sys/fs/nfsclient/nfs_clstate.c

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