/*
* Return root of a filesystem
*/
static int
nfs_root(struct mount *mp, int flags, struct vnode **vpp)
{
struct vnode *vp;
struct nfsmount *nmp;
struct nfsnode *np;
int error;
nmp = VFSTONFS(mp);
error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np, flags);
if (error)
return error;
vp = NFSTOV(np);
/*
* Get transfer parameters and attributes for root vnode once.
*/
mtx_lock(&nmp->nm_mtx);
if ((nmp->nm_state & NFSSTA_GOTFSINFO) == 0 &&
(nmp->nm_flag & NFSMNT_NFSV3)) {
mtx_unlock(&nmp->nm_mtx);
nfs_fsinfo(nmp, vp, curthread->td_ucred, curthread);
} else
mtx_unlock(&nmp->nm_mtx);
if (vp->v_type == VNON)
vp->v_type = VDIR;
vp->v_vflag |= VV_ROOT;
*vpp = vp;
return (0);
}
static int
nfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
{
struct vnode *vp;
struct nfs_statfs *sfp;
struct nfsmount *nmp = VFSTONFS(mp);
thread_t td = curthread;
int error = 0, retattr;
struct nfsnode *np;
struct nfsm_info info;
info.mrep = NULL;
info.v3 = (nmp->nm_flag & NFSMNT_NFSV3);
lwkt_gettoken(&nmp->nm_token);
#ifndef nolint
sfp = NULL;
#endif
error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np, NULL);
if (error) {
lwkt_reltoken(&nmp->nm_token);
return (error);
}
vp = NFSTOV(np);
/* ignore the passed cred */
cred = crget();
cred->cr_ngroups = 1;
if (info.v3 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0)
(void)nfs_fsinfo(nmp, vp, td);
nfsstats.rpccnt[NFSPROC_FSSTAT]++;
nfsm_reqhead(&info, vp, NFSPROC_FSSTAT, NFSX_FH(info.v3));
ERROROUT(nfsm_fhtom(&info, vp));
NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_FSSTAT, td, cred, &error));
if (info.v3) {
ERROROUT(nfsm_postop_attr(&info, vp, &retattr,
NFS_LATTR_NOSHRINK));
}
if (error) {
if (info.mrep != NULL)
m_freem(info.mrep);
goto nfsmout;
}
NULLOUT(sfp = nfsm_dissect(&info, NFSX_STATFS(info.v3)));
sbp->f_flag = nmp->nm_flag;
sbp->f_owner = nmp->nm_cred->cr_ruid;
if (info.v3) {
sbp->f_bsize = NFS_FABLKSIZE;
sbp->f_frsize = NFS_FABLKSIZE;
sbp->f_blocks = (fxdr_hyper(&sfp->sf_tbytes) /
((u_quad_t)NFS_FABLKSIZE));
sbp->f_bfree = (fxdr_hyper(&sfp->sf_fbytes) /
((u_quad_t)NFS_FABLKSIZE));
sbp->f_bavail = (fxdr_hyper(&sfp->sf_abytes) /
((u_quad_t)NFS_FABLKSIZE));
sbp->f_files = fxdr_hyper(&sfp->sf_tfiles);
sbp->f_ffree = fxdr_hyper(&sfp->sf_ffiles);
sbp->f_favail = fxdr_hyper(&sfp->sf_afiles);
} else {
sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
sbp->f_files = 0;
sbp->f_ffree = 0;
sbp->f_favail = 0;
}
sbp->f_syncreads = 0;
sbp->f_syncwrites = 0;
sbp->f_asyncreads = 0;
sbp->f_asyncwrites = 0;
sbp->f_type = mp->mnt_vfc->vfc_typenum;
m_freem(info.mrep);
info.mrep = NULL;
nfsmout:
vput(vp);
crfree(cred);
lwkt_reltoken(&nmp->nm_token);
return (error);
}
/*
* nfs statfs call
*/
int
nfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
{
struct vnode *vp;
struct nfs_statfs *sfp;
struct nfsmount *nmp = VFSTONFS(mp);
thread_t td = curthread;
int error = 0, retattr;
struct nfsnode *np;
u_quad_t tquad;
struct nfsm_info info;
info.mrep = NULL;
info.v3 = (nmp->nm_flag & NFSMNT_NFSV3);
lwkt_gettoken(&nmp->nm_token);
#ifndef nolint
sfp = NULL;
#endif
error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np, NULL);
if (error) {
lwkt_reltoken(&nmp->nm_token);
return (error);
}
vp = NFSTOV(np);
/* ignore the passed cred */
cred = crget();
cred->cr_ngroups = 1;
if (info.v3 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0)
(void)nfs_fsinfo(nmp, vp, td);
nfsstats.rpccnt[NFSPROC_FSSTAT]++;
nfsm_reqhead(&info, vp, NFSPROC_FSSTAT, NFSX_FH(info.v3));
ERROROUT(nfsm_fhtom(&info, vp));
NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_FSSTAT, td, cred, &error));
if (info.v3) {
ERROROUT(nfsm_postop_attr(&info, vp, &retattr,
NFS_LATTR_NOSHRINK));
}
if (error) {
if (info.mrep != NULL)
m_freem(info.mrep);
goto nfsmout;
}
NULLOUT(sfp = nfsm_dissect(&info, NFSX_STATFS(info.v3)));
sbp->f_flags = nmp->nm_flag;
if (info.v3) {
sbp->f_bsize = NFS_FABLKSIZE;
tquad = fxdr_hyper(&sfp->sf_tbytes);
sbp->f_blocks = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
tquad = fxdr_hyper(&sfp->sf_fbytes);
sbp->f_bfree = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
tquad = fxdr_hyper(&sfp->sf_abytes);
sbp->f_bavail = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
sbp->f_files = (fxdr_unsigned(int32_t,
sfp->sf_tfiles.nfsuquad[1]) & 0x7fffffff);
sbp->f_ffree = (fxdr_unsigned(int32_t,
sfp->sf_ffiles.nfsuquad[1]) & 0x7fffffff);
} else {
sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
sbp->f_files = 0;
sbp->f_ffree = 0;
}
/*
* Some values are pre-set in mnt_stat. Note in particular f_iosize
* cannot be changed once the filesystem is mounted as it is used
* as the basis for BIOs.
*/
if (sbp != &mp->mnt_stat) {
sbp->f_type = mp->mnt_vfc->vfc_typenum;
bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN);
sbp->f_iosize = mp->mnt_stat.f_iosize;
}
m_freem(info.mrep);
info.mrep = NULL;
nfsmout:
vput(vp);
crfree(cred);
lwkt_reltoken(&nmp->nm_token);
return (error);
}
int
nfs_nget(struct mount *mntp, nfsfh_t *fhp, int fhsize, struct nfsnode **npp)
{
struct nfsnode *np, *np2;
struct nfsnodehashhead *nhpp;
struct vnode *vp;
int error;
int lkflags;
struct nfsmount *nmp;
/*
* Calculate nfs mount point and figure out whether the rslock should
* be interruptable or not.
*/
nmp = VFSTONFS(mntp);
if (nmp->nm_flag & NFSMNT_INT)
lkflags = LK_PCATCH;
else
lkflags = 0;
lwkt_gettoken(&nfsnhash_token);
retry:
nhpp = NFSNOHASH(fnv_32_buf(fhp->fh_bytes, fhsize, FNV1_32_INIT));
loop:
for (np = nhpp->lh_first; np; np = np->n_hash.le_next) {
if (mntp != NFSTOV(np)->v_mount || np->n_fhsize != fhsize ||
bcmp((caddr_t)fhp, (caddr_t)np->n_fhp, fhsize)) {
continue;
}
vp = NFSTOV(np);
if (vget(vp, LK_EXCLUSIVE))
goto loop;
for (np = nhpp->lh_first; np; np = np->n_hash.le_next) {
if (mntp == NFSTOV(np)->v_mount &&
np->n_fhsize == fhsize &&
bcmp((caddr_t)fhp, (caddr_t)np->n_fhp, fhsize) == 0
) {
break;
}
}
if (np == NULL || NFSTOV(np) != vp) {
vput(vp);
goto loop;
}
*npp = np;
lwkt_reltoken(&nfsnhash_token);
return(0);
}
/*
* Obtain a lock to prevent a race condition if the getnewvnode()
* or MALLOC() below happens to block.
*/
if (lockmgr(&nfsnhash_lock, LK_EXCLUSIVE | LK_SLEEPFAIL))
goto loop;
/*
* Allocate before getnewvnode since doing so afterward
* might cause a bogus v_data pointer to get dereferenced
* elsewhere if objcache should block.
*/
np = objcache_get(nfsnode_objcache, M_WAITOK);
error = getnewvnode(VT_NFS, mntp, &vp, 0, 0);
if (error) {
lockmgr(&nfsnhash_lock, LK_RELEASE);
*npp = NULL;
objcache_put(nfsnode_objcache, np);
lwkt_reltoken(&nfsnhash_token);
return (error);
}
/*
* Initialize most of (np).
*/
bzero(np, sizeof (*np));
if (fhsize > NFS_SMALLFH) {
MALLOC(np->n_fhp, nfsfh_t *, fhsize, M_NFSBIGFH, M_WAITOK);
} else {
/*
* Look up a vnode/nfsnode by file handle.
* Callers must check for mount points!!
* In all cases, a pointer to a
* nfsnode structure is returned.
*/
int
nfs_nget(
mount_t mp,
nfsnode_t dnp,
struct componentname *cnp,
u_char *fhp,
int fhsize,
struct nfs_vattr *nvap,
u_int64_t *xidp,
uint32_t auth,
int flags,
nfsnode_t *npp)
{
nfsnode_t np;
struct nfsnodehashhead *nhpp;
vnode_t vp;
int error, nfsvers;
mount_t mp2;
struct vnode_fsparam vfsp;
uint32_t vid;
FSDBG_TOP(263, mp, dnp, flags, npp);
/* Check for unmount in progress */
if (!mp || vfs_isforce(mp)) {
*npp = NULL;
error = ENXIO;
FSDBG_BOT(263, mp, dnp, 0xd1e, error);
return (error);
}
nfsvers = VFSTONFS(mp)->nm_vers;
nhpp = NFSNOHASH(nfs_hash(fhp, fhsize));
loop:
lck_mtx_lock(nfs_node_hash_mutex);
for (np = nhpp->lh_first; np != 0; np = np->n_hash.le_next) {
mp2 = (np->n_hflag & NHINIT) ? np->n_mount : NFSTOMP(np);
if (mp != mp2 || np->n_fhsize != fhsize ||
bcmp(fhp, np->n_fhp, fhsize))
continue;
if (nvap && (nvap->nva_flags & NFS_FFLAG_TRIGGER_REFERRAL) &&
cnp && (cnp->cn_namelen > (fhsize - (int)sizeof(dnp)))) {
/* The name was too long to fit in the file handle. Check it against the node's name. */
int namecmp = 0;
const char *vname = vnode_getname(NFSTOV(np));
if (vname) {
if (cnp->cn_namelen != (int)strlen(vname))
namecmp = 1;
else
namecmp = strncmp(vname, cnp->cn_nameptr, cnp->cn_namelen);
vnode_putname(vname);
}
if (namecmp) /* full name didn't match */
continue;
}
FSDBG(263, dnp, np, np->n_flag, 0xcace0000);
/* if the node is locked, sleep on it */
if ((np->n_hflag & NHLOCKED) && !(flags & NG_NOCREATE)) {
np->n_hflag |= NHLOCKWANT;
FSDBG(263, dnp, np, np->n_flag, 0xcace2222);
msleep(np, nfs_node_hash_mutex, PDROP | PINOD, "nfs_nget", NULL);
FSDBG(263, dnp, np, np->n_flag, 0xcace3333);
goto loop;
}
vp = NFSTOV(np);
vid = vnode_vid(vp);
lck_mtx_unlock(nfs_node_hash_mutex);
if ((error = vnode_getwithvid(vp, vid))) {
/*
* If vnode is being reclaimed or has already
* changed identity, no need to wait.
*/
FSDBG_BOT(263, dnp, *npp, 0xcace0d1e, error);
return (error);
}
if ((error = nfs_node_lock(np))) {
/* this only fails if the node is now unhashed */
/* so let's see if we can find/create it again */
FSDBG(263, dnp, *npp, 0xcaced1e2, error);
vnode_put(vp);
if (flags & NG_NOCREATE) {
*npp = 0;
FSDBG_BOT(263, dnp, *npp, 0xcaced1e0, ENOENT);
return (ENOENT);
}
goto loop;
}
/* update attributes */
if (nvap)
error = nfs_loadattrcache(np, nvap, xidp, 0);
if (error) {
nfs_node_unlock(np);
vnode_put(vp);
} else {
if (dnp && cnp && (flags & NG_MAKEENTRY))
cache_enter(NFSTOV(dnp), vp, cnp);
/*
* Update the vnode if the name/and or the parent has
* changed. We need to do this so that if getattrlist is
* called asking for ATTR_CMN_NAME, that the "most"
* correct name is being returned. In addition for
* monitored vnodes we need to kick the vnode out of the
* name cache. We do this so that if there are hard
* links in the same directory the link will not be
* found and a lookup will get us here to return the
* name of the current link. In addition by removing the
* name from the name cache the old name will not be
* found after a rename done on another client or the
* server. The principle reason to do this is because
* Finder is asking for notifications on a directory.
* The directory changes, Finder gets notified, reads
* the directory (which we have purged) and for each
* entry returned calls getattrlist with the name
* returned from readdir. gettattrlist has to call
* namei/lookup to resolve the name, because its not in
* the cache we end up here. We need to update the name
* so Finder will get the name it called us with.
*
* We had an imperfect solution with respect to case
* sensitivity. There is a test that is run in
* FileBuster that does renames from some name to
* another name differing only in case. It then reads
* the directory looking for the new name, after it
* finds that new name, it ask gettattrlist to verify
* that the name is the new name. Usually that works,
* but renames generate fsevents and fseventsd will do a
* lookup on the name via lstat. Since that test renames
* old name to new name back and forth there is a race
* that an fsevent will be behind and will access the
* file by the old name, on a case insensitive file
* system that will work. Problem is if we do a case
* sensitive compare, we're going to change the name,
* which the test's getattrlist verification step is
* going to fail. So we will check the case sensitivity
* of the file system and do the appropriate compare. In
* a rare instance for non homogeneous file systems
* w.r.t. pathconf we will use case sensitive compares.
* That could break if the file system is actually case
* insensitive.
*
* Note that V2 does not know the case, so we just
* assume case sensitivity.
*
* This is clearly not perfect due to races, but this is
* as good as its going to get. You can defeat the
* handling of hard links simply by doing:
*
* while :; do ls -l > /dev/null; done
*
* in a terminal window. Even a single ls -l can cause a
* race.
*
* <rant>What we really need is for the caller, that
* knows the name being used is valid since it got it
* from a readdir to use that name and not ask for the
* ATTR_CMN_NAME</rant>
*/
if (dnp && cnp && (vp != NFSTOV(dnp))) {
int update_flags = (vnode_ismonitored((NFSTOV(dnp)))) ? VNODE_UPDATE_CACHE : 0;
int (*cmp)(const char *s1, const char *s2, size_t n);
cmp = nfs_case_insensitive(mp) ? strncasecmp : strncmp;
if (vp->v_name && cnp->cn_namelen && (*cmp)(cnp->cn_nameptr, vp->v_name, cnp->cn_namelen))
update_flags |= VNODE_UPDATE_NAME;
if ((vp->v_name == NULL && cnp->cn_namelen != 0) || (vp->v_name != NULL && cnp->cn_namelen == 0))
update_flags |= VNODE_UPDATE_NAME;
if (vnode_parent(vp) != NFSTOV(dnp))
update_flags |= VNODE_UPDATE_PARENT;
if (update_flags) {
NFS_NODE_DBG("vnode_update_identity old name %s new name %.*s update flags = %x\n",
vp->v_name, cnp->cn_namelen, cnp->cn_nameptr ? cnp->cn_nameptr : "", update_flags);
vnode_update_identity(vp, NFSTOV(dnp), cnp->cn_nameptr, cnp->cn_namelen, 0, update_flags);
}
}
*npp = np;
}
FSDBG_BOT(263, dnp, *npp, 0xcace0000, error);
return(error);
}
FSDBG(263, mp, dnp, npp, 0xaaaaaaaa);
if (flags & NG_NOCREATE) {
lck_mtx_unlock(nfs_node_hash_mutex);
*npp = 0;
FSDBG_BOT(263, dnp, *npp, 0x80000001, ENOENT);
return (ENOENT);
}
/*
* allocate and initialize nfsnode and stick it in the hash
* before calling getnewvnode(). Anyone finding it in the
* hash before initialization is complete will wait for it.
*/
MALLOC_ZONE(np, nfsnode_t, sizeof *np, M_NFSNODE, M_WAITOK);
if (!np) {
lck_mtx_unlock(nfs_node_hash_mutex);
*npp = 0;
FSDBG_BOT(263, dnp, *npp, 0x80000001, ENOMEM);
return (ENOMEM);
}
bzero(np, sizeof *np);
np->n_hflag |= (NHINIT | NHLOCKED);
np->n_mount = mp;
np->n_auth = auth;
TAILQ_INIT(&np->n_opens);
TAILQ_INIT(&np->n_lock_owners);
TAILQ_INIT(&np->n_locks);
np->n_dlink.tqe_next = NFSNOLIST;
np->n_dreturn.tqe_next = NFSNOLIST;
np->n_monlink.le_next = NFSNOLIST;
/* ugh... need to keep track of ".zfs" directories to workaround server bugs */
if ((nvap->nva_type == VDIR) && cnp && (cnp->cn_namelen == 4) &&
(cnp->cn_nameptr[0] == '.') && (cnp->cn_nameptr[1] == 'z') &&
(cnp->cn_nameptr[2] == 'f') && (cnp->cn_nameptr[3] == 's'))
np->n_flag |= NISDOTZFS;
if (dnp && (dnp->n_flag & NISDOTZFS))
np->n_flag |= NISDOTZFSCHILD;
if (dnp && cnp && ((cnp->cn_namelen != 2) ||
(cnp->cn_nameptr[0] != '.') || (cnp->cn_nameptr[1] != '.'))) {
vnode_t dvp = NFSTOV(dnp);
if (!vnode_get(dvp)) {
if (!vnode_ref(dvp))
np->n_parent = dvp;
vnode_put(dvp);
}
}
/* setup node's file handle */
if (fhsize > NFS_SMALLFH) {
MALLOC_ZONE(np->n_fhp, u_char *,
fhsize, M_NFSBIGFH, M_WAITOK);
if (!np->n_fhp) {
lck_mtx_unlock(nfs_node_hash_mutex);
FREE_ZONE(np, sizeof *np, M_NFSNODE);
*npp = 0;
FSDBG_BOT(263, dnp, *npp, 0x80000002, ENOMEM);
return (ENOMEM);
}
} else {
/*
* nfs statfs call
*/
static int
nfs_statfs(struct mount *mp, struct statfs *sbp)
{
struct vnode *vp;
struct thread *td;
struct nfs_statfs *sfp;
caddr_t bpos, dpos;
struct nfsmount *nmp = VFSTONFS(mp);
int error = 0, v3 = (nmp->nm_flag & NFSMNT_NFSV3), retattr;
struct mbuf *mreq, *mrep, *md, *mb;
struct nfsnode *np;
u_quad_t tquad;
td = curthread;
#ifndef nolint
sfp = NULL;
#endif
error = vfs_busy(mp, MBF_NOWAIT);
if (error)
return (error);
error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np, LK_EXCLUSIVE);
if (error) {
vfs_unbusy(mp);
return (error);
}
vp = NFSTOV(np);
mtx_lock(&nmp->nm_mtx);
if (v3 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
mtx_unlock(&nmp->nm_mtx);
(void)nfs_fsinfo(nmp, vp, td->td_ucred, td);
} else
mtx_unlock(&nmp->nm_mtx);
nfsstats.rpccnt[NFSPROC_FSSTAT]++;
mreq = nfsm_reqhead(vp, NFSPROC_FSSTAT, NFSX_FH(v3));
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(vp, v3);
nfsm_request(vp, NFSPROC_FSSTAT, td, td->td_ucred);
if (v3)
nfsm_postop_attr(vp, retattr);
if (error) {
if (mrep != NULL)
m_freem(mrep);
goto nfsmout;
}
sfp = nfsm_dissect(struct nfs_statfs *, NFSX_STATFS(v3));
mtx_lock(&nmp->nm_mtx);
sbp->f_iosize = nfs_iosize(nmp);
mtx_unlock(&nmp->nm_mtx);
if (v3) {
sbp->f_bsize = NFS_FABLKSIZE;
tquad = fxdr_hyper(&sfp->sf_tbytes);
sbp->f_blocks = tquad / NFS_FABLKSIZE;
tquad = fxdr_hyper(&sfp->sf_fbytes);
sbp->f_bfree = tquad / NFS_FABLKSIZE;
tquad = fxdr_hyper(&sfp->sf_abytes);
sbp->f_bavail = tquad / NFS_FABLKSIZE;
sbp->f_files = (fxdr_unsigned(int32_t,
sfp->sf_tfiles.nfsuquad[1]) & 0x7fffffff);
sbp->f_ffree = (fxdr_unsigned(int32_t,
sfp->sf_ffiles.nfsuquad[1]) & 0x7fffffff);
} else {
sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
sbp->f_files = 0;
sbp->f_ffree = 0;
}
m_freem(mrep);
nfsmout:
vput(vp);
vfs_unbusy(mp);
return (error);
}
/*
* Common code for mount and mountroot
*/
static int
mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam,
char *hst, struct vnode **vpp, struct ucred *cred, int nametimeo,
int negnametimeo)
{
struct nfsmount *nmp;
struct nfsnode *np;
int error;
struct vattr attrs;
if (mp->mnt_flag & MNT_UPDATE) {
nmp = VFSTONFS(mp);
printf("%s: MNT_UPDATE is no longer handled here\n", __func__);
free(nam, M_SONAME);
return (0);
} else {
nmp = uma_zalloc(nfsmount_zone, M_WAITOK);
bzero((caddr_t)nmp, sizeof (struct nfsmount));
TAILQ_INIT(&nmp->nm_bufq);
mp->mnt_data = nmp;
nmp->nm_getinfo = nfs_getnlminfo;
nmp->nm_vinvalbuf = nfs_vinvalbuf;
}
vfs_getnewfsid(mp);
nmp->nm_mountp = mp;
mtx_init(&nmp->nm_mtx, "NFSmount lock", NULL, MTX_DEF);
/*
* V2 can only handle 32 bit filesizes. A 4GB-1 limit may be too
* high, depending on whether we end up with negative offsets in
* the client or server somewhere. 2GB-1 may be safer.
*
* For V3, nfs_fsinfo will adjust this as necessary. Assume maximum
* that we can handle until we find out otherwise.
*/
if ((argp->flags & NFSMNT_NFSV3) == 0)
nmp->nm_maxfilesize = 0xffffffffLL;
else
nmp->nm_maxfilesize = OFF_MAX;
nmp->nm_timeo = NFS_TIMEO;
nmp->nm_retry = NFS_RETRANS;
if ((argp->flags & NFSMNT_NFSV3) && argp->sotype == SOCK_STREAM) {
nmp->nm_wsize = nmp->nm_rsize = NFS_MAXDATA;
} else {
nmp->nm_wsize = NFS_WSIZE;
nmp->nm_rsize = NFS_RSIZE;
}
nmp->nm_wcommitsize = hibufspace / (desiredvnodes / 1000);
nmp->nm_readdirsize = NFS_READDIRSIZE;
nmp->nm_numgrps = NFS_MAXGRPS;
nmp->nm_readahead = NFS_DEFRAHEAD;
nmp->nm_deadthresh = NFS_MAXDEADTHRESH;
nmp->nm_nametimeo = nametimeo;
nmp->nm_negnametimeo = negnametimeo;
nmp->nm_tprintf_delay = nfs_tprintf_delay;
if (nmp->nm_tprintf_delay < 0)
nmp->nm_tprintf_delay = 0;
nmp->nm_tprintf_initial_delay = nfs_tprintf_initial_delay;
if (nmp->nm_tprintf_initial_delay < 0)
nmp->nm_tprintf_initial_delay = 0;
nmp->nm_fhsize = argp->fhsize;
bcopy((caddr_t)argp->fh, (caddr_t)nmp->nm_fh, argp->fhsize);
bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN);
nmp->nm_nam = nam;
/* Set up the sockets and per-host congestion */
nmp->nm_sotype = argp->sotype;
nmp->nm_soproto = argp->proto;
nmp->nm_rpcops = &nfs_rpcops;
nfs_decode_args(mp, nmp, argp, hst);
/*
* For Connection based sockets (TCP,...) defer the connect until
* the first request, in case the server is not responding.
*/
if (nmp->nm_sotype == SOCK_DGRAM &&
(error = nfs_connect(nmp)))
goto bad;
/*
* This is silly, but it has to be set so that vinifod() works.
* We do not want to do an nfs_statfs() here since we can get
* stuck on a dead server and we are holding a lock on the mount
* point.
*/
mtx_lock(&nmp->nm_mtx);
mp->mnt_stat.f_iosize = nfs_iosize(nmp);
mtx_unlock(&nmp->nm_mtx);
/*
* A reference count is needed on the nfsnode representing the
* remote root. If this object is not persistent, then backward
* traversals of the mount point (i.e. "..") will not work if
* the nfsnode gets flushed out of the cache. Ufs does not have
* this problem, because one can identify root inodes by their
* number == ROOTINO (2).
*/
error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np, LK_EXCLUSIVE);
if (error)
goto bad;
*vpp = NFSTOV(np);
/*
* Get file attributes and transfer parameters for the
* mountpoint. This has the side effect of filling in
* (*vpp)->v_type with the correct value.
*/
if (argp->flags & NFSMNT_NFSV3)
nfs_fsinfo(nmp, *vpp, curthread->td_ucred, curthread);
else
VOP_GETATTR(*vpp, &attrs, curthread->td_ucred);
/*
* Lose the lock but keep the ref.
*/
VOP_UNLOCK(*vpp, 0);
return (0);
bad:
nfs_disconnect(nmp);
mtx_destroy(&nmp->nm_mtx);
uma_zfree(nfsmount_zone, nmp);
free(nam, M_SONAME);
return (error);
}
/*
* Look up a vnode/nfsnode by file handle and store the pointer in *npp.
* Callers must check for mount points!!
* An error number is returned.
*/
int
nfs_nget(struct mount *mnt, nfsfh_t *fh, int fhsize, struct nfsnode **npp)
{
struct nfsmount *nmp;
struct nfsnode *np, find, *np2;
struct vnode *vp, *nvp;
struct proc *p = curproc; /* XXX */
int error;
nmp = VFSTONFS(mnt);
loop:
rw_enter_write(&nfs_hashlock);
find.n_fhp = fh;
find.n_fhsize = fhsize;
np = RB_FIND(nfs_nodetree, &nmp->nm_ntree, &find);
if (np != NULL) {
rw_exit_write(&nfs_hashlock);
vp = NFSTOV(np);
error = vget(vp, LK_EXCLUSIVE, p);
if (error)
goto loop;
*npp = np;
return (0);
}
/*
* getnewvnode() could recycle a vnode, potentially formerly
* owned by NFS. This will cause a VOP_RECLAIM() to happen,
* which will cause recursive locking, so we unlock before
* calling getnewvnode() lock again afterwards, but must check
* to see if this nfsnode has been added while we did not hold
* the lock.
*/
rw_exit_write(&nfs_hashlock);
error = getnewvnode(VT_NFS, mnt, &nfs_vops, &nvp);
/* note that we don't have this vnode set up completely yet */
rw_enter_write(&nfs_hashlock);
if (error) {
*npp = NULL;
rw_exit_write(&nfs_hashlock);
return (error);
}
nvp->v_flag |= VLARVAL;
np = RB_FIND(nfs_nodetree, &nmp->nm_ntree, &find);
if (np != NULL) {
vgone(nvp);
rw_exit_write(&nfs_hashlock);
goto loop;
}
vp = nvp;
np = pool_get(&nfs_node_pool, PR_WAITOK | PR_ZERO);
vp->v_data = np;
/* we now have an nfsnode on this vnode */
vp->v_flag &= ~VLARVAL;
np->n_vnode = vp;
rw_init(&np->n_commitlock, "nfs_commitlk");
/*
* Are we getting the root? If so, make sure the vnode flags
* are correct
*/
if ((fhsize == nmp->nm_fhsize) && !bcmp(fh, nmp->nm_fh, fhsize)) {
if (vp->v_type == VNON)
vp->v_type = VDIR;
vp->v_flag |= VROOT;
}
np->n_fhp = &np->n_fh;
bcopy(fh, np->n_fhp, fhsize);
np->n_fhsize = fhsize;
np2 = RB_INSERT(nfs_nodetree, &nmp->nm_ntree, np);
KASSERT(np2 == NULL);
np->n_accstamp = -1;
rw_exit(&nfs_hashlock);
*npp = np;
return (0);
}
/*
* Look up a vnode/nfsnode by file handle.
* Callers must check for mount points!!
* In all cases, a pointer to a
* nfsnode structure is returned.
*/
int
nfs_nget1(struct mount *mntp, nfsfh_t *fhp, int fhsize, struct nfsnode **npp,
int lkflags)
{
struct nfsnode *np;
struct vnode *vp;
struct nfsmount *nmp = VFSTONFS(mntp);
int error;
struct fh_match fhm;
fhm.fhm_fhp = fhp;
fhm.fhm_fhsize = fhsize;
loop:
rw_enter(&nmp->nm_rbtlock, RW_READER);
np = rb_tree_find_node(&nmp->nm_rbtree, &fhm);
if (np != NULL) {
vp = NFSTOV(np);
mutex_enter(vp->v_interlock);
rw_exit(&nmp->nm_rbtlock);
error = vget(vp, LK_EXCLUSIVE | lkflags);
if (error == EBUSY)
return error;
if (error)
goto loop;
*npp = np;
return(0);
}
rw_exit(&nmp->nm_rbtlock);
error = getnewvnode(VT_NFS, mntp, nfsv2_vnodeop_p, NULL, &vp);
if (error) {
*npp = 0;
return (error);
}
np = pool_get(&nfs_node_pool, PR_WAITOK);
memset(np, 0, sizeof *np);
np->n_vnode = vp;
/*
* Insert the nfsnode in the hash queue for its new file handle
*/
if (fhsize > NFS_SMALLFH) {
np->n_fhp = kmem_alloc(fhsize, KM_SLEEP);
} else
np->n_fhp = &np->n_fh;
memcpy(np->n_fhp, fhp, fhsize);
np->n_fhsize = fhsize;
np->n_accstamp = -1;
np->n_vattr = pool_get(&nfs_vattr_pool, PR_WAITOK);
rw_enter(&nmp->nm_rbtlock, RW_WRITER);
if (NULL != rb_tree_find_node(&nmp->nm_rbtree, &fhm)) {
rw_exit(&nmp->nm_rbtlock);
if (fhsize > NFS_SMALLFH) {
kmem_free(np->n_fhp, fhsize);
}
pool_put(&nfs_vattr_pool, np->n_vattr);
pool_put(&nfs_node_pool, np);
ungetnewvnode(vp);
goto loop;
}
vp->v_data = np;
genfs_node_init(vp, &nfs_genfsops);
/*
* Initalize read/write creds to useful values. VOP_OPEN will
* overwrite these.
*/
np->n_rcred = curlwp->l_cred;
kauth_cred_hold(np->n_rcred);
np->n_wcred = curlwp->l_cred;
kauth_cred_hold(np->n_wcred);
VOP_LOCK(vp, LK_EXCLUSIVE);
NFS_INVALIDATE_ATTRCACHE(np);
uvm_vnp_setsize(vp, 0);
(void)rb_tree_insert_node(&nmp->nm_rbtree, np);
rw_exit(&nmp->nm_rbtlock);
*npp = np;
return (0);
}
