/* $NetBSD: nfs_clrpcops.c,v 1.5 2025/02/27 08:39:53 andvar Exp $ */
/*-
* Copyright (c) 1989, 1993
* The Regents of the University of California. All rights reserved.
*
* This code is derived from software contributed to Berkeley by
* Rick Macklem at The University of Guelph.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 4. Neither the name of the University nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
*/
/*
* Rpc op calls, generally called from the vnode op calls or through the
* buffer cache, for NFS v2, 3 and 4.
* These do not normally make any changes to vnode arguments or use
* structures that might change between the VFS variants. The returned
* arguments are all at the end, after the NFSPROC_T *p one.
*/
/*
* nfscl_getsameserver() can return one of three values:
* NFSDSP_USETHISSESSION - Use this session for the DS.
* NFSDSP_SEQTHISSESSION - Use the nfsclds_sequence field of this dsp for new
* session.
* NFSDSP_NOTFOUND - No matching server was found.
*/
enum nfsclds_state {
NFSDSP_USETHISSESSION = 0,
NFSDSP_SEQTHISSESSION = 1,
NFSDSP_NOTFOUND = 2,
};
/*
* nfs access rpc op.
* For nfs version 3 and 4, use the access rpc to check accessibility. If file
* modes are changed on the server, accesses might still fail later.
*/
APPLESTATIC int
nfsrpc_access(vnode_t vp, int acmode, struct ucred *cred,
NFSPROC_T *p, struct nfsvattr *nap, int *attrflagp)
{
int error;
u_int32_t mode, rmode;
if (acmode & VREAD)
mode = NFSACCESS_READ;
else
mode = 0;
if (vnode_vtype(vp) == VDIR) {
if (acmode & VWRITE)
mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND |
NFSACCESS_DELETE);
if (acmode & VEXEC)
mode |= NFSACCESS_LOOKUP;
} else {
if (acmode & VWRITE)
mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
if (acmode & VEXEC)
mode |= NFSACCESS_EXECUTE;
}
/*
* Now, just call nfsrpc_accessrpc() to do the actual RPC.
*/
error = nfsrpc_accessrpc(vp, mode, cred, p, nap, attrflagp, &rmode,
NULL);
/*
* The NFS V3 spec does not clarify whether or not
* the returned access bits can be a superset of
* the ones requested, so...
*/
if (!error && (rmode & mode) != mode)
error = EACCES;
return (error);
}
/*
* The actual rpc, separated out for Darwin.
*/
APPLESTATIC int
nfsrpc_accessrpc(vnode_t vp, u_int32_t mode, struct ucred *cred,
NFSPROC_T *p, struct nfsvattr *nap, int *attrflagp, u_int32_t *rmodep,
void *stuff)
{
u_int32_t *tl;
u_int32_t supported, rmode;
int error;
struct nfsrv_descript nfsd, *nd = &nfsd;
nfsattrbit_t attrbits;
/*
* It's not obvious what should be done about
* unsupported access modes. For now, be paranoid
* and clear the unsupported ones.
*/
rmode &= supported;
*rmodep = rmode;
} else
error = nd->nd_repstat;
nfsmout:
mbuf_freem(nd->nd_mrep);
return (error);
}
/*
* For NFSv4, Open Ops are only done on Regular Files.
*/
if (vnode_vtype(vp) != VREG)
return (0);
mode = 0;
if (amode & FREAD)
mode |= NFSV4OPEN_ACCESSREAD;
if (amode & FWRITE)
mode |= NFSV4OPEN_ACCESSWRITE;
nfhp = np->n_fhp;
retrycnt = 0;
#ifdef notdef
{ char name[100]; int namel;
namel = (np->n_v4->n4_namelen < 100) ? np->n_v4->n4_namelen : 99;
bcopy(NFS4NODENAME(np->n_v4), name, namel);
name[namel] = '\0';
printf("rpcopen p=0x%x name=%s",p->p_pid,name);
if (nfhp->nfh_len > 0) printf(" fh=0x%x\n",nfhp->nfh_fh[12]);
else printf(" fhl=0\n");
}
#endif
do {
dp = NULL;
error = nfscl_open(vp, nfhp->nfh_fh, nfhp->nfh_len, mode, 1,
cred, p, NULL, &op, &newone, &ret, 1);
if (error) {
return (error);
}
if (nmp->nm_clp != NULL)
clidrev = nmp->nm_clp->nfsc_clientidrev;
else
clidrev = 0;
if (ret == NFSCLOPEN_DOOPEN) {
if (np->n_v4 != NULL) {
error = nfsrpc_openrpc(nmp, vp, np->n_v4->n4_data,
np->n_v4->n4_fhlen, np->n_fhp->nfh_fh,
np->n_fhp->nfh_len, mode, op,
NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, &dp,
0, 0x0, cred, p, 0, 0);
if (dp != NULL) {
#ifdef APPLE
OSBitAndAtomic((int32_t)~NDELEGMOD, (UInt32 *)&np->n_flag);
#else
NFSLOCKNODE(np);
np->n_flag &= ~NDELEGMOD;
/*
* Invalidate the attribute cache, so that
* attributes that pre-date the issue of a
* delegation are not cached, since the
* cached attributes will remain valid while
* the delegation is held.
*/
NFSINVALATTRCACHE(np);
NFSUNLOCKNODE(np);
#endif
(void) nfscl_deleg(nmp->nm_mountp,
op->nfso_own->nfsow_clp,
nfhp->nfh_fh, nfhp->nfh_len, cred, p, &dp);
}
} else {
error = EIO;
}
newnfs_copyincred(cred, &op->nfso_cred);
} else if (ret == NFSCLOPEN_SETCRED)
/*
* This is a new local open on a delegation. It needs
* to have credentials so that an open can be done
* against the server during recovery.
*/
newnfs_copyincred(cred, &op->nfso_cred);
/*
* nfso_opencnt is the count of how many VOP_OPEN()s have
* been done on this Open successfully and a VOP_CLOSE()
* is expected for each of these.
* If error is non-zero, don't increment it, since the Open
* hasn't succeeded yet.
*/
if (!error)
op->nfso_opencnt++;
nfscl_openrelease(op, error, newone);
if (error == NFSERR_GRACE || error == NFSERR_STALECLIENTID ||
error == NFSERR_STALEDONTRECOVER || error == NFSERR_DELAY ||
error == NFSERR_BADSESSION) {
(void) nfs_catnap(PZERO, error, "nfs_open");
} else if ((error == NFSERR_EXPIRED || error == NFSERR_BADSTATEID)
&& clidrev != 0) {
expireret = nfscl_hasexpired(nmp->nm_clp, clidrev, p);
retrycnt++;
}
} while (error == NFSERR_GRACE || error == NFSERR_STALECLIENTID ||
error == NFSERR_STALEDONTRECOVER || error == NFSERR_DELAY ||
error == NFSERR_BADSESSION ||
((error == NFSERR_EXPIRED || error == NFSERR_BADSTATEID) &&
expireret == 0 && clidrev != 0 && retrycnt < 4));
if (error && retrycnt >= 4)
error = EIO;
return (error);
}
/*
* the actual open rpc
*/
APPLESTATIC int
nfsrpc_openrpc(struct nfsmount *nmp, vnode_t vp, u_int8_t *nfhp, int fhlen,
u_int8_t *newfhp, int newfhlen, u_int32_t mode, struct nfsclopen *op,
u_int8_t *name, int namelen, struct nfscldeleg **dpp,
int reclaim, u_int32_t delegtype, struct ucred *cred, NFSPROC_T *p,
int syscred, int recursed)
{
u_int32_t *tl;
struct nfsrv_descript nfsd, *nd = &nfsd;
struct nfscldeleg *dp, *ndp = NULL;
struct nfsvattr nfsva;
u_int32_t rflags, deleg;
nfsattrbit_t attrbits;
int error, ret, acesize, limitby;
/*
* If the server is handing out delegations, but we didn't
* get one because an OpenConfirm was required, try the
* Open again, to get a delegation. This is a harmless no-op,
* from a server's point of view.
*/
if (!reclaim && (rflags & NFSV4OPEN_RESULTCONFIRM) &&
(op->nfso_own->nfsow_clp->nfsc_flags & NFSCLFLAGS_GOTDELEG)
&& !error && dp == NULL && ndp == NULL && !recursed) {
do {
ret = nfsrpc_openrpc(nmp, vp, nfhp, fhlen, newfhp,
newfhlen, mode, op, name, namelen, &ndp, 0, 0x0,
cred, p, syscred, 1);
if (ret == NFSERR_DELAY)
(void) nfs_catnap(PZERO, ret, "nfs_open2");
} while (ret == NFSERR_DELAY);
if (ret) {
if (ndp != NULL) {
FREE((caddr_t)ndp, M_NFSCLDELEG);
ndp = NULL;
}
if (ret == NFSERR_STALECLIENTID ||
ret == NFSERR_STALEDONTRECOVER ||
ret == NFSERR_BADSESSION)
error = ret;
}
}
}
if (nd->nd_repstat != 0 && error == 0)
error = nd->nd_repstat;
if (error == NFSERR_STALECLIENTID || error == NFSERR_BADSESSION)
nfscl_initiate_recovery(op->nfso_own->nfsow_clp);
nfsmout:
if (!error)
*dpp = ndp;
else if (ndp != NULL)
FREE((caddr_t)ndp, M_NFSCLDELEG);
mbuf_freem(nd->nd_mrep);
return (error);
}
/*
* V4 Close operation.
*/
APPLESTATIC int
nfsrpc_close(vnode_t vp, int doclose, NFSPROC_T *p)
{
struct nfsclclient *clp;
int error;
if (vnode_vtype(vp) != VREG)
return (0);
if (doclose)
error = nfscl_doclose(vp, &clp, p);
else
error = nfscl_getclose(vp, &clp);
if (error)
return (error);
nfscl_clientrelease(clp);
return (0);
}
/*
* Close the open.
*/
APPLESTATIC void
nfsrpc_doclose(struct nfsmount *nmp, struct nfsclopen *op, NFSPROC_T *p)
{
struct nfsrv_descript nfsd, *nd = &nfsd;
struct nfscllockowner *lp, *nlp;
struct nfscllock *lop, *nlop;
struct ucred *tcred;
u_int64_t off = 0, len = 0;
u_int32_t type = NFSV4LOCKT_READ;
int error, do_unlock, trycnt;
tcred = newnfs_getcred();
newnfs_copycred(&op->nfso_cred, tcred);
/*
* (Theoretically this could be done in the same
* compound as the close, but having multiple
* sequenced Ops in the same compound might be
* too scary for some servers.)
*/
if (op->nfso_posixlock) {
off = 0;
len = NFS64BITSSET;
type = NFSV4LOCKT_READ;
}
/*
* Since this function is only called from VOP_INACTIVE(), no
* other thread will be manipulating this Open. As such, the
* lock lists are not being changed by other threads, so it should
* be safe to do this without locking.
*/
LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
do_unlock = 1;
LIST_FOREACH_SAFE(lop, &lp->nfsl_lock, nfslo_list, nlop) {
if (op->nfso_posixlock == 0) {
off = lop->nfslo_first;
len = lop->nfslo_end - lop->nfslo_first;
if (lop->nfslo_type == F_WRLCK)
type = NFSV4LOCKT_WRITE;
else
type = NFSV4LOCKT_READ;
}
if (do_unlock) {
trycnt = 0;
do {
error = nfsrpc_locku(nd, nmp, lp, off,
len, type, tcred, p, 0);
if ((nd->nd_repstat == NFSERR_GRACE ||
nd->nd_repstat == NFSERR_DELAY) &&
error == 0)
(void) nfs_catnap(PZERO,
(int)nd->nd_repstat,
"nfs_close");
} while ((nd->nd_repstat == NFSERR_GRACE ||
nd->nd_repstat == NFSERR_DELAY) &&
error == 0 && trycnt++ < 5);
if (op->nfso_posixlock)
do_unlock = 0;
}
nfscl_freelock(lop, 0);
}
/*
* Do a ReleaseLockOwner.
* The lock owner name nfsl_owner may be used by other opens for
* other files but the lock_owner4 name that nfsrpc_rellockown()
* puts on the wire has the file handle for this file appended
* to it, so it can be done now.
*/
(void)nfsrpc_rellockown(nmp, lp, lp->nfsl_open->nfso_fh,
lp->nfsl_open->nfso_fhlen, tcred, p);
}
/*
* There could be other Opens for different files on the same
* OpenOwner, so locking is required.
*/
NFSLOCKCLSTATE();
nfscl_lockexcl(&op->nfso_own->nfsow_rwlock, NFSCLSTATEMUTEXPTR);
NFSUNLOCKCLSTATE();
do {
error = nfscl_tryclose(op, tcred, nmp, p);
if (error == NFSERR_GRACE)
(void) nfs_catnap(PZERO, error, "nfs_close");
} while (error == NFSERR_GRACE);
NFSLOCKCLSTATE();
nfscl_lockunlock(&op->nfso_own->nfsow_rwlock);
/*
* The actual Close RPC.
*/
APPLESTATIC int
nfsrpc_closerpc(struct nfsrv_descript *nd, struct nfsmount *nmp,
struct nfsclopen *op, struct ucred *cred, NFSPROC_T *p,
int syscred)
{
u_int32_t *tl;
int error;
/*
* Allocate a single session structure for NFSv4.0, because some of
* the fields are used by NFSv4.0 although it doesn't do a session.
*/
dsp = malloc(sizeof(struct nfsclds), M_NFSCLDS, M_WAITOK | M_ZERO);
mtx_init(&dsp->nfsclds_mtx, "nfsds", NULL, MTX_DEF);
mtx_init(&dsp->nfsclds_sess.nfsess_mtx, "nfssession", NULL, MTX_DEF);
NFSLOCKMNT(nmp);
TAILQ_INSERT_HEAD(&nmp->nm_sess, dsp, nfsclds_list);
NFSUNLOCKMNT(nmp);
/*
* Do a readlink rpc.
*/
APPLESTATIC int
nfsrpc_readlink(vnode_t vp, struct uio *uiop, struct ucred *cred,
NFSPROC_T *p, struct nfsvattr *nap, int *attrflagp, void *stuff)
{
u_int32_t *tl;
struct nfsrv_descript nfsd, *nd = &nfsd;
struct nfsnode *np = VTONFS(vp);
nfsattrbit_t attrbits;
int error, len, cangetattr = 1;
*attrflagp = 0;
NFSCL_REQSTART(nd, NFSPROC_READLINK, vp);
if (nd->nd_flag & ND_NFSV4) {
/*
* And do a Getattr op.
*/
NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
*tl = txdr_unsigned(NFSV4OP_GETATTR);
NFSGETATTR_ATTRBIT(&attrbits);
(void) nfsrv_putattrbit(nd, &attrbits);
}
error = nfscl_request(nd, vp, p, cred, stuff);
if (error)
return (error);
if (nd->nd_flag & ND_NFSV3)
error = nfscl_postop_attr(nd, nap, attrflagp, stuff);
if (!nd->nd_repstat && !error) {
NFSM_STRSIZ(len, NFS_MAXPATHLEN);
/*
* This seems weird to me, but must have been added to
* FreeBSD for some reason. The only thing I can think of
* is that there was/is some server that replies with
* more link data than it should?
*/
if (len == NFS_MAXPATHLEN) {
NFSLOCKNODE(np);
if (np->n_size > 0 && np->n_size < NFS_MAXPATHLEN) {
len = np->n_size;
cangetattr = 0;
}
NFSUNLOCKNODE(np);
}
error = nfsm_mbufuio(nd, uiop, len);
if ((nd->nd_flag & ND_NFSV4) && !error && cangetattr)
error = nfscl_postop_attr(nd, nap, attrflagp, stuff);
}
if (nd->nd_repstat && !error)
error = nd->nd_repstat;
nfsmout:
mbuf_freem(nd->nd_mrep);
return (error);
}
/*
* The actual read RPC.
*/
static int
nfsrpc_readrpc(vnode_t vp, struct uio *uiop, struct ucred *cred,
nfsv4stateid_t *stateidp, NFSPROC_T *p, struct nfsvattr *nap,
int *attrflagp, void *stuff)
{
u_int32_t *tl;
int error = 0, len, retlen, tsiz, eof = 0;
struct nfsrv_descript nfsd;
struct nfsmount *nmp = VFSTONFS(vnode_mount(vp));
struct nfsrv_descript *nd = &nfsd;
int rsize;
off_t tmp_off;
*attrflagp = 0;
tsiz = uio_uio_resid(uiop);
tmp_off = uiop->uio_offset + tsiz;
NFSLOCKMNT(nmp);
if (tmp_off > nmp->nm_maxfilesize || tmp_off < uiop->uio_offset) {
NFSUNLOCKMNT(nmp);
return (EFBIG);
}
rsize = nmp->nm_rsize;
NFSUNLOCKMNT(nmp);
nd->nd_mrep = NULL;
while (tsiz > 0) {
*attrflagp = 0;
len = (tsiz > rsize) ? rsize : tsiz;
NFSCL_REQSTART(nd, NFSPROC_READ, vp);
if (nd->nd_flag & ND_NFSV4)
nfsm_stateidtom(nd, stateidp, NFSSTATEID_PUTSTATEID);
NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED * 3);
if (nd->nd_flag & ND_NFSV2) {
*tl++ = txdr_unsigned(uiop->uio_offset);
*tl++ = txdr_unsigned(len);
*tl = 0;
} else {
txdr_hyper(uiop->uio_offset, tl);
*(tl + 2) = txdr_unsigned(len);
}
/*
* Since I can't do a Getattr for NFSv4 for Write, there
* doesn't seem any point in doing one here, either.
* (See the comment in nfsrpc_writerpc() for more info.)
*/
error = nfscl_request(nd, vp, p, cred, stuff);
if (error)
return (error);
if (nd->nd_flag & ND_NFSV3) {
error = nfscl_postop_attr(nd, nap, attrflagp, stuff);
} else if (!nd->nd_repstat && (nd->nd_flag & ND_NFSV2)) {
error = nfsm_loadattr(nd, nap);
if (!error)
*attrflagp = 1;
}
if (nd->nd_repstat || error) {
if (!error)
error = nd->nd_repstat;
goto nfsmout;
}
if (nd->nd_flag & ND_NFSV3) {
NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
eof = fxdr_unsigned(int, *(tl + 1));
} else if (nd->nd_flag & ND_NFSV4) {
NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
eof = fxdr_unsigned(int, *tl);
}
NFSM_STRSIZ(retlen, len);
error = nfsm_mbufuio(nd, uiop, retlen);
if (error)
goto nfsmout;
mbuf_freem(nd->nd_mrep);
nd->nd_mrep = NULL;
tsiz -= retlen;
if (!(nd->nd_flag & ND_NFSV2)) {
if (eof || retlen == 0)
tsiz = 0;
} else if (retlen < len)
tsiz = 0;
}
return (0);
nfsmout:
mbuf_freem(nd->nd_mrep);
return (error);
}
/*
* nfs write operation
* When called_from_strategy != 0, it should return EIO for an error that
* indicates recovery is in progress, so that the buffer will be left
* dirty and be written back to the server later. If it loops around,
* the recovery thread could get stuck waiting for the buffer and recovery
* will then deadlock.
*/
APPLESTATIC int
nfsrpc_write(vnode_t vp, struct uio *uiop, int *iomode, int *must_commit,
struct ucred *cred, NFSPROC_T *p, struct nfsvattr *nap, int *attrflagp,
void *stuff, int called_from_strategy)
{
int error, expireret = 0, retrycnt, nostateid;
u_int32_t clidrev = 0;
struct nfsmount *nmp = VFSTONFS(vnode_mount(vp));
struct nfsnode *np = VTONFS(vp);
struct ucred *newcred;
struct nfsfh *nfhp = NULL;
nfsv4stateid_t stateid;
void *lckp;
/*
* If there is no stateid for NFSv4, it means this is an
* extraneous write after close. Basically a poorly
* implemented buffer cache. Just don't do the write.
*/
if (nostateid)
error = 0;
else
error = nfsrpc_writerpc(vp, uiop, iomode, must_commit,
newcred, &stateid, p, nap, attrflagp, stuff);
if (error == NFSERR_STALESTATEID || error == NFSERR_BADSESSION)
nfscl_initiate_recovery(nmp->nm_clp);
if (lckp != NULL)
nfscl_lockderef(lckp);
if (error == NFSERR_GRACE || error == NFSERR_STALESTATEID ||
error == NFSERR_STALEDONTRECOVER || error == NFSERR_DELAY ||
error == NFSERR_OLDSTATEID || error == NFSERR_BADSESSION) {
(void) nfs_catnap(PZERO, error, "nfs_write");
} else if ((error == NFSERR_EXPIRED ||
error == NFSERR_BADSTATEID) && clidrev != 0) {
expireret = nfscl_hasexpired(nmp->nm_clp, clidrev, p);
}
retrycnt++;
} while (error == NFSERR_GRACE || error == NFSERR_DELAY ||
((error == NFSERR_STALESTATEID || error == NFSERR_BADSESSION ||
error == NFSERR_STALEDONTRECOVER) && called_from_strategy == 0) ||
(error == NFSERR_OLDSTATEID && retrycnt < 20) ||
((error == NFSERR_EXPIRED || error == NFSERR_BADSTATEID) &&
expireret == 0 && clidrev != 0 && retrycnt < 4));
if (error != 0 && (retrycnt >= 4 ||
((error == NFSERR_STALESTATEID || error == NFSERR_BADSESSION ||
error == NFSERR_STALEDONTRECOVER) && called_from_strategy != 0)))
error = EIO;
if (NFSHASNFSV4(nmp))
NFSFREECRED(newcred);
return (error);
}
/*
* The actual write RPC.
*/
static int
nfsrpc_writerpc(vnode_t vp, struct uio *uiop, int *iomode,
int *must_commit, struct ucred *cred, nfsv4stateid_t *stateidp,
NFSPROC_T *p, struct nfsvattr *nap, int *attrflagp, void *stuff)
{
u_int32_t *tl;
struct nfsmount *nmp = VFSTONFS(vnode_mount(vp));
struct nfsnode *np = VTONFS(vp);
int error = 0, len, tsiz, rlen, commit, committed = NFSWRITE_FILESYNC;
int wccflag = 0, wsize;
int32_t backup;
struct nfsrv_descript nfsd;
struct nfsrv_descript *nd = &nfsd;
nfsattrbit_t attrbits;
off_t tmp_off;
KASSERT(uiop->uio_iovcnt == 1, ("nfs: writerpc iovcnt > 1"));
*attrflagp = 0;
tsiz = uio_uio_resid(uiop);
tmp_off = uiop->uio_offset + tsiz;
NFSLOCKMNT(nmp);
if (tmp_off > nmp->nm_maxfilesize || tmp_off < uiop->uio_offset) {
NFSUNLOCKMNT(nmp);
return (EFBIG);
}
wsize = nmp->nm_wsize;
NFSUNLOCKMNT(nmp);
nd->nd_mrep = NULL; /* NFSv2 sometimes does a write with */
nd->nd_repstat = 0; /* uio_resid == 0, so the while is not done */
while (tsiz > 0) {
*attrflagp = 0;
len = (tsiz > wsize) ? wsize : tsiz;
NFSCL_REQSTART(nd, NFSPROC_WRITE, vp);
if (nd->nd_flag & ND_NFSV4) {
nfsm_stateidtom(nd, stateidp, NFSSTATEID_PUTSTATEID);
NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER+2*NFSX_UNSIGNED);
txdr_hyper(uiop->uio_offset, tl);
tl += 2;
*tl++ = txdr_unsigned(*iomode);
*tl = txdr_unsigned(len);
} else if (nd->nd_flag & ND_NFSV3) {
NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER+3*NFSX_UNSIGNED);
txdr_hyper(uiop->uio_offset, tl);
tl += 2;
*tl++ = txdr_unsigned(len);
*tl++ = txdr_unsigned(*iomode);
*tl = txdr_unsigned(len);
} else {
u_int32_t x;
NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
/*
* Not sure why someone changed this, since the
* RFC clearly states that "beginoffset" and
* "totalcount" are ignored, but it wouldn't
* surprise me if there's a busted server out there.
*/
/* Set both "begin" and "current" to non-garbage. */
x = txdr_unsigned((u_int32_t)uiop->uio_offset);
*tl++ = x; /* "begin offset" */
*tl++ = x; /* "current offset" */
x = txdr_unsigned(len);
*tl++ = x; /* total to this offset */
*tl = x; /* size of this write */
}
nfsm_uiombuf(nd, uiop, len);
/*
* Although it is tempting to do a normal Getattr Op in the
* NFSv4 compound, the result can be a nearly hung client
* system if the Getattr asks for Owner and/or OwnerGroup.
* It occurs when the client can't map either the Owner or
* Owner_group name in the Getattr reply to a uid/gid. When
* there is a cache miss, the kernel does an upcall to the
* nfsuserd. Then, it can try and read the local /etc/passwd
* or /etc/group file. It can then block in getnewbuf(),
* waiting for dirty writes to be pushed to the NFS server.
* The only reason this doesn't result in a complete
* deadlock, is that the upcall times out and allows
* the write to complete. However, progress is so slow
* that it might just as well be deadlocked.
* As such, we get the rest of the attributes, but not
* Owner or Owner_group.
* nb: nfscl_loadattrcache() needs to be told that these
* partial attributes from a write rpc are being
* passed in, via a argument flag.
*/
if (nd->nd_flag & ND_NFSV4) {
NFSWRITEGETATTR_ATTRBIT(&attrbits);
NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
*tl = txdr_unsigned(NFSV4OP_GETATTR);
(void) nfsrv_putattrbit(nd, &attrbits);
}
error = nfscl_request(nd, vp, p, cred, stuff);
if (error)
return (error);
if (nd->nd_repstat) {
/*
* In case the rpc gets retried, roll
* the uio fileds changed by nfsm_uiombuf()
* back.
*/
uiop->uio_offset -= len;
uio_uio_resid_add(uiop, len);
uio_iov_base_add(uiop, -len);
uio_iov_len_add(uiop, len);
}
if (nd->nd_flag & (ND_NFSV3 | ND_NFSV4)) {
error = nfscl_wcc_data(nd, vp, nap, attrflagp,
&wccflag, stuff);
if (error)
goto nfsmout;
}
if (!nd->nd_repstat) {
if (nd->nd_flag & (ND_NFSV3 | ND_NFSV4)) {
NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED
+ NFSX_VERF);
rlen = fxdr_unsigned(int, *tl++);
if (rlen == 0) {
error = NFSERR_IO;
goto nfsmout;
} else if (rlen < len) {
backup = len - rlen;
uio_iov_base_add(uiop, -(backup));
uio_iov_len_add(uiop, backup);
uiop->uio_offset -= backup;
uio_uio_resid_add(uiop, backup);
len = rlen;
}
commit = fxdr_unsigned(int, *tl++);
/*
* Return the lowest commitment level
* obtained by any of the RPCs.
*/
if (committed == NFSWRITE_FILESYNC)
committed = commit;
else if (committed == NFSWRITE_DATASYNC &&
commit == NFSWRITE_UNSTABLE)
committed = commit;
NFSLOCKMNT(nmp);
if (!NFSHASWRITEVERF(nmp)) {
NFSBCOPY((caddr_t)tl,
(caddr_t)&nmp->nm_verf[0],
NFSX_VERF);
NFSSETWRITEVERF(nmp);
} else if (NFSBCMP(tl, nmp->nm_verf,
NFSX_VERF)) {
*must_commit = 1;
NFSBCOPY(tl, nmp->nm_verf, NFSX_VERF);
}
NFSUNLOCKMNT(nmp);
}
if (nd->nd_flag & ND_NFSV4)
NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
if (nd->nd_flag & (ND_NFSV2 | ND_NFSV4)) {
error = nfsm_loadattr(nd, nap);
if (!error)
*attrflagp = NFS_LATTR_NOSHRINK;
}
} else {
error = nd->nd_repstat;
}
if (error)
goto nfsmout;
NFSWRITERPC_SETTIME(wccflag, np, nap, (nd->nd_flag & ND_NFSV4));
mbuf_freem(nd->nd_mrep);
nd->nd_mrep = NULL;
tsiz -= len;
}
nfsmout:
mbuf_freem(nd->nd_mrep);
*iomode = committed;
if (nd->nd_repstat && !error)
error = nd->nd_repstat;
return (error);
}
/*
* nfs mknod rpc
* For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
* mode set to specify the file type and the size field for rdev.
*/
APPLESTATIC int
nfsrpc_mknod(vnode_t dvp, char *name, int namelen, struct vattr *vap,
u_int32_t rdev, enum vtype vtyp, struct ucred *cred, NFSPROC_T *p,
struct nfsvattr *dnap, struct nfsvattr *nnap, struct nfsfh **nfhpp,
int *attrflagp, int *dattrflagp, void *dstuff)
{
u_int32_t *tl;
int error = 0;
struct nfsrv_descript nfsd, *nd = &nfsd;
nfsattrbit_t attrbits;
nmp = VFSTONFS(dvp->v_mount);
np = VTONFS(dvp);
*unlockedp = 0;
*nfhpp = NULL;
*dpp = NULL;
*attrflagp = 0;
*dattrflagp = 0;
if (namelen > NFS_MAXNAMLEN)
return (ENAMETOOLONG);
NFSCL_REQSTART(nd, NFSPROC_CREATE, dvp);
/*
* For V4, this is actually an Open op.
*/
NFSM_BUILD(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
*tl++ = txdr_unsigned(owp->nfsow_seqid);
*tl++ = txdr_unsigned(NFSV4OPEN_ACCESSWRITE |
NFSV4OPEN_ACCESSREAD);
*tl++ = txdr_unsigned(NFSV4OPEN_DENYNONE);
*tl++ = NFSMNT_MDSSESSION(nmp)->nfsess_clientid.lval[0];
*tl = NFSMNT_MDSSESSION(nmp)->nfsess_clientid.lval[1];
(void) nfsm_strtom(nd, owp->nfsow_owner, NFSV4CL_LOCKNAMELEN);
NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
*tl++ = txdr_unsigned(NFSV4OPEN_CREATE);
if (fmode & O_EXCL) {
if (NFSHASNFSV4N(nmp)) {
if (NFSHASSESSPERSIST(nmp)) {
/* Use GUARDED for persistent sessions. */
*tl = txdr_unsigned(NFSCREATE_GUARDED);
nfscl_fillsattr(nd, vap, dvp, 0, 0);
} else {
/* Otherwise, use EXCLUSIVE4_1. */
*tl = txdr_unsigned(NFSCREATE_EXCLUSIVE41);
NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
*tl++ = cverf.lval[0];
*tl = cverf.lval[1];
nfscl_fillsattr(nd, vap, dvp, 0, 0);
}
} else {
/* NFSv4.0 */
*tl = txdr_unsigned(NFSCREATE_EXCLUSIVE);
NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
*tl++ = cverf.lval[0];
*tl = cverf.lval[1];
}
} else {
*tl = txdr_unsigned(NFSCREATE_UNCHECKED);
nfscl_fillsattr(nd, vap, dvp, 0, 0);
}
NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
*tl = txdr_unsigned(NFSV4OPEN_CLAIMNULL);
(void) nfsm_strtom(nd, name, namelen);
/* Get the new file's handle and attributes. */
NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
*tl++ = txdr_unsigned(NFSV4OP_GETFH);
*tl = txdr_unsigned(NFSV4OP_GETATTR);
NFSGETATTR_ATTRBIT(&attrbits);
(void) nfsrv_putattrbit(nd, &attrbits);
/* Get the directory's post-op attributes. */
NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
*tl = txdr_unsigned(NFSV4OP_PUTFH);
(void) nfsm_fhtom(nd, np->n_fhp->nfh_fh, np->n_fhp->nfh_len, 0);
NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
*tl = txdr_unsigned(NFSV4OP_GETATTR);
(void) nfsrv_putattrbit(nd, &attrbits);
error = nfscl_request(nd, dvp, p, cred, dstuff);
if (error)
return (error);
NFSCL_INCRSEQID(owp->nfsow_seqid, nd);
if (nd->nd_repstat == 0) {
NFSM_DISSECT(tl, u_int32_t *, NFSX_STATEID +
6 * NFSX_UNSIGNED);
stateid.seqid = *tl++;
stateid.other[0] = *tl++;
stateid.other[1] = *tl++;
stateid.other[2] = *tl;
rflags = fxdr_unsigned(u_int32_t, *(tl + 6));
(void) nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
deleg = fxdr_unsigned(int, *tl);
if (deleg == NFSV4OPEN_DELEGATEREAD ||
deleg == NFSV4OPEN_DELEGATEWRITE) {
if (!(owp->nfsow_clp->nfsc_flags &
NFSCLFLAGS_FIRSTDELEG))
owp->nfsow_clp->nfsc_flags |=
(NFSCLFLAGS_FIRSTDELEG | NFSCLFLAGS_GOTDELEG);
MALLOC(dp, struct nfscldeleg *,
sizeof (struct nfscldeleg) + NFSX_V4FHMAX,
M_NFSCLDELEG, M_WAITOK);
LIST_INIT(&dp->nfsdl_owner);
LIST_INIT(&dp->nfsdl_lock);
dp->nfsdl_clp = owp->nfsow_clp;
newnfs_copyincred(cred, &dp->nfsdl_cred);
nfscl_lockinit(&dp->nfsdl_rwlock);
NFSM_DISSECT(tl, u_int32_t *, NFSX_STATEID +
NFSX_UNSIGNED);
dp->nfsdl_stateid.seqid = *tl++;
dp->nfsdl_stateid.other[0] = *tl++;
dp->nfsdl_stateid.other[1] = *tl++;
dp->nfsdl_stateid.other[2] = *tl++;
ret = fxdr_unsigned(int, *tl);
if (deleg == NFSV4OPEN_DELEGATEWRITE) {
dp->nfsdl_flags = NFSCLDL_WRITE;
/*
* Indicates how much the file can grow.
*/
NFSM_DISSECT(tl, u_int32_t *,
3 * NFSX_UNSIGNED);
limitby = fxdr_unsigned(int, *tl++);
switch (limitby) {
case NFSV4OPEN_LIMITSIZE:
dp->nfsdl_sizelimit = fxdr_hyper(tl);
break;
case NFSV4OPEN_LIMITBLOCKS:
dp->nfsdl_sizelimit =
fxdr_unsigned(u_int64_t, *tl++);
dp->nfsdl_sizelimit *=
fxdr_unsigned(u_int64_t, *tl);
break;
default:
error = NFSERR_BADXDR;
goto nfsmout;
}
} else {
dp->nfsdl_flags = NFSCLDL_READ;
}
if (ret)
dp->nfsdl_flags |= NFSCLDL_RECALL;
error = nfsrv_dissectace(nd, &dp->nfsdl_ace, &ret,
&acesize, p);
if (error)
goto nfsmout;
} else if (deleg != NFSV4OPEN_DELEGATENONE) {
error = NFSERR_BADXDR;
goto nfsmout;
}
error = nfscl_mtofh(nd, nfhpp, nnap, attrflagp);
if (error)
goto nfsmout;
/* Get rid of the PutFH and Getattr status values. */
NFSM_DISSECT(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
/* Load the directory attributes. */
error = nfsm_loadattr(nd, dnap);
if (error)
goto nfsmout;
*dattrflagp = 1;
if (dp != NULL && *attrflagp) {
dp->nfsdl_change = nnap->na_filerev;
dp->nfsdl_modtime = nnap->na_mtime;
dp->nfsdl_flags |= NFSCLDL_MODTIMESET;
}
/*
* We can now complete the Open state.
*/
nfhp = *nfhpp;
if (dp != NULL) {
dp->nfsdl_fhlen = nfhp->nfh_len;
NFSBCOPY(nfhp->nfh_fh, dp->nfsdl_fh, nfhp->nfh_len);
}
/*
* Get an Open structure that will be
* attached to the OpenOwner, acquired already.
*/
error = nfscl_open(dvp, nfhp->nfh_fh, nfhp->nfh_len,
(NFSV4OPEN_ACCESSWRITE | NFSV4OPEN_ACCESSREAD), 0,
cred, p, NULL, &op, &newone, NULL, 0);
if (error)
goto nfsmout;
op->nfso_stateid = stateid;
newnfs_copyincred(cred, &op->nfso_cred);
if ((rflags & NFSV4OPEN_RESULTCONFIRM)) {
do {
ret = nfsrpc_openconfirm(dvp, nfhp->nfh_fh,
nfhp->nfh_len, op, cred, p);
if (ret == NFSERR_DELAY)
(void) nfs_catnap(PZERO, ret, "nfs_create");
} while (ret == NFSERR_DELAY);
error = ret;
}
/*
* If the server is handing out delegations, but we didn't
* get one because an OpenConfirm was required, try the
* Open again, to get a delegation. This is a harmless no-op,
* from a server's point of view.
*/
if ((rflags & NFSV4OPEN_RESULTCONFIRM) &&
(owp->nfsow_clp->nfsc_flags & NFSCLFLAGS_GOTDELEG) &&
!error && dp == NULL) {
do {
ret = nfsrpc_openrpc(VFSTONFS(vnode_mount(dvp)), dvp,
np->n_fhp->nfh_fh, np->n_fhp->nfh_len,
nfhp->nfh_fh, nfhp->nfh_len,
(NFSV4OPEN_ACCESSWRITE | NFSV4OPEN_ACCESSREAD), op,
name, namelen, &dp, 0, 0x0, cred, p, 0, 1);
if (ret == NFSERR_DELAY)
(void) nfs_catnap(PZERO, ret, "nfs_crt2");
} while (ret == NFSERR_DELAY);
if (ret) {
if (dp != NULL) {
FREE((caddr_t)dp, M_NFSCLDELEG);
dp = NULL;
}
if (ret == NFSERR_STALECLIENTID ||
ret == NFSERR_STALEDONTRECOVER ||
ret == NFSERR_BADSESSION)
error = ret;
}
}
nfscl_openrelease(op, error, newone);
*unlockedp = 1;
}
if (nd->nd_repstat != 0 && error == 0)
error = nd->nd_repstat;
if (error == NFSERR_STALECLIENTID || error == NFSERR_BADSESSION)
nfscl_initiate_recovery(owp->nfsow_clp);
nfsmout:
if (!error)
*dpp = dp;
else if (dp != NULL)
FREE((caddr_t)dp, M_NFSCLDELEG);
mbuf_freem(nd->nd_mrep);
return (error);
}
/*
* nfs symbolic link create rpc
*/
APPLESTATIC int
nfsrpc_symlink(vnode_t dvp, char *name, int namelen, char *target,
struct vattr *vap, struct ucred *cred, NFSPROC_T *p, struct nfsvattr *dnap,
struct nfsvattr *nnap, struct nfsfh **nfhpp, int *attrflagp,
int *dattrflagp, void *dstuff)
{
u_int32_t *tl;
struct nfsrv_descript nfsd, *nd = &nfsd;
struct nfsmount *nmp;
int slen, error = 0;
*nfhpp = NULL;
*attrflagp = 0;
*dattrflagp = 0;
nmp = VFSTONFS(vnode_mount(dvp));
slen = strlen(target);
if (slen > NFS_MAXPATHLEN || namelen > NFS_MAXNAMLEN)
return (ENAMETOOLONG);
NFSCL_REQSTART(nd, NFSPROC_SYMLINK, dvp);
if (nd->nd_flag & ND_NFSV4) {
NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
*tl = txdr_unsigned(NFLNK);
(void) nfsm_strtom(nd, target, slen);
}
(void) nfsm_strtom(nd, name, namelen);
if (nd->nd_flag & (ND_NFSV3 | ND_NFSV4))
nfscl_fillsattr(nd, vap, dvp, 0, 0);
if (!(nd->nd_flag & ND_NFSV4))
(void) nfsm_strtom(nd, target, slen);
if (nd->nd_flag & ND_NFSV2)
nfscl_fillsattr(nd, vap, dvp, NFSSATTR_SIZENEG1, 0);
error = nfscl_request(nd, dvp, p, cred, dstuff);
if (error)
return (error);
if (nd->nd_flag & ND_NFSV4)
error = nfscl_wcc_data(nd, dvp, dnap, dattrflagp, NULL, dstuff);
if ((nd->nd_flag & ND_NFSV3) && !error) {
if (!nd->nd_repstat)
error = nfscl_mtofh(nd, nfhpp, nnap, attrflagp);
if (!error)
error = nfscl_wcc_data(nd, dvp, dnap, dattrflagp,
NULL, dstuff);
}
if (nd->nd_repstat && !error)
error = nd->nd_repstat;
mbuf_freem(nd->nd_mrep);
/*
* Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
* Only do this if vfs.nfs.ignore_eexist is set.
* Never do this for NFSv4.1 or later minor versions, since sessions
* should guarantee "exactly once" RPC semantics.
*/
if (error == EEXIST && nfsignore_eexist != 0 && (!NFSHASNFSV4(nmp) ||
nmp->nm_minorvers == 0))
error = 0;
return (error);
}
/*
* nfs make dir rpc
*/
APPLESTATIC int
nfsrpc_mkdir(vnode_t dvp, char *name, int namelen, struct vattr *vap,
struct ucred *cred, NFSPROC_T *p, struct nfsvattr *dnap,
struct nfsvattr *nnap, struct nfsfh **nfhpp, int *attrflagp,
int *dattrflagp, void *dstuff)
{
u_int32_t *tl;
struct nfsrv_descript nfsd, *nd = &nfsd;
nfsattrbit_t attrbits;
int error = 0;
struct nfsfh *fhp;
struct nfsmount *nmp;
*nfhpp = NULL;
*attrflagp = 0;
*dattrflagp = 0;
nmp = VFSTONFS(vnode_mount(dvp));
fhp = VTONFS(dvp)->n_fhp;
if (namelen > NFS_MAXNAMLEN)
return (ENAMETOOLONG);
NFSCL_REQSTART(nd, NFSPROC_MKDIR, dvp);
if (nd->nd_flag & ND_NFSV4) {
NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
*tl = txdr_unsigned(NFDIR);
}
(void) nfsm_strtom(nd, name, namelen);
nfscl_fillsattr(nd, vap, dvp, NFSSATTR_SIZENEG1, 0);
if (nd->nd_flag & ND_NFSV4) {
NFSGETATTR_ATTRBIT(&attrbits);
NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
*tl++ = txdr_unsigned(NFSV4OP_GETFH);
*tl = txdr_unsigned(NFSV4OP_GETATTR);
(void) nfsrv_putattrbit(nd, &attrbits);
NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
*tl = txdr_unsigned(NFSV4OP_PUTFH);
(void) nfsm_fhtom(nd, fhp->nfh_fh, fhp->nfh_len, 0);
NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
*tl = txdr_unsigned(NFSV4OP_GETATTR);
(void) nfsrv_putattrbit(nd, &attrbits);
}
error = nfscl_request(nd, dvp, p, cred, dstuff);
if (error)
return (error);
if (nd->nd_flag & ND_NFSV4)
error = nfscl_wcc_data(nd, dvp, dnap, dattrflagp, NULL, dstuff);
if (!nd->nd_repstat && !error) {
if (nd->nd_flag & ND_NFSV4) {
NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
}
if (!error)
error = nfscl_mtofh(nd, nfhpp, nnap, attrflagp);
if (error == 0 && (nd->nd_flag & ND_NFSV4) != 0) {
/* Get rid of the PutFH and Getattr status values. */
NFSM_DISSECT(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
/* Load the directory attributes. */
error = nfsm_loadattr(nd, dnap);
if (error == 0)
*dattrflagp = 1;
}
}
if ((nd->nd_flag & ND_NFSV3) && !error)
error = nfscl_wcc_data(nd, dvp, dnap, dattrflagp, NULL, dstuff);
if (nd->nd_repstat && !error)
error = nd->nd_repstat;
nfsmout:
mbuf_freem(nd->nd_mrep);
/*
* Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
* Only do this if vfs.nfs.ignore_eexist is set.
* Never do this for NFSv4.1 or later minor versions, since sessions
* should guarantee "exactly once" RPC semantics.
*/
if (error == EEXIST && nfsignore_eexist != 0 && (!NFSHASNFSV4(nmp) ||
nmp->nm_minorvers == 0))
error = 0;
return (error);
}
/*
* nfs remove directory call
*/
APPLESTATIC int
nfsrpc_rmdir(vnode_t dvp, char *name, int namelen, struct ucred *cred,
NFSPROC_T *p, struct nfsvattr *dnap, int *dattrflagp, void *dstuff)
{
struct nfsrv_descript nfsd, *nd = &nfsd;
int error = 0;
*dattrflagp = 0;
if (namelen > NFS_MAXNAMLEN)
return (ENAMETOOLONG);
NFSCL_REQSTART(nd, NFSPROC_RMDIR, dvp);
(void) nfsm_strtom(nd, name, namelen);
error = nfscl_request(nd, dvp, p, cred, dstuff);
if (error)
return (error);
if (nd->nd_flag & (ND_NFSV3 | ND_NFSV4))
error = nfscl_wcc_data(nd, dvp, dnap, dattrflagp, NULL, dstuff);
if (nd->nd_repstat && !error)
error = nd->nd_repstat;
mbuf_freem(nd->nd_mrep);
/*
* Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
*/
if (error == ENOENT)
error = 0;
return (error);
}
/*
* Readdir rpc.
* Always returns with either uio_resid unchanged, if you are at the
* end of the directory, or uio_resid == 0, with all DIRBLKSIZ chunks
* filled in.
* I felt this would allow caching of directory blocks more easily
* than returning a partially filled block.
* Directory offset cookies:
* Oh my, what to do with them...
* I can think of three ways to deal with them:
* 1 - have the layer above these RPCs maintain a map between logical
* directory byte offsets and the NFS directory offset cookies
* 2 - pass the opaque directory offset cookies up into userland
* and let the libc functions deal with them, via the system call
* 3 - return them to userland in the "struct dirent", so future versions
* of libc can use them and do whatever is necessary to make things work
* above these rpc calls, in the meantime
* For now, I do #3 by "hiding" the directory offset cookies after the
* d_name field in struct dirent. This is space inside d_reclen that
* will be ignored by anything that doesn't know about them.
* The directory offset cookies are filled in as the last 8 bytes of
* each directory entry, after d_name. Someday, the userland libc
* functions may be able to use these. In the meantime, it satisfies
* OpenBSD's requirements for cookies being returned.
* If expects the directory offset cookie for the read to be in uio_offset
* and returns the one for the next entry after this directory block in
* there, as well.
*/
APPLESTATIC int
nfsrpc_readdir(vnode_t vp, struct uio *uiop, nfsuint64 *cookiep,
struct ucred *cred, NFSPROC_T *p, struct nfsvattr *nap, int *attrflagp,
int *eofp, void *stuff)
{
int len, left;
struct dirent *dp = NULL;
u_int32_t *tl;
nfsquad_t cookie, ncookie;
struct nfsmount *nmp = VFSTONFS(vnode_mount(vp));
struct nfsnode *dnp = VTONFS(vp);
struct nfsvattr nfsva;
struct nfsrv_descript nfsd, *nd = &nfsd;
int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
int reqsize, tryformoredirs = 1, readsize, eof = 0, gotmnton = 0;
long dotfileid, dotdotfileid = 0;
u_int32_t fakefileno = 0xffffffff, rderr;
char *cp;
nfsattrbit_t attrbits, dattrbits;
u_int32_t *tl2 = NULL;
size_t tresid;
/*
* There is no point in reading a lot more than uio_resid, however
* adding one additional DIRBLKSIZ makes sense. Since uio_resid
* and nm_readdirsize are both exact multiples of DIRBLKSIZ, this
* will never make readsize > nm_readdirsize.
*/
readsize = nmp->nm_readdirsize;
if (readsize > uio_uio_resid(uiop))
readsize = uio_uio_resid(uiop) + DIRBLKSIZ;
/*
* For NFSv4, first create the "." and ".." entries.
*/
if (NFSHASNFSV4(nmp)) {
reqsize = 6 * NFSX_UNSIGNED;
NFSGETATTR_ATTRBIT(&dattrbits);
NFSZERO_ATTRBIT(&attrbits);
NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_FILEID);
NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TYPE);
if (NFSISSET_ATTRBIT(&dnp->n_vattr.na_suppattr,
NFSATTRBIT_MOUNTEDONFILEID)) {
NFSSETBIT_ATTRBIT(&attrbits,
NFSATTRBIT_MOUNTEDONFILEID);
gotmnton = 1;
} else {
/*
* Must fake it. Use the fileno, except when the
* fsid is != to that of the directory. For that
* case, generate a fake fileno that is not the same.
*/
NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_FSID);
gotmnton = 0;
}
/*
* Joy, oh joy. For V4 we get to hand craft '.' and '..'.
*/
if (uiop->uio_offset == 0) {
NFSCL_REQSTART(nd, NFSPROC_LOOKUPP, vp);
NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
*tl++ = txdr_unsigned(NFSV4OP_GETFH);
*tl = txdr_unsigned(NFSV4OP_GETATTR);
(void) nfsrv_putattrbit(nd, &attrbits);
error = nfscl_request(nd, vp, p, cred, stuff);
if (error)
return (error);
dotfileid = 0; /* Fake out the compiler. */
if ((nd->nd_flag & ND_NOMOREDATA) == 0) {
error = nfsm_loadattr(nd, &nfsva);
if (error != 0)
goto nfsmout;
dotfileid = nfsva.na_fileid;
}
if (nd->nd_repstat == 0) {
NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
len = fxdr_unsigned(int, *(tl + 4));
if (len > 0 && len <= NFSX_V4FHMAX)
error = nfsm_advance(nd, NFSM_RNDUP(len), -1);
else
error = EPERM;
if (!error) {
NFSM_DISSECT(tl, u_int32_t *, 2*NFSX_UNSIGNED);
nfsva.na_mntonfileno = 0xffffffff;
error = nfsv4_loadattr(nd, NULL, &nfsva, NULL,
NULL, 0, NULL, NULL, NULL, NULL, NULL, 0,
NULL, NULL, NULL, p, cred);
if (error) {
dotdotfileid = dotfileid;
} else if (gotmnton) {
if (nfsva.na_mntonfileno != 0xffffffff)
dotdotfileid = nfsva.na_mntonfileno;
else
dotdotfileid = nfsva.na_fileid;
} else if (nfsva.na_filesid[0] ==
dnp->n_vattr.na_filesid[0] &&
nfsva.na_filesid[1] ==
dnp->n_vattr.na_filesid[1]) {
dotdotfileid = nfsva.na_fileid;
} else {
do {
fakefileno--;
} while (fakefileno ==
nfsva.na_fileid);
dotdotfileid = fakefileno;
}
}
} else if (nd->nd_repstat == NFSERR_NOENT) {
/*
* Lookupp returns NFSERR_NOENT when we are
* at the root, so just use the current dir.
*/
nd->nd_repstat = 0;
dotdotfileid = dotfileid;
} else {
error = nd->nd_repstat;
}
mbuf_freem(nd->nd_mrep);
if (error)
return (error);
nd->nd_mrep = NULL;
dp = (struct dirent *) CAST_DOWN(caddr_t, uio_iov_base(uiop));
dp->d_type = DT_DIR;
dp->d_fileno = dotfileid;
dp->d_namlen = 1;
dp->d_name[0] = '.';
dp->d_name[1] = '\0';
dp->d_reclen = DIRENT_SIZE(dp) + NFSX_HYPER;
/*
* Just make these offset cookie 0.
*/
tl = (u_int32_t *)&dp->d_name[4];
*tl++ = 0;
*tl = 0;
blksiz += dp->d_reclen;
uio_uio_resid_add(uiop, -(dp->d_reclen));
uiop->uio_offset += dp->d_reclen;
uio_iov_base_add(uiop, dp->d_reclen);
uio_iov_len_add(uiop, -(dp->d_reclen));
dp = (struct dirent *) CAST_DOWN(caddr_t, uio_iov_base(uiop));
dp->d_type = DT_DIR;
dp->d_fileno = dotdotfileid;
dp->d_namlen = 2;
dp->d_name[0] = '.';
dp->d_name[1] = '.';
dp->d_name[2] = '\0';
dp->d_reclen = DIRENT_SIZE(dp) + NFSX_HYPER;
/*
* Just make these offset cookie 0.
*/
tl = (u_int32_t *)&dp->d_name[4];
*tl++ = 0;
*tl = 0;
blksiz += dp->d_reclen;
uio_uio_resid_add(uiop, -(dp->d_reclen));
uiop->uio_offset += dp->d_reclen;
uio_iov_base_add(uiop, dp->d_reclen);
uio_iov_len_add(uiop, -(dp->d_reclen));
}
NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_RDATTRERROR);
} else {
reqsize = 5 * NFSX_UNSIGNED;
}
/* loop through the dir entries, doctoring them to 4bsd form */
while (more_dirs && bigenough) {
if (nd->nd_flag & ND_NFSV4) {
NFSM_DISSECT(tl, u_int32_t *, 3*NFSX_UNSIGNED);
ncookie.lval[0] = *tl++;
ncookie.lval[1] = *tl++;
len = fxdr_unsigned(int, *tl);
} else if (nd->nd_flag & ND_NFSV3) {
NFSM_DISSECT(tl, u_int32_t *, 3*NFSX_UNSIGNED);
nfsva.na_fileid = fxdr_hyper(tl);
tl += 2;
len = fxdr_unsigned(int, *tl);
} else {
NFSM_DISSECT(tl, u_int32_t *, 2*NFSX_UNSIGNED);
nfsva.na_fileid =
fxdr_unsigned(long, *tl++);
len = fxdr_unsigned(int, *tl);
}
if (len <= 0 || len > NFS_MAXNAMLEN) {
error = EBADRPC;
goto nfsmout;
}
tlen = NFSM_RNDUP(len);
if (tlen == len)
tlen += 4; /* To ensure null termination */
left = DIRBLKSIZ - blksiz;
if ((int)(tlen + DIRHDSIZ + NFSX_HYPER) > left) {
dp->d_reclen += left;
uio_iov_base_add(uiop, left);
uio_iov_len_add(uiop, -(left));
uio_uio_resid_add(uiop, -(left));
uiop->uio_offset += left;
blksiz = 0;
}
if ((int)(tlen + DIRHDSIZ + NFSX_HYPER) > uio_uio_resid(uiop))
bigenough = 0;
if (bigenough) {
dp = (struct dirent *) CAST_DOWN(caddr_t, uio_iov_base(uiop));
dp->d_namlen = len;
dp->d_reclen = tlen + DIRHDSIZ + NFSX_HYPER;
dp->d_type = DT_UNKNOWN;
blksiz += dp->d_reclen;
if (blksiz == DIRBLKSIZ)
blksiz = 0;
uio_uio_resid_add(uiop, -(DIRHDSIZ));
uiop->uio_offset += DIRHDSIZ;
uio_iov_base_add(uiop, DIRHDSIZ);
uio_iov_len_add(uiop, -(DIRHDSIZ));
error = nfsm_mbufuio(nd, uiop, len);
if (error)
goto nfsmout;
cp = CAST_DOWN(caddr_t, uio_iov_base(uiop));
tlen -= len;
*cp = '\0'; /* null terminate */
cp += tlen; /* points to cookie storage */
tl2 = (u_int32_t *)cp;
uio_iov_base_add(uiop, (tlen + NFSX_HYPER));
uio_iov_len_add(uiop, -(tlen + NFSX_HYPER));
uio_uio_resid_add(uiop, -(tlen + NFSX_HYPER));
uiop->uio_offset += (tlen + NFSX_HYPER);
} else {
error = nfsm_advance(nd, NFSM_RNDUP(len), -1);
if (error)
goto nfsmout;
}
if (nd->nd_flag & ND_NFSV4) {
rderr = 0;
nfsva.na_mntonfileno = 0xffffffff;
error = nfsv4_loadattr(nd, NULL, &nfsva, NULL,
NULL, 0, NULL, NULL, NULL, NULL, NULL, 0,
NULL, NULL, &rderr, p, cred);
if (error)
goto nfsmout;
NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
} else if (nd->nd_flag & ND_NFSV3) {
NFSM_DISSECT(tl, u_int32_t *, 3*NFSX_UNSIGNED);
ncookie.lval[0] = *tl++;
ncookie.lval[1] = *tl++;
} else {
NFSM_DISSECT(tl, u_int32_t *, 2*NFSX_UNSIGNED);
ncookie.lval[0] = 0;
ncookie.lval[1] = *tl++;
}
if (bigenough) {
if (nd->nd_flag & ND_NFSV4) {
if (rderr) {
dp->d_fileno = 0;
} else {
if (gotmnton) {
if (nfsva.na_mntonfileno != 0xffffffff)
dp->d_fileno = nfsva.na_mntonfileno;
else
dp->d_fileno = nfsva.na_fileid;
} else if (nfsva.na_filesid[0] ==
dnp->n_vattr.na_filesid[0] &&
nfsva.na_filesid[1] ==
dnp->n_vattr.na_filesid[1]) {
dp->d_fileno = nfsva.na_fileid;
} else {
do {
fakefileno--;
} while (fakefileno ==
nfsva.na_fileid);
dp->d_fileno = fakefileno;
}
dp->d_type = vtonfs_dtype(nfsva.na_type);
}
} else {
dp->d_fileno = nfsva.na_fileid;
}
*tl2++ = cookiep->nfsuquad[0] = cookie.lval[0] =
ncookie.lval[0];
*tl2 = cookiep->nfsuquad[1] = cookie.lval[1] =
ncookie.lval[1];
}
more_dirs = fxdr_unsigned(int, *tl);
}
/*
* If at end of rpc data, get the eof boolean
*/
if (!more_dirs) {
NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
eof = fxdr_unsigned(int, *tl);
if (tryformoredirs)
more_dirs = !eof;
if (nd->nd_flag & ND_NFSV4) {
error = nfscl_postop_attr(nd, nap, attrflagp,
stuff);
if (error)
goto nfsmout;
}
}
mbuf_freem(nd->nd_mrep);
nd->nd_mrep = NULL;
}
/*
* Fill last record, iff any, out to a multiple of DIRBLKSIZ
* by increasing d_reclen for the last record.
*/
if (blksiz > 0) {
left = DIRBLKSIZ - blksiz;
dp->d_reclen += left;
uio_iov_base_add(uiop, left);
uio_iov_len_add(uiop, -(left));
uio_uio_resid_add(uiop, -(left));
uiop->uio_offset += left;
}
/*
* If returning no data, assume end of file.
* If not bigenough, return not end of file, since you aren't
* returning all the data
* Otherwise, return the eof flag from the server.
*/
if (eofp) {
if (tresid == ((size_t)(uio_uio_resid(uiop))))
*eofp = 1;
else if (!bigenough)
*eofp = 0;
else
*eofp = eof;
}
/*
* For NFSv4, first create the "." and ".." entries.
*/
if (NFSHASNFSV4(nmp)) {
NFSGETATTR_ATTRBIT(&dattrbits);
NFSZERO_ATTRBIT(&attrbits);
NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_FILEID);
if (NFSISSET_ATTRBIT(&dnp->n_vattr.na_suppattr,
NFSATTRBIT_MOUNTEDONFILEID)) {
NFSSETBIT_ATTRBIT(&attrbits,
NFSATTRBIT_MOUNTEDONFILEID);
gotmnton = 1;
} else {
/*
* Must fake it. Use the fileno, except when the
* fsid is != to that of the directory. For that
* case, generate a fake fileno that is not the same.
*/
NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_FSID);
gotmnton = 0;
}
/*
* Joy, oh joy. For V4 we get to hand craft '.' and '..'.
*/
if (uiop->uio_offset == 0) {
NFSCL_REQSTART(nd, NFSPROC_LOOKUPP, vp);
NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
*tl++ = txdr_unsigned(NFSV4OP_GETFH);
*tl = txdr_unsigned(NFSV4OP_GETATTR);
(void) nfsrv_putattrbit(nd, &attrbits);
error = nfscl_request(nd, vp, p, cred, stuff);
if (error)
return (error);
dotfileid = 0; /* Fake out the compiler. */
if ((nd->nd_flag & ND_NOMOREDATA) == 0) {
error = nfsm_loadattr(nd, &nfsva);
if (error != 0)
goto nfsmout;
dctime = nfsva.na_ctime;
dotfileid = nfsva.na_fileid;
}
if (nd->nd_repstat == 0) {
NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
len = fxdr_unsigned(int, *(tl + 4));
if (len > 0 && len <= NFSX_V4FHMAX)
error = nfsm_advance(nd, NFSM_RNDUP(len), -1);
else
error = EPERM;
if (!error) {
NFSM_DISSECT(tl, u_int32_t *, 2*NFSX_UNSIGNED);
nfsva.na_mntonfileno = 0xffffffff;
error = nfsv4_loadattr(nd, NULL, &nfsva, NULL,
NULL, 0, NULL, NULL, NULL, NULL, NULL, 0,
NULL, NULL, NULL, p, cred);
if (error) {
dotdotfileid = dotfileid;
} else if (gotmnton) {
if (nfsva.na_mntonfileno != 0xffffffff)
dotdotfileid = nfsva.na_mntonfileno;
else
dotdotfileid = nfsva.na_fileid;
} else if (nfsva.na_filesid[0] ==
dnp->n_vattr.na_filesid[0] &&
nfsva.na_filesid[1] ==
dnp->n_vattr.na_filesid[1]) {
dotdotfileid = nfsva.na_fileid;
} else {
do {
fakefileno--;
} while (fakefileno ==
nfsva.na_fileid);
dotdotfileid = fakefileno;
}
}
} else if (nd->nd_repstat == NFSERR_NOENT) {
/*
* Lookupp returns NFSERR_NOENT when we are
* at the root, so just use the current dir.
*/
nd->nd_repstat = 0;
dotdotfileid = dotfileid;
} else {
error = nd->nd_repstat;
}
mbuf_freem(nd->nd_mrep);
if (error)
return (error);
nd->nd_mrep = NULL;
dp = (struct dirent *)uio_iov_base(uiop);
dp->d_type = DT_DIR;
dp->d_fileno = dotfileid;
dp->d_namlen = 1;
dp->d_name[0] = '.';
dp->d_name[1] = '\0';
dp->d_reclen = DIRENT_SIZE(dp) + NFSX_HYPER;
/*
* Just make these offset cookie 0.
*/
tl = (u_int32_t *)&dp->d_name[4];
*tl++ = 0;
*tl = 0;
blksiz += dp->d_reclen;
uio_uio_resid_add(uiop, -(dp->d_reclen));
uiop->uio_offset += dp->d_reclen;
uio_iov_base_add(uiop, dp->d_reclen);
uio_iov_len_add(uiop, -(dp->d_reclen));
dp = (struct dirent *)uio_iov_base(uiop);
dp->d_type = DT_DIR;
dp->d_fileno = dotdotfileid;
dp->d_namlen = 2;
dp->d_name[0] = '.';
dp->d_name[1] = '.';
dp->d_name[2] = '\0';
dp->d_reclen = DIRENT_SIZE(dp) + NFSX_HYPER;
/*
* Just make these offset cookie 0.
*/
tl = (u_int32_t *)&dp->d_name[4];
*tl++ = 0;
*tl = 0;
blksiz += dp->d_reclen;
uio_uio_resid_add(uiop, -(dp->d_reclen));
uiop->uio_offset += dp->d_reclen;
uio_iov_base_add(uiop, dp->d_reclen);
uio_iov_len_add(uiop, -(dp->d_reclen));
}
NFSREADDIRPLUS_ATTRBIT(&attrbits);
if (gotmnton)
NFSSETBIT_ATTRBIT(&attrbits,
NFSATTRBIT_MOUNTEDONFILEID);
}
if (bigenough) {
if (nd->nd_flag & ND_NFSV4) {
if (rderr) {
dp->d_fileno = 0;
} else if (gotmnton) {
if (nfsva.na_mntonfileno != 0xffffffff)
dp->d_fileno = nfsva.na_mntonfileno;
else
dp->d_fileno = nfsva.na_fileid;
} else if (nfsva.na_filesid[0] ==
dnp->n_vattr.na_filesid[0] &&
nfsva.na_filesid[1] ==
dnp->n_vattr.na_filesid[1]) {
dp->d_fileno = nfsva.na_fileid;
} else {
do {
fakefileno--;
} while (fakefileno ==
nfsva.na_fileid);
dp->d_fileno = fakefileno;
}
} else {
dp->d_fileno = fileno;
}
*tl2++ = cookiep->nfsuquad[0] = cookie.lval[0] =
ncookie.lval[0];
*tl2 = cookiep->nfsuquad[1] = cookie.lval[1] =
ncookie.lval[1];
if (nfhp != NULL) {
if (NFSRV_CMPFH(nfhp->nfh_fh, nfhp->nfh_len,
dnp->n_fhp->nfh_fh, dnp->n_fhp->nfh_len)) {
VREF(vp);
newvp = vp;
unlocknewvp = 0;
FREE((caddr_t)nfhp, M_NFSFH);
np = dnp;
} else if (isdotdot != 0) {
/*
* Skip doing a nfscl_nget() call for "..".
* There's a race between acquiring the nfs
* node here and lookups that look for the
* directory being read (in the parent).
* It would try to get a lock on ".." here,
* owning the lock on the directory being
* read. Lookup will hold the lock on ".."
* and try to acquire the lock on the
* directory being read.
* If the directory is unlocked/relocked,
* then there is a LOR with the buflock
* vp is relocked.
*/
free(nfhp, M_NFSFH);
} else {
error = nfscl_nget(vnode_mount(vp), vp,
nfhp, cnp, p, &np, NULL, LK_EXCLUSIVE);
if (!error) {
newvp = NFSTOV(np);
unlocknewvp = 1;
}
}
nfhp = NULL;
if (newvp != NULLVP) {
error = nfscl_loadattrcache(&newvp,
&nfsva, NULL, NULL, 0, 0);
if (error) {
if (unlocknewvp)
vput(newvp);
else
vrele(newvp);
goto nfsmout;
}
dp->d_type =
vtonfs_dtype(np->n_vattr.na_type);
ndp->ni_vp = newvp;
NFSCNHASH(cnp, HASHINIT);
if (cnp->cn_namelen <= NCHNAMLEN &&
(newvp->v_type != VDIR ||
dctime.tv_sec != 0)) {
cache_enter_time(ndp->ni_dvp,
ndp->ni_vp, cnp,
&nfsva.na_ctime,
newvp->v_type != VDIR ? NULL :
&dctime);
}
if (unlocknewvp)
vput(newvp);
else
vrele(newvp);
newvp = NULLVP;
}
}
} else if (nfhp != NULL) {
FREE((caddr_t)nfhp, M_NFSFH);
}
NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
more_dirs = fxdr_unsigned(int, *tl);
}
/*
* If at end of rpc data, get the eof boolean
*/
if (!more_dirs) {
NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
eof = fxdr_unsigned(int, *tl);
if (tryformoredirs)
more_dirs = !eof;
if (nd->nd_flag & ND_NFSV4) {
error = nfscl_postop_attr(nd, nap, attrflagp,
stuff);
if (error)
goto nfsmout;
}
}
mbuf_freem(nd->nd_mrep);
nd->nd_mrep = NULL;
}
/*
* Fill last record, iff any, out to a multiple of DIRBLKSIZ
* by increasing d_reclen for the last record.
*/
if (blksiz > 0) {
left = DIRBLKSIZ - blksiz;
dp->d_reclen += left;
uio_iov_base_add(uiop, left);
uio_iov_len_add(uiop, -(left));
uio_uio_resid_add(uiop, -(left));
uiop->uio_offset += left;
}
/*
* If returning no data, assume end of file.
* If not bigenough, return not end of file, since you aren't
* returning all the data
* Otherwise, return the eof flag from the server.
*/
if (eofp != NULL) {
if (tresid == uio_uio_resid(uiop))
*eofp = 1;
else if (!bigenough)
*eofp = 0;
else
*eofp = eof;
}
/*
* NFS byte range lock rpc.
* (Mostly just calls one of the three lower level RPC routines.)
*/
APPLESTATIC int
nfsrpc_advlock(vnode_t vp, off_t size, int op, struct flock *fl,
int reclaim, struct ucred *cred, NFSPROC_T *p, void *id, int flags)
{
struct nfscllockowner *lp;
struct nfsclclient *clp;
struct nfsfh *nfhp;
struct nfsrv_descript nfsd, *nd = &nfsd;
struct nfsmount *nmp = VFSTONFS(vnode_mount(vp));
u_int64_t off, len;
off_t start, end;
u_int32_t clidrev = 0;
int error = 0, newone = 0, expireret = 0, retrycnt, donelocally;
int callcnt, dorpc;
/*
* Convert the flock structure into a start and end and do POSIX
* bounds checking.
*/
switch (fl->l_whence) {
case SEEK_SET:
case SEEK_CUR:
/*
* Caller is responsible for adding any necessary offset
* when SEEK_CUR is used.
*/
start = fl->l_start;
off = fl->l_start;
break;
case SEEK_END:
start = size + fl->l_start;
off = size + fl->l_start;
break;
default:
return (EINVAL);
}
if (start < 0)
return (EINVAL);
if (fl->l_len != 0) {
end = start + fl->l_len - 1;
if (end < start)
return (EINVAL);
}
len = fl->l_len;
if (len == 0)
len = NFS64BITSSET;
retrycnt = 0;
do {
nd->nd_repstat = 0;
if (op == F_GETLK) {
error = nfscl_getcl(vnode_mount(vp), cred, p, 1, &clp);
if (error)
return (error);
error = nfscl_lockt(vp, clp, off, len, fl, p, id, flags);
if (!error) {
clidrev = clp->nfsc_clientidrev;
error = nfsrpc_lockt(nd, vp, clp, off, len, fl, cred,
p, id, flags);
} else if (error == -1) {
error = 0;
}
nfscl_clientrelease(clp);
} else if (op == F_UNLCK && fl->l_type == F_UNLCK) {
/*
* We must loop around for all lockowner cases.
*/
callcnt = 0;
error = nfscl_getcl(vnode_mount(vp), cred, p, 1, &clp);
if (error)
return (error);
do {
error = nfscl_relbytelock(vp, off, len, cred, p, callcnt,
clp, id, flags, &lp, &dorpc);
/*
* If it returns a NULL lp, we're done.
*/
if (lp == NULL) {
if (callcnt == 0)
nfscl_clientrelease(clp);
else
nfscl_releasealllocks(clp, vp, p, id, flags);
return (error);
}
if (nmp->nm_clp != NULL)
clidrev = nmp->nm_clp->nfsc_clientidrev;
else
clidrev = 0;
/*
* If the server doesn't support Posix lock semantics,
* only allow locks on the entire file, since it won't
* handle overlapping byte ranges.
* There might still be a problem when a lock
* upgrade/downgrade (read<->write) occurs, since the
* server "might" expect an unlock first?
*/
if (dorpc && (lp->nfsl_open->nfso_posixlock ||
(off == 0 && len == NFS64BITSSET))) {
/*
* Since the lock records will go away, we must
* wait for grace and delay here.
*/
do {
error = nfsrpc_locku(nd, nmp, lp, off, len,
NFSV4LOCKT_READ, cred, p, 0);
if ((nd->nd_repstat == NFSERR_GRACE ||
nd->nd_repstat == NFSERR_DELAY) &&
error == 0)
(void) nfs_catnap(PZERO, (int)nd->nd_repstat,
"nfs_advlock");
} while ((nd->nd_repstat == NFSERR_GRACE ||
nd->nd_repstat == NFSERR_DELAY) && error == 0);
}
callcnt++;
} while (error == 0 && nd->nd_repstat == 0);
nfscl_releasealllocks(clp, vp, p, id, flags);
} else if (op == F_SETLK) {
error = nfscl_getbytelock(vp, off, len, fl->l_type, cred, p,
NULL, 0, id, flags, NULL, NULL, &lp, &newone, &donelocally);
if (error || donelocally) {
return (error);
}
if (nmp->nm_clp != NULL)
clidrev = nmp->nm_clp->nfsc_clientidrev;
else
clidrev = 0;
nfhp = VTONFS(vp)->n_fhp;
if (!lp->nfsl_open->nfso_posixlock &&
(off != 0 || len != NFS64BITSSET)) {
error = EINVAL;
} else {
error = nfsrpc_lock(nd, nmp, vp, nfhp->nfh_fh,
nfhp->nfh_len, lp, newone, reclaim, off,
len, fl->l_type, cred, p, 0);
}
if (!error)
error = nd->nd_repstat;
nfscl_lockrelease(lp, error, newone);
} else {
error = EINVAL;
}
if (!error)
error = nd->nd_repstat;
if (error == NFSERR_GRACE || error == NFSERR_STALESTATEID ||
error == NFSERR_STALEDONTRECOVER ||
error == NFSERR_STALECLIENTID || error == NFSERR_DELAY ||
error == NFSERR_BADSESSION) {
(void) nfs_catnap(PZERO, error, "nfs_advlock");
} else if ((error == NFSERR_EXPIRED || error == NFSERR_BADSTATEID)
&& clidrev != 0) {
expireret = nfscl_hasexpired(nmp->nm_clp, clidrev, p);
retrycnt++;
}
} while (error == NFSERR_GRACE ||
error == NFSERR_STALECLIENTID || error == NFSERR_DELAY ||
error == NFSERR_STALEDONTRECOVER || error == NFSERR_STALESTATEID ||
error == NFSERR_BADSESSION ||
((error == NFSERR_EXPIRED || error == NFSERR_BADSTATEID) &&
expireret == 0 && clidrev != 0 && retrycnt < 4));
if (error && retrycnt >= 4)
error = EIO;
return (error);
}
/*
* The lower level routine for the LockT case.
*/
APPLESTATIC int
nfsrpc_lockt(struct nfsrv_descript *nd, vnode_t vp,
struct nfsclclient *clp, u_int64_t off, u_int64_t len, struct flock *fl,
struct ucred *cred, NFSPROC_T *p, void *id, int flags)
{
u_int32_t *tl;
int error, type, size;
uint8_t own[NFSV4CL_LOCKNAMELEN + NFSX_V4FHMAX];
struct nfsnode *np;
struct nfsmount *nmp;
nmp = VFSTONFS(vp->v_mount);
NFSCL_REQSTART(nd, NFSPROC_LOCKT, vp);
NFSM_BUILD(tl, u_int32_t *, 7 * NFSX_UNSIGNED);
if (fl->l_type == F_RDLCK)
*tl++ = txdr_unsigned(NFSV4LOCKT_READ);
else
*tl++ = txdr_unsigned(NFSV4LOCKT_WRITE);
txdr_hyper(off, tl);
tl += 2;
txdr_hyper(len, tl);
tl += 2;
*tl++ = NFSMNT_MDSSESSION(nmp)->nfsess_clientid.lval[0];
*tl = NFSMNT_MDSSESSION(nmp)->nfsess_clientid.lval[1];
nfscl_filllockowner(id, own, flags);
np = VTONFS(vp);
NFSBCOPY(np->n_fhp->nfh_fh, &own[NFSV4CL_LOCKNAMELEN],
np->n_fhp->nfh_len);
(void)nfsm_strtom(nd, own, NFSV4CL_LOCKNAMELEN + np->n_fhp->nfh_len);
error = nfscl_request(nd, vp, p, cred, NULL);
if (error)
return (error);
if (nd->nd_repstat == 0) {
fl->l_type = F_UNLCK;
} else if (nd->nd_repstat == NFSERR_DENIED) {
nd->nd_repstat = 0;
fl->l_whence = SEEK_SET;
NFSM_DISSECT(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
fl->l_start = fxdr_hyper(tl);
tl += 2;
len = fxdr_hyper(tl);
tl += 2;
if (len == NFS64BITSSET)
fl->l_len = 0;
else
fl->l_len = len;
type = fxdr_unsigned(int, *tl++);
if (type == NFSV4LOCKT_WRITE)
fl->l_type = F_WRLCK;
else
fl->l_type = F_RDLCK;
/*
* XXX For now, I have no idea what to do with the
* conflicting lock_owner, so I'll just set the pid == 0
* and skip over the lock_owner.
*/
fl->l_pid = (pid_t)0;
tl += 2;
size = fxdr_unsigned(int, *tl);
if (size < 0 || size > NFSV4_OPAQUELIMIT)
error = EBADRPC;
if (!error)
error = nfsm_advance(nd, NFSM_RNDUP(size), -1);
} else if (nd->nd_repstat == NFSERR_STALECLIENTID ||
nd->nd_repstat == NFSERR_BADSESSION)
nfscl_initiate_recovery(clp);
nfsmout:
mbuf_freem(nd->nd_mrep);
return (error);
}
/*
* Lower level function that performs the LockU RPC.
*/
static int
nfsrpc_locku(struct nfsrv_descript *nd, struct nfsmount *nmp,
struct nfscllockowner *lp, u_int64_t off, u_int64_t len,
u_int32_t type, struct ucred *cred, NFSPROC_T *p, int syscred)
{
u_int32_t *tl;
int error;
/*
* Now we know how many stripe indices and addresses, so
* we can allocate the structure the correct size.
*/
i = (stripecnt * sizeof(uint8_t)) / sizeof(struct nfsclds *)
+ 1;
NFSCL_DEBUG(4, "stripeindices=%d\n", i);
ndi = malloc(sizeof(*ndi) + (addrcnt + i) *
sizeof(struct nfsclds *), M_NFSDEVINFO, M_WAITOK | M_ZERO);
NFSBCOPY(deviceid, ndi->nfsdi_deviceid, NFSX_V4DEVICEID);
ndi->nfsdi_refcnt = 0;
ndi->nfsdi_stripecnt = stripecnt;
ndi->nfsdi_addrcnt = addrcnt;
/* Fill in the stripe indices. */
for (i = 0; i < stripecnt; i++) {
stripeindex = fxdr_unsigned(uint8_t, *tl++);
NFSCL_DEBUG(4, "stripeind=%d\n", stripeindex);
if (stripeindex >= addrcnt) {
printf("NFS devinfo stripeindex %d: too big\n",
(int)stripeindex);
error = NFSERR_BADXDR;
goto nfsmout;
}
nfsfldi_setstripeindex(ndi, i, stripeindex);
}
/* Now, dissect the server address(es). */
safilled = 0;
for (i = 0; i < addrcnt; i++) {
NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED);
cnt = fxdr_unsigned(uint32_t, *tl);
if (cnt == 0) {
printf("NFS devinfo 0 len addrlist\n");
error = NFSERR_BADXDR;
goto nfsmout;
}
dspp = nfsfldi_addr(ndi, i);
pos = arc4random() % cnt; /* Choose one. */
safilled = 0;
for (j = 0; j < cnt; j++) {
error = nfsv4_getipaddr(nd, &ss, &isudp);
if (error != 0 && error != EPERM) {
error = NFSERR_BADXDR;
goto nfsmout;
}
if (error == 0 && isudp == 0) {
/*
* The algorithm is:
* - use "pos" entry if it is of the
* same af_family or none of them
* is of the same af_family
* else
* - use the first one of the same
* af_family.
*/
if ((safilled == 0 && ss.ss_family ==
nmp->nm_nam->sa_family) ||
(j == pos &&
(safilled == 0 || ss.ss_family ==
nmp->nm_nam->sa_family)) ||
(safilled == 1 && ss.ss_family ==
nmp->nm_nam->sa_family)) {
error = nfsrpc_fillsa(nmp, &ss,
&dsp, p);
if (error == 0) {
*dspp = dsp;
if (ss.ss_family ==
nmp->nm_nam->sa_family)
safilled = 2;
else
safilled = 1;
}
}
}
}
if (safilled == 0)
break;
}
/* And the notify bits. */
NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED);
if (safilled != 0) {
bitcnt = fxdr_unsigned(int, *tl);
if (bitcnt > 0) {
NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED);
if (notifybitsp != NULL)
*notifybitsp =
fxdr_unsigned(uint32_t, *tl);
}
*ndip = ndi;
} else
error = EPERM;
}
if (nd->nd_repstat != 0)
error = nd->nd_repstat;
nfsmout:
if (error != 0 && ndi != NULL)
nfscl_freedevinfo(ndi);
mbuf_freem(nd->nd_mrep);
return (error);
}
/*
* Do the NFSv4.1 LayoutCommit.
*/
int
nfsrpc_layoutcommit(struct nfsmount *nmp, uint8_t *fh, int fhlen, int reclaim,
uint64_t off, uint64_t len, uint64_t lastbyte, nfsv4stateid_t *stateidp,
int layouttype, int layoutupdatecnt, uint8_t *layp, struct ucred *cred,
NFSPROC_T *p, void *stuff)
{
uint32_t *tl;
struct nfsrv_descript nfsd, *nd = &nfsd;
int error, outcnt, i;
uint8_t *cp;
/*
* Acquire a layout and devinfo, if possible. The caller must have acquired
* a reference count on the nfsclclient structure before calling this.
* Return the layout in lypp with a reference count on it, if successful.
*/
static int
nfsrpc_getlayout(struct nfsmount *nmp, vnode_t vp, struct nfsfh *nfhp,
int iomode, uint32_t *notifybitsp, nfsv4stateid_t *stateidp, uint64_t off,
struct nfscllayout **lypp, struct ucred *cred, NFSPROC_T *p)
{
struct nfscllayout *lyp;
struct nfsclflayout *flp, *tflp;
struct nfscldevinfo *dip;
struct nfsclflayouthead flh;
int error = 0, islocked, layoutlen, recalled, retonclose;
nfsv4stateid_t stateid;
*lypp = NULL;
/*
* If lyp is returned non-NULL, there will be a refcnt (shared lock)
* on it, iff flp != NULL or a lock (exclusive lock) on it iff
* flp == NULL.
*/
lyp = nfscl_getlayout(nmp->nm_clp, nfhp->nfh_fh, nfhp->nfh_len,
off, &flp, &recalled);
islocked = 0;
if (lyp == NULL || flp == NULL) {
if (recalled != 0)
return (EIO);
LIST_INIT(&flh);
layoutlen = NFSMNT_MDSSESSION(nmp)->nfsess_maxcache -
(NFSX_STATEID + 3 * NFSX_UNSIGNED);
if (lyp == NULL) {
stateid.seqid = 0;
stateid.other[0] = stateidp->other[0];
stateid.other[1] = stateidp->other[1];
stateid.other[2] = stateidp->other[2];
error = nfsrpc_layoutget(nmp, nfhp->nfh_fh,
nfhp->nfh_len, iomode, (uint64_t)0, INT64_MAX,
(uint64_t)0, layoutlen, &stateid, &retonclose,
&flh, cred, p, NULL);
} else {
islocked = 1;
stateid.seqid = lyp->nfsly_stateid.seqid;
stateid.other[0] = lyp->nfsly_stateid.other[0];
stateid.other[1] = lyp->nfsly_stateid.other[1];
stateid.other[2] = lyp->nfsly_stateid.other[2];
error = nfsrpc_layoutget(nmp, nfhp->nfh_fh,
nfhp->nfh_len, iomode, off, INT64_MAX,
(uint64_t)0, layoutlen, &stateid, &retonclose,
&flh, cred, p, NULL);
}
if (error == 0)
LIST_FOREACH(tflp, &flh, nfsfl_list) {
error = nfscl_adddevinfo(nmp, NULL, tflp);
if (error != 0) {
error = nfsrpc_getdeviceinfo(nmp,
tflp->nfsfl_dev,
NFSLAYOUT_NFSV4_1_FILES,
notifybitsp, &dip, cred, p);
if (error != 0)
break;
error = nfscl_adddevinfo(nmp, dip,
tflp);
if (error != 0)
printf(
"getlayout: cannot add\n");
}
}
if (error == 0) {
/*
* nfscl_layout() always returns with the nfsly_lock
* set to a refcnt (shared lock).
*/
error = nfscl_layout(nmp, vp, nfhp->nfh_fh,
nfhp->nfh_len, &stateid, retonclose, &flh, &lyp,
cred, p);
if (error == 0)
*lypp = lyp;
} else if (islocked != 0)
nfsv4_unlock(&lyp->nfsly_lock, 0);
} else
*lypp = lyp;
return (error);
}
/*
* Do a TCP connection plus exchange id and create session.
* If successful, a "struct nfsclds" is linked into the list for the
* mount point and a pointer to it is returned.
*/
static int
nfsrpc_fillsa(struct nfsmount *nmp, struct sockaddr_storage *ssp,
struct nfsclds **dspp, NFSPROC_T *p)
{
struct sockaddr_in *msad, *sad, *ssd;
struct sockaddr_in6 *msad6, *sad6, *ssd6;
struct nfsclclient *clp;
struct nfssockreq *nrp;
struct nfsclds *dsp, *tdsp;
int error;
enum nfsclds_state retv;
uint32_t sequenceid;
/*
* Check to see if we already have a session for this
* address that is usable for a DS.
* Note that the MDS's address is in a different place
* than the sessions already acquired for DS's.
*/
msad = (struct sockaddr_in *)nmp->nm_sockreq.nr_nam;
tdsp = TAILQ_FIRST(&nmp->nm_sess);
while (tdsp != NULL) {
if (msad != NULL && msad->sin_family == AF_INET &&
ssd->sin_addr.s_addr == msad->sin_addr.s_addr &&
ssd->sin_port == msad->sin_port &&
(tdsp->nfsclds_flags & NFSCLDS_DS) != 0) {
*dspp = tdsp;
NFSUNLOCKMNT(nmp);
NFSCL_DEBUG(4, "fnd same addr\n");
return (0);
}
tdsp = TAILQ_NEXT(tdsp, nfsclds_list);
if (tdsp != NULL && tdsp->nfsclds_sockp != NULL)
msad = (struct sockaddr_in *)
tdsp->nfsclds_sockp->nr_nam;
else
msad = NULL;
}
NFSUNLOCKMNT(nmp);
/* No IP address match, so look for new/trunked one. */
sad = malloc(sizeof(*sad), M_SONAME, M_WAITOK | M_ZERO);
sad->sin_len = sizeof(*sad);
sad->sin_family = AF_INET;
sad->sin_port = ssd->sin_port;
sad->sin_addr.s_addr = ssd->sin_addr.s_addr;
nrp = malloc(sizeof(*nrp), M_NFSSOCKREQ, M_WAITOK | M_ZERO);
nrp->nr_nam = (struct sockaddr *)sad;
} else if (ssp->ss_family == AF_INET6) {
ssd6 = (struct sockaddr_in6 *)ssp;
NFSLOCKMNT(nmp);
/*
* Check to see if we already have a session for this
* address that is usable for a DS.
* Note that the MDS's address is in a different place
* than the sessions already acquired for DS's.
*/
msad6 = (struct sockaddr_in6 *)nmp->nm_sockreq.nr_nam;
tdsp = TAILQ_FIRST(&nmp->nm_sess);
while (tdsp != NULL) {
if (msad6 != NULL && msad6->sin6_family == AF_INET6 &&
IN6_ARE_ADDR_EQUAL(&ssd6->sin6_addr,
&msad6->sin6_addr) &&
ssd6->sin6_port == msad6->sin6_port &&
(tdsp->nfsclds_flags & NFSCLDS_DS) != 0) {
*dspp = tdsp;
NFSUNLOCKMNT(nmp);
return (0);
}
tdsp = TAILQ_NEXT(tdsp, nfsclds_list);
if (tdsp != NULL && tdsp->nfsclds_sockp != NULL)
msad6 = (struct sockaddr_in6 *)
tdsp->nfsclds_sockp->nr_nam;
else
msad6 = NULL;
}
NFSUNLOCKMNT(nmp);
/*
* Use the credentials that were used for the mount, which are
* in nmp->nm_sockreq.nr_cred for newnfs_connect() etc.
* Ref. counting the credentials with crhold() is probably not
* necessary, since nm_sockreq.nr_cred won't be crfree()'d until
* unmount, but I did it anyhow.
*/
nrp->nr_cred = crhold(nmp->nm_sockreq.nr_cred);
error = newnfs_connect(nmp, nrp, NULL, p, 0);
NFSCL_DEBUG(3, "DS connect=%d\n", error);
/* Now, do the exchangeid and create session. */
if (error == 0)
error = nfsrpc_exchangeid(nmp, clp, nrp, NFSV4EXCH_USEPNFSDS,
&dsp, nrp->nr_cred, p);
NFSCL_DEBUG(3, "DS exchangeid=%d\n", error);
if (error == 0) {
dsp->nfsclds_sockp = nrp;
NFSLOCKMNT(nmp);
retv = nfscl_getsameserver(nmp, dsp, &tdsp);
NFSCL_DEBUG(3, "getsame ret=%d\n", retv);
if (retv == NFSDSP_USETHISSESSION) {
NFSUNLOCKMNT(nmp);
/*
* If there is already a session for this server,
* use it.
*/
(void)newnfs_disconnect(nrp);
nfscl_freenfsclds(dsp);
*dspp = tdsp;
return (0);
}
if (retv == NFSDSP_SEQTHISSESSION)
sequenceid = tdsp->nfsclds_sess.nfsess_sequenceid;
else
sequenceid = dsp->nfsclds_sess.nfsess_sequenceid;
NFSUNLOCKMNT(nmp);
error = nfsrpc_createsession(nmp, &dsp->nfsclds_sess,
nrp, sequenceid, 0, nrp->nr_cred, p);
NFSCL_DEBUG(3, "DS createsess=%d\n", error);
} else {
NFSFREECRED(nrp->nr_cred);
NFSFREEMUTEX(&nrp->nr_mtx);
free(nrp->nr_nam, M_SONAME);
free(nrp, M_NFSSOCKREQ);
}
if (error == 0) {
NFSCL_DEBUG(3, "add DS session\n");
/*
* Put it at the end of the list. That way the list
* is ordered by when the entry was added. This matters
* since the one done first is the one that should be
* used for sequencid'ing any subsequent create sessions.
*/
NFSLOCKMNT(nmp);
TAILQ_INSERT_TAIL(&nmp->nm_sess, dsp, nfsclds_list);
NFSUNLOCKMNT(nmp);
*dspp = dsp;
} else if (dsp != NULL)
nfscl_freenfsclds(dsp);
return (error);
}
/*
* Do the NFSv4.1 Reclaim Complete.
*/
int
nfsrpc_reclaimcomplete(struct nfsmount *nmp, struct ucred *cred, NFSPROC_T *p)
{
uint32_t *tl;
struct nfsrv_descript nfsd;
struct nfsrv_descript *nd = &nfsd;
int error;
/*
* Initialize the slot tables for a session.
*/
static void
nfscl_initsessionslots(struct nfsclsession *sep)
{
int i;
for (i = 0; i < NFSV4_CBSLOTS; i++) {
m_freem(sep->nfsess_cbslots[i].nfssl_reply);
NFSBZERO(&sep->nfsess_cbslots[i], sizeof(struct nfsslot));
}
for (i = 0; i < 64; i++)
sep->nfsess_slotseq[i] = 0;
sep->nfsess_slots = 0;
}
/*
* Called to try and do an I/O operation via an NFSv4.1 Data Server (DS).
*/
int
nfscl_doiods(vnode_t vp, struct uio *uiop, int *iomode, int *must_commit,
uint32_t rwaccess, struct ucred *cred, NFSPROC_T *p)
{
struct nfsnode *np = VTONFS(vp);
struct nfsmount *nmp = VFSTONFS(vnode_mount(vp));
struct nfscllayout *layp;
struct nfscldevinfo *dip;
struct nfsclflayout *rflp;
nfsv4stateid_t stateid;
struct ucred *newcred;
uint64_t lastbyte, len, off, oresid, xfer;
int eof, error, iolaymode, recalled;
void *lckp;
if (!NFSHASPNFS(nmp) || nfscl_enablecallb == 0 || nfs_numnfscbd == 0 ||
(np->n_flag & NNOLAYOUT) != 0)
return (EIO);
/* Now, get a reference cnt on the clientid for this mount. */
if (nfscl_getref(nmp) == 0)
return (EIO);
/* Find an appropriate stateid. */
newcred = NFSNEWCRED(cred);
error = nfscl_getstateid(vp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len,
rwaccess, 1, newcred, p, &stateid, &lckp);
if (error != 0) {
NFSFREECRED(newcred);
nfscl_relref(nmp);
return (error);
}
/* Search for a layout for this file. */
off = uiop->uio_offset;
layp = nfscl_getlayout(nmp->nm_clp, np->n_fhp->nfh_fh,
np->n_fhp->nfh_len, off, &rflp, &recalled);
if (layp == NULL || rflp == NULL) {
if (recalled != 0) {
NFSFREECRED(newcred);
nfscl_relref(nmp);
return (EIO);
}
if (layp != NULL) {
nfscl_rellayout(layp, (rflp == NULL) ? 1 : 0);
layp = NULL;
}
/* Try and get a Layout, if it is supported. */
if (rwaccess == NFSV4OPEN_ACCESSWRITE ||
(np->n_flag & NWRITEOPENED) != 0)
iolaymode = NFSLAYOUTIOMODE_RW;
else
iolaymode = NFSLAYOUTIOMODE_READ;
error = nfsrpc_getlayout(nmp, vp, np->n_fhp, iolaymode,
NULL, &stateid, off, &layp, newcred, p);
if (error != 0) {
NFSLOCKNODE(np);
np->n_flag |= NNOLAYOUT;
NFSUNLOCKNODE(np);
if (lckp != NULL)
nfscl_lockderef(lckp);
NFSFREECRED(newcred);
if (layp != NULL)
nfscl_rellayout(layp, 0);
nfscl_relref(nmp);
return (error);
}
}
/*
* Loop around finding a layout that works for the first part of
* this I/O operation, and then call the function that actually
* does the RPC.
*/
eof = 0;
len = (uint64_t)uiop->uio_resid;
while (len > 0 && error == 0 && eof == 0) {
off = uiop->uio_offset;
error = nfscl_findlayoutforio(layp, off, rwaccess, &rflp);
if (error == 0) {
oresid = xfer = (uint64_t)uiop->uio_resid;
if (xfer > (rflp->nfsfl_end - rflp->nfsfl_off))
xfer = rflp->nfsfl_end - rflp->nfsfl_off;
dip = nfscl_getdevinfo(nmp->nm_clp, rflp->nfsfl_dev,
rflp->nfsfl_devp);
if (dip != NULL) {
error = nfscl_doflayoutio(vp, uiop, iomode,
must_commit, &eof, &stateid, rwaccess, dip,
layp, rflp, off, xfer, newcred, p);
nfscl_reldevinfo(dip);
lastbyte = off + xfer - 1;
if (error == 0) {
NFSLOCKCLSTATE();
if (lastbyte > layp->nfsly_lastbyte)
layp->nfsly_lastbyte = lastbyte;
NFSUNLOCKCLSTATE();
}
} else
error = EIO;
if (error == 0)
len -= (oresid - (uint64_t)uiop->uio_resid);
}
}
if (lckp != NULL)
nfscl_lockderef(lckp);
NFSFREECRED(newcred);
nfscl_rellayout(layp, 0);
nfscl_relref(nmp);
return (error);
}
/*
* Find a file layout that will handle the first bytes of the requested
* range and return the information from it needed to to the I/O operation.
*/
int
nfscl_findlayoutforio(struct nfscllayout *lyp, uint64_t off, uint32_t rwaccess,
struct nfsclflayout **retflpp)
{
struct nfsclflayout *flp, *nflp, *rflp;
uint32_t rw;
rflp = NULL;
rw = rwaccess;
/* For reading, do the Read list first and then the Write list. */
do {
if (rw == NFSV4OPEN_ACCESSREAD)
flp = LIST_FIRST(&lyp->nfsly_flayread);
else
flp = LIST_FIRST(&lyp->nfsly_flayrw);
while (flp != NULL) {
nflp = LIST_NEXT(flp, nfsfl_list);
if (flp->nfsfl_off > off)
break;
if (flp->nfsfl_end > off &&
(rflp == NULL || rflp->nfsfl_end < flp->nfsfl_end))
rflp = flp;
flp = nflp;
}
if (rw == NFSV4OPEN_ACCESSREAD)
rw = NFSV4OPEN_ACCESSWRITE;
else
rw = 0;
} while (rw != 0);
if (rflp != NULL) {
/* This one covers the most bytes starting at off. */
*retflpp = rflp;
return (0);
}
return (EIO);
}
/*
* Do I/O using an NFSv4.1 file layout.
*/
static int
nfscl_doflayoutio(vnode_t vp, struct uio *uiop, int *iomode, int *must_commit,
int *eofp, nfsv4stateid_t *stateidp, int rwflag, struct nfscldevinfo *dp,
struct nfscllayout *lyp, struct nfsclflayout *flp, uint64_t off,
uint64_t len, struct ucred *cred, NFSPROC_T *p)
{
uint64_t io_off, rel_off, stripe_unit_size, transfer, xfer;
int commit_thru_mds, error = 0, stripe_index, stripe_pos;
struct nfsnode *np;
struct nfsfh *fhp;
struct nfsclds **dspp;
/*
* Search the list of nfsclds structures for one with the same
* server.
*/
cur_dsp = NULL;
TAILQ_FOREACH(dsp, &nmp->nm_sess, nfsclds_list) {
if (dsp->nfsclds_servownlen == newdsp->nfsclds_servownlen &&
dsp->nfsclds_servownlen != 0 &&
!NFSBCMP(dsp->nfsclds_serverown, newdsp->nfsclds_serverown,
dsp->nfsclds_servownlen)) {
NFSCL_DEBUG(4, "fnd same fdsp=%p dsp=%p flg=0x%x\n",
TAILQ_FIRST(&nmp->nm_sess), dsp,
dsp->nfsclds_flags);
/* Server major id matches. */
if ((dsp->nfsclds_flags & NFSCLDS_DS) != 0) {
*retdspp = dsp;
return (NFSDSP_USETHISSESSION);
}
/*
* Note the first match, so it can be used for
* sequence'ing new sessions.
*/
if (cur_dsp == NULL)
cur_dsp = dsp;
}
}
if (cur_dsp != NULL) {
*retdspp = cur_dsp;
return (NFSDSP_SEQTHISSESSION);
}
return (NFSDSP_NOTFOUND);
}
#ifdef notyet
/*
* NFS commit rpc to a DS.
*/
static int
nfsrpc_commitds(vnode_t vp, uint64_t offset, int cnt, struct nfsclds *dsp,
struct nfsfh *fhp, struct ucred *cred, NFSPROC_T *p, void *stuff)
{
uint32_t *tl;
struct nfsrv_descript nfsd, *nd = &nfsd;
struct nfsmount *nmp = VFSTONFS(vnode_mount(vp));
struct nfssockreq *nrp;
int error;
