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/smp_lock.h>
66 #include <linux/completion.h>
67 #include <linux/buffer_head.h> /* for sync_blockdev() */
68 #include <linux/bio.h>
69 #include <linux/suspend.h>
70 #include "jfs_incore.h"
71 #include "jfs_filsys.h"
72 #include "jfs_metapage.h"
73 #include "jfs_txnmgr.h"
74 #include "jfs_debug.h"
78 * lbuf's ready to be redriven. Protected by log_redrive_lock (jfsIO thread)
80 static struct lbuf *log_redrive_list;
81 static DEFINE_SPINLOCK(log_redrive_lock);
82 DECLARE_WAIT_QUEUE_HEAD(jfs_IO_thread_wait);
86 * log read/write serialization (per log)
88 #define LOG_LOCK_INIT(log) init_MUTEX(&(log)->loglock)
89 #define LOG_LOCK(log) down(&((log)->loglock))
90 #define LOG_UNLOCK(log) up(&((log)->loglock))
94 * log group commit serialization (per log)
97 #define LOGGC_LOCK_INIT(log) spin_lock_init(&(log)->gclock)
98 #define LOGGC_LOCK(log) spin_lock_irq(&(log)->gclock)
99 #define LOGGC_UNLOCK(log) spin_unlock_irq(&(log)->gclock)
100 #define LOGGC_WAKEUP(tblk) wake_up_all(&(tblk)->gcwait)
103 * log sync serialization (per log)
105 #define LOGSYNC_DELTA(logsize) min((logsize)/8, 128*LOGPSIZE)
106 #define LOGSYNC_BARRIER(logsize) ((logsize)/4)
108 #define LOGSYNC_DELTA(logsize) min((logsize)/4, 256*LOGPSIZE)
109 #define LOGSYNC_BARRIER(logsize) ((logsize)/2)
114 * log buffer cache synchronization
116 static DEFINE_SPINLOCK(jfsLCacheLock);
118 #define LCACHE_LOCK(flags) spin_lock_irqsave(&jfsLCacheLock, flags)
119 #define LCACHE_UNLOCK(flags) spin_unlock_irqrestore(&jfsLCacheLock, flags)
122 * See __SLEEP_COND in jfs_locks.h
124 #define LCACHE_SLEEP_COND(wq, cond, flags) \
128 __SLEEP_COND(wq, cond, LCACHE_LOCK(flags), LCACHE_UNLOCK(flags)); \
131 #define LCACHE_WAKEUP(event) wake_up(event)
135 * lbuf buffer cache (lCache) control
137 /* log buffer manager pageout control (cumulative, inclusive) */
138 #define lbmREAD 0x0001
139 #define lbmWRITE 0x0002 /* enqueue at tail of write queue;
140 * init pageout if at head of queue;
142 #define lbmRELEASE 0x0004 /* remove from write queue
143 * at completion of pageout;
144 * do not free/recycle it yet:
145 * caller will free it;
147 #define lbmSYNC 0x0008 /* do not return to freelist
148 * when removed from write queue;
150 #define lbmFREE 0x0010 /* return to freelist
151 * at completion of pageout;
152 * the buffer may be recycled;
154 #define lbmDONE 0x0020
155 #define lbmERROR 0x0040
156 #define lbmGC 0x0080 /* lbmIODone to perform post-GC processing
159 #define lbmDIRECT 0x0100
162 * Global list of active external journals
164 static LIST_HEAD(jfs_external_logs);
165 static struct jfs_log *dummy_log = NULL;
166 static DECLARE_MUTEX(jfs_log_sem);
169 * external references
171 extern void txLazyUnlock(struct tblock * tblk);
172 extern int jfs_stop_threads;
173 extern struct completion jfsIOwait;
174 extern int jfs_tlocks_low;
179 static int lmWriteRecord(struct jfs_log * log, struct tblock * tblk,
180 struct lrd * lrd, struct tlock * tlck);
182 static int lmNextPage(struct jfs_log * log);
183 static int lmLogFileSystem(struct jfs_log * log, struct jfs_sb_info *sbi,
186 static int open_inline_log(struct super_block *sb);
187 static int open_dummy_log(struct super_block *sb);
188 static int lbmLogInit(struct jfs_log * log);
189 static void lbmLogShutdown(struct jfs_log * log);
190 static struct lbuf *lbmAllocate(struct jfs_log * log, int);
191 static void lbmFree(struct lbuf * bp);
192 static void lbmfree(struct lbuf * bp);
193 static int lbmRead(struct jfs_log * log, int pn, struct lbuf ** bpp);
194 static void lbmWrite(struct jfs_log * log, struct lbuf * bp, int flag, int cant_block);
195 static void lbmDirectWrite(struct jfs_log * log, struct lbuf * bp, int flag);
196 static int lbmIOWait(struct lbuf * bp, int flag);
197 static bio_end_io_t lbmIODone;
198 static void lbmStartIO(struct lbuf * bp);
199 static void lmGCwrite(struct jfs_log * log, int cant_block);
200 static int lmLogSync(struct jfs_log * log, int nosyncwait);
207 #ifdef CONFIG_JFS_STATISTICS
208 static struct lmStat {
209 uint commit; /* # of commit */
210 uint pagedone; /* # of page written */
211 uint submitted; /* # of pages submitted */
212 uint full_page; /* # of full pages submitted */
213 uint partial_page; /* # of partial pages submitted */
221 * FUNCTION: write a log record;
225 * RETURN: lsn - offset to the next log record to write (end-of-log);
228 * note: todo: log error handler
230 int lmLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
235 struct metapage *mp = NULL;
237 jfs_info("lmLog: log:0x%p tblk:0x%p, lrd:0x%p tlck:0x%p",
238 log, tblk, lrd, tlck);
242 /* log by (out-of-transaction) JFS ? */
246 /* log from page ? */
248 tlck->type & tlckBTROOT || (mp = tlck->mp) == NULL)
252 * initialize/update page/transaction recovery lsn
259 * initialize page lsn if first log write of the page
266 /* insert page at tail of logsynclist */
267 list_add_tail(&mp->synclist, &log->synclist);
271 * initialize/update lsn of tblock of the page
273 * transaction inherits oldest lsn of pages associated
274 * with allocation/deallocation of resources (their
275 * log records are used to reconstruct allocation map
276 * at recovery time: inode for inode allocation map,
277 * B+-tree index of extent descriptors for block
279 * allocation map pages inherit transaction lsn at
280 * commit time to allow forwarding log syncpt past log
281 * records associated with allocation/deallocation of
282 * resources only after persistent map of these map pages
283 * have been updated and propagated to home.
286 * initialize transaction lsn:
288 if (tblk->lsn == 0) {
289 /* inherit lsn of its first page logged */
293 /* insert tblock after the page on logsynclist */
294 list_add(&tblk->synclist, &mp->synclist);
297 * update transaction lsn:
300 /* inherit oldest/smallest lsn of page */
301 logdiff(diffp, mp->lsn, log);
302 logdiff(difft, tblk->lsn, log);
304 /* update tblock lsn with page lsn */
307 /* move tblock after page on logsynclist */
308 list_move(&tblk->synclist, &mp->synclist);
315 * write the log record
318 lsn = lmWriteRecord(log, tblk, lrd, tlck);
321 * forward log syncpt if log reached next syncpt trigger
323 logdiff(diffp, lsn, log);
324 if (diffp >= log->nextsync)
325 lsn = lmLogSync(log, 0);
327 /* update end-of-log lsn */
332 /* return end-of-log address */
338 * NAME: lmWriteRecord()
340 * FUNCTION: move the log record to current log page
342 * PARAMETER: cd - commit descriptor
344 * RETURN: end-of-log address
346 * serialization: LOG_LOCK() held on entry/exit
349 lmWriteRecord(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
352 int lsn = 0; /* end-of-log address */
353 struct lbuf *bp; /* dst log page buffer */
354 struct logpage *lp; /* dst log page */
355 caddr_t dst; /* destination address in log page */
356 int dstoffset; /* end-of-log offset in log page */
357 int freespace; /* free space in log page */
358 caddr_t p; /* src meta-data page */
361 int nbytes; /* number of bytes to move */
364 struct linelock *linelock;
371 /* retrieve destination log page to write */
372 bp = (struct lbuf *) log->bp;
373 lp = (struct logpage *) bp->l_ldata;
374 dstoffset = log->eor;
376 /* any log data to write ? */
381 * move log record data
383 /* retrieve source meta-data page to log */
384 if (tlck->flag & tlckPAGELOCK) {
385 p = (caddr_t) (tlck->mp->data);
386 linelock = (struct linelock *) & tlck->lock;
388 /* retrieve source in-memory inode to log */
389 else if (tlck->flag & tlckINODELOCK) {
390 if (tlck->type & tlckDTREE)
391 p = (caddr_t) &JFS_IP(tlck->ip)->i_dtroot;
393 p = (caddr_t) &JFS_IP(tlck->ip)->i_xtroot;
394 linelock = (struct linelock *) & tlck->lock;
397 else if (tlck->flag & tlckINLINELOCK) {
399 inlinelock = (struct inlinelock *) & tlck;
400 p = (caddr_t) & inlinelock->pxd;
401 linelock = (struct linelock *) & tlck;
403 #endif /* _JFS_WIP */
405 jfs_err("lmWriteRecord: UFO tlck:0x%p", tlck);
406 return 0; /* Probably should trap */
408 l2linesize = linelock->l2linesize;
411 ASSERT(linelock->index <= linelock->maxcnt);
414 for (i = 0; i < linelock->index; i++, lv++) {
419 if (dstoffset >= LOGPSIZE - LOGPTLRSIZE) {
420 /* page become full: move on to next page */
424 lp = (struct logpage *) bp->l_ldata;
425 dstoffset = LOGPHDRSIZE;
429 * move log vector data
431 src = (u8 *) p + (lv->offset << l2linesize);
432 srclen = lv->length << l2linesize;
435 freespace = (LOGPSIZE - LOGPTLRSIZE) - dstoffset;
436 nbytes = min(freespace, srclen);
437 dst = (caddr_t) lp + dstoffset;
438 memcpy(dst, src, nbytes);
441 /* is page not full ? */
442 if (dstoffset < LOGPSIZE - LOGPTLRSIZE)
445 /* page become full: move on to next page */
448 bp = (struct lbuf *) log->bp;
449 lp = (struct logpage *) bp->l_ldata;
450 dstoffset = LOGPHDRSIZE;
457 * move log vector descriptor
460 lvd = (struct lvd *) ((caddr_t) lp + dstoffset);
461 lvd->offset = cpu_to_le16(lv->offset);
462 lvd->length = cpu_to_le16(lv->length);
464 jfs_info("lmWriteRecord: lv offset:%d length:%d",
465 lv->offset, lv->length);
468 if ((i = linelock->next)) {
469 linelock = (struct linelock *) lid_to_tlock(i);
474 * move log record descriptor
477 lrd->length = cpu_to_le16(len);
483 freespace = (LOGPSIZE - LOGPTLRSIZE) - dstoffset;
484 nbytes = min(freespace, srclen);
485 dst = (caddr_t) lp + dstoffset;
486 memcpy(dst, src, nbytes);
491 /* are there more to move than freespace of page ? */
496 * end of log record descriptor
499 /* update last log record eor */
500 log->eor = dstoffset;
501 bp->l_eor = dstoffset;
502 lsn = (log->page << L2LOGPSIZE) + dstoffset;
504 if (lrd->type & cpu_to_le16(LOG_COMMIT)) {
506 jfs_info("wr: tclsn:0x%x, beor:0x%x", tblk->clsn,
509 INCREMENT(lmStat.commit); /* # of commit */
512 * enqueue tblock for group commit:
514 * enqueue tblock of non-trivial/synchronous COMMIT
515 * at tail of group commit queue
516 * (trivial/asynchronous COMMITs are ignored by
521 /* init tblock gc state */
522 tblk->flag = tblkGC_QUEUE;
524 tblk->pn = log->page;
525 tblk->eor = log->eor;
527 /* enqueue transaction to commit queue */
528 list_add_tail(&tblk->cqueue, &log->cqueue);
533 jfs_info("lmWriteRecord: lrd:0x%04x bp:0x%p pn:%d eor:0x%x",
534 le16_to_cpu(lrd->type), log->bp, log->page, dstoffset);
536 /* page not full ? */
537 if (dstoffset < LOGPSIZE - LOGPTLRSIZE)
541 /* page become full: move on to next page */
544 bp = (struct lbuf *) log->bp;
545 lp = (struct logpage *) bp->l_ldata;
546 dstoffset = LOGPHDRSIZE;
557 * FUNCTION: write current page and allocate next page.
563 * serialization: LOG_LOCK() held on entry/exit
565 static int lmNextPage(struct jfs_log * log)
568 int lspn; /* log sequence page number */
569 int pn; /* current page number */
574 /* get current log page number and log sequence page number */
577 lp = (struct logpage *) bp->l_ldata;
578 lspn = le32_to_cpu(lp->h.page);
583 * write or queue the full page at the tail of write queue
585 /* get the tail tblk on commit queue */
586 if (list_empty(&log->cqueue))
589 tblk = list_entry(log->cqueue.prev, struct tblock, cqueue);
591 /* every tblk who has COMMIT record on the current page,
592 * and has not been committed, must be on commit queue
593 * since tblk is queued at commit queueu at the time
594 * of writing its COMMIT record on the page before
595 * page becomes full (even though the tblk thread
596 * who wrote COMMIT record may have been suspended
600 /* is page bound with outstanding tail tblk ? */
601 if (tblk && tblk->pn == pn) {
602 /* mark tblk for end-of-page */
603 tblk->flag |= tblkGC_EOP;
605 if (log->cflag & logGC_PAGEOUT) {
606 /* if page is not already on write queue,
607 * just enqueue (no lbmWRITE to prevent redrive)
608 * buffer to wqueue to ensure correct serial order
609 * of the pages since log pages will be added
612 if (bp->l_wqnext == NULL)
613 lbmWrite(log, bp, 0, 0);
616 * No current GC leader, initiate group commit
618 log->cflag |= logGC_PAGEOUT;
622 /* page is not bound with outstanding tblk:
623 * init write or mark it to be redriven (lbmWRITE)
626 /* finalize the page */
627 bp->l_ceor = bp->l_eor;
628 lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_ceor);
629 lbmWrite(log, bp, lbmWRITE | lbmRELEASE | lbmFREE, 0);
634 * allocate/initialize next page
636 /* if log wraps, the first data page of log is 2
637 * (0 never used, 1 is superblock).
639 log->page = (pn == log->size - 1) ? 2 : pn + 1;
640 log->eor = LOGPHDRSIZE; /* ? valid page empty/full at logRedo() */
642 /* allocate/initialize next log page buffer */
643 nextbp = lbmAllocate(log, log->page);
644 nextbp->l_eor = log->eor;
647 /* initialize next log page */
648 lp = (struct logpage *) nextbp->l_ldata;
649 lp->h.page = lp->t.page = cpu_to_le32(lspn + 1);
650 lp->h.eor = lp->t.eor = cpu_to_le16(LOGPHDRSIZE);
657 * NAME: lmGroupCommit()
659 * FUNCTION: group commit
660 * initiate pageout of the pages with COMMIT in the order of
661 * page number - redrive pageout of the page at the head of
662 * pageout queue until full page has been written.
667 * LOGGC_LOCK serializes log group commit queue, and
668 * transaction blocks on the commit queue.
669 * N.B. LOG_LOCK is NOT held during lmGroupCommit().
671 int lmGroupCommit(struct jfs_log * log, struct tblock * tblk)
677 /* group committed already ? */
678 if (tblk->flag & tblkGC_COMMITTED) {
679 if (tblk->flag & tblkGC_ERROR)
685 jfs_info("lmGroup Commit: tblk = 0x%p, gcrtc = %d", tblk, log->gcrtc);
687 if (tblk->xflag & COMMIT_LAZY)
688 tblk->flag |= tblkGC_LAZY;
690 if ((!(log->cflag & logGC_PAGEOUT)) && (!list_empty(&log->cqueue)) &&
691 (!(tblk->xflag & COMMIT_LAZY) || test_bit(log_FLUSH, &log->flag)
692 || jfs_tlocks_low)) {
694 * No pageout in progress
696 * start group commit as its group leader.
698 log->cflag |= logGC_PAGEOUT;
703 if (tblk->xflag & COMMIT_LAZY) {
705 * Lazy transactions can leave now
711 /* lmGCwrite gives up LOGGC_LOCK, check again */
713 if (tblk->flag & tblkGC_COMMITTED) {
714 if (tblk->flag & tblkGC_ERROR)
721 /* upcount transaction waiting for completion
724 tblk->flag |= tblkGC_READY;
726 __SLEEP_COND(tblk->gcwait, (tblk->flag & tblkGC_COMMITTED),
727 LOGGC_LOCK(log), LOGGC_UNLOCK(log));
729 /* removed from commit queue */
730 if (tblk->flag & tblkGC_ERROR)
740 * FUNCTION: group commit write
741 * initiate write of log page, building a group of all transactions
742 * with commit records on that page.
747 * LOGGC_LOCK must be held by caller.
748 * N.B. LOG_LOCK is NOT held during lmGroupCommit().
750 static void lmGCwrite(struct jfs_log * log, int cant_write)
754 int gcpn; /* group commit page number */
756 struct tblock *xtblk = NULL;
759 * build the commit group of a log page
761 * scan commit queue and make a commit group of all
762 * transactions with COMMIT records on the same log page.
764 /* get the head tblk on the commit queue */
765 gcpn = list_entry(log->cqueue.next, struct tblock, cqueue)->pn;
767 list_for_each_entry(tblk, &log->cqueue, cqueue) {
768 if (tblk->pn != gcpn)
773 /* state transition: (QUEUE, READY) -> COMMIT */
774 tblk->flag |= tblkGC_COMMIT;
776 tblk = xtblk; /* last tblk of the page */
779 * pageout to commit transactions on the log page.
781 bp = (struct lbuf *) tblk->bp;
782 lp = (struct logpage *) bp->l_ldata;
783 /* is page already full ? */
784 if (tblk->flag & tblkGC_EOP) {
785 /* mark page to free at end of group commit of the page */
786 tblk->flag &= ~tblkGC_EOP;
787 tblk->flag |= tblkGC_FREE;
788 bp->l_ceor = bp->l_eor;
789 lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_ceor);
790 lbmWrite(log, bp, lbmWRITE | lbmRELEASE | lbmGC,
792 INCREMENT(lmStat.full_page);
794 /* page is not yet full */
796 bp->l_ceor = tblk->eor; /* ? bp->l_ceor = bp->l_eor; */
797 lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_ceor);
798 lbmWrite(log, bp, lbmWRITE | lbmGC, cant_write);
799 INCREMENT(lmStat.partial_page);
806 * FUNCTION: group commit post-processing
807 * Processes transactions after their commit records have been written
808 * to disk, redriving log I/O if necessary.
813 * This routine is called a interrupt time by lbmIODone
815 static void lmPostGC(struct lbuf * bp)
818 struct jfs_log *log = bp->l_log;
820 struct tblock *tblk, *temp;
823 spin_lock_irqsave(&log->gclock, flags);
825 * current pageout of group commit completed.
827 * remove/wakeup transactions from commit queue who were
828 * group committed with the current log page
830 list_for_each_entry_safe(tblk, temp, &log->cqueue, cqueue) {
831 if (!(tblk->flag & tblkGC_COMMIT))
833 /* if transaction was marked GC_COMMIT then
834 * it has been shipped in the current pageout
835 * and made it to disk - it is committed.
838 if (bp->l_flag & lbmERROR)
839 tblk->flag |= tblkGC_ERROR;
841 /* remove it from the commit queue */
842 list_del(&tblk->cqueue);
843 tblk->flag &= ~tblkGC_QUEUE;
845 if (tblk == log->flush_tblk) {
846 /* we can stop flushing the log now */
847 clear_bit(log_FLUSH, &log->flag);
848 log->flush_tblk = NULL;
851 jfs_info("lmPostGC: tblk = 0x%p, flag = 0x%x", tblk,
854 if (!(tblk->xflag & COMMIT_FORCE))
856 * Hand tblk over to lazy commit thread
860 /* state transition: COMMIT -> COMMITTED */
861 tblk->flag |= tblkGC_COMMITTED;
863 if (tblk->flag & tblkGC_READY)
869 /* was page full before pageout ?
870 * (and this is the last tblk bound with the page)
872 if (tblk->flag & tblkGC_FREE)
874 /* did page become full after pageout ?
875 * (and this is the last tblk bound with the page)
877 else if (tblk->flag & tblkGC_EOP) {
878 /* finalize the page */
879 lp = (struct logpage *) bp->l_ldata;
880 bp->l_ceor = bp->l_eor;
881 lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_eor);
882 jfs_info("lmPostGC: calling lbmWrite");
883 lbmWrite(log, bp, lbmWRITE | lbmRELEASE | lbmFREE,
889 /* are there any transactions who have entered lnGroupCommit()
890 * (whose COMMITs are after that of the last log page written.
891 * They are waiting for new group commit (above at (SLEEP 1))
892 * or lazy transactions are on a full (queued) log page,
893 * select the latest ready transaction as new group leader and
894 * wake her up to lead her group.
896 if ((!list_empty(&log->cqueue)) &&
897 ((log->gcrtc > 0) || (tblk->bp->l_wqnext != NULL) ||
898 test_bit(log_FLUSH, &log->flag) || jfs_tlocks_low))
900 * Call lmGCwrite with new group leader
904 /* no transaction are ready yet (transactions are only just
905 * queued (GC_QUEUE) and not entered for group commit yet).
906 * the first transaction entering group commit
907 * will elect herself as new group leader.
910 log->cflag &= ~logGC_PAGEOUT;
913 spin_unlock_irqrestore(&log->gclock, flags);
920 * FUNCTION: write log SYNCPT record for specified log
921 * if new sync address is available
922 * (normally the case if sync() is executed by back-ground
924 * if not, explicitly run jfs_blogsync() to initiate
925 * getting of new sync address.
926 * calculate new value of i_nextsync which determines when
927 * this code is called again.
929 * this is called only from lmLog().
931 * PARAMETER: ip - pointer to logs inode.
935 * serialization: LOG_LOCK() held on entry/exit
937 static int lmLogSync(struct jfs_log * log, int nosyncwait)
940 int written; /* written since last syncpt */
941 int free; /* free space left available */
942 int delta; /* additional delta to write normally */
943 int more; /* additional write granted */
946 struct logsyncblk *lp;
951 /* if last sync is same as last syncpt,
952 * invoke sync point forward processing to update sync.
955 if (log->sync == log->syncpt) {
957 /* ToDo: push dirty metapages out to disk */
960 if (list_empty(&log->synclist))
961 log->sync = log->lsn;
963 lp = list_entry(log->synclist.next,
964 struct logsyncblk, synclist);
971 /* if sync is different from last syncpt,
972 * write a SYNCPT record with syncpt = sync.
973 * reset syncpt = sync
975 if (log->sync != log->syncpt) {
976 struct jfs_sb_info *sbi;
979 * We need to make sure all of the "written" metapages
980 * actually make it to disk
982 list_for_each_entry(sbi, &log->sb_list, log_list) {
983 if (sbi->flag & JFS_NOINTEGRITY)
985 filemap_fdatawrite(sbi->ipbmap->i_mapping);
986 filemap_fdatawrite(sbi->ipimap->i_mapping);
987 filemap_fdatawrite(sbi->sb->s_bdev->bd_inode->i_mapping);
989 list_for_each_entry(sbi, &log->sb_list, log_list) {
990 if (sbi->flag & JFS_NOINTEGRITY)
992 filemap_fdatawait(sbi->ipbmap->i_mapping);
993 filemap_fdatawait(sbi->ipimap->i_mapping);
994 filemap_fdatawait(sbi->sb->s_bdev->bd_inode->i_mapping);
999 lrd.type = cpu_to_le16(LOG_SYNCPT);
1001 lrd.log.syncpt.sync = cpu_to_le32(log->sync);
1002 lsn = lmWriteRecord(log, NULL, &lrd, NULL);
1004 log->syncpt = log->sync;
1009 * setup next syncpt trigger (SWAG)
1011 logsize = log->logsize;
1013 logdiff(written, lsn, log);
1014 free = logsize - written;
1015 delta = LOGSYNC_DELTA(logsize);
1016 more = min(free / 2, delta);
1017 if (more < 2 * LOGPSIZE) {
1018 jfs_warn("\n ... Log Wrap ... Log Wrap ... Log Wrap ...\n");
1022 * option 1 - panic ? No.!
1023 * option 2 - shutdown file systems
1024 * associated with log ?
1025 * option 3 - extend log ?
1028 * option 4 - second chance
1030 * mark log wrapped, and continue.
1031 * when all active transactions are completed,
1032 * mark log vaild for recovery.
1033 * if crashed during invalid state, log state
1034 * implies invald log, forcing fsck().
1036 /* mark log state log wrap in log superblock */
1037 /* log->state = LOGWRAP; */
1039 /* reset sync point computation */
1040 log->syncpt = log->sync = lsn;
1041 log->nextsync = delta;
1043 /* next syncpt trigger = written + more */
1044 log->nextsync = written + more;
1046 /* return if lmLogSync() from outside of transaction, e.g., sync() */
1050 /* if number of bytes written from last sync point is more
1051 * than 1/4 of the log size, stop new transactions from
1052 * starting until all current transactions are completed
1053 * by setting syncbarrier flag.
1055 if (written > LOGSYNC_BARRIER(logsize) && logsize > 32 * LOGPSIZE) {
1056 set_bit(log_SYNCBARRIER, &log->flag);
1057 jfs_info("log barrier on: lsn=0x%x syncpt=0x%x", lsn,
1060 * We may have to initiate group commit
1062 jfs_flush_journal(log, 0);
1072 * FUNCTION: open the log on first open;
1073 * insert filesystem in the active list of the log.
1075 * PARAMETER: ipmnt - file system mount inode
1076 * iplog - log inode (out)
1082 int lmLogOpen(struct super_block *sb)
1085 struct block_device *bdev;
1086 struct jfs_log *log;
1087 struct jfs_sb_info *sbi = JFS_SBI(sb);
1089 if (sbi->flag & JFS_NOINTEGRITY)
1090 return open_dummy_log(sb);
1092 if (sbi->mntflag & JFS_INLINELOG)
1093 return open_inline_log(sb);
1096 list_for_each_entry(log, &jfs_external_logs, journal_list) {
1097 if (log->bdev->bd_dev == sbi->logdev) {
1098 if (memcmp(log->uuid, sbi->loguuid,
1099 sizeof(log->uuid))) {
1100 jfs_warn("wrong uuid on JFS journal\n");
1105 * add file system to log active file system list
1107 if ((rc = lmLogFileSystem(log, sbi, 1))) {
1115 if (!(log = kmalloc(sizeof(struct jfs_log), GFP_KERNEL))) {
1119 memset(log, 0, sizeof(struct jfs_log));
1120 INIT_LIST_HEAD(&log->sb_list);
1123 * external log as separate logical volume
1125 * file systems to log may have n-to-1 relationship;
1128 bdev = open_by_devnum(sbi->logdev, FMODE_READ|FMODE_WRITE);
1130 rc = -PTR_ERR(bdev);
1134 if ((rc = bd_claim(bdev, log))) {
1139 memcpy(log->uuid, sbi->loguuid, sizeof(log->uuid));
1144 if ((rc = lmLogInit(log)))
1147 list_add(&log->journal_list, &jfs_external_logs);
1150 * add file system to log active file system list
1152 if ((rc = lmLogFileSystem(log, sbi, 1)))
1157 list_add(&sbi->log_list, &log->sb_list);
1167 shutdown: /* unwind lbmLogInit() */
1168 list_del(&log->journal_list);
1169 lbmLogShutdown(log);
1174 close: /* close external log device */
1177 free: /* free log descriptor */
1181 jfs_warn("lmLogOpen: exit(%d)", rc);
1185 static int open_inline_log(struct super_block *sb)
1187 struct jfs_log *log;
1190 if (!(log = kmalloc(sizeof(struct jfs_log), GFP_KERNEL)))
1192 memset(log, 0, sizeof(struct jfs_log));
1193 INIT_LIST_HEAD(&log->sb_list);
1195 set_bit(log_INLINELOG, &log->flag);
1196 log->bdev = sb->s_bdev;
1197 log->base = addressPXD(&JFS_SBI(sb)->logpxd);
1198 log->size = lengthPXD(&JFS_SBI(sb)->logpxd) >>
1199 (L2LOGPSIZE - sb->s_blocksize_bits);
1200 log->l2bsize = sb->s_blocksize_bits;
1201 ASSERT(L2LOGPSIZE >= sb->s_blocksize_bits);
1206 if ((rc = lmLogInit(log))) {
1208 jfs_warn("lmLogOpen: exit(%d)", rc);
1212 list_add(&JFS_SBI(sb)->log_list, &log->sb_list);
1213 JFS_SBI(sb)->log = log;
1218 static int open_dummy_log(struct super_block *sb)
1224 dummy_log = kmalloc(sizeof(struct jfs_log), GFP_KERNEL);
1229 memset(dummy_log, 0, sizeof(struct jfs_log));
1230 INIT_LIST_HEAD(&dummy_log->sb_list);
1231 dummy_log->no_integrity = 1;
1232 /* Make up some stuff */
1233 dummy_log->base = 0;
1234 dummy_log->size = 1024;
1235 rc = lmLogInit(dummy_log);
1244 LOG_LOCK(dummy_log);
1245 list_add(&JFS_SBI(sb)->log_list, &dummy_log->sb_list);
1246 JFS_SBI(sb)->log = dummy_log;
1247 LOG_UNLOCK(dummy_log);
1256 * FUNCTION: log initialization at first log open.
1258 * logredo() (or logformat()) should have been run previously.
1259 * initialize the log from log superblock.
1260 * set the log state in the superblock to LOGMOUNT and
1261 * write SYNCPT log record.
1263 * PARAMETER: log - log structure
1266 * -EINVAL - bad log magic number or superblock dirty
1267 * error returned from logwait()
1269 * serialization: single first open thread
1271 int lmLogInit(struct jfs_log * log)
1275 struct logsuper *logsuper;
1276 struct lbuf *bpsuper;
1281 jfs_info("lmLogInit: log:0x%p", log);
1283 /* initialize the group commit serialization lock */
1284 LOGGC_LOCK_INIT(log);
1286 /* allocate/initialize the log write serialization lock */
1289 LOGSYNC_LOCK_INIT(log);
1291 INIT_LIST_HEAD(&log->synclist);
1293 init_waitqueue_head(&log->syncwait);
1295 INIT_LIST_HEAD(&log->cqueue);
1296 log->flush_tblk = NULL;
1301 * initialize log i/o
1303 if ((rc = lbmLogInit(log)))
1306 if (!test_bit(log_INLINELOG, &log->flag))
1307 log->l2bsize = L2LOGPSIZE;
1309 /* check for disabled journaling to disk */
1310 if (log->no_integrity) {
1312 * Journal pages will still be filled. When the time comes
1313 * to actually do the I/O, the write is not done, and the
1314 * endio routine is called directly.
1316 bp = lbmAllocate(log , 0);
1318 bp->l_pn = bp->l_eor = 0;
1321 * validate log superblock
1323 if ((rc = lbmRead(log, 1, &bpsuper)))
1326 logsuper = (struct logsuper *) bpsuper->l_ldata;
1328 if (logsuper->magic != cpu_to_le32(LOGMAGIC)) {
1329 jfs_warn("*** Log Format Error ! ***");
1334 /* logredo() should have been run successfully. */
1335 if (logsuper->state != cpu_to_le32(LOGREDONE)) {
1336 jfs_warn("*** Log Is Dirty ! ***");
1341 /* initialize log from log superblock */
1342 if (test_bit(log_INLINELOG,&log->flag)) {
1343 if (log->size != le32_to_cpu(logsuper->size)) {
1347 jfs_info("lmLogInit: inline log:0x%p base:0x%Lx "
1349 (unsigned long long) log->base, log->size);
1351 if (memcmp(logsuper->uuid, log->uuid, 16)) {
1352 jfs_warn("wrong uuid on JFS log device");
1355 log->size = le32_to_cpu(logsuper->size);
1356 log->l2bsize = le32_to_cpu(logsuper->l2bsize);
1357 jfs_info("lmLogInit: external log:0x%p base:0x%Lx "
1359 (unsigned long long) log->base, log->size);
1362 log->page = le32_to_cpu(logsuper->end) / LOGPSIZE;
1363 log->eor = le32_to_cpu(logsuper->end) - (LOGPSIZE * log->page);
1366 * initialize for log append write mode
1368 /* establish current/end-of-log page/buffer */
1369 if ((rc = lbmRead(log, log->page, &bp)))
1372 lp = (struct logpage *) bp->l_ldata;
1374 jfs_info("lmLogInit: lsn:0x%x page:%d eor:%d:%d",
1375 le32_to_cpu(logsuper->end), log->page, log->eor,
1376 le16_to_cpu(lp->h.eor));
1379 bp->l_pn = log->page;
1380 bp->l_eor = log->eor;
1382 /* if current page is full, move on to next page */
1383 if (log->eor >= LOGPSIZE - LOGPTLRSIZE)
1387 * initialize log syncpoint
1390 * write the first SYNCPT record with syncpoint = 0
1391 * (i.e., log redo up to HERE !);
1392 * remove current page from lbm write queue at end of pageout
1393 * (to write log superblock update), but do not release to
1398 lrd.type = cpu_to_le16(LOG_SYNCPT);
1400 lrd.log.syncpt.sync = 0;
1401 lsn = lmWriteRecord(log, NULL, &lrd, NULL);
1403 bp->l_ceor = bp->l_eor;
1404 lp = (struct logpage *) bp->l_ldata;
1405 lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_eor);
1406 lbmWrite(log, bp, lbmWRITE | lbmSYNC, 0);
1407 if ((rc = lbmIOWait(bp, 0)))
1411 * update/write superblock
1413 logsuper->state = cpu_to_le32(LOGMOUNT);
1414 log->serial = le32_to_cpu(logsuper->serial) + 1;
1415 logsuper->serial = cpu_to_le32(log->serial);
1416 lbmDirectWrite(log, bpsuper, lbmWRITE | lbmRELEASE | lbmSYNC);
1417 if ((rc = lbmIOWait(bpsuper, lbmFREE)))
1421 /* initialize logsync parameters */
1422 log->logsize = (log->size - 2) << L2LOGPSIZE;
1425 log->sync = log->syncpt;
1426 log->nextsync = LOGSYNC_DELTA(log->logsize);
1428 jfs_info("lmLogInit: lsn:0x%x syncpt:0x%x sync:0x%x",
1429 log->lsn, log->syncpt, log->sync);
1432 * initialize for lazy/group commit
1441 errout30: /* release log page */
1443 bp->l_wqnext = NULL;
1446 errout20: /* release log superblock */
1449 errout10: /* unwind lbmLogInit() */
1450 lbmLogShutdown(log);
1452 jfs_warn("lmLogInit: exit(%d)", rc);
1458 * NAME: lmLogClose()
1460 * FUNCTION: remove file system <ipmnt> from active list of log <iplog>
1461 * and close it on last close.
1463 * PARAMETER: sb - superblock
1465 * RETURN: errors from subroutines
1469 int lmLogClose(struct super_block *sb)
1471 struct jfs_sb_info *sbi = JFS_SBI(sb);
1472 struct jfs_log *log = sbi->log;
1473 struct block_device *bdev;
1476 jfs_info("lmLogClose: log:0x%p", log);
1480 list_del(&sbi->log_list);
1485 * We need to make sure all of the "written" metapages
1486 * actually make it to disk
1488 sync_blockdev(sb->s_bdev);
1490 if (test_bit(log_INLINELOG, &log->flag)) {
1492 * in-line log in host file system
1494 rc = lmLogShutdown(log);
1499 if (!log->no_integrity)
1500 lmLogFileSystem(log, sbi, 0);
1502 if (!list_empty(&log->sb_list))
1506 * TODO: ensure that the dummy_log is in a state to allow
1507 * lbmLogShutdown to deallocate all the buffers and call
1508 * kfree against dummy_log. For now, leave dummy_log & its
1509 * buffers in memory, and resuse if another no-integrity mount
1512 if (log->no_integrity)
1516 * external log as separate logical volume
1518 list_del(&log->journal_list);
1520 rc = lmLogShutdown(log);
1529 jfs_info("lmLogClose: exit(%d)", rc);
1535 * NAME: jfs_flush_journal()
1537 * FUNCTION: initiate write of any outstanding transactions to the journal
1538 * and optionally wait until they are all written to disk
1540 * wait == 0 flush until latest txn is committed, don't wait
1541 * wait == 1 flush until latest txn is committed, wait
1542 * wait > 1 flush until all txn's are complete, wait
1544 void jfs_flush_journal(struct jfs_log *log, int wait)
1547 struct tblock *target = NULL;
1549 /* jfs_write_inode may call us during read-only mount */
1553 jfs_info("jfs_flush_journal: log:0x%p wait=%d", log, wait);
1557 if (!list_empty(&log->cqueue)) {
1559 * This ensures that we will keep writing to the journal as long
1560 * as there are unwritten commit records
1562 target = list_entry(log->cqueue.prev, struct tblock, cqueue);
1564 if (test_bit(log_FLUSH, &log->flag)) {
1566 * We're already flushing.
1567 * if flush_tblk is NULL, we are flushing everything,
1568 * so leave it that way. Otherwise, update it to the
1569 * latest transaction
1571 if (log->flush_tblk)
1572 log->flush_tblk = target;
1574 /* Only flush until latest transaction is committed */
1575 log->flush_tblk = target;
1576 set_bit(log_FLUSH, &log->flag);
1579 * Initiate I/O on outstanding transactions
1581 if (!(log->cflag & logGC_PAGEOUT)) {
1582 log->cflag |= logGC_PAGEOUT;
1587 if ((wait > 1) || test_bit(log_SYNCBARRIER, &log->flag)) {
1588 /* Flush until all activity complete */
1589 set_bit(log_FLUSH, &log->flag);
1590 log->flush_tblk = NULL;
1593 if (wait && target && !(target->flag & tblkGC_COMMITTED)) {
1594 DECLARE_WAITQUEUE(__wait, current);
1596 add_wait_queue(&target->gcwait, &__wait);
1597 set_current_state(TASK_UNINTERRUPTIBLE);
1600 current->state = TASK_RUNNING;
1602 remove_wait_queue(&target->gcwait, &__wait);
1610 * If there was recent activity, we may need to wait
1611 * for the lazycommit thread to catch up
1613 if ((!list_empty(&log->cqueue)) || !list_empty(&log->synclist)) {
1614 for (i = 0; i < 800; i++) { /* Too much? */
1615 current->state = TASK_INTERRUPTIBLE;
1616 schedule_timeout(HZ / 4);
1617 if (list_empty(&log->cqueue) &&
1618 list_empty(&log->synclist))
1622 assert(list_empty(&log->cqueue));
1623 assert(list_empty(&log->synclist));
1624 clear_bit(log_FLUSH, &log->flag);
1628 * NAME: lmLogShutdown()
1630 * FUNCTION: log shutdown at last LogClose().
1632 * write log syncpt record.
1633 * update super block to set redone flag to 0.
1635 * PARAMETER: log - log inode
1637 * RETURN: 0 - success
1639 * serialization: single last close thread
1641 int lmLogShutdown(struct jfs_log * log)
1646 struct logsuper *logsuper;
1647 struct lbuf *bpsuper;
1651 jfs_info("lmLogShutdown: log:0x%p", log);
1653 jfs_flush_journal(log, 2);
1656 * write the last SYNCPT record with syncpoint = 0
1657 * (i.e., log redo up to HERE !)
1661 lrd.type = cpu_to_le16(LOG_SYNCPT);
1663 lrd.log.syncpt.sync = 0;
1665 lsn = lmWriteRecord(log, NULL, &lrd, NULL);
1667 lp = (struct logpage *) bp->l_ldata;
1668 lp->h.eor = lp->t.eor = cpu_to_le16(bp->l_eor);
1669 lbmWrite(log, log->bp, lbmWRITE | lbmRELEASE | lbmSYNC, 0);
1670 lbmIOWait(log->bp, lbmFREE);
1673 * synchronous update log superblock
1674 * mark log state as shutdown cleanly
1675 * (i.e., Log does not need to be replayed).
1677 if ((rc = lbmRead(log, 1, &bpsuper)))
1680 logsuper = (struct logsuper *) bpsuper->l_ldata;
1681 logsuper->state = cpu_to_le32(LOGREDONE);
1682 logsuper->end = cpu_to_le32(lsn);
1683 lbmDirectWrite(log, bpsuper, lbmWRITE | lbmRELEASE | lbmSYNC);
1684 rc = lbmIOWait(bpsuper, lbmFREE);
1686 jfs_info("lmLogShutdown: lsn:0x%x page:%d eor:%d",
1687 lsn, log->page, log->eor);
1691 * shutdown per log i/o
1693 lbmLogShutdown(log);
1696 jfs_warn("lmLogShutdown: exit(%d)", rc);
1703 * NAME: lmLogFileSystem()
1705 * FUNCTION: insert (<activate> = true)/remove (<activate> = false)
1706 * file system into/from log active file system list.
1708 * PARAMETE: log - pointer to logs inode.
1709 * fsdev - kdev_t of filesystem.
1710 * serial - pointer to returned log serial number
1711 * activate - insert/remove device from active list.
1713 * RETURN: 0 - success
1714 * errors returned by vms_iowait().
1716 static int lmLogFileSystem(struct jfs_log * log, struct jfs_sb_info *sbi,
1721 struct logsuper *logsuper;
1722 struct lbuf *bpsuper;
1723 char *uuid = sbi->uuid;
1726 * insert/remove file system device to log active file system list.
1728 if ((rc = lbmRead(log, 1, &bpsuper)))
1731 logsuper = (struct logsuper *) bpsuper->l_ldata;
1733 for (i = 0; i < MAX_ACTIVE; i++)
1734 if (!memcmp(logsuper->active[i].uuid, NULL_UUID, 16)) {
1735 memcpy(logsuper->active[i].uuid, uuid, 16);
1739 if (i == MAX_ACTIVE) {
1740 jfs_warn("Too many file systems sharing journal!");
1742 return -EMFILE; /* Is there a better rc? */
1745 for (i = 0; i < MAX_ACTIVE; i++)
1746 if (!memcmp(logsuper->active[i].uuid, uuid, 16)) {
1747 memcpy(logsuper->active[i].uuid, NULL_UUID, 16);
1750 if (i == MAX_ACTIVE) {
1751 jfs_warn("Somebody stomped on the journal!");
1759 * synchronous write log superblock:
1761 * write sidestream bypassing write queue:
1762 * at file system mount, log super block is updated for
1763 * activation of the file system before any log record
1764 * (MOUNT record) of the file system, and at file system
1765 * unmount, all meta data for the file system has been
1766 * flushed before log super block is updated for deactivation
1767 * of the file system.
1769 lbmDirectWrite(log, bpsuper, lbmWRITE | lbmRELEASE | lbmSYNC);
1770 rc = lbmIOWait(bpsuper, lbmFREE);
1776 * log buffer manager (lbm)
1777 * ------------------------
1779 * special purpose buffer manager supporting log i/o requirements.
1781 * per log write queue:
1782 * log pageout occurs in serial order by fifo write queue and
1783 * restricting to a single i/o in pregress at any one time.
1784 * a circular singly-linked list
1785 * (log->wrqueue points to the tail, and buffers are linked via
1786 * bp->wrqueue field), and
1787 * maintains log page in pageout ot waiting for pageout in serial pageout.
1793 * initialize per log I/O setup at lmLogInit()
1795 static int lbmLogInit(struct jfs_log * log)
1800 jfs_info("lbmLogInit: log:0x%p", log);
1802 /* initialize current buffer cursor */
1805 /* initialize log device write queue */
1809 * Each log has its own buffer pages allocated to it. These are
1810 * not managed by the page cache. This ensures that a transaction
1811 * writing to the log does not block trying to allocate a page from
1812 * the page cache (for the log). This would be bad, since page
1813 * allocation waits on the kswapd thread that may be committing inodes
1814 * which would cause log activity. Was that clear? I'm trying to
1815 * avoid deadlock here.
1817 init_waitqueue_head(&log->free_wait);
1819 log->lbuf_free = NULL;
1821 for (i = 0; i < LOGPAGES; i++) {
1822 lbuf = kmalloc(sizeof(struct lbuf), GFP_KERNEL);
1825 lbuf->l_ldata = (char *) get_zeroed_page(GFP_KERNEL);
1826 if (lbuf->l_ldata == 0) {
1831 init_waitqueue_head(&lbuf->l_ioevent);
1833 lbuf->l_freelist = log->lbuf_free;
1834 log->lbuf_free = lbuf;
1840 lbmLogShutdown(log);
1848 * finalize per log I/O setup at lmLogShutdown()
1850 static void lbmLogShutdown(struct jfs_log * log)
1854 jfs_info("lbmLogShutdown: log:0x%p", log);
1856 lbuf = log->lbuf_free;
1858 struct lbuf *next = lbuf->l_freelist;
1859 free_page((unsigned long) lbuf->l_ldata);
1871 * allocate an empty log buffer
1873 static struct lbuf *lbmAllocate(struct jfs_log * log, int pn)
1876 unsigned long flags;
1879 * recycle from log buffer freelist if any
1882 LCACHE_SLEEP_COND(log->free_wait, (bp = log->lbuf_free), flags);
1883 log->lbuf_free = bp->l_freelist;
1884 LCACHE_UNLOCK(flags);
1888 bp->l_wqnext = NULL;
1889 bp->l_freelist = NULL;
1892 bp->l_blkno = log->base + (pn << (L2LOGPSIZE - log->l2bsize));
1902 * release a log buffer to freelist
1904 static void lbmFree(struct lbuf * bp)
1906 unsigned long flags;
1912 LCACHE_UNLOCK(flags);
1915 static void lbmfree(struct lbuf * bp)
1917 struct jfs_log *log = bp->l_log;
1919 assert(bp->l_wqnext == NULL);
1922 * return the buffer to head of freelist
1924 bp->l_freelist = log->lbuf_free;
1925 log->lbuf_free = bp;
1927 wake_up(&log->free_wait);
1935 * FUNCTION: add a log buffer to the the log redrive list
1941 * Takes log_redrive_lock.
1943 static inline void lbmRedrive(struct lbuf *bp)
1945 unsigned long flags;
1947 spin_lock_irqsave(&log_redrive_lock, flags);
1948 bp->l_redrive_next = log_redrive_list;
1949 log_redrive_list = bp;
1950 spin_unlock_irqrestore(&log_redrive_lock, flags);
1952 wake_up(&jfs_IO_thread_wait);
1959 static int lbmRead(struct jfs_log * log, int pn, struct lbuf ** bpp)
1965 * allocate a log buffer
1967 *bpp = bp = lbmAllocate(log, pn);
1968 jfs_info("lbmRead: bp:0x%p pn:0x%x", bp, pn);
1970 bp->l_flag |= lbmREAD;
1972 bio = bio_alloc(GFP_NOFS, 1);
1974 bio->bi_sector = bp->l_blkno << (log->l2bsize - 9);
1975 bio->bi_bdev = log->bdev;
1976 bio->bi_io_vec[0].bv_page = virt_to_page(bp->l_ldata);
1977 bio->bi_io_vec[0].bv_len = LOGPSIZE;
1978 bio->bi_io_vec[0].bv_offset = 0;
1982 bio->bi_size = LOGPSIZE;
1984 bio->bi_end_io = lbmIODone;
1985 bio->bi_private = bp;
1986 submit_bio(READ_SYNC, bio);
1988 wait_event(bp->l_ioevent, (bp->l_flag != lbmREAD));
1997 * buffer at head of pageout queue stays after completion of
1998 * partial-page pageout and redriven by explicit initiation of
1999 * pageout by caller until full-page pageout is completed and
2002 * device driver i/o done redrives pageout of new buffer at
2003 * head of pageout queue when current buffer at head of pageout
2004 * queue is released at the completion of its full-page pageout.
2006 * LOGGC_LOCK() serializes lbmWrite() by lmNextPage() and lmGroupCommit().
2007 * LCACHE_LOCK() serializes xflag between lbmWrite() and lbmIODone()
2009 static void lbmWrite(struct jfs_log * log, struct lbuf * bp, int flag,
2013 unsigned long flags;
2015 jfs_info("lbmWrite: bp:0x%p flag:0x%x pn:0x%x", bp, flag, bp->l_pn);
2017 /* map the logical block address to physical block address */
2019 log->base + (bp->l_pn << (L2LOGPSIZE - log->l2bsize));
2021 LCACHE_LOCK(flags); /* disable+lock */
2024 * initialize buffer for device driver
2029 * insert bp at tail of write queue associated with log
2031 * (request is either for bp already/currently at head of queue
2032 * or new bp to be inserted at tail)
2036 /* is buffer not already on write queue ? */
2037 if (bp->l_wqnext == NULL) {
2038 /* insert at tail of wqueue */
2044 bp->l_wqnext = tail->l_wqnext;
2045 tail->l_wqnext = bp;
2051 /* is buffer at head of wqueue and for write ? */
2052 if ((bp != tail->l_wqnext) || !(flag & lbmWRITE)) {
2053 LCACHE_UNLOCK(flags); /* unlock+enable */
2057 LCACHE_UNLOCK(flags); /* unlock+enable */
2061 else if (flag & lbmSYNC)
2074 * initiate pageout bypassing write queue for sidestream
2075 * (e.g., log superblock) write;
2077 static void lbmDirectWrite(struct jfs_log * log, struct lbuf * bp, int flag)
2079 jfs_info("lbmDirectWrite: bp:0x%p flag:0x%x pn:0x%x",
2080 bp, flag, bp->l_pn);
2083 * initialize buffer for device driver
2085 bp->l_flag = flag | lbmDIRECT;
2087 /* map the logical block address to physical block address */
2089 log->base + (bp->l_pn << (L2LOGPSIZE - log->l2bsize));
2092 * initiate pageout of the page
2099 * NAME: lbmStartIO()
2101 * FUNCTION: Interface to DD strategy routine
2105 * serialization: LCACHE_LOCK() is NOT held during log i/o;
2107 static void lbmStartIO(struct lbuf * bp)
2110 struct jfs_log *log = bp->l_log;
2112 jfs_info("lbmStartIO\n");
2114 bio = bio_alloc(GFP_NOFS, 1);
2115 bio->bi_sector = bp->l_blkno << (log->l2bsize - 9);
2116 bio->bi_bdev = log->bdev;
2117 bio->bi_io_vec[0].bv_page = virt_to_page(bp->l_ldata);
2118 bio->bi_io_vec[0].bv_len = LOGPSIZE;
2119 bio->bi_io_vec[0].bv_offset = 0;
2123 bio->bi_size = LOGPSIZE;
2125 bio->bi_end_io = lbmIODone;
2126 bio->bi_private = bp;
2128 /* check if journaling to disk has been disabled */
2129 if (!log->no_integrity) {
2130 submit_bio(WRITE_SYNC, bio);
2131 INCREMENT(lmStat.submitted);
2135 lbmIODone(bio, 0, 0); /* 2nd argument appears to not be used => 0
2136 * 3rd argument appears to not be used => 0
2145 static int lbmIOWait(struct lbuf * bp, int flag)
2147 unsigned long flags;
2150 jfs_info("lbmIOWait1: bp:0x%p flag:0x%x:0x%x", bp, bp->l_flag, flag);
2152 LCACHE_LOCK(flags); /* disable+lock */
2154 LCACHE_SLEEP_COND(bp->l_ioevent, (bp->l_flag & lbmDONE), flags);
2156 rc = (bp->l_flag & lbmERROR) ? -EIO : 0;
2161 LCACHE_UNLOCK(flags); /* unlock+enable */
2163 jfs_info("lbmIOWait2: bp:0x%p flag:0x%x:0x%x", bp, bp->l_flag, flag);
2170 * executed at INTIODONE level
2172 static int lbmIODone(struct bio *bio, unsigned int bytes_done, int error)
2174 struct lbuf *bp = bio->bi_private;
2175 struct lbuf *nextbp, *tail;
2176 struct jfs_log *log;
2177 unsigned long flags;
2183 * get back jfs buffer bound to the i/o buffer
2185 jfs_info("lbmIODone: bp:0x%p flag:0x%x", bp, bp->l_flag);
2187 LCACHE_LOCK(flags); /* disable+lock */
2189 bp->l_flag |= lbmDONE;
2191 if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
2192 bp->l_flag |= lbmERROR;
2194 jfs_err("lbmIODone: I/O error in JFS log");
2202 if (bp->l_flag & lbmREAD) {
2203 bp->l_flag &= ~lbmREAD;
2205 LCACHE_UNLOCK(flags); /* unlock+enable */
2207 /* wakeup I/O initiator */
2208 LCACHE_WAKEUP(&bp->l_ioevent);
2214 * pageout completion
2216 * the bp at the head of write queue has completed pageout.
2218 * if single-commit/full-page pageout, remove the current buffer
2219 * from head of pageout queue, and redrive pageout with
2220 * the new buffer at head of pageout queue;
2221 * otherwise, the partial-page pageout buffer stays at
2222 * the head of pageout queue to be redriven for pageout
2223 * by lmGroupCommit() until full-page pageout is completed.
2225 bp->l_flag &= ~lbmWRITE;
2226 INCREMENT(lmStat.pagedone);
2228 /* update committed lsn */
2230 log->clsn = (bp->l_pn << L2LOGPSIZE) + bp->l_ceor;
2232 if (bp->l_flag & lbmDIRECT) {
2233 LCACHE_WAKEUP(&bp->l_ioevent);
2234 LCACHE_UNLOCK(flags);
2240 /* single element queue */
2242 /* remove head buffer of full-page pageout
2243 * from log device write queue
2245 if (bp->l_flag & lbmRELEASE) {
2247 bp->l_wqnext = NULL;
2250 /* multi element queue */
2252 /* remove head buffer of full-page pageout
2253 * from log device write queue
2255 if (bp->l_flag & lbmRELEASE) {
2256 nextbp = tail->l_wqnext = bp->l_wqnext;
2257 bp->l_wqnext = NULL;
2260 * redrive pageout of next page at head of write queue:
2261 * redrive next page without any bound tblk
2262 * (i.e., page w/o any COMMIT records), or
2263 * first page of new group commit which has been
2264 * queued after current page (subsequent pageout
2265 * is performed synchronously, except page without
2266 * any COMMITs) by lmGroupCommit() as indicated
2269 if (nextbp->l_flag & lbmWRITE) {
2271 * We can't do the I/O at interrupt time.
2272 * The jfsIO thread can do it
2280 * synchronous pageout:
2282 * buffer has not necessarily been removed from write queue
2283 * (e.g., synchronous write of partial-page with COMMIT):
2284 * leave buffer for i/o initiator to dispose
2286 if (bp->l_flag & lbmSYNC) {
2287 LCACHE_UNLOCK(flags); /* unlock+enable */
2289 /* wakeup I/O initiator */
2290 LCACHE_WAKEUP(&bp->l_ioevent);
2294 * Group Commit pageout:
2296 else if (bp->l_flag & lbmGC) {
2297 LCACHE_UNLOCK(flags);
2302 * asynchronous pageout:
2304 * buffer must have been removed from write queue:
2305 * insert buffer at head of freelist where it can be recycled
2308 assert(bp->l_flag & lbmRELEASE);
2309 assert(bp->l_flag & lbmFREE);
2312 LCACHE_UNLOCK(flags); /* unlock+enable */
2318 int jfsIOWait(void *arg)
2324 complete(&jfsIOwait);
2327 DECLARE_WAITQUEUE(wq, current);
2329 spin_lock_irq(&log_redrive_lock);
2330 while ((bp = log_redrive_list) != 0) {
2331 log_redrive_list = bp->l_redrive_next;
2332 bp->l_redrive_next = NULL;
2333 spin_unlock_irq(&log_redrive_lock);
2335 spin_lock_irq(&log_redrive_lock);
2337 if (current->flags & PF_FREEZE) {
2338 spin_unlock_irq(&log_redrive_lock);
2339 refrigerator(PF_FREEZE);
2341 add_wait_queue(&jfs_IO_thread_wait, &wq);
2342 set_current_state(TASK_INTERRUPTIBLE);
2343 spin_unlock_irq(&log_redrive_lock);
2345 current->state = TASK_RUNNING;
2346 remove_wait_queue(&jfs_IO_thread_wait, &wq);
2348 } while (!jfs_stop_threads);
2350 jfs_info("jfsIOWait being killed!");
2351 complete_and_exit(&jfsIOwait, 0);
2355 * NAME: lmLogFormat()/jfs_logform()
2357 * FUNCTION: format file system log
2361 * logAddress - start address of log space in FS block
2362 * logSize - length of log space in FS block;
2364 * RETURN: 0 - success
2367 * XXX: We're synchronously writing one page at a time. This needs to
2368 * be improved by writing multiple pages at once.
2370 int lmLogFormat(struct jfs_log *log, s64 logAddress, int logSize)
2373 struct jfs_sb_info *sbi;
2374 struct logsuper *logsuper;
2376 int lspn; /* log sequence page number */
2377 struct lrd *lrd_ptr;
2381 jfs_info("lmLogFormat: logAddress:%Ld logSize:%d",
2382 (long long)logAddress, logSize);
2384 sbi = list_entry(log->sb_list.next, struct jfs_sb_info, log_list);
2386 /* allocate a log buffer */
2387 bp = lbmAllocate(log, 1);
2389 npages = logSize >> sbi->l2nbperpage;
2394 * page 0 - reserved;
2395 * page 1 - log superblock;
2396 * page 2 - log data page: A SYNC log record is written
2397 * into this page at logform time;
2398 * pages 3-N - log data page: set to empty log data pages;
2401 * init log superblock: log page 1
2403 logsuper = (struct logsuper *) bp->l_ldata;
2405 logsuper->magic = cpu_to_le32(LOGMAGIC);
2406 logsuper->version = cpu_to_le32(LOGVERSION);
2407 logsuper->state = cpu_to_le32(LOGREDONE);
2408 logsuper->flag = cpu_to_le32(sbi->mntflag); /* ? */
2409 logsuper->size = cpu_to_le32(npages);
2410 logsuper->bsize = cpu_to_le32(sbi->bsize);
2411 logsuper->l2bsize = cpu_to_le32(sbi->l2bsize);
2412 logsuper->end = cpu_to_le32(2 * LOGPSIZE + LOGPHDRSIZE + LOGRDSIZE);
2414 bp->l_flag = lbmWRITE | lbmSYNC | lbmDIRECT;
2415 bp->l_blkno = logAddress + sbi->nbperpage;
2417 if ((rc = lbmIOWait(bp, 0)))
2421 * init pages 2 to npages-1 as log data pages:
2423 * log page sequence number (lpsn) initialization:
2426 * +-----+-----+=====+=====+===.....===+=====+
2428 * <--- N page circular file ---->
2430 * the N (= npages-2) data pages of the log is maintained as
2431 * a circular file for the log records;
2432 * lpsn grows by 1 monotonically as each log page is written
2433 * to the circular file of the log;
2434 * and setLogpage() will not reset the page number even if
2435 * the eor is equal to LOGPHDRSIZE. In order for binary search
2436 * still work in find log end process, we have to simulate the
2437 * log wrap situation at the log format time.
2438 * The 1st log page written will have the highest lpsn. Then
2439 * the succeeding log pages will have ascending order of
2440 * the lspn starting from 0, ... (N-2)
2442 lp = (struct logpage *) bp->l_ldata;
2444 * initialize 1st log page to be written: lpsn = N - 1,
2445 * write a SYNCPT log record is written to this page
2447 lp->h.page = lp->t.page = cpu_to_le32(npages - 3);
2448 lp->h.eor = lp->t.eor = cpu_to_le16(LOGPHDRSIZE + LOGRDSIZE);
2450 lrd_ptr = (struct lrd *) &lp->data;
2451 lrd_ptr->logtid = 0;
2452 lrd_ptr->backchain = 0;
2453 lrd_ptr->type = cpu_to_le16(LOG_SYNCPT);
2454 lrd_ptr->length = 0;
2455 lrd_ptr->log.syncpt.sync = 0;
2457 bp->l_blkno += sbi->nbperpage;
2458 bp->l_flag = lbmWRITE | lbmSYNC | lbmDIRECT;
2460 if ((rc = lbmIOWait(bp, 0)))
2464 * initialize succeeding log pages: lpsn = 0, 1, ..., (N-2)
2466 for (lspn = 0; lspn < npages - 3; lspn++) {
2467 lp->h.page = lp->t.page = cpu_to_le32(lspn);
2468 lp->h.eor = lp->t.eor = cpu_to_le16(LOGPHDRSIZE);
2470 bp->l_blkno += sbi->nbperpage;
2471 bp->l_flag = lbmWRITE | lbmSYNC | lbmDIRECT;
2473 if ((rc = lbmIOWait(bp, 0)))
2482 /* release the buffer */
2488 #ifdef CONFIG_JFS_STATISTICS
2489 int jfs_lmstats_read(char *buffer, char **start, off_t offset, int length,
2490 int *eof, void *data)
2495 len += sprintf(buffer,
2496 "JFS Logmgr stats\n"
2497 "================\n"
2499 "writes submitted = %d\n"
2500 "writes completed = %d\n"
2501 "full pages submitted = %d\n"
2502 "partial pages submitted = %d\n",
2507 lmStat.partial_page);
2510 *start = buffer + begin;
2523 #endif /* CONFIG_JFS_STATISTICS */