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/export.h>
71 #include <linux/delay.h>
72 #include <linux/mutex.h>
73 #include <linux/seq_file.h>
74 #include <linux/slab.h>
75 #include "jfs_incore.h"
76 #include "jfs_filsys.h"
77 #include "jfs_metapage.h"
78 #include "jfs_superblock.h"
79 #include "jfs_txnmgr.h"
80 #include "jfs_debug.h"
84 * lbuf's ready to be redriven. Protected by log_redrive_lock (jfsIO thread)
86 static struct lbuf
*log_redrive_list
;
87 static DEFINE_SPINLOCK(log_redrive_lock
);
91 * log read/write serialization (per log)
93 #define LOG_LOCK_INIT(log) mutex_init(&(log)->loglock)
94 #define LOG_LOCK(log) mutex_lock(&((log)->loglock))
95 #define LOG_UNLOCK(log) mutex_unlock(&((log)->loglock))
99 * log group commit serialization (per log)
102 #define LOGGC_LOCK_INIT(log) spin_lock_init(&(log)->gclock)
103 #define LOGGC_LOCK(log) spin_lock_irq(&(log)->gclock)
104 #define LOGGC_UNLOCK(log) spin_unlock_irq(&(log)->gclock)
105 #define LOGGC_WAKEUP(tblk) wake_up_all(&(tblk)->gcwait)
108 * log sync serialization (per log)
110 #define LOGSYNC_DELTA(logsize) min((logsize)/8, 128*LOGPSIZE)
111 #define LOGSYNC_BARRIER(logsize) ((logsize)/4)
113 #define LOGSYNC_DELTA(logsize) min((logsize)/4, 256*LOGPSIZE)
114 #define LOGSYNC_BARRIER(logsize) ((logsize)/2)
119 * log buffer cache synchronization
121 static DEFINE_SPINLOCK(jfsLCacheLock
);
123 #define LCACHE_LOCK(flags) spin_lock_irqsave(&jfsLCacheLock, flags)
124 #define LCACHE_UNLOCK(flags) spin_unlock_irqrestore(&jfsLCacheLock, flags)
127 * See __SLEEP_COND in jfs_locks.h
129 #define LCACHE_SLEEP_COND(wq, cond, flags) \
133 __SLEEP_COND(wq, cond, LCACHE_LOCK(flags), LCACHE_UNLOCK(flags)); \
136 #define LCACHE_WAKEUP(event) wake_up(event)
140 * lbuf buffer cache (lCache) control
142 /* log buffer manager pageout control (cumulative, inclusive) */
143 #define lbmREAD 0x0001
144 #define lbmWRITE 0x0002 /* enqueue at tail of write queue;
145 * init pageout if at head of queue;
147 #define lbmRELEASE 0x0004 /* remove from write queue
148 * at completion of pageout;
149 * do not free/recycle it yet:
150 * caller will free it;
152 #define lbmSYNC 0x0008 /* do not return to freelist
153 * when removed from write queue;
155 #define lbmFREE 0x0010 /* return to freelist
156 * at completion of pageout;
157 * the buffer may be recycled;
159 #define lbmDONE 0x0020
160 #define lbmERROR 0x0040
161 #define lbmGC 0x0080 /* lbmIODone to perform post-GC processing
164 #define lbmDIRECT 0x0100
167 * Global list of active external journals
169 static LIST_HEAD(jfs_external_logs
);
170 static struct jfs_log
*dummy_log
;
171 static DEFINE_MUTEX(jfs_log_mutex
);
176 static int lmWriteRecord(struct jfs_log
* log
, struct tblock
* tblk
,
177 struct lrd
* lrd
, struct tlock
* tlck
);
179 static int lmNextPage(struct jfs_log
* log
);
180 static int lmLogFileSystem(struct jfs_log
* log
, struct jfs_sb_info
*sbi
,
183 static int open_inline_log(struct super_block
*sb
);
184 static int open_dummy_log(struct super_block
*sb
);
185 static int lbmLogInit(struct jfs_log
* log
);
186 static void lbmLogShutdown(struct jfs_log
* log
);
187 static struct lbuf
*lbmAllocate(struct jfs_log
* log
, int);
188 static void lbmFree(struct lbuf
* bp
);
189 static void lbmfree(struct lbuf
* bp
);
190 static int lbmRead(struct jfs_log
* log
, int pn
, struct lbuf
** bpp
);
191 static void lbmWrite(struct jfs_log
* log
, struct lbuf
* bp
, int flag
, int cant_block
);
192 static void lbmDirectWrite(struct jfs_log
* log
, struct lbuf
* bp
, int flag
);
193 static int lbmIOWait(struct lbuf
* bp
, int flag
);
194 static bio_end_io_t lbmIODone
;
195 static void lbmStartIO(struct lbuf
* bp
);
196 static void lmGCwrite(struct jfs_log
* log
, int cant_block
);
197 static int lmLogSync(struct jfs_log
* log
, int hard_sync
);
204 #ifdef CONFIG_JFS_STATISTICS
205 static struct lmStat
{
206 uint commit
; /* # of commit */
207 uint pagedone
; /* # of page written */
208 uint submitted
; /* # of pages submitted */
209 uint full_page
; /* # of full pages submitted */
210 uint partial_page
; /* # of partial pages submitted */
214 static void write_special_inodes(struct jfs_log
*log
,
215 int (*writer
)(struct address_space
*))
217 struct jfs_sb_info
*sbi
;
219 list_for_each_entry(sbi
, &log
->sb_list
, log_list
) {
220 writer(sbi
->ipbmap
->i_mapping
);
221 writer(sbi
->ipimap
->i_mapping
);
222 writer(sbi
->direct_inode
->i_mapping
);
229 * FUNCTION: write a log record;
233 * RETURN: lsn - offset to the next log record to write (end-of-log);
236 * note: todo: log error handler
238 int lmLog(struct jfs_log
* log
, struct tblock
* tblk
, struct lrd
* lrd
,
243 struct metapage
*mp
= NULL
;
246 jfs_info("lmLog: log:0x%p tblk:0x%p, lrd:0x%p tlck:0x%p",
247 log
, tblk
, lrd
, tlck
);
251 /* log by (out-of-transaction) JFS ? */
255 /* log from page ? */
257 tlck
->type
& tlckBTROOT
|| (mp
= tlck
->mp
) == NULL
)
261 * initialize/update page/transaction recovery lsn
265 LOGSYNC_LOCK(log
, flags
);
268 * initialize page lsn if first log write of the page
275 /* insert page at tail of logsynclist */
276 list_add_tail(&mp
->synclist
, &log
->synclist
);
280 * initialize/update lsn of tblock of the page
282 * transaction inherits oldest lsn of pages associated
283 * with allocation/deallocation of resources (their
284 * log records are used to reconstruct allocation map
285 * at recovery time: inode for inode allocation map,
286 * B+-tree index of extent descriptors for block
288 * allocation map pages inherit transaction lsn at
289 * commit time to allow forwarding log syncpt past log
290 * records associated with allocation/deallocation of
291 * resources only after persistent map of these map pages
292 * have been updated and propagated to home.
295 * initialize transaction lsn:
297 if (tblk
->lsn
== 0) {
298 /* inherit lsn of its first page logged */
302 /* insert tblock after the page on logsynclist */
303 list_add(&tblk
->synclist
, &mp
->synclist
);
306 * update transaction lsn:
309 /* inherit oldest/smallest lsn of page */
310 logdiff(diffp
, mp
->lsn
, log
);
311 logdiff(difft
, tblk
->lsn
, log
);
313 /* update tblock lsn with page lsn */
316 /* move tblock after page on logsynclist */
317 list_move(&tblk
->synclist
, &mp
->synclist
);
321 LOGSYNC_UNLOCK(log
, flags
);
324 * write the log record
327 lsn
= lmWriteRecord(log
, tblk
, lrd
, tlck
);
330 * forward log syncpt if log reached next syncpt trigger
332 logdiff(diffp
, lsn
, log
);
333 if (diffp
>= log
->nextsync
)
334 lsn
= lmLogSync(log
, 0);
336 /* update end-of-log lsn */
341 /* return end-of-log address */
346 * NAME: lmWriteRecord()
348 * FUNCTION: move the log record to current log page
350 * PARAMETER: cd - commit descriptor
352 * RETURN: end-of-log address
354 * serialization: LOG_LOCK() held on entry/exit
357 lmWriteRecord(struct jfs_log
* log
, struct tblock
* tblk
, struct lrd
* lrd
,
360 int lsn
= 0; /* end-of-log address */
361 struct lbuf
*bp
; /* dst log page buffer */
362 struct logpage
*lp
; /* dst log page */
363 caddr_t dst
; /* destination address in log page */
364 int dstoffset
; /* end-of-log offset in log page */
365 int freespace
; /* free space in log page */
366 caddr_t p
; /* src meta-data page */
369 int nbytes
; /* number of bytes to move */
372 struct linelock
*linelock
;
379 /* retrieve destination log page to write */
380 bp
= (struct lbuf
*) log
->bp
;
381 lp
= (struct logpage
*) bp
->l_ldata
;
382 dstoffset
= log
->eor
;
384 /* any log data to write ? */
389 * move log record data
391 /* retrieve source meta-data page to log */
392 if (tlck
->flag
& tlckPAGELOCK
) {
393 p
= (caddr_t
) (tlck
->mp
->data
);
394 linelock
= (struct linelock
*) & tlck
->lock
;
396 /* retrieve source in-memory inode to log */
397 else if (tlck
->flag
& tlckINODELOCK
) {
398 if (tlck
->type
& tlckDTREE
)
399 p
= (caddr_t
) &JFS_IP(tlck
->ip
)->i_dtroot
;
401 p
= (caddr_t
) &JFS_IP(tlck
->ip
)->i_xtroot
;
402 linelock
= (struct linelock
*) & tlck
->lock
;
405 else if (tlck
->flag
& tlckINLINELOCK
) {
407 inlinelock
= (struct inlinelock
*) & tlck
;
408 p
= (caddr_t
) & inlinelock
->pxd
;
409 linelock
= (struct linelock
*) & tlck
;
411 #endif /* _JFS_WIP */
413 jfs_err("lmWriteRecord: UFO tlck:0x%p", tlck
);
414 return 0; /* Probably should trap */
416 l2linesize
= linelock
->l2linesize
;
419 ASSERT(linelock
->index
<= linelock
->maxcnt
);
422 for (i
= 0; i
< linelock
->index
; i
++, lv
++) {
427 if (dstoffset
>= LOGPSIZE
- LOGPTLRSIZE
) {
428 /* page become full: move on to next page */
432 lp
= (struct logpage
*) bp
->l_ldata
;
433 dstoffset
= LOGPHDRSIZE
;
437 * move log vector data
439 src
= (u8
*) p
+ (lv
->offset
<< l2linesize
);
440 srclen
= lv
->length
<< l2linesize
;
443 freespace
= (LOGPSIZE
- LOGPTLRSIZE
) - dstoffset
;
444 nbytes
= min(freespace
, srclen
);
445 dst
= (caddr_t
) lp
+ dstoffset
;
446 memcpy(dst
, src
, nbytes
);
449 /* is page not full ? */
450 if (dstoffset
< LOGPSIZE
- LOGPTLRSIZE
)
453 /* page become full: move on to next page */
456 bp
= (struct lbuf
*) log
->bp
;
457 lp
= (struct logpage
*) bp
->l_ldata
;
458 dstoffset
= LOGPHDRSIZE
;
465 * move log vector descriptor
468 lvd
= (struct lvd
*) ((caddr_t
) lp
+ dstoffset
);
469 lvd
->offset
= cpu_to_le16(lv
->offset
);
470 lvd
->length
= cpu_to_le16(lv
->length
);
472 jfs_info("lmWriteRecord: lv offset:%d length:%d",
473 lv
->offset
, lv
->length
);
476 if ((i
= linelock
->next
)) {
477 linelock
= (struct linelock
*) lid_to_tlock(i
);
482 * move log record descriptor
485 lrd
->length
= cpu_to_le16(len
);
491 freespace
= (LOGPSIZE
- LOGPTLRSIZE
) - dstoffset
;
492 nbytes
= min(freespace
, srclen
);
493 dst
= (caddr_t
) lp
+ dstoffset
;
494 memcpy(dst
, src
, nbytes
);
499 /* are there more to move than freespace of page ? */
504 * end of log record descriptor
507 /* update last log record eor */
508 log
->eor
= dstoffset
;
509 bp
->l_eor
= dstoffset
;
510 lsn
= (log
->page
<< L2LOGPSIZE
) + dstoffset
;
512 if (lrd
->type
& cpu_to_le16(LOG_COMMIT
)) {
514 jfs_info("wr: tclsn:0x%x, beor:0x%x", tblk
->clsn
,
517 INCREMENT(lmStat
.commit
); /* # of commit */
520 * enqueue tblock for group commit:
522 * enqueue tblock of non-trivial/synchronous COMMIT
523 * at tail of group commit queue
524 * (trivial/asynchronous COMMITs are ignored by
529 /* init tblock gc state */
530 tblk
->flag
= tblkGC_QUEUE
;
532 tblk
->pn
= log
->page
;
533 tblk
->eor
= log
->eor
;
535 /* enqueue transaction to commit queue */
536 list_add_tail(&tblk
->cqueue
, &log
->cqueue
);
541 jfs_info("lmWriteRecord: lrd:0x%04x bp:0x%p pn:%d eor:0x%x",
542 le16_to_cpu(lrd
->type
), log
->bp
, log
->page
, dstoffset
);
544 /* page not full ? */
545 if (dstoffset
< LOGPSIZE
- LOGPTLRSIZE
)
549 /* page become full: move on to next page */
552 bp
= (struct lbuf
*) log
->bp
;
553 lp
= (struct logpage
*) bp
->l_ldata
;
554 dstoffset
= LOGPHDRSIZE
;
565 * FUNCTION: write current page and allocate next page.
571 * serialization: LOG_LOCK() held on entry/exit
573 static int lmNextPage(struct jfs_log
* log
)
576 int lspn
; /* log sequence page number */
577 int pn
; /* current page number */
582 /* get current log page number and log sequence page number */
585 lp
= (struct logpage
*) bp
->l_ldata
;
586 lspn
= le32_to_cpu(lp
->h
.page
);
591 * write or queue the full page at the tail of write queue
593 /* get the tail tblk on commit queue */
594 if (list_empty(&log
->cqueue
))
597 tblk
= list_entry(log
->cqueue
.prev
, struct tblock
, cqueue
);
599 /* every tblk who has COMMIT record on the current page,
600 * and has not been committed, must be on commit queue
601 * since tblk is queued at commit queueu at the time
602 * of writing its COMMIT record on the page before
603 * page becomes full (even though the tblk thread
604 * who wrote COMMIT record may have been suspended
608 /* is page bound with outstanding tail tblk ? */
609 if (tblk
&& tblk
->pn
== pn
) {
610 /* mark tblk for end-of-page */
611 tblk
->flag
|= tblkGC_EOP
;
613 if (log
->cflag
& logGC_PAGEOUT
) {
614 /* if page is not already on write queue,
615 * just enqueue (no lbmWRITE to prevent redrive)
616 * buffer to wqueue to ensure correct serial order
617 * of the pages since log pages will be added
620 if (bp
->l_wqnext
== NULL
)
621 lbmWrite(log
, bp
, 0, 0);
624 * No current GC leader, initiate group commit
626 log
->cflag
|= logGC_PAGEOUT
;
630 /* page is not bound with outstanding tblk:
631 * init write or mark it to be redriven (lbmWRITE)
634 /* finalize the page */
635 bp
->l_ceor
= bp
->l_eor
;
636 lp
->h
.eor
= lp
->t
.eor
= cpu_to_le16(bp
->l_ceor
);
637 lbmWrite(log
, bp
, lbmWRITE
| lbmRELEASE
| lbmFREE
, 0);
642 * allocate/initialize next page
644 /* if log wraps, the first data page of log is 2
645 * (0 never used, 1 is superblock).
647 log
->page
= (pn
== log
->size
- 1) ? 2 : pn
+ 1;
648 log
->eor
= LOGPHDRSIZE
; /* ? valid page empty/full at logRedo() */
650 /* allocate/initialize next log page buffer */
651 nextbp
= lbmAllocate(log
, log
->page
);
652 nextbp
->l_eor
= log
->eor
;
655 /* initialize next log page */
656 lp
= (struct logpage
*) nextbp
->l_ldata
;
657 lp
->h
.page
= lp
->t
.page
= cpu_to_le32(lspn
+ 1);
658 lp
->h
.eor
= lp
->t
.eor
= cpu_to_le16(LOGPHDRSIZE
);
665 * NAME: lmGroupCommit()
667 * FUNCTION: group commit
668 * initiate pageout of the pages with COMMIT in the order of
669 * page number - redrive pageout of the page at the head of
670 * pageout queue until full page has been written.
675 * LOGGC_LOCK serializes log group commit queue, and
676 * transaction blocks on the commit queue.
677 * N.B. LOG_LOCK is NOT held during lmGroupCommit().
679 int lmGroupCommit(struct jfs_log
* log
, struct tblock
* tblk
)
685 /* group committed already ? */
686 if (tblk
->flag
& tblkGC_COMMITTED
) {
687 if (tblk
->flag
& tblkGC_ERROR
)
693 jfs_info("lmGroup Commit: tblk = 0x%p, gcrtc = %d", tblk
, log
->gcrtc
);
695 if (tblk
->xflag
& COMMIT_LAZY
)
696 tblk
->flag
|= tblkGC_LAZY
;
698 if ((!(log
->cflag
& logGC_PAGEOUT
)) && (!list_empty(&log
->cqueue
)) &&
699 (!(tblk
->xflag
& COMMIT_LAZY
) || test_bit(log_FLUSH
, &log
->flag
)
700 || jfs_tlocks_low
)) {
702 * No pageout in progress
704 * start group commit as its group leader.
706 log
->cflag
|= logGC_PAGEOUT
;
711 if (tblk
->xflag
& COMMIT_LAZY
) {
713 * Lazy transactions can leave now
719 /* lmGCwrite gives up LOGGC_LOCK, check again */
721 if (tblk
->flag
& tblkGC_COMMITTED
) {
722 if (tblk
->flag
& tblkGC_ERROR
)
729 /* upcount transaction waiting for completion
732 tblk
->flag
|= tblkGC_READY
;
734 __SLEEP_COND(tblk
->gcwait
, (tblk
->flag
& tblkGC_COMMITTED
),
735 LOGGC_LOCK(log
), LOGGC_UNLOCK(log
));
737 /* removed from commit queue */
738 if (tblk
->flag
& tblkGC_ERROR
)
748 * FUNCTION: group commit write
749 * initiate write of log page, building a group of all transactions
750 * with commit records on that page.
755 * LOGGC_LOCK must be held by caller.
756 * N.B. LOG_LOCK is NOT held during lmGroupCommit().
758 static void lmGCwrite(struct jfs_log
* log
, int cant_write
)
762 int gcpn
; /* group commit page number */
764 struct tblock
*xtblk
= NULL
;
767 * build the commit group of a log page
769 * scan commit queue and make a commit group of all
770 * transactions with COMMIT records on the same log page.
772 /* get the head tblk on the commit queue */
773 gcpn
= list_entry(log
->cqueue
.next
, struct tblock
, cqueue
)->pn
;
775 list_for_each_entry(tblk
, &log
->cqueue
, cqueue
) {
776 if (tblk
->pn
!= gcpn
)
781 /* state transition: (QUEUE, READY) -> COMMIT */
782 tblk
->flag
|= tblkGC_COMMIT
;
784 tblk
= xtblk
; /* last tblk of the page */
787 * pageout to commit transactions on the log page.
789 bp
= (struct lbuf
*) tblk
->bp
;
790 lp
= (struct logpage
*) bp
->l_ldata
;
791 /* is page already full ? */
792 if (tblk
->flag
& tblkGC_EOP
) {
793 /* mark page to free at end of group commit of the page */
794 tblk
->flag
&= ~tblkGC_EOP
;
795 tblk
->flag
|= tblkGC_FREE
;
796 bp
->l_ceor
= bp
->l_eor
;
797 lp
->h
.eor
= lp
->t
.eor
= cpu_to_le16(bp
->l_ceor
);
798 lbmWrite(log
, bp
, lbmWRITE
| lbmRELEASE
| lbmGC
,
800 INCREMENT(lmStat
.full_page
);
802 /* page is not yet full */
804 bp
->l_ceor
= tblk
->eor
; /* ? bp->l_ceor = bp->l_eor; */
805 lp
->h
.eor
= lp
->t
.eor
= cpu_to_le16(bp
->l_ceor
);
806 lbmWrite(log
, bp
, lbmWRITE
| lbmGC
, cant_write
);
807 INCREMENT(lmStat
.partial_page
);
814 * FUNCTION: group commit post-processing
815 * Processes transactions after their commit records have been written
816 * to disk, redriving log I/O if necessary.
821 * This routine is called a interrupt time by lbmIODone
823 static void lmPostGC(struct lbuf
* bp
)
826 struct jfs_log
*log
= bp
->l_log
;
828 struct tblock
*tblk
, *temp
;
831 spin_lock_irqsave(&log
->gclock
, flags
);
833 * current pageout of group commit completed.
835 * remove/wakeup transactions from commit queue who were
836 * group committed with the current log page
838 list_for_each_entry_safe(tblk
, temp
, &log
->cqueue
, cqueue
) {
839 if (!(tblk
->flag
& tblkGC_COMMIT
))
841 /* if transaction was marked GC_COMMIT then
842 * it has been shipped in the current pageout
843 * and made it to disk - it is committed.
846 if (bp
->l_flag
& lbmERROR
)
847 tblk
->flag
|= tblkGC_ERROR
;
849 /* remove it from the commit queue */
850 list_del(&tblk
->cqueue
);
851 tblk
->flag
&= ~tblkGC_QUEUE
;
853 if (tblk
== log
->flush_tblk
) {
854 /* we can stop flushing the log now */
855 clear_bit(log_FLUSH
, &log
->flag
);
856 log
->flush_tblk
= NULL
;
859 jfs_info("lmPostGC: tblk = 0x%p, flag = 0x%x", tblk
,
862 if (!(tblk
->xflag
& COMMIT_FORCE
))
864 * Hand tblk over to lazy commit thread
868 /* state transition: COMMIT -> COMMITTED */
869 tblk
->flag
|= tblkGC_COMMITTED
;
871 if (tblk
->flag
& tblkGC_READY
)
877 /* was page full before pageout ?
878 * (and this is the last tblk bound with the page)
880 if (tblk
->flag
& tblkGC_FREE
)
882 /* did page become full after pageout ?
883 * (and this is the last tblk bound with the page)
885 else if (tblk
->flag
& tblkGC_EOP
) {
886 /* finalize the page */
887 lp
= (struct logpage
*) bp
->l_ldata
;
888 bp
->l_ceor
= bp
->l_eor
;
889 lp
->h
.eor
= lp
->t
.eor
= cpu_to_le16(bp
->l_eor
);
890 jfs_info("lmPostGC: calling lbmWrite");
891 lbmWrite(log
, bp
, lbmWRITE
| lbmRELEASE
| lbmFREE
,
897 /* are there any transactions who have entered lnGroupCommit()
898 * (whose COMMITs are after that of the last log page written.
899 * They are waiting for new group commit (above at (SLEEP 1))
900 * or lazy transactions are on a full (queued) log page,
901 * select the latest ready transaction as new group leader and
902 * wake her up to lead her group.
904 if ((!list_empty(&log
->cqueue
)) &&
905 ((log
->gcrtc
> 0) || (tblk
->bp
->l_wqnext
!= NULL
) ||
906 test_bit(log_FLUSH
, &log
->flag
) || jfs_tlocks_low
))
908 * Call lmGCwrite with new group leader
912 /* no transaction are ready yet (transactions are only just
913 * queued (GC_QUEUE) and not entered for group commit yet).
914 * the first transaction entering group commit
915 * will elect herself as new group leader.
918 log
->cflag
&= ~logGC_PAGEOUT
;
921 spin_unlock_irqrestore(&log
->gclock
, flags
);
928 * FUNCTION: write log SYNCPT record for specified log
929 * if new sync address is available
930 * (normally the case if sync() is executed by back-ground
932 * calculate new value of i_nextsync which determines when
933 * this code is called again.
935 * PARAMETERS: log - log structure
936 * hard_sync - 1 to force all metadata to be written
940 * serialization: LOG_LOCK() held on entry/exit
942 static int lmLogSync(struct jfs_log
* log
, int hard_sync
)
945 int written
; /* written since last syncpt */
946 int free
; /* free space left available */
947 int delta
; /* additional delta to write normally */
948 int more
; /* additional write granted */
951 struct logsyncblk
*lp
;
954 /* push dirty metapages out to disk */
956 write_special_inodes(log
, filemap_fdatawrite
);
958 write_special_inodes(log
, filemap_flush
);
963 /* if last sync is same as last syncpt,
964 * invoke sync point forward processing to update sync.
967 if (log
->sync
== log
->syncpt
) {
968 LOGSYNC_LOCK(log
, flags
);
969 if (list_empty(&log
->synclist
))
970 log
->sync
= log
->lsn
;
972 lp
= list_entry(log
->synclist
.next
,
973 struct logsyncblk
, synclist
);
976 LOGSYNC_UNLOCK(log
, flags
);
980 /* if sync is different from last syncpt,
981 * write a SYNCPT record with syncpt = sync.
982 * reset syncpt = sync
984 if (log
->sync
!= log
->syncpt
) {
987 lrd
.type
= cpu_to_le16(LOG_SYNCPT
);
989 lrd
.log
.syncpt
.sync
= cpu_to_le32(log
->sync
);
990 lsn
= lmWriteRecord(log
, NULL
, &lrd
, NULL
);
992 log
->syncpt
= log
->sync
;
997 * setup next syncpt trigger (SWAG)
999 logsize
= log
->logsize
;
1001 logdiff(written
, lsn
, log
);
1002 free
= logsize
- written
;
1003 delta
= LOGSYNC_DELTA(logsize
);
1004 more
= min(free
/ 2, delta
);
1005 if (more
< 2 * LOGPSIZE
) {
1006 jfs_warn("\n ... Log Wrap ... Log Wrap ... Log Wrap ...\n");
1010 * option 1 - panic ? No.!
1011 * option 2 - shutdown file systems
1012 * associated with log ?
1013 * option 3 - extend log ?
1014 * option 4 - second chance
1016 * mark log wrapped, and continue.
1017 * when all active transactions are completed,
1018 * mark log valid for recovery.
1019 * if crashed during invalid state, log state
1020 * implies invalid log, forcing fsck().
1022 /* mark log state log wrap in log superblock */
1023 /* log->state = LOGWRAP; */
1025 /* reset sync point computation */
1026 log
->syncpt
= log
->sync
= lsn
;
1027 log
->nextsync
= delta
;
1029 /* next syncpt trigger = written + more */
1030 log
->nextsync
= written
+ more
;
1032 /* if number of bytes written from last sync point is more
1033 * than 1/4 of the log size, stop new transactions from
1034 * starting until all current transactions are completed
1035 * by setting syncbarrier flag.
1037 if (!test_bit(log_SYNCBARRIER
, &log
->flag
) &&
1038 (written
> LOGSYNC_BARRIER(logsize
)) && log
->active
) {
1039 set_bit(log_SYNCBARRIER
, &log
->flag
);
1040 jfs_info("log barrier on: lsn=0x%x syncpt=0x%x", lsn
,
1043 * We may have to initiate group commit
1045 jfs_flush_journal(log
, 0);
1054 * FUNCTION: write log SYNCPT record for specified log
1056 * PARAMETERS: log - log structure
1057 * hard_sync - set to 1 to force metadata to be written
1059 void jfs_syncpt(struct jfs_log
*log
, int hard_sync
)
1061 if (!test_bit(log_QUIESCE
, &log
->flag
))
1062 lmLogSync(log
, hard_sync
);
1069 * FUNCTION: open the log on first open;
1070 * insert filesystem in the active list of the log.
1072 * PARAMETER: ipmnt - file system mount inode
1073 * iplog - log inode (out)
1079 int lmLogOpen(struct super_block
*sb
)
1082 struct block_device
*bdev
;
1083 struct jfs_log
*log
;
1084 struct jfs_sb_info
*sbi
= JFS_SBI(sb
);
1086 if (sbi
->flag
& JFS_NOINTEGRITY
)
1087 return open_dummy_log(sb
);
1089 if (sbi
->mntflag
& JFS_INLINELOG
)
1090 return open_inline_log(sb
);
1092 mutex_lock(&jfs_log_mutex
);
1093 list_for_each_entry(log
, &jfs_external_logs
, journal_list
) {
1094 if (log
->bdev
->bd_dev
== sbi
->logdev
) {
1095 if (memcmp(log
->uuid
, sbi
->loguuid
,
1096 sizeof(log
->uuid
))) {
1097 jfs_warn("wrong uuid on JFS journal");
1098 mutex_unlock(&jfs_log_mutex
);
1102 * add file system to log active file system list
1104 if ((rc
= lmLogFileSystem(log
, sbi
, 1))) {
1105 mutex_unlock(&jfs_log_mutex
);
1112 if (!(log
= kzalloc(sizeof(struct jfs_log
), GFP_KERNEL
))) {
1113 mutex_unlock(&jfs_log_mutex
);
1116 INIT_LIST_HEAD(&log
->sb_list
);
1117 init_waitqueue_head(&log
->syncwait
);
1120 * external log as separate logical volume
1122 * file systems to log may have n-to-1 relationship;
1125 bdev
= blkdev_get_by_dev(sbi
->logdev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
,
1133 memcpy(log
->uuid
, sbi
->loguuid
, sizeof(log
->uuid
));
1138 if ((rc
= lmLogInit(log
)))
1141 list_add(&log
->journal_list
, &jfs_external_logs
);
1144 * add file system to log active file system list
1146 if ((rc
= lmLogFileSystem(log
, sbi
, 1)))
1151 list_add(&sbi
->log_list
, &log
->sb_list
);
1155 mutex_unlock(&jfs_log_mutex
);
1161 shutdown
: /* unwind lbmLogInit() */
1162 list_del(&log
->journal_list
);
1163 lbmLogShutdown(log
);
1165 close
: /* close external log device */
1166 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
);
1168 free
: /* free log descriptor */
1169 mutex_unlock(&jfs_log_mutex
);
1172 jfs_warn("lmLogOpen: exit(%d)", rc
);
1176 static int open_inline_log(struct super_block
*sb
)
1178 struct jfs_log
*log
;
1181 if (!(log
= kzalloc(sizeof(struct jfs_log
), GFP_KERNEL
)))
1183 INIT_LIST_HEAD(&log
->sb_list
);
1184 init_waitqueue_head(&log
->syncwait
);
1186 set_bit(log_INLINELOG
, &log
->flag
);
1187 log
->bdev
= sb
->s_bdev
;
1188 log
->base
= addressPXD(&JFS_SBI(sb
)->logpxd
);
1189 log
->size
= lengthPXD(&JFS_SBI(sb
)->logpxd
) >>
1190 (L2LOGPSIZE
- sb
->s_blocksize_bits
);
1191 log
->l2bsize
= sb
->s_blocksize_bits
;
1192 ASSERT(L2LOGPSIZE
>= sb
->s_blocksize_bits
);
1197 if ((rc
= lmLogInit(log
))) {
1199 jfs_warn("lmLogOpen: exit(%d)", rc
);
1203 list_add(&JFS_SBI(sb
)->log_list
, &log
->sb_list
);
1204 JFS_SBI(sb
)->log
= log
;
1209 static int open_dummy_log(struct super_block
*sb
)
1213 mutex_lock(&jfs_log_mutex
);
1215 dummy_log
= kzalloc(sizeof(struct jfs_log
), GFP_KERNEL
);
1217 mutex_unlock(&jfs_log_mutex
);
1220 INIT_LIST_HEAD(&dummy_log
->sb_list
);
1221 init_waitqueue_head(&dummy_log
->syncwait
);
1222 dummy_log
->no_integrity
= 1;
1223 /* Make up some stuff */
1224 dummy_log
->base
= 0;
1225 dummy_log
->size
= 1024;
1226 rc
= lmLogInit(dummy_log
);
1230 mutex_unlock(&jfs_log_mutex
);
1235 LOG_LOCK(dummy_log
);
1236 list_add(&JFS_SBI(sb
)->log_list
, &dummy_log
->sb_list
);
1237 JFS_SBI(sb
)->log
= dummy_log
;
1238 LOG_UNLOCK(dummy_log
);
1239 mutex_unlock(&jfs_log_mutex
);
1247 * FUNCTION: log initialization at first log open.
1249 * logredo() (or logformat()) should have been run previously.
1250 * initialize the log from log superblock.
1251 * set the log state in the superblock to LOGMOUNT and
1252 * write SYNCPT log record.
1254 * PARAMETER: log - log structure
1257 * -EINVAL - bad log magic number or superblock dirty
1258 * error returned from logwait()
1260 * serialization: single first open thread
1262 int lmLogInit(struct jfs_log
* log
)
1266 struct logsuper
*logsuper
;
1267 struct lbuf
*bpsuper
;
1272 jfs_info("lmLogInit: log:0x%p", log
);
1274 /* initialize the group commit serialization lock */
1275 LOGGC_LOCK_INIT(log
);
1277 /* allocate/initialize the log write serialization lock */
1280 LOGSYNC_LOCK_INIT(log
);
1282 INIT_LIST_HEAD(&log
->synclist
);
1284 INIT_LIST_HEAD(&log
->cqueue
);
1285 log
->flush_tblk
= NULL
;
1290 * initialize log i/o
1292 if ((rc
= lbmLogInit(log
)))
1295 if (!test_bit(log_INLINELOG
, &log
->flag
))
1296 log
->l2bsize
= L2LOGPSIZE
;
1298 /* check for disabled journaling to disk */
1299 if (log
->no_integrity
) {
1301 * Journal pages will still be filled. When the time comes
1302 * to actually do the I/O, the write is not done, and the
1303 * endio routine is called directly.
1305 bp
= lbmAllocate(log
, 0);
1307 bp
->l_pn
= bp
->l_eor
= 0;
1310 * validate log superblock
1312 if ((rc
= lbmRead(log
, 1, &bpsuper
)))
1315 logsuper
= (struct logsuper
*) bpsuper
->l_ldata
;
1317 if (logsuper
->magic
!= cpu_to_le32(LOGMAGIC
)) {
1318 jfs_warn("*** Log Format Error ! ***");
1323 /* logredo() should have been run successfully. */
1324 if (logsuper
->state
!= cpu_to_le32(LOGREDONE
)) {
1325 jfs_warn("*** Log Is Dirty ! ***");
1330 /* initialize log from log superblock */
1331 if (test_bit(log_INLINELOG
,&log
->flag
)) {
1332 if (log
->size
!= le32_to_cpu(logsuper
->size
)) {
1336 jfs_info("lmLogInit: inline log:0x%p base:0x%Lx size:0x%x",
1337 log
, (unsigned long long)log
->base
, log
->size
);
1339 if (memcmp(logsuper
->uuid
, log
->uuid
, 16)) {
1340 jfs_warn("wrong uuid on JFS log device");
1343 log
->size
= le32_to_cpu(logsuper
->size
);
1344 log
->l2bsize
= le32_to_cpu(logsuper
->l2bsize
);
1345 jfs_info("lmLogInit: external log:0x%p base:0x%Lx size:0x%x",
1346 log
, (unsigned long long)log
->base
, log
->size
);
1349 log
->page
= le32_to_cpu(logsuper
->end
) / LOGPSIZE
;
1350 log
->eor
= le32_to_cpu(logsuper
->end
) - (LOGPSIZE
* log
->page
);
1353 * initialize for log append write mode
1355 /* establish current/end-of-log page/buffer */
1356 if ((rc
= lbmRead(log
, log
->page
, &bp
)))
1359 lp
= (struct logpage
*) bp
->l_ldata
;
1361 jfs_info("lmLogInit: lsn:0x%x page:%d eor:%d:%d",
1362 le32_to_cpu(logsuper
->end
), log
->page
, log
->eor
,
1363 le16_to_cpu(lp
->h
.eor
));
1366 bp
->l_pn
= log
->page
;
1367 bp
->l_eor
= log
->eor
;
1369 /* if current page is full, move on to next page */
1370 if (log
->eor
>= LOGPSIZE
- LOGPTLRSIZE
)
1374 * initialize log syncpoint
1377 * write the first SYNCPT record with syncpoint = 0
1378 * (i.e., log redo up to HERE !);
1379 * remove current page from lbm write queue at end of pageout
1380 * (to write log superblock update), but do not release to
1385 lrd
.type
= cpu_to_le16(LOG_SYNCPT
);
1387 lrd
.log
.syncpt
.sync
= 0;
1388 lsn
= lmWriteRecord(log
, NULL
, &lrd
, NULL
);
1390 bp
->l_ceor
= bp
->l_eor
;
1391 lp
= (struct logpage
*) bp
->l_ldata
;
1392 lp
->h
.eor
= lp
->t
.eor
= cpu_to_le16(bp
->l_eor
);
1393 lbmWrite(log
, bp
, lbmWRITE
| lbmSYNC
, 0);
1394 if ((rc
= lbmIOWait(bp
, 0)))
1398 * update/write superblock
1400 logsuper
->state
= cpu_to_le32(LOGMOUNT
);
1401 log
->serial
= le32_to_cpu(logsuper
->serial
) + 1;
1402 logsuper
->serial
= cpu_to_le32(log
->serial
);
1403 lbmDirectWrite(log
, bpsuper
, lbmWRITE
| lbmRELEASE
| lbmSYNC
);
1404 if ((rc
= lbmIOWait(bpsuper
, lbmFREE
)))
1408 /* initialize logsync parameters */
1409 log
->logsize
= (log
->size
- 2) << L2LOGPSIZE
;
1412 log
->sync
= log
->syncpt
;
1413 log
->nextsync
= LOGSYNC_DELTA(log
->logsize
);
1415 jfs_info("lmLogInit: lsn:0x%x syncpt:0x%x sync:0x%x",
1416 log
->lsn
, log
->syncpt
, log
->sync
);
1419 * initialize for lazy/group commit
1428 errout30
: /* release log page */
1430 bp
->l_wqnext
= NULL
;
1433 errout20
: /* release log superblock */
1436 errout10
: /* unwind lbmLogInit() */
1437 lbmLogShutdown(log
);
1439 jfs_warn("lmLogInit: exit(%d)", rc
);
1445 * NAME: lmLogClose()
1447 * FUNCTION: remove file system <ipmnt> from active list of log <iplog>
1448 * and close it on last close.
1450 * PARAMETER: sb - superblock
1452 * RETURN: errors from subroutines
1456 int lmLogClose(struct super_block
*sb
)
1458 struct jfs_sb_info
*sbi
= JFS_SBI(sb
);
1459 struct jfs_log
*log
= sbi
->log
;
1460 struct block_device
*bdev
;
1463 jfs_info("lmLogClose: log:0x%p", log
);
1465 mutex_lock(&jfs_log_mutex
);
1467 list_del(&sbi
->log_list
);
1472 * We need to make sure all of the "written" metapages
1473 * actually make it to disk
1475 sync_blockdev(sb
->s_bdev
);
1477 if (test_bit(log_INLINELOG
, &log
->flag
)) {
1479 * in-line log in host file system
1481 rc
= lmLogShutdown(log
);
1486 if (!log
->no_integrity
)
1487 lmLogFileSystem(log
, sbi
, 0);
1489 if (!list_empty(&log
->sb_list
))
1493 * TODO: ensure that the dummy_log is in a state to allow
1494 * lbmLogShutdown to deallocate all the buffers and call
1495 * kfree against dummy_log. For now, leave dummy_log & its
1496 * buffers in memory, and resuse if another no-integrity mount
1499 if (log
->no_integrity
)
1503 * external log as separate logical volume
1505 list_del(&log
->journal_list
);
1507 rc
= lmLogShutdown(log
);
1509 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
);
1514 mutex_unlock(&jfs_log_mutex
);
1515 jfs_info("lmLogClose: exit(%d)", rc
);
1521 * NAME: jfs_flush_journal()
1523 * FUNCTION: initiate write of any outstanding transactions to the journal
1524 * and optionally wait until they are all written to disk
1526 * wait == 0 flush until latest txn is committed, don't wait
1527 * wait == 1 flush until latest txn is committed, wait
1528 * wait > 1 flush until all txn's are complete, wait
1530 void jfs_flush_journal(struct jfs_log
*log
, int wait
)
1533 struct tblock
*target
= NULL
;
1535 /* jfs_write_inode may call us during read-only mount */
1539 jfs_info("jfs_flush_journal: log:0x%p wait=%d", log
, wait
);
1543 if (!list_empty(&log
->cqueue
)) {
1545 * This ensures that we will keep writing to the journal as long
1546 * as there are unwritten commit records
1548 target
= list_entry(log
->cqueue
.prev
, struct tblock
, cqueue
);
1550 if (test_bit(log_FLUSH
, &log
->flag
)) {
1552 * We're already flushing.
1553 * if flush_tblk is NULL, we are flushing everything,
1554 * so leave it that way. Otherwise, update it to the
1555 * latest transaction
1557 if (log
->flush_tblk
)
1558 log
->flush_tblk
= target
;
1560 /* Only flush until latest transaction is committed */
1561 log
->flush_tblk
= target
;
1562 set_bit(log_FLUSH
, &log
->flag
);
1565 * Initiate I/O on outstanding transactions
1567 if (!(log
->cflag
& logGC_PAGEOUT
)) {
1568 log
->cflag
|= logGC_PAGEOUT
;
1573 if ((wait
> 1) || test_bit(log_SYNCBARRIER
, &log
->flag
)) {
1574 /* Flush until all activity complete */
1575 set_bit(log_FLUSH
, &log
->flag
);
1576 log
->flush_tblk
= NULL
;
1579 if (wait
&& target
&& !(target
->flag
& tblkGC_COMMITTED
)) {
1580 DECLARE_WAITQUEUE(__wait
, current
);
1582 add_wait_queue(&target
->gcwait
, &__wait
);
1583 set_current_state(TASK_UNINTERRUPTIBLE
);
1587 remove_wait_queue(&target
->gcwait
, &__wait
);
1594 write_special_inodes(log
, filemap_fdatawrite
);
1597 * If there was recent activity, we may need to wait
1598 * for the lazycommit thread to catch up
1600 if ((!list_empty(&log
->cqueue
)) || !list_empty(&log
->synclist
)) {
1601 for (i
= 0; i
< 200; i
++) { /* Too much? */
1603 write_special_inodes(log
, filemap_fdatawrite
);
1604 if (list_empty(&log
->cqueue
) &&
1605 list_empty(&log
->synclist
))
1609 assert(list_empty(&log
->cqueue
));
1611 #ifdef CONFIG_JFS_DEBUG
1612 if (!list_empty(&log
->synclist
)) {
1613 struct logsyncblk
*lp
;
1615 printk(KERN_ERR
"jfs_flush_journal: synclist not empty\n");
1616 list_for_each_entry(lp
, &log
->synclist
, synclist
) {
1617 if (lp
->xflag
& COMMIT_PAGE
) {
1618 struct metapage
*mp
= (struct metapage
*)lp
;
1619 print_hex_dump(KERN_ERR
, "metapage: ",
1620 DUMP_PREFIX_ADDRESS
, 16, 4,
1621 mp
, sizeof(struct metapage
), 0);
1622 print_hex_dump(KERN_ERR
, "page: ",
1623 DUMP_PREFIX_ADDRESS
, 16,
1624 sizeof(long), mp
->page
,
1625 sizeof(struct page
), 0);
1627 print_hex_dump(KERN_ERR
, "tblock:",
1628 DUMP_PREFIX_ADDRESS
, 16, 4,
1629 lp
, sizeof(struct tblock
), 0);
1633 WARN_ON(!list_empty(&log
->synclist
));
1635 clear_bit(log_FLUSH
, &log
->flag
);
1639 * NAME: lmLogShutdown()
1641 * FUNCTION: log shutdown at last LogClose().
1643 * write log syncpt record.
1644 * update super block to set redone flag to 0.
1646 * PARAMETER: log - log inode
1648 * RETURN: 0 - success
1650 * serialization: single last close thread
1652 int lmLogShutdown(struct jfs_log
* log
)
1657 struct logsuper
*logsuper
;
1658 struct lbuf
*bpsuper
;
1662 jfs_info("lmLogShutdown: log:0x%p", log
);
1664 jfs_flush_journal(log
, 2);
1667 * write the last SYNCPT record with syncpoint = 0
1668 * (i.e., log redo up to HERE !)
1672 lrd
.type
= cpu_to_le16(LOG_SYNCPT
);
1674 lrd
.log
.syncpt
.sync
= 0;
1676 lsn
= lmWriteRecord(log
, NULL
, &lrd
, NULL
);
1678 lp
= (struct logpage
*) bp
->l_ldata
;
1679 lp
->h
.eor
= lp
->t
.eor
= cpu_to_le16(bp
->l_eor
);
1680 lbmWrite(log
, log
->bp
, lbmWRITE
| lbmRELEASE
| lbmSYNC
, 0);
1681 lbmIOWait(log
->bp
, lbmFREE
);
1685 * synchronous update log superblock
1686 * mark log state as shutdown cleanly
1687 * (i.e., Log does not need to be replayed).
1689 if ((rc
= lbmRead(log
, 1, &bpsuper
)))
1692 logsuper
= (struct logsuper
*) bpsuper
->l_ldata
;
1693 logsuper
->state
= cpu_to_le32(LOGREDONE
);
1694 logsuper
->end
= cpu_to_le32(lsn
);
1695 lbmDirectWrite(log
, bpsuper
, lbmWRITE
| lbmRELEASE
| lbmSYNC
);
1696 rc
= lbmIOWait(bpsuper
, lbmFREE
);
1698 jfs_info("lmLogShutdown: lsn:0x%x page:%d eor:%d",
1699 lsn
, log
->page
, log
->eor
);
1703 * shutdown per log i/o
1705 lbmLogShutdown(log
);
1708 jfs_warn("lmLogShutdown: exit(%d)", rc
);
1715 * NAME: lmLogFileSystem()
1717 * FUNCTION: insert (<activate> = true)/remove (<activate> = false)
1718 * file system into/from log active file system list.
1720 * PARAMETE: log - pointer to logs inode.
1721 * fsdev - kdev_t of filesystem.
1722 * serial - pointer to returned log serial number
1723 * activate - insert/remove device from active list.
1725 * RETURN: 0 - success
1726 * errors returned by vms_iowait().
1728 static int lmLogFileSystem(struct jfs_log
* log
, struct jfs_sb_info
*sbi
,
1733 struct logsuper
*logsuper
;
1734 struct lbuf
*bpsuper
;
1735 char *uuid
= sbi
->uuid
;
1738 * insert/remove file system device to log active file system list.
1740 if ((rc
= lbmRead(log
, 1, &bpsuper
)))
1743 logsuper
= (struct logsuper
*) bpsuper
->l_ldata
;
1745 for (i
= 0; i
< MAX_ACTIVE
; i
++)
1746 if (!memcmp(logsuper
->active
[i
].uuid
, NULL_UUID
, 16)) {
1747 memcpy(logsuper
->active
[i
].uuid
, uuid
, 16);
1751 if (i
== MAX_ACTIVE
) {
1752 jfs_warn("Too many file systems sharing journal!");
1754 return -EMFILE
; /* Is there a better rc? */
1757 for (i
= 0; i
< MAX_ACTIVE
; i
++)
1758 if (!memcmp(logsuper
->active
[i
].uuid
, uuid
, 16)) {
1759 memcpy(logsuper
->active
[i
].uuid
, NULL_UUID
, 16);
1762 if (i
== MAX_ACTIVE
) {
1763 jfs_warn("Somebody stomped on the journal!");
1771 * synchronous write log superblock:
1773 * write sidestream bypassing write queue:
1774 * at file system mount, log super block is updated for
1775 * activation of the file system before any log record
1776 * (MOUNT record) of the file system, and at file system
1777 * unmount, all meta data for the file system has been
1778 * flushed before log super block is updated for deactivation
1779 * of the file system.
1781 lbmDirectWrite(log
, bpsuper
, lbmWRITE
| lbmRELEASE
| lbmSYNC
);
1782 rc
= lbmIOWait(bpsuper
, lbmFREE
);
1788 * log buffer manager (lbm)
1789 * ------------------------
1791 * special purpose buffer manager supporting log i/o requirements.
1793 * per log write queue:
1794 * log pageout occurs in serial order by fifo write queue and
1795 * restricting to a single i/o in pregress at any one time.
1796 * a circular singly-linked list
1797 * (log->wrqueue points to the tail, and buffers are linked via
1798 * bp->wrqueue field), and
1799 * maintains log page in pageout ot waiting for pageout in serial pageout.
1805 * initialize per log I/O setup at lmLogInit()
1807 static int lbmLogInit(struct jfs_log
* log
)
1812 jfs_info("lbmLogInit: log:0x%p", log
);
1814 /* initialize current buffer cursor */
1817 /* initialize log device write queue */
1821 * Each log has its own buffer pages allocated to it. These are
1822 * not managed by the page cache. This ensures that a transaction
1823 * writing to the log does not block trying to allocate a page from
1824 * the page cache (for the log). This would be bad, since page
1825 * allocation waits on the kswapd thread that may be committing inodes
1826 * which would cause log activity. Was that clear? I'm trying to
1827 * avoid deadlock here.
1829 init_waitqueue_head(&log
->free_wait
);
1831 log
->lbuf_free
= NULL
;
1833 for (i
= 0; i
< LOGPAGES
;) {
1836 struct page
*page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
1840 buffer
= page_address(page
);
1841 for (offset
= 0; offset
< PAGE_SIZE
; offset
+= LOGPSIZE
) {
1842 lbuf
= kmalloc(sizeof(struct lbuf
), GFP_KERNEL
);
1848 if (offset
) /* we already have one reference */
1850 lbuf
->l_offset
= offset
;
1851 lbuf
->l_ldata
= buffer
+ offset
;
1852 lbuf
->l_page
= page
;
1854 init_waitqueue_head(&lbuf
->l_ioevent
);
1856 lbuf
->l_freelist
= log
->lbuf_free
;
1857 log
->lbuf_free
= lbuf
;
1865 lbmLogShutdown(log
);
1873 * finalize per log I/O setup at lmLogShutdown()
1875 static void lbmLogShutdown(struct jfs_log
* log
)
1879 jfs_info("lbmLogShutdown: log:0x%p", log
);
1881 lbuf
= log
->lbuf_free
;
1883 struct lbuf
*next
= lbuf
->l_freelist
;
1884 __free_page(lbuf
->l_page
);
1894 * allocate an empty log buffer
1896 static struct lbuf
*lbmAllocate(struct jfs_log
* log
, int pn
)
1899 unsigned long flags
;
1902 * recycle from log buffer freelist if any
1905 LCACHE_SLEEP_COND(log
->free_wait
, (bp
= log
->lbuf_free
), flags
);
1906 log
->lbuf_free
= bp
->l_freelist
;
1907 LCACHE_UNLOCK(flags
);
1911 bp
->l_wqnext
= NULL
;
1912 bp
->l_freelist
= NULL
;
1915 bp
->l_blkno
= log
->base
+ (pn
<< (L2LOGPSIZE
- log
->l2bsize
));
1925 * release a log buffer to freelist
1927 static void lbmFree(struct lbuf
* bp
)
1929 unsigned long flags
;
1935 LCACHE_UNLOCK(flags
);
1938 static void lbmfree(struct lbuf
* bp
)
1940 struct jfs_log
*log
= bp
->l_log
;
1942 assert(bp
->l_wqnext
== NULL
);
1945 * return the buffer to head of freelist
1947 bp
->l_freelist
= log
->lbuf_free
;
1948 log
->lbuf_free
= bp
;
1950 wake_up(&log
->free_wait
);
1958 * FUNCTION: add a log buffer to the log redrive list
1964 * Takes log_redrive_lock.
1966 static inline void lbmRedrive(struct lbuf
*bp
)
1968 unsigned long flags
;
1970 spin_lock_irqsave(&log_redrive_lock
, flags
);
1971 bp
->l_redrive_next
= log_redrive_list
;
1972 log_redrive_list
= bp
;
1973 spin_unlock_irqrestore(&log_redrive_lock
, flags
);
1975 wake_up_process(jfsIOthread
);
1982 static int lbmRead(struct jfs_log
* log
, int pn
, struct lbuf
** bpp
)
1988 * allocate a log buffer
1990 *bpp
= bp
= lbmAllocate(log
, pn
);
1991 jfs_info("lbmRead: bp:0x%p pn:0x%x", bp
, pn
);
1993 bp
->l_flag
|= lbmREAD
;
1995 bio
= bio_alloc(GFP_NOFS
, 1);
1997 bio
->bi_iter
.bi_sector
= bp
->l_blkno
<< (log
->l2bsize
- 9);
1998 bio
->bi_bdev
= log
->bdev
;
2000 bio_add_page(bio
, bp
->l_page
, LOGPSIZE
, bp
->l_offset
);
2001 BUG_ON(bio
->bi_iter
.bi_size
!= LOGPSIZE
);
2003 bio
->bi_end_io
= lbmIODone
;
2004 bio
->bi_private
= bp
;
2005 bio
->bi_opf
= REQ_OP_READ
;
2006 /*check if journaling to disk has been disabled*/
2007 if (log
->no_integrity
) {
2008 bio
->bi_iter
.bi_size
= 0;
2014 wait_event(bp
->l_ioevent
, (bp
->l_flag
!= lbmREAD
));
2023 * buffer at head of pageout queue stays after completion of
2024 * partial-page pageout and redriven by explicit initiation of
2025 * pageout by caller until full-page pageout is completed and
2028 * device driver i/o done redrives pageout of new buffer at
2029 * head of pageout queue when current buffer at head of pageout
2030 * queue is released at the completion of its full-page pageout.
2032 * LOGGC_LOCK() serializes lbmWrite() by lmNextPage() and lmGroupCommit().
2033 * LCACHE_LOCK() serializes xflag between lbmWrite() and lbmIODone()
2035 static void lbmWrite(struct jfs_log
* log
, struct lbuf
* bp
, int flag
,
2039 unsigned long flags
;
2041 jfs_info("lbmWrite: bp:0x%p flag:0x%x pn:0x%x", bp
, flag
, bp
->l_pn
);
2043 /* map the logical block address to physical block address */
2045 log
->base
+ (bp
->l_pn
<< (L2LOGPSIZE
- log
->l2bsize
));
2047 LCACHE_LOCK(flags
); /* disable+lock */
2050 * initialize buffer for device driver
2055 * insert bp at tail of write queue associated with log
2057 * (request is either for bp already/currently at head of queue
2058 * or new bp to be inserted at tail)
2062 /* is buffer not already on write queue ? */
2063 if (bp
->l_wqnext
== NULL
) {
2064 /* insert at tail of wqueue */
2070 bp
->l_wqnext
= tail
->l_wqnext
;
2071 tail
->l_wqnext
= bp
;
2077 /* is buffer at head of wqueue and for write ? */
2078 if ((bp
!= tail
->l_wqnext
) || !(flag
& lbmWRITE
)) {
2079 LCACHE_UNLOCK(flags
); /* unlock+enable */
2083 LCACHE_UNLOCK(flags
); /* unlock+enable */
2087 else if (flag
& lbmSYNC
)
2100 * initiate pageout bypassing write queue for sidestream
2101 * (e.g., log superblock) write;
2103 static void lbmDirectWrite(struct jfs_log
* log
, struct lbuf
* bp
, int flag
)
2105 jfs_info("lbmDirectWrite: bp:0x%p flag:0x%x pn:0x%x",
2106 bp
, flag
, bp
->l_pn
);
2109 * initialize buffer for device driver
2111 bp
->l_flag
= flag
| lbmDIRECT
;
2113 /* map the logical block address to physical block address */
2115 log
->base
+ (bp
->l_pn
<< (L2LOGPSIZE
- log
->l2bsize
));
2118 * initiate pageout of the page
2125 * NAME: lbmStartIO()
2127 * FUNCTION: Interface to DD strategy routine
2131 * serialization: LCACHE_LOCK() is NOT held during log i/o;
2133 static void lbmStartIO(struct lbuf
* bp
)
2136 struct jfs_log
*log
= bp
->l_log
;
2138 jfs_info("lbmStartIO");
2140 bio
= bio_alloc(GFP_NOFS
, 1);
2141 bio
->bi_iter
.bi_sector
= bp
->l_blkno
<< (log
->l2bsize
- 9);
2142 bio
->bi_bdev
= log
->bdev
;
2144 bio_add_page(bio
, bp
->l_page
, LOGPSIZE
, bp
->l_offset
);
2145 BUG_ON(bio
->bi_iter
.bi_size
!= LOGPSIZE
);
2147 bio
->bi_end_io
= lbmIODone
;
2148 bio
->bi_private
= bp
;
2149 bio
->bi_opf
= REQ_OP_WRITE
| REQ_SYNC
;
2151 /* check if journaling to disk has been disabled */
2152 if (log
->no_integrity
) {
2153 bio
->bi_iter
.bi_size
= 0;
2157 INCREMENT(lmStat
.submitted
);
2165 static int lbmIOWait(struct lbuf
* bp
, int flag
)
2167 unsigned long flags
;
2170 jfs_info("lbmIOWait1: bp:0x%p flag:0x%x:0x%x", bp
, bp
->l_flag
, flag
);
2172 LCACHE_LOCK(flags
); /* disable+lock */
2174 LCACHE_SLEEP_COND(bp
->l_ioevent
, (bp
->l_flag
& lbmDONE
), flags
);
2176 rc
= (bp
->l_flag
& lbmERROR
) ? -EIO
: 0;
2181 LCACHE_UNLOCK(flags
); /* unlock+enable */
2183 jfs_info("lbmIOWait2: bp:0x%p flag:0x%x:0x%x", bp
, bp
->l_flag
, flag
);
2190 * executed at INTIODONE level
2192 static void lbmIODone(struct bio
*bio
)
2194 struct lbuf
*bp
= bio
->bi_private
;
2195 struct lbuf
*nextbp
, *tail
;
2196 struct jfs_log
*log
;
2197 unsigned long flags
;
2200 * get back jfs buffer bound to the i/o buffer
2202 jfs_info("lbmIODone: bp:0x%p flag:0x%x", bp
, bp
->l_flag
);
2204 LCACHE_LOCK(flags
); /* disable+lock */
2206 bp
->l_flag
|= lbmDONE
;
2208 if (bio
->bi_error
) {
2209 bp
->l_flag
|= lbmERROR
;
2211 jfs_err("lbmIODone: I/O error in JFS log");
2219 if (bp
->l_flag
& lbmREAD
) {
2220 bp
->l_flag
&= ~lbmREAD
;
2222 LCACHE_UNLOCK(flags
); /* unlock+enable */
2224 /* wakeup I/O initiator */
2225 LCACHE_WAKEUP(&bp
->l_ioevent
);
2231 * pageout completion
2233 * the bp at the head of write queue has completed pageout.
2235 * if single-commit/full-page pageout, remove the current buffer
2236 * from head of pageout queue, and redrive pageout with
2237 * the new buffer at head of pageout queue;
2238 * otherwise, the partial-page pageout buffer stays at
2239 * the head of pageout queue to be redriven for pageout
2240 * by lmGroupCommit() until full-page pageout is completed.
2242 bp
->l_flag
&= ~lbmWRITE
;
2243 INCREMENT(lmStat
.pagedone
);
2245 /* update committed lsn */
2247 log
->clsn
= (bp
->l_pn
<< L2LOGPSIZE
) + bp
->l_ceor
;
2249 if (bp
->l_flag
& lbmDIRECT
) {
2250 LCACHE_WAKEUP(&bp
->l_ioevent
);
2251 LCACHE_UNLOCK(flags
);
2257 /* single element queue */
2259 /* remove head buffer of full-page pageout
2260 * from log device write queue
2262 if (bp
->l_flag
& lbmRELEASE
) {
2264 bp
->l_wqnext
= NULL
;
2267 /* multi element queue */
2269 /* remove head buffer of full-page pageout
2270 * from log device write queue
2272 if (bp
->l_flag
& lbmRELEASE
) {
2273 nextbp
= tail
->l_wqnext
= bp
->l_wqnext
;
2274 bp
->l_wqnext
= NULL
;
2277 * redrive pageout of next page at head of write queue:
2278 * redrive next page without any bound tblk
2279 * (i.e., page w/o any COMMIT records), or
2280 * first page of new group commit which has been
2281 * queued after current page (subsequent pageout
2282 * is performed synchronously, except page without
2283 * any COMMITs) by lmGroupCommit() as indicated
2286 if (nextbp
->l_flag
& lbmWRITE
) {
2288 * We can't do the I/O at interrupt time.
2289 * The jfsIO thread can do it
2297 * synchronous pageout:
2299 * buffer has not necessarily been removed from write queue
2300 * (e.g., synchronous write of partial-page with COMMIT):
2301 * leave buffer for i/o initiator to dispose
2303 if (bp
->l_flag
& lbmSYNC
) {
2304 LCACHE_UNLOCK(flags
); /* unlock+enable */
2306 /* wakeup I/O initiator */
2307 LCACHE_WAKEUP(&bp
->l_ioevent
);
2311 * Group Commit pageout:
2313 else if (bp
->l_flag
& lbmGC
) {
2314 LCACHE_UNLOCK(flags
);
2319 * asynchronous pageout:
2321 * buffer must have been removed from write queue:
2322 * insert buffer at head of freelist where it can be recycled
2325 assert(bp
->l_flag
& lbmRELEASE
);
2326 assert(bp
->l_flag
& lbmFREE
);
2329 LCACHE_UNLOCK(flags
); /* unlock+enable */
2333 int jfsIOWait(void *arg
)
2338 spin_lock_irq(&log_redrive_lock
);
2339 while ((bp
= log_redrive_list
)) {
2340 log_redrive_list
= bp
->l_redrive_next
;
2341 bp
->l_redrive_next
= NULL
;
2342 spin_unlock_irq(&log_redrive_lock
);
2344 spin_lock_irq(&log_redrive_lock
);
2347 if (freezing(current
)) {
2348 spin_unlock_irq(&log_redrive_lock
);
2351 set_current_state(TASK_INTERRUPTIBLE
);
2352 spin_unlock_irq(&log_redrive_lock
);
2355 } while (!kthread_should_stop());
2357 jfs_info("jfsIOWait being killed!");
2362 * NAME: lmLogFormat()/jfs_logform()
2364 * FUNCTION: format file system log
2368 * logAddress - start address of log space in FS block
2369 * logSize - length of log space in FS block;
2371 * RETURN: 0 - success
2374 * XXX: We're synchronously writing one page at a time. This needs to
2375 * be improved by writing multiple pages at once.
2377 int lmLogFormat(struct jfs_log
*log
, s64 logAddress
, int logSize
)
2380 struct jfs_sb_info
*sbi
;
2381 struct logsuper
*logsuper
;
2383 int lspn
; /* log sequence page number */
2384 struct lrd
*lrd_ptr
;
2388 jfs_info("lmLogFormat: logAddress:%Ld logSize:%d",
2389 (long long)logAddress
, logSize
);
2391 sbi
= list_entry(log
->sb_list
.next
, struct jfs_sb_info
, log_list
);
2393 /* allocate a log buffer */
2394 bp
= lbmAllocate(log
, 1);
2396 npages
= logSize
>> sbi
->l2nbperpage
;
2401 * page 0 - reserved;
2402 * page 1 - log superblock;
2403 * page 2 - log data page: A SYNC log record is written
2404 * into this page at logform time;
2405 * pages 3-N - log data page: set to empty log data pages;
2408 * init log superblock: log page 1
2410 logsuper
= (struct logsuper
*) bp
->l_ldata
;
2412 logsuper
->magic
= cpu_to_le32(LOGMAGIC
);
2413 logsuper
->version
= cpu_to_le32(LOGVERSION
);
2414 logsuper
->state
= cpu_to_le32(LOGREDONE
);
2415 logsuper
->flag
= cpu_to_le32(sbi
->mntflag
); /* ? */
2416 logsuper
->size
= cpu_to_le32(npages
);
2417 logsuper
->bsize
= cpu_to_le32(sbi
->bsize
);
2418 logsuper
->l2bsize
= cpu_to_le32(sbi
->l2bsize
);
2419 logsuper
->end
= cpu_to_le32(2 * LOGPSIZE
+ LOGPHDRSIZE
+ LOGRDSIZE
);
2421 bp
->l_flag
= lbmWRITE
| lbmSYNC
| lbmDIRECT
;
2422 bp
->l_blkno
= logAddress
+ sbi
->nbperpage
;
2424 if ((rc
= lbmIOWait(bp
, 0)))
2428 * init pages 2 to npages-1 as log data pages:
2430 * log page sequence number (lpsn) initialization:
2433 * +-----+-----+=====+=====+===.....===+=====+
2435 * <--- N page circular file ---->
2437 * the N (= npages-2) data pages of the log is maintained as
2438 * a circular file for the log records;
2439 * lpsn grows by 1 monotonically as each log page is written
2440 * to the circular file of the log;
2441 * and setLogpage() will not reset the page number even if
2442 * the eor is equal to LOGPHDRSIZE. In order for binary search
2443 * still work in find log end process, we have to simulate the
2444 * log wrap situation at the log format time.
2445 * The 1st log page written will have the highest lpsn. Then
2446 * the succeeding log pages will have ascending order of
2447 * the lspn starting from 0, ... (N-2)
2449 lp
= (struct logpage
*) bp
->l_ldata
;
2451 * initialize 1st log page to be written: lpsn = N - 1,
2452 * write a SYNCPT log record is written to this page
2454 lp
->h
.page
= lp
->t
.page
= cpu_to_le32(npages
- 3);
2455 lp
->h
.eor
= lp
->t
.eor
= cpu_to_le16(LOGPHDRSIZE
+ LOGRDSIZE
);
2457 lrd_ptr
= (struct lrd
*) &lp
->data
;
2458 lrd_ptr
->logtid
= 0;
2459 lrd_ptr
->backchain
= 0;
2460 lrd_ptr
->type
= cpu_to_le16(LOG_SYNCPT
);
2461 lrd_ptr
->length
= 0;
2462 lrd_ptr
->log
.syncpt
.sync
= 0;
2464 bp
->l_blkno
+= sbi
->nbperpage
;
2465 bp
->l_flag
= lbmWRITE
| lbmSYNC
| lbmDIRECT
;
2467 if ((rc
= lbmIOWait(bp
, 0)))
2471 * initialize succeeding log pages: lpsn = 0, 1, ..., (N-2)
2473 for (lspn
= 0; lspn
< npages
- 3; lspn
++) {
2474 lp
->h
.page
= lp
->t
.page
= cpu_to_le32(lspn
);
2475 lp
->h
.eor
= lp
->t
.eor
= cpu_to_le16(LOGPHDRSIZE
);
2477 bp
->l_blkno
+= sbi
->nbperpage
;
2478 bp
->l_flag
= lbmWRITE
| lbmSYNC
| lbmDIRECT
;
2480 if ((rc
= lbmIOWait(bp
, 0)))
2489 /* release the buffer */
2495 #ifdef CONFIG_JFS_STATISTICS
2496 static int jfs_lmstats_proc_show(struct seq_file
*m
, void *v
)
2499 "JFS Logmgr stats\n"
2500 "================\n"
2502 "writes submitted = %d\n"
2503 "writes completed = %d\n"
2504 "full pages submitted = %d\n"
2505 "partial pages submitted = %d\n",
2510 lmStat
.partial_page
);
2514 static int jfs_lmstats_proc_open(struct inode
*inode
, struct file
*file
)
2516 return single_open(file
, jfs_lmstats_proc_show
, NULL
);
2519 const struct file_operations jfs_lmstats_proc_fops
= {
2520 .open
= jfs_lmstats_proc_open
,
2522 .llseek
= seq_lseek
,
2523 .release
= single_release
,
2525 #endif /* CONFIG_JFS_STATISTICS */