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_clstate.c

Version: -  FREEBSD  -  FREEBSD-13-STABLE  -  FREEBSD-13-0  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  l41  -  OPENBSD  -  linux-2.6  -  MK84  -  PLAN9  -  xnu-8792 
SearchContext: -  none  -  3  -  10 

    1 /*-
    2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
    3  *
    4  * Copyright (c) 2009 Rick Macklem, University of Guelph
    5  * All rights reserved.
    6  *
    7  * Redistribution and use in source and binary forms, with or without
    8  * modification, are permitted provided that the following conditions
    9  * are met:
   10  * 1. Redistributions of source code must retain the above copyright
   11  *    notice, this list of conditions and the following disclaimer.
   12  * 2. Redistributions in binary form must reproduce the above copyright
   13  *    notice, this list of conditions and the following disclaimer in the
   14  *    documentation and/or other materials provided with the distribution.
   15  *
   16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
   17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
   20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   26  * SUCH DAMAGE.
   27  *
   28  */
   29 
   30 #include <sys/cdefs.h>
   31 __FBSDID("$FreeBSD$");
   32 
   33 /*
   34  * These functions implement the client side state handling for NFSv4.
   35  * NFSv4 state handling:
   36  * - A lockowner is used to determine lock contention, so it
   37  *   corresponds directly to a Posix pid. (1 to 1 mapping)
   38  * - The correct granularity of an OpenOwner is not nearly so
   39  *   obvious. An OpenOwner does the following:
   40  *   - provides a serial sequencing of Open/Close/Lock-with-new-lockowner
   41  *   - is used to check for Open/Share contention (not applicable to
   42  *     this client, since all Opens are Deny_None)
   43  *   As such, I considered both extreme.
   44  *   1 OpenOwner per ClientID - Simple to manage, but fully serializes
   45  *   all Open, Close and Lock (with a new lockowner) Ops.
   46  *   1 OpenOwner for each Open - This one results in an OpenConfirm for
   47  *   every Open, for most servers.
   48  *   So, I chose to use the same mapping as I did for LockOwnwers.
   49  *   The main concern here is that you can end up with multiple Opens
   50  *   for the same File Handle, but on different OpenOwners (opens
   51  *   inherited from parents, grandparents...) and you do not know
   52  *   which of these the vnodeop close applies to. This is handled by
   53  *   delaying the Close Op(s) until all of the Opens have been closed.
   54  *   (It is not yet obvious if this is the correct granularity.)
   55  * - How the code handles serialization:
   56  *   - For the ClientId, it uses an exclusive lock while getting its
   57  *     SetClientId and during recovery. Otherwise, it uses a shared
   58  *     lock via a reference count.
   59  *   - For the rest of the data structures, it uses an SMP mutex
   60  *     (once the nfs client is SMP safe) and doesn't sleep while
   61  *     manipulating the linked lists.
   62  *   - The serialization of Open/Close/Lock/LockU falls out in the
   63  *     "wash", since OpenOwners and LockOwners are both mapped from
   64  *     Posix pid. In other words, there is only one Posix pid using
   65  *     any given owner, so that owner is serialized. (If you change
   66  *     the granularity of the OpenOwner, then code must be added to
   67  *     serialize Ops on the OpenOwner.)
   68  * - When to get rid of OpenOwners and LockOwners.
   69  *   - The function nfscl_cleanup_common() is executed after a process exits.
   70  *     It goes through the client list looking for all Open and Lock Owners.
   71  *     When one is found, it is marked "defunct" or in the case of
   72  *     an OpenOwner without any Opens, freed.
   73  *     The renew thread scans for defunct Owners and gets rid of them,
   74  *     if it can. The LockOwners will also be deleted when the
   75  *     associated Open is closed.
   76  *   - If the LockU or Close Op(s) fail during close in a way
   77  *     that could be recovered upon retry, they are relinked to the
   78  *     ClientId's defunct open list and retried by the renew thread
   79  *     until they succeed or an unmount/recovery occurs.
   80  *     (Since we are done with them, they do not need to be recovered.)
   81  */
   82 
   83 #include <fs/nfs/nfsport.h>
   84 
   85 /*
   86  * Global variables
   87  */
   88 extern struct nfsstatsv1 nfsstatsv1;
   89 extern struct nfsreqhead nfsd_reqq;
   90 extern u_int32_t newnfs_false, newnfs_true;
   91 extern int nfscl_debuglevel;
   92 extern int nfscl_enablecallb;
   93 extern int nfs_numnfscbd;
   94 NFSREQSPINLOCK;
   95 NFSCLSTATEMUTEX;
   96 int nfscl_inited = 0;
   97 struct nfsclhead nfsclhead;     /* Head of clientid list */
   98 int nfscl_deleghighwater = NFSCLDELEGHIGHWATER;
   99 int nfscl_layouthighwater = NFSCLLAYOUTHIGHWATER;
  100 
  101 static int nfscl_delegcnt = 0;
  102 static int nfscl_layoutcnt = 0;
  103 static int nfscl_getopen(struct nfsclownerhead *, u_int8_t *, int, u_int8_t *,
  104     u_int8_t *, u_int32_t, struct nfscllockowner **, struct nfsclopen **);
  105 static void nfscl_clrelease(struct nfsclclient *);
  106 static void nfscl_unlinkopen(struct nfsclopen *);
  107 static void nfscl_cleanclient(struct nfsclclient *);
  108 static void nfscl_expireclient(struct nfsclclient *, struct nfsmount *,
  109     struct ucred *, NFSPROC_T *);
  110 static int nfscl_expireopen(struct nfsclclient *, struct nfsclopen *,
  111     struct nfsmount *, struct ucred *, NFSPROC_T *);
  112 static void nfscl_recover(struct nfsclclient *, struct ucred *, NFSPROC_T *);
  113 static void nfscl_insertlock(struct nfscllockowner *, struct nfscllock *,
  114     struct nfscllock *, int);
  115 static int nfscl_updatelock(struct nfscllockowner *, struct nfscllock **,
  116     struct nfscllock **, int);
  117 static void nfscl_delegreturnall(struct nfsclclient *, NFSPROC_T *,
  118     struct nfscldeleghead *);
  119 static u_int32_t nfscl_nextcbident(void);
  120 static mount_t nfscl_getmnt(int, uint8_t *, u_int32_t, struct nfsclclient **);
  121 static struct nfsclclient *nfscl_getclnt(u_int32_t);
  122 static struct nfsclclient *nfscl_getclntsess(uint8_t *);
  123 static struct nfscldeleg *nfscl_finddeleg(struct nfsclclient *, u_int8_t *,
  124     int);
  125 static void nfscl_retoncloselayout(vnode_t, struct nfsclclient *, uint8_t *,
  126     int, struct nfsclrecalllayout **, struct nfscllayout **);
  127 static void nfscl_reldevinfo_locked(struct nfscldevinfo *);
  128 static struct nfscllayout *nfscl_findlayout(struct nfsclclient *, u_int8_t *,
  129     int);
  130 static struct nfscldevinfo *nfscl_finddevinfo(struct nfsclclient *, uint8_t *);
  131 static int nfscl_checkconflict(struct nfscllockownerhead *, struct nfscllock *,
  132     u_int8_t *, struct nfscllock **);
  133 static void nfscl_freealllocks(struct nfscllockownerhead *, int);
  134 static int nfscl_localconflict(struct nfsclclient *, u_int8_t *, int,
  135     struct nfscllock *, u_int8_t *, struct nfscldeleg *, struct nfscllock **);
  136 static void nfscl_newopen(struct nfsclclient *, struct nfscldeleg *,
  137     struct nfsclowner **, struct nfsclowner **, struct nfsclopen **,
  138     struct nfsclopen **, u_int8_t *, u_int8_t *, int, struct ucred *, int *);
  139 static int nfscl_moveopen(vnode_t , struct nfsclclient *,
  140     struct nfsmount *, struct nfsclopen *, struct nfsclowner *,
  141     struct nfscldeleg *, struct ucred *, NFSPROC_T *);
  142 static void nfscl_totalrecall(struct nfsclclient *);
  143 static int nfscl_relock(vnode_t , struct nfsclclient *, struct nfsmount *,
  144     struct nfscllockowner *, struct nfscllock *, struct ucred *, NFSPROC_T *);
  145 static int nfscl_tryopen(struct nfsmount *, vnode_t , u_int8_t *, int,
  146     u_int8_t *, int, u_int32_t, struct nfsclopen *, u_int8_t *, int,
  147     struct nfscldeleg **, int, u_int32_t, struct ucred *, NFSPROC_T *);
  148 static int nfscl_trylock(struct nfsmount *, vnode_t , u_int8_t *,
  149     int, struct nfscllockowner *, int, int, u_int64_t, u_int64_t, short,
  150     struct ucred *, NFSPROC_T *);
  151 static int nfsrpc_reopen(struct nfsmount *, u_int8_t *, int, u_int32_t,
  152     struct nfsclopen *, struct nfscldeleg **, struct ucred *, NFSPROC_T *);
  153 static void nfscl_freedeleg(struct nfscldeleghead *, struct nfscldeleg *,
  154     bool);
  155 static int nfscl_errmap(struct nfsrv_descript *, u_int32_t);
  156 static void nfscl_cleanup_common(struct nfsclclient *, u_int8_t *);
  157 static int nfscl_recalldeleg(struct nfsclclient *, struct nfsmount *,
  158     struct nfscldeleg *, vnode_t, struct ucred *, NFSPROC_T *, int,
  159     vnode_t *);
  160 static void nfscl_freeopenowner(struct nfsclowner *, int);
  161 static void nfscl_cleandeleg(struct nfscldeleg *);
  162 static int nfscl_trydelegreturn(struct nfscldeleg *, struct ucred *,
  163     struct nfsmount *, NFSPROC_T *);
  164 static void nfscl_emptylockowner(struct nfscllockowner *,
  165     struct nfscllockownerfhhead *);
  166 static void nfscl_mergeflayouts(struct nfsclflayouthead *,
  167     struct nfsclflayouthead *);
  168 static int nfscl_layoutrecall(int, struct nfscllayout *, uint32_t, uint64_t,
  169     uint64_t, uint32_t, uint32_t, uint32_t, char *, struct nfsclrecalllayout *);
  170 static int nfscl_seq(uint32_t, uint32_t);
  171 static void nfscl_layoutreturn(struct nfsmount *, struct nfscllayout *,
  172     struct ucred *, NFSPROC_T *);
  173 static void nfscl_dolayoutcommit(struct nfsmount *, struct nfscllayout *,
  174     struct ucred *, NFSPROC_T *);
  175 
  176 static short nfscberr_null[] = {
  177         0,
  178         0,
  179 };
  180 
  181 static short nfscberr_getattr[] = {
  182         NFSERR_RESOURCE,
  183         NFSERR_BADHANDLE,
  184         NFSERR_BADXDR,
  185         NFSERR_RESOURCE,
  186         NFSERR_SERVERFAULT,
  187         0,
  188 };
  189 
  190 static short nfscberr_recall[] = {
  191         NFSERR_RESOURCE,
  192         NFSERR_BADHANDLE,
  193         NFSERR_BADSTATEID,
  194         NFSERR_BADXDR,
  195         NFSERR_RESOURCE,
  196         NFSERR_SERVERFAULT,
  197         0,
  198 };
  199 
  200 static short *nfscl_cberrmap[] = {
  201         nfscberr_null,
  202         nfscberr_null,
  203         nfscberr_null,
  204         nfscberr_getattr,
  205         nfscberr_recall
  206 };
  207 
  208 #define NETFAMILY(clp) \
  209                 (((clp)->nfsc_flags & NFSCLFLAGS_AFINET6) ? AF_INET6 : AF_INET)
  210 
  211 /*
  212  * Called for an open operation.
  213  * If the nfhp argument is NULL, just get an openowner.
  214  */
  215 int
  216 nfscl_open(vnode_t vp, u_int8_t *nfhp, int fhlen, u_int32_t amode, int usedeleg,
  217     struct ucred *cred, NFSPROC_T *p, struct nfsclowner **owpp,
  218     struct nfsclopen **opp, int *newonep, int *retp, int lockit, bool firstref)
  219 {
  220         struct nfsclclient *clp;
  221         struct nfsclowner *owp, *nowp;
  222         struct nfsclopen *op = NULL, *nop = NULL;
  223         struct nfscldeleg *dp;
  224         struct nfsclownerhead *ohp;
  225         u_int8_t own[NFSV4CL_LOCKNAMELEN];
  226         int ret;
  227 
  228         if (newonep != NULL)
  229                 *newonep = 0;
  230         if (opp != NULL)
  231                 *opp = NULL;
  232         if (owpp != NULL)
  233                 *owpp = NULL;
  234 
  235         /*
  236          * Might need one or both of these, so MALLOC them now, to
  237          * avoid a tsleep() in MALLOC later.
  238          */
  239         nowp = malloc(sizeof (struct nfsclowner),
  240             M_NFSCLOWNER, M_WAITOK);
  241         if (nfhp != NULL)
  242             nop = malloc(sizeof (struct nfsclopen) +
  243                 fhlen - 1, M_NFSCLOPEN, M_WAITOK);
  244         ret = nfscl_getcl(vnode_mount(vp), cred, p, 1, firstref, &clp);
  245         if (ret != 0) {
  246                 free(nowp, M_NFSCLOWNER);
  247                 if (nop != NULL)
  248                         free(nop, M_NFSCLOPEN);
  249                 return (ret);
  250         }
  251 
  252         /*
  253          * Get the Open iff it already exists.
  254          * If none found, add the new one or return error, depending upon
  255          * "create".
  256          */
  257         NFSLOCKCLSTATE();
  258         dp = NULL;
  259         /* First check the delegation list */
  260         if (nfhp != NULL && usedeleg) {
  261                 LIST_FOREACH(dp, NFSCLDELEGHASH(clp, nfhp, fhlen), nfsdl_hash) {
  262                         if (dp->nfsdl_fhlen == fhlen &&
  263                             !NFSBCMP(nfhp, dp->nfsdl_fh, fhlen)) {
  264                                 if (!(amode & NFSV4OPEN_ACCESSWRITE) ||
  265                                     (dp->nfsdl_flags & NFSCLDL_WRITE))
  266                                         break;
  267                                 dp = NULL;
  268                                 break;
  269                         }
  270                 }
  271         }
  272 
  273         if (dp != NULL) {
  274                 nfscl_filllockowner(p->td_proc, own, F_POSIX);
  275                 ohp = &dp->nfsdl_owner;
  276         } else {
  277                 /* For NFSv4.1 and this option, use a single open_owner. */
  278                 if (NFSHASONEOPENOWN(VFSTONFS(vnode_mount(vp))))
  279                         nfscl_filllockowner(NULL, own, F_POSIX);
  280                 else
  281                         nfscl_filllockowner(p->td_proc, own, F_POSIX);
  282                 ohp = &clp->nfsc_owner;
  283         }
  284         /* Now, search for an openowner */
  285         LIST_FOREACH(owp, ohp, nfsow_list) {
  286                 if (!NFSBCMP(owp->nfsow_owner, own, NFSV4CL_LOCKNAMELEN))
  287                         break;
  288         }
  289 
  290         /*
  291          * Create a new open, as required.
  292          */
  293         nfscl_newopen(clp, dp, &owp, &nowp, &op, &nop, own, nfhp, fhlen,
  294             cred, newonep);
  295 
  296         /*
  297          * Now, check the mode on the open and return the appropriate
  298          * value.
  299          */
  300         if (retp != NULL) {
  301                 if (nfhp != NULL && dp != NULL && nop == NULL)
  302                         /* new local open on delegation */
  303                         *retp = NFSCLOPEN_SETCRED;
  304                 else
  305                         *retp = NFSCLOPEN_OK;
  306         }
  307         if (op != NULL && (amode & ~(op->nfso_mode))) {
  308                 op->nfso_mode |= amode;
  309                 if (retp != NULL && dp == NULL)
  310                         *retp = NFSCLOPEN_DOOPEN;
  311         }
  312 
  313         /*
  314          * Serialize modifications to the open owner for multiple threads
  315          * within the same process using a read/write sleep lock.
  316          * For NFSv4.1 and a single OpenOwner, allow concurrent open operations
  317          * by acquiring a shared lock.  The close operations still use an
  318          * exclusive lock for this case.
  319          */
  320         if (lockit != 0) {
  321                 if (NFSHASONEOPENOWN(VFSTONFS(vnode_mount(vp)))) {
  322                         /*
  323                          * Get a shared lock on the OpenOwner, but first
  324                          * wait for any pending exclusive lock, so that the
  325                          * exclusive locker gets priority.
  326                          */
  327                         nfsv4_lock(&owp->nfsow_rwlock, 0, NULL,
  328                             NFSCLSTATEMUTEXPTR, NULL);
  329                         nfsv4_getref(&owp->nfsow_rwlock, NULL,
  330                             NFSCLSTATEMUTEXPTR, NULL);
  331                 } else
  332                         nfscl_lockexcl(&owp->nfsow_rwlock, NFSCLSTATEMUTEXPTR);
  333         }
  334         NFSUNLOCKCLSTATE();
  335         if (nowp != NULL)
  336                 free(nowp, M_NFSCLOWNER);
  337         if (nop != NULL)
  338                 free(nop, M_NFSCLOPEN);
  339         if (owpp != NULL)
  340                 *owpp = owp;
  341         if (opp != NULL)
  342                 *opp = op;
  343         return (0);
  344 }
  345 
  346 /*
  347  * Create a new open, as required.
  348  */
  349 static void
  350 nfscl_newopen(struct nfsclclient *clp, struct nfscldeleg *dp,
  351     struct nfsclowner **owpp, struct nfsclowner **nowpp, struct nfsclopen **opp,
  352     struct nfsclopen **nopp, u_int8_t *own, u_int8_t *fhp, int fhlen,
  353     struct ucred *cred, int *newonep)
  354 {
  355         struct nfsclowner *owp = *owpp, *nowp;
  356         struct nfsclopen *op, *nop;
  357 
  358         if (nowpp != NULL)
  359                 nowp = *nowpp;
  360         else
  361                 nowp = NULL;
  362         if (nopp != NULL)
  363                 nop = *nopp;
  364         else
  365                 nop = NULL;
  366         if (owp == NULL && nowp != NULL) {
  367                 NFSBCOPY(own, nowp->nfsow_owner, NFSV4CL_LOCKNAMELEN);
  368                 LIST_INIT(&nowp->nfsow_open);
  369                 nowp->nfsow_clp = clp;
  370                 nowp->nfsow_seqid = 0;
  371                 nowp->nfsow_defunct = 0;
  372                 nfscl_lockinit(&nowp->nfsow_rwlock);
  373                 if (dp != NULL) {
  374                         nfsstatsv1.cllocalopenowners++;
  375                         LIST_INSERT_HEAD(&dp->nfsdl_owner, nowp, nfsow_list);
  376                 } else {
  377                         nfsstatsv1.clopenowners++;
  378                         LIST_INSERT_HEAD(&clp->nfsc_owner, nowp, nfsow_list);
  379                 }
  380                 owp = *owpp = nowp;
  381                 *nowpp = NULL;
  382                 if (newonep != NULL)
  383                         *newonep = 1;
  384         }
  385 
  386          /* If an fhp has been specified, create an Open as well. */
  387         if (fhp != NULL) {
  388                 /* and look for the correct open, based upon FH */
  389                 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
  390                         if (op->nfso_fhlen == fhlen &&
  391                             !NFSBCMP(op->nfso_fh, fhp, fhlen))
  392                                 break;
  393                 }
  394                 if (op == NULL && nop != NULL) {
  395                         nop->nfso_own = owp;
  396                         nop->nfso_mode = 0;
  397                         nop->nfso_opencnt = 0;
  398                         nop->nfso_posixlock = 1;
  399                         nop->nfso_fhlen = fhlen;
  400                         NFSBCOPY(fhp, nop->nfso_fh, fhlen);
  401                         LIST_INIT(&nop->nfso_lock);
  402                         nop->nfso_stateid.seqid = 0;
  403                         nop->nfso_stateid.other[0] = 0;
  404                         nop->nfso_stateid.other[1] = 0;
  405                         nop->nfso_stateid.other[2] = 0;
  406                         KASSERT(cred != NULL, ("%s: cred NULL\n", __func__));
  407                         newnfs_copyincred(cred, &nop->nfso_cred);
  408                         if (dp != NULL) {
  409                                 TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
  410                                 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp,
  411                                     nfsdl_list);
  412                                 dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
  413                                 nfsstatsv1.cllocalopens++;
  414                         } else {
  415                                 nfsstatsv1.clopens++;
  416                         }
  417                         LIST_INSERT_HEAD(&owp->nfsow_open, nop, nfso_list);
  418                         *opp = nop;
  419                         *nopp = NULL;
  420                         if (newonep != NULL)
  421                                 *newonep = 1;
  422                 } else {
  423                         *opp = op;
  424                 }
  425         }
  426 }
  427 
  428 /*
  429  * Called to find/add a delegation to a client.
  430  */
  431 int
  432 nfscl_deleg(mount_t mp, struct nfsclclient *clp, u_int8_t *nfhp,
  433     int fhlen, struct ucred *cred, NFSPROC_T *p, struct nfscldeleg **dpp)
  434 {
  435         struct nfscldeleg *dp = *dpp, *tdp;
  436 
  437         /*
  438          * First, if we have received a Read delegation for a file on a
  439          * read/write file system, just return it, because they aren't
  440          * useful, imho.
  441          */
  442         if (mp != NULL && dp != NULL && !NFSMNT_RDONLY(mp) &&
  443             (dp->nfsdl_flags & NFSCLDL_READ)) {
  444                 (void) nfscl_trydelegreturn(dp, cred, VFSTONFS(mp), p);
  445                 free(dp, M_NFSCLDELEG);
  446                 *dpp = NULL;
  447                 return (0);
  448         }
  449 
  450         /* Look for the correct deleg, based upon FH */
  451         NFSLOCKCLSTATE();
  452         tdp = nfscl_finddeleg(clp, nfhp, fhlen);
  453         if (tdp == NULL) {
  454                 if (dp == NULL) {
  455                         NFSUNLOCKCLSTATE();
  456                         return (NFSERR_BADSTATEID);
  457                 }
  458                 *dpp = NULL;
  459                 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list);
  460                 LIST_INSERT_HEAD(NFSCLDELEGHASH(clp, nfhp, fhlen), dp,
  461                     nfsdl_hash);
  462                 dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
  463                 nfsstatsv1.cldelegates++;
  464                 nfscl_delegcnt++;
  465         } else {
  466                 /*
  467                  * Delegation already exists, what do we do if a new one??
  468                  */
  469                 if (dp != NULL) {
  470                         printf("Deleg already exists!\n");
  471                         free(dp, M_NFSCLDELEG);
  472                         *dpp = NULL;
  473                 } else {
  474                         *dpp = tdp;
  475                 }
  476         }
  477         NFSUNLOCKCLSTATE();
  478         return (0);
  479 }
  480 
  481 /*
  482  * Find a delegation for this file handle. Return NULL upon failure.
  483  */
  484 static struct nfscldeleg *
  485 nfscl_finddeleg(struct nfsclclient *clp, u_int8_t *fhp, int fhlen)
  486 {
  487         struct nfscldeleg *dp;
  488 
  489         LIST_FOREACH(dp, NFSCLDELEGHASH(clp, fhp, fhlen), nfsdl_hash) {
  490             if (dp->nfsdl_fhlen == fhlen &&
  491                 !NFSBCMP(dp->nfsdl_fh, fhp, fhlen))
  492                 break;
  493         }
  494         return (dp);
  495 }
  496 
  497 /*
  498  * Get a stateid for an I/O operation. First, look for an open and iff
  499  * found, return either a lockowner stateid or the open stateid.
  500  * If no Open is found, just return error and the special stateid of all zeros.
  501  */
  502 int
  503 nfscl_getstateid(vnode_t vp, u_int8_t *nfhp, int fhlen, u_int32_t mode,
  504     int fords, struct ucred *cred, NFSPROC_T *p, nfsv4stateid_t *stateidp,
  505     void **lckpp)
  506 {
  507         struct nfsclclient *clp;
  508         struct nfsclowner *owp;
  509         struct nfsclopen *op = NULL, *top;
  510         struct nfscllockowner *lp;
  511         struct nfscldeleg *dp;
  512         struct nfsnode *np;
  513         struct nfsmount *nmp;
  514         u_int8_t own[NFSV4CL_LOCKNAMELEN];
  515         int error, done;
  516 
  517         *lckpp = NULL;
  518         /*
  519          * Initially, just set the special stateid of all zeros.
  520          * (Don't do this for a DS, since the special stateid can't be used.)
  521          */
  522         if (fords == 0) {
  523                 stateidp->seqid = 0;
  524                 stateidp->other[0] = 0;
  525                 stateidp->other[1] = 0;
  526                 stateidp->other[2] = 0;
  527         }
  528         if (vnode_vtype(vp) != VREG)
  529                 return (EISDIR);
  530         np = VTONFS(vp);
  531         nmp = VFSTONFS(vnode_mount(vp));
  532         NFSLOCKCLSTATE();
  533         clp = nfscl_findcl(nmp);
  534         if (clp == NULL) {
  535                 NFSUNLOCKCLSTATE();
  536                 return (EACCES);
  537         }
  538 
  539         /*
  540          * Wait for recovery to complete.
  541          */
  542         while ((clp->nfsc_flags & NFSCLFLAGS_RECVRINPROG))
  543                 (void) nfsmsleep(&clp->nfsc_flags, NFSCLSTATEMUTEXPTR,
  544                     PZERO, "nfsrecvr", NULL);
  545 
  546         /*
  547          * First, look for a delegation.
  548          */
  549         LIST_FOREACH(dp, NFSCLDELEGHASH(clp, nfhp, fhlen), nfsdl_hash) {
  550                 if (dp->nfsdl_fhlen == fhlen &&
  551                     !NFSBCMP(nfhp, dp->nfsdl_fh, fhlen)) {
  552                         if (!(mode & NFSV4OPEN_ACCESSWRITE) ||
  553                             (dp->nfsdl_flags & NFSCLDL_WRITE)) {
  554                                 stateidp->seqid = dp->nfsdl_stateid.seqid;
  555                                 stateidp->other[0] = dp->nfsdl_stateid.other[0];
  556                                 stateidp->other[1] = dp->nfsdl_stateid.other[1];
  557                                 stateidp->other[2] = dp->nfsdl_stateid.other[2];
  558                                 if (!(np->n_flag & NDELEGRECALL)) {
  559                                         TAILQ_REMOVE(&clp->nfsc_deleg, dp,
  560                                             nfsdl_list);
  561                                         TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp,
  562                                             nfsdl_list);
  563                                         dp->nfsdl_timestamp = NFSD_MONOSEC +
  564                                             120;
  565                                         dp->nfsdl_rwlock.nfslock_usecnt++;
  566                                         *lckpp = (void *)&dp->nfsdl_rwlock;
  567                                 }
  568                                 NFSUNLOCKCLSTATE();
  569                                 return (0);
  570                         }
  571                         break;
  572                 }
  573         }
  574 
  575         if (p != NULL) {
  576                 /*
  577                  * If p != NULL, we want to search the parentage tree
  578                  * for a matching OpenOwner and use that.
  579                  */
  580                 if (NFSHASONEOPENOWN(VFSTONFS(vnode_mount(vp))))
  581                         nfscl_filllockowner(NULL, own, F_POSIX);
  582                 else
  583                         nfscl_filllockowner(p->td_proc, own, F_POSIX);
  584                 lp = NULL;
  585                 error = nfscl_getopen(&clp->nfsc_owner, nfhp, fhlen, own, own,
  586                     mode, &lp, &op);
  587                 if (error == 0 && lp != NULL && fords == 0) {
  588                         /* Don't return a lock stateid for a DS. */
  589                         stateidp->seqid =
  590                             lp->nfsl_stateid.seqid;
  591                         stateidp->other[0] =
  592                             lp->nfsl_stateid.other[0];
  593                         stateidp->other[1] =
  594                             lp->nfsl_stateid.other[1];
  595                         stateidp->other[2] =
  596                             lp->nfsl_stateid.other[2];
  597                         NFSUNLOCKCLSTATE();
  598                         return (0);
  599                 }
  600         }
  601         if (op == NULL) {
  602                 /* If not found, just look for any OpenOwner that will work. */
  603                 top = NULL;
  604                 done = 0;
  605                 owp = LIST_FIRST(&clp->nfsc_owner);
  606                 while (!done && owp != NULL) {
  607                         LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
  608                                 if (op->nfso_fhlen == fhlen &&
  609                                     !NFSBCMP(op->nfso_fh, nfhp, fhlen)) {
  610                                         if (top == NULL && (op->nfso_mode &
  611                                             NFSV4OPEN_ACCESSWRITE) != 0 &&
  612                                             (mode & NFSV4OPEN_ACCESSREAD) != 0)
  613                                                 top = op;
  614                                         if ((mode & op->nfso_mode) == mode) {
  615                                                 done = 1;
  616                                                 break;
  617                                         }
  618                                 }
  619                         }
  620                         if (!done)
  621                                 owp = LIST_NEXT(owp, nfsow_list);
  622                 }
  623                 if (!done) {
  624                         NFSCL_DEBUG(2, "openmode top=%p\n", top);
  625                         if (top == NULL || NFSHASOPENMODE(nmp)) {
  626                                 NFSUNLOCKCLSTATE();
  627                                 return (ENOENT);
  628                         } else
  629                                 op = top;
  630                 }
  631                 /*
  632                  * For read aheads or write behinds, use the open cred.
  633                  * A read ahead or write behind is indicated by p == NULL.
  634                  */
  635                 if (p == NULL)
  636                         newnfs_copycred(&op->nfso_cred, cred);
  637         }
  638 
  639         /*
  640          * No lock stateid, so return the open stateid.
  641          */
  642         stateidp->seqid = op->nfso_stateid.seqid;
  643         stateidp->other[0] = op->nfso_stateid.other[0];
  644         stateidp->other[1] = op->nfso_stateid.other[1];
  645         stateidp->other[2] = op->nfso_stateid.other[2];
  646         NFSUNLOCKCLSTATE();
  647         return (0);
  648 }
  649 
  650 /*
  651  * Search for a matching file, mode and, optionally, lockowner.
  652  */
  653 static int
  654 nfscl_getopen(struct nfsclownerhead *ohp, u_int8_t *nfhp, int fhlen,
  655     u_int8_t *openown, u_int8_t *lockown, u_int32_t mode,
  656     struct nfscllockowner **lpp, struct nfsclopen **opp)
  657 {
  658         struct nfsclowner *owp;
  659         struct nfsclopen *op, *rop, *rop2;
  660         struct nfscllockowner *lp;
  661         int keep_looping;
  662 
  663         if (lpp != NULL)
  664                 *lpp = NULL;
  665         /*
  666          * rop will be set to the open to be returned. There are three
  667          * variants of this, all for an open of the correct file:
  668          * 1 - A match of lockown.
  669          * 2 - A match of the openown, when no lockown match exists.
  670          * 3 - A match for any open, if no openown or lockown match exists.
  671          * Looking for #2 over #3 probably isn't necessary, but since
  672          * RFC3530 is vague w.r.t. the relationship between openowners and
  673          * lockowners, I think this is the safer way to go.
  674          */
  675         rop = NULL;
  676         rop2 = NULL;
  677         keep_looping = 1;
  678         /* Search the client list */
  679         owp = LIST_FIRST(ohp);
  680         while (owp != NULL && keep_looping != 0) {
  681                 /* and look for the correct open */
  682                 op = LIST_FIRST(&owp->nfsow_open);
  683                 while (op != NULL && keep_looping != 0) {
  684                         if (op->nfso_fhlen == fhlen &&
  685                             !NFSBCMP(op->nfso_fh, nfhp, fhlen)
  686                             && (op->nfso_mode & mode) == mode) {
  687                                 if (lpp != NULL) {
  688                                         /* Now look for a matching lockowner. */
  689                                         LIST_FOREACH(lp, &op->nfso_lock,
  690                                             nfsl_list) {
  691                                                 if (!NFSBCMP(lp->nfsl_owner,
  692                                                     lockown,
  693                                                     NFSV4CL_LOCKNAMELEN)) {
  694                                                         *lpp = lp;
  695                                                         rop = op;
  696                                                         keep_looping = 0;
  697                                                         break;
  698                                                 }
  699                                         }
  700                                 }
  701                                 if (rop == NULL && !NFSBCMP(owp->nfsow_owner,
  702                                     openown, NFSV4CL_LOCKNAMELEN)) {
  703                                         rop = op;
  704                                         if (lpp == NULL)
  705                                                 keep_looping = 0;
  706                                 }
  707                                 if (rop2 == NULL)
  708                                         rop2 = op;
  709                         }
  710                         op = LIST_NEXT(op, nfso_list);
  711                 }
  712                 owp = LIST_NEXT(owp, nfsow_list);
  713         }
  714         if (rop == NULL)
  715                 rop = rop2;
  716         if (rop == NULL)
  717                 return (EBADF);
  718         *opp = rop;
  719         return (0);
  720 }
  721 
  722 /*
  723  * Release use of an open owner. Called when open operations are done
  724  * with the open owner.
  725  */
  726 void
  727 nfscl_ownerrelease(struct nfsmount *nmp, struct nfsclowner *owp,
  728     __unused int error, __unused int candelete, int unlocked)
  729 {
  730 
  731         if (owp == NULL)
  732                 return;
  733         NFSLOCKCLSTATE();
  734         if (unlocked == 0) {
  735                 if (NFSHASONEOPENOWN(nmp))
  736                         nfsv4_relref(&owp->nfsow_rwlock);
  737                 else
  738                         nfscl_lockunlock(&owp->nfsow_rwlock);
  739         }
  740         nfscl_clrelease(owp->nfsow_clp);
  741         NFSUNLOCKCLSTATE();
  742 }
  743 
  744 /*
  745  * Release use of an open structure under an open owner.
  746  */
  747 void
  748 nfscl_openrelease(struct nfsmount *nmp, struct nfsclopen *op, int error,
  749     int candelete)
  750 {
  751         struct nfsclclient *clp;
  752         struct nfsclowner *owp;
  753 
  754         if (op == NULL)
  755                 return;
  756         NFSLOCKCLSTATE();
  757         owp = op->nfso_own;
  758         if (NFSHASONEOPENOWN(nmp))
  759                 nfsv4_relref(&owp->nfsow_rwlock);
  760         else
  761                 nfscl_lockunlock(&owp->nfsow_rwlock);
  762         clp = owp->nfsow_clp;
  763         if (error && candelete && op->nfso_opencnt == 0)
  764                 nfscl_freeopen(op, 0, true);
  765         nfscl_clrelease(clp);
  766         NFSUNLOCKCLSTATE();
  767 }
  768 
  769 /*
  770  * Called to get a clientid structure. It will optionally lock the
  771  * client data structures to do the SetClientId/SetClientId_confirm,
  772  * but will release that lock and return the clientid with a reference
  773  * count on it.
  774  * If the "cred" argument is NULL, a new clientid should not be created.
  775  * If the "p" argument is NULL, a SetClientID/SetClientIDConfirm cannot
  776  * be done.
  777  * The start_renewthread argument tells nfscl_getcl() to start a renew
  778  * thread if this creates a new clp.
  779  * It always clpp with a reference count on it, unless returning an error.
  780  */
  781 int
  782 nfscl_getcl(struct mount *mp, struct ucred *cred, NFSPROC_T *p,
  783     int start_renewthread, bool firstref, struct nfsclclient **clpp)
  784 {
  785         struct nfsclclient *clp;
  786         struct nfsclclient *newclp = NULL;
  787         struct nfsmount *nmp;
  788         char uuid[HOSTUUIDLEN];
  789         int igotlock = 0, error, trystalecnt, clidinusedelay, i;
  790         u_int16_t idlen = 0;
  791 
  792         nmp = VFSTONFS(mp);
  793         if (cred != NULL) {
  794                 getcredhostuuid(cred, uuid, sizeof uuid);
  795                 idlen = strlen(uuid);
  796                 if (idlen > 0)
  797                         idlen += sizeof (u_int64_t);
  798                 else
  799                         idlen += sizeof (u_int64_t) + 16; /* 16 random bytes */
  800                 newclp = malloc(
  801                     sizeof (struct nfsclclient) + idlen - 1, M_NFSCLCLIENT,
  802                     M_WAITOK | M_ZERO);
  803         }
  804         NFSLOCKCLSTATE();
  805         /*
  806          * If a forced dismount is already in progress, don't
  807          * allocate a new clientid and get out now. For the case where
  808          * clp != NULL, this is a harmless optimization.
  809          */
  810         if (NFSCL_FORCEDISM(mp)) {
  811                 NFSUNLOCKCLSTATE();
  812                 if (newclp != NULL)
  813                         free(newclp, M_NFSCLCLIENT);
  814                 return (EBADF);
  815         }
  816         clp = nmp->nm_clp;
  817         if (clp == NULL) {
  818                 if (newclp == NULL) {
  819                         NFSUNLOCKCLSTATE();
  820                         return (EACCES);
  821                 }
  822                 clp = newclp;
  823                 clp->nfsc_idlen = idlen;
  824                 LIST_INIT(&clp->nfsc_owner);
  825                 TAILQ_INIT(&clp->nfsc_deleg);
  826                 TAILQ_INIT(&clp->nfsc_layout);
  827                 LIST_INIT(&clp->nfsc_devinfo);
  828                 for (i = 0; i < NFSCLDELEGHASHSIZE; i++)
  829                         LIST_INIT(&clp->nfsc_deleghash[i]);
  830                 for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++)
  831                         LIST_INIT(&clp->nfsc_layouthash[i]);
  832                 clp->nfsc_flags = NFSCLFLAGS_INITED;
  833                 clp->nfsc_clientidrev = 1;
  834                 clp->nfsc_cbident = nfscl_nextcbident();
  835                 nfscl_fillclid(nmp->nm_clval, uuid, clp->nfsc_id,
  836                     clp->nfsc_idlen);
  837                 LIST_INSERT_HEAD(&nfsclhead, clp, nfsc_list);
  838                 nmp->nm_clp = clp;
  839                 clp->nfsc_nmp = nmp;
  840                 NFSUNLOCKCLSTATE();
  841                 if (start_renewthread != 0)
  842                         nfscl_start_renewthread(clp);
  843         } else {
  844                 NFSUNLOCKCLSTATE();
  845                 if (newclp != NULL)
  846                         free(newclp, M_NFSCLCLIENT);
  847         }
  848         NFSLOCKCLSTATE();
  849         while ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0 && !igotlock &&
  850             !NFSCL_FORCEDISM(mp))
  851                 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
  852                     NFSCLSTATEMUTEXPTR, mp);
  853         if (igotlock == 0) {
  854                 /*
  855                  * Call nfsv4_lock() with "iwantlock == 0" on the firstref so
  856                  * that it will wait for a pending exclusive lock request.
  857                  * This gives the exclusive lock request priority over this
  858                  * shared lock request.
  859                  * An exclusive lock on nfsc_lock is used mainly for server
  860                  * crash recoveries and delegation recalls.
  861                  */
  862                 if (firstref)
  863                         nfsv4_lock(&clp->nfsc_lock, 0, NULL, NFSCLSTATEMUTEXPTR,
  864                             mp);
  865                 nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
  866         }
  867         if (igotlock == 0 && NFSCL_FORCEDISM(mp)) {
  868                 /*
  869                  * Both nfsv4_lock() and nfsv4_getref() know to check
  870                  * for NFSCL_FORCEDISM() and return without sleeping to
  871                  * wait for the exclusive lock to be released, since it
  872                  * might be held by nfscl_umount() and we need to get out
  873                  * now for that case and not wait until nfscl_umount()
  874                  * releases it.
  875                  */
  876                 NFSUNLOCKCLSTATE();
  877                 return (EBADF);
  878         }
  879         NFSUNLOCKCLSTATE();
  880 
  881         /*
  882          * If it needs a clientid, do the setclientid now.
  883          */
  884         if ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0) {
  885                 if (!igotlock)
  886                         panic("nfscl_clget");
  887                 if (p == NULL || cred == NULL) {
  888                         NFSLOCKCLSTATE();
  889                         nfsv4_unlock(&clp->nfsc_lock, 0);
  890                         NFSUNLOCKCLSTATE();
  891                         return (EACCES);
  892                 }
  893                 /*
  894                  * If RFC3530 Sec. 14.2.33 is taken literally,
  895                  * NFSERR_CLIDINUSE will be returned persistently for the
  896                  * case where a new mount of the same file system is using
  897                  * a different principal. In practice, NFSERR_CLIDINUSE is
  898                  * only returned when there is outstanding unexpired state
  899                  * on the clientid. As such, try for twice the lease
  900                  * interval, if we know what that is. Otherwise, make a
  901                  * wild ass guess.
  902                  * The case of returning NFSERR_STALECLIENTID is far less
  903                  * likely, but might occur if there is a significant delay
  904                  * between doing the SetClientID and SetClientIDConfirm Ops,
  905                  * such that the server throws away the clientid before
  906                  * receiving the SetClientIDConfirm.
  907                  */
  908                 if (clp->nfsc_renew > 0)
  909                         clidinusedelay = NFSCL_LEASE(clp->nfsc_renew) * 2;
  910                 else
  911                         clidinusedelay = 120;
  912                 trystalecnt = 3;
  913                 do {
  914                         error = nfsrpc_setclient(nmp, clp, 0, cred, p);
  915                         if (error == NFSERR_STALECLIENTID ||
  916                             error == NFSERR_STALEDONTRECOVER ||
  917                             error == NFSERR_BADSESSION ||
  918                             error == NFSERR_CLIDINUSE) {
  919                                 (void) nfs_catnap(PZERO, error, "nfs_setcl");
  920                         }
  921                 } while (((error == NFSERR_STALECLIENTID ||
  922                      error == NFSERR_BADSESSION ||
  923                      error == NFSERR_STALEDONTRECOVER) && --trystalecnt > 0) ||
  924                     (error == NFSERR_CLIDINUSE && --clidinusedelay > 0));
  925                 if (error) {
  926                         NFSLOCKCLSTATE();
  927                         nfsv4_unlock(&clp->nfsc_lock, 0);
  928                         NFSUNLOCKCLSTATE();
  929                         return (error);
  930                 }
  931                 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
  932         }
  933         if (igotlock) {
  934                 NFSLOCKCLSTATE();
  935                 nfsv4_unlock(&clp->nfsc_lock, 1);
  936                 NFSUNLOCKCLSTATE();
  937         }
  938 
  939         *clpp = clp;
  940         return (0);
  941 }
  942 
  943 /*
  944  * Get a reference to a clientid and return it, if valid.
  945  */
  946 struct nfsclclient *
  947 nfscl_findcl(struct nfsmount *nmp)
  948 {
  949         struct nfsclclient *clp;
  950 
  951         clp = nmp->nm_clp;
  952         if (clp == NULL || !(clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID))
  953                 return (NULL);
  954         return (clp);
  955 }
  956 
  957 /*
  958  * Release the clientid structure. It may be locked or reference counted.
  959  */
  960 static void
  961 nfscl_clrelease(struct nfsclclient *clp)
  962 {
  963 
  964         if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK)
  965                 nfsv4_unlock(&clp->nfsc_lock, 0);
  966         else
  967                 nfsv4_relref(&clp->nfsc_lock);
  968 }
  969 
  970 /*
  971  * External call for nfscl_clrelease.
  972  */
  973 void
  974 nfscl_clientrelease(struct nfsclclient *clp)
  975 {
  976 
  977         NFSLOCKCLSTATE();
  978         if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK)
  979                 nfsv4_unlock(&clp->nfsc_lock, 0);
  980         else
  981                 nfsv4_relref(&clp->nfsc_lock);
  982         NFSUNLOCKCLSTATE();
  983 }
  984 
  985 /*
  986  * Called when wanting to lock a byte region.
  987  */
  988 int
  989 nfscl_getbytelock(vnode_t vp, u_int64_t off, u_int64_t len,
  990     short type, struct ucred *cred, NFSPROC_T *p, struct nfsclclient *rclp,
  991     int recovery, void *id, int flags, u_int8_t *rownp, u_int8_t *ropenownp,
  992     struct nfscllockowner **lpp, int *newonep, int *donelocallyp)
  993 {
  994         struct nfscllockowner *lp;
  995         struct nfsclopen *op;
  996         struct nfsclclient *clp;
  997         struct nfscllockowner *nlp;
  998         struct nfscllock *nlop, *otherlop;
  999         struct nfscldeleg *dp = NULL, *ldp = NULL;
 1000         struct nfscllockownerhead *lhp = NULL;
 1001         struct nfsnode *np;
 1002         u_int8_t own[NFSV4CL_LOCKNAMELEN], *ownp, openown[NFSV4CL_LOCKNAMELEN];
 1003         u_int8_t *openownp;
 1004         int error = 0, ret, donelocally = 0;
 1005         u_int32_t mode;
 1006 
 1007         /* For Lock Ops, the open mode doesn't matter, so use 0 to match any. */
 1008         mode = 0;
 1009         np = VTONFS(vp);
 1010         *lpp = NULL;
 1011         lp = NULL;
 1012         *newonep = 0;
 1013         *donelocallyp = 0;
 1014 
 1015         /*
 1016          * Might need these, so MALLOC them now, to
 1017          * avoid a tsleep() in MALLOC later.
 1018          */
 1019         nlp = malloc(
 1020             sizeof (struct nfscllockowner), M_NFSCLLOCKOWNER, M_WAITOK);
 1021         otherlop = malloc(
 1022             sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
 1023         nlop = malloc(
 1024             sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
 1025         nlop->nfslo_type = type;
 1026         nlop->nfslo_first = off;
 1027         if (len == NFS64BITSSET) {
 1028                 nlop->nfslo_end = NFS64BITSSET;
 1029         } else {
 1030                 nlop->nfslo_end = off + len;
 1031                 if (nlop->nfslo_end <= nlop->nfslo_first)
 1032                         error = NFSERR_INVAL;
 1033         }
 1034 
 1035         if (!error) {
 1036                 if (recovery)
 1037                         clp = rclp;
 1038                 else
 1039                         error = nfscl_getcl(vnode_mount(vp), cred, p, 1, true,
 1040                             &clp);
 1041         }
 1042         if (error) {
 1043                 free(nlp, M_NFSCLLOCKOWNER);
 1044                 free(otherlop, M_NFSCLLOCK);
 1045                 free(nlop, M_NFSCLLOCK);
 1046                 return (error);
 1047         }
 1048 
 1049         op = NULL;
 1050         if (recovery) {
 1051                 ownp = rownp;
 1052                 openownp = ropenownp;
 1053         } else {
 1054                 nfscl_filllockowner(id, own, flags);
 1055                 ownp = own;
 1056                 if (NFSHASONEOPENOWN(VFSTONFS(vnode_mount(vp))))
 1057                         nfscl_filllockowner(NULL, openown, F_POSIX);
 1058                 else
 1059                         nfscl_filllockowner(p->td_proc, openown, F_POSIX);
 1060                 openownp = openown;
 1061         }
 1062         if (!recovery) {
 1063                 NFSLOCKCLSTATE();
 1064                 /*
 1065                  * First, search for a delegation. If one exists for this file,
 1066                  * the lock can be done locally against it, so long as there
 1067                  * isn't a local lock conflict.
 1068                  */
 1069                 ldp = dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
 1070                     np->n_fhp->nfh_len);
 1071                 /* Just sanity check for correct type of delegation */
 1072                 if (dp != NULL && ((dp->nfsdl_flags &
 1073                     (NFSCLDL_RECALL | NFSCLDL_DELEGRET)) != 0 ||
 1074                      (type == F_WRLCK &&
 1075                       (dp->nfsdl_flags & NFSCLDL_WRITE) == 0)))
 1076                         dp = NULL;
 1077         }
 1078         if (dp != NULL) {
 1079                 /* Now, find an open and maybe a lockowner. */
 1080                 ret = nfscl_getopen(&dp->nfsdl_owner, np->n_fhp->nfh_fh,
 1081                     np->n_fhp->nfh_len, openownp, ownp, mode, NULL, &op);
 1082                 if (ret)
 1083                         ret = nfscl_getopen(&clp->nfsc_owner,
 1084                             np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp,
 1085                             ownp, mode, NULL, &op);
 1086                 if (!ret) {
 1087                         lhp = &dp->nfsdl_lock;
 1088                         TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
 1089                         TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list);
 1090                         dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
 1091                         donelocally = 1;
 1092                 } else {
 1093                         dp = NULL;
 1094                 }
 1095         }
 1096         if (!donelocally) {
 1097                 /*
 1098                  * Get the related Open and maybe lockowner.
 1099                  */
 1100                 error = nfscl_getopen(&clp->nfsc_owner,
 1101                     np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp,
 1102                     ownp, mode, &lp, &op);
 1103                 if (!error)
 1104                         lhp = &op->nfso_lock;
 1105         }
 1106         if (!error && !recovery)
 1107                 error = nfscl_localconflict(clp, np->n_fhp->nfh_fh,
 1108                     np->n_fhp->nfh_len, nlop, ownp, ldp, NULL);
 1109         if (error) {
 1110                 if (!recovery) {
 1111                         nfscl_clrelease(clp);
 1112                         NFSUNLOCKCLSTATE();
 1113                 }
 1114                 free(nlp, M_NFSCLLOCKOWNER);
 1115                 free(otherlop, M_NFSCLLOCK);
 1116                 free(nlop, M_NFSCLLOCK);
 1117                 return (error);
 1118         }
 1119 
 1120         /*
 1121          * Ok, see if a lockowner exists and create one, as required.
 1122          */
 1123         if (lp == NULL)
 1124                 LIST_FOREACH(lp, lhp, nfsl_list) {
 1125                         if (!NFSBCMP(lp->nfsl_owner, ownp, NFSV4CL_LOCKNAMELEN))
 1126                                 break;
 1127                 }
 1128         if (lp == NULL) {
 1129                 NFSBCOPY(ownp, nlp->nfsl_owner, NFSV4CL_LOCKNAMELEN);
 1130                 if (recovery)
 1131                         NFSBCOPY(ropenownp, nlp->nfsl_openowner,
 1132                             NFSV4CL_LOCKNAMELEN);
 1133                 else
 1134                         NFSBCOPY(op->nfso_own->nfsow_owner, nlp->nfsl_openowner,
 1135                             NFSV4CL_LOCKNAMELEN);
 1136                 nlp->nfsl_seqid = 0;
 1137                 nlp->nfsl_lockflags = flags;
 1138                 nlp->nfsl_inprog = NULL;
 1139                 nfscl_lockinit(&nlp->nfsl_rwlock);
 1140                 LIST_INIT(&nlp->nfsl_lock);
 1141                 if (donelocally) {
 1142                         nlp->nfsl_open = NULL;
 1143                         nfsstatsv1.cllocallockowners++;
 1144                 } else {
 1145                         nlp->nfsl_open = op;
 1146                         nfsstatsv1.cllockowners++;
 1147                 }
 1148                 LIST_INSERT_HEAD(lhp, nlp, nfsl_list);
 1149                 lp = nlp;
 1150                 nlp = NULL;
 1151                 *newonep = 1;
 1152         }
 1153 
 1154         /*
 1155          * Now, update the byte ranges for locks.
 1156          */
 1157         ret = nfscl_updatelock(lp, &nlop, &otherlop, donelocally);
 1158         if (!ret)
 1159                 donelocally = 1;
 1160         if (donelocally) {
 1161                 *donelocallyp = 1;
 1162                 if (!recovery)
 1163                         nfscl_clrelease(clp);
 1164         } else {
 1165                 /*
 1166                  * Serial modifications on the lock owner for multiple threads
 1167                  * for the same process using a read/write lock.
 1168                  */
 1169                 if (!recovery)
 1170                         nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR);
 1171         }
 1172         if (!recovery)
 1173                 NFSUNLOCKCLSTATE();
 1174 
 1175         if (nlp)
 1176                 free(nlp, M_NFSCLLOCKOWNER);
 1177         if (nlop)
 1178                 free(nlop, M_NFSCLLOCK);
 1179         if (otherlop)
 1180                 free(otherlop, M_NFSCLLOCK);
 1181 
 1182         *lpp = lp;
 1183         return (0);
 1184 }
 1185 
 1186 /*
 1187  * Called to unlock a byte range, for LockU.
 1188  */
 1189 int
 1190 nfscl_relbytelock(vnode_t vp, u_int64_t off, u_int64_t len,
 1191     __unused struct ucred *cred, NFSPROC_T *p, int callcnt,
 1192     struct nfsclclient *clp, void *id, int flags,
 1193     struct nfscllockowner **lpp, int *dorpcp)
 1194 {
 1195         struct nfscllockowner *lp;
 1196         struct nfsclowner *owp;
 1197         struct nfsclopen *op;
 1198         struct nfscllock *nlop, *other_lop = NULL;
 1199         struct nfscldeleg *dp;
 1200         struct nfsnode *np;
 1201         u_int8_t own[NFSV4CL_LOCKNAMELEN];
 1202         int ret = 0, fnd;
 1203 
 1204         np = VTONFS(vp);
 1205         *lpp = NULL;
 1206         *dorpcp = 0;
 1207 
 1208         /*
 1209          * Might need these, so MALLOC them now, to
 1210          * avoid a tsleep() in MALLOC later.
 1211          */
 1212         nlop = malloc(
 1213             sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
 1214         nlop->nfslo_type = F_UNLCK;
 1215         nlop->nfslo_first = off;
 1216         if (len == NFS64BITSSET) {
 1217                 nlop->nfslo_end = NFS64BITSSET;
 1218         } else {
 1219                 nlop->nfslo_end = off + len;
 1220                 if (nlop->nfslo_end <= nlop->nfslo_first) {
 1221                         free(nlop, M_NFSCLLOCK);
 1222                         return (NFSERR_INVAL);
 1223                 }
 1224         }
 1225         if (callcnt == 0) {
 1226                 other_lop = malloc(
 1227                     sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
 1228                 *other_lop = *nlop;
 1229         }
 1230         nfscl_filllockowner(id, own, flags);
 1231         dp = NULL;
 1232         NFSLOCKCLSTATE();
 1233         if (callcnt == 0)
 1234                 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
 1235                     np->n_fhp->nfh_len);
 1236 
 1237         /*
 1238          * First, unlock any local regions on a delegation.
 1239          */
 1240         if (dp != NULL) {
 1241                 /* Look for this lockowner. */
 1242                 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
 1243                         if (!NFSBCMP(lp->nfsl_owner, own,
 1244                             NFSV4CL_LOCKNAMELEN))
 1245                                 break;
 1246                 }
 1247                 if (lp != NULL)
 1248                         /* Use other_lop, so nlop is still available */
 1249                         (void)nfscl_updatelock(lp, &other_lop, NULL, 1);
 1250         }
 1251 
 1252         /*
 1253          * Now, find a matching open/lockowner that hasn't already been done,
 1254          * as marked by nfsl_inprog.
 1255          */
 1256         lp = NULL;
 1257         fnd = 0;
 1258         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
 1259             LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
 1260                 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
 1261                     !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
 1262                     LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
 1263                         if (lp->nfsl_inprog == NULL &&
 1264                             !NFSBCMP(lp->nfsl_owner, own,
 1265                              NFSV4CL_LOCKNAMELEN)) {
 1266                                 fnd = 1;
 1267                                 break;
 1268                         }
 1269                     }
 1270                     if (fnd)
 1271                         break;
 1272                 }
 1273             }
 1274             if (fnd)
 1275                 break;
 1276         }
 1277 
 1278         if (lp != NULL) {
 1279                 ret = nfscl_updatelock(lp, &nlop, NULL, 0);
 1280                 if (ret)
 1281                         *dorpcp = 1;
 1282                 /*
 1283                  * Serial modifications on the lock owner for multiple
 1284                  * threads for the same process using a read/write lock.
 1285                  */
 1286                 lp->nfsl_inprog = p;
 1287                 nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR);
 1288                 *lpp = lp;
 1289         }
 1290         NFSUNLOCKCLSTATE();
 1291         if (nlop)
 1292                 free(nlop, M_NFSCLLOCK);
 1293         if (other_lop)
 1294                 free(other_lop, M_NFSCLLOCK);
 1295         return (0);
 1296 }
 1297 
 1298 /*
 1299  * Release all lockowners marked in progess for this process and file.
 1300  */
 1301 void
 1302 nfscl_releasealllocks(struct nfsclclient *clp, vnode_t vp, NFSPROC_T *p,
 1303     void *id, int flags)
 1304 {
 1305         struct nfsclowner *owp;
 1306         struct nfsclopen *op;
 1307         struct nfscllockowner *lp;
 1308         struct nfsnode *np;
 1309         u_int8_t own[NFSV4CL_LOCKNAMELEN];
 1310 
 1311         np = VTONFS(vp);
 1312         nfscl_filllockowner(id, own, flags);
 1313         NFSLOCKCLSTATE();
 1314         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
 1315             LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
 1316                 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
 1317                     !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
 1318                     LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
 1319                         if (lp->nfsl_inprog == p &&
 1320                             !NFSBCMP(lp->nfsl_owner, own,
 1321                             NFSV4CL_LOCKNAMELEN)) {
 1322                             lp->nfsl_inprog = NULL;
 1323                             nfscl_lockunlock(&lp->nfsl_rwlock);
 1324                         }
 1325                     }
 1326                 }
 1327             }
 1328         }
 1329         nfscl_clrelease(clp);
 1330         NFSUNLOCKCLSTATE();
 1331 }
 1332 
 1333 /*
 1334  * Called to find out if any bytes within the byte range specified are
 1335  * write locked by the calling process. Used to determine if flushing
 1336  * is required before a LockU.
 1337  * If in doubt, return 1, so the flush will occur.
 1338  */
 1339 int
 1340 nfscl_checkwritelocked(vnode_t vp, struct flock *fl,
 1341     struct ucred *cred, NFSPROC_T *p, void *id, int flags)
 1342 {
 1343         struct nfsclowner *owp;
 1344         struct nfscllockowner *lp;
 1345         struct nfsclopen *op;
 1346         struct nfsclclient *clp;
 1347         struct nfscllock *lop;
 1348         struct nfscldeleg *dp;
 1349         struct nfsnode *np;
 1350         u_int64_t off, end;
 1351         u_int8_t own[NFSV4CL_LOCKNAMELEN];
 1352         int error = 0;
 1353 
 1354         np = VTONFS(vp);
 1355         switch (fl->l_whence) {
 1356         case SEEK_SET:
 1357         case SEEK_CUR:
 1358                 /*
 1359                  * Caller is responsible for adding any necessary offset
 1360                  * when SEEK_CUR is used.
 1361                  */
 1362                 off = fl->l_start;
 1363                 break;
 1364         case SEEK_END:
 1365                 off = np->n_size + fl->l_start;
 1366                 break;
 1367         default:
 1368                 return (1);
 1369         }
 1370         if (fl->l_len != 0) {
 1371                 end = off + fl->l_len;
 1372                 if (end < off)
 1373                         return (1);
 1374         } else {
 1375                 end = NFS64BITSSET;
 1376         }
 1377 
 1378         error = nfscl_getcl(vnode_mount(vp), cred, p, 1, true, &clp);
 1379         if (error)
 1380                 return (1);
 1381         nfscl_filllockowner(id, own, flags);
 1382         NFSLOCKCLSTATE();
 1383 
 1384         /*
 1385          * First check the delegation locks.
 1386          */
 1387         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
 1388         if (dp != NULL) {
 1389                 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
 1390                         if (!NFSBCMP(lp->nfsl_owner, own,
 1391                             NFSV4CL_LOCKNAMELEN))
 1392                                 break;
 1393                 }
 1394                 if (lp != NULL) {
 1395                         LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
 1396                                 if (lop->nfslo_first >= end)
 1397                                         break;
 1398                                 if (lop->nfslo_end <= off)
 1399                                         continue;
 1400                                 if (lop->nfslo_type == F_WRLCK) {
 1401                                         nfscl_clrelease(clp);
 1402                                         NFSUNLOCKCLSTATE();
 1403                                         return (1);
 1404                                 }
 1405                         }
 1406                 }
 1407         }
 1408 
 1409         /*
 1410          * Now, check state against the server.
 1411          */
 1412         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
 1413             LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
 1414                 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
 1415                     !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
 1416                     LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
 1417                         if (!NFSBCMP(lp->nfsl_owner, own,
 1418                             NFSV4CL_LOCKNAMELEN))
 1419                             break;
 1420                     }
 1421                     if (lp != NULL) {
 1422                         LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
 1423                             if (lop->nfslo_first >= end)
 1424                                 break;
 1425                             if (lop->nfslo_end <= off)
 1426                                 continue;
 1427                             if (lop->nfslo_type == F_WRLCK) {
 1428                                 nfscl_clrelease(clp);
 1429                                 NFSUNLOCKCLSTATE();
 1430                                 return (1);
 1431                             }
 1432                         }
 1433                     }
 1434                 }
 1435             }
 1436         }
 1437         nfscl_clrelease(clp);
 1438         NFSUNLOCKCLSTATE();
 1439         return (0);
 1440 }
 1441 
 1442 /*
 1443  * Release a byte range lock owner structure.
 1444  */
 1445 void
 1446 nfscl_lockrelease(struct nfscllockowner *lp, int error, int candelete)
 1447 {
 1448         struct nfsclclient *clp;
 1449 
 1450         if (lp == NULL)
 1451                 return;
 1452         NFSLOCKCLSTATE();
 1453         clp = lp->nfsl_open->nfso_own->nfsow_clp;
 1454         if (error != 0 && candelete &&
 1455             (lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED) == 0)
 1456                 nfscl_freelockowner(lp, 0);
 1457         else
 1458                 nfscl_lockunlock(&lp->nfsl_rwlock);
 1459         nfscl_clrelease(clp);
 1460         NFSUNLOCKCLSTATE();
 1461 }
 1462 
 1463 /*
 1464  * Unlink the open structure.
 1465  */
 1466 static void
 1467 nfscl_unlinkopen(struct nfsclopen *op)
 1468 {
 1469 
 1470         LIST_REMOVE(op, nfso_list);
 1471 }
 1472 
 1473 /*
 1474  * Free up an open structure and any associated byte range lock structures.
 1475  */
 1476 void
 1477 nfscl_freeopen(struct nfsclopen *op, int local, bool unlink)
 1478 {
 1479 
 1480         if (unlink)
 1481                 nfscl_unlinkopen(op);
 1482         nfscl_freealllocks(&op->nfso_lock, local);
 1483         free(op, M_NFSCLOPEN);
 1484         if (local)
 1485                 nfsstatsv1.cllocalopens--;
 1486         else
 1487                 nfsstatsv1.clopens--;
 1488 }
 1489 
 1490 /*
 1491  * Free up all lock owners and associated locks.
 1492  */
 1493 static void
 1494 nfscl_freealllocks(struct nfscllockownerhead *lhp, int local)
 1495 {
 1496         struct nfscllockowner *lp, *nlp;
 1497 
 1498         LIST_FOREACH_SAFE(lp, lhp, nfsl_list, nlp) {
 1499                 if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED))
 1500                         panic("nfscllckw");
 1501                 nfscl_freelockowner(lp, local);
 1502         }
 1503 }
 1504 
 1505 /*
 1506  * Called for an Open when NFSERR_EXPIRED is received from the server.
 1507  * If there are no byte range locks nor a Share Deny lost, try to do a
 1508  * fresh Open. Otherwise, free the open.
 1509  */
 1510 static int
 1511 nfscl_expireopen(struct nfsclclient *clp, struct nfsclopen *op,
 1512     struct nfsmount *nmp, struct ucred *cred, NFSPROC_T *p)
 1513 {
 1514         struct nfscllockowner *lp;
 1515         struct nfscldeleg *dp;
 1516         int mustdelete = 0, error;
 1517 
 1518         /*
 1519          * Look for any byte range lock(s).
 1520          */
 1521         LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
 1522                 if (!LIST_EMPTY(&lp->nfsl_lock)) {
 1523                         mustdelete = 1;
 1524                         break;
 1525                 }
 1526         }
 1527 
 1528         /*
 1529          * If no byte range lock(s) nor a Share deny, try to re-open.
 1530          */
 1531         if (!mustdelete && (op->nfso_mode & NFSLCK_DENYBITS) == 0) {
 1532                 newnfs_copycred(&op->nfso_cred, cred);
 1533                 dp = NULL;
 1534                 error = nfsrpc_reopen(nmp, op->nfso_fh,
 1535                     op->nfso_fhlen, op->nfso_mode, op, &dp, cred, p);
 1536                 if (error) {
 1537                         mustdelete = 1;
 1538                         if (dp != NULL) {
 1539                                 free(dp, M_NFSCLDELEG);
 1540                                 dp = NULL;
 1541                         }
 1542                 }
 1543                 if (dp != NULL)
 1544                         nfscl_deleg(nmp->nm_mountp, clp, op->nfso_fh,
 1545                             op->nfso_fhlen, cred, p, &dp);
 1546         }
 1547 
 1548         /*
 1549          * If a byte range lock or Share deny or couldn't re-open, free it.
 1550          */
 1551         if (mustdelete)
 1552                 nfscl_freeopen(op, 0, true);
 1553         return (mustdelete);
 1554 }
 1555 
 1556 /*
 1557  * Free up an open owner structure.
 1558  */
 1559 static void
 1560 nfscl_freeopenowner(struct nfsclowner *owp, int local)
 1561 {
 1562         int owned;
 1563 
 1564         /*
 1565          * Make sure the NFSCLSTATE mutex is held, to avoid races with
 1566          * calls in nfscl_renewthread() that do not hold a reference
 1567          * count on the nfsclclient and just the mutex.
 1568          * The mutex will not be held for calls done with the exclusive
 1569          * nfsclclient lock held, in particular, nfscl_hasexpired()
 1570          * and nfscl_recalldeleg() might do this.
 1571          */
 1572         owned = mtx_owned(NFSCLSTATEMUTEXPTR);
 1573         if (owned == 0)
 1574                 NFSLOCKCLSTATE();
 1575         LIST_REMOVE(owp, nfsow_list);
 1576         if (owned == 0)
 1577                 NFSUNLOCKCLSTATE();
 1578         free(owp, M_NFSCLOWNER);
 1579         if (local)
 1580                 nfsstatsv1.cllocalopenowners--;
 1581         else
 1582                 nfsstatsv1.clopenowners--;
 1583 }
 1584 
 1585 /*
 1586  * Free up a byte range lock owner structure.
 1587  */
 1588 void
 1589 nfscl_freelockowner(struct nfscllockowner *lp, int local)
 1590 {
 1591         struct nfscllock *lop, *nlop;
 1592         int owned;
 1593 
 1594         /*
 1595          * Make sure the NFSCLSTATE mutex is held, to avoid races with
 1596          * calls in nfscl_renewthread() that do not hold a reference
 1597          * count on the nfsclclient and just the mutex.
 1598          * The mutex will not be held for calls done with the exclusive
 1599          * nfsclclient lock held, in particular, nfscl_hasexpired()
 1600          * and nfscl_recalldeleg() might do this.
 1601          */
 1602         owned = mtx_owned(NFSCLSTATEMUTEXPTR);
 1603         if (owned == 0)
 1604                 NFSLOCKCLSTATE();
 1605         LIST_REMOVE(lp, nfsl_list);
 1606         if (owned == 0)
 1607                 NFSUNLOCKCLSTATE();
 1608         LIST_FOREACH_SAFE(lop, &lp->nfsl_lock, nfslo_list, nlop) {
 1609                 nfscl_freelock(lop, local);
 1610         }
 1611         free(lp, M_NFSCLLOCKOWNER);
 1612         if (local)
 1613                 nfsstatsv1.cllocallockowners--;
 1614         else
 1615                 nfsstatsv1.cllockowners--;
 1616 }
 1617 
 1618 /*
 1619  * Free up a byte range lock structure.
 1620  */
 1621 void
 1622 nfscl_freelock(struct nfscllock *lop, int local)
 1623 {
 1624 
 1625         LIST_REMOVE(lop, nfslo_list);
 1626         free(lop, M_NFSCLLOCK);
 1627         if (local)
 1628                 nfsstatsv1.cllocallocks--;
 1629         else
 1630                 nfsstatsv1.cllocks--;
 1631 }
 1632 
 1633 /*
 1634  * Clean out the state related to a delegation.
 1635  */
 1636 static void
 1637 nfscl_cleandeleg(struct nfscldeleg *dp)
 1638 {
 1639         struct nfsclowner *owp, *nowp;
 1640         struct nfsclopen *op;
 1641 
 1642         LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) {
 1643                 op = LIST_FIRST(&owp->nfsow_open);
 1644                 if (op != NULL) {
 1645                         if (LIST_NEXT(op, nfso_list) != NULL)
 1646                                 panic("nfscleandel");
 1647                         nfscl_freeopen(op, 1, true);
 1648                 }
 1649                 nfscl_freeopenowner(owp, 1);
 1650         }
 1651         nfscl_freealllocks(&dp->nfsdl_lock, 1);
 1652 }
 1653 
 1654 /*
 1655  * Free a delegation.
 1656  */
 1657 static void
 1658 nfscl_freedeleg(struct nfscldeleghead *hdp, struct nfscldeleg *dp, bool freeit)
 1659 {
 1660 
 1661         TAILQ_REMOVE(hdp, dp, nfsdl_list);
 1662         LIST_REMOVE(dp, nfsdl_hash);
 1663         if (freeit)
 1664                 free(dp, M_NFSCLDELEG);
 1665         nfsstatsv1.cldelegates--;
 1666         nfscl_delegcnt--;
 1667 }
 1668 
 1669 /*
 1670  * Free up all state related to this client structure.
 1671  */
 1672 static void
 1673 nfscl_cleanclient(struct nfsclclient *clp)
 1674 {
 1675         struct nfsclowner *owp, *nowp;
 1676         struct nfsclopen *op, *nop;
 1677         struct nfscllayout *lyp, *nlyp;
 1678         struct nfscldevinfo *dip, *ndip;
 1679 
 1680         TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp)
 1681                 nfscl_freelayout(lyp);
 1682 
 1683         LIST_FOREACH_SAFE(dip, &clp->nfsc_devinfo, nfsdi_list, ndip)
 1684                 nfscl_freedevinfo(dip);
 1685 
 1686         /* Now, all the OpenOwners, etc. */
 1687         LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
 1688                 LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
 1689                         nfscl_freeopen(op, 0, true);
 1690                 }
 1691                 nfscl_freeopenowner(owp, 0);
 1692         }
 1693 }
 1694 
 1695 /*
 1696  * Called when an NFSERR_EXPIRED is received from the server.
 1697  */
 1698 static void
 1699 nfscl_expireclient(struct nfsclclient *clp, struct nfsmount *nmp,
 1700     struct ucred *cred, NFSPROC_T *p)
 1701 {
 1702         struct nfsclowner *owp, *nowp, *towp;
 1703         struct nfsclopen *op, *nop, *top;
 1704         struct nfscldeleg *dp, *ndp;
 1705         int ret, printed = 0;
 1706 
 1707         /*
 1708          * First, merge locally issued Opens into the list for the server.
 1709          */
 1710         dp = TAILQ_FIRST(&clp->nfsc_deleg);
 1711         while (dp != NULL) {
 1712             ndp = TAILQ_NEXT(dp, nfsdl_list);
 1713             owp = LIST_FIRST(&dp->nfsdl_owner);
 1714             while (owp != NULL) {
 1715                 nowp = LIST_NEXT(owp, nfsow_list);
 1716                 op = LIST_FIRST(&owp->nfsow_open);
 1717                 if (op != NULL) {
 1718                     if (LIST_NEXT(op, nfso_list) != NULL)
 1719                         panic("nfsclexp");
 1720                     LIST_FOREACH(towp, &clp->nfsc_owner, nfsow_list) {
 1721                         if (!NFSBCMP(towp->nfsow_owner, owp->nfsow_owner,
 1722                             NFSV4CL_LOCKNAMELEN))
 1723                             break;
 1724                     }
 1725                     if (towp != NULL) {
 1726                         /* Merge opens in */
 1727                         LIST_FOREACH(top, &towp->nfsow_open, nfso_list) {
 1728                             if (top->nfso_fhlen == op->nfso_fhlen &&
 1729                                 !NFSBCMP(top->nfso_fh, op->nfso_fh,
 1730                                  op->nfso_fhlen)) {
 1731                                 top->nfso_mode |= op->nfso_mode;
 1732                                 top->nfso_opencnt += op->nfso_opencnt;
 1733                                 break;
 1734                             }
 1735                         }
 1736                         if (top == NULL) {
 1737                             /* Just add the open to the owner list */
 1738                             LIST_REMOVE(op, nfso_list);
 1739                             op->nfso_own = towp;
 1740                             LIST_INSERT_HEAD(&towp->nfsow_open, op, nfso_list);
 1741                             nfsstatsv1.cllocalopens--;
 1742                             nfsstatsv1.clopens++;
 1743                         }
 1744                     } else {
 1745                         /* Just add the openowner to the client list */
 1746                         LIST_REMOVE(owp, nfsow_list);
 1747                         owp->nfsow_clp = clp;
 1748                         LIST_INSERT_HEAD(&clp->nfsc_owner, owp, nfsow_list);
 1749                         nfsstatsv1.cllocalopenowners--;
 1750                         nfsstatsv1.clopenowners++;
 1751                         nfsstatsv1.cllocalopens--;
 1752                         nfsstatsv1.clopens++;
 1753                     }
 1754                 }
 1755                 owp = nowp;
 1756             }
 1757             if (!printed && !LIST_EMPTY(&dp->nfsdl_lock)) {
 1758                 printed = 1;
 1759                 printf("nfsv4 expired locks lost\n");
 1760             }
 1761             nfscl_cleandeleg(dp);
 1762             nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
 1763             dp = ndp;
 1764         }
 1765         if (!TAILQ_EMPTY(&clp->nfsc_deleg))
 1766             panic("nfsclexp");
 1767 
 1768         /*
 1769          * Now, try and reopen against the server.
 1770          */
 1771         LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
 1772                 owp->nfsow_seqid = 0;
 1773                 LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
 1774                         ret = nfscl_expireopen(clp, op, nmp, cred, p);
 1775                         if (ret && !printed) {
 1776                                 printed = 1;
 1777                                 printf("nfsv4 expired locks lost\n");
 1778                         }
 1779                 }
 1780                 if (LIST_EMPTY(&owp->nfsow_open))
 1781                         nfscl_freeopenowner(owp, 0);
 1782         }
 1783 }
 1784 
 1785 /*
 1786  * This function must be called after the process represented by "own" has
 1787  * exited. Must be called with CLSTATE lock held.
 1788  */
 1789 static void
 1790 nfscl_cleanup_common(struct nfsclclient *clp, u_int8_t *own)
 1791 {
 1792         struct nfsclowner *owp, *nowp;
 1793         struct nfscllockowner *lp;
 1794         struct nfscldeleg *dp;
 1795 
 1796         /* First, get rid of local locks on delegations. */
 1797         TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
 1798                 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
 1799                     if (!NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
 1800                         if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED))
 1801                             panic("nfscllckw");
 1802                         nfscl_freelockowner(lp, 1);
 1803                         break;
 1804                     }
 1805                 }
 1806         }
 1807         owp = LIST_FIRST(&clp->nfsc_owner);
 1808         while (owp != NULL) {
 1809                 nowp = LIST_NEXT(owp, nfsow_list);
 1810                 if (!NFSBCMP(owp->nfsow_owner, own,
 1811                     NFSV4CL_LOCKNAMELEN)) {
 1812                         /*
 1813                          * If there are children that haven't closed the
 1814                          * file descriptors yet, the opens will still be
 1815                          * here. For that case, let the renew thread clear
 1816                          * out the OpenOwner later.
 1817                          */
 1818                         if (LIST_EMPTY(&owp->nfsow_open))
 1819                                 nfscl_freeopenowner(owp, 0);
 1820                         else
 1821                                 owp->nfsow_defunct = 1;
 1822                         break;
 1823                 }
 1824                 owp = nowp;
 1825         }
 1826 }
 1827 
 1828 /*
 1829  * Find open/lock owners for processes that have exited.
 1830  */
 1831 static void
 1832 nfscl_cleanupkext(struct nfsclclient *clp, struct nfscllockownerfhhead *lhp)
 1833 {
 1834         struct nfsclowner *owp, *nowp;
 1835         struct nfsclopen *op;
 1836         struct nfscllockowner *lp, *nlp;
 1837         struct nfscldeleg *dp;
 1838         uint8_t own[NFSV4CL_LOCKNAMELEN];
 1839 
 1840         NFSPROCLISTLOCK();
 1841         NFSLOCKCLSTATE();
 1842         LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
 1843                 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
 1844                         LIST_FOREACH_SAFE(lp, &op->nfso_lock, nfsl_list, nlp) {
 1845                                 if (LIST_EMPTY(&lp->nfsl_lock))
 1846                                         nfscl_emptylockowner(lp, lhp);
 1847                         }
 1848                 }
 1849                 if (nfscl_procdoesntexist(owp->nfsow_owner)) {
 1850                         memcpy(own, owp->nfsow_owner, NFSV4CL_LOCKNAMELEN);
 1851                         nfscl_cleanup_common(clp, own);
 1852                 }
 1853         }
 1854 
 1855         /*
 1856          * For the single open_owner case, these lock owners need to be
 1857          * checked to see if they still exist separately.
 1858          * This is because nfscl_procdoesntexist() never returns true for
 1859          * the single open_owner so that the above doesn't ever call
 1860          * nfscl_cleanup_common().
 1861          */
 1862         TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
 1863                 LIST_FOREACH_SAFE(lp, &dp->nfsdl_lock, nfsl_list, nlp) {
 1864                         if (nfscl_procdoesntexist(lp->nfsl_owner)) {
 1865                                 memcpy(own, lp->nfsl_owner,
 1866                                     NFSV4CL_LOCKNAMELEN);
 1867                                 nfscl_cleanup_common(clp, own);
 1868                         }
 1869                 }
 1870         }
 1871         NFSUNLOCKCLSTATE();
 1872         NFSPROCLISTUNLOCK();
 1873 }
 1874 
 1875 /*
 1876  * Take the empty lock owner and move it to the local lhp list if the
 1877  * associated process no longer exists.
 1878  */
 1879 static void
 1880 nfscl_emptylockowner(struct nfscllockowner *lp,
 1881     struct nfscllockownerfhhead *lhp)
 1882 {
 1883         struct nfscllockownerfh *lfhp, *mylfhp;
 1884         struct nfscllockowner *nlp;
 1885         int fnd_it;
 1886 
 1887         /* If not a Posix lock owner, just return. */
 1888         if ((lp->nfsl_lockflags & F_POSIX) == 0)
 1889                 return;
 1890 
 1891         fnd_it = 0;
 1892         mylfhp = NULL;
 1893         /*
 1894          * First, search to see if this lock owner is already in the list.
 1895          * If it is, then the associated process no longer exists.
 1896          */
 1897         SLIST_FOREACH(lfhp, lhp, nfslfh_list) {
 1898                 if (lfhp->nfslfh_len == lp->nfsl_open->nfso_fhlen &&
 1899                     !NFSBCMP(lfhp->nfslfh_fh, lp->nfsl_open->nfso_fh,
 1900                     lfhp->nfslfh_len))
 1901                         mylfhp = lfhp;
 1902                 LIST_FOREACH(nlp, &lfhp->nfslfh_lock, nfsl_list)
 1903                         if (!NFSBCMP(nlp->nfsl_owner, lp->nfsl_owner,
 1904                             NFSV4CL_LOCKNAMELEN))
 1905                                 fnd_it = 1;
 1906         }
 1907         /* If not found, check if process still exists. */
 1908         if (fnd_it == 0 && nfscl_procdoesntexist(lp->nfsl_owner) == 0)
 1909                 return;
 1910 
 1911         /* Move the lock owner over to the local list. */
 1912         if (mylfhp == NULL) {
 1913                 mylfhp = malloc(sizeof(struct nfscllockownerfh), M_TEMP,
 1914                     M_NOWAIT);
 1915                 if (mylfhp == NULL)
 1916                         return;
 1917                 mylfhp->nfslfh_len = lp->nfsl_open->nfso_fhlen;
 1918                 NFSBCOPY(lp->nfsl_open->nfso_fh, mylfhp->nfslfh_fh,
 1919                     mylfhp->nfslfh_len);
 1920                 LIST_INIT(&mylfhp->nfslfh_lock);
 1921                 SLIST_INSERT_HEAD(lhp, mylfhp, nfslfh_list);
 1922         }
 1923         LIST_REMOVE(lp, nfsl_list);
 1924         LIST_INSERT_HEAD(&mylfhp->nfslfh_lock, lp, nfsl_list);
 1925 }
 1926 
 1927 static int      fake_global;    /* Used to force visibility of MNTK_UNMOUNTF */
 1928 /*
 1929  * Called from nfs umount to free up the clientid.
 1930  */
 1931 void
 1932 nfscl_umount(struct nfsmount *nmp, NFSPROC_T *p, struct nfscldeleghead *dhp)
 1933 {
 1934         struct nfsclclient *clp;
 1935         struct ucred *cred;
 1936         int igotlock;
 1937 
 1938         /*
 1939          * For the case that matters, this is the thread that set
 1940          * MNTK_UNMOUNTF, so it will see it set. The code that follows is
 1941          * done to ensure that any thread executing nfscl_getcl() after
 1942          * this time, will see MNTK_UNMOUNTF set. nfscl_getcl() uses the
 1943          * mutex for NFSLOCKCLSTATE(), so it is "m" for the following
 1944          * explanation, courtesy of Alan Cox.
 1945          * What follows is a snippet from Alan Cox's email at:
 1946          * https://docs.FreeBSD.org/cgi/mid.cgi?BANLkTikR3d65zPHo9==08ZfJ2vmqZucEvw
 1947          * 
 1948          * 1. Set MNTK_UNMOUNTF
 1949          * 2. Acquire a standard FreeBSD mutex "m".
 1950          * 3. Update some data structures.
 1951          * 4. Release mutex "m".
 1952          * 
 1953          * Then, other threads that acquire "m" after step 4 has occurred will
 1954          * see MNTK_UNMOUNTF as set.  But, other threads that beat thread X to
 1955          * step 2 may or may not see MNTK_UNMOUNTF as set.
 1956          */
 1957         NFSLOCKCLSTATE();
 1958         if ((nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) {
 1959                 fake_global++;
 1960                 NFSUNLOCKCLSTATE();
 1961                 NFSLOCKCLSTATE();
 1962         }
 1963 
 1964         clp = nmp->nm_clp;
 1965         if (clp != NULL) {
 1966                 if ((clp->nfsc_flags & NFSCLFLAGS_INITED) == 0)
 1967                         panic("nfscl umount");
 1968         
 1969                 /*
 1970                  * First, handshake with the nfscl renew thread, to terminate
 1971                  * it.
 1972                  */
 1973                 clp->nfsc_flags |= NFSCLFLAGS_UMOUNT;
 1974                 while (clp->nfsc_flags & NFSCLFLAGS_HASTHREAD)
 1975                         (void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT,
 1976                             "nfsclumnt", hz);
 1977         
 1978                 /*
 1979                  * Now, get the exclusive lock on the client state, so
 1980                  * that no uses of the state are still in progress.
 1981                  */
 1982                 do {
 1983                         igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
 1984                             NFSCLSTATEMUTEXPTR, NULL);
 1985                 } while (!igotlock);
 1986                 NFSUNLOCKCLSTATE();
 1987         
 1988                 /*
 1989                  * Free up all the state. It will expire on the server, but
 1990                  * maybe we should do a SetClientId/SetClientIdConfirm so
 1991                  * the server throws it away?
 1992                  */
 1993                 LIST_REMOVE(clp, nfsc_list);
 1994                 nfscl_delegreturnall(clp, p, dhp);
 1995                 cred = newnfs_getcred();
 1996                 if (NFSHASNFSV4N(nmp)) {
 1997                         (void)nfsrpc_destroysession(nmp, clp, cred, p);
 1998                         (void)nfsrpc_destroyclient(nmp, clp, cred, p);
 1999                 } else
 2000                         (void)nfsrpc_setclient(nmp, clp, 0, cred, p);
 2001                 nfscl_cleanclient(clp);
 2002                 nmp->nm_clp = NULL;
 2003                 NFSFREECRED(cred);
 2004                 free(clp, M_NFSCLCLIENT);
 2005         } else
 2006                 NFSUNLOCKCLSTATE();
 2007 }
 2008 
 2009 /*
 2010  * This function is called when a server replies with NFSERR_STALECLIENTID
 2011  * NFSERR_STALESTATEID or NFSERR_BADSESSION. It traverses the clientid lists,
 2012  * doing Opens and Locks with reclaim. If these fail, it deletes the
 2013  * corresponding state.
 2014  */
 2015 static void
 2016 nfscl_recover(struct nfsclclient *clp, struct ucred *cred, NFSPROC_T *p)
 2017 {
 2018         struct nfsclowner *owp, *nowp;
 2019         struct nfsclopen *op, *nop;
 2020         struct nfscllockowner *lp, *nlp;
 2021         struct nfscllock *lop, *nlop;
 2022         struct nfscldeleg *dp, *ndp, *tdp;
 2023         struct nfsmount *nmp;
 2024         struct ucred *tcred;
 2025         struct nfsclopenhead extra_open;
 2026         struct nfscldeleghead extra_deleg;
 2027         struct nfsreq *rep;
 2028         u_int64_t len;
 2029         u_int32_t delegtype = NFSV4OPEN_DELEGATEWRITE, mode;
 2030         int i, igotlock = 0, error, trycnt, firstlock;
 2031         struct nfscllayout *lyp, *nlyp;
 2032         bool recovered_one;
 2033 
 2034         /*
 2035          * First, lock the client structure, so everyone else will
 2036          * block when trying to use state.
 2037          */
 2038         NFSLOCKCLSTATE();
 2039         clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG;
 2040         do {
 2041                 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
 2042                     NFSCLSTATEMUTEXPTR, NULL);
 2043         } while (!igotlock);
 2044         NFSUNLOCKCLSTATE();
 2045 
 2046         nmp = clp->nfsc_nmp;
 2047         if (nmp == NULL)
 2048                 panic("nfscl recover");
 2049 
 2050         /*
 2051          * For now, just get rid of all layouts. There may be a need
 2052          * to do LayoutCommit Ops with reclaim == true later.
 2053          */
 2054         TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp)
 2055                 nfscl_freelayout(lyp);
 2056         TAILQ_INIT(&clp->nfsc_layout);
 2057         for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++)
 2058                 LIST_INIT(&clp->nfsc_layouthash[i]);
 2059 
 2060         trycnt = 5;
 2061         do {
 2062                 error = nfsrpc_setclient(nmp, clp, 1, cred, p);
 2063         } while ((error == NFSERR_STALECLIENTID ||
 2064              error == NFSERR_BADSESSION ||
 2065              error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
 2066         if (error) {
 2067                 NFSLOCKCLSTATE();
 2068                 clp->nfsc_flags &= ~(NFSCLFLAGS_RECOVER |
 2069                     NFSCLFLAGS_RECVRINPROG);
 2070                 wakeup(&clp->nfsc_flags);
 2071                 nfsv4_unlock(&clp->nfsc_lock, 0);
 2072                 NFSUNLOCKCLSTATE();
 2073                 return;
 2074         }
 2075         clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
 2076         clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
 2077 
 2078         /*
 2079          * Mark requests already queued on the server, so that they don't
 2080          * initiate another recovery cycle. Any requests already in the
 2081          * queue that handle state information will have the old stale
 2082          * clientid/stateid and will get a NFSERR_STALESTATEID,
 2083          * NFSERR_STALECLIENTID or NFSERR_BADSESSION reply from the server.
 2084          * This will be translated to NFSERR_STALEDONTRECOVER when
 2085          * R_DONTRECOVER is set.
 2086          */
 2087         NFSLOCKREQ();
 2088         TAILQ_FOREACH(rep, &nfsd_reqq, r_chain) {
 2089                 if (rep->r_nmp == nmp)
 2090                         rep->r_flags |= R_DONTRECOVER;
 2091         }
 2092         NFSUNLOCKREQ();
 2093 
 2094         /*
 2095          * Now, mark all delegations "need reclaim".
 2096          */
 2097         TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list)
 2098                 dp->nfsdl_flags |= NFSCLDL_NEEDRECLAIM;
 2099 
 2100         TAILQ_INIT(&extra_deleg);
 2101         LIST_INIT(&extra_open);
 2102         /*
 2103          * Now traverse the state lists, doing Open and Lock Reclaims.
 2104          */
 2105         tcred = newnfs_getcred();
 2106         recovered_one = false;
 2107         owp = LIST_FIRST(&clp->nfsc_owner);
 2108         while (owp != NULL) {
 2109             nowp = LIST_NEXT(owp, nfsow_list);
 2110             owp->nfsow_seqid = 0;
 2111             op = LIST_FIRST(&owp->nfsow_open);
 2112             while (op != NULL) {
 2113                 nop = LIST_NEXT(op, nfso_list);
 2114                 if (error != NFSERR_NOGRACE && error != NFSERR_BADSESSION) {
 2115                     /* Search for a delegation to reclaim with the open */
 2116                     TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
 2117                         if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
 2118                             continue;
 2119                         if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
 2120                             mode = NFSV4OPEN_ACCESSWRITE;
 2121                             delegtype = NFSV4OPEN_DELEGATEWRITE;
 2122                         } else {
 2123                             mode = NFSV4OPEN_ACCESSREAD;
 2124                             delegtype = NFSV4OPEN_DELEGATEREAD;
 2125                         }
 2126                         if ((op->nfso_mode & mode) == mode &&
 2127                             op->nfso_fhlen == dp->nfsdl_fhlen &&
 2128                             !NFSBCMP(op->nfso_fh, dp->nfsdl_fh, op->nfso_fhlen))
 2129                             break;
 2130                     }
 2131                     ndp = dp;
 2132                     if (dp == NULL)
 2133                         delegtype = NFSV4OPEN_DELEGATENONE;
 2134                     newnfs_copycred(&op->nfso_cred, tcred);
 2135                     error = nfscl_tryopen(nmp, NULL, op->nfso_fh,
 2136                         op->nfso_fhlen, op->nfso_fh, op->nfso_fhlen,
 2137                         op->nfso_mode, op, NULL, 0, &ndp, 1, delegtype,
 2138                         tcred, p);
 2139                     if (!error) {
 2140                         recovered_one = true;
 2141                         /* Handle any replied delegation */
 2142                         if (ndp != NULL && ((ndp->nfsdl_flags & NFSCLDL_WRITE)
 2143                             || NFSMNT_RDONLY(nmp->nm_mountp))) {
 2144                             if ((ndp->nfsdl_flags & NFSCLDL_WRITE))
 2145                                 mode = NFSV4OPEN_ACCESSWRITE;
 2146                             else
 2147                                 mode = NFSV4OPEN_ACCESSREAD;
 2148                             TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
 2149                                 if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
 2150                                     continue;
 2151                                 if ((op->nfso_mode & mode) == mode &&
 2152                                     op->nfso_fhlen == dp->nfsdl_fhlen &&
 2153                                     !NFSBCMP(op->nfso_fh, dp->nfsdl_fh,
 2154                                     op->nfso_fhlen)) {
 2155                                     dp->nfsdl_stateid = ndp->nfsdl_stateid;
 2156                                     dp->nfsdl_sizelimit = ndp->nfsdl_sizelimit;
 2157                                     dp->nfsdl_ace = ndp->nfsdl_ace;
 2158                                     dp->nfsdl_change = ndp->nfsdl_change;
 2159                                     dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
 2160                                     if ((ndp->nfsdl_flags & NFSCLDL_RECALL))
 2161                                         dp->nfsdl_flags |= NFSCLDL_RECALL;
 2162                                     free(ndp, M_NFSCLDELEG);
 2163                                     ndp = NULL;
 2164                                     break;
 2165                                 }
 2166                             }
 2167                         }
 2168                         if (ndp != NULL)
 2169                             TAILQ_INSERT_HEAD(&extra_deleg, ndp, nfsdl_list);
 2170 
 2171                         /* and reclaim all byte range locks */
 2172                         lp = LIST_FIRST(&op->nfso_lock);
 2173                         while (lp != NULL) {
 2174                             nlp = LIST_NEXT(lp, nfsl_list);
 2175                             lp->nfsl_seqid = 0;
 2176                             firstlock = 1;
 2177                             lop = LIST_FIRST(&lp->nfsl_lock);
 2178                             while (lop != NULL) {
 2179                                 nlop = LIST_NEXT(lop, nfslo_list);
 2180                                 if (lop->nfslo_end == NFS64BITSSET)
 2181                                     len = NFS64BITSSET;
 2182                                 else
 2183                                     len = lop->nfslo_end - lop->nfslo_first;
 2184                                 error = nfscl_trylock(nmp, NULL,
 2185                                     op->nfso_fh, op->nfso_fhlen, lp,
 2186                                     firstlock, 1, lop->nfslo_first, len,
 2187                                     lop->nfslo_type, tcred, p);
 2188                                 if (error != 0)
 2189                                     nfscl_freelock(lop, 0);
 2190                                 else
 2191                                     firstlock = 0;
 2192                                 lop = nlop;
 2193                             }
 2194                             /* If no locks, but a lockowner, just delete it. */
 2195                             if (LIST_EMPTY(&lp->nfsl_lock))
 2196                                 nfscl_freelockowner(lp, 0);
 2197                             lp = nlp;
 2198                         }
 2199                     } else if (error == NFSERR_NOGRACE && !recovered_one &&
 2200                         NFSHASNFSV4N(nmp)) {
 2201                         /*
 2202                          * For NFSv4.1/4.2, the NFSERR_EXPIRED case will
 2203                          * actually end up here, since the client will do
 2204                          * a recovery for NFSERR_BADSESSION, but will get
 2205                          * an NFSERR_NOGRACE reply for the first "reclaim"
 2206                          * attempt.
 2207                          * So, call nfscl_expireclient() to recover the
 2208                          * opens as best we can and then do a reclaim
 2209                          * complete and return.
 2210                          */
 2211                         nfsrpc_reclaimcomplete(nmp, cred, p);
 2212                         nfscl_expireclient(clp, nmp, tcred, p);
 2213                         goto out;
 2214                     }
 2215                 }
 2216                 if (error != 0 && error != NFSERR_BADSESSION)
 2217                     nfscl_freeopen(op, 0, true);
 2218                 op = nop;
 2219             }
 2220             owp = nowp;
 2221         }
 2222 
 2223         /*
 2224          * Now, try and get any delegations not yet reclaimed by cobbling
 2225          * to-gether an appropriate open.
 2226          */
 2227         nowp = NULL;
 2228         dp = TAILQ_FIRST(&clp->nfsc_deleg);
 2229         while (dp != NULL) {
 2230             ndp = TAILQ_NEXT(dp, nfsdl_list);
 2231             if ((dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM)) {
 2232                 if (nowp == NULL) {
 2233                     nowp = malloc(
 2234                         sizeof (struct nfsclowner), M_NFSCLOWNER, M_WAITOK);
 2235                     /*
 2236                      * Name must be as long an largest possible
 2237                      * NFSV4CL_LOCKNAMELEN. 12 for now.
 2238                      */
 2239                     NFSBCOPY("RECLAIMDELEG", nowp->nfsow_owner,
 2240                         NFSV4CL_LOCKNAMELEN);
 2241                     LIST_INIT(&nowp->nfsow_open);
 2242                     nowp->nfsow_clp = clp;
 2243                     nowp->nfsow_seqid = 0;
 2244                     nowp->nfsow_defunct = 0;
 2245                     nfscl_lockinit(&nowp->nfsow_rwlock);
 2246                 }
 2247                 nop = NULL;
 2248                 if (error != NFSERR_NOGRACE && error != NFSERR_BADSESSION) {
 2249                     nop = malloc(sizeof (struct nfsclopen) +
 2250                         dp->nfsdl_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
 2251                     nop->nfso_own = nowp;
 2252                     if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
 2253                         nop->nfso_mode = NFSV4OPEN_ACCESSWRITE;
 2254                         delegtype = NFSV4OPEN_DELEGATEWRITE;
 2255                     } else {
 2256                         nop->nfso_mode = NFSV4OPEN_ACCESSREAD;
 2257                         delegtype = NFSV4OPEN_DELEGATEREAD;
 2258                     }
 2259                     nop->nfso_opencnt = 0;
 2260                     nop->nfso_posixlock = 1;
 2261                     nop->nfso_fhlen = dp->nfsdl_fhlen;
 2262                     NFSBCOPY(dp->nfsdl_fh, nop->nfso_fh, dp->nfsdl_fhlen);
 2263                     LIST_INIT(&nop->nfso_lock);
 2264                     nop->nfso_stateid.seqid = 0;
 2265                     nop->nfso_stateid.other[0] = 0;
 2266                     nop->nfso_stateid.other[1] = 0;
 2267                     nop->nfso_stateid.other[2] = 0;
 2268                     newnfs_copycred(&dp->nfsdl_cred, tcred);
 2269                     newnfs_copyincred(tcred, &nop->nfso_cred);
 2270                     tdp = NULL;
 2271                     error = nfscl_tryopen(nmp, NULL, nop->nfso_fh,
 2272                         nop->nfso_fhlen, nop->nfso_fh, nop->nfso_fhlen,
 2273                         nop->nfso_mode, nop, NULL, 0, &tdp, 1,
 2274                         delegtype, tcred, p);
 2275                     if (tdp != NULL) {
 2276                         if ((tdp->nfsdl_flags & NFSCLDL_WRITE))
 2277                             mode = NFSV4OPEN_ACCESSWRITE;
 2278                         else
 2279                             mode = NFSV4OPEN_ACCESSREAD;
 2280                         if ((nop->nfso_mode & mode) == mode &&
 2281                             nop->nfso_fhlen == tdp->nfsdl_fhlen &&
 2282                             !NFSBCMP(nop->nfso_fh, tdp->nfsdl_fh,
 2283                             nop->nfso_fhlen)) {
 2284                             dp->nfsdl_stateid = tdp->nfsdl_stateid;
 2285                             dp->nfsdl_sizelimit = tdp->nfsdl_sizelimit;
 2286                             dp->nfsdl_ace = tdp->nfsdl_ace;
 2287                             dp->nfsdl_change = tdp->nfsdl_change;
 2288                             dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
 2289                             if ((tdp->nfsdl_flags & NFSCLDL_RECALL))
 2290                                 dp->nfsdl_flags |= NFSCLDL_RECALL;
 2291                             free(tdp, M_NFSCLDELEG);
 2292                         } else {
 2293                             TAILQ_INSERT_HEAD(&extra_deleg, tdp, nfsdl_list);
 2294                         }
 2295                     }
 2296                 }
 2297                 if (error) {
 2298                     if (nop != NULL)
 2299                         free(nop, M_NFSCLOPEN);
 2300                     if (error == NFSERR_NOGRACE && !recovered_one &&
 2301                         NFSHASNFSV4N(nmp)) {
 2302                         /*
 2303                          * For NFSv4.1/4.2, the NFSERR_EXPIRED case will
 2304                          * actually end up here, since the client will do
 2305                          * a recovery for NFSERR_BADSESSION, but will get
 2306                          * an NFSERR_NOGRACE reply for the first "reclaim"
 2307                          * attempt.
 2308                          * So, call nfscl_expireclient() to recover the
 2309                          * opens as best we can and then do a reclaim
 2310                          * complete and return.
 2311                          */
 2312                         nfsrpc_reclaimcomplete(nmp, cred, p);
 2313                         nfscl_expireclient(clp, nmp, tcred, p);
 2314                         free(nowp, M_NFSCLOWNER);
 2315                         goto out;
 2316                     }
 2317                     /*
 2318                      * Couldn't reclaim it, so throw the state
 2319                      * away. Ouch!!
 2320                      */
 2321                     nfscl_cleandeleg(dp);
 2322                     nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
 2323                 } else {
 2324                     recovered_one = true;
 2325                     LIST_INSERT_HEAD(&extra_open, nop, nfso_list);
 2326                 }
 2327             }
 2328             dp = ndp;
 2329         }
 2330 
 2331         /*
 2332          * Now, get rid of extra Opens and Delegations.
 2333          */
 2334         LIST_FOREACH_SAFE(op, &extra_open, nfso_list, nop) {
 2335                 do {
 2336                         newnfs_copycred(&op->nfso_cred, tcred);
 2337                         error = nfscl_tryclose(op, tcred, nmp, p, true);
 2338                         if (error == NFSERR_GRACE)
 2339                                 (void) nfs_catnap(PZERO, error, "nfsexcls");
 2340                 } while (error == NFSERR_GRACE);
 2341                 LIST_REMOVE(op, nfso_list);
 2342                 free(op, M_NFSCLOPEN);
 2343         }
 2344         if (nowp != NULL)
 2345                 free(nowp, M_NFSCLOWNER);
 2346 
 2347         TAILQ_FOREACH_SAFE(dp, &extra_deleg, nfsdl_list, ndp) {
 2348                 do {
 2349                         newnfs_copycred(&dp->nfsdl_cred, tcred);
 2350                         error = nfscl_trydelegreturn(dp, tcred, nmp, p);
 2351                         if (error == NFSERR_GRACE)
 2352                                 (void) nfs_catnap(PZERO, error, "nfsexdlg");
 2353                 } while (error == NFSERR_GRACE);
 2354                 TAILQ_REMOVE(&extra_deleg, dp, nfsdl_list);
 2355                 free(dp, M_NFSCLDELEG);
 2356         }
 2357 
 2358         /* For NFSv4.1 or later, do a RECLAIM_COMPLETE. */
 2359         if (NFSHASNFSV4N(nmp))
 2360                 (void)nfsrpc_reclaimcomplete(nmp, cred, p);
 2361 
 2362 out:
 2363         NFSLOCKCLSTATE();
 2364         clp->nfsc_flags &= ~NFSCLFLAGS_RECVRINPROG;
 2365         wakeup(&clp->nfsc_flags);
 2366         nfsv4_unlock(&clp->nfsc_lock, 0);
 2367         NFSUNLOCKCLSTATE();
 2368         NFSFREECRED(tcred);
 2369 }
 2370 
 2371 /*
 2372  * This function is called when a server replies with NFSERR_EXPIRED.
 2373  * It deletes all state for the client and does a fresh SetClientId/confirm.
 2374  * XXX Someday it should post a signal to the process(es) that hold the
 2375  * state, so they know that lock state has been lost.
 2376  */
 2377 int
 2378 nfscl_hasexpired(struct nfsclclient *clp, u_int32_t clidrev, NFSPROC_T *p)
 2379 {
 2380         struct nfsmount *nmp;
 2381         struct ucred *cred;
 2382         int igotlock = 0, error, trycnt;
 2383 
 2384         /*
 2385          * If the clientid has gone away or a new SetClientid has already
 2386          * been done, just return ok.
 2387          */
 2388         if (clp == NULL || clidrev != clp->nfsc_clientidrev)
 2389                 return (0);
 2390 
 2391         /*
 2392          * First, lock the client structure, so everyone else will
 2393          * block when trying to use state. Also, use NFSCLFLAGS_EXPIREIT so
 2394          * that only one thread does the work.
 2395          */
 2396         NFSLOCKCLSTATE();
 2397         clp->nfsc_flags |= NFSCLFLAGS_EXPIREIT;
 2398         do {
 2399                 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
 2400                     NFSCLSTATEMUTEXPTR, NULL);
 2401         } while (!igotlock && (clp->nfsc_flags & NFSCLFLAGS_EXPIREIT));
 2402         if ((clp->nfsc_flags & NFSCLFLAGS_EXPIREIT) == 0) {
 2403                 if (igotlock)
 2404                         nfsv4_unlock(&clp->nfsc_lock, 0);
 2405                 NFSUNLOCKCLSTATE();
 2406                 return (0);
 2407         }
 2408         clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG;
 2409         NFSUNLOCKCLSTATE();
 2410 
 2411         nmp = clp->nfsc_nmp;
 2412         if (nmp == NULL)
 2413                 panic("nfscl expired");
 2414         cred = newnfs_getcred();
 2415         trycnt = 5;
 2416         do {
 2417                 error = nfsrpc_setclient(nmp, clp, 0, cred, p);
 2418         } while ((error == NFSERR_STALECLIENTID ||
 2419              error == NFSERR_BADSESSION ||
 2420              error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
 2421         if (error) {
 2422                 NFSLOCKCLSTATE();
 2423                 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
 2424         } else {
 2425                 /*
 2426                  * Expire the state for the client.
 2427                  */
 2428                 nfscl_expireclient(clp, nmp, cred, p);
 2429                 NFSLOCKCLSTATE();
 2430                 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
 2431                 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
 2432         }
 2433         clp->nfsc_flags &= ~(NFSCLFLAGS_EXPIREIT | NFSCLFLAGS_RECVRINPROG);
 2434         wakeup(&clp->nfsc_flags);
 2435         nfsv4_unlock(&clp->nfsc_lock, 0);
 2436         NFSUNLOCKCLSTATE();
 2437         NFSFREECRED(cred);
 2438         return (error);
 2439 }
 2440 
 2441 /*
 2442  * This function inserts a lock in the list after insert_lop.
 2443  */
 2444 static void
 2445 nfscl_insertlock(struct nfscllockowner *lp, struct nfscllock *new_lop,
 2446     struct nfscllock *insert_lop, int local)
 2447 {
 2448 
 2449         if ((struct nfscllockowner *)insert_lop == lp)
 2450                 LIST_INSERT_HEAD(&lp->nfsl_lock, new_lop, nfslo_list);
 2451         else
 2452                 LIST_INSERT_AFTER(insert_lop, new_lop, nfslo_list);
 2453         if (local)
 2454                 nfsstatsv1.cllocallocks++;
 2455         else
 2456                 nfsstatsv1.cllocks++;
 2457 }
 2458 
 2459 /*
 2460  * This function updates the locking for a lock owner and given file. It
 2461  * maintains a list of lock ranges ordered on increasing file offset that
 2462  * are NFSCLLOCK_READ or NFSCLLOCK_WRITE and non-overlapping (aka POSIX style).
 2463  * It always adds new_lop to the list and sometimes uses the one pointed
 2464  * at by other_lopp.
 2465  * Returns 1 if the locks were modified, 0 otherwise.
 2466  */
 2467 static int
 2468 nfscl_updatelock(struct nfscllockowner *lp, struct nfscllock **new_lopp,
 2469     struct nfscllock **other_lopp, int local)
 2470 {
 2471         struct nfscllock *new_lop = *new_lopp;
 2472         struct nfscllock *lop, *tlop, *ilop;
 2473         struct nfscllock *other_lop;
 2474         int unlock = 0, modified = 0;
 2475         u_int64_t tmp;
 2476 
 2477         /*
 2478          * Work down the list until the lock is merged.
 2479          */
 2480         if (new_lop->nfslo_type == F_UNLCK)
 2481                 unlock = 1;
 2482         ilop = (struct nfscllock *)lp;
 2483         lop = LIST_FIRST(&lp->nfsl_lock);
 2484         while (lop != NULL) {
 2485             /*
 2486              * Only check locks for this file that aren't before the start of
 2487              * new lock's range.
 2488              */
 2489             if (lop->nfslo_end >= new_lop->nfslo_first) {
 2490                 if (new_lop->nfslo_end < lop->nfslo_first) {
 2491                     /*
 2492                      * If the new lock ends before the start of the
 2493                      * current lock's range, no merge, just insert
 2494                      * the new lock.
 2495                      */
 2496                     break;
 2497                 }
 2498                 if (new_lop->nfslo_type == lop->nfslo_type ||
 2499                     (new_lop->nfslo_first <= lop->nfslo_first &&
 2500                      new_lop->nfslo_end >= lop->nfslo_end)) {
 2501                     /*
 2502                      * This lock can be absorbed by the new lock/unlock.
 2503                      * This happens when it covers the entire range
 2504                      * of the old lock or is contiguous
 2505                      * with the old lock and is of the same type or an
 2506                      * unlock.
 2507                      */
 2508                     if (new_lop->nfslo_type != lop->nfslo_type ||
 2509                         new_lop->nfslo_first != lop->nfslo_first ||
 2510                         new_lop->nfslo_end != lop->nfslo_end)
 2511                         modified = 1;
 2512                     if (lop->nfslo_first < new_lop->nfslo_first)
 2513                         new_lop->nfslo_first = lop->nfslo_first;
 2514                     if (lop->nfslo_end > new_lop->nfslo_end)
 2515                         new_lop->nfslo_end = lop->nfslo_end;
 2516                     tlop = lop;
 2517                     lop = LIST_NEXT(lop, nfslo_list);
 2518                     nfscl_freelock(tlop, local);
 2519                     continue;
 2520                 }
 2521 
 2522                 /*
 2523                  * All these cases are for contiguous locks that are not the
 2524                  * same type, so they can't be merged.
 2525                  */
 2526                 if (new_lop->nfslo_first <= lop->nfslo_first) {
 2527                     /*
 2528                      * This case is where the new lock overlaps with the
 2529                      * first part of the old lock. Move the start of the
 2530                      * old lock to just past the end of the new lock. The
 2531                      * new lock will be inserted in front of the old, since
 2532                      * ilop hasn't been updated. (We are done now.)
 2533                      */
 2534                     if (lop->nfslo_first != new_lop->nfslo_end) {
 2535                         lop->nfslo_first = new_lop->nfslo_end;
 2536                         modified = 1;
 2537                     }
 2538                     break;
 2539                 }
 2540                 if (new_lop->nfslo_end >= lop->nfslo_end) {
 2541                     /*
 2542                      * This case is where the new lock overlaps with the
 2543                      * end of the old lock's range. Move the old lock's
 2544                      * end to just before the new lock's first and insert
 2545                      * the new lock after the old lock.
 2546                      * Might not be done yet, since the new lock could
 2547                      * overlap further locks with higher ranges.
 2548                      */
 2549                     if (lop->nfslo_end != new_lop->nfslo_first) {
 2550                         lop->nfslo_end = new_lop->nfslo_first;
 2551                         modified = 1;
 2552                     }
 2553                     ilop = lop;
 2554                     lop = LIST_NEXT(lop, nfslo_list);
 2555                     continue;
 2556                 }
 2557                 /*
 2558                  * The final case is where the new lock's range is in the
 2559                  * middle of the current lock's and splits the current lock
 2560                  * up. Use *other_lopp to handle the second part of the
 2561                  * split old lock range. (We are done now.)
 2562                  * For unlock, we use new_lop as other_lop and tmp, since
 2563                  * other_lop and new_lop are the same for this case.
 2564                  * We noted the unlock case above, so we don't need
 2565                  * new_lop->nfslo_type any longer.
 2566                  */
 2567                 tmp = new_lop->nfslo_first;
 2568                 if (unlock) {
 2569                     other_lop = new_lop;
 2570                     *new_lopp = NULL;
 2571                 } else {
 2572                     other_lop = *other_lopp;
 2573                     *other_lopp = NULL;
 2574                 }
 2575                 other_lop->nfslo_first = new_lop->nfslo_end;
 2576                 other_lop->nfslo_end = lop->nfslo_end;
 2577                 other_lop->nfslo_type = lop->nfslo_type;
 2578                 lop->nfslo_end = tmp;
 2579                 nfscl_insertlock(lp, other_lop, lop, local);
 2580                 ilop = lop;
 2581                 modified = 1;
 2582                 break;
 2583             }
 2584             ilop = lop;
 2585             lop = LIST_NEXT(lop, nfslo_list);
 2586             if (lop == NULL)
 2587                 break;
 2588         }
 2589 
 2590         /*
 2591          * Insert the new lock in the list at the appropriate place.
 2592          */
 2593         if (!unlock) {
 2594                 nfscl_insertlock(lp, new_lop, ilop, local);
 2595                 *new_lopp = NULL;
 2596                 modified = 1;
 2597         }
 2598         return (modified);
 2599 }
 2600 
 2601 /*
 2602  * This function must be run as a kernel thread.
 2603  * It does Renew Ops and recovery, when required.
 2604  */
 2605 void
 2606 nfscl_renewthread(struct nfsclclient *clp, NFSPROC_T *p)
 2607 {
 2608         struct nfsclowner *owp, *nowp;
 2609         struct nfsclopen *op;
 2610         struct nfscllockowner *lp, *nlp;
 2611         struct nfscldeleghead dh;
 2612         struct nfscldeleg *dp, *ndp;
 2613         struct ucred *cred;
 2614         u_int32_t clidrev;
 2615         int error, cbpathdown, islept, igotlock, ret, clearok;
 2616         uint32_t recover_done_time = 0;
 2617         time_t mytime;
 2618         static time_t prevsec = 0;
 2619         struct nfscllockownerfh *lfhp, *nlfhp;
 2620         struct nfscllockownerfhhead lfh;
 2621         struct nfscllayout *lyp, *nlyp;
 2622         struct nfscldevinfo *dip, *ndip;
 2623         struct nfscllayouthead rlh;
 2624         struct nfsclrecalllayout *recallp;
 2625         struct nfsclds *dsp;
 2626         struct mount *mp;
 2627         vnode_t vp;
 2628 
 2629         cred = newnfs_getcred();
 2630         NFSLOCKCLSTATE();
 2631         clp->nfsc_flags |= NFSCLFLAGS_HASTHREAD;
 2632         mp = clp->nfsc_nmp->nm_mountp;
 2633         NFSUNLOCKCLSTATE();
 2634         for(;;) {
 2635                 newnfs_setroot(cred);
 2636                 cbpathdown = 0;
 2637                 if (clp->nfsc_flags & NFSCLFLAGS_RECOVER) {
 2638                         /*
 2639                          * Only allow one recover within 1/2 of the lease
 2640                          * duration (nfsc_renew).
 2641                          */
 2642                         if (recover_done_time < NFSD_MONOSEC) {
 2643                                 recover_done_time = NFSD_MONOSEC +
 2644                                     clp->nfsc_renew;
 2645                                 NFSCL_DEBUG(1, "Doing recovery..\n");
 2646                                 nfscl_recover(clp, cred, p);
 2647                         } else {
 2648                                 NFSCL_DEBUG(1, "Clear Recovery dt=%u ms=%jd\n",
 2649                                     recover_done_time, (intmax_t)NFSD_MONOSEC);
 2650                                 NFSLOCKCLSTATE();
 2651                                 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
 2652                                 NFSUNLOCKCLSTATE();
 2653                         }
 2654                 }
 2655                 if (clp->nfsc_expire <= NFSD_MONOSEC &&
 2656                     (clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID)) {
 2657                         clp->nfsc_expire = NFSD_MONOSEC + clp->nfsc_renew;
 2658                         clidrev = clp->nfsc_clientidrev;
 2659                         error = nfsrpc_renew(clp, NULL, cred, p);
 2660                         if (error == NFSERR_CBPATHDOWN)
 2661                             cbpathdown = 1;
 2662                         else if (error == NFSERR_STALECLIENTID) {
 2663                             NFSLOCKCLSTATE();
 2664                             clp->nfsc_flags |= NFSCLFLAGS_RECOVER;
 2665                             NFSUNLOCKCLSTATE();
 2666                         } else if (error == NFSERR_EXPIRED)
 2667                             (void) nfscl_hasexpired(clp, clidrev, p);
 2668                 }
 2669 
 2670 checkdsrenew:
 2671                 if (NFSHASNFSV4N(clp->nfsc_nmp)) {
 2672                         /* Do renews for any DS sessions. */
 2673                         NFSLOCKMNT(clp->nfsc_nmp);
 2674                         /* Skip first entry, since the MDS is handled above. */
 2675                         dsp = TAILQ_FIRST(&clp->nfsc_nmp->nm_sess);
 2676                         if (dsp != NULL)
 2677                                 dsp = TAILQ_NEXT(dsp, nfsclds_list);
 2678                         while (dsp != NULL) {
 2679                                 if (dsp->nfsclds_expire <= NFSD_MONOSEC &&
 2680                                     dsp->nfsclds_sess.nfsess_defunct == 0) {
 2681                                         dsp->nfsclds_expire = NFSD_MONOSEC +
 2682                                             clp->nfsc_renew;
 2683                                         NFSUNLOCKMNT(clp->nfsc_nmp);
 2684                                         (void)nfsrpc_renew(clp, dsp, cred, p);
 2685                                         goto checkdsrenew;
 2686                                 }
 2687                                 dsp = TAILQ_NEXT(dsp, nfsclds_list);
 2688                         }
 2689                         NFSUNLOCKMNT(clp->nfsc_nmp);
 2690                 }
 2691 
 2692                 TAILQ_INIT(&dh);
 2693                 NFSLOCKCLSTATE();
 2694                 if (cbpathdown)
 2695                         /* It's a Total Recall! */
 2696                         nfscl_totalrecall(clp);
 2697 
 2698                 /*
 2699                  * Now, handle defunct owners.
 2700                  */
 2701                 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
 2702                         if (LIST_EMPTY(&owp->nfsow_open)) {
 2703                                 if (owp->nfsow_defunct != 0)
 2704                                         nfscl_freeopenowner(owp, 0);
 2705                         }
 2706                 }
 2707 
 2708                 /*
 2709                  * Do the recall on any delegations. To avoid trouble, always
 2710                  * come back up here after having slept.
 2711                  */
 2712                 igotlock = 0;
 2713 tryagain:
 2714                 dp = TAILQ_FIRST(&clp->nfsc_deleg);
 2715                 while (dp != NULL) {
 2716                         ndp = TAILQ_NEXT(dp, nfsdl_list);
 2717                         if ((dp->nfsdl_flags & NFSCLDL_RECALL)) {
 2718                                 /*
 2719                                  * Wait for outstanding I/O ops to be done.
 2720                                  */
 2721                                 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
 2722                                     if (igotlock) {
 2723                                         nfsv4_unlock(&clp->nfsc_lock, 0);
 2724                                         igotlock = 0;
 2725                                     }
 2726                                     dp->nfsdl_rwlock.nfslock_lock |=
 2727                                         NFSV4LOCK_WANTED;
 2728                                     msleep(&dp->nfsdl_rwlock,
 2729                                         NFSCLSTATEMUTEXPTR, PVFS, "nfscld",
 2730                                         5 * hz);
 2731                                     if (NFSCL_FORCEDISM(mp))
 2732                                         goto terminate;
 2733                                     goto tryagain;
 2734                                 }
 2735                                 while (!igotlock) {
 2736                                     igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
 2737                                         &islept, NFSCLSTATEMUTEXPTR, mp);
 2738                                     if (igotlock == 0 && NFSCL_FORCEDISM(mp))
 2739                                         goto terminate;
 2740                                     if (islept)
 2741                                         goto tryagain;
 2742                                 }
 2743                                 NFSUNLOCKCLSTATE();
 2744                                 newnfs_copycred(&dp->nfsdl_cred, cred);
 2745                                 ret = nfscl_recalldeleg(clp, clp->nfsc_nmp, dp,
 2746                                     NULL, cred, p, 1, &vp);
 2747                                 if (!ret) {
 2748                                     nfscl_cleandeleg(dp);
 2749                                     TAILQ_REMOVE(&clp->nfsc_deleg, dp,
 2750                                         nfsdl_list);
 2751                                     LIST_REMOVE(dp, nfsdl_hash);
 2752                                     TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
 2753                                     nfscl_delegcnt--;
 2754                                     nfsstatsv1.cldelegates--;
 2755                                 }
 2756                                 NFSLOCKCLSTATE();
 2757                                 /*
 2758                                  * The nfsc_lock must be released before doing
 2759                                  * vrele(), since it might call nfs_inactive().
 2760                                  * For the unlikely case where the vnode failed
 2761                                  * to be acquired by nfscl_recalldeleg(), a
 2762                                  * VOP_RECLAIM() should be in progress and it
 2763                                  * will return the delegation.
 2764                                  */
 2765                                 nfsv4_unlock(&clp->nfsc_lock, 0);
 2766                                 igotlock = 0;
 2767                                 if (vp != NULL) {
 2768                                         NFSUNLOCKCLSTATE();
 2769                                         vrele(vp);
 2770                                         NFSLOCKCLSTATE();
 2771                                 }
 2772                                 goto tryagain;
 2773                         }
 2774                         dp = ndp;
 2775                 }
 2776 
 2777                 /*
 2778                  * Clear out old delegations, if we are above the high water
 2779                  * mark. Only clear out ones with no state related to them.
 2780                  * The tailq list is in LRU order.
 2781                  */
 2782                 dp = TAILQ_LAST(&clp->nfsc_deleg, nfscldeleghead);
 2783                 while (nfscl_delegcnt > nfscl_deleghighwater && dp != NULL) {
 2784                     ndp = TAILQ_PREV(dp, nfscldeleghead, nfsdl_list);
 2785                     if (dp->nfsdl_rwlock.nfslock_usecnt == 0 &&
 2786                         dp->nfsdl_rwlock.nfslock_lock == 0 &&
 2787                         dp->nfsdl_timestamp < NFSD_MONOSEC &&
 2788                         (dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_ZAPPED |
 2789                           NFSCLDL_NEEDRECLAIM | NFSCLDL_DELEGRET)) == 0) {
 2790                         clearok = 1;
 2791                         LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
 2792                             op = LIST_FIRST(&owp->nfsow_open);
 2793                             if (op != NULL) {
 2794                                 clearok = 0;
 2795                                 break;
 2796                             }
 2797                         }
 2798                         if (clearok) {
 2799                             LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
 2800                                 if (!LIST_EMPTY(&lp->nfsl_lock)) {
 2801                                     clearok = 0;
 2802                                     break;
 2803                                 }
 2804                             }
 2805                         }
 2806                         if (clearok) {
 2807                             TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
 2808                             LIST_REMOVE(dp, nfsdl_hash);
 2809                             TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
 2810                             nfscl_delegcnt--;
 2811                             nfsstatsv1.cldelegates--;
 2812                         }
 2813                     }
 2814                     dp = ndp;
 2815                 }
 2816                 if (igotlock)
 2817                         nfsv4_unlock(&clp->nfsc_lock, 0);
 2818 
 2819                 /*
 2820                  * Do the recall on any layouts. To avoid trouble, always
 2821                  * come back up here after having slept.
 2822                  */
 2823                 TAILQ_INIT(&rlh);
 2824 tryagain2:
 2825                 TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp) {
 2826                         if ((lyp->nfsly_flags & NFSLY_RECALL) != 0) {
 2827                                 /*
 2828                                  * Wait for outstanding I/O ops to be done.
 2829                                  */
 2830                                 if (lyp->nfsly_lock.nfslock_usecnt > 0 ||
 2831                                     (lyp->nfsly_lock.nfslock_lock &
 2832                                      NFSV4LOCK_LOCK) != 0) {
 2833                                         lyp->nfsly_lock.nfslock_lock |=
 2834                                             NFSV4LOCK_WANTED;
 2835                                         msleep(&lyp->nfsly_lock.nfslock_lock,
 2836                                             NFSCLSTATEMUTEXPTR, PVFS, "nfslyp",
 2837                                             5 * hz);
 2838                                         if (NFSCL_FORCEDISM(mp))
 2839                                             goto terminate;
 2840                                         goto tryagain2;
 2841                                 }
 2842                                 /* Move the layout to the recall list. */
 2843                                 TAILQ_REMOVE(&clp->nfsc_layout, lyp,
 2844                                     nfsly_list);
 2845                                 LIST_REMOVE(lyp, nfsly_hash);
 2846                                 TAILQ_INSERT_HEAD(&rlh, lyp, nfsly_list);
 2847 
 2848                                 /* Handle any layout commits. */
 2849                                 if (!NFSHASNOLAYOUTCOMMIT(clp->nfsc_nmp) &&
 2850                                     (lyp->nfsly_flags & NFSLY_WRITTEN) != 0) {
 2851                                         lyp->nfsly_flags &= ~NFSLY_WRITTEN;
 2852                                         NFSUNLOCKCLSTATE();
 2853                                         NFSCL_DEBUG(3, "do layoutcommit\n");
 2854                                         nfscl_dolayoutcommit(clp->nfsc_nmp, lyp,
 2855                                             cred, p);
 2856                                         NFSLOCKCLSTATE();
 2857                                         goto tryagain2;
 2858                                 }
 2859                         }
 2860                 }
 2861 
 2862                 /* Now, look for stale layouts. */
 2863                 lyp = TAILQ_LAST(&clp->nfsc_layout, nfscllayouthead);
 2864                 while (lyp != NULL) {
 2865                         nlyp = TAILQ_PREV(lyp, nfscllayouthead, nfsly_list);
 2866                         if (lyp->nfsly_timestamp < NFSD_MONOSEC &&
 2867                             (lyp->nfsly_flags & (NFSLY_RECALL |
 2868                              NFSLY_RETONCLOSE)) == 0 &&
 2869                             lyp->nfsly_lock.nfslock_usecnt == 0 &&
 2870                             lyp->nfsly_lock.nfslock_lock == 0) {
 2871                                 NFSCL_DEBUG(4, "ret stale lay=%d\n",
 2872                                     nfscl_layoutcnt);
 2873                                 recallp = malloc(sizeof(*recallp),
 2874                                     M_NFSLAYRECALL, M_NOWAIT);
 2875                                 if (recallp == NULL)
 2876                                         break;
 2877                                 (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE,
 2878                                     lyp, NFSLAYOUTIOMODE_ANY, 0, UINT64_MAX,
 2879                                     lyp->nfsly_stateid.seqid, 0, 0, NULL,
 2880                                     recallp);
 2881                         }
 2882                         lyp = nlyp;
 2883                 }
 2884 
 2885                 /*
 2886                  * Free up any unreferenced device info structures.
 2887                  */
 2888                 LIST_FOREACH_SAFE(dip, &clp->nfsc_devinfo, nfsdi_list, ndip) {
 2889                         if (dip->nfsdi_layoutrefs == 0 &&
 2890                             dip->nfsdi_refcnt == 0) {
 2891                                 NFSCL_DEBUG(4, "freeing devinfo\n");
 2892                                 LIST_REMOVE(dip, nfsdi_list);
 2893                                 nfscl_freedevinfo(dip);
 2894                         }
 2895                 }
 2896                 NFSUNLOCKCLSTATE();
 2897 
 2898                 /* Do layout return(s), as required. */
 2899                 TAILQ_FOREACH_SAFE(lyp, &rlh, nfsly_list, nlyp) {
 2900                         TAILQ_REMOVE(&rlh, lyp, nfsly_list);
 2901                         NFSCL_DEBUG(4, "ret layout\n");
 2902                         nfscl_layoutreturn(clp->nfsc_nmp, lyp, cred, p);
 2903                         if ((lyp->nfsly_flags & NFSLY_RETONCLOSE) != 0) {
 2904                                 NFSLOCKCLSTATE();
 2905                                 lyp->nfsly_flags |= NFSLY_RETURNED;
 2906                                 wakeup(lyp);
 2907                                 NFSUNLOCKCLSTATE();
 2908                         } else
 2909                                 nfscl_freelayout(lyp);
 2910                 }
 2911 
 2912                 /*
 2913                  * Delegreturn any delegations cleaned out or recalled.
 2914                  */
 2915                 TAILQ_FOREACH_SAFE(dp, &dh, nfsdl_list, ndp) {
 2916                         newnfs_copycred(&dp->nfsdl_cred, cred);
 2917                         (void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
 2918                         TAILQ_REMOVE(&dh, dp, nfsdl_list);
 2919                         free(dp, M_NFSCLDELEG);
 2920                 }
 2921 
 2922                 SLIST_INIT(&lfh);
 2923                 /*
 2924                  * Call nfscl_cleanupkext() once per second to check for
 2925                  * open/lock owners where the process has exited.
 2926                  */
 2927                 mytime = NFSD_MONOSEC;
 2928                 if (prevsec != mytime) {
 2929                         prevsec = mytime;
 2930                         nfscl_cleanupkext(clp, &lfh);
 2931                 }
 2932 
 2933                 /*
 2934                  * Do a ReleaseLockOwner for all lock owners where the
 2935                  * associated process no longer exists, as found by
 2936                  * nfscl_cleanupkext().
 2937                  */
 2938                 newnfs_setroot(cred);
 2939                 SLIST_FOREACH_SAFE(lfhp, &lfh, nfslfh_list, nlfhp) {
 2940                         LIST_FOREACH_SAFE(lp, &lfhp->nfslfh_lock, nfsl_list,
 2941                             nlp) {
 2942                                 (void)nfsrpc_rellockown(clp->nfsc_nmp, lp,
 2943                                     lfhp->nfslfh_fh, lfhp->nfslfh_len, cred,
 2944                                     p);
 2945                                 nfscl_freelockowner(lp, 0);
 2946                         }
 2947                         free(lfhp, M_TEMP);
 2948                 }
 2949                 SLIST_INIT(&lfh);
 2950 
 2951                 NFSLOCKCLSTATE();
 2952                 if ((clp->nfsc_flags & NFSCLFLAGS_RECOVER) == 0)
 2953                         (void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT, "nfscl",
 2954                             hz);
 2955 terminate:
 2956                 if (clp->nfsc_flags & NFSCLFLAGS_UMOUNT) {
 2957                         clp->nfsc_flags &= ~NFSCLFLAGS_HASTHREAD;
 2958                         NFSUNLOCKCLSTATE();
 2959                         NFSFREECRED(cred);
 2960                         wakeup((caddr_t)clp);
 2961                         return;
 2962                 }
 2963                 NFSUNLOCKCLSTATE();
 2964         }
 2965 }
 2966 
 2967 /*
 2968  * Initiate state recovery. Called when NFSERR_STALECLIENTID,
 2969  * NFSERR_STALESTATEID or NFSERR_BADSESSION is received.
 2970  */
 2971 void
 2972 nfscl_initiate_recovery(struct nfsclclient *clp)
 2973 {
 2974 
 2975         if (clp == NULL)
 2976                 return;
 2977         NFSLOCKCLSTATE();
 2978         clp->nfsc_flags |= NFSCLFLAGS_RECOVER;
 2979         NFSUNLOCKCLSTATE();
 2980         wakeup((caddr_t)clp);
 2981 }
 2982 
 2983 /*
 2984  * Dump out the state stuff for debugging.
 2985  */
 2986 void
 2987 nfscl_dumpstate(struct nfsmount *nmp, int openowner, int opens,
 2988     int lockowner, int locks)
 2989 {
 2990         struct nfsclclient *clp;
 2991         struct nfsclowner *owp;
 2992         struct nfsclopen *op;
 2993         struct nfscllockowner *lp;
 2994         struct nfscllock *lop;
 2995         struct nfscldeleg *dp;
 2996 
 2997         clp = nmp->nm_clp;
 2998         if (clp == NULL) {
 2999                 printf("nfscl dumpstate NULL clp\n");
 3000                 return;
 3001         }
 3002         NFSLOCKCLSTATE();
 3003         TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
 3004           LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
 3005             if (openowner && !LIST_EMPTY(&owp->nfsow_open))
 3006                 printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
 3007                     owp->nfsow_owner[0], owp->nfsow_owner[1],
 3008                     owp->nfsow_owner[2], owp->nfsow_owner[3],
 3009                     owp->nfsow_seqid);
 3010             LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
 3011                 if (opens)
 3012                     printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
 3013                         op->nfso_stateid.other[0], op->nfso_stateid.other[1],
 3014                         op->nfso_stateid.other[2], op->nfso_opencnt,
 3015                         op->nfso_fh[12]);
 3016                 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
 3017                     if (lockowner)
 3018                         printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
 3019                             lp->nfsl_owner[0], lp->nfsl_owner[1],
 3020                             lp->nfsl_owner[2], lp->nfsl_owner[3],
 3021                             lp->nfsl_seqid,
 3022                             lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
 3023                             lp->nfsl_stateid.other[2]);
 3024                     LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
 3025                         if (locks)
 3026 #ifdef __FreeBSD__
 3027                             printf("lck typ=%d fst=%ju end=%ju\n",
 3028                                 lop->nfslo_type, (intmax_t)lop->nfslo_first,
 3029                                 (intmax_t)lop->nfslo_end);
 3030 #else
 3031                             printf("lck typ=%d fst=%qd end=%qd\n",
 3032                                 lop->nfslo_type, lop->nfslo_first,
 3033                                 lop->nfslo_end);
 3034 #endif
 3035                     }
 3036                 }
 3037             }
 3038           }
 3039         }
 3040         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
 3041             if (openowner && !LIST_EMPTY(&owp->nfsow_open))
 3042                 printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
 3043                     owp->nfsow_owner[0], owp->nfsow_owner[1],
 3044                     owp->nfsow_owner[2], owp->nfsow_owner[3],
 3045                     owp->nfsow_seqid);
 3046             LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
 3047                 if (opens)
 3048                     printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
 3049                         op->nfso_stateid.other[0], op->nfso_stateid.other[1],
 3050                         op->nfso_stateid.other[2], op->nfso_opencnt,
 3051                         op->nfso_fh[12]);
 3052                 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
 3053                     if (lockowner)
 3054                         printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
 3055                             lp->nfsl_owner[0], lp->nfsl_owner[1],
 3056                             lp->nfsl_owner[2], lp->nfsl_owner[3],
 3057                             lp->nfsl_seqid,
 3058                             lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
 3059                             lp->nfsl_stateid.other[2]);
 3060                     LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
 3061                         if (locks)
 3062 #ifdef __FreeBSD__
 3063                             printf("lck typ=%d fst=%ju end=%ju\n",
 3064                                 lop->nfslo_type, (intmax_t)lop->nfslo_first,
 3065                                 (intmax_t)lop->nfslo_end);
 3066 #else
 3067                             printf("lck typ=%d fst=%qd end=%qd\n",
 3068                                 lop->nfslo_type, lop->nfslo_first,
 3069                                 lop->nfslo_end);
 3070 #endif
 3071                     }
 3072                 }
 3073             }
 3074         }
 3075         NFSUNLOCKCLSTATE();
 3076 }
 3077 
 3078 /*
 3079  * Check for duplicate open owners and opens.
 3080  * (Only used as a diagnostic aid.)
 3081  */
 3082 void
 3083 nfscl_dupopen(vnode_t vp, int dupopens)
 3084 {
 3085         struct nfsclclient *clp;
 3086         struct nfsclowner *owp, *owp2;
 3087         struct nfsclopen *op, *op2;
 3088         struct nfsfh *nfhp;
 3089 
 3090         clp = VFSTONFS(vnode_mount(vp))->nm_clp;
 3091         if (clp == NULL) {
 3092                 printf("nfscl dupopen NULL clp\n");
 3093                 return;
 3094         }
 3095         nfhp = VTONFS(vp)->n_fhp;
 3096         NFSLOCKCLSTATE();
 3097 
 3098         /*
 3099          * First, search for duplicate owners.
 3100          * These should never happen!
 3101          */
 3102         LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
 3103             LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
 3104                 if (owp != owp2 &&
 3105                     !NFSBCMP(owp->nfsow_owner, owp2->nfsow_owner,
 3106                     NFSV4CL_LOCKNAMELEN)) {
 3107                         NFSUNLOCKCLSTATE();
 3108                         printf("DUP OWNER\n");
 3109                         nfscl_dumpstate(VFSTONFS(vnode_mount(vp)), 1, 1, 0, 0);
 3110                         return;
 3111                 }
 3112             }
 3113         }
 3114 
 3115         /*
 3116          * Now, search for duplicate stateids.
 3117          * These shouldn't happen, either.
 3118          */
 3119         LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
 3120             LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
 3121                 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
 3122                     LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
 3123                         if (op != op2 &&
 3124                             (op->nfso_stateid.other[0] != 0 ||
 3125                              op->nfso_stateid.other[1] != 0 ||
 3126                              op->nfso_stateid.other[2] != 0) &&
 3127                             op->nfso_stateid.other[0] == op2->nfso_stateid.other[0] &&
 3128                             op->nfso_stateid.other[1] == op2->nfso_stateid.other[1] &&
 3129                             op->nfso_stateid.other[2] == op2->nfso_stateid.other[2]) {
 3130                             NFSUNLOCKCLSTATE();
 3131                             printf("DUP STATEID\n");
 3132                             nfscl_dumpstate(VFSTONFS(vnode_mount(vp)), 1, 1, 0,
 3133                                 0);
 3134                             return;
 3135                         }
 3136                     }
 3137                 }
 3138             }
 3139         }
 3140 
 3141         /*
 3142          * Now search for duplicate opens.
 3143          * Duplicate opens for the same owner
 3144          * should never occur. Other duplicates are
 3145          * possible and are checked for if "dupopens"
 3146          * is true.
 3147          */
 3148         LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
 3149             LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
 3150                 if (nfhp->nfh_len == op2->nfso_fhlen &&
 3151                     !NFSBCMP(nfhp->nfh_fh, op2->nfso_fh, nfhp->nfh_len)) {
 3152                     LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
 3153                         LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
 3154                             if (op != op2 && nfhp->nfh_len == op->nfso_fhlen &&
 3155                                 !NFSBCMP(nfhp->nfh_fh, op->nfso_fh, nfhp->nfh_len) &&
 3156                                 (!NFSBCMP(op->nfso_own->nfsow_owner,
 3157                                  op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN) ||
 3158                                  dupopens)) {
 3159                                 if (!NFSBCMP(op->nfso_own->nfsow_owner,
 3160                                     op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
 3161                                     NFSUNLOCKCLSTATE();
 3162                                     printf("BADDUP OPEN\n");
 3163                                 } else {
 3164                                     NFSUNLOCKCLSTATE();
 3165                                     printf("DUP OPEN\n");
 3166                                 }
 3167                                 nfscl_dumpstate(VFSTONFS(vnode_mount(vp)), 1, 1,
 3168                                     0, 0);
 3169                                 return;
 3170                             }
 3171                         }
 3172                     }
 3173                 }
 3174             }
 3175         }
 3176         NFSUNLOCKCLSTATE();
 3177 }
 3178 
 3179 /*
 3180  * During close, find an open that needs to be dereferenced and
 3181  * dereference it. If there are no more opens for this file,
 3182  * log a message to that effect.
 3183  * Opens aren't actually Close'd until VOP_INACTIVE() is performed
 3184  * on the file's vnode.
 3185  * This is the safe way, since it is difficult to identify
 3186  * which open the close is for and I/O can be performed after the
 3187  * close(2) system call when a file is mmap'd.
 3188  * If it returns 0 for success, there will be a referenced
 3189  * clp returned via clpp.
 3190  */
 3191 int
 3192 nfscl_getclose(vnode_t vp, struct nfsclclient **clpp)
 3193 {
 3194         struct nfsclclient *clp;
 3195         struct nfsclowner *owp;
 3196         struct nfsclopen *op;
 3197         struct nfscldeleg *dp;
 3198         struct nfsfh *nfhp;
 3199         int error, notdecr;
 3200 
 3201         error = nfscl_getcl(vnode_mount(vp), NULL, NULL, 1, true, &clp);
 3202         if (error)
 3203                 return (error);
 3204         *clpp = clp;
 3205 
 3206         nfhp = VTONFS(vp)->n_fhp;
 3207         notdecr = 1;
 3208         NFSLOCKCLSTATE();
 3209         /*
 3210          * First, look for one under a delegation that was locally issued
 3211          * and just decrement the opencnt for it. Since all my Opens against
 3212          * the server are DENY_NONE, I don't see a problem with hanging
 3213          * onto them. (It is much easier to use one of the extant Opens
 3214          * that I already have on the server when a Delegation is recalled
 3215          * than to do fresh Opens.) Someday, I might need to rethink this, but.
 3216          */
 3217         dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
 3218         if (dp != NULL) {
 3219                 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
 3220                         op = LIST_FIRST(&owp->nfsow_open);
 3221                         if (op != NULL) {
 3222                                 /*
 3223                                  * Since a delegation is for a file, there
 3224                                  * should never be more than one open for
 3225                                  * each openowner.
 3226                                  */
 3227                                 if (LIST_NEXT(op, nfso_list) != NULL)
 3228                                         panic("nfscdeleg opens");
 3229                                 if (notdecr && op->nfso_opencnt > 0) {
 3230                                         notdecr = 0;
 3231                                         op->nfso_opencnt--;
 3232                                         break;
 3233                                 }
 3234                         }
 3235                 }
 3236         }
 3237 
 3238         /* Now process the opens against the server. */
 3239         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
 3240                 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
 3241                         if (op->nfso_fhlen == nfhp->nfh_len &&
 3242                             !NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
 3243                             nfhp->nfh_len)) {
 3244                                 /* Found an open, decrement cnt if possible */
 3245                                 if (notdecr && op->nfso_opencnt > 0) {
 3246                                         notdecr = 0;
 3247                                         op->nfso_opencnt--;
 3248                                 }
 3249                                 /*
 3250                                  * There are more opens, so just return.
 3251                                  */
 3252                                 if (op->nfso_opencnt > 0) {
 3253                                         NFSUNLOCKCLSTATE();
 3254                                         return (0);
 3255                                 }
 3256                         }
 3257                 }
 3258         }
 3259         NFSUNLOCKCLSTATE();
 3260         if (notdecr)
 3261                 printf("nfscl: never fnd open\n");
 3262         return (0);
 3263 }
 3264 
 3265 int
 3266 nfscl_doclose(vnode_t vp, struct nfsclclient **clpp, NFSPROC_T *p)
 3267 {
 3268         struct nfsclclient *clp;
 3269         struct nfsmount *nmp;
 3270         struct nfsclowner *owp, *nowp;
 3271         struct nfsclopen *op, *nop;
 3272         struct nfsclopenhead delayed;
 3273         struct nfscldeleg *dp;
 3274         struct nfsfh *nfhp;
 3275         struct nfsclrecalllayout *recallp;
 3276         struct nfscllayout *lyp;
 3277         int error;
 3278 
 3279         error = nfscl_getcl(vnode_mount(vp), NULL, NULL, 1, true, &clp);
 3280         if (error)
 3281                 return (error);
 3282         *clpp = clp;
 3283 
 3284         nmp = VFSTONFS(vnode_mount(vp));
 3285         nfhp = VTONFS(vp)->n_fhp;
 3286         recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_WAITOK);
 3287         NFSLOCKCLSTATE();
 3288         /*
 3289          * First get rid of the local Open structures, which should be no
 3290          * longer in use.
 3291          */
 3292         dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
 3293         if (dp != NULL) {
 3294                 LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) {
 3295                         op = LIST_FIRST(&owp->nfsow_open);
 3296                         if (op != NULL) {
 3297                                 KASSERT((op->nfso_opencnt == 0),
 3298                                     ("nfscl: bad open cnt on deleg"));
 3299                                 nfscl_freeopen(op, 1, true);
 3300                         }
 3301                         nfscl_freeopenowner(owp, 1);
 3302                 }
 3303         }
 3304 
 3305         /* Return any layouts marked return on close. */
 3306         nfscl_retoncloselayout(vp, clp, nfhp->nfh_fh, nfhp->nfh_len, &recallp,
 3307             &lyp);
 3308 
 3309         /* Now process the opens against the server. */
 3310         LIST_INIT(&delayed);
 3311 lookformore:
 3312         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
 3313                 op = LIST_FIRST(&owp->nfsow_open);
 3314                 while (op != NULL) {
 3315                         if (op->nfso_fhlen == nfhp->nfh_len &&
 3316                             !NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
 3317                             nfhp->nfh_len)) {
 3318                                 /* Found an open, close it. */
 3319                                 KASSERT((op->nfso_opencnt == 0),
 3320                                     ("nfscl: bad open cnt on server"));
 3321                                 NFSUNLOCKCLSTATE();
 3322                                 if (NFSHASNFSV4N(nmp))
 3323                                         nfsrpc_doclose(nmp, op, p, false,
 3324                                             true);
 3325                                 else
 3326                                         nfsrpc_doclose(nmp, op, p, true,
 3327                                             true);
 3328                                 NFSLOCKCLSTATE();
 3329                                 if (error == NFSERR_DELAY) {
 3330                                         nfscl_unlinkopen(op);
 3331                                         op->nfso_own = NULL;
 3332                                         LIST_INSERT_HEAD(&delayed, op,
 3333                                             nfso_list);
 3334                                 }
 3335                                 goto lookformore;
 3336                         }
 3337                         op = LIST_NEXT(op, nfso_list);
 3338                 }
 3339         }
 3340         nfscl_clrelease(clp);
 3341 
 3342         /* Now, wait for any layout that is returned upon close. */
 3343         if (lyp != NULL) {
 3344                 while ((lyp->nfsly_flags & NFSLY_RETURNED) == 0) {
 3345                         if (NFSCL_FORCEDISM(vnode_mount(vp))) {
 3346                                 lyp = NULL;
 3347                                 break;
 3348                         }
 3349                         msleep(lyp, NFSCLSTATEMUTEXPTR, PZERO, "nfslroc", hz);
 3350                 }
 3351                 if (lyp != NULL)
 3352                         nfscl_freelayout(lyp);
 3353         }
 3354 
 3355         NFSUNLOCKCLSTATE();
 3356         /*
 3357          * recallp has been set NULL by nfscl_retoncloselayout() if it was
 3358          * used by the function, but calling free() with a NULL pointer is ok.
 3359          */
 3360         free(recallp, M_NFSLAYRECALL);
 3361 
 3362         /* Now, loop retrying the delayed closes. */
 3363         LIST_FOREACH_SAFE(op, &delayed, nfso_list, nop) {
 3364                 nfsrpc_doclose(nmp, op, p, true, false);
 3365                 LIST_REMOVE(op, nfso_list);
 3366                 nfscl_freeopen(op, 0, false);
 3367         }
 3368         return (0);
 3369 }
 3370 
 3371 /*
 3372  * Return all delegations on this client.
 3373  * (Must be called with client sleep lock.)
 3374  */
 3375 static void
 3376 nfscl_delegreturnall(struct nfsclclient *clp, NFSPROC_T *p,
 3377     struct nfscldeleghead *dhp)
 3378 {
 3379         struct nfscldeleg *dp, *ndp;
 3380         struct ucred *cred;
 3381 
 3382         cred = newnfs_getcred();
 3383         TAILQ_FOREACH_SAFE(dp, &clp->nfsc_deleg, nfsdl_list, ndp) {
 3384                 nfscl_cleandeleg(dp);
 3385                 (void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
 3386                 if (dhp != NULL) {
 3387                         nfscl_freedeleg(&clp->nfsc_deleg, dp, false);
 3388                         TAILQ_INSERT_HEAD(dhp, dp, nfsdl_list);
 3389                 } else
 3390                         nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
 3391         }
 3392         NFSFREECRED(cred);
 3393 }
 3394 
 3395 /*
 3396  * Return any delegation for this vp.
 3397  */
 3398 void
 3399 nfscl_delegreturnvp(vnode_t vp, NFSPROC_T *p)
 3400 {
 3401         struct nfsclclient *clp;
 3402         struct nfscldeleg *dp;
 3403         struct ucred *cred;
 3404         struct nfsnode *np;
 3405 
 3406         np = VTONFS(vp);
 3407         cred = newnfs_getcred();
 3408         dp = NULL;
 3409         NFSLOCKCLSTATE();
 3410         clp = VFSTONFS(vp->v_mount)->nm_clp;
 3411         if (clp != NULL)
 3412                 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
 3413                     np->n_fhp->nfh_len);
 3414         if (dp != NULL) {
 3415                 nfscl_cleandeleg(dp);
 3416                 nfscl_freedeleg(&clp->nfsc_deleg, dp, false);
 3417                 NFSUNLOCKCLSTATE();
 3418                 newnfs_copycred(&dp->nfsdl_cred, cred);
 3419                 nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
 3420                 free(dp, M_NFSCLDELEG);
 3421         } else
 3422                 NFSUNLOCKCLSTATE();
 3423         NFSFREECRED(cred);
 3424 }
 3425 
 3426 /*
 3427  * Do a callback RPC.
 3428  */
 3429 void
 3430 nfscl_docb(struct nfsrv_descript *nd, NFSPROC_T *p)
 3431 {
 3432         int clist, gotseq_ok, i, j, k, op, rcalls;
 3433         u_int32_t *tl;
 3434         struct nfsclclient *clp;
 3435         struct nfscldeleg *dp = NULL;
 3436         int numops, taglen = -1, error = 0, trunc __unused;
 3437         u_int32_t minorvers = 0, retops = 0, *retopsp = NULL, *repp, cbident;
 3438         u_char tag[NFSV4_SMALLSTR + 1], *tagstr;
 3439         vnode_t vp = NULL;
 3440         struct nfsnode *np;
 3441         struct vattr va;
 3442         struct nfsfh *nfhp;
 3443         mount_t mp;
 3444         nfsattrbit_t attrbits, rattrbits;
 3445         nfsv4stateid_t stateid;
 3446         uint32_t seqid, slotid = 0, highslot, cachethis __unused;
 3447         uint8_t sessionid[NFSX_V4SESSIONID];
 3448         struct mbuf *rep;
 3449         struct nfscllayout *lyp;
 3450         uint64_t filesid[2], len, off;
 3451         int changed, gotone, laytype, recalltype;
 3452         uint32_t iomode;
 3453         struct nfsclrecalllayout *recallp = NULL;
 3454         struct nfsclsession *tsep;
 3455 
 3456         gotseq_ok = 0;
 3457         nfsrvd_rephead(nd);
 3458         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
 3459         taglen = fxdr_unsigned(int, *tl);
 3460         if (taglen < 0 || taglen > NFSV4_OPAQUELIMIT) {
 3461                 error = EBADRPC;
 3462                 taglen = -1;
 3463                 goto nfsmout;
 3464         }
 3465         if (taglen <= NFSV4_SMALLSTR)
 3466                 tagstr = tag;
 3467         else
 3468                 tagstr = malloc(taglen + 1, M_TEMP, M_WAITOK);
 3469         error = nfsrv_mtostr(nd, tagstr, taglen);
 3470         if (error) {
 3471                 if (taglen > NFSV4_SMALLSTR)
 3472                         free(tagstr, M_TEMP);
 3473                 taglen = -1;
 3474                 goto nfsmout;
 3475         }
 3476         (void) nfsm_strtom(nd, tag, taglen);
 3477         if (taglen > NFSV4_SMALLSTR) {
 3478                 free(tagstr, M_TEMP);
 3479         }
 3480         NFSM_BUILD(retopsp, u_int32_t *, NFSX_UNSIGNED);
 3481         NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
 3482         minorvers = fxdr_unsigned(u_int32_t, *tl++);
 3483         if (minorvers != NFSV4_MINORVERSION && minorvers != NFSV41_MINORVERSION)
 3484                 nd->nd_repstat = NFSERR_MINORVERMISMATCH;
 3485         cbident = fxdr_unsigned(u_int32_t, *tl++);
 3486         if (nd->nd_repstat)
 3487                 numops = 0;
 3488         else
 3489                 numops = fxdr_unsigned(int, *tl);
 3490         /*
 3491          * Loop around doing the sub ops.
 3492          */
 3493         for (i = 0; i < numops; i++) {
 3494                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
 3495                 NFSM_BUILD(repp, u_int32_t *, 2 * NFSX_UNSIGNED);
 3496                 *repp++ = *tl;
 3497                 op = fxdr_unsigned(int, *tl);
 3498                 nd->nd_procnum = op;
 3499                 if (i == 0 && op != NFSV4OP_CBSEQUENCE && minorvers !=
 3500                     NFSV4_MINORVERSION) {
 3501                     nd->nd_repstat = NFSERR_OPNOTINSESS;
 3502                     *repp = nfscl_errmap(nd, minorvers);
 3503                     retops++;
 3504                     break;
 3505                 }
 3506                 if (op < NFSV4OP_CBGETATTR ||
 3507                    (op > NFSV4OP_CBRECALL && minorvers == NFSV4_MINORVERSION) ||
 3508                    (op > NFSV4OP_CBNOTIFYDEVID &&
 3509                     minorvers == NFSV41_MINORVERSION)) {
 3510                     nd->nd_repstat = NFSERR_OPILLEGAL;
 3511                     *repp = nfscl_errmap(nd, minorvers);
 3512                     retops++;
 3513                     break;
 3514                 }
 3515                 if (op < NFSV41_CBNOPS)
 3516                         nfsstatsv1.cbrpccnt[nd->nd_procnum]++;
 3517                 switch (op) {
 3518                 case NFSV4OP_CBGETATTR:
 3519                         NFSCL_DEBUG(4, "cbgetattr\n");
 3520                         mp = NULL;
 3521                         vp = NULL;
 3522                         error = nfsm_getfh(nd, &nfhp);
 3523                         if (!error)
 3524                                 error = nfsrv_getattrbits(nd, &attrbits,
 3525                                     NULL, NULL);
 3526                         if (!error) {
 3527                                 mp = nfscl_getmnt(minorvers, sessionid, cbident,
 3528                                     &clp);
 3529                                 if (mp == NULL)
 3530                                         error = NFSERR_SERVERFAULT;
 3531                         }
 3532                         if (!error) {
 3533                                 error = nfscl_ngetreopen(mp, nfhp->nfh_fh,
 3534                                     nfhp->nfh_len, p, &np);
 3535                                 if (!error)
 3536                                         vp = NFSTOV(np);
 3537                         }
 3538                         if (!error) {
 3539                                 NFSZERO_ATTRBIT(&rattrbits);
 3540                                 NFSLOCKCLSTATE();
 3541                                 dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
 3542                                     nfhp->nfh_len);
 3543                                 if (dp != NULL) {
 3544                                         if (NFSISSET_ATTRBIT(&attrbits,
 3545                                             NFSATTRBIT_SIZE)) {
 3546                                                 if (vp != NULL)
 3547                                                         va.va_size = np->n_size;
 3548                                                 else
 3549                                                         va.va_size =
 3550                                                             dp->nfsdl_size;
 3551                                                 NFSSETBIT_ATTRBIT(&rattrbits,
 3552                                                     NFSATTRBIT_SIZE);
 3553                                         }
 3554                                         if (NFSISSET_ATTRBIT(&attrbits,
 3555                                             NFSATTRBIT_CHANGE)) {
 3556                                                 va.va_filerev =
 3557                                                     dp->nfsdl_change;
 3558                                                 if (vp == NULL ||
 3559                                                     (np->n_flag & NDELEGMOD))
 3560                                                         va.va_filerev++;
 3561                                                 NFSSETBIT_ATTRBIT(&rattrbits,
 3562                                                     NFSATTRBIT_CHANGE);
 3563                                         }
 3564                                 } else
 3565                                         error = NFSERR_SERVERFAULT;
 3566                                 NFSUNLOCKCLSTATE();
 3567                         }
 3568                         if (vp != NULL)
 3569                                 vrele(vp);
 3570                         if (mp != NULL)
 3571                                 vfs_unbusy(mp);
 3572                         if (nfhp != NULL)
 3573                                 free(nfhp, M_NFSFH);
 3574                         if (!error)
 3575                                 (void) nfsv4_fillattr(nd, NULL, NULL, NULL, &va,
 3576                                     NULL, 0, &rattrbits, NULL, p, 0, 0, 0, 0,
 3577                                     (uint64_t)0, NULL);
 3578                         break;
 3579                 case NFSV4OP_CBRECALL:
 3580                         NFSCL_DEBUG(4, "cbrecall\n");
 3581                         NFSM_DISSECT(tl, u_int32_t *, NFSX_STATEID +
 3582                             NFSX_UNSIGNED);
 3583                         stateid.seqid = *tl++;
 3584                         NFSBCOPY((caddr_t)tl, (caddr_t)stateid.other,
 3585                             NFSX_STATEIDOTHER);
 3586                         tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
 3587                         trunc = fxdr_unsigned(int, *tl);
 3588                         error = nfsm_getfh(nd, &nfhp);
 3589                         if (!error) {
 3590                                 NFSLOCKCLSTATE();
 3591                                 if (minorvers == NFSV4_MINORVERSION)
 3592                                         clp = nfscl_getclnt(cbident);
 3593                                 else
 3594                                         clp = nfscl_getclntsess(sessionid);
 3595                                 if (clp != NULL) {
 3596                                         dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
 3597                                             nfhp->nfh_len);
 3598                                         if (dp != NULL && (dp->nfsdl_flags &
 3599                                             NFSCLDL_DELEGRET) == 0) {
 3600                                                 dp->nfsdl_flags |=
 3601                                                     NFSCLDL_RECALL;
 3602                                                 wakeup((caddr_t)clp);
 3603                                         }
 3604                                 } else {
 3605                                         error = NFSERR_SERVERFAULT;
 3606                                 }
 3607                                 NFSUNLOCKCLSTATE();
 3608                         }
 3609                         if (nfhp != NULL)
 3610                                 free(nfhp, M_NFSFH);
 3611                         break;
 3612                 case NFSV4OP_CBLAYOUTRECALL:
 3613                         NFSCL_DEBUG(4, "cblayrec\n");
 3614                         nfhp = NULL;
 3615                         NFSM_DISSECT(tl, uint32_t *, 4 * NFSX_UNSIGNED);
 3616                         laytype = fxdr_unsigned(int, *tl++);
 3617                         iomode = fxdr_unsigned(uint32_t, *tl++);
 3618                         if (newnfs_true == *tl++)
 3619                                 changed = 1;
 3620                         else
 3621                                 changed = 0;
 3622                         recalltype = fxdr_unsigned(int, *tl);
 3623                         NFSCL_DEBUG(4, "layt=%d iom=%d ch=%d rectyp=%d\n",
 3624                             laytype, iomode, changed, recalltype);
 3625                         recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL,
 3626                             M_WAITOK);
 3627                         if (laytype != NFSLAYOUT_NFSV4_1_FILES &&
 3628                             laytype != NFSLAYOUT_FLEXFILE)
 3629                                 error = NFSERR_NOMATCHLAYOUT;
 3630                         else if (recalltype == NFSLAYOUTRETURN_FILE) {
 3631                                 error = nfsm_getfh(nd, &nfhp);
 3632                                 NFSCL_DEBUG(4, "retfile getfh=%d\n", error);
 3633                                 if (error != 0)
 3634                                         goto nfsmout;
 3635                                 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_HYPER +
 3636                                     NFSX_STATEID);
 3637                                 off = fxdr_hyper(tl); tl += 2;
 3638                                 len = fxdr_hyper(tl); tl += 2;
 3639                                 stateid.seqid = fxdr_unsigned(uint32_t, *tl++);
 3640                                 NFSBCOPY(tl, stateid.other, NFSX_STATEIDOTHER);
 3641                                 if (minorvers == NFSV4_MINORVERSION)
 3642                                         error = NFSERR_NOTSUPP;
 3643                                 NFSCL_DEBUG(4, "off=%ju len=%ju sq=%u err=%d\n",
 3644                                     (uintmax_t)off, (uintmax_t)len,
 3645                                     stateid.seqid, error);
 3646                                 if (error == 0) {
 3647                                         NFSLOCKCLSTATE();
 3648                                         clp = nfscl_getclntsess(sessionid);
 3649                                         NFSCL_DEBUG(4, "cbly clp=%p\n", clp);
 3650                                         if (clp != NULL) {
 3651                                                 lyp = nfscl_findlayout(clp,
 3652                                                     nfhp->nfh_fh,
 3653                                                     nfhp->nfh_len);
 3654                                                 NFSCL_DEBUG(4, "cblyp=%p\n",
 3655                                                     lyp);
 3656                                                 if (lyp != NULL &&
 3657                                                     (lyp->nfsly_flags &
 3658                                                      (NFSLY_FILES |
 3659                                                       NFSLY_FLEXFILE)) != 0 &&
 3660                                                     !NFSBCMP(stateid.other,
 3661                                                     lyp->nfsly_stateid.other,
 3662                                                     NFSX_STATEIDOTHER)) {
 3663                                                         error =
 3664                                                             nfscl_layoutrecall(
 3665                                                             recalltype,
 3666                                                             lyp, iomode, off,
 3667                                                             len, stateid.seqid,
 3668                                                             0, 0, NULL,
 3669                                                             recallp);
 3670                                                         if (error == 0 &&
 3671                                                             stateid.seqid >
 3672                                                             lyp->nfsly_stateid.seqid)
 3673                                                                 lyp->nfsly_stateid.seqid =
 3674                                                                     stateid.seqid;
 3675                                                         recallp = NULL;
 3676                                                         wakeup(clp);
 3677                                                         NFSCL_DEBUG(4,
 3678                                                             "aft layrcal=%d "
 3679                                                             "layseqid=%d\n",
 3680                                                             error,
 3681                                                             lyp->nfsly_stateid.seqid);
 3682                                                 } else
 3683                                                         error =
 3684                                                           NFSERR_NOMATCHLAYOUT;
 3685                                         } else
 3686                                                 error = NFSERR_NOMATCHLAYOUT;
 3687                                         NFSUNLOCKCLSTATE();
 3688                                 }
 3689                                 free(nfhp, M_NFSFH);
 3690                         } else if (recalltype == NFSLAYOUTRETURN_FSID) {
 3691                                 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_HYPER);
 3692                                 filesid[0] = fxdr_hyper(tl); tl += 2;
 3693                                 filesid[1] = fxdr_hyper(tl); tl += 2;
 3694                                 gotone = 0;
 3695                                 NFSLOCKCLSTATE();
 3696                                 clp = nfscl_getclntsess(sessionid);
 3697                                 if (clp != NULL) {
 3698                                         TAILQ_FOREACH(lyp, &clp->nfsc_layout,
 3699                                             nfsly_list) {
 3700                                                 if (lyp->nfsly_filesid[0] ==
 3701                                                     filesid[0] &&
 3702                                                     lyp->nfsly_filesid[1] ==
 3703                                                     filesid[1]) {
 3704                                                         error =
 3705                                                             nfscl_layoutrecall(
 3706                                                             recalltype,
 3707                                                             lyp, iomode, 0,
 3708                                                             UINT64_MAX,
 3709                                                             lyp->nfsly_stateid.seqid,
 3710                                                             0, 0, NULL,
 3711                                                             recallp);
 3712                                                         recallp = NULL;
 3713                                                         gotone = 1;
 3714                                                 }
 3715                                         }
 3716                                         if (gotone != 0)
 3717                                                 wakeup(clp);
 3718                                         else
 3719                                                 error = NFSERR_NOMATCHLAYOUT;
 3720                                 } else
 3721                                         error = NFSERR_NOMATCHLAYOUT;
 3722                                 NFSUNLOCKCLSTATE();
 3723                         } else if (recalltype == NFSLAYOUTRETURN_ALL) {
 3724                                 gotone = 0;
 3725                                 NFSLOCKCLSTATE();
 3726                                 clp = nfscl_getclntsess(sessionid);
 3727                                 if (clp != NULL) {
 3728                                         TAILQ_FOREACH(lyp, &clp->nfsc_layout,
 3729                                             nfsly_list) {
 3730                                                 error = nfscl_layoutrecall(
 3731                                                     recalltype, lyp, iomode, 0,
 3732                                                     UINT64_MAX,
 3733                                                     lyp->nfsly_stateid.seqid,
 3734                                                     0, 0, NULL, recallp);
 3735                                                 recallp = NULL;
 3736                                                 gotone = 1;
 3737                                         }
 3738                                         if (gotone != 0)
 3739                                                 wakeup(clp);
 3740                                         else
 3741                                                 error = NFSERR_NOMATCHLAYOUT;
 3742                                 } else
 3743                                         error = NFSERR_NOMATCHLAYOUT;
 3744                                 NFSUNLOCKCLSTATE();
 3745                         } else
 3746                                 error = NFSERR_NOMATCHLAYOUT;
 3747                         if (recallp != NULL) {
 3748                                 free(recallp, M_NFSLAYRECALL);
 3749                                 recallp = NULL;
 3750                         }
 3751                         break;
 3752                 case NFSV4OP_CBSEQUENCE:
 3753                         if (i != 0) {
 3754                             error = NFSERR_SEQUENCEPOS;
 3755                             break;
 3756                         }
 3757                         NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID +
 3758                             5 * NFSX_UNSIGNED);
 3759                         bcopy(tl, sessionid, NFSX_V4SESSIONID);
 3760                         tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
 3761                         seqid = fxdr_unsigned(uint32_t, *tl++);
 3762                         slotid = fxdr_unsigned(uint32_t, *tl++);
 3763                         highslot = fxdr_unsigned(uint32_t, *tl++);
 3764                         cachethis = *tl++;
 3765                         /* Throw away the referring call stuff. */
 3766                         clist = fxdr_unsigned(int, *tl);
 3767                         for (j = 0; j < clist; j++) {
 3768                                 NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID +
 3769                                     NFSX_UNSIGNED);
 3770                                 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
 3771                                 rcalls = fxdr_unsigned(int, *tl);
 3772                                 for (k = 0; k < rcalls; k++) {
 3773                                         NFSM_DISSECT(tl, uint32_t *,
 3774                                             2 * NFSX_UNSIGNED);
 3775                                 }
 3776                         }
 3777                         NFSLOCKCLSTATE();
 3778                         clp = nfscl_getclntsess(sessionid);
 3779                         if (clp == NULL)
 3780                                 error = NFSERR_SERVERFAULT;
 3781                         if (error == 0) {
 3782                                 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
 3783                                 error = nfsv4_seqsession(seqid, slotid,
 3784                                     highslot, tsep->nfsess_cbslots, &rep,
 3785                                     tsep->nfsess_backslots);
 3786                         }
 3787                         NFSUNLOCKCLSTATE();
 3788                         if (error == 0 || error == NFSERR_REPLYFROMCACHE) {
 3789                                 gotseq_ok = 1;
 3790                                 if (rep != NULL) {
 3791                                         /*
 3792                                          * Handle a reply for a retried
 3793                                          * callback.  The reply will be
 3794                                          * re-inserted in the session cache
 3795                                          * by the nfsv4_seqsess_cacherep() call
 3796                                          * after out:
 3797                                          */
 3798                                         KASSERT(error == NFSERR_REPLYFROMCACHE,
 3799                                             ("cbsequence: non-NULL rep"));
 3800                                         NFSCL_DEBUG(4, "Got cbretry\n");
 3801                                         m_freem(nd->nd_mreq);
 3802                                         nd->nd_mreq = rep;
 3803                                         rep = NULL;
 3804                                         goto out;
 3805                                 }
 3806                                 NFSM_BUILD(tl, uint32_t *,
 3807                                     NFSX_V4SESSIONID + 4 * NFSX_UNSIGNED);
 3808                                 bcopy(sessionid, tl, NFSX_V4SESSIONID);
 3809                                 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
 3810                                 *tl++ = txdr_unsigned(seqid);
 3811                                 *tl++ = txdr_unsigned(slotid);
 3812                                 *tl++ = txdr_unsigned(NFSV4_CBSLOTS - 1);
 3813                                 *tl = txdr_unsigned(NFSV4_CBSLOTS - 1);
 3814                         }
 3815                         break;
 3816                 default:
 3817                         if (i == 0 && minorvers == NFSV41_MINORVERSION)
 3818                                 error = NFSERR_OPNOTINSESS;
 3819                         else {
 3820                                 NFSCL_DEBUG(1, "unsupp callback %d\n", op);
 3821                                 error = NFSERR_NOTSUPP;
 3822                         }
 3823                         break;
 3824                 }
 3825                 if (error) {
 3826                         if (error == EBADRPC || error == NFSERR_BADXDR) {
 3827                                 nd->nd_repstat = NFSERR_BADXDR;
 3828                         } else {
 3829                                 nd->nd_repstat = error;
 3830                         }
 3831                         error = 0;
 3832                 }
 3833                 retops++;
 3834                 if (nd->nd_repstat) {
 3835                         *repp = nfscl_errmap(nd, minorvers);
 3836                         break;
 3837                 } else
 3838                         *repp = 0;      /* NFS4_OK */
 3839         }
 3840 nfsmout:
 3841         if (recallp != NULL)
 3842                 free(recallp, M_NFSLAYRECALL);
 3843         if (error) {
 3844                 if (error == EBADRPC || error == NFSERR_BADXDR)
 3845                         nd->nd_repstat = NFSERR_BADXDR;
 3846                 else
 3847                         printf("nfsv4 comperr1=%d\n", error);
 3848         }
 3849         if (taglen == -1) {
 3850                 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
 3851                 *tl++ = 0;
 3852                 *tl = 0;
 3853         } else {
 3854                 *retopsp = txdr_unsigned(retops);
 3855         }
 3856         *nd->nd_errp = nfscl_errmap(nd, minorvers);
 3857 out:
 3858         if (gotseq_ok != 0) {
 3859                 rep = m_copym(nd->nd_mreq, 0, M_COPYALL, M_WAITOK);
 3860                 NFSLOCKCLSTATE();
 3861                 clp = nfscl_getclntsess(sessionid);
 3862                 if (clp != NULL) {
 3863                         tsep = nfsmnt_mdssession(clp->nfsc_nmp);
 3864                         nfsv4_seqsess_cacherep(slotid, tsep->nfsess_cbslots,
 3865                             NFSERR_OK, &rep);
 3866                         NFSUNLOCKCLSTATE();
 3867                 } else {
 3868                         NFSUNLOCKCLSTATE();
 3869                         m_freem(rep);
 3870                 }
 3871         }
 3872 }
 3873 
 3874 /*
 3875  * Generate the next cbident value. Basically just increment a static value
 3876  * and then check that it isn't already in the list, if it has wrapped around.
 3877  */
 3878 static u_int32_t
 3879 nfscl_nextcbident(void)
 3880 {
 3881         struct nfsclclient *clp;
 3882         int matched;
 3883         static u_int32_t nextcbident = 0;
 3884         static int haswrapped = 0;
 3885 
 3886         nextcbident++;
 3887         if (nextcbident == 0)
 3888                 haswrapped = 1;
 3889         if (haswrapped) {
 3890                 /*
 3891                  * Search the clientid list for one already using this cbident.
 3892                  */
 3893                 do {
 3894                         matched = 0;
 3895                         NFSLOCKCLSTATE();
 3896                         LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
 3897                                 if (clp->nfsc_cbident == nextcbident) {
 3898                                         matched = 1;
 3899                                         break;
 3900                                 }
 3901                         }
 3902                         NFSUNLOCKCLSTATE();
 3903                         if (matched == 1)
 3904                                 nextcbident++;
 3905                 } while (matched);
 3906         }
 3907         return (nextcbident);
 3908 }
 3909 
 3910 /*
 3911  * Get the mount point related to a given cbident or session and busy it.
 3912  */
 3913 static mount_t
 3914 nfscl_getmnt(int minorvers, uint8_t *sessionid, u_int32_t cbident,
 3915     struct nfsclclient **clpp)
 3916 {
 3917         struct nfsclclient *clp;
 3918         mount_t mp;
 3919         int error;
 3920         struct nfsclsession *tsep;
 3921 
 3922         *clpp = NULL;
 3923         NFSLOCKCLSTATE();
 3924         LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
 3925                 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
 3926                 if (minorvers == NFSV4_MINORVERSION) {
 3927                         if (clp->nfsc_cbident == cbident)
 3928                                 break;
 3929                 } else if (!NFSBCMP(tsep->nfsess_sessionid, sessionid,
 3930                     NFSX_V4SESSIONID))
 3931                         break;
 3932         }
 3933         if (clp == NULL) {
 3934                 NFSUNLOCKCLSTATE();
 3935                 return (NULL);
 3936         }
 3937         mp = clp->nfsc_nmp->nm_mountp;
 3938         vfs_ref(mp);
 3939         NFSUNLOCKCLSTATE();
 3940         error = vfs_busy(mp, 0);
 3941         vfs_rel(mp);
 3942         if (error != 0)
 3943                 return (NULL);
 3944         *clpp = clp;
 3945         return (mp);
 3946 }
 3947 
 3948 /*
 3949  * Get the clientid pointer related to a given cbident.
 3950  */
 3951 static struct nfsclclient *
 3952 nfscl_getclnt(u_int32_t cbident)
 3953 {
 3954         struct nfsclclient *clp;
 3955 
 3956         LIST_FOREACH(clp, &nfsclhead, nfsc_list)
 3957                 if (clp->nfsc_cbident == cbident)
 3958                         break;
 3959         return (clp);
 3960 }
 3961 
 3962 /*
 3963  * Get the clientid pointer related to a given sessionid.
 3964  */
 3965 static struct nfsclclient *
 3966 nfscl_getclntsess(uint8_t *sessionid)
 3967 {
 3968         struct nfsclclient *clp;
 3969         struct nfsclsession *tsep;
 3970 
 3971         LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
 3972                 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
 3973                 if (!NFSBCMP(tsep->nfsess_sessionid, sessionid,
 3974                     NFSX_V4SESSIONID))
 3975                         break;
 3976         }
 3977         return (clp);
 3978 }
 3979 
 3980 /*
 3981  * Search for a lock conflict locally on the client. A conflict occurs if
 3982  * - not same owner and overlapping byte range and at least one of them is
 3983  *   a write lock or this is an unlock.
 3984  */
 3985 static int
 3986 nfscl_localconflict(struct nfsclclient *clp, u_int8_t *fhp, int fhlen,
 3987     struct nfscllock *nlop, u_int8_t *own, struct nfscldeleg *dp,
 3988     struct nfscllock **lopp)
 3989 {
 3990         struct nfsclowner *owp;
 3991         struct nfsclopen *op;
 3992         int ret;
 3993 
 3994         if (dp != NULL) {
 3995                 ret = nfscl_checkconflict(&dp->nfsdl_lock, nlop, own, lopp);
 3996                 if (ret)
 3997                         return (ret);
 3998         }
 3999         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
 4000                 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
 4001                         if (op->nfso_fhlen == fhlen &&
 4002                             !NFSBCMP(op->nfso_fh, fhp, fhlen)) {
 4003                                 ret = nfscl_checkconflict(&op->nfso_lock, nlop,
 4004                                     own, lopp);
 4005                                 if (ret)
 4006                                         return (ret);
 4007                         }
 4008                 }
 4009         }
 4010         return (0);
 4011 }
 4012 
 4013 static int
 4014 nfscl_checkconflict(struct nfscllockownerhead *lhp, struct nfscllock *nlop,
 4015     u_int8_t *own, struct nfscllock **lopp)
 4016 {
 4017         struct nfscllockowner *lp;
 4018         struct nfscllock *lop;
 4019 
 4020         LIST_FOREACH(lp, lhp, nfsl_list) {
 4021                 if (NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
 4022                         LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
 4023                                 if (lop->nfslo_first >= nlop->nfslo_end)
 4024                                         break;
 4025                                 if (lop->nfslo_end <= nlop->nfslo_first)
 4026                                         continue;
 4027                                 if (lop->nfslo_type == F_WRLCK ||
 4028                                     nlop->nfslo_type == F_WRLCK ||
 4029                                     nlop->nfslo_type == F_UNLCK) {
 4030                                         if (lopp != NULL)
 4031                                                 *lopp = lop;
 4032                                         return (NFSERR_DENIED);
 4033                                 }
 4034                         }
 4035                 }
 4036         }
 4037         return (0);
 4038 }
 4039 
 4040 /*
 4041  * Check for a local conflicting lock.
 4042  */
 4043 int
 4044 nfscl_lockt(vnode_t vp, struct nfsclclient *clp, u_int64_t off,
 4045     u_int64_t len, struct flock *fl, NFSPROC_T *p, void *id, int flags)
 4046 {
 4047         struct nfscllock *lop, nlck;
 4048         struct nfscldeleg *dp;
 4049         struct nfsnode *np;
 4050         u_int8_t own[NFSV4CL_LOCKNAMELEN];
 4051         int error;
 4052 
 4053         nlck.nfslo_type = fl->l_type;
 4054         nlck.nfslo_first = off;
 4055         if (len == NFS64BITSSET) {
 4056                 nlck.nfslo_end = NFS64BITSSET;
 4057         } else {
 4058                 nlck.nfslo_end = off + len;
 4059                 if (nlck.nfslo_end <= nlck.nfslo_first)
 4060                         return (NFSERR_INVAL);
 4061         }
 4062         np = VTONFS(vp);
 4063         nfscl_filllockowner(id, own, flags);
 4064         NFSLOCKCLSTATE();
 4065         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
 4066         error = nfscl_localconflict(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len,
 4067             &nlck, own, dp, &lop);
 4068         if (error != 0) {
 4069                 fl->l_whence = SEEK_SET;
 4070                 fl->l_start = lop->nfslo_first;
 4071                 if (lop->nfslo_end == NFS64BITSSET)
 4072                         fl->l_len = 0;
 4073                 else
 4074                         fl->l_len = lop->nfslo_end - lop->nfslo_first;
 4075                 fl->l_pid = (pid_t)0;
 4076                 fl->l_type = lop->nfslo_type;
 4077                 error = -1;                     /* no RPC required */
 4078         } else if (dp != NULL && ((dp->nfsdl_flags & NFSCLDL_WRITE) ||
 4079             fl->l_type == F_RDLCK)) {
 4080                 /*
 4081                  * The delegation ensures that there isn't a conflicting
 4082                  * lock on the server, so return -1 to indicate an RPC
 4083                  * isn't required.
 4084                  */
 4085                 fl->l_type = F_UNLCK;
 4086                 error = -1;
 4087         }
 4088         NFSUNLOCKCLSTATE();
 4089         return (error);
 4090 }
 4091 
 4092 /*
 4093  * Handle Recall of a delegation.
 4094  * The clp must be exclusive locked when this is called.
 4095  */
 4096 static int
 4097 nfscl_recalldeleg(struct nfsclclient *clp, struct nfsmount *nmp,
 4098     struct nfscldeleg *dp, vnode_t vp, struct ucred *cred, NFSPROC_T *p,
 4099     int called_from_renewthread, vnode_t *vpp)
 4100 {
 4101         struct nfsclowner *owp, *lowp, *nowp;
 4102         struct nfsclopen *op, *lop;
 4103         struct nfscllockowner *lp;
 4104         struct nfscllock *lckp;
 4105         struct nfsnode *np;
 4106         int error = 0, ret;
 4107 
 4108         if (vp == NULL) {
 4109                 KASSERT(vpp != NULL, ("nfscl_recalldeleg: vpp NULL"));
 4110                 *vpp = NULL;
 4111                 /*
 4112                  * First, get a vnode for the file. This is needed to do RPCs.
 4113                  */
 4114                 ret = nfscl_ngetreopen(nmp->nm_mountp, dp->nfsdl_fh,
 4115                     dp->nfsdl_fhlen, p, &np);
 4116                 if (ret) {
 4117                         /*
 4118                          * File isn't open, so nothing to move over to the
 4119                          * server.
 4120                          */
 4121                         return (0);
 4122                 }
 4123                 vp = NFSTOV(np);
 4124                 *vpp = vp;
 4125         } else {
 4126                 np = VTONFS(vp);
 4127         }
 4128         dp->nfsdl_flags &= ~NFSCLDL_MODTIMESET;
 4129 
 4130         /*
 4131          * Ok, if it's a write delegation, flush data to the server, so
 4132          * that close/open consistency is retained.
 4133          */
 4134         ret = 0;
 4135         NFSLOCKNODE(np);
 4136         if ((dp->nfsdl_flags & NFSCLDL_WRITE) && (np->n_flag & NMODIFIED)) {
 4137                 np->n_flag |= NDELEGRECALL;
 4138                 NFSUNLOCKNODE(np);
 4139                 ret = ncl_flush(vp, MNT_WAIT, p, 1, called_from_renewthread);
 4140                 NFSLOCKNODE(np);
 4141                 np->n_flag &= ~NDELEGRECALL;
 4142         }
 4143         NFSINVALATTRCACHE(np);
 4144         NFSUNLOCKNODE(np);
 4145         if (ret == EIO && called_from_renewthread != 0) {
 4146                 /*
 4147                  * If the flush failed with EIO for the renew thread,
 4148                  * return now, so that the dirty buffer will be flushed
 4149                  * later.
 4150                  */
 4151                 return (ret);
 4152         }
 4153 
 4154         /*
 4155          * Now, for each openowner with opens issued locally, move them
 4156          * over to state against the server.
 4157          */
 4158         LIST_FOREACH(lowp, &dp->nfsdl_owner, nfsow_list) {
 4159                 lop = LIST_FIRST(&lowp->nfsow_open);
 4160                 if (lop != NULL) {
 4161                         if (LIST_NEXT(lop, nfso_list) != NULL)
 4162                                 panic("nfsdlg mult opens");
 4163                         /*
 4164                          * Look for the same openowner against the server.
 4165                          */
 4166                         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
 4167                                 if (!NFSBCMP(lowp->nfsow_owner,
 4168                                     owp->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
 4169                                         newnfs_copycred(&dp->nfsdl_cred, cred);
 4170                                         ret = nfscl_moveopen(vp, clp, nmp, lop,
 4171                                             owp, dp, cred, p);
 4172                                         if (ret == NFSERR_STALECLIENTID ||
 4173                                             ret == NFSERR_STALEDONTRECOVER ||
 4174                                             ret == NFSERR_BADSESSION)
 4175                                                 return (ret);
 4176                                         if (ret) {
 4177                                                 nfscl_freeopen(lop, 1, true);
 4178                                                 if (!error)
 4179                                                         error = ret;
 4180                                         }
 4181                                         break;
 4182                                 }
 4183                         }
 4184 
 4185                         /*
 4186                          * If no openowner found, create one and get an open
 4187                          * for it.
 4188                          */
 4189                         if (owp == NULL) {
 4190                                 nowp = malloc(
 4191                                     sizeof (struct nfsclowner), M_NFSCLOWNER,
 4192                                     M_WAITOK);
 4193                                 nfscl_newopen(clp, NULL, &owp, &nowp, &op, 
 4194                                     NULL, lowp->nfsow_owner, dp->nfsdl_fh,
 4195                                     dp->nfsdl_fhlen, NULL, NULL);
 4196                                 newnfs_copycred(&dp->nfsdl_cred, cred);
 4197                                 ret = nfscl_moveopen(vp, clp, nmp, lop,
 4198                                     owp, dp, cred, p);
 4199                                 if (ret) {
 4200                                         nfscl_freeopenowner(owp, 0);
 4201                                         if (ret == NFSERR_STALECLIENTID ||
 4202                                             ret == NFSERR_STALEDONTRECOVER ||
 4203                                             ret == NFSERR_BADSESSION)
 4204                                                 return (ret);
 4205                                         if (ret) {
 4206                                                 nfscl_freeopen(lop, 1, true);
 4207                                                 if (!error)
 4208                                                         error = ret;
 4209                                         }
 4210                                 }
 4211                         }
 4212                 }
 4213         }
 4214 
 4215         /*
 4216          * Now, get byte range locks for any locks done locally.
 4217          */
 4218         LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
 4219                 LIST_FOREACH(lckp, &lp->nfsl_lock, nfslo_list) {
 4220                         newnfs_copycred(&dp->nfsdl_cred, cred);
 4221                         ret = nfscl_relock(vp, clp, nmp, lp, lckp, cred, p);
 4222                         if (ret == NFSERR_STALESTATEID ||
 4223                             ret == NFSERR_STALEDONTRECOVER ||
 4224                             ret == NFSERR_STALECLIENTID ||
 4225                             ret == NFSERR_BADSESSION)
 4226                                 return (ret);
 4227                         if (ret && !error)
 4228                                 error = ret;
 4229                 }
 4230         }
 4231         return (error);
 4232 }
 4233 
 4234 /*
 4235  * Move a locally issued open over to an owner on the state list.
 4236  * SIDE EFFECT: If it needs to sleep (do an rpc), it unlocks clstate and
 4237  * returns with it unlocked.
 4238  */
 4239 static int
 4240 nfscl_moveopen(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp,
 4241     struct nfsclopen *lop, struct nfsclowner *owp, struct nfscldeleg *dp,
 4242     struct ucred *cred, NFSPROC_T *p)
 4243 {
 4244         struct nfsclopen *op, *nop;
 4245         struct nfscldeleg *ndp;
 4246         struct nfsnode *np;
 4247         int error = 0, newone;
 4248 
 4249         /*
 4250          * First, look for an appropriate open, If found, just increment the
 4251          * opencnt in it.
 4252          */
 4253         LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
 4254                 if ((op->nfso_mode & lop->nfso_mode) == lop->nfso_mode &&
 4255                     op->nfso_fhlen == lop->nfso_fhlen &&
 4256                     !NFSBCMP(op->nfso_fh, lop->nfso_fh, op->nfso_fhlen)) {
 4257                         op->nfso_opencnt += lop->nfso_opencnt;
 4258                         nfscl_freeopen(lop, 1, true);
 4259                         return (0);
 4260                 }
 4261         }
 4262 
 4263         /* No appropriate open, so we have to do one against the server. */
 4264         np = VTONFS(vp);
 4265         nop = malloc(sizeof (struct nfsclopen) +
 4266             lop->nfso_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
 4267         newone = 0;
 4268         nfscl_newopen(clp, NULL, &owp, NULL, &op, &nop, owp->nfsow_owner,
 4269             lop->nfso_fh, lop->nfso_fhlen, cred, &newone);
 4270         ndp = dp;
 4271         error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data, np->n_v4->n4_fhlen,
 4272             lop->nfso_fh, lop->nfso_fhlen, lop->nfso_mode, op,
 4273             NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, &ndp, 0, 0, cred, p);
 4274         if (error) {
 4275                 if (newone)
 4276                         nfscl_freeopen(op, 0, true);
 4277         } else {
 4278                 op->nfso_mode |= lop->nfso_mode;
 4279                 op->nfso_opencnt += lop->nfso_opencnt;
 4280                 nfscl_freeopen(lop, 1, true);
 4281         }
 4282         if (nop != NULL)
 4283                 free(nop, M_NFSCLOPEN);
 4284         if (ndp != NULL) {
 4285                 /*
 4286                  * What should I do with the returned delegation, since the
 4287                  * delegation is being recalled? For now, just printf and
 4288                  * through it away.
 4289                  */
 4290                 printf("Moveopen returned deleg\n");
 4291                 free(ndp, M_NFSCLDELEG);
 4292         }
 4293         return (error);
 4294 }
 4295 
 4296 /*
 4297  * Recall all delegations on this client.
 4298  */
 4299 static void
 4300 nfscl_totalrecall(struct nfsclclient *clp)
 4301 {
 4302         struct nfscldeleg *dp;
 4303 
 4304         TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
 4305                 if ((dp->nfsdl_flags & NFSCLDL_DELEGRET) == 0)
 4306                         dp->nfsdl_flags |= NFSCLDL_RECALL;
 4307         }
 4308 }
 4309 
 4310 /*
 4311  * Relock byte ranges. Called for delegation recall and state expiry.
 4312  */
 4313 static int
 4314 nfscl_relock(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp,
 4315     struct nfscllockowner *lp, struct nfscllock *lop, struct ucred *cred,
 4316     NFSPROC_T *p)
 4317 {
 4318         struct nfscllockowner *nlp;
 4319         struct nfsfh *nfhp;
 4320         u_int64_t off, len;
 4321         int error, newone, donelocally;
 4322 
 4323         off = lop->nfslo_first;
 4324         len = lop->nfslo_end - lop->nfslo_first;
 4325         error = nfscl_getbytelock(vp, off, len, lop->nfslo_type, cred, p,
 4326             clp, 1, NULL, lp->nfsl_lockflags, lp->nfsl_owner,
 4327             lp->nfsl_openowner, &nlp, &newone, &donelocally);
 4328         if (error || donelocally)
 4329                 return (error);
 4330         nfhp = VTONFS(vp)->n_fhp;
 4331         error = nfscl_trylock(nmp, vp, nfhp->nfh_fh,
 4332             nfhp->nfh_len, nlp, newone, 0, off,
 4333             len, lop->nfslo_type, cred, p);
 4334         if (error)
 4335                 nfscl_freelockowner(nlp, 0);
 4336         return (error);
 4337 }
 4338 
 4339 /*
 4340  * Called to re-open a file. Basically get a vnode for the file handle
 4341  * and then call nfsrpc_openrpc() to do the rest.
 4342  */
 4343 static int
 4344 nfsrpc_reopen(struct nfsmount *nmp, u_int8_t *fhp, int fhlen,
 4345     u_int32_t mode, struct nfsclopen *op, struct nfscldeleg **dpp,
 4346     struct ucred *cred, NFSPROC_T *p)
 4347 {
 4348         struct nfsnode *np;
 4349         vnode_t vp;
 4350         int error;
 4351 
 4352         error = nfscl_ngetreopen(nmp->nm_mountp, fhp, fhlen, p, &np);
 4353         if (error)
 4354                 return (error);
 4355         vp = NFSTOV(np);
 4356         if (np->n_v4 != NULL) {
 4357                 error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data,
 4358                     np->n_v4->n4_fhlen, fhp, fhlen, mode, op,
 4359                     NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, dpp, 0, 0,
 4360                     cred, p);
 4361         } else {
 4362                 error = EINVAL;
 4363         }
 4364         vrele(vp);
 4365         return (error);
 4366 }
 4367 
 4368 /*
 4369  * Try an open against the server. Just call nfsrpc_openrpc(), retrying while
 4370  * NFSERR_DELAY. Also, try system credentials, if the passed in credentials
 4371  * fail.
 4372  */
 4373 static int
 4374 nfscl_tryopen(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen,
 4375     u_int8_t *newfhp, int newfhlen, u_int32_t mode, struct nfsclopen *op,
 4376     u_int8_t *name, int namelen, struct nfscldeleg **ndpp,
 4377     int reclaim, u_int32_t delegtype, struct ucred *cred, NFSPROC_T *p)
 4378 {
 4379         int error;
 4380 
 4381         do {
 4382                 error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp, newfhlen,
 4383                     mode, op, name, namelen, ndpp, reclaim, delegtype, cred, p,
 4384                     0, 0);
 4385                 if (error == NFSERR_DELAY)
 4386                         (void) nfs_catnap(PZERO, error, "nfstryop");
 4387         } while (error == NFSERR_DELAY);
 4388         if (error == EAUTH || error == EACCES) {
 4389                 /* Try again using system credentials */
 4390                 newnfs_setroot(cred);
 4391                 do {
 4392                     error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp,
 4393                         newfhlen, mode, op, name, namelen, ndpp, reclaim,
 4394                         delegtype, cred, p, 1, 0);
 4395                     if (error == NFSERR_DELAY)
 4396                         (void) nfs_catnap(PZERO, error, "nfstryop");
 4397                 } while (error == NFSERR_DELAY);
 4398         }
 4399         return (error);
 4400 }
 4401 
 4402 /*
 4403  * Try a byte range lock. Just loop on nfsrpc_lock() while it returns
 4404  * NFSERR_DELAY. Also, retry with system credentials, if the provided
 4405  * cred don't work.
 4406  */
 4407 static int
 4408 nfscl_trylock(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp,
 4409     int fhlen, struct nfscllockowner *nlp, int newone, int reclaim,
 4410     u_int64_t off, u_int64_t len, short type, struct ucred *cred, NFSPROC_T *p)
 4411 {
 4412         struct nfsrv_descript nfsd, *nd = &nfsd;
 4413         int error;
 4414 
 4415         do {
 4416                 error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp, newone,
 4417                     reclaim, off, len, type, cred, p, 0);
 4418                 if (!error && nd->nd_repstat == NFSERR_DELAY)
 4419                         (void) nfs_catnap(PZERO, (int)nd->nd_repstat,
 4420                             "nfstrylck");
 4421         } while (!error && nd->nd_repstat == NFSERR_DELAY);
 4422         if (!error)
 4423                 error = nd->nd_repstat;
 4424         if (error == EAUTH || error == EACCES) {
 4425                 /* Try again using root credentials */
 4426                 newnfs_setroot(cred);
 4427                 do {
 4428                         error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp,
 4429                             newone, reclaim, off, len, type, cred, p, 1);
 4430                         if (!error && nd->nd_repstat == NFSERR_DELAY)
 4431                                 (void) nfs_catnap(PZERO, (int)nd->nd_repstat,
 4432                                     "nfstrylck");
 4433                 } while (!error && nd->nd_repstat == NFSERR_DELAY);
 4434                 if (!error)
 4435                         error = nd->nd_repstat;
 4436         }
 4437         return (error);
 4438 }
 4439 
 4440 /*
 4441  * Try a delegreturn against the server. Just call nfsrpc_delegreturn(),
 4442  * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
 4443  * credentials fail.
 4444  */
 4445 static int
 4446 nfscl_trydelegreturn(struct nfscldeleg *dp, struct ucred *cred,
 4447     struct nfsmount *nmp, NFSPROC_T *p)
 4448 {
 4449         int error;
 4450 
 4451         do {
 4452                 error = nfsrpc_delegreturn(dp, cred, nmp, p, 0);
 4453                 if (error == NFSERR_DELAY)
 4454                         (void) nfs_catnap(PZERO, error, "nfstrydp");
 4455         } while (error == NFSERR_DELAY);
 4456         if (error == EAUTH || error == EACCES) {
 4457                 /* Try again using system credentials */
 4458                 newnfs_setroot(cred);
 4459                 do {
 4460                         error = nfsrpc_delegreturn(dp, cred, nmp, p, 1);
 4461                         if (error == NFSERR_DELAY)
 4462                                 (void) nfs_catnap(PZERO, error, "nfstrydp");
 4463                 } while (error == NFSERR_DELAY);
 4464         }
 4465         return (error);
 4466 }
 4467 
 4468 /*
 4469  * Try a close against the server. Just call nfsrpc_closerpc(),
 4470  * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
 4471  * credentials fail.
 4472  */
 4473 int
 4474 nfscl_tryclose(struct nfsclopen *op, struct ucred *cred,
 4475     struct nfsmount *nmp, NFSPROC_T *p, bool loop_on_delayed)
 4476 {
 4477         struct nfsrv_descript nfsd, *nd = &nfsd;
 4478         int error;
 4479 
 4480         do {
 4481                 error = nfsrpc_closerpc(nd, nmp, op, cred, p, 0);
 4482                 if (loop_on_delayed && error == NFSERR_DELAY)
 4483                         (void) nfs_catnap(PZERO, error, "nfstrycl");
 4484         } while (loop_on_delayed && error == NFSERR_DELAY);
 4485         if (error == EAUTH || error == EACCES) {
 4486                 /* Try again using system credentials */
 4487                 newnfs_setroot(cred);
 4488                 do {
 4489                         error = nfsrpc_closerpc(nd, nmp, op, cred, p, 1);
 4490                         if (loop_on_delayed && error == NFSERR_DELAY)
 4491                                 (void) nfs_catnap(PZERO, error, "nfstrycl");
 4492                 } while (loop_on_delayed && error == NFSERR_DELAY);
 4493         }
 4494         return (error);
 4495 }
 4496 
 4497 /*
 4498  * Decide if a delegation on a file permits close without flushing writes
 4499  * to the server. This might be a big performance win in some environments.
 4500  * (Not useful until the client does caching on local stable storage.)
 4501  */
 4502 int
 4503 nfscl_mustflush(vnode_t vp)
 4504 {
 4505         struct nfsclclient *clp;
 4506         struct nfscldeleg *dp;
 4507         struct nfsnode *np;
 4508         struct nfsmount *nmp;
 4509 
 4510         np = VTONFS(vp);
 4511         nmp = VFSTONFS(vnode_mount(vp));
 4512         if (!NFSHASNFSV4(nmp))
 4513                 return (1);
 4514         NFSLOCKCLSTATE();
 4515         clp = nfscl_findcl(nmp);
 4516         if (clp == NULL) {
 4517                 NFSUNLOCKCLSTATE();
 4518                 return (1);
 4519         }
 4520         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
 4521         if (dp != NULL && (dp->nfsdl_flags &
 4522             (NFSCLDL_WRITE | NFSCLDL_RECALL | NFSCLDL_DELEGRET)) ==
 4523              NFSCLDL_WRITE &&
 4524             (dp->nfsdl_sizelimit >= np->n_size ||
 4525              !NFSHASSTRICT3530(nmp))) {
 4526                 NFSUNLOCKCLSTATE();
 4527                 return (0);
 4528         }
 4529         NFSUNLOCKCLSTATE();
 4530         return (1);
 4531 }
 4532 
 4533 /*
 4534  * See if a (write) delegation exists for this file.
 4535  */
 4536 int
 4537 nfscl_nodeleg(vnode_t vp, int writedeleg)
 4538 {
 4539         struct nfsclclient *clp;
 4540         struct nfscldeleg *dp;
 4541         struct nfsnode *np;
 4542         struct nfsmount *nmp;
 4543 
 4544         np = VTONFS(vp);
 4545         nmp = VFSTONFS(vnode_mount(vp));
 4546         if (!NFSHASNFSV4(nmp))
 4547                 return (1);
 4548         NFSLOCKCLSTATE();
 4549         clp = nfscl_findcl(nmp);
 4550         if (clp == NULL) {
 4551                 NFSUNLOCKCLSTATE();
 4552                 return (1);
 4553         }
 4554         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
 4555         if (dp != NULL &&
 4556             (dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_DELEGRET)) == 0 &&
 4557             (writedeleg == 0 || (dp->nfsdl_flags & NFSCLDL_WRITE) ==
 4558              NFSCLDL_WRITE)) {
 4559                 NFSUNLOCKCLSTATE();
 4560                 return (0);
 4561         }
 4562         NFSUNLOCKCLSTATE();
 4563         return (1);
 4564 }
 4565 
 4566 /*
 4567  * Look for an associated delegation that should be DelegReturned.
 4568  */
 4569 int
 4570 nfscl_removedeleg(vnode_t vp, NFSPROC_T *p, nfsv4stateid_t *stp)
 4571 {
 4572         struct nfsclclient *clp;
 4573         struct nfscldeleg *dp;
 4574         struct nfsclowner *owp;
 4575         struct nfscllockowner *lp;
 4576         struct nfsmount *nmp;
 4577         struct mount *mp;
 4578         struct ucred *cred;
 4579         struct nfsnode *np;
 4580         int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
 4581 
 4582         nmp = VFSTONFS(vnode_mount(vp));
 4583         if (NFSHASPNFS(nmp))
 4584                 return (retcnt);
 4585         np = VTONFS(vp);
 4586         mp = nmp->nm_mountp;
 4587         NFSLOCKCLSTATE();
 4588         /*
 4589          * Loop around waiting for:
 4590          * - outstanding I/O operations on delegations to complete
 4591          * - for a delegation on vp that has state, lock the client and
 4592          *   do a recall
 4593          * - return delegation with no state
 4594          */
 4595         while (1) {
 4596                 clp = nfscl_findcl(nmp);
 4597                 if (clp == NULL) {
 4598                         NFSUNLOCKCLSTATE();
 4599                         return (retcnt);
 4600                 }
 4601                 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
 4602                     np->n_fhp->nfh_len);
 4603                 if (dp != NULL) {
 4604                     /*
 4605                      * Wait for outstanding I/O ops to be done.
 4606                      */
 4607                     if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
 4608                         if (igotlock) {
 4609                             nfsv4_unlock(&clp->nfsc_lock, 0);
 4610                             igotlock = 0;
 4611                         }
 4612                         dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
 4613                         msleep(&dp->nfsdl_rwlock, NFSCLSTATEMUTEXPTR, PZERO,
 4614                             "nfscld", hz);
 4615                         if (NFSCL_FORCEDISM(mp)) {
 4616                             dp->nfsdl_flags &= ~NFSCLDL_DELEGRET;
 4617                             NFSUNLOCKCLSTATE();
 4618                             return (0);
 4619                         }
 4620                         continue;
 4621                     }
 4622                     needsrecall = 0;
 4623                     LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
 4624                         if (!LIST_EMPTY(&owp->nfsow_open)) {
 4625                             needsrecall = 1;
 4626                             break;
 4627                         }
 4628                     }
 4629                     if (!needsrecall) {
 4630                         LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
 4631                             if (!LIST_EMPTY(&lp->nfsl_lock)) {
 4632                                 needsrecall = 1;
 4633                                 break;
 4634                             }
 4635                         }
 4636                     }
 4637                     if (needsrecall && !triedrecall) {
 4638                         dp->nfsdl_flags |= NFSCLDL_DELEGRET;
 4639                         islept = 0;
 4640                         while (!igotlock) {
 4641                             igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
 4642                                 &islept, NFSCLSTATEMUTEXPTR, mp);
 4643                             if (NFSCL_FORCEDISM(mp)) {
 4644                                 dp->nfsdl_flags &= ~NFSCLDL_DELEGRET;
 4645                                 if (igotlock)
 4646                                     nfsv4_unlock(&clp->nfsc_lock, 0);
 4647                                 NFSUNLOCKCLSTATE();
 4648                                 return (0);
 4649                             }
 4650                             if (islept)
 4651                                 break;
 4652                         }
 4653                         if (islept)
 4654                             continue;
 4655                         NFSUNLOCKCLSTATE();
 4656                         cred = newnfs_getcred();
 4657                         newnfs_copycred(&dp->nfsdl_cred, cred);
 4658                         nfscl_recalldeleg(clp, nmp, dp, vp, cred, p, 0, NULL);
 4659                         NFSFREECRED(cred);
 4660                         triedrecall = 1;
 4661                         NFSLOCKCLSTATE();
 4662                         nfsv4_unlock(&clp->nfsc_lock, 0);
 4663                         igotlock = 0;
 4664                         continue;
 4665                     }
 4666                     *stp = dp->nfsdl_stateid;
 4667                     retcnt = 1;
 4668                     nfscl_cleandeleg(dp);
 4669                     nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
 4670                 }
 4671                 if (igotlock)
 4672                     nfsv4_unlock(&clp->nfsc_lock, 0);
 4673                 NFSUNLOCKCLSTATE();
 4674                 return (retcnt);
 4675         }
 4676 }
 4677 
 4678 /*
 4679  * Look for associated delegation(s) that should be DelegReturned.
 4680  */
 4681 int
 4682 nfscl_renamedeleg(vnode_t fvp, nfsv4stateid_t *fstp, int *gotfdp, vnode_t tvp,
 4683     nfsv4stateid_t *tstp, int *gottdp, NFSPROC_T *p)
 4684 {
 4685         struct nfsclclient *clp;
 4686         struct nfscldeleg *dp;
 4687         struct nfsclowner *owp;
 4688         struct nfscllockowner *lp;
 4689         struct nfsmount *nmp;
 4690         struct mount *mp;
 4691         struct ucred *cred;
 4692         struct nfsnode *np;
 4693         int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
 4694 
 4695         nmp = VFSTONFS(vnode_mount(fvp));
 4696         *gotfdp = 0;
 4697         *gottdp = 0;
 4698         if (NFSHASPNFS(nmp))
 4699                 return (retcnt);
 4700         mp = nmp->nm_mountp;
 4701         NFSLOCKCLSTATE();
 4702         /*
 4703          * Loop around waiting for:
 4704          * - outstanding I/O operations on delegations to complete
 4705          * - for a delegation on fvp that has state, lock the client and
 4706          *   do a recall
 4707          * - return delegation(s) with no state.
 4708          */
 4709         while (1) {
 4710                 clp = nfscl_findcl(nmp);
 4711                 if (clp == NULL) {
 4712                         NFSUNLOCKCLSTATE();
 4713                         return (retcnt);
 4714                 }
 4715                 np = VTONFS(fvp);
 4716                 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
 4717                     np->n_fhp->nfh_len);
 4718                 if (dp != NULL && *gotfdp == 0) {
 4719                     /*
 4720                      * Wait for outstanding I/O ops to be done.
 4721                      */
 4722                     if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
 4723                         if (igotlock) {
 4724                             nfsv4_unlock(&clp->nfsc_lock, 0);
 4725                             igotlock = 0;
 4726                         }
 4727                         dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
 4728                         msleep(&dp->nfsdl_rwlock, NFSCLSTATEMUTEXPTR, PZERO,
 4729                             "nfscld", hz);
 4730                         if (NFSCL_FORCEDISM(mp)) {
 4731                             dp->nfsdl_flags &= ~NFSCLDL_DELEGRET;
 4732                             NFSUNLOCKCLSTATE();
 4733                             *gotfdp = 0;
 4734                             *gottdp = 0;
 4735                             return (0);
 4736                         }
 4737                         continue;
 4738                     }
 4739                     needsrecall = 0;
 4740                     LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
 4741                         if (!LIST_EMPTY(&owp->nfsow_open)) {
 4742                             needsrecall = 1;
 4743                             break;
 4744                         }
 4745                     }
 4746                     if (!needsrecall) {
 4747                         LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
 4748                             if (!LIST_EMPTY(&lp->nfsl_lock)) {
 4749                                 needsrecall = 1;
 4750                                 break;
 4751                             }
 4752                         }
 4753                     }
 4754                     if (needsrecall && !triedrecall) {
 4755                         dp->nfsdl_flags |= NFSCLDL_DELEGRET;
 4756                         islept = 0;
 4757                         while (!igotlock) {
 4758                             igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
 4759                                 &islept, NFSCLSTATEMUTEXPTR, mp);
 4760                             if (NFSCL_FORCEDISM(mp)) {
 4761                                 dp->nfsdl_flags &= ~NFSCLDL_DELEGRET;
 4762                                 if (igotlock)
 4763                                     nfsv4_unlock(&clp->nfsc_lock, 0);
 4764                                 NFSUNLOCKCLSTATE();
 4765                                 *gotfdp = 0;
 4766                                 *gottdp = 0;
 4767                                 return (0);
 4768                             }
 4769                             if (islept)
 4770                                 break;
 4771                         }
 4772                         if (islept)
 4773                             continue;
 4774                         NFSUNLOCKCLSTATE();
 4775                         cred = newnfs_getcred();
 4776                         newnfs_copycred(&dp->nfsdl_cred, cred);
 4777                         nfscl_recalldeleg(clp, nmp, dp, fvp, cred, p, 0, NULL);
 4778                         NFSFREECRED(cred);
 4779                         triedrecall = 1;
 4780                         NFSLOCKCLSTATE();
 4781                         nfsv4_unlock(&clp->nfsc_lock, 0);
 4782                         igotlock = 0;
 4783                         continue;
 4784                     }
 4785                     *fstp = dp->nfsdl_stateid;
 4786                     retcnt++;
 4787                     *gotfdp = 1;
 4788                     nfscl_cleandeleg(dp);
 4789                     nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
 4790                 }
 4791                 if (igotlock) {
 4792                     nfsv4_unlock(&clp->nfsc_lock, 0);
 4793                     igotlock = 0;
 4794                 }
 4795                 if (tvp != NULL) {
 4796                     np = VTONFS(tvp);
 4797                     dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
 4798                         np->n_fhp->nfh_len);
 4799                     if (dp != NULL && *gottdp == 0) {
 4800                         /*
 4801                          * Wait for outstanding I/O ops to be done.
 4802                          */
 4803                         if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
 4804                             dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
 4805                             msleep(&dp->nfsdl_rwlock, NFSCLSTATEMUTEXPTR, PZERO,
 4806                                 "nfscld", hz);
 4807                             if (NFSCL_FORCEDISM(mp)) {
 4808                                 NFSUNLOCKCLSTATE();
 4809                                 *gotfdp = 0;
 4810                                 *gottdp = 0;
 4811                                 return (0);
 4812                             }
 4813                             continue;
 4814                         }
 4815                         LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
 4816                             if (!LIST_EMPTY(&owp->nfsow_open)) {
 4817                                 NFSUNLOCKCLSTATE();
 4818                                 return (retcnt);
 4819                             }
 4820                         }
 4821                         LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
 4822                             if (!LIST_EMPTY(&lp->nfsl_lock)) {
 4823                                 NFSUNLOCKCLSTATE();
 4824                                 return (retcnt);
 4825                             }
 4826                         }
 4827                         *tstp = dp->nfsdl_stateid;
 4828                         retcnt++;
 4829                         *gottdp = 1;
 4830                         nfscl_cleandeleg(dp);
 4831                         nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
 4832                     }
 4833                 }
 4834                 NFSUNLOCKCLSTATE();
 4835                 return (retcnt);
 4836         }
 4837 }
 4838 
 4839 /*
 4840  * Get a reference on the clientid associated with the mount point.
 4841  * Return 1 if success, 0 otherwise.
 4842  */
 4843 int
 4844 nfscl_getref(struct nfsmount *nmp)
 4845 {
 4846         struct nfsclclient *clp;
 4847         int ret;
 4848 
 4849         NFSLOCKCLSTATE();
 4850         clp = nfscl_findcl(nmp);
 4851         if (clp == NULL) {
 4852                 NFSUNLOCKCLSTATE();
 4853                 return (0);
 4854         }
 4855         nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, nmp->nm_mountp);
 4856         ret = 1;
 4857         if (NFSCL_FORCEDISM(nmp->nm_mountp))
 4858                 ret = 0;
 4859         NFSUNLOCKCLSTATE();
 4860         return (ret);
 4861 }
 4862 
 4863 /*
 4864  * Release a reference on a clientid acquired with the above call.
 4865  */
 4866 void
 4867 nfscl_relref(struct nfsmount *nmp)
 4868 {
 4869         struct nfsclclient *clp;
 4870 
 4871         NFSLOCKCLSTATE();
 4872         clp = nfscl_findcl(nmp);
 4873         if (clp == NULL) {
 4874                 NFSUNLOCKCLSTATE();
 4875                 return;
 4876         }
 4877         nfsv4_relref(&clp->nfsc_lock);
 4878         NFSUNLOCKCLSTATE();
 4879 }
 4880 
 4881 /*
 4882  * Save the size attribute in the delegation, since the nfsnode
 4883  * is going away.
 4884  */
 4885 void
 4886 nfscl_reclaimnode(vnode_t vp)
 4887 {
 4888         struct nfsclclient *clp;
 4889         struct nfscldeleg *dp;
 4890         struct nfsnode *np = VTONFS(vp);
 4891         struct nfsmount *nmp;
 4892 
 4893         nmp = VFSTONFS(vnode_mount(vp));
 4894         if (!NFSHASNFSV4(nmp))
 4895                 return;
 4896         NFSLOCKCLSTATE();
 4897         clp = nfscl_findcl(nmp);
 4898         if (clp == NULL) {
 4899                 NFSUNLOCKCLSTATE();
 4900                 return;
 4901         }
 4902         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
 4903         if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE))
 4904                 dp->nfsdl_size = np->n_size;
 4905         NFSUNLOCKCLSTATE();
 4906 }
 4907 
 4908 /*
 4909  * Get the saved size attribute in the delegation, since it is a
 4910  * newly allocated nfsnode.
 4911  */
 4912 void
 4913 nfscl_newnode(vnode_t vp)
 4914 {
 4915         struct nfsclclient *clp;
 4916         struct nfscldeleg *dp;
 4917         struct nfsnode *np = VTONFS(vp);
 4918         struct nfsmount *nmp;
 4919 
 4920         nmp = VFSTONFS(vnode_mount(vp));
 4921         if (!NFSHASNFSV4(nmp))
 4922                 return;
 4923         NFSLOCKCLSTATE();
 4924         clp = nfscl_findcl(nmp);
 4925         if (clp == NULL) {
 4926                 NFSUNLOCKCLSTATE();
 4927                 return;
 4928         }
 4929         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
 4930         if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE))
 4931                 np->n_size = dp->nfsdl_size;
 4932         NFSUNLOCKCLSTATE();
 4933 }
 4934 
 4935 /*
 4936  * If there is a valid write delegation for this file, set the modtime
 4937  * to the local clock time.
 4938  */
 4939 void
 4940 nfscl_delegmodtime(vnode_t vp)
 4941 {
 4942         struct nfsclclient *clp;
 4943         struct nfscldeleg *dp;
 4944         struct nfsnode *np = VTONFS(vp);
 4945         struct nfsmount *nmp;
 4946 
 4947         nmp = VFSTONFS(vnode_mount(vp));
 4948         if (!NFSHASNFSV4(nmp))
 4949                 return;
 4950         NFSLOCKCLSTATE();
 4951         clp = nfscl_findcl(nmp);
 4952         if (clp == NULL) {
 4953                 NFSUNLOCKCLSTATE();
 4954                 return;
 4955         }
 4956         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
 4957         if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE)) {
 4958                 nanotime(&dp->nfsdl_modtime);
 4959                 dp->nfsdl_flags |= NFSCLDL_MODTIMESET;
 4960         }
 4961         NFSUNLOCKCLSTATE();
 4962 }
 4963 
 4964 /*
 4965  * If there is a valid write delegation for this file with a modtime set,
 4966  * put that modtime in mtime.
 4967  */
 4968 void
 4969 nfscl_deleggetmodtime(vnode_t vp, struct timespec *mtime)
 4970 {
 4971         struct nfsclclient *clp;
 4972         struct nfscldeleg *dp;
 4973         struct nfsnode *np = VTONFS(vp);
 4974         struct nfsmount *nmp;
 4975 
 4976         nmp = VFSTONFS(vnode_mount(vp));
 4977         if (!NFSHASNFSV4(nmp))
 4978                 return;
 4979         NFSLOCKCLSTATE();
 4980         clp = nfscl_findcl(nmp);
 4981         if (clp == NULL) {
 4982                 NFSUNLOCKCLSTATE();
 4983                 return;
 4984         }
 4985         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
 4986         if (dp != NULL &&
 4987             (dp->nfsdl_flags & (NFSCLDL_WRITE | NFSCLDL_MODTIMESET)) ==
 4988             (NFSCLDL_WRITE | NFSCLDL_MODTIMESET))
 4989                 *mtime = dp->nfsdl_modtime;
 4990         NFSUNLOCKCLSTATE();
 4991 }
 4992 
 4993 static int
 4994 nfscl_errmap(struct nfsrv_descript *nd, u_int32_t minorvers)
 4995 {
 4996         short *defaulterrp, *errp;
 4997 
 4998         if (!nd->nd_repstat)
 4999                 return (0);
 5000         if (nd->nd_procnum == NFSPROC_NOOP)
 5001                 return (txdr_unsigned(nd->nd_repstat & 0xffff));
 5002         if (nd->nd_repstat == EBADRPC)
 5003                 return (txdr_unsigned(NFSERR_BADXDR));
 5004         if (nd->nd_repstat == NFSERR_MINORVERMISMATCH ||
 5005             nd->nd_repstat == NFSERR_OPILLEGAL)
 5006                 return (txdr_unsigned(nd->nd_repstat));
 5007         if (nd->nd_repstat >= NFSERR_BADIOMODE && nd->nd_repstat < 20000 &&
 5008             minorvers > NFSV4_MINORVERSION) {
 5009                 /* NFSv4.n error. */
 5010                 return (txdr_unsigned(nd->nd_repstat));
 5011         }
 5012         if (nd->nd_procnum < NFSV4OP_CBNOPS)
 5013                 errp = defaulterrp = nfscl_cberrmap[nd->nd_procnum];
 5014         else
 5015                 return (txdr_unsigned(nd->nd_repstat));
 5016         while (*++errp)
 5017                 if (*errp == (short)nd->nd_repstat)
 5018                         return (txdr_unsigned(nd->nd_repstat));
 5019         return (txdr_unsigned(*defaulterrp));
 5020 }
 5021 
 5022 /*
 5023  * Called to find/add a layout to a client.
 5024  * This function returns the layout with a refcnt (shared lock) upon
 5025  * success (returns 0) or with no lock/refcnt on the layout when an
 5026  * error is returned.
 5027  * If a layout is passed in via lypp, it is locked (exclusively locked).
 5028  */
 5029 int
 5030 nfscl_layout(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen,
 5031     nfsv4stateid_t *stateidp, int layouttype, int retonclose,
 5032     struct nfsclflayouthead *fhlp, struct nfscllayout **lypp,
 5033     struct ucred *cred, NFSPROC_T *p)
 5034 {
 5035         struct nfsclclient *clp;
 5036         struct nfscllayout *lyp, *tlyp;
 5037         struct nfsclflayout *flp;
 5038         struct nfsnode *np = VTONFS(vp);
 5039         mount_t mp;
 5040         int layout_passed_in;
 5041 
 5042         mp = nmp->nm_mountp;
 5043         layout_passed_in = 1;
 5044         tlyp = NULL;
 5045         lyp = *lypp;
 5046         if (lyp == NULL) {
 5047                 layout_passed_in = 0;
 5048                 tlyp = malloc(sizeof(*tlyp) + fhlen - 1, M_NFSLAYOUT,
 5049                     M_WAITOK | M_ZERO);
 5050         }
 5051 
 5052         NFSLOCKCLSTATE();
 5053         clp = nmp->nm_clp;
 5054         if (clp == NULL) {
 5055                 if (layout_passed_in != 0)
 5056                         nfsv4_unlock(&lyp->nfsly_lock, 0);
 5057                 NFSUNLOCKCLSTATE();
 5058                 if (tlyp != NULL)
 5059                         free(tlyp, M_NFSLAYOUT);
 5060                 return (EPERM);
 5061         }
 5062         if (lyp == NULL) {
 5063                 /*
 5064                  * Although no lyp was passed in, another thread might have
 5065                  * allocated one. If one is found, just increment it's ref
 5066                  * count and return it.
 5067                  */
 5068                 lyp = nfscl_findlayout(clp, fhp, fhlen);
 5069                 if (lyp == NULL) {
 5070                         lyp = tlyp;
 5071                         tlyp = NULL;
 5072                         lyp->nfsly_stateid.seqid = stateidp->seqid;
 5073                         lyp->nfsly_stateid.other[0] = stateidp->other[0];
 5074                         lyp->nfsly_stateid.other[1] = stateidp->other[1];
 5075                         lyp->nfsly_stateid.other[2] = stateidp->other[2];
 5076                         lyp->nfsly_lastbyte = 0;
 5077                         LIST_INIT(&lyp->nfsly_flayread);
 5078                         LIST_INIT(&lyp->nfsly_flayrw);
 5079                         LIST_INIT(&lyp->nfsly_recall);
 5080                         lyp->nfsly_filesid[0] = np->n_vattr.na_filesid[0];
 5081                         lyp->nfsly_filesid[1] = np->n_vattr.na_filesid[1];
 5082                         lyp->nfsly_clp = clp;
 5083                         if (layouttype == NFSLAYOUT_FLEXFILE)
 5084                                 lyp->nfsly_flags = NFSLY_FLEXFILE;
 5085                         else
 5086                                 lyp->nfsly_flags = NFSLY_FILES;
 5087                         if (retonclose != 0)
 5088                                 lyp->nfsly_flags |= NFSLY_RETONCLOSE;
 5089                         lyp->nfsly_fhlen = fhlen;
 5090                         NFSBCOPY(fhp, lyp->nfsly_fh, fhlen);
 5091                         TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
 5092                         LIST_INSERT_HEAD(NFSCLLAYOUTHASH(clp, fhp, fhlen), lyp,
 5093                             nfsly_hash);
 5094                         lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
 5095                         nfscl_layoutcnt++;
 5096                 } else {
 5097                         if (retonclose != 0)
 5098                                 lyp->nfsly_flags |= NFSLY_RETONCLOSE;
 5099                         if (stateidp->seqid > lyp->nfsly_stateid.seqid)
 5100                                 lyp->nfsly_stateid.seqid = stateidp->seqid;
 5101                         TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list);
 5102                         TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
 5103                         lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
 5104                 }
 5105                 nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
 5106                 if (NFSCL_FORCEDISM(mp)) {
 5107                         NFSUNLOCKCLSTATE();
 5108                         if (tlyp != NULL)
 5109                                 free(tlyp, M_NFSLAYOUT);
 5110                         return (EPERM);
 5111                 }
 5112                 *lypp = lyp;
 5113         } else if (stateidp->seqid > lyp->nfsly_stateid.seqid)
 5114                 lyp->nfsly_stateid.seqid = stateidp->seqid;
 5115 
 5116         /* Merge the new list of File Layouts into the list. */
 5117         flp = LIST_FIRST(fhlp);
 5118         if (flp != NULL) {
 5119                 if (flp->nfsfl_iomode == NFSLAYOUTIOMODE_READ)
 5120                         nfscl_mergeflayouts(&lyp->nfsly_flayread, fhlp);
 5121                 else
 5122                         nfscl_mergeflayouts(&lyp->nfsly_flayrw, fhlp);
 5123         }
 5124         if (layout_passed_in != 0)
 5125                 nfsv4_unlock(&lyp->nfsly_lock, 1);
 5126         NFSUNLOCKCLSTATE();
 5127         if (tlyp != NULL)
 5128                 free(tlyp, M_NFSLAYOUT);
 5129         return (0);
 5130 }
 5131 
 5132 /*
 5133  * Search for a layout by MDS file handle.
 5134  * If one is found, it is returned with a refcnt (shared lock) iff
 5135  * retflpp returned non-NULL and locked (exclusive locked) iff retflpp is
 5136  * returned NULL.
 5137  */
 5138 struct nfscllayout *
 5139 nfscl_getlayout(struct nfsclclient *clp, uint8_t *fhp, int fhlen,
 5140     uint64_t off, uint32_t rwaccess, struct nfsclflayout **retflpp,
 5141     int *recalledp)
 5142 {
 5143         struct nfscllayout *lyp;
 5144         mount_t mp;
 5145         int error, igotlock;
 5146 
 5147         mp = clp->nfsc_nmp->nm_mountp;
 5148         *recalledp = 0;
 5149         *retflpp = NULL;
 5150         NFSLOCKCLSTATE();
 5151         lyp = nfscl_findlayout(clp, fhp, fhlen);
 5152         if (lyp != NULL) {
 5153                 if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
 5154                         TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list);
 5155                         TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
 5156                         lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
 5157                         error = nfscl_findlayoutforio(lyp, off, rwaccess,
 5158                             retflpp);
 5159                         if (error == 0)
 5160                                 nfsv4_getref(&lyp->nfsly_lock, NULL,
 5161                                     NFSCLSTATEMUTEXPTR, mp);
 5162                         else {
 5163                                 do {
 5164                                         igotlock = nfsv4_lock(&lyp->nfsly_lock,
 5165                                             1, NULL, NFSCLSTATEMUTEXPTR, mp);
 5166                                 } while (igotlock == 0 && !NFSCL_FORCEDISM(mp));
 5167                                 *retflpp = NULL;
 5168                         }
 5169                         if (NFSCL_FORCEDISM(mp)) {
 5170                                 lyp = NULL;
 5171                                 *recalledp = 1;
 5172                         }
 5173                 } else {
 5174                         lyp = NULL;
 5175                         *recalledp = 1;
 5176                 }
 5177         }
 5178         NFSUNLOCKCLSTATE();
 5179         return (lyp);
 5180 }
 5181 
 5182 /*
 5183  * Search for a layout by MDS file handle. If one is found, mark in to be
 5184  * recalled, if it already marked "return on close".
 5185  */
 5186 static void
 5187 nfscl_retoncloselayout(vnode_t vp, struct nfsclclient *clp, uint8_t *fhp,
 5188     int fhlen, struct nfsclrecalllayout **recallpp, struct nfscllayout **lypp)
 5189 {
 5190         struct nfscllayout *lyp;
 5191         uint32_t iomode;
 5192 
 5193         *lypp = NULL;
 5194         if (vp->v_type != VREG || !NFSHASPNFS(VFSTONFS(vnode_mount(vp))) ||
 5195             nfscl_enablecallb == 0 || nfs_numnfscbd == 0 ||
 5196             (VTONFS(vp)->n_flag & NNOLAYOUT) != 0)
 5197                 return;
 5198         lyp = nfscl_findlayout(clp, fhp, fhlen);
 5199         if (lyp != NULL && (lyp->nfsly_flags & NFSLY_RETONCLOSE) != 0) {
 5200                 if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
 5201                         iomode = 0;
 5202                         if (!LIST_EMPTY(&lyp->nfsly_flayread))
 5203                                 iomode |= NFSLAYOUTIOMODE_READ;
 5204                         if (!LIST_EMPTY(&lyp->nfsly_flayrw))
 5205                                 iomode |= NFSLAYOUTIOMODE_RW;
 5206                         nfscl_layoutrecall(NFSLAYOUTRETURN_FILE, lyp, iomode,
 5207                             0, UINT64_MAX, lyp->nfsly_stateid.seqid, 0, 0, NULL,
 5208                             *recallpp);
 5209                         NFSCL_DEBUG(4, "retoncls recall iomode=%d\n", iomode);
 5210                         *recallpp = NULL;
 5211                 }
 5212 
 5213                 /* Now, wake up renew thread to do LayoutReturn. */
 5214                 wakeup(clp);
 5215                 *lypp = lyp;
 5216         }
 5217 }
 5218 
 5219 /*
 5220  * Mark the layout to be recalled and with an error.
 5221  * Also, disable the dsp from further use.
 5222  */
 5223 void
 5224 nfscl_dserr(uint32_t op, uint32_t stat, struct nfscldevinfo *dp,
 5225     struct nfscllayout *lyp, struct nfsclds *dsp)
 5226 {
 5227         struct nfsclrecalllayout *recallp;
 5228         uint32_t iomode;
 5229 
 5230         printf("DS being disabled, error=%d\n", stat);
 5231         /* Set up the return of the layout. */
 5232         recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_WAITOK);
 5233         iomode = 0;
 5234         NFSLOCKCLSTATE();
 5235         if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
 5236                 if (!LIST_EMPTY(&lyp->nfsly_flayread))
 5237                         iomode |= NFSLAYOUTIOMODE_READ;
 5238                 if (!LIST_EMPTY(&lyp->nfsly_flayrw))
 5239                         iomode |= NFSLAYOUTIOMODE_RW;
 5240                 (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE, lyp, iomode,
 5241                     0, UINT64_MAX, lyp->nfsly_stateid.seqid, stat, op,
 5242                     dp->nfsdi_deviceid, recallp);
 5243                 NFSUNLOCKCLSTATE();
 5244                 NFSCL_DEBUG(4, "nfscl_dserr recall iomode=%d\n", iomode);
 5245         } else {
 5246                 NFSUNLOCKCLSTATE();
 5247                 free(recallp, M_NFSLAYRECALL);
 5248         }
 5249 
 5250         /* And shut the TCP connection down. */
 5251         nfscl_cancelreqs(dsp);
 5252 }
 5253 
 5254 /*
 5255  * Cancel all RPCs for this "dsp" by closing the connection.
 5256  * Also, mark the session as defunct.
 5257  * If NFSCLDS_SAMECONN is set, the connection is shared with other DSs and
 5258  * cannot be shut down.
 5259  */
 5260 void
 5261 nfscl_cancelreqs(struct nfsclds *dsp)
 5262 {
 5263         struct __rpc_client *cl;
 5264         static int non_event;
 5265 
 5266         NFSLOCKDS(dsp);
 5267         if ((dsp->nfsclds_flags & (NFSCLDS_CLOSED | NFSCLDS_SAMECONN)) == 0 &&
 5268             dsp->nfsclds_sockp != NULL &&
 5269             dsp->nfsclds_sockp->nr_client != NULL) {
 5270                 dsp->nfsclds_flags |= NFSCLDS_CLOSED;
 5271                 cl = dsp->nfsclds_sockp->nr_client;
 5272                 dsp->nfsclds_sess.nfsess_defunct = 1;
 5273                 NFSUNLOCKDS(dsp);
 5274                 CLNT_CLOSE(cl);
 5275                 /*
 5276                  * This 1sec sleep is done to reduce the number of reconnect
 5277                  * attempts made on the DS while it has failed.
 5278                  */
 5279                 tsleep(&non_event, PVFS, "ndscls", hz);
 5280                 return;
 5281         }
 5282         NFSUNLOCKDS(dsp);
 5283 }
 5284 
 5285 /*
 5286  * Dereference a layout.
 5287  */
 5288 void
 5289 nfscl_rellayout(struct nfscllayout *lyp, int exclocked)
 5290 {
 5291 
 5292         NFSLOCKCLSTATE();
 5293         if (exclocked != 0)
 5294                 nfsv4_unlock(&lyp->nfsly_lock, 0);
 5295         else
 5296                 nfsv4_relref(&lyp->nfsly_lock);
 5297         NFSUNLOCKCLSTATE();
 5298 }
 5299 
 5300 /*
 5301  * Search for a devinfo by deviceid. If one is found, return it after
 5302  * acquiring a reference count on it.
 5303  */
 5304 struct nfscldevinfo *
 5305 nfscl_getdevinfo(struct nfsclclient *clp, uint8_t *deviceid,
 5306     struct nfscldevinfo *dip)
 5307 {
 5308 
 5309         NFSLOCKCLSTATE();
 5310         if (dip == NULL)
 5311                 dip = nfscl_finddevinfo(clp, deviceid);
 5312         if (dip != NULL)
 5313                 dip->nfsdi_refcnt++;
 5314         NFSUNLOCKCLSTATE();
 5315         return (dip);
 5316 }
 5317 
 5318 /*
 5319  * Dereference a devinfo structure.
 5320  */
 5321 static void
 5322 nfscl_reldevinfo_locked(struct nfscldevinfo *dip)
 5323 {
 5324 
 5325         dip->nfsdi_refcnt--;
 5326         if (dip->nfsdi_refcnt == 0)
 5327                 wakeup(&dip->nfsdi_refcnt);
 5328 }
 5329 
 5330 /*
 5331  * Dereference a devinfo structure.
 5332  */
 5333 void
 5334 nfscl_reldevinfo(struct nfscldevinfo *dip)
 5335 {
 5336 
 5337         NFSLOCKCLSTATE();
 5338         nfscl_reldevinfo_locked(dip);
 5339         NFSUNLOCKCLSTATE();
 5340 }
 5341 
 5342 /*
 5343  * Find a layout for this file handle. Return NULL upon failure.
 5344  */
 5345 static struct nfscllayout *
 5346 nfscl_findlayout(struct nfsclclient *clp, u_int8_t *fhp, int fhlen)
 5347 {
 5348         struct nfscllayout *lyp;
 5349 
 5350         LIST_FOREACH(lyp, NFSCLLAYOUTHASH(clp, fhp, fhlen), nfsly_hash)
 5351                 if (lyp->nfsly_fhlen == fhlen &&
 5352                     !NFSBCMP(lyp->nfsly_fh, fhp, fhlen))
 5353                         break;
 5354         return (lyp);
 5355 }
 5356 
 5357 /*
 5358  * Find a devinfo for this deviceid. Return NULL upon failure.
 5359  */
 5360 static struct nfscldevinfo *
 5361 nfscl_finddevinfo(struct nfsclclient *clp, uint8_t *deviceid)
 5362 {
 5363         struct nfscldevinfo *dip;
 5364 
 5365         LIST_FOREACH(dip, &clp->nfsc_devinfo, nfsdi_list)
 5366                 if (NFSBCMP(dip->nfsdi_deviceid, deviceid, NFSX_V4DEVICEID)
 5367                     == 0)
 5368                         break;
 5369         return (dip);
 5370 }
 5371 
 5372 /*
 5373  * Merge the new file layout list into the main one, maintaining it in
 5374  * increasing offset order.
 5375  */
 5376 static void
 5377 nfscl_mergeflayouts(struct nfsclflayouthead *fhlp,
 5378     struct nfsclflayouthead *newfhlp)
 5379 {
 5380         struct nfsclflayout *flp, *nflp, *prevflp, *tflp;
 5381 
 5382         flp = LIST_FIRST(fhlp);
 5383         prevflp = NULL;
 5384         LIST_FOREACH_SAFE(nflp, newfhlp, nfsfl_list, tflp) {
 5385                 while (flp != NULL && flp->nfsfl_off < nflp->nfsfl_off) {
 5386                         prevflp = flp;
 5387                         flp = LIST_NEXT(flp, nfsfl_list);
 5388                 }
 5389                 if (prevflp == NULL)
 5390                         LIST_INSERT_HEAD(fhlp, nflp, nfsfl_list);
 5391                 else
 5392                         LIST_INSERT_AFTER(prevflp, nflp, nfsfl_list);
 5393                 prevflp = nflp;
 5394         }
 5395 }
 5396 
 5397 /*
 5398  * Add this nfscldevinfo to the client, if it doesn't already exist.
 5399  * This function consumes the structure pointed at by dip, if not NULL.
 5400  */
 5401 int
 5402 nfscl_adddevinfo(struct nfsmount *nmp, struct nfscldevinfo *dip, int ind,
 5403     struct nfsclflayout *flp)
 5404 {
 5405         struct nfsclclient *clp;
 5406         struct nfscldevinfo *tdip;
 5407         uint8_t *dev;
 5408 
 5409         NFSLOCKCLSTATE();
 5410         clp = nmp->nm_clp;
 5411         if (clp == NULL) {
 5412                 NFSUNLOCKCLSTATE();
 5413                 if (dip != NULL)
 5414                         free(dip, M_NFSDEVINFO);
 5415                 return (ENODEV);
 5416         }
 5417         if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
 5418                 dev = flp->nfsfl_dev;
 5419         else
 5420                 dev = flp->nfsfl_ffm[ind].dev;
 5421         tdip = nfscl_finddevinfo(clp, dev);
 5422         if (tdip != NULL) {
 5423                 tdip->nfsdi_layoutrefs++;
 5424                 if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
 5425                         flp->nfsfl_devp = tdip;
 5426                 else
 5427                         flp->nfsfl_ffm[ind].devp = tdip;
 5428                 nfscl_reldevinfo_locked(tdip);
 5429                 NFSUNLOCKCLSTATE();
 5430                 if (dip != NULL)
 5431                         free(dip, M_NFSDEVINFO);
 5432                 return (0);
 5433         }
 5434         if (dip != NULL) {
 5435                 LIST_INSERT_HEAD(&clp->nfsc_devinfo, dip, nfsdi_list);
 5436                 dip->nfsdi_layoutrefs = 1;
 5437                 if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
 5438                         flp->nfsfl_devp = dip;
 5439                 else
 5440                         flp->nfsfl_ffm[ind].devp = dip;
 5441         }
 5442         NFSUNLOCKCLSTATE();
 5443         if (dip == NULL)
 5444                 return (ENODEV);
 5445         return (0);
 5446 }
 5447 
 5448 /*
 5449  * Free up a layout structure and associated file layout structure(s).
 5450  */
 5451 void
 5452 nfscl_freelayout(struct nfscllayout *layp)
 5453 {
 5454         struct nfsclflayout *flp, *nflp;
 5455         struct nfsclrecalllayout *rp, *nrp;
 5456 
 5457         LIST_FOREACH_SAFE(flp, &layp->nfsly_flayread, nfsfl_list, nflp) {
 5458                 LIST_REMOVE(flp, nfsfl_list);
 5459                 nfscl_freeflayout(flp);
 5460         }
 5461         LIST_FOREACH_SAFE(flp, &layp->nfsly_flayrw, nfsfl_list, nflp) {
 5462                 LIST_REMOVE(flp, nfsfl_list);
 5463                 nfscl_freeflayout(flp);
 5464         }
 5465         LIST_FOREACH_SAFE(rp, &layp->nfsly_recall, nfsrecly_list, nrp) {
 5466                 LIST_REMOVE(rp, nfsrecly_list);
 5467                 free(rp, M_NFSLAYRECALL);
 5468         }
 5469         nfscl_layoutcnt--;
 5470         free(layp, M_NFSLAYOUT);
 5471 }
 5472 
 5473 /*
 5474  * Free up a file layout structure.
 5475  */
 5476 void
 5477 nfscl_freeflayout(struct nfsclflayout *flp)
 5478 {
 5479         int i, j;
 5480 
 5481         if ((flp->nfsfl_flags & NFSFL_FILE) != 0) {
 5482                 for (i = 0; i < flp->nfsfl_fhcnt; i++)
 5483                         free(flp->nfsfl_fh[i], M_NFSFH);
 5484                 if (flp->nfsfl_devp != NULL)
 5485                         flp->nfsfl_devp->nfsdi_layoutrefs--;
 5486         }
 5487         if ((flp->nfsfl_flags & NFSFL_FLEXFILE) != 0)
 5488                 for (i = 0; i < flp->nfsfl_mirrorcnt; i++) {
 5489                         for (j = 0; j < flp->nfsfl_ffm[i].fhcnt; j++)
 5490                                 free(flp->nfsfl_ffm[i].fh[j], M_NFSFH);
 5491                         if (flp->nfsfl_ffm[i].devp != NULL)     
 5492                                 flp->nfsfl_ffm[i].devp->nfsdi_layoutrefs--;     
 5493                 }
 5494         free(flp, M_NFSFLAYOUT);
 5495 }
 5496 
 5497 /*
 5498  * Free up a file layout devinfo structure.
 5499  */
 5500 void
 5501 nfscl_freedevinfo(struct nfscldevinfo *dip)
 5502 {
 5503 
 5504         free(dip, M_NFSDEVINFO);
 5505 }
 5506 
 5507 /*
 5508  * Mark any layouts that match as recalled.
 5509  */
 5510 static int
 5511 nfscl_layoutrecall(int recalltype, struct nfscllayout *lyp, uint32_t iomode,
 5512     uint64_t off, uint64_t len, uint32_t stateseqid, uint32_t stat, uint32_t op,
 5513     char *devid, struct nfsclrecalllayout *recallp)
 5514 {
 5515         struct nfsclrecalllayout *rp, *orp;
 5516 
 5517         recallp->nfsrecly_recalltype = recalltype;
 5518         recallp->nfsrecly_iomode = iomode;
 5519         recallp->nfsrecly_stateseqid = stateseqid;
 5520         recallp->nfsrecly_off = off;
 5521         recallp->nfsrecly_len = len;
 5522         recallp->nfsrecly_stat = stat;
 5523         recallp->nfsrecly_op = op;
 5524         if (devid != NULL)
 5525                 NFSBCOPY(devid, recallp->nfsrecly_devid, NFSX_V4DEVICEID);
 5526         /*
 5527          * Order the list as file returns first, followed by fsid and any
 5528          * returns, both in increasing stateseqid order.
 5529          * Note that the seqids wrap around, so 1 is after 0xffffffff.
 5530          * (I'm not sure this is correct because I find RFC5661 confusing
 5531          *  on this, but hopefully it will work ok.)
 5532          */
 5533         orp = NULL;
 5534         LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) {
 5535                 orp = rp;
 5536                 if ((recalltype == NFSLAYOUTRETURN_FILE &&
 5537                      (rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE ||
 5538                       nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) ||
 5539                     (recalltype != NFSLAYOUTRETURN_FILE &&
 5540                      rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE &&
 5541                      nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) {
 5542                         LIST_INSERT_BEFORE(rp, recallp, nfsrecly_list);
 5543                         break;
 5544                 }
 5545 
 5546                 /*
 5547                  * Put any error return on all the file returns that will
 5548                  * preceed this one.
 5549                  */
 5550                 if (rp->nfsrecly_recalltype == NFSLAYOUTRETURN_FILE &&
 5551                    stat != 0 && rp->nfsrecly_stat == 0) {
 5552                         rp->nfsrecly_stat = stat;
 5553                         rp->nfsrecly_op = op;
 5554                         if (devid != NULL)
 5555                                 NFSBCOPY(devid, rp->nfsrecly_devid,
 5556                                     NFSX_V4DEVICEID);
 5557                 }
 5558         }
 5559         if (rp == NULL) {
 5560                 if (orp == NULL)
 5561                         LIST_INSERT_HEAD(&lyp->nfsly_recall, recallp,
 5562                             nfsrecly_list);
 5563                 else
 5564                         LIST_INSERT_AFTER(orp, recallp, nfsrecly_list);
 5565         }
 5566         lyp->nfsly_flags |= NFSLY_RECALL;
 5567         wakeup(lyp->nfsly_clp);
 5568         return (0);
 5569 }
 5570 
 5571 /*
 5572  * Compare the two seqids for ordering. The trick is that the seqids can
 5573  * wrap around from 0xffffffff->0, so check for the cases where one
 5574  * has wrapped around.
 5575  * Return 1 if seqid1 comes before seqid2, 0 otherwise.
 5576  */
 5577 static int
 5578 nfscl_seq(uint32_t seqid1, uint32_t seqid2)
 5579 {
 5580 
 5581         if (seqid2 > seqid1 && (seqid2 - seqid1) >= 0x7fffffff)
 5582                 /* seqid2 has wrapped around. */
 5583                 return (0);
 5584         if (seqid1 > seqid2 && (seqid1 - seqid2) >= 0x7fffffff)
 5585                 /* seqid1 has wrapped around. */
 5586                 return (1);
 5587         if (seqid1 <= seqid2)
 5588                 return (1);
 5589         return (0);
 5590 }
 5591 
 5592 /*
 5593  * Do a layout return for each of the recalls.
 5594  */
 5595 static void
 5596 nfscl_layoutreturn(struct nfsmount *nmp, struct nfscllayout *lyp,
 5597     struct ucred *cred, NFSPROC_T *p)
 5598 {
 5599         struct nfsclrecalllayout *rp;
 5600         nfsv4stateid_t stateid;
 5601         int layouttype;
 5602 
 5603         NFSBCOPY(lyp->nfsly_stateid.other, stateid.other, NFSX_STATEIDOTHER);
 5604         stateid.seqid = lyp->nfsly_stateid.seqid;
 5605         if ((lyp->nfsly_flags & NFSLY_FILES) != 0)
 5606                 layouttype = NFSLAYOUT_NFSV4_1_FILES;
 5607         else
 5608                 layouttype = NFSLAYOUT_FLEXFILE;
 5609         LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) {
 5610                 (void)nfsrpc_layoutreturn(nmp, lyp->nfsly_fh,
 5611                     lyp->nfsly_fhlen, 0, layouttype,
 5612                     rp->nfsrecly_iomode, rp->nfsrecly_recalltype,
 5613                     rp->nfsrecly_off, rp->nfsrecly_len,
 5614                     &stateid, cred, p, rp->nfsrecly_stat, rp->nfsrecly_op,
 5615                     rp->nfsrecly_devid);
 5616         }
 5617 }
 5618 
 5619 /*
 5620  * Do the layout commit for a file layout.
 5621  */
 5622 static void
 5623 nfscl_dolayoutcommit(struct nfsmount *nmp, struct nfscllayout *lyp,
 5624     struct ucred *cred, NFSPROC_T *p)
 5625 {
 5626         struct nfsclflayout *flp;
 5627         uint64_t len;
 5628         int error, layouttype;
 5629 
 5630         if ((lyp->nfsly_flags & NFSLY_FILES) != 0)
 5631                 layouttype = NFSLAYOUT_NFSV4_1_FILES;
 5632         else
 5633                 layouttype = NFSLAYOUT_FLEXFILE;
 5634         LIST_FOREACH(flp, &lyp->nfsly_flayrw, nfsfl_list) {
 5635                 if (layouttype == NFSLAYOUT_FLEXFILE &&
 5636                     (flp->nfsfl_fflags & NFSFLEXFLAG_NO_LAYOUTCOMMIT) != 0) {
 5637                         NFSCL_DEBUG(4, "Flex file: no layoutcommit\n");
 5638                         /* If not supported, don't bother doing it. */
 5639                         NFSLOCKMNT(nmp);
 5640                         nmp->nm_state |= NFSSTA_NOLAYOUTCOMMIT;
 5641                         NFSUNLOCKMNT(nmp);
 5642                         break;
 5643                 } else if (flp->nfsfl_off <= lyp->nfsly_lastbyte) {
 5644                         len = flp->nfsfl_end - flp->nfsfl_off;
 5645                         error = nfsrpc_layoutcommit(nmp, lyp->nfsly_fh,
 5646                             lyp->nfsly_fhlen, 0, flp->nfsfl_off, len,
 5647                             lyp->nfsly_lastbyte, &lyp->nfsly_stateid,
 5648                             layouttype, cred, p, NULL);
 5649                         NFSCL_DEBUG(4, "layoutcommit err=%d\n", error);
 5650                         if (error == NFSERR_NOTSUPP) {
 5651                                 /* If not supported, don't bother doing it. */
 5652                                 NFSLOCKMNT(nmp);
 5653                                 nmp->nm_state |= NFSSTA_NOLAYOUTCOMMIT;
 5654                                 NFSUNLOCKMNT(nmp);
 5655                                 break;
 5656                         }
 5657                 }
 5658         }
 5659 }
 5660 
 5661 /*
 5662  * Commit all layouts for a file (vnode).
 5663  */
 5664 int
 5665 nfscl_layoutcommit(vnode_t vp, NFSPROC_T *p)
 5666 {
 5667         struct nfsclclient *clp;
 5668         struct nfscllayout *lyp;
 5669         struct nfsnode *np = VTONFS(vp);
 5670         mount_t mp;
 5671         struct nfsmount *nmp;
 5672 
 5673         mp = vnode_mount(vp);
 5674         nmp = VFSTONFS(mp);
 5675         if (NFSHASNOLAYOUTCOMMIT(nmp))
 5676                 return (0);
 5677         NFSLOCKCLSTATE();
 5678         clp = nmp->nm_clp;
 5679         if (clp == NULL) {
 5680                 NFSUNLOCKCLSTATE();
 5681                 return (EPERM);
 5682         }
 5683         lyp = nfscl_findlayout(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
 5684         if (lyp == NULL) {
 5685                 NFSUNLOCKCLSTATE();
 5686                 return (EPERM);
 5687         }
 5688         nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
 5689         if (NFSCL_FORCEDISM(mp)) {
 5690                 NFSUNLOCKCLSTATE();
 5691                 return (EPERM);
 5692         }
 5693 tryagain:
 5694         if ((lyp->nfsly_flags & NFSLY_WRITTEN) != 0) {
 5695                 lyp->nfsly_flags &= ~NFSLY_WRITTEN;
 5696                 NFSUNLOCKCLSTATE();
 5697                 NFSCL_DEBUG(4, "do layoutcommit2\n");
 5698                 nfscl_dolayoutcommit(clp->nfsc_nmp, lyp, NFSPROCCRED(p), p);
 5699                 NFSLOCKCLSTATE();
 5700                 goto tryagain;
 5701         }
 5702         nfsv4_relref(&lyp->nfsly_lock);
 5703         NFSUNLOCKCLSTATE();
 5704         return (0);
 5705 }
 5706 

Cache object: 58ac8d1e11099083bc42837d6b583e39


[ 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.