int
cgbfree(struct uufsd *disk, ufs2_daddr_t bno, long size)
{
u_int8_t *blksfree;
struct fs *fs;
struct cg *cgp;
ufs1_daddr_t fragno, cgbno;
int i, cg, blk, frags, bbase;
fs = &disk->d_fs;
cg = dtog(fs, bno);
if (cgread1(disk, cg) != 1)
return (-1);
cgp = &disk->d_cg;
cgbno = dtogd(fs, bno);
blksfree = cg_blksfree(cgp);
if (size == fs->fs_bsize) {
fragno = fragstoblks(fs, cgbno);
ffs_setblock(fs, blksfree, fragno);
ffs_clusteracct(fs, cgp, fragno, 1);
cgp->cg_cs.cs_nbfree++;
fs->fs_cstotal.cs_nbfree++;
fs->fs_cs(fs, cg).cs_nbfree++;
} else {
bbase = cgbno - fragnum(fs, cgbno);
/*
* decrement the counts associated with the old frags
*/
blk = blkmap(fs, blksfree, bbase);
ffs_fragacct(fs, blk, cgp->cg_frsum, -1);
/*
* deallocate the fragment
*/
frags = numfrags(fs, size);
for (i = 0; i < frags; i++)
setbit(blksfree, cgbno + i);
cgp->cg_cs.cs_nffree += i;
fs->fs_cstotal.cs_nffree += i;
fs->fs_cs(fs, cg).cs_nffree += i;
/*
* add back in counts associated with the new frags
*/
blk = blkmap(fs, blksfree, bbase);
ffs_fragacct(fs, blk, cgp->cg_frsum, 1);
/*
* if a complete block has been reassembled, account for it
*/
fragno = fragstoblks(fs, bbase);
if (ffs_isblock(fs, blksfree, fragno)) {
cgp->cg_cs.cs_nffree -= fs->fs_frag;
fs->fs_cstotal.cs_nffree -= fs->fs_frag;
fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
ffs_clusteracct(fs, cgp, fragno, 1);
cgp->cg_cs.cs_nbfree++;
fs->fs_cstotal.cs_nbfree++;
fs->fs_cs(fs, cg).cs_nbfree++;
}
}
return cgwrite(disk);
}
/*
* Allocate a block in a cylinder group.
*
* This algorithm implements the following policy:
* 1) allocate the requested block.
* 2) allocate a rotationally optimal block in the same cylinder.
* 3) allocate the next available block on the block rotor for the
* specified cylinder group.
* Note that this routine only allocates fs_bsize blocks; these
* blocks may be fragmented by the routine that allocates them.
*/
static daddr_t
ffs_alloccgblk(struct inode *ip, struct buf *bp, daddr_t bpref)
{
struct cg *cgp;
daddr_t blkno;
int32_t bno;
struct fs *fs = ip->i_fs;
const int needswap = UFS_FSNEEDSWAP(fs);
u_int8_t *blksfree;
cgp = (struct cg *)bp->b_data;
blksfree = cg_blksfree(cgp, needswap);
if (bpref == 0 || dtog(fs, bpref) != ufs_rw32(cgp->cg_cgx, needswap)) {
bpref = ufs_rw32(cgp->cg_rotor, needswap);
} else {
bpref = ffs_blknum(fs, bpref);
bno = dtogd(fs, bpref);
/*
* if the requested block is available, use it
*/
if (ffs_isblock(fs, blksfree, ffs_fragstoblks(fs, bno)))
goto gotit;
}
/*
* Take the next available one in this cylinder group.
*/
bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag);
if (bno < 0)
return (0);
cgp->cg_rotor = ufs_rw32(bno, needswap);
gotit:
blkno = ffs_fragstoblks(fs, bno);
ffs_clrblock(fs, blksfree, (long)blkno);
ffs_clusteracct(fs, cgp, blkno, -1);
ufs_add32(cgp->cg_cs.cs_nbfree, -1, needswap);
fs->fs_cstotal.cs_nbfree--;
fs->fs_cs(fs, ufs_rw32(cgp->cg_cgx, needswap)).cs_nbfree--;
fs->fs_fmod = 1;
blkno = ufs_rw32(cgp->cg_cgx, needswap) * fs->fs_fpg + bno;
return (blkno);
}
ufs2_daddr_t
cgballoc(struct uufsd *disk)
{
u_int8_t *blksfree;
struct cg *cgp;
struct fs *fs;
long bno;
fs = &disk->d_fs;
cgp = &disk->d_cg;
blksfree = cg_blksfree(cgp);
for (bno = 0; bno < fs->fs_fpg / fs->fs_frag; bno++)
if (ffs_isblock(fs, blksfree, bno))
goto gotit;
return (0);
gotit:
fs->fs_cs(fs, cgp->cg_cgx).cs_nbfree--;
ffs_clrblock(fs, blksfree, (long)bno);
ffs_clusteracct(fs, cgp, bno, -1);
cgp->cg_cs.cs_nbfree--;
fs->fs_cstotal.cs_nbfree--;
fs->fs_fmod = 1;
return (cgbase(fs, cgp->cg_cgx) + blkstofrags(fs, bno));
}
/*
* Free a block or fragment.
*
* The specified block or fragment is placed back in the
* free map. If a fragment is deallocated, a possible
* block reassembly is checked.
*/
void
ffs_blkfree(struct inode *ip, daddr_t bno, long size)
{
struct cg *cgp;
struct buf *bp;
int32_t fragno, cgbno;
int i, error, cg, blk, frags, bbase;
struct fs *fs = ip->i_fs;
const int needswap = UFS_FSNEEDSWAP(fs);
if (size > fs->fs_bsize || ffs_fragoff(fs, size) != 0 ||
ffs_fragnum(fs, bno) + ffs_numfrags(fs, size) > fs->fs_frag) {
errx(1, "blkfree: bad size: bno %lld bsize %d size %ld",
(long long)bno, fs->fs_bsize, size);
}
cg = dtog(fs, bno);
if (bno >= fs->fs_size) {
warnx("bad block %lld, ino %llu", (long long)bno,
(unsigned long long)ip->i_number);
return;
}
error = bread(ip->i_devvp, FFS_FSBTODB(fs, cgtod(fs, cg)),
(int)fs->fs_cgsize, 0, &bp);
if (error) {
brelse(bp, 0);
return;
}
cgp = (struct cg *)bp->b_data;
if (!cg_chkmagic(cgp, needswap)) {
brelse(bp, 0);
return;
}
cgbno = dtogd(fs, bno);
if (size == fs->fs_bsize) {
fragno = ffs_fragstoblks(fs, cgbno);
if (!ffs_isfreeblock(fs, cg_blksfree(cgp, needswap), fragno)) {
errx(1, "blkfree: freeing free block %lld",
(long long)bno);
}
ffs_setblock(fs, cg_blksfree(cgp, needswap), fragno);
ffs_clusteracct(fs, cgp, fragno, 1);
ufs_add32(cgp->cg_cs.cs_nbfree, 1, needswap);
fs->fs_cstotal.cs_nbfree++;
fs->fs_cs(fs, cg).cs_nbfree++;
} else {
bbase = cgbno - ffs_fragnum(fs, cgbno);
/*
* decrement the counts associated with the old frags
*/
blk = blkmap(fs, cg_blksfree(cgp, needswap), bbase);
ffs_fragacct(fs, blk, cgp->cg_frsum, -1, needswap);
/*
* deallocate the fragment
*/
frags = ffs_numfrags(fs, size);
for (i = 0; i < frags; i++) {
if (isset(cg_blksfree(cgp, needswap), cgbno + i)) {
errx(1, "blkfree: freeing free frag: block %lld",
(long long)(cgbno + i));
}
setbit(cg_blksfree(cgp, needswap), cgbno + i);
}
ufs_add32(cgp->cg_cs.cs_nffree, i, needswap);
fs->fs_cstotal.cs_nffree += i;
fs->fs_cs(fs, cg).cs_nffree += i;
/*
* add back in counts associated with the new frags
*/
blk = blkmap(fs, cg_blksfree(cgp, needswap), bbase);
ffs_fragacct(fs, blk, cgp->cg_frsum, 1, needswap);
/*
* if a complete block has been reassembled, account for it
*/
fragno = ffs_fragstoblks(fs, bbase);
if (ffs_isblock(fs, cg_blksfree(cgp, needswap), fragno)) {
ufs_add32(cgp->cg_cs.cs_nffree, -fs->fs_frag, needswap);
fs->fs_cstotal.cs_nffree -= fs->fs_frag;
fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
ffs_clusteracct(fs, cgp, fragno, 1);
ufs_add32(cgp->cg_cs.cs_nbfree, 1, needswap);
fs->fs_cstotal.cs_nbfree++;
fs->fs_cs(fs, cg).cs_nbfree++;
}
}
fs->fs_fmod = 1;
bdwrite(bp);
}
/*
* Find a suitable location for the journal in the filesystem.
*
* Our strategy here is to look for a contiguous block of free space
* at least "logfile" MB in size (plus room for any indirect blocks).
* We start at the middle of the filesystem and check each cylinder
* group working outwards. If "logfile" MB is not available as a
* single contigous chunk, then return the address and size of the
* largest chunk found.
*
* XXX
* At what stage does the search fail? Is if the largest space we could
* find is less than a quarter the requested space reasonable? If the
* search fails entirely, return a block address if "0" it indicate this.
*/
void
wapbl_find_log_start(struct mount *mp, struct vnode *vp, off_t logsize,
daddr_t *addr, daddr_t *indir_addr, size_t *size)
{
struct ufsmount *ump = VFSTOUFS(mp);
struct fs *fs = ump->um_fs;
struct vnode *devvp = ump->um_devvp;
struct cg *cgp;
struct buf *bp;
uint8_t *blksfree;
daddr_t blkno, best_addr, start_addr;
daddr_t desired_blks, min_desired_blks;
daddr_t freeblks, best_blks;
int bpcg, cg, error, fixedsize, indir_blks, n, s;
#ifdef FFS_EI
const int needswap = UFS_FSNEEDSWAP(fs);
#endif
if (logsize == 0) {
fixedsize = 0; /* We can adjust the size if tight */
logsize = lfragtosize(fs, fs->fs_dsize) /
UFS_WAPBL_JOURNAL_SCALE;
DPRINTF("suggested log size = %lld\n", logsize);
logsize = max(logsize, UFS_WAPBL_MIN_JOURNAL_SIZE);
logsize = min(logsize, UFS_WAPBL_MAX_JOURNAL_SIZE);
DPRINTF("adjusted log size = %lld\n", logsize);
} else {
fixedsize = 1;
DPRINTF("fixed log size = %lld\n", logsize);
}
desired_blks = logsize / fs->fs_bsize;
DPRINTF("desired blocks = %lld\n", desired_blks);
/* add in number of indirect blocks needed */
indir_blks = 0;
if (desired_blks >= NDADDR) {
struct indir indirs[NIADDR + 2];
int num;
error = ufs_getlbns(vp, desired_blks, indirs, &num);
if (error) {
printf("%s: ufs_getlbns failed, error %d!\n",
__func__, error);
goto bad;
}
switch (num) {
case 2:
indir_blks = 1; /* 1st level indirect */
break;
case 3:
indir_blks = 1 + /* 1st level indirect */
1 + /* 2nd level indirect */
indirs[1].in_off + 1; /* extra 1st level indirect */
break;
default:
printf("%s: unexpected numlevels %d from ufs_getlbns\n",
__func__, num);
*size = 0;
goto bad;
}
desired_blks += indir_blks;
}
DPRINTF("desired blocks = %lld (including indirect)\n",
desired_blks);
/*
* If a specific size wasn't requested, allow for a smaller log
* if we're really tight for space...
*/
min_desired_blks = desired_blks;
if (!fixedsize)
min_desired_blks = desired_blks / 4;
/* Look at number of blocks per CG. If it's too small, bail early. */
bpcg = fragstoblks(fs, fs->fs_fpg);
if (min_desired_blks > bpcg) {
printf("ffs_wapbl: cylinder group size of %lld MB "
" is not big enough for journal\n",
lblktosize(fs, bpcg) / (1024 * 1024));
goto bad;
}
/*
* Start with the middle cylinder group, and search outwards in
* both directions until we either find the requested log size
* or reach the start/end of the file system. If we reach the
* start/end without finding enough space for the full requested
* log size, use the largest extent found if it is large enough
* to satisfy the our minimum size.
*
* XXX
* Can we just use the cluster contigsum stuff (esp on UFS2)
* here to simplify this search code?
*/
best_addr = 0;
best_blks = 0;
for (cg = fs->fs_ncg / 2, s = 0, n = 1;
best_blks < desired_blks && cg >= 0 && cg < fs->fs_ncg;
s++, n = -n, cg += n * s) {
DPRINTF("check cg %d of %d\n", cg, fs->fs_ncg);
error = bread(devvp, fsbtodb(fs, cgtod(fs, cg)),
fs->fs_cgsize, &bp);
if (error) {
continue;
}
cgp = (struct cg *)bp->b_data;
if (!cg_chkmagic(cgp)) {
brelse(bp);
continue;
}
blksfree = cg_blksfree(cgp);
for (blkno = 0; blkno < bpcg;) {
/* look for next free block */
/* XXX use scanc() and fragtbl[] here? */
for (; blkno < bpcg - min_desired_blks; blkno++)
if (ffs_isblock(fs, blksfree, blkno))
break;
/* past end of search space in this CG? */
if (blkno >= bpcg - min_desired_blks)
break;
/* count how many free blocks in this extent */
start_addr = blkno;
for (freeblks = 0; blkno < bpcg; blkno++, freeblks++)
if (!ffs_isblock(fs, blksfree, blkno))
break;
if (freeblks > best_blks) {
best_blks = freeblks;
best_addr = blkstofrags(fs, start_addr) +
cgbase(fs, cg);
if (freeblks >= desired_blks) {
DPRINTF("found len %lld"
" at offset %lld in gc\n",
freeblks, start_addr);
break;
}
}
}
brelse(bp);
}
DPRINTF("best found len = %lld, wanted %lld"
" at addr %lld\n", best_blks, desired_blks, best_addr);
if (best_blks < min_desired_blks) {
*addr = 0;
*indir_addr = 0;
} else {
/* put indirect blocks at start, and data blocks after */
*addr = best_addr + blkstofrags(fs, indir_blks);
*indir_addr = best_addr;
}
*size = min(desired_blks, best_blks) - indir_blks;
return;
bad:
*addr = 0;
*indir_addr = 0;
*size = 0;
return;
}