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 bool nfscl_checkown(struct nfsclowner *, struct nfsclopen *, uint8_t *,
 106     uint8_t *, struct nfscllockowner **, struct nfsclopen **,
 107     struct nfsclopen **);
 108 static void nfscl_clrelease(struct nfsclclient *);
 109 static void nfscl_cleanclient(struct nfsclclient *);
 110 static void nfscl_expireclient(struct nfsclclient *, struct nfsmount *,
 111     struct ucred *, NFSPROC_T *);
 112 static int nfscl_expireopen(struct nfsclclient *, struct nfsclopen *,
 113     struct nfsmount *, struct ucred *, NFSPROC_T *);
 114 static void nfscl_recover(struct nfsclclient *, bool *, struct ucred *,
 115     NFSPROC_T *);
 116 static void nfscl_insertlock(struct nfscllockowner *, struct nfscllock *,
 117     struct nfscllock *, int);
 118 static int nfscl_updatelock(struct nfscllockowner *, struct nfscllock **,
 119     struct nfscllock **, int);
 120 static void nfscl_delegreturnall(struct nfsclclient *, NFSPROC_T *);
 121 static u_int32_t nfscl_nextcbident(void);
 122 static mount_t nfscl_getmnt(int, uint8_t *, u_int32_t, struct nfsclclient **);
 123 static struct nfsclclient *nfscl_getclnt(u_int32_t);
 124 static struct nfsclclient *nfscl_getclntsess(uint8_t *);
 125 static struct nfscldeleg *nfscl_finddeleg(struct nfsclclient *, u_int8_t *,
 126     int);
 127 static void nfscl_retoncloselayout(vnode_t, struct nfsclclient *, uint8_t *,
 128     int, struct nfsclrecalllayout **);
 129 static void nfscl_reldevinfo_locked(struct nfscldevinfo *);
 130 static struct nfscllayout *nfscl_findlayout(struct nfsclclient *, u_int8_t *,
 131     int);
 132 static struct nfscldevinfo *nfscl_finddevinfo(struct nfsclclient *, uint8_t *);
 133 static int nfscl_checkconflict(struct nfscllockownerhead *, struct nfscllock *,
 134     u_int8_t *, struct nfscllock **);
 135 static void nfscl_freealllocks(struct nfscllockownerhead *, int);
 136 static int nfscl_localconflict(struct nfsclclient *, u_int8_t *, int,
 137     struct nfscllock *, u_int8_t *, struct nfscldeleg *, struct nfscllock **);
 138 static void nfscl_newopen(struct nfsclclient *, struct nfscldeleg *,
 139     struct nfsclowner **, struct nfsclowner **, struct nfsclopen **,
 140     struct nfsclopen **, u_int8_t *, u_int8_t *, int, struct ucred *, int *);
 141 static int nfscl_moveopen(vnode_t , struct nfsclclient *,
 142     struct nfsmount *, struct nfsclopen *, struct nfsclowner *,
 143     struct nfscldeleg *, struct ucred *, NFSPROC_T *);
 144 static void nfscl_totalrecall(struct nfsclclient *);
 145 static int nfscl_relock(vnode_t , struct nfsclclient *, struct nfsmount *,
 146     struct nfscllockowner *, struct nfscllock *, struct ucred *, NFSPROC_T *);
 147 static int nfscl_tryopen(struct nfsmount *, vnode_t , u_int8_t *, int,
 148     u_int8_t *, int, u_int32_t, struct nfsclopen *, u_int8_t *, int,
 149     struct nfscldeleg **, int, u_int32_t, struct ucred *, NFSPROC_T *);
 150 static int nfscl_trylock(struct nfsmount *, vnode_t , u_int8_t *,
 151     int, struct nfscllockowner *, int, int, u_int64_t, u_int64_t, short,
 152     struct ucred *, NFSPROC_T *);
 153 static int nfsrpc_reopen(struct nfsmount *, u_int8_t *, int, u_int32_t,
 154     struct nfsclopen *, struct nfscldeleg **, struct ucred *, NFSPROC_T *);
 155 static void nfscl_freedeleg(struct nfscldeleghead *, struct nfscldeleg *);
 156 static int nfscl_errmap(struct nfsrv_descript *, u_int32_t);
 157 static void nfscl_cleanup_common(struct nfsclclient *, u_int8_t *);
 158 static int nfscl_recalldeleg(struct nfsclclient *, struct nfsmount *,
 159     struct nfscldeleg *, vnode_t, struct ucred *, NFSPROC_T *, int);
 160 static void nfscl_freeopenowner(struct nfsclowner *, int);
 161 static void nfscl_cleandeleg(struct nfscldeleg *);
 162 static int nfscl_trydelegreturn(struct nfscldeleg *, struct ucred *,
 163     struct nfsmount *, NFSPROC_T *);
 164 static void nfscl_emptylockowner(struct nfscllockowner *,
 165     struct nfscllockownerfhhead *);
 166 static void nfscl_mergeflayouts(struct nfsclflayouthead *,
 167     struct nfsclflayouthead *);
 168 static int nfscl_layoutrecall(int, struct nfscllayout *, uint32_t, uint64_t,
 169     uint64_t, uint32_t, uint32_t, uint32_t, char *, struct nfsclrecalllayout *);
 170 static int nfscl_seq(uint32_t, uint32_t);
 171 static void nfscl_layoutreturn(struct nfsmount *, struct nfscllayout *,
 172     struct ucred *, NFSPROC_T *);
 173 static void nfscl_dolayoutcommit(struct nfsmount *, struct nfscllayout *,
 174     struct ucred *, NFSPROC_T *);
 175
 176 static short nfscberr_null[] = {
 177         0,
 178         0,
 179 };
 180
 181 static short nfscberr_getattr[] = {
 182         NFSERR_RESOURCE,
 183         NFSERR_BADHANDLE,
 184         NFSERR_BADXDR,
 185         NFSERR_RESOURCE,
 186         NFSERR_SERVERFAULT,
 187         0,
 188 };
 189
 190 static short nfscberr_recall[] = {
 191         NFSERR_RESOURCE,
 192         NFSERR_BADHANDLE,
 193         NFSERR_BADSTATEID,
 194         NFSERR_BADXDR,
 195         NFSERR_RESOURCE,
 196         NFSERR_SERVERFAULT,
 197         0,
 198 };
 199
 200 static short *nfscl_cberrmap[] = {
 201         nfscberr_null,
 202         nfscberr_null,
 203         nfscberr_null,
 204         nfscberr_getattr,
 205         nfscberr_recall
 206 };
 207
 208 #define NETFAMILY(clp) \
 209                 (((clp)->nfsc_flags & NFSCLFLAGS_AFINET6) ? AF_INET6 : AF_INET)
 210
 211 /*
 212  * Called for an open operation.
 213  * If the nfhp argument is NULL, just get an openowner.
 214  */
 215 int
 216 nfscl_open(vnode_t vp, u_int8_t *nfhp, int fhlen, u_int32_t amode, int usedeleg,
 217     struct ucred *cred, NFSPROC_T *p, struct nfsclowner **owpp,
 218     struct nfsclopen **opp, int *newonep, int *retp, int lockit)
 219 {
 220         struct nfsclclient *clp;
 221         struct nfsclowner *owp, *nowp;
 222         struct nfsclopen *op = NULL, *nop = NULL;
 223         struct nfscldeleg *dp;
 224         struct nfsclownerhead *ohp;
 225         u_int8_t own[NFSV4CL_LOCKNAMELEN];
 226         int ret;
 227
 228         if (newonep != NULL)
 229                 *newonep = 0;
 230         if (opp != NULL)
 231                 *opp = NULL;
 232         if (owpp != NULL)
 233                 *owpp = NULL;
 234
 235         /*
 236          * Might need one or both of these, so MALLOC them now, to
 237          * avoid a tsleep() in MALLOC later.
 238          */
 239         nowp = malloc(sizeof (struct nfsclowner),
 240             M_NFSCLOWNER, M_WAITOK);
 241         if (nfhp != NULL)
 242             nop = malloc(sizeof (struct nfsclopen) +
 243                 fhlen - 1, M_NFSCLOPEN, M_WAITOK);
 244         ret = nfscl_getcl(vp->v_mount, cred, p, 1, &clp);
 245         if (ret != 0) {
 246                 free(nowp, M_NFSCLOWNER);
 247                 if (nop != NULL)
 248                         free(nop, M_NFSCLOPEN);
 249                 return (ret);
 250         }
 251
 252         /*
 253          * Get the Open iff it already exists.
 254          * If none found, add the new one or return error, depending upon
 255          * "create".
 256          */
 257         NFSLOCKCLSTATE();
 258         dp = NULL;
 259         /* First check the delegation list */
 260         if (nfhp != NULL && usedeleg) {
 261                 LIST_FOREACH(dp, NFSCLDELEGHASH(clp, nfhp, fhlen), nfsdl_hash) {
 262                         if (dp->nfsdl_fhlen == fhlen &&
 263                             !NFSBCMP(nfhp, dp->nfsdl_fh, fhlen)) {
 264                                 if (!(amode & NFSV4OPEN_ACCESSWRITE) ||
 265                                     (dp->nfsdl_flags & NFSCLDL_WRITE))
 266                                         break;
 267                                 dp = NULL;
 268                                 break;
 269                         }
 270                 }
 271         }
 272
 273         /* For NFSv4.1/4.2 and this option, use a single open_owner. */
 274         if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount)))
 275                 nfscl_filllockowner(NULL, own, F_POSIX);
 276         else
 277                 nfscl_filllockowner(p->td_proc, own, F_POSIX);
 278         if (dp != NULL)
 279                 ohp = &dp->nfsdl_owner;
 280         else
 281                 ohp = &clp->nfsc_owner;
 282         /* Now, search for an openowner */
 283         LIST_FOREACH(owp, ohp, nfsow_list) {
 284                 if (!NFSBCMP(owp->nfsow_owner, own, NFSV4CL_LOCKNAMELEN))
 285                         break;
 286         }
 287
 288         /*
 289          * Create a new open, as required.
 290          */
 291         nfscl_newopen(clp, dp, &owp, &nowp, &op, &nop, own, nfhp, fhlen,
 292             cred, newonep);
 293
 294         /*
 295          * Now, check the mode on the open and return the appropriate
 296          * value.
 297          */
 298         if (retp != NULL) {
 299                 if (nfhp != NULL && dp != NULL && nop == NULL)
 300                         /* new local open on delegation */
 301                         *retp = NFSCLOPEN_SETCRED;
 302                 else
 303                         *retp = NFSCLOPEN_OK;
 304         }
 305         if (op != NULL && (amode & ~(op->nfso_mode))) {
 306                 op->nfso_mode |= amode;
 307                 if (retp != NULL && dp == NULL)
 308                         *retp = NFSCLOPEN_DOOPEN;
 309         }
 310
 311         /*
 312          * Serialize modifications to the open owner for multiple threads
 313          * within the same process using a read/write sleep lock.
 314          * For NFSv4.1 and a single OpenOwner, allow concurrent open operations
 315          * by acquiring a shared lock.  The close operations still use an
 316          * exclusive lock for this case.
 317          */
 318         if (lockit != 0) {
 319                 if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount))) {
 320                         /*
 321                          * Get a shared lock on the OpenOwner, but first
 322                          * wait for any pending exclusive lock, so that the
 323                          * exclusive locker gets priority.
 324                          */
 325                         nfsv4_lock(&owp->nfsow_rwlock, 0, NULL,
 326                             NFSCLSTATEMUTEXPTR, NULL);
 327                         nfsv4_getref(&owp->nfsow_rwlock, NULL,
 328                             NFSCLSTATEMUTEXPTR, NULL);
 329                 } else
 330                         nfscl_lockexcl(&owp->nfsow_rwlock, NFSCLSTATEMUTEXPTR);
 331         }
 332         NFSUNLOCKCLSTATE();
 333         if (nowp != NULL)
 334                 free(nowp, M_NFSCLOWNER);
 335         if (nop != NULL)
 336                 free(nop, M_NFSCLOPEN);
 337         if (owpp != NULL)
 338                 *owpp = owp;
 339         if (opp != NULL)
 340                 *opp = op;
 341         return (0);
 342 }
 343
 344 /*
 345  * Create a new open, as required.
 346  */
 347 static void
 348 nfscl_newopen(struct nfsclclient *clp, struct nfscldeleg *dp,
 349     struct nfsclowner **owpp, struct nfsclowner **nowpp, struct nfsclopen **opp,
 350     struct nfsclopen **nopp, u_int8_t *own, u_int8_t *fhp, int fhlen,
 351     struct ucred *cred, int *newonep)
 352 {
 353         struct nfsclowner *owp = *owpp, *nowp;
 354         struct nfsclopen *op, *nop;
 355
 356         if (nowpp != NULL)
 357                 nowp = *nowpp;
 358         else
 359                 nowp = NULL;
 360         if (nopp != NULL)
 361                 nop = *nopp;
 362         else
 363                 nop = NULL;
 364         if (owp == NULL && nowp != NULL) {
 365                 NFSBCOPY(own, nowp->nfsow_owner, NFSV4CL_LOCKNAMELEN);
 366                 LIST_INIT(&nowp->nfsow_open);
 367                 nowp->nfsow_clp = clp;
 368                 nowp->nfsow_seqid = 0;
 369                 nowp->nfsow_defunct = 0;
 370                 nfscl_lockinit(&nowp->nfsow_rwlock);
 371                 if (dp != NULL) {
 372                         nfsstatsv1.cllocalopenowners++;
 373                         LIST_INSERT_HEAD(&dp->nfsdl_owner, nowp, nfsow_list);
 374                 } else {
 375                         nfsstatsv1.clopenowners++;
 376                         LIST_INSERT_HEAD(&clp->nfsc_owner, nowp, nfsow_list);
 377                 }
 378                 owp = *owpp = nowp;
 379                 *nowpp = NULL;
 380                 if (newonep != NULL)
 381                         *newonep = 1;
 382         }
 383
 384          /* If an fhp has been specified, create an Open as well. */
 385         if (fhp != NULL) {
 386                 /* and look for the correct open, based upon FH */
 387                 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
 388                         if (op->nfso_fhlen == fhlen &&
 389                             !NFSBCMP(op->nfso_fh, fhp, fhlen))
 390                                 break;
 391                 }
 392                 if (op == NULL && nop != NULL) {
 393                         nop->nfso_own = owp;
 394                         nop->nfso_mode = 0;
 395                         nop->nfso_opencnt = 0;
 396                         nop->nfso_posixlock = 1;
 397                         nop->nfso_fhlen = fhlen;
 398                         NFSBCOPY(fhp, nop->nfso_fh, fhlen);
 399                         LIST_INIT(&nop->nfso_lock);
 400                         nop->nfso_stateid.seqid = 0;
 401                         nop->nfso_stateid.other[0] = 0;
 402                         nop->nfso_stateid.other[1] = 0;
 403                         nop->nfso_stateid.other[2] = 0;
 404                         KASSERT(cred != NULL, ("%s: cred NULL\n", __func__));
 405                         newnfs_copyincred(cred, &nop->nfso_cred);
 406                         if (dp != NULL) {
 407                                 TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
 408                                 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp,
 409                                     nfsdl_list);
 410                                 dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
 411                                 nfsstatsv1.cllocalopens++;
 412                         } else {
 413                                 nfsstatsv1.clopens++;
 414                         }
 415                         LIST_INSERT_HEAD(&owp->nfsow_open, nop, nfso_list);
 416                         *opp = nop;
 417                         *nopp = NULL;
 418                         if (newonep != NULL)
 419                                 *newonep = 1;
 420                 } else {
 421                         *opp = op;
 422                 }
 423         }
 424 }
 425
 426 /*
 427  * Called to find/add a delegation to a client.
 428  */
 429 int
 430 nfscl_deleg(mount_t mp, struct nfsclclient *clp, u_int8_t *nfhp,
 431     int fhlen, struct ucred *cred, NFSPROC_T *p, struct nfscldeleg **dpp)
 432 {
 433         struct nfscldeleg *dp = *dpp, *tdp;
 434
 435         /*
 436          * First, if we have received a Read delegation for a file on a
 437          * read/write file system, just return it, because they aren't
 438          * useful, imho.
 439          */
 440         if (mp != NULL && dp != NULL && !NFSMNT_RDONLY(mp) &&
 441             (dp->nfsdl_flags & NFSCLDL_READ)) {
 442                 (void) nfscl_trydelegreturn(dp, cred, VFSTONFS(mp), p);
 443                 free(dp, M_NFSCLDELEG);
 444                 *dpp = NULL;
 445                 return (0);
 446         }
 447
 448         /* Look for the correct deleg, based upon FH */
 449         NFSLOCKCLSTATE();
 450         tdp = nfscl_finddeleg(clp, nfhp, fhlen);
 451         if (tdp == NULL) {
 452                 if (dp == NULL) {
 453                         NFSUNLOCKCLSTATE();
 454                         return (NFSERR_BADSTATEID);
 455                 }
 456                 *dpp = NULL;
 457                 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list);
 458                 LIST_INSERT_HEAD(NFSCLDELEGHASH(clp, nfhp, fhlen), dp,
 459                     nfsdl_hash);
 460                 dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
 461                 nfsstatsv1.cldelegates++;
 462                 nfscl_delegcnt++;
 463         } else {
 464                 /*
 465                  * Delegation already exists, what do we do if a new one??
 466                  */
 467                 if (dp != NULL) {
 468                         printf("Deleg already exists!\n");
 469                         free(dp, M_NFSCLDELEG);
 470                         *dpp = NULL;
 471                 } else {
 472                         *dpp = tdp;
 473                 }
 474         }
 475         NFSUNLOCKCLSTATE();
 476         return (0);
 477 }
 478
 479 /*
 480  * Find a delegation for this file handle. Return NULL upon failure.
 481  */
 482 static struct nfscldeleg *
 483 nfscl_finddeleg(struct nfsclclient *clp, u_int8_t *fhp, int fhlen)
 484 {
 485         struct nfscldeleg *dp;
 486
 487         LIST_FOREACH(dp, NFSCLDELEGHASH(clp, fhp, fhlen), nfsdl_hash) {
 488             if (dp->nfsdl_fhlen == fhlen &&
 489                 !NFSBCMP(dp->nfsdl_fh, fhp, fhlen))
 490                 break;
 491         }
 492         return (dp);
 493 }
 494
 495 /*
 496  * Get a stateid for an I/O operation. First, look for an open and iff
 497  * found, return either a lockowner stateid or the open stateid.
 498  * If no Open is found, just return error and the special stateid of all zeros.
 499  */
 500 int
 501 nfscl_getstateid(vnode_t vp, u_int8_t *nfhp, int fhlen, u_int32_t mode,
 502     int fords, struct ucred *cred, NFSPROC_T *p, nfsv4stateid_t *stateidp,
 503     void **lckpp)
 504 {
 505         struct nfsclclient *clp;
 506         struct nfsclowner *owp;
 507         struct nfsclopen *op = NULL, *top;
 508         struct nfscllockowner *lp;
 509         struct nfscldeleg *dp;
 510         struct nfsnode *np;
 511         struct nfsmount *nmp;
 512         u_int8_t own[NFSV4CL_LOCKNAMELEN];
 513         int error;
 514         bool done;
 515
 516         *lckpp = NULL;
 517         /*
 518          * Initially, just set the special stateid of all zeros.
 519          * (Don't do this for a DS, since the special stateid can't be used.)
 520          */
 521         if (fords == 0) {
 522                 stateidp->seqid = 0;
 523                 stateidp->other[0] = 0;
 524                 stateidp->other[1] = 0;
 525                 stateidp->other[2] = 0;
 526         }
 527         if (vnode_vtype(vp) != VREG)
 528                 return (EISDIR);
 529         np = VTONFS(vp);
 530         nmp = VFSTONFS(vp->v_mount);
 531         NFSLOCKCLSTATE();
 532         clp = nfscl_findcl(nmp);
 533         if (clp == NULL) {
 534                 NFSUNLOCKCLSTATE();
 535                 return (EACCES);
 536         }
 537
 538         /*
 539          * Wait for recovery to complete.
 540          */
 541         while ((clp->nfsc_flags & NFSCLFLAGS_RECVRINPROG))
 542                 (void) nfsmsleep(&clp->nfsc_flags, NFSCLSTATEMUTEXPTR,
 543                     PZERO, "nfsrecvr", NULL);
 544
 545         /*
 546          * First, look for a delegation.
 547          */
 548         LIST_FOREACH(dp, NFSCLDELEGHASH(clp, nfhp, fhlen), nfsdl_hash) {
 549                 if (dp->nfsdl_fhlen == fhlen &&
 550                     !NFSBCMP(nfhp, dp->nfsdl_fh, fhlen)) {
 551                         if (!(mode & NFSV4OPEN_ACCESSWRITE) ||
 552                             (dp->nfsdl_flags & NFSCLDL_WRITE)) {
 553                                 stateidp->seqid = dp->nfsdl_stateid.seqid;
 554                                 stateidp->other[0] = dp->nfsdl_stateid.other[0];
 555                                 stateidp->other[1] = dp->nfsdl_stateid.other[1];
 556                                 stateidp->other[2] = dp->nfsdl_stateid.other[2];
 557                                 if (!(np->n_flag & NDELEGRECALL)) {
 558                                         TAILQ_REMOVE(&clp->nfsc_deleg, dp,
 559                                             nfsdl_list);
 560                                         TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp,
 561                                             nfsdl_list);
 562                                         dp->nfsdl_timestamp = NFSD_MONOSEC +
 563                                             120;
 564                                         dp->nfsdl_rwlock.nfslock_usecnt++;
 565                                         *lckpp = (void *)&dp->nfsdl_rwlock;
 566                                 }
 567                                 NFSUNLOCKCLSTATE();
 568                                 return (0);
 569                         }
 570                         break;
 571                 }
 572         }
 573
 574         if (p != NULL) {
 575                 /*
 576                  * If p != NULL, we want to search the parentage tree
 577                  * for a matching OpenOwner and use that.
 578                  */
 579                 if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount)))
 580                         nfscl_filllockowner(NULL, own, F_POSIX);
 581                 else
 582                         nfscl_filllockowner(p->td_proc, own, F_POSIX);
 583                 lp = NULL;
 584                 error = nfscl_getopen(&clp->nfsc_owner, nfhp, fhlen, own, own,
 585                     mode, &lp, &op);
 586                 if (error == 0 && lp != NULL && fords == 0) {
 587                         /* Don't return a lock stateid for a DS. */
 588                         stateidp->seqid =
 589                             lp->nfsl_stateid.seqid;
 590                         stateidp->other[0] =
 591                             lp->nfsl_stateid.other[0];
 592                         stateidp->other[1] =
 593                             lp->nfsl_stateid.other[1];
 594                         stateidp->other[2] =
 595                             lp->nfsl_stateid.other[2];
 596                         NFSUNLOCKCLSTATE();
 597                         return (0);
 598                 }
 599         }
 600         if (op == NULL) {
 601                 /* If not found, just look for any OpenOwner that will work. */
 602                 top = NULL;
 603                 done = false;
 604                 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
 605                         LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
 606                                 if (op->nfso_fhlen == fhlen &&
 607                                     !NFSBCMP(op->nfso_fh, nfhp, fhlen)) {
 608                                         if (top == NULL && (op->nfso_mode &
 609                                             NFSV4OPEN_ACCESSWRITE) != 0 &&
 610                                             (mode & NFSV4OPEN_ACCESSREAD) != 0)
 611                                                 top = op;
 612                                         if ((mode & op->nfso_mode) == mode) {
 613                                                 done = true;
 614                                                 break;
 615                                         }
 616                                 }
 617                         }
 618                         if (done)
 619                                 break;
 620                 }
 621                 if (!done) {
 622                         NFSCL_DEBUG(2, "openmode top=%p\n", top);
 623                         if (top == NULL || NFSHASOPENMODE(nmp)) {
 624                                 NFSUNLOCKCLSTATE();
 625                                 return (ENOENT);
 626                         } else
 627                                 op = top;
 628                 }
 629                 /*
 630                  * For read aheads or write behinds, use the open cred.
 631                  * A read ahead or write behind is indicated by p == NULL.
 632                  */
 633                 if (p == NULL)
 634                         newnfs_copycred(&op->nfso_cred, cred);
 635         }
 636
 637         /*
 638          * No lock stateid, so return the open stateid.
 639          */
 640         stateidp->seqid = op->nfso_stateid.seqid;
 641         stateidp->other[0] = op->nfso_stateid.other[0];
 642         stateidp->other[1] = op->nfso_stateid.other[1];
 643         stateidp->other[2] = op->nfso_stateid.other[2];
 644         NFSUNLOCKCLSTATE();
 645         return (0);
 646 }
 647
 648 /*
 649  * Search for a matching file, mode and, optionally, lockowner.
 650  */
 651 static int
 652 nfscl_getopen(struct nfsclownerhead *ohp, u_int8_t *nfhp, int fhlen,
 653     u_int8_t *openown, u_int8_t *lockown, u_int32_t mode,
 654     struct nfscllockowner **lpp, struct nfsclopen **opp)
 655 {
 656         struct nfsclowner *owp;
 657         struct nfsclopen *op, *rop, *rop2;
 658         bool 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 = true;
 675         /* Search the client list */
 676         LIST_FOREACH(owp, ohp, nfsow_list) {
 677                 /* and look for the correct open */
 678                 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
 679                         if (op->nfso_fhlen == fhlen &&
 680                             !NFSBCMP(op->nfso_fh, nfhp, fhlen)
 681                             && (op->nfso_mode & mode) == mode)
 682                                 keep_looping = nfscl_checkown(owp, op, openown,
 683                                     lockown, lpp, &rop, &rop2);
 684                         if (!keep_looping)
 685                                 break;
 686                 }
 687                 if (!keep_looping)
 688                         break;
 689         }
 690         if (rop == NULL)
 691                 rop = rop2;
 692         if (rop == NULL)
 693                 return (EBADF);
 694         *opp = rop;
 695         return (0);
 696 }
 697
 698 /* Check for an owner match. */
 699 static bool
 700 nfscl_checkown(struct nfsclowner *owp, struct nfsclopen *op, uint8_t *openown,
 701     uint8_t *lockown, struct nfscllockowner **lpp, struct nfsclopen **ropp,
 702     struct nfsclopen **ropp2)
 703 {
 704         struct nfscllockowner *lp;
 705         bool keep_looping;
 706
 707         keep_looping = true;
 708         if (lpp != NULL) {
 709                 /* Now look for a matching lockowner. */
 710                 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
 711                         if (!NFSBCMP(lp->nfsl_owner, lockown,
 712                             NFSV4CL_LOCKNAMELEN)) {
 713                                 *lpp = lp;
 714                                 *ropp = op;
 715                                 return (false);
 716                         }
 717                 }
 718         }
 719         if (*ropp == NULL && !NFSBCMP(owp->nfsow_owner, openown,
 720             NFSV4CL_LOCKNAMELEN)) {
 721                 *ropp = op;
 722                 if (lpp == NULL)
 723                         keep_looping = false;
 724         }
 725         if (*ropp2 == NULL)
 726                 *ropp2 = op;
 727         return (keep_looping);
 728 }
 729
 730 /*
 731  * Release use of an open owner. Called when open operations are done
 732  * with the open owner.
 733  */
 734 void
 735 nfscl_ownerrelease(struct nfsmount *nmp, struct nfsclowner *owp,
 736     __unused int error, __unused int candelete, int unlocked)
 737 {
 738
 739         if (owp == NULL)
 740                 return;
 741         NFSLOCKCLSTATE();
 742         if (unlocked == 0) {
 743                 if (NFSHASONEOPENOWN(nmp))
 744                         nfsv4_relref(&owp->nfsow_rwlock);
 745                 else
 746                         nfscl_lockunlock(&owp->nfsow_rwlock);
 747         }
 748         nfscl_clrelease(owp->nfsow_clp);
 749         NFSUNLOCKCLSTATE();
 750 }
 751
 752 /*
 753  * Release use of an open structure under an open owner.
 754  */
 755 void
 756 nfscl_openrelease(struct nfsmount *nmp, struct nfsclopen *op, int error,
 757     int candelete)
 758 {
 759         struct nfsclclient *clp;
 760         struct nfsclowner *owp;
 761
 762         if (op == NULL)
 763                 return;
 764         NFSLOCKCLSTATE();
 765         owp = op->nfso_own;
 766         if (NFSHASONEOPENOWN(nmp))
 767                 nfsv4_relref(&owp->nfsow_rwlock);
 768         else
 769                 nfscl_lockunlock(&owp->nfsow_rwlock);
 770         clp = owp->nfsow_clp;
 771         if (error && candelete && op->nfso_opencnt == 0)
 772                 nfscl_freeopen(op, 0);
 773         nfscl_clrelease(clp);
 774         NFSUNLOCKCLSTATE();
 775 }
 776
 777 /*
 778  * Called to get a clientid structure. It will optionally lock the
 779  * client data structures to do the SetClientId/SetClientId_confirm,
 780  * but will release that lock and return the clientid with a reference
 781  * count on it.
 782  * If the "cred" argument is NULL, a new clientid should not be created.
 783  * If the "p" argument is NULL, a SetClientID/SetClientIDConfirm cannot
 784  * be done.
 785  * The start_renewthread argument tells nfscl_getcl() to start a renew
 786  * thread if this creates a new clp.
 787  * It always clpp with a reference count on it, unless returning an error.
 788  */
 789 int
 790 nfscl_getcl(struct mount *mp, struct ucred *cred, NFSPROC_T *p,
 791     int start_renewthread, struct nfsclclient **clpp)
 792 {
 793         struct nfsclclient *clp;
 794         struct nfsclclient *newclp = NULL;
 795         struct nfsmount *nmp;
 796         char uuid[HOSTUUIDLEN];
 797         int igotlock = 0, error, trystalecnt, clidinusedelay, i;
 798         u_int16_t idlen = 0;
 799
 800         nmp = VFSTONFS(mp);
 801         if (cred != NULL) {
 802                 getcredhostuuid(cred, uuid, sizeof uuid);
 803                 idlen = strlen(uuid);
 804                 if (idlen > 0)
 805                         idlen += sizeof (u_int64_t);
 806                 else
 807                         idlen += sizeof (u_int64_t) + 16; /* 16 random bytes */
 808                 newclp = malloc(
 809                     sizeof (struct nfsclclient) + idlen - 1, M_NFSCLCLIENT,
 810                     M_WAITOK | M_ZERO);
 811         }
 812         NFSLOCKCLSTATE();
 813         /*
 814          * If a forced dismount is already in progress, don't
 815          * allocate a new clientid and get out now. For the case where
 816          * clp != NULL, this is a harmless optimization.
 817          */
 818         if (NFSCL_FORCEDISM(mp)) {
 819                 NFSUNLOCKCLSTATE();
 820                 if (newclp != NULL)
 821                         free(newclp, M_NFSCLCLIENT);
 822                 return (EBADF);
 823         }
 824         clp = nmp->nm_clp;
 825         if (clp == NULL) {
 826                 if (newclp == NULL) {
 827                         NFSUNLOCKCLSTATE();
 828                         return (EACCES);
 829                 }
 830                 clp = newclp;
 831                 clp->nfsc_idlen = idlen;
 832                 LIST_INIT(&clp->nfsc_owner);
 833                 TAILQ_INIT(&clp->nfsc_deleg);
 834                 TAILQ_INIT(&clp->nfsc_layout);
 835                 LIST_INIT(&clp->nfsc_devinfo);
 836                 for (i = 0; i < NFSCLDELEGHASHSIZE; i++)
 837                         LIST_INIT(&clp->nfsc_deleghash[i]);
 838                 for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++)
 839                         LIST_INIT(&clp->nfsc_layouthash[i]);
 840                 clp->nfsc_flags = NFSCLFLAGS_INITED;
 841                 clp->nfsc_clientidrev = 1;
 842                 clp->nfsc_cbident = nfscl_nextcbident();
 843                 nfscl_fillclid(nmp->nm_clval, uuid, clp->nfsc_id,
 844                     clp->nfsc_idlen);
 845                 LIST_INSERT_HEAD(&nfsclhead, clp, nfsc_list);
 846                 nmp->nm_clp = clp;
 847                 clp->nfsc_nmp = nmp;
 848                 NFSUNLOCKCLSTATE();
 849                 if (start_renewthread != 0)
 850                         nfscl_start_renewthread(clp);
 851         } else {
 852                 NFSUNLOCKCLSTATE();
 853                 if (newclp != NULL)
 854                         free(newclp, M_NFSCLCLIENT);
 855         }
 856         NFSLOCKCLSTATE();
 857         while ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0 && !igotlock &&
 858             !NFSCL_FORCEDISM(mp))
 859                 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
 860                     NFSCLSTATEMUTEXPTR, mp);
 861         if (igotlock == 0) {
 862                 /*
 863                  * Call nfsv4_lock() with "iwantlock == 0" so that it will
 864                  * wait for a pending exclusive lock request.  This gives the
 865                  * exclusive lock request priority over this shared lock
 866                  * request.
 867                  * An exclusive lock on nfsc_lock is used mainly for server
 868                  * crash recoveries.
 869                  */
 870                 nfsv4_lock(&clp->nfsc_lock, 0, NULL, NFSCLSTATEMUTEXPTR, mp);
 871                 nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
 872         }
 873         if (igotlock == 0 && NFSCL_FORCEDISM(mp)) {
 874                 /*
 875                  * Both nfsv4_lock() and nfsv4_getref() know to check
 876                  * for NFSCL_FORCEDISM() and return without sleeping to
 877                  * wait for the exclusive lock to be released, since it
 878                  * might be held by nfscl_umount() and we need to get out
 879                  * now for that case and not wait until nfscl_umount()
 880                  * releases it.
 881                  */
 882                 NFSUNLOCKCLSTATE();
 883                 return (EBADF);
 884         }
 885         NFSUNLOCKCLSTATE();
 886
 887         /*
 888          * If it needs a clientid, do the setclientid now.
 889          */
 890         if ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0) {
 891                 if (!igotlock)
 892                         panic("nfscl_clget");
 893                 if (p == NULL || cred == NULL) {
 894                         NFSLOCKCLSTATE();
 895                         nfsv4_unlock(&clp->nfsc_lock, 0);
 896                         NFSUNLOCKCLSTATE();
 897                         return (EACCES);
 898                 }
 899                 /*
 900                  * If RFC3530 Sec. 14.2.33 is taken literally,
 901                  * NFSERR_CLIDINUSE will be returned persistently for the
 902                  * case where a new mount of the same file system is using
 903                  * a different principal. In practice, NFSERR_CLIDINUSE is
 904                  * only returned when there is outstanding unexpired state
 905                  * on the clientid. As such, try for twice the lease
 906                  * interval, if we know what that is. Otherwise, make a
 907                  * wild ass guess.
 908                  * The case of returning NFSERR_STALECLIENTID is far less
 909                  * likely, but might occur if there is a significant delay
 910                  * between doing the SetClientID and SetClientIDConfirm Ops,
 911                  * such that the server throws away the clientid before
 912                  * receiving the SetClientIDConfirm.
 913                  */
 914                 if (clp->nfsc_renew > 0)
 915                         clidinusedelay = NFSCL_LEASE(clp->nfsc_renew) * 2;
 916                 else
 917                         clidinusedelay = 120;
 918                 trystalecnt = 3;
 919                 do {
 920                         error = nfsrpc_setclient(nmp, clp, 0, NULL, cred, p);
 921                         if (error == NFSERR_STALECLIENTID ||
 922                             error == NFSERR_STALEDONTRECOVER ||
 923                             error == NFSERR_BADSESSION ||
 924                             error == NFSERR_CLIDINUSE) {
 925                                 (void) nfs_catnap(PZERO, error, "nfs_setcl");
 926                         }
 927                 } while (((error == NFSERR_STALECLIENTID ||
 928                      error == NFSERR_BADSESSION ||
 929                      error == NFSERR_STALEDONTRECOVER) && --trystalecnt > 0) ||
 930                     (error == NFSERR_CLIDINUSE && --clidinusedelay > 0));
 931                 if (error) {
 932                         NFSLOCKCLSTATE();
 933                         nfsv4_unlock(&clp->nfsc_lock, 0);
 934                         NFSUNLOCKCLSTATE();
 935                         return (error);
 936                 }
 937                 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
 938         }
 939         if (igotlock) {
 940                 NFSLOCKCLSTATE();
 941                 nfsv4_unlock(&clp->nfsc_lock, 1);
 942                 NFSUNLOCKCLSTATE();
 943         }
 944
 945         *clpp = clp;
 946         return (0);
 947 }
 948
 949 /*
 950  * Get a reference to a clientid and return it, if valid.
 951  */
 952 struct nfsclclient *
 953 nfscl_findcl(struct nfsmount *nmp)
 954 {
 955         struct nfsclclient *clp;
 956
 957         clp = nmp->nm_clp;
 958         if (clp == NULL || !(clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID))
 959                 return (NULL);
 960         return (clp);
 961 }
 962
 963 /*
 964  * Release the clientid structure. It may be locked or reference counted.
 965  */
 966 static void
 967 nfscl_clrelease(struct nfsclclient *clp)
 968 {
 969
 970         if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK)
 971                 nfsv4_unlock(&clp->nfsc_lock, 0);
 972         else
 973                 nfsv4_relref(&clp->nfsc_lock);
 974 }
 975
 976 /*
 977  * External call for nfscl_clrelease.
 978  */
 979 void
 980 nfscl_clientrelease(struct nfsclclient *clp)
 981 {
 982
 983         NFSLOCKCLSTATE();
 984         if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK)
 985                 nfsv4_unlock(&clp->nfsc_lock, 0);
 986         else
 987                 nfsv4_relref(&clp->nfsc_lock);
 988         NFSUNLOCKCLSTATE();
 989 }
 990
 991 /*
 992  * Called when wanting to lock a byte region.
 993  */
 994 int
 995 nfscl_getbytelock(vnode_t vp, u_int64_t off, u_int64_t len,
 996     short type, struct ucred *cred, NFSPROC_T *p, struct nfsclclient *rclp,
 997     int recovery, void *id, int flags, u_int8_t *rownp, u_int8_t *ropenownp,
 998     struct nfscllockowner **lpp, int *newonep, int *donelocallyp)
 999 {
1000         struct nfscllockowner *lp;
1001         struct nfsclopen *op;
1002         struct nfsclclient *clp;
1003         struct nfscllockowner *nlp;
1004         struct nfscllock *nlop, *otherlop;
1005         struct nfscldeleg *dp = NULL, *ldp = NULL;
1006         struct nfscllockownerhead *lhp = NULL;
1007         struct nfsnode *np;
1008         u_int8_t own[NFSV4CL_LOCKNAMELEN], *ownp, openown[NFSV4CL_LOCKNAMELEN];
1009         u_int8_t *openownp;
1010         int error = 0, ret, donelocally = 0;
1011         u_int32_t mode;
1012
1013         /* For Lock Ops, the open mode doesn't matter, so use 0 to match any. */
1014         mode = 0;
1015         np = VTONFS(vp);
1016         *lpp = NULL;
1017         lp = NULL;
1018         *newonep = 0;
1019         *donelocallyp = 0;
1020
1021         /*
1022          * Might need these, so MALLOC them now, to
1023          * avoid a tsleep() in MALLOC later.
1024          */
1025         nlp = malloc(
1026             sizeof (struct nfscllockowner), M_NFSCLLOCKOWNER, M_WAITOK);
1027         otherlop = malloc(
1028             sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1029         nlop = malloc(
1030             sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1031         nlop->nfslo_type = type;
1032         nlop->nfslo_first = off;
1033         if (len == NFS64BITSSET) {
1034                 nlop->nfslo_end = NFS64BITSSET;
1035         } else {
1036                 nlop->nfslo_end = off + len;
1037                 if (nlop->nfslo_end <= nlop->nfslo_first)
1038                         error = NFSERR_INVAL;
1039         }
1040
1041         if (!error) {
1042                 if (recovery)
1043                         clp = rclp;
1044                 else
1045                         error = nfscl_getcl(vp->v_mount, cred, p, 1, &clp);
1046         }
1047         if (error) {
1048                 free(nlp, M_NFSCLLOCKOWNER);
1049                 free(otherlop, M_NFSCLLOCK);
1050                 free(nlop, M_NFSCLLOCK);
1051                 return (error);
1052         }
1053
1054         op = NULL;
1055         if (recovery) {
1056                 ownp = rownp;
1057                 openownp = ropenownp;
1058         } else {
1059                 nfscl_filllockowner(id, own, flags);
1060                 ownp = own;
1061                 if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount)))
1062                         nfscl_filllockowner(NULL, openown, F_POSIX);
1063                 else
1064                         nfscl_filllockowner(p->td_proc, openown, F_POSIX);
1065                 openownp = openown;
1066         }
1067         if (!recovery) {
1068                 NFSLOCKCLSTATE();
1069                 /*
1070                  * First, search for a delegation. If one exists for this file,
1071                  * the lock can be done locally against it, so long as there
1072                  * isn't a local lock conflict.
1073                  */
1074                 ldp = dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
1075                     np->n_fhp->nfh_len);
1076                 /* Just sanity check for correct type of delegation */
1077                 if (dp != NULL && ((dp->nfsdl_flags &
1078                     (NFSCLDL_RECALL | NFSCLDL_DELEGRET)) != 0 ||
1079                      (type == F_WRLCK &&
1080                       (dp->nfsdl_flags & NFSCLDL_WRITE) == 0)))
1081                         dp = NULL;
1082         }
1083         if (dp != NULL) {
1084                 /* Now, find an open and maybe a lockowner. */
1085                 ret = nfscl_getopen(&dp->nfsdl_owner, np->n_fhp->nfh_fh,
1086                     np->n_fhp->nfh_len, openownp, ownp, mode, NULL, &op);
1087                 if (ret)
1088                         ret = nfscl_getopen(&clp->nfsc_owner,
1089                             np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp,
1090                             ownp, mode, NULL, &op);
1091                 if (!ret) {
1092                         lhp = &dp->nfsdl_lock;
1093                         TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
1094                         TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list);
1095                         dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
1096                         donelocally = 1;
1097                 } else {
1098                         dp = NULL;
1099                 }
1100         }
1101         if (!donelocally) {
1102                 /*
1103                  * Get the related Open and maybe lockowner.
1104                  */
1105                 error = nfscl_getopen(&clp->nfsc_owner,
1106                     np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp,
1107                     ownp, mode, &lp, &op);
1108                 if (!error)
1109                         lhp = &op->nfso_lock;
1110         }
1111         if (!error && !recovery)
1112                 error = nfscl_localconflict(clp, np->n_fhp->nfh_fh,
1113                     np->n_fhp->nfh_len, nlop, ownp, ldp, NULL);
1114         if (error) {
1115                 if (!recovery) {
1116                         nfscl_clrelease(clp);
1117                         NFSUNLOCKCLSTATE();
1118                 }
1119                 free(nlp, M_NFSCLLOCKOWNER);
1120                 free(otherlop, M_NFSCLLOCK);
1121                 free(nlop, M_NFSCLLOCK);
1122                 return (error);
1123         }
1124
1125         /*
1126          * Ok, see if a lockowner exists and create one, as required.
1127          */
1128         if (lp == NULL)
1129                 LIST_FOREACH(lp, lhp, nfsl_list) {
1130                         if (!NFSBCMP(lp->nfsl_owner, ownp, NFSV4CL_LOCKNAMELEN))
1131                                 break;
1132                 }
1133         if (lp == NULL) {
1134                 NFSBCOPY(ownp, nlp->nfsl_owner, NFSV4CL_LOCKNAMELEN);
1135                 if (recovery)
1136                         NFSBCOPY(ropenownp, nlp->nfsl_openowner,
1137                             NFSV4CL_LOCKNAMELEN);
1138                 else
1139                         NFSBCOPY(op->nfso_own->nfsow_owner, nlp->nfsl_openowner,
1140                             NFSV4CL_LOCKNAMELEN);
1141                 nlp->nfsl_seqid = 0;
1142                 nlp->nfsl_lockflags = flags;
1143                 nlp->nfsl_inprog = NULL;
1144                 nfscl_lockinit(&nlp->nfsl_rwlock);
1145                 LIST_INIT(&nlp->nfsl_lock);
1146                 if (donelocally) {
1147                         nlp->nfsl_open = NULL;
1148                         nfsstatsv1.cllocallockowners++;
1149                 } else {
1150                         nlp->nfsl_open = op;
1151                         nfsstatsv1.cllockowners++;
1152                 }
1153                 LIST_INSERT_HEAD(lhp, nlp, nfsl_list);
1154                 lp = nlp;
1155                 nlp = NULL;
1156                 *newonep = 1;
1157         }
1158
1159         /*
1160          * Now, update the byte ranges for locks.
1161          */
1162         ret = nfscl_updatelock(lp, &nlop, &otherlop, donelocally);
1163         if (!ret)
1164                 donelocally = 1;
1165         if (donelocally) {
1166                 *donelocallyp = 1;
1167                 if (!recovery)
1168                         nfscl_clrelease(clp);
1169         } else {
1170                 /*
1171                  * Serial modifications on the lock owner for multiple threads
1172                  * for the same process using a read/write lock.
1173                  */
1174                 if (!recovery)
1175                         nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR);
1176         }
1177         if (!recovery)
1178                 NFSUNLOCKCLSTATE();
1179
1180         if (nlp)
1181                 free(nlp, M_NFSCLLOCKOWNER);
1182         if (nlop)
1183                 free(nlop, M_NFSCLLOCK);
1184         if (otherlop)
1185                 free(otherlop, M_NFSCLLOCK);
1186
1187         *lpp = lp;
1188         return (0);
1189 }
1190
1191 /*
1192  * Called to unlock a byte range, for LockU.
1193  */
1194 int
1195 nfscl_relbytelock(vnode_t vp, u_int64_t off, u_int64_t len,
1196     __unused struct ucred *cred, NFSPROC_T *p, int callcnt,
1197     struct nfsclclient *clp, void *id, int flags,
1198     struct nfscllockowner **lpp, int *dorpcp)
1199 {
1200         struct nfscllockowner *lp;
1201         struct nfsclowner *owp;
1202         struct nfsclopen *op;
1203         struct nfscllock *nlop, *other_lop = NULL;
1204         struct nfscldeleg *dp;
1205         struct nfsnode *np;
1206         u_int8_t own[NFSV4CL_LOCKNAMELEN];
1207         int ret = 0, fnd;
1208
1209         np = VTONFS(vp);
1210         *lpp = NULL;
1211         *dorpcp = 0;
1212
1213         /*
1214          * Might need these, so MALLOC them now, to
1215          * avoid a tsleep() in MALLOC later.
1216          */
1217         nlop = malloc(
1218             sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1219         nlop->nfslo_type = F_UNLCK;
1220         nlop->nfslo_first = off;
1221         if (len == NFS64BITSSET) {
1222                 nlop->nfslo_end = NFS64BITSSET;
1223         } else {
1224                 nlop->nfslo_end = off + len;
1225                 if (nlop->nfslo_end <= nlop->nfslo_first) {
1226                         free(nlop, M_NFSCLLOCK);
1227                         return (NFSERR_INVAL);
1228                 }
1229         }
1230         if (callcnt == 0) {
1231                 other_lop = malloc(
1232                     sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1233                 *other_lop = *nlop;
1234         }
1235         nfscl_filllockowner(id, own, flags);
1236         dp = NULL;
1237         NFSLOCKCLSTATE();
1238         if (callcnt == 0)
1239                 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
1240                     np->n_fhp->nfh_len);
1241
1242         /*
1243          * First, unlock any local regions on a delegation.
1244          */
1245         if (dp != NULL) {
1246                 /* Look for this lockowner. */
1247                 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
1248                         if (!NFSBCMP(lp->nfsl_owner, own,
1249                             NFSV4CL_LOCKNAMELEN))
1250                                 break;
1251                 }
1252                 if (lp != NULL)
1253                         /* Use other_lop, so nlop is still available */
1254                         (void)nfscl_updatelock(lp, &other_lop, NULL, 1);
1255         }
1256
1257         /*
1258          * Now, find a matching open/lockowner that hasn't already been done,
1259          * as marked by nfsl_inprog.
1260          */
1261         lp = NULL;
1262         fnd = 0;
1263         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
1264             LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1265                 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1266                     !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1267                     LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1268                         if (lp->nfsl_inprog == NULL &&
1269                             !NFSBCMP(lp->nfsl_owner, own,
1270                              NFSV4CL_LOCKNAMELEN)) {
1271                                 fnd = 1;
1272                                 break;
1273                         }
1274                     }
1275                     if (fnd)
1276                         break;
1277                 }
1278             }
1279             if (fnd)
1280                 break;
1281         }
1282
1283         if (lp != NULL) {
1284                 ret = nfscl_updatelock(lp, &nlop, NULL, 0);
1285                 if (ret)
1286                         *dorpcp = 1;
1287                 /*
1288                  * Serial modifications on the lock owner for multiple
1289                  * threads for the same process using a read/write lock.
1290                  */
1291                 lp->nfsl_inprog = p;
1292                 nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR);
1293                 *lpp = lp;
1294         }
1295         NFSUNLOCKCLSTATE();
1296         if (nlop)
1297                 free(nlop, M_NFSCLLOCK);
1298         if (other_lop)
1299                 free(other_lop, M_NFSCLLOCK);
1300         return (0);
1301 }
1302
1303 /*
1304  * Release all lockowners marked in progess for this process and file.
1305  */
1306 void
1307 nfscl_releasealllocks(struct nfsclclient *clp, vnode_t vp, NFSPROC_T *p,
1308     void *id, int flags)
1309 {
1310         struct nfsclowner *owp;
1311         struct nfsclopen *op;
1312         struct nfscllockowner *lp;
1313         struct nfsnode *np;
1314         u_int8_t own[NFSV4CL_LOCKNAMELEN];
1315
1316         np = VTONFS(vp);
1317         nfscl_filllockowner(id, own, flags);
1318         NFSLOCKCLSTATE();
1319         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
1320             LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1321                 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1322                     !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1323                     LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1324                         if (lp->nfsl_inprog == p &&
1325                             !NFSBCMP(lp->nfsl_owner, own,
1326                             NFSV4CL_LOCKNAMELEN)) {
1327                             lp->nfsl_inprog = NULL;
1328                             nfscl_lockunlock(&lp->nfsl_rwlock);
1329                         }
1330                     }
1331                 }
1332             }
1333         }
1334         nfscl_clrelease(clp);
1335         NFSUNLOCKCLSTATE();
1336 }
1337
1338 /*
1339  * Called to find out if any bytes within the byte range specified are
1340  * write locked by the calling process. Used to determine if flushing
1341  * is required before a LockU.
1342  * If in doubt, return 1, so the flush will occur.
1343  */
1344 int
1345 nfscl_checkwritelocked(vnode_t vp, struct flock *fl,
1346     struct ucred *cred, NFSPROC_T *p, void *id, int flags)
1347 {
1348         struct nfsclowner *owp;
1349         struct nfscllockowner *lp;
1350         struct nfsclopen *op;
1351         struct nfsclclient *clp;
1352         struct nfscllock *lop;
1353         struct nfscldeleg *dp;
1354         struct nfsnode *np;
1355         u_int64_t off, end;
1356         u_int8_t own[NFSV4CL_LOCKNAMELEN];
1357         int error = 0;
1358
1359         np = VTONFS(vp);
1360         switch (fl->l_whence) {
1361         case SEEK_SET:
1362         case SEEK_CUR:
1363                 /*
1364                  * Caller is responsible for adding any necessary offset
1365                  * when SEEK_CUR is used.
1366                  */
1367                 off = fl->l_start;
1368                 break;
1369         case SEEK_END:
1370                 off = np->n_size + fl->l_start;
1371                 break;
1372         default:
1373                 return (1);
1374         }
1375         if (fl->l_len != 0) {
1376                 end = off + fl->l_len;
1377                 if (end < off)
1378                         return (1);
1379         } else {
1380                 end = NFS64BITSSET;
1381         }
1382
1383         error = nfscl_getcl(vp->v_mount, cred, p, 1, &clp);
1384         if (error)
1385                 return (1);
1386         nfscl_filllockowner(id, own, flags);
1387         NFSLOCKCLSTATE();
1388
1389         /*
1390          * First check the delegation locks.
1391          */
1392         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
1393         if (dp != NULL) {
1394                 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
1395                         if (!NFSBCMP(lp->nfsl_owner, own,
1396                             NFSV4CL_LOCKNAMELEN))
1397                                 break;
1398                 }
1399                 if (lp != NULL) {
1400                         LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
1401                                 if (lop->nfslo_first >= end)
1402                                         break;
1403                                 if (lop->nfslo_end <= off)
1404                                         continue;
1405                                 if (lop->nfslo_type == F_WRLCK) {
1406                                         nfscl_clrelease(clp);
1407                                         NFSUNLOCKCLSTATE();
1408                                         return (1);
1409                                 }
1410                         }
1411                 }
1412         }
1413
1414         /*
1415          * Now, check state against the server.
1416          */
1417         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
1418             LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1419                 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1420                     !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1421                     LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1422                         if (!NFSBCMP(lp->nfsl_owner, own,
1423                             NFSV4CL_LOCKNAMELEN))
1424                             break;
1425                     }
1426                     if (lp != NULL) {
1427                         LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
1428                             if (lop->nfslo_first >= end)
1429                                 break;
1430                             if (lop->nfslo_end <= off)
1431                                 continue;
1432                             if (lop->nfslo_type == F_WRLCK) {
1433                                 nfscl_clrelease(clp);
1434                                 NFSUNLOCKCLSTATE();
1435                                 return (1);
1436                             }
1437                         }
1438                     }
1439                 }
1440             }
1441         }
1442         nfscl_clrelease(clp);
1443         NFSUNLOCKCLSTATE();
1444         return (0);
1445 }
1446
1447 /*
1448  * Release a byte range lock owner structure.
1449  */
1450 void
1451 nfscl_lockrelease(struct nfscllockowner *lp, int error, int candelete)
1452 {
1453         struct nfsclclient *clp;
1454
1455         if (lp == NULL)
1456                 return;
1457         NFSLOCKCLSTATE();
1458         clp = lp->nfsl_open->nfso_own->nfsow_clp;
1459         if (error != 0 && candelete &&
1460             (lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED) == 0)
1461                 nfscl_freelockowner(lp, 0);
1462         else
1463                 nfscl_lockunlock(&lp->nfsl_rwlock);
1464         nfscl_clrelease(clp);
1465         NFSUNLOCKCLSTATE();
1466 }
1467
1468 /*
1469  * Free up an open structure and any associated byte range lock structures.
1470  */
1471 void
1472 nfscl_freeopen(struct nfsclopen *op, int local)
1473 {
1474
1475         LIST_REMOVE(op, nfso_list);
1476         nfscl_freealllocks(&op->nfso_lock, local);
1477         free(op, M_NFSCLOPEN);
1478         if (local)
1479                 nfsstatsv1.cllocalopens--;
1480         else
1481                 nfsstatsv1.clopens--;
1482 }
1483
1484 /*
1485  * Free up all lock owners and associated locks.
1486  */
1487 static void
1488 nfscl_freealllocks(struct nfscllockownerhead *lhp, int local)
1489 {
1490         struct nfscllockowner *lp, *nlp;
1491
1492         LIST_FOREACH_SAFE(lp, lhp, nfsl_list, nlp) {
1493                 if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED))
1494                         panic("nfscllckw");
1495                 nfscl_freelockowner(lp, local);
1496         }
1497 }
1498
1499 /*
1500  * Called for an Open when NFSERR_EXPIRED is received from the server.
1501  * If there are no byte range locks nor a Share Deny lost, try to do a
1502  * fresh Open. Otherwise, free the open.
1503  */
1504 static int
1505 nfscl_expireopen(struct nfsclclient *clp, struct nfsclopen *op,
1506     struct nfsmount *nmp, struct ucred *cred, NFSPROC_T *p)
1507 {
1508         struct nfscllockowner *lp;
1509         struct nfscldeleg *dp;
1510         int mustdelete = 0, error;
1511
1512         /*
1513          * Look for any byte range lock(s).
1514          */
1515         LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1516                 if (!LIST_EMPTY(&lp->nfsl_lock)) {
1517                         mustdelete = 1;
1518                         break;
1519                 }
1520         }
1521
1522         /*
1523          * If no byte range lock(s) nor a Share deny, try to re-open.
1524          */
1525         if (!mustdelete && (op->nfso_mode & NFSLCK_DENYBITS) == 0) {
1526                 newnfs_copycred(&op->nfso_cred, cred);
1527                 dp = NULL;
1528                 error = nfsrpc_reopen(nmp, op->nfso_fh,
1529                     op->nfso_fhlen, op->nfso_mode, op, &dp, cred, p);
1530                 if (error) {
1531                         mustdelete = 1;
1532                         if (dp != NULL) {
1533                                 free(dp, M_NFSCLDELEG);
1534                                 dp = NULL;
1535                         }
1536                 }
1537                 if (dp != NULL)
1538                         nfscl_deleg(nmp->nm_mountp, clp, op->nfso_fh,
1539                             op->nfso_fhlen, cred, p, &dp);
1540         }
1541
1542         /*
1543          * If a byte range lock or Share deny or couldn't re-open, free it.
1544          */
1545         if (mustdelete)
1546                 nfscl_freeopen(op, 0);
1547         return (mustdelete);
1548 }
1549
1550 /*
1551  * Free up an open owner structure.
1552  */
1553 static void
1554 nfscl_freeopenowner(struct nfsclowner *owp, int local)
1555 {
1556
1557         LIST_REMOVE(owp, nfsow_list);
1558         free(owp, M_NFSCLOWNER);
1559         if (local)
1560                 nfsstatsv1.cllocalopenowners--;
1561         else
1562                 nfsstatsv1.clopenowners--;
1563 }
1564
1565 /*
1566  * Free up a byte range lock owner structure.
1567  */
1568 void
1569 nfscl_freelockowner(struct nfscllockowner *lp, int local)
1570 {
1571         struct nfscllock *lop, *nlop;
1572
1573         LIST_REMOVE(lp, nfsl_list);
1574         LIST_FOREACH_SAFE(lop, &lp->nfsl_lock, nfslo_list, nlop) {
1575                 nfscl_freelock(lop, local);
1576         }
1577         free(lp, M_NFSCLLOCKOWNER);
1578         if (local)
1579                 nfsstatsv1.cllocallockowners--;
1580         else
1581                 nfsstatsv1.cllockowners--;
1582 }
1583
1584 /*
1585  * Free up a byte range lock structure.
1586  */
1587 void
1588 nfscl_freelock(struct nfscllock *lop, int local)
1589 {
1590
1591         LIST_REMOVE(lop, nfslo_list);
1592         free(lop, M_NFSCLLOCK);
1593         if (local)
1594                 nfsstatsv1.cllocallocks--;
1595         else
1596                 nfsstatsv1.cllocks--;
1597 }
1598
1599 /*
1600  * Clean out the state related to a delegation.
1601  */
1602 static void
1603 nfscl_cleandeleg(struct nfscldeleg *dp)
1604 {
1605         struct nfsclowner *owp, *nowp;
1606         struct nfsclopen *op;
1607
1608         LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) {
1609                 op = LIST_FIRST(&owp->nfsow_open);
1610                 if (op != NULL) {
1611                         if (LIST_NEXT(op, nfso_list) != NULL)
1612                                 panic("nfscleandel");
1613                         nfscl_freeopen(op, 1);
1614                 }
1615                 nfscl_freeopenowner(owp, 1);
1616         }
1617         nfscl_freealllocks(&dp->nfsdl_lock, 1);
1618 }
1619
1620 /*
1621  * Free a delegation.
1622  */
1623 static void
1624 nfscl_freedeleg(struct nfscldeleghead *hdp, struct nfscldeleg *dp)
1625 {
1626
1627         TAILQ_REMOVE(hdp, dp, nfsdl_list);
1628         LIST_REMOVE(dp, nfsdl_hash);
1629         free(dp, M_NFSCLDELEG);
1630         nfsstatsv1.cldelegates--;
1631         nfscl_delegcnt--;
1632 }
1633
1634 /*
1635  * Free up all state related to this client structure.
1636  */
1637 static void
1638 nfscl_cleanclient(struct nfsclclient *clp)
1639 {
1640         struct nfsclowner *owp, *nowp;
1641         struct nfsclopen *op, *nop;
1642         struct nfscllayout *lyp, *nlyp;
1643         struct nfscldevinfo *dip, *ndip;
1644
1645         TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp)
1646                 nfscl_freelayout(lyp);
1647
1648         LIST_FOREACH_SAFE(dip, &clp->nfsc_devinfo, nfsdi_list, ndip)
1649                 nfscl_freedevinfo(dip);
1650
1651         /* Now, all the OpenOwners, etc. */
1652         LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1653                 LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
1654                         nfscl_freeopen(op, 0);
1655                 }
1656                 nfscl_freeopenowner(owp, 0);
1657         }
1658 }
1659
1660 /*
1661  * Called when an NFSERR_EXPIRED is received from the server.
1662  */
1663 static void
1664 nfscl_expireclient(struct nfsclclient *clp, struct nfsmount *nmp,
1665     struct ucred *cred, NFSPROC_T *p)
1666 {
1667         struct nfsclowner *owp, *nowp, *towp;
1668         struct nfsclopen *op, *nop, *top;
1669         struct nfscldeleg *dp, *ndp;
1670         int ret, printed = 0;
1671
1672         /*
1673          * First, merge locally issued Opens into the list for the server.
1674          */
1675         dp = TAILQ_FIRST(&clp->nfsc_deleg);
1676         while (dp != NULL) {
1677             ndp = TAILQ_NEXT(dp, nfsdl_list);
1678             owp = LIST_FIRST(&dp->nfsdl_owner);
1679             while (owp != NULL) {
1680                 nowp = LIST_NEXT(owp, nfsow_list);
1681                 op = LIST_FIRST(&owp->nfsow_open);
1682                 if (op != NULL) {
1683                     if (LIST_NEXT(op, nfso_list) != NULL)
1684                         panic("nfsclexp");
1685                     LIST_FOREACH(towp, &clp->nfsc_owner, nfsow_list) {
1686                         if (!NFSBCMP(towp->nfsow_owner, owp->nfsow_owner,
1687                             NFSV4CL_LOCKNAMELEN))
1688                             break;
1689                     }
1690                     if (towp != NULL) {
1691                         /* Merge opens in */
1692                         LIST_FOREACH(top, &towp->nfsow_open, nfso_list) {
1693                             if (top->nfso_fhlen == op->nfso_fhlen &&
1694                                 !NFSBCMP(top->nfso_fh, op->nfso_fh,
1695                                  op->nfso_fhlen)) {
1696                                 top->nfso_mode |= op->nfso_mode;
1697                                 top->nfso_opencnt += op->nfso_opencnt;
1698                                 break;
1699                             }
1700                         }
1701                         if (top == NULL) {
1702                             /* Just add the open to the owner list */
1703                             LIST_REMOVE(op, nfso_list);
1704                             op->nfso_own = towp;
1705                             LIST_INSERT_HEAD(&towp->nfsow_open, op, nfso_list);
1706                             nfsstatsv1.cllocalopens--;
1707                             nfsstatsv1.clopens++;
1708                         }
1709                     } else {
1710                         /* Just add the openowner to the client list */
1711                         LIST_REMOVE(owp, nfsow_list);
1712                         owp->nfsow_clp = clp;
1713                         LIST_INSERT_HEAD(&clp->nfsc_owner, owp, nfsow_list);
1714                         nfsstatsv1.cllocalopenowners--;
1715                         nfsstatsv1.clopenowners++;
1716                         nfsstatsv1.cllocalopens--;
1717                         nfsstatsv1.clopens++;
1718                     }
1719                 }
1720                 owp = nowp;
1721             }
1722             if (!printed && !LIST_EMPTY(&dp->nfsdl_lock)) {
1723                 printed = 1;
1724                 printf("nfsv4 expired locks lost\n");
1725             }
1726             nfscl_cleandeleg(dp);
1727             nfscl_freedeleg(&clp->nfsc_deleg, dp);
1728             dp = ndp;
1729         }
1730         if (!TAILQ_EMPTY(&clp->nfsc_deleg))
1731             panic("nfsclexp");
1732
1733         /*
1734          * Now, try and reopen against the server.
1735          */
1736         LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1737                 owp->nfsow_seqid = 0;
1738                 LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
1739                         ret = nfscl_expireopen(clp, op, nmp, cred, p);
1740                         if (ret && !printed) {
1741                                 printed = 1;
1742                                 printf("nfsv4 expired locks lost\n");
1743                         }
1744                 }
1745                 if (LIST_EMPTY(&owp->nfsow_open))
1746                         nfscl_freeopenowner(owp, 0);
1747         }
1748 }
1749
1750 /*
1751  * This function must be called after the process represented by "own" has
1752  * exited. Must be called with CLSTATE lock held.
1753  */
1754 static void
1755 nfscl_cleanup_common(struct nfsclclient *clp, u_int8_t *own)
1756 {
1757         struct nfsclowner *owp, *nowp;
1758         struct nfscllockowner *lp, *nlp;
1759         struct nfscldeleg *dp;
1760
1761         /* First, get rid of local locks on delegations. */
1762         TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
1763                 LIST_FOREACH_SAFE(lp, &dp->nfsdl_lock, nfsl_list, nlp) {
1764                     if (!NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
1765                         if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED))
1766                             panic("nfscllckw");
1767                         nfscl_freelockowner(lp, 1);
1768                     }
1769                 }
1770         }
1771         owp = LIST_FIRST(&clp->nfsc_owner);
1772         while (owp != NULL) {
1773                 nowp = LIST_NEXT(owp, nfsow_list);
1774                 if (!NFSBCMP(owp->nfsow_owner, own,
1775                     NFSV4CL_LOCKNAMELEN)) {
1776                         /*
1777                          * If there are children that haven't closed the
1778                          * file descriptors yet, the opens will still be
1779                          * here. For that case, let the renew thread clear
1780                          * out the OpenOwner later.
1781                          */
1782                         if (LIST_EMPTY(&owp->nfsow_open))
1783                                 nfscl_freeopenowner(owp, 0);
1784                         else
1785                                 owp->nfsow_defunct = 1;
1786                 }
1787                 owp = nowp;
1788         }
1789 }
1790
1791 /*
1792  * Find open/lock owners for processes that have exited.
1793  */
1794 static void
1795 nfscl_cleanupkext(struct nfsclclient *clp, struct nfscllockownerfhhead *lhp)
1796 {
1797         struct nfsclowner *owp, *nowp;
1798         struct nfsclopen *op;
1799         struct nfscllockowner *lp, *nlp;
1800         struct nfscldeleg *dp;
1801
1802         /*
1803          * All the pidhash locks must be acquired, since they are sx locks
1804          * and must be acquired before the mutexes.  The pid(s) that will
1805          * be used aren't known yet, so all the locks need to be acquired.
1806          * Fortunately, this function is only performed once/sec.
1807          */
1808         pidhash_slockall();
1809         NFSLOCKCLSTATE();
1810         LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1811                 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1812                         LIST_FOREACH_SAFE(lp, &op->nfso_lock, nfsl_list, nlp) {
1813                                 if (LIST_EMPTY(&lp->nfsl_lock))
1814                                         nfscl_emptylockowner(lp, lhp);
1815                         }
1816                 }
1817                 if (nfscl_procdoesntexist(owp->nfsow_owner))
1818                         nfscl_cleanup_common(clp, owp->nfsow_owner);
1819         }
1820
1821         /*
1822          * For the single open_owner case, these lock owners need to be
1823          * checked to see if they still exist separately.
1824          * This is because nfscl_procdoesntexist() never returns true for
1825          * the single open_owner so that the above doesn't ever call
1826          * nfscl_cleanup_common().
1827          */
1828         TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
1829                 LIST_FOREACH_SAFE(lp, &dp->nfsdl_lock, nfsl_list, nlp) {
1830                         if (nfscl_procdoesntexist(lp->nfsl_owner))
1831                                 nfscl_cleanup_common(clp, lp->nfsl_owner);
1832                 }
1833         }
1834         NFSUNLOCKCLSTATE();
1835         pidhash_sunlockall();
1836 }
1837
1838 /*
1839  * Take the empty lock owner and move it to the local lhp list if the
1840  * associated process no longer exists.
1841  */
1842 static void
1843 nfscl_emptylockowner(struct nfscllockowner *lp,
1844     struct nfscllockownerfhhead *lhp)
1845 {
1846         struct nfscllockownerfh *lfhp, *mylfhp;
1847         struct nfscllockowner *nlp;
1848         int fnd_it;
1849
1850         /* If not a Posix lock owner, just return. */
1851         if ((lp->nfsl_lockflags & F_POSIX) == 0)
1852                 return;
1853
1854         fnd_it = 0;
1855         mylfhp = NULL;
1856         /*
1857          * First, search to see if this lock owner is already in the list.
1858          * If it is, then the associated process no longer exists.
1859          */
1860         SLIST_FOREACH(lfhp, lhp, nfslfh_list) {
1861                 if (lfhp->nfslfh_len == lp->nfsl_open->nfso_fhlen &&
1862                     !NFSBCMP(lfhp->nfslfh_fh, lp->nfsl_open->nfso_fh,
1863                     lfhp->nfslfh_len))
1864                         mylfhp = lfhp;
1865                 LIST_FOREACH(nlp, &lfhp->nfslfh_lock, nfsl_list)
1866                         if (!NFSBCMP(nlp->nfsl_owner, lp->nfsl_owner,
1867                             NFSV4CL_LOCKNAMELEN))
1868                                 fnd_it = 1;
1869         }
1870         /* If not found, check if process still exists. */
1871         if (fnd_it == 0 && nfscl_procdoesntexist(lp->nfsl_owner) == 0)
1872                 return;
1873
1874         /* Move the lock owner over to the local list. */
1875         if (mylfhp == NULL) {
1876                 mylfhp = malloc(sizeof(struct nfscllockownerfh), M_TEMP,
1877                     M_NOWAIT);
1878                 if (mylfhp == NULL)
1879                         return;
1880                 mylfhp->nfslfh_len = lp->nfsl_open->nfso_fhlen;
1881                 NFSBCOPY(lp->nfsl_open->nfso_fh, mylfhp->nfslfh_fh,
1882                     mylfhp->nfslfh_len);
1883                 LIST_INIT(&mylfhp->nfslfh_lock);
1884                 SLIST_INSERT_HEAD(lhp, mylfhp, nfslfh_list);
1885         }
1886         LIST_REMOVE(lp, nfsl_list);
1887         LIST_INSERT_HEAD(&mylfhp->nfslfh_lock, lp, nfsl_list);
1888 }
1889
1890 static int      fake_global;    /* Used to force visibility of MNTK_UNMOUNTF */
1891 /*
1892  * Called from nfs umount to free up the clientid.
1893  */
1894 void
1895 nfscl_umount(struct nfsmount *nmp, NFSPROC_T *p)
1896 {
1897         struct nfsclclient *clp;
1898         struct ucred *cred;
1899         int igotlock;
1900
1901         /*
1902          * For the case that matters, this is the thread that set
1903          * MNTK_UNMOUNTF, so it will see it set. The code that follows is
1904          * done to ensure that any thread executing nfscl_getcl() after
1905          * this time, will see MNTK_UNMOUNTF set. nfscl_getcl() uses the
1906          * mutex for NFSLOCKCLSTATE(), so it is "m" for the following
1907          * explanation, courtesy of Alan Cox.
1908          * What follows is a snippet from Alan Cox's email at:
1909          * https://docs.FreeBSD.org/cgi/mid.cgi?BANLkTikR3d65zPHo9==08ZfJ2vmqZucEvw
1910          *
1911          * 1. Set MNTK_UNMOUNTF
1912          * 2. Acquire a standard FreeBSD mutex "m".
1913          * 3. Update some data structures.
1914          * 4. Release mutex "m".
1915          *
1916          * Then, other threads that acquire "m" after step 4 has occurred will
1917          * see MNTK_UNMOUNTF as set.  But, other threads that beat thread X to
1918          * step 2 may or may not see MNTK_UNMOUNTF as set.
1919          */
1920         NFSLOCKCLSTATE();
1921         if ((nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) {
1922                 fake_global++;
1923                 NFSUNLOCKCLSTATE();
1924                 NFSLOCKCLSTATE();
1925         }
1926
1927         clp = nmp->nm_clp;
1928         if (clp != NULL) {
1929                 if ((clp->nfsc_flags & NFSCLFLAGS_INITED) == 0)
1930                         panic("nfscl umount");
1931
1932                 /*
1933                  * First, handshake with the nfscl renew thread, to terminate
1934                  * it.
1935                  */
1936                 clp->nfsc_flags |= NFSCLFLAGS_UMOUNT;
1937                 while (clp->nfsc_flags & NFSCLFLAGS_HASTHREAD)
1938                         (void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT,
1939                             "nfsclumnt", hz);
1940
1941                 /*
1942                  * Now, get the exclusive lock on the client state, so
1943                  * that no uses of the state are still in progress.
1944                  */
1945                 do {
1946                         igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
1947                             NFSCLSTATEMUTEXPTR, NULL);
1948                 } while (!igotlock);
1949                 NFSUNLOCKCLSTATE();
1950
1951                 /*
1952                  * Free up all the state. It will expire on the server, but
1953                  * maybe we should do a SetClientId/SetClientIdConfirm so
1954                  * the server throws it away?
1955                  */
1956                 LIST_REMOVE(clp, nfsc_list);
1957                 nfscl_delegreturnall(clp, p);
1958                 cred = newnfs_getcred();
1959                 if (NFSHASNFSV4N(nmp)) {
1960                         (void)nfsrpc_destroysession(nmp, clp, cred, p);
1961                         (void)nfsrpc_destroyclient(nmp, clp, cred, p);
1962                 } else
1963                         (void)nfsrpc_setclient(nmp, clp, 0, NULL, cred, p);
1964                 nfscl_cleanclient(clp);
1965                 nmp->nm_clp = NULL;
1966                 NFSFREECRED(cred);
1967                 free(clp, M_NFSCLCLIENT);
1968         } else
1969                 NFSUNLOCKCLSTATE();
1970 }
1971
1972 /*
1973  * This function is called when a server replies with NFSERR_STALECLIENTID
1974  * NFSERR_STALESTATEID or NFSERR_BADSESSION. It traverses the clientid lists,
1975  * doing Opens and Locks with reclaim. If these fail, it deletes the
1976  * corresponding state.
1977  */
1978 static void
1979 nfscl_recover(struct nfsclclient *clp, bool *retokp, struct ucred *cred,
1980     NFSPROC_T *p)
1981 {
1982         struct nfsclowner *owp, *nowp;
1983         struct nfsclopen *op, *nop;
1984         struct nfscllockowner *lp, *nlp;
1985         struct nfscllock *lop, *nlop;
1986         struct nfscldeleg *dp, *ndp, *tdp;
1987         struct nfsmount *nmp;
1988         struct ucred *tcred;
1989         struct nfsclopenhead extra_open;
1990         struct nfscldeleghead extra_deleg;
1991         struct nfsreq *rep;
1992         u_int64_t len;
1993         u_int32_t delegtype = NFSV4OPEN_DELEGATEWRITE, mode;
1994         int i, igotlock = 0, error, trycnt, firstlock;
1995         struct nfscllayout *lyp, *nlyp;
1996
1997         /*
1998          * First, lock the client structure, so everyone else will
1999          * block when trying to use state.
2000          */
2001         NFSLOCKCLSTATE();
2002         clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG;
2003         do {
2004                 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
2005                     NFSCLSTATEMUTEXPTR, NULL);
2006         } while (!igotlock);
2007         NFSUNLOCKCLSTATE();
2008
2009         nmp = clp->nfsc_nmp;
2010         if (nmp == NULL)
2011                 panic("nfscl recover");
2012
2013         /*
2014          * For now, just get rid of all layouts. There may be a need
2015          * to do LayoutCommit Ops with reclaim == true later.
2016          */
2017         TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp)
2018                 nfscl_freelayout(lyp);
2019         TAILQ_INIT(&clp->nfsc_layout);
2020         for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++)
2021                 LIST_INIT(&clp->nfsc_layouthash[i]);
2022
2023         trycnt = 5;
2024         tcred = NULL;
2025         do {
2026                 error = nfsrpc_setclient(nmp, clp, 1, retokp, cred, p);
2027         } while ((error == NFSERR_STALECLIENTID ||
2028              error == NFSERR_BADSESSION ||
2029              error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
2030         if (error) {
2031                 NFSLOCKCLSTATE();
2032                 clp->nfsc_flags &= ~(NFSCLFLAGS_RECOVER |
2033                     NFSCLFLAGS_RECVRINPROG);
2034                 wakeup(&clp->nfsc_flags);
2035                 nfsv4_unlock(&clp->nfsc_lock, 0);
2036                 NFSUNLOCKCLSTATE();
2037                 return;
2038         }
2039         clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
2040         clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2041
2042         /*
2043          * Mark requests already queued on the server, so that they don't
2044          * initiate another recovery cycle. Any requests already in the
2045          * queue that handle state information will have the old stale
2046          * clientid/stateid and will get a NFSERR_STALESTATEID,
2047          * NFSERR_STALECLIENTID or NFSERR_BADSESSION reply from the server.
2048          * This will be translated to NFSERR_STALEDONTRECOVER when
2049          * R_DONTRECOVER is set.
2050          */
2051         NFSLOCKREQ();
2052         TAILQ_FOREACH(rep, &nfsd_reqq, r_chain) {
2053                 if (rep->r_nmp == nmp)
2054                         rep->r_flags |= R_DONTRECOVER;
2055         }
2056         NFSUNLOCKREQ();
2057
2058         /*
2059          * If nfsrpc_setclient() returns *retokp == true,
2060          * no more recovery is needed.
2061          */
2062         if (*retokp)
2063                 goto out;
2064
2065         /*
2066          * Now, mark all delegations "need reclaim".
2067          */
2068         TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list)
2069                 dp->nfsdl_flags |= NFSCLDL_NEEDRECLAIM;
2070
2071         TAILQ_INIT(&extra_deleg);
2072         LIST_INIT(&extra_open);
2073         /*
2074          * Now traverse the state lists, doing Open and Lock Reclaims.
2075          */
2076         tcred = newnfs_getcred();
2077         owp = LIST_FIRST(&clp->nfsc_owner);
2078         while (owp != NULL) {
2079             nowp = LIST_NEXT(owp, nfsow_list);
2080             owp->nfsow_seqid = 0;
2081             op = LIST_FIRST(&owp->nfsow_open);
2082             while (op != NULL) {
2083                 nop = LIST_NEXT(op, nfso_list);
2084                 if (error != NFSERR_NOGRACE && error != NFSERR_BADSESSION) {
2085                     /* Search for a delegation to reclaim with the open */
2086                     TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2087                         if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
2088                             continue;
2089                         if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
2090                             mode = NFSV4OPEN_ACCESSWRITE;
2091                             delegtype = NFSV4OPEN_DELEGATEWRITE;
2092                         } else {
2093                             mode = NFSV4OPEN_ACCESSREAD;
2094                             delegtype = NFSV4OPEN_DELEGATEREAD;
2095                         }
2096                         if ((op->nfso_mode & mode) == mode &&
2097                             op->nfso_fhlen == dp->nfsdl_fhlen &&
2098                             !NFSBCMP(op->nfso_fh, dp->nfsdl_fh, op->nfso_fhlen))
2099                             break;
2100                     }
2101                     ndp = dp;
2102                     if (dp == NULL)
2103                         delegtype = NFSV4OPEN_DELEGATENONE;
2104                     newnfs_copycred(&op->nfso_cred, tcred);
2105                     error = nfscl_tryopen(nmp, NULL, op->nfso_fh,
2106                         op->nfso_fhlen, op->nfso_fh, op->nfso_fhlen,
2107                         op->nfso_mode, op, NULL, 0, &ndp, 1, delegtype,
2108                         tcred, p);
2109                     if (!error) {
2110                         /* Handle any replied delegation */
2111                         if (ndp != NULL && ((ndp->nfsdl_flags & NFSCLDL_WRITE)
2112                             || NFSMNT_RDONLY(nmp->nm_mountp))) {
2113                             if ((ndp->nfsdl_flags & NFSCLDL_WRITE))
2114                                 mode = NFSV4OPEN_ACCESSWRITE;
2115                             else
2116                                 mode = NFSV4OPEN_ACCESSREAD;
2117                             TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2118                                 if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
2119                                     continue;
2120                                 if ((op->nfso_mode & mode) == mode &&
2121                                     op->nfso_fhlen == dp->nfsdl_fhlen &&
2122                                     !NFSBCMP(op->nfso_fh, dp->nfsdl_fh,
2123                                     op->nfso_fhlen)) {
2124                                     dp->nfsdl_stateid = ndp->nfsdl_stateid;
2125                                     dp->nfsdl_sizelimit = ndp->nfsdl_sizelimit;
2126                                     dp->nfsdl_ace = ndp->nfsdl_ace;
2127                                     dp->nfsdl_change = ndp->nfsdl_change;
2128                                     dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
2129                                     if ((ndp->nfsdl_flags & NFSCLDL_RECALL))
2130                                         dp->nfsdl_flags |= NFSCLDL_RECALL;
2131                                     free(ndp, M_NFSCLDELEG);
2132                                     ndp = NULL;
2133                                     break;
2134                                 }
2135                             }
2136                         }
2137                         if (ndp != NULL)
2138                             TAILQ_INSERT_HEAD(&extra_deleg, ndp, nfsdl_list);
2139
2140                         /* and reclaim all byte range locks */
2141                         lp = LIST_FIRST(&op->nfso_lock);
2142                         while (lp != NULL) {
2143                             nlp = LIST_NEXT(lp, nfsl_list);
2144                             lp->nfsl_seqid = 0;
2145                             firstlock = 1;
2146                             lop = LIST_FIRST(&lp->nfsl_lock);
2147                             while (lop != NULL) {
2148                                 nlop = LIST_NEXT(lop, nfslo_list);
2149                                 if (lop->nfslo_end == NFS64BITSSET)
2150                                     len = NFS64BITSSET;
2151                                 else
2152                                     len = lop->nfslo_end - lop->nfslo_first;
2153                                 error = nfscl_trylock(nmp, NULL,
2154                                     op->nfso_fh, op->nfso_fhlen, lp,
2155                                     firstlock, 1, lop->nfslo_first, len,
2156                                     lop->nfslo_type, tcred, p);
2157                                 if (error != 0)
2158                                     nfscl_freelock(lop, 0);
2159                                 else
2160                                     firstlock = 0;
2161                                 lop = nlop;
2162                             }
2163                             /* If no locks, but a lockowner, just delete it. */
2164                             if (LIST_EMPTY(&lp->nfsl_lock))
2165                                 nfscl_freelockowner(lp, 0);
2166                             lp = nlp;
2167                         }
2168                     }
2169                 }
2170                 if (error != 0 && error != NFSERR_BADSESSION)
2171                     nfscl_freeopen(op, 0);
2172                 op = nop;
2173             }
2174             owp = nowp;
2175         }
2176
2177         /*
2178          * Now, try and get any delegations not yet reclaimed by cobbling
2179          * to-gether an appropriate open.
2180          */
2181         nowp = NULL;
2182         dp = TAILQ_FIRST(&clp->nfsc_deleg);
2183         while (dp != NULL) {
2184             ndp = TAILQ_NEXT(dp, nfsdl_list);
2185             if ((dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM)) {
2186                 if (nowp == NULL) {
2187                     nowp = malloc(
2188                         sizeof (struct nfsclowner), M_NFSCLOWNER, M_WAITOK);
2189                     /*
2190                      * Name must be as long an largest possible
2191                      * NFSV4CL_LOCKNAMELEN. 12 for now.
2192                      */
2193                     NFSBCOPY("RECLAIMDELEG", nowp->nfsow_owner,
2194                         NFSV4CL_LOCKNAMELEN);
2195                     LIST_INIT(&nowp->nfsow_open);
2196                     nowp->nfsow_clp = clp;
2197                     nowp->nfsow_seqid = 0;
2198                     nowp->nfsow_defunct = 0;
2199                     nfscl_lockinit(&nowp->nfsow_rwlock);
2200                 }
2201                 nop = NULL;
2202                 if (error != NFSERR_NOGRACE && error != NFSERR_BADSESSION) {
2203                     nop = malloc(sizeof (struct nfsclopen) +
2204                         dp->nfsdl_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
2205                     nop->nfso_own = nowp;
2206                     if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
2207                         nop->nfso_mode = NFSV4OPEN_ACCESSWRITE;
2208                         delegtype = NFSV4OPEN_DELEGATEWRITE;
2209                     } else {
2210                         nop->nfso_mode = NFSV4OPEN_ACCESSREAD;
2211                         delegtype = NFSV4OPEN_DELEGATEREAD;
2212                     }
2213                     nop->nfso_opencnt = 0;
2214                     nop->nfso_posixlock = 1;
2215                     nop->nfso_fhlen = dp->nfsdl_fhlen;
2216                     NFSBCOPY(dp->nfsdl_fh, nop->nfso_fh, dp->nfsdl_fhlen);
2217                     LIST_INIT(&nop->nfso_lock);
2218                     nop->nfso_stateid.seqid = 0;
2219                     nop->nfso_stateid.other[0] = 0;
2220                     nop->nfso_stateid.other[1] = 0;
2221                     nop->nfso_stateid.other[2] = 0;
2222                     newnfs_copycred(&dp->nfsdl_cred, tcred);
2223                     newnfs_copyincred(tcred, &nop->nfso_cred);
2224                     tdp = NULL;
2225                     error = nfscl_tryopen(nmp, NULL, nop->nfso_fh,
2226                         nop->nfso_fhlen, nop->nfso_fh, nop->nfso_fhlen,
2227                         nop->nfso_mode, nop, NULL, 0, &tdp, 1,
2228                         delegtype, tcred, p);
2229                     if (tdp != NULL) {
2230                         if ((tdp->nfsdl_flags & NFSCLDL_WRITE))
2231                             mode = NFSV4OPEN_ACCESSWRITE;
2232                         else
2233                             mode = NFSV4OPEN_ACCESSREAD;
2234                         if ((nop->nfso_mode & mode) == mode &&
2235                             nop->nfso_fhlen == tdp->nfsdl_fhlen &&
2236                             !NFSBCMP(nop->nfso_fh, tdp->nfsdl_fh,
2237                             nop->nfso_fhlen)) {
2238                             dp->nfsdl_stateid = tdp->nfsdl_stateid;
2239                             dp->nfsdl_sizelimit = tdp->nfsdl_sizelimit;
2240                             dp->nfsdl_ace = tdp->nfsdl_ace;
2241                             dp->nfsdl_change = tdp->nfsdl_change;
2242                             dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
2243                             if ((tdp->nfsdl_flags & NFSCLDL_RECALL))
2244                                 dp->nfsdl_flags |= NFSCLDL_RECALL;
2245                             free(tdp, M_NFSCLDELEG);
2246                         } else {
2247                             TAILQ_INSERT_HEAD(&extra_deleg, tdp, nfsdl_list);
2248                         }
2249                     }
2250                 }
2251                 if (error) {
2252                     if (nop != NULL)
2253                         free(nop, M_NFSCLOPEN);
2254                     /*
2255                      * Couldn't reclaim it, so throw the state
2256                      * away. Ouch!!
2257                      */
2258                     nfscl_cleandeleg(dp);
2259                     nfscl_freedeleg(&clp->nfsc_deleg, dp);
2260                 } else {
2261                     LIST_INSERT_HEAD(&extra_open, nop, nfso_list);
2262                 }
2263             }
2264             dp = ndp;
2265         }
2266
2267         /*
2268          * Now, get rid of extra Opens and Delegations.
2269          */
2270         LIST_FOREACH_SAFE(op, &extra_open, nfso_list, nop) {
2271                 do {
2272                         newnfs_copycred(&op->nfso_cred, tcred);
2273                         error = nfscl_tryclose(op, tcred, nmp, p);
2274                         if (error == NFSERR_GRACE)
2275                                 (void) nfs_catnap(PZERO, error, "nfsexcls");
2276                 } while (error == NFSERR_GRACE);
2277                 LIST_REMOVE(op, nfso_list);
2278                 free(op, M_NFSCLOPEN);
2279         }
2280         if (nowp != NULL)
2281                 free(nowp, M_NFSCLOWNER);
2282
2283         TAILQ_FOREACH_SAFE(dp, &extra_deleg, nfsdl_list, ndp) {
2284                 do {
2285                         newnfs_copycred(&dp->nfsdl_cred, tcred);
2286                         error = nfscl_trydelegreturn(dp, tcred, nmp, p);
2287                         if (error == NFSERR_GRACE)
2288                                 (void) nfs_catnap(PZERO, error, "nfsexdlg");
2289                 } while (error == NFSERR_GRACE);
2290                 TAILQ_REMOVE(&extra_deleg, dp, nfsdl_list);
2291                 free(dp, M_NFSCLDELEG);
2292         }
2293
2294         /* For NFSv4.1 or later, do a RECLAIM_COMPLETE. */
2295         if (NFSHASNFSV4N(nmp))
2296                 (void)nfsrpc_reclaimcomplete(nmp, cred, p);
2297
2298 out:
2299         NFSLOCKCLSTATE();
2300         clp->nfsc_flags &= ~NFSCLFLAGS_RECVRINPROG;
2301         wakeup(&clp->nfsc_flags);
2302         nfsv4_unlock(&clp->nfsc_lock, 0);
2303         NFSUNLOCKCLSTATE();
2304         if (tcred != NULL)
2305                 NFSFREECRED(tcred);
2306 }
2307
2308 /*
2309  * This function is called when a server replies with NFSERR_EXPIRED.
2310  * It deletes all state for the client and does a fresh SetClientId/confirm.
2311  * XXX Someday it should post a signal to the process(es) that hold the
2312  * state, so they know that lock state has been lost.
2313  */
2314 int
2315 nfscl_hasexpired(struct nfsclclient *clp, u_int32_t clidrev, NFSPROC_T *p)
2316 {
2317         struct nfsmount *nmp;
2318         struct ucred *cred;
2319         int igotlock = 0, error, trycnt;
2320
2321         /*
2322          * If the clientid has gone away or a new SetClientid has already
2323          * been done, just return ok.
2324          */
2325         if (clp == NULL || clidrev != clp->nfsc_clientidrev)
2326                 return (0);
2327
2328         /*
2329          * First, lock the client structure, so everyone else will
2330          * block when trying to use state. Also, use NFSCLFLAGS_EXPIREIT so
2331          * that only one thread does the work.
2332          */
2333         NFSLOCKCLSTATE();
2334         clp->nfsc_flags |= NFSCLFLAGS_EXPIREIT;
2335         do {
2336                 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
2337                     NFSCLSTATEMUTEXPTR, NULL);
2338         } while (!igotlock && (clp->nfsc_flags & NFSCLFLAGS_EXPIREIT));
2339         if ((clp->nfsc_flags & NFSCLFLAGS_EXPIREIT) == 0) {
2340                 if (igotlock)
2341                         nfsv4_unlock(&clp->nfsc_lock, 0);
2342                 NFSUNLOCKCLSTATE();
2343                 return (0);
2344         }
2345         clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG;
2346         NFSUNLOCKCLSTATE();
2347
2348         nmp = clp->nfsc_nmp;
2349         if (nmp == NULL)
2350                 panic("nfscl expired");
2351         cred = newnfs_getcred();
2352         trycnt = 5;
2353         do {
2354                 error = nfsrpc_setclient(nmp, clp, 0, NULL, cred, p);
2355         } while ((error == NFSERR_STALECLIENTID ||
2356              error == NFSERR_BADSESSION ||
2357              error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
2358         if (error) {
2359                 NFSLOCKCLSTATE();
2360                 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2361         } else {
2362                 /*
2363                  * Expire the state for the client.
2364                  */
2365                 nfscl_expireclient(clp, nmp, cred, p);
2366                 NFSLOCKCLSTATE();
2367                 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
2368                 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2369         }
2370         clp->nfsc_flags &= ~(NFSCLFLAGS_EXPIREIT | NFSCLFLAGS_RECVRINPROG);
2371         wakeup(&clp->nfsc_flags);
2372         nfsv4_unlock(&clp->nfsc_lock, 0);
2373         NFSUNLOCKCLSTATE();
2374         NFSFREECRED(cred);
2375         return (error);
2376 }
2377
2378 /*
2379  * This function inserts a lock in the list after insert_lop.
2380  */
2381 static void
2382 nfscl_insertlock(struct nfscllockowner *lp, struct nfscllock *new_lop,
2383     struct nfscllock *insert_lop, int local)
2384 {
2385
2386         if ((struct nfscllockowner *)insert_lop == lp)
2387                 LIST_INSERT_HEAD(&lp->nfsl_lock, new_lop, nfslo_list);
2388         else
2389                 LIST_INSERT_AFTER(insert_lop, new_lop, nfslo_list);
2390         if (local)
2391                 nfsstatsv1.cllocallocks++;
2392         else
2393                 nfsstatsv1.cllocks++;
2394 }
2395
2396 /*
2397  * This function updates the locking for a lock owner and given file. It
2398  * maintains a list of lock ranges ordered on increasing file offset that
2399  * are NFSCLLOCK_READ or NFSCLLOCK_WRITE and non-overlapping (aka POSIX style).
2400  * It always adds new_lop to the list and sometimes uses the one pointed
2401  * at by other_lopp.
2402  * Returns 1 if the locks were modified, 0 otherwise.
2403  */
2404 static int
2405 nfscl_updatelock(struct nfscllockowner *lp, struct nfscllock **new_lopp,
2406     struct nfscllock **other_lopp, int local)
2407 {
2408         struct nfscllock *new_lop = *new_lopp;
2409         struct nfscllock *lop, *tlop, *ilop;
2410         struct nfscllock *other_lop;
2411         int unlock = 0, modified = 0;
2412         u_int64_t tmp;
2413
2414         /*
2415          * Work down the list until the lock is merged.
2416          */
2417         if (new_lop->nfslo_type == F_UNLCK)
2418                 unlock = 1;
2419         ilop = (struct nfscllock *)lp;
2420         lop = LIST_FIRST(&lp->nfsl_lock);
2421         while (lop != NULL) {
2422             /*
2423              * Only check locks for this file that aren't before the start of
2424              * new lock's range.
2425              */
2426             if (lop->nfslo_end >= new_lop->nfslo_first) {
2427                 if (new_lop->nfslo_end < lop->nfslo_first) {
2428                     /*
2429                      * If the new lock ends before the start of the
2430                      * current lock's range, no merge, just insert
2431                      * the new lock.
2432                      */
2433                     break;
2434                 }
2435                 if (new_lop->nfslo_type == lop->nfslo_type ||
2436                     (new_lop->nfslo_first <= lop->nfslo_first &&
2437                      new_lop->nfslo_end >= lop->nfslo_end)) {
2438                     /*
2439                      * This lock can be absorbed by the new lock/unlock.
2440                      * This happens when it covers the entire range
2441                      * of the old lock or is contiguous
2442                      * with the old lock and is of the same type or an
2443                      * unlock.
2444                      */
2445                     if (new_lop->nfslo_type != lop->nfslo_type ||
2446                         new_lop->nfslo_first != lop->nfslo_first ||
2447                         new_lop->nfslo_end != lop->nfslo_end)
2448                         modified = 1;
2449                     if (lop->nfslo_first < new_lop->nfslo_first)
2450                         new_lop->nfslo_first = lop->nfslo_first;
2451                     if (lop->nfslo_end > new_lop->nfslo_end)
2452                         new_lop->nfslo_end = lop->nfslo_end;
2453                     tlop = lop;
2454                     lop = LIST_NEXT(lop, nfslo_list);
2455                     nfscl_freelock(tlop, local);
2456                     continue;
2457                 }
2458
2459                 /*
2460                  * All these cases are for contiguous locks that are not the
2461                  * same type, so they can't be merged.
2462                  */
2463                 if (new_lop->nfslo_first <= lop->nfslo_first) {
2464                     /*
2465                      * This case is where the new lock overlaps with the
2466                      * first part of the old lock. Move the start of the
2467                      * old lock to just past the end of the new lock. The
2468                      * new lock will be inserted in front of the old, since
2469                      * ilop hasn't been updated. (We are done now.)
2470                      */
2471                     if (lop->nfslo_first != new_lop->nfslo_end) {
2472                         lop->nfslo_first = new_lop->nfslo_end;
2473                         modified = 1;
2474                     }
2475                     break;
2476                 }
2477                 if (new_lop->nfslo_end >= lop->nfslo_end) {
2478                     /*
2479                      * This case is where the new lock overlaps with the
2480                      * end of the old lock's range. Move the old lock's
2481                      * end to just before the new lock's first and insert
2482                      * the new lock after the old lock.
2483                      * Might not be done yet, since the new lock could
2484                      * overlap further locks with higher ranges.
2485                      */
2486                     if (lop->nfslo_end != new_lop->nfslo_first) {
2487                         lop->nfslo_end = new_lop->nfslo_first;
2488                         modified = 1;
2489                     }
2490                     ilop = lop;
2491                     lop = LIST_NEXT(lop, nfslo_list);
2492                     continue;
2493                 }
2494                 /*
2495                  * The final case is where the new lock's range is in the
2496                  * middle of the current lock's and splits the current lock
2497                  * up. Use *other_lopp to handle the second part of the
2498                  * split old lock range. (We are done now.)
2499                  * For unlock, we use new_lop as other_lop and tmp, since
2500                  * other_lop and new_lop are the same for this case.
2501                  * We noted the unlock case above, so we don't need
2502                  * new_lop->nfslo_type any longer.
2503                  */
2504                 tmp = new_lop->nfslo_first;
2505                 if (unlock) {
2506                     other_lop = new_lop;
2507                     *new_lopp = NULL;
2508                 } else {
2509                     other_lop = *other_lopp;
2510                     *other_lopp = NULL;
2511                 }
2512                 other_lop->nfslo_first = new_lop->nfslo_end;
2513                 other_lop->nfslo_end = lop->nfslo_end;
2514                 other_lop->nfslo_type = lop->nfslo_type;
2515                 lop->nfslo_end = tmp;
2516                 nfscl_insertlock(lp, other_lop, lop, local);
2517                 ilop = lop;
2518                 modified = 1;
2519                 break;
2520             }
2521             ilop = lop;
2522             lop = LIST_NEXT(lop, nfslo_list);
2523             if (lop == NULL)
2524                 break;
2525         }
2526
2527         /*
2528          * Insert the new lock in the list at the appropriate place.
2529          */
2530         if (!unlock) {
2531                 nfscl_insertlock(lp, new_lop, ilop, local);
2532                 *new_lopp = NULL;
2533                 modified = 1;
2534         }
2535         return (modified);
2536 }
2537
2538 /*
2539  * This function must be run as a kernel thread.
2540  * It does Renew Ops and recovery, when required.
2541  */
2542 void
2543 nfscl_renewthread(struct nfsclclient *clp, NFSPROC_T *p)
2544 {
2545         struct nfsclowner *owp, *nowp;
2546         struct nfsclopen *op;
2547         struct nfscllockowner *lp, *nlp;
2548         struct nfscldeleghead dh;
2549         struct nfscldeleg *dp, *ndp;
2550         struct ucred *cred;
2551         u_int32_t clidrev;
2552         int error, cbpathdown, islept, igotlock, ret, clearok;
2553         uint32_t recover_done_time = 0;
2554         time_t mytime;
2555         static time_t prevsec = 0;
2556         struct nfscllockownerfh *lfhp, *nlfhp;
2557         struct nfscllockownerfhhead lfh;
2558         struct nfscllayout *lyp, *nlyp;
2559         struct nfscldevinfo *dip, *ndip;
2560         struct nfscllayouthead rlh;
2561         struct nfsclrecalllayout *recallp;
2562         struct nfsclds *dsp;
2563         bool retok;
2564         struct mount *mp;
2565
2566         cred = newnfs_getcred();
2567         NFSLOCKCLSTATE();
2568         clp->nfsc_flags |= NFSCLFLAGS_HASTHREAD;
2569         mp = clp->nfsc_nmp->nm_mountp;
2570         NFSUNLOCKCLSTATE();
2571         for(;;) {
2572                 newnfs_setroot(cred);
2573                 cbpathdown = 0;
2574                 if (clp->nfsc_flags & NFSCLFLAGS_RECOVER) {
2575                         /*
2576                          * Only allow one full recover within 1/2 of the lease
2577                          * duration (nfsc_renew).
2578                          * retok is value/result.  If passed in set to true,
2579                          * it indicates only a CreateSession operation should
2580                          * be attempted.
2581                          * If it is returned true, it indicates that the
2582                          * recovery only required a CreateSession.
2583                          */
2584                         retok = true;
2585                         if (recover_done_time < NFSD_MONOSEC) {
2586                                 recover_done_time = NFSD_MONOSEC +
2587                                     clp->nfsc_renew;
2588                                 retok = false;
2589                         }
2590                         NFSCL_DEBUG(1, "Doing recovery, only "
2591                             "createsession=%d\n", retok);
2592                         nfscl_recover(clp, &retok, cred, p);
2593                 }
2594                 if (clp->nfsc_expire <= NFSD_MONOSEC &&
2595                     (clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID)) {
2596                         clp->nfsc_expire = NFSD_MONOSEC + clp->nfsc_renew;
2597                         clidrev = clp->nfsc_clientidrev;
2598                         error = nfsrpc_renew(clp, NULL, cred, p);
2599                         if (error == NFSERR_CBPATHDOWN)
2600                             cbpathdown = 1;
2601                         else if (error == NFSERR_STALECLIENTID ||
2602                             error == NFSERR_BADSESSION) {
2603                             NFSLOCKCLSTATE();
2604                             clp->nfsc_flags |= NFSCLFLAGS_RECOVER;
2605                             NFSUNLOCKCLSTATE();
2606                         } else if (error == NFSERR_EXPIRED)
2607                             (void) nfscl_hasexpired(clp, clidrev, p);
2608                 }
2609
2610 checkdsrenew:
2611                 if (NFSHASNFSV4N(clp->nfsc_nmp)) {
2612                         /* Do renews for any DS sessions. */
2613                         NFSLOCKMNT(clp->nfsc_nmp);
2614                         /* Skip first entry, since the MDS is handled above. */
2615                         dsp = TAILQ_FIRST(&clp->nfsc_nmp->nm_sess);
2616                         if (dsp != NULL)
2617                                 dsp = TAILQ_NEXT(dsp, nfsclds_list);
2618                         while (dsp != NULL) {
2619                                 if (dsp->nfsclds_expire <= NFSD_MONOSEC &&
2620                                     dsp->nfsclds_sess.nfsess_defunct == 0) {
2621                                         dsp->nfsclds_expire = NFSD_MONOSEC +
2622                                             clp->nfsc_renew;
2623                                         NFSUNLOCKMNT(clp->nfsc_nmp);
2624                                         (void)nfsrpc_renew(clp, dsp, cred, p);
2625                                         goto checkdsrenew;
2626                                 }
2627                                 dsp = TAILQ_NEXT(dsp, nfsclds_list);
2628                         }
2629                         NFSUNLOCKMNT(clp->nfsc_nmp);
2630                 }
2631
2632                 TAILQ_INIT(&dh);
2633                 NFSLOCKCLSTATE();
2634                 if (cbpathdown)
2635                         /* It's a Total Recall! */
2636                         nfscl_totalrecall(clp);
2637
2638                 /*
2639                  * Now, handle defunct owners.
2640                  */
2641                 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
2642                         if (LIST_EMPTY(&owp->nfsow_open)) {
2643                                 if (owp->nfsow_defunct != 0)
2644                                         nfscl_freeopenowner(owp, 0);
2645                         }
2646                 }
2647
2648                 /*
2649                  * Do the recall on any delegations. To avoid trouble, always
2650                  * come back up here after having slept.
2651                  */
2652                 igotlock = 0;
2653 tryagain:
2654                 dp = TAILQ_FIRST(&clp->nfsc_deleg);
2655                 while (dp != NULL) {
2656                         ndp = TAILQ_NEXT(dp, nfsdl_list);
2657                         if ((dp->nfsdl_flags & NFSCLDL_RECALL)) {
2658                                 /*
2659                                  * Wait for outstanding I/O ops to be done.
2660                                  */
2661                                 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
2662                                     if (igotlock) {
2663                                         nfsv4_unlock(&clp->nfsc_lock, 0);
2664                                         igotlock = 0;
2665                                     }
2666                                     dp->nfsdl_rwlock.nfslock_lock |=
2667                                         NFSV4LOCK_WANTED;
2668                                     msleep(&dp->nfsdl_rwlock,
2669                                         NFSCLSTATEMUTEXPTR, PVFS, "nfscld",
2670                                         5 * hz);
2671                                     if (NFSCL_FORCEDISM(mp))
2672                                         goto terminate;
2673                                     goto tryagain;
2674                                 }
2675                                 while (!igotlock) {
2676                                     igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
2677                                         &islept, NFSCLSTATEMUTEXPTR, mp);
2678                                     if (igotlock == 0 && NFSCL_FORCEDISM(mp))
2679                                         goto terminate;
2680                                     if (islept)
2681                                         goto tryagain;
2682                                 }
2683                                 NFSUNLOCKCLSTATE();
2684                                 newnfs_copycred(&dp->nfsdl_cred, cred);
2685                                 ret = nfscl_recalldeleg(clp, clp->nfsc_nmp, dp,
2686                                     NULL, cred, p, 1);
2687                                 if (!ret) {
2688                                     nfscl_cleandeleg(dp);
2689                                     TAILQ_REMOVE(&clp->nfsc_deleg, dp,
2690                                         nfsdl_list);
2691                                     LIST_REMOVE(dp, nfsdl_hash);
2692                                     TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
2693                                     nfscl_delegcnt--;
2694                                     nfsstatsv1.cldelegates--;
2695                                 }
2696                                 NFSLOCKCLSTATE();
2697                         }
2698                         dp = ndp;
2699                 }
2700
2701                 /*
2702                  * Clear out old delegations, if we are above the high water
2703                  * mark. Only clear out ones with no state related to them.
2704                  * The tailq list is in LRU order.
2705                  */
2706                 dp = TAILQ_LAST(&clp->nfsc_deleg, nfscldeleghead);
2707                 while (nfscl_delegcnt > nfscl_deleghighwater && dp != NULL) {
2708                     ndp = TAILQ_PREV(dp, nfscldeleghead, nfsdl_list);
2709                     if (dp->nfsdl_rwlock.nfslock_usecnt == 0 &&
2710                         dp->nfsdl_rwlock.nfslock_lock == 0 &&
2711                         dp->nfsdl_timestamp < NFSD_MONOSEC &&
2712                         (dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_ZAPPED |
2713                           NFSCLDL_NEEDRECLAIM | NFSCLDL_DELEGRET)) == 0) {
2714                         clearok = 1;
2715                         LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
2716                             op = LIST_FIRST(&owp->nfsow_open);
2717                             if (op != NULL) {
2718                                 clearok = 0;
2719                                 break;
2720                             }
2721                         }
2722                         if (clearok) {
2723                             LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
2724                                 if (!LIST_EMPTY(&lp->nfsl_lock)) {
2725                                     clearok = 0;
2726                                     break;
2727                                 }
2728                             }
2729                         }
2730                         if (clearok) {
2731                             TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
2732                             LIST_REMOVE(dp, nfsdl_hash);
2733                             TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
2734                             nfscl_delegcnt--;
2735                             nfsstatsv1.cldelegates--;
2736                         }
2737                     }
2738                     dp = ndp;
2739                 }
2740                 if (igotlock)
2741                         nfsv4_unlock(&clp->nfsc_lock, 0);
2742
2743                 /*
2744                  * Do the recall on any layouts. To avoid trouble, always
2745                  * come back up here after having slept.
2746                  */
2747                 TAILQ_INIT(&rlh);
2748 tryagain2:
2749                 TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp) {
2750                         if ((lyp->nfsly_flags & NFSLY_RECALL) != 0) {
2751                                 /*
2752                                  * Wait for outstanding I/O ops to be done.
2753                                  */
2754                                 if (lyp->nfsly_lock.nfslock_usecnt > 0 ||
2755                                     (lyp->nfsly_lock.nfslock_lock &
2756                                      NFSV4LOCK_LOCK) != 0) {
2757                                         lyp->nfsly_lock.nfslock_lock |=
2758                                             NFSV4LOCK_WANTED;
2759                                         msleep(&lyp->nfsly_lock.nfslock_lock,
2760                                             NFSCLSTATEMUTEXPTR, PVFS, "nfslyp",
2761                                             5 * hz);
2762                                         if (NFSCL_FORCEDISM(mp))
2763                                             goto terminate;
2764                                         goto tryagain2;
2765                                 }
2766                                 /* Move the layout to the recall list. */
2767                                 TAILQ_REMOVE(&clp->nfsc_layout, lyp,
2768                                     nfsly_list);
2769                                 LIST_REMOVE(lyp, nfsly_hash);
2770                                 TAILQ_INSERT_HEAD(&rlh, lyp, nfsly_list);
2771
2772                                 /* Handle any layout commits. */
2773                                 if (!NFSHASNOLAYOUTCOMMIT(clp->nfsc_nmp) &&
2774                                     (lyp->nfsly_flags & NFSLY_WRITTEN) != 0) {
2775                                         lyp->nfsly_flags &= ~NFSLY_WRITTEN;
2776                                         NFSUNLOCKCLSTATE();
2777                                         NFSCL_DEBUG(3, "do layoutcommit\n");
2778                                         nfscl_dolayoutcommit(clp->nfsc_nmp, lyp,
2779                                             cred, p);
2780                                         NFSLOCKCLSTATE();
2781                                         goto tryagain2;
2782                                 }
2783                         }
2784                 }
2785
2786                 /* Now, look for stale layouts. */
2787                 lyp = TAILQ_LAST(&clp->nfsc_layout, nfscllayouthead);
2788                 while (lyp != NULL) {
2789                         nlyp = TAILQ_PREV(lyp, nfscllayouthead, nfsly_list);
2790                         if (lyp->nfsly_timestamp < NFSD_MONOSEC &&
2791                             (lyp->nfsly_flags & NFSLY_RECALL) == 0 &&
2792                             lyp->nfsly_lock.nfslock_usecnt == 0 &&
2793                             lyp->nfsly_lock.nfslock_lock == 0) {
2794                                 NFSCL_DEBUG(4, "ret stale lay=%d\n",
2795                                     nfscl_layoutcnt);
2796                                 recallp = malloc(sizeof(*recallp),
2797                                     M_NFSLAYRECALL, M_NOWAIT);
2798                                 if (recallp == NULL)
2799                                         break;
2800                                 (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE,
2801                                     lyp, NFSLAYOUTIOMODE_ANY, 0, UINT64_MAX,
2802                                     lyp->nfsly_stateid.seqid, 0, 0, NULL,
2803                                     recallp);
2804                         }
2805                         lyp = nlyp;
2806                 }
2807
2808                 /*
2809                  * Free up any unreferenced device info structures.
2810                  */
2811                 LIST_FOREACH_SAFE(dip, &clp->nfsc_devinfo, nfsdi_list, ndip) {
2812                         if (dip->nfsdi_layoutrefs == 0 &&
2813                             dip->nfsdi_refcnt == 0) {
2814                                 NFSCL_DEBUG(4, "freeing devinfo\n");
2815                                 LIST_REMOVE(dip, nfsdi_list);
2816                                 nfscl_freedevinfo(dip);
2817                         }
2818                 }
2819                 NFSUNLOCKCLSTATE();
2820
2821                 /* Do layout return(s), as required. */
2822                 TAILQ_FOREACH_SAFE(lyp, &rlh, nfsly_list, nlyp) {
2823                         TAILQ_REMOVE(&rlh, lyp, nfsly_list);
2824                         NFSCL_DEBUG(4, "ret layout\n");
2825                         nfscl_layoutreturn(clp->nfsc_nmp, lyp, cred, p);
2826                         nfscl_freelayout(lyp);
2827                 }
2828
2829                 /*
2830                  * Delegreturn any delegations cleaned out or recalled.
2831                  */
2832                 TAILQ_FOREACH_SAFE(dp, &dh, nfsdl_list, ndp) {
2833                         newnfs_copycred(&dp->nfsdl_cred, cred);
2834                         (void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
2835                         TAILQ_REMOVE(&dh, dp, nfsdl_list);
2836                         free(dp, M_NFSCLDELEG);
2837                 }
2838
2839                 SLIST_INIT(&lfh);
2840                 /*
2841                  * Call nfscl_cleanupkext() once per second to check for
2842                  * open/lock owners where the process has exited.
2843                  */
2844                 mytime = NFSD_MONOSEC;
2845                 if (prevsec != mytime) {
2846                         prevsec = mytime;
2847                         nfscl_cleanupkext(clp, &lfh);
2848                 }
2849
2850                 /*
2851                  * Do a ReleaseLockOwner for all lock owners where the
2852                  * associated process no longer exists, as found by
2853                  * nfscl_cleanupkext().
2854                  */
2855                 newnfs_setroot(cred);
2856                 SLIST_FOREACH_SAFE(lfhp, &lfh, nfslfh_list, nlfhp) {
2857                         LIST_FOREACH_SAFE(lp, &lfhp->nfslfh_lock, nfsl_list,
2858                             nlp) {
2859                                 (void)nfsrpc_rellockown(clp->nfsc_nmp, lp,
2860                                     lfhp->nfslfh_fh, lfhp->nfslfh_len, cred,
2861                                     p);
2862                                 nfscl_freelockowner(lp, 0);
2863                         }
2864                         free(lfhp, M_TEMP);
2865                 }
2866                 SLIST_INIT(&lfh);
2867
2868                 NFSLOCKCLSTATE();
2869                 if ((clp->nfsc_flags & NFSCLFLAGS_RECOVER) == 0)
2870                         (void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT, "nfscl",
2871                             hz);
2872 terminate:
2873                 if (clp->nfsc_flags & NFSCLFLAGS_UMOUNT) {
2874                         clp->nfsc_flags &= ~NFSCLFLAGS_HASTHREAD;
2875                         NFSUNLOCKCLSTATE();
2876                         NFSFREECRED(cred);
2877                         wakeup((caddr_t)clp);
2878                         return;
2879                 }
2880                 NFSUNLOCKCLSTATE();
2881         }
2882 }
2883
2884 /*
2885  * Initiate state recovery. Called when NFSERR_STALECLIENTID,
2886  * NFSERR_STALESTATEID or NFSERR_BADSESSION is received.
2887  */
2888 void
2889 nfscl_initiate_recovery(struct nfsclclient *clp)
2890 {
2891
2892         if (clp == NULL)
2893                 return;
2894         NFSLOCKCLSTATE();
2895         clp->nfsc_flags |= NFSCLFLAGS_RECOVER;
2896         NFSUNLOCKCLSTATE();
2897         wakeup((caddr_t)clp);
2898 }
2899
2900 /*
2901  * Dump out the state stuff for debugging.
2902  */
2903 void
2904 nfscl_dumpstate(struct nfsmount *nmp, int openowner, int opens,
2905     int lockowner, int locks)
2906 {
2907         struct nfsclclient *clp;
2908         struct nfsclowner *owp;
2909         struct nfsclopen *op;
2910         struct nfscllockowner *lp;
2911         struct nfscllock *lop;
2912         struct nfscldeleg *dp;
2913
2914         clp = nmp->nm_clp;
2915         if (clp == NULL) {
2916                 printf("nfscl dumpstate NULL clp\n");
2917                 return;
2918         }
2919         NFSLOCKCLSTATE();
2920         TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2921           LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
2922             if (openowner && !LIST_EMPTY(&owp->nfsow_open))
2923                 printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
2924                     owp->nfsow_owner[0], owp->nfsow_owner[1],
2925                     owp->nfsow_owner[2], owp->nfsow_owner[3],
2926                     owp->nfsow_seqid);
2927             LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
2928                 if (opens)
2929                     printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
2930                         op->nfso_stateid.other[0], op->nfso_stateid.other[1],
2931                         op->nfso_stateid.other[2], op->nfso_opencnt,
2932                         op->nfso_fh[12]);
2933                 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
2934                     if (lockowner)
2935                         printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
2936                             lp->nfsl_owner[0], lp->nfsl_owner[1],
2937                             lp->nfsl_owner[2], lp->nfsl_owner[3],
2938                             lp->nfsl_seqid,
2939                             lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
2940                             lp->nfsl_stateid.other[2]);
2941                     LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
2942                         if (locks)
2943 #ifdef __FreeBSD__
2944                             printf("lck typ=%d fst=%ju end=%ju\n",
2945                                 lop->nfslo_type, (intmax_t)lop->nfslo_first,
2946                                 (intmax_t)lop->nfslo_end);
2947 #else
2948                             printf("lck typ=%d fst=%qd end=%qd\n",
2949                                 lop->nfslo_type, lop->nfslo_first,
2950                                 lop->nfslo_end);
2951 #endif
2952                     }
2953                 }
2954             }
2955           }
2956         }
2957         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
2958             if (openowner && !LIST_EMPTY(&owp->nfsow_open))
2959                 printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
2960                     owp->nfsow_owner[0], owp->nfsow_owner[1],
2961                     owp->nfsow_owner[2], owp->nfsow_owner[3],
2962                     owp->nfsow_seqid);
2963             LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
2964                 if (opens)
2965                     printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
2966                         op->nfso_stateid.other[0], op->nfso_stateid.other[1],
2967                         op->nfso_stateid.other[2], op->nfso_opencnt,
2968                         op->nfso_fh[12]);
2969                 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
2970                     if (lockowner)
2971                         printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
2972                             lp->nfsl_owner[0], lp->nfsl_owner[1],
2973                             lp->nfsl_owner[2], lp->nfsl_owner[3],
2974                             lp->nfsl_seqid,
2975                             lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
2976                             lp->nfsl_stateid.other[2]);
2977                     LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
2978                         if (locks)
2979 #ifdef __FreeBSD__
2980                             printf("lck typ=%d fst=%ju end=%ju\n",
2981                                 lop->nfslo_type, (intmax_t)lop->nfslo_first,
2982                                 (intmax_t)lop->nfslo_end);
2983 #else
2984                             printf("lck typ=%d fst=%qd end=%qd\n",
2985                                 lop->nfslo_type, lop->nfslo_first,
2986                                 lop->nfslo_end);
2987 #endif
2988                     }
2989                 }
2990             }
2991         }
2992         NFSUNLOCKCLSTATE();
2993 }
2994
2995 /*
2996  * Check for duplicate open owners and opens.
2997  * (Only used as a diagnostic aid.)
2998  */
2999 void
3000 nfscl_dupopen(vnode_t vp, int dupopens)
3001 {
3002         struct nfsclclient *clp;
3003         struct nfsclowner *owp, *owp2;
3004         struct nfsclopen *op, *op2;
3005         struct nfsfh *nfhp;
3006
3007         clp = VFSTONFS(vp->v_mount)->nm_clp;
3008         if (clp == NULL) {
3009                 printf("nfscl dupopen NULL clp\n");
3010                 return;
3011         }
3012         nfhp = VTONFS(vp)->n_fhp;
3013         NFSLOCKCLSTATE();
3014
3015         /*
3016          * First, search for duplicate owners.
3017          * These should never happen!
3018          */
3019         LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3020             LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3021                 if (owp != owp2 &&
3022                     !NFSBCMP(owp->nfsow_owner, owp2->nfsow_owner,
3023                     NFSV4CL_LOCKNAMELEN)) {
3024                         NFSUNLOCKCLSTATE();
3025                         printf("DUP OWNER\n");
3026                         nfscl_dumpstate(VFSTONFS(vp->v_mount), 1, 1, 0, 0);
3027                         return;
3028                 }
3029             }
3030         }
3031
3032         /*
3033          * Now, search for duplicate stateids.
3034          * These shouldn't happen, either.
3035          */
3036         LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3037             LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
3038                 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3039                     LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3040                         if (op != op2 &&
3041                             (op->nfso_stateid.other[0] != 0 ||
3042                              op->nfso_stateid.other[1] != 0 ||
3043                              op->nfso_stateid.other[2] != 0) &&
3044                             op->nfso_stateid.other[0] == op2->nfso_stateid.other[0] &&
3045                             op->nfso_stateid.other[1] == op2->nfso_stateid.other[1] &&
3046                             op->nfso_stateid.other[2] == op2->nfso_stateid.other[2]) {
3047                             NFSUNLOCKCLSTATE();
3048                             printf("DUP STATEID\n");
3049                             nfscl_dumpstate(VFSTONFS(vp->v_mount), 1, 1, 0, 0);
3050                             return;
3051                         }
3052                     }
3053                 }
3054             }
3055         }
3056
3057         /*
3058          * Now search for duplicate opens.
3059          * Duplicate opens for the same owner
3060          * should never occur. Other duplicates are
3061          * possible and are checked for if "dupopens"
3062          * is true.
3063          */
3064         LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3065             LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
3066                 if (nfhp->nfh_len == op2->nfso_fhlen &&
3067                     !NFSBCMP(nfhp->nfh_fh, op2->nfso_fh, nfhp->nfh_len)) {
3068                     LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3069                         LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3070                             if (op != op2 && nfhp->nfh_len == op->nfso_fhlen &&
3071                                 !NFSBCMP(nfhp->nfh_fh, op->nfso_fh, nfhp->nfh_len) &&
3072                                 (!NFSBCMP(op->nfso_own->nfsow_owner,
3073                                  op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN) ||
3074                                  dupopens)) {
3075                                 if (!NFSBCMP(op->nfso_own->nfsow_owner,
3076                                     op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
3077                                     NFSUNLOCKCLSTATE();
3078                                     printf("BADDUP OPEN\n");
3079                                 } else {
3080                                     NFSUNLOCKCLSTATE();
3081                                     printf("DUP OPEN\n");
3082                                 }
3083                                 nfscl_dumpstate(VFSTONFS(vp->v_mount), 1, 1, 0,
3084                                     0);
3085                                 return;
3086                             }
3087                         }
3088                     }
3089                 }
3090             }
3091         }
3092         NFSUNLOCKCLSTATE();
3093 }
3094
3095 /*
3096  * During close, find an open that needs to be dereferenced and
3097  * dereference it. If there are no more opens for this file,
3098  * log a message to that effect.
3099  * Opens aren't actually Close'd until VOP_INACTIVE() is performed
3100  * on the file's vnode.
3101  * This is the safe way, since it is difficult to identify
3102  * which open the close is for and I/O can be performed after the
3103  * close(2) system call when a file is mmap'd.
3104  * If it returns 0 for success, there will be a referenced
3105  * clp returned via clpp.
3106  */
3107 int
3108 nfscl_getclose(vnode_t vp, struct nfsclclient **clpp)
3109 {
3110         struct nfsclclient *clp;
3111         struct nfsclowner *owp;
3112         struct nfsclopen *op;
3113         struct nfscldeleg *dp;
3114         struct nfsfh *nfhp;
3115         int error, notdecr;
3116
3117         error = nfscl_getcl(vp->v_mount, NULL, NULL, 1, &clp);
3118         if (error)
3119                 return (error);
3120         *clpp = clp;
3121
3122         nfhp = VTONFS(vp)->n_fhp;
3123         notdecr = 1;
3124         NFSLOCKCLSTATE();
3125         /*
3126          * First, look for one under a delegation that was locally issued
3127          * and just decrement the opencnt for it. Since all my Opens against
3128          * the server are DENY_NONE, I don't see a problem with hanging
3129          * onto them. (It is much easier to use one of the extant Opens
3130          * that I already have on the server when a Delegation is recalled
3131          * than to do fresh Opens.) Someday, I might need to rethink this, but.
3132          */
3133         dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
3134         if (dp != NULL) {
3135                 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
3136                         op = LIST_FIRST(&owp->nfsow_open);
3137                         if (op != NULL) {
3138                                 /*
3139                                  * Since a delegation is for a file, there
3140                                  * should never be more than one open for
3141                                  * each openowner.
3142                                  */
3143                                 if (LIST_NEXT(op, nfso_list) != NULL)
3144                                         panic("nfscdeleg opens");
3145                                 if (notdecr && op->nfso_opencnt > 0) {
3146                                         notdecr = 0;
3147                                         op->nfso_opencnt--;
3148                                         break;
3149                                 }
3150                         }
3151                 }
3152         }
3153
3154         /* Now process the opens against the server. */
3155         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3156                 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3157                         if (op->nfso_fhlen == nfhp->nfh_len &&
3158                             !NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
3159                             nfhp->nfh_len)) {
3160                                 /* Found an open, decrement cnt if possible */
3161                                 if (notdecr && op->nfso_opencnt > 0) {
3162                                         notdecr = 0;
3163                                         op->nfso_opencnt--;
3164                                 }
3165                                 /*
3166                                  * There are more opens, so just return.
3167                                  */
3168                                 if (op->nfso_opencnt > 0) {
3169                                         NFSUNLOCKCLSTATE();
3170                                         return (0);
3171                                 }
3172                         }
3173                 }
3174         }
3175         NFSUNLOCKCLSTATE();
3176         if (notdecr)
3177                 printf("nfscl: never fnd open\n");
3178         return (0);
3179 }
3180
3181 int
3182 nfscl_doclose(vnode_t vp, struct nfsclclient **clpp, NFSPROC_T *p)
3183 {
3184         struct nfsclclient *clp;
3185         struct nfsclowner *owp, *nowp;
3186         struct nfsclopen *op;
3187         struct nfscldeleg *dp;
3188         struct nfsfh *nfhp;
3189         struct nfsclrecalllayout *recallp;
3190         int error;
3191
3192         error = nfscl_getcl(vp->v_mount, NULL, NULL, 1, &clp);
3193         if (error)
3194                 return (error);
3195         *clpp = clp;
3196
3197         nfhp = VTONFS(vp)->n_fhp;
3198         recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_WAITOK);
3199         NFSLOCKCLSTATE();
3200         /*
3201          * First get rid of the local Open structures, which should be no
3202          * longer in use.
3203          */
3204         dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
3205         if (dp != NULL) {
3206                 LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) {
3207                         op = LIST_FIRST(&owp->nfsow_open);
3208                         if (op != NULL) {
3209                                 KASSERT((op->nfso_opencnt == 0),
3210                                     ("nfscl: bad open cnt on deleg"));
3211                                 nfscl_freeopen(op, 1);
3212                         }
3213                         nfscl_freeopenowner(owp, 1);
3214                 }
3215         }
3216
3217         /* Return any layouts marked return on close. */
3218         nfscl_retoncloselayout(vp, clp, nfhp->nfh_fh, nfhp->nfh_len, &recallp);
3219
3220         /* Now process the opens against the server. */
3221 lookformore:
3222         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3223                 op = LIST_FIRST(&owp->nfsow_open);
3224                 while (op != NULL) {
3225                         if (op->nfso_fhlen == nfhp->nfh_len &&
3226                             !NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
3227                             nfhp->nfh_len)) {
3228                                 /* Found an open, close it. */
3229 #ifdef DIAGNOSTIC
3230                                 KASSERT((op->nfso_opencnt == 0),
3231                                     ("nfscl: bad open cnt on server (%d)",
3232                                      op->nfso_opencnt));
3233 #endif
3234                                 NFSUNLOCKCLSTATE();
3235                                 nfsrpc_doclose(VFSTONFS(vp->v_mount), op, p);
3236                                 NFSLOCKCLSTATE();
3237                                 goto lookformore;
3238                         }
3239                         op = LIST_NEXT(op, nfso_list);
3240                 }
3241         }
3242         NFSUNLOCKCLSTATE();
3243         /*
3244          * recallp has been set NULL by nfscl_retoncloselayout() if it was
3245          * used by the function, but calling free() with a NULL pointer is ok.
3246          */
3247         free(recallp, M_NFSLAYRECALL);
3248         return (0);
3249 }
3250
3251 /*
3252  * Return all delegations on this client.
3253  * (Must be called with client sleep lock.)
3254  */
3255 static void
3256 nfscl_delegreturnall(struct nfsclclient *clp, NFSPROC_T *p)
3257 {
3258         struct nfscldeleg *dp, *ndp;
3259         struct ucred *cred;
3260
3261         cred = newnfs_getcred();
3262         TAILQ_FOREACH_SAFE(dp, &clp->nfsc_deleg, nfsdl_list, ndp) {
3263                 nfscl_cleandeleg(dp);
3264                 (void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
3265                 nfscl_freedeleg(&clp->nfsc_deleg, dp);
3266         }
3267         NFSFREECRED(cred);
3268 }
3269
3270 /*
3271  * Do a callback RPC.
3272  */
3273 void
3274 nfscl_docb(struct nfsrv_descript *nd, NFSPROC_T *p)
3275 {
3276         int clist, gotseq_ok, i, j, k, op, rcalls;
3277         u_int32_t *tl;
3278         struct nfsclclient *clp;
3279         struct nfscldeleg *dp = NULL;
3280         int numops, taglen = -1, error = 0, trunc __unused;
3281         u_int32_t minorvers = 0, retops = 0, *retopsp = NULL, *repp, cbident;
3282         u_char tag[NFSV4_SMALLSTR + 1], *tagstr;
3283         vnode_t vp = NULL;
3284         struct nfsnode *np;
3285         struct vattr va;
3286         struct nfsfh *nfhp;
3287         mount_t mp;
3288         nfsattrbit_t attrbits, rattrbits;
3289         nfsv4stateid_t stateid;
3290         uint32_t seqid, slotid = 0, highslot, cachethis __unused;
3291         uint8_t sessionid[NFSX_V4SESSIONID];
3292         struct mbuf *rep;
3293         struct nfscllayout *lyp;
3294         uint64_t filesid[2], len, off;
3295         int changed, gotone, laytype, recalltype;
3296         uint32_t iomode;
3297         struct nfsclrecalllayout *recallp = NULL;
3298         struct nfsclsession *tsep;
3299
3300         gotseq_ok = 0;
3301         nfsrvd_rephead(nd);
3302         NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3303         taglen = fxdr_unsigned(int, *tl);
3304         if (taglen < 0) {
3305                 error = EBADRPC;
3306                 goto nfsmout;
3307         }
3308         if (taglen <= NFSV4_SMALLSTR)
3309                 tagstr = tag;
3310         else
3311                 tagstr = malloc(taglen + 1, M_TEMP, M_WAITOK);
3312         error = nfsrv_mtostr(nd, tagstr, taglen);
3313         if (error) {
3314                 if (taglen > NFSV4_SMALLSTR)
3315                         free(tagstr, M_TEMP);
3316                 taglen = -1;
3317                 goto nfsmout;
3318         }
3319         (void) nfsm_strtom(nd, tag, taglen);
3320         if (taglen > NFSV4_SMALLSTR) {
3321                 free(tagstr, M_TEMP);
3322         }
3323         NFSM_BUILD(retopsp, u_int32_t *, NFSX_UNSIGNED);
3324         NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3325         minorvers = fxdr_unsigned(u_int32_t, *tl++);
3326         if (minorvers != NFSV4_MINORVERSION &&
3327             minorvers != NFSV41_MINORVERSION &&
3328             minorvers != NFSV42_MINORVERSION)
3329                 nd->nd_repstat = NFSERR_MINORVERMISMATCH;
3330         cbident = fxdr_unsigned(u_int32_t, *tl++);
3331         if (nd->nd_repstat)
3332                 numops = 0;
3333         else
3334                 numops = fxdr_unsigned(int, *tl);
3335         /*
3336          * Loop around doing the sub ops.
3337          */
3338         for (i = 0; i < numops; i++) {
3339                 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3340                 NFSM_BUILD(repp, u_int32_t *, 2 * NFSX_UNSIGNED);
3341                 *repp++ = *tl;
3342                 op = fxdr_unsigned(int, *tl);
3343                 if (op < NFSV4OP_CBGETATTR ||
3344                    (op > NFSV4OP_CBRECALL && minorvers == NFSV4_MINORVERSION) ||
3345                    (op > NFSV4OP_CBNOTIFYDEVID &&
3346                     minorvers == NFSV41_MINORVERSION) ||
3347                    (op > NFSV4OP_CBOFFLOAD &&
3348                     minorvers == NFSV42_MINORVERSION)) {
3349                     nd->nd_repstat = NFSERR_OPILLEGAL;
3350                     *repp = nfscl_errmap(nd, minorvers);
3351                     retops++;
3352                     break;
3353                 }
3354                 nd->nd_procnum = op;
3355                 if (op < NFSV42_CBNOPS)
3356                         nfsstatsv1.cbrpccnt[nd->nd_procnum]++;
3357                 switch (op) {
3358                 case NFSV4OP_CBGETATTR:
3359                         NFSCL_DEBUG(4, "cbgetattr\n");
3360                         mp = NULL;
3361                         vp = NULL;
3362                         error = nfsm_getfh(nd, &nfhp);
3363                         if (!error)
3364                                 error = nfsrv_getattrbits(nd, &attrbits,
3365                                     NULL, NULL);
3366                         if (error == 0 && i == 0 &&
3367                             minorvers != NFSV4_MINORVERSION)
3368                                 error = NFSERR_OPNOTINSESS;
3369                         if (!error) {
3370                                 mp = nfscl_getmnt(minorvers, sessionid, cbident,
3371                                     &clp);
3372                                 if (mp == NULL)
3373                                         error = NFSERR_SERVERFAULT;
3374                         }
3375                         if (!error) {
3376                                 error = nfscl_ngetreopen(mp, nfhp->nfh_fh,
3377                                     nfhp->nfh_len, p, &np);
3378                                 if (!error)
3379                                         vp = NFSTOV(np);
3380                         }
3381                         if (!error) {
3382                                 NFSZERO_ATTRBIT(&rattrbits);
3383                                 NFSLOCKCLSTATE();
3384                                 dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
3385                                     nfhp->nfh_len);
3386                                 if (dp != NULL) {
3387                                         if (NFSISSET_ATTRBIT(&attrbits,
3388                                             NFSATTRBIT_SIZE)) {
3389                                                 if (vp != NULL)
3390                                                         va.va_size = np->n_size;
3391                                                 else
3392                                                         va.va_size =
3393                                                             dp->nfsdl_size;
3394                                                 NFSSETBIT_ATTRBIT(&rattrbits,
3395                                                     NFSATTRBIT_SIZE);
3396                                         }
3397                                         if (NFSISSET_ATTRBIT(&attrbits,
3398                                             NFSATTRBIT_CHANGE)) {
3399                                                 va.va_filerev =
3400                                                     dp->nfsdl_change;
3401                                                 if (vp == NULL ||
3402                                                     (np->n_flag & NDELEGMOD))
3403                                                         va.va_filerev++;
3404                                                 NFSSETBIT_ATTRBIT(&rattrbits,
3405                                                     NFSATTRBIT_CHANGE);
3406                                         }
3407                                 } else
3408                                         error = NFSERR_SERVERFAULT;
3409                                 NFSUNLOCKCLSTATE();
3410                         }
3411                         if (vp != NULL)
3412                                 vrele(vp);
3413                         if (mp != NULL)
3414                                 vfs_unbusy(mp);
3415                         if (nfhp != NULL)
3416                                 free(nfhp, M_NFSFH);
3417                         if (!error)
3418                                 (void) nfsv4_fillattr(nd, NULL, NULL, NULL, &va,
3419                                     NULL, 0, &rattrbits, NULL, p, 0, 0, 0, 0,
3420                                     (uint64_t)0, NULL);
3421                         break;
3422                 case NFSV4OP_CBRECALL:
3423                         NFSCL_DEBUG(4, "cbrecall\n");
3424                         NFSM_DISSECT(tl, u_int32_t *, NFSX_STATEID +
3425                             NFSX_UNSIGNED);
3426                         stateid.seqid = *tl++;
3427                         NFSBCOPY((caddr_t)tl, (caddr_t)stateid.other,
3428                             NFSX_STATEIDOTHER);
3429                         tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
3430                         trunc = fxdr_unsigned(int, *tl);
3431                         error = nfsm_getfh(nd, &nfhp);
3432                         if (error == 0 && i == 0 &&
3433                             minorvers != NFSV4_MINORVERSION)
3434                                 error = NFSERR_OPNOTINSESS;
3435                         if (!error) {
3436                                 NFSLOCKCLSTATE();
3437                                 if (minorvers == NFSV4_MINORVERSION)
3438                                         clp = nfscl_getclnt(cbident);
3439                                 else
3440                                         clp = nfscl_getclntsess(sessionid);
3441                                 if (clp != NULL) {
3442                                         dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
3443                                             nfhp->nfh_len);
3444                                         if (dp != NULL && (dp->nfsdl_flags &
3445                                             NFSCLDL_DELEGRET) == 0) {
3446                                                 dp->nfsdl_flags |=
3447                                                     NFSCLDL_RECALL;
3448                                                 wakeup((caddr_t)clp);
3449                                         }
3450                                 } else {
3451                                         error = NFSERR_SERVERFAULT;
3452                                 }
3453                                 NFSUNLOCKCLSTATE();
3454                         }
3455                         if (nfhp != NULL)
3456                                 free(nfhp, M_NFSFH);
3457                         break;
3458                 case NFSV4OP_CBLAYOUTRECALL:
3459                         NFSCL_DEBUG(4, "cblayrec\n");
3460                         nfhp = NULL;
3461                         NFSM_DISSECT(tl, uint32_t *, 4 * NFSX_UNSIGNED);
3462                         laytype = fxdr_unsigned(int, *tl++);
3463                         iomode = fxdr_unsigned(uint32_t, *tl++);
3464                         if (newnfs_true == *tl++)
3465                                 changed = 1;
3466                         else
3467                                 changed = 0;
3468                         recalltype = fxdr_unsigned(int, *tl);
3469                         NFSCL_DEBUG(4, "layt=%d iom=%d ch=%d rectyp=%d\n",
3470                             laytype, iomode, changed, recalltype);
3471                         recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL,
3472                             M_WAITOK);
3473                         if (laytype != NFSLAYOUT_NFSV4_1_FILES &&
3474                             laytype != NFSLAYOUT_FLEXFILE)
3475                                 error = NFSERR_NOMATCHLAYOUT;
3476                         else if (recalltype == NFSLAYOUTRETURN_FILE) {
3477                                 error = nfsm_getfh(nd, &nfhp);
3478                                 NFSCL_DEBUG(4, "retfile getfh=%d\n", error);
3479                                 if (error != 0)
3480                                         goto nfsmout;
3481                                 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_HYPER +
3482                                     NFSX_STATEID);
3483                                 off = fxdr_hyper(tl); tl += 2;
3484                                 len = fxdr_hyper(tl); tl += 2;
3485                                 stateid.seqid = fxdr_unsigned(uint32_t, *tl++);
3486                                 NFSBCOPY(tl, stateid.other, NFSX_STATEIDOTHER);
3487                                 if (minorvers == NFSV4_MINORVERSION)
3488                                         error = NFSERR_NOTSUPP;
3489                                 else if (i == 0)
3490                                         error = NFSERR_OPNOTINSESS;
3491                                 NFSCL_DEBUG(4, "off=%ju len=%ju sq=%u err=%d\n",
3492                                     (uintmax_t)off, (uintmax_t)len,
3493                                     stateid.seqid, error);
3494                                 if (error == 0) {
3495                                         NFSLOCKCLSTATE();
3496                                         clp = nfscl_getclntsess(sessionid);
3497                                         NFSCL_DEBUG(4, "cbly clp=%p\n", clp);
3498                                         if (clp != NULL) {
3499                                                 lyp = nfscl_findlayout(clp,
3500                                                     nfhp->nfh_fh,
3501                                                     nfhp->nfh_len);
3502                                                 NFSCL_DEBUG(4, "cblyp=%p\n",
3503                                                     lyp);
3504                                                 if (lyp != NULL &&
3505                                                     (lyp->nfsly_flags &
3506                                                      (NFSLY_FILES |
3507                                                       NFSLY_FLEXFILE)) != 0 &&
3508                                                     !NFSBCMP(stateid.other,
3509                                                     lyp->nfsly_stateid.other,
3510                                                     NFSX_STATEIDOTHER)) {
3511                                                         error =
3512                                                             nfscl_layoutrecall(
3513                                                             recalltype,
3514                                                             lyp, iomode, off,
3515                                                             len, stateid.seqid,
3516                                                             0, 0, NULL,
3517                                                             recallp);
3518                                                         if (error == 0 &&
3519                                                             stateid.seqid >
3520                                                             lyp->nfsly_stateid.seqid)
3521                                                                 lyp->nfsly_stateid.seqid =
3522                                                                     stateid.seqid;
3523                                                         recallp = NULL;
3524                                                         wakeup(clp);
3525                                                         NFSCL_DEBUG(4,
3526                                                             "aft layrcal=%d "
3527                                                             "layseqid=%d\n",
3528                                                             error,
3529                                                             lyp->nfsly_stateid.seqid);
3530                                                 } else
3531                                                         error =
3532                                                           NFSERR_NOMATCHLAYOUT;
3533                                         } else
3534                                                 error = NFSERR_NOMATCHLAYOUT;
3535                                         NFSUNLOCKCLSTATE();
3536                                 }
3537                                 free(nfhp, M_NFSFH);
3538                         } else if (recalltype == NFSLAYOUTRETURN_FSID) {
3539                                 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_HYPER);
3540                                 filesid[0] = fxdr_hyper(tl); tl += 2;
3541                                 filesid[1] = fxdr_hyper(tl); tl += 2;
3542                                 gotone = 0;
3543                                 NFSLOCKCLSTATE();
3544                                 clp = nfscl_getclntsess(sessionid);
3545                                 if (clp != NULL) {
3546                                         TAILQ_FOREACH(lyp, &clp->nfsc_layout,
3547                                             nfsly_list) {
3548                                                 if (lyp->nfsly_filesid[0] ==
3549                                                     filesid[0] &&
3550                                                     lyp->nfsly_filesid[1] ==
3551                                                     filesid[1]) {
3552                                                         error =
3553                                                             nfscl_layoutrecall(
3554                                                             recalltype,
3555                                                             lyp, iomode, 0,
3556                                                             UINT64_MAX,
3557                                                             lyp->nfsly_stateid.seqid,
3558                                                             0, 0, NULL,
3559                                                             recallp);
3560                                                         recallp = NULL;
3561                                                         gotone = 1;
3562                                                 }
3563                                         }
3564                                         if (gotone != 0)
3565                                                 wakeup(clp);
3566                                         else
3567                                                 error = NFSERR_NOMATCHLAYOUT;
3568                                 } else
3569                                         error = NFSERR_NOMATCHLAYOUT;
3570                                 NFSUNLOCKCLSTATE();
3571                         } else if (recalltype == NFSLAYOUTRETURN_ALL) {
3572                                 gotone = 0;
3573                                 NFSLOCKCLSTATE();
3574                                 clp = nfscl_getclntsess(sessionid);
3575                                 if (clp != NULL) {
3576                                         TAILQ_FOREACH(lyp, &clp->nfsc_layout,
3577                                             nfsly_list) {
3578                                                 error = nfscl_layoutrecall(
3579                                                     recalltype, lyp, iomode, 0,
3580                                                     UINT64_MAX,
3581                                                     lyp->nfsly_stateid.seqid,
3582                                                     0, 0, NULL, recallp);
3583                                                 recallp = NULL;
3584                                                 gotone = 1;
3585                                         }
3586                                         if (gotone != 0)
3587                                                 wakeup(clp);
3588                                         else
3589                                                 error = NFSERR_NOMATCHLAYOUT;
3590                                 } else
3591                                         error = NFSERR_NOMATCHLAYOUT;
3592                                 NFSUNLOCKCLSTATE();
3593                         } else
3594                                 error = NFSERR_NOMATCHLAYOUT;
3595                         if (recallp != NULL) {
3596                                 free(recallp, M_NFSLAYRECALL);
3597                                 recallp = NULL;
3598                         }
3599                         break;
3600                 case NFSV4OP_CBSEQUENCE:
3601                         NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID +
3602                             5 * NFSX_UNSIGNED);
3603                         bcopy(tl, sessionid, NFSX_V4SESSIONID);
3604                         tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3605                         seqid = fxdr_unsigned(uint32_t, *tl++);
3606                         slotid = fxdr_unsigned(uint32_t, *tl++);
3607                         highslot = fxdr_unsigned(uint32_t, *tl++);
3608                         cachethis = *tl++;
3609                         /* Throw away the referring call stuff. */
3610                         clist = fxdr_unsigned(int, *tl);
3611                         for (j = 0; j < clist; j++) {
3612                                 NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID +
3613                                     NFSX_UNSIGNED);
3614                                 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3615                                 rcalls = fxdr_unsigned(int, *tl);
3616                                 for (k = 0; k < rcalls; k++) {
3617                                         NFSM_DISSECT(tl, uint32_t *,
3618                                             2 * NFSX_UNSIGNED);
3619                                 }
3620                         }
3621                         NFSLOCKCLSTATE();
3622                         if (i == 0) {
3623                                 clp = nfscl_getclntsess(sessionid);
3624                                 if (clp == NULL)
3625                                         error = NFSERR_SERVERFAULT;
3626                         } else
3627                                 error = NFSERR_SEQUENCEPOS;
3628                         if (error == 0) {
3629                                 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3630                                 error = nfsv4_seqsession(seqid, slotid,
3631                                     highslot, tsep->nfsess_cbslots, &rep,
3632                                     tsep->nfsess_backslots);
3633                         }
3634                         NFSUNLOCKCLSTATE();
3635                         if (error == 0 || error == NFSERR_REPLYFROMCACHE) {
3636                                 gotseq_ok = 1;
3637                                 if (rep != NULL) {
3638                                         /*
3639                                          * Handle a reply for a retried
3640                                          * callback.  The reply will be
3641                                          * re-inserted in the session cache
3642                                          * by the nfsv4_seqsess_cacherep() call
3643                                          * after out:
3644                                          */
3645                                         KASSERT(error == NFSERR_REPLYFROMCACHE,
3646                                             ("cbsequence: non-NULL rep"));
3647                                         NFSCL_DEBUG(4, "Got cbretry\n");
3648                                         m_freem(nd->nd_mreq);
3649                                         nd->nd_mreq = rep;
3650                                         rep = NULL;
3651                                         goto out;
3652                                 }
3653                                 NFSM_BUILD(tl, uint32_t *,
3654                                     NFSX_V4SESSIONID + 4 * NFSX_UNSIGNED);
3655                                 bcopy(sessionid, tl, NFSX_V4SESSIONID);
3656                                 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3657                                 *tl++ = txdr_unsigned(seqid);
3658                                 *tl++ = txdr_unsigned(slotid);
3659                                 *tl++ = txdr_unsigned(NFSV4_CBSLOTS - 1);
3660                                 *tl = txdr_unsigned(NFSV4_CBSLOTS - 1);
3661                         }
3662                         break;
3663                 default:
3664                         if (i == 0 && minorvers != NFSV4_MINORVERSION)
3665                                 error = NFSERR_OPNOTINSESS;
3666                         else {
3667                                 NFSCL_DEBUG(1, "unsupp callback %d\n", op);
3668                                 error = NFSERR_NOTSUPP;
3669                         }
3670                         break;
3671                 }
3672                 if (error) {
3673                         if (error == EBADRPC || error == NFSERR_BADXDR) {
3674                                 nd->nd_repstat = NFSERR_BADXDR;
3675                         } else {
3676                                 nd->nd_repstat = error;
3677                         }
3678                         error = 0;
3679                 }
3680                 retops++;
3681                 if (nd->nd_repstat) {
3682                         *repp = nfscl_errmap(nd, minorvers);
3683                         break;
3684                 } else
3685                         *repp = 0;      /* NFS4_OK */
3686         }
3687 nfsmout:
3688         if (recallp != NULL)
3689                 free(recallp, M_NFSLAYRECALL);
3690         if (error) {
3691                 if (error == EBADRPC || error == NFSERR_BADXDR)
3692                         nd->nd_repstat = NFSERR_BADXDR;
3693                 else
3694                         printf("nfsv4 comperr1=%d\n", error);
3695         }
3696         if (taglen == -1) {
3697                 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
3698                 *tl++ = 0;
3699                 *tl = 0;
3700         } else {
3701                 *retopsp = txdr_unsigned(retops);
3702         }
3703         *nd->nd_errp = nfscl_errmap(nd, minorvers);
3704 out:
3705         if (gotseq_ok != 0) {
3706                 rep = m_copym(nd->nd_mreq, 0, M_COPYALL, M_WAITOK);
3707                 NFSLOCKCLSTATE();
3708                 clp = nfscl_getclntsess(sessionid);
3709                 if (clp != NULL) {
3710                         tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3711                         nfsv4_seqsess_cacherep(slotid, tsep->nfsess_cbslots,
3712                             NFSERR_OK, &rep);
3713                         NFSUNLOCKCLSTATE();
3714                 } else {
3715                         NFSUNLOCKCLSTATE();
3716                         m_freem(rep);
3717                 }
3718         }
3719 }
3720
3721 /*
3722  * Generate the next cbident value. Basically just increment a static value
3723  * and then check that it isn't already in the list, if it has wrapped around.
3724  */
3725 static u_int32_t
3726 nfscl_nextcbident(void)
3727 {
3728         struct nfsclclient *clp;
3729         int matched;
3730         static u_int32_t nextcbident = 0;
3731         static int haswrapped = 0;
3732
3733         nextcbident++;
3734         if (nextcbident == 0)
3735                 haswrapped = 1;
3736         if (haswrapped) {
3737                 /*
3738                  * Search the clientid list for one already using this cbident.
3739                  */
3740                 do {
3741                         matched = 0;
3742                         NFSLOCKCLSTATE();
3743                         LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3744                                 if (clp->nfsc_cbident == nextcbident) {
3745                                         matched = 1;
3746                                         break;
3747                                 }
3748                         }
3749                         NFSUNLOCKCLSTATE();
3750                         if (matched == 1)
3751                                 nextcbident++;
3752                 } while (matched);
3753         }
3754         return (nextcbident);
3755 }
3756
3757 /*
3758  * Get the mount point related to a given cbident or session and busy it.
3759  */
3760 static mount_t
3761 nfscl_getmnt(int minorvers, uint8_t *sessionid, u_int32_t cbident,
3762     struct nfsclclient **clpp)
3763 {
3764         struct nfsclclient *clp;
3765         mount_t mp;
3766         int error;
3767         struct nfsclsession *tsep;
3768
3769         *clpp = NULL;
3770         NFSLOCKCLSTATE();
3771         LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3772                 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3773                 if (minorvers == NFSV4_MINORVERSION) {
3774                         if (clp->nfsc_cbident == cbident)
3775                                 break;
3776                 } else if (!NFSBCMP(tsep->nfsess_sessionid, sessionid,
3777                     NFSX_V4SESSIONID))
3778                         break;
3779         }
3780         if (clp == NULL) {
3781                 NFSUNLOCKCLSTATE();
3782                 return (NULL);
3783         }
3784         mp = clp->nfsc_nmp->nm_mountp;
3785         vfs_ref(mp);
3786         NFSUNLOCKCLSTATE();
3787         error = vfs_busy(mp, 0);
3788         vfs_rel(mp);
3789         if (error != 0)
3790                 return (NULL);
3791         *clpp = clp;
3792         return (mp);
3793 }
3794
3795 /*
3796  * Get the clientid pointer related to a given cbident.
3797  */
3798 static struct nfsclclient *
3799 nfscl_getclnt(u_int32_t cbident)
3800 {
3801         struct nfsclclient *clp;
3802
3803         LIST_FOREACH(clp, &nfsclhead, nfsc_list)
3804                 if (clp->nfsc_cbident == cbident)
3805                         break;
3806         return (clp);
3807 }
3808
3809 /*
3810  * Get the clientid pointer related to a given sessionid.
3811  */
3812 static struct nfsclclient *
3813 nfscl_getclntsess(uint8_t *sessionid)
3814 {
3815         struct nfsclclient *clp;
3816         struct nfsclsession *tsep;
3817
3818         LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3819                 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3820                 if (!NFSBCMP(tsep->nfsess_sessionid, sessionid,
3821                     NFSX_V4SESSIONID))
3822                         break;
3823         }
3824         return (clp);
3825 }
3826
3827 /*
3828  * Search for a lock conflict locally on the client. A conflict occurs if
3829  * - not same owner and overlapping byte range and at least one of them is
3830  *   a write lock or this is an unlock.
3831  */
3832 static int
3833 nfscl_localconflict(struct nfsclclient *clp, u_int8_t *fhp, int fhlen,
3834     struct nfscllock *nlop, u_int8_t *own, struct nfscldeleg *dp,
3835     struct nfscllock **lopp)
3836 {
3837         struct nfsclowner *owp;
3838         struct nfsclopen *op;
3839         int ret;
3840
3841         if (dp != NULL) {
3842                 ret = nfscl_checkconflict(&dp->nfsdl_lock, nlop, own, lopp);
3843                 if (ret)
3844                         return (ret);
3845         }
3846         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3847                 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3848                         if (op->nfso_fhlen == fhlen &&
3849                             !NFSBCMP(op->nfso_fh, fhp, fhlen)) {
3850                                 ret = nfscl_checkconflict(&op->nfso_lock, nlop,
3851                                     own, lopp);
3852                                 if (ret)
3853                                         return (ret);
3854                         }
3855                 }
3856         }
3857         return (0);
3858 }
3859
3860 static int
3861 nfscl_checkconflict(struct nfscllockownerhead *lhp, struct nfscllock *nlop,
3862     u_int8_t *own, struct nfscllock **lopp)
3863 {
3864         struct nfscllockowner *lp;
3865         struct nfscllock *lop;
3866
3867         LIST_FOREACH(lp, lhp, nfsl_list) {
3868                 if (NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
3869                         LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
3870                                 if (lop->nfslo_first >= nlop->nfslo_end)
3871                                         break;
3872                                 if (lop->nfslo_end <= nlop->nfslo_first)
3873                                         continue;
3874                                 if (lop->nfslo_type == F_WRLCK ||
3875                                     nlop->nfslo_type == F_WRLCK ||
3876                                     nlop->nfslo_type == F_UNLCK) {
3877                                         if (lopp != NULL)
3878                                                 *lopp = lop;
3879                                         return (NFSERR_DENIED);
3880                                 }
3881                         }
3882                 }
3883         }
3884         return (0);
3885 }
3886
3887 /*
3888  * Check for a local conflicting lock.
3889  */
3890 int
3891 nfscl_lockt(vnode_t vp, struct nfsclclient *clp, u_int64_t off,
3892     u_int64_t len, struct flock *fl, NFSPROC_T *p, void *id, int flags)
3893 {
3894         struct nfscllock *lop, nlck;
3895         struct nfscldeleg *dp;
3896         struct nfsnode *np;
3897         u_int8_t own[NFSV4CL_LOCKNAMELEN];
3898         int error;
3899
3900         nlck.nfslo_type = fl->l_type;
3901         nlck.nfslo_first = off;
3902         if (len == NFS64BITSSET) {
3903                 nlck.nfslo_end = NFS64BITSSET;
3904         } else {
3905                 nlck.nfslo_end = off + len;
3906                 if (nlck.nfslo_end <= nlck.nfslo_first)
3907                         return (NFSERR_INVAL);
3908         }
3909         np = VTONFS(vp);
3910         nfscl_filllockowner(id, own, flags);
3911         NFSLOCKCLSTATE();
3912         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
3913         error = nfscl_localconflict(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len,
3914             &nlck, own, dp, &lop);
3915         if (error != 0) {
3916                 fl->l_whence = SEEK_SET;
3917                 fl->l_start = lop->nfslo_first;
3918                 if (lop->nfslo_end == NFS64BITSSET)
3919                         fl->l_len = 0;
3920                 else
3921                         fl->l_len = lop->nfslo_end - lop->nfslo_first;
3922                 fl->l_pid = (pid_t)0;
3923                 fl->l_type = lop->nfslo_type;
3924                 error = -1;                     /* no RPC required */
3925         } else if (dp != NULL && ((dp->nfsdl_flags & NFSCLDL_WRITE) ||
3926             fl->l_type == F_RDLCK)) {
3927                 /*
3928                  * The delegation ensures that there isn't a conflicting
3929                  * lock on the server, so return -1 to indicate an RPC
3930                  * isn't required.
3931                  */
3932                 fl->l_type = F_UNLCK;
3933                 error = -1;
3934         }
3935         NFSUNLOCKCLSTATE();
3936         return (error);
3937 }
3938
3939 /*
3940  * Handle Recall of a delegation.
3941  * The clp must be exclusive locked when this is called.
3942  */
3943 static int
3944 nfscl_recalldeleg(struct nfsclclient *clp, struct nfsmount *nmp,
3945     struct nfscldeleg *dp, vnode_t vp, struct ucred *cred, NFSPROC_T *p,
3946     int called_from_renewthread)
3947 {
3948         struct nfsclowner *owp, *lowp, *nowp;
3949         struct nfsclopen *op, *lop;
3950         struct nfscllockowner *lp;
3951         struct nfscllock *lckp;
3952         struct nfsnode *np;
3953         int error = 0, ret, gotvp = 0;
3954
3955         if (vp == NULL) {
3956                 /*
3957                  * First, get a vnode for the file. This is needed to do RPCs.
3958                  */
3959                 ret = nfscl_ngetreopen(nmp->nm_mountp, dp->nfsdl_fh,
3960                     dp->nfsdl_fhlen, p, &np);
3961                 if (ret) {
3962                         /*
3963                          * File isn't open, so nothing to move over to the
3964                          * server.
3965                          */
3966                         return (0);
3967                 }
3968                 vp = NFSTOV(np);
3969                 gotvp = 1;
3970         } else {
3971                 np = VTONFS(vp);
3972         }
3973         dp->nfsdl_flags &= ~NFSCLDL_MODTIMESET;
3974
3975         /*
3976          * Ok, if it's a write delegation, flush data to the server, so
3977          * that close/open consistency is retained.
3978          */
3979         ret = 0;
3980         NFSLOCKNODE(np);
3981         if ((dp->nfsdl_flags & NFSCLDL_WRITE) && (np->n_flag & NMODIFIED)) {
3982                 np->n_flag |= NDELEGRECALL;
3983                 NFSUNLOCKNODE(np);
3984                 ret = ncl_flush(vp, MNT_WAIT, p, 1, called_from_renewthread);
3985                 NFSLOCKNODE(np);
3986                 np->n_flag &= ~NDELEGRECALL;
3987         }
3988         NFSINVALATTRCACHE(np);
3989         NFSUNLOCKNODE(np);
3990         if (ret == EIO && called_from_renewthread != 0) {
3991                 /*
3992                  * If the flush failed with EIO for the renew thread,
3993                  * return now, so that the dirty buffer will be flushed
3994                  * later.
3995                  */
3996                 if (gotvp != 0)
3997                         vrele(vp);
3998                 return (ret);
3999         }
4000
4001         /*
4002          * Now, for each openowner with opens issued locally, move them
4003          * over to state against the server.
4004          */
4005         LIST_FOREACH(lowp, &dp->nfsdl_owner, nfsow_list) {
4006                 lop = LIST_FIRST(&lowp->nfsow_open);
4007                 if (lop != NULL) {
4008                         if (LIST_NEXT(lop, nfso_list) != NULL)
4009                                 panic("nfsdlg mult opens");
4010                         /*
4011                          * Look for the same openowner against the server.
4012                          */
4013                         LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
4014                                 if (!NFSBCMP(lowp->nfsow_owner,
4015                                     owp->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
4016                                         newnfs_copycred(&dp->nfsdl_cred, cred);
4017                                         ret = nfscl_moveopen(vp, clp, nmp, lop,
4018                                             owp, dp, cred, p);
4019                                         if (ret == NFSERR_STALECLIENTID ||
4020                                             ret == NFSERR_STALEDONTRECOVER ||
4021                                             ret == NFSERR_BADSESSION) {
4022                                                 if (gotvp)
4023                                                         vrele(vp);
4024                                                 return (ret);
4025                                         }
4026                                         if (ret) {
4027                                                 nfscl_freeopen(lop, 1);
4028                                                 if (!error)
4029                                                         error = ret;
4030                                         }
4031                                         break;
4032                                 }
4033                         }
4034
4035                         /*
4036                          * If no openowner found, create one and get an open
4037                          * for it.
4038                          */
4039                         if (owp == NULL) {
4040                                 nowp = malloc(
4041                                     sizeof (struct nfsclowner), M_NFSCLOWNER,
4042                                     M_WAITOK);
4043                                 nfscl_newopen(clp, NULL, &owp, &nowp, &op,
4044                                     NULL, lowp->nfsow_owner, dp->nfsdl_fh,
4045                                     dp->nfsdl_fhlen, NULL, NULL);
4046                                 newnfs_copycred(&dp->nfsdl_cred, cred);
4047                                 ret = nfscl_moveopen(vp, clp, nmp, lop,
4048                                     owp, dp, cred, p);
4049                                 if (ret) {
4050                                         nfscl_freeopenowner(owp, 0);
4051                                         if (ret == NFSERR_STALECLIENTID ||
4052                                             ret == NFSERR_STALEDONTRECOVER ||
4053                                             ret == NFSERR_BADSESSION) {
4054                                                 if (gotvp)
4055                                                         vrele(vp);
4056                                                 return (ret);
4057                                         }
4058                                         if (ret) {
4059                                                 nfscl_freeopen(lop, 1);
4060                                                 if (!error)
4061                                                         error = ret;
4062                                         }
4063                                 }
4064                         }
4065                 }
4066         }
4067
4068         /*
4069          * Now, get byte range locks for any locks done locally.
4070          */
4071         LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4072                 LIST_FOREACH(lckp, &lp->nfsl_lock, nfslo_list) {
4073                         newnfs_copycred(&dp->nfsdl_cred, cred);
4074                         ret = nfscl_relock(vp, clp, nmp, lp, lckp, cred, p);
4075                         if (ret == NFSERR_STALESTATEID ||
4076                             ret == NFSERR_STALEDONTRECOVER ||
4077                             ret == NFSERR_STALECLIENTID ||
4078                             ret == NFSERR_BADSESSION) {
4079                                 if (gotvp)
4080                                         vrele(vp);
4081                                 return (ret);
4082                         }
4083                         if (ret && !error)
4084                                 error = ret;
4085                 }
4086         }
4087         if (gotvp)
4088                 vrele(vp);
4089         return (error);
4090 }
4091
4092 /*
4093  * Move a locally issued open over to an owner on the state list.
4094  * SIDE EFFECT: If it needs to sleep (do an rpc), it unlocks clstate and
4095  * returns with it unlocked.
4096  */
4097 static int
4098 nfscl_moveopen(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp,
4099     struct nfsclopen *lop, struct nfsclowner *owp, struct nfscldeleg *dp,
4100     struct ucred *cred, NFSPROC_T *p)
4101 {
4102         struct nfsclopen *op, *nop;
4103         struct nfscldeleg *ndp;
4104         struct nfsnode *np;
4105         int error = 0, newone;
4106
4107         /*
4108          * First, look for an appropriate open, If found, just increment the
4109          * opencnt in it.
4110          */
4111         LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
4112                 if ((op->nfso_mode & lop->nfso_mode) == lop->nfso_mode &&
4113                     op->nfso_fhlen == lop->nfso_fhlen &&
4114                     !NFSBCMP(op->nfso_fh, lop->nfso_fh, op->nfso_fhlen)) {
4115                         op->nfso_opencnt += lop->nfso_opencnt;
4116                         nfscl_freeopen(lop, 1);
4117                         return (0);
4118                 }
4119         }
4120
4121         /* No appropriate open, so we have to do one against the server. */
4122         np = VTONFS(vp);
4123         nop = malloc(sizeof (struct nfsclopen) +
4124             lop->nfso_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
4125         newone = 0;
4126         nfscl_newopen(clp, NULL, &owp, NULL, &op, &nop, owp->nfsow_owner,
4127             lop->nfso_fh, lop->nfso_fhlen, cred, &newone);
4128         ndp = dp;
4129         error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data, np->n_v4->n4_fhlen,
4130             lop->nfso_fh, lop->nfso_fhlen, lop->nfso_mode, op,
4131             NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, &ndp, 0, 0, cred, p);
4132         if (error) {
4133                 if (newone)
4134                         nfscl_freeopen(op, 0);
4135         } else {
4136                 op->nfso_mode |= lop->nfso_mode;
4137                 op->nfso_opencnt += lop->nfso_opencnt;
4138                 nfscl_freeopen(lop, 1);
4139         }
4140         if (nop != NULL)
4141                 free(nop, M_NFSCLOPEN);
4142         if (ndp != NULL) {
4143                 /*
4144                  * What should I do with the returned delegation, since the
4145                  * delegation is being recalled? For now, just printf and
4146                  * through it away.
4147                  */
4148                 printf("Moveopen returned deleg\n");
4149                 free(ndp, M_NFSCLDELEG);
4150         }
4151         return (error);
4152 }
4153
4154 /*
4155  * Recall all delegations on this client.
4156  */
4157 static void
4158 nfscl_totalrecall(struct nfsclclient *clp)
4159 {
4160         struct nfscldeleg *dp;
4161
4162         TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
4163                 if ((dp->nfsdl_flags & NFSCLDL_DELEGRET) == 0)
4164                         dp->nfsdl_flags |= NFSCLDL_RECALL;
4165         }
4166 }
4167
4168 /*
4169  * Relock byte ranges. Called for delegation recall and state expiry.
4170  */
4171 static int
4172 nfscl_relock(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp,
4173     struct nfscllockowner *lp, struct nfscllock *lop, struct ucred *cred,
4174     NFSPROC_T *p)
4175 {
4176         struct nfscllockowner *nlp;
4177         struct nfsfh *nfhp;
4178         u_int64_t off, len;
4179         int error, newone, donelocally;
4180
4181         off = lop->nfslo_first;
4182         len = lop->nfslo_end - lop->nfslo_first;
4183         error = nfscl_getbytelock(vp, off, len, lop->nfslo_type, cred, p,
4184             clp, 1, NULL, lp->nfsl_lockflags, lp->nfsl_owner,
4185             lp->nfsl_openowner, &nlp, &newone, &donelocally);
4186         if (error || donelocally)
4187                 return (error);
4188         nfhp = VTONFS(vp)->n_fhp;
4189         error = nfscl_trylock(nmp, vp, nfhp->nfh_fh,
4190             nfhp->nfh_len, nlp, newone, 0, off,
4191             len, lop->nfslo_type, cred, p);
4192         if (error)
4193                 nfscl_freelockowner(nlp, 0);
4194         return (error);
4195 }
4196
4197 /*
4198  * Called to re-open a file. Basically get a vnode for the file handle
4199  * and then call nfsrpc_openrpc() to do the rest.
4200  */
4201 static int
4202 nfsrpc_reopen(struct nfsmount *nmp, u_int8_t *fhp, int fhlen,
4203     u_int32_t mode, struct nfsclopen *op, struct nfscldeleg **dpp,
4204     struct ucred *cred, NFSPROC_T *p)
4205 {
4206         struct nfsnode *np;
4207         vnode_t vp;
4208         int error;
4209
4210         error = nfscl_ngetreopen(nmp->nm_mountp, fhp, fhlen, p, &np);
4211         if (error)
4212                 return (error);
4213         vp = NFSTOV(np);
4214         if (np->n_v4 != NULL) {
4215                 error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data,
4216                     np->n_v4->n4_fhlen, fhp, fhlen, mode, op,
4217                     NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, dpp, 0, 0,
4218                     cred, p);
4219         } else {
4220                 error = EINVAL;
4221         }
4222         vrele(vp);
4223         return (error);
4224 }
4225
4226 /*
4227  * Try an open against the server. Just call nfsrpc_openrpc(), retrying while
4228  * NFSERR_DELAY. Also, try system credentials, if the passed in credentials
4229  * fail.
4230  */
4231 static int
4232 nfscl_tryopen(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen,
4233     u_int8_t *newfhp, int newfhlen, u_int32_t mode, struct nfsclopen *op,
4234     u_int8_t *name, int namelen, struct nfscldeleg **ndpp,
4235     int reclaim, u_int32_t delegtype, struct ucred *cred, NFSPROC_T *p)
4236 {
4237         int error;
4238
4239         do {
4240                 error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp, newfhlen,
4241                     mode, op, name, namelen, ndpp, reclaim, delegtype, cred, p,
4242                     0, 0);
4243                 if (error == NFSERR_DELAY)
4244                         (void) nfs_catnap(PZERO, error, "nfstryop");
4245         } while (error == NFSERR_DELAY);
4246         if (error == EAUTH || error == EACCES) {
4247                 /* Try again using system credentials */
4248                 newnfs_setroot(cred);
4249                 do {
4250                     error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp,
4251                         newfhlen, mode, op, name, namelen, ndpp, reclaim,
4252                         delegtype, cred, p, 1, 0);
4253                     if (error == NFSERR_DELAY)
4254                         (void) nfs_catnap(PZERO, error, "nfstryop");
4255                 } while (error == NFSERR_DELAY);
4256         }
4257         return (error);
4258 }
4259
4260 /*
4261  * Try a byte range lock. Just loop on nfsrpc_lock() while it returns
4262  * NFSERR_DELAY. Also, retry with system credentials, if the provided
4263  * cred don't work.
4264  */
4265 static int
4266 nfscl_trylock(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp,
4267     int fhlen, struct nfscllockowner *nlp, int newone, int reclaim,
4268     u_int64_t off, u_int64_t len, short type, struct ucred *cred, NFSPROC_T *p)
4269 {
4270         struct nfsrv_descript nfsd, *nd = &nfsd;
4271         int error;
4272
4273         do {
4274                 error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp, newone,
4275                     reclaim, off, len, type, cred, p, 0);
4276                 if (!error && nd->nd_repstat == NFSERR_DELAY)
4277                         (void) nfs_catnap(PZERO, (int)nd->nd_repstat,
4278                             "nfstrylck");
4279         } while (!error && nd->nd_repstat == NFSERR_DELAY);
4280         if (!error)
4281                 error = nd->nd_repstat;
4282         if (error == EAUTH || error == EACCES) {
4283                 /* Try again using root credentials */
4284                 newnfs_setroot(cred);
4285                 do {
4286                         error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp,
4287                             newone, reclaim, off, len, type, cred, p, 1);
4288                         if (!error && nd->nd_repstat == NFSERR_DELAY)
4289                                 (void) nfs_catnap(PZERO, (int)nd->nd_repstat,
4290                                     "nfstrylck");
4291                 } while (!error && nd->nd_repstat == NFSERR_DELAY);
4292                 if (!error)
4293                         error = nd->nd_repstat;
4294         }
4295         return (error);
4296 }
4297
4298 /*
4299  * Try a delegreturn against the server. Just call nfsrpc_delegreturn(),
4300  * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
4301  * credentials fail.
4302  */
4303 static int
4304 nfscl_trydelegreturn(struct nfscldeleg *dp, struct ucred *cred,
4305     struct nfsmount *nmp, NFSPROC_T *p)
4306 {
4307         int error;
4308
4309         do {
4310                 error = nfsrpc_delegreturn(dp, cred, nmp, p, 0);
4311                 if (error == NFSERR_DELAY)
4312                         (void) nfs_catnap(PZERO, error, "nfstrydp");
4313         } while (error == NFSERR_DELAY);
4314         if (error == EAUTH || error == EACCES) {
4315                 /* Try again using system credentials */
4316                 newnfs_setroot(cred);
4317                 do {
4318                         error = nfsrpc_delegreturn(dp, cred, nmp, p, 1);
4319                         if (error == NFSERR_DELAY)
4320                                 (void) nfs_catnap(PZERO, error, "nfstrydp");
4321                 } while (error == NFSERR_DELAY);
4322         }
4323         return (error);
4324 }
4325
4326 /*
4327  * Try a close against the server. Just call nfsrpc_closerpc(),
4328  * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
4329  * credentials fail.
4330  */
4331 int
4332 nfscl_tryclose(struct nfsclopen *op, struct ucred *cred,
4333     struct nfsmount *nmp, NFSPROC_T *p)
4334 {
4335         struct nfsrv_descript nfsd, *nd = &nfsd;
4336         int error;
4337
4338         do {
4339                 error = nfsrpc_closerpc(nd, nmp, op, cred, p, 0);
4340                 if (error == NFSERR_DELAY)
4341                         (void) nfs_catnap(PZERO, error, "nfstrycl");
4342         } while (error == NFSERR_DELAY);
4343         if (error == EAUTH || error == EACCES) {
4344                 /* Try again using system credentials */
4345                 newnfs_setroot(cred);
4346                 do {
4347                         error = nfsrpc_closerpc(nd, nmp, op, cred, p, 1);
4348                         if (error == NFSERR_DELAY)
4349                                 (void) nfs_catnap(PZERO, error, "nfstrycl");
4350                 } while (error == NFSERR_DELAY);
4351         }
4352         return (error);
4353 }
4354
4355 /*
4356  * Decide if a delegation on a file permits close without flushing writes
4357  * to the server. This might be a big performance win in some environments.
4358  * (Not useful until the client does caching on local stable storage.)
4359  */
4360 int
4361 nfscl_mustflush(vnode_t vp)
4362 {
4363         struct nfsclclient *clp;
4364         struct nfscldeleg *dp;
4365         struct nfsnode *np;
4366         struct nfsmount *nmp;
4367
4368         np = VTONFS(vp);
4369         nmp = VFSTONFS(vp->v_mount);
4370         if (!NFSHASNFSV4(nmp))
4371                 return (1);
4372         NFSLOCKCLSTATE();
4373         clp = nfscl_findcl(nmp);
4374         if (clp == NULL) {
4375                 NFSUNLOCKCLSTATE();
4376                 return (1);
4377         }
4378         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4379         if (dp != NULL && (dp->nfsdl_flags &
4380             (NFSCLDL_WRITE | NFSCLDL_RECALL | NFSCLDL_DELEGRET)) ==
4381              NFSCLDL_WRITE &&
4382             (dp->nfsdl_sizelimit >= np->n_size ||
4383              !NFSHASSTRICT3530(nmp))) {
4384                 NFSUNLOCKCLSTATE();
4385                 return (0);
4386         }
4387         NFSUNLOCKCLSTATE();
4388         return (1);
4389 }
4390
4391 /*
4392  * See if a (write) delegation exists for this file.
4393  */
4394 int
4395 nfscl_nodeleg(vnode_t vp, int writedeleg)
4396 {
4397         struct nfsclclient *clp;
4398         struct nfscldeleg *dp;
4399         struct nfsnode *np;
4400         struct nfsmount *nmp;
4401
4402         np = VTONFS(vp);
4403         nmp = VFSTONFS(vp->v_mount);
4404         if (!NFSHASNFSV4(nmp))
4405                 return (1);
4406         NFSLOCKCLSTATE();
4407         clp = nfscl_findcl(nmp);
4408         if (clp == NULL) {
4409                 NFSUNLOCKCLSTATE();
4410                 return (1);
4411         }
4412         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4413         if (dp != NULL &&
4414             (dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_DELEGRET)) == 0 &&
4415             (writedeleg == 0 || (dp->nfsdl_flags & NFSCLDL_WRITE) ==
4416              NFSCLDL_WRITE)) {
4417                 NFSUNLOCKCLSTATE();
4418                 return (0);
4419         }
4420         NFSUNLOCKCLSTATE();
4421         return (1);
4422 }
4423
4424 /*
4425  * Look for an associated delegation that should be DelegReturned.
4426  */
4427 int
4428 nfscl_removedeleg(vnode_t vp, NFSPROC_T *p, nfsv4stateid_t *stp)
4429 {
4430         struct nfsclclient *clp;
4431         struct nfscldeleg *dp;
4432         struct nfsclowner *owp;
4433         struct nfscllockowner *lp;
4434         struct nfsmount *nmp;
4435         struct ucred *cred;
4436         struct nfsnode *np;
4437         int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
4438
4439         nmp = VFSTONFS(vp->v_mount);
4440         np = VTONFS(vp);
4441         NFSLOCKCLSTATE();
4442         /*
4443          * Loop around waiting for:
4444          * - outstanding I/O operations on delegations to complete
4445          * - for a delegation on vp that has state, lock the client and
4446          *   do a recall
4447          * - return delegation with no state
4448          */
4449         while (1) {
4450                 clp = nfscl_findcl(nmp);
4451                 if (clp == NULL) {
4452                         NFSUNLOCKCLSTATE();
4453                         return (retcnt);
4454                 }
4455                 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4456                     np->n_fhp->nfh_len);
4457                 if (dp != NULL) {
4458                     /*
4459                      * Wait for outstanding I/O ops to be done.
4460                      */
4461                     if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4462                         if (igotlock) {
4463                             nfsv4_unlock(&clp->nfsc_lock, 0);
4464                             igotlock = 0;
4465                         }
4466                         dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4467                         (void) nfsmsleep(&dp->nfsdl_rwlock,
4468                             NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4469                         continue;
4470                     }
4471                     needsrecall = 0;
4472                     LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4473                         if (!LIST_EMPTY(&owp->nfsow_open)) {
4474                             needsrecall = 1;
4475                             break;
4476                         }
4477                     }
4478                     if (!needsrecall) {
4479                         LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4480                             if (!LIST_EMPTY(&lp->nfsl_lock)) {
4481                                 needsrecall = 1;
4482                                 break;
4483                             }
4484                         }
4485                     }
4486                     if (needsrecall && !triedrecall) {
4487                         dp->nfsdl_flags |= NFSCLDL_DELEGRET;
4488                         islept = 0;
4489                         while (!igotlock) {
4490                             igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
4491                                 &islept, NFSCLSTATEMUTEXPTR, NULL);
4492                             if (islept)
4493                                 break;
4494                         }
4495                         if (islept)
4496                             continue;
4497                         NFSUNLOCKCLSTATE();
4498                         cred = newnfs_getcred();
4499                         newnfs_copycred(&dp->nfsdl_cred, cred);
4500                         (void) nfscl_recalldeleg(clp, nmp, dp, vp, cred, p, 0);
4501                         NFSFREECRED(cred);
4502                         triedrecall = 1;
4503                         NFSLOCKCLSTATE();
4504                         nfsv4_unlock(&clp->nfsc_lock, 0);
4505                         igotlock = 0;
4506                         continue;
4507                     }
4508                     *stp = dp->nfsdl_stateid;
4509                     retcnt = 1;
4510                     nfscl_cleandeleg(dp);
4511                     nfscl_freedeleg(&clp->nfsc_deleg, dp);
4512                 }
4513                 if (igotlock)
4514                     nfsv4_unlock(&clp->nfsc_lock, 0);
4515                 NFSUNLOCKCLSTATE();
4516                 return (retcnt);
4517         }
4518 }
4519
4520 /*
4521  * Look for associated delegation(s) that should be DelegReturned.
4522  */
4523 int
4524 nfscl_renamedeleg(vnode_t fvp, nfsv4stateid_t *fstp, int *gotfdp, vnode_t tvp,
4525     nfsv4stateid_t *tstp, int *gottdp, NFSPROC_T *p)
4526 {
4527         struct nfsclclient *clp;
4528         struct nfscldeleg *dp;
4529         struct nfsclowner *owp;
4530         struct nfscllockowner *lp;
4531         struct nfsmount *nmp;
4532         struct ucred *cred;
4533         struct nfsnode *np;
4534         int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
4535
4536         nmp = VFSTONFS(fvp->v_mount);
4537         *gotfdp = 0;
4538         *gottdp = 0;
4539         NFSLOCKCLSTATE();
4540         /*
4541          * Loop around waiting for:
4542          * - outstanding I/O operations on delegations to complete
4543          * - for a delegation on fvp that has state, lock the client and
4544          *   do a recall
4545          * - return delegation(s) with no state.
4546          */
4547         while (1) {
4548                 clp = nfscl_findcl(nmp);
4549                 if (clp == NULL) {
4550                         NFSUNLOCKCLSTATE();
4551                         return (retcnt);
4552                 }
4553                 np = VTONFS(fvp);
4554                 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4555                     np->n_fhp->nfh_len);
4556                 if (dp != NULL && *gotfdp == 0) {
4557                     /*
4558                      * Wait for outstanding I/O ops to be done.
4559                      */
4560                     if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4561                         if (igotlock) {
4562                             nfsv4_unlock(&clp->nfsc_lock, 0);
4563                             igotlock = 0;
4564                         }
4565                         dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4566                         (void) nfsmsleep(&dp->nfsdl_rwlock,
4567                             NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4568                         continue;
4569                     }
4570                     needsrecall = 0;
4571                     LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4572                         if (!LIST_EMPTY(&owp->nfsow_open)) {
4573                             needsrecall = 1;
4574                             break;
4575                         }
4576                     }
4577                     if (!needsrecall) {
4578                         LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4579                             if (!LIST_EMPTY(&lp->nfsl_lock)) {
4580                                 needsrecall = 1;
4581                                 break;
4582                             }
4583                         }
4584                     }
4585                     if (needsrecall && !triedrecall) {
4586                         dp->nfsdl_flags |= NFSCLDL_DELEGRET;
4587                         islept = 0;
4588                         while (!igotlock) {
4589                             igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
4590                                 &islept, NFSCLSTATEMUTEXPTR, NULL);
4591                             if (islept)
4592                                 break;
4593                         }
4594                         if (islept)
4595                             continue;
4596                         NFSUNLOCKCLSTATE();
4597                         cred = newnfs_getcred();
4598                         newnfs_copycred(&dp->nfsdl_cred, cred);
4599                         (void) nfscl_recalldeleg(clp, nmp, dp, fvp, cred, p, 0);
4600                         NFSFREECRED(cred);
4601                         triedrecall = 1;
4602                         NFSLOCKCLSTATE();
4603                         nfsv4_unlock(&clp->nfsc_lock, 0);
4604                         igotlock = 0;
4605                         continue;
4606                     }
4607                     *fstp = dp->nfsdl_stateid;
4608                     retcnt++;
4609                     *gotfdp = 1;
4610                     nfscl_cleandeleg(dp);
4611                     nfscl_freedeleg(&clp->nfsc_deleg, dp);
4612                 }
4613                 if (igotlock) {
4614                     nfsv4_unlock(&clp->nfsc_lock, 0);
4615                     igotlock = 0;
4616                 }
4617                 if (tvp != NULL) {
4618                     np = VTONFS(tvp);
4619                     dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4620                         np->n_fhp->nfh_len);
4621                     if (dp != NULL && *gottdp == 0) {
4622                         /*
4623                          * Wait for outstanding I/O ops to be done.
4624                          */
4625                         if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4626                             dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4627                             (void) nfsmsleep(&dp->nfsdl_rwlock,
4628                                 NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4629                             continue;
4630                         }
4631                         LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4632                             if (!LIST_EMPTY(&owp->nfsow_open)) {
4633                                 NFSUNLOCKCLSTATE();
4634                                 return (retcnt);
4635                             }
4636                         }
4637                         LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4638                             if (!LIST_EMPTY(&lp->nfsl_lock)) {
4639                                 NFSUNLOCKCLSTATE();
4640                                 return (retcnt);
4641                             }
4642                         }
4643                         *tstp = dp->nfsdl_stateid;
4644                         retcnt++;
4645                         *gottdp = 1;
4646                         nfscl_cleandeleg(dp);
4647                         nfscl_freedeleg(&clp->nfsc_deleg, dp);
4648                     }
4649                 }
4650                 NFSUNLOCKCLSTATE();
4651                 return (retcnt);
4652         }
4653 }
4654
4655 /*
4656  * Get a reference on the clientid associated with the mount point.
4657  * Return 1 if success, 0 otherwise.
4658  */
4659 int
4660 nfscl_getref(struct nfsmount *nmp)
4661 {
4662         struct nfsclclient *clp;
4663
4664         NFSLOCKCLSTATE();
4665         clp = nfscl_findcl(nmp);
4666         if (clp == NULL) {
4667                 NFSUNLOCKCLSTATE();
4668                 return (0);
4669         }
4670         nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, NULL);
4671         NFSUNLOCKCLSTATE();
4672         return (1);
4673 }
4674
4675 /*
4676  * Release a reference on a clientid acquired with the above call.
4677  */
4678 void
4679 nfscl_relref(struct nfsmount *nmp)
4680 {
4681         struct nfsclclient *clp;
4682
4683         NFSLOCKCLSTATE();
4684         clp = nfscl_findcl(nmp);
4685         if (clp == NULL) {
4686                 NFSUNLOCKCLSTATE();
4687                 return;
4688         }
4689         nfsv4_relref(&clp->nfsc_lock);
4690         NFSUNLOCKCLSTATE();
4691 }
4692
4693 /*
4694  * Save the size attribute in the delegation, since the nfsnode
4695  * is going away.
4696  */
4697 void
4698 nfscl_reclaimnode(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                 dp->nfsdl_size = np->n_size;
4717         NFSUNLOCKCLSTATE();
4718 }
4719
4720 /*
4721  * Get the saved size attribute in the delegation, since it is a
4722  * newly allocated nfsnode.
4723  */
4724 void
4725 nfscl_newnode(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                 np->n_size = dp->nfsdl_size;
4744         NFSUNLOCKCLSTATE();
4745 }
4746
4747 /*
4748  * If there is a valid write delegation for this file, set the modtime
4749  * to the local clock time.
4750  */
4751 void
4752 nfscl_delegmodtime(vnode_t vp)
4753 {
4754         struct nfsclclient *clp;
4755         struct nfscldeleg *dp;
4756         struct nfsnode *np = VTONFS(vp);
4757         struct nfsmount *nmp;
4758
4759         nmp = VFSTONFS(vp->v_mount);
4760         if (!NFSHASNFSV4(nmp))
4761                 return;
4762         NFSLOCKCLSTATE();
4763         clp = nfscl_findcl(nmp);
4764         if (clp == NULL) {
4765                 NFSUNLOCKCLSTATE();
4766                 return;
4767         }
4768         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4769         if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE)) {
4770                 nanotime(&dp->nfsdl_modtime);
4771                 dp->nfsdl_flags |= NFSCLDL_MODTIMESET;
4772         }
4773         NFSUNLOCKCLSTATE();
4774 }
4775
4776 /*
4777  * If there is a valid write delegation for this file with a modtime set,
4778  * put that modtime in mtime.
4779  */
4780 void
4781 nfscl_deleggetmodtime(vnode_t vp, struct timespec *mtime)
4782 {
4783         struct nfsclclient *clp;
4784         struct nfscldeleg *dp;
4785         struct nfsnode *np = VTONFS(vp);
4786         struct nfsmount *nmp;
4787
4788         nmp = VFSTONFS(vp->v_mount);
4789         if (!NFSHASNFSV4(nmp))
4790                 return;
4791         NFSLOCKCLSTATE();
4792         clp = nfscl_findcl(nmp);
4793         if (clp == NULL) {
4794                 NFSUNLOCKCLSTATE();
4795                 return;
4796         }
4797         dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4798         if (dp != NULL &&
4799             (dp->nfsdl_flags & (NFSCLDL_WRITE | NFSCLDL_MODTIMESET)) ==
4800             (NFSCLDL_WRITE | NFSCLDL_MODTIMESET))
4801                 *mtime = dp->nfsdl_modtime;
4802         NFSUNLOCKCLSTATE();
4803 }
4804
4805 static int
4806 nfscl_errmap(struct nfsrv_descript *nd, u_int32_t minorvers)
4807 {
4808         short *defaulterrp, *errp;
4809
4810         if (!nd->nd_repstat)
4811                 return (0);
4812         if (nd->nd_procnum == NFSPROC_NOOP)
4813                 return (txdr_unsigned(nd->nd_repstat & 0xffff));
4814         if (nd->nd_repstat == EBADRPC)
4815                 return (txdr_unsigned(NFSERR_BADXDR));
4816         if (nd->nd_repstat == NFSERR_MINORVERMISMATCH ||
4817             nd->nd_repstat == NFSERR_OPILLEGAL)
4818                 return (txdr_unsigned(nd->nd_repstat));
4819         if (nd->nd_repstat >= NFSERR_BADIOMODE && nd->nd_repstat < 20000 &&
4820             minorvers > NFSV4_MINORVERSION) {
4821                 /* NFSv4.n error. */
4822                 return (txdr_unsigned(nd->nd_repstat));
4823         }
4824         if (nd->nd_procnum < NFSV4OP_CBNOPS)
4825                 errp = defaulterrp = nfscl_cberrmap[nd->nd_procnum];
4826         else
4827                 return (txdr_unsigned(nd->nd_repstat));
4828         while (*++errp)
4829                 if (*errp == (short)nd->nd_repstat)
4830                         return (txdr_unsigned(nd->nd_repstat));
4831         return (txdr_unsigned(*defaulterrp));
4832 }
4833
4834 /*
4835  * Called to find/add a layout to a client.
4836  * This function returns the layout with a refcnt (shared lock) upon
4837  * success (returns 0) or with no lock/refcnt on the layout when an
4838  * error is returned.
4839  * If a layout is passed in via lypp, it is locked (exclusively locked).
4840  */
4841 int
4842 nfscl_layout(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen,
4843     nfsv4stateid_t *stateidp, int layouttype, int retonclose,
4844     struct nfsclflayouthead *fhlp, struct nfscllayout **lypp,
4845     struct ucred *cred, NFSPROC_T *p)
4846 {
4847         struct nfsclclient *clp;
4848         struct nfscllayout *lyp, *tlyp;
4849         struct nfsclflayout *flp;
4850         struct nfsnode *np = VTONFS(vp);
4851         mount_t mp;
4852         int layout_passed_in;
4853
4854         mp = nmp->nm_mountp;
4855         layout_passed_in = 1;
4856         tlyp = NULL;
4857         lyp = *lypp;
4858         if (lyp == NULL) {
4859                 layout_passed_in = 0;
4860                 tlyp = malloc(sizeof(*tlyp) + fhlen - 1, M_NFSLAYOUT,
4861                     M_WAITOK | M_ZERO);
4862         }
4863
4864         NFSLOCKCLSTATE();
4865         clp = nmp->nm_clp;
4866         if (clp == NULL) {
4867                 if (layout_passed_in != 0)
4868                         nfsv4_unlock(&lyp->nfsly_lock, 0);
4869                 NFSUNLOCKCLSTATE();
4870                 if (tlyp != NULL)
4871                         free(tlyp, M_NFSLAYOUT);
4872                 return (EPERM);
4873         }
4874         if (lyp == NULL) {
4875                 /*
4876                  * Although no lyp was passed in, another thread might have
4877                  * allocated one. If one is found, just increment it's ref
4878                  * count and return it.
4879                  */
4880                 lyp = nfscl_findlayout(clp, fhp, fhlen);
4881                 if (lyp == NULL) {
4882                         lyp = tlyp;
4883                         tlyp = NULL;
4884                         lyp->nfsly_stateid.seqid = stateidp->seqid;
4885                         lyp->nfsly_stateid.other[0] = stateidp->other[0];
4886                         lyp->nfsly_stateid.other[1] = stateidp->other[1];
4887                         lyp->nfsly_stateid.other[2] = stateidp->other[2];
4888                         lyp->nfsly_lastbyte = 0;
4889                         LIST_INIT(&lyp->nfsly_flayread);
4890                         LIST_INIT(&lyp->nfsly_flayrw);
4891                         LIST_INIT(&lyp->nfsly_recall);
4892                         lyp->nfsly_filesid[0] = np->n_vattr.na_filesid[0];
4893                         lyp->nfsly_filesid[1] = np->n_vattr.na_filesid[1];
4894                         lyp->nfsly_clp = clp;
4895                         if (layouttype == NFSLAYOUT_FLEXFILE)
4896                                 lyp->nfsly_flags = NFSLY_FLEXFILE;
4897                         else
4898                                 lyp->nfsly_flags = NFSLY_FILES;
4899                         if (retonclose != 0)
4900                                 lyp->nfsly_flags |= NFSLY_RETONCLOSE;
4901                         lyp->nfsly_fhlen = fhlen;
4902                         NFSBCOPY(fhp, lyp->nfsly_fh, fhlen);
4903                         TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
4904                         LIST_INSERT_HEAD(NFSCLLAYOUTHASH(clp, fhp, fhlen), lyp,
4905                             nfsly_hash);
4906                         lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
4907                         nfscl_layoutcnt++;
4908                 } else {
4909                         if (retonclose != 0)
4910                                 lyp->nfsly_flags |= NFSLY_RETONCLOSE;
4911                         if (stateidp->seqid > lyp->nfsly_stateid.seqid)
4912                                 lyp->nfsly_stateid.seqid = stateidp->seqid;
4913                         TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list);
4914                         TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
4915                         lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
4916                 }
4917                 nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
4918                 if (NFSCL_FORCEDISM(mp)) {
4919                         NFSUNLOCKCLSTATE();
4920                         if (tlyp != NULL)
4921                                 free(tlyp, M_NFSLAYOUT);
4922                         return (EPERM);
4923                 }
4924                 *lypp = lyp;
4925         } else if (stateidp->seqid > lyp->nfsly_stateid.seqid)
4926                 lyp->nfsly_stateid.seqid = stateidp->seqid;
4927
4928         /* Merge the new list of File Layouts into the list. */
4929         flp = LIST_FIRST(fhlp);
4930         if (flp != NULL) {
4931                 if (flp->nfsfl_iomode == NFSLAYOUTIOMODE_READ)
4932                         nfscl_mergeflayouts(&lyp->nfsly_flayread, fhlp);
4933                 else
4934                         nfscl_mergeflayouts(&lyp->nfsly_flayrw, fhlp);
4935         }
4936         if (layout_passed_in != 0)
4937                 nfsv4_unlock(&lyp->nfsly_lock, 1);
4938         NFSUNLOCKCLSTATE();
4939         if (tlyp != NULL)
4940                 free(tlyp, M_NFSLAYOUT);
4941         return (0);
4942 }
4943
4944 /*
4945  * Search for a layout by MDS file handle.
4946  * If one is found, it is returned with a refcnt (shared lock) iff
4947  * retflpp returned non-NULL and locked (exclusive locked) iff retflpp is
4948  * returned NULL.
4949  */
4950 struct nfscllayout *
4951 nfscl_getlayout(struct nfsclclient *clp, uint8_t *fhp, int fhlen,
4952     uint64_t off, struct nfsclflayout **retflpp, int *recalledp)
4953 {
4954         struct nfscllayout *lyp;
4955         mount_t mp;
4956         int error, igotlock;
4957
4958         mp = clp->nfsc_nmp->nm_mountp;
4959         *recalledp = 0;
4960         *retflpp = NULL;
4961         NFSLOCKCLSTATE();
4962         lyp = nfscl_findlayout(clp, fhp, fhlen);
4963         if (lyp != NULL) {
4964                 if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
4965                         TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list);
4966                         TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
4967                         lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
4968                         error = nfscl_findlayoutforio(lyp, off,
4969                             NFSV4OPEN_ACCESSREAD, retflpp);
4970                         if (error == 0)
4971                                 nfsv4_getref(&lyp->nfsly_lock, NULL,
4972                                     NFSCLSTATEMUTEXPTR, mp);
4973                         else {
4974                                 do {
4975                                         igotlock = nfsv4_lock(&lyp->nfsly_lock,
4976                                             1, NULL, NFSCLSTATEMUTEXPTR, mp);
4977                                 } while (igotlock == 0 && !NFSCL_FORCEDISM(mp));
4978                                 *retflpp = NULL;
4979                         }
4980                         if (NFSCL_FORCEDISM(mp)) {
4981                                 lyp = NULL;
4982                                 *recalledp = 1;
4983                         }
4984                 } else {
4985                         lyp = NULL;
4986                         *recalledp = 1;
4987                 }
4988         }
4989         NFSUNLOCKCLSTATE();
4990         return (lyp);
4991 }
4992
4993 /*
4994  * Search for a layout by MDS file handle. If one is found, mark in to be
4995  * recalled, if it already marked "return on close".
4996  */
4997 static void
4998 nfscl_retoncloselayout(vnode_t vp, struct nfsclclient *clp, uint8_t *fhp,
4999     int fhlen, struct nfsclrecalllayout **recallpp)
5000 {
5001         struct nfscllayout *lyp;
5002         uint32_t iomode;
5003
5004         if (vp->v_type != VREG || !NFSHASPNFS(VFSTONFS(vp->v_mount)) ||
5005             nfscl_enablecallb == 0 || nfs_numnfscbd == 0 ||
5006             (VTONFS(vp)->n_flag & NNOLAYOUT) != 0)
5007                 return;
5008         lyp = nfscl_findlayout(clp, fhp, fhlen);
5009         if (lyp != NULL && (lyp->nfsly_flags & (NFSLY_RETONCLOSE |
5010             NFSLY_RECALL)) == NFSLY_RETONCLOSE) {
5011                 iomode = 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, 0, 0, NULL,
5018                     *recallpp);
5019                 NFSCL_DEBUG(4, "retoncls recall iomode=%d\n", iomode);
5020                 *recallpp = NULL;
5021         }
5022 }
5023
5024 /*
5025  * Mark the layout to be recalled and with an error.
5026  * Also, disable the dsp from further use.
5027  */
5028 void
5029 nfscl_dserr(uint32_t op, uint32_t stat, struct nfscldevinfo *dp,
5030     struct nfscllayout *lyp, struct nfsclds *dsp)
5031 {
5032         struct nfsclrecalllayout *recallp;
5033         uint32_t iomode;
5034
5035         printf("DS being disabled, error=%d\n", stat);
5036         /* Set up the return of the layout. */
5037         recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_WAITOK);
5038         iomode = 0;
5039         NFSLOCKCLSTATE();
5040         if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
5041                 if (!LIST_EMPTY(&lyp->nfsly_flayread))
5042                         iomode |= NFSLAYOUTIOMODE_READ;
5043                 if (!LIST_EMPTY(&lyp->nfsly_flayrw))
5044                         iomode |= NFSLAYOUTIOMODE_RW;
5045                 (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE, lyp, iomode,
5046                     0, UINT64_MAX, lyp->nfsly_stateid.seqid, stat, op,
5047                     dp->nfsdi_deviceid, recallp);
5048                 NFSUNLOCKCLSTATE();
5049                 NFSCL_DEBUG(4, "nfscl_dserr recall iomode=%d\n", iomode);
5050         } else {
5051                 NFSUNLOCKCLSTATE();
5052                 free(recallp, M_NFSLAYRECALL);
5053         }
5054
5055         /* And shut the TCP connection down. */
5056         nfscl_cancelreqs(dsp);
5057 }
5058
5059 /*
5060  * Cancel all RPCs for this "dsp" by closing the connection.
5061  * Also, mark the session as defunct.
5062  * If NFSCLDS_SAMECONN is set, the connection is shared with other DSs and
5063  * cannot be shut down.
5064  */
5065 void
5066 nfscl_cancelreqs(struct nfsclds *dsp)
5067 {
5068         struct __rpc_client *cl;
5069         static int non_event;
5070
5071         NFSLOCKDS(dsp);
5072         if ((dsp->nfsclds_flags & (NFSCLDS_CLOSED | NFSCLDS_SAMECONN)) == 0 &&
5073             dsp->nfsclds_sockp != NULL &&
5074             dsp->nfsclds_sockp->nr_client != NULL) {
5075                 dsp->nfsclds_flags |= NFSCLDS_CLOSED;
5076                 cl = dsp->nfsclds_sockp->nr_client;
5077                 dsp->nfsclds_sess.nfsess_defunct = 1;
5078                 NFSUNLOCKDS(dsp);
5079                 CLNT_CLOSE(cl);
5080                 /*
5081                  * This 1sec sleep is done to reduce the number of reconnect
5082                  * attempts made on the DS while it has failed.
5083                  */
5084                 tsleep(&non_event, PVFS, "ndscls", hz);
5085                 return;
5086         }
5087         NFSUNLOCKDS(dsp);
5088 }
5089
5090 /*
5091  * Dereference a layout.
5092  */
5093 void
5094 nfscl_rellayout(struct nfscllayout *lyp, int exclocked)
5095 {
5096
5097         NFSLOCKCLSTATE();
5098         if (exclocked != 0)
5099                 nfsv4_unlock(&lyp->nfsly_lock, 0);
5100         else
5101                 nfsv4_relref(&lyp->nfsly_lock);
5102         NFSUNLOCKCLSTATE();
5103 }
5104
5105 /*
5106  * Search for a devinfo by deviceid. If one is found, return it after
5107  * acquiring a reference count on it.
5108  */
5109 struct nfscldevinfo *
5110 nfscl_getdevinfo(struct nfsclclient *clp, uint8_t *deviceid,
5111     struct nfscldevinfo *dip)
5112 {
5113
5114         NFSLOCKCLSTATE();
5115         if (dip == NULL)
5116                 dip = nfscl_finddevinfo(clp, deviceid);
5117         if (dip != NULL)
5118                 dip->nfsdi_refcnt++;
5119         NFSUNLOCKCLSTATE();
5120         return (dip);
5121 }
5122
5123 /*
5124  * Dereference a devinfo structure.
5125  */
5126 static void
5127 nfscl_reldevinfo_locked(struct nfscldevinfo *dip)
5128 {
5129
5130         dip->nfsdi_refcnt--;
5131         if (dip->nfsdi_refcnt == 0)
5132                 wakeup(&dip->nfsdi_refcnt);
5133 }
5134
5135 /*
5136  * Dereference a devinfo structure.
5137  */
5138 void
5139 nfscl_reldevinfo(struct nfscldevinfo *dip)
5140 {
5141
5142         NFSLOCKCLSTATE();
5143         nfscl_reldevinfo_locked(dip);
5144         NFSUNLOCKCLSTATE();
5145 }
5146
5147 /*
5148  * Find a layout for this file handle. Return NULL upon failure.
5149  */
5150 static struct nfscllayout *
5151 nfscl_findlayout(struct nfsclclient *clp, u_int8_t *fhp, int fhlen)
5152 {
5153         struct nfscllayout *lyp;
5154
5155         LIST_FOREACH(lyp, NFSCLLAYOUTHASH(clp, fhp, fhlen), nfsly_hash)
5156                 if (lyp->nfsly_fhlen == fhlen &&
5157                     !NFSBCMP(lyp->nfsly_fh, fhp, fhlen))
5158                         break;
5159         return (lyp);
5160 }
5161
5162 /*
5163  * Find a devinfo for this deviceid. Return NULL upon failure.
5164  */
5165 static struct nfscldevinfo *
5166 nfscl_finddevinfo(struct nfsclclient *clp, uint8_t *deviceid)
5167 {
5168         struct nfscldevinfo *dip;
5169
5170         LIST_FOREACH(dip, &clp->nfsc_devinfo, nfsdi_list)
5171                 if (NFSBCMP(dip->nfsdi_deviceid, deviceid, NFSX_V4DEVICEID)
5172                     == 0)
5173                         break;
5174         return (dip);
5175 }
5176
5177 /*
5178  * Merge the new file layout list into the main one, maintaining it in
5179  * increasing offset order.
5180  */
5181 static void
5182 nfscl_mergeflayouts(struct nfsclflayouthead *fhlp,
5183     struct nfsclflayouthead *newfhlp)
5184 {
5185         struct nfsclflayout *flp, *nflp, *prevflp, *tflp;
5186
5187         flp = LIST_FIRST(fhlp);
5188         prevflp = NULL;
5189         LIST_FOREACH_SAFE(nflp, newfhlp, nfsfl_list, tflp) {
5190                 while (flp != NULL && flp->nfsfl_off < nflp->nfsfl_off) {
5191                         prevflp = flp;
5192                         flp = LIST_NEXT(flp, nfsfl_list);
5193                 }
5194                 if (prevflp == NULL)
5195                         LIST_INSERT_HEAD(fhlp, nflp, nfsfl_list);
5196                 else
5197                         LIST_INSERT_AFTER(prevflp, nflp, nfsfl_list);
5198                 prevflp = nflp;
5199         }
5200 }
5201
5202 /*
5203  * Add this nfscldevinfo to the client, if it doesn't already exist.
5204  * This function consumes the structure pointed at by dip, if not NULL.
5205  */
5206 int
5207 nfscl_adddevinfo(struct nfsmount *nmp, struct nfscldevinfo *dip, int ind,
5208     struct nfsclflayout *flp)
5209 {
5210         struct nfsclclient *clp;
5211         struct nfscldevinfo *tdip;
5212         uint8_t *dev;
5213
5214         NFSLOCKCLSTATE();
5215         clp = nmp->nm_clp;
5216         if (clp == NULL) {
5217                 NFSUNLOCKCLSTATE();
5218                 if (dip != NULL)
5219                         free(dip, M_NFSDEVINFO);
5220                 return (ENODEV);
5221         }
5222         if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5223                 dev = flp->nfsfl_dev;
5224         else
5225                 dev = flp->nfsfl_ffm[ind].dev;
5226         tdip = nfscl_finddevinfo(clp, dev);
5227         if (tdip != NULL) {
5228                 tdip->nfsdi_layoutrefs++;
5229                 if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5230                         flp->nfsfl_devp = tdip;
5231                 else
5232                         flp->nfsfl_ffm[ind].devp = tdip;
5233                 nfscl_reldevinfo_locked(tdip);
5234                 NFSUNLOCKCLSTATE();
5235                 if (dip != NULL)
5236                         free(dip, M_NFSDEVINFO);
5237                 return (0);
5238         }
5239         if (dip != NULL) {
5240                 LIST_INSERT_HEAD(&clp->nfsc_devinfo, dip, nfsdi_list);
5241                 dip->nfsdi_layoutrefs = 1;
5242                 if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5243                         flp->nfsfl_devp = dip;
5244                 else
5245                         flp->nfsfl_ffm[ind].devp = dip;
5246         }
5247         NFSUNLOCKCLSTATE();
5248         if (dip == NULL)
5249                 return (ENODEV);
5250         return (0);
5251 }
5252
5253 /*
5254  * Free up a layout structure and associated file layout structure(s).
5255  */
5256 void
5257 nfscl_freelayout(struct nfscllayout *layp)
5258 {
5259         struct nfsclflayout *flp, *nflp;
5260         struct nfsclrecalllayout *rp, *nrp;
5261
5262         LIST_FOREACH_SAFE(flp, &layp->nfsly_flayread, nfsfl_list, nflp) {
5263                 LIST_REMOVE(flp, nfsfl_list);
5264                 nfscl_freeflayout(flp);
5265         }
5266         LIST_FOREACH_SAFE(flp, &layp->nfsly_flayrw, nfsfl_list, nflp) {
5267                 LIST_REMOVE(flp, nfsfl_list);
5268                 nfscl_freeflayout(flp);
5269         }
5270         LIST_FOREACH_SAFE(rp, &layp->nfsly_recall, nfsrecly_list, nrp) {
5271                 LIST_REMOVE(rp, nfsrecly_list);
5272                 free(rp, M_NFSLAYRECALL);
5273         }
5274         nfscl_layoutcnt--;
5275         free(layp, M_NFSLAYOUT);
5276 }
5277
5278 /*
5279  * Free up a file layout structure.
5280  */
5281 void
5282 nfscl_freeflayout(struct nfsclflayout *flp)
5283 {
5284         int i, j;
5285
5286         if ((flp->nfsfl_flags & NFSFL_FILE) != 0) {
5287                 for (i = 0; i < flp->nfsfl_fhcnt; i++)
5288                         free(flp->nfsfl_fh[i], M_NFSFH);
5289                 if (flp->nfsfl_devp != NULL)
5290                         flp->nfsfl_devp->nfsdi_layoutrefs--;
5291         }
5292         if ((flp->nfsfl_flags & NFSFL_FLEXFILE) != 0)
5293                 for (i = 0; i < flp->nfsfl_mirrorcnt; i++) {
5294                         for (j = 0; j < flp->nfsfl_ffm[i].fhcnt; j++)
5295                                 free(flp->nfsfl_ffm[i].fh[j], M_NFSFH);
5296                         if (flp->nfsfl_ffm[i].devp != NULL)
5297                                 flp->nfsfl_ffm[i].devp->nfsdi_layoutrefs--;
5298                 }
5299         free(flp, M_NFSFLAYOUT);
5300 }
5301
5302 /*
5303  * Free up a file layout devinfo structure.
5304  */
5305 void
5306 nfscl_freedevinfo(struct nfscldevinfo *dip)
5307 {
5308
5309         free(dip, M_NFSDEVINFO);
5310 }
5311
5312 /*
5313  * Mark any layouts that match as recalled.
5314  */
5315 static int
5316 nfscl_layoutrecall(int recalltype, struct nfscllayout *lyp, uint32_t iomode,
5317     uint64_t off, uint64_t len, uint32_t stateseqid, uint32_t stat, uint32_t op,
5318     char *devid, struct nfsclrecalllayout *recallp)
5319 {
5320         struct nfsclrecalllayout *rp, *orp;
5321
5322         recallp->nfsrecly_recalltype = recalltype;
5323         recallp->nfsrecly_iomode = iomode;
5324         recallp->nfsrecly_stateseqid = stateseqid;
5325         recallp->nfsrecly_off = off;
5326         recallp->nfsrecly_len = len;
5327         recallp->nfsrecly_stat = stat;
5328         recallp->nfsrecly_op = op;
5329         if (devid != NULL)
5330                 NFSBCOPY(devid, recallp->nfsrecly_devid, NFSX_V4DEVICEID);
5331         /*
5332          * Order the list as file returns first, followed by fsid and any
5333          * returns, both in increasing stateseqid order.
5334          * Note that the seqids wrap around, so 1 is after 0xffffffff.
5335          * (I'm not sure this is correct because I find RFC5661 confusing
5336          *  on this, but hopefully it will work ok.)
5337          */
5338         orp = NULL;
5339         LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) {
5340                 orp = rp;
5341                 if ((recalltype == NFSLAYOUTRETURN_FILE &&
5342                      (rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE ||
5343                       nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) ||
5344                     (recalltype != NFSLAYOUTRETURN_FILE &&
5345                      rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE &&
5346                      nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) {
5347                         LIST_INSERT_BEFORE(rp, recallp, nfsrecly_list);
5348                         break;
5349                 }
5350
5351                 /*
5352                  * Put any error return on all the file returns that will
5353                  * preceed this one.
5354                  */
5355                 if (rp->nfsrecly_recalltype == NFSLAYOUTRETURN_FILE &&
5356                    stat != 0 && rp->nfsrecly_stat == 0) {
5357                         rp->nfsrecly_stat = stat;
5358                         rp->nfsrecly_op = op;
5359                         if (devid != NULL)
5360                                 NFSBCOPY(devid, rp->nfsrecly_devid,
5361                                     NFSX_V4DEVICEID);
5362                 }
5363         }
5364         if (rp == NULL) {
5365                 if (orp == NULL)
5366                         LIST_INSERT_HEAD(&lyp->nfsly_recall, recallp,
5367                             nfsrecly_list);
5368                 else
5369                         LIST_INSERT_AFTER(orp, recallp, nfsrecly_list);
5370         }
5371         lyp->nfsly_flags |= NFSLY_RECALL;
5372         wakeup(lyp->nfsly_clp);
5373         return (0);
5374 }
5375
5376 /*
5377  * Compare the two seqids for ordering. The trick is that the seqids can
5378  * wrap around from 0xffffffff->0, so check for the cases where one
5379  * has wrapped around.
5380  * Return 1 if seqid1 comes before seqid2, 0 otherwise.
5381  */
5382 static int
5383 nfscl_seq(uint32_t seqid1, uint32_t seqid2)
5384 {
5385
5386         if (seqid2 > seqid1 && (seqid2 - seqid1) >= 0x7fffffff)
5387                 /* seqid2 has wrapped around. */
5388                 return (0);
5389         if (seqid1 > seqid2 && (seqid1 - seqid2) >= 0x7fffffff)
5390                 /* seqid1 has wrapped around. */
5391                 return (1);
5392         if (seqid1 <= seqid2)
5393                 return (1);
5394         return (0);
5395 }
5396
5397 /*
5398  * Do a layout return for each of the recalls.
5399  */
5400 static void
5401 nfscl_layoutreturn(struct nfsmount *nmp, struct nfscllayout *lyp,
5402     struct ucred *cred, NFSPROC_T *p)
5403 {
5404         struct nfsclrecalllayout *rp;
5405         nfsv4stateid_t stateid;
5406         int layouttype;
5407
5408         NFSBCOPY(lyp->nfsly_stateid.other, stateid.other, NFSX_STATEIDOTHER);
5409         stateid.seqid = lyp->nfsly_stateid.seqid;
5410         if ((lyp->nfsly_flags & NFSLY_FILES) != 0)
5411                 layouttype = NFSLAYOUT_NFSV4_1_FILES;
5412         else
5413                 layouttype = NFSLAYOUT_FLEXFILE;
5414         LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) {
5415                 (void)nfsrpc_layoutreturn(nmp, lyp->nfsly_fh,
5416                     lyp->nfsly_fhlen, 0, layouttype,
5417                     rp->nfsrecly_iomode, rp->nfsrecly_recalltype,
5418                     rp->nfsrecly_off, rp->nfsrecly_len,
5419                     &stateid, cred, p, rp->nfsrecly_stat, rp->nfsrecly_op,
5420                     rp->nfsrecly_devid);
5421         }
5422 }
5423
5424 /*
5425  * Do the layout commit for a file layout.
5426  */
5427 static void
5428 nfscl_dolayoutcommit(struct nfsmount *nmp, struct nfscllayout *lyp,
5429     struct ucred *cred, NFSPROC_T *p)
5430 {
5431         struct nfsclflayout *flp;
5432         uint64_t len;
5433         int error, layouttype;
5434
5435         if ((lyp->nfsly_flags & NFSLY_FILES) != 0)
5436                 layouttype = NFSLAYOUT_NFSV4_1_FILES;
5437         else
5438                 layouttype = NFSLAYOUT_FLEXFILE;
5439         LIST_FOREACH(flp, &lyp->nfsly_flayrw, nfsfl_list) {
5440                 if (layouttype == NFSLAYOUT_FLEXFILE &&
5441                     (flp->nfsfl_fflags & NFSFLEXFLAG_NO_LAYOUTCOMMIT) != 0) {
5442                         NFSCL_DEBUG(4, "Flex file: no layoutcommit\n");
5443                         /* If not supported, don't bother doing it. */
5444                         NFSLOCKMNT(nmp);
5445                         nmp->nm_state |= NFSSTA_NOLAYOUTCOMMIT;
5446                         NFSUNLOCKMNT(nmp);
5447                         break;
5448                 } else if (flp->nfsfl_off <= lyp->nfsly_lastbyte) {
5449                         len = flp->nfsfl_end - flp->nfsfl_off;
5450                         error = nfsrpc_layoutcommit(nmp, lyp->nfsly_fh,
5451                             lyp->nfsly_fhlen, 0, flp->nfsfl_off, len,
5452                             lyp->nfsly_lastbyte, &lyp->nfsly_stateid,
5453                             layouttype, cred, p, NULL);
5454                         NFSCL_DEBUG(4, "layoutcommit err=%d\n", error);
5455                         if (error == NFSERR_NOTSUPP) {
5456                                 /* If not supported, don't bother doing it. */
5457                                 NFSLOCKMNT(nmp);
5458                                 nmp->nm_state |= NFSSTA_NOLAYOUTCOMMIT;
5459                                 NFSUNLOCKMNT(nmp);
5460                                 break;
5461                         }
5462                 }
5463         }
5464 }
5465
5466 /*
5467  * Commit all layouts for a file (vnode).
5468  */
5469 int
5470 nfscl_layoutcommit(vnode_t vp, NFSPROC_T *p)
5471 {
5472         struct nfsclclient *clp;
5473         struct nfscllayout *lyp;
5474         struct nfsnode *np = VTONFS(vp);
5475         mount_t mp;
5476         struct nfsmount *nmp;
5477
5478         mp = vp->v_mount;
5479         nmp = VFSTONFS(mp);
5480         if (NFSHASNOLAYOUTCOMMIT(nmp))
5481                 return (0);
5482         NFSLOCKCLSTATE();
5483         clp = nmp->nm_clp;
5484         if (clp == NULL) {
5485                 NFSUNLOCKCLSTATE();
5486                 return (EPERM);
5487         }
5488         lyp = nfscl_findlayout(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
5489         if (lyp == NULL) {
5490                 NFSUNLOCKCLSTATE();
5491                 return (EPERM);
5492         }
5493         nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
5494         if (NFSCL_FORCEDISM(mp)) {
5495                 NFSUNLOCKCLSTATE();
5496                 return (EPERM);
5497         }
5498 tryagain:
5499         if ((lyp->nfsly_flags & NFSLY_WRITTEN) != 0) {
5500                 lyp->nfsly_flags &= ~NFSLY_WRITTEN;
5501                 NFSUNLOCKCLSTATE();
5502                 NFSCL_DEBUG(4, "do layoutcommit2\n");
5503                 nfscl_dolayoutcommit(clp->nfsc_nmp, lyp, NFSPROCCRED(p), p);
5504                 NFSLOCKCLSTATE();
5505                 goto tryagain;
5506         }
5507         nfsv4_relref(&lyp->nfsly_lock);
5508         NFSUNLOCKCLSTATE();
5509         return (0);
5510 }