int
cluster_read_gb(struct vnode *vp, u_quad_t filesize, daddr_t lblkno, long size,
struct ucred *cred, long totread, int seqcount, int gbflags,
struct buf **bpp)
{
struct buf *bp, *rbp, *reqbp;
struct bufobj *bo;
daddr_t blkno, origblkno;
int maxra, racluster;
int error, ncontig;
int i;
error = 0;
bo = &vp->v_bufobj;
if (!unmapped_buf_allowed)
gbflags &= ~GB_UNMAPPED;
/*
* Try to limit the amount of read-ahead by a few
* ad-hoc parameters. This needs work!!!
*/
racluster = vp->v_mount->mnt_iosize_max / size;
maxra = seqcount;
maxra = min(read_max, maxra);
maxra = min(nbuf/8, maxra);
if (((u_quad_t)(lblkno + maxra + 1) * size) > filesize)
maxra = (filesize / size) - lblkno;
/*
* get the requested block
*/
*bpp = reqbp = bp = getblk(vp, lblkno, size, 0, 0, gbflags);
origblkno = lblkno;
/*
* if it is in the cache, then check to see if the reads have been
* sequential. If they have, then try some read-ahead, otherwise
* back-off on prospective read-aheads.
*/
if (bp->b_flags & B_CACHE) {
if (!seqcount) {
return 0;
} else if ((bp->b_flags & B_RAM) == 0) {
return 0;
} else {
bp->b_flags &= ~B_RAM;
BO_LOCK(bo);
for (i = 1; i < maxra; i++) {
/*
* Stop if the buffer does not exist or it
* is invalid (about to go away?)
*/
rbp = gbincore(&vp->v_bufobj, lblkno+i);
if (rbp == NULL || (rbp->b_flags & B_INVAL))
break;
/*
* Set another read-ahead mark so we know
* to check again. (If we can lock the
* buffer without waiting)
*/
if ((((i % racluster) == (racluster - 1)) ||
(i == (maxra - 1)))
&& (0 == BUF_LOCK(rbp,
LK_EXCLUSIVE | LK_NOWAIT, NULL))) {
rbp->b_flags |= B_RAM;
BUF_UNLOCK(rbp);
}
}
BO_UNLOCK(bo);
if (i >= maxra) {
return 0;
}
lblkno += i;
}
reqbp = bp = NULL;
/*
* If it isn't in the cache, then get a chunk from
* disk if sequential, otherwise just get the block.
*/
} else {
off_t firstread = bp->b_offset;
int nblks;
long minread;
KASSERT(bp->b_offset != NOOFFSET,
("cluster_read: no buffer offset"));
ncontig = 0;
/*
* Adjust totread if needed
*/
minread = read_min * size;
if (minread > totread)
totread = minread;
/*
* Compute the total number of blocks that we should read
* synchronously.
*/
if (firstread + totread > filesize)
totread = filesize - firstread;
nblks = howmany(totread, size);
if (nblks > racluster)
nblks = racluster;
/*
* Now compute the number of contiguous blocks.
*/
if (nblks > 1) {
error = VOP_BMAP(vp, lblkno, NULL,
&blkno, &ncontig, NULL);
/*
* If this failed to map just do the original block.
*/
if (error || blkno == -1)
ncontig = 0;
}
/*
* If we have contiguous data available do a cluster
* otherwise just read the requested block.
*/
if (ncontig) {
/* Account for our first block. */
ncontig = min(ncontig + 1, nblks);
if (ncontig < nblks)
nblks = ncontig;
bp = cluster_rbuild(vp, filesize, lblkno,
blkno, size, nblks, gbflags, bp);
lblkno += (bp->b_bufsize / size);
} else {
bp->b_flags |= B_RAM;
bp->b_iocmd = BIO_READ;
lblkno += 1;
}
}
/*
* handle the synchronous read so that it is available ASAP.
*/
if (bp) {
if ((bp->b_flags & B_CLUSTER) == 0) {
vfs_busy_pages(bp, 0);
}
bp->b_flags &= ~B_INVAL;
bp->b_ioflags &= ~BIO_ERROR;
if ((bp->b_flags & B_ASYNC) || bp->b_iodone != NULL)
BUF_KERNPROC(bp);
bp->b_iooffset = dbtob(bp->b_blkno);
bstrategy(bp);
curthread->td_ru.ru_inblock++;
}
/*
* If we have been doing sequential I/O, then do some read-ahead.
*/
while (lblkno < (origblkno + maxra)) {
error = VOP_BMAP(vp, lblkno, NULL, &blkno, &ncontig, NULL);
if (error)
break;
if (blkno == -1)
break;
/*
* We could throttle ncontig here by maxra but we might as
* well read the data if it is contiguous. We're throttled
* by racluster anyway.
*/
if (ncontig) {
ncontig = min(ncontig + 1, racluster);
rbp = cluster_rbuild(vp, filesize, lblkno, blkno,
size, ncontig, gbflags, NULL);
lblkno += (rbp->b_bufsize / size);
if (rbp->b_flags & B_DELWRI) {
bqrelse(rbp);
continue;
}
} else {
rbp = getblk(vp, lblkno, size, 0, 0, gbflags);
lblkno += 1;
if (rbp->b_flags & B_DELWRI) {
bqrelse(rbp);
continue;
}
rbp->b_flags |= B_ASYNC | B_RAM;
rbp->b_iocmd = BIO_READ;
rbp->b_blkno = blkno;
}
if (rbp->b_flags & B_CACHE) {
rbp->b_flags &= ~B_ASYNC;
bqrelse(rbp);
continue;
}
if ((rbp->b_flags & B_CLUSTER) == 0) {
vfs_busy_pages(rbp, 0);
}
rbp->b_flags &= ~B_INVAL;
rbp->b_ioflags &= ~BIO_ERROR;
if ((rbp->b_flags & B_ASYNC) || rbp->b_iodone != NULL)
BUF_KERNPROC(rbp);
rbp->b_iooffset = dbtob(rbp->b_blkno);
bstrategy(rbp);
curthread->td_ru.ru_inblock++;
}
if (reqbp)
return (bufwait(reqbp));
else
return (error);
}
/*
* If blocks are contiguous on disk, use this to provide clustered
* read ahead. We will read as many blocks as possible sequentially
* and then parcel them up into logical blocks in the buffer hash table.
*/
static struct buf *
cluster_rbuild(struct vnode *vp, u_quad_t filesize, daddr_t lbn,
daddr_t blkno, long size, int run, int gbflags, struct buf *fbp)
{
struct bufobj *bo;
struct buf *bp, *tbp;
daddr_t bn;
off_t off;
long tinc, tsize;
int i, inc, j, toff;
KASSERT(size == vp->v_mount->mnt_stat.f_iosize,
("cluster_rbuild: size %ld != filesize %jd\n",
size, (intmax_t)vp->v_mount->mnt_stat.f_iosize));
/*
* avoid a division
*/
while ((u_quad_t) size * (lbn + run) > filesize) {
--run;
}
if (fbp) {
tbp = fbp;
tbp->b_iocmd = BIO_READ;
} else {
tbp = getblk(vp, lbn, size, 0, 0, gbflags);
if (tbp->b_flags & B_CACHE)
return tbp;
tbp->b_flags |= B_ASYNC | B_RAM;
tbp->b_iocmd = BIO_READ;
}
tbp->b_blkno = blkno;
if( (tbp->b_flags & B_MALLOC) ||
((tbp->b_flags & B_VMIO) == 0) || (run <= 1) )
return tbp;
bp = trypbuf(&cluster_pbuf_freecnt);
if (bp == 0)
return tbp;
/*
* We are synthesizing a buffer out of vm_page_t's, but
* if the block size is not page aligned then the starting
* address may not be either. Inherit the b_data offset
* from the original buffer.
*/
bp->b_flags = B_ASYNC | B_CLUSTER | B_VMIO;
if ((gbflags & GB_UNMAPPED) != 0) {
bp->b_flags |= B_UNMAPPED;
bp->b_data = unmapped_buf;
} else {
bp->b_data = (char *)((vm_offset_t)bp->b_data |
((vm_offset_t)tbp->b_data & PAGE_MASK));
}
bp->b_iocmd = BIO_READ;
bp->b_iodone = cluster_callback;
bp->b_blkno = blkno;
bp->b_lblkno = lbn;
bp->b_offset = tbp->b_offset;
KASSERT(bp->b_offset != NOOFFSET, ("cluster_rbuild: no buffer offset"));
pbgetvp(vp, bp);
TAILQ_INIT(&bp->b_cluster.cluster_head);
bp->b_bcount = 0;
bp->b_bufsize = 0;
bp->b_npages = 0;
inc = btodb(size);
bo = &vp->v_bufobj;
for (bn = blkno, i = 0; i < run; ++i, bn += inc) {
if (i != 0) {
if ((bp->b_npages * PAGE_SIZE) +
round_page(size) > vp->v_mount->mnt_iosize_max) {
break;
}
tbp = getblk(vp, lbn + i, size, 0, 0, GB_LOCK_NOWAIT |
(gbflags & GB_UNMAPPED));
/* Don't wait around for locked bufs. */
if (tbp == NULL)
break;
/*
* Stop scanning if the buffer is fully valid
* (marked B_CACHE), or locked (may be doing a
* background write), or if the buffer is not
* VMIO backed. The clustering code can only deal
* with VMIO-backed buffers.
*/
BO_LOCK(bo);
if ((tbp->b_vflags & BV_BKGRDINPROG) ||
(tbp->b_flags & B_CACHE) ||
(tbp->b_flags & B_VMIO) == 0) {
BO_UNLOCK(bo);
bqrelse(tbp);
break;
}
BO_UNLOCK(bo);
/*
* The buffer must be completely invalid in order to
* take part in the cluster. If it is partially valid
* then we stop.
*/
off = tbp->b_offset;
tsize = size;
VM_OBJECT_LOCK(tbp->b_bufobj->bo_object);
for (j = 0; tsize > 0; j++) {
toff = off & PAGE_MASK;
tinc = tsize;
if (toff + tinc > PAGE_SIZE)
tinc = PAGE_SIZE - toff;
VM_OBJECT_LOCK_ASSERT(tbp->b_pages[j]->object,
MA_OWNED);
if ((tbp->b_pages[j]->valid &
vm_page_bits(toff, tinc)) != 0)
break;
off += tinc;
tsize -= tinc;
}
VM_OBJECT_UNLOCK(tbp->b_bufobj->bo_object);
if (tsize > 0) {
bqrelse(tbp);
break;
}
/*
* Set a read-ahead mark as appropriate
*/
if ((fbp && (i == 1)) || (i == (run - 1)))
tbp->b_flags |= B_RAM;
/*
* Set the buffer up for an async read (XXX should
* we do this only if we do not wind up brelse()ing?).
* Set the block number if it isn't set, otherwise
* if it is make sure it matches the block number we
* expect.
*/
tbp->b_flags |= B_ASYNC;
tbp->b_iocmd = BIO_READ;
if (tbp->b_blkno == tbp->b_lblkno) {
tbp->b_blkno = bn;
} else if (tbp->b_blkno != bn) {
brelse(tbp);
break;
}
}
/*
* XXX fbp from caller may not be B_ASYNC, but we are going
* to biodone() it in cluster_callback() anyway
*/
BUF_KERNPROC(tbp);
TAILQ_INSERT_TAIL(&bp->b_cluster.cluster_head,
tbp, b_cluster.cluster_entry);
VM_OBJECT_LOCK(tbp->b_bufobj->bo_object);
for (j = 0; j < tbp->b_npages; j += 1) {
vm_page_t m;
m = tbp->b_pages[j];
vm_page_io_start(m);
vm_object_pip_add(m->object, 1);
if ((bp->b_npages == 0) ||
(bp->b_pages[bp->b_npages-1] != m)) {
bp->b_pages[bp->b_npages] = m;
bp->b_npages++;
}
if (m->valid == VM_PAGE_BITS_ALL)
tbp->b_pages[j] = bogus_page;
}
VM_OBJECT_UNLOCK(tbp->b_bufobj->bo_object);
/*
* Don't inherit tbp->b_bufsize as it may be larger due to
* a non-page-aligned size. Instead just aggregate using
* 'size'.
*/
if (tbp->b_bcount != size)
printf("warning: tbp->b_bcount wrong %ld vs %ld\n", tbp->b_bcount, size);
if (tbp->b_bufsize != size)
printf("warning: tbp->b_bufsize wrong %ld vs %ld\n", tbp->b_bufsize, size);
bp->b_bcount += size;
bp->b_bufsize += size;
}
/*
* Fully valid pages in the cluster are already good and do not need
* to be re-read from disk. Replace the page with bogus_page
*/
VM_OBJECT_LOCK(bp->b_bufobj->bo_object);
for (j = 0; j < bp->b_npages; j++) {
VM_OBJECT_LOCK_ASSERT(bp->b_pages[j]->object, MA_OWNED);
if (bp->b_pages[j]->valid == VM_PAGE_BITS_ALL)
bp->b_pages[j] = bogus_page;
}
VM_OBJECT_UNLOCK(bp->b_bufobj->bo_object);
if (bp->b_bufsize > bp->b_kvasize)
panic("cluster_rbuild: b_bufsize(%ld) > b_kvasize(%d)\n",
bp->b_bufsize, bp->b_kvasize);
bp->b_kvasize = bp->b_bufsize;
if ((bp->b_flags & B_UNMAPPED) == 0) {
pmap_qenter(trunc_page((vm_offset_t) bp->b_data),
(vm_page_t *)bp->b_pages, bp->b_npages);
}
return (bp);
}
/*
* Truncate the inode oip to at most length size, freeing the
* disk blocks.
*/
int
ext2_truncate(struct vnode *vp, off_t length, int flags, struct ucred *cred,
struct thread *td)
{
struct vnode *ovp = vp;
int32_t lastblock;
struct inode *oip;
int32_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
uint32_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
struct m_ext2fs *fs;
struct buf *bp;
int offset, size, level;
e4fs_daddr_t count, nblocks, blocksreleased = 0;
int error, i, allerror;
off_t osize;
#ifdef INVARIANTS
struct bufobj *bo;
#endif
oip = VTOI(ovp);
#ifdef INVARIANTS
bo = &ovp->v_bufobj;
#endif
ASSERT_VOP_LOCKED(vp, "ext2_truncate");
if (length < 0)
return (EINVAL);
if (ovp->v_type == VLNK &&
oip->i_size < ovp->v_mount->mnt_maxsymlinklen) {
#ifdef INVARIANTS
if (length != 0)
panic("ext2_truncate: partial truncate of symlink");
#endif
bzero((char *)&oip->i_shortlink, (u_int)oip->i_size);
oip->i_size = 0;
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ext2_update(ovp, 1));
}
if (oip->i_size == length) {
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ext2_update(ovp, 0));
}
fs = oip->i_e2fs;
osize = oip->i_size;
/*
* Lengthen the size of the file. We must ensure that the
* last byte of the file is allocated. Since the smallest
* value of osize is 0, length will be at least 1.
*/
if (osize < length) {
if (length > oip->i_e2fs->e2fs_maxfilesize)
return (EFBIG);
vnode_pager_setsize(ovp, length);
offset = blkoff(fs, length - 1);
lbn = lblkno(fs, length - 1);
flags |= BA_CLRBUF;
error = ext2_balloc(oip, lbn, offset + 1, cred, &bp, flags);
if (error) {
vnode_pager_setsize(vp, osize);
return (error);
}
oip->i_size = length;
if (bp->b_bufsize == fs->e2fs_bsize)
bp->b_flags |= B_CLUSTEROK;
if (flags & IO_SYNC)
bwrite(bp);
else if (DOINGASYNC(ovp))
bdwrite(bp);
else
bawrite(bp);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ext2_update(ovp, !DOINGASYNC(ovp)));
}
/*
* Shorten the size of the file. If the file is not being
* truncated to a block boundry, the contents of the
* partial block following the end of the file must be
* zero'ed in case it ever become accessible again because
* of subsequent file growth.
*/
/* I don't understand the comment above */
offset = blkoff(fs, length);
if (offset == 0) {
oip->i_size = length;
} else {
lbn = lblkno(fs, length);
flags |= BA_CLRBUF;
error = ext2_balloc(oip, lbn, offset, cred, &bp, flags);
if (error)
return (error);
oip->i_size = length;
size = blksize(fs, oip, lbn);
bzero((char *)bp->b_data + offset, (u_int)(size - offset));
allocbuf(bp, size);
if (bp->b_bufsize == fs->e2fs_bsize)
bp->b_flags |= B_CLUSTEROK;
if (flags & IO_SYNC)
bwrite(bp);
else if (DOINGASYNC(ovp))
bdwrite(bp);
else
bawrite(bp);
}
/*
* Calculate index into inode's block list of
* last direct and indirect blocks (if any)
* which we want to keep. Lastblock is -1 when
* the file is truncated to 0.
*/
lastblock = lblkno(fs, length + fs->e2fs_bsize - 1) - 1;
lastiblock[SINGLE] = lastblock - NDADDR;
lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
nblocks = btodb(fs->e2fs_bsize);
/*
* Update file and block pointers on disk before we start freeing
* blocks. If we crash before free'ing blocks below, the blocks
* will be returned to the free list. lastiblock values are also
* normalized to -1 for calls to ext2_indirtrunc below.
*/
for (level = TRIPLE; level >= SINGLE; level--) {
oldblks[NDADDR + level] = oip->i_ib[level];
if (lastiblock[level] < 0) {
oip->i_ib[level] = 0;
lastiblock[level] = -1;
}
}
for (i = 0; i < NDADDR; i++) {
oldblks[i] = oip->i_db[i];
if (i > lastblock)
oip->i_db[i] = 0;
}
oip->i_flag |= IN_CHANGE | IN_UPDATE;
allerror = ext2_update(ovp, !DOINGASYNC(ovp));
/*
* Having written the new inode to disk, save its new configuration
* and put back the old block pointers long enough to process them.
* Note that we save the new block configuration so we can check it
* when we are done.
*/
for (i = 0; i < NDADDR; i++) {
newblks[i] = oip->i_db[i];
oip->i_db[i] = oldblks[i];
}
for (i = 0; i < NIADDR; i++) {
newblks[NDADDR + i] = oip->i_ib[i];
oip->i_ib[i] = oldblks[NDADDR + i];
}
oip->i_size = osize;
error = vtruncbuf(ovp, cred, length, (int)fs->e2fs_bsize);
if (error && (allerror == 0))
allerror = error;
vnode_pager_setsize(ovp, length);
/*
* Indirect blocks first.
*/
indir_lbn[SINGLE] = -NDADDR;
indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
for (level = TRIPLE; level >= SINGLE; level--) {
bn = oip->i_ib[level];
if (bn != 0) {
error = ext2_indirtrunc(oip, indir_lbn[level],
fsbtodb(fs, bn), lastiblock[level], level, &count);
if (error)
allerror = error;
blocksreleased += count;
if (lastiblock[level] < 0) {
oip->i_ib[level] = 0;
ext2_blkfree(oip, bn, fs->e2fs_fsize);
blocksreleased += nblocks;
}
}
if (lastiblock[level] >= 0)
goto done;
}
/*
* All whole direct blocks or frags.
*/
for (i = NDADDR - 1; i > lastblock; i--) {
long bsize;
bn = oip->i_db[i];
if (bn == 0)
continue;
oip->i_db[i] = 0;
bsize = blksize(fs, oip, i);
ext2_blkfree(oip, bn, bsize);
blocksreleased += btodb(bsize);
}
if (lastblock < 0)
goto done;
/*
* Finally, look for a change in size of the
* last direct block; release any frags.
*/
bn = oip->i_db[lastblock];
if (bn != 0) {
long oldspace, newspace;
/*
* Calculate amount of space we're giving
* back as old block size minus new block size.
*/
oldspace = blksize(fs, oip, lastblock);
oip->i_size = length;
newspace = blksize(fs, oip, lastblock);
if (newspace == 0)
panic("ext2_truncate: newspace");
if (oldspace - newspace > 0) {
/*
* Block number of space to be free'd is
* the old block # plus the number of frags
* required for the storage we're keeping.
*/
bn += numfrags(fs, newspace);
ext2_blkfree(oip, bn, oldspace - newspace);
blocksreleased += btodb(oldspace - newspace);
}
}
done:
#ifdef INVARIANTS
for (level = SINGLE; level <= TRIPLE; level++)
if (newblks[NDADDR + level] != oip->i_ib[level])
panic("itrunc1");
for (i = 0; i < NDADDR; i++)
if (newblks[i] != oip->i_db[i])
panic("itrunc2");
BO_LOCK(bo);
if (length == 0 && (bo->bo_dirty.bv_cnt != 0 ||
bo->bo_clean.bv_cnt != 0))
panic("itrunc3");
BO_UNLOCK(bo);
#endif /* INVARIANTS */
/*
* Put back the real size.
*/
oip->i_size = length;
if (oip->i_blocks >= blocksreleased)
oip->i_blocks -= blocksreleased;
else /* sanity */
oip->i_blocks = 0;
oip->i_flag |= IN_CHANGE;
vnode_pager_setsize(ovp, length);
return (allerror);
}
/*
* Unmount the filesystem described by mp.
*/
static int
msdosfs_unmount(struct mount *mp, int mntflags)
{
struct msdosfsmount *pmp;
int error, flags;
error = flags = 0;
pmp = VFSTOMSDOSFS(mp);
if ((pmp->pm_flags & MSDOSFSMNT_RONLY) == 0)
error = msdosfs_sync(mp, MNT_WAIT);
if ((mntflags & MNT_FORCE) != 0)
flags |= FORCECLOSE;
else if (error != 0)
return (error);
error = vflush(mp, 0, flags, curthread);
if (error != 0 && error != ENXIO)
return (error);
if ((pmp->pm_flags & MSDOSFSMNT_RONLY) == 0) {
error = markvoldirty(pmp, 0);
if (error && error != ENXIO) {
(void)markvoldirty(pmp, 1);
return (error);
}
}
if (pmp->pm_flags & MSDOSFSMNT_KICONV && msdosfs_iconv) {
if (pmp->pm_w2u)
msdosfs_iconv->close(pmp->pm_w2u);
if (pmp->pm_u2w)
msdosfs_iconv->close(pmp->pm_u2w);
if (pmp->pm_d2u)
msdosfs_iconv->close(pmp->pm_d2u);
if (pmp->pm_u2d)
msdosfs_iconv->close(pmp->pm_u2d);
}
#ifdef MSDOSFS_DEBUG
{
struct vnode *vp = pmp->pm_devvp;
struct bufobj *bo;
bo = &vp->v_bufobj;
BO_LOCK(bo);
VI_LOCK(vp);
vn_printf(vp,
"msdosfs_umount(): just before calling VOP_CLOSE()\n");
printf("freef %p, freeb %p, mount %p\n",
TAILQ_NEXT(vp, v_actfreelist), vp->v_actfreelist.tqe_prev,
vp->v_mount);
printf("cleanblkhd %p, dirtyblkhd %p, numoutput %ld, type %d\n",
TAILQ_FIRST(&vp->v_bufobj.bo_clean.bv_hd),
TAILQ_FIRST(&vp->v_bufobj.bo_dirty.bv_hd),
vp->v_bufobj.bo_numoutput, vp->v_type);
VI_UNLOCK(vp);
BO_UNLOCK(bo);
}
#endif
DROP_GIANT();
if (pmp->pm_devvp->v_type == VCHR && pmp->pm_devvp->v_rdev != NULL)
pmp->pm_devvp->v_rdev->si_mountpt = NULL;
g_topology_lock();
g_vfs_close(pmp->pm_cp);
g_topology_unlock();
PICKUP_GIANT();
vrele(pmp->pm_devvp);
dev_rel(pmp->pm_dev);
free(pmp->pm_inusemap, M_MSDOSFSFAT);
if (pmp->pm_flags & MSDOSFS_LARGEFS)
msdosfs_fileno_free(mp);
lockdestroy(&pmp->pm_fatlock);
free(pmp, M_MSDOSFSMNT);
mp->mnt_data = NULL;
MNT_ILOCK(mp);
mp->mnt_flag &= ~MNT_LOCAL;
MNT_IUNLOCK(mp);
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;
}
