/*
* EOPNOTSUPP is no longer legal. For local media VFS's that do not
* implement their own VOP_PUTPAGES, their VOP_PUTPAGES should call to
* vnode_pager_generic_putpages() to implement the previous behaviour.
*
* Caller has already cleared the pmap modified bits, if any.
*
* All other FS's should use the bypass to get to the local media
* backing vp's VOP_PUTPAGES.
*/
static void
vnode_pager_putpages(vm_object_t object, vm_page_t *m, int count,
int sync, int *rtvals)
{
int rtval;
struct vnode *vp;
int bytes = count * PAGE_SIZE;
/*
* Force synchronous operation if we are extremely low on memory
* to prevent a low-memory deadlock. VOP operations often need to
* allocate more memory to initiate the I/O ( i.e. do a BMAP
* operation ). The swapper handles the case by limiting the amount
* of asynchronous I/O, but that sort of solution doesn't scale well
* for the vnode pager without a lot of work.
*
* Also, the backing vnode's iodone routine may not wake the pageout
* daemon up. This should be probably be addressed XXX.
*/
if ((vmstats.v_free_count + vmstats.v_cache_count) <
vmstats.v_pageout_free_min) {
sync |= OBJPC_SYNC;
}
/*
* Call device-specific putpages function
*/
vp = object->handle;
rtval = VOP_PUTPAGES(vp, m, bytes, sync, rtvals, 0);
if (rtval == EOPNOTSUPP) {
kprintf("vnode_pager: *** WARNING *** stale FS putpages\n");
rtval = vnode_pager_generic_putpages( vp, m, bytes, sync, rtvals);
}
}
/*
* EOPNOTSUPP is no longer legal. For local media VFS's that do not
* implement their own VOP_PUTPAGES, their VOP_PUTPAGES should call to
* vnode_pager_generic_putpages() to implement the previous behaviour.
*
* All other FS's should use the bypass to get to the local media
* backing vp's VOP_PUTPAGES.
*/
static void
vnode_pager_putpages(vm_object_t object, vm_page_t *m, int count,
int flags, int *rtvals)
{
int rtval;
struct vnode *vp;
int bytes = count * PAGE_SIZE;
/*
* Force synchronous operation if we are extremely low on memory
* to prevent a low-memory deadlock. VOP operations often need to
* allocate more memory to initiate the I/O ( i.e. do a BMAP
* operation ). The swapper handles the case by limiting the amount
* of asynchronous I/O, but that sort of solution doesn't scale well
* for the vnode pager without a lot of work.
*
* Also, the backing vnode's iodone routine may not wake the pageout
* daemon up. This should be probably be addressed XXX.
*/
if (vm_cnt.v_free_count < vm_cnt.v_pageout_free_min)
flags |= VM_PAGER_PUT_SYNC;
/*
* Call device-specific putpages function
*/
vp = object->handle;
VM_OBJECT_WUNLOCK(object);
rtval = VOP_PUTPAGES(vp, m, bytes, flags, rtvals);
KASSERT(rtval != EOPNOTSUPP,
("vnode_pager: stale FS putpages\n"));
VM_OBJECT_WLOCK(object);
}
static int
vn_put(struct uvm_object *uobj, voff_t offlo, voff_t offhi, int flags)
{
struct vnode *vp = (struct vnode *)uobj;
return VOP_PUTPAGES(vp, offlo, offhi, flags);
}
int
RUMP_VOP_PUTPAGES(struct vnode *vp,
off_t offlo,
off_t offhi,
int flags)
{
int error;
rump_schedule();
error = VOP_PUTPAGES(vp, offlo, offhi, flags);
rump_unschedule();
return error;
}
/*
* Handes the read/write request given in 'bp' using the vnode's VOP_READ
* and VOP_WRITE operations.
*
* 'obp' is a pointer to the original request fed to the vnd device.
*/
static void
handle_with_rdwr(struct vnd_softc *vnd, const struct buf *obp, struct buf *bp)
{
bool doread;
off_t offset;
size_t len, resid;
struct vnode *vp;
doread = bp->b_flags & B_READ;
offset = obp->b_rawblkno * vnd->sc_dkdev.dk_label->d_secsize;
len = bp->b_bcount;
vp = vnd->sc_vp;
#if defined(DEBUG)
if (vnddebug & VDB_IO)
printf("vnd (rdwr): vp %p, %s, rawblkno 0x%" PRIx64
", secsize %d, offset %" PRIu64
", bcount %d\n",
vp, doread ? "read" : "write", obp->b_rawblkno,
vnd->sc_dkdev.dk_label->d_secsize, offset,
bp->b_bcount);
#endif
/* Issue the read or write operation. */
bp->b_error =
vn_rdwr(doread ? UIO_READ : UIO_WRITE,
vp, bp->b_data, len, offset, UIO_SYSSPACE,
IO_ADV_ENCODE(POSIX_FADV_NOREUSE), vnd->sc_cred, &resid, NULL);
bp->b_resid = resid;
mutex_enter(vp->v_interlock);
(void) VOP_PUTPAGES(vp, 0, 0,
PGO_ALLPAGES | PGO_CLEANIT | PGO_FREE | PGO_SYNCIO);
/* We need to increase the number of outputs on the vnode if
* there was any write to it. */
if (!doread) {
mutex_enter(vp->v_interlock);
vp->v_numoutput++;
mutex_exit(vp->v_interlock);
}
biodone(bp);
}
static int
unionfs_putpages(void *v)
{
struct vop_putpages_args /* {
struct vnode *a_vp;
voff_t a_offlo;
voff_t a_offhi;
int a_flags;
} */ *ap = v;
struct vnode *vp = ap->a_vp, *tvp;
struct unionfs_node *unp;
KASSERT(mutex_owned(vp->v_interlock));
unp = VTOUNIONFS(vp);
tvp = (unp->un_uppervp != NULLVP ? unp->un_uppervp : unp->un_lowervp);
KASSERT(tvp->v_interlock == vp->v_interlock);
if (ap->a_flags & PGO_RECLAIM) {
mutex_exit(vp->v_interlock);
return 0;
}
return VOP_PUTPAGES(tvp, ap->a_offlo, ap->a_offhi, ap->a_flags);
}
/*
* Read/write clusters from/to backing store.
* For persistent snapshots must be called with cl == 0. off is the
* offset into the snapshot.
*/
static int
fss_bs_io(struct fss_softc *sc, fss_io_type rw,
u_int32_t cl, off_t off, int len, void *data)
{
int error;
off += FSS_CLTOB(sc, cl);
vn_lock(sc->sc_bs_vp, LK_EXCLUSIVE|LK_RETRY);
error = vn_rdwr((rw == FSS_READ ? UIO_READ : UIO_WRITE), sc->sc_bs_vp,
data, len, off, UIO_SYSSPACE,
IO_ADV_ENCODE(POSIX_FADV_NOREUSE) | IO_NODELOCKED,
sc->sc_bs_lwp->l_cred, NULL, NULL);
if (error == 0) {
mutex_enter(sc->sc_bs_vp->v_interlock);
error = VOP_PUTPAGES(sc->sc_bs_vp, trunc_page(off),
round_page(off+len), PGO_CLEANIT | PGO_FREE | PGO_SYNCIO);
}
VOP_UNLOCK(sc->sc_bs_vp);
return error;
}
/*
* Truncate the inode oip to at most length size, freeing the
* disk blocks.
*/
int
ffs_truncate(struct vnode *ovp, off_t length, int ioflag, kauth_cred_t cred)
{
daddr_t lastblock;
struct inode *oip = VTOI(ovp);
daddr_t bn, lastiblock[UFS_NIADDR], indir_lbn[UFS_NIADDR];
daddr_t blks[UFS_NDADDR + UFS_NIADDR];
struct fs *fs;
int offset, pgoffset, level;
int64_t count, blocksreleased = 0;
int i, aflag, nblocks;
int error, allerror = 0;
off_t osize;
int sync;
struct ufsmount *ump = oip->i_ump;
if (ovp->v_type == VCHR || ovp->v_type == VBLK ||
ovp->v_type == VFIFO || ovp->v_type == VSOCK) {
KASSERT(oip->i_size == 0);
return 0;
}
if (length < 0)
return (EINVAL);
if (ovp->v_type == VLNK &&
(oip->i_size < ump->um_maxsymlinklen ||
(ump->um_maxsymlinklen == 0 && DIP(oip, blocks) == 0))) {
KDASSERT(length == 0);
memset(SHORTLINK(oip), 0, (size_t)oip->i_size);
oip->i_size = 0;
DIP_ASSIGN(oip, size, 0);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ffs_update(ovp, NULL, NULL, 0));
}
if (oip->i_size == length) {
/* still do a uvm_vnp_setsize() as writesize may be larger */
uvm_vnp_setsize(ovp, length);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ffs_update(ovp, NULL, NULL, 0));
}
fs = oip->i_fs;
if (length > ump->um_maxfilesize)
return (EFBIG);
if ((oip->i_flags & SF_SNAPSHOT) != 0)
ffs_snapremove(ovp);
osize = oip->i_size;
aflag = ioflag & IO_SYNC ? B_SYNC : 0;
/*
* 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 (ffs_lblkno(fs, osize) < UFS_NDADDR &&
ffs_lblkno(fs, osize) != ffs_lblkno(fs, length) &&
ffs_blkroundup(fs, osize) != osize) {
off_t eob;
eob = ffs_blkroundup(fs, osize);
uvm_vnp_setwritesize(ovp, eob);
error = ufs_balloc_range(ovp, osize, eob - osize,
cred, aflag);
if (error) {
(void) ffs_truncate(ovp, osize,
ioflag & IO_SYNC, cred);
return error;
}
if (ioflag & IO_SYNC) {
mutex_enter(ovp->v_interlock);
VOP_PUTPAGES(ovp,
trunc_page(osize & fs->fs_bmask),
round_page(eob), PGO_CLEANIT | PGO_SYNCIO |
PGO_JOURNALLOCKED);
}
}
uvm_vnp_setwritesize(ovp, length);
error = ufs_balloc_range(ovp, length - 1, 1, cred, aflag);
if (error) {
(void) ffs_truncate(ovp, osize, ioflag & IO_SYNC, cred);
return (error);
}
uvm_vnp_setsize(ovp, length);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
KASSERT(ovp->v_size == oip->i_size);
return (ffs_update(ovp, NULL, NULL, 0));
}
/*
* When truncating a regular file down to a non-block-aligned size,
* we must zero the part of last block which is past the new EOF.
* We must synchronously flush the zeroed pages to disk
* since the new pages will be invalidated as soon as we
* inform the VM system of the new, smaller size.
* We must do this before acquiring the GLOCK, since fetching
* the pages will acquire the GLOCK internally.
* So there is a window where another thread could see a whole
* zeroed page past EOF, but that's life.
*/
offset = ffs_blkoff(fs, length);
pgoffset = length & PAGE_MASK;
if (ovp->v_type == VREG && (pgoffset != 0 || offset != 0) &&
osize > length) {
daddr_t lbn;
voff_t eoz;
int size;
if (offset != 0) {
error = ufs_balloc_range(ovp, length - 1, 1, cred,
aflag);
if (error)
return error;
}
lbn = ffs_lblkno(fs, length);
size = ffs_blksize(fs, oip, lbn);
eoz = MIN(MAX(ffs_lblktosize(fs, lbn) + size, round_page(pgoffset)),
osize);
ubc_zerorange(&ovp->v_uobj, length, eoz - length,
UBC_UNMAP_FLAG(ovp));
if (round_page(eoz) > round_page(length)) {
mutex_enter(ovp->v_interlock);
error = VOP_PUTPAGES(ovp, round_page(length),
round_page(eoz),
PGO_CLEANIT | PGO_DEACTIVATE | PGO_JOURNALLOCKED |
((ioflag & IO_SYNC) ? PGO_SYNCIO : 0));
if (error)
return error;
}
}
genfs_node_wrlock(ovp);
oip->i_size = length;
DIP_ASSIGN(oip, size, length);
uvm_vnp_setsize(ovp, length);
/*
* 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 = ffs_lblkno(fs, length + fs->fs_bsize - 1) - 1;
lastiblock[SINGLE] = lastblock - UFS_NDADDR;
lastiblock[DOUBLE] = lastiblock[SINGLE] - FFS_NINDIR(fs);
lastiblock[TRIPLE] = lastiblock[DOUBLE] - FFS_NINDIR(fs) * FFS_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.
*/
sync = 0;
for (level = TRIPLE; level >= SINGLE; level--) {
blks[UFS_NDADDR + level] = DIP(oip, ib[level]);
if (lastiblock[level] < 0 && blks[UFS_NDADDR + level] != 0) {
sync = 1;
DIP_ASSIGN(oip, ib[level], 0);
lastiblock[level] = -1;
}
}
for (i = 0; i < UFS_NDADDR; i++) {
blks[i] = DIP(oip, db[i]);
if (i > lastblock && blks[i] != 0) {
sync = 1;
DIP_ASSIGN(oip, db[i], 0);
}
}
oip->i_flag |= IN_CHANGE | IN_UPDATE;
if (sync) {
error = ffs_update(ovp, NULL, NULL, UPDATE_WAIT);
if (error && !allerror)
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 < UFS_NDADDR; i++) {
bn = DIP(oip, db[i]);
DIP_ASSIGN(oip, db[i], blks[i]);
blks[i] = bn;
}
for (i = 0; i < UFS_NIADDR; i++) {
bn = DIP(oip, ib[i]);
DIP_ASSIGN(oip, ib[i], blks[UFS_NDADDR + i]);
blks[UFS_NDADDR + i] = bn;
}
oip->i_size = osize;
DIP_ASSIGN(oip, size, osize);
error = vtruncbuf(ovp, lastblock + 1, 0, 0);
if (error && !allerror)
allerror = error;
/*
* Indirect blocks first.
*/
indir_lbn[SINGLE] = -UFS_NDADDR;
indir_lbn[DOUBLE] = indir_lbn[SINGLE] - FFS_NINDIR(fs) - 1;
indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - FFS_NINDIR(fs) * FFS_NINDIR(fs) - 1;
for (level = TRIPLE; level >= SINGLE; level--) {
if (oip->i_ump->um_fstype == UFS1)
bn = ufs_rw32(oip->i_ffs1_ib[level],UFS_FSNEEDSWAP(fs));
else
bn = ufs_rw64(oip->i_ffs2_ib[level],UFS_FSNEEDSWAP(fs));
if (bn != 0) {
error = ffs_indirtrunc(oip, indir_lbn[level],
FFS_FSBTODB(fs, bn), lastiblock[level], level, &count);
if (error)
allerror = error;
blocksreleased += count;
if (lastiblock[level] < 0) {
DIP_ASSIGN(oip, ib[level], 0);
if (oip->i_ump->um_mountp->mnt_wapbl) {
UFS_WAPBL_REGISTER_DEALLOCATION(
oip->i_ump->um_mountp,
FFS_FSBTODB(fs, bn), fs->fs_bsize);
} else
ffs_blkfree(fs, oip->i_devvp, bn,
fs->fs_bsize, oip->i_number);
blocksreleased += nblocks;
}
}
if (lastiblock[level] >= 0)
goto done;
}
/*
* All whole direct blocks or frags.
*/
for (i = UFS_NDADDR - 1; i > lastblock; i--) {
long bsize;
if (oip->i_ump->um_fstype == UFS1)
bn = ufs_rw32(oip->i_ffs1_db[i], UFS_FSNEEDSWAP(fs));
else
bn = ufs_rw64(oip->i_ffs2_db[i], UFS_FSNEEDSWAP(fs));
if (bn == 0)
continue;
DIP_ASSIGN(oip, db[i], 0);
bsize = ffs_blksize(fs, oip, i);
if ((oip->i_ump->um_mountp->mnt_wapbl) &&
(ovp->v_type != VREG)) {
UFS_WAPBL_REGISTER_DEALLOCATION(oip->i_ump->um_mountp,
FFS_FSBTODB(fs, bn), bsize);
} else
ffs_blkfree(fs, oip->i_devvp, bn, bsize, oip->i_number);
blocksreleased += btodb(bsize);
}
if (lastblock < 0)
goto done;
/*
* Finally, look for a change in size of the
* last direct block; release any frags.
*/
if (oip->i_ump->um_fstype == UFS1)
bn = ufs_rw32(oip->i_ffs1_db[lastblock], UFS_FSNEEDSWAP(fs));
else
bn = ufs_rw64(oip->i_ffs2_db[lastblock], UFS_FSNEEDSWAP(fs));
if (bn != 0) {
long oldspace, newspace;
/*
* Calculate amount of space we're giving
* back as old block size minus new block size.
*/
oldspace = ffs_blksize(fs, oip, lastblock);
oip->i_size = length;
DIP_ASSIGN(oip, size, length);
newspace = ffs_blksize(fs, oip, lastblock);
if (newspace == 0)
panic("itrunc: 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 += ffs_numfrags(fs, newspace);
if ((oip->i_ump->um_mountp->mnt_wapbl) &&
(ovp->v_type != VREG)) {
UFS_WAPBL_REGISTER_DEALLOCATION(
oip->i_ump->um_mountp, FFS_FSBTODB(fs, bn),
oldspace - newspace);
} else
ffs_blkfree(fs, oip->i_devvp, bn,
oldspace - newspace, oip->i_number);
blocksreleased += btodb(oldspace - newspace);
}
}
done:
#ifdef DIAGNOSTIC
for (level = SINGLE; level <= TRIPLE; level++)
if (blks[UFS_NDADDR + level] != DIP(oip, ib[level]))
panic("itrunc1");
for (i = 0; i < UFS_NDADDR; i++)
if (blks[i] != DIP(oip, db[i]))
panic("itrunc2");
if (length == 0 &&
(!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd)))
panic("itrunc3");
#endif /* DIAGNOSTIC */
/*
* Put back the real size.
*/
oip->i_size = length;
DIP_ASSIGN(oip, size, length);
DIP_ADD(oip, blocks, -blocksreleased);
genfs_node_unlock(ovp);
oip->i_flag |= IN_CHANGE;
UFS_WAPBL_UPDATE(ovp, NULL, NULL, 0);
#if defined(QUOTA) || defined(QUOTA2)
(void) chkdq(oip, -blocksreleased, NOCRED, 0);
#endif
KASSERT(ovp->v_type != VREG || ovp->v_size == oip->i_size);
return (allerror);
}
int
do_posix_fadvise(int fd, off_t offset, off_t len, int advice)
{
file_t *fp;
vnode_t *vp;
off_t endoffset;
int error;
CTASSERT(POSIX_FADV_NORMAL == UVM_ADV_NORMAL);
CTASSERT(POSIX_FADV_RANDOM == UVM_ADV_RANDOM);
CTASSERT(POSIX_FADV_SEQUENTIAL == UVM_ADV_SEQUENTIAL);
if (len == 0) {
endoffset = INT64_MAX;
} else if (len > 0 && (INT64_MAX - offset) >= len) {
endoffset = offset + len;
} else {
return EINVAL;
}
if ((fp = fd_getfile(fd)) == NULL) {
return EBADF;
}
if (fp->f_type != DTYPE_VNODE) {
if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
error = ESPIPE;
} else {
error = EOPNOTSUPP;
}
fd_putfile(fd);
return error;
}
switch (advice) {
case POSIX_FADV_WILLNEED:
case POSIX_FADV_DONTNEED:
vp = fp->f_vnode;
if (vp->v_type != VREG && vp->v_type != VBLK) {
fd_putfile(fd);
return 0;
}
break;
}
switch (advice) {
case POSIX_FADV_NORMAL:
case POSIX_FADV_RANDOM:
case POSIX_FADV_SEQUENTIAL:
/*
* We ignore offset and size. Must lock the file to
* do this, as f_advice is sub-word sized.
*/
mutex_enter(&fp->f_lock);
fp->f_advice = (u_char)advice;
mutex_exit(&fp->f_lock);
error = 0;
break;
case POSIX_FADV_WILLNEED:
vp = fp->f_vnode;
error = uvm_readahead(&vp->v_uobj, offset, endoffset - offset);
break;
case POSIX_FADV_DONTNEED:
vp = fp->f_vnode;
/*
* Align the region to page boundaries as VOP_PUTPAGES expects
* by shrinking it. We shrink instead of expand because we
* do not want to deactivate cache outside of the requested
* region. It means that if the specified region is smaller
* than PAGE_SIZE, we do nothing.
*/
if (round_page(offset) < trunc_page(endoffset) &&
offset <= round_page(offset)) {
mutex_enter(vp->v_interlock);
error = VOP_PUTPAGES(vp,
round_page(offset), trunc_page(endoffset),
PGO_DEACTIVATE | PGO_CLEANIT);
} else {
error = 0;
}
break;
case POSIX_FADV_NOREUSE:
/* Not implemented yet. */
error = 0;
break;
default:
error = EINVAL;
break;
}
fd_putfile(fd);
return error;
}
/* ARGSUSED */
int
ffs_full_fsync(struct vnode *vp, int flags)
{
int error, i, uflags;
struct mount *mp;
KASSERT(vp->v_tag == VT_UFS);
KASSERT(VTOI(vp) != NULL);
KASSERT(vp->v_type != VCHR && vp->v_type != VBLK);
error = 0;
uflags = UPDATE_CLOSE | ((flags & FSYNC_WAIT) ? UPDATE_WAIT : 0);
mp = vp->v_mount;
/*
* Flush all dirty data associated with the vnode.
*/
if (vp->v_type == VREG) {
int pflags = PGO_ALLPAGES | PGO_CLEANIT;
if ((flags & FSYNC_WAIT))
pflags |= PGO_SYNCIO;
if (fstrans_getstate(mp) == FSTRANS_SUSPENDING)
pflags |= PGO_FREE;
mutex_enter(vp->v_interlock);
error = VOP_PUTPAGES(vp, 0, 0, pflags);
if (error)
return error;
}
#ifdef WAPBL
if (mp && mp->mnt_wapbl) {
/*
* Don't bother writing out metadata if the syncer is
* making the request. We will let the sync vnode
* write it out in a single burst through a call to
* VFS_SYNC().
*/
if ((flags & (FSYNC_DATAONLY | FSYNC_LAZY)) != 0)
return 0;
if ((VTOI(vp)->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE
| IN_MODIFY | IN_MODIFIED | IN_ACCESSED)) != 0) {
error = UFS_WAPBL_BEGIN(mp);
if (error)
return error;
error = ffs_update(vp, NULL, NULL, uflags);
UFS_WAPBL_END(mp);
}
if (error || (flags & FSYNC_NOLOG) != 0)
return error;
/*
* Don't flush the log if the vnode being flushed
* contains no dirty buffers that could be in the log.
*/
if (!LIST_EMPTY(&vp->v_dirtyblkhd)) {
error = wapbl_flush(mp->mnt_wapbl, 0);
if (error)
return error;
}
if ((flags & FSYNC_WAIT) != 0) {
mutex_enter(vp->v_interlock);
while (vp->v_numoutput != 0)
cv_wait(&vp->v_cv, vp->v_interlock);
mutex_exit(vp->v_interlock);
}
return error;
}
#endif /* WAPBL */
error = vflushbuf(vp, (flags & FSYNC_WAIT) != 0);
if (error == 0)
error = ffs_update(vp, NULL, NULL, uflags);
if (error == 0 && (flags & FSYNC_CACHE) != 0) {
i = 1;
(void)VOP_IOCTL(VTOI(vp)->i_devvp, DIOCCACHESYNC, &i, FWRITE,
kauth_cred_get());
}
return error;
}
int
ffs_fsync(void *v)
{
struct vop_fsync_args /* {
struct vnode *a_vp;
kauth_cred_t a_cred;
int a_flags;
off_t a_offlo;
off_t a_offhi;
struct lwp *a_l;
} */ *ap = v;
struct buf *bp;
int num, error, i;
struct indir ia[NIADDR + 1];
int bsize;
daddr_t blk_high;
struct vnode *vp;
struct mount *mp;
vp = ap->a_vp;
mp = vp->v_mount;
fstrans_start(mp, FSTRANS_LAZY);
if ((ap->a_offlo == 0 && ap->a_offhi == 0) || (vp->v_type != VREG)) {
error = ffs_full_fsync(vp, ap->a_flags);
goto out;
}
bsize = mp->mnt_stat.f_iosize;
blk_high = ap->a_offhi / bsize;
if (ap->a_offhi % bsize != 0)
blk_high++;
/*
* First, flush all pages in range.
*/
mutex_enter(vp->v_interlock);
error = VOP_PUTPAGES(vp, trunc_page(ap->a_offlo),
round_page(ap->a_offhi), PGO_CLEANIT |
((ap->a_flags & FSYNC_WAIT) ? PGO_SYNCIO : 0));
if (error) {
goto out;
}
#ifdef WAPBL
KASSERT(vp->v_type == VREG);
if (mp->mnt_wapbl) {
/*
* Don't bother writing out metadata if the syncer is
* making the request. We will let the sync vnode
* write it out in a single burst through a call to
* VFS_SYNC().
*/
if ((ap->a_flags & (FSYNC_DATAONLY | FSYNC_LAZY)) != 0) {
fstrans_done(mp);
return 0;
}
error = 0;
if (vp->v_tag == VT_UFS && VTOI(vp)->i_flag &
(IN_ACCESS | IN_CHANGE | IN_UPDATE | IN_MODIFY |
IN_MODIFIED | IN_ACCESSED)) {
error = UFS_WAPBL_BEGIN(mp);
if (error) {
fstrans_done(mp);
return error;
}
error = ffs_update(vp, NULL, NULL, UPDATE_CLOSE |
((ap->a_flags & FSYNC_WAIT) ? UPDATE_WAIT : 0));
UFS_WAPBL_END(mp);
}
if (error || (ap->a_flags & FSYNC_NOLOG) != 0) {
fstrans_done(mp);
return error;
}
error = wapbl_flush(mp->mnt_wapbl, 0);
fstrans_done(mp);
return error;
}
#endif /* WAPBL */
/*
* Then, flush indirect blocks.
*/
if (blk_high >= NDADDR) {
error = ufs_getlbns(vp, blk_high, ia, &num);
if (error)
goto out;
mutex_enter(&bufcache_lock);
for (i = 0; i < num; i++) {
if ((bp = incore(vp, ia[i].in_lbn)) == NULL)
continue;
if ((bp->b_cflags & BC_BUSY) != 0 ||
(bp->b_oflags & BO_DELWRI) == 0)
continue;
bp->b_cflags |= BC_BUSY | BC_VFLUSH;
mutex_exit(&bufcache_lock);
bawrite(bp);
mutex_enter(&bufcache_lock);
}
mutex_exit(&bufcache_lock);
}
if (ap->a_flags & FSYNC_WAIT) {
mutex_enter(vp->v_interlock);
while (vp->v_numoutput > 0)
cv_wait(&vp->v_cv, vp->v_interlock);
mutex_exit(vp->v_interlock);
}
error = ffs_update(vp, NULL, NULL, UPDATE_CLOSE |
(((ap->a_flags & (FSYNC_WAIT | FSYNC_DATAONLY)) == FSYNC_WAIT)
? UPDATE_WAIT : 0));
if (error == 0 && ap->a_flags & FSYNC_CACHE) {
int l = 0;
VOP_IOCTL(VTOI(vp)->i_devvp, DIOCCACHESYNC, &l, FWRITE,
curlwp->l_cred);
}
out:
fstrans_done(mp);
return error;
}
int
lfs_truncate(struct vnode *ovp, off_t length, int ioflag, kauth_cred_t cred)
{
daddr_t lastblock;
struct inode *oip = VTOI(ovp);
daddr_t bn, lbn, lastiblock[ULFS_NIADDR], indir_lbn[ULFS_NIADDR];
/* XXX ondisk32 */
int32_t newblks[ULFS_NDADDR + ULFS_NIADDR];
struct lfs *fs;
struct buf *bp;
int offset, size, level;
daddr_t count, rcount;
daddr_t blocksreleased = 0, real_released = 0;
int i, nblocks;
int aflags, error, allerror = 0;
off_t osize;
long lastseg;
size_t bc;
int obufsize, odb;
int usepc;
if (ovp->v_type == VCHR || ovp->v_type == VBLK ||
ovp->v_type == VFIFO || ovp->v_type == VSOCK) {
KASSERT(oip->i_size == 0);
return 0;
}
if (length < 0)
return (EINVAL);
/*
* Just return and not update modification times.
*/
if (oip->i_size == length) {
/* still do a uvm_vnp_setsize() as writesize may be larger */
uvm_vnp_setsize(ovp, length);
return (0);
}
fs = oip->i_lfs;
if (ovp->v_type == VLNK &&
(oip->i_size < fs->um_maxsymlinklen ||
(fs->um_maxsymlinklen == 0 &&
oip->i_ffs1_blocks == 0))) {
#ifdef DIAGNOSTIC
if (length != 0)
panic("lfs_truncate: partial truncate of symlink");
#endif
memset((char *)SHORTLINK(oip), 0, (u_int)oip->i_size);
oip->i_size = oip->i_ffs1_size = 0;
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (lfs_update(ovp, NULL, NULL, 0));
}
if (oip->i_size == length) {
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (lfs_update(ovp, NULL, NULL, 0));
}
lfs_imtime(fs);
osize = oip->i_size;
usepc = (ovp->v_type == VREG && ovp != fs->lfs_ivnode);
ASSERT_NO_SEGLOCK(fs);
/*
* 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->um_maxfilesize)
return (EFBIG);
aflags = B_CLRBUF;
if (ioflag & IO_SYNC)
aflags |= B_SYNC;
if (usepc) {
if (lfs_lblkno(fs, osize) < ULFS_NDADDR &&
lfs_lblkno(fs, osize) != lfs_lblkno(fs, length) &&
lfs_blkroundup(fs, osize) != osize) {
off_t eob;
eob = lfs_blkroundup(fs, osize);
uvm_vnp_setwritesize(ovp, eob);
error = ulfs_balloc_range(ovp, osize,
eob - osize, cred, aflags);
if (error) {
(void) lfs_truncate(ovp, osize,
ioflag & IO_SYNC, cred);
return error;
}
if (ioflag & IO_SYNC) {
mutex_enter(ovp->v_interlock);
VOP_PUTPAGES(ovp,
trunc_page(osize & lfs_sb_getbmask(fs)),
round_page(eob),
PGO_CLEANIT | PGO_SYNCIO);
}
}
uvm_vnp_setwritesize(ovp, length);
error = ulfs_balloc_range(ovp, length - 1, 1, cred,
aflags);
if (error) {
(void) lfs_truncate(ovp, osize,
ioflag & IO_SYNC, cred);
return error;
}
uvm_vnp_setsize(ovp, length);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
KASSERT(ovp->v_size == oip->i_size);
oip->i_lfs_hiblk = lfs_lblkno(fs, oip->i_size + lfs_sb_getbsize(fs) - 1) - 1;
return (lfs_update(ovp, NULL, NULL, 0));
} else {
error = lfs_reserve(fs, ovp, NULL,
lfs_btofsb(fs, (ULFS_NIADDR + 2) << lfs_sb_getbshift(fs)));
if (error)
return (error);
error = lfs_balloc(ovp, length - 1, 1, cred,
aflags, &bp);
lfs_reserve(fs, ovp, NULL,
-lfs_btofsb(fs, (ULFS_NIADDR + 2) << lfs_sb_getbshift(fs)));
if (error)
return (error);
oip->i_ffs1_size = oip->i_size = length;
uvm_vnp_setsize(ovp, length);
(void) VOP_BWRITE(bp->b_vp, bp);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
oip->i_lfs_hiblk = lfs_lblkno(fs, oip->i_size + lfs_sb_getbsize(fs) - 1) - 1;
return (lfs_update(ovp, NULL, NULL, 0));
}
}
if ((error = lfs_reserve(fs, ovp, NULL,
lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs)))) != 0)
return (error);
/*
* 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 = lfs_blkoff(fs, length);
lastseg = -1;
bc = 0;
if (ovp != fs->lfs_ivnode)
lfs_seglock(fs, SEGM_PROT);
if (offset == 0) {
oip->i_size = oip->i_ffs1_size = length;
} else if (!usepc) {
lbn = lfs_lblkno(fs, length);
aflags = B_CLRBUF;
if (ioflag & IO_SYNC)
aflags |= B_SYNC;
error = lfs_balloc(ovp, length - 1, 1, cred, aflags, &bp);
if (error) {
lfs_reserve(fs, ovp, NULL,
-lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs)));
goto errout;
}
obufsize = bp->b_bufsize;
odb = lfs_btofsb(fs, bp->b_bcount);
oip->i_size = oip->i_ffs1_size = length;
size = lfs_blksize(fs, oip, lbn);
if (ovp->v_type != VDIR)
memset((char *)bp->b_data + offset, 0,
(u_int)(size - offset));
allocbuf(bp, size, 1);
if ((bp->b_flags & B_LOCKED) != 0 && bp->b_iodone == NULL) {
mutex_enter(&lfs_lock);
locked_queue_bytes -= obufsize - bp->b_bufsize;
mutex_exit(&lfs_lock);
}
if (bp->b_oflags & BO_DELWRI) {
lfs_sb_addavail(fs, odb - lfs_btofsb(fs, size));
/* XXX shouldn't this wake up on lfs_availsleep? */
}
(void) VOP_BWRITE(bp->b_vp, bp);
} else { /* vp->v_type == VREG && length < osize && offset != 0 */
/*
* When truncating a regular file down to a non-block-aligned
* size, we must zero the part of last block which is past
* the new EOF. We must synchronously flush the zeroed pages
* to disk since the new pages will be invalidated as soon
* as we inform the VM system of the new, smaller size.
* We must do this before acquiring the GLOCK, since fetching
* the pages will acquire the GLOCK internally.
* So there is a window where another thread could see a whole
* zeroed page past EOF, but that's life.
*/
daddr_t xlbn;
voff_t eoz;
aflags = ioflag & IO_SYNC ? B_SYNC : 0;
error = ulfs_balloc_range(ovp, length - 1, 1, cred, aflags);
if (error) {
lfs_reserve(fs, ovp, NULL,
-lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs)));
goto errout;
}
xlbn = lfs_lblkno(fs, length);
size = lfs_blksize(fs, oip, xlbn);
eoz = MIN(lfs_lblktosize(fs, xlbn) + size, osize);
ubc_zerorange(&ovp->v_uobj, length, eoz - length,
UBC_UNMAP_FLAG(ovp));
if (round_page(eoz) > round_page(length)) {
mutex_enter(ovp->v_interlock);
error = VOP_PUTPAGES(ovp, round_page(length),
round_page(eoz),
PGO_CLEANIT | PGO_DEACTIVATE |
((ioflag & IO_SYNC) ? PGO_SYNCIO : 0));
if (error) {
lfs_reserve(fs, ovp, NULL,
-lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs)));
goto errout;
}
}
}
genfs_node_wrlock(ovp);
oip->i_size = oip->i_ffs1_size = length;
uvm_vnp_setsize(ovp, length);
/*
* 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.
*/
/* Avoid sign overflow - XXX assumes that off_t is a quad_t. */
if (length > QUAD_MAX - lfs_sb_getbsize(fs))
lastblock = lfs_lblkno(fs, QUAD_MAX - lfs_sb_getbsize(fs));
else
lastblock = lfs_lblkno(fs, length + lfs_sb_getbsize(fs) - 1) - 1;
lastiblock[SINGLE] = lastblock - ULFS_NDADDR;
lastiblock[DOUBLE] = lastiblock[SINGLE] - LFS_NINDIR(fs);
lastiblock[TRIPLE] = lastiblock[DOUBLE] - LFS_NINDIR(fs) * LFS_NINDIR(fs);
nblocks = lfs_btofsb(fs, lfs_sb_getbsize(fs));
/*
* Record changed file and block pointers before we start
* freeing blocks. lastiblock values are also normalized to -1
* for calls to lfs_indirtrunc below.
*/
memcpy((void *)newblks, (void *)&oip->i_ffs1_db[0], sizeof newblks);
for (level = TRIPLE; level >= SINGLE; level--)
if (lastiblock[level] < 0) {
newblks[ULFS_NDADDR+level] = 0;
lastiblock[level] = -1;
}
for (i = ULFS_NDADDR - 1; i > lastblock; i--)
newblks[i] = 0;
oip->i_size = oip->i_ffs1_size = osize;
error = lfs_vtruncbuf(ovp, lastblock + 1, false, 0);
if (error && !allerror)
allerror = error;
/*
* Indirect blocks first.
*/
indir_lbn[SINGLE] = -ULFS_NDADDR;
indir_lbn[DOUBLE] = indir_lbn[SINGLE] - LFS_NINDIR(fs) - 1;
indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - LFS_NINDIR(fs) * LFS_NINDIR(fs) - 1;
for (level = TRIPLE; level >= SINGLE; level--) {
bn = oip->i_ffs1_ib[level];
if (bn != 0) {
error = lfs_indirtrunc(oip, indir_lbn[level],
bn, lastiblock[level],
level, &count, &rcount,
&lastseg, &bc);
if (error)
allerror = error;
real_released += rcount;
blocksreleased += count;
if (lastiblock[level] < 0) {
if (oip->i_ffs1_ib[level] > 0)
real_released += nblocks;
blocksreleased += nblocks;
oip->i_ffs1_ib[level] = 0;
lfs_blkfree(fs, oip, bn, lfs_sb_getbsize(fs),
&lastseg, &bc);
lfs_deregister_block(ovp, bn);
}
}
if (lastiblock[level] >= 0)
goto done;
}
/*
* All whole direct blocks or frags.
*/
for (i = ULFS_NDADDR - 1; i > lastblock; i--) {
long bsize, obsize;
bn = oip->i_ffs1_db[i];
if (bn == 0)
continue;
bsize = lfs_blksize(fs, oip, i);
if (oip->i_ffs1_db[i] > 0) {
/* Check for fragment size changes */
obsize = oip->i_lfs_fragsize[i];
real_released += lfs_btofsb(fs, obsize);
oip->i_lfs_fragsize[i] = 0;
} else
obsize = 0;
blocksreleased += lfs_btofsb(fs, bsize);
oip->i_ffs1_db[i] = 0;
lfs_blkfree(fs, oip, bn, obsize, &lastseg, &bc);
lfs_deregister_block(ovp, bn);
}
if (lastblock < 0)
goto done;
/*
* Finally, look for a change in size of the
* last direct block; release any frags.
*/
bn = oip->i_ffs1_db[lastblock];
if (bn != 0) {
long oldspace, newspace;
#if 0
long olddspace;
#endif
/*
* Calculate amount of space we're giving
* back as old block size minus new block size.
*/
oldspace = lfs_blksize(fs, oip, lastblock);
#if 0
olddspace = oip->i_lfs_fragsize[lastblock];
#endif
oip->i_size = oip->i_ffs1_size = length;
newspace = lfs_blksize(fs, oip, lastblock);
if (newspace == 0)
panic("itrunc: newspace");
if (oldspace - newspace > 0) {
blocksreleased += lfs_btofsb(fs, oldspace - newspace);
}
#if 0
if (bn > 0 && olddspace - newspace > 0) {
/* No segment accounting here, just vnode */
real_released += lfs_btofsb(fs, olddspace - newspace);
}
#endif
}
done:
/* Finish segment accounting corrections */
lfs_update_seguse(fs, oip, lastseg, bc);
#ifdef DIAGNOSTIC
for (level = SINGLE; level <= TRIPLE; level++)
if ((newblks[ULFS_NDADDR + level] == 0) !=
((oip->i_ffs1_ib[level]) == 0)) {
panic("lfs itrunc1");
}
for (i = 0; i < ULFS_NDADDR; i++)
if ((newblks[i] == 0) != (oip->i_ffs1_db[i] == 0)) {
panic("lfs itrunc2");
}
if (length == 0 &&
(!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd)))
panic("lfs itrunc3");
#endif /* DIAGNOSTIC */
/*
* Put back the real size.
*/
oip->i_size = oip->i_ffs1_size = length;
oip->i_lfs_effnblks -= blocksreleased;
oip->i_ffs1_blocks -= real_released;
mutex_enter(&lfs_lock);
lfs_sb_addbfree(fs, blocksreleased);
mutex_exit(&lfs_lock);
#ifdef DIAGNOSTIC
if (oip->i_size == 0 &&
(oip->i_ffs1_blocks != 0 || oip->i_lfs_effnblks != 0)) {
printf("lfs_truncate: truncate to 0 but %d blks/%jd effblks\n",
oip->i_ffs1_blocks, (intmax_t)oip->i_lfs_effnblks);
panic("lfs_truncate: persistent blocks");
}
#endif
/*
* If we truncated to zero, take us off the paging queue.
*/
mutex_enter(&lfs_lock);
if (oip->i_size == 0 && oip->i_flags & IN_PAGING) {
oip->i_flags &= ~IN_PAGING;
TAILQ_REMOVE(&fs->lfs_pchainhd, oip, i_lfs_pchain);
}
mutex_exit(&lfs_lock);
oip->i_flag |= IN_CHANGE;
#if defined(LFS_QUOTA) || defined(LFS_QUOTA2)
(void) lfs_chkdq(oip, -blocksreleased, NOCRED, 0);
#endif
lfs_reserve(fs, ovp, NULL,
-lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs)));
genfs_node_unlock(ovp);
errout:
oip->i_lfs_hiblk = lfs_lblkno(fs, oip->i_size + lfs_sb_getbsize(fs) - 1) - 1;
if (ovp != fs->lfs_ivnode)
lfs_segunlock(fs);
return (allerror ? allerror : error);
}
* Destroy any in core blocks past the truncation length.
* Inlined from vtruncbuf, so that lfs_avail could be updated.
* We take the seglock to prevent cleaning from occurring while we are
* invalidating blocks.
*/
static int
lfs_vtruncbuf(struct vnode *vp, daddr_t lbn, bool catch, int slptimeo)
{
struct buf *bp, *nbp;
int error;
struct lfs *fs;
voff_t off;
off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift);
mutex_enter(vp->v_interlock);
error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO);
if (error)
return error;
fs = VTOI(vp)->i_lfs;
ASSERT_SEGLOCK(fs);
mutex_enter(&bufcache_lock);
restart:
for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
nbp = LIST_NEXT(bp, b_vnbufs);
if (bp->b_lblkno < lbn)
continue;
error = bbusy(bp, catch, slptimeo, NULL);
if (error == EPASSTHROUGH)