jfs_txnmgr.c

linux 内核源代码

 * function:	log from maplock of freed data extents;
 */
static void mapLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd,
		   struct tlock * tlck)
{
	struct pxd_lock *pxdlock;
	int i, nlock;
	pxd_t *pxd;

	/*
	 *	page relocation: free the source page extent
	 *
	 * a maplock for txUpdateMap() for free of the page
	 * has been formatted at txLock() time saving the src
	 * relocated page address;
	 */
	if (tlck->type & tlckRELOCATE) {
		/* log LOG_NOREDOPAGE of the old relocated page
		 * for logredo() to start NoRedoPage filter;
		 */
		lrd->type = cpu_to_le16(LOG_NOREDOPAGE);
		pxdlock = (struct pxd_lock *) & tlck->lock;
		pxd = &lrd->log.redopage.pxd;
		*pxd = pxdlock->pxd;
		lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));

		/* (N.B. currently, logredo() does NOT update bmap
		 * for free of the page itself for (LOG_XTREE|LOG_NOREDOPAGE);
		 * if page free from relocation, LOG_UPDATEMAP log is
		 * specifically generated now for logredo()
		 * to update bmap for free of src relocated page;
		 * (new flag LOG_RELOCATE may be introduced which will
		 * inform logredo() to start NORedoPage filter and also
		 * update block allocation map at the same time, thus
		 * avoiding an extra log write);
		 */
		lrd->type = cpu_to_le16(LOG_UPDATEMAP);
		lrd->log.updatemap.type = cpu_to_le16(LOG_FREEPXD);
		lrd->log.updatemap.nxd = cpu_to_le16(1);
		lrd->log.updatemap.pxd = pxdlock->pxd;
		lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL));

		/* a maplock for txUpdateMap() for free of the page
		 * has been formatted at txLock() time;
		 */
		tlck->flag |= tlckUPDATEMAP;
		return;
	}
	/*

	 * Otherwise it's not a relocate request
	 *
	 */
	else {
		/* log LOG_UPDATEMAP for logredo() to update bmap for
		 * free of truncated/relocated delta extent of the data;
		 * e.g.: external EA extent, relocated/truncated extent
		 * from xtTailgate();
		 */
		lrd->type = cpu_to_le16(LOG_UPDATEMAP);
		pxdlock = (struct pxd_lock *) & tlck->lock;
		nlock = pxdlock->index;
		for (i = 0; i < nlock; i++, pxdlock++) {
			if (pxdlock->flag & mlckALLOCPXD)
				lrd->log.updatemap.type =
				    cpu_to_le16(LOG_ALLOCPXD);
			else
				lrd->log.updatemap.type =
				    cpu_to_le16(LOG_FREEPXD);
			lrd->log.updatemap.nxd = cpu_to_le16(1);
			lrd->log.updatemap.pxd = pxdlock->pxd;
			lrd->backchain =
			    cpu_to_le32(lmLog(log, tblk, lrd, NULL));
			jfs_info("mapLog: xaddr:0x%lx xlen:0x%x",
				 (ulong) addressPXD(&pxdlock->pxd),
				 lengthPXD(&pxdlock->pxd));
		}

		/* update bmap */
		tlck->flag |= tlckUPDATEMAP;
	}
}

/*
 *	txEA()
 *
 * function:	acquire maplock for EA/ACL extents or
 *		set COMMIT_INLINE flag;
 */
void txEA(tid_t tid, struct inode *ip, dxd_t * oldea, dxd_t * newea)
{
	struct tlock *tlck = NULL;
	struct pxd_lock *maplock = NULL, *pxdlock = NULL;

	/*
	 * format maplock for alloc of new EA extent
	 */
	if (newea) {
		/* Since the newea could be a completely zeroed entry we need to
		 * check for the two flags which indicate we should actually
		 * commit new EA data
		 */
		if (newea->flag & DXD_EXTENT) {
			tlck = txMaplock(tid, ip, tlckMAP);
			maplock = (struct pxd_lock *) & tlck->lock;
			pxdlock = (struct pxd_lock *) maplock;
			pxdlock->flag = mlckALLOCPXD;
			PXDaddress(&pxdlock->pxd, addressDXD(newea));
			PXDlength(&pxdlock->pxd, lengthDXD(newea));
			pxdlock++;
			maplock->index = 1;
		} else if (newea->flag & DXD_INLINE) {
			tlck = NULL;

			set_cflag(COMMIT_Inlineea, ip);
		}
	}

	/*
	 * format maplock for free of old EA extent
	 */
	if (!test_cflag(COMMIT_Nolink, ip) && oldea->flag & DXD_EXTENT) {
		if (tlck == NULL) {
			tlck = txMaplock(tid, ip, tlckMAP);
			maplock = (struct pxd_lock *) & tlck->lock;
			pxdlock = (struct pxd_lock *) maplock;
			maplock->index = 0;
		}
		pxdlock->flag = mlckFREEPXD;
		PXDaddress(&pxdlock->pxd, addressDXD(oldea));
		PXDlength(&pxdlock->pxd, lengthDXD(oldea));
		maplock->index++;
	}
}

/*
 *	txForce()
 *
 * function: synchronously write pages locked by transaction
 *	     after txLog() but before txUpdateMap();
 */
static void txForce(struct tblock * tblk)
{
	struct tlock *tlck;
	lid_t lid, next;
	struct metapage *mp;

	/*
	 * reverse the order of transaction tlocks in
	 * careful update order of address index pages
	 * (right to left, bottom up)
	 */
	tlck = lid_to_tlock(tblk->next);
	lid = tlck->next;
	tlck->next = 0;
	while (lid) {
		tlck = lid_to_tlock(lid);
		next = tlck->next;
		tlck->next = tblk->next;
		tblk->next = lid;
		lid = next;
	}

	/*
	 * synchronously write the page, and
	 * hold the page for txUpdateMap();
	 */
	for (lid = tblk->next; lid; lid = next) {
		tlck = lid_to_tlock(lid);
		next = tlck->next;

		if ((mp = tlck->mp) != NULL &&
		    (tlck->type & tlckBTROOT) == 0) {
			assert(mp->xflag & COMMIT_PAGE);

			if (tlck->flag & tlckWRITEPAGE) {
				tlck->flag &= ~tlckWRITEPAGE;

				/* do not release page to freelist */
				force_metapage(mp);
#if 0
				/*
				 * The "right" thing to do here is to
				 * synchronously write the metadata.
				 * With the current implementation this
				 * is hard since write_metapage requires
				 * us to kunmap & remap the page.  If we
				 * have tlocks pointing into the metadata
				 * pages, we don't want to do this.  I think
				 * we can get by with synchronously writing
				 * the pages when they are released.
				 */
				assert(mp->nohomeok);
				set_bit(META_dirty, &mp->flag);
				set_bit(META_sync, &mp->flag);
#endif
			}
		}
	}
}

/*
 *	txUpdateMap()
 *
 * function:	update persistent allocation map (and working map
 *		if appropriate);
 *
 * parameter:
 */
static void txUpdateMap(struct tblock * tblk)
{
	struct inode *ip;
	struct inode *ipimap;
	lid_t lid;
	struct tlock *tlck;
	struct maplock *maplock;
	struct pxd_lock pxdlock;
	int maptype;
	int k, nlock;
	struct metapage *mp = NULL;

	ipimap = JFS_SBI(tblk->sb)->ipimap;

	maptype = (tblk->xflag & COMMIT_PMAP) ? COMMIT_PMAP : COMMIT_PWMAP;


	/*
	 *	update block allocation map
	 *
	 * update allocation state in pmap (and wmap) and
	 * update lsn of the pmap page;
	 */
	/*
	 * scan each tlock/page of transaction for block allocation/free:
	 *
	 * for each tlock/page of transaction, update map.
	 *  ? are there tlock for pmap and pwmap at the same time ?
	 */
	for (lid = tblk->next; lid; lid = tlck->next) {
		tlck = lid_to_tlock(lid);

		if ((tlck->flag & tlckUPDATEMAP) == 0)
			continue;

		if (tlck->flag & tlckFREEPAGE) {
			/*
			 * Another thread may attempt to reuse freed space
			 * immediately, so we want to get rid of the metapage
			 * before anyone else has a chance to get it.
			 * Lock metapage, update maps, then invalidate
			 * the metapage.
			 */
			mp = tlck->mp;
			ASSERT(mp->xflag & COMMIT_PAGE);
			grab_metapage(mp);
		}

		/*
		 * extent list:
		 * . in-line PXD list:
		 * . out-of-line XAD list:
		 */
		maplock = (struct maplock *) & tlck->lock;
		nlock = maplock->index;

		for (k = 0; k < nlock; k++, maplock++) {
			/*
			 * allocate blocks in persistent map:
			 *
			 * blocks have been allocated from wmap at alloc time;
			 */
			if (maplock->flag & mlckALLOC) {
				txAllocPMap(ipimap, maplock, tblk);
			}
			/*
			 * free blocks in persistent and working map:
			 * blocks will be freed in pmap and then in wmap;
			 *
			 * ? tblock specifies the PMAP/PWMAP based upon
			 * transaction
			 *
			 * free blocks in persistent map:
			 * blocks will be freed from wmap at last reference
			 * release of the object for regular files;
			 *
			 * Alway free blocks from both persistent & working
			 * maps for directories
			 */
			else {	/* (maplock->flag & mlckFREE) */

				if (tlck->flag & tlckDIRECTORY)
					txFreeMap(ipimap, maplock,
						  tblk, COMMIT_PWMAP);
				else
					txFreeMap(ipimap, maplock,
						  tblk, maptype);
			}
		}
		if (tlck->flag & tlckFREEPAGE) {
			if (!(tblk->flag & tblkGC_LAZY)) {
				/* This is equivalent to txRelease */
				ASSERT(mp->lid == lid);
				tlck->mp->lid = 0;
			}
			assert(mp->nohomeok == 1);
			metapage_homeok(mp);
			discard_metapage(mp);
			tlck->mp = NULL;
		}
	}
	/*
	 *	update inode allocation map
	 *
	 * update allocation state in pmap and
	 * update lsn of the pmap page;
	 * update in-memory inode flag/state
	 *
	 * unlock mapper/write lock
	 */
	if (tblk->xflag & COMMIT_CREATE) {
		diUpdatePMap(ipimap, tblk->ino, false, tblk);
		/* update persistent block allocation map
		 * for the allocation of inode extent;
		 */
		pxdlock.flag = mlckALLOCPXD;
		pxdlock.pxd = tblk->u.ixpxd;
		pxdlock.index = 1;
		txAllocPMap(ipimap, (struct maplock *) & pxdlock, tblk);
	} else if (tblk->xflag & COMMIT_DELETE) {
		ip = tblk->u.ip;
		diUpdatePMap(ipimap, ip->i_ino, true, tblk);
		iput(ip);
	}
}

/*
 *	txAllocPMap()
 *
 * function: allocate from persistent map;
 *
 * parameter:
 *	ipbmap	-
 *	malock	-
 *		xad list:
 *		pxd:
 *
 *	maptype -
 *		allocate from persistent map;
 *		free from persistent map;
 *		(e.g., tmp file - free from working map at releae
 *		 of last reference);
 *		free from persistent and working map;
 *
 *	lsn	- log sequence number;
 */
static void txAllocPMap(struct inode *ip, struct maplock * maplock,
			struct tblock * tblk)
{
	struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
	struct xdlistlock *xadlistlock;
	xad_t *xad;
	s64 xaddr;
	int xlen;
	struct pxd_lock *pxdlock;
	struct xdlistlock *pxdlistlock;
	pxd_t *pxd;
	int n;

	/*
	 * allocate from persistent map;
	 */
	if (maplock->flag & mlckALLOCXADLIST) {
		xadlistlock = (struct xdlistlock *) maplock;
		xad = xadlistlock->xdlist;
		for (n = 0; n < xadlistlock->count; n++, xad++) {
			if (xad->flag & (XAD_NEW | XAD_EXTENDED)) {
				xaddr = addressXAD(xad);
				xlen = lengthXAD(xad);
				dbUpdatePMap(ipbmap, false, xaddr,
					     (s64) xlen, tblk);
				xad->flag &= ~(XAD_NEW | XAD_EXTENDED);
				jfs_info("allocPMap: xaddr:0x%lx xlen:%d",
					 (ulong) xaddr, xlen);
			}
		}
	} else if (maplock->flag & mlckALLOCPXD) {
		pxdlock = (struct pxd_lock *) maplock;
		xaddr = addressPXD(&pxdlock->pxd);
		xlen = lengthPXD(&pxdlock->pxd);
		dbUpdatePMap(ipbmap, false, xaddr, (s64) xlen, tblk);
		jfs_info("allocPMap: xaddr:0x%lx xlen:%d", (ulong) xaddr, xlen);
	} else {		/* (maplock->flag & mlckALLOCPXDLIST) */

		pxdlistlock = (struct xdlistlock *) maplock;
		pxd = pxdlistlock->xdlist;
		for (n = 0; n < pxdlistlock->count; n++, pxd++) {
			xaddr = addressPXD(pxd);
			xlen = lengthPXD(pxd);
			dbUpdatePMap(ipbmap, false, xaddr, (s64) xlen,
				     tblk);
			jfs_info("allocPMap: xaddr:0x%lx xlen:%d",
				 (ulong) xaddr, xlen);
		}
	}
}

/*
 *	txFreeMap()
 *
 * function:	free from persistent and/or working map;
 *
 * todo: optimization
 */
void txFreeMap(struct inode *ip,
	       struct maplock * maplock, struct tblock * tblk, int maptype)
{
	struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
	struct xdlistlock *xadlistlock;
	xad_t *xad;
	s64 xaddr;
	int xlen;
	struct pxd_lock *pxdlock;
	struct xdlistlock *pxdlistlock;
	pxd_t *pxd;
	int n;

	jfs_info("txFreeMap: tblk:0x%p maplock:0x%p maptype:0x%x",
		 tblk, maplock, maptype);

	/*
	 * free from persistent map;
	 */
	if (maptype == COMMIT_PMAP || maptype == COMMIT_PWMAP) {
		if (maplock->flag & mlckFREEXADLIST) {
			xadlistlock = (struct xdlistlock *) maplock;
			xad = xadlistlock->xdlist;
			for (n = 0; n < xadlistlock->count; n++, xad++) {
				if (!(xad->flag & XAD_NEW)) {
					xaddr = addressXAD(xad);
					xlen = lengthXAD(xad);
					dbUpdatePMap(ipbmap, true, xaddr,
						     (s64) xlen, tblk);
					jfs_info("freePMap: xaddr:0x%lx "
						 "xlen:%d",
						 (ulong) xaddr, xlen);
				}
			}
		} else if (maplock->flag & mlckFREEPXD) {
			pxdlock = (struct pxd_lock *) maplock;
			xaddr = addressPXD(&pxdlock->pxd);
			xlen = lengthPXD(&pxdlock->pxd);
			dbUpdatePMap(ipbmap, true, xaddr, (s64) xlen,
				     tblk);
			jfs_info("freePMap: xaddr:0x%lx xlen:%d",
				 (ulong) xaddr, xlen);
		} else {	/* (maplock->flag & mlckALLOCPXDLIST) */

			pxdlistlock = (struct xdlistlock *) maplock;
			pxd = pxdlistlock->xdlist;
			for (n = 0; n < pxdlistlock->count; n++, pxd++) {
				xaddr = addressPXD(pxd);
				xlen = lengthPXD(pxd);
				dbUpdatePMap(ipbmap, true, xaddr,
					     (s64) xlen, tblk);
				jfs_info("freePMap: xaddr:0x%lx xlen:%d",
					 (ulong) xaddr, xlen);
			}
		}
	}

	/*
	 * free from working map;
	 */
	if (maptype == COMMIT_PWMAP || maptype == COMMIT_WMAP) {
		if (maplock->flag & mlckFREEXADLIST) {
			xadlistlock = (struct xdlistlock *) maplock;
			xad = xadlistlock->xdlist;
			for (n = 0; n < xadlistlock->count; n++, xad++) {
				xaddr = addressXAD(xad);
				xlen = lengthXAD(xad);
				dbFree(ip, xaddr, (s64) xlen);
				xad->flag = 0;
				jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
					 (ulong) xaddr, xlen);
			}
		} else if (maplock->flag & mlckFREEPXD) {
			pxdlock = (struct pxd_lock *) maplock;
			xaddr = addressPXD(&pxdlock->pxd);
			xlen = lengthPXD(&pxdlock->pxd);
			dbFree(ip, xaddr, (s64) xlen);
			jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
				 (ulong) xaddr, xlen);
		} else {	/* (maplock->flag & mlckFREEPXDLIST) */

			pxdlistlock = (struct xdlistlock *) maplock;
			pxd = pxdlistlock->xdlist;
			for (n = 0; n < pxdlistlock->count; n++, pxd++) {
				xaddr = addressPXD(pxd);
				xlen = lengthPXD(pxd);
				dbFree(ip, xaddr, (s64) xlen);
				jfs_info("freeWMap: xaddr:0x%lx xlen:%d",
					 (ulong) xaddr, xlen);
			}
		}
	}
}

/*
 *	txFreelock()
 *
 * function:	remove tlock from inode anonymous locklist
 */
void txFreelock(struct inode *ip)
{
	struct jfs_inode_info *jfs_ip = JFS_IP(ip);
	struct tlock *xtlck, *tlck;
	lid_t xlid = 0, lid;

	if (!jfs_ip->atlhead)