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>
49 #include <sys/namei.h>
51 #include <sys/socket.h>
53 #include <sys/malloc.h>
54 #include <sys/sysent.h>
55 #include <sys/syscall.h>
56 #include <sys/sysproto.h>
58 #include <sys/objcache.h>
62 #include <vm/vm_object.h>
63 #include <vm/vm_extern.h>
73 #include "nfsm_subs.h"
76 #include <netinet/in.h>
78 MALLOC_DEFINE(M_NFSMOUNT, "NFS mount", "NFS mount");
81 * Data items converted to xdr at startup, since they are constant
82 * This is kinda hokey, but may save a little time doing byte swaps
84 u_int32_t nfs_xdrneg1;
85 u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers;
86 u_int32_t rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr;
87 u_int32_t rpc_auth_kerb;
88 u_int32_t nfs_prog, nfs_true, nfs_false;
90 /* And other global data */
91 static enum vtype nv2tov_type[8]= {
92 VNON, VREG, VDIR, VBLK, VCHR, VLNK, VNON, VNON
94 enum vtype nv3tov_type[8]= {
95 VNON, VREG, VDIR, VBLK, VCHR, VLNK, VSOCK, VFIFO
101 * Protect master lists only. Primary protection uses the per-mount
102 * and per nfssvc_sock tokens.
104 struct lwkt_token nfs_token = LWKT_TOKEN_INITIALIZER(unp_token);
106 static long nfs_pbuf_freecnt = -1; /* start out unlimited */
108 struct nfsmount_head nfs_mountq = TAILQ_HEAD_INITIALIZER(nfs_mountq);
109 struct nfssvc_sockhead nfssvc_sockhead;
110 int nfssvc_sockhead_flag;
111 struct nfsd_head nfsd_head;
113 struct nfs_bufq nfs_bufq;
114 struct nqfhhashhead *nqfhhashtbl;
117 static int nfs_prev_nfssvc_sy_narg;
118 static sy_call_t *nfs_prev_nfssvc_sy_call;
123 * Mapping of old NFS Version 2 RPC numbers to generic numbers.
125 int nfsv3_procid[NFS_NPROCS] = {
154 #endif /* NFS_NOSERVER */
156 * and the reverse mapping from generic to Version 2 procedure numbers
158 int nfsv2_procid[NFS_NPROCS] = {
189 * Maps errno values to nfs error numbers.
190 * Use NFSERR_IO as the catch all for ones not specifically defined in
193 static u_char nfsrv_v2errmap[ELAST] = {
194 NFSERR_PERM, NFSERR_NOENT, NFSERR_IO, NFSERR_IO, NFSERR_IO,
195 NFSERR_NXIO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
196 NFSERR_IO, NFSERR_IO, NFSERR_ACCES, NFSERR_IO, NFSERR_IO,
197 NFSERR_IO, NFSERR_EXIST, NFSERR_IO, NFSERR_NODEV, NFSERR_NOTDIR,
198 NFSERR_ISDIR, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
199 NFSERR_IO, NFSERR_FBIG, NFSERR_NOSPC, NFSERR_IO, NFSERR_ROFS,
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_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_NAMETOL, NFSERR_IO, NFSERR_IO,
207 NFSERR_NOTEMPTY, NFSERR_IO, NFSERR_IO, NFSERR_DQUOT, NFSERR_STALE,
208 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
209 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
210 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
211 NFSERR_IO /* << Last is 86 */
215 * Maps errno values to nfs error numbers.
216 * Although it is not obvious whether or not NFS clients really care if
217 * a returned error value is in the specified list for the procedure, the
218 * safest thing to do is filter them appropriately. For Version 2, the
219 * X/Open XNFS document is the only specification that defines error values
220 * for each RPC (The RFC simply lists all possible error values for all RPCs),
221 * so I have decided to not do this for Version 2.
222 * The first entry is the default error return and the rest are the valid
223 * errors for that RPC in increasing numeric order.
225 static short nfsv3err_null[] = {
230 static short nfsv3err_getattr[] = {
239 static short nfsv3err_setattr[] = {
255 static short nfsv3err_lookup[] = {
268 static short nfsv3err_access[] = {
277 static short nfsv3err_readlink[] = {
289 static short nfsv3err_read[] = {
301 static short nfsv3err_write[] = {
316 static short nfsv3err_create[] = {
333 static short nfsv3err_mkdir[] = {
350 static short nfsv3err_symlink[] = {
367 static short nfsv3err_mknod[] = {
385 static short nfsv3err_remove[] = {
399 static short nfsv3err_rmdir[] = {
417 static short nfsv3err_rename[] = {
440 static short nfsv3err_link[] = {
460 static short nfsv3err_readdir[] = {
473 static short nfsv3err_readdirplus[] = {
487 static short nfsv3err_fsstat[] = {
496 static short nfsv3err_fsinfo[] = {
504 static short nfsv3err_pathconf[] = {
512 static short nfsv3err_commit[] = {
521 static short *nfsrv_v3errmap[] = {
539 nfsv3err_readdirplus,
546 #endif /* NFS_NOSERVER */
549 * This needs to return a monotonically increasing or close to monotonically
550 * increasing result, otherwise the write gathering queues won't work
559 return ((u_quad_t)tv.tv_sec * 1000000 + (u_quad_t)tv.tv_usec);
563 * Called once to initialize data structures...
566 nfs_init(struct vfsconf *vfsp)
568 callout_init_mp(&nfs_timer_handle);
569 nfsmount_objcache = objcache_create_simple(M_NFSMOUNT, sizeof(struct nfsmount));
571 nfs_mount_type = vfsp->vfc_typenum;
573 rpc_vers = txdr_unsigned(RPC_VER2);
574 rpc_call = txdr_unsigned(RPC_CALL);
575 rpc_reply = txdr_unsigned(RPC_REPLY);
576 rpc_msgdenied = txdr_unsigned(RPC_MSGDENIED);
577 rpc_msgaccepted = txdr_unsigned(RPC_MSGACCEPTED);
578 rpc_mismatch = txdr_unsigned(RPC_MISMATCH);
579 rpc_autherr = txdr_unsigned(RPC_AUTHERR);
580 rpc_auth_unix = txdr_unsigned(RPCAUTH_UNIX);
581 rpc_auth_kerb = txdr_unsigned(RPCAUTH_KERB4);
582 nfs_prog = txdr_unsigned(NFS_PROG);
583 nfs_true = txdr_unsigned(TRUE);
584 nfs_false = txdr_unsigned(FALSE);
585 nfs_xdrneg1 = txdr_unsigned(-1);
586 nfs_ticks = (hz * NFS_TICKINTVL + 500) / 1000;
589 nfs_nhinit(); /* Init the nfsnode table */
591 nfsrv_init(0); /* Init server data structures */
592 nfsrv_initcache(); /* Init the server request cache */
596 * Mainly for vkernel operation. If memory is severely limited
598 if (nfs_maxasyncbio > nmbclusters * MCLBYTES / NFS_MAXDATA / 3)
599 nfs_maxasyncbio = nmbclusters * MCLBYTES / NFS_MAXDATA / 3;
600 if (nfs_maxasyncbio < 4)
604 * Initialize reply list and start timer
606 nfs_timer_callout(0);
608 #if 1 /* XXX this isn't really needed */
609 nfs_prev_nfssvc_sy_narg = sysent[SYS_nfssvc].sy_narg;
610 sysent[SYS_nfssvc].sy_narg = 2;
612 nfs_prev_nfssvc_sy_call = sysent[SYS_nfssvc].sy_call;
613 sysent[SYS_nfssvc].sy_call = (sy_call_t *)sys_nfssvc;
615 nfs_pbuf_freecnt = nswbuf_kva / 2 + 1;
621 nfs_uninit(struct vfsconf *vfsp)
623 callout_stop(&nfs_timer_handle);
625 sysent[SYS_nfssvc].sy_narg = nfs_prev_nfssvc_sy_narg;
626 sysent[SYS_nfssvc].sy_call = nfs_prev_nfssvc_sy_call;
627 nfs_nhdestroy(); /* Destroy the nfsnode table */
629 nfsrv_destroycache(); /* Destroy the server request cache */
631 objcache_destroy(nfsmount_objcache);
636 * Attribute cache routines.
637 * nfs_loadattrcache() - loads or updates the cache contents from attributes
638 * that are on the mbuf list
639 * nfs_getattrcache() - returns valid attributes if found in cache, returns
644 * Load the attribute cache (that lives in the nfsnode entry) with
645 * the values on the mbuf list. Load *vaper with the attributes. vaper
648 * As a side effect n_mtime, which we use to determine if the file was
649 * modified by some other host, is set to the attribute timestamp and
650 * NRMODIFIED is set if the two values differ.
652 * WARNING: the mtime loaded into vaper does not necessarily represent
653 * n_mtime or n_attr.mtime due to NACC and NUPD.
656 nfs_loadattrcache(struct vnode *vp, struct mbuf **mdp, caddr_t *dposp,
657 struct vattr *vaper, int lattr_flags)
660 struct nfs_fattr *fp;
670 struct timespec mtime;
671 int v3 = NFS_ISV3(vp);
674 t1 = (mtod(md, caddr_t) + md->m_len) - *dposp;
675 if ((error = nfsm_disct(mdp, dposp, NFSX_FATTR(v3), t1, &cp2)) != 0)
677 fp = (struct nfs_fattr *)cp2;
679 vtyp = nfsv3tov_type(fp->fa_type);
680 vmode = fxdr_unsigned(u_short, fp->fa_mode);
681 rmajor = (int)fxdr_unsigned(int, fp->fa3_rdev.specdata1);
682 rminor = (int)fxdr_unsigned(int, fp->fa3_rdev.specdata2);
683 fxdr_nfsv3time(&fp->fa3_mtime, &mtime);
685 vtyp = nfsv2tov_type(fp->fa_type);
686 vmode = fxdr_unsigned(u_short, fp->fa_mode);
690 * The duplicate information returned in fa_type and fa_mode
691 * is an ambiguity in the NFS version 2 protocol.
693 * VREG should be taken literally as a regular file. If a
694 * server intents to return some type information differently
695 * in the upper bits of the mode field (e.g. for sockets, or
696 * FIFOs), NFSv2 mandates fa_type to be VNON. Anyway, we
697 * leave the examination of the mode bits even in the VREG
698 * case to avoid breakage for bogus servers, but we make sure
699 * that there are actually type bits set in the upper part of
700 * fa_mode (and failing that, trust the va_type field).
702 * NFSv3 cleared the issue, and requires fa_mode to not
703 * contain any type information (while also introduing sockets
704 * and FIFOs for fa_type).
706 if (vtyp == VNON || (vtyp == VREG && (vmode & S_IFMT) != 0))
707 vtyp = IFTOVT(vmode);
708 rdev = fxdr_unsigned(int32_t, fp->fa2_rdev);
709 rmajor = umajor(rdev);
710 rminor = uminor(rdev);
711 fxdr_nfsv2time(&fp->fa2_mtime, &mtime);
714 * Really ugly NFSv2 kludge.
716 if (vtyp == VCHR && rdev == (dev_t)0xffffffff)
721 * If v_type == VNON it is a new node, so fill in the v_type,
722 * n_mtime fields. Check to see if it represents a special
723 * device, and if so, check for a possible alias. Once the
724 * correct vnode has been obtained, fill in the rest of the
728 if (vp->v_type != vtyp) {
729 nfs_setvtype(vp, vtyp);
730 if (vp->v_type == VFIFO) {
731 vp->v_ops = &vp->v_mount->mnt_vn_fifo_ops;
732 } else if (vp->v_type == VCHR || vp->v_type == VBLK) {
733 vp->v_ops = &vp->v_mount->mnt_vn_spec_ops;
734 addaliasu(vp, rmajor, rminor);
736 vp->v_ops = &vp->v_mount->mnt_vn_use_ops;
738 np->n_mtime = mtime.tv_sec;
739 } else if (np->n_mtime != mtime.tv_sec) {
741 * If we haven't modified the file locally and the server
742 * timestamp does not match, then the server probably
743 * modified the file. We must flag this condition so
744 * the proper syncnronization can be done. We do not
745 * try to synchronize the state here because that
746 * could lead to an endless recursion.
748 * XXX loadattrcache can be set during the reply to a write,
749 * before the write timestamp is properly processed. To
750 * avoid unconditionally setting the rmodified bit (which
751 * has the effect of flushing the cache), we only do this
752 * check if the lmodified bit is not set.
754 np->n_mtime = mtime.tv_sec;
755 if ((lattr_flags & NFS_LATTR_NOMTIMECHECK) == 0)
756 np->n_flag |= NRMODIFIED;
760 vap->va_mode = (vmode & 07777);
761 vap->va_rmajor = rmajor;
762 vap->va_rminor = rminor;
763 vap->va_mtime = mtime;
764 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
766 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
767 vap->va_uid = fxdr_unsigned(uid_t, fp->fa_uid);
768 vap->va_gid = fxdr_unsigned(gid_t, fp->fa_gid);
769 vap->va_size = fxdr_hyper(&fp->fa3_size);
770 vap->va_blocksize = NFS_FABLKSIZE;
771 vap->va_bytes = fxdr_hyper(&fp->fa3_used);
772 vap->va_fileid = fxdr_hyper(&fp->fa3_fileid);
773 fxdr_nfsv3time(&fp->fa3_atime, &vap->va_atime);
774 fxdr_nfsv3time(&fp->fa3_ctime, &vap->va_ctime);
778 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
779 vap->va_uid = fxdr_unsigned(uid_t, fp->fa_uid);
780 vap->va_gid = fxdr_unsigned(gid_t, fp->fa_gid);
781 vap->va_size = fxdr_unsigned(u_int32_t, fp->fa2_size);
782 vap->va_blocksize = fxdr_unsigned(int32_t, fp->fa2_blocksize);
783 vap->va_bytes = (u_quad_t)fxdr_unsigned(int32_t, fp->fa2_blocks)
785 vap->va_fileid = fxdr_unsigned(int32_t, fp->fa2_fileid);
786 fxdr_nfsv2time(&fp->fa2_atime, &vap->va_atime);
788 vap->va_ctime.tv_sec = fxdr_unsigned(u_int32_t,
789 fp->fa2_ctime.nfsv2_sec);
790 vap->va_ctime.tv_nsec = 0;
791 vap->va_gen = fxdr_unsigned(u_int32_t,fp->fa2_ctime.nfsv2_usec);
794 np->n_attrstamp = time_uptime;
795 if (vap->va_size != np->n_size) {
796 if (vap->va_type == VREG) {
798 * Get rid of all the junk we had before and just
799 * set NRMODIFIED if NLMODIFIED is 0. Depend on
800 * occassionally flushing our dirty buffers to
801 * clear both the NLMODIFIED and NRMODIFIED flags.
803 if ((np->n_flag & NLMODIFIED) == 0)
804 np->n_flag |= NRMODIFIED;
806 if ((lattr_flags & NFS_LATTR_NOSHRINK) &&
807 vap->va_size < np->n_size) {
809 * We've been told not to shrink the file;
810 * zero np->n_attrstamp to indicate that
811 * the attributes are stale.
813 * This occurs primarily due to recursive
814 * NFS ops that are executed during periods
815 * where we cannot safely reduce the size of
818 * Additionally, write rpcs are broken down
819 * into buffers and np->n_size is
820 * pre-extended. Setting NRMODIFIED here
821 * can result in n_size getting reset to a
822 * lower value, which is NOT what we want.
823 * XXX this needs to be cleaned up a lot
826 vap->va_size = np->n_size;
828 if ((np->n_flag & NLMODIFIED) == 0)
829 np->n_flag |= NRMODIFIED;
830 } else if (np->n_flag & NLMODIFIED) {
832 * We've modified the file: Use the larger
833 * of our size, and the server's size. At
834 * this point the cache coherency is all
835 * shot to hell. To try to handle multiple
836 * clients appending to the file at the same
837 * time mark that the server has changed
838 * the file if the server's notion of the
839 * file size is larger then our notion.
841 * XXX this needs work.
843 if (vap->va_size < np->n_size) {
844 vap->va_size = np->n_size;
846 np->n_size = vap->va_size;
847 np->n_flag |= NRMODIFIED;
851 * Someone changed the file's size on the
852 * server and there are no local changes
853 * to get in the way, set the size and mark
856 np->n_size = vap->va_size;
857 np->n_flag |= NRMODIFIED;
859 nvnode_pager_setsize(vp, np->n_size, XXX);
862 np->n_size = vap->va_size;
866 bcopy((caddr_t)vap, (caddr_t)vaper, sizeof(*vap));
867 if (np->n_flag & NCHG) {
868 if (np->n_flag & NACC)
869 vaper->va_atime = np->n_atim;
870 if (np->n_flag & NUPD)
871 vaper->va_mtime = np->n_mtim;
878 #include <sys/sysctl.h>
879 SYSCTL_DECL(_vfs_nfs);
880 static int nfs_acdebug;
881 SYSCTL_INT(_vfs_nfs, OID_AUTO, acdebug, CTLFLAG_RW, &nfs_acdebug, 0, "");
885 * Check the time stamp
886 * If the cache is valid, copy contents to *vap and return 0
887 * otherwise return an error
890 nfs_getattrcache(struct vnode *vp, struct vattr *vaper)
894 struct nfsmount *nmp;
899 nmp = VFSTONFS(vp->v_mount);
902 * Dynamic timeout based on how recently the file was modified.
903 * n_mtime is always valid.
905 timeo = (get_approximate_time_t() - np->n_mtime) / 60;
909 kprintf("nfs_getattrcache: initial timeo = %d\n", timeo);
912 if (vap->va_type == VDIR) {
913 if ((np->n_flag & NLMODIFIED) || timeo < nmp->nm_acdirmin)
914 timeo = nmp->nm_acdirmin;
915 else if (timeo > nmp->nm_acdirmax)
916 timeo = nmp->nm_acdirmax;
918 if ((np->n_flag & NLMODIFIED) || timeo < nmp->nm_acregmin)
919 timeo = nmp->nm_acregmin;
920 else if (timeo > nmp->nm_acregmax)
921 timeo = nmp->nm_acregmax;
926 kprintf("acregmin %d; acregmax %d; acdirmin %d; acdirmax %d\n",
927 nmp->nm_acregmin, nmp->nm_acregmax,
928 nmp->nm_acdirmin, nmp->nm_acdirmax);
931 kprintf("nfs_getattrcache: age = %d; final timeo = %d\n",
932 (int)(time_uptime - np->n_attrstamp), timeo);
935 if (np->n_attrstamp == 0 || (time_uptime - np->n_attrstamp) >= timeo) {
936 nfsstats.attrcache_misses++;
939 nfsstats.attrcache_hits++;
942 * Our attribute cache can be stale due to modifications made on
943 * this host. XXX this is a bad hack. We need a more deterministic
944 * means of finding out which np fields are valid verses attr cache
945 * fields. We really should update the vattr info on the fly when
946 * making local changes.
948 if (vap->va_size != np->n_size) {
949 if (vap->va_type == VREG) {
950 if (np->n_flag & NLMODIFIED)
951 vap->va_size = np->n_size;
952 nfs_meta_setsize(vp, curthread, vap->va_size, 0);
954 np->n_size = vap->va_size;
957 bcopy((caddr_t)vap, (caddr_t)vaper, sizeof(struct vattr));
958 if (np->n_flag & NCHG) {
959 if (np->n_flag & NACC)
960 vaper->va_atime = np->n_atim;
961 if (np->n_flag & NUPD)
962 vaper->va_mtime = np->n_mtim;
970 * Set up nameidata for a lookup() call and do it.
972 * If pubflag is set, this call is done for a lookup operation on the
973 * public filehandle. In that case we allow crossing mountpoints and
974 * absolute pathnames. However, the caller is expected to check that
975 * the lookup result is within the public fs, and deny access if
978 * dirp may be set whether an error is returned or not, and must be
979 * released by the caller.
981 * On return nd->nl_nch usually points to the target ncp, which may represent
984 * NOTE: the caller must call nlookup_done(nd) unconditionally on return
988 nfs_namei(struct nlookupdata *nd, struct ucred *cred, int nflags,
989 struct vnode **dvpp, struct vnode **vpp,
990 fhandle_t *fhp, int len,
991 struct nfssvc_sock *slp, struct sockaddr *nam, struct mbuf **mdp,
992 caddr_t *dposp, struct vnode **dirpp, struct thread *td,
993 int kerbflag, int pubflag)
997 char *fromcp, *tocp, *cp;
1004 namebuf = objcache_get(namei_oc, M_WAITOK);
1008 * Copy the name from the mbuf list to namebuf.
1013 rem = mtod(md, caddr_t) + md->m_len - fromcp;
1014 for (i = 0; i < len; i++) {
1021 fromcp = mtod(md, caddr_t);
1024 if (*fromcp == '\0' || (!pubflag && *fromcp == '/')) {
1028 *tocp++ = *fromcp++;
1034 len = nfsm_rndup(len)-len;
1038 else if ((error = nfs_adv(mdp, dposp, len, rem)) != 0)
1043 * Extract and set starting directory. The returned dp is refd
1046 error = nfsrv_fhtovp(fhp, FALSE, &mp, &dp, cred, slp,
1047 nam, &rdonly, kerbflag, pubflag);
1050 if (dp->v_type != VDIR) {
1057 * Set return directory. Reference to dp is implicitly transfered
1058 * to the returned pointer. This must be set before we potentially
1064 * read-only - NLC_DELETE, NLC_RENAME_DST are disallowed. NLC_CREATE
1065 * is passed through to nlookup() and will be disallowed
1066 * if the file does not already exist.
1069 nflags |= NLC_NFS_RDONLY;
1070 if (nflags & (NLC_DELETE | NLC_RENAME_DST)) {
1077 * Oh joy. For WebNFS, handle those pesky '%' escapes,
1078 * and the 'native path' indicator.
1081 cp = objcache_get(namei_oc, M_WAITOK);
1084 if ((unsigned char)*fromcp >= WEBNFS_SPECCHAR_START) {
1085 switch ((unsigned char)*fromcp) {
1086 case WEBNFS_NATIVE_CHAR:
1088 * 'Native' path for us is the same
1089 * as a path according to the NFS spec,
1090 * just skip the escape char.
1095 * More may be added in the future, range 0x80-0xff
1099 objcache_put(namei_oc, cp);
1104 * Translate the '%' escapes, URL-style.
1106 while (*fromcp != '\0') {
1107 if (*fromcp == WEBNFS_ESC_CHAR) {
1108 if (fromcp[1] != '\0' && fromcp[2] != '\0') {
1110 *tocp++ = HEXSTRTOI(fromcp);
1115 objcache_put(namei_oc, cp);
1119 *tocp++ = *fromcp++;
1122 objcache_put(namei_oc, namebuf);
1127 * Setup for search. We need to get a start directory from dp. Note
1128 * that dp is ref'd, but we no longer 'own' the ref (*dirpp owns it).
1131 nflags |= NLC_NFS_NOSOFTLINKTRAV;
1132 nflags |= NLC_NOCROSSMOUNT;
1136 * We need a starting ncp from the directory vnode dp. dp must not
1137 * be locked. The returned ncp will be refd but not locked.
1139 * If no suitable ncp is found we instruct cache_fromdvp() to create
1140 * one. If this fails the directory has probably been removed while
1141 * the target was chdir'd into it and any further lookup will fail.
1143 if ((error = cache_fromdvp(dp, cred, 1, &nch)) != 0)
1145 nlookup_init_raw(nd, namebuf, UIO_SYSSPACE, nflags, cred, &nch);
1149 * Ok, do the lookup.
1151 error = nlookup(nd);
1154 * If no error occured return the requested dvpp and vpp. If
1155 * NLC_CREATE was specified nd->nl_nch may represent a negative
1156 * cache hit in which case we do not attempt to obtain the vp.
1160 if (nd->nl_nch.ncp->nc_parent) {
1162 nch.ncp = nch.ncp->nc_parent;
1165 error = cache_vget(&nch, nd->nl_cred,
1166 LK_EXCLUSIVE, dvpp);
1172 if (vpp && nd->nl_nch.ncp->nc_vp) {
1173 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, vpp);
1176 if (dvpp && *dvpp) {
1191 objcache_put(namei_oc, namebuf);
1196 * nfsrv_fhtovp() - convert a fh to a vnode ptr (optionally locked)
1197 * - look up fsid in mount list (if not found ret error)
1198 * - get vp and export rights by calling VFS_FHTOVP()
1199 * - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
1200 * - if not lockflag unlock it with vn_unlock()
1203 nfsrv_fhtovp(fhandle_t *fhp, int lockflag,
1204 struct mount **mpp, struct vnode **vpp,
1205 struct ucred *cred, struct nfssvc_sock *slp, struct sockaddr *nam,
1206 int *rdonlyp, int kerbflag, int pubflag)
1210 struct ucred *credanon;
1212 #ifdef MNT_EXNORESPORT /* XXX needs mountd and /etc/exports help yet */
1213 struct sockaddr_int *saddr;
1219 if (nfs_ispublicfh(fhp)) {
1220 if (!pubflag || !nfs_pub.np_valid)
1222 fhp = &nfs_pub.np_handle;
1225 mp = *mpp = vfs_getvfs(&fhp->fh_fsid);
1228 error = VFS_CHECKEXP(mp, nam, &exflags, &credanon);
1233 error = VFS_FHTOVP(mp, NULL, &fhp->fh_fid, vpp);
1237 #ifdef MNT_EXNORESPORT
1238 if (!(exflags & (MNT_EXNORESPORT|MNT_EXPUBLIC))) {
1239 saddr = (struct sockaddr_in *)nam;
1240 if (saddr->sin_family == AF_INET &&
1241 ntohs(saddr->sin_port) >= IPPORT_RESERVED) {
1244 return (NFSERR_AUTHERR | AUTH_TOOWEAK);
1249 * Check/setup credentials.
1251 if (exflags & MNT_EXKERB) {
1255 return (NFSERR_AUTHERR | AUTH_TOOWEAK);
1257 } else if (kerbflag) {
1260 return (NFSERR_AUTHERR | AUTH_TOOWEAK);
1261 } else if (cred->cr_uid == 0 || (exflags & MNT_EXPORTANON)) {
1262 cred->cr_uid = credanon->cr_uid;
1263 for (i = 0; i < credanon->cr_ngroups && i < NGROUPS; i++)
1264 cred->cr_groups[i] = credanon->cr_groups[i];
1265 cred->cr_ngroups = i;
1267 if (exflags & MNT_EXRDONLY)
1278 * WebNFS: check if a filehandle is a public filehandle. For v3, this
1279 * means a length of 0, for v2 it means all zeroes. nfsm_srvmtofh has
1280 * transformed this to all zeroes in both cases, so check for it.
1283 nfs_ispublicfh(fhandle_t *fhp)
1285 char *cp = (char *)fhp;
1288 for (i = 0; i < NFSX_V3FH; i++)
1294 #endif /* NFS_NOSERVER */
1296 * This function compares two net addresses by family and returns TRUE
1297 * if they are the same host.
1298 * If there is any doubt, return FALSE.
1299 * The AF_INET family is handled as a special case so that address mbufs
1300 * don't need to be saved to store "struct in_addr", which is only 4 bytes.
1303 netaddr_match(int family, union nethostaddr *haddr, struct sockaddr *nam)
1305 struct sockaddr_in *inetaddr;
1309 inetaddr = (struct sockaddr_in *)nam;
1310 if (inetaddr->sin_family == AF_INET &&
1311 inetaddr->sin_addr.s_addr == haddr->had_inetaddr)
1320 static nfsuint64 nfs_nullcookie = { { 0, 0 } };
1322 * This function finds the directory cookie that corresponds to the
1323 * logical byte offset given.
1326 nfs_getcookie(struct nfsnode *np, off_t off, int add)
1328 struct nfsdmap *dp, *dp2;
1331 pos = (uoff_t)off / NFS_DIRBLKSIZ;
1332 if (pos == 0 || off < 0) {
1335 panic("nfs getcookie add at <= 0");
1337 return (&nfs_nullcookie);
1340 dp = np->n_cookies.lh_first;
1343 dp = kmalloc(sizeof(struct nfsdmap), M_NFSDIROFF,
1345 dp->ndm_eocookie = 0;
1346 LIST_INSERT_HEAD(&np->n_cookies, dp, ndm_list);
1350 while (pos >= NFSNUMCOOKIES) {
1351 pos -= NFSNUMCOOKIES;
1352 if (dp->ndm_list.le_next) {
1353 if (!add && dp->ndm_eocookie < NFSNUMCOOKIES &&
1354 pos >= dp->ndm_eocookie)
1356 dp = dp->ndm_list.le_next;
1358 dp2 = kmalloc(sizeof(struct nfsdmap), M_NFSDIROFF,
1360 dp2->ndm_eocookie = 0;
1361 LIST_INSERT_AFTER(dp, dp2, ndm_list);
1366 if (pos >= dp->ndm_eocookie) {
1368 dp->ndm_eocookie = pos + 1;
1372 return (&dp->ndm_cookies[pos]);
1376 * Invalidate cached directory information, except for the actual directory
1377 * blocks (which are invalidated separately).
1378 * Done mainly to avoid the use of stale offset cookies.
1381 nfs_invaldir(struct vnode *vp)
1383 struct nfsnode *np = VTONFS(vp);
1386 if (vp->v_type != VDIR)
1387 panic("nfs: invaldir not dir");
1389 np->n_direofoffset = 0;
1390 np->n_cookieverf.nfsuquad[0] = 0;
1391 np->n_cookieverf.nfsuquad[1] = 0;
1392 if (np->n_cookies.lh_first)
1393 np->n_cookies.lh_first->ndm_eocookie = 0;
1397 * Set the v_type field for an NFS client's vnode and initialize for
1398 * buffer cache operations if necessary.
1401 nfs_setvtype(struct vnode *vp, enum vtype vtyp)
1410 * Needs VMIO, size not yet known, and blocksize
1411 * is not really relevant if we are passing a
1414 vinitvmio(vp, 0, PAGE_SIZE, -1);
1422 * The write verifier has changed (probably due to a server reboot), so all
1423 * B_NEEDCOMMIT blocks will have to be written again. Since they are on the
1424 * dirty block list as B_DELWRI, all this takes is clearing the B_NEEDCOMMIT
1425 * and B_CLUSTEROK flags. Once done the new write verifier can be set for the
1428 * B_CLUSTEROK must be cleared along with B_NEEDCOMMIT because stage 1 data
1429 * writes are not clusterable.
1432 static int nfs_clearcommit_bp(struct buf *bp, void *data __unused);
1433 static int nfs_clearcommit_callback(struct mount *mp, struct vnode *vp,
1434 void *data __unused);
1437 nfs_clearcommit(struct mount *mp)
1439 vsyncscan(mp, VMSC_NOWAIT, nfs_clearcommit_callback, NULL);
1443 nfs_clearcommit_callback(struct mount *mp, struct vnode *vp,
1444 void *data __unused)
1446 lwkt_gettoken(&vp->v_token);
1447 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
1448 nfs_clearcommit_bp, NULL);
1449 lwkt_reltoken(&vp->v_token);
1455 nfs_clearcommit_bp(struct buf *bp, void *data __unused)
1457 if (BUF_LOCKINUSE(bp) == 0 &&
1458 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
1459 == (B_DELWRI | B_NEEDCOMMIT)) {
1460 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
1465 #ifndef NFS_NOSERVER
1467 * Map errnos to NFS error numbers. For Version 3 also filter out error
1468 * numbers not specified for the associated procedure.
1471 nfsrv_errmap(struct nfsrv_descript *nd, int err)
1473 short *defaulterrp, *errp;
1475 if (nd->nd_flag & ND_NFSV3) {
1476 if (nd->nd_procnum <= NFSPROC_COMMIT) {
1477 errp = defaulterrp = nfsrv_v3errmap[nd->nd_procnum];
1481 else if (*errp > err)
1484 return ((int)*defaulterrp);
1486 return (err & 0xffff);
1489 return ((int)nfsrv_v2errmap[err - 1]);
1494 * Sort the group list in increasing numerical order.
1495 * (Insertion sort by Chris Torek, who was grossed out by the bubble sort
1496 * that used to be here.)
1499 nfsrvw_sort(gid_t *list, int num)
1504 /* Insertion sort. */
1505 for (i = 1; i < num; i++) {
1507 /* find correct slot for value v, moving others up */
1508 for (j = i; --j >= 0 && v < list[j];)
1509 list[j + 1] = list[j];
1515 * copy credentials making sure that the result can be compared with bcmp().
1518 nfsrv_setcred(struct ucred *incred, struct ucred *outcred)
1522 bzero((caddr_t)outcred, sizeof (struct ucred));
1523 outcred->cr_ref = 1;
1524 outcred->cr_uid = incred->cr_uid;
1525 outcred->cr_ngroups = incred->cr_ngroups;
1526 for (i = 0; i < incred->cr_ngroups; i++)
1527 outcred->cr_groups[i] = incred->cr_groups[i];
1528 nfsrvw_sort(outcred->cr_groups, outcred->cr_ngroups);
1530 #endif /* NFS_NOSERVER */
1533 * Hold a ucred in nfs_node. Discard prison information, otherwise
1534 * prisons might stick around indefinitely due to NFS node caching.
1537 nfs_crhold(struct ucred *cred)
1540 if (cred->cr_prison) {
1542 prison_free(cred->cr_prison);
1543 cred->cr_prison = NULL;
1545 cred = crhold(cred);
1552 * Return whether two ucreds are the same insofar as NFS cares about.
1555 nfs_crsame(struct ucred *cr1, struct ucred *cr2)
1558 if (cr1 == NULL || cr2 == NULL)
1559 return (cr1 == cr2);
1560 if (cr1->cr_uid != cr2->cr_uid ||
1561 cr1->cr_ruid != cr2->cr_ruid ||
1562 cr1->cr_rgid != cr2->cr_rgid ||
1563 cr1->cr_ngroups != cr2->cr_ngroups) {
1566 if (bcmp(cr1->cr_groups, cr2->cr_groups,
1567 cr1->cr_ngroups * sizeof(cr1->cr_groups[0])) != 0) {