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 * 3. All advertising materials mentioning features or use of this software
17 * must display the following acknowledgement:
18 * This product includes software developed by the University of
19 * California, Berkeley and its contributors.
20 * 4. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * @(#)nfs_subs.c 8.8 (Berkeley) 5/22/95
37 * $FreeBSD: /repoman/r/ncvs/src/sys/nfsclient/nfs_subs.c,v 1.128 2004/04/14 23:23:55 peadar Exp $
38 * $DragonFly: src/sys/vfs/nfs/nfs_subs.c,v 1.35 2006/03/27 16:18:39 dillon Exp $
42 * These functions support the macros and help fiddle mbuf chains for
43 * the nfs op functions. They do things like create the rpc header and
44 * copy data between mbuf chains and uio lists.
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/kernel.h>
51 #include <sys/mount.h>
52 #include <sys/vnode.h>
53 #include <sys/nlookup.h>
54 #include <sys/namei.h>
56 #include <sys/socket.h>
58 #include <sys/malloc.h>
59 #include <sys/sysent.h>
60 #include <sys/syscall.h>
64 #include <vm/vm_object.h>
65 #include <vm/vm_extern.h>
66 #include <vm/vm_zone.h>
76 #include "nfsm_subs.h"
79 #include <netinet/in.h>
82 * Data items converted to xdr at startup, since they are constant
83 * This is kinda hokey, but may save a little time doing byte swaps
85 u_int32_t nfs_xdrneg1;
86 u_int32_t rpc_call, rpc_vers, rpc_reply, rpc_msgdenied, rpc_autherr,
87 rpc_mismatch, rpc_auth_unix, rpc_msgaccepted,
89 u_int32_t nfs_prog, nfs_true, nfs_false;
91 /* And other global data */
92 static u_int32_t nfs_xid = 0;
93 static enum vtype nv2tov_type[8]= {
94 VNON, VREG, VDIR, VBLK, VCHR, VLNK, VNON, VNON
96 enum vtype nv3tov_type[8]= {
97 VNON, VREG, VDIR, VBLK, VCHR, VLNK, VSOCK, VFIFO
101 int nfs_pbuf_freecnt = -1; /* start out unlimited */
103 struct nfs_reqq nfs_reqq;
104 struct nfssvc_sockhead nfssvc_sockhead;
105 int nfssvc_sockhead_flag;
106 struct nfsd_head nfsd_head;
108 struct nfs_bufq nfs_bufq;
109 struct nqfhhashhead *nqfhhashtbl;
112 static int nfs_prev_nfssvc_sy_narg;
113 static sy_call_t *nfs_prev_nfssvc_sy_call;
118 * Mapping of old NFS Version 2 RPC numbers to generic numbers.
120 int nfsv3_procid[NFS_NPROCS] = {
149 #endif /* NFS_NOSERVER */
151 * and the reverse mapping from generic to Version 2 procedure numbers
153 int nfsv2_procid[NFS_NPROCS] = {
184 * Maps errno values to nfs error numbers.
185 * Use NFSERR_IO as the catch all for ones not specifically defined in
188 static u_char nfsrv_v2errmap[ELAST] = {
189 NFSERR_PERM, NFSERR_NOENT, NFSERR_IO, NFSERR_IO, NFSERR_IO,
190 NFSERR_NXIO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
191 NFSERR_IO, NFSERR_IO, NFSERR_ACCES, NFSERR_IO, NFSERR_IO,
192 NFSERR_IO, NFSERR_EXIST, NFSERR_IO, NFSERR_NODEV, NFSERR_NOTDIR,
193 NFSERR_ISDIR, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
194 NFSERR_IO, NFSERR_FBIG, NFSERR_NOSPC, NFSERR_IO, NFSERR_ROFS,
195 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
196 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
197 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
198 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
199 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
200 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
201 NFSERR_IO, NFSERR_IO, NFSERR_NAMETOL, NFSERR_IO, NFSERR_IO,
202 NFSERR_NOTEMPTY, NFSERR_IO, NFSERR_IO, NFSERR_DQUOT, NFSERR_STALE,
203 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
204 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
205 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
206 NFSERR_IO /* << Last is 86 */
210 * Maps errno values to nfs error numbers.
211 * Although it is not obvious whether or not NFS clients really care if
212 * a returned error value is in the specified list for the procedure, the
213 * safest thing to do is filter them appropriately. For Version 2, the
214 * X/Open XNFS document is the only specification that defines error values
215 * for each RPC (The RFC simply lists all possible error values for all RPCs),
216 * so I have decided to not do this for Version 2.
217 * The first entry is the default error return and the rest are the valid
218 * errors for that RPC in increasing numeric order.
220 static short nfsv3err_null[] = {
225 static short nfsv3err_getattr[] = {
234 static short nfsv3err_setattr[] = {
250 static short nfsv3err_lookup[] = {
263 static short nfsv3err_access[] = {
272 static short nfsv3err_readlink[] = {
284 static short nfsv3err_read[] = {
296 static short nfsv3err_write[] = {
311 static short nfsv3err_create[] = {
328 static short nfsv3err_mkdir[] = {
345 static short nfsv3err_symlink[] = {
362 static short nfsv3err_mknod[] = {
380 static short nfsv3err_remove[] = {
394 static short nfsv3err_rmdir[] = {
412 static short nfsv3err_rename[] = {
435 static short nfsv3err_link[] = {
455 static short nfsv3err_readdir[] = {
468 static short nfsv3err_readdirplus[] = {
482 static short nfsv3err_fsstat[] = {
491 static short nfsv3err_fsinfo[] = {
499 static short nfsv3err_pathconf[] = {
507 static short nfsv3err_commit[] = {
516 static short *nfsrv_v3errmap[] = {
534 nfsv3err_readdirplus,
541 #endif /* NFS_NOSERVER */
543 extern struct nfsrtt nfsrtt;
544 extern struct nfsstats nfsstats;
545 extern nfstype nfsv2_type[9];
546 extern nfstype nfsv3_type[9];
547 extern struct nfsnodehashhead *nfsnodehashtbl;
548 extern u_long nfsnodehash;
551 extern int nfssvc(struct proc *, struct nfssvc_args *, int *);
553 LIST_HEAD(nfsnodehashhead, nfsnode);
556 * This needs to return a monotonically increasing or close to monotonically
557 * increasing result, otherwise the write gathering queues won't work
566 return ((u_quad_t)tv.tv_sec * 1000000 + (u_quad_t)tv.tv_usec);
570 * Create the header for an rpc request packet
571 * The hsiz is the size of the rest of the nfs request header.
572 * (just used to decide if a cluster is a good idea)
575 nfsm_reqh(struct vnode *vp, u_long procid, int hsiz, caddr_t *bposp)
580 mb = m_getl(hsiz, MB_WAIT, MT_DATA, 0, NULL);
582 bpos = mtod(mb, caddr_t);
584 /* Finally, return values */
590 * Build the RPC header and fill in the authorization info.
591 * The authorization string argument is only used when the credentials
592 * come from outside of the kernel.
593 * Returns the head of the mbuf list.
596 nfsm_rpchead(struct ucred *cr, int nmflag, int procid, int auth_type,
597 int auth_len, char *auth_str, int verf_len, char *verf_str,
598 struct mbuf *mrest, int mrest_len, struct mbuf **mbp,
605 struct mbuf *mreq, *mb2;
606 int siz, grpsiz, authsiz, dsiz;
608 authsiz = nfsm_rndup(auth_len);
609 dsiz = authsiz + 10 * NFSX_UNSIGNED;
610 mb = m_getl(dsiz, MB_WAIT, MT_DATA, M_PKTHDR, NULL);
611 if (dsiz < MINCLSIZE) {
615 MH_ALIGN(mb, 8 * NFSX_UNSIGNED);
617 mb->m_len = mb->m_pkthdr.len = 0;
619 bpos = mtod(mb, caddr_t);
622 * First the RPC header.
624 nfsm_build(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
626 /* Get a pretty random xid to start with */
630 * Skip zero xid if it should ever happen.
635 *tl++ = *xidp = txdr_unsigned(nfs_xid);
638 *tl++ = txdr_unsigned(NFS_PROG);
639 if (nmflag & NFSMNT_NFSV3)
640 *tl++ = txdr_unsigned(NFS_VER3);
642 *tl++ = txdr_unsigned(NFS_VER2);
643 if (nmflag & NFSMNT_NFSV3)
644 *tl++ = txdr_unsigned(procid);
646 *tl++ = txdr_unsigned(nfsv2_procid[procid]);
649 * And then the authorization cred.
651 *tl++ = txdr_unsigned(auth_type);
652 *tl = txdr_unsigned(authsiz);
655 nfsm_build(tl, u_int32_t *, auth_len);
656 *tl++ = 0; /* stamp ?? */
657 *tl++ = 0; /* NULL hostname */
658 *tl++ = txdr_unsigned(cr->cr_uid);
659 *tl++ = txdr_unsigned(cr->cr_groups[0]);
660 grpsiz = (auth_len >> 2) - 5;
661 *tl++ = txdr_unsigned(grpsiz);
662 for (i = 1; i <= grpsiz; i++)
663 *tl++ = txdr_unsigned(cr->cr_groups[i]);
668 if (M_TRAILINGSPACE(mb) == 0) {
669 mb2 = m_getl(siz, MB_WAIT, MT_DATA, 0, NULL);
673 bpos = mtod(mb, caddr_t);
675 i = min(siz, M_TRAILINGSPACE(mb));
676 bcopy(auth_str, bpos, i);
682 if ((siz = (nfsm_rndup(auth_len) - auth_len)) > 0) {
683 for (i = 0; i < siz; i++)
691 * And the verifier...
693 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
695 *tl++ = txdr_unsigned(RPCAUTH_KERB4);
696 *tl = txdr_unsigned(verf_len);
699 if (M_TRAILINGSPACE(mb) == 0) {
700 mb2 = m_getl(siz, MB_WAIT, MT_DATA, 0, NULL);
704 bpos = mtod(mb, caddr_t);
706 i = min(siz, M_TRAILINGSPACE(mb));
707 bcopy(verf_str, bpos, i);
713 if ((siz = (nfsm_rndup(verf_len) - verf_len)) > 0) {
714 for (i = 0; i < siz; i++)
719 *tl++ = txdr_unsigned(RPCAUTH_NULL);
723 mreq->m_pkthdr.len = authsiz + 10 * NFSX_UNSIGNED + mrest_len;
724 mreq->m_pkthdr.rcvif = (struct ifnet *)0;
730 * copies mbuf chain to the uio scatter/gather list
733 nfsm_mbuftouio(struct mbuf **mrep, struct uio *uiop, int siz, caddr_t *dpos)
735 char *mbufcp, *uiocp;
743 len = mtod(mp, caddr_t)+mp->m_len-mbufcp;
744 rem = nfsm_rndup(siz)-siz;
746 if (uiop->uio_iovcnt <= 0 || uiop->uio_iov == NULL)
748 left = uiop->uio_iov->iov_len;
749 uiocp = uiop->uio_iov->iov_base;
758 mbufcp = mtod(mp, caddr_t);
761 xfer = (left > len) ? len : left;
764 if (uiop->uio_iov->iov_op != NULL)
765 (*(uiop->uio_iov->iov_op))
766 (mbufcp, uiocp, xfer);
769 if (uiop->uio_segflg == UIO_SYSSPACE)
770 bcopy(mbufcp, uiocp, xfer);
772 copyout(mbufcp, uiocp, xfer);
777 uiop->uio_offset += xfer;
778 uiop->uio_resid -= xfer;
780 if (uiop->uio_iov->iov_len <= siz) {
784 uiop->uio_iov->iov_base += uiosiz;
785 uiop->uio_iov->iov_len -= uiosiz;
793 error = nfs_adv(mrep, dpos, rem, len);
801 * copies a uio scatter/gather list to an mbuf chain.
802 * NOTE: can ony handle iovcnt == 1
805 nfsm_uiotombuf(struct uio *uiop, struct mbuf **mq, int siz, caddr_t *bpos)
808 struct mbuf *mp, *mp2;
809 int xfer, left, mlen;
811 boolean_t getcluster;
815 if (uiop->uio_iovcnt != 1)
816 panic("nfsm_uiotombuf: iovcnt != 1");
819 if (siz >= MINCLSIZE)
823 rem = nfsm_rndup(siz) - siz;
826 left = uiop->uio_iov->iov_len;
827 uiocp = uiop->uio_iov->iov_base;
832 mlen = M_TRAILINGSPACE(mp);
835 mp = m_getcl(MB_WAIT, MT_DATA, 0);
837 mp = m_get(MB_WAIT, MT_DATA);
841 mlen = M_TRAILINGSPACE(mp);
843 xfer = (left > mlen) ? mlen : left;
846 if (uiop->uio_iov->iov_op != NULL)
847 (*(uiop->uio_iov->iov_op))
848 (uiocp, mtod(mp, caddr_t)+mp->m_len, xfer);
851 if (uiop->uio_segflg == UIO_SYSSPACE)
852 bcopy(uiocp, mtod(mp, caddr_t)+mp->m_len, xfer);
854 copyin(uiocp, mtod(mp, caddr_t)+mp->m_len, xfer);
858 uiop->uio_offset += xfer;
859 uiop->uio_resid -= xfer;
861 uiop->uio_iov->iov_base += uiosiz;
862 uiop->uio_iov->iov_len -= uiosiz;
866 if (rem > M_TRAILINGSPACE(mp)) {
867 MGET(mp, MB_WAIT, MT_DATA);
871 cp = mtod(mp, caddr_t)+mp->m_len;
872 for (left = 0; left < rem; left++)
877 *bpos = mtod(mp, caddr_t)+mp->m_len;
883 * Help break down an mbuf chain by setting the first siz bytes contiguous
884 * pointed to by returned val.
885 * This is used by the macros nfsm_dissect and nfsm_dissecton for tough
886 * cases. (The macros use the vars. dpos and dpos2)
889 nfsm_disct(struct mbuf **mdp, caddr_t *dposp, int siz, int left, caddr_t *cp2)
891 struct mbuf *mp, *mp2;
897 *mdp = mp = mp->m_next;
901 *dposp = mtod(mp, caddr_t);
906 } else if (mp->m_next == NULL) {
908 } else if (siz > MHLEN) {
909 panic("nfs S too big");
911 MGET(mp2, MB_WAIT, MT_DATA);
912 mp2->m_next = mp->m_next;
916 *cp2 = p = mtod(mp, caddr_t);
917 bcopy(*dposp, p, left); /* Copy what was left */
921 /* Loop around copying up the siz2 bytes */
925 xfer = (siz2 > mp2->m_len) ? mp2->m_len : siz2;
927 bcopy(mtod(mp2, caddr_t), p, xfer);
938 *dposp = mtod(mp2, caddr_t);
944 * Advance the position in the mbuf chain.
947 nfs_adv(struct mbuf **mdp, caddr_t *dposp, int offs, int left)
962 *dposp = mtod(m, caddr_t)+offs;
967 * Copy a string into mbufs for the hard cases...
970 nfsm_strtmbuf(struct mbuf **mb, char **bpos, const char *cp, long siz)
972 struct mbuf *m1 = NULL, *m2;
973 long left, xfer, len, tlen;
979 left = M_TRAILINGSPACE(m2);
981 tl = ((u_int32_t *)(*bpos));
982 *tl++ = txdr_unsigned(siz);
984 left -= NFSX_UNSIGNED;
985 m2->m_len += NFSX_UNSIGNED;
987 bcopy(cp, (caddr_t) tl, left);
994 /* Loop around adding mbufs */
998 m1 = m_getl(siz, MB_WAIT, MT_DATA, 0, &msize);
1002 tl = mtod(m1, u_int32_t *);
1005 *tl++ = txdr_unsigned(siz);
1006 m1->m_len -= NFSX_UNSIGNED;
1007 tlen = NFSX_UNSIGNED;
1010 if (siz < m1->m_len) {
1011 len = nfsm_rndup(siz);
1014 *(tl+(xfer>>2)) = 0;
1016 xfer = len = m1->m_len;
1018 bcopy(cp, (caddr_t) tl, xfer);
1019 m1->m_len = len+tlen;
1024 *bpos = mtod(m1, caddr_t)+m1->m_len;
1029 * Called once to initialize data structures...
1032 nfs_init(struct vfsconf *vfsp)
1036 callout_init(&nfs_timer_handle);
1037 nfsmount_zone = zinit("NFSMOUNT", sizeof(struct nfsmount), 0, 0, 1);
1039 nfs_mount_type = vfsp->vfc_typenum;
1041 rpc_vers = txdr_unsigned(RPC_VER2);
1042 rpc_call = txdr_unsigned(RPC_CALL);
1043 rpc_reply = txdr_unsigned(RPC_REPLY);
1044 rpc_msgdenied = txdr_unsigned(RPC_MSGDENIED);
1045 rpc_msgaccepted = txdr_unsigned(RPC_MSGACCEPTED);
1046 rpc_mismatch = txdr_unsigned(RPC_MISMATCH);
1047 rpc_autherr = txdr_unsigned(RPC_AUTHERR);
1048 rpc_auth_unix = txdr_unsigned(RPCAUTH_UNIX);
1049 rpc_auth_kerb = txdr_unsigned(RPCAUTH_KERB4);
1050 nfs_prog = txdr_unsigned(NFS_PROG);
1051 nfs_true = txdr_unsigned(TRUE);
1052 nfs_false = txdr_unsigned(FALSE);
1053 nfs_xdrneg1 = txdr_unsigned(-1);
1054 nfs_ticks = (hz * NFS_TICKINTVL + 500) / 1000;
1057 /* Ensure async daemons disabled */
1058 for (i = 0; i < NFS_MAXASYNCDAEMON; i++) {
1059 nfs_iodwant[i] = NULL;
1060 nfs_iodmount[i] = (struct nfsmount *)0;
1062 nfs_nhinit(); /* Init the nfsnode table */
1063 #ifndef NFS_NOSERVER
1064 nfsrv_init(0); /* Init server data structures */
1065 nfsrv_initcache(); /* Init the server request cache */
1069 * Initialize reply list and start timer
1071 TAILQ_INIT(&nfs_reqq);
1075 nfs_prev_nfssvc_sy_narg = sysent[SYS_nfssvc].sy_narg;
1076 sysent[SYS_nfssvc].sy_narg = 2;
1077 nfs_prev_nfssvc_sy_call = sysent[SYS_nfssvc].sy_call;
1078 sysent[SYS_nfssvc].sy_call = (sy_call_t *)nfssvc;
1080 nfs_pbuf_freecnt = nswbuf / 2 + 1;
1086 nfs_uninit(struct vfsconf *vfsp)
1088 callout_stop(&nfs_timer_handle);
1089 nfs_mount_type = -1;
1090 sysent[SYS_nfssvc].sy_narg = nfs_prev_nfssvc_sy_narg;
1091 sysent[SYS_nfssvc].sy_call = nfs_prev_nfssvc_sy_call;
1096 * Attribute cache routines.
1097 * nfs_loadattrcache() - loads or updates the cache contents from attributes
1098 * that are on the mbuf list
1099 * nfs_getattrcache() - returns valid attributes if found in cache, returns
1104 * Load the attribute cache (that lives in the nfsnode entry) with
1105 * the values on the mbuf list. Load *vaper with the attributes. vaper
1108 * As a side effect n_mtime, which we use to determine if the file was
1109 * modified by some other host, is set to the attribute timestamp and
1110 * NRMODIFIED is set if the two values differ.
1112 * WARNING: the mtime loaded into vaper does not necessarily represent
1113 * n_mtime or n_attr.mtime due to NACC and NUPD.
1116 nfs_loadattrcache(struct vnode **vpp, struct mbuf **mdp, caddr_t *dposp,
1117 struct vattr *vaper, int lattr_flags)
1119 struct vnode *vp = *vpp;
1121 struct nfs_fattr *fp;
1130 struct timespec mtime;
1131 int v3 = NFS_ISV3(vp);
1134 t1 = (mtod(md, caddr_t) + md->m_len) - *dposp;
1135 if ((error = nfsm_disct(mdp, dposp, NFSX_FATTR(v3), t1, &cp2)) != 0)
1137 fp = (struct nfs_fattr *)cp2;
1139 vtyp = nfsv3tov_type(fp->fa_type);
1140 vmode = fxdr_unsigned(u_short, fp->fa_mode);
1141 rdev = makeudev(fxdr_unsigned(int, fp->fa3_rdev.specdata1),
1142 fxdr_unsigned(int, fp->fa3_rdev.specdata2));
1143 fxdr_nfsv3time(&fp->fa3_mtime, &mtime);
1145 vtyp = nfsv2tov_type(fp->fa_type);
1146 vmode = fxdr_unsigned(u_short, fp->fa_mode);
1150 * The duplicate information returned in fa_type and fa_mode
1151 * is an ambiguity in the NFS version 2 protocol.
1153 * VREG should be taken literally as a regular file. If a
1154 * server intents to return some type information differently
1155 * in the upper bits of the mode field (e.g. for sockets, or
1156 * FIFOs), NFSv2 mandates fa_type to be VNON. Anyway, we
1157 * leave the examination of the mode bits even in the VREG
1158 * case to avoid breakage for bogus servers, but we make sure
1159 * that there are actually type bits set in the upper part of
1160 * fa_mode (and failing that, trust the va_type field).
1162 * NFSv3 cleared the issue, and requires fa_mode to not
1163 * contain any type information (while also introduing sockets
1164 * and FIFOs for fa_type).
1166 if (vtyp == VNON || (vtyp == VREG && (vmode & S_IFMT) != 0))
1167 vtyp = IFTOVT(vmode);
1168 rdev = fxdr_unsigned(int32_t, fp->fa2_rdev);
1169 fxdr_nfsv2time(&fp->fa2_mtime, &mtime);
1172 * Really ugly NFSv2 kludge.
1174 if (vtyp == VCHR && rdev == (udev_t)0xffffffff)
1179 * If v_type == VNON it is a new node, so fill in the v_type,
1180 * n_mtime fields. Check to see if it represents a special
1181 * device, and if so, check for a possible alias. Once the
1182 * correct vnode has been obtained, fill in the rest of the
1186 if (vp->v_type != vtyp) {
1188 if (vp->v_type == VFIFO) {
1189 vp->v_ops = &vp->v_mount->mnt_vn_fifo_ops;
1190 } else if (vp->v_type == VCHR || vp->v_type == VBLK) {
1191 vp->v_ops = &vp->v_mount->mnt_vn_spec_ops;
1192 addaliasu(vp, rdev);
1194 vp->v_ops = &vp->v_mount->mnt_vn_use_ops;
1196 np->n_mtime = mtime.tv_sec;
1197 } else if (np->n_mtime != mtime.tv_sec) {
1199 * If we haven't modified the file locally and the server
1200 * timestamp does not match, then the server probably
1201 * modified the file. We must flag this condition so
1202 * the proper syncnronization can be done. We do not
1203 * try to synchronize the state here because that
1204 * could lead to an endless recursion.
1206 * XXX loadattrcache can be set during the reply to a write,
1207 * before the write timestamp is properly processed. To
1208 * avoid unconditionally setting the rmodified bit (which
1209 * has the effect of flushing the cache), we only do this
1210 * check if the lmodified bit is not set.
1212 np->n_mtime = mtime.tv_sec;
1213 if ((lattr_flags & NFS_LATTR_NOMTIMECHECK) == 0)
1214 np->n_flag |= NRMODIFIED;
1217 vap->va_type = vtyp;
1218 vap->va_mode = (vmode & 07777);
1219 vap->va_rdev = rdev;
1220 vap->va_mtime = mtime;
1221 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
1223 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
1224 vap->va_uid = fxdr_unsigned(uid_t, fp->fa_uid);
1225 vap->va_gid = fxdr_unsigned(gid_t, fp->fa_gid);
1226 vap->va_size = fxdr_hyper(&fp->fa3_size);
1227 vap->va_blocksize = NFS_FABLKSIZE;
1228 vap->va_bytes = fxdr_hyper(&fp->fa3_used);
1229 vap->va_fileid = fxdr_unsigned(int32_t,
1230 fp->fa3_fileid.nfsuquad[1]);
1231 fxdr_nfsv3time(&fp->fa3_atime, &vap->va_atime);
1232 fxdr_nfsv3time(&fp->fa3_ctime, &vap->va_ctime);
1234 vap->va_filerev = 0;
1236 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
1237 vap->va_uid = fxdr_unsigned(uid_t, fp->fa_uid);
1238 vap->va_gid = fxdr_unsigned(gid_t, fp->fa_gid);
1239 vap->va_size = fxdr_unsigned(u_int32_t, fp->fa2_size);
1240 vap->va_blocksize = fxdr_unsigned(int32_t, fp->fa2_blocksize);
1241 vap->va_bytes = (u_quad_t)fxdr_unsigned(int32_t, fp->fa2_blocks)
1243 vap->va_fileid = fxdr_unsigned(int32_t, fp->fa2_fileid);
1244 fxdr_nfsv2time(&fp->fa2_atime, &vap->va_atime);
1246 vap->va_ctime.tv_sec = fxdr_unsigned(u_int32_t,
1247 fp->fa2_ctime.nfsv2_sec);
1248 vap->va_ctime.tv_nsec = 0;
1249 vap->va_gen = fxdr_unsigned(u_int32_t,fp->fa2_ctime.nfsv2_usec);
1250 vap->va_filerev = 0;
1252 np->n_attrstamp = time_second;
1253 if (vap->va_size != np->n_size) {
1254 if (vap->va_type == VREG) {
1255 if ((lattr_flags & NFS_LATTR_NOSHRINK) &&
1256 vap->va_size < np->n_size) {
1258 * We've been told not to shrink the file;
1259 * zero np->n_attrstamp to indicate that
1260 * the attributes are stale.
1262 * This occurs primarily due to recursive
1263 * NFS ops that are executed during periods
1264 * where we cannot safely reduce the size of
1267 * Additionally, write rpcs are broken down
1268 * into buffers and np->n_size is
1269 * pre-extended. Setting NRMODIFIED here
1270 * can result in n_size getting reset to a
1271 * lower value, which is NOT what we want.
1272 * XXX this needs to be cleaned up a lot
1275 vap->va_size = np->n_size;
1276 np->n_attrstamp = 0;
1277 if ((np->n_flag & NLMODIFIED) == 0)
1278 np->n_flag |= NRMODIFIED;
1279 } else if (np->n_flag & NLMODIFIED) {
1281 * We've modified the file: Use the larger
1282 * of our size, and the server's size. At
1283 * this point the cache coherency is all
1284 * shot to hell. To try to handle multiple
1285 * clients appending to the file at the same
1286 * time mark that the server has changed
1287 * the file if the server's notion of the
1288 * file size is larger then our notion.
1290 * XXX this needs work.
1292 if (vap->va_size < np->n_size) {
1293 vap->va_size = np->n_size;
1295 np->n_size = vap->va_size;
1296 np->n_flag |= NRMODIFIED;
1300 * Someone changed the file's size on the
1301 * server and there are no local changes
1302 * to get in the way, set the size and mark
1305 np->n_size = vap->va_size;
1306 np->n_flag |= NRMODIFIED;
1308 vnode_pager_setsize(vp, np->n_size);
1310 np->n_size = vap->va_size;
1313 if (vaper != NULL) {
1314 bcopy((caddr_t)vap, (caddr_t)vaper, sizeof(*vap));
1315 if (np->n_flag & NCHG) {
1316 if (np->n_flag & NACC)
1317 vaper->va_atime = np->n_atim;
1318 if (np->n_flag & NUPD)
1319 vaper->va_mtime = np->n_mtim;
1326 #include <sys/sysctl.h>
1327 SYSCTL_DECL(_vfs_nfs);
1328 static int nfs_acdebug;
1329 SYSCTL_INT(_vfs_nfs, OID_AUTO, acdebug, CTLFLAG_RW, &nfs_acdebug, 0, "");
1333 * Check the time stamp
1334 * If the cache is valid, copy contents to *vap and return 0
1335 * otherwise return an error
1338 nfs_getattrcache(struct vnode *vp, struct vattr *vaper)
1342 struct nfsmount *nmp;
1347 nmp = VFSTONFS(vp->v_mount);
1350 * Dynamic timeout based on how recently the file was modified.
1351 * n_mtime is always valid.
1353 timeo = (get_approximate_time_t() - np->n_mtime) / 60;
1357 printf("nfs_getattrcache: initial timeo = %d\n", timeo);
1360 if (vap->va_type == VDIR) {
1361 if ((np->n_flag & NLMODIFIED) || timeo < nmp->nm_acdirmin)
1362 timeo = nmp->nm_acdirmin;
1363 else if (timeo > nmp->nm_acdirmax)
1364 timeo = nmp->nm_acdirmax;
1366 if ((np->n_flag & NLMODIFIED) || timeo < nmp->nm_acregmin)
1367 timeo = nmp->nm_acregmin;
1368 else if (timeo > nmp->nm_acregmax)
1369 timeo = nmp->nm_acregmax;
1373 if (nfs_acdebug > 2)
1374 printf("acregmin %d; acregmax %d; acdirmin %d; acdirmax %d\n",
1375 nmp->nm_acregmin, nmp->nm_acregmax,
1376 nmp->nm_acdirmin, nmp->nm_acdirmax);
1379 printf("nfs_getattrcache: age = %d; final timeo = %d\n",
1380 (int)(time_second - np->n_attrstamp), timeo);
1383 if (np->n_attrstamp == 0 || (time_second - np->n_attrstamp) >= timeo) {
1384 nfsstats.attrcache_misses++;
1387 nfsstats.attrcache_hits++;
1390 * Our attribute cache can be stale due to modifications made on
1391 * this host. XXX this is a bad hack. We need a more deterministic
1392 * means of finding out which np fields are valid verses attr cache
1393 * fields. We really should update the vattr info on the fly when
1394 * making local changes.
1396 if (vap->va_size != np->n_size) {
1397 if (vap->va_type == VREG) {
1398 if (np->n_flag & NLMODIFIED) {
1399 if (vap->va_size < np->n_size)
1400 vap->va_size = np->n_size;
1402 np->n_size = vap->va_size;
1404 np->n_size = vap->va_size;
1406 vnode_pager_setsize(vp, np->n_size);
1408 np->n_size = vap->va_size;
1411 bcopy((caddr_t)vap, (caddr_t)vaper, sizeof(struct vattr));
1412 if (np->n_flag & NCHG) {
1413 if (np->n_flag & NACC)
1414 vaper->va_atime = np->n_atim;
1415 if (np->n_flag & NUPD)
1416 vaper->va_mtime = np->n_mtim;
1421 #ifndef NFS_NOSERVER
1424 * Set up nameidata for a lookup() call and do it.
1426 * If pubflag is set, this call is done for a lookup operation on the
1427 * public filehandle. In that case we allow crossing mountpoints and
1428 * absolute pathnames. However, the caller is expected to check that
1429 * the lookup result is within the public fs, and deny access if
1432 * dirp may be set whether an error is returned or not, and must be
1433 * released by the caller.
1435 * On return nd->nl_ncp usually points to the target ncp, which may represent
1438 * NOTE: the caller must call nlookup_done(nd) unconditionally on return
1442 nfs_namei(struct nlookupdata *nd, struct ucred *cred, int nameiop,
1443 struct vnode **dvpp, struct vnode **vpp,
1444 fhandle_t *fhp, int len,
1445 struct nfssvc_sock *slp, struct sockaddr *nam, struct mbuf **mdp,
1446 caddr_t *dposp, struct vnode **dirpp, struct thread *td,
1447 int kerbflag, int pubflag)
1452 char *fromcp, *tocp, *cp;
1454 struct namecache *ncp;
1458 namebuf = zalloc(namei_zone);
1463 * Copy the name from the mbuf list to namebuf.
1468 rem = mtod(md, caddr_t) + md->m_len - fromcp;
1469 for (i = 0; i < len; i++) {
1476 fromcp = mtod(md, caddr_t);
1479 if (*fromcp == '\0' || (!pubflag && *fromcp == '/')) {
1483 *tocp++ = *fromcp++;
1489 len = nfsm_rndup(len)-len;
1493 else if ((error = nfs_adv(mdp, dposp, len, rem)) != 0)
1498 * Extract and set starting directory. The returned dp is refd
1501 error = nfsrv_fhtovp(fhp, FALSE, &dp, cred, slp,
1502 nam, &rdonly, kerbflag, pubflag);
1505 if (dp->v_type != VDIR) {
1512 * Set return directory. Reference to dp is implicitly transfered
1513 * to the returned pointer. This must be set before we potentially
1520 * Oh joy. For WebNFS, handle those pesky '%' escapes,
1521 * and the 'native path' indicator.
1523 cp = zalloc(namei_zone);
1526 if ((unsigned char)*fromcp >= WEBNFS_SPECCHAR_START) {
1527 switch ((unsigned char)*fromcp) {
1528 case WEBNFS_NATIVE_CHAR:
1530 * 'Native' path for us is the same
1531 * as a path according to the NFS spec,
1532 * just skip the escape char.
1537 * More may be added in the future, range 0x80-0xff
1541 zfree(namei_zone, cp);
1546 * Translate the '%' escapes, URL-style.
1548 while (*fromcp != '\0') {
1549 if (*fromcp == WEBNFS_ESC_CHAR) {
1550 if (fromcp[1] != '\0' && fromcp[2] != '\0') {
1552 *tocp++ = HEXSTRTOI(fromcp);
1557 zfree(namei_zone, cp);
1561 *tocp++ = *fromcp++;
1564 zfree(namei_zone, namebuf);
1569 * Setup for search. We need to get a start directory from dp. Note
1570 * that dp is ref'd, but we no longer 'own' the ref (*dirpp owns it).
1573 flags |= NLC_NFS_NOSOFTLINKTRAV;
1574 flags |= NLC_NOCROSSMOUNT;
1577 flags |= NLC_NFS_RDONLY;
1578 if (nameiop == NAMEI_CREATE || nameiop == NAMEI_RENAME)
1579 flags |= NLC_CREATE;
1582 * We need a starting ncp from the directory vnode dp. dp must not
1583 * be locked. The returned ncp will be refd but not locked.
1585 * If no suitable ncp is found we instruct cache_fromdvp() to create
1586 * one. If this fails the directory has probably been removed while
1587 * the target was chdir'd into it and any further lookup will fail.
1589 if ((ncp = cache_fromdvp(dp, cred, 1)) == NULL) {
1593 nlookup_init_raw(nd, namebuf, UIO_SYSSPACE, flags, cred, ncp);
1597 * Ok, do the lookup.
1599 error = nlookup(nd);
1602 * If no error occured return the requested dvpp and vpp. If
1603 * NLC_CREATE was specified nd->nl_ncp may represent a negative
1604 * cache hit in which case we do not attempt to obtain the vp.
1609 if (ncp->nc_parent &&
1610 ncp->nc_parent->nc_mount == ncp->nc_mount) {
1611 error = cache_vget(ncp->nc_parent, nd->nl_cred,
1612 LK_EXCLUSIVE, dvpp);
1617 if (vpp && ncp->nc_vp) {
1618 error = cache_vget(ncp, nd->nl_cred, LK_EXCLUSIVE, vpp);
1621 if (dvpp && *dvpp) {
1636 zfree(namei_zone, namebuf);
1641 * A fiddled version of m_adj() that ensures null fill to a long
1642 * boundary and only trims off the back end
1645 nfsm_adj(struct mbuf *mp, int len, int nul)
1652 * Trim from tail. Scan the mbuf chain,
1653 * calculating its length and finding the last mbuf.
1654 * If the adjustment only affects this mbuf, then just
1655 * adjust and return. Otherwise, rescan and truncate
1656 * after the remaining size.
1662 if (m->m_next == (struct mbuf *)0)
1666 if (m->m_len > len) {
1669 cp = mtod(m, caddr_t)+m->m_len-nul;
1670 for (i = 0; i < nul; i++)
1679 * Correct length for chain is "count".
1680 * Find the mbuf with last data, adjust its length,
1681 * and toss data from remaining mbufs on chain.
1683 for (m = mp; m; m = m->m_next) {
1684 if (m->m_len >= count) {
1687 cp = mtod(m, caddr_t)+m->m_len-nul;
1688 for (i = 0; i < nul; i++)
1695 for (m = m->m_next;m;m = m->m_next)
1700 * Make these functions instead of macros, so that the kernel text size
1701 * doesn't get too big...
1704 nfsm_srvwcc(struct nfsrv_descript *nfsd, int before_ret,
1705 struct vattr *before_vap, int after_ret, struct vattr *after_vap,
1706 struct mbuf **mbp, char **bposp)
1708 struct mbuf *mb = *mbp, *mb2;
1709 char *bpos = *bposp;
1713 * before_ret is 0 if before_vap is valid, non-zero if it isn't.
1716 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1719 nfsm_build(tl, u_int32_t *, 7 * NFSX_UNSIGNED);
1721 txdr_hyper(before_vap->va_size, tl);
1723 txdr_nfsv3time(&(before_vap->va_mtime), tl);
1725 txdr_nfsv3time(&(before_vap->va_ctime), tl);
1729 nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp);
1733 nfsm_srvpostopattr(struct nfsrv_descript *nfsd, int after_ret,
1734 struct vattr *after_vap, struct mbuf **mbp, char **bposp)
1736 struct mbuf *mb = *mbp, *mb2;
1737 char *bpos = *bposp;
1739 struct nfs_fattr *fp;
1742 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1745 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_V3FATTR);
1747 fp = (struct nfs_fattr *)tl;
1748 nfsm_srvfattr(nfsd, after_vap, fp);
1755 nfsm_srvfattr(struct nfsrv_descript *nfsd, struct vattr *vap,
1756 struct nfs_fattr *fp)
1759 fp->fa_nlink = txdr_unsigned(vap->va_nlink);
1760 fp->fa_uid = txdr_unsigned(vap->va_uid);
1761 fp->fa_gid = txdr_unsigned(vap->va_gid);
1762 if (nfsd->nd_flag & ND_NFSV3) {
1763 fp->fa_type = vtonfsv3_type(vap->va_type);
1764 fp->fa_mode = vtonfsv3_mode(vap->va_mode);
1765 txdr_hyper(vap->va_size, &fp->fa3_size);
1766 txdr_hyper(vap->va_bytes, &fp->fa3_used);
1767 fp->fa3_rdev.specdata1 = txdr_unsigned(umajor(vap->va_rdev));
1768 fp->fa3_rdev.specdata2 = txdr_unsigned(uminor(vap->va_rdev));
1769 fp->fa3_fsid.nfsuquad[0] = 0;
1770 fp->fa3_fsid.nfsuquad[1] = txdr_unsigned(vap->va_fsid);
1771 fp->fa3_fileid.nfsuquad[0] = 0;
1772 fp->fa3_fileid.nfsuquad[1] = txdr_unsigned(vap->va_fileid);
1773 txdr_nfsv3time(&vap->va_atime, &fp->fa3_atime);
1774 txdr_nfsv3time(&vap->va_mtime, &fp->fa3_mtime);
1775 txdr_nfsv3time(&vap->va_ctime, &fp->fa3_ctime);
1777 fp->fa_type = vtonfsv2_type(vap->va_type);
1778 fp->fa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1779 fp->fa2_size = txdr_unsigned(vap->va_size);
1780 fp->fa2_blocksize = txdr_unsigned(vap->va_blocksize);
1781 if (vap->va_type == VFIFO)
1782 fp->fa2_rdev = 0xffffffff;
1784 fp->fa2_rdev = txdr_unsigned(vap->va_rdev);
1785 fp->fa2_blocks = txdr_unsigned(vap->va_bytes / NFS_FABLKSIZE);
1786 fp->fa2_fsid = txdr_unsigned(vap->va_fsid);
1787 fp->fa2_fileid = txdr_unsigned(vap->va_fileid);
1788 txdr_nfsv2time(&vap->va_atime, &fp->fa2_atime);
1789 txdr_nfsv2time(&vap->va_mtime, &fp->fa2_mtime);
1790 txdr_nfsv2time(&vap->va_ctime, &fp->fa2_ctime);
1795 * nfsrv_fhtovp() - convert a fh to a vnode ptr (optionally locked)
1796 * - look up fsid in mount list (if not found ret error)
1797 * - get vp and export rights by calling VFS_FHTOVP()
1798 * - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
1799 * - if not lockflag unlock it with VOP_UNLOCK()
1802 nfsrv_fhtovp(fhandle_t *fhp, int lockflag, struct vnode **vpp,
1803 struct ucred *cred, struct nfssvc_sock *slp, struct sockaddr *nam,
1804 int *rdonlyp, int kerbflag, int pubflag)
1806 struct thread *td = curthread; /* XXX */
1809 struct ucred *credanon;
1811 #ifdef MNT_EXNORESPORT /* XXX needs mountd and /etc/exports help yet */
1812 struct sockaddr_int *saddr;
1815 *vpp = (struct vnode *)0;
1817 if (nfs_ispublicfh(fhp)) {
1818 if (!pubflag || !nfs_pub.np_valid)
1820 fhp = &nfs_pub.np_handle;
1823 mp = vfs_getvfs(&fhp->fh_fsid);
1826 error = VFS_CHECKEXP(mp, nam, &exflags, &credanon);
1829 error = VFS_FHTOVP(mp, &fhp->fh_fid, vpp);
1832 #ifdef MNT_EXNORESPORT
1833 if (!(exflags & (MNT_EXNORESPORT|MNT_EXPUBLIC))) {
1834 saddr = (struct sockaddr_in *)nam;
1835 if (saddr->sin_family == AF_INET &&
1836 ntohs(saddr->sin_port) >= IPPORT_RESERVED) {
1839 return (NFSERR_AUTHERR | AUTH_TOOWEAK);
1844 * Check/setup credentials.
1846 if (exflags & MNT_EXKERB) {
1850 return (NFSERR_AUTHERR | AUTH_TOOWEAK);
1852 } else if (kerbflag) {
1855 return (NFSERR_AUTHERR | AUTH_TOOWEAK);
1856 } else if (cred->cr_uid == 0 || (exflags & MNT_EXPORTANON)) {
1857 cred->cr_uid = credanon->cr_uid;
1858 for (i = 0; i < credanon->cr_ngroups && i < NGROUPS; i++)
1859 cred->cr_groups[i] = credanon->cr_groups[i];
1860 cred->cr_ngroups = i;
1862 if (exflags & MNT_EXRDONLY)
1867 nfsrv_object_create(*vpp);
1870 VOP_UNLOCK(*vpp, 0, td);
1876 * WebNFS: check if a filehandle is a public filehandle. For v3, this
1877 * means a length of 0, for v2 it means all zeroes. nfsm_srvmtofh has
1878 * transformed this to all zeroes in both cases, so check for it.
1881 nfs_ispublicfh(fhandle_t *fhp)
1883 char *cp = (char *)fhp;
1886 for (i = 0; i < NFSX_V3FH; i++)
1892 #endif /* NFS_NOSERVER */
1894 * This function compares two net addresses by family and returns TRUE
1895 * if they are the same host.
1896 * If there is any doubt, return FALSE.
1897 * The AF_INET family is handled as a special case so that address mbufs
1898 * don't need to be saved to store "struct in_addr", which is only 4 bytes.
1901 netaddr_match(int family, union nethostaddr *haddr, struct sockaddr *nam)
1903 struct sockaddr_in *inetaddr;
1907 inetaddr = (struct sockaddr_in *)nam;
1908 if (inetaddr->sin_family == AF_INET &&
1909 inetaddr->sin_addr.s_addr == haddr->had_inetaddr)
1918 static nfsuint64 nfs_nullcookie = { { 0, 0 } };
1920 * This function finds the directory cookie that corresponds to the
1921 * logical byte offset given.
1924 nfs_getcookie(struct nfsnode *np, off_t off, int add)
1926 struct nfsdmap *dp, *dp2;
1929 pos = (uoff_t)off / NFS_DIRBLKSIZ;
1930 if (pos == 0 || off < 0) {
1933 panic("nfs getcookie add at <= 0");
1935 return (&nfs_nullcookie);
1938 dp = np->n_cookies.lh_first;
1941 MALLOC(dp, struct nfsdmap *, sizeof (struct nfsdmap),
1942 M_NFSDIROFF, M_WAITOK);
1943 dp->ndm_eocookie = 0;
1944 LIST_INSERT_HEAD(&np->n_cookies, dp, ndm_list);
1946 return ((nfsuint64 *)0);
1948 while (pos >= NFSNUMCOOKIES) {
1949 pos -= NFSNUMCOOKIES;
1950 if (dp->ndm_list.le_next) {
1951 if (!add && dp->ndm_eocookie < NFSNUMCOOKIES &&
1952 pos >= dp->ndm_eocookie)
1953 return ((nfsuint64 *)0);
1954 dp = dp->ndm_list.le_next;
1956 MALLOC(dp2, struct nfsdmap *, sizeof (struct nfsdmap),
1957 M_NFSDIROFF, M_WAITOK);
1958 dp2->ndm_eocookie = 0;
1959 LIST_INSERT_AFTER(dp, dp2, ndm_list);
1962 return ((nfsuint64 *)0);
1964 if (pos >= dp->ndm_eocookie) {
1966 dp->ndm_eocookie = pos + 1;
1968 return ((nfsuint64 *)0);
1970 return (&dp->ndm_cookies[pos]);
1974 * Invalidate cached directory information, except for the actual directory
1975 * blocks (which are invalidated separately).
1976 * Done mainly to avoid the use of stale offset cookies.
1979 nfs_invaldir(struct vnode *vp)
1981 struct nfsnode *np = VTONFS(vp);
1984 if (vp->v_type != VDIR)
1985 panic("nfs: invaldir not dir");
1987 np->n_direofoffset = 0;
1988 np->n_cookieverf.nfsuquad[0] = 0;
1989 np->n_cookieverf.nfsuquad[1] = 0;
1990 if (np->n_cookies.lh_first)
1991 np->n_cookies.lh_first->ndm_eocookie = 0;
1995 * The write verifier has changed (probably due to a server reboot), so all
1996 * B_NEEDCOMMIT blocks will have to be written again. Since they are on the
1997 * dirty block list as B_DELWRI, all this takes is clearing the B_NEEDCOMMIT
1998 * and B_CLUSTEROK flags. Once done the new write verifier can be set for the
2001 * B_CLUSTEROK must be cleared along with B_NEEDCOMMIT because stage 1 data
2002 * writes are not clusterable.
2005 static int nfs_clearcommit_bp(struct buf *bp, void *data __unused);
2008 nfs_clearcommit(struct mount *mp)
2010 struct vnode *vp, *nvp;
2013 lwkt_gettoken(&ilock, &mntvnode_token);
2015 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp; vp = nvp) {
2016 nvp = TAILQ_NEXT(vp, v_nmntvnodes); /* ZZZ */
2017 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
2018 nfs_clearcommit_bp, NULL);
2021 lwkt_reltoken(&ilock);
2025 nfs_clearcommit_bp(struct buf *bp, void *data __unused)
2027 if (BUF_REFCNT(bp) == 0 &&
2028 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2029 == (B_DELWRI | B_NEEDCOMMIT)) {
2030 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
2035 #ifndef NFS_NOSERVER
2037 * Map errnos to NFS error numbers. For Version 3 also filter out error
2038 * numbers not specified for the associated procedure.
2041 nfsrv_errmap(struct nfsrv_descript *nd, int err)
2043 short *defaulterrp, *errp;
2045 if (nd->nd_flag & ND_NFSV3) {
2046 if (nd->nd_procnum <= NFSPROC_COMMIT) {
2047 errp = defaulterrp = nfsrv_v3errmap[nd->nd_procnum];
2051 else if (*errp > err)
2054 return ((int)*defaulterrp);
2056 return (err & 0xffff);
2059 return ((int)nfsrv_v2errmap[err - 1]);
2064 nfsrv_object_create(struct vnode *vp)
2066 struct thread *td = curthread;
2068 if (vp == NULL || vp->v_type != VREG)
2070 return (vfs_object_create(vp, td));
2074 * Sort the group list in increasing numerical order.
2075 * (Insertion sort by Chris Torek, who was grossed out by the bubble sort
2076 * that used to be here.)
2079 nfsrvw_sort(gid_t *list, int num)
2084 /* Insertion sort. */
2085 for (i = 1; i < num; i++) {
2087 /* find correct slot for value v, moving others up */
2088 for (j = i; --j >= 0 && v < list[j];)
2089 list[j + 1] = list[j];
2095 * copy credentials making sure that the result can be compared with bcmp().
2098 nfsrv_setcred(struct ucred *incred, struct ucred *outcred)
2102 bzero((caddr_t)outcred, sizeof (struct ucred));
2103 outcred->cr_ref = 1;
2104 outcred->cr_uid = incred->cr_uid;
2105 outcred->cr_ngroups = incred->cr_ngroups;
2106 for (i = 0; i < incred->cr_ngroups; i++)
2107 outcred->cr_groups[i] = incred->cr_groups[i];
2108 nfsrvw_sort(outcred->cr_groups, outcred->cr_ngroups);
2110 #endif /* NFS_NOSERVER */