/*
* Rewrite an existing directory entry to point at the inode
* supplied. The parameters describing the directory entry are
* set up by a call to namei.
*/
int
ufs_dirrewrite(struct inode *dp, struct inode *oip, ufsino_t newinum,
int newtype, int isrmdir)
{
struct buf *bp;
struct direct *ep;
struct vnode *vdp = ITOV(dp);
int error;
error = UFS_BUFATOFF(dp, (off_t)dp->i_offset, (char **)&ep, &bp);
if (error)
return (error);
ep->d_ino = newinum;
if (vdp->v_mount->mnt_maxsymlinklen > 0)
ep->d_type = newtype;
oip->i_effnlink--;
if (DOINGSOFTDEP(vdp)) {
softdep_change_linkcnt(oip, 0);
softdep_setup_directory_change(bp, dp, oip, newinum, isrmdir);
bdwrite(bp);
} else {
DIP_ADD(oip, nlink, -1);
oip->i_flag |= IN_CHANGE;
UFS_WAPBL_UPDATE(oip, MNT_WAIT);
if (DOINGASYNC(vdp)) {
bdwrite(bp);
error = 0;
} else {
error = VOP_BWRITE(bp);
}
}
dp->i_flag |= IN_CHANGE | IN_UPDATE;
UFS_WAPBL_UPDATE(dp, MNT_WAIT);
return (error);
}
/*
* Update the access, modified, and inode change times as specified by the
* IN_ACCESS, IN_UPDATE, and IN_CHANGE flags respectively. Write the inode
* to disk if the IN_MODIFIED flag is set (it may be set initially, or by
* the timestamp update). The IN_LAZYMOD flag is set to force a write
* later if not now. If we write now, then clear both IN_MODIFIED and
* IN_LAZYMOD to reflect the presumably successful write, and if waitfor is
* set, then wait for the write to complete.
*/
int
ffs_update(struct vnode *vp, int waitfor)
{
struct fs *fs;
struct buf *bp;
struct inode *ip;
int error;
ufs_itimes(vp);
ip = VTOI(vp);
if ((ip->i_flag & IN_MODIFIED) == 0 && waitfor == 0)
return (0);
ip->i_flag &= ~(IN_LAZYMOD | IN_MODIFIED);
fs = ip->i_fs;
if (fs->fs_ronly)
return (0);
/*
* The vnode type is usually set to VBAD if an unrecoverable I/O
* error has occured (such as when reading the inode). Clear the
* modified bits but do not write anything out in this case.
*/
if (vp->v_type == VBAD)
return (0);
/*
* Ensure that uid and gid are correct. This is a temporary
* fix until fsck has been changed to do the update.
*/
if (fs->fs_inodefmt < FS_44INODEFMT) { /* XXX */
ip->i_din.di_ouid = ip->i_uid; /* XXX */
ip->i_din.di_ogid = ip->i_gid; /* XXX */
} /* XXX */
error = bread(ip->i_devvp,
fsbtodoff(fs, ino_to_fsba(fs, ip->i_number)),
(int)fs->fs_bsize, &bp);
if (error) {
brelse(bp);
return (error);
}
if (DOINGSOFTDEP(vp))
softdep_update_inodeblock(ip, bp, waitfor);
else if (ip->i_effnlink != ip->i_nlink)
panic("ffs_update: bad link cnt");
*((struct ufs1_dinode *)bp->b_data +
ino_to_fsbo(fs, ip->i_number)) = ip->i_din;
if (waitfor && !DOINGASYNC(vp)) {
return (bwrite(bp));
} else if (vm_page_count_severe() || buf_dirty_count_severe()) {
return (bwrite(bp));
} else {
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
return (0);
}
}
/*
* ffs_balloc(struct vnode *a_vp, ufs_daddr_t a_lbn, int a_size,
* struct ucred *a_cred, int a_flags, struct buf *a_bpp)
*
* Balloc defines the structure of filesystem storage by allocating
* the physical blocks on a device given the inode and the logical
* block number in a file.
*
* NOTE: B_CLRBUF - this flag tells balloc to clear invalid portions
* of the buffer. However, any dirty bits will override missing
* valid bits. This case occurs when writable mmaps are truncated
* and then extended.
*/
int
ffs_balloc(struct vop_balloc_args *ap)
{
struct inode *ip;
ufs_daddr_t lbn;
int size;
struct ucred *cred;
int flags;
struct fs *fs;
ufs_daddr_t nb;
struct buf *bp, *nbp, *dbp;
struct vnode *vp;
struct indir indirs[NIADDR + 2];
ufs_daddr_t newb, *bap, pref;
int deallocated, osize, nsize, num, i, error;
ufs_daddr_t *allocib, *blkp, *allocblk, allociblk[NIADDR + 1];
ufs_daddr_t *lbns_remfree, lbns[NIADDR + 1];
int unwindidx;
int seqcount;
vp = ap->a_vp;
ip = VTOI(vp);
fs = ip->i_fs;
lbn = lblkno(fs, ap->a_startoffset);
size = blkoff(fs, ap->a_startoffset) + ap->a_size;
if (size > fs->fs_bsize)
panic("ffs_balloc: blk too big");
*ap->a_bpp = NULL;
if (lbn < 0)
return (EFBIG);
cred = ap->a_cred;
flags = ap->a_flags;
/*
* The vnode must be locked for us to be able to safely mess
* around with the inode.
*/
if (vn_islocked(vp) != LK_EXCLUSIVE) {
panic("ffs_balloc: vnode %p not exclusively locked!", vp);
}
/*
* If the next write will extend the file into a new block,
* and the file is currently composed of a fragment
* this fragment has to be extended to be a full block.
*/
nb = lblkno(fs, ip->i_size);
if (nb < NDADDR && nb < lbn) {
/*
* The filesize prior to this write can fit in direct
* blocks (ex. fragmentation is possibly done)
* we are now extending the file write beyond
* the block which has end of the file prior to this write.
*/
osize = blksize(fs, ip, nb);
/*
* osize gives disk allocated size in the last block. It is
* either in fragments or a file system block size.
*/
if (osize < fs->fs_bsize && osize > 0) {
/* A few fragments are already allocated, since the
* current extends beyond this block allocated the
* complete block as fragments are on in last block.
*/
error = ffs_realloccg(ip, nb,
ffs_blkpref(ip, nb, (int)nb, &ip->i_db[0]),
osize, (int)fs->fs_bsize, cred, &bp);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, nb,
dofftofsb(fs, bp->b_bio2.bio_offset),
ip->i_db[nb], fs->fs_bsize, osize, bp);
/* adjust the inode size, we just grew */
ip->i_size = smalllblktosize(fs, nb + 1);
ip->i_db[nb] = dofftofsb(fs, bp->b_bio2.bio_offset);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (flags & B_SYNC)
bwrite(bp);
else
bawrite(bp);
/* bp is already released here */
}
}
/*
* The first NDADDR blocks are direct blocks
*/
if (lbn < NDADDR) {
nb = ip->i_db[lbn];
if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) {
error = bread(vp, lblktodoff(fs, lbn), fs->fs_bsize, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_bio2.bio_offset = fsbtodoff(fs, nb);
*ap->a_bpp = bp;
return (0);
}
if (nb != 0) {
/*
* Consider need to reallocate a fragment.
*/
osize = fragroundup(fs, blkoff(fs, ip->i_size));
nsize = fragroundup(fs, size);
if (nsize <= osize) {
error = bread(vp, lblktodoff(fs, lbn),
osize, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_bio2.bio_offset = fsbtodoff(fs, nb);
} else {
/*
* NOTE: ffs_realloccg() issues a bread().
*/
error = ffs_realloccg(ip, lbn,
ffs_blkpref(ip, lbn, (int)lbn,
&ip->i_db[0]), osize, nsize, cred, &bp);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, lbn,
dofftofsb(fs, bp->b_bio2.bio_offset),
nb, nsize, osize, bp);
}
} else {
if (ip->i_size < smalllblktosize(fs, lbn + 1))
nsize = fragroundup(fs, size);
else
nsize = fs->fs_bsize;
error = ffs_alloc(ip, lbn,
ffs_blkpref(ip, lbn, (int)lbn, &ip->i_db[0]),
nsize, cred, &newb);
if (error)
return (error);
bp = getblk(vp, lblktodoff(fs, lbn), nsize, 0, 0);
bp->b_bio2.bio_offset = fsbtodoff(fs, newb);
if (flags & B_CLRBUF)
vfs_bio_clrbuf(bp);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, lbn, newb, 0,
nsize, 0, bp);
}
ip->i_db[lbn] = dofftofsb(fs, bp->b_bio2.bio_offset);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
*ap->a_bpp = bp;
return (0);
}
/*
* Determine the number of levels of indirection.
*/
pref = 0;
if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0)
return(error);
#ifdef DIAGNOSTIC
if (num < 1)
panic ("ffs_balloc: ufs_bmaparray returned indirect block");
#endif
/*
* Get a handle on the data block buffer before working through
* indirect blocks to avoid a deadlock between the VM system holding
* a locked VM page and issuing a BMAP (which tries to lock the
* indirect blocks), and the filesystem holding a locked indirect
* block and then trying to read a data block (which tries to lock
* the underlying VM pages).
*/
dbp = getblk(vp, lblktodoff(fs, lbn), fs->fs_bsize, 0, 0);
/*
* Setup undo history
*/
allocib = NULL;
allocblk = allociblk;
lbns_remfree = lbns;
unwindidx = -1;
/*
* Fetch the first indirect block directly from the inode, allocating
* one if necessary.
*/
--num;
nb = ip->i_ib[indirs[0].in_off];
if (nb == 0) {
pref = ffs_blkpref(ip, lbn, 0, NULL);
/*
* If the filesystem has run out of space we can skip the
* full fsync/undo of the main [fail] case since no undo
* history has been built yet. Hence the goto fail2.
*/
if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
cred, &newb)) != 0)
goto fail2;
nb = newb;
*allocblk++ = nb;
*lbns_remfree++ = indirs[1].in_lbn;
bp = getblk(vp, lblktodoff(fs, indirs[1].in_lbn),
fs->fs_bsize, 0, 0);
bp->b_bio2.bio_offset = fsbtodoff(fs, nb);
vfs_bio_clrbuf(bp);
if (DOINGSOFTDEP(vp)) {
softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off,
newb, 0, fs->fs_bsize, 0, bp);
bdwrite(bp);
} else {
/*
* Write synchronously so that indirect blocks
* never point at garbage.
*/
if (DOINGASYNC(vp))
bdwrite(bp);
else if ((error = bwrite(bp)) != 0)
goto fail;
}
allocib = &ip->i_ib[indirs[0].in_off];
*allocib = nb;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
}
/*
* Fetch through the indirect blocks, allocating as necessary.
*/
for (i = 1;;) {
error = bread(vp, lblktodoff(fs, indirs[i].in_lbn), (int)fs->fs_bsize, &bp);
if (error) {
brelse(bp);
goto fail;
}
bap = (ufs_daddr_t *)bp->b_data;
nb = bap[indirs[i].in_off];
if (i == num)
break;
i += 1;
if (nb != 0) {
bqrelse(bp);
continue;
}
if (pref == 0)
pref = ffs_blkpref(ip, lbn, 0, NULL);
if ((error =
ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, &newb)) != 0) {
brelse(bp);
goto fail;
}
nb = newb;
*allocblk++ = nb;
*lbns_remfree++ = indirs[i].in_lbn;
nbp = getblk(vp, lblktodoff(fs, indirs[i].in_lbn),
fs->fs_bsize, 0, 0);
nbp->b_bio2.bio_offset = fsbtodoff(fs, nb);
vfs_bio_clrbuf(nbp);
if (DOINGSOFTDEP(vp)) {
softdep_setup_allocindir_meta(nbp, ip, bp,
indirs[i - 1].in_off, nb);
bdwrite(nbp);
} else {
/*
* Write synchronously so that indirect blocks
* never point at garbage.
*/
if ((error = bwrite(nbp)) != 0) {
brelse(bp);
goto fail;
}
}
bap[indirs[i - 1].in_off] = nb;
if (allocib == NULL && unwindidx < 0)
unwindidx = i - 1;
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & B_SYNC) {
bwrite(bp);
} else {
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
}
}
/*
* Get the data block, allocating if necessary. We have already
* called getblk() on the data block buffer, dbp. If we have to
* allocate it and B_CLRBUF has been set the inference is an intention
* to zero out the related disk blocks, so we do not have to issue
* a read. Instead we simply call vfs_bio_clrbuf(). If B_CLRBUF is
* not set the caller intends to overwrite the entire contents of the
* buffer and we don't waste time trying to clean up the contents.
*
* bp references the current indirect block. When allocating,
* the block must be updated.
*/
if (nb == 0) {
pref = ffs_blkpref(ip, lbn, indirs[i].in_off, &bap[0]);
error = ffs_alloc(ip,
lbn, pref, (int)fs->fs_bsize, cred, &newb);
if (error) {
brelse(bp);
goto fail;
}
nb = newb;
*allocblk++ = nb;
*lbns_remfree++ = lbn;
dbp->b_bio2.bio_offset = fsbtodoff(fs, nb);
if (flags & B_CLRBUF)
vfs_bio_clrbuf(dbp);
if (DOINGSOFTDEP(vp))
softdep_setup_allocindir_page(ip, lbn, bp,
indirs[i].in_off, nb, 0, dbp);
bap[indirs[i].in_off] = nb;
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & B_SYNC) {
bwrite(bp);
} else {
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
}
*ap->a_bpp = dbp;
return (0);
}
brelse(bp);
/*
* At this point all related indirect blocks have been allocated
* if necessary and released. bp is no longer valid. dbp holds
* our getblk()'d data block.
*
* XXX we previously performed a cluster_read operation here.
*/
if (flags & B_CLRBUF) {
/*
* If B_CLRBUF is set we must validate the invalid portions
* of the buffer. This typically requires a read-before-
* write. The strategy call will fill in bio_offset in that
* case.
*
* If we hit this case we do a cluster read if possible
* since nearby data blocks are likely to be accessed soon
* too.
*/
if ((dbp->b_flags & B_CACHE) == 0) {
bqrelse(dbp);
seqcount = (flags & B_SEQMASK) >> B_SEQSHIFT;
if (seqcount &&
(vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
error = cluster_read(vp, (off_t)ip->i_size,
lblktodoff(fs, lbn),
(int)fs->fs_bsize,
fs->fs_bsize,
seqcount * BKVASIZE,
&dbp);
} else {
error = bread(vp, lblktodoff(fs, lbn),
(int)fs->fs_bsize, &dbp);
}
if (error)
goto fail;
} else {
/*
* Balloc defines the structure of filesystem storage
* by allocating the physical blocks on a device given
* the inode and the logical block number in a file.
* This is the allocation strategy for UFS1. Below is
* the allocation strategy for UFS2.
*/
int
ffs_balloc_ufs1(struct vnode *vp, off_t startoffset, int size,
struct ucred *cred, int flags, struct buf **bpp)
{
struct inode *ip;
struct ufs1_dinode *dp;
ufs_lbn_t lbn, lastlbn;
struct fs *fs;
ufs1_daddr_t nb;
struct buf *bp, *nbp;
struct ufsmount *ump;
struct indir indirs[NIADDR + 2];
int deallocated, osize, nsize, num, i, error;
ufs2_daddr_t newb;
ufs1_daddr_t *bap, pref;
ufs1_daddr_t *allocib, *blkp, *allocblk, allociblk[NIADDR + 1];
ufs2_daddr_t *lbns_remfree, lbns[NIADDR + 1];
int unwindidx = -1;
int saved_inbdflush;
static struct timeval lastfail;
static int curfail;
int reclaimed;
ip = VTOI(vp);
dp = ip->i_din1;
fs = ip->i_fs;
ump = ip->i_ump;
lbn = lblkno(fs, startoffset);
size = blkoff(fs, startoffset) + size;
reclaimed = 0;
if (size > fs->fs_bsize)
panic("ffs_balloc_ufs1: blk too big");
*bpp = NULL;
if (flags & IO_EXT)
return (EOPNOTSUPP);
if (lbn < 0)
return (EFBIG);
if (DOINGSOFTDEP(vp))
softdep_prealloc(vp, MNT_WAIT);
/*
* If the next write will extend the file into a new block,
* and the file is currently composed of a fragment
* this fragment has to be extended to be a full block.
*/
lastlbn = lblkno(fs, ip->i_size);
if (lastlbn < NDADDR && lastlbn < lbn) {
nb = lastlbn;
osize = blksize(fs, ip, nb);
if (osize < fs->fs_bsize && osize > 0) {
UFS_LOCK(ump);
error = ffs_realloccg(ip, nb, dp->di_db[nb],
ffs_blkpref_ufs1(ip, lastlbn, (int)nb,
&dp->di_db[0]), osize, (int)fs->fs_bsize, flags,
cred, &bp);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, nb,
dbtofsb(fs, bp->b_blkno), dp->di_db[nb],
fs->fs_bsize, osize, bp);
ip->i_size = smalllblktosize(fs, nb + 1);
dp->di_size = ip->i_size;
dp->di_db[nb] = dbtofsb(fs, bp->b_blkno);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (flags & IO_SYNC)
bwrite(bp);
else
bawrite(bp);
}
}
/*
* The first NDADDR blocks are direct blocks
*/
if (lbn < NDADDR) {
if (flags & BA_METAONLY)
panic("ffs_balloc_ufs1: BA_METAONLY for direct block");
nb = dp->di_db[lbn];
if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) {
error = bread(vp, lbn, fs->fs_bsize, NOCRED, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_blkno = fsbtodb(fs, nb);
*bpp = bp;
return (0);
}
if (nb != 0) {
/*
* Consider need to reallocate a fragment.
*/
osize = fragroundup(fs, blkoff(fs, ip->i_size));
nsize = fragroundup(fs, size);
if (nsize <= osize) {
error = bread(vp, lbn, osize, NOCRED, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_blkno = fsbtodb(fs, nb);
} else {
UFS_LOCK(ump);
error = ffs_realloccg(ip, lbn, dp->di_db[lbn],
ffs_blkpref_ufs1(ip, lbn, (int)lbn,
&dp->di_db[0]), osize, nsize, flags,
cred, &bp);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, lbn,
dbtofsb(fs, bp->b_blkno), nb,
nsize, osize, bp);
}
} else {
if (ip->i_size < smalllblktosize(fs, lbn + 1))
nsize = fragroundup(fs, size);
else
nsize = fs->fs_bsize;
UFS_LOCK(ump);
error = ffs_alloc(ip, lbn,
ffs_blkpref_ufs1(ip, lbn, (int)lbn, &dp->di_db[0]),
nsize, flags, cred, &newb);
if (error)
return (error);
bp = getblk(vp, lbn, nsize, 0, 0, 0);
bp->b_blkno = fsbtodb(fs, newb);
if (flags & BA_CLRBUF)
vfs_bio_clrbuf(bp);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, lbn, newb, 0,
nsize, 0, bp);
}
dp->di_db[lbn] = dbtofsb(fs, bp->b_blkno);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
*bpp = bp;
return (0);
}
/*
* Determine the number of levels of indirection.
*/
pref = 0;
if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0)
return(error);
#ifdef INVARIANTS
if (num < 1)
panic ("ffs_balloc_ufs1: ufs_getlbns returned indirect block");
#endif
saved_inbdflush = curthread_pflags_set(TDP_INBDFLUSH);
/*
* Fetch the first indirect block allocating if necessary.
*/
--num;
nb = dp->di_ib[indirs[0].in_off];
allocib = NULL;
allocblk = allociblk;
lbns_remfree = lbns;
if (nb == 0) {
UFS_LOCK(ump);
pref = ffs_blkpref_ufs1(ip, lbn, 0, (ufs1_daddr_t *)0);
if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
flags, cred, &newb)) != 0) {
curthread_pflags_restore(saved_inbdflush);
return (error);
}
nb = newb;
*allocblk++ = nb;
*lbns_remfree++ = indirs[1].in_lbn;
bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0, 0);
bp->b_blkno = fsbtodb(fs, nb);
vfs_bio_clrbuf(bp);
if (DOINGSOFTDEP(vp)) {
softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off,
newb, 0, fs->fs_bsize, 0, bp);
bdwrite(bp);
} else {
/*
* Write synchronously so that indirect blocks
* never point at garbage.
*/
if (DOINGASYNC(vp))
bdwrite(bp);
else if ((error = bwrite(bp)) != 0)
goto fail;
}
allocib = &dp->di_ib[indirs[0].in_off];
*allocib = nb;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
}
/*
* Fetch through the indirect blocks, allocating as necessary.
*/
retry:
for (i = 1;;) {
error = bread(vp,
indirs[i].in_lbn, (int)fs->fs_bsize, NOCRED, &bp);
if (error) {
brelse(bp);
goto fail;
}
bap = (ufs1_daddr_t *)bp->b_data;
nb = bap[indirs[i].in_off];
if (i == num)
break;
i += 1;
if (nb != 0) {
bqrelse(bp);
continue;
}
UFS_LOCK(ump);
if (pref == 0)
pref = ffs_blkpref_ufs1(ip, lbn, 0, (ufs1_daddr_t *)0);
if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
flags | IO_BUFLOCKED, cred, &newb)) != 0) {
brelse(bp);
if (++reclaimed == 1) {
UFS_LOCK(ump);
softdep_request_cleanup(fs, vp, cred,
FLUSH_BLOCKS_WAIT);
UFS_UNLOCK(ump);
goto retry;
}
if (ppsratecheck(&lastfail, &curfail, 1)) {
ffs_fserr(fs, ip->i_number, "filesystem full");
uprintf("\n%s: write failed, filesystem "
"is full\n", fs->fs_fsmnt);
}
goto fail;
}
nb = newb;
*allocblk++ = nb;
*lbns_remfree++ = indirs[i].in_lbn;
nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0, 0);
nbp->b_blkno = fsbtodb(fs, nb);
vfs_bio_clrbuf(nbp);
if (DOINGSOFTDEP(vp)) {
softdep_setup_allocindir_meta(nbp, ip, bp,
indirs[i - 1].in_off, nb);
bdwrite(nbp);
} else {
/*
* Write synchronously so that indirect blocks
* never point at garbage.
*/
if ((error = bwrite(nbp)) != 0) {
brelse(bp);
goto fail;
}
}
bap[indirs[i - 1].in_off] = nb;
if (allocib == NULL && unwindidx < 0)
unwindidx = i - 1;
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & IO_SYNC) {
bwrite(bp);
} else {
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
}
}
/*
* If asked only for the indirect block, then return it.
*/
if (flags & BA_METAONLY) {
curthread_pflags_restore(saved_inbdflush);
*bpp = bp;
return (0);
}
/*
* Get the data block, allocating if necessary.
*/
if (nb == 0) {
UFS_LOCK(ump);
pref = ffs_blkpref_ufs1(ip, lbn, indirs[i].in_off, &bap[0]);
error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
flags | IO_BUFLOCKED, cred, &newb);
if (error) {
brelse(bp);
if (++reclaimed == 1) {
UFS_LOCK(ump);
softdep_request_cleanup(fs, vp, cred,
FLUSH_BLOCKS_WAIT);
UFS_UNLOCK(ump);
goto retry;
}
if (ppsratecheck(&lastfail, &curfail, 1)) {
ffs_fserr(fs, ip->i_number, "filesystem full");
uprintf("\n%s: write failed, filesystem "
"is full\n", fs->fs_fsmnt);
}
goto fail;
}
nb = newb;
*allocblk++ = nb;
*lbns_remfree++ = lbn;
nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0, 0);
nbp->b_blkno = fsbtodb(fs, nb);
if (flags & BA_CLRBUF)
vfs_bio_clrbuf(nbp);
if (DOINGSOFTDEP(vp))
softdep_setup_allocindir_page(ip, lbn, bp,
indirs[i].in_off, nb, 0, nbp);
bap[indirs[i].in_off] = nb;
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & IO_SYNC) {
bwrite(bp);
} else {
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
}
curthread_pflags_restore(saved_inbdflush);
*bpp = nbp;
return (0);
}
brelse(bp);
if (flags & BA_CLRBUF) {
int seqcount = (flags & BA_SEQMASK) >> BA_SEQSHIFT;
if (seqcount && (vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
error = cluster_read(vp, ip->i_size, lbn,
(int)fs->fs_bsize, NOCRED,
MAXBSIZE, seqcount, &nbp);
} else {
error = bread(vp, lbn, (int)fs->fs_bsize, NOCRED, &nbp);
}
if (error) {
brelse(nbp);
goto fail;
}
} else {
/*
* Update the access, modified, and inode change times as specified by the
* IN_ACCESS, IN_UPDATE, and IN_CHANGE flags respectively. Write the inode
* to disk if the IN_MODIFIED flag is set (it may be set initially, or by
* the timestamp update). The IN_LAZYMOD flag is set to force a write
* later if not now. The IN_LAZYACCESS is set instead of IN_MODIFIED if the fs
* is currently being suspended (or is suspended) and vnode has been accessed.
* If we write now, then clear IN_MODIFIED, IN_LAZYACCESS and IN_LAZYMOD to
* reflect the presumably successful write, and if waitfor is set, then wait
* for the write to complete.
*/
int
ffs_update (vnode *vp, int waitfor)
{
int error = 0;
print("HARVEY TODO: %s\n", __func__);
#if 0
struct fs *fs;
struct buf *bp;
struct inode *ip;
int flags, error;
ASSERT_VOP_ELOCKED(vp, "ffs_update");
ufs_itimes(vp);
ip = VTOI(vp);
if ((ip->i_flag & IN_MODIFIED) == 0 && waitfor == 0)
return (0);
ip->i_flag &= ~(IN_LAZYACCESS | IN_LAZYMOD | IN_MODIFIED);
fs = ITOFS(ip);
if (fs->fs_ronly && ITOUMP(ip)->um_fsckpid == 0)
return (0);
/*
* If we are updating a snapshot and another process is currently
* writing the buffer containing the inode for this snapshot then
* a deadlock can occur when it tries to check the snapshot to see
* if that block needs to be copied. Thus when updating a snapshot
* we check to see if the buffer is already locked, and if it is
* we drop the snapshot lock until the buffer has been written
* and is available to us. We have to grab a reference to the
* snapshot vnode to prevent it from being removed while we are
* waiting for the buffer.
*/
flags = 0;
if (IS_SNAPSHOT(ip))
flags = GB_LOCK_NOWAIT;
loop:
error = breadn_flags(ITODEVVP(ip),
fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
(int) fs->fs_bsize, 0, 0, 0, NOCRED, flags, &bp);
if (error != 0) {
if (error != EBUSY)
return (error);
KASSERT((IS_SNAPSHOT(ip)), ("EBUSY from non-snapshot"));
/*
* Wait for our inode block to become available.
*
* Hold a reference to the vnode to protect against
* ffs_snapgone(). Since we hold a reference, it can only
* get reclaimed (VI_DOOMED flag) in a forcible downgrade
* or unmount. For an unmount, the entire filesystem will be
* gone, so we cannot attempt to touch anything associated
* with it while the vnode is unlocked; all we can do is
* pause briefly and try again. If when we relock the vnode
* we discover that it has been reclaimed, updating it is no
* longer necessary and we can just return an error.
*/
vref(vp);
VOP_UNLOCK(vp, 0);
pause("ffsupd", 1);
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
vrele(vp);
if ((vp->v_iflag & VI_DOOMED) != 0)
return (ENOENT);
goto loop;
}
if (DOINGSOFTDEP(vp))
softdep_update_inodeblock(ip, bp, waitfor);
else if (ip->i_effnlink != ip->i_nlink)
panic("ffs_update: bad link cnt");
if (I_IS_UFS1(ip)) {
*((struct ufs1_dinode *)bp->b_data +
ino_to_fsbo(fs, ip->i_number)) = *ip->i_din1;
/* XXX: FIX? The entropy here is desirable, but the harvesting may be expensive */
random_harvest_queue(&(ip->i_din1), sizeof(ip->i_din1), 1, RANDOM_FS_ATIME);
} else {
*((struct ufs2_dinode *)bp->b_data +
ino_to_fsbo(fs, ip->i_number)) = *ip->i_din2;
/* XXX: FIX? The entropy here is desirable, but the harvesting may be expensive */
random_harvest_queue(&(ip->i_din2), sizeof(ip->i_din2), 1, RANDOM_FS_ATIME);
}
if (waitfor)
error = bwrite(bp);
else if (vm_page_count_severe() || buf_dirty_count_severe()) {
bawrite(bp);
error = 0;
} else {
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
error = 0;
}
#endif // 0
return (error);
}
/*
* Balloc defines the structure of file system storage
* by allocating the physical blocks on a device given
* the inode and the logical block number in a file.
*/
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);
}
/*
* Write a directory entry after a call to namei, using the parameters
* that it left in nameidata. The argument dirp is the new directory
* entry contents. Dvp is a pointer to the directory to be written,
* which was left locked by namei. Remaining parameters (dp->i_offset,
* dp->i_count) indicate how the space for the new entry is to be obtained.
* Non-null bp indicates that a directory is being created (for the
* soft dependency code).
*/
int
ufs_direnter(struct vnode *dvp, struct vnode *tvp, struct direct *dirp,
struct componentname *cnp, struct buf *newdirbp)
{
struct ucred *cr;
struct proc *p;
int newentrysize;
struct inode *dp;
struct buf *bp;
u_int dsize;
struct direct *ep, *nep;
int error, ret, blkoff, loc, spacefree, flags;
char *dirbuf;
UFS_WAPBL_JLOCK_ASSERT(dvp->v_mount);
error = 0;
cr = cnp->cn_cred;
p = cnp->cn_proc;
dp = VTOI(dvp);
newentrysize = DIRSIZ(FSFMT(dvp), dirp);
if (dp->i_count == 0) {
/*
* If dp->i_count is 0, then namei could find no
* space in the directory. Here, dp->i_offset will
* be on a directory block boundary and we will write the
* new entry into a fresh block.
*/
if (dp->i_offset & (DIRBLKSIZ - 1))
panic("ufs_direnter: newblk");
flags = B_CLRBUF;
if (!DOINGSOFTDEP(dvp))
flags |= B_SYNC;
if ((error = UFS_BUF_ALLOC(dp, (off_t)dp->i_offset, DIRBLKSIZ,
cr, flags, &bp)) != 0) {
if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
bdwrite(newdirbp);
return (error);
}
DIP_ASSIGN(dp, size, dp->i_offset + DIRBLKSIZ);
dp->i_flag |= IN_CHANGE | IN_UPDATE;
uvm_vnp_setsize(dvp, DIP(dp, size));
dirp->d_reclen = DIRBLKSIZ;
blkoff = dp->i_offset &
(VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_iosize - 1);
memcpy(bp->b_data + blkoff, dirp, newentrysize);
#ifdef UFS_DIRHASH
if (dp->i_dirhash != NULL) {
ufsdirhash_newblk(dp, dp->i_offset);
ufsdirhash_add(dp, dirp, dp->i_offset);
ufsdirhash_checkblock(dp, (char *)bp->b_data + blkoff,
dp->i_offset);
}
#endif
if (DOINGSOFTDEP(dvp)) {
/*
* Ensure that the entire newly allocated block is a
* valid directory so that future growth within the
* block does not have to ensure that the block is
* written before the inode.
*/
blkoff += DIRBLKSIZ;
while (blkoff < bp->b_bcount) {
((struct direct *)
(bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
blkoff += DIRBLKSIZ;
}
if (softdep_setup_directory_add(bp, dp, dp->i_offset,
dirp->d_ino, newdirbp, 1) == 0) {
bdwrite(bp);
return (UFS_UPDATE(dp, 0));
}
/* We have just allocated a directory block in an
* indirect block. Rather than tracking when it gets
* claimed by the inode, we simply do a VOP_FSYNC
* now to ensure that it is there (in case the user
* does a future fsync). Note that we have to unlock
* the inode for the entry that we just entered, as
* the VOP_FSYNC may need to lock other inodes which
* can lead to deadlock if we also hold a lock on
* the newly entered node.
*/
if ((error = VOP_BWRITE(bp)))
return (error);
if (tvp != NULL)
VOP_UNLOCK(tvp, 0);
error = VOP_FSYNC(dvp, p->p_ucred, MNT_WAIT);
if (tvp != NULL)
vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY, p);
return (error);
}
error = VOP_BWRITE(bp);
ret = UFS_UPDATE(dp, !DOINGSOFTDEP(dvp));
if (error == 0)
return (ret);
return (error);
}
/*
* If dp->i_count is non-zero, then namei found space for the new
* entry in the range dp->i_offset to dp->i_offset + dp->i_count
* in the directory. To use this space, we may have to compact
* the entries located there, by copying them together towards the
* beginning of the block, leaving the free space in one usable
* chunk at the end.
*/
/*
* Increase size of directory if entry eats into new space.
* This should never push the size past a new multiple of
* DIRBLKSIZE.
*
* N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
*/
if (dp->i_offset + dp->i_count > DIP(dp, size)) {
DIP_ASSIGN(dp, size, dp->i_offset + dp->i_count);
dp->i_flag |= IN_CHANGE | IN_UPDATE;
UFS_WAPBL_UPDATE(dp, MNT_WAIT);
}
/*
* Get the block containing the space for the new directory entry.
*/
if ((error = UFS_BUFATOFF(dp, (off_t)dp->i_offset, &dirbuf, &bp))
!= 0) {
if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
bdwrite(newdirbp);
return (error);
}
/*
* Find space for the new entry. In the simple case, the entry at
* offset base will have the space. If it does not, then namei
* arranged that compacting the region dp->i_offset to
* dp->i_offset + dp->i_count would yield the space.
*/
ep = (struct direct *)dirbuf;
dsize = ep->d_ino ? DIRSIZ(FSFMT(dvp), ep) : 0;
spacefree = ep->d_reclen - dsize;
for (loc = ep->d_reclen; loc < dp->i_count; ) {
nep = (struct direct *)(dirbuf + loc);
/* Trim the existing slot (NB: dsize may be zero). */
ep->d_reclen = dsize;
ep = (struct direct *)((char *)ep + dsize);
/* Read nep->d_reclen now as the memmove() may clobber it. */
loc += nep->d_reclen;
if (nep->d_ino == 0) {
/*
* A mid-block unused entry. Such entries are
* never created by the kernel, but fsck_ffs
* can create them (and it doesn't fix them).
*
* Add up the free space, and initialise the
* relocated entry since we don't memmove it.
*/
spacefree += nep->d_reclen;
ep->d_ino = 0;
dsize = 0;
continue;
}
dsize = DIRSIZ(FSFMT(dvp), nep);
spacefree += nep->d_reclen - dsize;
#ifdef UFS_DIRHASH
if (dp->i_dirhash != NULL)
ufsdirhash_move(dp, nep,
dp->i_offset + ((char *)nep - dirbuf),
dp->i_offset + ((char *)ep - dirbuf));
#endif
if (DOINGSOFTDEP(dvp))
softdep_change_directoryentry_offset(dp, dirbuf,
(caddr_t)nep, (caddr_t)ep, dsize);
else
memmove(ep, nep, dsize);
}
/*
* Here, `ep' points to a directory entry containing `dsize' in-use
* bytes followed by `spacefree' unused bytes. If ep->d_ino == 0,
* then the entry is completely unused (dsize == 0). The value
* of ep->d_reclen is always indeterminate.
*
* Update the pointer fields in the previous entry (if any),
* copy in the new entry, and write out the block.
*/
if (ep->d_ino == 0) {
if (spacefree + dsize < newentrysize)
panic("ufs_direnter: compact1");
dirp->d_reclen = spacefree + dsize;
} else {
if (spacefree < newentrysize)
panic("ufs_direnter: compact2");
dirp->d_reclen = spacefree;
ep->d_reclen = dsize;
ep = (struct direct *)((char *)ep + dsize);
}
#ifdef UFS_DIRHASH
if (dp->i_dirhash != NULL && (ep->d_ino == 0 ||
dirp->d_reclen == spacefree))
ufsdirhash_add(dp, dirp, dp->i_offset + ((char *)ep - dirbuf));
#endif
memcpy(ep, dirp, newentrysize);
#ifdef UFS_DIRHASH
if (dp->i_dirhash != NULL)
ufsdirhash_checkblock(dp, dirbuf -
(dp->i_offset & (DIRBLKSIZ - 1)),
dp->i_offset & ~(DIRBLKSIZ - 1));
#endif
if (DOINGSOFTDEP(dvp)) {
(void)softdep_setup_directory_add(bp, dp,
dp->i_offset + (caddr_t)ep - dirbuf,
dirp->d_ino, newdirbp, 0);
bdwrite(bp);
} else {
error = VOP_BWRITE(bp);
}
dp->i_flag |= IN_CHANGE | IN_UPDATE;
/*
* If all went well, and the directory can be shortened, proceed
* with the truncation. Note that we have to unlock the inode for
* the entry that we just entered, as the truncation may need to
* lock other inodes which can lead to deadlock if we also hold a
* lock on the newly entered node.
*/
if (error == 0 && dp->i_endoff && dp->i_endoff < DIP(dp, size)) {
if (tvp != NULL)
VOP_UNLOCK(tvp, 0);
#ifdef UFS_DIRHASH
if (dp->i_dirhash != NULL)
ufsdirhash_dirtrunc(dp, dp->i_endoff);
#endif
error = UFS_TRUNCATE(dp, (off_t)dp->i_endoff, IO_SYNC, cr);
if (tvp != NULL)
vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY, p);
}
UFS_WAPBL_UPDATE(dp, MNT_WAIT);
return (error);
}
/*
* Look up a FFS dinode number to find its incore vnode, otherwise read it
* in from disk. If it is in core, wait for the lock bit to clear, then
* return the inode locked. Detection and handling of mount points must be
* done by the calling routine.
*/
int
ffs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
{
struct fs *fs;
struct inode *ip;
struct ufs1_dinode *dp1;
#ifdef FFS2
struct ufs2_dinode *dp2;
#endif
struct ufsmount *ump;
struct buf *bp;
struct vnode *vp;
dev_t dev;
int error;
ump = VFSTOUFS(mp);
dev = ump->um_dev;
retry:
if ((*vpp = ufs_ihashget(dev, ino)) != NULL)
return (0);
/* Allocate a new vnode/inode. */
if ((error = getnewvnode(VT_UFS, mp, ffs_vnodeop_p, &vp)) != 0) {
*vpp = NULL;
return (error);
}
#ifdef VFSDEBUG
vp->v_flag |= VLOCKSWORK;
#endif
/* XXX - we use the same pool for ffs and mfs */
ip = pool_get(&ffs_ino_pool, PR_WAITOK);
bzero((caddr_t)ip, sizeof(struct inode));
lockinit(&ip->i_lock, PINOD, "inode", 0, 0);
ip->i_ump = ump;
VREF(ip->i_devvp);
vp->v_data = ip;
ip->i_vnode = vp;
ip->i_fs = fs = ump->um_fs;
ip->i_dev = dev;
ip->i_number = ino;
ip->i_vtbl = &ffs_vtbl;
/*
* Put it onto its hash chain and lock it so that other requests for
* this inode will block if they arrive while we are sleeping waiting
* for old data structures to be purged or for the contents of the
* disk portion of this inode to be read.
*/
error = ufs_ihashins(ip);
if (error) {
/*
* VOP_INACTIVE will treat this as a stale file
* and recycle it quickly
*/
vrele(vp);
if (error == EEXIST)
goto retry;
return (error);
}
/* Read in the disk contents for the inode, copy into the inode. */
error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
(int)fs->fs_bsize, NOCRED, &bp);
if (error) {
/*
* The inode does not contain anything useful, so it would
* be misleading to leave it on its hash chain. With mode
* still zero, it will be unlinked and returned to the free
* list by vput().
*/
vput(vp);
brelse(bp);
*vpp = NULL;
return (error);
}
#ifdef FFS2
if (ip->i_ump->um_fstype == UM_UFS2) {
ip->i_din2 = pool_get(&ffs_dinode2_pool, PR_WAITOK);
dp2 = (struct ufs2_dinode *) bp->b_data + ino_to_fsbo(fs, ino);
*ip->i_din2 = *dp2;
} else
#endif
{
ip->i_din1 = pool_get(&ffs_dinode1_pool, PR_WAITOK);
dp1 = (struct ufs1_dinode *) bp->b_data + ino_to_fsbo(fs, ino);
*ip->i_din1 = *dp1;
}
brelse(bp);
if (DOINGSOFTDEP(vp))
softdep_load_inodeblock(ip);
else
ip->i_effnlink = DIP(ip, nlink);
/*
* Initialize the vnode from the inode, check for aliases.
* Note that the underlying vnode may have changed.
*/
error = ufs_vinit(mp, ffs_specop_p, FFS_FIFOOPS, &vp);
if (error) {
vput(vp);
*vpp = NULL;
return (error);
}
/*
* Set up a generation number for this inode if it does not
* already have one. This should only happen on old filesystems.
*/
if (DIP(ip, gen) == 0) {
DIP_ASSIGN(ip, gen, arc4random() & INT_MAX);
if (DIP(ip, gen) == 0 || DIP(ip, gen) == -1)
DIP_ASSIGN(ip, gen, 1); /* Shouldn't happen */
if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0)
ip->i_flag |= IN_MODIFIED;
}
/*
* Ensure that uid and gid are correct. This is a temporary
* fix until fsck has been changed to do the update.
*/
if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_inodefmt < FS_44INODEFMT) {
ip->i_ffs1_uid = ip->i_din1->di_ouid;
ip->i_ffs1_gid = ip->i_din1->di_ogid;
}
*vpp = vp;
return (0);
}
/*
* Rmdir system call.
*/
int
ufs_rmdir(void *v)
{
struct vop_rmdir_args *ap = v;
struct vnode *vp = ap->a_vp;
struct vnode *dvp = ap->a_dvp;
struct componentname *cnp = ap->a_cnp;
struct inode *ip, *dp;
int error;
ip = VTOI(vp);
dp = VTOI(dvp);
/*
* No rmdir "." or of mounted on directories.
*/
if (dp == ip || vp->v_mountedhere != NULL) {
if (dp == ip)
vrele(dvp);
else
vput(dvp);
vput(vp);
return (EINVAL);
}
/*
* Do not remove a directory that is in the process of being renamed.
* Verify the directory is empty (and valid). Rmdir ".." will not be
* valid since ".." will contain a reference to the current directory
* and thus be non-empty.
*/
error = 0;
if (ip->i_flag & IN_RENAME) {
error = EINVAL;
goto out;
}
if (ip->i_effnlink != 2 ||
!ufs_dirempty(ip, dp->i_number, cnp->cn_cred)) {
error = ENOTEMPTY;
goto out;
}
if ((DIP(dp, flags) & APPEND) ||
(DIP(ip, flags) & (IMMUTABLE | APPEND))) {
error = EPERM;
goto out;
}
/*
* Delete reference to directory before purging
* inode. If we crash in between, the directory
* will be reattached to lost+found,
*/
dp->i_effnlink--;
ip->i_effnlink--;
if (DOINGSOFTDEP(vp)) {
softdep_change_linkcnt(dp, 0);
softdep_change_linkcnt(ip, 0);
}
if ((error = ufs_dirremove(dvp, ip, cnp->cn_flags, 1)) != 0) {
dp->i_effnlink++;
ip->i_effnlink++;
if (DOINGSOFTDEP(vp)) {
softdep_change_linkcnt(dp, 0);
softdep_change_linkcnt(ip, 0);
}
goto out;
}
VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
cache_purge(dvp);
/*
* Truncate inode. The only stuff left in the directory is "." and
* "..". The "." reference is inconsequential since we are quashing
* it. The soft dependency code will arrange to do these operations
* after the parent directory entry has been deleted on disk, so
* when running with that code we avoid doing them now.
*/
if (!DOINGSOFTDEP(vp)) {
int ioflag;
DIP_ADD(dp, nlink, -1);
dp->i_flag |= IN_CHANGE;
DIP_ADD(ip, nlink, -1);
ip->i_flag |= IN_CHANGE;
ioflag = DOINGASYNC(vp) ? 0 : IO_SYNC;
error = UFS_TRUNCATE(ip, (off_t)0, ioflag, cnp->cn_cred);
}
cache_purge(vp);
#ifdef UFS_DIRHASH
/* Kill any active hash; i_effnlink == 0, so it will not come back. */
if (ip->i_dirhash != NULL)
ufsdirhash_free(ip);
#endif
out:
VN_KNOTE(vp, NOTE_DELETE);
vput(dvp);
vput(vp);
return (error);
}
/*
* Mkdir system call
*/
int
ufs_mkdir(void *v)
{
struct vop_mkdir_args *ap = v;
struct vnode *dvp = ap->a_dvp;
struct vattr *vap = ap->a_vap;
struct componentname *cnp = ap->a_cnp;
struct inode *ip, *dp;
struct vnode *tvp;
struct buf *bp;
struct direct newdir;
struct dirtemplate dirtemplate, *dtp;
int error, dmode, blkoff;
#ifdef DIAGNOSTIC
if ((cnp->cn_flags & HASBUF) == 0)
panic("ufs_mkdir: no name");
#endif
dp = VTOI(dvp);
if ((nlink_t) DIP(dp, nlink) >= LINK_MAX) {
error = EMLINK;
goto out;
}
dmode = vap->va_mode & 0777;
dmode |= IFDIR;
/*
* Must simulate part of ufs_makeinode here to acquire the inode,
* but not have it entered in the parent directory. The entry is
* made later after writing "." and ".." entries.
*/
if ((error = UFS_INODE_ALLOC(dp, dmode, cnp->cn_cred, &tvp)) != 0)
goto out;
ip = VTOI(tvp);
DIP_ASSIGN(ip, uid, cnp->cn_cred->cr_uid);
DIP_ASSIGN(ip, gid, DIP(dp, gid));
if ((error = getinoquota(ip)) ||
(error = ufs_quota_alloc_inode(ip, cnp->cn_cred))) {
pool_put(&namei_pool, cnp->cn_pnbuf);
UFS_INODE_FREE(ip, ip->i_number, dmode);
vput(tvp);
vput(dvp);
return (error);
}
ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
DIP_ASSIGN(ip, mode, dmode);
tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */
ip->i_effnlink = 2;
DIP_ASSIGN(ip, nlink, 2);
if (DOINGSOFTDEP(tvp))
softdep_change_linkcnt(ip, 0);
/*
* Bump link count in parent directory to reflect work done below.
* Should be done before reference is create so cleanup is
* possible if we crash.
*/
dp->i_effnlink++;
DIP_ADD(dp, nlink, 1);
dp->i_flag |= IN_CHANGE;
if (DOINGSOFTDEP(dvp))
softdep_change_linkcnt(dp, 0);
if ((error = UFS_UPDATE(dp, !DOINGSOFTDEP(dvp))) != 0)
goto bad;
/*
* Initialize directory with "." and ".." from static template.
*/
if (dvp->v_mount->mnt_maxsymlinklen > 0)
dtp = &mastertemplate;
else
dtp = (struct dirtemplate *)&omastertemplate;
dirtemplate = *dtp;
dirtemplate.dot_ino = ip->i_number;
dirtemplate.dotdot_ino = dp->i_number;
if ((error = UFS_BUF_ALLOC(ip, (off_t)0, DIRBLKSIZ, cnp->cn_cred,
B_CLRBUF, &bp)) != 0)
goto bad;
DIP_ASSIGN(ip, size, DIRBLKSIZ);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
uvm_vnp_setsize(tvp, DIP(ip, size));
bcopy((caddr_t)&dirtemplate, (caddr_t)bp->b_data, sizeof dirtemplate);
if (DOINGSOFTDEP(tvp)) {
/*
* Ensure that the entire newly allocated block is a
* valid directory so that future growth within the
* block does not have to ensure that the block is
* written before the inode
*/
blkoff = DIRBLKSIZ;
while (blkoff < bp->b_bcount) {
((struct direct *)
(bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
blkoff += DIRBLKSIZ;
}
}
if ((error = UFS_UPDATE(ip, !DOINGSOFTDEP(tvp))) != 0) {
(void)VOP_BWRITE(bp);
goto bad;
}
/*
* Directory set up, now install its entry in the parent directory.
*
* If we are not doing soft dependencies, then we must write out the
* buffer containing the new directory body before entering the new
* name in the parent. If we are doing soft dependencies, then the
* buffer containing the new directory body will be passed to and
* released in the soft dependency code after the code has attached
* an appropriate ordering dependency to the buffer which ensures that
* the buffer is written before the new name is written in the parent.
*/
if (!DOINGSOFTDEP(dvp) && ((error = VOP_BWRITE(bp)) != 0))
goto bad;
ufs_makedirentry(ip, cnp, &newdir);
error = ufs_direnter(dvp, tvp, &newdir, cnp, bp);
bad:
if (error == 0) {
VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
*ap->a_vpp = tvp;
} else {
dp->i_effnlink--;
DIP_ADD(dp, nlink, -1);
dp->i_flag |= IN_CHANGE;
if (DOINGSOFTDEP(dvp))
softdep_change_linkcnt(dp, 0);
/*
* No need to do an explicit VOP_TRUNCATE here, vrele will
* do this for us because we set the link count to 0.
*/
ip->i_effnlink = 0;
DIP_ASSIGN(ip, nlink, 0);
ip->i_flag |= IN_CHANGE;
if (DOINGSOFTDEP(tvp))
softdep_change_linkcnt(ip, 0);
vput(tvp);
}
out:
pool_put(&namei_pool, cnp->cn_pnbuf);
vput(dvp);
return (error);
}
/*
* Last reference to an inode. If necessary, write or delete it.
*/
int
ufs_inactive(void *v)
{
struct vop_inactive_args *ap = v;
struct vnode *vp = ap->a_vp;
struct inode *ip = VTOI(vp);
struct proc *p = ap->a_p;
mode_t mode;
int error = 0;
#ifdef DIAGNOSTIC
extern int prtactive;
if (prtactive && vp->v_usecount != 0)
vprint("ffs_inactive: pushing active", vp);
#endif
/*
* Ignore inodes related to stale file handles.
*/
if (ip->i_din1 == NULL || DIP(ip, mode) == 0)
goto out;
if (DIP(ip, nlink) <= 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
if (getinoquota(ip) == 0)
(void)ufs_quota_free_inode(ip, NOCRED);
error = UFS_TRUNCATE(ip, (off_t)0, IO_EXT | IO_NORMAL, NOCRED);
DIP_ASSIGN(ip, rdev, 0);
mode = DIP(ip, mode);
DIP_ASSIGN(ip, mode, 0);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
/*
* Setting the mode to zero needs to wait for the inode to be
* written just as does a change to the link count. So, rather
* than creating a new entry point to do the same thing, we
* just use softdep_change_linkcnt(). Also, we can't let
* softdep co-opt us to help on its worklist, as we may end up
* trying to recycle vnodes and getting to this same point a
* couple of times, blowing the kernel stack. However, this
* could be optimized by checking if we are coming from
* vrele(), vput() or vclean() (by checking for VXLOCK) and
* just avoiding the co-opt to happen in the last case.
*/
if (DOINGSOFTDEP(vp))
softdep_change_linkcnt(ip, 1);
UFS_INODE_FREE(ip, ip->i_number, mode);
}
if (ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) {
UFS_UPDATE(ip, 0);
}
out:
VOP_UNLOCK(vp, 0, p);
/*
* If we are done with the inode, reclaim it
* so that it can be reused immediately.
*/
if (ip->i_din1 == NULL || DIP(ip, mode) == 0)
vrecycle(vp, p);
return (error);
}
/*
* Last reference to an inode. If necessary, write or delete it.
*/
int
ufs_inactive(void *v)
{
struct vop_inactive_args /* {
struct vnode *a_vp;
struct bool *a_recycle;
} */ *ap = v;
struct vnode *vp = ap->a_vp;
struct inode *ip = VTOI(vp);
struct mount *transmp;
mode_t mode;
int error = 0;
int logged = 0;
UFS_WAPBL_JUNLOCK_ASSERT(vp->v_mount);
transmp = vp->v_mount;
fstrans_start(transmp, FSTRANS_SHARED);
/*
* Ignore inodes related to stale file handles.
*/
if (ip->i_mode == 0)
goto out;
if (ip->i_ffs_effnlink == 0 && DOINGSOFTDEP(vp))
softdep_releasefile(ip);
if (ip->i_nlink <= 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
error = UFS_WAPBL_BEGIN(vp->v_mount);
if (error)
goto out;
logged = 1;
#ifdef QUOTA
(void)chkiq(ip, -1, NOCRED, 0);
#endif
#ifdef UFS_EXTATTR
ufs_extattr_vnode_inactive(vp, curlwp);
#endif
if (ip->i_size != 0) {
/*
* When journaling, only truncate one indirect block
* at a time
*/
if (vp->v_mount->mnt_wapbl) {
uint64_t incr = MNINDIR(ip->i_ump) <<
vp->v_mount->mnt_fs_bshift; /* Power of 2 */
uint64_t base = NDADDR <<
vp->v_mount->mnt_fs_bshift;
while (!error && ip->i_size > base + incr) {
/*
* round down to next full indirect
* block boundary.
*/
uint64_t nsize = base +
((ip->i_size - base - 1) &
~(incr - 1));
error = UFS_TRUNCATE(vp, nsize, 0,
NOCRED);
if (error)
break;
UFS_WAPBL_END(vp->v_mount);
error = UFS_WAPBL_BEGIN(vp->v_mount);
if (error)
goto out;
}
}
if (!error)
error = UFS_TRUNCATE(vp, (off_t)0, 0, NOCRED);
}
/*
* Setting the mode to zero needs to wait for the inode
* to be written just as does a change to the link count.
* So, rather than creating a new entry point to do the
* same thing, we just use softdep_change_linkcnt().
*/
DIP_ASSIGN(ip, rdev, 0);
mode = ip->i_mode;
ip->i_mode = 0;
DIP_ASSIGN(ip, mode, 0);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
mutex_enter(&vp->v_interlock);
vp->v_iflag |= VI_FREEING;
mutex_exit(&vp->v_interlock);
if (DOINGSOFTDEP(vp))
softdep_change_linkcnt(ip);
UFS_VFREE(vp, ip->i_number, mode);
}
if (ip->i_flag & (IN_CHANGE | IN_UPDATE | IN_MODIFIED)) {
if (!logged++) {
int err;
err = UFS_WAPBL_BEGIN(vp->v_mount);
if (err)
goto out;
}
UFS_UPDATE(vp, NULL, NULL, 0);
}
if (logged)
UFS_WAPBL_END(vp->v_mount);
out:
/*
* If we are done with the inode, reclaim it
* so that it can be reused immediately.
*/
*ap->a_recycle = (ip->i_mode == 0);
VOP_UNLOCK(vp, 0);
fstrans_done(transmp);
return (error);
}
/*
* Balloc defines the structure of file system storage
* by allocating the physical blocks on a device given
* the inode and the logical block number in a file.
* This is the allocation strategy for UFS2. Above is
* the allocation strategy for UFS1.
*/
int
ffs_balloc_ufs2(vnode *vp, off_t startoffset, int size,
Ucred *cred, int flags, Buf **bpp)
{
int error = 0;
print("HARVEY TODO: %s\n", __func__);
#if 0
struct inode *ip;
struct ufs2_dinode *dp;
ufs_lbn_t lbn, lastlbn;
struct fs *fs;
struct buf *bp, *nbp;
struct ufsmount *ump;
struct indir indirs[UFS_NIADDR + 2];
ufs2_daddr_t nb, newb, *bap, pref;
ufs2_daddr_t *allocib, *blkp, *allocblk, allociblk[UFS_NIADDR + 1];
ufs2_daddr_t *lbns_remfree, lbns[UFS_NIADDR + 1];
int deallocated, osize, nsize, num, i, error;
int unwindidx = -1;
int saved_inbdflush;
static struct timeval lastfail;
static int curfail;
int gbflags, reclaimed;
ip = VTOI(vp);
dp = ip->i_din2;
fs = ITOFS(ip);
ump = ITOUMP(ip);
lbn = lblkno(fs, startoffset);
size = blkoff(fs, startoffset) + size;
reclaimed = 0;
if (size > fs->fs_bsize)
panic("ffs_balloc_ufs2: blk too big");
*bpp = nil;
if (lbn < 0)
return (EFBIG);
gbflags = (flags & BA_UNMAPPED) != 0 ? GB_UNMAPPED : 0;
if (DOINGSOFTDEP(vp))
softdep_prealloc(vp, MNT_WAIT);
/*
* Check for allocating external data.
*/
if (flags & IO_EXT) {
if (lbn >= UFS_NXADDR)
return (EFBIG);
/*
* If the next write will extend the data into a new block,
* and the data is currently composed of a fragment
* this fragment has to be extended to be a full block.
*/
lastlbn = lblkno(fs, dp->di_extsize);
if (lastlbn < lbn) {
nb = lastlbn;
osize = sblksize(fs, dp->di_extsize, nb);
if (osize < fs->fs_bsize && osize > 0) {
UFS_LOCK(ump);
error = ffs_realloccg(ip, -1 - nb,
dp->di_extb[nb],
ffs_blkpref_ufs2(ip, lastlbn, (int)nb,
&dp->di_extb[0]), osize,
(int)fs->fs_bsize, flags, cred, &bp);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocext(ip, nb,
dbtofsb(fs, bp->b_blkno),
dp->di_extb[nb],
fs->fs_bsize, osize, bp);
dp->di_extsize = smalllblktosize(fs, nb + 1);
dp->di_extb[nb] = dbtofsb(fs, bp->b_blkno);
bp->b_xflags |= BX_ALTDATA;
ip->i_flag |= IN_CHANGE;
if (flags & IO_SYNC)
bwrite(bp);
else
bawrite(bp);
}
}
/*
* All blocks are direct blocks
*/
if (flags & BA_METAONLY)
panic("ffs_balloc_ufs2: BA_METAONLY for ext block");
nb = dp->di_extb[lbn];
if (nb != 0 && dp->di_extsize >= smalllblktosize(fs, lbn + 1)) {
error = bread_gb(vp, -1 - lbn, fs->fs_bsize, NOCRED,
gbflags, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_blkno = fsbtodb(fs, nb);
bp->b_xflags |= BX_ALTDATA;
*bpp = bp;
return (0);
}
if (nb != 0) {
/*
* Consider need to reallocate a fragment.
*/
osize = fragroundup(fs, blkoff(fs, dp->di_extsize));
nsize = fragroundup(fs, size);
if (nsize <= osize) {
error = bread_gb(vp, -1 - lbn, osize, NOCRED,
gbflags, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_blkno = fsbtodb(fs, nb);
bp->b_xflags |= BX_ALTDATA;
} else {
UFS_LOCK(ump);
error = ffs_realloccg(ip, -1 - lbn,
dp->di_extb[lbn],
ffs_blkpref_ufs2(ip, lbn, (int)lbn,
&dp->di_extb[0]), osize, nsize, flags,
cred, &bp);
if (error)
return (error);
bp->b_xflags |= BX_ALTDATA;
if (DOINGSOFTDEP(vp))
softdep_setup_allocext(ip, lbn,
dbtofsb(fs, bp->b_blkno), nb,
nsize, osize, bp);
}
} else {
if (dp->di_extsize < smalllblktosize(fs, lbn + 1))
nsize = fragroundup(fs, size);
else
nsize = fs->fs_bsize;
UFS_LOCK(ump);
error = ffs_alloc(ip, lbn,
ffs_blkpref_ufs2(ip, lbn, (int)lbn, &dp->di_extb[0]),
nsize, flags, cred, &newb);
if (error)
return (error);
bp = getblk(vp, -1 - lbn, nsize, 0, 0, gbflags);
bp->b_blkno = fsbtodb(fs, newb);
bp->b_xflags |= BX_ALTDATA;
if (flags & BA_CLRBUF)
vfs_bio_clrbuf(bp);
if (DOINGSOFTDEP(vp))
softdep_setup_allocext(ip, lbn, newb, 0,
nsize, 0, bp);
}
dp->di_extb[lbn] = dbtofsb(fs, bp->b_blkno);
ip->i_flag |= IN_CHANGE;
*bpp = bp;
return (0);
}
/*
* If the next write will extend the file into a new block,
* and the file is currently composed of a fragment
* this fragment has to be extended to be a full block.
*/
lastlbn = lblkno(fs, ip->i_size);
if (lastlbn < UFS_NDADDR && lastlbn < lbn) {
nb = lastlbn;
osize = blksize(fs, ip, nb);
if (osize < fs->fs_bsize && osize > 0) {
UFS_LOCK(ump);
error = ffs_realloccg(ip, nb, dp->di_db[nb],
ffs_blkpref_ufs2(ip, lastlbn, (int)nb,
&dp->di_db[0]), osize, (int)fs->fs_bsize,
flags, cred, &bp);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, nb,
dbtofsb(fs, bp->b_blkno),
dp->di_db[nb],
fs->fs_bsize, osize, bp);
ip->i_size = smalllblktosize(fs, nb + 1);
dp->di_size = ip->i_size;
dp->di_db[nb] = dbtofsb(fs, bp->b_blkno);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (flags & IO_SYNC)
bwrite(bp);
else
bawrite(bp);
}
}
/*
* The first UFS_NDADDR blocks are direct blocks
*/
if (lbn < UFS_NDADDR) {
if (flags & BA_METAONLY)
panic("ffs_balloc_ufs2: BA_METAONLY for direct block");
nb = dp->di_db[lbn];
if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) {
error = bread_gb(vp, lbn, fs->fs_bsize, NOCRED,
gbflags, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_blkno = fsbtodb(fs, nb);
*bpp = bp;
return (0);
}
if (nb != 0) {
/*
* Consider need to reallocate a fragment.
*/
osize = fragroundup(fs, blkoff(fs, ip->i_size));
nsize = fragroundup(fs, size);
if (nsize <= osize) {
error = bread_gb(vp, lbn, osize, NOCRED,
gbflags, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_blkno = fsbtodb(fs, nb);
} else {
UFS_LOCK(ump);
error = ffs_realloccg(ip, lbn, dp->di_db[lbn],
ffs_blkpref_ufs2(ip, lbn, (int)lbn,
&dp->di_db[0]), osize, nsize, flags,
cred, &bp);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, lbn,
dbtofsb(fs, bp->b_blkno), nb,
nsize, osize, bp);
}
} else {
if (ip->i_size < smalllblktosize(fs, lbn + 1))
nsize = fragroundup(fs, size);
else
nsize = fs->fs_bsize;
UFS_LOCK(ump);
error = ffs_alloc(ip, lbn,
ffs_blkpref_ufs2(ip, lbn, (int)lbn,
&dp->di_db[0]), nsize, flags, cred, &newb);
if (error)
return (error);
bp = getblk(vp, lbn, nsize, 0, 0, gbflags);
bp->b_blkno = fsbtodb(fs, newb);
if (flags & BA_CLRBUF)
vfs_bio_clrbuf(bp);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, lbn, newb, 0,
nsize, 0, bp);
}
dp->di_db[lbn] = dbtofsb(fs, bp->b_blkno);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
*bpp = bp;
return (0);
}
/*
* Determine the number of levels of indirection.
*/
pref = 0;
if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0)
return(error);
#ifdef INVARIANTS
if (num < 1)
panic ("ffs_balloc_ufs2: ufs_getlbns returned indirect block");
#endif
saved_inbdflush = curthread_pflags_set(TDP_INBDFLUSH);
/*
* Fetch the first indirect block allocating if necessary.
*/
--num;
nb = dp->di_ib[indirs[0].in_off];
allocib = nil;
allocblk = allociblk;
lbns_remfree = lbns;
if (nb == 0) {
UFS_LOCK(ump);
pref = ffs_blkpref_ufs2(ip, lbn, -indirs[0].in_off - 1,
(ufs2_daddr_t *)0);
if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
flags, cred, &newb)) != 0) {
curthread_pflags_restore(saved_inbdflush);
return (error);
}
pref = newb + fs->fs_frag;
nb = newb;
MPASS(allocblk < allociblk + nitems(allociblk));
MPASS(lbns_remfree < lbns + nitems(lbns));
*allocblk++ = nb;
*lbns_remfree++ = indirs[1].in_lbn;
bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0,
GB_UNMAPPED);
bp->b_blkno = fsbtodb(fs, nb);
vfs_bio_clrbuf(bp);
if (DOINGSOFTDEP(vp)) {
softdep_setup_allocdirect(ip,
UFS_NDADDR + indirs[0].in_off, newb, 0,
fs->fs_bsize, 0, bp);
bdwrite(bp);
} else if ((flags & IO_SYNC) == 0 && DOINGASYNC(vp)) {
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
} else {
if ((error = bwrite(bp)) != 0)
goto fail;
}
allocib = &dp->di_ib[indirs[0].in_off];
*allocib = nb;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
}
/*
* Fetch through the indirect blocks, allocating as necessary.
*/
retry:
for (i = 1;;) {
error = bread(vp,
indirs[i].in_lbn, (int)fs->fs_bsize, NOCRED, &bp);
if (error) {
brelse(bp);
goto fail;
}
bap = (ufs2_daddr_t *)bp->b_data;
nb = bap[indirs[i].in_off];
if (i == num)
break;
i += 1;
if (nb != 0) {
bqrelse(bp);
continue;
}
UFS_LOCK(ump);
/*
* If parent indirect has just been allocated, try to cluster
* immediately following it.
*/
if (pref == 0)
pref = ffs_blkpref_ufs2(ip, lbn, i - num - 1,
(ufs2_daddr_t *)0);
if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
flags | IO_BUFLOCKED, cred, &newb)) != 0) {
brelse(bp);
if (DOINGSOFTDEP(vp) && ++reclaimed == 1) {
UFS_LOCK(ump);
softdep_request_cleanup(fs, vp, cred,
FLUSH_BLOCKS_WAIT);
UFS_UNLOCK(ump);
goto retry;
}
if (ppsratecheck(&lastfail, &curfail, 1)) {
ffs_fserr(fs, ip->i_number, "filesystem full");
uprintf("\n%s: write failed, filesystem "
"is full\n", fs->fs_fsmnt);
}
goto fail;
}
pref = newb + fs->fs_frag;
nb = newb;
MPASS(allocblk < allociblk + nitems(allociblk));
MPASS(lbns_remfree < lbns + nitems(lbns));
*allocblk++ = nb;
*lbns_remfree++ = indirs[i].in_lbn;
nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0,
GB_UNMAPPED);
nbp->b_blkno = fsbtodb(fs, nb);
vfs_bio_clrbuf(nbp);
if (DOINGSOFTDEP(vp)) {
softdep_setup_allocindir_meta(nbp, ip, bp,
indirs[i - 1].in_off, nb);
bdwrite(nbp);
} else if ((flags & IO_SYNC) == 0 && DOINGASYNC(vp)) {
if (nbp->b_bufsize == fs->fs_bsize)
nbp->b_flags |= B_CLUSTEROK;
bdwrite(nbp);
} else {
if ((error = bwrite(nbp)) != 0) {
brelse(bp);
goto fail;
}
}
bap[indirs[i - 1].in_off] = nb;
if (allocib == nil && unwindidx < 0)
unwindidx = i - 1;
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & IO_SYNC) {
bwrite(bp);
} else {
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
}
}
/*
* If asked only for the indirect block, then return it.
*/
if (flags & BA_METAONLY) {
curthread_pflags_restore(saved_inbdflush);
*bpp = bp;
return (0);
}
/*
* Get the data block, allocating if necessary.
*/
if (nb == 0) {
UFS_LOCK(ump);
/*
* If allocating metadata at the front of the cylinder
* group and parent indirect block has just been allocated,
* then cluster next to it if it is the first indirect in
* the file. Otherwise it has been allocated in the metadata
* area, so we want to find our own place out in the data area.
*/
if (pref == 0 || (lbn > UFS_NDADDR && fs->fs_metaspace != 0))
pref = ffs_blkpref_ufs2(ip, lbn, indirs[i].in_off,
&bap[0]);
error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
flags | IO_BUFLOCKED, cred, &newb);
if (error) {
brelse(bp);
if (DOINGSOFTDEP(vp) && ++reclaimed == 1) {
UFS_LOCK(ump);
softdep_request_cleanup(fs, vp, cred,
FLUSH_BLOCKS_WAIT);
UFS_UNLOCK(ump);
goto retry;
}
if (ppsratecheck(&lastfail, &curfail, 1)) {
ffs_fserr(fs, ip->i_number, "filesystem full");
uprintf("\n%s: write failed, filesystem "
"is full\n", fs->fs_fsmnt);
}
goto fail;
}
nb = newb;
MPASS(allocblk < allociblk + nitems(allociblk));
MPASS(lbns_remfree < lbns + nitems(lbns));
*allocblk++ = nb;
*lbns_remfree++ = lbn;
nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0, gbflags);
nbp->b_blkno = fsbtodb(fs, nb);
if (flags & BA_CLRBUF)
vfs_bio_clrbuf(nbp);
if (DOINGSOFTDEP(vp))
softdep_setup_allocindir_page(ip, lbn, bp,
indirs[i].in_off, nb, 0, nbp);
bap[indirs[i].in_off] = nb;
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & IO_SYNC) {
bwrite(bp);
} else {
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
}
curthread_pflags_restore(saved_inbdflush);
*bpp = nbp;
return (0);
}
brelse(bp);
/*
* If requested clear invalid portions of the buffer. If we
* have to do a read-before-write (typical if BA_CLRBUF is set),
* try to do some read-ahead in the sequential case to reduce
* the number of I/O transactions.
*/
if (flags & BA_CLRBUF) {
int seqcount = (flags & BA_SEQMASK) >> BA_SEQSHIFT;
if (seqcount != 0 &&
(vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0 &&
!(vm_page_count_severe() || buf_dirty_count_severe())) {
error = cluster_read(vp, ip->i_size, lbn,
(int)fs->fs_bsize, NOCRED,
MAXBSIZE, seqcount, gbflags, &nbp);
} else {
error = bread_gb(vp, lbn, (int)fs->fs_bsize,
NOCRED, gbflags, &nbp);
}
if (error) {
brelse(nbp);
goto fail;
}
} else {
/*
* Last reference to an inode. If necessary, write or delete it.
*/
int
ufs_inactive(void *v)
{
struct vop_inactive_args *ap = v;
struct vnode *vp = ap->a_vp;
struct inode *ip = VTOI(vp);
struct fs *fs = ip->i_fs;
struct proc *p = curproc;
mode_t mode;
int error = 0, logged = 0, truncate_error = 0;
#ifdef DIAGNOSTIC
extern int prtactive;
if (prtactive && vp->v_usecount != 0)
vprint("ufs_inactive: pushing active", vp);
#endif
UFS_WAPBL_JUNLOCK_ASSERT(vp->v_mount);
/*
* Ignore inodes related to stale file handles.
*/
if (ip->i_din1 == NULL || DIP(ip, mode) == 0)
goto out;
if (DIP(ip, nlink) <= 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
error = UFS_WAPBL_BEGIN(vp->v_mount);
if (error)
goto out;
logged = 1;
if (getinoquota(ip) == 0)
(void)ufs_quota_free_inode(ip, NOCRED);
if (DIP(ip, size) != 0 && vp->v_mount->mnt_wapbl) {
/*
* When journaling, only truncate one indirect block at
* a time.
*/
uint64_t incr = MNINDIR(ip->i_ump) << fs->fs_bshift;
uint64_t base = NDADDR << fs->fs_bshift;
while (!error && DIP(ip, size) > base + incr) {
/*
* round down to next full indirect block
* boundary.
*/
uint64_t nsize = base +
((DIP(ip, size) - base - 1) &
~(incr - 1));
error = UFS_TRUNCATE(ip, nsize, 0, NOCRED);
if (error)
break;
UFS_WAPBL_END(vp->v_mount);
error = UFS_WAPBL_BEGIN(vp->v_mount);
if (error)
goto out;
}
}
if (error == 0) {
truncate_error = UFS_TRUNCATE(ip, (off_t)0, 0, NOCRED);
/* XXX pedro: remove me */
if (truncate_error)
printf("UFS_TRUNCATE()=%d\n", truncate_error);
}
DIP_ASSIGN(ip, rdev, 0);
mode = DIP(ip, mode);
DIP_ASSIGN(ip, mode, 0);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
/*
* Setting the mode to zero needs to wait for the inode to be
* written just as does a change to the link count. So, rather
* than creating a new entry point to do the same thing, we
* just use softdep_change_linkcnt(). Also, we can't let
* softdep co-opt us to help on its worklist, as we may end up
* trying to recycle vnodes and getting to this same point a
* couple of times, blowing the kernel stack. However, this
* could be optimized by checking if we are coming from
* vrele(), vput() or vclean() (by checking for VXLOCK) and
* just avoiding the co-opt to happen in the last case.
*/
if (DOINGSOFTDEP(vp))
softdep_change_linkcnt(ip, 1);
UFS_INODE_FREE(ip, ip->i_number, mode);
}
if (ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) {
if (!logged++) {
int err;
err = UFS_WAPBL_BEGIN(vp->v_mount);
if (err) {
error = err;
goto out;
}
}
UFS_UPDATE(ip, 0);
}
if (logged)
UFS_WAPBL_END(vp->v_mount);
out:
VOP_UNLOCK(vp, 0);
/*
* If we are done with the inode, reclaim it
* so that it can be reused immediately.
*/
if (error == 0 && truncate_error == 0 &&
(ip->i_din1 == NULL || DIP(ip, mode) == 0))
vrecycle(vp, p);
return (truncate_error ? truncate_error : 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);
}
/*
* Update the access, modified, and inode change times as specified by the
* IN_ACCESS, IN_UPDATE, and IN_CHANGE flags respectively. The IN_MODIFIED
* flag is used to specify that the inode needs to be updated but that the
* times have already been set. The access and modified times are taken from
* the second and third parameters; the inode change time is always taken
* from the current time. If waitfor is set, then wait for the disk write
* of the inode to complete.
*/
int
ffs_update(struct inode *ip, struct timespec *atime,
struct timespec *mtime, int waitfor)
{
struct vnode *vp;
struct fs *fs;
struct buf *bp;
int error;
struct timespec ts;
vp = ITOV(ip);
if (vp->v_mount->mnt_flag & MNT_RDONLY) {
ip->i_flag &=
~(IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE);
return (0);
}
if ((ip->i_flag &
(IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
waitfor != MNT_WAIT)
return (0);
getnanotime(&ts);
if (ip->i_flag & IN_ACCESS) {
DIP_ASSIGN(ip, atime, atime ? atime->tv_sec : ts.tv_sec);
DIP_ASSIGN(ip, atimensec, atime ? atime->tv_nsec : ts.tv_nsec);
}
if (ip->i_flag & IN_UPDATE) {
DIP_ASSIGN(ip, mtime, mtime ? mtime->tv_sec : ts.tv_sec);
DIP_ASSIGN(ip, mtimensec, mtime ? mtime->tv_nsec : ts.tv_nsec);
ip->i_modrev++;
}
if (ip->i_flag & IN_CHANGE) {
DIP_ASSIGN(ip, ctime, ts.tv_sec);
DIP_ASSIGN(ip, ctimensec, ts.tv_nsec);
}
ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE);
fs = ip->i_fs;
/*
* Ensure that uid and gid are correct. This is a temporary
* fix until fsck has been changed to do the update.
*/
if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_inodefmt < FS_44INODEFMT) {
ip->i_din1->di_ouid = ip->i_ffs1_uid;
ip->i_din1->di_ogid = ip->i_ffs1_gid;
}
error = bread(ip->i_devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
(int)fs->fs_bsize, &bp);
if (error) {
brelse(bp);
return (error);
}
if (DOINGSOFTDEP(vp))
softdep_update_inodeblock(ip, bp, waitfor);
else if (ip->i_effnlink != DIP(ip, nlink))
panic("ffs_update: bad link cnt");
#ifdef FFS2
if (ip->i_ump->um_fstype == UM_UFS2)
*((struct ufs2_dinode *)bp->b_data +
ino_to_fsbo(fs, ip->i_number)) = *ip->i_din2;
else
#endif
*((struct ufs1_dinode *)bp->b_data +
ino_to_fsbo(fs, ip->i_number)) = *ip->i_din1;
if (waitfor && !DOINGASYNC(vp)) {
return (bwrite(bp));
} else {
bdwrite(bp);
return (0);
}
}
/*
* Allocate a new inode.
*/
int
ufs_makeinode(int mode, struct vnode *dvp, struct vnode **vpp,
struct componentname *cnp)
{
struct inode *ip, *pdir;
struct direct newdir;
struct vnode *tvp;
int error;
pdir = VTOI(dvp);
#ifdef DIAGNOSTIC
if ((cnp->cn_flags & HASBUF) == 0)
panic("ufs_makeinode: no name");
#endif
*vpp = NULL;
if ((mode & IFMT) == 0)
mode |= IFREG;
if ((error = UFS_INODE_ALLOC(pdir, mode, cnp->cn_cred, &tvp)) != 0) {
pool_put(&namei_pool, cnp->cn_pnbuf);
vput(dvp);
return (error);
}
ip = VTOI(tvp);
DIP_ASSIGN(ip, gid, DIP(pdir, gid));
DIP_ASSIGN(ip, uid, cnp->cn_cred->cr_uid);
if ((error = getinoquota(ip)) ||
(error = ufs_quota_alloc_inode(ip, cnp->cn_cred))) {
pool_put(&namei_pool, cnp->cn_pnbuf);
UFS_INODE_FREE(ip, ip->i_number, mode);
vput(tvp);
vput(dvp);
return (error);
}
ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE;
DIP_ASSIGN(ip, mode, mode);
tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */
ip->i_effnlink = 1;
DIP_ASSIGN(ip, nlink, 1);
if (DOINGSOFTDEP(tvp))
softdep_change_linkcnt(ip, 0);
if ((DIP(ip, mode) & ISGID) &&
!groupmember(DIP(ip, gid), cnp->cn_cred) &&
suser_ucred(cnp->cn_cred))
DIP_AND(ip, mode, ~ISGID);
/*
* Make sure inode goes to disk before directory entry.
*/
if ((error = UFS_UPDATE(ip, !DOINGSOFTDEP(tvp))) != 0)
goto bad;
ufs_makedirentry(ip, cnp, &newdir);
if ((error = ufs_direnter(dvp, tvp, &newdir, cnp, NULL)) != 0)
goto bad;
if ((cnp->cn_flags & SAVESTART) == 0)
pool_put(&namei_pool, cnp->cn_pnbuf);
vput(dvp);
*vpp = tvp;
return (0);
bad:
/*
* Write error occurred trying to update the inode
* or the directory so must deallocate the inode.
*/
pool_put(&namei_pool, cnp->cn_pnbuf);
vput(dvp);
ip->i_effnlink = 0;
DIP_ASSIGN(ip, nlink, 0);
ip->i_flag |= IN_CHANGE;
if (DOINGSOFTDEP(tvp))
softdep_change_linkcnt(ip, 0);
tvp->v_type = VNON;
vput(tvp);
return (error);
}
/*
* Truncate the inode oip to at most length size, freeing the
* disk blocks.
*/
int
ffs_truncate(struct vnode *vp, off_t length, int flags, struct ucred *cred)
{
struct vnode *ovp = vp;
ufs_daddr_t lastblock;
struct inode *oip;
ufs_daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
ufs_daddr_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
struct fs *fs;
struct buf *bp;
int offset, size, level;
long count, nblocks, blocksreleased = 0;
int i;
int aflags, error, allerror;
off_t osize;
oip = VTOI(ovp);
fs = oip->i_fs;
if (length < 0)
return (EINVAL);
if (length > fs->fs_maxfilesize)
return (EFBIG);
if (ovp->v_type == VLNK &&
(oip->i_size < ovp->v_mount->mnt_maxsymlinklen || oip->i_din.di_blocks == 0)) {
#ifdef DIAGNOSTIC
if (length != 0)
panic("ffs_truncate: partial truncate of symlink");
#endif /* DIAGNOSTIC */
bzero((char *)&oip->i_shortlink, (uint)oip->i_size);
oip->i_size = 0;
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ffs_update(ovp, 1));
}
if (oip->i_size == length) {
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ffs_update(ovp, 0));
}
if (fs->fs_ronly)
panic("ffs_truncate: read-only filesystem");
#ifdef QUOTA
error = ufs_getinoquota(oip);
if (error)
return (error);
#endif
ovp->v_lasta = ovp->v_clen = ovp->v_cstart = ovp->v_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, MNT_WAIT, 0)) != 0)
return (error);
} else {
#ifdef QUOTA
(void) ufs_chkdq(oip, -oip->i_blocks, NOCRED, 0);
#endif
softdep_setup_freeblocks(oip, length);
vinvalbuf(ovp, 0, 0, 0);
nvnode_pager_setsize(ovp, 0, fs->fs_bsize, 0);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ffs_update(ovp, 0));
}
}
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.
*
* nvextendbuf() only breads the old buffer. The blocksize
* of the new buffer must be specified so it knows how large
* to make the VM object.
*/
if (osize < length) {
nvextendbuf(vp, osize, length,
blkoffsize(fs, oip, osize), /* oblksize */
blkoffresize(fs, length), /* nblksize */
blkoff(fs, osize),
blkoff(fs, length),
0);
aflags = B_CLRBUF;
if (flags & IO_SYNC)
aflags |= B_SYNC;
/* BALLOC will reallocate the fragment at the old EOF */
error = VOP_BALLOC(ovp, length - 1, 1, cred, aflags, &bp);
if (error)
return (error);
oip->i_size = length;
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
if (aflags & B_SYNC)
bwrite(bp);
else
bawrite(bp);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ffs_update(ovp, 1));
}
/*
* Shorten the size of the file.
*
* NOTE: The block size specified in nvtruncbuf() is the blocksize
* of the buffer containing length prior to any reallocation
* of the block.
*/
allerror = nvtruncbuf(ovp, length, blkoffsize(fs, oip, length),
blkoff(fs, length), 0);
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 = VOP_BALLOC(ovp, 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.
*
* nvtruncbuf() may have re-dirtied the underlying block
* as part of its truncation zeroing code. To avoid a
* 'locking against myself' panic in the second fsync we
* can simply undirty the bp since the redirtying was
* related to areas of the buffer that we are going to
* throw away anyway, and we will b*write() the remainder
* anyway down below.
*/
if (DOINGSOFTDEP(ovp) && lbn < NDADDR &&
fragroundup(fs, blkoff(fs, length)) < fs->fs_bsize) {
bundirty(bp);
error = VOP_FSYNC(ovp, MNT_WAIT, 0);
if (error) {
bdwrite(bp);
return (error);
}
}
oip->i_size = length;
size = blksize(fs, oip, lbn);
#if 0
/* remove - nvtruncbuf deals with this */
if (ovp->v_type != VDIR)
bzero((char *)bp->b_data + offset,
(uint)(size - offset));
#endif
/* Kirk's code has reallocbuf(bp, size, 1) here */
allocbuf(bp, size);
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
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.
*/
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;
error = ffs_update(ovp, 1);
if (error && allerror == 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.
*/
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;
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 = ffs_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;
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 = oip->i_db[i];
if (bn == 0)
continue;
oip->i_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 = 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("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] != oip->i_ib[level])
panic("ffs_truncate1");
for (i = 0; i < NDADDR; i++)
if (newblks[i] != oip->i_db[i])
panic("ffs_truncate2");
if (length == 0 && !RB_EMPTY(&ovp->v_rbdirty_tree))
panic("ffs_truncate3");
#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;
#ifdef QUOTA
(void) ufs_chkdq(oip, -blocksreleased, NOCRED, 0);
#endif
return (allerror);
}
/*
* link vnode call
*/
int
ufs_link(void *v)
{
struct vop_link_args *ap = v;
struct vnode *dvp = ap->a_dvp;
struct vnode *vp = ap->a_vp;
struct componentname *cnp = ap->a_cnp;
struct proc *p = cnp->cn_proc;
struct inode *ip;
struct direct newdir;
int error;
#ifdef DIAGNOSTIC
if ((cnp->cn_flags & HASBUF) == 0)
panic("ufs_link: no name");
#endif
if (vp->v_type == VDIR) {
VOP_ABORTOP(dvp, cnp);
error = EPERM;
goto out2;
}
if (dvp->v_mount != vp->v_mount) {
VOP_ABORTOP(dvp, cnp);
error = EXDEV;
goto out2;
}
if (dvp != vp && (error = vn_lock(vp, LK_EXCLUSIVE, p))) {
VOP_ABORTOP(dvp, cnp);
goto out2;
}
ip = VTOI(vp);
if ((nlink_t) DIP(ip, nlink) >= LINK_MAX) {
VOP_ABORTOP(dvp, cnp);
error = EMLINK;
goto out1;
}
if (DIP(ip, flags) & (IMMUTABLE | APPEND)) {
VOP_ABORTOP(dvp, cnp);
error = EPERM;
goto out1;
}
ip->i_effnlink++;
DIP_ADD(ip, nlink, 1);
ip->i_flag |= IN_CHANGE;
if (DOINGSOFTDEP(vp))
softdep_change_linkcnt(ip, 0);
if ((error = UFS_UPDATE(ip, !DOINGSOFTDEP(vp))) == 0) {
ufs_makedirentry(ip, cnp, &newdir);
error = ufs_direnter(dvp, vp, &newdir, cnp, NULL);
}
if (error) {
ip->i_effnlink--;
DIP_ADD(ip, nlink, -1);
ip->i_flag |= IN_CHANGE;
if (DOINGSOFTDEP(vp))
softdep_change_linkcnt(ip, 0);
}
pool_put(&namei_pool, cnp->cn_pnbuf);
VN_KNOTE(vp, NOTE_LINK);
VN_KNOTE(dvp, NOTE_WRITE);
out1:
if (dvp != vp)
VOP_UNLOCK(vp, 0);
out2:
vput(dvp);
return (error);
}
/*
* Rename system call.
* rename("foo", "bar");
* is essentially
* unlink("bar");
* link("foo", "bar");
* unlink("foo");
* but ``atomically''. Can't do full commit without saving state in the
* inode on disk which isn't feasible at this time. Best we can do is
* always guarantee the target exists.
*
* Basic algorithm is:
*
* 1) Bump link count on source while we're linking it to the
* target. This also ensure the inode won't be deleted out
* from underneath us while we work (it may be truncated by
* a concurrent `trunc' or `open' for creation).
* 2) Link source to destination. If destination already exists,
* delete it first.
* 3) Unlink source reference to inode if still around. If a
* directory was moved and the parent of the destination
* is different from the source, patch the ".." entry in the
* directory.
*/
int
ufs_rename(void *v)
{
struct vop_rename_args *ap = v;
struct vnode *tvp = ap->a_tvp;
struct vnode *tdvp = ap->a_tdvp;
struct vnode *fvp = ap->a_fvp;
struct vnode *fdvp = ap->a_fdvp;
struct componentname *tcnp = ap->a_tcnp;
struct componentname *fcnp = ap->a_fcnp;
struct proc *p = fcnp->cn_proc;
struct inode *ip, *xp, *dp;
struct direct newdir;
int doingdirectory = 0, oldparent = 0, newparent = 0;
int error = 0;
#ifdef DIAGNOSTIC
if ((tcnp->cn_flags & HASBUF) == 0 ||
(fcnp->cn_flags & HASBUF) == 0)
panic("ufs_rename: no name");
#endif
/*
* Check for cross-device rename.
*/
if ((fvp->v_mount != tdvp->v_mount) ||
(tvp && (fvp->v_mount != tvp->v_mount))) {
error = EXDEV;
abortit:
VOP_ABORTOP(tdvp, tcnp);
if (tdvp == tvp)
vrele(tdvp);
else
vput(tdvp);
if (tvp)
vput(tvp);
VOP_ABORTOP(fdvp, fcnp);
vrele(fdvp);
vrele(fvp);
return (error);
}
if (tvp && ((DIP(VTOI(tvp), flags) & (IMMUTABLE | APPEND)) ||
(DIP(VTOI(tdvp), flags) & APPEND))) {
error = EPERM;
goto abortit;
}
/*
* Check if just deleting a link name or if we've lost a race.
* If another process completes the same rename after we've looked
* up the source and have blocked looking up the target, then the
* source and target inodes may be identical now although the
* names were never linked.
*/
if (fvp == tvp) {
if (fvp->v_type == VDIR) {
/*
* Linked directories are impossible, so we must
* have lost the race. Pretend that the rename
* completed before the lookup.
*/
error = ENOENT;
goto abortit;
}
/* Release destination completely. */
VOP_ABORTOP(tdvp, tcnp);
vput(tdvp);
vput(tvp);
/*
* Delete source. There is another race now that everything
* is unlocked, but this doesn't cause any new complications.
* relookup() may find a file that is unrelated to the
* original one, or it may fail. Too bad.
*/
vrele(fvp);
fcnp->cn_flags &= ~MODMASK;
fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
if ((fcnp->cn_flags & SAVESTART) == 0)
panic("ufs_rename: lost from startdir");
fcnp->cn_nameiop = DELETE;
if ((error = vfs_relookup(fdvp, &fvp, fcnp)) != 0)
return (error); /* relookup did vrele() */
vrele(fdvp);
return (VOP_REMOVE(fdvp, fvp, fcnp));
}
if ((error = vn_lock(fvp, LK_EXCLUSIVE, p)) != 0)
goto abortit;
/* fvp, tdvp, tvp now locked */
dp = VTOI(fdvp);
ip = VTOI(fvp);
if ((nlink_t) DIP(ip, nlink) >= LINK_MAX) {
VOP_UNLOCK(fvp, 0);
error = EMLINK;
goto abortit;
}
if ((DIP(ip, flags) & (IMMUTABLE | APPEND)) ||
(DIP(dp, flags) & APPEND)) {
VOP_UNLOCK(fvp, 0);
error = EPERM;
goto abortit;
}
if ((DIP(ip, mode) & IFMT) == IFDIR) {
error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred);
if (!error && tvp)
error = VOP_ACCESS(tvp, VWRITE, tcnp->cn_cred);
if (error) {
VOP_UNLOCK(fvp, 0);
error = EACCES;
goto abortit;
}
/*
* Avoid ".", "..", and aliases of "." for obvious reasons.
*/
if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
dp == ip ||
(fcnp->cn_flags & ISDOTDOT) ||
(tcnp->cn_flags & ISDOTDOT) ||
(ip->i_flag & IN_RENAME)) {
VOP_UNLOCK(fvp, 0);
error = EINVAL;
goto abortit;
}
ip->i_flag |= IN_RENAME;
oldparent = dp->i_number;
doingdirectory = 1;
}
VN_KNOTE(fdvp, NOTE_WRITE); /* XXX right place? */
/*
* When the target exists, both the directory
* and target vnodes are returned locked.
*/
dp = VTOI(tdvp);
xp = NULL;
if (tvp)
xp = VTOI(tvp);
/*
* 1) Bump link count while we're moving stuff
* around. If we crash somewhere before
* completing our work, the link count
* may be wrong, but correctable.
*/
ip->i_effnlink++;
DIP_ADD(ip, nlink, 1);
ip->i_flag |= IN_CHANGE;
if (DOINGSOFTDEP(fvp))
softdep_change_linkcnt(ip, 0);
if ((error = UFS_UPDATE(ip, !DOINGSOFTDEP(fvp))) != 0) {
VOP_UNLOCK(fvp, 0);
goto bad;
}
/*
* If ".." must be changed (ie the directory gets a new
* parent) then the source directory must not be in the
* directory hierarchy above the target, as this would
* orphan everything below the source directory. Also
* the user must have write permission in the source so
* as to be able to change "..". We must repeat the call
* to namei, as the parent directory is unlocked by the
* call to checkpath().
*/
error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred);
VOP_UNLOCK(fvp, 0);
/* tdvp and tvp locked */
if (oldparent != dp->i_number)
newparent = dp->i_number;
if (doingdirectory && newparent) {
if (error) /* write access check above */
goto bad;
if (xp != NULL)
vput(tvp);
/*
* Compensate for the reference ufs_checkpath() loses.
*/
vref(tdvp);
/* Only tdvp is locked */
if ((error = ufs_checkpath(ip, dp, tcnp->cn_cred)) != 0) {
vrele(tdvp);
goto out;
}
if ((tcnp->cn_flags & SAVESTART) == 0)
panic("ufs_rename: lost to startdir");
if ((error = vfs_relookup(tdvp, &tvp, tcnp)) != 0)
goto out;
vrele(tdvp); /* relookup() acquired a reference */
dp = VTOI(tdvp);
xp = NULL;
if (tvp)
xp = VTOI(tvp);
}
/*
* 2) If target doesn't exist, link the target
* to the source and unlink the source.
* Otherwise, rewrite the target directory
* entry to reference the source inode and
* expunge the original entry's existence.
*/
if (xp == NULL) {
if (dp->i_dev != ip->i_dev)
panic("rename: EXDEV");
/*
* Account for ".." in new directory.
* When source and destination have the same
* parent we don't fool with the link count.
*/
if (doingdirectory && newparent) {
if ((nlink_t) DIP(dp, nlink) >= LINK_MAX) {
error = EMLINK;
goto bad;
}
dp->i_effnlink++;
DIP_ADD(dp, nlink, 1);
dp->i_flag |= IN_CHANGE;
if (DOINGSOFTDEP(tdvp))
softdep_change_linkcnt(dp, 0);
if ((error = UFS_UPDATE(dp, !DOINGSOFTDEP(tdvp)))
!= 0) {
dp->i_effnlink--;
DIP_ADD(dp, nlink, -1);
dp->i_flag |= IN_CHANGE;
if (DOINGSOFTDEP(tdvp))
softdep_change_linkcnt(dp, 0);
goto bad;
}
}
ufs_makedirentry(ip, tcnp, &newdir);
if ((error = ufs_direnter(tdvp, NULL, &newdir, tcnp, NULL)) != 0) {
if (doingdirectory && newparent) {
dp->i_effnlink--;
DIP_ADD(dp, nlink, -1);
dp->i_flag |= IN_CHANGE;
if (DOINGSOFTDEP(tdvp))
softdep_change_linkcnt(dp, 0);
(void)UFS_UPDATE(dp, 1);
}
goto bad;
}
VN_KNOTE(tdvp, NOTE_WRITE);
vput(tdvp);
} else {
if (xp->i_dev != dp->i_dev || xp->i_dev != ip->i_dev)
panic("rename: EXDEV");
/*
* Short circuit rename(foo, foo).
*/
if (xp->i_number == ip->i_number)
panic("ufs_rename: same file");
/*
* If the parent directory is "sticky", then the user must
* own the parent directory, or the destination of the rename,
* otherwise the destination may not be changed (except by
* root). This implements append-only directories.
*/
if ((DIP(dp, mode) & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 &&
tcnp->cn_cred->cr_uid != DIP(dp, uid) &&
DIP(xp, uid )!= tcnp->cn_cred->cr_uid) {
error = EPERM;
goto bad;
}
/*
* Target must be empty if a directory and have no links
* to it. Also, ensure source and target are compatible
* (both directories, or both not directories).
*/
if ((DIP(xp, mode) & IFMT) == IFDIR) {
if (xp->i_effnlink > 2 ||
!ufs_dirempty(xp, dp->i_number, tcnp->cn_cred)) {
error = ENOTEMPTY;
goto bad;
}
if (!doingdirectory) {
error = ENOTDIR;
goto bad;
}
cache_purge(tdvp);
} else if (doingdirectory) {
error = EISDIR;
goto bad;
}
if ((error = ufs_dirrewrite(dp, xp, ip->i_number,
IFTODT(DIP(ip, mode)), (doingdirectory && newparent) ?
newparent : doingdirectory)) != 0)
goto bad;
if (doingdirectory) {
if (!newparent) {
dp->i_effnlink--;
if (DOINGSOFTDEP(tdvp))
softdep_change_linkcnt(dp, 0);
}
xp->i_effnlink--;
if (DOINGSOFTDEP(tvp))
softdep_change_linkcnt(xp, 0);
}
if (doingdirectory && !DOINGSOFTDEP(tvp)) {
/*
* Truncate inode. The only stuff left in the directory
* is "." and "..". The "." reference is inconsequential
* since we are quashing it. We have removed the "."
* reference and the reference in the parent directory,
* but there may be other hard links. The soft
* dependency code will arrange to do these operations
* after the parent directory entry has been deleted on
* disk, so when running with that code we avoid doing
* them now.
*/
if (!newparent) {
DIP_ADD(dp, nlink, -1);
dp->i_flag |= IN_CHANGE;
}
DIP_ADD(xp, nlink, -1);
xp->i_flag |= IN_CHANGE;
if ((error = UFS_TRUNCATE(VTOI(tvp), (off_t)0, IO_SYNC,
tcnp->cn_cred)) != 0)
goto bad;
}
VN_KNOTE(tdvp, NOTE_WRITE);
vput(tdvp);
VN_KNOTE(tvp, NOTE_DELETE);
vput(tvp);
xp = NULL;
}
/*
* 3) Unlink the source.
*/
fcnp->cn_flags &= ~MODMASK;
fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
if ((fcnp->cn_flags & SAVESTART) == 0)
panic("ufs_rename: lost from startdir");
if ((error = vfs_relookup(fdvp, &fvp, fcnp)) != 0) {
vrele(ap->a_fvp);
return (error);
}
vrele(fdvp);
if (fvp == NULL) {
/*
* From name has disappeared.
*/
if (doingdirectory)
panic("ufs_rename: lost dir entry");
vrele(ap->a_fvp);
return (0);
}
xp = VTOI(fvp);
dp = VTOI(fdvp);
/*
* Ensure that the directory entry still exists and has not
* changed while the new name has been entered. If the source is
* a file then the entry may have been unlinked or renamed. In
* either case there is no further work to be done. If the source
* is a directory then it cannot have been rmdir'ed; the IN_RENAME
* flag ensures that it cannot be moved by another rename or removed
* by a rmdir.
*/
if (xp != ip) {
if (doingdirectory)
panic("ufs_rename: lost dir entry");
} else {
/*
* If the source is a directory with a
* new parent, the link count of the old
* parent directory must be decremented
* and ".." set to point to the new parent.
*/
if (doingdirectory && newparent) {
xp->i_offset = mastertemplate.dot_reclen;
ufs_dirrewrite(xp, dp, newparent, DT_DIR, 0);
cache_purge(fdvp);
}
error = ufs_dirremove(fdvp, xp, fcnp->cn_flags, 0);
xp->i_flag &= ~IN_RENAME;
}
VN_KNOTE(fvp, NOTE_RENAME);
if (dp)
vput(fdvp);
if (xp)
vput(fvp);
vrele(ap->a_fvp);
return (error);
bad:
if (xp)
vput(ITOV(xp));
vput(ITOV(dp));
out:
vrele(fdvp);
if (doingdirectory)
ip->i_flag &= ~IN_RENAME;
if (vn_lock(fvp, LK_EXCLUSIVE, p) == 0) {
ip->i_effnlink--;
DIP_ADD(ip, nlink, -1);
ip->i_flag |= IN_CHANGE;
ip->i_flag &= ~IN_RENAME;
if (DOINGSOFTDEP(fvp))
softdep_change_linkcnt(ip, 0);
vput(fvp);
} else
vrele(fvp);
return (error);
}
/*
* Truncate the inode ip to at most length size, freeing the
* disk blocks.
*/
int
ffs_truncate(vnode *vp, off_t length, int flags, Ucred *cred)
{
print("HARVEY TODO: %s\n", __func__);
#if 0
struct inode *ip;
ufs2_daddr_t bn, lbn, lastblock, lastiblock[UFS_NIADDR];
ufs2_daddr_t indir_lbn[UFS_NIADDR], oldblks[UFS_NDADDR + UFS_NIADDR];
ufs2_daddr_t newblks[UFS_NDADDR + UFS_NIADDR];
ufs2_daddr_t count, blocksreleased = 0, datablocks, blkno;
struct bufobj *bo;
struct fs *fs;
struct buf *bp;
struct ufsmount *ump;
int softdeptrunc, journaltrunc;
int needextclean, extblocks;
int offset, size, level, nblocks;
int i, error, allerror, indiroff, waitforupdate;
off_t osize;
ip = VTOI(vp);
ump = VFSTOUFS(vp->v_mount);
fs = ump->um_fs;
bo = &vp->v_bufobj;
ASSERT_VOP_LOCKED(vp, "ffs_truncate");
if (length < 0)
return (EINVAL);
if (length > fs->fs_maxfilesize)
return (EFBIG);
#ifdef QUOTA
error = getinoquota(ip);
if (error)
return (error);
#endif
/*
* Historically clients did not have to specify which data
* they were truncating. So, if not specified, we assume
* traditional behavior, e.g., just the normal data.
*/
if ((flags & (IO_EXT | IO_NORMAL)) == 0)
flags |= IO_NORMAL;
if (!DOINGSOFTDEP(vp) && !DOINGASYNC(vp))
flags |= IO_SYNC;
waitforupdate = (flags & IO_SYNC) != 0 || !DOINGASYNC(vp);
/*
* If we are truncating the extended-attributes, and cannot
* do it with soft updates, then do it slowly here. If we are
* truncating both the extended attributes and the file contents
* (e.g., the file is being unlinked), then pick it off with
* soft updates below.
*/
allerror = 0;
needextclean = 0;
softdeptrunc = 0;
journaltrunc = DOINGSUJ(vp);
if (journaltrunc == 0 && DOINGSOFTDEP(vp) && length == 0)
softdeptrunc = !softdep_slowdown(vp);
extblocks = 0;
datablocks = DIP(ip, i_blocks);
if (fs->fs_magic == FS_UFS2_MAGIC && ip->i_din2->di_extsize > 0) {
extblocks = btodb(fragroundup(fs, ip->i_din2->di_extsize));
datablocks -= extblocks;
}
if ((flags & IO_EXT) && extblocks > 0) {
if (length != 0)
panic("ffs_truncate: partial trunc of extdata");
if (softdeptrunc || journaltrunc) {
if ((flags & IO_NORMAL) == 0)
goto extclean;
needextclean = 1;
} else {
if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
return (error);
#ifdef QUOTA
(void) chkdq(ip, -extblocks, NOCRED, 0);
#endif
vinvalbuf(vp, V_ALT, 0, 0);
vn_pages_remove(vp,
OFF_TO_IDX(lblktosize(fs, -extblocks)), 0);
osize = ip->i_din2->di_extsize;
ip->i_din2->di_blocks -= extblocks;
ip->i_din2->di_extsize = 0;
for (i = 0; i < UFS_NXADDR; i++) {
oldblks[i] = ip->i_din2->di_extb[i];
ip->i_din2->di_extb[i] = 0;
}
ip->i_flag |= IN_CHANGE;
if ((error = ffs_update(vp, waitforupdate)))
return (error);
for (i = 0; i < UFS_NXADDR; i++) {
if (oldblks[i] == 0)
continue;
ffs_blkfree(ump, fs, ITODEVVP(ip), oldblks[i],
sblksize(fs, osize, i), ip->i_number,
vp->v_type, nil);
}
}
}
if ((flags & IO_NORMAL) == 0)
return (0);
if (vp->v_type == VLNK &&
(ip->i_size < vp->v_mount->mnt_maxsymlinklen ||
datablocks == 0)) {
#ifdef INVARIANTS
if (length != 0)
panic("ffs_truncate: partial truncate of symlink");
#endif
bzero(SHORTLINK(ip), (uint)ip->i_size);
ip->i_size = 0;
DIP_SET(ip, i_size, 0);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (needextclean)
goto extclean;
return (ffs_update(vp, waitforupdate));
}
if (ip->i_size == length) {
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (needextclean)
goto extclean;
return (ffs_update(vp, 0));
}
if (fs->fs_ronly)
panic("ffs_truncate: read-only filesystem");
if (IS_SNAPSHOT(ip))
ffs_snapremove(vp);
vp->v_lasta = vp->v_clen = vp->v_cstart = vp->v_lastw = 0;
osize = ip->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) {
vnode_pager_setsize(vp, length);
flags |= BA_CLRBUF;
error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &bp);
if (error) {
vnode_pager_setsize(vp, osize);
return (error);
}
ip->i_size = length;
DIP_SET(ip, i_size, length);
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
if (flags & IO_SYNC)
bwrite(bp);
else if (DOINGASYNC(vp))
bdwrite(bp);
else
bawrite(bp);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ffs_update(vp, waitforupdate));
}
/*
* Lookup block number for a given offset. Zero length files
* have no blocks, so return a blkno of -1.
*/
lbn = lblkno(fs, length - 1);
if (length == 0) {
blkno = -1;
} else if (lbn < UFS_NDADDR) {
blkno = DIP(ip, i_db[lbn]);
} else {
error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn), fs->fs_bsize,
cred, BA_METAONLY, &bp);
if (error)
return (error);
indiroff = (lbn - UFS_NDADDR) % NINDIR(fs);
if (I_IS_UFS1(ip))
blkno = ((ufs1_daddr_t *)(bp->b_data))[indiroff];
else
blkno = ((ufs2_daddr_t *)(bp->b_data))[indiroff];
/*
* If the block number is non-zero, then the indirect block
* must have been previously allocated and need not be written.
* If the block number is zero, then we may have allocated
* the indirect block and hence need to write it out.
*/
if (blkno != 0)
brelse(bp);
else if (flags & IO_SYNC)
bwrite(bp);
else
bdwrite(bp);
}
/*
* If the block number at the new end of the file is zero,
* then we must allocate it to ensure that the last block of
* the file is allocated. Soft updates does not handle this
* case, so here we have to clean up the soft updates data
* structures describing the allocation past the truncation
* point. Finding and deallocating those structures is a lot of
* work. Since partial truncation with a hole at the end occurs
* rarely, we solve the problem by syncing the file so that it
* will have no soft updates data structures left.
*/
if (blkno == 0 && (error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
return (error);
if (blkno != 0 && DOINGSOFTDEP(vp)) {
if (softdeptrunc == 0 && journaltrunc == 0) {
/*
* If soft updates cannot handle this truncation,
* clean up soft dependency data structures and
* fall through to the synchronous truncation.
*/
if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
return (error);
} else {
flags = IO_NORMAL | (needextclean ? IO_EXT: 0);
if (journaltrunc)
softdep_journal_freeblocks(ip, cred, length,
flags);
else
softdep_setup_freeblocks(ip, length, flags);
ASSERT_VOP_LOCKED(vp, "ffs_truncate1");
if (journaltrunc == 0) {
ip->i_flag |= IN_CHANGE | IN_UPDATE;
error = ffs_update(vp, 0);
}
return (error);
}
}
/*
* Shorten the size of the file. If the last block of the
* shortened file is unallocated, we must allocate it.
* Additionally, 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 (blkno != 0 && offset == 0) {
ip->i_size = length;
DIP_SET(ip, i_size, length);
} else {
lbn = lblkno(fs, length);
flags |= BA_CLRBUF;
error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &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(vp) && lbn < UFS_NDADDR &&
fragroundup(fs, blkoff(fs, length)) < fs->fs_bsize &&
(error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
return (error);
ip->i_size = length;
DIP_SET(ip, i_size, length);
size = blksize(fs, ip, lbn);
if (vp->v_type != VDIR && offset != 0)
bzero((char *)bp->b_data + offset,
(uint)(size - offset));
/* Kirk's code has reallocbuf(bp, size, 1) here */
allocbuf(bp, size);
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
if (flags & IO_SYNC)
bwrite(bp);
else if (DOINGASYNC(vp))
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->fs_bsize - 1) - 1;
lastiblock[SINGLE] = lastblock - UFS_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[UFS_NDADDR + level] = DIP(ip, i_ib[level]);
if (lastiblock[level] < 0) {
DIP_SET(ip, i_ib[level], 0);
lastiblock[level] = -1;
}
}
for (i = 0; i < UFS_NDADDR; i++) {
oldblks[i] = DIP(ip, i_db[i]);
if (i > lastblock)
DIP_SET(ip, i_db[i], 0);
}
ip->i_flag |= IN_CHANGE | IN_UPDATE;
allerror = ffs_update(vp, waitforupdate);
/*
* 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 < UFS_NDADDR; i++) {
newblks[i] = DIP(ip, i_db[i]);
DIP_SET(ip, i_db[i], oldblks[i]);
}
for (i = 0; i < UFS_NIADDR; i++) {
newblks[UFS_NDADDR + i] = DIP(ip, i_ib[i]);
DIP_SET(ip, i_ib[i], oldblks[UFS_NDADDR + i]);
}
ip->i_size = osize;
DIP_SET(ip, i_size, osize);
error = vtruncbuf(vp, cred, length, fs->fs_bsize);
if (error && (allerror == 0))
allerror = error;
/*
* Indirect blocks first.
*/
indir_lbn[SINGLE] = -UFS_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(ip, i_ib[level]);
if (bn != 0) {
error = ffs_indirtrunc(ip, indir_lbn[level],
fsbtodb(fs, bn), lastiblock[level], level, &count);
if (error)
allerror = error;
blocksreleased += count;
if (lastiblock[level] < 0) {
DIP_SET(ip, i_ib[level], 0);
ffs_blkfree(ump, fs, ump->um_devvp, bn,
fs->fs_bsize, ip->i_number,
vp->v_type, nil);
blocksreleased += nblocks;
}
}
if (lastiblock[level] >= 0)
goto done;
}
/*
* All whole direct blocks or frags.
*/
for (i = UFS_NDADDR - 1; i > lastblock; i--) {
long bsize;
bn = DIP(ip, i_db[i]);
if (bn == 0)
continue;
DIP_SET(ip, i_db[i], 0);
bsize = blksize(fs, ip, i);
ffs_blkfree(ump, fs, ump->um_devvp, bn, bsize, ip->i_number,
vp->v_type, nil);
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(ip, 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, ip, lastblock);
ip->i_size = length;
DIP_SET(ip, i_size, length);
newspace = blksize(fs, ip, 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(ump, fs, ump->um_devvp, bn,
oldspace - newspace, ip->i_number, vp->v_type, nil);
blocksreleased += btodb(oldspace - newspace);
}
}
done:
#ifdef INVARIANTS
for (level = SINGLE; level <= TRIPLE; level++)
if (newblks[UFS_NDADDR + level] != DIP(ip, i_ib[level]))
panic("ffs_truncate1");
for (i = 0; i < UFS_NDADDR; i++)
if (newblks[i] != DIP(ip, i_db[i]))
panic("ffs_truncate2");
BO_LOCK(bo);
if (length == 0 &&
(fs->fs_magic != FS_UFS2_MAGIC || ip->i_din2->di_extsize == 0) &&
(bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
panic("ffs_truncate3");
BO_UNLOCK(bo);
#endif /* INVARIANTS */
/*
* Put back the real size.
*/
ip->i_size = length;
DIP_SET(ip, i_size, length);
if (DIP(ip, i_blocks) >= blocksreleased)
DIP_SET(ip, i_blocks, DIP(ip, i_blocks) - blocksreleased);
else /* sanity */
DIP_SET(ip, i_blocks, 0);
ip->i_flag |= IN_CHANGE;
#ifdef QUOTA
(void) chkdq(ip, -blocksreleased, NOCRED, 0);
#endif
return (allerror);
extclean:
if (journaltrunc)
softdep_journal_freeblocks(ip, cred, length, IO_EXT);
else
softdep_setup_freeblocks(ip, length, IO_EXT);
return (ffs_update(vp, waitforupdate));
#endif // 0
return 0;
}
/*
* Remove a directory entry after a call to namei, using
* the parameters which it left in nameidata. The entry
* dp->i_offset contains the offset into the directory of the
* entry to be eliminated. The dp->i_count field contains the
* size of the previous record in the directory. If this
* is 0, the first entry is being deleted, so we need only
* zero the inode number to mark the entry as free. If the
* entry is not the first in the directory, we must reclaim
* the space of the now empty record by adding the record size
* to the size of the previous entry.
*/
int
ufs_dirremove(struct vnode *dvp, struct inode *ip, int flags, int isrmdir)
{
struct inode *dp;
struct direct *ep;
struct buf *bp;
int error;
UFS_WAPBL_JLOCK_ASSERT(dvp->v_mount);
dp = VTOI(dvp);
if ((error = UFS_BUFATOFF(dp,
(off_t)(dp->i_offset - dp->i_count), (char **)&ep, &bp)) != 0)
return (error);
#ifdef UFS_DIRHASH
/*
* Remove the dirhash entry. This is complicated by the fact
* that `ep' is the previous entry when dp->i_count != 0.
*/
if (dp->i_dirhash != NULL)
ufsdirhash_remove(dp, (dp->i_count == 0) ? ep :
(struct direct *)((char *)ep + ep->d_reclen), dp->i_offset);
#endif
if (dp->i_count == 0) {
/*
* First entry in block: set d_ino to zero.
*/
ep->d_ino = 0;
} else {
/*
* Collapse new free space into previous entry.
*/
ep->d_reclen += dp->i_reclen;
}
#ifdef UFS_DIRHASH
if (dp->i_dirhash != NULL)
ufsdirhash_checkblock(dp, (char *)ep -
((dp->i_offset - dp->i_count) & (DIRBLKSIZ - 1)),
dp->i_offset & ~(DIRBLKSIZ - 1));
#endif
if (DOINGSOFTDEP(dvp)) {
if (ip) {
ip->i_effnlink--;
softdep_change_linkcnt(ip, 0);
softdep_setup_remove(bp, dp, ip, isrmdir);
}
if (softdep_slowdown(dvp)) {
error = bwrite(bp);
} else {
bdwrite(bp);
error = 0;
}
} else {
if (ip) {
ip->i_effnlink--;
DIP_ADD(ip, nlink, -1);
ip->i_flag |= IN_CHANGE;
UFS_WAPBL_UPDATE(ip, 0);
}
if (DOINGASYNC(dvp) && dp->i_count != 0) {
bdwrite(bp);
error = 0;
} else
error = bwrite(bp);
}
dp->i_flag |= IN_CHANGE | IN_UPDATE;
UFS_WAPBL_UPDATE(dp, 0);
return (error);
}
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);
}