sys/fs/nfsclient/nfs_clstate.c

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