1 /*
2 * Copyright (C) International Business Machines Corp., 2000-2004
3 * Portions Copyright (C) Christoph Hellwig, 2001-2002
4 *
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.
9 *
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.
14 *
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
18 */
19
20 /*
21 * jfs_logmgr.c: log manager
22 *
23 * for related information, see transaction manager (jfs_txnmgr.c), and
24 * recovery manager (jfs_logredo.c).
25 *
26 * note: for detail, RTFS.
27 *
28 * log buffer manager:
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
35 *
36 * group commit:
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.
40 *
41 * TBD lazy commit:
42 * transactions are committed asynchronously when the log page
43 * containing it COMMIT is paged out when it becomes full;
44 *
45 * serialization:
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;
49 *
50 * TBD log integrity:
51 * careful-write (ping-pong) of last logpage to recover from crash
52 * in overwrite.
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).
56 *
57 * alternatives:
58 * lsn - 64-bit monotonically increasing integer vs
59 * 32-bit lspn and page eor.
60 */
61
62 #include <linux/fs.h>
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"
75
76
77 /*
78 * lbuf's ready to be redriven. Protected by log_redrive_lock (jfsIO thread)
79 */
80 static struct lbuf *log_redrive_list;
81 static DEFINE_SPINLOCK(log_redrive_lock);
82 DECLARE_WAIT_QUEUE_HEAD(jfs_IO_thread_wait);
83
84
85 /*
86 * log read/write serialization (per log)
87 */
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))
91
92
93 /*
94 * log group commit serialization (per log)
95 */
96
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)
101
102 /*
103 * log sync serialization (per log)
104 */
105 #define LOGSYNC_DELTA(logsize) min((logsize)/8, 128*LOGPSIZE)
106 #define LOGSYNC_BARRIER(logsize) ((logsize)/4)
107 /*
108 #define LOGSYNC_DELTA(logsize) min((logsize)/4, 256*LOGPSIZE)
109 #define LOGSYNC_BARRIER(logsize) ((logsize)/2)
110 */
111
112
113 /*
114 * log buffer cache synchronization
115 */
116 static DEFINE_SPINLOCK(jfsLCacheLock);
117
118 #define LCACHE_LOCK(flags) spin_lock_irqsave(&jfsLCacheLock, flags)
119 #define LCACHE_UNLOCK(flags) spin_unlock_irqrestore(&jfsLCacheLock, flags)
120
121 /*
122 * See __SLEEP_COND in jfs_locks.h
123 */
124 #define LCACHE_SLEEP_COND(wq, cond, flags) \
125 do { \
126 if (cond) \
127 break; \
128 __SLEEP_COND(wq, cond, LCACHE_LOCK(flags), LCACHE_UNLOCK(flags)); \
129 } while (0)
130
131 #define LCACHE_WAKEUP(event) wake_up(event)
132
133
134 /*
135 * lbuf buffer cache (lCache) control
136 */
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;
141 */
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;
146 */
147 #define lbmSYNC 0x0008 /* do not return to freelist
148 * when removed from write queue;
149 */
150 #define lbmFREE 0x0010 /* return to freelist
151 * at completion of pageout;
152 * the buffer may be recycled;
153 */
154 #define lbmDONE 0x0020
155 #define lbmERROR 0x0040
156 #define lbmGC 0x0080 /* lbmIODone to perform post-GC processing
157 * of log page
158 */
159 #define lbmDIRECT 0x0100
160
161 /*
162 * Global list of active external journals
163 */
164 static LIST_HEAD(jfs_external_logs);
165 static struct jfs_log *dummy_log = NULL;
166 static DECLARE_MUTEX(jfs_log_sem);
167
168 /*
169 * external references
170 */
171 extern void txLazyUnlock(struct tblock * tblk);
172 extern int jfs_stop_threads;
173 extern struct completion jfsIOwait;
174 extern int jfs_tlocks_low;
175
176 /*
177 * forward references
178 */
179 static int lmWriteRecord(struct jfs_log * log, struct tblock * tblk,
180 struct lrd * lrd, struct tlock * tlck);
181
182 static int lmNextPage(struct jfs_log * log);
183 static int lmLogFileSystem(struct jfs_log * log, struct jfs_sb_info *sbi,
184 int activate);
185
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);
201
202
203
204 /*
205 * statistics
206 */
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 */
214 } lmStat;
215 #endif
216
217
218 /*
219 * NAME: lmLog()
220 *
221 * FUNCTION: write a log record;
222 *
223 * PARAMETER:
224 *
225 * RETURN: lsn - offset to the next log record to write (end-of-log);
226 * -1 - error;
227 *
228 * note: todo: log error handler
229 */
230 int lmLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
231 struct tlock * tlck)
232 {
233 int lsn;
234 int diffp, difft;
235 struct metapage *mp = NULL;
236
237 jfs_info("lmLog: log:0x%p tblk:0x%p, lrd:0x%p tlck:0x%p",
238 log, tblk, lrd, tlck);
239
240 LOG_LOCK(log);
241
242 /* log by (out-of-transaction) JFS ? */
243 if (tblk == NULL)
244 goto writeRecord;
245
246 /* log from page ? */
247 if (tlck == NULL ||
248 tlck->type & tlckBTROOT || (mp = tlck->mp) == NULL)
249 goto writeRecord;
250
251 /*
252 * initialize/update page/transaction recovery lsn
253 */
254 lsn = log->lsn;
255
256 LOGSYNC_LOCK(log);
257
258 /*
259 * initialize page lsn if first log write of the page
260 */
261 if (mp->lsn == 0) {
262 mp->log = log;
263 mp->lsn = lsn;
264 log->count++;
265
266 /* insert page at tail of logsynclist */
267 list_add_tail(&mp->synclist, &log->synclist);
268 }
269
270 /*
271 * initialize/update lsn of tblock of the page
272 *
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
278 * allocation map);
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.
284 */
285 /*
286 * initialize transaction lsn:
287 */
288 if (tblk->lsn == 0) {
289 /* inherit lsn of its first page logged */
290 tblk->lsn = mp->lsn;
291 log->count++;
292
293 /* insert tblock after the page on logsynclist */
294 list_add(&tblk->synclist, &mp->synclist);
295 }
296 /*
297 * update transaction lsn:
298 */
299 else {
300 /* inherit oldest/smallest lsn of page */
301 logdiff(diffp, mp->lsn, log);
302 logdiff(difft, tblk->lsn, log);
303 if (diffp < difft) {
304 /* update tblock lsn with page lsn */
305 tblk->lsn = mp->lsn;
306
307 /* move tblock after page on logsynclist */
308 list_move(&tblk->synclist, &mp->synclist);
309 }
310 }
311
312 LOGSYNC_UNLOCK(log);
313
314 /*
315 * write the log record
316 */
317 writeRecord:
318 lsn = lmWriteRecord(log, tblk, lrd, tlck);
319
320 /*
321 * forward log syncpt if log reached next syncpt trigger
322 */
323 logdiff(diffp, lsn, log);
324 if (diffp >= log->nextsync)
325 lsn = lmLogSync(log, 0);
326
327 /* update end-of-log lsn */
328 log->lsn = lsn;
329
330 LOG_UNLOCK(log);
331
332 /* return end-of-log address */
333 return lsn;
334 }
335
336
337 /*
338 * NAME: lmWriteRecord()
339 *
340 * FUNCTION: move the log record to current log page
341 *
342 * PARAMETER: cd - commit descriptor
343 *
344 * RETURN: end-of-log address
345 *
346 * serialization: LOG_LOCK() held on entry/exit
347 */
348 static int
349 lmWriteRecord(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
350 struct tlock * tlck)
351 {
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 */
359 caddr_t src;
360 int srclen;
361 int nbytes; /* number of bytes to move */
362 int i;
363 int len;
364 struct linelock *linelock;
365 struct lv *lv;
366 struct lvd *lvd;
367 int l2linesize;
368
369 len = 0;
370
371 /* retrieve destination log page to write */
372 bp = (struct lbuf *) log->bp;
373 lp = (struct logpage *) bp->l_ldata;
374 dstoffset = log->eor;
375
376 /* any log data to write ? */
377 if (tlck == NULL)
378 goto moveLrd;
379
380 /*
381 * move log record data
382 */
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;
387 }
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;
392 else
393 p = (caddr_t) &JFS_IP(tlck->ip)->i_xtroot;
394 linelock = (struct linelock *) & tlck->lock;
395 }
396 #ifdef _JFS_WIP
397 else if (tlck->flag & tlckINLINELOCK) {
398
399 inlinelock = (struct inlinelock *) & tlck;
400 p = (caddr_t) & inlinelock->pxd;
401 linelock = (struct linelock *) & tlck;
402 }
403 #endif /* _JFS_WIP */
404 else {
405 jfs_err("lmWriteRecord: UFO tlck:0x%p", tlck);
406 return 0; /* Probably should trap */
407 }
408 l2linesize = linelock->l2linesize;
409
410 moveData:
411 ASSERT(linelock->index <= linelock->maxcnt);
412
413 lv = linelock->lv;
414 for (i = 0; i < linelock->index; i++, lv++) {
415 if (lv->length == 0)
416 continue;
417
418 /* is page full ? */
419 if (dstoffset >= LOGPSIZE - LOGPTLRSIZE) {
420 /* page become full: move on to next page */
421 lmNextPage(log);
422
423 bp = log->bp;
424 lp = (struct logpage *) bp->l_ldata;
425 dstoffset = LOGPHDRSIZE;
426 }
427
428 /*
429 * move log vector data
430 */
431 src = (u8 *) p + (lv->offset << l2linesize);
432 srclen = lv->length << l2linesize;
433 len += srclen;
434 while (srclen > 0) {
435 freespace = (LOGPSIZE - LOGPTLRSIZE) - dstoffset;
436 nbytes = min(freespace, srclen);
437 dst = (caddr_t) lp + dstoffset;
438 memcpy(dst, src, nbytes);
439 dstoffset += nbytes;
440
441 /* is page not full ? */
442 if (dstoffset < LOGPSIZE - LOGPTLRSIZE)
443 break;
444
445 /* page become full: move on to next page */
446 lmNextPage(log);
447
448 bp = (struct lbuf *) log->bp;
449 lp = (struct logpage *) bp->l_ldata;
450 dstoffset = LOGPHDRSIZE;
451
452 srclen -= nbytes;
453 src += nbytes;
454 }
455
456 /*
457 * move log vector descriptor
458 */
459 len += 4;
460 lvd = (struct lvd *) ((caddr_t) lp + dstoffset);
461 lvd->offset = cpu_to_le16(lv->offset);
462 lvd->length = cpu_to_le16(lv->length);
463 dstoffset += 4;
464 jfs_info("lmWriteRecord: lv offset:%d length:%d",
465 lv->offset, lv->length);
466 }
467
468 if ((i = linelock->next)) {
469 linelock = (struct linelock *) lid_to_tlock(i);
470 goto moveData;
471 }
472
473 /*
474 * move log record descriptor
475 */
476 moveLrd:
477 lrd->length = cpu_to_le16(len);
478
479 src = (caddr_t) lrd;
480 srclen = LOGRDSIZE;
481
482 while (srclen > 0) {
483 freespace = (LOGPSIZE - LOGPTLRSIZE) - dstoffset;
484 nbytes = min(freespace, srclen);
485 dst = (caddr_t) lp + dstoffset;
486 memcpy(dst, src, nbytes);
487
488 dstoffset += nbytes;
489 srclen -= nbytes;
490
491 /* are there more to move than freespace of page ? */
492 if (srclen)
493 goto pageFull;
494
495 /*
496 * end of log record descriptor
497 */
498
499 /* update last log record eor */
500 log->eor = dstoffset;
501 bp->l_eor = dstoffset;
502 lsn = (log->page << L2LOGPSIZE) + dstoffset;
503
504 if (lrd->type & cpu_to_le16(LOG_COMMIT)) {
505 tblk->clsn = lsn;
506 jfs_info("wr: tclsn:0x%x, beor:0x%x", tblk->clsn,
507 bp->l_eor);
508
509 INCREMENT(lmStat.commit); /* # of commit */
510
511 /*
512 * enqueue tblock for group commit:
513 *
514 * enqueue tblock of non-trivial/synchronous COMMIT
515 * at tail of group commit queue
516 * (trivial/asynchronous COMMITs are ignored by
517 * group commit.)
518 */
519 LOGGC_LOCK(log);
520
521 /* init tblock gc state */
522 tblk->flag = tblkGC_QUEUE;
523 tblk->bp = log->bp;
524 tblk->pn = log->page;
525 tblk->eor = log->eor;
526
527 /* enqueue transaction to commit queue */
528 list_add_tail(&tblk->cqueue, &log->cqueue);
529
530 LOGGC_UNLOCK(log);
531 }
532
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);
535
536 /* page not full ? */
537 if (dstoffset < LOGPSIZE - LOGPTLRSIZE)
538 return lsn;
539
540 pageFull:
541 /* page become full: move on to next page */
542 lmNextPage(log);
543
544 bp = (struct lbuf *) log->bp;
545 lp = (struct logpage *) bp->l_ldata;
546 dstoffset = LOGPHDRSIZE;
547 src += nbytes;
548 }
549
550 return lsn;
551 }
552
553
554 /*
555 * NAME: lmNextPage()
556 *
557 * FUNCTION: write current page and allocate next page.
558 *
559 * PARAMETER: log
560 *
561 * RETURN: 0
562 *
563 * serialization: LOG_LOCK() held on entry/exit
564 */
565 static int lmNextPage(struct jfs_log * log)
566 {
567 struct logpage *lp;
568 int lspn; /* log sequence page number */
569 int pn; /* current page number */
570 struct lbuf *bp;
571 struct lbuf *nextbp;
572 struct tblock *tblk;
573
574 /* get current log page number and log sequence page number */
575 pn = log->page;
576 bp = log->bp;
577 lp = (struct logpage *) bp->l_ldata;
578 lspn = le32_to_cpu(lp->h.page);
579
580 LOGGC_LOCK(log);
581
582 /*
583 * write or queue the full page at the tail of write queue
584 */
585 /* get the tail tblk on commit queue */
586 if (list_empty(&log->cqueue))
587 tblk = NULL;
588 else
589 tblk = list_entry(log->cqueue.prev, struct tblock, cqueue);
590
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
597 * currently);
598 */
599
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;
604
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
610 * continuously
611 */
612 if (bp->l_wqnext == NULL)
613 lbmWrite(log, bp, 0, 0);
614 } else {
615 /*
616 * No current GC leader, initiate group commit
617 */
618 log->cflag |= logGC_PAGEOUT;
619 lmGCwrite(log, 0);
620 }
621 }
622 /* page is not bound with outstanding tblk:
623 * init write or mark it to be redriven (lbmWRITE)
624 */
625 else {
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);
630 }
631 LOGGC_UNLOCK(log);
632
633 /*
634 * allocate/initialize next page
635 */
636 /* if log wraps, the first data page of log is 2
637 * (0 never used, 1 is superblock).
638 */
639 log->page = (pn == log->size - 1) ? 2 : pn + 1;
640 log->eor = LOGPHDRSIZE; /* ? valid page empty/full at logRedo() */
641
642 /* allocate/initialize next log page buffer */
643 nextbp = lbmAllocate(log, log->page);
644 nextbp->l_eor = log->eor;
645 log->bp = nextbp;
646
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);
651
652 return 0;
653 }
654
655
656 /*
657 * NAME: lmGroupCommit()
658 *
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.
663 *
664 * RETURN:
665 *
666 * NOTE:
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().
670 */
671 int lmGroupCommit(struct jfs_log * log, struct tblock * tblk)
672 {
673 int rc = 0;
674
675 LOGGC_LOCK(log);
676
677 /* group committed already ? */
678 if (tblk->flag & tblkGC_COMMITTED) {
679 if (tblk->flag & tblkGC_ERROR)
680 rc = -EIO;
681
682 LOGGC_UNLOCK(log);
683 return rc;
684 }
685 jfs_info("lmGroup Commit: tblk = 0x%p, gcrtc = %d", tblk, log->gcrtc);
686
687 if (tblk->xflag & COMMIT_LAZY)
688 tblk->flag |= tblkGC_LAZY;
689
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)) {
693 /*
694 * No pageout in progress
695 *
696 * start group commit as its group leader.
697 */
698 log->cflag |= logGC_PAGEOUT;
699
700 lmGCwrite(log, 0);
701 }
702
703 if (tblk->xflag & COMMIT_LAZY) {
704 /*
705 * Lazy transactions can leave now
706 */
707 LOGGC_UNLOCK(log);
708 return 0;
709 }
710
711 /* lmGCwrite gives up LOGGC_LOCK, check again */
712
713 if (tblk->flag & tblkGC_COMMITTED) {
714 if (tblk->flag & tblkGC_ERROR)
715 rc = -EIO;
716
717 LOGGC_UNLOCK(log);
718 return rc;
719 }
720
721 /* upcount transaction waiting for completion
722 */
723 log->gcrtc++;
724 tblk->flag |= tblkGC_READY;
725
726 __SLEEP_COND(tblk->gcwait, (tblk->flag & tblkGC_COMMITTED),
727 LOGGC_LOCK(log), LOGGC_UNLOCK(log));
728
729 /* removed from commit queue */
730 if (tblk->flag & tblkGC_ERROR)
731 rc = -EIO;
732
733 LOGGC_UNLOCK(log);
734 return rc;
735 }
736
737 /*
738 * NAME: lmGCwrite()
739 *
740 * FUNCTION: group commit write
741 * initiate write of log page, building a group of all transactions
742 * with commit records on that page.
743 *
744 * RETURN: None
745 *
746 * NOTE:
747 * LOGGC_LOCK must be held by caller.
748 * N.B. LOG_LOCK is NOT held during lmGroupCommit().
749 */
750 static void lmGCwrite(struct jfs_log * log, int cant_write)
751 {
752 struct lbuf *bp;
753 struct logpage *lp;
754 int gcpn; /* group commit page number */
755 struct tblock *tblk;
756 struct tblock *xtblk = NULL;
757
758 /*
759 * build the commit group of a log page
760 *
761 * scan commit queue and make a commit group of all
762 * transactions with COMMIT records on the same log page.
763 */
764 /* get the head tblk on the commit queue */
765 gcpn = list_entry(log->cqueue.next, struct tblock, cqueue)->pn;
766
767 list_for_each_entry(tblk, &log->cqueue, cqueue) {
768 if (tblk->pn != gcpn)
769 break;
770
771 xtblk = tblk;
772
773 /* state transition: (QUEUE, READY) -> COMMIT */
774 tblk->flag |= tblkGC_COMMIT;
775 }
776 tblk = xtblk; /* last tblk of the page */
777
778 /*
779 * pageout to commit transactions on the log page.
780 */
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,
791 cant_write);
792 INCREMENT(lmStat.full_page);
793 }
794 /* page is not yet full */
795 else {
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);
800 }
801 }
802
803 /*
804 * NAME: lmPostGC()
805 *
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.
809 *
810 * RETURN: None
811 *
812 * NOTE:
813 * This routine is called a interrupt time by lbmIODone
814 */
815 static void lmPostGC(struct lbuf * bp)
816 {
817 unsigned long flags;
818 struct jfs_log *log = bp->l_log;
819 struct logpage *lp;
820 struct tblock *tblk, *temp;
821
822 //LOGGC_LOCK(log);
823 spin_lock_irqsave(&log->gclock, flags);
824 /*
825 * current pageout of group commit completed.
826 *
827 * remove/wakeup transactions from commit queue who were
828 * group committed with the current log page
829 */
830 list_for_each_entry_safe(tblk, temp, &log->cqueue, cqueue) {
831 if (!(tblk->flag & tblkGC_COMMIT))
832 break;
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.
836 */
837
838 if (bp->l_flag & lbmERROR)
839 tblk->flag |= tblkGC_ERROR;
840
841 /* remove it from the commit queue */
842 list_del(&tblk->cqueue);
843 tblk->flag &= ~tblkGC_QUEUE;
844
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;
849 }
850
851 jfs_info("lmPostGC: tblk = 0x%p, flag = 0x%x", tblk,
852 tblk->flag);
853
854 if (!(tblk->xflag & COMMIT_FORCE))
855 /*
856 * Hand tblk over to lazy commit thread
857 */
858 txLazyUnlock(tblk);
859 else {
860 /* state transition: COMMIT -> COMMITTED */
861 tblk->flag |= tblkGC_COMMITTED;
862
863 if (tblk->flag & tblkGC_READY)
864 log->gcrtc--;
865
866 LOGGC_WAKEUP(tblk);
867 }
868
869 /* was page full before pageout ?
870 * (and this is the last tblk bound with the page)
871 */
872 if (tblk->flag & tblkGC_FREE)
873 lbmFree(bp);
874 /* did page become full after pageout ?
875 * (and this is the last tblk bound with the page)
876 */
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,
884 1);
885 }
886
887 }
888
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.
895 */
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))
899 /*
900 * Call lmGCwrite with new group leader
901 */
902 lmGCwrite(log, 1);
903
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.
908 */
909 else
910 log->cflag &= ~logGC_PAGEOUT;
911
912 //LOGGC_UNLOCK(log);
913 spin_unlock_irqrestore(&log->gclock, flags);
914 return;
915 }
916
917 /*
918 * NAME: lmLogSync()
919 *
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
923 * process).
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.
928 *
929 * this is called only from lmLog().
930 *
931 * PARAMETER: ip - pointer to logs inode.
932 *
933 * RETURN: 0
934 *
935 * serialization: LOG_LOCK() held on entry/exit
936 */
937 static int lmLogSync(struct jfs_log * log, int nosyncwait)
938 {
939 int logsize;
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 */
944 struct lrd lrd;
945 int lsn;
946 struct logsyncblk *lp;
947
948 /*
949 * forward syncpt
950 */
951 /* if last sync is same as last syncpt,
952 * invoke sync point forward processing to update sync.
953 */
954
955 if (log->sync == log->syncpt) {
956 LOGSYNC_LOCK(log);
957 /* ToDo: push dirty metapages out to disk */
958 // bmLogSync(log);
959
960 if (list_empty(&log->synclist))
961 log->sync = log->lsn;
962 else {
963 lp = list_entry(log->synclist.next,
964 struct logsyncblk, synclist);
965 log->sync = lp->lsn;
966 }
967 LOGSYNC_UNLOCK(log);
968
969 }
970
971 /* if sync is different from last syncpt,
972 * write a SYNCPT record with syncpt = sync.
973 * reset syncpt = sync
974 */
975 if (log->sync != log->syncpt) {
976 struct jfs_sb_info *sbi;
977
978 /*
979 * We need to make sure all of the "written" metapages
980 * actually make it to disk
981 */
982 list_for_each_entry(sbi, &log->sb_list, log_list) {
983 if (sbi->flag & JFS_NOINTEGRITY)
984 continue;
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);
988 }
989 list_for_each_entry(sbi, &log->sb_list, log_list) {
990 if (sbi->flag & JFS_NOINTEGRITY)
991 continue;
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);
995 }
996
997 lrd.logtid = 0;
998 lrd.backchain = 0;
999 lrd.type = cpu_to_le16(LOG_SYNCPT);
1000 lrd.length = 0;
1001 lrd.log.syncpt.sync = cpu_to_le32(log->sync);
1002 lsn = lmWriteRecord(log, NULL, &lrd, NULL);
1003
1004 log->syncpt = log->sync;
1005 } else
1006 lsn = log->lsn;
1007
1008 /*
1009 * setup next syncpt trigger (SWAG)
1010 */
1011 logsize = log->logsize;
1012
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");
1019 /*
1020 * log wrapping
1021 *
1022 * option 1 - panic ? No.!
1023 * option 2 - shutdown file systems
1024 * associated with log ?
1025 * option 3 - extend log ?
1026 */
1027 /*
1028 * option 4 - second chance
1029 *
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().
1035 */
1036 /* mark log state log wrap in log superblock */
1037 /* log->state = LOGWRAP; */
1038
1039 /* reset sync point computation */
1040 log->syncpt = log->sync = lsn;
1041 log->nextsync = delta;
1042 } else
1043 /* next syncpt trigger = written + more */
1044 log->nextsync = written + more;
1045
1046 /* return if lmLogSync() from outside of transaction, e.g., sync() */
1047 if (nosyncwait)
1048 return lsn;
1049
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.
1054 */
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,
1058 log->syncpt);
1059 /*
1060 * We may have to initiate group commit
1061 */
1062 jfs_flush_journal(log, 0);
1063 }
1064
1065 return lsn;
1066 }
1067
1068
1069 /*
1070 * NAME: lmLogOpen()
1071 *
1072 * FUNCTION: open the log on first open;
1073 * insert filesystem in the active list of the log.
1074 *
1075 * PARAMETER: ipmnt - file system mount inode
1076 * iplog - log inode (out)
1077 *
1078 * RETURN:
1079 *
1080 * serialization:
1081 */
1082 int lmLogOpen(struct super_block *sb)
1083 {
1084 int rc;
1085 struct block_device *bdev;
1086 struct jfs_log *log;
1087 struct jfs_sb_info *sbi = JFS_SBI(sb);
1088
1089 if (sbi->flag & JFS_NOINTEGRITY)
1090 return open_dummy_log(sb);
1091
1092 if (sbi->mntflag & JFS_INLINELOG)
1093 return open_inline_log(sb);
1094
1095 down(&jfs_log_sem);
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");
1101 up(&jfs_log_sem);
1102 return -EINVAL;
1103 }
1104 /*
1105 * add file system to log active file system list
1106 */
1107 if ((rc = lmLogFileSystem(log, sbi, 1))) {
1108 up(&jfs_log_sem);
1109 return rc;
1110 }
1111 goto journal_found;
1112 }
1113 }
1114
1115 if (!(log = kmalloc(sizeof(struct jfs_log), GFP_KERNEL))) {
1116 up(&jfs_log_sem);
1117 return -ENOMEM;
1118 }
1119 memset(log, 0, sizeof(struct jfs_log));
1120 INIT_LIST_HEAD(&log->sb_list);
1121
1122 /*
1123 * external log as separate logical volume
1124 *
1125 * file systems to log may have n-to-1 relationship;
1126 */
1127
1128 bdev = open_by_devnum(sbi->logdev, FMODE_READ|FMODE_WRITE);
1129 if (IS_ERR(bdev)) {
1130 rc = -PTR_ERR(bdev);
1131 goto free;
1132 }
1133
1134 if ((rc = bd_claim(bdev, log))) {
1135 goto close;
1136 }
1137
1138 log->bdev = bdev;
1139 memcpy(log->uuid, sbi->loguuid, sizeof(log->uuid));
1140
1141 /*
1142 * initialize log:
1143 */
1144 if ((rc = lmLogInit(log)))
1145 goto unclaim;
1146
1147 list_add(&log->journal_list, &jfs_external_logs);
1148
1149 /*
1150 * add file system to log active file system list
1151 */
1152 if ((rc = lmLogFileSystem(log, sbi, 1)))
1153 goto shutdown;
1154
1155 journal_found:
1156 LOG_LOCK(log);
1157 list_add(&sbi->log_list, &log->sb_list);
1158 sbi->log = log;
1159 LOG_UNLOCK(log);
1160
1161 up(&jfs_log_sem);
1162 return 0;
1163
1164 /*
1165 * unwind on error
1166 */
1167 shutdown: /* unwind lbmLogInit() */
1168 list_del(&log->journal_list);
1169 lbmLogShutdown(log);
1170
1171 unclaim:
1172 bd_release(bdev);
1173
1174 close: /* close external log device */
1175 blkdev_put(bdev);
1176
1177 free: /* free log descriptor */
1178 up(&jfs_log_sem);
1179 kfree(log);
1180
1181 jfs_warn("lmLogOpen: exit(%d)", rc);
1182 return rc;
1183 }
1184
1185 static int open_inline_log(struct super_block *sb)
1186 {
1187 struct jfs_log *log;
1188 int rc;
1189
1190 if (!(log = kmalloc(sizeof(struct jfs_log), GFP_KERNEL)))
1191 return -ENOMEM;
1192 memset(log, 0, sizeof(struct jfs_log));
1193 INIT_LIST_HEAD(&log->sb_list);
1194
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);
1202
1203 /*
1204 * initialize log.
1205 */
1206 if ((rc = lmLogInit(log))) {
1207 kfree(log);
1208 jfs_warn("lmLogOpen: exit(%d)", rc);
1209 return rc;
1210 }
1211
1212 list_add(&JFS_SBI(sb)->log_list, &log->sb_list);
1213 JFS_SBI(sb)->log = log;
1214
1215 return rc;
1216 }
1217
1218 static int open_dummy_log(struct super_block *sb)
1219 {
1220 int rc;
1221
1222 down(&jfs_log_sem);
1223 if (!dummy_log) {
1224 dummy_log = kmalloc(sizeof(struct jfs_log), GFP_KERNEL);
1225 if (!dummy_log) {
1226 up(&jfs_log_sem);
1227 return -ENOMEM;
1228 }
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);
1236 if (rc) {
1237 kfree(dummy_log);
1238 dummy_log = NULL;
1239 up(&jfs_log_sem);
1240 return rc;
1241 }
1242 }
1243
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);
1248 up(&jfs_log_sem);
1249
1250 return 0;
1251 }
1252
1253 /*
1254 * NAME: lmLogInit()
1255 *
1256 * FUNCTION: log initialization at first log open.
1257 *
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.
1262 *
1263 * PARAMETER: log - log structure
1264 *
1265 * RETURN: 0 - if ok
1266 * -EINVAL - bad log magic number or superblock dirty
1267 * error returned from logwait()
1268 *
1269 * serialization: single first open thread
1270 */
1271 int lmLogInit(struct jfs_log * log)
1272 {
1273 int rc = 0;
1274 struct lrd lrd;
1275 struct logsuper *logsuper;
1276 struct lbuf *bpsuper;
1277 struct lbuf *bp;
1278 struct logpage *lp;
1279 int lsn = 0;
1280
1281 jfs_info("lmLogInit: log:0x%p", log);
1282
1283 /* initialize the group commit serialization lock */
1284 LOGGC_LOCK_INIT(log);
1285
1286 /* allocate/initialize the log write serialization lock */
1287 LOG_LOCK_INIT(log);
1288
1289 LOGSYNC_LOCK_INIT(log);
1290
1291 INIT_LIST_HEAD(&log->synclist);
1292
1293 init_waitqueue_head(&log->syncwait);
1294
1295 INIT_LIST_HEAD(&log->cqueue);
1296 log->flush_tblk = NULL;
1297
1298 log->count = 0;
1299
1300 /*
1301 * initialize log i/o
1302 */
1303 if ((rc = lbmLogInit(log)))
1304 return rc;
1305
1306 if (!test_bit(log_INLINELOG, &log->flag))
1307 log->l2bsize = L2LOGPSIZE;
1308
1309 /* check for disabled journaling to disk */
1310 if (log->no_integrity) {
1311 /*
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.
1315 */
1316 bp = lbmAllocate(log , 0);
1317 log->bp = bp;
1318 bp->l_pn = bp->l_eor = 0;
1319 } else {
1320 /*
1321 * validate log superblock
1322 */
1323 if ((rc = lbmRead(log, 1, &bpsuper)))
1324 goto errout10;
1325
1326 logsuper = (struct logsuper *) bpsuper->l_ldata;
1327
1328 if (logsuper->magic != cpu_to_le32(LOGMAGIC)) {
1329 jfs_warn("*** Log Format Error ! ***");
1330 rc = -EINVAL;
1331 goto errout20;
1332 }
1333
1334 /* logredo() should have been run successfully. */
1335 if (logsuper->state != cpu_to_le32(LOGREDONE)) {
1336 jfs_warn("*** Log Is Dirty ! ***");
1337 rc = -EINVAL;
1338 goto errout20;
1339 }
1340
1341 /* initialize log from log superblock */
1342 if (test_bit(log_INLINELOG,&log->flag)) {
1343 if (log->size != le32_to_cpu(logsuper->size)) {
1344 rc = -EINVAL;
1345 goto errout20;
1346 }
1347 jfs_info("lmLogInit: inline log:0x%p base:0x%Lx "
1348 "size:0x%x", log,
1349 (unsigned long long) log->base, log->size);
1350 } else {
1351 if (memcmp(logsuper->uuid, log->uuid, 16)) {
1352 jfs_warn("wrong uuid on JFS log device");
1353 goto errout20;
1354 }
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 "
1358 "size:0x%x", log,
1359 (unsigned long long) log->base, log->size);
1360 }
1361
1362 log->page = le32_to_cpu(logsuper->end) / LOGPSIZE;
1363 log->eor = le32_to_cpu(logsuper->end) - (LOGPSIZE * log->page);
1364
1365 /*
1366 * initialize for log append write mode
1367 */
1368 /* establish current/end-of-log page/buffer */
1369 if ((rc = lbmRead(log, log->page, &bp)))
1370 goto errout20;
1371
1372 lp = (struct logpage *) bp->l_ldata;
1373
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));
1377
1378 log->bp = bp;
1379 bp->l_pn = log->page;
1380 bp->l_eor = log->eor;
1381
1382 /* if current page is full, move on to next page */
1383 if (log->eor >= LOGPSIZE - LOGPTLRSIZE)
1384 lmNextPage(log);
1385
1386 /*
1387 * initialize log syncpoint
1388 */
1389 /*
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
1394 * freelist;
1395 */
1396 lrd.logtid = 0;
1397 lrd.backchain = 0;
1398 lrd.type = cpu_to_le16(LOG_SYNCPT);
1399 lrd.length = 0;
1400 lrd.log.syncpt.sync = 0;
1401 lsn = lmWriteRecord(log, NULL, &lrd, NULL);
1402 bp = log->bp;
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)))
1408 goto errout30;
1409
1410 /*
1411 * update/write superblock
1412 */
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)))
1418 goto errout30;
1419 }
1420
1421 /* initialize logsync parameters */
1422 log->logsize = (log->size - 2) << L2LOGPSIZE;
1423 log->lsn = lsn;
1424 log->syncpt = lsn;
1425 log->sync = log->syncpt;
1426 log->nextsync = LOGSYNC_DELTA(log->logsize);
1427
1428 jfs_info("lmLogInit: lsn:0x%x syncpt:0x%x sync:0x%x",
1429 log->lsn, log->syncpt, log->sync);
1430
1431 /*
1432 * initialize for lazy/group commit
1433 */
1434 log->clsn = lsn;
1435
1436 return 0;
1437
1438 /*
1439 * unwind on error
1440 */
1441 errout30: /* release log page */
1442 log->wqueue = NULL;
1443 bp->l_wqnext = NULL;
1444 lbmFree(bp);
1445
1446 errout20: /* release log superblock */
1447 lbmFree(bpsuper);
1448
1449 errout10: /* unwind lbmLogInit() */
1450 lbmLogShutdown(log);
1451
1452 jfs_warn("lmLogInit: exit(%d)", rc);
1453 return rc;
1454 }
1455
1456
1457 /*
1458 * NAME: lmLogClose()
1459 *
1460 * FUNCTION: remove file system <ipmnt> from active list of log <iplog>
1461 * and close it on last close.
1462 *
1463 * PARAMETER: sb - superblock
1464 *
1465 * RETURN: errors from subroutines
1466 *
1467 * serialization:
1468 */
1469 int lmLogClose(struct super_block *sb)
1470 {
1471 struct jfs_sb_info *sbi = JFS_SBI(sb);
1472 struct jfs_log *log = sbi->log;
1473 struct block_device *bdev;
1474 int rc = 0;
1475
1476 jfs_info("lmLogClose: log:0x%p", log);
1477
1478 down(&jfs_log_sem);
1479 LOG_LOCK(log);
1480 list_del(&sbi->log_list);
1481 LOG_UNLOCK(log);
1482 sbi->log = NULL;
1483
1484 /*
1485 * We need to make sure all of the "written" metapages
1486 * actually make it to disk
1487 */
1488 sync_blockdev(sb->s_bdev);
1489
1490 if (test_bit(log_INLINELOG, &log->flag)) {
1491 /*
1492 * in-line log in host file system
1493 */
1494 rc = lmLogShutdown(log);
1495 kfree(log);
1496 goto out;
1497 }
1498
1499 if (!log->no_integrity)
1500 lmLogFileSystem(log, sbi, 0);
1501
1502 if (!list_empty(&log->sb_list))
1503 goto out;
1504
1505 /*
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
1510 * is requested.
1511 */
1512 if (log->no_integrity)
1513 goto out;
1514
1515 /*
1516 * external log as separate logical volume
1517 */
1518 list_del(&log->journal_list);
1519 bdev = log->bdev;
1520 rc = lmLogShutdown(log);
1521
1522 bd_release(bdev);
1523 blkdev_put(bdev);
1524
1525 kfree(log);
1526
1527 out:
1528 up(&jfs_log_sem);
1529 jfs_info("lmLogClose: exit(%d)", rc);
1530 return rc;
1531 }
1532
1533
1534 /*
1535 * NAME: jfs_flush_journal()
1536 *
1537 * FUNCTION: initiate write of any outstanding transactions to the journal
1538 * and optionally wait until they are all written to disk
1539 *
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
1543 */
1544 void jfs_flush_journal(struct jfs_log *log, int wait)
1545 {
1546 int i;
1547 struct tblock *target = NULL;
1548
1549 /* jfs_write_inode may call us during read-only mount */
1550 if (!log)
1551 return;
1552
1553 jfs_info("jfs_flush_journal: log:0x%p wait=%d", log, wait);
1554
1555 LOGGC_LOCK(log);
1556
1557 if (!list_empty(&log->cqueue)) {
1558 /*
1559 * This ensures that we will keep writing to the journal as long
1560 * as there are unwritten commit records
1561 */
1562 target = list_entry(log->cqueue.prev, struct tblock, cqueue);
1563
1564 if (test_bit(log_FLUSH, &log->flag)) {
1565 /*
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
1570 */
1571 if (log->flush_tblk)
1572 log->flush_tblk = target;
1573 } else {
1574 /* Only flush until latest transaction is committed */
1575 log->flush_tblk = target;
1576 set_bit(log_FLUSH, &log->flag);
1577
1578 /*
1579 * Initiate I/O on outstanding transactions
1580 */
1581 if (!(log->cflag & logGC_PAGEOUT)) {
1582 log->cflag |= logGC_PAGEOUT;
1583 lmGCwrite(log, 0);
1584 }
1585 }
1586 }
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;
1591 }
1592
1593 if (wait && target && !(target->flag & tblkGC_COMMITTED)) {
1594 DECLARE_WAITQUEUE(__wait, current);
1595
1596 add_wait_queue(&target->gcwait, &__wait);
1597 set_current_state(TASK_UNINTERRUPTIBLE);
1598 LOGGC_UNLOCK(log);
1599 schedule();
1600 current->state = TASK_RUNNING;
1601 LOGGC_LOCK(log);
1602 remove_wait_queue(&target->gcwait, &__wait);
1603 }
1604 LOGGC_UNLOCK(log);
1605
1606 if (wait < 2)
1607 return;
1608
1609 /*
1610 * If there was recent activity, we may need to wait
1611 * for the lazycommit thread to catch up
1612 */
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))
1619 break;
1620 }
1621 }
1622 assert(list_empty(&log->cqueue));
1623 assert(list_empty(&log->synclist));
1624 clear_bit(log_FLUSH, &log->flag);
1625 }
1626
1627 /*
1628 * NAME: lmLogShutdown()
1629 *
1630 * FUNCTION: log shutdown at last LogClose().
1631 *
1632 * write log syncpt record.
1633 * update super block to set redone flag to 0.
1634 *
1635 * PARAMETER: log - log inode
1636 *
1637 * RETURN: 0 - success
1638 *
1639 * serialization: single last close thread
1640 */
1641 int lmLogShutdown(struct jfs_log * log)
1642 {
1643 int rc;
1644 struct lrd lrd;
1645 int lsn;
1646 struct logsuper *logsuper;
1647 struct lbuf *bpsuper;
1648 struct lbuf *bp;
1649 struct logpage *lp;
1650
1651 jfs_info("lmLogShutdown: log:0x%p", log);
1652
1653 jfs_flush_journal(log, 2);
1654
1655 /*
1656 * write the last SYNCPT record with syncpoint = 0
1657 * (i.e., log redo up to HERE !)
1658 */
1659 lrd.logtid = 0;
1660 lrd.backchain = 0;
1661 lrd.type = cpu_to_le16(LOG_SYNCPT);
1662 lrd.length = 0;
1663 lrd.log.syncpt.sync = 0;
1664
1665 lsn = lmWriteRecord(log, NULL, &lrd, NULL);
1666 bp = log->bp;
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);
1671
1672 /*
1673 * synchronous update log superblock
1674 * mark log state as shutdown cleanly
1675 * (i.e., Log does not need to be replayed).
1676 */
1677 if ((rc = lbmRead(log, 1, &bpsuper)))
1678 goto out;
1679
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);
1685
1686 jfs_info("lmLogShutdown: lsn:0x%x page:%d eor:%d",
1687 lsn, log->page, log->eor);
1688
1689 out:
1690 /*
1691 * shutdown per log i/o
1692 */
1693 lbmLogShutdown(log);
1694
1695 if (rc) {
1696 jfs_warn("lmLogShutdown: exit(%d)", rc);
1697 }
1698 return rc;
1699 }
1700
1701
1702 /*
1703 * NAME: lmLogFileSystem()
1704 *
1705 * FUNCTION: insert (<activate> = true)/remove (<activate> = false)
1706 * file system into/from log active file system list.
1707 *
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.
1712 *
1713 * RETURN: 0 - success
1714 * errors returned by vms_iowait().
1715 */
1716 static int lmLogFileSystem(struct jfs_log * log, struct jfs_sb_info *sbi,
1717 int activate)
1718 {
1719 int rc = 0;
1720 int i;
1721 struct logsuper *logsuper;
1722 struct lbuf *bpsuper;
1723 char *uuid = sbi->uuid;
1724
1725 /*
1726 * insert/remove file system device to log active file system list.
1727 */
1728 if ((rc = lbmRead(log, 1, &bpsuper)))
1729 return rc;
1730
1731 logsuper = (struct logsuper *) bpsuper->l_ldata;
1732 if (activate) {
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);
1736 sbi->aggregate = i;
1737 break;
1738 }
1739 if (i == MAX_ACTIVE) {
1740 jfs_warn("Too many file systems sharing journal!");
1741 lbmFree(bpsuper);
1742 return -EMFILE; /* Is there a better rc? */
1743 }
1744 } else {
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);
1748 break;
1749 }
1750 if (i == MAX_ACTIVE) {
1751 jfs_warn("Somebody stomped on the journal!");
1752 lbmFree(bpsuper);
1753 return -EIO;
1754 }
1755
1756 }
1757
1758 /*
1759 * synchronous write log superblock:
1760 *
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.
1768 */
1769 lbmDirectWrite(log, bpsuper, lbmWRITE | lbmRELEASE | lbmSYNC);
1770 rc = lbmIOWait(bpsuper, lbmFREE);
1771
1772 return rc;
1773 }
1774
1775 /*
1776 * log buffer manager (lbm)
1777 * ------------------------
1778 *
1779 * special purpose buffer manager supporting log i/o requirements.
1780 *
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.
1788 */
1789
1790 /*
1791 * lbmLogInit()
1792 *
1793 * initialize per log I/O setup at lmLogInit()
1794 */
1795 static int lbmLogInit(struct jfs_log * log)
1796 { /* log inode */
1797 int i;
1798 struct lbuf *lbuf;
1799
1800 jfs_info("lbmLogInit: log:0x%p", log);
1801
1802 /* initialize current buffer cursor */
1803 log->bp = NULL;
1804
1805 /* initialize log device write queue */
1806 log->wqueue = NULL;
1807
1808 /*
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.
1816 */
1817 init_waitqueue_head(&log->free_wait);
1818
1819 log->lbuf_free = NULL;
1820
1821 for (i = 0; i < LOGPAGES; i++) {
1822 lbuf = kmalloc(sizeof(struct lbuf), GFP_KERNEL);
1823 if (lbuf == 0)
1824 goto error;
1825 lbuf->l_ldata = (char *) get_zeroed_page(GFP_KERNEL);
1826 if (lbuf->l_ldata == 0) {
1827 kfree(lbuf);
1828 goto error;
1829 }
1830 lbuf->l_log = log;
1831 init_waitqueue_head(&lbuf->l_ioevent);
1832
1833 lbuf->l_freelist = log->lbuf_free;
1834 log->lbuf_free = lbuf;
1835 }
1836
1837 return (0);
1838
1839 error:
1840 lbmLogShutdown(log);
1841 return -ENOMEM;
1842 }
1843
1844
1845 /*
1846 * lbmLogShutdown()
1847 *
1848 * finalize per log I/O setup at lmLogShutdown()
1849 */
1850 static void lbmLogShutdown(struct jfs_log * log)
1851 {
1852 struct lbuf *lbuf;
1853
1854 jfs_info("lbmLogShutdown: log:0x%p", log);
1855
1856 lbuf = log->lbuf_free;
1857 while (lbuf) {
1858 struct lbuf *next = lbuf->l_freelist;
1859 free_page((unsigned long) lbuf->l_ldata);
1860 kfree(lbuf);
1861 lbuf = next;
1862 }
1863
1864 log->bp = NULL;
1865 }
1866
1867
1868 /*
1869 * lbmAllocate()
1870 *
1871 * allocate an empty log buffer
1872 */
1873 static struct lbuf *lbmAllocate(struct jfs_log * log, int pn)
1874 {
1875 struct lbuf *bp;
1876 unsigned long flags;
1877
1878 /*
1879 * recycle from log buffer freelist if any
1880 */
1881 LCACHE_LOCK(flags);
1882 LCACHE_SLEEP_COND(log->free_wait, (bp = log->lbuf_free), flags);
1883 log->lbuf_free = bp->l_freelist;
1884 LCACHE_UNLOCK(flags);
1885
1886 bp->l_flag = 0;
1887
1888 bp->l_wqnext = NULL;
1889 bp->l_freelist = NULL;
1890
1891 bp->l_pn = pn;
1892 bp->l_blkno = log->base + (pn << (L2LOGPSIZE - log->l2bsize));
1893 bp->l_ceor = 0;
1894
1895 return bp;
1896 }
1897
1898
1899 /*
1900 * lbmFree()
1901 *
1902 * release a log buffer to freelist
1903 */
1904 static void lbmFree(struct lbuf * bp)
1905 {
1906 unsigned long flags;
1907
1908 LCACHE_LOCK(flags);
1909
1910 lbmfree(bp);
1911
1912 LCACHE_UNLOCK(flags);
1913 }
1914
1915 static void lbmfree(struct lbuf * bp)
1916 {
1917 struct jfs_log *log = bp->l_log;
1918
1919 assert(bp->l_wqnext == NULL);
1920
1921 /*
1922 * return the buffer to head of freelist
1923 */
1924 bp->l_freelist = log->lbuf_free;
1925 log->lbuf_free = bp;
1926
1927 wake_up(&log->free_wait);
1928 return;
1929 }
1930
1931
1932 /*
1933 * NAME: lbmRedrive
1934 *
1935 * FUNCTION: add a log buffer to the the log redrive list
1936 *
1937 * PARAMETER:
1938 * bp - log buffer
1939 *
1940 * NOTES:
1941 * Takes log_redrive_lock.
1942 */
1943 static inline void lbmRedrive(struct lbuf *bp)
1944 {
1945 unsigned long flags;
1946
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);
1951
1952 wake_up(&jfs_IO_thread_wait);
1953 }
1954
1955
1956 /*
1957 * lbmRead()
1958 */
1959 static int lbmRead(struct jfs_log * log, int pn, struct lbuf ** bpp)
1960 {
1961 struct bio *bio;
1962 struct lbuf *bp;
1963
1964 /*
1965 * allocate a log buffer
1966 */
1967 *bpp = bp = lbmAllocate(log, pn);
1968 jfs_info("lbmRead: bp:0x%p pn:0x%x", bp, pn);
1969
1970 bp->l_flag |= lbmREAD;
1971
1972 bio = bio_alloc(GFP_NOFS, 1);
1973
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;
1979
1980 bio->bi_vcnt = 1;
1981 bio->bi_idx = 0;
1982 bio->bi_size = LOGPSIZE;
1983
1984 bio->bi_end_io = lbmIODone;
1985 bio->bi_private = bp;
1986 submit_bio(READ_SYNC, bio);
1987
1988 wait_event(bp->l_ioevent, (bp->l_flag != lbmREAD));
1989
1990 return 0;
1991 }
1992
1993
1994 /*
1995 * lbmWrite()
1996 *
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
2000 * released.
2001 *
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.
2005 *
2006 * LOGGC_LOCK() serializes lbmWrite() by lmNextPage() and lmGroupCommit().
2007 * LCACHE_LOCK() serializes xflag between lbmWrite() and lbmIODone()
2008 */
2009 static void lbmWrite(struct jfs_log * log, struct lbuf * bp, int flag,
2010 int cant_block)
2011 {
2012 struct lbuf *tail;
2013 unsigned long flags;
2014
2015 jfs_info("lbmWrite: bp:0x%p flag:0x%x pn:0x%x", bp, flag, bp->l_pn);
2016
2017 /* map the logical block address to physical block address */
2018 bp->l_blkno =
2019 log->base + (bp->l_pn << (L2LOGPSIZE - log->l2bsize));
2020
2021 LCACHE_LOCK(flags); /* disable+lock */
2022
2023 /*
2024 * initialize buffer for device driver
2025 */
2026 bp->l_flag = flag;
2027
2028 /*
2029 * insert bp at tail of write queue associated with log
2030 *
2031 * (request is either for bp already/currently at head of queue
2032 * or new bp to be inserted at tail)
2033 */
2034 tail = log->wqueue;
2035
2036 /* is buffer not already on write queue ? */
2037 if (bp->l_wqnext == NULL) {
2038 /* insert at tail of wqueue */
2039 if (tail == NULL) {
2040 log->wqueue = bp;
2041 bp->l_wqnext = bp;
2042 } else {
2043 log->wqueue = bp;
2044 bp->l_wqnext = tail->l_wqnext;
2045 tail->l_wqnext = bp;
2046 }
2047
2048 tail = bp;
2049 }
2050
2051 /* is buffer at head of wqueue and for write ? */
2052 if ((bp != tail->l_wqnext) || !(flag & lbmWRITE)) {
2053 LCACHE_UNLOCK(flags); /* unlock+enable */
2054 return;
2055 }
2056
2057 LCACHE_UNLOCK(flags); /* unlock+enable */
2058
2059 if (cant_block)
2060 lbmRedrive(bp);
2061 else if (flag & lbmSYNC)
2062 lbmStartIO(bp);
2063 else {
2064 LOGGC_UNLOCK(log);
2065 lbmStartIO(bp);
2066 LOGGC_LOCK(log);
2067 }
2068 }
2069
2070
2071 /*
2072 * lbmDirectWrite()
2073 *
2074 * initiate pageout bypassing write queue for sidestream
2075 * (e.g., log superblock) write;
2076 */
2077 static void lbmDirectWrite(struct jfs_log * log, struct lbuf * bp, int flag)
2078 {
2079 jfs_info("lbmDirectWrite: bp:0x%p flag:0x%x pn:0x%x",
2080 bp, flag, bp->l_pn);
2081
2082 /*
2083 * initialize buffer for device driver
2084 */
2085 bp->l_flag = flag | lbmDIRECT;
2086
2087 /* map the logical block address to physical block address */
2088 bp->l_blkno =
2089 log->base + (bp->l_pn << (L2LOGPSIZE - log->l2bsize));
2090
2091 /*
2092 * initiate pageout of the page
2093 */
2094 lbmStartIO(bp);
2095 }
2096
2097
2098 /*
2099 * NAME: lbmStartIO()
2100 *
2101 * FUNCTION: Interface to DD strategy routine
2102 *
2103 * RETURN: none
2104 *
2105 * serialization: LCACHE_LOCK() is NOT held during log i/o;
2106 */
2107 static void lbmStartIO(struct lbuf * bp)
2108 {
2109 struct bio *bio;
2110 struct jfs_log *log = bp->l_log;
2111
2112 jfs_info("lbmStartIO\n");
2113
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;
2120
2121 bio->bi_vcnt = 1;
2122 bio->bi_idx = 0;
2123 bio->bi_size = LOGPSIZE;
2124
2125 bio->bi_end_io = lbmIODone;
2126 bio->bi_private = bp;
2127
2128 /* check if journaling to disk has been disabled */
2129 if (!log->no_integrity) {
2130 submit_bio(WRITE_SYNC, bio);
2131 INCREMENT(lmStat.submitted);
2132 }
2133 else {
2134 bio->bi_size = 0;
2135 lbmIODone(bio, 0, 0); /* 2nd argument appears to not be used => 0
2136 * 3rd argument appears to not be used => 0
2137 */
2138 }
2139 }
2140
2141
2142 /*
2143 * lbmIOWait()
2144 */
2145 static int lbmIOWait(struct lbuf * bp, int flag)
2146 {
2147 unsigned long flags;
2148 int rc = 0;
2149
2150 jfs_info("lbmIOWait1: bp:0x%p flag:0x%x:0x%x", bp, bp->l_flag, flag);
2151
2152 LCACHE_LOCK(flags); /* disable+lock */
2153
2154 LCACHE_SLEEP_COND(bp->l_ioevent, (bp->l_flag & lbmDONE), flags);
2155
2156 rc = (bp->l_flag & lbmERROR) ? -EIO : 0;
2157
2158 if (flag & lbmFREE)
2159 lbmfree(bp);
2160
2161 LCACHE_UNLOCK(flags); /* unlock+enable */
2162
2163 jfs_info("lbmIOWait2: bp:0x%p flag:0x%x:0x%x", bp, bp->l_flag, flag);
2164 return rc;
2165 }
2166
2167 /*
2168 * lbmIODone()
2169 *
2170 * executed at INTIODONE level
2171 */
2172 static int lbmIODone(struct bio *bio, unsigned int bytes_done, int error)
2173 {
2174 struct lbuf *bp = bio->bi_private;
2175 struct lbuf *nextbp, *tail;
2176 struct jfs_log *log;
2177 unsigned long flags;
2178
2179 if (bio->bi_size)
2180 return 1;
2181
2182 /*
2183 * get back jfs buffer bound to the i/o buffer
2184 */
2185 jfs_info("lbmIODone: bp:0x%p flag:0x%x", bp, bp->l_flag);
2186
2187 LCACHE_LOCK(flags); /* disable+lock */
2188
2189 bp->l_flag |= lbmDONE;
2190
2191 if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
2192 bp->l_flag |= lbmERROR;
2193
2194 jfs_err("lbmIODone: I/O error in JFS log");
2195 }
2196
2197 bio_put(bio);
2198
2199 /*
2200 * pagein completion
2201 */
2202 if (bp->l_flag & lbmREAD) {
2203 bp->l_flag &= ~lbmREAD;
2204
2205 LCACHE_UNLOCK(flags); /* unlock+enable */
2206
2207 /* wakeup I/O initiator */
2208 LCACHE_WAKEUP(&bp->l_ioevent);
2209
2210 return 0;
2211 }
2212
2213 /*
2214 * pageout completion
2215 *
2216 * the bp at the head of write queue has completed pageout.
2217 *
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.
2224 */
2225 bp->l_flag &= ~lbmWRITE;
2226 INCREMENT(lmStat.pagedone);
2227
2228 /* update committed lsn */
2229 log = bp->l_log;
2230 log->clsn = (bp->l_pn << L2LOGPSIZE) + bp->l_ceor;
2231
2232 if (bp->l_flag & lbmDIRECT) {
2233 LCACHE_WAKEUP(&bp->l_ioevent);
2234 LCACHE_UNLOCK(flags);
2235 return 0;
2236 }
2237
2238 tail = log->wqueue;
2239
2240 /* single element queue */
2241 if (bp == tail) {
2242 /* remove head buffer of full-page pageout
2243 * from log device write queue
2244 */
2245 if (bp->l_flag & lbmRELEASE) {
2246 log->wqueue = NULL;
2247 bp->l_wqnext = NULL;
2248 }
2249 }
2250 /* multi element queue */
2251 else {
2252 /* remove head buffer of full-page pageout
2253 * from log device write queue
2254 */
2255 if (bp->l_flag & lbmRELEASE) {
2256 nextbp = tail->l_wqnext = bp->l_wqnext;
2257 bp->l_wqnext = NULL;
2258
2259 /*
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
2267 * by lbmWRITE flag;
2268 */
2269 if (nextbp->l_flag & lbmWRITE) {
2270 /*
2271 * We can't do the I/O at interrupt time.
2272 * The jfsIO thread can do it
2273 */
2274 lbmRedrive(nextbp);
2275 }
2276 }
2277 }
2278
2279 /*
2280 * synchronous pageout:
2281 *
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
2285 */
2286 if (bp->l_flag & lbmSYNC) {
2287 LCACHE_UNLOCK(flags); /* unlock+enable */
2288
2289 /* wakeup I/O initiator */
2290 LCACHE_WAKEUP(&bp->l_ioevent);
2291 }
2292
2293 /*
2294 * Group Commit pageout:
2295 */
2296 else if (bp->l_flag & lbmGC) {
2297 LCACHE_UNLOCK(flags);
2298 lmPostGC(bp);
2299 }
2300
2301 /*
2302 * asynchronous pageout:
2303 *
2304 * buffer must have been removed from write queue:
2305 * insert buffer at head of freelist where it can be recycled
2306 */
2307 else {
2308 assert(bp->l_flag & lbmRELEASE);
2309 assert(bp->l_flag & lbmFREE);
2310 lbmfree(bp);
2311
2312 LCACHE_UNLOCK(flags); /* unlock+enable */
2313 }
2314
2315 return 0;
2316 }
2317
2318 int jfsIOWait(void *arg)
2319 {
2320 struct lbuf *bp;
2321
2322 daemonize("jfsIO");
2323
2324 complete(&jfsIOwait);
2325
2326 do {
2327 DECLARE_WAITQUEUE(wq, current);
2328
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);
2334 lbmStartIO(bp);
2335 spin_lock_irq(&log_redrive_lock);
2336 }
2337 if (current->flags & PF_FREEZE) {
2338 spin_unlock_irq(&log_redrive_lock);
2339 refrigerator(PF_FREEZE);
2340 } else {
2341 add_wait_queue(&jfs_IO_thread_wait, &wq);
2342 set_current_state(TASK_INTERRUPTIBLE);
2343 spin_unlock_irq(&log_redrive_lock);
2344 schedule();
2345 current->state = TASK_RUNNING;
2346 remove_wait_queue(&jfs_IO_thread_wait, &wq);
2347 }
2348 } while (!jfs_stop_threads);
2349
2350 jfs_info("jfsIOWait being killed!");
2351 complete_and_exit(&jfsIOwait, 0);
2352 }
2353
2354 /*
2355 * NAME: lmLogFormat()/jfs_logform()
2356 *
2357 * FUNCTION: format file system log
2358 *
2359 * PARAMETERS:
2360 * log - volume log
2361 * logAddress - start address of log space in FS block
2362 * logSize - length of log space in FS block;
2363 *
2364 * RETURN: 0 - success
2365 * -EIO - i/o error
2366 *
2367 * XXX: We're synchronously writing one page at a time. This needs to
2368 * be improved by writing multiple pages at once.
2369 */
2370 int lmLogFormat(struct jfs_log *log, s64 logAddress, int logSize)
2371 {
2372 int rc = -EIO;
2373 struct jfs_sb_info *sbi;
2374 struct logsuper *logsuper;
2375 struct logpage *lp;
2376 int lspn; /* log sequence page number */
2377 struct lrd *lrd_ptr;
2378 int npages = 0;
2379 struct lbuf *bp;
2380
2381 jfs_info("lmLogFormat: logAddress:%Ld logSize:%d",
2382 (long long)logAddress, logSize);
2383
2384 sbi = list_entry(log->sb_list.next, struct jfs_sb_info, log_list);
2385
2386 /* allocate a log buffer */
2387 bp = lbmAllocate(log, 1);
2388
2389 npages = logSize >> sbi->l2nbperpage;
2390
2391 /*
2392 * log space:
2393 *
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;
2399 */
2400 /*
2401 * init log superblock: log page 1
2402 */
2403 logsuper = (struct logsuper *) bp->l_ldata;
2404
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);
2413
2414 bp->l_flag = lbmWRITE | lbmSYNC | lbmDIRECT;
2415 bp->l_blkno = logAddress + sbi->nbperpage;
2416 lbmStartIO(bp);
2417 if ((rc = lbmIOWait(bp, 0)))
2418 goto exit;
2419
2420 /*
2421 * init pages 2 to npages-1 as log data pages:
2422 *
2423 * log page sequence number (lpsn) initialization:
2424 *
2425 * pn: 0 1 2 3 n-1
2426 * +-----+-----+=====+=====+===.....===+=====+
2427 * lspn: N-1 0 1 N-2
2428 * <--- N page circular file ---->
2429 *
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)
2441 */
2442 lp = (struct logpage *) bp->l_ldata;
2443 /*
2444 * initialize 1st log page to be written: lpsn = N - 1,
2445 * write a SYNCPT log record is written to this page
2446 */
2447 lp->h.page = lp->t.page = cpu_to_le32(npages - 3);
2448 lp->h.eor = lp->t.eor = cpu_to_le16(LOGPHDRSIZE + LOGRDSIZE);
2449
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;
2456
2457 bp->l_blkno += sbi->nbperpage;
2458 bp->l_flag = lbmWRITE | lbmSYNC | lbmDIRECT;
2459 lbmStartIO(bp);
2460 if ((rc = lbmIOWait(bp, 0)))
2461 goto exit;
2462
2463 /*
2464 * initialize succeeding log pages: lpsn = 0, 1, ..., (N-2)
2465 */
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);
2469
2470 bp->l_blkno += sbi->nbperpage;
2471 bp->l_flag = lbmWRITE | lbmSYNC | lbmDIRECT;
2472 lbmStartIO(bp);
2473 if ((rc = lbmIOWait(bp, 0)))
2474 goto exit;
2475 }
2476
2477 rc = 0;
2478 exit:
2479 /*
2480 * finalize log
2481 */
2482 /* release the buffer */
2483 lbmFree(bp);
2484
2485 return rc;
2486 }
2487
2488 #ifdef CONFIG_JFS_STATISTICS
2489 int jfs_lmstats_read(char *buffer, char **start, off_t offset, int length,
2490 int *eof, void *data)
2491 {
2492 int len = 0;
2493 off_t begin;
2494
2495 len += sprintf(buffer,
2496 "JFS Logmgr stats\n"
2497 "================\n"
2498 "commits = %d\n"
2499 "writes submitted = %d\n"
2500 "writes completed = %d\n"
2501 "full pages submitted = %d\n"
2502 "partial pages submitted = %d\n",
2503 lmStat.commit,
2504 lmStat.submitted,
2505 lmStat.pagedone,
2506 lmStat.full_page,
2507 lmStat.partial_page);
2508
2509 begin = offset;
2510 *start = buffer + begin;
2511 len -= begin;
2512
2513 if (len > length)
2514 len = length;
2515 else
2516 *eof = 1;
2517
2518 if (len < 0)
2519 len = 0;
2520
2521 return len;
2522 }
2523 #endif /* CONFIG_JFS_STATISTICS */
2524
|
This page was automatically generated by the
LXR engine.
|