2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * Rick Macklem at The University of Guelph.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. All advertising materials mentioning features or use of this software
17 * must display the following acknowledgement:
18 * This product includes software developed by the University of
19 * California, Berkeley and its contributors.
20 * 4. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * @(#)nfs_subs.c 8.8 (Berkeley) 5/22/95
37 * $FreeBSD: /repoman/r/ncvs/src/sys/nfsclient/nfs_subs.c,v 1.128 2004/04/14 23:23:55 peadar Exp $
38 * $DragonFly: src/sys/vfs/nfs/nfs_subs.c,v 1.48 2008/09/17 21:44:24 dillon Exp $
42 * These functions support the macros and help fiddle mbuf chains for
43 * the nfs op functions. They do things like create the rpc header and
44 * copy data between mbuf chains and uio lists.
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/kernel.h>
51 #include <sys/mount.h>
52 #include <sys/vnode.h>
53 #include <sys/nlookup.h>
54 #include <sys/namei.h>
56 #include <sys/socket.h>
58 #include <sys/malloc.h>
59 #include <sys/sysent.h>
60 #include <sys/syscall.h>
62 #include <sys/objcache.h>
65 #include <vm/vm_object.h>
66 #include <vm/vm_extern.h>
67 #include <vm/vm_zone.h>
77 #include "nfsm_subs.h"
80 #include <netinet/in.h>
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
102 static int nfs_pbuf_freecnt = -1; /* start out unlimited */
104 struct nfsmount_head nfs_mountq = TAILQ_HEAD_INITIALIZER(nfs_mountq);
105 struct nfssvc_sockhead nfssvc_sockhead;
106 int nfssvc_sockhead_flag;
107 struct nfsd_head nfsd_head;
109 struct nfs_bufq nfs_bufq;
110 struct nqfhhashhead *nqfhhashtbl;
113 static int nfs_prev_nfssvc_sy_narg;
114 static sy_call_t *nfs_prev_nfssvc_sy_call;
119 * Mapping of old NFS Version 2 RPC numbers to generic numbers.
121 int nfsv3_procid[NFS_NPROCS] = {
150 #endif /* NFS_NOSERVER */
152 * and the reverse mapping from generic to Version 2 procedure numbers
154 int nfsv2_procid[NFS_NPROCS] = {
185 * Maps errno values to nfs error numbers.
186 * Use NFSERR_IO as the catch all for ones not specifically defined in
189 static u_char nfsrv_v2errmap[ELAST] = {
190 NFSERR_PERM, NFSERR_NOENT, NFSERR_IO, NFSERR_IO, NFSERR_IO,
191 NFSERR_NXIO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
192 NFSERR_IO, NFSERR_IO, NFSERR_ACCES, NFSERR_IO, NFSERR_IO,
193 NFSERR_IO, NFSERR_EXIST, NFSERR_IO, NFSERR_NODEV, NFSERR_NOTDIR,
194 NFSERR_ISDIR, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
195 NFSERR_IO, NFSERR_FBIG, NFSERR_NOSPC, NFSERR_IO, NFSERR_ROFS,
196 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
197 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
198 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
199 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
200 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
201 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
202 NFSERR_IO, NFSERR_IO, NFSERR_NAMETOL, NFSERR_IO, NFSERR_IO,
203 NFSERR_NOTEMPTY, NFSERR_IO, NFSERR_IO, NFSERR_DQUOT, NFSERR_STALE,
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 /* << Last is 86 */
211 * Maps errno values to nfs error numbers.
212 * Although it is not obvious whether or not NFS clients really care if
213 * a returned error value is in the specified list for the procedure, the
214 * safest thing to do is filter them appropriately. For Version 2, the
215 * X/Open XNFS document is the only specification that defines error values
216 * for each RPC (The RFC simply lists all possible error values for all RPCs),
217 * so I have decided to not do this for Version 2.
218 * The first entry is the default error return and the rest are the valid
219 * errors for that RPC in increasing numeric order.
221 static short nfsv3err_null[] = {
226 static short nfsv3err_getattr[] = {
235 static short nfsv3err_setattr[] = {
251 static short nfsv3err_lookup[] = {
264 static short nfsv3err_access[] = {
273 static short nfsv3err_readlink[] = {
285 static short nfsv3err_read[] = {
297 static short nfsv3err_write[] = {
312 static short nfsv3err_create[] = {
329 static short nfsv3err_mkdir[] = {
346 static short nfsv3err_symlink[] = {
363 static short nfsv3err_mknod[] = {
381 static short nfsv3err_remove[] = {
395 static short nfsv3err_rmdir[] = {
413 static short nfsv3err_rename[] = {
436 static short nfsv3err_link[] = {
456 static short nfsv3err_readdir[] = {
469 static short nfsv3err_readdirplus[] = {
483 static short nfsv3err_fsstat[] = {
492 static short nfsv3err_fsinfo[] = {
500 static short nfsv3err_pathconf[] = {
508 static short nfsv3err_commit[] = {
517 static short *nfsrv_v3errmap[] = {
535 nfsv3err_readdirplus,
542 #endif /* NFS_NOSERVER */
545 extern int sys_nfssvc(struct proc *, struct nfssvc_args *, int *);
548 * This needs to return a monotonically increasing or close to monotonically
549 * increasing result, otherwise the write gathering queues won't work
558 return ((u_quad_t)tv.tv_sec * 1000000 + (u_quad_t)tv.tv_usec);
562 * Called once to initialize data structures...
565 nfs_init(struct vfsconf *vfsp)
567 callout_init(&nfs_timer_handle);
568 nfsmount_zone = zinit("NFSMOUNT", sizeof(struct nfsmount), 0, 0, 1);
570 nfs_mount_type = vfsp->vfc_typenum;
572 rpc_vers = txdr_unsigned(RPC_VER2);
573 rpc_call = txdr_unsigned(RPC_CALL);
574 rpc_reply = txdr_unsigned(RPC_REPLY);
575 rpc_msgdenied = txdr_unsigned(RPC_MSGDENIED);
576 rpc_msgaccepted = txdr_unsigned(RPC_MSGACCEPTED);
577 rpc_mismatch = txdr_unsigned(RPC_MISMATCH);
578 rpc_autherr = txdr_unsigned(RPC_AUTHERR);
579 rpc_auth_unix = txdr_unsigned(RPCAUTH_UNIX);
580 rpc_auth_kerb = txdr_unsigned(RPCAUTH_KERB4);
581 nfs_prog = txdr_unsigned(NFS_PROG);
582 nfs_true = txdr_unsigned(TRUE);
583 nfs_false = txdr_unsigned(FALSE);
584 nfs_xdrneg1 = txdr_unsigned(-1);
585 nfs_ticks = (hz * NFS_TICKINTVL + 500) / 1000;
588 nfs_nhinit(); /* Init the nfsnode table */
590 nfsrv_init(0); /* Init server data structures */
591 nfsrv_initcache(); /* Init the server request cache */
595 * Initialize reply list and start timer
599 nfs_prev_nfssvc_sy_narg = sysent[SYS_nfssvc].sy_narg;
600 sysent[SYS_nfssvc].sy_narg = 2;
601 nfs_prev_nfssvc_sy_call = sysent[SYS_nfssvc].sy_call;
602 sysent[SYS_nfssvc].sy_call = (sy_call_t *)sys_nfssvc;
604 nfs_pbuf_freecnt = nswbuf / 2 + 1;
610 nfs_uninit(struct vfsconf *vfsp)
612 callout_stop(&nfs_timer_handle);
614 sysent[SYS_nfssvc].sy_narg = nfs_prev_nfssvc_sy_narg;
615 sysent[SYS_nfssvc].sy_call = nfs_prev_nfssvc_sy_call;
620 * Attribute cache routines.
621 * nfs_loadattrcache() - loads or updates the cache contents from attributes
622 * that are on the mbuf list
623 * nfs_getattrcache() - returns valid attributes if found in cache, returns
628 * Load the attribute cache (that lives in the nfsnode entry) with
629 * the values on the mbuf list. Load *vaper with the attributes. vaper
632 * As a side effect n_mtime, which we use to determine if the file was
633 * modified by some other host, is set to the attribute timestamp and
634 * NRMODIFIED is set if the two values differ.
636 * WARNING: the mtime loaded into vaper does not necessarily represent
637 * n_mtime or n_attr.mtime due to NACC and NUPD.
640 nfs_loadattrcache(struct vnode **vpp, struct mbuf **mdp, caddr_t *dposp,
641 struct vattr *vaper, int lattr_flags)
643 struct vnode *vp = *vpp;
645 struct nfs_fattr *fp;
655 struct timespec mtime;
656 int v3 = NFS_ISV3(vp);
659 t1 = (mtod(md, caddr_t) + md->m_len) - *dposp;
660 if ((error = nfsm_disct(mdp, dposp, NFSX_FATTR(v3), t1, &cp2)) != 0)
662 fp = (struct nfs_fattr *)cp2;
664 vtyp = nfsv3tov_type(fp->fa_type);
665 vmode = fxdr_unsigned(u_short, fp->fa_mode);
666 rmajor = (int)fxdr_unsigned(int, fp->fa3_rdev.specdata1);
667 rminor = (int)fxdr_unsigned(int, fp->fa3_rdev.specdata2);
668 fxdr_nfsv3time(&fp->fa3_mtime, &mtime);
670 vtyp = nfsv2tov_type(fp->fa_type);
671 vmode = fxdr_unsigned(u_short, fp->fa_mode);
675 * The duplicate information returned in fa_type and fa_mode
676 * is an ambiguity in the NFS version 2 protocol.
678 * VREG should be taken literally as a regular file. If a
679 * server intents to return some type information differently
680 * in the upper bits of the mode field (e.g. for sockets, or
681 * FIFOs), NFSv2 mandates fa_type to be VNON. Anyway, we
682 * leave the examination of the mode bits even in the VREG
683 * case to avoid breakage for bogus servers, but we make sure
684 * that there are actually type bits set in the upper part of
685 * fa_mode (and failing that, trust the va_type field).
687 * NFSv3 cleared the issue, and requires fa_mode to not
688 * contain any type information (while also introduing sockets
689 * and FIFOs for fa_type).
691 if (vtyp == VNON || (vtyp == VREG && (vmode & S_IFMT) != 0))
692 vtyp = IFTOVT(vmode);
693 rdev = fxdr_unsigned(int32_t, fp->fa2_rdev);
694 rmajor = umajor(rdev);
695 rminor = uminor(rdev);
696 fxdr_nfsv2time(&fp->fa2_mtime, &mtime);
699 * Really ugly NFSv2 kludge.
701 if (vtyp == VCHR && rdev == (udev_t)0xffffffff)
706 * If v_type == VNON it is a new node, so fill in the v_type,
707 * n_mtime fields. Check to see if it represents a special
708 * device, and if so, check for a possible alias. Once the
709 * correct vnode has been obtained, fill in the rest of the
713 if (vp->v_type != vtyp) {
714 nfs_setvtype(vp, vtyp);
715 if (vp->v_type == VFIFO) {
716 vp->v_ops = &vp->v_mount->mnt_vn_fifo_ops;
717 } else if (vp->v_type == VCHR || vp->v_type == VBLK) {
718 vp->v_ops = &vp->v_mount->mnt_vn_spec_ops;
719 addaliasu(vp, rmajor, rminor);
721 vp->v_ops = &vp->v_mount->mnt_vn_use_ops;
723 np->n_mtime = mtime.tv_sec;
724 } else if (np->n_mtime != mtime.tv_sec) {
726 * If we haven't modified the file locally and the server
727 * timestamp does not match, then the server probably
728 * modified the file. We must flag this condition so
729 * the proper syncnronization can be done. We do not
730 * try to synchronize the state here because that
731 * could lead to an endless recursion.
733 * XXX loadattrcache can be set during the reply to a write,
734 * before the write timestamp is properly processed. To
735 * avoid unconditionally setting the rmodified bit (which
736 * has the effect of flushing the cache), we only do this
737 * check if the lmodified bit is not set.
739 np->n_mtime = mtime.tv_sec;
740 if ((lattr_flags & NFS_LATTR_NOMTIMECHECK) == 0)
741 np->n_flag |= NRMODIFIED;
745 vap->va_mode = (vmode & 07777);
746 vap->va_rmajor = rmajor;
747 vap->va_rminor = rminor;
748 vap->va_mtime = mtime;
749 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
751 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
752 vap->va_uid = fxdr_unsigned(uid_t, fp->fa_uid);
753 vap->va_gid = fxdr_unsigned(gid_t, fp->fa_gid);
754 vap->va_size = fxdr_hyper(&fp->fa3_size);
755 vap->va_blocksize = NFS_FABLKSIZE;
756 vap->va_bytes = fxdr_hyper(&fp->fa3_used);
757 vap->va_fileid = fxdr_hyper(&fp->fa3_fileid);
758 fxdr_nfsv3time(&fp->fa3_atime, &vap->va_atime);
759 fxdr_nfsv3time(&fp->fa3_ctime, &vap->va_ctime);
763 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
764 vap->va_uid = fxdr_unsigned(uid_t, fp->fa_uid);
765 vap->va_gid = fxdr_unsigned(gid_t, fp->fa_gid);
766 vap->va_size = fxdr_unsigned(u_int32_t, fp->fa2_size);
767 vap->va_blocksize = fxdr_unsigned(int32_t, fp->fa2_blocksize);
768 vap->va_bytes = (u_quad_t)fxdr_unsigned(int32_t, fp->fa2_blocks)
770 vap->va_fileid = fxdr_unsigned(int32_t, fp->fa2_fileid);
771 fxdr_nfsv2time(&fp->fa2_atime, &vap->va_atime);
773 vap->va_ctime.tv_sec = fxdr_unsigned(u_int32_t,
774 fp->fa2_ctime.nfsv2_sec);
775 vap->va_ctime.tv_nsec = 0;
776 vap->va_gen = fxdr_unsigned(u_int32_t,fp->fa2_ctime.nfsv2_usec);
779 np->n_attrstamp = time_second;
780 if (vap->va_size != np->n_size) {
781 if (vap->va_type == VREG) {
782 if ((lattr_flags & NFS_LATTR_NOSHRINK) &&
783 vap->va_size < np->n_size) {
785 * We've been told not to shrink the file;
786 * zero np->n_attrstamp to indicate that
787 * the attributes are stale.
789 * This occurs primarily due to recursive
790 * NFS ops that are executed during periods
791 * where we cannot safely reduce the size of
794 * Additionally, write rpcs are broken down
795 * into buffers and np->n_size is
796 * pre-extended. Setting NRMODIFIED here
797 * can result in n_size getting reset to a
798 * lower value, which is NOT what we want.
799 * XXX this needs to be cleaned up a lot
802 vap->va_size = np->n_size;
804 if ((np->n_flag & NLMODIFIED) == 0)
805 np->n_flag |= NRMODIFIED;
806 } else if (np->n_flag & NLMODIFIED) {
808 * We've modified the file: Use the larger
809 * of our size, and the server's size. At
810 * this point the cache coherency is all
811 * shot to hell. To try to handle multiple
812 * clients appending to the file at the same
813 * time mark that the server has changed
814 * the file if the server's notion of the
815 * file size is larger then our notion.
817 * XXX this needs work.
819 if (vap->va_size < np->n_size) {
820 vap->va_size = np->n_size;
822 np->n_size = vap->va_size;
823 np->n_flag |= NRMODIFIED;
827 * Someone changed the file's size on the
828 * server and there are no local changes
829 * to get in the way, set the size and mark
832 np->n_size = vap->va_size;
833 np->n_flag |= NRMODIFIED;
835 vnode_pager_setsize(vp, np->n_size);
837 np->n_size = vap->va_size;
841 bcopy((caddr_t)vap, (caddr_t)vaper, sizeof(*vap));
842 if (np->n_flag & NCHG) {
843 if (np->n_flag & NACC)
844 vaper->va_atime = np->n_atim;
845 if (np->n_flag & NUPD)
846 vaper->va_mtime = np->n_mtim;
853 #include <sys/sysctl.h>
854 SYSCTL_DECL(_vfs_nfs);
855 static int nfs_acdebug;
856 SYSCTL_INT(_vfs_nfs, OID_AUTO, acdebug, CTLFLAG_RW, &nfs_acdebug, 0, "");
860 * Check the time stamp
861 * If the cache is valid, copy contents to *vap and return 0
862 * otherwise return an error
865 nfs_getattrcache(struct vnode *vp, struct vattr *vaper)
869 struct nfsmount *nmp;
874 nmp = VFSTONFS(vp->v_mount);
877 * Dynamic timeout based on how recently the file was modified.
878 * n_mtime is always valid.
880 timeo = (get_approximate_time_t() - np->n_mtime) / 60;
884 kprintf("nfs_getattrcache: initial timeo = %d\n", timeo);
887 if (vap->va_type == VDIR) {
888 if ((np->n_flag & NLMODIFIED) || timeo < nmp->nm_acdirmin)
889 timeo = nmp->nm_acdirmin;
890 else if (timeo > nmp->nm_acdirmax)
891 timeo = nmp->nm_acdirmax;
893 if ((np->n_flag & NLMODIFIED) || timeo < nmp->nm_acregmin)
894 timeo = nmp->nm_acregmin;
895 else if (timeo > nmp->nm_acregmax)
896 timeo = nmp->nm_acregmax;
901 kprintf("acregmin %d; acregmax %d; acdirmin %d; acdirmax %d\n",
902 nmp->nm_acregmin, nmp->nm_acregmax,
903 nmp->nm_acdirmin, nmp->nm_acdirmax);
906 kprintf("nfs_getattrcache: age = %d; final timeo = %d\n",
907 (int)(time_second - np->n_attrstamp), timeo);
910 if (np->n_attrstamp == 0 || (time_second - np->n_attrstamp) >= timeo) {
911 nfsstats.attrcache_misses++;
914 nfsstats.attrcache_hits++;
917 * Our attribute cache can be stale due to modifications made on
918 * this host. XXX this is a bad hack. We need a more deterministic
919 * means of finding out which np fields are valid verses attr cache
920 * fields. We really should update the vattr info on the fly when
921 * making local changes.
923 if (vap->va_size != np->n_size) {
924 if (vap->va_type == VREG) {
925 if (np->n_flag & NLMODIFIED) {
926 if (vap->va_size < np->n_size)
927 vap->va_size = np->n_size;
929 np->n_size = vap->va_size;
931 np->n_size = vap->va_size;
933 vnode_pager_setsize(vp, np->n_size);
935 np->n_size = vap->va_size;
938 bcopy((caddr_t)vap, (caddr_t)vaper, sizeof(struct vattr));
939 if (np->n_flag & NCHG) {
940 if (np->n_flag & NACC)
941 vaper->va_atime = np->n_atim;
942 if (np->n_flag & NUPD)
943 vaper->va_mtime = np->n_mtim;
951 * Set up nameidata for a lookup() call and do it.
953 * If pubflag is set, this call is done for a lookup operation on the
954 * public filehandle. In that case we allow crossing mountpoints and
955 * absolute pathnames. However, the caller is expected to check that
956 * the lookup result is within the public fs, and deny access if
959 * dirp may be set whether an error is returned or not, and must be
960 * released by the caller.
962 * On return nd->nl_nch usually points to the target ncp, which may represent
965 * NOTE: the caller must call nlookup_done(nd) unconditionally on return
969 nfs_namei(struct nlookupdata *nd, struct ucred *cred, int nflags,
970 struct vnode **dvpp, struct vnode **vpp,
971 fhandle_t *fhp, int len,
972 struct nfssvc_sock *slp, struct sockaddr *nam, struct mbuf **mdp,
973 caddr_t *dposp, struct vnode **dirpp, struct thread *td,
974 int kerbflag, int pubflag)
978 char *fromcp, *tocp, *cp;
985 namebuf = objcache_get(namei_oc, M_WAITOK);
989 * Copy the name from the mbuf list to namebuf.
994 rem = mtod(md, caddr_t) + md->m_len - fromcp;
995 for (i = 0; i < len; i++) {
1002 fromcp = mtod(md, caddr_t);
1005 if (*fromcp == '\0' || (!pubflag && *fromcp == '/')) {
1009 *tocp++ = *fromcp++;
1015 len = nfsm_rndup(len)-len;
1019 else if ((error = nfs_adv(mdp, dposp, len, rem)) != 0)
1024 * Extract and set starting directory. The returned dp is refd
1027 error = nfsrv_fhtovp(fhp, FALSE, &mp, &dp, cred, slp,
1028 nam, &rdonly, kerbflag, pubflag);
1031 if (dp->v_type != VDIR) {
1038 * Set return directory. Reference to dp is implicitly transfered
1039 * to the returned pointer. This must be set before we potentially
1046 * Oh joy. For WebNFS, handle those pesky '%' escapes,
1047 * and the 'native path' indicator.
1049 cp = objcache_get(namei_oc, M_WAITOK);
1052 if ((unsigned char)*fromcp >= WEBNFS_SPECCHAR_START) {
1053 switch ((unsigned char)*fromcp) {
1054 case WEBNFS_NATIVE_CHAR:
1056 * 'Native' path for us is the same
1057 * as a path according to the NFS spec,
1058 * just skip the escape char.
1063 * More may be added in the future, range 0x80-0xff
1067 objcache_put(namei_oc, cp);
1072 * Translate the '%' escapes, URL-style.
1074 while (*fromcp != '\0') {
1075 if (*fromcp == WEBNFS_ESC_CHAR) {
1076 if (fromcp[1] != '\0' && fromcp[2] != '\0') {
1078 *tocp++ = HEXSTRTOI(fromcp);
1083 objcache_put(namei_oc, cp);
1087 *tocp++ = *fromcp++;
1090 objcache_put(namei_oc, namebuf);
1095 * Setup for search. We need to get a start directory from dp. Note
1096 * that dp is ref'd, but we no longer 'own' the ref (*dirpp owns it).
1099 nflags |= NLC_NFS_NOSOFTLINKTRAV;
1100 nflags |= NLC_NOCROSSMOUNT;
1103 nflags |= NLC_NFS_RDONLY;
1106 * We need a starting ncp from the directory vnode dp. dp must not
1107 * be locked. The returned ncp will be refd but not locked.
1109 * If no suitable ncp is found we instruct cache_fromdvp() to create
1110 * one. If this fails the directory has probably been removed while
1111 * the target was chdir'd into it and any further lookup will fail.
1113 if ((error = cache_fromdvp(dp, cred, 1, &nch)) != 0)
1115 nlookup_init_raw(nd, namebuf, UIO_SYSSPACE, nflags, cred, &nch);
1119 * Ok, do the lookup.
1121 error = nlookup(nd);
1124 * If no error occured return the requested dvpp and vpp. If
1125 * NLC_CREATE was specified nd->nl_nch may represent a negative
1126 * cache hit in which case we do not attempt to obtain the vp.
1130 if (nd->nl_nch.ncp->nc_parent) {
1132 nch.ncp = nch.ncp->nc_parent;
1133 error = cache_vget(&nch, nd->nl_cred,
1134 LK_EXCLUSIVE, dvpp);
1139 if (vpp && nd->nl_nch.ncp->nc_vp) {
1140 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, vpp);
1143 if (dvpp && *dvpp) {
1158 objcache_put(namei_oc, namebuf);
1163 * nfsrv_fhtovp() - convert a fh to a vnode ptr (optionally locked)
1164 * - look up fsid in mount list (if not found ret error)
1165 * - get vp and export rights by calling VFS_FHTOVP()
1166 * - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
1167 * - if not lockflag unlock it with vn_unlock()
1170 nfsrv_fhtovp(fhandle_t *fhp, int lockflag,
1171 struct mount **mpp, struct vnode **vpp,
1172 struct ucred *cred, struct nfssvc_sock *slp, struct sockaddr *nam,
1173 int *rdonlyp, int kerbflag, int pubflag)
1177 struct ucred *credanon;
1179 #ifdef MNT_EXNORESPORT /* XXX needs mountd and /etc/exports help yet */
1180 struct sockaddr_int *saddr;
1186 if (nfs_ispublicfh(fhp)) {
1187 if (!pubflag || !nfs_pub.np_valid)
1189 fhp = &nfs_pub.np_handle;
1192 mp = *mpp = vfs_getvfs(&fhp->fh_fsid);
1195 error = VFS_CHECKEXP(mp, nam, &exflags, &credanon);
1198 error = VFS_FHTOVP(mp, NULL, &fhp->fh_fid, vpp);
1201 #ifdef MNT_EXNORESPORT
1202 if (!(exflags & (MNT_EXNORESPORT|MNT_EXPUBLIC))) {
1203 saddr = (struct sockaddr_in *)nam;
1204 if (saddr->sin_family == AF_INET &&
1205 ntohs(saddr->sin_port) >= IPPORT_RESERVED) {
1208 return (NFSERR_AUTHERR | AUTH_TOOWEAK);
1213 * Check/setup credentials.
1215 if (exflags & MNT_EXKERB) {
1219 return (NFSERR_AUTHERR | AUTH_TOOWEAK);
1221 } else if (kerbflag) {
1224 return (NFSERR_AUTHERR | AUTH_TOOWEAK);
1225 } else if (cred->cr_uid == 0 || (exflags & MNT_EXPORTANON)) {
1226 cred->cr_uid = credanon->cr_uid;
1227 for (i = 0; i < credanon->cr_ngroups && i < NGROUPS; i++)
1228 cred->cr_groups[i] = credanon->cr_groups[i];
1229 cred->cr_ngroups = i;
1231 if (exflags & MNT_EXRDONLY)
1242 * WebNFS: check if a filehandle is a public filehandle. For v3, this
1243 * means a length of 0, for v2 it means all zeroes. nfsm_srvmtofh has
1244 * transformed this to all zeroes in both cases, so check for it.
1247 nfs_ispublicfh(fhandle_t *fhp)
1249 char *cp = (char *)fhp;
1252 for (i = 0; i < NFSX_V3FH; i++)
1258 #endif /* NFS_NOSERVER */
1260 * This function compares two net addresses by family and returns TRUE
1261 * if they are the same host.
1262 * If there is any doubt, return FALSE.
1263 * The AF_INET family is handled as a special case so that address mbufs
1264 * don't need to be saved to store "struct in_addr", which is only 4 bytes.
1267 netaddr_match(int family, union nethostaddr *haddr, struct sockaddr *nam)
1269 struct sockaddr_in *inetaddr;
1273 inetaddr = (struct sockaddr_in *)nam;
1274 if (inetaddr->sin_family == AF_INET &&
1275 inetaddr->sin_addr.s_addr == haddr->had_inetaddr)
1284 static nfsuint64 nfs_nullcookie = { { 0, 0 } };
1286 * This function finds the directory cookie that corresponds to the
1287 * logical byte offset given.
1290 nfs_getcookie(struct nfsnode *np, off_t off, int add)
1292 struct nfsdmap *dp, *dp2;
1295 pos = (uoff_t)off / NFS_DIRBLKSIZ;
1296 if (pos == 0 || off < 0) {
1299 panic("nfs getcookie add at <= 0");
1301 return (&nfs_nullcookie);
1304 dp = np->n_cookies.lh_first;
1307 MALLOC(dp, struct nfsdmap *, sizeof (struct nfsdmap),
1308 M_NFSDIROFF, M_WAITOK);
1309 dp->ndm_eocookie = 0;
1310 LIST_INSERT_HEAD(&np->n_cookies, dp, ndm_list);
1314 while (pos >= NFSNUMCOOKIES) {
1315 pos -= NFSNUMCOOKIES;
1316 if (dp->ndm_list.le_next) {
1317 if (!add && dp->ndm_eocookie < NFSNUMCOOKIES &&
1318 pos >= dp->ndm_eocookie)
1320 dp = dp->ndm_list.le_next;
1322 MALLOC(dp2, struct nfsdmap *, sizeof (struct nfsdmap),
1323 M_NFSDIROFF, M_WAITOK);
1324 dp2->ndm_eocookie = 0;
1325 LIST_INSERT_AFTER(dp, dp2, ndm_list);
1330 if (pos >= dp->ndm_eocookie) {
1332 dp->ndm_eocookie = pos + 1;
1336 return (&dp->ndm_cookies[pos]);
1340 * Invalidate cached directory information, except for the actual directory
1341 * blocks (which are invalidated separately).
1342 * Done mainly to avoid the use of stale offset cookies.
1345 nfs_invaldir(struct vnode *vp)
1347 struct nfsnode *np = VTONFS(vp);
1350 if (vp->v_type != VDIR)
1351 panic("nfs: invaldir not dir");
1353 np->n_direofoffset = 0;
1354 np->n_cookieverf.nfsuquad[0] = 0;
1355 np->n_cookieverf.nfsuquad[1] = 0;
1356 if (np->n_cookies.lh_first)
1357 np->n_cookies.lh_first->ndm_eocookie = 0;
1361 * Set the v_type field for an NFS client's vnode and initialize for
1362 * buffer cache operations if necessary.
1365 nfs_setvtype(struct vnode *vp, enum vtype vtyp)
1373 vinitvmio(vp, 0); /* needs VMIO, size not yet known */
1381 * The write verifier has changed (probably due to a server reboot), so all
1382 * B_NEEDCOMMIT blocks will have to be written again. Since they are on the
1383 * dirty block list as B_DELWRI, all this takes is clearing the B_NEEDCOMMIT
1384 * and B_CLUSTEROK flags. Once done the new write verifier can be set for the
1387 * B_CLUSTEROK must be cleared along with B_NEEDCOMMIT because stage 1 data
1388 * writes are not clusterable.
1391 static int nfs_clearcommit_bp(struct buf *bp, void *data __unused);
1392 static int nfs_clearcommit_callback(struct mount *mp, struct vnode *vp,
1393 void *data __unused);
1396 nfs_clearcommit(struct mount *mp)
1398 vmntvnodescan(mp, VMSC_NOWAIT, nfs_clearcommit_callback, NULL, NULL);
1402 nfs_clearcommit_callback(struct mount *mp, struct vnode *vp,
1403 void *data __unused)
1408 lwkt_gettoken(&vlock, &vp->v_token);
1409 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
1410 nfs_clearcommit_bp, NULL);
1411 lwkt_reltoken(&vlock);
1417 nfs_clearcommit_bp(struct buf *bp, void *data __unused)
1419 if (BUF_REFCNT(bp) == 0 &&
1420 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
1421 == (B_DELWRI | B_NEEDCOMMIT)) {
1422 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
1427 #ifndef NFS_NOSERVER
1429 * Map errnos to NFS error numbers. For Version 3 also filter out error
1430 * numbers not specified for the associated procedure.
1433 nfsrv_errmap(struct nfsrv_descript *nd, int err)
1435 short *defaulterrp, *errp;
1437 if (nd->nd_flag & ND_NFSV3) {
1438 if (nd->nd_procnum <= NFSPROC_COMMIT) {
1439 errp = defaulterrp = nfsrv_v3errmap[nd->nd_procnum];
1443 else if (*errp > err)
1446 return ((int)*defaulterrp);
1448 return (err & 0xffff);
1451 return ((int)nfsrv_v2errmap[err - 1]);
1456 * Sort the group list in increasing numerical order.
1457 * (Insertion sort by Chris Torek, who was grossed out by the bubble sort
1458 * that used to be here.)
1461 nfsrvw_sort(gid_t *list, int num)
1466 /* Insertion sort. */
1467 for (i = 1; i < num; i++) {
1469 /* find correct slot for value v, moving others up */
1470 for (j = i; --j >= 0 && v < list[j];)
1471 list[j + 1] = list[j];
1477 * copy credentials making sure that the result can be compared with bcmp().
1480 nfsrv_setcred(struct ucred *incred, struct ucred *outcred)
1484 bzero((caddr_t)outcred, sizeof (struct ucred));
1485 outcred->cr_ref = 1;
1486 outcred->cr_uid = incred->cr_uid;
1487 outcred->cr_ngroups = incred->cr_ngroups;
1488 for (i = 0; i < incred->cr_ngroups; i++)
1489 outcred->cr_groups[i] = incred->cr_groups[i];
1490 nfsrvw_sort(outcred->cr_groups, outcred->cr_ngroups);
1492 #endif /* NFS_NOSERVER */