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  *   - When a process exits, it calls nfscl_cleanup(), which goes
  68  *     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 NFSREQSPINLOCK;
  90 NFSCLSTATEMUTEX;
  91 int nfscl_inited = 0;
  92 struct nfsclhead nfsclhead;     /* Head of clientid list */
  93 int nfscl_deleghighwater = NFSCLDELEGHIGHWATER;
  94 #endif  /* !APPLEKEXT */
  95
  96 static int nfscl_delegcnt = 0;
  97 static int nfscl_getopen(struct nfsclownerhead *, u_int8_t *, int, u_int8_t *,
  98     u_int8_t *, u_int32_t, struct nfscllockowner **, struct nfsclopen **);
  99 static void nfscl_clrelease(struct nfsclclient *);
 100 static void nfscl_cleanclient(struct nfsclclient *);
 101 static void nfscl_expireclient(struct nfsclclient *, struct nfsmount *,
 102     struct ucred *, NFSPROC_T *);
 103 static int nfscl_expireopen(struct nfsclclient *, struct nfsclopen *,
 104     struct nfsmount *, struct ucred *, NFSPROC_T *);
 105 static void nfscl_recover(struct nfsclclient *, struct ucred *, NFSPROC_T *);
 106 static void nfscl_insertlock(struct nfscllockowner *, struct nfscllock *,
 107     struct nfscllock *, int);
 108 static int nfscl_updatelock(struct nfscllockowner *, struct nfscllock **,
 109     struct nfscllock **, int);
 110 static void nfscl_delegreturnall(struct nfsclclient *, NFSPROC_T *);
 111 static u_int32_t nfscl_nextcbident(void);
 112 static mount_t nfscl_getmnt(u_int32_t);
 113 static struct nfscldeleg *nfscl_finddeleg(struct nfsclclient *, u_int8_t *,
 114     int);
 115 static int nfscl_checkconflict(struct nfscllockownerhead *, struct nfscllock *,
 116     u_int8_t *, struct nfscllock **);
 117 static void nfscl_freealllocks(struct nfscllockownerhead *, int);
 118 static int nfscl_localconflict(struct nfsclclient *, u_int8_t *, int,
 119     struct nfscllock *, u_int8_t *, struct nfscldeleg *, struct nfscllock **);
 120 static void nfscl_newopen(struct nfsclclient *, struct nfscldeleg *,
 121     struct nfsclowner **, struct nfsclowner **, struct nfsclopen **,
 122     struct nfsclopen **, u_int8_t *, u_int8_t *, int, int *);
 123 static int nfscl_moveopen(vnode_t , struct nfsclclient *,
 124     struct nfsmount *, struct nfsclopen *, struct nfsclowner *,
 125     struct nfscldeleg *, struct ucred *, NFSPROC_T *);
 126 static void nfscl_totalrecall(struct nfsclclient *);
 127 static int nfscl_relock(vnode_t , struct nfsclclient *, struct nfsmount *,
 128     struct nfscllockowner *, struct nfscllock *, struct ucred *, NFSPROC_T *);
 129 static int nfscl_tryopen(struct nfsmount *, vnode_t , u_int8_t *, int,
 130     u_int8_t *, int, u_int32_t, struct nfsclopen *, u_int8_t *, int,
 131     struct nfscldeleg **, int, u_int32_t, struct ucred *, NFSPROC_T *);
 132 static int nfscl_trylock(struct nfsmount *, vnode_t , u_int8_t *,
 133     int, struct nfscllockowner *, int, int, u_int64_t, u_int64_t, short,
 134     struct ucred *, NFSPROC_T *);
 135 static int nfsrpc_reopen(struct nfsmount *, u_int8_t *, int, u_int32_t,
 136     struct nfsclopen *, struct nfscldeleg **, struct ucred *, NFSPROC_T *);
 137 static void nfscl_freedeleg(struct nfscldeleghead *, struct nfscldeleg *);
 138 static int nfscl_errmap(struct nfsrv_descript *);
 139 static void nfscl_cleanup_common(struct nfsclclient *, u_int8_t *);
 140 static int nfscl_recalldeleg(struct nfsclclient *, struct nfsmount *,
 141     struct nfscldeleg *, vnode_t, struct ucred *, NFSPROC_T *, int);
 142 static void nfscl_freeopenowner(struct nfsclowner *, int);
 143 static void nfscl_cleandeleg(struct nfscldeleg *);
 144 static int nfscl_trydelegreturn(struct nfscldeleg *, struct ucred *,
 145     struct nfsmount *, NFSPROC_T *);
 146
 147 static short nfscberr_null[] = {
 148         0,
 149         0,
 150 };
 151
 152 static short nfscberr_getattr[] = {
 153         NFSERR_RESOURCE,
 154         NFSERR_BADHANDLE,
 155         NFSERR_BADXDR,
 156         NFSERR_RESOURCE,
 157         NFSERR_SERVERFAULT,
 158         0,
 159 };
 160
 161 static short nfscberr_recall[] = {
 162         NFSERR_RESOURCE,
 163         NFSERR_BADHANDLE,
 164         NFSERR_BADSTATEID,
 165         NFSERR_BADXDR,
 166         NFSERR_RESOURCE,
 167         NFSERR_SERVERFAULT,
 168         0,
 169 };
 170
 171 static short *nfscl_cberrmap[] = {
 172         nfscberr_null,
 173         nfscberr_null,
 174         nfscberr_null,
 175         nfscberr_getattr,
 176         nfscberr_recall
 177 };
 178
 179 #define NETFAMILY(clp) \
 180                 (((clp)->nfsc_flags & NFSCLFLAGS_AFINET6) ? AF_INET6 : AF_INET)
 181
 182 /*
 183  * Called for an open operation.
 184  * If the nfhp argument is NULL, just get an openowner.
 185  */
 186 APPLESTATIC int
 187 nfscl_open(vnode_t vp, u_int8_t *nfhp, int fhlen, u_int32_t amode, int usedeleg,
 188     struct ucred *cred, NFSPROC_T *p, struct nfsclowner **owpp,
 189     struct nfsclopen **opp, int *newonep, int *retp, int lockit)
 190 {
 191         struct nfsclclient *clp;
 192         struct nfsclowner *owp, *nowp;
 193         struct nfsclopen *op = NULL, *nop = NULL;
 194         struct nfscldeleg *dp;
 195         struct nfsclownerhead *ohp;
 196         u_int8_t own[NFSV4CL_LOCKNAMELEN];
 197         int ret;
 198
 199         if (newonep != NULL)
 200                 *newonep = 0;
 201         if (opp != NULL)
 202                 *opp = NULL;
 203         if (owpp != NULL)
 204                 *owpp = NULL;
 205
 206         /*
 207          * Might need one or both of these, so MALLOC them now, to
 208          * avoid a tsleep() in MALLOC later.
 209          */
 210         MALLOC(nowp, struct nfsclowner *, sizeof (struct nfsclowner),
 211             M_NFSCLOWNER, M_WAITOK);
 212         if (nfhp != NULL)
 213             MALLOC(nop, struct nfsclopen *, sizeof (struct nfsclopen) +
 214                 fhlen - 1, M_NFSCLOPEN, M_WAITOK);
 215         ret = nfscl_getcl(vp, cred, p, &clp);
 216         if (ret != 0) {
 217                 FREE((caddr_t)nowp, M_NFSCLOWNER);
 218                 if (nop != NULL)
 219                         FREE((caddr_t)nop, M_NFSCLOPEN);
 220                 return (ret);
 221         }
 222
 223         /*
 224          * Get the Open iff it already exists.
 225          * If none found, add the new one or return error, depending upon
 226          * "create".
 227          */
 228         nfscl_filllockowner(p->td_proc, own, F_POSIX);
 229         NFSLOCKCLSTATE();
 230         dp = NULL;
 231         /* First check the delegation list */
 232         if (nfhp != NULL && usedeleg) {
 233                 LIST_FOREACH(dp, NFSCLDELEGHASH(clp, nfhp, fhlen), nfsdl_hash) {
 234                         if (dp->nfsdl_fhlen == fhlen &&
 235                             !NFSBCMP(nfhp, dp->nfsdl_fh, fhlen)) {
 236                                 if (!(amode & NFSV4OPEN_ACCESSWRITE) ||
 237                                     (dp->nfsdl_flags & NFSCLDL_WRITE))
 238                                         break;
 239                                 dp = NULL;
 240                                 break;
 241                         }
 242                 }
 243         }
 244
 245         if (dp != NULL)
 246                 ohp = &dp->nfsdl_owner;
 247         else
 248                 ohp = &clp->nfsc_owner;
 249         /* Now, search for an openowner */
 250         LIST_FOREACH(owp, ohp, nfsow_list) {
 251                 if (!NFSBCMP(owp->nfsow_owner, own, NFSV4CL_LOCKNAMELEN))
 252                         break;
 253         }
 254
 255         /*
 256          * Create a new open, as required.
 257          */
 258         nfscl_newopen(clp, dp, &owp, &nowp, &op, &nop, own, nfhp, fhlen,
 259             newonep);
 260
 261         /*
 262          * Serialize modifications to the open owner for multiple threads
 263          * within the same process using a read/write sleep lock.
 264          */
 265         if (lockit)
 266                 nfscl_lockexcl(&owp->nfsow_rwlock, NFSCLSTATEMUTEXPTR);
 267         NFSUNLOCKCLSTATE();
 268         if (nowp != NULL)
 269                 FREE((caddr_t)nowp, M_NFSCLOWNER);
 270         if (nop != NULL)
 271                 FREE((caddr_t)nop, M_NFSCLOPEN);
 272         if (owpp != NULL)
 273                 *owpp = owp;
 274         if (opp != NULL)
 275                 *opp = op;
 276         if (retp != NULL) {
 277                 if (nfhp != NULL && dp != NULL && nop == NULL)
 278                         /* new local open on delegation */
 279                         *retp = NFSCLOPEN_SETCRED;
 280                 else
 281                         *retp = NFSCLOPEN_OK;
 282         }
 283
 284         /*
 285          * Now, check the mode on the open and return the appropriate
 286          * value.
 287          */
 288         if (op != NULL && (amode & ~(op->nfso_mode))) {
 289                 op->nfso_mode |= amode;
 290                 if (retp != NULL && dp == NULL)
 291                         *retp = NFSCLOPEN_DOOPEN;
 292         }
 293         return (0);
 294 }
 295
 296 /*
 297  * Create a new open, as required.
 298  */
 299 static void
 300 nfscl_newopen(struct nfsclclient *clp, struct nfscldeleg *dp,
 301     struct nfsclowner **owpp, struct nfsclowner **nowpp, struct nfsclopen **opp,
 302     struct nfsclopen **nopp, u_int8_t *own, u_int8_t *fhp, int fhlen,
 303     int *newonep)
 304 {
 305         struct nfsclowner *owp = *owpp, *nowp;
 306         struct nfsclopen *op, *nop;
 307
 308         if (nowpp != NULL)
 309                 nowp = *nowpp;
 310         else
 311                 nowp = NULL;
 312         if (nopp != NULL)
 313                 nop = *nopp;
 314         else
 315                 nop = NULL;
 316         if (owp == NULL && nowp != NULL) {
 317                 NFSBCOPY(own, nowp->nfsow_owner, NFSV4CL_LOCKNAMELEN);
 318                 LIST_INIT(&nowp->nfsow_open);
 319                 nowp->nfsow_clp = clp;
 320                 nowp->nfsow_seqid = 0;
 321                 nowp->nfsow_defunct = 0;
 322                 nfscl_lockinit(&nowp->nfsow_rwlock);
 323                 if (dp != NULL) {
 324                         newnfsstats.cllocalopenowners++;
 325                         LIST_INSERT_HEAD(&dp->nfsdl_owner, nowp, nfsow_list);
 326                 } else {
 327                         newnfsstats.clopenowners++;
 328                         LIST_INSERT_HEAD(&clp->nfsc_owner, nowp, nfsow_list);
 329                 }
 330                 owp = *owpp = nowp;
 331                 *nowpp = NULL;
 332                 if (newonep != NULL)
 333                         *newonep = 1;
 334         }
 335
 336          /* If an fhp has been specified, create an Open as well. */
 337         if (fhp != NULL) {
 338                 /* and look for the correct open, based upon FH */
 339                 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
 340                         if (op->nfso_fhlen == fhlen &&
 341                             !NFSBCMP(op->nfso_fh, fhp, fhlen))
 342                                 break;
 343                 }
 344                 if (op == NULL && nop != NULL) {
 345                         nop->nfso_own = owp;
 346                         nop->nfso_mode = 0;
 347                         nop->nfso_opencnt = 0;
 348                         nop->nfso_posixlock = 1;
 349                         nop->nfso_fhlen = fhlen;
 350                         NFSBCOPY(fhp, nop->nfso_fh, fhlen);
 351                         LIST_INIT(&nop->nfso_lock);
 352                         nop->nfso_stateid.seqid = 0;
 353                         nop->nfso_stateid.other[0] = 0;
 354                         nop->nfso_stateid.other[1] = 0;
 355                         nop->nfso_stateid.other[2] = 0;
 356                         if (dp != NULL) {
 357                                 TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
 358                                 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp,
 359                                     nfsdl_list);
 360                                 dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
 361                                 newnfsstats.cllocalopens++;
 362                         } else {
 363                                 newnfsstats.clopens++;
 364                         }
 365                         LIST_INSERT_HEAD(&owp->nfsow_open, nop, nfso_list);
 366                         *opp = nop;
 367                         *nopp = NULL;
 368                         if (newonep != NULL)
 369                                 *newonep = 1;
 370                 } else {
 371                         *opp = op;
 372                 }
 373         }
 374 }
 375
 376 /*
 377  * Called to find/add a delegation to a client.
 378  */
 379 APPLESTATIC int
 380 nfscl_deleg(mount_t mp, struct nfsclclient *clp, u_int8_t *nfhp,
 381     int fhlen, struct ucred *cred, NFSPROC_T *p, struct nfscldeleg **dpp)
 382 {
 383         struct nfscldeleg *dp = *dpp, *tdp;
 384
 385         /*
 386          * First, if we have received a Read delegation for a file on a
 387          * read/write file system, just return it, because they aren't
 388          * useful, imho.
 389          */
 390         if (mp != NULL && dp != NULL && !NFSMNT_RDONLY(mp) &&
 391             (dp->nfsdl_flags & NFSCLDL_READ)) {
 392                 (void) nfscl_trydelegreturn(dp, cred, VFSTONFS(mp), p);
 393                 FREE((caddr_t)dp, M_NFSCLDELEG);
 394                 *dpp = NULL;
 395                 return (0);
 396         }
 397
 398         /* Look for the correct deleg, based upon FH */
 399         NFSLOCKCLSTATE();
 400         tdp = nfscl_finddeleg(clp, nfhp, fhlen);
 401         if (tdp == NULL) {
 402                 if (dp == NULL) {
 403                         NFSUNLOCKCLSTATE();
 404                         return (NFSERR_BADSTATEID);
 405                 }
 406                 *dpp = NULL;
 407                 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list);
 408                 LIST_INSERT_HEAD(NFSCLDELEGHASH(clp, nfhp, fhlen), dp,
 409                     nfsdl_hash);
 410                 dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
 411                 newnfsstats.cldelegates++;
 412                 nfscl_delegcnt++;
 413         } else {
 414                 /*
 415                  * Delegation already exists, what do we do if a new one??
 416                  */
 417                 if (dp != NULL) {
 418                         printf("Deleg already exists!\n");
 419                         FREE((caddr_t)dp, M_NFSCLDELEG);
 420                         *dpp = NULL;
 421                 } else {
 422                         *dpp = tdp;
 423                 }
 424         }
 425         NFSUNLOCKCLSTATE();
 426         return (0);
 427 }
 428
 429 /*
 430  * Find a delegation for this file handle. Return NULL upon failure.
 431  */
 432 static struct nfscldeleg *
 433 nfscl_finddeleg(struct nfsclclient *clp, u_int8_t *fhp, int fhlen)
 434 {
 435         struct nfscldeleg *dp;
 436
 437         LIST_FOREACH(dp, NFSCLDELEGHASH(clp, fhp, fhlen), nfsdl_hash) {
 438             if (dp->nfsdl_fhlen == fhlen &&
 439                 !NFSBCMP(dp->nfsdl_fh, fhp, fhlen))
 440                 break;
 441         }
 442         return (dp);
 443 }
 444
 445 /*
 446  * Get a stateid for an I/O operation. First, look for an open and iff
 447  * found, return either a lockowner stateid or the open stateid.
 448  * If no Open is found, just return error and the special stateid of all zeros.
 449  */
 450 APPLESTATIC int
 451 nfscl_getstateid(vnode_t vp, u_int8_t *nfhp, int fhlen, u_int32_t mode,
 452     struct ucred *cred, NFSPROC_T *p, nfsv4stateid_t *stateidp,
 453     void **lckpp)
 454 {
 455         struct nfsclclient *clp;
 456         struct nfsclowner *owp;
 457         struct nfsclopen *op = NULL;
 458         struct nfscllockowner *lp;
 459         struct nfscldeleg *dp;
 460         struct nfsnode *np;
 461         u_int8_t own[NFSV4CL_LOCKNAMELEN];
 462         int error, done;
 463
 464         *lckpp = NULL;
 465         /*
 466          * Initially, just set the special stateid of all zeros.
 467          */
 468         stateidp->seqid = 0;
 469         stateidp->other[0] = 0;
 470         stateidp->other[1] = 0;
 471         stateidp->other[2] = 0;
 472         if (vnode_vtype(vp) != VREG)
 473                 return (EISDIR);
 474         np = VTONFS(vp);
 475         NFSLOCKCLSTATE();
 476         clp = nfscl_findcl(VFSTONFS(vnode_mount(vp)));
 477         if (clp == NULL) {
 478                 NFSUNLOCKCLSTATE();
 479                 return (EACCES);
 480         }
 481
 482         /*
 483          * Wait for recovery to complete.
 484          */
 485         while ((clp->nfsc_flags & NFSCLFLAGS_RECVRINPROG))
 486                 (void) nfsmsleep(&clp->nfsc_flags, NFSCLSTATEMUTEXPTR,
 487                     PZERO, "nfsrecvr", NULL);
 488
 489         /*
 490          * First, look for a delegation.
 491          */
 492         LIST_FOREACH(dp, NFSCLDELEGHASH(clp, nfhp, fhlen), nfsdl_hash) {
 493                 if (dp->nfsdl_fhlen == fhlen &&
 494                     !NFSBCMP(nfhp, dp->nfsdl_fh, fhlen)) {
 495                         if (!(mode & NFSV4OPEN_ACCESSWRITE) ||
 496                             (dp->nfsdl_flags & NFSCLDL_WRITE)) {
 497                                 stateidp->seqid = dp->nfsdl_stateid.seqid;
 498                                 stateidp->other[0] = dp->nfsdl_stateid.other[0];
 499                                 stateidp->other[1] = dp->nfsdl_stateid.other[1];
 500                                 stateidp->other[2] = dp->nfsdl_stateid.other[2];
 501                                 if (!(np->n_flag & NDELEGRECALL)) {
 502                                         TAILQ_REMOVE(&clp->nfsc_deleg, dp,
 503                                             nfsdl_list);
 504                                         TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp,
 505                                             nfsdl_list);
 506                                         dp->nfsdl_timestamp = NFSD_MONOSEC +
 507                                             120;
 508                                         dp->nfsdl_rwlock.nfslock_usecnt++;
 509                                         *lckpp = (void *)&dp->nfsdl_rwlock;
 510                                 }
 511                                 NFSUNLOCKCLSTATE();
 512                                 return (0);
 513                         }
 514                         break;
 515                 }
 516         }
 517
 518         if (p != NULL) {
 519                 /*
 520                  * If p != NULL, we want to search the parentage tree
 521                  * for a matching OpenOwner and use that.
 522                  */
 523                 nfscl_filllockowner(p->td_proc, own, F_POSIX);
 524                 lp = NULL;
 525                 error = nfscl_getopen(&clp->nfsc_owner, nfhp, fhlen, own, own,
 526                     mode, &lp, &op);
 527                 if (error == 0 && lp != NULL) {
 528                         stateidp->seqid =
 529                             lp->nfsl_stateid.seqid;
 530                         stateidp->other[0] =
 531                             lp->nfsl_stateid.other[0];
 532                         stateidp->other[1] =
 533                             lp->nfsl_stateid.other[1];
 534                         stateidp->other[2] =
 535                             lp->nfsl_stateid.other[2];
 536                         NFSUNLOCKCLSTATE();
 537                         return (0);
 538                 }
 539         }
 540         if (op == NULL) {
 541                 /* If not found, just look for any OpenOwner that will work. */
 542                 done = 0;
 543                 owp = LIST_FIRST(&clp->nfsc_owner);
 544                 while (!done && owp != NULL) {
 545                         LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
 546                                 if (op->nfso_fhlen == fhlen &&
 547                                     !NFSBCMP(op->nfso_fh, nfhp, fhlen) &&
 548                                     (mode & op->nfso_mode) == mode) {
 549                                         done = 1;
 550                                         break;
 551                                 }
 552                         }
 553                         if (!done)
 554                                 owp = LIST_NEXT(owp, nfsow_list);
 555                 }
 556                 if (!done) {
 557                         NFSUNLOCKCLSTATE();
 558                         return (ENOENT);
 559                 }
 560                 /* for read aheads or write behinds, use the open cred */
 561                 newnfs_copycred(&op->nfso_cred, cred);
 562         }
 563
 564         /*
 565          * No lock stateid, so return the open stateid.
 566          */
 567         stateidp->seqid = op->nfso_stateid.seqid;
 568         stateidp->other[0] = op->nfso_stateid.other[0];
 569         stateidp->other[1] = op->nfso_stateid.other[1];
 570         stateidp->other[2] = op->nfso_stateid.other[2];
 571         NFSUNLOCKCLSTATE();
 572         return (0);
 573 }
 574
 575 /*
 576  * Search for a matching file, mode and, optionally, lockowner.
 577  */
 578 static int
 579 nfscl_getopen(struct nfsclownerhead *ohp, u_int8_t *nfhp, int fhlen,
 580     u_int8_t *openown, u_int8_t *lockown, u_int32_t mode,
 581     struct nfscllockowner **lpp, struct nfsclopen **opp)
 582 {
 583         struct nfsclowner *owp;
 584         struct nfsclopen *op, *rop, *rop2;
 585         struct nfscllockowner *lp;
 586         int keep_looping;
 587
 588         if (lpp != NULL)
 589                 *lpp = NULL;
 590         /*
 591          * rop will be set to the open to be returned. There are three
 592          * variants of this, all for an open of the correct file:
 593          * 1 - A match of lockown.
 594          * 2 - A match of the openown, when no lockown match exists.
 595          * 3 - A match for any open, if no openown or lockown match exists.
 596          * Looking for #2 over #3 probably isn't necessary, but since
 597          * RFC3530 is vague w.r.t. the relationship between openowners and
 598          * lockowners, I think this is the safer way to go.
 599          */
 600         rop = NULL;
 601         rop2 = NULL;
 602         keep_looping = 1;
 603         /* Search the client list */
 604         owp = LIST_FIRST(ohp);
 605         while (owp != NULL && keep_looping != 0) {
 606                 /* and look for the correct open */
 607                 op = LIST_FIRST(&owp->nfsow_open);
 608                 while (op != NULL && keep_looping != 0) {
 609                         if (op->nfso_fhlen == fhlen &&
 610                             !NFSBCMP(op->nfso_fh, nfhp, fhlen)
 611                             && (op->nfso_mode & mode) == mode) {
 612                                 if (lpp != NULL) {
 613                                         /* Now look for a matching lockowner. */
 614                                         LIST_FOREACH(lp, &op->nfso_lock,
 615                                             nfsl_list) {
 616                                                 if (!NFSBCMP(lp->nfsl_owner,
 617                                                     lockown,
 618                                                     NFSV4CL_LOCKNAMELEN)) {
 619                                                         *lpp = lp;
 620                                                         rop = op;
 621                                                         keep_looping = 0;
 622                                                         break;
 623                                                 }
 624                                         }
 625                                 }
 626                                 if (rop == NULL && !NFSBCMP(owp->nfsow_owner,
 627                                     openown, NFSV4CL_LOCKNAMELEN)) {
 628                                         rop = op;
 629                                         if (lpp == NULL)
 630                                                 keep_looping = 0;
 631                                 }
 632                                 if (rop2 == NULL)
 633                                         rop2 = op;
 634                         }
 635                         op = LIST_NEXT(op, nfso_list);
 636                 }
 637                 owp = LIST_NEXT(owp, nfsow_list);
 638         }
 639         if (rop == NULL)
 640                 rop = rop2;
 641         if (rop == NULL)
 642                 return (EBADF);
 643         *opp = rop;
 644         return (0);
 645 }
 646
 647 /*
 648  * Release use of an open owner. Called when open operations are done
 649  * with the open owner.
 650  */
 651 APPLESTATIC void
 652 nfscl_ownerrelease(struct nfsclowner *owp, __unused int error,
 653     __unused int candelete, int unlocked)
 654 {
 655
 656         if (owp == NULL)
 657                 return;
 658         NFSLOCKCLSTATE();
 659         if (!unlocked)
 660                 nfscl_lockunlock(&owp->nfsow_rwlock);
 661         nfscl_clrelease(owp->nfsow_clp);
 662         NFSUNLOCKCLSTATE();
 663 }
 664
 665 /*
 666  * Release use of an open structure under an open owner.
 667  */
 668 APPLESTATIC void
 669 nfscl_openrelease(struct nfsclopen *op, int error, int candelete)
 670 {
 671         struct nfsclclient *clp;
 672         struct nfsclowner *owp;
 673
 674         if (op == NULL)
 675                 return;
 676         NFSLOCKCLSTATE();
 677         owp = op->nfso_own;
 678         nfscl_lockunlock(&owp->nfsow_rwlock);
 679         clp = owp->nfsow_clp;
 680         if (error && candelete && op->nfso_opencnt == 0)
 681                 nfscl_freeopen(op, 0);
 682         nfscl_clrelease(clp);
 683         NFSUNLOCKCLSTATE();
 684 }
 685
 686 /*
 687  * Called to get a clientid structure. It will optionally lock the
 688  * client data structures to do the SetClientId/SetClientId_confirm,
 689  * but will release that lock and return the clientid with a refernce
 690  * count on it.
 691  * If the "cred" argument is NULL, a new clientid should not be created.
 692  * If the "p" argument is NULL, a SetClientID/SetClientIDConfirm cannot
 693  * be done.
 694  * It always clpp with a reference count on it, unless returning an error.
 695  */
 696 APPLESTATIC int
 697 nfscl_getcl(vnode_t vp, struct ucred *cred, NFSPROC_T *p,
 698     struct nfsclclient **clpp)
 699 {
 700         struct nfsclclient *clp;
 701         struct nfsclclient *newclp = NULL;
 702         struct nfscllockowner *lp, *nlp;
 703         struct mount *mp;
 704         struct nfsmount *nmp;
 705         char uuid[HOSTUUIDLEN];
 706         int igotlock = 0, error, trystalecnt, clidinusedelay, i;
 707         u_int16_t idlen = 0;
 708
 709         mp = vnode_mount(vp);
 710         nmp = VFSTONFS(mp);
 711         if (cred != NULL) {
 712                 getcredhostuuid(cred, uuid, sizeof uuid);
 713                 idlen = strlen(uuid);
 714                 if (idlen > 0)
 715                         idlen += sizeof (u_int64_t);
 716                 else
 717                         idlen += sizeof (u_int64_t) + 16; /* 16 random bytes */
 718                 MALLOC(newclp, struct nfsclclient *,
 719                     sizeof (struct nfsclclient) + idlen - 1, M_NFSCLCLIENT,
 720                     M_WAITOK);
 721         }
 722         NFSLOCKCLSTATE();
 723         /*
 724          * If a forced dismount is already in progress, don't
 725          * allocate a new clientid and get out now. For the case where
 726          * clp != NULL, this is a harmless optimization.
 727          */
 728         if ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) {
 729                 NFSUNLOCKCLSTATE();
 730                 if (newclp != NULL)
 731                         free(newclp, M_NFSCLCLIENT);
 732                 return (EBADF);
 733         }
 734         clp = nmp->nm_clp;
 735         if (clp == NULL) {
 736                 if (newclp == NULL) {
 737                         NFSUNLOCKCLSTATE();
 738                         return (EACCES);
 739                 }
 740                 clp = newclp;
 741                 NFSBZERO((caddr_t)clp, sizeof(struct nfsclclient) + idlen - 1);
 742                 clp->nfsc_idlen = idlen;
 743                 LIST_INIT(&clp->nfsc_owner);
 744                 TAILQ_INIT(&clp->nfsc_deleg);
 745                 for (i = 0; i < NFSCLDELEGHASHSIZE; i++)
 746                         LIST_INIT(&clp->nfsc_deleghash[i]);
 747                 LIST_INIT(&clp->nfsc_defunctlockowner);
 748                 clp->nfsc_flags = NFSCLFLAGS_INITED;
 749                 clp->nfsc_clientidrev = 1;
 750                 clp->nfsc_cbident = nfscl_nextcbident();
 751                 nfscl_fillclid(nmp->nm_clval, uuid, clp->nfsc_id,
 752                     clp->nfsc_idlen);
 753                 LIST_INSERT_HEAD(&nfsclhead, clp, nfsc_list);
 754                 nmp->nm_clp = clp;
 755                 clp->nfsc_nmp = nmp;
 756                 NFSUNLOCKCLSTATE();
 757                 nfscl_start_renewthread(clp);
 758         } else {
 759                 NFSUNLOCKCLSTATE();
 760                 if (newclp != NULL)
 761                         FREE((caddr_t)newclp, M_NFSCLCLIENT);
 762         }
 763         NFSLOCKCLSTATE();
 764         while ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0 && !igotlock &&
 765             (mp->mnt_kern_flag & MNTK_UNMOUNTF) == 0)
 766                 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
 767                     NFSCLSTATEMUTEXPTR, mp);
 768         if (!igotlock)
 769                 nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
 770         if (igotlock == 0 && (mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) {
 771                 /*
 772                  * Both nfsv4_lock() and nfsv4_getref() know to check
 773                  * for MNTK_UNMOUNTF and return without sleeping to
 774                  * wait for the exclusive lock to be released, since it
 775                  * might be held by nfscl_umount() and we need to get out
 776                  * now for that case and not wait until nfscl_umount()
 777                  * releases it.
 778                  */
 779                 NFSUNLOCKCLSTATE();
 780                 return (EBADF);
 781         }
 782         NFSUNLOCKCLSTATE();
 783
 784         /*
 785          * If it needs a clientid, do the setclientid now.
 786          */
 787         if ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0) {
 788                 if (!igotlock)
 789                         panic("nfscl_clget");
 790                 if (p == NULL || cred == NULL) {
 791                         NFSLOCKCLSTATE();
 792                         nfsv4_unlock(&clp->nfsc_lock, 0);
 793                         NFSUNLOCKCLSTATE();
 794                         return (EACCES);
 795                 }
 796                 /* get rid of defunct lockowners */
 797                 LIST_FOREACH_SAFE(lp, &clp->nfsc_defunctlockowner, nfsl_list,
 798                     nlp) {
 799                         nfscl_freelockowner(lp, 0);
 800                 }
 801                 /*
 802                  * If RFC3530 Sec. 14.2.33 is taken literally,
 803                  * NFSERR_CLIDINUSE will be returned persistently for the
 804                  * case where a new mount of the same file system is using
 805                  * a different principal. In practice, NFSERR_CLIDINUSE is
 806                  * only returned when there is outstanding unexpired state
 807                  * on the clientid. As such, try for twice the lease
 808                  * interval, if we know what that is. Otherwise, make a
 809                  * wild ass guess.
 810                  * The case of returning NFSERR_STALECLIENTID is far less
 811                  * likely, but might occur if there is a significant delay
 812                  * between doing the SetClientID and SetClientIDConfirm Ops,
 813                  * such that the server throws away the clientid before
 814                  * receiving the SetClientIDConfirm.
 815                  */
 816                 if (clp->nfsc_renew > 0)
 817                         clidinusedelay = NFSCL_LEASE(clp->nfsc_renew) * 2;
 818                 else
 819                         clidinusedelay = 120;
 820                 trystalecnt = 3;
 821                 do {
 822                         error = nfsrpc_setclient(VFSTONFS(vnode_mount(vp)),
 823                             clp, cred, p);
 824                         if (error == NFSERR_STALECLIENTID ||
 825                             error == NFSERR_STALEDONTRECOVER ||
 826                             error == NFSERR_CLIDINUSE) {
 827                                 (void) nfs_catnap(PZERO, error, "nfs_setcl");
 828                         }
 829                 } while (((error == NFSERR_STALECLIENTID ||
 830                      error == NFSERR_STALEDONTRECOVER) && --trystalecnt > 0) ||
 831                     (error == NFSERR_CLIDINUSE && --clidinusedelay > 0));
 832                 if (error) {
 833                         NFSLOCKCLSTATE();
 834                         nfsv4_unlock(&clp->nfsc_lock, 0);
 835                         NFSUNLOCKCLSTATE();
 836                         return (error);
 837                 }
 838                 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
 839         }
 840         if (igotlock) {
 841                 NFSLOCKCLSTATE();
 842                 nfsv4_unlock(&clp->nfsc_lock, 1);
 843                 NFSUNLOCKCLSTATE();
 844         }
 845
 846         *clpp = clp;
 847         return (0);
 848 }
 849
 850 /*
 851  * Get a reference to a clientid and return it, if valid.
 852  */
 853 APPLESTATIC struct nfsclclient *
 854 nfscl_findcl(struct nfsmount *nmp)
 855 {
 856         struct nfsclclient *clp;
 857
 858         clp = nmp->nm_clp;
 859         if (clp == NULL || !(clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID))
 860                 return (NULL);
 861         return (clp);
 862 }
 863
 864 /*
 865  * Release the clientid structure. It may be locked or reference counted.
 866  */
 867 static void
 868 nfscl_clrelease(struct nfsclclient *clp)
 869 {
 870
 871         if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK)
 872                 nfsv4_unlock(&clp->nfsc_lock, 0);
 873         else
 874                 nfsv4_relref(&clp->nfsc_lock);
 875 }
 876
 877 /*
 878  * External call for nfscl_clrelease.
 879  */
 880 APPLESTATIC void
 881 nfscl_clientrelease(struct nfsclclient *clp)
 882 {
 883
 884         NFSLOCKCLSTATE();
 885         if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK)
 886                 nfsv4_unlock(&clp->nfsc_lock, 0);
 887         else
 888                 nfsv4_relref(&clp->nfsc_lock);
 889         NFSUNLOCKCLSTATE();
 890 }
 891
 892 /*
 893  * Called when wanting to lock a byte region.
 894  */
 895 APPLESTATIC int
 896 nfscl_getbytelock(vnode_t vp, u_int64_t off, u_int64_t len,
 897     short type, struct ucred *cred, NFSPROC_T *p, struct nfsclclient *rclp,
 898     int recovery, void *id, int flags, u_int8_t *rownp, u_int8_t *ropenownp,
 899     struct nfscllockowner **lpp, int *newonep, int *donelocallyp)
 900 {
 901         struct nfscllockowner *lp;
 902         struct nfsclopen *op;
 903         struct nfsclclient *clp;
 904         struct nfscllockowner *nlp;
 905         struct nfscllock *nlop, *otherlop;
 906         struct nfscldeleg *dp = NULL, *ldp = NULL;
 907         struct nfscllockownerhead *lhp = NULL;
 908         struct nfsnode *np;
 909         u_int8_t own[NFSV4CL_LOCKNAMELEN], *ownp, openown[NFSV4CL_LOCKNAMELEN];
 910         u_int8_t *openownp;
 911         int error = 0, ret, donelocally = 0;
 912         u_int32_t mode;
 913
 914         /* For Lock Ops, the open mode doesn't matter, so use 0 to match any. */
 915         mode = 0;
 916         np = VTONFS(vp);
 917         *lpp = NULL;
 918         lp = NULL;
 919         *newonep = 0;
 920         *donelocallyp = 0;
 921
 922         /*
 923          * Might need these, so MALLOC them now, to
 924          * avoid a tsleep() in MALLOC later.
 925          */
 926         MALLOC(nlp, struct nfscllockowner *,
 927             sizeof (struct nfscllockowner), M_NFSCLLOCKOWNER, M_WAITOK);
 928         MALLOC(otherlop, struct nfscllock *,
 929             sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
 930         MALLOC(nlop, struct nfscllock *,
 931             sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
 932         nlop->nfslo_type = type;
 933         nlop->nfslo_first = off;
 934         if (len == NFS64BITSSET) {
 935                 nlop->nfslo_end = NFS64BITSSET;
 936         } else {
 937                 nlop->nfslo_end = off + len;
 938                 if (nlop->nfslo_end <= nlop->nfslo_first)
 939                         error = NFSERR_INVAL;
 940         }
 941
 942         if (!error) {
 943                 if (recovery)
 944                         clp = rclp;
 945                 else
 946                         error = nfscl_getcl(vp, cred, p, &clp);
 947         }
 948         if (error) {
 949                 FREE((caddr_t)nlp, M_NFSCLLOCKOWNER);
 950                 FREE((caddr_t)otherlop, M_NFSCLLOCK);
 951                 FREE((caddr_t)nlop, M_NFSCLLOCK);
 952                 return (error);
 953         }
 954
 955         op = NULL;
 956         if (recovery) {
 957                 ownp = rownp;
 958                 openownp = ropenownp;
 959         } else {
 960                 nfscl_filllockowner(id, own, flags);
 961                 ownp = own;
 962                 nfscl_filllockowner(p->td_proc, openown, F_POSIX);
 963                 openownp = openown;
 964         }
 965         if (!recovery) {
 966                 NFSLOCKCLSTATE();
 967                 /*
 968                  * First, search for a delegation. If one exists for this file,
 969                  * the lock can be done locally against it, so long as there
 970                  * isn't a local lock conflict.
 971                  */
 972                 ldp = dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
 973                     np->n_fhp->nfh_len);
 974                 /* Just sanity check for correct type of delegation */
 975                 if (dp != NULL && ((dp->nfsdl_flags &
 976                     (NFSCLDL_RECALL | NFSCLDL_DELEGRET)) != 0 ||
 977                      (type == F_WRLCK &&
 978                       (dp->nfsdl_flags & NFSCLDL_WRITE) == 0)))
 979                         dp = NULL;
 980         }
 981         if (dp != NULL) {
 982                 /* Now, find an open and maybe a lockowner. */
 983                 ret = nfscl_getopen(&dp->nfsdl_owner, np->n_fhp->nfh_fh,
 984                     np->n_fhp->nfh_len, openownp, ownp, mode, NULL, &op);
 985                 if (ret)
 986                         ret = nfscl_getopen(&clp->nfsc_owner,
 987                             np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp,
 988                             ownp, mode, NULL, &op);
 989                 if (!ret) {
 990                         lhp = &dp->nfsdl_lock;
 991                         TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
 992                         TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list);
 993                         dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
 994                         donelocally = 1;
 995                 } else {
 996                         dp = NULL;
 997                 }
 998         }
 999         if (!donelocally) {
1000                 /*
1001                  * Get the related Open and maybe lockowner.
1002                  */
1003                 error = nfscl_getopen(&clp->nfsc_owner,
1004                     np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp,
1005                     ownp, mode, &lp, &op);
1006                 if (!error)
1007                         lhp = &op->nfso_lock;
1008         }
1009         if (!error && !recovery)
1010                 error = nfscl_localconflict(clp, np->n_fhp->nfh_fh,
1011                     np->n_fhp->nfh_len, nlop, ownp, ldp, NULL);
1012         if (error) {
1013                 if (!recovery) {
1014                         nfscl_clrelease(clp);
1015                         NFSUNLOCKCLSTATE();
1016                 }
1017                 FREE((caddr_t)nlp, M_NFSCLLOCKOWNER);
1018                 FREE((caddr_t)otherlop, M_NFSCLLOCK);
1019                 FREE((caddr_t)nlop, M_NFSCLLOCK);
1020                 return (error);
1021         }
1022
1023         /*
1024          * Ok, see if a lockowner exists and create one, as required.
1025          */
1026         if (lp == NULL)
1027                 LIST_FOREACH(lp, lhp, nfsl_list) {
1028                         if (!NFSBCMP(lp->nfsl_owner, ownp, NFSV4CL_LOCKNAMELEN))
1029                                 break;
1030                 }
1031         if (lp == NULL) {
1032                 NFSBCOPY(ownp, nlp->nfsl_owner, NFSV4CL_LOCKNAMELEN);
1033                 if (recovery)
1034                         NFSBCOPY(ropenownp, nlp->nfsl_openowner,
1035                             NFSV4CL_LOCKNAMELEN);
1036                 else
1037                         NFSBCOPY(op->nfso_own->nfsow_owner, nlp->nfsl_openowner,
1038                             NFSV4CL_LOCKNAMELEN);
1039                 nlp->nfsl_seqid = 0;
1040                 nlp->nfsl_defunct = 0;
1041                 nlp->nfsl_inprog = NULL;
1042                 nfscl_lockinit(&nlp->nfsl_rwlock);
1043                 LIST_INIT(&nlp->nfsl_lock);
1044                 if (donelocally) {
1045                         nlp->nfsl_open = NULL;
1046                         newnfsstats.cllocallockowners++;
1047                 } else {
1048                         nlp->nfsl_open = op;
1049                         newnfsstats.cllockowners++;
1050                 }
1051                 LIST_INSERT_HEAD(lhp, nlp, nfsl_list);
1052                 lp = nlp;
1053                 nlp = NULL;
1054                 *newonep = 1;
1055         }
1056
1057         /*
1058          * Now, update the byte ranges for locks.
1059          */
1060         ret = nfscl_updatelock(lp, &nlop, &otherlop, donelocally);
1061         if (!ret)
1062                 donelocally = 1;
1063         if (donelocally) {
1064                 *donelocallyp = 1;
1065                 if (!recovery)
1066                         nfscl_clrelease(clp);
1067         } else {
1068                 /*
1069                  * Serial modifications on the lock owner for multiple threads
1070                  * for the same process using a read/write lock.
1071                  */
1072                 if (!recovery)
1073                         nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR);
1074         }
1075         if (!recovery)
1076                 NFSUNLOCKCLSTATE();
1077
1078         if (nlp)
1079                 FREE((caddr_t)nlp, M_NFSCLLOCKOWNER);
1080         if (nlop)
1081                 FREE((caddr_t)nlop, M_NFSCLLOCK);
1082         if (otherlop)
1083                 FREE((caddr_t)otherlop, M_NFSCLLOCK);
1084
1085         *lpp = lp;
1086         return (0);
1087 }
1088
1089 /*
1090  * Called to unlock a byte range, for LockU.
1091  */
1092 APPLESTATIC int
1093 nfscl_relbytelock(vnode_t vp, u_int64_t off, u_int64_t len,
1094     __unused struct ucred *cred, NFSPROC_T *p, int callcnt,
1095     struct nfsclclient *clp, void *id, int flags,
1096     struct nfscllockowner **lpp, int *dorpcp)
1097 {
1098         struct nfscllockowner *lp;
1099         struct nfsclowner *owp;
1100         struct nfsclopen *op;
1101         struct nfscllock *nlop, *other_lop = NULL;
1102         struct nfscldeleg *dp;
1103         struct nfsnode *np;
1104         u_int8_t own[NFSV4CL_LOCKNAMELEN];
1105         int ret = 0, fnd;
1106
1107         np = VTONFS(vp);
1108         *lpp = NULL;
1109         *dorpcp = 0;
1110
1111         /*
1112          * Might need these, so MALLOC them now, to
1113          * avoid a tsleep() in MALLOC later.
1114          */
1115         MALLOC(nlop, struct nfscllock *,
1116             sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1117         nlop->nfslo_type = F_UNLCK;
1118         nlop->nfslo_first = off;
1119         if (len == NFS64BITSSET) {
1120                 nlop->nfslo_end = NFS64BITSSET;
1121         } else {
1122                 nlop->nfslo_end = off + len;
1123                 if (nlop->nfslo_end <= nlop->nfslo_first) {
1124                         FREE((caddr_t)nlop, M_NFSCLLOCK);
1125                         return (NFSERR_INVAL);
1126                 }
1127         }
1128         if (callcnt == 0) {
1129                 MALLOC(other_lop, struct nfscllock *,
1130                     sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1131                 *other_lop = *nlop;
1132         }
1133         nfscl_filllockowner(id, own, flags);
1134         dp = NULL;
1135         NFSLOCKCLSTATE();
1136         if (callcnt == 0)
1137                 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
1138                     np->n_fhp->nfh_len);
1139
1140         /*
1141          * First, unlock any local regions on a delegation.
1142          */
1143         if (dp != NULL) {
1144                 /* Look for this lockowner. */
1145                 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
1146                         if (!NFSBCMP(lp->nfsl_owner, own,
1147                             NFSV4CL_LOCKNAMELEN))
1148                                 break;
1149                 }
1150                 if (lp != NULL)
1151                         /* Use other_lop, so nlop is still available */
1152                         (void)nfscl_updatelock(lp, &other_lop, NULL, 1);
1153         }
1154
1155         /*
1156          * Now, find a matching open/lockowner that hasn't already been done,
1157          * as marked by nfsl_inprog.
1158          */
1159         lp = NULL;
1160         fnd = 0;
1161         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
1162             LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1163                 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1164                     !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1165                     LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1166                         if (lp->nfsl_inprog == NULL &&
1167                             !NFSBCMP(lp->nfsl_owner, own,
1168                              NFSV4CL_LOCKNAMELEN)) {
1169                                 fnd = 1;
1170                                 break;
1171                         }
1172                     }
1173                     if (fnd)
1174                         break;
1175                 }
1176             }
1177             if (fnd)
1178                 break;
1179         }
1180
1181         if (lp != NULL) {
1182                 ret = nfscl_updatelock(lp, &nlop, NULL, 0);
1183                 if (ret)
1184                         *dorpcp = 1;
1185                 /*
1186                  * Serial modifications on the lock owner for multiple
1187                  * threads for the same process using a read/write lock.
1188                  */
1189                 lp->nfsl_inprog = p;
1190                 nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR);
1191                 *lpp = lp;
1192         }
1193         NFSUNLOCKCLSTATE();
1194         if (nlop)
1195                 FREE((caddr_t)nlop, M_NFSCLLOCK);
1196         if (other_lop)
1197                 FREE((caddr_t)other_lop, M_NFSCLLOCK);
1198         return (0);
1199 }
1200
1201 /*
1202  * Release all lockowners marked in progess for this process and file.
1203  */
1204 APPLESTATIC void
1205 nfscl_releasealllocks(struct nfsclclient *clp, vnode_t vp, NFSPROC_T *p,
1206     void *id, int flags)
1207 {
1208         struct nfsclowner *owp;
1209         struct nfsclopen *op;
1210         struct nfscllockowner *lp;
1211         struct nfsnode *np;
1212         u_int8_t own[NFSV4CL_LOCKNAMELEN];
1213
1214         np = VTONFS(vp);
1215         nfscl_filllockowner(id, own, flags);
1216         NFSLOCKCLSTATE();
1217         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
1218             LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1219                 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1220                     !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1221                     LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1222                         if (lp->nfsl_inprog == p &&
1223                             !NFSBCMP(lp->nfsl_owner, own,
1224                             NFSV4CL_LOCKNAMELEN)) {
1225                             lp->nfsl_inprog = NULL;
1226                             nfscl_lockunlock(&lp->nfsl_rwlock);
1227                         }
1228                     }
1229                 }
1230             }
1231         }
1232         nfscl_clrelease(clp);
1233         NFSUNLOCKCLSTATE();
1234 }
1235
1236 /*
1237  * Called to find out if any bytes within the byte range specified are
1238  * write locked by the calling process. Used to determine if flushing
1239  * is required before a LockU.
1240  * If in doubt, return 1, so the flush will occur.
1241  */
1242 APPLESTATIC int
1243 nfscl_checkwritelocked(vnode_t vp, struct flock *fl,
1244     struct ucred *cred, NFSPROC_T *p, void *id, int flags)
1245 {
1246         struct nfsclowner *owp;
1247         struct nfscllockowner *lp;
1248         struct nfsclopen *op;
1249         struct nfsclclient *clp;
1250         struct nfscllock *lop;
1251         struct nfscldeleg *dp;
1252         struct nfsnode *np;
1253         u_int64_t off, end;
1254         u_int8_t own[NFSV4CL_LOCKNAMELEN];
1255         int error = 0;
1256
1257         np = VTONFS(vp);
1258         switch (fl->l_whence) {
1259         case SEEK_SET:
1260         case SEEK_CUR:
1261                 /*
1262                  * Caller is responsible for adding any necessary offset
1263                  * when SEEK_CUR is used.
1264                  */
1265                 off = fl->l_start;
1266                 break;
1267         case SEEK_END:
1268                 off = np->n_size + fl->l_start;
1269                 break;
1270         default:
1271                 return (1);
1272         };
1273         if (fl->l_len != 0) {
1274                 end = off + fl->l_len;
1275                 if (end < off)
1276                         return (1);
1277         } else {
1278                 end = NFS64BITSSET;
1279         }
1280
1281         error = nfscl_getcl(vp, cred, p, &clp);
1282         if (error)
1283                 return (1);
1284         nfscl_filllockowner(id, own, flags);
1285         NFSLOCKCLSTATE();
1286
1287         /*
1288          * First check the delegation locks.
1289          */
1290         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
1291         if (dp != NULL) {
1292                 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
1293                         if (!NFSBCMP(lp->nfsl_owner, own,
1294                             NFSV4CL_LOCKNAMELEN))
1295                                 break;
1296                 }
1297                 if (lp != NULL) {
1298                         LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
1299                                 if (lop->nfslo_first >= end)
1300                                         break;
1301                                 if (lop->nfslo_end <= off)
1302                                         continue;
1303                                 if (lop->nfslo_type == F_WRLCK) {
1304                                         nfscl_clrelease(clp);
1305                                         NFSUNLOCKCLSTATE();
1306                                         return (1);
1307                                 }
1308                         }
1309                 }
1310         }
1311
1312         /*
1313          * Now, check state against the server.
1314          */
1315         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
1316             LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1317                 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1318                     !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1319                     LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1320                         if (!NFSBCMP(lp->nfsl_owner, own,
1321                             NFSV4CL_LOCKNAMELEN))
1322                             break;
1323                     }
1324                     if (lp != NULL) {
1325                         LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
1326                             if (lop->nfslo_first >= end)
1327                                 break;
1328                             if (lop->nfslo_end <= off)
1329                                 continue;
1330                             if (lop->nfslo_type == F_WRLCK) {
1331                                 nfscl_clrelease(clp);
1332                                 NFSUNLOCKCLSTATE();
1333                                 return (1);
1334                             }
1335                         }
1336                     }
1337                 }
1338             }
1339         }
1340         nfscl_clrelease(clp);
1341         NFSUNLOCKCLSTATE();
1342         return (0);
1343 }
1344
1345 /*
1346  * Release a byte range lock owner structure.
1347  */
1348 APPLESTATIC void
1349 nfscl_lockrelease(struct nfscllockowner *lp, int error, int candelete)
1350 {
1351         struct nfsclclient *clp;
1352
1353         if (lp == NULL)
1354                 return;
1355         NFSLOCKCLSTATE();
1356         clp = lp->nfsl_open->nfso_own->nfsow_clp;
1357         if (error != 0 && candelete &&
1358             (lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED) == 0)
1359                 nfscl_freelockowner(lp, 0);
1360         else
1361                 nfscl_lockunlock(&lp->nfsl_rwlock);
1362         nfscl_clrelease(clp);
1363         NFSUNLOCKCLSTATE();
1364 }
1365
1366 /*
1367  * Free up an open structure and any associated byte range lock structures.
1368  */
1369 APPLESTATIC void
1370 nfscl_freeopen(struct nfsclopen *op, int local)
1371 {
1372
1373         LIST_REMOVE(op, nfso_list);
1374         nfscl_freealllocks(&op->nfso_lock, local);
1375         FREE((caddr_t)op, M_NFSCLOPEN);
1376         if (local)
1377                 newnfsstats.cllocalopens--;
1378         else
1379                 newnfsstats.clopens--;
1380 }
1381
1382 /*
1383  * Free up all lock owners and associated locks.
1384  */
1385 static void
1386 nfscl_freealllocks(struct nfscllockownerhead *lhp, int local)
1387 {
1388         struct nfscllockowner *lp, *nlp;
1389
1390         LIST_FOREACH_SAFE(lp, lhp, nfsl_list, nlp) {
1391                 if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED))
1392                         panic("nfscllckw");
1393                 nfscl_freelockowner(lp, local);
1394         }
1395 }
1396
1397 /*
1398  * Called for an Open when NFSERR_EXPIRED is received from the server.
1399  * If there are no byte range locks nor a Share Deny lost, try to do a
1400  * fresh Open. Otherwise, free the open.
1401  */
1402 static int
1403 nfscl_expireopen(struct nfsclclient *clp, struct nfsclopen *op,
1404     struct nfsmount *nmp, struct ucred *cred, NFSPROC_T *p)
1405 {
1406         struct nfscllockowner *lp;
1407         struct nfscldeleg *dp;
1408         int mustdelete = 0, error;
1409
1410         /*
1411          * Look for any byte range lock(s).
1412          */
1413         LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1414                 if (!LIST_EMPTY(&lp->nfsl_lock)) {
1415                         mustdelete = 1;
1416                         break;
1417                 }
1418         }
1419
1420         /*
1421          * If no byte range lock(s) nor a Share deny, try to re-open.
1422          */
1423         if (!mustdelete && (op->nfso_mode & NFSLCK_DENYBITS) == 0) {
1424                 newnfs_copycred(&op->nfso_cred, cred);
1425                 dp = NULL;
1426                 error = nfsrpc_reopen(nmp, op->nfso_fh,
1427                     op->nfso_fhlen, op->nfso_mode, op, &dp, cred, p);
1428                 if (error) {
1429                         mustdelete = 1;
1430                         if (dp != NULL) {
1431                                 FREE((caddr_t)dp, M_NFSCLDELEG);
1432                                 dp = NULL;
1433                         }
1434                 }
1435                 if (dp != NULL)
1436                         nfscl_deleg(nmp->nm_mountp, clp, op->nfso_fh,
1437                             op->nfso_fhlen, cred, p, &dp);
1438         }
1439
1440         /*
1441          * If a byte range lock or Share deny or couldn't re-open, free it.
1442          */
1443         if (mustdelete)
1444                 nfscl_freeopen(op, 0);
1445         return (mustdelete);
1446 }
1447
1448 /*
1449  * Free up an open owner structure.
1450  */
1451 static void
1452 nfscl_freeopenowner(struct nfsclowner *owp, int local)
1453 {
1454
1455         LIST_REMOVE(owp, nfsow_list);
1456         FREE((caddr_t)owp, M_NFSCLOWNER);
1457         if (local)
1458                 newnfsstats.cllocalopenowners--;
1459         else
1460                 newnfsstats.clopenowners--;
1461 }
1462
1463 /*
1464  * Free up a byte range lock owner structure.
1465  */
1466 APPLESTATIC void
1467 nfscl_freelockowner(struct nfscllockowner *lp, int local)
1468 {
1469         struct nfscllock *lop, *nlop;
1470
1471         LIST_REMOVE(lp, nfsl_list);
1472         LIST_FOREACH_SAFE(lop, &lp->nfsl_lock, nfslo_list, nlop) {
1473                 nfscl_freelock(lop, local);
1474         }
1475         FREE((caddr_t)lp, M_NFSCLLOCKOWNER);
1476         if (local)
1477                 newnfsstats.cllocallockowners--;
1478         else
1479                 newnfsstats.cllockowners--;
1480 }
1481
1482 /*
1483  * Free up a byte range lock structure.
1484  */
1485 APPLESTATIC void
1486 nfscl_freelock(struct nfscllock *lop, int local)
1487 {
1488
1489         LIST_REMOVE(lop, nfslo_list);
1490         FREE((caddr_t)lop, M_NFSCLLOCK);
1491         if (local)
1492                 newnfsstats.cllocallocks--;
1493         else
1494                 newnfsstats.cllocks--;
1495 }
1496
1497 /*
1498  * Clean out the state related to a delegation.
1499  */
1500 static void
1501 nfscl_cleandeleg(struct nfscldeleg *dp)
1502 {
1503         struct nfsclowner *owp, *nowp;
1504         struct nfsclopen *op;
1505
1506         LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) {
1507                 op = LIST_FIRST(&owp->nfsow_open);
1508                 if (op != NULL) {
1509                         if (LIST_NEXT(op, nfso_list) != NULL)
1510                                 panic("nfscleandel");
1511                         nfscl_freeopen(op, 1);
1512                 }
1513                 nfscl_freeopenowner(owp, 1);
1514         }
1515         nfscl_freealllocks(&dp->nfsdl_lock, 1);
1516 }
1517
1518 /*
1519  * Free a delegation.
1520  */
1521 static void
1522 nfscl_freedeleg(struct nfscldeleghead *hdp, struct nfscldeleg *dp)
1523 {
1524
1525         TAILQ_REMOVE(hdp, dp, nfsdl_list);
1526         LIST_REMOVE(dp, nfsdl_hash);
1527         FREE((caddr_t)dp, M_NFSCLDELEG);
1528         newnfsstats.cldelegates--;
1529         nfscl_delegcnt--;
1530 }
1531
1532 /*
1533  * Free up all state related to this client structure.
1534  */
1535 static void
1536 nfscl_cleanclient(struct nfsclclient *clp)
1537 {
1538         struct nfsclowner *owp, *nowp;
1539         struct nfsclopen *op, *nop;
1540         struct nfscllockowner *lp, *nlp;
1541
1542
1543         /* get rid of defunct lockowners */
1544         LIST_FOREACH_SAFE(lp, &clp->nfsc_defunctlockowner, nfsl_list, nlp) {
1545                 nfscl_freelockowner(lp, 0);
1546         }
1547
1548         /* Now, all the OpenOwners, etc. */
1549         LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1550                 LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
1551                         nfscl_freeopen(op, 0);
1552                 }
1553                 nfscl_freeopenowner(owp, 0);
1554         }
1555 }
1556
1557 /*
1558  * Called when an NFSERR_EXPIRED is received from the server.
1559  */
1560 static void
1561 nfscl_expireclient(struct nfsclclient *clp, struct nfsmount *nmp,
1562     struct ucred *cred, NFSPROC_T *p)
1563 {
1564         struct nfsclowner *owp, *nowp, *towp;
1565         struct nfsclopen *op, *nop, *top;
1566         struct nfscldeleg *dp, *ndp;
1567         int ret, printed = 0;
1568
1569         /*
1570          * First, merge locally issued Opens into the list for the server.
1571          */
1572         dp = TAILQ_FIRST(&clp->nfsc_deleg);
1573         while (dp != NULL) {
1574             ndp = TAILQ_NEXT(dp, nfsdl_list);
1575             owp = LIST_FIRST(&dp->nfsdl_owner);
1576             while (owp != NULL) {
1577                 nowp = LIST_NEXT(owp, nfsow_list);
1578                 op = LIST_FIRST(&owp->nfsow_open);
1579                 if (op != NULL) {
1580                     if (LIST_NEXT(op, nfso_list) != NULL)
1581                         panic("nfsclexp");
1582                     LIST_FOREACH(towp, &clp->nfsc_owner, nfsow_list) {
1583                         if (!NFSBCMP(towp->nfsow_owner, owp->nfsow_owner,
1584                             NFSV4CL_LOCKNAMELEN))
1585                             break;
1586                     }
1587                     if (towp != NULL) {
1588                         /* Merge opens in */
1589                         LIST_FOREACH(top, &towp->nfsow_open, nfso_list) {
1590                             if (top->nfso_fhlen == op->nfso_fhlen &&
1591                                 !NFSBCMP(top->nfso_fh, op->nfso_fh,
1592                                  op->nfso_fhlen)) {
1593                                 top->nfso_mode |= op->nfso_mode;
1594                                 top->nfso_opencnt += op->nfso_opencnt;
1595                                 break;
1596                             }
1597                         }
1598                         if (top == NULL) {
1599                             /* Just add the open to the owner list */
1600                             LIST_REMOVE(op, nfso_list);
1601                             op->nfso_own = towp;
1602                             LIST_INSERT_HEAD(&towp->nfsow_open, op, nfso_list);
1603                             newnfsstats.cllocalopens--;
1604                             newnfsstats.clopens++;
1605                         }
1606                     } else {
1607                         /* Just add the openowner to the client list */
1608                         LIST_REMOVE(owp, nfsow_list);
1609                         owp->nfsow_clp = clp;
1610                         LIST_INSERT_HEAD(&clp->nfsc_owner, owp, nfsow_list);
1611                         newnfsstats.cllocalopenowners--;
1612                         newnfsstats.clopenowners++;
1613                         newnfsstats.cllocalopens--;
1614                         newnfsstats.clopens++;
1615                     }
1616                 }
1617                 owp = nowp;
1618             }
1619             if (!printed && !LIST_EMPTY(&dp->nfsdl_lock)) {
1620                 printed = 1;
1621                 printf("nfsv4 expired locks lost\n");
1622             }
1623             nfscl_cleandeleg(dp);
1624             nfscl_freedeleg(&clp->nfsc_deleg, dp);
1625             dp = ndp;
1626         }
1627         if (!TAILQ_EMPTY(&clp->nfsc_deleg))
1628             panic("nfsclexp");
1629
1630         /*
1631          * Now, try and reopen against the server.
1632          */
1633         LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1634                 owp->nfsow_seqid = 0;
1635                 LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
1636                         ret = nfscl_expireopen(clp, op, nmp, cred, p);
1637                         if (ret && !printed) {
1638                                 printed = 1;
1639                                 printf("nfsv4 expired locks lost\n");
1640                         }
1641                 }
1642                 if (LIST_EMPTY(&owp->nfsow_open))
1643                         nfscl_freeopenowner(owp, 0);
1644         }
1645 }
1646
1647 #ifndef __FreeBSD__
1648 /*
1649  * Called from exit() upon process termination.
1650  */
1651 APPLESTATIC void
1652 nfscl_cleanup(NFSPROC_T *p)
1653 {
1654         struct nfsclclient *clp;
1655         u_int8_t own[NFSV4CL_LOCKNAMELEN];
1656
1657         if (!nfscl_inited)
1658                 return;
1659         nfscl_filllockowner(p->td_proc, own, F_POSIX);
1660
1661         NFSLOCKCLSTATE();
1662         /*
1663          * Loop through all the clientids, looking for the OpenOwners.
1664          */
1665         LIST_FOREACH(clp, &nfsclhead, nfsc_list)
1666                 nfscl_cleanup_common(clp, own);
1667         NFSUNLOCKCLSTATE();
1668 }
1669 #endif  /* !__FreeBSD__ */
1670
1671 /*
1672  * Common code used by nfscl_cleanup() and nfscl_cleanupkext().
1673  * Must be called with CLSTATE lock held.
1674  */
1675 static void
1676 nfscl_cleanup_common(struct nfsclclient *clp, u_int8_t *own)
1677 {
1678         struct nfsclowner *owp, *nowp;
1679         struct nfsclopen *op;
1680         struct nfscllockowner *lp, *nlp;
1681         struct nfscldeleg *dp;
1682
1683         /* First, get rid of local locks on delegations. */
1684         TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
1685                 LIST_FOREACH_SAFE(lp, &dp->nfsdl_lock, nfsl_list, nlp) {
1686                     if (!NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
1687                         if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED))
1688                             panic("nfscllckw");
1689                         nfscl_freelockowner(lp, 1);
1690                     }
1691                 }
1692         }
1693         owp = LIST_FIRST(&clp->nfsc_owner);
1694         while (owp != NULL) {
1695                 nowp = LIST_NEXT(owp, nfsow_list);
1696                 if (!NFSBCMP(owp->nfsow_owner, own,
1697                     NFSV4CL_LOCKNAMELEN)) {
1698                         /*
1699                          * If there are children that haven't closed the
1700                          * file descriptors yet, the opens will still be
1701                          * here. For that case, let the renew thread clear
1702                          * out the OpenOwner later.
1703                          */
1704                         if (LIST_EMPTY(&owp->nfsow_open))
1705                                 nfscl_freeopenowner(owp, 0);
1706                         else
1707                                 owp->nfsow_defunct = 1;
1708                 } else {
1709                         /* look for lockowners on other opens */
1710                         LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1711                                 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1712                                         if (!NFSBCMP(lp->nfsl_owner, own,
1713                                             NFSV4CL_LOCKNAMELEN))
1714                                                 lp->nfsl_defunct = 1;
1715                                 }
1716                         }
1717                 }
1718                 owp = nowp;
1719         }
1720
1721         /* and check the defunct list */
1722         LIST_FOREACH(lp, &clp->nfsc_defunctlockowner, nfsl_list) {
1723                 if (!NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN))
1724                     lp->nfsl_defunct = 1;
1725         }
1726 }
1727
1728 #if defined(APPLEKEXT) || defined(__FreeBSD__)
1729 /*
1730  * Simulate the call nfscl_cleanup() by looking for open owners associated
1731  * with processes that no longer exist, since a call to nfscl_cleanup()
1732  * can't be patched into exit().
1733  */
1734 static void
1735 nfscl_cleanupkext(struct nfsclclient *clp)
1736 {
1737         struct nfsclowner *owp, *nowp;
1738         struct nfscllockowner *lp;
1739
1740         NFSPROCLISTLOCK();
1741         NFSLOCKCLSTATE();
1742         LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1743                 if (nfscl_procdoesntexist(owp->nfsow_owner))
1744                         nfscl_cleanup_common(clp, owp->nfsow_owner);
1745         }
1746
1747         /* and check the defunct list */
1748         LIST_FOREACH(lp, &clp->nfsc_defunctlockowner, nfsl_list) {
1749                 if (nfscl_procdoesntexist(lp->nfsl_owner))
1750                         lp->nfsl_defunct = 1;
1751         }
1752         NFSUNLOCKCLSTATE();
1753         NFSPROCLISTUNLOCK();
1754 }
1755 #endif  /* APPLEKEXT || __FreeBSD__ */
1756
1757 static int      fake_global;    /* Used to force visibility of MNTK_UNMOUNTF */
1758 /*
1759  * Called from nfs umount to free up the clientid.
1760  */
1761 APPLESTATIC void
1762 nfscl_umount(struct nfsmount *nmp, NFSPROC_T *p)
1763 {
1764         struct nfsclclient *clp;
1765         struct ucred *cred;
1766         int igotlock;
1767
1768         /*
1769          * For the case that matters, this is the thread that set
1770          * MNTK_UNMOUNTF, so it will see it set. The code that follows is
1771          * done to ensure that any thread executing nfscl_getcl() after
1772          * this time, will see MNTK_UNMOUNTF set. nfscl_getcl() uses the
1773          * mutex for NFSLOCKCLSTATE(), so it is "m" for the following
1774          * explanation, courtesy of Alan Cox.
1775          * What follows is a snippet from Alan Cox's email at:
1776          * http://docs.FreeBSD.org/cgi/
1777          *     mid.cgi?BANLkTikR3d65zPHo9==08ZfJ2vmqZucEvw
1778          *
1779          * 1. Set MNTK_UNMOUNTF
1780          * 2. Acquire a standard FreeBSD mutex "m".
1781          * 3. Update some data structures.
1782          * 4. Release mutex "m".
1783          *
1784          * Then, other threads that acquire "m" after step 4 has occurred will
1785          * see MNTK_UNMOUNTF as set.  But, other threads that beat thread X to
1786          * step 2 may or may not see MNTK_UNMOUNTF as set.
1787          */
1788         NFSLOCKCLSTATE();
1789         if ((nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) {
1790                 fake_global++;
1791                 NFSUNLOCKCLSTATE();
1792                 NFSLOCKCLSTATE();
1793         }
1794
1795         clp = nmp->nm_clp;
1796         if (clp != NULL) {
1797                 if ((clp->nfsc_flags & NFSCLFLAGS_INITED) == 0)
1798                         panic("nfscl umount");
1799
1800                 /*
1801                  * First, handshake with the nfscl renew thread, to terminate
1802                  * it.
1803                  */
1804                 clp->nfsc_flags |= NFSCLFLAGS_UMOUNT;
1805                 while (clp->nfsc_flags & NFSCLFLAGS_HASTHREAD)
1806                         (void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT,
1807                             "nfsclumnt", hz);
1808
1809                 /*
1810                  * Now, get the exclusive lock on the client state, so
1811                  * that no uses of the state are still in progress.
1812                  */
1813                 do {
1814                         igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
1815                             NFSCLSTATEMUTEXPTR, NULL);
1816                 } while (!igotlock);
1817                 NFSUNLOCKCLSTATE();
1818
1819                 /*
1820                  * Free up all the state. It will expire on the server, but
1821                  * maybe we should do a SetClientId/SetClientIdConfirm so
1822                  * the server throws it away?
1823                  */
1824                 LIST_REMOVE(clp, nfsc_list);
1825                 nfscl_delegreturnall(clp, p);
1826                 cred = newnfs_getcred();
1827                 (void) nfsrpc_setclient(nmp, clp, cred, p);
1828                 nfscl_cleanclient(clp);
1829                 nmp->nm_clp = NULL;
1830                 NFSFREECRED(cred);
1831                 FREE((caddr_t)clp, M_NFSCLCLIENT);
1832         } else
1833                 NFSUNLOCKCLSTATE();
1834 }
1835
1836 /*
1837  * This function is called when a server replies with NFSERR_STALECLIENTID
1838  * or NFSERR_STALESTATEID. It traverses the clientid lists, doing Opens
1839  * and Locks with reclaim. If these fail, it deletes the corresponding state.
1840  */
1841 static void
1842 nfscl_recover(struct nfsclclient *clp, struct ucred *cred, NFSPROC_T *p)
1843 {
1844         struct nfsclowner *owp, *nowp;
1845         struct nfsclopen *op, *nop;
1846         struct nfscllockowner *lp, *nlp;
1847         struct nfscllock *lop, *nlop;
1848         struct nfscldeleg *dp, *ndp, *tdp;
1849         struct nfsmount *nmp;
1850         struct ucred *tcred;
1851         struct nfsclopenhead extra_open;
1852         struct nfscldeleghead extra_deleg;
1853         struct nfsreq *rep;
1854         u_int64_t len;
1855         u_int32_t delegtype = NFSV4OPEN_DELEGATEWRITE, mode;
1856         int igotlock = 0, error, trycnt, firstlock, s;
1857
1858         /*
1859          * First, lock the client structure, so everyone else will
1860          * block when trying to use state.
1861          */
1862         NFSLOCKCLSTATE();
1863         clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG;
1864         do {
1865                 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
1866                     NFSCLSTATEMUTEXPTR, NULL);
1867         } while (!igotlock);
1868         NFSUNLOCKCLSTATE();
1869
1870         nmp = clp->nfsc_nmp;
1871         if (nmp == NULL)
1872                 panic("nfscl recover");
1873         trycnt = 5;
1874         do {
1875                 error = nfsrpc_setclient(nmp, clp, cred, p);
1876         } while ((error == NFSERR_STALECLIENTID ||
1877              error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
1878         if (error) {
1879                 nfscl_cleanclient(clp);
1880                 NFSLOCKCLSTATE();
1881                 clp->nfsc_flags &= ~(NFSCLFLAGS_HASCLIENTID |
1882                     NFSCLFLAGS_RECOVER | NFSCLFLAGS_RECVRINPROG);
1883                 wakeup(&clp->nfsc_flags);
1884                 nfsv4_unlock(&clp->nfsc_lock, 0);
1885                 NFSUNLOCKCLSTATE();
1886                 return;
1887         }
1888         clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
1889         clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
1890
1891         /*
1892          * Mark requests already queued on the server, so that they don't
1893          * initiate another recovery cycle. Any requests already in the
1894          * queue that handle state information will have the old stale
1895          * clientid/stateid and will get a NFSERR_STALESTATEID or
1896          * NFSERR_STALECLIENTID reply from the server. This will be
1897          * translated to NFSERR_STALEDONTRECOVER when R_DONTRECOVER is set.
1898          */
1899         s = splsoftclock();
1900         NFSLOCKREQ();
1901         TAILQ_FOREACH(rep, &nfsd_reqq, r_chain) {
1902                 if (rep->r_nmp == nmp)
1903                         rep->r_flags |= R_DONTRECOVER;
1904         }
1905         NFSUNLOCKREQ();
1906         splx(s);
1907
1908         /* get rid of defunct lockowners */
1909         LIST_FOREACH_SAFE(lp, &clp->nfsc_defunctlockowner, nfsl_list, nlp) {
1910                 nfscl_freelockowner(lp, 0);
1911         }
1912
1913         /*
1914          * Now, mark all delegations "need reclaim".
1915          */
1916         TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list)
1917                 dp->nfsdl_flags |= NFSCLDL_NEEDRECLAIM;
1918
1919         TAILQ_INIT(&extra_deleg);
1920         LIST_INIT(&extra_open);
1921         /*
1922          * Now traverse the state lists, doing Open and Lock Reclaims.
1923          */
1924         tcred = newnfs_getcred();
1925         owp = LIST_FIRST(&clp->nfsc_owner);
1926         while (owp != NULL) {
1927             nowp = LIST_NEXT(owp, nfsow_list);
1928             owp->nfsow_seqid = 0;
1929             op = LIST_FIRST(&owp->nfsow_open);
1930             while (op != NULL) {
1931                 nop = LIST_NEXT(op, nfso_list);
1932                 if (error != NFSERR_NOGRACE) {
1933                     /* Search for a delegation to reclaim with the open */
1934                     TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
1935                         if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
1936                             continue;
1937                         if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
1938                             mode = NFSV4OPEN_ACCESSWRITE;
1939                             delegtype = NFSV4OPEN_DELEGATEWRITE;
1940                         } else {
1941                             mode = NFSV4OPEN_ACCESSREAD;
1942                             delegtype = NFSV4OPEN_DELEGATEREAD;
1943                         }
1944                         if ((op->nfso_mode & mode) == mode &&
1945                             op->nfso_fhlen == dp->nfsdl_fhlen &&
1946                             !NFSBCMP(op->nfso_fh, dp->nfsdl_fh, op->nfso_fhlen))
1947                             break;
1948                     }
1949                     ndp = dp;
1950                     if (dp == NULL)
1951                         delegtype = NFSV4OPEN_DELEGATENONE;
1952                     newnfs_copycred(&op->nfso_cred, tcred);
1953                     error = nfscl_tryopen(nmp, NULL, op->nfso_fh,
1954                         op->nfso_fhlen, op->nfso_fh, op->nfso_fhlen,
1955                         op->nfso_mode, op, NULL, 0, &ndp, 1, delegtype,
1956                         tcred, p);
1957                     if (!error) {
1958                         /* Handle any replied delegation */
1959                         if (ndp != NULL && ((ndp->nfsdl_flags & NFSCLDL_WRITE)
1960                             || NFSMNT_RDONLY(nmp->nm_mountp))) {
1961                             if ((ndp->nfsdl_flags & NFSCLDL_WRITE))
1962                                 mode = NFSV4OPEN_ACCESSWRITE;
1963                             else
1964                                 mode = NFSV4OPEN_ACCESSREAD;
1965                             TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
1966                                 if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
1967                                     continue;
1968                                 if ((op->nfso_mode & mode) == mode &&
1969                                     op->nfso_fhlen == dp->nfsdl_fhlen &&
1970                                     !NFSBCMP(op->nfso_fh, dp->nfsdl_fh,
1971                                     op->nfso_fhlen)) {
1972                                     dp->nfsdl_stateid = ndp->nfsdl_stateid;
1973                                     dp->nfsdl_sizelimit = ndp->nfsdl_sizelimit;
1974                                     dp->nfsdl_ace = ndp->nfsdl_ace;
1975                                     dp->nfsdl_change = ndp->nfsdl_change;
1976                                     dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
1977                                     if ((ndp->nfsdl_flags & NFSCLDL_RECALL))
1978                                         dp->nfsdl_flags |= NFSCLDL_RECALL;
1979                                     FREE((caddr_t)ndp, M_NFSCLDELEG);
1980                                     ndp = NULL;
1981                                     break;
1982                                 }
1983                             }
1984                         }
1985                         if (ndp != NULL)
1986                             TAILQ_INSERT_HEAD(&extra_deleg, ndp, nfsdl_list);
1987
1988                         /* and reclaim all byte range locks */
1989                         lp = LIST_FIRST(&op->nfso_lock);
1990                         while (lp != NULL) {
1991                             nlp = LIST_NEXT(lp, nfsl_list);
1992                             lp->nfsl_seqid = 0;
1993                             firstlock = 1;
1994                             lop = LIST_FIRST(&lp->nfsl_lock);
1995                             while (lop != NULL) {
1996                                 nlop = LIST_NEXT(lop, nfslo_list);
1997                                 if (lop->nfslo_end == NFS64BITSSET)
1998                                     len = NFS64BITSSET;
1999                                 else
2000                                     len = lop->nfslo_end - lop->nfslo_first;
2001                                 if (error != NFSERR_NOGRACE)
2002                                     error = nfscl_trylock(nmp, NULL,
2003                                         op->nfso_fh, op->nfso_fhlen, lp,
2004                                         firstlock, 1, lop->nfslo_first, len,
2005                                         lop->nfslo_type, tcred, p);
2006                                 if (error != 0)
2007                                     nfscl_freelock(lop, 0);
2008                                 else
2009                                     firstlock = 0;
2010                                 lop = nlop;
2011                             }
2012                             /* If no locks, but a lockowner, just delete it. */
2013                             if (LIST_EMPTY(&lp->nfsl_lock))
2014                                 nfscl_freelockowner(lp, 0);
2015                             lp = nlp;
2016                         }
2017                     } else {
2018                         nfscl_freeopen(op, 0);
2019                     }
2020                 }
2021                 op = nop;
2022             }
2023             owp = nowp;
2024         }
2025
2026         /*
2027          * Now, try and get any delegations not yet reclaimed by cobbling
2028          * to-gether an appropriate open.
2029          */
2030         nowp = NULL;
2031         dp = TAILQ_FIRST(&clp->nfsc_deleg);
2032         while (dp != NULL) {
2033             ndp = TAILQ_NEXT(dp, nfsdl_list);
2034             if ((dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM)) {
2035                 if (nowp == NULL) {
2036                     MALLOC(nowp, struct nfsclowner *,
2037                         sizeof (struct nfsclowner), M_NFSCLOWNER, M_WAITOK);
2038                     /*
2039                      * Name must be as long an largest possible
2040                      * NFSV4CL_LOCKNAMELEN. 12 for now.
2041                      */
2042                     NFSBCOPY("RECLAIMDELEG", nowp->nfsow_owner,
2043                         NFSV4CL_LOCKNAMELEN);
2044                     LIST_INIT(&nowp->nfsow_open);
2045                     nowp->nfsow_clp = clp;
2046                     nowp->nfsow_seqid = 0;
2047                     nowp->nfsow_defunct = 0;
2048                     nfscl_lockinit(&nowp->nfsow_rwlock);
2049                 }
2050                 nop = NULL;
2051                 if (error != NFSERR_NOGRACE) {
2052                     MALLOC(nop, struct nfsclopen *, sizeof (struct nfsclopen) +
2053                         dp->nfsdl_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
2054                     nop->nfso_own = nowp;
2055                     if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
2056                         nop->nfso_mode = NFSV4OPEN_ACCESSWRITE;
2057                         delegtype = NFSV4OPEN_DELEGATEWRITE;
2058                     } else {
2059                         nop->nfso_mode = NFSV4OPEN_ACCESSREAD;
2060                         delegtype = NFSV4OPEN_DELEGATEREAD;
2061                     }
2062                     nop->nfso_opencnt = 0;
2063                     nop->nfso_posixlock = 1;
2064                     nop->nfso_fhlen = dp->nfsdl_fhlen;
2065                     NFSBCOPY(dp->nfsdl_fh, nop->nfso_fh, dp->nfsdl_fhlen);
2066                     LIST_INIT(&nop->nfso_lock);
2067                     nop->nfso_stateid.seqid = 0;
2068                     nop->nfso_stateid.other[0] = 0;
2069                     nop->nfso_stateid.other[1] = 0;
2070                     nop->nfso_stateid.other[2] = 0;
2071                     newnfs_copycred(&dp->nfsdl_cred, tcred);
2072                     newnfs_copyincred(tcred, &nop->nfso_cred);
2073                     tdp = NULL;
2074                     error = nfscl_tryopen(nmp, NULL, nop->nfso_fh,
2075                         nop->nfso_fhlen, nop->nfso_fh, nop->nfso_fhlen,
2076                         nop->nfso_mode, nop, NULL, 0, &tdp, 1,
2077                         delegtype, tcred, p);
2078                     if (tdp != NULL) {
2079                         if ((tdp->nfsdl_flags & NFSCLDL_WRITE))
2080                             mode = NFSV4OPEN_ACCESSWRITE;
2081                         else
2082                             mode = NFSV4OPEN_ACCESSREAD;
2083                         if ((nop->nfso_mode & mode) == mode &&
2084                             nop->nfso_fhlen == tdp->nfsdl_fhlen &&
2085                             !NFSBCMP(nop->nfso_fh, tdp->nfsdl_fh,
2086                             nop->nfso_fhlen)) {
2087                             dp->nfsdl_stateid = tdp->nfsdl_stateid;
2088                             dp->nfsdl_sizelimit = tdp->nfsdl_sizelimit;
2089                             dp->nfsdl_ace = tdp->nfsdl_ace;
2090                             dp->nfsdl_change = tdp->nfsdl_change;
2091                             dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
2092                             if ((tdp->nfsdl_flags & NFSCLDL_RECALL))
2093                                 dp->nfsdl_flags |= NFSCLDL_RECALL;
2094                             FREE((caddr_t)tdp, M_NFSCLDELEG);
2095                         } else {
2096                             TAILQ_INSERT_HEAD(&extra_deleg, tdp, nfsdl_list);
2097                         }
2098                     }
2099                 }
2100                 if (error) {
2101                     if (nop != NULL)
2102                         FREE((caddr_t)nop, M_NFSCLOPEN);
2103                     /*
2104                      * Couldn't reclaim it, so throw the state
2105                      * away. Ouch!!
2106                      */
2107                     nfscl_cleandeleg(dp);
2108                     nfscl_freedeleg(&clp->nfsc_deleg, dp);
2109                 } else {
2110                     LIST_INSERT_HEAD(&extra_open, nop, nfso_list);
2111                 }
2112             }
2113             dp = ndp;
2114         }
2115
2116         /*
2117          * Now, get rid of extra Opens and Delegations.
2118          */
2119         LIST_FOREACH_SAFE(op, &extra_open, nfso_list, nop) {
2120                 do {
2121                         newnfs_copycred(&op->nfso_cred, tcred);
2122                         error = nfscl_tryclose(op, tcred, nmp, p);
2123                         if (error == NFSERR_GRACE)
2124                                 (void) nfs_catnap(PZERO, error, "nfsexcls");
2125                 } while (error == NFSERR_GRACE);
2126                 LIST_REMOVE(op, nfso_list);
2127                 FREE((caddr_t)op, M_NFSCLOPEN);
2128         }
2129         if (nowp != NULL)
2130                 FREE((caddr_t)nowp, M_NFSCLOWNER);
2131
2132         TAILQ_FOREACH_SAFE(dp, &extra_deleg, nfsdl_list, ndp) {
2133                 do {
2134                         newnfs_copycred(&dp->nfsdl_cred, tcred);
2135                         error = nfscl_trydelegreturn(dp, tcred, nmp, p);
2136                         if (error == NFSERR_GRACE)
2137                                 (void) nfs_catnap(PZERO, error, "nfsexdlg");
2138                 } while (error == NFSERR_GRACE);
2139                 TAILQ_REMOVE(&extra_deleg, dp, nfsdl_list);
2140                 FREE((caddr_t)dp, M_NFSCLDELEG);
2141         }
2142
2143         NFSLOCKCLSTATE();
2144         clp->nfsc_flags &= ~NFSCLFLAGS_RECVRINPROG;
2145         wakeup(&clp->nfsc_flags);
2146         nfsv4_unlock(&clp->nfsc_lock, 0);
2147         NFSUNLOCKCLSTATE();
2148         NFSFREECRED(tcred);
2149 }
2150
2151 /*
2152  * This function is called when a server replies with NFSERR_EXPIRED.
2153  * It deletes all state for the client and does a fresh SetClientId/confirm.
2154  * XXX Someday it should post a signal to the process(es) that hold the
2155  * state, so they know that lock state has been lost.
2156  */
2157 APPLESTATIC int
2158 nfscl_hasexpired(struct nfsclclient *clp, u_int32_t clidrev, NFSPROC_T *p)
2159 {
2160         struct nfscllockowner *lp, *nlp;
2161         struct nfsmount *nmp;
2162         struct ucred *cred;
2163         int igotlock = 0, error, trycnt;
2164
2165         /*
2166          * If the clientid has gone away or a new SetClientid has already
2167          * been done, just return ok.
2168          */
2169         if (clp == NULL || clidrev != clp->nfsc_clientidrev)
2170                 return (0);
2171
2172         /*
2173          * First, lock the client structure, so everyone else will
2174          * block when trying to use state. Also, use NFSCLFLAGS_EXPIREIT so
2175          * that only one thread does the work.
2176          */
2177         NFSLOCKCLSTATE();
2178         clp->nfsc_flags |= NFSCLFLAGS_EXPIREIT;
2179         do {
2180                 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
2181                     NFSCLSTATEMUTEXPTR, NULL);
2182         } while (!igotlock && (clp->nfsc_flags & NFSCLFLAGS_EXPIREIT));
2183         if ((clp->nfsc_flags & NFSCLFLAGS_EXPIREIT) == 0) {
2184                 if (igotlock)
2185                         nfsv4_unlock(&clp->nfsc_lock, 0);
2186                 NFSUNLOCKCLSTATE();
2187                 return (0);
2188         }
2189         clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG;
2190         NFSUNLOCKCLSTATE();
2191
2192         nmp = clp->nfsc_nmp;
2193         if (nmp == NULL)
2194                 panic("nfscl expired");
2195         cred = newnfs_getcred();
2196         trycnt = 5;
2197         do {
2198                 error = nfsrpc_setclient(nmp, clp, cred, p);
2199         } while ((error == NFSERR_STALECLIENTID ||
2200              error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
2201         if (error) {
2202                 /*
2203                  * Clear out any state.
2204                  */
2205                 nfscl_cleanclient(clp);
2206                 NFSLOCKCLSTATE();
2207                 clp->nfsc_flags &= ~(NFSCLFLAGS_HASCLIENTID |
2208                     NFSCLFLAGS_RECOVER);
2209         } else {
2210                 /* get rid of defunct lockowners */
2211                 LIST_FOREACH_SAFE(lp, &clp->nfsc_defunctlockowner, nfsl_list,
2212                     nlp) {
2213                         nfscl_freelockowner(lp, 0);
2214                 }
2215
2216                 /*
2217                  * Expire the state for the client.
2218                  */
2219                 nfscl_expireclient(clp, nmp, cred, p);
2220                 NFSLOCKCLSTATE();
2221                 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
2222                 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2223         }
2224         clp->nfsc_flags &= ~(NFSCLFLAGS_EXPIREIT | NFSCLFLAGS_RECVRINPROG);
2225         wakeup(&clp->nfsc_flags);
2226         nfsv4_unlock(&clp->nfsc_lock, 0);
2227         NFSUNLOCKCLSTATE();
2228         NFSFREECRED(cred);
2229         return (error);
2230 }
2231
2232 /*
2233  * This function inserts a lock in the list after insert_lop.
2234  */
2235 static void
2236 nfscl_insertlock(struct nfscllockowner *lp, struct nfscllock *new_lop,
2237     struct nfscllock *insert_lop, int local)
2238 {
2239
2240         if ((struct nfscllockowner *)insert_lop == lp)
2241                 LIST_INSERT_HEAD(&lp->nfsl_lock, new_lop, nfslo_list);
2242         else
2243                 LIST_INSERT_AFTER(insert_lop, new_lop, nfslo_list);
2244         if (local)
2245                 newnfsstats.cllocallocks++;
2246         else
2247                 newnfsstats.cllocks++;
2248 }
2249
2250 /*
2251  * This function updates the locking for a lock owner and given file. It
2252  * maintains a list of lock ranges ordered on increasing file offset that
2253  * are NFSCLLOCK_READ or NFSCLLOCK_WRITE and non-overlapping (aka POSIX style).
2254  * It always adds new_lop to the list and sometimes uses the one pointed
2255  * at by other_lopp.
2256  * Returns 1 if the locks were modified, 0 otherwise.
2257  */
2258 static int
2259 nfscl_updatelock(struct nfscllockowner *lp, struct nfscllock **new_lopp,
2260     struct nfscllock **other_lopp, int local)
2261 {
2262         struct nfscllock *new_lop = *new_lopp;
2263         struct nfscllock *lop, *tlop, *ilop;
2264         struct nfscllock *other_lop;
2265         int unlock = 0, modified = 0;
2266         u_int64_t tmp;
2267
2268         /*
2269          * Work down the list until the lock is merged.
2270          */
2271         if (new_lop->nfslo_type == F_UNLCK)
2272                 unlock = 1;
2273         ilop = (struct nfscllock *)lp;
2274         lop = LIST_FIRST(&lp->nfsl_lock);
2275         while (lop != NULL) {
2276             /*
2277              * Only check locks for this file that aren't before the start of
2278              * new lock's range.
2279              */
2280             if (lop->nfslo_end >= new_lop->nfslo_first) {
2281                 if (new_lop->nfslo_end < lop->nfslo_first) {
2282                     /*
2283                      * If the new lock ends before the start of the
2284                      * current lock's range, no merge, just insert
2285                      * the new lock.
2286                      */
2287                     break;
2288                 }
2289                 if (new_lop->nfslo_type == lop->nfslo_type ||
2290                     (new_lop->nfslo_first <= lop->nfslo_first &&
2291                      new_lop->nfslo_end >= lop->nfslo_end)) {
2292                     /*
2293                      * This lock can be absorbed by the new lock/unlock.
2294                      * This happens when it covers the entire range
2295                      * of the old lock or is contiguous
2296                      * with the old lock and is of the same type or an
2297                      * unlock.
2298                      */
2299                     if (new_lop->nfslo_type != lop->nfslo_type ||
2300                         new_lop->nfslo_first != lop->nfslo_first ||
2301                         new_lop->nfslo_end != lop->nfslo_end)
2302                         modified = 1;
2303                     if (lop->nfslo_first < new_lop->nfslo_first)
2304                         new_lop->nfslo_first = lop->nfslo_first;
2305                     if (lop->nfslo_end > new_lop->nfslo_end)
2306                         new_lop->nfslo_end = lop->nfslo_end;
2307                     tlop = lop;
2308                     lop = LIST_NEXT(lop, nfslo_list);
2309                     nfscl_freelock(tlop, local);
2310                     continue;
2311                 }
2312
2313                 /*
2314                  * All these cases are for contiguous locks that are not the
2315                  * same type, so they can't be merged.
2316                  */
2317                 if (new_lop->nfslo_first <= lop->nfslo_first) {
2318                     /*
2319                      * This case is where the new lock overlaps with the
2320                      * first part of the old lock. Move the start of the
2321                      * old lock to just past the end of the new lock. The
2322                      * new lock will be inserted in front of the old, since
2323                      * ilop hasn't been updated. (We are done now.)
2324                      */
2325                     if (lop->nfslo_first != new_lop->nfslo_end) {
2326                         lop->nfslo_first = new_lop->nfslo_end;
2327                         modified = 1;
2328                     }
2329                     break;
2330                 }
2331                 if (new_lop->nfslo_end >= lop->nfslo_end) {
2332                     /*
2333                      * This case is where the new lock overlaps with the
2334                      * end of the old lock's range. Move the old lock's
2335                      * end to just before the new lock's first and insert
2336                      * the new lock after the old lock.
2337                      * Might not be done yet, since the new lock could
2338                      * overlap further locks with higher ranges.
2339                      */
2340                     if (lop->nfslo_end != new_lop->nfslo_first) {
2341                         lop->nfslo_end = new_lop->nfslo_first;
2342                         modified = 1;
2343                     }
2344                     ilop = lop;
2345                     lop = LIST_NEXT(lop, nfslo_list);
2346                     continue;
2347                 }
2348                 /*
2349                  * The final case is where the new lock's range is in the
2350                  * middle of the current lock's and splits the current lock
2351                  * up. Use *other_lopp to handle the second part of the
2352                  * split old lock range. (We are done now.)
2353                  * For unlock, we use new_lop as other_lop and tmp, since
2354                  * other_lop and new_lop are the same for this case.
2355                  * We noted the unlock case above, so we don't need
2356                  * new_lop->nfslo_type any longer.
2357                  */
2358                 tmp = new_lop->nfslo_first;
2359                 if (unlock) {
2360                     other_lop = new_lop;
2361                     *new_lopp = NULL;
2362                 } else {
2363                     other_lop = *other_lopp;
2364                     *other_lopp = NULL;
2365                 }
2366                 other_lop->nfslo_first = new_lop->nfslo_end;
2367                 other_lop->nfslo_end = lop->nfslo_end;
2368                 other_lop->nfslo_type = lop->nfslo_type;
2369                 lop->nfslo_end = tmp;
2370                 nfscl_insertlock(lp, other_lop, lop, local);
2371                 ilop = lop;
2372                 modified = 1;
2373                 break;
2374             }
2375             ilop = lop;
2376             lop = LIST_NEXT(lop, nfslo_list);
2377             if (lop == NULL)
2378                 break;
2379         }
2380
2381         /*
2382          * Insert the new lock in the list at the appropriate place.
2383          */
2384         if (!unlock) {
2385                 nfscl_insertlock(lp, new_lop, ilop, local);
2386                 *new_lopp = NULL;
2387                 modified = 1;
2388         }
2389         return (modified);
2390 }
2391
2392 /*
2393  * This function must be run as a kernel thread.
2394  * It does Renew Ops and recovery, when required.
2395  */
2396 APPLESTATIC void
2397 nfscl_renewthread(struct nfsclclient *clp, NFSPROC_T *p)
2398 {
2399         struct nfsclowner *owp, *nowp;
2400         struct nfsclopen *op;
2401         struct nfscllockowner *lp, *nlp, *olp;
2402         struct nfscldeleghead dh;
2403         struct nfscllockownerhead lh;
2404         struct nfscldeleg *dp, *ndp;
2405         struct ucred *cred;
2406         u_int32_t clidrev;
2407         int error, cbpathdown, islept, igotlock, ret, clearok;
2408         uint32_t recover_done_time = 0;
2409
2410         cred = newnfs_getcred();
2411         NFSLOCKCLSTATE();
2412         clp->nfsc_flags |= NFSCLFLAGS_HASTHREAD;
2413         NFSUNLOCKCLSTATE();
2414         for(;;) {
2415                 newnfs_setroot(cred);
2416                 cbpathdown = 0;
2417                 if (clp->nfsc_flags & NFSCLFLAGS_RECOVER) {
2418                         /*
2419                          * Only allow one recover within 1/2 of the lease
2420                          * duration (nfsc_renew).
2421                          */
2422                         if (recover_done_time < NFSD_MONOSEC) {
2423                                 recover_done_time = NFSD_MONOSEC +
2424                                     clp->nfsc_renew;
2425                                 nfscl_recover(clp, cred, p);
2426                         } else {
2427                                 NFSLOCKCLSTATE();
2428                                 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2429                                 NFSUNLOCKCLSTATE();
2430                         }
2431                 }
2432                 if (clp->nfsc_expire <= NFSD_MONOSEC &&
2433                     (clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID)) {
2434                         clp->nfsc_expire = NFSD_MONOSEC + clp->nfsc_renew;
2435                         clidrev = clp->nfsc_clientidrev;
2436                         error = nfsrpc_renew(clp, cred, p);
2437                         if (error == NFSERR_CBPATHDOWN)
2438                             cbpathdown = 1;
2439                         else if (error == NFSERR_STALECLIENTID) {
2440                             NFSLOCKCLSTATE();
2441                             clp->nfsc_flags |= NFSCLFLAGS_RECOVER;
2442                             NFSUNLOCKCLSTATE();
2443                         } else if (error == NFSERR_EXPIRED)
2444                             (void) nfscl_hasexpired(clp, clidrev, p);
2445                 }
2446
2447                 LIST_INIT(&lh);
2448                 TAILQ_INIT(&dh);
2449                 NFSLOCKCLSTATE();
2450                 if (cbpathdown)
2451                         /* It's a Total Recall! */
2452                         nfscl_totalrecall(clp);
2453
2454                 /*
2455                  * Now, handle defunct owners.
2456                  */
2457                 owp = LIST_FIRST(&clp->nfsc_owner);
2458                 while (owp != NULL) {
2459                     nowp = LIST_NEXT(owp, nfsow_list);
2460                     if (LIST_EMPTY(&owp->nfsow_open)) {
2461                         if (owp->nfsow_defunct)
2462                             nfscl_freeopenowner(owp, 0);
2463                     } else {
2464                         LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
2465                             lp = LIST_FIRST(&op->nfso_lock);
2466                             while (lp != NULL) {
2467                                 nlp = LIST_NEXT(lp, nfsl_list);
2468                                 if (lp->nfsl_defunct &&
2469                                     LIST_EMPTY(&lp->nfsl_lock)) {
2470                                     LIST_FOREACH(olp, &lh, nfsl_list) {
2471                                         if (!NFSBCMP(olp->nfsl_owner,
2472                                             lp->nfsl_owner,NFSV4CL_LOCKNAMELEN))
2473                                             break;
2474                                     }
2475                                     if (olp == NULL) {
2476                                         LIST_REMOVE(lp, nfsl_list);
2477                                         LIST_INSERT_HEAD(&lh, lp, nfsl_list);
2478                                     } else {
2479                                         nfscl_freelockowner(lp, 0);
2480                                     }
2481                                 }
2482                                 lp = nlp;
2483                             }
2484                         }
2485                     }
2486                     owp = nowp;
2487                 }
2488
2489                 /* also search the defunct list */
2490                 lp = LIST_FIRST(&clp->nfsc_defunctlockowner);
2491                 while (lp != NULL) {
2492                     nlp = LIST_NEXT(lp, nfsl_list);
2493                     if (lp->nfsl_defunct) {
2494                         LIST_FOREACH(olp, &lh, nfsl_list) {
2495                             if (!NFSBCMP(olp->nfsl_owner, lp->nfsl_owner,
2496                                 NFSV4CL_LOCKNAMELEN))
2497                                 break;
2498                         }
2499                         if (olp == NULL) {
2500                             LIST_REMOVE(lp, nfsl_list);
2501                             LIST_INSERT_HEAD(&lh, lp, nfsl_list);
2502                         } else {
2503                             nfscl_freelockowner(lp, 0);
2504                         }
2505                     }
2506                     lp = nlp;
2507                 }
2508                 /* and release defunct lock owners */
2509                 LIST_FOREACH_SAFE(lp, &lh, nfsl_list, nlp) {
2510                     nfscl_freelockowner(lp, 0);
2511                 }
2512
2513                 /*
2514                  * Do the recall on any delegations. To avoid trouble, always
2515                  * come back up here after having slept.
2516                  */
2517                 igotlock = 0;
2518 tryagain:
2519                 dp = TAILQ_FIRST(&clp->nfsc_deleg);
2520                 while (dp != NULL) {
2521                         ndp = TAILQ_NEXT(dp, nfsdl_list);
2522                         if ((dp->nfsdl_flags & NFSCLDL_RECALL)) {
2523                                 /*
2524                                  * Wait for outstanding I/O ops to be done.
2525                                  */
2526                                 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
2527                                     if (igotlock) {
2528                                         nfsv4_unlock(&clp->nfsc_lock, 0);
2529                                         igotlock = 0;
2530                                     }
2531                                     dp->nfsdl_rwlock.nfslock_lock |=
2532                                         NFSV4LOCK_WANTED;
2533                                     (void) nfsmsleep(&dp->nfsdl_rwlock,
2534                                         NFSCLSTATEMUTEXPTR, PZERO, "nfscld",
2535                                         NULL);
2536                                     goto tryagain;
2537                                 }
2538                                 while (!igotlock) {
2539                                     igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
2540                                         &islept, NFSCLSTATEMUTEXPTR, NULL);
2541                                     if (islept)
2542                                         goto tryagain;
2543                                 }
2544                                 NFSUNLOCKCLSTATE();
2545                                 newnfs_copycred(&dp->nfsdl_cred, cred);
2546                                 ret = nfscl_recalldeleg(clp, clp->nfsc_nmp, dp,
2547                                     NULL, cred, p, 1);
2548                                 if (!ret) {
2549                                     nfscl_cleandeleg(dp);
2550                                     TAILQ_REMOVE(&clp->nfsc_deleg, dp,
2551                                         nfsdl_list);
2552                                     LIST_REMOVE(dp, nfsdl_hash);
2553                                     TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
2554                                     nfscl_delegcnt--;
2555                                     newnfsstats.cldelegates--;
2556                                 }
2557                                 NFSLOCKCLSTATE();
2558                         }
2559                         dp = ndp;
2560                 }
2561
2562                 /*
2563                  * Clear out old delegations, if we are above the high water
2564                  * mark. Only clear out ones with no state related to them.
2565                  * The tailq list is in LRU order.
2566                  */
2567                 dp = TAILQ_LAST(&clp->nfsc_deleg, nfscldeleghead);
2568                 while (nfscl_delegcnt > nfscl_deleghighwater && dp != NULL) {
2569                     ndp = TAILQ_PREV(dp, nfscldeleghead, nfsdl_list);
2570                     if (dp->nfsdl_rwlock.nfslock_usecnt == 0 &&
2571                         dp->nfsdl_rwlock.nfslock_lock == 0 &&
2572                         dp->nfsdl_timestamp < NFSD_MONOSEC &&
2573                         (dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_ZAPPED |
2574                           NFSCLDL_NEEDRECLAIM | NFSCLDL_DELEGRET)) == 0) {
2575                         clearok = 1;
2576                         LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
2577                             op = LIST_FIRST(&owp->nfsow_open);
2578                             if (op != NULL) {
2579                                 clearok = 0;
2580                                 break;
2581                             }
2582                         }
2583                         if (clearok) {
2584                             LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
2585                                 if (!LIST_EMPTY(&lp->nfsl_lock)) {
2586                                     clearok = 0;
2587                                     break;
2588                                 }
2589                             }
2590                         }
2591                         if (clearok) {
2592                             TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
2593                             LIST_REMOVE(dp, nfsdl_hash);
2594                             TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
2595                             nfscl_delegcnt--;
2596                             newnfsstats.cldelegates--;
2597                         }
2598                     }
2599                     dp = ndp;
2600                 }
2601                 if (igotlock)
2602                         nfsv4_unlock(&clp->nfsc_lock, 0);
2603                 NFSUNLOCKCLSTATE();
2604
2605                 /*
2606                  * Delegreturn any delegations cleaned out or recalled.
2607                  */
2608                 TAILQ_FOREACH_SAFE(dp, &dh, nfsdl_list, ndp) {
2609                         newnfs_copycred(&dp->nfsdl_cred, cred);
2610                         (void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
2611                         TAILQ_REMOVE(&dh, dp, nfsdl_list);
2612                         FREE((caddr_t)dp, M_NFSCLDELEG);
2613                 }
2614
2615 #if defined(APPLEKEXT) || defined(__FreeBSD__)
2616                 /*
2617                  * Simulate the calls to nfscl_cleanup() when a process
2618                  * exits, since the call can't be patched into exit().
2619                  */
2620                 {
2621                         struct timespec mytime;
2622                         static time_t prevsec = 0;
2623
2624                         NFSGETNANOTIME(&mytime);
2625                         if (prevsec != mytime.tv_sec) {
2626                                 prevsec = mytime.tv_sec;
2627                                 nfscl_cleanupkext(clp);
2628                         }
2629                 }
2630 #endif  /* APPLEKEXT || __FreeBSD__ */
2631
2632                 NFSLOCKCLSTATE();
2633                 if ((clp->nfsc_flags & NFSCLFLAGS_RECOVER) == 0)
2634                         (void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT, "nfscl",
2635                             hz);
2636                 if (clp->nfsc_flags & NFSCLFLAGS_UMOUNT) {
2637                         clp->nfsc_flags &= ~NFSCLFLAGS_HASTHREAD;
2638                         NFSUNLOCKCLSTATE();
2639                         NFSFREECRED(cred);
2640                         wakeup((caddr_t)clp);
2641                         return;
2642                 }
2643                 NFSUNLOCKCLSTATE();
2644         }
2645 }
2646
2647 /*
2648  * Initiate state recovery. Called when NFSERR_STALECLIENTID or
2649  * NFSERR_STALESTATEID is received.
2650  */
2651 APPLESTATIC void
2652 nfscl_initiate_recovery(struct nfsclclient *clp)
2653 {
2654
2655         if (clp == NULL)
2656                 return;
2657         NFSLOCKCLSTATE();
2658         clp->nfsc_flags |= NFSCLFLAGS_RECOVER;
2659         NFSUNLOCKCLSTATE();
2660         wakeup((caddr_t)clp);
2661 }
2662
2663 /*
2664  * Dump out the state stuff for debugging.
2665  */
2666 APPLESTATIC void
2667 nfscl_dumpstate(struct nfsmount *nmp, int openowner, int opens,
2668     int lockowner, int locks)
2669 {
2670         struct nfsclclient *clp;
2671         struct nfsclowner *owp;
2672         struct nfsclopen *op;
2673         struct nfscllockowner *lp;
2674         struct nfscllock *lop;
2675         struct nfscldeleg *dp;
2676
2677         clp = nmp->nm_clp;
2678         if (clp == NULL) {
2679                 printf("nfscl dumpstate NULL clp\n");
2680                 return;
2681         }
2682         NFSLOCKCLSTATE();
2683         TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2684           LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
2685             if (openowner && !LIST_EMPTY(&owp->nfsow_open))
2686                 printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
2687                     owp->nfsow_owner[0], owp->nfsow_owner[1],
2688                     owp->nfsow_owner[2], owp->nfsow_owner[3],
2689                     owp->nfsow_seqid);
2690             LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
2691                 if (opens)
2692                     printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
2693                         op->nfso_stateid.other[0], op->nfso_stateid.other[1],
2694                         op->nfso_stateid.other[2], op->nfso_opencnt,
2695                         op->nfso_fh[12]);
2696                 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
2697                     if (lockowner)
2698                         printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
2699                             lp->nfsl_owner[0], lp->nfsl_owner[1],
2700                             lp->nfsl_owner[2], lp->nfsl_owner[3],
2701                             lp->nfsl_seqid,
2702                             lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
2703                             lp->nfsl_stateid.other[2]);
2704                     LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
2705                         if (locks)
2706 #ifdef __FreeBSD__
2707                             printf("lck typ=%d fst=%ju end=%ju\n",
2708                                 lop->nfslo_type, (intmax_t)lop->nfslo_first,
2709                                 (intmax_t)lop->nfslo_end);
2710 #else
2711                             printf("lck typ=%d fst=%qd end=%qd\n",
2712                                 lop->nfslo_type, lop->nfslo_first,
2713                                 lop->nfslo_end);
2714 #endif
2715                     }
2716                 }
2717             }
2718           }
2719         }
2720         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
2721             if (openowner && !LIST_EMPTY(&owp->nfsow_open))
2722                 printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
2723                     owp->nfsow_owner[0], owp->nfsow_owner[1],
2724                     owp->nfsow_owner[2], owp->nfsow_owner[3],
2725                     owp->nfsow_seqid);
2726             LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
2727                 if (opens)
2728                     printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
2729                         op->nfso_stateid.other[0], op->nfso_stateid.other[1],
2730                         op->nfso_stateid.other[2], op->nfso_opencnt,
2731                         op->nfso_fh[12]);
2732                 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
2733                     if (lockowner)
2734                         printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
2735                             lp->nfsl_owner[0], lp->nfsl_owner[1],
2736                             lp->nfsl_owner[2], lp->nfsl_owner[3],
2737                             lp->nfsl_seqid,
2738                             lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
2739                             lp->nfsl_stateid.other[2]);
2740                     LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
2741                         if (locks)
2742 #ifdef __FreeBSD__
2743                             printf("lck typ=%d fst=%ju end=%ju\n",
2744                                 lop->nfslo_type, (intmax_t)lop->nfslo_first,
2745                                 (intmax_t)lop->nfslo_end);
2746 #else
2747                             printf("lck typ=%d fst=%qd end=%qd\n",
2748                                 lop->nfslo_type, lop->nfslo_first,
2749                                 lop->nfslo_end);
2750 #endif
2751                     }
2752                 }
2753             }
2754         }
2755         NFSUNLOCKCLSTATE();
2756 }
2757
2758 /*
2759  * Check for duplicate open owners and opens.
2760  * (Only used as a diagnostic aid.)
2761  */
2762 APPLESTATIC void
2763 nfscl_dupopen(vnode_t vp, int dupopens)
2764 {
2765         struct nfsclclient *clp;
2766         struct nfsclowner *owp, *owp2;
2767         struct nfsclopen *op, *op2;
2768         struct nfsfh *nfhp;
2769
2770         clp = VFSTONFS(vnode_mount(vp))->nm_clp;
2771         if (clp == NULL) {
2772                 printf("nfscl dupopen NULL clp\n");
2773                 return;
2774         }
2775         nfhp = VTONFS(vp)->n_fhp;
2776         NFSLOCKCLSTATE();
2777
2778         /*
2779          * First, search for duplicate owners.
2780          * These should never happen!
2781          */
2782         LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
2783             LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
2784                 if (owp != owp2 &&
2785                     !NFSBCMP(owp->nfsow_owner, owp2->nfsow_owner,
2786                     NFSV4CL_LOCKNAMELEN)) {
2787                         NFSUNLOCKCLSTATE();
2788                         printf("DUP OWNER\n");
2789                         nfscl_dumpstate(VFSTONFS(vnode_mount(vp)), 1, 1, 0, 0);
2790                         return;
2791                 }
2792             }
2793         }
2794
2795         /*
2796          * Now, search for duplicate stateids.
2797          * These shouldn't happen, either.
2798          */
2799         LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
2800             LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
2801                 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
2802                     LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
2803                         if (op != op2 &&
2804                             (op->nfso_stateid.other[0] != 0 ||
2805                              op->nfso_stateid.other[1] != 0 ||
2806                              op->nfso_stateid.other[2] != 0) &&
2807                             op->nfso_stateid.other[0] == op2->nfso_stateid.other[0] &&
2808                             op->nfso_stateid.other[1] == op2->nfso_stateid.other[1] &&
2809                             op->nfso_stateid.other[2] == op2->nfso_stateid.other[2]) {
2810                             NFSUNLOCKCLSTATE();
2811                             printf("DUP STATEID\n");
2812                             nfscl_dumpstate(VFSTONFS(vnode_mount(vp)), 1, 1, 0,
2813                                 0);
2814                             return;
2815                         }
2816                     }
2817                 }
2818             }
2819         }
2820
2821         /*
2822          * Now search for duplicate opens.
2823          * Duplicate opens for the same owner
2824          * should never occur. Other duplicates are
2825          * possible and are checked for if "dupopens"
2826          * is true.
2827          */
2828         LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
2829             LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
2830                 if (nfhp->nfh_len == op2->nfso_fhlen &&
2831                     !NFSBCMP(nfhp->nfh_fh, op2->nfso_fh, nfhp->nfh_len)) {
2832                     LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
2833                         LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
2834                             if (op != op2 && nfhp->nfh_len == op->nfso_fhlen &&
2835                                 !NFSBCMP(nfhp->nfh_fh, op->nfso_fh, nfhp->nfh_len) &&
2836                                 (!NFSBCMP(op->nfso_own->nfsow_owner,
2837                                  op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN) ||
2838                                  dupopens)) {
2839                                 if (!NFSBCMP(op->nfso_own->nfsow_owner,
2840                                     op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
2841                                     NFSUNLOCKCLSTATE();
2842                                     printf("BADDUP OPEN\n");
2843                                 } else {
2844                                     NFSUNLOCKCLSTATE();
2845                                     printf("DUP OPEN\n");
2846                                 }
2847                                 nfscl_dumpstate(VFSTONFS(vnode_mount(vp)), 1, 1,
2848                                     0, 0);
2849                                 return;
2850                             }
2851                         }
2852                     }
2853                 }
2854             }
2855         }
2856         NFSUNLOCKCLSTATE();
2857 }
2858
2859 /*
2860  * During close, find an open that needs to be dereferenced and
2861  * dereference it. If there are no more opens for this file,
2862  * log a message to that effect.
2863  * Opens aren't actually Close'd until VOP_INACTIVE() is performed
2864  * on the file's vnode.
2865  * This is the safe way, since it is difficult to identify
2866  * which open the close is for and I/O can be performed after the
2867  * close(2) system call when a file is mmap'd.
2868  * If it returns 0 for success, there will be a referenced
2869  * clp returned via clpp.
2870  */
2871 APPLESTATIC int
2872 nfscl_getclose(vnode_t vp, struct nfsclclient **clpp)
2873 {
2874         struct nfsclclient *clp;
2875         struct nfsclowner *owp;
2876         struct nfsclopen *op;
2877         struct nfscldeleg *dp;
2878         struct nfsfh *nfhp;
2879         int error, notdecr;
2880
2881         error = nfscl_getcl(vp, NULL, NULL, &clp);
2882         if (error)
2883                 return (error);
2884         *clpp = clp;
2885
2886         nfhp = VTONFS(vp)->n_fhp;
2887         notdecr = 1;
2888         NFSLOCKCLSTATE();
2889         /*
2890          * First, look for one under a delegation that was locally issued
2891          * and just decrement the opencnt for it. Since all my Opens against
2892          * the server are DENY_NONE, I don't see a problem with hanging
2893          * onto them. (It is much easier to use one of the extant Opens
2894          * that I already have on the server when a Delegation is recalled
2895          * than to do fresh Opens.) Someday, I might need to rethink this, but.
2896          */
2897         dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
2898         if (dp != NULL) {
2899                 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
2900                         op = LIST_FIRST(&owp->nfsow_open);
2901                         if (op != NULL) {
2902                                 /*
2903                                  * Since a delegation is for a file, there
2904                                  * should never be more than one open for
2905                                  * each openowner.
2906                                  */
2907                                 if (LIST_NEXT(op, nfso_list) != NULL)
2908                                         panic("nfscdeleg opens");
2909                                 if (notdecr && op->nfso_opencnt > 0) {
2910                                         notdecr = 0;
2911                                         op->nfso_opencnt--;
2912                                         break;
2913                                 }
2914                         }
2915                 }
2916         }
2917
2918         /* Now process the opens against the server. */
2919         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
2920                 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
2921                         if (op->nfso_fhlen == nfhp->nfh_len &&
2922                             !NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
2923                             nfhp->nfh_len)) {
2924                                 /* Found an open, decrement cnt if possible */
2925                                 if (notdecr && op->nfso_opencnt > 0) {
2926                                         notdecr = 0;
2927                                         op->nfso_opencnt--;
2928                                 }
2929                                 /*
2930                                  * There are more opens, so just return.
2931                                  */
2932                                 if (op->nfso_opencnt > 0) {
2933                                         NFSUNLOCKCLSTATE();
2934                                         return (0);
2935                                 }
2936                         }
2937                 }
2938         }
2939         NFSUNLOCKCLSTATE();
2940         if (notdecr)
2941                 printf("nfscl: never fnd open\n");
2942         return (0);
2943 }
2944
2945 APPLESTATIC int
2946 nfscl_doclose(vnode_t vp, struct nfsclclient **clpp, NFSPROC_T *p)
2947 {
2948         struct nfsclclient *clp;
2949         struct nfsclowner *owp, *nowp;
2950         struct nfsclopen *op;
2951         struct nfscldeleg *dp;
2952         struct nfsfh *nfhp;
2953         int error;
2954
2955         error = nfscl_getcl(vp, NULL, NULL, &clp);
2956         if (error)
2957                 return (error);
2958         *clpp = clp;
2959
2960         nfhp = VTONFS(vp)->n_fhp;
2961         NFSLOCKCLSTATE();
2962         /*
2963          * First get rid of the local Open structures, which should be no
2964          * longer in use.
2965          */
2966         dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
2967         if (dp != NULL) {
2968                 LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) {
2969                         op = LIST_FIRST(&owp->nfsow_open);
2970                         if (op != NULL) {
2971                                 KASSERT((op->nfso_opencnt == 0),
2972                                     ("nfscl: bad open cnt on deleg"));
2973                                 nfscl_freeopen(op, 1);
2974                         }
2975                         nfscl_freeopenowner(owp, 1);
2976                 }
2977         }
2978
2979         /* Now process the opens against the server. */
2980 lookformore:
2981         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
2982                 op = LIST_FIRST(&owp->nfsow_open);
2983                 while (op != NULL) {
2984                         if (op->nfso_fhlen == nfhp->nfh_len &&
2985                             !NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
2986                             nfhp->nfh_len)) {
2987                                 /* Found an open, close it. */
2988                                 KASSERT((op->nfso_opencnt == 0),
2989                                     ("nfscl: bad open cnt on server"));
2990                                 NFSUNLOCKCLSTATE();
2991                                 nfsrpc_doclose(VFSTONFS(vnode_mount(vp)), op,
2992                                     p);
2993                                 NFSLOCKCLSTATE();
2994                                 goto lookformore;
2995                         }
2996                         op = LIST_NEXT(op, nfso_list);
2997                 }
2998         }
2999         NFSUNLOCKCLSTATE();
3000         return (0);
3001 }
3002
3003 /*
3004  * Return all delegations on this client.
3005  * (Must be called with client sleep lock.)
3006  */
3007 static void
3008 nfscl_delegreturnall(struct nfsclclient *clp, NFSPROC_T *p)
3009 {
3010         struct nfscldeleg *dp, *ndp;
3011         struct ucred *cred;
3012
3013         cred = newnfs_getcred();
3014         TAILQ_FOREACH_SAFE(dp, &clp->nfsc_deleg, nfsdl_list, ndp) {
3015                 nfscl_cleandeleg(dp);
3016                 (void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
3017                 nfscl_freedeleg(&clp->nfsc_deleg, dp);
3018         }
3019         NFSFREECRED(cred);
3020 }
3021
3022 /*
3023  * Do a callback RPC.
3024  */
3025 APPLESTATIC void
3026 nfscl_docb(struct nfsrv_descript *nd, NFSPROC_T *p)
3027 {
3028         int i, op;
3029         u_int32_t *tl;
3030         struct nfsclclient *clp;
3031         struct nfscldeleg *dp = NULL;
3032         int numops, taglen = -1, error = 0, trunc, ret = 0;
3033         u_int32_t minorvers, retops = 0, *retopsp = NULL, *repp, cbident;
3034         u_char tag[NFSV4_SMALLSTR + 1], *tagstr;
3035         vnode_t vp = NULL;
3036         struct nfsnode *np;
3037         struct vattr va;
3038         struct nfsfh *nfhp;
3039         mount_t mp;
3040         nfsattrbit_t attrbits, rattrbits;
3041         nfsv4stateid_t stateid;
3042
3043         nfsrvd_rephead(nd);
3044         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3045         taglen = fxdr_unsigned(int, *tl);
3046         if (taglen < 0) {
3047                 error = EBADRPC;
3048                 goto nfsmout;
3049         }
3050         if (taglen <= NFSV4_SMALLSTR)
3051                 tagstr = tag;
3052         else
3053                 tagstr = malloc(taglen + 1, M_TEMP, M_WAITOK);
3054         error = nfsrv_mtostr(nd, tagstr, taglen);
3055         if (error) {
3056                 if (taglen > NFSV4_SMALLSTR)
3057                         free(tagstr, M_TEMP);
3058                 taglen = -1;
3059                 goto nfsmout;
3060         }
3061         (void) nfsm_strtom(nd, tag, taglen);
3062         if (taglen > NFSV4_SMALLSTR) {
3063                 free(tagstr, M_TEMP);
3064         }
3065         NFSM_BUILD(retopsp, u_int32_t *, NFSX_UNSIGNED);
3066         NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3067         minorvers = fxdr_unsigned(u_int32_t, *tl++);
3068         if (minorvers != NFSV4_MINORVERSION)
3069                 nd->nd_repstat = NFSERR_MINORVERMISMATCH;
3070         cbident = fxdr_unsigned(u_int32_t, *tl++);
3071         if (nd->nd_repstat)
3072                 numops = 0;
3073         else
3074                 numops = fxdr_unsigned(int, *tl);
3075         /*
3076          * Loop around doing the sub ops.
3077          */
3078         for (i = 0; i < numops; i++) {
3079                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3080                 NFSM_BUILD(repp, u_int32_t *, 2 * NFSX_UNSIGNED);
3081                 *repp++ = *tl;
3082                 op = fxdr_unsigned(int, *tl);
3083                 if (op < NFSV4OP_CBGETATTR || op > NFSV4OP_CBRECALL) {
3084                     nd->nd_repstat = NFSERR_OPILLEGAL;
3085                     *repp = nfscl_errmap(nd);
3086                     retops++;
3087                     break;
3088                 }
3089                 nd->nd_procnum = op;
3090                 newnfsstats.cbrpccnt[nd->nd_procnum]++;
3091                 switch (op) {
3092                 case NFSV4OP_CBGETATTR:
3093                         clp = NULL;
3094                         error = nfsm_getfh(nd, &nfhp);
3095                         if (!error)
3096                                 error = nfsrv_getattrbits(nd, &attrbits,
3097                                     NULL, NULL);
3098                         if (!error) {
3099                                 mp = nfscl_getmnt(cbident);
3100                                 if (mp == NULL)
3101                                         error = NFSERR_SERVERFAULT;
3102                         }
3103                         if (!error) {
3104                                 dp = NULL;
3105                                 NFSLOCKCLSTATE();
3106                                 clp = nfscl_findcl(VFSTONFS(mp));
3107                                 if (clp != NULL)
3108                                         dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
3109                                             nfhp->nfh_len);
3110                                 NFSUNLOCKCLSTATE();
3111                                 if (dp == NULL)
3112                                         error = NFSERR_SERVERFAULT;
3113                         }
3114                         if (!error) {
3115                                 ret = nfscl_ngetreopen(mp, nfhp->nfh_fh,
3116                                     nfhp->nfh_len, p, &np);
3117                                 if (!ret)
3118                                         vp = NFSTOV(np);
3119                         }
3120                         if (nfhp != NULL)
3121                                 FREE((caddr_t)nfhp, M_NFSFH);
3122                         if (!error) {
3123                                 NFSZERO_ATTRBIT(&rattrbits);
3124                                 if (NFSISSET_ATTRBIT(&attrbits,
3125                                     NFSATTRBIT_SIZE)) {
3126                                         if (!ret)
3127                                                 va.va_size = np->n_size;
3128                                         else
3129                                                 va.va_size = dp->nfsdl_size;
3130                                         NFSSETBIT_ATTRBIT(&rattrbits,
3131                                             NFSATTRBIT_SIZE);
3132                                 }
3133                                 if (NFSISSET_ATTRBIT(&attrbits,
3134                                     NFSATTRBIT_CHANGE)) {
3135                                         va.va_filerev = dp->nfsdl_change;
3136                                         if (ret || (np->n_flag & NDELEGMOD))
3137                                                 va.va_filerev++;
3138                                         NFSSETBIT_ATTRBIT(&rattrbits,
3139                                             NFSATTRBIT_CHANGE);
3140                                 }
3141                                 (void) nfsv4_fillattr(nd, NULL, NULL, NULL, &va,
3142                                     NULL, 0, &rattrbits, NULL, NULL, 0, 0, 0, 0,
3143                                     (uint64_t)0);
3144                                 if (!ret)
3145                                         vrele(vp);
3146                         }
3147                         break;
3148                 case NFSV4OP_CBRECALL:
3149                         clp = NULL;
3150                         NFSM_DISSECT(tl, u_int32_t *, NFSX_STATEID +
3151                             NFSX_UNSIGNED);
3152                         stateid.seqid = *tl++;
3153                         NFSBCOPY((caddr_t)tl, (caddr_t)stateid.other,
3154                             NFSX_STATEIDOTHER);
3155                         tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
3156                         trunc = fxdr_unsigned(int, *tl);
3157                         error = nfsm_getfh(nd, &nfhp);
3158                         if (!error) {
3159                                 mp = nfscl_getmnt(cbident);
3160                                 if (mp == NULL)
3161                                         error = NFSERR_SERVERFAULT;
3162                         }
3163                         if (!error) {
3164                                 NFSLOCKCLSTATE();
3165                                 clp = nfscl_findcl(VFSTONFS(mp));
3166                                 if (clp != NULL) {
3167                                         dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
3168                                             nfhp->nfh_len);
3169                                         if (dp != NULL && (dp->nfsdl_flags &
3170                                             NFSCLDL_DELEGRET) == 0) {
3171                                                 dp->nfsdl_flags |=
3172                                                     NFSCLDL_RECALL;
3173                                                 wakeup((caddr_t)clp);
3174                                         }
3175                                 } else {
3176                                         error = NFSERR_SERVERFAULT;
3177                                 }
3178                                 NFSUNLOCKCLSTATE();
3179                         }
3180                         if (nfhp != NULL)
3181                                 FREE((caddr_t)nfhp, M_NFSFH);
3182                         break;
3183                 };
3184                 if (error) {
3185                         if (error == EBADRPC || error == NFSERR_BADXDR) {
3186                                 nd->nd_repstat = NFSERR_BADXDR;
3187                         } else {
3188                                 nd->nd_repstat = error;
3189                         }
3190                         error = 0;
3191                 }
3192                 retops++;
3193                 if (nd->nd_repstat) {
3194                         *repp = nfscl_errmap(nd);
3195                         break;
3196                 } else
3197                         *repp = 0;      /* NFS4_OK */
3198         }
3199 nfsmout:
3200         if (error) {
3201                 if (error == EBADRPC || error == NFSERR_BADXDR)
3202                         nd->nd_repstat = NFSERR_BADXDR;
3203                 else
3204                         printf("nfsv4 comperr1=%d\n", error);
3205         }
3206         if (taglen == -1) {
3207                 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
3208                 *tl++ = 0;
3209                 *tl = 0;
3210         } else {
3211                 *retopsp = txdr_unsigned(retops);
3212         }
3213         *nd->nd_errp = nfscl_errmap(nd);
3214 }
3215
3216 /*
3217  * Generate the next cbident value. Basically just increment a static value
3218  * and then check that it isn't already in the list, if it has wrapped around.
3219  */
3220 static u_int32_t
3221 nfscl_nextcbident(void)
3222 {
3223         struct nfsclclient *clp;
3224         int matched;
3225         static u_int32_t nextcbident = 0;
3226         static int haswrapped = 0;
3227
3228         nextcbident++;
3229         if (nextcbident == 0)
3230                 haswrapped = 1;
3231         if (haswrapped) {
3232                 /*
3233                  * Search the clientid list for one already using this cbident.
3234                  */
3235                 do {
3236                         matched = 0;
3237                         NFSLOCKCLSTATE();
3238                         LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3239                                 if (clp->nfsc_cbident == nextcbident) {
3240                                         matched = 1;
3241                                         break;
3242                                 }
3243                         }
3244                         NFSUNLOCKCLSTATE();
3245                         if (matched == 1)
3246                                 nextcbident++;
3247                 } while (matched);
3248         }
3249         return (nextcbident);
3250 }
3251
3252 /*
3253  * Get the mount point related to a given cbident.
3254  */
3255 static mount_t
3256 nfscl_getmnt(u_int32_t cbident)
3257 {
3258         struct nfsclclient *clp;
3259         struct nfsmount *nmp;
3260
3261         NFSLOCKCLSTATE();
3262         LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3263                 if (clp->nfsc_cbident == cbident)
3264                         break;
3265         }
3266         if (clp == NULL) {
3267                 NFSUNLOCKCLSTATE();
3268                 return (NULL);
3269         }
3270         nmp = clp->nfsc_nmp;
3271         NFSUNLOCKCLSTATE();
3272         return (nmp->nm_mountp);
3273 }
3274
3275 /*
3276  * Search for a lock conflict locally on the client. A conflict occurs if
3277  * - not same owner and overlapping byte range and at least one of them is
3278  *   a write lock or this is an unlock.
3279  */
3280 static int
3281 nfscl_localconflict(struct nfsclclient *clp, u_int8_t *fhp, int fhlen,
3282     struct nfscllock *nlop, u_int8_t *own, struct nfscldeleg *dp,
3283     struct nfscllock **lopp)
3284 {
3285         struct nfsclowner *owp;
3286         struct nfsclopen *op;
3287         int ret;
3288
3289         if (dp != NULL) {
3290                 ret = nfscl_checkconflict(&dp->nfsdl_lock, nlop, own, lopp);
3291                 if (ret)
3292                         return (ret);
3293         }
3294         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3295                 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3296                         if (op->nfso_fhlen == fhlen &&
3297                             !NFSBCMP(op->nfso_fh, fhp, fhlen)) {
3298                                 ret = nfscl_checkconflict(&op->nfso_lock, nlop,
3299                                     own, lopp);
3300                                 if (ret)
3301                                         return (ret);
3302                         }
3303                 }
3304         }
3305         return (0);
3306 }
3307
3308 static int
3309 nfscl_checkconflict(struct nfscllockownerhead *lhp, struct nfscllock *nlop,
3310     u_int8_t *own, struct nfscllock **lopp)
3311 {
3312         struct nfscllockowner *lp;
3313         struct nfscllock *lop;
3314
3315         LIST_FOREACH(lp, lhp, nfsl_list) {
3316                 if (NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
3317                         LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
3318                                 if (lop->nfslo_first >= nlop->nfslo_end)
3319                                         break;
3320                                 if (lop->nfslo_end <= nlop->nfslo_first)
3321                                         continue;
3322                                 if (lop->nfslo_type == F_WRLCK ||
3323                                     nlop->nfslo_type == F_WRLCK ||
3324                                     nlop->nfslo_type == F_UNLCK) {
3325                                         if (lopp != NULL)
3326                                                 *lopp = lop;
3327                                         return (NFSERR_DENIED);
3328                                 }
3329                         }
3330                 }
3331         }
3332         return (0);
3333 }
3334
3335 /*
3336  * Check for a local conflicting lock.
3337  */
3338 APPLESTATIC int
3339 nfscl_lockt(vnode_t vp, struct nfsclclient *clp, u_int64_t off,
3340     u_int64_t len, struct flock *fl, NFSPROC_T *p, void *id, int flags)
3341 {
3342         struct nfscllock *lop, nlck;
3343         struct nfscldeleg *dp;
3344         struct nfsnode *np;
3345         u_int8_t own[NFSV4CL_LOCKNAMELEN];
3346         int error;
3347
3348         nlck.nfslo_type = fl->l_type;
3349         nlck.nfslo_first = off;
3350         if (len == NFS64BITSSET) {
3351                 nlck.nfslo_end = NFS64BITSSET;
3352         } else {
3353                 nlck.nfslo_end = off + len;
3354                 if (nlck.nfslo_end <= nlck.nfslo_first)
3355                         return (NFSERR_INVAL);
3356         }
3357         np = VTONFS(vp);
3358         nfscl_filllockowner(id, own, flags);
3359         NFSLOCKCLSTATE();
3360         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
3361         error = nfscl_localconflict(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len,
3362             &nlck, own, dp, &lop);
3363         if (error != 0) {
3364                 fl->l_whence = SEEK_SET;
3365                 fl->l_start = lop->nfslo_first;
3366                 if (lop->nfslo_end == NFS64BITSSET)
3367                         fl->l_len = 0;
3368                 else
3369                         fl->l_len = lop->nfslo_end - lop->nfslo_first;
3370                 fl->l_pid = (pid_t)0;
3371                 fl->l_type = lop->nfslo_type;
3372                 error = -1;                     /* no RPC required */
3373         } else if (dp != NULL && ((dp->nfsdl_flags & NFSCLDL_WRITE) ||
3374             fl->l_type == F_RDLCK)) {
3375                 /*
3376                  * The delegation ensures that there isn't a conflicting
3377                  * lock on the server, so return -1 to indicate an RPC
3378                  * isn't required.
3379                  */
3380                 fl->l_type = F_UNLCK;
3381                 error = -1;
3382         }
3383         NFSUNLOCKCLSTATE();
3384         return (error);
3385 }
3386
3387 /*
3388  * Handle Recall of a delegation.
3389  * The clp must be exclusive locked when this is called.
3390  */
3391 static int
3392 nfscl_recalldeleg(struct nfsclclient *clp, struct nfsmount *nmp,
3393     struct nfscldeleg *dp, vnode_t vp, struct ucred *cred, NFSPROC_T *p,
3394     int called_from_renewthread)
3395 {
3396         struct nfsclowner *owp, *lowp, *nowp;
3397         struct nfsclopen *op, *lop;
3398         struct nfscllockowner *lp;
3399         struct nfscllock *lckp;
3400         struct nfsnode *np;
3401         int error = 0, ret, gotvp = 0;
3402
3403         if (vp == NULL) {
3404                 /*
3405                  * First, get a vnode for the file. This is needed to do RPCs.
3406                  */
3407                 ret = nfscl_ngetreopen(nmp->nm_mountp, dp->nfsdl_fh,
3408                     dp->nfsdl_fhlen, p, &np);
3409                 if (ret) {
3410                         /*
3411                          * File isn't open, so nothing to move over to the
3412                          * server.
3413                          */
3414                         return (0);
3415                 }
3416                 vp = NFSTOV(np);
3417                 gotvp = 1;
3418         } else {
3419                 np = VTONFS(vp);
3420         }
3421         dp->nfsdl_flags &= ~NFSCLDL_MODTIMESET;
3422
3423         /*
3424          * Ok, if it's a write delegation, flush data to the server, so
3425          * that close/open consistency is retained.
3426          */
3427         ret = 0;
3428         NFSLOCKNODE(np);
3429         if ((dp->nfsdl_flags & NFSCLDL_WRITE) && (np->n_flag & NMODIFIED)) {
3430                 np->n_flag |= NDELEGRECALL;
3431                 NFSUNLOCKNODE(np);
3432                 ret = ncl_flush(vp, MNT_WAIT, cred, p, 1,
3433                     called_from_renewthread);
3434                 NFSLOCKNODE(np);
3435                 np->n_flag &= ~NDELEGRECALL;
3436         }
3437         NFSINVALATTRCACHE(np);
3438         NFSUNLOCKNODE(np);
3439         if (ret == EIO && called_from_renewthread != 0) {
3440                 /*
3441                  * If the flush failed with EIO for the renew thread,
3442                  * return now, so that the dirty buffer will be flushed
3443                  * later.
3444                  */
3445                 if (gotvp != 0)
3446                         vrele(vp);
3447                 return (ret);
3448         }
3449
3450         /*
3451          * Now, for each openowner with opens issued locally, move them
3452          * over to state against the server.
3453          */
3454         LIST_FOREACH(lowp, &dp->nfsdl_owner, nfsow_list) {
3455                 lop = LIST_FIRST(&lowp->nfsow_open);
3456                 if (lop != NULL) {
3457                         if (LIST_NEXT(lop, nfso_list) != NULL)
3458                                 panic("nfsdlg mult opens");
3459                         /*
3460                          * Look for the same openowner against the server.
3461                          */
3462                         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3463                                 if (!NFSBCMP(lowp->nfsow_owner,
3464                                     owp->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
3465                                         newnfs_copycred(&dp->nfsdl_cred, cred);
3466                                         ret = nfscl_moveopen(vp, clp, nmp, lop,
3467                                             owp, dp, cred, p);
3468                                         if (ret == NFSERR_STALECLIENTID ||
3469                                             ret == NFSERR_STALEDONTRECOVER) {
3470                                                 if (gotvp)
3471                                                         vrele(vp);
3472                                                 return (ret);
3473                                         }
3474                                         if (ret) {
3475                                                 nfscl_freeopen(lop, 1);
3476                                                 if (!error)
3477                                                         error = ret;
3478                                         }
3479                                         break;
3480                                 }
3481                         }
3482
3483                         /*
3484                          * If no openowner found, create one and get an open
3485                          * for it.
3486                          */
3487                         if (owp == NULL) {
3488                                 MALLOC(nowp, struct nfsclowner *,
3489                                     sizeof (struct nfsclowner), M_NFSCLOWNER,
3490                                     M_WAITOK);
3491                                 nfscl_newopen(clp, NULL, &owp, &nowp, &op,
3492                                     NULL, lowp->nfsow_owner, dp->nfsdl_fh,
3493                                     dp->nfsdl_fhlen, NULL);
3494                                 newnfs_copycred(&dp->nfsdl_cred, cred);
3495                                 ret = nfscl_moveopen(vp, clp, nmp, lop,
3496                                     owp, dp, cred, p);
3497                                 if (ret) {
3498                                         nfscl_freeopenowner(owp, 0);
3499                                         if (ret == NFSERR_STALECLIENTID ||
3500                                             ret == NFSERR_STALEDONTRECOVER) {
3501                                                 if (gotvp)
3502                                                         vrele(vp);
3503                                                 return (ret);
3504                                         }
3505                                         if (ret) {
3506                                                 nfscl_freeopen(lop, 1);
3507                                                 if (!error)
3508                                                         error = ret;
3509                                         }
3510                                 }
3511                         }
3512                 }
3513         }
3514
3515         /*
3516          * Now, get byte range locks for any locks done locally.
3517          */
3518         LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
3519                 LIST_FOREACH(lckp, &lp->nfsl_lock, nfslo_list) {
3520                         newnfs_copycred(&dp->nfsdl_cred, cred);
3521                         ret = nfscl_relock(vp, clp, nmp, lp, lckp, cred, p);
3522                         if (ret == NFSERR_STALESTATEID ||
3523                             ret == NFSERR_STALEDONTRECOVER ||
3524                             ret == NFSERR_STALECLIENTID) {
3525                                 if (gotvp)
3526                                         vrele(vp);
3527                                 return (ret);
3528                         }
3529                         if (ret && !error)
3530                                 error = ret;
3531                 }
3532         }
3533         if (gotvp)
3534                 vrele(vp);
3535         return (error);
3536 }
3537
3538 /*
3539  * Move a locally issued open over to an owner on the state list.
3540  * SIDE EFFECT: If it needs to sleep (do an rpc), it unlocks clstate and
3541  * returns with it unlocked.
3542  */
3543 static int
3544 nfscl_moveopen(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp,
3545     struct nfsclopen *lop, struct nfsclowner *owp, struct nfscldeleg *dp,
3546     struct ucred *cred, NFSPROC_T *p)
3547 {
3548         struct nfsclopen *op, *nop;
3549         struct nfscldeleg *ndp;
3550         struct nfsnode *np;
3551         int error = 0, newone;
3552
3553         /*
3554          * First, look for an appropriate open, If found, just increment the
3555          * opencnt in it.
3556          */
3557         LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3558                 if ((op->nfso_mode & lop->nfso_mode) == lop->nfso_mode &&
3559                     op->nfso_fhlen == lop->nfso_fhlen &&
3560                     !NFSBCMP(op->nfso_fh, lop->nfso_fh, op->nfso_fhlen)) {
3561                         op->nfso_opencnt += lop->nfso_opencnt;
3562                         nfscl_freeopen(lop, 1);
3563                         return (0);
3564                 }
3565         }
3566
3567         /* No appropriate open, so we have to do one against the server. */
3568         np = VTONFS(vp);
3569         MALLOC(nop, struct nfsclopen *, sizeof (struct nfsclopen) +
3570             lop->nfso_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
3571         newone = 0;
3572         nfscl_newopen(clp, NULL, &owp, NULL, &op, &nop, owp->nfsow_owner,
3573             lop->nfso_fh, lop->nfso_fhlen, &newone);
3574         ndp = dp;
3575         error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data, np->n_v4->n4_fhlen,
3576             lop->nfso_fh, lop->nfso_fhlen, lop->nfso_mode, op,
3577             NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, &ndp, 0, 0, cred, p);
3578         if (error) {
3579                 if (newone)
3580                         nfscl_freeopen(op, 0);
3581         } else {
3582                 if (newone)
3583                         newnfs_copyincred(cred, &op->nfso_cred);
3584                 op->nfso_mode |= lop->nfso_mode;
3585                 op->nfso_opencnt += lop->nfso_opencnt;
3586                 nfscl_freeopen(lop, 1);
3587         }
3588         if (nop != NULL)
3589                 FREE((caddr_t)nop, M_NFSCLOPEN);
3590         if (ndp != NULL) {
3591                 /*
3592                  * What should I do with the returned delegation, since the
3593                  * delegation is being recalled? For now, just printf and
3594                  * through it away.
3595                  */
3596                 printf("Moveopen returned deleg\n");
3597                 FREE((caddr_t)ndp, M_NFSCLDELEG);
3598         }
3599         return (error);
3600 }
3601
3602 /*
3603  * Recall all delegations on this client.
3604  */
3605 static void
3606 nfscl_totalrecall(struct nfsclclient *clp)
3607 {
3608         struct nfscldeleg *dp;
3609
3610         TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
3611                 if ((dp->nfsdl_flags & NFSCLDL_DELEGRET) == 0)
3612                         dp->nfsdl_flags |= NFSCLDL_RECALL;
3613         }
3614 }
3615
3616 /*
3617  * Relock byte ranges. Called for delegation recall and state expiry.
3618  */
3619 static int
3620 nfscl_relock(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp,
3621     struct nfscllockowner *lp, struct nfscllock *lop, struct ucred *cred,
3622     NFSPROC_T *p)
3623 {
3624         struct nfscllockowner *nlp;
3625         struct nfsfh *nfhp;
3626         u_int64_t off, len;
3627         u_int32_t clidrev = 0;
3628         int error, newone, donelocally;
3629
3630         off = lop->nfslo_first;
3631         len = lop->nfslo_end - lop->nfslo_first;
3632         error = nfscl_getbytelock(vp, off, len, lop->nfslo_type, cred, p,
3633             clp, 1, NULL, 0, lp->nfsl_owner, lp->nfsl_openowner, &nlp, &newone,
3634             &donelocally);
3635         if (error || donelocally)
3636                 return (error);
3637         if (nmp->nm_clp != NULL)
3638                 clidrev = nmp->nm_clp->nfsc_clientidrev;
3639         else
3640                 clidrev = 0;
3641         nfhp = VTONFS(vp)->n_fhp;
3642         error = nfscl_trylock(nmp, vp, nfhp->nfh_fh,
3643             nfhp->nfh_len, nlp, newone, 0, off,
3644             len, lop->nfslo_type, cred, p);
3645         if (error)
3646                 nfscl_freelockowner(nlp, 0);
3647         return (error);
3648 }
3649
3650 /*
3651  * Called to re-open a file. Basically get a vnode for the file handle
3652  * and then call nfsrpc_openrpc() to do the rest.
3653  */
3654 static int
3655 nfsrpc_reopen(struct nfsmount *nmp, u_int8_t *fhp, int fhlen,
3656     u_int32_t mode, struct nfsclopen *op, struct nfscldeleg **dpp,
3657     struct ucred *cred, NFSPROC_T *p)
3658 {
3659         struct nfsnode *np;
3660         vnode_t vp;
3661         int error;
3662
3663         error = nfscl_ngetreopen(nmp->nm_mountp, fhp, fhlen, p, &np);
3664         if (error)
3665                 return (error);
3666         vp = NFSTOV(np);
3667         if (np->n_v4 != NULL) {
3668                 error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data,
3669                     np->n_v4->n4_fhlen, fhp, fhlen, mode, op,
3670                     NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, dpp, 0, 0,
3671                     cred, p);
3672         } else {
3673                 error = EINVAL;
3674         }
3675         vrele(vp);
3676         return (error);
3677 }
3678
3679 /*
3680  * Try an open against the server. Just call nfsrpc_openrpc(), retrying while
3681  * NFSERR_DELAY. Also, try system credentials, if the passed in credentials
3682  * fail.
3683  */
3684 static int
3685 nfscl_tryopen(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen,
3686     u_int8_t *newfhp, int newfhlen, u_int32_t mode, struct nfsclopen *op,
3687     u_int8_t *name, int namelen, struct nfscldeleg **ndpp,
3688     int reclaim, u_int32_t delegtype, struct ucred *cred, NFSPROC_T *p)
3689 {
3690         int error;
3691
3692         do {
3693                 error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp, newfhlen,
3694                     mode, op, name, namelen, ndpp, reclaim, delegtype, cred, p,
3695                     0, 0);
3696                 if (error == NFSERR_DELAY)
3697                         (void) nfs_catnap(PZERO, error, "nfstryop");
3698         } while (error == NFSERR_DELAY);
3699         if (error == EAUTH || error == EACCES) {
3700                 /* Try again using system credentials */
3701                 newnfs_setroot(cred);
3702                 do {
3703                     error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp,
3704                         newfhlen, mode, op, name, namelen, ndpp, reclaim,
3705                         delegtype, cred, p, 1, 0);
3706                     if (error == NFSERR_DELAY)
3707                         (void) nfs_catnap(PZERO, error, "nfstryop");
3708                 } while (error == NFSERR_DELAY);
3709         }
3710         return (error);
3711 }
3712
3713 /*
3714  * Try a byte range lock. Just loop on nfsrpc_lock() while it returns
3715  * NFSERR_DELAY. Also, retry with system credentials, if the provided
3716  * cred don't work.
3717  */
3718 static int
3719 nfscl_trylock(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp,
3720     int fhlen, struct nfscllockowner *nlp, int newone, int reclaim,
3721     u_int64_t off, u_int64_t len, short type, struct ucred *cred, NFSPROC_T *p)
3722 {
3723         struct nfsrv_descript nfsd, *nd = &nfsd;
3724         int error;
3725
3726         do {
3727                 error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp, newone,
3728                     reclaim, off, len, type, cred, p, 0);
3729                 if (!error && nd->nd_repstat == NFSERR_DELAY)
3730                         (void) nfs_catnap(PZERO, (int)nd->nd_repstat,
3731                             "nfstrylck");
3732         } while (!error && nd->nd_repstat == NFSERR_DELAY);
3733         if (!error)
3734                 error = nd->nd_repstat;
3735         if (error == EAUTH || error == EACCES) {
3736                 /* Try again using root credentials */
3737                 newnfs_setroot(cred);
3738                 do {
3739                         error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp,
3740                             newone, reclaim, off, len, type, cred, p, 1);
3741                         if (!error && nd->nd_repstat == NFSERR_DELAY)
3742                                 (void) nfs_catnap(PZERO, (int)nd->nd_repstat,
3743                                     "nfstrylck");
3744                 } while (!error && nd->nd_repstat == NFSERR_DELAY);
3745                 if (!error)
3746                         error = nd->nd_repstat;
3747         }
3748         return (error);
3749 }
3750
3751 /*
3752  * Try a delegreturn against the server. Just call nfsrpc_delegreturn(),
3753  * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
3754  * credentials fail.
3755  */
3756 static int
3757 nfscl_trydelegreturn(struct nfscldeleg *dp, struct ucred *cred,
3758     struct nfsmount *nmp, NFSPROC_T *p)
3759 {
3760         int error;
3761
3762         do {
3763                 error = nfsrpc_delegreturn(dp, cred, nmp, p, 0);
3764                 if (error == NFSERR_DELAY)
3765                         (void) nfs_catnap(PZERO, error, "nfstrydp");
3766         } while (error == NFSERR_DELAY);
3767         if (error == EAUTH || error == EACCES) {
3768                 /* Try again using system credentials */
3769                 newnfs_setroot(cred);
3770                 do {
3771                         error = nfsrpc_delegreturn(dp, cred, nmp, p, 1);
3772                         if (error == NFSERR_DELAY)
3773                                 (void) nfs_catnap(PZERO, error, "nfstrydp");
3774                 } while (error == NFSERR_DELAY);
3775         }
3776         return (error);
3777 }
3778
3779 /*
3780  * Try a close against the server. Just call nfsrpc_closerpc(),
3781  * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
3782  * credentials fail.
3783  */
3784 APPLESTATIC int
3785 nfscl_tryclose(struct nfsclopen *op, struct ucred *cred,
3786     struct nfsmount *nmp, NFSPROC_T *p)
3787 {
3788         struct nfsrv_descript nfsd, *nd = &nfsd;
3789         int error;
3790
3791         do {
3792                 error = nfsrpc_closerpc(nd, nmp, op, cred, p, 0);
3793                 if (error == NFSERR_DELAY)
3794                         (void) nfs_catnap(PZERO, error, "nfstrycl");
3795         } while (error == NFSERR_DELAY);
3796         if (error == EAUTH || error == EACCES) {
3797                 /* Try again using system credentials */
3798                 newnfs_setroot(cred);
3799                 do {
3800                         error = nfsrpc_closerpc(nd, nmp, op, cred, p, 1);
3801                         if (error == NFSERR_DELAY)
3802                                 (void) nfs_catnap(PZERO, error, "nfstrycl");
3803                 } while (error == NFSERR_DELAY);
3804         }
3805         return (error);
3806 }
3807
3808 /*
3809  * Decide if a delegation on a file permits close without flushing writes
3810  * to the server. This might be a big performance win in some environments.
3811  * (Not useful until the client does caching on local stable storage.)
3812  */
3813 APPLESTATIC int
3814 nfscl_mustflush(vnode_t vp)
3815 {
3816         struct nfsclclient *clp;
3817         struct nfscldeleg *dp;
3818         struct nfsnode *np;
3819         struct nfsmount *nmp;
3820
3821         np = VTONFS(vp);
3822         nmp = VFSTONFS(vnode_mount(vp));
3823         if (!NFSHASNFSV4(nmp))
3824                 return (1);
3825         NFSLOCKCLSTATE();
3826         clp = nfscl_findcl(nmp);
3827         if (clp == NULL) {
3828                 NFSUNLOCKCLSTATE();
3829                 return (1);
3830         }
3831         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
3832         if (dp != NULL && (dp->nfsdl_flags &
3833             (NFSCLDL_WRITE | NFSCLDL_RECALL | NFSCLDL_DELEGRET)) ==
3834              NFSCLDL_WRITE &&
3835             (dp->nfsdl_sizelimit >= np->n_size ||
3836              !NFSHASSTRICT3530(nmp))) {
3837                 NFSUNLOCKCLSTATE();
3838                 return (0);
3839         }
3840         NFSUNLOCKCLSTATE();
3841         return (1);
3842 }
3843
3844 /*
3845  * See if a (write) delegation exists for this file.
3846  */
3847 APPLESTATIC int
3848 nfscl_nodeleg(vnode_t vp, int writedeleg)
3849 {
3850         struct nfsclclient *clp;
3851         struct nfscldeleg *dp;
3852         struct nfsnode *np;
3853         struct nfsmount *nmp;
3854
3855         np = VTONFS(vp);
3856         nmp = VFSTONFS(vnode_mount(vp));
3857         if (!NFSHASNFSV4(nmp))
3858                 return (1);
3859         NFSLOCKCLSTATE();
3860         clp = nfscl_findcl(nmp);
3861         if (clp == NULL) {
3862                 NFSUNLOCKCLSTATE();
3863                 return (1);
3864         }
3865         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
3866         if (dp != NULL &&
3867             (dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_DELEGRET)) == 0 &&
3868             (writedeleg == 0 || (dp->nfsdl_flags & NFSCLDL_WRITE) ==
3869              NFSCLDL_WRITE)) {
3870                 NFSUNLOCKCLSTATE();
3871                 return (0);
3872         }
3873         NFSUNLOCKCLSTATE();
3874         return (1);
3875 }
3876
3877 /*
3878  * Look for an associated delegation that should be DelegReturned.
3879  */
3880 APPLESTATIC int
3881 nfscl_removedeleg(vnode_t vp, NFSPROC_T *p, nfsv4stateid_t *stp)
3882 {
3883         struct nfsclclient *clp;
3884         struct nfscldeleg *dp;
3885         struct nfsclowner *owp;
3886         struct nfscllockowner *lp;
3887         struct nfsmount *nmp;
3888         struct ucred *cred;
3889         struct nfsnode *np;
3890         int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
3891
3892         nmp = VFSTONFS(vnode_mount(vp));
3893         np = VTONFS(vp);
3894         NFSLOCKCLSTATE();
3895         /*
3896          * Loop around waiting for:
3897          * - outstanding I/O operations on delegations to complete
3898          * - for a delegation on vp that has state, lock the client and
3899          *   do a recall
3900          * - return delegation with no state
3901          */
3902         while (1) {
3903                 clp = nfscl_findcl(nmp);
3904                 if (clp == NULL) {
3905                         NFSUNLOCKCLSTATE();
3906                         return (retcnt);
3907                 }
3908                 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
3909                     np->n_fhp->nfh_len);
3910                 if (dp != NULL) {
3911                     /*
3912                      * Wait for outstanding I/O ops to be done.
3913                      */
3914                     if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
3915                         if (igotlock) {
3916                             nfsv4_unlock(&clp->nfsc_lock, 0);
3917                             igotlock = 0;
3918                         }
3919                         dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
3920                         (void) nfsmsleep(&dp->nfsdl_rwlock,
3921                             NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
3922                         continue;
3923                     }
3924                     needsrecall = 0;
3925                     LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
3926                         if (!LIST_EMPTY(&owp->nfsow_open)) {
3927                             needsrecall = 1;
3928                             break;
3929                         }
3930                     }
3931                     if (!needsrecall) {
3932                         LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
3933                             if (!LIST_EMPTY(&lp->nfsl_lock)) {
3934                                 needsrecall = 1;
3935                                 break;
3936                             }
3937                         }
3938                     }
3939                     if (needsrecall && !triedrecall) {
3940                         dp->nfsdl_flags |= NFSCLDL_DELEGRET;
3941                         islept = 0;
3942                         while (!igotlock) {
3943                             igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
3944                                 &islept, NFSCLSTATEMUTEXPTR, NULL);
3945                             if (islept)
3946                                 break;
3947                         }
3948                         if (islept)
3949                             continue;
3950                         NFSUNLOCKCLSTATE();
3951                         cred = newnfs_getcred();
3952                         newnfs_copycred(&dp->nfsdl_cred, cred);
3953                         (void) nfscl_recalldeleg(clp, nmp, dp, vp, cred, p, 0);
3954                         NFSFREECRED(cred);
3955                         triedrecall = 1;
3956                         NFSLOCKCLSTATE();
3957                         nfsv4_unlock(&clp->nfsc_lock, 0);
3958                         igotlock = 0;
3959                         continue;
3960                     }
3961                     *stp = dp->nfsdl_stateid;
3962                     retcnt = 1;
3963                     nfscl_cleandeleg(dp);
3964                     nfscl_freedeleg(&clp->nfsc_deleg, dp);
3965                 }
3966                 if (igotlock)
3967                     nfsv4_unlock(&clp->nfsc_lock, 0);
3968                 NFSUNLOCKCLSTATE();
3969                 return (retcnt);
3970         }
3971 }
3972
3973 /*
3974  * Look for associated delegation(s) that should be DelegReturned.
3975  */
3976 APPLESTATIC int
3977 nfscl_renamedeleg(vnode_t fvp, nfsv4stateid_t *fstp, int *gotfdp, vnode_t tvp,
3978     nfsv4stateid_t *tstp, int *gottdp, NFSPROC_T *p)
3979 {
3980         struct nfsclclient *clp;
3981         struct nfscldeleg *dp;
3982         struct nfsclowner *owp;
3983         struct nfscllockowner *lp;
3984         struct nfsmount *nmp;
3985         struct ucred *cred;
3986         struct nfsnode *np;
3987         int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
3988
3989         nmp = VFSTONFS(vnode_mount(fvp));
3990         *gotfdp = 0;
3991         *gottdp = 0;
3992         NFSLOCKCLSTATE();
3993         /*
3994          * Loop around waiting for:
3995          * - outstanding I/O operations on delegations to complete
3996          * - for a delegation on fvp that has state, lock the client and
3997          *   do a recall
3998          * - return delegation(s) with no state.
3999          */
4000         while (1) {
4001                 clp = nfscl_findcl(nmp);
4002                 if (clp == NULL) {
4003                         NFSUNLOCKCLSTATE();
4004                         return (retcnt);
4005                 }
4006                 np = VTONFS(fvp);
4007                 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4008                     np->n_fhp->nfh_len);
4009                 if (dp != NULL && *gotfdp == 0) {
4010                     /*
4011                      * Wait for outstanding I/O ops to be done.
4012                      */
4013                     if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4014                         if (igotlock) {
4015                             nfsv4_unlock(&clp->nfsc_lock, 0);
4016                             igotlock = 0;
4017                         }
4018                         dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4019                         (void) nfsmsleep(&dp->nfsdl_rwlock,
4020                             NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4021                         continue;
4022                     }
4023                     needsrecall = 0;
4024                     LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4025                         if (!LIST_EMPTY(&owp->nfsow_open)) {
4026                             needsrecall = 1;
4027                             break;
4028                         }
4029                     }
4030                     if (!needsrecall) {
4031                         LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4032                             if (!LIST_EMPTY(&lp->nfsl_lock)) {
4033                                 needsrecall = 1;
4034                                 break;
4035                             }
4036                         }
4037                     }
4038                     if (needsrecall && !triedrecall) {
4039                         dp->nfsdl_flags |= NFSCLDL_DELEGRET;
4040                         islept = 0;
4041                         while (!igotlock) {
4042                             igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
4043                                 &islept, NFSCLSTATEMUTEXPTR, NULL);
4044                             if (islept)
4045                                 break;
4046                         }
4047                         if (islept)
4048                             continue;
4049                         NFSUNLOCKCLSTATE();
4050                         cred = newnfs_getcred();
4051                         newnfs_copycred(&dp->nfsdl_cred, cred);
4052                         (void) nfscl_recalldeleg(clp, nmp, dp, fvp, cred, p, 0);
4053                         NFSFREECRED(cred);
4054                         triedrecall = 1;
4055                         NFSLOCKCLSTATE();
4056                         nfsv4_unlock(&clp->nfsc_lock, 0);
4057                         igotlock = 0;
4058                         continue;
4059                     }
4060                     *fstp = dp->nfsdl_stateid;
4061                     retcnt++;
4062                     *gotfdp = 1;
4063                     nfscl_cleandeleg(dp);
4064                     nfscl_freedeleg(&clp->nfsc_deleg, dp);
4065                 }
4066                 if (igotlock) {
4067                     nfsv4_unlock(&clp->nfsc_lock, 0);
4068                     igotlock = 0;
4069                 }
4070                 if (tvp != NULL) {
4071                     np = VTONFS(tvp);
4072                     dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4073                         np->n_fhp->nfh_len);
4074                     if (dp != NULL && *gottdp == 0) {
4075                         /*
4076                          * Wait for outstanding I/O ops to be done.
4077                          */
4078                         if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4079                             dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4080                             (void) nfsmsleep(&dp->nfsdl_rwlock,
4081                                 NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4082                             continue;
4083                         }
4084                         LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4085                             if (!LIST_EMPTY(&owp->nfsow_open)) {
4086                                 NFSUNLOCKCLSTATE();
4087                                 return (retcnt);
4088                             }
4089                         }
4090                         LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4091                             if (!LIST_EMPTY(&lp->nfsl_lock)) {
4092                                 NFSUNLOCKCLSTATE();
4093                                 return (retcnt);
4094                             }
4095                         }
4096                         *tstp = dp->nfsdl_stateid;
4097                         retcnt++;
4098                         *gottdp = 1;
4099                         nfscl_cleandeleg(dp);
4100                         nfscl_freedeleg(&clp->nfsc_deleg, dp);
4101                     }
4102                 }
4103                 NFSUNLOCKCLSTATE();
4104                 return (retcnt);
4105         }
4106 }
4107
4108 /*
4109  * Get a reference on the clientid associated with the mount point.
4110  * Return 1 if success, 0 otherwise.
4111  */
4112 APPLESTATIC int
4113 nfscl_getref(struct nfsmount *nmp)
4114 {
4115         struct nfsclclient *clp;
4116
4117         NFSLOCKCLSTATE();
4118         clp = nfscl_findcl(nmp);
4119         if (clp == NULL) {
4120                 NFSUNLOCKCLSTATE();
4121                 return (0);
4122         }
4123         nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, NULL);
4124         NFSUNLOCKCLSTATE();
4125         return (1);
4126 }
4127
4128 /*
4129  * Release a reference on a clientid acquired with the above call.
4130  */
4131 APPLESTATIC void
4132 nfscl_relref(struct nfsmount *nmp)
4133 {
4134         struct nfsclclient *clp;
4135
4136         NFSLOCKCLSTATE();
4137         clp = nfscl_findcl(nmp);
4138         if (clp == NULL) {
4139                 NFSUNLOCKCLSTATE();
4140                 return;
4141         }
4142         nfsv4_relref(&clp->nfsc_lock);
4143         NFSUNLOCKCLSTATE();
4144 }
4145
4146 /*
4147  * Save the size attribute in the delegation, since the nfsnode
4148  * is going away.
4149  */
4150 APPLESTATIC void
4151 nfscl_reclaimnode(vnode_t vp)
4152 {
4153         struct nfsclclient *clp;
4154         struct nfscldeleg *dp;
4155         struct nfsnode *np = VTONFS(vp);
4156         struct nfsmount *nmp;
4157
4158         nmp = VFSTONFS(vnode_mount(vp));
4159         if (!NFSHASNFSV4(nmp))
4160                 return;
4161         NFSLOCKCLSTATE();
4162         clp = nfscl_findcl(nmp);
4163         if (clp == NULL) {
4164                 NFSUNLOCKCLSTATE();
4165                 return;
4166         }
4167         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4168         if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE))
4169                 dp->nfsdl_size = np->n_size;
4170         NFSUNLOCKCLSTATE();
4171 }
4172
4173 /*
4174  * Get the saved size attribute in the delegation, since it is a
4175  * newly allocated nfsnode.
4176  */
4177 APPLESTATIC void
4178 nfscl_newnode(vnode_t vp)
4179 {
4180         struct nfsclclient *clp;
4181         struct nfscldeleg *dp;
4182         struct nfsnode *np = VTONFS(vp);
4183         struct nfsmount *nmp;
4184
4185         nmp = VFSTONFS(vnode_mount(vp));
4186         if (!NFSHASNFSV4(nmp))
4187                 return;
4188         NFSLOCKCLSTATE();
4189         clp = nfscl_findcl(nmp);
4190         if (clp == NULL) {
4191                 NFSUNLOCKCLSTATE();
4192                 return;
4193         }
4194         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4195         if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE))
4196                 np->n_size = dp->nfsdl_size;
4197         NFSUNLOCKCLSTATE();
4198 }
4199
4200 /*
4201  * If there is a valid write delegation for this file, set the modtime
4202  * to the local clock time.
4203  */
4204 APPLESTATIC void
4205 nfscl_delegmodtime(vnode_t vp)
4206 {
4207         struct nfsclclient *clp;
4208         struct nfscldeleg *dp;
4209         struct nfsnode *np = VTONFS(vp);
4210         struct nfsmount *nmp;
4211
4212         nmp = VFSTONFS(vnode_mount(vp));
4213         if (!NFSHASNFSV4(nmp))
4214                 return;
4215         NFSLOCKCLSTATE();
4216         clp = nfscl_findcl(nmp);
4217         if (clp == NULL) {
4218                 NFSUNLOCKCLSTATE();
4219                 return;
4220         }
4221         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4222         if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE)) {
4223                 NFSGETNANOTIME(&dp->nfsdl_modtime);
4224                 dp->nfsdl_flags |= NFSCLDL_MODTIMESET;
4225         }
4226         NFSUNLOCKCLSTATE();
4227 }
4228
4229 /*
4230  * If there is a valid write delegation for this file with a modtime set,
4231  * put that modtime in mtime.
4232  */
4233 APPLESTATIC void
4234 nfscl_deleggetmodtime(vnode_t vp, struct timespec *mtime)
4235 {
4236         struct nfsclclient *clp;
4237         struct nfscldeleg *dp;
4238         struct nfsnode *np = VTONFS(vp);
4239         struct nfsmount *nmp;
4240
4241         nmp = VFSTONFS(vnode_mount(vp));
4242         if (!NFSHASNFSV4(nmp))
4243                 return;
4244         NFSLOCKCLSTATE();
4245         clp = nfscl_findcl(nmp);
4246         if (clp == NULL) {
4247                 NFSUNLOCKCLSTATE();
4248                 return;
4249         }
4250         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4251         if (dp != NULL &&
4252             (dp->nfsdl_flags & (NFSCLDL_WRITE | NFSCLDL_MODTIMESET)) ==
4253             (NFSCLDL_WRITE | NFSCLDL_MODTIMESET))
4254                 *mtime = dp->nfsdl_modtime;
4255         NFSUNLOCKCLSTATE();
4256 }
4257
4258 static int
4259 nfscl_errmap(struct nfsrv_descript *nd)
4260 {
4261         short *defaulterrp, *errp;
4262
4263         if (!nd->nd_repstat)
4264                 return (0);
4265         if (nd->nd_procnum == NFSPROC_NOOP)
4266                 return (txdr_unsigned(nd->nd_repstat & 0xffff));
4267         if (nd->nd_repstat == EBADRPC)
4268                 return (txdr_unsigned(NFSERR_BADXDR));
4269         if (nd->nd_repstat == NFSERR_MINORVERMISMATCH ||
4270             nd->nd_repstat == NFSERR_OPILLEGAL)
4271                 return (txdr_unsigned(nd->nd_repstat));
4272         errp = defaulterrp = nfscl_cberrmap[nd->nd_procnum];
4273         while (*++errp)
4274                 if (*errp == (short)nd->nd_repstat)
4275                         return (txdr_unsigned(nd->nd_repstat));
4276         return (txdr_unsigned(*defaulterrp));
4277 }
4278