static int
ext2_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn,
daddr_t lastbn, int level, e4fs_daddr_t *countp)
{
struct buf *bp;
struct m_ext2fs *fs = ip->i_e2fs;
struct vnode *vp;
e2fs_daddr_t *bap, *copy;
int i, nblocks, error = 0, allerror = 0;
e2fs_lbn_t nb, nlbn, last;
e4fs_daddr_t blkcount, factor, blocksreleased = 0;
/*
* Calculate index in current block of last
* block to be kept. -1 indicates the entire
* block so we need not calculate the index.
*/
factor = 1;
for (i = SINGLE; i < level; i++)
factor *= NINDIR(fs);
last = lastbn;
if (lastbn > 0)
last /= factor;
nblocks = btodb(fs->e2fs_bsize);
/*
* Get buffer of block pointers, zero those entries corresponding
* to blocks to be free'd, and update on disk copy first. Since
* double(triple) indirect before single(double) indirect, calls
* to bmap on these blocks will fail. However, we already have
* the on disk address, so we have to set the b_blkno field
* explicitly instead of letting bread do everything for us.
*/
vp = ITOV(ip);
bp = getblk(vp, lbn, (int)fs->e2fs_bsize, 0, 0, 0);
if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0) {
bp->b_iocmd = BIO_READ;
if (bp->b_bcount > bp->b_bufsize)
panic("ext2_indirtrunc: bad buffer size");
bp->b_blkno = dbn;
vfs_busy_pages(bp, 0);
bp->b_iooffset = dbtob(bp->b_blkno);
bstrategy(bp);
error = bufwait(bp);
}
if (error) {
brelse(bp);
*countp = 0;
return (error);
}
bap = (e2fs_daddr_t *)bp->b_data;
copy = malloc(fs->e2fs_bsize, M_TEMP, M_WAITOK);
bcopy((caddr_t)bap, (caddr_t)copy, (u_int)fs->e2fs_bsize);
bzero((caddr_t)&bap[last + 1],
(NINDIR(fs) - (last + 1)) * sizeof(e2fs_daddr_t));
if (last == -1)
bp->b_flags |= B_INVAL;
if (DOINGASYNC(vp)) {
bdwrite(bp);
} else {
error = bwrite(bp);
if (error)
allerror = error;
}
bap = copy;
/*
* Recursively free totally unused blocks.
*/
for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
i--, nlbn += factor) {
nb = bap[i];
if (nb == 0)
continue;
if (level > SINGLE) {
if ((error = ext2_indirtrunc(ip, nlbn,
fsbtodb(fs, nb), (int32_t)-1, level - 1, &blkcount)) != 0)
allerror = error;
blocksreleased += blkcount;
}
ext2_blkfree(ip, nb, fs->e2fs_bsize);
blocksreleased += nblocks;
}
/*
* Recursively free last partial block.
*/
if (level > SINGLE && lastbn >= 0) {
last = lastbn % factor;
nb = bap[i];
if (nb != 0) {
if ((error = ext2_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
last, level - 1, &blkcount)) != 0)
allerror = error;
blocksreleased += blkcount;
}
}
free(copy, M_TEMP);
*countp = blocksreleased;
return (allerror);
}
/*
* 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.
*/
int
ext2_balloc(struct inode *ip, e2fs_lbn_t lbn, int size, struct ucred *cred,
struct buf **bpp, int flags)
{
struct m_ext2fs *fs;
struct ext2mount *ump;
struct buf *bp, *nbp;
struct vnode *vp = ITOV(ip);
struct indir indirs[NIADDR + 2];
e4fs_daddr_t nb, newb;
e2fs_daddr_t *bap, pref;
int osize, nsize, num, i, error;
*bpp = NULL;
if (lbn < 0)
return (EFBIG);
fs = ip->i_e2fs;
ump = ip->i_ump;
/*
* check if this is a sequential block allocation.
* If so, increment next_alloc fields to allow ext2_blkpref
* to make a good guess
*/
if (lbn == ip->i_next_alloc_block + 1) {
ip->i_next_alloc_block++;
ip->i_next_alloc_goal++;
}
/*
* The first NDADDR blocks are direct blocks
*/
if (lbn < NDADDR) {
nb = ip->i_db[lbn];
/* no new block is to be allocated, and no need to expand
the file */
if (nb != 0 && ip->i_size >= (lbn + 1) * fs->e2fs_bsize) {
error = bread(vp, lbn, fs->e2fs_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 {
/* Godmar thinks: this shouldn't happen w/o fragments */
printf("nsize %d(%d) > osize %d(%d) nb %d\n",
(int)nsize, (int)size, (int)osize,
(int)ip->i_size, (int)nb);
panic(
"ext2_balloc: Something is terribly wrong");
/*
* please note there haven't been any changes from here on -
* FFS seems to work.
*/
}
} else {
if (ip->i_size < (lbn + 1) * fs->e2fs_bsize)
nsize = fragroundup(fs, size);
else
nsize = fs->e2fs_bsize;
EXT2_LOCK(ump);
error = ext2_alloc(ip, lbn,
ext2_blkpref(ip, lbn, (int)lbn, &ip->i_db[0], 0),
nsize, 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);
}
ip->i_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 = ext2_getlbns(vp, lbn, indirs, &num)) != 0)
return (error);
#ifdef INVARIANTS
if (num < 1)
panic ("ext2_balloc: ext2_getlbns returned indirect block");
#endif
/*
* Fetch the first indirect block allocating if necessary.
*/
--num;
nb = ip->i_ib[indirs[0].in_off];
if (nb == 0) {
EXT2_LOCK(ump);
pref = ext2_blkpref(ip, lbn, indirs[0].in_off +
EXT2_NDIR_BLOCKS, &ip->i_db[0], 0);
if ((error = ext2_alloc(ip, lbn, pref, fs->e2fs_bsize, cred,
&newb)))
return (error);
nb = newb;
bp = getblk(vp, indirs[1].in_lbn, fs->e2fs_bsize, 0, 0, 0);
bp->b_blkno = fsbtodb(fs, newb);
vfs_bio_clrbuf(bp);
/*
* Write synchronously so that indirect blocks
* never point at garbage.
*/
if ((error = bwrite(bp)) != 0) {
ext2_blkfree(ip, nb, fs->e2fs_bsize);
return (error);
}
ip->i_ib[indirs[0].in_off] = newb;
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->e2fs_bsize, NOCRED, &bp);
if (error) {
brelse(bp);
return (error);
}
bap = (e2fs_daddr_t *)bp->b_data;
nb = bap[indirs[i].in_off];
if (i == num)
break;
i += 1;
if (nb != 0) {
bqrelse(bp);
continue;
}
EXT2_LOCK(ump);
if (pref == 0)
pref = ext2_blkpref(ip, lbn, indirs[i].in_off, bap,
bp->b_lblkno);
error = ext2_alloc(ip, lbn, pref, (int)fs->e2fs_bsize, cred, &newb);
if (error) {
brelse(bp);
return (error);
}
nb = newb;
nbp = getblk(vp, indirs[i].in_lbn, fs->e2fs_bsize, 0, 0, 0);
nbp->b_blkno = fsbtodb(fs, nb);
vfs_bio_clrbuf(nbp);
/*
* Write synchronously so that indirect blocks
* never point at garbage.
*/
if ((error = bwrite(nbp)) != 0) {
ext2_blkfree(ip, nb, fs->e2fs_bsize);
EXT2_UNLOCK(ump);
brelse(bp);
return (error);
}
bap[indirs[i - 1].in_off] = nb;
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & IO_SYNC) {
bwrite(bp);
} else {
if (bp->b_bufsize == fs->e2fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
}
}
/*
* Get the data block, allocating if necessary.
*/
if (nb == 0) {
EXT2_LOCK(ump);
pref = ext2_blkpref(ip, lbn, indirs[i].in_off, &bap[0],
bp->b_lblkno);
if ((error = ext2_alloc(ip,
lbn, pref, (int)fs->e2fs_bsize, cred, &newb)) != 0) {
brelse(bp);
return (error);
}
nb = newb;
nbp = getblk(vp, lbn, fs->e2fs_bsize, 0, 0, 0);
nbp->b_blkno = fsbtodb(fs, nb);
if (flags & BA_CLRBUF)
vfs_bio_clrbuf(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->e2fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
}
*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->e2fs_bsize, NOCRED,
MAXBSIZE, seqcount, 0, &nbp);
} else {
error = bread(vp, lbn, (int)fs->e2fs_bsize, NOCRED, &nbp);
}
if (error) {
brelse(nbp);
return (error);
}
} else {
/*
* Truncate the inode oip to at most length size, freeing the
* disk blocks.
*/
int
ext2_truncate(struct vnode *vp, off_t length, int flags, struct ucred *cred,
struct thread *td)
{
struct vnode *ovp = vp;
int32_t lastblock;
struct inode *oip;
int32_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
uint32_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
struct m_ext2fs *fs;
struct buf *bp;
int offset, size, level;
e4fs_daddr_t count, nblocks, blocksreleased = 0;
int error, i, allerror;
off_t osize;
#ifdef INVARIANTS
struct bufobj *bo;
#endif
oip = VTOI(ovp);
#ifdef INVARIANTS
bo = &ovp->v_bufobj;
#endif
ASSERT_VOP_LOCKED(vp, "ext2_truncate");
if (length < 0)
return (EINVAL);
if (ovp->v_type == VLNK &&
oip->i_size < ovp->v_mount->mnt_maxsymlinklen) {
#ifdef INVARIANTS
if (length != 0)
panic("ext2_truncate: partial truncate of symlink");
#endif
bzero((char *)&oip->i_shortlink, (u_int)oip->i_size);
oip->i_size = 0;
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ext2_update(ovp, 1));
}
if (oip->i_size == length) {
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ext2_update(ovp, 0));
}
fs = oip->i_e2fs;
osize = oip->i_size;
/*
* Lengthen the size of the file. We must ensure that the
* last byte of the file is allocated. Since the smallest
* value of osize is 0, length will be at least 1.
*/
if (osize < length) {
if (length > oip->i_e2fs->e2fs_maxfilesize)
return (EFBIG);
vnode_pager_setsize(ovp, length);
offset = blkoff(fs, length - 1);
lbn = lblkno(fs, length - 1);
flags |= BA_CLRBUF;
error = ext2_balloc(oip, lbn, offset + 1, cred, &bp, flags);
if (error) {
vnode_pager_setsize(vp, osize);
return (error);
}
oip->i_size = length;
if (bp->b_bufsize == fs->e2fs_bsize)
bp->b_flags |= B_CLUSTEROK;
if (flags & IO_SYNC)
bwrite(bp);
else if (DOINGASYNC(ovp))
bdwrite(bp);
else
bawrite(bp);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ext2_update(ovp, !DOINGASYNC(ovp)));
}
/*
* Shorten the size of the file. If the file is not being
* truncated to a block boundry, the contents of the
* partial block following the end of the file must be
* zero'ed in case it ever become accessible again because
* of subsequent file growth.
*/
/* I don't understand the comment above */
offset = blkoff(fs, length);
if (offset == 0) {
oip->i_size = length;
} else {
lbn = lblkno(fs, length);
flags |= BA_CLRBUF;
error = ext2_balloc(oip, lbn, offset, cred, &bp, flags);
if (error)
return (error);
oip->i_size = length;
size = blksize(fs, oip, lbn);
bzero((char *)bp->b_data + offset, (u_int)(size - offset));
allocbuf(bp, size);
if (bp->b_bufsize == fs->e2fs_bsize)
bp->b_flags |= B_CLUSTEROK;
if (flags & IO_SYNC)
bwrite(bp);
else if (DOINGASYNC(ovp))
bdwrite(bp);
else
bawrite(bp);
}
/*
* Calculate index into inode's block list of
* last direct and indirect blocks (if any)
* which we want to keep. Lastblock is -1 when
* the file is truncated to 0.
*/
lastblock = lblkno(fs, length + fs->e2fs_bsize - 1) - 1;
lastiblock[SINGLE] = lastblock - NDADDR;
lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
nblocks = btodb(fs->e2fs_bsize);
/*
* Update file and block pointers on disk before we start freeing
* blocks. If we crash before free'ing blocks below, the blocks
* will be returned to the free list. lastiblock values are also
* normalized to -1 for calls to ext2_indirtrunc below.
*/
for (level = TRIPLE; level >= SINGLE; level--) {
oldblks[NDADDR + level] = oip->i_ib[level];
if (lastiblock[level] < 0) {
oip->i_ib[level] = 0;
lastiblock[level] = -1;
}
}
for (i = 0; i < NDADDR; i++) {
oldblks[i] = oip->i_db[i];
if (i > lastblock)
oip->i_db[i] = 0;
}
oip->i_flag |= IN_CHANGE | IN_UPDATE;
allerror = ext2_update(ovp, !DOINGASYNC(ovp));
/*
* Having written the new inode to disk, save its new configuration
* and put back the old block pointers long enough to process them.
* Note that we save the new block configuration so we can check it
* when we are done.
*/
for (i = 0; i < NDADDR; i++) {
newblks[i] = oip->i_db[i];
oip->i_db[i] = oldblks[i];
}
for (i = 0; i < NIADDR; i++) {
newblks[NDADDR + i] = oip->i_ib[i];
oip->i_ib[i] = oldblks[NDADDR + i];
}
oip->i_size = osize;
error = vtruncbuf(ovp, cred, length, (int)fs->e2fs_bsize);
if (error && (allerror == 0))
allerror = error;
vnode_pager_setsize(ovp, length);
/*
* Indirect blocks first.
*/
indir_lbn[SINGLE] = -NDADDR;
indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
for (level = TRIPLE; level >= SINGLE; level--) {
bn = oip->i_ib[level];
if (bn != 0) {
error = ext2_indirtrunc(oip, indir_lbn[level],
fsbtodb(fs, bn), lastiblock[level], level, &count);
if (error)
allerror = error;
blocksreleased += count;
if (lastiblock[level] < 0) {
oip->i_ib[level] = 0;
ext2_blkfree(oip, bn, fs->e2fs_fsize);
blocksreleased += nblocks;
}
}
if (lastiblock[level] >= 0)
goto done;
}
/*
* All whole direct blocks or frags.
*/
for (i = NDADDR - 1; i > lastblock; i--) {
long bsize;
bn = oip->i_db[i];
if (bn == 0)
continue;
oip->i_db[i] = 0;
bsize = blksize(fs, oip, i);
ext2_blkfree(oip, bn, bsize);
blocksreleased += btodb(bsize);
}
if (lastblock < 0)
goto done;
/*
* Finally, look for a change in size of the
* last direct block; release any frags.
*/
bn = oip->i_db[lastblock];
if (bn != 0) {
long oldspace, newspace;
/*
* Calculate amount of space we're giving
* back as old block size minus new block size.
*/
oldspace = blksize(fs, oip, lastblock);
oip->i_size = length;
newspace = blksize(fs, oip, lastblock);
if (newspace == 0)
panic("ext2_truncate: newspace");
if (oldspace - newspace > 0) {
/*
* Block number of space to be free'd is
* the old block # plus the number of frags
* required for the storage we're keeping.
*/
bn += numfrags(fs, newspace);
ext2_blkfree(oip, bn, oldspace - newspace);
blocksreleased += btodb(oldspace - newspace);
}
}
done:
#ifdef INVARIANTS
for (level = SINGLE; level <= TRIPLE; level++)
if (newblks[NDADDR + level] != oip->i_ib[level])
panic("itrunc1");
for (i = 0; i < NDADDR; i++)
if (newblks[i] != oip->i_db[i])
panic("itrunc2");
BO_LOCK(bo);
if (length == 0 && (bo->bo_dirty.bv_cnt != 0 ||
bo->bo_clean.bv_cnt != 0))
panic("itrunc3");
BO_UNLOCK(bo);
#endif /* INVARIANTS */
/*
* Put back the real size.
*/
oip->i_size = length;
if (oip->i_blocks >= blocksreleased)
oip->i_blocks -= blocksreleased;
else /* sanity */
oip->i_blocks = 0;
oip->i_flag |= IN_CHANGE;
vnode_pager_setsize(ovp, length);
return (allerror);
}
/*
* ext2_reallocblks(struct vnode *a_vp, struct cluster_save *a_buflist)
*/
int
ext2_reallocblks(struct vop_reallocblks_args *ap)
{
#ifndef FANCY_REALLOC
/* kprintf("ext2_reallocblks not implemented\n"); */
return ENOSPC;
#else
struct ext2_sb_info *fs;
struct inode *ip;
struct vnode *vp;
struct buf *sbp, *ebp;
daddr_t *bap, *sbap, *ebap;
struct cluster_save *buflist;
daddr_t start_lbn, end_lbn, soff, eoff, newblk, blkno;
struct indir start_ap[NIADDR + 1], end_ap[NIADDR + 1], *idp;
int i, len, start_lvl, end_lvl, pref, ssize;
vp = ap->a_vp;
ip = VTOI(vp);
fs = ip->i_e2fs;
#ifdef UNKLAR
if (fs->fs_contigsumsize <= 0)
return (ENOSPC);
#endif
buflist = ap->a_buflist;
len = buflist->bs_nchildren;
start_lbn = lblkno(fs, buflist->bs_children[0]->b_loffset);
end_lbn = start_lbn + len - 1;
#if DIAGNOSTIC
for (i = 1; i < len; i++) {
if (buflist->bs_children[i]->b_loffset != lblktodoff(fs, start_lbn) + lblktodoff(fs, i))
panic("ext2_reallocblks: non-cluster");
}
#endif
/*
* If the latest allocation is in a new block group, assume that
* the filesystem has decided to move and do not force it back to
* the previous block group.
*/
if (dtog(fs, dofftofsb(fs, buflist->bs_children[0]->b_bio2.bio_offset)) !=
dtog(fs, dofftofsb(fs, buflist->bs_children[len - 1]->b_bio2.bio_offset)))
return (ENOSPC);
if (ext2_getlbns(vp, start_lbn, start_ap, &start_lvl) ||
ext2_getlbns(vp, end_lbn, end_ap, &end_lvl))
return (ENOSPC);
/*
* Get the starting offset and block map for the first block.
*/
if (start_lvl == 0) {
sbap = &ip->i_db[0];
soff = start_lbn;
} else {
idp = &start_ap[start_lvl - 1];
if (bread(vp, lblktodoff(fs, idp->in_lbn), (int)fs->s_blocksize, NOCRED, &sbp)) {
brelse(sbp);
return (ENOSPC);
}
sbap = (daddr_t *)sbp->b_data;
soff = idp->in_off;
}
/*
* Find the preferred location for the cluster.
*/
pref = ext2_blkpref(ip, start_lbn, soff, sbap);
/*
* If the block range spans two block maps, get the second map.
*/
if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) {
ssize = len;
} else {
#if DIAGNOSTIC
if (start_ap[start_lvl-1].in_lbn == idp->in_lbn)
panic("ext2_reallocblk: start == end");
#endif
ssize = len - (idp->in_off + 1);
if (bread(vp, lblktodoff(fs, idp->in_lbn), (int)fs->s_blocksize, NOCRED, &ebp))
goto fail;
ebap = (daddr_t *)ebp->b_data;
}
/*
* Search the block map looking for an allocation of the desired size.
*/
if ((newblk = (daddr_t)ext2_hashalloc(ip, dtog(fs, pref), (long)pref,
len, (u_long (*)())ext2_clusteralloc)) == 0)
goto fail;
/*
* We have found a new contiguous block.
*
* First we have to replace the old block pointers with the new
* block pointers in the inode and indirect blocks associated
* with the file.
*/
blkno = newblk;
for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->s_frags_per_block) {
if (i == ssize)
bap = ebap;
#if DIAGNOSTIC
if (buflist->bs_children[i]->b_bio2.bio_offset != fsbtodoff(fs, *bap))
panic("ext2_reallocblks: alloc mismatch");
#endif
*bap++ = blkno;
}
/*
* Next we must write out the modified inode and indirect blocks.
* For strict correctness, the writes should be synchronous since
* the old block values may have been written to disk. In practise
* they are almost never written, but if we are concerned about
* strict correctness, the `doasyncfree' flag should be set to zero.
*
* The test on `doasyncfree' should be changed to test a flag
* that shows whether the associated buffers and inodes have
* been written. The flag should be set when the cluster is
* started and cleared whenever the buffer or inode is flushed.
* We can then check below to see if it is set, and do the
* synchronous write only when it has been cleared.
*/
if (sbap != &ip->i_db[0]) {
if (doasyncfree)
bdwrite(sbp);
else
bwrite(sbp);
} else {
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (!doasyncfree)
EXT2_UPDATE(vp, 1);
}
if (ssize < len)
if (doasyncfree)
bdwrite(ebp);
else
bwrite(ebp);
/*
* Last, free the old blocks and assign the new blocks to the buffers.
*/
for (blkno = newblk, i = 0; i < len; i++, blkno += fs->s_frags_per_block) {
ext2_blkfree(ip, dofftofsb(fs, buflist->bs_children[i]->b_bio2.bio_offset),
fs->s_blocksize);
buflist->bs_children[i]->b_bio2.bio_offset = fsbtodoff(fs, blkno);
}
return (0);
fail:
if (ssize < len)
brelse(ebp);
if (sbap != &ip->i_db[0])
brelse(sbp);
return (ENOSPC);
#endif /* FANCY_REALLOC */
}
static int
ext2_indirtrunc(struct inode *ip, daddr_t lbn, off_t doffset, daddr_t lastbn,
int level, long *countp)
{
int i;
struct buf *bp;
struct ext2_sb_info *fs = ip->i_e2fs;
daddr_t *bap;
struct vnode *vp;
daddr_t *copy, nb, nlbn, last;
long blkcount, factor;
int nblocks, blocksreleased = 0;
int error = 0, allerror = 0;
/*
* Calculate index in current block of last
* block to be kept. -1 indicates the entire
* block so we need not calculate the index.
*/
factor = 1;
for (i = SINGLE; i < level; i++)
factor *= NINDIR(fs);
last = lastbn;
if (lastbn > 0)
last /= factor;
nblocks = btodb(fs->s_blocksize);
/*
* Get buffer of block pointers, zero those entries corresponding
* to blocks to be free'd, and update on disk copy first. Since
* double(triple) indirect before single(double) indirect, calls
* to bmap on these blocks will fail. However, we already have
* the on disk address, so we have to set the bio_offset field
* explicitly instead of letting bread do everything for us.
*/
vp = ITOV(ip);
bp = getblk(vp, lblktodoff(fs, lbn), (int)fs->s_blocksize, 0, 0);
if ((bp->b_flags & B_CACHE) == 0) {
bp->b_flags &= ~(B_ERROR | B_INVAL);
bp->b_cmd = BUF_CMD_READ;
if (bp->b_bcount > bp->b_bufsize)
panic("ext2_indirtrunc: bad buffer size");
bp->b_bio2.bio_offset = doffset;
bp->b_bio1.bio_done = biodone_sync;
bp->b_bio1.bio_flags |= BIO_SYNC;
vfs_busy_pages(bp->b_vp, bp);
vn_strategy(vp, &bp->b_bio1);
error = biowait(&bp->b_bio1, "biord");
}
if (error) {
brelse(bp);
*countp = 0;
return (error);
}
bap = (daddr_t *)bp->b_data;
MALLOC(copy, daddr_t *, fs->s_blocksize, M_TEMP, M_WAITOK);
bcopy((caddr_t)bap, (caddr_t)copy, (u_int)fs->s_blocksize);
bzero((caddr_t)&bap[last + 1],
(u_int)(NINDIR(fs) - (last + 1)) * sizeof (daddr_t));
if (last == -1)
bp->b_flags |= B_INVAL;
error = bwrite(bp);
if (error)
allerror = error;
bap = copy;
/*
* Recursively free totally unused blocks.
*/
for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
i--, nlbn += factor) {
nb = bap[i];
if (nb == 0)
continue;
if (level > SINGLE) {
if ((error = ext2_indirtrunc(ip, nlbn,
fsbtodoff(fs, nb), (daddr_t)-1, level - 1, &blkcount)) != 0)
allerror = error;
blocksreleased += blkcount;
}
ext2_blkfree(ip, nb, fs->s_blocksize);
blocksreleased += nblocks;
}
/*
* Recursively free last partial block.
*/
if (level > SINGLE && lastbn >= 0) {
last = lastbn % factor;
nb = bap[i];
if (nb != 0) {
error = ext2_indirtrunc(ip, nlbn, fsbtodoff(fs, nb),
last, level - 1, &blkcount);
if (error)
allerror = error;
blocksreleased += blkcount;
}
}
FREE(copy, M_TEMP);
*countp = blocksreleased;
return (allerror);
}
/*
* Truncate the inode oip to at most length size, freeing the
* disk blocks.
*/
int
ext2_truncate(struct vnode *vp, off_t length, int flags, struct ucred *cred)
{
struct vnode *ovp = vp;
daddr_t lastblock;
struct inode *oip;
daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
daddr_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
struct ext2_sb_info *fs;
struct buf *bp;
int offset, size, level;
long count, nblocks, blocksreleased = 0;
int i;
int aflags, error, allerror;
off_t osize;
/*
kprintf("ext2_truncate called %d to %d\n", VTOI(ovp)->i_number, length);
*/ /*
* negative file sizes will totally break the code below and
* are not meaningful anyways.
*/
if (length < 0)
return EFBIG;
oip = VTOI(ovp);
if (ovp->v_type == VLNK &&
oip->i_size < ovp->v_mount->mnt_maxsymlinklen) {
#if DIAGNOSTIC
if (length != 0)
panic("ext2_truncate: partial truncate of symlink");
#endif
bzero((char *)&oip->i_shortlink, (u_int)oip->i_size);
oip->i_size = 0;
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (EXT2_UPDATE(ovp, 1));
}
if (oip->i_size == length) {
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (EXT2_UPDATE(ovp, 0));
}
#if QUOTA
if ((error = ext2_getinoquota(oip)) != 0)
return (error);
#endif
fs = oip->i_e2fs;
osize = oip->i_size;
ext2_discard_prealloc(oip);
/*
* Lengthen the size of the file. We must ensure that the
* last byte of the file is allocated. Since the smallest
* value of osize is 0, length will be at least 1.
*/
if (osize < length) {
offset = blkoff(fs, length - 1);
lbn = lblkno(fs, length - 1);
aflags = B_CLRBUF;
if (flags & IO_SYNC)
aflags |= B_SYNC;
vnode_pager_setsize(ovp, length);
error = ext2_balloc(oip, lbn, offset + 1, cred, &bp, aflags);
if (error) {
vnode_pager_setsize(ovp, osize);
return (error);
}
oip->i_size = length;
if (aflags & IO_SYNC)
bwrite(bp);
else
bawrite(bp);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (EXT2_UPDATE(ovp, 1));
}
/*
* Shorten the size of the file. If the file is not being
* truncated to a block boundry, the contents of the
* partial block following the end of the file must be
* zero'ed in case it ever become accessable again because
* of subsequent file growth.
*/
/* I don't understand the comment above */
offset = blkoff(fs, length);
if (offset == 0) {
oip->i_size = length;
} else {
lbn = lblkno(fs, length);
aflags = B_CLRBUF;
if (flags & IO_SYNC)
aflags |= B_SYNC;
error = ext2_balloc(oip, lbn, offset, cred, &bp, aflags);
if (error)
return (error);
oip->i_size = length;
size = blksize(fs, oip, lbn);
bzero((char *)bp->b_data + offset, (u_int)(size - offset));
allocbuf(bp, size);
if (aflags & IO_SYNC)
bwrite(bp);
else
bawrite(bp);
}
/*
* Calculate index into inode's block list of
* last direct and indirect blocks (if any)
* which we want to keep. Lastblock is -1 when
* the file is truncated to 0.
*/
lastblock = lblkno(fs, length + fs->s_blocksize - 1) - 1;
lastiblock[SINGLE] = lastblock - NDADDR;
lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
nblocks = btodb(fs->s_blocksize);
/*
* Update file and block pointers on disk before we start freeing
* blocks. If we crash before free'ing blocks below, the blocks
* will be returned to the free list. lastiblock values are also
* normalized to -1 for calls to ext2_indirtrunc below.
*/
bcopy((caddr_t)&oip->i_db[0], (caddr_t)oldblks, sizeof oldblks);
for (level = TRIPLE; level >= SINGLE; level--)
if (lastiblock[level] < 0) {
oip->i_ib[level] = 0;
lastiblock[level] = -1;
}
for (i = NDADDR - 1; i > lastblock; i--)
oip->i_db[i] = 0;
oip->i_flag |= IN_CHANGE | IN_UPDATE;
allerror = EXT2_UPDATE(ovp, 1);
/*
* Having written the new inode to disk, save its new configuration
* and put back the old block pointers long enough to process them.
* Note that we save the new block configuration so we can check it
* when we are done.
*/
bcopy((caddr_t)&oip->i_db[0], (caddr_t)newblks, sizeof newblks);
bcopy((caddr_t)oldblks, (caddr_t)&oip->i_db[0], sizeof oldblks);
oip->i_size = osize;
error = vtruncbuf(ovp, length, (int)fs->s_blocksize);
if (error && (allerror == 0))
allerror = error;
/*
* Indirect blocks first.
*/
indir_lbn[SINGLE] = -NDADDR;
indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
for (level = TRIPLE; level >= SINGLE; level--) {
bn = oip->i_ib[level];
if (bn != 0) {
error = ext2_indirtrunc(oip, indir_lbn[level],
fsbtodoff(fs, bn), lastiblock[level], level, &count);
if (error)
allerror = error;
blocksreleased += count;
if (lastiblock[level] < 0) {
oip->i_ib[level] = 0;
ext2_blkfree(oip, bn, fs->s_frag_size);
blocksreleased += nblocks;
}
}
if (lastiblock[level] >= 0)
goto done;
}
/*
* All whole direct blocks or frags.
*/
for (i = NDADDR - 1; i > lastblock; i--) {
long bsize;
bn = oip->i_db[i];
if (bn == 0)
continue;
oip->i_db[i] = 0;
bsize = blksize(fs, oip, i);
ext2_blkfree(oip, bn, bsize);
blocksreleased += btodb(bsize);
}
if (lastblock < 0)
goto done;
/*
* Finally, look for a change in size of the
* last direct block; release any frags.
*/
bn = oip->i_db[lastblock];
if (bn != 0) {
long oldspace, newspace;
/*
* Calculate amount of space we're giving
* back as old block size minus new block size.
*/
oldspace = blksize(fs, oip, lastblock);
oip->i_size = length;
newspace = blksize(fs, oip, lastblock);
if (newspace == 0)
panic("itrunc: newspace");
if (oldspace - newspace > 0) {
/*
* Block number of space to be free'd is
* the old block # plus the number of frags
* required for the storage we're keeping.
*/
bn += numfrags(fs, newspace);
ext2_blkfree(oip, bn, oldspace - newspace);
blocksreleased += btodb(oldspace - newspace);
}
}
done:
#if DIAGNOSTIC
for (level = SINGLE; level <= TRIPLE; level++)
if (newblks[NDADDR + level] != oip->i_ib[level])
panic("itrunc1");
for (i = 0; i < NDADDR; i++)
if (newblks[i] != oip->i_db[i])
panic("itrunc2");
if (length == 0 && (!RB_EMPTY(&ovp->v_rbdirty_tree) ||
!RB_EMPTY(&ovp->v_rbclean_tree)))
panic("itrunc3");
#endif /* DIAGNOSTIC */
/*
* Put back the real size.
*/
oip->i_size = length;
oip->i_blocks -= blocksreleased;
if (oip->i_blocks < 0) /* sanity */
oip->i_blocks = 0;
oip->i_flag |= IN_CHANGE;
vnode_pager_setsize(ovp, length);
#if QUOTA
ext2_chkdq(oip, -blocksreleased, NOCRED, 0);
#endif
return (allerror);
}
int
ext2_reallocblks(struct vop_reallocblks_args *ap)
{
struct m_ext2fs *fs;
struct inode *ip;
struct vnode *vp;
struct buf *sbp, *ebp;
uint32_t *bap, *sbap, *ebap = 0;
struct ext2mount *ump;
struct cluster_save *buflist;
struct indir start_ap[NIADDR + 1], end_ap[NIADDR + 1], *idp;
e2fs_lbn_t start_lbn, end_lbn;
int soff;
e2fs_daddr_t newblk, blkno;
int i, len, start_lvl, end_lvl, pref, ssize;
if (doreallocblks == 0)
return (ENOSPC);
vp = ap->a_vp;
ip = VTOI(vp);
fs = ip->i_e2fs;
ump = ip->i_ump;
if (fs->e2fs_contigsumsize <= 0)
return (ENOSPC);
buflist = ap->a_buflist;
len = buflist->bs_nchildren;
start_lbn = buflist->bs_children[0]->b_lblkno;
end_lbn = start_lbn + len - 1;
#ifdef INVARIANTS
for (i = 1; i < len; i++)
if (buflist->bs_children[i]->b_lblkno != start_lbn + i)
panic("ext2_reallocblks: non-cluster");
#endif
/*
* If the cluster crosses the boundary for the first indirect
* block, leave space for the indirect block. Indirect blocks
* are initially laid out in a position after the last direct
* block. Block reallocation would usually destroy locality by
* moving the indirect block out of the way to make room for
* data blocks if we didn't compensate here. We should also do
* this for other indirect block boundaries, but it is only
* important for the first one.
*/
if (start_lbn < NDADDR && end_lbn >= NDADDR)
return (ENOSPC);
/*
* If the latest allocation is in a new cylinder group, assume that
* the filesystem has decided to move and do not force it back to
* the previous cylinder group.
*/
if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) !=
dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno)))
return (ENOSPC);
if (ext2_getlbns(vp, start_lbn, start_ap, &start_lvl) ||
ext2_getlbns(vp, end_lbn, end_ap, &end_lvl))
return (ENOSPC);
/*
* Get the starting offset and block map for the first block.
*/
if (start_lvl == 0) {
sbap = &ip->i_db[0];
soff = start_lbn;
} else {
idp = &start_ap[start_lvl - 1];
if (bread(vp, idp->in_lbn, (int)fs->e2fs_bsize, NOCRED, &sbp)) {
brelse(sbp);
return (ENOSPC);
}
sbap = (u_int *)sbp->b_data;
soff = idp->in_off;
}
/*
* If the block range spans two block maps, get the second map.
*/
if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) {
ssize = len;
} else {
#ifdef INVARIANTS
if (start_ap[start_lvl-1].in_lbn == idp->in_lbn)
panic("ext2_reallocblks: start == end");
#endif
ssize = len - (idp->in_off + 1);
if (bread(vp, idp->in_lbn, (int)fs->e2fs_bsize, NOCRED, &ebp))
goto fail;
ebap = (u_int *)ebp->b_data;
}
/*
* Find the preferred location for the cluster.
*/
EXT2_LOCK(ump);
pref = ext2_blkpref(ip, start_lbn, soff, sbap, 0);
/*
* Search the block map looking for an allocation of the desired size.
*/
if ((newblk = (e2fs_daddr_t)ext2_hashalloc(ip, dtog(fs, pref), pref,
len, ext2_clusteralloc)) == 0) {
EXT2_UNLOCK(ump);
goto fail;
}
/*
* We have found a new contiguous block.
*
* First we have to replace the old block pointers with the new
* block pointers in the inode and indirect blocks associated
* with the file.
*/
#ifdef DEBUG
printf("realloc: ino %d, lbns %jd-%jd\n\told:", ip->i_number,
(intmax_t)start_lbn, (intmax_t)end_lbn);
#endif /* DEBUG */
blkno = newblk;
for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->e2fs_fpb) {
if (i == ssize) {
bap = ebap;
soff = -i;
}
#ifdef INVARIANTS
if (buflist->bs_children[i]->b_blkno != fsbtodb(fs, *bap))
panic("ext2_reallocblks: alloc mismatch");
#endif
#ifdef DEBUG
printf(" %d,", *bap);
#endif /* DEBUG */
*bap++ = blkno;
}
/*
* Next we must write out the modified inode and indirect blocks.
* For strict correctness, the writes should be synchronous since
* the old block values may have been written to disk. In practise
* they are almost never written, but if we are concerned about
* strict correctness, the `doasyncfree' flag should be set to zero.
*
* The test on `doasyncfree' should be changed to test a flag
* that shows whether the associated buffers and inodes have
* been written. The flag should be set when the cluster is
* started and cleared whenever the buffer or inode is flushed.
* We can then check below to see if it is set, and do the
* synchronous write only when it has been cleared.
*/
if (sbap != &ip->i_db[0]) {
if (doasyncfree)
bdwrite(sbp);
else
bwrite(sbp);
} else {
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (!doasyncfree)
ext2_update(vp, 1);
}
if (ssize < len) {
if (doasyncfree)
bdwrite(ebp);
else
bwrite(ebp);
}
/*
* Last, free the old blocks and assign the new blocks to the buffers.
*/
#ifdef DEBUG
printf("\n\tnew:");
#endif /* DEBUG */
for (blkno = newblk, i = 0; i < len; i++, blkno += fs->e2fs_fpb) {
ext2_blkfree(ip, dbtofsb(fs, buflist->bs_children[i]->b_blkno),
fs->e2fs_bsize);
buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno);
#ifdef DEBUG
printf(" %d,", blkno);
#endif /* DEBUG */
}
#ifdef DEBUG
printf("\n");
#endif /* DEBUG */
return (0);
fail:
if (ssize < len)
brelse(ebp);
if (sbap != &ip->i_db[0])
brelse(sbp);
return (ENOSPC);
}