2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2009 Rick Macklem, University of Guelph
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
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.
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
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
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.)
83 #include <fs/nfs/nfsport.h>
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;
97 struct nfsclhead nfsclhead; /* Head of clientid list */
98 int nfscl_deleghighwater = NFSCLDELEGHIGHWATER;
99 int nfscl_layouthighwater = NFSCLLAYOUTHIGHWATER;
101 static int nfscl_delegcnt = 0;
102 static int nfscl_layoutcnt = 0;
103 static int nfscl_getopen(struct nfsclownerhead *, struct nfsclopenhash *,
104 u_int8_t *, int, u_int8_t *, u_int8_t *, u_int32_t,
105 struct nfscllockowner **, struct nfsclopen **);
106 static bool nfscl_checkown(struct nfsclowner *, struct nfsclopen *, uint8_t *,
107 uint8_t *, struct nfscllockowner **, struct nfsclopen **,
108 struct nfsclopen **);
109 static void nfscl_clrelease(struct nfsclclient *);
110 static void nfscl_cleanclient(struct nfsclclient *);
111 static void nfscl_expireclient(struct nfsclclient *, struct nfsmount *,
112 struct ucred *, NFSPROC_T *);
113 static int nfscl_expireopen(struct nfsclclient *, struct nfsclopen *,
114 struct nfsmount *, struct ucred *, NFSPROC_T *);
115 static void nfscl_recover(struct nfsclclient *, bool *, struct ucred *,
117 static void nfscl_insertlock(struct nfscllockowner *, struct nfscllock *,
118 struct nfscllock *, int);
119 static int nfscl_updatelock(struct nfscllockowner *, struct nfscllock **,
120 struct nfscllock **, int);
121 static void nfscl_delegreturnall(struct nfsclclient *, NFSPROC_T *);
122 static u_int32_t nfscl_nextcbident(void);
123 static mount_t nfscl_getmnt(int, uint8_t *, u_int32_t, struct nfsclclient **);
124 static struct nfsclclient *nfscl_getclnt(u_int32_t);
125 static struct nfsclclient *nfscl_getclntsess(uint8_t *);
126 static struct nfscldeleg *nfscl_finddeleg(struct nfsclclient *, u_int8_t *,
128 static void nfscl_retoncloselayout(vnode_t, struct nfsclclient *, uint8_t *,
129 int, struct nfsclrecalllayout **);
130 static void nfscl_reldevinfo_locked(struct nfscldevinfo *);
131 static struct nfscllayout *nfscl_findlayout(struct nfsclclient *, u_int8_t *,
133 static struct nfscldevinfo *nfscl_finddevinfo(struct nfsclclient *, uint8_t *);
134 static int nfscl_checkconflict(struct nfscllockownerhead *, struct nfscllock *,
135 u_int8_t *, struct nfscllock **);
136 static void nfscl_freealllocks(struct nfscllockownerhead *, int);
137 static int nfscl_localconflict(struct nfsclclient *, u_int8_t *, int,
138 struct nfscllock *, u_int8_t *, struct nfscldeleg *, struct nfscllock **);
139 static void nfscl_newopen(struct nfsclclient *, struct nfscldeleg *,
140 struct nfsclowner **, struct nfsclowner **, struct nfsclopen **,
141 struct nfsclopen **, u_int8_t *, u_int8_t *, int, struct ucred *, int *);
142 static int nfscl_moveopen(vnode_t , struct nfsclclient *,
143 struct nfsmount *, struct nfsclopen *, struct nfsclowner *,
144 struct nfscldeleg *, struct ucred *, NFSPROC_T *);
145 static void nfscl_totalrecall(struct nfsclclient *);
146 static int nfscl_relock(vnode_t , struct nfsclclient *, struct nfsmount *,
147 struct nfscllockowner *, struct nfscllock *, struct ucred *, NFSPROC_T *);
148 static int nfscl_tryopen(struct nfsmount *, vnode_t , u_int8_t *, int,
149 u_int8_t *, int, u_int32_t, struct nfsclopen *, u_int8_t *, int,
150 struct nfscldeleg **, int, u_int32_t, struct ucred *, NFSPROC_T *);
151 static int nfscl_trylock(struct nfsmount *, vnode_t , u_int8_t *,
152 int, struct nfscllockowner *, int, int, u_int64_t, u_int64_t, short,
153 struct ucred *, NFSPROC_T *);
154 static int nfsrpc_reopen(struct nfsmount *, u_int8_t *, int, u_int32_t,
155 struct nfsclopen *, struct nfscldeleg **, struct ucred *, NFSPROC_T *);
156 static void nfscl_freedeleg(struct nfscldeleghead *, struct nfscldeleg *,
158 static int nfscl_errmap(struct nfsrv_descript *, u_int32_t);
159 static void nfscl_cleanup_common(struct nfsclclient *, u_int8_t *);
160 static int nfscl_recalldeleg(struct nfsclclient *, struct nfsmount *,
161 struct nfscldeleg *, vnode_t, struct ucred *, NFSPROC_T *, int,
163 static void nfscl_freeopenowner(struct nfsclowner *, int);
164 static void nfscl_cleandeleg(struct nfscldeleg *);
165 static int nfscl_trydelegreturn(struct nfscldeleg *, struct ucred *,
166 struct nfsmount *, NFSPROC_T *);
167 static void nfscl_emptylockowner(struct nfscllockowner *,
168 struct nfscllockownerfhhead *);
169 static void nfscl_mergeflayouts(struct nfsclflayouthead *,
170 struct nfsclflayouthead *);
171 static int nfscl_layoutrecall(int, struct nfscllayout *, uint32_t, uint64_t,
172 uint64_t, uint32_t, uint32_t, uint32_t, char *, struct nfsclrecalllayout *);
173 static int nfscl_seq(uint32_t, uint32_t);
174 static void nfscl_layoutreturn(struct nfsmount *, struct nfscllayout *,
175 struct ucred *, NFSPROC_T *);
176 static void nfscl_dolayoutcommit(struct nfsmount *, struct nfscllayout *,
177 struct ucred *, NFSPROC_T *);
179 static short nfscberr_null[] = {
184 static short nfscberr_getattr[] = {
193 static short nfscberr_recall[] = {
203 static short *nfscl_cberrmap[] = {
211 #define NETFAMILY(clp) \
212 (((clp)->nfsc_flags & NFSCLFLAGS_AFINET6) ? AF_INET6 : AF_INET)
215 * Called for an open operation.
216 * If the nfhp argument is NULL, just get an openowner.
219 nfscl_open(vnode_t vp, u_int8_t *nfhp, int fhlen, u_int32_t amode, int usedeleg,
220 struct ucred *cred, NFSPROC_T *p, struct nfsclowner **owpp,
221 struct nfsclopen **opp, int *newonep, int *retp, int lockit)
223 struct nfsclclient *clp;
224 struct nfsclowner *owp, *nowp;
225 struct nfsclopen *op = NULL, *nop = NULL;
226 struct nfscldeleg *dp;
227 struct nfsclownerhead *ohp;
228 u_int8_t own[NFSV4CL_LOCKNAMELEN];
239 * Might need one or both of these, so MALLOC them now, to
240 * avoid a tsleep() in MALLOC later.
242 nowp = malloc(sizeof (struct nfsclowner),
243 M_NFSCLOWNER, M_WAITOK);
245 nop = malloc(sizeof (struct nfsclopen) +
246 fhlen - 1, M_NFSCLOPEN, M_WAITOK);
247 nop->nfso_hash.le_prev = NULL;
249 ret = nfscl_getcl(vp->v_mount, cred, p, 1, &clp);
251 free(nowp, M_NFSCLOWNER);
253 free(nop, M_NFSCLOPEN);
258 * Get the Open iff it already exists.
259 * If none found, add the new one or return error, depending upon
264 /* First check the delegation list */
265 if (nfhp != NULL && usedeleg) {
266 LIST_FOREACH(dp, NFSCLDELEGHASH(clp, nfhp, fhlen), nfsdl_hash) {
267 if (dp->nfsdl_fhlen == fhlen &&
268 !NFSBCMP(nfhp, dp->nfsdl_fh, fhlen)) {
269 if (!(amode & NFSV4OPEN_ACCESSWRITE) ||
270 (dp->nfsdl_flags & NFSCLDL_WRITE))
278 /* For NFSv4.1/4.2 and this option, use a single open_owner. */
279 if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount)))
280 nfscl_filllockowner(NULL, own, F_POSIX);
282 nfscl_filllockowner(p->td_proc, own, F_POSIX);
284 ohp = &dp->nfsdl_owner;
286 ohp = &clp->nfsc_owner;
287 /* Now, search for an openowner */
288 LIST_FOREACH(owp, ohp, nfsow_list) {
289 if (!NFSBCMP(owp->nfsow_owner, own, NFSV4CL_LOCKNAMELEN))
294 * Create a new open, as required.
296 nfscl_newopen(clp, dp, &owp, &nowp, &op, &nop, own, nfhp, fhlen,
300 * Now, check the mode on the open and return the appropriate
304 if (nfhp != NULL && dp != NULL && nop == NULL)
305 /* new local open on delegation */
306 *retp = NFSCLOPEN_SETCRED;
308 *retp = NFSCLOPEN_OK;
310 if (op != NULL && (amode & ~(op->nfso_mode))) {
311 op->nfso_mode |= amode;
312 if (retp != NULL && dp == NULL)
313 *retp = NFSCLOPEN_DOOPEN;
317 * Serialize modifications to the open owner for multiple threads
318 * within the same process using a read/write sleep lock.
319 * For NFSv4.1 and a single OpenOwner, allow concurrent open operations
320 * by acquiring a shared lock. The close operations still use an
321 * exclusive lock for this case.
324 if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount))) {
326 * Get a shared lock on the OpenOwner, but first
327 * wait for any pending exclusive lock, so that the
328 * exclusive locker gets priority.
330 nfsv4_lock(&owp->nfsow_rwlock, 0, NULL,
331 NFSCLSTATEMUTEXPTR, NULL);
332 nfsv4_getref(&owp->nfsow_rwlock, NULL,
333 NFSCLSTATEMUTEXPTR, NULL);
335 nfscl_lockexcl(&owp->nfsow_rwlock, NFSCLSTATEMUTEXPTR);
339 free(nowp, M_NFSCLOWNER);
341 free(nop, M_NFSCLOPEN);
350 * Create a new open, as required.
353 nfscl_newopen(struct nfsclclient *clp, struct nfscldeleg *dp,
354 struct nfsclowner **owpp, struct nfsclowner **nowpp, struct nfsclopen **opp,
355 struct nfsclopen **nopp, u_int8_t *own, u_int8_t *fhp, int fhlen,
356 struct ucred *cred, int *newonep)
358 struct nfsclowner *owp = *owpp, *nowp;
359 struct nfsclopen *op, *nop;
369 if (owp == NULL && nowp != NULL) {
370 NFSBCOPY(own, nowp->nfsow_owner, NFSV4CL_LOCKNAMELEN);
371 LIST_INIT(&nowp->nfsow_open);
372 nowp->nfsow_clp = clp;
373 nowp->nfsow_seqid = 0;
374 nowp->nfsow_defunct = 0;
375 nfscl_lockinit(&nowp->nfsow_rwlock);
377 nfsstatsv1.cllocalopenowners++;
378 LIST_INSERT_HEAD(&dp->nfsdl_owner, nowp, nfsow_list);
380 nfsstatsv1.clopenowners++;
381 LIST_INSERT_HEAD(&clp->nfsc_owner, nowp, nfsow_list);
389 /* If an fhp has been specified, create an Open as well. */
391 /* and look for the correct open, based upon FH */
392 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
393 if (op->nfso_fhlen == fhlen &&
394 !NFSBCMP(op->nfso_fh, fhp, fhlen))
397 if (op == NULL && nop != NULL) {
400 nop->nfso_opencnt = 0;
401 nop->nfso_posixlock = 1;
402 nop->nfso_fhlen = fhlen;
403 NFSBCOPY(fhp, nop->nfso_fh, fhlen);
404 LIST_INIT(&nop->nfso_lock);
405 nop->nfso_stateid.seqid = 0;
406 nop->nfso_stateid.other[0] = 0;
407 nop->nfso_stateid.other[1] = 0;
408 nop->nfso_stateid.other[2] = 0;
409 KASSERT(cred != NULL, ("%s: cred NULL\n", __func__));
410 newnfs_copyincred(cred, &nop->nfso_cred);
412 TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
413 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp,
415 dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
416 nfsstatsv1.cllocalopens++;
418 LIST_INSERT_HEAD(NFSCLOPENHASH(clp, fhp, fhlen),
420 nfsstatsv1.clopens++;
422 LIST_INSERT_HEAD(&owp->nfsow_open, nop, nfso_list);
434 * Called to find/add a delegation to a client.
437 nfscl_deleg(mount_t mp, struct nfsclclient *clp, u_int8_t *nfhp,
438 int fhlen, struct ucred *cred, NFSPROC_T *p, struct nfscldeleg **dpp)
440 struct nfscldeleg *dp = *dpp, *tdp;
443 * First, if we have received a Read delegation for a file on a
444 * read/write file system, just return it, because they aren't
447 if (mp != NULL && dp != NULL && !NFSMNT_RDONLY(mp) &&
448 (dp->nfsdl_flags & NFSCLDL_READ)) {
449 (void) nfscl_trydelegreturn(dp, cred, VFSTONFS(mp), p);
450 free(dp, M_NFSCLDELEG);
455 /* Look for the correct deleg, based upon FH */
457 tdp = nfscl_finddeleg(clp, nfhp, fhlen);
461 return (NFSERR_BADSTATEID);
464 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list);
465 LIST_INSERT_HEAD(NFSCLDELEGHASH(clp, nfhp, fhlen), dp,
467 dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
468 nfsstatsv1.cldelegates++;
472 * Delegation already exists, what do we do if a new one??
475 printf("Deleg already exists!\n");
476 free(dp, M_NFSCLDELEG);
487 * Find a delegation for this file handle. Return NULL upon failure.
489 static struct nfscldeleg *
490 nfscl_finddeleg(struct nfsclclient *clp, u_int8_t *fhp, int fhlen)
492 struct nfscldeleg *dp;
494 LIST_FOREACH(dp, NFSCLDELEGHASH(clp, fhp, fhlen), nfsdl_hash) {
495 if (dp->nfsdl_fhlen == fhlen &&
496 !NFSBCMP(dp->nfsdl_fh, fhp, fhlen))
503 * Get a stateid for an I/O operation. First, look for an open and iff
504 * found, return either a lockowner stateid or the open stateid.
505 * If no Open is found, just return error and the special stateid of all zeros.
508 nfscl_getstateid(vnode_t vp, u_int8_t *nfhp, int fhlen, u_int32_t mode,
509 int fords, struct ucred *cred, NFSPROC_T *p, nfsv4stateid_t *stateidp,
512 struct nfsclclient *clp;
513 struct nfsclopen *op = NULL, *top;
514 struct nfsclopenhash *oph;
515 struct nfscllockowner *lp;
516 struct nfscldeleg *dp;
518 struct nfsmount *nmp;
519 u_int8_t own[NFSV4CL_LOCKNAMELEN];
525 * Initially, just set the special stateid of all zeros.
526 * (Don't do this for a DS, since the special stateid can't be used.)
530 stateidp->other[0] = 0;
531 stateidp->other[1] = 0;
532 stateidp->other[2] = 0;
534 if (vnode_vtype(vp) != VREG)
537 nmp = VFSTONFS(vp->v_mount);
539 clp = nfscl_findcl(nmp);
546 * Wait for recovery to complete.
548 while ((clp->nfsc_flags & NFSCLFLAGS_RECVRINPROG))
549 (void) nfsmsleep(&clp->nfsc_flags, NFSCLSTATEMUTEXPTR,
550 PZERO, "nfsrecvr", NULL);
553 * First, look for a delegation.
555 LIST_FOREACH(dp, NFSCLDELEGHASH(clp, nfhp, fhlen), nfsdl_hash) {
556 if (dp->nfsdl_fhlen == fhlen &&
557 !NFSBCMP(nfhp, dp->nfsdl_fh, fhlen)) {
558 if (!(mode & NFSV4OPEN_ACCESSWRITE) ||
559 (dp->nfsdl_flags & NFSCLDL_WRITE)) {
560 stateidp->seqid = dp->nfsdl_stateid.seqid;
561 stateidp->other[0] = dp->nfsdl_stateid.other[0];
562 stateidp->other[1] = dp->nfsdl_stateid.other[1];
563 stateidp->other[2] = dp->nfsdl_stateid.other[2];
564 if (!(np->n_flag & NDELEGRECALL)) {
565 TAILQ_REMOVE(&clp->nfsc_deleg, dp,
567 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp,
569 dp->nfsdl_timestamp = NFSD_MONOSEC +
571 dp->nfsdl_rwlock.nfslock_usecnt++;
572 *lckpp = (void *)&dp->nfsdl_rwlock;
583 * If p != NULL, we want to search the parentage tree
584 * for a matching OpenOwner and use that.
586 if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount)))
587 nfscl_filllockowner(NULL, own, F_POSIX);
589 nfscl_filllockowner(p->td_proc, own, F_POSIX);
591 error = nfscl_getopen(NULL, clp->nfsc_openhash, nfhp, fhlen,
592 own, own, mode, &lp, &op);
593 if (error == 0 && lp != NULL && fords == 0) {
594 /* Don't return a lock stateid for a DS. */
596 lp->nfsl_stateid.seqid;
598 lp->nfsl_stateid.other[0];
600 lp->nfsl_stateid.other[1];
602 lp->nfsl_stateid.other[2];
608 /* If not found, just look for any OpenOwner that will work. */
611 oph = NFSCLOPENHASH(clp, nfhp, fhlen);
612 LIST_FOREACH(op, oph, nfso_hash) {
613 if (op->nfso_fhlen == fhlen &&
614 !NFSBCMP(op->nfso_fh, nfhp, fhlen)) {
615 if (top == NULL && (op->nfso_mode &
616 NFSV4OPEN_ACCESSWRITE) != 0 &&
617 (mode & NFSV4OPEN_ACCESSREAD) != 0)
619 if ((mode & op->nfso_mode) == mode) {
620 /* LRU order the hash list. */
621 LIST_REMOVE(op, nfso_hash);
622 LIST_INSERT_HEAD(oph, op, nfso_hash);
629 NFSCL_DEBUG(2, "openmode top=%p\n", top);
630 if (top == NULL || NFSHASOPENMODE(nmp)) {
637 * For read aheads or write behinds, use the open cred.
638 * A read ahead or write behind is indicated by p == NULL.
641 newnfs_copycred(&op->nfso_cred, cred);
645 * No lock stateid, so return the open stateid.
647 stateidp->seqid = op->nfso_stateid.seqid;
648 stateidp->other[0] = op->nfso_stateid.other[0];
649 stateidp->other[1] = op->nfso_stateid.other[1];
650 stateidp->other[2] = op->nfso_stateid.other[2];
656 * Search for a matching file, mode and, optionally, lockowner.
659 nfscl_getopen(struct nfsclownerhead *ohp, struct nfsclopenhash *ohashp,
660 u_int8_t *nfhp, int fhlen, u_int8_t *openown, u_int8_t *lockown,
661 u_int32_t mode, struct nfscllockowner **lpp, struct nfsclopen **opp)
663 struct nfsclowner *owp;
664 struct nfsclopen *op, *rop, *rop2;
665 struct nfsclopenhash *oph;
668 KASSERT(ohp == NULL || ohashp == NULL, ("nfscl_getopen: "
669 "only one of ohp and ohashp can be set"));
673 * rop will be set to the open to be returned. There are three
674 * variants of this, all for an open of the correct file:
675 * 1 - A match of lockown.
676 * 2 - A match of the openown, when no lockown match exists.
677 * 3 - A match for any open, if no openown or lockown match exists.
678 * Looking for #2 over #3 probably isn't necessary, but since
679 * RFC3530 is vague w.r.t. the relationship between openowners and
680 * lockowners, I think this is the safer way to go.
685 /* Search the client list */
686 if (ohashp == NULL) {
687 /* Search the local opens on the delegation. */
688 LIST_FOREACH(owp, ohp, nfsow_list) {
689 /* and look for the correct open */
690 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
691 if (op->nfso_fhlen == fhlen &&
692 !NFSBCMP(op->nfso_fh, nfhp, fhlen)
693 && (op->nfso_mode & mode) == mode)
694 keep_looping = nfscl_checkown(owp, op, openown,
695 lockown, lpp, &rop, &rop2);
703 /* Search for matching opens on the hash list. */
704 oph = &ohashp[NFSCLOPENHASHFUNC(nfhp, fhlen)];
705 LIST_FOREACH(op, oph, nfso_hash) {
706 if (op->nfso_fhlen == fhlen &&
707 !NFSBCMP(op->nfso_fh, nfhp, fhlen)
708 && (op->nfso_mode & mode) == mode)
709 keep_looping = nfscl_checkown(op->nfso_own, op,
710 openown, lockown, lpp, &rop, &rop2);
712 /* LRU order the hash list. */
713 LIST_REMOVE(op, nfso_hash);
714 LIST_INSERT_HEAD(oph, op, nfso_hash);
727 /* Check for an owner match. */
729 nfscl_checkown(struct nfsclowner *owp, struct nfsclopen *op, uint8_t *openown,
730 uint8_t *lockown, struct nfscllockowner **lpp, struct nfsclopen **ropp,
731 struct nfsclopen **ropp2)
733 struct nfscllockowner *lp;
738 /* Now look for a matching lockowner. */
739 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
740 if (!NFSBCMP(lp->nfsl_owner, lockown,
741 NFSV4CL_LOCKNAMELEN)) {
748 if (*ropp == NULL && !NFSBCMP(owp->nfsow_owner, openown,
749 NFSV4CL_LOCKNAMELEN)) {
752 keep_looping = false;
756 return (keep_looping);
760 * Release use of an open owner. Called when open operations are done
761 * with the open owner.
764 nfscl_ownerrelease(struct nfsmount *nmp, struct nfsclowner *owp,
765 __unused int error, __unused int candelete, int unlocked)
772 if (NFSHASONEOPENOWN(nmp))
773 nfsv4_relref(&owp->nfsow_rwlock);
775 nfscl_lockunlock(&owp->nfsow_rwlock);
777 nfscl_clrelease(owp->nfsow_clp);
782 * Release use of an open structure under an open owner.
785 nfscl_openrelease(struct nfsmount *nmp, struct nfsclopen *op, int error,
788 struct nfsclclient *clp;
789 struct nfsclowner *owp;
795 if (NFSHASONEOPENOWN(nmp))
796 nfsv4_relref(&owp->nfsow_rwlock);
798 nfscl_lockunlock(&owp->nfsow_rwlock);
799 clp = owp->nfsow_clp;
800 if (error && candelete && op->nfso_opencnt == 0)
801 nfscl_freeopen(op, 0);
802 nfscl_clrelease(clp);
807 * Called to get a clientid structure. It will optionally lock the
808 * client data structures to do the SetClientId/SetClientId_confirm,
809 * but will release that lock and return the clientid with a reference
811 * If the "cred" argument is NULL, a new clientid should not be created.
812 * If the "p" argument is NULL, a SetClientID/SetClientIDConfirm cannot
814 * The start_renewthread argument tells nfscl_getcl() to start a renew
815 * thread if this creates a new clp.
816 * It always clpp with a reference count on it, unless returning an error.
819 nfscl_getcl(struct mount *mp, struct ucred *cred, NFSPROC_T *p,
820 int start_renewthread, struct nfsclclient **clpp)
822 struct nfsclclient *clp;
823 struct nfsclclient *newclp = NULL;
824 struct nfsmount *nmp;
825 char uuid[HOSTUUIDLEN];
826 int igotlock = 0, error, trystalecnt, clidinusedelay, i;
831 getcredhostuuid(cred, uuid, sizeof uuid);
832 idlen = strlen(uuid);
834 idlen += sizeof (u_int64_t);
836 idlen += sizeof (u_int64_t) + 16; /* 16 random bytes */
838 sizeof (struct nfsclclient) + idlen - 1, M_NFSCLCLIENT,
843 * If a forced dismount is already in progress, don't
844 * allocate a new clientid and get out now. For the case where
845 * clp != NULL, this is a harmless optimization.
847 if (NFSCL_FORCEDISM(mp)) {
850 free(newclp, M_NFSCLCLIENT);
855 if (newclp == NULL) {
860 clp->nfsc_idlen = idlen;
861 LIST_INIT(&clp->nfsc_owner);
862 TAILQ_INIT(&clp->nfsc_deleg);
863 TAILQ_INIT(&clp->nfsc_layout);
864 LIST_INIT(&clp->nfsc_devinfo);
865 for (i = 0; i < NFSCLDELEGHASHSIZE; i++)
866 LIST_INIT(&clp->nfsc_deleghash[i]);
867 for (i = 0; i < NFSCLOPENHASHSIZE; i++)
868 LIST_INIT(&clp->nfsc_openhash[i]);
869 for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++)
870 LIST_INIT(&clp->nfsc_layouthash[i]);
871 clp->nfsc_flags = NFSCLFLAGS_INITED;
872 clp->nfsc_clientidrev = 1;
873 clp->nfsc_cbident = nfscl_nextcbident();
874 nfscl_fillclid(nmp->nm_clval, uuid, clp->nfsc_id,
876 LIST_INSERT_HEAD(&nfsclhead, clp, nfsc_list);
880 if (start_renewthread != 0)
881 nfscl_start_renewthread(clp);
885 free(newclp, M_NFSCLCLIENT);
888 while ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0 && !igotlock &&
889 !NFSCL_FORCEDISM(mp))
890 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
891 NFSCLSTATEMUTEXPTR, mp);
894 * Call nfsv4_lock() with "iwantlock == 0" so that it will
895 * wait for a pending exclusive lock request. This gives the
896 * exclusive lock request priority over this shared lock
898 * An exclusive lock on nfsc_lock is used mainly for server
901 nfsv4_lock(&clp->nfsc_lock, 0, NULL, NFSCLSTATEMUTEXPTR, mp);
902 nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
904 if (igotlock == 0 && NFSCL_FORCEDISM(mp)) {
906 * Both nfsv4_lock() and nfsv4_getref() know to check
907 * for NFSCL_FORCEDISM() and return without sleeping to
908 * wait for the exclusive lock to be released, since it
909 * might be held by nfscl_umount() and we need to get out
910 * now for that case and not wait until nfscl_umount()
919 * If it needs a clientid, do the setclientid now.
921 if ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0) {
923 panic("nfscl_clget");
924 if (p == NULL || cred == NULL) {
926 nfsv4_unlock(&clp->nfsc_lock, 0);
931 * If RFC3530 Sec. 14.2.33 is taken literally,
932 * NFSERR_CLIDINUSE will be returned persistently for the
933 * case where a new mount of the same file system is using
934 * a different principal. In practice, NFSERR_CLIDINUSE is
935 * only returned when there is outstanding unexpired state
936 * on the clientid. As such, try for twice the lease
937 * interval, if we know what that is. Otherwise, make a
939 * The case of returning NFSERR_STALECLIENTID is far less
940 * likely, but might occur if there is a significant delay
941 * between doing the SetClientID and SetClientIDConfirm Ops,
942 * such that the server throws away the clientid before
943 * receiving the SetClientIDConfirm.
945 if (clp->nfsc_renew > 0)
946 clidinusedelay = NFSCL_LEASE(clp->nfsc_renew) * 2;
948 clidinusedelay = 120;
951 error = nfsrpc_setclient(nmp, clp, 0, NULL, cred, p);
952 if (error == NFSERR_STALECLIENTID ||
953 error == NFSERR_STALEDONTRECOVER ||
954 error == NFSERR_BADSESSION ||
955 error == NFSERR_CLIDINUSE) {
956 (void) nfs_catnap(PZERO, error, "nfs_setcl");
958 } while (((error == NFSERR_STALECLIENTID ||
959 error == NFSERR_BADSESSION ||
960 error == NFSERR_STALEDONTRECOVER) && --trystalecnt > 0) ||
961 (error == NFSERR_CLIDINUSE && --clidinusedelay > 0));
964 nfsv4_unlock(&clp->nfsc_lock, 0);
968 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
972 nfsv4_unlock(&clp->nfsc_lock, 1);
981 * Get a reference to a clientid and return it, if valid.
984 nfscl_findcl(struct nfsmount *nmp)
986 struct nfsclclient *clp;
989 if (clp == NULL || !(clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID))
995 * Release the clientid structure. It may be locked or reference counted.
998 nfscl_clrelease(struct nfsclclient *clp)
1001 if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK)
1002 nfsv4_unlock(&clp->nfsc_lock, 0);
1004 nfsv4_relref(&clp->nfsc_lock);
1008 * External call for nfscl_clrelease.
1011 nfscl_clientrelease(struct nfsclclient *clp)
1015 if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK)
1016 nfsv4_unlock(&clp->nfsc_lock, 0);
1018 nfsv4_relref(&clp->nfsc_lock);
1023 * Called when wanting to lock a byte region.
1026 nfscl_getbytelock(vnode_t vp, u_int64_t off, u_int64_t len,
1027 short type, struct ucred *cred, NFSPROC_T *p, struct nfsclclient *rclp,
1028 int recovery, void *id, int flags, u_int8_t *rownp, u_int8_t *ropenownp,
1029 struct nfscllockowner **lpp, int *newonep, int *donelocallyp)
1031 struct nfscllockowner *lp;
1032 struct nfsclopen *op;
1033 struct nfsclclient *clp;
1034 struct nfscllockowner *nlp;
1035 struct nfscllock *nlop, *otherlop;
1036 struct nfscldeleg *dp = NULL, *ldp = NULL;
1037 struct nfscllockownerhead *lhp = NULL;
1039 u_int8_t own[NFSV4CL_LOCKNAMELEN], *ownp, openown[NFSV4CL_LOCKNAMELEN];
1041 int error = 0, ret, donelocally = 0;
1044 /* For Lock Ops, the open mode doesn't matter, so use 0 to match any. */
1053 * Might need these, so MALLOC them now, to
1054 * avoid a tsleep() in MALLOC later.
1057 sizeof (struct nfscllockowner), M_NFSCLLOCKOWNER, M_WAITOK);
1059 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1061 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1062 nlop->nfslo_type = type;
1063 nlop->nfslo_first = off;
1064 if (len == NFS64BITSSET) {
1065 nlop->nfslo_end = NFS64BITSSET;
1067 nlop->nfslo_end = off + len;
1068 if (nlop->nfslo_end <= nlop->nfslo_first)
1069 error = NFSERR_INVAL;
1076 error = nfscl_getcl(vp->v_mount, cred, p, 1, &clp);
1079 free(nlp, M_NFSCLLOCKOWNER);
1080 free(otherlop, M_NFSCLLOCK);
1081 free(nlop, M_NFSCLLOCK);
1088 openownp = ropenownp;
1090 nfscl_filllockowner(id, own, flags);
1092 if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount)))
1093 nfscl_filllockowner(NULL, openown, F_POSIX);
1095 nfscl_filllockowner(p->td_proc, openown, F_POSIX);
1101 * First, search for a delegation. If one exists for this file,
1102 * the lock can be done locally against it, so long as there
1103 * isn't a local lock conflict.
1105 ldp = dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
1106 np->n_fhp->nfh_len);
1107 /* Just sanity check for correct type of delegation */
1108 if (dp != NULL && ((dp->nfsdl_flags &
1109 (NFSCLDL_RECALL | NFSCLDL_DELEGRET)) != 0 ||
1111 (dp->nfsdl_flags & NFSCLDL_WRITE) == 0)))
1115 /* Now, find an open and maybe a lockowner. */
1116 ret = nfscl_getopen(&dp->nfsdl_owner, NULL, np->n_fhp->nfh_fh,
1117 np->n_fhp->nfh_len, openownp, ownp, mode, NULL, &op);
1119 ret = nfscl_getopen(NULL, clp->nfsc_openhash,
1120 np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp,
1121 ownp, mode, NULL, &op);
1123 lhp = &dp->nfsdl_lock;
1124 TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
1125 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list);
1126 dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
1134 * Get the related Open and maybe lockowner.
1136 error = nfscl_getopen(NULL, clp->nfsc_openhash,
1137 np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp,
1138 ownp, mode, &lp, &op);
1140 lhp = &op->nfso_lock;
1142 if (!error && !recovery)
1143 error = nfscl_localconflict(clp, np->n_fhp->nfh_fh,
1144 np->n_fhp->nfh_len, nlop, ownp, ldp, NULL);
1147 nfscl_clrelease(clp);
1150 free(nlp, M_NFSCLLOCKOWNER);
1151 free(otherlop, M_NFSCLLOCK);
1152 free(nlop, M_NFSCLLOCK);
1157 * Ok, see if a lockowner exists and create one, as required.
1160 LIST_FOREACH(lp, lhp, nfsl_list) {
1161 if (!NFSBCMP(lp->nfsl_owner, ownp, NFSV4CL_LOCKNAMELEN))
1165 NFSBCOPY(ownp, nlp->nfsl_owner, NFSV4CL_LOCKNAMELEN);
1167 NFSBCOPY(ropenownp, nlp->nfsl_openowner,
1168 NFSV4CL_LOCKNAMELEN);
1170 NFSBCOPY(op->nfso_own->nfsow_owner, nlp->nfsl_openowner,
1171 NFSV4CL_LOCKNAMELEN);
1172 nlp->nfsl_seqid = 0;
1173 nlp->nfsl_lockflags = flags;
1174 nlp->nfsl_inprog = NULL;
1175 nfscl_lockinit(&nlp->nfsl_rwlock);
1176 LIST_INIT(&nlp->nfsl_lock);
1178 nlp->nfsl_open = NULL;
1179 nfsstatsv1.cllocallockowners++;
1181 nlp->nfsl_open = op;
1182 nfsstatsv1.cllockowners++;
1184 LIST_INSERT_HEAD(lhp, nlp, nfsl_list);
1191 * Now, update the byte ranges for locks.
1193 ret = nfscl_updatelock(lp, &nlop, &otherlop, donelocally);
1199 nfscl_clrelease(clp);
1202 * Serial modifications on the lock owner for multiple threads
1203 * for the same process using a read/write lock.
1206 nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR);
1212 free(nlp, M_NFSCLLOCKOWNER);
1214 free(nlop, M_NFSCLLOCK);
1216 free(otherlop, M_NFSCLLOCK);
1223 * Called to unlock a byte range, for LockU.
1226 nfscl_relbytelock(vnode_t vp, u_int64_t off, u_int64_t len,
1227 __unused struct ucred *cred, NFSPROC_T *p, int callcnt,
1228 struct nfsclclient *clp, void *id, int flags,
1229 struct nfscllockowner **lpp, int *dorpcp)
1231 struct nfscllockowner *lp;
1232 struct nfsclowner *owp;
1233 struct nfsclopen *op;
1234 struct nfscllock *nlop, *other_lop = NULL;
1235 struct nfscldeleg *dp;
1237 u_int8_t own[NFSV4CL_LOCKNAMELEN];
1245 * Might need these, so MALLOC them now, to
1246 * avoid a tsleep() in MALLOC later.
1249 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1250 nlop->nfslo_type = F_UNLCK;
1251 nlop->nfslo_first = off;
1252 if (len == NFS64BITSSET) {
1253 nlop->nfslo_end = NFS64BITSSET;
1255 nlop->nfslo_end = off + len;
1256 if (nlop->nfslo_end <= nlop->nfslo_first) {
1257 free(nlop, M_NFSCLLOCK);
1258 return (NFSERR_INVAL);
1263 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1266 nfscl_filllockowner(id, own, flags);
1270 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
1271 np->n_fhp->nfh_len);
1274 * First, unlock any local regions on a delegation.
1277 /* Look for this lockowner. */
1278 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
1279 if (!NFSBCMP(lp->nfsl_owner, own,
1280 NFSV4CL_LOCKNAMELEN))
1284 /* Use other_lop, so nlop is still available */
1285 (void)nfscl_updatelock(lp, &other_lop, NULL, 1);
1289 * Now, find a matching open/lockowner that hasn't already been done,
1290 * as marked by nfsl_inprog.
1294 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
1295 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1296 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1297 !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1298 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1299 if (lp->nfsl_inprog == NULL &&
1300 !NFSBCMP(lp->nfsl_owner, own,
1301 NFSV4CL_LOCKNAMELEN)) {
1315 ret = nfscl_updatelock(lp, &nlop, NULL, 0);
1319 * Serial modifications on the lock owner for multiple
1320 * threads for the same process using a read/write lock.
1322 lp->nfsl_inprog = p;
1323 nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR);
1328 free(nlop, M_NFSCLLOCK);
1330 free(other_lop, M_NFSCLLOCK);
1335 * Release all lockowners marked in progess for this process and file.
1338 nfscl_releasealllocks(struct nfsclclient *clp, vnode_t vp, NFSPROC_T *p,
1339 void *id, int flags)
1341 struct nfsclowner *owp;
1342 struct nfsclopen *op;
1343 struct nfscllockowner *lp;
1345 u_int8_t own[NFSV4CL_LOCKNAMELEN];
1348 nfscl_filllockowner(id, own, flags);
1350 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
1351 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1352 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1353 !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1354 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1355 if (lp->nfsl_inprog == p &&
1356 !NFSBCMP(lp->nfsl_owner, own,
1357 NFSV4CL_LOCKNAMELEN)) {
1358 lp->nfsl_inprog = NULL;
1359 nfscl_lockunlock(&lp->nfsl_rwlock);
1365 nfscl_clrelease(clp);
1370 * Called to find out if any bytes within the byte range specified are
1371 * write locked by the calling process. Used to determine if flushing
1372 * is required before a LockU.
1373 * If in doubt, return 1, so the flush will occur.
1376 nfscl_checkwritelocked(vnode_t vp, struct flock *fl,
1377 struct ucred *cred, NFSPROC_T *p, void *id, int flags)
1379 struct nfsclowner *owp;
1380 struct nfscllockowner *lp;
1381 struct nfsclopen *op;
1382 struct nfsclclient *clp;
1383 struct nfscllock *lop;
1384 struct nfscldeleg *dp;
1387 u_int8_t own[NFSV4CL_LOCKNAMELEN];
1391 switch (fl->l_whence) {
1395 * Caller is responsible for adding any necessary offset
1396 * when SEEK_CUR is used.
1401 off = np->n_size + fl->l_start;
1406 if (fl->l_len != 0) {
1407 end = off + fl->l_len;
1414 error = nfscl_getcl(vp->v_mount, cred, p, 1, &clp);
1417 nfscl_filllockowner(id, own, flags);
1421 * First check the delegation locks.
1423 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
1425 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
1426 if (!NFSBCMP(lp->nfsl_owner, own,
1427 NFSV4CL_LOCKNAMELEN))
1431 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
1432 if (lop->nfslo_first >= end)
1434 if (lop->nfslo_end <= off)
1436 if (lop->nfslo_type == F_WRLCK) {
1437 nfscl_clrelease(clp);
1446 * Now, check state against the server.
1448 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
1449 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1450 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1451 !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1452 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1453 if (!NFSBCMP(lp->nfsl_owner, own,
1454 NFSV4CL_LOCKNAMELEN))
1458 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
1459 if (lop->nfslo_first >= end)
1461 if (lop->nfslo_end <= off)
1463 if (lop->nfslo_type == F_WRLCK) {
1464 nfscl_clrelease(clp);
1473 nfscl_clrelease(clp);
1479 * Release a byte range lock owner structure.
1482 nfscl_lockrelease(struct nfscllockowner *lp, int error, int candelete)
1484 struct nfsclclient *clp;
1489 clp = lp->nfsl_open->nfso_own->nfsow_clp;
1490 if (error != 0 && candelete &&
1491 (lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED) == 0)
1492 nfscl_freelockowner(lp, 0);
1494 nfscl_lockunlock(&lp->nfsl_rwlock);
1495 nfscl_clrelease(clp);
1500 * Free up an open structure and any associated byte range lock structures.
1503 nfscl_freeopen(struct nfsclopen *op, int local)
1506 LIST_REMOVE(op, nfso_list);
1507 if (op->nfso_hash.le_prev != NULL)
1508 LIST_REMOVE(op, nfso_hash);
1509 nfscl_freealllocks(&op->nfso_lock, local);
1510 free(op, M_NFSCLOPEN);
1512 nfsstatsv1.cllocalopens--;
1514 nfsstatsv1.clopens--;
1518 * Free up all lock owners and associated locks.
1521 nfscl_freealllocks(struct nfscllockownerhead *lhp, int local)
1523 struct nfscllockowner *lp, *nlp;
1525 LIST_FOREACH_SAFE(lp, lhp, nfsl_list, nlp) {
1526 if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED))
1528 nfscl_freelockowner(lp, local);
1533 * Called for an Open when NFSERR_EXPIRED is received from the server.
1534 * If there are no byte range locks nor a Share Deny lost, try to do a
1535 * fresh Open. Otherwise, free the open.
1538 nfscl_expireopen(struct nfsclclient *clp, struct nfsclopen *op,
1539 struct nfsmount *nmp, struct ucred *cred, NFSPROC_T *p)
1541 struct nfscllockowner *lp;
1542 struct nfscldeleg *dp;
1543 int mustdelete = 0, error;
1546 * Look for any byte range lock(s).
1548 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1549 if (!LIST_EMPTY(&lp->nfsl_lock)) {
1556 * If no byte range lock(s) nor a Share deny, try to re-open.
1558 if (!mustdelete && (op->nfso_mode & NFSLCK_DENYBITS) == 0) {
1559 newnfs_copycred(&op->nfso_cred, cred);
1561 error = nfsrpc_reopen(nmp, op->nfso_fh,
1562 op->nfso_fhlen, op->nfso_mode, op, &dp, cred, p);
1566 free(dp, M_NFSCLDELEG);
1571 nfscl_deleg(nmp->nm_mountp, clp, op->nfso_fh,
1572 op->nfso_fhlen, cred, p, &dp);
1576 * If a byte range lock or Share deny or couldn't re-open, free it.
1579 nfscl_freeopen(op, 0);
1580 return (mustdelete);
1584 * Free up an open owner structure.
1587 nfscl_freeopenowner(struct nfsclowner *owp, int local)
1590 LIST_REMOVE(owp, nfsow_list);
1591 free(owp, M_NFSCLOWNER);
1593 nfsstatsv1.cllocalopenowners--;
1595 nfsstatsv1.clopenowners--;
1599 * Free up a byte range lock owner structure.
1602 nfscl_freelockowner(struct nfscllockowner *lp, int local)
1604 struct nfscllock *lop, *nlop;
1606 LIST_REMOVE(lp, nfsl_list);
1607 LIST_FOREACH_SAFE(lop, &lp->nfsl_lock, nfslo_list, nlop) {
1608 nfscl_freelock(lop, local);
1610 free(lp, M_NFSCLLOCKOWNER);
1612 nfsstatsv1.cllocallockowners--;
1614 nfsstatsv1.cllockowners--;
1618 * Free up a byte range lock structure.
1621 nfscl_freelock(struct nfscllock *lop, int local)
1624 LIST_REMOVE(lop, nfslo_list);
1625 free(lop, M_NFSCLLOCK);
1627 nfsstatsv1.cllocallocks--;
1629 nfsstatsv1.cllocks--;
1633 * Clean out the state related to a delegation.
1636 nfscl_cleandeleg(struct nfscldeleg *dp)
1638 struct nfsclowner *owp, *nowp;
1639 struct nfsclopen *op;
1641 LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) {
1642 op = LIST_FIRST(&owp->nfsow_open);
1644 if (LIST_NEXT(op, nfso_list) != NULL)
1645 panic("nfscleandel");
1646 nfscl_freeopen(op, 1);
1648 nfscl_freeopenowner(owp, 1);
1650 nfscl_freealllocks(&dp->nfsdl_lock, 1);
1654 * Free a delegation.
1657 nfscl_freedeleg(struct nfscldeleghead *hdp, struct nfscldeleg *dp, bool freeit)
1660 TAILQ_REMOVE(hdp, dp, nfsdl_list);
1661 LIST_REMOVE(dp, nfsdl_hash);
1663 free(dp, M_NFSCLDELEG);
1664 nfsstatsv1.cldelegates--;
1669 * Free up all state related to this client structure.
1672 nfscl_cleanclient(struct nfsclclient *clp)
1674 struct nfsclowner *owp, *nowp;
1675 struct nfsclopen *op, *nop;
1676 struct nfscllayout *lyp, *nlyp;
1677 struct nfscldevinfo *dip, *ndip;
1679 TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp)
1680 nfscl_freelayout(lyp);
1682 LIST_FOREACH_SAFE(dip, &clp->nfsc_devinfo, nfsdi_list, ndip)
1683 nfscl_freedevinfo(dip);
1685 /* Now, all the OpenOwners, etc. */
1686 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1687 LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
1688 nfscl_freeopen(op, 0);
1690 nfscl_freeopenowner(owp, 0);
1695 * Called when an NFSERR_EXPIRED is received from the server.
1698 nfscl_expireclient(struct nfsclclient *clp, struct nfsmount *nmp,
1699 struct ucred *cred, NFSPROC_T *p)
1701 struct nfsclowner *owp, *nowp, *towp;
1702 struct nfsclopen *op, *nop, *top;
1703 struct nfscldeleg *dp, *ndp;
1704 int ret, printed = 0;
1707 * First, merge locally issued Opens into the list for the server.
1709 dp = TAILQ_FIRST(&clp->nfsc_deleg);
1710 while (dp != NULL) {
1711 ndp = TAILQ_NEXT(dp, nfsdl_list);
1712 owp = LIST_FIRST(&dp->nfsdl_owner);
1713 while (owp != NULL) {
1714 nowp = LIST_NEXT(owp, nfsow_list);
1715 op = LIST_FIRST(&owp->nfsow_open);
1717 if (LIST_NEXT(op, nfso_list) != NULL)
1719 LIST_FOREACH(towp, &clp->nfsc_owner, nfsow_list) {
1720 if (!NFSBCMP(towp->nfsow_owner, owp->nfsow_owner,
1721 NFSV4CL_LOCKNAMELEN))
1725 /* Merge opens in */
1726 LIST_FOREACH(top, &towp->nfsow_open, nfso_list) {
1727 if (top->nfso_fhlen == op->nfso_fhlen &&
1728 !NFSBCMP(top->nfso_fh, op->nfso_fh,
1730 top->nfso_mode |= op->nfso_mode;
1731 top->nfso_opencnt += op->nfso_opencnt;
1736 /* Just add the open to the owner list */
1737 LIST_REMOVE(op, nfso_list);
1738 op->nfso_own = towp;
1739 LIST_INSERT_HEAD(&towp->nfsow_open, op, nfso_list);
1740 LIST_INSERT_HEAD(NFSCLOPENHASH(clp, op->nfso_fh,
1741 op->nfso_fhlen), op, nfso_hash);
1742 nfsstatsv1.cllocalopens--;
1743 nfsstatsv1.clopens++;
1746 /* Just add the openowner to the client list */
1747 LIST_REMOVE(owp, nfsow_list);
1748 owp->nfsow_clp = clp;
1749 LIST_INSERT_HEAD(&clp->nfsc_owner, owp, nfsow_list);
1750 LIST_INSERT_HEAD(NFSCLOPENHASH(clp, op->nfso_fh,
1751 op->nfso_fhlen), op, nfso_hash);
1752 nfsstatsv1.cllocalopenowners--;
1753 nfsstatsv1.clopenowners++;
1754 nfsstatsv1.cllocalopens--;
1755 nfsstatsv1.clopens++;
1760 if (!printed && !LIST_EMPTY(&dp->nfsdl_lock)) {
1762 printf("nfsv4 expired locks lost\n");
1764 nfscl_cleandeleg(dp);
1765 nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
1768 if (!TAILQ_EMPTY(&clp->nfsc_deleg))
1772 * Now, try and reopen against the server.
1774 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1775 owp->nfsow_seqid = 0;
1776 LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
1777 ret = nfscl_expireopen(clp, op, nmp, cred, p);
1778 if (ret && !printed) {
1780 printf("nfsv4 expired locks lost\n");
1783 if (LIST_EMPTY(&owp->nfsow_open))
1784 nfscl_freeopenowner(owp, 0);
1789 * This function must be called after the process represented by "own" has
1790 * exited. Must be called with CLSTATE lock held.
1793 nfscl_cleanup_common(struct nfsclclient *clp, u_int8_t *own)
1795 struct nfsclowner *owp, *nowp;
1796 struct nfscllockowner *lp, *nlp;
1797 struct nfscldeleg *dp;
1799 /* First, get rid of local locks on delegations. */
1800 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
1801 LIST_FOREACH_SAFE(lp, &dp->nfsdl_lock, nfsl_list, nlp) {
1802 if (!NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
1803 if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED))
1805 nfscl_freelockowner(lp, 1);
1809 owp = LIST_FIRST(&clp->nfsc_owner);
1810 while (owp != NULL) {
1811 nowp = LIST_NEXT(owp, nfsow_list);
1812 if (!NFSBCMP(owp->nfsow_owner, own,
1813 NFSV4CL_LOCKNAMELEN)) {
1815 * If there are children that haven't closed the
1816 * file descriptors yet, the opens will still be
1817 * here. For that case, let the renew thread clear
1818 * out the OpenOwner later.
1820 if (LIST_EMPTY(&owp->nfsow_open))
1821 nfscl_freeopenowner(owp, 0);
1823 owp->nfsow_defunct = 1;
1830 * Find open/lock owners for processes that have exited.
1833 nfscl_cleanupkext(struct nfsclclient *clp, struct nfscllockownerfhhead *lhp)
1835 struct nfsclowner *owp, *nowp;
1836 struct nfsclopen *op;
1837 struct nfscllockowner *lp, *nlp;
1838 struct nfscldeleg *dp;
1841 * All the pidhash locks must be acquired, since they are sx locks
1842 * and must be acquired before the mutexes. The pid(s) that will
1843 * be used aren't known yet, so all the locks need to be acquired.
1844 * Fortunately, this function is only performed once/sec.
1848 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1849 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1850 LIST_FOREACH_SAFE(lp, &op->nfso_lock, nfsl_list, nlp) {
1851 if (LIST_EMPTY(&lp->nfsl_lock))
1852 nfscl_emptylockowner(lp, lhp);
1855 if (nfscl_procdoesntexist(owp->nfsow_owner))
1856 nfscl_cleanup_common(clp, owp->nfsow_owner);
1860 * For the single open_owner case, these lock owners need to be
1861 * checked to see if they still exist separately.
1862 * This is because nfscl_procdoesntexist() never returns true for
1863 * the single open_owner so that the above doesn't ever call
1864 * nfscl_cleanup_common().
1866 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
1867 LIST_FOREACH_SAFE(lp, &dp->nfsdl_lock, nfsl_list, nlp) {
1868 if (nfscl_procdoesntexist(lp->nfsl_owner))
1869 nfscl_cleanup_common(clp, lp->nfsl_owner);
1873 pidhash_sunlockall();
1877 * Take the empty lock owner and move it to the local lhp list if the
1878 * associated process no longer exists.
1881 nfscl_emptylockowner(struct nfscllockowner *lp,
1882 struct nfscllockownerfhhead *lhp)
1884 struct nfscllockownerfh *lfhp, *mylfhp;
1885 struct nfscllockowner *nlp;
1888 /* If not a Posix lock owner, just return. */
1889 if ((lp->nfsl_lockflags & F_POSIX) == 0)
1895 * First, search to see if this lock owner is already in the list.
1896 * If it is, then the associated process no longer exists.
1898 SLIST_FOREACH(lfhp, lhp, nfslfh_list) {
1899 if (lfhp->nfslfh_len == lp->nfsl_open->nfso_fhlen &&
1900 !NFSBCMP(lfhp->nfslfh_fh, lp->nfsl_open->nfso_fh,
1903 LIST_FOREACH(nlp, &lfhp->nfslfh_lock, nfsl_list)
1904 if (!NFSBCMP(nlp->nfsl_owner, lp->nfsl_owner,
1905 NFSV4CL_LOCKNAMELEN))
1908 /* If not found, check if process still exists. */
1909 if (fnd_it == 0 && nfscl_procdoesntexist(lp->nfsl_owner) == 0)
1912 /* Move the lock owner over to the local list. */
1913 if (mylfhp == NULL) {
1914 mylfhp = malloc(sizeof(struct nfscllockownerfh), M_TEMP,
1918 mylfhp->nfslfh_len = lp->nfsl_open->nfso_fhlen;
1919 NFSBCOPY(lp->nfsl_open->nfso_fh, mylfhp->nfslfh_fh,
1920 mylfhp->nfslfh_len);
1921 LIST_INIT(&mylfhp->nfslfh_lock);
1922 SLIST_INSERT_HEAD(lhp, mylfhp, nfslfh_list);
1924 LIST_REMOVE(lp, nfsl_list);
1925 LIST_INSERT_HEAD(&mylfhp->nfslfh_lock, lp, nfsl_list);
1928 static int fake_global; /* Used to force visibility of MNTK_UNMOUNTF */
1930 * Called from nfs umount to free up the clientid.
1933 nfscl_umount(struct nfsmount *nmp, NFSPROC_T *p)
1935 struct nfsclclient *clp;
1940 * For the case that matters, this is the thread that set
1941 * MNTK_UNMOUNTF, so it will see it set. The code that follows is
1942 * done to ensure that any thread executing nfscl_getcl() after
1943 * this time, will see MNTK_UNMOUNTF set. nfscl_getcl() uses the
1944 * mutex for NFSLOCKCLSTATE(), so it is "m" for the following
1945 * explanation, courtesy of Alan Cox.
1946 * What follows is a snippet from Alan Cox's email at:
1947 * https://docs.FreeBSD.org/cgi/mid.cgi?BANLkTikR3d65zPHo9==08ZfJ2vmqZucEvw
1949 * 1. Set MNTK_UNMOUNTF
1950 * 2. Acquire a standard FreeBSD mutex "m".
1951 * 3. Update some data structures.
1952 * 4. Release mutex "m".
1954 * Then, other threads that acquire "m" after step 4 has occurred will
1955 * see MNTK_UNMOUNTF as set. But, other threads that beat thread X to
1956 * step 2 may or may not see MNTK_UNMOUNTF as set.
1959 if ((nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) {
1967 if ((clp->nfsc_flags & NFSCLFLAGS_INITED) == 0)
1968 panic("nfscl umount");
1971 * First, handshake with the nfscl renew thread, to terminate
1974 clp->nfsc_flags |= NFSCLFLAGS_UMOUNT;
1975 while (clp->nfsc_flags & NFSCLFLAGS_HASTHREAD)
1976 (void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT,
1980 * Now, get the exclusive lock on the client state, so
1981 * that no uses of the state are still in progress.
1984 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
1985 NFSCLSTATEMUTEXPTR, NULL);
1986 } while (!igotlock);
1990 * Free up all the state. It will expire on the server, but
1991 * maybe we should do a SetClientId/SetClientIdConfirm so
1992 * the server throws it away?
1994 LIST_REMOVE(clp, nfsc_list);
1995 nfscl_delegreturnall(clp, p);
1996 cred = newnfs_getcred();
1997 if (NFSHASNFSV4N(nmp)) {
1998 (void)nfsrpc_destroysession(nmp, clp, cred, p);
1999 (void)nfsrpc_destroyclient(nmp, clp, cred, p);
2001 (void)nfsrpc_setclient(nmp, clp, 0, NULL, cred, p);
2002 nfscl_cleanclient(clp);
2005 free(clp, M_NFSCLCLIENT);
2011 * This function is called when a server replies with NFSERR_STALECLIENTID
2012 * NFSERR_STALESTATEID or NFSERR_BADSESSION. It traverses the clientid lists,
2013 * doing Opens and Locks with reclaim. If these fail, it deletes the
2014 * corresponding state.
2017 nfscl_recover(struct nfsclclient *clp, bool *retokp, struct ucred *cred,
2020 struct nfsclowner *owp, *nowp;
2021 struct nfsclopen *op, *nop;
2022 struct nfscllockowner *lp, *nlp;
2023 struct nfscllock *lop, *nlop;
2024 struct nfscldeleg *dp, *ndp, *tdp;
2025 struct nfsmount *nmp;
2026 struct ucred *tcred;
2027 struct nfsclopenhead extra_open;
2028 struct nfscldeleghead extra_deleg;
2031 u_int32_t delegtype = NFSV4OPEN_DELEGATEWRITE, mode;
2032 int i, igotlock = 0, error, trycnt, firstlock;
2033 struct nfscllayout *lyp, *nlyp;
2037 * First, lock the client structure, so everyone else will
2038 * block when trying to use state.
2041 clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG;
2043 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
2044 NFSCLSTATEMUTEXPTR, NULL);
2045 } while (!igotlock);
2048 nmp = clp->nfsc_nmp;
2050 panic("nfscl recover");
2053 * For now, just get rid of all layouts. There may be a need
2054 * to do LayoutCommit Ops with reclaim == true later.
2056 TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp)
2057 nfscl_freelayout(lyp);
2058 TAILQ_INIT(&clp->nfsc_layout);
2059 for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++)
2060 LIST_INIT(&clp->nfsc_layouthash[i]);
2065 error = nfsrpc_setclient(nmp, clp, 1, retokp, cred, p);
2066 } while ((error == NFSERR_STALECLIENTID ||
2067 error == NFSERR_BADSESSION ||
2068 error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
2071 clp->nfsc_flags &= ~(NFSCLFLAGS_RECOVER |
2072 NFSCLFLAGS_RECVRINPROG);
2073 wakeup(&clp->nfsc_flags);
2074 nfsv4_unlock(&clp->nfsc_lock, 0);
2078 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
2079 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2082 * Mark requests already queued on the server, so that they don't
2083 * initiate another recovery cycle. Any requests already in the
2084 * queue that handle state information will have the old stale
2085 * clientid/stateid and will get a NFSERR_STALESTATEID,
2086 * NFSERR_STALECLIENTID or NFSERR_BADSESSION reply from the server.
2087 * This will be translated to NFSERR_STALEDONTRECOVER when
2088 * R_DONTRECOVER is set.
2091 TAILQ_FOREACH(rep, &nfsd_reqq, r_chain) {
2092 if (rep->r_nmp == nmp)
2093 rep->r_flags |= R_DONTRECOVER;
2098 * If nfsrpc_setclient() returns *retokp == true,
2099 * no more recovery is needed.
2105 * Now, mark all delegations "need reclaim".
2107 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list)
2108 dp->nfsdl_flags |= NFSCLDL_NEEDRECLAIM;
2110 TAILQ_INIT(&extra_deleg);
2111 LIST_INIT(&extra_open);
2113 * Now traverse the state lists, doing Open and Lock Reclaims.
2115 tcred = newnfs_getcred();
2116 recovered_one = false;
2117 owp = LIST_FIRST(&clp->nfsc_owner);
2118 while (owp != NULL) {
2119 nowp = LIST_NEXT(owp, nfsow_list);
2120 owp->nfsow_seqid = 0;
2121 op = LIST_FIRST(&owp->nfsow_open);
2122 while (op != NULL) {
2123 nop = LIST_NEXT(op, nfso_list);
2124 if (error != NFSERR_NOGRACE && error != NFSERR_BADSESSION) {
2125 /* Search for a delegation to reclaim with the open */
2126 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2127 if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
2129 if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
2130 mode = NFSV4OPEN_ACCESSWRITE;
2131 delegtype = NFSV4OPEN_DELEGATEWRITE;
2133 mode = NFSV4OPEN_ACCESSREAD;
2134 delegtype = NFSV4OPEN_DELEGATEREAD;
2136 if ((op->nfso_mode & mode) == mode &&
2137 op->nfso_fhlen == dp->nfsdl_fhlen &&
2138 !NFSBCMP(op->nfso_fh, dp->nfsdl_fh, op->nfso_fhlen))
2143 delegtype = NFSV4OPEN_DELEGATENONE;
2144 newnfs_copycred(&op->nfso_cred, tcred);
2145 error = nfscl_tryopen(nmp, NULL, op->nfso_fh,
2146 op->nfso_fhlen, op->nfso_fh, op->nfso_fhlen,
2147 op->nfso_mode, op, NULL, 0, &ndp, 1, delegtype,
2150 recovered_one = true;
2151 /* Handle any replied delegation */
2152 if (ndp != NULL && ((ndp->nfsdl_flags & NFSCLDL_WRITE)
2153 || NFSMNT_RDONLY(nmp->nm_mountp))) {
2154 if ((ndp->nfsdl_flags & NFSCLDL_WRITE))
2155 mode = NFSV4OPEN_ACCESSWRITE;
2157 mode = NFSV4OPEN_ACCESSREAD;
2158 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2159 if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
2161 if ((op->nfso_mode & mode) == mode &&
2162 op->nfso_fhlen == dp->nfsdl_fhlen &&
2163 !NFSBCMP(op->nfso_fh, dp->nfsdl_fh,
2165 dp->nfsdl_stateid = ndp->nfsdl_stateid;
2166 dp->nfsdl_sizelimit = ndp->nfsdl_sizelimit;
2167 dp->nfsdl_ace = ndp->nfsdl_ace;
2168 dp->nfsdl_change = ndp->nfsdl_change;
2169 dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
2170 if ((ndp->nfsdl_flags & NFSCLDL_RECALL))
2171 dp->nfsdl_flags |= NFSCLDL_RECALL;
2172 free(ndp, M_NFSCLDELEG);
2179 TAILQ_INSERT_HEAD(&extra_deleg, ndp, nfsdl_list);
2181 /* and reclaim all byte range locks */
2182 lp = LIST_FIRST(&op->nfso_lock);
2183 while (lp != NULL) {
2184 nlp = LIST_NEXT(lp, nfsl_list);
2187 lop = LIST_FIRST(&lp->nfsl_lock);
2188 while (lop != NULL) {
2189 nlop = LIST_NEXT(lop, nfslo_list);
2190 if (lop->nfslo_end == NFS64BITSSET)
2193 len = lop->nfslo_end - lop->nfslo_first;
2194 error = nfscl_trylock(nmp, NULL,
2195 op->nfso_fh, op->nfso_fhlen, lp,
2196 firstlock, 1, lop->nfslo_first, len,
2197 lop->nfslo_type, tcred, p);
2199 nfscl_freelock(lop, 0);
2204 /* If no locks, but a lockowner, just delete it. */
2205 if (LIST_EMPTY(&lp->nfsl_lock))
2206 nfscl_freelockowner(lp, 0);
2209 } else if (error == NFSERR_NOGRACE && !recovered_one &&
2210 NFSHASNFSV4N(nmp)) {
2212 * For NFSv4.1/4.2, the NFSERR_EXPIRED case will
2213 * actually end up here, since the client will do
2214 * a recovery for NFSERR_BADSESSION, but will get
2215 * an NFSERR_NOGRACE reply for the first "reclaim"
2217 * So, call nfscl_expireclient() to recover the
2218 * opens as best we can and then do a reclaim
2219 * complete and return.
2221 nfsrpc_reclaimcomplete(nmp, cred, p);
2222 nfscl_expireclient(clp, nmp, tcred, p);
2226 if (error != 0 && error != NFSERR_BADSESSION)
2227 nfscl_freeopen(op, 0);
2234 * Now, try and get any delegations not yet reclaimed by cobbling
2235 * to-gether an appropriate open.
2238 dp = TAILQ_FIRST(&clp->nfsc_deleg);
2239 while (dp != NULL) {
2240 ndp = TAILQ_NEXT(dp, nfsdl_list);
2241 if ((dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM)) {
2244 sizeof (struct nfsclowner), M_NFSCLOWNER, M_WAITOK);
2246 * Name must be as long an largest possible
2247 * NFSV4CL_LOCKNAMELEN. 12 for now.
2249 NFSBCOPY("RECLAIMDELEG", nowp->nfsow_owner,
2250 NFSV4CL_LOCKNAMELEN);
2251 LIST_INIT(&nowp->nfsow_open);
2252 nowp->nfsow_clp = clp;
2253 nowp->nfsow_seqid = 0;
2254 nowp->nfsow_defunct = 0;
2255 nfscl_lockinit(&nowp->nfsow_rwlock);
2258 if (error != NFSERR_NOGRACE && error != NFSERR_BADSESSION) {
2259 nop = malloc(sizeof (struct nfsclopen) +
2260 dp->nfsdl_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
2261 nop->nfso_own = nowp;
2262 if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
2263 nop->nfso_mode = NFSV4OPEN_ACCESSWRITE;
2264 delegtype = NFSV4OPEN_DELEGATEWRITE;
2266 nop->nfso_mode = NFSV4OPEN_ACCESSREAD;
2267 delegtype = NFSV4OPEN_DELEGATEREAD;
2269 nop->nfso_opencnt = 0;
2270 nop->nfso_posixlock = 1;
2271 nop->nfso_fhlen = dp->nfsdl_fhlen;
2272 NFSBCOPY(dp->nfsdl_fh, nop->nfso_fh, dp->nfsdl_fhlen);
2273 LIST_INIT(&nop->nfso_lock);
2274 nop->nfso_stateid.seqid = 0;
2275 nop->nfso_stateid.other[0] = 0;
2276 nop->nfso_stateid.other[1] = 0;
2277 nop->nfso_stateid.other[2] = 0;
2278 newnfs_copycred(&dp->nfsdl_cred, tcred);
2279 newnfs_copyincred(tcred, &nop->nfso_cred);
2281 error = nfscl_tryopen(nmp, NULL, nop->nfso_fh,
2282 nop->nfso_fhlen, nop->nfso_fh, nop->nfso_fhlen,
2283 nop->nfso_mode, nop, NULL, 0, &tdp, 1,
2284 delegtype, tcred, p);
2286 if ((tdp->nfsdl_flags & NFSCLDL_WRITE))
2287 mode = NFSV4OPEN_ACCESSWRITE;
2289 mode = NFSV4OPEN_ACCESSREAD;
2290 if ((nop->nfso_mode & mode) == mode &&
2291 nop->nfso_fhlen == tdp->nfsdl_fhlen &&
2292 !NFSBCMP(nop->nfso_fh, tdp->nfsdl_fh,
2294 dp->nfsdl_stateid = tdp->nfsdl_stateid;
2295 dp->nfsdl_sizelimit = tdp->nfsdl_sizelimit;
2296 dp->nfsdl_ace = tdp->nfsdl_ace;
2297 dp->nfsdl_change = tdp->nfsdl_change;
2298 dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
2299 if ((tdp->nfsdl_flags & NFSCLDL_RECALL))
2300 dp->nfsdl_flags |= NFSCLDL_RECALL;
2301 free(tdp, M_NFSCLDELEG);
2303 TAILQ_INSERT_HEAD(&extra_deleg, tdp, nfsdl_list);
2309 free(nop, M_NFSCLOPEN);
2310 if (error == NFSERR_NOGRACE && !recovered_one &&
2311 NFSHASNFSV4N(nmp)) {
2313 * For NFSv4.1/4.2, the NFSERR_EXPIRED case will
2314 * actually end up here, since the client will do
2315 * a recovery for NFSERR_BADSESSION, but will get
2316 * an NFSERR_NOGRACE reply for the first "reclaim"
2318 * So, call nfscl_expireclient() to recover the
2319 * opens as best we can and then do a reclaim
2320 * complete and return.
2322 nfsrpc_reclaimcomplete(nmp, cred, p);
2323 nfscl_expireclient(clp, nmp, tcred, p);
2324 free(nowp, M_NFSCLOWNER);
2328 * Couldn't reclaim it, so throw the state
2331 nfscl_cleandeleg(dp);
2332 nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
2334 recovered_one = true;
2335 LIST_INSERT_HEAD(&extra_open, nop, nfso_list);
2342 * Now, get rid of extra Opens and Delegations.
2344 LIST_FOREACH_SAFE(op, &extra_open, nfso_list, nop) {
2346 newnfs_copycred(&op->nfso_cred, tcred);
2347 error = nfscl_tryclose(op, tcred, nmp, p);
2348 if (error == NFSERR_GRACE)
2349 (void) nfs_catnap(PZERO, error, "nfsexcls");
2350 } while (error == NFSERR_GRACE);
2351 LIST_REMOVE(op, nfso_list);
2352 free(op, M_NFSCLOPEN);
2355 free(nowp, M_NFSCLOWNER);
2357 TAILQ_FOREACH_SAFE(dp, &extra_deleg, nfsdl_list, ndp) {
2359 newnfs_copycred(&dp->nfsdl_cred, tcred);
2360 error = nfscl_trydelegreturn(dp, tcred, nmp, p);
2361 if (error == NFSERR_GRACE)
2362 (void) nfs_catnap(PZERO, error, "nfsexdlg");
2363 } while (error == NFSERR_GRACE);
2364 TAILQ_REMOVE(&extra_deleg, dp, nfsdl_list);
2365 free(dp, M_NFSCLDELEG);
2368 /* For NFSv4.1 or later, do a RECLAIM_COMPLETE. */
2369 if (NFSHASNFSV4N(nmp))
2370 (void)nfsrpc_reclaimcomplete(nmp, cred, p);
2374 clp->nfsc_flags &= ~NFSCLFLAGS_RECVRINPROG;
2375 wakeup(&clp->nfsc_flags);
2376 nfsv4_unlock(&clp->nfsc_lock, 0);
2383 * This function is called when a server replies with NFSERR_EXPIRED.
2384 * It deletes all state for the client and does a fresh SetClientId/confirm.
2385 * XXX Someday it should post a signal to the process(es) that hold the
2386 * state, so they know that lock state has been lost.
2389 nfscl_hasexpired(struct nfsclclient *clp, u_int32_t clidrev, NFSPROC_T *p)
2391 struct nfsmount *nmp;
2393 int igotlock = 0, error, trycnt;
2396 * If the clientid has gone away or a new SetClientid has already
2397 * been done, just return ok.
2399 if (clp == NULL || clidrev != clp->nfsc_clientidrev)
2403 * First, lock the client structure, so everyone else will
2404 * block when trying to use state. Also, use NFSCLFLAGS_EXPIREIT so
2405 * that only one thread does the work.
2408 clp->nfsc_flags |= NFSCLFLAGS_EXPIREIT;
2410 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
2411 NFSCLSTATEMUTEXPTR, NULL);
2412 } while (!igotlock && (clp->nfsc_flags & NFSCLFLAGS_EXPIREIT));
2413 if ((clp->nfsc_flags & NFSCLFLAGS_EXPIREIT) == 0) {
2415 nfsv4_unlock(&clp->nfsc_lock, 0);
2419 clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG;
2422 nmp = clp->nfsc_nmp;
2424 panic("nfscl expired");
2425 cred = newnfs_getcred();
2428 error = nfsrpc_setclient(nmp, clp, 0, NULL, cred, p);
2429 } while ((error == NFSERR_STALECLIENTID ||
2430 error == NFSERR_BADSESSION ||
2431 error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
2434 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2437 * Expire the state for the client.
2439 nfscl_expireclient(clp, nmp, cred, p);
2441 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
2442 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2444 clp->nfsc_flags &= ~(NFSCLFLAGS_EXPIREIT | NFSCLFLAGS_RECVRINPROG);
2445 wakeup(&clp->nfsc_flags);
2446 nfsv4_unlock(&clp->nfsc_lock, 0);
2453 * This function inserts a lock in the list after insert_lop.
2456 nfscl_insertlock(struct nfscllockowner *lp, struct nfscllock *new_lop,
2457 struct nfscllock *insert_lop, int local)
2460 if ((struct nfscllockowner *)insert_lop == lp)
2461 LIST_INSERT_HEAD(&lp->nfsl_lock, new_lop, nfslo_list);
2463 LIST_INSERT_AFTER(insert_lop, new_lop, nfslo_list);
2465 nfsstatsv1.cllocallocks++;
2467 nfsstatsv1.cllocks++;
2471 * This function updates the locking for a lock owner and given file. It
2472 * maintains a list of lock ranges ordered on increasing file offset that
2473 * are NFSCLLOCK_READ or NFSCLLOCK_WRITE and non-overlapping (aka POSIX style).
2474 * It always adds new_lop to the list and sometimes uses the one pointed
2476 * Returns 1 if the locks were modified, 0 otherwise.
2479 nfscl_updatelock(struct nfscllockowner *lp, struct nfscllock **new_lopp,
2480 struct nfscllock **other_lopp, int local)
2482 struct nfscllock *new_lop = *new_lopp;
2483 struct nfscllock *lop, *tlop, *ilop;
2484 struct nfscllock *other_lop;
2485 int unlock = 0, modified = 0;
2489 * Work down the list until the lock is merged.
2491 if (new_lop->nfslo_type == F_UNLCK)
2493 ilop = (struct nfscllock *)lp;
2494 lop = LIST_FIRST(&lp->nfsl_lock);
2495 while (lop != NULL) {
2497 * Only check locks for this file that aren't before the start of
2500 if (lop->nfslo_end >= new_lop->nfslo_first) {
2501 if (new_lop->nfslo_end < lop->nfslo_first) {
2503 * If the new lock ends before the start of the
2504 * current lock's range, no merge, just insert
2509 if (new_lop->nfslo_type == lop->nfslo_type ||
2510 (new_lop->nfslo_first <= lop->nfslo_first &&
2511 new_lop->nfslo_end >= lop->nfslo_end)) {
2513 * This lock can be absorbed by the new lock/unlock.
2514 * This happens when it covers the entire range
2515 * of the old lock or is contiguous
2516 * with the old lock and is of the same type or an
2519 if (new_lop->nfslo_type != lop->nfslo_type ||
2520 new_lop->nfslo_first != lop->nfslo_first ||
2521 new_lop->nfslo_end != lop->nfslo_end)
2523 if (lop->nfslo_first < new_lop->nfslo_first)
2524 new_lop->nfslo_first = lop->nfslo_first;
2525 if (lop->nfslo_end > new_lop->nfslo_end)
2526 new_lop->nfslo_end = lop->nfslo_end;
2528 lop = LIST_NEXT(lop, nfslo_list);
2529 nfscl_freelock(tlop, local);
2534 * All these cases are for contiguous locks that are not the
2535 * same type, so they can't be merged.
2537 if (new_lop->nfslo_first <= lop->nfslo_first) {
2539 * This case is where the new lock overlaps with the
2540 * first part of the old lock. Move the start of the
2541 * old lock to just past the end of the new lock. The
2542 * new lock will be inserted in front of the old, since
2543 * ilop hasn't been updated. (We are done now.)
2545 if (lop->nfslo_first != new_lop->nfslo_end) {
2546 lop->nfslo_first = new_lop->nfslo_end;
2551 if (new_lop->nfslo_end >= lop->nfslo_end) {
2553 * This case is where the new lock overlaps with the
2554 * end of the old lock's range. Move the old lock's
2555 * end to just before the new lock's first and insert
2556 * the new lock after the old lock.
2557 * Might not be done yet, since the new lock could
2558 * overlap further locks with higher ranges.
2560 if (lop->nfslo_end != new_lop->nfslo_first) {
2561 lop->nfslo_end = new_lop->nfslo_first;
2565 lop = LIST_NEXT(lop, nfslo_list);
2569 * The final case is where the new lock's range is in the
2570 * middle of the current lock's and splits the current lock
2571 * up. Use *other_lopp to handle the second part of the
2572 * split old lock range. (We are done now.)
2573 * For unlock, we use new_lop as other_lop and tmp, since
2574 * other_lop and new_lop are the same for this case.
2575 * We noted the unlock case above, so we don't need
2576 * new_lop->nfslo_type any longer.
2578 tmp = new_lop->nfslo_first;
2580 other_lop = new_lop;
2583 other_lop = *other_lopp;
2586 other_lop->nfslo_first = new_lop->nfslo_end;
2587 other_lop->nfslo_end = lop->nfslo_end;
2588 other_lop->nfslo_type = lop->nfslo_type;
2589 lop->nfslo_end = tmp;
2590 nfscl_insertlock(lp, other_lop, lop, local);
2596 lop = LIST_NEXT(lop, nfslo_list);
2602 * Insert the new lock in the list at the appropriate place.
2605 nfscl_insertlock(lp, new_lop, ilop, local);
2613 * This function must be run as a kernel thread.
2614 * It does Renew Ops and recovery, when required.
2617 nfscl_renewthread(struct nfsclclient *clp, NFSPROC_T *p)
2619 struct nfsclowner *owp, *nowp;
2620 struct nfsclopen *op;
2621 struct nfscllockowner *lp, *nlp;
2622 struct nfscldeleghead dh;
2623 struct nfscldeleg *dp, *ndp;
2626 int error, cbpathdown, islept, igotlock, ret, clearok;
2627 uint32_t recover_done_time = 0;
2629 static time_t prevsec = 0;
2630 struct nfscllockownerfh *lfhp, *nlfhp;
2631 struct nfscllockownerfhhead lfh;
2632 struct nfscllayout *lyp, *nlyp;
2633 struct nfscldevinfo *dip, *ndip;
2634 struct nfscllayouthead rlh;
2635 struct nfsclrecalllayout *recallp;
2636 struct nfsclds *dsp;
2641 cred = newnfs_getcred();
2643 clp->nfsc_flags |= NFSCLFLAGS_HASTHREAD;
2644 mp = clp->nfsc_nmp->nm_mountp;
2647 newnfs_setroot(cred);
2649 if (clp->nfsc_flags & NFSCLFLAGS_RECOVER) {
2651 * Only allow one full recover within 1/2 of the lease
2652 * duration (nfsc_renew).
2653 * retok is value/result. If passed in set to true,
2654 * it indicates only a CreateSession operation should
2656 * If it is returned true, it indicates that the
2657 * recovery only required a CreateSession.
2660 if (recover_done_time < NFSD_MONOSEC) {
2661 recover_done_time = NFSD_MONOSEC +
2665 NFSCL_DEBUG(1, "Doing recovery, only "
2666 "createsession=%d\n", retok);
2667 nfscl_recover(clp, &retok, cred, p);
2669 if (clp->nfsc_expire <= NFSD_MONOSEC &&
2670 (clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID)) {
2671 clp->nfsc_expire = NFSD_MONOSEC + clp->nfsc_renew;
2672 clidrev = clp->nfsc_clientidrev;
2673 error = nfsrpc_renew(clp, NULL, cred, p);
2674 if (error == NFSERR_CBPATHDOWN)
2676 else if (error == NFSERR_STALECLIENTID ||
2677 error == NFSERR_BADSESSION) {
2679 clp->nfsc_flags |= NFSCLFLAGS_RECOVER;
2681 } else if (error == NFSERR_EXPIRED)
2682 (void) nfscl_hasexpired(clp, clidrev, p);
2686 if (NFSHASNFSV4N(clp->nfsc_nmp)) {
2687 /* Do renews for any DS sessions. */
2688 NFSLOCKMNT(clp->nfsc_nmp);
2689 /* Skip first entry, since the MDS is handled above. */
2690 dsp = TAILQ_FIRST(&clp->nfsc_nmp->nm_sess);
2692 dsp = TAILQ_NEXT(dsp, nfsclds_list);
2693 while (dsp != NULL) {
2694 if (dsp->nfsclds_expire <= NFSD_MONOSEC &&
2695 dsp->nfsclds_sess.nfsess_defunct == 0) {
2696 dsp->nfsclds_expire = NFSD_MONOSEC +
2698 NFSUNLOCKMNT(clp->nfsc_nmp);
2699 (void)nfsrpc_renew(clp, dsp, cred, p);
2702 dsp = TAILQ_NEXT(dsp, nfsclds_list);
2704 NFSUNLOCKMNT(clp->nfsc_nmp);
2710 /* It's a Total Recall! */
2711 nfscl_totalrecall(clp);
2714 * Now, handle defunct owners.
2716 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
2717 if (LIST_EMPTY(&owp->nfsow_open)) {
2718 if (owp->nfsow_defunct != 0)
2719 nfscl_freeopenowner(owp, 0);
2724 * Do the recall on any delegations. To avoid trouble, always
2725 * come back up here after having slept.
2729 dp = TAILQ_FIRST(&clp->nfsc_deleg);
2730 while (dp != NULL) {
2731 ndp = TAILQ_NEXT(dp, nfsdl_list);
2732 if ((dp->nfsdl_flags & NFSCLDL_RECALL)) {
2734 * Wait for outstanding I/O ops to be done.
2736 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
2738 nfsv4_unlock(&clp->nfsc_lock, 0);
2741 dp->nfsdl_rwlock.nfslock_lock |=
2743 msleep(&dp->nfsdl_rwlock,
2744 NFSCLSTATEMUTEXPTR, PVFS, "nfscld",
2746 if (NFSCL_FORCEDISM(mp))
2751 igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
2752 &islept, NFSCLSTATEMUTEXPTR, mp);
2753 if (igotlock == 0 && NFSCL_FORCEDISM(mp))
2759 newnfs_copycred(&dp->nfsdl_cred, cred);
2760 ret = nfscl_recalldeleg(clp, clp->nfsc_nmp, dp,
2761 NULL, cred, p, 1, &vp);
2763 nfscl_cleandeleg(dp);
2764 TAILQ_REMOVE(&clp->nfsc_deleg, dp,
2766 LIST_REMOVE(dp, nfsdl_hash);
2767 TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
2769 nfsstatsv1.cldelegates--;
2773 * The nfsc_lock must be released before doing
2774 * vrele(), since it might call nfs_inactive().
2775 * For the unlikely case where the vnode failed
2776 * to be acquired by nfscl_recalldeleg(), a
2777 * VOP_RECLAIM() should be in progress and it
2778 * will return the delegation.
2780 nfsv4_unlock(&clp->nfsc_lock, 0);
2793 * Clear out old delegations, if we are above the high water
2794 * mark. Only clear out ones with no state related to them.
2795 * The tailq list is in LRU order.
2797 dp = TAILQ_LAST(&clp->nfsc_deleg, nfscldeleghead);
2798 while (nfscl_delegcnt > nfscl_deleghighwater && dp != NULL) {
2799 ndp = TAILQ_PREV(dp, nfscldeleghead, nfsdl_list);
2800 if (dp->nfsdl_rwlock.nfslock_usecnt == 0 &&
2801 dp->nfsdl_rwlock.nfslock_lock == 0 &&
2802 dp->nfsdl_timestamp < NFSD_MONOSEC &&
2803 (dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_ZAPPED |
2804 NFSCLDL_NEEDRECLAIM | NFSCLDL_DELEGRET)) == 0) {
2806 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
2807 op = LIST_FIRST(&owp->nfsow_open);
2814 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
2815 if (!LIST_EMPTY(&lp->nfsl_lock)) {
2822 TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
2823 LIST_REMOVE(dp, nfsdl_hash);
2824 TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
2826 nfsstatsv1.cldelegates--;
2832 nfsv4_unlock(&clp->nfsc_lock, 0);
2835 * Do the recall on any layouts. To avoid trouble, always
2836 * come back up here after having slept.
2840 TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp) {
2841 if ((lyp->nfsly_flags & NFSLY_RECALL) != 0) {
2843 * Wait for outstanding I/O ops to be done.
2845 if (lyp->nfsly_lock.nfslock_usecnt > 0 ||
2846 (lyp->nfsly_lock.nfslock_lock &
2847 NFSV4LOCK_LOCK) != 0) {
2848 lyp->nfsly_lock.nfslock_lock |=
2850 msleep(&lyp->nfsly_lock.nfslock_lock,
2851 NFSCLSTATEMUTEXPTR, PVFS, "nfslyp",
2853 if (NFSCL_FORCEDISM(mp))
2857 /* Move the layout to the recall list. */
2858 TAILQ_REMOVE(&clp->nfsc_layout, lyp,
2860 LIST_REMOVE(lyp, nfsly_hash);
2861 TAILQ_INSERT_HEAD(&rlh, lyp, nfsly_list);
2863 /* Handle any layout commits. */
2864 if (!NFSHASNOLAYOUTCOMMIT(clp->nfsc_nmp) &&
2865 (lyp->nfsly_flags & NFSLY_WRITTEN) != 0) {
2866 lyp->nfsly_flags &= ~NFSLY_WRITTEN;
2868 NFSCL_DEBUG(3, "do layoutcommit\n");
2869 nfscl_dolayoutcommit(clp->nfsc_nmp, lyp,
2877 /* Now, look for stale layouts. */
2878 lyp = TAILQ_LAST(&clp->nfsc_layout, nfscllayouthead);
2879 while (lyp != NULL) {
2880 nlyp = TAILQ_PREV(lyp, nfscllayouthead, nfsly_list);
2881 if (lyp->nfsly_timestamp < NFSD_MONOSEC &&
2882 (lyp->nfsly_flags & NFSLY_RECALL) == 0 &&
2883 lyp->nfsly_lock.nfslock_usecnt == 0 &&
2884 lyp->nfsly_lock.nfslock_lock == 0) {
2885 NFSCL_DEBUG(4, "ret stale lay=%d\n",
2887 recallp = malloc(sizeof(*recallp),
2888 M_NFSLAYRECALL, M_NOWAIT);
2889 if (recallp == NULL)
2891 (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE,
2892 lyp, NFSLAYOUTIOMODE_ANY, 0, UINT64_MAX,
2893 lyp->nfsly_stateid.seqid, 0, 0, NULL,
2900 * Free up any unreferenced device info structures.
2902 LIST_FOREACH_SAFE(dip, &clp->nfsc_devinfo, nfsdi_list, ndip) {
2903 if (dip->nfsdi_layoutrefs == 0 &&
2904 dip->nfsdi_refcnt == 0) {
2905 NFSCL_DEBUG(4, "freeing devinfo\n");
2906 LIST_REMOVE(dip, nfsdi_list);
2907 nfscl_freedevinfo(dip);
2912 /* Do layout return(s), as required. */
2913 TAILQ_FOREACH_SAFE(lyp, &rlh, nfsly_list, nlyp) {
2914 TAILQ_REMOVE(&rlh, lyp, nfsly_list);
2915 NFSCL_DEBUG(4, "ret layout\n");
2916 nfscl_layoutreturn(clp->nfsc_nmp, lyp, cred, p);
2917 nfscl_freelayout(lyp);
2921 * Delegreturn any delegations cleaned out or recalled.
2923 TAILQ_FOREACH_SAFE(dp, &dh, nfsdl_list, ndp) {
2924 newnfs_copycred(&dp->nfsdl_cred, cred);
2925 (void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
2926 TAILQ_REMOVE(&dh, dp, nfsdl_list);
2927 free(dp, M_NFSCLDELEG);
2932 * Call nfscl_cleanupkext() once per second to check for
2933 * open/lock owners where the process has exited.
2935 mytime = NFSD_MONOSEC;
2936 if (prevsec != mytime) {
2938 nfscl_cleanupkext(clp, &lfh);
2942 * Do a ReleaseLockOwner for all lock owners where the
2943 * associated process no longer exists, as found by
2944 * nfscl_cleanupkext().
2946 newnfs_setroot(cred);
2947 SLIST_FOREACH_SAFE(lfhp, &lfh, nfslfh_list, nlfhp) {
2948 LIST_FOREACH_SAFE(lp, &lfhp->nfslfh_lock, nfsl_list,
2950 (void)nfsrpc_rellockown(clp->nfsc_nmp, lp,
2951 lfhp->nfslfh_fh, lfhp->nfslfh_len, cred,
2953 nfscl_freelockowner(lp, 0);
2960 if ((clp->nfsc_flags & NFSCLFLAGS_RECOVER) == 0)
2961 (void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT, "nfscl",
2964 if (clp->nfsc_flags & NFSCLFLAGS_UMOUNT) {
2965 clp->nfsc_flags &= ~NFSCLFLAGS_HASTHREAD;
2968 wakeup((caddr_t)clp);
2976 * Initiate state recovery. Called when NFSERR_STALECLIENTID,
2977 * NFSERR_STALESTATEID or NFSERR_BADSESSION is received.
2980 nfscl_initiate_recovery(struct nfsclclient *clp)
2986 clp->nfsc_flags |= NFSCLFLAGS_RECOVER;
2988 wakeup((caddr_t)clp);
2992 * Dump out the state stuff for debugging.
2995 nfscl_dumpstate(struct nfsmount *nmp, int openowner, int opens,
2996 int lockowner, int locks)
2998 struct nfsclclient *clp;
2999 struct nfsclowner *owp;
3000 struct nfsclopen *op;
3001 struct nfscllockowner *lp;
3002 struct nfscllock *lop;
3003 struct nfscldeleg *dp;
3007 printf("nfscl dumpstate NULL clp\n");
3011 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
3012 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
3013 if (openowner && !LIST_EMPTY(&owp->nfsow_open))
3014 printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
3015 owp->nfsow_owner[0], owp->nfsow_owner[1],
3016 owp->nfsow_owner[2], owp->nfsow_owner[3],
3018 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3020 printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
3021 op->nfso_stateid.other[0], op->nfso_stateid.other[1],
3022 op->nfso_stateid.other[2], op->nfso_opencnt,
3024 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
3026 printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
3027 lp->nfsl_owner[0], lp->nfsl_owner[1],
3028 lp->nfsl_owner[2], lp->nfsl_owner[3],
3030 lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
3031 lp->nfsl_stateid.other[2]);
3032 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
3035 printf("lck typ=%d fst=%ju end=%ju\n",
3036 lop->nfslo_type, (intmax_t)lop->nfslo_first,
3037 (intmax_t)lop->nfslo_end);
3039 printf("lck typ=%d fst=%qd end=%qd\n",
3040 lop->nfslo_type, lop->nfslo_first,
3048 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3049 if (openowner && !LIST_EMPTY(&owp->nfsow_open))
3050 printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
3051 owp->nfsow_owner[0], owp->nfsow_owner[1],
3052 owp->nfsow_owner[2], owp->nfsow_owner[3],
3054 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3056 printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
3057 op->nfso_stateid.other[0], op->nfso_stateid.other[1],
3058 op->nfso_stateid.other[2], op->nfso_opencnt,
3060 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
3062 printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
3063 lp->nfsl_owner[0], lp->nfsl_owner[1],
3064 lp->nfsl_owner[2], lp->nfsl_owner[3],
3066 lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
3067 lp->nfsl_stateid.other[2]);
3068 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
3071 printf("lck typ=%d fst=%ju end=%ju\n",
3072 lop->nfslo_type, (intmax_t)lop->nfslo_first,
3073 (intmax_t)lop->nfslo_end);
3075 printf("lck typ=%d fst=%qd end=%qd\n",
3076 lop->nfslo_type, lop->nfslo_first,
3087 * Check for duplicate open owners and opens.
3088 * (Only used as a diagnostic aid.)
3091 nfscl_dupopen(vnode_t vp, int dupopens)
3093 struct nfsclclient *clp;
3094 struct nfsclowner *owp, *owp2;
3095 struct nfsclopen *op, *op2;
3098 clp = VFSTONFS(vp->v_mount)->nm_clp;
3100 printf("nfscl dupopen NULL clp\n");
3103 nfhp = VTONFS(vp)->n_fhp;
3107 * First, search for duplicate owners.
3108 * These should never happen!
3110 LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3111 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3113 !NFSBCMP(owp->nfsow_owner, owp2->nfsow_owner,
3114 NFSV4CL_LOCKNAMELEN)) {
3116 printf("DUP OWNER\n");
3117 nfscl_dumpstate(VFSTONFS(vp->v_mount), 1, 1, 0, 0);
3124 * Now, search for duplicate stateids.
3125 * These shouldn't happen, either.
3127 LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3128 LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
3129 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3130 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3132 (op->nfso_stateid.other[0] != 0 ||
3133 op->nfso_stateid.other[1] != 0 ||
3134 op->nfso_stateid.other[2] != 0) &&
3135 op->nfso_stateid.other[0] == op2->nfso_stateid.other[0] &&
3136 op->nfso_stateid.other[1] == op2->nfso_stateid.other[1] &&
3137 op->nfso_stateid.other[2] == op2->nfso_stateid.other[2]) {
3139 printf("DUP STATEID\n");
3140 nfscl_dumpstate(VFSTONFS(vp->v_mount), 1, 1, 0, 0);
3149 * Now search for duplicate opens.
3150 * Duplicate opens for the same owner
3151 * should never occur. Other duplicates are
3152 * possible and are checked for if "dupopens"
3155 LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3156 LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
3157 if (nfhp->nfh_len == op2->nfso_fhlen &&
3158 !NFSBCMP(nfhp->nfh_fh, op2->nfso_fh, nfhp->nfh_len)) {
3159 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3160 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3161 if (op != op2 && nfhp->nfh_len == op->nfso_fhlen &&
3162 !NFSBCMP(nfhp->nfh_fh, op->nfso_fh, nfhp->nfh_len) &&
3163 (!NFSBCMP(op->nfso_own->nfsow_owner,
3164 op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN) ||
3166 if (!NFSBCMP(op->nfso_own->nfsow_owner,
3167 op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
3169 printf("BADDUP OPEN\n");
3172 printf("DUP OPEN\n");
3174 nfscl_dumpstate(VFSTONFS(vp->v_mount), 1, 1, 0,
3187 * During close, find an open that needs to be dereferenced and
3188 * dereference it. If there are no more opens for this file,
3189 * log a message to that effect.
3190 * Opens aren't actually Close'd until VOP_INACTIVE() is performed
3191 * on the file's vnode.
3192 * This is the safe way, since it is difficult to identify
3193 * which open the close is for and I/O can be performed after the
3194 * close(2) system call when a file is mmap'd.
3195 * If it returns 0 for success, there will be a referenced
3196 * clp returned via clpp.
3199 nfscl_getclose(vnode_t vp, struct nfsclclient **clpp)
3201 struct nfsclclient *clp;
3202 struct nfsclowner *owp;
3203 struct nfsclopen *op;
3204 struct nfscldeleg *dp;
3208 error = nfscl_getcl(vp->v_mount, NULL, NULL, 1, &clp);
3213 nfhp = VTONFS(vp)->n_fhp;
3217 * First, look for one under a delegation that was locally issued
3218 * and just decrement the opencnt for it. Since all my Opens against
3219 * the server are DENY_NONE, I don't see a problem with hanging
3220 * onto them. (It is much easier to use one of the extant Opens
3221 * that I already have on the server when a Delegation is recalled
3222 * than to do fresh Opens.) Someday, I might need to rethink this, but.
3224 dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
3226 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
3227 op = LIST_FIRST(&owp->nfsow_open);
3230 * Since a delegation is for a file, there
3231 * should never be more than one open for
3234 if (LIST_NEXT(op, nfso_list) != NULL)
3235 panic("nfscdeleg opens");
3236 if (notdecr && op->nfso_opencnt > 0) {
3245 /* Now process the opens against the server. */
3246 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3247 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3248 if (op->nfso_fhlen == nfhp->nfh_len &&
3249 !NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
3251 /* Found an open, decrement cnt if possible */
3252 if (notdecr && op->nfso_opencnt > 0) {
3257 * There are more opens, so just return.
3259 if (op->nfso_opencnt > 0) {
3268 printf("nfscl: never fnd open\n");
3273 nfscl_doclose(vnode_t vp, struct nfsclclient **clpp, NFSPROC_T *p)
3275 struct nfsclclient *clp;
3276 struct nfsclowner *owp, *nowp;
3277 struct nfsclopen *op;
3278 struct nfscldeleg *dp;
3280 struct nfsclrecalllayout *recallp;
3283 error = nfscl_getcl(vp->v_mount, NULL, NULL, 1, &clp);
3288 nfhp = VTONFS(vp)->n_fhp;
3289 recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_WAITOK);
3292 * First get rid of the local Open structures, which should be no
3295 dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
3297 LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) {
3298 op = LIST_FIRST(&owp->nfsow_open);
3300 KASSERT((op->nfso_opencnt == 0),
3301 ("nfscl: bad open cnt on deleg"));
3302 nfscl_freeopen(op, 1);
3304 nfscl_freeopenowner(owp, 1);
3308 /* Return any layouts marked return on close. */
3309 nfscl_retoncloselayout(vp, clp, nfhp->nfh_fh, nfhp->nfh_len, &recallp);
3311 /* Now process the opens against the server. */
3313 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3314 op = LIST_FIRST(&owp->nfsow_open);
3315 while (op != NULL) {
3316 if (op->nfso_fhlen == nfhp->nfh_len &&
3317 !NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
3319 /* Found an open, close it. */
3321 KASSERT((op->nfso_opencnt == 0),
3322 ("nfscl: bad open cnt on server (%d)",
3326 nfsrpc_doclose(VFSTONFS(vp->v_mount), op, p);
3330 op = LIST_NEXT(op, nfso_list);
3335 * recallp has been set NULL by nfscl_retoncloselayout() if it was
3336 * used by the function, but calling free() with a NULL pointer is ok.
3338 free(recallp, M_NFSLAYRECALL);
3343 * Return all delegations on this client.
3344 * (Must be called with client sleep lock.)
3347 nfscl_delegreturnall(struct nfsclclient *clp, NFSPROC_T *p)
3349 struct nfscldeleg *dp, *ndp;
3352 cred = newnfs_getcred();
3353 TAILQ_FOREACH_SAFE(dp, &clp->nfsc_deleg, nfsdl_list, ndp) {
3354 nfscl_cleandeleg(dp);
3355 (void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
3356 nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
3362 * Return any delegation for this vp.
3365 nfscl_delegreturnvp(vnode_t vp, NFSPROC_T *p)
3367 struct nfsclclient *clp;
3368 struct nfscldeleg *dp;
3373 cred = newnfs_getcred();
3376 clp = VFSTONFS(vp->v_mount)->nm_clp;
3378 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
3379 np->n_fhp->nfh_len);
3381 nfscl_cleandeleg(dp);
3382 nfscl_freedeleg(&clp->nfsc_deleg, dp, false);
3384 newnfs_copycred(&dp->nfsdl_cred, cred);
3385 nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
3386 free(dp, M_NFSCLDELEG);
3393 * Do a callback RPC.
3396 nfscl_docb(struct nfsrv_descript *nd, NFSPROC_T *p)
3398 int clist, gotseq_ok, i, j, k, op, rcalls;
3400 struct nfsclclient *clp;
3401 struct nfscldeleg *dp = NULL;
3402 int numops, taglen = -1, error = 0, trunc __unused;
3403 u_int32_t minorvers = 0, retops = 0, *retopsp = NULL, *repp, cbident;
3404 u_char tag[NFSV4_SMALLSTR + 1], *tagstr;
3410 nfsattrbit_t attrbits, rattrbits;
3411 nfsv4stateid_t stateid;
3412 uint32_t seqid, slotid = 0, highslot, cachethis __unused;
3413 uint8_t sessionid[NFSX_V4SESSIONID];
3415 struct nfscllayout *lyp;
3416 uint64_t filesid[2], len, off;
3417 int changed, gotone, laytype, recalltype;
3419 struct nfsclrecalllayout *recallp = NULL;
3420 struct nfsclsession *tsep;
3424 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3425 taglen = fxdr_unsigned(int, *tl);
3430 if (taglen <= NFSV4_SMALLSTR)
3433 tagstr = malloc(taglen + 1, M_TEMP, M_WAITOK);
3434 error = nfsrv_mtostr(nd, tagstr, taglen);
3436 if (taglen > NFSV4_SMALLSTR)
3437 free(tagstr, M_TEMP);
3441 (void) nfsm_strtom(nd, tag, taglen);
3442 if (taglen > NFSV4_SMALLSTR) {
3443 free(tagstr, M_TEMP);
3445 NFSM_BUILD(retopsp, u_int32_t *, NFSX_UNSIGNED);
3446 NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3447 minorvers = fxdr_unsigned(u_int32_t, *tl++);
3448 if (minorvers != NFSV4_MINORVERSION &&
3449 minorvers != NFSV41_MINORVERSION &&
3450 minorvers != NFSV42_MINORVERSION)
3451 nd->nd_repstat = NFSERR_MINORVERMISMATCH;
3452 cbident = fxdr_unsigned(u_int32_t, *tl++);
3456 numops = fxdr_unsigned(int, *tl);
3458 * Loop around doing the sub ops.
3460 for (i = 0; i < numops; i++) {
3461 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3462 NFSM_BUILD(repp, u_int32_t *, 2 * NFSX_UNSIGNED);
3464 op = fxdr_unsigned(int, *tl);
3465 if (op < NFSV4OP_CBGETATTR ||
3466 (op > NFSV4OP_CBRECALL && minorvers == NFSV4_MINORVERSION) ||
3467 (op > NFSV4OP_CBNOTIFYDEVID &&
3468 minorvers == NFSV41_MINORVERSION) ||
3469 (op > NFSV4OP_CBOFFLOAD &&
3470 minorvers == NFSV42_MINORVERSION)) {
3471 nd->nd_repstat = NFSERR_OPILLEGAL;
3472 *repp = nfscl_errmap(nd, minorvers);
3476 nd->nd_procnum = op;
3477 if (op < NFSV42_CBNOPS)
3478 nfsstatsv1.cbrpccnt[nd->nd_procnum]++;
3480 case NFSV4OP_CBGETATTR:
3481 NFSCL_DEBUG(4, "cbgetattr\n");
3484 error = nfsm_getfh(nd, &nfhp);
3486 error = nfsrv_getattrbits(nd, &attrbits,
3488 if (error == 0 && i == 0 &&
3489 minorvers != NFSV4_MINORVERSION)
3490 error = NFSERR_OPNOTINSESS;
3492 mp = nfscl_getmnt(minorvers, sessionid, cbident,
3495 error = NFSERR_SERVERFAULT;
3498 error = nfscl_ngetreopen(mp, nfhp->nfh_fh,
3499 nfhp->nfh_len, p, &np);
3504 NFSZERO_ATTRBIT(&rattrbits);
3506 dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
3509 if (NFSISSET_ATTRBIT(&attrbits,
3512 va.va_size = np->n_size;
3516 NFSSETBIT_ATTRBIT(&rattrbits,
3519 if (NFSISSET_ATTRBIT(&attrbits,
3520 NFSATTRBIT_CHANGE)) {
3524 (np->n_flag & NDELEGMOD))
3526 NFSSETBIT_ATTRBIT(&rattrbits,
3530 error = NFSERR_SERVERFAULT;
3538 free(nfhp, M_NFSFH);
3540 (void) nfsv4_fillattr(nd, NULL, NULL, NULL, &va,
3541 NULL, 0, &rattrbits, NULL, p, 0, 0, 0, 0,
3544 case NFSV4OP_CBRECALL:
3545 NFSCL_DEBUG(4, "cbrecall\n");
3546 NFSM_DISSECT(tl, u_int32_t *, NFSX_STATEID +
3548 stateid.seqid = *tl++;
3549 NFSBCOPY((caddr_t)tl, (caddr_t)stateid.other,
3551 tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
3552 trunc = fxdr_unsigned(int, *tl);
3553 error = nfsm_getfh(nd, &nfhp);
3554 if (error == 0 && i == 0 &&
3555 minorvers != NFSV4_MINORVERSION)
3556 error = NFSERR_OPNOTINSESS;
3559 if (minorvers == NFSV4_MINORVERSION)
3560 clp = nfscl_getclnt(cbident);
3562 clp = nfscl_getclntsess(sessionid);
3564 dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
3566 if (dp != NULL && (dp->nfsdl_flags &
3567 NFSCLDL_DELEGRET) == 0) {
3570 wakeup((caddr_t)clp);
3573 error = NFSERR_SERVERFAULT;
3578 free(nfhp, M_NFSFH);
3580 case NFSV4OP_CBLAYOUTRECALL:
3581 NFSCL_DEBUG(4, "cblayrec\n");
3583 NFSM_DISSECT(tl, uint32_t *, 4 * NFSX_UNSIGNED);
3584 laytype = fxdr_unsigned(int, *tl++);
3585 iomode = fxdr_unsigned(uint32_t, *tl++);
3586 if (newnfs_true == *tl++)
3590 recalltype = fxdr_unsigned(int, *tl);
3591 NFSCL_DEBUG(4, "layt=%d iom=%d ch=%d rectyp=%d\n",
3592 laytype, iomode, changed, recalltype);
3593 recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL,
3595 if (laytype != NFSLAYOUT_NFSV4_1_FILES &&
3596 laytype != NFSLAYOUT_FLEXFILE)
3597 error = NFSERR_NOMATCHLAYOUT;
3598 else if (recalltype == NFSLAYOUTRETURN_FILE) {
3599 error = nfsm_getfh(nd, &nfhp);
3600 NFSCL_DEBUG(4, "retfile getfh=%d\n", error);
3603 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_HYPER +
3605 off = fxdr_hyper(tl); tl += 2;
3606 len = fxdr_hyper(tl); tl += 2;
3607 stateid.seqid = fxdr_unsigned(uint32_t, *tl++);
3608 NFSBCOPY(tl, stateid.other, NFSX_STATEIDOTHER);
3609 if (minorvers == NFSV4_MINORVERSION)
3610 error = NFSERR_NOTSUPP;
3612 error = NFSERR_OPNOTINSESS;
3613 NFSCL_DEBUG(4, "off=%ju len=%ju sq=%u err=%d\n",
3614 (uintmax_t)off, (uintmax_t)len,
3615 stateid.seqid, error);
3618 clp = nfscl_getclntsess(sessionid);
3619 NFSCL_DEBUG(4, "cbly clp=%p\n", clp);
3621 lyp = nfscl_findlayout(clp,
3624 NFSCL_DEBUG(4, "cblyp=%p\n",
3629 NFSLY_FLEXFILE)) != 0 &&
3630 !NFSBCMP(stateid.other,
3631 lyp->nfsly_stateid.other,
3632 NFSX_STATEIDOTHER)) {
3642 lyp->nfsly_stateid.seqid)
3643 lyp->nfsly_stateid.seqid =
3651 lyp->nfsly_stateid.seqid);
3654 NFSERR_NOMATCHLAYOUT;
3656 error = NFSERR_NOMATCHLAYOUT;
3659 free(nfhp, M_NFSFH);
3660 } else if (recalltype == NFSLAYOUTRETURN_FSID) {
3661 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_HYPER);
3662 filesid[0] = fxdr_hyper(tl); tl += 2;
3663 filesid[1] = fxdr_hyper(tl); tl += 2;
3666 clp = nfscl_getclntsess(sessionid);
3668 TAILQ_FOREACH(lyp, &clp->nfsc_layout,
3670 if (lyp->nfsly_filesid[0] ==
3672 lyp->nfsly_filesid[1] ==
3679 lyp->nfsly_stateid.seqid,
3689 error = NFSERR_NOMATCHLAYOUT;
3691 error = NFSERR_NOMATCHLAYOUT;
3693 } else if (recalltype == NFSLAYOUTRETURN_ALL) {
3696 clp = nfscl_getclntsess(sessionid);
3698 TAILQ_FOREACH(lyp, &clp->nfsc_layout,
3700 error = nfscl_layoutrecall(
3701 recalltype, lyp, iomode, 0,
3703 lyp->nfsly_stateid.seqid,
3704 0, 0, NULL, recallp);
3711 error = NFSERR_NOMATCHLAYOUT;
3713 error = NFSERR_NOMATCHLAYOUT;
3716 error = NFSERR_NOMATCHLAYOUT;
3717 if (recallp != NULL) {
3718 free(recallp, M_NFSLAYRECALL);
3722 case NFSV4OP_CBSEQUENCE:
3723 NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID +
3725 bcopy(tl, sessionid, NFSX_V4SESSIONID);
3726 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3727 seqid = fxdr_unsigned(uint32_t, *tl++);
3728 slotid = fxdr_unsigned(uint32_t, *tl++);
3729 highslot = fxdr_unsigned(uint32_t, *tl++);
3731 /* Throw away the referring call stuff. */
3732 clist = fxdr_unsigned(int, *tl);
3733 for (j = 0; j < clist; j++) {
3734 NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID +
3736 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3737 rcalls = fxdr_unsigned(int, *tl);
3738 for (k = 0; k < rcalls; k++) {
3739 NFSM_DISSECT(tl, uint32_t *,
3745 clp = nfscl_getclntsess(sessionid);
3747 error = NFSERR_SERVERFAULT;
3749 error = NFSERR_SEQUENCEPOS;
3751 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3752 error = nfsv4_seqsession(seqid, slotid,
3753 highslot, tsep->nfsess_cbslots, &rep,
3754 tsep->nfsess_backslots);
3757 if (error == 0 || error == NFSERR_REPLYFROMCACHE) {
3761 * Handle a reply for a retried
3762 * callback. The reply will be
3763 * re-inserted in the session cache
3764 * by the nfsv4_seqsess_cacherep() call
3767 KASSERT(error == NFSERR_REPLYFROMCACHE,
3768 ("cbsequence: non-NULL rep"));
3769 NFSCL_DEBUG(4, "Got cbretry\n");
3770 m_freem(nd->nd_mreq);
3775 NFSM_BUILD(tl, uint32_t *,
3776 NFSX_V4SESSIONID + 4 * NFSX_UNSIGNED);
3777 bcopy(sessionid, tl, NFSX_V4SESSIONID);
3778 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3779 *tl++ = txdr_unsigned(seqid);
3780 *tl++ = txdr_unsigned(slotid);
3781 *tl++ = txdr_unsigned(NFSV4_CBSLOTS - 1);
3782 *tl = txdr_unsigned(NFSV4_CBSLOTS - 1);
3786 if (i == 0 && minorvers != NFSV4_MINORVERSION)
3787 error = NFSERR_OPNOTINSESS;
3789 NFSCL_DEBUG(1, "unsupp callback %d\n", op);
3790 error = NFSERR_NOTSUPP;
3795 if (error == EBADRPC || error == NFSERR_BADXDR) {
3796 nd->nd_repstat = NFSERR_BADXDR;
3798 nd->nd_repstat = error;
3803 if (nd->nd_repstat) {
3804 *repp = nfscl_errmap(nd, minorvers);
3807 *repp = 0; /* NFS4_OK */
3810 if (recallp != NULL)
3811 free(recallp, M_NFSLAYRECALL);
3813 if (error == EBADRPC || error == NFSERR_BADXDR)
3814 nd->nd_repstat = NFSERR_BADXDR;
3816 printf("nfsv4 comperr1=%d\n", error);
3819 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
3823 *retopsp = txdr_unsigned(retops);
3825 *nd->nd_errp = nfscl_errmap(nd, minorvers);
3827 if (gotseq_ok != 0) {
3828 rep = m_copym(nd->nd_mreq, 0, M_COPYALL, M_WAITOK);
3830 clp = nfscl_getclntsess(sessionid);
3832 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3833 nfsv4_seqsess_cacherep(slotid, tsep->nfsess_cbslots,
3844 * Generate the next cbident value. Basically just increment a static value
3845 * and then check that it isn't already in the list, if it has wrapped around.
3848 nfscl_nextcbident(void)
3850 struct nfsclclient *clp;
3852 static u_int32_t nextcbident = 0;
3853 static int haswrapped = 0;
3856 if (nextcbident == 0)
3860 * Search the clientid list for one already using this cbident.
3865 LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3866 if (clp->nfsc_cbident == nextcbident) {
3876 return (nextcbident);
3880 * Get the mount point related to a given cbident or session and busy it.
3883 nfscl_getmnt(int minorvers, uint8_t *sessionid, u_int32_t cbident,
3884 struct nfsclclient **clpp)
3886 struct nfsclclient *clp;
3889 struct nfsclsession *tsep;
3893 LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3894 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3895 if (minorvers == NFSV4_MINORVERSION) {
3896 if (clp->nfsc_cbident == cbident)
3898 } else if (!NFSBCMP(tsep->nfsess_sessionid, sessionid,
3906 mp = clp->nfsc_nmp->nm_mountp;
3909 error = vfs_busy(mp, 0);
3918 * Get the clientid pointer related to a given cbident.
3920 static struct nfsclclient *
3921 nfscl_getclnt(u_int32_t cbident)
3923 struct nfsclclient *clp;
3925 LIST_FOREACH(clp, &nfsclhead, nfsc_list)
3926 if (clp->nfsc_cbident == cbident)
3932 * Get the clientid pointer related to a given sessionid.
3934 static struct nfsclclient *
3935 nfscl_getclntsess(uint8_t *sessionid)
3937 struct nfsclclient *clp;
3938 struct nfsclsession *tsep;
3940 LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3941 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3942 if (!NFSBCMP(tsep->nfsess_sessionid, sessionid,
3950 * Search for a lock conflict locally on the client. A conflict occurs if
3951 * - not same owner and overlapping byte range and at least one of them is
3952 * a write lock or this is an unlock.
3955 nfscl_localconflict(struct nfsclclient *clp, u_int8_t *fhp, int fhlen,
3956 struct nfscllock *nlop, u_int8_t *own, struct nfscldeleg *dp,
3957 struct nfscllock **lopp)
3959 struct nfsclowner *owp;
3960 struct nfsclopen *op;
3964 ret = nfscl_checkconflict(&dp->nfsdl_lock, nlop, own, lopp);
3968 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3969 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3970 if (op->nfso_fhlen == fhlen &&
3971 !NFSBCMP(op->nfso_fh, fhp, fhlen)) {
3972 ret = nfscl_checkconflict(&op->nfso_lock, nlop,
3983 nfscl_checkconflict(struct nfscllockownerhead *lhp, struct nfscllock *nlop,
3984 u_int8_t *own, struct nfscllock **lopp)
3986 struct nfscllockowner *lp;
3987 struct nfscllock *lop;
3989 LIST_FOREACH(lp, lhp, nfsl_list) {
3990 if (NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
3991 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
3992 if (lop->nfslo_first >= nlop->nfslo_end)
3994 if (lop->nfslo_end <= nlop->nfslo_first)
3996 if (lop->nfslo_type == F_WRLCK ||
3997 nlop->nfslo_type == F_WRLCK ||
3998 nlop->nfslo_type == F_UNLCK) {
4001 return (NFSERR_DENIED);
4010 * Check for a local conflicting lock.
4013 nfscl_lockt(vnode_t vp, struct nfsclclient *clp, u_int64_t off,
4014 u_int64_t len, struct flock *fl, NFSPROC_T *p, void *id, int flags)
4016 struct nfscllock *lop, nlck;
4017 struct nfscldeleg *dp;
4019 u_int8_t own[NFSV4CL_LOCKNAMELEN];
4022 nlck.nfslo_type = fl->l_type;
4023 nlck.nfslo_first = off;
4024 if (len == NFS64BITSSET) {
4025 nlck.nfslo_end = NFS64BITSSET;
4027 nlck.nfslo_end = off + len;
4028 if (nlck.nfslo_end <= nlck.nfslo_first)
4029 return (NFSERR_INVAL);
4032 nfscl_filllockowner(id, own, flags);
4034 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4035 error = nfscl_localconflict(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len,
4036 &nlck, own, dp, &lop);
4038 fl->l_whence = SEEK_SET;
4039 fl->l_start = lop->nfslo_first;
4040 if (lop->nfslo_end == NFS64BITSSET)
4043 fl->l_len = lop->nfslo_end - lop->nfslo_first;
4044 fl->l_pid = (pid_t)0;
4045 fl->l_type = lop->nfslo_type;
4046 error = -1; /* no RPC required */
4047 } else if (dp != NULL && ((dp->nfsdl_flags & NFSCLDL_WRITE) ||
4048 fl->l_type == F_RDLCK)) {
4050 * The delegation ensures that there isn't a conflicting
4051 * lock on the server, so return -1 to indicate an RPC
4054 fl->l_type = F_UNLCK;
4062 * Handle Recall of a delegation.
4063 * The clp must be exclusive locked when this is called.
4066 nfscl_recalldeleg(struct nfsclclient *clp, struct nfsmount *nmp,
4067 struct nfscldeleg *dp, vnode_t vp, struct ucred *cred, NFSPROC_T *p,
4068 int called_from_renewthread, vnode_t *vpp)
4070 struct nfsclowner *owp, *lowp, *nowp;
4071 struct nfsclopen *op, *lop;
4072 struct nfscllockowner *lp;
4073 struct nfscllock *lckp;
4078 KASSERT(vpp != NULL, ("nfscl_recalldeleg: vpp NULL"));
4081 * First, get a vnode for the file. This is needed to do RPCs.
4083 ret = nfscl_ngetreopen(nmp->nm_mountp, dp->nfsdl_fh,
4084 dp->nfsdl_fhlen, p, &np);
4087 * File isn't open, so nothing to move over to the
4097 dp->nfsdl_flags &= ~NFSCLDL_MODTIMESET;
4100 * Ok, if it's a write delegation, flush data to the server, so
4101 * that close/open consistency is retained.
4105 if ((dp->nfsdl_flags & NFSCLDL_WRITE) && (np->n_flag & NMODIFIED)) {
4106 np->n_flag |= NDELEGRECALL;
4108 ret = ncl_flush(vp, MNT_WAIT, p, 1, called_from_renewthread);
4110 np->n_flag &= ~NDELEGRECALL;
4112 NFSINVALATTRCACHE(np);
4114 if (ret == EIO && called_from_renewthread != 0) {
4116 * If the flush failed with EIO for the renew thread,
4117 * return now, so that the dirty buffer will be flushed
4124 * Now, for each openowner with opens issued locally, move them
4125 * over to state against the server.
4127 LIST_FOREACH(lowp, &dp->nfsdl_owner, nfsow_list) {
4128 lop = LIST_FIRST(&lowp->nfsow_open);
4130 if (LIST_NEXT(lop, nfso_list) != NULL)
4131 panic("nfsdlg mult opens");
4133 * Look for the same openowner against the server.
4135 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
4136 if (!NFSBCMP(lowp->nfsow_owner,
4137 owp->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
4138 newnfs_copycred(&dp->nfsdl_cred, cred);
4139 ret = nfscl_moveopen(vp, clp, nmp, lop,
4141 if (ret == NFSERR_STALECLIENTID ||
4142 ret == NFSERR_STALEDONTRECOVER ||
4143 ret == NFSERR_BADSESSION)
4146 nfscl_freeopen(lop, 1);
4155 * If no openowner found, create one and get an open
4160 sizeof (struct nfsclowner), M_NFSCLOWNER,
4162 nfscl_newopen(clp, NULL, &owp, &nowp, &op,
4163 NULL, lowp->nfsow_owner, dp->nfsdl_fh,
4164 dp->nfsdl_fhlen, NULL, NULL);
4165 newnfs_copycred(&dp->nfsdl_cred, cred);
4166 ret = nfscl_moveopen(vp, clp, nmp, lop,
4169 nfscl_freeopenowner(owp, 0);
4170 if (ret == NFSERR_STALECLIENTID ||
4171 ret == NFSERR_STALEDONTRECOVER ||
4172 ret == NFSERR_BADSESSION)
4175 nfscl_freeopen(lop, 1);
4185 * Now, get byte range locks for any locks done locally.
4187 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4188 LIST_FOREACH(lckp, &lp->nfsl_lock, nfslo_list) {
4189 newnfs_copycred(&dp->nfsdl_cred, cred);
4190 ret = nfscl_relock(vp, clp, nmp, lp, lckp, cred, p);
4191 if (ret == NFSERR_STALESTATEID ||
4192 ret == NFSERR_STALEDONTRECOVER ||
4193 ret == NFSERR_STALECLIENTID ||
4194 ret == NFSERR_BADSESSION)
4204 * Move a locally issued open over to an owner on the state list.
4205 * SIDE EFFECT: If it needs to sleep (do an rpc), it unlocks clstate and
4206 * returns with it unlocked.
4209 nfscl_moveopen(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp,
4210 struct nfsclopen *lop, struct nfsclowner *owp, struct nfscldeleg *dp,
4211 struct ucred *cred, NFSPROC_T *p)
4213 struct nfsclopen *op, *nop;
4214 struct nfscldeleg *ndp;
4216 int error = 0, newone;
4219 * First, look for an appropriate open, If found, just increment the
4222 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
4223 if ((op->nfso_mode & lop->nfso_mode) == lop->nfso_mode &&
4224 op->nfso_fhlen == lop->nfso_fhlen &&
4225 !NFSBCMP(op->nfso_fh, lop->nfso_fh, op->nfso_fhlen)) {
4226 op->nfso_opencnt += lop->nfso_opencnt;
4227 nfscl_freeopen(lop, 1);
4232 /* No appropriate open, so we have to do one against the server. */
4234 nop = malloc(sizeof (struct nfsclopen) +
4235 lop->nfso_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
4236 nop->nfso_hash.le_prev = NULL;
4238 nfscl_newopen(clp, NULL, &owp, NULL, &op, &nop, owp->nfsow_owner,
4239 lop->nfso_fh, lop->nfso_fhlen, cred, &newone);
4241 error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data, np->n_v4->n4_fhlen,
4242 lop->nfso_fh, lop->nfso_fhlen, lop->nfso_mode, op,
4243 NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, &ndp, 0, 0, cred, p);
4246 nfscl_freeopen(op, 0);
4248 op->nfso_mode |= lop->nfso_mode;
4249 op->nfso_opencnt += lop->nfso_opencnt;
4250 nfscl_freeopen(lop, 1);
4253 free(nop, M_NFSCLOPEN);
4256 * What should I do with the returned delegation, since the
4257 * delegation is being recalled? For now, just printf and
4260 printf("Moveopen returned deleg\n");
4261 free(ndp, M_NFSCLDELEG);
4267 * Recall all delegations on this client.
4270 nfscl_totalrecall(struct nfsclclient *clp)
4272 struct nfscldeleg *dp;
4274 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
4275 if ((dp->nfsdl_flags & NFSCLDL_DELEGRET) == 0)
4276 dp->nfsdl_flags |= NFSCLDL_RECALL;
4281 * Relock byte ranges. Called for delegation recall and state expiry.
4284 nfscl_relock(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp,
4285 struct nfscllockowner *lp, struct nfscllock *lop, struct ucred *cred,
4288 struct nfscllockowner *nlp;
4291 int error, newone, donelocally;
4293 off = lop->nfslo_first;
4294 len = lop->nfslo_end - lop->nfslo_first;
4295 error = nfscl_getbytelock(vp, off, len, lop->nfslo_type, cred, p,
4296 clp, 1, NULL, lp->nfsl_lockflags, lp->nfsl_owner,
4297 lp->nfsl_openowner, &nlp, &newone, &donelocally);
4298 if (error || donelocally)
4300 nfhp = VTONFS(vp)->n_fhp;
4301 error = nfscl_trylock(nmp, vp, nfhp->nfh_fh,
4302 nfhp->nfh_len, nlp, newone, 0, off,
4303 len, lop->nfslo_type, cred, p);
4305 nfscl_freelockowner(nlp, 0);
4310 * Called to re-open a file. Basically get a vnode for the file handle
4311 * and then call nfsrpc_openrpc() to do the rest.
4314 nfsrpc_reopen(struct nfsmount *nmp, u_int8_t *fhp, int fhlen,
4315 u_int32_t mode, struct nfsclopen *op, struct nfscldeleg **dpp,
4316 struct ucred *cred, NFSPROC_T *p)
4322 error = nfscl_ngetreopen(nmp->nm_mountp, fhp, fhlen, p, &np);
4326 if (np->n_v4 != NULL) {
4327 error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data,
4328 np->n_v4->n4_fhlen, fhp, fhlen, mode, op,
4329 NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, dpp, 0, 0,
4339 * Try an open against the server. Just call nfsrpc_openrpc(), retrying while
4340 * NFSERR_DELAY. Also, try system credentials, if the passed in credentials
4344 nfscl_tryopen(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen,
4345 u_int8_t *newfhp, int newfhlen, u_int32_t mode, struct nfsclopen *op,
4346 u_int8_t *name, int namelen, struct nfscldeleg **ndpp,
4347 int reclaim, u_int32_t delegtype, struct ucred *cred, NFSPROC_T *p)
4352 error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp, newfhlen,
4353 mode, op, name, namelen, ndpp, reclaim, delegtype, cred, p,
4355 if (error == NFSERR_DELAY)
4356 (void) nfs_catnap(PZERO, error, "nfstryop");
4357 } while (error == NFSERR_DELAY);
4358 if (error == EAUTH || error == EACCES) {
4359 /* Try again using system credentials */
4360 newnfs_setroot(cred);
4362 error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp,
4363 newfhlen, mode, op, name, namelen, ndpp, reclaim,
4364 delegtype, cred, p, 1, 0);
4365 if (error == NFSERR_DELAY)
4366 (void) nfs_catnap(PZERO, error, "nfstryop");
4367 } while (error == NFSERR_DELAY);
4373 * Try a byte range lock. Just loop on nfsrpc_lock() while it returns
4374 * NFSERR_DELAY. Also, retry with system credentials, if the provided
4378 nfscl_trylock(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp,
4379 int fhlen, struct nfscllockowner *nlp, int newone, int reclaim,
4380 u_int64_t off, u_int64_t len, short type, struct ucred *cred, NFSPROC_T *p)
4382 struct nfsrv_descript nfsd, *nd = &nfsd;
4386 error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp, newone,
4387 reclaim, off, len, type, cred, p, 0);
4388 if (!error && nd->nd_repstat == NFSERR_DELAY)
4389 (void) nfs_catnap(PZERO, (int)nd->nd_repstat,
4391 } while (!error && nd->nd_repstat == NFSERR_DELAY);
4393 error = nd->nd_repstat;
4394 if (error == EAUTH || error == EACCES) {
4395 /* Try again using root credentials */
4396 newnfs_setroot(cred);
4398 error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp,
4399 newone, reclaim, off, len, type, cred, p, 1);
4400 if (!error && nd->nd_repstat == NFSERR_DELAY)
4401 (void) nfs_catnap(PZERO, (int)nd->nd_repstat,
4403 } while (!error && nd->nd_repstat == NFSERR_DELAY);
4405 error = nd->nd_repstat;
4411 * Try a delegreturn against the server. Just call nfsrpc_delegreturn(),
4412 * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
4416 nfscl_trydelegreturn(struct nfscldeleg *dp, struct ucred *cred,
4417 struct nfsmount *nmp, NFSPROC_T *p)
4422 error = nfsrpc_delegreturn(dp, cred, nmp, p, 0);
4423 if (error == NFSERR_DELAY)
4424 (void) nfs_catnap(PZERO, error, "nfstrydp");
4425 } while (error == NFSERR_DELAY);
4426 if (error == EAUTH || error == EACCES) {
4427 /* Try again using system credentials */
4428 newnfs_setroot(cred);
4430 error = nfsrpc_delegreturn(dp, cred, nmp, p, 1);
4431 if (error == NFSERR_DELAY)
4432 (void) nfs_catnap(PZERO, error, "nfstrydp");
4433 } while (error == NFSERR_DELAY);
4439 * Try a close against the server. Just call nfsrpc_closerpc(),
4440 * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
4444 nfscl_tryclose(struct nfsclopen *op, struct ucred *cred,
4445 struct nfsmount *nmp, NFSPROC_T *p)
4447 struct nfsrv_descript nfsd, *nd = &nfsd;
4451 error = nfsrpc_closerpc(nd, nmp, op, cred, p, 0);
4452 if (error == NFSERR_DELAY)
4453 (void) nfs_catnap(PZERO, error, "nfstrycl");
4454 } while (error == NFSERR_DELAY);
4455 if (error == EAUTH || error == EACCES) {
4456 /* Try again using system credentials */
4457 newnfs_setroot(cred);
4459 error = nfsrpc_closerpc(nd, nmp, op, cred, p, 1);
4460 if (error == NFSERR_DELAY)
4461 (void) nfs_catnap(PZERO, error, "nfstrycl");
4462 } while (error == NFSERR_DELAY);
4468 * Decide if a delegation on a file permits close without flushing writes
4469 * to the server. This might be a big performance win in some environments.
4470 * (Not useful until the client does caching on local stable storage.)
4473 nfscl_mustflush(vnode_t vp)
4475 struct nfsclclient *clp;
4476 struct nfscldeleg *dp;
4478 struct nfsmount *nmp;
4481 nmp = VFSTONFS(vp->v_mount);
4482 if (!NFSHASNFSV4(nmp))
4485 clp = nfscl_findcl(nmp);
4490 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4491 if (dp != NULL && (dp->nfsdl_flags &
4492 (NFSCLDL_WRITE | NFSCLDL_RECALL | NFSCLDL_DELEGRET)) ==
4494 (dp->nfsdl_sizelimit >= np->n_size ||
4495 !NFSHASSTRICT3530(nmp))) {
4504 * See if a (write) delegation exists for this file.
4507 nfscl_nodeleg(vnode_t vp, int writedeleg)
4509 struct nfsclclient *clp;
4510 struct nfscldeleg *dp;
4512 struct nfsmount *nmp;
4515 nmp = VFSTONFS(vp->v_mount);
4516 if (!NFSHASNFSV4(nmp))
4519 clp = nfscl_findcl(nmp);
4524 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4526 (dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_DELEGRET)) == 0 &&
4527 (writedeleg == 0 || (dp->nfsdl_flags & NFSCLDL_WRITE) ==
4537 * Look for an associated delegation that should be DelegReturned.
4540 nfscl_removedeleg(vnode_t vp, NFSPROC_T *p, nfsv4stateid_t *stp)
4542 struct nfsclclient *clp;
4543 struct nfscldeleg *dp;
4544 struct nfsclowner *owp;
4545 struct nfscllockowner *lp;
4546 struct nfsmount *nmp;
4549 int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
4551 nmp = VFSTONFS(vp->v_mount);
4555 * Loop around waiting for:
4556 * - outstanding I/O operations on delegations to complete
4557 * - for a delegation on vp that has state, lock the client and
4559 * - return delegation with no state
4562 clp = nfscl_findcl(nmp);
4567 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4568 np->n_fhp->nfh_len);
4571 * Wait for outstanding I/O ops to be done.
4573 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4575 nfsv4_unlock(&clp->nfsc_lock, 0);
4578 dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4579 (void) nfsmsleep(&dp->nfsdl_rwlock,
4580 NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4584 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4585 if (!LIST_EMPTY(&owp->nfsow_open)) {
4591 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4592 if (!LIST_EMPTY(&lp->nfsl_lock)) {
4598 if (needsrecall && !triedrecall) {
4599 dp->nfsdl_flags |= NFSCLDL_DELEGRET;
4602 igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
4603 &islept, NFSCLSTATEMUTEXPTR, NULL);
4610 cred = newnfs_getcred();
4611 newnfs_copycred(&dp->nfsdl_cred, cred);
4612 nfscl_recalldeleg(clp, nmp, dp, vp, cred, p, 0, NULL);
4616 nfsv4_unlock(&clp->nfsc_lock, 0);
4620 *stp = dp->nfsdl_stateid;
4622 nfscl_cleandeleg(dp);
4623 nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
4626 nfsv4_unlock(&clp->nfsc_lock, 0);
4633 * Look for associated delegation(s) that should be DelegReturned.
4636 nfscl_renamedeleg(vnode_t fvp, nfsv4stateid_t *fstp, int *gotfdp, vnode_t tvp,
4637 nfsv4stateid_t *tstp, int *gottdp, NFSPROC_T *p)
4639 struct nfsclclient *clp;
4640 struct nfscldeleg *dp;
4641 struct nfsclowner *owp;
4642 struct nfscllockowner *lp;
4643 struct nfsmount *nmp;
4646 int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
4648 nmp = VFSTONFS(fvp->v_mount);
4653 * Loop around waiting for:
4654 * - outstanding I/O operations on delegations to complete
4655 * - for a delegation on fvp that has state, lock the client and
4657 * - return delegation(s) with no state.
4660 clp = nfscl_findcl(nmp);
4666 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4667 np->n_fhp->nfh_len);
4668 if (dp != NULL && *gotfdp == 0) {
4670 * Wait for outstanding I/O ops to be done.
4672 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4674 nfsv4_unlock(&clp->nfsc_lock, 0);
4677 dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4678 (void) nfsmsleep(&dp->nfsdl_rwlock,
4679 NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4683 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4684 if (!LIST_EMPTY(&owp->nfsow_open)) {
4690 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4691 if (!LIST_EMPTY(&lp->nfsl_lock)) {
4697 if (needsrecall && !triedrecall) {
4698 dp->nfsdl_flags |= NFSCLDL_DELEGRET;
4701 igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
4702 &islept, NFSCLSTATEMUTEXPTR, NULL);
4709 cred = newnfs_getcred();
4710 newnfs_copycred(&dp->nfsdl_cred, cred);
4711 nfscl_recalldeleg(clp, nmp, dp, fvp, cred, p, 0, NULL);
4715 nfsv4_unlock(&clp->nfsc_lock, 0);
4719 *fstp = dp->nfsdl_stateid;
4722 nfscl_cleandeleg(dp);
4723 nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
4726 nfsv4_unlock(&clp->nfsc_lock, 0);
4731 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4732 np->n_fhp->nfh_len);
4733 if (dp != NULL && *gottdp == 0) {
4735 * Wait for outstanding I/O ops to be done.
4737 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4738 dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4739 (void) nfsmsleep(&dp->nfsdl_rwlock,
4740 NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4743 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4744 if (!LIST_EMPTY(&owp->nfsow_open)) {
4749 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4750 if (!LIST_EMPTY(&lp->nfsl_lock)) {
4755 *tstp = dp->nfsdl_stateid;
4758 nfscl_cleandeleg(dp);
4759 nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
4768 * Get a reference on the clientid associated with the mount point.
4769 * Return 1 if success, 0 otherwise.
4772 nfscl_getref(struct nfsmount *nmp)
4774 struct nfsclclient *clp;
4777 clp = nfscl_findcl(nmp);
4782 nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, NULL);
4788 * Release a reference on a clientid acquired with the above call.
4791 nfscl_relref(struct nfsmount *nmp)
4793 struct nfsclclient *clp;
4796 clp = nfscl_findcl(nmp);
4801 nfsv4_relref(&clp->nfsc_lock);
4806 * Save the size attribute in the delegation, since the nfsnode
4810 nfscl_reclaimnode(vnode_t vp)
4812 struct nfsclclient *clp;
4813 struct nfscldeleg *dp;
4814 struct nfsnode *np = VTONFS(vp);
4815 struct nfsmount *nmp;
4817 nmp = VFSTONFS(vp->v_mount);
4818 if (!NFSHASNFSV4(nmp))
4821 clp = nfscl_findcl(nmp);
4826 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4827 if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE))
4828 dp->nfsdl_size = np->n_size;
4833 * Get the saved size attribute in the delegation, since it is a
4834 * newly allocated nfsnode.
4837 nfscl_newnode(vnode_t vp)
4839 struct nfsclclient *clp;
4840 struct nfscldeleg *dp;
4841 struct nfsnode *np = VTONFS(vp);
4842 struct nfsmount *nmp;
4844 nmp = VFSTONFS(vp->v_mount);
4845 if (!NFSHASNFSV4(nmp))
4848 clp = nfscl_findcl(nmp);
4853 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4854 if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE))
4855 np->n_size = dp->nfsdl_size;
4860 * If there is a valid write delegation for this file, set the modtime
4861 * to the local clock time.
4864 nfscl_delegmodtime(vnode_t vp)
4866 struct nfsclclient *clp;
4867 struct nfscldeleg *dp;
4868 struct nfsnode *np = VTONFS(vp);
4869 struct nfsmount *nmp;
4871 nmp = VFSTONFS(vp->v_mount);
4872 if (!NFSHASNFSV4(nmp))
4875 clp = nfscl_findcl(nmp);
4880 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4881 if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE)) {
4882 nanotime(&dp->nfsdl_modtime);
4883 dp->nfsdl_flags |= NFSCLDL_MODTIMESET;
4889 * If there is a valid write delegation for this file with a modtime set,
4890 * put that modtime in mtime.
4893 nfscl_deleggetmodtime(vnode_t vp, struct timespec *mtime)
4895 struct nfsclclient *clp;
4896 struct nfscldeleg *dp;
4897 struct nfsnode *np = VTONFS(vp);
4898 struct nfsmount *nmp;
4900 nmp = VFSTONFS(vp->v_mount);
4901 if (!NFSHASNFSV4(nmp))
4904 clp = nfscl_findcl(nmp);
4909 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4911 (dp->nfsdl_flags & (NFSCLDL_WRITE | NFSCLDL_MODTIMESET)) ==
4912 (NFSCLDL_WRITE | NFSCLDL_MODTIMESET))
4913 *mtime = dp->nfsdl_modtime;
4918 nfscl_errmap(struct nfsrv_descript *nd, u_int32_t minorvers)
4920 short *defaulterrp, *errp;
4922 if (!nd->nd_repstat)
4924 if (nd->nd_procnum == NFSPROC_NOOP)
4925 return (txdr_unsigned(nd->nd_repstat & 0xffff));
4926 if (nd->nd_repstat == EBADRPC)
4927 return (txdr_unsigned(NFSERR_BADXDR));
4928 if (nd->nd_repstat == NFSERR_MINORVERMISMATCH ||
4929 nd->nd_repstat == NFSERR_OPILLEGAL)
4930 return (txdr_unsigned(nd->nd_repstat));
4931 if (nd->nd_repstat >= NFSERR_BADIOMODE && nd->nd_repstat < 20000 &&
4932 minorvers > NFSV4_MINORVERSION) {
4933 /* NFSv4.n error. */
4934 return (txdr_unsigned(nd->nd_repstat));
4936 if (nd->nd_procnum < NFSV4OP_CBNOPS)
4937 errp = defaulterrp = nfscl_cberrmap[nd->nd_procnum];
4939 return (txdr_unsigned(nd->nd_repstat));
4941 if (*errp == (short)nd->nd_repstat)
4942 return (txdr_unsigned(nd->nd_repstat));
4943 return (txdr_unsigned(*defaulterrp));
4947 * Called to find/add a layout to a client.
4948 * This function returns the layout with a refcnt (shared lock) upon
4949 * success (returns 0) or with no lock/refcnt on the layout when an
4950 * error is returned.
4951 * If a layout is passed in via lypp, it is locked (exclusively locked).
4954 nfscl_layout(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen,
4955 nfsv4stateid_t *stateidp, int layouttype, int retonclose,
4956 struct nfsclflayouthead *fhlp, struct nfscllayout **lypp,
4957 struct ucred *cred, NFSPROC_T *p)
4959 struct nfsclclient *clp;
4960 struct nfscllayout *lyp, *tlyp;
4961 struct nfsclflayout *flp;
4962 struct nfsnode *np = VTONFS(vp);
4964 int layout_passed_in;
4966 mp = nmp->nm_mountp;
4967 layout_passed_in = 1;
4971 layout_passed_in = 0;
4972 tlyp = malloc(sizeof(*tlyp) + fhlen - 1, M_NFSLAYOUT,
4979 if (layout_passed_in != 0)
4980 nfsv4_unlock(&lyp->nfsly_lock, 0);
4983 free(tlyp, M_NFSLAYOUT);
4988 * Although no lyp was passed in, another thread might have
4989 * allocated one. If one is found, just increment it's ref
4990 * count and return it.
4992 lyp = nfscl_findlayout(clp, fhp, fhlen);
4996 lyp->nfsly_stateid.seqid = stateidp->seqid;
4997 lyp->nfsly_stateid.other[0] = stateidp->other[0];
4998 lyp->nfsly_stateid.other[1] = stateidp->other[1];
4999 lyp->nfsly_stateid.other[2] = stateidp->other[2];
5000 lyp->nfsly_lastbyte = 0;
5001 LIST_INIT(&lyp->nfsly_flayread);
5002 LIST_INIT(&lyp->nfsly_flayrw);
5003 LIST_INIT(&lyp->nfsly_recall);
5004 lyp->nfsly_filesid[0] = np->n_vattr.na_filesid[0];
5005 lyp->nfsly_filesid[1] = np->n_vattr.na_filesid[1];
5006 lyp->nfsly_clp = clp;
5007 if (layouttype == NFSLAYOUT_FLEXFILE)
5008 lyp->nfsly_flags = NFSLY_FLEXFILE;
5010 lyp->nfsly_flags = NFSLY_FILES;
5011 if (retonclose != 0)
5012 lyp->nfsly_flags |= NFSLY_RETONCLOSE;
5013 lyp->nfsly_fhlen = fhlen;
5014 NFSBCOPY(fhp, lyp->nfsly_fh, fhlen);
5015 TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
5016 LIST_INSERT_HEAD(NFSCLLAYOUTHASH(clp, fhp, fhlen), lyp,
5018 lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
5021 if (retonclose != 0)
5022 lyp->nfsly_flags |= NFSLY_RETONCLOSE;
5023 if (stateidp->seqid > lyp->nfsly_stateid.seqid)
5024 lyp->nfsly_stateid.seqid = stateidp->seqid;
5025 TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list);
5026 TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
5027 lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
5029 nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
5030 if (NFSCL_FORCEDISM(mp)) {
5033 free(tlyp, M_NFSLAYOUT);
5037 } else if (stateidp->seqid > lyp->nfsly_stateid.seqid)
5038 lyp->nfsly_stateid.seqid = stateidp->seqid;
5040 /* Merge the new list of File Layouts into the list. */
5041 flp = LIST_FIRST(fhlp);
5043 if (flp->nfsfl_iomode == NFSLAYOUTIOMODE_READ)
5044 nfscl_mergeflayouts(&lyp->nfsly_flayread, fhlp);
5046 nfscl_mergeflayouts(&lyp->nfsly_flayrw, fhlp);
5048 if (layout_passed_in != 0)
5049 nfsv4_unlock(&lyp->nfsly_lock, 1);
5052 free(tlyp, M_NFSLAYOUT);
5057 * Search for a layout by MDS file handle.
5058 * If one is found, it is returned with a refcnt (shared lock) iff
5059 * retflpp returned non-NULL and locked (exclusive locked) iff retflpp is
5062 struct nfscllayout *
5063 nfscl_getlayout(struct nfsclclient *clp, uint8_t *fhp, int fhlen,
5064 uint64_t off, struct nfsclflayout **retflpp, int *recalledp)
5066 struct nfscllayout *lyp;
5068 int error, igotlock;
5070 mp = clp->nfsc_nmp->nm_mountp;
5074 lyp = nfscl_findlayout(clp, fhp, fhlen);
5076 if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
5077 TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list);
5078 TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
5079 lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
5080 error = nfscl_findlayoutforio(lyp, off,
5081 NFSV4OPEN_ACCESSREAD, retflpp);
5083 nfsv4_getref(&lyp->nfsly_lock, NULL,
5084 NFSCLSTATEMUTEXPTR, mp);
5087 igotlock = nfsv4_lock(&lyp->nfsly_lock,
5088 1, NULL, NFSCLSTATEMUTEXPTR, mp);
5089 } while (igotlock == 0 && !NFSCL_FORCEDISM(mp));
5092 if (NFSCL_FORCEDISM(mp)) {
5106 * Search for a layout by MDS file handle. If one is found, mark in to be
5107 * recalled, if it already marked "return on close".
5110 nfscl_retoncloselayout(vnode_t vp, struct nfsclclient *clp, uint8_t *fhp,
5111 int fhlen, struct nfsclrecalllayout **recallpp)
5113 struct nfscllayout *lyp;
5116 if (vp->v_type != VREG || !NFSHASPNFS(VFSTONFS(vp->v_mount)) ||
5117 nfscl_enablecallb == 0 || nfs_numnfscbd == 0 ||
5118 (VTONFS(vp)->n_flag & NNOLAYOUT) != 0)
5120 lyp = nfscl_findlayout(clp, fhp, fhlen);
5121 if (lyp != NULL && (lyp->nfsly_flags & (NFSLY_RETONCLOSE |
5122 NFSLY_RECALL)) == NFSLY_RETONCLOSE) {
5124 if (!LIST_EMPTY(&lyp->nfsly_flayread))
5125 iomode |= NFSLAYOUTIOMODE_READ;
5126 if (!LIST_EMPTY(&lyp->nfsly_flayrw))
5127 iomode |= NFSLAYOUTIOMODE_RW;
5128 (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE, lyp, iomode,
5129 0, UINT64_MAX, lyp->nfsly_stateid.seqid, 0, 0, NULL,
5131 NFSCL_DEBUG(4, "retoncls recall iomode=%d\n", iomode);
5137 * Mark the layout to be recalled and with an error.
5138 * Also, disable the dsp from further use.
5141 nfscl_dserr(uint32_t op, uint32_t stat, struct nfscldevinfo *dp,
5142 struct nfscllayout *lyp, struct nfsclds *dsp)
5144 struct nfsclrecalllayout *recallp;
5147 printf("DS being disabled, error=%d\n", stat);
5148 /* Set up the return of the layout. */
5149 recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_WAITOK);
5152 if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
5153 if (!LIST_EMPTY(&lyp->nfsly_flayread))
5154 iomode |= NFSLAYOUTIOMODE_READ;
5155 if (!LIST_EMPTY(&lyp->nfsly_flayrw))
5156 iomode |= NFSLAYOUTIOMODE_RW;
5157 (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE, lyp, iomode,
5158 0, UINT64_MAX, lyp->nfsly_stateid.seqid, stat, op,
5159 dp->nfsdi_deviceid, recallp);
5161 NFSCL_DEBUG(4, "nfscl_dserr recall iomode=%d\n", iomode);
5164 free(recallp, M_NFSLAYRECALL);
5167 /* And shut the TCP connection down. */
5168 nfscl_cancelreqs(dsp);
5172 * Cancel all RPCs for this "dsp" by closing the connection.
5173 * Also, mark the session as defunct.
5174 * If NFSCLDS_SAMECONN is set, the connection is shared with other DSs and
5175 * cannot be shut down.
5178 nfscl_cancelreqs(struct nfsclds *dsp)
5180 struct __rpc_client *cl;
5181 static int non_event;
5184 if ((dsp->nfsclds_flags & (NFSCLDS_CLOSED | NFSCLDS_SAMECONN)) == 0 &&
5185 dsp->nfsclds_sockp != NULL &&
5186 dsp->nfsclds_sockp->nr_client != NULL) {
5187 dsp->nfsclds_flags |= NFSCLDS_CLOSED;
5188 cl = dsp->nfsclds_sockp->nr_client;
5189 dsp->nfsclds_sess.nfsess_defunct = 1;
5193 * This 1sec sleep is done to reduce the number of reconnect
5194 * attempts made on the DS while it has failed.
5196 tsleep(&non_event, PVFS, "ndscls", hz);
5203 * Dereference a layout.
5206 nfscl_rellayout(struct nfscllayout *lyp, int exclocked)
5211 nfsv4_unlock(&lyp->nfsly_lock, 0);
5213 nfsv4_relref(&lyp->nfsly_lock);
5218 * Search for a devinfo by deviceid. If one is found, return it after
5219 * acquiring a reference count on it.
5221 struct nfscldevinfo *
5222 nfscl_getdevinfo(struct nfsclclient *clp, uint8_t *deviceid,
5223 struct nfscldevinfo *dip)
5228 dip = nfscl_finddevinfo(clp, deviceid);
5230 dip->nfsdi_refcnt++;
5236 * Dereference a devinfo structure.
5239 nfscl_reldevinfo_locked(struct nfscldevinfo *dip)
5242 dip->nfsdi_refcnt--;
5243 if (dip->nfsdi_refcnt == 0)
5244 wakeup(&dip->nfsdi_refcnt);
5248 * Dereference a devinfo structure.
5251 nfscl_reldevinfo(struct nfscldevinfo *dip)
5255 nfscl_reldevinfo_locked(dip);
5260 * Find a layout for this file handle. Return NULL upon failure.
5262 static struct nfscllayout *
5263 nfscl_findlayout(struct nfsclclient *clp, u_int8_t *fhp, int fhlen)
5265 struct nfscllayout *lyp;
5267 LIST_FOREACH(lyp, NFSCLLAYOUTHASH(clp, fhp, fhlen), nfsly_hash)
5268 if (lyp->nfsly_fhlen == fhlen &&
5269 !NFSBCMP(lyp->nfsly_fh, fhp, fhlen))
5275 * Find a devinfo for this deviceid. Return NULL upon failure.
5277 static struct nfscldevinfo *
5278 nfscl_finddevinfo(struct nfsclclient *clp, uint8_t *deviceid)
5280 struct nfscldevinfo *dip;
5282 LIST_FOREACH(dip, &clp->nfsc_devinfo, nfsdi_list)
5283 if (NFSBCMP(dip->nfsdi_deviceid, deviceid, NFSX_V4DEVICEID)
5290 * Merge the new file layout list into the main one, maintaining it in
5291 * increasing offset order.
5294 nfscl_mergeflayouts(struct nfsclflayouthead *fhlp,
5295 struct nfsclflayouthead *newfhlp)
5297 struct nfsclflayout *flp, *nflp, *prevflp, *tflp;
5299 flp = LIST_FIRST(fhlp);
5301 LIST_FOREACH_SAFE(nflp, newfhlp, nfsfl_list, tflp) {
5302 while (flp != NULL && flp->nfsfl_off < nflp->nfsfl_off) {
5304 flp = LIST_NEXT(flp, nfsfl_list);
5306 if (prevflp == NULL)
5307 LIST_INSERT_HEAD(fhlp, nflp, nfsfl_list);
5309 LIST_INSERT_AFTER(prevflp, nflp, nfsfl_list);
5315 * Add this nfscldevinfo to the client, if it doesn't already exist.
5316 * This function consumes the structure pointed at by dip, if not NULL.
5319 nfscl_adddevinfo(struct nfsmount *nmp, struct nfscldevinfo *dip, int ind,
5320 struct nfsclflayout *flp)
5322 struct nfsclclient *clp;
5323 struct nfscldevinfo *tdip;
5331 free(dip, M_NFSDEVINFO);
5334 if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5335 dev = flp->nfsfl_dev;
5337 dev = flp->nfsfl_ffm[ind].dev;
5338 tdip = nfscl_finddevinfo(clp, dev);
5340 tdip->nfsdi_layoutrefs++;
5341 if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5342 flp->nfsfl_devp = tdip;
5344 flp->nfsfl_ffm[ind].devp = tdip;
5345 nfscl_reldevinfo_locked(tdip);
5348 free(dip, M_NFSDEVINFO);
5352 LIST_INSERT_HEAD(&clp->nfsc_devinfo, dip, nfsdi_list);
5353 dip->nfsdi_layoutrefs = 1;
5354 if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5355 flp->nfsfl_devp = dip;
5357 flp->nfsfl_ffm[ind].devp = dip;
5366 * Free up a layout structure and associated file layout structure(s).
5369 nfscl_freelayout(struct nfscllayout *layp)
5371 struct nfsclflayout *flp, *nflp;
5372 struct nfsclrecalllayout *rp, *nrp;
5374 LIST_FOREACH_SAFE(flp, &layp->nfsly_flayread, nfsfl_list, nflp) {
5375 LIST_REMOVE(flp, nfsfl_list);
5376 nfscl_freeflayout(flp);
5378 LIST_FOREACH_SAFE(flp, &layp->nfsly_flayrw, nfsfl_list, nflp) {
5379 LIST_REMOVE(flp, nfsfl_list);
5380 nfscl_freeflayout(flp);
5382 LIST_FOREACH_SAFE(rp, &layp->nfsly_recall, nfsrecly_list, nrp) {
5383 LIST_REMOVE(rp, nfsrecly_list);
5384 free(rp, M_NFSLAYRECALL);
5387 free(layp, M_NFSLAYOUT);
5391 * Free up a file layout structure.
5394 nfscl_freeflayout(struct nfsclflayout *flp)
5398 if ((flp->nfsfl_flags & NFSFL_FILE) != 0) {
5399 for (i = 0; i < flp->nfsfl_fhcnt; i++)
5400 free(flp->nfsfl_fh[i], M_NFSFH);
5401 if (flp->nfsfl_devp != NULL)
5402 flp->nfsfl_devp->nfsdi_layoutrefs--;
5404 if ((flp->nfsfl_flags & NFSFL_FLEXFILE) != 0)
5405 for (i = 0; i < flp->nfsfl_mirrorcnt; i++) {
5406 for (j = 0; j < flp->nfsfl_ffm[i].fhcnt; j++)
5407 free(flp->nfsfl_ffm[i].fh[j], M_NFSFH);
5408 if (flp->nfsfl_ffm[i].devp != NULL)
5409 flp->nfsfl_ffm[i].devp->nfsdi_layoutrefs--;
5411 free(flp, M_NFSFLAYOUT);
5415 * Free up a file layout devinfo structure.
5418 nfscl_freedevinfo(struct nfscldevinfo *dip)
5421 free(dip, M_NFSDEVINFO);
5425 * Mark any layouts that match as recalled.
5428 nfscl_layoutrecall(int recalltype, struct nfscllayout *lyp, uint32_t iomode,
5429 uint64_t off, uint64_t len, uint32_t stateseqid, uint32_t stat, uint32_t op,
5430 char *devid, struct nfsclrecalllayout *recallp)
5432 struct nfsclrecalllayout *rp, *orp;
5434 recallp->nfsrecly_recalltype = recalltype;
5435 recallp->nfsrecly_iomode = iomode;
5436 recallp->nfsrecly_stateseqid = stateseqid;
5437 recallp->nfsrecly_off = off;
5438 recallp->nfsrecly_len = len;
5439 recallp->nfsrecly_stat = stat;
5440 recallp->nfsrecly_op = op;
5442 NFSBCOPY(devid, recallp->nfsrecly_devid, NFSX_V4DEVICEID);
5444 * Order the list as file returns first, followed by fsid and any
5445 * returns, both in increasing stateseqid order.
5446 * Note that the seqids wrap around, so 1 is after 0xffffffff.
5447 * (I'm not sure this is correct because I find RFC5661 confusing
5448 * on this, but hopefully it will work ok.)
5451 LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) {
5453 if ((recalltype == NFSLAYOUTRETURN_FILE &&
5454 (rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE ||
5455 nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) ||
5456 (recalltype != NFSLAYOUTRETURN_FILE &&
5457 rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE &&
5458 nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) {
5459 LIST_INSERT_BEFORE(rp, recallp, nfsrecly_list);
5464 * Put any error return on all the file returns that will
5467 if (rp->nfsrecly_recalltype == NFSLAYOUTRETURN_FILE &&
5468 stat != 0 && rp->nfsrecly_stat == 0) {
5469 rp->nfsrecly_stat = stat;
5470 rp->nfsrecly_op = op;
5472 NFSBCOPY(devid, rp->nfsrecly_devid,
5478 LIST_INSERT_HEAD(&lyp->nfsly_recall, recallp,
5481 LIST_INSERT_AFTER(orp, recallp, nfsrecly_list);
5483 lyp->nfsly_flags |= NFSLY_RECALL;
5484 wakeup(lyp->nfsly_clp);
5489 * Compare the two seqids for ordering. The trick is that the seqids can
5490 * wrap around from 0xffffffff->0, so check for the cases where one
5491 * has wrapped around.
5492 * Return 1 if seqid1 comes before seqid2, 0 otherwise.
5495 nfscl_seq(uint32_t seqid1, uint32_t seqid2)
5498 if (seqid2 > seqid1 && (seqid2 - seqid1) >= 0x7fffffff)
5499 /* seqid2 has wrapped around. */
5501 if (seqid1 > seqid2 && (seqid1 - seqid2) >= 0x7fffffff)
5502 /* seqid1 has wrapped around. */
5504 if (seqid1 <= seqid2)
5510 * Do a layout return for each of the recalls.
5513 nfscl_layoutreturn(struct nfsmount *nmp, struct nfscllayout *lyp,
5514 struct ucred *cred, NFSPROC_T *p)
5516 struct nfsclrecalllayout *rp;
5517 nfsv4stateid_t stateid;
5520 NFSBCOPY(lyp->nfsly_stateid.other, stateid.other, NFSX_STATEIDOTHER);
5521 stateid.seqid = lyp->nfsly_stateid.seqid;
5522 if ((lyp->nfsly_flags & NFSLY_FILES) != 0)
5523 layouttype = NFSLAYOUT_NFSV4_1_FILES;
5525 layouttype = NFSLAYOUT_FLEXFILE;
5526 LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) {
5527 (void)nfsrpc_layoutreturn(nmp, lyp->nfsly_fh,
5528 lyp->nfsly_fhlen, 0, layouttype,
5529 rp->nfsrecly_iomode, rp->nfsrecly_recalltype,
5530 rp->nfsrecly_off, rp->nfsrecly_len,
5531 &stateid, cred, p, rp->nfsrecly_stat, rp->nfsrecly_op,
5532 rp->nfsrecly_devid);
5537 * Do the layout commit for a file layout.
5540 nfscl_dolayoutcommit(struct nfsmount *nmp, struct nfscllayout *lyp,
5541 struct ucred *cred, NFSPROC_T *p)
5543 struct nfsclflayout *flp;
5545 int error, layouttype;
5547 if ((lyp->nfsly_flags & NFSLY_FILES) != 0)
5548 layouttype = NFSLAYOUT_NFSV4_1_FILES;
5550 layouttype = NFSLAYOUT_FLEXFILE;
5551 LIST_FOREACH(flp, &lyp->nfsly_flayrw, nfsfl_list) {
5552 if (layouttype == NFSLAYOUT_FLEXFILE &&
5553 (flp->nfsfl_fflags & NFSFLEXFLAG_NO_LAYOUTCOMMIT) != 0) {
5554 NFSCL_DEBUG(4, "Flex file: no layoutcommit\n");
5555 /* If not supported, don't bother doing it. */
5557 nmp->nm_state |= NFSSTA_NOLAYOUTCOMMIT;
5560 } else if (flp->nfsfl_off <= lyp->nfsly_lastbyte) {
5561 len = flp->nfsfl_end - flp->nfsfl_off;
5562 error = nfsrpc_layoutcommit(nmp, lyp->nfsly_fh,
5563 lyp->nfsly_fhlen, 0, flp->nfsfl_off, len,
5564 lyp->nfsly_lastbyte, &lyp->nfsly_stateid,
5565 layouttype, cred, p, NULL);
5566 NFSCL_DEBUG(4, "layoutcommit err=%d\n", error);
5567 if (error == NFSERR_NOTSUPP) {
5568 /* If not supported, don't bother doing it. */
5570 nmp->nm_state |= NFSSTA_NOLAYOUTCOMMIT;
5579 * Commit all layouts for a file (vnode).
5582 nfscl_layoutcommit(vnode_t vp, NFSPROC_T *p)
5584 struct nfsclclient *clp;
5585 struct nfscllayout *lyp;
5586 struct nfsnode *np = VTONFS(vp);
5588 struct nfsmount *nmp;
5592 if (NFSHASNOLAYOUTCOMMIT(nmp))
5600 lyp = nfscl_findlayout(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
5605 nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
5606 if (NFSCL_FORCEDISM(mp)) {
5611 if ((lyp->nfsly_flags & NFSLY_WRITTEN) != 0) {
5612 lyp->nfsly_flags &= ~NFSLY_WRITTEN;
5614 NFSCL_DEBUG(4, "do layoutcommit2\n");
5615 nfscl_dolayoutcommit(clp->nfsc_nmp, lyp, NFSPROCCRED(p), p);
5619 nfsv4_relref(&lyp->nfsly_lock);