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/fs/nfsclient/nfs_clvnops.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  * 4. Neither the name of the University nor the names of its contributors
   17  *    may be used to endorse or promote products derived from this software
   18  *    without specific prior written permission.
   19  *
   20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   30  * SUCH DAMAGE.
   31  *
   32  *      from nfs_vnops.c        8.16 (Berkeley) 5/27/95
   33  */
   34 
   35 #include <sys/cdefs.h>
   36 __FBSDID("$FreeBSD: releng/10.1/sys/fs/nfsclient/nfs_clvnops.c 269283 2014-07-30 03:56:17Z kib $");
   37 
   38 /*
   39  * vnode op calls for Sun NFS version 2, 3 and 4
   40  */
   41 
   42 #include "opt_kdtrace.h"
   43 #include "opt_inet.h"
   44 
   45 #include <sys/param.h>
   46 #include <sys/kernel.h>
   47 #include <sys/systm.h>
   48 #include <sys/resourcevar.h>
   49 #include <sys/proc.h>
   50 #include <sys/mount.h>
   51 #include <sys/bio.h>
   52 #include <sys/buf.h>
   53 #include <sys/jail.h>
   54 #include <sys/malloc.h>
   55 #include <sys/mbuf.h>
   56 #include <sys/namei.h>
   57 #include <sys/socket.h>
   58 #include <sys/vnode.h>
   59 #include <sys/dirent.h>
   60 #include <sys/fcntl.h>
   61 #include <sys/lockf.h>
   62 #include <sys/stat.h>
   63 #include <sys/sysctl.h>
   64 #include <sys/signalvar.h>
   65 
   66 #include <vm/vm.h>
   67 #include <vm/vm_extern.h>
   68 #include <vm/vm_object.h>
   69 
   70 #include <fs/nfs/nfsport.h>
   71 #include <fs/nfsclient/nfsnode.h>
   72 #include <fs/nfsclient/nfsmount.h>
   73 #include <fs/nfsclient/nfs.h>
   74 #include <fs/nfsclient/nfs_kdtrace.h>
   75 
   76 #include <net/if.h>
   77 #include <netinet/in.h>
   78 #include <netinet/in_var.h>
   79 
   80 #include <nfs/nfs_lock.h>
   81 
   82 #ifdef KDTRACE_HOOKS
   83 #include <sys/dtrace_bsd.h>
   84 
   85 dtrace_nfsclient_accesscache_flush_probe_func_t
   86                 dtrace_nfscl_accesscache_flush_done_probe;
   87 uint32_t        nfscl_accesscache_flush_done_id;
   88 
   89 dtrace_nfsclient_accesscache_get_probe_func_t
   90                 dtrace_nfscl_accesscache_get_hit_probe,
   91                 dtrace_nfscl_accesscache_get_miss_probe;
   92 uint32_t        nfscl_accesscache_get_hit_id;
   93 uint32_t        nfscl_accesscache_get_miss_id;
   94 
   95 dtrace_nfsclient_accesscache_load_probe_func_t
   96                 dtrace_nfscl_accesscache_load_done_probe;
   97 uint32_t        nfscl_accesscache_load_done_id;
   98 #endif /* !KDTRACE_HOOKS */
   99 
  100 /* Defs */
  101 #define TRUE    1
  102 #define FALSE   0
  103 
  104 extern struct nfsstats newnfsstats;
  105 extern int nfsrv_useacl;
  106 extern int nfscl_debuglevel;
  107 MALLOC_DECLARE(M_NEWNFSREQ);
  108 
  109 /*
  110  * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
  111  * calls are not in getblk() and brelse() so that they would not be necessary
  112  * here.
  113  */
  114 #ifndef B_VMIO
  115 #define vfs_busy_pages(bp, f)
  116 #endif
  117 
  118 static vop_read_t       nfsfifo_read;
  119 static vop_write_t      nfsfifo_write;
  120 static vop_close_t      nfsfifo_close;
  121 static int      nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
  122                     struct thread *);
  123 static vop_lookup_t     nfs_lookup;
  124 static vop_create_t     nfs_create;
  125 static vop_mknod_t      nfs_mknod;
  126 static vop_open_t       nfs_open;
  127 static vop_pathconf_t   nfs_pathconf;
  128 static vop_close_t      nfs_close;
  129 static vop_access_t     nfs_access;
  130 static vop_getattr_t    nfs_getattr;
  131 static vop_setattr_t    nfs_setattr;
  132 static vop_read_t       nfs_read;
  133 static vop_fsync_t      nfs_fsync;
  134 static vop_remove_t     nfs_remove;
  135 static vop_link_t       nfs_link;
  136 static vop_rename_t     nfs_rename;
  137 static vop_mkdir_t      nfs_mkdir;
  138 static vop_rmdir_t      nfs_rmdir;
  139 static vop_symlink_t    nfs_symlink;
  140 static vop_readdir_t    nfs_readdir;
  141 static vop_strategy_t   nfs_strategy;
  142 static vop_lock1_t      nfs_lock1;
  143 static  int     nfs_lookitup(struct vnode *, char *, int,
  144                     struct ucred *, struct thread *, struct nfsnode **);
  145 static  int     nfs_sillyrename(struct vnode *, struct vnode *,
  146                     struct componentname *);
  147 static vop_access_t     nfsspec_access;
  148 static vop_readlink_t   nfs_readlink;
  149 static vop_print_t      nfs_print;
  150 static vop_advlock_t    nfs_advlock;
  151 static vop_advlockasync_t nfs_advlockasync;
  152 static vop_getacl_t nfs_getacl;
  153 static vop_setacl_t nfs_setacl;
  154 
  155 /*
  156  * Global vfs data structures for nfs
  157  */
  158 struct vop_vector newnfs_vnodeops = {
  159         .vop_default =          &default_vnodeops,
  160         .vop_access =           nfs_access,
  161         .vop_advlock =          nfs_advlock,
  162         .vop_advlockasync =     nfs_advlockasync,
  163         .vop_close =            nfs_close,
  164         .vop_create =           nfs_create,
  165         .vop_fsync =            nfs_fsync,
  166         .vop_getattr =          nfs_getattr,
  167         .vop_getpages =         ncl_getpages,
  168         .vop_putpages =         ncl_putpages,
  169         .vop_inactive =         ncl_inactive,
  170         .vop_link =             nfs_link,
  171         .vop_lock1 =            nfs_lock1,
  172         .vop_lookup =           nfs_lookup,
  173         .vop_mkdir =            nfs_mkdir,
  174         .vop_mknod =            nfs_mknod,
  175         .vop_open =             nfs_open,
  176         .vop_pathconf =         nfs_pathconf,
  177         .vop_print =            nfs_print,
  178         .vop_read =             nfs_read,
  179         .vop_readdir =          nfs_readdir,
  180         .vop_readlink =         nfs_readlink,
  181         .vop_reclaim =          ncl_reclaim,
  182         .vop_remove =           nfs_remove,
  183         .vop_rename =           nfs_rename,
  184         .vop_rmdir =            nfs_rmdir,
  185         .vop_setattr =          nfs_setattr,
  186         .vop_strategy =         nfs_strategy,
  187         .vop_symlink =          nfs_symlink,
  188         .vop_write =            ncl_write,
  189         .vop_getacl =           nfs_getacl,
  190         .vop_setacl =           nfs_setacl,
  191 };
  192 
  193 struct vop_vector newnfs_fifoops = {
  194         .vop_default =          &fifo_specops,
  195         .vop_access =           nfsspec_access,
  196         .vop_close =            nfsfifo_close,
  197         .vop_fsync =            nfs_fsync,
  198         .vop_getattr =          nfs_getattr,
  199         .vop_inactive =         ncl_inactive,
  200         .vop_print =            nfs_print,
  201         .vop_read =             nfsfifo_read,
  202         .vop_reclaim =          ncl_reclaim,
  203         .vop_setattr =          nfs_setattr,
  204         .vop_write =            nfsfifo_write,
  205 };
  206 
  207 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
  208     struct componentname *cnp, struct vattr *vap);
  209 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
  210     int namelen, struct ucred *cred, struct thread *td);
  211 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp,
  212     char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp,
  213     char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td);
  214 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp,
  215     struct componentname *scnp, struct sillyrename *sp);
  216 
  217 /*
  218  * Global variables
  219  */
  220 #define DIRHDSIZ        (sizeof (struct dirent) - (MAXNAMLEN + 1))
  221 
  222 SYSCTL_DECL(_vfs_nfs);
  223 
  224 static int      nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
  225 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
  226            &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
  227 
  228 static int      nfs_prime_access_cache = 0;
  229 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
  230            &nfs_prime_access_cache, 0,
  231            "Prime NFS ACCESS cache when fetching attributes");
  232 
  233 static int      newnfs_commit_on_close = 0;
  234 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW,
  235     &newnfs_commit_on_close, 0, "write+commit on close, else only write");
  236 
  237 static int      nfs_clean_pages_on_close = 1;
  238 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
  239            &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
  240 
  241 int newnfs_directio_enable = 0;
  242 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
  243            &newnfs_directio_enable, 0, "Enable NFS directio");
  244 
  245 int nfs_keep_dirty_on_error;
  246 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW,
  247     &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned");
  248 
  249 /*
  250  * This sysctl allows other processes to mmap a file that has been opened
  251  * O_DIRECT by a process.  In general, having processes mmap the file while
  252  * Direct IO is in progress can lead to Data Inconsistencies.  But, we allow
  253  * this by default to prevent DoS attacks - to prevent a malicious user from
  254  * opening up files O_DIRECT preventing other users from mmap'ing these
  255  * files.  "Protected" environments where stricter consistency guarantees are
  256  * required can disable this knob.  The process that opened the file O_DIRECT
  257  * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
  258  * meaningful.
  259  */
  260 int newnfs_directio_allow_mmap = 1;
  261 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
  262            &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
  263 
  264 #if 0
  265 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
  266            &newnfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
  267 
  268 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
  269            &newnfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
  270 #endif
  271 
  272 #define NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY                \
  273                          | NFSACCESS_EXTEND | NFSACCESS_EXECUTE \
  274                          | NFSACCESS_DELETE | NFSACCESS_LOOKUP)
  275 
  276 /*
  277  * SMP Locking Note :
  278  * The list of locks after the description of the lock is the ordering
  279  * of other locks acquired with the lock held.
  280  * np->n_mtx : Protects the fields in the nfsnode.
  281        VM Object Lock
  282        VI_MTX (acquired indirectly)
  283  * nmp->nm_mtx : Protects the fields in the nfsmount.
  284        rep->r_mtx
  285  * ncl_iod_mutex : Global lock, protects shared nfsiod state.
  286  * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
  287        nmp->nm_mtx
  288        rep->r_mtx
  289  * rep->r_mtx : Protects the fields in an nfsreq.
  290  */
  291 
  292 static int
  293 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td,
  294     struct ucred *cred, u_int32_t *retmode)
  295 {
  296         int error = 0, attrflag, i, lrupos;
  297         u_int32_t rmode;
  298         struct nfsnode *np = VTONFS(vp);
  299         struct nfsvattr nfsva;
  300 
  301         error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag,
  302             &rmode, NULL);
  303         if (attrflag)
  304                 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
  305         if (!error) {
  306                 lrupos = 0;
  307                 mtx_lock(&np->n_mtx);
  308                 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
  309                         if (np->n_accesscache[i].uid == cred->cr_uid) {
  310                                 np->n_accesscache[i].mode = rmode;
  311                                 np->n_accesscache[i].stamp = time_second;
  312                                 break;
  313                         }
  314                         if (i > 0 && np->n_accesscache[i].stamp <
  315                             np->n_accesscache[lrupos].stamp)
  316                                 lrupos = i;
  317                 }
  318                 if (i == NFS_ACCESSCACHESIZE) {
  319                         np->n_accesscache[lrupos].uid = cred->cr_uid;
  320                         np->n_accesscache[lrupos].mode = rmode;
  321                         np->n_accesscache[lrupos].stamp = time_second;
  322                 }
  323                 mtx_unlock(&np->n_mtx);
  324                 if (retmode != NULL)
  325                         *retmode = rmode;
  326                 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0);
  327         } else if (NFS_ISV4(vp)) {
  328                 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
  329         }
  330 #ifdef KDTRACE_HOOKS
  331         if (error != 0)
  332                 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0,
  333                     error);
  334 #endif
  335         return (error);
  336 }
  337 
  338 /*
  339  * nfs access vnode op.
  340  * For nfs version 2, just return ok. File accesses may fail later.
  341  * For nfs version 3, use the access rpc to check accessibility. If file modes
  342  * are changed on the server, accesses might still fail later.
  343  */
  344 static int
  345 nfs_access(struct vop_access_args *ap)
  346 {
  347         struct vnode *vp = ap->a_vp;
  348         int error = 0, i, gotahit;
  349         u_int32_t mode, wmode, rmode;
  350         int v34 = NFS_ISV34(vp);
  351         struct nfsnode *np = VTONFS(vp);
  352 
  353         /*
  354          * Disallow write attempts on filesystems mounted read-only;
  355          * unless the file is a socket, fifo, or a block or character
  356          * device resident on the filesystem.
  357          */
  358         if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS |
  359             VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL |
  360             VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) {
  361                 switch (vp->v_type) {
  362                 case VREG:
  363                 case VDIR:
  364                 case VLNK:
  365                         return (EROFS);
  366                 default:
  367                         break;
  368                 }
  369         }
  370         /*
  371          * For nfs v3 or v4, check to see if we have done this recently, and if
  372          * so return our cached result instead of making an ACCESS call.
  373          * If not, do an access rpc, otherwise you are stuck emulating
  374          * ufs_access() locally using the vattr. This may not be correct,
  375          * since the server may apply other access criteria such as
  376          * client uid-->server uid mapping that we do not know about.
  377          */
  378         if (v34) {
  379                 if (ap->a_accmode & VREAD)
  380                         mode = NFSACCESS_READ;
  381                 else
  382                         mode = 0;
  383                 if (vp->v_type != VDIR) {
  384                         if (ap->a_accmode & VWRITE)
  385                                 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
  386                         if (ap->a_accmode & VAPPEND)
  387                                 mode |= NFSACCESS_EXTEND;
  388                         if (ap->a_accmode & VEXEC)
  389                                 mode |= NFSACCESS_EXECUTE;
  390                         if (ap->a_accmode & VDELETE)
  391                                 mode |= NFSACCESS_DELETE;
  392                 } else {
  393                         if (ap->a_accmode & VWRITE)
  394                                 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
  395                         if (ap->a_accmode & VAPPEND)
  396                                 mode |= NFSACCESS_EXTEND;
  397                         if (ap->a_accmode & VEXEC)
  398                                 mode |= NFSACCESS_LOOKUP;
  399                         if (ap->a_accmode & VDELETE)
  400                                 mode |= NFSACCESS_DELETE;
  401                         if (ap->a_accmode & VDELETE_CHILD)
  402                                 mode |= NFSACCESS_MODIFY;
  403                 }
  404                 /* XXX safety belt, only make blanket request if caching */
  405                 if (nfsaccess_cache_timeout > 0) {
  406                         wmode = NFSACCESS_READ | NFSACCESS_MODIFY |
  407                                 NFSACCESS_EXTEND | NFSACCESS_EXECUTE |
  408                                 NFSACCESS_DELETE | NFSACCESS_LOOKUP;
  409                 } else {
  410                         wmode = mode;
  411                 }
  412 
  413                 /*
  414                  * Does our cached result allow us to give a definite yes to
  415                  * this request?
  416                  */
  417                 gotahit = 0;
  418                 mtx_lock(&np->n_mtx);
  419                 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
  420                         if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
  421                             if (time_second < (np->n_accesscache[i].stamp
  422                                 + nfsaccess_cache_timeout) &&
  423                                 (np->n_accesscache[i].mode & mode) == mode) {
  424                                 NFSINCRGLOBAL(newnfsstats.accesscache_hits);
  425                                 gotahit = 1;
  426                             }
  427                             break;
  428                         }
  429                 }
  430                 mtx_unlock(&np->n_mtx);
  431 #ifdef KDTRACE_HOOKS
  432                 if (gotahit != 0)
  433                         KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp,
  434                             ap->a_cred->cr_uid, mode);
  435                 else
  436                         KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp,
  437                             ap->a_cred->cr_uid, mode);
  438 #endif
  439                 if (gotahit == 0) {
  440                         /*
  441                          * Either a no, or a don't know.  Go to the wire.
  442                          */
  443                         NFSINCRGLOBAL(newnfsstats.accesscache_misses);
  444                         error = nfs34_access_otw(vp, wmode, ap->a_td,
  445                             ap->a_cred, &rmode);
  446                         if (!error &&
  447                             (rmode & mode) != mode)
  448                                 error = EACCES;
  449                 }
  450                 return (error);
  451         } else {
  452                 if ((error = nfsspec_access(ap)) != 0) {
  453                         return (error);
  454                 }
  455                 /*
  456                  * Attempt to prevent a mapped root from accessing a file
  457                  * which it shouldn't.  We try to read a byte from the file
  458                  * if the user is root and the file is not zero length.
  459                  * After calling nfsspec_access, we should have the correct
  460                  * file size cached.
  461                  */
  462                 mtx_lock(&np->n_mtx);
  463                 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
  464                     && VTONFS(vp)->n_size > 0) {
  465                         struct iovec aiov;
  466                         struct uio auio;
  467                         char buf[1];
  468 
  469                         mtx_unlock(&np->n_mtx);
  470                         aiov.iov_base = buf;
  471                         aiov.iov_len = 1;
  472                         auio.uio_iov = &aiov;
  473                         auio.uio_iovcnt = 1;
  474                         auio.uio_offset = 0;
  475                         auio.uio_resid = 1;
  476                         auio.uio_segflg = UIO_SYSSPACE;
  477                         auio.uio_rw = UIO_READ;
  478                         auio.uio_td = ap->a_td;
  479 
  480                         if (vp->v_type == VREG)
  481                                 error = ncl_readrpc(vp, &auio, ap->a_cred);
  482                         else if (vp->v_type == VDIR) {
  483                                 char* bp;
  484                                 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
  485                                 aiov.iov_base = bp;
  486                                 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
  487                                 error = ncl_readdirrpc(vp, &auio, ap->a_cred,
  488                                     ap->a_td);
  489                                 free(bp, M_TEMP);
  490                         } else if (vp->v_type == VLNK)
  491                                 error = ncl_readlinkrpc(vp, &auio, ap->a_cred);
  492                         else
  493                                 error = EACCES;
  494                 } else
  495                         mtx_unlock(&np->n_mtx);
  496                 return (error);
  497         }
  498 }
  499 
  500 
  501 /*
  502  * nfs open vnode op
  503  * Check to see if the type is ok
  504  * and that deletion is not in progress.
  505  * For paged in text files, you will need to flush the page cache
  506  * if consistency is lost.
  507  */
  508 /* ARGSUSED */
  509 static int
  510 nfs_open(struct vop_open_args *ap)
  511 {
  512         struct vnode *vp = ap->a_vp;
  513         struct nfsnode *np = VTONFS(vp);
  514         struct vattr vattr;
  515         int error;
  516         int fmode = ap->a_mode;
  517         struct ucred *cred;
  518 
  519         if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
  520                 return (EOPNOTSUPP);
  521 
  522         /*
  523          * For NFSv4, we need to do the Open Op before cache validation,
  524          * so that we conform to RFC3530 Sec. 9.3.1.
  525          */
  526         if (NFS_ISV4(vp)) {
  527                 error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td);
  528                 if (error) {
  529                         error = nfscl_maperr(ap->a_td, error, (uid_t)0,
  530                             (gid_t)0);
  531                         return (error);
  532                 }
  533         }
  534 
  535         /*
  536          * Now, if this Open will be doing reading, re-validate/flush the
  537          * cache, so that Close/Open coherency is maintained.
  538          */
  539         mtx_lock(&np->n_mtx);
  540         if (np->n_flag & NMODIFIED) {
  541                 mtx_unlock(&np->n_mtx);
  542                 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
  543                 if (error == EINTR || error == EIO) {
  544                         if (NFS_ISV4(vp))
  545                                 (void) nfsrpc_close(vp, 0, ap->a_td);
  546                         return (error);
  547                 }
  548                 mtx_lock(&np->n_mtx);
  549                 np->n_attrstamp = 0;
  550                 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
  551                 if (vp->v_type == VDIR)
  552                         np->n_direofoffset = 0;
  553                 mtx_unlock(&np->n_mtx);
  554                 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
  555                 if (error) {
  556                         if (NFS_ISV4(vp))
  557                                 (void) nfsrpc_close(vp, 0, ap->a_td);
  558                         return (error);
  559                 }
  560                 mtx_lock(&np->n_mtx);
  561                 np->n_mtime = vattr.va_mtime;
  562                 if (NFS_ISV4(vp))
  563                         np->n_change = vattr.va_filerev;
  564         } else {
  565                 mtx_unlock(&np->n_mtx);
  566                 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
  567                 if (error) {
  568                         if (NFS_ISV4(vp))
  569                                 (void) nfsrpc_close(vp, 0, ap->a_td);
  570                         return (error);
  571                 }
  572                 mtx_lock(&np->n_mtx);
  573                 if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) ||
  574                     NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
  575                         if (vp->v_type == VDIR)
  576                                 np->n_direofoffset = 0;
  577                         mtx_unlock(&np->n_mtx);
  578                         error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
  579                         if (error == EINTR || error == EIO) {
  580                                 if (NFS_ISV4(vp))
  581                                         (void) nfsrpc_close(vp, 0, ap->a_td);
  582                                 return (error);
  583                         }
  584                         mtx_lock(&np->n_mtx);
  585                         np->n_mtime = vattr.va_mtime;
  586                         if (NFS_ISV4(vp))
  587                                 np->n_change = vattr.va_filerev;
  588                 }
  589         }
  590 
  591         /*
  592          * If the object has >= 1 O_DIRECT active opens, we disable caching.
  593          */
  594         if (newnfs_directio_enable && (fmode & O_DIRECT) &&
  595             (vp->v_type == VREG)) {
  596                 if (np->n_directio_opens == 0) {
  597                         mtx_unlock(&np->n_mtx);
  598                         error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
  599                         if (error) {
  600                                 if (NFS_ISV4(vp))
  601                                         (void) nfsrpc_close(vp, 0, ap->a_td);
  602                                 return (error);
  603                         }
  604                         mtx_lock(&np->n_mtx);
  605                         np->n_flag |= NNONCACHE;
  606                 }
  607                 np->n_directio_opens++;
  608         }
  609 
  610         /* If opened for writing via NFSv4.1 or later, mark that for pNFS. */
  611         if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0)
  612                 np->n_flag |= NWRITEOPENED;
  613 
  614         /*
  615          * If this is an open for writing, capture a reference to the
  616          * credentials, so they can be used by ncl_putpages(). Using
  617          * these write credentials is preferable to the credentials of
  618          * whatever thread happens to be doing the VOP_PUTPAGES() since
  619          * the write RPCs are less likely to fail with EACCES.
  620          */
  621         if ((fmode & FWRITE) != 0) {
  622                 cred = np->n_writecred;
  623                 np->n_writecred = crhold(ap->a_cred);
  624         } else
  625                 cred = NULL;
  626         mtx_unlock(&np->n_mtx);
  627 
  628         if (cred != NULL)
  629                 crfree(cred);
  630         vnode_create_vobject(vp, vattr.va_size, ap->a_td);
  631         return (0);
  632 }
  633 
  634 /*
  635  * nfs close vnode op
  636  * What an NFS client should do upon close after writing is a debatable issue.
  637  * Most NFS clients push delayed writes to the server upon close, basically for
  638  * two reasons:
  639  * 1 - So that any write errors may be reported back to the client process
  640  *     doing the close system call. By far the two most likely errors are
  641  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
  642  * 2 - To put a worst case upper bound on cache inconsistency between
  643  *     multiple clients for the file.
  644  * There is also a consistency problem for Version 2 of the protocol w.r.t.
  645  * not being able to tell if other clients are writing a file concurrently,
  646  * since there is no way of knowing if the changed modify time in the reply
  647  * is only due to the write for this client.
  648  * (NFS Version 3 provides weak cache consistency data in the reply that
  649  *  should be sufficient to detect and handle this case.)
  650  *
  651  * The current code does the following:
  652  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
  653  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
  654  *                     or commit them (this satisfies 1 and 2 except for the
  655  *                     case where the server crashes after this close but
  656  *                     before the commit RPC, which is felt to be "good
  657  *                     enough". Changing the last argument to ncl_flush() to
  658  *                     a 1 would force a commit operation, if it is felt a
  659  *                     commit is necessary now.
  660  * for NFS Version 4 - flush the dirty buffers and commit them, if
  661  *                     nfscl_mustflush() says this is necessary.
  662  *                     It is necessary if there is no write delegation held,
  663  *                     in order to satisfy open/close coherency.
  664  *                     If the file isn't cached on local stable storage,
  665  *                     it may be necessary in order to detect "out of space"
  666  *                     errors from the server, if the write delegation
  667  *                     issued by the server doesn't allow the file to grow.
  668  */
  669 /* ARGSUSED */
  670 static int
  671 nfs_close(struct vop_close_args *ap)
  672 {
  673         struct vnode *vp = ap->a_vp;
  674         struct nfsnode *np = VTONFS(vp);
  675         struct nfsvattr nfsva;
  676         struct ucred *cred;
  677         int error = 0, ret, localcred = 0;
  678         int fmode = ap->a_fflag;
  679 
  680         if ((vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF))
  681                 return (0);
  682         /*
  683          * During shutdown, a_cred isn't valid, so just use root.
  684          */
  685         if (ap->a_cred == NOCRED) {
  686                 cred = newnfs_getcred();
  687                 localcred = 1;
  688         } else {
  689                 cred = ap->a_cred;
  690         }
  691         if (vp->v_type == VREG) {
  692             /*
  693              * Examine and clean dirty pages, regardless of NMODIFIED.
  694              * This closes a major hole in close-to-open consistency.
  695              * We want to push out all dirty pages (and buffers) on
  696              * close, regardless of whether they were dirtied by
  697              * mmap'ed writes or via write().
  698              */
  699             if (nfs_clean_pages_on_close && vp->v_object) {
  700                 VM_OBJECT_WLOCK(vp->v_object);
  701                 vm_object_page_clean(vp->v_object, 0, 0, 0);
  702                 VM_OBJECT_WUNLOCK(vp->v_object);
  703             }
  704             mtx_lock(&np->n_mtx);
  705             if (np->n_flag & NMODIFIED) {
  706                 mtx_unlock(&np->n_mtx);
  707                 if (NFS_ISV3(vp)) {
  708                     /*
  709                      * Under NFSv3 we have dirty buffers to dispose of.  We
  710                      * must flush them to the NFS server.  We have the option
  711                      * of waiting all the way through the commit rpc or just
  712                      * waiting for the initial write.  The default is to only
  713                      * wait through the initial write so the data is in the
  714                      * server's cache, which is roughly similar to the state
  715                      * a standard disk subsystem leaves the file in on close().
  716                      *
  717                      * We cannot clear the NMODIFIED bit in np->n_flag due to
  718                      * potential races with other processes, and certainly
  719                      * cannot clear it if we don't commit.
  720                      * These races occur when there is no longer the old
  721                      * traditional vnode locking implemented for Vnode Ops.
  722                      */
  723                     int cm = newnfs_commit_on_close ? 1 : 0;
  724                     error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td, cm, 0);
  725                     /* np->n_flag &= ~NMODIFIED; */
  726                 } else if (NFS_ISV4(vp)) { 
  727                         if (nfscl_mustflush(vp) != 0) {
  728                                 int cm = newnfs_commit_on_close ? 1 : 0;
  729                                 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td,
  730                                     cm, 0);
  731                                 /*
  732                                  * as above w.r.t races when clearing
  733                                  * NMODIFIED.
  734                                  * np->n_flag &= ~NMODIFIED;
  735                                  */
  736                         }
  737                 } else
  738                     error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
  739                 mtx_lock(&np->n_mtx);
  740             }
  741             /* 
  742              * Invalidate the attribute cache in all cases.
  743              * An open is going to fetch fresh attrs any way, other procs
  744              * on this node that have file open will be forced to do an 
  745              * otw attr fetch, but this is safe.
  746              * --> A user found that their RPC count dropped by 20% when
  747              *     this was commented out and I can't see any requirement
  748              *     for it, so I've disabled it when negative lookups are
  749              *     enabled. (What does this have to do with negative lookup
  750              *     caching? Well nothing, except it was reported by the
  751              *     same user that needed negative lookup caching and I wanted
  752              *     there to be a way to disable it to see if it
  753              *     is the cause of some caching/coherency issue that might
  754              *     crop up.)
  755              */
  756             if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) {
  757                     np->n_attrstamp = 0;
  758                     KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
  759             }
  760             if (np->n_flag & NWRITEERR) {
  761                 np->n_flag &= ~NWRITEERR;
  762                 error = np->n_error;
  763             }
  764             mtx_unlock(&np->n_mtx);
  765         }
  766 
  767         if (NFS_ISV4(vp)) {
  768                 /*
  769                  * Get attributes so "change" is up to date.
  770                  */
  771                 if (error == 0 && nfscl_mustflush(vp) != 0 &&
  772                     vp->v_type == VREG &&
  773                     (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) == 0) {
  774                         ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva,
  775                             NULL);
  776                         if (!ret) {
  777                                 np->n_change = nfsva.na_filerev;
  778                                 (void) nfscl_loadattrcache(&vp, &nfsva, NULL,
  779                                     NULL, 0, 0);
  780                         }
  781                 }
  782 
  783                 /*
  784                  * and do the close.
  785                  */
  786                 ret = nfsrpc_close(vp, 0, ap->a_td);
  787                 if (!error && ret)
  788                         error = ret;
  789                 if (error)
  790                         error = nfscl_maperr(ap->a_td, error, (uid_t)0,
  791                             (gid_t)0);
  792         }
  793         if (newnfs_directio_enable)
  794                 KASSERT((np->n_directio_asyncwr == 0),
  795                         ("nfs_close: dirty unflushed (%d) directio buffers\n",
  796                          np->n_directio_asyncwr));
  797         if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
  798                 mtx_lock(&np->n_mtx);
  799                 KASSERT((np->n_directio_opens > 0), 
  800                         ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
  801                 np->n_directio_opens--;
  802                 if (np->n_directio_opens == 0)
  803                         np->n_flag &= ~NNONCACHE;
  804                 mtx_unlock(&np->n_mtx);
  805         }
  806         if (localcred)
  807                 NFSFREECRED(cred);
  808         return (error);
  809 }
  810 
  811 /*
  812  * nfs getattr call from vfs.
  813  */
  814 static int
  815 nfs_getattr(struct vop_getattr_args *ap)
  816 {
  817         struct vnode *vp = ap->a_vp;
  818         struct thread *td = curthread;  /* XXX */
  819         struct nfsnode *np = VTONFS(vp);
  820         int error = 0;
  821         struct nfsvattr nfsva;
  822         struct vattr *vap = ap->a_vap;
  823         struct vattr vattr;
  824 
  825         /*
  826          * Update local times for special files.
  827          */
  828         mtx_lock(&np->n_mtx);
  829         if (np->n_flag & (NACC | NUPD))
  830                 np->n_flag |= NCHG;
  831         mtx_unlock(&np->n_mtx);
  832         /*
  833          * First look in the cache.
  834          */
  835         if (ncl_getattrcache(vp, &vattr) == 0) {
  836                 vap->va_type = vattr.va_type;
  837                 vap->va_mode = vattr.va_mode;
  838                 vap->va_nlink = vattr.va_nlink;
  839                 vap->va_uid = vattr.va_uid;
  840                 vap->va_gid = vattr.va_gid;
  841                 vap->va_fsid = vattr.va_fsid;
  842                 vap->va_fileid = vattr.va_fileid;
  843                 vap->va_size = vattr.va_size;
  844                 vap->va_blocksize = vattr.va_blocksize;
  845                 vap->va_atime = vattr.va_atime;
  846                 vap->va_mtime = vattr.va_mtime;
  847                 vap->va_ctime = vattr.va_ctime;
  848                 vap->va_gen = vattr.va_gen;
  849                 vap->va_flags = vattr.va_flags;
  850                 vap->va_rdev = vattr.va_rdev;
  851                 vap->va_bytes = vattr.va_bytes;
  852                 vap->va_filerev = vattr.va_filerev;
  853                 /*
  854                  * Get the local modify time for the case of a write
  855                  * delegation.
  856                  */
  857                 nfscl_deleggetmodtime(vp, &vap->va_mtime);
  858                 return (0);
  859         }
  860 
  861         if (NFS_ISV34(vp) && nfs_prime_access_cache &&
  862             nfsaccess_cache_timeout > 0) {
  863                 NFSINCRGLOBAL(newnfsstats.accesscache_misses);
  864                 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
  865                 if (ncl_getattrcache(vp, ap->a_vap) == 0) {
  866                         nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
  867                         return (0);
  868                 }
  869         }
  870         error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL);
  871         if (!error)
  872                 error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0);
  873         if (!error) {
  874                 /*
  875                  * Get the local modify time for the case of a write
  876                  * delegation.
  877                  */
  878                 nfscl_deleggetmodtime(vp, &vap->va_mtime);
  879         } else if (NFS_ISV4(vp)) {
  880                 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
  881         }
  882         return (error);
  883 }
  884 
  885 /*
  886  * nfs setattr call.
  887  */
  888 static int
  889 nfs_setattr(struct vop_setattr_args *ap)
  890 {
  891         struct vnode *vp = ap->a_vp;
  892         struct nfsnode *np = VTONFS(vp);
  893         struct thread *td = curthread;  /* XXX */
  894         struct vattr *vap = ap->a_vap;
  895         int error = 0;
  896         u_quad_t tsize;
  897 
  898 #ifndef nolint
  899         tsize = (u_quad_t)0;
  900 #endif
  901 
  902         /*
  903          * Setting of flags and marking of atimes are not supported.
  904          */
  905         if (vap->va_flags != VNOVAL)
  906                 return (EOPNOTSUPP);
  907 
  908         /*
  909          * Disallow write attempts if the filesystem is mounted read-only.
  910          */
  911         if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
  912             vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
  913             vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
  914             (vp->v_mount->mnt_flag & MNT_RDONLY))
  915                 return (EROFS);
  916         if (vap->va_size != VNOVAL) {
  917                 switch (vp->v_type) {
  918                 case VDIR:
  919                         return (EISDIR);
  920                 case VCHR:
  921                 case VBLK:
  922                 case VSOCK:
  923                 case VFIFO:
  924                         if (vap->va_mtime.tv_sec == VNOVAL &&
  925                             vap->va_atime.tv_sec == VNOVAL &&
  926                             vap->va_mode == (mode_t)VNOVAL &&
  927                             vap->va_uid == (uid_t)VNOVAL &&
  928                             vap->va_gid == (gid_t)VNOVAL)
  929                                 return (0);             
  930                         vap->va_size = VNOVAL;
  931                         break;
  932                 default:
  933                         /*
  934                          * Disallow write attempts if the filesystem is
  935                          * mounted read-only.
  936                          */
  937                         if (vp->v_mount->mnt_flag & MNT_RDONLY)
  938                                 return (EROFS);
  939                         /*
  940                          *  We run vnode_pager_setsize() early (why?),
  941                          * we must set np->n_size now to avoid vinvalbuf
  942                          * V_SAVE races that might setsize a lower
  943                          * value.
  944                          */
  945                         mtx_lock(&np->n_mtx);
  946                         tsize = np->n_size;
  947                         mtx_unlock(&np->n_mtx);
  948                         error = ncl_meta_setsize(vp, ap->a_cred, td,
  949                             vap->va_size);
  950                         mtx_lock(&np->n_mtx);
  951                         if (np->n_flag & NMODIFIED) {
  952                             tsize = np->n_size;
  953                             mtx_unlock(&np->n_mtx);
  954                             if (vap->va_size == 0)
  955                                 error = ncl_vinvalbuf(vp, 0, td, 1);
  956                             else
  957                                 error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
  958                             if (error) {
  959                                 vnode_pager_setsize(vp, tsize);
  960                                 return (error);
  961                             }
  962                             /*
  963                              * Call nfscl_delegmodtime() to set the modify time
  964                              * locally, as required.
  965                              */
  966                             nfscl_delegmodtime(vp);
  967                         } else
  968                             mtx_unlock(&np->n_mtx);
  969                         /*
  970                          * np->n_size has already been set to vap->va_size
  971                          * in ncl_meta_setsize(). We must set it again since
  972                          * nfs_loadattrcache() could be called through
  973                          * ncl_meta_setsize() and could modify np->n_size.
  974                          */
  975                         mtx_lock(&np->n_mtx);
  976                         np->n_vattr.na_size = np->n_size = vap->va_size;
  977                         mtx_unlock(&np->n_mtx);
  978                 };
  979         } else {
  980                 mtx_lock(&np->n_mtx);
  981                 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) && 
  982                     (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
  983                         mtx_unlock(&np->n_mtx);
  984                         if ((error = ncl_vinvalbuf(vp, V_SAVE, td, 1)) != 0 &&
  985                             (error == EINTR || error == EIO))
  986                                 return (error);
  987                 } else
  988                         mtx_unlock(&np->n_mtx);
  989         }
  990         error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
  991         if (error && vap->va_size != VNOVAL) {
  992                 mtx_lock(&np->n_mtx);
  993                 np->n_size = np->n_vattr.na_size = tsize;
  994                 vnode_pager_setsize(vp, tsize);
  995                 mtx_unlock(&np->n_mtx);
  996         }
  997         return (error);
  998 }
  999 
 1000 /*
 1001  * Do an nfs setattr rpc.
 1002  */
 1003 static int
 1004 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
 1005     struct thread *td)
 1006 {
 1007         struct nfsnode *np = VTONFS(vp);
 1008         int error, ret, attrflag, i;
 1009         struct nfsvattr nfsva;
 1010 
 1011         if (NFS_ISV34(vp)) {
 1012                 mtx_lock(&np->n_mtx);
 1013                 for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
 1014                         np->n_accesscache[i].stamp = 0;
 1015                 np->n_flag |= NDELEGMOD;
 1016                 mtx_unlock(&np->n_mtx);
 1017                 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
 1018         }
 1019         error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag,
 1020             NULL);
 1021         if (attrflag) {
 1022                 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
 1023                 if (ret && !error)
 1024                         error = ret;
 1025         }
 1026         if (error && NFS_ISV4(vp))
 1027                 error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
 1028         return (error);
 1029 }
 1030 
 1031 /*
 1032  * nfs lookup call, one step at a time...
 1033  * First look in cache
 1034  * If not found, unlock the directory nfsnode and do the rpc
 1035  */
 1036 static int
 1037 nfs_lookup(struct vop_lookup_args *ap)
 1038 {
 1039         struct componentname *cnp = ap->a_cnp;
 1040         struct vnode *dvp = ap->a_dvp;
 1041         struct vnode **vpp = ap->a_vpp;
 1042         struct mount *mp = dvp->v_mount;
 1043         int flags = cnp->cn_flags;
 1044         struct vnode *newvp;
 1045         struct nfsmount *nmp;
 1046         struct nfsnode *np, *newnp;
 1047         int error = 0, attrflag, dattrflag, ltype, ncticks;
 1048         struct thread *td = cnp->cn_thread;
 1049         struct nfsfh *nfhp;
 1050         struct nfsvattr dnfsva, nfsva;
 1051         struct vattr vattr;
 1052         struct timespec nctime;
 1053         
 1054         *vpp = NULLVP;
 1055         if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
 1056             (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
 1057                 return (EROFS);
 1058         if (dvp->v_type != VDIR)
 1059                 return (ENOTDIR);
 1060         nmp = VFSTONFS(mp);
 1061         np = VTONFS(dvp);
 1062 
 1063         /* For NFSv4, wait until any remove is done. */
 1064         mtx_lock(&np->n_mtx);
 1065         while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
 1066                 np->n_flag |= NREMOVEWANT;
 1067                 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
 1068         }
 1069         mtx_unlock(&np->n_mtx);
 1070 
 1071         if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0)
 1072                 return (error);
 1073         error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks);
 1074         if (error > 0 && error != ENOENT)
 1075                 return (error);
 1076         if (error == -1) {
 1077                 /*
 1078                  * Lookups of "." are special and always return the
 1079                  * current directory.  cache_lookup() already handles
 1080                  * associated locking bookkeeping, etc.
 1081                  */
 1082                 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
 1083                         /* XXX: Is this really correct? */
 1084                         if (cnp->cn_nameiop != LOOKUP &&
 1085                             (flags & ISLASTCN))
 1086                                 cnp->cn_flags |= SAVENAME;
 1087                         return (0);
 1088                 }
 1089 
 1090                 /*
 1091                  * We only accept a positive hit in the cache if the
 1092                  * change time of the file matches our cached copy.
 1093                  * Otherwise, we discard the cache entry and fallback
 1094                  * to doing a lookup RPC.  We also only trust cache
 1095                  * entries for less than nm_nametimeo seconds.
 1096                  *
 1097                  * To better handle stale file handles and attributes,
 1098                  * clear the attribute cache of this node if it is a
 1099                  * leaf component, part of an open() call, and not
 1100                  * locally modified before fetching the attributes.
 1101                  * This should allow stale file handles to be detected
 1102                  * here where we can fall back to a LOOKUP RPC to
 1103                  * recover rather than having nfs_open() detect the
 1104                  * stale file handle and failing open(2) with ESTALE.
 1105                  */
 1106                 newvp = *vpp;
 1107                 newnp = VTONFS(newvp);
 1108                 if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
 1109                     (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
 1110                     !(newnp->n_flag & NMODIFIED)) {
 1111                         mtx_lock(&newnp->n_mtx);
 1112                         newnp->n_attrstamp = 0;
 1113                         KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
 1114                         mtx_unlock(&newnp->n_mtx);
 1115                 }
 1116                 if (nfscl_nodeleg(newvp, 0) == 0 ||
 1117                     ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) &&
 1118                     VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
 1119                     timespeccmp(&vattr.va_ctime, &nctime, ==))) {
 1120                         NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
 1121                         if (cnp->cn_nameiop != LOOKUP &&
 1122                             (flags & ISLASTCN))
 1123                                 cnp->cn_flags |= SAVENAME;
 1124                         return (0);
 1125                 }
 1126                 cache_purge(newvp);
 1127                 if (dvp != newvp)
 1128                         vput(newvp);
 1129                 else 
 1130                         vrele(newvp);
 1131                 *vpp = NULLVP;
 1132         } else if (error == ENOENT) {
 1133                 if (dvp->v_iflag & VI_DOOMED)
 1134                         return (ENOENT);
 1135                 /*
 1136                  * We only accept a negative hit in the cache if the
 1137                  * modification time of the parent directory matches
 1138                  * the cached copy in the name cache entry.
 1139                  * Otherwise, we discard all of the negative cache
 1140                  * entries for this directory.  We also only trust
 1141                  * negative cache entries for up to nm_negnametimeo
 1142                  * seconds.
 1143                  */
 1144                 if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) &&
 1145                     VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
 1146                     timespeccmp(&vattr.va_mtime, &nctime, ==)) {
 1147                         NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
 1148                         return (ENOENT);
 1149                 }
 1150                 cache_purge_negative(dvp);
 1151         }
 1152 
 1153         error = 0;
 1154         newvp = NULLVP;
 1155         NFSINCRGLOBAL(newnfsstats.lookupcache_misses);
 1156         error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
 1157             cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
 1158             NULL);
 1159         if (dattrflag)
 1160                 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
 1161         if (error) {
 1162                 if (newvp != NULLVP) {
 1163                         vput(newvp);
 1164                         *vpp = NULLVP;
 1165                 }
 1166 
 1167                 if (error != ENOENT) {
 1168                         if (NFS_ISV4(dvp))
 1169                                 error = nfscl_maperr(td, error, (uid_t)0,
 1170                                     (gid_t)0);
 1171                         return (error);
 1172                 }
 1173 
 1174                 /* The requested file was not found. */
 1175                 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
 1176                     (flags & ISLASTCN)) {
 1177                         /*
 1178                          * XXX: UFS does a full VOP_ACCESS(dvp,
 1179                          * VWRITE) here instead of just checking
 1180                          * MNT_RDONLY.
 1181                          */
 1182                         if (mp->mnt_flag & MNT_RDONLY)
 1183                                 return (EROFS);
 1184                         cnp->cn_flags |= SAVENAME;
 1185                         return (EJUSTRETURN);
 1186                 }
 1187 
 1188                 if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE &&
 1189                     dattrflag) {
 1190                         /*
 1191                          * Cache the modification time of the parent
 1192                          * directory from the post-op attributes in
 1193                          * the name cache entry.  The negative cache
 1194                          * entry will be ignored once the directory
 1195                          * has changed.  Don't bother adding the entry
 1196                          * if the directory has already changed.
 1197                          */
 1198                         mtx_lock(&np->n_mtx);
 1199                         if (timespeccmp(&np->n_vattr.na_mtime,
 1200                             &dnfsva.na_mtime, ==)) {
 1201                                 mtx_unlock(&np->n_mtx);
 1202                                 cache_enter_time(dvp, NULL, cnp,
 1203                                     &dnfsva.na_mtime, NULL);
 1204                         } else
 1205                                 mtx_unlock(&np->n_mtx);
 1206                 }
 1207                 return (ENOENT);
 1208         }
 1209 
 1210         /*
 1211          * Handle RENAME case...
 1212          */
 1213         if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
 1214                 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
 1215                         FREE((caddr_t)nfhp, M_NFSFH);
 1216                         return (EISDIR);
 1217                 }
 1218                 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
 1219                     LK_EXCLUSIVE);
 1220                 if (error)
 1221                         return (error);
 1222                 newvp = NFSTOV(np);
 1223                 if (attrflag)
 1224                         (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
 1225                             0, 1);
 1226                 *vpp = newvp;
 1227                 cnp->cn_flags |= SAVENAME;
 1228                 return (0);
 1229         }
 1230 
 1231         if (flags & ISDOTDOT) {
 1232                 ltype = NFSVOPISLOCKED(dvp);
 1233                 error = vfs_busy(mp, MBF_NOWAIT);
 1234                 if (error != 0) {
 1235                         vfs_ref(mp);
 1236                         NFSVOPUNLOCK(dvp, 0);
 1237                         error = vfs_busy(mp, 0);
 1238                         NFSVOPLOCK(dvp, ltype | LK_RETRY);
 1239                         vfs_rel(mp);
 1240                         if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
 1241                                 vfs_unbusy(mp);
 1242                                 error = ENOENT;
 1243                         }
 1244                         if (error != 0)
 1245                                 return (error);
 1246                 }
 1247                 NFSVOPUNLOCK(dvp, 0);
 1248                 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
 1249                     cnp->cn_lkflags);
 1250                 if (error == 0)
 1251                         newvp = NFSTOV(np);
 1252                 vfs_unbusy(mp);
 1253                 if (newvp != dvp)
 1254                         NFSVOPLOCK(dvp, ltype | LK_RETRY);
 1255                 if (dvp->v_iflag & VI_DOOMED) {
 1256                         if (error == 0) {
 1257                                 if (newvp == dvp)
 1258                                         vrele(newvp);
 1259                                 else
 1260                                         vput(newvp);
 1261                         }
 1262                         error = ENOENT;
 1263                 }
 1264                 if (error != 0)
 1265                         return (error);
 1266                 if (attrflag)
 1267                         (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
 1268                             0, 1);
 1269         } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
 1270                 FREE((caddr_t)nfhp, M_NFSFH);
 1271                 VREF(dvp);
 1272                 newvp = dvp;
 1273                 if (attrflag)
 1274                         (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
 1275                             0, 1);
 1276         } else {
 1277                 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
 1278                     cnp->cn_lkflags);
 1279                 if (error)
 1280                         return (error);
 1281                 newvp = NFSTOV(np);
 1282                 if (attrflag)
 1283                         (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
 1284                             0, 1);
 1285                 else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
 1286                     !(np->n_flag & NMODIFIED)) {                        
 1287                         /*
 1288                          * Flush the attribute cache when opening a
 1289                          * leaf node to ensure that fresh attributes
 1290                          * are fetched in nfs_open() since we did not
 1291                          * fetch attributes from the LOOKUP reply.
 1292                          */
 1293                         mtx_lock(&np->n_mtx);
 1294                         np->n_attrstamp = 0;
 1295                         KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
 1296                         mtx_unlock(&np->n_mtx);
 1297                 }
 1298         }
 1299         if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
 1300                 cnp->cn_flags |= SAVENAME;
 1301         if ((cnp->cn_flags & MAKEENTRY) &&
 1302             (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) &&
 1303             attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0))
 1304                 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
 1305                     newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime);
 1306         *vpp = newvp;
 1307         return (0);
 1308 }
 1309 
 1310 /*
 1311  * nfs read call.
 1312  * Just call ncl_bioread() to do the work.
 1313  */
 1314 static int
 1315 nfs_read(struct vop_read_args *ap)
 1316 {
 1317         struct vnode *vp = ap->a_vp;
 1318 
 1319         switch (vp->v_type) {
 1320         case VREG:
 1321                 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
 1322         case VDIR:
 1323                 return (EISDIR);
 1324         default:
 1325                 return (EOPNOTSUPP);
 1326         }
 1327 }
 1328 
 1329 /*
 1330  * nfs readlink call
 1331  */
 1332 static int
 1333 nfs_readlink(struct vop_readlink_args *ap)
 1334 {
 1335         struct vnode *vp = ap->a_vp;
 1336 
 1337         if (vp->v_type != VLNK)
 1338                 return (EINVAL);
 1339         return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
 1340 }
 1341 
 1342 /*
 1343  * Do a readlink rpc.
 1344  * Called by ncl_doio() from below the buffer cache.
 1345  */
 1346 int
 1347 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
 1348 {
 1349         int error, ret, attrflag;
 1350         struct nfsvattr nfsva;
 1351 
 1352         error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
 1353             &attrflag, NULL);
 1354         if (attrflag) {
 1355                 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
 1356                 if (ret && !error)
 1357                         error = ret;
 1358         }
 1359         if (error && NFS_ISV4(vp))
 1360                 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
 1361         return (error);
 1362 }
 1363 
 1364 /*
 1365  * nfs read rpc call
 1366  * Ditto above
 1367  */
 1368 int
 1369 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
 1370 {
 1371         int error, ret, attrflag;
 1372         struct nfsvattr nfsva;
 1373         struct nfsmount *nmp;
 1374 
 1375         nmp = VFSTONFS(vnode_mount(vp));
 1376         error = EIO;
 1377         attrflag = 0;
 1378         if (NFSHASPNFS(nmp))
 1379                 error = nfscl_doiods(vp, uiop, NULL, NULL,
 1380                     NFSV4OPEN_ACCESSREAD, cred, uiop->uio_td);
 1381         NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error);
 1382         if (error != 0)
 1383                 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva,
 1384                     &attrflag, NULL);
 1385         if (attrflag) {
 1386                 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
 1387                 if (ret && !error)
 1388                         error = ret;
 1389         }
 1390         if (error && NFS_ISV4(vp))
 1391                 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
 1392         return (error);
 1393 }
 1394 
 1395 /*
 1396  * nfs write call
 1397  */
 1398 int
 1399 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
 1400     int *iomode, int *must_commit, int called_from_strategy)
 1401 {
 1402         struct nfsvattr nfsva;
 1403         int error, attrflag, ret;
 1404         struct nfsmount *nmp;
 1405 
 1406         nmp = VFSTONFS(vnode_mount(vp));
 1407         error = EIO;
 1408         attrflag = 0;
 1409         if (NFSHASPNFS(nmp))
 1410                 error = nfscl_doiods(vp, uiop, iomode, must_commit,
 1411                     NFSV4OPEN_ACCESSWRITE, cred, uiop->uio_td);
 1412         NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error);
 1413         if (error != 0)
 1414                 error = nfsrpc_write(vp, uiop, iomode, must_commit, cred,
 1415                     uiop->uio_td, &nfsva, &attrflag, NULL,
 1416                     called_from_strategy);
 1417         if (attrflag) {
 1418                 if (VTONFS(vp)->n_flag & ND_NFSV4)
 1419                         ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1,
 1420                             1);
 1421                 else
 1422                         ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
 1423                             1);
 1424                 if (ret && !error)
 1425                         error = ret;
 1426         }
 1427         if (DOINGASYNC(vp))
 1428                 *iomode = NFSWRITE_FILESYNC;
 1429         if (error && NFS_ISV4(vp))
 1430                 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
 1431         return (error);
 1432 }
 1433 
 1434 /*
 1435  * nfs mknod rpc
 1436  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
 1437  * mode set to specify the file type and the size field for rdev.
 1438  */
 1439 static int
 1440 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
 1441     struct vattr *vap)
 1442 {
 1443         struct nfsvattr nfsva, dnfsva;
 1444         struct vnode *newvp = NULL;
 1445         struct nfsnode *np = NULL, *dnp;
 1446         struct nfsfh *nfhp;
 1447         struct vattr vattr;
 1448         int error = 0, attrflag, dattrflag;
 1449         u_int32_t rdev;
 1450 
 1451         if (vap->va_type == VCHR || vap->va_type == VBLK)
 1452                 rdev = vap->va_rdev;
 1453         else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
 1454                 rdev = 0xffffffff;
 1455         else
 1456                 return (EOPNOTSUPP);
 1457         if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
 1458                 return (error);
 1459         error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
 1460             rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva,
 1461             &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
 1462         if (!error) {
 1463                 if (!nfhp)
 1464                         (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
 1465                             cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
 1466                             &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
 1467                             NULL);
 1468                 if (nfhp)
 1469                         error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
 1470                             cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
 1471         }
 1472         if (dattrflag)
 1473                 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
 1474         if (!error) {
 1475                 newvp = NFSTOV(np);
 1476                 if (attrflag != 0) {
 1477                         error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
 1478                             0, 1);
 1479                         if (error != 0)
 1480                                 vput(newvp);
 1481                 }
 1482         }
 1483         if (!error) {
 1484                 *vpp = newvp;
 1485         } else if (NFS_ISV4(dvp)) {
 1486                 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
 1487                     vap->va_gid);
 1488         }
 1489         dnp = VTONFS(dvp);
 1490         mtx_lock(&dnp->n_mtx);
 1491         dnp->n_flag |= NMODIFIED;
 1492         if (!dattrflag) {
 1493                 dnp->n_attrstamp = 0;
 1494                 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
 1495         }
 1496         mtx_unlock(&dnp->n_mtx);
 1497         return (error);
 1498 }
 1499 
 1500 /*
 1501  * nfs mknod vop
 1502  * just call nfs_mknodrpc() to do the work.
 1503  */
 1504 /* ARGSUSED */
 1505 static int
 1506 nfs_mknod(struct vop_mknod_args *ap)
 1507 {
 1508         return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
 1509 }
 1510 
 1511 static struct mtx nfs_cverf_mtx;
 1512 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex",
 1513     MTX_DEF);
 1514 
 1515 static nfsquad_t
 1516 nfs_get_cverf(void)
 1517 {
 1518         static nfsquad_t cverf;
 1519         nfsquad_t ret;
 1520         static int cverf_initialized = 0;
 1521 
 1522         mtx_lock(&nfs_cverf_mtx);
 1523         if (cverf_initialized == 0) {
 1524                 cverf.lval[0] = arc4random();
 1525                 cverf.lval[1] = arc4random();
 1526                 cverf_initialized = 1;
 1527         } else
 1528                 cverf.qval++;
 1529         ret = cverf;
 1530         mtx_unlock(&nfs_cverf_mtx);
 1531 
 1532         return (ret);
 1533 }
 1534 
 1535 /*
 1536  * nfs file create call
 1537  */
 1538 static int
 1539 nfs_create(struct vop_create_args *ap)
 1540 {
 1541         struct vnode *dvp = ap->a_dvp;
 1542         struct vattr *vap = ap->a_vap;
 1543         struct componentname *cnp = ap->a_cnp;
 1544         struct nfsnode *np = NULL, *dnp;
 1545         struct vnode *newvp = NULL;
 1546         struct nfsmount *nmp;
 1547         struct nfsvattr dnfsva, nfsva;
 1548         struct nfsfh *nfhp;
 1549         nfsquad_t cverf;
 1550         int error = 0, attrflag, dattrflag, fmode = 0;
 1551         struct vattr vattr;
 1552 
 1553         /*
 1554          * Oops, not for me..
 1555          */
 1556         if (vap->va_type == VSOCK)
 1557                 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
 1558 
 1559         if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
 1560                 return (error);
 1561         if (vap->va_vaflags & VA_EXCLUSIVE)
 1562                 fmode |= O_EXCL;
 1563         dnp = VTONFS(dvp);
 1564         nmp = VFSTONFS(vnode_mount(dvp));
 1565 again:
 1566         /* For NFSv4, wait until any remove is done. */
 1567         mtx_lock(&dnp->n_mtx);
 1568         while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
 1569                 dnp->n_flag |= NREMOVEWANT;
 1570                 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
 1571         }
 1572         mtx_unlock(&dnp->n_mtx);
 1573 
 1574         cverf = nfs_get_cverf();
 1575         error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
 1576             vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva,
 1577             &nfhp, &attrflag, &dattrflag, NULL);
 1578         if (!error) {
 1579                 if (nfhp == NULL)
 1580                         (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
 1581                             cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
 1582                             &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
 1583                             NULL);
 1584                 if (nfhp != NULL)
 1585                         error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
 1586                             cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
 1587         }
 1588         if (dattrflag)
 1589                 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
 1590         if (!error) {
 1591                 newvp = NFSTOV(np);
 1592                 if (attrflag == 0)
 1593                         error = nfsrpc_getattr(newvp, cnp->cn_cred,
 1594                             cnp->cn_thread, &nfsva, NULL);
 1595                 if (error == 0)
 1596                         error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
 1597                             0, 1);
 1598         }
 1599         if (error) {
 1600                 if (newvp != NULL) {
 1601                         vput(newvp);
 1602                         newvp = NULL;
 1603                 }
 1604                 if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
 1605                     error == NFSERR_NOTSUPP) {
 1606                         fmode &= ~O_EXCL;
 1607                         goto again;
 1608                 }
 1609         } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
 1610                 if (nfscl_checksattr(vap, &nfsva)) {
 1611                         /*
 1612                          * We are normally called with only a partially
 1613                          * initialized VAP. Since the NFSv3 spec says that
 1614                          * the server may use the file attributes to
 1615                          * store the verifier, the spec requires us to do a
 1616                          * SETATTR RPC. FreeBSD servers store the verifier in
 1617                          * atime, but we can't really assume that all servers
 1618                          * will so we ensure that our SETATTR sets both atime
 1619                          * and mtime.
 1620                          */
 1621                         if (vap->va_mtime.tv_sec == VNOVAL)
 1622                                 vfs_timestamp(&vap->va_mtime);
 1623                         if (vap->va_atime.tv_sec == VNOVAL)
 1624                                 vap->va_atime = vap->va_mtime;
 1625                         error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
 1626                             cnp->cn_thread, &nfsva, &attrflag, NULL);
 1627                         if (error && (vap->va_uid != (uid_t)VNOVAL ||
 1628                             vap->va_gid != (gid_t)VNOVAL)) {
 1629                                 /* try again without setting uid/gid */
 1630                                 vap->va_uid = (uid_t)VNOVAL;
 1631                                 vap->va_gid = (uid_t)VNOVAL;
 1632                                 error = nfsrpc_setattr(newvp, vap, NULL, 
 1633                                     cnp->cn_cred, cnp->cn_thread, &nfsva,
 1634                                     &attrflag, NULL);
 1635                         }
 1636                         if (attrflag)
 1637                                 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
 1638                                     NULL, 0, 1);
 1639                         if (error != 0)
 1640                                 vput(newvp);
 1641                 }
 1642         }
 1643         if (!error) {
 1644                 if ((cnp->cn_flags & MAKEENTRY) && attrflag)
 1645                         cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
 1646                             NULL);
 1647                 *ap->a_vpp = newvp;
 1648         } else if (NFS_ISV4(dvp)) {
 1649                 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
 1650                     vap->va_gid);
 1651         }
 1652         mtx_lock(&dnp->n_mtx);
 1653         dnp->n_flag |= NMODIFIED;
 1654         if (!dattrflag) {
 1655                 dnp->n_attrstamp = 0;
 1656                 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
 1657         }
 1658         mtx_unlock(&dnp->n_mtx);
 1659         return (error);
 1660 }
 1661 
 1662 /*
 1663  * nfs file remove call
 1664  * To try and make nfs semantics closer to ufs semantics, a file that has
 1665  * other processes using the vnode is renamed instead of removed and then
 1666  * removed later on the last close.
 1667  * - If v_usecount > 1
 1668  *        If a rename is not already in the works
 1669  *           call nfs_sillyrename() to set it up
 1670  *     else
 1671  *        do the remove rpc
 1672  */
 1673 static int
 1674 nfs_remove(struct vop_remove_args *ap)
 1675 {
 1676         struct vnode *vp = ap->a_vp;
 1677         struct vnode *dvp = ap->a_dvp;
 1678         struct componentname *cnp = ap->a_cnp;
 1679         struct nfsnode *np = VTONFS(vp);
 1680         int error = 0;
 1681         struct vattr vattr;
 1682 
 1683         KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
 1684         KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
 1685         if (vp->v_type == VDIR)
 1686                 error = EPERM;
 1687         else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
 1688             VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
 1689             vattr.va_nlink > 1)) {
 1690                 /*
 1691                  * Purge the name cache so that the chance of a lookup for
 1692                  * the name succeeding while the remove is in progress is
 1693                  * minimized. Without node locking it can still happen, such
 1694                  * that an I/O op returns ESTALE, but since you get this if
 1695                  * another host removes the file..
 1696                  */
 1697                 cache_purge(vp);
 1698                 /*
 1699                  * throw away biocache buffers, mainly to avoid
 1700                  * unnecessary delayed writes later.
 1701                  */
 1702                 error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1);
 1703                 /* Do the rpc */
 1704                 if (error != EINTR && error != EIO)
 1705                         error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
 1706                             cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
 1707                 /*
 1708                  * Kludge City: If the first reply to the remove rpc is lost..
 1709                  *   the reply to the retransmitted request will be ENOENT
 1710                  *   since the file was in fact removed
 1711                  *   Therefore, we cheat and return success.
 1712                  */
 1713                 if (error == ENOENT)
 1714                         error = 0;
 1715         } else if (!np->n_sillyrename)
 1716                 error = nfs_sillyrename(dvp, vp, cnp);
 1717         mtx_lock(&np->n_mtx);
 1718         np->n_attrstamp = 0;
 1719         mtx_unlock(&np->n_mtx);
 1720         KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
 1721         return (error);
 1722 }
 1723 
 1724 /*
 1725  * nfs file remove rpc called from nfs_inactive
 1726  */
 1727 int
 1728 ncl_removeit(struct sillyrename *sp, struct vnode *vp)
 1729 {
 1730         /*
 1731          * Make sure that the directory vnode is still valid.
 1732          * XXX we should lock sp->s_dvp here.
 1733          */
 1734         if (sp->s_dvp->v_type == VBAD)
 1735                 return (0);
 1736         return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
 1737             sp->s_cred, NULL));
 1738 }
 1739 
 1740 /*
 1741  * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
 1742  */
 1743 static int
 1744 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
 1745     int namelen, struct ucred *cred, struct thread *td)
 1746 {
 1747         struct nfsvattr dnfsva;
 1748         struct nfsnode *dnp = VTONFS(dvp);
 1749         int error = 0, dattrflag;
 1750 
 1751         mtx_lock(&dnp->n_mtx);
 1752         dnp->n_flag |= NREMOVEINPROG;
 1753         mtx_unlock(&dnp->n_mtx);
 1754         error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
 1755             &dattrflag, NULL);
 1756         mtx_lock(&dnp->n_mtx);
 1757         if ((dnp->n_flag & NREMOVEWANT)) {
 1758                 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
 1759                 mtx_unlock(&dnp->n_mtx);
 1760                 wakeup((caddr_t)dnp);
 1761         } else {
 1762                 dnp->n_flag &= ~NREMOVEINPROG;
 1763                 mtx_unlock(&dnp->n_mtx);
 1764         }
 1765         if (dattrflag)
 1766                 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
 1767         mtx_lock(&dnp->n_mtx);
 1768         dnp->n_flag |= NMODIFIED;
 1769         if (!dattrflag) {
 1770                 dnp->n_attrstamp = 0;
 1771                 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
 1772         }
 1773         mtx_unlock(&dnp->n_mtx);
 1774         if (error && NFS_ISV4(dvp))
 1775                 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
 1776         return (error);
 1777 }
 1778 
 1779 /*
 1780  * nfs file rename call
 1781  */
 1782 static int
 1783 nfs_rename(struct vop_rename_args *ap)
 1784 {
 1785         struct vnode *fvp = ap->a_fvp;
 1786         struct vnode *tvp = ap->a_tvp;
 1787         struct vnode *fdvp = ap->a_fdvp;
 1788         struct vnode *tdvp = ap->a_tdvp;
 1789         struct componentname *tcnp = ap->a_tcnp;
 1790         struct componentname *fcnp = ap->a_fcnp;
 1791         struct nfsnode *fnp = VTONFS(ap->a_fvp);
 1792         struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
 1793         struct nfsv4node *newv4 = NULL;
 1794         int error;
 1795 
 1796         KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
 1797             (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
 1798         /* Check for cross-device rename */
 1799         if ((fvp->v_mount != tdvp->v_mount) ||
 1800             (tvp && (fvp->v_mount != tvp->v_mount))) {
 1801                 error = EXDEV;
 1802                 goto out;
 1803         }
 1804 
 1805         if (fvp == tvp) {
 1806                 ncl_printf("nfs_rename: fvp == tvp (can't happen)\n");
 1807                 error = 0;
 1808                 goto out;
 1809         }
 1810         if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0)
 1811                 goto out;
 1812 
 1813         /*
 1814          * We have to flush B_DELWRI data prior to renaming
 1815          * the file.  If we don't, the delayed-write buffers
 1816          * can be flushed out later after the file has gone stale
 1817          * under NFSV3.  NFSV2 does not have this problem because
 1818          * ( as far as I can tell ) it flushes dirty buffers more
 1819          * often.
 1820          * 
 1821          * Skip the rename operation if the fsync fails, this can happen
 1822          * due to the server's volume being full, when we pushed out data
 1823          * that was written back to our cache earlier. Not checking for
 1824          * this condition can result in potential (silent) data loss.
 1825          */
 1826         error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
 1827         NFSVOPUNLOCK(fvp, 0);
 1828         if (!error && tvp)
 1829                 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
 1830         if (error)
 1831                 goto out;
 1832 
 1833         /*
 1834          * If the tvp exists and is in use, sillyrename it before doing the
 1835          * rename of the new file over it.
 1836          * XXX Can't sillyrename a directory.
 1837          */
 1838         if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
 1839                 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
 1840                 vput(tvp);
 1841                 tvp = NULL;
 1842         }
 1843 
 1844         error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
 1845             tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
 1846             tcnp->cn_thread);
 1847 
 1848         if (error == 0 && NFS_ISV4(tdvp)) {
 1849                 /*
 1850                  * For NFSv4, check to see if it is the same name and
 1851                  * replace the name, if it is different.
 1852                  */
 1853                 MALLOC(newv4, struct nfsv4node *,
 1854                     sizeof (struct nfsv4node) +
 1855                     tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
 1856                     M_NFSV4NODE, M_WAITOK);
 1857                 mtx_lock(&tdnp->n_mtx);
 1858                 mtx_lock(&fnp->n_mtx);
 1859                 if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
 1860                     (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
 1861                       NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
 1862                       tcnp->cn_namelen) ||
 1863                       tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
 1864                       NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
 1865                         tdnp->n_fhp->nfh_len))) {
 1866 #ifdef notdef
 1867 { char nnn[100]; int nnnl;
 1868 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
 1869 bcopy(tcnp->cn_nameptr, nnn, nnnl);
 1870 nnn[nnnl] = '\0';
 1871 printf("ren replace=%s\n",nnn);
 1872 }
 1873 #endif
 1874                         FREE((caddr_t)fnp->n_v4, M_NFSV4NODE);
 1875                         fnp->n_v4 = newv4;
 1876                         newv4 = NULL;
 1877                         fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
 1878                         fnp->n_v4->n4_namelen = tcnp->cn_namelen;
 1879                         NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
 1880                             tdnp->n_fhp->nfh_len);
 1881                         NFSBCOPY(tcnp->cn_nameptr,
 1882                             NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
 1883                 }
 1884                 mtx_unlock(&tdnp->n_mtx);
 1885                 mtx_unlock(&fnp->n_mtx);
 1886                 if (newv4 != NULL)
 1887                         FREE((caddr_t)newv4, M_NFSV4NODE);
 1888         }
 1889 
 1890         if (fvp->v_type == VDIR) {
 1891                 if (tvp != NULL && tvp->v_type == VDIR)
 1892                         cache_purge(tdvp);
 1893                 cache_purge(fdvp);
 1894         }
 1895 
 1896 out:
 1897         if (tdvp == tvp)
 1898                 vrele(tdvp);
 1899         else
 1900                 vput(tdvp);
 1901         if (tvp)
 1902                 vput(tvp);
 1903         vrele(fdvp);
 1904         vrele(fvp);
 1905         /*
 1906          * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
 1907          */
 1908         if (error == ENOENT)
 1909                 error = 0;
 1910         return (error);
 1911 }
 1912 
 1913 /*
 1914  * nfs file rename rpc called from nfs_remove() above
 1915  */
 1916 static int
 1917 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
 1918     struct sillyrename *sp)
 1919 {
 1920 
 1921         return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
 1922             sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
 1923             scnp->cn_thread));
 1924 }
 1925 
 1926 /*
 1927  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
 1928  */
 1929 static int
 1930 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
 1931     int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
 1932     int tnamelen, struct ucred *cred, struct thread *td)
 1933 {
 1934         struct nfsvattr fnfsva, tnfsva;
 1935         struct nfsnode *fdnp = VTONFS(fdvp);
 1936         struct nfsnode *tdnp = VTONFS(tdvp);
 1937         int error = 0, fattrflag, tattrflag;
 1938 
 1939         error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
 1940             tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
 1941             &tattrflag, NULL, NULL);
 1942         mtx_lock(&fdnp->n_mtx);
 1943         fdnp->n_flag |= NMODIFIED;
 1944         if (fattrflag != 0) {
 1945                 mtx_unlock(&fdnp->n_mtx);
 1946                 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1);
 1947         } else {
 1948                 fdnp->n_attrstamp = 0;
 1949                 mtx_unlock(&fdnp->n_mtx);
 1950                 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
 1951         }
 1952         mtx_lock(&tdnp->n_mtx);
 1953         tdnp->n_flag |= NMODIFIED;
 1954         if (tattrflag != 0) {
 1955                 mtx_unlock(&tdnp->n_mtx);
 1956                 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1);
 1957         } else {
 1958                 tdnp->n_attrstamp = 0;
 1959                 mtx_unlock(&tdnp->n_mtx);
 1960                 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
 1961         }
 1962         if (error && NFS_ISV4(fdvp))
 1963                 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
 1964         return (error);
 1965 }
 1966 
 1967 /*
 1968  * nfs hard link create call
 1969  */
 1970 static int
 1971 nfs_link(struct vop_link_args *ap)
 1972 {
 1973         struct vnode *vp = ap->a_vp;
 1974         struct vnode *tdvp = ap->a_tdvp;
 1975         struct componentname *cnp = ap->a_cnp;
 1976         struct nfsnode *np, *tdnp;
 1977         struct nfsvattr nfsva, dnfsva;
 1978         int error = 0, attrflag, dattrflag;
 1979 
 1980         /*
 1981          * Push all writes to the server, so that the attribute cache
 1982          * doesn't get "out of sync" with the server.
 1983          * XXX There should be a better way!
 1984          */
 1985         VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
 1986 
 1987         error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
 1988             cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag,
 1989             &dattrflag, NULL);
 1990         tdnp = VTONFS(tdvp);
 1991         mtx_lock(&tdnp->n_mtx);
 1992         tdnp->n_flag |= NMODIFIED;
 1993         if (dattrflag != 0) {
 1994                 mtx_unlock(&tdnp->n_mtx);
 1995                 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1);
 1996         } else {
 1997                 tdnp->n_attrstamp = 0;
 1998                 mtx_unlock(&tdnp->n_mtx);
 1999                 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
 2000         }
 2001         if (attrflag)
 2002                 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
 2003         else {
 2004                 np = VTONFS(vp);
 2005                 mtx_lock(&np->n_mtx);
 2006                 np->n_attrstamp = 0;
 2007                 mtx_unlock(&np->n_mtx);
 2008                 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
 2009         }
 2010         /*
 2011          * If negative lookup caching is enabled, I might as well
 2012          * add an entry for this node. Not necessary for correctness,
 2013          * but if negative caching is enabled, then the system
 2014          * must care about lookup caching hit rate, so...
 2015          */
 2016         if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
 2017             (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
 2018                 cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL);
 2019         }
 2020         if (error && NFS_ISV4(vp))
 2021                 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
 2022                     (gid_t)0);
 2023         return (error);
 2024 }
 2025 
 2026 /*
 2027  * nfs symbolic link create call
 2028  */
 2029 static int
 2030 nfs_symlink(struct vop_symlink_args *ap)
 2031 {
 2032         struct vnode *dvp = ap->a_dvp;
 2033         struct vattr *vap = ap->a_vap;
 2034         struct componentname *cnp = ap->a_cnp;
 2035         struct nfsvattr nfsva, dnfsva;
 2036         struct nfsfh *nfhp;
 2037         struct nfsnode *np = NULL, *dnp;
 2038         struct vnode *newvp = NULL;
 2039         int error = 0, attrflag, dattrflag, ret;
 2040 
 2041         vap->va_type = VLNK;
 2042         error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
 2043             ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva,
 2044             &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
 2045         if (nfhp) {
 2046                 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
 2047                     &np, NULL, LK_EXCLUSIVE);
 2048                 if (!ret)
 2049                         newvp = NFSTOV(np);
 2050                 else if (!error)
 2051                         error = ret;
 2052         }
 2053         if (newvp != NULL) {
 2054                 if (attrflag)
 2055                         (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
 2056                             0, 1);
 2057         } else if (!error) {
 2058                 /*
 2059                  * If we do not have an error and we could not extract the
 2060                  * newvp from the response due to the request being NFSv2, we
 2061                  * have to do a lookup in order to obtain a newvp to return.
 2062                  */
 2063                 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
 2064                     cnp->cn_cred, cnp->cn_thread, &np);
 2065                 if (!error)
 2066                         newvp = NFSTOV(np);
 2067         }
 2068         if (error) {
 2069                 if (newvp)
 2070                         vput(newvp);
 2071                 if (NFS_ISV4(dvp))
 2072                         error = nfscl_maperr(cnp->cn_thread, error,
 2073                             vap->va_uid, vap->va_gid);
 2074         } else {
 2075                 *ap->a_vpp = newvp;
 2076         }
 2077 
 2078         dnp = VTONFS(dvp);
 2079         mtx_lock(&dnp->n_mtx);
 2080         dnp->n_flag |= NMODIFIED;
 2081         if (dattrflag != 0) {
 2082                 mtx_unlock(&dnp->n_mtx);
 2083                 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
 2084         } else {
 2085                 dnp->n_attrstamp = 0;
 2086                 mtx_unlock(&dnp->n_mtx);
 2087                 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
 2088         }
 2089         /*
 2090          * If negative lookup caching is enabled, I might as well
 2091          * add an entry for this node. Not necessary for correctness,
 2092          * but if negative caching is enabled, then the system
 2093          * must care about lookup caching hit rate, so...
 2094          */
 2095         if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
 2096             (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
 2097                 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, NULL);
 2098         }
 2099         return (error);
 2100 }
 2101 
 2102 /*
 2103  * nfs make dir call
 2104  */
 2105 static int
 2106 nfs_mkdir(struct vop_mkdir_args *ap)
 2107 {
 2108         struct vnode *dvp = ap->a_dvp;
 2109         struct vattr *vap = ap->a_vap;
 2110         struct componentname *cnp = ap->a_cnp;
 2111         struct nfsnode *np = NULL, *dnp;
 2112         struct vnode *newvp = NULL;
 2113         struct vattr vattr;
 2114         struct nfsfh *nfhp;
 2115         struct nfsvattr nfsva, dnfsva;
 2116         int error = 0, attrflag, dattrflag, ret;
 2117 
 2118         if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
 2119                 return (error);
 2120         vap->va_type = VDIR;
 2121         error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
 2122             vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp,
 2123             &attrflag, &dattrflag, NULL);
 2124         dnp = VTONFS(dvp);
 2125         mtx_lock(&dnp->n_mtx);
 2126         dnp->n_flag |= NMODIFIED;
 2127         if (dattrflag != 0) {
 2128                 mtx_unlock(&dnp->n_mtx);
 2129                 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
 2130         } else {
 2131                 dnp->n_attrstamp = 0;
 2132                 mtx_unlock(&dnp->n_mtx);
 2133                 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
 2134         }
 2135         if (nfhp) {
 2136                 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
 2137                     &np, NULL, LK_EXCLUSIVE);
 2138                 if (!ret) {
 2139                         newvp = NFSTOV(np);
 2140                         if (attrflag)
 2141                            (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
 2142                                 NULL, 0, 1);
 2143                 } else if (!error)
 2144                         error = ret;
 2145         }
 2146         if (!error && newvp == NULL) {
 2147                 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
 2148                     cnp->cn_cred, cnp->cn_thread, &np);
 2149                 if (!error) {
 2150                         newvp = NFSTOV(np);
 2151                         if (newvp->v_type != VDIR)
 2152                                 error = EEXIST;
 2153                 }
 2154         }
 2155         if (error) {
 2156                 if (newvp)
 2157                         vput(newvp);
 2158                 if (NFS_ISV4(dvp))
 2159                         error = nfscl_maperr(cnp->cn_thread, error,
 2160                             vap->va_uid, vap->va_gid);
 2161         } else {
 2162                 /*
 2163                  * If negative lookup caching is enabled, I might as well
 2164                  * add an entry for this node. Not necessary for correctness,
 2165                  * but if negative caching is enabled, then the system
 2166                  * must care about lookup caching hit rate, so...
 2167                  */
 2168                 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
 2169                     (cnp->cn_flags & MAKEENTRY) &&
 2170                     attrflag != 0 && dattrflag != 0)
 2171                         cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
 2172                             &dnfsva.na_ctime);
 2173                 *ap->a_vpp = newvp;
 2174         }
 2175         return (error);
 2176 }
 2177 
 2178 /*
 2179  * nfs remove directory call
 2180  */
 2181 static int
 2182 nfs_rmdir(struct vop_rmdir_args *ap)
 2183 {
 2184         struct vnode *vp = ap->a_vp;
 2185         struct vnode *dvp = ap->a_dvp;
 2186         struct componentname *cnp = ap->a_cnp;
 2187         struct nfsnode *dnp;
 2188         struct nfsvattr dnfsva;
 2189         int error, dattrflag;
 2190 
 2191         if (dvp == vp)
 2192                 return (EINVAL);
 2193         error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
 2194             cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL);
 2195         dnp = VTONFS(dvp);
 2196         mtx_lock(&dnp->n_mtx);
 2197         dnp->n_flag |= NMODIFIED;
 2198         if (dattrflag != 0) {
 2199                 mtx_unlock(&dnp->n_mtx);
 2200                 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
 2201         } else {
 2202                 dnp->n_attrstamp = 0;
 2203                 mtx_unlock(&dnp->n_mtx);
 2204                 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
 2205         }
 2206 
 2207         cache_purge(dvp);
 2208         cache_purge(vp);
 2209         if (error && NFS_ISV4(dvp))
 2210                 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
 2211                     (gid_t)0);
 2212         /*
 2213          * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
 2214          */
 2215         if (error == ENOENT)
 2216                 error = 0;
 2217         return (error);
 2218 }
 2219 
 2220 /*
 2221  * nfs readdir call
 2222  */
 2223 static int
 2224 nfs_readdir(struct vop_readdir_args *ap)
 2225 {
 2226         struct vnode *vp = ap->a_vp;
 2227         struct nfsnode *np = VTONFS(vp);
 2228         struct uio *uio = ap->a_uio;
 2229         ssize_t tresid;
 2230         int error = 0;
 2231         struct vattr vattr;
 2232         
 2233         if (ap->a_eofflag != NULL)
 2234                 *ap->a_eofflag = 0;
 2235         if (vp->v_type != VDIR) 
 2236                 return(EPERM);
 2237 
 2238         /*
 2239          * First, check for hit on the EOF offset cache
 2240          */
 2241         if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
 2242             (np->n_flag & NMODIFIED) == 0) {
 2243                 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
 2244                         mtx_lock(&np->n_mtx);
 2245                         if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
 2246                             !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
 2247                                 mtx_unlock(&np->n_mtx);
 2248                                 NFSINCRGLOBAL(newnfsstats.direofcache_hits);
 2249                                 if (ap->a_eofflag != NULL)
 2250                                         *ap->a_eofflag = 1;
 2251                                 return (0);
 2252                         } else
 2253                                 mtx_unlock(&np->n_mtx);
 2254                 }
 2255         }
 2256 
 2257         /*
 2258          * Call ncl_bioread() to do the real work.
 2259          */
 2260         tresid = uio->uio_resid;
 2261         error = ncl_bioread(vp, uio, 0, ap->a_cred);
 2262 
 2263         if (!error && uio->uio_resid == tresid) {
 2264                 NFSINCRGLOBAL(newnfsstats.direofcache_misses);
 2265                 if (ap->a_eofflag != NULL)
 2266                         *ap->a_eofflag = 1;
 2267         }
 2268         return (error);
 2269 }
 2270 
 2271 /*
 2272  * Readdir rpc call.
 2273  * Called from below the buffer cache by ncl_doio().
 2274  */
 2275 int
 2276 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
 2277     struct thread *td)
 2278 {
 2279         struct nfsvattr nfsva;
 2280         nfsuint64 *cookiep, cookie;
 2281         struct nfsnode *dnp = VTONFS(vp);
 2282         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 2283         int error = 0, eof, attrflag;
 2284 
 2285         KASSERT(uiop->uio_iovcnt == 1 &&
 2286             (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
 2287             (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
 2288             ("nfs readdirrpc bad uio"));
 2289 
 2290         /*
 2291          * If there is no cookie, assume directory was stale.
 2292          */
 2293         ncl_dircookie_lock(dnp);
 2294         cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
 2295         if (cookiep) {
 2296                 cookie = *cookiep;
 2297                 ncl_dircookie_unlock(dnp);
 2298         } else {
 2299                 ncl_dircookie_unlock(dnp);              
 2300                 return (NFSERR_BAD_COOKIE);
 2301         }
 2302 
 2303         if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
 2304                 (void)ncl_fsinfo(nmp, vp, cred, td);
 2305 
 2306         error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
 2307             &attrflag, &eof, NULL);
 2308         if (attrflag)
 2309                 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
 2310 
 2311         if (!error) {
 2312                 /*
 2313                  * We are now either at the end of the directory or have filled
 2314                  * the block.
 2315                  */
 2316                 if (eof)
 2317                         dnp->n_direofoffset = uiop->uio_offset;
 2318                 else {
 2319                         if (uiop->uio_resid > 0)
 2320                                 ncl_printf("EEK! readdirrpc resid > 0\n");
 2321                         ncl_dircookie_lock(dnp);
 2322                         cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
 2323                         *cookiep = cookie;
 2324                         ncl_dircookie_unlock(dnp);
 2325                 }
 2326         } else if (NFS_ISV4(vp)) {
 2327                 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
 2328         }
 2329         return (error);
 2330 }
 2331 
 2332 /*
 2333  * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
 2334  */
 2335 int
 2336 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
 2337     struct thread *td)
 2338 {
 2339         struct nfsvattr nfsva;
 2340         nfsuint64 *cookiep, cookie;
 2341         struct nfsnode *dnp = VTONFS(vp);
 2342         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 2343         int error = 0, attrflag, eof;
 2344 
 2345         KASSERT(uiop->uio_iovcnt == 1 &&
 2346             (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
 2347             (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
 2348             ("nfs readdirplusrpc bad uio"));
 2349 
 2350         /*
 2351          * If there is no cookie, assume directory was stale.
 2352          */
 2353         ncl_dircookie_lock(dnp);
 2354         cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
 2355         if (cookiep) {
 2356                 cookie = *cookiep;
 2357                 ncl_dircookie_unlock(dnp);
 2358         } else {
 2359                 ncl_dircookie_unlock(dnp);
 2360                 return (NFSERR_BAD_COOKIE);
 2361         }
 2362 
 2363         if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
 2364                 (void)ncl_fsinfo(nmp, vp, cred, td);
 2365         error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
 2366             &attrflag, &eof, NULL);
 2367         if (attrflag)
 2368                 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
 2369 
 2370         if (!error) {
 2371                 /*
 2372                  * We are now either at end of the directory or have filled the
 2373                  * the block.
 2374                  */
 2375                 if (eof)
 2376                         dnp->n_direofoffset = uiop->uio_offset;
 2377                 else {
 2378                         if (uiop->uio_resid > 0)
 2379                                 ncl_printf("EEK! readdirplusrpc resid > 0\n");
 2380                         ncl_dircookie_lock(dnp);
 2381                         cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
 2382                         *cookiep = cookie;
 2383                         ncl_dircookie_unlock(dnp);
 2384                 }
 2385         } else if (NFS_ISV4(vp)) {
 2386                 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
 2387         }
 2388         return (error);
 2389 }
 2390 
 2391 /*
 2392  * Silly rename. To make the NFS filesystem that is stateless look a little
 2393  * more like the "ufs" a remove of an active vnode is translated to a rename
 2394  * to a funny looking filename that is removed by nfs_inactive on the
 2395  * nfsnode. There is the potential for another process on a different client
 2396  * to create the same funny name between the nfs_lookitup() fails and the
 2397  * nfs_rename() completes, but...
 2398  */
 2399 static int
 2400 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
 2401 {
 2402         struct sillyrename *sp;
 2403         struct nfsnode *np;
 2404         int error;
 2405         short pid;
 2406         unsigned int lticks;
 2407 
 2408         cache_purge(dvp);
 2409         np = VTONFS(vp);
 2410         KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
 2411         MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
 2412             M_NEWNFSREQ, M_WAITOK);
 2413         sp->s_cred = crhold(cnp->cn_cred);
 2414         sp->s_dvp = dvp;
 2415         VREF(dvp);
 2416 
 2417         /* 
 2418          * Fudge together a funny name.
 2419          * Changing the format of the funny name to accomodate more 
 2420          * sillynames per directory.
 2421          * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is 
 2422          * CPU ticks since boot.
 2423          */
 2424         pid = cnp->cn_thread->td_proc->p_pid;
 2425         lticks = (unsigned int)ticks;
 2426         for ( ; ; ) {
 2427                 sp->s_namlen = sprintf(sp->s_name, 
 2428                                        ".nfs.%08x.%04x4.4", lticks, 
 2429                                        pid);
 2430                 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
 2431                                  cnp->cn_thread, NULL))
 2432                         break;
 2433                 lticks++;
 2434         }
 2435         error = nfs_renameit(dvp, vp, cnp, sp);
 2436         if (error)
 2437                 goto bad;
 2438         error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
 2439                 cnp->cn_thread, &np);
 2440         np->n_sillyrename = sp;
 2441         return (0);
 2442 bad:
 2443         vrele(sp->s_dvp);
 2444         crfree(sp->s_cred);
 2445         free((caddr_t)sp, M_NEWNFSREQ);
 2446         return (error);
 2447 }
 2448 
 2449 /*
 2450  * Look up a file name and optionally either update the file handle or
 2451  * allocate an nfsnode, depending on the value of npp.
 2452  * npp == NULL  --> just do the lookup
 2453  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
 2454  *                      handled too
 2455  * *npp != NULL --> update the file handle in the vnode
 2456  */
 2457 static int
 2458 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
 2459     struct thread *td, struct nfsnode **npp)
 2460 {
 2461         struct vnode *newvp = NULL, *vp;
 2462         struct nfsnode *np, *dnp = VTONFS(dvp);
 2463         struct nfsfh *nfhp, *onfhp;
 2464         struct nfsvattr nfsva, dnfsva;
 2465         struct componentname cn;
 2466         int error = 0, attrflag, dattrflag;
 2467         u_int hash;
 2468 
 2469         error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
 2470             &nfhp, &attrflag, &dattrflag, NULL);
 2471         if (dattrflag)
 2472                 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
 2473         if (npp && !error) {
 2474                 if (*npp != NULL) {
 2475                     np = *npp;
 2476                     vp = NFSTOV(np);
 2477                     /*
 2478                      * For NFSv4, check to see if it is the same name and
 2479                      * replace the name, if it is different.
 2480                      */
 2481                     if (np->n_v4 != NULL && nfsva.na_type == VREG &&
 2482                         (np->n_v4->n4_namelen != len ||
 2483                          NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
 2484                          dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
 2485                          NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
 2486                          dnp->n_fhp->nfh_len))) {
 2487 #ifdef notdef
 2488 { char nnn[100]; int nnnl;
 2489 nnnl = (len < 100) ? len : 99;
 2490 bcopy(name, nnn, nnnl);
 2491 nnn[nnnl] = '\0';
 2492 printf("replace=%s\n",nnn);
 2493 }
 2494 #endif
 2495                             FREE((caddr_t)np->n_v4, M_NFSV4NODE);
 2496                             MALLOC(np->n_v4, struct nfsv4node *,
 2497                                 sizeof (struct nfsv4node) +
 2498                                 dnp->n_fhp->nfh_len + len - 1,
 2499                                 M_NFSV4NODE, M_WAITOK);
 2500                             np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
 2501                             np->n_v4->n4_namelen = len;
 2502                             NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
 2503                                 dnp->n_fhp->nfh_len);
 2504                             NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
 2505                     }
 2506                     hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
 2507                         FNV1_32_INIT);
 2508                     onfhp = np->n_fhp;
 2509                     /*
 2510                      * Rehash node for new file handle.
 2511                      */
 2512                     vfs_hash_rehash(vp, hash);
 2513                     np->n_fhp = nfhp;
 2514                     if (onfhp != NULL)
 2515                         FREE((caddr_t)onfhp, M_NFSFH);
 2516                     newvp = NFSTOV(np);
 2517                 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
 2518                     FREE((caddr_t)nfhp, M_NFSFH);
 2519                     VREF(dvp);
 2520                     newvp = dvp;
 2521                 } else {
 2522                     cn.cn_nameptr = name;
 2523                     cn.cn_namelen = len;
 2524                     error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
 2525                         &np, NULL, LK_EXCLUSIVE);
 2526                     if (error)
 2527                         return (error);
 2528                     newvp = NFSTOV(np);
 2529                 }
 2530                 if (!attrflag && *npp == NULL) {
 2531                         if (newvp == dvp)
 2532                                 vrele(newvp);
 2533                         else
 2534                                 vput(newvp);
 2535                         return (ENOENT);
 2536                 }
 2537                 if (attrflag)
 2538                         (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
 2539                             0, 1);
 2540         }
 2541         if (npp && *npp == NULL) {
 2542                 if (error) {
 2543                         if (newvp) {
 2544                                 if (newvp == dvp)
 2545                                         vrele(newvp);
 2546                                 else
 2547                                         vput(newvp);
 2548                         }
 2549                 } else
 2550                         *npp = np;
 2551         }
 2552         if (error && NFS_ISV4(dvp))
 2553                 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
 2554         return (error);
 2555 }
 2556 
 2557 /*
 2558  * Nfs Version 3 and 4 commit rpc
 2559  */
 2560 int
 2561 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
 2562    struct thread *td)
 2563 {
 2564         struct nfsvattr nfsva;
 2565         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 2566         int error, attrflag;
 2567 
 2568         mtx_lock(&nmp->nm_mtx);
 2569         if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
 2570                 mtx_unlock(&nmp->nm_mtx);
 2571                 return (0);
 2572         }
 2573         mtx_unlock(&nmp->nm_mtx);
 2574         error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva,
 2575             &attrflag, NULL);
 2576         if (attrflag != 0)
 2577                 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL,
 2578                     0, 1);
 2579         if (error != 0 && NFS_ISV4(vp))
 2580                 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
 2581         return (error);
 2582 }
 2583 
 2584 /*
 2585  * Strategy routine.
 2586  * For async requests when nfsiod(s) are running, queue the request by
 2587  * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
 2588  * request.
 2589  */
 2590 static int
 2591 nfs_strategy(struct vop_strategy_args *ap)
 2592 {
 2593         struct buf *bp = ap->a_bp;
 2594         struct ucred *cr;
 2595 
 2596         KASSERT(!(bp->b_flags & B_DONE),
 2597             ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
 2598         BUF_ASSERT_HELD(bp);
 2599 
 2600         if (bp->b_iocmd == BIO_READ)
 2601                 cr = bp->b_rcred;
 2602         else
 2603                 cr = bp->b_wcred;
 2604 
 2605         /*
 2606          * If the op is asynchronous and an i/o daemon is waiting
 2607          * queue the request, wake it up and wait for completion
 2608          * otherwise just do it ourselves.
 2609          */
 2610         if ((bp->b_flags & B_ASYNC) == 0 ||
 2611             ncl_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
 2612                 (void) ncl_doio(ap->a_vp, bp, cr, curthread, 1);
 2613         return (0);
 2614 }
 2615 
 2616 /*
 2617  * fsync vnode op. Just call ncl_flush() with commit == 1.
 2618  */
 2619 /* ARGSUSED */
 2620 static int
 2621 nfs_fsync(struct vop_fsync_args *ap)
 2622 {
 2623 
 2624         if (ap->a_vp->v_type != VREG) {
 2625                 /*
 2626                  * For NFS, metadata is changed synchronously on the server,
 2627                  * so there is nothing to flush. Also, ncl_flush() clears
 2628                  * the NMODIFIED flag and that shouldn't be done here for
 2629                  * directories.
 2630                  */
 2631                 return (0);
 2632         }
 2633         return (ncl_flush(ap->a_vp, ap->a_waitfor, NULL, ap->a_td, 1, 0));
 2634 }
 2635 
 2636 /*
 2637  * Flush all the blocks associated with a vnode.
 2638  *      Walk through the buffer pool and push any dirty pages
 2639  *      associated with the vnode.
 2640  * If the called_from_renewthread argument is TRUE, it has been called
 2641  * from the NFSv4 renew thread and, as such, cannot block indefinitely
 2642  * waiting for a buffer write to complete.
 2643  */
 2644 int
 2645 ncl_flush(struct vnode *vp, int waitfor, struct ucred *cred, struct thread *td,
 2646     int commit, int called_from_renewthread)
 2647 {
 2648         struct nfsnode *np = VTONFS(vp);
 2649         struct buf *bp;
 2650         int i;
 2651         struct buf *nbp;
 2652         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 2653         int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
 2654         int passone = 1, trycnt = 0;
 2655         u_quad_t off, endoff, toff;
 2656         struct ucred* wcred = NULL;
 2657         struct buf **bvec = NULL;
 2658         struct bufobj *bo;
 2659 #ifndef NFS_COMMITBVECSIZ
 2660 #define NFS_COMMITBVECSIZ       20
 2661 #endif
 2662         struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
 2663         int bvecsize = 0, bveccount;
 2664 
 2665         if (called_from_renewthread != 0)
 2666                 slptimeo = hz;
 2667         if (nmp->nm_flag & NFSMNT_INT)
 2668                 slpflag = PCATCH;
 2669         if (!commit)
 2670                 passone = 0;
 2671         bo = &vp->v_bufobj;
 2672         /*
 2673          * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
 2674          * server, but has not been committed to stable storage on the server
 2675          * yet. On the first pass, the byte range is worked out and the commit
 2676          * rpc is done. On the second pass, ncl_writebp() is called to do the
 2677          * job.
 2678          */
 2679 again:
 2680         off = (u_quad_t)-1;
 2681         endoff = 0;
 2682         bvecpos = 0;
 2683         if (NFS_ISV34(vp) && commit) {
 2684                 if (bvec != NULL && bvec != bvec_on_stack)
 2685                         free(bvec, M_TEMP);
 2686                 /*
 2687                  * Count up how many buffers waiting for a commit.
 2688                  */
 2689                 bveccount = 0;
 2690                 BO_LOCK(bo);
 2691                 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
 2692                         if (!BUF_ISLOCKED(bp) &&
 2693                             (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
 2694                                 == (B_DELWRI | B_NEEDCOMMIT))
 2695                                 bveccount++;
 2696                 }
 2697                 /*
 2698                  * Allocate space to remember the list of bufs to commit.  It is
 2699                  * important to use M_NOWAIT here to avoid a race with nfs_write.
 2700                  * If we can't get memory (for whatever reason), we will end up
 2701                  * committing the buffers one-by-one in the loop below.
 2702                  */
 2703                 if (bveccount > NFS_COMMITBVECSIZ) {
 2704                         /*
 2705                          * Release the vnode interlock to avoid a lock
 2706                          * order reversal.
 2707                          */
 2708                         BO_UNLOCK(bo);
 2709                         bvec = (struct buf **)
 2710                                 malloc(bveccount * sizeof(struct buf *),
 2711                                        M_TEMP, M_NOWAIT);
 2712                         BO_LOCK(bo);
 2713                         if (bvec == NULL) {
 2714                                 bvec = bvec_on_stack;
 2715                                 bvecsize = NFS_COMMITBVECSIZ;
 2716                         } else
 2717                                 bvecsize = bveccount;
 2718                 } else {
 2719                         bvec = bvec_on_stack;
 2720                         bvecsize = NFS_COMMITBVECSIZ;
 2721                 }
 2722                 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
 2723                         if (bvecpos >= bvecsize)
 2724                                 break;
 2725                         if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
 2726                                 nbp = TAILQ_NEXT(bp, b_bobufs);
 2727                                 continue;
 2728                         }
 2729                         if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
 2730                             (B_DELWRI | B_NEEDCOMMIT)) {
 2731                                 BUF_UNLOCK(bp);
 2732                                 nbp = TAILQ_NEXT(bp, b_bobufs);
 2733                                 continue;
 2734                         }
 2735                         BO_UNLOCK(bo);
 2736                         bremfree(bp);
 2737                         /*
 2738                          * Work out if all buffers are using the same cred
 2739                          * so we can deal with them all with one commit.
 2740                          *
 2741                          * NOTE: we are not clearing B_DONE here, so we have
 2742                          * to do it later on in this routine if we intend to
 2743                          * initiate I/O on the bp.
 2744                          *
 2745                          * Note: to avoid loopback deadlocks, we do not
 2746                          * assign b_runningbufspace.
 2747                          */
 2748                         if (wcred == NULL)
 2749                                 wcred = bp->b_wcred;
 2750                         else if (wcred != bp->b_wcred)
 2751                                 wcred = NOCRED;
 2752                         vfs_busy_pages(bp, 1);
 2753 
 2754                         BO_LOCK(bo);
 2755                         /*
 2756                          * bp is protected by being locked, but nbp is not
 2757                          * and vfs_busy_pages() may sleep.  We have to
 2758                          * recalculate nbp.
 2759                          */
 2760                         nbp = TAILQ_NEXT(bp, b_bobufs);
 2761 
 2762                         /*
 2763                          * A list of these buffers is kept so that the
 2764                          * second loop knows which buffers have actually
 2765                          * been committed. This is necessary, since there
 2766                          * may be a race between the commit rpc and new
 2767                          * uncommitted writes on the file.
 2768                          */
 2769                         bvec[bvecpos++] = bp;
 2770                         toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
 2771                                 bp->b_dirtyoff;
 2772                         if (toff < off)
 2773                                 off = toff;
 2774                         toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
 2775                         if (toff > endoff)
 2776                                 endoff = toff;
 2777                 }
 2778                 BO_UNLOCK(bo);
 2779         }
 2780         if (bvecpos > 0) {
 2781                 /*
 2782                  * Commit data on the server, as required.
 2783                  * If all bufs are using the same wcred, then use that with
 2784                  * one call for all of them, otherwise commit each one
 2785                  * separately.
 2786                  */
 2787                 if (wcred != NOCRED)
 2788                         retv = ncl_commit(vp, off, (int)(endoff - off),
 2789                                           wcred, td);
 2790                 else {
 2791                         retv = 0;
 2792                         for (i = 0; i < bvecpos; i++) {
 2793                                 off_t off, size;
 2794                                 bp = bvec[i];
 2795                                 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
 2796                                         bp->b_dirtyoff;
 2797                                 size = (u_quad_t)(bp->b_dirtyend
 2798                                                   - bp->b_dirtyoff);
 2799                                 retv = ncl_commit(vp, off, (int)size,
 2800                                                   bp->b_wcred, td);
 2801                                 if (retv) break;
 2802                         }
 2803                 }
 2804 
 2805                 if (retv == NFSERR_STALEWRITEVERF)
 2806                         ncl_clearcommit(vp->v_mount);
 2807 
 2808                 /*
 2809                  * Now, either mark the blocks I/O done or mark the
 2810                  * blocks dirty, depending on whether the commit
 2811                  * succeeded.
 2812                  */
 2813                 for (i = 0; i < bvecpos; i++) {
 2814                         bp = bvec[i];
 2815                         bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
 2816                         if (retv) {
 2817                                 /*
 2818                                  * Error, leave B_DELWRI intact
 2819                                  */
 2820                                 vfs_unbusy_pages(bp);
 2821                                 brelse(bp);
 2822                         } else {
 2823                                 /*
 2824                                  * Success, remove B_DELWRI ( bundirty() ).
 2825                                  *
 2826                                  * b_dirtyoff/b_dirtyend seem to be NFS
 2827                                  * specific.  We should probably move that
 2828                                  * into bundirty(). XXX
 2829                                  */
 2830                                 bufobj_wref(bo);
 2831                                 bp->b_flags |= B_ASYNC;
 2832                                 bundirty(bp);
 2833                                 bp->b_flags &= ~B_DONE;
 2834                                 bp->b_ioflags &= ~BIO_ERROR;
 2835                                 bp->b_dirtyoff = bp->b_dirtyend = 0;
 2836                                 bufdone(bp);
 2837                         }
 2838                 }
 2839         }
 2840 
 2841         /*
 2842          * Start/do any write(s) that are required.
 2843          */
 2844 loop:
 2845         BO_LOCK(bo);
 2846         TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
 2847                 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
 2848                         if (waitfor != MNT_WAIT || passone)
 2849                                 continue;
 2850 
 2851                         error = BUF_TIMELOCK(bp,
 2852                             LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
 2853                             BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo);
 2854                         if (error == 0) {
 2855                                 BUF_UNLOCK(bp);
 2856                                 goto loop;
 2857                         }
 2858                         if (error == ENOLCK) {
 2859                                 error = 0;
 2860                                 goto loop;
 2861                         }
 2862                         if (called_from_renewthread != 0) {
 2863                                 /*
 2864                                  * Return EIO so the flush will be retried
 2865                                  * later.
 2866                                  */
 2867                                 error = EIO;
 2868                                 goto done;
 2869                         }
 2870                         if (newnfs_sigintr(nmp, td)) {
 2871                                 error = EINTR;
 2872                                 goto done;
 2873                         }
 2874                         if (slpflag == PCATCH) {
 2875                                 slpflag = 0;
 2876                                 slptimeo = 2 * hz;
 2877                         }
 2878                         goto loop;
 2879                 }
 2880                 if ((bp->b_flags & B_DELWRI) == 0)
 2881                         panic("nfs_fsync: not dirty");
 2882                 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
 2883                         BUF_UNLOCK(bp);
 2884                         continue;
 2885                 }
 2886                 BO_UNLOCK(bo);
 2887                 bremfree(bp);
 2888                 if (passone || !commit)
 2889                     bp->b_flags |= B_ASYNC;
 2890                 else
 2891                     bp->b_flags |= B_ASYNC;
 2892                 bwrite(bp);
 2893                 if (newnfs_sigintr(nmp, td)) {
 2894                         error = EINTR;
 2895                         goto done;
 2896                 }
 2897                 goto loop;
 2898         }
 2899         if (passone) {
 2900                 passone = 0;
 2901                 BO_UNLOCK(bo);
 2902                 goto again;
 2903         }
 2904         if (waitfor == MNT_WAIT) {
 2905                 while (bo->bo_numoutput) {
 2906                         error = bufobj_wwait(bo, slpflag, slptimeo);
 2907                         if (error) {
 2908                             BO_UNLOCK(bo);
 2909                             if (called_from_renewthread != 0) {
 2910                                 /*
 2911                                  * Return EIO so that the flush will be
 2912                                  * retried later.
 2913                                  */
 2914                                 error = EIO;
 2915                                 goto done;
 2916                             }
 2917                             error = newnfs_sigintr(nmp, td);
 2918                             if (error)
 2919                                 goto done;
 2920                             if (slpflag == PCATCH) {
 2921                                 slpflag = 0;
 2922                                 slptimeo = 2 * hz;
 2923                             }
 2924                             BO_LOCK(bo);
 2925                         }
 2926                 }
 2927                 if (bo->bo_dirty.bv_cnt != 0 && commit) {
 2928                         BO_UNLOCK(bo);
 2929                         goto loop;
 2930                 }
 2931                 /*
 2932                  * Wait for all the async IO requests to drain
 2933                  */
 2934                 BO_UNLOCK(bo);
 2935                 mtx_lock(&np->n_mtx);
 2936                 while (np->n_directio_asyncwr > 0) {
 2937                         np->n_flag |= NFSYNCWAIT;
 2938                         error = newnfs_msleep(td, &np->n_directio_asyncwr,
 2939                             &np->n_mtx, slpflag | (PRIBIO + 1), 
 2940                             "nfsfsync", 0);
 2941                         if (error) {
 2942                                 if (newnfs_sigintr(nmp, td)) {
 2943                                         mtx_unlock(&np->n_mtx);
 2944                                         error = EINTR;  
 2945                                         goto done;
 2946                                 }
 2947                         }
 2948                 }
 2949                 mtx_unlock(&np->n_mtx);
 2950         } else
 2951                 BO_UNLOCK(bo);
 2952         if (NFSHASPNFS(nmp)) {
 2953                 nfscl_layoutcommit(vp, td);
 2954                 /*
 2955                  * Invalidate the attribute cache, since writes to a DS
 2956                  * won't update the size attribute.
 2957                  */
 2958                 mtx_lock(&np->n_mtx);
 2959                 np->n_attrstamp = 0;
 2960         } else
 2961                 mtx_lock(&np->n_mtx);
 2962         if (np->n_flag & NWRITEERR) {
 2963                 error = np->n_error;
 2964                 np->n_flag &= ~NWRITEERR;
 2965         }
 2966         if (commit && bo->bo_dirty.bv_cnt == 0 &&
 2967             bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
 2968                 np->n_flag &= ~NMODIFIED;
 2969         mtx_unlock(&np->n_mtx);
 2970 done:
 2971         if (bvec != NULL && bvec != bvec_on_stack)
 2972                 free(bvec, M_TEMP);
 2973         if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
 2974             (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
 2975              np->n_directio_asyncwr != 0) && trycnt++ < 5) {
 2976                 /* try, try again... */
 2977                 passone = 1;
 2978                 wcred = NULL;
 2979                 bvec = NULL;
 2980                 bvecsize = 0;
 2981 printf("try%d\n", trycnt);
 2982                 goto again;
 2983         }
 2984         return (error);
 2985 }
 2986 
 2987 /*
 2988  * NFS advisory byte-level locks.
 2989  */
 2990 static int
 2991 nfs_advlock(struct vop_advlock_args *ap)
 2992 {
 2993         struct vnode *vp = ap->a_vp;
 2994         struct ucred *cred;
 2995         struct nfsnode *np = VTONFS(ap->a_vp);
 2996         struct proc *p = (struct proc *)ap->a_id;
 2997         struct thread *td = curthread;  /* XXX */
 2998         struct vattr va;
 2999         int ret, error = EOPNOTSUPP;
 3000         u_quad_t size;
 3001         
 3002         if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) {
 3003                 if (vp->v_type != VREG)
 3004                         return (EINVAL);
 3005                 if ((ap->a_flags & F_POSIX) != 0)
 3006                         cred = p->p_ucred;
 3007                 else
 3008                         cred = td->td_ucred;
 3009                 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
 3010                 if (vp->v_iflag & VI_DOOMED) {
 3011                         NFSVOPUNLOCK(vp, 0);
 3012                         return (EBADF);
 3013                 }
 3014 
 3015                 /*
 3016                  * If this is unlocking a write locked region, flush and
 3017                  * commit them before unlocking. This is required by
 3018                  * RFC3530 Sec. 9.3.2.
 3019                  */
 3020                 if (ap->a_op == F_UNLCK &&
 3021                     nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id,
 3022                     ap->a_flags))
 3023                         (void) ncl_flush(vp, MNT_WAIT, cred, td, 1, 0);
 3024 
 3025                 /*
 3026                  * Loop around doing the lock op, while a blocking lock
 3027                  * must wait for the lock op to succeed.
 3028                  */
 3029                 do {
 3030                         ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
 3031                             ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags);
 3032                         if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
 3033                             ap->a_op == F_SETLK) {
 3034                                 NFSVOPUNLOCK(vp, 0);
 3035                                 error = nfs_catnap(PZERO | PCATCH, ret,
 3036                                     "ncladvl");
 3037                                 if (error)
 3038                                         return (EINTR);
 3039                                 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
 3040                                 if (vp->v_iflag & VI_DOOMED) {
 3041                                         NFSVOPUNLOCK(vp, 0);
 3042                                         return (EBADF);
 3043                                 }
 3044                         }
 3045                 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
 3046                      ap->a_op == F_SETLK);
 3047                 if (ret == NFSERR_DENIED) {
 3048                         NFSVOPUNLOCK(vp, 0);
 3049                         return (EAGAIN);
 3050                 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
 3051                         NFSVOPUNLOCK(vp, 0);
 3052                         return (ret);
 3053                 } else if (ret != 0) {
 3054                         NFSVOPUNLOCK(vp, 0);
 3055                         return (EACCES);
 3056                 }
 3057 
 3058                 /*
 3059                  * Now, if we just got a lock, invalidate data in the buffer
 3060                  * cache, as required, so that the coherency conforms with
 3061                  * RFC3530 Sec. 9.3.2.
 3062                  */
 3063                 if (ap->a_op == F_SETLK) {
 3064                         if ((np->n_flag & NMODIFIED) == 0) {
 3065                                 np->n_attrstamp = 0;
 3066                                 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
 3067                                 ret = VOP_GETATTR(vp, &va, cred);
 3068                         }
 3069                         if ((np->n_flag & NMODIFIED) || ret ||
 3070                             np->n_change != va.va_filerev) {
 3071                                 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
 3072                                 np->n_attrstamp = 0;
 3073                                 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
 3074                                 ret = VOP_GETATTR(vp, &va, cred);
 3075                                 if (!ret) {
 3076                                         np->n_mtime = va.va_mtime;
 3077                                         np->n_change = va.va_filerev;
 3078                                 }
 3079                         }
 3080                         /* Mark that a file lock has been acquired. */
 3081                         mtx_lock(&np->n_mtx);
 3082                         np->n_flag |= NHASBEENLOCKED;
 3083                         mtx_unlock(&np->n_mtx);
 3084                 }
 3085                 NFSVOPUNLOCK(vp, 0);
 3086                 return (0);
 3087         } else if (!NFS_ISV4(vp)) {
 3088                 error = NFSVOPLOCK(vp, LK_SHARED);
 3089                 if (error)
 3090                         return (error);
 3091                 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
 3092                         size = VTONFS(vp)->n_size;
 3093                         NFSVOPUNLOCK(vp, 0);
 3094                         error = lf_advlock(ap, &(vp->v_lockf), size);
 3095                 } else {
 3096                         if (nfs_advlock_p != NULL)
 3097                                 error = nfs_advlock_p(ap);
 3098                         else {
 3099                                 NFSVOPUNLOCK(vp, 0);
 3100                                 error = ENOLCK;
 3101                         }
 3102                 }
 3103                 if (error == 0 && ap->a_op == F_SETLK) {
 3104                         /* Mark that a file lock has been acquired. */
 3105                         mtx_lock(&np->n_mtx);
 3106                         np->n_flag |= NHASBEENLOCKED;
 3107                         mtx_unlock(&np->n_mtx);
 3108                 }
 3109         }
 3110         return (error);
 3111 }
 3112 
 3113 /*
 3114  * NFS advisory byte-level locks.
 3115  */
 3116 static int
 3117 nfs_advlockasync(struct vop_advlockasync_args *ap)
 3118 {
 3119         struct vnode *vp = ap->a_vp;
 3120         u_quad_t size;
 3121         int error;
 3122         
 3123         if (NFS_ISV4(vp))
 3124                 return (EOPNOTSUPP);
 3125         error = NFSVOPLOCK(vp, LK_SHARED);
 3126         if (error)
 3127                 return (error);
 3128         if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
 3129                 size = VTONFS(vp)->n_size;
 3130                 NFSVOPUNLOCK(vp, 0);
 3131                 error = lf_advlockasync(ap, &(vp->v_lockf), size);
 3132         } else {
 3133                 NFSVOPUNLOCK(vp, 0);
 3134                 error = EOPNOTSUPP;
 3135         }
 3136         return (error);
 3137 }
 3138 
 3139 /*
 3140  * Print out the contents of an nfsnode.
 3141  */
 3142 static int
 3143 nfs_print(struct vop_print_args *ap)
 3144 {
 3145         struct vnode *vp = ap->a_vp;
 3146         struct nfsnode *np = VTONFS(vp);
 3147 
 3148         ncl_printf("\tfileid %ld fsid 0x%x",
 3149            np->n_vattr.na_fileid, np->n_vattr.na_fsid);
 3150         if (vp->v_type == VFIFO)
 3151                 fifo_printinfo(vp);
 3152         printf("\n");
 3153         return (0);
 3154 }
 3155 
 3156 /*
 3157  * This is the "real" nfs::bwrite(struct buf*).
 3158  * We set B_CACHE if this is a VMIO buffer.
 3159  */
 3160 int
 3161 ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
 3162 {
 3163         int s;
 3164         int oldflags = bp->b_flags;
 3165 #if 0
 3166         int retv = 1;
 3167         off_t off;
 3168 #endif
 3169 
 3170         BUF_ASSERT_HELD(bp);
 3171 
 3172         if (bp->b_flags & B_INVAL) {
 3173                 brelse(bp);
 3174                 return(0);
 3175         }
 3176 
 3177         bp->b_flags |= B_CACHE;
 3178 
 3179         /*
 3180          * Undirty the bp.  We will redirty it later if the I/O fails.
 3181          */
 3182 
 3183         s = splbio();
 3184         bundirty(bp);
 3185         bp->b_flags &= ~B_DONE;
 3186         bp->b_ioflags &= ~BIO_ERROR;
 3187         bp->b_iocmd = BIO_WRITE;
 3188 
 3189         bufobj_wref(bp->b_bufobj);
 3190         curthread->td_ru.ru_oublock++;
 3191         splx(s);
 3192 
 3193         /*
 3194          * Note: to avoid loopback deadlocks, we do not
 3195          * assign b_runningbufspace.
 3196          */
 3197         vfs_busy_pages(bp, 1);
 3198 
 3199         BUF_KERNPROC(bp);
 3200         bp->b_iooffset = dbtob(bp->b_blkno);
 3201         bstrategy(bp);
 3202 
 3203         if( (oldflags & B_ASYNC) == 0) {
 3204                 int rtval = bufwait(bp);
 3205 
 3206                 if (oldflags & B_DELWRI) {
 3207                         s = splbio();
 3208                         reassignbuf(bp);
 3209                         splx(s);
 3210                 }
 3211                 brelse(bp);
 3212                 return (rtval);
 3213         }
 3214 
 3215         return (0);
 3216 }
 3217 
 3218 /*
 3219  * nfs special file access vnode op.
 3220  * Essentially just get vattr and then imitate iaccess() since the device is
 3221  * local to the client.
 3222  */
 3223 static int
 3224 nfsspec_access(struct vop_access_args *ap)
 3225 {
 3226         struct vattr *vap;
 3227         struct ucred *cred = ap->a_cred;
 3228         struct vnode *vp = ap->a_vp;
 3229         accmode_t accmode = ap->a_accmode;
 3230         struct vattr vattr;
 3231         int error;
 3232 
 3233         /*
 3234          * Disallow write attempts on filesystems mounted read-only;
 3235          * unless the file is a socket, fifo, or a block or character
 3236          * device resident on the filesystem.
 3237          */
 3238         if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
 3239                 switch (vp->v_type) {
 3240                 case VREG:
 3241                 case VDIR:
 3242                 case VLNK:
 3243                         return (EROFS);
 3244                 default:
 3245                         break;
 3246                 }
 3247         }
 3248         vap = &vattr;
 3249         error = VOP_GETATTR(vp, vap, cred);
 3250         if (error)
 3251                 goto out;
 3252         error  = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
 3253             accmode, cred, NULL);
 3254 out:
 3255         return error;
 3256 }
 3257 
 3258 /*
 3259  * Read wrapper for fifos.
 3260  */
 3261 static int
 3262 nfsfifo_read(struct vop_read_args *ap)
 3263 {
 3264         struct nfsnode *np = VTONFS(ap->a_vp);
 3265         int error;
 3266 
 3267         /*
 3268          * Set access flag.
 3269          */
 3270         mtx_lock(&np->n_mtx);
 3271         np->n_flag |= NACC;
 3272         vfs_timestamp(&np->n_atim);
 3273         mtx_unlock(&np->n_mtx);
 3274         error = fifo_specops.vop_read(ap);
 3275         return error;   
 3276 }
 3277 
 3278 /*
 3279  * Write wrapper for fifos.
 3280  */
 3281 static int
 3282 nfsfifo_write(struct vop_write_args *ap)
 3283 {
 3284         struct nfsnode *np = VTONFS(ap->a_vp);
 3285 
 3286         /*
 3287          * Set update flag.
 3288          */
 3289         mtx_lock(&np->n_mtx);
 3290         np->n_flag |= NUPD;
 3291         vfs_timestamp(&np->n_mtim);
 3292         mtx_unlock(&np->n_mtx);
 3293         return(fifo_specops.vop_write(ap));
 3294 }
 3295 
 3296 /*
 3297  * Close wrapper for fifos.
 3298  *
 3299  * Update the times on the nfsnode then do fifo close.
 3300  */
 3301 static int
 3302 nfsfifo_close(struct vop_close_args *ap)
 3303 {
 3304         struct vnode *vp = ap->a_vp;
 3305         struct nfsnode *np = VTONFS(vp);
 3306         struct vattr vattr;
 3307         struct timespec ts;
 3308 
 3309         mtx_lock(&np->n_mtx);
 3310         if (np->n_flag & (NACC | NUPD)) {
 3311                 vfs_timestamp(&ts);
 3312                 if (np->n_flag & NACC)
 3313                         np->n_atim = ts;
 3314                 if (np->n_flag & NUPD)
 3315                         np->n_mtim = ts;
 3316                 np->n_flag |= NCHG;
 3317                 if (vrefcnt(vp) == 1 &&
 3318                     (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
 3319                         VATTR_NULL(&vattr);
 3320                         if (np->n_flag & NACC)
 3321                                 vattr.va_atime = np->n_atim;
 3322                         if (np->n_flag & NUPD)
 3323                                 vattr.va_mtime = np->n_mtim;
 3324                         mtx_unlock(&np->n_mtx);
 3325                         (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
 3326                         goto out;
 3327                 }
 3328         }
 3329         mtx_unlock(&np->n_mtx);
 3330 out:
 3331         return (fifo_specops.vop_close(ap));
 3332 }
 3333 
 3334 /*
 3335  * Just call ncl_writebp() with the force argument set to 1.
 3336  *
 3337  * NOTE: B_DONE may or may not be set in a_bp on call.
 3338  */
 3339 static int
 3340 nfs_bwrite(struct buf *bp)
 3341 {
 3342 
 3343         return (ncl_writebp(bp, 1, curthread));
 3344 }
 3345 
 3346 struct buf_ops buf_ops_newnfs = {
 3347         .bop_name       =       "buf_ops_nfs",
 3348         .bop_write      =       nfs_bwrite,
 3349         .bop_strategy   =       bufstrategy,
 3350         .bop_sync       =       bufsync,
 3351         .bop_bdflush    =       bufbdflush,
 3352 };
 3353 
 3354 /*
 3355  * Cloned from vop_stdlock(), and then the ugly hack added.
 3356  */
 3357 static int
 3358 nfs_lock1(struct vop_lock1_args *ap)
 3359 {
 3360         struct vnode *vp = ap->a_vp;
 3361         int error = 0;
 3362 
 3363         /*
 3364          * Since vfs_hash_get() calls vget() and it will no longer work
 3365          * for FreeBSD8 with flags == 0, I can only think of this horrible
 3366          * hack to work around it. I call vfs_hash_get() with LK_EXCLOTHER
 3367          * and then handle it here. All I want for this case is a v_usecount
 3368          * on the vnode to use for recovery, while another thread might
 3369          * hold a lock on the vnode. I have the other threads blocked, so
 3370          * there isn't any race problem.
 3371          */
 3372         if ((ap->a_flags & LK_TYPE_MASK) == LK_EXCLOTHER) {
 3373                 if ((ap->a_flags & LK_INTERLOCK) == 0)
 3374                         panic("ncllock1");
 3375                 if ((vp->v_iflag & VI_DOOMED))
 3376                         error = ENOENT;
 3377                 VI_UNLOCK(vp);
 3378                 return (error);
 3379         }
 3380         return (_lockmgr_args(vp->v_vnlock, ap->a_flags, VI_MTX(vp),
 3381             LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, ap->a_file,
 3382             ap->a_line));
 3383 }
 3384 
 3385 static int
 3386 nfs_getacl(struct vop_getacl_args *ap)
 3387 {
 3388         int error;
 3389 
 3390         if (ap->a_type != ACL_TYPE_NFS4)
 3391                 return (EOPNOTSUPP);
 3392         error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
 3393             NULL);
 3394         if (error > NFSERR_STALE) {
 3395                 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
 3396                 error = EPERM;
 3397         }
 3398         return (error);
 3399 }
 3400 
 3401 static int
 3402 nfs_setacl(struct vop_setacl_args *ap)
 3403 {
 3404         int error;
 3405 
 3406         if (ap->a_type != ACL_TYPE_NFS4)
 3407                 return (EOPNOTSUPP);
 3408         error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
 3409             NULL);
 3410         if (error > NFSERR_STALE) {
 3411                 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
 3412                 error = EPERM;
 3413         }
 3414         return (error);
 3415 }
 3416 
 3417 /*
 3418  * Return POSIX pathconf information applicable to nfs filesystems.
 3419  */
 3420 static int
 3421 nfs_pathconf(struct vop_pathconf_args *ap)
 3422 {
 3423         struct nfsv3_pathconf pc;
 3424         struct nfsvattr nfsva;
 3425         struct vnode *vp = ap->a_vp;
 3426         struct thread *td = curthread;
 3427         int attrflag, error;
 3428 
 3429         if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX ||
 3430             ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED ||
 3431             ap->a_name == _PC_NO_TRUNC)) ||
 3432             (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) {
 3433                 /*
 3434                  * Since only the above 4 a_names are returned by the NFSv3
 3435                  * Pathconf RPC, there is no point in doing it for others.
 3436                  * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can
 3437                  * be used for _PC_NFS4_ACL as well.
 3438                  */
 3439                 error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva,
 3440                     &attrflag, NULL);
 3441                 if (attrflag != 0)
 3442                         (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
 3443                             1);
 3444                 if (error != 0)
 3445                         return (error);
 3446         } else {
 3447                 /*
 3448                  * For NFSv2 (or NFSv3 when not one of the above 4 a_names),
 3449                  * just fake them.
 3450                  */
 3451                 pc.pc_linkmax = LINK_MAX;
 3452                 pc.pc_namemax = NFS_MAXNAMLEN;
 3453                 pc.pc_notrunc = 1;
 3454                 pc.pc_chownrestricted = 1;
 3455                 pc.pc_caseinsensitive = 0;
 3456                 pc.pc_casepreserving = 1;
 3457                 error = 0;
 3458         }
 3459         switch (ap->a_name) {
 3460         case _PC_LINK_MAX:
 3461                 *ap->a_retval = pc.pc_linkmax;
 3462                 break;
 3463         case _PC_NAME_MAX:
 3464                 *ap->a_retval = pc.pc_namemax;
 3465                 break;
 3466         case _PC_PATH_MAX:
 3467                 *ap->a_retval = PATH_MAX;
 3468                 break;
 3469         case _PC_PIPE_BUF:
 3470                 *ap->a_retval = PIPE_BUF;
 3471                 break;
 3472         case _PC_CHOWN_RESTRICTED:
 3473                 *ap->a_retval = pc.pc_chownrestricted;
 3474                 break;
 3475         case _PC_NO_TRUNC:
 3476                 *ap->a_retval = pc.pc_notrunc;
 3477                 break;
 3478         case _PC_ACL_EXTENDED:
 3479                 *ap->a_retval = 0;
 3480                 break;
 3481         case _PC_ACL_NFS4:
 3482                 if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 &&
 3483                     NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL))
 3484                         *ap->a_retval = 1;
 3485                 else
 3486                         *ap->a_retval = 0;
 3487                 break;
 3488         case _PC_ACL_PATH_MAX:
 3489                 if (NFS_ISV4(vp))
 3490                         *ap->a_retval = ACL_MAX_ENTRIES;
 3491                 else
 3492                         *ap->a_retval = 3;
 3493                 break;
 3494         case _PC_MAC_PRESENT:
 3495                 *ap->a_retval = 0;
 3496                 break;
 3497         case _PC_ASYNC_IO:
 3498                 /* _PC_ASYNC_IO should have been handled by upper layers. */
 3499                 KASSERT(0, ("_PC_ASYNC_IO should not get here"));
 3500                 error = EINVAL;
 3501                 break;
 3502         case _PC_PRIO_IO:
 3503                 *ap->a_retval = 0;
 3504                 break;
 3505         case _PC_SYNC_IO:
 3506                 *ap->a_retval = 0;
 3507                 break;
 3508         case _PC_ALLOC_SIZE_MIN:
 3509                 *ap->a_retval = vp->v_mount->mnt_stat.f_bsize;
 3510                 break;
 3511         case _PC_FILESIZEBITS:
 3512                 if (NFS_ISV34(vp))
 3513                         *ap->a_retval = 64;
 3514                 else
 3515                         *ap->a_retval = 32;
 3516                 break;
 3517         case _PC_REC_INCR_XFER_SIZE:
 3518                 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
 3519                 break;
 3520         case _PC_REC_MAX_XFER_SIZE:
 3521                 *ap->a_retval = -1; /* means ``unlimited'' */
 3522                 break;
 3523         case _PC_REC_MIN_XFER_SIZE:
 3524                 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
 3525                 break;
 3526         case _PC_REC_XFER_ALIGN:
 3527                 *ap->a_retval = PAGE_SIZE;
 3528                 break;
 3529         case _PC_SYMLINK_MAX:
 3530                 *ap->a_retval = NFS_MAXPATHLEN;
 3531                 break;
 3532 
 3533         default:
 3534                 error = EINVAL;
 3535                 break;
 3536         }
 3537         return (error);
 3538 }
 3539 

Cache object: 1273ec92185f4ed81dfa1f4983a206be


[ 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.