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_clport.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  * SPDX-License-Identifier: BSD-3-Clause
    3  *
    4  * Copyright (c) 1989, 1993
    5  *      The Regents of the University of California.  All rights reserved.
    6  *
    7  * This code is derived from software contributed to Berkeley by
    8  * Rick Macklem at The University of Guelph.
    9  *
   10  * Redistribution and use in source and binary forms, with or without
   11  * modification, are permitted provided that the following conditions
   12  * are met:
   13  * 1. Redistributions of source code must retain the above copyright
   14  *    notice, this list of conditions and the following disclaimer.
   15  * 2. Redistributions in binary form must reproduce the above copyright
   16  *    notice, this list of conditions and the following disclaimer in the
   17  *    documentation and/or other materials provided with the distribution.
   18  * 3. Neither the name of the University nor the names of its contributors
   19  *    may be used to endorse or promote products derived from this software
   20  *    without specific prior written permission.
   21  *
   22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   32  * SUCH DAMAGE.
   33  *
   34  */
   35 
   36 #include <sys/cdefs.h>
   37 __FBSDID("$FreeBSD: stable/12/sys/fs/nfsclient/nfs_clport.c 370177 2021-07-23 11:53:47Z markj $");
   38 
   39 #include "opt_inet.h"
   40 #include "opt_inet6.h"
   41 
   42 #include <sys/capsicum.h>
   43 
   44 /*
   45  * generally, I don't like #includes inside .h files, but it seems to
   46  * be the easiest way to handle the port.
   47  */
   48 #include <sys/fail.h>
   49 #include <sys/hash.h>
   50 #include <sys/sysctl.h>
   51 #include <fs/nfs/nfsport.h>
   52 #include <netinet/in_fib.h>
   53 #include <netinet/if_ether.h>
   54 #include <netinet6/ip6_var.h>
   55 #include <net/if_types.h>
   56 
   57 #include <fs/nfsclient/nfs_kdtrace.h>
   58 
   59 #ifdef KDTRACE_HOOKS
   60 dtrace_nfsclient_attrcache_flush_probe_func_t
   61                 dtrace_nfscl_attrcache_flush_done_probe;
   62 uint32_t        nfscl_attrcache_flush_done_id;
   63 
   64 dtrace_nfsclient_attrcache_get_hit_probe_func_t
   65                 dtrace_nfscl_attrcache_get_hit_probe;
   66 uint32_t        nfscl_attrcache_get_hit_id;
   67 
   68 dtrace_nfsclient_attrcache_get_miss_probe_func_t
   69                 dtrace_nfscl_attrcache_get_miss_probe;
   70 uint32_t        nfscl_attrcache_get_miss_id;
   71 
   72 dtrace_nfsclient_attrcache_load_probe_func_t
   73                 dtrace_nfscl_attrcache_load_done_probe;
   74 uint32_t        nfscl_attrcache_load_done_id;
   75 #endif /* !KDTRACE_HOOKS */
   76 
   77 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
   78 extern struct vop_vector newnfs_vnodeops;
   79 extern struct vop_vector newnfs_fifoops;
   80 extern uma_zone_t newnfsnode_zone;
   81 extern struct buf_ops buf_ops_newnfs;
   82 extern int ncl_pbuf_freecnt;
   83 extern short nfsv4_cbport;
   84 extern int nfscl_enablecallb;
   85 extern int nfs_numnfscbd;
   86 extern int nfscl_inited;
   87 struct mtx ncl_iod_mutex;
   88 NFSDLOCKMUTEX;
   89 extern struct mtx nfsrv_dslock_mtx;
   90 
   91 extern void (*ncl_call_invalcaches)(struct vnode *);
   92 
   93 SYSCTL_DECL(_vfs_nfs);
   94 static int ncl_fileid_maxwarnings = 10;
   95 SYSCTL_INT(_vfs_nfs, OID_AUTO, fileid_maxwarnings, CTLFLAG_RWTUN,
   96     &ncl_fileid_maxwarnings, 0,
   97     "Limit fileid corruption warnings; 0 is off; -1 is unlimited");
   98 static volatile int ncl_fileid_nwarnings;
   99 
  100 static void nfscl_warn_fileid(struct nfsmount *, struct nfsvattr *,
  101     struct nfsvattr *);
  102 
  103 /*
  104  * Comparison function for vfs_hash functions.
  105  */
  106 int
  107 newnfs_vncmpf(struct vnode *vp, void *arg)
  108 {
  109         struct nfsfh *nfhp = (struct nfsfh *)arg;
  110         struct nfsnode *np = VTONFS(vp);
  111 
  112         if (np->n_fhp->nfh_len != nfhp->nfh_len ||
  113             NFSBCMP(np->n_fhp->nfh_fh, nfhp->nfh_fh, nfhp->nfh_len))
  114                 return (1);
  115         return (0);
  116 }
  117 
  118 /*
  119  * Look up a vnode/nfsnode by file handle.
  120  * Callers must check for mount points!!
  121  * In all cases, a pointer to a
  122  * nfsnode structure is returned.
  123  * This variant takes a "struct nfsfh *" as second argument and uses
  124  * that structure up, either by hanging off the nfsnode or FREEing it.
  125  */
  126 int
  127 nfscl_nget(struct mount *mntp, struct vnode *dvp, struct nfsfh *nfhp,
  128     struct componentname *cnp, struct thread *td, struct nfsnode **npp,
  129     void *stuff, int lkflags)
  130 {
  131         struct nfsnode *np, *dnp;
  132         struct vnode *vp, *nvp;
  133         struct nfsv4node *newd, *oldd;
  134         int error;
  135         u_int hash;
  136         struct nfsmount *nmp;
  137 
  138         nmp = VFSTONFS(mntp);
  139         dnp = VTONFS(dvp);
  140         *npp = NULL;
  141 
  142         hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, FNV1_32_INIT);
  143 
  144         error = vfs_hash_get(mntp, hash, lkflags,
  145             td, &nvp, newnfs_vncmpf, nfhp);
  146         if (error == 0 && nvp != NULL) {
  147                 /*
  148                  * I believe there is a slight chance that vgonel() could
  149                  * get called on this vnode between when NFSVOPLOCK() drops
  150                  * the VI_LOCK() and vget() acquires it again, so that it
  151                  * hasn't yet had v_usecount incremented. If this were to
  152                  * happen, the VI_DOOMED flag would be set, so check for
  153                  * that here. Since we now have the v_usecount incremented,
  154                  * we should be ok until we vrele() it, if the VI_DOOMED
  155                  * flag isn't set now.
  156                  */
  157                 VI_LOCK(nvp);
  158                 if ((nvp->v_iflag & VI_DOOMED)) {
  159                         VI_UNLOCK(nvp);
  160                         vrele(nvp);
  161                         error = ENOENT;
  162                 } else {
  163                         VI_UNLOCK(nvp);
  164                 }
  165         }
  166         if (error) {
  167                 free(nfhp, M_NFSFH);
  168                 return (error);
  169         }
  170         if (nvp != NULL) {
  171                 np = VTONFS(nvp);
  172                 /*
  173                  * For NFSv4, check to see if it is the same name and
  174                  * replace the name, if it is different.
  175                  */
  176                 oldd = newd = NULL;
  177                 if ((nmp->nm_flag & NFSMNT_NFSV4) && np->n_v4 != NULL &&
  178                     nvp->v_type == VREG &&
  179                     (np->n_v4->n4_namelen != cnp->cn_namelen ||
  180                      NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
  181                      cnp->cn_namelen) ||
  182                      dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
  183                      NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
  184                      dnp->n_fhp->nfh_len))) {
  185                     newd = malloc(
  186                         sizeof (struct nfsv4node) + dnp->n_fhp->nfh_len +
  187                         + cnp->cn_namelen - 1, M_NFSV4NODE, M_WAITOK);
  188                     NFSLOCKNODE(np);
  189                     if (newd != NULL && np->n_v4 != NULL && nvp->v_type == VREG
  190                         && (np->n_v4->n4_namelen != cnp->cn_namelen ||
  191                          NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
  192                          cnp->cn_namelen) ||
  193                          dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
  194                          NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
  195                          dnp->n_fhp->nfh_len))) {
  196                         oldd = np->n_v4;
  197                         np->n_v4 = newd;
  198                         newd = NULL;
  199                         np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
  200                         np->n_v4->n4_namelen = cnp->cn_namelen;
  201                         NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
  202                             dnp->n_fhp->nfh_len);
  203                         NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
  204                             cnp->cn_namelen);
  205                     }
  206                     NFSUNLOCKNODE(np);
  207                 }
  208                 if (newd != NULL)
  209                         free(newd, M_NFSV4NODE);
  210                 if (oldd != NULL)
  211                         free(oldd, M_NFSV4NODE);
  212                 *npp = np;
  213                 free(nfhp, M_NFSFH);
  214                 return (0);
  215         }
  216         np = uma_zalloc(newnfsnode_zone, M_WAITOK | M_ZERO);
  217 
  218         error = getnewvnode(nfs_vnode_tag, mntp, &newnfs_vnodeops, &nvp);
  219         if (error) {
  220                 uma_zfree(newnfsnode_zone, np);
  221                 free(nfhp, M_NFSFH);
  222                 return (error);
  223         }
  224         vp = nvp;
  225         KASSERT(vp->v_bufobj.bo_bsize != 0, ("nfscl_nget: bo_bsize == 0"));
  226         vp->v_bufobj.bo_ops = &buf_ops_newnfs;
  227         vp->v_data = np;
  228         np->n_vnode = vp;
  229         /* 
  230          * Initialize the mutex even if the vnode is going to be a loser.
  231          * This simplifies the logic in reclaim, which can then unconditionally
  232          * destroy the mutex (in the case of the loser, or if hash_insert
  233          * happened to return an error no special casing is needed).
  234          */
  235         mtx_init(&np->n_mtx, "NEWNFSnode lock", NULL, MTX_DEF | MTX_DUPOK);
  236         lockinit(&np->n_excl, PVFS, "nfsupg", VLKTIMEOUT, LK_NOSHARE |
  237             LK_CANRECURSE);
  238 
  239         /* 
  240          * Are we getting the root? If so, make sure the vnode flags
  241          * are correct 
  242          */
  243         if ((nfhp->nfh_len == nmp->nm_fhsize) &&
  244             !bcmp(nfhp->nfh_fh, nmp->nm_fh, nfhp->nfh_len)) {
  245                 if (vp->v_type == VNON)
  246                         vp->v_type = VDIR;
  247                 vp->v_vflag |= VV_ROOT;
  248         }
  249 
  250         vp->v_vflag |= VV_VMSIZEVNLOCK;
  251         
  252         np->n_fhp = nfhp;
  253         /*
  254          * For NFSv4, we have to attach the directory file handle and
  255          * file name, so that Open Ops can be done later.
  256          */
  257         if (nmp->nm_flag & NFSMNT_NFSV4) {
  258                 np->n_v4 = malloc(sizeof (struct nfsv4node)
  259                     + dnp->n_fhp->nfh_len + cnp->cn_namelen - 1, M_NFSV4NODE,
  260                     M_WAITOK);
  261                 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
  262                 np->n_v4->n4_namelen = cnp->cn_namelen;
  263                 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
  264                     dnp->n_fhp->nfh_len);
  265                 NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
  266                     cnp->cn_namelen);
  267         } else {
  268                 np->n_v4 = NULL;
  269         }
  270 
  271         /*
  272          * NFS supports recursive and shared locking.
  273          */
  274         lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL);
  275         VN_LOCK_AREC(vp);
  276         VN_LOCK_ASHARE(vp);
  277         error = insmntque(vp, mntp);
  278         if (error != 0) {
  279                 *npp = NULL;
  280                 mtx_destroy(&np->n_mtx);
  281                 lockdestroy(&np->n_excl);
  282                 free(nfhp, M_NFSFH);
  283                 if (np->n_v4 != NULL)
  284                         free(np->n_v4, M_NFSV4NODE);
  285                 uma_zfree(newnfsnode_zone, np);
  286                 return (error);
  287         }
  288         error = vfs_hash_insert(vp, hash, lkflags, 
  289             td, &nvp, newnfs_vncmpf, nfhp);
  290         if (error)
  291                 return (error);
  292         if (nvp != NULL) {
  293                 *npp = VTONFS(nvp);
  294                 /* vfs_hash_insert() vput()'s the losing vnode */
  295                 return (0);
  296         }
  297         *npp = np;
  298 
  299         return (0);
  300 }
  301 
  302 /*
  303  * Another variant of nfs_nget(). This one is only used by reopen. It
  304  * takes almost the same args as nfs_nget(), but only succeeds if an entry
  305  * exists in the cache. (Since files should already be "open" with a
  306  * vnode ref cnt on the node when reopen calls this, it should always
  307  * succeed.)
  308  * Also, don't get a vnode lock, since it may already be locked by some
  309  * other process that is handling it. This is ok, since all other threads
  310  * on the client are blocked by the nfsc_lock being exclusively held by the
  311  * caller of this function.
  312  */
  313 int
  314 nfscl_ngetreopen(struct mount *mntp, u_int8_t *fhp, int fhsize,
  315     struct thread *td, struct nfsnode **npp)
  316 {
  317         struct vnode *nvp;
  318         u_int hash;
  319         struct nfsfh *nfhp;
  320         int error;
  321 
  322         *npp = NULL;
  323         /* For forced dismounts, just return error. */
  324         if (NFSCL_FORCEDISM(mntp))
  325                 return (EINTR);
  326         nfhp = malloc(sizeof (struct nfsfh) + fhsize,
  327             M_NFSFH, M_WAITOK);
  328         bcopy(fhp, &nfhp->nfh_fh[0], fhsize);
  329         nfhp->nfh_len = fhsize;
  330 
  331         hash = fnv_32_buf(fhp, fhsize, FNV1_32_INIT);
  332 
  333         /*
  334          * First, try to get the vnode locked, but don't block for the lock.
  335          */
  336         error = vfs_hash_get(mntp, hash, (LK_EXCLUSIVE | LK_NOWAIT), td, &nvp,
  337             newnfs_vncmpf, nfhp);
  338         if (error == 0 && nvp != NULL) {
  339                 NFSVOPUNLOCK(nvp, 0);
  340         } else if (error == EBUSY) {
  341                 /*
  342                  * It is safe so long as a vflush() with
  343                  * FORCECLOSE has not been done. Since the Renew thread is
  344                  * stopped and the MNTK_UNMOUNTF flag is set before doing
  345                  * a vflush() with FORCECLOSE, we should be ok here.
  346                  */
  347                 if (NFSCL_FORCEDISM(mntp))
  348                         error = EINTR;
  349                 else {
  350                         vfs_hash_ref(mntp, hash, td, &nvp, newnfs_vncmpf, nfhp);
  351                         if (nvp == NULL) {
  352                                 error = ENOENT;
  353                         } else if ((nvp->v_iflag & VI_DOOMED) != 0) {
  354                                 error = ENOENT;
  355                                 vrele(nvp);
  356                         } else {
  357                                 error = 0;
  358                         }
  359                 }
  360         }
  361         free(nfhp, M_NFSFH);
  362         if (error)
  363                 return (error);
  364         if (nvp != NULL) {
  365                 *npp = VTONFS(nvp);
  366                 return (0);
  367         }
  368         return (EINVAL);
  369 }
  370 
  371 static void
  372 nfscl_warn_fileid(struct nfsmount *nmp, struct nfsvattr *oldnap,
  373     struct nfsvattr *newnap)
  374 {
  375         int off;
  376 
  377         if (ncl_fileid_maxwarnings >= 0 &&
  378             ncl_fileid_nwarnings >= ncl_fileid_maxwarnings)
  379                 return;
  380         off = 0;
  381         if (ncl_fileid_maxwarnings >= 0) {
  382                 if (++ncl_fileid_nwarnings >= ncl_fileid_maxwarnings)
  383                         off = 1;
  384         }
  385 
  386         printf("newnfs: server '%s' error: fileid changed. "
  387             "fsid %jx:%jx: expected fileid %#jx, got %#jx. "
  388             "(BROKEN NFS SERVER OR MIDDLEWARE)\n",
  389             nmp->nm_com.nmcom_hostname,
  390             (uintmax_t)nmp->nm_fsid[0],
  391             (uintmax_t)nmp->nm_fsid[1],
  392             (uintmax_t)oldnap->na_fileid,
  393             (uintmax_t)newnap->na_fileid);
  394 
  395         if (off)
  396                 printf("newnfs: Logged %d times about fileid corruption; "
  397                     "going quiet to avoid spamming logs excessively. (Limit "
  398                     "is: %d).\n", ncl_fileid_nwarnings,
  399                     ncl_fileid_maxwarnings);
  400 }
  401 
  402 void
  403 ncl_copy_vattr(struct vattr *dst, struct vattr *src)
  404 {
  405         dst->va_type = src->va_type;
  406         dst->va_mode = src->va_mode;
  407         dst->va_nlink = src->va_nlink;
  408         dst->va_uid = src->va_uid;
  409         dst->va_gid = src->va_gid;
  410         dst->va_fsid = src->va_fsid;
  411         dst->va_fileid = src->va_fileid;
  412         dst->va_size = src->va_size;
  413         dst->va_blocksize = src->va_blocksize;
  414         dst->va_atime = src->va_atime;
  415         dst->va_mtime = src->va_mtime;
  416         dst->va_ctime = src->va_ctime;
  417         dst->va_gen = src->va_gen;
  418         dst->va_flags = src->va_flags;
  419         dst->va_rdev = src->va_rdev;
  420         dst->va_bytes = src->va_bytes;
  421         dst->va_filerev = src->va_filerev;
  422 }
  423 
  424 /*
  425  * Load the attribute cache (that lives in the nfsnode entry) with
  426  * the attributes of the second argument and
  427  * Iff vaper not NULL
  428  *    copy the attributes to *vaper
  429  * Similar to nfs_loadattrcache(), except the attributes are passed in
  430  * instead of being parsed out of the mbuf list.
  431  */
  432 int
  433 nfscl_loadattrcache(struct vnode **vpp, struct nfsvattr *nap, void *nvaper,
  434     void *stuff, int writeattr, int dontshrink)
  435 {
  436         struct vnode *vp = *vpp;
  437         struct vattr *vap, *nvap = &nap->na_vattr, *vaper = nvaper;
  438         struct nfsnode *np;
  439         struct nfsmount *nmp;
  440         struct timespec mtime_save;
  441         int error, force_fid_err;
  442         dev_t topfsid;
  443 
  444         error = 0;
  445 
  446         /*
  447          * If v_type == VNON it is a new node, so fill in the v_type,
  448          * n_mtime fields. Check to see if it represents a special 
  449          * device, and if so, check for a possible alias. Once the
  450          * correct vnode has been obtained, fill in the rest of the
  451          * information.
  452          */
  453         np = VTONFS(vp);
  454         NFSLOCKNODE(np);
  455         if (vp->v_type != nvap->va_type) {
  456                 vp->v_type = nvap->va_type;
  457                 if (vp->v_type == VFIFO)
  458                         vp->v_op = &newnfs_fifoops;
  459                 np->n_mtime = nvap->va_mtime;
  460         }
  461         nmp = VFSTONFS(vp->v_mount);
  462         vap = &np->n_vattr.na_vattr;
  463         mtime_save = vap->va_mtime;
  464         if (writeattr) {
  465                 np->n_vattr.na_filerev = nap->na_filerev;
  466                 np->n_vattr.na_size = nap->na_size;
  467                 np->n_vattr.na_mtime = nap->na_mtime;
  468                 np->n_vattr.na_ctime = nap->na_ctime;
  469                 np->n_vattr.na_fsid = nap->na_fsid;
  470                 np->n_vattr.na_mode = nap->na_mode;
  471         } else {
  472                 force_fid_err = 0;
  473                 KFAIL_POINT_ERROR(DEBUG_FP, nfscl_force_fileid_warning,
  474                     force_fid_err);
  475                 /*
  476                  * BROKEN NFS SERVER OR MIDDLEWARE
  477                  *
  478                  * Certain NFS servers (certain old proprietary filers ca.
  479                  * 2006) or broken middleboxes (e.g. WAN accelerator products)
  480                  * will respond to GETATTR requests with results for a
  481                  * different fileid.
  482                  *
  483                  * The WAN accelerator we've observed not only serves stale
  484                  * cache results for a given file, it also occasionally serves
  485                  * results for wholly different files.  This causes surprising
  486                  * problems; for example the cached size attribute of a file
  487                  * may truncate down and then back up, resulting in zero
  488                  * regions in file contents read by applications.  We observed
  489                  * this reliably with Clang and .c files during parallel build.
  490                  * A pcap revealed packet fragmentation and GETATTR RPC
  491                  * responses with wholly wrong fileids.
  492                  */
  493                 if ((np->n_vattr.na_fileid != 0 &&
  494                      np->n_vattr.na_fileid != nap->na_fileid) ||
  495                     force_fid_err) {
  496                         nfscl_warn_fileid(nmp, &np->n_vattr, nap);
  497                         error = EIDRM;
  498                         goto out;
  499                 }
  500                 NFSBCOPY((caddr_t)nap, (caddr_t)&np->n_vattr,
  501                     sizeof (struct nfsvattr));
  502         }
  503 
  504         /*
  505          * For NFSv4, the server's export may be a tree of file systems
  506          * where a fileno is a unique value within each file system.
  507          * na_filesid[0,1] uniquely identify the server file system
  508          * and nm_fsid[0,1] is the value for the root file system mounted.
  509          * As such, the value of va_fsid generated by vn_fsid() represents
  510          * the root file system on the server and a different value for
  511          * va_fsid is needed for the other server file systems.  This
  512          * va_fsid is ideally unique for all of the server file systems,
  513          * so a 64bit hash on na_filesid[0,1] is calculated.
  514          * Although highly unlikely that the fnv_64_hash() will be
  515          * the same as the root, test for this case and recalculate the hash.
  516          */
  517         vn_fsid(vp, vap);
  518         if (NFSHASNFSV4(nmp) && NFSHASHASSETFSID(nmp) &&
  519             (nmp->nm_fsid[0] != np->n_vattr.na_filesid[0] ||
  520              nmp->nm_fsid[1] != np->n_vattr.na_filesid[1])) {
  521                 topfsid = vap->va_fsid;
  522                 vap->va_fsid = FNV1_64_INIT;
  523                 do {
  524                         vap->va_fsid = fnv_64_buf(np->n_vattr.na_filesid,
  525                             sizeof(np->n_vattr.na_filesid), vap->va_fsid);
  526                 } while (vap->va_fsid == topfsid);
  527         }
  528 
  529         np->n_attrstamp = time_second;
  530         if (vap->va_size != np->n_size) {
  531                 if (vap->va_type == VREG) {
  532                         if (dontshrink && vap->va_size < np->n_size) {
  533                                 /*
  534                                  * We've been told not to shrink the file;
  535                                  * zero np->n_attrstamp to indicate that
  536                                  * the attributes are stale.
  537                                  */
  538                                 vap->va_size = np->n_size;
  539                                 np->n_attrstamp = 0;
  540                                 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
  541                         } else if (np->n_flag & NMODIFIED) {
  542                                 /*
  543                                  * We've modified the file: Use the larger
  544                                  * of our size, and the server's size.
  545                                  */
  546                                 if (vap->va_size < np->n_size) {
  547                                         vap->va_size = np->n_size;
  548                                 } else {
  549                                         np->n_size = vap->va_size;
  550                                         np->n_flag |= NSIZECHANGED;
  551                                 }
  552                         } else {
  553                                 np->n_size = vap->va_size;
  554                                 np->n_flag |= NSIZECHANGED;
  555                         }
  556                 } else {
  557                         np->n_size = vap->va_size;
  558                 }
  559         }
  560         /*
  561          * The following checks are added to prevent a race between (say)
  562          * a READDIR+ and a WRITE. 
  563          * READDIR+, WRITE requests sent out.
  564          * READDIR+ resp, WRITE resp received on client.
  565          * However, the WRITE resp was handled before the READDIR+ resp
  566          * causing the post op attrs from the write to be loaded first
  567          * and the attrs from the READDIR+ to be loaded later. If this 
  568          * happens, we have stale attrs loaded into the attrcache.
  569          * We detect this by for the mtime moving back. We invalidate the 
  570          * attrcache when this happens.
  571          */
  572         if (timespeccmp(&mtime_save, &vap->va_mtime, >)) {
  573                 /* Size changed or mtime went backwards */
  574                 np->n_attrstamp = 0;
  575                 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
  576         }
  577         if (vaper != NULL) {
  578                 ncl_copy_vattr(vaper, vap);
  579                 if (np->n_flag & NCHG) {
  580                         if (np->n_flag & NACC)
  581                                 vaper->va_atime = np->n_atim;
  582                         if (np->n_flag & NUPD)
  583                                 vaper->va_mtime = np->n_mtim;
  584                 }
  585         }
  586 
  587 out:
  588 #ifdef KDTRACE_HOOKS
  589         if (np->n_attrstamp != 0)
  590                 KDTRACE_NFS_ATTRCACHE_LOAD_DONE(vp, vap, error);
  591 #endif
  592         (void)ncl_pager_setsize(vp, NULL);
  593         return (error);
  594 }
  595 
  596 /*
  597  * Call vnode_pager_setsize() if the size of the node changed, as
  598  * recorded in nfsnode vs. v_object, or delay the call if notifying
  599  * the pager is not possible at the moment.
  600  *
  601  * If nsizep is non-NULL, the call is delayed and the new node size is
  602  * provided.  Caller should itself call vnode_pager_setsize() if
  603  * function returned true.  If nsizep is NULL, function tries to call
  604  * vnode_pager_setsize() itself if needed and possible, and the nfs
  605  * node is unlocked unconditionally, the return value is not useful.
  606  */
  607 bool
  608 ncl_pager_setsize(struct vnode *vp, u_quad_t *nsizep)
  609 {
  610         struct nfsnode *np;
  611         vm_object_t object;
  612         struct vattr *vap;
  613         u_quad_t nsize;
  614         bool setnsize;
  615 
  616         np = VTONFS(vp);
  617         NFSASSERTNODE(np);
  618 
  619         vap = &np->n_vattr.na_vattr;
  620         nsize = vap->va_size;
  621         object = vp->v_object;
  622         setnsize = false;
  623 
  624         if (object != NULL && nsize != object->un_pager.vnp.vnp_size) {
  625                 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
  626                     (curthread->td_pflags2 & TDP2_SBPAGES) == 0)
  627                         setnsize = true;
  628                 else
  629                         np->n_flag |= NVNSETSZSKIP;
  630         }
  631         if (nsizep == NULL) {
  632                 NFSUNLOCKNODE(np);
  633                 if (setnsize)
  634                         vnode_pager_setsize(vp, nsize);
  635                 setnsize = false;
  636         } else {
  637                 *nsizep = nsize;
  638         }
  639         return (setnsize);
  640 }
  641 
  642 /*
  643  * Fill in the client id name. For these bytes:
  644  * 1 - they must be unique
  645  * 2 - they should be persistent across client reboots
  646  * 1 is more critical than 2
  647  * Use the mount point's unique id plus either the uuid or, if that
  648  * isn't set, random junk.
  649  */
  650 void
  651 nfscl_fillclid(u_int64_t clval, char *uuid, u_int8_t *cp, u_int16_t idlen)
  652 {
  653         int uuidlen;
  654 
  655         /*
  656          * First, put in the 64bit mount point identifier.
  657          */
  658         if (idlen >= sizeof (u_int64_t)) {
  659                 NFSBCOPY((caddr_t)&clval, cp, sizeof (u_int64_t));
  660                 cp += sizeof (u_int64_t);
  661                 idlen -= sizeof (u_int64_t);
  662         }
  663 
  664         /*
  665          * If uuid is non-zero length, use it.
  666          */
  667         uuidlen = strlen(uuid);
  668         if (uuidlen > 0 && idlen >= uuidlen) {
  669                 NFSBCOPY(uuid, cp, uuidlen);
  670                 cp += uuidlen;
  671                 idlen -= uuidlen;
  672         }
  673 
  674         /*
  675          * This only normally happens if the uuid isn't set.
  676          */
  677         while (idlen > 0) {
  678                 *cp++ = (u_int8_t)(arc4random() % 256);
  679                 idlen--;
  680         }
  681 }
  682 
  683 /*
  684  * Fill in a lock owner name. For now, pid + the process's creation time.
  685  */
  686 void
  687 nfscl_filllockowner(void *id, u_int8_t *cp, int flags)
  688 {
  689         union {
  690                 u_int32_t       lval;
  691                 u_int8_t        cval[4];
  692         } tl;
  693         struct proc *p;
  694 
  695         if (id == NULL) {
  696                 /* Return the single open_owner of all 0 bytes. */
  697                 bzero(cp, NFSV4CL_LOCKNAMELEN);
  698                 return;
  699         }
  700         if ((flags & F_POSIX) != 0) {
  701                 p = (struct proc *)id;
  702                 tl.lval = p->p_pid;
  703                 *cp++ = tl.cval[0];
  704                 *cp++ = tl.cval[1];
  705                 *cp++ = tl.cval[2];
  706                 *cp++ = tl.cval[3];
  707                 tl.lval = p->p_stats->p_start.tv_sec;
  708                 *cp++ = tl.cval[0];
  709                 *cp++ = tl.cval[1];
  710                 *cp++ = tl.cval[2];
  711                 *cp++ = tl.cval[3];
  712                 tl.lval = p->p_stats->p_start.tv_usec;
  713                 *cp++ = tl.cval[0];
  714                 *cp++ = tl.cval[1];
  715                 *cp++ = tl.cval[2];
  716                 *cp = tl.cval[3];
  717         } else if ((flags & F_FLOCK) != 0) {
  718                 bcopy(&id, cp, sizeof(id));
  719                 bzero(&cp[sizeof(id)], NFSV4CL_LOCKNAMELEN - sizeof(id));
  720         } else {
  721                 printf("nfscl_filllockowner: not F_POSIX or F_FLOCK\n");
  722                 bzero(cp, NFSV4CL_LOCKNAMELEN);
  723         }
  724 }
  725 
  726 /*
  727  * Find the parent process for the thread passed in as an argument.
  728  * If none exists, return NULL, otherwise return a thread for the parent.
  729  * (Can be any of the threads, since it is only used for td->td_proc.)
  730  */
  731 NFSPROC_T *
  732 nfscl_getparent(struct thread *td)
  733 {
  734         struct proc *p;
  735         struct thread *ptd;
  736 
  737         if (td == NULL)
  738                 return (NULL);
  739         p = td->td_proc;
  740         if (p->p_pid == 0)
  741                 return (NULL);
  742         p = p->p_pptr;
  743         if (p == NULL)
  744                 return (NULL);
  745         ptd = TAILQ_FIRST(&p->p_threads);
  746         return (ptd);
  747 }
  748 
  749 /*
  750  * Start up the renew kernel thread.
  751  */
  752 static void
  753 start_nfscl(void *arg)
  754 {
  755         struct nfsclclient *clp;
  756         struct thread *td;
  757 
  758         clp = (struct nfsclclient *)arg;
  759         td = TAILQ_FIRST(&clp->nfsc_renewthread->p_threads);
  760         nfscl_renewthread(clp, td);
  761         kproc_exit(0);
  762 }
  763 
  764 void
  765 nfscl_start_renewthread(struct nfsclclient *clp)
  766 {
  767 
  768         kproc_create(start_nfscl, (void *)clp, &clp->nfsc_renewthread, 0, 0,
  769             "nfscl");
  770 }
  771 
  772 /*
  773  * Handle wcc_data.
  774  * For NFSv4, it assumes that nfsv4_wccattr() was used to set up the getattr
  775  * as the first Op after PutFH.
  776  * (For NFSv4, the postop attributes are after the Op, so they can't be
  777  *  parsed here. A separate call to nfscl_postop_attr() is required.)
  778  */
  779 int
  780 nfscl_wcc_data(struct nfsrv_descript *nd, struct vnode *vp,
  781     struct nfsvattr *nap, int *flagp, int *wccflagp, void *stuff)
  782 {
  783         u_int32_t *tl;
  784         struct nfsnode *np = VTONFS(vp);
  785         struct nfsvattr nfsva;
  786         int error = 0;
  787 
  788         if (wccflagp != NULL)
  789                 *wccflagp = 0;
  790         if (nd->nd_flag & ND_NFSV3) {
  791                 *flagp = 0;
  792                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
  793                 if (*tl == newnfs_true) {
  794                         NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
  795                         if (wccflagp != NULL) {
  796                                 NFSLOCKNODE(np);
  797                                 *wccflagp = (np->n_mtime.tv_sec ==
  798                                     fxdr_unsigned(u_int32_t, *(tl + 2)) &&
  799                                     np->n_mtime.tv_nsec ==
  800                                     fxdr_unsigned(u_int32_t, *(tl + 3)));
  801                                 NFSUNLOCKNODE(np);
  802                         }
  803                 }
  804                 error = nfscl_postop_attr(nd, nap, flagp, stuff);
  805                 if (wccflagp != NULL && *flagp == 0)
  806                         *wccflagp = 0;
  807         } else if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR))
  808             == (ND_NFSV4 | ND_V4WCCATTR)) {
  809                 error = nfsv4_loadattr(nd, NULL, &nfsva, NULL,
  810                     NULL, 0, NULL, NULL, NULL, NULL, NULL, 0,
  811                     NULL, NULL, NULL, NULL, NULL);
  812                 if (error)
  813                         return (error);
  814                 /*
  815                  * Get rid of Op# and status for next op.
  816                  */
  817                 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
  818                 if (*++tl)
  819                         nd->nd_flag |= ND_NOMOREDATA;
  820                 if (wccflagp != NULL &&
  821                     nfsva.na_vattr.va_mtime.tv_sec != 0) {
  822                         NFSLOCKNODE(np);
  823                         *wccflagp = (np->n_mtime.tv_sec ==
  824                             nfsva.na_vattr.va_mtime.tv_sec &&
  825                             np->n_mtime.tv_nsec ==
  826                             nfsva.na_vattr.va_mtime.tv_sec);
  827                         NFSUNLOCKNODE(np);
  828                 }
  829         }
  830 nfsmout:
  831         return (error);
  832 }
  833 
  834 /*
  835  * Get postop attributes.
  836  */
  837 int
  838 nfscl_postop_attr(struct nfsrv_descript *nd, struct nfsvattr *nap, int *retp,
  839     void *stuff)
  840 {
  841         u_int32_t *tl;
  842         int error = 0;
  843 
  844         *retp = 0;
  845         if (nd->nd_flag & ND_NOMOREDATA)
  846                 return (error);
  847         if (nd->nd_flag & ND_NFSV3) {
  848                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
  849                 *retp = fxdr_unsigned(int, *tl);
  850         } else if (nd->nd_flag & ND_NFSV4) {
  851                 /*
  852                  * For NFSv4, the postop attr are at the end, so no point
  853                  * in looking if nd_repstat != 0.
  854                  */
  855                 if (!nd->nd_repstat) {
  856                         NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
  857                         if (*(tl + 1))
  858                                 /* should never happen since nd_repstat != 0 */
  859                                 nd->nd_flag |= ND_NOMOREDATA;
  860                         else
  861                                 *retp = 1;
  862                 }
  863         } else if (!nd->nd_repstat) {
  864                 /* For NFSv2, the attributes are here iff nd_repstat == 0 */
  865                 *retp = 1;
  866         }
  867         if (*retp) {
  868                 error = nfsm_loadattr(nd, nap);
  869                 if (error)
  870                         *retp = 0;
  871         }
  872 nfsmout:
  873         return (error);
  874 }
  875 
  876 /*
  877  * nfscl_request() - mostly a wrapper for newnfs_request().
  878  */
  879 int
  880 nfscl_request(struct nfsrv_descript *nd, struct vnode *vp, NFSPROC_T *p,
  881     struct ucred *cred, void *stuff)
  882 {
  883         int ret, vers;
  884         struct nfsmount *nmp;
  885 
  886         nmp = VFSTONFS(vp->v_mount);
  887         if (nd->nd_flag & ND_NFSV4)
  888                 vers = NFS_VER4;
  889         else if (nd->nd_flag & ND_NFSV3)
  890                 vers = NFS_VER3;
  891         else
  892                 vers = NFS_VER2;
  893         ret = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred,
  894                 NFS_PROG, vers, NULL, 1, NULL, NULL);
  895         return (ret);
  896 }
  897 
  898 /*
  899  * fill in this bsden's variant of statfs using nfsstatfs.
  900  */
  901 void
  902 nfscl_loadsbinfo(struct nfsmount *nmp, struct nfsstatfs *sfp, void *statfs)
  903 {
  904         struct statfs *sbp = (struct statfs *)statfs;
  905 
  906         if (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) {
  907                 sbp->f_bsize = NFS_FABLKSIZE;
  908                 sbp->f_blocks = sfp->sf_tbytes / NFS_FABLKSIZE;
  909                 sbp->f_bfree = sfp->sf_fbytes / NFS_FABLKSIZE;
  910                 /*
  911                  * Although sf_abytes is uint64_t and f_bavail is int64_t,
  912                  * the value after dividing by NFS_FABLKSIZE is small
  913                  * enough that it will fit in 63bits, so it is ok to
  914                  * assign it to f_bavail without fear that it will become
  915                  * negative.
  916                  */
  917                 sbp->f_bavail = sfp->sf_abytes / NFS_FABLKSIZE;
  918                 sbp->f_files = sfp->sf_tfiles;
  919                 /* Since f_ffree is int64_t, clip it to 63bits. */
  920                 if (sfp->sf_ffiles > INT64_MAX)
  921                         sbp->f_ffree = INT64_MAX;
  922                 else
  923                         sbp->f_ffree = sfp->sf_ffiles;
  924         } else if ((nmp->nm_flag & NFSMNT_NFSV4) == 0) {
  925                 /*
  926                  * The type casts to (int32_t) ensure that this code is
  927                  * compatible with the old NFS client, in that it will
  928                  * propagate bit31 to the high order bits. This may or may
  929                  * not be correct for NFSv2, but since it is a legacy
  930                  * environment, I'd rather retain backwards compatibility.
  931                  */
  932                 sbp->f_bsize = (int32_t)sfp->sf_bsize;
  933                 sbp->f_blocks = (int32_t)sfp->sf_blocks;
  934                 sbp->f_bfree = (int32_t)sfp->sf_bfree;
  935                 sbp->f_bavail = (int32_t)sfp->sf_bavail;
  936                 sbp->f_files = 0;
  937                 sbp->f_ffree = 0;
  938         }
  939 }
  940 
  941 /*
  942  * Use the fsinfo stuff to update the mount point.
  943  */
  944 void
  945 nfscl_loadfsinfo(struct nfsmount *nmp, struct nfsfsinfo *fsp)
  946 {
  947 
  948         if ((nmp->nm_wsize == 0 || fsp->fs_wtpref < nmp->nm_wsize) &&
  949             fsp->fs_wtpref >= NFS_FABLKSIZE)
  950                 nmp->nm_wsize = (fsp->fs_wtpref + NFS_FABLKSIZE - 1) &
  951                     ~(NFS_FABLKSIZE - 1);
  952         if (fsp->fs_wtmax < nmp->nm_wsize && fsp->fs_wtmax > 0) {
  953                 nmp->nm_wsize = fsp->fs_wtmax & ~(NFS_FABLKSIZE - 1);
  954                 if (nmp->nm_wsize == 0)
  955                         nmp->nm_wsize = fsp->fs_wtmax;
  956         }
  957         if (nmp->nm_wsize < NFS_FABLKSIZE)
  958                 nmp->nm_wsize = NFS_FABLKSIZE;
  959         if ((nmp->nm_rsize == 0 || fsp->fs_rtpref < nmp->nm_rsize) &&
  960             fsp->fs_rtpref >= NFS_FABLKSIZE)
  961                 nmp->nm_rsize = (fsp->fs_rtpref + NFS_FABLKSIZE - 1) &
  962                     ~(NFS_FABLKSIZE - 1);
  963         if (fsp->fs_rtmax < nmp->nm_rsize && fsp->fs_rtmax > 0) {
  964                 nmp->nm_rsize = fsp->fs_rtmax & ~(NFS_FABLKSIZE - 1);
  965                 if (nmp->nm_rsize == 0)
  966                         nmp->nm_rsize = fsp->fs_rtmax;
  967         }
  968         if (nmp->nm_rsize < NFS_FABLKSIZE)
  969                 nmp->nm_rsize = NFS_FABLKSIZE;
  970         if ((nmp->nm_readdirsize == 0 || fsp->fs_dtpref < nmp->nm_readdirsize)
  971             && fsp->fs_dtpref >= NFS_DIRBLKSIZ)
  972                 nmp->nm_readdirsize = (fsp->fs_dtpref + NFS_DIRBLKSIZ - 1) &
  973                     ~(NFS_DIRBLKSIZ - 1);
  974         if (fsp->fs_rtmax < nmp->nm_readdirsize && fsp->fs_rtmax > 0) {
  975                 nmp->nm_readdirsize = fsp->fs_rtmax & ~(NFS_DIRBLKSIZ - 1);
  976                 if (nmp->nm_readdirsize == 0)
  977                         nmp->nm_readdirsize = fsp->fs_rtmax;
  978         }
  979         if (nmp->nm_readdirsize < NFS_DIRBLKSIZ)
  980                 nmp->nm_readdirsize = NFS_DIRBLKSIZ;
  981         if (fsp->fs_maxfilesize > 0 &&
  982             fsp->fs_maxfilesize < nmp->nm_maxfilesize)
  983                 nmp->nm_maxfilesize = fsp->fs_maxfilesize;
  984         nmp->nm_mountp->mnt_stat.f_iosize = newnfs_iosize(nmp);
  985         nmp->nm_state |= NFSSTA_GOTFSINFO;
  986 }
  987 
  988 /*
  989  * Lookups source address which should be used to communicate with
  990  * @nmp and stores it inside @pdst.
  991  *
  992  * Returns 0 on success.
  993  */
  994 u_int8_t *
  995 nfscl_getmyip(struct nfsmount *nmp, struct in6_addr *paddr, int *isinet6p)
  996 {
  997 #if defined(INET6) || defined(INET)
  998         int error, fibnum;
  999 
 1000         fibnum = curthread->td_proc->p_fibnum;
 1001 #endif
 1002 #ifdef INET
 1003         if (nmp->nm_nam->sa_family == AF_INET) {
 1004                 struct sockaddr_in *sin;
 1005                 struct nhop4_extended nh_ext;
 1006 
 1007                 sin = (struct sockaddr_in *)nmp->nm_nam;
 1008                 CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
 1009                 error = fib4_lookup_nh_ext(fibnum, sin->sin_addr, 0, 0,
 1010                     &nh_ext);
 1011                 CURVNET_RESTORE();
 1012                 if (error != 0)
 1013                         return (NULL);
 1014 
 1015                 if (IN_LOOPBACK(ntohl(nh_ext.nh_src.s_addr))) {
 1016                         /* Ignore loopback addresses */
 1017                         return (NULL);
 1018                 }
 1019 
 1020                 *isinet6p = 0;
 1021                 *((struct in_addr *)paddr) = nh_ext.nh_src;
 1022 
 1023                 return (u_int8_t *)paddr;
 1024         }
 1025 #endif
 1026 #ifdef INET6
 1027         if (nmp->nm_nam->sa_family == AF_INET6) {
 1028                 struct sockaddr_in6 *sin6;
 1029 
 1030                 sin6 = (struct sockaddr_in6 *)nmp->nm_nam;
 1031 
 1032                 CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
 1033                 error = in6_selectsrc_addr(fibnum, &sin6->sin6_addr,
 1034                     sin6->sin6_scope_id, NULL, paddr, NULL);
 1035                 CURVNET_RESTORE();
 1036                 if (error != 0)
 1037                         return (NULL);
 1038 
 1039                 if (IN6_IS_ADDR_LOOPBACK(paddr))
 1040                         return (NULL);
 1041 
 1042                 /* Scope is embedded in */
 1043                 *isinet6p = 1;
 1044 
 1045                 return (u_int8_t *)paddr;
 1046         }
 1047 #endif
 1048         return (NULL);
 1049 }
 1050 
 1051 /*
 1052  * Copy NFS uid, gids from the cred structure.
 1053  */
 1054 void
 1055 newnfs_copyincred(struct ucred *cr, struct nfscred *nfscr)
 1056 {
 1057         int i;
 1058 
 1059         KASSERT(cr->cr_ngroups >= 0,
 1060             ("newnfs_copyincred: negative cr_ngroups"));
 1061         nfscr->nfsc_uid = cr->cr_uid;
 1062         nfscr->nfsc_ngroups = MIN(cr->cr_ngroups, NFS_MAXGRPS + 1);
 1063         for (i = 0; i < nfscr->nfsc_ngroups; i++)
 1064                 nfscr->nfsc_groups[i] = cr->cr_groups[i];
 1065 }
 1066 
 1067 
 1068 /*
 1069  * Do any client specific initialization.
 1070  */
 1071 void
 1072 nfscl_init(void)
 1073 {
 1074         static int inited = 0;
 1075 
 1076         if (inited)
 1077                 return;
 1078         inited = 1;
 1079         nfscl_inited = 1;
 1080         ncl_pbuf_freecnt = nswbuf / 2 + 1;
 1081 }
 1082 
 1083 /*
 1084  * Check each of the attributes to be set, to ensure they aren't already
 1085  * the correct value. Disable setting ones already correct.
 1086  */
 1087 int
 1088 nfscl_checksattr(struct vattr *vap, struct nfsvattr *nvap)
 1089 {
 1090 
 1091         if (vap->va_mode != (mode_t)VNOVAL) {
 1092                 if (vap->va_mode == nvap->na_mode)
 1093                         vap->va_mode = (mode_t)VNOVAL;
 1094         }
 1095         if (vap->va_uid != (uid_t)VNOVAL) {
 1096                 if (vap->va_uid == nvap->na_uid)
 1097                         vap->va_uid = (uid_t)VNOVAL;
 1098         }
 1099         if (vap->va_gid != (gid_t)VNOVAL) {
 1100                 if (vap->va_gid == nvap->na_gid)
 1101                         vap->va_gid = (gid_t)VNOVAL;
 1102         }
 1103         if (vap->va_size != VNOVAL) {
 1104                 if (vap->va_size == nvap->na_size)
 1105                         vap->va_size = VNOVAL;
 1106         }
 1107 
 1108         /*
 1109          * We are normally called with only a partially initialized
 1110          * VAP.  Since the NFSv3 spec says that server may use the
 1111          * file attributes to store the verifier, the spec requires
 1112          * us to do a SETATTR RPC. FreeBSD servers store the verifier
 1113          * in atime, but we can't really assume that all servers will
 1114          * so we ensure that our SETATTR sets both atime and mtime.
 1115          * Set the VA_UTIMES_NULL flag for this case, so that
 1116          * the server's time will be used.  This is needed to
 1117          * work around a bug in some Solaris servers, where
 1118          * setting the time TOCLIENT causes the Setattr RPC
 1119          * to return NFS_OK, but not set va_mode.
 1120          */
 1121         if (vap->va_mtime.tv_sec == VNOVAL) {
 1122                 vfs_timestamp(&vap->va_mtime);
 1123                 vap->va_vaflags |= VA_UTIMES_NULL;
 1124         }
 1125         if (vap->va_atime.tv_sec == VNOVAL)
 1126                 vap->va_atime = vap->va_mtime;
 1127         return (1);
 1128 }
 1129 
 1130 /*
 1131  * Map nfsv4 errors to errno.h errors.
 1132  * The uid and gid arguments are only used for NFSERR_BADOWNER and that
 1133  * error should only be returned for the Open, Create and Setattr Ops.
 1134  * As such, most calls can just pass in 0 for those arguments.
 1135  */
 1136 int
 1137 nfscl_maperr(struct thread *td, int error, uid_t uid, gid_t gid)
 1138 {
 1139         struct proc *p;
 1140 
 1141         if (error < 10000 || error >= NFSERR_STALEWRITEVERF)
 1142                 return (error);
 1143         if (td != NULL)
 1144                 p = td->td_proc;
 1145         else
 1146                 p = NULL;
 1147         switch (error) {
 1148         case NFSERR_BADOWNER:
 1149                 tprintf(p, LOG_INFO,
 1150                     "No name and/or group mapping for uid,gid:(%d,%d)\n",
 1151                     uid, gid);
 1152                 return (EPERM);
 1153         case NFSERR_BADNAME:
 1154         case NFSERR_BADCHAR:
 1155                 printf("nfsv4 char/name not handled by server\n");
 1156                 return (ENOENT);
 1157         case NFSERR_STALECLIENTID:
 1158         case NFSERR_STALESTATEID:
 1159         case NFSERR_EXPIRED:
 1160         case NFSERR_BADSTATEID:
 1161         case NFSERR_BADSESSION:
 1162                 printf("nfsv4 recover err returned %d\n", error);
 1163                 return (EIO);
 1164         case NFSERR_BADHANDLE:
 1165         case NFSERR_SERVERFAULT:
 1166         case NFSERR_BADTYPE:
 1167         case NFSERR_FHEXPIRED:
 1168         case NFSERR_RESOURCE:
 1169         case NFSERR_MOVED:
 1170         case NFSERR_NOFILEHANDLE:
 1171         case NFSERR_MINORVERMISMATCH:
 1172         case NFSERR_OLDSTATEID:
 1173         case NFSERR_BADSEQID:
 1174         case NFSERR_LEASEMOVED:
 1175         case NFSERR_RECLAIMBAD:
 1176         case NFSERR_BADXDR:
 1177         case NFSERR_OPILLEGAL:
 1178                 printf("nfsv4 client/server protocol prob err=%d\n",
 1179                     error);
 1180                 return (EIO);
 1181         default:
 1182                 tprintf(p, LOG_INFO, "nfsv4 err=%d\n", error);
 1183                 return (EIO);
 1184         };
 1185 }
 1186 
 1187 /*
 1188  * Check to see if the process for this owner exists. Return 1 if it doesn't
 1189  * and 0 otherwise.
 1190  */
 1191 int
 1192 nfscl_procdoesntexist(u_int8_t *own)
 1193 {
 1194         union {
 1195                 u_int32_t       lval;
 1196                 u_int8_t        cval[4];
 1197         } tl;
 1198         struct proc *p;
 1199         pid_t pid;
 1200         int i, ret = 0;
 1201 
 1202         /* For the single open_owner of all 0 bytes, just return 0. */
 1203         for (i = 0; i < NFSV4CL_LOCKNAMELEN; i++)
 1204                 if (own[i] != 0)
 1205                         break;
 1206         if (i == NFSV4CL_LOCKNAMELEN)
 1207                 return (0);
 1208 
 1209         tl.cval[0] = *own++;
 1210         tl.cval[1] = *own++;
 1211         tl.cval[2] = *own++;
 1212         tl.cval[3] = *own++;
 1213         pid = tl.lval;
 1214         p = pfind_locked(pid);
 1215         if (p == NULL)
 1216                 return (1);
 1217         if (p->p_stats == NULL) {
 1218                 PROC_UNLOCK(p);
 1219                 return (0);
 1220         }
 1221         tl.cval[0] = *own++;
 1222         tl.cval[1] = *own++;
 1223         tl.cval[2] = *own++;
 1224         tl.cval[3] = *own++;
 1225         if (tl.lval != p->p_stats->p_start.tv_sec) {
 1226                 ret = 1;
 1227         } else {
 1228                 tl.cval[0] = *own++;
 1229                 tl.cval[1] = *own++;
 1230                 tl.cval[2] = *own++;
 1231                 tl.cval[3] = *own;
 1232                 if (tl.lval != p->p_stats->p_start.tv_usec)
 1233                         ret = 1;
 1234         }
 1235         PROC_UNLOCK(p);
 1236         return (ret);
 1237 }
 1238 
 1239 /*
 1240  * - nfs pseudo system call for the client
 1241  */
 1242 /*
 1243  * MPSAFE
 1244  */
 1245 static int
 1246 nfssvc_nfscl(struct thread *td, struct nfssvc_args *uap)
 1247 {
 1248         struct file *fp;
 1249         struct nfscbd_args nfscbdarg;
 1250         struct nfsd_nfscbd_args nfscbdarg2;
 1251         struct nameidata nd;
 1252         struct nfscl_dumpmntopts dumpmntopts;
 1253         cap_rights_t rights;
 1254         char *buf;
 1255         int error;
 1256         struct mount *mp;
 1257         struct nfsmount *nmp;
 1258 
 1259         if (uap->flag & NFSSVC_CBADDSOCK) {
 1260                 error = copyin(uap->argp, (caddr_t)&nfscbdarg, sizeof(nfscbdarg));
 1261                 if (error)
 1262                         return (error);
 1263                 /*
 1264                  * Since we don't know what rights might be required,
 1265                  * pretend that we need them all. It is better to be too
 1266                  * careful than too reckless.
 1267                  */
 1268                 error = fget(td, nfscbdarg.sock,
 1269                     cap_rights_init(&rights, CAP_SOCK_CLIENT), &fp);
 1270                 if (error)
 1271                         return (error);
 1272                 if (fp->f_type != DTYPE_SOCKET) {
 1273                         fdrop(fp, td);
 1274                         return (EPERM);
 1275                 }
 1276                 error = nfscbd_addsock(fp);
 1277                 fdrop(fp, td);
 1278                 if (!error && nfscl_enablecallb == 0) {
 1279                         nfsv4_cbport = nfscbdarg.port;
 1280                         nfscl_enablecallb = 1;
 1281                 }
 1282         } else if (uap->flag & NFSSVC_NFSCBD) {
 1283                 if (uap->argp == NULL) 
 1284                         return (EINVAL);
 1285                 error = copyin(uap->argp, (caddr_t)&nfscbdarg2,
 1286                     sizeof(nfscbdarg2));
 1287                 if (error)
 1288                         return (error);
 1289                 error = nfscbd_nfsd(td, &nfscbdarg2);
 1290         } else if (uap->flag & NFSSVC_DUMPMNTOPTS) {
 1291                 error = copyin(uap->argp, &dumpmntopts, sizeof(dumpmntopts));
 1292                 if (error == 0 && (dumpmntopts.ndmnt_blen < 256 ||
 1293                     dumpmntopts.ndmnt_blen > 1024))
 1294                         error = EINVAL;
 1295                 if (error == 0)
 1296                         error = nfsrv_lookupfilename(&nd,
 1297                             dumpmntopts.ndmnt_fname, td);
 1298                 if (error == 0 && strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name,
 1299                     "nfs") != 0) {
 1300                         vput(nd.ni_vp);
 1301                         error = EINVAL;
 1302                 }
 1303                 if (error == 0) {
 1304                         buf = malloc(dumpmntopts.ndmnt_blen, M_TEMP, M_WAITOK |
 1305                             M_ZERO);
 1306                         nfscl_retopts(VFSTONFS(nd.ni_vp->v_mount), buf,
 1307                             dumpmntopts.ndmnt_blen);
 1308                         vput(nd.ni_vp);
 1309                         error = copyout(buf, dumpmntopts.ndmnt_buf,
 1310                             dumpmntopts.ndmnt_blen);
 1311                         free(buf, M_TEMP);
 1312                 }
 1313         } else if (uap->flag & NFSSVC_FORCEDISM) {
 1314                 buf = malloc(MNAMELEN + 1, M_TEMP, M_WAITOK);
 1315                 error = copyinstr(uap->argp, buf, MNAMELEN + 1, NULL);
 1316                 if (error == 0) {
 1317                         nmp = NULL;
 1318                         mtx_lock(&mountlist_mtx);
 1319                         TAILQ_FOREACH(mp, &mountlist, mnt_list) {
 1320                                 if (strcmp(mp->mnt_stat.f_mntonname, buf) ==
 1321                                     0 && strcmp(mp->mnt_stat.f_fstypename,
 1322                                     "nfs") == 0 && mp->mnt_data != NULL) {
 1323                                         nmp = VFSTONFS(mp);
 1324                                         NFSDDSLOCK();
 1325                                         if (nfsv4_findmirror(nmp) != NULL) {
 1326                                                 NFSDDSUNLOCK();
 1327                                                 error = ENXIO;
 1328                                                 nmp = NULL;
 1329                                                 break;
 1330                                         }
 1331                                         mtx_lock(&nmp->nm_mtx);
 1332                                         if ((nmp->nm_privflag &
 1333                                             NFSMNTP_FORCEDISM) == 0) {
 1334                                                 nmp->nm_privflag |= 
 1335                                                    (NFSMNTP_FORCEDISM |
 1336                                                     NFSMNTP_CANCELRPCS);
 1337                                                 mtx_unlock(&nmp->nm_mtx);
 1338                                         } else {
 1339                                                 mtx_unlock(&nmp->nm_mtx);
 1340                                                 nmp = NULL;
 1341                                         }
 1342                                         NFSDDSUNLOCK();
 1343                                         break;
 1344                                 }
 1345                         }
 1346                         mtx_unlock(&mountlist_mtx);
 1347 
 1348                         if (nmp != NULL) {
 1349                                 /*
 1350                                  * Call newnfs_nmcancelreqs() to cause
 1351                                  * any RPCs in progress on the mount point to
 1352                                  * fail.
 1353                                  * This will cause any process waiting for an
 1354                                  * RPC to complete while holding a vnode lock
 1355                                  * on the mounted-on vnode (such as "df" or
 1356                                  * a non-forced "umount") to fail.
 1357                                  * This will unlock the mounted-on vnode so
 1358                                  * a forced dismount can succeed.
 1359                                  * Then clear NFSMNTP_CANCELRPCS and wakeup(),
 1360                                  * so that nfs_unmount() can complete.
 1361                                  */
 1362                                 newnfs_nmcancelreqs(nmp);
 1363                                 mtx_lock(&nmp->nm_mtx);
 1364                                 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
 1365                                 wakeup(nmp);
 1366                                 mtx_unlock(&nmp->nm_mtx);
 1367                         } else if (error == 0)
 1368                                 error = EINVAL;
 1369                 }
 1370                 free(buf, M_TEMP);
 1371         } else {
 1372                 error = EINVAL;
 1373         }
 1374         return (error);
 1375 }
 1376 
 1377 extern int (*nfsd_call_nfscl)(struct thread *, struct nfssvc_args *);
 1378 
 1379 /*
 1380  * Called once to initialize data structures...
 1381  */
 1382 static int
 1383 nfscl_modevent(module_t mod, int type, void *data)
 1384 {
 1385         int error = 0;
 1386         static int loaded = 0;
 1387 
 1388         switch (type) {
 1389         case MOD_LOAD:
 1390                 if (loaded)
 1391                         return (0);
 1392                 newnfs_portinit();
 1393                 mtx_init(&ncl_iod_mutex, "ncl_iod_mutex", NULL, MTX_DEF);
 1394                 nfscl_init();
 1395                 NFSD_LOCK();
 1396                 nfsrvd_cbinit(0);
 1397                 NFSD_UNLOCK();
 1398                 ncl_call_invalcaches = ncl_invalcaches;
 1399                 nfsd_call_nfscl = nfssvc_nfscl;
 1400                 loaded = 1;
 1401                 break;
 1402 
 1403         case MOD_UNLOAD:
 1404                 if (nfs_numnfscbd != 0) {
 1405                         error = EBUSY;
 1406                         break;
 1407                 }
 1408 
 1409                 /*
 1410                  * XXX: Unloading of nfscl module is unsupported.
 1411                  */
 1412 #if 0
 1413                 ncl_call_invalcaches = NULL;
 1414                 nfsd_call_nfscl = NULL;
 1415                 /* and get rid of the mutexes */
 1416                 mtx_destroy(&ncl_iod_mutex);
 1417                 loaded = 0;
 1418                 break;
 1419 #else
 1420                 /* FALLTHROUGH */
 1421 #endif
 1422         default:
 1423                 error = EOPNOTSUPP;
 1424                 break;
 1425         }
 1426         return error;
 1427 }
 1428 static moduledata_t nfscl_mod = {
 1429         "nfscl",
 1430         nfscl_modevent,
 1431         NULL,
 1432 };
 1433 DECLARE_MODULE(nfscl, nfscl_mod, SI_SUB_VFS, SI_ORDER_FIRST);
 1434 
 1435 /* So that loader and kldload(2) can find us, wherever we are.. */
 1436 MODULE_VERSION(nfscl, 1);
 1437 MODULE_DEPEND(nfscl, nfscommon, 1, 1, 1);
 1438 MODULE_DEPEND(nfscl, krpc, 1, 1, 1);
 1439 MODULE_DEPEND(nfscl, nfssvc, 1, 1, 1);
 1440 MODULE_DEPEND(nfscl, nfslock, 1, 1, 1);
 1441 

Cache object: bfe9cf19c791ec22e6b965958ddf169c


[ 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.