/* ARGSUSED */
static int
ext2_read(struct vop_read_args *ap)
{
struct vnode *vp;
struct inode *ip;
struct uio *uio;
FS *fs;
struct buf *bp;
daddr_t lbn, nextlbn;
off_t nextloffset;
off_t bytesinfile;
long size, xfersize, blkoffset;
int error, orig_resid;
int seqcount = ap->a_ioflag >> 16;
vp = ap->a_vp;
ip = VTOI(vp);
uio = ap->a_uio;
#ifdef DIAGNOSTIC
if (uio->uio_rw != UIO_READ)
panic("ext2_read: mode");
if (vp->v_type == VLNK) {
if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
panic("ext2_read: short symlink");
} else if (vp->v_type != VREG && vp->v_type != VDIR)
panic("ext2_read: type %d", vp->v_type);
#endif
fs = ip->I_FS;
#if 0
if ((u_quad_t)uio->uio_offset > fs->fs_maxfilesize)
return (EFBIG);
#endif
orig_resid = uio->uio_resid;
for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
break;
lbn = lblkno(fs, uio->uio_offset);
nextlbn = lbn + 1;
nextloffset = lblktodoff(fs, nextlbn);
size = BLKSIZE(fs, ip, lbn);
blkoffset = blkoff(fs, uio->uio_offset);
xfersize = fs->s_frag_size - blkoffset;
if (uio->uio_resid < xfersize)
xfersize = uio->uio_resid;
if (bytesinfile < xfersize)
xfersize = bytesinfile;
if (nextloffset >= ip->i_size) {
error = bread(vp, lblktodoff(fs, lbn), size, &bp);
} else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
error = cluster_read(vp, (off_t)ip->i_size,
lblktodoff(fs, lbn), size,
uio->uio_resid,
(ap->a_ioflag >> 16) * BKVASIZE,
&bp);
} else if (seqcount > 1) {
/*
* Return buffer with the contents of block "offset" from the beginning of
* vnode "vp". If "res" is non-zero, fill it in with a pointer to the
* remaining space in the vnode.
*/
int
ffs_blkatoff(struct vnode *vp, off_t uoffset, char **res, struct buf **bpp)
{
struct inode *ip;
struct fs *fs;
struct buf *bp;
ufs_daddr_t lbn;
int bsize, error;
ip = VTOI(vp);
fs = ip->i_fs;
lbn = lblkno(fs, uoffset);
bsize = blksize(fs, ip, lbn);
*bpp = NULL;
error = bread(vp, lblktodoff(fs, lbn), bsize, &bp);
if (error) {
brelse(bp);
return (error);
}
if (res)
*res = (char *)bp->b_data + blkoff(fs, uoffset);
*bpp = bp;
return (0);
}
/*
* Return buffer with the contents of block "offset" from the beginning of
* directory "ip". If "res" is non-zero, fill it in with a pointer to the
* remaining space in the directory.
*/
int
ext2_blkatoff(struct vnode *vp, off_t offset, char **res, struct buf **bpp)
{
struct inode *ip;
struct ext2_sb_info *fs;
struct buf *bp;
daddr_t lbn;
int bsize, error;
ip = VTOI(vp);
fs = ip->i_e2fs;
lbn = lblkno(fs, offset);
bsize = blksize(fs, ip, lbn);
*bpp = NULL;
if ((error = bread(vp, lblktodoff(fs, lbn), bsize, &bp)) != 0) {
brelse(bp);
return (error);
}
if (res)
*res = (char *)bp->b_data + blkoff(fs, offset);
*bpp = bp;
return (0);
}
/*
* Release blocks associated with the inode ip and stored in the indirect
* block bn. Blocks are free'd in LIFO order up to (but not including)
* lastbn. If level is greater than SINGLE, the block is an indirect block
* and recursive calls to indirtrunc must be used to cleanse other indirect
* blocks.
*
* NB: triple indirect blocks are untested.
*/
static int
ffs_indirtrunc(struct inode *ip, ufs_daddr_t lbn, ufs_daddr_t dbn,
ufs_daddr_t lastbn, int level, long *countp)
{
int i;
struct buf *bp;
struct fs *fs = ip->i_fs;
ufs_daddr_t *bap;
struct vnode *vp;
ufs_daddr_t *copy = NULL, 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->fs_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 bio_offset field
* explicitly instead of letting bread do everything for us.
*/
vp = ITOV(ip);
bp = getblk(vp, lblktodoff(fs, lbn), (int)fs->fs_bsize, 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("ffs_indirtrunc: bad buffer size");
/*
* BIO is bio2 which chains back to bio1. We wait
* on bio1.
*/
bp->b_bio2.bio_offset = dbtodoff(fs, dbn);
bp->b_bio1.bio_done = biodone_sync;
bp->b_bio1.bio_flags |= BIO_SYNC;
vfs_busy_pages(vp, bp);
/*
* Access the block device layer using the device vnode
* and the translated block number (bio2) instead of the
* file vnode (vp) and logical block number (bio1).
*
* Even though we are bypassing the vnode layer, we still
* want the vnode state to indicate that an I/O on its behalf
* is in progress.
*/
bio_start_transaction(&bp->b_bio1, &vp->v_track_read);
vn_strategy(ip->i_devvp, &bp->b_bio2);
error = biowait(&bp->b_bio1, "biord");
}
if (error) {
brelse(bp);
*countp = 0;
return (error);
}
bap = (ufs_daddr_t *)bp->b_data;
if (lastbn != -1) {
copy = kmalloc(fs->fs_bsize, M_TEMP, M_WAITOK);
bcopy((caddr_t)bap, (caddr_t)copy, (uint)fs->fs_bsize);
bzero((caddr_t)&bap[last + 1],
(uint)(NINDIR(fs) - (last + 1)) * sizeof (ufs_daddr_t));
if (DOINGASYNC(vp)) {
bawrite(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 = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
(ufs_daddr_t)-1, level - 1, &blkcount)) != 0)
allerror = error;
blocksreleased += blkcount;
}
ffs_blkfree(ip, nb, fs->fs_bsize);
blocksreleased += nblocks;
}
/*
* Recursively free last partial block.
*/
if (level > SINGLE && lastbn >= 0) {
last = lastbn % factor;
nb = bap[i];
if (nb != 0) {
error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
last, level - 1, &blkcount);
if (error)
allerror = error;
blocksreleased += blkcount;
}
}
if (copy != NULL) {
kfree(copy, M_TEMP);
} else {
bp->b_flags |= B_INVAL | B_NOCACHE;
brelse(bp);
}
*countp = blocksreleased;
return (allerror);
}
if (uio->uio_resid < xfersize)
xfersize = uio->uio_resid;
if (bytesinfile < xfersize)
xfersize = bytesinfile;
if (nextloffset >= ip->i_size) {
error = bread(vp, lblktodoff(fs, lbn), size, &bp);
} else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
error = cluster_read(vp, (off_t)ip->i_size,
lblktodoff(fs, lbn), size,
uio->uio_resid,
(ap->a_ioflag >> 16) * BKVASIZE,
&bp);
} else if (seqcount > 1) {
int nextsize = BLKSIZE(fs, ip, nextlbn);
error = breadn(vp, lblktodoff(fs, lbn),
size, &nextloffset, &nextsize, 1, &bp);
} else {
error = bread(vp, lblktodoff(fs, lbn), size, &bp);
}
if (error) {
brelse(bp);
bp = NULL;
break;
}
/*
* We should only get non-zero b_resid when an I/O error
* has occurred, which should cause us to break above.
* However, if the short read did not cause an error,
* then we want to ensure that we do not uiomove bad
/*
* 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 {
/*
* Return buffer with the contents of block "offset" from the beginning of
* vnode "vp". If "res" is non-zero, fill it in with a pointer to the
* remaining space in the vnode.
*
* This version includes a read-ahead optimization.
*/
int
ffs_blkatoff_ra(struct vnode *vp, off_t uoffset, char **res, struct buf **bpp,
int seqcount)
{
struct inode *ip;
struct fs *fs;
struct buf *bp;
ufs_daddr_t lbn;
ufs_daddr_t nextlbn;
off_t base_loffset;
off_t next_loffset;
int bsize, error;
int nextbsize;
ip = VTOI(vp);
fs = ip->i_fs;
lbn = lblkno(fs, uoffset);
base_loffset = lblktodoff(fs, lbn);
bsize = blksize(fs, ip, lbn);
nextlbn = lbn + 1;
next_loffset = lblktodoff(fs, nextlbn);
*bpp = NULL;
if (next_loffset >= ip->i_size) {
/*
* Do not do readahead if this is the last block,
* bsize might represent a fragment.
*/
error = bread(vp, base_loffset, bsize, &bp);
} else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
/*
* Try to cluster if we allowed to.
*/
error = cluster_read(vp, (off_t)ip->i_size,
base_loffset, bsize,
bsize, seqcount * BKVASIZE, &bp);
} else if (seqcount > 1) {
/*
* Faked read ahead
*/
nextbsize = blksize(fs, ip, nextlbn);
error = breadn(vp, base_loffset, bsize,
&next_loffset, &nextbsize, 1, &bp);
} else {
/*
* Failing all of the above, just read what the
* user asked for. Interestingly, the same as
* the first option above.
*/
error = bread(vp, base_loffset, bsize, &bp);
}
if (error) {
brelse(bp);
return (error);
}
if (res)
*res = (char *)bp->b_data + (int)(uoffset - base_loffset);
*bpp = bp;
return (0);
}
/*
* 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);
}
/*
* Indirect blocks are now on the vnode for the file. They are given negative
* logical block numbers. Indirect blocks are addressed by the negative
* address of the first data block to which they point. Double indirect blocks
* are addressed by one less than the address of the first indirect block to
* which they point. Triple indirect blocks are addressed by one less than
* the address of the first double indirect block to which they point.
*
* ext2_bmaparray does the bmap conversion, and if requested returns the
* array of logical blocks which must be traversed to get to a block.
* Each entry contains the offset into that block that gets you to the
* next block and the disk address of the block (if it is assigned).
*/
static
int
ext2_bmaparray(struct vnode *vp, ext2_daddr_t bn, ext2_daddr_t *bnp,
struct indir *ap, int *nump, int *runp, int *runb)
{
struct inode *ip;
struct buf *bp;
struct ext2_mount *ump;
struct mount *mp;
struct ext2_sb_info *fs;
struct indir a[NIADDR+1], *xap;
ext2_daddr_t daddr;
long metalbn;
int error, maxrun, num;
ip = VTOI(vp);
mp = vp->v_mount;
ump = VFSTOEXT2(mp);
fs = ip->i_e2fs;
#ifdef DIAGNOSTIC
if ((ap != NULL && nump == NULL) || (ap == NULL && nump != NULL))
panic("ext2_bmaparray: invalid arguments");
#endif
if (runp) {
*runp = 0;
}
if (runb) {
*runb = 0;
}
maxrun = mp->mnt_iosize_max / mp->mnt_stat.f_iosize - 1;
xap = ap == NULL ? a : ap;
if (!nump)
nump = #
error = ext2_getlbns(vp, bn, xap, nump);
if (error)
return (error);
num = *nump;
if (num == 0) {
*bnp = blkptrtodb(ump, ip->i_db[bn]);
if (*bnp == 0)
*bnp = -1;
else if (runp) {
daddr_t bnb = bn;
for (++bn; bn < NDADDR && *runp < maxrun &&
is_sequential(ump, ip->i_db[bn - 1], ip->i_db[bn]);
++bn, ++*runp);
bn = bnb;
if (runb && (bn > 0)) {
for (--bn; (bn >= 0) && (*runb < maxrun) &&
is_sequential(ump, ip->i_db[bn],
ip->i_db[bn+1]);
--bn, ++*runb);
}
}
return (0);
}
/* Get disk address out of indirect block array */
daddr = ip->i_ib[xap->in_off];
for (bp = NULL, ++xap; --num; ++xap) {
/*
* Exit the loop if there is no disk address assigned yet and
* the indirect block isn't in the cache, or if we were
* looking for an indirect block and we've found it.
*/
metalbn = xap->in_lbn;
if ((daddr == 0 &&
!findblk(vp, dbtodoff(fs, metalbn), FINDBLK_TEST)) ||
metalbn == bn) {
break;
}
/*
* If we get here, we've either got the block in the cache
* or we have a disk address for it, go fetch it.
*/
if (bp)
bqrelse(bp);
xap->in_exists = 1;
bp = getblk(vp, lblktodoff(fs, metalbn),
mp->mnt_stat.f_iosize, 0, 0);
if ((bp->b_flags & B_CACHE) == 0) {
#ifdef DIAGNOSTIC
if (!daddr)
panic("ext2_bmaparray: indirect block not in cache");
#endif
/*
* This runs through ext2_strategy using bio2 to
* cache the disk offset, then comes back through
* bio1. So we want to wait on bio1
*/
bp->b_bio1.bio_done = biodone_sync;
bp->b_bio1.bio_flags |= BIO_SYNC;
bp->b_bio2.bio_offset = fsbtodoff(fs, daddr);
bp->b_flags &= ~(B_INVAL|B_ERROR);
bp->b_cmd = BUF_CMD_READ;
vfs_busy_pages(bp->b_vp, bp);
vn_strategy(bp->b_vp, &bp->b_bio1);
error = biowait(&bp->b_bio1, "biord");
if (error) {
brelse(bp);
return (error);
}
}
daddr = ((ext2_daddr_t *)bp->b_data)[xap->in_off];
if (num == 1 && daddr && runp) {
for (bn = xap->in_off + 1;
bn < MNINDIR(ump) && *runp < maxrun &&
is_sequential(ump,
((ext2_daddr_t *)bp->b_data)[bn - 1],
((ext2_daddr_t *)bp->b_data)[bn]);
++bn, ++*runp);
bn = xap->in_off;
if (runb && bn) {
for(--bn; bn >= 0 && *runb < maxrun &&
is_sequential(ump, ((daddr_t *)bp->b_data)[bn],
((daddr_t *)bp->b_data)[bn+1]);
--bn, ++*runb);
}
}
}
if (bp)
bqrelse(bp);
daddr = blkptrtodb(ump, daddr);
*bnp = daddr == 0 ? -1 : daddr;
return (0);
}
