jfs_logmgr.c

jfs-2.4-1.1.7.tar.gz jfs 2.4－1.1.7 源码

	lbmfree(bp);

	LCACHE_UNLOCK(flags);
}

static void lbmfree(struct lbuf * bp)
{
	struct jfs_log *log = bp->l_log;

	assert(bp->l_wqnext == NULL);

	/*
	 * return the buffer to head of freelist
	 */
	bp->l_freelist = log->lbuf_free;
	log->lbuf_free = bp;

	wake_up(&log->free_wait);
	return;
}


/*
 * NAME:	lbmRedrive
 *
 * FUNCTION:	add a log buffer to the the log redrive list
 *
 * PARAMETER:
 *     bp	- log buffer
 *
 * NOTES:
 *	Takes log_redrive_lock.
 */
static inline void lbmRedrive(struct lbuf *bp)
{
	unsigned long flags;

	spin_lock_irqsave(&log_redrive_lock, flags);
	bp->l_redrive_next = log_redrive_list;
	log_redrive_list = bp;
	spin_unlock_irqrestore(&log_redrive_lock, flags);

	wake_up(&jfs_IO_thread_wait);
}


/*
 *	lbmRead()
 */
static int lbmRead(struct jfs_log * log, int pn, struct lbuf ** bpp)
{
	struct lbuf *bp;

	/*
	 * allocate a log buffer
	 */
	*bpp = bp = lbmAllocate(log, pn);
	jfs_info("lbmRead: bp:0x%p pn:0x%x", bp, pn);

	bp->l_flag |= lbmREAD;
	bp->l_bh.b_reqnext = NULL;
	clear_bit(BH_Uptodate, &bp->l_bh.b_state);
	lock_buffer(&bp->l_bh);
	set_bit(BH_Mapped, &bp->l_bh.b_state);
	set_bit(BH_Req, &bp->l_bh.b_state);
	bp->l_bh.b_rdev = bp->l_bh.b_dev;
	bp->l_bh.b_rsector = bp->l_blkno << (log->l2bsize - 9);
	generic_make_request(READ, &bp->l_bh);
	run_task_queue(&tq_disk);

	wait_event(bp->l_ioevent, (bp->l_flag != lbmREAD));

	return 0;
}


/*
 *	lbmWrite()
 *
 * buffer at head of pageout queue stays after completion of
 * partial-page pageout and redriven by explicit initiation of
 * pageout by caller until full-page pageout is completed and
 * released.
 *
 * device driver i/o done redrives pageout of new buffer at
 * head of pageout queue when current buffer at head of pageout
 * queue is released at the completion of its full-page pageout.
 *
 * LOGGC_LOCK() serializes lbmWrite() by lmNextPage() and lmGroupCommit().
 * LCACHE_LOCK() serializes xflag between lbmWrite() and lbmIODone()
 */
static void lbmWrite(struct jfs_log * log, struct lbuf * bp, int flag,
		     int cant_block)
{
	struct lbuf *tail;
	unsigned long flags;

	jfs_info("lbmWrite: bp:0x%p flag:0x%x pn:0x%x", bp, flag, bp->l_pn);

	/* map the logical block address to physical block address */
	bp->l_blkno =
	    log->base + (bp->l_pn << (L2LOGPSIZE - log->l2bsize));

	LCACHE_LOCK(flags);		/* disable+lock */

	/*
	 * initialize buffer for device driver
	 */
	bp->l_flag = flag;

	/*
	 *      insert bp at tail of write queue associated with log
	 *
	 * (request is either for bp already/currently at head of queue
	 * or new bp to be inserted at tail)
	 */
	tail = log->wqueue;

	/* is buffer not already on write queue ? */
	if (bp->l_wqnext == NULL) {
		/* insert at tail of wqueue */
		if (tail == NULL) {
			log->wqueue = bp;
			bp->l_wqnext = bp;
		} else {
			log->wqueue = bp;
			bp->l_wqnext = tail->l_wqnext;
			tail->l_wqnext = bp;
		}

		tail = bp;
	}

	/* is buffer at head of wqueue and for write ? */
	if ((bp != tail->l_wqnext) || !(flag & lbmWRITE)) {
		LCACHE_UNLOCK(flags);	/* unlock+enable */
		return;
	}

	LCACHE_UNLOCK(flags);	/* unlock+enable */

	if (cant_block)
		lbmRedrive(bp);
	else if (flag & lbmSYNC)
		lbmStartIO(bp);
	else {
		LOGGC_UNLOCK(log);
		lbmStartIO(bp);
		LOGGC_LOCK(log);
	}
}


/*
 *	lbmDirectWrite()
 *
 * initiate pageout bypassing write queue for sidestream
 * (e.g., log superblock) write;
 */
static void lbmDirectWrite(struct jfs_log * log, struct lbuf * bp, int flag)
{
	jfs_info("lbmDirectWrite: bp:0x%p flag:0x%x pn:0x%x",
		 bp, flag, bp->l_pn);

	/*
	 * initialize buffer for device driver
	 */
	bp->l_flag = flag | lbmDIRECT;

	/* map the logical block address to physical block address */
	bp->l_blkno =
	    log->base + (bp->l_pn << (L2LOGPSIZE - log->l2bsize));

	/*
	 *      initiate pageout of the page
	 */
	lbmStartIO(bp);
}


/*
 * NAME:	lbmStartIO()
 *
 * FUNCTION:	Interface to DD strategy routine
 *
 * RETURN:      none
 *
 * serialization: LCACHE_LOCK() is NOT held during log i/o;
 */
static void lbmStartIO(struct lbuf * bp)
{
	jfs_info("lbmStartIO");

//	lock_buffer(&bp->l_bh);
	assert(!test_bit(BH_Lock, &bp->l_bh.b_state));
	set_bit(BH_Lock, &bp->l_bh.b_state);

	if (bp->l_log->no_integrity)
		/* Don't really do I/O */
		lbmIODone(&bp->l_bh, 1);
	else {
		bp->l_bh.b_reqnext = NULL;
		set_bit(BH_Dirty, &bp->l_bh.b_state);
		set_bit(BH_Mapped, &bp->l_bh.b_state);
		set_bit(BH_Req, &bp->l_bh.b_state);
		bp->l_bh.b_rdev = bp->l_bh.b_dev;
		bp->l_bh.b_rsector = bp->l_blkno << (bp->l_log->l2bsize - 9);
		generic_make_request(WRITE, &bp->l_bh);

		INCREMENT(lmStat.submitted);
		run_task_queue(&tq_disk);
	}
}


/*
 *	lbmIOWait()
 */
static int lbmIOWait(struct lbuf * bp, int flag)
{
	unsigned long flags;
	int rc = 0;

	jfs_info("lbmIOWait1: bp:0x%p flag:0x%x:0x%x", bp, bp->l_flag, flag);

	LCACHE_LOCK(flags);		/* disable+lock */

	LCACHE_SLEEP_COND(bp->l_ioevent, (bp->l_flag & lbmDONE), flags);

	rc = (bp->l_flag & lbmERROR) ? -EIO : 0;

	if (flag & lbmFREE)
		lbmfree(bp);

	LCACHE_UNLOCK(flags);	/* unlock+enable */

	jfs_info("lbmIOWait2: bp:0x%p flag:0x%x:0x%x", bp, bp->l_flag, flag);
	return rc;
}

/*
 *	lbmIODone()
 *
 * executed at INTIODONE level
 */
static void lbmIODone(struct buffer_head *bh, int uptodate)
{
	struct lbuf *bp = bh->b_private;
	struct lbuf *nextbp, *tail;
	struct jfs_log *log;
	unsigned long flags;

	/*
	 * get back jfs buffer bound to the i/o buffer
	 */
	jfs_info("lbmIODone: bp:0x%p flag:0x%x", bp, bp->l_flag);

	LCACHE_LOCK(flags);		/* disable+lock */

	unlock_buffer(&bp->l_bh);
	bp->l_flag |= lbmDONE;

	if (!uptodate) {
		bp->l_flag |= lbmERROR;

		jfs_err("lbmIODone: I/O error in JFS log");
	}

	/*
	 *      pagein completion
	 */
	if (bp->l_flag & lbmREAD) {
		bp->l_flag &= ~lbmREAD;

		LCACHE_UNLOCK(flags);	/* unlock+enable */

		/* wakeup I/O initiator */
		LCACHE_WAKEUP(&bp->l_ioevent);

		return;
	}

	/*
	 *      pageout completion
	 *
	 * the bp at the head of write queue has completed pageout.
	 *
	 * if single-commit/full-page pageout, remove the current buffer
	 * from head of pageout queue, and redrive pageout with
	 * the new buffer at head of pageout queue;
	 * otherwise, the partial-page pageout buffer stays at
	 * the head of pageout queue to be redriven for pageout
	 * by lmGroupCommit() until full-page pageout is completed.
	 */
	bp->l_flag &= ~lbmWRITE;
	INCREMENT(lmStat.pagedone);

	/* update committed lsn */
	log = bp->l_log;
	log->clsn = (bp->l_pn << L2LOGPSIZE) + bp->l_ceor;

	if (bp->l_flag & lbmDIRECT) {
		LCACHE_WAKEUP(&bp->l_ioevent);
		LCACHE_UNLOCK(flags);
		return;
	}

	tail = log->wqueue;

	/* single element queue */
	if (bp == tail) {
		/* remove head buffer of full-page pageout
		 * from log device write queue
		 */
		if (bp->l_flag & lbmRELEASE) {
			log->wqueue = NULL;
			bp->l_wqnext = NULL;
		}
	}
	/* multi element queue */
	else {
		/* remove head buffer of full-page pageout
		 * from log device write queue
		 */
		if (bp->l_flag & lbmRELEASE) {
			nextbp = tail->l_wqnext = bp->l_wqnext;
			bp->l_wqnext = NULL;

			/*
			 * redrive pageout of next page at head of write queue:
			 * redrive next page without any bound tblk
			 * (i.e., page w/o any COMMIT records), or
			 * first page of new group commit which has been
			 * queued after current page (subsequent pageout
			 * is performed synchronously, except page without
			 * any COMMITs) by lmGroupCommit() as indicated
			 * by lbmWRITE flag;
			 */
			if (nextbp->l_flag & lbmWRITE) {
				/*
				 * We can't do the I/O at interrupt time.
				 * The jfsIO thread can do it
				 */
				lbmRedrive(nextbp);
			}
		}
	}

	/*
	 *      synchronous pageout:
	 *
	 * buffer has not necessarily been removed from write queue
	 * (e.g., synchronous write of partial-page with COMMIT):
	 * leave buffer for i/o initiator to dispose
	 */
	if (bp->l_flag & lbmSYNC) {
		LCACHE_UNLOCK(flags);	/* unlock+enable */

		/* wakeup I/O initiator */
		LCACHE_WAKEUP(&bp->l_ioevent);
	}

	/*
	 *      Group Commit pageout:
	 */
	else if (bp->l_flag & lbmGC) {
		LCACHE_UNLOCK(flags);
		lmPostGC(bp);
	}

	/*
	 *      asynchronous pageout:
	 *
	 * buffer must have been removed from write queue:
	 * insert buffer at head of freelist where it can be recycled
	 */
	else {
		assert(bp->l_flag & lbmRELEASE);
		assert(bp->l_flag & lbmFREE);
		lbmfree(bp);

		LCACHE_UNLOCK(flags);	/* unlock+enable */
	}
}

int jfsIOWait(void *arg)
{
	struct lbuf *bp;

	lock_kernel();

	daemonize();
	current->tty = NULL;
	strcpy(current->comm, "jfsIO");

	unlock_kernel();

	spin_lock_irq(&current->Sigmask_lock);
	sigfillset(&current->blocked);
	Recalc_sigpending(current);
	spin_unlock_irq(&current->Sigmask_lock);

	complete(&jfsIOwait);

	do {
		DECLARE_WAITQUEUE(wq, current);

		spin_lock_irq(&log_redrive_lock);
		while ((bp = log_redrive_list)) {
			log_redrive_list = bp->l_redrive_next;
			bp->l_redrive_next = NULL;
			spin_unlock_irq(&log_redrive_lock);
			lbmStartIO(bp);
			spin_lock_irq(&log_redrive_lock);
		}
		add_wait_queue(&jfs_IO_thread_wait, &wq);
		set_current_state(TASK_INTERRUPTIBLE);
		spin_unlock_irq(&log_redrive_lock);
		schedule();
		current->state = TASK_RUNNING;
		remove_wait_queue(&jfs_IO_thread_wait, &wq);
	} while (!jfs_stop_threads);

	jfs_info("jfsIOWait being killed!");
	complete(&jfsIOwait);
	return 0;
}

/*
 * NAME:	lmLogFormat()/jfs_logform()
 *
 * FUNCTION:	format file system log
 *
 * PARAMETERS:
 *      log	- volume log
 *	logAddress - start address of log space in FS block
 *	logSize	- length of log space in FS block;
 *
 * RETURN:	0	- success
 *		-EIO	- i/o error
 *
 * XXX: We're synchronously writing one page at a time.  This needs to
 *	be improved by writing multiple pages at once.
 */
int lmLogFormat(struct jfs_log *log, s64 logAddress, int logSize)
{
	int rc = -EIO;
	struct jfs_sb_info *sbi = JFS_SBI(log->sb);
	struct logsuper *logsuper;
	struct logpage *lp;
	int lspn;		/* log sequence page number */
	struct lrd *lrd_ptr;
	int npages = 0;
	struct lbuf *bp;

	jfs_info("lmLogFormat: logAddress:%Ld logSize:%d",
		 (long long)logAddress, logSize);

	/* allocate a log buffer */
	bp = lbmAllocate(log, 1);

	npages = logSize >> sbi->l2nbperpage;

	/*
	 *      log space:
	 *
	 * page 0 - reserved;
	 * page 1 - log superblock;
	 * page 2 - log data page: A SYNC log record is written
	 *          into this page at logform time;
	 * pages 3-N - log data page: set to empty log data pages;
	 */
	/*
	 *      init log superblock: log page 1
	 */
	logsuper = (struct logsuper *) bp->l_ldata;

	logsuper->magic = cpu_to_le32(LOGMAGIC);
	logsuper->version = cpu_to_le32(LOGVERSION);
	logsuper->state = cpu_to_le32(LOGREDONE);
	logsuper->flag = cpu_to_le32(sbi->mntflag);	/* ? */
	logsuper->size = cpu_to_le32(npages);
	logsuper->bsize = cpu_to_le32(sbi->bsize);
	logsuper->l2bsize = cpu_to_le32(sbi->l2bsize);
	logsuper->end = cpu_to_le32(2 * LOGPSIZE + LOGPHDRSIZE + LOGRDSIZE);

	bp->l_flag = lbmWRITE | lbmSYNC | lbmDIRECT;
	bp->l_blkno = logAddress + sbi->nbperpage;
	lbmStartIO(bp);
	if ((rc = lbmIOWait(bp, 0)))
		goto exit;

	/*
	 *      init pages 2 to npages-1 as log data pages:
	 *
	 * log page sequence number (lpsn) initialization:
	 *
	 * pn:   0     1     2     3                 n-1
	 *       +-----+-----+=====+=====+===.....===+=====+
	 * lspn:             N-1   0     1           N-2
	 *                   <--- N page circular file ---->
	 *
	 * the N (= npages-2) data pages of the log is maintained as
	 * a circular file for the log records;
	 * lpsn grows by 1 monotonically as each log page is written
	 * to the circular file of the log;
	 * and setLogpage() will not reset the page number even if
	 * the eor is equal to LOGPHDRSIZE. In order for binary search
	 * still work in find log end process, we have to simulate the
	 * log wrap situation at the log format time.
	 * The 1st log page written will have the highest lpsn. Then
	 * the succeeding log pages will have ascending order of
	 * the lspn starting from 0, ... (N-2)
	 */
	lp = (struct logpage *) bp->l_ldata;
	/*
	 * initialize 1st log page to be written: lpsn = N - 1,
	 * write a SYNCPT log record is written to this page
	 */
	lp->h.page = lp->t.page = cpu_to_le32(npages - 3);
	lp->h.eor = lp->t.eor = cpu_to_le16(LOGPHDRSIZE + LOGRDSIZE);

	lrd_ptr = (struct lrd *) &lp->data;
	lrd_ptr->logtid = 0;
	lrd_ptr->backchain = 0;
	lrd_ptr->type = cpu_to_le16(LOG_SYNCPT);
	lrd_ptr->length = 0;
	lrd_ptr->log.syncpt.sync = 0;

	bp->l_blkno += sbi->nbperpage;
	bp->l_flag = lbmWRITE | lbmSYNC | lbmDIRECT;
	lbmStartIO(bp);
	if ((rc = lbmIOWait(bp, 0)))
		goto exit;

	/*
	 *      initialize succeeding log pages: lpsn = 0, 1, ..., (N-2)
	 */
	for (lspn = 0; lspn < npages - 3; lspn++) {
		lp->h.page = lp->t.page = cpu_to_le32(lspn);
		lp->h.eor = lp->t.eor = cpu_to_le16(LOGPHDRSIZE);

		bp->l_blkno += sbi->nbperpage;
		bp->l_flag = lbmWRITE | lbmSYNC | lbmDIRECT;
		lbmStartIO(bp);
		if ((rc = lbmIOWait(bp, 0)))
			goto exit;
	}

	rc = 0;
exit:
	/*
	 *      finalize log
	 */
	/* release the buffer */
	lbmFree(bp);

	return rc;
}

#ifdef CONFIG_JFS_STATISTICS
int jfs_lmstats_read(char *buffer, char **start, off_t offset, int length,
		      int *eof, void *data)
{
	int len = 0;
	off_t begin;

	len += sprintf(buffer,
		       "JFS Logmgr stats\n"
		       "================\n"
		       "commits = %d\n"
		       "writes submitted = %d\n"
		       "writes completed = %d\n"
		       "full pages submitted = %d\n"
		       "partial pages submitted = %d\n",
		       lmStat.commit,
		       lmStat.submitted,
		       lmStat.pagedone,
		       lmStat.full_page,
		       lmStat.partial_page);

	begin = offset;
	*start = buffer + begin;
	len -= begin;

	if (len > length)
		len = length;
	else
		*eof = 1;

	if (len < 0)
		len = 0;

	return len;
}
#endif /* CONFIG_JFS_STATISTICS */