/*
* Allocate a block in the filesystem.
*
* A preference may be optionally specified. If a preference is given
* the following hierarchy is used to allocate a block:
* 1) allocate the requested block.
* 2) allocate a rotationally optimal block in the same cylinder.
* 3) allocate a block in the same cylinder group.
* 4) quadradically rehash into other cylinder groups, until an
* available block is located.
* If no block preference is given the following hierarchy is used
* to allocate a block:
* 1) allocate a block in the cylinder group that contains the
* inode for the file.
* 2) quadradically rehash into other cylinder groups, until an
* available block is located.
*/
int
ext2_alloc(struct inode *ip, daddr_t lbn, e4fs_daddr_t bpref, int size,
struct ucred *cred, e4fs_daddr_t *bnp)
{
struct m_ext2fs *fs;
struct ext2mount *ump;
int32_t bno;
int cg;
*bnp = 0;
fs = ip->i_e2fs;
ump = ip->i_ump;
mtx_assert(EXT2_MTX(ump), MA_OWNED);
#ifdef INVARIANTS
if ((u_int)size > fs->e2fs_bsize || blkoff(fs, size) != 0) {
vn_printf(ip->i_devvp, "bsize = %lu, size = %d, fs = %s\n",
(long unsigned int)fs->e2fs_bsize, size, fs->e2fs_fsmnt);
panic("ext2_alloc: bad size");
}
if (cred == NOCRED)
panic("ext2_alloc: missing credential");
#endif /* INVARIANTS */
if (size == fs->e2fs_bsize && fs->e2fs->e2fs_fbcount == 0)
goto nospace;
if (cred->cr_uid != 0 &&
fs->e2fs->e2fs_fbcount < fs->e2fs->e2fs_rbcount)
goto nospace;
if (bpref >= fs->e2fs->e2fs_bcount)
bpref = 0;
if (bpref == 0)
cg = ino_to_cg(fs, ip->i_number);
else
cg = dtog(fs, bpref);
bno = (daddr_t)ext2_hashalloc(ip, cg, bpref, fs->e2fs_bsize,
ext2_alloccg);
if (bno > 0) {
/* set next_alloc fields as done in block_getblk */
ip->i_next_alloc_block = lbn;
ip->i_next_alloc_goal = bno;
ip->i_blocks += btodb(fs->e2fs_bsize);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
*bnp = bno;
return (0);
}
nospace:
EXT2_UNLOCK(ump);
ext2_fserr(fs, cred->cr_uid, "filesystem full");
uprintf("\n%s: write failed, filesystem is full\n", fs->e2fs_fsmnt);
return (ENOSPC);
}
/*
* 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 */
}
/*
* Allocate an inode in the filesystem.
*
*/
int
ext2_valloc(struct vnode *pvp, int mode, struct ucred *cred, struct vnode **vpp)
{
struct timespec ts;
struct inode *pip;
struct m_ext2fs *fs;
struct inode *ip;
struct ext2mount *ump;
ino_t ino, ipref;
int i, error, cg;
*vpp = NULL;
pip = VTOI(pvp);
fs = pip->i_e2fs;
ump = pip->i_ump;
EXT2_LOCK(ump);
if (fs->e2fs->e2fs_ficount == 0)
goto noinodes;
/*
* If it is a directory then obtain a cylinder group based on
* ext2_dirpref else obtain it using ino_to_cg. The preferred inode is
* always the next inode.
*/
if ((mode & IFMT) == IFDIR) {
cg = ext2_dirpref(pip);
if (fs->e2fs_contigdirs[cg] < 255)
fs->e2fs_contigdirs[cg]++;
} else {
cg = ino_to_cg(fs, pip->i_number);
if (fs->e2fs_contigdirs[cg] > 0)
fs->e2fs_contigdirs[cg]--;
}
ipref = cg * fs->e2fs->e2fs_ipg + 1;
ino = (ino_t)ext2_hashalloc(pip, cg, (long)ipref, mode, ext2_nodealloccg);
if (ino == 0)
goto noinodes;
error = VFS_VGET(pvp->v_mount, ino, LK_EXCLUSIVE, vpp);
if (error) {
ext2_vfree(pvp, ino, mode);
return (error);
}
ip = VTOI(*vpp);
/*
* The question is whether using VGET was such good idea at all:
* Linux doesn't read the old inode in when it is allocating a
* new one. I will set at least i_size and i_blocks to zero.
*/
ip->i_size = 0;
ip->i_blocks = 0;
ip->i_mode = 0;
ip->i_flags = 0;
/* now we want to make sure that the block pointers are zeroed out */
for (i = 0; i < NDADDR; i++)
ip->i_db[i] = 0;
for (i = 0; i < NIADDR; i++)
ip->i_ib[i] = 0;
/*
* Set up a new generation number for this inode.
* XXX check if this makes sense in ext2
*/
if (ip->i_gen == 0 || ++ip->i_gen == 0)
ip->i_gen = random() / 2 + 1;
vfs_timestamp(&ts);
ip->i_birthtime = ts.tv_sec;
ip->i_birthnsec = ts.tv_nsec;
/*
printf("ext2_valloc: allocated inode %d\n", ino);
*/
return (0);
noinodes:
EXT2_UNLOCK(ump);
ext2_fserr(fs, cred->cr_uid, "out of inodes");
uprintf("\n%s: create/symlink failed, no inodes free\n", fs->e2fs_fsmnt);
return (ENOSPC);
}
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);
}