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 #include <fs/nfs/nfsport.h>
  84
  85 /*
  86  * Global variables
  87  */
  88 extern struct nfsstatsv1 nfsstatsv1;
  89 extern struct nfsreqhead nfsd_reqq;
  90 extern u_int32_t newnfs_false, newnfs_true;
  91 extern int nfscl_debuglevel;
  92 extern int nfscl_enablecallb;
  93 extern int nfs_numnfscbd;
  94 NFSREQSPINLOCK;
  95 NFSCLSTATEMUTEX;
  96 int nfscl_inited = 0;
  97 struct nfsclhead nfsclhead;     /* Head of clientid list */
  98 int nfscl_deleghighwater = NFSCLDELEGHIGHWATER;
  99 int nfscl_layouthighwater = NFSCLLAYOUTHIGHWATER;
 100
 101 static int nfscl_delegcnt = 0;
 102 static int nfscl_layoutcnt = 0;
 103 static int nfscl_getopen(struct nfsclownerhead *, u_int8_t *, int, u_int8_t *,
 104     u_int8_t *, u_int32_t, struct nfscllockowner **, struct nfsclopen **);
 105 static void nfscl_clrelease(struct nfsclclient *);
 106 static void nfscl_cleanclient(struct nfsclclient *);
 107 static void nfscl_expireclient(struct nfsclclient *, struct nfsmount *,
 108     struct ucred *, NFSPROC_T *);
 109 static int nfscl_expireopen(struct nfsclclient *, struct nfsclopen *,
 110     struct nfsmount *, struct ucred *, NFSPROC_T *);
 111 static void nfscl_recover(struct nfsclclient *, struct ucred *, NFSPROC_T *);
 112 static void nfscl_insertlock(struct nfscllockowner *, struct nfscllock *,
 113     struct nfscllock *, int);
 114 static int nfscl_updatelock(struct nfscllockowner *, struct nfscllock **,
 115     struct nfscllock **, int);
 116 static void nfscl_delegreturnall(struct nfsclclient *, NFSPROC_T *);
 117 static u_int32_t nfscl_nextcbident(void);
 118 static mount_t nfscl_getmnt(int, uint8_t *, u_int32_t, struct nfsclclient **);
 119 static struct nfsclclient *nfscl_getclnt(u_int32_t);
 120 static struct nfsclclient *nfscl_getclntsess(uint8_t *);
 121 static struct nfscldeleg *nfscl_finddeleg(struct nfsclclient *, u_int8_t *,
 122     int);
 123 static void nfscl_retoncloselayout(vnode_t, struct nfsclclient *, uint8_t *,
 124     int, struct nfsclrecalllayout **);
 125 static void nfscl_reldevinfo_locked(struct nfscldevinfo *);
 126 static struct nfscllayout *nfscl_findlayout(struct nfsclclient *, u_int8_t *,
 127     int);
 128 static struct nfscldevinfo *nfscl_finddevinfo(struct nfsclclient *, uint8_t *);
 129 static int nfscl_checkconflict(struct nfscllockownerhead *, struct nfscllock *,
 130     u_int8_t *, struct nfscllock **);
 131 static void nfscl_freealllocks(struct nfscllockownerhead *, int);
 132 static int nfscl_localconflict(struct nfsclclient *, u_int8_t *, int,
 133     struct nfscllock *, u_int8_t *, struct nfscldeleg *, struct nfscllock **);
 134 static void nfscl_newopen(struct nfsclclient *, struct nfscldeleg *,
 135     struct nfsclowner **, struct nfsclowner **, struct nfsclopen **,
 136     struct nfsclopen **, u_int8_t *, u_int8_t *, int, struct ucred *, int *);
 137 static int nfscl_moveopen(vnode_t , struct nfsclclient *,
 138     struct nfsmount *, struct nfsclopen *, struct nfsclowner *,
 139     struct nfscldeleg *, struct ucred *, NFSPROC_T *);
 140 static void nfscl_totalrecall(struct nfsclclient *);
 141 static int nfscl_relock(vnode_t , struct nfsclclient *, struct nfsmount *,
 142     struct nfscllockowner *, struct nfscllock *, struct ucred *, NFSPROC_T *);
 143 static int nfscl_tryopen(struct nfsmount *, vnode_t , u_int8_t *, int,
 144     u_int8_t *, int, u_int32_t, struct nfsclopen *, u_int8_t *, int,
 145     struct nfscldeleg **, int, u_int32_t, struct ucred *, NFSPROC_T *);
 146 static int nfscl_trylock(struct nfsmount *, vnode_t , u_int8_t *,
 147     int, struct nfscllockowner *, int, int, u_int64_t, u_int64_t, short,
 148     struct ucred *, NFSPROC_T *);
 149 static int nfsrpc_reopen(struct nfsmount *, u_int8_t *, int, u_int32_t,
 150     struct nfsclopen *, struct nfscldeleg **, struct ucred *, NFSPROC_T *);
 151 static void nfscl_freedeleg(struct nfscldeleghead *, struct nfscldeleg *,
 152     bool);
 153 static int nfscl_errmap(struct nfsrv_descript *, u_int32_t);
 154 static void nfscl_cleanup_common(struct nfsclclient *, u_int8_t *);
 155 static int nfscl_recalldeleg(struct nfsclclient *, struct nfsmount *,
 156     struct nfscldeleg *, vnode_t, struct ucred *, NFSPROC_T *, int,
 157     vnode_t *);
 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 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 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 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 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 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 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 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 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 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 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 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 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 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 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 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 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, bool freeit)
1615 {
1616
1617         TAILQ_REMOVE(hdp, dp, nfsdl_list);
1618         LIST_REMOVE(dp, nfsdl_hash);
1619         if (freeit)
1620                 free(dp, M_NFSCLDELEG);
1621         nfsstatsv1.cldelegates--;
1622         nfscl_delegcnt--;
1623 }
1624
1625 /*
1626  * Free up all state related to this client structure.
1627  */
1628 static void
1629 nfscl_cleanclient(struct nfsclclient *clp)
1630 {
1631         struct nfsclowner *owp, *nowp;
1632         struct nfsclopen *op, *nop;
1633         struct nfscllayout *lyp, *nlyp;
1634         struct nfscldevinfo *dip, *ndip;
1635
1636         TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp)
1637                 nfscl_freelayout(lyp);
1638
1639         LIST_FOREACH_SAFE(dip, &clp->nfsc_devinfo, nfsdi_list, ndip)
1640                 nfscl_freedevinfo(dip);
1641
1642         /* Now, all the OpenOwners, etc. */
1643         LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1644                 LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
1645                         nfscl_freeopen(op, 0);
1646                 }
1647                 nfscl_freeopenowner(owp, 0);
1648         }
1649 }
1650
1651 /*
1652  * Called when an NFSERR_EXPIRED is received from the server.
1653  */
1654 static void
1655 nfscl_expireclient(struct nfsclclient *clp, struct nfsmount *nmp,
1656     struct ucred *cred, NFSPROC_T *p)
1657 {
1658         struct nfsclowner *owp, *nowp, *towp;
1659         struct nfsclopen *op, *nop, *top;
1660         struct nfscldeleg *dp, *ndp;
1661         int ret, printed = 0;
1662
1663         /*
1664          * First, merge locally issued Opens into the list for the server.
1665          */
1666         dp = TAILQ_FIRST(&clp->nfsc_deleg);
1667         while (dp != NULL) {
1668             ndp = TAILQ_NEXT(dp, nfsdl_list);
1669             owp = LIST_FIRST(&dp->nfsdl_owner);
1670             while (owp != NULL) {
1671                 nowp = LIST_NEXT(owp, nfsow_list);
1672                 op = LIST_FIRST(&owp->nfsow_open);
1673                 if (op != NULL) {
1674                     if (LIST_NEXT(op, nfso_list) != NULL)
1675                         panic("nfsclexp");
1676                     LIST_FOREACH(towp, &clp->nfsc_owner, nfsow_list) {
1677                         if (!NFSBCMP(towp->nfsow_owner, owp->nfsow_owner,
1678                             NFSV4CL_LOCKNAMELEN))
1679                             break;
1680                     }
1681                     if (towp != NULL) {
1682                         /* Merge opens in */
1683                         LIST_FOREACH(top, &towp->nfsow_open, nfso_list) {
1684                             if (top->nfso_fhlen == op->nfso_fhlen &&
1685                                 !NFSBCMP(top->nfso_fh, op->nfso_fh,
1686                                  op->nfso_fhlen)) {
1687                                 top->nfso_mode |= op->nfso_mode;
1688                                 top->nfso_opencnt += op->nfso_opencnt;
1689                                 break;
1690                             }
1691                         }
1692                         if (top == NULL) {
1693                             /* Just add the open to the owner list */
1694                             LIST_REMOVE(op, nfso_list);
1695                             op->nfso_own = towp;
1696                             LIST_INSERT_HEAD(&towp->nfsow_open, op, nfso_list);
1697                             nfsstatsv1.cllocalopens--;
1698                             nfsstatsv1.clopens++;
1699                         }
1700                     } else {
1701                         /* Just add the openowner to the client list */
1702                         LIST_REMOVE(owp, nfsow_list);
1703                         owp->nfsow_clp = clp;
1704                         LIST_INSERT_HEAD(&clp->nfsc_owner, owp, nfsow_list);
1705                         nfsstatsv1.cllocalopenowners--;
1706                         nfsstatsv1.clopenowners++;
1707                         nfsstatsv1.cllocalopens--;
1708                         nfsstatsv1.clopens++;
1709                     }
1710                 }
1711                 owp = nowp;
1712             }
1713             if (!printed && !LIST_EMPTY(&dp->nfsdl_lock)) {
1714                 printed = 1;
1715                 printf("nfsv4 expired locks lost\n");
1716             }
1717             nfscl_cleandeleg(dp);
1718             nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
1719             dp = ndp;
1720         }
1721         if (!TAILQ_EMPTY(&clp->nfsc_deleg))
1722             panic("nfsclexp");
1723
1724         /*
1725          * Now, try and reopen against the server.
1726          */
1727         LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1728                 owp->nfsow_seqid = 0;
1729                 LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
1730                         ret = nfscl_expireopen(clp, op, nmp, cred, p);
1731                         if (ret && !printed) {
1732                                 printed = 1;
1733                                 printf("nfsv4 expired locks lost\n");
1734                         }
1735                 }
1736                 if (LIST_EMPTY(&owp->nfsow_open))
1737                         nfscl_freeopenowner(owp, 0);
1738         }
1739 }
1740
1741 /*
1742  * This function must be called after the process represented by "own" has
1743  * exited. Must be called with CLSTATE lock held.
1744  */
1745 static void
1746 nfscl_cleanup_common(struct nfsclclient *clp, u_int8_t *own)
1747 {
1748         struct nfsclowner *owp, *nowp;
1749         struct nfscllockowner *lp, *nlp;
1750         struct nfscldeleg *dp;
1751
1752         /* First, get rid of local locks on delegations. */
1753         TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
1754                 LIST_FOREACH_SAFE(lp, &dp->nfsdl_lock, nfsl_list, nlp) {
1755                     if (!NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
1756                         if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED))
1757                             panic("nfscllckw");
1758                         nfscl_freelockowner(lp, 1);
1759                     }
1760                 }
1761         }
1762         owp = LIST_FIRST(&clp->nfsc_owner);
1763         while (owp != NULL) {
1764                 nowp = LIST_NEXT(owp, nfsow_list);
1765                 if (!NFSBCMP(owp->nfsow_owner, own,
1766                     NFSV4CL_LOCKNAMELEN)) {
1767                         /*
1768                          * If there are children that haven't closed the
1769                          * file descriptors yet, the opens will still be
1770                          * here. For that case, let the renew thread clear
1771                          * out the OpenOwner later.
1772                          */
1773                         if (LIST_EMPTY(&owp->nfsow_open))
1774                                 nfscl_freeopenowner(owp, 0);
1775                         else
1776                                 owp->nfsow_defunct = 1;
1777                 }
1778                 owp = nowp;
1779         }
1780 }
1781
1782 /*
1783  * Find open/lock owners for processes that have exited.
1784  */
1785 static void
1786 nfscl_cleanupkext(struct nfsclclient *clp, struct nfscllockownerfhhead *lhp)
1787 {
1788         struct nfsclowner *owp, *nowp;
1789         struct nfsclopen *op;
1790         struct nfscllockowner *lp, *nlp;
1791         struct nfscldeleg *dp;
1792
1793         NFSPROCLISTLOCK();
1794         NFSLOCKCLSTATE();
1795         LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1796                 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1797                         LIST_FOREACH_SAFE(lp, &op->nfso_lock, nfsl_list, nlp) {
1798                                 if (LIST_EMPTY(&lp->nfsl_lock))
1799                                         nfscl_emptylockowner(lp, lhp);
1800                         }
1801                 }
1802                 if (nfscl_procdoesntexist(owp->nfsow_owner))
1803                         nfscl_cleanup_common(clp, owp->nfsow_owner);
1804         }
1805
1806         /*
1807          * For the single open_owner case, these lock owners need to be
1808          * checked to see if they still exist separately.
1809          * This is because nfscl_procdoesntexist() never returns true for
1810          * the single open_owner so that the above doesn't ever call
1811          * nfscl_cleanup_common().
1812          */
1813         TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
1814                 LIST_FOREACH_SAFE(lp, &dp->nfsdl_lock, nfsl_list, nlp) {
1815                         if (nfscl_procdoesntexist(lp->nfsl_owner))
1816                                 nfscl_cleanup_common(clp, lp->nfsl_owner);
1817                 }
1818         }
1819         NFSUNLOCKCLSTATE();
1820         NFSPROCLISTUNLOCK();
1821 }
1822
1823 /*
1824  * Take the empty lock owner and move it to the local lhp list if the
1825  * associated process no longer exists.
1826  */
1827 static void
1828 nfscl_emptylockowner(struct nfscllockowner *lp,
1829     struct nfscllockownerfhhead *lhp)
1830 {
1831         struct nfscllockownerfh *lfhp, *mylfhp;
1832         struct nfscllockowner *nlp;
1833         int fnd_it;
1834
1835         /* If not a Posix lock owner, just return. */
1836         if ((lp->nfsl_lockflags & F_POSIX) == 0)
1837                 return;
1838
1839         fnd_it = 0;
1840         mylfhp = NULL;
1841         /*
1842          * First, search to see if this lock owner is already in the list.
1843          * If it is, then the associated process no longer exists.
1844          */
1845         SLIST_FOREACH(lfhp, lhp, nfslfh_list) {
1846                 if (lfhp->nfslfh_len == lp->nfsl_open->nfso_fhlen &&
1847                     !NFSBCMP(lfhp->nfslfh_fh, lp->nfsl_open->nfso_fh,
1848                     lfhp->nfslfh_len))
1849                         mylfhp = lfhp;
1850                 LIST_FOREACH(nlp, &lfhp->nfslfh_lock, nfsl_list)
1851                         if (!NFSBCMP(nlp->nfsl_owner, lp->nfsl_owner,
1852                             NFSV4CL_LOCKNAMELEN))
1853                                 fnd_it = 1;
1854         }
1855         /* If not found, check if process still exists. */
1856         if (fnd_it == 0 && nfscl_procdoesntexist(lp->nfsl_owner) == 0)
1857                 return;
1858
1859         /* Move the lock owner over to the local list. */
1860         if (mylfhp == NULL) {
1861                 mylfhp = malloc(sizeof(struct nfscllockownerfh), M_TEMP,
1862                     M_NOWAIT);
1863                 if (mylfhp == NULL)
1864                         return;
1865                 mylfhp->nfslfh_len = lp->nfsl_open->nfso_fhlen;
1866                 NFSBCOPY(lp->nfsl_open->nfso_fh, mylfhp->nfslfh_fh,
1867                     mylfhp->nfslfh_len);
1868                 LIST_INIT(&mylfhp->nfslfh_lock);
1869                 SLIST_INSERT_HEAD(lhp, mylfhp, nfslfh_list);
1870         }
1871         LIST_REMOVE(lp, nfsl_list);
1872         LIST_INSERT_HEAD(&mylfhp->nfslfh_lock, lp, nfsl_list);
1873 }
1874
1875 static int      fake_global;    /* Used to force visibility of MNTK_UNMOUNTF */
1876 /*
1877  * Called from nfs umount to free up the clientid.
1878  */
1879 void
1880 nfscl_umount(struct nfsmount *nmp, NFSPROC_T *p)
1881 {
1882         struct nfsclclient *clp;
1883         struct ucred *cred;
1884         int igotlock;
1885
1886         /*
1887          * For the case that matters, this is the thread that set
1888          * MNTK_UNMOUNTF, so it will see it set. The code that follows is
1889          * done to ensure that any thread executing nfscl_getcl() after
1890          * this time, will see MNTK_UNMOUNTF set. nfscl_getcl() uses the
1891          * mutex for NFSLOCKCLSTATE(), so it is "m" for the following
1892          * explanation, courtesy of Alan Cox.
1893          * What follows is a snippet from Alan Cox's email at:
1894          * https://docs.FreeBSD.org/cgi/mid.cgi?BANLkTikR3d65zPHo9==08ZfJ2vmqZucEvw
1895          *
1896          * 1. Set MNTK_UNMOUNTF
1897          * 2. Acquire a standard FreeBSD mutex "m".
1898          * 3. Update some data structures.
1899          * 4. Release mutex "m".
1900          *
1901          * Then, other threads that acquire "m" after step 4 has occurred will
1902          * see MNTK_UNMOUNTF as set.  But, other threads that beat thread X to
1903          * step 2 may or may not see MNTK_UNMOUNTF as set.
1904          */
1905         NFSLOCKCLSTATE();
1906         if ((nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) {
1907                 fake_global++;
1908                 NFSUNLOCKCLSTATE();
1909                 NFSLOCKCLSTATE();
1910         }
1911
1912         clp = nmp->nm_clp;
1913         if (clp != NULL) {
1914                 if ((clp->nfsc_flags & NFSCLFLAGS_INITED) == 0)
1915                         panic("nfscl umount");
1916
1917                 /*
1918                  * First, handshake with the nfscl renew thread, to terminate
1919                  * it.
1920                  */
1921                 clp->nfsc_flags |= NFSCLFLAGS_UMOUNT;
1922                 while (clp->nfsc_flags & NFSCLFLAGS_HASTHREAD)
1923                         (void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT,
1924                             "nfsclumnt", hz);
1925
1926                 /*
1927                  * Now, get the exclusive lock on the client state, so
1928                  * that no uses of the state are still in progress.
1929                  */
1930                 do {
1931                         igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
1932                             NFSCLSTATEMUTEXPTR, NULL);
1933                 } while (!igotlock);
1934                 NFSUNLOCKCLSTATE();
1935
1936                 /*
1937                  * Free up all the state. It will expire on the server, but
1938                  * maybe we should do a SetClientId/SetClientIdConfirm so
1939                  * the server throws it away?
1940                  */
1941                 LIST_REMOVE(clp, nfsc_list);
1942                 nfscl_delegreturnall(clp, p);
1943                 cred = newnfs_getcred();
1944                 if (NFSHASNFSV4N(nmp)) {
1945                         (void)nfsrpc_destroysession(nmp, clp, cred, p);
1946                         (void)nfsrpc_destroyclient(nmp, clp, cred, p);
1947                 } else
1948                         (void)nfsrpc_setclient(nmp, clp, 0, cred, p);
1949                 nfscl_cleanclient(clp);
1950                 nmp->nm_clp = NULL;
1951                 NFSFREECRED(cred);
1952                 free(clp, M_NFSCLCLIENT);
1953         } else
1954                 NFSUNLOCKCLSTATE();
1955 }
1956
1957 /*
1958  * This function is called when a server replies with NFSERR_STALECLIENTID
1959  * NFSERR_STALESTATEID or NFSERR_BADSESSION. It traverses the clientid lists,
1960  * doing Opens and Locks with reclaim. If these fail, it deletes the
1961  * corresponding state.
1962  */
1963 static void
1964 nfscl_recover(struct nfsclclient *clp, struct ucred *cred, NFSPROC_T *p)
1965 {
1966         struct nfsclowner *owp, *nowp;
1967         struct nfsclopen *op, *nop;
1968         struct nfscllockowner *lp, *nlp;
1969         struct nfscllock *lop, *nlop;
1970         struct nfscldeleg *dp, *ndp, *tdp;
1971         struct nfsmount *nmp;
1972         struct ucred *tcred;
1973         struct nfsclopenhead extra_open;
1974         struct nfscldeleghead extra_deleg;
1975         struct nfsreq *rep;
1976         u_int64_t len;
1977         u_int32_t delegtype = NFSV4OPEN_DELEGATEWRITE, mode;
1978         int i, igotlock = 0, error, trycnt, firstlock;
1979         struct nfscllayout *lyp, *nlyp;
1980
1981         /*
1982          * First, lock the client structure, so everyone else will
1983          * block when trying to use state.
1984          */
1985         NFSLOCKCLSTATE();
1986         clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG;
1987         do {
1988                 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
1989                     NFSCLSTATEMUTEXPTR, NULL);
1990         } while (!igotlock);
1991         NFSUNLOCKCLSTATE();
1992
1993         nmp = clp->nfsc_nmp;
1994         if (nmp == NULL)
1995                 panic("nfscl recover");
1996
1997         /*
1998          * For now, just get rid of all layouts. There may be a need
1999          * to do LayoutCommit Ops with reclaim == true later.
2000          */
2001         TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp)
2002                 nfscl_freelayout(lyp);
2003         TAILQ_INIT(&clp->nfsc_layout);
2004         for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++)
2005                 LIST_INIT(&clp->nfsc_layouthash[i]);
2006
2007         trycnt = 5;
2008         do {
2009                 error = nfsrpc_setclient(nmp, clp, 1, cred, p);
2010         } while ((error == NFSERR_STALECLIENTID ||
2011              error == NFSERR_BADSESSION ||
2012              error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
2013         if (error) {
2014                 NFSLOCKCLSTATE();
2015                 clp->nfsc_flags &= ~(NFSCLFLAGS_RECOVER |
2016                     NFSCLFLAGS_RECVRINPROG);
2017                 wakeup(&clp->nfsc_flags);
2018                 nfsv4_unlock(&clp->nfsc_lock, 0);
2019                 NFSUNLOCKCLSTATE();
2020                 return;
2021         }
2022         clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
2023         clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2024
2025         /*
2026          * Mark requests already queued on the server, so that they don't
2027          * initiate another recovery cycle. Any requests already in the
2028          * queue that handle state information will have the old stale
2029          * clientid/stateid and will get a NFSERR_STALESTATEID,
2030          * NFSERR_STALECLIENTID or NFSERR_BADSESSION reply from the server.
2031          * This will be translated to NFSERR_STALEDONTRECOVER when
2032          * R_DONTRECOVER is set.
2033          */
2034         NFSLOCKREQ();
2035         TAILQ_FOREACH(rep, &nfsd_reqq, r_chain) {
2036                 if (rep->r_nmp == nmp)
2037                         rep->r_flags |= R_DONTRECOVER;
2038         }
2039         NFSUNLOCKREQ();
2040
2041         /*
2042          * Now, mark all delegations "need reclaim".
2043          */
2044         TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list)
2045                 dp->nfsdl_flags |= NFSCLDL_NEEDRECLAIM;
2046
2047         TAILQ_INIT(&extra_deleg);
2048         LIST_INIT(&extra_open);
2049         /*
2050          * Now traverse the state lists, doing Open and Lock Reclaims.
2051          */
2052         tcred = newnfs_getcred();
2053         owp = LIST_FIRST(&clp->nfsc_owner);
2054         while (owp != NULL) {
2055             nowp = LIST_NEXT(owp, nfsow_list);
2056             owp->nfsow_seqid = 0;
2057             op = LIST_FIRST(&owp->nfsow_open);
2058             while (op != NULL) {
2059                 nop = LIST_NEXT(op, nfso_list);
2060                 if (error != NFSERR_NOGRACE && error != NFSERR_BADSESSION) {
2061                     /* Search for a delegation to reclaim with the open */
2062                     TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2063                         if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
2064                             continue;
2065                         if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
2066                             mode = NFSV4OPEN_ACCESSWRITE;
2067                             delegtype = NFSV4OPEN_DELEGATEWRITE;
2068                         } else {
2069                             mode = NFSV4OPEN_ACCESSREAD;
2070                             delegtype = NFSV4OPEN_DELEGATEREAD;
2071                         }
2072                         if ((op->nfso_mode & mode) == mode &&
2073                             op->nfso_fhlen == dp->nfsdl_fhlen &&
2074                             !NFSBCMP(op->nfso_fh, dp->nfsdl_fh, op->nfso_fhlen))
2075                             break;
2076                     }
2077                     ndp = dp;
2078                     if (dp == NULL)
2079                         delegtype = NFSV4OPEN_DELEGATENONE;
2080                     newnfs_copycred(&op->nfso_cred, tcred);
2081                     error = nfscl_tryopen(nmp, NULL, op->nfso_fh,
2082                         op->nfso_fhlen, op->nfso_fh, op->nfso_fhlen,
2083                         op->nfso_mode, op, NULL, 0, &ndp, 1, delegtype,
2084                         tcred, p);
2085                     if (!error) {
2086                         /* Handle any replied delegation */
2087                         if (ndp != NULL && ((ndp->nfsdl_flags & NFSCLDL_WRITE)
2088                             || NFSMNT_RDONLY(nmp->nm_mountp))) {
2089                             if ((ndp->nfsdl_flags & NFSCLDL_WRITE))
2090                                 mode = NFSV4OPEN_ACCESSWRITE;
2091                             else
2092                                 mode = NFSV4OPEN_ACCESSREAD;
2093                             TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2094                                 if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
2095                                     continue;
2096                                 if ((op->nfso_mode & mode) == mode &&
2097                                     op->nfso_fhlen == dp->nfsdl_fhlen &&
2098                                     !NFSBCMP(op->nfso_fh, dp->nfsdl_fh,
2099                                     op->nfso_fhlen)) {
2100                                     dp->nfsdl_stateid = ndp->nfsdl_stateid;
2101                                     dp->nfsdl_sizelimit = ndp->nfsdl_sizelimit;
2102                                     dp->nfsdl_ace = ndp->nfsdl_ace;
2103                                     dp->nfsdl_change = ndp->nfsdl_change;
2104                                     dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
2105                                     if ((ndp->nfsdl_flags & NFSCLDL_RECALL))
2106                                         dp->nfsdl_flags |= NFSCLDL_RECALL;
2107                                     free(ndp, M_NFSCLDELEG);
2108                                     ndp = NULL;
2109                                     break;
2110                                 }
2111                             }
2112                         }
2113                         if (ndp != NULL)
2114                             TAILQ_INSERT_HEAD(&extra_deleg, ndp, nfsdl_list);
2115
2116                         /* and reclaim all byte range locks */
2117                         lp = LIST_FIRST(&op->nfso_lock);
2118                         while (lp != NULL) {
2119                             nlp = LIST_NEXT(lp, nfsl_list);
2120                             lp->nfsl_seqid = 0;
2121                             firstlock = 1;
2122                             lop = LIST_FIRST(&lp->nfsl_lock);
2123                             while (lop != NULL) {
2124                                 nlop = LIST_NEXT(lop, nfslo_list);
2125                                 if (lop->nfslo_end == NFS64BITSSET)
2126                                     len = NFS64BITSSET;
2127                                 else
2128                                     len = lop->nfslo_end - lop->nfslo_first;
2129                                 error = nfscl_trylock(nmp, NULL,
2130                                     op->nfso_fh, op->nfso_fhlen, lp,
2131                                     firstlock, 1, lop->nfslo_first, len,
2132                                     lop->nfslo_type, tcred, p);
2133                                 if (error != 0)
2134                                     nfscl_freelock(lop, 0);
2135                                 else
2136                                     firstlock = 0;
2137                                 lop = nlop;
2138                             }
2139                             /* If no locks, but a lockowner, just delete it. */
2140                             if (LIST_EMPTY(&lp->nfsl_lock))
2141                                 nfscl_freelockowner(lp, 0);
2142                             lp = nlp;
2143                         }
2144                     }
2145                 }
2146                 if (error != 0 && error != NFSERR_BADSESSION)
2147                     nfscl_freeopen(op, 0);
2148                 op = nop;
2149             }
2150             owp = nowp;
2151         }
2152
2153         /*
2154          * Now, try and get any delegations not yet reclaimed by cobbling
2155          * to-gether an appropriate open.
2156          */
2157         nowp = NULL;
2158         dp = TAILQ_FIRST(&clp->nfsc_deleg);
2159         while (dp != NULL) {
2160             ndp = TAILQ_NEXT(dp, nfsdl_list);
2161             if ((dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM)) {
2162                 if (nowp == NULL) {
2163                     nowp = malloc(
2164                         sizeof (struct nfsclowner), M_NFSCLOWNER, M_WAITOK);
2165                     /*
2166                      * Name must be as long an largest possible
2167                      * NFSV4CL_LOCKNAMELEN. 12 for now.
2168                      */
2169                     NFSBCOPY("RECLAIMDELEG", nowp->nfsow_owner,
2170                         NFSV4CL_LOCKNAMELEN);
2171                     LIST_INIT(&nowp->nfsow_open);
2172                     nowp->nfsow_clp = clp;
2173                     nowp->nfsow_seqid = 0;
2174                     nowp->nfsow_defunct = 0;
2175                     nfscl_lockinit(&nowp->nfsow_rwlock);
2176                 }
2177                 nop = NULL;
2178                 if (error != NFSERR_NOGRACE && error != NFSERR_BADSESSION) {
2179                     nop = malloc(sizeof (struct nfsclopen) +
2180                         dp->nfsdl_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
2181                     nop->nfso_own = nowp;
2182                     if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
2183                         nop->nfso_mode = NFSV4OPEN_ACCESSWRITE;
2184                         delegtype = NFSV4OPEN_DELEGATEWRITE;
2185                     } else {
2186                         nop->nfso_mode = NFSV4OPEN_ACCESSREAD;
2187                         delegtype = NFSV4OPEN_DELEGATEREAD;
2188                     }
2189                     nop->nfso_opencnt = 0;
2190                     nop->nfso_posixlock = 1;
2191                     nop->nfso_fhlen = dp->nfsdl_fhlen;
2192                     NFSBCOPY(dp->nfsdl_fh, nop->nfso_fh, dp->nfsdl_fhlen);
2193                     LIST_INIT(&nop->nfso_lock);
2194                     nop->nfso_stateid.seqid = 0;
2195                     nop->nfso_stateid.other[0] = 0;
2196                     nop->nfso_stateid.other[1] = 0;
2197                     nop->nfso_stateid.other[2] = 0;
2198                     newnfs_copycred(&dp->nfsdl_cred, tcred);
2199                     newnfs_copyincred(tcred, &nop->nfso_cred);
2200                     tdp = NULL;
2201                     error = nfscl_tryopen(nmp, NULL, nop->nfso_fh,
2202                         nop->nfso_fhlen, nop->nfso_fh, nop->nfso_fhlen,
2203                         nop->nfso_mode, nop, NULL, 0, &tdp, 1,
2204                         delegtype, tcred, p);
2205                     if (tdp != NULL) {
2206                         if ((tdp->nfsdl_flags & NFSCLDL_WRITE))
2207                             mode = NFSV4OPEN_ACCESSWRITE;
2208                         else
2209                             mode = NFSV4OPEN_ACCESSREAD;
2210                         if ((nop->nfso_mode & mode) == mode &&
2211                             nop->nfso_fhlen == tdp->nfsdl_fhlen &&
2212                             !NFSBCMP(nop->nfso_fh, tdp->nfsdl_fh,
2213                             nop->nfso_fhlen)) {
2214                             dp->nfsdl_stateid = tdp->nfsdl_stateid;
2215                             dp->nfsdl_sizelimit = tdp->nfsdl_sizelimit;
2216                             dp->nfsdl_ace = tdp->nfsdl_ace;
2217                             dp->nfsdl_change = tdp->nfsdl_change;
2218                             dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
2219                             if ((tdp->nfsdl_flags & NFSCLDL_RECALL))
2220                                 dp->nfsdl_flags |= NFSCLDL_RECALL;
2221                             free(tdp, M_NFSCLDELEG);
2222                         } else {
2223                             TAILQ_INSERT_HEAD(&extra_deleg, tdp, nfsdl_list);
2224                         }
2225                     }
2226                 }
2227                 if (error) {
2228                     if (nop != NULL)
2229                         free(nop, M_NFSCLOPEN);
2230                     /*
2231                      * Couldn't reclaim it, so throw the state
2232                      * away. Ouch!!
2233                      */
2234                     nfscl_cleandeleg(dp);
2235                     nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
2236                 } else {
2237                     LIST_INSERT_HEAD(&extra_open, nop, nfso_list);
2238                 }
2239             }
2240             dp = ndp;
2241         }
2242
2243         /*
2244          * Now, get rid of extra Opens and Delegations.
2245          */
2246         LIST_FOREACH_SAFE(op, &extra_open, nfso_list, nop) {
2247                 do {
2248                         newnfs_copycred(&op->nfso_cred, tcred);
2249                         error = nfscl_tryclose(op, tcred, nmp, p);
2250                         if (error == NFSERR_GRACE)
2251                                 (void) nfs_catnap(PZERO, error, "nfsexcls");
2252                 } while (error == NFSERR_GRACE);
2253                 LIST_REMOVE(op, nfso_list);
2254                 free(op, M_NFSCLOPEN);
2255         }
2256         if (nowp != NULL)
2257                 free(nowp, M_NFSCLOWNER);
2258
2259         TAILQ_FOREACH_SAFE(dp, &extra_deleg, nfsdl_list, ndp) {
2260                 do {
2261                         newnfs_copycred(&dp->nfsdl_cred, tcred);
2262                         error = nfscl_trydelegreturn(dp, tcred, nmp, p);
2263                         if (error == NFSERR_GRACE)
2264                                 (void) nfs_catnap(PZERO, error, "nfsexdlg");
2265                 } while (error == NFSERR_GRACE);
2266                 TAILQ_REMOVE(&extra_deleg, dp, nfsdl_list);
2267                 free(dp, M_NFSCLDELEG);
2268         }
2269
2270         /* For NFSv4.1 or later, do a RECLAIM_COMPLETE. */
2271         if (NFSHASNFSV4N(nmp))
2272                 (void)nfsrpc_reclaimcomplete(nmp, cred, p);
2273
2274         NFSLOCKCLSTATE();
2275         clp->nfsc_flags &= ~NFSCLFLAGS_RECVRINPROG;
2276         wakeup(&clp->nfsc_flags);
2277         nfsv4_unlock(&clp->nfsc_lock, 0);
2278         NFSUNLOCKCLSTATE();
2279         NFSFREECRED(tcred);
2280 }
2281
2282 /*
2283  * This function is called when a server replies with NFSERR_EXPIRED.
2284  * It deletes all state for the client and does a fresh SetClientId/confirm.
2285  * XXX Someday it should post a signal to the process(es) that hold the
2286  * state, so they know that lock state has been lost.
2287  */
2288 int
2289 nfscl_hasexpired(struct nfsclclient *clp, u_int32_t clidrev, NFSPROC_T *p)
2290 {
2291         struct nfsmount *nmp;
2292         struct ucred *cred;
2293         int igotlock = 0, error, trycnt;
2294
2295         /*
2296          * If the clientid has gone away or a new SetClientid has already
2297          * been done, just return ok.
2298          */
2299         if (clp == NULL || clidrev != clp->nfsc_clientidrev)
2300                 return (0);
2301
2302         /*
2303          * First, lock the client structure, so everyone else will
2304          * block when trying to use state. Also, use NFSCLFLAGS_EXPIREIT so
2305          * that only one thread does the work.
2306          */
2307         NFSLOCKCLSTATE();
2308         clp->nfsc_flags |= NFSCLFLAGS_EXPIREIT;
2309         do {
2310                 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
2311                     NFSCLSTATEMUTEXPTR, NULL);
2312         } while (!igotlock && (clp->nfsc_flags & NFSCLFLAGS_EXPIREIT));
2313         if ((clp->nfsc_flags & NFSCLFLAGS_EXPIREIT) == 0) {
2314                 if (igotlock)
2315                         nfsv4_unlock(&clp->nfsc_lock, 0);
2316                 NFSUNLOCKCLSTATE();
2317                 return (0);
2318         }
2319         clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG;
2320         NFSUNLOCKCLSTATE();
2321
2322         nmp = clp->nfsc_nmp;
2323         if (nmp == NULL)
2324                 panic("nfscl expired");
2325         cred = newnfs_getcred();
2326         trycnt = 5;
2327         do {
2328                 error = nfsrpc_setclient(nmp, clp, 0, cred, p);
2329         } while ((error == NFSERR_STALECLIENTID ||
2330              error == NFSERR_BADSESSION ||
2331              error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
2332         if (error) {
2333                 NFSLOCKCLSTATE();
2334                 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2335         } else {
2336                 /*
2337                  * Expire the state for the client.
2338                  */
2339                 nfscl_expireclient(clp, nmp, cred, p);
2340                 NFSLOCKCLSTATE();
2341                 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
2342                 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2343         }
2344         clp->nfsc_flags &= ~(NFSCLFLAGS_EXPIREIT | NFSCLFLAGS_RECVRINPROG);
2345         wakeup(&clp->nfsc_flags);
2346         nfsv4_unlock(&clp->nfsc_lock, 0);
2347         NFSUNLOCKCLSTATE();
2348         NFSFREECRED(cred);
2349         return (error);
2350 }
2351
2352 /*
2353  * This function inserts a lock in the list after insert_lop.
2354  */
2355 static void
2356 nfscl_insertlock(struct nfscllockowner *lp, struct nfscllock *new_lop,
2357     struct nfscllock *insert_lop, int local)
2358 {
2359
2360         if ((struct nfscllockowner *)insert_lop == lp)
2361                 LIST_INSERT_HEAD(&lp->nfsl_lock, new_lop, nfslo_list);
2362         else
2363                 LIST_INSERT_AFTER(insert_lop, new_lop, nfslo_list);
2364         if (local)
2365                 nfsstatsv1.cllocallocks++;
2366         else
2367                 nfsstatsv1.cllocks++;
2368 }
2369
2370 /*
2371  * This function updates the locking for a lock owner and given file. It
2372  * maintains a list of lock ranges ordered on increasing file offset that
2373  * are NFSCLLOCK_READ or NFSCLLOCK_WRITE and non-overlapping (aka POSIX style).
2374  * It always adds new_lop to the list and sometimes uses the one pointed
2375  * at by other_lopp.
2376  * Returns 1 if the locks were modified, 0 otherwise.
2377  */
2378 static int
2379 nfscl_updatelock(struct nfscllockowner *lp, struct nfscllock **new_lopp,
2380     struct nfscllock **other_lopp, int local)
2381 {
2382         struct nfscllock *new_lop = *new_lopp;
2383         struct nfscllock *lop, *tlop, *ilop;
2384         struct nfscllock *other_lop;
2385         int unlock = 0, modified = 0;
2386         u_int64_t tmp;
2387
2388         /*
2389          * Work down the list until the lock is merged.
2390          */
2391         if (new_lop->nfslo_type == F_UNLCK)
2392                 unlock = 1;
2393         ilop = (struct nfscllock *)lp;
2394         lop = LIST_FIRST(&lp->nfsl_lock);
2395         while (lop != NULL) {
2396             /*
2397              * Only check locks for this file that aren't before the start of
2398              * new lock's range.
2399              */
2400             if (lop->nfslo_end >= new_lop->nfslo_first) {
2401                 if (new_lop->nfslo_end < lop->nfslo_first) {
2402                     /*
2403                      * If the new lock ends before the start of the
2404                      * current lock's range, no merge, just insert
2405                      * the new lock.
2406                      */
2407                     break;
2408                 }
2409                 if (new_lop->nfslo_type == lop->nfslo_type ||
2410                     (new_lop->nfslo_first <= lop->nfslo_first &&
2411                      new_lop->nfslo_end >= lop->nfslo_end)) {
2412                     /*
2413                      * This lock can be absorbed by the new lock/unlock.
2414                      * This happens when it covers the entire range
2415                      * of the old lock or is contiguous
2416                      * with the old lock and is of the same type or an
2417                      * unlock.
2418                      */
2419                     if (new_lop->nfslo_type != lop->nfslo_type ||
2420                         new_lop->nfslo_first != lop->nfslo_first ||
2421                         new_lop->nfslo_end != lop->nfslo_end)
2422                         modified = 1;
2423                     if (lop->nfslo_first < new_lop->nfslo_first)
2424                         new_lop->nfslo_first = lop->nfslo_first;
2425                     if (lop->nfslo_end > new_lop->nfslo_end)
2426                         new_lop->nfslo_end = lop->nfslo_end;
2427                     tlop = lop;
2428                     lop = LIST_NEXT(lop, nfslo_list);
2429                     nfscl_freelock(tlop, local);
2430                     continue;
2431                 }
2432
2433                 /*
2434                  * All these cases are for contiguous locks that are not the
2435                  * same type, so they can't be merged.
2436                  */
2437                 if (new_lop->nfslo_first <= lop->nfslo_first) {
2438                     /*
2439                      * This case is where the new lock overlaps with the
2440                      * first part of the old lock. Move the start of the
2441                      * old lock to just past the end of the new lock. The
2442                      * new lock will be inserted in front of the old, since
2443                      * ilop hasn't been updated. (We are done now.)
2444                      */
2445                     if (lop->nfslo_first != new_lop->nfslo_end) {
2446                         lop->nfslo_first = new_lop->nfslo_end;
2447                         modified = 1;
2448                     }
2449                     break;
2450                 }
2451                 if (new_lop->nfslo_end >= lop->nfslo_end) {
2452                     /*
2453                      * This case is where the new lock overlaps with the
2454                      * end of the old lock's range. Move the old lock's
2455                      * end to just before the new lock's first and insert
2456                      * the new lock after the old lock.
2457                      * Might not be done yet, since the new lock could
2458                      * overlap further locks with higher ranges.
2459                      */
2460                     if (lop->nfslo_end != new_lop->nfslo_first) {
2461                         lop->nfslo_end = new_lop->nfslo_first;
2462                         modified = 1;
2463                     }
2464                     ilop = lop;
2465                     lop = LIST_NEXT(lop, nfslo_list);
2466                     continue;
2467                 }
2468                 /*
2469                  * The final case is where the new lock's range is in the
2470                  * middle of the current lock's and splits the current lock
2471                  * up. Use *other_lopp to handle the second part of the
2472                  * split old lock range. (We are done now.)
2473                  * For unlock, we use new_lop as other_lop and tmp, since
2474                  * other_lop and new_lop are the same for this case.
2475                  * We noted the unlock case above, so we don't need
2476                  * new_lop->nfslo_type any longer.
2477                  */
2478                 tmp = new_lop->nfslo_first;
2479                 if (unlock) {
2480                     other_lop = new_lop;
2481                     *new_lopp = NULL;
2482                 } else {
2483                     other_lop = *other_lopp;
2484                     *other_lopp = NULL;
2485                 }
2486                 other_lop->nfslo_first = new_lop->nfslo_end;
2487                 other_lop->nfslo_end = lop->nfslo_end;
2488                 other_lop->nfslo_type = lop->nfslo_type;
2489                 lop->nfslo_end = tmp;
2490                 nfscl_insertlock(lp, other_lop, lop, local);
2491                 ilop = lop;
2492                 modified = 1;
2493                 break;
2494             }
2495             ilop = lop;
2496             lop = LIST_NEXT(lop, nfslo_list);
2497             if (lop == NULL)
2498                 break;
2499         }
2500
2501         /*
2502          * Insert the new lock in the list at the appropriate place.
2503          */
2504         if (!unlock) {
2505                 nfscl_insertlock(lp, new_lop, ilop, local);
2506                 *new_lopp = NULL;
2507                 modified = 1;
2508         }
2509         return (modified);
2510 }
2511
2512 /*
2513  * This function must be run as a kernel thread.
2514  * It does Renew Ops and recovery, when required.
2515  */
2516 void
2517 nfscl_renewthread(struct nfsclclient *clp, NFSPROC_T *p)
2518 {
2519         struct nfsclowner *owp, *nowp;
2520         struct nfsclopen *op;
2521         struct nfscllockowner *lp, *nlp;
2522         struct nfscldeleghead dh;
2523         struct nfscldeleg *dp, *ndp;
2524         struct ucred *cred;
2525         u_int32_t clidrev;
2526         int error, cbpathdown, islept, igotlock, ret, clearok;
2527         uint32_t recover_done_time = 0;
2528         time_t mytime;
2529         static time_t prevsec = 0;
2530         struct nfscllockownerfh *lfhp, *nlfhp;
2531         struct nfscllockownerfhhead lfh;
2532         struct nfscllayout *lyp, *nlyp;
2533         struct nfscldevinfo *dip, *ndip;
2534         struct nfscllayouthead rlh;
2535         struct nfsclrecalllayout *recallp;
2536         struct nfsclds *dsp;
2537         struct mount *mp;
2538         vnode_t vp;
2539
2540         cred = newnfs_getcred();
2541         NFSLOCKCLSTATE();
2542         clp->nfsc_flags |= NFSCLFLAGS_HASTHREAD;
2543         mp = clp->nfsc_nmp->nm_mountp;
2544         NFSUNLOCKCLSTATE();
2545         for(;;) {
2546                 newnfs_setroot(cred);
2547                 cbpathdown = 0;
2548                 if (clp->nfsc_flags & NFSCLFLAGS_RECOVER) {
2549                         /*
2550                          * Only allow one recover within 1/2 of the lease
2551                          * duration (nfsc_renew).
2552                          */
2553                         if (recover_done_time < NFSD_MONOSEC) {
2554                                 recover_done_time = NFSD_MONOSEC +
2555                                     clp->nfsc_renew;
2556                                 NFSCL_DEBUG(1, "Doing recovery..\n");
2557                                 nfscl_recover(clp, cred, p);
2558                         } else {
2559                                 NFSCL_DEBUG(1, "Clear Recovery dt=%u ms=%jd\n",
2560                                     recover_done_time, (intmax_t)NFSD_MONOSEC);
2561                                 NFSLOCKCLSTATE();
2562                                 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2563                                 NFSUNLOCKCLSTATE();
2564                         }
2565                 }
2566                 if (clp->nfsc_expire <= NFSD_MONOSEC &&
2567                     (clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID)) {
2568                         clp->nfsc_expire = NFSD_MONOSEC + clp->nfsc_renew;
2569                         clidrev = clp->nfsc_clientidrev;
2570                         error = nfsrpc_renew(clp, NULL, cred, p);
2571                         if (error == NFSERR_CBPATHDOWN)
2572                             cbpathdown = 1;
2573                         else if (error == NFSERR_STALECLIENTID ||
2574                             error == NFSERR_BADSESSION) {
2575                             NFSLOCKCLSTATE();
2576                             clp->nfsc_flags |= NFSCLFLAGS_RECOVER;
2577                             NFSUNLOCKCLSTATE();
2578                         } else if (error == NFSERR_EXPIRED)
2579                             (void) nfscl_hasexpired(clp, clidrev, p);
2580                 }
2581
2582 checkdsrenew:
2583                 if (NFSHASNFSV4N(clp->nfsc_nmp)) {
2584                         /* Do renews for any DS sessions. */
2585                         NFSLOCKMNT(clp->nfsc_nmp);
2586                         /* Skip first entry, since the MDS is handled above. */
2587                         dsp = TAILQ_FIRST(&clp->nfsc_nmp->nm_sess);
2588                         if (dsp != NULL)
2589                                 dsp = TAILQ_NEXT(dsp, nfsclds_list);
2590                         while (dsp != NULL) {
2591                                 if (dsp->nfsclds_expire <= NFSD_MONOSEC &&
2592                                     dsp->nfsclds_sess.nfsess_defunct == 0) {
2593                                         dsp->nfsclds_expire = NFSD_MONOSEC +
2594                                             clp->nfsc_renew;
2595                                         NFSUNLOCKMNT(clp->nfsc_nmp);
2596                                         (void)nfsrpc_renew(clp, dsp, cred, p);
2597                                         goto checkdsrenew;
2598                                 }
2599                                 dsp = TAILQ_NEXT(dsp, nfsclds_list);
2600                         }
2601                         NFSUNLOCKMNT(clp->nfsc_nmp);
2602                 }
2603
2604                 TAILQ_INIT(&dh);
2605                 NFSLOCKCLSTATE();
2606                 if (cbpathdown)
2607                         /* It's a Total Recall! */
2608                         nfscl_totalrecall(clp);
2609
2610                 /*
2611                  * Now, handle defunct owners.
2612                  */
2613                 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
2614                         if (LIST_EMPTY(&owp->nfsow_open)) {
2615                                 if (owp->nfsow_defunct != 0)
2616                                         nfscl_freeopenowner(owp, 0);
2617                         }
2618                 }
2619
2620                 /*
2621                  * Do the recall on any delegations. To avoid trouble, always
2622                  * come back up here after having slept.
2623                  */
2624                 igotlock = 0;
2625 tryagain:
2626                 dp = TAILQ_FIRST(&clp->nfsc_deleg);
2627                 while (dp != NULL) {
2628                         ndp = TAILQ_NEXT(dp, nfsdl_list);
2629                         if ((dp->nfsdl_flags & NFSCLDL_RECALL)) {
2630                                 /*
2631                                  * Wait for outstanding I/O ops to be done.
2632                                  */
2633                                 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
2634                                     if (igotlock) {
2635                                         nfsv4_unlock(&clp->nfsc_lock, 0);
2636                                         igotlock = 0;
2637                                     }
2638                                     dp->nfsdl_rwlock.nfslock_lock |=
2639                                         NFSV4LOCK_WANTED;
2640                                     msleep(&dp->nfsdl_rwlock,
2641                                         NFSCLSTATEMUTEXPTR, PVFS, "nfscld",
2642                                         5 * hz);
2643                                     if (NFSCL_FORCEDISM(mp))
2644                                         goto terminate;
2645                                     goto tryagain;
2646                                 }
2647                                 while (!igotlock) {
2648                                     igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
2649                                         &islept, NFSCLSTATEMUTEXPTR, mp);
2650                                     if (igotlock == 0 && NFSCL_FORCEDISM(mp))
2651                                         goto terminate;
2652                                     if (islept)
2653                                         goto tryagain;
2654                                 }
2655                                 NFSUNLOCKCLSTATE();
2656                                 newnfs_copycred(&dp->nfsdl_cred, cred);
2657                                 ret = nfscl_recalldeleg(clp, clp->nfsc_nmp, dp,
2658                                     NULL, cred, p, 1, &vp);
2659                                 if (!ret) {
2660                                     nfscl_cleandeleg(dp);
2661                                     TAILQ_REMOVE(&clp->nfsc_deleg, dp,
2662                                         nfsdl_list);
2663                                     LIST_REMOVE(dp, nfsdl_hash);
2664                                     TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
2665                                     nfscl_delegcnt--;
2666                                     nfsstatsv1.cldelegates--;
2667                                 }
2668                                 NFSLOCKCLSTATE();
2669                                 /*
2670                                  * The nfsc_lock must be released before doing
2671                                  * vrele(), since it might call nfs_inactive().
2672                                  * For the unlikely case where the vnode failed
2673                                  * to be acquired by nfscl_recalldeleg(), a
2674                                  * VOP_RECLAIM() should be in progress and it
2675                                  * will return the delegation.
2676                                  */
2677                                 nfsv4_unlock(&clp->nfsc_lock, 0);
2678                                 igotlock = 0;
2679                                 if (vp != NULL) {
2680                                         NFSUNLOCKCLSTATE();
2681                                         vrele(vp);
2682                                         NFSLOCKCLSTATE();
2683                                 }
2684                                 goto tryagain;
2685                         }
2686                         dp = ndp;
2687                 }
2688
2689                 /*
2690                  * Clear out old delegations, if we are above the high water
2691                  * mark. Only clear out ones with no state related to them.
2692                  * The tailq list is in LRU order.
2693                  */
2694                 dp = TAILQ_LAST(&clp->nfsc_deleg, nfscldeleghead);
2695                 while (nfscl_delegcnt > nfscl_deleghighwater && dp != NULL) {
2696                     ndp = TAILQ_PREV(dp, nfscldeleghead, nfsdl_list);
2697                     if (dp->nfsdl_rwlock.nfslock_usecnt == 0 &&
2698                         dp->nfsdl_rwlock.nfslock_lock == 0 &&
2699                         dp->nfsdl_timestamp < NFSD_MONOSEC &&
2700                         (dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_ZAPPED |
2701                           NFSCLDL_NEEDRECLAIM | NFSCLDL_DELEGRET)) == 0) {
2702                         clearok = 1;
2703                         LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
2704                             op = LIST_FIRST(&owp->nfsow_open);
2705                             if (op != NULL) {
2706                                 clearok = 0;
2707                                 break;
2708                             }
2709                         }
2710                         if (clearok) {
2711                             LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
2712                                 if (!LIST_EMPTY(&lp->nfsl_lock)) {
2713                                     clearok = 0;
2714                                     break;
2715                                 }
2716                             }
2717                         }
2718                         if (clearok) {
2719                             TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
2720                             LIST_REMOVE(dp, nfsdl_hash);
2721                             TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
2722                             nfscl_delegcnt--;
2723                             nfsstatsv1.cldelegates--;
2724                         }
2725                     }
2726                     dp = ndp;
2727                 }
2728                 if (igotlock)
2729                         nfsv4_unlock(&clp->nfsc_lock, 0);
2730
2731                 /*
2732                  * Do the recall on any layouts. To avoid trouble, always
2733                  * come back up here after having slept.
2734                  */
2735                 TAILQ_INIT(&rlh);
2736 tryagain2:
2737                 TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp) {
2738                         if ((lyp->nfsly_flags & NFSLY_RECALL) != 0) {
2739                                 /*
2740                                  * Wait for outstanding I/O ops to be done.
2741                                  */
2742                                 if (lyp->nfsly_lock.nfslock_usecnt > 0 ||
2743                                     (lyp->nfsly_lock.nfslock_lock &
2744                                      NFSV4LOCK_LOCK) != 0) {
2745                                         lyp->nfsly_lock.nfslock_lock |=
2746                                             NFSV4LOCK_WANTED;
2747                                         msleep(&lyp->nfsly_lock.nfslock_lock,
2748                                             NFSCLSTATEMUTEXPTR, PVFS, "nfslyp",
2749                                             5 * hz);
2750                                         if (NFSCL_FORCEDISM(mp))
2751                                             goto terminate;
2752                                         goto tryagain2;
2753                                 }
2754                                 /* Move the layout to the recall list. */
2755                                 TAILQ_REMOVE(&clp->nfsc_layout, lyp,
2756                                     nfsly_list);
2757                                 LIST_REMOVE(lyp, nfsly_hash);
2758                                 TAILQ_INSERT_HEAD(&rlh, lyp, nfsly_list);
2759
2760                                 /* Handle any layout commits. */
2761                                 if (!NFSHASNOLAYOUTCOMMIT(clp->nfsc_nmp) &&
2762                                     (lyp->nfsly_flags & NFSLY_WRITTEN) != 0) {
2763                                         lyp->nfsly_flags &= ~NFSLY_WRITTEN;
2764                                         NFSUNLOCKCLSTATE();
2765                                         NFSCL_DEBUG(3, "do layoutcommit\n");
2766                                         nfscl_dolayoutcommit(clp->nfsc_nmp, lyp,
2767                                             cred, p);
2768                                         NFSLOCKCLSTATE();
2769                                         goto tryagain2;
2770                                 }
2771                         }
2772                 }
2773
2774                 /* Now, look for stale layouts. */
2775                 lyp = TAILQ_LAST(&clp->nfsc_layout, nfscllayouthead);
2776                 while (lyp != NULL) {
2777                         nlyp = TAILQ_PREV(lyp, nfscllayouthead, nfsly_list);
2778                         if (lyp->nfsly_timestamp < NFSD_MONOSEC &&
2779                             (lyp->nfsly_flags & NFSLY_RECALL) == 0 &&
2780                             lyp->nfsly_lock.nfslock_usecnt == 0 &&
2781                             lyp->nfsly_lock.nfslock_lock == 0) {
2782                                 NFSCL_DEBUG(4, "ret stale lay=%d\n",
2783                                     nfscl_layoutcnt);
2784                                 recallp = malloc(sizeof(*recallp),
2785                                     M_NFSLAYRECALL, M_NOWAIT);
2786                                 if (recallp == NULL)
2787                                         break;
2788                                 (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE,
2789                                     lyp, NFSLAYOUTIOMODE_ANY, 0, UINT64_MAX,
2790                                     lyp->nfsly_stateid.seqid, 0, 0, NULL,
2791                                     recallp);
2792                         }
2793                         lyp = nlyp;
2794                 }
2795
2796                 /*
2797                  * Free up any unreferenced device info structures.
2798                  */
2799                 LIST_FOREACH_SAFE(dip, &clp->nfsc_devinfo, nfsdi_list, ndip) {
2800                         if (dip->nfsdi_layoutrefs == 0 &&
2801                             dip->nfsdi_refcnt == 0) {
2802                                 NFSCL_DEBUG(4, "freeing devinfo\n");
2803                                 LIST_REMOVE(dip, nfsdi_list);
2804                                 nfscl_freedevinfo(dip);
2805                         }
2806                 }
2807                 NFSUNLOCKCLSTATE();
2808
2809                 /* Do layout return(s), as required. */
2810                 TAILQ_FOREACH_SAFE(lyp, &rlh, nfsly_list, nlyp) {
2811                         TAILQ_REMOVE(&rlh, lyp, nfsly_list);
2812                         NFSCL_DEBUG(4, "ret layout\n");
2813                         nfscl_layoutreturn(clp->nfsc_nmp, lyp, cred, p);
2814                         nfscl_freelayout(lyp);
2815                 }
2816
2817                 /*
2818                  * Delegreturn any delegations cleaned out or recalled.
2819                  */
2820                 TAILQ_FOREACH_SAFE(dp, &dh, nfsdl_list, ndp) {
2821                         newnfs_copycred(&dp->nfsdl_cred, cred);
2822                         (void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
2823                         TAILQ_REMOVE(&dh, dp, nfsdl_list);
2824                         free(dp, M_NFSCLDELEG);
2825                 }
2826
2827                 SLIST_INIT(&lfh);
2828                 /*
2829                  * Call nfscl_cleanupkext() once per second to check for
2830                  * open/lock owners where the process has exited.
2831                  */
2832                 mytime = NFSD_MONOSEC;
2833                 if (prevsec != mytime) {
2834                         prevsec = mytime;
2835                         nfscl_cleanupkext(clp, &lfh);
2836                 }
2837
2838                 /*
2839                  * Do a ReleaseLockOwner for all lock owners where the
2840                  * associated process no longer exists, as found by
2841                  * nfscl_cleanupkext().
2842                  */
2843                 newnfs_setroot(cred);
2844                 SLIST_FOREACH_SAFE(lfhp, &lfh, nfslfh_list, nlfhp) {
2845                         LIST_FOREACH_SAFE(lp, &lfhp->nfslfh_lock, nfsl_list,
2846                             nlp) {
2847                                 (void)nfsrpc_rellockown(clp->nfsc_nmp, lp,
2848                                     lfhp->nfslfh_fh, lfhp->nfslfh_len, cred,
2849                                     p);
2850                                 nfscl_freelockowner(lp, 0);
2851                         }
2852                         free(lfhp, M_TEMP);
2853                 }
2854                 SLIST_INIT(&lfh);
2855
2856                 NFSLOCKCLSTATE();
2857                 if ((clp->nfsc_flags & NFSCLFLAGS_RECOVER) == 0)
2858                         (void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT, "nfscl",
2859                             hz);
2860 terminate:
2861                 if (clp->nfsc_flags & NFSCLFLAGS_UMOUNT) {
2862                         clp->nfsc_flags &= ~NFSCLFLAGS_HASTHREAD;
2863                         NFSUNLOCKCLSTATE();
2864                         NFSFREECRED(cred);
2865                         wakeup((caddr_t)clp);
2866                         return;
2867                 }
2868                 NFSUNLOCKCLSTATE();
2869         }
2870 }
2871
2872 /*
2873  * Initiate state recovery. Called when NFSERR_STALECLIENTID,
2874  * NFSERR_STALESTATEID or NFSERR_BADSESSION is received.
2875  */
2876 void
2877 nfscl_initiate_recovery(struct nfsclclient *clp)
2878 {
2879
2880         if (clp == NULL)
2881                 return;
2882         NFSLOCKCLSTATE();
2883         clp->nfsc_flags |= NFSCLFLAGS_RECOVER;
2884         NFSUNLOCKCLSTATE();
2885         wakeup((caddr_t)clp);
2886 }
2887
2888 /*
2889  * Dump out the state stuff for debugging.
2890  */
2891 void
2892 nfscl_dumpstate(struct nfsmount *nmp, int openowner, int opens,
2893     int lockowner, int locks)
2894 {
2895         struct nfsclclient *clp;
2896         struct nfsclowner *owp;
2897         struct nfsclopen *op;
2898         struct nfscllockowner *lp;
2899         struct nfscllock *lop;
2900         struct nfscldeleg *dp;
2901
2902         clp = nmp->nm_clp;
2903         if (clp == NULL) {
2904                 printf("nfscl dumpstate NULL clp\n");
2905                 return;
2906         }
2907         NFSLOCKCLSTATE();
2908         TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2909           LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
2910             if (openowner && !LIST_EMPTY(&owp->nfsow_open))
2911                 printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
2912                     owp->nfsow_owner[0], owp->nfsow_owner[1],
2913                     owp->nfsow_owner[2], owp->nfsow_owner[3],
2914                     owp->nfsow_seqid);
2915             LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
2916                 if (opens)
2917                     printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
2918                         op->nfso_stateid.other[0], op->nfso_stateid.other[1],
2919                         op->nfso_stateid.other[2], op->nfso_opencnt,
2920                         op->nfso_fh[12]);
2921                 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
2922                     if (lockowner)
2923                         printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
2924                             lp->nfsl_owner[0], lp->nfsl_owner[1],
2925                             lp->nfsl_owner[2], lp->nfsl_owner[3],
2926                             lp->nfsl_seqid,
2927                             lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
2928                             lp->nfsl_stateid.other[2]);
2929                     LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
2930                         if (locks)
2931 #ifdef __FreeBSD__
2932                             printf("lck typ=%d fst=%ju end=%ju\n",
2933                                 lop->nfslo_type, (intmax_t)lop->nfslo_first,
2934                                 (intmax_t)lop->nfslo_end);
2935 #else
2936                             printf("lck typ=%d fst=%qd end=%qd\n",
2937                                 lop->nfslo_type, lop->nfslo_first,
2938                                 lop->nfslo_end);
2939 #endif
2940                     }
2941                 }
2942             }
2943           }
2944         }
2945         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
2946             if (openowner && !LIST_EMPTY(&owp->nfsow_open))
2947                 printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
2948                     owp->nfsow_owner[0], owp->nfsow_owner[1],
2949                     owp->nfsow_owner[2], owp->nfsow_owner[3],
2950                     owp->nfsow_seqid);
2951             LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
2952                 if (opens)
2953                     printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
2954                         op->nfso_stateid.other[0], op->nfso_stateid.other[1],
2955                         op->nfso_stateid.other[2], op->nfso_opencnt,
2956                         op->nfso_fh[12]);
2957                 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
2958                     if (lockowner)
2959                         printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
2960                             lp->nfsl_owner[0], lp->nfsl_owner[1],
2961                             lp->nfsl_owner[2], lp->nfsl_owner[3],
2962                             lp->nfsl_seqid,
2963                             lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
2964                             lp->nfsl_stateid.other[2]);
2965                     LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
2966                         if (locks)
2967 #ifdef __FreeBSD__
2968                             printf("lck typ=%d fst=%ju end=%ju\n",
2969                                 lop->nfslo_type, (intmax_t)lop->nfslo_first,
2970                                 (intmax_t)lop->nfslo_end);
2971 #else
2972                             printf("lck typ=%d fst=%qd end=%qd\n",
2973                                 lop->nfslo_type, lop->nfslo_first,
2974                                 lop->nfslo_end);
2975 #endif
2976                     }
2977                 }
2978             }
2979         }
2980         NFSUNLOCKCLSTATE();
2981 }
2982
2983 /*
2984  * Check for duplicate open owners and opens.
2985  * (Only used as a diagnostic aid.)
2986  */
2987 void
2988 nfscl_dupopen(vnode_t vp, int dupopens)
2989 {
2990         struct nfsclclient *clp;
2991         struct nfsclowner *owp, *owp2;
2992         struct nfsclopen *op, *op2;
2993         struct nfsfh *nfhp;
2994
2995         clp = VFSTONFS(vnode_mount(vp))->nm_clp;
2996         if (clp == NULL) {
2997                 printf("nfscl dupopen NULL clp\n");
2998                 return;
2999         }
3000         nfhp = VTONFS(vp)->n_fhp;
3001         NFSLOCKCLSTATE();
3002
3003         /*
3004          * First, search for duplicate owners.
3005          * These should never happen!
3006          */
3007         LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3008             LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3009                 if (owp != owp2 &&
3010                     !NFSBCMP(owp->nfsow_owner, owp2->nfsow_owner,
3011                     NFSV4CL_LOCKNAMELEN)) {
3012                         NFSUNLOCKCLSTATE();
3013                         printf("DUP OWNER\n");
3014                         nfscl_dumpstate(VFSTONFS(vnode_mount(vp)), 1, 1, 0, 0);
3015                         return;
3016                 }
3017             }
3018         }
3019
3020         /*
3021          * Now, search for duplicate stateids.
3022          * These shouldn't happen, either.
3023          */
3024         LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3025             LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
3026                 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3027                     LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3028                         if (op != op2 &&
3029                             (op->nfso_stateid.other[0] != 0 ||
3030                              op->nfso_stateid.other[1] != 0 ||
3031                              op->nfso_stateid.other[2] != 0) &&
3032                             op->nfso_stateid.other[0] == op2->nfso_stateid.other[0] &&
3033                             op->nfso_stateid.other[1] == op2->nfso_stateid.other[1] &&
3034                             op->nfso_stateid.other[2] == op2->nfso_stateid.other[2]) {
3035                             NFSUNLOCKCLSTATE();
3036                             printf("DUP STATEID\n");
3037                             nfscl_dumpstate(VFSTONFS(vnode_mount(vp)), 1, 1, 0,
3038                                 0);
3039                             return;
3040                         }
3041                     }
3042                 }
3043             }
3044         }
3045
3046         /*
3047          * Now search for duplicate opens.
3048          * Duplicate opens for the same owner
3049          * should never occur. Other duplicates are
3050          * possible and are checked for if "dupopens"
3051          * is true.
3052          */
3053         LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3054             LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
3055                 if (nfhp->nfh_len == op2->nfso_fhlen &&
3056                     !NFSBCMP(nfhp->nfh_fh, op2->nfso_fh, nfhp->nfh_len)) {
3057                     LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3058                         LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3059                             if (op != op2 && nfhp->nfh_len == op->nfso_fhlen &&
3060                                 !NFSBCMP(nfhp->nfh_fh, op->nfso_fh, nfhp->nfh_len) &&
3061                                 (!NFSBCMP(op->nfso_own->nfsow_owner,
3062                                  op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN) ||
3063                                  dupopens)) {
3064                                 if (!NFSBCMP(op->nfso_own->nfsow_owner,
3065                                     op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
3066                                     NFSUNLOCKCLSTATE();
3067                                     printf("BADDUP OPEN\n");
3068                                 } else {
3069                                     NFSUNLOCKCLSTATE();
3070                                     printf("DUP OPEN\n");
3071                                 }
3072                                 nfscl_dumpstate(VFSTONFS(vnode_mount(vp)), 1, 1,
3073                                     0, 0);
3074                                 return;
3075                             }
3076                         }
3077                     }
3078                 }
3079             }
3080         }
3081         NFSUNLOCKCLSTATE();
3082 }
3083
3084 /*
3085  * During close, find an open that needs to be dereferenced and
3086  * dereference it. If there are no more opens for this file,
3087  * log a message to that effect.
3088  * Opens aren't actually Close'd until VOP_INACTIVE() is performed
3089  * on the file's vnode.
3090  * This is the safe way, since it is difficult to identify
3091  * which open the close is for and I/O can be performed after the
3092  * close(2) system call when a file is mmap'd.
3093  * If it returns 0 for success, there will be a referenced
3094  * clp returned via clpp.
3095  */
3096 int
3097 nfscl_getclose(vnode_t vp, struct nfsclclient **clpp)
3098 {
3099         struct nfsclclient *clp;
3100         struct nfsclowner *owp;
3101         struct nfsclopen *op;
3102         struct nfscldeleg *dp;
3103         struct nfsfh *nfhp;
3104         int error, notdecr;
3105
3106         error = nfscl_getcl(vnode_mount(vp), NULL, NULL, 1, &clp);
3107         if (error)
3108                 return (error);
3109         *clpp = clp;
3110
3111         nfhp = VTONFS(vp)->n_fhp;
3112         notdecr = 1;
3113         NFSLOCKCLSTATE();
3114         /*
3115          * First, look for one under a delegation that was locally issued
3116          * and just decrement the opencnt for it. Since all my Opens against
3117          * the server are DENY_NONE, I don't see a problem with hanging
3118          * onto them. (It is much easier to use one of the extant Opens
3119          * that I already have on the server when a Delegation is recalled
3120          * than to do fresh Opens.) Someday, I might need to rethink this, but.
3121          */
3122         dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
3123         if (dp != NULL) {
3124                 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
3125                         op = LIST_FIRST(&owp->nfsow_open);
3126                         if (op != NULL) {
3127                                 /*
3128                                  * Since a delegation is for a file, there
3129                                  * should never be more than one open for
3130                                  * each openowner.
3131                                  */
3132                                 if (LIST_NEXT(op, nfso_list) != NULL)
3133                                         panic("nfscdeleg opens");
3134                                 if (notdecr && op->nfso_opencnt > 0) {
3135                                         notdecr = 0;
3136                                         op->nfso_opencnt--;
3137                                         break;
3138                                 }
3139                         }
3140                 }
3141         }
3142
3143         /* Now process the opens against the server. */
3144         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3145                 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3146                         if (op->nfso_fhlen == nfhp->nfh_len &&
3147                             !NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
3148                             nfhp->nfh_len)) {
3149                                 /* Found an open, decrement cnt if possible */
3150                                 if (notdecr && op->nfso_opencnt > 0) {
3151                                         notdecr = 0;
3152                                         op->nfso_opencnt--;
3153                                 }
3154                                 /*
3155                                  * There are more opens, so just return.
3156                                  */
3157                                 if (op->nfso_opencnt > 0) {
3158                                         NFSUNLOCKCLSTATE();
3159                                         return (0);
3160                                 }
3161                         }
3162                 }
3163         }
3164         NFSUNLOCKCLSTATE();
3165         if (notdecr)
3166                 printf("nfscl: never fnd open\n");
3167         return (0);
3168 }
3169
3170 int
3171 nfscl_doclose(vnode_t vp, struct nfsclclient **clpp, NFSPROC_T *p)
3172 {
3173         struct nfsclclient *clp;
3174         struct nfsclowner *owp, *nowp;
3175         struct nfsclopen *op;
3176         struct nfscldeleg *dp;
3177         struct nfsfh *nfhp;
3178         struct nfsclrecalllayout *recallp;
3179         int error;
3180
3181         error = nfscl_getcl(vnode_mount(vp), NULL, NULL, 1, &clp);
3182         if (error)
3183                 return (error);
3184         *clpp = clp;
3185
3186         nfhp = VTONFS(vp)->n_fhp;
3187         recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_WAITOK);
3188         NFSLOCKCLSTATE();
3189         /*
3190          * First get rid of the local Open structures, which should be no
3191          * longer in use.
3192          */
3193         dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
3194         if (dp != NULL) {
3195                 LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) {
3196                         op = LIST_FIRST(&owp->nfsow_open);
3197                         if (op != NULL) {
3198                                 KASSERT((op->nfso_opencnt == 0),
3199                                     ("nfscl: bad open cnt on deleg"));
3200                                 nfscl_freeopen(op, 1);
3201                         }
3202                         nfscl_freeopenowner(owp, 1);
3203                 }
3204         }
3205
3206         /* Return any layouts marked return on close. */
3207         nfscl_retoncloselayout(vp, clp, nfhp->nfh_fh, nfhp->nfh_len, &recallp);
3208
3209         /* Now process the opens against the server. */
3210 lookformore:
3211         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3212                 op = LIST_FIRST(&owp->nfsow_open);
3213                 while (op != NULL) {
3214                         if (op->nfso_fhlen == nfhp->nfh_len &&
3215                             !NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
3216                             nfhp->nfh_len)) {
3217                                 /* Found an open, close it. */
3218                                 KASSERT((op->nfso_opencnt == 0),
3219                                     ("nfscl: bad open cnt on server"));
3220                                 NFSUNLOCKCLSTATE();
3221                                 nfsrpc_doclose(VFSTONFS(vnode_mount(vp)), op,
3222                                     p);
3223                                 NFSLOCKCLSTATE();
3224                                 goto lookformore;
3225                         }
3226                         op = LIST_NEXT(op, nfso_list);
3227                 }
3228         }
3229         NFSUNLOCKCLSTATE();
3230         /*
3231          * recallp has been set NULL by nfscl_retoncloselayout() if it was
3232          * used by the function, but calling free() with a NULL pointer is ok.
3233          */
3234         free(recallp, M_NFSLAYRECALL);
3235         return (0);
3236 }
3237
3238 /*
3239  * Return all delegations on this client.
3240  * (Must be called with client sleep lock.)
3241  */
3242 static void
3243 nfscl_delegreturnall(struct nfsclclient *clp, NFSPROC_T *p)
3244 {
3245         struct nfscldeleg *dp, *ndp;
3246         struct ucred *cred;
3247
3248         cred = newnfs_getcred();
3249         TAILQ_FOREACH_SAFE(dp, &clp->nfsc_deleg, nfsdl_list, ndp) {
3250                 nfscl_cleandeleg(dp);
3251                 (void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
3252                 nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
3253         }
3254         NFSFREECRED(cred);
3255 }
3256
3257 /*
3258  * Return any delegation for this vp.
3259  */
3260 void
3261 nfscl_delegreturnvp(vnode_t vp, NFSPROC_T *p)
3262 {
3263         struct nfsclclient *clp;
3264         struct nfscldeleg *dp;
3265         struct ucred *cred;
3266         struct nfsnode *np;
3267
3268         np = VTONFS(vp);
3269         cred = newnfs_getcred();
3270         dp = NULL;
3271         NFSLOCKCLSTATE();
3272         clp = VFSTONFS(vp->v_mount)->nm_clp;
3273         if (clp != NULL)
3274                 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
3275                     np->n_fhp->nfh_len);
3276         if (dp != NULL) {
3277                 nfscl_cleandeleg(dp);
3278                 nfscl_freedeleg(&clp->nfsc_deleg, dp, false);
3279                 NFSUNLOCKCLSTATE();
3280                 newnfs_copycred(&dp->nfsdl_cred, cred);
3281                 nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
3282                 free(dp, M_NFSCLDELEG);
3283         } else
3284                 NFSUNLOCKCLSTATE();
3285         NFSFREECRED(cred);
3286 }
3287
3288 /*
3289  * Do a callback RPC.
3290  */
3291 void
3292 nfscl_docb(struct nfsrv_descript *nd, NFSPROC_T *p)
3293 {
3294         int clist, gotseq_ok, i, j, k, op, rcalls;
3295         u_int32_t *tl;
3296         struct nfsclclient *clp;
3297         struct nfscldeleg *dp = NULL;
3298         int numops, taglen = -1, error = 0, trunc __unused;
3299         u_int32_t minorvers = 0, retops = 0, *retopsp = NULL, *repp, cbident;
3300         u_char tag[NFSV4_SMALLSTR + 1], *tagstr;
3301         vnode_t vp = NULL;
3302         struct nfsnode *np;
3303         struct vattr va;
3304         struct nfsfh *nfhp;
3305         mount_t mp;
3306         nfsattrbit_t attrbits, rattrbits;
3307         nfsv4stateid_t stateid;
3308         uint32_t seqid, slotid = 0, highslot, cachethis __unused;
3309         uint8_t sessionid[NFSX_V4SESSIONID];
3310         struct mbuf *rep;
3311         struct nfscllayout *lyp;
3312         uint64_t filesid[2], len, off;
3313         int changed, gotone, laytype, recalltype;
3314         uint32_t iomode;
3315         struct nfsclrecalllayout *recallp = NULL;
3316         struct nfsclsession *tsep;
3317
3318         gotseq_ok = 0;
3319         nfsrvd_rephead(nd);
3320         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3321         taglen = fxdr_unsigned(int, *tl);
3322         if (taglen < 0) {
3323                 error = EBADRPC;
3324                 goto nfsmout;
3325         }
3326         if (taglen <= NFSV4_SMALLSTR)
3327                 tagstr = tag;
3328         else
3329                 tagstr = malloc(taglen + 1, M_TEMP, M_WAITOK);
3330         error = nfsrv_mtostr(nd, tagstr, taglen);
3331         if (error) {
3332                 if (taglen > NFSV4_SMALLSTR)
3333                         free(tagstr, M_TEMP);
3334                 taglen = -1;
3335                 goto nfsmout;
3336         }
3337         (void) nfsm_strtom(nd, tag, taglen);
3338         if (taglen > NFSV4_SMALLSTR) {
3339                 free(tagstr, M_TEMP);
3340         }
3341         NFSM_BUILD(retopsp, u_int32_t *, NFSX_UNSIGNED);
3342         NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3343         minorvers = fxdr_unsigned(u_int32_t, *tl++);
3344         if (minorvers != NFSV4_MINORVERSION && minorvers != NFSV41_MINORVERSION)
3345                 nd->nd_repstat = NFSERR_MINORVERMISMATCH;
3346         cbident = fxdr_unsigned(u_int32_t, *tl++);
3347         if (nd->nd_repstat)
3348                 numops = 0;
3349         else
3350                 numops = fxdr_unsigned(int, *tl);
3351         /*
3352          * Loop around doing the sub ops.
3353          */
3354         for (i = 0; i < numops; i++) {
3355                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3356                 NFSM_BUILD(repp, u_int32_t *, 2 * NFSX_UNSIGNED);
3357                 *repp++ = *tl;
3358                 op = fxdr_unsigned(int, *tl);
3359                 if (op < NFSV4OP_CBGETATTR ||
3360                    (op > NFSV4OP_CBRECALL && minorvers == NFSV4_MINORVERSION) ||
3361                    (op > NFSV4OP_CBNOTIFYDEVID &&
3362                     minorvers == NFSV41_MINORVERSION)) {
3363                     nd->nd_repstat = NFSERR_OPILLEGAL;
3364                     *repp = nfscl_errmap(nd, minorvers);
3365                     retops++;
3366                     break;
3367                 }
3368                 nd->nd_procnum = op;
3369                 if (op < NFSV41_CBNOPS)
3370                         nfsstatsv1.cbrpccnt[nd->nd_procnum]++;
3371                 switch (op) {
3372                 case NFSV4OP_CBGETATTR:
3373                         NFSCL_DEBUG(4, "cbgetattr\n");
3374                         mp = NULL;
3375                         vp = NULL;
3376                         error = nfsm_getfh(nd, &nfhp);
3377                         if (!error)
3378                                 error = nfsrv_getattrbits(nd, &attrbits,
3379                                     NULL, NULL);
3380                         if (error == 0 && i == 0 &&
3381                             minorvers != NFSV4_MINORVERSION)
3382                                 error = NFSERR_OPNOTINSESS;
3383                         if (!error) {
3384                                 mp = nfscl_getmnt(minorvers, sessionid, cbident,
3385                                     &clp);
3386                                 if (mp == NULL)
3387                                         error = NFSERR_SERVERFAULT;
3388                         }
3389                         if (!error) {
3390                                 error = nfscl_ngetreopen(mp, nfhp->nfh_fh,
3391                                     nfhp->nfh_len, p, &np);
3392                                 if (!error)
3393                                         vp = NFSTOV(np);
3394                         }
3395                         if (!error) {
3396                                 NFSZERO_ATTRBIT(&rattrbits);
3397                                 NFSLOCKCLSTATE();
3398                                 dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
3399                                     nfhp->nfh_len);
3400                                 if (dp != NULL) {
3401                                         if (NFSISSET_ATTRBIT(&attrbits,
3402                                             NFSATTRBIT_SIZE)) {
3403                                                 if (vp != NULL)
3404                                                         va.va_size = np->n_size;
3405                                                 else
3406                                                         va.va_size =
3407                                                             dp->nfsdl_size;
3408                                                 NFSSETBIT_ATTRBIT(&rattrbits,
3409                                                     NFSATTRBIT_SIZE);
3410                                         }
3411                                         if (NFSISSET_ATTRBIT(&attrbits,
3412                                             NFSATTRBIT_CHANGE)) {
3413                                                 va.va_filerev =
3414                                                     dp->nfsdl_change;
3415                                                 if (vp == NULL ||
3416                                                     (np->n_flag & NDELEGMOD))
3417                                                         va.va_filerev++;
3418                                                 NFSSETBIT_ATTRBIT(&rattrbits,
3419                                                     NFSATTRBIT_CHANGE);
3420                                         }
3421                                 } else
3422                                         error = NFSERR_SERVERFAULT;
3423                                 NFSUNLOCKCLSTATE();
3424                         }
3425                         if (vp != NULL)
3426                                 vrele(vp);
3427                         if (mp != NULL)
3428                                 vfs_unbusy(mp);
3429                         if (nfhp != NULL)
3430                                 free(nfhp, M_NFSFH);
3431                         if (!error)
3432                                 (void) nfsv4_fillattr(nd, NULL, NULL, NULL, &va,
3433                                     NULL, 0, &rattrbits, NULL, p, 0, 0, 0, 0,
3434                                     (uint64_t)0, NULL);
3435                         break;
3436                 case NFSV4OP_CBRECALL:
3437                         NFSCL_DEBUG(4, "cbrecall\n");
3438                         NFSM_DISSECT(tl, u_int32_t *, NFSX_STATEID +
3439                             NFSX_UNSIGNED);
3440                         stateid.seqid = *tl++;
3441                         NFSBCOPY((caddr_t)tl, (caddr_t)stateid.other,
3442                             NFSX_STATEIDOTHER);
3443                         tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
3444                         trunc = fxdr_unsigned(int, *tl);
3445                         error = nfsm_getfh(nd, &nfhp);
3446                         if (error == 0 && i == 0 &&
3447                             minorvers != NFSV4_MINORVERSION)
3448                                 error = NFSERR_OPNOTINSESS;
3449                         if (!error) {
3450                                 NFSLOCKCLSTATE();
3451                                 if (minorvers == NFSV4_MINORVERSION)
3452                                         clp = nfscl_getclnt(cbident);
3453                                 else
3454                                         clp = nfscl_getclntsess(sessionid);
3455                                 if (clp != NULL) {
3456                                         dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
3457                                             nfhp->nfh_len);
3458                                         if (dp != NULL && (dp->nfsdl_flags &
3459                                             NFSCLDL_DELEGRET) == 0) {
3460                                                 dp->nfsdl_flags |=
3461                                                     NFSCLDL_RECALL;
3462                                                 wakeup((caddr_t)clp);
3463                                         }
3464                                 } else {
3465                                         error = NFSERR_SERVERFAULT;
3466                                 }
3467                                 NFSUNLOCKCLSTATE();
3468                         }
3469                         if (nfhp != NULL)
3470                                 free(nfhp, M_NFSFH);
3471                         break;
3472                 case NFSV4OP_CBLAYOUTRECALL:
3473                         NFSCL_DEBUG(4, "cblayrec\n");
3474                         nfhp = NULL;
3475                         NFSM_DISSECT(tl, uint32_t *, 4 * NFSX_UNSIGNED);
3476                         laytype = fxdr_unsigned(int, *tl++);
3477                         iomode = fxdr_unsigned(uint32_t, *tl++);
3478                         if (newnfs_true == *tl++)
3479                                 changed = 1;
3480                         else
3481                                 changed = 0;
3482                         recalltype = fxdr_unsigned(int, *tl);
3483                         NFSCL_DEBUG(4, "layt=%d iom=%d ch=%d rectyp=%d\n",
3484                             laytype, iomode, changed, recalltype);
3485                         recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL,
3486                             M_WAITOK);
3487                         if (laytype != NFSLAYOUT_NFSV4_1_FILES &&
3488                             laytype != NFSLAYOUT_FLEXFILE)
3489                                 error = NFSERR_NOMATCHLAYOUT;
3490                         else if (recalltype == NFSLAYOUTRETURN_FILE) {
3491                                 error = nfsm_getfh(nd, &nfhp);
3492                                 NFSCL_DEBUG(4, "retfile getfh=%d\n", error);
3493                                 if (error != 0)
3494                                         goto nfsmout;
3495                                 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_HYPER +
3496                                     NFSX_STATEID);
3497                                 off = fxdr_hyper(tl); tl += 2;
3498                                 len = fxdr_hyper(tl); tl += 2;
3499                                 stateid.seqid = fxdr_unsigned(uint32_t, *tl++);
3500                                 NFSBCOPY(tl, stateid.other, NFSX_STATEIDOTHER);
3501                                 if (minorvers == NFSV4_MINORVERSION)
3502                                         error = NFSERR_NOTSUPP;
3503                                 else if (i == 0)
3504                                         error = NFSERR_OPNOTINSESS;
3505                                 NFSCL_DEBUG(4, "off=%ju len=%ju sq=%u err=%d\n",
3506                                     (uintmax_t)off, (uintmax_t)len,
3507                                     stateid.seqid, error);
3508                                 if (error == 0) {
3509                                         NFSLOCKCLSTATE();
3510                                         clp = nfscl_getclntsess(sessionid);
3511                                         NFSCL_DEBUG(4, "cbly clp=%p\n", clp);
3512                                         if (clp != NULL) {
3513                                                 lyp = nfscl_findlayout(clp,
3514                                                     nfhp->nfh_fh,
3515                                                     nfhp->nfh_len);
3516                                                 NFSCL_DEBUG(4, "cblyp=%p\n",
3517                                                     lyp);
3518                                                 if (lyp != NULL &&
3519                                                     (lyp->nfsly_flags &
3520                                                      (NFSLY_FILES |
3521                                                       NFSLY_FLEXFILE)) != 0 &&
3522                                                     !NFSBCMP(stateid.other,
3523                                                     lyp->nfsly_stateid.other,
3524                                                     NFSX_STATEIDOTHER)) {
3525                                                         error =
3526                                                             nfscl_layoutrecall(
3527                                                             recalltype,
3528                                                             lyp, iomode, off,
3529                                                             len, stateid.seqid,
3530                                                             0, 0, NULL,
3531                                                             recallp);
3532                                                         if (error == 0 &&
3533                                                             stateid.seqid >
3534                                                             lyp->nfsly_stateid.seqid)
3535                                                                 lyp->nfsly_stateid.seqid =
3536                                                                     stateid.seqid;
3537                                                         recallp = NULL;
3538                                                         wakeup(clp);
3539                                                         NFSCL_DEBUG(4,
3540                                                             "aft layrcal=%d "
3541                                                             "layseqid=%d\n",
3542                                                             error,
3543                                                             lyp->nfsly_stateid.seqid);
3544                                                 } else
3545                                                         error =
3546                                                           NFSERR_NOMATCHLAYOUT;
3547                                         } else
3548                                                 error = NFSERR_NOMATCHLAYOUT;
3549                                         NFSUNLOCKCLSTATE();
3550                                 }
3551                                 free(nfhp, M_NFSFH);
3552                         } else if (recalltype == NFSLAYOUTRETURN_FSID) {
3553                                 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_HYPER);
3554                                 filesid[0] = fxdr_hyper(tl); tl += 2;
3555                                 filesid[1] = fxdr_hyper(tl); tl += 2;
3556                                 gotone = 0;
3557                                 NFSLOCKCLSTATE();
3558                                 clp = nfscl_getclntsess(sessionid);
3559                                 if (clp != NULL) {
3560                                         TAILQ_FOREACH(lyp, &clp->nfsc_layout,
3561                                             nfsly_list) {
3562                                                 if (lyp->nfsly_filesid[0] ==
3563                                                     filesid[0] &&
3564                                                     lyp->nfsly_filesid[1] ==
3565                                                     filesid[1]) {
3566                                                         error =
3567                                                             nfscl_layoutrecall(
3568                                                             recalltype,
3569                                                             lyp, iomode, 0,
3570                                                             UINT64_MAX,
3571                                                             lyp->nfsly_stateid.seqid,
3572                                                             0, 0, NULL,
3573                                                             recallp);
3574                                                         recallp = NULL;
3575                                                         gotone = 1;
3576                                                 }
3577                                         }
3578                                         if (gotone != 0)
3579                                                 wakeup(clp);
3580                                         else
3581                                                 error = NFSERR_NOMATCHLAYOUT;
3582                                 } else
3583                                         error = NFSERR_NOMATCHLAYOUT;
3584                                 NFSUNLOCKCLSTATE();
3585                         } else if (recalltype == NFSLAYOUTRETURN_ALL) {
3586                                 gotone = 0;
3587                                 NFSLOCKCLSTATE();
3588                                 clp = nfscl_getclntsess(sessionid);
3589                                 if (clp != NULL) {
3590                                         TAILQ_FOREACH(lyp, &clp->nfsc_layout,
3591                                             nfsly_list) {
3592                                                 error = nfscl_layoutrecall(
3593                                                     recalltype, lyp, iomode, 0,
3594                                                     UINT64_MAX,
3595                                                     lyp->nfsly_stateid.seqid,
3596                                                     0, 0, NULL, recallp);
3597                                                 recallp = NULL;
3598                                                 gotone = 1;
3599                                         }
3600                                         if (gotone != 0)
3601                                                 wakeup(clp);
3602                                         else
3603                                                 error = NFSERR_NOMATCHLAYOUT;
3604                                 } else
3605                                         error = NFSERR_NOMATCHLAYOUT;
3606                                 NFSUNLOCKCLSTATE();
3607                         } else
3608                                 error = NFSERR_NOMATCHLAYOUT;
3609                         if (recallp != NULL) {
3610                                 free(recallp, M_NFSLAYRECALL);
3611                                 recallp = NULL;
3612                         }
3613                         break;
3614                 case NFSV4OP_CBSEQUENCE:
3615                         NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID +
3616                             5 * NFSX_UNSIGNED);
3617                         bcopy(tl, sessionid, NFSX_V4SESSIONID);
3618                         tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3619                         seqid = fxdr_unsigned(uint32_t, *tl++);
3620                         slotid = fxdr_unsigned(uint32_t, *tl++);
3621                         highslot = fxdr_unsigned(uint32_t, *tl++);
3622                         cachethis = *tl++;
3623                         /* Throw away the referring call stuff. */
3624                         clist = fxdr_unsigned(int, *tl);
3625                         for (j = 0; j < clist; j++) {
3626                                 NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID +
3627                                     NFSX_UNSIGNED);
3628                                 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3629                                 rcalls = fxdr_unsigned(int, *tl);
3630                                 for (k = 0; k < rcalls; k++) {
3631                                         NFSM_DISSECT(tl, uint32_t *,
3632                                             2 * NFSX_UNSIGNED);
3633                                 }
3634                         }
3635                         NFSLOCKCLSTATE();
3636                         if (i == 0) {
3637                                 clp = nfscl_getclntsess(sessionid);
3638                                 if (clp == NULL)
3639                                         error = NFSERR_SERVERFAULT;
3640                         } else
3641                                 error = NFSERR_SEQUENCEPOS;
3642                         if (error == 0) {
3643                                 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3644                                 error = nfsv4_seqsession(seqid, slotid,
3645                                     highslot, tsep->nfsess_cbslots, &rep,
3646                                     tsep->nfsess_backslots);
3647                         }
3648                         NFSUNLOCKCLSTATE();
3649                         if (error == 0 || error == NFSERR_REPLYFROMCACHE) {
3650                                 gotseq_ok = 1;
3651                                 if (rep != NULL) {
3652                                         /*
3653                                          * Handle a reply for a retried
3654                                          * callback.  The reply will be
3655                                          * re-inserted in the session cache
3656                                          * by the nfsv4_seqsess_cacherep() call
3657                                          * after out:
3658                                          */
3659                                         KASSERT(error == NFSERR_REPLYFROMCACHE,
3660                                             ("cbsequence: non-NULL rep"));
3661                                         NFSCL_DEBUG(4, "Got cbretry\n");
3662                                         m_freem(nd->nd_mreq);
3663                                         nd->nd_mreq = rep;
3664                                         rep = NULL;
3665                                         goto out;
3666                                 }
3667                                 NFSM_BUILD(tl, uint32_t *,
3668                                     NFSX_V4SESSIONID + 4 * NFSX_UNSIGNED);
3669                                 bcopy(sessionid, tl, NFSX_V4SESSIONID);
3670                                 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3671                                 *tl++ = txdr_unsigned(seqid);
3672                                 *tl++ = txdr_unsigned(slotid);
3673                                 *tl++ = txdr_unsigned(NFSV4_CBSLOTS - 1);
3674                                 *tl = txdr_unsigned(NFSV4_CBSLOTS - 1);
3675                         }
3676                         break;
3677                 default:
3678                         if (i == 0 && minorvers == NFSV41_MINORVERSION)
3679                                 error = NFSERR_OPNOTINSESS;
3680                         else {
3681                                 NFSCL_DEBUG(1, "unsupp callback %d\n", op);
3682                                 error = NFSERR_NOTSUPP;
3683                         }
3684                         break;
3685                 }
3686                 if (error) {
3687                         if (error == EBADRPC || error == NFSERR_BADXDR) {
3688                                 nd->nd_repstat = NFSERR_BADXDR;
3689                         } else {
3690                                 nd->nd_repstat = error;
3691                         }
3692                         error = 0;
3693                 }
3694                 retops++;
3695                 if (nd->nd_repstat) {
3696                         *repp = nfscl_errmap(nd, minorvers);
3697                         break;
3698                 } else
3699                         *repp = 0;      /* NFS4_OK */
3700         }
3701 nfsmout:
3702         if (recallp != NULL)
3703                 free(recallp, M_NFSLAYRECALL);
3704         if (error) {
3705                 if (error == EBADRPC || error == NFSERR_BADXDR)
3706                         nd->nd_repstat = NFSERR_BADXDR;
3707                 else
3708                         printf("nfsv4 comperr1=%d\n", error);
3709         }
3710         if (taglen == -1) {
3711                 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
3712                 *tl++ = 0;
3713                 *tl = 0;
3714         } else {
3715                 *retopsp = txdr_unsigned(retops);
3716         }
3717         *nd->nd_errp = nfscl_errmap(nd, minorvers);
3718 out:
3719         if (gotseq_ok != 0) {
3720                 rep = m_copym(nd->nd_mreq, 0, M_COPYALL, M_WAITOK);
3721                 NFSLOCKCLSTATE();
3722                 clp = nfscl_getclntsess(sessionid);
3723                 if (clp != NULL) {
3724                         tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3725                         nfsv4_seqsess_cacherep(slotid, tsep->nfsess_cbslots,
3726                             NFSERR_OK, &rep);
3727                         NFSUNLOCKCLSTATE();
3728                 } else {
3729                         NFSUNLOCKCLSTATE();
3730                         m_freem(rep);
3731                 }
3732         }
3733 }
3734
3735 /*
3736  * Generate the next cbident value. Basically just increment a static value
3737  * and then check that it isn't already in the list, if it has wrapped around.
3738  */
3739 static u_int32_t
3740 nfscl_nextcbident(void)
3741 {
3742         struct nfsclclient *clp;
3743         int matched;
3744         static u_int32_t nextcbident = 0;
3745         static int haswrapped = 0;
3746
3747         nextcbident++;
3748         if (nextcbident == 0)
3749                 haswrapped = 1;
3750         if (haswrapped) {
3751                 /*
3752                  * Search the clientid list for one already using this cbident.
3753                  */
3754                 do {
3755                         matched = 0;
3756                         NFSLOCKCLSTATE();
3757                         LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3758                                 if (clp->nfsc_cbident == nextcbident) {
3759                                         matched = 1;
3760                                         break;
3761                                 }
3762                         }
3763                         NFSUNLOCKCLSTATE();
3764                         if (matched == 1)
3765                                 nextcbident++;
3766                 } while (matched);
3767         }
3768         return (nextcbident);
3769 }
3770
3771 /*
3772  * Get the mount point related to a given cbident or session and busy it.
3773  */
3774 static mount_t
3775 nfscl_getmnt(int minorvers, uint8_t *sessionid, u_int32_t cbident,
3776     struct nfsclclient **clpp)
3777 {
3778         struct nfsclclient *clp;
3779         mount_t mp;
3780         int error;
3781         struct nfsclsession *tsep;
3782
3783         *clpp = NULL;
3784         NFSLOCKCLSTATE();
3785         LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3786                 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3787                 if (minorvers == NFSV4_MINORVERSION) {
3788                         if (clp->nfsc_cbident == cbident)
3789                                 break;
3790                 } else if (!NFSBCMP(tsep->nfsess_sessionid, sessionid,
3791                     NFSX_V4SESSIONID))
3792                         break;
3793         }
3794         if (clp == NULL) {
3795                 NFSUNLOCKCLSTATE();
3796                 return (NULL);
3797         }
3798         mp = clp->nfsc_nmp->nm_mountp;
3799         vfs_ref(mp);
3800         NFSUNLOCKCLSTATE();
3801         error = vfs_busy(mp, 0);
3802         vfs_rel(mp);
3803         if (error != 0)
3804                 return (NULL);
3805         *clpp = clp;
3806         return (mp);
3807 }
3808
3809 /*
3810  * Get the clientid pointer related to a given cbident.
3811  */
3812 static struct nfsclclient *
3813 nfscl_getclnt(u_int32_t cbident)
3814 {
3815         struct nfsclclient *clp;
3816
3817         LIST_FOREACH(clp, &nfsclhead, nfsc_list)
3818                 if (clp->nfsc_cbident == cbident)
3819                         break;
3820         return (clp);
3821 }
3822
3823 /*
3824  * Get the clientid pointer related to a given sessionid.
3825  */
3826 static struct nfsclclient *
3827 nfscl_getclntsess(uint8_t *sessionid)
3828 {
3829         struct nfsclclient *clp;
3830         struct nfsclsession *tsep;
3831
3832         LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3833                 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3834                 if (!NFSBCMP(tsep->nfsess_sessionid, sessionid,
3835                     NFSX_V4SESSIONID))
3836                         break;
3837         }
3838         return (clp);
3839 }
3840
3841 /*
3842  * Search for a lock conflict locally on the client. A conflict occurs if
3843  * - not same owner and overlapping byte range and at least one of them is
3844  *   a write lock or this is an unlock.
3845  */
3846 static int
3847 nfscl_localconflict(struct nfsclclient *clp, u_int8_t *fhp, int fhlen,
3848     struct nfscllock *nlop, u_int8_t *own, struct nfscldeleg *dp,
3849     struct nfscllock **lopp)
3850 {
3851         struct nfsclowner *owp;
3852         struct nfsclopen *op;
3853         int ret;
3854
3855         if (dp != NULL) {
3856                 ret = nfscl_checkconflict(&dp->nfsdl_lock, nlop, own, lopp);
3857                 if (ret)
3858                         return (ret);
3859         }
3860         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3861                 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3862                         if (op->nfso_fhlen == fhlen &&
3863                             !NFSBCMP(op->nfso_fh, fhp, fhlen)) {
3864                                 ret = nfscl_checkconflict(&op->nfso_lock, nlop,
3865                                     own, lopp);
3866                                 if (ret)
3867                                         return (ret);
3868                         }
3869                 }
3870         }
3871         return (0);
3872 }
3873
3874 static int
3875 nfscl_checkconflict(struct nfscllockownerhead *lhp, struct nfscllock *nlop,
3876     u_int8_t *own, struct nfscllock **lopp)
3877 {
3878         struct nfscllockowner *lp;
3879         struct nfscllock *lop;
3880
3881         LIST_FOREACH(lp, lhp, nfsl_list) {
3882                 if (NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
3883                         LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
3884                                 if (lop->nfslo_first >= nlop->nfslo_end)
3885                                         break;
3886                                 if (lop->nfslo_end <= nlop->nfslo_first)
3887                                         continue;
3888                                 if (lop->nfslo_type == F_WRLCK ||
3889                                     nlop->nfslo_type == F_WRLCK ||
3890                                     nlop->nfslo_type == F_UNLCK) {
3891                                         if (lopp != NULL)
3892                                                 *lopp = lop;
3893                                         return (NFSERR_DENIED);
3894                                 }
3895                         }
3896                 }
3897         }
3898         return (0);
3899 }
3900
3901 /*
3902  * Check for a local conflicting lock.
3903  */
3904 int
3905 nfscl_lockt(vnode_t vp, struct nfsclclient *clp, u_int64_t off,
3906     u_int64_t len, struct flock *fl, NFSPROC_T *p, void *id, int flags)
3907 {
3908         struct nfscllock *lop, nlck;
3909         struct nfscldeleg *dp;
3910         struct nfsnode *np;
3911         u_int8_t own[NFSV4CL_LOCKNAMELEN];
3912         int error;
3913
3914         nlck.nfslo_type = fl->l_type;
3915         nlck.nfslo_first = off;
3916         if (len == NFS64BITSSET) {
3917                 nlck.nfslo_end = NFS64BITSSET;
3918         } else {
3919                 nlck.nfslo_end = off + len;
3920                 if (nlck.nfslo_end <= nlck.nfslo_first)
3921                         return (NFSERR_INVAL);
3922         }
3923         np = VTONFS(vp);
3924         nfscl_filllockowner(id, own, flags);
3925         NFSLOCKCLSTATE();
3926         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
3927         error = nfscl_localconflict(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len,
3928             &nlck, own, dp, &lop);
3929         if (error != 0) {
3930                 fl->l_whence = SEEK_SET;
3931                 fl->l_start = lop->nfslo_first;
3932                 if (lop->nfslo_end == NFS64BITSSET)
3933                         fl->l_len = 0;
3934                 else
3935                         fl->l_len = lop->nfslo_end - lop->nfslo_first;
3936                 fl->l_pid = (pid_t)0;
3937                 fl->l_type = lop->nfslo_type;
3938                 error = -1;                     /* no RPC required */
3939         } else if (dp != NULL && ((dp->nfsdl_flags & NFSCLDL_WRITE) ||
3940             fl->l_type == F_RDLCK)) {
3941                 /*
3942                  * The delegation ensures that there isn't a conflicting
3943                  * lock on the server, so return -1 to indicate an RPC
3944                  * isn't required.
3945                  */
3946                 fl->l_type = F_UNLCK;
3947                 error = -1;
3948         }
3949         NFSUNLOCKCLSTATE();
3950         return (error);
3951 }
3952
3953 /*
3954  * Handle Recall of a delegation.
3955  * The clp must be exclusive locked when this is called.
3956  */
3957 static int
3958 nfscl_recalldeleg(struct nfsclclient *clp, struct nfsmount *nmp,
3959     struct nfscldeleg *dp, vnode_t vp, struct ucred *cred, NFSPROC_T *p,
3960     int called_from_renewthread, vnode_t *vpp)
3961 {
3962         struct nfsclowner *owp, *lowp, *nowp;
3963         struct nfsclopen *op, *lop;
3964         struct nfscllockowner *lp;
3965         struct nfscllock *lckp;
3966         struct nfsnode *np;
3967         int error = 0, ret;
3968
3969         if (vp == NULL) {
3970                 KASSERT(vpp != NULL, ("nfscl_recalldeleg: vpp NULL"));
3971                 *vpp = NULL;
3972                 /*
3973                  * First, get a vnode for the file. This is needed to do RPCs.
3974                  */
3975                 ret = nfscl_ngetreopen(nmp->nm_mountp, dp->nfsdl_fh,
3976                     dp->nfsdl_fhlen, p, &np);
3977                 if (ret) {
3978                         /*
3979                          * File isn't open, so nothing to move over to the
3980                          * server.
3981                          */
3982                         return (0);
3983                 }
3984                 vp = NFSTOV(np);
3985                 *vpp = vp;
3986         } else {
3987                 np = VTONFS(vp);
3988         }
3989         dp->nfsdl_flags &= ~NFSCLDL_MODTIMESET;
3990
3991         /*
3992          * Ok, if it's a write delegation, flush data to the server, so
3993          * that close/open consistency is retained.
3994          */
3995         ret = 0;
3996         NFSLOCKNODE(np);
3997         if ((dp->nfsdl_flags & NFSCLDL_WRITE) && (np->n_flag & NMODIFIED)) {
3998                 np->n_flag |= NDELEGRECALL;
3999                 NFSUNLOCKNODE(np);
4000                 ret = ncl_flush(vp, MNT_WAIT, p, 1, called_from_renewthread);
4001                 NFSLOCKNODE(np);
4002                 np->n_flag &= ~NDELEGRECALL;
4003         }
4004         NFSINVALATTRCACHE(np);
4005         NFSUNLOCKNODE(np);
4006         if (ret == EIO && called_from_renewthread != 0) {
4007                 /*
4008                  * If the flush failed with EIO for the renew thread,
4009                  * return now, so that the dirty buffer will be flushed
4010                  * later.
4011                  */
4012                 return (ret);
4013         }
4014
4015         /*
4016          * Now, for each openowner with opens issued locally, move them
4017          * over to state against the server.
4018          */
4019         LIST_FOREACH(lowp, &dp->nfsdl_owner, nfsow_list) {
4020                 lop = LIST_FIRST(&lowp->nfsow_open);
4021                 if (lop != NULL) {
4022                         if (LIST_NEXT(lop, nfso_list) != NULL)
4023                                 panic("nfsdlg mult opens");
4024                         /*
4025                          * Look for the same openowner against the server.
4026                          */
4027                         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
4028                                 if (!NFSBCMP(lowp->nfsow_owner,
4029                                     owp->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
4030                                         newnfs_copycred(&dp->nfsdl_cred, cred);
4031                                         ret = nfscl_moveopen(vp, clp, nmp, lop,
4032                                             owp, dp, cred, p);
4033                                         if (ret == NFSERR_STALECLIENTID ||
4034                                             ret == NFSERR_STALEDONTRECOVER ||
4035                                             ret == NFSERR_BADSESSION)
4036                                                 return (ret);
4037                                         if (ret) {
4038                                                 nfscl_freeopen(lop, 1);
4039                                                 if (!error)
4040                                                         error = ret;
4041                                         }
4042                                         break;
4043                                 }
4044                         }
4045
4046                         /*
4047                          * If no openowner found, create one and get an open
4048                          * for it.
4049                          */
4050                         if (owp == NULL) {
4051                                 nowp = malloc(
4052                                     sizeof (struct nfsclowner), M_NFSCLOWNER,
4053                                     M_WAITOK);
4054                                 nfscl_newopen(clp, NULL, &owp, &nowp, &op,
4055                                     NULL, lowp->nfsow_owner, dp->nfsdl_fh,
4056                                     dp->nfsdl_fhlen, NULL, NULL);
4057                                 newnfs_copycred(&dp->nfsdl_cred, cred);
4058                                 ret = nfscl_moveopen(vp, clp, nmp, lop,
4059                                     owp, dp, cred, p);
4060                                 if (ret) {
4061                                         nfscl_freeopenowner(owp, 0);
4062                                         if (ret == NFSERR_STALECLIENTID ||
4063                                             ret == NFSERR_STALEDONTRECOVER ||
4064                                             ret == NFSERR_BADSESSION)
4065                                                 return (ret);
4066                                         if (ret) {
4067                                                 nfscl_freeopen(lop, 1);
4068                                                 if (!error)
4069                                                         error = ret;
4070                                         }
4071                                 }
4072                         }
4073                 }
4074         }
4075
4076         /*
4077          * Now, get byte range locks for any locks done locally.
4078          */
4079         LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4080                 LIST_FOREACH(lckp, &lp->nfsl_lock, nfslo_list) {
4081                         newnfs_copycred(&dp->nfsdl_cred, cred);
4082                         ret = nfscl_relock(vp, clp, nmp, lp, lckp, cred, p);
4083                         if (ret == NFSERR_STALESTATEID ||
4084                             ret == NFSERR_STALEDONTRECOVER ||
4085                             ret == NFSERR_STALECLIENTID ||
4086                             ret == NFSERR_BADSESSION)
4087                                 return (ret);
4088                         if (ret && !error)
4089                                 error = ret;
4090                 }
4091         }
4092         return (error);
4093 }
4094
4095 /*
4096  * Move a locally issued open over to an owner on the state list.
4097  * SIDE EFFECT: If it needs to sleep (do an rpc), it unlocks clstate and
4098  * returns with it unlocked.
4099  */
4100 static int
4101 nfscl_moveopen(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp,
4102     struct nfsclopen *lop, struct nfsclowner *owp, struct nfscldeleg *dp,
4103     struct ucred *cred, NFSPROC_T *p)
4104 {
4105         struct nfsclopen *op, *nop;
4106         struct nfscldeleg *ndp;
4107         struct nfsnode *np;
4108         int error = 0, newone;
4109
4110         /*
4111          * First, look for an appropriate open, If found, just increment the
4112          * opencnt in it.
4113          */
4114         LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
4115                 if ((op->nfso_mode & lop->nfso_mode) == lop->nfso_mode &&
4116                     op->nfso_fhlen == lop->nfso_fhlen &&
4117                     !NFSBCMP(op->nfso_fh, lop->nfso_fh, op->nfso_fhlen)) {
4118                         op->nfso_opencnt += lop->nfso_opencnt;
4119                         nfscl_freeopen(lop, 1);
4120                         return (0);
4121                 }
4122         }
4123
4124         /* No appropriate open, so we have to do one against the server. */
4125         np = VTONFS(vp);
4126         nop = malloc(sizeof (struct nfsclopen) +
4127             lop->nfso_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
4128         newone = 0;
4129         nfscl_newopen(clp, NULL, &owp, NULL, &op, &nop, owp->nfsow_owner,
4130             lop->nfso_fh, lop->nfso_fhlen, cred, &newone);
4131         ndp = dp;
4132         error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data, np->n_v4->n4_fhlen,
4133             lop->nfso_fh, lop->nfso_fhlen, lop->nfso_mode, op,
4134             NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, &ndp, 0, 0, cred, p);
4135         if (error) {
4136                 if (newone)
4137                         nfscl_freeopen(op, 0);
4138         } else {
4139                 op->nfso_mode |= lop->nfso_mode;
4140                 op->nfso_opencnt += lop->nfso_opencnt;
4141                 nfscl_freeopen(lop, 1);
4142         }
4143         if (nop != NULL)
4144                 free(nop, M_NFSCLOPEN);
4145         if (ndp != NULL) {
4146                 /*
4147                  * What should I do with the returned delegation, since the
4148                  * delegation is being recalled? For now, just printf and
4149                  * through it away.
4150                  */
4151                 printf("Moveopen returned deleg\n");
4152                 free(ndp, M_NFSCLDELEG);
4153         }
4154         return (error);
4155 }
4156
4157 /*
4158  * Recall all delegations on this client.
4159  */
4160 static void
4161 nfscl_totalrecall(struct nfsclclient *clp)
4162 {
4163         struct nfscldeleg *dp;
4164
4165         TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
4166                 if ((dp->nfsdl_flags & NFSCLDL_DELEGRET) == 0)
4167                         dp->nfsdl_flags |= NFSCLDL_RECALL;
4168         }
4169 }
4170
4171 /*
4172  * Relock byte ranges. Called for delegation recall and state expiry.
4173  */
4174 static int
4175 nfscl_relock(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp,
4176     struct nfscllockowner *lp, struct nfscllock *lop, struct ucred *cred,
4177     NFSPROC_T *p)
4178 {
4179         struct nfscllockowner *nlp;
4180         struct nfsfh *nfhp;
4181         u_int64_t off, len;
4182         int error, newone, donelocally;
4183
4184         off = lop->nfslo_first;
4185         len = lop->nfslo_end - lop->nfslo_first;
4186         error = nfscl_getbytelock(vp, off, len, lop->nfslo_type, cred, p,
4187             clp, 1, NULL, lp->nfsl_lockflags, lp->nfsl_owner,
4188             lp->nfsl_openowner, &nlp, &newone, &donelocally);
4189         if (error || donelocally)
4190                 return (error);
4191         nfhp = VTONFS(vp)->n_fhp;
4192         error = nfscl_trylock(nmp, vp, nfhp->nfh_fh,
4193             nfhp->nfh_len, nlp, newone, 0, off,
4194             len, lop->nfslo_type, cred, p);
4195         if (error)
4196                 nfscl_freelockowner(nlp, 0);
4197         return (error);
4198 }
4199
4200 /*
4201  * Called to re-open a file. Basically get a vnode for the file handle
4202  * and then call nfsrpc_openrpc() to do the rest.
4203  */
4204 static int
4205 nfsrpc_reopen(struct nfsmount *nmp, u_int8_t *fhp, int fhlen,
4206     u_int32_t mode, struct nfsclopen *op, struct nfscldeleg **dpp,
4207     struct ucred *cred, NFSPROC_T *p)
4208 {
4209         struct nfsnode *np;
4210         vnode_t vp;
4211         int error;
4212
4213         error = nfscl_ngetreopen(nmp->nm_mountp, fhp, fhlen, p, &np);
4214         if (error)
4215                 return (error);
4216         vp = NFSTOV(np);
4217         if (np->n_v4 != NULL) {
4218                 error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data,
4219                     np->n_v4->n4_fhlen, fhp, fhlen, mode, op,
4220                     NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, dpp, 0, 0,
4221                     cred, p);
4222         } else {
4223                 error = EINVAL;
4224         }
4225         vrele(vp);
4226         return (error);
4227 }
4228
4229 /*
4230  * Try an open against the server. Just call nfsrpc_openrpc(), retrying while
4231  * NFSERR_DELAY. Also, try system credentials, if the passed in credentials
4232  * fail.
4233  */
4234 static int
4235 nfscl_tryopen(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen,
4236     u_int8_t *newfhp, int newfhlen, u_int32_t mode, struct nfsclopen *op,
4237     u_int8_t *name, int namelen, struct nfscldeleg **ndpp,
4238     int reclaim, u_int32_t delegtype, struct ucred *cred, NFSPROC_T *p)
4239 {
4240         int error;
4241
4242         do {
4243                 error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp, newfhlen,
4244                     mode, op, name, namelen, ndpp, reclaim, delegtype, cred, p,
4245                     0, 0);
4246                 if (error == NFSERR_DELAY)
4247                         (void) nfs_catnap(PZERO, error, "nfstryop");
4248         } while (error == NFSERR_DELAY);
4249         if (error == EAUTH || error == EACCES) {
4250                 /* Try again using system credentials */
4251                 newnfs_setroot(cred);
4252                 do {
4253                     error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp,
4254                         newfhlen, mode, op, name, namelen, ndpp, reclaim,
4255                         delegtype, cred, p, 1, 0);
4256                     if (error == NFSERR_DELAY)
4257                         (void) nfs_catnap(PZERO, error, "nfstryop");
4258                 } while (error == NFSERR_DELAY);
4259         }
4260         return (error);
4261 }
4262
4263 /*
4264  * Try a byte range lock. Just loop on nfsrpc_lock() while it returns
4265  * NFSERR_DELAY. Also, retry with system credentials, if the provided
4266  * cred don't work.
4267  */
4268 static int
4269 nfscl_trylock(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp,
4270     int fhlen, struct nfscllockowner *nlp, int newone, int reclaim,
4271     u_int64_t off, u_int64_t len, short type, struct ucred *cred, NFSPROC_T *p)
4272 {
4273         struct nfsrv_descript nfsd, *nd = &nfsd;
4274         int error;
4275
4276         do {
4277                 error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp, newone,
4278                     reclaim, off, len, type, cred, p, 0);
4279                 if (!error && nd->nd_repstat == NFSERR_DELAY)
4280                         (void) nfs_catnap(PZERO, (int)nd->nd_repstat,
4281                             "nfstrylck");
4282         } while (!error && nd->nd_repstat == NFSERR_DELAY);
4283         if (!error)
4284                 error = nd->nd_repstat;
4285         if (error == EAUTH || error == EACCES) {
4286                 /* Try again using root credentials */
4287                 newnfs_setroot(cred);
4288                 do {
4289                         error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp,
4290                             newone, reclaim, off, len, type, cred, p, 1);
4291                         if (!error && nd->nd_repstat == NFSERR_DELAY)
4292                                 (void) nfs_catnap(PZERO, (int)nd->nd_repstat,
4293                                     "nfstrylck");
4294                 } while (!error && nd->nd_repstat == NFSERR_DELAY);
4295                 if (!error)
4296                         error = nd->nd_repstat;
4297         }
4298         return (error);
4299 }
4300
4301 /*
4302  * Try a delegreturn against the server. Just call nfsrpc_delegreturn(),
4303  * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
4304  * credentials fail.
4305  */
4306 static int
4307 nfscl_trydelegreturn(struct nfscldeleg *dp, struct ucred *cred,
4308     struct nfsmount *nmp, NFSPROC_T *p)
4309 {
4310         int error;
4311
4312         do {
4313                 error = nfsrpc_delegreturn(dp, cred, nmp, p, 0);
4314                 if (error == NFSERR_DELAY)
4315                         (void) nfs_catnap(PZERO, error, "nfstrydp");
4316         } while (error == NFSERR_DELAY);
4317         if (error == EAUTH || error == EACCES) {
4318                 /* Try again using system credentials */
4319                 newnfs_setroot(cred);
4320                 do {
4321                         error = nfsrpc_delegreturn(dp, cred, nmp, p, 1);
4322                         if (error == NFSERR_DELAY)
4323                                 (void) nfs_catnap(PZERO, error, "nfstrydp");
4324                 } while (error == NFSERR_DELAY);
4325         }
4326         return (error);
4327 }
4328
4329 /*
4330  * Try a close against the server. Just call nfsrpc_closerpc(),
4331  * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
4332  * credentials fail.
4333  */
4334 int
4335 nfscl_tryclose(struct nfsclopen *op, struct ucred *cred,
4336     struct nfsmount *nmp, NFSPROC_T *p)
4337 {
4338         struct nfsrv_descript nfsd, *nd = &nfsd;
4339         int error;
4340
4341         do {
4342                 error = nfsrpc_closerpc(nd, nmp, op, cred, p, 0);
4343                 if (error == NFSERR_DELAY)
4344                         (void) nfs_catnap(PZERO, error, "nfstrycl");
4345         } while (error == NFSERR_DELAY);
4346         if (error == EAUTH || error == EACCES) {
4347                 /* Try again using system credentials */
4348                 newnfs_setroot(cred);
4349                 do {
4350                         error = nfsrpc_closerpc(nd, nmp, op, cred, p, 1);
4351                         if (error == NFSERR_DELAY)
4352                                 (void) nfs_catnap(PZERO, error, "nfstrycl");
4353                 } while (error == NFSERR_DELAY);
4354         }
4355         return (error);
4356 }
4357
4358 /*
4359  * Decide if a delegation on a file permits close without flushing writes
4360  * to the server. This might be a big performance win in some environments.
4361  * (Not useful until the client does caching on local stable storage.)
4362  */
4363 int
4364 nfscl_mustflush(vnode_t vp)
4365 {
4366         struct nfsclclient *clp;
4367         struct nfscldeleg *dp;
4368         struct nfsnode *np;
4369         struct nfsmount *nmp;
4370
4371         np = VTONFS(vp);
4372         nmp = VFSTONFS(vnode_mount(vp));
4373         if (!NFSHASNFSV4(nmp))
4374                 return (1);
4375         NFSLOCKCLSTATE();
4376         clp = nfscl_findcl(nmp);
4377         if (clp == NULL) {
4378                 NFSUNLOCKCLSTATE();
4379                 return (1);
4380         }
4381         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4382         if (dp != NULL && (dp->nfsdl_flags &
4383             (NFSCLDL_WRITE | NFSCLDL_RECALL | NFSCLDL_DELEGRET)) ==
4384              NFSCLDL_WRITE &&
4385             (dp->nfsdl_sizelimit >= np->n_size ||
4386              !NFSHASSTRICT3530(nmp))) {
4387                 NFSUNLOCKCLSTATE();
4388                 return (0);
4389         }
4390         NFSUNLOCKCLSTATE();
4391         return (1);
4392 }
4393
4394 /*
4395  * See if a (write) delegation exists for this file.
4396  */
4397 int
4398 nfscl_nodeleg(vnode_t vp, int writedeleg)
4399 {
4400         struct nfsclclient *clp;
4401         struct nfscldeleg *dp;
4402         struct nfsnode *np;
4403         struct nfsmount *nmp;
4404
4405         np = VTONFS(vp);
4406         nmp = VFSTONFS(vnode_mount(vp));
4407         if (!NFSHASNFSV4(nmp))
4408                 return (1);
4409         NFSLOCKCLSTATE();
4410         clp = nfscl_findcl(nmp);
4411         if (clp == NULL) {
4412                 NFSUNLOCKCLSTATE();
4413                 return (1);
4414         }
4415         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4416         if (dp != NULL &&
4417             (dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_DELEGRET)) == 0 &&
4418             (writedeleg == 0 || (dp->nfsdl_flags & NFSCLDL_WRITE) ==
4419              NFSCLDL_WRITE)) {
4420                 NFSUNLOCKCLSTATE();
4421                 return (0);
4422         }
4423         NFSUNLOCKCLSTATE();
4424         return (1);
4425 }
4426
4427 /*
4428  * Look for an associated delegation that should be DelegReturned.
4429  */
4430 int
4431 nfscl_removedeleg(vnode_t vp, NFSPROC_T *p, nfsv4stateid_t *stp)
4432 {
4433         struct nfsclclient *clp;
4434         struct nfscldeleg *dp;
4435         struct nfsclowner *owp;
4436         struct nfscllockowner *lp;
4437         struct nfsmount *nmp;
4438         struct ucred *cred;
4439         struct nfsnode *np;
4440         int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
4441
4442         nmp = VFSTONFS(vnode_mount(vp));
4443         np = VTONFS(vp);
4444         NFSLOCKCLSTATE();
4445         /*
4446          * Loop around waiting for:
4447          * - outstanding I/O operations on delegations to complete
4448          * - for a delegation on vp that has state, lock the client and
4449          *   do a recall
4450          * - return delegation with no state
4451          */
4452         while (1) {
4453                 clp = nfscl_findcl(nmp);
4454                 if (clp == NULL) {
4455                         NFSUNLOCKCLSTATE();
4456                         return (retcnt);
4457                 }
4458                 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4459                     np->n_fhp->nfh_len);
4460                 if (dp != NULL) {
4461                     /*
4462                      * Wait for outstanding I/O ops to be done.
4463                      */
4464                     if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4465                         if (igotlock) {
4466                             nfsv4_unlock(&clp->nfsc_lock, 0);
4467                             igotlock = 0;
4468                         }
4469                         dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4470                         (void) nfsmsleep(&dp->nfsdl_rwlock,
4471                             NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4472                         continue;
4473                     }
4474                     needsrecall = 0;
4475                     LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4476                         if (!LIST_EMPTY(&owp->nfsow_open)) {
4477                             needsrecall = 1;
4478                             break;
4479                         }
4480                     }
4481                     if (!needsrecall) {
4482                         LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4483                             if (!LIST_EMPTY(&lp->nfsl_lock)) {
4484                                 needsrecall = 1;
4485                                 break;
4486                             }
4487                         }
4488                     }
4489                     if (needsrecall && !triedrecall) {
4490                         dp->nfsdl_flags |= NFSCLDL_DELEGRET;
4491                         islept = 0;
4492                         while (!igotlock) {
4493                             igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
4494                                 &islept, NFSCLSTATEMUTEXPTR, NULL);
4495                             if (islept)
4496                                 break;
4497                         }
4498                         if (islept)
4499                             continue;
4500                         NFSUNLOCKCLSTATE();
4501                         cred = newnfs_getcred();
4502                         newnfs_copycred(&dp->nfsdl_cred, cred);
4503                         nfscl_recalldeleg(clp, nmp, dp, vp, cred, p, 0, NULL);
4504                         NFSFREECRED(cred);
4505                         triedrecall = 1;
4506                         NFSLOCKCLSTATE();
4507                         nfsv4_unlock(&clp->nfsc_lock, 0);
4508                         igotlock = 0;
4509                         continue;
4510                     }
4511                     *stp = dp->nfsdl_stateid;
4512                     retcnt = 1;
4513                     nfscl_cleandeleg(dp);
4514                     nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
4515                 }
4516                 if (igotlock)
4517                     nfsv4_unlock(&clp->nfsc_lock, 0);
4518                 NFSUNLOCKCLSTATE();
4519                 return (retcnt);
4520         }
4521 }
4522
4523 /*
4524  * Look for associated delegation(s) that should be DelegReturned.
4525  */
4526 int
4527 nfscl_renamedeleg(vnode_t fvp, nfsv4stateid_t *fstp, int *gotfdp, vnode_t tvp,
4528     nfsv4stateid_t *tstp, int *gottdp, NFSPROC_T *p)
4529 {
4530         struct nfsclclient *clp;
4531         struct nfscldeleg *dp;
4532         struct nfsclowner *owp;
4533         struct nfscllockowner *lp;
4534         struct nfsmount *nmp;
4535         struct ucred *cred;
4536         struct nfsnode *np;
4537         int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
4538
4539         nmp = VFSTONFS(vnode_mount(fvp));
4540         *gotfdp = 0;
4541         *gottdp = 0;
4542         NFSLOCKCLSTATE();
4543         /*
4544          * Loop around waiting for:
4545          * - outstanding I/O operations on delegations to complete
4546          * - for a delegation on fvp that has state, lock the client and
4547          *   do a recall
4548          * - return delegation(s) with no state.
4549          */
4550         while (1) {
4551                 clp = nfscl_findcl(nmp);
4552                 if (clp == NULL) {
4553                         NFSUNLOCKCLSTATE();
4554                         return (retcnt);
4555                 }
4556                 np = VTONFS(fvp);
4557                 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4558                     np->n_fhp->nfh_len);
4559                 if (dp != NULL && *gotfdp == 0) {
4560                     /*
4561                      * Wait for outstanding I/O ops to be done.
4562                      */
4563                     if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4564                         if (igotlock) {
4565                             nfsv4_unlock(&clp->nfsc_lock, 0);
4566                             igotlock = 0;
4567                         }
4568                         dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4569                         (void) nfsmsleep(&dp->nfsdl_rwlock,
4570                             NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4571                         continue;
4572                     }
4573                     needsrecall = 0;
4574                     LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4575                         if (!LIST_EMPTY(&owp->nfsow_open)) {
4576                             needsrecall = 1;
4577                             break;
4578                         }
4579                     }
4580                     if (!needsrecall) {
4581                         LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4582                             if (!LIST_EMPTY(&lp->nfsl_lock)) {
4583                                 needsrecall = 1;
4584                                 break;
4585                             }
4586                         }
4587                     }
4588                     if (needsrecall && !triedrecall) {
4589                         dp->nfsdl_flags |= NFSCLDL_DELEGRET;
4590                         islept = 0;
4591                         while (!igotlock) {
4592                             igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
4593                                 &islept, NFSCLSTATEMUTEXPTR, NULL);
4594                             if (islept)
4595                                 break;
4596                         }
4597                         if (islept)
4598                             continue;
4599                         NFSUNLOCKCLSTATE();
4600                         cred = newnfs_getcred();
4601                         newnfs_copycred(&dp->nfsdl_cred, cred);
4602                         nfscl_recalldeleg(clp, nmp, dp, fvp, cred, p, 0, NULL);
4603                         NFSFREECRED(cred);
4604                         triedrecall = 1;
4605                         NFSLOCKCLSTATE();
4606                         nfsv4_unlock(&clp->nfsc_lock, 0);
4607                         igotlock = 0;
4608                         continue;
4609                     }
4610                     *fstp = dp->nfsdl_stateid;
4611                     retcnt++;
4612                     *gotfdp = 1;
4613                     nfscl_cleandeleg(dp);
4614                     nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
4615                 }
4616                 if (igotlock) {
4617                     nfsv4_unlock(&clp->nfsc_lock, 0);
4618                     igotlock = 0;
4619                 }
4620                 if (tvp != NULL) {
4621                     np = VTONFS(tvp);
4622                     dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4623                         np->n_fhp->nfh_len);
4624                     if (dp != NULL && *gottdp == 0) {
4625                         /*
4626                          * Wait for outstanding I/O ops to be done.
4627                          */
4628                         if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4629                             dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4630                             (void) nfsmsleep(&dp->nfsdl_rwlock,
4631                                 NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4632                             continue;
4633                         }
4634                         LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4635                             if (!LIST_EMPTY(&owp->nfsow_open)) {
4636                                 NFSUNLOCKCLSTATE();
4637                                 return (retcnt);
4638                             }
4639                         }
4640                         LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4641                             if (!LIST_EMPTY(&lp->nfsl_lock)) {
4642                                 NFSUNLOCKCLSTATE();
4643                                 return (retcnt);
4644                             }
4645                         }
4646                         *tstp = dp->nfsdl_stateid;
4647                         retcnt++;
4648                         *gottdp = 1;
4649                         nfscl_cleandeleg(dp);
4650                         nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
4651                     }
4652                 }
4653                 NFSUNLOCKCLSTATE();
4654                 return (retcnt);
4655         }
4656 }
4657
4658 /*
4659  * Get a reference on the clientid associated with the mount point.
4660  * Return 1 if success, 0 otherwise.
4661  */
4662 int
4663 nfscl_getref(struct nfsmount *nmp)
4664 {
4665         struct nfsclclient *clp;
4666
4667         NFSLOCKCLSTATE();
4668         clp = nfscl_findcl(nmp);
4669         if (clp == NULL) {
4670                 NFSUNLOCKCLSTATE();
4671                 return (0);
4672         }
4673         nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, NULL);
4674         NFSUNLOCKCLSTATE();
4675         return (1);
4676 }
4677
4678 /*
4679  * Release a reference on a clientid acquired with the above call.
4680  */
4681 void
4682 nfscl_relref(struct nfsmount *nmp)
4683 {
4684         struct nfsclclient *clp;
4685
4686         NFSLOCKCLSTATE();
4687         clp = nfscl_findcl(nmp);
4688         if (clp == NULL) {
4689                 NFSUNLOCKCLSTATE();
4690                 return;
4691         }
4692         nfsv4_relref(&clp->nfsc_lock);
4693         NFSUNLOCKCLSTATE();
4694 }
4695
4696 /*
4697  * Save the size attribute in the delegation, since the nfsnode
4698  * is going away.
4699  */
4700 void
4701 nfscl_reclaimnode(vnode_t vp)
4702 {
4703         struct nfsclclient *clp;
4704         struct nfscldeleg *dp;
4705         struct nfsnode *np = VTONFS(vp);
4706         struct nfsmount *nmp;
4707
4708         nmp = VFSTONFS(vnode_mount(vp));
4709         if (!NFSHASNFSV4(nmp))
4710                 return;
4711         NFSLOCKCLSTATE();
4712         clp = nfscl_findcl(nmp);
4713         if (clp == NULL) {
4714                 NFSUNLOCKCLSTATE();
4715                 return;
4716         }
4717         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4718         if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE))
4719                 dp->nfsdl_size = np->n_size;
4720         NFSUNLOCKCLSTATE();
4721 }
4722
4723 /*
4724  * Get the saved size attribute in the delegation, since it is a
4725  * newly allocated nfsnode.
4726  */
4727 void
4728 nfscl_newnode(vnode_t vp)
4729 {
4730         struct nfsclclient *clp;
4731         struct nfscldeleg *dp;
4732         struct nfsnode *np = VTONFS(vp);
4733         struct nfsmount *nmp;
4734
4735         nmp = VFSTONFS(vnode_mount(vp));
4736         if (!NFSHASNFSV4(nmp))
4737                 return;
4738         NFSLOCKCLSTATE();
4739         clp = nfscl_findcl(nmp);
4740         if (clp == NULL) {
4741                 NFSUNLOCKCLSTATE();
4742                 return;
4743         }
4744         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4745         if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE))
4746                 np->n_size = dp->nfsdl_size;
4747         NFSUNLOCKCLSTATE();
4748 }
4749
4750 /*
4751  * If there is a valid write delegation for this file, set the modtime
4752  * to the local clock time.
4753  */
4754 void
4755 nfscl_delegmodtime(vnode_t vp)
4756 {
4757         struct nfsclclient *clp;
4758         struct nfscldeleg *dp;
4759         struct nfsnode *np = VTONFS(vp);
4760         struct nfsmount *nmp;
4761
4762         nmp = VFSTONFS(vnode_mount(vp));
4763         if (!NFSHASNFSV4(nmp))
4764                 return;
4765         NFSLOCKCLSTATE();
4766         clp = nfscl_findcl(nmp);
4767         if (clp == NULL) {
4768                 NFSUNLOCKCLSTATE();
4769                 return;
4770         }
4771         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4772         if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE)) {
4773                 nanotime(&dp->nfsdl_modtime);
4774                 dp->nfsdl_flags |= NFSCLDL_MODTIMESET;
4775         }
4776         NFSUNLOCKCLSTATE();
4777 }
4778
4779 /*
4780  * If there is a valid write delegation for this file with a modtime set,
4781  * put that modtime in mtime.
4782  */
4783 void
4784 nfscl_deleggetmodtime(vnode_t vp, struct timespec *mtime)
4785 {
4786         struct nfsclclient *clp;
4787         struct nfscldeleg *dp;
4788         struct nfsnode *np = VTONFS(vp);
4789         struct nfsmount *nmp;
4790
4791         nmp = VFSTONFS(vnode_mount(vp));
4792         if (!NFSHASNFSV4(nmp))
4793                 return;
4794         NFSLOCKCLSTATE();
4795         clp = nfscl_findcl(nmp);
4796         if (clp == NULL) {
4797                 NFSUNLOCKCLSTATE();
4798                 return;
4799         }
4800         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4801         if (dp != NULL &&
4802             (dp->nfsdl_flags & (NFSCLDL_WRITE | NFSCLDL_MODTIMESET)) ==
4803             (NFSCLDL_WRITE | NFSCLDL_MODTIMESET))
4804                 *mtime = dp->nfsdl_modtime;
4805         NFSUNLOCKCLSTATE();
4806 }
4807
4808 static int
4809 nfscl_errmap(struct nfsrv_descript *nd, u_int32_t minorvers)
4810 {
4811         short *defaulterrp, *errp;
4812
4813         if (!nd->nd_repstat)
4814                 return (0);
4815         if (nd->nd_procnum == NFSPROC_NOOP)
4816                 return (txdr_unsigned(nd->nd_repstat & 0xffff));
4817         if (nd->nd_repstat == EBADRPC)
4818                 return (txdr_unsigned(NFSERR_BADXDR));
4819         if (nd->nd_repstat == NFSERR_MINORVERMISMATCH ||
4820             nd->nd_repstat == NFSERR_OPILLEGAL)
4821                 return (txdr_unsigned(nd->nd_repstat));
4822         if (nd->nd_repstat >= NFSERR_BADIOMODE && nd->nd_repstat < 20000 &&
4823             minorvers > NFSV4_MINORVERSION) {
4824                 /* NFSv4.n error. */
4825                 return (txdr_unsigned(nd->nd_repstat));
4826         }
4827         if (nd->nd_procnum < NFSV4OP_CBNOPS)
4828                 errp = defaulterrp = nfscl_cberrmap[nd->nd_procnum];
4829         else
4830                 return (txdr_unsigned(nd->nd_repstat));
4831         while (*++errp)
4832                 if (*errp == (short)nd->nd_repstat)
4833                         return (txdr_unsigned(nd->nd_repstat));
4834         return (txdr_unsigned(*defaulterrp));
4835 }
4836
4837 /*
4838  * Called to find/add a layout to a client.
4839  * This function returns the layout with a refcnt (shared lock) upon
4840  * success (returns 0) or with no lock/refcnt on the layout when an
4841  * error is returned.
4842  * If a layout is passed in via lypp, it is locked (exclusively locked).
4843  */
4844 int
4845 nfscl_layout(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen,
4846     nfsv4stateid_t *stateidp, int layouttype, int retonclose,
4847     struct nfsclflayouthead *fhlp, struct nfscllayout **lypp,
4848     struct ucred *cred, NFSPROC_T *p)
4849 {
4850         struct nfsclclient *clp;
4851         struct nfscllayout *lyp, *tlyp;
4852         struct nfsclflayout *flp;
4853         struct nfsnode *np = VTONFS(vp);
4854         mount_t mp;
4855         int layout_passed_in;
4856
4857         mp = nmp->nm_mountp;
4858         layout_passed_in = 1;
4859         tlyp = NULL;
4860         lyp = *lypp;
4861         if (lyp == NULL) {
4862                 layout_passed_in = 0;
4863                 tlyp = malloc(sizeof(*tlyp) + fhlen - 1, M_NFSLAYOUT,
4864                     M_WAITOK | M_ZERO);
4865         }
4866
4867         NFSLOCKCLSTATE();
4868         clp = nmp->nm_clp;
4869         if (clp == NULL) {
4870                 if (layout_passed_in != 0)
4871                         nfsv4_unlock(&lyp->nfsly_lock, 0);
4872                 NFSUNLOCKCLSTATE();
4873                 if (tlyp != NULL)
4874                         free(tlyp, M_NFSLAYOUT);
4875                 return (EPERM);
4876         }
4877         if (lyp == NULL) {
4878                 /*
4879                  * Although no lyp was passed in, another thread might have
4880                  * allocated one. If one is found, just increment it's ref
4881                  * count and return it.
4882                  */
4883                 lyp = nfscl_findlayout(clp, fhp, fhlen);
4884                 if (lyp == NULL) {
4885                         lyp = tlyp;
4886                         tlyp = NULL;
4887                         lyp->nfsly_stateid.seqid = stateidp->seqid;
4888                         lyp->nfsly_stateid.other[0] = stateidp->other[0];
4889                         lyp->nfsly_stateid.other[1] = stateidp->other[1];
4890                         lyp->nfsly_stateid.other[2] = stateidp->other[2];
4891                         lyp->nfsly_lastbyte = 0;
4892                         LIST_INIT(&lyp->nfsly_flayread);
4893                         LIST_INIT(&lyp->nfsly_flayrw);
4894                         LIST_INIT(&lyp->nfsly_recall);
4895                         lyp->nfsly_filesid[0] = np->n_vattr.na_filesid[0];
4896                         lyp->nfsly_filesid[1] = np->n_vattr.na_filesid[1];
4897                         lyp->nfsly_clp = clp;
4898                         if (layouttype == NFSLAYOUT_FLEXFILE)
4899                                 lyp->nfsly_flags = NFSLY_FLEXFILE;
4900                         else
4901                                 lyp->nfsly_flags = NFSLY_FILES;
4902                         if (retonclose != 0)
4903                                 lyp->nfsly_flags |= NFSLY_RETONCLOSE;
4904                         lyp->nfsly_fhlen = fhlen;
4905                         NFSBCOPY(fhp, lyp->nfsly_fh, fhlen);
4906                         TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
4907                         LIST_INSERT_HEAD(NFSCLLAYOUTHASH(clp, fhp, fhlen), lyp,
4908                             nfsly_hash);
4909                         lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
4910                         nfscl_layoutcnt++;
4911                 } else {
4912                         if (retonclose != 0)
4913                                 lyp->nfsly_flags |= NFSLY_RETONCLOSE;
4914                         if (stateidp->seqid > lyp->nfsly_stateid.seqid)
4915                                 lyp->nfsly_stateid.seqid = stateidp->seqid;
4916                         TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list);
4917                         TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
4918                         lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
4919                 }
4920                 nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
4921                 if (NFSCL_FORCEDISM(mp)) {
4922                         NFSUNLOCKCLSTATE();
4923                         if (tlyp != NULL)
4924                                 free(tlyp, M_NFSLAYOUT);
4925                         return (EPERM);
4926                 }
4927                 *lypp = lyp;
4928         } else if (stateidp->seqid > lyp->nfsly_stateid.seqid)
4929                 lyp->nfsly_stateid.seqid = stateidp->seqid;
4930
4931         /* Merge the new list of File Layouts into the list. */
4932         flp = LIST_FIRST(fhlp);
4933         if (flp != NULL) {
4934                 if (flp->nfsfl_iomode == NFSLAYOUTIOMODE_READ)
4935                         nfscl_mergeflayouts(&lyp->nfsly_flayread, fhlp);
4936                 else
4937                         nfscl_mergeflayouts(&lyp->nfsly_flayrw, fhlp);
4938         }
4939         if (layout_passed_in != 0)
4940                 nfsv4_unlock(&lyp->nfsly_lock, 1);
4941         NFSUNLOCKCLSTATE();
4942         if (tlyp != NULL)
4943                 free(tlyp, M_NFSLAYOUT);
4944         return (0);
4945 }
4946
4947 /*
4948  * Search for a layout by MDS file handle.
4949  * If one is found, it is returned with a refcnt (shared lock) iff
4950  * retflpp returned non-NULL and locked (exclusive locked) iff retflpp is
4951  * returned NULL.
4952  */
4953 struct nfscllayout *
4954 nfscl_getlayout(struct nfsclclient *clp, uint8_t *fhp, int fhlen,
4955     uint64_t off, struct nfsclflayout **retflpp, int *recalledp)
4956 {
4957         struct nfscllayout *lyp;
4958         mount_t mp;
4959         int error, igotlock;
4960
4961         mp = clp->nfsc_nmp->nm_mountp;
4962         *recalledp = 0;
4963         *retflpp = NULL;
4964         NFSLOCKCLSTATE();
4965         lyp = nfscl_findlayout(clp, fhp, fhlen);
4966         if (lyp != NULL) {
4967                 if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
4968                         TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list);
4969                         TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
4970                         lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
4971                         error = nfscl_findlayoutforio(lyp, off,
4972                             NFSV4OPEN_ACCESSREAD, retflpp);
4973                         if (error == 0)
4974                                 nfsv4_getref(&lyp->nfsly_lock, NULL,
4975                                     NFSCLSTATEMUTEXPTR, mp);
4976                         else {
4977                                 do {
4978                                         igotlock = nfsv4_lock(&lyp->nfsly_lock,
4979                                             1, NULL, NFSCLSTATEMUTEXPTR, mp);
4980                                 } while (igotlock == 0 && !NFSCL_FORCEDISM(mp));
4981                                 *retflpp = NULL;
4982                         }
4983                         if (NFSCL_FORCEDISM(mp)) {
4984                                 lyp = NULL;
4985                                 *recalledp = 1;
4986                         }
4987                 } else {
4988                         lyp = NULL;
4989                         *recalledp = 1;
4990                 }
4991         }
4992         NFSUNLOCKCLSTATE();
4993         return (lyp);
4994 }
4995
4996 /*
4997  * Search for a layout by MDS file handle. If one is found, mark in to be
4998  * recalled, if it already marked "return on close".
4999  */
5000 static void
5001 nfscl_retoncloselayout(vnode_t vp, struct nfsclclient *clp, uint8_t *fhp,
5002     int fhlen, struct nfsclrecalllayout **recallpp)
5003 {
5004         struct nfscllayout *lyp;
5005         uint32_t iomode;
5006
5007         if (vp->v_type != VREG || !NFSHASPNFS(VFSTONFS(vnode_mount(vp))) ||
5008             nfscl_enablecallb == 0 || nfs_numnfscbd == 0 ||
5009             (VTONFS(vp)->n_flag & NNOLAYOUT) != 0)
5010                 return;
5011         lyp = nfscl_findlayout(clp, fhp, fhlen);
5012         if (lyp != NULL && (lyp->nfsly_flags & (NFSLY_RETONCLOSE |
5013             NFSLY_RECALL)) == NFSLY_RETONCLOSE) {
5014                 iomode = 0;
5015                 if (!LIST_EMPTY(&lyp->nfsly_flayread))
5016                         iomode |= NFSLAYOUTIOMODE_READ;
5017                 if (!LIST_EMPTY(&lyp->nfsly_flayrw))
5018                         iomode |= NFSLAYOUTIOMODE_RW;
5019                 (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE, lyp, iomode,
5020                     0, UINT64_MAX, lyp->nfsly_stateid.seqid, 0, 0, NULL,
5021                     *recallpp);
5022                 NFSCL_DEBUG(4, "retoncls recall iomode=%d\n", iomode);
5023                 *recallpp = NULL;
5024         }
5025 }
5026
5027 /*
5028  * Mark the layout to be recalled and with an error.
5029  * Also, disable the dsp from further use.
5030  */
5031 void
5032 nfscl_dserr(uint32_t op, uint32_t stat, struct nfscldevinfo *dp,
5033     struct nfscllayout *lyp, struct nfsclds *dsp)
5034 {
5035         struct nfsclrecalllayout *recallp;
5036         uint32_t iomode;
5037
5038         printf("DS being disabled, error=%d\n", stat);
5039         /* Set up the return of the layout. */
5040         recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_WAITOK);
5041         iomode = 0;
5042         NFSLOCKCLSTATE();
5043         if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
5044                 if (!LIST_EMPTY(&lyp->nfsly_flayread))
5045                         iomode |= NFSLAYOUTIOMODE_READ;
5046                 if (!LIST_EMPTY(&lyp->nfsly_flayrw))
5047                         iomode |= NFSLAYOUTIOMODE_RW;
5048                 (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE, lyp, iomode,
5049                     0, UINT64_MAX, lyp->nfsly_stateid.seqid, stat, op,
5050                     dp->nfsdi_deviceid, recallp);
5051                 NFSUNLOCKCLSTATE();
5052                 NFSCL_DEBUG(4, "nfscl_dserr recall iomode=%d\n", iomode);
5053         } else {
5054                 NFSUNLOCKCLSTATE();
5055                 free(recallp, M_NFSLAYRECALL);
5056         }
5057
5058         /* And shut the TCP connection down. */
5059         nfscl_cancelreqs(dsp);
5060 }
5061
5062 /*
5063  * Cancel all RPCs for this "dsp" by closing the connection.
5064  * Also, mark the session as defunct.
5065  * If NFSCLDS_SAMECONN is set, the connection is shared with other DSs and
5066  * cannot be shut down.
5067  */
5068 void
5069 nfscl_cancelreqs(struct nfsclds *dsp)
5070 {
5071         struct __rpc_client *cl;
5072         static int non_event;
5073
5074         NFSLOCKDS(dsp);
5075         if ((dsp->nfsclds_flags & (NFSCLDS_CLOSED | NFSCLDS_SAMECONN)) == 0 &&
5076             dsp->nfsclds_sockp != NULL &&
5077             dsp->nfsclds_sockp->nr_client != NULL) {
5078                 dsp->nfsclds_flags |= NFSCLDS_CLOSED;
5079                 cl = dsp->nfsclds_sockp->nr_client;
5080                 dsp->nfsclds_sess.nfsess_defunct = 1;
5081                 NFSUNLOCKDS(dsp);
5082                 CLNT_CLOSE(cl);
5083                 /*
5084                  * This 1sec sleep is done to reduce the number of reconnect
5085                  * attempts made on the DS while it has failed.
5086                  */
5087                 tsleep(&non_event, PVFS, "ndscls", hz);
5088                 return;
5089         }
5090         NFSUNLOCKDS(dsp);
5091 }
5092
5093 /*
5094  * Dereference a layout.
5095  */
5096 void
5097 nfscl_rellayout(struct nfscllayout *lyp, int exclocked)
5098 {
5099
5100         NFSLOCKCLSTATE();
5101         if (exclocked != 0)
5102                 nfsv4_unlock(&lyp->nfsly_lock, 0);
5103         else
5104                 nfsv4_relref(&lyp->nfsly_lock);
5105         NFSUNLOCKCLSTATE();
5106 }
5107
5108 /*
5109  * Search for a devinfo by deviceid. If one is found, return it after
5110  * acquiring a reference count on it.
5111  */
5112 struct nfscldevinfo *
5113 nfscl_getdevinfo(struct nfsclclient *clp, uint8_t *deviceid,
5114     struct nfscldevinfo *dip)
5115 {
5116
5117         NFSLOCKCLSTATE();
5118         if (dip == NULL)
5119                 dip = nfscl_finddevinfo(clp, deviceid);
5120         if (dip != NULL)
5121                 dip->nfsdi_refcnt++;
5122         NFSUNLOCKCLSTATE();
5123         return (dip);
5124 }
5125
5126 /*
5127  * Dereference a devinfo structure.
5128  */
5129 static void
5130 nfscl_reldevinfo_locked(struct nfscldevinfo *dip)
5131 {
5132
5133         dip->nfsdi_refcnt--;
5134         if (dip->nfsdi_refcnt == 0)
5135                 wakeup(&dip->nfsdi_refcnt);
5136 }
5137
5138 /*
5139  * Dereference a devinfo structure.
5140  */
5141 void
5142 nfscl_reldevinfo(struct nfscldevinfo *dip)
5143 {
5144
5145         NFSLOCKCLSTATE();
5146         nfscl_reldevinfo_locked(dip);
5147         NFSUNLOCKCLSTATE();
5148 }
5149
5150 /*
5151  * Find a layout for this file handle. Return NULL upon failure.
5152  */
5153 static struct nfscllayout *
5154 nfscl_findlayout(struct nfsclclient *clp, u_int8_t *fhp, int fhlen)
5155 {
5156         struct nfscllayout *lyp;
5157
5158         LIST_FOREACH(lyp, NFSCLLAYOUTHASH(clp, fhp, fhlen), nfsly_hash)
5159                 if (lyp->nfsly_fhlen == fhlen &&
5160                     !NFSBCMP(lyp->nfsly_fh, fhp, fhlen))
5161                         break;
5162         return (lyp);
5163 }
5164
5165 /*
5166  * Find a devinfo for this deviceid. Return NULL upon failure.
5167  */
5168 static struct nfscldevinfo *
5169 nfscl_finddevinfo(struct nfsclclient *clp, uint8_t *deviceid)
5170 {
5171         struct nfscldevinfo *dip;
5172
5173         LIST_FOREACH(dip, &clp->nfsc_devinfo, nfsdi_list)
5174                 if (NFSBCMP(dip->nfsdi_deviceid, deviceid, NFSX_V4DEVICEID)
5175                     == 0)
5176                         break;
5177         return (dip);
5178 }
5179
5180 /*
5181  * Merge the new file layout list into the main one, maintaining it in
5182  * increasing offset order.
5183  */
5184 static void
5185 nfscl_mergeflayouts(struct nfsclflayouthead *fhlp,
5186     struct nfsclflayouthead *newfhlp)
5187 {
5188         struct nfsclflayout *flp, *nflp, *prevflp, *tflp;
5189
5190         flp = LIST_FIRST(fhlp);
5191         prevflp = NULL;
5192         LIST_FOREACH_SAFE(nflp, newfhlp, nfsfl_list, tflp) {
5193                 while (flp != NULL && flp->nfsfl_off < nflp->nfsfl_off) {
5194                         prevflp = flp;
5195                         flp = LIST_NEXT(flp, nfsfl_list);
5196                 }
5197                 if (prevflp == NULL)
5198                         LIST_INSERT_HEAD(fhlp, nflp, nfsfl_list);
5199                 else
5200                         LIST_INSERT_AFTER(prevflp, nflp, nfsfl_list);
5201                 prevflp = nflp;
5202         }
5203 }
5204
5205 /*
5206  * Add this nfscldevinfo to the client, if it doesn't already exist.
5207  * This function consumes the structure pointed at by dip, if not NULL.
5208  */
5209 int
5210 nfscl_adddevinfo(struct nfsmount *nmp, struct nfscldevinfo *dip, int ind,
5211     struct nfsclflayout *flp)
5212 {
5213         struct nfsclclient *clp;
5214         struct nfscldevinfo *tdip;
5215         uint8_t *dev;
5216
5217         NFSLOCKCLSTATE();
5218         clp = nmp->nm_clp;
5219         if (clp == NULL) {
5220                 NFSUNLOCKCLSTATE();
5221                 if (dip != NULL)
5222                         free(dip, M_NFSDEVINFO);
5223                 return (ENODEV);
5224         }
5225         if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5226                 dev = flp->nfsfl_dev;
5227         else
5228                 dev = flp->nfsfl_ffm[ind].dev;
5229         tdip = nfscl_finddevinfo(clp, dev);
5230         if (tdip != NULL) {
5231                 tdip->nfsdi_layoutrefs++;
5232                 if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5233                         flp->nfsfl_devp = tdip;
5234                 else
5235                         flp->nfsfl_ffm[ind].devp = tdip;
5236                 nfscl_reldevinfo_locked(tdip);
5237                 NFSUNLOCKCLSTATE();
5238                 if (dip != NULL)
5239                         free(dip, M_NFSDEVINFO);
5240                 return (0);
5241         }
5242         if (dip != NULL) {
5243                 LIST_INSERT_HEAD(&clp->nfsc_devinfo, dip, nfsdi_list);
5244                 dip->nfsdi_layoutrefs = 1;
5245                 if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5246                         flp->nfsfl_devp = dip;
5247                 else
5248                         flp->nfsfl_ffm[ind].devp = dip;
5249         }
5250         NFSUNLOCKCLSTATE();
5251         if (dip == NULL)
5252                 return (ENODEV);
5253         return (0);
5254 }
5255
5256 /*
5257  * Free up a layout structure and associated file layout structure(s).
5258  */
5259 void
5260 nfscl_freelayout(struct nfscllayout *layp)
5261 {
5262         struct nfsclflayout *flp, *nflp;
5263         struct nfsclrecalllayout *rp, *nrp;
5264
5265         LIST_FOREACH_SAFE(flp, &layp->nfsly_flayread, nfsfl_list, nflp) {
5266                 LIST_REMOVE(flp, nfsfl_list);
5267                 nfscl_freeflayout(flp);
5268         }
5269         LIST_FOREACH_SAFE(flp, &layp->nfsly_flayrw, nfsfl_list, nflp) {
5270                 LIST_REMOVE(flp, nfsfl_list);
5271                 nfscl_freeflayout(flp);
5272         }
5273         LIST_FOREACH_SAFE(rp, &layp->nfsly_recall, nfsrecly_list, nrp) {
5274                 LIST_REMOVE(rp, nfsrecly_list);
5275                 free(rp, M_NFSLAYRECALL);
5276         }
5277         nfscl_layoutcnt--;
5278         free(layp, M_NFSLAYOUT);
5279 }
5280
5281 /*
5282  * Free up a file layout structure.
5283  */
5284 void
5285 nfscl_freeflayout(struct nfsclflayout *flp)
5286 {
5287         int i, j;
5288
5289         if ((flp->nfsfl_flags & NFSFL_FILE) != 0) {
5290                 for (i = 0; i < flp->nfsfl_fhcnt; i++)
5291                         free(flp->nfsfl_fh[i], M_NFSFH);
5292                 if (flp->nfsfl_devp != NULL)
5293                         flp->nfsfl_devp->nfsdi_layoutrefs--;
5294         }
5295         if ((flp->nfsfl_flags & NFSFL_FLEXFILE) != 0)
5296                 for (i = 0; i < flp->nfsfl_mirrorcnt; i++) {
5297                         for (j = 0; j < flp->nfsfl_ffm[i].fhcnt; j++)
5298                                 free(flp->nfsfl_ffm[i].fh[j], M_NFSFH);
5299                         if (flp->nfsfl_ffm[i].devp != NULL)
5300                                 flp->nfsfl_ffm[i].devp->nfsdi_layoutrefs--;
5301                 }
5302         free(flp, M_NFSFLAYOUT);
5303 }
5304
5305 /*
5306  * Free up a file layout devinfo structure.
5307  */
5308 void
5309 nfscl_freedevinfo(struct nfscldevinfo *dip)
5310 {
5311
5312         free(dip, M_NFSDEVINFO);
5313 }
5314
5315 /*
5316  * Mark any layouts that match as recalled.
5317  */
5318 static int
5319 nfscl_layoutrecall(int recalltype, struct nfscllayout *lyp, uint32_t iomode,
5320     uint64_t off, uint64_t len, uint32_t stateseqid, uint32_t stat, uint32_t op,
5321     char *devid, struct nfsclrecalllayout *recallp)
5322 {
5323         struct nfsclrecalllayout *rp, *orp;
5324
5325         recallp->nfsrecly_recalltype = recalltype;
5326         recallp->nfsrecly_iomode = iomode;
5327         recallp->nfsrecly_stateseqid = stateseqid;
5328         recallp->nfsrecly_off = off;
5329         recallp->nfsrecly_len = len;
5330         recallp->nfsrecly_stat = stat;
5331         recallp->nfsrecly_op = op;
5332         if (devid != NULL)
5333                 NFSBCOPY(devid, recallp->nfsrecly_devid, NFSX_V4DEVICEID);
5334         /*
5335          * Order the list as file returns first, followed by fsid and any
5336          * returns, both in increasing stateseqid order.
5337          * Note that the seqids wrap around, so 1 is after 0xffffffff.
5338          * (I'm not sure this is correct because I find RFC5661 confusing
5339          *  on this, but hopefully it will work ok.)
5340          */
5341         orp = NULL;
5342         LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) {
5343                 orp = rp;
5344                 if ((recalltype == NFSLAYOUTRETURN_FILE &&
5345                      (rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE ||
5346                       nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) ||
5347                     (recalltype != NFSLAYOUTRETURN_FILE &&
5348                      rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE &&
5349                      nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) {
5350                         LIST_INSERT_BEFORE(rp, recallp, nfsrecly_list);
5351                         break;
5352                 }
5353
5354                 /*
5355                  * Put any error return on all the file returns that will
5356                  * preceed this one.
5357                  */
5358                 if (rp->nfsrecly_recalltype == NFSLAYOUTRETURN_FILE &&
5359                    stat != 0 && rp->nfsrecly_stat == 0) {
5360                         rp->nfsrecly_stat = stat;
5361                         rp->nfsrecly_op = op;
5362                         if (devid != NULL)
5363                                 NFSBCOPY(devid, rp->nfsrecly_devid,
5364                                     NFSX_V4DEVICEID);
5365                 }
5366         }
5367         if (rp == NULL) {
5368                 if (orp == NULL)
5369                         LIST_INSERT_HEAD(&lyp->nfsly_recall, recallp,
5370                             nfsrecly_list);
5371                 else
5372                         LIST_INSERT_AFTER(orp, recallp, nfsrecly_list);
5373         }
5374         lyp->nfsly_flags |= NFSLY_RECALL;
5375         wakeup(lyp->nfsly_clp);
5376         return (0);
5377 }
5378
5379 /*
5380  * Compare the two seqids for ordering. The trick is that the seqids can
5381  * wrap around from 0xffffffff->0, so check for the cases where one
5382  * has wrapped around.
5383  * Return 1 if seqid1 comes before seqid2, 0 otherwise.
5384  */
5385 static int
5386 nfscl_seq(uint32_t seqid1, uint32_t seqid2)
5387 {
5388
5389         if (seqid2 > seqid1 && (seqid2 - seqid1) >= 0x7fffffff)
5390                 /* seqid2 has wrapped around. */
5391                 return (0);
5392         if (seqid1 > seqid2 && (seqid1 - seqid2) >= 0x7fffffff)
5393                 /* seqid1 has wrapped around. */
5394                 return (1);
5395         if (seqid1 <= seqid2)
5396                 return (1);
5397         return (0);
5398 }
5399
5400 /*
5401  * Do a layout return for each of the recalls.
5402  */
5403 static void
5404 nfscl_layoutreturn(struct nfsmount *nmp, struct nfscllayout *lyp,
5405     struct ucred *cred, NFSPROC_T *p)
5406 {
5407         struct nfsclrecalllayout *rp;
5408         nfsv4stateid_t stateid;
5409         int layouttype;
5410
5411         NFSBCOPY(lyp->nfsly_stateid.other, stateid.other, NFSX_STATEIDOTHER);
5412         stateid.seqid = lyp->nfsly_stateid.seqid;
5413         if ((lyp->nfsly_flags & NFSLY_FILES) != 0)
5414                 layouttype = NFSLAYOUT_NFSV4_1_FILES;
5415         else
5416                 layouttype = NFSLAYOUT_FLEXFILE;
5417         LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) {
5418                 (void)nfsrpc_layoutreturn(nmp, lyp->nfsly_fh,
5419                     lyp->nfsly_fhlen, 0, layouttype,
5420                     rp->nfsrecly_iomode, rp->nfsrecly_recalltype,
5421                     rp->nfsrecly_off, rp->nfsrecly_len,
5422                     &stateid, cred, p, rp->nfsrecly_stat, rp->nfsrecly_op,
5423                     rp->nfsrecly_devid);
5424         }
5425 }
5426
5427 /*
5428  * Do the layout commit for a file layout.
5429  */
5430 static void
5431 nfscl_dolayoutcommit(struct nfsmount *nmp, struct nfscllayout *lyp,
5432     struct ucred *cred, NFSPROC_T *p)
5433 {
5434         struct nfsclflayout *flp;
5435         uint64_t len;
5436         int error, layouttype;
5437
5438         if ((lyp->nfsly_flags & NFSLY_FILES) != 0)
5439                 layouttype = NFSLAYOUT_NFSV4_1_FILES;
5440         else
5441                 layouttype = NFSLAYOUT_FLEXFILE;
5442         LIST_FOREACH(flp, &lyp->nfsly_flayrw, nfsfl_list) {
5443                 if (layouttype == NFSLAYOUT_FLEXFILE &&
5444                     (flp->nfsfl_fflags & NFSFLEXFLAG_NO_LAYOUTCOMMIT) != 0) {
5445                         NFSCL_DEBUG(4, "Flex file: no layoutcommit\n");
5446                         /* If not supported, don't bother doing it. */
5447                         NFSLOCKMNT(nmp);
5448                         nmp->nm_state |= NFSSTA_NOLAYOUTCOMMIT;
5449                         NFSUNLOCKMNT(nmp);
5450                         break;
5451                 } else if (flp->nfsfl_off <= lyp->nfsly_lastbyte) {
5452                         len = flp->nfsfl_end - flp->nfsfl_off;
5453                         error = nfsrpc_layoutcommit(nmp, lyp->nfsly_fh,
5454                             lyp->nfsly_fhlen, 0, flp->nfsfl_off, len,
5455                             lyp->nfsly_lastbyte, &lyp->nfsly_stateid,
5456                             layouttype, cred, p, NULL);
5457                         NFSCL_DEBUG(4, "layoutcommit err=%d\n", error);
5458                         if (error == NFSERR_NOTSUPP) {
5459                                 /* If not supported, don't bother doing it. */
5460                                 NFSLOCKMNT(nmp);
5461                                 nmp->nm_state |= NFSSTA_NOLAYOUTCOMMIT;
5462                                 NFSUNLOCKMNT(nmp);
5463                                 break;
5464                         }
5465                 }
5466         }
5467 }
5468
5469 /*
5470  * Commit all layouts for a file (vnode).
5471  */
5472 int
5473 nfscl_layoutcommit(vnode_t vp, NFSPROC_T *p)
5474 {
5475         struct nfsclclient *clp;
5476         struct nfscllayout *lyp;
5477         struct nfsnode *np = VTONFS(vp);
5478         mount_t mp;
5479         struct nfsmount *nmp;
5480
5481         mp = vnode_mount(vp);
5482         nmp = VFSTONFS(mp);
5483         if (NFSHASNOLAYOUTCOMMIT(nmp))
5484                 return (0);
5485         NFSLOCKCLSTATE();
5486         clp = nmp->nm_clp;
5487         if (clp == NULL) {
5488                 NFSUNLOCKCLSTATE();
5489                 return (EPERM);
5490         }
5491         lyp = nfscl_findlayout(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
5492         if (lyp == NULL) {
5493                 NFSUNLOCKCLSTATE();
5494                 return (EPERM);
5495         }
5496         nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
5497         if (NFSCL_FORCEDISM(mp)) {
5498                 NFSUNLOCKCLSTATE();
5499                 return (EPERM);
5500         }
5501 tryagain:
5502         if ((lyp->nfsly_flags & NFSLY_WRITTEN) != 0) {
5503                 lyp->nfsly_flags &= ~NFSLY_WRITTEN;
5504                 NFSUNLOCKCLSTATE();
5505                 NFSCL_DEBUG(4, "do layoutcommit2\n");
5506                 nfscl_dolayoutcommit(clp->nfsc_nmp, lyp, NFSPROCCRED(p), p);
5507                 NFSLOCKCLSTATE();
5508                 goto tryagain;
5509         }
5510         nfsv4_relref(&lyp->nfsly_lock);
5511         NFSUNLOCKCLSTATE();
5512         return (0);
5513 }
5514