/*
* 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);
}
/*
* Truncate the inode oip to at most length size, freeing the
* disk blocks.
*/
int
ffs_truncate(struct inode *oip, off_t length, int flags, struct ucred *cred)
{
struct vnode *ovp;
daddr64_t lastblock;
daddr64_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
daddr64_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
struct fs *fs;
struct buf *bp;
int offset, size, level;
long count, nblocks, vflags, blocksreleased = 0;
int i, aflags, error, allerror, indirect = 0;
off_t osize;
extern int num_indirdep;
extern int max_indirdep;
if (length < 0)
return (EINVAL);
ovp = ITOV(oip);
if (ovp->v_type != VREG &&
ovp->v_type != VDIR &&
ovp->v_type != VLNK)
return (0);
if (DIP(oip, size) == length)
return (0);
if (ovp->v_type == VLNK &&
(DIP(oip, size) < ovp->v_mount->mnt_maxsymlinklen ||
(ovp->v_mount->mnt_maxsymlinklen == 0 &&
oip->i_din1->di_blocks == 0))) {
#ifdef DIAGNOSTIC
if (length != 0)
panic("ffs_truncate: partial truncate of symlink");
#endif
memset(SHORTLINK(oip), 0, (size_t) DIP(oip, size));
DIP_ASSIGN(oip, size, 0);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (UFS_UPDATE(oip, MNT_WAIT));
}
if ((error = getinoquota(oip)) != 0)
return (error);
uvm_vnp_setsize(ovp, length);
oip->i_ci.ci_lasta = oip->i_ci.ci_clen
= oip->i_ci.ci_cstart = oip->i_ci.ci_lastw = 0;
if (DOINGSOFTDEP(ovp)) {
if (length > 0 || softdep_slowdown(ovp)) {
/*
* If a file is only partially truncated, then
* we have to clean up the data structures
* describing the allocation past the truncation
* point. Finding and deallocating those structures
* is a lot of work. Since partial truncation occurs
* rarely, we solve the problem by syncing the file
* so that it will have no data structures left.
*/
if ((error = VOP_FSYNC(ovp, cred, MNT_WAIT)) != 0)
return (error);
} else {
(void)ufs_quota_free_blocks(oip, DIP(oip, blocks),
NOCRED);
softdep_setup_freeblocks(oip, length);
(void) vinvalbuf(ovp, 0, cred, curproc, 0, 0);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (UFS_UPDATE(oip, 0));
}
}
fs = oip->i_fs;
osize = DIP(oip, 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 > fs->fs_maxfilesize)
return (EFBIG);
aflags = B_CLRBUF;
if (flags & IO_SYNC)
aflags |= B_SYNC;
error = UFS_BUF_ALLOC(oip, length - 1, 1,
cred, aflags, &bp);
if (error)
return (error);
if (bp->b_lblkno >= NDADDR)
indirect = 1;
DIP_ASSIGN(oip, size, length);
uvm_vnp_setsize(ovp, length);
(void) uvm_vnp_uncache(ovp);
if (aflags & B_SYNC)
bwrite(bp);
else
bawrite(bp);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
error = UFS_UPDATE(oip, MNT_WAIT);
if (DOINGSOFTDEP(ovp) && num_indirdep > max_indirdep)
if (indirect) {
/*
* If the number of pending indirect block
* dependencies is sufficiently close to the
* maximum number of simultaneously mappable
* buffers force a sync on the vnode to prevent
* buffer cache exhaustion.
*/
VOP_FSYNC(ovp, curproc->p_ucred, MNT_WAIT);
}
return (error);
}
uvm_vnp_setsize(ovp, length);
/*
* Shorten the size of the file. If the file is not being
* truncated to a block boundary, the contents of the
* partial block following the end of the file must be
* zero'ed in case it ever becomes accessible again because
* of subsequent file growth. Directories however are not
* zero'ed as they should grow back initialized to empty.
*/
offset = blkoff(fs, length);
if (offset == 0) {
DIP_ASSIGN(oip, size, length);
} else {
lbn = lblkno(fs, length);
aflags = B_CLRBUF;
if (flags & IO_SYNC)
aflags |= B_SYNC;
error = UFS_BUF_ALLOC(oip, length - 1, 1,
cred, aflags, &bp);
if (error)
return (error);
/*
* When we are doing soft updates and the UFS_BALLOC
* above fills in a direct block hole with a full sized
* block that will be truncated down to a fragment below,
* we must flush out the block dependency with an FSYNC
* so that we do not get a soft updates inconsistency
* when we create the fragment below.
*/
if (DOINGSOFTDEP(ovp) && lbn < NDADDR &&
fragroundup(fs, blkoff(fs, length)) < fs->fs_bsize &&
(error = VOP_FSYNC(ovp, cred, MNT_WAIT)) != 0)
return (error);
DIP_ASSIGN(oip, size, length);
size = blksize(fs, oip, lbn);
(void) uvm_vnp_uncache(ovp);
if (ovp->v_type != VDIR)
bzero((char *)bp->b_data + offset,
(u_int)(size - offset));
bp->b_bcount = size;
if (aflags & B_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->fs_bsize - 1) - 1;
lastiblock[SINGLE] = lastblock - NDADDR;
lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
nblocks = btodb(fs->fs_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 ffs_indirtrunc below.
*/
for (level = TRIPLE; level >= SINGLE; level--) {
oldblks[NDADDR + level] = DIP(oip, ib[level]);
if (lastiblock[level] < 0) {
DIP_ASSIGN(oip, ib[level], 0);
lastiblock[level] = -1;
}
}
for (i = 0; i < NDADDR; i++) {
oldblks[i] = DIP(oip, db[i]);
if (i > lastblock)
DIP_ASSIGN(oip, db[i], 0);
}
oip->i_flag |= IN_CHANGE | IN_UPDATE;
if ((error = UFS_UPDATE(oip, MNT_WAIT)) != 0)
allerror = error;
/*
* 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] = DIP(oip, db[i]);
DIP_ASSIGN(oip, db[i], oldblks[i]);
}
for (i = 0; i < NIADDR; i++) {
newblks[NDADDR + i] = DIP(oip, ib[i]);
DIP_ASSIGN(oip, ib[i], oldblks[NDADDR + i]);
}
DIP_ASSIGN(oip, size, osize);
vflags = ((length > 0) ? V_SAVE : 0) | V_SAVEMETA;
allerror = vinvalbuf(ovp, vflags, cred, curproc, 0, 0);
/*
* 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 = DIP(oip, ib[level]);
if (bn != 0) {
error = ffs_indirtrunc(oip, indir_lbn[level],
fsbtodb(fs, bn), lastiblock[level], level, &count);
if (error)
allerror = error;
blocksreleased += count;
if (lastiblock[level] < 0) {
DIP_ASSIGN(oip, ib[level], 0);
ffs_blkfree(oip, bn, fs->fs_bsize);
blocksreleased += nblocks;
}
}
if (lastiblock[level] >= 0)
goto done;
}
/*
* All whole direct blocks or frags.
*/
for (i = NDADDR - 1; i > lastblock; i--) {
long bsize;
bn = DIP(oip, db[i]);
if (bn == 0)
continue;
DIP_ASSIGN(oip, db[i], 0);
bsize = blksize(fs, oip, i);
ffs_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 = DIP(oip, 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);
DIP_ASSIGN(oip, size, length);
newspace = blksize(fs, oip, lastblock);
if (newspace == 0)
panic("ffs_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);
ffs_blkfree(oip, bn, oldspace - newspace);
blocksreleased += btodb(oldspace - newspace);
}
}
done:
#ifdef DIAGNOSTIC
for (level = SINGLE; level <= TRIPLE; level++)
if (newblks[NDADDR + level] != DIP(oip, ib[level]))
panic("ffs_truncate1");
for (i = 0; i < NDADDR; i++)
if (newblks[i] != DIP(oip, db[i]))
panic("ffs_truncate2");
#endif /* DIAGNOSTIC */
/*
* Put back the real size.
*/
DIP_ASSIGN(oip, size, length);
DIP_ADD(oip, blocks, -blocksreleased);
oip->i_flag |= IN_CHANGE;
(void)ufs_quota_free_blocks(oip, blocksreleased, NOCRED);
return (allerror);
}
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);
}