cregit-Linux how code gets into the kernel

Release 4.10 fs/jfs/jfs_dmap.c

Directory: fs/jfs
/*
 *   Copyright (C) International Business Machines Corp., 2000-2004
 *   Portions Copyright (C) Tino Reichardt, 2012
 *
 *   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/slab.h>
#include "jfs_incore.h"
#include "jfs_superblock.h"
#include "jfs_dmap.h"
#include "jfs_imap.h"
#include "jfs_lock.h"
#include "jfs_metapage.h"
#include "jfs_debug.h"
#include "jfs_discard.h"

/*
 *      SERIALIZATION of the Block Allocation Map.
 *
 *      the working state of the block allocation map is accessed in
 *      two directions:
 *
 *      1) allocation and free requests that start at the dmap
 *         level and move up through the dmap control pages (i.e.
 *         the vast majority of requests).
 *
 *      2) allocation requests that start at dmap control page
 *         level and work down towards the dmaps.
 *
 *      the serialization scheme used here is as follows.
 *
 *      requests which start at the bottom are serialized against each
 *      other through buffers and each requests holds onto its buffers
 *      as it works it way up from a single dmap to the required level
 *      of dmap control page.
 *      requests that start at the top are serialized against each other
 *      and request that start from the bottom by the multiple read/single
 *      write inode lock of the bmap inode. requests starting at the top
 *      take this lock in write mode while request starting at the bottom
 *      take the lock in read mode.  a single top-down request may proceed
 *      exclusively while multiple bottoms-up requests may proceed
 *      simultaneously (under the protection of busy buffers).
 *
 *      in addition to information found in dmaps and dmap control pages,
 *      the working state of the block allocation map also includes read/
 *      write information maintained in the bmap descriptor (i.e. total
 *      free block count, allocation group level free block counts).
 *      a single exclusive lock (BMAP_LOCK) is used to guard this information
 *      in the face of multiple-bottoms up requests.
 *      (lock ordering: IREAD_LOCK, BMAP_LOCK);
 *
 *      accesses to the persistent state of the block allocation map (limited
 *      to the persistent bitmaps in dmaps) is guarded by (busy) buffers.
 */


#define BMAP_LOCK_INIT(bmp)	mutex_init(&bmp->db_bmaplock)

#define BMAP_LOCK(bmp)		mutex_lock(&bmp->db_bmaplock)

#define BMAP_UNLOCK(bmp)	mutex_unlock(&bmp->db_bmaplock)

/*
 * forward references
 */
static void dbAllocBits(struct bmap * bmp, struct dmap * dp, s64 blkno,
			int nblocks);
static void dbSplit(dmtree_t * tp, int leafno, int splitsz, int newval);
static int dbBackSplit(dmtree_t * tp, int leafno);
static int dbJoin(dmtree_t * tp, int leafno, int newval);
static void dbAdjTree(dmtree_t * tp, int leafno, int newval);
static int dbAdjCtl(struct bmap * bmp, s64 blkno, int newval, int alloc,
		    int level);
static int dbAllocAny(struct bmap * bmp, s64 nblocks, int l2nb, s64 * results);
static int dbAllocNext(struct bmap * bmp, struct dmap * dp, s64 blkno,
		       int nblocks);
static int dbAllocNear(struct bmap * bmp, struct dmap * dp, s64 blkno,
		       int nblocks,
		       int l2nb, s64 * results);
static int dbAllocDmap(struct bmap * bmp, struct dmap * dp, s64 blkno,
		       int nblocks);
static int dbAllocDmapLev(struct bmap * bmp, struct dmap * dp, int nblocks,
			  int l2nb,
			  s64 * results);
static int dbAllocAG(struct bmap * bmp, int agno, s64 nblocks, int l2nb,
		     s64 * results);
static int dbAllocCtl(struct bmap * bmp, s64 nblocks, int l2nb, s64 blkno,
		      s64 * results);
static int dbExtend(struct inode *ip, s64 blkno, s64 nblocks, s64 addnblocks);
static int dbFindBits(u32 word, int l2nb);
static int dbFindCtl(struct bmap * bmp, int l2nb, int level, s64 * blkno);
static int dbFindLeaf(dmtree_t * tp, int l2nb, int *leafidx);
static int dbFreeBits(struct bmap * bmp, struct dmap * dp, s64 blkno,
		      int nblocks);
static int dbFreeDmap(struct bmap * bmp, struct dmap * dp, s64 blkno,
		      int nblocks);
static int dbMaxBud(u8 * cp);
static int blkstol2(s64 nb);

static int cntlz(u32 value);
static int cnttz(u32 word);

static int dbAllocDmapBU(struct bmap * bmp, struct dmap * dp, s64 blkno,
			 int nblocks);
static int dbInitDmap(struct dmap * dp, s64 blkno, int nblocks);
static int dbInitDmapTree(struct dmap * dp);
static int dbInitTree(struct dmaptree * dtp);
static int dbInitDmapCtl(struct dmapctl * dcp, int level, int i);
static int dbGetL2AGSize(s64 nblocks);

/*
 *      buddy table
 *
 * table used for determining buddy sizes within characters of
 * dmap bitmap words.  the characters themselves serve as indexes
 * into the table, with the table elements yielding the maximum
 * binary buddy of free bits within the character.
 */

static const s8 budtab[256] = {
	3, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
	2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
	2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
	2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
	2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
	2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
	2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0,
	2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, -1
};

/*
 * NAME:        dbMount()
 *
 * FUNCTION:    initializate the block allocation map.
 *
 *              memory is allocated for the in-core bmap descriptor and
 *              the in-core descriptor is initialized from disk.
 *
 * PARAMETERS:
 *      ipbmap  - pointer to in-core inode for the block map.
 *
 * RETURN VALUES:
 *      0       - success
 *      -ENOMEM - insufficient memory
 *      -EIO    - i/o error
 */



int dbMount(struct inode *ipbmap)
{
	struct bmap *bmp;
	struct dbmap_disk *dbmp_le;
	struct metapage *mp;
	int i;

	/*
         * allocate/initialize the in-memory bmap descriptor
         */
	/* allocate memory for the in-memory bmap descriptor */
	bmp = kmalloc(sizeof(struct bmap), GFP_KERNEL);
	if (bmp == NULL)
		return -ENOMEM;

	/* read the on-disk bmap descriptor. */
	mp = read_metapage(ipbmap,
			   BMAPBLKNO << JFS_SBI(ipbmap->i_sb)->l2nbperpage,
			   PSIZE, 0);
	if (mp == NULL) {
		kfree(bmp);
		return -EIO;
	}

	/* copy the on-disk bmap descriptor to its in-memory version. */
	dbmp_le = (struct dbmap_disk *) mp->data;
	bmp->db_mapsize = le64_to_cpu(dbmp_le->dn_mapsize);
	bmp->db_nfree = le64_to_cpu(dbmp_le->dn_nfree);
	bmp->db_l2nbperpage = le32_to_cpu(dbmp_le->dn_l2nbperpage);
	bmp->db_numag = le32_to_cpu(dbmp_le->dn_numag);
	bmp->db_maxlevel = le32_to_cpu(dbmp_le->dn_maxlevel);
	bmp->db_maxag = le32_to_cpu(dbmp_le->dn_maxag);
	bmp->db_agpref = le32_to_cpu(dbmp_le->dn_agpref);
	bmp->db_aglevel = le32_to_cpu(dbmp_le->dn_aglevel);
	bmp->db_agheight = le32_to_cpu(dbmp_le->dn_agheight);
	bmp->db_agwidth = le32_to_cpu(dbmp_le->dn_agwidth);
	bmp->db_agstart = le32_to_cpu(dbmp_le->dn_agstart);
	bmp->db_agl2size = le32_to_cpu(dbmp_le->dn_agl2size);
	for (i = 0; i < MAXAG; i++)
		bmp->db_agfree[i] = le64_to_cpu(dbmp_le->dn_agfree[i]);
	bmp->db_agsize = le64_to_cpu(dbmp_le->dn_agsize);
	bmp->db_maxfreebud = dbmp_le->dn_maxfreebud;

	/* release the buffer. */
	release_metapage(mp);

	/* bind the bmap inode and the bmap descriptor to each other. */
	bmp->db_ipbmap = ipbmap;
	JFS_SBI(ipbmap->i_sb)->bmap = bmp;

	memset(bmp->db_active, 0, sizeof(bmp->db_active));

	/*
         * allocate/initialize the bmap lock
         */
	BMAP_LOCK_INIT(bmp);

	return (0);
}



Contributors
PersonTokensPropCommitsCommitProp
dave kleikampdave kleikamp33599.41%583.33%


daniel mackdaniel mack20.59%116.67%


Total337100.00%6100.00%



/*
 * NAME:        dbUnmount()
 *
 * FUNCTION:    terminate the block allocation map in preparation for
 *              file system unmount.
 *
 *              the in-core bmap descriptor is written to disk and
 *              the memory for this descriptor is freed.
 *
 * PARAMETERS:
 *      ipbmap  - pointer to in-core inode for the block map.
 *
 * RETURN VALUES:
 *      0       - success
 *      -EIO    - i/o error
 */



int dbUnmount(struct inode *ipbmap, int mounterror)
{
	struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap;

	if (!(mounterror || isReadOnly(ipbmap)))
		dbSync(ipbmap);

	/*
         * Invalidate the page cache buffers
         */
	truncate_inode_pages(ipbmap->i_mapping, 0);

	/* free the memory for the in-memory bmap. */
	kfree(bmp);

	return (0);
}


Contributors
PersonTokensPropCommitsCommitProp
dave kleikampdave kleikamp65100.00%3100.00%


Total65100.00%3100.00%



/*
 *      dbSync()
 */



int dbSync(struct inode *ipbmap)
{
	struct dbmap_disk *dbmp_le;
	struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap;
	struct metapage *mp;
	int i;

	/*
         * write bmap global control page
         */
	/* get the buffer for the on-disk bmap descriptor. */
	mp = read_metapage(ipbmap,
			   BMAPBLKNO << JFS_SBI(ipbmap->i_sb)->l2nbperpage,
			   PSIZE, 0);
	if (mp == NULL) {
		jfs_err("dbSync: read_metapage failed!");
		return -EIO;
	}
	/* copy the in-memory version of the bmap to the on-disk version */
	dbmp_le = (struct dbmap_disk *) mp->data;
	dbmp_le->dn_mapsize = cpu_to_le64(bmp->db_mapsize);
	dbmp_le->dn_nfree = cpu_to_le64(bmp->db_nfree);
	dbmp_le->dn_l2nbperpage = cpu_to_le32(bmp->db_l2nbperpage);
	dbmp_le->dn_numag = cpu_to_le32(bmp->db_numag);
	dbmp_le->dn_maxlevel = cpu_to_le32(bmp->db_maxlevel);
	dbmp_le->dn_maxag = cpu_to_le32(bmp->db_maxag);
	dbmp_le->dn_agpref = cpu_to_le32(bmp->db_agpref);
	dbmp_le->dn_aglevel = cpu_to_le32(bmp->db_aglevel);
	dbmp_le->dn_agheight = cpu_to_le32(bmp->db_agheight);
	dbmp_le->dn_agwidth = cpu_to_le32(bmp->db_agwidth);
	dbmp_le->dn_agstart = cpu_to_le32(bmp->db_agstart);
	dbmp_le->dn_agl2size = cpu_to_le32(bmp->db_agl2size);
	for (i = 0; i < MAXAG; i++)
		dbmp_le->dn_agfree[i] = cpu_to_le64(bmp->db_agfree[i]);
	dbmp_le->dn_agsize = cpu_to_le64(bmp->db_agsize);
	dbmp_le->dn_maxfreebud = bmp->db_maxfreebud;

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

	/*
         * write out dirty pages of bmap
         */
	filemap_write_and_wait(ipbmap->i_mapping);

	diWriteSpecial(ipbmap, 0);

	return (0);
}


Contributors
PersonTokensPropCommitsCommitProp
dave kleikampdave kleikamp29097.64%666.67%


andrew mortonandrew morton41.35%111.11%


daniel mackdaniel mack20.67%111.11%


hirofumi ogawahirofumi ogawa10.34%111.11%


Total297100.00%9100.00%



/*
 * NAME:        dbFree()
 *
 * FUNCTION:    free the specified block range from the working block
 *              allocation map.
 *
 *              the blocks will be free from the working map one dmap
 *              at a time.
 *
 * PARAMETERS:
 *      ip      - pointer to in-core inode;
 *      blkno   - starting block number to be freed.
 *      nblocks - number of blocks to be freed.
 *
 * RETURN VALUES:
 *      0       - success
 *      -EIO    - i/o error
 */



int dbFree(struct inode *ip, s64 blkno, s64 nblocks)
{
	struct metapage *mp;
	struct dmap *dp;
	int nb, rc;
	s64 lblkno, rem;
	struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
	struct bmap *bmp = JFS_SBI(ip->i_sb)->bmap;
	struct super_block *sb = ipbmap->i_sb;

	IREAD_LOCK(ipbmap, RDWRLOCK_DMAP);

	/* block to be freed better be within the mapsize. */
	if (unlikely((blkno == 0) || (blkno + nblocks > bmp->db_mapsize))) {
		IREAD_UNLOCK(ipbmap);
		printk(KERN_ERR "blkno = %Lx, nblocks = %Lx\n",
		       (unsigned long long) blkno,
		       (unsigned long long) nblocks);
		jfs_error(ip->i_sb, "block to be freed is outside the map\n");
		return -EIO;
	}

	/**
         * TRIM the blocks, when mounted with discard option
         */
	if (JFS_SBI(sb)->flag & JFS_DISCARD)
		if (JFS_SBI(sb)->minblks_trim <= nblocks)
			jfs_issue_discard(ipbmap, blkno, nblocks);

	/*
         * free the blocks a dmap at a time.
         */
	mp = NULL;
	for (rem = nblocks; rem > 0; rem -= nb, blkno += nb) {
		/* release previous dmap if any */
		if (mp) {
			write_metapage(mp);
		}

		/* get the buffer for the current dmap. */
		lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage);
		mp = read_metapage(ipbmap, lblkno, PSIZE, 0);
		if (mp == NULL) {
			IREAD_UNLOCK(ipbmap);
			return -EIO;
		}
		dp = (struct dmap *) mp->data;

		/* determine the number of blocks to be freed from
                 * this dmap.
                 */
		nb = min(rem, BPERDMAP - (blkno & (BPERDMAP - 1)));

		/* free the blocks. */
		if ((rc = dbFreeDmap(bmp, dp, blkno, nb))) {
			jfs_error(ip->i_sb, "error in block map\n");
			release_metapage(mp);
			IREAD_UNLOCK(ipbmap);
			return (rc);
		}
	}

	/* write the last buffer. */
	write_metapage(mp);

	IREAD_UNLOCK(ipbmap);

	return (0);
}



Contributors
PersonTokensPropCommitsCommitProp
dave kleikampdave kleikamp30187.50%777.78%


tino reichardttino reichardt4111.92%111.11%


joe perchesjoe perches20.58%111.11%


Total344100.00%9100.00%



/*
 * NAME:        dbUpdatePMap()
 *
 * FUNCTION:    update the allocation state (free or allocate) of the
 *              specified block range in the persistent block allocation map.
 *
 *              the blocks will be updated in the persistent map one
 *              dmap at a time.
 *
 * PARAMETERS:
 *      ipbmap  - pointer to in-core inode for the block map.
 *      free    - 'true' if block range is to be freed from the persistent
 *                map; 'false' if it is to be allocated.
 *      blkno   - starting block number of the range.
 *      nblocks - number of contiguous blocks in the range.
 *      tblk    - transaction block;
 *
 * RETURN VALUES:
 *      0       - success
 *      -EIO    - i/o error
 */


int

dbUpdatePMap(struct inode *ipbmap,
	     int free, s64 blkno, s64 nblocks, struct tblock * tblk)
{
	int nblks, dbitno, wbitno, rbits;
	int word, nbits, nwords;
	struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap;
	s64 lblkno, rem, lastlblkno;
	u32 mask;
	struct dmap *dp;
	struct metapage *mp;
	struct jfs_log *log;
	int lsn, difft, diffp;
	unsigned long flags;

	/* the blocks better be within the mapsize. */
	if (blkno + nblocks > bmp->db_mapsize) {
		printk(KERN_ERR "blkno = %Lx, nblocks = %Lx\n",
		       (unsigned long long) blkno,
		       (unsigned long long) nblocks);
		jfs_error(ipbmap->i_sb, "blocks are outside the map\n");
		return -EIO;
	}

	/* compute delta of transaction lsn from log syncpt */
	lsn = tblk->lsn;
	log = (struct jfs_log *) JFS_SBI(tblk->sb)->log;
	logdiff(difft, lsn, log);

	/*
         * update the block state a dmap at a time.
         */
	mp = NULL;
	lastlblkno = 0;
	for (rem = nblocks; rem > 0; rem -= nblks, blkno += nblks) {
		/* get the buffer for the current dmap. */
		lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage);
		if (lblkno != lastlblkno) {
			if (mp) {
				write_metapage(mp);
			}

			mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE,
					   0);
			if (mp == NULL)
				return -EIO;
			metapage_wait_for_io(mp);
		}
		dp = (struct dmap *) mp->data;

		/* determine the bit number and word within the dmap of
                 * the starting block.  also determine how many blocks
                 * are to be updated within this dmap.
                 */
		dbitno = blkno & (BPERDMAP - 1);
		word = dbitno >> L2DBWORD;
		nblks = min(rem, (s64)BPERDMAP - dbitno);

		/* update the bits of the dmap words. the first and last
                 * words may only have a subset of their bits updated. if
                 * this is the case, we'll work against that word (i.e.
                 * partial first and/or last) only in a single pass.  a
                 * single pass will also be used to update all words that
                 * are to have all their bits updated.
                 */
		for (rbits = nblks; rbits > 0;
		     rbits -= nbits, dbitno += nbits) {
			/* determine the bit number within the word and
                         * the number of bits within the word.
                         */
			wbitno = dbitno & (DBWORD - 1);
			nbits = min(rbits, DBWORD - wbitno);

			/* check if only part of the word is to be updated. */
			if (nbits < DBWORD) {
				/* update (free or allocate) the bits
                                 * in this word.
                                 */
				mask =
				    (ONES << (DBWORD - nbits) >> wbitno);
				if (free)
					dp->pmap[word] &=
					    cpu_to_le32(~mask);
				else
					dp->pmap[word] |=
					    cpu_to_le32(mask);

				word += 1;
			} else {
				/* one or more words are to have all
                                 * their bits updated.  determine how
                                 * many words and how many bits.
                                 */
				nwords = rbits >> L2DBWORD;
				nbits = nwords << L2DBWORD;

				/* update (free or allocate) the bits
                                 * in these words.
                                 */
				if (free)
					memset(&dp->pmap[word], 0,
					       nwords * 4);
				else
					memset(&dp->pmap[word], (int) ONES,
					       nwords * 4);

				word += nwords;
			}
		}

		/*
                 * update dmap lsn
                 */
		if (lblkno == lastlblkno)
			continue;

		lastlblkno = lblkno;

		LOGSYNC_LOCK(log, flags);
		if (mp->lsn != 0) {
			/* inherit older/smaller lsn */
			logdiff(diffp, mp->lsn, log);
			if (difft < diffp) {
				mp->lsn = lsn;

				/* move bp after tblock in logsync list */
				list_move(&mp->synclist, &tblk->synclist);
			}

			/* inherit younger/larger clsn */
			logdiff(difft, tblk->clsn, log);
			logdiff(diffp, mp->clsn, log);
			if (difft > diffp)
				mp->clsn = tblk->clsn;
		} else {
			mp->log = log;
			mp->lsn = lsn;

			/* insert bp after tblock in logsync list */
			log->count++;
			list_add(&mp->synclist, &tblk->synclist);

			mp->clsn = tblk->clsn;
		}
		LOGSYNC_UNLOCK(log, flags);
	}

	/* write the last buffer. */
	if (mp) {
		write_metapage(mp);
	}

	return (0);
}



Contributors
PersonTokensPropCommitsCommitProp
dave kleikampdave kleikamp63599.84%888.89%


joe perchesjoe perches10.16%111.11%


Total636100.00%9100.00%



/*
 * NAME:        dbNextAG()
 *
 * FUNCTION:    find the preferred allocation group for new allocations.
 *
 *              Within the allocation groups, we maintain a preferred
 *              allocation group which consists of a group with at least
 *              average free space.  It is the preferred group that we target
 *              new inode allocation towards.  The tie-in between inode
 *              allocation and block allocation occurs as we allocate the
 *              first (data) block of an inode and specify the inode (block)
 *              as the allocation hint for this block.
 *
 *              We try to avoid having more than one open file growing in
 *              an allocation group, as this will lead to fragmentation.
 *              This differs from the old OS/2 method of trying to keep
 *              empty ags around for large allocations.
 *
 * PARAMETERS:
 *      ipbmap  - pointer to in-core inode for the block map.
 *
 * RETURN VALUES:
 *      the preferred allocation group number.
 */



int dbNextAG(struct inode *ipbmap)
{
	s64 avgfree;
	int agpref;
	s64 hwm = 0;
	int i;
	int next_best = -1;
	struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap;

	BMAP_LOCK(bmp);

	/* determine the average number of free blocks within the ags. */
	avgfree = (u32)bmp->db_nfree / bmp->db_numag;

	/*
         * if the current preferred ag does not have an active allocator
         * and has at least average freespace, return it
         */
	agpref = bmp->db_agpref;
	if ((atomic_read(&bmp->db_active[agpref]) == 0) &&
	    (bmp->db_agfree[agpref] >= avgfree))
		goto unlock;

	/* From the last preferred ag, find the next one with at least
         * average free space.
         */
	for (i = 0 ; i < bmp->db_numag; i++, agpref++) {
		if (agpref == bmp->db_numag)
			agpref = 0;

		if (atomic_read(&bmp->db_active[agpref]))
			/* open file is currently growing in this ag */
			continue;
		if (bmp->db_agfree[agpref] >= avgfree) {
			/* Return this one */
			bmp->db_agpref = agpref;
			goto unlock;
		} else if (bmp->db_agfree[agpref] > hwm) {
			/* Less than avg. freespace, but best so far */
			hwm = bmp->db_agfree[agpref];
			next_best = agpref;
		}
	}

	/*
         * If no inactive ag was found with average freespace, use the
         * next best
         */
	if (next_best != -1)
		bmp->db_agpref = next_best;
	/* else leave db_agpref unchanged */
unlock:
	BMAP_UNLOCK(bmp);

	/* return the preferred group.
         */
	return (bmp->db_agpref);
}


Contributors
PersonTokensPropCommitsCommitProp
dave kleikampdave kleikamp230100.00%4100.00%


Total230100.00%4100.00%



/*
 * NAME:        dbAlloc()
 *
 * FUNCTION:    attempt to allocate a specified number of contiguous free
 *              blocks from the working allocation block map.
 *
 *              the block allocation policy uses hints and a multi-step
 *              approach.
 *
 *              for allocation requests smaller than the number of blocks
 *              per dmap, we first try to allocate the new blocks
 *              immediately following the hint.  if these blocks are not
 *              available, we try to allocate blocks near the hint.  if
 *              no blocks near the hint are available, we next try to
 *              allocate within the same dmap as contains the hint.
 *
 *              if no blocks are available in the dmap or the allocation
 *              request is larger than the dmap size, we try to allocate
 *              within the same allocation group as contains the hint. if
 *              this does not succeed, we finally try to allocate anywhere
 *              within the aggregate.
 *
 *              we also try to allocate anywhere within the aggregate for
 *              for allocation requests larger than the allocation group
 *              size or requests that specify no hint value.
 *
 * PARAMETERS:
 *      ip      - pointer to in-core inode;
 *      hint    - allocation hint.
 *      nblocks - number of contiguous blocks in the range.
 *      results - on successful return, set to the starting block number
 *                of the newly allocated contiguous range.
 *
 * RETURN VALUES:
 *      0       - success
 *      -ENOSPC - insufficient disk resources
 *      -EIO    - i/o error
 */



int dbAlloc(struct inode *ip, s64 hint, s64 nblocks, s64 * results)
{
	int rc, agno;
	struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
	struct bmap *bmp;
	struct metapage *mp;
	s64 lblkno, blkno;
	struct dmap *dp;
	int l2nb;
	s64 mapSize;
	int writers;

	/* assert that nblocks is valid */
	assert(nblocks > 0);

	/* get the log2 number of blocks to be allocated.
         * if the number of blocks is not a log2 multiple,
         * it will be rounded up to the next log2 multiple.
         */
	l2nb = BLKSTOL2(nblocks);

	bmp = JFS_SBI(ip->i_sb)->bmap;

	mapSize = bmp->db_mapsize;

	/* the hint should be within the map */
	if (hint >= mapSize) {
		jfs_error(ip->i_sb, "the hint is outside the map\n");
		return -EIO;
	}

	/* if the number of blocks to be allocated is greater than the
         * allocation group size, try to allocate anywhere.
         */
	if (l2nb > bmp->db_agl2size) {
		IWRITE_LOCK(ipbmap, RDWRLOCK_DMAP);

		rc = dbAllocAny(bmp, nblocks, l2nb, results);