jfs_imap.c

jfs-2.4-1.1.7.tar.gz jfs 2.4－1.1.7 源码

	extno = ino >> L2INOSPEREXT;
	bitno = ino & (INOSPEREXT - 1);
	mask = HIGHORDER >> bitno;

	if (!(le32_to_cpu(iagp->wmap[extno]) & mask)) {
		jfs_error(ip->i_sb,
			  "diFree: wmap shows inode already free");
	}

	if (!addressPXD(&iagp->inoext[extno])) {
		release_metapage(mp);
		IREAD_UNLOCK(ipimap);
		AG_UNLOCK(imap, agno);
		jfs_error(ip->i_sb, "diFree: invalid inoext");
		return -EIO;
	}

	/* compute the bitmap for the extent reflecting the freed inode.
	 */
	bitmap = le32_to_cpu(iagp->wmap[extno]) & ~mask;

	if (imap->im_agctl[agno].numfree > imap->im_agctl[agno].numinos) {
		release_metapage(mp);
		IREAD_UNLOCK(ipimap);
		AG_UNLOCK(imap, agno);
		jfs_error(ip->i_sb, "diFree: numfree > numinos");
		return -EIO;
	}
	/*
	 *      inode extent still has some inodes or below low water mark:
	 *      keep the inode extent;
	 */
	if (bitmap ||
	    imap->im_agctl[agno].numfree < 96 ||
	    (imap->im_agctl[agno].numfree < 288 &&
	     (((imap->im_agctl[agno].numfree * 100) /
	       imap->im_agctl[agno].numinos) <= 25))) {
		/* if the iag currently has no free inodes (i.e.,
		 * the inode being freed is the first free inode of iag),
		 * insert the iag at head of the inode free list for the ag.
		 */
		if (iagp->nfreeinos == 0) {
			/* check if there are any iags on the ag inode
			 * free list.  if so, read the first one so that
			 * we can link the current iag onto the list at
			 * the head.
			 */
			if ((fwd = imap->im_agctl[agno].inofree) >= 0) {
				/* read the iag that currently is the head
				 * of the list.
				 */
				if ((rc = diIAGRead(imap, fwd, &amp))) {
					IREAD_UNLOCK(ipimap);
					AG_UNLOCK(imap, agno);
					release_metapage(mp);
					return (rc);
				}
				aiagp = (struct iag *) amp->data;

				/* make current head point back to the iag.
				 */
				aiagp->inofreeback = cpu_to_le32(iagno);

				write_metapage(amp);
			}

			/* iag points forward to current head and iag
			 * becomes the new head of the list.
			 */
			iagp->inofreefwd =
			    cpu_to_le32(imap->im_agctl[agno].inofree);
			iagp->inofreeback = -1;
			imap->im_agctl[agno].inofree = iagno;
		}
		IREAD_UNLOCK(ipimap);

		/* update the free inode summary map for the extent if
		 * freeing the inode means the extent will now have free
		 * inodes (i.e., the inode being freed is the first free 
		 * inode of extent),
		 */
		if (iagp->wmap[extno] == ONES) {
			sword = extno >> L2EXTSPERSUM;
			bitno = extno & (EXTSPERSUM - 1);
			iagp->inosmap[sword] &=
			    cpu_to_le32(~(HIGHORDER >> bitno));
		}

		/* update the bitmap.
		 */
		iagp->wmap[extno] = cpu_to_le32(bitmap);
		DBG_DIFREE(imap, inum);

		/* update the free inode counts at the iag, ag and
		 * map level.
		 */
		iagp->nfreeinos =
		    cpu_to_le32(le32_to_cpu(iagp->nfreeinos) + 1);
		imap->im_agctl[agno].numfree += 1;
		atomic_inc(&imap->im_numfree);

		/* release the AG inode map lock
		 */
		AG_UNLOCK(imap, agno);

		/* write the iag */
		write_metapage(mp);

		return (0);
	}


	/*
	 *      inode extent has become free and above low water mark:
	 *      free the inode extent;
	 */

	/*
	 *      prepare to update iag list(s) (careful update step 1)
	 */
	amp = bmp = cmp = dmp = NULL;
	fwd = back = -1;

	/* check if the iag currently has no free extents.  if so,
	 * it will be placed on the head of the ag extent free list.
	 */
	if (iagp->nfreeexts == 0) {
		/* check if the ag extent free list has any iags.
		 * if so, read the iag at the head of the list now.
		 * this (head) iag will be updated later to reflect
		 * the addition of the current iag at the head of
		 * the list.
		 */
		if ((fwd = imap->im_agctl[agno].extfree) >= 0) {
			if ((rc = diIAGRead(imap, fwd, &amp)))
				goto error_out;
			aiagp = (struct iag *) amp->data;
		}
	} else {
		/* iag has free extents. check if the addition of a free
		 * extent will cause all extents to be free within this
		 * iag.  if so, the iag will be removed from the ag extent
		 * free list and placed on the inode map's free iag list.
		 */
		if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG - 1)) {
			/* in preparation for removing the iag from the
			 * ag extent free list, read the iags preceeding
			 * and following the iag on the ag extent free
			 * list.
			 */
			if ((fwd = le32_to_cpu(iagp->extfreefwd)) >= 0) {
				if ((rc = diIAGRead(imap, fwd, &amp)))
					goto error_out;
				aiagp = (struct iag *) amp->data;
			}

			if ((back = le32_to_cpu(iagp->extfreeback)) >= 0) {
				if ((rc = diIAGRead(imap, back, &bmp)))
					goto error_out;
				biagp = (struct iag *) bmp->data;
			}
		}
	}

	/* remove the iag from the ag inode free list if freeing
	 * this extent cause the iag to have no free inodes.
	 */
	if (iagp->nfreeinos == cpu_to_le32(INOSPEREXT - 1)) {
		int inofreeback = le32_to_cpu(iagp->inofreeback);
		int inofreefwd = le32_to_cpu(iagp->inofreefwd);

		/* in preparation for removing the iag from the
		 * ag inode free list, read the iags preceeding
		 * and following the iag on the ag inode free
		 * list.  before reading these iags, we must make
		 * sure that we already don't have them in hand
		 * from up above, since re-reading an iag (buffer)
		 * we are currently holding would cause a deadlock.
		 */
		if (inofreefwd >= 0) {

			if (inofreefwd == fwd)
				ciagp = (struct iag *) amp->data;
			else if (inofreefwd == back)
				ciagp = (struct iag *) bmp->data;
			else {
				if ((rc =
				     diIAGRead(imap, inofreefwd, &cmp)))
					goto error_out;
				ciagp = (struct iag *) cmp->data;
			}
			assert(ciagp != NULL);
		}

		if (inofreeback >= 0) {
			if (inofreeback == fwd)
				diagp = (struct iag *) amp->data;
			else if (inofreeback == back)
				diagp = (struct iag *) bmp->data;
			else {
				if ((rc =
				     diIAGRead(imap, inofreeback, &dmp)))
					goto error_out;
				diagp = (struct iag *) dmp->data;
			}
			assert(diagp != NULL);
		}
	}

	IREAD_UNLOCK(ipimap);

	/*
	 * invalidate any page of the inode extent freed from buffer cache;
	 */
	freepxd = iagp->inoext[extno];
	invalidate_pxd_metapages(ip, freepxd);

	/*
	 *      update iag list(s) (careful update step 2)
	 */
	/* add the iag to the ag extent free list if this is the
	 * first free extent for the iag.
	 */
	if (iagp->nfreeexts == 0) {
		if (fwd >= 0)
			aiagp->extfreeback = cpu_to_le32(iagno);

		iagp->extfreefwd =
		    cpu_to_le32(imap->im_agctl[agno].extfree);
		iagp->extfreeback = -1;
		imap->im_agctl[agno].extfree = iagno;
	} else {
		/* remove the iag from the ag extent list if all extents
		 * are now free and place it on the inode map iag free list.
		 */
		if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG - 1)) {
			if (fwd >= 0)
				aiagp->extfreeback = iagp->extfreeback;

			if (back >= 0)
				biagp->extfreefwd = iagp->extfreefwd;
			else
				imap->im_agctl[agno].extfree =
				    le32_to_cpu(iagp->extfreefwd);

			iagp->extfreefwd = iagp->extfreeback = -1;

			IAGFREE_LOCK(imap);
			iagp->iagfree = cpu_to_le32(imap->im_freeiag);
			imap->im_freeiag = iagno;
			IAGFREE_UNLOCK(imap);
		}
	}

	/* remove the iag from the ag inode free list if freeing
	 * this extent causes the iag to have no free inodes.
	 */
	if (iagp->nfreeinos == cpu_to_le32(INOSPEREXT - 1)) {
		if ((int) le32_to_cpu(iagp->inofreefwd) >= 0)
			ciagp->inofreeback = iagp->inofreeback;

		if ((int) le32_to_cpu(iagp->inofreeback) >= 0)
			diagp->inofreefwd = iagp->inofreefwd;
		else
			imap->im_agctl[agno].inofree =
			    le32_to_cpu(iagp->inofreefwd);

		iagp->inofreefwd = iagp->inofreeback = -1;
	}

	/* update the inode extent address and working map 
	 * to reflect the free extent.
	 * the permanent map should have been updated already 
	 * for the inode being freed.
	 */
	if (iagp->pmap[extno] != 0) {
		jfs_error(ip->i_sb, "diFree: the pmap does not show inode free");
	}
	iagp->wmap[extno] = 0;
	DBG_DIFREE(imap, inum);
	PXDlength(&iagp->inoext[extno], 0);
	PXDaddress(&iagp->inoext[extno], 0);

	/* update the free extent and free inode summary maps
	 * to reflect the freed extent.
	 * the inode summary map is marked to indicate no inodes 
	 * available for the freed extent.
	 */
	sword = extno >> L2EXTSPERSUM;
	bitno = extno & (EXTSPERSUM - 1);
	mask = HIGHORDER >> bitno;
	iagp->inosmap[sword] |= cpu_to_le32(mask);
	iagp->extsmap[sword] &= cpu_to_le32(~mask);

	/* update the number of free inodes and number of free extents
	 * for the iag.
	 */
	iagp->nfreeinos = cpu_to_le32(le32_to_cpu(iagp->nfreeinos) -
				      (INOSPEREXT - 1));
	iagp->nfreeexts = cpu_to_le32(le32_to_cpu(iagp->nfreeexts) + 1);

	/* update the number of free inodes and backed inodes
	 * at the ag and inode map level.
	 */
	imap->im_agctl[agno].numfree -= (INOSPEREXT - 1);
	imap->im_agctl[agno].numinos -= INOSPEREXT;
	atomic_sub(INOSPEREXT - 1, &imap->im_numfree);
	atomic_sub(INOSPEREXT, &imap->im_numinos);

	if (amp)
		write_metapage(amp);
	if (bmp)
		write_metapage(bmp);
	if (cmp)
		write_metapage(cmp);
	if (dmp)
		write_metapage(dmp);

	/*
	 * start transaction to update block allocation map
	 * for the inode extent freed;
	 *
	 * N.B. AG_LOCK is released and iag will be released below, and 
	 * other thread may allocate inode from/reusing the ixad freed
	 * BUT with new/different backing inode extent from the extent 
	 * to be freed by the transaction;  
	 */
	tid = txBegin(ipimap->i_sb, COMMIT_FORCE);
	down(&JFS_IP(ipimap)->commit_sem);

	/* acquire tlock of the iag page of the freed ixad 
	 * to force the page NOHOMEOK (even though no data is
	 * logged from the iag page) until NOREDOPAGE|FREEXTENT log 
	 * for the free of the extent is committed;
	 * write FREEXTENT|NOREDOPAGE log record
	 * N.B. linelock is overlaid as freed extent descriptor;
	 */
	tlck = txLock(tid, ipimap, mp, tlckINODE | tlckFREE);
	pxdlock = (struct pxd_lock *) & tlck->lock;
	pxdlock->flag = mlckFREEPXD;
	pxdlock->pxd = freepxd;
	pxdlock->index = 1;

	write_metapage(mp);

	iplist[0] = ipimap;

	/*
	 * logredo needs the IAG number and IAG extent index in order
	 * to ensure that the IMap is consistent.  The least disruptive
	 * way to pass these values through  to the transaction manager
	 * is in the iplist array.  
	 * 
	 * It's not pretty, but it works.
	 */
	iplist[1] = (struct inode *) (size_t)iagno;
	iplist[2] = (struct inode *) (size_t)extno;

	rc = txCommit(tid, 1, &iplist[0], COMMIT_FORCE);

	txEnd(tid);
	up(&JFS_IP(ipimap)->commit_sem);

	/* unlock the AG inode map information */
	AG_UNLOCK(imap, agno);

	return (0);

      error_out:
	IREAD_UNLOCK(ipimap);

	if (amp)
		release_metapage(amp);
	if (bmp)
		release_metapage(bmp);
	if (cmp)
		release_metapage(cmp);
	if (dmp)
		release_metapage(dmp);

	AG_UNLOCK(imap, agno);

	release_metapage(mp);

	return (rc);
}

/*
 * There are several places in the diAlloc* routines where we initialize
 * the inode.
 */
static inline void
diInitInode(struct inode *ip, int iagno, int ino, int extno, struct iag * iagp)
{
	struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
	struct jfs_inode_info *jfs_ip = JFS_IP(ip);

	ip->i_ino = (iagno << L2INOSPERIAG) + ino;
	DBG_DIALLOC(JFS_IP(ipimap)->i_imap, ip->i_ino);
	jfs_ip->ixpxd = iagp->inoext[extno];
	jfs_ip->agno = BLKTOAG(le64_to_cpu(iagp->agstart), sbi);
	jfs_ip->active_ag = -1;
}


/*
 * NAME:        diAlloc(pip,dir,ip)
 *
 * FUNCTION:    allocate a disk inode from the inode working map 
 *		for a fileset or aggregate.
 *
 * PARAMETERS:
 *      pip  	- pointer to incore inode for the parent inode.
 *      dir  	- TRUE if the new disk inode is for a directory.
 *      ip  	- pointer to a new inode
 *
 * RETURN VALUES:
 *      0       - success.
 *      -ENOSPC	- insufficient disk resources.
 *      -EIO  	- i/o error.
 */
int diAlloc(struct inode *pip, boolean_t dir, struct inode *ip)
{
	int rc, ino, iagno, addext, extno, bitno, sword;
	int nwords, rem, i, agno;
	u32 mask, inosmap, extsmap;
	struct inode *ipimap;
	struct metapage *mp;
	ino_t inum;
	struct iag *iagp;
	struct inomap *imap;

	/* get the pointers to the inode map inode and the
	 * corresponding imap control structure.
	 */
	ipimap = JFS_SBI(pip->i_sb)->ipimap;
	imap = JFS_IP(ipimap)->i_imap;
	JFS_IP(ip)->ipimap = ipimap;
	JFS_IP(ip)->fileset = FILESYSTEM_I;

	/* for a directory, the allocation policy is to start 
	 * at the ag level using the preferred ag.
	 */
	if (dir == TRUE) {
		agno = dbNextAG(JFS_SBI(pip->i_sb)->ipbmap);
		AG_LOCK(imap, agno);
		goto tryag;
	}

	/* for files, the policy starts off by trying to allocate from
	 * the same iag containing the parent disk inode:
	 * try to allocate the new disk inode close to the parent disk
	 * inode, using parent disk inode number + 1 as the allocation
	 * hint.  (we use a left-to-right policy to attempt to avoid
	 * moving backward on the disk.)  compute the hint within the
	 * file system and the iag.
	 */

	/* get the ag number of this iag */
	agno = JFS_IP(pip)->agno;

	if (atomic_read(&JFS_SBI(pip->i_sb)->bmap->db_active[agno])) {
		/*
		 * There is an open file actively growing.  We want to
		 * allocate new inodes from a different ag to avoid
		 * fragmentation problems.
		 */
		agno = dbNextAG(JFS_SBI(pip->i_sb)->ipbmap);
		AG_LOCK(imap, agno);
		goto tryag;
	}

	inum = pip->i_ino + 1;
	ino = inum & (INOSPERIAG - 1);

	/* back off the the hint if it is outside of the iag */
	if (ino == 0)
		inum = pip->i_ino;

	/* lock the AG inode map information */