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/5.0/sys/nfsclient/nfs_vnops.c 104908 2002-10-11 14:58:34Z mike $");
   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 <nfs/rpcv2.h>
   73 #include <nfs/nfsproto.h>
   74 #include <nfsclient/nfs.h>
   75 #include <nfsclient/nfsnode.h>
   76 #include <nfsclient/nfsmount.h>
   77 #include <nfsclient/nfs_lock.h>
   78 #include <nfs/xdr_subs.h>
   79 #include <nfsclient/nfsm_subs.h>
   80 
   81 #include <net/if.h>
   82 #include <netinet/in.h>
   83 #include <netinet/in_var.h>
   84 
   85 /* Defs */
   86 #define TRUE    1
   87 #define FALSE   0
   88 
   89 /*
   90  * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
   91  * calls are not in getblk() and brelse() so that they would not be necessary
   92  * here.
   93  */
   94 #ifndef B_VMIO
   95 #define vfs_busy_pages(bp, f)
   96 #endif
   97 
   98 static int      nfsspec_read(struct vop_read_args *);
   99 static int      nfsspec_write(struct vop_write_args *);
  100 static int      nfsfifo_read(struct vop_read_args *);
  101 static int      nfsfifo_write(struct vop_write_args *);
  102 static int      nfsspec_close(struct vop_close_args *);
  103 static int      nfsfifo_close(struct vop_close_args *);
  104 static int      nfs_flush(struct vnode *, struct ucred *, int, struct thread *,
  105                     int);
  106 static int      nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
  107                     struct thread *);
  108 static  int     nfs_lookup(struct vop_lookup_args *);
  109 static  int     nfs_create(struct vop_create_args *);
  110 static  int     nfs_mknod(struct vop_mknod_args *);
  111 static  int     nfs_open(struct vop_open_args *);
  112 static  int     nfs_close(struct vop_close_args *);
  113 static  int     nfs_access(struct vop_access_args *);
  114 static  int     nfs_getattr(struct vop_getattr_args *);
  115 static  int     nfs_setattr(struct vop_setattr_args *);
  116 static  int     nfs_read(struct vop_read_args *);
  117 static  int     nfs_fsync(struct vop_fsync_args *);
  118 static  int     nfs_remove(struct vop_remove_args *);
  119 static  int     nfs_link(struct vop_link_args *);
  120 static  int     nfs_rename(struct vop_rename_args *);
  121 static  int     nfs_mkdir(struct vop_mkdir_args *);
  122 static  int     nfs_rmdir(struct vop_rmdir_args *);
  123 static  int     nfs_symlink(struct vop_symlink_args *);
  124 static  int     nfs_readdir(struct vop_readdir_args *);
  125 static  int     nfs_strategy(struct vop_strategy_args *);
  126 static  int     nfs_lookitup(struct vnode *, const char *, int,
  127                     struct ucred *, struct thread *, struct nfsnode **);
  128 static  int     nfs_sillyrename(struct vnode *, struct vnode *,
  129                     struct componentname *);
  130 static int      nfsspec_access(struct vop_access_args *);
  131 static int      nfs_readlink(struct vop_readlink_args *);
  132 static int      nfs_print(struct vop_print_args *);
  133 static int      nfs_advlock(struct vop_advlock_args *);
  134 
  135 /*
  136  * Global vfs data structures for nfs
  137  */
  138 vop_t **nfsv2_vnodeop_p;
  139 static struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = {
  140         { &vop_default_desc,            (vop_t *) vop_defaultop },
  141         { &vop_access_desc,             (vop_t *) nfs_access },
  142         { &vop_advlock_desc,            (vop_t *) nfs_advlock },
  143         { &vop_close_desc,              (vop_t *) nfs_close },
  144         { &vop_create_desc,             (vop_t *) nfs_create },
  145         { &vop_fsync_desc,              (vop_t *) nfs_fsync },
  146         { &vop_getattr_desc,            (vop_t *) nfs_getattr },
  147         { &vop_getpages_desc,           (vop_t *) nfs_getpages },
  148         { &vop_putpages_desc,           (vop_t *) nfs_putpages },
  149         { &vop_inactive_desc,           (vop_t *) nfs_inactive },
  150         { &vop_islocked_desc,           (vop_t *) vop_stdislocked },
  151         { &vop_lease_desc,              (vop_t *) vop_null },
  152         { &vop_link_desc,               (vop_t *) nfs_link },
  153         { &vop_lock_desc,               (vop_t *) vop_sharedlock },
  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_unlock_desc,             (vop_t *) vop_stdunlock },
  170         { &vop_write_desc,              (vop_t *) nfs_write },
  171         { NULL, NULL }
  172 };
  173 static struct vnodeopv_desc nfsv2_vnodeop_opv_desc =
  174         { &nfsv2_vnodeop_p, nfsv2_vnodeop_entries };
  175 VNODEOP_SET(nfsv2_vnodeop_opv_desc);
  176 
  177 /*
  178  * Special device vnode ops
  179  */
  180 vop_t **spec_nfsv2nodeop_p;
  181 static struct vnodeopv_entry_desc nfsv2_specop_entries[] = {
  182         { &vop_default_desc,            (vop_t *) spec_vnoperate },
  183         { &vop_access_desc,             (vop_t *) nfsspec_access },
  184         { &vop_close_desc,              (vop_t *) nfsspec_close },
  185         { &vop_fsync_desc,              (vop_t *) nfs_fsync },
  186         { &vop_getattr_desc,            (vop_t *) nfs_getattr },
  187         { &vop_islocked_desc,           (vop_t *) vop_stdislocked },
  188         { &vop_inactive_desc,           (vop_t *) nfs_inactive },
  189         { &vop_lock_desc,               (vop_t *) vop_sharedlock },
  190         { &vop_print_desc,              (vop_t *) nfs_print },
  191         { &vop_read_desc,               (vop_t *) nfsspec_read },
  192         { &vop_reclaim_desc,            (vop_t *) nfs_reclaim },
  193         { &vop_setattr_desc,            (vop_t *) nfs_setattr },
  194         { &vop_unlock_desc,             (vop_t *) vop_stdunlock },
  195         { &vop_write_desc,              (vop_t *) nfsspec_write },
  196         { NULL, NULL }
  197 };
  198 static struct vnodeopv_desc spec_nfsv2nodeop_opv_desc =
  199         { &spec_nfsv2nodeop_p, nfsv2_specop_entries };
  200 VNODEOP_SET(spec_nfsv2nodeop_opv_desc);
  201 
  202 vop_t **fifo_nfsv2nodeop_p;
  203 static struct vnodeopv_entry_desc nfsv2_fifoop_entries[] = {
  204         { &vop_default_desc,            (vop_t *) fifo_vnoperate },
  205         { &vop_access_desc,             (vop_t *) nfsspec_access },
  206         { &vop_close_desc,              (vop_t *) nfsfifo_close },
  207         { &vop_fsync_desc,              (vop_t *) nfs_fsync },
  208         { &vop_getattr_desc,            (vop_t *) nfs_getattr },
  209         { &vop_inactive_desc,           (vop_t *) nfs_inactive },
  210         { &vop_islocked_desc,           (vop_t *) vop_stdislocked },
  211         { &vop_lock_desc,               (vop_t *) vop_sharedlock },
  212         { &vop_print_desc,              (vop_t *) nfs_print },
  213         { &vop_read_desc,               (vop_t *) nfsfifo_read },
  214         { &vop_reclaim_desc,            (vop_t *) nfs_reclaim },
  215         { &vop_setattr_desc,            (vop_t *) nfs_setattr },
  216         { &vop_unlock_desc,             (vop_t *) vop_stdunlock },
  217         { &vop_write_desc,              (vop_t *) nfsfifo_write },
  218         { NULL, NULL }
  219 };
  220 static struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc =
  221         { &fifo_nfsv2nodeop_p, nfsv2_fifoop_entries };
  222 VNODEOP_SET(fifo_nfsv2nodeop_opv_desc);
  223 
  224 static int      nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
  225                              struct componentname *cnp, struct vattr *vap);
  226 static int      nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
  227                               struct ucred *cred, struct thread *td);
  228 static int      nfs_renamerpc(struct vnode *fdvp, const char *fnameptr,
  229                               int fnamelen, struct vnode *tdvp,
  230                               const char *tnameptr, int tnamelen,
  231                               struct ucred *cred, struct thread *td);
  232 static int      nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
  233                              struct sillyrename *sp);
  234 
  235 /*
  236  * Global variables
  237  */
  238 struct proc     *nfs_iodwant[NFS_MAXASYNCDAEMON];
  239 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
  240 int              nfs_numasync = 0;
  241 #define DIRHDSIZ        (sizeof (struct dirent) - (MAXNAMLEN + 1))
  242 
  243 SYSCTL_DECL(_vfs_nfs);
  244 
  245 static int      nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
  246 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
  247            &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
  248 
  249 static int      nfsv3_commit_on_close = 0;
  250 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
  251            &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
  252 #if 0
  253 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
  254            &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
  255 
  256 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
  257            &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
  258 #endif
  259 
  260 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY          \
  261                          | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE     \
  262                          | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
  263 static int
  264 nfs3_access_otw(struct vnode *vp, int wmode, struct thread *td,
  265     struct ucred *cred)
  266 {
  267         const int v3 = 1;
  268         u_int32_t *tl;
  269         int error = 0, attrflag;
  270 
  271         struct mbuf *mreq, *mrep, *md, *mb;
  272         caddr_t bpos, dpos;
  273         u_int32_t rmode;
  274         struct nfsnode *np = VTONFS(vp);
  275 
  276         nfsstats.rpccnt[NFSPROC_ACCESS]++;
  277         mreq = nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
  278         mb = mreq;
  279         bpos = mtod(mb, caddr_t);
  280         nfsm_fhtom(vp, v3);
  281         tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
  282         *tl = txdr_unsigned(wmode);
  283         nfsm_request(vp, NFSPROC_ACCESS, td, cred);
  284         nfsm_postop_attr(vp, attrflag);
  285         if (!error) {
  286                 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
  287                 rmode = fxdr_unsigned(u_int32_t, *tl);
  288                 np->n_mode = rmode;
  289                 np->n_modeuid = cred->cr_uid;
  290                 np->n_modestamp = time_second;
  291         }
  292         m_freem(mrep);
  293 nfsmout:
  294         return error;
  295 }
  296 
  297 /*
  298  * nfs access vnode op.
  299  * For nfs version 2, just return ok. File accesses may fail later.
  300  * For nfs version 3, use the access rpc to check accessibility. If file modes
  301  * are changed on the server, accesses might still fail later.
  302  */
  303 static int
  304 nfs_access(struct vop_access_args *ap)
  305 {
  306         struct vnode *vp = ap->a_vp;
  307         int error = 0;
  308         u_int32_t mode, wmode;
  309         int v3 = NFS_ISV3(vp);
  310         struct nfsnode *np = VTONFS(vp);
  311 
  312         /*
  313          * Disallow write attempts on filesystems mounted read-only;
  314          * unless the file is a socket, fifo, or a block or character
  315          * device resident on the filesystem.
  316          */
  317         if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
  318                 switch (vp->v_type) {
  319                 case VREG:
  320                 case VDIR:
  321                 case VLNK:
  322                         return (EROFS);
  323                 default:
  324                         break;
  325                 }
  326         }
  327         /*
  328          * For nfs v3, check to see if we have done this recently, and if
  329          * so return our cached result instead of making an ACCESS call.
  330          * If not, do an access rpc, otherwise you are stuck emulating
  331          * ufs_access() locally using the vattr. This may not be correct,
  332          * since the server may apply other access criteria such as
  333          * client uid-->server uid mapping that we do not know about.
  334          */
  335         if (v3) {
  336                 if (ap->a_mode & VREAD)
  337                         mode = NFSV3ACCESS_READ;
  338                 else
  339                         mode = 0;
  340                 if (vp->v_type != VDIR) {
  341                         if (ap->a_mode & VWRITE)
  342                                 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
  343                         if (ap->a_mode & VEXEC)
  344                                 mode |= NFSV3ACCESS_EXECUTE;
  345                 } else {
  346                         if (ap->a_mode & VWRITE)
  347                                 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
  348                                          NFSV3ACCESS_DELETE);
  349                         if (ap->a_mode & VEXEC)
  350                                 mode |= NFSV3ACCESS_LOOKUP;
  351                 }
  352                 /* XXX safety belt, only make blanket request if caching */
  353                 if (nfsaccess_cache_timeout > 0) {
  354                         wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
  355                                 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
  356                                 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
  357                 } else {
  358                         wmode = mode;
  359                 }
  360 
  361                 /*
  362                  * Does our cached result allow us to give a definite yes to
  363                  * this request?
  364                  */
  365                 if ((time_second < (np->n_modestamp + nfsaccess_cache_timeout)) &&
  366                     (ap->a_cred->cr_uid == np->n_modeuid) &&
  367                     ((np->n_mode & mode) == mode)) {
  368                         nfsstats.accesscache_hits++;
  369                 } else {
  370                         /*
  371                          * Either a no, or a don't know.  Go to the wire.
  372                          */
  373                         nfsstats.accesscache_misses++;
  374                         error = nfs3_access_otw(vp, wmode, ap->a_td,ap->a_cred);
  375                         if (!error) {
  376                                 if ((np->n_mode & mode) != mode) {
  377                                         error = EACCES;
  378                                 }
  379                         }
  380                 }
  381                 return (error);
  382         } else {
  383                 if ((error = nfsspec_access(ap)) != 0)
  384                         return (error);
  385 
  386                 /*
  387                  * Attempt to prevent a mapped root from accessing a file
  388                  * which it shouldn't.  We try to read a byte from the file
  389                  * if the user is root and the file is not zero length.
  390                  * After calling nfsspec_access, we should have the correct
  391                  * file size cached.
  392                  */
  393                 if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD)
  394                     && VTONFS(vp)->n_size > 0) {
  395                         struct iovec aiov;
  396                         struct uio auio;
  397                         char buf[1];
  398 
  399                         aiov.iov_base = buf;
  400                         aiov.iov_len = 1;
  401                         auio.uio_iov = &aiov;
  402                         auio.uio_iovcnt = 1;
  403                         auio.uio_offset = 0;
  404                         auio.uio_resid = 1;
  405                         auio.uio_segflg = UIO_SYSSPACE;
  406                         auio.uio_rw = UIO_READ;
  407                         auio.uio_td = ap->a_td;
  408 
  409                         if (vp->v_type == VREG)
  410                                 error = nfs_readrpc(vp, &auio, ap->a_cred);
  411                         else if (vp->v_type == VDIR) {
  412                                 char* bp;
  413                                 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
  414                                 aiov.iov_base = bp;
  415                                 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
  416                                 error = nfs_readdirrpc(vp, &auio, ap->a_cred);
  417                                 free(bp, M_TEMP);
  418                         } else if (vp->v_type == VLNK)
  419                                 error = nfs_readlinkrpc(vp, &auio, ap->a_cred);
  420                         else
  421                                 error = EACCES;
  422                 }
  423                 return (error);
  424         }
  425 }
  426 
  427 /*
  428  * nfs open vnode op
  429  * Check to see if the type is ok
  430  * and that deletion is not in progress.
  431  * For paged in text files, you will need to flush the page cache
  432  * if consistency is lost.
  433  */
  434 /* ARGSUSED */
  435 static int
  436 nfs_open(struct vop_open_args *ap)
  437 {
  438         struct vnode *vp = ap->a_vp;
  439         struct nfsnode *np = VTONFS(vp);
  440         struct vattr vattr;
  441         int error;
  442 
  443         if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
  444 #ifdef DIAGNOSTIC
  445                 printf("open eacces vtyp=%d\n", vp->v_type);
  446 #endif
  447                 return (EACCES);
  448         }
  449         /*
  450          * Get a valid lease. If cached data is stale, flush it.
  451          */
  452         if (np->n_flag & NMODIFIED) {
  453                 if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
  454                         ap->a_td, 1)) == EINTR)
  455                         return (error);
  456                 np->n_attrstamp = 0;
  457                 if (vp->v_type == VDIR)
  458                         np->n_direofoffset = 0;
  459                 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
  460                 if (error)
  461                         return (error);
  462                 np->n_mtime = vattr.va_mtime.tv_sec;
  463         } else {
  464                 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
  465                 if (error)
  466                         return (error);
  467                 if (np->n_mtime != vattr.va_mtime.tv_sec) {
  468                         if (vp->v_type == VDIR)
  469                                 np->n_direofoffset = 0;
  470                         if ((error = nfs_vinvalbuf(vp, V_SAVE,
  471                                 ap->a_cred, ap->a_td, 1)) == EINTR)
  472                                 return (error);
  473                         np->n_mtime = vattr.va_mtime.tv_sec;
  474                 }
  475         }
  476         np->n_attrstamp = 0; /* For Open/Close consistency */
  477         return (0);
  478 }
  479 
  480 /*
  481  * nfs close vnode op
  482  * What an NFS client should do upon close after writing is a debatable issue.
  483  * Most NFS clients push delayed writes to the server upon close, basically for
  484  * two reasons:
  485  * 1 - So that any write errors may be reported back to the client process
  486  *     doing the close system call. By far the two most likely errors are
  487  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
  488  * 2 - To put a worst case upper bound on cache inconsistency between
  489  *     multiple clients for the file.
  490  * There is also a consistency problem for Version 2 of the protocol w.r.t.
  491  * not being able to tell if other clients are writing a file concurrently,
  492  * since there is no way of knowing if the changed modify time in the reply
  493  * is only due to the write for this client.
  494  * (NFS Version 3 provides weak cache consistency data in the reply that
  495  *  should be sufficient to detect and handle this case.)
  496  *
  497  * The current code does the following:
  498  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
  499  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
  500  *                     or commit them (this satisfies 1 and 2 except for the
  501  *                     case where the server crashes after this close but
  502  *                     before the commit RPC, which is felt to be "good
  503  *                     enough". Changing the last argument to nfs_flush() to
  504  *                     a 1 would force a commit operation, if it is felt a
  505  *                     commit is necessary now.
  506  */
  507 /* ARGSUSED */
  508 static int
  509 nfs_close(struct vop_close_args *ap)
  510 {
  511         struct vnode *vp = ap->a_vp;
  512         struct nfsnode *np = VTONFS(vp);
  513         int error = 0;
  514 
  515         if (vp->v_type == VREG) {
  516             if (np->n_flag & NMODIFIED) {
  517                 if (NFS_ISV3(vp)) {
  518                     /*
  519                      * Under NFSv3 we have dirty buffers to dispose of.  We
  520                      * must flush them to the NFS server.  We have the option
  521                      * of waiting all the way through the commit rpc or just
  522                      * waiting for the initial write.  The default is to only
  523                      * wait through the initial write so the data is in the
  524                      * server's cache, which is roughly similar to the state
  525                      * a standard disk subsystem leaves the file in on close().
  526                      *
  527                      * We cannot clear the NMODIFIED bit in np->n_flag due to
  528                      * potential races with other processes, and certainly
  529                      * cannot clear it if we don't commit.
  530                      */
  531                     int cm = nfsv3_commit_on_close ? 1 : 0;
  532                     error = nfs_flush(vp, ap->a_cred, MNT_WAIT, ap->a_td, cm);
  533                     /* np->n_flag &= ~NMODIFIED; */
  534                 } else {
  535                     error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_td, 1);
  536                 }
  537                 np->n_attrstamp = 0;
  538             }
  539             if (np->n_flag & NWRITEERR) {
  540                 np->n_flag &= ~NWRITEERR;
  541                 error = np->n_error;
  542             }
  543         }
  544         return (error);
  545 }
  546 
  547 /*
  548  * nfs getattr call from vfs.
  549  */
  550 static int
  551 nfs_getattr(struct vop_getattr_args *ap)
  552 {
  553         struct vnode *vp = ap->a_vp;
  554         struct nfsnode *np = VTONFS(vp);
  555         caddr_t bpos, dpos;
  556         int error = 0;
  557         struct mbuf *mreq, *mrep, *md, *mb;
  558         int v3 = NFS_ISV3(vp);
  559 
  560         /*
  561          * Update local times for special files.
  562          */
  563         if (np->n_flag & (NACC | NUPD))
  564                 np->n_flag |= NCHG;
  565         /*
  566          * First look in the cache.
  567          */
  568         if (nfs_getattrcache(vp, ap->a_vap) == 0)
  569                 return (0);
  570 
  571         if (v3 && nfsaccess_cache_timeout > 0) {
  572                 nfsstats.accesscache_misses++;
  573                 nfs3_access_otw(vp, NFSV3ACCESS_ALL, ap->a_td, ap->a_cred);
  574                 if (nfs_getattrcache(vp, ap->a_vap) == 0)
  575                         return (0);
  576         }
  577 
  578         nfsstats.rpccnt[NFSPROC_GETATTR]++;
  579         mreq = nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
  580         mb = mreq;
  581         bpos = mtod(mb, caddr_t);
  582         nfsm_fhtom(vp, v3);
  583         nfsm_request(vp, NFSPROC_GETATTR, ap->a_td, ap->a_cred);
  584         if (!error) {
  585                 nfsm_loadattr(vp, ap->a_vap);
  586         }
  587         m_freem(mrep);
  588 nfsmout:
  589         return (error);
  590 }
  591 
  592 /*
  593  * nfs setattr call.
  594  */
  595 static int
  596 nfs_setattr(struct vop_setattr_args *ap)
  597 {
  598         struct vnode *vp = ap->a_vp;
  599         struct nfsnode *np = VTONFS(vp);
  600         struct vattr *vap = ap->a_vap;
  601         int error = 0;
  602         u_quad_t tsize;
  603 
  604 #ifndef nolint
  605         tsize = (u_quad_t)0;
  606 #endif
  607 
  608         /*
  609          * Setting of flags is not supported.
  610          */
  611         if (vap->va_flags != VNOVAL)
  612                 return (EOPNOTSUPP);
  613 
  614         /*
  615          * Disallow write attempts if the filesystem is mounted read-only.
  616          */
  617         if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
  618             vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
  619             vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
  620             (vp->v_mount->mnt_flag & MNT_RDONLY))
  621                 return (EROFS);
  622         if (vap->va_size != VNOVAL) {
  623                 switch (vp->v_type) {
  624                 case VDIR:
  625                         return (EISDIR);
  626                 case VCHR:
  627                 case VBLK:
  628                 case VSOCK:
  629                 case VFIFO:
  630                         if (vap->va_mtime.tv_sec == VNOVAL &&
  631                             vap->va_atime.tv_sec == VNOVAL &&
  632                             vap->va_mode == (mode_t)VNOVAL &&
  633                             vap->va_uid == (uid_t)VNOVAL &&
  634                             vap->va_gid == (gid_t)VNOVAL)
  635                                 return (0);
  636                         vap->va_size = VNOVAL;
  637                         break;
  638                 default:
  639                         /*
  640                          * Disallow write attempts if the filesystem is
  641                          * mounted read-only.
  642                          */
  643                         if (vp->v_mount->mnt_flag & MNT_RDONLY)
  644                                 return (EROFS);
  645 
  646                         /*
  647                          *  We run vnode_pager_setsize() early (why?),
  648                          * we must set np->n_size now to avoid vinvalbuf
  649                          * V_SAVE races that might setsize a lower
  650                          * value.
  651                          */
  652 
  653                         tsize = np->n_size;
  654                         error = nfs_meta_setsize(vp, ap->a_cred, 
  655                                                 ap->a_td, vap->va_size);
  656 
  657                         if (np->n_flag & NMODIFIED) {
  658                             if (vap->va_size == 0)
  659                                 error = nfs_vinvalbuf(vp, 0,
  660                                         ap->a_cred, ap->a_td, 1);
  661                             else
  662                                 error = nfs_vinvalbuf(vp, V_SAVE,
  663                                         ap->a_cred, ap->a_td, 1);
  664                             if (error) {
  665                                 vnode_pager_setsize(vp, np->n_size);
  666                                 return (error);
  667                             }
  668                         }
  669                         np->n_vattr.va_size = vap->va_size;
  670                 };
  671         } else if ((vap->va_mtime.tv_sec != VNOVAL ||
  672                 vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NMODIFIED) &&
  673                 vp->v_type == VREG &&
  674                 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
  675                  ap->a_td, 1)) == EINTR)
  676                 return (error);
  677         error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_td);
  678         if (error && vap->va_size != VNOVAL) {
  679                 np->n_size = np->n_vattr.va_size = tsize;
  680                 vnode_pager_setsize(vp, np->n_size);
  681         }
  682         return (error);
  683 }
  684 
  685 /*
  686  * Do an nfs setattr rpc.
  687  */
  688 static int
  689 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
  690     struct thread *td)
  691 {
  692         struct nfsv2_sattr *sp;
  693         caddr_t bpos, dpos;
  694         u_int32_t *tl;
  695         int error = 0, wccflag = NFSV3_WCCRATTR;
  696         struct mbuf *mreq, *mrep, *md, *mb;
  697         int v3 = NFS_ISV3(vp);
  698 
  699         nfsstats.rpccnt[NFSPROC_SETATTR]++;
  700         mreq = nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
  701         mb = mreq;
  702         bpos = mtod(mb, caddr_t);
  703         nfsm_fhtom(vp, v3);
  704         if (v3) {
  705                 nfsm_v3attrbuild(vap, TRUE);
  706                 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
  707                 *tl = nfs_false;
  708         } else {
  709                 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
  710                 if (vap->va_mode == (mode_t)VNOVAL)
  711                         sp->sa_mode = nfs_xdrneg1;
  712                 else
  713                         sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
  714                 if (vap->va_uid == (uid_t)VNOVAL)
  715                         sp->sa_uid = nfs_xdrneg1;
  716                 else
  717                         sp->sa_uid = txdr_unsigned(vap->va_uid);
  718                 if (vap->va_gid == (gid_t)VNOVAL)
  719                         sp->sa_gid = nfs_xdrneg1;
  720                 else
  721                         sp->sa_gid = txdr_unsigned(vap->va_gid);
  722                 sp->sa_size = txdr_unsigned(vap->va_size);
  723                 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
  724                 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
  725         }
  726         nfsm_request(vp, NFSPROC_SETATTR, td, cred);
  727         if (v3) {
  728                 nfsm_wcc_data(vp, wccflag);
  729         } else
  730                 nfsm_loadattr(vp, NULL);
  731         m_freem(mrep);
  732 nfsmout:
  733         return (error);
  734 }
  735 
  736 /*
  737  * nfs lookup call, one step at a time...
  738  * First look in cache
  739  * If not found, unlock the directory nfsnode and do the rpc
  740  */
  741 static int
  742 nfs_lookup(struct vop_lookup_args *ap)
  743 {
  744         struct componentname *cnp = ap->a_cnp;
  745         struct vnode *dvp = ap->a_dvp;
  746         struct vnode **vpp = ap->a_vpp;
  747         int flags = cnp->cn_flags;
  748         struct vnode *newvp;
  749         struct nfsmount *nmp;
  750         caddr_t bpos, dpos;
  751         struct mbuf *mreq, *mrep, *md, *mb;
  752         long len;
  753         nfsfh_t *fhp;
  754         struct nfsnode *np;
  755         int lockparent, wantparent, error = 0, attrflag, fhsize;
  756         int v3 = NFS_ISV3(dvp);
  757         struct thread *td = cnp->cn_thread;
  758 
  759         *vpp = NULLVP;
  760         cnp->cn_flags &= ~PDIRUNLOCK;
  761         if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
  762             (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
  763                 return (EROFS);
  764         if (dvp->v_type != VDIR)
  765                 return (ENOTDIR);
  766         lockparent = flags & LOCKPARENT;
  767         wantparent = flags & (LOCKPARENT|WANTPARENT);
  768         nmp = VFSTONFS(dvp->v_mount);
  769         np = VTONFS(dvp);
  770         if ((error = cache_lookup(dvp, vpp, cnp)) && error != ENOENT) {
  771                 struct vattr vattr;
  772                 int vpid;
  773 
  774                 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) {
  775                         *vpp = NULLVP;
  776                         return (error);
  777                 }
  778 
  779                 vhold(*vpp);
  780                 newvp = *vpp;
  781                 vpid = newvp->v_id;
  782                 /*
  783                  * See the comment starting `Step through' in ufs/ufs_lookup.c
  784                  * for an explanation of the locking protocol
  785                  */
  786                 if (dvp == newvp) {
  787                         VREF(newvp);
  788                         error = 0;
  789                 } else if (flags & ISDOTDOT) {
  790                         VOP_UNLOCK(dvp, 0, td);
  791                         cnp->cn_flags |= PDIRUNLOCK;
  792                         error = vget(newvp, LK_EXCLUSIVE, td);
  793                         if (!error && lockparent && (flags & ISLASTCN)) {
  794                                 error = vn_lock(dvp, LK_EXCLUSIVE, td);
  795                                 if (error == 0)
  796                                         cnp->cn_flags &= ~PDIRUNLOCK;
  797                         }
  798                 } else {
  799                         error = vget(newvp, LK_EXCLUSIVE, td);
  800                         if (!lockparent || error || !(flags & ISLASTCN)) {
  801                                 VOP_UNLOCK(dvp, 0, td);
  802                                 cnp->cn_flags |= PDIRUNLOCK;
  803                         }
  804                 }
  805                 if (!error) {
  806                         if (vpid == newvp->v_id) {
  807                            if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, td)
  808                             && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) {
  809                                 nfsstats.lookupcache_hits++;
  810                                 if (cnp->cn_nameiop != LOOKUP &&
  811                                     (flags & ISLASTCN))
  812                                         cnp->cn_flags |= SAVENAME;
  813                                 vdrop(newvp);
  814                                 return (0);
  815                            }
  816                            cache_purge(newvp);
  817                         }
  818                         vput(newvp);
  819                         if (lockparent && dvp != newvp && (flags & ISLASTCN))
  820                                 VOP_UNLOCK(dvp, 0, td);
  821                 }
  822                 vdrop(newvp);
  823                 error = vn_lock(dvp, LK_EXCLUSIVE, td);
  824                 *vpp = NULLVP;
  825                 if (error) {
  826                         cnp->cn_flags |= PDIRUNLOCK;
  827                         return (error);
  828                 }
  829                 cnp->cn_flags &= ~PDIRUNLOCK;
  830         }
  831         error = 0;
  832         newvp = NULLVP;
  833         nfsstats.lookupcache_misses++;
  834         nfsstats.rpccnt[NFSPROC_LOOKUP]++;
  835         len = cnp->cn_namelen;
  836         mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
  837                 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
  838         mb = mreq;
  839         bpos = mtod(mb, caddr_t);
  840         nfsm_fhtom(dvp, v3);
  841         nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
  842         nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_thread, cnp->cn_cred);
  843         if (error) {
  844                 nfsm_postop_attr(dvp, attrflag);
  845                 m_freem(mrep);
  846                 goto nfsmout;
  847         }
  848         nfsm_getfh(fhp, fhsize, v3);
  849 
  850         /*
  851          * Handle RENAME case...
  852          */
  853         if (cnp->cn_nameiop == RENAME && wantparent && (flags & ISLASTCN)) {
  854                 if (NFS_CMPFH(np, fhp, fhsize)) {
  855                         m_freem(mrep);
  856                         return (EISDIR);
  857                 }
  858                 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
  859                 if (error) {
  860                         m_freem(mrep);
  861                         return (error);
  862                 }
  863                 newvp = NFSTOV(np);
  864                 if (v3) {
  865                         nfsm_postop_attr(newvp, attrflag);
  866                         nfsm_postop_attr(dvp, attrflag);
  867                 } else
  868                         nfsm_loadattr(newvp, NULL);
  869                 *vpp = newvp;
  870                 m_freem(mrep);
  871                 cnp->cn_flags |= SAVENAME;
  872                 if (!lockparent) {
  873                         VOP_UNLOCK(dvp, 0, td);
  874                         cnp->cn_flags |= PDIRUNLOCK;
  875                 }
  876                 return (0);
  877         }
  878 
  879         if (flags & ISDOTDOT) {
  880                 VOP_UNLOCK(dvp, 0, td);
  881                 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
  882                 if (error) {
  883                         vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td);
  884                         return (error);
  885                 }
  886                 newvp = NFSTOV(np);
  887                 if (lockparent && (flags & ISLASTCN)) {
  888                         error = vn_lock(dvp, LK_EXCLUSIVE, td);
  889                         if (error) {
  890                                 cnp->cn_flags |= PDIRUNLOCK;
  891                                 vput(newvp);
  892                                 return (error);
  893                         }
  894                 } else
  895                         cnp->cn_flags |= PDIRUNLOCK;
  896         } else if (NFS_CMPFH(np, fhp, fhsize)) {
  897                 VREF(dvp);
  898                 newvp = dvp;
  899         } else {
  900                 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
  901                 if (error) {
  902                         m_freem(mrep);
  903                         return (error);
  904                 }
  905                 if (!lockparent || !(flags & ISLASTCN)) {
  906                         cnp->cn_flags |= PDIRUNLOCK;
  907                         VOP_UNLOCK(dvp, 0, td);
  908                 }
  909                 newvp = NFSTOV(np);
  910         }
  911         if (v3) {
  912                 nfsm_postop_attr(newvp, attrflag);
  913                 nfsm_postop_attr(dvp, attrflag);
  914         } else
  915                 nfsm_loadattr(newvp, NULL);
  916         if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
  917                 cnp->cn_flags |= SAVENAME;
  918         if ((cnp->cn_flags & MAKEENTRY) &&
  919             (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
  920                 np->n_ctime = np->n_vattr.va_ctime.tv_sec;
  921                 cache_enter(dvp, newvp, cnp);
  922         }
  923         *vpp = newvp;
  924         m_freem(mrep);
  925 nfsmout:
  926         if (error) {
  927                 if (newvp != NULLVP) {
  928                         vrele(newvp);
  929                         *vpp = NULLVP;
  930                 }
  931                 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
  932                     (flags & ISLASTCN) && error == ENOENT) {
  933                         if (!lockparent) {
  934                                 VOP_UNLOCK(dvp, 0, td);
  935                                 cnp->cn_flags |= PDIRUNLOCK;
  936                         }
  937                         if (dvp->v_mount->mnt_flag & MNT_RDONLY)
  938                                 error = EROFS;
  939                         else
  940                                 error = EJUSTRETURN;
  941                 }
  942                 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
  943                         cnp->cn_flags |= SAVENAME;
  944         }
  945         return (error);
  946 }
  947 
  948 /*
  949  * nfs read call.
  950  * Just call nfs_bioread() to do the work.
  951  */
  952 static int
  953 nfs_read(struct vop_read_args *ap)
  954 {
  955         struct vnode *vp = ap->a_vp;
  956 
  957         if (vp->v_type != VREG)
  958                 return (EPERM);
  959         return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
  960 }
  961 
  962 /*
  963  * nfs readlink call
  964  */
  965 static int
  966 nfs_readlink(struct vop_readlink_args *ap)
  967 {
  968         struct vnode *vp = ap->a_vp;
  969 
  970         if (vp->v_type != VLNK)
  971                 return (EINVAL);
  972         return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred));
  973 }
  974 
  975 /*
  976  * Do a readlink rpc.
  977  * Called by nfs_doio() from below the buffer cache.
  978  */
  979 int
  980 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
  981 {
  982         caddr_t bpos, dpos;
  983         int error = 0, len, attrflag;
  984         struct mbuf *mreq, *mrep, *md, *mb;
  985         int v3 = NFS_ISV3(vp);
  986 
  987         nfsstats.rpccnt[NFSPROC_READLINK]++;
  988         mreq = nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
  989         mb = mreq;
  990         bpos = mtod(mb, caddr_t);
  991         nfsm_fhtom(vp, v3);
  992         nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, cred);
  993         if (v3)
  994                 nfsm_postop_attr(vp, attrflag);
  995         if (!error) {
  996                 nfsm_strsiz(len, NFS_MAXPATHLEN);
  997                 if (len == NFS_MAXPATHLEN) {
  998                         struct nfsnode *np = VTONFS(vp);
  999                         if (np->n_size && np->n_size < NFS_MAXPATHLEN)
 1000                                 len = np->n_size;
 1001                 }
 1002                 nfsm_mtouio(uiop, len);
 1003         }
 1004         m_freem(mrep);
 1005 nfsmout:
 1006         return (error);
 1007 }
 1008 
 1009 /*
 1010  * nfs read rpc call
 1011  * Ditto above
 1012  */
 1013 int
 1014 nfs_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
 1015 {
 1016         u_int32_t *tl;
 1017         caddr_t bpos, dpos;
 1018         struct mbuf *mreq, *mrep, *md, *mb;
 1019         struct nfsmount *nmp;
 1020         int error = 0, len, retlen, tsiz, eof, attrflag;
 1021         int v3 = NFS_ISV3(vp);
 1022 
 1023 #ifndef nolint
 1024         eof = 0;
 1025 #endif
 1026         nmp = VFSTONFS(vp->v_mount);
 1027         tsiz = uiop->uio_resid;
 1028         if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
 1029                 return (EFBIG);
 1030         while (tsiz > 0) {
 1031                 nfsstats.rpccnt[NFSPROC_READ]++;
 1032                 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
 1033                 mreq = nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
 1034                 mb = mreq;
 1035                 bpos = mtod(mb, caddr_t);
 1036                 nfsm_fhtom(vp, v3);
 1037                 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED * 3);
 1038                 if (v3) {
 1039                         txdr_hyper(uiop->uio_offset, tl);
 1040                         *(tl + 2) = txdr_unsigned(len);
 1041                 } else {
 1042                         *tl++ = txdr_unsigned(uiop->uio_offset);
 1043                         *tl++ = txdr_unsigned(len);
 1044                         *tl = 0;
 1045                 }
 1046                 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, cred);
 1047                 if (v3) {
 1048                         nfsm_postop_attr(vp, attrflag);
 1049                         if (error) {
 1050                                 m_freem(mrep);
 1051                                 goto nfsmout;
 1052                         }
 1053                         tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED);
 1054                         eof = fxdr_unsigned(int, *(tl + 1));
 1055                 } else
 1056                         nfsm_loadattr(vp, NULL);
 1057                 nfsm_strsiz(retlen, nmp->nm_rsize);
 1058                 nfsm_mtouio(uiop, retlen);
 1059                 m_freem(mrep);
 1060                 tsiz -= retlen;
 1061                 if (v3) {
 1062                         if (eof || retlen == 0) {
 1063                                 tsiz = 0;
 1064                         }
 1065                 } else if (retlen < len) {
 1066                         tsiz = 0;
 1067                 }
 1068         }
 1069 nfsmout:
 1070         return (error);
 1071 }
 1072 
 1073 /*
 1074  * nfs write call
 1075  */
 1076 int
 1077 nfs_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
 1078     int *iomode, int *must_commit)
 1079 {
 1080         u_int32_t *tl;
 1081         int32_t backup;
 1082         caddr_t bpos, dpos;
 1083         struct mbuf *mreq, *mrep, *md, *mb;
 1084         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 1085         int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
 1086         int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
 1087 
 1088 #ifndef DIAGNOSTIC
 1089         if (uiop->uio_iovcnt != 1)
 1090                 panic("nfs: writerpc iovcnt > 1");
 1091 #endif
 1092         *must_commit = 0;
 1093         tsiz = uiop->uio_resid;
 1094         if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
 1095                 return (EFBIG);
 1096         while (tsiz > 0) {
 1097                 nfsstats.rpccnt[NFSPROC_WRITE]++;
 1098                 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
 1099                 mreq = nfsm_reqhead(vp, NFSPROC_WRITE,
 1100                         NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
 1101                 mb = mreq;
 1102                 bpos = mtod(mb, caddr_t);
 1103                 nfsm_fhtom(vp, v3);
 1104                 if (v3) {
 1105                         tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
 1106                         txdr_hyper(uiop->uio_offset, tl);
 1107                         tl += 2;
 1108                         *tl++ = txdr_unsigned(len);
 1109                         *tl++ = txdr_unsigned(*iomode);
 1110                         *tl = txdr_unsigned(len);
 1111                 } else {
 1112                         u_int32_t x;
 1113 
 1114                         tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
 1115                         /* Set both "begin" and "current" to non-garbage. */
 1116                         x = txdr_unsigned((u_int32_t)uiop->uio_offset);
 1117                         *tl++ = x;      /* "begin offset" */
 1118                         *tl++ = x;      /* "current offset" */
 1119                         x = txdr_unsigned(len);
 1120                         *tl++ = x;      /* total to this offset */
 1121                         *tl = x;        /* size of this write */
 1122                 }
 1123                 nfsm_uiotom(uiop, len);
 1124                 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, cred);
 1125                 if (v3) {
 1126                         wccflag = NFSV3_WCCCHK;
 1127                         nfsm_wcc_data(vp, wccflag);
 1128                         if (!error) {
 1129                                 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED
 1130                                         + NFSX_V3WRITEVERF);
 1131                                 rlen = fxdr_unsigned(int, *tl++);
 1132                                 if (rlen == 0) {
 1133                                         error = NFSERR_IO;
 1134                                         m_freem(mrep);
 1135                                         break;
 1136                                 } else if (rlen < len) {
 1137                                         backup = len - rlen;
 1138                                         uiop->uio_iov->iov_base =
 1139                                             (char *)uiop->uio_iov->iov_base -
 1140                                             backup;
 1141                                         uiop->uio_iov->iov_len += backup;
 1142                                         uiop->uio_offset -= backup;
 1143                                         uiop->uio_resid += backup;
 1144                                         len = rlen;
 1145                                 }
 1146                                 commit = fxdr_unsigned(int, *tl++);
 1147 
 1148                                 /*
 1149                                  * Return the lowest committment level
 1150                                  * obtained by any of the RPCs.
 1151                                  */
 1152                                 if (committed == NFSV3WRITE_FILESYNC)
 1153                                         committed = commit;
 1154                                 else if (committed == NFSV3WRITE_DATASYNC &&
 1155                                         commit == NFSV3WRITE_UNSTABLE)
 1156                                         committed = commit;
 1157                                 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
 1158                                     bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
 1159                                         NFSX_V3WRITEVERF);
 1160                                     nmp->nm_state |= NFSSTA_HASWRITEVERF;
 1161                                 } else if (bcmp((caddr_t)tl,
 1162                                     (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
 1163                                     *must_commit = 1;
 1164                                     bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
 1165                                         NFSX_V3WRITEVERF);
 1166                                 }
 1167                         }
 1168                 } else
 1169                     nfsm_loadattr(vp, NULL);
 1170                 if (wccflag)
 1171                     VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime.tv_sec;
 1172                 m_freem(mrep);
 1173                 if (error)
 1174                         break;
 1175                 tsiz -= len;
 1176         }
 1177 nfsmout:
 1178         if (vp->v_mount->mnt_flag & MNT_ASYNC)
 1179                 committed = NFSV3WRITE_FILESYNC;
 1180         *iomode = committed;
 1181         if (error)
 1182                 uiop->uio_resid = tsiz;
 1183         return (error);
 1184 }
 1185 
 1186 /*
 1187  * nfs mknod rpc
 1188  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
 1189  * mode set to specify the file type and the size field for rdev.
 1190  */
 1191 static int
 1192 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
 1193     struct vattr *vap)
 1194 {
 1195         struct nfsv2_sattr *sp;
 1196         u_int32_t *tl;
 1197         struct vnode *newvp = NULL;
 1198         struct nfsnode *np = NULL;
 1199         struct vattr vattr;
 1200         caddr_t bpos, dpos;
 1201         int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
 1202         struct mbuf *mreq, *mrep, *md, *mb;
 1203         u_int32_t rdev;
 1204         int v3 = NFS_ISV3(dvp);
 1205 
 1206         if (vap->va_type == VCHR || vap->va_type == VBLK)
 1207                 rdev = txdr_unsigned(vap->va_rdev);
 1208         else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
 1209                 rdev = nfs_xdrneg1;
 1210         else {
 1211                 return (EOPNOTSUPP);
 1212         }
 1213         if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
 1214                 return (error);
 1215         }
 1216         nfsstats.rpccnt[NFSPROC_MKNOD]++;
 1217         mreq = nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
 1218                 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
 1219         mb = mreq;
 1220         bpos = mtod(mb, caddr_t);
 1221         nfsm_fhtom(dvp, v3);
 1222         nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
 1223         if (v3) {
 1224                 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
 1225                 *tl++ = vtonfsv3_type(vap->va_type);
 1226                 nfsm_v3attrbuild(vap, FALSE);
 1227                 if (vap->va_type == VCHR || vap->va_type == VBLK) {
 1228                         tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
 1229                         *tl++ = txdr_unsigned(umajor(vap->va_rdev));
 1230                         *tl = txdr_unsigned(uminor(vap->va_rdev));
 1231                 }
 1232         } else {
 1233                 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
 1234                 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
 1235                 sp->sa_uid = nfs_xdrneg1;
 1236                 sp->sa_gid = nfs_xdrneg1;
 1237                 sp->sa_size = rdev;
 1238                 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
 1239                 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
 1240         }
 1241         nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_thread, cnp->cn_cred);
 1242         if (!error) {
 1243                 nfsm_mtofh(dvp, newvp, v3, gotvp);
 1244                 if (!gotvp) {
 1245                         if (newvp) {
 1246                                 vput(newvp);
 1247                                 newvp = NULL;
 1248                         }
 1249                         error = nfs_lookitup(dvp, cnp->cn_nameptr,
 1250                             cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
 1251                         if (!error)
 1252                                 newvp = NFSTOV(np);
 1253                 }
 1254         }
 1255         if (v3)
 1256                 nfsm_wcc_data(dvp, wccflag);
 1257         m_freem(mrep);
 1258 nfsmout:
 1259         if (error) {
 1260                 if (newvp)
 1261                         vput(newvp);
 1262         } else {
 1263                 if (cnp->cn_flags & MAKEENTRY)
 1264                         cache_enter(dvp, newvp, cnp);
 1265                 *vpp = newvp;
 1266         }
 1267         VTONFS(dvp)->n_flag |= NMODIFIED;
 1268         if (!wccflag)
 1269                 VTONFS(dvp)->n_attrstamp = 0;
 1270         return (error);
 1271 }
 1272 
 1273 /*
 1274  * nfs mknod vop
 1275  * just call nfs_mknodrpc() to do the work.
 1276  */
 1277 /* ARGSUSED */
 1278 static int
 1279 nfs_mknod(struct vop_mknod_args *ap)
 1280 {
 1281 
 1282         return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
 1283 }
 1284 
 1285 static u_long create_verf;
 1286 /*
 1287  * nfs file create call
 1288  */
 1289 static int
 1290 nfs_create(struct vop_create_args *ap)
 1291 {
 1292         struct vnode *dvp = ap->a_dvp;
 1293         struct vattr *vap = ap->a_vap;
 1294         struct componentname *cnp = ap->a_cnp;
 1295         struct nfsv2_sattr *sp;
 1296         u_int32_t *tl;
 1297         struct nfsnode *np = NULL;
 1298         struct vnode *newvp = NULL;
 1299         caddr_t bpos, dpos;
 1300         int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
 1301         struct mbuf *mreq, *mrep, *md, *mb;
 1302         struct vattr vattr;
 1303         int v3 = NFS_ISV3(dvp);
 1304 
 1305         /*
 1306          * Oops, not for me..
 1307          */
 1308         if (vap->va_type == VSOCK)
 1309                 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
 1310 
 1311         if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
 1312                 return (error);
 1313         }
 1314         if (vap->va_vaflags & VA_EXCLUSIVE)
 1315                 fmode |= O_EXCL;
 1316 again:
 1317         nfsstats.rpccnt[NFSPROC_CREATE]++;
 1318         mreq = nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
 1319                 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
 1320         mb = mreq;
 1321         bpos = mtod(mb, caddr_t);
 1322         nfsm_fhtom(dvp, v3);
 1323         nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
 1324         if (v3) {
 1325                 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
 1326                 if (fmode & O_EXCL) {
 1327                         *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
 1328                         tl = nfsm_build(u_int32_t *, NFSX_V3CREATEVERF);
 1329 #ifdef INET
 1330                         if (!TAILQ_EMPTY(&in_ifaddrhead))
 1331                                 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrhead))->sin_addr.s_addr;
 1332                         else
 1333 #endif
 1334                                 *tl++ = create_verf;
 1335                         *tl = ++create_verf;
 1336                 } else {
 1337                         *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
 1338                         nfsm_v3attrbuild(vap, FALSE);
 1339                 }
 1340         } else {
 1341                 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
 1342                 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
 1343                 sp->sa_uid = nfs_xdrneg1;
 1344                 sp->sa_gid = nfs_xdrneg1;
 1345                 sp->sa_size = 0;
 1346                 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
 1347                 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
 1348         }
 1349         nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_thread, cnp->cn_cred);
 1350         if (!error) {
 1351                 nfsm_mtofh(dvp, newvp, v3, gotvp);
 1352                 if (!gotvp) {
 1353                         if (newvp) {
 1354                                 vput(newvp);
 1355                                 newvp = NULL;
 1356                         }
 1357                         error = nfs_lookitup(dvp, cnp->cn_nameptr,
 1358                             cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
 1359                         if (!error)
 1360                                 newvp = NFSTOV(np);
 1361                 }
 1362         }
 1363         if (v3)
 1364                 nfsm_wcc_data(dvp, wccflag);
 1365         m_freem(mrep);
 1366 nfsmout:
 1367         if (error) {
 1368                 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
 1369                         fmode &= ~O_EXCL;
 1370                         goto again;
 1371                 }
 1372                 if (newvp)
 1373                         vput(newvp);
 1374         } else if (v3 && (fmode & O_EXCL)) {
 1375                 /*
 1376                  * We are normally called with only a partially initialized
 1377                  * VAP.  Since the NFSv3 spec says that server may use the
 1378                  * file attributes to store the verifier, the spec requires
 1379                  * us to do a SETATTR RPC. FreeBSD servers store the verifier
 1380                  * in atime, but we can't really assume that all servers will
 1381                  * so we ensure that our SETATTR sets both atime and mtime.
 1382                  */
 1383                 if (vap->va_mtime.tv_sec == VNOVAL)
 1384                         vfs_timestamp(&vap->va_mtime);
 1385                 if (vap->va_atime.tv_sec == VNOVAL)
 1386                         vap->va_atime = vap->va_mtime;
 1387                 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_thread);
 1388         }
 1389         if (!error) {
 1390                 if (cnp->cn_flags & MAKEENTRY)
 1391                         cache_enter(dvp, newvp, cnp);
 1392                 *ap->a_vpp = newvp;
 1393         }
 1394         VTONFS(dvp)->n_flag |= NMODIFIED;
 1395         if (!wccflag)
 1396                 VTONFS(dvp)->n_attrstamp = 0;
 1397         return (error);
 1398 }
 1399 
 1400 /*
 1401  * nfs file remove call
 1402  * To try and make nfs semantics closer to ufs semantics, a file that has
 1403  * other processes using the vnode is renamed instead of removed and then
 1404  * removed later on the last close.
 1405  * - If v_usecount > 1
 1406  *        If a rename is not already in the works
 1407  *           call nfs_sillyrename() to set it up
 1408  *     else
 1409  *        do the remove rpc
 1410  */
 1411 static int
 1412 nfs_remove(struct vop_remove_args *ap)
 1413 {
 1414         struct vnode *vp = ap->a_vp;
 1415         struct vnode *dvp = ap->a_dvp;
 1416         struct componentname *cnp = ap->a_cnp;
 1417         struct nfsnode *np = VTONFS(vp);
 1418         int error = 0;
 1419         struct vattr vattr;
 1420 
 1421 #ifndef DIAGNOSTIC
 1422         if ((cnp->cn_flags & HASBUF) == 0)
 1423                 panic("nfs_remove: no name");
 1424         if (vrefcnt(vp) < 1)
 1425                 panic("nfs_remove: bad v_usecount");
 1426 #endif
 1427         if (vp->v_type == VDIR)
 1428                 error = EPERM;
 1429         else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
 1430             VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_thread) == 0 &&
 1431             vattr.va_nlink > 1)) {
 1432                 /*
 1433                  * Purge the name cache so that the chance of a lookup for
 1434                  * the name succeeding while the remove is in progress is
 1435                  * minimized. Without node locking it can still happen, such
 1436                  * that an I/O op returns ESTALE, but since you get this if
 1437                  * another host removes the file..
 1438                  */
 1439                 cache_purge(vp);
 1440                 /*
 1441                  * throw away biocache buffers, mainly to avoid
 1442                  * unnecessary delayed writes later.
 1443                  */
 1444                 error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, cnp->cn_thread, 1);
 1445                 /* Do the rpc */
 1446                 if (error != EINTR)
 1447                         error = nfs_removerpc(dvp, cnp->cn_nameptr,
 1448                                 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
 1449                 /*
 1450                  * Kludge City: If the first reply to the remove rpc is lost..
 1451                  *   the reply to the retransmitted request will be ENOENT
 1452                  *   since the file was in fact removed
 1453                  *   Therefore, we cheat and return success.
 1454                  */
 1455                 if (error == ENOENT)
 1456                         error = 0;
 1457         } else if (!np->n_sillyrename)
 1458                 error = nfs_sillyrename(dvp, vp, cnp);
 1459         np->n_attrstamp = 0;
 1460         return (error);
 1461 }
 1462 
 1463 /*
 1464  * nfs file remove rpc called from nfs_inactive
 1465  */
 1466 int
 1467 nfs_removeit(struct sillyrename *sp)
 1468 {
 1469 
 1470         return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
 1471                 NULL));
 1472 }
 1473 
 1474 /*
 1475  * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
 1476  */
 1477 static int
 1478 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
 1479     struct ucred *cred, struct thread *td)
 1480 {
 1481         caddr_t bpos, dpos;
 1482         int error = 0, wccflag = NFSV3_WCCRATTR;
 1483         struct mbuf *mreq, *mrep, *md, *mb;
 1484         int v3 = NFS_ISV3(dvp);
 1485 
 1486         nfsstats.rpccnt[NFSPROC_REMOVE]++;
 1487         mreq = nfsm_reqhead(dvp, NFSPROC_REMOVE,
 1488                 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
 1489         mb = mreq;
 1490         bpos = mtod(mb, caddr_t);
 1491         nfsm_fhtom(dvp, v3);
 1492         nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
 1493         nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
 1494         if (v3)
 1495                 nfsm_wcc_data(dvp, wccflag);
 1496         m_freem(mrep);
 1497 nfsmout:
 1498         VTONFS(dvp)->n_flag |= NMODIFIED;
 1499         if (!wccflag)
 1500                 VTONFS(dvp)->n_attrstamp = 0;
 1501         return (error);
 1502 }
 1503 
 1504 /*
 1505  * nfs file rename call
 1506  */
 1507 static int
 1508 nfs_rename(struct vop_rename_args *ap)
 1509 {
 1510         struct vnode *fvp = ap->a_fvp;
 1511         struct vnode *tvp = ap->a_tvp;
 1512         struct vnode *fdvp = ap->a_fdvp;
 1513         struct vnode *tdvp = ap->a_tdvp;
 1514         struct componentname *tcnp = ap->a_tcnp;
 1515         struct componentname *fcnp = ap->a_fcnp;
 1516         int error;
 1517 
 1518 #ifndef DIAGNOSTIC
 1519         if ((tcnp->cn_flags & HASBUF) == 0 ||
 1520             (fcnp->cn_flags & HASBUF) == 0)
 1521                 panic("nfs_rename: no name");
 1522 #endif
 1523         /* Check for cross-device rename */
 1524         if ((fvp->v_mount != tdvp->v_mount) ||
 1525             (tvp && (fvp->v_mount != tvp->v_mount))) {
 1526                 error = EXDEV;
 1527                 goto out;
 1528         }
 1529 
 1530         /*
 1531          * We have to flush B_DELWRI data prior to renaming
 1532          * the file.  If we don't, the delayed-write buffers
 1533          * can be flushed out later after the file has gone stale
 1534          * under NFSV3.  NFSV2 does not have this problem because
 1535          * ( as far as I can tell ) it flushes dirty buffers more
 1536          * often.
 1537          */
 1538 
 1539         VOP_FSYNC(fvp, fcnp->cn_cred, MNT_WAIT, fcnp->cn_thread);
 1540         if (tvp)
 1541             VOP_FSYNC(tvp, tcnp->cn_cred, MNT_WAIT, tcnp->cn_thread);
 1542 
 1543         /*
 1544          * If the tvp exists and is in use, sillyrename it before doing the
 1545          * rename of the new file over it.
 1546          * XXX Can't sillyrename a directory.
 1547          */
 1548         if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
 1549                 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
 1550                 vput(tvp);
 1551                 tvp = NULL;
 1552         }
 1553 
 1554         error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
 1555                 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
 1556                 tcnp->cn_thread);
 1557 
 1558         if (fvp->v_type == VDIR) {
 1559                 if (tvp != NULL && tvp->v_type == VDIR)
 1560                         cache_purge(tdvp);
 1561                 cache_purge(fdvp);
 1562         }
 1563 
 1564 out:
 1565         if (tdvp == tvp)
 1566                 vrele(tdvp);
 1567         else
 1568                 vput(tdvp);
 1569         if (tvp)
 1570                 vput(tvp);
 1571         vrele(fdvp);
 1572         vrele(fvp);
 1573         /*
 1574          * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
 1575          */
 1576         if (error == ENOENT)
 1577                 error = 0;
 1578         return (error);
 1579 }
 1580 
 1581 /*
 1582  * nfs file rename rpc called from nfs_remove() above
 1583  */
 1584 static int
 1585 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
 1586     struct sillyrename *sp)
 1587 {
 1588 
 1589         return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, sdvp,
 1590             sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_thread));
 1591 }
 1592 
 1593 /*
 1594  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
 1595  */
 1596 static int
 1597 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
 1598     struct vnode *tdvp, const char *tnameptr, int tnamelen, struct ucred *cred,
 1599     struct thread *td)
 1600 {
 1601         caddr_t bpos, dpos;
 1602         int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
 1603         struct mbuf *mreq, *mrep, *md, *mb;
 1604         int v3 = NFS_ISV3(fdvp);
 1605 
 1606         nfsstats.rpccnt[NFSPROC_RENAME]++;
 1607         mreq = nfsm_reqhead(fdvp, NFSPROC_RENAME,
 1608                 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
 1609                 nfsm_rndup(tnamelen));
 1610         mb = mreq;
 1611         bpos = mtod(mb, caddr_t);
 1612         nfsm_fhtom(fdvp, v3);
 1613         nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
 1614         nfsm_fhtom(tdvp, v3);
 1615         nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
 1616         nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
 1617         if (v3) {
 1618                 nfsm_wcc_data(fdvp, fwccflag);
 1619                 nfsm_wcc_data(tdvp, twccflag);
 1620         }
 1621         m_freem(mrep);
 1622 nfsmout:
 1623         VTONFS(fdvp)->n_flag |= NMODIFIED;
 1624         VTONFS(tdvp)->n_flag |= NMODIFIED;
 1625         if (!fwccflag)
 1626                 VTONFS(fdvp)->n_attrstamp = 0;
 1627         if (!twccflag)
 1628                 VTONFS(tdvp)->n_attrstamp = 0;
 1629         return (error);
 1630 }
 1631 
 1632 /*
 1633  * nfs hard link create call
 1634  */
 1635 static int
 1636 nfs_link(struct vop_link_args *ap)
 1637 {
 1638         struct vnode *vp = ap->a_vp;
 1639         struct vnode *tdvp = ap->a_tdvp;
 1640         struct componentname *cnp = ap->a_cnp;
 1641         caddr_t bpos, dpos;
 1642         int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
 1643         struct mbuf *mreq, *mrep, *md, *mb;
 1644         int v3;
 1645 
 1646         if (vp->v_mount != tdvp->v_mount) {
 1647                 return (EXDEV);
 1648         }
 1649 
 1650         /*
 1651          * Push all writes to the server, so that the attribute cache
 1652          * doesn't get "out of sync" with the server.
 1653          * XXX There should be a better way!
 1654          */
 1655         VOP_FSYNC(vp, cnp->cn_cred, MNT_WAIT, cnp->cn_thread);
 1656 
 1657         v3 = NFS_ISV3(vp);
 1658         nfsstats.rpccnt[NFSPROC_LINK]++;
 1659         mreq = nfsm_reqhead(vp, NFSPROC_LINK,
 1660                 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
 1661         mb = mreq;
 1662         bpos = mtod(mb, caddr_t);
 1663         nfsm_fhtom(vp, v3);
 1664         nfsm_fhtom(tdvp, v3);
 1665         nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
 1666         nfsm_request(vp, NFSPROC_LINK, cnp->cn_thread, cnp->cn_cred);
 1667         if (v3) {
 1668                 nfsm_postop_attr(vp, attrflag);
 1669                 nfsm_wcc_data(tdvp, wccflag);
 1670         }
 1671         m_freem(mrep);
 1672 nfsmout:
 1673         VTONFS(tdvp)->n_flag |= NMODIFIED;
 1674         if (!attrflag)
 1675                 VTONFS(vp)->n_attrstamp = 0;
 1676         if (!wccflag)
 1677                 VTONFS(tdvp)->n_attrstamp = 0;
 1678         /*
 1679          * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
 1680          */
 1681         if (error == EEXIST)
 1682                 error = 0;
 1683         return (error);
 1684 }
 1685 
 1686 /*
 1687  * nfs symbolic link create call
 1688  */
 1689 static int
 1690 nfs_symlink(struct vop_symlink_args *ap)
 1691 {
 1692         struct vnode *dvp = ap->a_dvp;
 1693         struct vattr *vap = ap->a_vap;
 1694         struct componentname *cnp = ap->a_cnp;
 1695         struct nfsv2_sattr *sp;
 1696         caddr_t bpos, dpos;
 1697         int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
 1698         struct mbuf *mreq, *mrep, *md, *mb;
 1699         struct vnode *newvp = NULL;
 1700         int v3 = NFS_ISV3(dvp);
 1701 
 1702         nfsstats.rpccnt[NFSPROC_SYMLINK]++;
 1703         slen = strlen(ap->a_target);
 1704         mreq = nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
 1705             nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
 1706         mb = mreq;
 1707         bpos = mtod(mb, caddr_t);
 1708         nfsm_fhtom(dvp, v3);
 1709         nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
 1710         if (v3) {
 1711                 nfsm_v3attrbuild(vap, FALSE);
 1712         }
 1713         nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
 1714         if (!v3) {
 1715                 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
 1716                 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
 1717                 sp->sa_uid = nfs_xdrneg1;
 1718                 sp->sa_gid = nfs_xdrneg1;
 1719                 sp->sa_size = nfs_xdrneg1;
 1720                 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
 1721                 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
 1722         }
 1723 
 1724         /*
 1725          * Issue the NFS request and get the rpc response.
 1726          *
 1727          * Only NFSv3 responses returning an error of 0 actually return
 1728          * a file handle that can be converted into newvp without having
 1729          * to do an extra lookup rpc.
 1730          */
 1731         nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_thread, cnp->cn_cred);
 1732         if (v3) {
 1733                 if (error == 0)
 1734                         nfsm_mtofh(dvp, newvp, v3, gotvp);
 1735                 nfsm_wcc_data(dvp, wccflag);
 1736         }
 1737 
 1738         /*
 1739          * out code jumps -> here, mrep is also freed.
 1740          */
 1741 
 1742         m_freem(mrep);
 1743 nfsmout:
 1744 
 1745         /*
 1746          * If we get an EEXIST error, silently convert it to no-error
 1747          * in case of an NFS retry.
 1748          */
 1749         if (error == EEXIST)
 1750                 error = 0;
 1751 
 1752         /*
 1753          * If we do not have (or no longer have) an error, and we could
 1754          * not extract the newvp from the response due to the request being
 1755          * NFSv2 or the error being EEXIST.  We have to do a lookup in order
 1756          * to obtain a newvp to return.
 1757          */
 1758         if (error == 0 && newvp == NULL) {
 1759                 struct nfsnode *np = NULL;
 1760 
 1761                 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
 1762                     cnp->cn_cred, cnp->cn_thread, &np);
 1763                 if (!error)
 1764                         newvp = NFSTOV(np);
 1765         }
 1766         if (error) {
 1767                 if (newvp)
 1768                         vput(newvp);
 1769         } else {
 1770                 *ap->a_vpp = newvp;
 1771         }
 1772         VTONFS(dvp)->n_flag |= NMODIFIED;
 1773         if (!wccflag)
 1774                 VTONFS(dvp)->n_attrstamp = 0;
 1775         return (error);
 1776 }
 1777 
 1778 /*
 1779  * nfs make dir call
 1780  */
 1781 static int
 1782 nfs_mkdir(struct vop_mkdir_args *ap)
 1783 {
 1784         struct vnode *dvp = ap->a_dvp;
 1785         struct vattr *vap = ap->a_vap;
 1786         struct componentname *cnp = ap->a_cnp;
 1787         struct nfsv2_sattr *sp;
 1788         int len;
 1789         struct nfsnode *np = NULL;
 1790         struct vnode *newvp = NULL;
 1791         caddr_t bpos, dpos;
 1792         int error = 0, wccflag = NFSV3_WCCRATTR;
 1793         int gotvp = 0;
 1794         struct mbuf *mreq, *mrep, *md, *mb;
 1795         struct vattr vattr;
 1796         int v3 = NFS_ISV3(dvp);
 1797 
 1798         if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
 1799                 return (error);
 1800         }
 1801         len = cnp->cn_namelen;
 1802         nfsstats.rpccnt[NFSPROC_MKDIR]++;
 1803         mreq = nfsm_reqhead(dvp, NFSPROC_MKDIR,
 1804           NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
 1805         mb = mreq;
 1806         bpos = mtod(mb, caddr_t);
 1807         nfsm_fhtom(dvp, v3);
 1808         nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
 1809         if (v3) {
 1810                 nfsm_v3attrbuild(vap, FALSE);
 1811         } else {
 1812                 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
 1813                 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
 1814                 sp->sa_uid = nfs_xdrneg1;
 1815                 sp->sa_gid = nfs_xdrneg1;
 1816                 sp->sa_size = nfs_xdrneg1;
 1817                 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
 1818                 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
 1819         }
 1820         nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_thread, cnp->cn_cred);
 1821         if (!error)
 1822                 nfsm_mtofh(dvp, newvp, v3, gotvp);
 1823         if (v3)
 1824                 nfsm_wcc_data(dvp, wccflag);
 1825         m_freem(mrep);
 1826 nfsmout:
 1827         VTONFS(dvp)->n_flag |= NMODIFIED;
 1828         if (!wccflag)
 1829                 VTONFS(dvp)->n_attrstamp = 0;
 1830         /*
 1831          * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
 1832          * if we can succeed in looking up the directory.
 1833          */
 1834         if (error == EEXIST || (!error && !gotvp)) {
 1835                 if (newvp) {
 1836                         vrele(newvp);
 1837                         newvp = NULL;
 1838                 }
 1839                 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
 1840                         cnp->cn_thread, &np);
 1841                 if (!error) {
 1842                         newvp = NFSTOV(np);
 1843                         if (newvp->v_type != VDIR)
 1844                                 error = EEXIST;
 1845                 }
 1846         }
 1847         if (error) {
 1848                 if (newvp)
 1849                         vrele(newvp);
 1850         } else
 1851                 *ap->a_vpp = newvp;
 1852         return (error);
 1853 }
 1854 
 1855 /*
 1856  * nfs remove directory call
 1857  */
 1858 static int
 1859 nfs_rmdir(struct vop_rmdir_args *ap)
 1860 {
 1861         struct vnode *vp = ap->a_vp;
 1862         struct vnode *dvp = ap->a_dvp;
 1863         struct componentname *cnp = ap->a_cnp;
 1864         caddr_t bpos, dpos;
 1865         int error = 0, wccflag = NFSV3_WCCRATTR;
 1866         struct mbuf *mreq, *mrep, *md, *mb;
 1867         int v3 = NFS_ISV3(dvp);
 1868 
 1869         if (dvp == vp)
 1870                 return (EINVAL);
 1871         nfsstats.rpccnt[NFSPROC_RMDIR]++;
 1872         mreq = nfsm_reqhead(dvp, NFSPROC_RMDIR,
 1873                 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
 1874         mb = mreq;
 1875         bpos = mtod(mb, caddr_t);
 1876         nfsm_fhtom(dvp, v3);
 1877         nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
 1878         nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_thread, cnp->cn_cred);
 1879         if (v3)
 1880                 nfsm_wcc_data(dvp, wccflag);
 1881         m_freem(mrep);
 1882 nfsmout:
 1883         VTONFS(dvp)->n_flag |= NMODIFIED;
 1884         if (!wccflag)
 1885                 VTONFS(dvp)->n_attrstamp = 0;
 1886         cache_purge(dvp);
 1887         cache_purge(vp);
 1888         /*
 1889          * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
 1890          */
 1891         if (error == ENOENT)
 1892                 error = 0;
 1893         return (error);
 1894 }
 1895 
 1896 /*
 1897  * nfs readdir call
 1898  */
 1899 static int
 1900 nfs_readdir(struct vop_readdir_args *ap)
 1901 {
 1902         struct vnode *vp = ap->a_vp;
 1903         struct nfsnode *np = VTONFS(vp);
 1904         struct uio *uio = ap->a_uio;
 1905         int tresid, error;
 1906         struct vattr vattr;
 1907 
 1908         if (vp->v_type != VDIR)
 1909                 return (EPERM);
 1910         /*
 1911          * First, check for hit on the EOF offset cache
 1912          */
 1913         if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
 1914             (np->n_flag & NMODIFIED) == 0) {
 1915                 if (VOP_GETATTR(vp, &vattr, ap->a_cred, uio->uio_td) == 0 &&
 1916                         np->n_mtime == vattr.va_mtime.tv_sec) {
 1917                         nfsstats.direofcache_hits++;
 1918                         return (0);
 1919                 }
 1920         }
 1921 
 1922         /*
 1923          * Call nfs_bioread() to do the real work.
 1924          */
 1925         tresid = uio->uio_resid;
 1926         error = nfs_bioread(vp, uio, 0, ap->a_cred);
 1927 
 1928         if (!error && uio->uio_resid == tresid)
 1929                 nfsstats.direofcache_misses++;
 1930         return (error);
 1931 }
 1932 
 1933 /*
 1934  * Readdir rpc call.
 1935  * Called from below the buffer cache by nfs_doio().
 1936  */
 1937 int
 1938 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
 1939 {
 1940         int len, left;
 1941         struct dirent *dp = NULL;
 1942         u_int32_t *tl;
 1943         caddr_t cp;
 1944         nfsuint64 *cookiep;
 1945         caddr_t bpos, dpos;
 1946         struct mbuf *mreq, *mrep, *md, *mb;
 1947         nfsuint64 cookie;
 1948         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 1949         struct nfsnode *dnp = VTONFS(vp);
 1950         u_quad_t fileno;
 1951         int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
 1952         int attrflag;
 1953         int v3 = NFS_ISV3(vp);
 1954 
 1955 #ifndef DIAGNOSTIC
 1956         if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
 1957                 (uiop->uio_resid & (DIRBLKSIZ - 1)))
 1958                 panic("nfs readdirrpc bad uio");
 1959 #endif
 1960 
 1961         /*
 1962          * If there is no cookie, assume directory was stale.
 1963          */
 1964         cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
 1965         if (cookiep)
 1966                 cookie = *cookiep;
 1967         else
 1968                 return (NFSERR_BAD_COOKIE);
 1969         /*
 1970          * Loop around doing readdir rpc's of size nm_readdirsize
 1971          * truncated to a multiple of DIRBLKSIZ.
 1972          * The stopping criteria is EOF or buffer full.
 1973          */
 1974         while (more_dirs && bigenough) {
 1975                 nfsstats.rpccnt[NFSPROC_READDIR]++;
 1976                 mreq = nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
 1977                         NFSX_READDIR(v3));
 1978                 mb = mreq;
 1979                 bpos = mtod(mb, caddr_t);
 1980                 nfsm_fhtom(vp, v3);
 1981                 if (v3) {
 1982                         tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
 1983                         *tl++ = cookie.nfsuquad[0];
 1984                         *tl++ = cookie.nfsuquad[1];
 1985                         *tl++ = dnp->n_cookieverf.nfsuquad[0];
 1986                         *tl++ = dnp->n_cookieverf.nfsuquad[1];
 1987                 } else {
 1988                         tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
 1989                         *tl++ = cookie.nfsuquad[0];
 1990                 }
 1991                 *tl = txdr_unsigned(nmp->nm_readdirsize);
 1992                 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, cred);
 1993                 if (v3) {
 1994                         nfsm_postop_attr(vp, attrflag);
 1995                         if (!error) {
 1996                                 tl = nfsm_dissect(u_int32_t *,
 1997                                     2 * NFSX_UNSIGNED);
 1998                                 dnp->n_cookieverf.nfsuquad[0] = *tl++;
 1999                                 dnp->n_cookieverf.nfsuquad[1] = *tl;
 2000                         } else {
 2001                                 m_freem(mrep);
 2002                                 goto nfsmout;
 2003                         }
 2004                 }
 2005                 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2006                 more_dirs = fxdr_unsigned(int, *tl);
 2007 
 2008                 /* loop thru the dir entries, doctoring them to 4bsd form */
 2009                 while (more_dirs && bigenough) {
 2010                         if (v3) {
 2011                                 tl = nfsm_dissect(u_int32_t *,
 2012                                     3 * NFSX_UNSIGNED);
 2013                                 fileno = fxdr_hyper(tl);
 2014                                 len = fxdr_unsigned(int, *(tl + 2));
 2015                         } else {
 2016                                 tl = nfsm_dissect(u_int32_t *,
 2017                                     2 * NFSX_UNSIGNED);
 2018                                 fileno = fxdr_unsigned(u_quad_t, *tl++);
 2019                                 len = fxdr_unsigned(int, *tl);
 2020                         }
 2021                         if (len <= 0 || len > NFS_MAXNAMLEN) {
 2022                                 error = EBADRPC;
 2023                                 m_freem(mrep);
 2024                                 goto nfsmout;
 2025                         }
 2026                         tlen = nfsm_rndup(len);
 2027                         if (tlen == len)
 2028                                 tlen += 4;      /* To ensure null termination */
 2029                         left = DIRBLKSIZ - blksiz;
 2030                         if ((tlen + DIRHDSIZ) > left) {
 2031                                 dp->d_reclen += left;
 2032                                 uiop->uio_iov->iov_base =
 2033                                     (char *)uiop->uio_iov->iov_base + left;
 2034                                 uiop->uio_iov->iov_len -= left;
 2035                                 uiop->uio_offset += left;
 2036                                 uiop->uio_resid -= left;
 2037                                 blksiz = 0;
 2038                         }
 2039                         if ((tlen + DIRHDSIZ) > uiop->uio_resid)
 2040                                 bigenough = 0;
 2041                         if (bigenough) {
 2042                                 dp = (struct dirent *)uiop->uio_iov->iov_base;
 2043                                 dp->d_fileno = (int)fileno;
 2044                                 dp->d_namlen = len;
 2045                                 dp->d_reclen = tlen + DIRHDSIZ;
 2046                                 dp->d_type = DT_UNKNOWN;
 2047                                 blksiz += dp->d_reclen;
 2048                                 if (blksiz == DIRBLKSIZ)
 2049                                         blksiz = 0;
 2050                                 uiop->uio_offset += DIRHDSIZ;
 2051                                 uiop->uio_resid -= DIRHDSIZ;
 2052                                 uiop->uio_iov->iov_base =
 2053                                     (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
 2054                                 uiop->uio_iov->iov_len -= DIRHDSIZ;
 2055                                 nfsm_mtouio(uiop, len);
 2056                                 cp = uiop->uio_iov->iov_base;
 2057                                 tlen -= len;
 2058                                 *cp = '\0';     /* null terminate */
 2059                                 uiop->uio_iov->iov_base =
 2060                                     (char *)uiop->uio_iov->iov_base + tlen;
 2061                                 uiop->uio_iov->iov_len -= tlen;
 2062                                 uiop->uio_offset += tlen;
 2063                                 uiop->uio_resid -= tlen;
 2064                         } else
 2065                                 nfsm_adv(nfsm_rndup(len));
 2066                         if (v3) {
 2067                                 tl = nfsm_dissect(u_int32_t *,
 2068                                     3 * NFSX_UNSIGNED);
 2069                         } else {
 2070                                 tl = nfsm_dissect(u_int32_t *,
 2071                                     2 * NFSX_UNSIGNED);
 2072                         }
 2073                         if (bigenough) {
 2074                                 cookie.nfsuquad[0] = *tl++;
 2075                                 if (v3)
 2076                                         cookie.nfsuquad[1] = *tl++;
 2077                         } else if (v3)
 2078                                 tl += 2;
 2079                         else
 2080                                 tl++;
 2081                         more_dirs = fxdr_unsigned(int, *tl);
 2082                 }
 2083                 /*
 2084                  * If at end of rpc data, get the eof boolean
 2085                  */
 2086                 if (!more_dirs) {
 2087                         tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2088                         more_dirs = (fxdr_unsigned(int, *tl) == 0);
 2089                 }
 2090                 m_freem(mrep);
 2091         }
 2092         /*
 2093          * Fill last record, iff any, out to a multiple of DIRBLKSIZ
 2094          * by increasing d_reclen for the last record.
 2095          */
 2096         if (blksiz > 0) {
 2097                 left = DIRBLKSIZ - blksiz;
 2098                 dp->d_reclen += left;
 2099                 uiop->uio_iov->iov_base =
 2100                     (char *)uiop->uio_iov->iov_base + left;
 2101                 uiop->uio_iov->iov_len -= left;
 2102                 uiop->uio_offset += left;
 2103                 uiop->uio_resid -= left;
 2104         }
 2105 
 2106         /*
 2107          * We are now either at the end of the directory or have filled the
 2108          * block.
 2109          */
 2110         if (bigenough)
 2111                 dnp->n_direofoffset = uiop->uio_offset;
 2112         else {
 2113                 if (uiop->uio_resid > 0)
 2114                         printf("EEK! readdirrpc resid > 0\n");
 2115                 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
 2116                 *cookiep = cookie;
 2117         }
 2118 nfsmout:
 2119         return (error);
 2120 }
 2121 
 2122 /*
 2123  * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
 2124  */
 2125 int
 2126 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
 2127 {
 2128         int len, left;
 2129         struct dirent *dp;
 2130         u_int32_t *tl;
 2131         caddr_t cp;
 2132         struct vnode *newvp;
 2133         nfsuint64 *cookiep;
 2134         caddr_t bpos, dpos, dpossav1, dpossav2;
 2135         struct mbuf *mreq, *mrep, *md, *mb, *mdsav1, *mdsav2;
 2136         struct nameidata nami, *ndp = &nami;
 2137         struct componentname *cnp = &ndp->ni_cnd;
 2138         nfsuint64 cookie;
 2139         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 2140         struct nfsnode *dnp = VTONFS(vp), *np;
 2141         nfsfh_t *fhp;
 2142         u_quad_t fileno;
 2143         int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
 2144         int attrflag, fhsize;
 2145 
 2146 #ifndef nolint
 2147         dp = NULL;
 2148 #endif
 2149 #ifndef DIAGNOSTIC
 2150         if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
 2151                 (uiop->uio_resid & (DIRBLKSIZ - 1)))
 2152                 panic("nfs readdirplusrpc bad uio");
 2153 #endif
 2154         ndp->ni_dvp = vp;
 2155         newvp = NULLVP;
 2156 
 2157         /*
 2158          * If there is no cookie, assume directory was stale.
 2159          */
 2160         cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
 2161         if (cookiep)
 2162                 cookie = *cookiep;
 2163         else
 2164                 return (NFSERR_BAD_COOKIE);
 2165         /*
 2166          * Loop around doing readdir rpc's of size nm_readdirsize
 2167          * truncated to a multiple of DIRBLKSIZ.
 2168          * The stopping criteria is EOF or buffer full.
 2169          */
 2170         while (more_dirs && bigenough) {
 2171                 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
 2172                 mreq = nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
 2173                         NFSX_FH(1) + 6 * NFSX_UNSIGNED);
 2174                 mb = mreq;
 2175                 bpos = mtod(mb, caddr_t);
 2176                 nfsm_fhtom(vp, 1);
 2177                 tl = nfsm_build(u_int32_t *, 6 * NFSX_UNSIGNED);
 2178                 *tl++ = cookie.nfsuquad[0];
 2179                 *tl++ = cookie.nfsuquad[1];
 2180                 *tl++ = dnp->n_cookieverf.nfsuquad[0];
 2181                 *tl++ = dnp->n_cookieverf.nfsuquad[1];
 2182                 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
 2183                 *tl = txdr_unsigned(nmp->nm_rsize);
 2184                 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, cred);
 2185                 nfsm_postop_attr(vp, attrflag);
 2186                 if (error) {
 2187                         m_freem(mrep);
 2188                         goto nfsmout;
 2189                 }
 2190                 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
 2191                 dnp->n_cookieverf.nfsuquad[0] = *tl++;
 2192                 dnp->n_cookieverf.nfsuquad[1] = *tl++;
 2193                 more_dirs = fxdr_unsigned(int, *tl);
 2194 
 2195                 /* loop thru the dir entries, doctoring them to 4bsd form */
 2196                 while (more_dirs && bigenough) {
 2197                         tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
 2198                         fileno = fxdr_hyper(tl);
 2199                         len = fxdr_unsigned(int, *(tl + 2));
 2200                         if (len <= 0 || len > NFS_MAXNAMLEN) {
 2201                                 error = EBADRPC;
 2202                                 m_freem(mrep);
 2203                                 goto nfsmout;
 2204                         }
 2205                         tlen = nfsm_rndup(len);
 2206                         if (tlen == len)
 2207                                 tlen += 4;      /* To ensure null termination*/
 2208                         left = DIRBLKSIZ - blksiz;
 2209                         if ((tlen + DIRHDSIZ) > left) {
 2210                                 dp->d_reclen += left;
 2211                                 uiop->uio_iov->iov_base =
 2212                                     (char *)uiop->uio_iov->iov_base + left;
 2213                                 uiop->uio_iov->iov_len -= left;
 2214                                 uiop->uio_offset += left;
 2215                                 uiop->uio_resid -= left;
 2216                                 blksiz = 0;
 2217                         }
 2218                         if ((tlen + DIRHDSIZ) > uiop->uio_resid)
 2219                                 bigenough = 0;
 2220                         if (bigenough) {
 2221                                 dp = (struct dirent *)uiop->uio_iov->iov_base;
 2222                                 dp->d_fileno = (int)fileno;
 2223                                 dp->d_namlen = len;
 2224                                 dp->d_reclen = tlen + DIRHDSIZ;
 2225                                 dp->d_type = DT_UNKNOWN;
 2226                                 blksiz += dp->d_reclen;
 2227                                 if (blksiz == DIRBLKSIZ)
 2228                                         blksiz = 0;
 2229                                 uiop->uio_offset += DIRHDSIZ;
 2230                                 uiop->uio_resid -= DIRHDSIZ;
 2231                                 uiop->uio_iov->iov_base =
 2232                                     (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
 2233                                 uiop->uio_iov->iov_len -= DIRHDSIZ;
 2234                                 cnp->cn_nameptr = uiop->uio_iov->iov_base;
 2235                                 cnp->cn_namelen = len;
 2236                                 nfsm_mtouio(uiop, len);
 2237                                 cp = uiop->uio_iov->iov_base;
 2238                                 tlen -= len;
 2239                                 *cp = '\0';
 2240                                 uiop->uio_iov->iov_base =
 2241                                     (char *)uiop->uio_iov->iov_base + tlen;
 2242                                 uiop->uio_iov->iov_len -= tlen;
 2243                                 uiop->uio_offset += tlen;
 2244                                 uiop->uio_resid -= tlen;
 2245                         } else
 2246                                 nfsm_adv(nfsm_rndup(len));
 2247                         tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
 2248                         if (bigenough) {
 2249                                 cookie.nfsuquad[0] = *tl++;
 2250                                 cookie.nfsuquad[1] = *tl++;
 2251                         } else
 2252                                 tl += 2;
 2253 
 2254                         /*
 2255                          * Since the attributes are before the file handle
 2256                          * (sigh), we must skip over the attributes and then
 2257                          * come back and get them.
 2258                          */
 2259                         attrflag = fxdr_unsigned(int, *tl);
 2260                         if (attrflag) {
 2261                             dpossav1 = dpos;
 2262                             mdsav1 = md;
 2263                             nfsm_adv(NFSX_V3FATTR);
 2264                             tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2265                             doit = fxdr_unsigned(int, *tl);
 2266                             if (doit) {
 2267                                 nfsm_getfh(fhp, fhsize, 1);
 2268                                 if (NFS_CMPFH(dnp, fhp, fhsize)) {
 2269                                     VREF(vp);
 2270                                     newvp = vp;
 2271                                     np = dnp;
 2272                                 } else {
 2273                                     error = nfs_nget(vp->v_mount, fhp,
 2274                                         fhsize, &np);
 2275                                     if (error)
 2276                                         doit = 0;
 2277                                     else
 2278                                         newvp = NFSTOV(np);
 2279                                 }
 2280                             }
 2281                             if (doit && bigenough) {
 2282                                 dpossav2 = dpos;
 2283                                 dpos = dpossav1;
 2284                                 mdsav2 = md;
 2285                                 md = mdsav1;
 2286                                 nfsm_loadattr(newvp, NULL);
 2287                                 dpos = dpossav2;
 2288                                 md = mdsav2;
 2289                                 dp->d_type =
 2290                                     IFTODT(VTTOIF(np->n_vattr.va_type));
 2291                                 ndp->ni_vp = newvp;
 2292                                 cache_enter(ndp->ni_dvp, ndp->ni_vp, cnp);
 2293                             }
 2294                         } else {
 2295                             /* Just skip over the file handle */
 2296                             tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2297                             i = fxdr_unsigned(int, *tl);
 2298                             nfsm_adv(nfsm_rndup(i));
 2299                         }
 2300                         if (newvp != NULLVP) {
 2301                             if (newvp == vp)
 2302                                 vrele(newvp);
 2303                             else
 2304                                 vput(newvp);
 2305                             newvp = NULLVP;
 2306                         }
 2307                         tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2308                         more_dirs = fxdr_unsigned(int, *tl);
 2309                 }
 2310                 /*
 2311                  * If at end of rpc data, get the eof boolean
 2312                  */
 2313                 if (!more_dirs) {
 2314                         tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2315                         more_dirs = (fxdr_unsigned(int, *tl) == 0);
 2316                 }
 2317                 m_freem(mrep);
 2318         }
 2319         /*
 2320          * Fill last record, iff any, out to a multiple of DIRBLKSIZ
 2321          * by increasing d_reclen for the last record.
 2322          */
 2323         if (blksiz > 0) {
 2324                 left = DIRBLKSIZ - blksiz;
 2325                 dp->d_reclen += left;
 2326                 uiop->uio_iov->iov_base =
 2327                     (char *)uiop->uio_iov->iov_base + left;
 2328                 uiop->uio_iov->iov_len -= left;
 2329                 uiop->uio_offset += left;
 2330                 uiop->uio_resid -= left;
 2331         }
 2332 
 2333         /*
 2334          * We are now either at the end of the directory or have filled the
 2335          * block.
 2336          */
 2337         if (bigenough)
 2338                 dnp->n_direofoffset = uiop->uio_offset;
 2339         else {
 2340                 if (uiop->uio_resid > 0)
 2341                         printf("EEK! readdirplusrpc resid > 0\n");
 2342                 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
 2343                 *cookiep = cookie;
 2344         }
 2345 nfsmout:
 2346         if (newvp != NULLVP) {
 2347                 if (newvp == vp)
 2348                         vrele(newvp);
 2349                 else
 2350                         vput(newvp);
 2351                 newvp = NULLVP;
 2352         }
 2353         return (error);
 2354 }
 2355 
 2356 /*
 2357  * Silly rename. To make the NFS filesystem that is stateless look a little
 2358  * more like the "ufs" a remove of an active vnode is translated to a rename
 2359  * to a funny looking filename that is removed by nfs_inactive on the
 2360  * nfsnode. There is the potential for another process on a different client
 2361  * to create the same funny name between the nfs_lookitup() fails and the
 2362  * nfs_rename() completes, but...
 2363  */
 2364 static int
 2365 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
 2366 {
 2367         struct sillyrename *sp;
 2368         struct nfsnode *np;
 2369         int error;
 2370         short pid;
 2371 
 2372         cache_purge(dvp);
 2373         np = VTONFS(vp);
 2374 #ifndef DIAGNOSTIC
 2375         if (vp->v_type == VDIR)
 2376                 panic("nfs: sillyrename dir");
 2377 #endif
 2378         MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
 2379                 M_NFSREQ, M_WAITOK);
 2380         sp->s_cred = crhold(cnp->cn_cred);
 2381         sp->s_dvp = dvp;
 2382         VREF(dvp);
 2383 
 2384         /* Fudge together a funny name */
 2385         pid = cnp->cn_thread->td_proc->p_pid;
 2386         sp->s_namlen = sprintf(sp->s_name, ".nfsA%04x4.4", pid);
 2387 
 2388         /* Try lookitups until we get one that isn't there */
 2389         while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
 2390                 cnp->cn_thread, NULL) == 0) {
 2391                 sp->s_name[4]++;
 2392                 if (sp->s_name[4] > 'z') {
 2393                         error = EINVAL;
 2394                         goto bad;
 2395                 }
 2396         }
 2397         error = nfs_renameit(dvp, cnp, sp);
 2398         if (error)
 2399                 goto bad;
 2400         error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
 2401                 cnp->cn_thread, &np);
 2402         np->n_sillyrename = sp;
 2403         return (0);
 2404 bad:
 2405         vrele(sp->s_dvp);
 2406         crfree(sp->s_cred);
 2407         free((caddr_t)sp, M_NFSREQ);
 2408         return (error);
 2409 }
 2410 
 2411 /*
 2412  * Look up a file name and optionally either update the file handle or
 2413  * allocate an nfsnode, depending on the value of npp.
 2414  * npp == NULL  --> just do the lookup
 2415  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
 2416  *                      handled too
 2417  * *npp != NULL --> update the file handle in the vnode
 2418  */
 2419 static int
 2420 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
 2421     struct thread *td, struct nfsnode **npp)
 2422 {
 2423         struct vnode *newvp = NULL;
 2424         struct nfsnode *np, *dnp = VTONFS(dvp);
 2425         caddr_t bpos, dpos;
 2426         int error = 0, fhlen, attrflag;
 2427         struct mbuf *mreq, *mrep, *md, *mb;
 2428         nfsfh_t *nfhp;
 2429         int v3 = NFS_ISV3(dvp);
 2430 
 2431         nfsstats.rpccnt[NFSPROC_LOOKUP]++;
 2432         mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
 2433                 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
 2434         mb = mreq;
 2435         bpos = mtod(mb, caddr_t);
 2436         nfsm_fhtom(dvp, v3);
 2437         nfsm_strtom(name, len, NFS_MAXNAMLEN);
 2438         nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
 2439         if (npp && !error) {
 2440                 nfsm_getfh(nfhp, fhlen, v3);
 2441                 if (*npp) {
 2442                     np = *npp;
 2443                     if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
 2444                         free((caddr_t)np->n_fhp, M_NFSBIGFH);
 2445                         np->n_fhp = &np->n_fh;
 2446                     } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
 2447                         np->n_fhp =(nfsfh_t *)malloc(fhlen, M_NFSBIGFH, M_WAITOK);
 2448                     bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
 2449                     np->n_fhsize = fhlen;
 2450                     newvp = NFSTOV(np);
 2451                 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
 2452                     VREF(dvp);
 2453                     newvp = dvp;
 2454                 } else {
 2455                     error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
 2456                     if (error) {
 2457                         m_freem(mrep);
 2458                         return (error);
 2459                     }
 2460                     newvp = NFSTOV(np);
 2461                 }
 2462                 if (v3) {
 2463                         nfsm_postop_attr(newvp, attrflag);
 2464                         if (!attrflag && *npp == NULL) {
 2465                                 m_freem(mrep);
 2466                                 if (newvp == dvp)
 2467                                         vrele(newvp);
 2468                                 else
 2469                                         vput(newvp);
 2470                                 return (ENOENT);
 2471                         }
 2472                 } else
 2473                         nfsm_loadattr(newvp, NULL);
 2474         }
 2475         m_freem(mrep);
 2476 nfsmout:
 2477         if (npp && *npp == NULL) {
 2478                 if (error) {
 2479                         if (newvp) {
 2480                                 if (newvp == dvp)
 2481                                         vrele(newvp);
 2482                                 else
 2483                                         vput(newvp);
 2484                         }
 2485                 } else
 2486                         *npp = np;
 2487         }
 2488         return (error);
 2489 }
 2490 
 2491 /*
 2492  * Nfs Version 3 commit rpc
 2493  */
 2494 int
 2495 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
 2496     struct thread *td)
 2497 {
 2498         u_int32_t *tl;
 2499         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 2500         caddr_t bpos, dpos;
 2501         int error = 0, wccflag = NFSV3_WCCRATTR;
 2502         struct mbuf *mreq, *mrep, *md, *mb;
 2503 
 2504         if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
 2505                 return (0);
 2506         nfsstats.rpccnt[NFSPROC_COMMIT]++;
 2507         mreq = nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
 2508         mb = mreq;
 2509         bpos = mtod(mb, caddr_t);
 2510         nfsm_fhtom(vp, 1);
 2511         tl = nfsm_build(u_int32_t *, 3 * NFSX_UNSIGNED);
 2512         txdr_hyper(offset, tl);
 2513         tl += 2;
 2514         *tl = txdr_unsigned(cnt);
 2515         nfsm_request(vp, NFSPROC_COMMIT, td, cred);
 2516         nfsm_wcc_data(vp, wccflag);
 2517         if (!error) {
 2518                 tl = nfsm_dissect(u_int32_t *, NFSX_V3WRITEVERF);
 2519                 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
 2520                         NFSX_V3WRITEVERF)) {
 2521                         bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
 2522                                 NFSX_V3WRITEVERF);
 2523                         error = NFSERR_STALEWRITEVERF;
 2524                 }
 2525         }
 2526         m_freem(mrep);
 2527 nfsmout:
 2528         return (error);
 2529 }
 2530 
 2531 /*
 2532  * Strategy routine.
 2533  * For async requests when nfsiod(s) are running, queue the request by
 2534  * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
 2535  * request.
 2536  */
 2537 static int
 2538 nfs_strategy(struct vop_strategy_args *ap)
 2539 {
 2540         struct buf *bp = ap->a_bp;
 2541         struct ucred *cr;
 2542         struct thread *td;
 2543         int error = 0;
 2544 
 2545         KASSERT(!(bp->b_flags & B_DONE), ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
 2546         KASSERT(BUF_REFCNT(bp) > 0, ("nfs_strategy: buffer %p not locked", bp));
 2547 
 2548         if (bp->b_flags & B_PHYS)
 2549                 panic("nfs physio");
 2550 
 2551         if (bp->b_flags & B_ASYNC)
 2552                 td = NULL;
 2553         else
 2554                 td = curthread; /* XXX */
 2555 
 2556         if (bp->b_iocmd == BIO_READ)
 2557                 cr = bp->b_rcred;
 2558         else
 2559                 cr = bp->b_wcred;
 2560 
 2561         /*
 2562          * If the op is asynchronous and an i/o daemon is waiting
 2563          * queue the request, wake it up and wait for completion
 2564          * otherwise just do it ourselves.
 2565          */
 2566         if ((bp->b_flags & B_ASYNC) == 0 ||
 2567                 nfs_asyncio(bp, NOCRED, td))
 2568                 error = nfs_doio(bp, cr, td);
 2569         return (error);
 2570 }
 2571 
 2572 /*
 2573  * fsync vnode op. Just call nfs_flush() with commit == 1.
 2574  */
 2575 /* ARGSUSED */
 2576 static int
 2577 nfs_fsync(struct vop_fsync_args *ap)
 2578 {
 2579 
 2580         return (nfs_flush(ap->a_vp, ap->a_cred, ap->a_waitfor, ap->a_td, 1));
 2581 }
 2582 
 2583 /*
 2584  * Flush all the blocks associated with a vnode.
 2585  *      Walk through the buffer pool and push any dirty pages
 2586  *      associated with the vnode.
 2587  */
 2588 static int
 2589 nfs_flush(struct vnode *vp, struct ucred *cred, int waitfor, struct thread *td,
 2590     int commit)
 2591 {
 2592         struct nfsnode *np = VTONFS(vp);
 2593         struct buf *bp;
 2594         int i;
 2595         struct buf *nbp;
 2596         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 2597         int s, error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
 2598         int passone = 1;
 2599         u_quad_t off, endoff, toff;
 2600         struct ucred* wcred = NULL;
 2601         struct buf **bvec = NULL;
 2602 #ifndef NFS_COMMITBVECSIZ
 2603 #define NFS_COMMITBVECSIZ       20
 2604 #endif
 2605         struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
 2606         int bvecsize = 0, bveccount;
 2607 
 2608         if (nmp->nm_flag & NFSMNT_INT)
 2609                 slpflag = PCATCH;
 2610         if (!commit)
 2611                 passone = 0;
 2612         /*
 2613          * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
 2614          * server, but nas not been committed to stable storage on the server
 2615          * yet. On the first pass, the byte range is worked out and the commit
 2616          * rpc is done. On the second pass, nfs_writebp() is called to do the
 2617          * job.
 2618          */
 2619 again:
 2620         off = (u_quad_t)-1;
 2621         endoff = 0;
 2622         bvecpos = 0;
 2623         if (NFS_ISV3(vp) && commit) {
 2624                 s = splbio();
 2625                 /*
 2626                  * Count up how many buffers waiting for a commit.
 2627                  */
 2628                 bveccount = 0;
 2629                 VI_LOCK(vp);
 2630                 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
 2631                         nbp = TAILQ_NEXT(bp, b_vnbufs);
 2632                         if (BUF_REFCNT(bp) == 0 &&
 2633                             (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
 2634                                 == (B_DELWRI | B_NEEDCOMMIT))
 2635                                 bveccount++;
 2636                 }
 2637                 /*
 2638                  * Allocate space to remember the list of bufs to commit.  It is
 2639                  * important to use M_NOWAIT here to avoid a race with nfs_write.
 2640                  * If we can't get memory (for whatever reason), we will end up
 2641                  * committing the buffers one-by-one in the loop below.
 2642                  */
 2643                 if (bvec != NULL && bvec != bvec_on_stack)
 2644                         free(bvec, M_TEMP);
 2645                 if (bveccount > NFS_COMMITBVECSIZ) {
 2646                         bvec = (struct buf **)
 2647                                 malloc(bveccount * sizeof(struct buf *),
 2648                                        M_TEMP, M_NOWAIT);
 2649                         if (bvec == NULL) {
 2650                                 bvec = bvec_on_stack;
 2651                                 bvecsize = NFS_COMMITBVECSIZ;
 2652                         } else
 2653                                 bvecsize = bveccount;
 2654                 } else {
 2655                         bvec = bvec_on_stack;
 2656                         bvecsize = NFS_COMMITBVECSIZ;
 2657                 }
 2658                 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
 2659                         if (bvecpos >= bvecsize)
 2660                                 break;
 2661                         VI_UNLOCK(vp);
 2662                         if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
 2663                             (B_DELWRI | B_NEEDCOMMIT) ||
 2664                             BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
 2665                                 VI_LOCK(vp);
 2666                                 nbp = TAILQ_NEXT(bp, b_vnbufs);
 2667                                 continue;
 2668                         }
 2669                         bremfree(bp);
 2670                         /*
 2671                          * Work out if all buffers are using the same cred
 2672                          * so we can deal with them all with one commit.
 2673                          *
 2674                          * NOTE: we are not clearing B_DONE here, so we have
 2675                          * to do it later on in this routine if we intend to
 2676                          * initiate I/O on the bp.
 2677                          *
 2678                          * Note: to avoid loopback deadlocks, we do not
 2679                          * assign b_runningbufspace.
 2680                          */
 2681                         if (wcred == NULL)
 2682                                 wcred = bp->b_wcred;
 2683                         else if (wcred != bp->b_wcred)
 2684                                 wcred = NOCRED;
 2685                         bp->b_flags |= B_WRITEINPROG;
 2686                         vfs_busy_pages(bp, 1);
 2687 
 2688                         VI_LOCK(vp);
 2689                         /*
 2690                          * bp is protected by being locked, but nbp is not
 2691                          * and vfs_busy_pages() may sleep.  We have to
 2692                          * recalculate nbp.
 2693                          */
 2694                         nbp = TAILQ_NEXT(bp, b_vnbufs);
 2695 
 2696                         /*
 2697                          * A list of these buffers is kept so that the
 2698                          * second loop knows which buffers have actually
 2699                          * been committed. This is necessary, since there
 2700                          * may be a race between the commit rpc and new
 2701                          * uncommitted writes on the file.
 2702                          */
 2703                         bvec[bvecpos++] = bp;
 2704                         toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
 2705                                 bp->b_dirtyoff;
 2706                         if (toff < off)
 2707                                 off = toff;
 2708                         toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
 2709                         if (toff > endoff)
 2710                                 endoff = toff;
 2711                 }
 2712                 splx(s);
 2713                 VI_UNLOCK(vp);
 2714         }
 2715         if (bvecpos > 0) {
 2716                 /*
 2717                  * Commit data on the server, as required.
 2718                  * If all bufs are using the same wcred, then use that with
 2719                  * one call for all of them, otherwise commit each one
 2720                  * separately.
 2721                  */
 2722                 if (wcred != NOCRED)
 2723                         retv = nfs_commit(vp, off, (int)(endoff - off),
 2724                                           wcred, td);
 2725                 else {
 2726                         retv = 0;
 2727                         for (i = 0; i < bvecpos; i++) {
 2728                                 off_t off, size;
 2729                                 bp = bvec[i];
 2730                                 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
 2731                                         bp->b_dirtyoff;
 2732                                 size = (u_quad_t)(bp->b_dirtyend
 2733                                                   - bp->b_dirtyoff);
 2734                                 retv = nfs_commit(vp, off, (int)size,
 2735                                                   bp->b_wcred, td);
 2736                                 if (retv) break;
 2737                         }
 2738                 }
 2739 
 2740                 if (retv == NFSERR_STALEWRITEVERF)
 2741                         nfs_clearcommit(vp->v_mount);
 2742 
 2743                 /*
 2744                  * Now, either mark the blocks I/O done or mark the
 2745                  * blocks dirty, depending on whether the commit
 2746                  * succeeded.
 2747                  */
 2748                 for (i = 0; i < bvecpos; i++) {
 2749                         bp = bvec[i];
 2750                         bp->b_flags &= ~(B_NEEDCOMMIT | B_WRITEINPROG | B_CLUSTEROK);
 2751                         if (retv) {
 2752                                 /*
 2753                                  * Error, leave B_DELWRI intact
 2754                                  */
 2755                                 vfs_unbusy_pages(bp);
 2756                                 brelse(bp);
 2757                         } else {
 2758                                 /*
 2759                                  * Success, remove B_DELWRI ( bundirty() ).
 2760                                  *
 2761                                  * b_dirtyoff/b_dirtyend seem to be NFS
 2762                                  * specific.  We should probably move that
 2763                                  * into bundirty(). XXX
 2764                                  */
 2765                                 s = splbio();
 2766                                 VI_LOCK(vp);
 2767                                 vp->v_numoutput++;
 2768                                 VI_UNLOCK(vp);
 2769                                 bp->b_flags |= B_ASYNC;
 2770                                 bundirty(bp);
 2771                                 bp->b_flags &= ~B_DONE;
 2772                                 bp->b_ioflags &= ~BIO_ERROR;
 2773                                 bp->b_dirtyoff = bp->b_dirtyend = 0;
 2774                                 splx(s);
 2775                                 bufdone(bp);
 2776                         }
 2777                 }
 2778         }
 2779 
 2780         /*
 2781          * Start/do any write(s) that are required.
 2782          */
 2783 loop:
 2784         s = splbio();
 2785         VI_LOCK(vp);
 2786         for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
 2787                 nbp = TAILQ_NEXT(bp, b_vnbufs);
 2788                 VI_UNLOCK(vp);
 2789                 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
 2790                         if (waitfor != MNT_WAIT || passone) {
 2791                                 VI_LOCK(vp);
 2792                                 continue;
 2793                         }
 2794                         error = BUF_TIMELOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL,
 2795                             "nfsfsync", slpflag, slptimeo);
 2796                         splx(s);
 2797                         if (error == 0)
 2798                                 panic("nfs_fsync: inconsistent lock");
 2799                         if (error == ENOLCK)
 2800                                 goto loop;
 2801                         if (nfs_sigintr(nmp, NULL, td)) {
 2802                                 error = EINTR;
 2803                                 goto done;
 2804                         }
 2805                         if (slpflag == PCATCH) {
 2806                                 slpflag = 0;
 2807                                 slptimeo = 2 * hz;
 2808                         }
 2809                         goto loop;
 2810                 }
 2811                 if ((bp->b_flags & B_DELWRI) == 0)
 2812                         panic("nfs_fsync: not dirty");
 2813                 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
 2814                         BUF_UNLOCK(bp);
 2815                         VI_LOCK(vp);
 2816                         continue;
 2817                 }
 2818                 bremfree(bp);
 2819                 if (passone || !commit)
 2820                     bp->b_flags |= B_ASYNC;
 2821                 else
 2822                     bp->b_flags |= B_ASYNC | B_WRITEINPROG;
 2823                 splx(s);
 2824                 BUF_WRITE(bp);
 2825                 goto loop;
 2826         }
 2827         splx(s);
 2828         if (passone) {
 2829                 passone = 0;
 2830                 VI_UNLOCK(vp);
 2831                 goto again;
 2832         }
 2833         if (waitfor == MNT_WAIT) {
 2834                 while (vp->v_numoutput) {
 2835                         vp->v_iflag |= VI_BWAIT;
 2836                         error = msleep((caddr_t)&vp->v_numoutput, VI_MTX(vp),
 2837                                 slpflag | (PRIBIO + 1), "nfsfsync", slptimeo);
 2838                         if (error) {
 2839                             if (nfs_sigintr(nmp, NULL, td)) {
 2840                                 VI_UNLOCK(vp);
 2841                                 error = EINTR;
 2842                                 goto done;
 2843                             }
 2844                             if (slpflag == PCATCH) {
 2845                                 slpflag = 0;
 2846                                 slptimeo = 2 * hz;
 2847                             }
 2848                         }
 2849                 }
 2850                 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) && commit) {
 2851                         VI_UNLOCK(vp);
 2852                         goto loop;
 2853                 }
 2854         }
 2855         VI_UNLOCK(vp);
 2856         if (np->n_flag & NWRITEERR) {
 2857                 error = np->n_error;
 2858                 np->n_flag &= ~NWRITEERR;
 2859         }
 2860 done:
 2861         if (bvec != NULL && bvec != bvec_on_stack)
 2862                 free(bvec, M_TEMP);
 2863         return (error);
 2864 }
 2865 
 2866 /*
 2867  * NFS advisory byte-level locks.
 2868  */
 2869 static int
 2870 nfs_advlock(struct vop_advlock_args *ap)
 2871 {
 2872 
 2873         if ((VFSTONFS(ap->a_vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
 2874                 struct nfsnode *np = VTONFS(ap->a_vp);
 2875 
 2876                 return (lf_advlock(ap, &(np->n_lockf), np->n_size));
 2877         }
 2878         return (nfs_dolock(ap));
 2879 }
 2880 
 2881 /*
 2882  * Print out the contents of an nfsnode.
 2883  */
 2884 static int
 2885 nfs_print(struct vop_print_args *ap)
 2886 {
 2887         struct vnode *vp = ap->a_vp;
 2888         struct nfsnode *np = VTONFS(vp);
 2889 
 2890         printf("tag %s fileid %ld fsid 0x%x",
 2891                vp->v_tag, np->n_vattr.va_fileid, np->n_vattr.va_fsid);
 2892         if (vp->v_type == VFIFO)
 2893                 fifo_printinfo(vp);
 2894         printf("\n");
 2895         return (0);
 2896 }
 2897 
 2898 /*
 2899  * This is the "real" nfs::bwrite(struct buf*).
 2900  * B_WRITEINPROG isn't set unless the force flag is one and it
 2901  * handles the B_NEEDCOMMIT flag.
 2902  * We set B_CACHE if this is a VMIO buffer.
 2903  */
 2904 int
 2905 nfs_writebp(struct buf *bp, int force, struct thread *td)
 2906 {
 2907         int s;
 2908         int oldflags = bp->b_flags;
 2909 #if 0
 2910         int retv = 1;
 2911         off_t off;
 2912 #endif
 2913 
 2914         if (BUF_REFCNT(bp) == 0)
 2915                 panic("bwrite: buffer is not locked???");
 2916 
 2917         if (bp->b_flags & B_INVAL) {
 2918                 brelse(bp);
 2919                 return(0);
 2920         }
 2921 
 2922         bp->b_flags |= B_CACHE;
 2923 
 2924         /*
 2925          * Undirty the bp.  We will redirty it later if the I/O fails.
 2926          */
 2927 
 2928         s = splbio();
 2929         bundirty(bp);
 2930         bp->b_flags &= ~B_DONE;
 2931         bp->b_ioflags &= ~BIO_ERROR;
 2932         bp->b_iocmd = BIO_WRITE;
 2933 
 2934         VI_LOCK(bp->b_vp);
 2935         bp->b_vp->v_numoutput++;
 2936         VI_UNLOCK(bp->b_vp);
 2937         curthread->td_proc->p_stats->p_ru.ru_oublock++;
 2938         splx(s);
 2939 
 2940         /*
 2941          * Note: to avoid loopback deadlocks, we do not
 2942          * assign b_runningbufspace.
 2943          */
 2944         vfs_busy_pages(bp, 1);
 2945 
 2946         if (force)
 2947                 bp->b_flags |= B_WRITEINPROG;
 2948         BUF_KERNPROC(bp);
 2949         BUF_STRATEGY(bp);
 2950 
 2951         if( (oldflags & B_ASYNC) == 0) {
 2952                 int rtval = bufwait(bp);
 2953 
 2954                 if (oldflags & B_DELWRI) {
 2955                         s = splbio();
 2956                         reassignbuf(bp, bp->b_vp);
 2957                         splx(s);
 2958                 }
 2959 
 2960                 brelse(bp);
 2961                 return (rtval);
 2962         }
 2963 
 2964         return (0);
 2965 }
 2966 
 2967 /*
 2968  * nfs special file access vnode op.
 2969  * Essentially just get vattr and then imitate iaccess() since the device is
 2970  * local to the client.
 2971  */
 2972 static int
 2973 nfsspec_access(struct vop_access_args *ap)
 2974 {
 2975         struct vattr *vap;
 2976         gid_t *gp;
 2977         struct ucred *cred = ap->a_cred;
 2978         struct vnode *vp = ap->a_vp;
 2979         mode_t mode = ap->a_mode;
 2980         struct vattr vattr;
 2981         int i;
 2982         int error;
 2983 
 2984         /*
 2985          * Disallow write attempts on filesystems mounted read-only;
 2986          * unless the file is a socket, fifo, or a block or character
 2987          * device resident on the filesystem.
 2988          */
 2989         if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
 2990                 switch (vp->v_type) {
 2991                 case VREG:
 2992                 case VDIR:
 2993                 case VLNK:
 2994                         return (EROFS);
 2995                 default:
 2996                         break;
 2997                 }
 2998         }
 2999         /*
 3000          * If you're the super-user,
 3001          * you always get access.
 3002          */
 3003         if (cred->cr_uid == 0)
 3004                 return (0);
 3005         vap = &vattr;
 3006         error = VOP_GETATTR(vp, vap, cred, ap->a_td);
 3007         if (error)
 3008                 return (error);
 3009         /*
 3010          * Access check is based on only one of owner, group, public.
 3011          * If not owner, then check group. If not a member of the
 3012          * group, then check public access.
 3013          */
 3014         if (cred->cr_uid != vap->va_uid) {
 3015                 mode >>= 3;
 3016                 gp = cred->cr_groups;
 3017                 for (i = 0; i < cred->cr_ngroups; i++, gp++)
 3018                         if (vap->va_gid == *gp)
 3019                                 goto found;
 3020                 mode >>= 3;
 3021 found:
 3022                 ;
 3023         }
 3024         error = (vap->va_mode & mode) == mode ? 0 : EACCES;
 3025         return (error);
 3026 }
 3027 
 3028 /*
 3029  * Read wrapper for special devices.
 3030  */
 3031 static int
 3032 nfsspec_read(struct vop_read_args *ap)
 3033 {
 3034         struct nfsnode *np = VTONFS(ap->a_vp);
 3035 
 3036         /*
 3037          * Set access flag.
 3038          */
 3039         np->n_flag |= NACC;
 3040         getnanotime(&np->n_atim);
 3041         return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap));
 3042 }
 3043 
 3044 /*
 3045  * Write wrapper for special devices.
 3046  */
 3047 static int
 3048 nfsspec_write(struct vop_write_args *ap)
 3049 {
 3050         struct nfsnode *np = VTONFS(ap->a_vp);
 3051 
 3052         /*
 3053          * Set update flag.
 3054          */
 3055         np->n_flag |= NUPD;
 3056         getnanotime(&np->n_mtim);
 3057         return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap));
 3058 }
 3059 
 3060 /*
 3061  * Close wrapper for special devices.
 3062  *
 3063  * Update the times on the nfsnode then do device close.
 3064  */
 3065 static int
 3066 nfsspec_close(struct vop_close_args *ap)
 3067 {
 3068         struct vnode *vp = ap->a_vp;
 3069         struct nfsnode *np = VTONFS(vp);
 3070         struct vattr vattr;
 3071 
 3072         if (np->n_flag & (NACC | NUPD)) {
 3073                 np->n_flag |= NCHG;
 3074                 if (vrefcnt(vp) == 1 &&
 3075                     (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
 3076                         VATTR_NULL(&vattr);
 3077                         if (np->n_flag & NACC)
 3078                                 vattr.va_atime = np->n_atim;
 3079                         if (np->n_flag & NUPD)
 3080                                 vattr.va_mtime = np->n_mtim;
 3081                         (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_td);
 3082                 }
 3083         }
 3084         return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap));
 3085 }
 3086 
 3087 /*
 3088  * Read wrapper for fifos.
 3089  */
 3090 static int
 3091 nfsfifo_read(struct vop_read_args *ap)
 3092 {
 3093         struct nfsnode *np = VTONFS(ap->a_vp);
 3094 
 3095         /*
 3096          * Set access flag.
 3097          */
 3098         np->n_flag |= NACC;
 3099         getnanotime(&np->n_atim);
 3100         return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap));
 3101 }
 3102 
 3103 /*
 3104  * Write wrapper for fifos.
 3105  */
 3106 static int
 3107 nfsfifo_write(struct vop_write_args *ap)
 3108 {
 3109         struct nfsnode *np = VTONFS(ap->a_vp);
 3110 
 3111         /*
 3112          * Set update flag.
 3113          */
 3114         np->n_flag |= NUPD;
 3115         getnanotime(&np->n_mtim);
 3116         return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap));
 3117 }
 3118 
 3119 /*
 3120  * Close wrapper for fifos.
 3121  *
 3122  * Update the times on the nfsnode then do fifo close.
 3123  */
 3124 static int
 3125 nfsfifo_close(struct vop_close_args *ap)
 3126 {
 3127         struct vnode *vp = ap->a_vp;
 3128         struct nfsnode *np = VTONFS(vp);
 3129         struct vattr vattr;
 3130         struct timespec ts;
 3131 
 3132         if (np->n_flag & (NACC | NUPD)) {
 3133                 getnanotime(&ts);
 3134                 if (np->n_flag & NACC)
 3135                         np->n_atim = ts;
 3136                 if (np->n_flag & NUPD)
 3137                         np->n_mtim = ts;
 3138                 np->n_flag |= NCHG;
 3139                 if (vrefcnt(vp) == 1 &&
 3140                     (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
 3141                         VATTR_NULL(&vattr);
 3142                         if (np->n_flag & NACC)
 3143                                 vattr.va_atime = np->n_atim;
 3144                         if (np->n_flag & NUPD)
 3145                                 vattr.va_mtime = np->n_mtim;
 3146                         (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_td);
 3147                 }
 3148         }
 3149         return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap));
 3150 }
 3151 

Cache object: 0066fb20d516be9f2ca9aefa75c67504


[ 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.