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 *, bool *, struct ucred *,
 112     NFSPROC_T *);
 113 static void nfscl_insertlock(struct nfscllockowner *, struct nfscllock *,
 114     struct nfscllock *, int);
 115 static int nfscl_updatelock(struct nfscllockowner *, struct nfscllock **,
 116     struct nfscllock **, int);
 117 static void nfscl_delegreturnall(struct nfsclclient *, NFSPROC_T *);
 118 static u_int32_t nfscl_nextcbident(void);
 119 static mount_t nfscl_getmnt(int, uint8_t *, u_int32_t, struct nfsclclient **);
 120 static struct nfsclclient *nfscl_getclnt(u_int32_t);
 121 static struct nfsclclient *nfscl_getclntsess(uint8_t *);
 122 static struct nfscldeleg *nfscl_finddeleg(struct nfsclclient *, u_int8_t *,
 123     int);
 124 static void nfscl_retoncloselayout(vnode_t, struct nfsclclient *, uint8_t *,
 125     int, struct nfsclrecalllayout **);
 126 static void nfscl_reldevinfo_locked(struct nfscldevinfo *);
 127 static struct nfscllayout *nfscl_findlayout(struct nfsclclient *, u_int8_t *,
 128     int);
 129 static struct nfscldevinfo *nfscl_finddevinfo(struct nfsclclient *, uint8_t *);
 130 static int nfscl_checkconflict(struct nfscllockownerhead *, struct nfscllock *,
 131     u_int8_t *, struct nfscllock **);
 132 static void nfscl_freealllocks(struct nfscllockownerhead *, int);
 133 static int nfscl_localconflict(struct nfsclclient *, u_int8_t *, int,
 134     struct nfscllock *, u_int8_t *, struct nfscldeleg *, struct nfscllock **);
 135 static void nfscl_newopen(struct nfsclclient *, struct nfscldeleg *,
 136     struct nfsclowner **, struct nfsclowner **, struct nfsclopen **,
 137     struct nfsclopen **, u_int8_t *, u_int8_t *, int, struct ucred *, int *);
 138 static int nfscl_moveopen(vnode_t , struct nfsclclient *,
 139     struct nfsmount *, struct nfsclopen *, struct nfsclowner *,
 140     struct nfscldeleg *, struct ucred *, NFSPROC_T *);
 141 static void nfscl_totalrecall(struct nfsclclient *);
 142 static int nfscl_relock(vnode_t , struct nfsclclient *, struct nfsmount *,
 143     struct nfscllockowner *, struct nfscllock *, struct ucred *, NFSPROC_T *);
 144 static int nfscl_tryopen(struct nfsmount *, vnode_t , u_int8_t *, int,
 145     u_int8_t *, int, u_int32_t, struct nfsclopen *, u_int8_t *, int,
 146     struct nfscldeleg **, int, u_int32_t, struct ucred *, NFSPROC_T *);
 147 static int nfscl_trylock(struct nfsmount *, vnode_t , u_int8_t *,
 148     int, struct nfscllockowner *, int, int, u_int64_t, u_int64_t, short,
 149     struct ucred *, NFSPROC_T *);
 150 static int nfsrpc_reopen(struct nfsmount *, u_int8_t *, int, u_int32_t,
 151     struct nfsclopen *, struct nfscldeleg **, struct ucred *, NFSPROC_T *);
 152 static void nfscl_freedeleg(struct nfscldeleghead *, struct nfscldeleg *);
 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 static void nfscl_freeopenowner(struct nfsclowner *, int);
 158 static void nfscl_cleandeleg(struct nfscldeleg *);
 159 static int nfscl_trydelegreturn(struct nfscldeleg *, struct ucred *,
 160     struct nfsmount *, NFSPROC_T *);
 161 static void nfscl_emptylockowner(struct nfscllockowner *,
 162     struct nfscllockownerfhhead *);
 163 static void nfscl_mergeflayouts(struct nfsclflayouthead *,
 164     struct nfsclflayouthead *);
 165 static int nfscl_layoutrecall(int, struct nfscllayout *, uint32_t, uint64_t,
 166     uint64_t, uint32_t, uint32_t, uint32_t, char *, struct nfsclrecalllayout *);
 167 static int nfscl_seq(uint32_t, uint32_t);
 168 static void nfscl_layoutreturn(struct nfsmount *, struct nfscllayout *,
 169     struct ucred *, NFSPROC_T *);
 170 static void nfscl_dolayoutcommit(struct nfsmount *, struct nfscllayout *,
 171     struct ucred *, NFSPROC_T *);
 172
 173 static short nfscberr_null[] = {
 174         0,
 175         0,
 176 };
 177
 178 static short nfscberr_getattr[] = {
 179         NFSERR_RESOURCE,
 180         NFSERR_BADHANDLE,
 181         NFSERR_BADXDR,
 182         NFSERR_RESOURCE,
 183         NFSERR_SERVERFAULT,
 184         0,
 185 };
 186
 187 static short nfscberr_recall[] = {
 188         NFSERR_RESOURCE,
 189         NFSERR_BADHANDLE,
 190         NFSERR_BADSTATEID,
 191         NFSERR_BADXDR,
 192         NFSERR_RESOURCE,
 193         NFSERR_SERVERFAULT,
 194         0,
 195 };
 196
 197 static short *nfscl_cberrmap[] = {
 198         nfscberr_null,
 199         nfscberr_null,
 200         nfscberr_null,
 201         nfscberr_getattr,
 202         nfscberr_recall
 203 };
 204
 205 #define NETFAMILY(clp) \
 206                 (((clp)->nfsc_flags & NFSCLFLAGS_AFINET6) ? AF_INET6 : AF_INET)
 207
 208 /*
 209  * Called for an open operation.
 210  * If the nfhp argument is NULL, just get an openowner.
 211  */
 212 int
 213 nfscl_open(vnode_t vp, u_int8_t *nfhp, int fhlen, u_int32_t amode, int usedeleg,
 214     struct ucred *cred, NFSPROC_T *p, struct nfsclowner **owpp,
 215     struct nfsclopen **opp, int *newonep, int *retp, int lockit)
 216 {
 217         struct nfsclclient *clp;
 218         struct nfsclowner *owp, *nowp;
 219         struct nfsclopen *op = NULL, *nop = NULL;
 220         struct nfscldeleg *dp;
 221         struct nfsclownerhead *ohp;
 222         u_int8_t own[NFSV4CL_LOCKNAMELEN];
 223         int ret;
 224
 225         if (newonep != NULL)
 226                 *newonep = 0;
 227         if (opp != NULL)
 228                 *opp = NULL;
 229         if (owpp != NULL)
 230                 *owpp = NULL;
 231
 232         /*
 233          * Might need one or both of these, so MALLOC them now, to
 234          * avoid a tsleep() in MALLOC later.
 235          */
 236         nowp = malloc(sizeof (struct nfsclowner),
 237             M_NFSCLOWNER, M_WAITOK);
 238         if (nfhp != NULL)
 239             nop = malloc(sizeof (struct nfsclopen) +
 240                 fhlen - 1, M_NFSCLOPEN, M_WAITOK);
 241         ret = nfscl_getcl(vp->v_mount, cred, p, 1, &clp);
 242         if (ret != 0) {
 243                 free(nowp, M_NFSCLOWNER);
 244                 if (nop != NULL)
 245                         free(nop, M_NFSCLOPEN);
 246                 return (ret);
 247         }
 248
 249         /*
 250          * Get the Open iff it already exists.
 251          * If none found, add the new one or return error, depending upon
 252          * "create".
 253          */
 254         NFSLOCKCLSTATE();
 255         dp = NULL;
 256         /* First check the delegation list */
 257         if (nfhp != NULL && usedeleg) {
 258                 LIST_FOREACH(dp, NFSCLDELEGHASH(clp, nfhp, fhlen), nfsdl_hash) {
 259                         if (dp->nfsdl_fhlen == fhlen &&
 260                             !NFSBCMP(nfhp, dp->nfsdl_fh, fhlen)) {
 261                                 if (!(amode & NFSV4OPEN_ACCESSWRITE) ||
 262                                     (dp->nfsdl_flags & NFSCLDL_WRITE))
 263                                         break;
 264                                 dp = NULL;
 265                                 break;
 266                         }
 267                 }
 268         }
 269
 270         if (dp != NULL) {
 271                 nfscl_filllockowner(p->td_proc, own, F_POSIX);
 272                 ohp = &dp->nfsdl_owner;
 273         } else {
 274                 /* For NFSv4.1 and this option, use a single open_owner. */
 275                 if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount)))
 276                         nfscl_filllockowner(NULL, own, F_POSIX);
 277                 else
 278                         nfscl_filllockowner(p->td_proc, own, F_POSIX);
 279                 ohp = &clp->nfsc_owner;
 280         }
 281         /* Now, search for an openowner */
 282         LIST_FOREACH(owp, ohp, nfsow_list) {
 283                 if (!NFSBCMP(owp->nfsow_owner, own, NFSV4CL_LOCKNAMELEN))
 284                         break;
 285         }
 286
 287         /*
 288          * Create a new open, as required.
 289          */
 290         nfscl_newopen(clp, dp, &owp, &nowp, &op, &nop, own, nfhp, fhlen,
 291             cred, newonep);
 292
 293         /*
 294          * Now, check the mode on the open and return the appropriate
 295          * value.
 296          */
 297         if (retp != NULL) {
 298                 if (nfhp != NULL && dp != NULL && nop == NULL)
 299                         /* new local open on delegation */
 300                         *retp = NFSCLOPEN_SETCRED;
 301                 else
 302                         *retp = NFSCLOPEN_OK;
 303         }
 304         if (op != NULL && (amode & ~(op->nfso_mode))) {
 305                 op->nfso_mode |= amode;
 306                 if (retp != NULL && dp == NULL)
 307                         *retp = NFSCLOPEN_DOOPEN;
 308         }
 309
 310         /*
 311          * Serialize modifications to the open owner for multiple threads
 312          * within the same process using a read/write sleep lock.
 313          * For NFSv4.1 and a single OpenOwner, allow concurrent open operations
 314          * by acquiring a shared lock.  The close operations still use an
 315          * exclusive lock for this case.
 316          */
 317         if (lockit != 0) {
 318                 if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount))) {
 319                         /*
 320                          * Get a shared lock on the OpenOwner, but first
 321                          * wait for any pending exclusive lock, so that the
 322                          * exclusive locker gets priority.
 323                          */
 324                         nfsv4_lock(&owp->nfsow_rwlock, 0, NULL,
 325                             NFSCLSTATEMUTEXPTR, NULL);
 326                         nfsv4_getref(&owp->nfsow_rwlock, NULL,
 327                             NFSCLSTATEMUTEXPTR, NULL);
 328                 } else
 329                         nfscl_lockexcl(&owp->nfsow_rwlock, NFSCLSTATEMUTEXPTR);
 330         }
 331         NFSUNLOCKCLSTATE();
 332         if (nowp != NULL)
 333                 free(nowp, M_NFSCLOWNER);
 334         if (nop != NULL)
 335                 free(nop, M_NFSCLOPEN);
 336         if (owpp != NULL)
 337                 *owpp = owp;
 338         if (opp != NULL)
 339                 *opp = op;
 340         return (0);
 341 }
 342
 343 /*
 344  * Create a new open, as required.
 345  */
 346 static void
 347 nfscl_newopen(struct nfsclclient *clp, struct nfscldeleg *dp,
 348     struct nfsclowner **owpp, struct nfsclowner **nowpp, struct nfsclopen **opp,
 349     struct nfsclopen **nopp, u_int8_t *own, u_int8_t *fhp, int fhlen,
 350     struct ucred *cred, int *newonep)
 351 {
 352         struct nfsclowner *owp = *owpp, *nowp;
 353         struct nfsclopen *op, *nop;
 354
 355         if (nowpp != NULL)
 356                 nowp = *nowpp;
 357         else
 358                 nowp = NULL;
 359         if (nopp != NULL)
 360                 nop = *nopp;
 361         else
 362                 nop = NULL;
 363         if (owp == NULL && nowp != NULL) {
 364                 NFSBCOPY(own, nowp->nfsow_owner, NFSV4CL_LOCKNAMELEN);
 365                 LIST_INIT(&nowp->nfsow_open);
 366                 nowp->nfsow_clp = clp;
 367                 nowp->nfsow_seqid = 0;
 368                 nowp->nfsow_defunct = 0;
 369                 nfscl_lockinit(&nowp->nfsow_rwlock);
 370                 if (dp != NULL) {
 371                         nfsstatsv1.cllocalopenowners++;
 372                         LIST_INSERT_HEAD(&dp->nfsdl_owner, nowp, nfsow_list);
 373                 } else {
 374                         nfsstatsv1.clopenowners++;
 375                         LIST_INSERT_HEAD(&clp->nfsc_owner, nowp, nfsow_list);
 376                 }
 377                 owp = *owpp = nowp;
 378                 *nowpp = NULL;
 379                 if (newonep != NULL)
 380                         *newonep = 1;
 381         }
 382
 383          /* If an fhp has been specified, create an Open as well. */
 384         if (fhp != NULL) {
 385                 /* and look for the correct open, based upon FH */
 386                 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
 387                         if (op->nfso_fhlen == fhlen &&
 388                             !NFSBCMP(op->nfso_fh, fhp, fhlen))
 389                                 break;
 390                 }
 391                 if (op == NULL && nop != NULL) {
 392                         nop->nfso_own = owp;
 393                         nop->nfso_mode = 0;
 394                         nop->nfso_opencnt = 0;
 395                         nop->nfso_posixlock = 1;
 396                         nop->nfso_fhlen = fhlen;
 397                         NFSBCOPY(fhp, nop->nfso_fh, fhlen);
 398                         LIST_INIT(&nop->nfso_lock);
 399                         nop->nfso_stateid.seqid = 0;
 400                         nop->nfso_stateid.other[0] = 0;
 401                         nop->nfso_stateid.other[1] = 0;
 402                         nop->nfso_stateid.other[2] = 0;
 403                         KASSERT(cred != NULL, ("%s: cred NULL\n", __func__));
 404                         newnfs_copyincred(cred, &nop->nfso_cred);
 405                         if (dp != NULL) {
 406                                 TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
 407                                 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp,
 408                                     nfsdl_list);
 409                                 dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
 410                                 nfsstatsv1.cllocalopens++;
 411                         } else {
 412                                 nfsstatsv1.clopens++;
 413                         }
 414                         LIST_INSERT_HEAD(&owp->nfsow_open, nop, nfso_list);
 415                         *opp = nop;
 416                         *nopp = NULL;
 417                         if (newonep != NULL)
 418                                 *newonep = 1;
 419                 } else {
 420                         *opp = op;
 421                 }
 422         }
 423 }
 424
 425 /*
 426  * Called to find/add a delegation to a client.
 427  */
 428 int
 429 nfscl_deleg(mount_t mp, struct nfsclclient *clp, u_int8_t *nfhp,
 430     int fhlen, struct ucred *cred, NFSPROC_T *p, struct nfscldeleg **dpp)
 431 {
 432         struct nfscldeleg *dp = *dpp, *tdp;
 433
 434         /*
 435          * First, if we have received a Read delegation for a file on a
 436          * read/write file system, just return it, because they aren't
 437          * useful, imho.
 438          */
 439         if (mp != NULL && dp != NULL && !NFSMNT_RDONLY(mp) &&
 440             (dp->nfsdl_flags & NFSCLDL_READ)) {
 441                 (void) nfscl_trydelegreturn(dp, cred, VFSTONFS(mp), p);
 442                 free(dp, M_NFSCLDELEG);
 443                 *dpp = NULL;
 444                 return (0);
 445         }
 446
 447         /* Look for the correct deleg, based upon FH */
 448         NFSLOCKCLSTATE();
 449         tdp = nfscl_finddeleg(clp, nfhp, fhlen);
 450         if (tdp == NULL) {
 451                 if (dp == NULL) {
 452                         NFSUNLOCKCLSTATE();
 453                         return (NFSERR_BADSTATEID);
 454                 }
 455                 *dpp = NULL;
 456                 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list);
 457                 LIST_INSERT_HEAD(NFSCLDELEGHASH(clp, nfhp, fhlen), dp,
 458                     nfsdl_hash);
 459                 dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
 460                 nfsstatsv1.cldelegates++;
 461                 nfscl_delegcnt++;
 462         } else {
 463                 /*
 464                  * Delegation already exists, what do we do if a new one??
 465                  */
 466                 if (dp != NULL) {
 467                         printf("Deleg already exists!\n");
 468                         free(dp, M_NFSCLDELEG);
 469                         *dpp = NULL;
 470                 } else {
 471                         *dpp = tdp;
 472                 }
 473         }
 474         NFSUNLOCKCLSTATE();
 475         return (0);
 476 }
 477
 478 /*
 479  * Find a delegation for this file handle. Return NULL upon failure.
 480  */
 481 static struct nfscldeleg *
 482 nfscl_finddeleg(struct nfsclclient *clp, u_int8_t *fhp, int fhlen)
 483 {
 484         struct nfscldeleg *dp;
 485
 486         LIST_FOREACH(dp, NFSCLDELEGHASH(clp, fhp, fhlen), nfsdl_hash) {
 487             if (dp->nfsdl_fhlen == fhlen &&
 488                 !NFSBCMP(dp->nfsdl_fh, fhp, fhlen))
 489                 break;
 490         }
 491         return (dp);
 492 }
 493
 494 /*
 495  * Get a stateid for an I/O operation. First, look for an open and iff
 496  * found, return either a lockowner stateid or the open stateid.
 497  * If no Open is found, just return error and the special stateid of all zeros.
 498  */
 499 int
 500 nfscl_getstateid(vnode_t vp, u_int8_t *nfhp, int fhlen, u_int32_t mode,
 501     int fords, struct ucred *cred, NFSPROC_T *p, nfsv4stateid_t *stateidp,
 502     void **lckpp)
 503 {
 504         struct nfsclclient *clp;
 505         struct nfsclowner *owp;
 506         struct nfsclopen *op = NULL, *top;
 507         struct nfscllockowner *lp;
 508         struct nfscldeleg *dp;
 509         struct nfsnode *np;
 510         struct nfsmount *nmp;
 511         u_int8_t own[NFSV4CL_LOCKNAMELEN];
 512         int error, done;
 513
 514         *lckpp = NULL;
 515         /*
 516          * Initially, just set the special stateid of all zeros.
 517          * (Don't do this for a DS, since the special stateid can't be used.)
 518          */
 519         if (fords == 0) {
 520                 stateidp->seqid = 0;
 521                 stateidp->other[0] = 0;
 522                 stateidp->other[1] = 0;
 523                 stateidp->other[2] = 0;
 524         }
 525         if (vnode_vtype(vp) != VREG)
 526                 return (EISDIR);
 527         np = VTONFS(vp);
 528         nmp = VFSTONFS(vp->v_mount);
 529         NFSLOCKCLSTATE();
 530         clp = nfscl_findcl(nmp);
 531         if (clp == NULL) {
 532                 NFSUNLOCKCLSTATE();
 533                 return (EACCES);
 534         }
 535
 536         /*
 537          * Wait for recovery to complete.
 538          */
 539         while ((clp->nfsc_flags & NFSCLFLAGS_RECVRINPROG))
 540                 (void) nfsmsleep(&clp->nfsc_flags, NFSCLSTATEMUTEXPTR,
 541                     PZERO, "nfsrecvr", NULL);
 542
 543         /*
 544          * First, look for a delegation.
 545          */
 546         LIST_FOREACH(dp, NFSCLDELEGHASH(clp, nfhp, fhlen), nfsdl_hash) {
 547                 if (dp->nfsdl_fhlen == fhlen &&
 548                     !NFSBCMP(nfhp, dp->nfsdl_fh, fhlen)) {
 549                         if (!(mode & NFSV4OPEN_ACCESSWRITE) ||
 550                             (dp->nfsdl_flags & NFSCLDL_WRITE)) {
 551                                 stateidp->seqid = dp->nfsdl_stateid.seqid;
 552                                 stateidp->other[0] = dp->nfsdl_stateid.other[0];
 553                                 stateidp->other[1] = dp->nfsdl_stateid.other[1];
 554                                 stateidp->other[2] = dp->nfsdl_stateid.other[2];
 555                                 if (!(np->n_flag & NDELEGRECALL)) {
 556                                         TAILQ_REMOVE(&clp->nfsc_deleg, dp,
 557                                             nfsdl_list);
 558                                         TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp,
 559                                             nfsdl_list);
 560                                         dp->nfsdl_timestamp = NFSD_MONOSEC +
 561                                             120;
 562                                         dp->nfsdl_rwlock.nfslock_usecnt++;
 563                                         *lckpp = (void *)&dp->nfsdl_rwlock;
 564                                 }
 565                                 NFSUNLOCKCLSTATE();
 566                                 return (0);
 567                         }
 568                         break;
 569                 }
 570         }
 571
 572         if (p != NULL) {
 573                 /*
 574                  * If p != NULL, we want to search the parentage tree
 575                  * for a matching OpenOwner and use that.
 576                  */
 577                 if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount)))
 578                         nfscl_filllockowner(NULL, own, F_POSIX);
 579                 else
 580                         nfscl_filllockowner(p->td_proc, own, F_POSIX);
 581                 lp = NULL;
 582                 error = nfscl_getopen(&clp->nfsc_owner, nfhp, fhlen, own, own,
 583                     mode, &lp, &op);
 584                 if (error == 0 && lp != NULL && fords == 0) {
 585                         /* Don't return a lock stateid for a DS. */
 586                         stateidp->seqid =
 587                             lp->nfsl_stateid.seqid;
 588                         stateidp->other[0] =
 589                             lp->nfsl_stateid.other[0];
 590                         stateidp->other[1] =
 591                             lp->nfsl_stateid.other[1];
 592                         stateidp->other[2] =
 593                             lp->nfsl_stateid.other[2];
 594                         NFSUNLOCKCLSTATE();
 595                         return (0);
 596                 }
 597         }
 598         if (op == NULL) {
 599                 /* If not found, just look for any OpenOwner that will work. */
 600                 top = NULL;
 601                 done = 0;
 602                 owp = LIST_FIRST(&clp->nfsc_owner);
 603                 while (!done && owp != NULL) {
 604                         LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
 605                                 if (op->nfso_fhlen == fhlen &&
 606                                     !NFSBCMP(op->nfso_fh, nfhp, fhlen)) {
 607                                         if (top == NULL && (op->nfso_mode &
 608                                             NFSV4OPEN_ACCESSWRITE) != 0 &&
 609                                             (mode & NFSV4OPEN_ACCESSREAD) != 0)
 610                                                 top = op;
 611                                         if ((mode & op->nfso_mode) == mode) {
 612                                                 done = 1;
 613                                                 break;
 614                                         }
 615                                 }
 616                         }
 617                         if (!done)
 618                                 owp = LIST_NEXT(owp, nfsow_list);
 619                 }
 620                 if (!done) {
 621                         NFSCL_DEBUG(2, "openmode top=%p\n", top);
 622                         if (top == NULL || NFSHASOPENMODE(nmp)) {
 623                                 NFSUNLOCKCLSTATE();
 624                                 return (ENOENT);
 625                         } else
 626                                 op = top;
 627                 }
 628                 /*
 629                  * For read aheads or write behinds, use the open cred.
 630                  * A read ahead or write behind is indicated by p == NULL.
 631                  */
 632                 if (p == NULL)
 633                         newnfs_copycred(&op->nfso_cred, cred);
 634         }
 635
 636         /*
 637          * No lock stateid, so return the open stateid.
 638          */
 639         stateidp->seqid = op->nfso_stateid.seqid;
 640         stateidp->other[0] = op->nfso_stateid.other[0];
 641         stateidp->other[1] = op->nfso_stateid.other[1];
 642         stateidp->other[2] = op->nfso_stateid.other[2];
 643         NFSUNLOCKCLSTATE();
 644         return (0);
 645 }
 646
 647 /*
 648  * Search for a matching file, mode and, optionally, lockowner.
 649  */
 650 static int
 651 nfscl_getopen(struct nfsclownerhead *ohp, u_int8_t *nfhp, int fhlen,
 652     u_int8_t *openown, u_int8_t *lockown, u_int32_t mode,
 653     struct nfscllockowner **lpp, struct nfsclopen **opp)
 654 {
 655         struct nfsclowner *owp;
 656         struct nfsclopen *op, *rop, *rop2;
 657         struct nfscllockowner *lp;
 658         int keep_looping;
 659
 660         if (lpp != NULL)
 661                 *lpp = NULL;
 662         /*
 663          * rop will be set to the open to be returned. There are three
 664          * variants of this, all for an open of the correct file:
 665          * 1 - A match of lockown.
 666          * 2 - A match of the openown, when no lockown match exists.
 667          * 3 - A match for any open, if no openown or lockown match exists.
 668          * Looking for #2 over #3 probably isn't necessary, but since
 669          * RFC3530 is vague w.r.t. the relationship between openowners and
 670          * lockowners, I think this is the safer way to go.
 671          */
 672         rop = NULL;
 673         rop2 = NULL;
 674         keep_looping = 1;
 675         /* Search the client list */
 676         owp = LIST_FIRST(ohp);
 677         while (owp != NULL && keep_looping != 0) {
 678                 /* and look for the correct open */
 679                 op = LIST_FIRST(&owp->nfsow_open);
 680                 while (op != NULL && keep_looping != 0) {
 681                         if (op->nfso_fhlen == fhlen &&
 682                             !NFSBCMP(op->nfso_fh, nfhp, fhlen)
 683                             && (op->nfso_mode & mode) == mode) {
 684                                 if (lpp != NULL) {
 685                                         /* Now look for a matching lockowner. */
 686                                         LIST_FOREACH(lp, &op->nfso_lock,
 687                                             nfsl_list) {
 688                                                 if (!NFSBCMP(lp->nfsl_owner,
 689                                                     lockown,
 690                                                     NFSV4CL_LOCKNAMELEN)) {
 691                                                         *lpp = lp;
 692                                                         rop = op;
 693                                                         keep_looping = 0;
 694                                                         break;
 695                                                 }
 696                                         }
 697                                 }
 698                                 if (rop == NULL && !NFSBCMP(owp->nfsow_owner,
 699                                     openown, NFSV4CL_LOCKNAMELEN)) {
 700                                         rop = op;
 701                                         if (lpp == NULL)
 702                                                 keep_looping = 0;
 703                                 }
 704                                 if (rop2 == NULL)
 705                                         rop2 = op;
 706                         }
 707                         op = LIST_NEXT(op, nfso_list);
 708                 }
 709                 owp = LIST_NEXT(owp, nfsow_list);
 710         }
 711         if (rop == NULL)
 712                 rop = rop2;
 713         if (rop == NULL)
 714                 return (EBADF);
 715         *opp = rop;
 716         return (0);
 717 }
 718
 719 /*
 720  * Release use of an open owner. Called when open operations are done
 721  * with the open owner.
 722  */
 723 void
 724 nfscl_ownerrelease(struct nfsmount *nmp, struct nfsclowner *owp,
 725     __unused int error, __unused int candelete, int unlocked)
 726 {
 727
 728         if (owp == NULL)
 729                 return;
 730         NFSLOCKCLSTATE();
 731         if (unlocked == 0) {
 732                 if (NFSHASONEOPENOWN(nmp))
 733                         nfsv4_relref(&owp->nfsow_rwlock);
 734                 else
 735                         nfscl_lockunlock(&owp->nfsow_rwlock);
 736         }
 737         nfscl_clrelease(owp->nfsow_clp);
 738         NFSUNLOCKCLSTATE();
 739 }
 740
 741 /*
 742  * Release use of an open structure under an open owner.
 743  */
 744 void
 745 nfscl_openrelease(struct nfsmount *nmp, struct nfsclopen *op, int error,
 746     int candelete)
 747 {
 748         struct nfsclclient *clp;
 749         struct nfsclowner *owp;
 750
 751         if (op == NULL)
 752                 return;
 753         NFSLOCKCLSTATE();
 754         owp = op->nfso_own;
 755         if (NFSHASONEOPENOWN(nmp))
 756                 nfsv4_relref(&owp->nfsow_rwlock);
 757         else
 758                 nfscl_lockunlock(&owp->nfsow_rwlock);
 759         clp = owp->nfsow_clp;
 760         if (error && candelete && op->nfso_opencnt == 0)
 761                 nfscl_freeopen(op, 0);
 762         nfscl_clrelease(clp);
 763         NFSUNLOCKCLSTATE();
 764 }
 765
 766 /*
 767  * Called to get a clientid structure. It will optionally lock the
 768  * client data structures to do the SetClientId/SetClientId_confirm,
 769  * but will release that lock and return the clientid with a reference
 770  * count on it.
 771  * If the "cred" argument is NULL, a new clientid should not be created.
 772  * If the "p" argument is NULL, a SetClientID/SetClientIDConfirm cannot
 773  * be done.
 774  * The start_renewthread argument tells nfscl_getcl() to start a renew
 775  * thread if this creates a new clp.
 776  * It always clpp with a reference count on it, unless returning an error.
 777  */
 778 int
 779 nfscl_getcl(struct mount *mp, struct ucred *cred, NFSPROC_T *p,
 780     int start_renewthread, struct nfsclclient **clpp)
 781 {
 782         struct nfsclclient *clp;
 783         struct nfsclclient *newclp = NULL;
 784         struct nfsmount *nmp;
 785         char uuid[HOSTUUIDLEN];
 786         int igotlock = 0, error, trystalecnt, clidinusedelay, i;
 787         u_int16_t idlen = 0;
 788
 789         nmp = VFSTONFS(mp);
 790         if (cred != NULL) {
 791                 getcredhostuuid(cred, uuid, sizeof uuid);
 792                 idlen = strlen(uuid);
 793                 if (idlen > 0)
 794                         idlen += sizeof (u_int64_t);
 795                 else
 796                         idlen += sizeof (u_int64_t) + 16; /* 16 random bytes */
 797                 newclp = malloc(
 798                     sizeof (struct nfsclclient) + idlen - 1, M_NFSCLCLIENT,
 799                     M_WAITOK | M_ZERO);
 800         }
 801         NFSLOCKCLSTATE();
 802         /*
 803          * If a forced dismount is already in progress, don't
 804          * allocate a new clientid and get out now. For the case where
 805          * clp != NULL, this is a harmless optimization.
 806          */
 807         if (NFSCL_FORCEDISM(mp)) {
 808                 NFSUNLOCKCLSTATE();
 809                 if (newclp != NULL)
 810                         free(newclp, M_NFSCLCLIENT);
 811                 return (EBADF);
 812         }
 813         clp = nmp->nm_clp;
 814         if (clp == NULL) {
 815                 if (newclp == NULL) {
 816                         NFSUNLOCKCLSTATE();
 817                         return (EACCES);
 818                 }
 819                 clp = newclp;
 820                 clp->nfsc_idlen = idlen;
 821                 LIST_INIT(&clp->nfsc_owner);
 822                 TAILQ_INIT(&clp->nfsc_deleg);
 823                 TAILQ_INIT(&clp->nfsc_layout);
 824                 LIST_INIT(&clp->nfsc_devinfo);
 825                 for (i = 0; i < NFSCLDELEGHASHSIZE; i++)
 826                         LIST_INIT(&clp->nfsc_deleghash[i]);
 827                 for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++)
 828                         LIST_INIT(&clp->nfsc_layouthash[i]);
 829                 clp->nfsc_flags = NFSCLFLAGS_INITED;
 830                 clp->nfsc_clientidrev = 1;
 831                 clp->nfsc_cbident = nfscl_nextcbident();
 832                 nfscl_fillclid(nmp->nm_clval, uuid, clp->nfsc_id,
 833                     clp->nfsc_idlen);
 834                 LIST_INSERT_HEAD(&nfsclhead, clp, nfsc_list);
 835                 nmp->nm_clp = clp;
 836                 clp->nfsc_nmp = nmp;
 837                 NFSUNLOCKCLSTATE();
 838                 if (start_renewthread != 0)
 839                         nfscl_start_renewthread(clp);
 840         } else {
 841                 NFSUNLOCKCLSTATE();
 842                 if (newclp != NULL)
 843                         free(newclp, M_NFSCLCLIENT);
 844         }
 845         NFSLOCKCLSTATE();
 846         while ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0 && !igotlock &&
 847             !NFSCL_FORCEDISM(mp))
 848                 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
 849                     NFSCLSTATEMUTEXPTR, mp);
 850         if (igotlock == 0) {
 851                 /*
 852                  * Call nfsv4_lock() with "iwantlock == 0" so that it will
 853                  * wait for a pending exclusive lock request.  This gives the
 854                  * exclusive lock request priority over this shared lock
 855                  * request.
 856                  * An exclusive lock on nfsc_lock is used mainly for server
 857                  * crash recoveries.
 858                  */
 859                 nfsv4_lock(&clp->nfsc_lock, 0, NULL, NFSCLSTATEMUTEXPTR, mp);
 860                 nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
 861         }
 862         if (igotlock == 0 && NFSCL_FORCEDISM(mp)) {
 863                 /*
 864                  * Both nfsv4_lock() and nfsv4_getref() know to check
 865                  * for NFSCL_FORCEDISM() and return without sleeping to
 866                  * wait for the exclusive lock to be released, since it
 867                  * might be held by nfscl_umount() and we need to get out
 868                  * now for that case and not wait until nfscl_umount()
 869                  * releases it.
 870                  */
 871                 NFSUNLOCKCLSTATE();
 872                 return (EBADF);
 873         }
 874         NFSUNLOCKCLSTATE();
 875
 876         /*
 877          * If it needs a clientid, do the setclientid now.
 878          */
 879         if ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0) {
 880                 if (!igotlock)
 881                         panic("nfscl_clget");
 882                 if (p == NULL || cred == NULL) {
 883                         NFSLOCKCLSTATE();
 884                         nfsv4_unlock(&clp->nfsc_lock, 0);
 885                         NFSUNLOCKCLSTATE();
 886                         return (EACCES);
 887                 }
 888                 /*
 889                  * If RFC3530 Sec. 14.2.33 is taken literally,
 890                  * NFSERR_CLIDINUSE will be returned persistently for the
 891                  * case where a new mount of the same file system is using
 892                  * a different principal. In practice, NFSERR_CLIDINUSE is
 893                  * only returned when there is outstanding unexpired state
 894                  * on the clientid. As such, try for twice the lease
 895                  * interval, if we know what that is. Otherwise, make a
 896                  * wild ass guess.
 897                  * The case of returning NFSERR_STALECLIENTID is far less
 898                  * likely, but might occur if there is a significant delay
 899                  * between doing the SetClientID and SetClientIDConfirm Ops,
 900                  * such that the server throws away the clientid before
 901                  * receiving the SetClientIDConfirm.
 902                  */
 903                 if (clp->nfsc_renew > 0)
 904                         clidinusedelay = NFSCL_LEASE(clp->nfsc_renew) * 2;
 905                 else
 906                         clidinusedelay = 120;
 907                 trystalecnt = 3;
 908                 do {
 909                         error = nfsrpc_setclient(nmp, clp, 0, NULL, cred, p);
 910                         if (error == NFSERR_STALECLIENTID ||
 911                             error == NFSERR_STALEDONTRECOVER ||
 912                             error == NFSERR_BADSESSION ||
 913                             error == NFSERR_CLIDINUSE) {
 914                                 (void) nfs_catnap(PZERO, error, "nfs_setcl");
 915                         }
 916                 } while (((error == NFSERR_STALECLIENTID ||
 917                      error == NFSERR_BADSESSION ||
 918                      error == NFSERR_STALEDONTRECOVER) && --trystalecnt > 0) ||
 919                     (error == NFSERR_CLIDINUSE && --clidinusedelay > 0));
 920                 if (error) {
 921                         NFSLOCKCLSTATE();
 922                         nfsv4_unlock(&clp->nfsc_lock, 0);
 923                         NFSUNLOCKCLSTATE();
 924                         return (error);
 925                 }
 926                 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
 927         }
 928         if (igotlock) {
 929                 NFSLOCKCLSTATE();
 930                 nfsv4_unlock(&clp->nfsc_lock, 1);
 931                 NFSUNLOCKCLSTATE();
 932         }
 933
 934         *clpp = clp;
 935         return (0);
 936 }
 937
 938 /*
 939  * Get a reference to a clientid and return it, if valid.
 940  */
 941 struct nfsclclient *
 942 nfscl_findcl(struct nfsmount *nmp)
 943 {
 944         struct nfsclclient *clp;
 945
 946         clp = nmp->nm_clp;
 947         if (clp == NULL || !(clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID))
 948                 return (NULL);
 949         return (clp);
 950 }
 951
 952 /*
 953  * Release the clientid structure. It may be locked or reference counted.
 954  */
 955 static void
 956 nfscl_clrelease(struct nfsclclient *clp)
 957 {
 958
 959         if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK)
 960                 nfsv4_unlock(&clp->nfsc_lock, 0);
 961         else
 962                 nfsv4_relref(&clp->nfsc_lock);
 963 }
 964
 965 /*
 966  * External call for nfscl_clrelease.
 967  */
 968 void
 969 nfscl_clientrelease(struct nfsclclient *clp)
 970 {
 971
 972         NFSLOCKCLSTATE();
 973         if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK)
 974                 nfsv4_unlock(&clp->nfsc_lock, 0);
 975         else
 976                 nfsv4_relref(&clp->nfsc_lock);
 977         NFSUNLOCKCLSTATE();
 978 }
 979
 980 /*
 981  * Called when wanting to lock a byte region.
 982  */
 983 int
 984 nfscl_getbytelock(vnode_t vp, u_int64_t off, u_int64_t len,
 985     short type, struct ucred *cred, NFSPROC_T *p, struct nfsclclient *rclp,
 986     int recovery, void *id, int flags, u_int8_t *rownp, u_int8_t *ropenownp,
 987     struct nfscllockowner **lpp, int *newonep, int *donelocallyp)
 988 {
 989         struct nfscllockowner *lp;
 990         struct nfsclopen *op;
 991         struct nfsclclient *clp;
 992         struct nfscllockowner *nlp;
 993         struct nfscllock *nlop, *otherlop;
 994         struct nfscldeleg *dp = NULL, *ldp = NULL;
 995         struct nfscllockownerhead *lhp = NULL;
 996         struct nfsnode *np;
 997         u_int8_t own[NFSV4CL_LOCKNAMELEN], *ownp, openown[NFSV4CL_LOCKNAMELEN];
 998         u_int8_t *openownp;
 999         int error = 0, ret, donelocally = 0;
1000         u_int32_t mode;
1001
1002         /* For Lock Ops, the open mode doesn't matter, so use 0 to match any. */
1003         mode = 0;
1004         np = VTONFS(vp);
1005         *lpp = NULL;
1006         lp = NULL;
1007         *newonep = 0;
1008         *donelocallyp = 0;
1009
1010         /*
1011          * Might need these, so MALLOC them now, to
1012          * avoid a tsleep() in MALLOC later.
1013          */
1014         nlp = malloc(
1015             sizeof (struct nfscllockowner), M_NFSCLLOCKOWNER, M_WAITOK);
1016         otherlop = malloc(
1017             sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1018         nlop = malloc(
1019             sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1020         nlop->nfslo_type = type;
1021         nlop->nfslo_first = off;
1022         if (len == NFS64BITSSET) {
1023                 nlop->nfslo_end = NFS64BITSSET;
1024         } else {
1025                 nlop->nfslo_end = off + len;
1026                 if (nlop->nfslo_end <= nlop->nfslo_first)
1027                         error = NFSERR_INVAL;
1028         }
1029
1030         if (!error) {
1031                 if (recovery)
1032                         clp = rclp;
1033                 else
1034                         error = nfscl_getcl(vp->v_mount, cred, p, 1, &clp);
1035         }
1036         if (error) {
1037                 free(nlp, M_NFSCLLOCKOWNER);
1038                 free(otherlop, M_NFSCLLOCK);
1039                 free(nlop, M_NFSCLLOCK);
1040                 return (error);
1041         }
1042
1043         op = NULL;
1044         if (recovery) {
1045                 ownp = rownp;
1046                 openownp = ropenownp;
1047         } else {
1048                 nfscl_filllockowner(id, own, flags);
1049                 ownp = own;
1050                 if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount)))
1051                         nfscl_filllockowner(NULL, openown, F_POSIX);
1052                 else
1053                         nfscl_filllockowner(p->td_proc, openown, F_POSIX);
1054                 openownp = openown;
1055         }
1056         if (!recovery) {
1057                 NFSLOCKCLSTATE();
1058                 /*
1059                  * First, search for a delegation. If one exists for this file,
1060                  * the lock can be done locally against it, so long as there
1061                  * isn't a local lock conflict.
1062                  */
1063                 ldp = dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
1064                     np->n_fhp->nfh_len);
1065                 /* Just sanity check for correct type of delegation */
1066                 if (dp != NULL && ((dp->nfsdl_flags &
1067                     (NFSCLDL_RECALL | NFSCLDL_DELEGRET)) != 0 ||
1068                      (type == F_WRLCK &&
1069                       (dp->nfsdl_flags & NFSCLDL_WRITE) == 0)))
1070                         dp = NULL;
1071         }
1072         if (dp != NULL) {
1073                 /* Now, find an open and maybe a lockowner. */
1074                 ret = nfscl_getopen(&dp->nfsdl_owner, np->n_fhp->nfh_fh,
1075                     np->n_fhp->nfh_len, openownp, ownp, mode, NULL, &op);
1076                 if (ret)
1077                         ret = nfscl_getopen(&clp->nfsc_owner,
1078                             np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp,
1079                             ownp, mode, NULL, &op);
1080                 if (!ret) {
1081                         lhp = &dp->nfsdl_lock;
1082                         TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
1083                         TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list);
1084                         dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
1085                         donelocally = 1;
1086                 } else {
1087                         dp = NULL;
1088                 }
1089         }
1090         if (!donelocally) {
1091                 /*
1092                  * Get the related Open and maybe lockowner.
1093                  */
1094                 error = nfscl_getopen(&clp->nfsc_owner,
1095                     np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp,
1096                     ownp, mode, &lp, &op);
1097                 if (!error)
1098                         lhp = &op->nfso_lock;
1099         }
1100         if (!error && !recovery)
1101                 error = nfscl_localconflict(clp, np->n_fhp->nfh_fh,
1102                     np->n_fhp->nfh_len, nlop, ownp, ldp, NULL);
1103         if (error) {
1104                 if (!recovery) {
1105                         nfscl_clrelease(clp);
1106                         NFSUNLOCKCLSTATE();
1107                 }
1108                 free(nlp, M_NFSCLLOCKOWNER);
1109                 free(otherlop, M_NFSCLLOCK);
1110                 free(nlop, M_NFSCLLOCK);
1111                 return (error);
1112         }
1113
1114         /*
1115          * Ok, see if a lockowner exists and create one, as required.
1116          */
1117         if (lp == NULL)
1118                 LIST_FOREACH(lp, lhp, nfsl_list) {
1119                         if (!NFSBCMP(lp->nfsl_owner, ownp, NFSV4CL_LOCKNAMELEN))
1120                                 break;
1121                 }
1122         if (lp == NULL) {
1123                 NFSBCOPY(ownp, nlp->nfsl_owner, NFSV4CL_LOCKNAMELEN);
1124                 if (recovery)
1125                         NFSBCOPY(ropenownp, nlp->nfsl_openowner,
1126                             NFSV4CL_LOCKNAMELEN);
1127                 else
1128                         NFSBCOPY(op->nfso_own->nfsow_owner, nlp->nfsl_openowner,
1129                             NFSV4CL_LOCKNAMELEN);
1130                 nlp->nfsl_seqid = 0;
1131                 nlp->nfsl_lockflags = flags;
1132                 nlp->nfsl_inprog = NULL;
1133                 nfscl_lockinit(&nlp->nfsl_rwlock);
1134                 LIST_INIT(&nlp->nfsl_lock);
1135                 if (donelocally) {
1136                         nlp->nfsl_open = NULL;
1137                         nfsstatsv1.cllocallockowners++;
1138                 } else {
1139                         nlp->nfsl_open = op;
1140                         nfsstatsv1.cllockowners++;
1141                 }
1142                 LIST_INSERT_HEAD(lhp, nlp, nfsl_list);
1143                 lp = nlp;
1144                 nlp = NULL;
1145                 *newonep = 1;
1146         }
1147
1148         /*
1149          * Now, update the byte ranges for locks.
1150          */
1151         ret = nfscl_updatelock(lp, &nlop, &otherlop, donelocally);
1152         if (!ret)
1153                 donelocally = 1;
1154         if (donelocally) {
1155                 *donelocallyp = 1;
1156                 if (!recovery)
1157                         nfscl_clrelease(clp);
1158         } else {
1159                 /*
1160                  * Serial modifications on the lock owner for multiple threads
1161                  * for the same process using a read/write lock.
1162                  */
1163                 if (!recovery)
1164                         nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR);
1165         }
1166         if (!recovery)
1167                 NFSUNLOCKCLSTATE();
1168
1169         if (nlp)
1170                 free(nlp, M_NFSCLLOCKOWNER);
1171         if (nlop)
1172                 free(nlop, M_NFSCLLOCK);
1173         if (otherlop)
1174                 free(otherlop, M_NFSCLLOCK);
1175
1176         *lpp = lp;
1177         return (0);
1178 }
1179
1180 /*
1181  * Called to unlock a byte range, for LockU.
1182  */
1183 int
1184 nfscl_relbytelock(vnode_t vp, u_int64_t off, u_int64_t len,
1185     __unused struct ucred *cred, NFSPROC_T *p, int callcnt,
1186     struct nfsclclient *clp, void *id, int flags,
1187     struct nfscllockowner **lpp, int *dorpcp)
1188 {
1189         struct nfscllockowner *lp;
1190         struct nfsclowner *owp;
1191         struct nfsclopen *op;
1192         struct nfscllock *nlop, *other_lop = NULL;
1193         struct nfscldeleg *dp;
1194         struct nfsnode *np;
1195         u_int8_t own[NFSV4CL_LOCKNAMELEN];
1196         int ret = 0, fnd;
1197
1198         np = VTONFS(vp);
1199         *lpp = NULL;
1200         *dorpcp = 0;
1201
1202         /*
1203          * Might need these, so MALLOC them now, to
1204          * avoid a tsleep() in MALLOC later.
1205          */
1206         nlop = malloc(
1207             sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1208         nlop->nfslo_type = F_UNLCK;
1209         nlop->nfslo_first = off;
1210         if (len == NFS64BITSSET) {
1211                 nlop->nfslo_end = NFS64BITSSET;
1212         } else {
1213                 nlop->nfslo_end = off + len;
1214                 if (nlop->nfslo_end <= nlop->nfslo_first) {
1215                         free(nlop, M_NFSCLLOCK);
1216                         return (NFSERR_INVAL);
1217                 }
1218         }
1219         if (callcnt == 0) {
1220                 other_lop = malloc(
1221                     sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1222                 *other_lop = *nlop;
1223         }
1224         nfscl_filllockowner(id, own, flags);
1225         dp = NULL;
1226         NFSLOCKCLSTATE();
1227         if (callcnt == 0)
1228                 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
1229                     np->n_fhp->nfh_len);
1230
1231         /*
1232          * First, unlock any local regions on a delegation.
1233          */
1234         if (dp != NULL) {
1235                 /* Look for this lockowner. */
1236                 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
1237                         if (!NFSBCMP(lp->nfsl_owner, own,
1238                             NFSV4CL_LOCKNAMELEN))
1239                                 break;
1240                 }
1241                 if (lp != NULL)
1242                         /* Use other_lop, so nlop is still available */
1243                         (void)nfscl_updatelock(lp, &other_lop, NULL, 1);
1244         }
1245
1246         /*
1247          * Now, find a matching open/lockowner that hasn't already been done,
1248          * as marked by nfsl_inprog.
1249          */
1250         lp = NULL;
1251         fnd = 0;
1252         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
1253             LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1254                 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1255                     !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1256                     LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1257                         if (lp->nfsl_inprog == NULL &&
1258                             !NFSBCMP(lp->nfsl_owner, own,
1259                              NFSV4CL_LOCKNAMELEN)) {
1260                                 fnd = 1;
1261                                 break;
1262                         }
1263                     }
1264                     if (fnd)
1265                         break;
1266                 }
1267             }
1268             if (fnd)
1269                 break;
1270         }
1271
1272         if (lp != NULL) {
1273                 ret = nfscl_updatelock(lp, &nlop, NULL, 0);
1274                 if (ret)
1275                         *dorpcp = 1;
1276                 /*
1277                  * Serial modifications on the lock owner for multiple
1278                  * threads for the same process using a read/write lock.
1279                  */
1280                 lp->nfsl_inprog = p;
1281                 nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR);
1282                 *lpp = lp;
1283         }
1284         NFSUNLOCKCLSTATE();
1285         if (nlop)
1286                 free(nlop, M_NFSCLLOCK);
1287         if (other_lop)
1288                 free(other_lop, M_NFSCLLOCK);
1289         return (0);
1290 }
1291
1292 /*
1293  * Release all lockowners marked in progess for this process and file.
1294  */
1295 void
1296 nfscl_releasealllocks(struct nfsclclient *clp, vnode_t vp, NFSPROC_T *p,
1297     void *id, int flags)
1298 {
1299         struct nfsclowner *owp;
1300         struct nfsclopen *op;
1301         struct nfscllockowner *lp;
1302         struct nfsnode *np;
1303         u_int8_t own[NFSV4CL_LOCKNAMELEN];
1304
1305         np = VTONFS(vp);
1306         nfscl_filllockowner(id, own, flags);
1307         NFSLOCKCLSTATE();
1308         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
1309             LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1310                 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1311                     !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1312                     LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1313                         if (lp->nfsl_inprog == p &&
1314                             !NFSBCMP(lp->nfsl_owner, own,
1315                             NFSV4CL_LOCKNAMELEN)) {
1316                             lp->nfsl_inprog = NULL;
1317                             nfscl_lockunlock(&lp->nfsl_rwlock);
1318                         }
1319                     }
1320                 }
1321             }
1322         }
1323         nfscl_clrelease(clp);
1324         NFSUNLOCKCLSTATE();
1325 }
1326
1327 /*
1328  * Called to find out if any bytes within the byte range specified are
1329  * write locked by the calling process. Used to determine if flushing
1330  * is required before a LockU.
1331  * If in doubt, return 1, so the flush will occur.
1332  */
1333 int
1334 nfscl_checkwritelocked(vnode_t vp, struct flock *fl,
1335     struct ucred *cred, NFSPROC_T *p, void *id, int flags)
1336 {
1337         struct nfsclowner *owp;
1338         struct nfscllockowner *lp;
1339         struct nfsclopen *op;
1340         struct nfsclclient *clp;
1341         struct nfscllock *lop;
1342         struct nfscldeleg *dp;
1343         struct nfsnode *np;
1344         u_int64_t off, end;
1345         u_int8_t own[NFSV4CL_LOCKNAMELEN];
1346         int error = 0;
1347
1348         np = VTONFS(vp);
1349         switch (fl->l_whence) {
1350         case SEEK_SET:
1351         case SEEK_CUR:
1352                 /*
1353                  * Caller is responsible for adding any necessary offset
1354                  * when SEEK_CUR is used.
1355                  */
1356                 off = fl->l_start;
1357                 break;
1358         case SEEK_END:
1359                 off = np->n_size + fl->l_start;
1360                 break;
1361         default:
1362                 return (1);
1363         }
1364         if (fl->l_len != 0) {
1365                 end = off + fl->l_len;
1366                 if (end < off)
1367                         return (1);
1368         } else {
1369                 end = NFS64BITSSET;
1370         }
1371
1372         error = nfscl_getcl(vp->v_mount, cred, p, 1, &clp);
1373         if (error)
1374                 return (1);
1375         nfscl_filllockowner(id, own, flags);
1376         NFSLOCKCLSTATE();
1377
1378         /*
1379          * First check the delegation locks.
1380          */
1381         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
1382         if (dp != NULL) {
1383                 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
1384                         if (!NFSBCMP(lp->nfsl_owner, own,
1385                             NFSV4CL_LOCKNAMELEN))
1386                                 break;
1387                 }
1388                 if (lp != NULL) {
1389                         LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
1390                                 if (lop->nfslo_first >= end)
1391                                         break;
1392                                 if (lop->nfslo_end <= off)
1393                                         continue;
1394                                 if (lop->nfslo_type == F_WRLCK) {
1395                                         nfscl_clrelease(clp);
1396                                         NFSUNLOCKCLSTATE();
1397                                         return (1);
1398                                 }
1399                         }
1400                 }
1401         }
1402
1403         /*
1404          * Now, check state against the server.
1405          */
1406         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
1407             LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1408                 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1409                     !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1410                     LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1411                         if (!NFSBCMP(lp->nfsl_owner, own,
1412                             NFSV4CL_LOCKNAMELEN))
1413                             break;
1414                     }
1415                     if (lp != NULL) {
1416                         LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
1417                             if (lop->nfslo_first >= end)
1418                                 break;
1419                             if (lop->nfslo_end <= off)
1420                                 continue;
1421                             if (lop->nfslo_type == F_WRLCK) {
1422                                 nfscl_clrelease(clp);
1423                                 NFSUNLOCKCLSTATE();
1424                                 return (1);
1425                             }
1426                         }
1427                     }
1428                 }
1429             }
1430         }
1431         nfscl_clrelease(clp);
1432         NFSUNLOCKCLSTATE();
1433         return (0);
1434 }
1435
1436 /*
1437  * Release a byte range lock owner structure.
1438  */
1439 void
1440 nfscl_lockrelease(struct nfscllockowner *lp, int error, int candelete)
1441 {
1442         struct nfsclclient *clp;
1443
1444         if (lp == NULL)
1445                 return;
1446         NFSLOCKCLSTATE();
1447         clp = lp->nfsl_open->nfso_own->nfsow_clp;
1448         if (error != 0 && candelete &&
1449             (lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED) == 0)
1450                 nfscl_freelockowner(lp, 0);
1451         else
1452                 nfscl_lockunlock(&lp->nfsl_rwlock);
1453         nfscl_clrelease(clp);
1454         NFSUNLOCKCLSTATE();
1455 }
1456
1457 /*
1458  * Free up an open structure and any associated byte range lock structures.
1459  */
1460 void
1461 nfscl_freeopen(struct nfsclopen *op, int local)
1462 {
1463
1464         LIST_REMOVE(op, nfso_list);
1465         nfscl_freealllocks(&op->nfso_lock, local);
1466         free(op, M_NFSCLOPEN);
1467         if (local)
1468                 nfsstatsv1.cllocalopens--;
1469         else
1470                 nfsstatsv1.clopens--;
1471 }
1472
1473 /*
1474  * Free up all lock owners and associated locks.
1475  */
1476 static void
1477 nfscl_freealllocks(struct nfscllockownerhead *lhp, int local)
1478 {
1479         struct nfscllockowner *lp, *nlp;
1480
1481         LIST_FOREACH_SAFE(lp, lhp, nfsl_list, nlp) {
1482                 if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED))
1483                         panic("nfscllckw");
1484                 nfscl_freelockowner(lp, local);
1485         }
1486 }
1487
1488 /*
1489  * Called for an Open when NFSERR_EXPIRED is received from the server.
1490  * If there are no byte range locks nor a Share Deny lost, try to do a
1491  * fresh Open. Otherwise, free the open.
1492  */
1493 static int
1494 nfscl_expireopen(struct nfsclclient *clp, struct nfsclopen *op,
1495     struct nfsmount *nmp, struct ucred *cred, NFSPROC_T *p)
1496 {
1497         struct nfscllockowner *lp;
1498         struct nfscldeleg *dp;
1499         int mustdelete = 0, error;
1500
1501         /*
1502          * Look for any byte range lock(s).
1503          */
1504         LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1505                 if (!LIST_EMPTY(&lp->nfsl_lock)) {
1506                         mustdelete = 1;
1507                         break;
1508                 }
1509         }
1510
1511         /*
1512          * If no byte range lock(s) nor a Share deny, try to re-open.
1513          */
1514         if (!mustdelete && (op->nfso_mode & NFSLCK_DENYBITS) == 0) {
1515                 newnfs_copycred(&op->nfso_cred, cred);
1516                 dp = NULL;
1517                 error = nfsrpc_reopen(nmp, op->nfso_fh,
1518                     op->nfso_fhlen, op->nfso_mode, op, &dp, cred, p);
1519                 if (error) {
1520                         mustdelete = 1;
1521                         if (dp != NULL) {
1522                                 free(dp, M_NFSCLDELEG);
1523                                 dp = NULL;
1524                         }
1525                 }
1526                 if (dp != NULL)
1527                         nfscl_deleg(nmp->nm_mountp, clp, op->nfso_fh,
1528                             op->nfso_fhlen, cred, p, &dp);
1529         }
1530
1531         /*
1532          * If a byte range lock or Share deny or couldn't re-open, free it.
1533          */
1534         if (mustdelete)
1535                 nfscl_freeopen(op, 0);
1536         return (mustdelete);
1537 }
1538
1539 /*
1540  * Free up an open owner structure.
1541  */
1542 static void
1543 nfscl_freeopenowner(struct nfsclowner *owp, int local)
1544 {
1545
1546         LIST_REMOVE(owp, nfsow_list);
1547         free(owp, M_NFSCLOWNER);
1548         if (local)
1549                 nfsstatsv1.cllocalopenowners--;
1550         else
1551                 nfsstatsv1.clopenowners--;
1552 }
1553
1554 /*
1555  * Free up a byte range lock owner structure.
1556  */
1557 void
1558 nfscl_freelockowner(struct nfscllockowner *lp, int local)
1559 {
1560         struct nfscllock *lop, *nlop;
1561
1562         LIST_REMOVE(lp, nfsl_list);
1563         LIST_FOREACH_SAFE(lop, &lp->nfsl_lock, nfslo_list, nlop) {
1564                 nfscl_freelock(lop, local);
1565         }
1566         free(lp, M_NFSCLLOCKOWNER);
1567         if (local)
1568                 nfsstatsv1.cllocallockowners--;
1569         else
1570                 nfsstatsv1.cllockowners--;
1571 }
1572
1573 /*
1574  * Free up a byte range lock structure.
1575  */
1576 void
1577 nfscl_freelock(struct nfscllock *lop, int local)
1578 {
1579
1580         LIST_REMOVE(lop, nfslo_list);
1581         free(lop, M_NFSCLLOCK);
1582         if (local)
1583                 nfsstatsv1.cllocallocks--;
1584         else
1585                 nfsstatsv1.cllocks--;
1586 }
1587
1588 /*
1589  * Clean out the state related to a delegation.
1590  */
1591 static void
1592 nfscl_cleandeleg(struct nfscldeleg *dp)
1593 {
1594         struct nfsclowner *owp, *nowp;
1595         struct nfsclopen *op;
1596
1597         LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) {
1598                 op = LIST_FIRST(&owp->nfsow_open);
1599                 if (op != NULL) {
1600                         if (LIST_NEXT(op, nfso_list) != NULL)
1601                                 panic("nfscleandel");
1602                         nfscl_freeopen(op, 1);
1603                 }
1604                 nfscl_freeopenowner(owp, 1);
1605         }
1606         nfscl_freealllocks(&dp->nfsdl_lock, 1);
1607 }
1608
1609 /*
1610  * Free a delegation.
1611  */
1612 static void
1613 nfscl_freedeleg(struct nfscldeleghead *hdp, struct nfscldeleg *dp)
1614 {
1615
1616         TAILQ_REMOVE(hdp, dp, nfsdl_list);
1617         LIST_REMOVE(dp, nfsdl_hash);
1618         free(dp, M_NFSCLDELEG);
1619         nfsstatsv1.cldelegates--;
1620         nfscl_delegcnt--;
1621 }
1622
1623 /*
1624  * Free up all state related to this client structure.
1625  */
1626 static void
1627 nfscl_cleanclient(struct nfsclclient *clp)
1628 {
1629         struct nfsclowner *owp, *nowp;
1630         struct nfsclopen *op, *nop;
1631         struct nfscllayout *lyp, *nlyp;
1632         struct nfscldevinfo *dip, *ndip;
1633
1634         TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp)
1635                 nfscl_freelayout(lyp);
1636
1637         LIST_FOREACH_SAFE(dip, &clp->nfsc_devinfo, nfsdi_list, ndip)
1638                 nfscl_freedevinfo(dip);
1639
1640         /* Now, all the OpenOwners, etc. */
1641         LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1642                 LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
1643                         nfscl_freeopen(op, 0);
1644                 }
1645                 nfscl_freeopenowner(owp, 0);
1646         }
1647 }
1648
1649 /*
1650  * Called when an NFSERR_EXPIRED is received from the server.
1651  */
1652 static void
1653 nfscl_expireclient(struct nfsclclient *clp, struct nfsmount *nmp,
1654     struct ucred *cred, NFSPROC_T *p)
1655 {
1656         struct nfsclowner *owp, *nowp, *towp;
1657         struct nfsclopen *op, *nop, *top;
1658         struct nfscldeleg *dp, *ndp;
1659         int ret, printed = 0;
1660
1661         /*
1662          * First, merge locally issued Opens into the list for the server.
1663          */
1664         dp = TAILQ_FIRST(&clp->nfsc_deleg);
1665         while (dp != NULL) {
1666             ndp = TAILQ_NEXT(dp, nfsdl_list);
1667             owp = LIST_FIRST(&dp->nfsdl_owner);
1668             while (owp != NULL) {
1669                 nowp = LIST_NEXT(owp, nfsow_list);
1670                 op = LIST_FIRST(&owp->nfsow_open);
1671                 if (op != NULL) {
1672                     if (LIST_NEXT(op, nfso_list) != NULL)
1673                         panic("nfsclexp");
1674                     LIST_FOREACH(towp, &clp->nfsc_owner, nfsow_list) {
1675                         if (!NFSBCMP(towp->nfsow_owner, owp->nfsow_owner,
1676                             NFSV4CL_LOCKNAMELEN))
1677                             break;
1678                     }
1679                     if (towp != NULL) {
1680                         /* Merge opens in */
1681                         LIST_FOREACH(top, &towp->nfsow_open, nfso_list) {
1682                             if (top->nfso_fhlen == op->nfso_fhlen &&
1683                                 !NFSBCMP(top->nfso_fh, op->nfso_fh,
1684                                  op->nfso_fhlen)) {
1685                                 top->nfso_mode |= op->nfso_mode;
1686                                 top->nfso_opencnt += op->nfso_opencnt;
1687                                 break;
1688                             }
1689                         }
1690                         if (top == NULL) {
1691                             /* Just add the open to the owner list */
1692                             LIST_REMOVE(op, nfso_list);
1693                             op->nfso_own = towp;
1694                             LIST_INSERT_HEAD(&towp->nfsow_open, op, nfso_list);
1695                             nfsstatsv1.cllocalopens--;
1696                             nfsstatsv1.clopens++;
1697                         }
1698                     } else {
1699                         /* Just add the openowner to the client list */
1700                         LIST_REMOVE(owp, nfsow_list);
1701                         owp->nfsow_clp = clp;
1702                         LIST_INSERT_HEAD(&clp->nfsc_owner, owp, nfsow_list);
1703                         nfsstatsv1.cllocalopenowners--;
1704                         nfsstatsv1.clopenowners++;
1705                         nfsstatsv1.cllocalopens--;
1706                         nfsstatsv1.clopens++;
1707                     }
1708                 }
1709                 owp = nowp;
1710             }
1711             if (!printed && !LIST_EMPTY(&dp->nfsdl_lock)) {
1712                 printed = 1;
1713                 printf("nfsv4 expired locks lost\n");
1714             }
1715             nfscl_cleandeleg(dp);
1716             nfscl_freedeleg(&clp->nfsc_deleg, dp);
1717             dp = ndp;
1718         }
1719         if (!TAILQ_EMPTY(&clp->nfsc_deleg))
1720             panic("nfsclexp");
1721
1722         /*
1723          * Now, try and reopen against the server.
1724          */
1725         LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1726                 owp->nfsow_seqid = 0;
1727                 LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
1728                         ret = nfscl_expireopen(clp, op, nmp, cred, p);
1729                         if (ret && !printed) {
1730                                 printed = 1;
1731                                 printf("nfsv4 expired locks lost\n");
1732                         }
1733                 }
1734                 if (LIST_EMPTY(&owp->nfsow_open))
1735                         nfscl_freeopenowner(owp, 0);
1736         }
1737 }
1738
1739 /*
1740  * This function must be called after the process represented by "own" has
1741  * exited. Must be called with CLSTATE lock held.
1742  */
1743 static void
1744 nfscl_cleanup_common(struct nfsclclient *clp, u_int8_t *own)
1745 {
1746         struct nfsclowner *owp, *nowp;
1747         struct nfscllockowner *lp, *nlp;
1748         struct nfscldeleg *dp;
1749
1750         /* First, get rid of local locks on delegations. */
1751         TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
1752                 LIST_FOREACH_SAFE(lp, &dp->nfsdl_lock, nfsl_list, nlp) {
1753                     if (!NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
1754                         if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED))
1755                             panic("nfscllckw");
1756                         nfscl_freelockowner(lp, 1);
1757                     }
1758                 }
1759         }
1760         owp = LIST_FIRST(&clp->nfsc_owner);
1761         while (owp != NULL) {
1762                 nowp = LIST_NEXT(owp, nfsow_list);
1763                 if (!NFSBCMP(owp->nfsow_owner, own,
1764                     NFSV4CL_LOCKNAMELEN)) {
1765                         /*
1766                          * If there are children that haven't closed the
1767                          * file descriptors yet, the opens will still be
1768                          * here. For that case, let the renew thread clear
1769                          * out the OpenOwner later.
1770                          */
1771                         if (LIST_EMPTY(&owp->nfsow_open))
1772                                 nfscl_freeopenowner(owp, 0);
1773                         else
1774                                 owp->nfsow_defunct = 1;
1775                 }
1776                 owp = nowp;
1777         }
1778 }
1779
1780 /*
1781  * Find open/lock owners for processes that have exited.
1782  */
1783 static void
1784 nfscl_cleanupkext(struct nfsclclient *clp, struct nfscllockownerfhhead *lhp)
1785 {
1786         struct nfsclowner *owp, *nowp;
1787         struct nfsclopen *op;
1788         struct nfscllockowner *lp, *nlp;
1789         struct nfscldeleg *dp;
1790
1791         /*
1792          * All the pidhash locks must be acquired, since they are sx locks
1793          * and must be acquired before the mutexes.  The pid(s) that will
1794          * be used aren't known yet, so all the locks need to be acquired.
1795          * Fortunately, this function is only performed once/sec.
1796          */
1797         pidhash_slockall();
1798         NFSLOCKCLSTATE();
1799         LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1800                 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1801                         LIST_FOREACH_SAFE(lp, &op->nfso_lock, nfsl_list, nlp) {
1802                                 if (LIST_EMPTY(&lp->nfsl_lock))
1803                                         nfscl_emptylockowner(lp, lhp);
1804                         }
1805                 }
1806                 if (nfscl_procdoesntexist(owp->nfsow_owner))
1807                         nfscl_cleanup_common(clp, owp->nfsow_owner);
1808         }
1809
1810         /*
1811          * For the single open_owner case, these lock owners need to be
1812          * checked to see if they still exist separately.
1813          * This is because nfscl_procdoesntexist() never returns true for
1814          * the single open_owner so that the above doesn't ever call
1815          * nfscl_cleanup_common().
1816          */
1817         TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
1818                 LIST_FOREACH_SAFE(lp, &dp->nfsdl_lock, nfsl_list, nlp) {
1819                         if (nfscl_procdoesntexist(lp->nfsl_owner))
1820                                 nfscl_cleanup_common(clp, lp->nfsl_owner);
1821                 }
1822         }
1823         NFSUNLOCKCLSTATE();
1824         pidhash_sunlockall();
1825 }
1826
1827 /*
1828  * Take the empty lock owner and move it to the local lhp list if the
1829  * associated process no longer exists.
1830  */
1831 static void
1832 nfscl_emptylockowner(struct nfscllockowner *lp,
1833     struct nfscllockownerfhhead *lhp)
1834 {
1835         struct nfscllockownerfh *lfhp, *mylfhp;
1836         struct nfscllockowner *nlp;
1837         int fnd_it;
1838
1839         /* If not a Posix lock owner, just return. */
1840         if ((lp->nfsl_lockflags & F_POSIX) == 0)
1841                 return;
1842
1843         fnd_it = 0;
1844         mylfhp = NULL;
1845         /*
1846          * First, search to see if this lock owner is already in the list.
1847          * If it is, then the associated process no longer exists.
1848          */
1849         SLIST_FOREACH(lfhp, lhp, nfslfh_list) {
1850                 if (lfhp->nfslfh_len == lp->nfsl_open->nfso_fhlen &&
1851                     !NFSBCMP(lfhp->nfslfh_fh, lp->nfsl_open->nfso_fh,
1852                     lfhp->nfslfh_len))
1853                         mylfhp = lfhp;
1854                 LIST_FOREACH(nlp, &lfhp->nfslfh_lock, nfsl_list)
1855                         if (!NFSBCMP(nlp->nfsl_owner, lp->nfsl_owner,
1856                             NFSV4CL_LOCKNAMELEN))
1857                                 fnd_it = 1;
1858         }
1859         /* If not found, check if process still exists. */
1860         if (fnd_it == 0 && nfscl_procdoesntexist(lp->nfsl_owner) == 0)
1861                 return;
1862
1863         /* Move the lock owner over to the local list. */
1864         if (mylfhp == NULL) {
1865                 mylfhp = malloc(sizeof(struct nfscllockownerfh), M_TEMP,
1866                     M_NOWAIT);
1867                 if (mylfhp == NULL)
1868                         return;
1869                 mylfhp->nfslfh_len = lp->nfsl_open->nfso_fhlen;
1870                 NFSBCOPY(lp->nfsl_open->nfso_fh, mylfhp->nfslfh_fh,
1871                     mylfhp->nfslfh_len);
1872                 LIST_INIT(&mylfhp->nfslfh_lock);
1873                 SLIST_INSERT_HEAD(lhp, mylfhp, nfslfh_list);
1874         }
1875         LIST_REMOVE(lp, nfsl_list);
1876         LIST_INSERT_HEAD(&mylfhp->nfslfh_lock, lp, nfsl_list);
1877 }
1878
1879 static int      fake_global;    /* Used to force visibility of MNTK_UNMOUNTF */
1880 /*
1881  * Called from nfs umount to free up the clientid.
1882  */
1883 void
1884 nfscl_umount(struct nfsmount *nmp, NFSPROC_T *p)
1885 {
1886         struct nfsclclient *clp;
1887         struct ucred *cred;
1888         int igotlock;
1889
1890         /*
1891          * For the case that matters, this is the thread that set
1892          * MNTK_UNMOUNTF, so it will see it set. The code that follows is
1893          * done to ensure that any thread executing nfscl_getcl() after
1894          * this time, will see MNTK_UNMOUNTF set. nfscl_getcl() uses the
1895          * mutex for NFSLOCKCLSTATE(), so it is "m" for the following
1896          * explanation, courtesy of Alan Cox.
1897          * What follows is a snippet from Alan Cox's email at:
1898          * https://docs.FreeBSD.org/cgi/mid.cgi?BANLkTikR3d65zPHo9==08ZfJ2vmqZucEvw
1899          *
1900          * 1. Set MNTK_UNMOUNTF
1901          * 2. Acquire a standard FreeBSD mutex "m".
1902          * 3. Update some data structures.
1903          * 4. Release mutex "m".
1904          *
1905          * Then, other threads that acquire "m" after step 4 has occurred will
1906          * see MNTK_UNMOUNTF as set.  But, other threads that beat thread X to
1907          * step 2 may or may not see MNTK_UNMOUNTF as set.
1908          */
1909         NFSLOCKCLSTATE();
1910         if ((nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) {
1911                 fake_global++;
1912                 NFSUNLOCKCLSTATE();
1913                 NFSLOCKCLSTATE();
1914         }
1915
1916         clp = nmp->nm_clp;
1917         if (clp != NULL) {
1918                 if ((clp->nfsc_flags & NFSCLFLAGS_INITED) == 0)
1919                         panic("nfscl umount");
1920
1921                 /*
1922                  * First, handshake with the nfscl renew thread, to terminate
1923                  * it.
1924                  */
1925                 clp->nfsc_flags |= NFSCLFLAGS_UMOUNT;
1926                 while (clp->nfsc_flags & NFSCLFLAGS_HASTHREAD)
1927                         (void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT,
1928                             "nfsclumnt", hz);
1929
1930                 /*
1931                  * Now, get the exclusive lock on the client state, so
1932                  * that no uses of the state are still in progress.
1933                  */
1934                 do {
1935                         igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
1936                             NFSCLSTATEMUTEXPTR, NULL);
1937                 } while (!igotlock);
1938                 NFSUNLOCKCLSTATE();
1939
1940                 /*
1941                  * Free up all the state. It will expire on the server, but
1942                  * maybe we should do a SetClientId/SetClientIdConfirm so
1943                  * the server throws it away?
1944                  */
1945                 LIST_REMOVE(clp, nfsc_list);
1946                 nfscl_delegreturnall(clp, p);
1947                 cred = newnfs_getcred();
1948                 if (NFSHASNFSV4N(nmp)) {
1949                         (void)nfsrpc_destroysession(nmp, clp, cred, p);
1950                         (void)nfsrpc_destroyclient(nmp, clp, cred, p);
1951                 } else
1952                         (void)nfsrpc_setclient(nmp, clp, 0, NULL, cred, p);
1953                 nfscl_cleanclient(clp);
1954                 nmp->nm_clp = NULL;
1955                 NFSFREECRED(cred);
1956                 free(clp, M_NFSCLCLIENT);
1957         } else
1958                 NFSUNLOCKCLSTATE();
1959 }
1960
1961 /*
1962  * This function is called when a server replies with NFSERR_STALECLIENTID
1963  * NFSERR_STALESTATEID or NFSERR_BADSESSION. It traverses the clientid lists,
1964  * doing Opens and Locks with reclaim. If these fail, it deletes the
1965  * corresponding state.
1966  */
1967 static void
1968 nfscl_recover(struct nfsclclient *clp, bool *retokp, struct ucred *cred,
1969     NFSPROC_T *p)
1970 {
1971         struct nfsclowner *owp, *nowp;
1972         struct nfsclopen *op, *nop;
1973         struct nfscllockowner *lp, *nlp;
1974         struct nfscllock *lop, *nlop;
1975         struct nfscldeleg *dp, *ndp, *tdp;
1976         struct nfsmount *nmp;
1977         struct ucred *tcred;
1978         struct nfsclopenhead extra_open;
1979         struct nfscldeleghead extra_deleg;
1980         struct nfsreq *rep;
1981         u_int64_t len;
1982         u_int32_t delegtype = NFSV4OPEN_DELEGATEWRITE, mode;
1983         int i, igotlock = 0, error, trycnt, firstlock;
1984         struct nfscllayout *lyp, *nlyp;
1985
1986         /*
1987          * First, lock the client structure, so everyone else will
1988          * block when trying to use state.
1989          */
1990         NFSLOCKCLSTATE();
1991         clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG;
1992         do {
1993                 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
1994                     NFSCLSTATEMUTEXPTR, NULL);
1995         } while (!igotlock);
1996         NFSUNLOCKCLSTATE();
1997
1998         nmp = clp->nfsc_nmp;
1999         if (nmp == NULL)
2000                 panic("nfscl recover");
2001
2002         /*
2003          * For now, just get rid of all layouts. There may be a need
2004          * to do LayoutCommit Ops with reclaim == true later.
2005          */
2006         TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp)
2007                 nfscl_freelayout(lyp);
2008         TAILQ_INIT(&clp->nfsc_layout);
2009         for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++)
2010                 LIST_INIT(&clp->nfsc_layouthash[i]);
2011
2012         trycnt = 5;
2013         tcred = NULL;
2014         do {
2015                 error = nfsrpc_setclient(nmp, clp, 1, retokp, cred, p);
2016         } while ((error == NFSERR_STALECLIENTID ||
2017              error == NFSERR_BADSESSION ||
2018              error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
2019         if (error) {
2020                 NFSLOCKCLSTATE();
2021                 clp->nfsc_flags &= ~(NFSCLFLAGS_RECOVER |
2022                     NFSCLFLAGS_RECVRINPROG);
2023                 wakeup(&clp->nfsc_flags);
2024                 nfsv4_unlock(&clp->nfsc_lock, 0);
2025                 NFSUNLOCKCLSTATE();
2026                 return;
2027         }
2028         clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
2029         clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2030
2031         /*
2032          * Mark requests already queued on the server, so that they don't
2033          * initiate another recovery cycle. Any requests already in the
2034          * queue that handle state information will have the old stale
2035          * clientid/stateid and will get a NFSERR_STALESTATEID,
2036          * NFSERR_STALECLIENTID or NFSERR_BADSESSION reply from the server.
2037          * This will be translated to NFSERR_STALEDONTRECOVER when
2038          * R_DONTRECOVER is set.
2039          */
2040         NFSLOCKREQ();
2041         TAILQ_FOREACH(rep, &nfsd_reqq, r_chain) {
2042                 if (rep->r_nmp == nmp)
2043                         rep->r_flags |= R_DONTRECOVER;
2044         }
2045         NFSUNLOCKREQ();
2046
2047         /*
2048          * If nfsrpc_setclient() returns *retokp == true,
2049          * no more recovery is needed.
2050          */
2051         if (*retokp)
2052                 goto out;
2053
2054         /*
2055          * Now, mark all delegations "need reclaim".
2056          */
2057         TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list)
2058                 dp->nfsdl_flags |= NFSCLDL_NEEDRECLAIM;
2059
2060         TAILQ_INIT(&extra_deleg);
2061         LIST_INIT(&extra_open);
2062         /*
2063          * Now traverse the state lists, doing Open and Lock Reclaims.
2064          */
2065         tcred = newnfs_getcred();
2066         owp = LIST_FIRST(&clp->nfsc_owner);
2067         while (owp != NULL) {
2068             nowp = LIST_NEXT(owp, nfsow_list);
2069             owp->nfsow_seqid = 0;
2070             op = LIST_FIRST(&owp->nfsow_open);
2071             while (op != NULL) {
2072                 nop = LIST_NEXT(op, nfso_list);
2073                 if (error != NFSERR_NOGRACE && error != NFSERR_BADSESSION) {
2074                     /* Search for a delegation to reclaim with the open */
2075                     TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2076                         if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
2077                             continue;
2078                         if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
2079                             mode = NFSV4OPEN_ACCESSWRITE;
2080                             delegtype = NFSV4OPEN_DELEGATEWRITE;
2081                         } else {
2082                             mode = NFSV4OPEN_ACCESSREAD;
2083                             delegtype = NFSV4OPEN_DELEGATEREAD;
2084                         }
2085                         if ((op->nfso_mode & mode) == mode &&
2086                             op->nfso_fhlen == dp->nfsdl_fhlen &&
2087                             !NFSBCMP(op->nfso_fh, dp->nfsdl_fh, op->nfso_fhlen))
2088                             break;
2089                     }
2090                     ndp = dp;
2091                     if (dp == NULL)
2092                         delegtype = NFSV4OPEN_DELEGATENONE;
2093                     newnfs_copycred(&op->nfso_cred, tcred);
2094                     error = nfscl_tryopen(nmp, NULL, op->nfso_fh,
2095                         op->nfso_fhlen, op->nfso_fh, op->nfso_fhlen,
2096                         op->nfso_mode, op, NULL, 0, &ndp, 1, delegtype,
2097                         tcred, p);
2098                     if (!error) {
2099                         /* Handle any replied delegation */
2100                         if (ndp != NULL && ((ndp->nfsdl_flags & NFSCLDL_WRITE)
2101                             || NFSMNT_RDONLY(nmp->nm_mountp))) {
2102                             if ((ndp->nfsdl_flags & NFSCLDL_WRITE))
2103                                 mode = NFSV4OPEN_ACCESSWRITE;
2104                             else
2105                                 mode = NFSV4OPEN_ACCESSREAD;
2106                             TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2107                                 if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
2108                                     continue;
2109                                 if ((op->nfso_mode & mode) == mode &&
2110                                     op->nfso_fhlen == dp->nfsdl_fhlen &&
2111                                     !NFSBCMP(op->nfso_fh, dp->nfsdl_fh,
2112                                     op->nfso_fhlen)) {
2113                                     dp->nfsdl_stateid = ndp->nfsdl_stateid;
2114                                     dp->nfsdl_sizelimit = ndp->nfsdl_sizelimit;
2115                                     dp->nfsdl_ace = ndp->nfsdl_ace;
2116                                     dp->nfsdl_change = ndp->nfsdl_change;
2117                                     dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
2118                                     if ((ndp->nfsdl_flags & NFSCLDL_RECALL))
2119                                         dp->nfsdl_flags |= NFSCLDL_RECALL;
2120                                     free(ndp, M_NFSCLDELEG);
2121                                     ndp = NULL;
2122                                     break;
2123                                 }
2124                             }
2125                         }
2126                         if (ndp != NULL)
2127                             TAILQ_INSERT_HEAD(&extra_deleg, ndp, nfsdl_list);
2128
2129                         /* and reclaim all byte range locks */
2130                         lp = LIST_FIRST(&op->nfso_lock);
2131                         while (lp != NULL) {
2132                             nlp = LIST_NEXT(lp, nfsl_list);
2133                             lp->nfsl_seqid = 0;
2134                             firstlock = 1;
2135                             lop = LIST_FIRST(&lp->nfsl_lock);
2136                             while (lop != NULL) {
2137                                 nlop = LIST_NEXT(lop, nfslo_list);
2138                                 if (lop->nfslo_end == NFS64BITSSET)
2139                                     len = NFS64BITSSET;
2140                                 else
2141                                     len = lop->nfslo_end - lop->nfslo_first;
2142                                 error = nfscl_trylock(nmp, NULL,
2143                                     op->nfso_fh, op->nfso_fhlen, lp,
2144                                     firstlock, 1, lop->nfslo_first, len,
2145                                     lop->nfslo_type, tcred, p);
2146                                 if (error != 0)
2147                                     nfscl_freelock(lop, 0);
2148                                 else
2149                                     firstlock = 0;
2150                                 lop = nlop;
2151                             }
2152                             /* If no locks, but a lockowner, just delete it. */
2153                             if (LIST_EMPTY(&lp->nfsl_lock))
2154                                 nfscl_freelockowner(lp, 0);
2155                             lp = nlp;
2156                         }
2157                     }
2158                 }
2159                 if (error != 0 && error != NFSERR_BADSESSION)
2160                     nfscl_freeopen(op, 0);
2161                 op = nop;
2162             }
2163             owp = nowp;
2164         }
2165
2166         /*
2167          * Now, try and get any delegations not yet reclaimed by cobbling
2168          * to-gether an appropriate open.
2169          */
2170         nowp = NULL;
2171         dp = TAILQ_FIRST(&clp->nfsc_deleg);
2172         while (dp != NULL) {
2173             ndp = TAILQ_NEXT(dp, nfsdl_list);
2174             if ((dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM)) {
2175                 if (nowp == NULL) {
2176                     nowp = malloc(
2177                         sizeof (struct nfsclowner), M_NFSCLOWNER, M_WAITOK);
2178                     /*
2179                      * Name must be as long an largest possible
2180                      * NFSV4CL_LOCKNAMELEN. 12 for now.
2181                      */
2182                     NFSBCOPY("RECLAIMDELEG", nowp->nfsow_owner,
2183                         NFSV4CL_LOCKNAMELEN);
2184                     LIST_INIT(&nowp->nfsow_open);
2185                     nowp->nfsow_clp = clp;
2186                     nowp->nfsow_seqid = 0;
2187                     nowp->nfsow_defunct = 0;
2188                     nfscl_lockinit(&nowp->nfsow_rwlock);
2189                 }
2190                 nop = NULL;
2191                 if (error != NFSERR_NOGRACE && error != NFSERR_BADSESSION) {
2192                     nop = malloc(sizeof (struct nfsclopen) +
2193                         dp->nfsdl_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
2194                     nop->nfso_own = nowp;
2195                     if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
2196                         nop->nfso_mode = NFSV4OPEN_ACCESSWRITE;
2197                         delegtype = NFSV4OPEN_DELEGATEWRITE;
2198                     } else {
2199                         nop->nfso_mode = NFSV4OPEN_ACCESSREAD;
2200                         delegtype = NFSV4OPEN_DELEGATEREAD;
2201                     }
2202                     nop->nfso_opencnt = 0;
2203                     nop->nfso_posixlock = 1;
2204                     nop->nfso_fhlen = dp->nfsdl_fhlen;
2205                     NFSBCOPY(dp->nfsdl_fh, nop->nfso_fh, dp->nfsdl_fhlen);
2206                     LIST_INIT(&nop->nfso_lock);
2207                     nop->nfso_stateid.seqid = 0;
2208                     nop->nfso_stateid.other[0] = 0;
2209                     nop->nfso_stateid.other[1] = 0;
2210                     nop->nfso_stateid.other[2] = 0;
2211                     newnfs_copycred(&dp->nfsdl_cred, tcred);
2212                     newnfs_copyincred(tcred, &nop->nfso_cred);
2213                     tdp = NULL;
2214                     error = nfscl_tryopen(nmp, NULL, nop->nfso_fh,
2215                         nop->nfso_fhlen, nop->nfso_fh, nop->nfso_fhlen,
2216                         nop->nfso_mode, nop, NULL, 0, &tdp, 1,
2217                         delegtype, tcred, p);
2218                     if (tdp != NULL) {
2219                         if ((tdp->nfsdl_flags & NFSCLDL_WRITE))
2220                             mode = NFSV4OPEN_ACCESSWRITE;
2221                         else
2222                             mode = NFSV4OPEN_ACCESSREAD;
2223                         if ((nop->nfso_mode & mode) == mode &&
2224                             nop->nfso_fhlen == tdp->nfsdl_fhlen &&
2225                             !NFSBCMP(nop->nfso_fh, tdp->nfsdl_fh,
2226                             nop->nfso_fhlen)) {
2227                             dp->nfsdl_stateid = tdp->nfsdl_stateid;
2228                             dp->nfsdl_sizelimit = tdp->nfsdl_sizelimit;
2229                             dp->nfsdl_ace = tdp->nfsdl_ace;
2230                             dp->nfsdl_change = tdp->nfsdl_change;
2231                             dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
2232                             if ((tdp->nfsdl_flags & NFSCLDL_RECALL))
2233                                 dp->nfsdl_flags |= NFSCLDL_RECALL;
2234                             free(tdp, M_NFSCLDELEG);
2235                         } else {
2236                             TAILQ_INSERT_HEAD(&extra_deleg, tdp, nfsdl_list);
2237                         }
2238                     }
2239                 }
2240                 if (error) {
2241                     if (nop != NULL)
2242                         free(nop, M_NFSCLOPEN);
2243                     /*
2244                      * Couldn't reclaim it, so throw the state
2245                      * away. Ouch!!
2246                      */
2247                     nfscl_cleandeleg(dp);
2248                     nfscl_freedeleg(&clp->nfsc_deleg, dp);
2249                 } else {
2250                     LIST_INSERT_HEAD(&extra_open, nop, nfso_list);
2251                 }
2252             }
2253             dp = ndp;
2254         }
2255
2256         /*
2257          * Now, get rid of extra Opens and Delegations.
2258          */
2259         LIST_FOREACH_SAFE(op, &extra_open, nfso_list, nop) {
2260                 do {
2261                         newnfs_copycred(&op->nfso_cred, tcred);
2262                         error = nfscl_tryclose(op, tcred, nmp, p);
2263                         if (error == NFSERR_GRACE)
2264                                 (void) nfs_catnap(PZERO, error, "nfsexcls");
2265                 } while (error == NFSERR_GRACE);
2266                 LIST_REMOVE(op, nfso_list);
2267                 free(op, M_NFSCLOPEN);
2268         }
2269         if (nowp != NULL)
2270                 free(nowp, M_NFSCLOWNER);
2271
2272         TAILQ_FOREACH_SAFE(dp, &extra_deleg, nfsdl_list, ndp) {
2273                 do {
2274                         newnfs_copycred(&dp->nfsdl_cred, tcred);
2275                         error = nfscl_trydelegreturn(dp, tcred, nmp, p);
2276                         if (error == NFSERR_GRACE)
2277                                 (void) nfs_catnap(PZERO, error, "nfsexdlg");
2278                 } while (error == NFSERR_GRACE);
2279                 TAILQ_REMOVE(&extra_deleg, dp, nfsdl_list);
2280                 free(dp, M_NFSCLDELEG);
2281         }
2282
2283         /* For NFSv4.1 or later, do a RECLAIM_COMPLETE. */
2284         if (NFSHASNFSV4N(nmp))
2285                 (void)nfsrpc_reclaimcomplete(nmp, cred, p);
2286
2287 out:
2288         NFSLOCKCLSTATE();
2289         clp->nfsc_flags &= ~NFSCLFLAGS_RECVRINPROG;
2290         wakeup(&clp->nfsc_flags);
2291         nfsv4_unlock(&clp->nfsc_lock, 0);
2292         NFSUNLOCKCLSTATE();
2293         if (tcred != NULL)
2294                 NFSFREECRED(tcred);
2295 }
2296
2297 /*
2298  * This function is called when a server replies with NFSERR_EXPIRED.
2299  * It deletes all state for the client and does a fresh SetClientId/confirm.
2300  * XXX Someday it should post a signal to the process(es) that hold the
2301  * state, so they know that lock state has been lost.
2302  */
2303 int
2304 nfscl_hasexpired(struct nfsclclient *clp, u_int32_t clidrev, NFSPROC_T *p)
2305 {
2306         struct nfsmount *nmp;
2307         struct ucred *cred;
2308         int igotlock = 0, error, trycnt;
2309
2310         /*
2311          * If the clientid has gone away or a new SetClientid has already
2312          * been done, just return ok.
2313          */
2314         if (clp == NULL || clidrev != clp->nfsc_clientidrev)
2315                 return (0);
2316
2317         /*
2318          * First, lock the client structure, so everyone else will
2319          * block when trying to use state. Also, use NFSCLFLAGS_EXPIREIT so
2320          * that only one thread does the work.
2321          */
2322         NFSLOCKCLSTATE();
2323         clp->nfsc_flags |= NFSCLFLAGS_EXPIREIT;
2324         do {
2325                 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
2326                     NFSCLSTATEMUTEXPTR, NULL);
2327         } while (!igotlock && (clp->nfsc_flags & NFSCLFLAGS_EXPIREIT));
2328         if ((clp->nfsc_flags & NFSCLFLAGS_EXPIREIT) == 0) {
2329                 if (igotlock)
2330                         nfsv4_unlock(&clp->nfsc_lock, 0);
2331                 NFSUNLOCKCLSTATE();
2332                 return (0);
2333         }
2334         clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG;
2335         NFSUNLOCKCLSTATE();
2336
2337         nmp = clp->nfsc_nmp;
2338         if (nmp == NULL)
2339                 panic("nfscl expired");
2340         cred = newnfs_getcred();
2341         trycnt = 5;
2342         do {
2343                 error = nfsrpc_setclient(nmp, clp, 0, NULL, cred, p);
2344         } while ((error == NFSERR_STALECLIENTID ||
2345              error == NFSERR_BADSESSION ||
2346              error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
2347         if (error) {
2348                 NFSLOCKCLSTATE();
2349                 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2350         } else {
2351                 /*
2352                  * Expire the state for the client.
2353                  */
2354                 nfscl_expireclient(clp, nmp, cred, p);
2355                 NFSLOCKCLSTATE();
2356                 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
2357                 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2358         }
2359         clp->nfsc_flags &= ~(NFSCLFLAGS_EXPIREIT | NFSCLFLAGS_RECVRINPROG);
2360         wakeup(&clp->nfsc_flags);
2361         nfsv4_unlock(&clp->nfsc_lock, 0);
2362         NFSUNLOCKCLSTATE();
2363         NFSFREECRED(cred);
2364         return (error);
2365 }
2366
2367 /*
2368  * This function inserts a lock in the list after insert_lop.
2369  */
2370 static void
2371 nfscl_insertlock(struct nfscllockowner *lp, struct nfscllock *new_lop,
2372     struct nfscllock *insert_lop, int local)
2373 {
2374
2375         if ((struct nfscllockowner *)insert_lop == lp)
2376                 LIST_INSERT_HEAD(&lp->nfsl_lock, new_lop, nfslo_list);
2377         else
2378                 LIST_INSERT_AFTER(insert_lop, new_lop, nfslo_list);
2379         if (local)
2380                 nfsstatsv1.cllocallocks++;
2381         else
2382                 nfsstatsv1.cllocks++;
2383 }
2384
2385 /*
2386  * This function updates the locking for a lock owner and given file. It
2387  * maintains a list of lock ranges ordered on increasing file offset that
2388  * are NFSCLLOCK_READ or NFSCLLOCK_WRITE and non-overlapping (aka POSIX style).
2389  * It always adds new_lop to the list and sometimes uses the one pointed
2390  * at by other_lopp.
2391  * Returns 1 if the locks were modified, 0 otherwise.
2392  */
2393 static int
2394 nfscl_updatelock(struct nfscllockowner *lp, struct nfscllock **new_lopp,
2395     struct nfscllock **other_lopp, int local)
2396 {
2397         struct nfscllock *new_lop = *new_lopp;
2398         struct nfscllock *lop, *tlop, *ilop;
2399         struct nfscllock *other_lop;
2400         int unlock = 0, modified = 0;
2401         u_int64_t tmp;
2402
2403         /*
2404          * Work down the list until the lock is merged.
2405          */
2406         if (new_lop->nfslo_type == F_UNLCK)
2407                 unlock = 1;
2408         ilop = (struct nfscllock *)lp;
2409         lop = LIST_FIRST(&lp->nfsl_lock);
2410         while (lop != NULL) {
2411             /*
2412              * Only check locks for this file that aren't before the start of
2413              * new lock's range.
2414              */
2415             if (lop->nfslo_end >= new_lop->nfslo_first) {
2416                 if (new_lop->nfslo_end < lop->nfslo_first) {
2417                     /*
2418                      * If the new lock ends before the start of the
2419                      * current lock's range, no merge, just insert
2420                      * the new lock.
2421                      */
2422                     break;
2423                 }
2424                 if (new_lop->nfslo_type == lop->nfslo_type ||
2425                     (new_lop->nfslo_first <= lop->nfslo_first &&
2426                      new_lop->nfslo_end >= lop->nfslo_end)) {
2427                     /*
2428                      * This lock can be absorbed by the new lock/unlock.
2429                      * This happens when it covers the entire range
2430                      * of the old lock or is contiguous
2431                      * with the old lock and is of the same type or an
2432                      * unlock.
2433                      */
2434                     if (new_lop->nfslo_type != lop->nfslo_type ||
2435                         new_lop->nfslo_first != lop->nfslo_first ||
2436                         new_lop->nfslo_end != lop->nfslo_end)
2437                         modified = 1;
2438                     if (lop->nfslo_first < new_lop->nfslo_first)
2439                         new_lop->nfslo_first = lop->nfslo_first;
2440                     if (lop->nfslo_end > new_lop->nfslo_end)
2441                         new_lop->nfslo_end = lop->nfslo_end;
2442                     tlop = lop;
2443                     lop = LIST_NEXT(lop, nfslo_list);
2444                     nfscl_freelock(tlop, local);
2445                     continue;
2446                 }
2447
2448                 /*
2449                  * All these cases are for contiguous locks that are not the
2450                  * same type, so they can't be merged.
2451                  */
2452                 if (new_lop->nfslo_first <= lop->nfslo_first) {
2453                     /*
2454                      * This case is where the new lock overlaps with the
2455                      * first part of the old lock. Move the start of the
2456                      * old lock to just past the end of the new lock. The
2457                      * new lock will be inserted in front of the old, since
2458                      * ilop hasn't been updated. (We are done now.)
2459                      */
2460                     if (lop->nfslo_first != new_lop->nfslo_end) {
2461                         lop->nfslo_first = new_lop->nfslo_end;
2462                         modified = 1;
2463                     }
2464                     break;
2465                 }
2466                 if (new_lop->nfslo_end >= lop->nfslo_end) {
2467                     /*
2468                      * This case is where the new lock overlaps with the
2469                      * end of the old lock's range. Move the old lock's
2470                      * end to just before the new lock's first and insert
2471                      * the new lock after the old lock.
2472                      * Might not be done yet, since the new lock could
2473                      * overlap further locks with higher ranges.
2474                      */
2475                     if (lop->nfslo_end != new_lop->nfslo_first) {
2476                         lop->nfslo_end = new_lop->nfslo_first;
2477                         modified = 1;
2478                     }
2479                     ilop = lop;
2480                     lop = LIST_NEXT(lop, nfslo_list);
2481                     continue;
2482                 }
2483                 /*
2484                  * The final case is where the new lock's range is in the
2485                  * middle of the current lock's and splits the current lock
2486                  * up. Use *other_lopp to handle the second part of the
2487                  * split old lock range. (We are done now.)
2488                  * For unlock, we use new_lop as other_lop and tmp, since
2489                  * other_lop and new_lop are the same for this case.
2490                  * We noted the unlock case above, so we don't need
2491                  * new_lop->nfslo_type any longer.
2492                  */
2493                 tmp = new_lop->nfslo_first;
2494                 if (unlock) {
2495                     other_lop = new_lop;
2496                     *new_lopp = NULL;
2497                 } else {
2498                     other_lop = *other_lopp;
2499                     *other_lopp = NULL;
2500                 }
2501                 other_lop->nfslo_first = new_lop->nfslo_end;
2502                 other_lop->nfslo_end = lop->nfslo_end;
2503                 other_lop->nfslo_type = lop->nfslo_type;
2504                 lop->nfslo_end = tmp;
2505                 nfscl_insertlock(lp, other_lop, lop, local);
2506                 ilop = lop;
2507                 modified = 1;
2508                 break;
2509             }
2510             ilop = lop;
2511             lop = LIST_NEXT(lop, nfslo_list);
2512             if (lop == NULL)
2513                 break;
2514         }
2515
2516         /*
2517          * Insert the new lock in the list at the appropriate place.
2518          */
2519         if (!unlock) {
2520                 nfscl_insertlock(lp, new_lop, ilop, local);
2521                 *new_lopp = NULL;
2522                 modified = 1;
2523         }
2524         return (modified);
2525 }
2526
2527 /*
2528  * This function must be run as a kernel thread.
2529  * It does Renew Ops and recovery, when required.
2530  */
2531 void
2532 nfscl_renewthread(struct nfsclclient *clp, NFSPROC_T *p)
2533 {
2534         struct nfsclowner *owp, *nowp;
2535         struct nfsclopen *op;
2536         struct nfscllockowner *lp, *nlp;
2537         struct nfscldeleghead dh;
2538         struct nfscldeleg *dp, *ndp;
2539         struct ucred *cred;
2540         u_int32_t clidrev;
2541         int error, cbpathdown, islept, igotlock, ret, clearok;
2542         uint32_t recover_done_time = 0;
2543         time_t mytime;
2544         static time_t prevsec = 0;
2545         struct nfscllockownerfh *lfhp, *nlfhp;
2546         struct nfscllockownerfhhead lfh;
2547         struct nfscllayout *lyp, *nlyp;
2548         struct nfscldevinfo *dip, *ndip;
2549         struct nfscllayouthead rlh;
2550         struct nfsclrecalllayout *recallp;
2551         struct nfsclds *dsp;
2552         bool retok;
2553
2554         cred = newnfs_getcred();
2555         NFSLOCKCLSTATE();
2556         clp->nfsc_flags |= NFSCLFLAGS_HASTHREAD;
2557         NFSUNLOCKCLSTATE();
2558         for(;;) {
2559                 newnfs_setroot(cred);
2560                 cbpathdown = 0;
2561                 if (clp->nfsc_flags & NFSCLFLAGS_RECOVER) {
2562                         /*
2563                          * Only allow one full recover within 1/2 of the lease
2564                          * duration (nfsc_renew).
2565                          * retok is value/result.  If passed in set to true,
2566                          * it indicates only a CreateSession operation should
2567                          * be attempted.
2568                          * If it is returned true, it indicates that the
2569                          * recovery only required a CreateSession.
2570                          */
2571                         retok = true;
2572                         if (recover_done_time < NFSD_MONOSEC) {
2573                                 recover_done_time = NFSD_MONOSEC +
2574                                     clp->nfsc_renew;
2575                                 retok = false;
2576                         }
2577                         NFSCL_DEBUG(1, "Doing recovery, only "
2578                             "createsession=%d\n", retok);
2579                         nfscl_recover(clp, &retok, cred, p);
2580                 }
2581                 if (clp->nfsc_expire <= NFSD_MONOSEC &&
2582                     (clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID)) {
2583                         clp->nfsc_expire = NFSD_MONOSEC + clp->nfsc_renew;
2584                         clidrev = clp->nfsc_clientidrev;
2585                         error = nfsrpc_renew(clp, NULL, cred, p);
2586                         if (error == NFSERR_CBPATHDOWN)
2587                             cbpathdown = 1;
2588                         else if (error == NFSERR_STALECLIENTID ||
2589                             error == NFSERR_BADSESSION) {
2590                             NFSLOCKCLSTATE();
2591                             clp->nfsc_flags |= NFSCLFLAGS_RECOVER;
2592                             NFSUNLOCKCLSTATE();
2593                         } else if (error == NFSERR_EXPIRED)
2594                             (void) nfscl_hasexpired(clp, clidrev, p);
2595                 }
2596
2597 checkdsrenew:
2598                 if (NFSHASNFSV4N(clp->nfsc_nmp)) {
2599                         /* Do renews for any DS sessions. */
2600                         NFSLOCKMNT(clp->nfsc_nmp);
2601                         /* Skip first entry, since the MDS is handled above. */
2602                         dsp = TAILQ_FIRST(&clp->nfsc_nmp->nm_sess);
2603                         if (dsp != NULL)
2604                                 dsp = TAILQ_NEXT(dsp, nfsclds_list);
2605                         while (dsp != NULL) {
2606                                 if (dsp->nfsclds_expire <= NFSD_MONOSEC &&
2607                                     dsp->nfsclds_sess.nfsess_defunct == 0) {
2608                                         dsp->nfsclds_expire = NFSD_MONOSEC +
2609                                             clp->nfsc_renew;
2610                                         NFSUNLOCKMNT(clp->nfsc_nmp);
2611                                         (void)nfsrpc_renew(clp, dsp, cred, p);
2612                                         goto checkdsrenew;
2613                                 }
2614                                 dsp = TAILQ_NEXT(dsp, nfsclds_list);
2615                         }
2616                         NFSUNLOCKMNT(clp->nfsc_nmp);
2617                 }
2618
2619                 TAILQ_INIT(&dh);
2620                 NFSLOCKCLSTATE();
2621                 if (cbpathdown)
2622                         /* It's a Total Recall! */
2623                         nfscl_totalrecall(clp);
2624
2625                 /*
2626                  * Now, handle defunct owners.
2627                  */
2628                 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
2629                         if (LIST_EMPTY(&owp->nfsow_open)) {
2630                                 if (owp->nfsow_defunct != 0)
2631                                         nfscl_freeopenowner(owp, 0);
2632                         }
2633                 }
2634
2635                 /*
2636                  * Do the recall on any delegations. To avoid trouble, always
2637                  * come back up here after having slept.
2638                  */
2639                 igotlock = 0;
2640 tryagain:
2641                 dp = TAILQ_FIRST(&clp->nfsc_deleg);
2642                 while (dp != NULL) {
2643                         ndp = TAILQ_NEXT(dp, nfsdl_list);
2644                         if ((dp->nfsdl_flags & NFSCLDL_RECALL)) {
2645                                 /*
2646                                  * Wait for outstanding I/O ops to be done.
2647                                  */
2648                                 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
2649                                     if (igotlock) {
2650                                         nfsv4_unlock(&clp->nfsc_lock, 0);
2651                                         igotlock = 0;
2652                                     }
2653                                     dp->nfsdl_rwlock.nfslock_lock |=
2654                                         NFSV4LOCK_WANTED;
2655                                     (void) nfsmsleep(&dp->nfsdl_rwlock,
2656                                         NFSCLSTATEMUTEXPTR, PZERO, "nfscld",
2657                                         NULL);
2658                                     goto tryagain;
2659                                 }
2660                                 while (!igotlock) {
2661                                     igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
2662                                         &islept, NFSCLSTATEMUTEXPTR, NULL);
2663                                     if (islept)
2664                                         goto tryagain;
2665                                 }
2666                                 NFSUNLOCKCLSTATE();
2667                                 newnfs_copycred(&dp->nfsdl_cred, cred);
2668                                 ret = nfscl_recalldeleg(clp, clp->nfsc_nmp, dp,
2669                                     NULL, cred, p, 1);
2670                                 if (!ret) {
2671                                     nfscl_cleandeleg(dp);
2672                                     TAILQ_REMOVE(&clp->nfsc_deleg, dp,
2673                                         nfsdl_list);
2674                                     LIST_REMOVE(dp, nfsdl_hash);
2675                                     TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
2676                                     nfscl_delegcnt--;
2677                                     nfsstatsv1.cldelegates--;
2678                                 }
2679                                 NFSLOCKCLSTATE();
2680                         }
2681                         dp = ndp;
2682                 }
2683
2684                 /*
2685                  * Clear out old delegations, if we are above the high water
2686                  * mark. Only clear out ones with no state related to them.
2687                  * The tailq list is in LRU order.
2688                  */
2689                 dp = TAILQ_LAST(&clp->nfsc_deleg, nfscldeleghead);
2690                 while (nfscl_delegcnt > nfscl_deleghighwater && dp != NULL) {
2691                     ndp = TAILQ_PREV(dp, nfscldeleghead, nfsdl_list);
2692                     if (dp->nfsdl_rwlock.nfslock_usecnt == 0 &&
2693                         dp->nfsdl_rwlock.nfslock_lock == 0 &&
2694                         dp->nfsdl_timestamp < NFSD_MONOSEC &&
2695                         (dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_ZAPPED |
2696                           NFSCLDL_NEEDRECLAIM | NFSCLDL_DELEGRET)) == 0) {
2697                         clearok = 1;
2698                         LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
2699                             op = LIST_FIRST(&owp->nfsow_open);
2700                             if (op != NULL) {
2701                                 clearok = 0;
2702                                 break;
2703                             }
2704                         }
2705                         if (clearok) {
2706                             LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
2707                                 if (!LIST_EMPTY(&lp->nfsl_lock)) {
2708                                     clearok = 0;
2709                                     break;
2710                                 }
2711                             }
2712                         }
2713                         if (clearok) {
2714                             TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
2715                             LIST_REMOVE(dp, nfsdl_hash);
2716                             TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
2717                             nfscl_delegcnt--;
2718                             nfsstatsv1.cldelegates--;
2719                         }
2720                     }
2721                     dp = ndp;
2722                 }
2723                 if (igotlock)
2724                         nfsv4_unlock(&clp->nfsc_lock, 0);
2725
2726                 /*
2727                  * Do the recall on any layouts. To avoid trouble, always
2728                  * come back up here after having slept.
2729                  */
2730                 TAILQ_INIT(&rlh);
2731 tryagain2:
2732                 TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp) {
2733                         if ((lyp->nfsly_flags & NFSLY_RECALL) != 0) {
2734                                 /*
2735                                  * Wait for outstanding I/O ops to be done.
2736                                  */
2737                                 if (lyp->nfsly_lock.nfslock_usecnt > 0 ||
2738                                     (lyp->nfsly_lock.nfslock_lock &
2739                                      NFSV4LOCK_LOCK) != 0) {
2740                                         lyp->nfsly_lock.nfslock_lock |=
2741                                             NFSV4LOCK_WANTED;
2742                                         nfsmsleep(&lyp->nfsly_lock.nfslock_lock,
2743                                             NFSCLSTATEMUTEXPTR, PZERO, "nfslyp",
2744                                             NULL);
2745                                         goto tryagain2;
2746                                 }
2747                                 /* Move the layout to the recall list. */
2748                                 TAILQ_REMOVE(&clp->nfsc_layout, lyp,
2749                                     nfsly_list);
2750                                 LIST_REMOVE(lyp, nfsly_hash);
2751                                 TAILQ_INSERT_HEAD(&rlh, lyp, nfsly_list);
2752
2753                                 /* Handle any layout commits. */
2754                                 if (!NFSHASNOLAYOUTCOMMIT(clp->nfsc_nmp) &&
2755                                     (lyp->nfsly_flags & NFSLY_WRITTEN) != 0) {
2756                                         lyp->nfsly_flags &= ~NFSLY_WRITTEN;
2757                                         NFSUNLOCKCLSTATE();
2758                                         NFSCL_DEBUG(3, "do layoutcommit\n");
2759                                         nfscl_dolayoutcommit(clp->nfsc_nmp, lyp,
2760                                             cred, p);
2761                                         NFSLOCKCLSTATE();
2762                                         goto tryagain2;
2763                                 }
2764                         }
2765                 }
2766
2767                 /* Now, look for stale layouts. */
2768                 lyp = TAILQ_LAST(&clp->nfsc_layout, nfscllayouthead);
2769                 while (lyp != NULL) {
2770                         nlyp = TAILQ_PREV(lyp, nfscllayouthead, nfsly_list);
2771                         if (lyp->nfsly_timestamp < NFSD_MONOSEC &&
2772                             (lyp->nfsly_flags & NFSLY_RECALL) == 0 &&
2773                             lyp->nfsly_lock.nfslock_usecnt == 0 &&
2774                             lyp->nfsly_lock.nfslock_lock == 0) {
2775                                 NFSCL_DEBUG(4, "ret stale lay=%d\n",
2776                                     nfscl_layoutcnt);
2777                                 recallp = malloc(sizeof(*recallp),
2778                                     M_NFSLAYRECALL, M_NOWAIT);
2779                                 if (recallp == NULL)
2780                                         break;
2781                                 (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE,
2782                                     lyp, NFSLAYOUTIOMODE_ANY, 0, UINT64_MAX,
2783                                     lyp->nfsly_stateid.seqid, 0, 0, NULL,
2784                                     recallp);
2785                         }
2786                         lyp = nlyp;
2787                 }
2788
2789                 /*
2790                  * Free up any unreferenced device info structures.
2791                  */
2792                 LIST_FOREACH_SAFE(dip, &clp->nfsc_devinfo, nfsdi_list, ndip) {
2793                         if (dip->nfsdi_layoutrefs == 0 &&
2794                             dip->nfsdi_refcnt == 0) {
2795                                 NFSCL_DEBUG(4, "freeing devinfo\n");
2796                                 LIST_REMOVE(dip, nfsdi_list);
2797                                 nfscl_freedevinfo(dip);
2798                         }
2799                 }
2800                 NFSUNLOCKCLSTATE();
2801
2802                 /* Do layout return(s), as required. */
2803                 TAILQ_FOREACH_SAFE(lyp, &rlh, nfsly_list, nlyp) {
2804                         TAILQ_REMOVE(&rlh, lyp, nfsly_list);
2805                         NFSCL_DEBUG(4, "ret layout\n");
2806                         nfscl_layoutreturn(clp->nfsc_nmp, lyp, cred, p);
2807                         nfscl_freelayout(lyp);
2808                 }
2809
2810                 /*
2811                  * Delegreturn any delegations cleaned out or recalled.
2812                  */
2813                 TAILQ_FOREACH_SAFE(dp, &dh, nfsdl_list, ndp) {
2814                         newnfs_copycred(&dp->nfsdl_cred, cred);
2815                         (void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
2816                         TAILQ_REMOVE(&dh, dp, nfsdl_list);
2817                         free(dp, M_NFSCLDELEG);
2818                 }
2819
2820                 SLIST_INIT(&lfh);
2821                 /*
2822                  * Call nfscl_cleanupkext() once per second to check for
2823                  * open/lock owners where the process has exited.
2824                  */
2825                 mytime = NFSD_MONOSEC;
2826                 if (prevsec != mytime) {
2827                         prevsec = mytime;
2828                         nfscl_cleanupkext(clp, &lfh);
2829                 }
2830
2831                 /*
2832                  * Do a ReleaseLockOwner for all lock owners where the
2833                  * associated process no longer exists, as found by
2834                  * nfscl_cleanupkext().
2835                  */
2836                 newnfs_setroot(cred);
2837                 SLIST_FOREACH_SAFE(lfhp, &lfh, nfslfh_list, nlfhp) {
2838                         LIST_FOREACH_SAFE(lp, &lfhp->nfslfh_lock, nfsl_list,
2839                             nlp) {
2840                                 (void)nfsrpc_rellockown(clp->nfsc_nmp, lp,
2841                                     lfhp->nfslfh_fh, lfhp->nfslfh_len, cred,
2842                                     p);
2843                                 nfscl_freelockowner(lp, 0);
2844                         }
2845                         free(lfhp, M_TEMP);
2846                 }
2847                 SLIST_INIT(&lfh);
2848
2849                 NFSLOCKCLSTATE();
2850                 if ((clp->nfsc_flags & NFSCLFLAGS_RECOVER) == 0)
2851                         (void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT, "nfscl",
2852                             hz);
2853                 if (clp->nfsc_flags & NFSCLFLAGS_UMOUNT) {
2854                         clp->nfsc_flags &= ~NFSCLFLAGS_HASTHREAD;
2855                         NFSUNLOCKCLSTATE();
2856                         NFSFREECRED(cred);
2857                         wakeup((caddr_t)clp);
2858                         return;
2859                 }
2860                 NFSUNLOCKCLSTATE();
2861         }
2862 }
2863
2864 /*
2865  * Initiate state recovery. Called when NFSERR_STALECLIENTID,
2866  * NFSERR_STALESTATEID or NFSERR_BADSESSION is received.
2867  */
2868 void
2869 nfscl_initiate_recovery(struct nfsclclient *clp)
2870 {
2871
2872         if (clp == NULL)
2873                 return;
2874         NFSLOCKCLSTATE();
2875         clp->nfsc_flags |= NFSCLFLAGS_RECOVER;
2876         NFSUNLOCKCLSTATE();
2877         wakeup((caddr_t)clp);
2878 }
2879
2880 /*
2881  * Dump out the state stuff for debugging.
2882  */
2883 void
2884 nfscl_dumpstate(struct nfsmount *nmp, int openowner, int opens,
2885     int lockowner, int locks)
2886 {
2887         struct nfsclclient *clp;
2888         struct nfsclowner *owp;
2889         struct nfsclopen *op;
2890         struct nfscllockowner *lp;
2891         struct nfscllock *lop;
2892         struct nfscldeleg *dp;
2893
2894         clp = nmp->nm_clp;
2895         if (clp == NULL) {
2896                 printf("nfscl dumpstate NULL clp\n");
2897                 return;
2898         }
2899         NFSLOCKCLSTATE();
2900         TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2901           LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
2902             if (openowner && !LIST_EMPTY(&owp->nfsow_open))
2903                 printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
2904                     owp->nfsow_owner[0], owp->nfsow_owner[1],
2905                     owp->nfsow_owner[2], owp->nfsow_owner[3],
2906                     owp->nfsow_seqid);
2907             LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
2908                 if (opens)
2909                     printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
2910                         op->nfso_stateid.other[0], op->nfso_stateid.other[1],
2911                         op->nfso_stateid.other[2], op->nfso_opencnt,
2912                         op->nfso_fh[12]);
2913                 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
2914                     if (lockowner)
2915                         printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
2916                             lp->nfsl_owner[0], lp->nfsl_owner[1],
2917                             lp->nfsl_owner[2], lp->nfsl_owner[3],
2918                             lp->nfsl_seqid,
2919                             lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
2920                             lp->nfsl_stateid.other[2]);
2921                     LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
2922                         if (locks)
2923 #ifdef __FreeBSD__
2924                             printf("lck typ=%d fst=%ju end=%ju\n",
2925                                 lop->nfslo_type, (intmax_t)lop->nfslo_first,
2926                                 (intmax_t)lop->nfslo_end);
2927 #else
2928                             printf("lck typ=%d fst=%qd end=%qd\n",
2929                                 lop->nfslo_type, lop->nfslo_first,
2930                                 lop->nfslo_end);
2931 #endif
2932                     }
2933                 }
2934             }
2935           }
2936         }
2937         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
2938             if (openowner && !LIST_EMPTY(&owp->nfsow_open))
2939                 printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
2940                     owp->nfsow_owner[0], owp->nfsow_owner[1],
2941                     owp->nfsow_owner[2], owp->nfsow_owner[3],
2942                     owp->nfsow_seqid);
2943             LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
2944                 if (opens)
2945                     printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
2946                         op->nfso_stateid.other[0], op->nfso_stateid.other[1],
2947                         op->nfso_stateid.other[2], op->nfso_opencnt,
2948                         op->nfso_fh[12]);
2949                 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
2950                     if (lockowner)
2951                         printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
2952                             lp->nfsl_owner[0], lp->nfsl_owner[1],
2953                             lp->nfsl_owner[2], lp->nfsl_owner[3],
2954                             lp->nfsl_seqid,
2955                             lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
2956                             lp->nfsl_stateid.other[2]);
2957                     LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
2958                         if (locks)
2959 #ifdef __FreeBSD__
2960                             printf("lck typ=%d fst=%ju end=%ju\n",
2961                                 lop->nfslo_type, (intmax_t)lop->nfslo_first,
2962                                 (intmax_t)lop->nfslo_end);
2963 #else
2964                             printf("lck typ=%d fst=%qd end=%qd\n",
2965                                 lop->nfslo_type, lop->nfslo_first,
2966                                 lop->nfslo_end);
2967 #endif
2968                     }
2969                 }
2970             }
2971         }
2972         NFSUNLOCKCLSTATE();
2973 }
2974
2975 /*
2976  * Check for duplicate open owners and opens.
2977  * (Only used as a diagnostic aid.)
2978  */
2979 void
2980 nfscl_dupopen(vnode_t vp, int dupopens)
2981 {
2982         struct nfsclclient *clp;
2983         struct nfsclowner *owp, *owp2;
2984         struct nfsclopen *op, *op2;
2985         struct nfsfh *nfhp;
2986
2987         clp = VFSTONFS(vp->v_mount)->nm_clp;
2988         if (clp == NULL) {
2989                 printf("nfscl dupopen NULL clp\n");
2990                 return;
2991         }
2992         nfhp = VTONFS(vp)->n_fhp;
2993         NFSLOCKCLSTATE();
2994
2995         /*
2996          * First, search for duplicate owners.
2997          * These should never happen!
2998          */
2999         LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3000             LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3001                 if (owp != owp2 &&
3002                     !NFSBCMP(owp->nfsow_owner, owp2->nfsow_owner,
3003                     NFSV4CL_LOCKNAMELEN)) {
3004                         NFSUNLOCKCLSTATE();
3005                         printf("DUP OWNER\n");
3006                         nfscl_dumpstate(VFSTONFS(vp->v_mount), 1, 1, 0, 0);
3007                         return;
3008                 }
3009             }
3010         }
3011
3012         /*
3013          * Now, search for duplicate stateids.
3014          * These shouldn't happen, either.
3015          */
3016         LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3017             LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
3018                 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3019                     LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3020                         if (op != op2 &&
3021                             (op->nfso_stateid.other[0] != 0 ||
3022                              op->nfso_stateid.other[1] != 0 ||
3023                              op->nfso_stateid.other[2] != 0) &&
3024                             op->nfso_stateid.other[0] == op2->nfso_stateid.other[0] &&
3025                             op->nfso_stateid.other[1] == op2->nfso_stateid.other[1] &&
3026                             op->nfso_stateid.other[2] == op2->nfso_stateid.other[2]) {
3027                             NFSUNLOCKCLSTATE();
3028                             printf("DUP STATEID\n");
3029                             nfscl_dumpstate(VFSTONFS(vp->v_mount), 1, 1, 0, 0);
3030                             return;
3031                         }
3032                     }
3033                 }
3034             }
3035         }
3036
3037         /*
3038          * Now search for duplicate opens.
3039          * Duplicate opens for the same owner
3040          * should never occur. Other duplicates are
3041          * possible and are checked for if "dupopens"
3042          * is true.
3043          */
3044         LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3045             LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
3046                 if (nfhp->nfh_len == op2->nfso_fhlen &&
3047                     !NFSBCMP(nfhp->nfh_fh, op2->nfso_fh, nfhp->nfh_len)) {
3048                     LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3049                         LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3050                             if (op != op2 && nfhp->nfh_len == op->nfso_fhlen &&
3051                                 !NFSBCMP(nfhp->nfh_fh, op->nfso_fh, nfhp->nfh_len) &&
3052                                 (!NFSBCMP(op->nfso_own->nfsow_owner,
3053                                  op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN) ||
3054                                  dupopens)) {
3055                                 if (!NFSBCMP(op->nfso_own->nfsow_owner,
3056                                     op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
3057                                     NFSUNLOCKCLSTATE();
3058                                     printf("BADDUP OPEN\n");
3059                                 } else {
3060                                     NFSUNLOCKCLSTATE();
3061                                     printf("DUP OPEN\n");
3062                                 }
3063                                 nfscl_dumpstate(VFSTONFS(vp->v_mount), 1, 1, 0,
3064                                     0);
3065                                 return;
3066                             }
3067                         }
3068                     }
3069                 }
3070             }
3071         }
3072         NFSUNLOCKCLSTATE();
3073 }
3074
3075 /*
3076  * During close, find an open that needs to be dereferenced and
3077  * dereference it. If there are no more opens for this file,
3078  * log a message to that effect.
3079  * Opens aren't actually Close'd until VOP_INACTIVE() is performed
3080  * on the file's vnode.
3081  * This is the safe way, since it is difficult to identify
3082  * which open the close is for and I/O can be performed after the
3083  * close(2) system call when a file is mmap'd.
3084  * If it returns 0 for success, there will be a referenced
3085  * clp returned via clpp.
3086  */
3087 int
3088 nfscl_getclose(vnode_t vp, struct nfsclclient **clpp)
3089 {
3090         struct nfsclclient *clp;
3091         struct nfsclowner *owp;
3092         struct nfsclopen *op;
3093         struct nfscldeleg *dp;
3094         struct nfsfh *nfhp;
3095         int error, notdecr;
3096
3097         error = nfscl_getcl(vp->v_mount, NULL, NULL, 1, &clp);
3098         if (error)
3099                 return (error);
3100         *clpp = clp;
3101
3102         nfhp = VTONFS(vp)->n_fhp;
3103         notdecr = 1;
3104         NFSLOCKCLSTATE();
3105         /*
3106          * First, look for one under a delegation that was locally issued
3107          * and just decrement the opencnt for it. Since all my Opens against
3108          * the server are DENY_NONE, I don't see a problem with hanging
3109          * onto them. (It is much easier to use one of the extant Opens
3110          * that I already have on the server when a Delegation is recalled
3111          * than to do fresh Opens.) Someday, I might need to rethink this, but.
3112          */
3113         dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
3114         if (dp != NULL) {
3115                 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
3116                         op = LIST_FIRST(&owp->nfsow_open);
3117                         if (op != NULL) {
3118                                 /*
3119                                  * Since a delegation is for a file, there
3120                                  * should never be more than one open for
3121                                  * each openowner.
3122                                  */
3123                                 if (LIST_NEXT(op, nfso_list) != NULL)
3124                                         panic("nfscdeleg opens");
3125                                 if (notdecr && op->nfso_opencnt > 0) {
3126                                         notdecr = 0;
3127                                         op->nfso_opencnt--;
3128                                         break;
3129                                 }
3130                         }
3131                 }
3132         }
3133
3134         /* Now process the opens against the server. */
3135         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3136                 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3137                         if (op->nfso_fhlen == nfhp->nfh_len &&
3138                             !NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
3139                             nfhp->nfh_len)) {
3140                                 /* Found an open, decrement cnt if possible */
3141                                 if (notdecr && op->nfso_opencnt > 0) {
3142                                         notdecr = 0;
3143                                         op->nfso_opencnt--;
3144                                 }
3145                                 /*
3146                                  * There are more opens, so just return.
3147                                  */
3148                                 if (op->nfso_opencnt > 0) {
3149                                         NFSUNLOCKCLSTATE();
3150                                         return (0);
3151                                 }
3152                         }
3153                 }
3154         }
3155         NFSUNLOCKCLSTATE();
3156         if (notdecr)
3157                 printf("nfscl: never fnd open\n");
3158         return (0);
3159 }
3160
3161 int
3162 nfscl_doclose(vnode_t vp, struct nfsclclient **clpp, NFSPROC_T *p)
3163 {
3164         struct nfsclclient *clp;
3165         struct nfsclowner *owp, *nowp;
3166         struct nfsclopen *op;
3167         struct nfscldeleg *dp;
3168         struct nfsfh *nfhp;
3169         struct nfsclrecalllayout *recallp;
3170         int error;
3171
3172         error = nfscl_getcl(vp->v_mount, NULL, NULL, 1, &clp);
3173         if (error)
3174                 return (error);
3175         *clpp = clp;
3176
3177         nfhp = VTONFS(vp)->n_fhp;
3178         recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_WAITOK);
3179         NFSLOCKCLSTATE();
3180         /*
3181          * First get rid of the local Open structures, which should be no
3182          * longer in use.
3183          */
3184         dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
3185         if (dp != NULL) {
3186                 LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) {
3187                         op = LIST_FIRST(&owp->nfsow_open);
3188                         if (op != NULL) {
3189                                 KASSERT((op->nfso_opencnt == 0),
3190                                     ("nfscl: bad open cnt on deleg"));
3191                                 nfscl_freeopen(op, 1);
3192                         }
3193                         nfscl_freeopenowner(owp, 1);
3194                 }
3195         }
3196
3197         /* Return any layouts marked return on close. */
3198         nfscl_retoncloselayout(vp, clp, nfhp->nfh_fh, nfhp->nfh_len, &recallp);
3199
3200         /* Now process the opens against the server. */
3201 lookformore:
3202         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3203                 op = LIST_FIRST(&owp->nfsow_open);
3204                 while (op != NULL) {
3205                         if (op->nfso_fhlen == nfhp->nfh_len &&
3206                             !NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
3207                             nfhp->nfh_len)) {
3208                                 /* Found an open, close it. */
3209 #ifdef DIAGNOSTIC
3210                                 KASSERT((op->nfso_opencnt == 0),
3211                                     ("nfscl: bad open cnt on server (%d)",
3212                                      op->nfso_opencnt));
3213 #endif
3214                                 NFSUNLOCKCLSTATE();
3215                                 nfsrpc_doclose(VFSTONFS(vp->v_mount), op, p);
3216                                 NFSLOCKCLSTATE();
3217                                 goto lookformore;
3218                         }
3219                         op = LIST_NEXT(op, nfso_list);
3220                 }
3221         }
3222         NFSUNLOCKCLSTATE();
3223         /*
3224          * recallp has been set NULL by nfscl_retoncloselayout() if it was
3225          * used by the function, but calling free() with a NULL pointer is ok.
3226          */
3227         free(recallp, M_NFSLAYRECALL);
3228         return (0);
3229 }
3230
3231 /*
3232  * Return all delegations on this client.
3233  * (Must be called with client sleep lock.)
3234  */
3235 static void
3236 nfscl_delegreturnall(struct nfsclclient *clp, NFSPROC_T *p)
3237 {
3238         struct nfscldeleg *dp, *ndp;
3239         struct ucred *cred;
3240
3241         cred = newnfs_getcred();
3242         TAILQ_FOREACH_SAFE(dp, &clp->nfsc_deleg, nfsdl_list, ndp) {
3243                 nfscl_cleandeleg(dp);
3244                 (void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
3245                 nfscl_freedeleg(&clp->nfsc_deleg, dp);
3246         }
3247         NFSFREECRED(cred);
3248 }
3249
3250 /*
3251  * Do a callback RPC.
3252  */
3253 void
3254 nfscl_docb(struct nfsrv_descript *nd, NFSPROC_T *p)
3255 {
3256         int clist, gotseq_ok, i, j, k, op, rcalls;
3257         u_int32_t *tl;
3258         struct nfsclclient *clp;
3259         struct nfscldeleg *dp = NULL;
3260         int numops, taglen = -1, error = 0, trunc __unused;
3261         u_int32_t minorvers = 0, retops = 0, *retopsp = NULL, *repp, cbident;
3262         u_char tag[NFSV4_SMALLSTR + 1], *tagstr;
3263         vnode_t vp = NULL;
3264         struct nfsnode *np;
3265         struct vattr va;
3266         struct nfsfh *nfhp;
3267         mount_t mp;
3268         nfsattrbit_t attrbits, rattrbits;
3269         nfsv4stateid_t stateid;
3270         uint32_t seqid, slotid = 0, highslot, cachethis __unused;
3271         uint8_t sessionid[NFSX_V4SESSIONID];
3272         struct mbuf *rep;
3273         struct nfscllayout *lyp;
3274         uint64_t filesid[2], len, off;
3275         int changed, gotone, laytype, recalltype;
3276         uint32_t iomode;
3277         struct nfsclrecalllayout *recallp = NULL;
3278         struct nfsclsession *tsep;
3279
3280         gotseq_ok = 0;
3281         nfsrvd_rephead(nd);
3282         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3283         taglen = fxdr_unsigned(int, *tl);
3284         if (taglen < 0) {
3285                 error = EBADRPC;
3286                 goto nfsmout;
3287         }
3288         if (taglen <= NFSV4_SMALLSTR)
3289                 tagstr = tag;
3290         else
3291                 tagstr = malloc(taglen + 1, M_TEMP, M_WAITOK);
3292         error = nfsrv_mtostr(nd, tagstr, taglen);
3293         if (error) {
3294                 if (taglen > NFSV4_SMALLSTR)
3295                         free(tagstr, M_TEMP);
3296                 taglen = -1;
3297                 goto nfsmout;
3298         }
3299         (void) nfsm_strtom(nd, tag, taglen);
3300         if (taglen > NFSV4_SMALLSTR) {
3301                 free(tagstr, M_TEMP);
3302         }
3303         NFSM_BUILD(retopsp, u_int32_t *, NFSX_UNSIGNED);
3304         NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3305         minorvers = fxdr_unsigned(u_int32_t, *tl++);
3306         if (minorvers != NFSV4_MINORVERSION &&
3307             minorvers != NFSV41_MINORVERSION &&
3308             minorvers != NFSV42_MINORVERSION)
3309                 nd->nd_repstat = NFSERR_MINORVERMISMATCH;
3310         cbident = fxdr_unsigned(u_int32_t, *tl++);
3311         if (nd->nd_repstat)
3312                 numops = 0;
3313         else
3314                 numops = fxdr_unsigned(int, *tl);
3315         /*
3316          * Loop around doing the sub ops.
3317          */
3318         for (i = 0; i < numops; i++) {
3319                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3320                 NFSM_BUILD(repp, u_int32_t *, 2 * NFSX_UNSIGNED);
3321                 *repp++ = *tl;
3322                 op = fxdr_unsigned(int, *tl);
3323                 if (op < NFSV4OP_CBGETATTR ||
3324                    (op > NFSV4OP_CBRECALL && minorvers == NFSV4_MINORVERSION) ||
3325                    (op > NFSV4OP_CBNOTIFYDEVID &&
3326                     minorvers == NFSV41_MINORVERSION) ||
3327                    (op > NFSV4OP_CBOFFLOAD &&
3328                     minorvers == NFSV42_MINORVERSION)) {
3329                     nd->nd_repstat = NFSERR_OPILLEGAL;
3330                     *repp = nfscl_errmap(nd, minorvers);
3331                     retops++;
3332                     break;
3333                 }
3334                 nd->nd_procnum = op;
3335                 if (op < NFSV42_CBNOPS)
3336                         nfsstatsv1.cbrpccnt[nd->nd_procnum]++;
3337                 switch (op) {
3338                 case NFSV4OP_CBGETATTR:
3339                         NFSCL_DEBUG(4, "cbgetattr\n");
3340                         mp = NULL;
3341                         vp = NULL;
3342                         error = nfsm_getfh(nd, &nfhp);
3343                         if (!error)
3344                                 error = nfsrv_getattrbits(nd, &attrbits,
3345                                     NULL, NULL);
3346                         if (error == 0 && i == 0 &&
3347                             minorvers != NFSV4_MINORVERSION)
3348                                 error = NFSERR_OPNOTINSESS;
3349                         if (!error) {
3350                                 mp = nfscl_getmnt(minorvers, sessionid, cbident,
3351                                     &clp);
3352                                 if (mp == NULL)
3353                                         error = NFSERR_SERVERFAULT;
3354                         }
3355                         if (!error) {
3356                                 error = nfscl_ngetreopen(mp, nfhp->nfh_fh,
3357                                     nfhp->nfh_len, p, &np);
3358                                 if (!error)
3359                                         vp = NFSTOV(np);
3360                         }
3361                         if (!error) {
3362                                 NFSZERO_ATTRBIT(&rattrbits);
3363                                 NFSLOCKCLSTATE();
3364                                 dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
3365                                     nfhp->nfh_len);
3366                                 if (dp != NULL) {
3367                                         if (NFSISSET_ATTRBIT(&attrbits,
3368                                             NFSATTRBIT_SIZE)) {
3369                                                 if (vp != NULL)
3370                                                         va.va_size = np->n_size;
3371                                                 else
3372                                                         va.va_size =
3373                                                             dp->nfsdl_size;
3374                                                 NFSSETBIT_ATTRBIT(&rattrbits,
3375                                                     NFSATTRBIT_SIZE);
3376                                         }
3377                                         if (NFSISSET_ATTRBIT(&attrbits,
3378                                             NFSATTRBIT_CHANGE)) {
3379                                                 va.va_filerev =
3380                                                     dp->nfsdl_change;
3381                                                 if (vp == NULL ||
3382                                                     (np->n_flag & NDELEGMOD))
3383                                                         va.va_filerev++;
3384                                                 NFSSETBIT_ATTRBIT(&rattrbits,
3385                                                     NFSATTRBIT_CHANGE);
3386                                         }
3387                                 } else
3388                                         error = NFSERR_SERVERFAULT;
3389                                 NFSUNLOCKCLSTATE();
3390                         }
3391                         if (vp != NULL)
3392                                 vrele(vp);
3393                         if (mp != NULL)
3394                                 vfs_unbusy(mp);
3395                         if (nfhp != NULL)
3396                                 free(nfhp, M_NFSFH);
3397                         if (!error)
3398                                 (void) nfsv4_fillattr(nd, NULL, NULL, NULL, &va,
3399                                     NULL, 0, &rattrbits, NULL, p, 0, 0, 0, 0,
3400                                     (uint64_t)0, NULL);
3401                         break;
3402                 case NFSV4OP_CBRECALL:
3403                         NFSCL_DEBUG(4, "cbrecall\n");
3404                         NFSM_DISSECT(tl, u_int32_t *, NFSX_STATEID +
3405                             NFSX_UNSIGNED);
3406                         stateid.seqid = *tl++;
3407                         NFSBCOPY((caddr_t)tl, (caddr_t)stateid.other,
3408                             NFSX_STATEIDOTHER);
3409                         tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
3410                         trunc = fxdr_unsigned(int, *tl);
3411                         error = nfsm_getfh(nd, &nfhp);
3412                         if (error == 0 && i == 0 &&
3413                             minorvers != NFSV4_MINORVERSION)
3414                                 error = NFSERR_OPNOTINSESS;
3415                         if (!error) {
3416                                 NFSLOCKCLSTATE();
3417                                 if (minorvers == NFSV4_MINORVERSION)
3418                                         clp = nfscl_getclnt(cbident);
3419                                 else
3420                                         clp = nfscl_getclntsess(sessionid);
3421                                 if (clp != NULL) {
3422                                         dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
3423                                             nfhp->nfh_len);
3424                                         if (dp != NULL && (dp->nfsdl_flags &
3425                                             NFSCLDL_DELEGRET) == 0) {
3426                                                 dp->nfsdl_flags |=
3427                                                     NFSCLDL_RECALL;
3428                                                 wakeup((caddr_t)clp);
3429                                         }
3430                                 } else {
3431                                         error = NFSERR_SERVERFAULT;
3432                                 }
3433                                 NFSUNLOCKCLSTATE();
3434                         }
3435                         if (nfhp != NULL)
3436                                 free(nfhp, M_NFSFH);
3437                         break;
3438                 case NFSV4OP_CBLAYOUTRECALL:
3439                         NFSCL_DEBUG(4, "cblayrec\n");
3440                         nfhp = NULL;
3441                         NFSM_DISSECT(tl, uint32_t *, 4 * NFSX_UNSIGNED);
3442                         laytype = fxdr_unsigned(int, *tl++);
3443                         iomode = fxdr_unsigned(uint32_t, *tl++);
3444                         if (newnfs_true == *tl++)
3445                                 changed = 1;
3446                         else
3447                                 changed = 0;
3448                         recalltype = fxdr_unsigned(int, *tl);
3449                         NFSCL_DEBUG(4, "layt=%d iom=%d ch=%d rectyp=%d\n",
3450                             laytype, iomode, changed, recalltype);
3451                         recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL,
3452                             M_WAITOK);
3453                         if (laytype != NFSLAYOUT_NFSV4_1_FILES &&
3454                             laytype != NFSLAYOUT_FLEXFILE)
3455                                 error = NFSERR_NOMATCHLAYOUT;
3456                         else if (recalltype == NFSLAYOUTRETURN_FILE) {
3457                                 error = nfsm_getfh(nd, &nfhp);
3458                                 NFSCL_DEBUG(4, "retfile getfh=%d\n", error);
3459                                 if (error != 0)
3460                                         goto nfsmout;
3461                                 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_HYPER +
3462                                     NFSX_STATEID);
3463                                 off = fxdr_hyper(tl); tl += 2;
3464                                 len = fxdr_hyper(tl); tl += 2;
3465                                 stateid.seqid = fxdr_unsigned(uint32_t, *tl++);
3466                                 NFSBCOPY(tl, stateid.other, NFSX_STATEIDOTHER);
3467                                 if (minorvers == NFSV4_MINORVERSION)
3468                                         error = NFSERR_NOTSUPP;
3469                                 else if (i == 0)
3470                                         error = NFSERR_OPNOTINSESS;
3471                                 NFSCL_DEBUG(4, "off=%ju len=%ju sq=%u err=%d\n",
3472                                     (uintmax_t)off, (uintmax_t)len,
3473                                     stateid.seqid, error);
3474                                 if (error == 0) {
3475                                         NFSLOCKCLSTATE();
3476                                         clp = nfscl_getclntsess(sessionid);
3477                                         NFSCL_DEBUG(4, "cbly clp=%p\n", clp);
3478                                         if (clp != NULL) {
3479                                                 lyp = nfscl_findlayout(clp,
3480                                                     nfhp->nfh_fh,
3481                                                     nfhp->nfh_len);
3482                                                 NFSCL_DEBUG(4, "cblyp=%p\n",
3483                                                     lyp);
3484                                                 if (lyp != NULL &&
3485                                                     (lyp->nfsly_flags &
3486                                                      (NFSLY_FILES |
3487                                                       NFSLY_FLEXFILE)) != 0 &&
3488                                                     !NFSBCMP(stateid.other,
3489                                                     lyp->nfsly_stateid.other,
3490                                                     NFSX_STATEIDOTHER)) {
3491                                                         error =
3492                                                             nfscl_layoutrecall(
3493                                                             recalltype,
3494                                                             lyp, iomode, off,
3495                                                             len, stateid.seqid,
3496                                                             0, 0, NULL,
3497                                                             recallp);
3498                                                         recallp = NULL;
3499                                                         wakeup(clp);
3500                                                         NFSCL_DEBUG(4,
3501                                                             "aft layrcal=%d\n",
3502                                                             error);
3503                                                 } else
3504                                                         error =
3505                                                           NFSERR_NOMATCHLAYOUT;
3506                                         } else
3507                                                 error = NFSERR_NOMATCHLAYOUT;
3508                                         NFSUNLOCKCLSTATE();
3509                                 }
3510                                 free(nfhp, M_NFSFH);
3511                         } else if (recalltype == NFSLAYOUTRETURN_FSID) {
3512                                 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_HYPER);
3513                                 filesid[0] = fxdr_hyper(tl); tl += 2;
3514                                 filesid[1] = fxdr_hyper(tl); tl += 2;
3515                                 gotone = 0;
3516                                 NFSLOCKCLSTATE();
3517                                 clp = nfscl_getclntsess(sessionid);
3518                                 if (clp != NULL) {
3519                                         TAILQ_FOREACH(lyp, &clp->nfsc_layout,
3520                                             nfsly_list) {
3521                                                 if (lyp->nfsly_filesid[0] ==
3522                                                     filesid[0] &&
3523                                                     lyp->nfsly_filesid[1] ==
3524                                                     filesid[1]) {
3525                                                         error =
3526                                                             nfscl_layoutrecall(
3527                                                             recalltype,
3528                                                             lyp, iomode, 0,
3529                                                             UINT64_MAX,
3530                                                             lyp->nfsly_stateid.seqid,
3531                                                             0, 0, NULL,
3532                                                             recallp);
3533                                                         recallp = NULL;
3534                                                         gotone = 1;
3535                                                 }
3536                                         }
3537                                         if (gotone != 0)
3538                                                 wakeup(clp);
3539                                         else
3540                                                 error = NFSERR_NOMATCHLAYOUT;
3541                                 } else
3542                                         error = NFSERR_NOMATCHLAYOUT;
3543                                 NFSUNLOCKCLSTATE();
3544                         } else if (recalltype == NFSLAYOUTRETURN_ALL) {
3545                                 gotone = 0;
3546                                 NFSLOCKCLSTATE();
3547                                 clp = nfscl_getclntsess(sessionid);
3548                                 if (clp != NULL) {
3549                                         TAILQ_FOREACH(lyp, &clp->nfsc_layout,
3550                                             nfsly_list) {
3551                                                 error = nfscl_layoutrecall(
3552                                                     recalltype, lyp, iomode, 0,
3553                                                     UINT64_MAX,
3554                                                     lyp->nfsly_stateid.seqid,
3555                                                     0, 0, NULL, recallp);
3556                                                 recallp = NULL;
3557                                                 gotone = 1;
3558                                         }
3559                                         if (gotone != 0)
3560                                                 wakeup(clp);
3561                                         else
3562                                                 error = NFSERR_NOMATCHLAYOUT;
3563                                 } else
3564                                         error = NFSERR_NOMATCHLAYOUT;
3565                                 NFSUNLOCKCLSTATE();
3566                         } else
3567                                 error = NFSERR_NOMATCHLAYOUT;
3568                         if (recallp != NULL) {
3569                                 free(recallp, M_NFSLAYRECALL);
3570                                 recallp = NULL;
3571                         }
3572                         break;
3573                 case NFSV4OP_CBSEQUENCE:
3574                         NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID +
3575                             5 * NFSX_UNSIGNED);
3576                         bcopy(tl, sessionid, NFSX_V4SESSIONID);
3577                         tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3578                         seqid = fxdr_unsigned(uint32_t, *tl++);
3579                         slotid = fxdr_unsigned(uint32_t, *tl++);
3580                         highslot = fxdr_unsigned(uint32_t, *tl++);
3581                         cachethis = *tl++;
3582                         /* Throw away the referring call stuff. */
3583                         clist = fxdr_unsigned(int, *tl);
3584                         for (j = 0; j < clist; j++) {
3585                                 NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID +
3586                                     NFSX_UNSIGNED);
3587                                 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3588                                 rcalls = fxdr_unsigned(int, *tl);
3589                                 for (k = 0; k < rcalls; k++) {
3590                                         NFSM_DISSECT(tl, uint32_t *,
3591                                             2 * NFSX_UNSIGNED);
3592                                 }
3593                         }
3594                         NFSLOCKCLSTATE();
3595                         if (i == 0) {
3596                                 clp = nfscl_getclntsess(sessionid);
3597                                 if (clp == NULL)
3598                                         error = NFSERR_SERVERFAULT;
3599                         } else
3600                                 error = NFSERR_SEQUENCEPOS;
3601                         if (error == 0) {
3602                                 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3603                                 error = nfsv4_seqsession(seqid, slotid,
3604                                     highslot, tsep->nfsess_cbslots, &rep,
3605                                     tsep->nfsess_backslots);
3606                         }
3607                         NFSUNLOCKCLSTATE();
3608                         if (error == 0 || error == NFSERR_REPLYFROMCACHE) {
3609                                 gotseq_ok = 1;
3610                                 if (rep != NULL) {
3611                                         /*
3612                                          * Handle a reply for a retried
3613                                          * callback.  The reply will be
3614                                          * re-inserted in the session cache
3615                                          * by the nfsv4_seqsess_cacherep() call
3616                                          * after out:
3617                                          */
3618                                         KASSERT(error == NFSERR_REPLYFROMCACHE,
3619                                             ("cbsequence: non-NULL rep"));
3620                                         NFSCL_DEBUG(4, "Got cbretry\n");
3621                                         m_freem(nd->nd_mreq);
3622                                         nd->nd_mreq = rep;
3623                                         rep = NULL;
3624                                         goto out;
3625                                 }
3626                                 NFSM_BUILD(tl, uint32_t *,
3627                                     NFSX_V4SESSIONID + 4 * NFSX_UNSIGNED);
3628                                 bcopy(sessionid, tl, NFSX_V4SESSIONID);
3629                                 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3630                                 *tl++ = txdr_unsigned(seqid);
3631                                 *tl++ = txdr_unsigned(slotid);
3632                                 *tl++ = txdr_unsigned(NFSV4_CBSLOTS - 1);
3633                                 *tl = txdr_unsigned(NFSV4_CBSLOTS - 1);
3634                         }
3635                         break;
3636                 default:
3637                         if (i == 0 && minorvers != NFSV4_MINORVERSION)
3638                                 error = NFSERR_OPNOTINSESS;
3639                         else {
3640                                 NFSCL_DEBUG(1, "unsupp callback %d\n", op);
3641                                 error = NFSERR_NOTSUPP;
3642                         }
3643                         break;
3644                 }
3645                 if (error) {
3646                         if (error == EBADRPC || error == NFSERR_BADXDR) {
3647                                 nd->nd_repstat = NFSERR_BADXDR;
3648                         } else {
3649                                 nd->nd_repstat = error;
3650                         }
3651                         error = 0;
3652                 }
3653                 retops++;
3654                 if (nd->nd_repstat) {
3655                         *repp = nfscl_errmap(nd, minorvers);
3656                         break;
3657                 } else
3658                         *repp = 0;      /* NFS4_OK */
3659         }
3660 nfsmout:
3661         if (recallp != NULL)
3662                 free(recallp, M_NFSLAYRECALL);
3663         if (error) {
3664                 if (error == EBADRPC || error == NFSERR_BADXDR)
3665                         nd->nd_repstat = NFSERR_BADXDR;
3666                 else
3667                         printf("nfsv4 comperr1=%d\n", error);
3668         }
3669         if (taglen == -1) {
3670                 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
3671                 *tl++ = 0;
3672                 *tl = 0;
3673         } else {
3674                 *retopsp = txdr_unsigned(retops);
3675         }
3676         *nd->nd_errp = nfscl_errmap(nd, minorvers);
3677 out:
3678         if (gotseq_ok != 0) {
3679                 rep = m_copym(nd->nd_mreq, 0, M_COPYALL, M_WAITOK);
3680                 NFSLOCKCLSTATE();
3681                 clp = nfscl_getclntsess(sessionid);
3682                 if (clp != NULL) {
3683                         tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3684                         nfsv4_seqsess_cacherep(slotid, tsep->nfsess_cbslots,
3685                             NFSERR_OK, &rep);
3686                         NFSUNLOCKCLSTATE();
3687                 } else {
3688                         NFSUNLOCKCLSTATE();
3689                         m_freem(rep);
3690                 }
3691         }
3692 }
3693
3694 /*
3695  * Generate the next cbident value. Basically just increment a static value
3696  * and then check that it isn't already in the list, if it has wrapped around.
3697  */
3698 static u_int32_t
3699 nfscl_nextcbident(void)
3700 {
3701         struct nfsclclient *clp;
3702         int matched;
3703         static u_int32_t nextcbident = 0;
3704         static int haswrapped = 0;
3705
3706         nextcbident++;
3707         if (nextcbident == 0)
3708                 haswrapped = 1;
3709         if (haswrapped) {
3710                 /*
3711                  * Search the clientid list for one already using this cbident.
3712                  */
3713                 do {
3714                         matched = 0;
3715                         NFSLOCKCLSTATE();
3716                         LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3717                                 if (clp->nfsc_cbident == nextcbident) {
3718                                         matched = 1;
3719                                         break;
3720                                 }
3721                         }
3722                         NFSUNLOCKCLSTATE();
3723                         if (matched == 1)
3724                                 nextcbident++;
3725                 } while (matched);
3726         }
3727         return (nextcbident);
3728 }
3729
3730 /*
3731  * Get the mount point related to a given cbident or session and busy it.
3732  */
3733 static mount_t
3734 nfscl_getmnt(int minorvers, uint8_t *sessionid, u_int32_t cbident,
3735     struct nfsclclient **clpp)
3736 {
3737         struct nfsclclient *clp;
3738         mount_t mp;
3739         int error;
3740         struct nfsclsession *tsep;
3741
3742         *clpp = NULL;
3743         NFSLOCKCLSTATE();
3744         LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3745                 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3746                 if (minorvers == NFSV4_MINORVERSION) {
3747                         if (clp->nfsc_cbident == cbident)
3748                                 break;
3749                 } else if (!NFSBCMP(tsep->nfsess_sessionid, sessionid,
3750                     NFSX_V4SESSIONID))
3751                         break;
3752         }
3753         if (clp == NULL) {
3754                 NFSUNLOCKCLSTATE();
3755                 return (NULL);
3756         }
3757         mp = clp->nfsc_nmp->nm_mountp;
3758         vfs_ref(mp);
3759         NFSUNLOCKCLSTATE();
3760         error = vfs_busy(mp, 0);
3761         vfs_rel(mp);
3762         if (error != 0)
3763                 return (NULL);
3764         *clpp = clp;
3765         return (mp);
3766 }
3767
3768 /*
3769  * Get the clientid pointer related to a given cbident.
3770  */
3771 static struct nfsclclient *
3772 nfscl_getclnt(u_int32_t cbident)
3773 {
3774         struct nfsclclient *clp;
3775
3776         LIST_FOREACH(clp, &nfsclhead, nfsc_list)
3777                 if (clp->nfsc_cbident == cbident)
3778                         break;
3779         return (clp);
3780 }
3781
3782 /*
3783  * Get the clientid pointer related to a given sessionid.
3784  */
3785 static struct nfsclclient *
3786 nfscl_getclntsess(uint8_t *sessionid)
3787 {
3788         struct nfsclclient *clp;
3789         struct nfsclsession *tsep;
3790
3791         LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3792                 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3793                 if (!NFSBCMP(tsep->nfsess_sessionid, sessionid,
3794                     NFSX_V4SESSIONID))
3795                         break;
3796         }
3797         return (clp);
3798 }
3799
3800 /*
3801  * Search for a lock conflict locally on the client. A conflict occurs if
3802  * - not same owner and overlapping byte range and at least one of them is
3803  *   a write lock or this is an unlock.
3804  */
3805 static int
3806 nfscl_localconflict(struct nfsclclient *clp, u_int8_t *fhp, int fhlen,
3807     struct nfscllock *nlop, u_int8_t *own, struct nfscldeleg *dp,
3808     struct nfscllock **lopp)
3809 {
3810         struct nfsclowner *owp;
3811         struct nfsclopen *op;
3812         int ret;
3813
3814         if (dp != NULL) {
3815                 ret = nfscl_checkconflict(&dp->nfsdl_lock, nlop, own, lopp);
3816                 if (ret)
3817                         return (ret);
3818         }
3819         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3820                 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3821                         if (op->nfso_fhlen == fhlen &&
3822                             !NFSBCMP(op->nfso_fh, fhp, fhlen)) {
3823                                 ret = nfscl_checkconflict(&op->nfso_lock, nlop,
3824                                     own, lopp);
3825                                 if (ret)
3826                                         return (ret);
3827                         }
3828                 }
3829         }
3830         return (0);
3831 }
3832
3833 static int
3834 nfscl_checkconflict(struct nfscllockownerhead *lhp, struct nfscllock *nlop,
3835     u_int8_t *own, struct nfscllock **lopp)
3836 {
3837         struct nfscllockowner *lp;
3838         struct nfscllock *lop;
3839
3840         LIST_FOREACH(lp, lhp, nfsl_list) {
3841                 if (NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
3842                         LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
3843                                 if (lop->nfslo_first >= nlop->nfslo_end)
3844                                         break;
3845                                 if (lop->nfslo_end <= nlop->nfslo_first)
3846                                         continue;
3847                                 if (lop->nfslo_type == F_WRLCK ||
3848                                     nlop->nfslo_type == F_WRLCK ||
3849                                     nlop->nfslo_type == F_UNLCK) {
3850                                         if (lopp != NULL)
3851                                                 *lopp = lop;
3852                                         return (NFSERR_DENIED);
3853                                 }
3854                         }
3855                 }
3856         }
3857         return (0);
3858 }
3859
3860 /*
3861  * Check for a local conflicting lock.
3862  */
3863 int
3864 nfscl_lockt(vnode_t vp, struct nfsclclient *clp, u_int64_t off,
3865     u_int64_t len, struct flock *fl, NFSPROC_T *p, void *id, int flags)
3866 {
3867         struct nfscllock *lop, nlck;
3868         struct nfscldeleg *dp;
3869         struct nfsnode *np;
3870         u_int8_t own[NFSV4CL_LOCKNAMELEN];
3871         int error;
3872
3873         nlck.nfslo_type = fl->l_type;
3874         nlck.nfslo_first = off;
3875         if (len == NFS64BITSSET) {
3876                 nlck.nfslo_end = NFS64BITSSET;
3877         } else {
3878                 nlck.nfslo_end = off + len;
3879                 if (nlck.nfslo_end <= nlck.nfslo_first)
3880                         return (NFSERR_INVAL);
3881         }
3882         np = VTONFS(vp);
3883         nfscl_filllockowner(id, own, flags);
3884         NFSLOCKCLSTATE();
3885         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
3886         error = nfscl_localconflict(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len,
3887             &nlck, own, dp, &lop);
3888         if (error != 0) {
3889                 fl->l_whence = SEEK_SET;
3890                 fl->l_start = lop->nfslo_first;
3891                 if (lop->nfslo_end == NFS64BITSSET)
3892                         fl->l_len = 0;
3893                 else
3894                         fl->l_len = lop->nfslo_end - lop->nfslo_first;
3895                 fl->l_pid = (pid_t)0;
3896                 fl->l_type = lop->nfslo_type;
3897                 error = -1;                     /* no RPC required */
3898         } else if (dp != NULL && ((dp->nfsdl_flags & NFSCLDL_WRITE) ||
3899             fl->l_type == F_RDLCK)) {
3900                 /*
3901                  * The delegation ensures that there isn't a conflicting
3902                  * lock on the server, so return -1 to indicate an RPC
3903                  * isn't required.
3904                  */
3905                 fl->l_type = F_UNLCK;
3906                 error = -1;
3907         }
3908         NFSUNLOCKCLSTATE();
3909         return (error);
3910 }
3911
3912 /*
3913  * Handle Recall of a delegation.
3914  * The clp must be exclusive locked when this is called.
3915  */
3916 static int
3917 nfscl_recalldeleg(struct nfsclclient *clp, struct nfsmount *nmp,
3918     struct nfscldeleg *dp, vnode_t vp, struct ucred *cred, NFSPROC_T *p,
3919     int called_from_renewthread)
3920 {
3921         struct nfsclowner *owp, *lowp, *nowp;
3922         struct nfsclopen *op, *lop;
3923         struct nfscllockowner *lp;
3924         struct nfscllock *lckp;
3925         struct nfsnode *np;
3926         int error = 0, ret, gotvp = 0;
3927
3928         if (vp == NULL) {
3929                 /*
3930                  * First, get a vnode for the file. This is needed to do RPCs.
3931                  */
3932                 ret = nfscl_ngetreopen(nmp->nm_mountp, dp->nfsdl_fh,
3933                     dp->nfsdl_fhlen, p, &np);
3934                 if (ret) {
3935                         /*
3936                          * File isn't open, so nothing to move over to the
3937                          * server.
3938                          */
3939                         return (0);
3940                 }
3941                 vp = NFSTOV(np);
3942                 gotvp = 1;
3943         } else {
3944                 np = VTONFS(vp);
3945         }
3946         dp->nfsdl_flags &= ~NFSCLDL_MODTIMESET;
3947
3948         /*
3949          * Ok, if it's a write delegation, flush data to the server, so
3950          * that close/open consistency is retained.
3951          */
3952         ret = 0;
3953         NFSLOCKNODE(np);
3954         if ((dp->nfsdl_flags & NFSCLDL_WRITE) && (np->n_flag & NMODIFIED)) {
3955                 np->n_flag |= NDELEGRECALL;
3956                 NFSUNLOCKNODE(np);
3957                 ret = ncl_flush(vp, MNT_WAIT, p, 1, called_from_renewthread);
3958                 NFSLOCKNODE(np);
3959                 np->n_flag &= ~NDELEGRECALL;
3960         }
3961         NFSINVALATTRCACHE(np);
3962         NFSUNLOCKNODE(np);
3963         if (ret == EIO && called_from_renewthread != 0) {
3964                 /*
3965                  * If the flush failed with EIO for the renew thread,
3966                  * return now, so that the dirty buffer will be flushed
3967                  * later.
3968                  */
3969                 if (gotvp != 0)
3970                         vrele(vp);
3971                 return (ret);
3972         }
3973
3974         /*
3975          * Now, for each openowner with opens issued locally, move them
3976          * over to state against the server.
3977          */
3978         LIST_FOREACH(lowp, &dp->nfsdl_owner, nfsow_list) {
3979                 lop = LIST_FIRST(&lowp->nfsow_open);
3980                 if (lop != NULL) {
3981                         if (LIST_NEXT(lop, nfso_list) != NULL)
3982                                 panic("nfsdlg mult opens");
3983                         /*
3984                          * Look for the same openowner against the server.
3985                          */
3986                         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3987                                 if (!NFSBCMP(lowp->nfsow_owner,
3988                                     owp->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
3989                                         newnfs_copycred(&dp->nfsdl_cred, cred);
3990                                         ret = nfscl_moveopen(vp, clp, nmp, lop,
3991                                             owp, dp, cred, p);
3992                                         if (ret == NFSERR_STALECLIENTID ||
3993                                             ret == NFSERR_STALEDONTRECOVER ||
3994                                             ret == NFSERR_BADSESSION) {
3995                                                 if (gotvp)
3996                                                         vrele(vp);
3997                                                 return (ret);
3998                                         }
3999                                         if (ret) {
4000                                                 nfscl_freeopen(lop, 1);
4001                                                 if (!error)
4002                                                         error = ret;
4003                                         }
4004                                         break;
4005                                 }
4006                         }
4007
4008                         /*
4009                          * If no openowner found, create one and get an open
4010                          * for it.
4011                          */
4012                         if (owp == NULL) {
4013                                 nowp = malloc(
4014                                     sizeof (struct nfsclowner), M_NFSCLOWNER,
4015                                     M_WAITOK);
4016                                 nfscl_newopen(clp, NULL, &owp, &nowp, &op,
4017                                     NULL, lowp->nfsow_owner, dp->nfsdl_fh,
4018                                     dp->nfsdl_fhlen, NULL, NULL);
4019                                 newnfs_copycred(&dp->nfsdl_cred, cred);
4020                                 ret = nfscl_moveopen(vp, clp, nmp, lop,
4021                                     owp, dp, cred, p);
4022                                 if (ret) {
4023                                         nfscl_freeopenowner(owp, 0);
4024                                         if (ret == NFSERR_STALECLIENTID ||
4025                                             ret == NFSERR_STALEDONTRECOVER ||
4026                                             ret == NFSERR_BADSESSION) {
4027                                                 if (gotvp)
4028                                                         vrele(vp);
4029                                                 return (ret);
4030                                         }
4031                                         if (ret) {
4032                                                 nfscl_freeopen(lop, 1);
4033                                                 if (!error)
4034                                                         error = ret;
4035                                         }
4036                                 }
4037                         }
4038                 }
4039         }
4040
4041         /*
4042          * Now, get byte range locks for any locks done locally.
4043          */
4044         LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4045                 LIST_FOREACH(lckp, &lp->nfsl_lock, nfslo_list) {
4046                         newnfs_copycred(&dp->nfsdl_cred, cred);
4047                         ret = nfscl_relock(vp, clp, nmp, lp, lckp, cred, p);
4048                         if (ret == NFSERR_STALESTATEID ||
4049                             ret == NFSERR_STALEDONTRECOVER ||
4050                             ret == NFSERR_STALECLIENTID ||
4051                             ret == NFSERR_BADSESSION) {
4052                                 if (gotvp)
4053                                         vrele(vp);
4054                                 return (ret);
4055                         }
4056                         if (ret && !error)
4057                                 error = ret;
4058                 }
4059         }
4060         if (gotvp)
4061                 vrele(vp);
4062         return (error);
4063 }
4064
4065 /*
4066  * Move a locally issued open over to an owner on the state list.
4067  * SIDE EFFECT: If it needs to sleep (do an rpc), it unlocks clstate and
4068  * returns with it unlocked.
4069  */
4070 static int
4071 nfscl_moveopen(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp,
4072     struct nfsclopen *lop, struct nfsclowner *owp, struct nfscldeleg *dp,
4073     struct ucred *cred, NFSPROC_T *p)
4074 {
4075         struct nfsclopen *op, *nop;
4076         struct nfscldeleg *ndp;
4077         struct nfsnode *np;
4078         int error = 0, newone;
4079
4080         /*
4081          * First, look for an appropriate open, If found, just increment the
4082          * opencnt in it.
4083          */
4084         LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
4085                 if ((op->nfso_mode & lop->nfso_mode) == lop->nfso_mode &&
4086                     op->nfso_fhlen == lop->nfso_fhlen &&
4087                     !NFSBCMP(op->nfso_fh, lop->nfso_fh, op->nfso_fhlen)) {
4088                         op->nfso_opencnt += lop->nfso_opencnt;
4089                         nfscl_freeopen(lop, 1);
4090                         return (0);
4091                 }
4092         }
4093
4094         /* No appropriate open, so we have to do one against the server. */
4095         np = VTONFS(vp);
4096         nop = malloc(sizeof (struct nfsclopen) +
4097             lop->nfso_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
4098         newone = 0;
4099         nfscl_newopen(clp, NULL, &owp, NULL, &op, &nop, owp->nfsow_owner,
4100             lop->nfso_fh, lop->nfso_fhlen, cred, &newone);
4101         ndp = dp;
4102         error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data, np->n_v4->n4_fhlen,
4103             lop->nfso_fh, lop->nfso_fhlen, lop->nfso_mode, op,
4104             NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, &ndp, 0, 0, cred, p);
4105         if (error) {
4106                 if (newone)
4107                         nfscl_freeopen(op, 0);
4108         } else {
4109                 op->nfso_mode |= lop->nfso_mode;
4110                 op->nfso_opencnt += lop->nfso_opencnt;
4111                 nfscl_freeopen(lop, 1);
4112         }
4113         if (nop != NULL)
4114                 free(nop, M_NFSCLOPEN);
4115         if (ndp != NULL) {
4116                 /*
4117                  * What should I do with the returned delegation, since the
4118                  * delegation is being recalled? For now, just printf and
4119                  * through it away.
4120                  */
4121                 printf("Moveopen returned deleg\n");
4122                 free(ndp, M_NFSCLDELEG);
4123         }
4124         return (error);
4125 }
4126
4127 /*
4128  * Recall all delegations on this client.
4129  */
4130 static void
4131 nfscl_totalrecall(struct nfsclclient *clp)
4132 {
4133         struct nfscldeleg *dp;
4134
4135         TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
4136                 if ((dp->nfsdl_flags & NFSCLDL_DELEGRET) == 0)
4137                         dp->nfsdl_flags |= NFSCLDL_RECALL;
4138         }
4139 }
4140
4141 /*
4142  * Relock byte ranges. Called for delegation recall and state expiry.
4143  */
4144 static int
4145 nfscl_relock(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp,
4146     struct nfscllockowner *lp, struct nfscllock *lop, struct ucred *cred,
4147     NFSPROC_T *p)
4148 {
4149         struct nfscllockowner *nlp;
4150         struct nfsfh *nfhp;
4151         u_int64_t off, len;
4152         int error, newone, donelocally;
4153
4154         off = lop->nfslo_first;
4155         len = lop->nfslo_end - lop->nfslo_first;
4156         error = nfscl_getbytelock(vp, off, len, lop->nfslo_type, cred, p,
4157             clp, 1, NULL, lp->nfsl_lockflags, lp->nfsl_owner,
4158             lp->nfsl_openowner, &nlp, &newone, &donelocally);
4159         if (error || donelocally)
4160                 return (error);
4161         nfhp = VTONFS(vp)->n_fhp;
4162         error = nfscl_trylock(nmp, vp, nfhp->nfh_fh,
4163             nfhp->nfh_len, nlp, newone, 0, off,
4164             len, lop->nfslo_type, cred, p);
4165         if (error)
4166                 nfscl_freelockowner(nlp, 0);
4167         return (error);
4168 }
4169
4170 /*
4171  * Called to re-open a file. Basically get a vnode for the file handle
4172  * and then call nfsrpc_openrpc() to do the rest.
4173  */
4174 static int
4175 nfsrpc_reopen(struct nfsmount *nmp, u_int8_t *fhp, int fhlen,
4176     u_int32_t mode, struct nfsclopen *op, struct nfscldeleg **dpp,
4177     struct ucred *cred, NFSPROC_T *p)
4178 {
4179         struct nfsnode *np;
4180         vnode_t vp;
4181         int error;
4182
4183         error = nfscl_ngetreopen(nmp->nm_mountp, fhp, fhlen, p, &np);
4184         if (error)
4185                 return (error);
4186         vp = NFSTOV(np);
4187         if (np->n_v4 != NULL) {
4188                 error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data,
4189                     np->n_v4->n4_fhlen, fhp, fhlen, mode, op,
4190                     NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, dpp, 0, 0,
4191                     cred, p);
4192         } else {
4193                 error = EINVAL;
4194         }
4195         vrele(vp);
4196         return (error);
4197 }
4198
4199 /*
4200  * Try an open against the server. Just call nfsrpc_openrpc(), retrying while
4201  * NFSERR_DELAY. Also, try system credentials, if the passed in credentials
4202  * fail.
4203  */
4204 static int
4205 nfscl_tryopen(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen,
4206     u_int8_t *newfhp, int newfhlen, u_int32_t mode, struct nfsclopen *op,
4207     u_int8_t *name, int namelen, struct nfscldeleg **ndpp,
4208     int reclaim, u_int32_t delegtype, struct ucred *cred, NFSPROC_T *p)
4209 {
4210         int error;
4211
4212         do {
4213                 error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp, newfhlen,
4214                     mode, op, name, namelen, ndpp, reclaim, delegtype, cred, p,
4215                     0, 0);
4216                 if (error == NFSERR_DELAY)
4217                         (void) nfs_catnap(PZERO, error, "nfstryop");
4218         } while (error == NFSERR_DELAY);
4219         if (error == EAUTH || error == EACCES) {
4220                 /* Try again using system credentials */
4221                 newnfs_setroot(cred);
4222                 do {
4223                     error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp,
4224                         newfhlen, mode, op, name, namelen, ndpp, reclaim,
4225                         delegtype, cred, p, 1, 0);
4226                     if (error == NFSERR_DELAY)
4227                         (void) nfs_catnap(PZERO, error, "nfstryop");
4228                 } while (error == NFSERR_DELAY);
4229         }
4230         return (error);
4231 }
4232
4233 /*
4234  * Try a byte range lock. Just loop on nfsrpc_lock() while it returns
4235  * NFSERR_DELAY. Also, retry with system credentials, if the provided
4236  * cred don't work.
4237  */
4238 static int
4239 nfscl_trylock(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp,
4240     int fhlen, struct nfscllockowner *nlp, int newone, int reclaim,
4241     u_int64_t off, u_int64_t len, short type, struct ucred *cred, NFSPROC_T *p)
4242 {
4243         struct nfsrv_descript nfsd, *nd = &nfsd;
4244         int error;
4245
4246         do {
4247                 error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp, newone,
4248                     reclaim, off, len, type, cred, p, 0);
4249                 if (!error && nd->nd_repstat == NFSERR_DELAY)
4250                         (void) nfs_catnap(PZERO, (int)nd->nd_repstat,
4251                             "nfstrylck");
4252         } while (!error && nd->nd_repstat == NFSERR_DELAY);
4253         if (!error)
4254                 error = nd->nd_repstat;
4255         if (error == EAUTH || error == EACCES) {
4256                 /* Try again using root credentials */
4257                 newnfs_setroot(cred);
4258                 do {
4259                         error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp,
4260                             newone, reclaim, off, len, type, cred, p, 1);
4261                         if (!error && nd->nd_repstat == NFSERR_DELAY)
4262                                 (void) nfs_catnap(PZERO, (int)nd->nd_repstat,
4263                                     "nfstrylck");
4264                 } while (!error && nd->nd_repstat == NFSERR_DELAY);
4265                 if (!error)
4266                         error = nd->nd_repstat;
4267         }
4268         return (error);
4269 }
4270
4271 /*
4272  * Try a delegreturn against the server. Just call nfsrpc_delegreturn(),
4273  * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
4274  * credentials fail.
4275  */
4276 static int
4277 nfscl_trydelegreturn(struct nfscldeleg *dp, struct ucred *cred,
4278     struct nfsmount *nmp, NFSPROC_T *p)
4279 {
4280         int error;
4281
4282         do {
4283                 error = nfsrpc_delegreturn(dp, cred, nmp, p, 0);
4284                 if (error == NFSERR_DELAY)
4285                         (void) nfs_catnap(PZERO, error, "nfstrydp");
4286         } while (error == NFSERR_DELAY);
4287         if (error == EAUTH || error == EACCES) {
4288                 /* Try again using system credentials */
4289                 newnfs_setroot(cred);
4290                 do {
4291                         error = nfsrpc_delegreturn(dp, cred, nmp, p, 1);
4292                         if (error == NFSERR_DELAY)
4293                                 (void) nfs_catnap(PZERO, error, "nfstrydp");
4294                 } while (error == NFSERR_DELAY);
4295         }
4296         return (error);
4297 }
4298
4299 /*
4300  * Try a close against the server. Just call nfsrpc_closerpc(),
4301  * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
4302  * credentials fail.
4303  */
4304 int
4305 nfscl_tryclose(struct nfsclopen *op, struct ucred *cred,
4306     struct nfsmount *nmp, NFSPROC_T *p)
4307 {
4308         struct nfsrv_descript nfsd, *nd = &nfsd;
4309         int error;
4310
4311         do {
4312                 error = nfsrpc_closerpc(nd, nmp, op, cred, p, 0);
4313                 if (error == NFSERR_DELAY)
4314                         (void) nfs_catnap(PZERO, error, "nfstrycl");
4315         } while (error == NFSERR_DELAY);
4316         if (error == EAUTH || error == EACCES) {
4317                 /* Try again using system credentials */
4318                 newnfs_setroot(cred);
4319                 do {
4320                         error = nfsrpc_closerpc(nd, nmp, op, cred, p, 1);
4321                         if (error == NFSERR_DELAY)
4322                                 (void) nfs_catnap(PZERO, error, "nfstrycl");
4323                 } while (error == NFSERR_DELAY);
4324         }
4325         return (error);
4326 }
4327
4328 /*
4329  * Decide if a delegation on a file permits close without flushing writes
4330  * to the server. This might be a big performance win in some environments.
4331  * (Not useful until the client does caching on local stable storage.)
4332  */
4333 int
4334 nfscl_mustflush(vnode_t vp)
4335 {
4336         struct nfsclclient *clp;
4337         struct nfscldeleg *dp;
4338         struct nfsnode *np;
4339         struct nfsmount *nmp;
4340
4341         np = VTONFS(vp);
4342         nmp = VFSTONFS(vp->v_mount);
4343         if (!NFSHASNFSV4(nmp))
4344                 return (1);
4345         NFSLOCKCLSTATE();
4346         clp = nfscl_findcl(nmp);
4347         if (clp == NULL) {
4348                 NFSUNLOCKCLSTATE();
4349                 return (1);
4350         }
4351         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4352         if (dp != NULL && (dp->nfsdl_flags &
4353             (NFSCLDL_WRITE | NFSCLDL_RECALL | NFSCLDL_DELEGRET)) ==
4354              NFSCLDL_WRITE &&
4355             (dp->nfsdl_sizelimit >= np->n_size ||
4356              !NFSHASSTRICT3530(nmp))) {
4357                 NFSUNLOCKCLSTATE();
4358                 return (0);
4359         }
4360         NFSUNLOCKCLSTATE();
4361         return (1);
4362 }
4363
4364 /*
4365  * See if a (write) delegation exists for this file.
4366  */
4367 int
4368 nfscl_nodeleg(vnode_t vp, int writedeleg)
4369 {
4370         struct nfsclclient *clp;
4371         struct nfscldeleg *dp;
4372         struct nfsnode *np;
4373         struct nfsmount *nmp;
4374
4375         np = VTONFS(vp);
4376         nmp = VFSTONFS(vp->v_mount);
4377         if (!NFSHASNFSV4(nmp))
4378                 return (1);
4379         NFSLOCKCLSTATE();
4380         clp = nfscl_findcl(nmp);
4381         if (clp == NULL) {
4382                 NFSUNLOCKCLSTATE();
4383                 return (1);
4384         }
4385         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4386         if (dp != NULL &&
4387             (dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_DELEGRET)) == 0 &&
4388             (writedeleg == 0 || (dp->nfsdl_flags & NFSCLDL_WRITE) ==
4389              NFSCLDL_WRITE)) {
4390                 NFSUNLOCKCLSTATE();
4391                 return (0);
4392         }
4393         NFSUNLOCKCLSTATE();
4394         return (1);
4395 }
4396
4397 /*
4398  * Look for an associated delegation that should be DelegReturned.
4399  */
4400 int
4401 nfscl_removedeleg(vnode_t vp, NFSPROC_T *p, nfsv4stateid_t *stp)
4402 {
4403         struct nfsclclient *clp;
4404         struct nfscldeleg *dp;
4405         struct nfsclowner *owp;
4406         struct nfscllockowner *lp;
4407         struct nfsmount *nmp;
4408         struct ucred *cred;
4409         struct nfsnode *np;
4410         int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
4411
4412         nmp = VFSTONFS(vp->v_mount);
4413         np = VTONFS(vp);
4414         NFSLOCKCLSTATE();
4415         /*
4416          * Loop around waiting for:
4417          * - outstanding I/O operations on delegations to complete
4418          * - for a delegation on vp that has state, lock the client and
4419          *   do a recall
4420          * - return delegation with no state
4421          */
4422         while (1) {
4423                 clp = nfscl_findcl(nmp);
4424                 if (clp == NULL) {
4425                         NFSUNLOCKCLSTATE();
4426                         return (retcnt);
4427                 }
4428                 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4429                     np->n_fhp->nfh_len);
4430                 if (dp != NULL) {
4431                     /*
4432                      * Wait for outstanding I/O ops to be done.
4433                      */
4434                     if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4435                         if (igotlock) {
4436                             nfsv4_unlock(&clp->nfsc_lock, 0);
4437                             igotlock = 0;
4438                         }
4439                         dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4440                         (void) nfsmsleep(&dp->nfsdl_rwlock,
4441                             NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4442                         continue;
4443                     }
4444                     needsrecall = 0;
4445                     LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4446                         if (!LIST_EMPTY(&owp->nfsow_open)) {
4447                             needsrecall = 1;
4448                             break;
4449                         }
4450                     }
4451                     if (!needsrecall) {
4452                         LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4453                             if (!LIST_EMPTY(&lp->nfsl_lock)) {
4454                                 needsrecall = 1;
4455                                 break;
4456                             }
4457                         }
4458                     }
4459                     if (needsrecall && !triedrecall) {
4460                         dp->nfsdl_flags |= NFSCLDL_DELEGRET;
4461                         islept = 0;
4462                         while (!igotlock) {
4463                             igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
4464                                 &islept, NFSCLSTATEMUTEXPTR, NULL);
4465                             if (islept)
4466                                 break;
4467                         }
4468                         if (islept)
4469                             continue;
4470                         NFSUNLOCKCLSTATE();
4471                         cred = newnfs_getcred();
4472                         newnfs_copycred(&dp->nfsdl_cred, cred);
4473                         (void) nfscl_recalldeleg(clp, nmp, dp, vp, cred, p, 0);
4474                         NFSFREECRED(cred);
4475                         triedrecall = 1;
4476                         NFSLOCKCLSTATE();
4477                         nfsv4_unlock(&clp->nfsc_lock, 0);
4478                         igotlock = 0;
4479                         continue;
4480                     }
4481                     *stp = dp->nfsdl_stateid;
4482                     retcnt = 1;
4483                     nfscl_cleandeleg(dp);
4484                     nfscl_freedeleg(&clp->nfsc_deleg, dp);
4485                 }
4486                 if (igotlock)
4487                     nfsv4_unlock(&clp->nfsc_lock, 0);
4488                 NFSUNLOCKCLSTATE();
4489                 return (retcnt);
4490         }
4491 }
4492
4493 /*
4494  * Look for associated delegation(s) that should be DelegReturned.
4495  */
4496 int
4497 nfscl_renamedeleg(vnode_t fvp, nfsv4stateid_t *fstp, int *gotfdp, vnode_t tvp,
4498     nfsv4stateid_t *tstp, int *gottdp, NFSPROC_T *p)
4499 {
4500         struct nfsclclient *clp;
4501         struct nfscldeleg *dp;
4502         struct nfsclowner *owp;
4503         struct nfscllockowner *lp;
4504         struct nfsmount *nmp;
4505         struct ucred *cred;
4506         struct nfsnode *np;
4507         int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
4508
4509         nmp = VFSTONFS(fvp->v_mount);
4510         *gotfdp = 0;
4511         *gottdp = 0;
4512         NFSLOCKCLSTATE();
4513         /*
4514          * Loop around waiting for:
4515          * - outstanding I/O operations on delegations to complete
4516          * - for a delegation on fvp that has state, lock the client and
4517          *   do a recall
4518          * - return delegation(s) with no state.
4519          */
4520         while (1) {
4521                 clp = nfscl_findcl(nmp);
4522                 if (clp == NULL) {
4523                         NFSUNLOCKCLSTATE();
4524                         return (retcnt);
4525                 }
4526                 np = VTONFS(fvp);
4527                 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4528                     np->n_fhp->nfh_len);
4529                 if (dp != NULL && *gotfdp == 0) {
4530                     /*
4531                      * Wait for outstanding I/O ops to be done.
4532                      */
4533                     if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4534                         if (igotlock) {
4535                             nfsv4_unlock(&clp->nfsc_lock, 0);
4536                             igotlock = 0;
4537                         }
4538                         dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4539                         (void) nfsmsleep(&dp->nfsdl_rwlock,
4540                             NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4541                         continue;
4542                     }
4543                     needsrecall = 0;
4544                     LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4545                         if (!LIST_EMPTY(&owp->nfsow_open)) {
4546                             needsrecall = 1;
4547                             break;
4548                         }
4549                     }
4550                     if (!needsrecall) {
4551                         LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4552                             if (!LIST_EMPTY(&lp->nfsl_lock)) {
4553                                 needsrecall = 1;
4554                                 break;
4555                             }
4556                         }
4557                     }
4558                     if (needsrecall && !triedrecall) {
4559                         dp->nfsdl_flags |= NFSCLDL_DELEGRET;
4560                         islept = 0;
4561                         while (!igotlock) {
4562                             igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
4563                                 &islept, NFSCLSTATEMUTEXPTR, NULL);
4564                             if (islept)
4565                                 break;
4566                         }
4567                         if (islept)
4568                             continue;
4569                         NFSUNLOCKCLSTATE();
4570                         cred = newnfs_getcred();
4571                         newnfs_copycred(&dp->nfsdl_cred, cred);
4572                         (void) nfscl_recalldeleg(clp, nmp, dp, fvp, cred, p, 0);
4573                         NFSFREECRED(cred);
4574                         triedrecall = 1;
4575                         NFSLOCKCLSTATE();
4576                         nfsv4_unlock(&clp->nfsc_lock, 0);
4577                         igotlock = 0;
4578                         continue;
4579                     }
4580                     *fstp = dp->nfsdl_stateid;
4581                     retcnt++;
4582                     *gotfdp = 1;
4583                     nfscl_cleandeleg(dp);
4584                     nfscl_freedeleg(&clp->nfsc_deleg, dp);
4585                 }
4586                 if (igotlock) {
4587                     nfsv4_unlock(&clp->nfsc_lock, 0);
4588                     igotlock = 0;
4589                 }
4590                 if (tvp != NULL) {
4591                     np = VTONFS(tvp);
4592                     dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4593                         np->n_fhp->nfh_len);
4594                     if (dp != NULL && *gottdp == 0) {
4595                         /*
4596                          * Wait for outstanding I/O ops to be done.
4597                          */
4598                         if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4599                             dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4600                             (void) nfsmsleep(&dp->nfsdl_rwlock,
4601                                 NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4602                             continue;
4603                         }
4604                         LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4605                             if (!LIST_EMPTY(&owp->nfsow_open)) {
4606                                 NFSUNLOCKCLSTATE();
4607                                 return (retcnt);
4608                             }
4609                         }
4610                         LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4611                             if (!LIST_EMPTY(&lp->nfsl_lock)) {
4612                                 NFSUNLOCKCLSTATE();
4613                                 return (retcnt);
4614                             }
4615                         }
4616                         *tstp = dp->nfsdl_stateid;
4617                         retcnt++;
4618                         *gottdp = 1;
4619                         nfscl_cleandeleg(dp);
4620                         nfscl_freedeleg(&clp->nfsc_deleg, dp);
4621                     }
4622                 }
4623                 NFSUNLOCKCLSTATE();
4624                 return (retcnt);
4625         }
4626 }
4627
4628 /*
4629  * Get a reference on the clientid associated with the mount point.
4630  * Return 1 if success, 0 otherwise.
4631  */
4632 int
4633 nfscl_getref(struct nfsmount *nmp)
4634 {
4635         struct nfsclclient *clp;
4636
4637         NFSLOCKCLSTATE();
4638         clp = nfscl_findcl(nmp);
4639         if (clp == NULL) {
4640                 NFSUNLOCKCLSTATE();
4641                 return (0);
4642         }
4643         nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, NULL);
4644         NFSUNLOCKCLSTATE();
4645         return (1);
4646 }
4647
4648 /*
4649  * Release a reference on a clientid acquired with the above call.
4650  */
4651 void
4652 nfscl_relref(struct nfsmount *nmp)
4653 {
4654         struct nfsclclient *clp;
4655
4656         NFSLOCKCLSTATE();
4657         clp = nfscl_findcl(nmp);
4658         if (clp == NULL) {
4659                 NFSUNLOCKCLSTATE();
4660                 return;
4661         }
4662         nfsv4_relref(&clp->nfsc_lock);
4663         NFSUNLOCKCLSTATE();
4664 }
4665
4666 /*
4667  * Save the size attribute in the delegation, since the nfsnode
4668  * is going away.
4669  */
4670 void
4671 nfscl_reclaimnode(vnode_t vp)
4672 {
4673         struct nfsclclient *clp;
4674         struct nfscldeleg *dp;
4675         struct nfsnode *np = VTONFS(vp);
4676         struct nfsmount *nmp;
4677
4678         nmp = VFSTONFS(vp->v_mount);
4679         if (!NFSHASNFSV4(nmp))
4680                 return;
4681         NFSLOCKCLSTATE();
4682         clp = nfscl_findcl(nmp);
4683         if (clp == NULL) {
4684                 NFSUNLOCKCLSTATE();
4685                 return;
4686         }
4687         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4688         if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE))
4689                 dp->nfsdl_size = np->n_size;
4690         NFSUNLOCKCLSTATE();
4691 }
4692
4693 /*
4694  * Get the saved size attribute in the delegation, since it is a
4695  * newly allocated nfsnode.
4696  */
4697 void
4698 nfscl_newnode(vnode_t vp)
4699 {
4700         struct nfsclclient *clp;
4701         struct nfscldeleg *dp;
4702         struct nfsnode *np = VTONFS(vp);
4703         struct nfsmount *nmp;
4704
4705         nmp = VFSTONFS(vp->v_mount);
4706         if (!NFSHASNFSV4(nmp))
4707                 return;
4708         NFSLOCKCLSTATE();
4709         clp = nfscl_findcl(nmp);
4710         if (clp == NULL) {
4711                 NFSUNLOCKCLSTATE();
4712                 return;
4713         }
4714         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4715         if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE))
4716                 np->n_size = dp->nfsdl_size;
4717         NFSUNLOCKCLSTATE();
4718 }
4719
4720 /*
4721  * If there is a valid write delegation for this file, set the modtime
4722  * to the local clock time.
4723  */
4724 void
4725 nfscl_delegmodtime(vnode_t vp)
4726 {
4727         struct nfsclclient *clp;
4728         struct nfscldeleg *dp;
4729         struct nfsnode *np = VTONFS(vp);
4730         struct nfsmount *nmp;
4731
4732         nmp = VFSTONFS(vp->v_mount);
4733         if (!NFSHASNFSV4(nmp))
4734                 return;
4735         NFSLOCKCLSTATE();
4736         clp = nfscl_findcl(nmp);
4737         if (clp == NULL) {
4738                 NFSUNLOCKCLSTATE();
4739                 return;
4740         }
4741         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4742         if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE)) {
4743                 nanotime(&dp->nfsdl_modtime);
4744                 dp->nfsdl_flags |= NFSCLDL_MODTIMESET;
4745         }
4746         NFSUNLOCKCLSTATE();
4747 }
4748
4749 /*
4750  * If there is a valid write delegation for this file with a modtime set,
4751  * put that modtime in mtime.
4752  */
4753 void
4754 nfscl_deleggetmodtime(vnode_t vp, struct timespec *mtime)
4755 {
4756         struct nfsclclient *clp;
4757         struct nfscldeleg *dp;
4758         struct nfsnode *np = VTONFS(vp);
4759         struct nfsmount *nmp;
4760
4761         nmp = VFSTONFS(vp->v_mount);
4762         if (!NFSHASNFSV4(nmp))
4763                 return;
4764         NFSLOCKCLSTATE();
4765         clp = nfscl_findcl(nmp);
4766         if (clp == NULL) {
4767                 NFSUNLOCKCLSTATE();
4768                 return;
4769         }
4770         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4771         if (dp != NULL &&
4772             (dp->nfsdl_flags & (NFSCLDL_WRITE | NFSCLDL_MODTIMESET)) ==
4773             (NFSCLDL_WRITE | NFSCLDL_MODTIMESET))
4774                 *mtime = dp->nfsdl_modtime;
4775         NFSUNLOCKCLSTATE();
4776 }
4777
4778 static int
4779 nfscl_errmap(struct nfsrv_descript *nd, u_int32_t minorvers)
4780 {
4781         short *defaulterrp, *errp;
4782
4783         if (!nd->nd_repstat)
4784                 return (0);
4785         if (nd->nd_procnum == NFSPROC_NOOP)
4786                 return (txdr_unsigned(nd->nd_repstat & 0xffff));
4787         if (nd->nd_repstat == EBADRPC)
4788                 return (txdr_unsigned(NFSERR_BADXDR));
4789         if (nd->nd_repstat == NFSERR_MINORVERMISMATCH ||
4790             nd->nd_repstat == NFSERR_OPILLEGAL)
4791                 return (txdr_unsigned(nd->nd_repstat));
4792         if (nd->nd_repstat >= NFSERR_BADIOMODE && nd->nd_repstat < 20000 &&
4793             minorvers > NFSV4_MINORVERSION) {
4794                 /* NFSv4.n error. */
4795                 return (txdr_unsigned(nd->nd_repstat));
4796         }
4797         if (nd->nd_procnum < NFSV4OP_CBNOPS)
4798                 errp = defaulterrp = nfscl_cberrmap[nd->nd_procnum];
4799         else
4800                 return (txdr_unsigned(nd->nd_repstat));
4801         while (*++errp)
4802                 if (*errp == (short)nd->nd_repstat)
4803                         return (txdr_unsigned(nd->nd_repstat));
4804         return (txdr_unsigned(*defaulterrp));
4805 }
4806
4807 /*
4808  * Called to find/add a layout to a client.
4809  * This function returns the layout with a refcnt (shared lock) upon
4810  * success (returns 0) or with no lock/refcnt on the layout when an
4811  * error is returned.
4812  * If a layout is passed in via lypp, it is locked (exclusively locked).
4813  */
4814 int
4815 nfscl_layout(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen,
4816     nfsv4stateid_t *stateidp, int layouttype, int retonclose,
4817     struct nfsclflayouthead *fhlp, struct nfscllayout **lypp,
4818     struct ucred *cred, NFSPROC_T *p)
4819 {
4820         struct nfsclclient *clp;
4821         struct nfscllayout *lyp, *tlyp;
4822         struct nfsclflayout *flp;
4823         struct nfsnode *np = VTONFS(vp);
4824         mount_t mp;
4825         int layout_passed_in;
4826
4827         mp = nmp->nm_mountp;
4828         layout_passed_in = 1;
4829         tlyp = NULL;
4830         lyp = *lypp;
4831         if (lyp == NULL) {
4832                 layout_passed_in = 0;
4833                 tlyp = malloc(sizeof(*tlyp) + fhlen - 1, M_NFSLAYOUT,
4834                     M_WAITOK | M_ZERO);
4835         }
4836
4837         NFSLOCKCLSTATE();
4838         clp = nmp->nm_clp;
4839         if (clp == NULL) {
4840                 if (layout_passed_in != 0)
4841                         nfsv4_unlock(&lyp->nfsly_lock, 0);
4842                 NFSUNLOCKCLSTATE();
4843                 if (tlyp != NULL)
4844                         free(tlyp, M_NFSLAYOUT);
4845                 return (EPERM);
4846         }
4847         if (lyp == NULL) {
4848                 /*
4849                  * Although no lyp was passed in, another thread might have
4850                  * allocated one. If one is found, just increment it's ref
4851                  * count and return it.
4852                  */
4853                 lyp = nfscl_findlayout(clp, fhp, fhlen);
4854                 if (lyp == NULL) {
4855                         lyp = tlyp;
4856                         tlyp = NULL;
4857                         lyp->nfsly_stateid.seqid = stateidp->seqid;
4858                         lyp->nfsly_stateid.other[0] = stateidp->other[0];
4859                         lyp->nfsly_stateid.other[1] = stateidp->other[1];
4860                         lyp->nfsly_stateid.other[2] = stateidp->other[2];
4861                         lyp->nfsly_lastbyte = 0;
4862                         LIST_INIT(&lyp->nfsly_flayread);
4863                         LIST_INIT(&lyp->nfsly_flayrw);
4864                         LIST_INIT(&lyp->nfsly_recall);
4865                         lyp->nfsly_filesid[0] = np->n_vattr.na_filesid[0];
4866                         lyp->nfsly_filesid[1] = np->n_vattr.na_filesid[1];
4867                         lyp->nfsly_clp = clp;
4868                         if (layouttype == NFSLAYOUT_FLEXFILE)
4869                                 lyp->nfsly_flags = NFSLY_FLEXFILE;
4870                         else
4871                                 lyp->nfsly_flags = NFSLY_FILES;
4872                         if (retonclose != 0)
4873                                 lyp->nfsly_flags |= NFSLY_RETONCLOSE;
4874                         lyp->nfsly_fhlen = fhlen;
4875                         NFSBCOPY(fhp, lyp->nfsly_fh, fhlen);
4876                         TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
4877                         LIST_INSERT_HEAD(NFSCLLAYOUTHASH(clp, fhp, fhlen), lyp,
4878                             nfsly_hash);
4879                         lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
4880                         nfscl_layoutcnt++;
4881                 } else {
4882                         if (retonclose != 0)
4883                                 lyp->nfsly_flags |= NFSLY_RETONCLOSE;
4884                         TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list);
4885                         TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
4886                         lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
4887                 }
4888                 nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
4889                 if (NFSCL_FORCEDISM(mp)) {
4890                         NFSUNLOCKCLSTATE();
4891                         if (tlyp != NULL)
4892                                 free(tlyp, M_NFSLAYOUT);
4893                         return (EPERM);
4894                 }
4895                 *lypp = lyp;
4896         } else
4897                 lyp->nfsly_stateid.seqid = stateidp->seqid;
4898
4899         /* Merge the new list of File Layouts into the list. */
4900         flp = LIST_FIRST(fhlp);
4901         if (flp != NULL) {
4902                 if (flp->nfsfl_iomode == NFSLAYOUTIOMODE_READ)
4903                         nfscl_mergeflayouts(&lyp->nfsly_flayread, fhlp);
4904                 else
4905                         nfscl_mergeflayouts(&lyp->nfsly_flayrw, fhlp);
4906         }
4907         if (layout_passed_in != 0)
4908                 nfsv4_unlock(&lyp->nfsly_lock, 1);
4909         NFSUNLOCKCLSTATE();
4910         if (tlyp != NULL)
4911                 free(tlyp, M_NFSLAYOUT);
4912         return (0);
4913 }
4914
4915 /*
4916  * Search for a layout by MDS file handle.
4917  * If one is found, it is returned with a refcnt (shared lock) iff
4918  * retflpp returned non-NULL and locked (exclusive locked) iff retflpp is
4919  * returned NULL.
4920  */
4921 struct nfscllayout *
4922 nfscl_getlayout(struct nfsclclient *clp, uint8_t *fhp, int fhlen,
4923     uint64_t off, struct nfsclflayout **retflpp, int *recalledp)
4924 {
4925         struct nfscllayout *lyp;
4926         mount_t mp;
4927         int error, igotlock;
4928
4929         mp = clp->nfsc_nmp->nm_mountp;
4930         *recalledp = 0;
4931         *retflpp = NULL;
4932         NFSLOCKCLSTATE();
4933         lyp = nfscl_findlayout(clp, fhp, fhlen);
4934         if (lyp != NULL) {
4935                 if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
4936                         TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list);
4937                         TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
4938                         lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
4939                         error = nfscl_findlayoutforio(lyp, off,
4940                             NFSV4OPEN_ACCESSREAD, retflpp);
4941                         if (error == 0)
4942                                 nfsv4_getref(&lyp->nfsly_lock, NULL,
4943                                     NFSCLSTATEMUTEXPTR, mp);
4944                         else {
4945                                 do {
4946                                         igotlock = nfsv4_lock(&lyp->nfsly_lock,
4947                                             1, NULL, NFSCLSTATEMUTEXPTR, mp);
4948                                 } while (igotlock == 0 && !NFSCL_FORCEDISM(mp));
4949                                 *retflpp = NULL;
4950                         }
4951                         if (NFSCL_FORCEDISM(mp)) {
4952                                 lyp = NULL;
4953                                 *recalledp = 1;
4954                         }
4955                 } else {
4956                         lyp = NULL;
4957                         *recalledp = 1;
4958                 }
4959         }
4960         NFSUNLOCKCLSTATE();
4961         return (lyp);
4962 }
4963
4964 /*
4965  * Search for a layout by MDS file handle. If one is found, mark in to be
4966  * recalled, if it already marked "return on close".
4967  */
4968 static void
4969 nfscl_retoncloselayout(vnode_t vp, struct nfsclclient *clp, uint8_t *fhp,
4970     int fhlen, struct nfsclrecalllayout **recallpp)
4971 {
4972         struct nfscllayout *lyp;
4973         uint32_t iomode;
4974
4975         if (vp->v_type != VREG || !NFSHASPNFS(VFSTONFS(vp->v_mount)) ||
4976             nfscl_enablecallb == 0 || nfs_numnfscbd == 0 ||
4977             (VTONFS(vp)->n_flag & NNOLAYOUT) != 0)
4978                 return;
4979         lyp = nfscl_findlayout(clp, fhp, fhlen);
4980         if (lyp != NULL && (lyp->nfsly_flags & (NFSLY_RETONCLOSE |
4981             NFSLY_RECALL)) == NFSLY_RETONCLOSE) {
4982                 iomode = 0;
4983                 if (!LIST_EMPTY(&lyp->nfsly_flayread))
4984                         iomode |= NFSLAYOUTIOMODE_READ;
4985                 if (!LIST_EMPTY(&lyp->nfsly_flayrw))
4986                         iomode |= NFSLAYOUTIOMODE_RW;
4987                 (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE, lyp, iomode,
4988                     0, UINT64_MAX, lyp->nfsly_stateid.seqid, 0, 0, NULL,
4989                     *recallpp);
4990                 NFSCL_DEBUG(4, "retoncls recall iomode=%d\n", iomode);
4991                 *recallpp = NULL;
4992         }
4993 }
4994
4995 /*
4996  * Mark the layout to be recalled and with an error.
4997  * Also, disable the dsp from further use.
4998  */
4999 void
5000 nfscl_dserr(uint32_t op, uint32_t stat, struct nfscldevinfo *dp,
5001     struct nfscllayout *lyp, struct nfsclds *dsp)
5002 {
5003         struct nfsclrecalllayout *recallp;
5004         uint32_t iomode;
5005
5006         printf("DS being disabled, error=%d\n", stat);
5007         /* Set up the return of the layout. */
5008         recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_WAITOK);
5009         iomode = 0;
5010         NFSLOCKCLSTATE();
5011         if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
5012                 if (!LIST_EMPTY(&lyp->nfsly_flayread))
5013                         iomode |= NFSLAYOUTIOMODE_READ;
5014                 if (!LIST_EMPTY(&lyp->nfsly_flayrw))
5015                         iomode |= NFSLAYOUTIOMODE_RW;
5016                 (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE, lyp, iomode,
5017                     0, UINT64_MAX, lyp->nfsly_stateid.seqid, stat, op,
5018                     dp->nfsdi_deviceid, recallp);
5019                 NFSUNLOCKCLSTATE();
5020                 NFSCL_DEBUG(4, "nfscl_dserr recall iomode=%d\n", iomode);
5021         } else {
5022                 NFSUNLOCKCLSTATE();
5023                 free(recallp, M_NFSLAYRECALL);
5024         }
5025
5026         /* And shut the TCP connection down. */
5027         nfscl_cancelreqs(dsp);
5028 }
5029
5030 /*
5031  * Cancel all RPCs for this "dsp" by closing the connection.
5032  * Also, mark the session as defunct.
5033  * If NFSCLDS_SAMECONN is set, the connection is shared with other DSs and
5034  * cannot be shut down.
5035  */
5036 void
5037 nfscl_cancelreqs(struct nfsclds *dsp)
5038 {
5039         struct __rpc_client *cl;
5040         static int non_event;
5041
5042         NFSLOCKDS(dsp);
5043         if ((dsp->nfsclds_flags & (NFSCLDS_CLOSED | NFSCLDS_SAMECONN)) == 0 &&
5044             dsp->nfsclds_sockp != NULL &&
5045             dsp->nfsclds_sockp->nr_client != NULL) {
5046                 dsp->nfsclds_flags |= NFSCLDS_CLOSED;
5047                 cl = dsp->nfsclds_sockp->nr_client;
5048                 dsp->nfsclds_sess.nfsess_defunct = 1;
5049                 NFSUNLOCKDS(dsp);
5050                 CLNT_CLOSE(cl);
5051                 /*
5052                  * This 1sec sleep is done to reduce the number of reconnect
5053                  * attempts made on the DS while it has failed.
5054                  */
5055                 tsleep(&non_event, PVFS, "ndscls", hz);
5056                 return;
5057         }
5058         NFSUNLOCKDS(dsp);
5059 }
5060
5061 /*
5062  * Dereference a layout.
5063  */
5064 void
5065 nfscl_rellayout(struct nfscllayout *lyp, int exclocked)
5066 {
5067
5068         NFSLOCKCLSTATE();
5069         if (exclocked != 0)
5070                 nfsv4_unlock(&lyp->nfsly_lock, 0);
5071         else
5072                 nfsv4_relref(&lyp->nfsly_lock);
5073         NFSUNLOCKCLSTATE();
5074 }
5075
5076 /*
5077  * Search for a devinfo by deviceid. If one is found, return it after
5078  * acquiring a reference count on it.
5079  */
5080 struct nfscldevinfo *
5081 nfscl_getdevinfo(struct nfsclclient *clp, uint8_t *deviceid,
5082     struct nfscldevinfo *dip)
5083 {
5084
5085         NFSLOCKCLSTATE();
5086         if (dip == NULL)
5087                 dip = nfscl_finddevinfo(clp, deviceid);
5088         if (dip != NULL)
5089                 dip->nfsdi_refcnt++;
5090         NFSUNLOCKCLSTATE();
5091         return (dip);
5092 }
5093
5094 /*
5095  * Dereference a devinfo structure.
5096  */
5097 static void
5098 nfscl_reldevinfo_locked(struct nfscldevinfo *dip)
5099 {
5100
5101         dip->nfsdi_refcnt--;
5102         if (dip->nfsdi_refcnt == 0)
5103                 wakeup(&dip->nfsdi_refcnt);
5104 }
5105
5106 /*
5107  * Dereference a devinfo structure.
5108  */
5109 void
5110 nfscl_reldevinfo(struct nfscldevinfo *dip)
5111 {
5112
5113         NFSLOCKCLSTATE();
5114         nfscl_reldevinfo_locked(dip);
5115         NFSUNLOCKCLSTATE();
5116 }
5117
5118 /*
5119  * Find a layout for this file handle. Return NULL upon failure.
5120  */
5121 static struct nfscllayout *
5122 nfscl_findlayout(struct nfsclclient *clp, u_int8_t *fhp, int fhlen)
5123 {
5124         struct nfscllayout *lyp;
5125
5126         LIST_FOREACH(lyp, NFSCLLAYOUTHASH(clp, fhp, fhlen), nfsly_hash)
5127                 if (lyp->nfsly_fhlen == fhlen &&
5128                     !NFSBCMP(lyp->nfsly_fh, fhp, fhlen))
5129                         break;
5130         return (lyp);
5131 }
5132
5133 /*
5134  * Find a devinfo for this deviceid. Return NULL upon failure.
5135  */
5136 static struct nfscldevinfo *
5137 nfscl_finddevinfo(struct nfsclclient *clp, uint8_t *deviceid)
5138 {
5139         struct nfscldevinfo *dip;
5140
5141         LIST_FOREACH(dip, &clp->nfsc_devinfo, nfsdi_list)
5142                 if (NFSBCMP(dip->nfsdi_deviceid, deviceid, NFSX_V4DEVICEID)
5143                     == 0)
5144                         break;
5145         return (dip);
5146 }
5147
5148 /*
5149  * Merge the new file layout list into the main one, maintaining it in
5150  * increasing offset order.
5151  */
5152 static void
5153 nfscl_mergeflayouts(struct nfsclflayouthead *fhlp,
5154     struct nfsclflayouthead *newfhlp)
5155 {
5156         struct nfsclflayout *flp, *nflp, *prevflp, *tflp;
5157
5158         flp = LIST_FIRST(fhlp);
5159         prevflp = NULL;
5160         LIST_FOREACH_SAFE(nflp, newfhlp, nfsfl_list, tflp) {
5161                 while (flp != NULL && flp->nfsfl_off < nflp->nfsfl_off) {
5162                         prevflp = flp;
5163                         flp = LIST_NEXT(flp, nfsfl_list);
5164                 }
5165                 if (prevflp == NULL)
5166                         LIST_INSERT_HEAD(fhlp, nflp, nfsfl_list);
5167                 else
5168                         LIST_INSERT_AFTER(prevflp, nflp, nfsfl_list);
5169                 prevflp = nflp;
5170         }
5171 }
5172
5173 /*
5174  * Add this nfscldevinfo to the client, if it doesn't already exist.
5175  * This function consumes the structure pointed at by dip, if not NULL.
5176  */
5177 int
5178 nfscl_adddevinfo(struct nfsmount *nmp, struct nfscldevinfo *dip, int ind,
5179     struct nfsclflayout *flp)
5180 {
5181         struct nfsclclient *clp;
5182         struct nfscldevinfo *tdip;
5183         uint8_t *dev;
5184
5185         NFSLOCKCLSTATE();
5186         clp = nmp->nm_clp;
5187         if (clp == NULL) {
5188                 NFSUNLOCKCLSTATE();
5189                 if (dip != NULL)
5190                         free(dip, M_NFSDEVINFO);
5191                 return (ENODEV);
5192         }
5193         if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5194                 dev = flp->nfsfl_dev;
5195         else
5196                 dev = flp->nfsfl_ffm[ind].dev;
5197         tdip = nfscl_finddevinfo(clp, dev);
5198         if (tdip != NULL) {
5199                 tdip->nfsdi_layoutrefs++;
5200                 if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5201                         flp->nfsfl_devp = tdip;
5202                 else
5203                         flp->nfsfl_ffm[ind].devp = tdip;
5204                 nfscl_reldevinfo_locked(tdip);
5205                 NFSUNLOCKCLSTATE();
5206                 if (dip != NULL)
5207                         free(dip, M_NFSDEVINFO);
5208                 return (0);
5209         }
5210         if (dip != NULL) {
5211                 LIST_INSERT_HEAD(&clp->nfsc_devinfo, dip, nfsdi_list);
5212                 dip->nfsdi_layoutrefs = 1;
5213                 if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5214                         flp->nfsfl_devp = dip;
5215                 else
5216                         flp->nfsfl_ffm[ind].devp = dip;
5217         }
5218         NFSUNLOCKCLSTATE();
5219         if (dip == NULL)
5220                 return (ENODEV);
5221         return (0);
5222 }
5223
5224 /*
5225  * Free up a layout structure and associated file layout structure(s).
5226  */
5227 void
5228 nfscl_freelayout(struct nfscllayout *layp)
5229 {
5230         struct nfsclflayout *flp, *nflp;
5231         struct nfsclrecalllayout *rp, *nrp;
5232
5233         LIST_FOREACH_SAFE(flp, &layp->nfsly_flayread, nfsfl_list, nflp) {
5234                 LIST_REMOVE(flp, nfsfl_list);
5235                 nfscl_freeflayout(flp);
5236         }
5237         LIST_FOREACH_SAFE(flp, &layp->nfsly_flayrw, nfsfl_list, nflp) {
5238                 LIST_REMOVE(flp, nfsfl_list);
5239                 nfscl_freeflayout(flp);
5240         }
5241         LIST_FOREACH_SAFE(rp, &layp->nfsly_recall, nfsrecly_list, nrp) {
5242                 LIST_REMOVE(rp, nfsrecly_list);
5243                 free(rp, M_NFSLAYRECALL);
5244         }
5245         nfscl_layoutcnt--;
5246         free(layp, M_NFSLAYOUT);
5247 }
5248
5249 /*
5250  * Free up a file layout structure.
5251  */
5252 void
5253 nfscl_freeflayout(struct nfsclflayout *flp)
5254 {
5255         int i, j;
5256
5257         if ((flp->nfsfl_flags & NFSFL_FILE) != 0) {
5258                 for (i = 0; i < flp->nfsfl_fhcnt; i++)
5259                         free(flp->nfsfl_fh[i], M_NFSFH);
5260                 if (flp->nfsfl_devp != NULL)
5261                         flp->nfsfl_devp->nfsdi_layoutrefs--;
5262         }
5263         if ((flp->nfsfl_flags & NFSFL_FLEXFILE) != 0)
5264                 for (i = 0; i < flp->nfsfl_mirrorcnt; i++) {
5265                         for (j = 0; j < flp->nfsfl_ffm[i].fhcnt; j++)
5266                                 free(flp->nfsfl_ffm[i].fh[j], M_NFSFH);
5267                         if (flp->nfsfl_ffm[i].devp != NULL)
5268                                 flp->nfsfl_ffm[i].devp->nfsdi_layoutrefs--;
5269                 }
5270         free(flp, M_NFSFLAYOUT);
5271 }
5272
5273 /*
5274  * Free up a file layout devinfo structure.
5275  */
5276 void
5277 nfscl_freedevinfo(struct nfscldevinfo *dip)
5278 {
5279
5280         free(dip, M_NFSDEVINFO);
5281 }
5282
5283 /*
5284  * Mark any layouts that match as recalled.
5285  */
5286 static int
5287 nfscl_layoutrecall(int recalltype, struct nfscllayout *lyp, uint32_t iomode,
5288     uint64_t off, uint64_t len, uint32_t stateseqid, uint32_t stat, uint32_t op,
5289     char *devid, struct nfsclrecalllayout *recallp)
5290 {
5291         struct nfsclrecalllayout *rp, *orp;
5292
5293         recallp->nfsrecly_recalltype = recalltype;
5294         recallp->nfsrecly_iomode = iomode;
5295         recallp->nfsrecly_stateseqid = stateseqid;
5296         recallp->nfsrecly_off = off;
5297         recallp->nfsrecly_len = len;
5298         recallp->nfsrecly_stat = stat;
5299         recallp->nfsrecly_op = op;
5300         if (devid != NULL)
5301                 NFSBCOPY(devid, recallp->nfsrecly_devid, NFSX_V4DEVICEID);
5302         /*
5303          * Order the list as file returns first, followed by fsid and any
5304          * returns, both in increasing stateseqid order.
5305          * Note that the seqids wrap around, so 1 is after 0xffffffff.
5306          * (I'm not sure this is correct because I find RFC5661 confusing
5307          *  on this, but hopefully it will work ok.)
5308          */
5309         orp = NULL;
5310         LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) {
5311                 orp = rp;
5312                 if ((recalltype == NFSLAYOUTRETURN_FILE &&
5313                      (rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE ||
5314                       nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) ||
5315                     (recalltype != NFSLAYOUTRETURN_FILE &&
5316                      rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE &&
5317                      nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) {
5318                         LIST_INSERT_BEFORE(rp, recallp, nfsrecly_list);
5319                         break;
5320                 }
5321
5322                 /*
5323                  * Put any error return on all the file returns that will
5324                  * preceed this one.
5325                  */
5326                 if (rp->nfsrecly_recalltype == NFSLAYOUTRETURN_FILE &&
5327                    stat != 0 && rp->nfsrecly_stat == 0) {
5328                         rp->nfsrecly_stat = stat;
5329                         rp->nfsrecly_op = op;
5330                         if (devid != NULL)
5331                                 NFSBCOPY(devid, rp->nfsrecly_devid,
5332                                     NFSX_V4DEVICEID);
5333                 }
5334         }
5335         if (rp == NULL) {
5336                 if (orp == NULL)
5337                         LIST_INSERT_HEAD(&lyp->nfsly_recall, recallp,
5338                             nfsrecly_list);
5339                 else
5340                         LIST_INSERT_AFTER(orp, recallp, nfsrecly_list);
5341         }
5342         lyp->nfsly_flags |= NFSLY_RECALL;
5343         wakeup(lyp->nfsly_clp);
5344         return (0);
5345 }
5346
5347 /*
5348  * Compare the two seqids for ordering. The trick is that the seqids can
5349  * wrap around from 0xffffffff->0, so check for the cases where one
5350  * has wrapped around.
5351  * Return 1 if seqid1 comes before seqid2, 0 otherwise.
5352  */
5353 static int
5354 nfscl_seq(uint32_t seqid1, uint32_t seqid2)
5355 {
5356
5357         if (seqid2 > seqid1 && (seqid2 - seqid1) >= 0x7fffffff)
5358                 /* seqid2 has wrapped around. */
5359                 return (0);
5360         if (seqid1 > seqid2 && (seqid1 - seqid2) >= 0x7fffffff)
5361                 /* seqid1 has wrapped around. */
5362                 return (1);
5363         if (seqid1 <= seqid2)
5364                 return (1);
5365         return (0);
5366 }
5367
5368 /*
5369  * Do a layout return for each of the recalls.
5370  */
5371 static void
5372 nfscl_layoutreturn(struct nfsmount *nmp, struct nfscllayout *lyp,
5373     struct ucred *cred, NFSPROC_T *p)
5374 {
5375         struct nfsclrecalllayout *rp;
5376         nfsv4stateid_t stateid;
5377         int layouttype;
5378
5379         NFSBCOPY(lyp->nfsly_stateid.other, stateid.other, NFSX_STATEIDOTHER);
5380         stateid.seqid = lyp->nfsly_stateid.seqid;
5381         if ((lyp->nfsly_flags & NFSLY_FILES) != 0)
5382                 layouttype = NFSLAYOUT_NFSV4_1_FILES;
5383         else
5384                 layouttype = NFSLAYOUT_FLEXFILE;
5385         LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) {
5386                 (void)nfsrpc_layoutreturn(nmp, lyp->nfsly_fh,
5387                     lyp->nfsly_fhlen, 0, layouttype,
5388                     rp->nfsrecly_iomode, rp->nfsrecly_recalltype,
5389                     rp->nfsrecly_off, rp->nfsrecly_len,
5390                     &stateid, cred, p, rp->nfsrecly_stat, rp->nfsrecly_op,
5391                     rp->nfsrecly_devid);
5392         }
5393 }
5394
5395 /*
5396  * Do the layout commit for a file layout.
5397  */
5398 static void
5399 nfscl_dolayoutcommit(struct nfsmount *nmp, struct nfscllayout *lyp,
5400     struct ucred *cred, NFSPROC_T *p)
5401 {
5402         struct nfsclflayout *flp;
5403         uint64_t len;
5404         int error, layouttype;
5405
5406         if ((lyp->nfsly_flags & NFSLY_FILES) != 0)
5407                 layouttype = NFSLAYOUT_NFSV4_1_FILES;
5408         else
5409                 layouttype = NFSLAYOUT_FLEXFILE;
5410         LIST_FOREACH(flp, &lyp->nfsly_flayrw, nfsfl_list) {
5411                 if (layouttype == NFSLAYOUT_FLEXFILE &&
5412                     (flp->nfsfl_fflags & NFSFLEXFLAG_NO_LAYOUTCOMMIT) != 0) {
5413                         NFSCL_DEBUG(4, "Flex file: no layoutcommit\n");
5414                         /* If not supported, don't bother doing it. */
5415                         NFSLOCKMNT(nmp);
5416                         nmp->nm_state |= NFSSTA_NOLAYOUTCOMMIT;
5417                         NFSUNLOCKMNT(nmp);
5418                         break;
5419                 } else if (flp->nfsfl_off <= lyp->nfsly_lastbyte) {
5420                         len = flp->nfsfl_end - flp->nfsfl_off;
5421                         error = nfsrpc_layoutcommit(nmp, lyp->nfsly_fh,
5422                             lyp->nfsly_fhlen, 0, flp->nfsfl_off, len,
5423                             lyp->nfsly_lastbyte, &lyp->nfsly_stateid,
5424                             layouttype, cred, p, NULL);
5425                         NFSCL_DEBUG(4, "layoutcommit err=%d\n", error);
5426                         if (error == NFSERR_NOTSUPP) {
5427                                 /* If not supported, don't bother doing it. */
5428                                 NFSLOCKMNT(nmp);
5429                                 nmp->nm_state |= NFSSTA_NOLAYOUTCOMMIT;
5430                                 NFSUNLOCKMNT(nmp);
5431                                 break;
5432                         }
5433                 }
5434         }
5435 }
5436
5437 /*
5438  * Commit all layouts for a file (vnode).
5439  */
5440 int
5441 nfscl_layoutcommit(vnode_t vp, NFSPROC_T *p)
5442 {
5443         struct nfsclclient *clp;
5444         struct nfscllayout *lyp;
5445         struct nfsnode *np = VTONFS(vp);
5446         mount_t mp;
5447         struct nfsmount *nmp;
5448
5449         mp = vp->v_mount;
5450         nmp = VFSTONFS(mp);
5451         if (NFSHASNOLAYOUTCOMMIT(nmp))
5452                 return (0);
5453         NFSLOCKCLSTATE();
5454         clp = nmp->nm_clp;
5455         if (clp == NULL) {
5456                 NFSUNLOCKCLSTATE();
5457                 return (EPERM);
5458         }
5459         lyp = nfscl_findlayout(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
5460         if (lyp == NULL) {
5461                 NFSUNLOCKCLSTATE();
5462                 return (EPERM);
5463         }
5464         nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
5465         if (NFSCL_FORCEDISM(mp)) {
5466                 NFSUNLOCKCLSTATE();
5467                 return (EPERM);
5468         }
5469 tryagain:
5470         if ((lyp->nfsly_flags & NFSLY_WRITTEN) != 0) {
5471                 lyp->nfsly_flags &= ~NFSLY_WRITTEN;
5472                 NFSUNLOCKCLSTATE();
5473                 NFSCL_DEBUG(4, "do layoutcommit2\n");
5474                 nfscl_dolayoutcommit(clp->nfsc_nmp, lyp, NFSPROCCRED(p), p);
5475                 NFSLOCKCLSTATE();
5476                 goto tryagain;
5477         }
5478         nfsv4_relref(&lyp->nfsly_lock);
5479         NFSUNLOCKCLSTATE();
5480         return (0);
5481 }