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>
57 #include <sys/objcache.h>
60 #include <vm/vm_object.h>
61 #include <vm/vm_extern.h>
71 #include "nfsm_subs.h"
74 #include <netinet/in.h>
76 MALLOC_DEFINE(M_NFSMOUNT, "NFS mount", "NFS mount");
79 * Data items converted to xdr at startup, since they are constant
80 * This is kinda hokey, but may save a little time doing byte swaps
82 u_int32_t nfs_xdrneg1;
83 u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers;
84 u_int32_t rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr;
85 u_int32_t rpc_auth_kerb;
86 u_int32_t nfs_prog, nfs_true, nfs_false;
88 /* And other global data */
89 static enum vtype nv2tov_type[8]= {
90 VNON, VREG, VDIR, VBLK, VCHR, VLNK, VNON, VNON
92 enum vtype nv3tov_type[8]= {
93 VNON, VREG, VDIR, VBLK, VCHR, VLNK, VSOCK, VFIFO
99 * Protect master lists only. Primary protection uses the per-mount
100 * and per nfssvc_sock tokens.
102 struct lwkt_token nfs_token = LWKT_TOKEN_INITIALIZER(unp_token);
104 static long nfs_pbuf_freecnt = -1; /* start out unlimited */
106 struct nfsmount_head nfs_mountq = TAILQ_HEAD_INITIALIZER(nfs_mountq);
107 struct nfssvc_sockhead nfssvc_sockhead;
108 int nfssvc_sockhead_flag;
109 struct nfsd_head nfsd_head;
111 struct nfs_bufq nfs_bufq;
112 struct nqfhhashhead *nqfhhashtbl;
115 static int nfs_prev_nfssvc_sy_narg;
116 static sy_call_t *nfs_prev_nfssvc_sy_call;
121 * Mapping of old NFS Version 2 RPC numbers to generic numbers.
123 int nfsv3_procid[NFS_NPROCS] = {
152 #endif /* NFS_NOSERVER */
154 * and the reverse mapping from generic to Version 2 procedure numbers
156 int nfsv2_procid[NFS_NPROCS] = {
187 * Maps errno values to nfs error numbers.
188 * Use NFSERR_IO as the catch all for ones not specifically defined in
191 static u_char nfsrv_v2errmap[ELAST] = {
192 NFSERR_PERM, NFSERR_NOENT, NFSERR_IO, NFSERR_IO, NFSERR_IO,
193 NFSERR_NXIO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
194 NFSERR_IO, NFSERR_IO, NFSERR_ACCES, NFSERR_IO, NFSERR_IO,
195 NFSERR_IO, NFSERR_EXIST, NFSERR_IO, NFSERR_NODEV, NFSERR_NOTDIR,
196 NFSERR_ISDIR, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
197 NFSERR_IO, NFSERR_FBIG, NFSERR_NOSPC, NFSERR_IO, NFSERR_ROFS,
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_IO, NFSERR_IO, NFSERR_IO,
204 NFSERR_IO, NFSERR_IO, NFSERR_NAMETOL, NFSERR_IO, NFSERR_IO,
205 NFSERR_NOTEMPTY, NFSERR_IO, NFSERR_IO, NFSERR_DQUOT, NFSERR_STALE,
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_IO, NFSERR_IO, NFSERR_IO,
209 NFSERR_IO /* << Last is 86 */
213 * Maps errno values to nfs error numbers.
214 * Although it is not obvious whether or not NFS clients really care if
215 * a returned error value is in the specified list for the procedure, the
216 * safest thing to do is filter them appropriately. For Version 2, the
217 * X/Open XNFS document is the only specification that defines error values
218 * for each RPC (The RFC simply lists all possible error values for all RPCs),
219 * so I have decided to not do this for Version 2.
220 * The first entry is the default error return and the rest are the valid
221 * errors for that RPC in increasing numeric order.
223 static short nfsv3err_null[] = {
228 static short nfsv3err_getattr[] = {
237 static short nfsv3err_setattr[] = {
253 static short nfsv3err_lookup[] = {
266 static short nfsv3err_access[] = {
275 static short nfsv3err_readlink[] = {
287 static short nfsv3err_read[] = {
299 static short nfsv3err_write[] = {
314 static short nfsv3err_create[] = {
331 static short nfsv3err_mkdir[] = {
348 static short nfsv3err_symlink[] = {
365 static short nfsv3err_mknod[] = {
383 static short nfsv3err_remove[] = {
397 static short nfsv3err_rmdir[] = {
415 static short nfsv3err_rename[] = {
438 static short nfsv3err_link[] = {
458 static short nfsv3err_readdir[] = {
471 static short nfsv3err_readdirplus[] = {
485 static short nfsv3err_fsstat[] = {
494 static short nfsv3err_fsinfo[] = {
502 static short nfsv3err_pathconf[] = {
510 static short nfsv3err_commit[] = {
519 static short *nfsrv_v3errmap[] = {
537 nfsv3err_readdirplus,
544 #endif /* NFS_NOSERVER */
547 extern int sys_nfssvc(struct proc *, struct nfssvc_args *, int *);
550 * This needs to return a monotonically increasing or close to monotonically
551 * increasing result, otherwise the write gathering queues won't work
560 return ((u_quad_t)tv.tv_sec * 1000000 + (u_quad_t)tv.tv_usec);
564 * Called once to initialize data structures...
567 nfs_init(struct vfsconf *vfsp)
569 callout_init(&nfs_timer_handle);
570 nfsmount_objcache = objcache_create_simple(M_NFSMOUNT, sizeof(struct nfsmount));
572 nfs_mount_type = vfsp->vfc_typenum;
574 rpc_vers = txdr_unsigned(RPC_VER2);
575 rpc_call = txdr_unsigned(RPC_CALL);
576 rpc_reply = txdr_unsigned(RPC_REPLY);
577 rpc_msgdenied = txdr_unsigned(RPC_MSGDENIED);
578 rpc_msgaccepted = txdr_unsigned(RPC_MSGACCEPTED);
579 rpc_mismatch = txdr_unsigned(RPC_MISMATCH);
580 rpc_autherr = txdr_unsigned(RPC_AUTHERR);
581 rpc_auth_unix = txdr_unsigned(RPCAUTH_UNIX);
582 rpc_auth_kerb = txdr_unsigned(RPCAUTH_KERB4);
583 nfs_prog = txdr_unsigned(NFS_PROG);
584 nfs_true = txdr_unsigned(TRUE);
585 nfs_false = txdr_unsigned(FALSE);
586 nfs_xdrneg1 = txdr_unsigned(-1);
587 nfs_ticks = (hz * NFS_TICKINTVL + 500) / 1000;
590 nfs_nhinit(); /* Init the nfsnode table */
592 nfsrv_init(0); /* Init server data structures */
593 nfsrv_initcache(); /* Init the server request cache */
597 * Mainly for vkernel operation. If memory is severely limited
599 if (nfs_maxasyncbio > nmbclusters * MCLBYTES / NFS_MAXDATA / 3)
600 nfs_maxasyncbio = nmbclusters * MCLBYTES / NFS_MAXDATA / 3;
601 if (nfs_maxasyncbio < 4)
605 * Initialize reply list and start timer
607 nfs_timer_callout(0);
609 nfs_prev_nfssvc_sy_narg = sysent[SYS_nfssvc].sy_narg;
610 sysent[SYS_nfssvc].sy_narg = 2;
611 nfs_prev_nfssvc_sy_call = sysent[SYS_nfssvc].sy_call;
612 sysent[SYS_nfssvc].sy_call = (sy_call_t *)sys_nfssvc;
614 nfs_pbuf_freecnt = nswbuf / 2 + 1;
620 nfs_uninit(struct vfsconf *vfsp)
622 callout_stop(&nfs_timer_handle);
624 sysent[SYS_nfssvc].sy_narg = nfs_prev_nfssvc_sy_narg;
625 sysent[SYS_nfssvc].sy_call = nfs_prev_nfssvc_sy_call;
630 * Attribute cache routines.
631 * nfs_loadattrcache() - loads or updates the cache contents from attributes
632 * that are on the mbuf list
633 * nfs_getattrcache() - returns valid attributes if found in cache, returns
638 * Load the attribute cache (that lives in the nfsnode entry) with
639 * the values on the mbuf list. Load *vaper with the attributes. vaper
642 * As a side effect n_mtime, which we use to determine if the file was
643 * modified by some other host, is set to the attribute timestamp and
644 * NRMODIFIED is set if the two values differ.
646 * WARNING: the mtime loaded into vaper does not necessarily represent
647 * n_mtime or n_attr.mtime due to NACC and NUPD.
650 nfs_loadattrcache(struct vnode *vp, struct mbuf **mdp, caddr_t *dposp,
651 struct vattr *vaper, int lattr_flags)
654 struct nfs_fattr *fp;
664 struct timespec mtime;
665 int v3 = NFS_ISV3(vp);
668 t1 = (mtod(md, caddr_t) + md->m_len) - *dposp;
669 if ((error = nfsm_disct(mdp, dposp, NFSX_FATTR(v3), t1, &cp2)) != 0)
671 fp = (struct nfs_fattr *)cp2;
673 vtyp = nfsv3tov_type(fp->fa_type);
674 vmode = fxdr_unsigned(u_short, fp->fa_mode);
675 rmajor = (int)fxdr_unsigned(int, fp->fa3_rdev.specdata1);
676 rminor = (int)fxdr_unsigned(int, fp->fa3_rdev.specdata2);
677 fxdr_nfsv3time(&fp->fa3_mtime, &mtime);
679 vtyp = nfsv2tov_type(fp->fa_type);
680 vmode = fxdr_unsigned(u_short, fp->fa_mode);
684 * The duplicate information returned in fa_type and fa_mode
685 * is an ambiguity in the NFS version 2 protocol.
687 * VREG should be taken literally as a regular file. If a
688 * server intents to return some type information differently
689 * in the upper bits of the mode field (e.g. for sockets, or
690 * FIFOs), NFSv2 mandates fa_type to be VNON. Anyway, we
691 * leave the examination of the mode bits even in the VREG
692 * case to avoid breakage for bogus servers, but we make sure
693 * that there are actually type bits set in the upper part of
694 * fa_mode (and failing that, trust the va_type field).
696 * NFSv3 cleared the issue, and requires fa_mode to not
697 * contain any type information (while also introduing sockets
698 * and FIFOs for fa_type).
700 if (vtyp == VNON || (vtyp == VREG && (vmode & S_IFMT) != 0))
701 vtyp = IFTOVT(vmode);
702 rdev = fxdr_unsigned(int32_t, fp->fa2_rdev);
703 rmajor = umajor(rdev);
704 rminor = uminor(rdev);
705 fxdr_nfsv2time(&fp->fa2_mtime, &mtime);
708 * Really ugly NFSv2 kludge.
710 if (vtyp == VCHR && rdev == (udev_t)0xffffffff)
715 * If v_type == VNON it is a new node, so fill in the v_type,
716 * n_mtime fields. Check to see if it represents a special
717 * device, and if so, check for a possible alias. Once the
718 * correct vnode has been obtained, fill in the rest of the
722 if (vp->v_type != vtyp) {
723 nfs_setvtype(vp, vtyp);
724 if (vp->v_type == VFIFO) {
725 vp->v_ops = &vp->v_mount->mnt_vn_fifo_ops;
726 } else if (vp->v_type == VCHR || vp->v_type == VBLK) {
727 vp->v_ops = &vp->v_mount->mnt_vn_spec_ops;
728 addaliasu(vp, rmajor, rminor);
730 vp->v_ops = &vp->v_mount->mnt_vn_use_ops;
732 np->n_mtime = mtime.tv_sec;
733 } else if (np->n_mtime != mtime.tv_sec) {
735 * If we haven't modified the file locally and the server
736 * timestamp does not match, then the server probably
737 * modified the file. We must flag this condition so
738 * the proper syncnronization can be done. We do not
739 * try to synchronize the state here because that
740 * could lead to an endless recursion.
742 * XXX loadattrcache can be set during the reply to a write,
743 * before the write timestamp is properly processed. To
744 * avoid unconditionally setting the rmodified bit (which
745 * has the effect of flushing the cache), we only do this
746 * check if the lmodified bit is not set.
748 np->n_mtime = mtime.tv_sec;
749 if ((lattr_flags & NFS_LATTR_NOMTIMECHECK) == 0)
750 np->n_flag |= NRMODIFIED;
754 vap->va_mode = (vmode & 07777);
755 vap->va_rmajor = rmajor;
756 vap->va_rminor = rminor;
757 vap->va_mtime = mtime;
758 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
760 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
761 vap->va_uid = fxdr_unsigned(uid_t, fp->fa_uid);
762 vap->va_gid = fxdr_unsigned(gid_t, fp->fa_gid);
763 vap->va_size = fxdr_hyper(&fp->fa3_size);
764 vap->va_blocksize = NFS_FABLKSIZE;
765 vap->va_bytes = fxdr_hyper(&fp->fa3_used);
766 vap->va_fileid = fxdr_hyper(&fp->fa3_fileid);
767 fxdr_nfsv3time(&fp->fa3_atime, &vap->va_atime);
768 fxdr_nfsv3time(&fp->fa3_ctime, &vap->va_ctime);
772 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
773 vap->va_uid = fxdr_unsigned(uid_t, fp->fa_uid);
774 vap->va_gid = fxdr_unsigned(gid_t, fp->fa_gid);
775 vap->va_size = fxdr_unsigned(u_int32_t, fp->fa2_size);
776 vap->va_blocksize = fxdr_unsigned(int32_t, fp->fa2_blocksize);
777 vap->va_bytes = (u_quad_t)fxdr_unsigned(int32_t, fp->fa2_blocks)
779 vap->va_fileid = fxdr_unsigned(int32_t, fp->fa2_fileid);
780 fxdr_nfsv2time(&fp->fa2_atime, &vap->va_atime);
782 vap->va_ctime.tv_sec = fxdr_unsigned(u_int32_t,
783 fp->fa2_ctime.nfsv2_sec);
784 vap->va_ctime.tv_nsec = 0;
785 vap->va_gen = fxdr_unsigned(u_int32_t,fp->fa2_ctime.nfsv2_usec);
788 np->n_attrstamp = time_second;
789 if (vap->va_size != np->n_size) {
790 if (vap->va_type == VREG) {
792 * Get rid of all the junk we had before and just
793 * set NRMODIFIED if NLMODIFIED is 0. Depend on
794 * occassionally flushing our dirty buffers to
795 * clear both the NLMODIFIED and NRMODIFIED flags.
797 if ((np->n_flag & NLMODIFIED) == 0)
798 np->n_flag |= NRMODIFIED;
800 if ((lattr_flags & NFS_LATTR_NOSHRINK) &&
801 vap->va_size < np->n_size) {
803 * We've been told not to shrink the file;
804 * zero np->n_attrstamp to indicate that
805 * the attributes are stale.
807 * This occurs primarily due to recursive
808 * NFS ops that are executed during periods
809 * where we cannot safely reduce the size of
812 * Additionally, write rpcs are broken down
813 * into buffers and np->n_size is
814 * pre-extended. Setting NRMODIFIED here
815 * can result in n_size getting reset to a
816 * lower value, which is NOT what we want.
817 * XXX this needs to be cleaned up a lot
820 vap->va_size = np->n_size;
822 if ((np->n_flag & NLMODIFIED) == 0)
823 np->n_flag |= NRMODIFIED;
824 } else if (np->n_flag & NLMODIFIED) {
826 * We've modified the file: Use the larger
827 * of our size, and the server's size. At
828 * this point the cache coherency is all
829 * shot to hell. To try to handle multiple
830 * clients appending to the file at the same
831 * time mark that the server has changed
832 * the file if the server's notion of the
833 * file size is larger then our notion.
835 * XXX this needs work.
837 if (vap->va_size < np->n_size) {
838 vap->va_size = np->n_size;
840 np->n_size = vap->va_size;
841 np->n_flag |= NRMODIFIED;
845 * Someone changed the file's size on the
846 * server and there are no local changes
847 * to get in the way, set the size and mark
850 np->n_size = vap->va_size;
851 np->n_flag |= NRMODIFIED;
853 nvnode_pager_setsize(vp, np->n_size, XXX);
856 np->n_size = vap->va_size;
860 bcopy((caddr_t)vap, (caddr_t)vaper, sizeof(*vap));
861 if (np->n_flag & NCHG) {
862 if (np->n_flag & NACC)
863 vaper->va_atime = np->n_atim;
864 if (np->n_flag & NUPD)
865 vaper->va_mtime = np->n_mtim;
872 #include <sys/sysctl.h>
873 SYSCTL_DECL(_vfs_nfs);
874 static int nfs_acdebug;
875 SYSCTL_INT(_vfs_nfs, OID_AUTO, acdebug, CTLFLAG_RW, &nfs_acdebug, 0, "");
879 * Check the time stamp
880 * If the cache is valid, copy contents to *vap and return 0
881 * otherwise return an error
884 nfs_getattrcache(struct vnode *vp, struct vattr *vaper)
888 struct nfsmount *nmp;
893 nmp = VFSTONFS(vp->v_mount);
896 * Dynamic timeout based on how recently the file was modified.
897 * n_mtime is always valid.
899 timeo = (get_approximate_time_t() - np->n_mtime) / 60;
903 kprintf("nfs_getattrcache: initial timeo = %d\n", timeo);
906 if (vap->va_type == VDIR) {
907 if ((np->n_flag & NLMODIFIED) || timeo < nmp->nm_acdirmin)
908 timeo = nmp->nm_acdirmin;
909 else if (timeo > nmp->nm_acdirmax)
910 timeo = nmp->nm_acdirmax;
912 if ((np->n_flag & NLMODIFIED) || timeo < nmp->nm_acregmin)
913 timeo = nmp->nm_acregmin;
914 else if (timeo > nmp->nm_acregmax)
915 timeo = nmp->nm_acregmax;
920 kprintf("acregmin %d; acregmax %d; acdirmin %d; acdirmax %d\n",
921 nmp->nm_acregmin, nmp->nm_acregmax,
922 nmp->nm_acdirmin, nmp->nm_acdirmax);
925 kprintf("nfs_getattrcache: age = %d; final timeo = %d\n",
926 (int)(time_second - np->n_attrstamp), timeo);
929 if (np->n_attrstamp == 0 || (time_second - np->n_attrstamp) >= timeo) {
930 nfsstats.attrcache_misses++;
933 nfsstats.attrcache_hits++;
936 * Our attribute cache can be stale due to modifications made on
937 * this host. XXX this is a bad hack. We need a more deterministic
938 * means of finding out which np fields are valid verses attr cache
939 * fields. We really should update the vattr info on the fly when
940 * making local changes.
942 if (vap->va_size != np->n_size) {
943 if (vap->va_type == VREG) {
944 if (np->n_flag & NLMODIFIED)
945 vap->va_size = np->n_size;
946 nfs_meta_setsize(vp, curthread, vap->va_size, 0);
948 np->n_size = vap->va_size;
951 bcopy((caddr_t)vap, (caddr_t)vaper, sizeof(struct vattr));
952 if (np->n_flag & NCHG) {
953 if (np->n_flag & NACC)
954 vaper->va_atime = np->n_atim;
955 if (np->n_flag & NUPD)
956 vaper->va_mtime = np->n_mtim;
964 * Set up nameidata for a lookup() call and do it.
966 * If pubflag is set, this call is done for a lookup operation on the
967 * public filehandle. In that case we allow crossing mountpoints and
968 * absolute pathnames. However, the caller is expected to check that
969 * the lookup result is within the public fs, and deny access if
972 * dirp may be set whether an error is returned or not, and must be
973 * released by the caller.
975 * On return nd->nl_nch usually points to the target ncp, which may represent
978 * NOTE: the caller must call nlookup_done(nd) unconditionally on return
982 nfs_namei(struct nlookupdata *nd, struct ucred *cred, int nflags,
983 struct vnode **dvpp, struct vnode **vpp,
984 fhandle_t *fhp, int len,
985 struct nfssvc_sock *slp, struct sockaddr *nam, struct mbuf **mdp,
986 caddr_t *dposp, struct vnode **dirpp, struct thread *td,
987 int kerbflag, int pubflag)
991 char *fromcp, *tocp, *cp;
998 namebuf = objcache_get(namei_oc, M_WAITOK);
1002 * Copy the name from the mbuf list to namebuf.
1007 rem = mtod(md, caddr_t) + md->m_len - fromcp;
1008 for (i = 0; i < len; i++) {
1015 fromcp = mtod(md, caddr_t);
1018 if (*fromcp == '\0' || (!pubflag && *fromcp == '/')) {
1022 *tocp++ = *fromcp++;
1028 len = nfsm_rndup(len)-len;
1032 else if ((error = nfs_adv(mdp, dposp, len, rem)) != 0)
1037 * Extract and set starting directory. The returned dp is refd
1040 error = nfsrv_fhtovp(fhp, FALSE, &mp, &dp, cred, slp,
1041 nam, &rdonly, kerbflag, pubflag);
1044 if (dp->v_type != VDIR) {
1051 * Set return directory. Reference to dp is implicitly transfered
1052 * to the returned pointer. This must be set before we potentially
1058 * read-only - NLC_DELETE, NLC_RENAME_DST are disallowed. NLC_CREATE
1059 * is passed through to nlookup() and will be disallowed
1060 * if the file does not already exist.
1063 nflags |= NLC_NFS_RDONLY;
1064 if (nflags & (NLC_DELETE | NLC_RENAME_DST)) {
1071 * Oh joy. For WebNFS, handle those pesky '%' escapes,
1072 * and the 'native path' indicator.
1075 cp = objcache_get(namei_oc, M_WAITOK);
1078 if ((unsigned char)*fromcp >= WEBNFS_SPECCHAR_START) {
1079 switch ((unsigned char)*fromcp) {
1080 case WEBNFS_NATIVE_CHAR:
1082 * 'Native' path for us is the same
1083 * as a path according to the NFS spec,
1084 * just skip the escape char.
1089 * More may be added in the future, range 0x80-0xff
1093 objcache_put(namei_oc, cp);
1098 * Translate the '%' escapes, URL-style.
1100 while (*fromcp != '\0') {
1101 if (*fromcp == WEBNFS_ESC_CHAR) {
1102 if (fromcp[1] != '\0' && fromcp[2] != '\0') {
1104 *tocp++ = HEXSTRTOI(fromcp);
1109 objcache_put(namei_oc, cp);
1113 *tocp++ = *fromcp++;
1116 objcache_put(namei_oc, namebuf);
1121 * Setup for search. We need to get a start directory from dp. Note
1122 * that dp is ref'd, but we no longer 'own' the ref (*dirpp owns it).
1125 nflags |= NLC_NFS_NOSOFTLINKTRAV;
1126 nflags |= NLC_NOCROSSMOUNT;
1130 * We need a starting ncp from the directory vnode dp. dp must not
1131 * be locked. The returned ncp will be refd but not locked.
1133 * If no suitable ncp is found we instruct cache_fromdvp() to create
1134 * one. If this fails the directory has probably been removed while
1135 * the target was chdir'd into it and any further lookup will fail.
1137 if ((error = cache_fromdvp(dp, cred, 1, &nch)) != 0)
1139 nlookup_init_raw(nd, namebuf, UIO_SYSSPACE, nflags, cred, &nch);
1143 * Ok, do the lookup.
1145 error = nlookup(nd);
1148 * If no error occured return the requested dvpp and vpp. If
1149 * NLC_CREATE was specified nd->nl_nch may represent a negative
1150 * cache hit in which case we do not attempt to obtain the vp.
1154 if (nd->nl_nch.ncp->nc_parent) {
1156 nch.ncp = nch.ncp->nc_parent;
1159 error = cache_vget(&nch, nd->nl_cred,
1160 LK_EXCLUSIVE, dvpp);
1166 if (vpp && nd->nl_nch.ncp->nc_vp) {
1167 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, vpp);
1170 if (dvpp && *dvpp) {
1185 objcache_put(namei_oc, namebuf);
1190 * nfsrv_fhtovp() - convert a fh to a vnode ptr (optionally locked)
1191 * - look up fsid in mount list (if not found ret error)
1192 * - get vp and export rights by calling VFS_FHTOVP()
1193 * - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
1194 * - if not lockflag unlock it with vn_unlock()
1197 nfsrv_fhtovp(fhandle_t *fhp, int lockflag,
1198 struct mount **mpp, struct vnode **vpp,
1199 struct ucred *cred, struct nfssvc_sock *slp, struct sockaddr *nam,
1200 int *rdonlyp, int kerbflag, int pubflag)
1204 struct ucred *credanon;
1206 #ifdef MNT_EXNORESPORT /* XXX needs mountd and /etc/exports help yet */
1207 struct sockaddr_int *saddr;
1213 if (nfs_ispublicfh(fhp)) {
1214 if (!pubflag || !nfs_pub.np_valid)
1216 fhp = &nfs_pub.np_handle;
1219 mp = *mpp = vfs_getvfs(&fhp->fh_fsid);
1222 error = VFS_CHECKEXP(mp, nam, &exflags, &credanon);
1225 error = VFS_FHTOVP(mp, NULL, &fhp->fh_fid, vpp);
1228 #ifdef MNT_EXNORESPORT
1229 if (!(exflags & (MNT_EXNORESPORT|MNT_EXPUBLIC))) {
1230 saddr = (struct sockaddr_in *)nam;
1231 if (saddr->sin_family == AF_INET &&
1232 ntohs(saddr->sin_port) >= IPPORT_RESERVED) {
1235 return (NFSERR_AUTHERR | AUTH_TOOWEAK);
1240 * Check/setup credentials.
1242 if (exflags & MNT_EXKERB) {
1246 return (NFSERR_AUTHERR | AUTH_TOOWEAK);
1248 } else if (kerbflag) {
1251 return (NFSERR_AUTHERR | AUTH_TOOWEAK);
1252 } else if (cred->cr_uid == 0 || (exflags & MNT_EXPORTANON)) {
1253 cred->cr_uid = credanon->cr_uid;
1254 for (i = 0; i < credanon->cr_ngroups && i < NGROUPS; i++)
1255 cred->cr_groups[i] = credanon->cr_groups[i];
1256 cred->cr_ngroups = i;
1258 if (exflags & MNT_EXRDONLY)
1269 * WebNFS: check if a filehandle is a public filehandle. For v3, this
1270 * means a length of 0, for v2 it means all zeroes. nfsm_srvmtofh has
1271 * transformed this to all zeroes in both cases, so check for it.
1274 nfs_ispublicfh(fhandle_t *fhp)
1276 char *cp = (char *)fhp;
1279 for (i = 0; i < NFSX_V3FH; i++)
1285 #endif /* NFS_NOSERVER */
1287 * This function compares two net addresses by family and returns TRUE
1288 * if they are the same host.
1289 * If there is any doubt, return FALSE.
1290 * The AF_INET family is handled as a special case so that address mbufs
1291 * don't need to be saved to store "struct in_addr", which is only 4 bytes.
1294 netaddr_match(int family, union nethostaddr *haddr, struct sockaddr *nam)
1296 struct sockaddr_in *inetaddr;
1300 inetaddr = (struct sockaddr_in *)nam;
1301 if (inetaddr->sin_family == AF_INET &&
1302 inetaddr->sin_addr.s_addr == haddr->had_inetaddr)
1311 static nfsuint64 nfs_nullcookie = { { 0, 0 } };
1313 * This function finds the directory cookie that corresponds to the
1314 * logical byte offset given.
1317 nfs_getcookie(struct nfsnode *np, off_t off, int add)
1319 struct nfsdmap *dp, *dp2;
1322 pos = (uoff_t)off / NFS_DIRBLKSIZ;
1323 if (pos == 0 || off < 0) {
1326 panic("nfs getcookie add at <= 0");
1328 return (&nfs_nullcookie);
1331 dp = np->n_cookies.lh_first;
1334 dp = kmalloc(sizeof(struct nfsdmap), M_NFSDIROFF,
1336 dp->ndm_eocookie = 0;
1337 LIST_INSERT_HEAD(&np->n_cookies, dp, ndm_list);
1341 while (pos >= NFSNUMCOOKIES) {
1342 pos -= NFSNUMCOOKIES;
1343 if (dp->ndm_list.le_next) {
1344 if (!add && dp->ndm_eocookie < NFSNUMCOOKIES &&
1345 pos >= dp->ndm_eocookie)
1347 dp = dp->ndm_list.le_next;
1349 dp2 = kmalloc(sizeof(struct nfsdmap), M_NFSDIROFF,
1351 dp2->ndm_eocookie = 0;
1352 LIST_INSERT_AFTER(dp, dp2, ndm_list);
1357 if (pos >= dp->ndm_eocookie) {
1359 dp->ndm_eocookie = pos + 1;
1363 return (&dp->ndm_cookies[pos]);
1367 * Invalidate cached directory information, except for the actual directory
1368 * blocks (which are invalidated separately).
1369 * Done mainly to avoid the use of stale offset cookies.
1372 nfs_invaldir(struct vnode *vp)
1374 struct nfsnode *np = VTONFS(vp);
1377 if (vp->v_type != VDIR)
1378 panic("nfs: invaldir not dir");
1380 np->n_direofoffset = 0;
1381 np->n_cookieverf.nfsuquad[0] = 0;
1382 np->n_cookieverf.nfsuquad[1] = 0;
1383 if (np->n_cookies.lh_first)
1384 np->n_cookies.lh_first->ndm_eocookie = 0;
1388 * Set the v_type field for an NFS client's vnode and initialize for
1389 * buffer cache operations if necessary.
1392 nfs_setvtype(struct vnode *vp, enum vtype vtyp)
1401 * Needs VMIO, size not yet known, and blocksize
1402 * is not really relevant if we are passing a
1405 vinitvmio(vp, 0, PAGE_SIZE, -1);
1413 * The write verifier has changed (probably due to a server reboot), so all
1414 * B_NEEDCOMMIT blocks will have to be written again. Since they are on the
1415 * dirty block list as B_DELWRI, all this takes is clearing the B_NEEDCOMMIT
1416 * and B_CLUSTEROK flags. Once done the new write verifier can be set for the
1419 * B_CLUSTEROK must be cleared along with B_NEEDCOMMIT because stage 1 data
1420 * writes are not clusterable.
1423 static int nfs_clearcommit_bp(struct buf *bp, void *data __unused);
1424 static int nfs_clearcommit_callback(struct mount *mp, struct vnode *vp,
1425 void *data __unused);
1428 nfs_clearcommit(struct mount *mp)
1430 vmntvnodescan(mp, VMSC_NOWAIT, nfs_clearcommit_callback, NULL, NULL);
1434 nfs_clearcommit_callback(struct mount *mp, struct vnode *vp,
1435 void *data __unused)
1438 lwkt_gettoken(&vp->v_token);
1439 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
1440 nfs_clearcommit_bp, NULL);
1441 lwkt_reltoken(&vp->v_token);
1447 nfs_clearcommit_bp(struct buf *bp, void *data __unused)
1449 if (BUF_REFCNT(bp) == 0 &&
1450 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
1451 == (B_DELWRI | B_NEEDCOMMIT)) {
1452 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
1457 #ifndef NFS_NOSERVER
1459 * Map errnos to NFS error numbers. For Version 3 also filter out error
1460 * numbers not specified for the associated procedure.
1463 nfsrv_errmap(struct nfsrv_descript *nd, int err)
1465 short *defaulterrp, *errp;
1467 if (nd->nd_flag & ND_NFSV3) {
1468 if (nd->nd_procnum <= NFSPROC_COMMIT) {
1469 errp = defaulterrp = nfsrv_v3errmap[nd->nd_procnum];
1473 else if (*errp > err)
1476 return ((int)*defaulterrp);
1478 return (err & 0xffff);
1481 return ((int)nfsrv_v2errmap[err - 1]);
1486 * Sort the group list in increasing numerical order.
1487 * (Insertion sort by Chris Torek, who was grossed out by the bubble sort
1488 * that used to be here.)
1491 nfsrvw_sort(gid_t *list, int num)
1496 /* Insertion sort. */
1497 for (i = 1; i < num; i++) {
1499 /* find correct slot for value v, moving others up */
1500 for (j = i; --j >= 0 && v < list[j];)
1501 list[j + 1] = list[j];
1507 * copy credentials making sure that the result can be compared with bcmp().
1510 nfsrv_setcred(struct ucred *incred, struct ucred *outcred)
1514 bzero((caddr_t)outcred, sizeof (struct ucred));
1515 outcred->cr_ref = 1;
1516 outcred->cr_uid = incred->cr_uid;
1517 outcred->cr_ngroups = incred->cr_ngroups;
1518 for (i = 0; i < incred->cr_ngroups; i++)
1519 outcred->cr_groups[i] = incred->cr_groups[i];
1520 nfsrvw_sort(outcred->cr_groups, outcred->cr_ngroups);
1522 #endif /* NFS_NOSERVER */