sys/fs/nfsclient/nfs_clstate.c

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