/*
* ffs_balloc(struct vnode *a_vp, ufs_daddr_t a_lbn, int a_size,
* struct ucred *a_cred, int a_flags, struct buf *a_bpp)
*
* Balloc defines the structure of filesystem storage by allocating
* the physical blocks on a device given the inode and the logical
* block number in a file.
*
* NOTE: B_CLRBUF - this flag tells balloc to clear invalid portions
* of the buffer. However, any dirty bits will override missing
* valid bits. This case occurs when writable mmaps are truncated
* and then extended.
*/
int
ffs_balloc(struct vop_balloc_args *ap)
{
struct inode *ip;
ufs_daddr_t lbn;
int size;
struct ucred *cred;
int flags;
struct fs *fs;
ufs_daddr_t nb;
struct buf *bp, *nbp, *dbp;
struct vnode *vp;
struct indir indirs[NIADDR + 2];
ufs_daddr_t newb, *bap, pref;
int deallocated, osize, nsize, num, i, error;
ufs_daddr_t *allocib, *blkp, *allocblk, allociblk[NIADDR + 1];
ufs_daddr_t *lbns_remfree, lbns[NIADDR + 1];
int unwindidx;
int seqcount;
vp = ap->a_vp;
ip = VTOI(vp);
fs = ip->i_fs;
lbn = lblkno(fs, ap->a_startoffset);
size = blkoff(fs, ap->a_startoffset) + ap->a_size;
if (size > fs->fs_bsize)
panic("ffs_balloc: blk too big");
*ap->a_bpp = NULL;
if (lbn < 0)
return (EFBIG);
cred = ap->a_cred;
flags = ap->a_flags;
/*
* The vnode must be locked for us to be able to safely mess
* around with the inode.
*/
if (vn_islocked(vp) != LK_EXCLUSIVE) {
panic("ffs_balloc: vnode %p not exclusively locked!", vp);
}
/*
* If the next write will extend the file into a new block,
* and the file is currently composed of a fragment
* this fragment has to be extended to be a full block.
*/
nb = lblkno(fs, ip->i_size);
if (nb < NDADDR && nb < lbn) {
/*
* The filesize prior to this write can fit in direct
* blocks (ex. fragmentation is possibly done)
* we are now extending the file write beyond
* the block which has end of the file prior to this write.
*/
osize = blksize(fs, ip, nb);
/*
* osize gives disk allocated size in the last block. It is
* either in fragments or a file system block size.
*/
if (osize < fs->fs_bsize && osize > 0) {
/* A few fragments are already allocated, since the
* current extends beyond this block allocated the
* complete block as fragments are on in last block.
*/
error = ffs_realloccg(ip, nb,
ffs_blkpref(ip, nb, (int)nb, &ip->i_db[0]),
osize, (int)fs->fs_bsize, cred, &bp);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, nb,
dofftofsb(fs, bp->b_bio2.bio_offset),
ip->i_db[nb], fs->fs_bsize, osize, bp);
/* adjust the inode size, we just grew */
ip->i_size = smalllblktosize(fs, nb + 1);
ip->i_db[nb] = dofftofsb(fs, bp->b_bio2.bio_offset);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (flags & B_SYNC)
bwrite(bp);
else
bawrite(bp);
/* bp is already released here */
}
}
/*
* The first NDADDR blocks are direct blocks
*/
if (lbn < NDADDR) {
nb = ip->i_db[lbn];
if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) {
error = bread(vp, lblktodoff(fs, lbn), fs->fs_bsize, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_bio2.bio_offset = fsbtodoff(fs, nb);
*ap->a_bpp = bp;
return (0);
}
if (nb != 0) {
/*
* Consider need to reallocate a fragment.
*/
osize = fragroundup(fs, blkoff(fs, ip->i_size));
nsize = fragroundup(fs, size);
if (nsize <= osize) {
error = bread(vp, lblktodoff(fs, lbn),
osize, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_bio2.bio_offset = fsbtodoff(fs, nb);
} else {
/*
* NOTE: ffs_realloccg() issues a bread().
*/
error = ffs_realloccg(ip, lbn,
ffs_blkpref(ip, lbn, (int)lbn,
&ip->i_db[0]), osize, nsize, cred, &bp);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, lbn,
dofftofsb(fs, bp->b_bio2.bio_offset),
nb, nsize, osize, bp);
}
} else {
if (ip->i_size < smalllblktosize(fs, lbn + 1))
nsize = fragroundup(fs, size);
else
nsize = fs->fs_bsize;
error = ffs_alloc(ip, lbn,
ffs_blkpref(ip, lbn, (int)lbn, &ip->i_db[0]),
nsize, cred, &newb);
if (error)
return (error);
bp = getblk(vp, lblktodoff(fs, lbn), nsize, 0, 0);
bp->b_bio2.bio_offset = fsbtodoff(fs, newb);
if (flags & B_CLRBUF)
vfs_bio_clrbuf(bp);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, lbn, newb, 0,
nsize, 0, bp);
}
ip->i_db[lbn] = dofftofsb(fs, bp->b_bio2.bio_offset);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
*ap->a_bpp = bp;
return (0);
}
/*
* Determine the number of levels of indirection.
*/
pref = 0;
if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0)
return(error);
#ifdef DIAGNOSTIC
if (num < 1)
panic ("ffs_balloc: ufs_bmaparray returned indirect block");
#endif
/*
* Get a handle on the data block buffer before working through
* indirect blocks to avoid a deadlock between the VM system holding
* a locked VM page and issuing a BMAP (which tries to lock the
* indirect blocks), and the filesystem holding a locked indirect
* block and then trying to read a data block (which tries to lock
* the underlying VM pages).
*/
dbp = getblk(vp, lblktodoff(fs, lbn), fs->fs_bsize, 0, 0);
/*
* Setup undo history
*/
allocib = NULL;
allocblk = allociblk;
lbns_remfree = lbns;
unwindidx = -1;
/*
* Fetch the first indirect block directly from the inode, allocating
* one if necessary.
*/
--num;
nb = ip->i_ib[indirs[0].in_off];
if (nb == 0) {
pref = ffs_blkpref(ip, lbn, 0, NULL);
/*
* If the filesystem has run out of space we can skip the
* full fsync/undo of the main [fail] case since no undo
* history has been built yet. Hence the goto fail2.
*/
if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
cred, &newb)) != 0)
goto fail2;
nb = newb;
*allocblk++ = nb;
*lbns_remfree++ = indirs[1].in_lbn;
bp = getblk(vp, lblktodoff(fs, indirs[1].in_lbn),
fs->fs_bsize, 0, 0);
bp->b_bio2.bio_offset = fsbtodoff(fs, nb);
vfs_bio_clrbuf(bp);
if (DOINGSOFTDEP(vp)) {
softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off,
newb, 0, fs->fs_bsize, 0, bp);
bdwrite(bp);
} else {
/*
* Write synchronously so that indirect blocks
* never point at garbage.
*/
if (DOINGASYNC(vp))
bdwrite(bp);
else if ((error = bwrite(bp)) != 0)
goto fail;
}
allocib = &ip->i_ib[indirs[0].in_off];
*allocib = nb;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
}
/*
* Fetch through the indirect blocks, allocating as necessary.
*/
for (i = 1;;) {
error = bread(vp, lblktodoff(fs, indirs[i].in_lbn), (int)fs->fs_bsize, &bp);
if (error) {
brelse(bp);
goto fail;
}
bap = (ufs_daddr_t *)bp->b_data;
nb = bap[indirs[i].in_off];
if (i == num)
break;
i += 1;
if (nb != 0) {
bqrelse(bp);
continue;
}
if (pref == 0)
pref = ffs_blkpref(ip, lbn, 0, NULL);
if ((error =
ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, &newb)) != 0) {
brelse(bp);
goto fail;
}
nb = newb;
*allocblk++ = nb;
*lbns_remfree++ = indirs[i].in_lbn;
nbp = getblk(vp, lblktodoff(fs, indirs[i].in_lbn),
fs->fs_bsize, 0, 0);
nbp->b_bio2.bio_offset = fsbtodoff(fs, nb);
vfs_bio_clrbuf(nbp);
if (DOINGSOFTDEP(vp)) {
softdep_setup_allocindir_meta(nbp, ip, bp,
indirs[i - 1].in_off, nb);
bdwrite(nbp);
} else {
/*
* Write synchronously so that indirect blocks
* never point at garbage.
*/
if ((error = bwrite(nbp)) != 0) {
brelse(bp);
goto fail;
}
}
bap[indirs[i - 1].in_off] = nb;
if (allocib == NULL && unwindidx < 0)
unwindidx = i - 1;
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & B_SYNC) {
bwrite(bp);
} else {
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
}
}
/*
* Get the data block, allocating if necessary. We have already
* called getblk() on the data block buffer, dbp. If we have to
* allocate it and B_CLRBUF has been set the inference is an intention
* to zero out the related disk blocks, so we do not have to issue
* a read. Instead we simply call vfs_bio_clrbuf(). If B_CLRBUF is
* not set the caller intends to overwrite the entire contents of the
* buffer and we don't waste time trying to clean up the contents.
*
* bp references the current indirect block. When allocating,
* the block must be updated.
*/
if (nb == 0) {
pref = ffs_blkpref(ip, lbn, indirs[i].in_off, &bap[0]);
error = ffs_alloc(ip,
lbn, pref, (int)fs->fs_bsize, cred, &newb);
if (error) {
brelse(bp);
goto fail;
}
nb = newb;
*allocblk++ = nb;
*lbns_remfree++ = lbn;
dbp->b_bio2.bio_offset = fsbtodoff(fs, nb);
if (flags & B_CLRBUF)
vfs_bio_clrbuf(dbp);
if (DOINGSOFTDEP(vp))
softdep_setup_allocindir_page(ip, lbn, bp,
indirs[i].in_off, nb, 0, dbp);
bap[indirs[i].in_off] = nb;
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & B_SYNC) {
bwrite(bp);
} else {
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
}
*ap->a_bpp = dbp;
return (0);
}
brelse(bp);
/*
* At this point all related indirect blocks have been allocated
* if necessary and released. bp is no longer valid. dbp holds
* our getblk()'d data block.
*
* XXX we previously performed a cluster_read operation here.
*/
if (flags & B_CLRBUF) {
/*
* If B_CLRBUF is set we must validate the invalid portions
* of the buffer. This typically requires a read-before-
* write. The strategy call will fill in bio_offset in that
* case.
*
* If we hit this case we do a cluster read if possible
* since nearby data blocks are likely to be accessed soon
* too.
*/
if ((dbp->b_flags & B_CACHE) == 0) {
bqrelse(dbp);
seqcount = (flags & B_SEQMASK) >> B_SEQSHIFT;
if (seqcount &&
(vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
error = cluster_read(vp, (off_t)ip->i_size,
lblktodoff(fs, lbn),
(int)fs->fs_bsize,
fs->fs_bsize,
seqcount * BKVASIZE,
&dbp);
} else {
error = bread(vp, lblktodoff(fs, lbn),
(int)fs->fs_bsize, &dbp);
}
if (error)
goto fail;
} else {
/*
* Balloc defines the structure of filesystem storage
* by allocating the physical blocks on a device given
* the inode and the logical block number in a file.
* This is the allocation strategy for UFS1. Below is
* the allocation strategy for UFS2.
*/
int
ffs_balloc_ufs1(struct vnode *vp, off_t startoffset, int size,
struct ucred *cred, int flags, struct buf **bpp)
{
struct inode *ip;
struct ufs1_dinode *dp;
ufs_lbn_t lbn, lastlbn;
struct fs *fs;
ufs1_daddr_t nb;
struct buf *bp, *nbp;
struct ufsmount *ump;
struct indir indirs[NIADDR + 2];
int deallocated, osize, nsize, num, i, error;
ufs2_daddr_t newb;
ufs1_daddr_t *bap, pref;
ufs1_daddr_t *allocib, *blkp, *allocblk, allociblk[NIADDR + 1];
ufs2_daddr_t *lbns_remfree, lbns[NIADDR + 1];
int unwindidx = -1;
int saved_inbdflush;
static struct timeval lastfail;
static int curfail;
int reclaimed;
ip = VTOI(vp);
dp = ip->i_din1;
fs = ip->i_fs;
ump = ip->i_ump;
lbn = lblkno(fs, startoffset);
size = blkoff(fs, startoffset) + size;
reclaimed = 0;
if (size > fs->fs_bsize)
panic("ffs_balloc_ufs1: blk too big");
*bpp = NULL;
if (flags & IO_EXT)
return (EOPNOTSUPP);
if (lbn < 0)
return (EFBIG);
if (DOINGSOFTDEP(vp))
softdep_prealloc(vp, MNT_WAIT);
/*
* If the next write will extend the file into a new block,
* and the file is currently composed of a fragment
* this fragment has to be extended to be a full block.
*/
lastlbn = lblkno(fs, ip->i_size);
if (lastlbn < NDADDR && lastlbn < lbn) {
nb = lastlbn;
osize = blksize(fs, ip, nb);
if (osize < fs->fs_bsize && osize > 0) {
UFS_LOCK(ump);
error = ffs_realloccg(ip, nb, dp->di_db[nb],
ffs_blkpref_ufs1(ip, lastlbn, (int)nb,
&dp->di_db[0]), osize, (int)fs->fs_bsize, flags,
cred, &bp);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, nb,
dbtofsb(fs, bp->b_blkno), dp->di_db[nb],
fs->fs_bsize, osize, bp);
ip->i_size = smalllblktosize(fs, nb + 1);
dp->di_size = ip->i_size;
dp->di_db[nb] = dbtofsb(fs, bp->b_blkno);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (flags & IO_SYNC)
bwrite(bp);
else
bawrite(bp);
}
}
/*
* The first NDADDR blocks are direct blocks
*/
if (lbn < NDADDR) {
if (flags & BA_METAONLY)
panic("ffs_balloc_ufs1: BA_METAONLY for direct block");
nb = dp->di_db[lbn];
if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) {
error = bread(vp, lbn, fs->fs_bsize, NOCRED, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_blkno = fsbtodb(fs, nb);
*bpp = bp;
return (0);
}
if (nb != 0) {
/*
* Consider need to reallocate a fragment.
*/
osize = fragroundup(fs, blkoff(fs, ip->i_size));
nsize = fragroundup(fs, size);
if (nsize <= osize) {
error = bread(vp, lbn, osize, NOCRED, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_blkno = fsbtodb(fs, nb);
} else {
UFS_LOCK(ump);
error = ffs_realloccg(ip, lbn, dp->di_db[lbn],
ffs_blkpref_ufs1(ip, lbn, (int)lbn,
&dp->di_db[0]), osize, nsize, flags,
cred, &bp);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, lbn,
dbtofsb(fs, bp->b_blkno), nb,
nsize, osize, bp);
}
} else {
if (ip->i_size < smalllblktosize(fs, lbn + 1))
nsize = fragroundup(fs, size);
else
nsize = fs->fs_bsize;
UFS_LOCK(ump);
error = ffs_alloc(ip, lbn,
ffs_blkpref_ufs1(ip, lbn, (int)lbn, &dp->di_db[0]),
nsize, flags, cred, &newb);
if (error)
return (error);
bp = getblk(vp, lbn, nsize, 0, 0, 0);
bp->b_blkno = fsbtodb(fs, newb);
if (flags & BA_CLRBUF)
vfs_bio_clrbuf(bp);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, lbn, newb, 0,
nsize, 0, bp);
}
dp->di_db[lbn] = dbtofsb(fs, bp->b_blkno);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
*bpp = bp;
return (0);
}
/*
* Determine the number of levels of indirection.
*/
pref = 0;
if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0)
return(error);
#ifdef INVARIANTS
if (num < 1)
panic ("ffs_balloc_ufs1: ufs_getlbns returned indirect block");
#endif
saved_inbdflush = curthread_pflags_set(TDP_INBDFLUSH);
/*
* Fetch the first indirect block allocating if necessary.
*/
--num;
nb = dp->di_ib[indirs[0].in_off];
allocib = NULL;
allocblk = allociblk;
lbns_remfree = lbns;
if (nb == 0) {
UFS_LOCK(ump);
pref = ffs_blkpref_ufs1(ip, lbn, 0, (ufs1_daddr_t *)0);
if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
flags, cred, &newb)) != 0) {
curthread_pflags_restore(saved_inbdflush);
return (error);
}
nb = newb;
*allocblk++ = nb;
*lbns_remfree++ = indirs[1].in_lbn;
bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0, 0);
bp->b_blkno = fsbtodb(fs, nb);
vfs_bio_clrbuf(bp);
if (DOINGSOFTDEP(vp)) {
softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off,
newb, 0, fs->fs_bsize, 0, bp);
bdwrite(bp);
} else {
/*
* Write synchronously so that indirect blocks
* never point at garbage.
*/
if (DOINGASYNC(vp))
bdwrite(bp);
else if ((error = bwrite(bp)) != 0)
goto fail;
}
allocib = &dp->di_ib[indirs[0].in_off];
*allocib = nb;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
}
/*
* Fetch through the indirect blocks, allocating as necessary.
*/
retry:
for (i = 1;;) {
error = bread(vp,
indirs[i].in_lbn, (int)fs->fs_bsize, NOCRED, &bp);
if (error) {
brelse(bp);
goto fail;
}
bap = (ufs1_daddr_t *)bp->b_data;
nb = bap[indirs[i].in_off];
if (i == num)
break;
i += 1;
if (nb != 0) {
bqrelse(bp);
continue;
}
UFS_LOCK(ump);
if (pref == 0)
pref = ffs_blkpref_ufs1(ip, lbn, 0, (ufs1_daddr_t *)0);
if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
flags | IO_BUFLOCKED, cred, &newb)) != 0) {
brelse(bp);
if (++reclaimed == 1) {
UFS_LOCK(ump);
softdep_request_cleanup(fs, vp, cred,
FLUSH_BLOCKS_WAIT);
UFS_UNLOCK(ump);
goto retry;
}
if (ppsratecheck(&lastfail, &curfail, 1)) {
ffs_fserr(fs, ip->i_number, "filesystem full");
uprintf("\n%s: write failed, filesystem "
"is full\n", fs->fs_fsmnt);
}
goto fail;
}
nb = newb;
*allocblk++ = nb;
*lbns_remfree++ = indirs[i].in_lbn;
nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0, 0);
nbp->b_blkno = fsbtodb(fs, nb);
vfs_bio_clrbuf(nbp);
if (DOINGSOFTDEP(vp)) {
softdep_setup_allocindir_meta(nbp, ip, bp,
indirs[i - 1].in_off, nb);
bdwrite(nbp);
} else {
/*
* Write synchronously so that indirect blocks
* never point at garbage.
*/
if ((error = bwrite(nbp)) != 0) {
brelse(bp);
goto fail;
}
}
bap[indirs[i - 1].in_off] = nb;
if (allocib == NULL && unwindidx < 0)
unwindidx = i - 1;
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & IO_SYNC) {
bwrite(bp);
} else {
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
}
}
/*
* If asked only for the indirect block, then return it.
*/
if (flags & BA_METAONLY) {
curthread_pflags_restore(saved_inbdflush);
*bpp = bp;
return (0);
}
/*
* Get the data block, allocating if necessary.
*/
if (nb == 0) {
UFS_LOCK(ump);
pref = ffs_blkpref_ufs1(ip, lbn, indirs[i].in_off, &bap[0]);
error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
flags | IO_BUFLOCKED, cred, &newb);
if (error) {
brelse(bp);
if (++reclaimed == 1) {
UFS_LOCK(ump);
softdep_request_cleanup(fs, vp, cred,
FLUSH_BLOCKS_WAIT);
UFS_UNLOCK(ump);
goto retry;
}
if (ppsratecheck(&lastfail, &curfail, 1)) {
ffs_fserr(fs, ip->i_number, "filesystem full");
uprintf("\n%s: write failed, filesystem "
"is full\n", fs->fs_fsmnt);
}
goto fail;
}
nb = newb;
*allocblk++ = nb;
*lbns_remfree++ = lbn;
nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0, 0);
nbp->b_blkno = fsbtodb(fs, nb);
if (flags & BA_CLRBUF)
vfs_bio_clrbuf(nbp);
if (DOINGSOFTDEP(vp))
softdep_setup_allocindir_page(ip, lbn, bp,
indirs[i].in_off, nb, 0, nbp);
bap[indirs[i].in_off] = nb;
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & IO_SYNC) {
bwrite(bp);
} else {
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
}
curthread_pflags_restore(saved_inbdflush);
*bpp = nbp;
return (0);
}
brelse(bp);
if (flags & BA_CLRBUF) {
int seqcount = (flags & BA_SEQMASK) >> BA_SEQSHIFT;
if (seqcount && (vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
error = cluster_read(vp, ip->i_size, lbn,
(int)fs->fs_bsize, NOCRED,
MAXBSIZE, seqcount, &nbp);
} else {
error = bread(vp, lbn, (int)fs->fs_bsize, NOCRED, &nbp);
}
if (error) {
brelse(nbp);
goto fail;
}
} else {
/*
* Balloc defines the structure of file system storage
* by allocating the physical blocks on a device given
* the inode and the logical block number in a file.
*/
int
ffs1_balloc(struct inode *ip, off_t startoffset, int size, struct ucred *cred,
int flags, struct buf **bpp)
{
daddr_t lbn, nb, newb, pref;
struct fs *fs;
struct buf *bp, *nbp;
struct vnode *vp;
struct proc *p;
struct indir indirs[NIADDR + 2];
int32_t *bap;
int deallocated, osize, nsize, num, i, error;
int32_t *allocib, *blkp, *allocblk, allociblk[NIADDR+1];
int unwindidx = -1;
vp = ITOV(ip);
fs = ip->i_fs;
p = curproc;
lbn = lblkno(fs, startoffset);
size = blkoff(fs, startoffset) + size;
if (size > fs->fs_bsize)
panic("ffs1_balloc: blk too big");
if (bpp != NULL)
*bpp = NULL;
if (lbn < 0)
return (EFBIG);
/*
* If the next write will extend the file into a new block,
* and the file is currently composed of a fragment
* this fragment has to be extended to be a full block.
*/
nb = lblkno(fs, ip->i_ffs1_size);
if (nb < NDADDR && nb < lbn) {
osize = blksize(fs, ip, nb);
if (osize < fs->fs_bsize && osize > 0) {
error = ffs_realloccg(ip, nb,
ffs1_blkpref(ip, nb, (int)nb, &ip->i_ffs1_db[0]),
osize, (int)fs->fs_bsize, cred, bpp, &newb);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, nb, newb,
ip->i_ffs1_db[nb], fs->fs_bsize, osize,
bpp ? *bpp : NULL);
ip->i_ffs1_size = lblktosize(fs, nb + 1);
uvm_vnp_setsize(vp, ip->i_ffs1_size);
ip->i_ffs1_db[nb] = newb;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (bpp != NULL) {
if (flags & B_SYNC)
bwrite(*bpp);
else
bawrite(*bpp);
}
}
}
/*
* The first NDADDR blocks are direct blocks
*/
if (lbn < NDADDR) {
nb = ip->i_ffs1_db[lbn];
if (nb != 0 && ip->i_ffs1_size >= lblktosize(fs, lbn + 1)) {
/*
* The block is an already-allocated direct block
* and the file already extends past this block,
* thus this must be a whole block.
* Just read the block (if requested).
*/
if (bpp != NULL) {
error = bread(vp, lbn, fs->fs_bsize, bpp);
if (error) {
brelse(*bpp);
return (error);
}
}
return (0);
}
if (nb != 0) {
/*
* Consider need to reallocate a fragment.
*/
osize = fragroundup(fs, blkoff(fs, ip->i_ffs1_size));
nsize = fragroundup(fs, size);
if (nsize <= osize) {
/*
* The existing block is already
* at least as big as we want.
* Just read the block (if requested).
*/
if (bpp != NULL) {
error = bread(vp, lbn, fs->fs_bsize,
bpp);
if (error) {
brelse(*bpp);
return (error);
}
(*bpp)->b_bcount = osize;
}
return (0);
} else {
/*
* The existing block is smaller than we
* want, grow it.
*/
error = ffs_realloccg(ip, lbn,
ffs1_blkpref(ip, lbn, (int)lbn,
&ip->i_ffs1_db[0]),
osize, nsize, cred, bpp, &newb);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, lbn,
newb, nb, nsize, osize,
bpp ? *bpp : NULL);
}
} else {
/*
* The block was not previously allocated,
* allocate a new block or fragment.
*/
if (ip->i_ffs1_size < lblktosize(fs, lbn + 1))
nsize = fragroundup(fs, size);
else
nsize = fs->fs_bsize;
error = ffs_alloc(ip, lbn,
ffs1_blkpref(ip, lbn, (int)lbn, &ip->i_ffs1_db[0]),
nsize, cred, &newb);
if (error)
return (error);
if (bpp != NULL) {
*bpp = getblk(vp, lbn, fs->fs_bsize, 0, 0);
if (nsize < fs->fs_bsize)
(*bpp)->b_bcount = nsize;
(*bpp)->b_blkno = fsbtodb(fs, newb);
if (flags & B_CLRBUF)
clrbuf(*bpp);
}
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, lbn, newb, 0,
nsize, 0, bpp ? *bpp : NULL);
}
ip->i_ffs1_db[lbn] = newb;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
return (0);
}
/*
* Determine the number of levels of indirection.
*/
pref = 0;
if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0)
return(error);
#ifdef DIAGNOSTIC
if (num < 1)
panic ("ffs1_balloc: ufs_bmaparray returned indirect block");
#endif
/*
* Fetch the first indirect block allocating if necessary.
*/
--num;
nb = ip->i_ffs1_ib[indirs[0].in_off];
allocib = NULL;
allocblk = allociblk;
if (nb == 0) {
pref = ffs1_blkpref(ip, lbn, -indirs[0].in_off - 1, NULL);
error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
cred, &newb);
if (error)
goto fail;
nb = newb;
*allocblk++ = nb;
bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0);
bp->b_blkno = fsbtodb(fs, nb);
clrbuf(bp);
if (DOINGSOFTDEP(vp)) {
softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off,
newb, 0, fs->fs_bsize, 0, bp);
bdwrite(bp);
} else {
/*
* Write synchronously so that indirect blocks
* never point at garbage.
*/
if ((error = bwrite(bp)) != 0)
goto fail;
}
allocib = &ip->i_ffs1_ib[indirs[0].in_off];
*allocib = nb;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
}
/*
* Fetch through the indirect blocks, allocating as necessary.
*/
for (i = 1;;) {
error = bread(vp, indirs[i].in_lbn, (int)fs->fs_bsize, &bp);
if (error) {
brelse(bp);
goto fail;
}
bap = (int32_t *)bp->b_data;
nb = bap[indirs[i].in_off];
if (i == num)
break;
i++;
if (nb != 0) {
brelse(bp);
continue;
}
if (pref == 0)
pref = ffs1_blkpref(ip, lbn, i - num - 1, NULL);
error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred,
&newb);
if (error) {
brelse(bp);
goto fail;
}
nb = newb;
*allocblk++ = nb;
nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0);
nbp->b_blkno = fsbtodb(fs, nb);
clrbuf(nbp);
if (DOINGSOFTDEP(vp)) {
softdep_setup_allocindir_meta(nbp, ip, bp,
indirs[i - 1].in_off, nb);
bdwrite(nbp);
} else {
/*
* Write synchronously so that indirect blocks
* never point at garbage.
*/
if ((error = bwrite(nbp)) != 0) {
brelse(bp);
goto fail;
}
}
bap[indirs[i - 1].in_off] = nb;
if (allocib == NULL && unwindidx < 0)
unwindidx = i - 1;
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & B_SYNC) {
bwrite(bp);
} else {
bdwrite(bp);
}
}
/*
* Get the data block, allocating if necessary.
*/
if (nb == 0) {
pref = ffs1_blkpref(ip, lbn, indirs[i].in_off, &bap[0]);
error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred,
&newb);
if (error) {
brelse(bp);
goto fail;
}
nb = newb;
*allocblk++ = nb;
if (bpp != NULL) {
nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0);
nbp->b_blkno = fsbtodb(fs, nb);
if (flags & B_CLRBUF)
clrbuf(nbp);
*bpp = nbp;
}
if (DOINGSOFTDEP(vp))
softdep_setup_allocindir_page(ip, lbn, bp,
indirs[i].in_off, nb, 0, bpp ? *bpp : NULL);
bap[indirs[i].in_off] = nb;
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & B_SYNC) {
bwrite(bp);
} else {
bdwrite(bp);
}
return (0);
}
brelse(bp);
if (bpp != NULL) {
if (flags & B_CLRBUF) {
error = bread(vp, lbn, (int)fs->fs_bsize, &nbp);
if (error) {
brelse(nbp);
goto fail;
}
} else {
nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0);
nbp->b_blkno = fsbtodb(fs, nb);
}
*bpp = nbp;
}
return (0);
fail:
/*
* If we have failed to allocate any blocks, simply return the error.
* This is the usual case and avoids the need to fsync the file.
*/
if (allocblk == allociblk && allocib == NULL && unwindidx == -1)
return (error);
/*
* If we have failed part way through block allocation, we have to
* deallocate any indirect blocks that we have allocated. We have to
* fsync the file before we start to get rid of all of its
* dependencies so that we do not leave them dangling. We have to sync
* it at the end so that the softdep code does not find any untracked
* changes. Although this is really slow, running out of disk space is
* not expected to be a common occurrence. The error return from fsync
* is ignored as we already have an error to return to the user.
*/
VOP_FSYNC(vp, p->p_ucred, MNT_WAIT, p);
for (deallocated = 0, blkp = allociblk; blkp < allocblk; blkp++) {
ffs_blkfree(ip, *blkp, fs->fs_bsize);
deallocated += fs->fs_bsize;
}
if (allocib != NULL) {
*allocib = 0;
} else if (unwindidx >= 0) {
int r;
r = bread(vp, indirs[unwindidx].in_lbn, (int)fs->fs_bsize, &bp);
if (r)
panic("Could not unwind indirect block, error %d", r);
bap = (int32_t *)bp->b_data;
bap[indirs[unwindidx].in_off] = 0;
if (flags & B_SYNC) {
bwrite(bp);
} else {
bdwrite(bp);
}
}
if (deallocated) {
/*
* Restore user's disk quota because allocation failed.
*/
(void)ufs_quota_free_blocks(ip, btodb(deallocated), cred);
ip->i_ffs1_blocks -= btodb(deallocated);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
}
VOP_FSYNC(vp, p->p_ucred, MNT_WAIT, p);
return (error);
}
/*
* Balloc defines the structure of file system storage
* by allocating the physical blocks on a device given
* the inode and the logical block number in a file.
* This is the allocation strategy for UFS2. Above is
* the allocation strategy for UFS1.
*/
int
ffs_balloc_ufs2(vnode *vp, off_t startoffset, int size,
Ucred *cred, int flags, Buf **bpp)
{
int error = 0;
print("HARVEY TODO: %s\n", __func__);
#if 0
struct inode *ip;
struct ufs2_dinode *dp;
ufs_lbn_t lbn, lastlbn;
struct fs *fs;
struct buf *bp, *nbp;
struct ufsmount *ump;
struct indir indirs[UFS_NIADDR + 2];
ufs2_daddr_t nb, newb, *bap, pref;
ufs2_daddr_t *allocib, *blkp, *allocblk, allociblk[UFS_NIADDR + 1];
ufs2_daddr_t *lbns_remfree, lbns[UFS_NIADDR + 1];
int deallocated, osize, nsize, num, i, error;
int unwindidx = -1;
int saved_inbdflush;
static struct timeval lastfail;
static int curfail;
int gbflags, reclaimed;
ip = VTOI(vp);
dp = ip->i_din2;
fs = ITOFS(ip);
ump = ITOUMP(ip);
lbn = lblkno(fs, startoffset);
size = blkoff(fs, startoffset) + size;
reclaimed = 0;
if (size > fs->fs_bsize)
panic("ffs_balloc_ufs2: blk too big");
*bpp = nil;
if (lbn < 0)
return (EFBIG);
gbflags = (flags & BA_UNMAPPED) != 0 ? GB_UNMAPPED : 0;
if (DOINGSOFTDEP(vp))
softdep_prealloc(vp, MNT_WAIT);
/*
* Check for allocating external data.
*/
if (flags & IO_EXT) {
if (lbn >= UFS_NXADDR)
return (EFBIG);
/*
* If the next write will extend the data into a new block,
* and the data is currently composed of a fragment
* this fragment has to be extended to be a full block.
*/
lastlbn = lblkno(fs, dp->di_extsize);
if (lastlbn < lbn) {
nb = lastlbn;
osize = sblksize(fs, dp->di_extsize, nb);
if (osize < fs->fs_bsize && osize > 0) {
UFS_LOCK(ump);
error = ffs_realloccg(ip, -1 - nb,
dp->di_extb[nb],
ffs_blkpref_ufs2(ip, lastlbn, (int)nb,
&dp->di_extb[0]), osize,
(int)fs->fs_bsize, flags, cred, &bp);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocext(ip, nb,
dbtofsb(fs, bp->b_blkno),
dp->di_extb[nb],
fs->fs_bsize, osize, bp);
dp->di_extsize = smalllblktosize(fs, nb + 1);
dp->di_extb[nb] = dbtofsb(fs, bp->b_blkno);
bp->b_xflags |= BX_ALTDATA;
ip->i_flag |= IN_CHANGE;
if (flags & IO_SYNC)
bwrite(bp);
else
bawrite(bp);
}
}
/*
* All blocks are direct blocks
*/
if (flags & BA_METAONLY)
panic("ffs_balloc_ufs2: BA_METAONLY for ext block");
nb = dp->di_extb[lbn];
if (nb != 0 && dp->di_extsize >= smalllblktosize(fs, lbn + 1)) {
error = bread_gb(vp, -1 - lbn, fs->fs_bsize, NOCRED,
gbflags, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_blkno = fsbtodb(fs, nb);
bp->b_xflags |= BX_ALTDATA;
*bpp = bp;
return (0);
}
if (nb != 0) {
/*
* Consider need to reallocate a fragment.
*/
osize = fragroundup(fs, blkoff(fs, dp->di_extsize));
nsize = fragroundup(fs, size);
if (nsize <= osize) {
error = bread_gb(vp, -1 - lbn, osize, NOCRED,
gbflags, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_blkno = fsbtodb(fs, nb);
bp->b_xflags |= BX_ALTDATA;
} else {
UFS_LOCK(ump);
error = ffs_realloccg(ip, -1 - lbn,
dp->di_extb[lbn],
ffs_blkpref_ufs2(ip, lbn, (int)lbn,
&dp->di_extb[0]), osize, nsize, flags,
cred, &bp);
if (error)
return (error);
bp->b_xflags |= BX_ALTDATA;
if (DOINGSOFTDEP(vp))
softdep_setup_allocext(ip, lbn,
dbtofsb(fs, bp->b_blkno), nb,
nsize, osize, bp);
}
} else {
if (dp->di_extsize < smalllblktosize(fs, lbn + 1))
nsize = fragroundup(fs, size);
else
nsize = fs->fs_bsize;
UFS_LOCK(ump);
error = ffs_alloc(ip, lbn,
ffs_blkpref_ufs2(ip, lbn, (int)lbn, &dp->di_extb[0]),
nsize, flags, cred, &newb);
if (error)
return (error);
bp = getblk(vp, -1 - lbn, nsize, 0, 0, gbflags);
bp->b_blkno = fsbtodb(fs, newb);
bp->b_xflags |= BX_ALTDATA;
if (flags & BA_CLRBUF)
vfs_bio_clrbuf(bp);
if (DOINGSOFTDEP(vp))
softdep_setup_allocext(ip, lbn, newb, 0,
nsize, 0, bp);
}
dp->di_extb[lbn] = dbtofsb(fs, bp->b_blkno);
ip->i_flag |= IN_CHANGE;
*bpp = bp;
return (0);
}
/*
* If the next write will extend the file into a new block,
* and the file is currently composed of a fragment
* this fragment has to be extended to be a full block.
*/
lastlbn = lblkno(fs, ip->i_size);
if (lastlbn < UFS_NDADDR && lastlbn < lbn) {
nb = lastlbn;
osize = blksize(fs, ip, nb);
if (osize < fs->fs_bsize && osize > 0) {
UFS_LOCK(ump);
error = ffs_realloccg(ip, nb, dp->di_db[nb],
ffs_blkpref_ufs2(ip, lastlbn, (int)nb,
&dp->di_db[0]), osize, (int)fs->fs_bsize,
flags, cred, &bp);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, nb,
dbtofsb(fs, bp->b_blkno),
dp->di_db[nb],
fs->fs_bsize, osize, bp);
ip->i_size = smalllblktosize(fs, nb + 1);
dp->di_size = ip->i_size;
dp->di_db[nb] = dbtofsb(fs, bp->b_blkno);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (flags & IO_SYNC)
bwrite(bp);
else
bawrite(bp);
}
}
/*
* The first UFS_NDADDR blocks are direct blocks
*/
if (lbn < UFS_NDADDR) {
if (flags & BA_METAONLY)
panic("ffs_balloc_ufs2: BA_METAONLY for direct block");
nb = dp->di_db[lbn];
if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) {
error = bread_gb(vp, lbn, fs->fs_bsize, NOCRED,
gbflags, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_blkno = fsbtodb(fs, nb);
*bpp = bp;
return (0);
}
if (nb != 0) {
/*
* Consider need to reallocate a fragment.
*/
osize = fragroundup(fs, blkoff(fs, ip->i_size));
nsize = fragroundup(fs, size);
if (nsize <= osize) {
error = bread_gb(vp, lbn, osize, NOCRED,
gbflags, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_blkno = fsbtodb(fs, nb);
} else {
UFS_LOCK(ump);
error = ffs_realloccg(ip, lbn, dp->di_db[lbn],
ffs_blkpref_ufs2(ip, lbn, (int)lbn,
&dp->di_db[0]), osize, nsize, flags,
cred, &bp);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, lbn,
dbtofsb(fs, bp->b_blkno), nb,
nsize, osize, bp);
}
} else {
if (ip->i_size < smalllblktosize(fs, lbn + 1))
nsize = fragroundup(fs, size);
else
nsize = fs->fs_bsize;
UFS_LOCK(ump);
error = ffs_alloc(ip, lbn,
ffs_blkpref_ufs2(ip, lbn, (int)lbn,
&dp->di_db[0]), nsize, flags, cred, &newb);
if (error)
return (error);
bp = getblk(vp, lbn, nsize, 0, 0, gbflags);
bp->b_blkno = fsbtodb(fs, newb);
if (flags & BA_CLRBUF)
vfs_bio_clrbuf(bp);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, lbn, newb, 0,
nsize, 0, bp);
}
dp->di_db[lbn] = dbtofsb(fs, bp->b_blkno);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
*bpp = bp;
return (0);
}
/*
* Determine the number of levels of indirection.
*/
pref = 0;
if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0)
return(error);
#ifdef INVARIANTS
if (num < 1)
panic ("ffs_balloc_ufs2: ufs_getlbns returned indirect block");
#endif
saved_inbdflush = curthread_pflags_set(TDP_INBDFLUSH);
/*
* Fetch the first indirect block allocating if necessary.
*/
--num;
nb = dp->di_ib[indirs[0].in_off];
allocib = nil;
allocblk = allociblk;
lbns_remfree = lbns;
if (nb == 0) {
UFS_LOCK(ump);
pref = ffs_blkpref_ufs2(ip, lbn, -indirs[0].in_off - 1,
(ufs2_daddr_t *)0);
if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
flags, cred, &newb)) != 0) {
curthread_pflags_restore(saved_inbdflush);
return (error);
}
pref = newb + fs->fs_frag;
nb = newb;
MPASS(allocblk < allociblk + nitems(allociblk));
MPASS(lbns_remfree < lbns + nitems(lbns));
*allocblk++ = nb;
*lbns_remfree++ = indirs[1].in_lbn;
bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0,
GB_UNMAPPED);
bp->b_blkno = fsbtodb(fs, nb);
vfs_bio_clrbuf(bp);
if (DOINGSOFTDEP(vp)) {
softdep_setup_allocdirect(ip,
UFS_NDADDR + indirs[0].in_off, newb, 0,
fs->fs_bsize, 0, bp);
bdwrite(bp);
} else if ((flags & IO_SYNC) == 0 && DOINGASYNC(vp)) {
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
} else {
if ((error = bwrite(bp)) != 0)
goto fail;
}
allocib = &dp->di_ib[indirs[0].in_off];
*allocib = nb;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
}
/*
* Fetch through the indirect blocks, allocating as necessary.
*/
retry:
for (i = 1;;) {
error = bread(vp,
indirs[i].in_lbn, (int)fs->fs_bsize, NOCRED, &bp);
if (error) {
brelse(bp);
goto fail;
}
bap = (ufs2_daddr_t *)bp->b_data;
nb = bap[indirs[i].in_off];
if (i == num)
break;
i += 1;
if (nb != 0) {
bqrelse(bp);
continue;
}
UFS_LOCK(ump);
/*
* If parent indirect has just been allocated, try to cluster
* immediately following it.
*/
if (pref == 0)
pref = ffs_blkpref_ufs2(ip, lbn, i - num - 1,
(ufs2_daddr_t *)0);
if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
flags | IO_BUFLOCKED, cred, &newb)) != 0) {
brelse(bp);
if (DOINGSOFTDEP(vp) && ++reclaimed == 1) {
UFS_LOCK(ump);
softdep_request_cleanup(fs, vp, cred,
FLUSH_BLOCKS_WAIT);
UFS_UNLOCK(ump);
goto retry;
}
if (ppsratecheck(&lastfail, &curfail, 1)) {
ffs_fserr(fs, ip->i_number, "filesystem full");
uprintf("\n%s: write failed, filesystem "
"is full\n", fs->fs_fsmnt);
}
goto fail;
}
pref = newb + fs->fs_frag;
nb = newb;
MPASS(allocblk < allociblk + nitems(allociblk));
MPASS(lbns_remfree < lbns + nitems(lbns));
*allocblk++ = nb;
*lbns_remfree++ = indirs[i].in_lbn;
nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0,
GB_UNMAPPED);
nbp->b_blkno = fsbtodb(fs, nb);
vfs_bio_clrbuf(nbp);
if (DOINGSOFTDEP(vp)) {
softdep_setup_allocindir_meta(nbp, ip, bp,
indirs[i - 1].in_off, nb);
bdwrite(nbp);
} else if ((flags & IO_SYNC) == 0 && DOINGASYNC(vp)) {
if (nbp->b_bufsize == fs->fs_bsize)
nbp->b_flags |= B_CLUSTEROK;
bdwrite(nbp);
} else {
if ((error = bwrite(nbp)) != 0) {
brelse(bp);
goto fail;
}
}
bap[indirs[i - 1].in_off] = nb;
if (allocib == nil && unwindidx < 0)
unwindidx = i - 1;
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & IO_SYNC) {
bwrite(bp);
} else {
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
}
}
/*
* If asked only for the indirect block, then return it.
*/
if (flags & BA_METAONLY) {
curthread_pflags_restore(saved_inbdflush);
*bpp = bp;
return (0);
}
/*
* Get the data block, allocating if necessary.
*/
if (nb == 0) {
UFS_LOCK(ump);
/*
* If allocating metadata at the front of the cylinder
* group and parent indirect block has just been allocated,
* then cluster next to it if it is the first indirect in
* the file. Otherwise it has been allocated in the metadata
* area, so we want to find our own place out in the data area.
*/
if (pref == 0 || (lbn > UFS_NDADDR && fs->fs_metaspace != 0))
pref = ffs_blkpref_ufs2(ip, lbn, indirs[i].in_off,
&bap[0]);
error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
flags | IO_BUFLOCKED, cred, &newb);
if (error) {
brelse(bp);
if (DOINGSOFTDEP(vp) && ++reclaimed == 1) {
UFS_LOCK(ump);
softdep_request_cleanup(fs, vp, cred,
FLUSH_BLOCKS_WAIT);
UFS_UNLOCK(ump);
goto retry;
}
if (ppsratecheck(&lastfail, &curfail, 1)) {
ffs_fserr(fs, ip->i_number, "filesystem full");
uprintf("\n%s: write failed, filesystem "
"is full\n", fs->fs_fsmnt);
}
goto fail;
}
nb = newb;
MPASS(allocblk < allociblk + nitems(allociblk));
MPASS(lbns_remfree < lbns + nitems(lbns));
*allocblk++ = nb;
*lbns_remfree++ = lbn;
nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0, gbflags);
nbp->b_blkno = fsbtodb(fs, nb);
if (flags & BA_CLRBUF)
vfs_bio_clrbuf(nbp);
if (DOINGSOFTDEP(vp))
softdep_setup_allocindir_page(ip, lbn, bp,
indirs[i].in_off, nb, 0, nbp);
bap[indirs[i].in_off] = nb;
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & IO_SYNC) {
bwrite(bp);
} else {
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
}
curthread_pflags_restore(saved_inbdflush);
*bpp = nbp;
return (0);
}
brelse(bp);
/*
* If requested clear invalid portions of the buffer. If we
* have to do a read-before-write (typical if BA_CLRBUF is set),
* try to do some read-ahead in the sequential case to reduce
* the number of I/O transactions.
*/
if (flags & BA_CLRBUF) {
int seqcount = (flags & BA_SEQMASK) >> BA_SEQSHIFT;
if (seqcount != 0 &&
(vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0 &&
!(vm_page_count_severe() || buf_dirty_count_severe())) {
error = cluster_read(vp, ip->i_size, lbn,
(int)fs->fs_bsize, NOCRED,
MAXBSIZE, seqcount, gbflags, &nbp);
} else {
error = bread_gb(vp, lbn, (int)fs->fs_bsize,
NOCRED, gbflags, &nbp);
}
if (error) {
brelse(nbp);
goto fail;
}
} else {
int
ffs2_balloc(struct inode *ip, off_t off, int size, struct ucred *cred,
int flags, struct buf **bpp)
{
daddr_t lbn, lastlbn, nb, newb, *blkp;
daddr_t pref, *allocblk, allociblk[NIADDR + 1];
daddr_t *bap, *allocib;
int deallocated, osize, nsize, num, i, error, unwindidx, r;
struct buf *bp, *nbp;
struct indir indirs[NIADDR + 2];
struct fs *fs;
struct vnode *vp;
struct proc *p;
vp = ITOV(ip);
fs = ip->i_fs;
p = curproc;
unwindidx = -1;
lbn = lblkno(fs, off);
size = blkoff(fs, off) + size;
if (size > fs->fs_bsize)
panic("ffs2_balloc: block too big");
if (bpp != NULL)
*bpp = NULL;
if (lbn < 0)
return (EFBIG);
/*
* If the next write will extend the file into a new block, and the
* file is currently composed of a fragment, this fragment has to be
* extended to be a full block.
*/
lastlbn = lblkno(fs, ip->i_ffs2_size);
if (lastlbn < NDADDR && lastlbn < lbn) {
nb = lastlbn;
osize = blksize(fs, ip, nb);
if (osize < fs->fs_bsize && osize > 0) {
error = ffs_realloccg(ip, nb, ffs2_blkpref(ip,
lastlbn, nb, &ip->i_ffs2_db[0]), osize,
(int) fs->fs_bsize, cred, bpp, &newb);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, nb, newb,
ip->i_ffs2_db[nb], fs->fs_bsize, osize,
bpp ? *bpp : NULL);
ip->i_ffs2_size = lblktosize(fs, nb + 1);
uvm_vnp_setsize(vp, ip->i_ffs2_size);
ip->i_ffs2_db[nb] = newb;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (bpp) {
if (flags & B_SYNC)
bwrite(*bpp);
else
bawrite(*bpp);
}
}
}
/*
* The first NDADDR blocks are direct.
*/
if (lbn < NDADDR) {
nb = ip->i_ffs2_db[lbn];
if (nb != 0 && ip->i_ffs2_size >= lblktosize(fs, lbn + 1)) {
/*
* The direct block is already allocated and the file
* extends past this block, thus this must be a whole
* block. Just read it, if requested.
*/
if (bpp != NULL) {
error = bread(vp, lbn, fs->fs_bsize, bpp);
if (error) {
brelse(*bpp);
return (error);
}
}
return (0);
}
if (nb != 0) {
/*
* Consider the need to allocate a fragment.
*/
osize = fragroundup(fs, blkoff(fs, ip->i_ffs2_size));
nsize = fragroundup(fs, size);
if (nsize <= osize) {
/*
* The existing block is already at least as
* big as we want. Just read it, if requested.
*/
if (bpp != NULL) {
error = bread(vp, lbn, fs->fs_bsize,
bpp);
if (error) {
brelse(*bpp);
return (error);
}
(*bpp)->b_bcount = osize;
}
return (0);
} else {
/*
* The existing block is smaller than we want,
* grow it.
*/
error = ffs_realloccg(ip, lbn,
ffs2_blkpref(ip, lbn, (int) lbn,
&ip->i_ffs2_db[0]), osize, nsize, cred,
bpp, &newb);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, lbn,
newb, nb, nsize, osize,
bpp ? *bpp : NULL);
}
} else {
/*
* The block was not previously allocated, allocate a
* new block or fragment.
*/
if (ip->i_ffs2_size < lblktosize(fs, lbn + 1))
nsize = fragroundup(fs, size);
else
nsize = fs->fs_bsize;
error = ffs_alloc(ip, lbn, ffs2_blkpref(ip, lbn,
(int) lbn, &ip->i_ffs2_db[0]), nsize, cred, &newb);
if (error)
return (error);
if (bpp != NULL) {
bp = getblk(vp, lbn, fs->fs_bsize, 0, 0);
if (nsize < fs->fs_bsize)
bp->b_bcount = nsize;
bp->b_blkno = fsbtodb(fs, newb);
if (flags & B_CLRBUF)
clrbuf(bp);
*bpp = bp;
}
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, lbn, newb, 0,
nsize, 0, bpp ? *bpp : NULL);
}
ip->i_ffs2_db[lbn] = newb;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
return (0);
}
/*
* Determine the number of levels of indirection.
*/
pref = 0;
error = ufs_getlbns(vp, lbn, indirs, &num);
if (error)
return (error);
#ifdef DIAGNOSTIC
if (num < 1)
panic("ffs2_balloc: ufs_bmaparray returned indirect block");
#endif
/*
* Fetch the first indirect block allocating it necessary.
*/
--num;
nb = ip->i_ffs2_ib[indirs[0].in_off];
allocib = NULL;
allocblk = allociblk;
if (nb == 0) {
pref = ffs2_blkpref(ip, lbn, -indirs[0].in_off - 1, NULL);
error = ffs_alloc(ip, lbn, pref, (int) fs->fs_bsize, cred,
&newb);
if (error)
goto fail;
nb = newb;
*allocblk++ = nb;
bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0);
bp->b_blkno = fsbtodb(fs, nb);
clrbuf(bp);
if (DOINGSOFTDEP(vp)) {
softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off,
newb, 0, fs->fs_bsize, 0, bp);
bdwrite(bp);
} else {
/*
* Write synchronously so that indirect blocks never
* point at garbage.
*/
error = bwrite(bp);
if (error)
goto fail;
}
unwindidx = 0;
allocib = &ip->i_ffs2_ib[indirs[0].in_off];
*allocib = nb;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
}
/*
* Fetch through the indirect blocks, allocating as necessary.
*/
for (i = 1;;) {
error = bread(vp, indirs[i].in_lbn, (int)fs->fs_bsize, &bp);
if (error) {
brelse(bp);
goto fail;
}
bap = (int64_t *) bp->b_data;
nb = bap[indirs[i].in_off];
if (i == num)
break;
i++;
if (nb != 0) {
brelse(bp);
continue;
}
if (pref == 0)
pref = ffs2_blkpref(ip, lbn, i - num - 1, NULL);
error = ffs_alloc(ip, lbn, pref, (int) fs->fs_bsize, cred,
&newb);
if (error) {
brelse(bp);
goto fail;
}
nb = newb;
*allocblk++ = nb;
nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0);
nbp->b_blkno = fsbtodb(fs, nb);
clrbuf(nbp);
if (DOINGSOFTDEP(vp)) {
softdep_setup_allocindir_meta(nbp, ip, bp,
indirs[i - 1].in_off, nb);
bdwrite(nbp);
} else {
/*
* Write synchronously so that indirect blocks never
* point at garbage.
*/
error = bwrite(nbp);
if (error) {
brelse(bp);
goto fail;
}
}
if (unwindidx < 0)
unwindidx = i - 1;
bap[indirs[i - 1].in_off] = nb;
/*
* If required, write synchronously, otherwise use delayed
* write.
*/
if (flags & B_SYNC)
bwrite(bp);
else
bdwrite(bp);
}
/*
* Get the data block, allocating if necessary.
*/
if (nb == 0) {
pref = ffs2_blkpref(ip, lbn, indirs[num].in_off, &bap[0]);
error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred,
&newb);
if (error) {
brelse(bp);
goto fail;
}
nb = newb;
*allocblk++ = nb;
if (bpp != NULL) {
nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0);
nbp->b_blkno = fsbtodb(fs, nb);
if (flags & B_CLRBUF)
clrbuf(nbp);
*bpp = nbp;
}
if (DOINGSOFTDEP(vp))
softdep_setup_allocindir_page(ip, lbn, bp,
indirs[num].in_off, nb, 0, bpp ? *bpp : NULL);
bap[indirs[num].in_off] = nb;
if (allocib == NULL && unwindidx < 0)
unwindidx = i - 1;
/*
* If required, write synchronously, otherwise use delayed
* write.
*/
if (flags & B_SYNC)
bwrite(bp);
else
bdwrite(bp);
return (0);
}
brelse(bp);
if (bpp != NULL) {
if (flags & B_CLRBUF) {
error = bread(vp, lbn, (int)fs->fs_bsize, &nbp);
if (error) {
brelse(nbp);
goto fail;
}
} else {
nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0);
nbp->b_blkno = fsbtodb(fs, nb);
clrbuf(nbp);
}
*bpp = nbp;
}
return (0);
fail:
/*
* If we have failed to allocate any blocks, simply return the error.
* This is the usual case and avoids the need to fsync the file.
*/
if (allocblk == allociblk && allocib == NULL && unwindidx == -1)
return (error);
/*
* If we have failed part way through block allocation, we have to
* deallocate any indirect blocks that we have allocated. We have to
* fsync the file before we start to get rid of all of its
* dependencies so that we do not leave them dangling. We have to sync
* it at the end so that the softdep code does not find any untracked
* changes. Although this is really slow, running out of disk space is
* not expected to be a common occurrence. The error return from fsync
* is ignored as we already have an error to return to the user.
*/
VOP_FSYNC(vp, p->p_ucred, MNT_WAIT, p);
if (unwindidx >= 0) {
/*
* First write out any buffers we've created to resolve their
* softdeps. This must be done in reverse order of creation so
* that we resolve the dependencies in one pass.
* Write the cylinder group buffers for these buffers too.
*/
for (i = num; i >= unwindidx; i--) {
if (i == 0)
break;
bp = getblk(vp, indirs[i].in_lbn, (int) fs->fs_bsize,
0, 0);
if (bp->b_flags & B_DELWRI) {
nb = fsbtodb(fs, cgtod(fs, dtog(fs,
dbtofsb(fs, bp->b_blkno))));
bwrite(bp);
bp = getblk(ip->i_devvp, nb,
(int) fs->fs_cgsize, 0, 0);
if (bp->b_flags & B_DELWRI)
bwrite(bp);
else {
bp->b_flags |= B_INVAL;
brelse(bp);
}
} else {
bp->b_flags |= B_INVAL;
brelse(bp);
}
}
if (DOINGSOFTDEP(vp) && unwindidx == 0) {
ip->i_flag |= IN_CHANGE | IN_UPDATE;
ffs_update(ip, 1);
}
/*
* Now that any dependencies that we created have been
* resolved, we can undo the partial allocation.
*/
if (unwindidx == 0) {
*allocib = 0;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (DOINGSOFTDEP(vp))
ffs_update(ip, 1);
} else {
r = bread(vp, indirs[unwindidx].in_lbn,
(int)fs->fs_bsize, &bp);
if (r)
panic("ffs2_balloc: unwind failed");
bap = (int64_t *) bp->b_data;
bap[indirs[unwindidx].in_off] = 0;
bwrite(bp);
}
for (i = unwindidx + 1; i <= num; i++) {
bp = getblk(vp, indirs[i].in_lbn, (int)fs->fs_bsize, 0,
0);
bp->b_flags |= B_INVAL;
brelse(bp);
}
}
for (deallocated = 0, blkp = allociblk; blkp < allocblk; blkp++) {
ffs_blkfree(ip, *blkp, fs->fs_bsize);
deallocated += fs->fs_bsize;
}
if (deallocated) {
/*
* Restore user's disk quota because allocation failed.
*/
(void) ufs_quota_free_blocks(ip, btodb(deallocated), cred);
ip->i_ffs2_blocks -= btodb(deallocated);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
}
VOP_FSYNC(vp, p->p_ucred, MNT_WAIT, p);
return (error);
}
