[linux.git] / fs / jfs / jfs_extent.c

/*
 *   Copyright (C) International Business Machines Corp., 2000-2004
 *
 *   This program is free software;  you can redistribute it and/or modify
 *   it under the terms of the GNU General Public License as published by
 *   the Free Software Foundation; either version 2 of the License, or
 *   (at your option) any later version.
 *
 *   This program is distributed in the hope that it will be useful,
 *   but WITHOUT ANY WARRANTY;  without even the implied warranty of
 *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See
 *   the GNU General Public License for more details.
 *
 *   You should have received a copy of the GNU General Public License
 *   along with this program;  if not, write to the Free Software
 *   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
 */

#include <linux/fs.h>
#include <linux/quotaops.h>
#include "jfs_incore.h"
#include "jfs_inode.h"
#include "jfs_superblock.h"
#include "jfs_dmap.h"
#include "jfs_extent.h"
#include "jfs_debug.h"

/*
 * forward references
 */
static int extBalloc(struct inode *, s64, s64 *, s64 *);
#ifdef _NOTYET
static int extBrealloc(struct inode *, s64, s64, s64 *, s64 *);
#endif
static s64 extRoundDown(s64 nb);

#define DPD(a)		(printk("(a): %d\n",(a)))
#define DPC(a)		(printk("(a): %c\n",(a)))
#define DPL1(a)					\
{						\
	if ((a) >> 32)				\
		printk("(a): %x%08x  ",(a));	\
	else					\
		printk("(a): %x  ",(a) << 32);	\
}
#define DPL(a)					\
{						\
	if ((a) >> 32)				\
		printk("(a): %x%08x\n",(a));	\
	else					\
		printk("(a): %x\n",(a) << 32);	\
}

#define DPD1(a)		(printk("(a): %d  ",(a)))
#define DPX(a)		(printk("(a): %08x\n",(a)))
#define DPX1(a)		(printk("(a): %08x  ",(a)))
#define DPS(a)		(printk("%s\n",(a)))
#define DPE(a)		(printk("\nENTERING: %s\n",(a)))
#define DPE1(a)		(printk("\nENTERING: %s",(a)))
#define DPS1(a)		(printk("  %s  ",(a)))


/*
 * NAME:	extAlloc()
 *
 * FUNCTION:	allocate an extent for a specified page range within a
 *		file.
 *
 * PARAMETERS:
 *	ip	- the inode of the file.
 *	xlen	- requested extent length.
 *	pno	- the starting page number with the file.
 *	xp	- pointer to an xad.  on entry, xad describes an
 *		  extent that is used as an allocation hint if the
 *		  xaddr of the xad is non-zero.  on successful exit,
 *		  the xad describes the newly allocated extent.
 *	abnr	- bool indicating whether the newly allocated extent
 *		  should be marked as allocated but not recorded.
 *
 * RETURN VALUES:
 *	0	- success
 *	-EIO	- i/o error.
 *	-ENOSPC	- insufficient disk resources.
 */
int
extAlloc(struct inode *ip, s64 xlen, s64 pno, xad_t * xp, bool abnr)
{
	struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
	s64 nxlen, nxaddr, xoff, hint, xaddr = 0;
	int rc;
	int xflag;

	/* This blocks if we are low on resources */
	txBeginAnon(ip->i_sb);

	/* Avoid race with jfs_commit_inode() */
	mutex_lock(&JFS_IP(ip)->commit_mutex);

	/* validate extent length */
	if (xlen > MAXXLEN)
		xlen = MAXXLEN;

	/* get the page's starting extent offset */
	xoff = pno << sbi->l2nbperpage;

	/* check if an allocation hint was provided */
	if ((hint = addressXAD(xp))) {
		/* get the size of the extent described by the hint */
		nxlen = lengthXAD(xp);

		/* check if the hint is for the portion of the file
		 * immediately previous to the current allocation
		 * request and if hint extent has the same abnr
		 * value as the current request.  if so, we can
		 * extend the hint extent to include the current
		 * extent if we can allocate the blocks immediately
		 * following the hint extent.
		 */
		if (offsetXAD(xp) + nxlen == xoff &&
		    abnr == ((xp->flag & XAD_NOTRECORDED) ? true : false))
			xaddr = hint + nxlen;

		/* adjust the hint to the last block of the extent */
		hint += (nxlen - 1);
	}

	/* allocate the disk blocks for the extent.  initially, extBalloc()
	 * will try to allocate disk blocks for the requested size (xlen).
	 * if this fails (xlen contiguous free blocks not available), it'll
	 * try to allocate a smaller number of blocks (producing a smaller
	 * extent), with this smaller number of blocks consisting of the
	 * requested number of blocks rounded down to the next smaller
	 * power of 2 number (i.e. 16 -> 8).  it'll continue to round down
	 * and retry the allocation until the number of blocks to allocate
	 * is smaller than the number of blocks per page.
	 */
	nxlen = xlen;
	if ((rc = extBalloc(ip, hint ? hint : INOHINT(ip), &nxlen, &nxaddr))) {
		mutex_unlock(&JFS_IP(ip)->commit_mutex);
		return (rc);
	}

	/* Allocate blocks to quota. */
	rc = dquot_alloc_block(ip, nxlen);
	if (rc) {
		dbFree(ip, nxaddr, (s64) nxlen);
		mutex_unlock(&JFS_IP(ip)->commit_mutex);
		return rc;
	}

	/* determine the value of the extent flag */
	xflag = abnr ? XAD_NOTRECORDED : 0;

	/* if we can extend the hint extent to cover the current request,
	 * extend it.  otherwise, insert a new extent to
	 * cover the current request.
	 */
	if (xaddr && xaddr == nxaddr)
		rc = xtExtend(0, ip, xoff, (int) nxlen, 0);
	else
		rc = xtInsert(0, ip, xflag, xoff, (int) nxlen, &nxaddr, 0);

	/* if the extend or insert failed,
	 * free the newly allocated blocks and return the error.
	 */
	if (rc) {
		dbFree(ip, nxaddr, nxlen);
		dquot_free_block(ip, nxlen);
		mutex_unlock(&JFS_IP(ip)->commit_mutex);
		return (rc);
	}

	/* set the results of the extent allocation */
	XADaddress(xp, nxaddr);
	XADlength(xp, nxlen);
	XADoffset(xp, xoff);
	xp->flag = xflag;

	mark_inode_dirty(ip);

	mutex_unlock(&JFS_IP(ip)->commit_mutex);
	/*
	 * COMMIT_SyncList flags an anonymous tlock on page that is on
	 * sync list.
	 * We need to commit the inode to get the page written disk.
	 */
	if (test_and_clear_cflag(COMMIT_Synclist,ip))
		jfs_commit_inode(ip, 0);

	return (0);
}


#ifdef _NOTYET
/*
 * NAME:	extRealloc()
 *
 * FUNCTION:	extend the allocation of a file extent containing a
 *		partial back last page.
 *
 * PARAMETERS:
 *	ip	- the inode of the file.
 *	cp	- cbuf for the partial backed last page.
 *	xlen	- request size of the resulting extent.
 *	xp	- pointer to an xad. on successful exit, the xad
 *		  describes the newly allocated extent.
 *	abnr	- bool indicating whether the newly allocated extent
 *		  should be marked as allocated but not recorded.
 *
 * RETURN VALUES:
 *	0	- success
 *	-EIO	- i/o error.
 *	-ENOSPC	- insufficient disk resources.
 */
int extRealloc(struct inode *ip, s64 nxlen, xad_t * xp, bool abnr)
{
	struct super_block *sb = ip->i_sb;
	s64 xaddr, xlen, nxaddr, delta, xoff;
	s64 ntail, nextend, ninsert;
	int rc, nbperpage = JFS_SBI(sb)->nbperpage;
	int xflag;

	/* This blocks if we are low on resources */
	txBeginAnon(ip->i_sb);

	mutex_lock(&JFS_IP(ip)->commit_mutex);
	/* validate extent length */
	if (nxlen > MAXXLEN)
		nxlen = MAXXLEN;

	/* get the extend (partial) page's disk block address and
	 * number of blocks.
	 */
	xaddr = addressXAD(xp);
	xlen = lengthXAD(xp);
	xoff = offsetXAD(xp);

	/* if the extend page is abnr and if the request is for
	 * the extent to be allocated and recorded,
	 * make the page allocated and recorded.
	 */
	if ((xp->flag & XAD_NOTRECORDED) && !abnr) {
		xp->flag = 0;
		if ((rc = xtUpdate(0, ip, xp)))
			goto exit;
	}

	/* try to allocated the request number of blocks for the
	 * extent.  dbRealloc() first tries to satisfy the request
	 * by extending the allocation in place. otherwise, it will
	 * try to allocate a new set of blocks large enough for the
	 * request.  in satisfying a request, dbReAlloc() may allocate
	 * less than what was request but will always allocate enough
	 * space as to satisfy the extend page.
	 */
	if ((rc = extBrealloc(ip, xaddr, xlen, &nxlen, &nxaddr)))
		goto exit;

	/* Allocat blocks to quota. */
	rc = dquot_alloc_block(ip, nxlen);
	if (rc) {
		dbFree(ip, nxaddr, (s64) nxlen);
		mutex_unlock(&JFS_IP(ip)->commit_mutex);
		return rc;
	}

	delta = nxlen - xlen;

	/* check if the extend page is not abnr but the request is abnr
	 * and the allocated disk space is for more than one page.  if this
	 * is the case, there is a miss match of abnr between the extend page
	 * and the one or more pages following the extend page.  as a result,
	 * two extents will have to be manipulated. the first will be that
	 * of the extent of the extend page and will be manipulated thru
	 * an xtExtend() or an xtTailgate(), depending upon whether the
	 * disk allocation occurred as an inplace extension.  the second
	 * extent will be manipulated (created) through an xtInsert() and
	 * will be for the pages following the extend page.
	 */
	if (abnr && (!(xp->flag & XAD_NOTRECORDED)) && (nxlen > nbperpage)) {
		ntail = nbperpage;
		nextend = ntail - xlen;
		ninsert = nxlen - nbperpage;

		xflag = XAD_NOTRECORDED;
	} else {
		ntail = nxlen;
		nextend = delta;
		ninsert = 0;

		xflag = xp->flag;
	}

	/* if we were able to extend the disk allocation in place,
	 * extend the extent.  otherwise, move the extent to a
	 * new disk location.
	 */
	if (xaddr == nxaddr) {
		/* extend the extent */
		if ((rc = xtExtend(0, ip, xoff + xlen, (int) nextend, 0))) {
			dbFree(ip, xaddr + xlen, delta);
			dquot_free_block(ip, nxlen);
			goto exit;
		}
	} else {
		/*
		 * move the extent to a new location:
		 *
		 * xtTailgate() accounts for relocated tail extent;
		 */
		if ((rc = xtTailgate(0, ip, xoff, (int) ntail, nxaddr, 0))) {
			dbFree(ip, nxaddr, nxlen);
			dquot_free_block(ip, nxlen);
			goto exit;
		}
	}


	/* check if we need to also insert a new extent */
	if (ninsert) {
		/* perform the insert.  if it fails, free the blocks
		 * to be inserted and make it appear that we only did
		 * the xtExtend() or xtTailgate() above.
		 */
		xaddr = nxaddr + ntail;
		if (xtInsert (0, ip, xflag, xoff + ntail, (int) ninsert,
			      &xaddr, 0)) {
			dbFree(ip, xaddr, (s64) ninsert);
			delta = nextend;
			nxlen = ntail;
			xflag = 0;
		}
	}

	/* set the return results */
	XADaddress(xp, nxaddr);
	XADlength(xp, nxlen);
	XADoffset(xp, xoff);
	xp->flag = xflag;

	mark_inode_dirty(ip);
exit:
	mutex_unlock(&JFS_IP(ip)->commit_mutex);
	return (rc);
}
#endif			/* _NOTYET */


/*
 * NAME:	extHint()
 *
 * FUNCTION:	produce an extent allocation hint for a file offset.
 *
 * PARAMETERS:
 *	ip	- the inode of the file.
 *	offset  - file offset for which the hint is needed.
 *	xp	- pointer to the xad that is to be filled in with
 *		  the hint.
 *
 * RETURN VALUES:
 *	0	- success
 *	-EIO	- i/o error.
 */
int extHint(struct inode *ip, s64 offset, xad_t * xp)
{
	struct super_block *sb = ip->i_sb;
	int nbperpage = JFS_SBI(sb)->nbperpage;
	s64 prev;
	int rc = 0;
	s64 xaddr;
	int xlen;
	int xflag;

	/* init the hint as "no hint provided" */
	XADaddress(xp, 0);

	/* determine the starting extent offset of the page previous
	 * to the page containing the offset.
	 */
	prev = ((offset & ~POFFSET) >> JFS_SBI(sb)->l2bsize) - nbperpage;

	/* if the offset is in the first page of the file, no hint provided.
	 */
	if (prev < 0)
		goto out;

	rc = xtLookup(ip, prev, nbperpage, &xflag, &xaddr, &xlen, 0);

	if ((rc == 0) && xlen) {
		if (xlen != nbperpage) {
			jfs_error(ip->i_sb, "extHint: corrupt xtree");
			rc = -EIO;
		}
		XADaddress(xp, xaddr);
		XADlength(xp, xlen);
		XADoffset(xp, prev);
		/*
		 * only preserve the abnr flag within the xad flags
		 * of the returned hint.
		 */
		xp->flag  = xflag & XAD_NOTRECORDED;
	} else
		rc = 0;

out:
	return (rc);
}


/*
 * NAME:	extRecord()
 *
 * FUNCTION:	change a page with a file from not recorded to recorded.
 *
 * PARAMETERS:
 *	ip	- inode of the file.
 *	cp	- cbuf of the file page.
 *
 * RETURN VALUES:
 *	0	- success
 *	-EIO	- i/o error.
 *	-ENOSPC	- insufficient disk resources.
 */
int extRecord(struct inode *ip, xad_t * xp)
{
	int rc;

	txBeginAnon(ip->i_sb);

	mutex_lock(&JFS_IP(ip)->commit_mutex);

	/* update the extent */
	rc = xtUpdate(0, ip, xp);

	mutex_unlock(&JFS_IP(ip)->commit_mutex);
	return rc;
}


#ifdef _NOTYET
/*
 * NAME:	extFill()
 *
 * FUNCTION:	allocate disk space for a file page that represents
 *		a file hole.
 *
 * PARAMETERS:
 *	ip	- the inode of the file.
 *	cp	- cbuf of the file page represent the hole.
 *
 * RETURN VALUES:
 *	0	- success
 *	-EIO	- i/o error.
 *	-ENOSPC	- insufficient disk resources.
 */
int extFill(struct inode *ip, xad_t * xp)
{
	int rc, nbperpage = JFS_SBI(ip->i_sb)->nbperpage;
	s64 blkno = offsetXAD(xp) >> ip->i_blkbits;

//	assert(ISSPARSE(ip));

	/* initialize the extent allocation hint */
	XADaddress(xp, 0);

	/* allocate an extent to fill the hole */
	if ((rc = extAlloc(ip, nbperpage, blkno, xp, false)))
		return (rc);

	assert(lengthPXD(xp) == nbperpage);

	return (0);
}
#endif			/* _NOTYET */


/*
 * NAME:	extBalloc()
 *
 * FUNCTION:	allocate disk blocks to form an extent.
 *
 *		initially, we will try to allocate disk blocks for the
 *		requested size (nblocks).  if this fails (nblocks
 *		contiguous free blocks not available), we'll try to allocate
 *		a smaller number of blocks (producing a smaller extent), with
 *		this smaller number of blocks consisting of the requested
 *		number of blocks rounded down to the next smaller power of 2
 *		number (i.e. 16 -> 8).  we'll continue to round down and
 *		retry the allocation until the number of blocks to allocate
 *		is smaller than the number of blocks per page.
 *
 * PARAMETERS:
 *	ip	 - the inode of the file.
 *	hint	 - disk block number to be used as an allocation hint.
 *	*nblocks - pointer to an s64 value.  on entry, this value specifies
 *		   the desired number of block to be allocated. on successful
 *		   exit, this value is set to the number of blocks actually
 *		   allocated.
 *	blkno	 - pointer to a block address that is filled in on successful
 *		   return with the starting block number of the newly
 *		   allocated block range.
 *
 * RETURN VALUES:
 *	0	- success
 *	-EIO	- i/o error.
 *	-ENOSPC	- insufficient disk resources.
 */
static int
extBalloc(struct inode *ip, s64 hint, s64 * nblocks, s64 * blkno)
{
	struct jfs_inode_info *ji = JFS_IP(ip);
	struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
	s64 nb, nblks, daddr, max;
	int rc, nbperpage = sbi->nbperpage;
	struct bmap *bmp = sbi->bmap;
	int ag;

	/* get the number of blocks to initially attempt to allocate.
	 * we'll first try the number of blocks requested unless this
	 * number is greater than the maximum number of contiguous free
	 * blocks in the map. in that case, we'll start off with the
	 * maximum free.
	 */
	max = (s64) 1 << bmp->db_maxfreebud;
	if (*nblocks >= max && *nblocks > nbperpage)
		nb = nblks = (max > nbperpage) ? max : nbperpage;
	else
		nb = nblks = *nblocks;

	/* try to allocate blocks */
	while ((rc = dbAlloc(ip, hint, nb, &daddr)) != 0) {
		/* if something other than an out of space error,
		 * stop and return this error.
		 */
		if (rc != -ENOSPC)
			return (rc);

		/* decrease the allocation request size */
		nb = min(nblks, extRoundDown(nb));

		/* give up if we cannot cover a page */
		if (nb < nbperpage)
			return (rc);
	}

	*nblocks = nb;
	*blkno = daddr;

	if (S_ISREG(ip->i_mode) && (ji->fileset == FILESYSTEM_I)) {
		ag = BLKTOAG(daddr, sbi);
		spin_lock_irq(&ji->ag_lock);
		if (ji->active_ag == -1) {
			atomic_inc(&bmp->db_active[ag]);
			ji->active_ag = ag;
		} else if (ji->active_ag != ag) {
			atomic_dec(&bmp->db_active[ji->active_ag]);
			atomic_inc(&bmp->db_active[ag]);
			ji->active_ag = ag;
		}
		spin_unlock_irq(&ji->ag_lock);
	}

	return (0);
}


#ifdef _NOTYET
/*
 * NAME:	extBrealloc()
 *
 * FUNCTION:	attempt to extend an extent's allocation.
 *
 *		Initially, we will try to extend the extent's allocation
 *		in place.  If this fails, we'll try to move the extent
 *		to a new set of blocks.  If moving the extent, we initially
 *		will try to allocate disk blocks for the requested size
 *		(newnblks).  if this fails (new contiguous free blocks not
 *		available), we'll try to allocate a smaller number of
 *		blocks (producing a smaller extent), with this smaller
 *		number of blocks consisting of the requested number of
 *		blocks rounded down to the next smaller power of 2
 *		number (i.e. 16 -> 8).  We'll continue to round down and
 *		retry the allocation until the number of blocks to allocate
 *		is smaller than the number of blocks per page.
 *
 * PARAMETERS:
 *	ip	 - the inode of the file.
 *	blkno	 - starting block number of the extents current allocation.
 *	nblks	 - number of blocks within the extents current allocation.
 *	newnblks - pointer to a s64 value.  on entry, this value is the
 *		   the new desired extent size (number of blocks).  on
 *		   successful exit, this value is set to the extent's actual
 *		   new size (new number of blocks).
 *	newblkno - the starting block number of the extents new allocation.
 *
 * RETURN VALUES:
 *	0	- success
 *	-EIO	- i/o error.
 *	-ENOSPC	- insufficient disk resources.
 */
static int
extBrealloc(struct inode *ip,
	    s64 blkno, s64 nblks, s64 * newnblks, s64 * newblkno)
{
	int rc;

	/* try to extend in place */
	if ((rc = dbExtend(ip, blkno, nblks, *newnblks - nblks)) == 0) {
		*newblkno = blkno;
		return (0);
	} else {
		if (rc != -ENOSPC)
			return (rc);
	}

	/* in place extension not possible.
	 * try to move the extent to a new set of blocks.
	 */
	return (extBalloc(ip, blkno, newnblks, newblkno));
}
#endif			/* _NOTYET */


/*
 * NAME:	extRoundDown()
 *
 * FUNCTION:	round down a specified number of blocks to the next
 *		smallest power of 2 number.
 *
 * PARAMETERS:
 *	nb	- the inode of the file.
 *
 * RETURN VALUES:
 *	next smallest power of 2 number.
 */
static s64 extRoundDown(s64 nb)
{
	int i;
	u64 m, k;

	for (i = 0, m = (u64) 1 << 63; i < 64; i++, m >>= 1) {
		if (m & nb)
			break;
	}

	i = 63 - i;
	k = (u64) 1 << i;
	k = ((k - 1) & nb) ? k : k >> 1;

	return (k);
}
Commit	Line	Data
1da177e4 LT	1	/*
	2	* Copyright (C) International Business Machines Corp., 2000-2004
	3	*
	4	* This program is free software; you can redistribute it and/or modify
	5	* it under the terms of the GNU General Public License as published by
63f83c9f	6	* the Free Software Foundation; either version 2 of the License, or
1da177e4	7	* (at your option) any later version.
63f83c9f	8	*
1da177e4 LT	9	* This program is distributed in the hope that it will be useful,
	10	* but WITHOUT ANY WARRANTY; without even the implied warranty of
	11	* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
	12	* the GNU General Public License for more details.
	13	*
	14	* You should have received a copy of the GNU General Public License
63f83c9f	15	* along with this program; if not, write to the Free Software
1da177e4 LT	16	* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
	17	*/
	18
	19	#include <linux/fs.h>
	20	#include <linux/quotaops.h>
	21	#include "jfs_incore.h"
1868f4aa	22	#include "jfs_inode.h"
1da177e4 LT	23	#include "jfs_superblock.h"
	24	#include "jfs_dmap.h"
	25	#include "jfs_extent.h"
	26	#include "jfs_debug.h"
	27
	28	/*
	29	* forward references
	30	*/
	31	static int extBalloc(struct inode , s64, s64 , s64 *);
	32	#ifdef _NOTYET
	33	static int extBrealloc(struct inode , s64, s64, s64 , s64 *);
	34	#endif
	35	static s64 extRoundDown(s64 nb);
	36
f720e3ba DK	37	#define DPD(a) (printk("(a): %d\n",(a)))
f720e3ba DK	38	#define DPC(a) (printk("(a): %c\n",(a)))
1da177e4 LT	39	#define DPL1(a) \
	40	{ \
	41	if ((a) >> 32) \
	42	printk("(a): %x%08x ",(a)); \
	43	else \
	44	printk("(a): %x ",(a) << 32); \
	45	}
	46	#define DPL(a) \
	47	{ \
	48	if ((a) >> 32) \
	49	printk("(a): %x%08x\n",(a)); \
	50	else \
	51	printk("(a): %x\n",(a) << 32); \
	52	}
	53
f720e3ba DK	54	#define DPD1(a) (printk("(a): %d ",(a)))
	55	#define DPX(a) (printk("(a): %08x\n",(a)))
	56	#define DPX1(a) (printk("(a): %08x ",(a)))
	57	#define DPS(a) (printk("%s\n",(a)))
	58	#define DPE(a) (printk("\nENTERING: %s\n",(a)))
	59	#define DPE1(a) (printk("\nENTERING: %s",(a)))
	60	#define DPS1(a) (printk(" %s ",(a)))
1da177e4 LT	61
	62
	63	/*
	64	* NAME: extAlloc()
	65	*
f720e3ba	66	* FUNCTION: allocate an extent for a specified page range within a
1da177e4 LT	67	* file.
	68	*
	69	* PARAMETERS:
	70	* ip - the inode of the file.
	71	* xlen - requested extent length.
	72	* pno - the starting page number with the file.
	73	* xp - pointer to an xad. on entry, xad describes an
	74	* extent that is used as an allocation hint if the
	75	* xaddr of the xad is non-zero. on successful exit,
	76	* the xad describes the newly allocated extent.
4d81715f	77	* abnr - bool indicating whether the newly allocated extent
1da177e4 LT	78	* should be marked as allocated but not recorded.
	79	*
	80	* RETURN VALUES:
f720e3ba DK	81	* 0 - success
	82	* -EIO - i/o error.
	83	* -ENOSPC - insufficient disk resources.
1da177e4 LT	84	*/
1da177e4 LT	85	int
4d81715f	86	extAlloc(struct inode ip, s64 xlen, s64 pno, xad_t xp, bool abnr)
1da177e4 LT	87	{
	88	struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
	89	s64 nxlen, nxaddr, xoff, hint, xaddr = 0;
	90	int rc;
	91	int xflag;
	92
	93	/* This blocks if we are low on resources */
	94	txBeginAnon(ip->i_sb);
	95
	96	/* Avoid race with jfs_commit_inode() */
1de87444	97	mutex_lock(&JFS_IP(ip)->commit_mutex);
1da177e4 LT	98
	99	/* validate extent length */
	100	if (xlen > MAXXLEN)
	101	xlen = MAXXLEN;
	102
	103	/* get the page's starting extent offset */
	104	xoff = pno << sbi->l2nbperpage;
	105
	106	/* check if an allocation hint was provided */
	107	if ((hint = addressXAD(xp))) {
	108	/* get the size of the extent described by the hint */
	109	nxlen = lengthXAD(xp);
	110
	111	/* check if the hint is for the portion of the file
	112	* immediately previous to the current allocation
	113	* request and if hint extent has the same abnr
	114	* value as the current request. if so, we can
	115	* extend the hint extent to include the current
	116	* extent if we can allocate the blocks immediately
	117	* following the hint extent.
	118	*/
	119	if (offsetXAD(xp) + nxlen == xoff &&
4d81715f	120	abnr == ((xp->flag & XAD_NOTRECORDED) ? true : false))
1da177e4 LT	121	xaddr = hint + nxlen;
	122
	123	/* adjust the hint to the last block of the extent */
	124	hint += (nxlen - 1);
	125	}
	126
	127	/* allocate the disk blocks for the extent. initially, extBalloc()
63f83c9f	128	* will try to allocate disk blocks for the requested size (xlen).
25985edc	129	* if this fails (xlen contiguous free blocks not available), it'll
1da177e4 LT	130	* try to allocate a smaller number of blocks (producing a smaller
	131	* extent), with this smaller number of blocks consisting of the
	132	* requested number of blocks rounded down to the next smaller
	133	* power of 2 number (i.e. 16 -> 8). it'll continue to round down
	134	* and retry the allocation until the number of blocks to allocate
	135	* is smaller than the number of blocks per page.
	136	*/
	137	nxlen = xlen;
	138	if ((rc = extBalloc(ip, hint ? hint : INOHINT(ip), &nxlen, &nxaddr))) {
1de87444	139	mutex_unlock(&JFS_IP(ip)->commit_mutex);
1da177e4 LT	140	return (rc);
	141	}
	142
	143	/* Allocate blocks to quota. */
5dd4056d CH	144	rc = dquot_alloc_block(ip, nxlen);
5dd4056d CH	145	if (rc) {
1da177e4	146	dbFree(ip, nxaddr, (s64) nxlen);
1de87444	147	mutex_unlock(&JFS_IP(ip)->commit_mutex);
5dd4056d	148	return rc;
1da177e4 LT	149	}
	150
	151	/* determine the value of the extent flag */
4d81715f	152	xflag = abnr ? XAD_NOTRECORDED : 0;
1da177e4	153
63f83c9f	154	/* if we can extend the hint extent to cover the current request,
1da177e4 LT	155	* extend it. otherwise, insert a new extent to
	156	* cover the current request.
	157	*/
	158	if (xaddr && xaddr == nxaddr)
	159	rc = xtExtend(0, ip, xoff, (int) nxlen, 0);
	160	else
	161	rc = xtInsert(0, ip, xflag, xoff, (int) nxlen, &nxaddr, 0);
	162
63f83c9f	163	/* if the extend or insert failed,
1da177e4 LT	164	* free the newly allocated blocks and return the error.
	165	*/
	166	if (rc) {
	167	dbFree(ip, nxaddr, nxlen);
5dd4056d	168	dquot_free_block(ip, nxlen);
1de87444	169	mutex_unlock(&JFS_IP(ip)->commit_mutex);
1da177e4 LT	170	return (rc);
	171	}
	172
	173	/* set the results of the extent allocation */
	174	XADaddress(xp, nxaddr);
	175	XADlength(xp, nxlen);
	176	XADoffset(xp, xoff);
	177	xp->flag = xflag;
	178
	179	mark_inode_dirty(ip);
	180
1de87444	181	mutex_unlock(&JFS_IP(ip)->commit_mutex);
1da177e4 LT	182	/*
	183	* COMMIT_SyncList flags an anonymous tlock on page that is on
	184	* sync list.
	185	* We need to commit the inode to get the page written disk.
	186	*/
	187	if (test_and_clear_cflag(COMMIT_Synclist,ip))
	188	jfs_commit_inode(ip, 0);
	189
	190	return (0);
	191	}
	192
	193
	194	#ifdef _NOTYET
	195	/*
f720e3ba	196	* NAME: extRealloc()
1da177e4	197	*
f720e3ba	198	* FUNCTION: extend the allocation of a file extent containing a
1da177e4 LT	199	* partial back last page.
	200	*
	201	* PARAMETERS:
	202	* ip - the inode of the file.
	203	* cp - cbuf for the partial backed last page.
	204	* xlen - request size of the resulting extent.
	205	* xp - pointer to an xad. on successful exit, the xad
	206	* describes the newly allocated extent.
4d81715f	207	* abnr - bool indicating whether the newly allocated extent
1da177e4 LT	208	* should be marked as allocated but not recorded.
	209	*
	210	* RETURN VALUES:
f720e3ba DK	211	* 0 - success
	212	* -EIO - i/o error.
	213	* -ENOSPC - insufficient disk resources.
1da177e4	214	*/
4d81715f	215	int extRealloc(struct inode ip, s64 nxlen, xad_t xp, bool abnr)
1da177e4 LT	216	{
	217	struct super_block *sb = ip->i_sb;
	218	s64 xaddr, xlen, nxaddr, delta, xoff;
	219	s64 ntail, nextend, ninsert;
	220	int rc, nbperpage = JFS_SBI(sb)->nbperpage;
	221	int xflag;
	222
	223	/* This blocks if we are low on resources */
	224	txBeginAnon(ip->i_sb);
	225
1de87444	226	mutex_lock(&JFS_IP(ip)->commit_mutex);
1da177e4 LT	227	/* validate extent length */
	228	if (nxlen > MAXXLEN)
	229	nxlen = MAXXLEN;
	230
	231	/* get the extend (partial) page's disk block address and
	232	* number of blocks.
	233	*/
	234	xaddr = addressXAD(xp);
	235	xlen = lengthXAD(xp);
	236	xoff = offsetXAD(xp);
	237
	238	/* if the extend page is abnr and if the request is for
63f83c9f	239	* the extent to be allocated and recorded,
1da177e4 LT	240	* make the page allocated and recorded.
	241	*/
	242	if ((xp->flag & XAD_NOTRECORDED) && !abnr) {
	243	xp->flag = 0;
	244	if ((rc = xtUpdate(0, ip, xp)))
	245	goto exit;
	246	}
	247
	248	/* try to allocated the request number of blocks for the
	249	* extent. dbRealloc() first tries to satisfy the request
	250	* by extending the allocation in place. otherwise, it will
	251	* try to allocate a new set of blocks large enough for the
	252	* request. in satisfying a request, dbReAlloc() may allocate
	253	* less than what was request but will always allocate enough
	254	* space as to satisfy the extend page.
	255	*/
	256	if ((rc = extBrealloc(ip, xaddr, xlen, &nxlen, &nxaddr)))
	257	goto exit;
	258
	259	/* Allocat blocks to quota. */
5dd4056d CH	260	rc = dquot_alloc_block(ip, nxlen);
5dd4056d CH	261	if (rc) {
1da177e4	262	dbFree(ip, nxaddr, (s64) nxlen);
1de87444	263	mutex_unlock(&JFS_IP(ip)->commit_mutex);
5dd4056d	264	return rc;
1da177e4 LT	265	}
	266
	267	delta = nxlen - xlen;
	268
	269	/* check if the extend page is not abnr but the request is abnr
	270	* and the allocated disk space is for more than one page. if this
	271	* is the case, there is a miss match of abnr between the extend page
	272	* and the one or more pages following the extend page. as a result,
	273	* two extents will have to be manipulated. the first will be that
	274	* of the extent of the extend page and will be manipulated thru
	275	* an xtExtend() or an xtTailgate(), depending upon whether the
	276	* disk allocation occurred as an inplace extension. the second
	277	* extent will be manipulated (created) through an xtInsert() and
	278	* will be for the pages following the extend page.
	279	*/
	280	if (abnr && (!(xp->flag & XAD_NOTRECORDED)) && (nxlen > nbperpage)) {
	281	ntail = nbperpage;
	282	nextend = ntail - xlen;
	283	ninsert = nxlen - nbperpage;
	284
	285	xflag = XAD_NOTRECORDED;
	286	} else {
	287	ntail = nxlen;
	288	nextend = delta;
	289	ninsert = 0;
	290
	291	xflag = xp->flag;
	292	}
	293
	294	/* if we were able to extend the disk allocation in place,
	295	* extend the extent. otherwise, move the extent to a
	296	* new disk location.
	297	*/
	298	if (xaddr == nxaddr) {
	299	/* extend the extent */
	300	if ((rc = xtExtend(0, ip, xoff + xlen, (int) nextend, 0))) {
	301	dbFree(ip, xaddr + xlen, delta);
5dd4056d	302	dquot_free_block(ip, nxlen);
1da177e4 LT	303	goto exit;
	304	}
	305	} else {
	306	/*
	307	* move the extent to a new location:
	308	*
	309	* xtTailgate() accounts for relocated tail extent;
	310	*/
	311	if ((rc = xtTailgate(0, ip, xoff, (int) ntail, nxaddr, 0))) {
	312	dbFree(ip, nxaddr, nxlen);
5dd4056d	313	dquot_free_block(ip, nxlen);
1da177e4 LT	314	goto exit;
	315	}
	316	}
	317
	318
	319	/* check if we need to also insert a new extent */
	320	if (ninsert) {
	321	/* perform the insert. if it fails, free the blocks
	322	* to be inserted and make it appear that we only did
	323	* the xtExtend() or xtTailgate() above.
	324	*/
	325	xaddr = nxaddr + ntail;
	326	if (xtInsert (0, ip, xflag, xoff + ntail, (int) ninsert,
	327	&xaddr, 0)) {
	328	dbFree(ip, xaddr, (s64) ninsert);
	329	delta = nextend;
	330	nxlen = ntail;
	331	xflag = 0;
	332	}
	333	}
	334
	335	/* set the return results */
	336	XADaddress(xp, nxaddr);
	337	XADlength(xp, nxlen);
	338	XADoffset(xp, xoff);
	339	xp->flag = xflag;
	340
	341	mark_inode_dirty(ip);
	342	exit:
1de87444	343	mutex_unlock(&JFS_IP(ip)->commit_mutex);
1da177e4 LT	344	return (rc);
	345	}
	346	#endif /* _NOTYET */
	347
	348
	349	/*
f720e3ba	350	* NAME: extHint()
1da177e4	351	*
f720e3ba	352	* FUNCTION: produce an extent allocation hint for a file offset.
1da177e4 LT	353	*
	354	* PARAMETERS:
	355	* ip - the inode of the file.
	356	* offset - file offset for which the hint is needed.
	357	* xp - pointer to the xad that is to be filled in with
	358	* the hint.
	359	*
	360	* RETURN VALUES:
f720e3ba DK	361	* 0 - success
f720e3ba DK	362	* -EIO - i/o error.
1da177e4 LT	363	*/
	364	int extHint(struct inode ip, s64 offset, xad_t xp)
	365	{
	366	struct super_block *sb = ip->i_sb;
fec1878f	367	int nbperpage = JFS_SBI(sb)->nbperpage;
1da177e4	368	s64 prev;
fec1878f DK	369	int rc = 0;
	370	s64 xaddr;
	371	int xlen;
	372	int xflag;
1da177e4 LT	373
	374	/* init the hint as "no hint provided" */
	375	XADaddress(xp, 0);
	376
	377	/* determine the starting extent offset of the page previous
	378	* to the page containing the offset.
	379	*/
	380	prev = ((offset & ~POFFSET) >> JFS_SBI(sb)->l2bsize) - nbperpage;
	381
fec1878f	382	/* if the offset is in the first page of the file, no hint provided.
1da177e4 LT	383	*/
1da177e4 LT	384	if (prev < 0)
fec1878f	385	goto out;
1da177e4	386
fec1878f	387	rc = xtLookup(ip, prev, nbperpage, &xflag, &xaddr, &xlen, 0);
1da177e4	388
fec1878f DK	389	if ((rc == 0) && xlen) {
	390	if (xlen != nbperpage) {
	391	jfs_error(ip->i_sb, "extHint: corrupt xtree");
	392	rc = -EIO;
	393	}
	394	XADaddress(xp, xaddr);
	395	XADlength(xp, xlen);
f7c52fd1	396	XADoffset(xp, prev);
fec1878f DK	397	/*
	398	* only preserve the abnr flag within the xad flags
	399	* of the returned hint.
	400	*/
	401	xp->flag = xflag & XAD_NOTRECORDED;
	402	} else
	403	rc = 0;
1da177e4	404
fec1878f DK	405	out:
fec1878f DK	406	return (rc);
1da177e4 LT	407	}
	408
	409
	410	/*
f720e3ba	411	* NAME: extRecord()
1da177e4	412	*
f720e3ba	413	* FUNCTION: change a page with a file from not recorded to recorded.
1da177e4 LT	414	*
	415	* PARAMETERS:
	416	* ip - inode of the file.
	417	* cp - cbuf of the file page.
	418	*
	419	* RETURN VALUES:
f720e3ba DK	420	* 0 - success
	421	* -EIO - i/o error.
	422	* -ENOSPC - insufficient disk resources.
1da177e4 LT	423	*/
	424	int extRecord(struct inode ip, xad_t xp)
	425	{
	426	int rc;
	427
	428	txBeginAnon(ip->i_sb);
	429
1de87444	430	mutex_lock(&JFS_IP(ip)->commit_mutex);
1da177e4 LT	431
	432	/* update the extent */
	433	rc = xtUpdate(0, ip, xp);
	434
1de87444	435	mutex_unlock(&JFS_IP(ip)->commit_mutex);
1da177e4 LT	436	return rc;
	437	}
	438
	439
	440	#ifdef _NOTYET
	441	/*
f720e3ba	442	* NAME: extFill()
1da177e4	443	*
f720e3ba	444	* FUNCTION: allocate disk space for a file page that represents
1da177e4 LT	445	* a file hole.
	446	*
	447	* PARAMETERS:
	448	* ip - the inode of the file.
	449	* cp - cbuf of the file page represent the hole.
	450	*
	451	* RETURN VALUES:
f720e3ba DK	452	* 0 - success
	453	* -EIO - i/o error.
	454	* -ENOSPC - insufficient disk resources.
1da177e4 LT	455	*/
	456	int extFill(struct inode ip, xad_t xp)
	457	{
	458	int rc, nbperpage = JFS_SBI(ip->i_sb)->nbperpage;
ba52de12	459	s64 blkno = offsetXAD(xp) >> ip->i_blkbits;
1da177e4	460
f720e3ba	461	// assert(ISSPARSE(ip));
1da177e4 LT	462
	463	/* initialize the extent allocation hint */
	464	XADaddress(xp, 0);
	465
	466	/* allocate an extent to fill the hole */
4d81715f	467	if ((rc = extAlloc(ip, nbperpage, blkno, xp, false)))
1da177e4 LT	468	return (rc);
	469
	470	assert(lengthPXD(xp) == nbperpage);
	471
	472	return (0);
	473	}
	474	#endif /* _NOTYET */
	475
	476
	477	/*
	478	* NAME: extBalloc()
	479	*
f720e3ba	480	* FUNCTION: allocate disk blocks to form an extent.
1da177e4 LT	481	*
1da177e4 LT	482	* initially, we will try to allocate disk blocks for the
63f83c9f	483	* requested size (nblocks). if this fails (nblocks
25985edc	484	* contiguous free blocks not available), we'll try to allocate
1da177e4 LT	485	* a smaller number of blocks (producing a smaller extent), with
	486	* this smaller number of blocks consisting of the requested
	487	* number of blocks rounded down to the next smaller power of 2
	488	* number (i.e. 16 -> 8). we'll continue to round down and
	489	* retry the allocation until the number of blocks to allocate
	490	* is smaller than the number of blocks per page.
63f83c9f	491	*
1da177e4 LT	492	* PARAMETERS:
	493	* ip - the inode of the file.
	494	* hint - disk block number to be used as an allocation hint.
	495	* *nblocks - pointer to an s64 value. on entry, this value specifies
	496	* the desired number of block to be allocated. on successful
	497	* exit, this value is set to the number of blocks actually
	498	* allocated.
	499	* blkno - pointer to a block address that is filled in on successful
63f83c9f	500	* return with the starting block number of the newly
1da177e4 LT	501	* allocated block range.
	502	*
	503	* RETURN VALUES:
f720e3ba DK	504	* 0 - success
	505	* -EIO - i/o error.
	506	* -ENOSPC - insufficient disk resources.
1da177e4 LT	507	*/
	508	static int
	509	extBalloc(struct inode ip, s64 hint, s64 nblocks, s64 * blkno)
	510	{
	511	struct jfs_inode_info *ji = JFS_IP(ip);
	512	struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
	513	s64 nb, nblks, daddr, max;
	514	int rc, nbperpage = sbi->nbperpage;
	515	struct bmap *bmp = sbi->bmap;
	516	int ag;
	517
	518	/* get the number of blocks to initially attempt to allocate.
	519	* we'll first try the number of blocks requested unless this
d6e05edc	520	* number is greater than the maximum number of contiguous free
63f83c9f	521	* blocks in the map. in that case, we'll start off with the
1da177e4 LT	522	* maximum free.
	523	*/
	524	max = (s64) 1 << bmp->db_maxfreebud;
	525	if (nblocks >= max && nblocks > nbperpage)
	526	nb = nblks = (max > nbperpage) ? max : nbperpage;
	527	else
	528	nb = nblks = *nblocks;
	529
	530	/* try to allocate blocks */
	531	while ((rc = dbAlloc(ip, hint, nb, &daddr)) != 0) {
	532	/* if something other than an out of space error,
	533	* stop and return this error.
	534	*/
	535	if (rc != -ENOSPC)
	536	return (rc);
	537
	538	/* decrease the allocation request size */
	539	nb = min(nblks, extRoundDown(nb));
	540
	541	/* give up if we cannot cover a page */
	542	if (nb < nbperpage)
	543	return (rc);
	544	}
	545
	546	*nblocks = nb;
	547	*blkno = daddr;
	548
	549	if (S_ISREG(ip->i_mode) && (ji->fileset == FILESYSTEM_I)) {
	550	ag = BLKTOAG(daddr, sbi);
	551	spin_lock_irq(&ji->ag_lock);
	552	if (ji->active_ag == -1) {
	553	atomic_inc(&bmp->db_active[ag]);
	554	ji->active_ag = ag;
	555	} else if (ji->active_ag != ag) {
	556	atomic_dec(&bmp->db_active[ji->active_ag]);
	557	atomic_inc(&bmp->db_active[ag]);
	558	ji->active_ag = ag;
	559	}
	560	spin_unlock_irq(&ji->ag_lock);
	561	}
	562
	563	return (0);
	564	}
	565
	566
	567	#ifdef _NOTYET
	568	/*
	569	* NAME: extBrealloc()
	570	*
f720e3ba	571	* FUNCTION: attempt to extend an extent's allocation.
1da177e4	572	*
63f83c9f DK	573	* Initially, we will try to extend the extent's allocation
	574	* in place. If this fails, we'll try to move the extent
	575	* to a new set of blocks. If moving the extent, we initially
1da177e4	576	* will try to allocate disk blocks for the requested size
63f83c9f	577	* (newnblks). if this fails (new contiguous free blocks not
25985edc	578	* available), we'll try to allocate a smaller number of
1da177e4 LT	579	* blocks (producing a smaller extent), with this smaller
	580	* number of blocks consisting of the requested number of
	581	* blocks rounded down to the next smaller power of 2
63f83c9f	582	* number (i.e. 16 -> 8). We'll continue to round down and
1da177e4 LT	583	* retry the allocation until the number of blocks to allocate
1da177e4 LT	584	* is smaller than the number of blocks per page.
63f83c9f	585	*
1da177e4 LT	586	* PARAMETERS:
1da177e4 LT	587	* ip - the inode of the file.
f720e3ba DK	588	* blkno - starting block number of the extents current allocation.
f720e3ba DK	589	* nblks - number of blocks within the extents current allocation.
1da177e4 LT	590	* newnblks - pointer to a s64 value. on entry, this value is the
	591	* the new desired extent size (number of blocks). on
	592	* successful exit, this value is set to the extent's actual
	593	* new size (new number of blocks).
	594	* newblkno - the starting block number of the extents new allocation.
	595	*
	596	* RETURN VALUES:
f720e3ba DK	597	* 0 - success
	598	* -EIO - i/o error.
	599	* -ENOSPC - insufficient disk resources.
1da177e4 LT	600	*/
	601	static int
	602	extBrealloc(struct inode *ip,
	603	s64 blkno, s64 nblks, s64 * newnblks, s64 * newblkno)
	604	{
	605	int rc;
	606
	607	/* try to extend in place */
	608	if ((rc = dbExtend(ip, blkno, nblks, *newnblks - nblks)) == 0) {
	609	*newblkno = blkno;
	610	return (0);
	611	} else {
	612	if (rc != -ENOSPC)
	613	return (rc);
	614	}
	615
63f83c9f	616	/* in place extension not possible.
1da177e4 LT	617	* try to move the extent to a new set of blocks.
	618	*/
	619	return (extBalloc(ip, blkno, newnblks, newblkno));
	620	}
	621	#endif /* _NOTYET */
	622
	623
	624	/*
f720e3ba	625	* NAME: extRoundDown()
1da177e4	626	*
f720e3ba	627	* FUNCTION: round down a specified number of blocks to the next
1da177e4 LT	628	* smallest power of 2 number.
	629	*
	630	* PARAMETERS:
	631	* nb - the inode of the file.
	632	*
	633	* RETURN VALUES:
f720e3ba	634	* next smallest power of 2 number.
1da177e4 LT	635	*/
	636	static s64 extRoundDown(s64 nb)
	637	{
	638	int i;
	639	u64 m, k;
	640
	641	for (i = 0, m = (u64) 1 << 63; i < 64; i++, m >>= 1) {
	642	if (m & nb)
	643	break;
	644	}
	645
	646	i = 63 - i;
	647	k = (u64) 1 << i;
	648	k = ((k - 1) & nb) ? k : k >> 1;
	649
	650	return (k);
	651	}