1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Copyright (C) International Business Machines Corp., 2000-2004
4 * Portions Copyright (C) Christoph Hellwig, 2001-2002
8 * jfs_logmgr.c: log manager
10 * for related information, see transaction manager (jfs_txnmgr.c), and
11 * recovery manager (jfs_logredo.c).
13 * note: for detail, RTFS.
16 * special purpose buffer manager supporting log i/o requirements.
17 * per log serial pageout of logpage
18 * queuing i/o requests and redrive i/o at iodone
19 * maintain current logpage buffer
20 * no caching since append only
21 * appropriate jfs buffer cache buffers as needed
24 * transactions which wrote COMMIT records in the same in-memory
25 * log page during the pageout of previous/current log page(s) are
26 * committed together by the pageout of the page.
29 * transactions are committed asynchronously when the log page
30 * containing it COMMIT is paged out when it becomes full;
33 * . a per log lock serialize log write.
34 * . a per log lock serialize group commit.
35 * . a per log lock serialize log open/close;
38 * careful-write (ping-pong) of last logpage to recover from crash
40 * detection of split (out-of-order) write of physical sectors
41 * of last logpage via timestamp at end of each sector
42 * with its mirror data array at trailer).
45 * lsn - 64-bit monotonically increasing integer vs
46 * 32-bit lspn and page eor.
50 #include <linux/blkdev.h>
51 #include <linux/interrupt.h>
52 #include <linux/completion.h>
53 #include <linux/kthread.h>
54 #include <linux/buffer_head.h> /* for sync_blockdev() */
55 #include <linux/bio.h>
56 #include <linux/freezer.h>
57 #include <linux/export.h>
58 #include <linux/delay.h>
59 #include <linux/mutex.h>
60 #include <linux/seq_file.h>
61 #include <linux/slab.h>
62 #include "jfs_incore.h"
63 #include "jfs_filsys.h"
64 #include "jfs_metapage.h"
65 #include "jfs_superblock.h"
66 #include "jfs_txnmgr.h"
67 #include "jfs_debug.h"
71 * lbuf's ready to be redriven. Protected by log_redrive_lock (jfsIO thread)
73 static struct lbuf
*log_redrive_list
;
74 static DEFINE_SPINLOCK(log_redrive_lock
);
78 * log read/write serialization (per log)
80 #define LOG_LOCK_INIT(log) mutex_init(&(log)->loglock)
81 #define LOG_LOCK(log) mutex_lock(&((log)->loglock))
82 #define LOG_UNLOCK(log) mutex_unlock(&((log)->loglock))
86 * log group commit serialization (per log)
89 #define LOGGC_LOCK_INIT(log) spin_lock_init(&(log)->gclock)
90 #define LOGGC_LOCK(log) spin_lock_irq(&(log)->gclock)
91 #define LOGGC_UNLOCK(log) spin_unlock_irq(&(log)->gclock)
92 #define LOGGC_WAKEUP(tblk) wake_up_all(&(tblk)->gcwait)
95 * log sync serialization (per log)
97 #define LOGSYNC_DELTA(logsize) min((logsize)/8, 128*LOGPSIZE)
98 #define LOGSYNC_BARRIER(logsize) ((logsize)/4)
100 #define LOGSYNC_DELTA(logsize) min((logsize)/4, 256*LOGPSIZE)
101 #define LOGSYNC_BARRIER(logsize) ((logsize)/2)
106 * log buffer cache synchronization
108 static DEFINE_SPINLOCK(jfsLCacheLock
);
110 #define LCACHE_LOCK(flags) spin_lock_irqsave(&jfsLCacheLock, flags)
111 #define LCACHE_UNLOCK(flags) spin_unlock_irqrestore(&jfsLCacheLock, flags)
114 * See __SLEEP_COND in jfs_locks.h
116 #define LCACHE_SLEEP_COND(wq, cond, flags) \
120 __SLEEP_COND(wq, cond, LCACHE_LOCK(flags), LCACHE_UNLOCK(flags)); \
123 #define LCACHE_WAKEUP(event) wake_up(event)
127 * lbuf buffer cache (lCache) control
129 /* log buffer manager pageout control (cumulative, inclusive) */
130 #define lbmREAD 0x0001
131 #define lbmWRITE 0x0002 /* enqueue at tail of write queue;
132 * init pageout if at head of queue;
134 #define lbmRELEASE 0x0004 /* remove from write queue
135 * at completion of pageout;
136 * do not free/recycle it yet:
137 * caller will free it;
139 #define lbmSYNC 0x0008 /* do not return to freelist
140 * when removed from write queue;
142 #define lbmFREE 0x0010 /* return to freelist
143 * at completion of pageout;
144 * the buffer may be recycled;
146 #define lbmDONE 0x0020
147 #define lbmERROR 0x0040
148 #define lbmGC 0x0080 /* lbmIODone to perform post-GC processing
151 #define lbmDIRECT 0x0100
154 * Global list of active external journals
156 static LIST_HEAD(jfs_external_logs
);
157 static struct jfs_log
*dummy_log
;
158 static DEFINE_MUTEX(jfs_log_mutex
);
163 static int lmWriteRecord(struct jfs_log
* log
, struct tblock
* tblk
,
164 struct lrd
* lrd
, struct tlock
* tlck
);
166 static int lmNextPage(struct jfs_log
* log
);
167 static int lmLogFileSystem(struct jfs_log
* log
, struct jfs_sb_info
*sbi
,
170 static int open_inline_log(struct super_block
*sb
);
171 static int open_dummy_log(struct super_block
*sb
);
172 static int lbmLogInit(struct jfs_log
* log
);
173 static void lbmLogShutdown(struct jfs_log
* log
);
174 static struct lbuf
*lbmAllocate(struct jfs_log
* log
, int);
175 static void lbmFree(struct lbuf
* bp
);
176 static void lbmfree(struct lbuf
* bp
);
177 static int lbmRead(struct jfs_log
* log
, int pn
, struct lbuf
** bpp
);
178 static void lbmWrite(struct jfs_log
* log
, struct lbuf
* bp
, int flag
, int cant_block
);
179 static void lbmDirectWrite(struct jfs_log
* log
, struct lbuf
* bp
, int flag
);
180 static int lbmIOWait(struct lbuf
* bp
, int flag
);
181 static bio_end_io_t lbmIODone
;
182 static void lbmStartIO(struct lbuf
* bp
);
183 static void lmGCwrite(struct jfs_log
* log
, int cant_block
);
184 static int lmLogSync(struct jfs_log
* log
, int hard_sync
);
191 #ifdef CONFIG_JFS_STATISTICS
192 static struct lmStat
{
193 uint commit
; /* # of commit */
194 uint pagedone
; /* # of page written */
195 uint submitted
; /* # of pages submitted */
196 uint full_page
; /* # of full pages submitted */
197 uint partial_page
; /* # of partial pages submitted */
201 static void write_special_inodes(struct jfs_log
*log
,
202 int (*writer
)(struct address_space
*))
204 struct jfs_sb_info
*sbi
;
206 list_for_each_entry(sbi
, &log
->sb_list
, log_list
) {
207 writer(sbi
->ipbmap
->i_mapping
);
208 writer(sbi
->ipimap
->i_mapping
);
209 writer(sbi
->direct_inode
->i_mapping
);
216 * FUNCTION: write a log record;
220 * RETURN: lsn - offset to the next log record to write (end-of-log);
223 * note: todo: log error handler
225 int lmLog(struct jfs_log
* log
, struct tblock
* tblk
, struct lrd
* lrd
,
230 struct metapage
*mp
= NULL
;
233 jfs_info("lmLog: log:0x%p tblk:0x%p, lrd:0x%p tlck:0x%p",
234 log
, tblk
, lrd
, tlck
);
238 /* log by (out-of-transaction) JFS ? */
242 /* log from page ? */
244 tlck
->type
& tlckBTROOT
|| (mp
= tlck
->mp
) == NULL
)
248 * initialize/update page/transaction recovery lsn
252 LOGSYNC_LOCK(log
, flags
);
255 * initialize page lsn if first log write of the page
262 /* insert page at tail of logsynclist */
263 list_add_tail(&mp
->synclist
, &log
->synclist
);
267 * initialize/update lsn of tblock of the page
269 * transaction inherits oldest lsn of pages associated
270 * with allocation/deallocation of resources (their
271 * log records are used to reconstruct allocation map
272 * at recovery time: inode for inode allocation map,
273 * B+-tree index of extent descriptors for block
275 * allocation map pages inherit transaction lsn at
276 * commit time to allow forwarding log syncpt past log
277 * records associated with allocation/deallocation of
278 * resources only after persistent map of these map pages
279 * have been updated and propagated to home.
282 * initialize transaction lsn:
284 if (tblk
->lsn
== 0) {
285 /* inherit lsn of its first page logged */
289 /* insert tblock after the page on logsynclist */
290 list_add(&tblk
->synclist
, &mp
->synclist
);
293 * update transaction lsn:
296 /* inherit oldest/smallest lsn of page */
297 logdiff(diffp
, mp
->lsn
, log
);
298 logdiff(difft
, tblk
->lsn
, log
);
300 /* update tblock lsn with page lsn */
303 /* move tblock after page on logsynclist */
304 list_move(&tblk
->synclist
, &mp
->synclist
);
308 LOGSYNC_UNLOCK(log
, flags
);
311 * write the log record
314 lsn
= lmWriteRecord(log
, tblk
, lrd
, tlck
);
317 * forward log syncpt if log reached next syncpt trigger
319 logdiff(diffp
, lsn
, log
);
320 if (diffp
>= log
->nextsync
)
321 lsn
= lmLogSync(log
, 0);
323 /* update end-of-log lsn */
328 /* return end-of-log address */
333 * NAME: lmWriteRecord()
335 * FUNCTION: move the log record to current log page
337 * PARAMETER: cd - commit descriptor
339 * RETURN: end-of-log address
341 * serialization: LOG_LOCK() held on entry/exit
344 lmWriteRecord(struct jfs_log
* log
, struct tblock
* tblk
, struct lrd
* lrd
,
347 int lsn
= 0; /* end-of-log address */
348 struct lbuf
*bp
; /* dst log page buffer */
349 struct logpage
*lp
; /* dst log page */
350 caddr_t dst
; /* destination address in log page */
351 int dstoffset
; /* end-of-log offset in log page */
352 int freespace
; /* free space in log page */
353 caddr_t p
; /* src meta-data page */
356 int nbytes
; /* number of bytes to move */
359 struct linelock
*linelock
;
366 /* retrieve destination log page to write */
367 bp
= (struct lbuf
*) log
->bp
;
368 lp
= (struct logpage
*) bp
->l_ldata
;
369 dstoffset
= log
->eor
;
371 /* any log data to write ? */
376 * move log record data
378 /* retrieve source meta-data page to log */
379 if (tlck
->flag
& tlckPAGELOCK
) {
380 p
= (caddr_t
) (tlck
->mp
->data
);
381 linelock
= (struct linelock
*) & tlck
->lock
;
383 /* retrieve source in-memory inode to log */
384 else if (tlck
->flag
& tlckINODELOCK
) {
385 if (tlck
->type
& tlckDTREE
)
386 p
= (caddr_t
) &JFS_IP(tlck
->ip
)->i_dtroot
;
388 p
= (caddr_t
) &JFS_IP(tlck
->ip
)->i_xtroot
;
389 linelock
= (struct linelock
*) & tlck
->lock
;
392 jfs_err("lmWriteRecord: UFO tlck:0x%p", tlck
);
393 return 0; /* Probably should trap */
395 l2linesize
= linelock
->l2linesize
;
398 ASSERT(linelock
->index
<= linelock
->maxcnt
);
401 for (i
= 0; i
< linelock
->index
; i
++, lv
++) {
406 if (dstoffset
>= LOGPSIZE
- LOGPTLRSIZE
) {
407 /* page become full: move on to next page */
411 lp
= (struct logpage
*) bp
->l_ldata
;
412 dstoffset
= LOGPHDRSIZE
;
416 * move log vector data
418 src
= (u8
*) p
+ (lv
->offset
<< l2linesize
);
419 srclen
= lv
->length
<< l2linesize
;
422 freespace
= (LOGPSIZE
- LOGPTLRSIZE
) - dstoffset
;
423 nbytes
= min(freespace
, srclen
);
424 dst
= (caddr_t
) lp
+ dstoffset
;
425 memcpy(dst
, src
, nbytes
);
428 /* is page not full ? */
429 if (dstoffset
< LOGPSIZE
- LOGPTLRSIZE
)
432 /* page become full: move on to next page */
435 bp
= (struct lbuf
*) log
->bp
;
436 lp
= (struct logpage
*) bp
->l_ldata
;
437 dstoffset
= LOGPHDRSIZE
;
444 * move log vector descriptor
447 lvd
= (struct lvd
*) ((caddr_t
) lp
+ dstoffset
);
448 lvd
->offset
= cpu_to_le16(lv
->offset
);
449 lvd
->length
= cpu_to_le16(lv
->length
);
451 jfs_info("lmWriteRecord: lv offset:%d length:%d",
452 lv
->offset
, lv
->length
);
455 if ((i
= linelock
->next
)) {
456 linelock
= (struct linelock
*) lid_to_tlock(i
);
461 * move log record descriptor
464 lrd
->length
= cpu_to_le16(len
);
470 freespace
= (LOGPSIZE
- LOGPTLRSIZE
) - dstoffset
;
471 nbytes
= min(freespace
, srclen
);
472 dst
= (caddr_t
) lp
+ dstoffset
;
473 memcpy(dst
, src
, nbytes
);
478 /* are there more to move than freespace of page ? */
483 * end of log record descriptor
486 /* update last log record eor */
487 log
->eor
= dstoffset
;
488 bp
->l_eor
= dstoffset
;
489 lsn
= (log
->page
<< L2LOGPSIZE
) + dstoffset
;
491 if (lrd
->type
& cpu_to_le16(LOG_COMMIT
)) {
493 jfs_info("wr: tclsn:0x%x, beor:0x%x", tblk
->clsn
,
496 INCREMENT(lmStat
.commit
); /* # of commit */
499 * enqueue tblock for group commit:
501 * enqueue tblock of non-trivial/synchronous COMMIT
502 * at tail of group commit queue
503 * (trivial/asynchronous COMMITs are ignored by
508 /* init tblock gc state */
509 tblk
->flag
= tblkGC_QUEUE
;
511 tblk
->pn
= log
->page
;
512 tblk
->eor
= log
->eor
;
514 /* enqueue transaction to commit queue */
515 list_add_tail(&tblk
->cqueue
, &log
->cqueue
);
520 jfs_info("lmWriteRecord: lrd:0x%04x bp:0x%p pn:%d eor:0x%x",
521 le16_to_cpu(lrd
->type
), log
->bp
, log
->page
, dstoffset
);
523 /* page not full ? */
524 if (dstoffset
< LOGPSIZE
- LOGPTLRSIZE
)
528 /* page become full: move on to next page */
531 bp
= (struct lbuf
*) log
->bp
;
532 lp
= (struct logpage
*) bp
->l_ldata
;
533 dstoffset
= LOGPHDRSIZE
;
544 * FUNCTION: write current page and allocate next page.
550 * serialization: LOG_LOCK() held on entry/exit
552 static int lmNextPage(struct jfs_log
* log
)
555 int lspn
; /* log sequence page number */
556 int pn
; /* current page number */
561 /* get current log page number and log sequence page number */
564 lp
= (struct logpage
*) bp
->l_ldata
;
565 lspn
= le32_to_cpu(lp
->h
.page
);
570 * write or queue the full page at the tail of write queue
572 /* get the tail tblk on commit queue */
573 if (list_empty(&log
->cqueue
))
576 tblk
= list_entry(log
->cqueue
.prev
, struct tblock
, cqueue
);
578 /* every tblk who has COMMIT record on the current page,
579 * and has not been committed, must be on commit queue
580 * since tblk is queued at commit queueu at the time
581 * of writing its COMMIT record on the page before
582 * page becomes full (even though the tblk thread
583 * who wrote COMMIT record may have been suspended
587 /* is page bound with outstanding tail tblk ? */
588 if (tblk
&& tblk
->pn
== pn
) {
589 /* mark tblk for end-of-page */
590 tblk
->flag
|= tblkGC_EOP
;
592 if (log
->cflag
& logGC_PAGEOUT
) {
593 /* if page is not already on write queue,
594 * just enqueue (no lbmWRITE to prevent redrive)
595 * buffer to wqueue to ensure correct serial order
596 * of the pages since log pages will be added
599 if (bp
->l_wqnext
== NULL
)
600 lbmWrite(log
, bp
, 0, 0);
603 * No current GC leader, initiate group commit
605 log
->cflag
|= logGC_PAGEOUT
;
609 /* page is not bound with outstanding tblk:
610 * init write or mark it to be redriven (lbmWRITE)
613 /* finalize the page */
614 bp
->l_ceor
= bp
->l_eor
;
615 lp
->h
.eor
= lp
->t
.eor
= cpu_to_le16(bp
->l_ceor
);
616 lbmWrite(log
, bp
, lbmWRITE
| lbmRELEASE
| lbmFREE
, 0);
621 * allocate/initialize next page
623 /* if log wraps, the first data page of log is 2
624 * (0 never used, 1 is superblock).
626 log
->page
= (pn
== log
->size
- 1) ? 2 : pn
+ 1;
627 log
->eor
= LOGPHDRSIZE
; /* ? valid page empty/full at logRedo() */
629 /* allocate/initialize next log page buffer */
630 nextbp
= lbmAllocate(log
, log
->page
);
631 nextbp
->l_eor
= log
->eor
;
634 /* initialize next log page */
635 lp
= (struct logpage
*) nextbp
->l_ldata
;
636 lp
->h
.page
= lp
->t
.page
= cpu_to_le32(lspn
+ 1);
637 lp
->h
.eor
= lp
->t
.eor
= cpu_to_le16(LOGPHDRSIZE
);
644 * NAME: lmGroupCommit()
646 * FUNCTION: group commit
647 * initiate pageout of the pages with COMMIT in the order of
648 * page number - redrive pageout of the page at the head of
649 * pageout queue until full page has been written.
654 * LOGGC_LOCK serializes log group commit queue, and
655 * transaction blocks on the commit queue.
656 * N.B. LOG_LOCK is NOT held during lmGroupCommit().
658 int lmGroupCommit(struct jfs_log
* log
, struct tblock
* tblk
)
664 /* group committed already ? */
665 if (tblk
->flag
& tblkGC_COMMITTED
) {
666 if (tblk
->flag
& tblkGC_ERROR
)
672 jfs_info("lmGroup Commit: tblk = 0x%p, gcrtc = %d", tblk
, log
->gcrtc
);
674 if (tblk
->xflag
& COMMIT_LAZY
)
675 tblk
->flag
|= tblkGC_LAZY
;
677 if ((!(log
->cflag
& logGC_PAGEOUT
)) && (!list_empty(&log
->cqueue
)) &&
678 (!(tblk
->xflag
& COMMIT_LAZY
) || test_bit(log_FLUSH
, &log
->flag
)
679 || jfs_tlocks_low
)) {
681 * No pageout in progress
683 * start group commit as its group leader.
685 log
->cflag
|= logGC_PAGEOUT
;
690 if (tblk
->xflag
& COMMIT_LAZY
) {
692 * Lazy transactions can leave now
698 /* lmGCwrite gives up LOGGC_LOCK, check again */
700 if (tblk
->flag
& tblkGC_COMMITTED
) {
701 if (tblk
->flag
& tblkGC_ERROR
)
708 /* upcount transaction waiting for completion
711 tblk
->flag
|= tblkGC_READY
;
713 __SLEEP_COND(tblk
->gcwait
, (tblk
->flag
& tblkGC_COMMITTED
),
714 LOGGC_LOCK(log
), LOGGC_UNLOCK(log
));
716 /* removed from commit queue */
717 if (tblk
->flag
& tblkGC_ERROR
)
727 * FUNCTION: group commit write
728 * initiate write of log page, building a group of all transactions
729 * with commit records on that page.
734 * LOGGC_LOCK must be held by caller.
735 * N.B. LOG_LOCK is NOT held during lmGroupCommit().
737 static void lmGCwrite(struct jfs_log
* log
, int cant_write
)
741 int gcpn
; /* group commit page number */
743 struct tblock
*xtblk
= NULL
;
746 * build the commit group of a log page
748 * scan commit queue and make a commit group of all
749 * transactions with COMMIT records on the same log page.
751 /* get the head tblk on the commit queue */
752 gcpn
= list_entry(log
->cqueue
.next
, struct tblock
, cqueue
)->pn
;
754 list_for_each_entry(tblk
, &log
->cqueue
, cqueue
) {
755 if (tblk
->pn
!= gcpn
)
760 /* state transition: (QUEUE, READY) -> COMMIT */
761 tblk
->flag
|= tblkGC_COMMIT
;
763 tblk
= xtblk
; /* last tblk of the page */
766 * pageout to commit transactions on the log page.
768 bp
= (struct lbuf
*) tblk
->bp
;
769 lp
= (struct logpage
*) bp
->l_ldata
;
770 /* is page already full ? */
771 if (tblk
->flag
& tblkGC_EOP
) {
772 /* mark page to free at end of group commit of the page */
773 tblk
->flag
&= ~tblkGC_EOP
;
774 tblk
->flag
|= tblkGC_FREE
;
775 bp
->l_ceor
= bp
->l_eor
;
776 lp
->h
.eor
= lp
->t
.eor
= cpu_to_le16(bp
->l_ceor
);
777 lbmWrite(log
, bp
, lbmWRITE
| lbmRELEASE
| lbmGC
,
779 INCREMENT(lmStat
.full_page
);
781 /* page is not yet full */
783 bp
->l_ceor
= tblk
->eor
; /* ? bp->l_ceor = bp->l_eor; */
784 lp
->h
.eor
= lp
->t
.eor
= cpu_to_le16(bp
->l_ceor
);
785 lbmWrite(log
, bp
, lbmWRITE
| lbmGC
, cant_write
);
786 INCREMENT(lmStat
.partial_page
);
793 * FUNCTION: group commit post-processing
794 * Processes transactions after their commit records have been written
795 * to disk, redriving log I/O if necessary.
800 * This routine is called a interrupt time by lbmIODone
802 static void lmPostGC(struct lbuf
* bp
)
805 struct jfs_log
*log
= bp
->l_log
;
807 struct tblock
*tblk
, *temp
;
810 spin_lock_irqsave(&log
->gclock
, flags
);
812 * current pageout of group commit completed.
814 * remove/wakeup transactions from commit queue who were
815 * group committed with the current log page
817 list_for_each_entry_safe(tblk
, temp
, &log
->cqueue
, cqueue
) {
818 if (!(tblk
->flag
& tblkGC_COMMIT
))
820 /* if transaction was marked GC_COMMIT then
821 * it has been shipped in the current pageout
822 * and made it to disk - it is committed.
825 if (bp
->l_flag
& lbmERROR
)
826 tblk
->flag
|= tblkGC_ERROR
;
828 /* remove it from the commit queue */
829 list_del(&tblk
->cqueue
);
830 tblk
->flag
&= ~tblkGC_QUEUE
;
832 if (tblk
== log
->flush_tblk
) {
833 /* we can stop flushing the log now */
834 clear_bit(log_FLUSH
, &log
->flag
);
835 log
->flush_tblk
= NULL
;
838 jfs_info("lmPostGC: tblk = 0x%p, flag = 0x%x", tblk
,
841 if (!(tblk
->xflag
& COMMIT_FORCE
))
843 * Hand tblk over to lazy commit thread
847 /* state transition: COMMIT -> COMMITTED */
848 tblk
->flag
|= tblkGC_COMMITTED
;
850 if (tblk
->flag
& tblkGC_READY
)
856 /* was page full before pageout ?
857 * (and this is the last tblk bound with the page)
859 if (tblk
->flag
& tblkGC_FREE
)
861 /* did page become full after pageout ?
862 * (and this is the last tblk bound with the page)
864 else if (tblk
->flag
& tblkGC_EOP
) {
865 /* finalize the page */
866 lp
= (struct logpage
*) bp
->l_ldata
;
867 bp
->l_ceor
= bp
->l_eor
;
868 lp
->h
.eor
= lp
->t
.eor
= cpu_to_le16(bp
->l_eor
);
869 jfs_info("lmPostGC: calling lbmWrite");
870 lbmWrite(log
, bp
, lbmWRITE
| lbmRELEASE
| lbmFREE
,
876 /* are there any transactions who have entered lnGroupCommit()
877 * (whose COMMITs are after that of the last log page written.
878 * They are waiting for new group commit (above at (SLEEP 1))
879 * or lazy transactions are on a full (queued) log page,
880 * select the latest ready transaction as new group leader and
881 * wake her up to lead her group.
883 if ((!list_empty(&log
->cqueue
)) &&
884 ((log
->gcrtc
> 0) || (tblk
->bp
->l_wqnext
!= NULL
) ||
885 test_bit(log_FLUSH
, &log
->flag
) || jfs_tlocks_low
))
887 * Call lmGCwrite with new group leader
891 /* no transaction are ready yet (transactions are only just
892 * queued (GC_QUEUE) and not entered for group commit yet).
893 * the first transaction entering group commit
894 * will elect herself as new group leader.
897 log
->cflag
&= ~logGC_PAGEOUT
;
900 spin_unlock_irqrestore(&log
->gclock
, flags
);
907 * FUNCTION: write log SYNCPT record for specified log
908 * if new sync address is available
909 * (normally the case if sync() is executed by back-ground
911 * calculate new value of i_nextsync which determines when
912 * this code is called again.
914 * PARAMETERS: log - log structure
915 * hard_sync - 1 to force all metadata to be written
919 * serialization: LOG_LOCK() held on entry/exit
921 static int lmLogSync(struct jfs_log
* log
, int hard_sync
)
924 int written
; /* written since last syncpt */
925 int free
; /* free space left available */
926 int delta
; /* additional delta to write normally */
927 int more
; /* additional write granted */
930 struct logsyncblk
*lp
;
933 /* push dirty metapages out to disk */
935 write_special_inodes(log
, filemap_fdatawrite
);
937 write_special_inodes(log
, filemap_flush
);
942 /* if last sync is same as last syncpt,
943 * invoke sync point forward processing to update sync.
946 if (log
->sync
== log
->syncpt
) {
947 LOGSYNC_LOCK(log
, flags
);
948 if (list_empty(&log
->synclist
))
949 log
->sync
= log
->lsn
;
951 lp
= list_entry(log
->synclist
.next
,
952 struct logsyncblk
, synclist
);
955 LOGSYNC_UNLOCK(log
, flags
);
959 /* if sync is different from last syncpt,
960 * write a SYNCPT record with syncpt = sync.
961 * reset syncpt = sync
963 if (log
->sync
!= log
->syncpt
) {
966 lrd
.type
= cpu_to_le16(LOG_SYNCPT
);
968 lrd
.log
.syncpt
.sync
= cpu_to_le32(log
->sync
);
969 lsn
= lmWriteRecord(log
, NULL
, &lrd
, NULL
);
971 log
->syncpt
= log
->sync
;
976 * setup next syncpt trigger (SWAG)
978 logsize
= log
->logsize
;
980 logdiff(written
, lsn
, log
);
981 free
= logsize
- written
;
982 delta
= LOGSYNC_DELTA(logsize
);
983 more
= min(free
/ 2, delta
);
984 if (more
< 2 * LOGPSIZE
) {
985 jfs_warn("\n ... Log Wrap ... Log Wrap ... Log Wrap ...\n");
989 * option 1 - panic ? No.!
990 * option 2 - shutdown file systems
991 * associated with log ?
992 * option 3 - extend log ?
993 * option 4 - second chance
995 * mark log wrapped, and continue.
996 * when all active transactions are completed,
997 * mark log valid for recovery.
998 * if crashed during invalid state, log state
999 * implies invalid log, forcing fsck().
1001 /* mark log state log wrap in log superblock */
1002 /* log->state = LOGWRAP; */
1004 /* reset sync point computation */
1005 log
->syncpt
= log
->sync
= lsn
;
1006 log
->nextsync
= delta
;
1008 /* next syncpt trigger = written + more */
1009 log
->nextsync
= written
+ more
;
1011 /* if number of bytes written from last sync point is more
1012 * than 1/4 of the log size, stop new transactions from
1013 * starting until all current transactions are completed
1014 * by setting syncbarrier flag.
1016 if (!test_bit(log_SYNCBARRIER
, &log
->flag
) &&
1017 (written
> LOGSYNC_BARRIER(logsize
)) && log
->active
) {
1018 set_bit(log_SYNCBARRIER
, &log
->flag
);
1019 jfs_info("log barrier on: lsn=0x%x syncpt=0x%x", lsn
,
1022 * We may have to initiate group commit
1024 jfs_flush_journal(log
, 0);
1033 * FUNCTION: write log SYNCPT record for specified log
1035 * PARAMETERS: log - log structure
1036 * hard_sync - set to 1 to force metadata to be written
1038 void jfs_syncpt(struct jfs_log
*log
, int hard_sync
)
1040 if (!test_bit(log_QUIESCE
, &log
->flag
))
1041 lmLogSync(log
, hard_sync
);
1048 * FUNCTION: open the log on first open;
1049 * insert filesystem in the active list of the log.
1051 * PARAMETER: ipmnt - file system mount inode
1052 * iplog - log inode (out)
1058 int lmLogOpen(struct super_block
*sb
)
1061 struct file
*bdev_file
;
1062 struct jfs_log
*log
;
1063 struct jfs_sb_info
*sbi
= JFS_SBI(sb
);
1065 if (sbi
->flag
& JFS_NOINTEGRITY
)
1066 return open_dummy_log(sb
);
1068 if (sbi
->mntflag
& JFS_INLINELOG
)
1069 return open_inline_log(sb
);
1071 mutex_lock(&jfs_log_mutex
);
1072 list_for_each_entry(log
, &jfs_external_logs
, journal_list
) {
1073 if (file_bdev(log
->bdev_file
)->bd_dev
== sbi
->logdev
) {
1074 if (!uuid_equal(&log
->uuid
, &sbi
->loguuid
)) {
1075 jfs_warn("wrong uuid on JFS journal");
1076 mutex_unlock(&jfs_log_mutex
);
1080 * add file system to log active file system list
1082 if ((rc
= lmLogFileSystem(log
, sbi
, 1))) {
1083 mutex_unlock(&jfs_log_mutex
);
1090 if (!(log
= kzalloc(sizeof(struct jfs_log
), GFP_KERNEL
))) {
1091 mutex_unlock(&jfs_log_mutex
);
1094 INIT_LIST_HEAD(&log
->sb_list
);
1095 init_waitqueue_head(&log
->syncwait
);
1098 * external log as separate logical volume
1100 * file systems to log may have n-to-1 relationship;
1103 bdev_file
= bdev_file_open_by_dev(sbi
->logdev
,
1104 BLK_OPEN_READ
| BLK_OPEN_WRITE
, log
, NULL
);
1105 if (IS_ERR(bdev_file
)) {
1106 rc
= PTR_ERR(bdev_file
);
1110 log
->bdev_file
= bdev_file
;
1111 uuid_copy(&log
->uuid
, &sbi
->loguuid
);
1116 if ((rc
= lmLogInit(log
)))
1119 list_add(&log
->journal_list
, &jfs_external_logs
);
1122 * add file system to log active file system list
1124 if ((rc
= lmLogFileSystem(log
, sbi
, 1)))
1129 list_add(&sbi
->log_list
, &log
->sb_list
);
1133 mutex_unlock(&jfs_log_mutex
);
1139 shutdown
: /* unwind lbmLogInit() */
1140 list_del(&log
->journal_list
);
1141 lbmLogShutdown(log
);
1143 close
: /* close external log device */
1144 bdev_fput(bdev_file
);
1146 free
: /* free log descriptor */
1147 mutex_unlock(&jfs_log_mutex
);
1150 jfs_warn("lmLogOpen: exit(%d)", rc
);
1154 static int open_inline_log(struct super_block
*sb
)
1156 struct jfs_log
*log
;
1159 if (!(log
= kzalloc(sizeof(struct jfs_log
), GFP_KERNEL
)))
1161 INIT_LIST_HEAD(&log
->sb_list
);
1162 init_waitqueue_head(&log
->syncwait
);
1164 set_bit(log_INLINELOG
, &log
->flag
);
1165 log
->bdev_file
= sb
->s_bdev_file
;
1166 log
->base
= addressPXD(&JFS_SBI(sb
)->logpxd
);
1167 log
->size
= lengthPXD(&JFS_SBI(sb
)->logpxd
) >>
1168 (L2LOGPSIZE
- sb
->s_blocksize_bits
);
1169 log
->l2bsize
= sb
->s_blocksize_bits
;
1170 ASSERT(L2LOGPSIZE
>= sb
->s_blocksize_bits
);
1175 if ((rc
= lmLogInit(log
))) {
1177 jfs_warn("lmLogOpen: exit(%d)", rc
);
1181 list_add(&JFS_SBI(sb
)->log_list
, &log
->sb_list
);
1182 JFS_SBI(sb
)->log
= log
;
1187 static int open_dummy_log(struct super_block
*sb
)
1191 mutex_lock(&jfs_log_mutex
);
1193 dummy_log
= kzalloc(sizeof(struct jfs_log
), GFP_KERNEL
);
1195 mutex_unlock(&jfs_log_mutex
);
1198 INIT_LIST_HEAD(&dummy_log
->sb_list
);
1199 init_waitqueue_head(&dummy_log
->syncwait
);
1200 dummy_log
->no_integrity
= 1;
1201 /* Make up some stuff */
1202 dummy_log
->base
= 0;
1203 dummy_log
->size
= 1024;
1204 rc
= lmLogInit(dummy_log
);
1208 mutex_unlock(&jfs_log_mutex
);
1213 LOG_LOCK(dummy_log
);
1214 list_add(&JFS_SBI(sb
)->log_list
, &dummy_log
->sb_list
);
1215 JFS_SBI(sb
)->log
= dummy_log
;
1216 LOG_UNLOCK(dummy_log
);
1217 mutex_unlock(&jfs_log_mutex
);
1225 * FUNCTION: log initialization at first log open.
1227 * logredo() (or logformat()) should have been run previously.
1228 * initialize the log from log superblock.
1229 * set the log state in the superblock to LOGMOUNT and
1230 * write SYNCPT log record.
1232 * PARAMETER: log - log structure
1235 * -EINVAL - bad log magic number or superblock dirty
1236 * error returned from logwait()
1238 * serialization: single first open thread
1240 int lmLogInit(struct jfs_log
* log
)
1244 struct logsuper
*logsuper
;
1245 struct lbuf
*bpsuper
;
1250 jfs_info("lmLogInit: log:0x%p", log
);
1252 /* initialize the group commit serialization lock */
1253 LOGGC_LOCK_INIT(log
);
1255 /* allocate/initialize the log write serialization lock */
1258 LOGSYNC_LOCK_INIT(log
);
1260 INIT_LIST_HEAD(&log
->synclist
);
1262 INIT_LIST_HEAD(&log
->cqueue
);
1263 log
->flush_tblk
= NULL
;
1268 * initialize log i/o
1270 if ((rc
= lbmLogInit(log
)))
1273 if (!test_bit(log_INLINELOG
, &log
->flag
))
1274 log
->l2bsize
= L2LOGPSIZE
;
1276 /* check for disabled journaling to disk */
1277 if (log
->no_integrity
) {
1279 * Journal pages will still be filled. When the time comes
1280 * to actually do the I/O, the write is not done, and the
1281 * endio routine is called directly.
1283 bp
= lbmAllocate(log
, 0);
1285 bp
->l_pn
= bp
->l_eor
= 0;
1288 * validate log superblock
1290 if ((rc
= lbmRead(log
, 1, &bpsuper
)))
1293 logsuper
= (struct logsuper
*) bpsuper
->l_ldata
;
1295 if (logsuper
->magic
!= cpu_to_le32(LOGMAGIC
)) {
1296 jfs_warn("*** Log Format Error ! ***");
1301 /* logredo() should have been run successfully. */
1302 if (logsuper
->state
!= cpu_to_le32(LOGREDONE
)) {
1303 jfs_warn("*** Log Is Dirty ! ***");
1308 /* initialize log from log superblock */
1309 if (test_bit(log_INLINELOG
,&log
->flag
)) {
1310 if (log
->size
!= le32_to_cpu(logsuper
->size
)) {
1314 jfs_info("lmLogInit: inline log:0x%p base:0x%Lx size:0x%x",
1315 log
, (unsigned long long)log
->base
, log
->size
);
1317 if (!uuid_equal(&logsuper
->uuid
, &log
->uuid
)) {
1318 jfs_warn("wrong uuid on JFS log device");
1322 log
->size
= le32_to_cpu(logsuper
->size
);
1323 log
->l2bsize
= le32_to_cpu(logsuper
->l2bsize
);
1324 jfs_info("lmLogInit: external log:0x%p base:0x%Lx size:0x%x",
1325 log
, (unsigned long long)log
->base
, log
->size
);
1328 log
->page
= le32_to_cpu(logsuper
->end
) / LOGPSIZE
;
1329 log
->eor
= le32_to_cpu(logsuper
->end
) - (LOGPSIZE
* log
->page
);
1332 * initialize for log append write mode
1334 /* establish current/end-of-log page/buffer */
1335 if ((rc
= lbmRead(log
, log
->page
, &bp
)))
1338 lp
= (struct logpage
*) bp
->l_ldata
;
1340 jfs_info("lmLogInit: lsn:0x%x page:%d eor:%d:%d",
1341 le32_to_cpu(logsuper
->end
), log
->page
, log
->eor
,
1342 le16_to_cpu(lp
->h
.eor
));
1345 bp
->l_pn
= log
->page
;
1346 bp
->l_eor
= log
->eor
;
1348 /* if current page is full, move on to next page */
1349 if (log
->eor
>= LOGPSIZE
- LOGPTLRSIZE
)
1353 * initialize log syncpoint
1356 * write the first SYNCPT record with syncpoint = 0
1357 * (i.e., log redo up to HERE !);
1358 * remove current page from lbm write queue at end of pageout
1359 * (to write log superblock update), but do not release to
1364 lrd
.type
= cpu_to_le16(LOG_SYNCPT
);
1366 lrd
.log
.syncpt
.sync
= 0;
1367 lsn
= lmWriteRecord(log
, NULL
, &lrd
, NULL
);
1369 bp
->l_ceor
= bp
->l_eor
;
1370 lp
= (struct logpage
*) bp
->l_ldata
;
1371 lp
->h
.eor
= lp
->t
.eor
= cpu_to_le16(bp
->l_eor
);
1372 lbmWrite(log
, bp
, lbmWRITE
| lbmSYNC
, 0);
1373 if ((rc
= lbmIOWait(bp
, 0)))
1377 * update/write superblock
1379 logsuper
->state
= cpu_to_le32(LOGMOUNT
);
1380 log
->serial
= le32_to_cpu(logsuper
->serial
) + 1;
1381 logsuper
->serial
= cpu_to_le32(log
->serial
);
1382 lbmDirectWrite(log
, bpsuper
, lbmWRITE
| lbmRELEASE
| lbmSYNC
);
1383 if ((rc
= lbmIOWait(bpsuper
, lbmFREE
)))
1387 /* initialize logsync parameters */
1388 log
->logsize
= (log
->size
- 2) << L2LOGPSIZE
;
1391 log
->sync
= log
->syncpt
;
1392 log
->nextsync
= LOGSYNC_DELTA(log
->logsize
);
1394 jfs_info("lmLogInit: lsn:0x%x syncpt:0x%x sync:0x%x",
1395 log
->lsn
, log
->syncpt
, log
->sync
);
1398 * initialize for lazy/group commit
1407 errout30
: /* release log page */
1409 bp
->l_wqnext
= NULL
;
1412 errout20
: /* release log superblock */
1415 errout10
: /* unwind lbmLogInit() */
1416 lbmLogShutdown(log
);
1418 jfs_warn("lmLogInit: exit(%d)", rc
);
1424 * NAME: lmLogClose()
1426 * FUNCTION: remove file system <ipmnt> from active list of log <iplog>
1427 * and close it on last close.
1429 * PARAMETER: sb - superblock
1431 * RETURN: errors from subroutines
1435 int lmLogClose(struct super_block
*sb
)
1437 struct jfs_sb_info
*sbi
= JFS_SBI(sb
);
1438 struct jfs_log
*log
= sbi
->log
;
1439 struct file
*bdev_file
;
1442 jfs_info("lmLogClose: log:0x%p", log
);
1444 mutex_lock(&jfs_log_mutex
);
1446 list_del(&sbi
->log_list
);
1451 * We need to make sure all of the "written" metapages
1452 * actually make it to disk
1454 sync_blockdev(sb
->s_bdev
);
1456 if (test_bit(log_INLINELOG
, &log
->flag
)) {
1458 * in-line log in host file system
1460 rc
= lmLogShutdown(log
);
1465 if (!log
->no_integrity
)
1466 lmLogFileSystem(log
, sbi
, 0);
1468 if (!list_empty(&log
->sb_list
))
1472 * TODO: ensure that the dummy_log is in a state to allow
1473 * lbmLogShutdown to deallocate all the buffers and call
1474 * kfree against dummy_log. For now, leave dummy_log & its
1475 * buffers in memory, and resuse if another no-integrity mount
1478 if (log
->no_integrity
)
1482 * external log as separate logical volume
1484 list_del(&log
->journal_list
);
1485 bdev_file
= log
->bdev_file
;
1486 rc
= lmLogShutdown(log
);
1488 bdev_fput(bdev_file
);
1493 mutex_unlock(&jfs_log_mutex
);
1494 jfs_info("lmLogClose: exit(%d)", rc
);
1500 * NAME: jfs_flush_journal()
1502 * FUNCTION: initiate write of any outstanding transactions to the journal
1503 * and optionally wait until they are all written to disk
1505 * wait == 0 flush until latest txn is committed, don't wait
1506 * wait == 1 flush until latest txn is committed, wait
1507 * wait > 1 flush until all txn's are complete, wait
1509 void jfs_flush_journal(struct jfs_log
*log
, int wait
)
1512 struct tblock
*target
= NULL
;
1514 /* jfs_write_inode may call us during read-only mount */
1518 jfs_info("jfs_flush_journal: log:0x%p wait=%d", log
, wait
);
1522 if (!list_empty(&log
->cqueue
)) {
1524 * This ensures that we will keep writing to the journal as long
1525 * as there are unwritten commit records
1527 target
= list_entry(log
->cqueue
.prev
, struct tblock
, cqueue
);
1529 if (test_bit(log_FLUSH
, &log
->flag
)) {
1531 * We're already flushing.
1532 * if flush_tblk is NULL, we are flushing everything,
1533 * so leave it that way. Otherwise, update it to the
1534 * latest transaction
1536 if (log
->flush_tblk
)
1537 log
->flush_tblk
= target
;
1539 /* Only flush until latest transaction is committed */
1540 log
->flush_tblk
= target
;
1541 set_bit(log_FLUSH
, &log
->flag
);
1544 * Initiate I/O on outstanding transactions
1546 if (!(log
->cflag
& logGC_PAGEOUT
)) {
1547 log
->cflag
|= logGC_PAGEOUT
;
1552 if ((wait
> 1) || test_bit(log_SYNCBARRIER
, &log
->flag
)) {
1553 /* Flush until all activity complete */
1554 set_bit(log_FLUSH
, &log
->flag
);
1555 log
->flush_tblk
= NULL
;
1558 if (wait
&& target
&& !(target
->flag
& tblkGC_COMMITTED
)) {
1559 DECLARE_WAITQUEUE(__wait
, current
);
1561 add_wait_queue(&target
->gcwait
, &__wait
);
1562 set_current_state(TASK_UNINTERRUPTIBLE
);
1566 remove_wait_queue(&target
->gcwait
, &__wait
);
1573 write_special_inodes(log
, filemap_fdatawrite
);
1576 * If there was recent activity, we may need to wait
1577 * for the lazycommit thread to catch up
1579 if ((!list_empty(&log
->cqueue
)) || !list_empty(&log
->synclist
)) {
1580 for (i
= 0; i
< 200; i
++) { /* Too much? */
1582 write_special_inodes(log
, filemap_fdatawrite
);
1583 if (list_empty(&log
->cqueue
) &&
1584 list_empty(&log
->synclist
))
1588 assert(list_empty(&log
->cqueue
));
1590 #ifdef CONFIG_JFS_DEBUG
1591 if (!list_empty(&log
->synclist
)) {
1592 struct logsyncblk
*lp
;
1594 printk(KERN_ERR
"jfs_flush_journal: synclist not empty\n");
1595 list_for_each_entry(lp
, &log
->synclist
, synclist
) {
1596 if (lp
->xflag
& COMMIT_PAGE
) {
1597 struct metapage
*mp
= (struct metapage
*)lp
;
1598 print_hex_dump(KERN_ERR
, "metapage: ",
1599 DUMP_PREFIX_ADDRESS
, 16, 4,
1600 mp
, sizeof(struct metapage
), 0);
1601 print_hex_dump(KERN_ERR
, "page: ",
1602 DUMP_PREFIX_ADDRESS
, 16,
1603 sizeof(long), mp
->folio
,
1604 sizeof(struct page
), 0);
1606 print_hex_dump(KERN_ERR
, "tblock:",
1607 DUMP_PREFIX_ADDRESS
, 16, 4,
1608 lp
, sizeof(struct tblock
), 0);
1612 WARN_ON(!list_empty(&log
->synclist
));
1614 clear_bit(log_FLUSH
, &log
->flag
);
1618 * NAME: lmLogShutdown()
1620 * FUNCTION: log shutdown at last LogClose().
1622 * write log syncpt record.
1623 * update super block to set redone flag to 0.
1625 * PARAMETER: log - log inode
1627 * RETURN: 0 - success
1629 * serialization: single last close thread
1631 int lmLogShutdown(struct jfs_log
* log
)
1636 struct logsuper
*logsuper
;
1637 struct lbuf
*bpsuper
;
1641 jfs_info("lmLogShutdown: log:0x%p", log
);
1643 jfs_flush_journal(log
, 2);
1646 * write the last SYNCPT record with syncpoint = 0
1647 * (i.e., log redo up to HERE !)
1651 lrd
.type
= cpu_to_le16(LOG_SYNCPT
);
1653 lrd
.log
.syncpt
.sync
= 0;
1655 lsn
= lmWriteRecord(log
, NULL
, &lrd
, NULL
);
1657 lp
= (struct logpage
*) bp
->l_ldata
;
1658 lp
->h
.eor
= lp
->t
.eor
= cpu_to_le16(bp
->l_eor
);
1659 lbmWrite(log
, log
->bp
, lbmWRITE
| lbmRELEASE
| lbmSYNC
, 0);
1660 lbmIOWait(log
->bp
, lbmFREE
);
1664 * synchronous update log superblock
1665 * mark log state as shutdown cleanly
1666 * (i.e., Log does not need to be replayed).
1668 if ((rc
= lbmRead(log
, 1, &bpsuper
)))
1671 logsuper
= (struct logsuper
*) bpsuper
->l_ldata
;
1672 logsuper
->state
= cpu_to_le32(LOGREDONE
);
1673 logsuper
->end
= cpu_to_le32(lsn
);
1674 lbmDirectWrite(log
, bpsuper
, lbmWRITE
| lbmRELEASE
| lbmSYNC
);
1675 rc
= lbmIOWait(bpsuper
, lbmFREE
);
1677 jfs_info("lmLogShutdown: lsn:0x%x page:%d eor:%d",
1678 lsn
, log
->page
, log
->eor
);
1682 * shutdown per log i/o
1684 lbmLogShutdown(log
);
1687 jfs_warn("lmLogShutdown: exit(%d)", rc
);
1694 * NAME: lmLogFileSystem()
1696 * FUNCTION: insert (<activate> = true)/remove (<activate> = false)
1697 * file system into/from log active file system list.
1699 * PARAMETE: log - pointer to logs inode.
1700 * fsdev - kdev_t of filesystem.
1701 * serial - pointer to returned log serial number
1702 * activate - insert/remove device from active list.
1704 * RETURN: 0 - success
1705 * errors returned by vms_iowait().
1707 static int lmLogFileSystem(struct jfs_log
* log
, struct jfs_sb_info
*sbi
,
1712 struct logsuper
*logsuper
;
1713 struct lbuf
*bpsuper
;
1714 uuid_t
*uuid
= &sbi
->uuid
;
1717 * insert/remove file system device to log active file system list.
1719 if ((rc
= lbmRead(log
, 1, &bpsuper
)))
1722 logsuper
= (struct logsuper
*) bpsuper
->l_ldata
;
1724 for (i
= 0; i
< MAX_ACTIVE
; i
++)
1725 if (uuid_is_null(&logsuper
->active
[i
].uuid
)) {
1726 uuid_copy(&logsuper
->active
[i
].uuid
, uuid
);
1730 if (i
== MAX_ACTIVE
) {
1731 jfs_warn("Too many file systems sharing journal!");
1733 return -EMFILE
; /* Is there a better rc? */
1736 for (i
= 0; i
< MAX_ACTIVE
; i
++)
1737 if (uuid_equal(&logsuper
->active
[i
].uuid
, uuid
)) {
1738 uuid_copy(&logsuper
->active
[i
].uuid
,
1742 if (i
== MAX_ACTIVE
) {
1743 jfs_warn("Somebody stomped on the journal!");
1751 * synchronous write log superblock:
1753 * write sidestream bypassing write queue:
1754 * at file system mount, log super block is updated for
1755 * activation of the file system before any log record
1756 * (MOUNT record) of the file system, and at file system
1757 * unmount, all meta data for the file system has been
1758 * flushed before log super block is updated for deactivation
1759 * of the file system.
1761 lbmDirectWrite(log
, bpsuper
, lbmWRITE
| lbmRELEASE
| lbmSYNC
);
1762 rc
= lbmIOWait(bpsuper
, lbmFREE
);
1768 * log buffer manager (lbm)
1769 * ------------------------
1771 * special purpose buffer manager supporting log i/o requirements.
1773 * per log write queue:
1774 * log pageout occurs in serial order by fifo write queue and
1775 * restricting to a single i/o in pregress at any one time.
1776 * a circular singly-linked list
1777 * (log->wrqueue points to the tail, and buffers are linked via
1778 * bp->wrqueue field), and
1779 * maintains log page in pageout ot waiting for pageout in serial pageout.
1785 * initialize per log I/O setup at lmLogInit()
1787 static int lbmLogInit(struct jfs_log
* log
)
1792 jfs_info("lbmLogInit: log:0x%p", log
);
1794 /* initialize current buffer cursor */
1797 /* initialize log device write queue */
1801 * Each log has its own buffer pages allocated to it. These are
1802 * not managed by the page cache. This ensures that a transaction
1803 * writing to the log does not block trying to allocate a page from
1804 * the page cache (for the log). This would be bad, since page
1805 * allocation waits on the kswapd thread that may be committing inodes
1806 * which would cause log activity. Was that clear? I'm trying to
1807 * avoid deadlock here.
1809 init_waitqueue_head(&log
->free_wait
);
1811 log
->lbuf_free
= NULL
;
1813 for (i
= 0; i
< LOGPAGES
;) {
1816 struct page
*page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
1820 buffer
= page_address(page
);
1821 for (offset
= 0; offset
< PAGE_SIZE
; offset
+= LOGPSIZE
) {
1822 lbuf
= kmalloc(sizeof(struct lbuf
), GFP_KERNEL
);
1828 if (offset
) /* we already have one reference */
1830 lbuf
->l_offset
= offset
;
1831 lbuf
->l_ldata
= buffer
+ offset
;
1832 lbuf
->l_page
= page
;
1834 init_waitqueue_head(&lbuf
->l_ioevent
);
1836 lbuf
->l_freelist
= log
->lbuf_free
;
1837 log
->lbuf_free
= lbuf
;
1845 lbmLogShutdown(log
);
1853 * finalize per log I/O setup at lmLogShutdown()
1855 static void lbmLogShutdown(struct jfs_log
* log
)
1859 jfs_info("lbmLogShutdown: log:0x%p", log
);
1861 lbuf
= log
->lbuf_free
;
1863 struct lbuf
*next
= lbuf
->l_freelist
;
1864 __free_page(lbuf
->l_page
);
1874 * allocate an empty log buffer
1876 static struct lbuf
*lbmAllocate(struct jfs_log
* log
, int pn
)
1879 unsigned long flags
;
1882 * recycle from log buffer freelist if any
1885 LCACHE_SLEEP_COND(log
->free_wait
, (bp
= log
->lbuf_free
), flags
);
1886 log
->lbuf_free
= bp
->l_freelist
;
1887 LCACHE_UNLOCK(flags
);
1891 bp
->l_wqnext
= NULL
;
1892 bp
->l_freelist
= NULL
;
1895 bp
->l_blkno
= log
->base
+ (pn
<< (L2LOGPSIZE
- log
->l2bsize
));
1905 * release a log buffer to freelist
1907 static void lbmFree(struct lbuf
* bp
)
1909 unsigned long flags
;
1915 LCACHE_UNLOCK(flags
);
1918 static void lbmfree(struct lbuf
* bp
)
1920 struct jfs_log
*log
= bp
->l_log
;
1922 assert(bp
->l_wqnext
== NULL
);
1925 * return the buffer to head of freelist
1927 bp
->l_freelist
= log
->lbuf_free
;
1928 log
->lbuf_free
= bp
;
1930 wake_up(&log
->free_wait
);
1938 * FUNCTION: add a log buffer to the log redrive list
1944 * Takes log_redrive_lock.
1946 static inline void lbmRedrive(struct lbuf
*bp
)
1948 unsigned long flags
;
1950 spin_lock_irqsave(&log_redrive_lock
, flags
);
1951 bp
->l_redrive_next
= log_redrive_list
;
1952 log_redrive_list
= bp
;
1953 spin_unlock_irqrestore(&log_redrive_lock
, flags
);
1955 wake_up_process(jfsIOthread
);
1962 static int lbmRead(struct jfs_log
* log
, int pn
, struct lbuf
** bpp
)
1968 * allocate a log buffer
1970 *bpp
= bp
= lbmAllocate(log
, pn
);
1971 jfs_info("lbmRead: bp:0x%p pn:0x%x", bp
, pn
);
1973 bp
->l_flag
|= lbmREAD
;
1975 bio
= bio_alloc(file_bdev(log
->bdev_file
), 1, REQ_OP_READ
, GFP_NOFS
);
1976 bio
->bi_iter
.bi_sector
= bp
->l_blkno
<< (log
->l2bsize
- 9);
1977 __bio_add_page(bio
, bp
->l_page
, LOGPSIZE
, bp
->l_offset
);
1978 BUG_ON(bio
->bi_iter
.bi_size
!= LOGPSIZE
);
1980 bio
->bi_end_io
= lbmIODone
;
1981 bio
->bi_private
= bp
;
1982 /*check if journaling to disk has been disabled*/
1983 if (log
->no_integrity
) {
1984 bio
->bi_iter
.bi_size
= 0;
1990 wait_event(bp
->l_ioevent
, (bp
->l_flag
!= lbmREAD
));
1999 * buffer at head of pageout queue stays after completion of
2000 * partial-page pageout and redriven by explicit initiation of
2001 * pageout by caller until full-page pageout is completed and
2004 * device driver i/o done redrives pageout of new buffer at
2005 * head of pageout queue when current buffer at head of pageout
2006 * queue is released at the completion of its full-page pageout.
2008 * LOGGC_LOCK() serializes lbmWrite() by lmNextPage() and lmGroupCommit().
2009 * LCACHE_LOCK() serializes xflag between lbmWrite() and lbmIODone()
2011 static void lbmWrite(struct jfs_log
* log
, struct lbuf
* bp
, int flag
,
2015 unsigned long flags
;
2017 jfs_info("lbmWrite: bp:0x%p flag:0x%x pn:0x%x", bp
, flag
, bp
->l_pn
);
2019 /* map the logical block address to physical block address */
2021 log
->base
+ (bp
->l_pn
<< (L2LOGPSIZE
- log
->l2bsize
));
2023 LCACHE_LOCK(flags
); /* disable+lock */
2026 * initialize buffer for device driver
2031 * insert bp at tail of write queue associated with log
2033 * (request is either for bp already/currently at head of queue
2034 * or new bp to be inserted at tail)
2038 /* is buffer not already on write queue ? */
2039 if (bp
->l_wqnext
== NULL
) {
2040 /* insert at tail of wqueue */
2046 bp
->l_wqnext
= tail
->l_wqnext
;
2047 tail
->l_wqnext
= bp
;
2053 /* is buffer at head of wqueue and for write ? */
2054 if ((bp
!= tail
->l_wqnext
) || !(flag
& lbmWRITE
)) {
2055 LCACHE_UNLOCK(flags
); /* unlock+enable */
2059 LCACHE_UNLOCK(flags
); /* unlock+enable */
2063 else if (flag
& lbmSYNC
)
2076 * initiate pageout bypassing write queue for sidestream
2077 * (e.g., log superblock) write;
2079 static void lbmDirectWrite(struct jfs_log
* log
, struct lbuf
* bp
, int flag
)
2081 jfs_info("lbmDirectWrite: bp:0x%p flag:0x%x pn:0x%x",
2082 bp
, flag
, bp
->l_pn
);
2085 * initialize buffer for device driver
2087 bp
->l_flag
= flag
| lbmDIRECT
;
2089 /* map the logical block address to physical block address */
2091 log
->base
+ (bp
->l_pn
<< (L2LOGPSIZE
- log
->l2bsize
));
2094 * initiate pageout of the page
2101 * NAME: lbmStartIO()
2103 * FUNCTION: Interface to DD strategy routine
2107 * serialization: LCACHE_LOCK() is NOT held during log i/o;
2109 static void lbmStartIO(struct lbuf
* bp
)
2112 struct jfs_log
*log
= bp
->l_log
;
2113 struct block_device
*bdev
= NULL
;
2115 jfs_info("lbmStartIO");
2117 if (!log
->no_integrity
)
2118 bdev
= file_bdev(log
->bdev_file
);
2120 bio
= bio_alloc(bdev
, 1, REQ_OP_WRITE
| REQ_SYNC
,
2122 bio
->bi_iter
.bi_sector
= bp
->l_blkno
<< (log
->l2bsize
- 9);
2123 __bio_add_page(bio
, bp
->l_page
, LOGPSIZE
, bp
->l_offset
);
2124 BUG_ON(bio
->bi_iter
.bi_size
!= LOGPSIZE
);
2126 bio
->bi_end_io
= lbmIODone
;
2127 bio
->bi_private
= bp
;
2129 /* check if journaling to disk has been disabled */
2130 if (log
->no_integrity
) {
2131 bio
->bi_iter
.bi_size
= 0;
2135 INCREMENT(lmStat
.submitted
);
2143 static int lbmIOWait(struct lbuf
* bp
, int flag
)
2145 unsigned long flags
;
2148 jfs_info("lbmIOWait1: bp:0x%p flag:0x%x:0x%x", bp
, bp
->l_flag
, flag
);
2150 LCACHE_LOCK(flags
); /* disable+lock */
2152 LCACHE_SLEEP_COND(bp
->l_ioevent
, (bp
->l_flag
& lbmDONE
), flags
);
2154 rc
= (bp
->l_flag
& lbmERROR
) ? -EIO
: 0;
2159 LCACHE_UNLOCK(flags
); /* unlock+enable */
2161 jfs_info("lbmIOWait2: bp:0x%p flag:0x%x:0x%x", bp
, bp
->l_flag
, flag
);
2168 * executed at INTIODONE level
2170 static void lbmIODone(struct bio
*bio
)
2172 struct lbuf
*bp
= bio
->bi_private
;
2173 struct lbuf
*nextbp
, *tail
;
2174 struct jfs_log
*log
;
2175 unsigned long flags
;
2178 * get back jfs buffer bound to the i/o buffer
2180 jfs_info("lbmIODone: bp:0x%p flag:0x%x", bp
, bp
->l_flag
);
2182 LCACHE_LOCK(flags
); /* disable+lock */
2184 bp
->l_flag
|= lbmDONE
;
2186 if (bio
->bi_status
) {
2187 bp
->l_flag
|= lbmERROR
;
2189 jfs_err("lbmIODone: I/O error in JFS log");
2197 if (bp
->l_flag
& lbmREAD
) {
2198 bp
->l_flag
&= ~lbmREAD
;
2200 LCACHE_UNLOCK(flags
); /* unlock+enable */
2202 /* wakeup I/O initiator */
2203 LCACHE_WAKEUP(&bp
->l_ioevent
);
2209 * pageout completion
2211 * the bp at the head of write queue has completed pageout.
2213 * if single-commit/full-page pageout, remove the current buffer
2214 * from head of pageout queue, and redrive pageout with
2215 * the new buffer at head of pageout queue;
2216 * otherwise, the partial-page pageout buffer stays at
2217 * the head of pageout queue to be redriven for pageout
2218 * by lmGroupCommit() until full-page pageout is completed.
2220 bp
->l_flag
&= ~lbmWRITE
;
2221 INCREMENT(lmStat
.pagedone
);
2223 /* update committed lsn */
2225 log
->clsn
= (bp
->l_pn
<< L2LOGPSIZE
) + bp
->l_ceor
;
2227 if (bp
->l_flag
& lbmDIRECT
) {
2228 LCACHE_WAKEUP(&bp
->l_ioevent
);
2229 LCACHE_UNLOCK(flags
);
2235 /* single element queue */
2237 /* remove head buffer of full-page pageout
2238 * from log device write queue
2240 if (bp
->l_flag
& lbmRELEASE
) {
2242 bp
->l_wqnext
= NULL
;
2245 /* multi element queue */
2247 /* remove head buffer of full-page pageout
2248 * from log device write queue
2250 if (bp
->l_flag
& lbmRELEASE
) {
2251 nextbp
= tail
->l_wqnext
= bp
->l_wqnext
;
2252 bp
->l_wqnext
= NULL
;
2255 * redrive pageout of next page at head of write queue:
2256 * redrive next page without any bound tblk
2257 * (i.e., page w/o any COMMIT records), or
2258 * first page of new group commit which has been
2259 * queued after current page (subsequent pageout
2260 * is performed synchronously, except page without
2261 * any COMMITs) by lmGroupCommit() as indicated
2264 if (nextbp
->l_flag
& lbmWRITE
) {
2266 * We can't do the I/O at interrupt time.
2267 * The jfsIO thread can do it
2275 * synchronous pageout:
2277 * buffer has not necessarily been removed from write queue
2278 * (e.g., synchronous write of partial-page with COMMIT):
2279 * leave buffer for i/o initiator to dispose
2281 if (bp
->l_flag
& lbmSYNC
) {
2282 LCACHE_UNLOCK(flags
); /* unlock+enable */
2284 /* wakeup I/O initiator */
2285 LCACHE_WAKEUP(&bp
->l_ioevent
);
2289 * Group Commit pageout:
2291 else if (bp
->l_flag
& lbmGC
) {
2292 LCACHE_UNLOCK(flags
);
2297 * asynchronous pageout:
2299 * buffer must have been removed from write queue:
2300 * insert buffer at head of freelist where it can be recycled
2303 assert(bp
->l_flag
& lbmRELEASE
);
2304 assert(bp
->l_flag
& lbmFREE
);
2307 LCACHE_UNLOCK(flags
); /* unlock+enable */
2311 int jfsIOWait(void *arg
)
2316 spin_lock_irq(&log_redrive_lock
);
2317 while ((bp
= log_redrive_list
)) {
2318 log_redrive_list
= bp
->l_redrive_next
;
2319 bp
->l_redrive_next
= NULL
;
2320 spin_unlock_irq(&log_redrive_lock
);
2322 spin_lock_irq(&log_redrive_lock
);
2325 if (freezing(current
)) {
2326 spin_unlock_irq(&log_redrive_lock
);
2329 set_current_state(TASK_INTERRUPTIBLE
);
2330 spin_unlock_irq(&log_redrive_lock
);
2333 } while (!kthread_should_stop());
2335 jfs_info("jfsIOWait being killed!");
2340 * NAME: lmLogFormat()/jfs_logform()
2342 * FUNCTION: format file system log
2346 * logAddress - start address of log space in FS block
2347 * logSize - length of log space in FS block;
2349 * RETURN: 0 - success
2352 * XXX: We're synchronously writing one page at a time. This needs to
2353 * be improved by writing multiple pages at once.
2355 int lmLogFormat(struct jfs_log
*log
, s64 logAddress
, int logSize
)
2358 struct jfs_sb_info
*sbi
;
2359 struct logsuper
*logsuper
;
2361 int lspn
; /* log sequence page number */
2362 struct lrd
*lrd_ptr
;
2366 jfs_info("lmLogFormat: logAddress:%Ld logSize:%d",
2367 (long long)logAddress
, logSize
);
2369 sbi
= list_entry(log
->sb_list
.next
, struct jfs_sb_info
, log_list
);
2371 /* allocate a log buffer */
2372 bp
= lbmAllocate(log
, 1);
2374 npages
= logSize
>> sbi
->l2nbperpage
;
2379 * page 0 - reserved;
2380 * page 1 - log superblock;
2381 * page 2 - log data page: A SYNC log record is written
2382 * into this page at logform time;
2383 * pages 3-N - log data page: set to empty log data pages;
2386 * init log superblock: log page 1
2388 logsuper
= (struct logsuper
*) bp
->l_ldata
;
2390 logsuper
->magic
= cpu_to_le32(LOGMAGIC
);
2391 logsuper
->version
= cpu_to_le32(LOGVERSION
);
2392 logsuper
->state
= cpu_to_le32(LOGREDONE
);
2393 logsuper
->flag
= cpu_to_le32(sbi
->mntflag
); /* ? */
2394 logsuper
->size
= cpu_to_le32(npages
);
2395 logsuper
->bsize
= cpu_to_le32(sbi
->bsize
);
2396 logsuper
->l2bsize
= cpu_to_le32(sbi
->l2bsize
);
2397 logsuper
->end
= cpu_to_le32(2 * LOGPSIZE
+ LOGPHDRSIZE
+ LOGRDSIZE
);
2399 bp
->l_flag
= lbmWRITE
| lbmSYNC
| lbmDIRECT
;
2400 bp
->l_blkno
= logAddress
+ sbi
->nbperpage
;
2402 if ((rc
= lbmIOWait(bp
, 0)))
2406 * init pages 2 to npages-1 as log data pages:
2408 * log page sequence number (lpsn) initialization:
2411 * +-----+-----+=====+=====+===.....===+=====+
2413 * <--- N page circular file ---->
2415 * the N (= npages-2) data pages of the log is maintained as
2416 * a circular file for the log records;
2417 * lpsn grows by 1 monotonically as each log page is written
2418 * to the circular file of the log;
2419 * and setLogpage() will not reset the page number even if
2420 * the eor is equal to LOGPHDRSIZE. In order for binary search
2421 * still work in find log end process, we have to simulate the
2422 * log wrap situation at the log format time.
2423 * The 1st log page written will have the highest lpsn. Then
2424 * the succeeding log pages will have ascending order of
2425 * the lspn starting from 0, ... (N-2)
2427 lp
= (struct logpage
*) bp
->l_ldata
;
2429 * initialize 1st log page to be written: lpsn = N - 1,
2430 * write a SYNCPT log record is written to this page
2432 lp
->h
.page
= lp
->t
.page
= cpu_to_le32(npages
- 3);
2433 lp
->h
.eor
= lp
->t
.eor
= cpu_to_le16(LOGPHDRSIZE
+ LOGRDSIZE
);
2435 lrd_ptr
= (struct lrd
*) &lp
->data
;
2436 lrd_ptr
->logtid
= 0;
2437 lrd_ptr
->backchain
= 0;
2438 lrd_ptr
->type
= cpu_to_le16(LOG_SYNCPT
);
2439 lrd_ptr
->length
= 0;
2440 lrd_ptr
->log
.syncpt
.sync
= 0;
2442 bp
->l_blkno
+= sbi
->nbperpage
;
2443 bp
->l_flag
= lbmWRITE
| lbmSYNC
| lbmDIRECT
;
2445 if ((rc
= lbmIOWait(bp
, 0)))
2449 * initialize succeeding log pages: lpsn = 0, 1, ..., (N-2)
2451 for (lspn
= 0; lspn
< npages
- 3; lspn
++) {
2452 lp
->h
.page
= lp
->t
.page
= cpu_to_le32(lspn
);
2453 lp
->h
.eor
= lp
->t
.eor
= cpu_to_le16(LOGPHDRSIZE
);
2455 bp
->l_blkno
+= sbi
->nbperpage
;
2456 bp
->l_flag
= lbmWRITE
| lbmSYNC
| lbmDIRECT
;
2458 if ((rc
= lbmIOWait(bp
, 0)))
2467 /* release the buffer */
2473 #ifdef CONFIG_JFS_STATISTICS
2474 int jfs_lmstats_proc_show(struct seq_file
*m
, void *v
)
2477 "JFS Logmgr stats\n"
2478 "================\n"
2480 "writes submitted = %d\n"
2481 "writes completed = %d\n"
2482 "full pages submitted = %d\n"
2483 "partial pages submitted = %d\n",
2488 lmStat
.partial_page
);
2491 #endif /* CONFIG_JFS_STATISTICS */