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. 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_subs.c 8.8 (Berkeley) 5/22/95
33 * $FreeBSD: /repoman/r/ncvs/src/sys/nfsclient/nfs_subs.c,v 1.128 2004/04/14 23:23:55 peadar Exp $
37 * These functions support the macros and help fiddle mbuf chains for
38 * the nfs op functions. They do things like create the rpc header and
39 * copy data between mbuf chains and uio lists.
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/kernel.h>
46 #include <sys/mount.h>
47 #include <sys/vnode.h>
48 #include <sys/nlookup.h>
50 #include <sys/socket.h>
52 #include <sys/malloc.h>
53 #include <sys/sysent.h>
54 #include <sys/syscall.h>
55 #include <sys/sysmsg.h>
57 #include <sys/objcache.h>
61 #include <vm/vm_object.h>
62 #include <vm/vm_extern.h>
72 #include "nfsm_subs.h"
75 #include <netinet/in.h>
78 * Data items converted to xdr at startup, since they are constant
79 * This is kinda hokey, but may save a little time doing byte swaps
81 u_int32_t nfs_xdrneg1;
82 u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers;
83 u_int32_t rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr;
84 u_int32_t rpc_auth_kerb;
85 u_int32_t nfs_prog, nfs_true, nfs_false;
87 /* And other global data */
88 static enum vtype nv2tov_type[8]= {
89 VNON, VREG, VDIR, VBLK, VCHR, VLNK, VNON, VNON
91 enum vtype nv3tov_type[8]= {
92 VNON, VREG, VDIR, VBLK, VCHR, VLNK, VSOCK, VFIFO
98 * Protect master lists only. Primary protection uses the per-mount
99 * and per nfssvc_sock tokens.
101 struct lwkt_token nfs_token = LWKT_TOKEN_INITIALIZER(unp_token);
103 static long nfs_pbuf_freecnt = -1; /* start out unlimited */
105 struct nfsmount_head nfs_mountq = TAILQ_HEAD_INITIALIZER(nfs_mountq);
106 struct nfssvc_sockhead nfssvc_sockhead;
107 int nfssvc_sockhead_flag;
108 struct nfsd_head nfsd_head;
110 struct nfs_bufq nfs_bufq;
111 struct nqfhhashhead *nqfhhashtbl;
114 static int nfs_prev_nfssvc_sy_narg;
115 static sy_call_t *nfs_prev_nfssvc_sy_call;
120 * Mapping of old NFS Version 2 RPC numbers to generic numbers.
122 int nfsv3_procid[NFS_NPROCS] = {
151 #endif /* NFS_NOSERVER */
153 * and the reverse mapping from generic to Version 2 procedure numbers
155 int nfsv2_procid[NFS_NPROCS] = {
186 * Maps errno values to nfs error numbers.
187 * Use NFSERR_IO as the catch all for ones not specifically defined in
190 static u_char nfsrv_v2errmap[ELAST] = {
191 NFSERR_PERM, NFSERR_NOENT, NFSERR_IO, NFSERR_IO, NFSERR_IO,
192 NFSERR_NXIO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
193 NFSERR_IO, NFSERR_IO, NFSERR_ACCES, NFSERR_IO, NFSERR_IO,
194 NFSERR_IO, NFSERR_EXIST, NFSERR_IO, NFSERR_NODEV, NFSERR_NOTDIR,
195 NFSERR_ISDIR, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
196 NFSERR_IO, NFSERR_FBIG, NFSERR_NOSPC, NFSERR_IO, NFSERR_ROFS,
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_IO, NFSERR_IO, NFSERR_IO,
202 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
203 NFSERR_IO, NFSERR_IO, NFSERR_NAMETOL, NFSERR_IO, NFSERR_IO,
204 NFSERR_NOTEMPTY, NFSERR_IO, NFSERR_IO, NFSERR_DQUOT, NFSERR_STALE,
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 /* << Last is 86 */
212 * Maps errno values to nfs error numbers.
213 * Although it is not obvious whether or not NFS clients really care if
214 * a returned error value is in the specified list for the procedure, the
215 * safest thing to do is filter them appropriately. For Version 2, the
216 * X/Open XNFS document is the only specification that defines error values
217 * for each RPC (The RFC simply lists all possible error values for all RPCs),
218 * so I have decided to not do this for Version 2.
219 * The first entry is the default error return and the rest are the valid
220 * errors for that RPC in increasing numeric order.
222 static short nfsv3err_null[] = {
227 static short nfsv3err_getattr[] = {
236 static short nfsv3err_setattr[] = {
252 static short nfsv3err_lookup[] = {
265 static short nfsv3err_access[] = {
274 static short nfsv3err_readlink[] = {
286 static short nfsv3err_read[] = {
298 static short nfsv3err_write[] = {
313 static short nfsv3err_create[] = {
330 static short nfsv3err_mkdir[] = {
347 static short nfsv3err_symlink[] = {
364 static short nfsv3err_mknod[] = {
382 static short nfsv3err_remove[] = {
396 static short nfsv3err_rmdir[] = {
414 static short nfsv3err_rename[] = {
437 static short nfsv3err_link[] = {
457 static short nfsv3err_readdir[] = {
470 static short nfsv3err_readdirplus[] = {
484 static short nfsv3err_fsstat[] = {
493 static short nfsv3err_fsinfo[] = {
501 static short nfsv3err_pathconf[] = {
509 static short nfsv3err_commit[] = {
518 static short *nfsrv_v3errmap[] = {
536 nfsv3err_readdirplus,
543 #endif /* NFS_NOSERVER */
546 * This needs to return a monotonically increasing or close to monotonically
547 * increasing result, otherwise the write gathering queues won't work
556 return ((u_quad_t)tv.tv_sec * 1000000 + (u_quad_t)tv.tv_usec);
560 * Called once to initialize data structures...
563 nfs_init(struct vfsconf *vfsp)
565 callout_init_mp(&nfs_timer_handle);
567 nfs_mount_type = vfsp->vfc_typenum;
569 rpc_vers = txdr_unsigned(RPC_VER2);
570 rpc_call = txdr_unsigned(RPC_CALL);
571 rpc_reply = txdr_unsigned(RPC_REPLY);
572 rpc_msgdenied = txdr_unsigned(RPC_MSGDENIED);
573 rpc_msgaccepted = txdr_unsigned(RPC_MSGACCEPTED);
574 rpc_mismatch = txdr_unsigned(RPC_MISMATCH);
575 rpc_autherr = txdr_unsigned(RPC_AUTHERR);
576 rpc_auth_unix = txdr_unsigned(RPCAUTH_UNIX);
577 rpc_auth_kerb = txdr_unsigned(RPCAUTH_KERB4);
578 nfs_prog = txdr_unsigned(NFS_PROG);
579 nfs_true = txdr_unsigned(TRUE);
580 nfs_false = txdr_unsigned(FALSE);
581 nfs_xdrneg1 = txdr_unsigned(-1);
582 nfs_ticks = (hz * NFS_TICKINTVL + 500) / 1000;
585 nfs_nhinit(); /* Init the nfsnode table */
587 nfsrv_init(0); /* Init server data structures */
588 nfsrv_initcache(); /* Init the server request cache */
592 * Mainly for vkernel operation. If memory is severely limited
594 if (nfs_maxasyncbio > nmbclusters * MCLBYTES / NFS_MAXDATA / 3)
595 nfs_maxasyncbio = nmbclusters * MCLBYTES / NFS_MAXDATA / 3;
596 if (nfs_maxasyncbio < 4)
600 * Initialize reply list and start timer
602 nfs_timer_callout(0);
604 #if 1 /* XXX this isn't really needed */
605 nfs_prev_nfssvc_sy_narg = sysent[SYS_nfssvc].sy_narg;
606 sysent[SYS_nfssvc].sy_narg = 2;
608 nfs_prev_nfssvc_sy_call = sysent[SYS_nfssvc].sy_call;
609 sysent[SYS_nfssvc].sy_call = (sy_call_t *)sys_nfssvc;
611 nfs_pbuf_freecnt = nswbuf_kva / 2 + 1;
617 nfs_uninit(struct vfsconf *vfsp)
619 callout_stop(&nfs_timer_handle);
621 sysent[SYS_nfssvc].sy_narg = nfs_prev_nfssvc_sy_narg;
622 sysent[SYS_nfssvc].sy_call = nfs_prev_nfssvc_sy_call;
623 nfs_nhdestroy(); /* Destroy the nfsnode table */
625 nfsrv_destroycache(); /* Destroy the server request cache */
631 * Attribute cache routines.
632 * nfs_loadattrcache() - loads or updates the cache contents from attributes
633 * that are on the mbuf list
634 * nfs_getattrcache() - returns valid attributes if found in cache, returns
639 * Load the attribute cache (that lives in the nfsnode entry) with
640 * the values on the mbuf list. Load *vaper with the attributes. vaper
643 * As a side effect n_mtime, which we use to determine if the file was
644 * modified by some other host, is set to the attribute timestamp and
645 * NRMODIFIED is set if the two values differ.
647 * WARNING: the mtime loaded into vaper does not necessarily represent
648 * n_mtime or n_attr.mtime due to NACC and NUPD.
651 nfs_loadattrcache(struct vnode *vp, struct mbuf **mdp, caddr_t *dposp,
652 struct vattr *vaper, int lattr_flags)
655 struct nfs_fattr *fp;
665 struct timespec mtime;
666 int v3 = NFS_ISV3(vp);
669 t1 = (mtod(md, caddr_t) + md->m_len) - *dposp;
670 if ((error = nfsm_disct(mdp, dposp, NFSX_FATTR(v3), t1, &cp2)) != 0)
672 fp = (struct nfs_fattr *)cp2;
674 vtyp = nfsv3tov_type(fp->fa_type);
675 vmode = fxdr_unsigned(u_short, fp->fa_mode);
676 rmajor = (int)fxdr_unsigned(int, fp->fa3_rdev.specdata1);
677 rminor = (int)fxdr_unsigned(int, fp->fa3_rdev.specdata2);
678 fxdr_nfsv3time(&fp->fa3_mtime, &mtime);
680 vtyp = nfsv2tov_type(fp->fa_type);
681 vmode = fxdr_unsigned(u_short, fp->fa_mode);
685 * The duplicate information returned in fa_type and fa_mode
686 * is an ambiguity in the NFS version 2 protocol.
688 * VREG should be taken literally as a regular file. If a
689 * server intents to return some type information differently
690 * in the upper bits of the mode field (e.g. for sockets, or
691 * FIFOs), NFSv2 mandates fa_type to be VNON. Anyway, we
692 * leave the examination of the mode bits even in the VREG
693 * case to avoid breakage for bogus servers, but we make sure
694 * that there are actually type bits set in the upper part of
695 * fa_mode (and failing that, trust the va_type field).
697 * NFSv3 cleared the issue, and requires fa_mode to not
698 * contain any type information (while also introduing sockets
699 * and FIFOs for fa_type).
701 if (vtyp == VNON || (vtyp == VREG && (vmode & S_IFMT) != 0))
702 vtyp = IFTOVT(vmode);
703 rdev = fxdr_unsigned(int32_t, fp->fa2_rdev);
704 rmajor = umajor(rdev);
705 rminor = uminor(rdev);
706 fxdr_nfsv2time(&fp->fa2_mtime, &mtime);
709 * Really ugly NFSv2 kludge.
711 if (vtyp == VCHR && rdev == (dev_t)0xffffffff)
716 * If v_type == VNON it is a new node, so fill in the v_type,
717 * n_mtime fields. Check to see if it represents a special
718 * device, and if so, check for a possible alias. Once the
719 * correct vnode has been obtained, fill in the rest of the
723 if (vp->v_type != vtyp) {
724 nfs_setvtype(vp, vtyp);
725 if (vp->v_type == VFIFO) {
726 vp->v_ops = &vp->v_mount->mnt_vn_fifo_ops;
727 } else if (vp->v_type == VCHR || vp->v_type == VBLK) {
728 vp->v_ops = &vp->v_mount->mnt_vn_spec_ops;
729 addaliasu(vp, rmajor, rminor);
731 vp->v_ops = &vp->v_mount->mnt_vn_use_ops;
733 np->n_mtime = mtime.tv_sec;
734 } else if (np->n_mtime != mtime.tv_sec) {
736 * If we haven't modified the file locally and the server
737 * timestamp does not match, then the server probably
738 * modified the file. We must flag this condition so
739 * the proper syncnronization can be done. We do not
740 * try to synchronize the state here because that
741 * could lead to an endless recursion.
743 * XXX loadattrcache can be set during the reply to a write,
744 * before the write timestamp is properly processed. To
745 * avoid unconditionally setting the rmodified bit (which
746 * has the effect of flushing the cache), we only do this
747 * check if the lmodified bit is not set.
749 np->n_mtime = mtime.tv_sec;
750 if ((lattr_flags & NFS_LATTR_NOMTIMECHECK) == 0)
751 np->n_flag |= NRMODIFIED;
755 vap->va_mode = (vmode & 07777);
756 vap->va_rmajor = rmajor;
757 vap->va_rminor = rminor;
758 vap->va_mtime = mtime;
759 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
761 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
762 vap->va_uid = fxdr_unsigned(uid_t, fp->fa_uid);
763 vap->va_gid = fxdr_unsigned(gid_t, fp->fa_gid);
764 vap->va_size = fxdr_hyper(&fp->fa3_size);
765 vap->va_blocksize = NFS_FABLKSIZE;
766 vap->va_bytes = fxdr_hyper(&fp->fa3_used);
767 vap->va_fileid = fxdr_hyper(&fp->fa3_fileid);
768 fxdr_nfsv3time(&fp->fa3_atime, &vap->va_atime);
769 fxdr_nfsv3time(&fp->fa3_ctime, &vap->va_ctime);
773 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
774 vap->va_uid = fxdr_unsigned(uid_t, fp->fa_uid);
775 vap->va_gid = fxdr_unsigned(gid_t, fp->fa_gid);
776 vap->va_size = fxdr_unsigned(u_int32_t, fp->fa2_size);
777 vap->va_blocksize = fxdr_unsigned(int32_t, fp->fa2_blocksize);
778 vap->va_bytes = (u_quad_t)fxdr_unsigned(int32_t, fp->fa2_blocks)
780 vap->va_fileid = fxdr_unsigned(int32_t, fp->fa2_fileid);
781 fxdr_nfsv2time(&fp->fa2_atime, &vap->va_atime);
783 vap->va_ctime.tv_sec = fxdr_unsigned(u_int32_t,
784 fp->fa2_ctime.nfsv2_sec);
785 vap->va_ctime.tv_nsec = 0;
786 vap->va_gen = fxdr_unsigned(u_int32_t,fp->fa2_ctime.nfsv2_usec);
789 np->n_attrstamp = time_uptime;
790 if (vap->va_size != np->n_size) {
791 if (vap->va_type == VREG) {
793 * Get rid of all the junk we had before and just
794 * set NRMODIFIED if NLMODIFIED is 0. Depend on
795 * occassionally flushing our dirty buffers to
796 * clear both the NLMODIFIED and NRMODIFIED flags.
798 if ((np->n_flag & NLMODIFIED) == 0)
799 np->n_flag |= NRMODIFIED;
801 if ((lattr_flags & NFS_LATTR_NOSHRINK) &&
802 vap->va_size < np->n_size) {
804 * We've been told not to shrink the file;
805 * zero np->n_attrstamp to indicate that
806 * the attributes are stale.
808 * This occurs primarily due to recursive
809 * NFS ops that are executed during periods
810 * where we cannot safely reduce the size of
813 * Additionally, write rpcs are broken down
814 * into buffers and np->n_size is
815 * pre-extended. Setting NRMODIFIED here
816 * can result in n_size getting reset to a
817 * lower value, which is NOT what we want.
818 * XXX this needs to be cleaned up a lot
821 vap->va_size = np->n_size;
823 if ((np->n_flag & NLMODIFIED) == 0)
824 np->n_flag |= NRMODIFIED;
825 } else if (np->n_flag & NLMODIFIED) {
827 * We've modified the file: Use the larger
828 * of our size, and the server's size. At
829 * this point the cache coherency is all
830 * shot to hell. To try to handle multiple
831 * clients appending to the file at the same
832 * time mark that the server has changed
833 * the file if the server's notion of the
834 * file size is larger then our notion.
836 * XXX this needs work.
838 if (vap->va_size < np->n_size) {
839 vap->va_size = np->n_size;
841 np->n_size = vap->va_size;
842 np->n_flag |= NRMODIFIED;
846 * Someone changed the file's size on the
847 * server and there are no local changes
848 * to get in the way, set the size and mark
851 np->n_size = vap->va_size;
852 np->n_flag |= NRMODIFIED;
854 nvnode_pager_setsize(vp, np->n_size, XXX);
857 np->n_size = vap->va_size;
861 bcopy((caddr_t)vap, (caddr_t)vaper, sizeof(*vap));
862 if (np->n_flag & NCHG) {
863 if (np->n_flag & NACC)
864 vaper->va_atime = np->n_atim;
865 if (np->n_flag & NUPD)
866 vaper->va_mtime = np->n_mtim;
873 #include <sys/sysctl.h>
874 SYSCTL_DECL(_vfs_nfs);
875 static int nfs_acdebug;
876 SYSCTL_INT(_vfs_nfs, OID_AUTO, acdebug, CTLFLAG_RW, &nfs_acdebug, 0, "");
880 * Check the time stamp
881 * If the cache is valid, copy contents to *vap and return 0
882 * otherwise return an error
885 nfs_getattrcache(struct vnode *vp, struct vattr *vaper)
889 struct nfsmount *nmp;
894 nmp = VFSTONFS(vp->v_mount);
897 * Dynamic timeout based on how recently the file was modified.
898 * n_mtime is always valid.
900 timeo = (get_approximate_time_t() - np->n_mtime) / 60;
904 kprintf("nfs_getattrcache: initial timeo = %d\n", timeo);
907 if (vap->va_type == VDIR) {
908 if ((np->n_flag & NLMODIFIED) || timeo < nmp->nm_acdirmin)
909 timeo = nmp->nm_acdirmin;
910 else if (timeo > nmp->nm_acdirmax)
911 timeo = nmp->nm_acdirmax;
913 if ((np->n_flag & NLMODIFIED) || timeo < nmp->nm_acregmin)
914 timeo = nmp->nm_acregmin;
915 else if (timeo > nmp->nm_acregmax)
916 timeo = nmp->nm_acregmax;
921 kprintf("acregmin %d; acregmax %d; acdirmin %d; acdirmax %d\n",
922 nmp->nm_acregmin, nmp->nm_acregmax,
923 nmp->nm_acdirmin, nmp->nm_acdirmax);
926 kprintf("nfs_getattrcache: age = %d; final timeo = %d\n",
927 (int)(time_uptime - np->n_attrstamp), timeo);
930 if (np->n_attrstamp == 0 || (time_uptime - np->n_attrstamp) >= timeo) {
931 nfsstats.attrcache_misses++;
934 nfsstats.attrcache_hits++;
937 * Our attribute cache can be stale due to modifications made on
938 * this host. XXX this is a bad hack. We need a more deterministic
939 * means of finding out which np fields are valid verses attr cache
940 * fields. We really should update the vattr info on the fly when
941 * making local changes.
943 if (vap->va_size != np->n_size) {
944 if (vap->va_type == VREG) {
945 if (np->n_flag & NLMODIFIED)
946 vap->va_size = np->n_size;
947 nfs_meta_setsize(vp, curthread, vap->va_size, 0);
949 np->n_size = vap->va_size;
952 bcopy((caddr_t)vap, (caddr_t)vaper, sizeof(struct vattr));
953 if (np->n_flag & NCHG) {
954 if (np->n_flag & NACC)
955 vaper->va_atime = np->n_atim;
956 if (np->n_flag & NUPD)
957 vaper->va_mtime = np->n_mtim;
965 * Set up nameidata for a lookup() call and do it.
967 * If pubflag is set, this call is done for a lookup operation on the
968 * public filehandle. In that case we allow crossing mountpoints and
969 * absolute pathnames. However, the caller is expected to check that
970 * the lookup result is within the public fs, and deny access if
973 * dirp may be set whether an error is returned or not, and must be
974 * released by the caller.
976 * On return nd->nl_nch usually points to the target ncp, which may represent
979 * NOTE: the caller must call nlookup_done(nd) unconditionally on return
983 nfs_namei(struct nlookupdata *nd, struct ucred *cred, int nflags,
984 struct vnode **dvpp, struct vnode **vpp,
985 fhandle_t *fhp, int len,
986 struct nfssvc_sock *slp, struct sockaddr *nam, struct mbuf **mdp,
987 caddr_t *dposp, struct vnode **dirpp, struct thread *td,
988 int kerbflag, int pubflag)
992 char *fromcp, *tocp, *cp;
1003 if (nd->nl_flags & NLC_HASBUF) {
1004 namebuf = nd->nl_path;
1005 nd->nl_flags &= ~NLC_HASBUF;
1009 namebuf = objcache_get(namei_oc, M_WAITOK);
1015 * Copy the name from the mbuf list to namebuf.
1021 rem = mtod(md, caddr_t) + md->m_len - fromcp;
1023 for (i = 0; i < len; i++) {
1030 fromcp = mtod(md, caddr_t);
1033 if (*fromcp == '\0' || (!pubflag && *fromcp == '/')) {
1037 *tocp++ = *fromcp++;
1043 len = nfsm_rndup(len)-len;
1047 else if ((error = nfs_adv(mdp, dposp, len, rem)) != 0)
1053 * Extract and set starting directory. The returned dp is refd
1056 error = nfsrv_fhtovp(fhp, FALSE, &mp, &dp, cred, slp,
1057 nam, &rdonly, kerbflag, pubflag);
1060 if (dp->v_type != VDIR) {
1067 * Set return directory. Reference to dp is implicitly transfered
1068 * to the returned pointer. This must be set before we potentially
1074 * read-only - NLC_DELETE, NLC_RENAME_DST are disallowed. NLC_CREATE
1075 * is passed through to nlookup() and will be disallowed
1076 * if the file does not already exist.
1079 nflags |= NLC_NFS_RDONLY;
1080 if (nflags & (NLC_DELETE | NLC_RENAME_DST)) {
1087 * Oh joy. For WebNFS, handle those pesky '%' escapes,
1088 * and the 'native path' indicator.
1091 cp = objcache_get(namei_oc, M_WAITOK);
1094 if ((unsigned char)*fromcp >= WEBNFS_SPECCHAR_START) {
1095 switch ((unsigned char)*fromcp) {
1096 case WEBNFS_NATIVE_CHAR:
1098 * 'Native' path for us is the same
1099 * as a path according to the NFS spec,
1100 * just skip the escape char.
1105 * More may be added in the future, range 0x80-0xff
1109 objcache_put(namei_oc, cp);
1114 * Translate the '%' escapes, URL-style.
1116 while (*fromcp != '\0') {
1117 if (*fromcp == WEBNFS_ESC_CHAR) {
1118 if (fromcp[1] != '\0' && fromcp[2] != '\0') {
1120 *tocp++ = HEXSTRTOI(fromcp);
1125 objcache_put(namei_oc, cp);
1129 *tocp++ = *fromcp++;
1132 objcache_put(namei_oc, namebuf);
1137 * Setup for search. We need to get a start directory from dp. Note
1138 * that dp is ref'd, but we no longer 'own' the ref (*dirpp owns it).
1141 nflags |= NLC_NFS_NOSOFTLINKTRAV;
1142 nflags |= NLC_NOCROSSMOUNT;
1146 * We need a starting ncp from the directory vnode dp. dp must not
1147 * be locked. The returned ncp will be refd but not locked.
1149 * If no suitable ncp is found we instruct cache_fromdvp() to create
1150 * one. If this fails the directory has probably been removed while
1151 * the target was chdir'd into it and any further lookup will fail.
1153 if ((error = cache_fromdvp(dp, cred, 1, &nch)) != 0)
1155 nlookup_init_raw(nd, namebuf, UIO_SYSSPACE, nflags, cred, &nch);
1159 * Ok, do the lookup.
1161 error = nlookup(nd);
1164 * If no error occured return the requested dvpp and vpp. If
1165 * NLC_CREATE was specified nd->nl_nch may represent a negative
1166 * cache hit in which case we do not attempt to obtain the vp.
1170 if (nd->nl_nch.ncp->nc_parent) {
1172 nch.ncp = nch.ncp->nc_parent;
1175 error = cache_vget(&nch, nd->nl_cred,
1176 LK_EXCLUSIVE, dvpp);
1182 if (vpp && nd->nl_nch.ncp->nc_vp) {
1183 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, vpp);
1186 if (dvpp && *dvpp) {
1201 objcache_put(namei_oc, namebuf);
1206 * nfsrv_fhtovp() - convert a fh to a vnode ptr (optionally locked)
1207 * - look up fsid in mount list (if not found ret error)
1208 * - get vp and export rights by calling VFS_FHTOVP()
1209 * - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
1210 * - if not lockflag unlock it with vn_unlock()
1213 nfsrv_fhtovp(fhandle_t *fhp, int lockflag,
1214 struct mount **mpp, struct vnode **vpp,
1215 struct ucred *cred, struct nfssvc_sock *slp, struct sockaddr *nam,
1216 int *rdonlyp, int kerbflag, int pubflag)
1220 struct ucred *credanon;
1222 #ifdef MNT_EXNORESPORT /* XXX needs mountd and /etc/exports help yet */
1223 struct sockaddr_int *saddr;
1229 if (nfs_ispublicfh(fhp)) {
1230 if (!pubflag || !nfs_pub.np_valid)
1232 fhp = &nfs_pub.np_handle;
1235 mp = *mpp = vfs_getvfs(&fhp->fh_fsid);
1238 error = VFS_CHECKEXP(mp, nam, &exflags, &credanon);
1243 error = VFS_FHTOVP(mp, NULL, &fhp->fh_fid, vpp);
1247 #ifdef MNT_EXNORESPORT
1248 if (!(exflags & (MNT_EXNORESPORT|MNT_EXPUBLIC))) {
1249 saddr = (struct sockaddr_in *)nam;
1250 if (saddr->sin_family == AF_INET &&
1251 ntohs(saddr->sin_port) >= IPPORT_RESERVED) {
1254 return (NFSERR_AUTHERR | AUTH_TOOWEAK);
1259 * Check/setup credentials.
1261 if (exflags & MNT_EXKERB) {
1265 return (NFSERR_AUTHERR | AUTH_TOOWEAK);
1267 } else if (kerbflag) {
1270 return (NFSERR_AUTHERR | AUTH_TOOWEAK);
1271 } else if (cred->cr_uid == 0 || (exflags & MNT_EXPORTANON)) {
1272 cred->cr_uid = credanon->cr_uid;
1273 for (i = 0; i < credanon->cr_ngroups && i < NGROUPS; i++)
1274 cred->cr_groups[i] = credanon->cr_groups[i];
1275 cred->cr_ngroups = i;
1277 if (exflags & MNT_EXRDONLY)
1288 * WebNFS: check if a filehandle is a public filehandle. For v3, this
1289 * means a length of 0, for v2 it means all zeroes. nfsm_srvmtofh has
1290 * transformed this to all zeroes in both cases, so check for it.
1293 nfs_ispublicfh(fhandle_t *fhp)
1295 char *cp = (char *)fhp;
1298 for (i = 0; i < NFSX_V3FH; i++)
1304 #endif /* NFS_NOSERVER */
1306 * This function compares two net addresses by family and returns TRUE
1307 * if they are the same host.
1308 * If there is any doubt, return FALSE.
1309 * The AF_INET family is handled as a special case so that address mbufs
1310 * don't need to be saved to store "struct in_addr", which is only 4 bytes.
1313 netaddr_match(int family, union nethostaddr *haddr, struct sockaddr *nam)
1315 struct sockaddr_in *inetaddr;
1319 inetaddr = (struct sockaddr_in *)nam;
1320 if (inetaddr->sin_family == AF_INET &&
1321 inetaddr->sin_addr.s_addr == haddr->had_inetaddr)
1330 static nfsuint64 nfs_nullcookie = { { 0, 0 } };
1332 * This function finds the directory cookie that corresponds to the
1333 * logical byte offset given.
1336 nfs_getcookie(struct nfsnode *np, off_t off, int add)
1338 struct nfsdmap *dp, *dp2;
1341 pos = (uoff_t)off / NFS_DIRBLKSIZ;
1342 if (pos == 0 || off < 0) {
1345 panic("nfs getcookie add at <= 0");
1347 return (&nfs_nullcookie);
1350 dp = np->n_cookies.lh_first;
1353 dp = kmalloc(sizeof(struct nfsdmap), M_NFSDIROFF,
1355 dp->ndm_eocookie = 0;
1356 LIST_INSERT_HEAD(&np->n_cookies, dp, ndm_list);
1360 while (pos >= NFSNUMCOOKIES) {
1361 pos -= NFSNUMCOOKIES;
1362 if (dp->ndm_list.le_next) {
1363 if (!add && dp->ndm_eocookie < NFSNUMCOOKIES &&
1364 pos >= dp->ndm_eocookie)
1366 dp = dp->ndm_list.le_next;
1368 dp2 = kmalloc(sizeof(struct nfsdmap), M_NFSDIROFF,
1370 dp2->ndm_eocookie = 0;
1371 LIST_INSERT_AFTER(dp, dp2, ndm_list);
1376 if (pos >= dp->ndm_eocookie) {
1378 dp->ndm_eocookie = pos + 1;
1382 return (&dp->ndm_cookies[pos]);
1386 * Invalidate cached directory information, except for the actual directory
1387 * blocks (which are invalidated separately).
1388 * Done mainly to avoid the use of stale offset cookies.
1391 nfs_invaldir(struct vnode *vp)
1393 struct nfsnode *np = VTONFS(vp);
1396 if (vp->v_type != VDIR)
1397 panic("nfs: invaldir not dir");
1399 np->n_direofoffset = 0;
1400 np->n_cookieverf.nfsuquad[0] = 0;
1401 np->n_cookieverf.nfsuquad[1] = 0;
1402 if (np->n_cookies.lh_first)
1403 np->n_cookies.lh_first->ndm_eocookie = 0;
1407 * Set the v_type field for an NFS client's vnode and initialize for
1408 * buffer cache operations if necessary.
1411 nfs_setvtype(struct vnode *vp, enum vtype vtyp)
1420 * Needs VMIO, size not yet known, and blocksize
1421 * is not really relevant if we are passing a
1424 vinitvmio(vp, 0, PAGE_SIZE, -1);
1432 * The write verifier has changed (probably due to a server reboot), so all
1433 * B_NEEDCOMMIT blocks will have to be written again. Since they are on the
1434 * dirty block list as B_DELWRI, all this takes is clearing the B_NEEDCOMMIT
1435 * and B_CLUSTEROK flags. Once done the new write verifier can be set for the
1438 * B_CLUSTEROK must be cleared along with B_NEEDCOMMIT because stage 1 data
1439 * writes are not clusterable.
1442 static int nfs_clearcommit_bp(struct buf *bp, void *data __unused);
1443 static int nfs_clearcommit_callback(struct mount *mp, struct vnode *vp,
1444 void *data __unused);
1447 nfs_clearcommit(struct mount *mp)
1449 vsyncscan(mp, VMSC_NOWAIT, nfs_clearcommit_callback, NULL);
1453 nfs_clearcommit_callback(struct mount *mp, struct vnode *vp,
1454 void *data __unused)
1456 lwkt_gettoken(&vp->v_token);
1457 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
1458 nfs_clearcommit_bp, NULL);
1459 lwkt_reltoken(&vp->v_token);
1465 nfs_clearcommit_bp(struct buf *bp, void *data __unused)
1467 if (BUF_LOCKINUSE(bp) == 0 &&
1468 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
1469 == (B_DELWRI | B_NEEDCOMMIT)) {
1470 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
1475 #ifndef NFS_NOSERVER
1477 * Map errnos to NFS error numbers. For Version 3 also filter out error
1478 * numbers not specified for the associated procedure.
1481 nfsrv_errmap(struct nfsrv_descript *nd, int err)
1483 short *defaulterrp, *errp;
1485 if (nd->nd_flag & ND_NFSV3) {
1486 if (nd->nd_procnum <= NFSPROC_COMMIT) {
1487 errp = defaulterrp = nfsrv_v3errmap[nd->nd_procnum];
1491 else if (*errp > err)
1494 return ((int)*defaulterrp);
1496 return (err & 0xffff);
1499 return ((int)nfsrv_v2errmap[err - 1]);
1504 * Sort the group list in increasing numerical order.
1505 * (Insertion sort by Chris Torek, who was grossed out by the bubble sort
1506 * that used to be here.)
1509 nfsrvw_sort(gid_t *list, int num)
1514 /* Insertion sort. */
1515 for (i = 1; i < num; i++) {
1517 /* find correct slot for value v, moving others up */
1518 for (j = i; --j >= 0 && v < list[j];)
1519 list[j + 1] = list[j];
1525 * copy credentials making sure that the result can be compared with bcmp().
1528 nfsrv_setcred(struct ucred *incred, struct ucred *outcred)
1532 bzero((caddr_t)outcred, sizeof (struct ucred));
1533 outcred->cr_ref = 1;
1534 outcred->cr_uid = incred->cr_uid;
1535 outcred->cr_ngroups = incred->cr_ngroups;
1536 for (i = 0; i < incred->cr_ngroups; i++)
1537 outcred->cr_groups[i] = incred->cr_groups[i];
1538 nfsrvw_sort(outcred->cr_groups, outcred->cr_ngroups);
1540 #endif /* NFS_NOSERVER */
1543 * Hold a ucred in nfs_node. Discard prison information, otherwise
1544 * prisons might stick around indefinitely due to NFS node caching.
1547 nfs_crhold(struct ucred *cred)
1550 if (cred->cr_prison) {
1552 prison_free(cred->cr_prison);
1553 cred->cr_prison = NULL;
1555 cred = crhold(cred);
1562 * Return whether two ucreds are the same insofar as NFS cares about.
1565 nfs_crsame(struct ucred *cr1, struct ucred *cr2)
1568 if (cr1 == NULL || cr2 == NULL)
1569 return (cr1 == cr2);
1570 if (cr1->cr_uid != cr2->cr_uid ||
1571 cr1->cr_ruid != cr2->cr_ruid ||
1572 cr1->cr_rgid != cr2->cr_rgid ||
1573 cr1->cr_ngroups != cr2->cr_ngroups) {
1576 if (bcmp(cr1->cr_groups, cr2->cr_groups,
1577 cr1->cr_ngroups * sizeof(cr1->cr_groups[0])) != 0) {