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);
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);
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);
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);
442 postat_ret = VOP_GETATTR(vp, vap, cred);
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);
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);
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_WAIT, MT_DATA);
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);
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);
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_WAIT, MT_DATA);
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))) {
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);
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 /* XXXRW: unlocked write. */
1181 nfsrvstats.srvvop_writes++;
1182 FREE((caddr_t)iv, M_TEMP);
1184 aftat_ret = VOP_GETATTR(vp, vap, cred);
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);
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) {
1488 error = vn_start_write(NULL, &mntp, V_WAIT);
1489 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1491 mvfslocked = VFS_LOCK_GIANT(mntp);
1494 error = VOP_WRITE(vp, uiop, ioflags, cred);
1495 /* XXXRW: 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);
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,
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);
1867 bzero((caddr_t)fhp, sizeof(nfh));
1868 fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsid;
1869 error = VOP_VPTOFH(nd.ni_vp, &fhp->fh_fid);
1871 error = VOP_GETATTR(nd.ni_vp, vap, cred);
1874 if (exclusive_flag && !error &&
1875 bcmp(cverf, (caddr_t)&vap->va_atime, NFSX_V3CREATEVERF))
1877 if (dirp == nd.ni_dvp)
1878 diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
1880 /* Drop the other locks to avoid deadlock. */
1882 if (nd.ni_dvp == nd.ni_vp)
1892 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY);
1893 diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
1894 VOP_UNLOCK(dirp, 0);
1898 nfsm_reply(NFSX_SRVFH(v3) + NFSX_FATTR(v3) + NFSX_WCCDATA(v3));
1901 nfsm_srvpostop_fh(fhp);
1902 nfsm_srvpostop_attr(0, vap);
1904 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1905 } else if (!error) {
1906 /* v2 non-error case. */
1907 nfsm_srvfhtom(fhp, v3);
1908 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
1909 nfsm_srvfillattr(vap, fp);
1915 if (nd.ni_dvp == nd.ni_vp)
1922 if (nd.ni_startdir) {
1923 vrele(nd.ni_startdir);
1924 nd.ni_startdir = NULL;
1928 NDFREE(&nd, NDF_ONLY_PNBUF);
1929 vn_finished_write(mp);
1930 VFS_UNLOCK_GIANT(vfslocked);
1935 * nfs v3 mknod service
1938 nfsrv_mknod(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
1939 struct thread *td, struct mbuf **mrq)
1941 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1942 struct sockaddr *nam = nfsd->nd_nam;
1943 caddr_t dpos = nfsd->nd_dpos;
1944 struct ucred *cred = nfsd->nd_cr;
1945 struct vattr va, dirfor, diraft;
1946 struct vattr *vap = &va;
1948 struct nameidata nd;
1950 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
1951 u_int32_t major, minor;
1953 struct mbuf *mb, *mreq;
1954 struct vnode *vp, *dirp = NULL;
1957 struct mount *mp = NULL;
1958 int v3 = (nfsd->nd_flag & ND_NFSV3);
1962 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1965 panic("nfsrv_mknod: v3 proc called on a v2 connection");
1968 fhp = &nfh.fh_generic;
1970 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
1974 vfslocked = VFS_LOCK_GIANT(mp);
1975 (void) vn_start_write(NULL, &mp, V_WAIT);
1976 vfs_rel(mp); /* The write holds a ref. */
1977 nfsm_srvnamesiz(len);
1979 nd.ni_cnd.cn_cred = cred;
1980 nd.ni_cnd.cn_nameiop = CREATE;
1981 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | SAVESTART | MPSAFE;
1984 * Handle nfs_namei() call. If an error occurs, the nd structure
1985 * is not valid. However, nfsm_*() routines may still jump to
1989 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
1990 &dirp, v3, &dirfor, &dirfor_ret, td, FALSE);
1991 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
1993 nfsm_reply(NFSX_WCCDATA(1));
1994 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1998 tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED);
1999 vtyp = nfsv3tov_type(*tl);
2000 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
2001 error = NFSERR_BADTYPE;
2006 if (vtyp == VCHR || vtyp == VBLK) {
2007 tl = nfsm_dissect_nonblock(u_int32_t *, 2 * NFSX_UNSIGNED);
2008 major = fxdr_unsigned(u_int32_t, *tl++);
2009 minor = fxdr_unsigned(u_int32_t, *tl);
2010 vap->va_rdev = makedev(major, minor);
2014 * Iff doesn't exist, create it.
2020 vap->va_type = vtyp;
2021 if (vap->va_mode == (mode_t)VNOVAL)
2023 if (vtyp == VSOCK) {
2024 vrele(nd.ni_startdir);
2025 nd.ni_startdir = NULL;
2026 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap);
2028 NDFREE(&nd, NDF_ONLY_PNBUF);
2030 if (vtyp != VFIFO && (error = suser_cred(cred, 0)))
2032 error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap);
2034 NDFREE(&nd, NDF_ONLY_PNBUF);
2041 * Release dvp prior to lookup
2046 nd.ni_cnd.cn_nameiop = LOOKUP;
2047 nd.ni_cnd.cn_flags &= ~(LOCKPARENT);
2048 nd.ni_cnd.cn_thread = td;
2049 nd.ni_cnd.cn_cred = td->td_ucred;
2050 tvfslocked = VFS_LOCK_GIANT(nd.ni_startdir->v_mount);
2052 nd.ni_cnd.cn_flags |= GIANTHELD;
2053 error = lookup(&nd);
2055 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
2056 nd.ni_cnd.cn_flags &= ~GIANTHELD;
2060 if (nd.ni_cnd.cn_flags & ISSYMLINK)
2065 * send response, cleanup, return.
2070 bzero((caddr_t)fhp, sizeof(nfh));
2071 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
2072 error = VOP_VPTOFH(vp, &fhp->fh_fid);
2074 error = VOP_GETATTR(vp, vap, cred);
2077 if (nd.ni_dvp == nd.ni_vp)
2088 if (nd.ni_startdir) {
2089 vrele(nd.ni_startdir);
2090 nd.ni_startdir = NULL;
2092 NDFREE(&nd, NDF_ONLY_PNBUF);
2094 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY);
2095 diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2096 VOP_UNLOCK(dirp, 0);
2099 nfsm_reply(NFSX_SRVFH(1) + NFSX_POSTOPATTR(1) + NFSX_WCCDATA(1));
2102 nfsm_srvpostop_fh(fhp);
2103 nfsm_srvpostop_attr(0, vap);
2105 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2107 vn_finished_write(mp);
2108 VFS_UNLOCK_GIANT(vfslocked);
2112 if (nd.ni_dvp == nd.ni_vp)
2122 vrele(nd.ni_startdir);
2123 NDFREE(&nd, NDF_ONLY_PNBUF);
2124 vn_finished_write(mp);
2125 VFS_UNLOCK_GIANT(vfslocked);
2130 * nfs remove service
2133 nfsrv_remove(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2134 struct thread *td, struct mbuf **mrq)
2136 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2137 struct sockaddr *nam = nfsd->nd_nam;
2138 caddr_t dpos = nfsd->nd_dpos;
2139 struct ucred *cred = nfsd->nd_cr;
2140 struct nameidata nd;
2142 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
2143 int v3 = (nfsd->nd_flag & ND_NFSV3);
2144 struct mbuf *mb, *mreq;
2146 struct vattr dirfor, diraft;
2149 struct mount *mp = NULL;
2152 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2156 fhp = &nfh.fh_generic;
2158 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
2162 vfslocked = VFS_LOCK_GIANT(mp);
2163 (void) vn_start_write(NULL, &mp, V_WAIT);
2164 vfs_rel(mp); /* The write holds a ref. */
2165 nfsm_srvnamesiz(len);
2167 nd.ni_cnd.cn_cred = cred;
2168 nd.ni_cnd.cn_nameiop = DELETE;
2169 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | MPSAFE;
2170 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
2171 &dirp, v3, &dirfor, &dirfor_ret, td, FALSE);
2172 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
2178 if (nd.ni_vp->v_type == VDIR) {
2179 error = EPERM; /* POSIX */
2183 * The root of a mounted filesystem cannot be deleted.
2185 if (nd.ni_vp->v_vflag & VV_ROOT) {
2191 error = VOP_REMOVE(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
2192 NDFREE(&nd, NDF_ONLY_PNBUF);
2196 if (dirp == nd.ni_dvp)
2197 diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2199 /* Drop the other locks to avoid deadlock. */
2201 if (nd.ni_dvp == nd.ni_vp)
2211 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY);
2212 diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2213 VOP_UNLOCK(dirp, 0);
2219 nfsm_reply(NFSX_WCCDATA(v3));
2221 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2225 NDFREE(&nd, NDF_ONLY_PNBUF);
2227 if (nd.ni_dvp == nd.ni_vp)
2234 vn_finished_write(mp);
2235 VFS_UNLOCK_GIANT(vfslocked);
2240 * nfs rename service
2243 nfsrv_rename(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2244 struct thread *td, struct mbuf **mrq)
2246 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2247 struct sockaddr *nam = nfsd->nd_nam;
2248 caddr_t dpos = nfsd->nd_dpos;
2249 struct ucred *cred = nfsd->nd_cr;
2251 int error = 0, len, len2, fdirfor_ret = 1, fdiraft_ret = 1;
2252 int tdirfor_ret = 1, tdiraft_ret = 1;
2253 int v3 = (nfsd->nd_flag & ND_NFSV3);
2254 struct mbuf *mb, *mreq;
2255 struct nameidata fromnd, tond;
2256 struct vnode *fvp, *tvp, *tdvp, *fdirp = NULL;
2257 struct vnode *tdirp = NULL;
2258 struct vattr fdirfor, fdiraft, tdirfor, tdiraft;
2260 fhandle_t *ffhp, *tfhp;
2262 struct mount *mp = NULL;
2265 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2270 ffhp = &fnfh.fh_generic;
2271 tfhp = &tnfh.fh_generic;
2274 * Clear fields incase goto nfsmout occurs from macro.
2280 nfsm_srvmtofh(ffhp);
2281 if ((mp = vfs_getvfs(&ffhp->fh_fsid)) == NULL) {
2285 vfslocked = VFS_LOCK_GIANT(mp);
2286 (void) vn_start_write(NULL, &mp, V_WAIT);
2287 vfs_rel(mp); /* The write holds a ref. */
2288 nfsm_srvnamesiz(len);
2290 * Remember our original uid so that we can reset cr_uid before
2291 * the second nfs_namei() call, in case it is remapped.
2293 saved_uid = cred->cr_uid;
2294 fromnd.ni_cnd.cn_cred = cred;
2295 fromnd.ni_cnd.cn_nameiop = DELETE;
2296 fromnd.ni_cnd.cn_flags = WANTPARENT | SAVESTART | MPSAFE;
2297 error = nfs_namei(&fromnd, ffhp, len, slp, nam, &md,
2298 &dpos, &fdirp, v3, &fdirfor, &fdirfor_ret, td, FALSE);
2299 vfslocked = nfsrv_lockedpair_nd(vfslocked, &fromnd);
2305 nfsm_reply(2 * NFSX_WCCDATA(v3));
2307 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
2308 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
2314 nfsm_srvmtofh(tfhp);
2315 nfsm_srvnamesiz(len2);
2316 cred->cr_uid = saved_uid;
2317 tond.ni_cnd.cn_cred = cred;
2318 tond.ni_cnd.cn_nameiop = RENAME;
2319 tond.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | NOCACHE | SAVESTART | MPSAFE;
2320 error = nfs_namei(&tond, tfhp, len2, slp, nam, &md,
2321 &dpos, &tdirp, v3, &tdirfor, &tdirfor_ret, td, FALSE);
2322 vfslocked = nfsrv_lockedpair_nd(vfslocked, &tond);
2333 if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
2339 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
2346 if (tvp->v_type == VDIR && tvp->v_mountedhere) {
2354 if (fvp->v_type == VDIR && fvp->v_mountedhere) {
2361 if (fvp->v_mount != tdvp->v_mount) {
2375 * If source is the same as the destination (that is the
2376 * same vnode with the same name in the same directory),
2377 * then there is nothing to do.
2379 if (fvp == tvp && fromnd.ni_dvp == tdvp &&
2380 fromnd.ni_cnd.cn_namelen == tond.ni_cnd.cn_namelen &&
2381 !bcmp(fromnd.ni_cnd.cn_nameptr, tond.ni_cnd.cn_nameptr,
2382 fromnd.ni_cnd.cn_namelen))
2387 * The VOP_RENAME function releases all vnode references &
2388 * locks prior to returning so we need to clear the pointers
2389 * to bypass cleanup code later on.
2391 error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd,
2392 tond.ni_dvp, tond.ni_vp, &tond.ni_cnd);
2393 fromnd.ni_dvp = NULL;
2394 fromnd.ni_vp = NULL;
2398 NDFREE(&fromnd, NDF_ONLY_PNBUF);
2399 NDFREE(&tond, NDF_ONLY_PNBUF);
2407 nfsm_reply(2 * NFSX_WCCDATA(v3));
2409 /* Release existing locks to prevent deadlock. */
2411 if (tond.ni_dvp == tond.ni_vp)
2422 vn_lock(fdirp, LK_EXCLUSIVE | LK_RETRY);
2423 fdiraft_ret = VOP_GETATTR(fdirp, &fdiraft, cred);
2424 VOP_UNLOCK(fdirp, 0);
2427 vn_lock(tdirp, LK_EXCLUSIVE | LK_RETRY);
2428 tdiraft_ret = VOP_GETATTR(tdirp, &tdiraft, cred);
2429 VOP_UNLOCK(tdirp, 0);
2431 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
2432 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
2439 * Clear out tond related fields
2442 if (tond.ni_dvp == tond.ni_vp)
2451 if (tond.ni_startdir)
2452 vrele(tond.ni_startdir);
2453 NDFREE(&tond, NDF_ONLY_PNBUF);
2455 * Clear out fromnd related fields
2459 if (fromnd.ni_startdir)
2460 vrele(fromnd.ni_startdir);
2461 NDFREE(&fromnd, NDF_ONLY_PNBUF);
2463 vrele(fromnd.ni_dvp);
2465 vrele(fromnd.ni_vp);
2467 vn_finished_write(mp);
2468 VFS_UNLOCK_GIANT(vfslocked);
2476 nfsrv_link(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2477 struct thread *td, struct mbuf **mrq)
2479 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2480 struct sockaddr *nam = nfsd->nd_nam;
2481 caddr_t dpos = nfsd->nd_dpos;
2482 struct ucred *cred = nfsd->nd_cr;
2483 struct nameidata nd;
2485 int error = 0, rdonly, len, dirfor_ret = 1, diraft_ret = 1;
2486 int getret = 1, v3 = (nfsd->nd_flag & ND_NFSV3);
2487 struct mbuf *mb, *mreq;
2488 struct vnode *vp = NULL, *xp, *dirp = NULL;
2489 struct vattr dirfor, diraft, at;
2491 fhandle_t *fhp, *dfhp;
2492 struct mount *mp = NULL;
2496 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2500 fhp = &nfh.fh_generic;
2501 dfhp = &dnfh.fh_generic;
2503 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
2507 vfslocked = VFS_LOCK_GIANT(mp);
2508 (void) vn_start_write(NULL, &mp, V_WAIT);
2509 vfs_rel(mp); /* The write holds a ref. */
2510 nfsm_srvmtofh(dfhp);
2511 nfsm_srvnamesiz(len);
2513 error = nfsrv_fhtovp(fhp, TRUE, &vp, &tvfslocked, cred, slp,
2514 nam, &rdonly, TRUE);
2515 vfslocked = nfsrv_lockedpair(vfslocked, tvfslocked);
2517 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2519 nfsm_srvpostop_attr(getret, &at);
2520 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2527 getret = VOP_GETATTR(vp, &at, cred);
2528 if (vp->v_type == VDIR) {
2529 error = EPERM; /* POSIX */
2533 nd.ni_cnd.cn_cred = cred;
2534 nd.ni_cnd.cn_nameiop = CREATE;
2535 nd.ni_cnd.cn_flags = LOCKPARENT | MPSAFE | MPSAFE;
2536 error = nfs_namei(&nd, dfhp, len, slp, nam, &md, &dpos,
2537 &dirp, v3, &dirfor, &dirfor_ret, td, FALSE);
2538 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
2556 if (vp->v_mount != xp->v_mount) {
2562 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2563 error = VOP_LINK(nd.ni_dvp, vp, &nd.ni_cnd);
2564 NDFREE(&nd, NDF_ONLY_PNBUF);
2569 getret = VOP_GETATTR(vp, &at, cred);
2572 if (dirp == nd.ni_dvp)
2573 diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2575 /* Release existing locks to prevent deadlock. */
2577 if (nd.ni_dvp == nd.ni_vp)
2587 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY);
2588 diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2589 VOP_UNLOCK(dirp, 0);
2593 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2595 nfsm_srvpostop_attr(getret, &at);
2596 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2602 NDFREE(&nd, NDF_ONLY_PNBUF);
2606 if (nd.ni_dvp == nd.ni_vp)
2615 vn_finished_write(mp);
2616 VFS_UNLOCK_GIANT(vfslocked);
2621 * nfs symbolic link service
2624 nfsrv_symlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2625 struct thread *td, struct mbuf **mrq)
2627 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2628 struct sockaddr *nam = nfsd->nd_nam;
2629 caddr_t dpos = nfsd->nd_dpos;
2630 struct ucred *cred = nfsd->nd_cr;
2631 struct vattr va, dirfor, diraft;
2632 struct nameidata nd;
2633 struct vattr *vap = &va;
2634 struct nfsv2_sattr *sp;
2635 char *bpos, *pathcp = NULL;
2638 int error = 0, len, len2, dirfor_ret = 1, diraft_ret = 1;
2639 int v3 = (nfsd->nd_flag & ND_NFSV3);
2640 struct mbuf *mb, *mreq;
2641 struct vnode *dirp = NULL;
2644 struct mount *mp = NULL;
2648 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2652 fhp = &nfh.fh_generic;
2654 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
2658 vfslocked = VFS_LOCK_GIANT(mp);
2659 (void) vn_start_write(NULL, &mp, V_WAIT);
2660 vfs_rel(mp); /* The write holds a ref. */
2661 nfsm_srvnamesiz(len);
2662 nd.ni_cnd.cn_cred = cred;
2663 nd.ni_cnd.cn_nameiop = CREATE;
2664 nd.ni_cnd.cn_flags = LOCKPARENT | SAVESTART | MPSAFE;
2665 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
2666 &dirp, v3, &dirfor, &dirfor_ret, td, FALSE);
2667 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
2672 nfsm_srvpathsiz(len2);
2680 MALLOC(pathcp, caddr_t, len2 + 1, M_TEMP, M_WAITOK);
2681 iv.iov_base = pathcp;
2683 io.uio_resid = len2;
2687 io.uio_segflg = UIO_SYSSPACE;
2688 io.uio_rw = UIO_READ;
2690 nfsm_mtouio(&io, len2);
2692 sp = nfsm_dissect_nonblock(struct nfsv2_sattr *, NFSX_V2SATTR);
2693 vap->va_mode = nfstov_mode(sp->sa_mode);
2695 *(pathcp + len2) = '\0';
2702 * issue symlink op. SAVESTART is set so the underlying path component
2703 * is only freed by the VOP if an error occurs.
2705 if (vap->va_mode == (mode_t)VNOVAL)
2707 error = VOP_SYMLINK(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap, pathcp);
2709 NDFREE(&nd, NDF_ONLY_PNBUF);
2714 * releases directory prior to potential lookup op.
2722 * Issue lookup. Leave SAVESTART set so we can easily free
2723 * the name buffer later on.
2725 * since LOCKPARENT is not set, ni_dvp will be garbage on
2726 * return whether an error occurs or not.
2728 nd.ni_cnd.cn_nameiop = LOOKUP;
2729 nd.ni_cnd.cn_flags &= ~(LOCKPARENT | FOLLOW);
2730 nd.ni_cnd.cn_flags |= (NOFOLLOW | LOCKLEAF);
2731 nd.ni_cnd.cn_thread = td;
2732 nd.ni_cnd.cn_cred = cred;
2733 tvfslocked = VFS_LOCK_GIANT(nd.ni_startdir->v_mount);
2735 nd.ni_cnd.cn_flags |= GIANTHELD;
2736 error = lookup(&nd);
2738 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
2739 nd.ni_cnd.cn_flags &= ~GIANTHELD;
2742 bzero((caddr_t)fhp, sizeof(nfh));
2743 fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsid;
2744 error = VOP_VPTOFH(nd.ni_vp, &fhp->fh_fid);
2746 error = VOP_GETATTR(nd.ni_vp, vap, cred);
2754 * These releases aren't strictly required, does even doing them
2755 * make any sense? XXX can nfsm_reply() block?
2758 FREE(pathcp, M_TEMP);
2762 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY);
2763 diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2764 VOP_UNLOCK(dirp, 0);
2766 if (nd.ni_startdir) {
2767 vrele(nd.ni_startdir);
2768 nd.ni_startdir = NULL;
2770 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2773 nfsm_srvpostop_fh(fhp);
2774 nfsm_srvpostop_attr(0, vap);
2776 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2782 NDFREE(&nd, NDF_ONLY_PNBUF);
2784 if (nd.ni_dvp == nd.ni_vp)
2792 vrele(nd.ni_startdir);
2796 FREE(pathcp, M_TEMP);
2798 vn_finished_write(mp);
2799 VFS_UNLOCK_GIANT(vfslocked);
2807 nfsrv_mkdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2808 struct thread *td, struct mbuf **mrq)
2810 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2811 struct sockaddr *nam = nfsd->nd_nam;
2812 caddr_t dpos = nfsd->nd_dpos;
2813 struct ucred *cred = nfsd->nd_cr;
2814 struct vattr va, dirfor, diraft;
2815 struct vattr *vap = &va;
2816 struct nfs_fattr *fp;
2817 struct nameidata nd;
2820 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
2821 int v3 = (nfsd->nd_flag & ND_NFSV3);
2822 struct mbuf *mb, *mreq;
2823 struct vnode *dirp = NULL;
2827 struct mount *mp = NULL;
2830 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2834 fhp = &nfh.fh_generic;
2836 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
2840 vfslocked = VFS_LOCK_GIANT(mp);
2841 (void) vn_start_write(NULL, &mp, V_WAIT);
2842 vfs_rel(mp); /* The write holds a ref. */
2843 nfsm_srvnamesiz(len);
2844 nd.ni_cnd.cn_cred = cred;
2845 nd.ni_cnd.cn_nameiop = CREATE;
2846 nd.ni_cnd.cn_flags = LOCKPARENT | MPSAFE;
2848 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
2849 &dirp, v3, &dirfor, &dirfor_ret, td, FALSE);
2850 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
2856 nfsm_reply(NFSX_WCCDATA(v3));
2858 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2866 tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED);
2867 vap->va_mode = nfstov_mode(*tl++);
2871 * At this point nd.ni_dvp is referenced and exclusively locked and
2872 * nd.ni_vp, if it exists, is referenced but not locked.
2875 vap->va_type = VDIR;
2876 if (nd.ni_vp != NULL) {
2877 NDFREE(&nd, NDF_ONLY_PNBUF);
2883 * Issue mkdir op. Since SAVESTART is not set, the pathname
2884 * component is freed by the VOP call. This will fill-in
2885 * nd.ni_vp, reference, and exclusively lock it.
2887 if (vap->va_mode == (mode_t)VNOVAL)
2889 error = VOP_MKDIR(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap);
2890 NDFREE(&nd, NDF_ONLY_PNBUF);
2897 bzero((caddr_t)fhp, sizeof(nfh));
2898 fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsid;
2899 error = VOP_VPTOFH(nd.ni_vp, &fhp->fh_fid);
2901 error = VOP_GETATTR(nd.ni_vp, vap, cred);
2905 if (dirp == nd.ni_dvp) {
2906 diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2908 /* Release existing locks to prevent deadlock. */
2910 NDFREE(&nd, NDF_ONLY_PNBUF);
2911 if (nd.ni_dvp == nd.ni_vp && vpexcl)
2924 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY);
2925 diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2926 VOP_UNLOCK(dirp, 0);
2929 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2932 nfsm_srvpostop_fh(fhp);
2933 nfsm_srvpostop_attr(0, vap);
2935 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2936 } else if (!error) {
2937 /* v2 non-error case. */
2938 nfsm_srvfhtom(fhp, v3);
2939 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
2940 nfsm_srvfillattr(vap, fp);
2947 NDFREE(&nd, NDF_ONLY_PNBUF);
2948 if (nd.ni_dvp == nd.ni_vp && vpexcl)
2961 vn_finished_write(mp);
2962 VFS_UNLOCK_GIANT(vfslocked);
2970 nfsrv_rmdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2971 struct thread *td, struct mbuf **mrq)
2973 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2974 struct sockaddr *nam = nfsd->nd_nam;
2975 caddr_t dpos = nfsd->nd_dpos;
2976 struct ucred *cred = nfsd->nd_cr;
2978 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
2979 int v3 = (nfsd->nd_flag & ND_NFSV3);
2980 struct mbuf *mb, *mreq;
2981 struct vnode *vp, *dirp = NULL;
2982 struct vattr dirfor, diraft;
2985 struct nameidata nd;
2986 struct mount *mp = NULL;
2989 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2993 fhp = &nfh.fh_generic;
2995 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
2999 vfslocked = VFS_LOCK_GIANT(mp);
3000 (void) vn_start_write(NULL, &mp, V_WAIT);
3001 vfs_rel(mp); /* The write holds a ref. */
3002 nfsm_srvnamesiz(len);
3003 nd.ni_cnd.cn_cred = cred;
3004 nd.ni_cnd.cn_nameiop = DELETE;
3005 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | MPSAFE;
3006 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
3007 &dirp, v3, &dirfor, &dirfor_ret, td, FALSE);
3008 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
3014 nfsm_reply(NFSX_WCCDATA(v3));
3016 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
3021 if (vp->v_type != VDIR) {
3026 * No rmdir "." please.
3028 if (nd.ni_dvp == vp) {
3033 * The root of a mounted filesystem cannot be deleted.
3035 if (vp->v_vflag & VV_ROOT)
3039 * Issue or abort op. Since SAVESTART is not set, path name
3040 * component is freed by the VOP after either.
3043 error = VOP_RMDIR(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
3044 NDFREE(&nd, NDF_ONLY_PNBUF);
3047 if (dirp == nd.ni_dvp)
3048 diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
3050 /* Release existing locks to prevent deadlock. */
3052 if (nd.ni_dvp == nd.ni_vp)
3061 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY);
3062 diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
3063 VOP_UNLOCK(dirp, 0);
3066 nfsm_reply(NFSX_WCCDATA(v3));
3069 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
3073 NDFREE(&nd, NDF_ONLY_PNBUF);
3075 if (nd.ni_dvp == nd.ni_vp)
3085 vn_finished_write(mp);
3086 VFS_UNLOCK_GIANT(vfslocked);
3091 * nfs readdir service
3092 * - mallocs what it thinks is enough to read
3093 * count rounded up to a multiple of NFS_DIRBLKSIZ <= NFS_MAXREADDIR
3094 * - calls VOP_READDIR()
3095 * - loops around building the reply
3096 * if the output generated exceeds count break out of loop
3097 * The nfsm_clget macro is used here so that the reply will be packed
3098 * tightly in mbuf clusters.
3099 * - it only knows that it has encountered eof when the VOP_READDIR()
3101 * - as such one readdir rpc will return eof false although you are there
3102 * and then the next will return eof
3103 * - it trims out records with d_fileno == 0
3104 * this doesn't matter for Unix clients, but they might confuse clients
3106 * NB: It is tempting to set eof to true if the VOP_READDIR() reads less
3107 * than requested, but this may not apply to all filesystems. For
3108 * example, client NFS does not { although it is never remote mounted
3110 * The alternate call nfsrv_readdirplus() does lookups as well.
3111 * PS: The NFS protocol spec. does not clarify what the "count" byte
3112 * argument is a count of.. just name strings and file id's or the
3113 * entire reply rpc or ...
3114 * I tried just file name and id sizes and it confused the Sun client,
3115 * so I am using the full rpc size now. The "paranoia.." comment refers
3116 * to including the status longwords that are not a part of the dir.
3117 * "entry" structures, but are in the rpc.
3121 u_int32_t fl_postopok;
3122 u_int32_t fl_fattr[NFSX_V3FATTR / sizeof (u_int32_t)];
3124 u_int32_t fl_fhsize;
3125 u_int32_t fl_nfh[NFSX_V3FH / sizeof (u_int32_t)];
3129 nfsrv_readdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3130 struct thread *td, struct mbuf **mrq)
3132 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3133 struct sockaddr *nam = nfsd->nd_nam;
3134 caddr_t dpos = nfsd->nd_dpos;
3135 struct ucred *cred = nfsd->nd_cr;
3142 struct mbuf *mb, *mreq;
3143 char *cpos, *cend, *rbuf;
3144 struct vnode *vp = NULL;
3150 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
3151 int siz, cnt, fullsiz, eofflag, rdonly, ncookies;
3152 int v3 = (nfsd->nd_flag & ND_NFSV3);
3153 u_quad_t off, toff, verf;
3154 u_long *cookies = NULL, *cookiep; /* needs to be int64_t or off_t */
3157 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3159 fhp = &nfh.fh_generic;
3162 tl = nfsm_dissect_nonblock(u_int32_t *, 5 * NFSX_UNSIGNED);
3163 toff = fxdr_hyper(tl);
3165 verf = fxdr_hyper(tl);
3168 tl = nfsm_dissect_nonblock(u_int32_t *, 2 * NFSX_UNSIGNED);
3169 toff = fxdr_unsigned(u_quad_t, *tl++);
3170 verf = 0; /* shut up gcc */
3173 cnt = fxdr_unsigned(int, *tl);
3174 siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
3175 xfer = NFS_SRVMAXDATA(nfsd);
3181 error = nfsrv_fhtovp(fhp, 1, &vp, &vfslocked, cred, slp,
3182 nam, &rdonly, TRUE);
3183 if (!error && vp->v_type != VDIR) {
3189 nfsm_reply(NFSX_UNSIGNED);
3191 nfsm_srvpostop_attr(getret, &at);
3197 * Obtain lock on vnode for this section of the code
3200 error = getret = VOP_GETATTR(vp, &at, cred);
3203 * XXX This check may be too strict for Solaris 2.5 clients.
3205 if (!error && toff && verf && verf != at.va_filerev)
3206 error = NFSERR_BAD_COOKIE;
3210 error = nfsrv_access(vp, VEXEC, cred, rdonly, td, 0);
3214 nfsm_reply(NFSX_POSTOPATTR(v3));
3216 nfsm_srvpostop_attr(getret, &at);
3223 * end section. Allocate rbuf and continue
3225 MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
3228 iv.iov_len = fullsiz;
3231 io.uio_offset = (off_t)off;
3232 io.uio_resid = fullsiz;
3233 io.uio_segflg = UIO_SYSSPACE;
3234 io.uio_rw = UIO_READ;
3237 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3239 free((caddr_t)cookies, M_TEMP);
3242 error = VOP_READDIR(vp, &io, cred, &eofflag, &ncookies, &cookies);
3243 off = (off_t)io.uio_offset;
3244 if (!cookies && !error)
3245 error = NFSERR_PERM;
3247 getret = VOP_GETATTR(vp, &at, cred);
3255 free((caddr_t)rbuf, M_TEMP);
3257 free((caddr_t)cookies, M_TEMP);
3258 nfsm_reply(NFSX_POSTOPATTR(v3));
3260 nfsm_srvpostop_attr(getret, &at);
3265 siz -= io.uio_resid;
3268 * If nothing read, return eof
3274 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) +
3277 nfsm_srvpostop_attr(getret, &at);
3278 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
3279 txdr_hyper(at.va_filerev, tl);
3282 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
3283 *tl++ = nfsrv_nfs_false;
3284 *tl = nfsrv_nfs_true;
3285 FREE((caddr_t)rbuf, M_TEMP);
3286 FREE((caddr_t)cookies, M_TEMP);
3293 * Check for degenerate cases of nothing useful read.
3294 * If so go try again
3298 dp = (struct dirent *)cpos;
3301 * For some reason FreeBSD's ufs_readdir() chooses to back the
3302 * directory offset up to a block boundary, so it is necessary to
3303 * skip over the records that precede the requested offset. This
3304 * requires the assumption that file offset cookies monotonically
3307 while (cpos < cend && ncookies > 0 &&
3308 (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
3309 ((u_quad_t)(*cookiep)) <= toff)) {
3310 cpos += dp->d_reclen;
3311 dp = (struct dirent *)cpos;
3315 if (cpos >= cend || ncookies == 0) {
3321 len = 3 * NFSX_UNSIGNED; /* paranoia, probably can be 0 */
3322 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + siz);
3324 nfsm_srvpostop_attr(getret, &at);
3325 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
3326 txdr_hyper(at.va_filerev, tl);
3330 be = bp + M_TRAILINGSPACE(mp);
3332 /* Loop through the records and build reply */
3333 while (cpos < cend && ncookies > 0) {
3334 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
3335 nlen = dp->d_namlen;
3336 rem = nfsm_rndup(nlen) - nlen;
3337 len += (4 * NFSX_UNSIGNED + nlen + rem);
3339 len += 2 * NFSX_UNSIGNED;
3345 * Build the directory record xdr from
3349 *tl = nfsrv_nfs_true;
3350 bp += NFSX_UNSIGNED;
3354 bp += NFSX_UNSIGNED;
3357 *tl = txdr_unsigned(dp->d_fileno);
3358 bp += NFSX_UNSIGNED;
3360 *tl = txdr_unsigned(nlen);
3361 bp += NFSX_UNSIGNED;
3363 /* And loop around copying the name */
3372 bcopy(cp, bp, tsiz);
3378 /* And null pad to an int32_t boundary. */
3379 for (i = 0; i < rem; i++)
3383 /* Finish off the record */
3386 bp += NFSX_UNSIGNED;
3389 *tl = txdr_unsigned(*cookiep);
3390 bp += NFSX_UNSIGNED;
3392 cpos += dp->d_reclen;
3393 dp = (struct dirent *)cpos;
3400 *tl = nfsrv_nfs_false;
3401 bp += NFSX_UNSIGNED;
3404 *tl = nfsrv_nfs_true;
3406 *tl = nfsrv_nfs_false;
3407 bp += NFSX_UNSIGNED;
3410 mp->m_len = bp - mtod(mp, caddr_t);
3412 mp->m_len += bp - bpos;
3413 FREE((caddr_t)rbuf, M_TEMP);
3414 FREE((caddr_t)cookies, M_TEMP);
3419 VFS_UNLOCK_GIANT(vfslocked);
3424 nfsrv_readdirplus(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3425 struct thread *td, struct mbuf **mrq)
3427 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3428 struct sockaddr *nam = nfsd->nd_nam;
3429 caddr_t dpos = nfsd->nd_dpos;
3430 struct ucred *cred = nfsd->nd_cr;
3437 struct mbuf *mb, *mreq;
3438 char *cpos, *cend, *rbuf;
3439 struct vnode *vp = NULL, *nvp;
3442 fhandle_t *fhp, *nfhp = (fhandle_t *)fl.fl_nfh;
3445 struct vattr va, at, *vap = &va;
3446 struct nfs_fattr *fp;
3447 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
3448 int siz, cnt, fullsiz, eofflag, rdonly, dirlen, ncookies;
3449 u_quad_t off, toff, verf;
3450 u_long *cookies = NULL, *cookiep; /* needs to be int64_t or off_t */
3451 int v3 = (nfsd->nd_flag & ND_NFSV3);
3454 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3457 panic("nfsrv_readdirplus: v3 proc called on a v2 connection");
3458 fhp = &nfh.fh_generic;
3460 tl = nfsm_dissect_nonblock(u_int32_t *, 6 * NFSX_UNSIGNED);
3461 toff = fxdr_hyper(tl);
3463 verf = fxdr_hyper(tl);
3465 siz = fxdr_unsigned(int, *tl++);
3466 cnt = fxdr_unsigned(int, *tl);
3468 siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
3469 xfer = NFS_SRVMAXDATA(nfsd);
3475 error = nfsrv_fhtovp(fhp, 1, &vp, &vfslocked, cred, slp,
3476 nam, &rdonly, TRUE);
3477 if (!error && vp->v_type != VDIR) {
3483 nfsm_reply(NFSX_UNSIGNED);
3484 nfsm_srvpostop_attr(getret, &at);
3488 error = getret = VOP_GETATTR(vp, &at, cred);
3491 * XXX This check may be too strict for Solaris 2.5 clients.
3493 if (!error && toff && verf && verf != at.va_filerev)
3494 error = NFSERR_BAD_COOKIE;
3497 error = nfsrv_access(vp, VEXEC, cred, rdonly, td, 0);
3501 nfsm_reply(NFSX_V3POSTOPATTR);
3502 nfsm_srvpostop_attr(getret, &at);
3507 MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
3510 iv.iov_len = fullsiz;
3513 io.uio_offset = (off_t)off;
3514 io.uio_resid = fullsiz;
3515 io.uio_segflg = UIO_SYSSPACE;
3516 io.uio_rw = UIO_READ;
3519 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3521 free((caddr_t)cookies, M_TEMP);
3524 error = VOP_READDIR(vp, &io, cred, &eofflag, &ncookies, &cookies);
3525 off = (u_quad_t)io.uio_offset;
3526 getret = VOP_GETATTR(vp, &at, cred);
3528 if (!cookies && !error)
3529 error = NFSERR_PERM;
3536 free((caddr_t)cookies, M_TEMP);
3537 free((caddr_t)rbuf, M_TEMP);
3538 nfsm_reply(NFSX_V3POSTOPATTR);
3539 nfsm_srvpostop_attr(getret, &at);
3544 siz -= io.uio_resid;
3547 * If nothing read, return eof
3553 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF +
3555 nfsm_srvpostop_attr(getret, &at);
3556 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
3557 txdr_hyper(at.va_filerev, tl);
3559 *tl++ = nfsrv_nfs_false;
3560 *tl = nfsrv_nfs_true;
3561 FREE((caddr_t)cookies, M_TEMP);
3562 FREE((caddr_t)rbuf, M_TEMP);
3569 * Check for degenerate cases of nothing useful read.
3570 * If so go try again
3574 dp = (struct dirent *)cpos;
3577 * For some reason FreeBSD's ufs_readdir() chooses to back the
3578 * directory offset up to a block boundary, so it is necessary to
3579 * skip over the records that precede the requested offset. This
3580 * requires the assumption that file offset cookies monotonically
3583 while (cpos < cend && ncookies > 0 &&
3584 (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
3585 ((u_quad_t)(*cookiep)) <= toff)) {
3586 cpos += dp->d_reclen;
3587 dp = (struct dirent *)cpos;
3591 if (cpos >= cend || ncookies == 0) {
3598 * Probe one of the directory entries to see if the filesystem
3601 if (VFS_VGET(vp->v_mount, dp->d_fileno, LK_EXCLUSIVE, &nvp) ==
3603 error = NFSERR_NOTSUPP;
3606 free((caddr_t)cookies, M_TEMP);
3607 free((caddr_t)rbuf, M_TEMP);
3608 nfsm_reply(NFSX_V3POSTOPATTR);
3609 nfsm_srvpostop_attr(getret, &at);
3616 dirlen = len = NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF +
3619 nfsm_srvpostop_attr(getret, &at);
3620 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
3621 txdr_hyper(at.va_filerev, tl);
3624 be = bp + M_TRAILINGSPACE(mp);
3626 /* Loop through the records and build reply */
3627 while (cpos < cend && ncookies > 0) {
3628 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
3629 nlen = dp->d_namlen;
3630 rem = nfsm_rndup(nlen)-nlen;
3633 * For readdir_and_lookup get the vnode using
3636 if (VFS_VGET(vp->v_mount, dp->d_fileno, LK_EXCLUSIVE,
3639 bzero((caddr_t)nfhp, NFSX_V3FH);
3641 nvp->v_mount->mnt_stat.f_fsid;
3643 * XXXRW: Assert the mountpoints are the same so that
3644 * we know that acquiring Giant based on the
3645 * directory is the right thing for the child.
3647 KASSERT(nvp->v_mount == vp->v_mount,
3648 ("nfsrv_readdirplus: nvp mount != vp mount"));
3649 if (VOP_VPTOFH(nvp, &nfhp->fh_fid)) {
3654 if (VOP_GETATTR(nvp, vap, cred)) {
3663 * If either the dircount or maxcount will be
3664 * exceeded, get out now. Both of these lengths
3665 * are calculated conservatively, including all
3668 len += (8 * NFSX_UNSIGNED + nlen + rem + NFSX_V3FH +
3670 dirlen += (6 * NFSX_UNSIGNED + nlen + rem);
3671 if (len > cnt || dirlen > fullsiz) {
3677 * Build the directory record xdr from
3680 fp = (struct nfs_fattr *)&fl.fl_fattr;
3681 nfsm_srvfillattr(vap, fp);
3682 fl.fl_fhsize = txdr_unsigned(NFSX_V3FH);
3683 fl.fl_fhok = nfsrv_nfs_true;
3684 fl.fl_postopok = nfsrv_nfs_true;
3685 fl.fl_off.nfsuquad[0] = 0;
3686 fl.fl_off.nfsuquad[1] = txdr_unsigned(*cookiep);
3689 *tl = nfsrv_nfs_true;
3690 bp += NFSX_UNSIGNED;
3693 bp += NFSX_UNSIGNED;
3695 *tl = txdr_unsigned(dp->d_fileno);
3696 bp += NFSX_UNSIGNED;
3698 *tl = txdr_unsigned(nlen);
3699 bp += NFSX_UNSIGNED;
3701 /* And loop around copying the name */
3706 if ((bp + xfer) > be)
3710 bcopy(cp, bp, tsiz);
3716 /* And null pad to an int32_t boundary. */
3717 for (i = 0; i < rem; i++)
3721 * Now copy the flrep structure out.
3723 xfer = sizeof (struct flrep);
3727 if ((bp + xfer) > be)
3731 bcopy(cp, bp, tsiz);
3739 cpos += dp->d_reclen;
3740 dp = (struct dirent *)cpos;
3747 *tl = nfsrv_nfs_false;
3748 bp += NFSX_UNSIGNED;
3751 *tl = nfsrv_nfs_true;
3753 *tl = nfsrv_nfs_false;
3754 bp += NFSX_UNSIGNED;
3757 mp->m_len = bp - mtod(mp, caddr_t);
3759 mp->m_len += bp - bpos;
3760 FREE((caddr_t)cookies, M_TEMP);
3761 FREE((caddr_t)rbuf, M_TEMP);
3765 VFS_UNLOCK_GIANT(vfslocked);
3770 * nfs commit service
3773 nfsrv_commit(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3774 struct thread *td, struct mbuf **mrq)
3776 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3777 struct sockaddr *nam = nfsd->nd_nam;
3778 caddr_t dpos = nfsd->nd_dpos;
3779 struct ucred *cred = nfsd->nd_cr;
3780 struct vattr bfor, aft;
3781 struct vnode *vp = NULL;
3786 int error = 0, rdonly, for_ret = 1, aft_ret = 1, cnt;
3787 struct mbuf *mb, *mreq;
3789 struct mount *mp = NULL;
3790 int v3 = (nfsd->nd_flag & ND_NFSV3);
3794 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3797 panic("nfsrv_commit: v3 proc called on a v2 connection");
3798 fhp = &nfh.fh_generic;
3800 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
3804 vfslocked = VFS_LOCK_GIANT(mp);
3805 (void) vn_start_write(NULL, &mp, V_WAIT);
3806 vfs_rel(mp); /* The write holds a ref. */
3807 tl = nfsm_dissect_nonblock(u_int32_t *, 3 * NFSX_UNSIGNED);
3810 * XXX At this time VOP_FSYNC() does not accept offset and byte
3811 * count parameters, so these arguments are useless (someday maybe).
3813 off = fxdr_hyper(tl);
3815 cnt = fxdr_unsigned(int, *tl);
3816 error = nfsrv_fhtovp(fhp, 1, &vp, &tvfslocked, cred, slp,
3817 nam, &rdonly, TRUE);
3818 vfslocked = nfsrv_lockedpair(vfslocked, tvfslocked);
3820 nfsm_reply(2 * NFSX_UNSIGNED);
3821 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3825 for_ret = VOP_GETATTR(vp, &bfor, cred);
3827 if (cnt > MAX_COMMIT_COUNT) {
3829 * Give up and do the whole thing
3832 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
3833 VM_OBJECT_LOCK(vp->v_object);
3834 vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
3835 VM_OBJECT_UNLOCK(vp->v_object);
3837 error = VOP_FSYNC(vp, MNT_WAIT, td);
3840 * Locate and synchronously write any buffers that fall
3841 * into the requested range. Note: we are assuming that
3842 * f_iosize is a power of 2.
3844 int iosize = vp->v_mount->mnt_stat.f_iosize;
3845 int iomask = iosize - 1;
3850 * Align to iosize boundry, super-align to page boundry.
3853 cnt += off & iomask;
3854 off &= ~(u_quad_t)iomask;
3856 if (off & PAGE_MASK) {
3857 cnt += off & PAGE_MASK;
3858 off &= ~(u_quad_t)PAGE_MASK;
3860 lblkno = off / iosize;
3863 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
3864 VM_OBJECT_LOCK(vp->v_object);
3865 vm_object_page_clean(vp->v_object, off / PAGE_SIZE, (cnt + PAGE_MASK) / PAGE_SIZE, OBJPC_SYNC);
3866 VM_OBJECT_UNLOCK(vp->v_object);
3875 * If we have a buffer and it is marked B_DELWRI we
3876 * have to lock and write it. Otherwise the prior
3877 * write is assumed to have already been committed.
3879 * gbincore() can return invalid buffers now so we
3880 * have to check that bit as well (though B_DELWRI
3881 * should not be set if B_INVAL is set there could be
3882 * a race here since we haven't locked the buffer).
3884 if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
3885 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
3886 LK_INTERLOCK, BO_MTX(bo)) == ENOLCK) {
3888 continue; /* retry */
3890 if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
3893 bp->b_flags &= ~B_ASYNC;
3909 aft_ret = VOP_GETATTR(vp, &aft, cred);
3913 nfsm_reply(NFSX_V3WCCDATA + NFSX_V3WRITEVERF);
3914 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3916 tl = nfsm_build(u_int32_t *, NFSX_V3WRITEVERF);
3917 if (nfsver.tv_sec == 0)
3919 *tl++ = txdr_unsigned(nfsver.tv_sec);
3920 *tl = txdr_unsigned(nfsver.tv_usec);
3927 vn_finished_write(mp);
3928 VFS_UNLOCK_GIANT(vfslocked);
3933 * nfs statfs service
3936 nfsrv_statfs(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3937 struct thread *td, struct mbuf **mrq)
3939 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3940 struct sockaddr *nam = nfsd->nd_nam;
3941 caddr_t dpos = nfsd->nd_dpos;
3942 struct ucred *cred = nfsd->nd_cr;
3944 struct nfs_statfs *sfp;
3946 int error = 0, rdonly, getret = 1;
3947 int v3 = (nfsd->nd_flag & ND_NFSV3);
3948 struct mbuf *mb, *mreq;
3949 struct vnode *vp = NULL;
3953 struct statfs statfs;
3957 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3959 fhp = &nfh.fh_generic;
3961 error = nfsrv_fhtovp(fhp, 1, &vp, &vfslocked, cred, slp,
3962 nam, &rdonly, TRUE);
3964 nfsm_reply(NFSX_UNSIGNED);
3966 nfsm_srvpostop_attr(getret, &at);
3971 error = VFS_STATFS(vp->v_mount, sf, td);
3972 getret = VOP_GETATTR(vp, &at, cred);
3975 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_STATFS(v3));
3977 nfsm_srvpostop_attr(getret, &at);
3982 sfp = nfsm_build(struct nfs_statfs *, NFSX_STATFS(v3));
3984 tval = (u_quad_t)sf->f_blocks;
3985 tval *= (u_quad_t)sf->f_bsize;
3986 txdr_hyper(tval, &sfp->sf_tbytes);
3987 tval = (u_quad_t)sf->f_bfree;
3988 tval *= (u_quad_t)sf->f_bsize;
3989 txdr_hyper(tval, &sfp->sf_fbytes);
3991 * Don't send negative values for available space,
3992 * since this field is unsigned in the NFS protocol.
3993 * Otherwise, the client would see absurdly high
3994 * numbers for free space.
3996 if (sf->f_bavail < 0)
3999 tval = (u_quad_t)sf->f_bavail;
4000 tval *= (u_quad_t)sf->f_bsize;
4001 txdr_hyper(tval, &sfp->sf_abytes);
4002 sfp->sf_tfiles.nfsuquad[0] = 0;
4003 sfp->sf_tfiles.nfsuquad[1] = txdr_unsigned(sf->f_files);
4004 sfp->sf_ffiles.nfsuquad[0] = 0;
4005 sfp->sf_ffiles.nfsuquad[1] = txdr_unsigned(sf->f_ffree);
4006 sfp->sf_afiles.nfsuquad[0] = 0;
4007 sfp->sf_afiles.nfsuquad[1] = txdr_unsigned(sf->f_ffree);
4008 sfp->sf_invarsec = 0;
4010 sfp->sf_tsize = txdr_unsigned(NFS_MAXDGRAMDATA);
4011 sfp->sf_bsize = txdr_unsigned(sf->f_bsize);
4012 sfp->sf_blocks = txdr_unsigned(sf->f_blocks);
4013 sfp->sf_bfree = txdr_unsigned(sf->f_bfree);
4014 if (sf->f_bavail < 0)
4017 sfp->sf_bavail = txdr_unsigned(sf->f_bavail);
4022 VFS_UNLOCK_GIANT(vfslocked);
4027 * nfs fsinfo service
4030 nfsrv_fsinfo(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
4031 struct thread *td, struct mbuf **mrq)
4033 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
4034 struct sockaddr *nam = nfsd->nd_nam;
4035 caddr_t dpos = nfsd->nd_dpos;
4036 struct ucred *cred = nfsd->nd_cr;
4037 struct nfsv3_fsinfo *sip;
4039 int error = 0, rdonly, getret = 1, pref;
4040 struct mbuf *mb, *mreq;
4041 struct vnode *vp = NULL;
4047 int v3 = (nfsd->nd_flag & ND_NFSV3);
4050 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
4052 panic("nfsrv_fsinfo: v3 proc called on a v2 connection");
4053 fhp = &nfh.fh_generic;
4056 error = nfsrv_fhtovp(fhp, 1, &vp, &vfslocked, cred, slp,
4057 nam, &rdonly, TRUE);
4059 nfsm_reply(NFSX_UNSIGNED);
4060 nfsm_srvpostop_attr(getret, &at);
4065 /* XXX Try to make a guess on the max file size. */
4066 VFS_STATFS(vp->v_mount, &sb, td);
4067 maxfsize = (u_quad_t)0x80000000 * sb.f_bsize - 1;
4069 getret = VOP_GETATTR(vp, &at, cred);
4072 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3FSINFO);
4073 nfsm_srvpostop_attr(getret, &at);
4074 sip = nfsm_build(struct nfsv3_fsinfo *, NFSX_V3FSINFO);
4078 * There should be filesystem VFS OP(s) to get this information.
4079 * For now, assume ufs.
4081 if (slp->ns_so->so_type == SOCK_DGRAM)
4082 pref = NFS_MAXDGRAMDATA;
4085 sip->fs_rtmax = txdr_unsigned(pref);
4086 sip->fs_rtpref = txdr_unsigned(pref);
4087 sip->fs_rtmult = txdr_unsigned(NFS_FABLKSIZE);
4088 sip->fs_wtmax = txdr_unsigned(pref);
4089 sip->fs_wtpref = txdr_unsigned(pref);
4090 sip->fs_wtmult = txdr_unsigned(NFS_FABLKSIZE);
4091 sip->fs_dtpref = txdr_unsigned(pref);
4092 txdr_hyper(maxfsize, &sip->fs_maxfilesize);
4093 sip->fs_timedelta.nfsv3_sec = 0;
4094 sip->fs_timedelta.nfsv3_nsec = txdr_unsigned(1);
4095 sip->fs_properties = txdr_unsigned(NFSV3FSINFO_LINK |
4096 NFSV3FSINFO_SYMLINK | NFSV3FSINFO_HOMOGENEOUS |
4097 NFSV3FSINFO_CANSETTIME);
4101 VFS_UNLOCK_GIANT(vfslocked);
4106 * nfs pathconf service
4109 nfsrv_pathconf(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
4110 struct thread *td, struct mbuf **mrq)
4112 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
4113 struct sockaddr *nam = nfsd->nd_nam;
4114 caddr_t dpos = nfsd->nd_dpos;
4115 struct ucred *cred = nfsd->nd_cr;
4116 struct nfsv3_pathconf *pc;
4118 int error = 0, rdonly, getret = 1;
4119 register_t linkmax, namemax, chownres, notrunc;
4120 struct mbuf *mb, *mreq;
4121 struct vnode *vp = NULL;
4125 int v3 = (nfsd->nd_flag & ND_NFSV3);
4128 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
4130 panic("nfsrv_pathconf: v3 proc called on a v2 connection");
4132 fhp = &nfh.fh_generic;
4134 error = nfsrv_fhtovp(fhp, 1, &vp, &vfslocked, cred, slp,
4135 nam, &rdonly, TRUE);
4137 nfsm_reply(NFSX_UNSIGNED);
4138 nfsm_srvpostop_attr(getret, &at);
4142 error = VOP_PATHCONF(vp, _PC_LINK_MAX, &linkmax);
4144 error = VOP_PATHCONF(vp, _PC_NAME_MAX, &namemax);
4146 error = VOP_PATHCONF(vp, _PC_CHOWN_RESTRICTED, &chownres);
4148 error = VOP_PATHCONF(vp, _PC_NO_TRUNC, ¬runc);
4149 getret = VOP_GETATTR(vp, &at, cred);
4152 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3PATHCONF);
4153 nfsm_srvpostop_attr(getret, &at);
4158 pc = nfsm_build(struct nfsv3_pathconf *, NFSX_V3PATHCONF);
4160 pc->pc_linkmax = txdr_unsigned(linkmax);
4161 pc->pc_namemax = txdr_unsigned(namemax);
4162 pc->pc_notrunc = txdr_unsigned(notrunc);
4163 pc->pc_chownrestricted = txdr_unsigned(chownres);
4166 * These should probably be supported by VOP_PATHCONF(), but
4167 * until msdosfs is exportable (why would you want to?), the
4168 * Unix defaults should be ok.
4170 pc->pc_caseinsensitive = nfsrv_nfs_false;
4171 pc->pc_casepreserving = nfsrv_nfs_true;
4175 VFS_UNLOCK_GIANT(vfslocked);
4180 * Null operation, used by clients to ping server
4184 nfsrv_null(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
4185 struct thread *td, struct mbuf **mrq)
4187 struct mbuf *mrep = nfsd->nd_mrep;
4189 int error = NFSERR_RETVOID;
4190 struct mbuf *mb, *mreq;
4192 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
4199 * No operation, used for obsolete procedures
4203 nfsrv_noop(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
4204 struct thread *td, struct mbuf **mrq)
4206 struct mbuf *mrep = nfsd->nd_mrep;
4209 struct mbuf *mb, *mreq;
4211 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
4212 if (nfsd->nd_repstat)
4213 error = nfsd->nd_repstat;
4215 error = EPROCUNAVAIL;
4223 * Perform access checking for vnodes obtained from file handles that would
4224 * refer to files already opened by a Unix client. You cannot just use
4225 * vn_writechk() and VOP_ACCESS() for two reasons.
4226 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
4228 * 2 - The owner is to be given access irrespective of mode bits for some
4229 * operations, so that processes that chmod after opening a file don't
4230 * break. I don't like this because it opens a security hole, but since
4231 * the nfs server opens a security hole the size of a barn door anyhow,
4234 * The exception to rule 2 is EPERM. If a file is IMMUTABLE, VOP_ACCESS()
4235 * will return EPERM instead of EACCESS. EPERM is always an error.
4238 nfsrv_access(struct vnode *vp, int flags, struct ucred *cred,
4239 int rdonly, struct thread *td, int override)
4244 VFS_ASSERT_GIANT(vp->v_mount);
4246 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
4248 if (flags & VWRITE) {
4249 /* Just vn_writechk() changed to check rdonly */
4251 * Disallow write attempts on read-only filesystems;
4252 * unless the file is a socket or a block or character
4253 * device resident on the filesystem.
4255 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
4256 switch (vp->v_type) {
4266 * If there's shared text associated with
4267 * the inode, we can't allow writing.
4269 if (vp->v_vflag & VV_TEXT)
4273 error = VOP_GETATTR(vp, &vattr, cred);
4276 error = VOP_ACCESS(vp, flags, cred, td);
4278 * Allow certain operations for the owner (reads and writes
4279 * on files that are already open).
4281 if (override && error == EACCES && cred->cr_uid == vattr.va_uid)