has been written.
672 * LOGGC_LOCK serializes log group commit queue, and
673 * transaction blocks on the commit queue.
674 * N.B. LOG_LOCK is NOT held during lmGroupCommit().
676 int lmGroupCommit(struct jfs_log
* log
, struct tblock
* tblk
)
682 /* group committed already ? */
683 if (tblk
->flag
& tblkGC_COMMITTED
) {
684 if (tblk
->flag
& tblkGC_ERROR
)
690 jfs_info("lmGroup Commit: tblk = 0x%p, gcrtc = %d", tblk
, log
->gcrtc
);
692 if (tblk
->xflag
& COMMIT_LAZY
)
693 tblk
->flag
|= tblkGC_LAZY
;
695 if ((!(log
->cflag
& logGC_PAGEOUT
)) && (!list_empty(&log
->cqueue
)) &&
696 (!(tblk
->xflag
& COMMIT_LAZY
) || test_bit(log_FLUSH
, &log
->flag
)
697 || jfs_tlocks_low
)) {
699 * No pageout in progress
701 * start group commit as its group leader.
703 log
->cflag
|= logGC_PAGEOUT
;
708 if (tblk
->xflag
& COMMIT_LAZY
) {
710 * Lazy transactions can leave now
716 /* lmGCwrite gives up LOGGC_LOCK, check again */
718 if (tblk
->flag
& tblkGC_COMMITTED
) {
719 if (tblk
->flag
& tblkGC_ERROR
)
726 /* upcount transaction waiting for completion
729 tblk
->flag
|= tblkGC_READY
;
731 __SLEEP_COND(tblk
->gcwait
, (tblk
->flag
& tblkGC_COMMITTED
),
732 LOGGC_LOCK(log
), LOGGC_UNLOCK(log
));
734 /* removed from commit queue */
735 if (tblk
->flag
& tblkGC_ERROR
)
745 * FUNCTION: group commit write
746 * initiate write of log page, building a group of all transactions
747 * with commit records on that page.
752 * LOGGC_LOCK must be held by caller.
753 * N.B. LOG_LOCK is NOT held during lmGroupCommit().
755 static void lmGCwrite(struct jfs_log
* log
, int cant_write
)
759 int gcpn
; /* group commit page number */
761 struct tblock
*xtblk
= NULL
;
764 * build the commit group of a log page
766 * scan commit queue and make a commit group of all
767 * transactions with COMMIT records on the same log page.
769 /* get the head tblk on the commit queue */
770 gcpn
= list_entry(log
->cqueue
.next
, struct tblock
, cqueue
)->pn
;
772 list_for_each_entry(tblk
, &log
->cqueue
, cqueue
) {
773 if (tblk
->pn
!= gcpn
)
778 /* state transition: (QUEUE, READY) -> COMMIT */
779 tblk
->flag
|= tblkGC_COMMIT
;
781 tblk
= xtblk
; /* last tblk of the page */
784 * pageout to commit transactions on the log page.
786 bp
= (struct lbuf
*) tblk
->bp
;
787 lp
= (struct logpage
*) bp
->l_ldata
;
788 /* is page already full ? */
789 if (tblk
->flag
& tblkGC_EOP
) {
790 /* mark page to free at end of group commit of the page */
791 tblk
->flag
&= ~tblkGC_EOP
;
792 tblk
->flag
|= tblkGC_FREE
;
793 bp
->l_ceor
= bp
->l_eor
;
794 lp
->h
.eor
= lp
->t
.eor
= cpu_to_le16(bp
->l_ceor
);
795 lbmWrite(log
, bp
, lbmWRITE
| lbmRELEASE
| lbmGC
,
797 INCREMENT(lmStat
.full_page
);
799 /* page is not yet full */
801 bp
->l_ceor
= tblk
->eor
; /* ? bp->l_ceor = bp->l_eor; */
802 lp
->h
.eor
= lp
->t
.eor
= cpu_to_le16(bp
->l_ceor
);
803 lbmWrite(log
, bp
, lbmWRITE
| lbmGC
, cant_write
);
804 INCREMENT(lmStat
.partial_page
);
811 * FUNCTION: group commit post-processing
812 * Processes transactions after their commit records have been written
813 * to disk, redriving log I/O if necessary.
818 * This routine is called a interrupt time by lbmIODone
820 static void lmPostGC(struct lbuf
* bp
)
823 struct jfs_log
*log
= bp
->l_log
;
825 struct tblock
*tblk
, *temp
;
828 spin_lock_irqsave(&log
->gclock
, flags
);
830 * current pageout of group commit completed.
832 * remove/wakeup transactions from commit queue who were
833 * group committed with the current log page
835 list_for_each_entry_safe(tblk
, temp
, &log
->cqueue
, cqueue
) {
836 if (!(tblk
->flag
& tblkGC_COMMIT
))
838 /* if transaction was marked GC_COMMIT then
839 * it has been shipped in the current pageout
840 * and made it to disk - it is committed.
843 if (bp
->l_flag
& lbmERROR
)
844 tblk
->flag
|= tblkGC_ERROR
;
846 /* remove it from the commit queue */
847 list_del(&tblk
->cqueue
);
848 tblk
->flag
&= ~tblkGC_QUEUE
;
850 if (tblk
== log
->flush_tblk
) {
851 /* we can stop flushing the log now */
852 clear_bit(log_FLUSH
, &log
->flag
);
853 log
->flush_tblk
= NULL
;
856 jfs_info("lmPostGC: tblk = 0x%p, flag = 0x%x", tblk
,
859 if (!(tblk
->xflag
& COMMIT_FORCE
))
861 * Hand tblk over to lazy commit thread
865 /* state transition: COMMIT -> COMMITTED */
866 tblk
->flag
|= tblkGC_COMMITTED
;
868 if (tblk
->flag
& tblkGC_READY
)
874 /* was page full before pageout ?
875 * (and this is the last tblk bound with the page)
877 if (tblk
->flag
& tblkGC_FREE
)
879 /* did page become full after pageout ?
880 * (and this is the last tblk bound with the page)
882 else if (tblk
->flag
& tblkGC_EOP
) {
883 /* finalize the page */
884 lp
= (struct logpage
*) bp
->l_ldata
;
885 bp
->l_ceor
= bp
->l_eor
;
886 lp
->h
.eor
= lp
->t
.eor
= cpu_to_le16(bp
->l_eor
);
887 jfs_info("lmPostGC: calling lbmWrite");
888 lbmWrite(log
, bp
, lbmWRITE
| lbmRELEASE
| lbmFREE
,
894 /* are there any transactions who have entered lnGroupCommit()
895 * (whose COMMITs are after that of the last log page written.
896 * They are waiting for new group commit (above at (SLEEP 1))
897 * or lazy transactions are on a full (queued) log page,
898 * select the latest ready transaction as new group leader and
899 * wake her up to lead her group.
901 if ((!list_empty(&log
->cqueue
)) &&
902 ((log
->gcrtc
> 0) || (tblk
->bp
->l_wqnext
!= NULL
) ||
903 test_bit(log_FLUSH
, &log
->flag
) || jfs_tlocks_low
))
905 * Call lmGCwrite with new group leader
909 /* no transaction are ready yet (transactions are only just
910 * queued (GC_QUEUE) and not entered for group commit yet).
911 * the first transaction entering group commit
912 * will elect herself as new group leader.
915 log
->cflag
&= ~logGC_PAGEOUT
;
918 spin_unlock_irqrestore(&log
->gclock
, flags
);
925 * FUNCTION: write log SYNCPT record for specified log
926 * if new sync address is available
927 * (normally the case if sync() is executed by back-ground
929 * calculate new value of i_nextsync which determines when
930 * this code is called again.
932 * PARAMETERS: log - log structure
933 * hard_sync - 1 to force all metadata to be written
937 * serialization: LOG_LOCK() held on entry/exit
939 static int lmLogSync(struct jfs_log
* log
, int hard_sync
)
942 int written
; /* written since last syncpt */
943 int free
; /* free space left available */
944 int delta
; /* additional delta to write normally */
945 int more
; /* additional write granted */
948 struct logsyncblk
*lp
;
951 /* push dirty metapages out to disk */
953 write_special_inodes(log
, filemap_fdatawrite
);
955 write_special_inodes(log
, filemap_flush
);
960 /* if last sync is same as last syncpt,
961 * invoke sync point forward processing to update sync.
964 if (log
->sync
== log
->syncpt
) {
965 LOGSYNC_LOCK(log
, flags
);
966 if (list_empty(&log
->synclist
))
967 log
->sync
= log
->lsn
;
969 lp
= list_entry(log
->synclist
.next
,
970 struct logsyncblk
, synclist
);
973 LOGSYNC_UNLOCK(log
, flags
);
977 /* if sync is different from last syncpt,
978 * write a SYNCPT record with syncpt = sync.
979 * reset syncpt = sync
981 if (log
->sync
!= log
->syncpt
) {
984 lrd
.type
= cpu_to_le16(LOG_SYNCPT
);
986 lrd
.log
.syncpt
.sync
= cpu_to_le32(log
->sync
);
987 lsn
= lmWriteRecord(log
, NULL
, &lrd
, NULL
);
989 log
->syncpt
= log
->sync
;
994 * setup next syncpt trigger (SWAG)
996 logsize
= log
->logsize
;
998 logdiff(written
, lsn
, log
);
999 free
= logsize
- written
;
1000 delta
= LOGSYNC_DELTA(logsize
);
1001 more
= min(free
/ 2, delta
);
1002 if (more
< 2 * LOGPSIZE
) {
1003 jfs_warn("\n ... Log Wrap ... Log Wrap ... Log Wrap ...\n");
1007 * option 1 - panic ? No.!
1008 * option 2 - shutdown file systems
1009 * associated with log ?
1010 * option 3 - extend log ?
1013 * option 4 - second chance
1015 * mark log wrapped, and continue.
1016 * when all active transactions are completed,
1017 * mark log vaild for recovery.
1018 * if crashed during invalid state, log state
1019 * implies invald log, forcing fsck().
1021 /* mark log state log wrap in log superblock */
1022 /* log->state = LOGWRAP; */
1024 /* reset sync point computation */
1025 log
->syncpt
= log
->sync
= lsn
;
1026 log
->nextsync
= delta
;
1028 /* next syncpt trigger = written + more */
1029 log
->nextsync
= written
+ more
;
1031 /* if number of bytes written from last sync point is more
1032 * than 1/4 of the log size, stop new transactions from
1033 * starting until all current transactions are completed
1034 * by setting syncbarrier flag.
1036 if (!test_bit(log_SYNCBARRIER
, &log
->flag
) &&
1037 (written
> LOGSYNC_BARRIER(logsize
)) && log
->active
) {
1038 set_bit(log_SYNCBARRIER
, &log
->flag
);
1039 jfs_info("log barrier on: lsn=0x%x syncpt=0x%x", lsn
,
1042 * We may have to initiate group commit
1044 jfs_flush_journal(log
, 0);
1053 * FUNCTION: write log SYNCPT record for specified log
1055 * PARAMETERS: log - log structure
1056 * hard_sync - set to 1 to force metadata to be written
1058 void jfs_syncpt(struct jfs_log
*log
, int hard_sync
)
1060 lmLogSync(log
, hard_sync
);
1067 * FUNCTION: open the log on first open;
1068 * insert filesystem in the active list of the log.
1070 * PARAMETER: ipmnt - file system mount inode
1071 * iplog - log inode (out)
1077 int lmLogOpen(struct super_block
*sb
)
1080 struct block_device
*bdev
;
1081 struct jfs_log
*log
;
1082 struct jfs_sb_info
*sbi
= JFS_SBI(sb
);
1084 if (sbi
->flag
& JFS_NOINTEGRITY
)
1085 return open_dummy_log(sb
);
1087 if (sbi
->mntflag
& JFS_INLINELOG
)
1088 return open_inline_log(sb
);
1090 mutex_lock(&jfs_log_mutex
);
1091 list_for_each_entry(log
, &jfs_external_logs
, journal_list
) {
1092 if (log
->bdev
->bd_dev
== sbi
->logdev
) {
1093 if (memcmp(log
->uuid
, sbi
->loguuid
,
1094 sizeof(log
->uuid
))) {
1095 jfs_warn("wrong uuid on JFS journal\n");
1096 mutex_unlock(&jfs_log_mutex
);
1100 * add file system to log active file system list
1102 if ((rc
= lmLogFileSystem(log
, sbi
, 1))) {
1103 mutex_unlock(&jfs_log_mutex
);
1110 if (!(log
= kzalloc(sizeof(struct jfs_log
), GFP_KERNEL
))) {
1111 mutex_unlock(&jfs_log_mutex
);
1114 INIT_LIST_HEAD(&log
->sb_list
);
1115 init_waitqueue_head(&log
->syncwait
);
1118 * external log as separate logical volume
1120 * file systems to log may have n-to-1 relationship;
1123 bdev
= open_by_devnum(sbi
->logdev
, FMODE_READ
|FMODE_WRITE
);
1125 rc
= -PTR_ERR(bdev
);
1129 if ((rc
= bd_claim(bdev
, log
))) {
1134 memcpy(log
->uuid
, sbi
->loguuid
, sizeof(log
->uuid
));
1139 if ((rc
= lmLogInit(log
)))
1142 list_add(&log
->journal_list
, &jfs_external_logs
);
1145 * add file system to log active file system list
1147 if ((rc
= lmLogFileSystem(log
, sbi
, 1)))
1152 list_add(&sbi
->log_list
, &log
->sb_list
);
1156 mutex_unlock(&jfs_log_mutex
);
1162 shutdown
: /* unwind lbmLogInit() */
1163 list_del(&log
->journal_list
);
1164 lbmLogShutdown(log
);
1169 close
: /* close external log device */
1172 free
: /* free log descriptor */
1173 mutex_unlock(&jfs_log_mutex
);
1176 jfs_warn("lmLogOpen: exit(%d)", rc
);
1180 static int open_inline_log(struct super_block
*sb
)
1182 struct jfs_log
*log
;
1185 if (!(log
= kzalloc(sizeof(struct jfs_log
), GFP_KERNEL
)))
1187 INIT_LIST_HEAD(&log
->sb_list
);
1188 init_waitqueue_head(&log
->syncwait
);
1190 set_bit(log_INLINELOG
, &log
->flag
);
1191 log
->bdev
= sb
->s_bdev
;
1192 log
->base
= addressPXD(&JFS_SBI(sb
)->logpxd
);
1193 log
->size
= lengthPXD(&JFS_SBI(sb
)->logpxd
) >>
1194 (L2LOGPSIZE
- sb
->s_blocksize_bits
);
1195 log
->l2bsize
= sb
->s_blocksize_bits
;
1196 ASSERT(L2LOGPSIZE
>= sb
->s_blocksize_bits
);
1201 if ((rc
= lmLogInit(log
))) {
1203 jfs_warn("lmLogOpen: exit(%d)", rc
);
1207 list_add(&JFS_SBI(sb
)->log_list
, &log
->sb_list
);
1208 JFS_SBI(sb
)->log
= log
;
1213 static int open_dummy_log(struct super_block
*sb
)
1217 mutex_lock(&jfs_log_mutex
);
1219 dummy_log
= kzalloc(sizeof(struct jfs_log
), GFP_KERNEL
);
1221 mutex_unlock(&jfs_log_mutex
);
1224 INIT_LIST_HEAD(&dummy_log
->sb_list
);
1225 init_waitqueue_head(&dummy_log
->syncwait
);
1226 dummy_log
->no_integrity
= 1;
1227 /* Make up some stuff */
1228 dummy_log
->base
= 0;
1229 dummy_log
->size
= 1024;
1230 rc
= lmLogInit(dummy_log
);
1234 mutex_unlock(&jfs_log_mutex
);
1239 LOG_LOCK(dummy_log
);
1240 list_add(&JFS_SBI(sb
)->log_list
, &dummy_log
->sb_list
);
1241 JFS_SBI(sb
)->log
= dummy_log
;
1242 LOG_UNLOCK(dummy_log
);
1243 mutex_unlock(&jfs_log_mutex
);
1251 * FUNCTION: log initialization at first log open.
1253 * logredo() (or logformat()) should have been run previously.
1254 * initialize the log from log superblock.
1255 * set the log state in the superblock to LOGMOUNT and
1256 * write SYNCPT log record.
1258 * PARAMETER: log - log structure
1261 * -EINVAL - bad log magic number or superblock dirty
1262 * error returned from logwait()
1264 * serialization: single first open thread
1266 int lmLogInit(struct jfs_log
* log
)
1270 struct logsuper
*logsuper
;
1271 struct lbuf
*bpsuper
;
1276 jfs_info("lmLogInit: log:0x%p", log
);
1278 /* initialize the group commit serialization lock */
1279 LOGGC_LOCK_INIT(log
);
1281 /* allocate/initialize the log write serialization lock */
1284 LOGSYNC_LOCK_INIT(log
);
1286 INIT_LIST_HEAD(&log
->synclist
);
1288 INIT_LIST_HEAD(&log
->cqueue
);
1289 log
->flush_tblk
= NULL
;
1294 * initialize log i/o
1296 if ((rc
= lbmLogInit(log
)))
1299 if (!test_bit(log_INLINELOG
, &log
->flag
))
1300 log
->l2bsize
= L2LOGPSIZE
;
1302 /* check for disabled journaling to disk */
1303 if (log
->no_integrity
) {
1305 * Journal pages will still be filled. When the time comes
1306 * to actually do the I/O, the write is not done, and the
1307 * endio routine is called directly.
1309 bp
= lbmAllocate(log
, 0);
1311 bp
->l_pn
= bp
->l_eor
= 0;
1314 * validate log superblock
1316 if ((rc
= lbmRead(log
, 1, &bpsuper
)))
1319 logsuper
= (struct logsuper
*) bpsuper
->l_ldata
;
1321 if (logsuper
->magic
!= cpu_to_le32(LOGMAGIC
)) {
1322 jfs_warn("*** Log Format Error ! ***");
1327 /* logredo() should have been run successfully. */
1328 if (logsuper
->state
!= cpu_to_le32(LOGREDONE
)) {
1329 jfs_warn("*** Log Is Dirty ! ***");
1334 /* initialize log from log superblock */
1335 if (test_bit(log_INLINELOG
,&log
->flag
)) {
1336 if (log
->size
!= le32_to_cpu(logsuper
->size
)) {
1340 jfs_info("lmLogInit: inline log:0x%p base:0x%Lx "
1342 (unsigned long long) log
->base
, log
->size
);
1344 if (memcmp(logsuper
->uuid
, log
->uuid
, 16)) {
1345 jfs_warn("wrong uuid on JFS log device");
1348 log
->size
= le32_to_cpu(logsuper
->size
);
1349 log
->l2bsize
= le32_to_cpu(logsuper
->l2bsize
);
1350 jfs_info("lmLogInit: external log:0x%p base:0x%Lx "
1352 (unsigned long long) log
->base
, log
->size
);
1355 log
->page
= le32_to_cpu(logsuper
->end
) / LOGPSIZE
;
1356 log
->eor
= le32_to_cpu(logsuper
->end
) - (LOGPSIZE
* log
->page
);
1359 * initialize for log append write mode
1361 /* establish current/end-of-log page/buffer */
1362 if ((rc
= lbmRead(log
, log
->page
, &bp
)))
1365 lp
= (struct logpage
*) bp
->l_ldata
;
1367 jfs_info("lmLogInit: lsn:0x%x page:%d eor:%d:%d",
1368 le32_to_cpu(logsuper
->end
), log
->page
, log
->eor
,
1369 le16_to_cpu(lp
->h
.eor
));
1372 bp
->l_pn
= log
->page
;
1373 bp
->l_eor
= log
->eor
;
1375 /* if current page is full, move on to next page */
1376 if (log
->eor
>= LOGPSIZE
- LOGPTLRSIZE
)
1380 * initialize log syncpoint
1383 * write the first SYNCPT record with syncpoint = 0
1384 * (i.e., log redo up to HERE !);
1385 * remove current page from lbm write queue at end of pageout
1386 * (to write log superblock update), but do not release to
1391 lrd
.type
= cpu_to_le16(LOG_SYNCPT
);
1393 lrd
.log
.syncpt
.sync
= 0;
1394 lsn
= lmWriteRecord(log
, NULL
, &lrd
, NULL
);
1396 bp
->l_ceor
= bp
->l_eor
;
1397 lp
= (struct logpage
*) bp
->l_ldata
;
1398 lp
->h
.eor
= lp
->t
.eor
= cpu_to_le16(bp
->l_eor
);
1399 lbmWrite(log
, bp
, lbmWRITE
| lbmSYNC
, 0);
1400 if ((rc
= lbmIOWait(bp
, 0)))
1404 * update/write superblock
1406 logsuper
->state
= cpu_to_le32(LOGMOUNT
);
1407 log
->serial
= le32_to_cpu(logsuper
->serial
) + 1;
1408 logsuper
->serial
= cpu_to_le32(log
->serial
);
1409 lbmDirectWrite(log
, bpsuper
, lbmWRITE
| lbmRELEASE
| lbmSYNC
);
1410 if ((rc
= lbmIOWait(bpsuper
, lbmFREE
)))
1414 /* initialize logsync parameters */
1415 log
->logsize
= (log
->size
- 2) << L2LOGPSIZE
;
1418 log
->sync
= log
->syncpt
;
1419 log
->nextsync
= LOGSYNC_DELTA(log
->logsize
);
1421 jfs_info("lmLogInit: lsn:0x%x syncpt:0x%x sync:0x%x",
1422 log
->lsn
, log
->syncpt
, log
->sync
);
1425 * initialize for lazy/group commit
1434 errout30
: /* release log page */
1436 bp
->l_wqnext
= NULL
;
1439 errout20
: /* release log superblock */
1442 errout10
: /* unwind lbmLogInit() */
1443 lbmLogShutdown(log
);
1445 jfs_warn("lmLogInit: exit(%d)", rc
);
1451 * NAME: lmLogClose()
1453 * FUNCTION: remove file system <ipmnt> from active list of log <iplog>
1454 * and close it on last close.
1456 * PARAMETER: sb - superblock
1458 * RETURN: errors from subroutines
1462 int lmLogClose(struct super_block
*sb
)
1464 struct jfs_sb_info
*sbi
= JFS_SBI(sb
);
1465 struct jfs_log
*log
= sbi
->log
;
1466 struct block_device
*bdev
;
1469 jfs_info("lmLogClose: log:0x%p", log
);
1471 mutex_lock(&jfs_log_mutex
);
1473 list_del(&sbi
->log_list
);
1478 * We need to make sure all of the "written" metapages
1479 * actually make it to disk
1481 sync_blockdev(sb
->s_bdev
);
1483 if (test_bit(log_INLINELOG
, &log
->flag
)) {
1485 * in-line log in host file system
1487 rc
= lmLogShutdown(log
);
1492 if (!log
->no_integrity
)
1493 lmLogFileSystem(log
, sbi
, 0);
1495 if (!list_empty(&log
->sb_list
))
1499 * TODO: ensure that the dummy_log is in a state to allow
1500 * lbmLogShutdown to deallocate all the buffers and call
1501 * kfree against dummy_log. For now, leave dummy_log & its
1502 * buffers in memory, and resuse if another no-integrity mount
1505 if (log
->no_integrity
)
1509 * external log as separate logical volume
1511 list_del(&log
->journal_list
);
1513 rc
= lmLogShutdown(log
);
1521 mutex_unlock(&jfs_log_mutex
);
1522 jfs_info("lmLogClose: exit(%d)", rc
);
1528 * NAME: jfs_flush_journal()
1530 * FUNCTION: initiate write of any outstanding transactions to the journal
1531 * and optionally wait until they are all written to disk
1533 * wait == 0 flush until latest txn is committed, don't wait
1534 * wait == 1 flush until latest txn is committed, wait
1535 * wait > 1 flush until all txn's are complete, wait
1537 void jfs_flush_journal(struct jfs_log
*log
, int wait
)
1540 struct tblock
*target
= NULL
;
1542 /* jfs_write_inode may call us during read-only mount */
1546 jfs_info("jfs_flush_journal: log:0x%p wait=%d", log
, wait
);
1550 if (!list_empty(&log
->cqueue
)) {
1552 * This ensures that we will keep writing to the journal as long
1553 * as there are unwritten commit records
1555 target
= list_entry(log
->cqueue
.prev
, struct tblock
, cqueue
);
1557 if (test_bit(log_FLUSH
, &log
->flag
)) {
1559 * We're already flushing.
1560 * if flush_tblk is NULL, we are flushing everything,
1561 * so leave it that way. Otherwise, update it to the
1562 * latest transaction
1564 if (log
->flush_tblk
)
1565 log
->flush_tblk
= target
;
1567 /* Only flush until latest transaction is committed */
1568 log
->flush_tblk
= target
;
1569 set_bit(log_FLUSH
, &log
->flag
);
1572 * Initiate I/O on outstanding transactions
1574 if (!(log
->cflag
& logGC_PAGEOUT
)) {
1575 log
->cflag
|= logGC_PAGEOUT
;
1580 if ((wait
> 1) || test_bit(log_SYNCBARRIER
, &log
->flag
)) {
1581 /* Flush until all activity complete */
1582 set_bit(log_FLUSH
, &log
->flag
);
1583 log
->flush_tblk
= NULL
;
1586 if (wait
&& target
&& !(target
->flag
& tblkGC_COMMITTED
)) {
1587 DECLARE_WAITQUEUE(__wait
, current
);
1589 add_wait_queue(&target
->gcwait
, &__wait
);
1590 set_current_state(TASK_UNINTERRUPTIBLE
);
1593 __set_current_state(TASK_RUNNING
);
1595 remove_wait_queue(&target
->gcwait
, &__wait
);
1602 write_special_inodes(log
, filemap_fdatawrite
);
1605 * If there was recent activity, we may need to wait
1606 * for the lazycommit thread to catch up
1608 if ((!list_empty(&log
->cqueue
)) || !list_empty(&log
->synclist
)) {
1609 for (i
= 0; i
< 200; i
++) { /* Too much? */
1611 write_special_inodes(log
, filemap_fdatawrite
);
1612 if (list_empty(&log
->cqueue
) &&
1613 list_empty(&log
->synclist
))
1617 assert(list_empty(&log
->cqueue
));
1619 #ifdef CONFIG_JFS_DEBUG
1620 if (!list_empty(&log
->synclist
)) {
1621 struct logsyncblk
*lp
;
1623 printk(KERN_ERR
"jfs_flush_journal: synclist not empty\n");
1624 list_for_each_entry(lp
, &log
->synclist
, synclist
) {
1625 if (lp
->xflag
& COMMIT_PAGE
) {
1626 struct metapage
*mp
= (struct metapage
*)lp
;
1627 print_hex_dump(KERN_ERR
, "metapage: ",
1628 DUMP_PREFIX_ADDRESS
, 16, 4,
1629 mp
, sizeof(struct metapage
), 0);
1630 print_hex_dump(KERN_ERR
, "page: ",
1631 DUMP_PREFIX_ADDRESS
, 16,
1632 sizeof(long), mp
->page
,
1633 sizeof(struct page
), 0);
1635 print_hex_dump(KERN_ERR
, "tblock:",
1636 DUMP_PREFIX_ADDRESS
, 16, 4,
1637 lp
, sizeof(struct tblock
), 0);
1641 WARN_ON(!list_empty(&log
->synclist
));
1643 clear_bit(log_FLUSH
, &log
->flag
);
1647 * NAME: lmLogShutdown()
1649 * FUNCTION: log shutdown at last LogClose().
1651 * write log syncpt record.
1652 * update super block to set redone flag to 0.
1654 * PARAMETER: log - log inode
1656 * RETURN: 0 - success
1658 * serialization: single last close thread
1660 int lmLogShutdown(struct jfs_log
* log
)
1665 struct logsuper
*logsuper
;
1666 struct lbuf
*bpsuper
;
1670 jfs_info("lmLogShutdown: log:0x%p", log
);
1672 jfs_flush_journal(log
, 2);
1675 * write the last SYNCPT record with syncpoint = 0
1676 * (i.e., log redo up to HERE !)
1680 lrd
.type
= cpu_to_le16(LOG_SYNCPT
);
1682 lrd
.log
.syncpt
.sync
= 0;
1684 lsn
= lmWriteRecord(log
, NULL
, &lrd
, NULL
);
1686 lp
= (struct logpage
*) bp
->l_ldata
;
1687 lp
->h
.eor
= lp
->t
.eor
= cpu_to_le16(bp
->l_eor
);
1688 lbmWrite(log
, log
->bp
, lbmWRITE
| lbmRELEASE
| lbmSYNC
, 0);
1689 lbmIOWait(log
->bp
, lbmFREE
);
1693 * synchronous update log superblock
1694 * mark log state as shutdown cleanly
1695 * (i.e., Log does not need to be replayed).
1697 if ((rc
= lbmRead(log
, 1, &bpsuper
)))
1700 logsuper
= (struct logsuper
*) bpsuper
->l_ldata
;
1701 logsuper
->state
= cpu_to_le32(LOGREDONE
);
1702 logsuper
->end
= cpu_to_le32(lsn
);
1703 lbmDirectWrite(log
, bpsuper
, lbmWRITE
| lbmRELEASE
| lbmSYNC
);
1704 rc
= lbmIOWait(bpsuper
, lbmFREE
);
1706 jfs_info("lmLogShutdown: lsn:0x%x page:%d eor:%d",
1707 lsn
, log
->page
, log
->eor
);
1711 * shutdown per log i/o
1713 lbmLogShutdown(log
);
1716 jfs_warn("lmLogShutdown: exit(%d)", rc
);
1723 * NAME: lmLogFileSystem()
1725 * FUNCTION: insert (<activate> = true)/remove (<activate> = false)
1726 * file system into/from log active file system list.
1728 * PARAMETE: log - pointer to logs inode.
1729 * fsdev - kdev_t of filesystem.
1730 * serial - pointer to returned log serial number
1731 * activate - insert/remove device from active list.
1733 * RETURN: 0 - success
1734 * errors returned by vms_iowait().
1736 static int lmLogFileSystem(struct jfs_log
* log
, struct jfs_sb_info
*sbi
,
1741 struct logsuper
*logsuper
;
1742 struct lbuf
*bpsuper
;
1743 char *uuid
= sbi
->uuid
;
1746 * insert/remove file system device to log active file system list.
1748 if ((rc
= lbmRead(log
, 1, &bpsuper
)))
1751 logsuper
= (struct logsuper
*) bpsuper
->l_ldata
;
1753 for (i
= 0; i
< MAX_ACTIVE
; i
++)
1754 if (!memcmp(logsuper
->active
[i
].uuid
, NULL_UUID
, 16)) {
1755 memcpy(logsuper
->active
[i
].uuid
, uuid
, 16);
1759 if (i
== MAX_ACTIVE
) {
1760 jfs_warn("Too many file systems sharing journal!");
1762 return -EMFILE
; /* Is there a better rc? */
1765 for (i
= 0; i
< MAX_ACTIVE
; i
++)
1766 if (!memcmp(logsuper
->active
[i
].uuid
, uuid
, 16)) {
1767 memcpy(logsuper
->active
[i
].uuid
, NULL_UUID
, 16);
1770 if (i
== MAX_ACTIVE
) {
1771 jfs_warn("Somebody stomped on the journal!");
1779 * synchronous write log superblock:
1781 * write sidestream bypassing write queue:
1782 * at file system mount, log super block is updated for
1783 * activation of the file system before any log record
1784 * (MOUNT record) of the file system, and at file system
1785 * unmount, all meta data for the file system has been
1786 * flushed before log super block is updated for deactivation
1787 * of the file system.
1789 lbmDirectWrite(log
, bpsuper
, lbmWRITE
| lbmRELEASE
| lbmSYNC
);
1790 rc
= lbmIOWait(bpsuper
, lbmFREE
);
1796 * log buffer manager (lbm)
1797 * ------------------------
1799 * special purpose buffer manager supporting log i/o requirements.
1801 * per log write queue:
1802 * log pageout occurs in serial order by fifo write queue and
1803 * restricting to a single i/o in pregress at any one time.
1804 * a circular singly-linked list
1805 * (log->wrqueue points to the tail, and buffers are linked via
1806 * bp->wrqueue field), and
1807 * maintains log page in pageout ot waiting for pageout in serial pageout.
1813 * initialize per log I/O setup at lmLogInit()
1815 static int lbmLogInit(struct jfs_log
* log
)
1820 jfs_info("lbmLogInit: log:0x%p", log
);
1822 /* initialize current buffer cursor */
1825 /* initialize log device write queue */
1829 * Each log has its own buffer pages allocated to it. These are
1830 * not managed by the page cache. This ensures that a transaction
1831 * writing to the log does not block trying to allocate a page from
1832 * the page cache (for the log). This would be bad, since page
1833 * allocation waits on the kswapd thread that may be committing inodes
1834 * which would cause log activity. Was that clear? I'm trying to
1835 * avoid deadlock here.
1837 init_waitqueue_head(&log
->free_wait
);
1839 log
->lbuf_free
= NULL
;
1841 for (i
= 0; i
< LOGPAGES
;) {
1846 buffer
= (char *) get_zeroed_page(GFP_KERNEL
);
1849 page
= virt_to_page(buffer
);
1850 for (offset
= 0; offset
< PAGE_SIZE
; offset
+= LOGPSIZE
) {
1851 lbuf
= kmalloc(sizeof(struct lbuf
), GFP_KERNEL
);
1854 free_page((unsigned long) buffer
);
1857 if (offset
) /* we already have one reference */
1859 lbuf
->l_offset
= offset
;
1860 lbuf
->l_ldata
= buffer
+ offset
;
1861 lbuf
->l_page
= page
;
1863 init_waitqueue_head(&lbuf
->l_ioevent
);
1865 lbuf
->l_freelist
= log
->lbuf_free
;
1866 log
->lbuf_free
= lbuf
;
1874 lbmLogShutdown(log
);
1882 * finalize per log I/O setup at lmLogShutdown()
1884 static void lbmLogShutdown(struct jfs_log
* log
)
1888 jfs_info("lbmLogShutdown: log:0x%p", log
);
1890 lbuf
= log
->lbuf_free
;
1892 struct lbuf
*next
= lbuf
->l_freelist
;
1893 __free_page(lbuf
->l_page
);
1903 * allocate an empty log buffer
1905 static struct lbuf
*lbmAllocate(struct jfs_log
* log
, int pn
)
1908 unsigned long flags
;
1911 * recycle from log buffer freelist if any
1914 LCACHE_SLEEP_COND(log
->free_wait
, (bp
= log
->lbuf_free
), flags
);
1915 log
->lbuf_free
= bp
->l_freelist
;
1916 LCACHE_UNLOCK(flags
);
1920 bp
->l_wqnext
= NULL
;
1921 bp
->l_freelist
= NULL
;
1924 bp
->l_blkno
= log
->base
+ (pn
<< (L2LOGPSIZE
- log
->l2bsize
));
1934 * release a log buffer to freelist
1936 static void lbmFree(struct lbuf
* bp
)
1938 unsigned long flags
;
1944 LCACHE_UNLOCK(flags
);
1947 static void lbmfree(struct lbuf
* bp
)
1949 struct jfs_log
*log
= bp
->l_log
;
1951 assert(bp
->l_wqnext
== NULL
);
1954 * return the buffer to head of freelist
1956 bp
->l_freelist
= log
->lbuf_free
;
1957 log
->lbuf_free
= bp
;
1959 wake_up(&log
->free_wait
);
1967 * FUNCTION: add a log buffer to the log redrive list
1973 * Takes log_redrive_lock.
1975 static inline void lbmRedrive(struct lbuf
*bp
)
1977 unsigned long flags
;
1979 spin_lock_irqsave(&log_redrive_lock
, flags
);
1980 bp
->l_redrive_next
= log_redrive_list
;
1981 log_redrive_list
= bp
;
1982 spin_unlock_irqrestore(&log_redrive_lock
, flags
);
1984 wake_up_process(jfsIOthread
);
1991 static int lbmRead(struct jfs_log
* log
, int pn
, struct lbuf
** bpp
)
1997 * allocate a log buffer
1999 *bpp
= bp
= lbmAllocate(log
, pn
);
2000 jfs_info("lbmRead: bp:0x%p pn:0x%x", bp
, pn
);
2002 bp
->l_flag
|= lbmREAD
;
2004 bio
= bio_alloc(GFP_NOFS
, 1);
2006 bio
->bi_sector
= bp
->l_blkno
<< (log
->l2bsize
- 9);
2007 bio
->bi_bdev
= log
->bdev
;
2008 bio
->bi_io_vec
[0].bv_page
= bp
->l_page
;
2009 bio
->bi_io_vec
[0].bv_len
= LOGPSIZE
;
2010 bio
->bi_io_vec
[0].bv_offset
= bp
->l_offset
;
2014 bio
->bi_size
= LOGPSIZE
;
2016 bio
->bi_end_io
= lbmIODone
;
2017 bio
->bi_private
= bp
;
2018 submit_bio(READ_SYNC
, bio
);
2020 wait_event(bp
->l_ioevent
, (bp
->l_flag
!= lbmREAD
));
2029 * buffer at head of pageout queue stays after completion of
2030 * partial-page pageout and redriven by explicit initiation of
2031 * pageout by caller until full-page pageout is completed and
2034 * device driver i/o done redrives pageout of new buffer at
2035 * head of pageout queue when current buffer at head of pageout
2036 * queue is released at the completion of its full-page pageout.
2038 * LOGGC_LOCK() serializes lbmWrite() by lmNextPage() and lmGroupCommit().
2039 * LCACHE_LOCK() serializes xflag between lbmWrite() and lbmIODone()
2041 static void lbmWrite(struct jfs_log
* log
, struct lbuf
* bp
, int flag
,
2045 unsigned long flags
;
2047 jfs_info("lbmWrite: bp:0x%p flag:0x%x pn:0x%x", bp
, flag
, bp
->l_pn
);
2049 /* map the logical block address to physical block address */
2051 log
->base
+ (bp
->l_pn
<< (L2LOGPSIZE
- log
->l2bsize
));
2053 LCACHE_LOCK(flags
); /* disable+lock */
2056 * initialize buffer for device driver
2061 * insert bp at tail of write queue associated with log
2063 * (request is either for bp already/currently at head of queue
2064 * or new bp to be inserted at tail)
2068 /* is buffer not already on write queue ? */
2069 if (bp
->l_wqnext
== NULL
) {
2070 /* insert at tail of wqueue */
2076 bp
->l_wqnext
= tail
->l_wqnext
;
2077 tail
->l_wqnext
= bp
;
2083 /* is buffer at head of wqueue and for write ? */
2084 if ((bp
!= tail
->l_wqnext
) || !(flag
& lbmWRITE
)) {
2085 LCACHE_UNLOCK(flags
); /* unlock+enable */
2089 LCACHE_UNLOCK(flags
); /* unlock+enable */
2093 else if (flag
& lbmSYNC
)
2106 * initiate pageout bypassing write queue for sidestream
2107 * (e.g., log superblock) write;
2109 static void lbmDirectWrite(struct jfs_log
* log
, struct lbuf
* bp
, int flag
)
2111 jfs_info("lbmDirectWrite: bp:0x%p flag:0x%x pn:0x%x",
2112 bp
, flag
, bp
->l_pn
);
2115 * initialize buffer for device driver
2117 bp
->l_flag
= flag
| lbmDIRECT
;
2119 /* map the logical block address to physical block address */
2121 log
->base
+ (bp
->l_pn
<< (L2LOGPSIZE
- log
->l2bsize
));
2124 * initiate pageout of the page
2131 * NAME: lbmStartIO()
2133 * FUNCTION: Interface to DD strategy routine
2137 * serialization: LCACHE_LOCK() is NOT held during log i/o;
2139 static void lbmStartIO(struct lbuf
* bp
)
2142 struct jfs_log
*log
= bp
->l_log
;
2144 jfs_info("lbmStartIO\n");
2146 bio
= bio_alloc(GFP_NOFS
, 1);
2147 bio
->bi_sector
= bp
->l_blkno
<< (log
->l2bsize
- 9);
2148 bio
->bi_bdev
= log
->bdev
;
2149 bio
->bi_io_vec
[0].bv_page
= bp
->l_page
;
2150 bio
->bi_io_vec
[0].bv_len
= LOGPSIZE
;
2151 bio
->bi_io_vec
[0].bv_offset
= bp
->l_offset
;
2155 bio
->bi_size
= LOGPSIZE
;
2157 bio
->bi_end_io
= lbmIODone
;
2158 bio
->bi_private
= bp
;
2160 /* check if journaling to disk has been disabled */
2161 if (log
->no_integrity
) {
2165 submit_bio(WRITE_SYNC
, bio
);
2166 INCREMENT(lmStat
.submitted
);
2174 static int lbmIOWait(struct lbuf
* bp
, int flag
)
2176 unsigned long flags
;
2179 jfs_info("lbmIOWait1: bp:0x%p flag:0x%x:0x%x", bp
, bp
->l_flag
, flag
);
2181 LCACHE_LOCK(flags
); /* disable+lock */
2183 LCACHE_SLEEP_COND(bp
->l_ioevent
, (bp
->l_flag
& lbmDONE
), flags
);
2185 rc
= (bp
->l_flag
& lbmERROR
) ? -EIO
: 0;
2190 LCACHE_UNLOCK(flags
); /* unlock+enable */
2192 jfs_info("lbmIOWait2: bp:0x%p flag:0x%x:0x%x", bp
, bp
->l_flag
, flag
);
2199 * executed at INTIODONE level
2201 static void lbmIODone(struct bio
*bio
, int error
)
2203 struct lbuf
*bp
= bio
->bi_private
;
2204 struct lbuf
*nextbp
, *tail
;
2205 struct jfs_log
*log
;
2206 unsigned long flags
;
2209 * get back jfs buffer bound to the i/o buffer
2211 jfs_info("lbmIODone: bp:0x%p flag:0x%x", bp
, bp
->l_flag
);
2213 LCACHE_LOCK(flags
); /* disable+lock */
2215 bp
->l_flag
|= lbmDONE
;
2217 if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
)) {
2218 bp
->l_flag
|= lbmERROR
;
2220 jfs_err("lbmIODone: I/O error in JFS log");
2228 if (bp
->l_flag
& lbmREAD
) {
2229 bp
->l_flag
&= ~lbmREAD
;
2231 LCACHE_UNLOCK(flags
); /* unlock+enable */
2233 /* wakeup I/O initiator */
2234 LCACHE_WAKEUP(&bp
->l_ioevent
);
2240 * pageout completion
2242 * the bp at the head of write queue has completed pageout.
2244 * if single-commit/full-page pageout, remove the current buffer
2245 * from head of pageout queue, and redrive pageout with
2246 * the new buffer at head of pageout queue;
2247 * otherwise, the partial-page pageout buffer stays at
2248 * the head of pageout queue to be redriven for pageout
2249 * by lmGroupCommit() until full-page pageout is completed.
2251 bp
->l_flag
&= ~lbmWRITE
;
2252 INCREMENT(lmStat
.pagedone
);
2254 /* update committed lsn */
2256 log
->clsn
= (bp
->l_pn
<< L2LOGPSIZE
) + bp
->l_ceor
;
2258 if (bp
->l_flag
& lbmDIRECT
) {
2259 LCACHE_WAKEUP(&bp
->l_ioevent
);
2260 LCACHE_UNLOCK(flags
);
2266 /* single element queue */
2268 /* remove head buffer of full-page pageout
2269 * from log device write queue
2271 if (bp
->l_flag
& lbmRELEASE
) {
2273 bp
->l_wqnext
= NULL
;
2276 /* multi element queue */
2278 /* remove head buffer of full-page pageout
2279 * from log device write queue
2281 if (bp
->l_flag
& lbmRELEASE
) {
2282 nextbp
= tail
->l_wqnext
= bp
->l_wqnext
;
2283 bp
->l_wqnext
= NULL
;
2286 * redrive pageout of next page at head of write queue:
2287 * redrive next page without any bound tblk
2288 * (i.e., page w/o any COMMIT records), or
2289 * first page of new group commit which has been
2290 * queued after current page (subsequent pageout
2291 * is performed synchronously, except page without
2292 * any COMMITs) by lmGroupCommit() as indicated
2295 if (nextbp
->l_flag
& lbmWRITE
) {
2297 * We can't do the I/O at interrupt time.
2298 * The jfsIO thread can do it
2306 * synchronous pageout:
2308 * buffer has not necessarily been removed from write queue
2309 * (e.g., synchronous write of partial-page with COMMIT):
2310 * leave buffer for i/o initiator to dispose
2312 if (bp
->l_flag
& lbmSYNC
) {
2313 LCACHE_UNLOCK(flags
); /* unlock+enable */
2315 /* wakeup I/O initiator */
2316 LCACHE_WAKEUP(&bp
->l_ioevent
);
2320 * Group Commit pageout:
2322 else if (bp
->l_flag
& lbmGC
) {
2323 LCACHE_UNLOCK(flags
);
2328 * asynchronous pageout:
2330 * buffer must have been removed from write queue:
2331 * insert buffer at head of freelist where it can be recycled
2334 assert(bp
->l_flag
& lbmRELEASE
);
2335 assert(bp
->l_flag
& lbmFREE
);
2338 LCACHE_UNLOCK(flags
); /* unlock+enable */
2342 int jfsIOWait(void *arg
)
2347 spin_lock_irq(&log_redrive_lock
);
2348 while ((bp
= log_redrive_list
)) {
2349 log_redrive_list
= bp
->l_redrive_next
;
2350 bp
->l_redrive_next
= NULL
;
2351 spin_unlock_irq(&log_redrive_lock
);
2353 spin_lock_irq(&log_redrive_lock
);
2356 if (freezing(current
)) {
2357 spin_unlock_irq(&log_redrive_lock
);
2360 set_current_state(TASK_INTERRUPTIBLE
);
2361 spin_unlock_irq(&log_redrive_lock
);
2363 __set_current_state(TASK_RUNNING
);
2365 } while (!kthread_should_stop());
2367 jfs_info("jfsIOWait being killed!");
2372 * NAME: lmLogFormat()/jfs_logform()
2374 * FUNCTION: format file system log
2378 * logAddress - start address of log space in FS block
2379 * logSize - length of log space in FS block;
2381 * RETURN: 0 - success
2384 * XXX: We're synchronously writing one page at a time. This needs to
2385 * be improved by writing multiple pages at once.
2387 int lmLogFormat(struct jfs_log
*log
, s64 logAddress
, int logSize
)
2390 struct jfs_sb_info
*sbi
;
2391 struct logsuper
*logsuper
;
2393 int lspn
; /* log sequence page number */
2394 struct lrd
*lrd_ptr
;
2398 jfs_info("lmLogFormat: logAddress:%Ld logSize:%d",
2399 (long long)logAddress
, logSize
);
2401 sbi
= list_entry(log
->sb_list
.next
, struct jfs_sb_info
, log_list
);
2403 /* allocate a log buffer */
2404 bp
= lbmAllocate(log
, 1);
2406 npages
= logSize
>> sbi
->l2nbperpage
;
2411 * page 0 - reserved;
2412 * page 1 - log superblock;
2413 * page 2 - log data page: A SYNC log record is written
2414 * into this page at logform time;
2415 * pages 3-N - log data page: set to empty log data pages;
2418 * init log superblock: log page 1
2420 logsuper
= (struct logsuper
*) bp
->l_ldata
;
2422 logsuper
->magic
= cpu_to_le32(LOGMAGIC
);
2423 logsuper
->version
= cpu_to_le32(LOGVERSION
);
2424 logsuper
->state
= cpu_to_le32(LOGREDONE
);
2425 logsuper
->flag
= cpu_to_le32(sbi
->mntflag
); /* ? */
2426 logsuper
->size
= cpu_to_le32(npages
);
2427 logsuper
->bsize
= cpu_to_le32(sbi
->bsize
);
2428 logsuper
->l2bsize
= cpu_to_le32(sbi
->l2bsize
);
2429 logsuper
->end
= cpu_to_le32(2 * LOGPSIZE
+ LOGPHDRSIZE
+ LOGRDSIZE
);
2431 bp
->l_flag
= lbmWRITE
| lbmSYNC
| lbmDIRECT
;
2432 bp
->l_blkno
= logAddress
+ sbi
->nbperpage
;
2434 if ((rc
= lbmIOWait(bp
, 0)))
2438 * init pages 2 to npages-1 as log data pages:
2440 * log page sequence number (lpsn) initialization:
2443 * +-----+-----+=====+=====+===.....===+=====+
2445 * <--- N page circular file ---->
2447 * the N (= npages-2) data pages of the log is maintained as
2448 * a circular file for the log records;
2449 * lpsn grows by 1 monotonically as each log page is written
2450 * to the circular file of the log;
2451 * and setLogpage() will not reset the page number even if
2452 * the eor is equal to LOGPHDRSIZE. In order for binary search
2453 * still work in find log end process, we have to simulate the
2454 * log wrap situation at the log format time.
2455 * The 1st log page written will have the highest lpsn. Then
2456 * the succeeding log pages will have ascending order of
2457 * the lspn starting from 0, ... (N-2)
2459 lp
= (struct logpage
*) bp
->l_ldata
;
2461 * initialize 1st log page to be written: lpsn = N - 1,
2462 * write a SYNCPT log record is written to this page
2464 lp
->h
.page
= lp
->t
.page
= cpu_to_le32(npages
- 3);
2465 lp
->h
.eor
= lp
->t
.eor
= cpu_to_le16(LOGPHDRSIZE
+ LOGRDSIZE
);
2467 lrd_ptr
= (struct lrd
*) &lp
->data
;
2468 lrd_ptr
->logtid
= 0;
2469 lrd_ptr
->backchain
= 0;
2470 lrd_ptr
->type
= cpu_to_le16(LOG_SYNCPT
);
2471 lrd_ptr
->length
= 0;
2472 lrd_ptr
->log
.syncpt
.sync
= 0;
2474 bp
->l_blkno
+= sbi
->nbperpage
;
2475 bp
->l_flag
= lbmWRITE
| lbmSYNC
| lbmDIRECT
;
2477 if ((rc
= lbmIOWait(bp
, 0)))
2481 * initialize succeeding log pages: lpsn = 0, 1, ..., (N-2)
2483 for (lspn
= 0; lspn
< npages
- 3; lspn
++) {
2484 lp
->h
.page
= lp
->t
.page
= cpu_to_le32(lspn
);
2485 lp
->h
.eor
= lp
->t
.eor
= cpu_to_le16(LOGPHDRSIZE
);
2487 bp
->l_blkno
+= sbi
->nbperpage
;
2488 bp
->l_flag
= lbmWRITE
| lbmSYNC
| lbmDIRECT
;
2490 if ((rc
= lbmIOWait(bp
, 0)))
2499 /* release the buffer */
2505 #ifdef CONFIG_JFS_STATISTICS
2506 int jfs_lmstats_read(char *buffer
, char **start
, off_t offset
, int length
,
2507 int *eof
, void *data
)
2512 len
+= sprintf(buffer
,
2513 "JFS Logmgr stats\n"
2514 "================\n"
2516 "writes submitted = %d\n"
2517 "writes completed = %d\n"
2518 "full pages submitted = %d\n"
2519 "partial pages submitted = %d\n",
2524 lmStat
.partial_page
);
2527 *start
= buffer
+ begin
;
2540 #endif /* CONFIG_JFS_STATISTICS */
This page took 0.07833 seconds and 6 git commands to generate.