void
vnode_pager_update_writecount(vm_object_t object, vm_offset_t start,
vm_offset_t end)
{
struct vnode *vp;
vm_ooffset_t old_wm;
VM_OBJECT_WLOCK(object);
if (object->type != OBJT_VNODE) {
VM_OBJECT_WUNLOCK(object);
return;
}
old_wm = object->un_pager.vnp.writemappings;
object->un_pager.vnp.writemappings += (vm_ooffset_t)end - start;
vp = object->handle;
if (old_wm == 0 && object->un_pager.vnp.writemappings != 0) {
ASSERT_VOP_ELOCKED(vp, "v_writecount inc");
VOP_ADD_WRITECOUNT(vp, 1);
CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
__func__, vp, vp->v_writecount);
} else if (old_wm != 0 && object->un_pager.vnp.writemappings == 0) {
ASSERT_VOP_ELOCKED(vp, "v_writecount dec");
VOP_ADD_WRITECOUNT(vp, -1);
CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
__func__, vp, vp->v_writecount);
}
VM_OBJECT_WUNLOCK(object);
}
static int
fuse_vnode_alloc(struct mount *mp,
struct thread *td,
uint64_t nodeid,
enum vtype vtyp,
struct vnode **vpp)
{
struct fuse_vnode_data *fvdat;
struct vnode *vp2;
int err = 0;
FS_DEBUG("been asked for vno #%ju\n", (uintmax_t)nodeid);
if (vtyp == VNON) {
return EINVAL;
}
*vpp = NULL;
err = vfs_hash_get(mp, fuse_vnode_hash(nodeid), LK_EXCLUSIVE, td, vpp,
fuse_vnode_cmp, &nodeid);
if (err)
return (err);
if (*vpp) {
MPASS((*vpp)->v_type == vtyp && (*vpp)->v_data != NULL);
FS_DEBUG("vnode taken from hash\n");
return (0);
}
fvdat = malloc(sizeof(*fvdat), M_FUSEVN, M_WAITOK | M_ZERO);
err = getnewvnode("fuse", mp, &fuse_vnops, vpp);
if (err) {
free(fvdat, M_FUSEVN);
return (err);
}
lockmgr((*vpp)->v_vnlock, LK_EXCLUSIVE, NULL);
fuse_vnode_init(*vpp, fvdat, nodeid, vtyp);
err = insmntque(*vpp, mp);
ASSERT_VOP_ELOCKED(*vpp, "fuse_vnode_alloc");
if (err) {
free(fvdat, M_FUSEVN);
*vpp = NULL;
return (err);
}
err = vfs_hash_insert(*vpp, fuse_vnode_hash(nodeid), LK_EXCLUSIVE,
td, &vp2, fuse_vnode_cmp, &nodeid);
if (err)
return (err);
if (vp2 != NULL) {
*vpp = vp2;
return (0);
}
ASSERT_VOP_ELOCKED(*vpp, "fuse_vnode_alloc");
return (0);
}
void
zfs_rmnode(znode_t *zp)
{
zfsvfs_t *zfsvfs = zp->z_zfsvfs;
objset_t *os = zfsvfs->z_os;
znode_t *xzp = NULL;
dmu_tx_t *tx;
uint64_t acl_obj;
uint64_t xattr_obj;
int error;
ASSERT(zp->z_links == 0);
ASSERT_VOP_ELOCKED(ZTOV(zp), __func__);
/*
* If this is an attribute directory, purge its contents.
*/
if (ZTOV(zp) != NULL && ZTOV(zp)->v_type == VDIR &&
(zp->z_pflags & ZFS_XATTR)) {
if (zfs_purgedir(zp) != 0) {
/*
* Not enough space to delete some xattrs.
* Leave it in the unlinked set.
*/
zfs_znode_dmu_fini(zp);
zfs_znode_free(zp);
return;
}
} else {
/*
* Free up all the data in the file. We don't do this for
* XATTR directories because we need truncate and remove to be
* in the same tx, like in zfs_znode_delete(). Otherwise, if
* we crash here we'll end up with an inconsistent truncated
* zap object in the delete queue. Note a truncated file is
* harmless since it only contains user data.
*/
error = dmu_free_long_range(os, zp->z_id, 0, DMU_OBJECT_END);
if (error) {
/*
* Not enough space. Leave the file in the unlinked
* set.
*/
zfs_znode_dmu_fini(zp);
zfs_znode_free(zp);
return;
}
}
/*
* If the file has extended attributes, we're going to unlink
* the xattr dir.
*/
error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
&xattr_obj, sizeof (xattr_obj));
if (error == 0 && xattr_obj) {
error = zfs_zget(zfsvfs, xattr_obj, &xzp);
ASSERT3S(error, ==, 0);
vn_lock(ZTOV(xzp), LK_EXCLUSIVE | LK_RETRY);
}
void
vnode_destroy_vobject(struct vnode *vp)
{
struct vm_object *obj;
obj = vp->v_object;
if (obj == NULL)
return;
ASSERT_VOP_ELOCKED(vp, "vnode_destroy_vobject");
VM_OBJECT_WLOCK(obj);
if (obj->ref_count == 0) {
/*
* don't double-terminate the object
*/
if ((obj->flags & OBJ_DEAD) == 0)
vm_object_terminate(obj);
else
VM_OBJECT_WUNLOCK(obj);
} else {
/*
* Woe to the process that tries to page now :-).
*/
vm_pager_deallocate(obj);
VM_OBJECT_WUNLOCK(obj);
}
vp->v_object = NULL;
}
void
fuse_vnode_open(struct vnode *vp, int32_t fuse_open_flags, struct thread *td)
{
/*
* Funcation is called for every vnode open.
* Merge fuse_open_flags it may be 0
*/
/*
* Ideally speaking, direct io should be enabled on
* fd's but do not see of any way of providing that
* this implementation.
*
* Also cannot think of a reason why would two
* different fd's on same vnode would like
* have DIRECT_IO turned on and off. But linux
* based implementation works on an fd not an
* inode and provides such a feature.
*
* XXXIP: Handle fd based DIRECT_IO
*/
if (fuse_open_flags & FOPEN_DIRECT_IO) {
ASSERT_VOP_ELOCKED(vp, __func__);
VTOFUD(vp)->flag |= FN_DIRECTIO;
fuse_io_invalbuf(vp, td);
} else {
if ((fuse_open_flags & FOPEN_KEEP_CACHE) == 0)
fuse_io_invalbuf(vp, td);
VTOFUD(vp)->flag &= ~FN_DIRECTIO;
}
if (vnode_vtype(vp) == VREG) {
/* XXXIP prevent getattr, by using cached node size */
vnode_create_vobject(vp, 0, td);
}
}
/*
* 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
ext2_update(struct vnode *vp, int waitfor)
{
struct m_ext2fs *fs;
struct buf *bp;
struct inode *ip;
int error;
ASSERT_VOP_ELOCKED(vp, "ext2_update");
ext2_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 = ip->i_e2fs;
if(fs->e2fs_ronly)
return (0);
if ((error = bread(ip->i_devvp,
fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
(int)fs->e2fs_bsize, NOCRED, &bp)) != 0) {
brelse(bp);
return (error);
}
ext2_i2ei(ip, (struct ext2fs_dinode *)((char *)bp->b_data +
EXT2_INODE_SIZE(fs) * ino_to_fsbo(fs, ip->i_number)));
if (waitfor && !DOINGASYNC(vp))
return (bwrite(bp));
else {
bdwrite(bp);
return (0);
}
}
/*
* The object must be locked.
*/
static void
vnode_pager_dealloc(vm_object_t object)
{
struct vnode *vp;
int refs;
vp = object->handle;
if (vp == NULL)
panic("vnode_pager_dealloc: pager already dealloced");
VM_OBJECT_ASSERT_WLOCKED(object);
vm_object_pip_wait(object, "vnpdea");
refs = object->ref_count;
object->handle = NULL;
object->type = OBJT_DEAD;
if (object->flags & OBJ_DISCONNECTWNT) {
vm_object_clear_flag(object, OBJ_DISCONNECTWNT);
wakeup(object);
}
ASSERT_VOP_ELOCKED(vp, "vnode_pager_dealloc");
if (object->un_pager.vnp.writemappings > 0) {
object->un_pager.vnp.writemappings = 0;
VOP_ADD_WRITECOUNT(vp, -1);
CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
__func__, vp, vp->v_writecount);
}
vp->v_object = NULL;
VOP_UNSET_TEXT(vp);
VM_OBJECT_WUNLOCK(object);
while (refs-- > 0)
vunref(vp);
VM_OBJECT_WLOCK(object);
}
static void
ncl_releasesillyrename(struct vnode *vp, struct thread *td)
{
struct nfsnode *np;
struct sillyrename *sp;
ASSERT_VOP_ELOCKED(vp, "releasesillyrename");
np = VTONFS(vp);
mtx_assert(&np->n_mtx, MA_OWNED);
if (vp->v_type != VDIR) {
sp = np->n_sillyrename;
np->n_sillyrename = NULL;
} else
sp = NULL;
if (sp != NULL) {
mtx_unlock(&np->n_mtx);
(void) ncl_vinvalbuf(vp, 0, td, 1);
/*
* Remove the silly file that was rename'd earlier
*/
ncl_removeit(sp, vp);
crfree(sp->s_cred);
sysmon_task_queue_sched(0, nfs_freesillyrename, sp);
mtx_lock(&np->n_mtx);
}
}
void
vnode_destroy_vobject(struct vnode *vp)
{
struct vm_object *obj;
obj = vp->v_object;
if (obj == NULL)
return;
ASSERT_VOP_ELOCKED(vp, "vnode_destroy_vobject");
VM_OBJECT_LOCK(obj);
if (obj->ref_count == 0) {
/*
* vclean() may be called twice. The first time
* removes the primary reference to the object,
* the second time goes one further and is a
* special-case to terminate the object.
*
* don't double-terminate the object
*/
if ((obj->flags & OBJ_DEAD) == 0)
vm_object_terminate(obj);
else
VM_OBJECT_UNLOCK(obj);
} else {
/*
* Woe to the process that tries to page now :-).
*/
vm_pager_deallocate(obj);
VM_OBJECT_UNLOCK(obj);
}
vp->v_object = NULL;
}
int
smbfs_smb_lookup(struct smbnode *dnp, const char *name, int nmlen,
struct smbfattr *fap, struct smb_cred *scred)
{
struct smbfs_fctx *ctx;
int error;
if (dnp == NULL || (dnp->n_ino == 2 && name == NULL)) {
bzero(fap, sizeof(*fap));
fap->fa_attr = SMB_FA_DIR;
fap->fa_ino = 2;
return 0;
}
MPASS(!(nmlen == 2 && name[0] == '.' && name[1] == '.'));
MPASS(!(nmlen == 1 && name[0] == '.'));
ASSERT_VOP_ELOCKED(dnp->n_vnode, "smbfs_smb_lookup");
error = smbfs_findopen(dnp, name, nmlen,
SMB_FA_SYSTEM | SMB_FA_HIDDEN | SMB_FA_DIR, scred, &ctx);
if (error)
return error;
ctx->f_flags |= SMBFS_RDD_FINDSINGLE;
error = smbfs_findnext(ctx, 1, scred);
if (error == 0) {
*fap = ctx->f_attr;
if (name == NULL)
fap->fa_ino = dnp->n_ino;
}
smbfs_findclose(ctx, scred);
return error;
}
void
fuse_internal_vnode_disappear(struct vnode *vp)
{
struct fuse_vnode_data *fvdat = VTOFUD(vp);
ASSERT_VOP_ELOCKED(vp, "fuse_internal_vnode_disappear");
fvdat->flag |= FN_REVOKED;
cache_purge(vp);
}
/*
* Flush and invalidate all dirty buffers. If another process is already
* doing the flush, just wait for completion.
*/
int
fuse_io_invalbuf(struct vnode *vp, struct thread *td)
{
struct fuse_vnode_data *fvdat = VTOFUD(vp);
int error = 0;
if (vp->v_iflag & VI_DOOMED)
return 0;
ASSERT_VOP_ELOCKED(vp, "fuse_io_invalbuf");
while (fvdat->flag & FN_FLUSHINPROG) {
struct proc *p = td->td_proc;
if (vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF)
return EIO;
fvdat->flag |= FN_FLUSHWANT;
tsleep(&fvdat->flag, PRIBIO + 2, "fusevinv", 2 * hz);
error = 0;
if (p != NULL) {
PROC_LOCK(p);
if (SIGNOTEMPTY(p->p_siglist) ||
SIGNOTEMPTY(td->td_siglist))
error = EINTR;
PROC_UNLOCK(p);
}
if (error == EINTR)
return EINTR;
}
fvdat->flag |= FN_FLUSHINPROG;
if (vp->v_bufobj.bo_object != NULL) {
VM_OBJECT_WLOCK(vp->v_bufobj.bo_object);
vm_object_page_clean(vp->v_bufobj.bo_object, 0, 0, OBJPC_SYNC);
VM_OBJECT_WUNLOCK(vp->v_bufobj.bo_object);
}
error = vinvalbuf(vp, V_SAVE, PCATCH, 0);
while (error) {
if (error == ERESTART || error == EINTR) {
fvdat->flag &= ~FN_FLUSHINPROG;
if (fvdat->flag & FN_FLUSHWANT) {
fvdat->flag &= ~FN_FLUSHWANT;
wakeup(&fvdat->flag);
}
return EINTR;
}
error = vinvalbuf(vp, V_SAVE, PCATCH, 0);
}
fvdat->flag &= ~FN_FLUSHINPROG;
if (fvdat->flag & FN_FLUSHWANT) {
fvdat->flag &= ~FN_FLUSHWANT;
wakeup(&fvdat->flag);
}
return (error);
}
int
fuse_vnode_savesize(struct vnode *vp, struct ucred *cred)
{
struct fuse_vnode_data *fvdat = VTOFUD(vp);
struct thread *td = curthread;
struct fuse_filehandle *fufh = NULL;
struct fuse_dispatcher fdi;
struct fuse_setattr_in *fsai;
int err = 0;
DEBUG("inode=%jd size=%jd\n", VTOI(vp), fvdat->filesize);
ASSERT_VOP_ELOCKED(vp, "fuse_io_extend");
if (fuse_isdeadfs(vp)) {
return EBADF;
}
if (vnode_vtype(vp) == VDIR) {
return EISDIR;
}
if (vfs_isrdonly(vnode_mount(vp))) {
return EROFS;
}
if (cred == NULL) {
cred = td->td_ucred;
}
fdisp_init(&fdi, sizeof(*fsai));
fdisp_make_vp(&fdi, FUSE_SETATTR, vp, td, cred);
fsai = fdi.indata;
fsai->valid = 0;
// Truncate to a new value.
fsai->size = fvdat->filesize;
fsai->valid |= FATTR_SIZE;
fuse_filehandle_getrw(vp, FUFH_WRONLY, &fufh);
if (fufh) {
fsai->fh = fufh->fh_id;
fsai->valid |= FATTR_FH;
}
err = fdisp_wait_answ(&fdi);
fdisp_destroy(&fdi);
if (err == 0)
fvdat->flag &= ~FN_SIZECHANGE;
fuse_invalidate_attr(vp);
return err;
}
int
fuse_vnode_setsize(struct vnode *vp, struct ucred *cred, off_t newsize)
{
struct fuse_vnode_data *fvdat = VTOFUD(vp);
off_t oldsize;
int err = 0;
ASSERT_VOP_ELOCKED(vp, "fuse_vnode_setsize");
oldsize = fvdat->filesize;
fvdat->filesize = newsize;
fvdat->flag |= FN_SIZECHANGE;
if (newsize < oldsize) {
err = vtruncbuf(vp, cred, newsize, fuse_iosize(vp));
}
vnode_pager_setsize(vp, newsize);
return err;
}
void
vnode_destroy_vobject(struct vnode *vp)
{
struct vm_object *obj;
obj = vp->v_object;
if (obj == NULL)
return;
ASSERT_VOP_ELOCKED(vp, "vnode_destroy_vobject");
VM_OBJECT_WLOCK(obj);
umtx_shm_object_terminated(obj);
if (obj->ref_count == 0) {
/*
* don't double-terminate the object
*/
if ((obj->flags & OBJ_DEAD) == 0) {
vm_object_terminate(obj);
} else {
/*
* Waiters were already handled during object
* termination. The exclusive vnode lock hopefully
* prevented new waiters from referencing the dying
* object.
*/
KASSERT((obj->flags & OBJ_DISCONNECTWNT) == 0,
("OBJ_DISCONNECTWNT set obj %p flags %x",
obj, obj->flags));
vp->v_object = NULL;
VM_OBJECT_WUNLOCK(obj);
}
} else {
/*
* Woe to the process that tries to page now :-).
*/
vm_pager_deallocate(obj);
VM_OBJECT_WUNLOCK(obj);
}
KASSERT(vp->v_object == NULL, ("vp %p obj %p", vp, vp->v_object));
}
int
fuse_vnode_setsize(struct vnode *vp, struct ucred *cred, off_t newsize)
{
struct fuse_vnode_data *fvdat = VTOFUD(vp);
off_t oldsize;
int err = 0;
FS_DEBUG("inode=%ju oldsize=%ju newsize=%ju\n",
(uintmax_t)VTOI(vp), (uintmax_t)fvdat->filesize,
(uintmax_t)newsize);
ASSERT_VOP_ELOCKED(vp, "fuse_vnode_setsize");
oldsize = fvdat->filesize;
fvdat->filesize = newsize;
fvdat->flag |= FN_SIZECHANGE;
if (newsize < oldsize) {
err = vtruncbuf(vp, cred, newsize, fuse_iosize(vp));
}
vnode_pager_setsize(vp, newsize);
fuse_invalidate_attr(vp);
return err;
}
static int
fuse_vnode_alloc(struct mount *mp,
struct thread *td,
uint64_t nodeid,
enum vtype vtyp,
struct vnode **vpp)
{
const int lkflags = LK_EXCLUSIVE | LK_RETRY;
struct fuse_vnode_data *fvdat;
struct vnode *vp2;
int err = 0;
DEBUG("been asked for vno #%ju\n", (uintmax_t)nodeid);
if (vtyp == VNON) {
return EINVAL;
}
*vpp = NULL;
err = vfs_hash_get(mp, fuse_vnode_hash(nodeid), lkflags, td, vpp,
fuse_vnode_cmp, &nodeid);
if (err)
return (err);
if (*vpp) {
MPASS((*vpp)->v_type == vtyp && (*vpp)->v_data != NULL);
DEBUG("vnode taken from hash\n");
return (0);
}
fvdat = malloc(sizeof(*fvdat), M_FUSEVN, M_WAITOK | M_ZERO);
err = getnewvnode("fuse", mp, &fuse_vnops, vpp);
if (err) {
free(fvdat, M_FUSEVN);
return (err);
}
vn_lock(*vpp, lkflags);
err = insmntque(*vpp, mp);
ASSERT_VOP_ELOCKED(*vpp, "fuse_vnode_alloc");
if (err) {
VOP_UNLOCK(*vpp, 0);
free(fvdat, M_FUSEVN);
*vpp = NULL;
return (err);
}
fuse_vnode_init(*vpp, fvdat, nodeid, vtyp);
err = vfs_hash_insert(*vpp, fuse_vnode_hash(nodeid), lkflags,
td, &vp2, fuse_vnode_cmp, &nodeid);
if (err) {
VOP_UNLOCK(*vpp, 0);
fuse_vnode_destroy(*vpp);
*vpp = NULL;
return (err);
}
/*
* XXXIP: Prevent silent vnode reuse. It may happen because several fuse
* filesystems ignore inode numbers
*/
KASSERT(vp2 == NULL,
("vfs hash collision for node #%ju\n", (uintmax_t)nodeid));
ASSERT_VOP_ELOCKED(*vpp, "fuse_vnode_alloc");
return (0);
}
/*
* Update the disk quota in the quota file.
*/
static int
dqsync(struct vnode *vp, struct dquot *dq)
{
uint8_t buf[sizeof(struct dqblk64)];
off_t base, recsize;
struct vnode *dqvp;
struct iovec aiov;
struct uio auio;
int error;
struct mount *mp;
struct ufsmount *ump;
#ifdef DEBUG_VFS_LOCKS
if (vp != NULL)
ASSERT_VOP_ELOCKED(vp, "dqsync");
#endif
mp = NULL;
error = 0;
if (dq == NODQUOT)
panic("dqsync: dquot");
if ((ump = dq->dq_ump) == NULL)
return (0);
UFS_LOCK(ump);
if ((dqvp = ump->um_quotas[dq->dq_type]) == NULLVP) {
if (vp == NULL) {
UFS_UNLOCK(ump);
return (0);
} else
panic("dqsync: file");
}
vref(dqvp);
UFS_UNLOCK(ump);
DQI_LOCK(dq);
if ((dq->dq_flags & DQ_MOD) == 0) {
DQI_UNLOCK(dq);
vrele(dqvp);
return (0);
}
DQI_UNLOCK(dq);
(void) vn_start_secondary_write(dqvp, &mp, V_WAIT);
if (vp != dqvp)
vn_lock(dqvp, LK_EXCLUSIVE | LK_RETRY);
DQI_LOCK(dq);
DQI_WAIT(dq, PINOD+2, "dqsync");
if ((dq->dq_flags & DQ_MOD) == 0)
goto out;
dq->dq_flags |= DQ_LOCK;
DQI_UNLOCK(dq);
/*
* Write the quota record to the quota file, performing any
* necessary conversions. See dqget() for additional details.
*/
if (ump->um_qflags[dq->dq_type] & QTF_64BIT) {
dq_dqb64(dq, (struct dqblk64 *)buf);
recsize = sizeof(struct dqblk64);
base = sizeof(struct dqhdr64);
} else {
dq_dqb32(dq, (struct dqblk32 *)buf);
recsize = sizeof(struct dqblk32);
base = 0;
}
auio.uio_iov = &aiov;
auio.uio_iovcnt = 1;
aiov.iov_base = buf;
aiov.iov_len = recsize;
auio.uio_resid = recsize;
auio.uio_offset = base + dq->dq_id * recsize;
auio.uio_segflg = UIO_SYSSPACE;
auio.uio_rw = UIO_WRITE;
auio.uio_td = (struct thread *)0;
error = VOP_WRITE(dqvp, &auio, 0, dq->dq_ump->um_cred[dq->dq_type]);
if (auio.uio_resid && error == 0)
error = EIO;
DQI_LOCK(dq);
DQI_WAKEUP(dq);
dq->dq_flags &= ~DQ_MOD;
out:
DQI_UNLOCK(dq);
if (vp != dqvp)
vput(dqvp);
else
vrele(dqvp);
vn_finished_secondary_write(mp);
return (error);
}
static int
tmpfs_rename(struct vop_rename_args *v)
{
struct vnode *fdvp = v->a_fdvp;
struct vnode *fvp = v->a_fvp;
struct componentname *fcnp = v->a_fcnp;
struct vnode *tdvp = v->a_tdvp;
struct vnode *tvp = v->a_tvp;
struct componentname *tcnp = v->a_tcnp;
struct mount *mp = NULL;
char *newname;
int error;
struct tmpfs_dirent *de;
struct tmpfs_mount *tmp;
struct tmpfs_node *fdnode;
struct tmpfs_node *fnode;
struct tmpfs_node *tnode;
struct tmpfs_node *tdnode;
MPASS(VOP_ISLOCKED(tdvp));
MPASS(IMPLIES(tvp != NULL, VOP_ISLOCKED(tvp)));
MPASS(fcnp->cn_flags & HASBUF);
MPASS(tcnp->cn_flags & HASBUF);
/* Disallow cross-device renames.
* XXX Why isn't this done by the caller? */
if (fvp->v_mount != tdvp->v_mount ||
(tvp != NULL && fvp->v_mount != tvp->v_mount)) {
error = EXDEV;
goto out;
}
/* If source and target are the same file, there is nothing to do. */
if (fvp == tvp) {
error = 0;
goto out;
}
/* If we need to move the directory between entries, lock the
* source so that we can safely operate on it. */
if (fdvp != tdvp && fdvp != tvp) {
if (vn_lock(fdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
mp = tdvp->v_mount;
error = vfs_busy(mp, 0);
if (error != 0) {
mp = NULL;
goto out;
}
error = tmpfs_rename_relock(fdvp, &fvp, tdvp, &tvp,
fcnp, tcnp);
if (error != 0) {
vfs_unbusy(mp);
return (error);
}
ASSERT_VOP_ELOCKED(fdvp,
"tmpfs_rename: fdvp not locked");
ASSERT_VOP_ELOCKED(tdvp,
"tmpfs_rename: tdvp not locked");
if (tvp != NULL)
ASSERT_VOP_ELOCKED(tvp,
"tmpfs_rename: tvp not locked");
if (fvp == tvp) {
error = 0;
goto out_locked;
}
}
}
tmp = VFS_TO_TMPFS(tdvp->v_mount);
tdnode = VP_TO_TMPFS_DIR(tdvp);
tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp);
fdnode = VP_TO_TMPFS_DIR(fdvp);
fnode = VP_TO_TMPFS_NODE(fvp);
de = tmpfs_dir_lookup(fdnode, fnode, fcnp);
/* Entry can disappear before we lock fdvp,
* also avoid manipulating '.' and '..' entries. */
if (de == NULL) {
if ((fcnp->cn_flags & ISDOTDOT) != 0 ||
(fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.'))
error = EINVAL;
else
error = ENOENT;
goto out_locked;
}
MPASS(de->td_node == fnode);
/* If re-naming a directory to another preexisting directory
* ensure that the target directory is empty so that its
* removal causes no side effects.
* Kern_rename guarantees the destination to be a directory
* if the source is one. */
if (tvp != NULL) {
MPASS(tnode != NULL);
if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
(tdnode->tn_flags & (APPEND | IMMUTABLE))) {
error = EPERM;
goto out_locked;
}
if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
if (tnode->tn_size > 0) {
error = ENOTEMPTY;
goto out_locked;
}
} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
error = ENOTDIR;
goto out_locked;
} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
error = EISDIR;
goto out_locked;
} else {
MPASS(fnode->tn_type != VDIR &&
tnode->tn_type != VDIR);
}
}
if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))
|| (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
error = EPERM;
goto out_locked;
}
/* Ensure that we have enough memory to hold the new name, if it
* has to be changed. */
if (fcnp->cn_namelen != tcnp->cn_namelen ||
bcmp(fcnp->cn_nameptr, tcnp->cn_nameptr, fcnp->cn_namelen) != 0) {
newname = malloc(tcnp->cn_namelen, M_TMPFSNAME, M_WAITOK);
} else
newname = NULL;
/* If the node is being moved to another directory, we have to do
* the move. */
if (fdnode != tdnode) {
/* In case we are moving a directory, we have to adjust its
* parent to point to the new parent. */
if (de->td_node->tn_type == VDIR) {
struct tmpfs_node *n;
/* Ensure the target directory is not a child of the
* directory being moved. Otherwise, we'd end up
* with stale nodes. */
n = tdnode;
/* TMPFS_LOCK garanties that no nodes are freed while
* traversing the list. Nodes can only be marked as
* removed: tn_parent == NULL. */
TMPFS_LOCK(tmp);
TMPFS_NODE_LOCK(n);
while (n != n->tn_dir.tn_parent) {
struct tmpfs_node *parent;
if (n == fnode) {
TMPFS_NODE_UNLOCK(n);
TMPFS_UNLOCK(tmp);
error = EINVAL;
if (newname != NULL)
free(newname, M_TMPFSNAME);
goto out_locked;
}
parent = n->tn_dir.tn_parent;
TMPFS_NODE_UNLOCK(n);
if (parent == NULL) {
n = NULL;
break;
}
TMPFS_NODE_LOCK(parent);
if (parent->tn_dir.tn_parent == NULL) {
TMPFS_NODE_UNLOCK(parent);
n = NULL;
break;
}
n = parent;
}
TMPFS_UNLOCK(tmp);
if (n == NULL) {
error = EINVAL;
if (newname != NULL)
free(newname, M_TMPFSNAME);
goto out_locked;
}
TMPFS_NODE_UNLOCK(n);
/* Adjust the parent pointer. */
TMPFS_VALIDATE_DIR(fnode);
TMPFS_NODE_LOCK(de->td_node);
de->td_node->tn_dir.tn_parent = tdnode;
TMPFS_NODE_UNLOCK(de->td_node);
/* As a result of changing the target of the '..'
* entry, the link count of the source and target
* directories has to be adjusted. */
TMPFS_NODE_LOCK(tdnode);
TMPFS_ASSERT_LOCKED(tdnode);
tdnode->tn_links++;
TMPFS_NODE_UNLOCK(tdnode);
TMPFS_NODE_LOCK(fdnode);
TMPFS_ASSERT_LOCKED(fdnode);
fdnode->tn_links--;
TMPFS_NODE_UNLOCK(fdnode);
}
}
/* Do the move: just remove the entry from the source directory
* and insert it into the target one. */
tmpfs_dir_detach(fdvp, de);
if (fcnp->cn_flags & DOWHITEOUT)
tmpfs_dir_whiteout_add(fdvp, fcnp);
if (tcnp->cn_flags & ISWHITEOUT)
tmpfs_dir_whiteout_remove(tdvp, tcnp);
/* If the name has changed, we need to make it effective by changing
* it in the directory entry. */
if (newname != NULL) {
MPASS(tcnp->cn_namelen <= MAXNAMLEN);
free(de->ud.td_name, M_TMPFSNAME);
de->ud.td_name = newname;
tmpfs_dirent_init(de, tcnp->cn_nameptr, tcnp->cn_namelen);
fnode->tn_status |= TMPFS_NODE_CHANGED;
tdnode->tn_status |= TMPFS_NODE_MODIFIED;
}
/* If we are overwriting an entry, we have to remove the old one
* from the target directory. */
if (tvp != NULL) {
struct tmpfs_dirent *tde;
/* Remove the old entry from the target directory. */
tde = tmpfs_dir_lookup(tdnode, tnode, tcnp);
tmpfs_dir_detach(tdvp, tde);
/* Free the directory entry we just deleted. Note that the
* node referred by it will not be removed until the vnode is
* really reclaimed. */
tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), tde);
}
tmpfs_dir_attach(tdvp, de);
cache_purge(fvp);
if (tvp != NULL)
cache_purge(tvp);
cache_purge_negative(tdvp);
error = 0;
out_locked:
if (fdvp != tdvp && fdvp != tvp)
VOP_UNLOCK(fdvp, 0);
out:
/* Release target nodes. */
/* XXX: I don't understand when tdvp can be the same as tvp, but
* other code takes care of this... */
if (tdvp == tvp)
vrele(tdvp);
else
vput(tdvp);
if (tvp != NULL)
vput(tvp);
/* Release source nodes. */
vrele(fdvp);
vrele(fvp);
if (mp != NULL)
vfs_unbusy(mp);
return error;
}
/*
struct vnop_create_args {
struct vnode *a_dvp;
struct vnode **a_vpp;
struct componentname *a_cnp;
struct vattr *a_vap;
};
*/
static int
fuse_vnop_create(struct vop_create_args *ap)
{
struct vnode *dvp = ap->a_dvp;
struct vnode **vpp = ap->a_vpp;
struct componentname *cnp = ap->a_cnp;
struct vattr *vap = ap->a_vap;
struct thread *td = cnp->cn_thread;
struct ucred *cred = cnp->cn_cred;
struct fuse_open_in *foi;
struct fuse_entry_out *feo;
struct fuse_dispatcher fdi;
struct fuse_dispatcher *fdip = &fdi;
int err;
struct mount *mp = vnode_mount(dvp);
uint64_t parentnid = VTOFUD(dvp)->nid;
mode_t mode = MAKEIMODE(vap->va_type, vap->va_mode);
uint64_t x_fh_id;
uint32_t x_open_flags;
fuse_trace_printf_vnop();
if (fuse_isdeadfs(dvp)) {
return ENXIO;
}
bzero(&fdi, sizeof(fdi));
/* XXX: Will we ever want devices ? */
if ((vap->va_type != VREG)) {
MPASS(vap->va_type != VFIFO);
goto bringup;
}
debug_printf("parent nid = %ju, mode = %x\n", (uintmax_t)parentnid,
mode);
fdisp_init(fdip, sizeof(*foi) + cnp->cn_namelen + 1);
if (!fsess_isimpl(mp, FUSE_CREATE)) {
debug_printf("eh, daemon doesn't implement create?\n");
return (EINVAL);
}
fdisp_make(fdip, FUSE_CREATE, vnode_mount(dvp), parentnid, td, cred);
foi = fdip->indata;
foi->mode = mode;
foi->flags = O_CREAT | O_RDWR;
memcpy((char *)fdip->indata + sizeof(*foi), cnp->cn_nameptr,
cnp->cn_namelen);
((char *)fdip->indata)[sizeof(*foi) + cnp->cn_namelen] = '\0';
err = fdisp_wait_answ(fdip);
if (err) {
if (err == ENOSYS)
fsess_set_notimpl(mp, FUSE_CREATE);
debug_printf("create: got err=%d from daemon\n", err);
goto out;
}
bringup:
feo = fdip->answ;
if ((err = fuse_internal_checkentry(feo, VREG))) {
goto out;
}
err = fuse_vnode_get(mp, feo->nodeid, dvp, vpp, cnp, VREG);
if (err) {
struct fuse_release_in *fri;
uint64_t nodeid = feo->nodeid;
uint64_t fh_id = ((struct fuse_open_out *)(feo + 1))->fh;
fdisp_init(fdip, sizeof(*fri));
fdisp_make(fdip, FUSE_RELEASE, mp, nodeid, td, cred);
fri = fdip->indata;
fri->fh = fh_id;
fri->flags = OFLAGS(mode);
fuse_insert_callback(fdip->tick, fuse_internal_forget_callback);
fuse_insert_message(fdip->tick);
return err;
}
ASSERT_VOP_ELOCKED(*vpp, "fuse_vnop_create");
fdip->answ = feo + 1;
x_fh_id = ((struct fuse_open_out *)(feo + 1))->fh;
x_open_flags = ((struct fuse_open_out *)(feo + 1))->open_flags;
fuse_filehandle_init(*vpp, FUFH_RDWR, NULL, x_fh_id);
fuse_vnode_open(*vpp, x_open_flags, td);
cache_purge_negative(dvp);
out:
fdisp_destroy(fdip);
return err;
}
/*
* 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);
}
/*
* Obtain a dquot structure for the specified identifier and quota file
* reading the information from the file if necessary.
*/
static int
dqget(struct vnode *vp, u_long id, struct ufsmount *ump, int type,
struct dquot **dqp)
{
uint8_t buf[sizeof(struct dqblk64)];
off_t base, recsize;
struct dquot *dq, *dq1;
struct dqhash *dqh;
struct vnode *dqvp;
struct iovec aiov;
struct uio auio;
int dqvplocked, error;
#ifdef DEBUG_VFS_LOCKS
if (vp != NULLVP)
ASSERT_VOP_ELOCKED(vp, "dqget");
#endif
if (vp != NULLVP && *dqp != NODQUOT) {
return (0);
}
/* XXX: Disallow negative id values to prevent the
* creation of 100GB+ quota data files.
*/
if ((int)id < 0)
return (EINVAL);
UFS_LOCK(ump);
dqvp = ump->um_quotas[type];
if (dqvp == NULLVP || (ump->um_qflags[type] & QTF_CLOSING)) {
*dqp = NODQUOT;
UFS_UNLOCK(ump);
return (EINVAL);
}
vref(dqvp);
UFS_UNLOCK(ump);
error = 0;
dqvplocked = 0;
/*
* Check the cache first.
*/
dqh = DQHASH(dqvp, id);
DQH_LOCK();
dq = dqhashfind(dqh, id, dqvp);
if (dq != NULL) {
DQH_UNLOCK();
hfound:
DQI_LOCK(dq);
DQI_WAIT(dq, PINOD+1, "dqget");
DQI_UNLOCK(dq);
if (dq->dq_ump == NULL) {
dqrele(vp, dq);
dq = NODQUOT;
error = EIO;
}
*dqp = dq;
if (dqvplocked)
vput(dqvp);
else
vrele(dqvp);
return (error);
}
/*
* Quota vnode lock is before DQ_LOCK. Acquire dqvp lock there
* since new dq will appear on the hash chain DQ_LOCKed.
*/
if (vp != dqvp) {
DQH_UNLOCK();
vn_lock(dqvp, LK_SHARED | LK_RETRY);
dqvplocked = 1;
DQH_LOCK();
/*
* Recheck the cache after sleep for quota vnode lock.
*/
dq = dqhashfind(dqh, id, dqvp);
if (dq != NULL) {
DQH_UNLOCK();
goto hfound;
}
}
/*
* Not in cache, allocate a new one or take it from the
* free list.
*/
if (TAILQ_FIRST(&dqfreelist) == NODQUOT &&
numdquot < MAXQUOTAS * desiredvnodes)
desireddquot += DQUOTINC;
if (numdquot < desireddquot) {
numdquot++;
DQH_UNLOCK();
dq1 = malloc(sizeof *dq1, M_DQUOT, M_WAITOK | M_ZERO);
mtx_init(&dq1->dq_lock, "dqlock", NULL, MTX_DEF);
DQH_LOCK();
/*
* Recheck the cache after sleep for memory.
*/
dq = dqhashfind(dqh, id, dqvp);
if (dq != NULL) {
numdquot--;
DQH_UNLOCK();
mtx_destroy(&dq1->dq_lock);
free(dq1, M_DQUOT);
goto hfound;
}
dq = dq1;
} else {
if ((dq = TAILQ_FIRST(&dqfreelist)) == NULL) {
DQH_UNLOCK();
tablefull("dquot");
*dqp = NODQUOT;
if (dqvplocked)
vput(dqvp);
else
vrele(dqvp);
return (EUSERS);
}
if (dq->dq_cnt || (dq->dq_flags & DQ_MOD))
panic("dqget: free dquot isn't %p", dq);
TAILQ_REMOVE(&dqfreelist, dq, dq_freelist);
if (dq->dq_ump != NULL)
LIST_REMOVE(dq, dq_hash);
}
/*
* Dq is put into hash already locked to prevent parallel
* usage while it is being read from file.
*/
dq->dq_flags = DQ_LOCK;
dq->dq_id = id;
dq->dq_type = type;
dq->dq_ump = ump;
LIST_INSERT_HEAD(dqh, dq, dq_hash);
DQREF(dq);
DQH_UNLOCK();
/*
* Read the requested quota record from the quota file, performing
* any necessary conversions.
*/
if (ump->um_qflags[type] & QTF_64BIT) {
recsize = sizeof(struct dqblk64);
base = sizeof(struct dqhdr64);
} else {
recsize = sizeof(struct dqblk32);
base = 0;
}
auio.uio_iov = &aiov;
auio.uio_iovcnt = 1;
aiov.iov_base = buf;
aiov.iov_len = recsize;
auio.uio_resid = recsize;
auio.uio_offset = base + id * recsize;
auio.uio_segflg = UIO_SYSSPACE;
auio.uio_rw = UIO_READ;
auio.uio_td = (struct thread *)0;
error = VOP_READ(dqvp, &auio, 0, ump->um_cred[type]);
if (auio.uio_resid == recsize && error == 0) {
bzero(&dq->dq_dqb, sizeof(dq->dq_dqb));
} else {
if (ump->um_qflags[type] & QTF_64BIT)
dqb64_dq((struct dqblk64 *)buf, dq);
else
dqb32_dq((struct dqblk32 *)buf, dq);
}
if (dqvplocked)
vput(dqvp);
else
vrele(dqvp);
/*
* I/O error in reading quota file, release
* quota structure and reflect problem to caller.
*/
if (error) {
DQH_LOCK();
dq->dq_ump = NULL;
LIST_REMOVE(dq, dq_hash);
DQH_UNLOCK();
DQI_LOCK(dq);
if (dq->dq_flags & DQ_WANT)
wakeup(dq);
dq->dq_flags = 0;
DQI_UNLOCK(dq);
dqrele(vp, dq);
*dqp = NODQUOT;
return (error);
}
DQI_LOCK(dq);
/*
* Check for no limit to enforce.
* Initialize time values if necessary.
*/
if (dq->dq_isoftlimit == 0 && dq->dq_bsoftlimit == 0 &&
dq->dq_ihardlimit == 0 && dq->dq_bhardlimit == 0)
dq->dq_flags |= DQ_FAKE;
if (dq->dq_id != 0) {
if (dq->dq_btime == 0) {
dq->dq_btime = time_second + ump->um_btime[type];
if (dq->dq_bsoftlimit &&
dq->dq_curblocks >= dq->dq_bsoftlimit)
dq->dq_flags |= DQ_MOD;
}
if (dq->dq_itime == 0) {
dq->dq_itime = time_second + ump->um_itime[type];
if (dq->dq_isoftlimit &&
dq->dq_curinodes >= dq->dq_isoftlimit)
dq->dq_flags |= DQ_MOD;
}
}
DQI_WAKEUP(dq);
DQI_UNLOCK(dq);
*dqp = dq;
return (0);
}
