2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * Rick Macklem at The University of Guelph.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * @(#)nfs_serv.c 8.8 (Berkeley) 7/31/95
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD$");
39 * nfs version 2 and 3 server calls to vnode ops
40 * - these routines generally have 3 phases
41 * 1 - break down and validate rpc request in mbuf list
42 * 2 - do the vnode ops for the request
43 * (surprisingly ?? many are very similar to syscalls in vfs_syscalls.c)
44 * 3 - build the rpc reply in an mbuf list
46 * - do not mix the phases, since the nfsm_?? macros can return failures
47 * on a bad rpc or similar and do not do any vrele() or vput()'s
49 * - the nfsm_reply() macro generates an nfs rpc reply with the nfs
50 * error number iff error != 0 whereas
51 * returning an error from the server function implies a fatal error
52 * such as a badly constructed rpc request that should be dropped without
54 * For nfsm_reply(), the case where error == EBADRPC is treated
55 * specially; after constructing a reply, it does an immediate
56 * `goto nfsmout' to avoid getting any V3 post-op status appended.
59 * Warning: always pay careful attention to resource cleanup on return
60 * and note that nfsm_*() macros can terminate a procedure on certain
63 * lookup() and namei()
64 * may return garbage in various structural fields/return elements
65 * if an error is returned, and may garbage up nd.ni_dvp even if no
66 * error is returned and you did not request LOCKPARENT or WANTPARENT.
68 * We use the ni_cnd.cn_flags 'HASBUF' flag to track whether the name
69 * buffer has been freed or not.
72 #include <sys/param.h>
73 #include <sys/systm.h>
75 #include <sys/namei.h>
76 #include <sys/unistd.h>
77 #include <sys/vnode.h>
78 #include <sys/mount.h>
79 #include <sys/socket.h>
80 #include <sys/socketvar.h>
81 #include <sys/malloc.h>
84 #include <sys/dirent.h>
86 #include <sys/kernel.h>
87 #include <sys/sysctl.h>
92 #include <vm/vm_extern.h>
93 #include <vm/vm_object.h>
95 #include <nfs/nfsproto.h>
96 #include <nfs/rpcv2.h>
97 #include <nfsserver/nfs.h>
98 #include <nfs/xdr_subs.h>
99 #include <nfsserver/nfsm_subs.h>
102 #define nfsdbprintf(info) printf info
104 #define nfsdbprintf(info)
107 #define MAX_COMMIT_COUNT (1024 * 1024)
109 #define NUM_HEURISTIC 1017
110 #define NHUSE_INIT 64
112 #define NHUSE_MAX 2048
114 static struct nfsheur {
115 struct vnode *nh_vp; /* vp to match (unreferenced pointer) */
116 off_t nh_nextr; /* next offset for sequential detection */
117 int nh_use; /* use count for selection */
118 int nh_seqcount; /* heuristic */
119 } nfsheur[NUM_HEURISTIC];
123 int nfsrvw_procrastinate = NFS_GATHERDELAY * 1000;
124 int nfsrvw_procrastinate_v3 = 0;
126 static struct timeval nfsver = { 0 };
128 SYSCTL_NODE(_vfs, OID_AUTO, nfsrv, CTLFLAG_RW, 0, "NFS server");
130 static int nfs_async;
131 static int nfs_commit_blks;
132 static int nfs_commit_miss;
133 SYSCTL_INT(_vfs_nfsrv, OID_AUTO, async, CTLFLAG_RW, &nfs_async, 0, "");
134 SYSCTL_INT(_vfs_nfsrv, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks, 0, "");
135 SYSCTL_INT(_vfs_nfsrv, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss, 0, "");
137 struct nfsrvstats nfsrvstats;
138 SYSCTL_STRUCT(_vfs_nfsrv, NFS_NFSRVSTATS, nfsrvstats, CTLFLAG_RW,
139 &nfsrvstats, nfsrvstats, "S,nfsrvstats");
141 static int nfsrv_access(struct vnode *, int, struct ucred *, int,
142 struct thread *, int);
143 static void nfsrvw_coalesce(struct nfsrv_descript *,
144 struct nfsrv_descript *);
147 * Clear nameidata fields that are tested in nsfmout cleanup code prior
148 * to using first nfsm macro (that might jump to the cleanup code).
152 ndclear(struct nameidata *nd)
155 nd->ni_cnd.cn_flags = 0;
158 nd->ni_startdir = NULL;
162 * Takes two vfslocked integers and returns with at most one
163 * reference to giant. The return value indicates whether giant
164 * is held by either lock. This simplifies nfsrv ops by allowing
165 * them to track only one vfslocked var.
168 nfsrv_lockedpair(int vfs1, int vfs2)
172 VFS_UNLOCK_GIANT(vfs2);
174 return (vfs1 | vfs2);
178 nfsrv_lockedpair_nd(int vfs1, struct nameidata *nd)
182 vfs2 = NDHASGIANT(nd);
184 return nfsrv_lockedpair(vfs1, vfs2);
188 * nfs v3 access service
191 nfsrv3_access(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
192 struct thread *td, struct mbuf **mrq)
194 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
195 struct sockaddr *nam = nfsd->nd_nam;
196 caddr_t dpos = nfsd->nd_dpos;
197 struct ucred *cred = nfsd->nd_cr;
198 struct vnode *vp = NULL;
203 int error = 0, rdonly, getret;
204 struct mbuf *mb, *mreq;
205 struct vattr vattr, *vap = &vattr;
206 u_long testmode, nfsmode;
207 int v3 = (nfsd->nd_flag & ND_NFSV3);
210 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
212 panic("nfsrv3_access: v3 proc called on a v2 connection");
214 fhp = &nfh.fh_generic;
216 tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED);
217 error = nfsrv_fhtovp(fhp, 1, &vp, &vfslocked, cred, slp,
220 nfsm_reply(NFSX_UNSIGNED);
221 nfsm_srvpostop_attr(1, NULL);
225 nfsmode = fxdr_unsigned(u_int32_t, *tl);
226 if ((nfsmode & NFSV3ACCESS_READ) &&
227 nfsrv_access(vp, VREAD, cred, rdonly, td, 0))
228 nfsmode &= ~NFSV3ACCESS_READ;
229 if (vp->v_type == VDIR)
230 testmode = (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
233 testmode = (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
234 if ((nfsmode & testmode) &&
235 nfsrv_access(vp, VWRITE, cred, rdonly, td, 0))
236 nfsmode &= ~testmode;
237 if (vp->v_type == VDIR)
238 testmode = NFSV3ACCESS_LOOKUP;
240 testmode = NFSV3ACCESS_EXECUTE;
241 if ((nfsmode & testmode) &&
242 nfsrv_access(vp, VEXEC, cred, rdonly, td, 0))
243 nfsmode &= ~testmode;
244 getret = VOP_GETATTR(vp, vap, cred, td);
247 nfsm_reply(NFSX_POSTOPATTR(1) + NFSX_UNSIGNED);
248 nfsm_srvpostop_attr(getret, vap);
249 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
250 *tl = txdr_unsigned(nfsmode);
254 VFS_UNLOCK_GIANT(vfslocked);
259 * nfs getattr service
262 nfsrv_getattr(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
263 struct thread *td, struct mbuf **mrq)
265 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
266 struct sockaddr *nam = nfsd->nd_nam;
267 caddr_t dpos = nfsd->nd_dpos;
268 struct ucred *cred = nfsd->nd_cr;
269 struct nfs_fattr *fp;
271 struct vattr *vap = &va;
272 struct vnode *vp = NULL;
276 int error = 0, rdonly;
277 struct mbuf *mb, *mreq;
280 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
282 fhp = &nfh.fh_generic;
284 error = nfsrv_fhtovp(fhp, 1, &vp, &vfslocked, cred, slp, nam,
291 error = VOP_GETATTR(vp, vap, cred, td);
294 nfsm_reply(NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
299 fp = nfsm_build(struct nfs_fattr *,
300 NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
301 nfsm_srvfillattr(vap, fp);
307 VFS_UNLOCK_GIANT(vfslocked);
312 * nfs setattr service
315 nfsrv_setattr(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
316 struct thread *td, struct mbuf **mrq)
318 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
319 struct sockaddr *nam = nfsd->nd_nam;
320 caddr_t dpos = nfsd->nd_dpos;
321 struct ucred *cred = nfsd->nd_cr;
322 struct vattr va, preat;
323 struct vattr *vap = &va;
324 struct nfsv2_sattr *sp;
325 struct nfs_fattr *fp;
326 struct vnode *vp = NULL;
331 int error = 0, rdonly, preat_ret = 1, postat_ret = 1;
332 int v3 = (nfsd->nd_flag & ND_NFSV3), gcheck = 0;
333 struct mbuf *mb, *mreq;
334 struct timespec guard = { 0, 0 };
335 struct mount *mp = NULL;
339 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
341 fhp = &nfh.fh_generic;
343 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
347 vfslocked = VFS_LOCK_GIANT(mp);
348 (void) vn_start_write(NULL, &mp, V_WAIT);
349 vfs_rel(mp); /* The write holds a ref. */
353 tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED);
354 gcheck = fxdr_unsigned(int, *tl);
356 tl = nfsm_dissect_nonblock(u_int32_t *, 2 * NFSX_UNSIGNED);
357 fxdr_nfsv3time(tl, &guard);
360 sp = nfsm_dissect_nonblock(struct nfsv2_sattr *, NFSX_V2SATTR);
362 * Nah nah nah nah na nah
363 * There is a bug in the Sun client that puts 0xffff in the mode
364 * field of sattr when it should put in 0xffffffff. The u_short
365 * doesn't sign extend.
366 * --> check the low order 2 bytes for 0xffff
368 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
369 vap->va_mode = nfstov_mode(sp->sa_mode);
370 if (sp->sa_uid != nfsrv_nfs_xdrneg1)
371 vap->va_uid = fxdr_unsigned(uid_t, sp->sa_uid);
372 if (sp->sa_gid != nfsrv_nfs_xdrneg1)
373 vap->va_gid = fxdr_unsigned(gid_t, sp->sa_gid);
374 if (sp->sa_size != nfsrv_nfs_xdrneg1)
375 vap->va_size = fxdr_unsigned(u_quad_t, sp->sa_size);
376 if (sp->sa_atime.nfsv2_sec != nfsrv_nfs_xdrneg1) {
378 fxdr_nfsv2time(&sp->sa_atime, &vap->va_atime);
380 vap->va_atime.tv_sec =
381 fxdr_unsigned(int32_t, sp->sa_atime.nfsv2_sec);
382 vap->va_atime.tv_nsec = 0;
385 if (sp->sa_mtime.nfsv2_sec != nfsrv_nfs_xdrneg1)
386 fxdr_nfsv2time(&sp->sa_mtime, &vap->va_mtime);
391 * Now that we have all the fields, lets do it.
393 error = nfsrv_fhtovp(fhp, 1, &vp, &tvfslocked, cred, slp,
395 vfslocked = nfsrv_lockedpair(vfslocked, tvfslocked);
397 nfsm_reply(2 * NFSX_UNSIGNED);
399 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap);
405 * vp now an active resource, pay careful attention to cleanup
408 error = preat_ret = VOP_GETATTR(vp, &preat, cred, td);
409 if (!error && gcheck &&
410 (preat.va_ctime.tv_sec != guard.tv_sec ||
411 preat.va_ctime.tv_nsec != guard.tv_nsec))
412 error = NFSERR_NOT_SYNC;
416 nfsm_reply(NFSX_WCCDATA(v3));
418 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap);
425 * If the size is being changed write acces is required, otherwise
426 * just check for a read only filesystem.
428 if (vap->va_size == ((u_quad_t)((quad_t) -1))) {
429 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
434 if (vp->v_type == VDIR) {
437 } else if ((error = nfsrv_access(vp, VWRITE, cred, rdonly,
441 error = VOP_SETATTR(vp, vap, cred, td);
442 postat_ret = VOP_GETATTR(vp, vap, cred, td);
450 nfsm_reply(NFSX_WCCORFATTR(v3));
452 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap);
454 /* v2 non-error case. */
455 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
456 nfsm_srvfillattr(vap, fp);
464 vn_finished_write(mp);
465 VFS_UNLOCK_GIANT(vfslocked);
473 nfsrv_lookup(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
474 struct thread *td, struct mbuf **mrq)
476 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
477 struct sockaddr *nam = nfsd->nd_nam;
478 caddr_t dpos = nfsd->nd_dpos;
479 struct ucred *cred = nfsd->nd_cr;
480 struct nfs_fattr *fp;
481 struct nameidata nd, ind, *ndp = &nd;
482 struct vnode *vp, *dirp = NULL;
486 int error = 0, len, dirattr_ret = 1;
487 int v3 = (nfsd->nd_flag & ND_NFSV3), pubflag;
488 struct mbuf *mb, *mreq;
489 struct vattr va, dirattr, *vap = &va;
493 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
497 fhp = &nfh.fh_generic;
499 nfsm_srvnamesiz(len);
501 pubflag = nfs_ispublicfh(fhp);
503 nd.ni_cnd.cn_cred = cred;
504 nd.ni_cnd.cn_nameiop = LOOKUP;
505 nd.ni_cnd.cn_flags = LOCKLEAF | SAVESTART | MPSAFE;
506 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
507 &dirp, v3, &dirattr, &dirattr_ret, td, pubflag);
508 vfslocked = NDHASGIANT(&nd);
511 * namei failure, only dirp to cleanup. Clear out garbarge from
512 * structure in case macros jump to nfsmout.
520 nfsm_reply(NFSX_POSTOPATTR(v3));
522 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
528 * Locate index file for public filehandle
530 * error is 0 on entry and 0 on exit from this block.
534 if (nd.ni_vp->v_type == VDIR && nfs_pub.np_index != NULL) {
536 * Setup call to lookup() to see if we can find
537 * the index file. Arguably, this doesn't belong
538 * in a kernel.. Ugh. If an error occurs, do not
539 * try to install an index file and then clear the
542 * When we replace nd with ind and redirect ndp,
543 * maintenance of ni_startdir and ni_vp shift to
544 * ind and we have to clean them up in the old nd.
545 * However, the cnd resource continues to be maintained
546 * via the original nd. Confused? You aren't alone!
549 VOP_UNLOCK(nd.ni_vp, 0, td);
550 ind.ni_pathlen = strlen(nfs_pub.np_index);
551 ind.ni_cnd.cn_nameptr = ind.ni_cnd.cn_pnbuf =
553 ind.ni_startdir = nd.ni_vp;
554 VREF(ind.ni_startdir);
555 ind.ni_cnd.cn_flags &= ~GIANTHELD;
556 tvfslocked = VFS_LOCK_GIANT(ind.ni_startdir->v_mount);
558 nd.ni_cnd.cn_flags |= GIANTHELD;
559 error = lookup(&ind);
561 vfslocked = nfsrv_lockedpair_nd(vfslocked, &ind);
562 ind.ni_cnd.cn_flags &= ~GIANTHELD;
566 * Found an index file. Get rid of
567 * the old references. transfer nd.ni_vp'
573 vrele(nd.ni_startdir);
574 nd.ni_startdir = NULL;
580 * If the public filehandle was used, check that this lookup
581 * didn't result in a filehandle outside the publicly exported
582 * filesystem. We clear the poor vp here to avoid lockups due
586 if (ndp->ni_vp->v_mount != nfs_pub.np_mount) {
594 * Resources at this point:
595 * ndp->ni_vp may not be NULL
599 nfsm_reply(NFSX_POSTOPATTR(v3));
601 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
607 * Get underlying attribute, then release remaining resources ( for
608 * the same potential blocking reason ) and reply.
611 bzero((caddr_t)fhp, sizeof(nfh));
612 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
613 error = VOP_VPTOFH(vp, &fhp->fh_fid);
615 error = VOP_GETATTR(vp, vap, cred, td);
618 vrele(ndp->ni_startdir);
621 ndp->ni_startdir = NULL;
623 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPORFATTR(v3) + NFSX_POSTOPATTR(v3));
626 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
630 nfsm_srvfhtom(fhp, v3);
632 nfsm_srvpostop_attr(0, vap);
633 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
635 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
636 nfsm_srvfillattr(vap, fp);
640 if (ndp->ni_vp || dirp || ndp->ni_startdir) {
645 if (ndp->ni_startdir)
646 vrele(ndp->ni_startdir);
648 NDFREE(&nd, NDF_ONLY_PNBUF);
649 VFS_UNLOCK_GIANT(vfslocked);
654 * nfs readlink service
657 nfsrv_readlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
658 struct thread *td, struct mbuf **mrq)
660 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
661 struct sockaddr *nam = nfsd->nd_nam;
662 caddr_t dpos = nfsd->nd_dpos;
663 struct ucred *cred = nfsd->nd_cr;
664 struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
665 struct iovec *ivp = iv;
669 int error = 0, rdonly, i, tlen, len, getret;
670 int v3 = (nfsd->nd_flag & ND_NFSV3);
671 struct mbuf *mb, *mp3, *nmp, *mreq;
672 struct vnode *vp = NULL;
676 struct uio io, *uiop = &io;
679 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
685 fhp = &nfh.fh_generic;
689 while (len < NFS_MAXPATHLEN) {
690 MGET(nmp, M_TRYWAIT, MT_DATA);
691 MCLGET(nmp, M_TRYWAIT);
692 nmp->m_len = NFSMSIZ(nmp);
699 if ((len + mp->m_len) > NFS_MAXPATHLEN) {
700 mp->m_len = NFS_MAXPATHLEN - len;
701 len = NFS_MAXPATHLEN;
704 ivp->iov_base = mtod(mp, caddr_t);
705 ivp->iov_len = mp->m_len;
710 uiop->uio_iovcnt = i;
711 uiop->uio_offset = 0;
712 uiop->uio_resid = len;
713 uiop->uio_rw = UIO_READ;
714 uiop->uio_segflg = UIO_SYSSPACE;
716 error = nfsrv_fhtovp(fhp, 1, &vp, &vfslocked, cred, slp,
719 nfsm_reply(2 * NFSX_UNSIGNED);
721 nfsm_srvpostop_attr(1, NULL);
725 if (vp->v_type != VLNK) {
731 error = VOP_READLINK(vp, uiop, cred);
732 getret = VOP_GETATTR(vp, &attr, cred, td);
735 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_UNSIGNED);
737 nfsm_srvpostop_attr(getret, &attr);
742 if (uiop->uio_resid > 0) {
743 len -= uiop->uio_resid;
744 tlen = nfsm_rndup(len);
745 nfsm_adj(mp3, NFS_MAXPATHLEN-tlen, tlen-len);
747 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
748 *tl = txdr_unsigned(len);
756 VFS_UNLOCK_GIANT(vfslocked);
764 nfsrv_read(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
765 struct thread *td, struct mbuf **mrq)
767 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
768 struct sockaddr *nam = nfsd->nd_nam;
769 caddr_t dpos = nfsd->nd_dpos;
770 struct ucred *cred = nfsd->nd_cr;
774 struct nfs_fattr *fp;
778 int error = 0, rdonly, cnt, len, left, siz, tlen, getret;
779 int v3 = (nfsd->nd_flag & ND_NFSV3), reqlen;
780 struct mbuf *mb, *mreq;
782 struct vnode *vp = NULL;
785 struct uio io, *uiop = &io;
786 struct vattr va, *vap = &va;
793 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
795 fhp = &nfh.fh_generic;
798 tl = nfsm_dissect_nonblock(u_int32_t *, 2 * NFSX_UNSIGNED);
799 off = fxdr_hyper(tl);
801 tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED);
802 off = (off_t)fxdr_unsigned(u_int32_t, *tl);
804 nfsm_srvstrsiz(reqlen, NFS_SRVMAXDATA(nfsd));
807 * Reference vp. If an error occurs, vp will be invalid, but we
808 * have to NULL it just in case. The macros might goto nfsmout
812 error = nfsrv_fhtovp(fhp, 1, &vp, &vfslocked, cred, slp,
816 nfsm_reply(2 * NFSX_UNSIGNED);
818 nfsm_srvpostop_attr(1, NULL);
823 if (vp->v_type != VREG) {
827 error = (vp->v_type == VDIR) ? EISDIR : EACCES;
830 if ((error = nfsrv_access(vp, VREAD, cred, rdonly,
832 error = nfsrv_access(vp, VEXEC, cred,
835 getret = VOP_GETATTR(vp, vap, cred, td);
841 nfsm_reply(NFSX_POSTOPATTR(v3));
843 nfsm_srvpostop_attr(getret, vap);
849 * Calculate byte count to read
852 if (off >= vap->va_size)
854 else if ((off + reqlen) > vap->va_size)
855 cnt = vap->va_size - off;
860 * Calculate seqcount for heuristic
868 * Locate best candidate
871 hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
875 if (nfsheur[hi].nh_vp == vp) {
879 if (nfsheur[hi].nh_use > 0)
880 --nfsheur[hi].nh_use;
881 hi = (hi + 1) % NUM_HEURISTIC;
882 if (nfsheur[hi].nh_use < nh->nh_use)
886 if (nh->nh_vp != vp) {
889 nh->nh_use = NHUSE_INIT;
897 * Calculate heuristic
900 if ((off == 0 && nh->nh_seqcount > 0) || off == nh->nh_nextr) {
901 if (++nh->nh_seqcount > IO_SEQMAX)
902 nh->nh_seqcount = IO_SEQMAX;
903 } else if (nh->nh_seqcount > 1) {
908 nh->nh_use += NHUSE_INC;
909 if (nh->nh_use > NHUSE_MAX)
910 nh->nh_use = NHUSE_MAX;
911 ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
914 nfsm_reply(NFSX_POSTOPORFATTR(v3) + 3 * NFSX_UNSIGNED+nfsm_rndup(cnt));
916 tl = nfsm_build(u_int32_t *, NFSX_V3FATTR + 4 * NFSX_UNSIGNED);
917 *tl++ = nfsrv_nfs_true;
918 fp = (struct nfs_fattr *)tl;
919 tl += (NFSX_V3FATTR / sizeof (u_int32_t));
921 tl = nfsm_build(u_int32_t *, NFSX_V2FATTR + NFSX_UNSIGNED);
922 fp = (struct nfs_fattr *)tl;
923 tl += (NFSX_V2FATTR / sizeof (u_int32_t));
925 len = left = nfsm_rndup(cnt);
928 * Generate the mbuf list with the uio_iov ref. to it.
933 siz = min(M_TRAILINGSPACE(m), left);
939 MGET(m, M_TRYWAIT, MT_DATA);
940 MCLGET(m, M_TRYWAIT);
946 MALLOC(iv, struct iovec *, i * sizeof (struct iovec),
948 uiop->uio_iov = iv2 = iv;
954 panic("nfsrv_read iov");
955 siz = min(M_TRAILINGSPACE(m), left);
957 iv->iov_base = mtod(m, caddr_t) + m->m_len;
966 uiop->uio_iovcnt = i;
967 uiop->uio_offset = off;
968 uiop->uio_resid = len;
969 uiop->uio_rw = UIO_READ;
970 uiop->uio_segflg = UIO_SYSSPACE;
971 error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
972 off = uiop->uio_offset;
974 FREE((caddr_t)iv2, M_TEMP);
975 if (error || (getret = VOP_GETATTR(vp, vap, cred, td))) {
981 nfsm_reply(NFSX_POSTOPATTR(v3));
983 nfsm_srvpostop_attr(getret, vap);
991 nfsm_srvfillattr(vap, fp);
992 tlen = len - uiop->uio_resid;
993 cnt = cnt < tlen ? cnt : tlen;
994 tlen = nfsm_rndup(cnt);
995 if (len != tlen || tlen != cnt)
996 nfsm_adj(mb, len - tlen, tlen - cnt);
998 *tl++ = txdr_unsigned(cnt);
1000 *tl++ = nfsrv_nfs_true;
1002 *tl++ = nfsrv_nfs_false;
1004 *tl = txdr_unsigned(cnt);
1008 VFS_UNLOCK_GIANT(vfslocked);
1016 nfsrv_write(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
1017 struct thread *td, struct mbuf **mrq)
1019 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1020 struct sockaddr *nam = nfsd->nd_nam;
1021 caddr_t dpos = nfsd->nd_dpos;
1022 struct ucred *cred = nfsd->nd_cr;
1026 struct nfs_fattr *fp;
1028 struct vattr va, forat;
1029 struct vattr *vap = &va;
1032 int error = 0, rdonly, len, forat_ret = 1;
1033 int ioflags, aftat_ret = 1, retlen = 0, zeroing, adjust;
1034 int stable = NFSV3WRITE_FILESYNC;
1035 int v3 = (nfsd->nd_flag & ND_NFSV3);
1036 struct mbuf *mb, *mreq;
1037 struct vnode *vp = NULL;
1040 struct uio io, *uiop = &io;
1042 struct mount *mntp = NULL;
1046 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1053 fhp = &nfh.fh_generic;
1055 if ((mntp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
1059 vfslocked = VFS_LOCK_GIANT(mntp);
1060 (void) vn_start_write(NULL, &mntp, V_WAIT);
1061 vfs_rel(mntp); /* The write holds a ref. */
1063 tl = nfsm_dissect_nonblock(u_int32_t *, 5 * NFSX_UNSIGNED);
1064 off = fxdr_hyper(tl);
1066 stable = fxdr_unsigned(int, *tl++);
1068 tl = nfsm_dissect_nonblock(u_int32_t *, 4 * NFSX_UNSIGNED);
1069 off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
1072 stable = NFSV3WRITE_UNSTABLE;
1074 retlen = len = fxdr_unsigned(int32_t, *tl);
1078 * For NFS Version 2, it is not obvious what a write of zero length
1079 * should do, but I might as well be consistent with Version 3,
1080 * which is to return ok so long as there are no permission problems.
1088 adjust = dpos - mtod(mp, caddr_t);
1089 mp->m_len -= adjust;
1090 if (mp->m_len > 0 && adjust > 0)
1091 mp->m_data += adjust;
1095 else if (mp->m_len > 0) {
1098 mp->m_len -= (i - len);
1107 if (len > NFS_MAXDATA || len < 0 || i < len) {
1109 nfsm_reply(2 * NFSX_UNSIGNED);
1111 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1115 error = nfsrv_fhtovp(fhp, 1, &vp, &tvfslocked, cred, slp,
1116 nam, &rdonly, TRUE);
1117 vfslocked = nfsrv_lockedpair(vfslocked, tvfslocked);
1120 nfsm_reply(2 * NFSX_UNSIGNED);
1122 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1127 forat_ret = VOP_GETATTR(vp, &forat, cred, td);
1128 if (vp->v_type != VREG) {
1132 error = (vp->v_type == VDIR) ? EISDIR : EACCES;
1135 error = nfsrv_access(vp, VWRITE, cred, rdonly, td, 1);
1139 nfsm_reply(NFSX_WCCDATA(v3));
1141 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1147 MALLOC(ivp, struct iovec *, cnt * sizeof (struct iovec), M_TEMP,
1149 uiop->uio_iov = iv = ivp;
1150 uiop->uio_iovcnt = cnt;
1153 if (mp->m_len > 0) {
1154 ivp->iov_base = mtod(mp, caddr_t);
1155 ivp->iov_len = mp->m_len;
1163 * The IO_METASYNC flag indicates that all metadata (and not just
1164 * enough to ensure data integrity) mus be written to stable storage
1166 * (IO_METASYNC is not yet implemented in 4.4BSD-Lite.)
1168 if (stable == NFSV3WRITE_UNSTABLE)
1169 ioflags = IO_NODELOCKED;
1170 else if (stable == NFSV3WRITE_DATASYNC)
1171 ioflags = (IO_SYNC | IO_NODELOCKED);
1173 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
1174 uiop->uio_resid = len;
1175 uiop->uio_rw = UIO_WRITE;
1176 uiop->uio_segflg = UIO_SYSSPACE;
1177 uiop->uio_td = NULL;
1178 uiop->uio_offset = off;
1179 error = VOP_WRITE(vp, uiop, ioflags, cred);
1180 /* Unlocked write. */
1181 nfsrvstats.srvvop_writes++;
1182 FREE((caddr_t)iv, M_TEMP);
1184 aftat_ret = VOP_GETATTR(vp, vap, cred, td);
1190 nfsm_reply(NFSX_PREOPATTR(v3) + NFSX_POSTOPORFATTR(v3) +
1191 2 * NFSX_UNSIGNED + NFSX_WRITEVERF(v3));
1193 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1198 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
1199 *tl++ = txdr_unsigned(retlen);
1201 * If nfs_async is set, then pretend the write was FILESYNC.
1203 if (stable == NFSV3WRITE_UNSTABLE && !nfs_async)
1204 *tl++ = txdr_unsigned(stable);
1206 *tl++ = txdr_unsigned(NFSV3WRITE_FILESYNC);
1208 * Actually, there is no need to txdr these fields,
1209 * but it may make the values more human readable,
1210 * for debugging purposes.
1212 if (nfsver.tv_sec == 0)
1214 *tl++ = txdr_unsigned(nfsver.tv_sec);
1215 *tl = txdr_unsigned(nfsver.tv_usec);
1216 } else if (!error) {
1217 /* v2 non-error case. */
1218 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
1219 nfsm_srvfillattr(vap, fp);
1225 vn_finished_write(mntp);
1226 VFS_UNLOCK_GIANT(vfslocked);
1231 * For the purposes of write gathering, we must decide if the credential
1232 * associated with two pending requests have equivilent privileges. Since
1233 * NFS only uses a subset of the BSD ucred -- the effective uid and group
1234 * IDs -- we have a compare routine that checks only the relevant fields.
1237 nfsrv_samecred(struct ucred *cr1, struct ucred *cr2)
1241 if (cr1->cr_uid != cr2->cr_uid)
1243 if (cr1->cr_ngroups != cr2->cr_ngroups)
1245 for (i = 0; i < cr1->cr_ngroups; i++) {
1246 if (cr1->cr_groups[i] != cr2->cr_groups[i])
1253 * NFS write service with write gathering support. Called when
1254 * nfsrvw_procrastinate > 0.
1255 * See: Chet Juszczak, "Improving the Write Performance of an NFS Server",
1256 * in Proc. of the Winter 1994 Usenix Conference, pg. 247-259, San Franscisco,
1260 nfsrv_writegather(struct nfsrv_descript **ndp, struct nfssvc_sock *slp,
1261 struct thread *td, struct mbuf **mrq)
1265 struct nfsrv_descript *wp, *nfsd, *owp, *swp;
1266 struct nfs_fattr *fp;
1269 struct nfsrvw_delayhash *wpp;
1271 struct vattr va, forat;
1274 int error = 0, rdonly, len, forat_ret = 1;
1275 int ioflags, aftat_ret = 1, s, adjust, v3, zeroing;
1276 struct mbuf *mb, *mreq, *mrep, *md;
1277 struct vnode *vp = NULL;
1278 struct uio io, *uiop = &io;
1280 struct mount *mntp = NULL;
1284 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1294 mrep = nfsd->nd_mrep;
1296 dpos = nfsd->nd_dpos;
1298 v3 = (nfsd->nd_flag & ND_NFSV3);
1299 LIST_INIT(&nfsd->nd_coalesce);
1300 nfsd->nd_mreq = NULL;
1301 nfsd->nd_stable = NFSV3WRITE_FILESYNC;
1302 cur_usec = nfs_curusec();
1303 nfsd->nd_time = cur_usec +
1304 (v3 ? nfsrvw_procrastinate_v3 : nfsrvw_procrastinate);
1307 * Now, get the write header..
1309 nfsm_srvmtofh(&nfsd->nd_fh);
1311 tl = nfsm_dissect_nonblock(u_int32_t *, 5 * NFSX_UNSIGNED);
1312 nfsd->nd_off = fxdr_hyper(tl);
1314 nfsd->nd_stable = fxdr_unsigned(int, *tl++);
1316 tl = nfsm_dissect_nonblock(u_int32_t *, 4 * NFSX_UNSIGNED);
1317 nfsd->nd_off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
1320 nfsd->nd_stable = NFSV3WRITE_UNSTABLE;
1322 len = fxdr_unsigned(int32_t, *tl);
1324 nfsd->nd_eoff = nfsd->nd_off + len;
1327 * Trim the header out of the mbuf list and trim off any trailing
1328 * junk so that the mbuf list has only the write data.
1336 adjust = dpos - mtod(mp, caddr_t);
1337 mp->m_len -= adjust;
1338 if (mp->m_len > 0 && adjust > 0)
1339 mp->m_data += adjust;
1346 mp->m_len -= (i - len);
1352 if (len > NFS_MAXDATA || len < 0 || i < len) {
1356 nfsm_writereply(2 * NFSX_UNSIGNED);
1358 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1359 nfsd->nd_mreq = mreq;
1360 nfsd->nd_mrep = NULL;
1365 * Add this entry to the hash and time queues.
1369 wp = LIST_FIRST(&slp->ns_tq);
1370 while (wp && wp->nd_time < nfsd->nd_time) {
1372 wp = LIST_NEXT(wp, nd_tq);
1374 NFS_DPF(WG, ("Q%03x", nfsd->nd_retxid & 0xfff));
1376 LIST_INSERT_AFTER(owp, nfsd, nd_tq);
1378 LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1380 if (nfsd->nd_mrep) {
1381 wpp = NWDELAYHASH(slp, nfsd->nd_fh.fh_fid.fid_data);
1383 wp = LIST_FIRST(wpp);
1385 bcmp((caddr_t)&nfsd->nd_fh,(caddr_t)&wp->nd_fh, NFSX_V3FH)){
1387 wp = LIST_NEXT(wp, nd_hash);
1389 while (wp && wp->nd_off < nfsd->nd_off &&
1390 !bcmp((caddr_t)&nfsd->nd_fh,(caddr_t)&wp->nd_fh, NFSX_V3FH)) {
1392 wp = LIST_NEXT(wp, nd_hash);
1395 LIST_INSERT_AFTER(owp, nfsd, nd_hash);
1398 * Search the hash list for overlapping entries and
1401 for(; nfsd && NFSW_CONTIG(owp, nfsd); nfsd = wp) {
1402 wp = LIST_NEXT(nfsd, nd_hash);
1403 if (nfsrv_samecred(owp->nd_cr, nfsd->nd_cr))
1404 nfsrvw_coalesce(owp, nfsd);
1407 LIST_INSERT_HEAD(wpp, nfsd, nd_hash);
1414 * Now, do VOP_WRITE()s for any one(s) that need to be done now
1415 * and generate the associated reply mbuf list(s).
1418 cur_usec = nfs_curusec();
1420 for (nfsd = LIST_FIRST(&slp->ns_tq); nfsd; nfsd = owp) {
1421 owp = LIST_NEXT(nfsd, nd_tq);
1422 if (nfsd->nd_time > cur_usec)
1426 NFS_DPF(WG, ("P%03x", nfsd->nd_retxid & 0xfff));
1427 LIST_REMOVE(nfsd, nd_tq);
1428 LIST_REMOVE(nfsd, nd_hash);
1430 mrep = nfsd->nd_mrep;
1431 nfsd->nd_mrep = NULL;
1433 v3 = (nfsd->nd_flag & ND_NFSV3);
1434 forat_ret = aftat_ret = 1;
1435 error = nfsrv_fhtovp(&nfsd->nd_fh, 1, &vp, &vfslocked, cred,
1436 slp, nfsd->nd_nam, &rdonly, TRUE);
1439 forat_ret = VOP_GETATTR(vp, &forat, cred, td);
1440 if (vp->v_type != VREG) {
1444 error = (vp->v_type == VDIR) ? EISDIR : EACCES;
1450 error = nfsrv_access(vp, VWRITE, cred, rdonly,
1452 if (nfsd->nd_stable == NFSV3WRITE_UNSTABLE)
1453 ioflags = IO_NODELOCKED;
1454 else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC)
1455 ioflags = (IO_SYNC | IO_NODELOCKED);
1457 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
1458 uiop->uio_rw = UIO_WRITE;
1459 uiop->uio_segflg = UIO_SYSSPACE;
1460 uiop->uio_td = NULL;
1461 uiop->uio_offset = nfsd->nd_off;
1462 uiop->uio_resid = nfsd->nd_eoff - nfsd->nd_off;
1463 if (uiop->uio_resid > 0) {
1471 uiop->uio_iovcnt = i;
1472 MALLOC(iov, struct iovec *, i * sizeof (struct iovec),
1474 uiop->uio_iov = ivp = iov;
1477 if (mp->m_len > 0) {
1478 ivp->iov_base = mtod(mp, caddr_t);
1479 ivp->iov_len = mp->m_len;
1486 if (vn_start_write(vp, &mntp, V_NOWAIT) != 0) {
1487 VOP_UNLOCK(vp, 0, td);
1488 error = vn_start_write(NULL, &mntp, V_WAIT);
1489 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1491 mvfslocked = VFS_LOCK_GIANT(mntp);
1494 error = VOP_WRITE(vp, uiop, ioflags, cred);
1495 /* Unlocked write. */
1496 nfsrvstats.srvvop_writes++;
1497 vn_finished_write(mntp);
1499 VFS_UNLOCK_GIANT(mvfslocked);
1500 FREE((caddr_t)iov, M_TEMP);
1504 aftat_ret = VOP_GETATTR(vp, &va, cred, td);
1508 VFS_UNLOCK_GIANT(vfslocked);
1510 * Loop around generating replies for all write rpcs that have
1511 * now been completed.
1515 NFS_DPF(WG, ("R%03x", nfsd->nd_retxid & 0xfff));
1517 nfsm_writereply(NFSX_WCCDATA(v3));
1519 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1522 nfsm_writereply(NFSX_PREOPATTR(v3) +
1523 NFSX_POSTOPORFATTR(v3) + 2 * NFSX_UNSIGNED +
1524 NFSX_WRITEVERF(v3));
1526 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1527 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
1528 *tl++ = txdr_unsigned(nfsd->nd_len);
1529 *tl++ = txdr_unsigned(swp->nd_stable);
1531 * Actually, there is no need to txdr these fields,
1532 * but it may make the values more human readable,
1533 * for debugging purposes.
1535 if (nfsver.tv_sec == 0)
1537 *tl++ = txdr_unsigned(nfsver.tv_sec);
1538 *tl = txdr_unsigned(nfsver.tv_usec);
1540 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
1541 nfsm_srvfillattr(&va, fp);
1544 nfsd->nd_mreq = mreq;
1546 panic("nfsrv_write: nd_mrep not free");
1549 * Done. Put it at the head of the timer queue so that
1550 * the final phase can return the reply.
1555 LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1557 nfsd = LIST_FIRST(&swp->nd_coalesce);
1559 LIST_REMOVE(nfsd, nd_tq);
1565 LIST_INSERT_HEAD(&slp->ns_tq, swp, nd_tq);
1572 * Search for a reply to return.
1575 LIST_FOREACH(nfsd, &slp->ns_tq, nd_tq)
1576 if (nfsd->nd_mreq) {
1577 NFS_DPF(WG, ("X%03x", nfsd->nd_retxid & 0xfff));
1578 LIST_REMOVE(nfsd, nd_tq);
1579 *mrq = nfsd->nd_mreq;
1588 * Coalesce the write request nfsd into owp. To do this we must:
1589 * - remove nfsd from the queues
1590 * - merge nfsd->nd_mrep into owp->nd_mrep
1591 * - update the nd_eoff and nd_stable for owp
1592 * - put nfsd on owp's nd_coalesce list
1593 * NB: Must be called at splsoftclock().
1596 nfsrvw_coalesce(struct nfsrv_descript *owp, struct nfsrv_descript *nfsd)
1600 struct nfsrv_descript *p;
1602 NFS_DPF(WG, ("C%03x-%03x",
1603 nfsd->nd_retxid & 0xfff, owp->nd_retxid & 0xfff));
1604 LIST_REMOVE(nfsd, nd_hash);
1605 LIST_REMOVE(nfsd, nd_tq);
1606 if (owp->nd_eoff < nfsd->nd_eoff) {
1607 overlap = owp->nd_eoff - nfsd->nd_off;
1609 panic("nfsrv_coalesce: bad off");
1611 m_adj(nfsd->nd_mrep, overlap);
1615 mp->m_next = nfsd->nd_mrep;
1616 owp->nd_eoff = nfsd->nd_eoff;
1618 m_freem(nfsd->nd_mrep);
1619 nfsd->nd_mrep = NULL;
1620 if (nfsd->nd_stable == NFSV3WRITE_FILESYNC)
1621 owp->nd_stable = NFSV3WRITE_FILESYNC;
1622 else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC &&
1623 owp->nd_stable == NFSV3WRITE_UNSTABLE)
1624 owp->nd_stable = NFSV3WRITE_DATASYNC;
1625 LIST_INSERT_HEAD(&owp->nd_coalesce, nfsd, nd_tq);
1628 * If nfsd had anything else coalesced into it, transfer them
1629 * to owp, otherwise their replies will never get sent.
1631 for (p = LIST_FIRST(&nfsd->nd_coalesce); p;
1632 p = LIST_FIRST(&nfsd->nd_coalesce)) {
1633 LIST_REMOVE(p, nd_tq);
1634 LIST_INSERT_HEAD(&owp->nd_coalesce, p, nd_tq);
1639 * nfs create service
1640 * now does a truncate to 0 length via. setattr if it already exists
1643 nfsrv_create(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
1644 struct thread *td, struct mbuf **mrq)
1646 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1647 struct sockaddr *nam = nfsd->nd_nam;
1648 caddr_t dpos = nfsd->nd_dpos;
1649 struct ucred *cred = nfsd->nd_cr;
1650 struct nfs_fattr *fp;
1651 struct vattr va, dirfor, diraft;
1652 struct vattr *vap = &va;
1653 struct nfsv2_sattr *sp;
1655 struct nameidata nd;
1657 int error = 0, rdev, len, tsize, dirfor_ret = 1, diraft_ret = 1;
1658 int v3 = (nfsd->nd_flag & ND_NFSV3), how, exclusive_flag = 0;
1660 struct mbuf *mb, *mreq;
1661 struct vnode *dirp = NULL;
1665 u_char cverf[NFSX_V3CREATEVERF];
1666 struct mount *mp = NULL;
1670 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1677 fhp = &nfh.fh_generic;
1679 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
1683 vfslocked = VFS_LOCK_GIANT(mp);
1684 (void) vn_start_write(NULL, &mp, V_WAIT);
1685 vfs_rel(mp); /* The write holds a ref. */
1686 nfsm_srvnamesiz(len);
1688 nd.ni_cnd.cn_cred = cred;
1689 nd.ni_cnd.cn_nameiop = CREATE;
1690 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | SAVESTART | MPSAFE;
1693 * Call namei and do initial cleanup to get a few things
1694 * out of the way. If we get an initial error we cleanup
1695 * and return here to avoid special-casing the invalid nd
1696 * structure through the rest of the case. dirp may be
1697 * set even if an error occurs, but the nd structure will not
1698 * be valid at all if an error occurs so we have to invalidate it
1699 * prior to calling nfsm_reply ( which might goto nfsmout ).
1701 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
1702 &dirp, v3, &dirfor, &dirfor_ret, td, FALSE);
1703 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
1709 nfsm_reply(NFSX_WCCDATA(v3));
1711 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1717 * No error. Continue. State:
1719 * startdir is valid ( we release this immediately )
1721 * nd.ni_vp may be valid
1722 * nd.ni_dvp is valid
1724 * The error state is set through the code and we may also do some
1725 * opportunistic releasing of vnodes to avoid holding locks through
1726 * NFS I/O. The cleanup at the end is a catch-all
1731 tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED);
1732 how = fxdr_unsigned(int, *tl);
1734 case NFSV3CREATE_GUARDED:
1740 case NFSV3CREATE_UNCHECKED:
1743 case NFSV3CREATE_EXCLUSIVE:
1744 cp = nfsm_dissect_nonblock(caddr_t, NFSX_V3CREATEVERF);
1745 bcopy(cp, cverf, NFSX_V3CREATEVERF);
1749 vap->va_type = VREG;
1751 sp = nfsm_dissect_nonblock(struct nfsv2_sattr *, NFSX_V2SATTR);
1752 vap->va_type = IFTOVT(fxdr_unsigned(u_int32_t, sp->sa_mode));
1753 if (vap->va_type == VNON)
1754 vap->va_type = VREG;
1755 vap->va_mode = nfstov_mode(sp->sa_mode);
1756 switch (vap->va_type) {
1758 tsize = fxdr_unsigned(int32_t, sp->sa_size);
1760 vap->va_size = (u_quad_t)tsize;
1765 rdev = fxdr_unsigned(long, sp->sa_size);
1773 * Iff doesn't exist, create it
1774 * otherwise just truncate to 0 length
1775 * should I set the mode too ?
1777 * The only possible error we can have at this point is EEXIST.
1778 * nd.ni_vp will also be non-NULL in that case.
1780 if (nd.ni_vp == NULL) {
1781 if (vap->va_mode == (mode_t)VNOVAL)
1783 if (vap->va_type == VREG || vap->va_type == VSOCK) {
1784 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap);
1786 NDFREE(&nd, NDF_ONLY_PNBUF);
1788 if (exclusive_flag) {
1791 bcopy(cverf, (caddr_t)&vap->va_atime,
1793 error = VOP_SETATTR(nd.ni_vp, vap, cred,
1797 } else if (vap->va_type == VCHR || vap->va_type == VBLK ||
1798 vap->va_type == VFIFO) {
1800 * NFSv2-specific code for creating device nodes
1803 * Handle SysV FIFO node special cases. All other
1804 * devices require super user to access.
1806 if (vap->va_type == VCHR && rdev == 0xffffffff)
1807 vap->va_type = VFIFO;
1808 if (vap->va_type != VFIFO &&
1809 (error = suser_cred(cred, 0))) {
1812 vap->va_rdev = rdev;
1813 error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap);
1815 NDFREE(&nd, NDF_ONLY_PNBUF);
1822 * release dvp prior to lookup
1829 * Even though LOCKPARENT was cleared, ni_dvp may
1832 nd.ni_cnd.cn_nameiop = LOOKUP;
1833 nd.ni_cnd.cn_flags &= ~(LOCKPARENT);
1834 nd.ni_cnd.cn_thread = td;
1835 nd.ni_cnd.cn_cred = cred;
1836 tvfslocked = VFS_LOCK_GIANT(nd.ni_startdir->v_mount);
1838 nd.ni_cnd.cn_flags |= GIANTHELD;
1839 error = lookup(&nd);
1841 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
1842 nd.ni_cnd.cn_flags &= ~GIANTHELD;
1846 if (nd.ni_cnd.cn_flags & ISSYMLINK) {
1854 if (vap->va_size != -1) {
1855 error = nfsrv_access(nd.ni_vp, VWRITE,
1856 cred, (nd.ni_cnd.cn_flags & RDONLY), td, 0);
1858 tempsize = vap->va_size;
1860 vap->va_size = tempsize;
1861 error = VOP_SETATTR(nd.ni_vp, vap, cred,
1868 bzero((caddr_t)fhp, sizeof(nfh));
1869 fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsid;
1870 error = VOP_VPTOFH(nd.ni_vp, &fhp->fh_fid);
1872 error = VOP_GETATTR(nd.ni_vp, vap, cred, td);
1875 if (exclusive_flag && !error &&
1876 bcmp(cverf, (caddr_t)&vap->va_atime, NFSX_V3CREATEVERF))
1878 if (dirp == nd.ni_dvp)
1879 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
1881 /* Drop the other locks to avoid deadlock. */
1883 if (nd.ni_dvp == nd.ni_vp)
1893 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY, td);
1894 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
1895 VOP_UNLOCK(dirp, 0, td);
1899 nfsm_reply(NFSX_SRVFH(v3) + NFSX_FATTR(v3) + NFSX_WCCDATA(v3));
1902 nfsm_srvpostop_fh(fhp);
1903 nfsm_srvpostop_attr(0, vap);
1905 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1906 } else if (!error) {
1907 /* v2 non-error case. */
1908 nfsm_srvfhtom(fhp, v3);
1909 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
1910 nfsm_srvfillattr(vap, fp);
1916 if (nd.ni_dvp == nd.ni_vp)
1923 if (nd.ni_startdir) {
1924 vrele(nd.ni_startdir);
1925 nd.ni_startdir = NULL;
1929 NDFREE(&nd, NDF_ONLY_PNBUF);
1930 vn_finished_write(mp);
1931 VFS_UNLOCK_GIANT(vfslocked);
1936 * nfs v3 mknod service
1939 nfsrv_mknod(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
1940 struct thread *td, struct mbuf **mrq)
1942 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1943 struct sockaddr *nam = nfsd->nd_nam;
1944 caddr_t dpos = nfsd->nd_dpos;
1945 struct ucred *cred = nfsd->nd_cr;
1946 struct vattr va, dirfor, diraft;
1947 struct vattr *vap = &va;
1949 struct nameidata nd;
1951 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
1952 u_int32_t major, minor;
1954 struct mbuf *mb, *mreq;
1955 struct vnode *vp, *dirp = NULL;
1958 struct mount *mp = NULL;
1959 int v3 = (nfsd->nd_flag & ND_NFSV3);
1963 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1966 panic("nfsrv_mknod: v3 proc called on a v2 connection");
1969 fhp = &nfh.fh_generic;
1971 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
1975 vfslocked = VFS_LOCK_GIANT(mp);
1976 (void) vn_start_write(NULL, &mp, V_WAIT);
1977 vfs_rel(mp); /* The write holds a ref. */
1978 nfsm_srvnamesiz(len);
1980 nd.ni_cnd.cn_cred = cred;
1981 nd.ni_cnd.cn_nameiop = CREATE;
1982 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | SAVESTART | MPSAFE;
1985 * Handle nfs_namei() call. If an error occurs, the nd structure
1986 * is not valid. However, nfsm_*() routines may still jump to
1990 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
1991 &dirp, v3, &dirfor, &dirfor_ret, td, FALSE);
1992 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
1994 nfsm_reply(NFSX_WCCDATA(1));
1995 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1999 tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED);
2000 vtyp = nfsv3tov_type(*tl);
2001 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
2002 error = NFSERR_BADTYPE;
2007 if (vtyp == VCHR || vtyp == VBLK) {
2008 tl = nfsm_dissect_nonblock(u_int32_t *, 2 * NFSX_UNSIGNED);
2009 major = fxdr_unsigned(u_int32_t, *tl++);
2010 minor = fxdr_unsigned(u_int32_t, *tl);
2011 vap->va_rdev = makedev(major, minor);
2015 * Iff doesn't exist, create it.
2021 vap->va_type = vtyp;
2022 if (vap->va_mode == (mode_t)VNOVAL)
2024 if (vtyp == VSOCK) {
2025 vrele(nd.ni_startdir);
2026 nd.ni_startdir = NULL;
2027 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap);
2029 NDFREE(&nd, NDF_ONLY_PNBUF);
2031 if (vtyp != VFIFO && (error = suser_cred(cred, 0)))
2033 error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap);
2035 NDFREE(&nd, NDF_ONLY_PNBUF);
2042 * Release dvp prior to lookup
2047 nd.ni_cnd.cn_nameiop = LOOKUP;
2048 nd.ni_cnd.cn_flags &= ~(LOCKPARENT);
2049 nd.ni_cnd.cn_thread = td;
2050 nd.ni_cnd.cn_cred = td->td_ucred;
2051 tvfslocked = VFS_LOCK_GIANT(nd.ni_startdir->v_mount);
2053 nd.ni_cnd.cn_flags |= GIANTHELD;
2054 error = lookup(&nd);
2056 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
2057 nd.ni_cnd.cn_flags &= ~GIANTHELD;
2061 if (nd.ni_cnd.cn_flags & ISSYMLINK)
2066 * send response, cleanup, return.
2071 bzero((caddr_t)fhp, sizeof(nfh));
2072 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
2073 error = VOP_VPTOFH(vp, &fhp->fh_fid);
2075 error = VOP_GETATTR(vp, vap, cred, td);
2078 if (nd.ni_dvp == nd.ni_vp)
2089 if (nd.ni_startdir) {
2090 vrele(nd.ni_startdir);
2091 nd.ni_startdir = NULL;
2093 NDFREE(&nd, NDF_ONLY_PNBUF);
2095 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY, td);
2096 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
2097 VOP_UNLOCK(dirp, 0, td);
2100 nfsm_reply(NFSX_SRVFH(1) + NFSX_POSTOPATTR(1) + NFSX_WCCDATA(1));
2103 nfsm_srvpostop_fh(fhp);
2104 nfsm_srvpostop_attr(0, vap);
2106 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2108 vn_finished_write(mp);
2109 VFS_UNLOCK_GIANT(vfslocked);
2113 if (nd.ni_dvp == nd.ni_vp)
2123 vrele(nd.ni_startdir);
2124 NDFREE(&nd, NDF_ONLY_PNBUF);
2125 vn_finished_write(mp);
2126 VFS_UNLOCK_GIANT(vfslocked);
2131 * nfs remove service
2134 nfsrv_remove(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2135 struct thread *td, struct mbuf **mrq)
2137 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2138 struct sockaddr *nam = nfsd->nd_nam;
2139 caddr_t dpos = nfsd->nd_dpos;
2140 struct ucred *cred = nfsd->nd_cr;
2141 struct nameidata nd;
2143 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
2144 int v3 = (nfsd->nd_flag & ND_NFSV3);
2145 struct mbuf *mb, *mreq;
2147 struct vattr dirfor, diraft;
2150 struct mount *mp = NULL;
2153 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2157 fhp = &nfh.fh_generic;
2159 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
2163 vfslocked = VFS_LOCK_GIANT(mp);
2164 (void) vn_start_write(NULL, &mp, V_WAIT);
2165 vfs_rel(mp); /* The write holds a ref. */
2166 nfsm_srvnamesiz(len);
2168 nd.ni_cnd.cn_cred = cred;
2169 nd.ni_cnd.cn_nameiop = DELETE;
2170 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | MPSAFE;
2171 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
2172 &dirp, v3, &dirfor, &dirfor_ret, td, FALSE);
2173 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
2179 if (nd.ni_vp->v_type == VDIR) {
2180 error = EPERM; /* POSIX */
2184 * The root of a mounted filesystem cannot be deleted.
2186 if (nd.ni_vp->v_vflag & VV_ROOT) {
2192 error = VOP_REMOVE(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
2193 NDFREE(&nd, NDF_ONLY_PNBUF);
2197 if (dirp == nd.ni_dvp)
2198 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
2200 /* Drop the other locks to avoid deadlock. */
2202 if (nd.ni_dvp == nd.ni_vp)
2212 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY, td);
2213 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
2214 VOP_UNLOCK(dirp, 0, td);
2220 nfsm_reply(NFSX_WCCDATA(v3));
2222 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2226 NDFREE(&nd, NDF_ONLY_PNBUF);
2228 if (nd.ni_dvp == nd.ni_vp)
2235 vn_finished_write(mp);
2236 VFS_UNLOCK_GIANT(vfslocked);
2241 * nfs rename service
2244 nfsrv_rename(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2245 struct thread *td, struct mbuf **mrq)
2247 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2248 struct sockaddr *nam = nfsd->nd_nam;
2249 caddr_t dpos = nfsd->nd_dpos;
2250 struct ucred *cred = nfsd->nd_cr;
2252 int error = 0, len, len2, fdirfor_ret = 1, fdiraft_ret = 1;
2253 int tdirfor_ret = 1, tdiraft_ret = 1;
2254 int v3 = (nfsd->nd_flag & ND_NFSV3);
2255 struct mbuf *mb, *mreq;
2256 struct nameidata fromnd, tond;
2257 struct vnode *fvp, *tvp, *tdvp, *fdirp = NULL;
2258 struct vnode *tdirp = NULL;
2259 struct vattr fdirfor, fdiraft, tdirfor, tdiraft;
2261 fhandle_t *ffhp, *tfhp;
2263 struct mount *mp = NULL;
2266 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2271 ffhp = &fnfh.fh_generic;
2272 tfhp = &tnfh.fh_generic;
2275 * Clear fields incase goto nfsmout occurs from macro.
2281 nfsm_srvmtofh(ffhp);
2282 if ((mp = vfs_getvfs(&ffhp->fh_fsid)) == NULL) {
2286 vfslocked = VFS_LOCK_GIANT(mp);
2287 (void) vn_start_write(NULL, &mp, V_WAIT);
2288 vfs_rel(mp); /* The write holds a ref. */
2289 nfsm_srvnamesiz(len);
2291 * Remember our original uid so that we can reset cr_uid before
2292 * the second nfs_namei() call, in case it is remapped.
2294 saved_uid = cred->cr_uid;
2295 fromnd.ni_cnd.cn_cred = cred;
2296 fromnd.ni_cnd.cn_nameiop = DELETE;
2297 fromnd.ni_cnd.cn_flags = WANTPARENT | SAVESTART | MPSAFE;
2298 error = nfs_namei(&fromnd, ffhp, len, slp, nam, &md,
2299 &dpos, &fdirp, v3, &fdirfor, &fdirfor_ret, td, FALSE);
2300 vfslocked = nfsrv_lockedpair_nd(vfslocked, &fromnd);
2306 nfsm_reply(2 * NFSX_WCCDATA(v3));
2308 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
2309 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
2315 nfsm_srvmtofh(tfhp);
2316 nfsm_srvnamesiz(len2);
2317 cred->cr_uid = saved_uid;
2318 tond.ni_cnd.cn_cred = cred;
2319 tond.ni_cnd.cn_nameiop = RENAME;
2320 tond.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | NOCACHE | SAVESTART | MPSAFE;
2321 error = nfs_namei(&tond, tfhp, len2, slp, nam, &md,
2322 &dpos, &tdirp, v3, &tdirfor, &tdirfor_ret, td, FALSE);
2323 vfslocked = nfsrv_lockedpair_nd(vfslocked, &tond);
2334 if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
2340 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
2347 if (tvp->v_type == VDIR && tvp->v_mountedhere) {
2355 if (fvp->v_type == VDIR && fvp->v_mountedhere) {
2362 if (fvp->v_mount != tdvp->v_mount) {
2376 * If source is the same as the destination (that is the
2377 * same vnode with the same name in the same directory),
2378 * then there is nothing to do.
2380 if (fvp == tvp && fromnd.ni_dvp == tdvp &&
2381 fromnd.ni_cnd.cn_namelen == tond.ni_cnd.cn_namelen &&
2382 !bcmp(fromnd.ni_cnd.cn_nameptr, tond.ni_cnd.cn_nameptr,
2383 fromnd.ni_cnd.cn_namelen))
2388 * The VOP_RENAME function releases all vnode references &
2389 * locks prior to returning so we need to clear the pointers
2390 * to bypass cleanup code later on.
2392 error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd,
2393 tond.ni_dvp, tond.ni_vp, &tond.ni_cnd);
2394 fromnd.ni_dvp = NULL;
2395 fromnd.ni_vp = NULL;
2399 NDFREE(&fromnd, NDF_ONLY_PNBUF);
2400 NDFREE(&tond, NDF_ONLY_PNBUF);
2408 nfsm_reply(2 * NFSX_WCCDATA(v3));
2410 /* Release existing locks to prevent deadlock. */
2412 if (tond.ni_dvp == tond.ni_vp)
2423 vn_lock(fdirp, LK_EXCLUSIVE | LK_RETRY, td);
2424 fdiraft_ret = VOP_GETATTR(fdirp, &fdiraft, cred, td);
2425 VOP_UNLOCK(fdirp, 0, td);
2428 vn_lock(tdirp, LK_EXCLUSIVE | LK_RETRY, td);
2429 tdiraft_ret = VOP_GETATTR(tdirp, &tdiraft, cred, td);
2430 VOP_UNLOCK(tdirp, 0, td);
2432 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
2433 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
2440 * Clear out tond related fields
2443 if (tond.ni_dvp == tond.ni_vp)
2452 if (tond.ni_startdir)
2453 vrele(tond.ni_startdir);
2454 NDFREE(&tond, NDF_ONLY_PNBUF);
2456 * Clear out fromnd related fields
2460 if (fromnd.ni_startdir)
2461 vrele(fromnd.ni_startdir);
2462 NDFREE(&fromnd, NDF_ONLY_PNBUF);
2464 vrele(fromnd.ni_dvp);
2466 vrele(fromnd.ni_vp);
2468 vn_finished_write(mp);
2469 VFS_UNLOCK_GIANT(vfslocked);
2477 nfsrv_link(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2478 struct thread *td, struct mbuf **mrq)
2480 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2481 struct sockaddr *nam = nfsd->nd_nam;
2482 caddr_t dpos = nfsd->nd_dpos;
2483 struct ucred *cred = nfsd->nd_cr;
2484 struct nameidata nd;
2486 int error = 0, rdonly, len, dirfor_ret = 1, diraft_ret = 1;
2487 int getret = 1, v3 = (nfsd->nd_flag & ND_NFSV3);
2488 struct mbuf *mb, *mreq;
2489 struct vnode *vp = NULL, *xp, *dirp = NULL;
2490 struct vattr dirfor, diraft, at;
2492 fhandle_t *fhp, *dfhp;
2493 struct mount *mp = NULL;
2497 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2501 fhp = &nfh.fh_generic;
2502 dfhp = &dnfh.fh_generic;
2504 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
2508 vfslocked = VFS_LOCK_GIANT(mp);
2509 (void) vn_start_write(NULL, &mp, V_WAIT);
2510 vfs_rel(mp); /* The write holds a ref. */
2511 nfsm_srvmtofh(dfhp);
2512 nfsm_srvnamesiz(len);
2514 error = nfsrv_fhtovp(fhp, TRUE, &vp, &tvfslocked, cred, slp,
2515 nam, &rdonly, TRUE);
2516 vfslocked = nfsrv_lockedpair(vfslocked, tvfslocked);
2518 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2520 nfsm_srvpostop_attr(getret, &at);
2521 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2528 getret = VOP_GETATTR(vp, &at, cred, td);
2529 if (vp->v_type == VDIR) {
2530 error = EPERM; /* POSIX */
2533 VOP_UNLOCK(vp, 0, td);
2534 nd.ni_cnd.cn_cred = cred;
2535 nd.ni_cnd.cn_nameiop = CREATE;
2536 nd.ni_cnd.cn_flags = LOCKPARENT | MPSAFE | MPSAFE;
2537 error = nfs_namei(&nd, dfhp, len, slp, nam, &md, &dpos,
2538 &dirp, v3, &dirfor, &dirfor_ret, td, FALSE);
2539 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
2557 if (vp->v_mount != xp->v_mount) {
2563 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
2564 error = VOP_LINK(nd.ni_dvp, vp, &nd.ni_cnd);
2565 NDFREE(&nd, NDF_ONLY_PNBUF);
2570 getret = VOP_GETATTR(vp, &at, cred, td);
2573 if (dirp == nd.ni_dvp)
2574 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
2576 /* Release existing locks to prevent deadlock. */
2578 if (nd.ni_dvp == nd.ni_vp)
2588 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY, td);
2589 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
2590 VOP_UNLOCK(dirp, 0, td);
2594 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2596 nfsm_srvpostop_attr(getret, &at);
2597 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2603 NDFREE(&nd, NDF_ONLY_PNBUF);
2607 if (nd.ni_dvp == nd.ni_vp)
2616 vn_finished_write(mp);
2617 VFS_UNLOCK_GIANT(vfslocked);
2622 * nfs symbolic link service
2625 nfsrv_symlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2626 struct thread *td, struct mbuf **mrq)
2628 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2629 struct sockaddr *nam = nfsd->nd_nam;
2630 caddr_t dpos = nfsd->nd_dpos;
2631 struct ucred *cred = nfsd->nd_cr;
2632 struct vattr va, dirfor, diraft;
2633 struct nameidata nd;
2634 struct vattr *vap = &va;
2635 struct nfsv2_sattr *sp;
2636 char *bpos, *pathcp = NULL;
2639 int error = 0, len, len2, dirfor_ret = 1, diraft_ret = 1;
2640 int v3 = (nfsd->nd_flag & ND_NFSV3);
2641 struct mbuf *mb, *mreq;
2642 struct vnode *dirp = NULL;
2645 struct mount *mp = NULL;
2649 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2653 fhp = &nfh.fh_generic;
2655 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
2659 vfslocked = VFS_LOCK_GIANT(mp);
2660 (void) vn_start_write(NULL, &mp, V_WAIT);
2661 vfs_rel(mp); /* The write holds a ref. */
2662 nfsm_srvnamesiz(len);
2663 nd.ni_cnd.cn_cred = cred;
2664 nd.ni_cnd.cn_nameiop = CREATE;
2665 nd.ni_cnd.cn_flags = LOCKPARENT | SAVESTART | MPSAFE;
2666 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
2667 &dirp, v3, &dirfor, &dirfor_ret, td, FALSE);
2668 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
2673 nfsm_srvpathsiz(len2);
2681 MALLOC(pathcp, caddr_t, len2 + 1, M_TEMP, M_WAITOK);
2682 iv.iov_base = pathcp;
2684 io.uio_resid = len2;
2688 io.uio_segflg = UIO_SYSSPACE;
2689 io.uio_rw = UIO_READ;
2691 nfsm_mtouio(&io, len2);
2693 sp = nfsm_dissect_nonblock(struct nfsv2_sattr *, NFSX_V2SATTR);
2694 vap->va_mode = nfstov_mode(sp->sa_mode);
2696 *(pathcp + len2) = '\0';
2703 * issue symlink op. SAVESTART is set so the underlying path component
2704 * is only freed by the VOP if an error occurs.
2706 if (vap->va_mode == (mode_t)VNOVAL)
2708 error = VOP_SYMLINK(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap, pathcp);
2710 NDFREE(&nd, NDF_ONLY_PNBUF);
2715 * releases directory prior to potential lookup op.
2723 * Issue lookup. Leave SAVESTART set so we can easily free
2724 * the name buffer later on.
2726 * since LOCKPARENT is not set, ni_dvp will be garbage on
2727 * return whether an error occurs or not.
2729 nd.ni_cnd.cn_nameiop = LOOKUP;
2730 nd.ni_cnd.cn_flags &= ~(LOCKPARENT | FOLLOW);
2731 nd.ni_cnd.cn_flags |= (NOFOLLOW | LOCKLEAF);
2732 nd.ni_cnd.cn_thread = td;
2733 nd.ni_cnd.cn_cred = cred;
2734 tvfslocked = VFS_LOCK_GIANT(nd.ni_startdir->v_mount);
2736 nd.ni_cnd.cn_flags |= GIANTHELD;
2737 error = lookup(&nd);
2739 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
2740 nd.ni_cnd.cn_flags &= ~GIANTHELD;
2743 bzero((caddr_t)fhp, sizeof(nfh));
2744 fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsid;
2745 error = VOP_VPTOFH(nd.ni_vp, &fhp->fh_fid);
2747 error = VOP_GETATTR(nd.ni_vp, vap, cred,
2756 * These releases aren't strictly required, does even doing them
2757 * make any sense? XXX can nfsm_reply() block?
2760 FREE(pathcp, M_TEMP);
2764 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY, td);
2765 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
2766 VOP_UNLOCK(dirp, 0, td);
2768 if (nd.ni_startdir) {
2769 vrele(nd.ni_startdir);
2770 nd.ni_startdir = NULL;
2772 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2775 nfsm_srvpostop_fh(fhp);
2776 nfsm_srvpostop_attr(0, vap);
2778 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2784 NDFREE(&nd, NDF_ONLY_PNBUF);
2786 if (nd.ni_dvp == nd.ni_vp)
2794 vrele(nd.ni_startdir);
2798 FREE(pathcp, M_TEMP);
2800 vn_finished_write(mp);
2801 VFS_UNLOCK_GIANT(vfslocked);
2809 nfsrv_mkdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2810 struct thread *td, struct mbuf **mrq)
2812 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2813 struct sockaddr *nam = nfsd->nd_nam;
2814 caddr_t dpos = nfsd->nd_dpos;
2815 struct ucred *cred = nfsd->nd_cr;
2816 struct vattr va, dirfor, diraft;
2817 struct vattr *vap = &va;
2818 struct nfs_fattr *fp;
2819 struct nameidata nd;
2822 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
2823 int v3 = (nfsd->nd_flag & ND_NFSV3);
2824 struct mbuf *mb, *mreq;
2825 struct vnode *dirp = NULL;
2829 struct mount *mp = NULL;
2832 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2836 fhp = &nfh.fh_generic;
2838 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
2842 vfslocked = VFS_LOCK_GIANT(mp);
2843 (void) vn_start_write(NULL, &mp, V_WAIT);
2844 vfs_rel(mp); /* The write holds a ref. */
2845 nfsm_srvnamesiz(len);
2846 nd.ni_cnd.cn_cred = cred;
2847 nd.ni_cnd.cn_nameiop = CREATE;
2848 nd.ni_cnd.cn_flags = LOCKPARENT | MPSAFE;
2850 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
2851 &dirp, v3, &dirfor, &dirfor_ret, td, FALSE);
2852 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
2858 nfsm_reply(NFSX_WCCDATA(v3));
2860 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2868 tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED);
2869 vap->va_mode = nfstov_mode(*tl++);
2873 * At this point nd.ni_dvp is referenced and exclusively locked and
2874 * nd.ni_vp, if it exists, is referenced but not locked.
2877 vap->va_type = VDIR;
2878 if (nd.ni_vp != NULL) {
2879 NDFREE(&nd, NDF_ONLY_PNBUF);
2885 * Issue mkdir op. Since SAVESTART is not set, the pathname
2886 * component is freed by the VOP call. This will fill-in
2887 * nd.ni_vp, reference, and exclusively lock it.
2889 if (vap->va_mode == (mode_t)VNOVAL)
2891 error = VOP_MKDIR(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap);
2892 NDFREE(&nd, NDF_ONLY_PNBUF);
2899 bzero((caddr_t)fhp, sizeof(nfh));
2900 fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsid;
2901 error = VOP_VPTOFH(nd.ni_vp, &fhp->fh_fid);
2903 error = VOP_GETATTR(nd.ni_vp, vap, cred, td);
2907 if (dirp == nd.ni_dvp) {
2908 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
2910 /* Release existing locks to prevent deadlock. */
2912 NDFREE(&nd, NDF_ONLY_PNBUF);
2913 if (nd.ni_dvp == nd.ni_vp && vpexcl)
2926 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY, td);
2927 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
2928 VOP_UNLOCK(dirp, 0, td);
2931 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2934 nfsm_srvpostop_fh(fhp);
2935 nfsm_srvpostop_attr(0, vap);
2937 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2938 } else if (!error) {
2939 /* v2 non-error case. */
2940 nfsm_srvfhtom(fhp, v3);
2941 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
2942 nfsm_srvfillattr(vap, fp);
2949 NDFREE(&nd, NDF_ONLY_PNBUF);
2950 if (nd.ni_dvp == nd.ni_vp && vpexcl)
2963 vn_finished_write(mp);
2964 VFS_UNLOCK_GIANT(vfslocked);
2972 nfsrv_rmdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2973 struct thread *td, struct mbuf **mrq)
2975 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2976 struct sockaddr *nam = nfsd->nd_nam;
2977 caddr_t dpos = nfsd->nd_dpos;
2978 struct ucred *cred = nfsd->nd_cr;
2980 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
2981 int v3 = (nfsd->nd_flag & ND_NFSV3);
2982 struct mbuf *mb, *mreq;
2983 struct vnode *vp, *dirp = NULL;
2984 struct vattr dirfor, diraft;
2987 struct nameidata nd;
2988 struct mount *mp = NULL;
2991 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2995 fhp = &nfh.fh_generic;
2997 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
3001 vfslocked = VFS_LOCK_GIANT(mp);
3002 (void) vn_start_write(NULL, &mp, V_WAIT);
3003 vfs_rel(mp); /* The write holds a ref. */
3004 nfsm_srvnamesiz(len);
3005 nd.ni_cnd.cn_cred = cred;
3006 nd.ni_cnd.cn_nameiop = DELETE;
3007 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | MPSAFE;
3008 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
3009 &dirp, v3, &dirfor, &dirfor_ret, td, FALSE);
3010 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
3016 nfsm_reply(NFSX_WCCDATA(v3));
3018 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
3023 if (vp->v_type != VDIR) {
3028 * No rmdir "." please.
3030 if (nd.ni_dvp == vp) {
3035 * The root of a mounted filesystem cannot be deleted.
3037 if (vp->v_vflag & VV_ROOT)
3041 * Issue or abort op. Since SAVESTART is not set, path name
3042 * component is freed by the VOP after either.
3045 error = VOP_RMDIR(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
3046 NDFREE(&nd, NDF_ONLY_PNBUF);
3049 if (dirp == nd.ni_dvp)
3050 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
3052 /* Release existing locks to prevent deadlock. */
3054 if (nd.ni_dvp == nd.ni_vp)
3063 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY, td);
3064 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, td);
3065 VOP_UNLOCK(dirp, 0, td);
3068 nfsm_reply(NFSX_WCCDATA(v3));
3071 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
3075 NDFREE(&nd, NDF_ONLY_PNBUF);
3077 if (nd.ni_dvp == nd.ni_vp)
3087 vn_finished_write(mp);
3088 VFS_UNLOCK_GIANT(vfslocked);
3093 * nfs readdir service
3094 * - mallocs what it thinks is enough to read
3095 * count rounded up to a multiple of NFS_DIRBLKSIZ <= NFS_MAXREADDIR
3096 * - calls VOP_READDIR()
3097 * - loops around building the reply
3098 * if the output generated exceeds count break out of loop
3099 * The nfsm_clget macro is used here so that the reply will be packed
3100 * tightly in mbuf clusters.
3101 * - it only knows that it has encountered eof when the VOP_READDIR()
3103 * - as such one readdir rpc will return eof false although you are there
3104 * and then the next will return eof
3105 * - it trims out records with d_fileno == 0
3106 * this doesn't matter for Unix clients, but they might confuse clients
3108 * NB: It is tempting to set eof to true if the VOP_READDIR() reads less
3109 * than requested, but this may not apply to all filesystems. For
3110 * example, client NFS does not { although it is never remote mounted
3112 * The alternate call nfsrv_readdirplus() does lookups as well.
3113 * PS: The NFS protocol spec. does not clarify what the "count" byte
3114 * argument is a count of.. just name strings and file id's or the
3115 * entire reply rpc or ...
3116 * I tried just file name and id sizes and it confused the Sun client,
3117 * so I am using the full rpc size now. The "paranoia.." comment refers
3118 * to including the status longwords that are not a part of the dir.
3119 * "entry" structures, but are in the rpc.
3123 u_int32_t fl_postopok;
3124 u_int32_t fl_fattr[NFSX_V3FATTR / sizeof (u_int32_t)];
3126 u_int32_t fl_fhsize;
3127 u_int32_t fl_nfh[NFSX_V3FH / sizeof (u_int32_t)];
3131 nfsrv_readdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3132 struct thread *td, struct mbuf **mrq)
3134 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3135 struct sockaddr *nam = nfsd->nd_nam;
3136 caddr_t dpos = nfsd->nd_dpos;
3137 struct ucred *cred = nfsd->nd_cr;
3144 struct mbuf *mb, *mreq;
3145 char *cpos, *cend, *rbuf;
3146 struct vnode *vp = NULL;
3152 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
3153 int siz, cnt, fullsiz, eofflag, rdonly, ncookies;
3154 int v3 = (nfsd->nd_flag & ND_NFSV3);
3155 u_quad_t off, toff, verf;
3156 u_long *cookies = NULL, *cookiep; /* needs to be int64_t or off_t */
3159 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3161 fhp = &nfh.fh_generic;
3164 tl = nfsm_dissect_nonblock(u_int32_t *, 5 * NFSX_UNSIGNED);
3165 toff = fxdr_hyper(tl);
3167 verf = fxdr_hyper(tl);
3170 tl = nfsm_dissect_nonblock(u_int32_t *, 2 * NFSX_UNSIGNED);
3171 toff = fxdr_unsigned(u_quad_t, *tl++);
3172 verf = 0; /* shut up gcc */
3175 cnt = fxdr_unsigned(int, *tl);
3176 siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
3177 xfer = NFS_SRVMAXDATA(nfsd);
3183 error = nfsrv_fhtovp(fhp, 1, &vp, &vfslocked, cred, slp,
3184 nam, &rdonly, TRUE);
3185 if (!error && vp->v_type != VDIR) {
3191 nfsm_reply(NFSX_UNSIGNED);
3193 nfsm_srvpostop_attr(getret, &at);
3199 * Obtain lock on vnode for this section of the code
3202 error = getret = VOP_GETATTR(vp, &at, cred, td);
3205 * XXX This check may be too strict for Solaris 2.5 clients.
3207 if (!error && toff && verf && verf != at.va_filerev)
3208 error = NFSERR_BAD_COOKIE;
3212 error = nfsrv_access(vp, VEXEC, cred, rdonly, td, 0);
3216 nfsm_reply(NFSX_POSTOPATTR(v3));
3218 nfsm_srvpostop_attr(getret, &at);
3222 VOP_UNLOCK(vp, 0, td);
3225 * end section. Allocate rbuf and continue
3227 MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
3230 iv.iov_len = fullsiz;
3233 io.uio_offset = (off_t)off;
3234 io.uio_resid = fullsiz;
3235 io.uio_segflg = UIO_SYSSPACE;
3236 io.uio_rw = UIO_READ;
3239 vn_lock(vp, LK_SHARED | LK_RETRY, td);
3241 free((caddr_t)cookies, M_TEMP);
3244 error = VOP_READDIR(vp, &io, cred, &eofflag, &ncookies, &cookies);
3245 off = (off_t)io.uio_offset;
3246 if (!cookies && !error)
3247 error = NFSERR_PERM;
3249 getret = VOP_GETATTR(vp, &at, cred, td);
3253 VOP_UNLOCK(vp, 0, td);
3257 free((caddr_t)rbuf, M_TEMP);
3259 free((caddr_t)cookies, M_TEMP);
3260 nfsm_reply(NFSX_POSTOPATTR(v3));
3262 nfsm_srvpostop_attr(getret, &at);
3267 siz -= io.uio_resid;
3270 * If nothing read, return eof
3276 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) +
3279 nfsm_srvpostop_attr(getret, &at);
3280 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
3281 txdr_hyper(at.va_filerev, tl);
3284 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
3285 *tl++ = nfsrv_nfs_false;
3286 *tl = nfsrv_nfs_true;
3287 FREE((caddr_t)rbuf, M_TEMP);
3288 FREE((caddr_t)cookies, M_TEMP);
3295 * Check for degenerate cases of nothing useful read.
3296 * If so go try again
3300 dp = (struct dirent *)cpos;
3303 * For some reason FreeBSD's ufs_readdir() chooses to back the
3304 * directory offset up to a block boundary, so it is necessary to
3305 * skip over the records that precede the requested offset. This
3306 * requires the assumption that file offset cookies monotonically
3309 while (cpos < cend && ncookies > 0 &&
3310 (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
3311 ((u_quad_t)(*cookiep)) <= toff)) {
3312 cpos += dp->d_reclen;
3313 dp = (struct dirent *)cpos;
3317 if (cpos >= cend || ncookies == 0) {
3323 len = 3 * NFSX_UNSIGNED; /* paranoia, probably can be 0 */
3324 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + siz);
3326 nfsm_srvpostop_attr(getret, &at);
3327 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
3328 txdr_hyper(at.va_filerev, tl);
3332 be = bp + M_TRAILINGSPACE(mp);
3334 /* Loop through the records and build reply */
3335 while (cpos < cend && ncookies > 0) {
3336 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
3337 nlen = dp->d_namlen;
3338 rem = nfsm_rndup(nlen) - nlen;
3339 len += (4 * NFSX_UNSIGNED + nlen + rem);
3341 len += 2 * NFSX_UNSIGNED;
3347 * Build the directory record xdr from
3351 *tl = nfsrv_nfs_true;
3352 bp += NFSX_UNSIGNED;
3356 bp += NFSX_UNSIGNED;
3359 *tl = txdr_unsigned(dp->d_fileno);
3360 bp += NFSX_UNSIGNED;
3362 *tl = txdr_unsigned(nlen);
3363 bp += NFSX_UNSIGNED;
3365 /* And loop around copying the name */
3374 bcopy(cp, bp, tsiz);
3380 /* And null pad to an int32_t boundary. */
3381 for (i = 0; i < rem; i++)
3385 /* Finish off the record */
3388 bp += NFSX_UNSIGNED;
3391 *tl = txdr_unsigned(*cookiep);
3392 bp += NFSX_UNSIGNED;
3394 cpos += dp->d_reclen;
3395 dp = (struct dirent *)cpos;
3402 *tl = nfsrv_nfs_false;
3403 bp += NFSX_UNSIGNED;
3406 *tl = nfsrv_nfs_true;
3408 *tl = nfsrv_nfs_false;
3409 bp += NFSX_UNSIGNED;
3412 mp->m_len = bp - mtod(mp, caddr_t);
3414 mp->m_len += bp - bpos;
3415 FREE((caddr_t)rbuf, M_TEMP);
3416 FREE((caddr_t)cookies, M_TEMP);
3421 VFS_UNLOCK_GIANT(vfslocked);
3426 nfsrv_readdirplus(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3427 struct thread *td, struct mbuf **mrq)
3429 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3430 struct sockaddr *nam = nfsd->nd_nam;
3431 caddr_t dpos = nfsd->nd_dpos;
3432 struct ucred *cred = nfsd->nd_cr;
3439 struct mbuf *mb, *mreq;
3440 char *cpos, *cend, *rbuf;
3441 struct vnode *vp = NULL, *nvp;
3444 fhandle_t *fhp, *nfhp = (fhandle_t *)fl.fl_nfh;
3447 struct vattr va, at, *vap = &va;
3448 struct nfs_fattr *fp;
3449 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
3450 int siz, cnt, fullsiz, eofflag, rdonly, dirlen, ncookies;
3451 u_quad_t off, toff, verf;
3452 u_long *cookies = NULL, *cookiep; /* needs to be int64_t or off_t */
3453 int v3 = (nfsd->nd_flag & ND_NFSV3);
3456 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3459 panic("nfsrv_readdirplus: v3 proc called on a v2 connection");
3460 fhp = &nfh.fh_generic;
3462 tl = nfsm_dissect_nonblock(u_int32_t *, 6 * NFSX_UNSIGNED);
3463 toff = fxdr_hyper(tl);
3465 verf = fxdr_hyper(tl);
3467 siz = fxdr_unsigned(int, *tl++);
3468 cnt = fxdr_unsigned(int, *tl);
3470 siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
3471 xfer = NFS_SRVMAXDATA(nfsd);
3477 error = nfsrv_fhtovp(fhp, 1, &vp, &vfslocked, cred, slp,
3478 nam, &rdonly, TRUE);
3479 if (!error && vp->v_type != VDIR) {
3485 nfsm_reply(NFSX_UNSIGNED);
3486 nfsm_srvpostop_attr(getret, &at);
3490 error = getret = VOP_GETATTR(vp, &at, cred, td);
3493 * XXX This check may be too strict for Solaris 2.5 clients.
3495 if (!error && toff && verf && verf != at.va_filerev)
3496 error = NFSERR_BAD_COOKIE;
3499 error = nfsrv_access(vp, VEXEC, cred, rdonly, td, 0);
3503 nfsm_reply(NFSX_V3POSTOPATTR);
3504 nfsm_srvpostop_attr(getret, &at);
3508 VOP_UNLOCK(vp, 0, td);
3509 MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
3512 iv.iov_len = fullsiz;
3515 io.uio_offset = (off_t)off;
3516 io.uio_resid = fullsiz;
3517 io.uio_segflg = UIO_SYSSPACE;
3518 io.uio_rw = UIO_READ;
3521 vn_lock(vp, LK_SHARED | LK_RETRY, td);
3523 free((caddr_t)cookies, M_TEMP);
3526 error = VOP_READDIR(vp, &io, cred, &eofflag, &ncookies, &cookies);
3527 off = (u_quad_t)io.uio_offset;
3528 getret = VOP_GETATTR(vp, &at, cred, td);
3529 VOP_UNLOCK(vp, 0, td);
3530 if (!cookies && !error)
3531 error = NFSERR_PERM;
3538 free((caddr_t)cookies, M_TEMP);
3539 free((caddr_t)rbuf, M_TEMP);
3540 nfsm_reply(NFSX_V3POSTOPATTR);
3541 nfsm_srvpostop_attr(getret, &at);
3546 siz -= io.uio_resid;
3549 * If nothing read, return eof
3555 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF +
3557 nfsm_srvpostop_attr(getret, &at);
3558 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
3559 txdr_hyper(at.va_filerev, tl);
3561 *tl++ = nfsrv_nfs_false;
3562 *tl = nfsrv_nfs_true;
3563 FREE((caddr_t)cookies, M_TEMP);
3564 FREE((caddr_t)rbuf, M_TEMP);
3571 * Check for degenerate cases of nothing useful read.
3572 * If so go try again
3576 dp = (struct dirent *)cpos;
3579 * For some reason FreeBSD's ufs_readdir() chooses to back the
3580 * directory offset up to a block boundary, so it is necessary to
3581 * skip over the records that precede the requested offset. This
3582 * requires the assumption that file offset cookies monotonically
3585 while (cpos < cend && ncookies > 0 &&
3586 (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
3587 ((u_quad_t)(*cookiep)) <= toff)) {
3588 cpos += dp->d_reclen;
3589 dp = (struct dirent *)cpos;
3593 if (cpos >= cend || ncookies == 0) {
3600 * Probe one of the directory entries to see if the filesystem
3603 error = VFS_VGET(vp->v_mount, dp->d_fileno, LK_EXCLUSIVE, &nvp);
3605 if (error == EOPNOTSUPP)
3606 error = NFSERR_NOTSUPP;
3608 error = NFSERR_SERVERFAULT;
3611 free((caddr_t)cookies, M_TEMP);
3612 free((caddr_t)rbuf, M_TEMP);
3613 nfsm_reply(NFSX_V3POSTOPATTR);
3614 nfsm_srvpostop_attr(getret, &at);
3621 dirlen = len = NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF +
3624 nfsm_srvpostop_attr(getret, &at);
3625 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
3626 txdr_hyper(at.va_filerev, tl);
3629 be = bp + M_TRAILINGSPACE(mp);
3631 /* Loop through the records and build reply */
3632 while (cpos < cend && ncookies > 0) {
3633 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
3634 nlen = dp->d_namlen;
3635 rem = nfsm_rndup(nlen)-nlen;
3638 * For readdir_and_lookup get the vnode using
3641 if (VFS_VGET(vp->v_mount, dp->d_fileno, LK_EXCLUSIVE,
3644 bzero((caddr_t)nfhp, NFSX_V3FH);
3646 nvp->v_mount->mnt_stat.f_fsid;
3648 * XXXRW: Assert the mountpoints are the same so that
3649 * we know that acquiring Giant based on the
3650 * directory is the right thing for the child.
3652 KASSERT(nvp->v_mount == vp->v_mount,
3653 ("nfsrv_readdirplus: nvp mount != vp mount"));
3654 if (VOP_VPTOFH(nvp, &nfhp->fh_fid)) {
3659 if (VOP_GETATTR(nvp, vap, cred, td)) {
3668 * If either the dircount or maxcount will be
3669 * exceeded, get out now. Both of these lengths
3670 * are calculated conservatively, including all
3673 len += (8 * NFSX_UNSIGNED + nlen + rem + NFSX_V3FH +
3675 dirlen += (6 * NFSX_UNSIGNED + nlen + rem);
3676 if (len > cnt || dirlen > fullsiz) {
3682 * Build the directory record xdr from
3685 fp = (struct nfs_fattr *)&fl.fl_fattr;
3686 nfsm_srvfillattr(vap, fp);
3687 fl.fl_fhsize = txdr_unsigned(NFSX_V3FH);
3688 fl.fl_fhok = nfsrv_nfs_true;
3689 fl.fl_postopok = nfsrv_nfs_true;
3690 fl.fl_off.nfsuquad[0] = 0;
3691 fl.fl_off.nfsuquad[1] = txdr_unsigned(*cookiep);
3694 *tl = nfsrv_nfs_true;
3695 bp += NFSX_UNSIGNED;
3698 bp += NFSX_UNSIGNED;
3700 *tl = txdr_unsigned(dp->d_fileno);
3701 bp += NFSX_UNSIGNED;
3703 *tl = txdr_unsigned(nlen);
3704 bp += NFSX_UNSIGNED;
3706 /* And loop around copying the name */
3711 if ((bp + xfer) > be)
3715 bcopy(cp, bp, tsiz);
3721 /* And null pad to an int32_t boundary. */
3722 for (i = 0; i < rem; i++)
3726 * Now copy the flrep structure out.
3728 xfer = sizeof (struct flrep);
3732 if ((bp + xfer) > be)
3736 bcopy(cp, bp, tsiz);
3744 cpos += dp->d_reclen;
3745 dp = (struct dirent *)cpos;
3752 *tl = nfsrv_nfs_false;
3753 bp += NFSX_UNSIGNED;
3756 *tl = nfsrv_nfs_true;
3758 *tl = nfsrv_nfs_false;
3759 bp += NFSX_UNSIGNED;
3762 mp->m_len = bp - mtod(mp, caddr_t);
3764 mp->m_len += bp - bpos;
3765 FREE((caddr_t)cookies, M_TEMP);
3766 FREE((caddr_t)rbuf, M_TEMP);
3770 VFS_UNLOCK_GIANT(vfslocked);
3775 * nfs commit service
3778 nfsrv_commit(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3779 struct thread *td, struct mbuf **mrq)
3781 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3782 struct sockaddr *nam = nfsd->nd_nam;
3783 caddr_t dpos = nfsd->nd_dpos;
3784 struct ucred *cred = nfsd->nd_cr;
3785 struct vattr bfor, aft;
3786 struct vnode *vp = NULL;
3791 int error = 0, rdonly, for_ret = 1, aft_ret = 1, cnt;
3792 struct mbuf *mb, *mreq;
3794 struct mount *mp = NULL;
3795 int v3 = (nfsd->nd_flag & ND_NFSV3);
3799 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3802 panic("nfsrv_commit: v3 proc called on a v2 connection");
3803 fhp = &nfh.fh_generic;
3805 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
3809 vfslocked = VFS_LOCK_GIANT(mp);
3810 (void) vn_start_write(NULL, &mp, V_WAIT);
3811 vfs_rel(mp); /* The write holds a ref. */
3812 tl = nfsm_dissect_nonblock(u_int32_t *, 3 * NFSX_UNSIGNED);
3815 * XXX At this time VOP_FSYNC() does not accept offset and byte
3816 * count parameters, so these arguments are useless (someday maybe).
3818 off = fxdr_hyper(tl);
3820 cnt = fxdr_unsigned(int, *tl);
3821 error = nfsrv_fhtovp(fhp, 1, &vp, &tvfslocked, cred, slp,
3822 nam, &rdonly, TRUE);
3823 vfslocked = nfsrv_lockedpair(vfslocked, tvfslocked);
3825 nfsm_reply(2 * NFSX_UNSIGNED);
3826 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3830 for_ret = VOP_GETATTR(vp, &bfor, cred, td);
3832 if (cnt > MAX_COMMIT_COUNT) {
3834 * Give up and do the whole thing
3837 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
3838 VM_OBJECT_LOCK(vp->v_object);
3839 vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
3840 VM_OBJECT_UNLOCK(vp->v_object);
3842 error = VOP_FSYNC(vp, MNT_WAIT, td);
3845 * Locate and synchronously write any buffers that fall
3846 * into the requested range. Note: we are assuming that
3847 * f_iosize is a power of 2.
3849 int iosize = vp->v_mount->mnt_stat.f_iosize;
3850 int iomask = iosize - 1;
3855 * Align to iosize boundry, super-align to page boundry.
3858 cnt += off & iomask;
3859 off &= ~(u_quad_t)iomask;
3861 if (off & PAGE_MASK) {
3862 cnt += off & PAGE_MASK;
3863 off &= ~(u_quad_t)PAGE_MASK;
3865 lblkno = off / iosize;
3868 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
3869 VM_OBJECT_LOCK(vp->v_object);
3870 vm_object_page_clean(vp->v_object, off / PAGE_SIZE, (cnt + PAGE_MASK) / PAGE_SIZE, OBJPC_SYNC);
3871 VM_OBJECT_UNLOCK(vp->v_object);
3880 * If we have a buffer and it is marked B_DELWRI we
3881 * have to lock and write it. Otherwise the prior
3882 * write is assumed to have already been committed.
3884 * gbincore() can return invalid buffers now so we
3885 * have to check that bit as well (though B_DELWRI
3886 * should not be set if B_INVAL is set there could be
3887 * a race here since we haven't locked the buffer).
3889 if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
3890 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
3891 LK_INTERLOCK, VI_MTX(vp)) == ENOLCK) {
3893 continue; /* retry */
3895 if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
3898 bp->b_flags &= ~B_ASYNC;
3915 aft_ret = VOP_GETATTR(vp, &aft, cred, td);
3919 nfsm_reply(NFSX_V3WCCDATA + NFSX_V3WRITEVERF);
3920 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3922 tl = nfsm_build(u_int32_t *, NFSX_V3WRITEVERF);
3923 if (nfsver.tv_sec == 0)
3925 *tl++ = txdr_unsigned(nfsver.tv_sec);
3926 *tl = txdr_unsigned(nfsver.tv_usec);
3933 vn_finished_write(mp);
3934 VFS_UNLOCK_GIANT(vfslocked);
3939 * nfs statfs service
3942 nfsrv_statfs(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3943 struct thread *td, struct mbuf **mrq)
3945 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3946 struct sockaddr *nam = nfsd->nd_nam;
3947 caddr_t dpos = nfsd->nd_dpos;
3948 struct ucred *cred = nfsd->nd_cr;
3950 struct nfs_statfs *sfp;
3952 int error = 0, rdonly, getret = 1;
3953 int v3 = (nfsd->nd_flag & ND_NFSV3);
3954 struct mbuf *mb, *mreq;
3955 struct vnode *vp = NULL;
3959 struct statfs statfs;
3963 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3965 fhp = &nfh.fh_generic;
3967 error = nfsrv_fhtovp(fhp, 1, &vp, &vfslocked, cred, slp,
3968 nam, &rdonly, TRUE);
3970 nfsm_reply(NFSX_UNSIGNED);
3972 nfsm_srvpostop_attr(getret, &at);
3977 error = VFS_STATFS(vp->v_mount, sf, td);
3978 getret = VOP_GETATTR(vp, &at, cred, td);
3981 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_STATFS(v3));
3983 nfsm_srvpostop_attr(getret, &at);
3988 sfp = nfsm_build(struct nfs_statfs *, NFSX_STATFS(v3));
3990 tval = (u_quad_t)sf->f_blocks;
3991 tval *= (u_quad_t)sf->f_bsize;
3992 txdr_hyper(tval, &sfp->sf_tbytes);
3993 tval = (u_quad_t)sf->f_bfree;
3994 tval *= (u_quad_t)sf->f_bsize;
3995 txdr_hyper(tval, &sfp->sf_fbytes);
3997 * Don't send negative values for available space,
3998 * since this field is unsigned in the NFS protocol.
3999 * Otherwise, the client would see absurdly high
4000 * numbers for free space.
4002 if (sf->f_bavail < 0)
4005 tval = (u_quad_t)sf->f_bavail;
4006 tval *= (u_quad_t)sf->f_bsize;
4007 txdr_hyper(tval, &sfp->sf_abytes);
4008 sfp->sf_tfiles.nfsuquad[0] = 0;
4009 sfp->sf_tfiles.nfsuquad[1] = txdr_unsigned(sf->f_files);
4010 sfp->sf_ffiles.nfsuquad[0] = 0;
4011 sfp->sf_ffiles.nfsuquad[1] = txdr_unsigned(sf->f_ffree);
4012 sfp->sf_afiles.nfsuquad[0] = 0;
4013 sfp->sf_afiles.nfsuquad[1] = txdr_unsigned(sf->f_ffree);
4014 sfp->sf_invarsec = 0;
4016 sfp->sf_tsize = txdr_unsigned(NFS_MAXDGRAMDATA);
4017 sfp->sf_bsize = txdr_unsigned(sf->f_bsize);
4018 sfp->sf_blocks = txdr_unsigned(sf->f_blocks);
4019 sfp->sf_bfree = txdr_unsigned(sf->f_bfree);
4020 if (sf->f_bavail < 0)
4023 sfp->sf_bavail = txdr_unsigned(sf->f_bavail);
4028 VFS_UNLOCK_GIANT(vfslocked);
4033 * nfs fsinfo service
4036 nfsrv_fsinfo(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
4037 struct thread *td, struct mbuf **mrq)
4039 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
4040 struct sockaddr *nam = nfsd->nd_nam;
4041 caddr_t dpos = nfsd->nd_dpos;
4042 struct ucred *cred = nfsd->nd_cr;
4043 struct nfsv3_fsinfo *sip;
4045 int error = 0, rdonly, getret = 1, pref;
4046 struct mbuf *mb, *mreq;
4047 struct vnode *vp = NULL;
4053 int v3 = (nfsd->nd_flag & ND_NFSV3);
4056 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
4058 panic("nfsrv_fsinfo: v3 proc called on a v2 connection");
4059 fhp = &nfh.fh_generic;
4062 error = nfsrv_fhtovp(fhp, 1, &vp, &vfslocked, cred, slp,
4063 nam, &rdonly, TRUE);
4065 nfsm_reply(NFSX_UNSIGNED);
4066 nfsm_srvpostop_attr(getret, &at);
4071 /* XXX Try to make a guess on the max file size. */
4072 VFS_STATFS(vp->v_mount, &sb, td);
4073 maxfsize = (u_quad_t)0x80000000 * sb.f_bsize - 1;
4075 getret = VOP_GETATTR(vp, &at, cred, td);
4078 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3FSINFO);
4079 nfsm_srvpostop_attr(getret, &at);
4080 sip = nfsm_build(struct nfsv3_fsinfo *, NFSX_V3FSINFO);
4084 * There should be filesystem VFS OP(s) to get this information.
4085 * For now, assume ufs.
4087 if (slp->ns_so->so_type == SOCK_DGRAM)
4088 pref = NFS_MAXDGRAMDATA;
4091 sip->fs_rtmax = txdr_unsigned(pref);
4092 sip->fs_rtpref = txdr_unsigned(pref);
4093 sip->fs_rtmult = txdr_unsigned(NFS_FABLKSIZE);
4094 sip->fs_wtmax = txdr_unsigned(pref);
4095 sip->fs_wtpref = txdr_unsigned(pref);
4096 sip->fs_wtmult = txdr_unsigned(NFS_FABLKSIZE);
4097 sip->fs_dtpref = txdr_unsigned(pref);
4098 txdr_hyper(maxfsize, &sip->fs_maxfilesize);
4099 sip->fs_timedelta.nfsv3_sec = 0;
4100 sip->fs_timedelta.nfsv3_nsec = txdr_unsigned(1);
4101 sip->fs_properties = txdr_unsigned(NFSV3FSINFO_LINK |
4102 NFSV3FSINFO_SYMLINK | NFSV3FSINFO_HOMOGENEOUS |
4103 NFSV3FSINFO_CANSETTIME);
4107 VFS_UNLOCK_GIANT(vfslocked);
4112 * nfs pathconf service
4115 nfsrv_pathconf(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
4116 struct thread *td, struct mbuf **mrq)
4118 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
4119 struct sockaddr *nam = nfsd->nd_nam;
4120 caddr_t dpos = nfsd->nd_dpos;
4121 struct ucred *cred = nfsd->nd_cr;
4122 struct nfsv3_pathconf *pc;
4124 int error = 0, rdonly, getret = 1;
4125 register_t linkmax, namemax, chownres, notrunc;
4126 struct mbuf *mb, *mreq;
4127 struct vnode *vp = NULL;
4131 int v3 = (nfsd->nd_flag & ND_NFSV3);
4134 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
4136 panic("nfsrv_pathconf: v3 proc called on a v2 connection");
4138 fhp = &nfh.fh_generic;
4140 error = nfsrv_fhtovp(fhp, 1, &vp, &vfslocked, cred, slp,
4141 nam, &rdonly, TRUE);
4143 nfsm_reply(NFSX_UNSIGNED);
4144 nfsm_srvpostop_attr(getret, &at);
4148 error = VOP_PATHCONF(vp, _PC_LINK_MAX, &linkmax);
4150 error = VOP_PATHCONF(vp, _PC_NAME_MAX, &namemax);
4152 error = VOP_PATHCONF(vp, _PC_CHOWN_RESTRICTED, &chownres);
4154 error = VOP_PATHCONF(vp, _PC_NO_TRUNC, ¬runc);
4155 getret = VOP_GETATTR(vp, &at, cred, td);
4158 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3PATHCONF);
4159 nfsm_srvpostop_attr(getret, &at);
4164 pc = nfsm_build(struct nfsv3_pathconf *, NFSX_V3PATHCONF);
4166 pc->pc_linkmax = txdr_unsigned(linkmax);
4167 pc->pc_namemax = txdr_unsigned(namemax);
4168 pc->pc_notrunc = txdr_unsigned(notrunc);
4169 pc->pc_chownrestricted = txdr_unsigned(chownres);
4172 * These should probably be supported by VOP_PATHCONF(), but
4173 * until msdosfs is exportable (why would you want to?), the
4174 * Unix defaults should be ok.
4176 pc->pc_caseinsensitive = nfsrv_nfs_false;
4177 pc->pc_casepreserving = nfsrv_nfs_true;
4181 VFS_UNLOCK_GIANT(vfslocked);
4186 * Null operation, used by clients to ping server
4190 nfsrv_null(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
4191 struct thread *td, struct mbuf **mrq)
4193 struct mbuf *mrep = nfsd->nd_mrep;
4195 int error = NFSERR_RETVOID;
4196 struct mbuf *mb, *mreq;
4198 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
4205 * No operation, used for obsolete procedures
4209 nfsrv_noop(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
4210 struct thread *td, struct mbuf **mrq)
4212 struct mbuf *mrep = nfsd->nd_mrep;
4215 struct mbuf *mb, *mreq;
4217 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
4218 if (nfsd->nd_repstat)
4219 error = nfsd->nd_repstat;
4221 error = EPROCUNAVAIL;
4229 * Perform access checking for vnodes obtained from file handles that would
4230 * refer to files already opened by a Unix client. You cannot just use
4231 * vn_writechk() and VOP_ACCESS() for two reasons.
4232 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
4234 * 2 - The owner is to be given access irrespective of mode bits for some
4235 * operations, so that processes that chmod after opening a file don't
4236 * break. I don't like this because it opens a security hole, but since
4237 * the nfs server opens a security hole the size of a barn door anyhow,
4240 * The exception to rule 2 is EPERM. If a file is IMMUTABLE, VOP_ACCESS()
4241 * will return EPERM instead of EACCESS. EPERM is always an error.
4244 nfsrv_access(struct vnode *vp, int flags, struct ucred *cred,
4245 int rdonly, struct thread *td, int override)
4250 VFS_ASSERT_GIANT(vp->v_mount);
4252 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
4254 if (flags & VWRITE) {
4255 /* Just vn_writechk() changed to check rdonly */
4257 * Disallow write attempts on read-only filesystems;
4258 * unless the file is a socket or a block or character
4259 * device resident on the filesystem.
4261 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
4262 switch (vp->v_type) {
4272 * If there's shared text associated with
4273 * the inode, we can't allow writing.
4275 if (vp->v_vflag & VV_TEXT)
4279 error = VOP_GETATTR(vp, &vattr, cred, td);
4282 error = VOP_ACCESS(vp, flags, cred, td);
4284 * Allow certain operations for the owner (reads and writes
4285 * on files that are already open).
4287 if (override && error == EACCES && cred->cr_uid == vattr.va_uid)