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 $
41 * These functions support the macros and help fiddle mbuf chains for
42 * the nfs op functions. They do things like create the rpc header and
43 * copy data between mbuf chains and uio lists.
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/kernel.h>
50 #include <sys/mount.h>
51 #include <sys/vnode.h>
52 #include <sys/nlookup.h>
53 #include <sys/namei.h>
55 #include <sys/socket.h>
57 #include <sys/malloc.h>
58 #include <sys/sysent.h>
59 #include <sys/syscall.h>
61 #include <sys/objcache.h>
64 #include <vm/vm_object.h>
65 #include <vm/vm_extern.h>
75 #include "nfsm_subs.h"
78 #include <netinet/in.h>
80 MALLOC_DEFINE(M_NFSMOUNT, "NFS mount", "NFS mount");
83 * Data items converted to xdr at startup, since they are constant
84 * This is kinda hokey, but may save a little time doing byte swaps
86 u_int32_t nfs_xdrneg1;
87 u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers;
88 u_int32_t rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr;
89 u_int32_t rpc_auth_kerb;
90 u_int32_t nfs_prog, nfs_true, nfs_false;
92 /* And other global data */
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
103 * Protect master lists only. Primary protection uses the per-mount
104 * and per nfssvc_sock tokens.
106 struct lwkt_token nfs_token = LWKT_TOKEN_INITIALIZER(unp_token);
108 static int nfs_pbuf_freecnt = -1; /* start out unlimited */
110 struct nfsmount_head nfs_mountq = TAILQ_HEAD_INITIALIZER(nfs_mountq);
111 struct nfssvc_sockhead nfssvc_sockhead;
112 int nfssvc_sockhead_flag;
113 struct nfsd_head nfsd_head;
115 struct nfs_bufq nfs_bufq;
116 struct nqfhhashhead *nqfhhashtbl;
119 static int nfs_prev_nfssvc_sy_narg;
120 static sy_call_t *nfs_prev_nfssvc_sy_call;
125 * Mapping of old NFS Version 2 RPC numbers to generic numbers.
127 int nfsv3_procid[NFS_NPROCS] = {
156 #endif /* NFS_NOSERVER */
158 * and the reverse mapping from generic to Version 2 procedure numbers
160 int nfsv2_procid[NFS_NPROCS] = {
191 * Maps errno values to nfs error numbers.
192 * Use NFSERR_IO as the catch all for ones not specifically defined in
195 static u_char nfsrv_v2errmap[ELAST] = {
196 NFSERR_PERM, NFSERR_NOENT, NFSERR_IO, NFSERR_IO, NFSERR_IO,
197 NFSERR_NXIO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
198 NFSERR_IO, NFSERR_IO, NFSERR_ACCES, NFSERR_IO, NFSERR_IO,
199 NFSERR_IO, NFSERR_EXIST, NFSERR_IO, NFSERR_NODEV, NFSERR_NOTDIR,
200 NFSERR_ISDIR, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
201 NFSERR_IO, NFSERR_FBIG, NFSERR_NOSPC, NFSERR_IO, NFSERR_ROFS,
202 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
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, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
207 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
208 NFSERR_IO, NFSERR_IO, NFSERR_NAMETOL, NFSERR_IO, NFSERR_IO,
209 NFSERR_NOTEMPTY, NFSERR_IO, NFSERR_IO, NFSERR_DQUOT, NFSERR_STALE,
210 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
211 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
212 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
213 NFSERR_IO /* << Last is 86 */
217 * Maps errno values to nfs error numbers.
218 * Although it is not obvious whether or not NFS clients really care if
219 * a returned error value is in the specified list for the procedure, the
220 * safest thing to do is filter them appropriately. For Version 2, the
221 * X/Open XNFS document is the only specification that defines error values
222 * for each RPC (The RFC simply lists all possible error values for all RPCs),
223 * so I have decided to not do this for Version 2.
224 * The first entry is the default error return and the rest are the valid
225 * errors for that RPC in increasing numeric order.
227 static short nfsv3err_null[] = {
232 static short nfsv3err_getattr[] = {
241 static short nfsv3err_setattr[] = {
257 static short nfsv3err_lookup[] = {
270 static short nfsv3err_access[] = {
279 static short nfsv3err_readlink[] = {
291 static short nfsv3err_read[] = {
303 static short nfsv3err_write[] = {
318 static short nfsv3err_create[] = {
335 static short nfsv3err_mkdir[] = {
352 static short nfsv3err_symlink[] = {
369 static short nfsv3err_mknod[] = {
387 static short nfsv3err_remove[] = {
401 static short nfsv3err_rmdir[] = {
419 static short nfsv3err_rename[] = {
442 static short nfsv3err_link[] = {
462 static short nfsv3err_readdir[] = {
475 static short nfsv3err_readdirplus[] = {
489 static short nfsv3err_fsstat[] = {
498 static short nfsv3err_fsinfo[] = {
506 static short nfsv3err_pathconf[] = {
514 static short nfsv3err_commit[] = {
523 static short *nfsrv_v3errmap[] = {
541 nfsv3err_readdirplus,
548 #endif /* NFS_NOSERVER */
551 extern int sys_nfssvc(struct proc *, struct nfssvc_args *, int *);
554 * This needs to return a monotonically increasing or close to monotonically
555 * increasing result, otherwise the write gathering queues won't work
564 return ((u_quad_t)tv.tv_sec * 1000000 + (u_quad_t)tv.tv_usec);
568 * Called once to initialize data structures...
571 nfs_init(struct vfsconf *vfsp)
573 callout_init(&nfs_timer_handle);
574 nfsmount_objcache = objcache_create_simple(M_NFSMOUNT, sizeof(struct nfsmount));
576 nfs_mount_type = vfsp->vfc_typenum;
578 rpc_vers = txdr_unsigned(RPC_VER2);
579 rpc_call = txdr_unsigned(RPC_CALL);
580 rpc_reply = txdr_unsigned(RPC_REPLY);
581 rpc_msgdenied = txdr_unsigned(RPC_MSGDENIED);
582 rpc_msgaccepted = txdr_unsigned(RPC_MSGACCEPTED);
583 rpc_mismatch = txdr_unsigned(RPC_MISMATCH);
584 rpc_autherr = txdr_unsigned(RPC_AUTHERR);
585 rpc_auth_unix = txdr_unsigned(RPCAUTH_UNIX);
586 rpc_auth_kerb = txdr_unsigned(RPCAUTH_KERB4);
587 nfs_prog = txdr_unsigned(NFS_PROG);
588 nfs_true = txdr_unsigned(TRUE);
589 nfs_false = txdr_unsigned(FALSE);
590 nfs_xdrneg1 = txdr_unsigned(-1);
591 nfs_ticks = (hz * NFS_TICKINTVL + 500) / 1000;
594 nfs_nhinit(); /* Init the nfsnode table */
596 nfsrv_init(0); /* Init server data structures */
597 nfsrv_initcache(); /* Init the server request cache */
601 * Mainly for vkernel operation. If memory is severely limited
603 if (nfs_maxasyncbio > nmbclusters * MCLBYTES / NFS_MAXDATA / 3)
604 nfs_maxasyncbio = nmbclusters * MCLBYTES / NFS_MAXDATA / 3;
605 if (nfs_maxasyncbio < 4)
609 * Initialize reply list and start timer
611 nfs_timer_callout(0);
613 nfs_prev_nfssvc_sy_narg = sysent[SYS_nfssvc].sy_narg;
614 sysent[SYS_nfssvc].sy_narg = 2;
615 nfs_prev_nfssvc_sy_call = sysent[SYS_nfssvc].sy_call;
616 sysent[SYS_nfssvc].sy_call = (sy_call_t *)sys_nfssvc;
618 nfs_pbuf_freecnt = nswbuf / 2 + 1;
624 nfs_uninit(struct vfsconf *vfsp)
626 callout_stop(&nfs_timer_handle);
628 sysent[SYS_nfssvc].sy_narg = nfs_prev_nfssvc_sy_narg;
629 sysent[SYS_nfssvc].sy_call = nfs_prev_nfssvc_sy_call;
634 * Attribute cache routines.
635 * nfs_loadattrcache() - loads or updates the cache contents from attributes
636 * that are on the mbuf list
637 * nfs_getattrcache() - returns valid attributes if found in cache, returns
642 * Load the attribute cache (that lives in the nfsnode entry) with
643 * the values on the mbuf list. Load *vaper with the attributes. vaper
646 * As a side effect n_mtime, which we use to determine if the file was
647 * modified by some other host, is set to the attribute timestamp and
648 * NRMODIFIED is set if the two values differ.
650 * WARNING: the mtime loaded into vaper does not necessarily represent
651 * n_mtime or n_attr.mtime due to NACC and NUPD.
654 nfs_loadattrcache(struct vnode *vp, struct mbuf **mdp, caddr_t *dposp,
655 struct vattr *vaper, int lattr_flags)
658 struct nfs_fattr *fp;
668 struct timespec mtime;
669 int v3 = NFS_ISV3(vp);
672 t1 = (mtod(md, caddr_t) + md->m_len) - *dposp;
673 if ((error = nfsm_disct(mdp, dposp, NFSX_FATTR(v3), t1, &cp2)) != 0)
675 fp = (struct nfs_fattr *)cp2;
677 vtyp = nfsv3tov_type(fp->fa_type);
678 vmode = fxdr_unsigned(u_short, fp->fa_mode);
679 rmajor = (int)fxdr_unsigned(int, fp->fa3_rdev.specdata1);
680 rminor = (int)fxdr_unsigned(int, fp->fa3_rdev.specdata2);
681 fxdr_nfsv3time(&fp->fa3_mtime, &mtime);
683 vtyp = nfsv2tov_type(fp->fa_type);
684 vmode = fxdr_unsigned(u_short, fp->fa_mode);
688 * The duplicate information returned in fa_type and fa_mode
689 * is an ambiguity in the NFS version 2 protocol.
691 * VREG should be taken literally as a regular file. If a
692 * server intents to return some type information differently
693 * in the upper bits of the mode field (e.g. for sockets, or
694 * FIFOs), NFSv2 mandates fa_type to be VNON. Anyway, we
695 * leave the examination of the mode bits even in the VREG
696 * case to avoid breakage for bogus servers, but we make sure
697 * that there are actually type bits set in the upper part of
698 * fa_mode (and failing that, trust the va_type field).
700 * NFSv3 cleared the issue, and requires fa_mode to not
701 * contain any type information (while also introduing sockets
702 * and FIFOs for fa_type).
704 if (vtyp == VNON || (vtyp == VREG && (vmode & S_IFMT) != 0))
705 vtyp = IFTOVT(vmode);
706 rdev = fxdr_unsigned(int32_t, fp->fa2_rdev);
707 rmajor = umajor(rdev);
708 rminor = uminor(rdev);
709 fxdr_nfsv2time(&fp->fa2_mtime, &mtime);
712 * Really ugly NFSv2 kludge.
714 if (vtyp == VCHR && rdev == (udev_t)0xffffffff)
719 * If v_type == VNON it is a new node, so fill in the v_type,
720 * n_mtime fields. Check to see if it represents a special
721 * device, and if so, check for a possible alias. Once the
722 * correct vnode has been obtained, fill in the rest of the
726 if (vp->v_type != vtyp) {
727 nfs_setvtype(vp, vtyp);
728 if (vp->v_type == VFIFO) {
729 vp->v_ops = &vp->v_mount->mnt_vn_fifo_ops;
730 } else if (vp->v_type == VCHR || vp->v_type == VBLK) {
731 vp->v_ops = &vp->v_mount->mnt_vn_spec_ops;
732 addaliasu(vp, rmajor, rminor);
734 vp->v_ops = &vp->v_mount->mnt_vn_use_ops;
736 np->n_mtime = mtime.tv_sec;
737 } else if (np->n_mtime != mtime.tv_sec) {
739 * If we haven't modified the file locally and the server
740 * timestamp does not match, then the server probably
741 * modified the file. We must flag this condition so
742 * the proper syncnronization can be done. We do not
743 * try to synchronize the state here because that
744 * could lead to an endless recursion.
746 * XXX loadattrcache can be set during the reply to a write,
747 * before the write timestamp is properly processed. To
748 * avoid unconditionally setting the rmodified bit (which
749 * has the effect of flushing the cache), we only do this
750 * check if the lmodified bit is not set.
752 np->n_mtime = mtime.tv_sec;
753 if ((lattr_flags & NFS_LATTR_NOMTIMECHECK) == 0)
754 np->n_flag |= NRMODIFIED;
758 vap->va_mode = (vmode & 07777);
759 vap->va_rmajor = rmajor;
760 vap->va_rminor = rminor;
761 vap->va_mtime = mtime;
762 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
764 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
765 vap->va_uid = fxdr_unsigned(uid_t, fp->fa_uid);
766 vap->va_gid = fxdr_unsigned(gid_t, fp->fa_gid);
767 vap->va_size = fxdr_hyper(&fp->fa3_size);
768 vap->va_blocksize = NFS_FABLKSIZE;
769 vap->va_bytes = fxdr_hyper(&fp->fa3_used);
770 vap->va_fileid = fxdr_hyper(&fp->fa3_fileid);
771 fxdr_nfsv3time(&fp->fa3_atime, &vap->va_atime);
772 fxdr_nfsv3time(&fp->fa3_ctime, &vap->va_ctime);
776 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
777 vap->va_uid = fxdr_unsigned(uid_t, fp->fa_uid);
778 vap->va_gid = fxdr_unsigned(gid_t, fp->fa_gid);
779 vap->va_size = fxdr_unsigned(u_int32_t, fp->fa2_size);
780 vap->va_blocksize = fxdr_unsigned(int32_t, fp->fa2_blocksize);
781 vap->va_bytes = (u_quad_t)fxdr_unsigned(int32_t, fp->fa2_blocks)
783 vap->va_fileid = fxdr_unsigned(int32_t, fp->fa2_fileid);
784 fxdr_nfsv2time(&fp->fa2_atime, &vap->va_atime);
786 vap->va_ctime.tv_sec = fxdr_unsigned(u_int32_t,
787 fp->fa2_ctime.nfsv2_sec);
788 vap->va_ctime.tv_nsec = 0;
789 vap->va_gen = fxdr_unsigned(u_int32_t,fp->fa2_ctime.nfsv2_usec);
792 np->n_attrstamp = time_second;
793 if (vap->va_size != np->n_size) {
794 if (vap->va_type == VREG) {
796 * Get rid of all the junk we had before and just
797 * set NRMODIFIED if NLMODIFIED is 0. Depend on
798 * occassionally flushing our dirty buffers to
799 * clear both the NLMODIFIED and NRMODIFIED flags.
801 if ((np->n_flag & NLMODIFIED) == 0)
802 np->n_flag |= NRMODIFIED;
804 if ((lattr_flags & NFS_LATTR_NOSHRINK) &&
805 vap->va_size < np->n_size) {
807 * We've been told not to shrink the file;
808 * zero np->n_attrstamp to indicate that
809 * the attributes are stale.
811 * This occurs primarily due to recursive
812 * NFS ops that are executed during periods
813 * where we cannot safely reduce the size of
816 * Additionally, write rpcs are broken down
817 * into buffers and np->n_size is
818 * pre-extended. Setting NRMODIFIED here
819 * can result in n_size getting reset to a
820 * lower value, which is NOT what we want.
821 * XXX this needs to be cleaned up a lot
824 vap->va_size = np->n_size;
826 if ((np->n_flag & NLMODIFIED) == 0)
827 np->n_flag |= NRMODIFIED;
828 } else if (np->n_flag & NLMODIFIED) {
830 * We've modified the file: Use the larger
831 * of our size, and the server's size. At
832 * this point the cache coherency is all
833 * shot to hell. To try to handle multiple
834 * clients appending to the file at the same
835 * time mark that the server has changed
836 * the file if the server's notion of the
837 * file size is larger then our notion.
839 * XXX this needs work.
841 if (vap->va_size < np->n_size) {
842 vap->va_size = np->n_size;
844 np->n_size = vap->va_size;
845 np->n_flag |= NRMODIFIED;
849 * Someone changed the file's size on the
850 * server and there are no local changes
851 * to get in the way, set the size and mark
854 np->n_size = vap->va_size;
855 np->n_flag |= NRMODIFIED;
857 nvnode_pager_setsize(vp, np->n_size, XXX);
860 np->n_size = vap->va_size;
864 bcopy((caddr_t)vap, (caddr_t)vaper, sizeof(*vap));
865 if (np->n_flag & NCHG) {
866 if (np->n_flag & NACC)
867 vaper->va_atime = np->n_atim;
868 if (np->n_flag & NUPD)
869 vaper->va_mtime = np->n_mtim;
876 #include <sys/sysctl.h>
877 SYSCTL_DECL(_vfs_nfs);
878 static int nfs_acdebug;
879 SYSCTL_INT(_vfs_nfs, OID_AUTO, acdebug, CTLFLAG_RW, &nfs_acdebug, 0, "");
883 * Check the time stamp
884 * If the cache is valid, copy contents to *vap and return 0
885 * otherwise return an error
888 nfs_getattrcache(struct vnode *vp, struct vattr *vaper)
892 struct nfsmount *nmp;
897 nmp = VFSTONFS(vp->v_mount);
900 * Dynamic timeout based on how recently the file was modified.
901 * n_mtime is always valid.
903 timeo = (get_approximate_time_t() - np->n_mtime) / 60;
907 kprintf("nfs_getattrcache: initial timeo = %d\n", timeo);
910 if (vap->va_type == VDIR) {
911 if ((np->n_flag & NLMODIFIED) || timeo < nmp->nm_acdirmin)
912 timeo = nmp->nm_acdirmin;
913 else if (timeo > nmp->nm_acdirmax)
914 timeo = nmp->nm_acdirmax;
916 if ((np->n_flag & NLMODIFIED) || timeo < nmp->nm_acregmin)
917 timeo = nmp->nm_acregmin;
918 else if (timeo > nmp->nm_acregmax)
919 timeo = nmp->nm_acregmax;
924 kprintf("acregmin %d; acregmax %d; acdirmin %d; acdirmax %d\n",
925 nmp->nm_acregmin, nmp->nm_acregmax,
926 nmp->nm_acdirmin, nmp->nm_acdirmax);
929 kprintf("nfs_getattrcache: age = %d; final timeo = %d\n",
930 (int)(time_second - np->n_attrstamp), timeo);
933 if (np->n_attrstamp == 0 || (time_second - np->n_attrstamp) >= timeo) {
934 nfsstats.attrcache_misses++;
937 nfsstats.attrcache_hits++;
940 * Our attribute cache can be stale due to modifications made on
941 * this host. XXX this is a bad hack. We need a more deterministic
942 * means of finding out which np fields are valid verses attr cache
943 * fields. We really should update the vattr info on the fly when
944 * making local changes.
946 if (vap->va_size != np->n_size) {
947 if (vap->va_type == VREG) {
948 if (np->n_flag & NLMODIFIED)
949 vap->va_size = np->n_size;
950 nfs_meta_setsize(vp, curthread, vap->va_size, 0);
952 np->n_size = vap->va_size;
955 bcopy((caddr_t)vap, (caddr_t)vaper, sizeof(struct vattr));
956 if (np->n_flag & NCHG) {
957 if (np->n_flag & NACC)
958 vaper->va_atime = np->n_atim;
959 if (np->n_flag & NUPD)
960 vaper->va_mtime = np->n_mtim;
968 * Set up nameidata for a lookup() call and do it.
970 * If pubflag is set, this call is done for a lookup operation on the
971 * public filehandle. In that case we allow crossing mountpoints and
972 * absolute pathnames. However, the caller is expected to check that
973 * the lookup result is within the public fs, and deny access if
976 * dirp may be set whether an error is returned or not, and must be
977 * released by the caller.
979 * On return nd->nl_nch usually points to the target ncp, which may represent
982 * NOTE: the caller must call nlookup_done(nd) unconditionally on return
986 nfs_namei(struct nlookupdata *nd, struct ucred *cred, int nflags,
987 struct vnode **dvpp, struct vnode **vpp,
988 fhandle_t *fhp, int len,
989 struct nfssvc_sock *slp, struct sockaddr *nam, struct mbuf **mdp,
990 caddr_t *dposp, struct vnode **dirpp, struct thread *td,
991 int kerbflag, int pubflag)
995 char *fromcp, *tocp, *cp;
1002 namebuf = objcache_get(namei_oc, M_WAITOK);
1006 * Copy the name from the mbuf list to namebuf.
1011 rem = mtod(md, caddr_t) + md->m_len - fromcp;
1012 for (i = 0; i < len; i++) {
1019 fromcp = mtod(md, caddr_t);
1022 if (*fromcp == '\0' || (!pubflag && *fromcp == '/')) {
1026 *tocp++ = *fromcp++;
1032 len = nfsm_rndup(len)-len;
1036 else if ((error = nfs_adv(mdp, dposp, len, rem)) != 0)
1041 * Extract and set starting directory. The returned dp is refd
1044 error = nfsrv_fhtovp(fhp, FALSE, &mp, &dp, cred, slp,
1045 nam, &rdonly, kerbflag, pubflag);
1048 if (dp->v_type != VDIR) {
1055 * Set return directory. Reference to dp is implicitly transfered
1056 * to the returned pointer. This must be set before we potentially
1062 * read-only - NLC_DELETE, NLC_RENAME_DST are disallowed. NLC_CREATE
1063 * is passed through to nlookup() and will be disallowed
1064 * if the file does not already exist.
1067 nflags |= NLC_NFS_RDONLY;
1068 if (nflags & (NLC_DELETE | NLC_RENAME_DST)) {
1075 * Oh joy. For WebNFS, handle those pesky '%' escapes,
1076 * and the 'native path' indicator.
1079 cp = objcache_get(namei_oc, M_WAITOK);
1082 if ((unsigned char)*fromcp >= WEBNFS_SPECCHAR_START) {
1083 switch ((unsigned char)*fromcp) {
1084 case WEBNFS_NATIVE_CHAR:
1086 * 'Native' path for us is the same
1087 * as a path according to the NFS spec,
1088 * just skip the escape char.
1093 * More may be added in the future, range 0x80-0xff
1097 objcache_put(namei_oc, cp);
1102 * Translate the '%' escapes, URL-style.
1104 while (*fromcp != '\0') {
1105 if (*fromcp == WEBNFS_ESC_CHAR) {
1106 if (fromcp[1] != '\0' && fromcp[2] != '\0') {
1108 *tocp++ = HEXSTRTOI(fromcp);
1113 objcache_put(namei_oc, cp);
1117 *tocp++ = *fromcp++;
1120 objcache_put(namei_oc, namebuf);
1125 * Setup for search. We need to get a start directory from dp. Note
1126 * that dp is ref'd, but we no longer 'own' the ref (*dirpp owns it).
1129 nflags |= NLC_NFS_NOSOFTLINKTRAV;
1130 nflags |= NLC_NOCROSSMOUNT;
1134 * We need a starting ncp from the directory vnode dp. dp must not
1135 * be locked. The returned ncp will be refd but not locked.
1137 * If no suitable ncp is found we instruct cache_fromdvp() to create
1138 * one. If this fails the directory has probably been removed while
1139 * the target was chdir'd into it and any further lookup will fail.
1141 if ((error = cache_fromdvp(dp, cred, 1, &nch)) != 0)
1143 nlookup_init_raw(nd, namebuf, UIO_SYSSPACE, nflags, cred, &nch);
1147 * Ok, do the lookup.
1149 error = nlookup(nd);
1152 * If no error occured return the requested dvpp and vpp. If
1153 * NLC_CREATE was specified nd->nl_nch may represent a negative
1154 * cache hit in which case we do not attempt to obtain the vp.
1158 if (nd->nl_nch.ncp->nc_parent) {
1160 nch.ncp = nch.ncp->nc_parent;
1163 error = cache_vget(&nch, nd->nl_cred,
1164 LK_EXCLUSIVE, dvpp);
1170 if (vpp && nd->nl_nch.ncp->nc_vp) {
1171 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, vpp);
1174 if (dvpp && *dvpp) {
1189 objcache_put(namei_oc, namebuf);
1194 * nfsrv_fhtovp() - convert a fh to a vnode ptr (optionally locked)
1195 * - look up fsid in mount list (if not found ret error)
1196 * - get vp and export rights by calling VFS_FHTOVP()
1197 * - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
1198 * - if not lockflag unlock it with vn_unlock()
1201 nfsrv_fhtovp(fhandle_t *fhp, int lockflag,
1202 struct mount **mpp, struct vnode **vpp,
1203 struct ucred *cred, struct nfssvc_sock *slp, struct sockaddr *nam,
1204 int *rdonlyp, int kerbflag, int pubflag)
1208 struct ucred *credanon;
1210 #ifdef MNT_EXNORESPORT /* XXX needs mountd and /etc/exports help yet */
1211 struct sockaddr_int *saddr;
1217 if (nfs_ispublicfh(fhp)) {
1218 if (!pubflag || !nfs_pub.np_valid)
1220 fhp = &nfs_pub.np_handle;
1223 mp = *mpp = vfs_getvfs(&fhp->fh_fsid);
1226 error = VFS_CHECKEXP(mp, nam, &exflags, &credanon);
1229 error = VFS_FHTOVP(mp, NULL, &fhp->fh_fid, vpp);
1232 #ifdef MNT_EXNORESPORT
1233 if (!(exflags & (MNT_EXNORESPORT|MNT_EXPUBLIC))) {
1234 saddr = (struct sockaddr_in *)nam;
1235 if (saddr->sin_family == AF_INET &&
1236 ntohs(saddr->sin_port) >= IPPORT_RESERVED) {
1239 return (NFSERR_AUTHERR | AUTH_TOOWEAK);
1244 * Check/setup credentials.
1246 if (exflags & MNT_EXKERB) {
1250 return (NFSERR_AUTHERR | AUTH_TOOWEAK);
1252 } else if (kerbflag) {
1255 return (NFSERR_AUTHERR | AUTH_TOOWEAK);
1256 } else if (cred->cr_uid == 0 || (exflags & MNT_EXPORTANON)) {
1257 cred->cr_uid = credanon->cr_uid;
1258 for (i = 0; i < credanon->cr_ngroups && i < NGROUPS; i++)
1259 cred->cr_groups[i] = credanon->cr_groups[i];
1260 cred->cr_ngroups = i;
1262 if (exflags & MNT_EXRDONLY)
1273 * WebNFS: check if a filehandle is a public filehandle. For v3, this
1274 * means a length of 0, for v2 it means all zeroes. nfsm_srvmtofh has
1275 * transformed this to all zeroes in both cases, so check for it.
1278 nfs_ispublicfh(fhandle_t *fhp)
1280 char *cp = (char *)fhp;
1283 for (i = 0; i < NFSX_V3FH; i++)
1289 #endif /* NFS_NOSERVER */
1291 * This function compares two net addresses by family and returns TRUE
1292 * if they are the same host.
1293 * If there is any doubt, return FALSE.
1294 * The AF_INET family is handled as a special case so that address mbufs
1295 * don't need to be saved to store "struct in_addr", which is only 4 bytes.
1298 netaddr_match(int family, union nethostaddr *haddr, struct sockaddr *nam)
1300 struct sockaddr_in *inetaddr;
1304 inetaddr = (struct sockaddr_in *)nam;
1305 if (inetaddr->sin_family == AF_INET &&
1306 inetaddr->sin_addr.s_addr == haddr->had_inetaddr)
1315 static nfsuint64 nfs_nullcookie = { { 0, 0 } };
1317 * This function finds the directory cookie that corresponds to the
1318 * logical byte offset given.
1321 nfs_getcookie(struct nfsnode *np, off_t off, int add)
1323 struct nfsdmap *dp, *dp2;
1326 pos = (uoff_t)off / NFS_DIRBLKSIZ;
1327 if (pos == 0 || off < 0) {
1330 panic("nfs getcookie add at <= 0");
1332 return (&nfs_nullcookie);
1335 dp = np->n_cookies.lh_first;
1338 dp = kmalloc(sizeof(struct nfsdmap), M_NFSDIROFF,
1340 dp->ndm_eocookie = 0;
1341 LIST_INSERT_HEAD(&np->n_cookies, dp, ndm_list);
1345 while (pos >= NFSNUMCOOKIES) {
1346 pos -= NFSNUMCOOKIES;
1347 if (dp->ndm_list.le_next) {
1348 if (!add && dp->ndm_eocookie < NFSNUMCOOKIES &&
1349 pos >= dp->ndm_eocookie)
1351 dp = dp->ndm_list.le_next;
1353 dp2 = kmalloc(sizeof(struct nfsdmap), M_NFSDIROFF,
1355 dp2->ndm_eocookie = 0;
1356 LIST_INSERT_AFTER(dp, dp2, ndm_list);
1361 if (pos >= dp->ndm_eocookie) {
1363 dp->ndm_eocookie = pos + 1;
1367 return (&dp->ndm_cookies[pos]);
1371 * Invalidate cached directory information, except for the actual directory
1372 * blocks (which are invalidated separately).
1373 * Done mainly to avoid the use of stale offset cookies.
1376 nfs_invaldir(struct vnode *vp)
1378 struct nfsnode *np = VTONFS(vp);
1381 if (vp->v_type != VDIR)
1382 panic("nfs: invaldir not dir");
1384 np->n_direofoffset = 0;
1385 np->n_cookieverf.nfsuquad[0] = 0;
1386 np->n_cookieverf.nfsuquad[1] = 0;
1387 if (np->n_cookies.lh_first)
1388 np->n_cookies.lh_first->ndm_eocookie = 0;
1392 * Set the v_type field for an NFS client's vnode and initialize for
1393 * buffer cache operations if necessary.
1396 nfs_setvtype(struct vnode *vp, enum vtype vtyp)
1405 * Needs VMIO, size not yet known, and blocksize
1406 * is not really relevant if we are passing a
1409 vinitvmio(vp, 0, PAGE_SIZE, -1);
1417 * The write verifier has changed (probably due to a server reboot), so all
1418 * B_NEEDCOMMIT blocks will have to be written again. Since they are on the
1419 * dirty block list as B_DELWRI, all this takes is clearing the B_NEEDCOMMIT
1420 * and B_CLUSTEROK flags. Once done the new write verifier can be set for the
1423 * B_CLUSTEROK must be cleared along with B_NEEDCOMMIT because stage 1 data
1424 * writes are not clusterable.
1427 static int nfs_clearcommit_bp(struct buf *bp, void *data __unused);
1428 static int nfs_clearcommit_callback(struct mount *mp, struct vnode *vp,
1429 void *data __unused);
1432 nfs_clearcommit(struct mount *mp)
1434 vmntvnodescan(mp, VMSC_NOWAIT, nfs_clearcommit_callback, NULL, NULL);
1438 nfs_clearcommit_callback(struct mount *mp, struct vnode *vp,
1439 void *data __unused)
1442 lwkt_gettoken(&vp->v_token);
1443 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
1444 nfs_clearcommit_bp, NULL);
1445 lwkt_reltoken(&vp->v_token);
1451 nfs_clearcommit_bp(struct buf *bp, void *data __unused)
1453 if (BUF_REFCNT(bp) == 0 &&
1454 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
1455 == (B_DELWRI | B_NEEDCOMMIT)) {
1456 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
1461 #ifndef NFS_NOSERVER
1463 * Map errnos to NFS error numbers. For Version 3 also filter out error
1464 * numbers not specified for the associated procedure.
1467 nfsrv_errmap(struct nfsrv_descript *nd, int err)
1469 short *defaulterrp, *errp;
1471 if (nd->nd_flag & ND_NFSV3) {
1472 if (nd->nd_procnum <= NFSPROC_COMMIT) {
1473 errp = defaulterrp = nfsrv_v3errmap[nd->nd_procnum];
1477 else if (*errp > err)
1480 return ((int)*defaulterrp);
1482 return (err & 0xffff);
1485 return ((int)nfsrv_v2errmap[err - 1]);
1490 * Sort the group list in increasing numerical order.
1491 * (Insertion sort by Chris Torek, who was grossed out by the bubble sort
1492 * that used to be here.)
1495 nfsrvw_sort(gid_t *list, int num)
1500 /* Insertion sort. */
1501 for (i = 1; i < num; i++) {
1503 /* find correct slot for value v, moving others up */
1504 for (j = i; --j >= 0 && v < list[j];)
1505 list[j + 1] = list[j];
1511 * copy credentials making sure that the result can be compared with bcmp().
1514 nfsrv_setcred(struct ucred *incred, struct ucred *outcred)
1518 bzero((caddr_t)outcred, sizeof (struct ucred));
1519 outcred->cr_ref = 1;
1520 outcred->cr_uid = incred->cr_uid;
1521 outcred->cr_ngroups = incred->cr_ngroups;
1522 for (i = 0; i < incred->cr_ngroups; i++)
1523 outcred->cr_groups[i] = incred->cr_groups[i];
1524 nfsrvw_sort(outcred->cr_groups, outcred->cr_ngroups);
1526 #endif /* NFS_NOSERVER */