The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/nfsclient/nfs_vnops.c

Version: -  FREEBSD  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-2  -  FREEBSD-11-1  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-4  -  FREEBSD-10-3  -  FREEBSD-10-2  -  FREEBSD-10-1  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-3  -  FREEBSD-9-2  -  FREEBSD-9-1  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-4  -  FREEBSD-8-3  -  FREEBSD-8-2  -  FREEBSD-8-1  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-4  -  FREEBSD-7-3  -  FREEBSD-7-2  -  FREEBSD-7-1  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-4  -  FREEBSD-6-3  -  FREEBSD-6-2  -  FREEBSD-6-1  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-5  -  FREEBSD-5-4  -  FREEBSD-5-3  -  FREEBSD-5-2  -  FREEBSD-5-1  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  linux-2.6  -  linux-2.4.22  -  MK83  -  MK84  -  PLAN9  -  DFBSD  -  NETBSD  -  NETBSD5  -  NETBSD4  -  NETBSD3  -  NETBSD20  -  OPENBSD  -  xnu-517  -  xnu-792  -  xnu-792.6.70  -  xnu-1228  -  xnu-1456.1.26  -  xnu-1699.24.8  -  xnu-2050.18.24  -  OPENSOLARIS  -  minix-3-1-1 
SearchContext: -  none  -  3  -  10 

    1 /*
    2  * Copyright (c) 1989, 1993
    3  *      The Regents of the University of California.  All rights reserved.
    4  *
    5  * This code is derived from software contributed to Berkeley by
    6  * Rick Macklem at The University of Guelph.
    7  *
    8  * Redistribution and use in source and binary forms, with or without
    9  * modification, are permitted provided that the following conditions
   10  * are met:
   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.
   23  *
   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
   34  * SUCH DAMAGE.
   35  *
   36  *      @(#)nfs_vnops.c 8.16 (Berkeley) 5/27/95
   37  */
   38 
   39 #include <sys/cdefs.h>
   40 __FBSDID("$FreeBSD: releng/5.2/sys/nfsclient/nfs_vnops.c 122698 2003-11-14 20:54:10Z alfred $");
   41 
   42 /*
   43  * vnode op calls for Sun NFS version 2 and 3
   44  */
   45 
   46 #include "opt_inet.h"
   47 
   48 #include <sys/param.h>
   49 #include <sys/kernel.h>
   50 #include <sys/systm.h>
   51 #include <sys/resourcevar.h>
   52 #include <sys/proc.h>
   53 #include <sys/mount.h>
   54 #include <sys/bio.h>
   55 #include <sys/buf.h>
   56 #include <sys/malloc.h>
   57 #include <sys/mbuf.h>
   58 #include <sys/namei.h>
   59 #include <sys/socket.h>
   60 #include <sys/vnode.h>
   61 #include <sys/dirent.h>
   62 #include <sys/fcntl.h>
   63 #include <sys/lockf.h>
   64 #include <sys/stat.h>
   65 #include <sys/sysctl.h>
   66 
   67 #include <vm/vm.h>
   68 #include <vm/vm_extern.h>
   69 
   70 #include <fs/fifofs/fifo.h>
   71 
   72 #include <rpc/rpcclnt.h>
   73 
   74 #include <nfs/rpcv2.h>
   75 #include <nfs/nfsproto.h>
   76 #include <nfsclient/nfs.h>
   77 #include <nfsclient/nfsnode.h>
   78 #include <nfsclient/nfsmount.h>
   79 #include <nfsclient/nfs_lock.h>
   80 #include <nfs/xdr_subs.h>
   81 #include <nfsclient/nfsm_subs.h>
   82 
   83 #include <net/if.h>
   84 #include <netinet/in.h>
   85 #include <netinet/in_var.h>
   86 
   87 /* Defs */
   88 #define TRUE    1
   89 #define FALSE   0
   90 
   91 /*
   92  * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
   93  * calls are not in getblk() and brelse() so that they would not be necessary
   94  * here.
   95  */
   96 #ifndef B_VMIO
   97 #define vfs_busy_pages(bp, f)
   98 #endif
   99 
  100 static int      nfsspec_read(struct vop_read_args *);
  101 static int      nfsspec_write(struct vop_write_args *);
  102 static int      nfsfifo_read(struct vop_read_args *);
  103 static int      nfsfifo_write(struct vop_write_args *);
  104 static int      nfsspec_close(struct vop_close_args *);
  105 static int      nfsfifo_close(struct vop_close_args *);
  106 static int      nfs_flush(struct vnode *, struct ucred *, int, struct thread *,
  107                     int);
  108 static int      nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
  109                     struct thread *);
  110 static  int     nfs_lookup(struct vop_lookup_args *);
  111 static  int     nfs_create(struct vop_create_args *);
  112 static  int     nfs_mknod(struct vop_mknod_args *);
  113 static  int     nfs_open(struct vop_open_args *);
  114 static  int     nfs_close(struct vop_close_args *);
  115 static  int     nfs_access(struct vop_access_args *);
  116 static  int     nfs_getattr(struct vop_getattr_args *);
  117 static  int     nfs_setattr(struct vop_setattr_args *);
  118 static  int     nfs_read(struct vop_read_args *);
  119 static  int     nfs_fsync(struct vop_fsync_args *);
  120 static  int     nfs_remove(struct vop_remove_args *);
  121 static  int     nfs_link(struct vop_link_args *);
  122 static  int     nfs_rename(struct vop_rename_args *);
  123 static  int     nfs_mkdir(struct vop_mkdir_args *);
  124 static  int     nfs_rmdir(struct vop_rmdir_args *);
  125 static  int     nfs_symlink(struct vop_symlink_args *);
  126 static  int     nfs_readdir(struct vop_readdir_args *);
  127 static  int     nfs_strategy(struct vop_strategy_args *);
  128 static  int     nfs_lookitup(struct vnode *, const char *, int,
  129                     struct ucred *, struct thread *, struct nfsnode **);
  130 static  int     nfs_sillyrename(struct vnode *, struct vnode *,
  131                     struct componentname *);
  132 static int      nfsspec_access(struct vop_access_args *);
  133 static int      nfs_readlink(struct vop_readlink_args *);
  134 static int      nfs_print(struct vop_print_args *);
  135 static int      nfs_advlock(struct vop_advlock_args *);
  136 
  137 /*
  138  * Global vfs data structures for nfs
  139  */
  140 vop_t **nfs_vnodeop_p;
  141 static struct vnodeopv_entry_desc nfs_vnodeop_entries[] = {
  142         { &vop_default_desc,            (vop_t *) vop_defaultop },
  143         { &vop_access_desc,             (vop_t *) nfs_access },
  144         { &vop_advlock_desc,            (vop_t *) nfs_advlock },
  145         { &vop_close_desc,              (vop_t *) nfs_close },
  146         { &vop_create_desc,             (vop_t *) nfs_create },
  147         { &vop_fsync_desc,              (vop_t *) nfs_fsync },
  148         { &vop_getattr_desc,            (vop_t *) nfs_getattr },
  149         { &vop_getpages_desc,           (vop_t *) nfs_getpages },
  150         { &vop_putpages_desc,           (vop_t *) nfs_putpages },
  151         { &vop_inactive_desc,           (vop_t *) nfs_inactive },
  152         { &vop_lease_desc,              (vop_t *) vop_null },
  153         { &vop_link_desc,               (vop_t *) nfs_link },
  154         { &vop_lookup_desc,             (vop_t *) nfs_lookup },
  155         { &vop_mkdir_desc,              (vop_t *) nfs_mkdir },
  156         { &vop_mknod_desc,              (vop_t *) nfs_mknod },
  157         { &vop_open_desc,               (vop_t *) nfs_open },
  158         { &vop_print_desc,              (vop_t *) nfs_print },
  159         { &vop_read_desc,               (vop_t *) nfs_read },
  160         { &vop_readdir_desc,            (vop_t *) nfs_readdir },
  161         { &vop_readlink_desc,           (vop_t *) nfs_readlink },
  162         { &vop_reclaim_desc,            (vop_t *) nfs_reclaim },
  163         { &vop_remove_desc,             (vop_t *) nfs_remove },
  164         { &vop_rename_desc,             (vop_t *) nfs_rename },
  165         { &vop_rmdir_desc,              (vop_t *) nfs_rmdir },
  166         { &vop_setattr_desc,            (vop_t *) nfs_setattr },
  167         { &vop_strategy_desc,           (vop_t *) nfs_strategy },
  168         { &vop_symlink_desc,            (vop_t *) nfs_symlink },
  169         { &vop_write_desc,              (vop_t *) nfs_write },
  170         { NULL, NULL }
  171 };
  172 static struct vnodeopv_desc nfs_vnodeop_opv_desc =
  173         { &nfs_vnodeop_p, nfs_vnodeop_entries };
  174 VNODEOP_SET(nfs_vnodeop_opv_desc);
  175 
  176 /*
  177  * Special device vnode ops
  178  */
  179 vop_t **spec_nfsnodeop_p;
  180 static struct vnodeopv_entry_desc nfs_specop_entries[] = {
  181         { &vop_default_desc,            (vop_t *) spec_vnoperate },
  182         { &vop_access_desc,             (vop_t *) nfsspec_access },
  183         { &vop_close_desc,              (vop_t *) nfsspec_close },
  184         { &vop_fsync_desc,              (vop_t *) nfs_fsync },
  185         { &vop_getattr_desc,            (vop_t *) nfs_getattr },
  186         { &vop_inactive_desc,           (vop_t *) nfs_inactive },
  187         { &vop_print_desc,              (vop_t *) nfs_print },
  188         { &vop_read_desc,               (vop_t *) nfsspec_read },
  189         { &vop_reclaim_desc,            (vop_t *) nfs_reclaim },
  190         { &vop_setattr_desc,            (vop_t *) nfs_setattr },
  191         { &vop_write_desc,              (vop_t *) nfsspec_write },
  192         { NULL, NULL }
  193 };
  194 static struct vnodeopv_desc spec_nfsnodeop_opv_desc =
  195         { &spec_nfsnodeop_p, nfs_specop_entries };
  196 VNODEOP_SET(spec_nfsnodeop_opv_desc);
  197 
  198 vop_t **fifo_nfsnodeop_p;
  199 static struct vnodeopv_entry_desc nfs_fifoop_entries[] = {
  200         { &vop_default_desc,            (vop_t *) fifo_vnoperate },
  201         { &vop_access_desc,             (vop_t *) nfsspec_access },
  202         { &vop_close_desc,              (vop_t *) nfsfifo_close },
  203         { &vop_fsync_desc,              (vop_t *) nfs_fsync },
  204         { &vop_getattr_desc,            (vop_t *) nfs_getattr },
  205         { &vop_inactive_desc,           (vop_t *) nfs_inactive },
  206         { &vop_print_desc,              (vop_t *) nfs_print },
  207         { &vop_read_desc,               (vop_t *) nfsfifo_read },
  208         { &vop_reclaim_desc,            (vop_t *) nfs_reclaim },
  209         { &vop_setattr_desc,            (vop_t *) nfs_setattr },
  210         { &vop_write_desc,              (vop_t *) nfsfifo_write },
  211         { NULL, NULL }
  212 };
  213 static struct vnodeopv_desc fifo_nfsnodeop_opv_desc =
  214         { &fifo_nfsnodeop_p, nfs_fifoop_entries };
  215 VNODEOP_SET(fifo_nfsnodeop_opv_desc);
  216 
  217 static int      nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
  218                              struct componentname *cnp, struct vattr *vap);
  219 static int      nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
  220                               struct ucred *cred, struct thread *td);
  221 static int      nfs_renamerpc(struct vnode *fdvp, const char *fnameptr,
  222                               int fnamelen, struct vnode *tdvp,
  223                               const char *tnameptr, int tnamelen,
  224                               struct ucred *cred, struct thread *td);
  225 static int      nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
  226                              struct sillyrename *sp);
  227 
  228 /*
  229  * Global variables
  230  */
  231 struct proc     *nfs_iodwant[NFS_MAXASYNCDAEMON];
  232 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
  233 int              nfs_numasync = 0;
  234 #define DIRHDSIZ        (sizeof (struct dirent) - (MAXNAMLEN + 1))
  235 
  236 SYSCTL_DECL(_vfs_nfs);
  237 
  238 static int      nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
  239 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
  240            &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
  241 
  242 static int      nfsv3_commit_on_close = 0;
  243 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
  244            &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
  245 #if 0
  246 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
  247            &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
  248 
  249 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
  250            &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
  251 #endif
  252 
  253 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY          \
  254                          | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE     \
  255                          | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
  256 static int
  257 nfs3_access_otw(struct vnode *vp, int wmode, struct thread *td,
  258     struct ucred *cred)
  259 {
  260         const int v3 = 1;
  261         u_int32_t *tl;
  262         int error = 0, attrflag;
  263 
  264         struct mbuf *mreq, *mrep, *md, *mb;
  265         caddr_t bpos, dpos;
  266         u_int32_t rmode;
  267         struct nfsnode *np = VTONFS(vp);
  268 
  269         nfsstats.rpccnt[NFSPROC_ACCESS]++;
  270         mreq = nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
  271         mb = mreq;
  272         bpos = mtod(mb, caddr_t);
  273         nfsm_fhtom(vp, v3);
  274         tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
  275         *tl = txdr_unsigned(wmode);
  276         nfsm_request(vp, NFSPROC_ACCESS, td, cred);
  277         nfsm_postop_attr(vp, attrflag);
  278         if (!error) {
  279                 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
  280                 rmode = fxdr_unsigned(u_int32_t, *tl);
  281                 np->n_mode = rmode;
  282                 np->n_modeuid = cred->cr_uid;
  283                 np->n_modestamp = time_second;
  284         }
  285         m_freem(mrep);
  286 nfsmout:
  287         return (error);
  288 }
  289 
  290 /*
  291  * nfs access vnode op.
  292  * For nfs version 2, just return ok. File accesses may fail later.
  293  * For nfs version 3, use the access rpc to check accessibility. If file modes
  294  * are changed on the server, accesses might still fail later.
  295  */
  296 static int
  297 nfs_access(struct vop_access_args *ap)
  298 {
  299         struct vnode *vp = ap->a_vp;
  300         int error = 0;
  301         u_int32_t mode, wmode;
  302         int v3 = NFS_ISV3(vp);
  303         struct nfsnode *np = VTONFS(vp);
  304 
  305         /*
  306          * Disallow write attempts on filesystems mounted read-only;
  307          * unless the file is a socket, fifo, or a block or character
  308          * device resident on the filesystem.
  309          */
  310         if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
  311                 switch (vp->v_type) {
  312                 case VREG:
  313                 case VDIR:
  314                 case VLNK:
  315                         return (EROFS);
  316                 default:
  317                         break;
  318                 }
  319         }
  320         /*
  321          * For nfs v3, check to see if we have done this recently, and if
  322          * so return our cached result instead of making an ACCESS call.
  323          * If not, do an access rpc, otherwise you are stuck emulating
  324          * ufs_access() locally using the vattr. This may not be correct,
  325          * since the server may apply other access criteria such as
  326          * client uid-->server uid mapping that we do not know about.
  327          */
  328         if (v3) {
  329                 if (ap->a_mode & VREAD)
  330                         mode = NFSV3ACCESS_READ;
  331                 else
  332                         mode = 0;
  333                 if (vp->v_type != VDIR) {
  334                         if (ap->a_mode & VWRITE)
  335                                 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
  336                         if (ap->a_mode & VEXEC)
  337                                 mode |= NFSV3ACCESS_EXECUTE;
  338                 } else {
  339                         if (ap->a_mode & VWRITE)
  340                                 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
  341                                          NFSV3ACCESS_DELETE);
  342                         if (ap->a_mode & VEXEC)
  343                                 mode |= NFSV3ACCESS_LOOKUP;
  344                 }
  345                 /* XXX safety belt, only make blanket request if caching */
  346                 if (nfsaccess_cache_timeout > 0) {
  347                         wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
  348                                 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
  349                                 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
  350                 } else {
  351                         wmode = mode;
  352                 }
  353 
  354                 /*
  355                  * Does our cached result allow us to give a definite yes to
  356                  * this request?
  357                  */
  358                 if ((time_second < (np->n_modestamp + nfsaccess_cache_timeout)) &&
  359                     (ap->a_cred->cr_uid == np->n_modeuid) &&
  360                     ((np->n_mode & mode) == mode)) {
  361                         nfsstats.accesscache_hits++;
  362                 } else {
  363                         /*
  364                          * Either a no, or a don't know.  Go to the wire.
  365                          */
  366                         nfsstats.accesscache_misses++;
  367                         error = nfs3_access_otw(vp, wmode, ap->a_td,ap->a_cred);
  368                         if (!error) {
  369                                 if ((np->n_mode & mode) != mode) {
  370                                         error = EACCES;
  371                                 }
  372                         }
  373                 }
  374                 return (error);
  375         } else {
  376                 if ((error = nfsspec_access(ap)) != 0)
  377                         return (error);
  378 
  379                 /*
  380                  * Attempt to prevent a mapped root from accessing a file
  381                  * which it shouldn't.  We try to read a byte from the file
  382                  * if the user is root and the file is not zero length.
  383                  * After calling nfsspec_access, we should have the correct
  384                  * file size cached.
  385                  */
  386                 if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD)
  387                     && VTONFS(vp)->n_size > 0) {
  388                         struct iovec aiov;
  389                         struct uio auio;
  390                         char buf[1];
  391 
  392                         aiov.iov_base = buf;
  393                         aiov.iov_len = 1;
  394                         auio.uio_iov = &aiov;
  395                         auio.uio_iovcnt = 1;
  396                         auio.uio_offset = 0;
  397                         auio.uio_resid = 1;
  398                         auio.uio_segflg = UIO_SYSSPACE;
  399                         auio.uio_rw = UIO_READ;
  400                         auio.uio_td = ap->a_td;
  401 
  402                         if (vp->v_type == VREG)
  403                                 error = nfs_readrpc(vp, &auio, ap->a_cred);
  404                         else if (vp->v_type == VDIR) {
  405                                 char* bp;
  406                                 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
  407                                 aiov.iov_base = bp;
  408                                 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
  409                                 error = nfs_readdirrpc(vp, &auio, ap->a_cred);
  410                                 free(bp, M_TEMP);
  411                         } else if (vp->v_type == VLNK)
  412                                 error = nfs_readlinkrpc(vp, &auio, ap->a_cred);
  413                         else
  414                                 error = EACCES;
  415                 }
  416                 return (error);
  417         }
  418 }
  419 
  420 /*
  421  * nfs open vnode op
  422  * Check to see if the type is ok
  423  * and that deletion is not in progress.
  424  * For paged in text files, you will need to flush the page cache
  425  * if consistency is lost.
  426  */
  427 /* ARGSUSED */
  428 static int
  429 nfs_open(struct vop_open_args *ap)
  430 {
  431         struct vnode *vp = ap->a_vp;
  432         struct nfsnode *np = VTONFS(vp);
  433         struct vattr vattr;
  434         int error;
  435 
  436         if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
  437 #ifdef DIAGNOSTIC
  438                 printf("open eacces vtyp=%d\n", vp->v_type);
  439 #endif
  440                 return (EOPNOTSUPP);
  441         }
  442         /*
  443          * Get a valid lease. If cached data is stale, flush it.
  444          */
  445         if (np->n_flag & NMODIFIED) {
  446                 if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
  447                         ap->a_td, 1)) == EINTR)
  448                         return (error);
  449                 np->n_attrstamp = 0;
  450                 if (vp->v_type == VDIR)
  451                         np->n_direofoffset = 0;
  452                 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
  453                 if (error)
  454                         return (error);
  455                 np->n_mtime = vattr.va_mtime.tv_sec;
  456         } else {
  457                 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
  458                 if (error)
  459                         return (error);
  460                 if (np->n_mtime != vattr.va_mtime.tv_sec) {
  461                         if (vp->v_type == VDIR)
  462                                 np->n_direofoffset = 0;
  463                         if ((error = nfs_vinvalbuf(vp, V_SAVE,
  464                                 ap->a_cred, ap->a_td, 1)) == EINTR)
  465                                 return (error);
  466                         np->n_mtime = vattr.va_mtime.tv_sec;
  467                 }
  468         }
  469         np->n_attrstamp = 0; /* For Open/Close consistency */
  470         return (0);
  471 }
  472 
  473 /*
  474  * nfs close vnode op
  475  * What an NFS client should do upon close after writing is a debatable issue.
  476  * Most NFS clients push delayed writes to the server upon close, basically for
  477  * two reasons:
  478  * 1 - So that any write errors may be reported back to the client process
  479  *     doing the close system call. By far the two most likely errors are
  480  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
  481  * 2 - To put a worst case upper bound on cache inconsistency between
  482  *     multiple clients for the file.
  483  * There is also a consistency problem for Version 2 of the protocol w.r.t.
  484  * not being able to tell if other clients are writing a file concurrently,
  485  * since there is no way of knowing if the changed modify time in the reply
  486  * is only due to the write for this client.
  487  * (NFS Version 3 provides weak cache consistency data in the reply that
  488  *  should be sufficient to detect and handle this case.)
  489  *
  490  * The current code does the following:
  491  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
  492  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
  493  *                     or commit them (this satisfies 1 and 2 except for the
  494  *                     case where the server crashes after this close but
  495  *                     before the commit RPC, which is felt to be "good
  496  *                     enough". Changing the last argument to nfs_flush() to
  497  *                     a 1 would force a commit operation, if it is felt a
  498  *                     commit is necessary now.
  499  */
  500 /* ARGSUSED */
  501 static int
  502 nfs_close(struct vop_close_args *ap)
  503 {
  504         struct vnode *vp = ap->a_vp;
  505         struct nfsnode *np = VTONFS(vp);
  506         int error = 0;
  507 
  508         if (vp->v_type == VREG) {
  509             if (np->n_flag & NMODIFIED) {
  510                 if (NFS_ISV3(vp)) {
  511                     /*
  512                      * Under NFSv3 we have dirty buffers to dispose of.  We
  513                      * must flush them to the NFS server.  We have the option
  514                      * of waiting all the way through the commit rpc or just
  515                      * waiting for the initial write.  The default is to only
  516                      * wait through the initial write so the data is in the
  517                      * server's cache, which is roughly similar to the state
  518                      * a standard disk subsystem leaves the file in on close().
  519                      *
  520                      * We cannot clear the NMODIFIED bit in np->n_flag due to
  521                      * potential races with other processes, and certainly
  522                      * cannot clear it if we don't commit.
  523                      */
  524                     int cm = nfsv3_commit_on_close ? 1 : 0;
  525                     error = nfs_flush(vp, ap->a_cred, MNT_WAIT, ap->a_td, cm);
  526                     /* np->n_flag &= ~NMODIFIED; */
  527                 } else {
  528                     vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, ap->a_td);
  529                     error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_td, 1);
  530                     VOP_UNLOCK(vp, 0, ap->a_td);
  531                 }
  532                 np->n_attrstamp = 0;
  533             }
  534             if (np->n_flag & NWRITEERR) {
  535                 np->n_flag &= ~NWRITEERR;
  536                 error = np->n_error;
  537             }
  538         }
  539         return (error);
  540 }
  541 
  542 /*
  543  * nfs getattr call from vfs.
  544  */
  545 static int
  546 nfs_getattr(struct vop_getattr_args *ap)
  547 {
  548         struct vnode *vp = ap->a_vp;
  549         struct nfsnode *np = VTONFS(vp);
  550         caddr_t bpos, dpos;
  551         int error = 0;
  552         struct mbuf *mreq, *mrep, *md, *mb;
  553         int v3 = NFS_ISV3(vp);
  554 
  555         /*
  556          * Update local times for special files.
  557          */
  558         if (np->n_flag & (NACC | NUPD))
  559                 np->n_flag |= NCHG;
  560         /*
  561          * First look in the cache.
  562          */
  563         if (nfs_getattrcache(vp, ap->a_vap) == 0)
  564                 return (0);
  565 
  566         if (v3 && nfsaccess_cache_timeout > 0) {
  567                 nfsstats.accesscache_misses++;
  568                 nfs3_access_otw(vp, NFSV3ACCESS_ALL, ap->a_td, ap->a_cred);
  569                 if (nfs_getattrcache(vp, ap->a_vap) == 0)
  570                         return (0);
  571         }
  572 
  573         nfsstats.rpccnt[NFSPROC_GETATTR]++;
  574         mreq = nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
  575         mb = mreq;
  576         bpos = mtod(mb, caddr_t);
  577         nfsm_fhtom(vp, v3);
  578         nfsm_request(vp, NFSPROC_GETATTR, ap->a_td, ap->a_cred);
  579         if (!error) {
  580                 nfsm_loadattr(vp, ap->a_vap);
  581         }
  582         m_freem(mrep);
  583 nfsmout:
  584         return (error);
  585 }
  586 
  587 /*
  588  * nfs setattr call.
  589  */
  590 static int
  591 nfs_setattr(struct vop_setattr_args *ap)
  592 {
  593         struct vnode *vp = ap->a_vp;
  594         struct nfsnode *np = VTONFS(vp);
  595         struct vattr *vap = ap->a_vap;
  596         int error = 0;
  597         u_quad_t tsize;
  598 
  599 #ifndef nolint
  600         tsize = (u_quad_t)0;
  601 #endif
  602 
  603         /*
  604          * Setting of flags is not supported.
  605          */
  606         if (vap->va_flags != VNOVAL)
  607                 return (EOPNOTSUPP);
  608 
  609         /*
  610          * Disallow write attempts if the filesystem is mounted read-only.
  611          */
  612         if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
  613             vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
  614             vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
  615             (vp->v_mount->mnt_flag & MNT_RDONLY))
  616                 return (EROFS);
  617         if (vap->va_size != VNOVAL) {
  618                 switch (vp->v_type) {
  619                 case VDIR:
  620                         return (EISDIR);
  621                 case VCHR:
  622                 case VBLK:
  623                 case VSOCK:
  624                 case VFIFO:
  625                         if (vap->va_mtime.tv_sec == VNOVAL &&
  626                             vap->va_atime.tv_sec == VNOVAL &&
  627                             vap->va_mode == (mode_t)VNOVAL &&
  628                             vap->va_uid == (uid_t)VNOVAL &&
  629                             vap->va_gid == (gid_t)VNOVAL)
  630                                 return (0);
  631                         vap->va_size = VNOVAL;
  632                         break;
  633                 default:
  634                         /*
  635                          * Disallow write attempts if the filesystem is
  636                          * mounted read-only.
  637                          */
  638                         if (vp->v_mount->mnt_flag & MNT_RDONLY)
  639                                 return (EROFS);
  640 
  641                         /*
  642                          *  We run vnode_pager_setsize() early (why?),
  643                          * we must set np->n_size now to avoid vinvalbuf
  644                          * V_SAVE races that might setsize a lower
  645                          * value.
  646                          */
  647 
  648                         tsize = np->n_size;
  649                         error = nfs_meta_setsize(vp, ap->a_cred, 
  650                                                 ap->a_td, vap->va_size);
  651 
  652                         if (np->n_flag & NMODIFIED) {
  653                             if (vap->va_size == 0)
  654                                 error = nfs_vinvalbuf(vp, 0,
  655                                         ap->a_cred, ap->a_td, 1);
  656                             else
  657                                 error = nfs_vinvalbuf(vp, V_SAVE,
  658                                         ap->a_cred, ap->a_td, 1);
  659                             if (error) {
  660                                 vnode_pager_setsize(vp, np->n_size);
  661                                 return (error);
  662                             }
  663                         }
  664                         /*
  665                          * np->n_size has already been set to vap->va_size
  666                          * in nfs_meta_setsize(). We must set it again since
  667                          * nfs_loadattrcache() could be called through
  668                          * nfs_meta_setsize() and could modify np->n_size.
  669                          */
  670                         np->n_vattr.va_size = np->n_size = vap->va_size;
  671                 };
  672         } else if ((vap->va_mtime.tv_sec != VNOVAL ||
  673                 vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NMODIFIED) &&
  674                 vp->v_type == VREG &&
  675                 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
  676                  ap->a_td, 1)) == EINTR)
  677                 return (error);
  678         error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_td);
  679         if (error && vap->va_size != VNOVAL) {
  680                 np->n_size = np->n_vattr.va_size = tsize;
  681                 vnode_pager_setsize(vp, np->n_size);
  682         }
  683         return (error);
  684 }
  685 
  686 /*
  687  * Do an nfs setattr rpc.
  688  */
  689 static int
  690 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
  691     struct thread *td)
  692 {
  693         struct nfsv2_sattr *sp;
  694         caddr_t bpos, dpos;
  695         u_int32_t *tl;
  696         int error = 0, wccflag = NFSV3_WCCRATTR;
  697         struct mbuf *mreq, *mrep, *md, *mb;
  698         int v3 = NFS_ISV3(vp);
  699 
  700         nfsstats.rpccnt[NFSPROC_SETATTR]++;
  701         mreq = nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
  702         mb = mreq;
  703         bpos = mtod(mb, caddr_t);
  704         nfsm_fhtom(vp, v3);
  705         if (v3) {
  706                 nfsm_v3attrbuild(vap, TRUE);
  707                 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
  708                 *tl = nfs_false;
  709         } else {
  710                 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
  711                 if (vap->va_mode == (mode_t)VNOVAL)
  712                         sp->sa_mode = nfs_xdrneg1;
  713                 else
  714                         sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
  715                 if (vap->va_uid == (uid_t)VNOVAL)
  716                         sp->sa_uid = nfs_xdrneg1;
  717                 else
  718                         sp->sa_uid = txdr_unsigned(vap->va_uid);
  719                 if (vap->va_gid == (gid_t)VNOVAL)
  720                         sp->sa_gid = nfs_xdrneg1;
  721                 else
  722                         sp->sa_gid = txdr_unsigned(vap->va_gid);
  723                 sp->sa_size = txdr_unsigned(vap->va_size);
  724                 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
  725                 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
  726         }
  727         nfsm_request(vp, NFSPROC_SETATTR, td, cred);
  728         if (v3) {
  729                 nfsm_wcc_data(vp, wccflag);
  730         } else
  731                 nfsm_loadattr(vp, NULL);
  732         m_freem(mrep);
  733 nfsmout:
  734         return (error);
  735 }
  736 
  737 /*
  738  * nfs lookup call, one step at a time...
  739  * First look in cache
  740  * If not found, unlock the directory nfsnode and do the rpc
  741  */
  742 static int
  743 nfs_lookup(struct vop_lookup_args *ap)
  744 {
  745         struct componentname *cnp = ap->a_cnp;
  746         struct vnode *dvp = ap->a_dvp;
  747         struct vnode **vpp = ap->a_vpp;
  748         int flags = cnp->cn_flags;
  749         struct vnode *newvp;
  750         struct nfsmount *nmp;
  751         caddr_t bpos, dpos;
  752         struct mbuf *mreq, *mrep, *md, *mb;
  753         long len;
  754         nfsfh_t *fhp;
  755         struct nfsnode *np;
  756         int lockparent, wantparent, error = 0, attrflag, fhsize;
  757         int v3 = NFS_ISV3(dvp);
  758         struct thread *td = cnp->cn_thread;
  759 
  760         *vpp = NULLVP;
  761         cnp->cn_flags &= ~PDIRUNLOCK;
  762         if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
  763             (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
  764                 return (EROFS);
  765         if (dvp->v_type != VDIR)
  766                 return (ENOTDIR);
  767         lockparent = flags & LOCKPARENT;
  768         wantparent = flags & (LOCKPARENT|WANTPARENT);
  769         nmp = VFSTONFS(dvp->v_mount);
  770         np = VTONFS(dvp);
  771         if ((error = cache_lookup(dvp, vpp, cnp)) && error != ENOENT) {
  772                 struct vattr vattr;
  773                 int vpid;
  774 
  775                 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) {
  776                         *vpp = NULLVP;
  777                         return (error);
  778                 }
  779 
  780                 vhold(*vpp);
  781                 newvp = *vpp;
  782                 vpid = newvp->v_id;
  783                 /*
  784                  * See the comment starting `Step through' in ufs/ufs_lookup.c
  785                  * for an explanation of the locking protocol
  786                  */
  787                 if (dvp == newvp) {
  788                         VREF(newvp);
  789                         error = 0;
  790                 } else if (flags & ISDOTDOT) {
  791                         VOP_UNLOCK(dvp, 0, td);
  792                         cnp->cn_flags |= PDIRUNLOCK;
  793                         error = vget(newvp, LK_EXCLUSIVE, td);
  794                         if (!error && lockparent && (flags & ISLASTCN)) {
  795                                 error = vn_lock(dvp, LK_EXCLUSIVE, td);
  796                                 if (error == 0)
  797                                         cnp->cn_flags &= ~PDIRUNLOCK;
  798                         }
  799                 } else {
  800                         error = vget(newvp, LK_EXCLUSIVE, td);
  801                         if (!lockparent || error || !(flags & ISLASTCN)) {
  802                                 VOP_UNLOCK(dvp, 0, td);
  803                                 cnp->cn_flags |= PDIRUNLOCK;
  804                         }
  805                 }
  806                 if (!error) {
  807                         if (vpid == newvp->v_id) {
  808                            if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, td)
  809                             && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) {
  810                                 nfsstats.lookupcache_hits++;
  811                                 if (cnp->cn_nameiop != LOOKUP &&
  812                                     (flags & ISLASTCN))
  813                                         cnp->cn_flags |= SAVENAME;
  814                                 vdrop(newvp);
  815                                 return (0);
  816                            }
  817                            cache_purge(newvp);
  818                         }
  819                         vput(newvp);
  820                         if (lockparent && dvp != newvp && (flags & ISLASTCN))
  821                                 VOP_UNLOCK(dvp, 0, td);
  822                 }
  823                 vdrop(newvp);
  824                 error = vn_lock(dvp, LK_EXCLUSIVE, td);
  825                 *vpp = NULLVP;
  826                 if (error) {
  827                         cnp->cn_flags |= PDIRUNLOCK;
  828                         return (error);
  829                 }
  830                 cnp->cn_flags &= ~PDIRUNLOCK;
  831         }
  832         error = 0;
  833         newvp = NULLVP;
  834         nfsstats.lookupcache_misses++;
  835         nfsstats.rpccnt[NFSPROC_LOOKUP]++;
  836         len = cnp->cn_namelen;
  837         mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
  838                 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
  839         mb = mreq;
  840         bpos = mtod(mb, caddr_t);
  841         nfsm_fhtom(dvp, v3);
  842         nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
  843         nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_thread, cnp->cn_cred);
  844         if (error) {
  845                 nfsm_postop_attr(dvp, attrflag);
  846                 m_freem(mrep);
  847                 goto nfsmout;
  848         }
  849         nfsm_getfh(fhp, fhsize, v3);
  850 
  851         /*
  852          * Handle RENAME case...
  853          */
  854         if (cnp->cn_nameiop == RENAME && wantparent && (flags & ISLASTCN)) {
  855                 if (NFS_CMPFH(np, fhp, fhsize)) {
  856                         m_freem(mrep);
  857                         return (EISDIR);
  858                 }
  859                 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
  860                 if (error) {
  861                         m_freem(mrep);
  862                         return (error);
  863                 }
  864                 newvp = NFSTOV(np);
  865                 if (v3) {
  866                         nfsm_postop_attr(newvp, attrflag);
  867                         nfsm_postop_attr(dvp, attrflag);
  868                 } else
  869                         nfsm_loadattr(newvp, NULL);
  870                 *vpp = newvp;
  871                 m_freem(mrep);
  872                 cnp->cn_flags |= SAVENAME;
  873                 if (!lockparent) {
  874                         VOP_UNLOCK(dvp, 0, td);
  875                         cnp->cn_flags |= PDIRUNLOCK;
  876                 }
  877                 return (0);
  878         }
  879 
  880         if (flags & ISDOTDOT) {
  881                 VOP_UNLOCK(dvp, 0, td);
  882                 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
  883                 if (error) {
  884                         vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td);
  885                         return (error);
  886                 }
  887                 newvp = NFSTOV(np);
  888                 if (lockparent && (flags & ISLASTCN)) {
  889                         error = vn_lock(dvp, LK_EXCLUSIVE, td);
  890                         if (error) {
  891                                 cnp->cn_flags |= PDIRUNLOCK;
  892                                 vput(newvp);
  893                                 return (error);
  894                         }
  895                 } else
  896                         cnp->cn_flags |= PDIRUNLOCK;
  897         } else if (NFS_CMPFH(np, fhp, fhsize)) {
  898                 VREF(dvp);
  899                 newvp = dvp;
  900         } else {
  901                 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
  902                 if (error) {
  903                         m_freem(mrep);
  904                         return (error);
  905                 }
  906                 if (!lockparent || !(flags & ISLASTCN)) {
  907                         cnp->cn_flags |= PDIRUNLOCK;
  908                         VOP_UNLOCK(dvp, 0, td);
  909                 }
  910                 newvp = NFSTOV(np);
  911         }
  912         if (v3) {
  913                 nfsm_postop_attr(newvp, attrflag);
  914                 nfsm_postop_attr(dvp, attrflag);
  915         } else
  916                 nfsm_loadattr(newvp, NULL);
  917         if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
  918                 cnp->cn_flags |= SAVENAME;
  919         if ((cnp->cn_flags & MAKEENTRY) &&
  920             (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
  921                 np->n_ctime = np->n_vattr.va_ctime.tv_sec;
  922                 cache_enter(dvp, newvp, cnp);
  923         }
  924         *vpp = newvp;
  925         m_freem(mrep);
  926 nfsmout:
  927         if (error) {
  928                 if (newvp != NULLVP) {
  929                         vrele(newvp);
  930                         *vpp = NULLVP;
  931                 }
  932                 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
  933                     (flags & ISLASTCN) && error == ENOENT) {
  934                         if (!lockparent) {
  935                                 VOP_UNLOCK(dvp, 0, td);
  936                                 cnp->cn_flags |= PDIRUNLOCK;
  937                         }
  938                         if (dvp->v_mount->mnt_flag & MNT_RDONLY)
  939                                 error = EROFS;
  940                         else
  941                                 error = EJUSTRETURN;
  942                 }
  943                 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
  944                         cnp->cn_flags |= SAVENAME;
  945         }
  946         return (error);
  947 }
  948 
  949 /*
  950  * nfs read call.
  951  * Just call nfs_bioread() to do the work.
  952  */
  953 static int
  954 nfs_read(struct vop_read_args *ap)
  955 {
  956         struct vnode *vp = ap->a_vp;
  957 
  958         switch (vp->v_type) {
  959         case VREG:
  960                 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
  961         case VDIR:
  962                 return (EISDIR);
  963         default:
  964                 return (EOPNOTSUPP);
  965         }
  966 }
  967 
  968 /*
  969  * nfs readlink call
  970  */
  971 static int
  972 nfs_readlink(struct vop_readlink_args *ap)
  973 {
  974         struct vnode *vp = ap->a_vp;
  975 
  976         if (vp->v_type != VLNK)
  977                 return (EINVAL);
  978         return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred));
  979 }
  980 
  981 /*
  982  * Do a readlink rpc.
  983  * Called by nfs_doio() from below the buffer cache.
  984  */
  985 int
  986 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
  987 {
  988         caddr_t bpos, dpos;
  989         int error = 0, len, attrflag;
  990         struct mbuf *mreq, *mrep, *md, *mb;
  991         int v3 = NFS_ISV3(vp);
  992 
  993         nfsstats.rpccnt[NFSPROC_READLINK]++;
  994         mreq = nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
  995         mb = mreq;
  996         bpos = mtod(mb, caddr_t);
  997         nfsm_fhtom(vp, v3);
  998         nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, cred);
  999         if (v3)
 1000                 nfsm_postop_attr(vp, attrflag);
 1001         if (!error) {
 1002                 nfsm_strsiz(len, NFS_MAXPATHLEN);
 1003                 if (len == NFS_MAXPATHLEN) {
 1004                         struct nfsnode *np = VTONFS(vp);
 1005                         if (np->n_size && np->n_size < NFS_MAXPATHLEN)
 1006                                 len = np->n_size;
 1007                 }
 1008                 nfsm_mtouio(uiop, len);
 1009         }
 1010         m_freem(mrep);
 1011 nfsmout:
 1012         return (error);
 1013 }
 1014 
 1015 /*
 1016  * nfs read rpc call
 1017  * Ditto above
 1018  */
 1019 int
 1020 nfs_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
 1021 {
 1022         u_int32_t *tl;
 1023         caddr_t bpos, dpos;
 1024         struct mbuf *mreq, *mrep, *md, *mb;
 1025         struct nfsmount *nmp;
 1026         int error = 0, len, retlen, tsiz, eof, attrflag;
 1027         int v3 = NFS_ISV3(vp);
 1028 
 1029 #ifndef nolint
 1030         eof = 0;
 1031 #endif
 1032         nmp = VFSTONFS(vp->v_mount);
 1033         tsiz = uiop->uio_resid;
 1034         if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
 1035                 return (EFBIG);
 1036         while (tsiz > 0) {
 1037                 nfsstats.rpccnt[NFSPROC_READ]++;
 1038                 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
 1039                 mreq = nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
 1040                 mb = mreq;
 1041                 bpos = mtod(mb, caddr_t);
 1042                 nfsm_fhtom(vp, v3);
 1043                 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED * 3);
 1044                 if (v3) {
 1045                         txdr_hyper(uiop->uio_offset, tl);
 1046                         *(tl + 2) = txdr_unsigned(len);
 1047                 } else {
 1048                         *tl++ = txdr_unsigned(uiop->uio_offset);
 1049                         *tl++ = txdr_unsigned(len);
 1050                         *tl = 0;
 1051                 }
 1052                 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, cred);
 1053                 if (v3) {
 1054                         nfsm_postop_attr(vp, attrflag);
 1055                         if (error) {
 1056                                 m_freem(mrep);
 1057                                 goto nfsmout;
 1058                         }
 1059                         tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED);
 1060                         eof = fxdr_unsigned(int, *(tl + 1));
 1061                 } else
 1062                         nfsm_loadattr(vp, NULL);
 1063                 nfsm_strsiz(retlen, nmp->nm_rsize);
 1064                 nfsm_mtouio(uiop, retlen);
 1065                 m_freem(mrep);
 1066                 tsiz -= retlen;
 1067                 if (v3) {
 1068                         if (eof || retlen == 0) {
 1069                                 tsiz = 0;
 1070                         }
 1071                 } else if (retlen < len) {
 1072                         tsiz = 0;
 1073                 }
 1074         }
 1075 nfsmout:
 1076         return (error);
 1077 }
 1078 
 1079 /*
 1080  * nfs write call
 1081  */
 1082 int
 1083 nfs_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
 1084     int *iomode, int *must_commit)
 1085 {
 1086         u_int32_t *tl;
 1087         int32_t backup;
 1088         caddr_t bpos, dpos;
 1089         struct mbuf *mreq, *mrep, *md, *mb;
 1090         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 1091         int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
 1092         int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
 1093 
 1094 #ifndef DIAGNOSTIC
 1095         if (uiop->uio_iovcnt != 1)
 1096                 panic("nfs: writerpc iovcnt > 1");
 1097 #endif
 1098         *must_commit = 0;
 1099         tsiz = uiop->uio_resid;
 1100         if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
 1101                 return (EFBIG);
 1102         while (tsiz > 0) {
 1103                 nfsstats.rpccnt[NFSPROC_WRITE]++;
 1104                 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
 1105                 mreq = nfsm_reqhead(vp, NFSPROC_WRITE,
 1106                         NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
 1107                 mb = mreq;
 1108                 bpos = mtod(mb, caddr_t);
 1109                 nfsm_fhtom(vp, v3);
 1110                 if (v3) {
 1111                         tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
 1112                         txdr_hyper(uiop->uio_offset, tl);
 1113                         tl += 2;
 1114                         *tl++ = txdr_unsigned(len);
 1115                         *tl++ = txdr_unsigned(*iomode);
 1116                         *tl = txdr_unsigned(len);
 1117                 } else {
 1118                         u_int32_t x;
 1119 
 1120                         tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
 1121                         /* Set both "begin" and "current" to non-garbage. */
 1122                         x = txdr_unsigned((u_int32_t)uiop->uio_offset);
 1123                         *tl++ = x;      /* "begin offset" */
 1124                         *tl++ = x;      /* "current offset" */
 1125                         x = txdr_unsigned(len);
 1126                         *tl++ = x;      /* total to this offset */
 1127                         *tl = x;        /* size of this write */
 1128                 }
 1129                 nfsm_uiotom(uiop, len);
 1130                 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, cred);
 1131                 if (v3) {
 1132                         wccflag = NFSV3_WCCCHK;
 1133                         nfsm_wcc_data(vp, wccflag);
 1134                         if (!error) {
 1135                                 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED
 1136                                         + NFSX_V3WRITEVERF);
 1137                                 rlen = fxdr_unsigned(int, *tl++);
 1138                                 if (rlen == 0) {
 1139                                         error = NFSERR_IO;
 1140                                         m_freem(mrep);
 1141                                         break;
 1142                                 } else if (rlen < len) {
 1143                                         backup = len - rlen;
 1144                                         uiop->uio_iov->iov_base =
 1145                                             (char *)uiop->uio_iov->iov_base -
 1146                                             backup;
 1147                                         uiop->uio_iov->iov_len += backup;
 1148                                         uiop->uio_offset -= backup;
 1149                                         uiop->uio_resid += backup;
 1150                                         len = rlen;
 1151                                 }
 1152                                 commit = fxdr_unsigned(int, *tl++);
 1153 
 1154                                 /*
 1155                                  * Return the lowest committment level
 1156                                  * obtained by any of the RPCs.
 1157                                  */
 1158                                 if (committed == NFSV3WRITE_FILESYNC)
 1159                                         committed = commit;
 1160                                 else if (committed == NFSV3WRITE_DATASYNC &&
 1161                                         commit == NFSV3WRITE_UNSTABLE)
 1162                                         committed = commit;
 1163                                 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
 1164                                     bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
 1165                                         NFSX_V3WRITEVERF);
 1166                                     nmp->nm_state |= NFSSTA_HASWRITEVERF;
 1167                                 } else if (bcmp((caddr_t)tl,
 1168                                     (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
 1169                                     *must_commit = 1;
 1170                                     bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
 1171                                         NFSX_V3WRITEVERF);
 1172                                 }
 1173                         }
 1174                 } else
 1175                     nfsm_loadattr(vp, NULL);
 1176                 if (wccflag)
 1177                     VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime.tv_sec;
 1178                 m_freem(mrep);
 1179                 if (error)
 1180                         break;
 1181                 tsiz -= len;
 1182         }
 1183 nfsmout:
 1184         if (vp->v_mount->mnt_flag & MNT_ASYNC)
 1185                 committed = NFSV3WRITE_FILESYNC;
 1186         *iomode = committed;
 1187         if (error)
 1188                 uiop->uio_resid = tsiz;
 1189         return (error);
 1190 }
 1191 
 1192 /*
 1193  * nfs mknod rpc
 1194  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
 1195  * mode set to specify the file type and the size field for rdev.
 1196  */
 1197 static int
 1198 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
 1199     struct vattr *vap)
 1200 {
 1201         struct nfsv2_sattr *sp;
 1202         u_int32_t *tl;
 1203         struct vnode *newvp = NULL;
 1204         struct nfsnode *np = NULL;
 1205         struct vattr vattr;
 1206         caddr_t bpos, dpos;
 1207         int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
 1208         struct mbuf *mreq, *mrep, *md, *mb;
 1209         u_int32_t rdev;
 1210         int v3 = NFS_ISV3(dvp);
 1211 
 1212         if (vap->va_type == VCHR || vap->va_type == VBLK)
 1213                 rdev = txdr_unsigned(vap->va_rdev);
 1214         else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
 1215                 rdev = nfs_xdrneg1;
 1216         else {
 1217                 return (EOPNOTSUPP);
 1218         }
 1219         if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
 1220                 return (error);
 1221         }
 1222         nfsstats.rpccnt[NFSPROC_MKNOD]++;
 1223         mreq = nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
 1224                 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
 1225         mb = mreq;
 1226         bpos = mtod(mb, caddr_t);
 1227         nfsm_fhtom(dvp, v3);
 1228         nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
 1229         if (v3) {
 1230                 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
 1231                 *tl++ = vtonfsv3_type(vap->va_type);
 1232                 nfsm_v3attrbuild(vap, FALSE);
 1233                 if (vap->va_type == VCHR || vap->va_type == VBLK) {
 1234                         tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
 1235                         *tl++ = txdr_unsigned(umajor(vap->va_rdev));
 1236                         *tl = txdr_unsigned(uminor(vap->va_rdev));
 1237                 }
 1238         } else {
 1239                 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
 1240                 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
 1241                 sp->sa_uid = nfs_xdrneg1;
 1242                 sp->sa_gid = nfs_xdrneg1;
 1243                 sp->sa_size = rdev;
 1244                 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
 1245                 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
 1246         }
 1247         nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_thread, cnp->cn_cred);
 1248         if (!error) {
 1249                 nfsm_mtofh(dvp, newvp, v3, gotvp);
 1250                 if (!gotvp) {
 1251                         if (newvp) {
 1252                                 vput(newvp);
 1253                                 newvp = NULL;
 1254                         }
 1255                         error = nfs_lookitup(dvp, cnp->cn_nameptr,
 1256                             cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
 1257                         if (!error)
 1258                                 newvp = NFSTOV(np);
 1259                 }
 1260         }
 1261         if (v3)
 1262                 nfsm_wcc_data(dvp, wccflag);
 1263         m_freem(mrep);
 1264 nfsmout:
 1265         if (error) {
 1266                 if (newvp)
 1267                         vput(newvp);
 1268         } else {
 1269                 if (cnp->cn_flags & MAKEENTRY)
 1270                         cache_enter(dvp, newvp, cnp);
 1271                 *vpp = newvp;
 1272         }
 1273         VTONFS(dvp)->n_flag |= NMODIFIED;
 1274         if (!wccflag)
 1275                 VTONFS(dvp)->n_attrstamp = 0;
 1276         return (error);
 1277 }
 1278 
 1279 /*
 1280  * nfs mknod vop
 1281  * just call nfs_mknodrpc() to do the work.
 1282  */
 1283 /* ARGSUSED */
 1284 static int
 1285 nfs_mknod(struct vop_mknod_args *ap)
 1286 {
 1287 
 1288         return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
 1289 }
 1290 
 1291 static u_long create_verf;
 1292 /*
 1293  * nfs file create call
 1294  */
 1295 static int
 1296 nfs_create(struct vop_create_args *ap)
 1297 {
 1298         struct vnode *dvp = ap->a_dvp;
 1299         struct vattr *vap = ap->a_vap;
 1300         struct componentname *cnp = ap->a_cnp;
 1301         struct nfsv2_sattr *sp;
 1302         u_int32_t *tl;
 1303         struct nfsnode *np = NULL;
 1304         struct vnode *newvp = NULL;
 1305         caddr_t bpos, dpos;
 1306         int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
 1307         struct mbuf *mreq, *mrep, *md, *mb;
 1308         struct vattr vattr;
 1309         int v3 = NFS_ISV3(dvp);
 1310 
 1311         /*
 1312          * Oops, not for me..
 1313          */
 1314         if (vap->va_type == VSOCK)
 1315                 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
 1316 
 1317         if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
 1318                 return (error);
 1319         }
 1320         if (vap->va_vaflags & VA_EXCLUSIVE)
 1321                 fmode |= O_EXCL;
 1322 again:
 1323         nfsstats.rpccnt[NFSPROC_CREATE]++;
 1324         mreq = nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
 1325                 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
 1326         mb = mreq;
 1327         bpos = mtod(mb, caddr_t);
 1328         nfsm_fhtom(dvp, v3);
 1329         nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
 1330         if (v3) {
 1331                 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
 1332                 if (fmode & O_EXCL) {
 1333                         *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
 1334                         tl = nfsm_build(u_int32_t *, NFSX_V3CREATEVERF);
 1335 #ifdef INET
 1336                         if (!TAILQ_EMPTY(&in_ifaddrhead))
 1337                                 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrhead))->sin_addr.s_addr;
 1338                         else
 1339 #endif
 1340                                 *tl++ = create_verf;
 1341                         *tl = ++create_verf;
 1342                 } else {
 1343                         *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
 1344                         nfsm_v3attrbuild(vap, FALSE);
 1345                 }
 1346         } else {
 1347                 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
 1348                 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
 1349                 sp->sa_uid = nfs_xdrneg1;
 1350                 sp->sa_gid = nfs_xdrneg1;
 1351                 sp->sa_size = 0;
 1352                 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
 1353                 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
 1354         }
 1355         nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_thread, cnp->cn_cred);
 1356         if (!error) {
 1357                 nfsm_mtofh(dvp, newvp, v3, gotvp);
 1358                 if (!gotvp) {
 1359                         if (newvp) {
 1360                                 vput(newvp);
 1361                                 newvp = NULL;
 1362                         }
 1363                         error = nfs_lookitup(dvp, cnp->cn_nameptr,
 1364                             cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
 1365                         if (!error)
 1366                                 newvp = NFSTOV(np);
 1367                 }
 1368         }
 1369         if (v3)
 1370                 nfsm_wcc_data(dvp, wccflag);
 1371         m_freem(mrep);
 1372 nfsmout:
 1373         if (error) {
 1374                 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
 1375                         fmode &= ~O_EXCL;
 1376                         goto again;
 1377                 }
 1378                 if (newvp)
 1379                         vput(newvp);
 1380         } else if (v3 && (fmode & O_EXCL)) {
 1381                 /*
 1382                  * We are normally called with only a partially initialized
 1383                  * VAP.  Since the NFSv3 spec says that server may use the
 1384                  * file attributes to store the verifier, the spec requires
 1385                  * us to do a SETATTR RPC. FreeBSD servers store the verifier
 1386                  * in atime, but we can't really assume that all servers will
 1387                  * so we ensure that our SETATTR sets both atime and mtime.
 1388                  */
 1389                 if (vap->va_mtime.tv_sec == VNOVAL)
 1390                         vfs_timestamp(&vap->va_mtime);
 1391                 if (vap->va_atime.tv_sec == VNOVAL)
 1392                         vap->va_atime = vap->va_mtime;
 1393                 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_thread);
 1394         }
 1395         if (!error) {
 1396                 if (cnp->cn_flags & MAKEENTRY)
 1397                         cache_enter(dvp, newvp, cnp);
 1398                 *ap->a_vpp = newvp;
 1399         }
 1400         VTONFS(dvp)->n_flag |= NMODIFIED;
 1401         if (!wccflag)
 1402                 VTONFS(dvp)->n_attrstamp = 0;
 1403         return (error);
 1404 }
 1405 
 1406 /*
 1407  * nfs file remove call
 1408  * To try and make nfs semantics closer to ufs semantics, a file that has
 1409  * other processes using the vnode is renamed instead of removed and then
 1410  * removed later on the last close.
 1411  * - If v_usecount > 1
 1412  *        If a rename is not already in the works
 1413  *           call nfs_sillyrename() to set it up
 1414  *     else
 1415  *        do the remove rpc
 1416  */
 1417 static int
 1418 nfs_remove(struct vop_remove_args *ap)
 1419 {
 1420         struct vnode *vp = ap->a_vp;
 1421         struct vnode *dvp = ap->a_dvp;
 1422         struct componentname *cnp = ap->a_cnp;
 1423         struct nfsnode *np = VTONFS(vp);
 1424         int error = 0;
 1425         struct vattr vattr;
 1426 
 1427 #ifndef DIAGNOSTIC
 1428         if ((cnp->cn_flags & HASBUF) == 0)
 1429                 panic("nfs_remove: no name");
 1430         if (vrefcnt(vp) < 1)
 1431                 panic("nfs_remove: bad v_usecount");
 1432 #endif
 1433         if (vp->v_type == VDIR)
 1434                 error = EPERM;
 1435         else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
 1436             VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_thread) == 0 &&
 1437             vattr.va_nlink > 1)) {
 1438                 /*
 1439                  * Purge the name cache so that the chance of a lookup for
 1440                  * the name succeeding while the remove is in progress is
 1441                  * minimized. Without node locking it can still happen, such
 1442                  * that an I/O op returns ESTALE, but since you get this if
 1443                  * another host removes the file..
 1444                  */
 1445                 cache_purge(vp);
 1446                 /*
 1447                  * throw away biocache buffers, mainly to avoid
 1448                  * unnecessary delayed writes later.
 1449                  */
 1450                 error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, cnp->cn_thread, 1);
 1451                 /* Do the rpc */
 1452                 if (error != EINTR)
 1453                         error = nfs_removerpc(dvp, cnp->cn_nameptr,
 1454                                 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
 1455                 /*
 1456                  * Kludge City: If the first reply to the remove rpc is lost..
 1457                  *   the reply to the retransmitted request will be ENOENT
 1458                  *   since the file was in fact removed
 1459                  *   Therefore, we cheat and return success.
 1460                  */
 1461                 if (error == ENOENT)
 1462                         error = 0;
 1463         } else if (!np->n_sillyrename)
 1464                 error = nfs_sillyrename(dvp, vp, cnp);
 1465         np->n_attrstamp = 0;
 1466         return (error);
 1467 }
 1468 
 1469 /*
 1470  * nfs file remove rpc called from nfs_inactive
 1471  */
 1472 int
 1473 nfs_removeit(struct sillyrename *sp)
 1474 {
 1475 
 1476         /*
 1477          * Make sure that the directory vnode is still valid.
 1478          * XXX we should lock sp->s_dvp here.
 1479          */
 1480         if (sp->s_dvp->v_type == VBAD)
 1481                 return (0);
 1482         return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
 1483                 NULL));
 1484 }
 1485 
 1486 /*
 1487  * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
 1488  */
 1489 static int
 1490 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
 1491     struct ucred *cred, struct thread *td)
 1492 {
 1493         caddr_t bpos, dpos;
 1494         int error = 0, wccflag = NFSV3_WCCRATTR;
 1495         struct mbuf *mreq, *mrep, *md, *mb;
 1496         int v3 = NFS_ISV3(dvp);
 1497 
 1498         nfsstats.rpccnt[NFSPROC_REMOVE]++;
 1499         mreq = nfsm_reqhead(dvp, NFSPROC_REMOVE,
 1500                 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
 1501         mb = mreq;
 1502         bpos = mtod(mb, caddr_t);
 1503         nfsm_fhtom(dvp, v3);
 1504         nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
 1505         nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
 1506         if (v3)
 1507                 nfsm_wcc_data(dvp, wccflag);
 1508         m_freem(mrep);
 1509 nfsmout:
 1510         VTONFS(dvp)->n_flag |= NMODIFIED;
 1511         if (!wccflag)
 1512                 VTONFS(dvp)->n_attrstamp = 0;
 1513         return (error);
 1514 }
 1515 
 1516 /*
 1517  * nfs file rename call
 1518  */
 1519 static int
 1520 nfs_rename(struct vop_rename_args *ap)
 1521 {
 1522         struct vnode *fvp = ap->a_fvp;
 1523         struct vnode *tvp = ap->a_tvp;
 1524         struct vnode *fdvp = ap->a_fdvp;
 1525         struct vnode *tdvp = ap->a_tdvp;
 1526         struct componentname *tcnp = ap->a_tcnp;
 1527         struct componentname *fcnp = ap->a_fcnp;
 1528         int error;
 1529 
 1530 #ifndef DIAGNOSTIC
 1531         if ((tcnp->cn_flags & HASBUF) == 0 ||
 1532             (fcnp->cn_flags & HASBUF) == 0)
 1533                 panic("nfs_rename: no name");
 1534 #endif
 1535         /* Check for cross-device rename */
 1536         if ((fvp->v_mount != tdvp->v_mount) ||
 1537             (tvp && (fvp->v_mount != tvp->v_mount))) {
 1538                 error = EXDEV;
 1539                 goto out;
 1540         }
 1541 
 1542         if (fvp == tvp) {
 1543                 printf("nfs_rename: fvp == tvp (can't happen)\n");
 1544                 error = 0;
 1545                 goto out;
 1546         }
 1547         if ((error = vn_lock(fvp, LK_EXCLUSIVE, fcnp->cn_thread)) != 0)
 1548                 goto out;
 1549 
 1550         /*
 1551          * We have to flush B_DELWRI data prior to renaming
 1552          * the file.  If we don't, the delayed-write buffers
 1553          * can be flushed out later after the file has gone stale
 1554          * under NFSV3.  NFSV2 does not have this problem because
 1555          * ( as far as I can tell ) it flushes dirty buffers more
 1556          * often.
 1557          */
 1558         VOP_FSYNC(fvp, fcnp->cn_cred, MNT_WAIT, fcnp->cn_thread);
 1559         VOP_UNLOCK(fvp, 0, fcnp->cn_thread);
 1560         if (tvp)
 1561             VOP_FSYNC(tvp, tcnp->cn_cred, MNT_WAIT, tcnp->cn_thread);
 1562 
 1563         /*
 1564          * If the tvp exists and is in use, sillyrename it before doing the
 1565          * rename of the new file over it.
 1566          * XXX Can't sillyrename a directory.
 1567          */
 1568         if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
 1569                 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
 1570                 vput(tvp);
 1571                 tvp = NULL;
 1572         }
 1573 
 1574         error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
 1575                 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
 1576                 tcnp->cn_thread);
 1577 
 1578         if (fvp->v_type == VDIR) {
 1579                 if (tvp != NULL && tvp->v_type == VDIR)
 1580                         cache_purge(tdvp);
 1581                 cache_purge(fdvp);
 1582         }
 1583 
 1584 out:
 1585         if (tdvp == tvp)
 1586                 vrele(tdvp);
 1587         else
 1588                 vput(tdvp);
 1589         if (tvp)
 1590                 vput(tvp);
 1591         vrele(fdvp);
 1592         vrele(fvp);
 1593         /*
 1594          * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
 1595          */
 1596         if (error == ENOENT)
 1597                 error = 0;
 1598         return (error);
 1599 }
 1600 
 1601 /*
 1602  * nfs file rename rpc called from nfs_remove() above
 1603  */
 1604 static int
 1605 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
 1606     struct sillyrename *sp)
 1607 {
 1608 
 1609         return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, sdvp,
 1610             sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_thread));
 1611 }
 1612 
 1613 /*
 1614  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
 1615  */
 1616 static int
 1617 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
 1618     struct vnode *tdvp, const char *tnameptr, int tnamelen, struct ucred *cred,
 1619     struct thread *td)
 1620 {
 1621         caddr_t bpos, dpos;
 1622         int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
 1623         struct mbuf *mreq, *mrep, *md, *mb;
 1624         int v3 = NFS_ISV3(fdvp);
 1625 
 1626         nfsstats.rpccnt[NFSPROC_RENAME]++;
 1627         mreq = nfsm_reqhead(fdvp, NFSPROC_RENAME,
 1628                 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
 1629                 nfsm_rndup(tnamelen));
 1630         mb = mreq;
 1631         bpos = mtod(mb, caddr_t);
 1632         nfsm_fhtom(fdvp, v3);
 1633         nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
 1634         nfsm_fhtom(tdvp, v3);
 1635         nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
 1636         nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
 1637         if (v3) {
 1638                 nfsm_wcc_data(fdvp, fwccflag);
 1639                 nfsm_wcc_data(tdvp, twccflag);
 1640         }
 1641         m_freem(mrep);
 1642 nfsmout:
 1643         VTONFS(fdvp)->n_flag |= NMODIFIED;
 1644         VTONFS(tdvp)->n_flag |= NMODIFIED;
 1645         if (!fwccflag)
 1646                 VTONFS(fdvp)->n_attrstamp = 0;
 1647         if (!twccflag)
 1648                 VTONFS(tdvp)->n_attrstamp = 0;
 1649         return (error);
 1650 }
 1651 
 1652 /*
 1653  * nfs hard link create call
 1654  */
 1655 static int
 1656 nfs_link(struct vop_link_args *ap)
 1657 {
 1658         struct vnode *vp = ap->a_vp;
 1659         struct vnode *tdvp = ap->a_tdvp;
 1660         struct componentname *cnp = ap->a_cnp;
 1661         caddr_t bpos, dpos;
 1662         int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
 1663         struct mbuf *mreq, *mrep, *md, *mb;
 1664         int v3;
 1665 
 1666         if (vp->v_mount != tdvp->v_mount) {
 1667                 return (EXDEV);
 1668         }
 1669 
 1670         /*
 1671          * Push all writes to the server, so that the attribute cache
 1672          * doesn't get "out of sync" with the server.
 1673          * XXX There should be a better way!
 1674          */
 1675         VOP_FSYNC(vp, cnp->cn_cred, MNT_WAIT, cnp->cn_thread);
 1676 
 1677         v3 = NFS_ISV3(vp);
 1678         nfsstats.rpccnt[NFSPROC_LINK]++;
 1679         mreq = nfsm_reqhead(vp, NFSPROC_LINK,
 1680                 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
 1681         mb = mreq;
 1682         bpos = mtod(mb, caddr_t);
 1683         nfsm_fhtom(vp, v3);
 1684         nfsm_fhtom(tdvp, v3);
 1685         nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
 1686         nfsm_request(vp, NFSPROC_LINK, cnp->cn_thread, cnp->cn_cred);
 1687         if (v3) {
 1688                 nfsm_postop_attr(vp, attrflag);
 1689                 nfsm_wcc_data(tdvp, wccflag);
 1690         }
 1691         m_freem(mrep);
 1692 nfsmout:
 1693         VTONFS(tdvp)->n_flag |= NMODIFIED;
 1694         if (!attrflag)
 1695                 VTONFS(vp)->n_attrstamp = 0;
 1696         if (!wccflag)
 1697                 VTONFS(tdvp)->n_attrstamp = 0;
 1698         /*
 1699          * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
 1700          */
 1701         if (error == EEXIST)
 1702                 error = 0;
 1703         return (error);
 1704 }
 1705 
 1706 /*
 1707  * nfs symbolic link create call
 1708  */
 1709 static int
 1710 nfs_symlink(struct vop_symlink_args *ap)
 1711 {
 1712         struct vnode *dvp = ap->a_dvp;
 1713         struct vattr *vap = ap->a_vap;
 1714         struct componentname *cnp = ap->a_cnp;
 1715         struct nfsv2_sattr *sp;
 1716         caddr_t bpos, dpos;
 1717         int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
 1718         struct mbuf *mreq, *mrep, *md, *mb;
 1719         struct vnode *newvp = NULL;
 1720         int v3 = NFS_ISV3(dvp);
 1721 
 1722         nfsstats.rpccnt[NFSPROC_SYMLINK]++;
 1723         slen = strlen(ap->a_target);
 1724         mreq = nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
 1725             nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
 1726         mb = mreq;
 1727         bpos = mtod(mb, caddr_t);
 1728         nfsm_fhtom(dvp, v3);
 1729         nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
 1730         if (v3) {
 1731                 nfsm_v3attrbuild(vap, FALSE);
 1732         }
 1733         nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
 1734         if (!v3) {
 1735                 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
 1736                 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
 1737                 sp->sa_uid = nfs_xdrneg1;
 1738                 sp->sa_gid = nfs_xdrneg1;
 1739                 sp->sa_size = nfs_xdrneg1;
 1740                 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
 1741                 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
 1742         }
 1743 
 1744         /*
 1745          * Issue the NFS request and get the rpc response.
 1746          *
 1747          * Only NFSv3 responses returning an error of 0 actually return
 1748          * a file handle that can be converted into newvp without having
 1749          * to do an extra lookup rpc.
 1750          */
 1751         nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_thread, cnp->cn_cred);
 1752         if (v3) {
 1753                 if (error == 0)
 1754                         nfsm_mtofh(dvp, newvp, v3, gotvp);
 1755                 nfsm_wcc_data(dvp, wccflag);
 1756         }
 1757 
 1758         /*
 1759          * out code jumps -> here, mrep is also freed.
 1760          */
 1761 
 1762         m_freem(mrep);
 1763 nfsmout:
 1764 
 1765         /*
 1766          * If we get an EEXIST error, silently convert it to no-error
 1767          * in case of an NFS retry.
 1768          */
 1769         if (error == EEXIST)
 1770                 error = 0;
 1771 
 1772         /*
 1773          * If we do not have (or no longer have) an error, and we could
 1774          * not extract the newvp from the response due to the request being
 1775          * NFSv2 or the error being EEXIST.  We have to do a lookup in order
 1776          * to obtain a newvp to return.
 1777          */
 1778         if (error == 0 && newvp == NULL) {
 1779                 struct nfsnode *np = NULL;
 1780 
 1781                 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
 1782                     cnp->cn_cred, cnp->cn_thread, &np);
 1783                 if (!error)
 1784                         newvp = NFSTOV(np);
 1785         }
 1786         if (error) {
 1787                 if (newvp)
 1788                         vput(newvp);
 1789         } else {
 1790                 *ap->a_vpp = newvp;
 1791         }
 1792         VTONFS(dvp)->n_flag |= NMODIFIED;
 1793         if (!wccflag)
 1794                 VTONFS(dvp)->n_attrstamp = 0;
 1795         return (error);
 1796 }
 1797 
 1798 /*
 1799  * nfs make dir call
 1800  */
 1801 static int
 1802 nfs_mkdir(struct vop_mkdir_args *ap)
 1803 {
 1804         struct vnode *dvp = ap->a_dvp;
 1805         struct vattr *vap = ap->a_vap;
 1806         struct componentname *cnp = ap->a_cnp;
 1807         struct nfsv2_sattr *sp;
 1808         int len;
 1809         struct nfsnode *np = NULL;
 1810         struct vnode *newvp = NULL;
 1811         caddr_t bpos, dpos;
 1812         int error = 0, wccflag = NFSV3_WCCRATTR;
 1813         int gotvp = 0;
 1814         struct mbuf *mreq, *mrep, *md, *mb;
 1815         struct vattr vattr;
 1816         int v3 = NFS_ISV3(dvp);
 1817 
 1818         if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
 1819                 return (error);
 1820         }
 1821         len = cnp->cn_namelen;
 1822         nfsstats.rpccnt[NFSPROC_MKDIR]++;
 1823         mreq = nfsm_reqhead(dvp, NFSPROC_MKDIR,
 1824           NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
 1825         mb = mreq;
 1826         bpos = mtod(mb, caddr_t);
 1827         nfsm_fhtom(dvp, v3);
 1828         nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
 1829         if (v3) {
 1830                 nfsm_v3attrbuild(vap, FALSE);
 1831         } else {
 1832                 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
 1833                 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
 1834                 sp->sa_uid = nfs_xdrneg1;
 1835                 sp->sa_gid = nfs_xdrneg1;
 1836                 sp->sa_size = nfs_xdrneg1;
 1837                 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
 1838                 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
 1839         }
 1840         nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_thread, cnp->cn_cred);
 1841         if (!error)
 1842                 nfsm_mtofh(dvp, newvp, v3, gotvp);
 1843         if (v3)
 1844                 nfsm_wcc_data(dvp, wccflag);
 1845         m_freem(mrep);
 1846 nfsmout:
 1847         VTONFS(dvp)->n_flag |= NMODIFIED;
 1848         if (!wccflag)
 1849                 VTONFS(dvp)->n_attrstamp = 0;
 1850         /*
 1851          * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
 1852          * if we can succeed in looking up the directory.
 1853          */
 1854         if (error == EEXIST || (!error && !gotvp)) {
 1855                 if (newvp) {
 1856                         vrele(newvp);
 1857                         newvp = NULL;
 1858                 }
 1859                 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
 1860                         cnp->cn_thread, &np);
 1861                 if (!error) {
 1862                         newvp = NFSTOV(np);
 1863                         if (newvp->v_type != VDIR)
 1864                                 error = EEXIST;
 1865                 }
 1866         }
 1867         if (error) {
 1868                 if (newvp)
 1869                         vrele(newvp);
 1870         } else
 1871                 *ap->a_vpp = newvp;
 1872         return (error);
 1873 }
 1874 
 1875 /*
 1876  * nfs remove directory call
 1877  */
 1878 static int
 1879 nfs_rmdir(struct vop_rmdir_args *ap)
 1880 {
 1881         struct vnode *vp = ap->a_vp;
 1882         struct vnode *dvp = ap->a_dvp;
 1883         struct componentname *cnp = ap->a_cnp;
 1884         caddr_t bpos, dpos;
 1885         int error = 0, wccflag = NFSV3_WCCRATTR;
 1886         struct mbuf *mreq, *mrep, *md, *mb;
 1887         int v3 = NFS_ISV3(dvp);
 1888 
 1889         if (dvp == vp)
 1890                 return (EINVAL);
 1891         nfsstats.rpccnt[NFSPROC_RMDIR]++;
 1892         mreq = nfsm_reqhead(dvp, NFSPROC_RMDIR,
 1893                 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
 1894         mb = mreq;
 1895         bpos = mtod(mb, caddr_t);
 1896         nfsm_fhtom(dvp, v3);
 1897         nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
 1898         nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_thread, cnp->cn_cred);
 1899         if (v3)
 1900                 nfsm_wcc_data(dvp, wccflag);
 1901         m_freem(mrep);
 1902 nfsmout:
 1903         VTONFS(dvp)->n_flag |= NMODIFIED;
 1904         if (!wccflag)
 1905                 VTONFS(dvp)->n_attrstamp = 0;
 1906         cache_purge(dvp);
 1907         cache_purge(vp);
 1908         /*
 1909          * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
 1910          */
 1911         if (error == ENOENT)
 1912                 error = 0;
 1913         return (error);
 1914 }
 1915 
 1916 /*
 1917  * nfs readdir call
 1918  */
 1919 static int
 1920 nfs_readdir(struct vop_readdir_args *ap)
 1921 {
 1922         struct vnode *vp = ap->a_vp;
 1923         struct nfsnode *np = VTONFS(vp);
 1924         struct uio *uio = ap->a_uio;
 1925         int tresid, error;
 1926         struct vattr vattr;
 1927 
 1928         if (vp->v_type != VDIR)
 1929                 return (EPERM);
 1930         /*
 1931          * First, check for hit on the EOF offset cache
 1932          */
 1933         if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
 1934             (np->n_flag & NMODIFIED) == 0) {
 1935                 if (VOP_GETATTR(vp, &vattr, ap->a_cred, uio->uio_td) == 0 &&
 1936                         np->n_mtime == vattr.va_mtime.tv_sec) {
 1937                         nfsstats.direofcache_hits++;
 1938                         return (0);
 1939                 }
 1940         }
 1941 
 1942         /*
 1943          * Call nfs_bioread() to do the real work.
 1944          */
 1945         tresid = uio->uio_resid;
 1946         error = nfs_bioread(vp, uio, 0, ap->a_cred);
 1947 
 1948         if (!error && uio->uio_resid == tresid)
 1949                 nfsstats.direofcache_misses++;
 1950         return (error);
 1951 }
 1952 
 1953 /*
 1954  * Readdir rpc call.
 1955  * Called from below the buffer cache by nfs_doio().
 1956  */
 1957 int
 1958 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
 1959 {
 1960         int len, left;
 1961         struct dirent *dp = NULL;
 1962         u_int32_t *tl;
 1963         caddr_t cp;
 1964         nfsuint64 *cookiep;
 1965         caddr_t bpos, dpos;
 1966         struct mbuf *mreq, *mrep, *md, *mb;
 1967         nfsuint64 cookie;
 1968         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 1969         struct nfsnode *dnp = VTONFS(vp);
 1970         u_quad_t fileno;
 1971         int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
 1972         int attrflag;
 1973         int v3 = NFS_ISV3(vp);
 1974 
 1975 #ifndef DIAGNOSTIC
 1976         if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
 1977                 (uiop->uio_resid & (DIRBLKSIZ - 1)))
 1978                 panic("nfs readdirrpc bad uio");
 1979 #endif
 1980 
 1981         /*
 1982          * If there is no cookie, assume directory was stale.
 1983          */
 1984         cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
 1985         if (cookiep)
 1986                 cookie = *cookiep;
 1987         else
 1988                 return (NFSERR_BAD_COOKIE);
 1989         /*
 1990          * Loop around doing readdir rpc's of size nm_readdirsize
 1991          * truncated to a multiple of DIRBLKSIZ.
 1992          * The stopping criteria is EOF or buffer full.
 1993          */
 1994         while (more_dirs && bigenough) {
 1995                 nfsstats.rpccnt[NFSPROC_READDIR]++;
 1996                 mreq = nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
 1997                         NFSX_READDIR(v3));
 1998                 mb = mreq;
 1999                 bpos = mtod(mb, caddr_t);
 2000                 nfsm_fhtom(vp, v3);
 2001                 if (v3) {
 2002                         tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
 2003                         *tl++ = cookie.nfsuquad[0];
 2004                         *tl++ = cookie.nfsuquad[1];
 2005                         *tl++ = dnp->n_cookieverf.nfsuquad[0];
 2006                         *tl++ = dnp->n_cookieverf.nfsuquad[1];
 2007                 } else {
 2008                         tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
 2009                         *tl++ = cookie.nfsuquad[0];
 2010                 }
 2011                 *tl = txdr_unsigned(nmp->nm_readdirsize);
 2012                 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, cred);
 2013                 if (v3) {
 2014                         nfsm_postop_attr(vp, attrflag);
 2015                         if (!error) {
 2016                                 tl = nfsm_dissect(u_int32_t *,
 2017                                     2 * NFSX_UNSIGNED);
 2018                                 dnp->n_cookieverf.nfsuquad[0] = *tl++;
 2019                                 dnp->n_cookieverf.nfsuquad[1] = *tl;
 2020                         } else {
 2021                                 m_freem(mrep);
 2022                                 goto nfsmout;
 2023                         }
 2024                 }
 2025                 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2026                 more_dirs = fxdr_unsigned(int, *tl);
 2027 
 2028                 /* loop thru the dir entries, doctoring them to 4bsd form */
 2029                 while (more_dirs && bigenough) {
 2030                         if (v3) {
 2031                                 tl = nfsm_dissect(u_int32_t *,
 2032                                     3 * NFSX_UNSIGNED);
 2033                                 fileno = fxdr_hyper(tl);
 2034                                 len = fxdr_unsigned(int, *(tl + 2));
 2035                         } else {
 2036                                 tl = nfsm_dissect(u_int32_t *,
 2037                                     2 * NFSX_UNSIGNED);
 2038                                 fileno = fxdr_unsigned(u_quad_t, *tl++);
 2039                                 len = fxdr_unsigned(int, *tl);
 2040                         }
 2041                         if (len <= 0 || len > NFS_MAXNAMLEN) {
 2042                                 error = EBADRPC;
 2043                                 m_freem(mrep);
 2044                                 goto nfsmout;
 2045                         }
 2046                         tlen = nfsm_rndup(len);
 2047                         if (tlen == len)
 2048                                 tlen += 4;      /* To ensure null termination */
 2049                         left = DIRBLKSIZ - blksiz;
 2050                         if ((tlen + DIRHDSIZ) > left) {
 2051                                 dp->d_reclen += left;
 2052                                 uiop->uio_iov->iov_base =
 2053                                     (char *)uiop->uio_iov->iov_base + left;
 2054                                 uiop->uio_iov->iov_len -= left;
 2055                                 uiop->uio_offset += left;
 2056                                 uiop->uio_resid -= left;
 2057                                 blksiz = 0;
 2058                         }
 2059                         if ((tlen + DIRHDSIZ) > uiop->uio_resid)
 2060                                 bigenough = 0;
 2061                         if (bigenough) {
 2062                                 dp = (struct dirent *)uiop->uio_iov->iov_base;
 2063                                 dp->d_fileno = (int)fileno;
 2064                                 dp->d_namlen = len;
 2065                                 dp->d_reclen = tlen + DIRHDSIZ;
 2066                                 dp->d_type = DT_UNKNOWN;
 2067                                 blksiz += dp->d_reclen;
 2068                                 if (blksiz == DIRBLKSIZ)
 2069                                         blksiz = 0;
 2070                                 uiop->uio_offset += DIRHDSIZ;
 2071                                 uiop->uio_resid -= DIRHDSIZ;
 2072                                 uiop->uio_iov->iov_base =
 2073                                     (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
 2074                                 uiop->uio_iov->iov_len -= DIRHDSIZ;
 2075                                 nfsm_mtouio(uiop, len);
 2076                                 cp = uiop->uio_iov->iov_base;
 2077                                 tlen -= len;
 2078                                 *cp = '\0';     /* null terminate */
 2079                                 uiop->uio_iov->iov_base =
 2080                                     (char *)uiop->uio_iov->iov_base + tlen;
 2081                                 uiop->uio_iov->iov_len -= tlen;
 2082                                 uiop->uio_offset += tlen;
 2083                                 uiop->uio_resid -= tlen;
 2084                         } else
 2085                                 nfsm_adv(nfsm_rndup(len));
 2086                         if (v3) {
 2087                                 tl = nfsm_dissect(u_int32_t *,
 2088                                     3 * NFSX_UNSIGNED);
 2089                         } else {
 2090                                 tl = nfsm_dissect(u_int32_t *,
 2091                                     2 * NFSX_UNSIGNED);
 2092                         }
 2093                         if (bigenough) {
 2094                                 cookie.nfsuquad[0] = *tl++;
 2095                                 if (v3)
 2096                                         cookie.nfsuquad[1] = *tl++;
 2097                         } else if (v3)
 2098                                 tl += 2;
 2099                         else
 2100                                 tl++;
 2101                         more_dirs = fxdr_unsigned(int, *tl);
 2102                 }
 2103                 /*
 2104                  * If at end of rpc data, get the eof boolean
 2105                  */
 2106                 if (!more_dirs) {
 2107                         tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2108                         more_dirs = (fxdr_unsigned(int, *tl) == 0);
 2109                 }
 2110                 m_freem(mrep);
 2111         }
 2112         /*
 2113          * Fill last record, iff any, out to a multiple of DIRBLKSIZ
 2114          * by increasing d_reclen for the last record.
 2115          */
 2116         if (blksiz > 0) {
 2117                 left = DIRBLKSIZ - blksiz;
 2118                 dp->d_reclen += left;
 2119                 uiop->uio_iov->iov_base =
 2120                     (char *)uiop->uio_iov->iov_base + left;
 2121                 uiop->uio_iov->iov_len -= left;
 2122                 uiop->uio_offset += left;
 2123                 uiop->uio_resid -= left;
 2124         }
 2125 
 2126         /*
 2127          * We are now either at the end of the directory or have filled the
 2128          * block.
 2129          */
 2130         if (bigenough)
 2131                 dnp->n_direofoffset = uiop->uio_offset;
 2132         else {
 2133                 if (uiop->uio_resid > 0)
 2134                         printf("EEK! readdirrpc resid > 0\n");
 2135                 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
 2136                 *cookiep = cookie;
 2137         }
 2138 nfsmout:
 2139         return (error);
 2140 }
 2141 
 2142 /*
 2143  * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
 2144  */
 2145 int
 2146 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
 2147 {
 2148         int len, left;
 2149         struct dirent *dp;
 2150         u_int32_t *tl;
 2151         caddr_t cp;
 2152         struct vnode *newvp;
 2153         nfsuint64 *cookiep;
 2154         caddr_t bpos, dpos, dpossav1, dpossav2;
 2155         struct mbuf *mreq, *mrep, *md, *mb, *mdsav1, *mdsav2;
 2156         struct nameidata nami, *ndp = &nami;
 2157         struct componentname *cnp = &ndp->ni_cnd;
 2158         nfsuint64 cookie;
 2159         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 2160         struct nfsnode *dnp = VTONFS(vp), *np;
 2161         nfsfh_t *fhp;
 2162         u_quad_t fileno;
 2163         int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
 2164         int attrflag, fhsize;
 2165 
 2166 #ifndef nolint
 2167         dp = NULL;
 2168 #endif
 2169 #ifndef DIAGNOSTIC
 2170         if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
 2171                 (uiop->uio_resid & (DIRBLKSIZ - 1)))
 2172                 panic("nfs readdirplusrpc bad uio");
 2173 #endif
 2174         ndp->ni_dvp = vp;
 2175         newvp = NULLVP;
 2176 
 2177         /*
 2178          * If there is no cookie, assume directory was stale.
 2179          */
 2180         cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
 2181         if (cookiep)
 2182                 cookie = *cookiep;
 2183         else
 2184                 return (NFSERR_BAD_COOKIE);
 2185         /*
 2186          * Loop around doing readdir rpc's of size nm_readdirsize
 2187          * truncated to a multiple of DIRBLKSIZ.
 2188          * The stopping criteria is EOF or buffer full.
 2189          */
 2190         while (more_dirs && bigenough) {
 2191                 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
 2192                 mreq = nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
 2193                         NFSX_FH(1) + 6 * NFSX_UNSIGNED);
 2194                 mb = mreq;
 2195                 bpos = mtod(mb, caddr_t);
 2196                 nfsm_fhtom(vp, 1);
 2197                 tl = nfsm_build(u_int32_t *, 6 * NFSX_UNSIGNED);
 2198                 *tl++ = cookie.nfsuquad[0];
 2199                 *tl++ = cookie.nfsuquad[1];
 2200                 *tl++ = dnp->n_cookieverf.nfsuquad[0];
 2201                 *tl++ = dnp->n_cookieverf.nfsuquad[1];
 2202                 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
 2203                 *tl = txdr_unsigned(nmp->nm_rsize);
 2204                 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, cred);
 2205                 nfsm_postop_attr(vp, attrflag);
 2206                 if (error) {
 2207                         m_freem(mrep);
 2208                         goto nfsmout;
 2209                 }
 2210                 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
 2211                 dnp->n_cookieverf.nfsuquad[0] = *tl++;
 2212                 dnp->n_cookieverf.nfsuquad[1] = *tl++;
 2213                 more_dirs = fxdr_unsigned(int, *tl);
 2214 
 2215                 /* loop thru the dir entries, doctoring them to 4bsd form */
 2216                 while (more_dirs && bigenough) {
 2217                         tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
 2218                         fileno = fxdr_hyper(tl);
 2219                         len = fxdr_unsigned(int, *(tl + 2));
 2220                         if (len <= 0 || len > NFS_MAXNAMLEN) {
 2221                                 error = EBADRPC;
 2222                                 m_freem(mrep);
 2223                                 goto nfsmout;
 2224                         }
 2225                         tlen = nfsm_rndup(len);
 2226                         if (tlen == len)
 2227                                 tlen += 4;      /* To ensure null termination*/
 2228                         left = DIRBLKSIZ - blksiz;
 2229                         if ((tlen + DIRHDSIZ) > left) {
 2230                                 dp->d_reclen += left;
 2231                                 uiop->uio_iov->iov_base =
 2232                                     (char *)uiop->uio_iov->iov_base + left;
 2233                                 uiop->uio_iov->iov_len -= left;
 2234                                 uiop->uio_offset += left;
 2235                                 uiop->uio_resid -= left;
 2236                                 blksiz = 0;
 2237                         }
 2238                         if ((tlen + DIRHDSIZ) > uiop->uio_resid)
 2239                                 bigenough = 0;
 2240                         if (bigenough) {
 2241                                 dp = (struct dirent *)uiop->uio_iov->iov_base;
 2242                                 dp->d_fileno = (int)fileno;
 2243                                 dp->d_namlen = len;
 2244                                 dp->d_reclen = tlen + DIRHDSIZ;
 2245                                 dp->d_type = DT_UNKNOWN;
 2246                                 blksiz += dp->d_reclen;
 2247                                 if (blksiz == DIRBLKSIZ)
 2248                                         blksiz = 0;
 2249                                 uiop->uio_offset += DIRHDSIZ;
 2250                                 uiop->uio_resid -= DIRHDSIZ;
 2251                                 uiop->uio_iov->iov_base =
 2252                                     (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
 2253                                 uiop->uio_iov->iov_len -= DIRHDSIZ;
 2254                                 cnp->cn_nameptr = uiop->uio_iov->iov_base;
 2255                                 cnp->cn_namelen = len;
 2256                                 nfsm_mtouio(uiop, len);
 2257                                 cp = uiop->uio_iov->iov_base;
 2258                                 tlen -= len;
 2259                                 *cp = '\0';
 2260                                 uiop->uio_iov->iov_base =
 2261                                     (char *)uiop->uio_iov->iov_base + tlen;
 2262                                 uiop->uio_iov->iov_len -= tlen;
 2263                                 uiop->uio_offset += tlen;
 2264                                 uiop->uio_resid -= tlen;
 2265                         } else
 2266                                 nfsm_adv(nfsm_rndup(len));
 2267                         tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
 2268                         if (bigenough) {
 2269                                 cookie.nfsuquad[0] = *tl++;
 2270                                 cookie.nfsuquad[1] = *tl++;
 2271                         } else
 2272                                 tl += 2;
 2273 
 2274                         /*
 2275                          * Since the attributes are before the file handle
 2276                          * (sigh), we must skip over the attributes and then
 2277                          * come back and get them.
 2278                          */
 2279                         attrflag = fxdr_unsigned(int, *tl);
 2280                         if (attrflag) {
 2281                             dpossav1 = dpos;
 2282                             mdsav1 = md;
 2283                             nfsm_adv(NFSX_V3FATTR);
 2284                             tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2285                             doit = fxdr_unsigned(int, *tl);
 2286                             if (doit) {
 2287                                 nfsm_getfh(fhp, fhsize, 1);
 2288                                 if (NFS_CMPFH(dnp, fhp, fhsize)) {
 2289                                     VREF(vp);
 2290                                     newvp = vp;
 2291                                     np = dnp;
 2292                                 } else {
 2293                                     error = nfs_nget(vp->v_mount, fhp,
 2294                                         fhsize, &np);
 2295                                     if (error)
 2296                                         doit = 0;
 2297                                     else
 2298                                         newvp = NFSTOV(np);
 2299                                 }
 2300                             }
 2301                             if (doit && bigenough) {
 2302                                 dpossav2 = dpos;
 2303                                 dpos = dpossav1;
 2304                                 mdsav2 = md;
 2305                                 md = mdsav1;
 2306                                 nfsm_loadattr(newvp, NULL);
 2307                                 dpos = dpossav2;
 2308                                 md = mdsav2;
 2309                                 dp->d_type =
 2310                                     IFTODT(VTTOIF(np->n_vattr.va_type));
 2311                                 ndp->ni_vp = newvp;
 2312                                 cache_enter(ndp->ni_dvp, ndp->ni_vp, cnp);
 2313                             }
 2314                         } else {
 2315                             /* Just skip over the file handle */
 2316                             tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2317                             i = fxdr_unsigned(int, *tl);
 2318                             nfsm_adv(nfsm_rndup(i));
 2319                         }
 2320                         if (newvp != NULLVP) {
 2321                             if (newvp == vp)
 2322                                 vrele(newvp);
 2323                             else
 2324                                 vput(newvp);
 2325                             newvp = NULLVP;
 2326                         }
 2327                         tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2328                         more_dirs = fxdr_unsigned(int, *tl);
 2329                 }
 2330                 /*
 2331                  * If at end of rpc data, get the eof boolean
 2332                  */
 2333                 if (!more_dirs) {
 2334                         tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2335                         more_dirs = (fxdr_unsigned(int, *tl) == 0);
 2336                 }
 2337                 m_freem(mrep);
 2338         }
 2339         /*
 2340          * Fill last record, iff any, out to a multiple of DIRBLKSIZ
 2341          * by increasing d_reclen for the last record.
 2342          */
 2343         if (blksiz > 0) {
 2344                 left = DIRBLKSIZ - blksiz;
 2345                 dp->d_reclen += left;
 2346                 uiop->uio_iov->iov_base =
 2347                     (char *)uiop->uio_iov->iov_base + left;
 2348                 uiop->uio_iov->iov_len -= left;
 2349                 uiop->uio_offset += left;
 2350                 uiop->uio_resid -= left;
 2351         }
 2352 
 2353         /*
 2354          * We are now either at the end of the directory or have filled the
 2355          * block.
 2356          */
 2357         if (bigenough)
 2358                 dnp->n_direofoffset = uiop->uio_offset;
 2359         else {
 2360                 if (uiop->uio_resid > 0)
 2361                         printf("EEK! readdirplusrpc resid > 0\n");
 2362                 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
 2363                 *cookiep = cookie;
 2364         }
 2365 nfsmout:
 2366         if (newvp != NULLVP) {
 2367                 if (newvp == vp)
 2368                         vrele(newvp);
 2369                 else
 2370                         vput(newvp);
 2371                 newvp = NULLVP;
 2372         }
 2373         return (error);
 2374 }
 2375 
 2376 /*
 2377  * Silly rename. To make the NFS filesystem that is stateless look a little
 2378  * more like the "ufs" a remove of an active vnode is translated to a rename
 2379  * to a funny looking filename that is removed by nfs_inactive on the
 2380  * nfsnode. There is the potential for another process on a different client
 2381  * to create the same funny name between the nfs_lookitup() fails and the
 2382  * nfs_rename() completes, but...
 2383  */
 2384 static int
 2385 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
 2386 {
 2387         struct sillyrename *sp;
 2388         struct nfsnode *np;
 2389         int error;
 2390         short pid;
 2391 
 2392         cache_purge(dvp);
 2393         np = VTONFS(vp);
 2394 #ifndef DIAGNOSTIC
 2395         if (vp->v_type == VDIR)
 2396                 panic("nfs: sillyrename dir");
 2397 #endif
 2398         MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
 2399                 M_NFSREQ, M_WAITOK);
 2400         sp->s_cred = crhold(cnp->cn_cred);
 2401         sp->s_dvp = dvp;
 2402         sp->s_removeit = nfs_removeit;
 2403         VREF(dvp);
 2404 
 2405         /* Fudge together a funny name */
 2406         pid = cnp->cn_thread->td_proc->p_pid;
 2407         sp->s_namlen = sprintf(sp->s_name, ".nfsA%04x4.4", pid);
 2408 
 2409         /* Try lookitups until we get one that isn't there */
 2410         while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
 2411                 cnp->cn_thread, NULL) == 0) {
 2412                 sp->s_name[4]++;
 2413                 if (sp->s_name[4] > 'z') {
 2414                         error = EINVAL;
 2415                         goto bad;
 2416                 }
 2417         }
 2418         error = nfs_renameit(dvp, cnp, sp);
 2419         if (error)
 2420                 goto bad;
 2421         error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
 2422                 cnp->cn_thread, &np);
 2423         np->n_sillyrename = sp;
 2424         return (0);
 2425 bad:
 2426         vrele(sp->s_dvp);
 2427         crfree(sp->s_cred);
 2428         free((caddr_t)sp, M_NFSREQ);
 2429         return (error);
 2430 }
 2431 
 2432 /*
 2433  * Look up a file name and optionally either update the file handle or
 2434  * allocate an nfsnode, depending on the value of npp.
 2435  * npp == NULL  --> just do the lookup
 2436  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
 2437  *                      handled too
 2438  * *npp != NULL --> update the file handle in the vnode
 2439  */
 2440 static int
 2441 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
 2442     struct thread *td, struct nfsnode **npp)
 2443 {
 2444         struct vnode *newvp = NULL;
 2445         struct nfsnode *np, *dnp = VTONFS(dvp);
 2446         caddr_t bpos, dpos;
 2447         int error = 0, fhlen, attrflag;
 2448         struct mbuf *mreq, *mrep, *md, *mb;
 2449         nfsfh_t *nfhp;
 2450         int v3 = NFS_ISV3(dvp);
 2451 
 2452         nfsstats.rpccnt[NFSPROC_LOOKUP]++;
 2453         mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
 2454                 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
 2455         mb = mreq;
 2456         bpos = mtod(mb, caddr_t);
 2457         nfsm_fhtom(dvp, v3);
 2458         nfsm_strtom(name, len, NFS_MAXNAMLEN);
 2459         nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
 2460         if (npp && !error) {
 2461                 nfsm_getfh(nfhp, fhlen, v3);
 2462                 if (*npp) {
 2463                     np = *npp;
 2464                     if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
 2465                         free((caddr_t)np->n_fhp, M_NFSBIGFH);
 2466                         np->n_fhp = &np->n_fh;
 2467                     } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
 2468                         np->n_fhp =(nfsfh_t *)malloc(fhlen, M_NFSBIGFH, M_WAITOK);
 2469                     bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
 2470                     np->n_fhsize = fhlen;
 2471                     newvp = NFSTOV(np);
 2472                 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
 2473                     VREF(dvp);
 2474                     newvp = dvp;
 2475                 } else {
 2476                     error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
 2477                     if (error) {
 2478                         m_freem(mrep);
 2479                         return (error);
 2480                     }
 2481                     newvp = NFSTOV(np);
 2482                 }
 2483                 if (v3) {
 2484                         nfsm_postop_attr(newvp, attrflag);
 2485                         if (!attrflag && *npp == NULL) {
 2486                                 m_freem(mrep);
 2487                                 if (newvp == dvp)
 2488                                         vrele(newvp);
 2489                                 else
 2490                                         vput(newvp);
 2491                                 return (ENOENT);
 2492                         }
 2493                 } else
 2494                         nfsm_loadattr(newvp, NULL);
 2495         }
 2496         m_freem(mrep);
 2497 nfsmout:
 2498         if (npp && *npp == NULL) {
 2499                 if (error) {
 2500                         if (newvp) {
 2501                                 if (newvp == dvp)
 2502                                         vrele(newvp);
 2503                                 else
 2504                                         vput(newvp);
 2505                         }
 2506                 } else
 2507                         *npp = np;
 2508         }
 2509         return (error);
 2510 }
 2511 
 2512 /*
 2513  * Nfs Version 3 commit rpc
 2514  */
 2515 int
 2516 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
 2517     struct thread *td)
 2518 {
 2519         u_int32_t *tl;
 2520         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 2521         caddr_t bpos, dpos;
 2522         int error = 0, wccflag = NFSV3_WCCRATTR;
 2523         struct mbuf *mreq, *mrep, *md, *mb;
 2524 
 2525         if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
 2526                 return (0);
 2527         nfsstats.rpccnt[NFSPROC_COMMIT]++;
 2528         mreq = nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
 2529         mb = mreq;
 2530         bpos = mtod(mb, caddr_t);
 2531         nfsm_fhtom(vp, 1);
 2532         tl = nfsm_build(u_int32_t *, 3 * NFSX_UNSIGNED);
 2533         txdr_hyper(offset, tl);
 2534         tl += 2;
 2535         *tl = txdr_unsigned(cnt);
 2536         nfsm_request(vp, NFSPROC_COMMIT, td, cred);
 2537         nfsm_wcc_data(vp, wccflag);
 2538         if (!error) {
 2539                 tl = nfsm_dissect(u_int32_t *, NFSX_V3WRITEVERF);
 2540                 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
 2541                         NFSX_V3WRITEVERF)) {
 2542                         bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
 2543                                 NFSX_V3WRITEVERF);
 2544                         error = NFSERR_STALEWRITEVERF;
 2545                 }
 2546         }
 2547         m_freem(mrep);
 2548 nfsmout:
 2549         return (error);
 2550 }
 2551 
 2552 /*
 2553  * Strategy routine.
 2554  * For async requests when nfsiod(s) are running, queue the request by
 2555  * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
 2556  * request.
 2557  */
 2558 static int
 2559 nfs_strategy(struct vop_strategy_args *ap)
 2560 {
 2561         struct buf *bp = ap->a_bp;
 2562         struct ucred *cr;
 2563         struct thread *td;
 2564         int error = 0;
 2565 
 2566         KASSERT(ap->a_vp == ap->a_bp->b_vp, ("%s(%p != %p)",
 2567             __func__, ap->a_vp, ap->a_bp->b_vp));
 2568         KASSERT(!(bp->b_flags & B_DONE), ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
 2569         KASSERT(BUF_REFCNT(bp) > 0, ("nfs_strategy: buffer %p not locked", bp));
 2570 
 2571         if (bp->b_flags & B_ASYNC)
 2572                 td = NULL;
 2573         else
 2574                 td = curthread; /* XXX */
 2575 
 2576         if (bp->b_iocmd == BIO_READ)
 2577                 cr = bp->b_rcred;
 2578         else
 2579                 cr = bp->b_wcred;
 2580 
 2581         /*
 2582          * If the op is asynchronous and an i/o daemon is waiting
 2583          * queue the request, wake it up and wait for completion
 2584          * otherwise just do it ourselves.
 2585          */
 2586         if ((bp->b_flags & B_ASYNC) == 0 ||
 2587                 nfs_asyncio(bp, NOCRED, td))
 2588                 error = nfs_doio(bp, cr, td);
 2589         return (error);
 2590 }
 2591 
 2592 /*
 2593  * fsync vnode op. Just call nfs_flush() with commit == 1.
 2594  */
 2595 /* ARGSUSED */
 2596 static int
 2597 nfs_fsync(struct vop_fsync_args *ap)
 2598 {
 2599 
 2600         return (nfs_flush(ap->a_vp, ap->a_cred, ap->a_waitfor, ap->a_td, 1));
 2601 }
 2602 
 2603 /*
 2604  * Flush all the blocks associated with a vnode.
 2605  *      Walk through the buffer pool and push any dirty pages
 2606  *      associated with the vnode.
 2607  */
 2608 static int
 2609 nfs_flush(struct vnode *vp, struct ucred *cred, int waitfor, struct thread *td,
 2610     int commit)
 2611 {
 2612         struct nfsnode *np = VTONFS(vp);
 2613         struct buf *bp;
 2614         int i;
 2615         struct buf *nbp;
 2616         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 2617         int s, error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
 2618         int passone = 1;
 2619         u_quad_t off, endoff, toff;
 2620         struct ucred* wcred = NULL;
 2621         struct buf **bvec = NULL;
 2622 #ifndef NFS_COMMITBVECSIZ
 2623 #define NFS_COMMITBVECSIZ       20
 2624 #endif
 2625         struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
 2626         int bvecsize = 0, bveccount;
 2627 
 2628         if (nmp->nm_flag & NFSMNT_INT)
 2629                 slpflag = PCATCH;
 2630         if (!commit)
 2631                 passone = 0;
 2632         /*
 2633          * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
 2634          * server, but nas not been committed to stable storage on the server
 2635          * yet. On the first pass, the byte range is worked out and the commit
 2636          * rpc is done. On the second pass, nfs_writebp() is called to do the
 2637          * job.
 2638          */
 2639 again:
 2640         off = (u_quad_t)-1;
 2641         endoff = 0;
 2642         bvecpos = 0;
 2643         if (NFS_ISV3(vp) && commit) {
 2644                 s = splbio();
 2645                 if (bvec != NULL && bvec != bvec_on_stack)
 2646                         free(bvec, M_TEMP);
 2647                 /*
 2648                  * Count up how many buffers waiting for a commit.
 2649                  */
 2650                 bveccount = 0;
 2651                 VI_LOCK(vp);
 2652                 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
 2653                         nbp = TAILQ_NEXT(bp, b_vnbufs);
 2654                         if (BUF_REFCNT(bp) == 0 &&
 2655                             (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
 2656                                 == (B_DELWRI | B_NEEDCOMMIT))
 2657                                 bveccount++;
 2658                 }
 2659                 /*
 2660                  * Allocate space to remember the list of bufs to commit.  It is
 2661                  * important to use M_NOWAIT here to avoid a race with nfs_write.
 2662                  * If we can't get memory (for whatever reason), we will end up
 2663                  * committing the buffers one-by-one in the loop below.
 2664                  */
 2665                 if (bveccount > NFS_COMMITBVECSIZ) {
 2666                         /*
 2667                          * Release the vnode interlock to avoid a lock
 2668                          * order reversal.
 2669                          */
 2670                         VI_UNLOCK(vp);
 2671                         bvec = (struct buf **)
 2672                                 malloc(bveccount * sizeof(struct buf *),
 2673                                        M_TEMP, M_NOWAIT);
 2674                         VI_LOCK(vp);
 2675                         if (bvec == NULL) {
 2676                                 bvec = bvec_on_stack;
 2677                                 bvecsize = NFS_COMMITBVECSIZ;
 2678                         } else
 2679                                 bvecsize = bveccount;
 2680                 } else {
 2681                         bvec = bvec_on_stack;
 2682                         bvecsize = NFS_COMMITBVECSIZ;
 2683                 }
 2684                 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
 2685                         if (bvecpos >= bvecsize)
 2686                                 break;
 2687                         if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
 2688                                 nbp = TAILQ_NEXT(bp, b_vnbufs);
 2689                                 continue;
 2690                         }
 2691                         if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
 2692                             (B_DELWRI | B_NEEDCOMMIT)) {
 2693                                 BUF_UNLOCK(bp);
 2694                                 nbp = TAILQ_NEXT(bp, b_vnbufs);
 2695                                 continue;
 2696                         }
 2697                         VI_UNLOCK(vp);
 2698                         bremfree(bp);
 2699                         /*
 2700                          * Work out if all buffers are using the same cred
 2701                          * so we can deal with them all with one commit.
 2702                          *
 2703                          * NOTE: we are not clearing B_DONE here, so we have
 2704                          * to do it later on in this routine if we intend to
 2705                          * initiate I/O on the bp.
 2706                          *
 2707                          * Note: to avoid loopback deadlocks, we do not
 2708                          * assign b_runningbufspace.
 2709                          */
 2710                         if (wcred == NULL)
 2711                                 wcred = bp->b_wcred;
 2712                         else if (wcred != bp->b_wcred)
 2713                                 wcred = NOCRED;
 2714                         bp->b_flags |= B_WRITEINPROG;
 2715                         vfs_busy_pages(bp, 1);
 2716 
 2717                         VI_LOCK(vp);
 2718                         /*
 2719                          * bp is protected by being locked, but nbp is not
 2720                          * and vfs_busy_pages() may sleep.  We have to
 2721                          * recalculate nbp.
 2722                          */
 2723                         nbp = TAILQ_NEXT(bp, b_vnbufs);
 2724 
 2725                         /*
 2726                          * A list of these buffers is kept so that the
 2727                          * second loop knows which buffers have actually
 2728                          * been committed. This is necessary, since there
 2729                          * may be a race between the commit rpc and new
 2730                          * uncommitted writes on the file.
 2731                          */
 2732                         bvec[bvecpos++] = bp;
 2733                         toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
 2734                                 bp->b_dirtyoff;
 2735                         if (toff < off)
 2736                                 off = toff;
 2737                         toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
 2738                         if (toff > endoff)
 2739                                 endoff = toff;
 2740                 }
 2741                 splx(s);
 2742                 VI_UNLOCK(vp);
 2743         }
 2744         if (bvecpos > 0) {
 2745                 /*
 2746                  * Commit data on the server, as required.
 2747                  * If all bufs are using the same wcred, then use that with
 2748                  * one call for all of them, otherwise commit each one
 2749                  * separately.
 2750                  */
 2751                 if (wcred != NOCRED)
 2752                         retv = nfs_commit(vp, off, (int)(endoff - off),
 2753                                           wcred, td);
 2754                 else {
 2755                         retv = 0;
 2756                         for (i = 0; i < bvecpos; i++) {
 2757                                 off_t off, size;
 2758                                 bp = bvec[i];
 2759                                 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
 2760                                         bp->b_dirtyoff;
 2761                                 size = (u_quad_t)(bp->b_dirtyend
 2762                                                   - bp->b_dirtyoff);
 2763                                 retv = nfs_commit(vp, off, (int)size,
 2764                                                   bp->b_wcred, td);
 2765                                 if (retv) break;
 2766                         }
 2767                 }
 2768 
 2769                 if (retv == NFSERR_STALEWRITEVERF)
 2770                         nfs_clearcommit(vp->v_mount);
 2771 
 2772                 /*
 2773                  * Now, either mark the blocks I/O done or mark the
 2774                  * blocks dirty, depending on whether the commit
 2775                  * succeeded.
 2776                  */
 2777                 for (i = 0; i < bvecpos; i++) {
 2778                         bp = bvec[i];
 2779                         bp->b_flags &= ~(B_NEEDCOMMIT | B_WRITEINPROG | B_CLUSTEROK);
 2780                         if (retv) {
 2781                                 /*
 2782                                  * Error, leave B_DELWRI intact
 2783                                  */
 2784                                 vfs_unbusy_pages(bp);
 2785                                 brelse(bp);
 2786                         } else {
 2787                                 /*
 2788                                  * Success, remove B_DELWRI ( bundirty() ).
 2789                                  *
 2790                                  * b_dirtyoff/b_dirtyend seem to be NFS
 2791                                  * specific.  We should probably move that
 2792                                  * into bundirty(). XXX
 2793                                  */
 2794                                 s = splbio();
 2795                                 VI_LOCK(vp);
 2796                                 vp->v_numoutput++;
 2797                                 VI_UNLOCK(vp);
 2798                                 bp->b_flags |= B_ASYNC;
 2799                                 bundirty(bp);
 2800                                 bp->b_flags &= ~B_DONE;
 2801                                 bp->b_ioflags &= ~BIO_ERROR;
 2802                                 bp->b_dirtyoff = bp->b_dirtyend = 0;
 2803                                 splx(s);
 2804                                 bufdone(bp);
 2805                         }
 2806                 }
 2807         }
 2808 
 2809         /*
 2810          * Start/do any write(s) that are required.
 2811          */
 2812 loop:
 2813         s = splbio();
 2814         VI_LOCK(vp);
 2815         for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
 2816                 nbp = TAILQ_NEXT(bp, b_vnbufs);
 2817                 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
 2818                         if (waitfor != MNT_WAIT || passone)
 2819                                 continue;
 2820 
 2821                         error = BUF_TIMELOCK(bp,
 2822                             LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
 2823                             VI_MTX(vp), "nfsfsync", slpflag, slptimeo);
 2824                         splx(s);
 2825                         if (error == 0)
 2826                                 panic("nfs_fsync: inconsistent lock");
 2827                         if (error == ENOLCK)
 2828                                 goto loop;
 2829                         if (nfs_sigintr(nmp, NULL, td)) {
 2830                                 error = EINTR;
 2831                                 goto done;
 2832                         }
 2833                         if (slpflag == PCATCH) {
 2834                                 slpflag = 0;
 2835                                 slptimeo = 2 * hz;
 2836                         }
 2837                         goto loop;
 2838                 }
 2839                 if ((bp->b_flags & B_DELWRI) == 0)
 2840                         panic("nfs_fsync: not dirty");
 2841                 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
 2842                         BUF_UNLOCK(bp);
 2843                         continue;
 2844                 }
 2845                 VI_UNLOCK(vp);
 2846                 bremfree(bp);
 2847                 if (passone || !commit)
 2848                     bp->b_flags |= B_ASYNC;
 2849                 else
 2850                     bp->b_flags |= B_ASYNC | B_WRITEINPROG;
 2851                 splx(s);
 2852                 BUF_WRITE(bp);
 2853                 goto loop;
 2854         }
 2855         splx(s);
 2856         if (passone) {
 2857                 passone = 0;
 2858                 VI_UNLOCK(vp);
 2859                 goto again;
 2860         }
 2861         if (waitfor == MNT_WAIT) {
 2862                 while (vp->v_numoutput) {
 2863                         vp->v_iflag |= VI_BWAIT;
 2864                         error = msleep((caddr_t)&vp->v_numoutput, VI_MTX(vp),
 2865                                 slpflag | (PRIBIO + 1), "nfsfsync", slptimeo);
 2866                         if (error) {
 2867                             VI_UNLOCK(vp);
 2868                             if (nfs_sigintr(nmp, NULL, td)) {
 2869                                 error = EINTR;
 2870                                 goto done;
 2871                             }
 2872                             if (slpflag == PCATCH) {
 2873                                 slpflag = 0;
 2874                                 slptimeo = 2 * hz;
 2875                             }
 2876                             VI_LOCK(vp);
 2877                         }
 2878                 }
 2879                 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) && commit) {
 2880                         VI_UNLOCK(vp);
 2881                         goto loop;
 2882                 }
 2883         }
 2884         VI_UNLOCK(vp);
 2885         if (np->n_flag & NWRITEERR) {
 2886                 error = np->n_error;
 2887                 np->n_flag &= ~NWRITEERR;
 2888         }
 2889 done:
 2890         if (bvec != NULL && bvec != bvec_on_stack)
 2891                 free(bvec, M_TEMP);
 2892         return (error);
 2893 }
 2894 
 2895 /*
 2896  * NFS advisory byte-level locks.
 2897  */
 2898 static int
 2899 nfs_advlock(struct vop_advlock_args *ap)
 2900 {
 2901 
 2902         if ((VFSTONFS(ap->a_vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
 2903                 struct nfsnode *np = VTONFS(ap->a_vp);
 2904 
 2905                 return (lf_advlock(ap, &(np->n_lockf), np->n_size));
 2906         }
 2907         return (nfs_dolock(ap));
 2908 }
 2909 
 2910 /*
 2911  * Print out the contents of an nfsnode.
 2912  */
 2913 static int
 2914 nfs_print(struct vop_print_args *ap)
 2915 {
 2916         struct vnode *vp = ap->a_vp;
 2917         struct nfsnode *np = VTONFS(vp);
 2918 
 2919         printf("\tfileid %ld fsid 0x%x",
 2920            np->n_vattr.va_fileid, np->n_vattr.va_fsid);
 2921         if (vp->v_type == VFIFO)
 2922                 fifo_printinfo(vp);
 2923         printf("\n");
 2924         return (0);
 2925 }
 2926 
 2927 /*
 2928  * This is the "real" nfs::bwrite(struct buf*).
 2929  * B_WRITEINPROG isn't set unless the force flag is one and it
 2930  * handles the B_NEEDCOMMIT flag.
 2931  * We set B_CACHE if this is a VMIO buffer.
 2932  */
 2933 int
 2934 nfs_writebp(struct buf *bp, int force, struct thread *td)
 2935 {
 2936         int s;
 2937         int oldflags = bp->b_flags;
 2938 #if 0
 2939         int retv = 1;
 2940         off_t off;
 2941 #endif
 2942 
 2943         if (BUF_REFCNT(bp) == 0)
 2944                 panic("bwrite: buffer is not locked???");
 2945 
 2946         if (bp->b_flags & B_INVAL) {
 2947                 brelse(bp);
 2948                 return(0);
 2949         }
 2950 
 2951         bp->b_flags |= B_CACHE;
 2952 
 2953         /*
 2954          * Undirty the bp.  We will redirty it later if the I/O fails.
 2955          */
 2956 
 2957         s = splbio();
 2958         bundirty(bp);
 2959         bp->b_flags &= ~B_DONE;
 2960         bp->b_ioflags &= ~BIO_ERROR;
 2961         bp->b_iocmd = BIO_WRITE;
 2962 
 2963         VI_LOCK(bp->b_vp);
 2964         bp->b_vp->v_numoutput++;
 2965         VI_UNLOCK(bp->b_vp);
 2966         curthread->td_proc->p_stats->p_ru.ru_oublock++;
 2967         splx(s);
 2968 
 2969         /*
 2970          * Note: to avoid loopback deadlocks, we do not
 2971          * assign b_runningbufspace.
 2972          */
 2973         vfs_busy_pages(bp, 1);
 2974 
 2975         if (force)
 2976                 bp->b_flags |= B_WRITEINPROG;
 2977         BUF_KERNPROC(bp);
 2978         bp->b_iooffset = dbtob(bp->b_blkno);
 2979         VOP_STRATEGY(bp->b_vp, bp);
 2980 
 2981         if( (oldflags & B_ASYNC) == 0) {
 2982                 int rtval = bufwait(bp);
 2983 
 2984                 if (oldflags & B_DELWRI) {
 2985                         s = splbio();
 2986                         reassignbuf(bp, bp->b_vp);
 2987                         splx(s);
 2988                 }
 2989 
 2990                 brelse(bp);
 2991                 return (rtval);
 2992         }
 2993 
 2994         return (0);
 2995 }
 2996 
 2997 /*
 2998  * nfs special file access vnode op.
 2999  * Essentially just get vattr and then imitate iaccess() since the device is
 3000  * local to the client.
 3001  */
 3002 static int
 3003 nfsspec_access(struct vop_access_args *ap)
 3004 {
 3005         struct vattr *vap;
 3006         struct ucred *cred = ap->a_cred;
 3007         struct vnode *vp = ap->a_vp;
 3008         mode_t mode = ap->a_mode;
 3009         struct vattr vattr;
 3010         int error;
 3011 
 3012         /*
 3013          * Disallow write attempts on filesystems mounted read-only;
 3014          * unless the file is a socket, fifo, or a block or character
 3015          * device resident on the filesystem.
 3016          */
 3017         if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
 3018                 switch (vp->v_type) {
 3019                 case VREG:
 3020                 case VDIR:
 3021                 case VLNK:
 3022                         return (EROFS);
 3023                 default:
 3024                         break;
 3025                 }
 3026         }
 3027         vap = &vattr;
 3028         error = VOP_GETATTR(vp, vap, cred, ap->a_td);
 3029         if (error)
 3030                 return (error);
 3031         return (vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
 3032             mode, cred, NULL));
 3033 }
 3034 
 3035 /*
 3036  * Read wrapper for special devices.
 3037  */
 3038 static int
 3039 nfsspec_read(struct vop_read_args *ap)
 3040 {
 3041         struct nfsnode *np = VTONFS(ap->a_vp);
 3042 
 3043         /*
 3044          * Set access flag.
 3045          */
 3046         np->n_flag |= NACC;
 3047         getnanotime(&np->n_atim);
 3048         return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap));
 3049 }
 3050 
 3051 /*
 3052  * Write wrapper for special devices.
 3053  */
 3054 static int
 3055 nfsspec_write(struct vop_write_args *ap)
 3056 {
 3057         struct nfsnode *np = VTONFS(ap->a_vp);
 3058 
 3059         /*
 3060          * Set update flag.
 3061          */
 3062         np->n_flag |= NUPD;
 3063         getnanotime(&np->n_mtim);
 3064         return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap));
 3065 }
 3066 
 3067 /*
 3068  * Close wrapper for special devices.
 3069  *
 3070  * Update the times on the nfsnode then do device close.
 3071  */
 3072 static int
 3073 nfsspec_close(struct vop_close_args *ap)
 3074 {
 3075         struct vnode *vp = ap->a_vp;
 3076         struct nfsnode *np = VTONFS(vp);
 3077         struct vattr vattr;
 3078 
 3079         if (np->n_flag & (NACC | NUPD)) {
 3080                 np->n_flag |= NCHG;
 3081                 if (vrefcnt(vp) == 1 &&
 3082                     (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
 3083                         VATTR_NULL(&vattr);
 3084                         if (np->n_flag & NACC)
 3085                                 vattr.va_atime = np->n_atim;
 3086                         if (np->n_flag & NUPD)
 3087                                 vattr.va_mtime = np->n_mtim;
 3088                         (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_td);
 3089                 }
 3090         }
 3091         return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap));
 3092 }
 3093 
 3094 /*
 3095  * Read wrapper for fifos.
 3096  */
 3097 static int
 3098 nfsfifo_read(struct vop_read_args *ap)
 3099 {
 3100         struct nfsnode *np = VTONFS(ap->a_vp);
 3101 
 3102         /*
 3103          * Set access flag.
 3104          */
 3105         np->n_flag |= NACC;
 3106         getnanotime(&np->n_atim);
 3107         return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap));
 3108 }
 3109 
 3110 /*
 3111  * Write wrapper for fifos.
 3112  */
 3113 static int
 3114 nfsfifo_write(struct vop_write_args *ap)
 3115 {
 3116         struct nfsnode *np = VTONFS(ap->a_vp);
 3117 
 3118         /*
 3119          * Set update flag.
 3120          */
 3121         np->n_flag |= NUPD;
 3122         getnanotime(&np->n_mtim);
 3123         return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap));
 3124 }
 3125 
 3126 /*
 3127  * Close wrapper for fifos.
 3128  *
 3129  * Update the times on the nfsnode then do fifo close.
 3130  */
 3131 static int
 3132 nfsfifo_close(struct vop_close_args *ap)
 3133 {
 3134         struct vnode *vp = ap->a_vp;
 3135         struct nfsnode *np = VTONFS(vp);
 3136         struct vattr vattr;
 3137         struct timespec ts;
 3138 
 3139         if (np->n_flag & (NACC | NUPD)) {
 3140                 getnanotime(&ts);
 3141                 if (np->n_flag & NACC)
 3142                         np->n_atim = ts;
 3143                 if (np->n_flag & NUPD)
 3144                         np->n_mtim = ts;
 3145                 np->n_flag |= NCHG;
 3146                 if (vrefcnt(vp) == 1 &&
 3147                     (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
 3148                         VATTR_NULL(&vattr);
 3149                         if (np->n_flag & NACC)
 3150                                 vattr.va_atime = np->n_atim;
 3151                         if (np->n_flag & NUPD)
 3152                                 vattr.va_mtime = np->n_mtim;
 3153                         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, ap->a_td);
 3154                         (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_td);
 3155                         VOP_UNLOCK(vp, 0, ap->a_td);
 3156                 }
 3157         }
 3158         return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap));
 3159 }
 3160 

Cache object: 17f5a2c12bde4bebe2a71e84a985fcd2


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.