/*
* Add a ref to a vnode's existing VM object, return the object or
* NULL if the vnode did not have one. This does not create the
* object (we can't since we don't know what the proper blocksize/boff
* is to match the VFS's use of the buffer cache).
*/
vm_object_t
vnode_pager_reference(struct vnode *vp)
{
vm_object_t object;
/*
* Prevent race condition when allocating the object. This
* can happen with NFS vnodes since the nfsnode isn't locked.
*
* Serialize potential vnode/object teardowns and interlocks
*/
lwkt_gettoken(&vp->v_token);
while (vp->v_flag & VOLOCK) {
vsetflags(vp, VOWANT);
tsleep(vp, 0, "vnpobj", 0);
}
vsetflags(vp, VOLOCK);
lwkt_reltoken(&vp->v_token);
/*
* Prevent race conditions against deallocation of the VM
* object.
*/
while ((object = vp->v_object) != NULL) {
vm_object_hold(object);
if ((object->flags & OBJ_DEAD) == 0)
break;
vm_object_dead_sleep(object, "vadead");
vm_object_drop(object);
}
/*
* The object is expected to exist, the caller will handle
* NULL returns if it does not.
*/
if (object) {
object->ref_count++;
vref(vp);
}
lwkt_gettoken(&vp->v_token);
vclrflags(vp, VOLOCK);
if (vp->v_flag & VOWANT) {
vclrflags(vp, VOWANT);
wakeup(vp);
}
lwkt_reltoken(&vp->v_token);
if (object)
vm_object_drop(object);
return (object);
}
/*
* This opens /dev/tty. Because multiple opens of /dev/tty only
* generate a single open to the actual tty, the file modes are
* locked to FREAD|FWRITE.
*/
static int
cttyopen(struct dev_open_args *ap)
{
struct proc *p = curproc;
struct vnode *ttyvp;
int error;
KKASSERT(p);
retry:
if ((ttyvp = cttyvp(p)) == NULL)
return (ENXIO);
if (ttyvp->v_flag & VCTTYISOPEN)
return (0);
/*
* Messy interlock, don't let the vnode go away while we try to
* lock it and check for race after we might have blocked.
*/
vhold(ttyvp);
vn_lock(ttyvp, LK_EXCLUSIVE | LK_RETRY);
if (ttyvp != cttyvp(p) || (ttyvp->v_flag & VCTTYISOPEN)) {
kprintf("Warning: cttyopen: race avoided\n");
vn_unlock(ttyvp);
vdrop(ttyvp);
goto retry;
}
vsetflags(ttyvp, VCTTYISOPEN);
error = VOP_OPEN(ttyvp, FREAD|FWRITE, ap->a_cred, NULL);
if (error)
vclrflags(ttyvp, VCTTYISOPEN);
vn_unlock(ttyvp);
vdrop(ttyvp);
return(error);
}
/*
* NOTE: VAGE is always cleared when calling VOP_OPEN().
*/
int
vop_open(struct vop_ops *ops, struct vnode *vp, int mode, struct ucred *cred,
struct file *fp)
{
struct vop_open_args ap;
VFS_MPLOCK_DECLARE;
int error;
/*
* Decrement 3-2-1-0. Does not decrement beyond 0
*/
if (vp->v_flag & VAGE0) {
vclrflags(vp, VAGE0);
} else if (vp->v_flag & VAGE1) {
vclrflags(vp, VAGE1);
vsetflags(vp, VAGE0);
}
ap.a_head.a_desc = &vop_open_desc;
ap.a_head.a_ops = ops;
ap.a_vp = vp;
ap.a_fp = fp;
ap.a_mode = mode;
ap.a_cred = cred;
VFS_MPLOCK1(vp->v_mount);
DO_OPS(ops, error, &ap, vop_open);
VFS_MPUNLOCK(vp->v_mount);
return(error);
}
/*
* This opens /dev/tty. Because multiple opens of /dev/tty only
* generate a single open to the actual tty, the file modes are
* locked to FREAD|FWRITE.
*/
static int
cttyopen(struct dev_open_args *ap)
{
struct proc *p = curproc;
struct vnode *ttyvp;
int error;
KKASSERT(p);
retry:
if ((ttyvp = cttyvp(p)) == NULL)
return (ENXIO);
if (ttyvp->v_flag & VCTTYISOPEN)
return (0);
/*
* Messy interlock, don't let the vnode go away while we try to
* lock it and check for race after we might have blocked.
*
* WARNING! The device open (devfs_spec_open()) temporarily
* releases the vnode lock on ttyvp when issuing the
* dev_dopen(), which means that the VCTTYISOPEn flag
* can race during the VOP_OPEN().
*
* If something does race we have to undo our potentially
* extra open.
*/
vhold(ttyvp);
vn_lock(ttyvp, LK_EXCLUSIVE | LK_RETRY);
if (ttyvp != cttyvp(p) || (ttyvp->v_flag & VCTTYISOPEN)) {
kprintf("Warning: cttyopen: race-1 avoided\n");
vn_unlock(ttyvp);
vdrop(ttyvp);
goto retry;
}
error = VOP_OPEN(ttyvp, FREAD|FWRITE, ap->a_cred, NULL);
/*
* Race against ctty close or change. This case has been validated
* and occurs every so often during synth builds.
*/
if (ttyvp != cttyvp(p) || (ttyvp->v_flag & VCTTYISOPEN)) {
if (error == 0)
VOP_CLOSE(ttyvp, FREAD|FWRITE, NULL);
vn_unlock(ttyvp);
vdrop(ttyvp);
goto retry;
}
if (error == 0)
vsetflags(ttyvp, VCTTYISOPEN);
vn_unlock(ttyvp);
vdrop(ttyvp);
return(error);
}
void
vsetobjdirty(struct vnode *vp)
{
struct syncer_ctx *ctx;
if ((vp->v_flag & VOBJDIRTY) == 0) {
ctx = vp->v_mount->mnt_syncer_ctx;
vsetflags(vp, VOBJDIRTY);
lwkt_gettoken(&ctx->sc_token);
if ((vp->v_flag & VONWORKLST) == 0)
vn_syncer_add(vp, syncdelay);
lwkt_reltoken(&ctx->sc_token);
}
}
/*
* Add an item to the syncer work queue.
*
* WARNING: Cannot get vp->v_token here if not already held, we must
* depend on the syncer_token (which might already be held by
* the caller) to protect v_synclist and VONWORKLST.
*
* MPSAFE
*/
void
vn_syncer_add(struct vnode *vp, int delay)
{
struct syncer_ctx *ctx;
int slot;
ctx = vn_get_syncer(vp);
lwkt_gettoken(&ctx->sc_token);
if (vp->v_flag & VONWORKLST)
LIST_REMOVE(vp, v_synclist);
if (delay > syncer_maxdelay - 2)
delay = syncer_maxdelay - 2;
slot = (ctx->syncer_delayno + delay) & ctx->syncer_mask;
LIST_INSERT_HEAD(&ctx->syncer_workitem_pending[slot], vp, v_synclist);
vsetflags(vp, VONWORKLST);
lwkt_reltoken(&ctx->sc_token);
}
/*
* Return locked root vnode of a filesystem
*/
static int
smbfs_root(struct mount *mp, struct vnode **vpp)
{
struct thread *td = curthread; /* XXX */
struct smbmount *smp = VFSTOSMBFS(mp);
struct vnode *vp;
struct smbnode *np;
struct smbfattr fattr;
struct ucred *cred;
struct smb_cred scred;
int error;
if (smp == NULL) {
SMBERROR("smp == NULL (bug in umount)\n");
return EINVAL;
}
if (smp->sm_root) {
*vpp = SMBTOV(smp->sm_root);
return vget(*vpp, LK_EXCLUSIVE | LK_RETRY);
}
if (td->td_proc)
cred = td->td_proc->p_ucred;
else
cred = proc0.p_ucred;
smb_makescred(&scred, td, cred);
error = smbfs_smb_lookup(NULL, NULL, 0, &fattr, &scred);
if (error)
return error;
error = smbfs_nget(mp, NULL, "TheRooT", 7, &fattr, &vp);
if (error)
return error;
vsetflags(vp, VROOT);
np = VTOSMB(vp);
smp->sm_root = np;
*vpp = vp;
return 0;
}
/*
* Mount the per-process file descriptors (/dev/fd)
*/
static int
fdesc_mount(struct mount *mp, char *path, caddr_t data, struct ucred *cred)
{
int error = 0;
struct fdescmount *fmp;
struct vnode *rvp;
if (path == NULL)
panic("fdesc_mount: cannot mount as root");
/*
* Update is a no-op
*/
if (mp->mnt_flag & MNT_UPDATE)
return (EOPNOTSUPP);
vfs_add_vnodeops(mp, &fdesc_vnode_vops, &mp->mnt_vn_norm_ops);
error = fdesc_allocvp(Froot, FD_ROOT, mp, &rvp);
if (error)
return (error);
fmp = kmalloc(sizeof(struct fdescmount), M_FDESCMNT, M_WAITOK); /* XXX */
rvp->v_type = VDIR;
vsetflags(rvp, VROOT);
fmp->f_root = rvp;
/* XXX -- don't mark as local to work around fts() problems */
/*mp->mnt_flag |= MNT_LOCAL;*/
mp->mnt_data = (qaddr_t) fmp;
vfs_getnewfsid(mp);
bzero(mp->mnt_stat.f_mntfromname, MNAMELEN);
bcopy("fdesc", mp->mnt_stat.f_mntfromname, sizeof("fdesc"));
fdesc_statfs(mp, &mp->mnt_stat, cred);
return (0);
}
/*
* Add an item to the syncer work queue.
*
* WARNING: Cannot get vp->v_token here if not already held, we must
* depend on the syncer_token (which might already be held by
* the caller) to protect v_synclist and VONWORKLST.
*
* MPSAFE
*/
void
vn_syncer_add(struct vnode *vp, int delay)
{
struct syncer_ctx *ctx;
int slot;
ctx = vp->v_mount->mnt_syncer_ctx;
lwkt_gettoken(&ctx->sc_token);
if (vp->v_flag & VONWORKLST)
LIST_REMOVE(vp, v_synclist);
if (delay <= 0) {
slot = -delay & ctx->syncer_mask;
} else {
if (delay > SYNCER_MAXDELAY - 2)
delay = SYNCER_MAXDELAY - 2;
slot = (ctx->syncer_delayno + delay) & ctx->syncer_mask;
}
LIST_INSERT_HEAD(&ctx->syncer_workitem_pending[slot], vp, v_synclist);
vsetflags(vp, VONWORKLST);
lwkt_reltoken(&ctx->sc_token);
}
static int
ckpt_thaw_proc(struct lwp *lp, struct file *fp)
{
struct proc *p = lp->lwp_proc;
Elf_Phdr *phdr = NULL;
Elf_Ehdr *ehdr = NULL;
int error;
size_t nbyte;
TRACE_ENTER;
ehdr = kmalloc(sizeof(Elf_Ehdr), M_TEMP, M_ZERO | M_WAITOK);
if ((error = elf_gethdr(fp, ehdr)) != 0)
goto done;
nbyte = sizeof(Elf_Phdr) * ehdr->e_phnum;
phdr = kmalloc(nbyte, M_TEMP, M_WAITOK);
/* fetch description of program writable mappings */
if ((error = elf_getphdrs(fp, phdr, nbyte)) != 0)
goto done;
/* fetch notes section containing register state */
if ((error = elf_getnotes(lp, fp, phdr->p_filesz)) != 0)
goto done;
/* fetch program text vnodes */
if ((error = elf_gettextvp(p, fp)) != 0)
goto done;
/* fetch signal disposition */
if ((error = elf_getsigs(lp, fp)) != 0) {
kprintf("failure in recovering signals\n");
goto done;
}
/* fetch open files */
if ((error = elf_getfiles(lp, fp)) != 0)
goto done;
/* handle mappings last in case we are reading from a socket */
error = elf_loadphdrs(fp, phdr, ehdr->e_phnum);
/*
* Set the textvp to the checkpoint file and mark the vnode so
* a future checkpointing of this checkpoint-restored program
* will copy out the contents of the mappings rather then trying
* to record the vnode info related to the checkpoint file, which
* is likely going to be destroyed when the program is re-checkpointed.
*/
if (error == 0 && fp->f_data && fp->f_type == DTYPE_VNODE) {
if (p->p_textvp)
vrele(p->p_textvp);
p->p_textvp = (struct vnode *)fp->f_data;
vsetflags(p->p_textvp, VCKPT);
vref(p->p_textvp);
}
done:
if (ehdr)
kfree(ehdr, M_TEMP);
if (phdr)
kfree(phdr, M_TEMP);
lwpsignal(p, lp, 35);
TRACE_EXIT;
return error;
}
/*
* Get the vnode associated with the given inode, allocating the vnode if
* necessary. The vnode will be returned exclusively locked.
*
* The caller must lock the inode (shared or exclusive).
*
* Great care must be taken to avoid deadlocks and vnode acquisition/reclaim
* races.
*/
struct vnode *
hammer2_igetv(hammer2_inode_t *ip, int *errorp)
{
hammer2_inode_data_t *ipdata;
hammer2_pfsmount_t *pmp;
struct vnode *vp;
ccms_state_t ostate;
pmp = ip->pmp;
KKASSERT(pmp != NULL);
*errorp = 0;
ipdata = &ip->chain->data->ipdata;
for (;;) {
/*
* Attempt to reuse an existing vnode assignment. It is
* possible to race a reclaim so the vget() may fail. The
* inode must be unlocked during the vget() to avoid a
* deadlock against a reclaim.
*/
vp = ip->vp;
if (vp) {
/*
* Inode must be unlocked during the vget() to avoid
* possible deadlocks, but leave the ip ref intact.
*
* vnode is held to prevent destruction during the
* vget(). The vget() can still fail if we lost
* a reclaim race on the vnode.
*/
vhold(vp);
ostate = hammer2_inode_lock_temp_release(ip);
if (vget(vp, LK_EXCLUSIVE)) {
vdrop(vp);
hammer2_inode_lock_temp_restore(ip, ostate);
continue;
}
hammer2_inode_lock_temp_restore(ip, ostate);
vdrop(vp);
/* vp still locked and ref from vget */
if (ip->vp != vp) {
kprintf("hammer2: igetv race %p/%p\n",
ip->vp, vp);
vput(vp);
continue;
}
*errorp = 0;
break;
}
/*
* No vnode exists, allocate a new vnode. Beware of
* allocation races. This function will return an
* exclusively locked and referenced vnode.
*/
*errorp = getnewvnode(VT_HAMMER2, pmp->mp, &vp, 0, 0);
if (*errorp) {
kprintf("hammer2: igetv getnewvnode failed %d\n",
*errorp);
vp = NULL;
break;
}
/*
* Lock the inode and check for an allocation race.
*/
ostate = hammer2_inode_lock_upgrade(ip);
if (ip->vp != NULL) {
vp->v_type = VBAD;
vx_put(vp);
hammer2_inode_lock_downgrade(ip, ostate);
continue;
}
switch (ipdata->type) {
case HAMMER2_OBJTYPE_DIRECTORY:
vp->v_type = VDIR;
break;
case HAMMER2_OBJTYPE_REGFILE:
vp->v_type = VREG;
vinitvmio(vp, ipdata->size,
HAMMER2_LBUFSIZE,
(int)ipdata->size & HAMMER2_LBUFMASK);
break;
case HAMMER2_OBJTYPE_SOFTLINK:
/*
* XXX for now we are using the generic file_read
* and file_write code so we need a buffer cache
* association.
*/
vp->v_type = VLNK;
vinitvmio(vp, ipdata->size,
HAMMER2_LBUFSIZE,
(int)ipdata->size & HAMMER2_LBUFMASK);
break;
case HAMMER2_OBJTYPE_CDEV:
vp->v_type = VCHR;
/* fall through */
case HAMMER2_OBJTYPE_BDEV:
vp->v_ops = &pmp->mp->mnt_vn_spec_ops;
if (ipdata->type != HAMMER2_OBJTYPE_CDEV)
vp->v_type = VBLK;
addaliasu(vp, ipdata->rmajor, ipdata->rminor);
break;
case HAMMER2_OBJTYPE_FIFO:
vp->v_type = VFIFO;
vp->v_ops = &pmp->mp->mnt_vn_fifo_ops;
break;
default:
panic("hammer2: unhandled objtype %d", ipdata->type);
break;
}
if (ip == pmp->iroot)
vsetflags(vp, VROOT);
vp->v_data = ip;
ip->vp = vp;
hammer2_inode_ref(ip); /* vp association */
hammer2_inode_lock_downgrade(ip, ostate);
break;
}
/*
* Return non-NULL vp and *errorp == 0, or NULL vp and *errorp != 0.
*/
if (hammer2_debug & 0x0002) {
kprintf("igetv vp %p refs 0x%08x aux 0x%08x\n",
vp, vp->v_refcnt, vp->v_auxrefs);
}
return (vp);
}
int
ntfs_vgetex(struct mount *mp, ino_t ino, u_int32_t attrtype, char *attrname,
u_long lkflags, u_long flags, struct thread *td,
struct vnode **vpp)
{
int error;
struct ntfsmount *ntmp;
struct ntnode *ip;
struct fnode *fp;
struct vnode *vp;
enum vtype f_type;
dprintf(("ntfs_vgetex: ino: %ju, attr: 0x%x:%s, lkf: 0x%lx, f: 0x%lx\n",
(uintmax_t) ino, attrtype, attrname?attrname:"", lkflags, flags));
ntmp = VFSTONTFS(mp);
*vpp = NULL;
/* Get ntnode */
error = ntfs_ntlookup(ntmp, ino, &ip);
if (error) {
kprintf("ntfs_vget: ntfs_ntget failed\n");
return (error);
}
/* It may be not initialized fully, so force load it */
if (!(flags & VG_DONTLOADIN) && !(ip->i_flag & IN_LOADED)) {
error = ntfs_loadntnode(ntmp, ip);
if(error) {
kprintf("ntfs_vget: CAN'T LOAD ATTRIBUTES FOR INO: %"PRId64"\n",
ip->i_number);
ntfs_ntput(ip);
return (error);
}
}
error = ntfs_fget(ntmp, ip, attrtype, attrname, &fp);
if (error) {
kprintf("ntfs_vget: ntfs_fget failed\n");
ntfs_ntput(ip);
return (error);
}
f_type = VINT;
if (!(flags & VG_DONTVALIDFN) && !(fp->f_flag & FN_VALID)) {
if ((ip->i_frflag & NTFS_FRFLAG_DIR) &&
(fp->f_attrtype == NTFS_A_DATA && fp->f_attrname == NULL)) {
f_type = VDIR;
} else if (flags & VG_EXT) {
f_type = VINT;
fp->f_size = fp->f_allocated = 0;
} else {
f_type = VREG;
error = ntfs_filesize(ntmp, fp,
&fp->f_size, &fp->f_allocated);
if (error) {
ntfs_ntput(ip);
return (error);
}
}
fp->f_flag |= FN_VALID;
}
if (FTOV(fp)) {
VGET(FTOV(fp), lkflags);
*vpp = FTOV(fp);
ntfs_ntput(ip);
return (0);
}
error = getnewvnode(VT_NTFS, ntmp->ntm_mountp, &vp, VLKTIMEOUT, 0);
if(error) {
ntfs_frele(fp);
ntfs_ntput(ip);
return (error);
}
dprintf(("ntfs_vget: vnode: %p for ntnode: %ju\n", vp, (uintmax_t)ino));
fp->f_vp = vp;
vp->v_data = fp;
vp->v_type = f_type;
if (ino == NTFS_ROOTINO)
vsetflags(vp, VROOT);
/*
* Normal files use the buffer cache
*/
if (f_type == VREG)
vinitvmio(vp, fp->f_size, PAGE_SIZE, -1);
ntfs_ntput(ip);
KKASSERT(lkflags & LK_TYPE_MASK);
/* XXX leave vnode locked exclusively from getnewvnode */
*vpp = vp;
return (0);
}
/*
* Common code for mount and mountroot
*/
int
ntfs_mountfs(struct vnode *devvp, struct mount *mp, struct ntfs_args *argsp,
struct ucred *cred)
{
struct buf *bp;
struct ntfsmount *ntmp;
cdev_t dev;
int error, ronly, ncount, i;
struct vnode *vp;
char cs_local[ICONV_CSNMAXLEN];
char cs_ntfs[ICONV_CSNMAXLEN];
/*
* Disallow multiple mounts of the same device.
* Disallow mounting of a device that is currently in use
* (except for root, which might share swap device for miniroot).
* Flush out any old buffers remaining from a previous use.
*/
error = vfs_mountedon(devvp);
if (error)
return (error);
ncount = vcount(devvp);
if (devvp->v_object)
ncount -= 1;
if (ncount > 1)
return (EBUSY);
VN_LOCK(devvp, LK_EXCLUSIVE | LK_RETRY);
error = vinvalbuf(devvp, V_SAVE, 0, 0);
VOP__UNLOCK(devvp, 0);
if (error)
return (error);
ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
VN_LOCK(devvp, LK_EXCLUSIVE | LK_RETRY);
error = VOP_OPEN(devvp, ronly ? FREAD : FREAD|FWRITE, FSCRED, NULL);
VOP__UNLOCK(devvp, 0);
if (error)
return (error);
dev = devvp->v_rdev;
bp = NULL;
error = bread(devvp, BBLOCK, BBSIZE, &bp);
if (error)
goto out;
ntmp = kmalloc(sizeof *ntmp, M_NTFSMNT, M_WAITOK | M_ZERO);
bcopy( bp->b_data, &ntmp->ntm_bootfile, sizeof(struct bootfile) );
/*
* We must not cache the boot block if its size is not exactly
* one cluster in order to avoid confusing the buffer cache when
* the boot file is read later by ntfs_readntvattr_plain(), which
* reads a cluster at a time.
*/
if (ntfs_cntob(1) != BBSIZE)
bp->b_flags |= B_NOCACHE;
brelse( bp );
bp = NULL;
if (strncmp(ntmp->ntm_bootfile.bf_sysid, NTFS_BBID, NTFS_BBIDLEN)) {
error = EINVAL;
dprintf(("ntfs_mountfs: invalid boot block\n"));
goto out;
}
{
int8_t cpr = ntmp->ntm_mftrecsz;
if( cpr > 0 )
ntmp->ntm_bpmftrec = ntmp->ntm_spc * cpr;
else
ntmp->ntm_bpmftrec = (1 << (-cpr)) / ntmp->ntm_bps;
}
dprintf(("ntfs_mountfs(): bps: %d, spc: %d, media: %x, mftrecsz: %d (%d sects)\n",
ntmp->ntm_bps,ntmp->ntm_spc,ntmp->ntm_bootfile.bf_media,
ntmp->ntm_mftrecsz,ntmp->ntm_bpmftrec));
dprintf(("ntfs_mountfs(): mftcn: 0x%x|0x%x\n",
(u_int32_t)ntmp->ntm_mftcn,(u_int32_t)ntmp->ntm_mftmirrcn));
ntmp->ntm_mountp = mp;
ntmp->ntm_dev = dev;
ntmp->ntm_devvp = devvp;
ntmp->ntm_uid = argsp->uid;
ntmp->ntm_gid = argsp->gid;
ntmp->ntm_mode = argsp->mode;
ntmp->ntm_flag = argsp->flag;
if (argsp->flag & NTFS_MFLAG_KICONV && ntfs_iconv) {
bcopy(argsp->cs_local, cs_local, sizeof(cs_local));
bcopy(argsp->cs_ntfs, cs_ntfs, sizeof(cs_ntfs));
ntfs_82u_init(ntmp, cs_local, cs_ntfs);
ntfs_u28_init(ntmp, NULL, cs_local, cs_ntfs);
} else {
ntfs_82u_init(ntmp, NULL, NULL);
ntfs_u28_init(ntmp, ntmp->ntm_82u, NULL, NULL);
}
mp->mnt_data = (qaddr_t)ntmp;
dprintf(("ntfs_mountfs(): case-%s,%s uid: %d, gid: %d, mode: %o\n",
(ntmp->ntm_flag & NTFS_MFLAG_CASEINS)?"insens.":"sens.",
(ntmp->ntm_flag & NTFS_MFLAG_ALLNAMES)?" allnames,":"",
ntmp->ntm_uid, ntmp->ntm_gid, ntmp->ntm_mode));
vfs_add_vnodeops(mp, &ntfs_vnode_vops, &mp->mnt_vn_norm_ops);
/*
* We read in some system nodes to do not allow
* reclaim them and to have everytime access to them.
*/
{
int pi[3] = { NTFS_MFTINO, NTFS_ROOTINO, NTFS_BITMAPINO };
for (i=0; i<3; i++) {
error = VFS_VGET(mp, NULL,
pi[i], &(ntmp->ntm_sysvn[pi[i]]));
if(error)
goto out1;
vsetflags(ntmp->ntm_sysvn[pi[i]], VSYSTEM);
vref(ntmp->ntm_sysvn[pi[i]]);
vput(ntmp->ntm_sysvn[pi[i]]);
}
}
/* read the Unicode lowercase --> uppercase translation table,
* if necessary */
if ((error = ntfs_toupper_use(mp, ntmp)))
goto out1;
/*
* Scan $BitMap and count free clusters
*/
error = ntfs_calccfree(ntmp, &ntmp->ntm_cfree);
if(error)
goto out1;
/*
* Read and translate to internal format attribute
* definition file.
*/
{
int num,j;
struct attrdef ad;
/* Open $AttrDef */
error = VFS_VGET(mp, NULL, NTFS_ATTRDEFINO, &vp);
if(error)
goto out1;
/* Count valid entries */
for(num=0;;num++) {
error = ntfs_readattr(ntmp, VTONT(vp),
NTFS_A_DATA, NULL,
num * sizeof(ad), sizeof(ad),
&ad, NULL);
if (error)
goto out1;
if (ad.ad_name[0] == 0)
break;
}
/* Alloc memory for attribute definitions */
ntmp->ntm_ad = kmalloc(num * sizeof(struct ntvattrdef),
M_NTFSMNT, M_WAITOK);
ntmp->ntm_adnum = num;
/* Read them and translate */
for(i=0;i<num;i++){
error = ntfs_readattr(ntmp, VTONT(vp),
NTFS_A_DATA, NULL,
i * sizeof(ad), sizeof(ad),
&ad, NULL);
if (error)
goto out1;
j = 0;
do {
ntmp->ntm_ad[i].ad_name[j] = ad.ad_name[j];
} while(ad.ad_name[j++]);
ntmp->ntm_ad[i].ad_namelen = j - 1;
ntmp->ntm_ad[i].ad_type = ad.ad_type;
}
vput(vp);
}
mp->mnt_stat.f_fsid.val[0] = dev2udev(dev);
mp->mnt_stat.f_fsid.val[1] = mp->mnt_vfc->vfc_typenum;
mp->mnt_maxsymlinklen = 0;
mp->mnt_flag |= MNT_LOCAL;
dev->si_mountpoint = mp;
return (0);
out1:
for(i=0;i<NTFS_SYSNODESNUM;i++)
if(ntmp->ntm_sysvn[i]) vrele(ntmp->ntm_sysvn[i]);
if (vflush(mp, 0, 0))
dprintf(("ntfs_mountfs: vflush failed\n"));
out:
dev->si_mountpoint = NULL;
if (bp)
brelse(bp);
vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
(void)VOP_CLOSE(devvp, ronly ? FREAD : FREAD|FWRITE, NULL);
vn_unlock(devvp);
return (error);
}
static int
devfs_spec_open(struct vop_open_args *ap)
{
struct vnode *vp = ap->a_vp;
struct vnode *orig_vp = NULL;
struct devfs_node *node = DEVFS_NODE(vp);
struct devfs_node *newnode;
cdev_t dev, ndev = NULL;
int error = 0;
if (node) {
if (node->d_dev == NULL)
return ENXIO;
if (!devfs_node_is_accessible(node))
return ENOENT;
}
if ((dev = vp->v_rdev) == NULL)
return ENXIO;
vn_lock(vp, LK_UPGRADE | LK_RETRY);
if (node && ap->a_fp) {
devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_open: -1.1-\n");
lockmgr(&devfs_lock, LK_EXCLUSIVE);
ndev = devfs_clone(dev, node->d_dir.d_name,
node->d_dir.d_namlen,
ap->a_mode, ap->a_cred);
if (ndev != NULL) {
newnode = devfs_create_device_node(
DEVFS_MNTDATA(vp->v_mount)->root_node,
ndev, NULL, NULL);
/* XXX: possibly destroy device if this happens */
if (newnode != NULL) {
dev = ndev;
devfs_link_dev(dev);
devfs_debug(DEVFS_DEBUG_DEBUG,
"parent here is: %s, node is: |%s|\n",
((node->parent->node_type == Nroot) ?
"ROOT!" : node->parent->d_dir.d_name),
newnode->d_dir.d_name);
devfs_debug(DEVFS_DEBUG_DEBUG,
"test: %s\n",
((struct devfs_node *)(TAILQ_LAST(DEVFS_DENODE_HEAD(node->parent), devfs_node_head)))->d_dir.d_name);
/*
* orig_vp is set to the original vp if we cloned.
*/
/* node->flags |= DEVFS_CLONED; */
devfs_allocv(&vp, newnode);
orig_vp = ap->a_vp;
ap->a_vp = vp;
}
}
lockmgr(&devfs_lock, LK_RELEASE);
}
devfs_debug(DEVFS_DEBUG_DEBUG,
"devfs_spec_open() called on %s! \n",
dev->si_name);
/*
* Make this field valid before any I/O in ->d_open
*/
if (!dev->si_iosize_max)
/* XXX: old DFLTPHYS == 64KB dependency */
dev->si_iosize_max = min(MAXPHYS,64*1024);
if (dev_dflags(dev) & D_TTY)
vsetflags(vp, VISTTY);
/*
* Open underlying device
*/
vn_unlock(vp);
error = dev_dopen(dev, ap->a_mode, S_IFCHR, ap->a_cred, ap->a_fp);
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
/*
* Clean up any cloned vp if we error out.
*/
if (error) {
if (orig_vp) {
vput(vp);
ap->a_vp = orig_vp;
/* orig_vp = NULL; */
}
return error;
}
/*
* This checks if the disk device is going to be opened for writing.
* It will be only allowed in the cases where securelevel permits it
* and it's not mounted R/W.
*/
if ((dev_dflags(dev) & D_DISK) && (ap->a_mode & FWRITE) &&
(ap->a_cred != FSCRED)) {
/* Very secure mode. No open for writing allowed */
if (securelevel >= 2)
return EPERM;
/*
* If it is mounted R/W, do not allow to open for writing.
* In the case it's mounted read-only but securelevel
* is >= 1, then do not allow opening for writing either.
*/
if (vfs_mountedon(vp)) {
if (!(dev->si_mountpoint->mnt_flag & MNT_RDONLY))
return EBUSY;
else if (securelevel >= 1)
return EPERM;
}
}
if (dev_dflags(dev) & D_TTY) {
if (dev->si_tty) {
struct tty *tp;
tp = dev->si_tty;
if (!tp->t_stop) {
devfs_debug(DEVFS_DEBUG_DEBUG,
"devfs: no t_stop\n");
tp->t_stop = nottystop;
}
}
}
if (vn_isdisk(vp, NULL)) {
if (!dev->si_bsize_phys)
dev->si_bsize_phys = DEV_BSIZE;
vinitvmio(vp, IDX_TO_OFF(INT_MAX), PAGE_SIZE, -1);
}
vop_stdopen(ap);
#if 0
if (node)
nanotime(&node->atime);
#endif
/*
* If we replaced the vp the vop_stdopen() call will have loaded
* it into fp->f_data and vref()d the vp, giving us two refs. So
* instead of just unlocking it here we have to vput() it.
*/
if (orig_vp)
vput(vp);
/* Ugly pty magic, to make pty devices appear once they are opened */
if (node && (node->flags & DEVFS_PTY) == DEVFS_PTY)
node->flags &= ~DEVFS_INVISIBLE;
if (ap->a_fp) {
KKASSERT(ap->a_fp->f_type == DTYPE_VNODE);
KKASSERT((ap->a_fp->f_flag & FMASK) == (ap->a_mode & FMASK));
ap->a_fp->f_ops = &devfs_dev_fileops;
KKASSERT(ap->a_fp->f_data == (void *)vp);
}
return 0;
}
int
ufs_quotaon(struct ucred *cred, struct mount *mp, int type, caddr_t fname)
{
struct ufsmount *ump = VFSTOUFS(mp);
struct vnode *vp, **vpp;
struct ufs_dquot *dq;
int error;
struct nlookupdata nd;
struct scaninfo scaninfo;
vpp = &ump->um_quotas[type];
error = nlookup_init(&nd, fname, UIO_USERSPACE, NLC_FOLLOW|NLC_LOCKVP);
if (error == 0)
error = vn_open(&nd, NULL, FREAD|FWRITE, 0);
if (error == 0 && nd.nl_open_vp->v_type != VREG)
error = EACCES;
if (error) {
nlookup_done(&nd);
return (error);
}
vp = nd.nl_open_vp;
nd.nl_open_vp = NULL;
nlookup_done(&nd);
vn_unlock(vp);
if (*vpp != vp)
ufs_quotaoff(mp, type);
ump->um_qflags[type] |= QTF_OPENING;
mp->mnt_flag |= MNT_QUOTA;
vsetflags(vp, VSYSTEM);
*vpp = vp;
/* XXX release duplicate vp if *vpp == vp? */
/*
* Save the credential of the process that turned on quotas.
* Set up the time limits for this quota.
*/
ump->um_cred[type] = crhold(cred);
ump->um_btime[type] = MAX_DQ_TIME;
ump->um_itime[type] = MAX_IQ_TIME;
if (ufs_dqget(NULLVP, 0, ump, type, &dq) == 0) {
if (dq->dq_btime > 0)
ump->um_btime[type] = dq->dq_btime;
if (dq->dq_itime > 0)
ump->um_itime[type] = dq->dq_itime;
ufs_dqrele(NULLVP, dq);
}
/*
* Search vnodes associated with this mount point,
* adding references to quota file being opened.
* NB: only need to add dquot's for inodes being modified.
*/
scaninfo.rescan = 1;
while (scaninfo.rescan) {
scaninfo.rescan = 0;
error = vmntvnodescan(mp, VMSC_GETVP,
NULL, ufs_quotaon_scan, &scaninfo);
if (error)
break;
}
ump->um_qflags[type] &= ~QTF_OPENING;
if (error)
ufs_quotaoff(mp, type);
return (error);
}
/* Return locked vnode to root of a filesystem */
static int
nwfs_root(struct mount *mp, struct vnode **vpp)
{
struct vnode *vp;
struct nwmount *nmp;
struct nwnode *np;
struct ncp_conn *conn;
struct nw_entry_info fattr;
struct thread *td = curthread; /* XXX */
struct ucred *cred;
int error, nsf, opt;
u_char vol;
KKASSERT(td->td_proc);
cred = td->td_proc->p_ucred;
nmp = VFSTONWFS(mp);
conn = NWFSTOCONN(nmp);
if (nmp->n_root) {
*vpp = NWTOV(nmp->n_root);
while (vget(*vpp, LK_EXCLUSIVE) != 0) /* XXX */
;
return 0;
}
error = ncp_lookup_volume(conn, nmp->m.mounted_vol, &vol,
&nmp->n_rootent.f_id, td, cred);
if (error)
return ENOENT;
nmp->n_volume = vol;
error = ncp_get_namespaces(conn, vol, &nsf, td, cred);
if (error)
return ENOENT;
if (nsf & NW_NSB_OS2) {
NCPVODEBUG("volume %s has os2 namespace\n",nmp->m.mounted_vol);
if ((nmp->m.flags & NWFS_MOUNT_NO_OS2) == 0) {
nmp->name_space = NW_NS_OS2;
nmp->m.nls.opt &= ~NWHP_DOS;
}
}
opt = nmp->m.nls.opt;
nsf = opt & (NWHP_UPPER | NWHP_LOWER);
if (opt & NWHP_DOS) {
if (nsf == (NWHP_UPPER | NWHP_LOWER)) {
nmp->m.nls.opt &= ~(NWHP_LOWER | NWHP_UPPER);
} else if (nsf == 0) {
nmp->m.nls.opt |= NWHP_LOWER;
}
} else {
if (nsf == (NWHP_UPPER | NWHP_LOWER)) {
nmp->m.nls.opt &= ~(NWHP_LOWER | NWHP_UPPER);
}
}
if (nmp->m.root_path[0]) {
nmp->m.root_path[0]--;
error = ncp_obtain_info(nmp, nmp->n_rootent.f_id,
-nmp->m.root_path[0], nmp->m.root_path, &fattr, td, cred);
if (error) {
NCPFATAL("Invalid root path specified\n");
return ENOENT;
}
nmp->n_rootent.f_parent = fattr.dirEntNum;
nmp->m.root_path[0]++;
error = ncp_obtain_info(nmp, nmp->n_rootent.f_id,
-nmp->m.root_path[0], nmp->m.root_path, &fattr, td, cred);
if (error) {
NCPFATAL("Invalid root path specified\n");
return ENOENT;
}
nmp->n_rootent.f_id = fattr.dirEntNum;
} else {
error = ncp_obtain_info(nmp, nmp->n_rootent.f_id,
0, NULL, &fattr, td, cred);
if (error) {
NCPFATAL("Can't obtain volume info\n");
return ENOENT;
}
fattr.nameLen = strlen(strcpy(fattr.entryName, NWFS_ROOTVOL));
nmp->n_rootent.f_parent = nmp->n_rootent.f_id;
}
error = nwfs_nget(mp, nmp->n_rootent, &fattr, NULL, &vp);
if (error)
return (error);
vsetflags(vp, VROOT);
np = VTONW(vp);
if (nmp->m.root_path[0] == 0)
np->n_flag |= NVOLUME;
nmp->n_root = np;
/* error = VOP_GETATTR(vp, &vattr);
if (error) {
vput(vp);
NCPFATAL("Can't get root directory entry\n");
return error;
}*/
*vpp = vp;
return (0);
}
/*
* Allocates a new vnode for the node node or returns a new reference to
* an existing one if the node had already a vnode referencing it. The
* resulting locked vnode is returned in *vpp.
*
* Returns zero on success or an appropriate error code on failure.
*
* The caller must ensure that node cannot go away (usually by holding
* the related directory entry).
*
* If dnode is non-NULL this routine avoids deadlocking against it but
* can return EAGAIN. Caller must try again. The dnode lock will cycle
* in this case, it remains locked on return in all cases. dnode must
* be shared-locked.
*/
int
tmpfs_alloc_vp(struct mount *mp,
struct tmpfs_node *dnode, struct tmpfs_node *node, int lkflag,
struct vnode **vpp)
{
int error = 0;
struct vnode *vp;
loop:
/*
* Interlocked extraction from node. This can race many things.
* We have to get a soft reference on the vnode while we hold
* the node locked, then acquire it properly and check for races.
*/
TMPFS_NODE_LOCK(node);
if ((vp = node->tn_vnode) != NULL) {
KKASSERT((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0);
vhold(vp);
TMPFS_NODE_UNLOCK(node);
if (dnode) {
/*
* Special-case handling to avoid deadlocking against
* dnode. This case has been validated and occurs
* every so often during synth builds.
*/
if (vget(vp, (lkflag & ~LK_RETRY) |
LK_NOWAIT |
LK_EXCLUSIVE) != 0) {
TMPFS_NODE_UNLOCK(dnode);
if (vget(vp, (lkflag & ~LK_RETRY) |
LK_SLEEPFAIL |
LK_EXCLUSIVE) == 0) {
vn_unlock(vp);
}
vdrop(vp);
TMPFS_NODE_LOCK_SH(dnode);
return EAGAIN;
}
} else {
/*
* Normal path
*/
if (vget(vp, lkflag | LK_EXCLUSIVE) != 0) {
vdrop(vp);
goto loop;
}
}
if (node->tn_vnode != vp) {
vput(vp);
vdrop(vp);
goto loop;
}
vdrop(vp);
goto out;
}
/* vp is NULL */
/*
* This should never happen.
*/
if (node->tn_vpstate & TMPFS_VNODE_DOOMED) {
TMPFS_NODE_UNLOCK(node);
error = ENOENT;
goto out;
}
/*
* Interlock against other calls to tmpfs_alloc_vp() trying to
* allocate and assign a vp to node.
*/
if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
node->tn_vpstate |= TMPFS_VNODE_WANT;
error = tsleep(&node->tn_vpstate, PINTERLOCKED | PCATCH,
"tmpfs_alloc_vp", 0);
TMPFS_NODE_UNLOCK(node);
if (error)
return error;
goto loop;
}
node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
TMPFS_NODE_UNLOCK(node);
/*
* Allocate a new vnode (may block). The ALLOCATING flag should
* prevent a race against someone else assigning node->tn_vnode.
*/
error = getnewvnode(VT_TMPFS, mp, &vp, VLKTIMEOUT, LK_CANRECURSE);
if (error != 0)
goto unlock;
KKASSERT(node->tn_vnode == NULL);
KKASSERT(vp != NULL);
vp->v_data = node;
vp->v_type = node->tn_type;
/* Type-specific initialization. */
switch (node->tn_type) {
case VBLK:
/* FALLTHROUGH */
case VCHR:
/* FALLTHROUGH */
case VSOCK:
break;
case VREG:
/*
* VMIO is mandatory. Tmpfs also supports KVABIO
* for its tmpfs_strategy().
*/
vsetflags(vp, VKVABIO);
vinitvmio(vp, node->tn_size, TMPFS_BLKSIZE, -1);
break;
case VLNK:
break;
case VFIFO:
vp->v_ops = &mp->mnt_vn_fifo_ops;
break;
case VDIR:
break;
default:
panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
}
unlock:
TMPFS_NODE_LOCK(node);
KKASSERT(node->tn_vpstate & TMPFS_VNODE_ALLOCATING);
node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING;
node->tn_vnode = vp;
if (node->tn_vpstate & TMPFS_VNODE_WANT) {
node->tn_vpstate &= ~TMPFS_VNODE_WANT;
TMPFS_NODE_UNLOCK(node);
wakeup(&node->tn_vpstate);
} else {
TMPFS_NODE_UNLOCK(node);
}
out:
*vpp = vp;
KKASSERT(IFF(error == 0, *vpp != NULL && vn_islocked(*vpp)));
return error;
}
/*
* Common code for vnode open operations. Check permissions, and call
* the VOP_NOPEN or VOP_NCREATE routine.
*
* The caller is responsible for setting up nd with nlookup_init() and
* for cleaning it up with nlookup_done(), whether we return an error
* or not.
*
* On success nd->nl_open_vp will hold a referenced and, if requested,
* locked vnode. A locked vnode is requested via NLC_LOCKVP. If fp
* is non-NULL the vnode will be installed in the file pointer.
*
* NOTE: The vnode is referenced just once on return whether or not it
* is also installed in the file pointer.
*/
int
vn_open(struct nlookupdata *nd, struct file *fp, int fmode, int cmode)
{
struct vnode *vp;
struct ucred *cred = nd->nl_cred;
struct vattr vat;
struct vattr *vap = &vat;
int error;
u_int flags;
uint64_t osize;
struct mount *mp;
/*
* Certain combinations are illegal
*/
if ((fmode & (FWRITE | O_TRUNC)) == O_TRUNC)
return(EACCES);
/*
* Lookup the path and create or obtain the vnode. After a
* successful lookup a locked nd->nl_nch will be returned.
*
* The result of this section should be a locked vnode.
*
* XXX with only a little work we should be able to avoid locking
* the vnode if FWRITE, O_CREAT, and O_TRUNC are *not* set.
*/
nd->nl_flags |= NLC_OPEN;
if (fmode & O_APPEND)
nd->nl_flags |= NLC_APPEND;
if (fmode & O_TRUNC)
nd->nl_flags |= NLC_TRUNCATE;
if (fmode & FREAD)
nd->nl_flags |= NLC_READ;
if (fmode & FWRITE)
nd->nl_flags |= NLC_WRITE;
if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
nd->nl_flags |= NLC_FOLLOW;
if (fmode & O_CREAT) {
/*
* CONDITIONAL CREATE FILE CASE
*
* Setting NLC_CREATE causes a negative hit to store
* the negative hit ncp and not return an error. Then
* nc_error or nc_vp may be checked to see if the ncp
* represents a negative hit. NLC_CREATE also requires
* write permission on the governing directory or EPERM
* is returned.
*/
nd->nl_flags |= NLC_CREATE;
nd->nl_flags |= NLC_REFDVP;
bwillinode(1);
error = nlookup(nd);
} else {
/*
* NORMAL OPEN FILE CASE
*/
error = nlookup(nd);
}
if (error)
return (error);
/*
* split case to allow us to re-resolve and retry the ncp in case
* we get ESTALE.
*/
again:
if (fmode & O_CREAT) {
if (nd->nl_nch.ncp->nc_vp == NULL) {
if ((error = ncp_writechk(&nd->nl_nch)) != 0)
return (error);
VATTR_NULL(vap);
vap->va_type = VREG;
vap->va_mode = cmode;
if (fmode & O_EXCL)
vap->va_vaflags |= VA_EXCLUSIVE;
error = VOP_NCREATE(&nd->nl_nch, nd->nl_dvp, &vp,
nd->nl_cred, vap);
if (error)
return (error);
fmode &= ~O_TRUNC;
/* locked vnode is returned */
} else {
if (fmode & O_EXCL) {
error = EEXIST;
} else {
error = cache_vget(&nd->nl_nch, cred,
LK_EXCLUSIVE, &vp);
}
if (error)
return (error);
fmode &= ~O_CREAT;
}
} else {
error = cache_vget(&nd->nl_nch, cred, LK_EXCLUSIVE, &vp);
if (error)
return (error);
}
/*
* We have a locked vnode and ncp now. Note that the ncp will
* be cleaned up by the caller if nd->nl_nch is left intact.
*/
if (vp->v_type == VLNK) {
error = EMLINK;
goto bad;
}
if (vp->v_type == VSOCK) {
error = EOPNOTSUPP;
goto bad;
}
if ((fmode & O_CREAT) == 0) {
if (fmode & (FWRITE | O_TRUNC)) {
if (vp->v_type == VDIR) {
error = EISDIR;
goto bad;
}
error = vn_writechk(vp, &nd->nl_nch);
if (error) {
/*
* Special stale handling, re-resolve the
* vnode.
*/
if (error == ESTALE) {
vput(vp);
vp = NULL;
cache_setunresolved(&nd->nl_nch);
error = cache_resolve(&nd->nl_nch, cred);
if (error == 0)
goto again;
}
goto bad;
}
}
}
if (fmode & O_TRUNC) {
vn_unlock(vp); /* XXX */
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); /* XXX */
osize = vp->v_filesize;
VATTR_NULL(vap);
vap->va_size = 0;
error = VOP_SETATTR(vp, vap, cred);
if (error)
goto bad;
error = VOP_GETATTR(vp, vap);
if (error)
goto bad;
mp = vq_vptomp(vp);
VFS_ACCOUNT(mp, vap->va_uid, vap->va_gid, -osize);
}
/*
* Set or clear VNSWAPCACHE on the vp based on nd->nl_nch.ncp->nc_flag.
* These particular bits a tracked all the way from the root.
*
* NOTE: Might not work properly on NFS servers due to the
* disconnected namecache.
*/
flags = nd->nl_nch.ncp->nc_flag;
if ((flags & (NCF_UF_CACHE | NCF_UF_PCACHE)) &&
(flags & (NCF_SF_NOCACHE | NCF_SF_PNOCACHE)) == 0) {
vsetflags(vp, VSWAPCACHE);
} else {
vclrflags(vp, VSWAPCACHE);
}
/*
* Setup the fp so VOP_OPEN can override it. No descriptor has been
* associated with the fp yet so we own it clean.
*
* f_nchandle inherits nl_nch. This used to be necessary only for
* directories but now we do it unconditionally so f*() ops
* such as fchmod() can access the actual namespace that was
* used to open the file.
*/
if (fp) {
if (nd->nl_flags & NLC_APPENDONLY)
fmode |= FAPPENDONLY;
fp->f_nchandle = nd->nl_nch;
cache_zero(&nd->nl_nch);
cache_unlock(&fp->f_nchandle);
}
/*
* Get rid of nl_nch. vn_open does not return it (it returns the
* vnode or the file pointer). Note: we can't leave nl_nch locked
* through the VOP_OPEN anyway since the VOP_OPEN may block, e.g.
* on /dev/ttyd0
*/
if (nd->nl_nch.ncp)
cache_put(&nd->nl_nch);
error = VOP_OPEN(vp, fmode, cred, fp);
if (error) {
/*
* setting f_ops to &badfileops will prevent the descriptor
* code from trying to close and release the vnode, since
* the open failed we do not want to call close.
*/
if (fp) {
fp->f_data = NULL;
fp->f_ops = &badfileops;
}
goto bad;
}
#if 0
/*
* Assert that VREG files have been setup for vmio.
*/
KASSERT(vp->v_type != VREG || vp->v_object != NULL,
("vn_open: regular file was not VMIO enabled!"));
#endif
/*
* Return the vnode. XXX needs some cleaning up. The vnode is
* only returned in the fp == NULL case.
*/
if (fp == NULL) {
nd->nl_open_vp = vp;
nd->nl_vp_fmode = fmode;
if ((nd->nl_flags & NLC_LOCKVP) == 0)
vn_unlock(vp);
} else {
vput(vp);
}
return (0);
bad:
if (vp)
vput(vp);
return (error);
}
int
vfs_mountroot_devfs(void)
{
struct vnode *vp;
struct nchandle nch;
struct nlookupdata nd;
struct mount *mp;
struct vfsconf *vfsp;
int error;
struct ucred *cred = proc0.p_ucred;
const char *devfs_path, *init_chroot;
char *dev_malloced = NULL;
if ((init_chroot = kgetenv("init_chroot")) != NULL) {
size_t l;
l = strlen(init_chroot) + sizeof("/dev");
dev_malloced = kmalloc(l, M_MOUNT, M_WAITOK);
ksnprintf(dev_malloced, l, "%s/dev", init_chroot);
devfs_path = dev_malloced;
} else {
devfs_path = "/dev";
}
/*
* Lookup the requested path and extract the nch and vnode.
*/
error = nlookup_init_raw(&nd,
devfs_path, UIO_SYSSPACE, NLC_FOLLOW,
cred, &rootnch);
if (error == 0) {
devfs_debug(DEVFS_DEBUG_DEBUG, "vfs_mountroot_devfs: nlookup_init is ok...\n");
if ((error = nlookup(&nd)) == 0) {
devfs_debug(DEVFS_DEBUG_DEBUG, "vfs_mountroot_devfs: nlookup is ok...\n");
if (nd.nl_nch.ncp->nc_vp == NULL) {
devfs_debug(DEVFS_DEBUG_SHOW, "vfs_mountroot_devfs: nlookup: simply not found\n");
error = ENOENT;
}
}
}
if (dev_malloced != NULL)
kfree(dev_malloced, M_MOUNT), dev_malloced = NULL;
devfs_path = NULL;
if (error) {
nlookup_done(&nd);
devfs_debug(DEVFS_DEBUG_SHOW, "vfs_mountroot_devfs: nlookup failed, error: %d\n", error);
return (error);
}
/*
* Extract the locked+refd ncp and cleanup the nd structure
*/
nch = nd.nl_nch;
cache_zero(&nd.nl_nch);
nlookup_done(&nd);
/*
* now we have the locked ref'd nch and unreferenced vnode.
*/
vp = nch.ncp->nc_vp;
if ((error = vget(vp, LK_EXCLUSIVE)) != 0) {
cache_put(&nch);
devfs_debug(DEVFS_DEBUG_SHOW, "vfs_mountroot_devfs: vget failed\n");
return (error);
}
cache_unlock(&nch);
if ((error = vinvalbuf(vp, V_SAVE, 0, 0)) != 0) {
cache_drop(&nch);
vput(vp);
devfs_debug(DEVFS_DEBUG_SHOW, "vfs_mountroot_devfs: vinvalbuf failed\n");
return (error);
}
if (vp->v_type != VDIR) {
cache_drop(&nch);
vput(vp);
devfs_debug(DEVFS_DEBUG_SHOW, "vfs_mountroot_devfs: vp is not VDIR\n");
return (ENOTDIR);
}
vfsp = vfsconf_find_by_name("devfs");
vsetflags(vp, VMOUNT);
/*
* Allocate and initialize the filesystem.
*/
mp = kmalloc(sizeof(struct mount), M_MOUNT, M_ZERO|M_WAITOK);
mount_init(mp);
vfs_busy(mp, LK_NOWAIT);
mp->mnt_op = vfsp->vfc_vfsops;
mp->mnt_vfc = vfsp;
vfsp->vfc_refcount++;
mp->mnt_stat.f_type = vfsp->vfc_typenum;
mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
mp->mnt_stat.f_owner = cred->cr_uid;
vn_unlock(vp);
/*
* Mount the filesystem.
*/
error = VFS_MOUNT(mp, "/dev", NULL, cred);
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
/*
* Put the new filesystem on the mount list after root. The mount
* point gets its own mnt_ncmountpt (unless the VFS already set one
* up) which represents the root of the mount. The lookup code
* detects the mount point going forward and checks the root of
* the mount going backwards.
*
* It is not necessary to invalidate or purge the vnode underneath
* because elements under the mount will be given their own glue
* namecache record.
*/
if (!error) {
if (mp->mnt_ncmountpt.ncp == NULL) {
/*
* allocate, then unlock, but leave the ref intact
*/
cache_allocroot(&mp->mnt_ncmountpt, mp, NULL);
cache_unlock(&mp->mnt_ncmountpt);
}
mp->mnt_ncmounton = nch; /* inherits ref */
nch.ncp->nc_flag |= NCF_ISMOUNTPT;
/* XXX get the root of the fs and cache_setvp(mnt_ncmountpt...) */
vclrflags(vp, VMOUNT);
mountlist_insert(mp, MNTINS_LAST);
vn_unlock(vp);
//checkdirs(&mp->mnt_ncmounton, &mp->mnt_ncmountpt);
error = vfs_allocate_syncvnode(mp);
if (error) {
devfs_debug(DEVFS_DEBUG_SHOW, "vfs_mountroot_devfs: vfs_allocate_syncvnode failed\n");
}
vfs_unbusy(mp);
error = VFS_START(mp, 0);
vrele(vp);
} else {
vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
vclrflags(vp, VMOUNT);
mp->mnt_vfc->vfc_refcount--;
vfs_unbusy(mp);
kfree(mp, M_MOUNT);
cache_drop(&nch);
vput(vp);
devfs_debug(DEVFS_DEBUG_SHOW, "vfs_mountroot_devfs: mount failed\n");
}
devfs_debug(DEVFS_DEBUG_DEBUG, "rootmount_devfs done with error: %d\n", error);
return (error);
}
/*
* Allocate a VM object for a vnode, typically a regular file vnode.
*
* Some additional information is required to generate a properly sized
* object which covers the entire buffer cache buffer straddling the file
* EOF. Userland does not see the extra pages as the VM fault code tests
* against v_filesize.
*/
vm_object_t
vnode_pager_alloc(void *handle, off_t length, vm_prot_t prot, off_t offset,
int blksize, int boff)
{
vm_object_t object;
struct vnode *vp;
off_t loffset;
vm_pindex_t lsize;
/*
* Pageout to vnode, no can do yet.
*/
if (handle == NULL)
return (NULL);
/*
* XXX hack - This initialization should be put somewhere else.
*/
if (vnode_pbuf_freecnt < 0) {
vnode_pbuf_freecnt = nswbuf / 2 + 1;
}
/*
* Serialize potential vnode/object teardowns and interlocks
*/
vp = (struct vnode *)handle;
lwkt_gettoken(&vp->v_token);
/*
* Prevent race condition when allocating the object. This
* can happen with NFS vnodes since the nfsnode isn't locked.
*/
while (vp->v_flag & VOLOCK) {
vsetflags(vp, VOWANT);
tsleep(vp, 0, "vnpobj", 0);
}
vsetflags(vp, VOLOCK);
lwkt_reltoken(&vp->v_token);
/*
* If the object is being terminated, wait for it to
* go away.
*/
while ((object = vp->v_object) != NULL) {
vm_object_hold(object);
if ((object->flags & OBJ_DEAD) == 0)
break;
vm_object_dead_sleep(object, "vadead");
vm_object_drop(object);
}
if (vp->v_sysref.refcnt <= 0)
panic("vnode_pager_alloc: no vnode reference");
/*
* Round up to the *next* block, then destroy the buffers in question.
* Since we are only removing some of the buffers we must rely on the
* scan count to determine whether a loop is necessary.
*
* Destroy any pages beyond the last buffer.
*/
if (boff < 0)
boff = (int)(length % blksize);
if (boff)
loffset = length + (blksize - boff);
else
loffset = length;
lsize = OFF_TO_IDX(round_page64(loffset));
if (object == NULL) {
/*
* And an object of the appropriate size
*/
object = vm_object_allocate_hold(OBJT_VNODE, lsize);
object->handle = handle;
vp->v_object = object;
vp->v_filesize = length;
if (vp->v_mount && (vp->v_mount->mnt_kern_flag & MNTK_NOMSYNC))
vm_object_set_flag(object, OBJ_NOMSYNC);
} else {
object->ref_count++;
if (object->size != lsize) {
kprintf("vnode_pager_alloc: Warning, objsize "
"mismatch %jd/%jd vp=%p obj=%p\n",
(intmax_t)object->size,
(intmax_t)lsize,
vp, object);
}
if (vp->v_filesize != length) {
kprintf("vnode_pager_alloc: Warning, filesize "
"mismatch %jd/%jd vp=%p obj=%p\n",
(intmax_t)vp->v_filesize,
(intmax_t)length,
vp, object);
}
}
vref(vp);
lwkt_gettoken(&vp->v_token);
vclrflags(vp, VOLOCK);
if (vp->v_flag & VOWANT) {
vclrflags(vp, VOWANT);
wakeup(vp);
}
lwkt_reltoken(&vp->v_token);
vm_object_drop(object);
return (object);
}
