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 *, u_int8_t *, int, u_int8_t *,
104 u_int8_t *, u_int32_t, struct nfscllockowner **, struct nfsclopen **);
105 static bool nfscl_checkown(struct nfsclowner *, struct nfsclopen *, uint8_t *,
106 uint8_t *, struct nfscllockowner **, struct nfsclopen **,
107 struct nfsclopen **);
108 static void nfscl_clrelease(struct nfsclclient *);
109 static void nfscl_cleanclient(struct nfsclclient *);
110 static void nfscl_expireclient(struct nfsclclient *, struct nfsmount *,
111 struct ucred *, NFSPROC_T *);
112 static int nfscl_expireopen(struct nfsclclient *, struct nfsclopen *,
113 struct nfsmount *, struct ucred *, NFSPROC_T *);
114 static void nfscl_recover(struct nfsclclient *, bool *, struct ucred *,
116 static void nfscl_insertlock(struct nfscllockowner *, struct nfscllock *,
117 struct nfscllock *, int);
118 static int nfscl_updatelock(struct nfscllockowner *, struct nfscllock **,
119 struct nfscllock **, int);
120 static void nfscl_delegreturnall(struct nfsclclient *, NFSPROC_T *);
121 static u_int32_t nfscl_nextcbident(void);
122 static mount_t nfscl_getmnt(int, uint8_t *, u_int32_t, struct nfsclclient **);
123 static struct nfsclclient *nfscl_getclnt(u_int32_t);
124 static struct nfsclclient *nfscl_getclntsess(uint8_t *);
125 static struct nfscldeleg *nfscl_finddeleg(struct nfsclclient *, u_int8_t *,
127 static void nfscl_retoncloselayout(vnode_t, struct nfsclclient *, uint8_t *,
128 int, struct nfsclrecalllayout **);
129 static void nfscl_reldevinfo_locked(struct nfscldevinfo *);
130 static struct nfscllayout *nfscl_findlayout(struct nfsclclient *, u_int8_t *,
132 static struct nfscldevinfo *nfscl_finddevinfo(struct nfsclclient *, uint8_t *);
133 static int nfscl_checkconflict(struct nfscllockownerhead *, struct nfscllock *,
134 u_int8_t *, struct nfscllock **);
135 static void nfscl_freealllocks(struct nfscllockownerhead *, int);
136 static int nfscl_localconflict(struct nfsclclient *, u_int8_t *, int,
137 struct nfscllock *, u_int8_t *, struct nfscldeleg *, struct nfscllock **);
138 static void nfscl_newopen(struct nfsclclient *, struct nfscldeleg *,
139 struct nfsclowner **, struct nfsclowner **, struct nfsclopen **,
140 struct nfsclopen **, u_int8_t *, u_int8_t *, int, struct ucred *, int *);
141 static int nfscl_moveopen(vnode_t , struct nfsclclient *,
142 struct nfsmount *, struct nfsclopen *, struct nfsclowner *,
143 struct nfscldeleg *, struct ucred *, NFSPROC_T *);
144 static void nfscl_totalrecall(struct nfsclclient *);
145 static int nfscl_relock(vnode_t , struct nfsclclient *, struct nfsmount *,
146 struct nfscllockowner *, struct nfscllock *, struct ucred *, NFSPROC_T *);
147 static int nfscl_tryopen(struct nfsmount *, vnode_t , u_int8_t *, int,
148 u_int8_t *, int, u_int32_t, struct nfsclopen *, u_int8_t *, int,
149 struct nfscldeleg **, int, u_int32_t, struct ucred *, NFSPROC_T *);
150 static int nfscl_trylock(struct nfsmount *, vnode_t , u_int8_t *,
151 int, struct nfscllockowner *, int, int, u_int64_t, u_int64_t, short,
152 struct ucred *, NFSPROC_T *);
153 static int nfsrpc_reopen(struct nfsmount *, u_int8_t *, int, u_int32_t,
154 struct nfsclopen *, struct nfscldeleg **, struct ucred *, NFSPROC_T *);
155 static void nfscl_freedeleg(struct nfscldeleghead *, struct nfscldeleg *);
156 static int nfscl_errmap(struct nfsrv_descript *, u_int32_t);
157 static void nfscl_cleanup_common(struct nfsclclient *, u_int8_t *);
158 static int nfscl_recalldeleg(struct nfsclclient *, struct nfsmount *,
159 struct nfscldeleg *, vnode_t, struct ucred *, NFSPROC_T *, int,
161 static void nfscl_freeopenowner(struct nfsclowner *, int);
162 static void nfscl_cleandeleg(struct nfscldeleg *);
163 static int nfscl_trydelegreturn(struct nfscldeleg *, struct ucred *,
164 struct nfsmount *, NFSPROC_T *);
165 static void nfscl_emptylockowner(struct nfscllockowner *,
166 struct nfscllockownerfhhead *);
167 static void nfscl_mergeflayouts(struct nfsclflayouthead *,
168 struct nfsclflayouthead *);
169 static int nfscl_layoutrecall(int, struct nfscllayout *, uint32_t, uint64_t,
170 uint64_t, uint32_t, uint32_t, uint32_t, char *, struct nfsclrecalllayout *);
171 static int nfscl_seq(uint32_t, uint32_t);
172 static void nfscl_layoutreturn(struct nfsmount *, struct nfscllayout *,
173 struct ucred *, NFSPROC_T *);
174 static void nfscl_dolayoutcommit(struct nfsmount *, struct nfscllayout *,
175 struct ucred *, NFSPROC_T *);
177 static short nfscberr_null[] = {
182 static short nfscberr_getattr[] = {
191 static short nfscberr_recall[] = {
201 static short *nfscl_cberrmap[] = {
209 #define NETFAMILY(clp) \
210 (((clp)->nfsc_flags & NFSCLFLAGS_AFINET6) ? AF_INET6 : AF_INET)
213 * Called for an open operation.
214 * If the nfhp argument is NULL, just get an openowner.
217 nfscl_open(vnode_t vp, u_int8_t *nfhp, int fhlen, u_int32_t amode, int usedeleg,
218 struct ucred *cred, NFSPROC_T *p, struct nfsclowner **owpp,
219 struct nfsclopen **opp, int *newonep, int *retp, int lockit)
221 struct nfsclclient *clp;
222 struct nfsclowner *owp, *nowp;
223 struct nfsclopen *op = NULL, *nop = NULL;
224 struct nfscldeleg *dp;
225 struct nfsclownerhead *ohp;
226 u_int8_t own[NFSV4CL_LOCKNAMELEN];
237 * Might need one or both of these, so MALLOC them now, to
238 * avoid a tsleep() in MALLOC later.
240 nowp = malloc(sizeof (struct nfsclowner),
241 M_NFSCLOWNER, M_WAITOK);
243 nop = malloc(sizeof (struct nfsclopen) +
244 fhlen - 1, M_NFSCLOPEN, M_WAITOK);
245 ret = nfscl_getcl(vp->v_mount, cred, p, 1, &clp);
247 free(nowp, M_NFSCLOWNER);
249 free(nop, M_NFSCLOPEN);
254 * Get the Open iff it already exists.
255 * If none found, add the new one or return error, depending upon
260 /* First check the delegation list */
261 if (nfhp != NULL && usedeleg) {
262 LIST_FOREACH(dp, NFSCLDELEGHASH(clp, nfhp, fhlen), nfsdl_hash) {
263 if (dp->nfsdl_fhlen == fhlen &&
264 !NFSBCMP(nfhp, dp->nfsdl_fh, fhlen)) {
265 if (!(amode & NFSV4OPEN_ACCESSWRITE) ||
266 (dp->nfsdl_flags & NFSCLDL_WRITE))
274 /* For NFSv4.1/4.2 and this option, use a single open_owner. */
275 if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount)))
276 nfscl_filllockowner(NULL, own, F_POSIX);
278 nfscl_filllockowner(p->td_proc, own, F_POSIX);
280 ohp = &dp->nfsdl_owner;
282 ohp = &clp->nfsc_owner;
283 /* Now, search for an openowner */
284 LIST_FOREACH(owp, ohp, nfsow_list) {
285 if (!NFSBCMP(owp->nfsow_owner, own, NFSV4CL_LOCKNAMELEN))
290 * Create a new open, as required.
292 nfscl_newopen(clp, dp, &owp, &nowp, &op, &nop, own, nfhp, fhlen,
296 * Now, check the mode on the open and return the appropriate
300 if (nfhp != NULL && dp != NULL && nop == NULL)
301 /* new local open on delegation */
302 *retp = NFSCLOPEN_SETCRED;
304 *retp = NFSCLOPEN_OK;
306 if (op != NULL && (amode & ~(op->nfso_mode))) {
307 op->nfso_mode |= amode;
308 if (retp != NULL && dp == NULL)
309 *retp = NFSCLOPEN_DOOPEN;
313 * Serialize modifications to the open owner for multiple threads
314 * within the same process using a read/write sleep lock.
315 * For NFSv4.1 and a single OpenOwner, allow concurrent open operations
316 * by acquiring a shared lock. The close operations still use an
317 * exclusive lock for this case.
320 if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount))) {
322 * Get a shared lock on the OpenOwner, but first
323 * wait for any pending exclusive lock, so that the
324 * exclusive locker gets priority.
326 nfsv4_lock(&owp->nfsow_rwlock, 0, NULL,
327 NFSCLSTATEMUTEXPTR, NULL);
328 nfsv4_getref(&owp->nfsow_rwlock, NULL,
329 NFSCLSTATEMUTEXPTR, NULL);
331 nfscl_lockexcl(&owp->nfsow_rwlock, NFSCLSTATEMUTEXPTR);
335 free(nowp, M_NFSCLOWNER);
337 free(nop, M_NFSCLOPEN);
346 * Create a new open, as required.
349 nfscl_newopen(struct nfsclclient *clp, struct nfscldeleg *dp,
350 struct nfsclowner **owpp, struct nfsclowner **nowpp, struct nfsclopen **opp,
351 struct nfsclopen **nopp, u_int8_t *own, u_int8_t *fhp, int fhlen,
352 struct ucred *cred, int *newonep)
354 struct nfsclowner *owp = *owpp, *nowp;
355 struct nfsclopen *op, *nop;
365 if (owp == NULL && nowp != NULL) {
366 NFSBCOPY(own, nowp->nfsow_owner, NFSV4CL_LOCKNAMELEN);
367 LIST_INIT(&nowp->nfsow_open);
368 nowp->nfsow_clp = clp;
369 nowp->nfsow_seqid = 0;
370 nowp->nfsow_defunct = 0;
371 nfscl_lockinit(&nowp->nfsow_rwlock);
373 nfsstatsv1.cllocalopenowners++;
374 LIST_INSERT_HEAD(&dp->nfsdl_owner, nowp, nfsow_list);
376 nfsstatsv1.clopenowners++;
377 LIST_INSERT_HEAD(&clp->nfsc_owner, nowp, nfsow_list);
385 /* If an fhp has been specified, create an Open as well. */
387 /* and look for the correct open, based upon FH */
388 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
389 if (op->nfso_fhlen == fhlen &&
390 !NFSBCMP(op->nfso_fh, fhp, fhlen))
393 if (op == NULL && nop != NULL) {
396 nop->nfso_opencnt = 0;
397 nop->nfso_posixlock = 1;
398 nop->nfso_fhlen = fhlen;
399 NFSBCOPY(fhp, nop->nfso_fh, fhlen);
400 LIST_INIT(&nop->nfso_lock);
401 nop->nfso_stateid.seqid = 0;
402 nop->nfso_stateid.other[0] = 0;
403 nop->nfso_stateid.other[1] = 0;
404 nop->nfso_stateid.other[2] = 0;
405 KASSERT(cred != NULL, ("%s: cred NULL\n", __func__));
406 newnfs_copyincred(cred, &nop->nfso_cred);
408 TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
409 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp,
411 dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
412 nfsstatsv1.cllocalopens++;
414 nfsstatsv1.clopens++;
416 LIST_INSERT_HEAD(&owp->nfsow_open, nop, nfso_list);
428 * Called to find/add a delegation to a client.
431 nfscl_deleg(mount_t mp, struct nfsclclient *clp, u_int8_t *nfhp,
432 int fhlen, struct ucred *cred, NFSPROC_T *p, struct nfscldeleg **dpp)
434 struct nfscldeleg *dp = *dpp, *tdp;
437 * First, if we have received a Read delegation for a file on a
438 * read/write file system, just return it, because they aren't
441 if (mp != NULL && dp != NULL && !NFSMNT_RDONLY(mp) &&
442 (dp->nfsdl_flags & NFSCLDL_READ)) {
443 (void) nfscl_trydelegreturn(dp, cred, VFSTONFS(mp), p);
444 free(dp, M_NFSCLDELEG);
449 /* Look for the correct deleg, based upon FH */
451 tdp = nfscl_finddeleg(clp, nfhp, fhlen);
455 return (NFSERR_BADSTATEID);
458 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list);
459 LIST_INSERT_HEAD(NFSCLDELEGHASH(clp, nfhp, fhlen), dp,
461 dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
462 nfsstatsv1.cldelegates++;
466 * Delegation already exists, what do we do if a new one??
469 printf("Deleg already exists!\n");
470 free(dp, M_NFSCLDELEG);
481 * Find a delegation for this file handle. Return NULL upon failure.
483 static struct nfscldeleg *
484 nfscl_finddeleg(struct nfsclclient *clp, u_int8_t *fhp, int fhlen)
486 struct nfscldeleg *dp;
488 LIST_FOREACH(dp, NFSCLDELEGHASH(clp, fhp, fhlen), nfsdl_hash) {
489 if (dp->nfsdl_fhlen == fhlen &&
490 !NFSBCMP(dp->nfsdl_fh, fhp, fhlen))
497 * Get a stateid for an I/O operation. First, look for an open and iff
498 * found, return either a lockowner stateid or the open stateid.
499 * If no Open is found, just return error and the special stateid of all zeros.
502 nfscl_getstateid(vnode_t vp, u_int8_t *nfhp, int fhlen, u_int32_t mode,
503 int fords, struct ucred *cred, NFSPROC_T *p, nfsv4stateid_t *stateidp,
506 struct nfsclclient *clp;
507 struct nfsclowner *owp;
508 struct nfsclopen *op = NULL, *top;
509 struct nfscllockowner *lp;
510 struct nfscldeleg *dp;
512 struct nfsmount *nmp;
513 u_int8_t own[NFSV4CL_LOCKNAMELEN];
519 * Initially, just set the special stateid of all zeros.
520 * (Don't do this for a DS, since the special stateid can't be used.)
524 stateidp->other[0] = 0;
525 stateidp->other[1] = 0;
526 stateidp->other[2] = 0;
528 if (vnode_vtype(vp) != VREG)
531 nmp = VFSTONFS(vp->v_mount);
533 clp = nfscl_findcl(nmp);
540 * Wait for recovery to complete.
542 while ((clp->nfsc_flags & NFSCLFLAGS_RECVRINPROG))
543 (void) nfsmsleep(&clp->nfsc_flags, NFSCLSTATEMUTEXPTR,
544 PZERO, "nfsrecvr", NULL);
547 * First, look for a delegation.
549 LIST_FOREACH(dp, NFSCLDELEGHASH(clp, nfhp, fhlen), nfsdl_hash) {
550 if (dp->nfsdl_fhlen == fhlen &&
551 !NFSBCMP(nfhp, dp->nfsdl_fh, fhlen)) {
552 if (!(mode & NFSV4OPEN_ACCESSWRITE) ||
553 (dp->nfsdl_flags & NFSCLDL_WRITE)) {
554 stateidp->seqid = dp->nfsdl_stateid.seqid;
555 stateidp->other[0] = dp->nfsdl_stateid.other[0];
556 stateidp->other[1] = dp->nfsdl_stateid.other[1];
557 stateidp->other[2] = dp->nfsdl_stateid.other[2];
558 if (!(np->n_flag & NDELEGRECALL)) {
559 TAILQ_REMOVE(&clp->nfsc_deleg, dp,
561 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp,
563 dp->nfsdl_timestamp = NFSD_MONOSEC +
565 dp->nfsdl_rwlock.nfslock_usecnt++;
566 *lckpp = (void *)&dp->nfsdl_rwlock;
577 * If p != NULL, we want to search the parentage tree
578 * for a matching OpenOwner and use that.
580 if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount)))
581 nfscl_filllockowner(NULL, own, F_POSIX);
583 nfscl_filllockowner(p->td_proc, own, F_POSIX);
585 error = nfscl_getopen(&clp->nfsc_owner, nfhp, fhlen, own, own,
587 if (error == 0 && lp != NULL && fords == 0) {
588 /* Don't return a lock stateid for a DS. */
590 lp->nfsl_stateid.seqid;
592 lp->nfsl_stateid.other[0];
594 lp->nfsl_stateid.other[1];
596 lp->nfsl_stateid.other[2];
602 /* If not found, just look for any OpenOwner that will work. */
605 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
606 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
607 if (op->nfso_fhlen == fhlen &&
608 !NFSBCMP(op->nfso_fh, nfhp, fhlen)) {
609 if (top == NULL && (op->nfso_mode &
610 NFSV4OPEN_ACCESSWRITE) != 0 &&
611 (mode & NFSV4OPEN_ACCESSREAD) != 0)
613 if ((mode & op->nfso_mode) == mode) {
623 NFSCL_DEBUG(2, "openmode top=%p\n", top);
624 if (top == NULL || NFSHASOPENMODE(nmp)) {
631 * For read aheads or write behinds, use the open cred.
632 * A read ahead or write behind is indicated by p == NULL.
635 newnfs_copycred(&op->nfso_cred, cred);
639 * No lock stateid, so return the open stateid.
641 stateidp->seqid = op->nfso_stateid.seqid;
642 stateidp->other[0] = op->nfso_stateid.other[0];
643 stateidp->other[1] = op->nfso_stateid.other[1];
644 stateidp->other[2] = op->nfso_stateid.other[2];
650 * Search for a matching file, mode and, optionally, lockowner.
653 nfscl_getopen(struct nfsclownerhead *ohp, u_int8_t *nfhp, int fhlen,
654 u_int8_t *openown, u_int8_t *lockown, u_int32_t mode,
655 struct nfscllockowner **lpp, struct nfsclopen **opp)
657 struct nfsclowner *owp;
658 struct nfsclopen *op, *rop, *rop2;
664 * rop will be set to the open to be returned. There are three
665 * variants of this, all for an open of the correct file:
666 * 1 - A match of lockown.
667 * 2 - A match of the openown, when no lockown match exists.
668 * 3 - A match for any open, if no openown or lockown match exists.
669 * Looking for #2 over #3 probably isn't necessary, but since
670 * RFC3530 is vague w.r.t. the relationship between openowners and
671 * lockowners, I think this is the safer way to go.
676 /* Search the client list */
677 LIST_FOREACH(owp, ohp, nfsow_list) {
678 /* and look for the correct open */
679 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
680 if (op->nfso_fhlen == fhlen &&
681 !NFSBCMP(op->nfso_fh, nfhp, fhlen)
682 && (op->nfso_mode & mode) == mode)
683 keep_looping = nfscl_checkown(owp, op, openown,
684 lockown, lpp, &rop, &rop2);
699 /* Check for an owner match. */
701 nfscl_checkown(struct nfsclowner *owp, struct nfsclopen *op, uint8_t *openown,
702 uint8_t *lockown, struct nfscllockowner **lpp, struct nfsclopen **ropp,
703 struct nfsclopen **ropp2)
705 struct nfscllockowner *lp;
710 /* Now look for a matching lockowner. */
711 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
712 if (!NFSBCMP(lp->nfsl_owner, lockown,
713 NFSV4CL_LOCKNAMELEN)) {
720 if (*ropp == NULL && !NFSBCMP(owp->nfsow_owner, openown,
721 NFSV4CL_LOCKNAMELEN)) {
724 keep_looping = false;
728 return (keep_looping);
732 * Release use of an open owner. Called when open operations are done
733 * with the open owner.
736 nfscl_ownerrelease(struct nfsmount *nmp, struct nfsclowner *owp,
737 __unused int error, __unused int candelete, int unlocked)
744 if (NFSHASONEOPENOWN(nmp))
745 nfsv4_relref(&owp->nfsow_rwlock);
747 nfscl_lockunlock(&owp->nfsow_rwlock);
749 nfscl_clrelease(owp->nfsow_clp);
754 * Release use of an open structure under an open owner.
757 nfscl_openrelease(struct nfsmount *nmp, struct nfsclopen *op, int error,
760 struct nfsclclient *clp;
761 struct nfsclowner *owp;
767 if (NFSHASONEOPENOWN(nmp))
768 nfsv4_relref(&owp->nfsow_rwlock);
770 nfscl_lockunlock(&owp->nfsow_rwlock);
771 clp = owp->nfsow_clp;
772 if (error && candelete && op->nfso_opencnt == 0)
773 nfscl_freeopen(op, 0);
774 nfscl_clrelease(clp);
779 * Called to get a clientid structure. It will optionally lock the
780 * client data structures to do the SetClientId/SetClientId_confirm,
781 * but will release that lock and return the clientid with a reference
783 * If the "cred" argument is NULL, a new clientid should not be created.
784 * If the "p" argument is NULL, a SetClientID/SetClientIDConfirm cannot
786 * The start_renewthread argument tells nfscl_getcl() to start a renew
787 * thread if this creates a new clp.
788 * It always clpp with a reference count on it, unless returning an error.
791 nfscl_getcl(struct mount *mp, struct ucred *cred, NFSPROC_T *p,
792 int start_renewthread, struct nfsclclient **clpp)
794 struct nfsclclient *clp;
795 struct nfsclclient *newclp = NULL;
796 struct nfsmount *nmp;
797 char uuid[HOSTUUIDLEN];
798 int igotlock = 0, error, trystalecnt, clidinusedelay, i;
803 getcredhostuuid(cred, uuid, sizeof uuid);
804 idlen = strlen(uuid);
806 idlen += sizeof (u_int64_t);
808 idlen += sizeof (u_int64_t) + 16; /* 16 random bytes */
810 sizeof (struct nfsclclient) + idlen - 1, M_NFSCLCLIENT,
815 * If a forced dismount is already in progress, don't
816 * allocate a new clientid and get out now. For the case where
817 * clp != NULL, this is a harmless optimization.
819 if (NFSCL_FORCEDISM(mp)) {
822 free(newclp, M_NFSCLCLIENT);
827 if (newclp == NULL) {
832 clp->nfsc_idlen = idlen;
833 LIST_INIT(&clp->nfsc_owner);
834 TAILQ_INIT(&clp->nfsc_deleg);
835 TAILQ_INIT(&clp->nfsc_layout);
836 LIST_INIT(&clp->nfsc_devinfo);
837 for (i = 0; i < NFSCLDELEGHASHSIZE; i++)
838 LIST_INIT(&clp->nfsc_deleghash[i]);
839 for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++)
840 LIST_INIT(&clp->nfsc_layouthash[i]);
841 clp->nfsc_flags = NFSCLFLAGS_INITED;
842 clp->nfsc_clientidrev = 1;
843 clp->nfsc_cbident = nfscl_nextcbident();
844 nfscl_fillclid(nmp->nm_clval, uuid, clp->nfsc_id,
846 LIST_INSERT_HEAD(&nfsclhead, clp, nfsc_list);
850 if (start_renewthread != 0)
851 nfscl_start_renewthread(clp);
855 free(newclp, M_NFSCLCLIENT);
858 while ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0 && !igotlock &&
859 !NFSCL_FORCEDISM(mp))
860 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
861 NFSCLSTATEMUTEXPTR, mp);
864 * Call nfsv4_lock() with "iwantlock == 0" so that it will
865 * wait for a pending exclusive lock request. This gives the
866 * exclusive lock request priority over this shared lock
868 * An exclusive lock on nfsc_lock is used mainly for server
871 nfsv4_lock(&clp->nfsc_lock, 0, NULL, NFSCLSTATEMUTEXPTR, mp);
872 nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
874 if (igotlock == 0 && NFSCL_FORCEDISM(mp)) {
876 * Both nfsv4_lock() and nfsv4_getref() know to check
877 * for NFSCL_FORCEDISM() and return without sleeping to
878 * wait for the exclusive lock to be released, since it
879 * might be held by nfscl_umount() and we need to get out
880 * now for that case and not wait until nfscl_umount()
889 * If it needs a clientid, do the setclientid now.
891 if ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0) {
893 panic("nfscl_clget");
894 if (p == NULL || cred == NULL) {
896 nfsv4_unlock(&clp->nfsc_lock, 0);
901 * If RFC3530 Sec. 14.2.33 is taken literally,
902 * NFSERR_CLIDINUSE will be returned persistently for the
903 * case where a new mount of the same file system is using
904 * a different principal. In practice, NFSERR_CLIDINUSE is
905 * only returned when there is outstanding unexpired state
906 * on the clientid. As such, try for twice the lease
907 * interval, if we know what that is. Otherwise, make a
909 * The case of returning NFSERR_STALECLIENTID is far less
910 * likely, but might occur if there is a significant delay
911 * between doing the SetClientID and SetClientIDConfirm Ops,
912 * such that the server throws away the clientid before
913 * receiving the SetClientIDConfirm.
915 if (clp->nfsc_renew > 0)
916 clidinusedelay = NFSCL_LEASE(clp->nfsc_renew) * 2;
918 clidinusedelay = 120;
921 error = nfsrpc_setclient(nmp, clp, 0, NULL, cred, p);
922 if (error == NFSERR_STALECLIENTID ||
923 error == NFSERR_STALEDONTRECOVER ||
924 error == NFSERR_BADSESSION ||
925 error == NFSERR_CLIDINUSE) {
926 (void) nfs_catnap(PZERO, error, "nfs_setcl");
928 } while (((error == NFSERR_STALECLIENTID ||
929 error == NFSERR_BADSESSION ||
930 error == NFSERR_STALEDONTRECOVER) && --trystalecnt > 0) ||
931 (error == NFSERR_CLIDINUSE && --clidinusedelay > 0));
934 nfsv4_unlock(&clp->nfsc_lock, 0);
938 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
942 nfsv4_unlock(&clp->nfsc_lock, 1);
951 * Get a reference to a clientid and return it, if valid.
954 nfscl_findcl(struct nfsmount *nmp)
956 struct nfsclclient *clp;
959 if (clp == NULL || !(clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID))
965 * Release the clientid structure. It may be locked or reference counted.
968 nfscl_clrelease(struct nfsclclient *clp)
971 if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK)
972 nfsv4_unlock(&clp->nfsc_lock, 0);
974 nfsv4_relref(&clp->nfsc_lock);
978 * External call for nfscl_clrelease.
981 nfscl_clientrelease(struct nfsclclient *clp)
985 if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK)
986 nfsv4_unlock(&clp->nfsc_lock, 0);
988 nfsv4_relref(&clp->nfsc_lock);
993 * Called when wanting to lock a byte region.
996 nfscl_getbytelock(vnode_t vp, u_int64_t off, u_int64_t len,
997 short type, struct ucred *cred, NFSPROC_T *p, struct nfsclclient *rclp,
998 int recovery, void *id, int flags, u_int8_t *rownp, u_int8_t *ropenownp,
999 struct nfscllockowner **lpp, int *newonep, int *donelocallyp)
1001 struct nfscllockowner *lp;
1002 struct nfsclopen *op;
1003 struct nfsclclient *clp;
1004 struct nfscllockowner *nlp;
1005 struct nfscllock *nlop, *otherlop;
1006 struct nfscldeleg *dp = NULL, *ldp = NULL;
1007 struct nfscllockownerhead *lhp = NULL;
1009 u_int8_t own[NFSV4CL_LOCKNAMELEN], *ownp, openown[NFSV4CL_LOCKNAMELEN];
1011 int error = 0, ret, donelocally = 0;
1014 /* For Lock Ops, the open mode doesn't matter, so use 0 to match any. */
1023 * Might need these, so MALLOC them now, to
1024 * avoid a tsleep() in MALLOC later.
1027 sizeof (struct nfscllockowner), M_NFSCLLOCKOWNER, M_WAITOK);
1029 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1031 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1032 nlop->nfslo_type = type;
1033 nlop->nfslo_first = off;
1034 if (len == NFS64BITSSET) {
1035 nlop->nfslo_end = NFS64BITSSET;
1037 nlop->nfslo_end = off + len;
1038 if (nlop->nfslo_end <= nlop->nfslo_first)
1039 error = NFSERR_INVAL;
1046 error = nfscl_getcl(vp->v_mount, cred, p, 1, &clp);
1049 free(nlp, M_NFSCLLOCKOWNER);
1050 free(otherlop, M_NFSCLLOCK);
1051 free(nlop, M_NFSCLLOCK);
1058 openownp = ropenownp;
1060 nfscl_filllockowner(id, own, flags);
1062 if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount)))
1063 nfscl_filllockowner(NULL, openown, F_POSIX);
1065 nfscl_filllockowner(p->td_proc, openown, F_POSIX);
1071 * First, search for a delegation. If one exists for this file,
1072 * the lock can be done locally against it, so long as there
1073 * isn't a local lock conflict.
1075 ldp = dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
1076 np->n_fhp->nfh_len);
1077 /* Just sanity check for correct type of delegation */
1078 if (dp != NULL && ((dp->nfsdl_flags &
1079 (NFSCLDL_RECALL | NFSCLDL_DELEGRET)) != 0 ||
1081 (dp->nfsdl_flags & NFSCLDL_WRITE) == 0)))
1085 /* Now, find an open and maybe a lockowner. */
1086 ret = nfscl_getopen(&dp->nfsdl_owner, np->n_fhp->nfh_fh,
1087 np->n_fhp->nfh_len, openownp, ownp, mode, NULL, &op);
1089 ret = nfscl_getopen(&clp->nfsc_owner,
1090 np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp,
1091 ownp, mode, NULL, &op);
1093 lhp = &dp->nfsdl_lock;
1094 TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
1095 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list);
1096 dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
1104 * Get the related Open and maybe lockowner.
1106 error = nfscl_getopen(&clp->nfsc_owner,
1107 np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp,
1108 ownp, mode, &lp, &op);
1110 lhp = &op->nfso_lock;
1112 if (!error && !recovery)
1113 error = nfscl_localconflict(clp, np->n_fhp->nfh_fh,
1114 np->n_fhp->nfh_len, nlop, ownp, ldp, NULL);
1117 nfscl_clrelease(clp);
1120 free(nlp, M_NFSCLLOCKOWNER);
1121 free(otherlop, M_NFSCLLOCK);
1122 free(nlop, M_NFSCLLOCK);
1127 * Ok, see if a lockowner exists and create one, as required.
1130 LIST_FOREACH(lp, lhp, nfsl_list) {
1131 if (!NFSBCMP(lp->nfsl_owner, ownp, NFSV4CL_LOCKNAMELEN))
1135 NFSBCOPY(ownp, nlp->nfsl_owner, NFSV4CL_LOCKNAMELEN);
1137 NFSBCOPY(ropenownp, nlp->nfsl_openowner,
1138 NFSV4CL_LOCKNAMELEN);
1140 NFSBCOPY(op->nfso_own->nfsow_owner, nlp->nfsl_openowner,
1141 NFSV4CL_LOCKNAMELEN);
1142 nlp->nfsl_seqid = 0;
1143 nlp->nfsl_lockflags = flags;
1144 nlp->nfsl_inprog = NULL;
1145 nfscl_lockinit(&nlp->nfsl_rwlock);
1146 LIST_INIT(&nlp->nfsl_lock);
1148 nlp->nfsl_open = NULL;
1149 nfsstatsv1.cllocallockowners++;
1151 nlp->nfsl_open = op;
1152 nfsstatsv1.cllockowners++;
1154 LIST_INSERT_HEAD(lhp, nlp, nfsl_list);
1161 * Now, update the byte ranges for locks.
1163 ret = nfscl_updatelock(lp, &nlop, &otherlop, donelocally);
1169 nfscl_clrelease(clp);
1172 * Serial modifications on the lock owner for multiple threads
1173 * for the same process using a read/write lock.
1176 nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR);
1182 free(nlp, M_NFSCLLOCKOWNER);
1184 free(nlop, M_NFSCLLOCK);
1186 free(otherlop, M_NFSCLLOCK);
1193 * Called to unlock a byte range, for LockU.
1196 nfscl_relbytelock(vnode_t vp, u_int64_t off, u_int64_t len,
1197 __unused struct ucred *cred, NFSPROC_T *p, int callcnt,
1198 struct nfsclclient *clp, void *id, int flags,
1199 struct nfscllockowner **lpp, int *dorpcp)
1201 struct nfscllockowner *lp;
1202 struct nfsclowner *owp;
1203 struct nfsclopen *op;
1204 struct nfscllock *nlop, *other_lop = NULL;
1205 struct nfscldeleg *dp;
1207 u_int8_t own[NFSV4CL_LOCKNAMELEN];
1215 * Might need these, so MALLOC them now, to
1216 * avoid a tsleep() in MALLOC later.
1219 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1220 nlop->nfslo_type = F_UNLCK;
1221 nlop->nfslo_first = off;
1222 if (len == NFS64BITSSET) {
1223 nlop->nfslo_end = NFS64BITSSET;
1225 nlop->nfslo_end = off + len;
1226 if (nlop->nfslo_end <= nlop->nfslo_first) {
1227 free(nlop, M_NFSCLLOCK);
1228 return (NFSERR_INVAL);
1233 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1236 nfscl_filllockowner(id, own, flags);
1240 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
1241 np->n_fhp->nfh_len);
1244 * First, unlock any local regions on a delegation.
1247 /* Look for this lockowner. */
1248 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
1249 if (!NFSBCMP(lp->nfsl_owner, own,
1250 NFSV4CL_LOCKNAMELEN))
1254 /* Use other_lop, so nlop is still available */
1255 (void)nfscl_updatelock(lp, &other_lop, NULL, 1);
1259 * Now, find a matching open/lockowner that hasn't already been done,
1260 * as marked by nfsl_inprog.
1264 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
1265 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1266 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1267 !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1268 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1269 if (lp->nfsl_inprog == NULL &&
1270 !NFSBCMP(lp->nfsl_owner, own,
1271 NFSV4CL_LOCKNAMELEN)) {
1285 ret = nfscl_updatelock(lp, &nlop, NULL, 0);
1289 * Serial modifications on the lock owner for multiple
1290 * threads for the same process using a read/write lock.
1292 lp->nfsl_inprog = p;
1293 nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR);
1298 free(nlop, M_NFSCLLOCK);
1300 free(other_lop, M_NFSCLLOCK);
1305 * Release all lockowners marked in progess for this process and file.
1308 nfscl_releasealllocks(struct nfsclclient *clp, vnode_t vp, NFSPROC_T *p,
1309 void *id, int flags)
1311 struct nfsclowner *owp;
1312 struct nfsclopen *op;
1313 struct nfscllockowner *lp;
1315 u_int8_t own[NFSV4CL_LOCKNAMELEN];
1318 nfscl_filllockowner(id, own, flags);
1320 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
1321 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1322 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1323 !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1324 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1325 if (lp->nfsl_inprog == p &&
1326 !NFSBCMP(lp->nfsl_owner, own,
1327 NFSV4CL_LOCKNAMELEN)) {
1328 lp->nfsl_inprog = NULL;
1329 nfscl_lockunlock(&lp->nfsl_rwlock);
1335 nfscl_clrelease(clp);
1340 * Called to find out if any bytes within the byte range specified are
1341 * write locked by the calling process. Used to determine if flushing
1342 * is required before a LockU.
1343 * If in doubt, return 1, so the flush will occur.
1346 nfscl_checkwritelocked(vnode_t vp, struct flock *fl,
1347 struct ucred *cred, NFSPROC_T *p, void *id, int flags)
1349 struct nfsclowner *owp;
1350 struct nfscllockowner *lp;
1351 struct nfsclopen *op;
1352 struct nfsclclient *clp;
1353 struct nfscllock *lop;
1354 struct nfscldeleg *dp;
1357 u_int8_t own[NFSV4CL_LOCKNAMELEN];
1361 switch (fl->l_whence) {
1365 * Caller is responsible for adding any necessary offset
1366 * when SEEK_CUR is used.
1371 off = np->n_size + fl->l_start;
1376 if (fl->l_len != 0) {
1377 end = off + fl->l_len;
1384 error = nfscl_getcl(vp->v_mount, cred, p, 1, &clp);
1387 nfscl_filllockowner(id, own, flags);
1391 * First check the delegation locks.
1393 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
1395 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
1396 if (!NFSBCMP(lp->nfsl_owner, own,
1397 NFSV4CL_LOCKNAMELEN))
1401 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
1402 if (lop->nfslo_first >= end)
1404 if (lop->nfslo_end <= off)
1406 if (lop->nfslo_type == F_WRLCK) {
1407 nfscl_clrelease(clp);
1416 * Now, check state against the server.
1418 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
1419 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1420 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1421 !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1422 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1423 if (!NFSBCMP(lp->nfsl_owner, own,
1424 NFSV4CL_LOCKNAMELEN))
1428 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
1429 if (lop->nfslo_first >= end)
1431 if (lop->nfslo_end <= off)
1433 if (lop->nfslo_type == F_WRLCK) {
1434 nfscl_clrelease(clp);
1443 nfscl_clrelease(clp);
1449 * Release a byte range lock owner structure.
1452 nfscl_lockrelease(struct nfscllockowner *lp, int error, int candelete)
1454 struct nfsclclient *clp;
1459 clp = lp->nfsl_open->nfso_own->nfsow_clp;
1460 if (error != 0 && candelete &&
1461 (lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED) == 0)
1462 nfscl_freelockowner(lp, 0);
1464 nfscl_lockunlock(&lp->nfsl_rwlock);
1465 nfscl_clrelease(clp);
1470 * Free up an open structure and any associated byte range lock structures.
1473 nfscl_freeopen(struct nfsclopen *op, int local)
1476 LIST_REMOVE(op, nfso_list);
1477 nfscl_freealllocks(&op->nfso_lock, local);
1478 free(op, M_NFSCLOPEN);
1480 nfsstatsv1.cllocalopens--;
1482 nfsstatsv1.clopens--;
1486 * Free up all lock owners and associated locks.
1489 nfscl_freealllocks(struct nfscllockownerhead *lhp, int local)
1491 struct nfscllockowner *lp, *nlp;
1493 LIST_FOREACH_SAFE(lp, lhp, nfsl_list, nlp) {
1494 if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED))
1496 nfscl_freelockowner(lp, local);
1501 * Called for an Open when NFSERR_EXPIRED is received from the server.
1502 * If there are no byte range locks nor a Share Deny lost, try to do a
1503 * fresh Open. Otherwise, free the open.
1506 nfscl_expireopen(struct nfsclclient *clp, struct nfsclopen *op,
1507 struct nfsmount *nmp, struct ucred *cred, NFSPROC_T *p)
1509 struct nfscllockowner *lp;
1510 struct nfscldeleg *dp;
1511 int mustdelete = 0, error;
1514 * Look for any byte range lock(s).
1516 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1517 if (!LIST_EMPTY(&lp->nfsl_lock)) {
1524 * If no byte range lock(s) nor a Share deny, try to re-open.
1526 if (!mustdelete && (op->nfso_mode & NFSLCK_DENYBITS) == 0) {
1527 newnfs_copycred(&op->nfso_cred, cred);
1529 error = nfsrpc_reopen(nmp, op->nfso_fh,
1530 op->nfso_fhlen, op->nfso_mode, op, &dp, cred, p);
1534 free(dp, M_NFSCLDELEG);
1539 nfscl_deleg(nmp->nm_mountp, clp, op->nfso_fh,
1540 op->nfso_fhlen, cred, p, &dp);
1544 * If a byte range lock or Share deny or couldn't re-open, free it.
1547 nfscl_freeopen(op, 0);
1548 return (mustdelete);
1552 * Free up an open owner structure.
1555 nfscl_freeopenowner(struct nfsclowner *owp, int local)
1558 LIST_REMOVE(owp, nfsow_list);
1559 free(owp, M_NFSCLOWNER);
1561 nfsstatsv1.cllocalopenowners--;
1563 nfsstatsv1.clopenowners--;
1567 * Free up a byte range lock owner structure.
1570 nfscl_freelockowner(struct nfscllockowner *lp, int local)
1572 struct nfscllock *lop, *nlop;
1574 LIST_REMOVE(lp, nfsl_list);
1575 LIST_FOREACH_SAFE(lop, &lp->nfsl_lock, nfslo_list, nlop) {
1576 nfscl_freelock(lop, local);
1578 free(lp, M_NFSCLLOCKOWNER);
1580 nfsstatsv1.cllocallockowners--;
1582 nfsstatsv1.cllockowners--;
1586 * Free up a byte range lock structure.
1589 nfscl_freelock(struct nfscllock *lop, int local)
1592 LIST_REMOVE(lop, nfslo_list);
1593 free(lop, M_NFSCLLOCK);
1595 nfsstatsv1.cllocallocks--;
1597 nfsstatsv1.cllocks--;
1601 * Clean out the state related to a delegation.
1604 nfscl_cleandeleg(struct nfscldeleg *dp)
1606 struct nfsclowner *owp, *nowp;
1607 struct nfsclopen *op;
1609 LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) {
1610 op = LIST_FIRST(&owp->nfsow_open);
1612 if (LIST_NEXT(op, nfso_list) != NULL)
1613 panic("nfscleandel");
1614 nfscl_freeopen(op, 1);
1616 nfscl_freeopenowner(owp, 1);
1618 nfscl_freealllocks(&dp->nfsdl_lock, 1);
1622 * Free a delegation.
1625 nfscl_freedeleg(struct nfscldeleghead *hdp, struct nfscldeleg *dp)
1628 TAILQ_REMOVE(hdp, dp, nfsdl_list);
1629 LIST_REMOVE(dp, nfsdl_hash);
1630 free(dp, M_NFSCLDELEG);
1631 nfsstatsv1.cldelegates--;
1636 * Free up all state related to this client structure.
1639 nfscl_cleanclient(struct nfsclclient *clp)
1641 struct nfsclowner *owp, *nowp;
1642 struct nfsclopen *op, *nop;
1643 struct nfscllayout *lyp, *nlyp;
1644 struct nfscldevinfo *dip, *ndip;
1646 TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp)
1647 nfscl_freelayout(lyp);
1649 LIST_FOREACH_SAFE(dip, &clp->nfsc_devinfo, nfsdi_list, ndip)
1650 nfscl_freedevinfo(dip);
1652 /* Now, all the OpenOwners, etc. */
1653 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1654 LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
1655 nfscl_freeopen(op, 0);
1657 nfscl_freeopenowner(owp, 0);
1662 * Called when an NFSERR_EXPIRED is received from the server.
1665 nfscl_expireclient(struct nfsclclient *clp, struct nfsmount *nmp,
1666 struct ucred *cred, NFSPROC_T *p)
1668 struct nfsclowner *owp, *nowp, *towp;
1669 struct nfsclopen *op, *nop, *top;
1670 struct nfscldeleg *dp, *ndp;
1671 int ret, printed = 0;
1674 * First, merge locally issued Opens into the list for the server.
1676 dp = TAILQ_FIRST(&clp->nfsc_deleg);
1677 while (dp != NULL) {
1678 ndp = TAILQ_NEXT(dp, nfsdl_list);
1679 owp = LIST_FIRST(&dp->nfsdl_owner);
1680 while (owp != NULL) {
1681 nowp = LIST_NEXT(owp, nfsow_list);
1682 op = LIST_FIRST(&owp->nfsow_open);
1684 if (LIST_NEXT(op, nfso_list) != NULL)
1686 LIST_FOREACH(towp, &clp->nfsc_owner, nfsow_list) {
1687 if (!NFSBCMP(towp->nfsow_owner, owp->nfsow_owner,
1688 NFSV4CL_LOCKNAMELEN))
1692 /* Merge opens in */
1693 LIST_FOREACH(top, &towp->nfsow_open, nfso_list) {
1694 if (top->nfso_fhlen == op->nfso_fhlen &&
1695 !NFSBCMP(top->nfso_fh, op->nfso_fh,
1697 top->nfso_mode |= op->nfso_mode;
1698 top->nfso_opencnt += op->nfso_opencnt;
1703 /* Just add the open to the owner list */
1704 LIST_REMOVE(op, nfso_list);
1705 op->nfso_own = towp;
1706 LIST_INSERT_HEAD(&towp->nfsow_open, op, nfso_list);
1707 nfsstatsv1.cllocalopens--;
1708 nfsstatsv1.clopens++;
1711 /* Just add the openowner to the client list */
1712 LIST_REMOVE(owp, nfsow_list);
1713 owp->nfsow_clp = clp;
1714 LIST_INSERT_HEAD(&clp->nfsc_owner, owp, nfsow_list);
1715 nfsstatsv1.cllocalopenowners--;
1716 nfsstatsv1.clopenowners++;
1717 nfsstatsv1.cllocalopens--;
1718 nfsstatsv1.clopens++;
1723 if (!printed && !LIST_EMPTY(&dp->nfsdl_lock)) {
1725 printf("nfsv4 expired locks lost\n");
1727 nfscl_cleandeleg(dp);
1728 nfscl_freedeleg(&clp->nfsc_deleg, dp);
1731 if (!TAILQ_EMPTY(&clp->nfsc_deleg))
1735 * Now, try and reopen against the server.
1737 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1738 owp->nfsow_seqid = 0;
1739 LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
1740 ret = nfscl_expireopen(clp, op, nmp, cred, p);
1741 if (ret && !printed) {
1743 printf("nfsv4 expired locks lost\n");
1746 if (LIST_EMPTY(&owp->nfsow_open))
1747 nfscl_freeopenowner(owp, 0);
1752 * This function must be called after the process represented by "own" has
1753 * exited. Must be called with CLSTATE lock held.
1756 nfscl_cleanup_common(struct nfsclclient *clp, u_int8_t *own)
1758 struct nfsclowner *owp, *nowp;
1759 struct nfscllockowner *lp, *nlp;
1760 struct nfscldeleg *dp;
1762 /* First, get rid of local locks on delegations. */
1763 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
1764 LIST_FOREACH_SAFE(lp, &dp->nfsdl_lock, nfsl_list, nlp) {
1765 if (!NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
1766 if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED))
1768 nfscl_freelockowner(lp, 1);
1772 owp = LIST_FIRST(&clp->nfsc_owner);
1773 while (owp != NULL) {
1774 nowp = LIST_NEXT(owp, nfsow_list);
1775 if (!NFSBCMP(owp->nfsow_owner, own,
1776 NFSV4CL_LOCKNAMELEN)) {
1778 * If there are children that haven't closed the
1779 * file descriptors yet, the opens will still be
1780 * here. For that case, let the renew thread clear
1781 * out the OpenOwner later.
1783 if (LIST_EMPTY(&owp->nfsow_open))
1784 nfscl_freeopenowner(owp, 0);
1786 owp->nfsow_defunct = 1;
1793 * Find open/lock owners for processes that have exited.
1796 nfscl_cleanupkext(struct nfsclclient *clp, struct nfscllockownerfhhead *lhp)
1798 struct nfsclowner *owp, *nowp;
1799 struct nfsclopen *op;
1800 struct nfscllockowner *lp, *nlp;
1801 struct nfscldeleg *dp;
1804 * All the pidhash locks must be acquired, since they are sx locks
1805 * and must be acquired before the mutexes. The pid(s) that will
1806 * be used aren't known yet, so all the locks need to be acquired.
1807 * Fortunately, this function is only performed once/sec.
1811 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1812 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1813 LIST_FOREACH_SAFE(lp, &op->nfso_lock, nfsl_list, nlp) {
1814 if (LIST_EMPTY(&lp->nfsl_lock))
1815 nfscl_emptylockowner(lp, lhp);
1818 if (nfscl_procdoesntexist(owp->nfsow_owner))
1819 nfscl_cleanup_common(clp, owp->nfsow_owner);
1823 * For the single open_owner case, these lock owners need to be
1824 * checked to see if they still exist separately.
1825 * This is because nfscl_procdoesntexist() never returns true for
1826 * the single open_owner so that the above doesn't ever call
1827 * nfscl_cleanup_common().
1829 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
1830 LIST_FOREACH_SAFE(lp, &dp->nfsdl_lock, nfsl_list, nlp) {
1831 if (nfscl_procdoesntexist(lp->nfsl_owner))
1832 nfscl_cleanup_common(clp, lp->nfsl_owner);
1836 pidhash_sunlockall();
1840 * Take the empty lock owner and move it to the local lhp list if the
1841 * associated process no longer exists.
1844 nfscl_emptylockowner(struct nfscllockowner *lp,
1845 struct nfscllockownerfhhead *lhp)
1847 struct nfscllockownerfh *lfhp, *mylfhp;
1848 struct nfscllockowner *nlp;
1851 /* If not a Posix lock owner, just return. */
1852 if ((lp->nfsl_lockflags & F_POSIX) == 0)
1858 * First, search to see if this lock owner is already in the list.
1859 * If it is, then the associated process no longer exists.
1861 SLIST_FOREACH(lfhp, lhp, nfslfh_list) {
1862 if (lfhp->nfslfh_len == lp->nfsl_open->nfso_fhlen &&
1863 !NFSBCMP(lfhp->nfslfh_fh, lp->nfsl_open->nfso_fh,
1866 LIST_FOREACH(nlp, &lfhp->nfslfh_lock, nfsl_list)
1867 if (!NFSBCMP(nlp->nfsl_owner, lp->nfsl_owner,
1868 NFSV4CL_LOCKNAMELEN))
1871 /* If not found, check if process still exists. */
1872 if (fnd_it == 0 && nfscl_procdoesntexist(lp->nfsl_owner) == 0)
1875 /* Move the lock owner over to the local list. */
1876 if (mylfhp == NULL) {
1877 mylfhp = malloc(sizeof(struct nfscllockownerfh), M_TEMP,
1881 mylfhp->nfslfh_len = lp->nfsl_open->nfso_fhlen;
1882 NFSBCOPY(lp->nfsl_open->nfso_fh, mylfhp->nfslfh_fh,
1883 mylfhp->nfslfh_len);
1884 LIST_INIT(&mylfhp->nfslfh_lock);
1885 SLIST_INSERT_HEAD(lhp, mylfhp, nfslfh_list);
1887 LIST_REMOVE(lp, nfsl_list);
1888 LIST_INSERT_HEAD(&mylfhp->nfslfh_lock, lp, nfsl_list);
1891 static int fake_global; /* Used to force visibility of MNTK_UNMOUNTF */
1893 * Called from nfs umount to free up the clientid.
1896 nfscl_umount(struct nfsmount *nmp, NFSPROC_T *p)
1898 struct nfsclclient *clp;
1903 * For the case that matters, this is the thread that set
1904 * MNTK_UNMOUNTF, so it will see it set. The code that follows is
1905 * done to ensure that any thread executing nfscl_getcl() after
1906 * this time, will see MNTK_UNMOUNTF set. nfscl_getcl() uses the
1907 * mutex for NFSLOCKCLSTATE(), so it is "m" for the following
1908 * explanation, courtesy of Alan Cox.
1909 * What follows is a snippet from Alan Cox's email at:
1910 * https://docs.FreeBSD.org/cgi/mid.cgi?BANLkTikR3d65zPHo9==08ZfJ2vmqZucEvw
1912 * 1. Set MNTK_UNMOUNTF
1913 * 2. Acquire a standard FreeBSD mutex "m".
1914 * 3. Update some data structures.
1915 * 4. Release mutex "m".
1917 * Then, other threads that acquire "m" after step 4 has occurred will
1918 * see MNTK_UNMOUNTF as set. But, other threads that beat thread X to
1919 * step 2 may or may not see MNTK_UNMOUNTF as set.
1922 if ((nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) {
1930 if ((clp->nfsc_flags & NFSCLFLAGS_INITED) == 0)
1931 panic("nfscl umount");
1934 * First, handshake with the nfscl renew thread, to terminate
1937 clp->nfsc_flags |= NFSCLFLAGS_UMOUNT;
1938 while (clp->nfsc_flags & NFSCLFLAGS_HASTHREAD)
1939 (void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT,
1943 * Now, get the exclusive lock on the client state, so
1944 * that no uses of the state are still in progress.
1947 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
1948 NFSCLSTATEMUTEXPTR, NULL);
1949 } while (!igotlock);
1953 * Free up all the state. It will expire on the server, but
1954 * maybe we should do a SetClientId/SetClientIdConfirm so
1955 * the server throws it away?
1957 LIST_REMOVE(clp, nfsc_list);
1958 nfscl_delegreturnall(clp, p);
1959 cred = newnfs_getcred();
1960 if (NFSHASNFSV4N(nmp)) {
1961 (void)nfsrpc_destroysession(nmp, clp, cred, p);
1962 (void)nfsrpc_destroyclient(nmp, clp, cred, p);
1964 (void)nfsrpc_setclient(nmp, clp, 0, NULL, cred, p);
1965 nfscl_cleanclient(clp);
1968 free(clp, M_NFSCLCLIENT);
1974 * This function is called when a server replies with NFSERR_STALECLIENTID
1975 * NFSERR_STALESTATEID or NFSERR_BADSESSION. It traverses the clientid lists,
1976 * doing Opens and Locks with reclaim. If these fail, it deletes the
1977 * corresponding state.
1980 nfscl_recover(struct nfsclclient *clp, bool *retokp, struct ucred *cred,
1983 struct nfsclowner *owp, *nowp;
1984 struct nfsclopen *op, *nop;
1985 struct nfscllockowner *lp, *nlp;
1986 struct nfscllock *lop, *nlop;
1987 struct nfscldeleg *dp, *ndp, *tdp;
1988 struct nfsmount *nmp;
1989 struct ucred *tcred;
1990 struct nfsclopenhead extra_open;
1991 struct nfscldeleghead extra_deleg;
1994 u_int32_t delegtype = NFSV4OPEN_DELEGATEWRITE, mode;
1995 int i, igotlock = 0, error, trycnt, firstlock;
1996 struct nfscllayout *lyp, *nlyp;
1999 * First, lock the client structure, so everyone else will
2000 * block when trying to use state.
2003 clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG;
2005 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
2006 NFSCLSTATEMUTEXPTR, NULL);
2007 } while (!igotlock);
2010 nmp = clp->nfsc_nmp;
2012 panic("nfscl recover");
2015 * For now, just get rid of all layouts. There may be a need
2016 * to do LayoutCommit Ops with reclaim == true later.
2018 TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp)
2019 nfscl_freelayout(lyp);
2020 TAILQ_INIT(&clp->nfsc_layout);
2021 for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++)
2022 LIST_INIT(&clp->nfsc_layouthash[i]);
2027 error = nfsrpc_setclient(nmp, clp, 1, retokp, cred, p);
2028 } while ((error == NFSERR_STALECLIENTID ||
2029 error == NFSERR_BADSESSION ||
2030 error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
2033 clp->nfsc_flags &= ~(NFSCLFLAGS_RECOVER |
2034 NFSCLFLAGS_RECVRINPROG);
2035 wakeup(&clp->nfsc_flags);
2036 nfsv4_unlock(&clp->nfsc_lock, 0);
2040 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
2041 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2044 * Mark requests already queued on the server, so that they don't
2045 * initiate another recovery cycle. Any requests already in the
2046 * queue that handle state information will have the old stale
2047 * clientid/stateid and will get a NFSERR_STALESTATEID,
2048 * NFSERR_STALECLIENTID or NFSERR_BADSESSION reply from the server.
2049 * This will be translated to NFSERR_STALEDONTRECOVER when
2050 * R_DONTRECOVER is set.
2053 TAILQ_FOREACH(rep, &nfsd_reqq, r_chain) {
2054 if (rep->r_nmp == nmp)
2055 rep->r_flags |= R_DONTRECOVER;
2060 * If nfsrpc_setclient() returns *retokp == true,
2061 * no more recovery is needed.
2067 * Now, mark all delegations "need reclaim".
2069 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list)
2070 dp->nfsdl_flags |= NFSCLDL_NEEDRECLAIM;
2072 TAILQ_INIT(&extra_deleg);
2073 LIST_INIT(&extra_open);
2075 * Now traverse the state lists, doing Open and Lock Reclaims.
2077 tcred = newnfs_getcred();
2078 owp = LIST_FIRST(&clp->nfsc_owner);
2079 while (owp != NULL) {
2080 nowp = LIST_NEXT(owp, nfsow_list);
2081 owp->nfsow_seqid = 0;
2082 op = LIST_FIRST(&owp->nfsow_open);
2083 while (op != NULL) {
2084 nop = LIST_NEXT(op, nfso_list);
2085 if (error != NFSERR_NOGRACE && error != NFSERR_BADSESSION) {
2086 /* Search for a delegation to reclaim with the open */
2087 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2088 if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
2090 if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
2091 mode = NFSV4OPEN_ACCESSWRITE;
2092 delegtype = NFSV4OPEN_DELEGATEWRITE;
2094 mode = NFSV4OPEN_ACCESSREAD;
2095 delegtype = NFSV4OPEN_DELEGATEREAD;
2097 if ((op->nfso_mode & mode) == mode &&
2098 op->nfso_fhlen == dp->nfsdl_fhlen &&
2099 !NFSBCMP(op->nfso_fh, dp->nfsdl_fh, op->nfso_fhlen))
2104 delegtype = NFSV4OPEN_DELEGATENONE;
2105 newnfs_copycred(&op->nfso_cred, tcred);
2106 error = nfscl_tryopen(nmp, NULL, op->nfso_fh,
2107 op->nfso_fhlen, op->nfso_fh, op->nfso_fhlen,
2108 op->nfso_mode, op, NULL, 0, &ndp, 1, delegtype,
2111 /* Handle any replied delegation */
2112 if (ndp != NULL && ((ndp->nfsdl_flags & NFSCLDL_WRITE)
2113 || NFSMNT_RDONLY(nmp->nm_mountp))) {
2114 if ((ndp->nfsdl_flags & NFSCLDL_WRITE))
2115 mode = NFSV4OPEN_ACCESSWRITE;
2117 mode = NFSV4OPEN_ACCESSREAD;
2118 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2119 if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
2121 if ((op->nfso_mode & mode) == mode &&
2122 op->nfso_fhlen == dp->nfsdl_fhlen &&
2123 !NFSBCMP(op->nfso_fh, dp->nfsdl_fh,
2125 dp->nfsdl_stateid = ndp->nfsdl_stateid;
2126 dp->nfsdl_sizelimit = ndp->nfsdl_sizelimit;
2127 dp->nfsdl_ace = ndp->nfsdl_ace;
2128 dp->nfsdl_change = ndp->nfsdl_change;
2129 dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
2130 if ((ndp->nfsdl_flags & NFSCLDL_RECALL))
2131 dp->nfsdl_flags |= NFSCLDL_RECALL;
2132 free(ndp, M_NFSCLDELEG);
2139 TAILQ_INSERT_HEAD(&extra_deleg, ndp, nfsdl_list);
2141 /* and reclaim all byte range locks */
2142 lp = LIST_FIRST(&op->nfso_lock);
2143 while (lp != NULL) {
2144 nlp = LIST_NEXT(lp, nfsl_list);
2147 lop = LIST_FIRST(&lp->nfsl_lock);
2148 while (lop != NULL) {
2149 nlop = LIST_NEXT(lop, nfslo_list);
2150 if (lop->nfslo_end == NFS64BITSSET)
2153 len = lop->nfslo_end - lop->nfslo_first;
2154 error = nfscl_trylock(nmp, NULL,
2155 op->nfso_fh, op->nfso_fhlen, lp,
2156 firstlock, 1, lop->nfslo_first, len,
2157 lop->nfslo_type, tcred, p);
2159 nfscl_freelock(lop, 0);
2164 /* If no locks, but a lockowner, just delete it. */
2165 if (LIST_EMPTY(&lp->nfsl_lock))
2166 nfscl_freelockowner(lp, 0);
2171 if (error != 0 && error != NFSERR_BADSESSION)
2172 nfscl_freeopen(op, 0);
2179 * Now, try and get any delegations not yet reclaimed by cobbling
2180 * to-gether an appropriate open.
2183 dp = TAILQ_FIRST(&clp->nfsc_deleg);
2184 while (dp != NULL) {
2185 ndp = TAILQ_NEXT(dp, nfsdl_list);
2186 if ((dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM)) {
2189 sizeof (struct nfsclowner), M_NFSCLOWNER, M_WAITOK);
2191 * Name must be as long an largest possible
2192 * NFSV4CL_LOCKNAMELEN. 12 for now.
2194 NFSBCOPY("RECLAIMDELEG", nowp->nfsow_owner,
2195 NFSV4CL_LOCKNAMELEN);
2196 LIST_INIT(&nowp->nfsow_open);
2197 nowp->nfsow_clp = clp;
2198 nowp->nfsow_seqid = 0;
2199 nowp->nfsow_defunct = 0;
2200 nfscl_lockinit(&nowp->nfsow_rwlock);
2203 if (error != NFSERR_NOGRACE && error != NFSERR_BADSESSION) {
2204 nop = malloc(sizeof (struct nfsclopen) +
2205 dp->nfsdl_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
2206 nop->nfso_own = nowp;
2207 if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
2208 nop->nfso_mode = NFSV4OPEN_ACCESSWRITE;
2209 delegtype = NFSV4OPEN_DELEGATEWRITE;
2211 nop->nfso_mode = NFSV4OPEN_ACCESSREAD;
2212 delegtype = NFSV4OPEN_DELEGATEREAD;
2214 nop->nfso_opencnt = 0;
2215 nop->nfso_posixlock = 1;
2216 nop->nfso_fhlen = dp->nfsdl_fhlen;
2217 NFSBCOPY(dp->nfsdl_fh, nop->nfso_fh, dp->nfsdl_fhlen);
2218 LIST_INIT(&nop->nfso_lock);
2219 nop->nfso_stateid.seqid = 0;
2220 nop->nfso_stateid.other[0] = 0;
2221 nop->nfso_stateid.other[1] = 0;
2222 nop->nfso_stateid.other[2] = 0;
2223 newnfs_copycred(&dp->nfsdl_cred, tcred);
2224 newnfs_copyincred(tcred, &nop->nfso_cred);
2226 error = nfscl_tryopen(nmp, NULL, nop->nfso_fh,
2227 nop->nfso_fhlen, nop->nfso_fh, nop->nfso_fhlen,
2228 nop->nfso_mode, nop, NULL, 0, &tdp, 1,
2229 delegtype, tcred, p);
2231 if ((tdp->nfsdl_flags & NFSCLDL_WRITE))
2232 mode = NFSV4OPEN_ACCESSWRITE;
2234 mode = NFSV4OPEN_ACCESSREAD;
2235 if ((nop->nfso_mode & mode) == mode &&
2236 nop->nfso_fhlen == tdp->nfsdl_fhlen &&
2237 !NFSBCMP(nop->nfso_fh, tdp->nfsdl_fh,
2239 dp->nfsdl_stateid = tdp->nfsdl_stateid;
2240 dp->nfsdl_sizelimit = tdp->nfsdl_sizelimit;
2241 dp->nfsdl_ace = tdp->nfsdl_ace;
2242 dp->nfsdl_change = tdp->nfsdl_change;
2243 dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
2244 if ((tdp->nfsdl_flags & NFSCLDL_RECALL))
2245 dp->nfsdl_flags |= NFSCLDL_RECALL;
2246 free(tdp, M_NFSCLDELEG);
2248 TAILQ_INSERT_HEAD(&extra_deleg, tdp, nfsdl_list);
2254 free(nop, M_NFSCLOPEN);
2256 * Couldn't reclaim it, so throw the state
2259 nfscl_cleandeleg(dp);
2260 nfscl_freedeleg(&clp->nfsc_deleg, dp);
2262 LIST_INSERT_HEAD(&extra_open, nop, nfso_list);
2269 * Now, get rid of extra Opens and Delegations.
2271 LIST_FOREACH_SAFE(op, &extra_open, nfso_list, nop) {
2273 newnfs_copycred(&op->nfso_cred, tcred);
2274 error = nfscl_tryclose(op, tcred, nmp, p);
2275 if (error == NFSERR_GRACE)
2276 (void) nfs_catnap(PZERO, error, "nfsexcls");
2277 } while (error == NFSERR_GRACE);
2278 LIST_REMOVE(op, nfso_list);
2279 free(op, M_NFSCLOPEN);
2282 free(nowp, M_NFSCLOWNER);
2284 TAILQ_FOREACH_SAFE(dp, &extra_deleg, nfsdl_list, ndp) {
2286 newnfs_copycred(&dp->nfsdl_cred, tcred);
2287 error = nfscl_trydelegreturn(dp, tcred, nmp, p);
2288 if (error == NFSERR_GRACE)
2289 (void) nfs_catnap(PZERO, error, "nfsexdlg");
2290 } while (error == NFSERR_GRACE);
2291 TAILQ_REMOVE(&extra_deleg, dp, nfsdl_list);
2292 free(dp, M_NFSCLDELEG);
2295 /* For NFSv4.1 or later, do a RECLAIM_COMPLETE. */
2296 if (NFSHASNFSV4N(nmp))
2297 (void)nfsrpc_reclaimcomplete(nmp, cred, p);
2301 clp->nfsc_flags &= ~NFSCLFLAGS_RECVRINPROG;
2302 wakeup(&clp->nfsc_flags);
2303 nfsv4_unlock(&clp->nfsc_lock, 0);
2310 * This function is called when a server replies with NFSERR_EXPIRED.
2311 * It deletes all state for the client and does a fresh SetClientId/confirm.
2312 * XXX Someday it should post a signal to the process(es) that hold the
2313 * state, so they know that lock state has been lost.
2316 nfscl_hasexpired(struct nfsclclient *clp, u_int32_t clidrev, NFSPROC_T *p)
2318 struct nfsmount *nmp;
2320 int igotlock = 0, error, trycnt;
2323 * If the clientid has gone away or a new SetClientid has already
2324 * been done, just return ok.
2326 if (clp == NULL || clidrev != clp->nfsc_clientidrev)
2330 * First, lock the client structure, so everyone else will
2331 * block when trying to use state. Also, use NFSCLFLAGS_EXPIREIT so
2332 * that only one thread does the work.
2335 clp->nfsc_flags |= NFSCLFLAGS_EXPIREIT;
2337 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
2338 NFSCLSTATEMUTEXPTR, NULL);
2339 } while (!igotlock && (clp->nfsc_flags & NFSCLFLAGS_EXPIREIT));
2340 if ((clp->nfsc_flags & NFSCLFLAGS_EXPIREIT) == 0) {
2342 nfsv4_unlock(&clp->nfsc_lock, 0);
2346 clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG;
2349 nmp = clp->nfsc_nmp;
2351 panic("nfscl expired");
2352 cred = newnfs_getcred();
2355 error = nfsrpc_setclient(nmp, clp, 0, NULL, cred, p);
2356 } while ((error == NFSERR_STALECLIENTID ||
2357 error == NFSERR_BADSESSION ||
2358 error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
2361 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2364 * Expire the state for the client.
2366 nfscl_expireclient(clp, nmp, cred, p);
2368 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
2369 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2371 clp->nfsc_flags &= ~(NFSCLFLAGS_EXPIREIT | NFSCLFLAGS_RECVRINPROG);
2372 wakeup(&clp->nfsc_flags);
2373 nfsv4_unlock(&clp->nfsc_lock, 0);
2380 * This function inserts a lock in the list after insert_lop.
2383 nfscl_insertlock(struct nfscllockowner *lp, struct nfscllock *new_lop,
2384 struct nfscllock *insert_lop, int local)
2387 if ((struct nfscllockowner *)insert_lop == lp)
2388 LIST_INSERT_HEAD(&lp->nfsl_lock, new_lop, nfslo_list);
2390 LIST_INSERT_AFTER(insert_lop, new_lop, nfslo_list);
2392 nfsstatsv1.cllocallocks++;
2394 nfsstatsv1.cllocks++;
2398 * This function updates the locking for a lock owner and given file. It
2399 * maintains a list of lock ranges ordered on increasing file offset that
2400 * are NFSCLLOCK_READ or NFSCLLOCK_WRITE and non-overlapping (aka POSIX style).
2401 * It always adds new_lop to the list and sometimes uses the one pointed
2403 * Returns 1 if the locks were modified, 0 otherwise.
2406 nfscl_updatelock(struct nfscllockowner *lp, struct nfscllock **new_lopp,
2407 struct nfscllock **other_lopp, int local)
2409 struct nfscllock *new_lop = *new_lopp;
2410 struct nfscllock *lop, *tlop, *ilop;
2411 struct nfscllock *other_lop;
2412 int unlock = 0, modified = 0;
2416 * Work down the list until the lock is merged.
2418 if (new_lop->nfslo_type == F_UNLCK)
2420 ilop = (struct nfscllock *)lp;
2421 lop = LIST_FIRST(&lp->nfsl_lock);
2422 while (lop != NULL) {
2424 * Only check locks for this file that aren't before the start of
2427 if (lop->nfslo_end >= new_lop->nfslo_first) {
2428 if (new_lop->nfslo_end < lop->nfslo_first) {
2430 * If the new lock ends before the start of the
2431 * current lock's range, no merge, just insert
2436 if (new_lop->nfslo_type == lop->nfslo_type ||
2437 (new_lop->nfslo_first <= lop->nfslo_first &&
2438 new_lop->nfslo_end >= lop->nfslo_end)) {
2440 * This lock can be absorbed by the new lock/unlock.
2441 * This happens when it covers the entire range
2442 * of the old lock or is contiguous
2443 * with the old lock and is of the same type or an
2446 if (new_lop->nfslo_type != lop->nfslo_type ||
2447 new_lop->nfslo_first != lop->nfslo_first ||
2448 new_lop->nfslo_end != lop->nfslo_end)
2450 if (lop->nfslo_first < new_lop->nfslo_first)
2451 new_lop->nfslo_first = lop->nfslo_first;
2452 if (lop->nfslo_end > new_lop->nfslo_end)
2453 new_lop->nfslo_end = lop->nfslo_end;
2455 lop = LIST_NEXT(lop, nfslo_list);
2456 nfscl_freelock(tlop, local);
2461 * All these cases are for contiguous locks that are not the
2462 * same type, so they can't be merged.
2464 if (new_lop->nfslo_first <= lop->nfslo_first) {
2466 * This case is where the new lock overlaps with the
2467 * first part of the old lock. Move the start of the
2468 * old lock to just past the end of the new lock. The
2469 * new lock will be inserted in front of the old, since
2470 * ilop hasn't been updated. (We are done now.)
2472 if (lop->nfslo_first != new_lop->nfslo_end) {
2473 lop->nfslo_first = new_lop->nfslo_end;
2478 if (new_lop->nfslo_end >= lop->nfslo_end) {
2480 * This case is where the new lock overlaps with the
2481 * end of the old lock's range. Move the old lock's
2482 * end to just before the new lock's first and insert
2483 * the new lock after the old lock.
2484 * Might not be done yet, since the new lock could
2485 * overlap further locks with higher ranges.
2487 if (lop->nfslo_end != new_lop->nfslo_first) {
2488 lop->nfslo_end = new_lop->nfslo_first;
2492 lop = LIST_NEXT(lop, nfslo_list);
2496 * The final case is where the new lock's range is in the
2497 * middle of the current lock's and splits the current lock
2498 * up. Use *other_lopp to handle the second part of the
2499 * split old lock range. (We are done now.)
2500 * For unlock, we use new_lop as other_lop and tmp, since
2501 * other_lop and new_lop are the same for this case.
2502 * We noted the unlock case above, so we don't need
2503 * new_lop->nfslo_type any longer.
2505 tmp = new_lop->nfslo_first;
2507 other_lop = new_lop;
2510 other_lop = *other_lopp;
2513 other_lop->nfslo_first = new_lop->nfslo_end;
2514 other_lop->nfslo_end = lop->nfslo_end;
2515 other_lop->nfslo_type = lop->nfslo_type;
2516 lop->nfslo_end = tmp;
2517 nfscl_insertlock(lp, other_lop, lop, local);
2523 lop = LIST_NEXT(lop, nfslo_list);
2529 * Insert the new lock in the list at the appropriate place.
2532 nfscl_insertlock(lp, new_lop, ilop, local);
2540 * This function must be run as a kernel thread.
2541 * It does Renew Ops and recovery, when required.
2544 nfscl_renewthread(struct nfsclclient *clp, NFSPROC_T *p)
2546 struct nfsclowner *owp, *nowp;
2547 struct nfsclopen *op;
2548 struct nfscllockowner *lp, *nlp;
2549 struct nfscldeleghead dh;
2550 struct nfscldeleg *dp, *ndp;
2553 int error, cbpathdown, islept, igotlock, ret, clearok;
2554 uint32_t recover_done_time = 0;
2556 static time_t prevsec = 0;
2557 struct nfscllockownerfh *lfhp, *nlfhp;
2558 struct nfscllockownerfhhead lfh;
2559 struct nfscllayout *lyp, *nlyp;
2560 struct nfscldevinfo *dip, *ndip;
2561 struct nfscllayouthead rlh;
2562 struct nfsclrecalllayout *recallp;
2563 struct nfsclds *dsp;
2568 cred = newnfs_getcred();
2570 clp->nfsc_flags |= NFSCLFLAGS_HASTHREAD;
2571 mp = clp->nfsc_nmp->nm_mountp;
2574 newnfs_setroot(cred);
2576 if (clp->nfsc_flags & NFSCLFLAGS_RECOVER) {
2578 * Only allow one full recover within 1/2 of the lease
2579 * duration (nfsc_renew).
2580 * retok is value/result. If passed in set to true,
2581 * it indicates only a CreateSession operation should
2583 * If it is returned true, it indicates that the
2584 * recovery only required a CreateSession.
2587 if (recover_done_time < NFSD_MONOSEC) {
2588 recover_done_time = NFSD_MONOSEC +
2592 NFSCL_DEBUG(1, "Doing recovery, only "
2593 "createsession=%d\n", retok);
2594 nfscl_recover(clp, &retok, cred, p);
2596 if (clp->nfsc_expire <= NFSD_MONOSEC &&
2597 (clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID)) {
2598 clp->nfsc_expire = NFSD_MONOSEC + clp->nfsc_renew;
2599 clidrev = clp->nfsc_clientidrev;
2600 error = nfsrpc_renew(clp, NULL, cred, p);
2601 if (error == NFSERR_CBPATHDOWN)
2603 else if (error == NFSERR_STALECLIENTID ||
2604 error == NFSERR_BADSESSION) {
2606 clp->nfsc_flags |= NFSCLFLAGS_RECOVER;
2608 } else if (error == NFSERR_EXPIRED)
2609 (void) nfscl_hasexpired(clp, clidrev, p);
2613 if (NFSHASNFSV4N(clp->nfsc_nmp)) {
2614 /* Do renews for any DS sessions. */
2615 NFSLOCKMNT(clp->nfsc_nmp);
2616 /* Skip first entry, since the MDS is handled above. */
2617 dsp = TAILQ_FIRST(&clp->nfsc_nmp->nm_sess);
2619 dsp = TAILQ_NEXT(dsp, nfsclds_list);
2620 while (dsp != NULL) {
2621 if (dsp->nfsclds_expire <= NFSD_MONOSEC &&
2622 dsp->nfsclds_sess.nfsess_defunct == 0) {
2623 dsp->nfsclds_expire = NFSD_MONOSEC +
2625 NFSUNLOCKMNT(clp->nfsc_nmp);
2626 (void)nfsrpc_renew(clp, dsp, cred, p);
2629 dsp = TAILQ_NEXT(dsp, nfsclds_list);
2631 NFSUNLOCKMNT(clp->nfsc_nmp);
2637 /* It's a Total Recall! */
2638 nfscl_totalrecall(clp);
2641 * Now, handle defunct owners.
2643 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
2644 if (LIST_EMPTY(&owp->nfsow_open)) {
2645 if (owp->nfsow_defunct != 0)
2646 nfscl_freeopenowner(owp, 0);
2651 * Do the recall on any delegations. To avoid trouble, always
2652 * come back up here after having slept.
2656 dp = TAILQ_FIRST(&clp->nfsc_deleg);
2657 while (dp != NULL) {
2658 ndp = TAILQ_NEXT(dp, nfsdl_list);
2659 if ((dp->nfsdl_flags & NFSCLDL_RECALL)) {
2661 * Wait for outstanding I/O ops to be done.
2663 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
2665 nfsv4_unlock(&clp->nfsc_lock, 0);
2668 dp->nfsdl_rwlock.nfslock_lock |=
2670 msleep(&dp->nfsdl_rwlock,
2671 NFSCLSTATEMUTEXPTR, PVFS, "nfscld",
2673 if (NFSCL_FORCEDISM(mp))
2678 igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
2679 &islept, NFSCLSTATEMUTEXPTR, mp);
2680 if (igotlock == 0 && NFSCL_FORCEDISM(mp))
2686 newnfs_copycred(&dp->nfsdl_cred, cred);
2687 ret = nfscl_recalldeleg(clp, clp->nfsc_nmp, dp,
2688 NULL, cred, p, 1, &vp);
2690 nfscl_cleandeleg(dp);
2691 TAILQ_REMOVE(&clp->nfsc_deleg, dp,
2693 LIST_REMOVE(dp, nfsdl_hash);
2694 TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
2696 nfsstatsv1.cldelegates--;
2700 * The nfsc_lock must be released before doing
2701 * vrele(), since it might call nfs_inactive().
2702 * For the unlikely case where the vnode failed
2703 * to be acquired by nfscl_recalldeleg(), a
2704 * VOP_RECLAIM() should be in progress and it
2705 * will return the delegation.
2707 nfsv4_unlock(&clp->nfsc_lock, 0);
2720 * Clear out old delegations, if we are above the high water
2721 * mark. Only clear out ones with no state related to them.
2722 * The tailq list is in LRU order.
2724 dp = TAILQ_LAST(&clp->nfsc_deleg, nfscldeleghead);
2725 while (nfscl_delegcnt > nfscl_deleghighwater && dp != NULL) {
2726 ndp = TAILQ_PREV(dp, nfscldeleghead, nfsdl_list);
2727 if (dp->nfsdl_rwlock.nfslock_usecnt == 0 &&
2728 dp->nfsdl_rwlock.nfslock_lock == 0 &&
2729 dp->nfsdl_timestamp < NFSD_MONOSEC &&
2730 (dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_ZAPPED |
2731 NFSCLDL_NEEDRECLAIM | NFSCLDL_DELEGRET)) == 0) {
2733 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
2734 op = LIST_FIRST(&owp->nfsow_open);
2741 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
2742 if (!LIST_EMPTY(&lp->nfsl_lock)) {
2749 TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
2750 LIST_REMOVE(dp, nfsdl_hash);
2751 TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
2753 nfsstatsv1.cldelegates--;
2759 nfsv4_unlock(&clp->nfsc_lock, 0);
2762 * Do the recall on any layouts. To avoid trouble, always
2763 * come back up here after having slept.
2767 TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp) {
2768 if ((lyp->nfsly_flags & NFSLY_RECALL) != 0) {
2770 * Wait for outstanding I/O ops to be done.
2772 if (lyp->nfsly_lock.nfslock_usecnt > 0 ||
2773 (lyp->nfsly_lock.nfslock_lock &
2774 NFSV4LOCK_LOCK) != 0) {
2775 lyp->nfsly_lock.nfslock_lock |=
2777 msleep(&lyp->nfsly_lock.nfslock_lock,
2778 NFSCLSTATEMUTEXPTR, PVFS, "nfslyp",
2780 if (NFSCL_FORCEDISM(mp))
2784 /* Move the layout to the recall list. */
2785 TAILQ_REMOVE(&clp->nfsc_layout, lyp,
2787 LIST_REMOVE(lyp, nfsly_hash);
2788 TAILQ_INSERT_HEAD(&rlh, lyp, nfsly_list);
2790 /* Handle any layout commits. */
2791 if (!NFSHASNOLAYOUTCOMMIT(clp->nfsc_nmp) &&
2792 (lyp->nfsly_flags & NFSLY_WRITTEN) != 0) {
2793 lyp->nfsly_flags &= ~NFSLY_WRITTEN;
2795 NFSCL_DEBUG(3, "do layoutcommit\n");
2796 nfscl_dolayoutcommit(clp->nfsc_nmp, lyp,
2804 /* Now, look for stale layouts. */
2805 lyp = TAILQ_LAST(&clp->nfsc_layout, nfscllayouthead);
2806 while (lyp != NULL) {
2807 nlyp = TAILQ_PREV(lyp, nfscllayouthead, nfsly_list);
2808 if (lyp->nfsly_timestamp < NFSD_MONOSEC &&
2809 (lyp->nfsly_flags & NFSLY_RECALL) == 0 &&
2810 lyp->nfsly_lock.nfslock_usecnt == 0 &&
2811 lyp->nfsly_lock.nfslock_lock == 0) {
2812 NFSCL_DEBUG(4, "ret stale lay=%d\n",
2814 recallp = malloc(sizeof(*recallp),
2815 M_NFSLAYRECALL, M_NOWAIT);
2816 if (recallp == NULL)
2818 (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE,
2819 lyp, NFSLAYOUTIOMODE_ANY, 0, UINT64_MAX,
2820 lyp->nfsly_stateid.seqid, 0, 0, NULL,
2827 * Free up any unreferenced device info structures.
2829 LIST_FOREACH_SAFE(dip, &clp->nfsc_devinfo, nfsdi_list, ndip) {
2830 if (dip->nfsdi_layoutrefs == 0 &&
2831 dip->nfsdi_refcnt == 0) {
2832 NFSCL_DEBUG(4, "freeing devinfo\n");
2833 LIST_REMOVE(dip, nfsdi_list);
2834 nfscl_freedevinfo(dip);
2839 /* Do layout return(s), as required. */
2840 TAILQ_FOREACH_SAFE(lyp, &rlh, nfsly_list, nlyp) {
2841 TAILQ_REMOVE(&rlh, lyp, nfsly_list);
2842 NFSCL_DEBUG(4, "ret layout\n");
2843 nfscl_layoutreturn(clp->nfsc_nmp, lyp, cred, p);
2844 nfscl_freelayout(lyp);
2848 * Delegreturn any delegations cleaned out or recalled.
2850 TAILQ_FOREACH_SAFE(dp, &dh, nfsdl_list, ndp) {
2851 newnfs_copycred(&dp->nfsdl_cred, cred);
2852 (void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
2853 TAILQ_REMOVE(&dh, dp, nfsdl_list);
2854 free(dp, M_NFSCLDELEG);
2859 * Call nfscl_cleanupkext() once per second to check for
2860 * open/lock owners where the process has exited.
2862 mytime = NFSD_MONOSEC;
2863 if (prevsec != mytime) {
2865 nfscl_cleanupkext(clp, &lfh);
2869 * Do a ReleaseLockOwner for all lock owners where the
2870 * associated process no longer exists, as found by
2871 * nfscl_cleanupkext().
2873 newnfs_setroot(cred);
2874 SLIST_FOREACH_SAFE(lfhp, &lfh, nfslfh_list, nlfhp) {
2875 LIST_FOREACH_SAFE(lp, &lfhp->nfslfh_lock, nfsl_list,
2877 (void)nfsrpc_rellockown(clp->nfsc_nmp, lp,
2878 lfhp->nfslfh_fh, lfhp->nfslfh_len, cred,
2880 nfscl_freelockowner(lp, 0);
2887 if ((clp->nfsc_flags & NFSCLFLAGS_RECOVER) == 0)
2888 (void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT, "nfscl",
2891 if (clp->nfsc_flags & NFSCLFLAGS_UMOUNT) {
2892 clp->nfsc_flags &= ~NFSCLFLAGS_HASTHREAD;
2895 wakeup((caddr_t)clp);
2903 * Initiate state recovery. Called when NFSERR_STALECLIENTID,
2904 * NFSERR_STALESTATEID or NFSERR_BADSESSION is received.
2907 nfscl_initiate_recovery(struct nfsclclient *clp)
2913 clp->nfsc_flags |= NFSCLFLAGS_RECOVER;
2915 wakeup((caddr_t)clp);
2919 * Dump out the state stuff for debugging.
2922 nfscl_dumpstate(struct nfsmount *nmp, int openowner, int opens,
2923 int lockowner, int locks)
2925 struct nfsclclient *clp;
2926 struct nfsclowner *owp;
2927 struct nfsclopen *op;
2928 struct nfscllockowner *lp;
2929 struct nfscllock *lop;
2930 struct nfscldeleg *dp;
2934 printf("nfscl dumpstate NULL clp\n");
2938 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2939 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
2940 if (openowner && !LIST_EMPTY(&owp->nfsow_open))
2941 printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
2942 owp->nfsow_owner[0], owp->nfsow_owner[1],
2943 owp->nfsow_owner[2], owp->nfsow_owner[3],
2945 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
2947 printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
2948 op->nfso_stateid.other[0], op->nfso_stateid.other[1],
2949 op->nfso_stateid.other[2], op->nfso_opencnt,
2951 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
2953 printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
2954 lp->nfsl_owner[0], lp->nfsl_owner[1],
2955 lp->nfsl_owner[2], lp->nfsl_owner[3],
2957 lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
2958 lp->nfsl_stateid.other[2]);
2959 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
2962 printf("lck typ=%d fst=%ju end=%ju\n",
2963 lop->nfslo_type, (intmax_t)lop->nfslo_first,
2964 (intmax_t)lop->nfslo_end);
2966 printf("lck typ=%d fst=%qd end=%qd\n",
2967 lop->nfslo_type, lop->nfslo_first,
2975 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
2976 if (openowner && !LIST_EMPTY(&owp->nfsow_open))
2977 printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
2978 owp->nfsow_owner[0], owp->nfsow_owner[1],
2979 owp->nfsow_owner[2], owp->nfsow_owner[3],
2981 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
2983 printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
2984 op->nfso_stateid.other[0], op->nfso_stateid.other[1],
2985 op->nfso_stateid.other[2], op->nfso_opencnt,
2987 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
2989 printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
2990 lp->nfsl_owner[0], lp->nfsl_owner[1],
2991 lp->nfsl_owner[2], lp->nfsl_owner[3],
2993 lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
2994 lp->nfsl_stateid.other[2]);
2995 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
2998 printf("lck typ=%d fst=%ju end=%ju\n",
2999 lop->nfslo_type, (intmax_t)lop->nfslo_first,
3000 (intmax_t)lop->nfslo_end);
3002 printf("lck typ=%d fst=%qd end=%qd\n",
3003 lop->nfslo_type, lop->nfslo_first,
3014 * Check for duplicate open owners and opens.
3015 * (Only used as a diagnostic aid.)
3018 nfscl_dupopen(vnode_t vp, int dupopens)
3020 struct nfsclclient *clp;
3021 struct nfsclowner *owp, *owp2;
3022 struct nfsclopen *op, *op2;
3025 clp = VFSTONFS(vp->v_mount)->nm_clp;
3027 printf("nfscl dupopen NULL clp\n");
3030 nfhp = VTONFS(vp)->n_fhp;
3034 * First, search for duplicate owners.
3035 * These should never happen!
3037 LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3038 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3040 !NFSBCMP(owp->nfsow_owner, owp2->nfsow_owner,
3041 NFSV4CL_LOCKNAMELEN)) {
3043 printf("DUP OWNER\n");
3044 nfscl_dumpstate(VFSTONFS(vp->v_mount), 1, 1, 0, 0);
3051 * Now, search for duplicate stateids.
3052 * These shouldn't happen, either.
3054 LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3055 LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
3056 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3057 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3059 (op->nfso_stateid.other[0] != 0 ||
3060 op->nfso_stateid.other[1] != 0 ||
3061 op->nfso_stateid.other[2] != 0) &&
3062 op->nfso_stateid.other[0] == op2->nfso_stateid.other[0] &&
3063 op->nfso_stateid.other[1] == op2->nfso_stateid.other[1] &&
3064 op->nfso_stateid.other[2] == op2->nfso_stateid.other[2]) {
3066 printf("DUP STATEID\n");
3067 nfscl_dumpstate(VFSTONFS(vp->v_mount), 1, 1, 0, 0);
3076 * Now search for duplicate opens.
3077 * Duplicate opens for the same owner
3078 * should never occur. Other duplicates are
3079 * possible and are checked for if "dupopens"
3082 LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3083 LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
3084 if (nfhp->nfh_len == op2->nfso_fhlen &&
3085 !NFSBCMP(nfhp->nfh_fh, op2->nfso_fh, nfhp->nfh_len)) {
3086 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3087 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3088 if (op != op2 && nfhp->nfh_len == op->nfso_fhlen &&
3089 !NFSBCMP(nfhp->nfh_fh, op->nfso_fh, nfhp->nfh_len) &&
3090 (!NFSBCMP(op->nfso_own->nfsow_owner,
3091 op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN) ||
3093 if (!NFSBCMP(op->nfso_own->nfsow_owner,
3094 op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
3096 printf("BADDUP OPEN\n");
3099 printf("DUP OPEN\n");
3101 nfscl_dumpstate(VFSTONFS(vp->v_mount), 1, 1, 0,
3114 * During close, find an open that needs to be dereferenced and
3115 * dereference it. If there are no more opens for this file,
3116 * log a message to that effect.
3117 * Opens aren't actually Close'd until VOP_INACTIVE() is performed
3118 * on the file's vnode.
3119 * This is the safe way, since it is difficult to identify
3120 * which open the close is for and I/O can be performed after the
3121 * close(2) system call when a file is mmap'd.
3122 * If it returns 0 for success, there will be a referenced
3123 * clp returned via clpp.
3126 nfscl_getclose(vnode_t vp, struct nfsclclient **clpp)
3128 struct nfsclclient *clp;
3129 struct nfsclowner *owp;
3130 struct nfsclopen *op;
3131 struct nfscldeleg *dp;
3135 error = nfscl_getcl(vp->v_mount, NULL, NULL, 1, &clp);
3140 nfhp = VTONFS(vp)->n_fhp;
3144 * First, look for one under a delegation that was locally issued
3145 * and just decrement the opencnt for it. Since all my Opens against
3146 * the server are DENY_NONE, I don't see a problem with hanging
3147 * onto them. (It is much easier to use one of the extant Opens
3148 * that I already have on the server when a Delegation is recalled
3149 * than to do fresh Opens.) Someday, I might need to rethink this, but.
3151 dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
3153 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
3154 op = LIST_FIRST(&owp->nfsow_open);
3157 * Since a delegation is for a file, there
3158 * should never be more than one open for
3161 if (LIST_NEXT(op, nfso_list) != NULL)
3162 panic("nfscdeleg opens");
3163 if (notdecr && op->nfso_opencnt > 0) {
3172 /* Now process the opens against the server. */
3173 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3174 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3175 if (op->nfso_fhlen == nfhp->nfh_len &&
3176 !NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
3178 /* Found an open, decrement cnt if possible */
3179 if (notdecr && op->nfso_opencnt > 0) {
3184 * There are more opens, so just return.
3186 if (op->nfso_opencnt > 0) {
3195 printf("nfscl: never fnd open\n");
3200 nfscl_doclose(vnode_t vp, struct nfsclclient **clpp, NFSPROC_T *p)
3202 struct nfsclclient *clp;
3203 struct nfsclowner *owp, *nowp;
3204 struct nfsclopen *op;
3205 struct nfscldeleg *dp;
3207 struct nfsclrecalllayout *recallp;
3210 error = nfscl_getcl(vp->v_mount, NULL, NULL, 1, &clp);
3215 nfhp = VTONFS(vp)->n_fhp;
3216 recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_WAITOK);
3219 * First get rid of the local Open structures, which should be no
3222 dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
3224 LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) {
3225 op = LIST_FIRST(&owp->nfsow_open);
3227 KASSERT((op->nfso_opencnt == 0),
3228 ("nfscl: bad open cnt on deleg"));
3229 nfscl_freeopen(op, 1);
3231 nfscl_freeopenowner(owp, 1);
3235 /* Return any layouts marked return on close. */
3236 nfscl_retoncloselayout(vp, clp, nfhp->nfh_fh, nfhp->nfh_len, &recallp);
3238 /* Now process the opens against the server. */
3240 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3241 op = LIST_FIRST(&owp->nfsow_open);
3242 while (op != NULL) {
3243 if (op->nfso_fhlen == nfhp->nfh_len &&
3244 !NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
3246 /* Found an open, close it. */
3248 KASSERT((op->nfso_opencnt == 0),
3249 ("nfscl: bad open cnt on server (%d)",
3253 nfsrpc_doclose(VFSTONFS(vp->v_mount), op, p);
3257 op = LIST_NEXT(op, nfso_list);
3262 * recallp has been set NULL by nfscl_retoncloselayout() if it was
3263 * used by the function, but calling free() with a NULL pointer is ok.
3265 free(recallp, M_NFSLAYRECALL);
3270 * Return all delegations on this client.
3271 * (Must be called with client sleep lock.)
3274 nfscl_delegreturnall(struct nfsclclient *clp, NFSPROC_T *p)
3276 struct nfscldeleg *dp, *ndp;
3279 cred = newnfs_getcred();
3280 TAILQ_FOREACH_SAFE(dp, &clp->nfsc_deleg, nfsdl_list, ndp) {
3281 nfscl_cleandeleg(dp);
3282 (void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
3283 nfscl_freedeleg(&clp->nfsc_deleg, dp);
3289 * Do a callback RPC.
3292 nfscl_docb(struct nfsrv_descript *nd, NFSPROC_T *p)
3294 int clist, gotseq_ok, i, j, k, op, rcalls;
3296 struct nfsclclient *clp;
3297 struct nfscldeleg *dp = NULL;
3298 int numops, taglen = -1, error = 0, trunc __unused;
3299 u_int32_t minorvers = 0, retops = 0, *retopsp = NULL, *repp, cbident;
3300 u_char tag[NFSV4_SMALLSTR + 1], *tagstr;
3306 nfsattrbit_t attrbits, rattrbits;
3307 nfsv4stateid_t stateid;
3308 uint32_t seqid, slotid = 0, highslot, cachethis __unused;
3309 uint8_t sessionid[NFSX_V4SESSIONID];
3311 struct nfscllayout *lyp;
3312 uint64_t filesid[2], len, off;
3313 int changed, gotone, laytype, recalltype;
3315 struct nfsclrecalllayout *recallp = NULL;
3316 struct nfsclsession *tsep;
3320 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3321 taglen = fxdr_unsigned(int, *tl);
3326 if (taglen <= NFSV4_SMALLSTR)
3329 tagstr = malloc(taglen + 1, M_TEMP, M_WAITOK);
3330 error = nfsrv_mtostr(nd, tagstr, taglen);
3332 if (taglen > NFSV4_SMALLSTR)
3333 free(tagstr, M_TEMP);
3337 (void) nfsm_strtom(nd, tag, taglen);
3338 if (taglen > NFSV4_SMALLSTR) {
3339 free(tagstr, M_TEMP);
3341 NFSM_BUILD(retopsp, u_int32_t *, NFSX_UNSIGNED);
3342 NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3343 minorvers = fxdr_unsigned(u_int32_t, *tl++);
3344 if (minorvers != NFSV4_MINORVERSION &&
3345 minorvers != NFSV41_MINORVERSION &&
3346 minorvers != NFSV42_MINORVERSION)
3347 nd->nd_repstat = NFSERR_MINORVERMISMATCH;
3348 cbident = fxdr_unsigned(u_int32_t, *tl++);
3352 numops = fxdr_unsigned(int, *tl);
3354 * Loop around doing the sub ops.
3356 for (i = 0; i < numops; i++) {
3357 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3358 NFSM_BUILD(repp, u_int32_t *, 2 * NFSX_UNSIGNED);
3360 op = fxdr_unsigned(int, *tl);
3361 if (op < NFSV4OP_CBGETATTR ||
3362 (op > NFSV4OP_CBRECALL && minorvers == NFSV4_MINORVERSION) ||
3363 (op > NFSV4OP_CBNOTIFYDEVID &&
3364 minorvers == NFSV41_MINORVERSION) ||
3365 (op > NFSV4OP_CBOFFLOAD &&
3366 minorvers == NFSV42_MINORVERSION)) {
3367 nd->nd_repstat = NFSERR_OPILLEGAL;
3368 *repp = nfscl_errmap(nd, minorvers);
3372 nd->nd_procnum = op;
3373 if (op < NFSV42_CBNOPS)
3374 nfsstatsv1.cbrpccnt[nd->nd_procnum]++;
3376 case NFSV4OP_CBGETATTR:
3377 NFSCL_DEBUG(4, "cbgetattr\n");
3380 error = nfsm_getfh(nd, &nfhp);
3382 error = nfsrv_getattrbits(nd, &attrbits,
3384 if (error == 0 && i == 0 &&
3385 minorvers != NFSV4_MINORVERSION)
3386 error = NFSERR_OPNOTINSESS;
3388 mp = nfscl_getmnt(minorvers, sessionid, cbident,
3391 error = NFSERR_SERVERFAULT;
3394 error = nfscl_ngetreopen(mp, nfhp->nfh_fh,
3395 nfhp->nfh_len, p, &np);
3400 NFSZERO_ATTRBIT(&rattrbits);
3402 dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
3405 if (NFSISSET_ATTRBIT(&attrbits,
3408 va.va_size = np->n_size;
3412 NFSSETBIT_ATTRBIT(&rattrbits,
3415 if (NFSISSET_ATTRBIT(&attrbits,
3416 NFSATTRBIT_CHANGE)) {
3420 (np->n_flag & NDELEGMOD))
3422 NFSSETBIT_ATTRBIT(&rattrbits,
3426 error = NFSERR_SERVERFAULT;
3434 free(nfhp, M_NFSFH);
3436 (void) nfsv4_fillattr(nd, NULL, NULL, NULL, &va,
3437 NULL, 0, &rattrbits, NULL, p, 0, 0, 0, 0,
3440 case NFSV4OP_CBRECALL:
3441 NFSCL_DEBUG(4, "cbrecall\n");
3442 NFSM_DISSECT(tl, u_int32_t *, NFSX_STATEID +
3444 stateid.seqid = *tl++;
3445 NFSBCOPY((caddr_t)tl, (caddr_t)stateid.other,
3447 tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
3448 trunc = fxdr_unsigned(int, *tl);
3449 error = nfsm_getfh(nd, &nfhp);
3450 if (error == 0 && i == 0 &&
3451 minorvers != NFSV4_MINORVERSION)
3452 error = NFSERR_OPNOTINSESS;
3455 if (minorvers == NFSV4_MINORVERSION)
3456 clp = nfscl_getclnt(cbident);
3458 clp = nfscl_getclntsess(sessionid);
3460 dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
3462 if (dp != NULL && (dp->nfsdl_flags &
3463 NFSCLDL_DELEGRET) == 0) {
3466 wakeup((caddr_t)clp);
3469 error = NFSERR_SERVERFAULT;
3474 free(nfhp, M_NFSFH);
3476 case NFSV4OP_CBLAYOUTRECALL:
3477 NFSCL_DEBUG(4, "cblayrec\n");
3479 NFSM_DISSECT(tl, uint32_t *, 4 * NFSX_UNSIGNED);
3480 laytype = fxdr_unsigned(int, *tl++);
3481 iomode = fxdr_unsigned(uint32_t, *tl++);
3482 if (newnfs_true == *tl++)
3486 recalltype = fxdr_unsigned(int, *tl);
3487 NFSCL_DEBUG(4, "layt=%d iom=%d ch=%d rectyp=%d\n",
3488 laytype, iomode, changed, recalltype);
3489 recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL,
3491 if (laytype != NFSLAYOUT_NFSV4_1_FILES &&
3492 laytype != NFSLAYOUT_FLEXFILE)
3493 error = NFSERR_NOMATCHLAYOUT;
3494 else if (recalltype == NFSLAYOUTRETURN_FILE) {
3495 error = nfsm_getfh(nd, &nfhp);
3496 NFSCL_DEBUG(4, "retfile getfh=%d\n", error);
3499 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_HYPER +
3501 off = fxdr_hyper(tl); tl += 2;
3502 len = fxdr_hyper(tl); tl += 2;
3503 stateid.seqid = fxdr_unsigned(uint32_t, *tl++);
3504 NFSBCOPY(tl, stateid.other, NFSX_STATEIDOTHER);
3505 if (minorvers == NFSV4_MINORVERSION)
3506 error = NFSERR_NOTSUPP;
3508 error = NFSERR_OPNOTINSESS;
3509 NFSCL_DEBUG(4, "off=%ju len=%ju sq=%u err=%d\n",
3510 (uintmax_t)off, (uintmax_t)len,
3511 stateid.seqid, error);
3514 clp = nfscl_getclntsess(sessionid);
3515 NFSCL_DEBUG(4, "cbly clp=%p\n", clp);
3517 lyp = nfscl_findlayout(clp,
3520 NFSCL_DEBUG(4, "cblyp=%p\n",
3525 NFSLY_FLEXFILE)) != 0 &&
3526 !NFSBCMP(stateid.other,
3527 lyp->nfsly_stateid.other,
3528 NFSX_STATEIDOTHER)) {
3538 lyp->nfsly_stateid.seqid)
3539 lyp->nfsly_stateid.seqid =
3547 lyp->nfsly_stateid.seqid);
3550 NFSERR_NOMATCHLAYOUT;
3552 error = NFSERR_NOMATCHLAYOUT;
3555 free(nfhp, M_NFSFH);
3556 } else if (recalltype == NFSLAYOUTRETURN_FSID) {
3557 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_HYPER);
3558 filesid[0] = fxdr_hyper(tl); tl += 2;
3559 filesid[1] = fxdr_hyper(tl); tl += 2;
3562 clp = nfscl_getclntsess(sessionid);
3564 TAILQ_FOREACH(lyp, &clp->nfsc_layout,
3566 if (lyp->nfsly_filesid[0] ==
3568 lyp->nfsly_filesid[1] ==
3575 lyp->nfsly_stateid.seqid,
3585 error = NFSERR_NOMATCHLAYOUT;
3587 error = NFSERR_NOMATCHLAYOUT;
3589 } else if (recalltype == NFSLAYOUTRETURN_ALL) {
3592 clp = nfscl_getclntsess(sessionid);
3594 TAILQ_FOREACH(lyp, &clp->nfsc_layout,
3596 error = nfscl_layoutrecall(
3597 recalltype, lyp, iomode, 0,
3599 lyp->nfsly_stateid.seqid,
3600 0, 0, NULL, recallp);
3607 error = NFSERR_NOMATCHLAYOUT;
3609 error = NFSERR_NOMATCHLAYOUT;
3612 error = NFSERR_NOMATCHLAYOUT;
3613 if (recallp != NULL) {
3614 free(recallp, M_NFSLAYRECALL);
3618 case NFSV4OP_CBSEQUENCE:
3619 NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID +
3621 bcopy(tl, sessionid, NFSX_V4SESSIONID);
3622 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3623 seqid = fxdr_unsigned(uint32_t, *tl++);
3624 slotid = fxdr_unsigned(uint32_t, *tl++);
3625 highslot = fxdr_unsigned(uint32_t, *tl++);
3627 /* Throw away the referring call stuff. */
3628 clist = fxdr_unsigned(int, *tl);
3629 for (j = 0; j < clist; j++) {
3630 NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID +
3632 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3633 rcalls = fxdr_unsigned(int, *tl);
3634 for (k = 0; k < rcalls; k++) {
3635 NFSM_DISSECT(tl, uint32_t *,
3641 clp = nfscl_getclntsess(sessionid);
3643 error = NFSERR_SERVERFAULT;
3645 error = NFSERR_SEQUENCEPOS;
3647 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3648 error = nfsv4_seqsession(seqid, slotid,
3649 highslot, tsep->nfsess_cbslots, &rep,
3650 tsep->nfsess_backslots);
3653 if (error == 0 || error == NFSERR_REPLYFROMCACHE) {
3657 * Handle a reply for a retried
3658 * callback. The reply will be
3659 * re-inserted in the session cache
3660 * by the nfsv4_seqsess_cacherep() call
3663 KASSERT(error == NFSERR_REPLYFROMCACHE,
3664 ("cbsequence: non-NULL rep"));
3665 NFSCL_DEBUG(4, "Got cbretry\n");
3666 m_freem(nd->nd_mreq);
3671 NFSM_BUILD(tl, uint32_t *,
3672 NFSX_V4SESSIONID + 4 * NFSX_UNSIGNED);
3673 bcopy(sessionid, tl, NFSX_V4SESSIONID);
3674 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3675 *tl++ = txdr_unsigned(seqid);
3676 *tl++ = txdr_unsigned(slotid);
3677 *tl++ = txdr_unsigned(NFSV4_CBSLOTS - 1);
3678 *tl = txdr_unsigned(NFSV4_CBSLOTS - 1);
3682 if (i == 0 && minorvers != NFSV4_MINORVERSION)
3683 error = NFSERR_OPNOTINSESS;
3685 NFSCL_DEBUG(1, "unsupp callback %d\n", op);
3686 error = NFSERR_NOTSUPP;
3691 if (error == EBADRPC || error == NFSERR_BADXDR) {
3692 nd->nd_repstat = NFSERR_BADXDR;
3694 nd->nd_repstat = error;
3699 if (nd->nd_repstat) {
3700 *repp = nfscl_errmap(nd, minorvers);
3703 *repp = 0; /* NFS4_OK */
3706 if (recallp != NULL)
3707 free(recallp, M_NFSLAYRECALL);
3709 if (error == EBADRPC || error == NFSERR_BADXDR)
3710 nd->nd_repstat = NFSERR_BADXDR;
3712 printf("nfsv4 comperr1=%d\n", error);
3715 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
3719 *retopsp = txdr_unsigned(retops);
3721 *nd->nd_errp = nfscl_errmap(nd, minorvers);
3723 if (gotseq_ok != 0) {
3724 rep = m_copym(nd->nd_mreq, 0, M_COPYALL, M_WAITOK);
3726 clp = nfscl_getclntsess(sessionid);
3728 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3729 nfsv4_seqsess_cacherep(slotid, tsep->nfsess_cbslots,
3740 * Generate the next cbident value. Basically just increment a static value
3741 * and then check that it isn't already in the list, if it has wrapped around.
3744 nfscl_nextcbident(void)
3746 struct nfsclclient *clp;
3748 static u_int32_t nextcbident = 0;
3749 static int haswrapped = 0;
3752 if (nextcbident == 0)
3756 * Search the clientid list for one already using this cbident.
3761 LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3762 if (clp->nfsc_cbident == nextcbident) {
3772 return (nextcbident);
3776 * Get the mount point related to a given cbident or session and busy it.
3779 nfscl_getmnt(int minorvers, uint8_t *sessionid, u_int32_t cbident,
3780 struct nfsclclient **clpp)
3782 struct nfsclclient *clp;
3785 struct nfsclsession *tsep;
3789 LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3790 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3791 if (minorvers == NFSV4_MINORVERSION) {
3792 if (clp->nfsc_cbident == cbident)
3794 } else if (!NFSBCMP(tsep->nfsess_sessionid, sessionid,
3802 mp = clp->nfsc_nmp->nm_mountp;
3805 error = vfs_busy(mp, 0);
3814 * Get the clientid pointer related to a given cbident.
3816 static struct nfsclclient *
3817 nfscl_getclnt(u_int32_t cbident)
3819 struct nfsclclient *clp;
3821 LIST_FOREACH(clp, &nfsclhead, nfsc_list)
3822 if (clp->nfsc_cbident == cbident)
3828 * Get the clientid pointer related to a given sessionid.
3830 static struct nfsclclient *
3831 nfscl_getclntsess(uint8_t *sessionid)
3833 struct nfsclclient *clp;
3834 struct nfsclsession *tsep;
3836 LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3837 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3838 if (!NFSBCMP(tsep->nfsess_sessionid, sessionid,
3846 * Search for a lock conflict locally on the client. A conflict occurs if
3847 * - not same owner and overlapping byte range and at least one of them is
3848 * a write lock or this is an unlock.
3851 nfscl_localconflict(struct nfsclclient *clp, u_int8_t *fhp, int fhlen,
3852 struct nfscllock *nlop, u_int8_t *own, struct nfscldeleg *dp,
3853 struct nfscllock **lopp)
3855 struct nfsclowner *owp;
3856 struct nfsclopen *op;
3860 ret = nfscl_checkconflict(&dp->nfsdl_lock, nlop, own, lopp);
3864 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3865 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3866 if (op->nfso_fhlen == fhlen &&
3867 !NFSBCMP(op->nfso_fh, fhp, fhlen)) {
3868 ret = nfscl_checkconflict(&op->nfso_lock, nlop,
3879 nfscl_checkconflict(struct nfscllockownerhead *lhp, struct nfscllock *nlop,
3880 u_int8_t *own, struct nfscllock **lopp)
3882 struct nfscllockowner *lp;
3883 struct nfscllock *lop;
3885 LIST_FOREACH(lp, lhp, nfsl_list) {
3886 if (NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
3887 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
3888 if (lop->nfslo_first >= nlop->nfslo_end)
3890 if (lop->nfslo_end <= nlop->nfslo_first)
3892 if (lop->nfslo_type == F_WRLCK ||
3893 nlop->nfslo_type == F_WRLCK ||
3894 nlop->nfslo_type == F_UNLCK) {
3897 return (NFSERR_DENIED);
3906 * Check for a local conflicting lock.
3909 nfscl_lockt(vnode_t vp, struct nfsclclient *clp, u_int64_t off,
3910 u_int64_t len, struct flock *fl, NFSPROC_T *p, void *id, int flags)
3912 struct nfscllock *lop, nlck;
3913 struct nfscldeleg *dp;
3915 u_int8_t own[NFSV4CL_LOCKNAMELEN];
3918 nlck.nfslo_type = fl->l_type;
3919 nlck.nfslo_first = off;
3920 if (len == NFS64BITSSET) {
3921 nlck.nfslo_end = NFS64BITSSET;
3923 nlck.nfslo_end = off + len;
3924 if (nlck.nfslo_end <= nlck.nfslo_first)
3925 return (NFSERR_INVAL);
3928 nfscl_filllockowner(id, own, flags);
3930 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
3931 error = nfscl_localconflict(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len,
3932 &nlck, own, dp, &lop);
3934 fl->l_whence = SEEK_SET;
3935 fl->l_start = lop->nfslo_first;
3936 if (lop->nfslo_end == NFS64BITSSET)
3939 fl->l_len = lop->nfslo_end - lop->nfslo_first;
3940 fl->l_pid = (pid_t)0;
3941 fl->l_type = lop->nfslo_type;
3942 error = -1; /* no RPC required */
3943 } else if (dp != NULL && ((dp->nfsdl_flags & NFSCLDL_WRITE) ||
3944 fl->l_type == F_RDLCK)) {
3946 * The delegation ensures that there isn't a conflicting
3947 * lock on the server, so return -1 to indicate an RPC
3950 fl->l_type = F_UNLCK;
3958 * Handle Recall of a delegation.
3959 * The clp must be exclusive locked when this is called.
3962 nfscl_recalldeleg(struct nfsclclient *clp, struct nfsmount *nmp,
3963 struct nfscldeleg *dp, vnode_t vp, struct ucred *cred, NFSPROC_T *p,
3964 int called_from_renewthread, vnode_t *vpp)
3966 struct nfsclowner *owp, *lowp, *nowp;
3967 struct nfsclopen *op, *lop;
3968 struct nfscllockowner *lp;
3969 struct nfscllock *lckp;
3974 KASSERT(vpp != NULL, ("nfscl_recalldeleg: vpp NULL"));
3977 * First, get a vnode for the file. This is needed to do RPCs.
3979 ret = nfscl_ngetreopen(nmp->nm_mountp, dp->nfsdl_fh,
3980 dp->nfsdl_fhlen, p, &np);
3983 * File isn't open, so nothing to move over to the
3993 dp->nfsdl_flags &= ~NFSCLDL_MODTIMESET;
3996 * Ok, if it's a write delegation, flush data to the server, so
3997 * that close/open consistency is retained.
4001 if ((dp->nfsdl_flags & NFSCLDL_WRITE) && (np->n_flag & NMODIFIED)) {
4002 np->n_flag |= NDELEGRECALL;
4004 ret = ncl_flush(vp, MNT_WAIT, p, 1, called_from_renewthread);
4006 np->n_flag &= ~NDELEGRECALL;
4008 NFSINVALATTRCACHE(np);
4010 if (ret == EIO && called_from_renewthread != 0) {
4012 * If the flush failed with EIO for the renew thread,
4013 * return now, so that the dirty buffer will be flushed
4020 * Now, for each openowner with opens issued locally, move them
4021 * over to state against the server.
4023 LIST_FOREACH(lowp, &dp->nfsdl_owner, nfsow_list) {
4024 lop = LIST_FIRST(&lowp->nfsow_open);
4026 if (LIST_NEXT(lop, nfso_list) != NULL)
4027 panic("nfsdlg mult opens");
4029 * Look for the same openowner against the server.
4031 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
4032 if (!NFSBCMP(lowp->nfsow_owner,
4033 owp->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
4034 newnfs_copycred(&dp->nfsdl_cred, cred);
4035 ret = nfscl_moveopen(vp, clp, nmp, lop,
4037 if (ret == NFSERR_STALECLIENTID ||
4038 ret == NFSERR_STALEDONTRECOVER ||
4039 ret == NFSERR_BADSESSION)
4042 nfscl_freeopen(lop, 1);
4051 * If no openowner found, create one and get an open
4056 sizeof (struct nfsclowner), M_NFSCLOWNER,
4058 nfscl_newopen(clp, NULL, &owp, &nowp, &op,
4059 NULL, lowp->nfsow_owner, dp->nfsdl_fh,
4060 dp->nfsdl_fhlen, NULL, NULL);
4061 newnfs_copycred(&dp->nfsdl_cred, cred);
4062 ret = nfscl_moveopen(vp, clp, nmp, lop,
4065 nfscl_freeopenowner(owp, 0);
4066 if (ret == NFSERR_STALECLIENTID ||
4067 ret == NFSERR_STALEDONTRECOVER ||
4068 ret == NFSERR_BADSESSION)
4071 nfscl_freeopen(lop, 1);
4081 * Now, get byte range locks for any locks done locally.
4083 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4084 LIST_FOREACH(lckp, &lp->nfsl_lock, nfslo_list) {
4085 newnfs_copycred(&dp->nfsdl_cred, cred);
4086 ret = nfscl_relock(vp, clp, nmp, lp, lckp, cred, p);
4087 if (ret == NFSERR_STALESTATEID ||
4088 ret == NFSERR_STALEDONTRECOVER ||
4089 ret == NFSERR_STALECLIENTID ||
4090 ret == NFSERR_BADSESSION)
4100 * Move a locally issued open over to an owner on the state list.
4101 * SIDE EFFECT: If it needs to sleep (do an rpc), it unlocks clstate and
4102 * returns with it unlocked.
4105 nfscl_moveopen(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp,
4106 struct nfsclopen *lop, struct nfsclowner *owp, struct nfscldeleg *dp,
4107 struct ucred *cred, NFSPROC_T *p)
4109 struct nfsclopen *op, *nop;
4110 struct nfscldeleg *ndp;
4112 int error = 0, newone;
4115 * First, look for an appropriate open, If found, just increment the
4118 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
4119 if ((op->nfso_mode & lop->nfso_mode) == lop->nfso_mode &&
4120 op->nfso_fhlen == lop->nfso_fhlen &&
4121 !NFSBCMP(op->nfso_fh, lop->nfso_fh, op->nfso_fhlen)) {
4122 op->nfso_opencnt += lop->nfso_opencnt;
4123 nfscl_freeopen(lop, 1);
4128 /* No appropriate open, so we have to do one against the server. */
4130 nop = malloc(sizeof (struct nfsclopen) +
4131 lop->nfso_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
4133 nfscl_newopen(clp, NULL, &owp, NULL, &op, &nop, owp->nfsow_owner,
4134 lop->nfso_fh, lop->nfso_fhlen, cred, &newone);
4136 error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data, np->n_v4->n4_fhlen,
4137 lop->nfso_fh, lop->nfso_fhlen, lop->nfso_mode, op,
4138 NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, &ndp, 0, 0, cred, p);
4141 nfscl_freeopen(op, 0);
4143 op->nfso_mode |= lop->nfso_mode;
4144 op->nfso_opencnt += lop->nfso_opencnt;
4145 nfscl_freeopen(lop, 1);
4148 free(nop, M_NFSCLOPEN);
4151 * What should I do with the returned delegation, since the
4152 * delegation is being recalled? For now, just printf and
4155 printf("Moveopen returned deleg\n");
4156 free(ndp, M_NFSCLDELEG);
4162 * Recall all delegations on this client.
4165 nfscl_totalrecall(struct nfsclclient *clp)
4167 struct nfscldeleg *dp;
4169 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
4170 if ((dp->nfsdl_flags & NFSCLDL_DELEGRET) == 0)
4171 dp->nfsdl_flags |= NFSCLDL_RECALL;
4176 * Relock byte ranges. Called for delegation recall and state expiry.
4179 nfscl_relock(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp,
4180 struct nfscllockowner *lp, struct nfscllock *lop, struct ucred *cred,
4183 struct nfscllockowner *nlp;
4186 int error, newone, donelocally;
4188 off = lop->nfslo_first;
4189 len = lop->nfslo_end - lop->nfslo_first;
4190 error = nfscl_getbytelock(vp, off, len, lop->nfslo_type, cred, p,
4191 clp, 1, NULL, lp->nfsl_lockflags, lp->nfsl_owner,
4192 lp->nfsl_openowner, &nlp, &newone, &donelocally);
4193 if (error || donelocally)
4195 nfhp = VTONFS(vp)->n_fhp;
4196 error = nfscl_trylock(nmp, vp, nfhp->nfh_fh,
4197 nfhp->nfh_len, nlp, newone, 0, off,
4198 len, lop->nfslo_type, cred, p);
4200 nfscl_freelockowner(nlp, 0);
4205 * Called to re-open a file. Basically get a vnode for the file handle
4206 * and then call nfsrpc_openrpc() to do the rest.
4209 nfsrpc_reopen(struct nfsmount *nmp, u_int8_t *fhp, int fhlen,
4210 u_int32_t mode, struct nfsclopen *op, struct nfscldeleg **dpp,
4211 struct ucred *cred, NFSPROC_T *p)
4217 error = nfscl_ngetreopen(nmp->nm_mountp, fhp, fhlen, p, &np);
4221 if (np->n_v4 != NULL) {
4222 error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data,
4223 np->n_v4->n4_fhlen, fhp, fhlen, mode, op,
4224 NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, dpp, 0, 0,
4234 * Try an open against the server. Just call nfsrpc_openrpc(), retrying while
4235 * NFSERR_DELAY. Also, try system credentials, if the passed in credentials
4239 nfscl_tryopen(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen,
4240 u_int8_t *newfhp, int newfhlen, u_int32_t mode, struct nfsclopen *op,
4241 u_int8_t *name, int namelen, struct nfscldeleg **ndpp,
4242 int reclaim, u_int32_t delegtype, struct ucred *cred, NFSPROC_T *p)
4247 error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp, newfhlen,
4248 mode, op, name, namelen, ndpp, reclaim, delegtype, cred, p,
4250 if (error == NFSERR_DELAY)
4251 (void) nfs_catnap(PZERO, error, "nfstryop");
4252 } while (error == NFSERR_DELAY);
4253 if (error == EAUTH || error == EACCES) {
4254 /* Try again using system credentials */
4255 newnfs_setroot(cred);
4257 error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp,
4258 newfhlen, mode, op, name, namelen, ndpp, reclaim,
4259 delegtype, cred, p, 1, 0);
4260 if (error == NFSERR_DELAY)
4261 (void) nfs_catnap(PZERO, error, "nfstryop");
4262 } while (error == NFSERR_DELAY);
4268 * Try a byte range lock. Just loop on nfsrpc_lock() while it returns
4269 * NFSERR_DELAY. Also, retry with system credentials, if the provided
4273 nfscl_trylock(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp,
4274 int fhlen, struct nfscllockowner *nlp, int newone, int reclaim,
4275 u_int64_t off, u_int64_t len, short type, struct ucred *cred, NFSPROC_T *p)
4277 struct nfsrv_descript nfsd, *nd = &nfsd;
4281 error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp, newone,
4282 reclaim, off, len, type, cred, p, 0);
4283 if (!error && nd->nd_repstat == NFSERR_DELAY)
4284 (void) nfs_catnap(PZERO, (int)nd->nd_repstat,
4286 } while (!error && nd->nd_repstat == NFSERR_DELAY);
4288 error = nd->nd_repstat;
4289 if (error == EAUTH || error == EACCES) {
4290 /* Try again using root credentials */
4291 newnfs_setroot(cred);
4293 error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp,
4294 newone, reclaim, off, len, type, cred, p, 1);
4295 if (!error && nd->nd_repstat == NFSERR_DELAY)
4296 (void) nfs_catnap(PZERO, (int)nd->nd_repstat,
4298 } while (!error && nd->nd_repstat == NFSERR_DELAY);
4300 error = nd->nd_repstat;
4306 * Try a delegreturn against the server. Just call nfsrpc_delegreturn(),
4307 * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
4311 nfscl_trydelegreturn(struct nfscldeleg *dp, struct ucred *cred,
4312 struct nfsmount *nmp, NFSPROC_T *p)
4317 error = nfsrpc_delegreturn(dp, cred, nmp, p, 0);
4318 if (error == NFSERR_DELAY)
4319 (void) nfs_catnap(PZERO, error, "nfstrydp");
4320 } while (error == NFSERR_DELAY);
4321 if (error == EAUTH || error == EACCES) {
4322 /* Try again using system credentials */
4323 newnfs_setroot(cred);
4325 error = nfsrpc_delegreturn(dp, cred, nmp, p, 1);
4326 if (error == NFSERR_DELAY)
4327 (void) nfs_catnap(PZERO, error, "nfstrydp");
4328 } while (error == NFSERR_DELAY);
4334 * Try a close against the server. Just call nfsrpc_closerpc(),
4335 * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
4339 nfscl_tryclose(struct nfsclopen *op, struct ucred *cred,
4340 struct nfsmount *nmp, NFSPROC_T *p)
4342 struct nfsrv_descript nfsd, *nd = &nfsd;
4346 error = nfsrpc_closerpc(nd, nmp, op, cred, p, 0);
4347 if (error == NFSERR_DELAY)
4348 (void) nfs_catnap(PZERO, error, "nfstrycl");
4349 } while (error == NFSERR_DELAY);
4350 if (error == EAUTH || error == EACCES) {
4351 /* Try again using system credentials */
4352 newnfs_setroot(cred);
4354 error = nfsrpc_closerpc(nd, nmp, op, cred, p, 1);
4355 if (error == NFSERR_DELAY)
4356 (void) nfs_catnap(PZERO, error, "nfstrycl");
4357 } while (error == NFSERR_DELAY);
4363 * Decide if a delegation on a file permits close without flushing writes
4364 * to the server. This might be a big performance win in some environments.
4365 * (Not useful until the client does caching on local stable storage.)
4368 nfscl_mustflush(vnode_t vp)
4370 struct nfsclclient *clp;
4371 struct nfscldeleg *dp;
4373 struct nfsmount *nmp;
4376 nmp = VFSTONFS(vp->v_mount);
4377 if (!NFSHASNFSV4(nmp))
4380 clp = nfscl_findcl(nmp);
4385 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4386 if (dp != NULL && (dp->nfsdl_flags &
4387 (NFSCLDL_WRITE | NFSCLDL_RECALL | NFSCLDL_DELEGRET)) ==
4389 (dp->nfsdl_sizelimit >= np->n_size ||
4390 !NFSHASSTRICT3530(nmp))) {
4399 * See if a (write) delegation exists for this file.
4402 nfscl_nodeleg(vnode_t vp, int writedeleg)
4404 struct nfsclclient *clp;
4405 struct nfscldeleg *dp;
4407 struct nfsmount *nmp;
4410 nmp = VFSTONFS(vp->v_mount);
4411 if (!NFSHASNFSV4(nmp))
4414 clp = nfscl_findcl(nmp);
4419 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4421 (dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_DELEGRET)) == 0 &&
4422 (writedeleg == 0 || (dp->nfsdl_flags & NFSCLDL_WRITE) ==
4432 * Look for an associated delegation that should be DelegReturned.
4435 nfscl_removedeleg(vnode_t vp, NFSPROC_T *p, nfsv4stateid_t *stp)
4437 struct nfsclclient *clp;
4438 struct nfscldeleg *dp;
4439 struct nfsclowner *owp;
4440 struct nfscllockowner *lp;
4441 struct nfsmount *nmp;
4444 int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
4446 nmp = VFSTONFS(vp->v_mount);
4450 * Loop around waiting for:
4451 * - outstanding I/O operations on delegations to complete
4452 * - for a delegation on vp that has state, lock the client and
4454 * - return delegation with no state
4457 clp = nfscl_findcl(nmp);
4462 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4463 np->n_fhp->nfh_len);
4466 * Wait for outstanding I/O ops to be done.
4468 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4470 nfsv4_unlock(&clp->nfsc_lock, 0);
4473 dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4474 (void) nfsmsleep(&dp->nfsdl_rwlock,
4475 NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4479 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4480 if (!LIST_EMPTY(&owp->nfsow_open)) {
4486 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4487 if (!LIST_EMPTY(&lp->nfsl_lock)) {
4493 if (needsrecall && !triedrecall) {
4494 dp->nfsdl_flags |= NFSCLDL_DELEGRET;
4497 igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
4498 &islept, NFSCLSTATEMUTEXPTR, NULL);
4505 cred = newnfs_getcred();
4506 newnfs_copycred(&dp->nfsdl_cred, cred);
4507 nfscl_recalldeleg(clp, nmp, dp, vp, cred, p, 0, NULL);
4511 nfsv4_unlock(&clp->nfsc_lock, 0);
4515 *stp = dp->nfsdl_stateid;
4517 nfscl_cleandeleg(dp);
4518 nfscl_freedeleg(&clp->nfsc_deleg, dp);
4521 nfsv4_unlock(&clp->nfsc_lock, 0);
4528 * Look for associated delegation(s) that should be DelegReturned.
4531 nfscl_renamedeleg(vnode_t fvp, nfsv4stateid_t *fstp, int *gotfdp, vnode_t tvp,
4532 nfsv4stateid_t *tstp, int *gottdp, NFSPROC_T *p)
4534 struct nfsclclient *clp;
4535 struct nfscldeleg *dp;
4536 struct nfsclowner *owp;
4537 struct nfscllockowner *lp;
4538 struct nfsmount *nmp;
4541 int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
4543 nmp = VFSTONFS(fvp->v_mount);
4548 * Loop around waiting for:
4549 * - outstanding I/O operations on delegations to complete
4550 * - for a delegation on fvp that has state, lock the client and
4552 * - return delegation(s) with no state.
4555 clp = nfscl_findcl(nmp);
4561 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4562 np->n_fhp->nfh_len);
4563 if (dp != NULL && *gotfdp == 0) {
4565 * Wait for outstanding I/O ops to be done.
4567 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4569 nfsv4_unlock(&clp->nfsc_lock, 0);
4572 dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4573 (void) nfsmsleep(&dp->nfsdl_rwlock,
4574 NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4578 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4579 if (!LIST_EMPTY(&owp->nfsow_open)) {
4585 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4586 if (!LIST_EMPTY(&lp->nfsl_lock)) {
4592 if (needsrecall && !triedrecall) {
4593 dp->nfsdl_flags |= NFSCLDL_DELEGRET;
4596 igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
4597 &islept, NFSCLSTATEMUTEXPTR, NULL);
4604 cred = newnfs_getcred();
4605 newnfs_copycred(&dp->nfsdl_cred, cred);
4606 nfscl_recalldeleg(clp, nmp, dp, fvp, cred, p, 0, NULL);
4610 nfsv4_unlock(&clp->nfsc_lock, 0);
4614 *fstp = dp->nfsdl_stateid;
4617 nfscl_cleandeleg(dp);
4618 nfscl_freedeleg(&clp->nfsc_deleg, dp);
4621 nfsv4_unlock(&clp->nfsc_lock, 0);
4626 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4627 np->n_fhp->nfh_len);
4628 if (dp != NULL && *gottdp == 0) {
4630 * Wait for outstanding I/O ops to be done.
4632 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4633 dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4634 (void) nfsmsleep(&dp->nfsdl_rwlock,
4635 NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4638 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4639 if (!LIST_EMPTY(&owp->nfsow_open)) {
4644 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4645 if (!LIST_EMPTY(&lp->nfsl_lock)) {
4650 *tstp = dp->nfsdl_stateid;
4653 nfscl_cleandeleg(dp);
4654 nfscl_freedeleg(&clp->nfsc_deleg, dp);
4663 * Get a reference on the clientid associated with the mount point.
4664 * Return 1 if success, 0 otherwise.
4667 nfscl_getref(struct nfsmount *nmp)
4669 struct nfsclclient *clp;
4672 clp = nfscl_findcl(nmp);
4677 nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, NULL);
4683 * Release a reference on a clientid acquired with the above call.
4686 nfscl_relref(struct nfsmount *nmp)
4688 struct nfsclclient *clp;
4691 clp = nfscl_findcl(nmp);
4696 nfsv4_relref(&clp->nfsc_lock);
4701 * Save the size attribute in the delegation, since the nfsnode
4705 nfscl_reclaimnode(vnode_t vp)
4707 struct nfsclclient *clp;
4708 struct nfscldeleg *dp;
4709 struct nfsnode *np = VTONFS(vp);
4710 struct nfsmount *nmp;
4712 nmp = VFSTONFS(vp->v_mount);
4713 if (!NFSHASNFSV4(nmp))
4716 clp = nfscl_findcl(nmp);
4721 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4722 if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE))
4723 dp->nfsdl_size = np->n_size;
4728 * Get the saved size attribute in the delegation, since it is a
4729 * newly allocated nfsnode.
4732 nfscl_newnode(vnode_t vp)
4734 struct nfsclclient *clp;
4735 struct nfscldeleg *dp;
4736 struct nfsnode *np = VTONFS(vp);
4737 struct nfsmount *nmp;
4739 nmp = VFSTONFS(vp->v_mount);
4740 if (!NFSHASNFSV4(nmp))
4743 clp = nfscl_findcl(nmp);
4748 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4749 if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE))
4750 np->n_size = dp->nfsdl_size;
4755 * If there is a valid write delegation for this file, set the modtime
4756 * to the local clock time.
4759 nfscl_delegmodtime(vnode_t vp)
4761 struct nfsclclient *clp;
4762 struct nfscldeleg *dp;
4763 struct nfsnode *np = VTONFS(vp);
4764 struct nfsmount *nmp;
4766 nmp = VFSTONFS(vp->v_mount);
4767 if (!NFSHASNFSV4(nmp))
4770 clp = nfscl_findcl(nmp);
4775 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4776 if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE)) {
4777 nanotime(&dp->nfsdl_modtime);
4778 dp->nfsdl_flags |= NFSCLDL_MODTIMESET;
4784 * If there is a valid write delegation for this file with a modtime set,
4785 * put that modtime in mtime.
4788 nfscl_deleggetmodtime(vnode_t vp, struct timespec *mtime)
4790 struct nfsclclient *clp;
4791 struct nfscldeleg *dp;
4792 struct nfsnode *np = VTONFS(vp);
4793 struct nfsmount *nmp;
4795 nmp = VFSTONFS(vp->v_mount);
4796 if (!NFSHASNFSV4(nmp))
4799 clp = nfscl_findcl(nmp);
4804 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4806 (dp->nfsdl_flags & (NFSCLDL_WRITE | NFSCLDL_MODTIMESET)) ==
4807 (NFSCLDL_WRITE | NFSCLDL_MODTIMESET))
4808 *mtime = dp->nfsdl_modtime;
4813 nfscl_errmap(struct nfsrv_descript *nd, u_int32_t minorvers)
4815 short *defaulterrp, *errp;
4817 if (!nd->nd_repstat)
4819 if (nd->nd_procnum == NFSPROC_NOOP)
4820 return (txdr_unsigned(nd->nd_repstat & 0xffff));
4821 if (nd->nd_repstat == EBADRPC)
4822 return (txdr_unsigned(NFSERR_BADXDR));
4823 if (nd->nd_repstat == NFSERR_MINORVERMISMATCH ||
4824 nd->nd_repstat == NFSERR_OPILLEGAL)
4825 return (txdr_unsigned(nd->nd_repstat));
4826 if (nd->nd_repstat >= NFSERR_BADIOMODE && nd->nd_repstat < 20000 &&
4827 minorvers > NFSV4_MINORVERSION) {
4828 /* NFSv4.n error. */
4829 return (txdr_unsigned(nd->nd_repstat));
4831 if (nd->nd_procnum < NFSV4OP_CBNOPS)
4832 errp = defaulterrp = nfscl_cberrmap[nd->nd_procnum];
4834 return (txdr_unsigned(nd->nd_repstat));
4836 if (*errp == (short)nd->nd_repstat)
4837 return (txdr_unsigned(nd->nd_repstat));
4838 return (txdr_unsigned(*defaulterrp));
4842 * Called to find/add a layout to a client.
4843 * This function returns the layout with a refcnt (shared lock) upon
4844 * success (returns 0) or with no lock/refcnt on the layout when an
4845 * error is returned.
4846 * If a layout is passed in via lypp, it is locked (exclusively locked).
4849 nfscl_layout(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen,
4850 nfsv4stateid_t *stateidp, int layouttype, int retonclose,
4851 struct nfsclflayouthead *fhlp, struct nfscllayout **lypp,
4852 struct ucred *cred, NFSPROC_T *p)
4854 struct nfsclclient *clp;
4855 struct nfscllayout *lyp, *tlyp;
4856 struct nfsclflayout *flp;
4857 struct nfsnode *np = VTONFS(vp);
4859 int layout_passed_in;
4861 mp = nmp->nm_mountp;
4862 layout_passed_in = 1;
4866 layout_passed_in = 0;
4867 tlyp = malloc(sizeof(*tlyp) + fhlen - 1, M_NFSLAYOUT,
4874 if (layout_passed_in != 0)
4875 nfsv4_unlock(&lyp->nfsly_lock, 0);
4878 free(tlyp, M_NFSLAYOUT);
4883 * Although no lyp was passed in, another thread might have
4884 * allocated one. If one is found, just increment it's ref
4885 * count and return it.
4887 lyp = nfscl_findlayout(clp, fhp, fhlen);
4891 lyp->nfsly_stateid.seqid = stateidp->seqid;
4892 lyp->nfsly_stateid.other[0] = stateidp->other[0];
4893 lyp->nfsly_stateid.other[1] = stateidp->other[1];
4894 lyp->nfsly_stateid.other[2] = stateidp->other[2];
4895 lyp->nfsly_lastbyte = 0;
4896 LIST_INIT(&lyp->nfsly_flayread);
4897 LIST_INIT(&lyp->nfsly_flayrw);
4898 LIST_INIT(&lyp->nfsly_recall);
4899 lyp->nfsly_filesid[0] = np->n_vattr.na_filesid[0];
4900 lyp->nfsly_filesid[1] = np->n_vattr.na_filesid[1];
4901 lyp->nfsly_clp = clp;
4902 if (layouttype == NFSLAYOUT_FLEXFILE)
4903 lyp->nfsly_flags = NFSLY_FLEXFILE;
4905 lyp->nfsly_flags = NFSLY_FILES;
4906 if (retonclose != 0)
4907 lyp->nfsly_flags |= NFSLY_RETONCLOSE;
4908 lyp->nfsly_fhlen = fhlen;
4909 NFSBCOPY(fhp, lyp->nfsly_fh, fhlen);
4910 TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
4911 LIST_INSERT_HEAD(NFSCLLAYOUTHASH(clp, fhp, fhlen), lyp,
4913 lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
4916 if (retonclose != 0)
4917 lyp->nfsly_flags |= NFSLY_RETONCLOSE;
4918 if (stateidp->seqid > lyp->nfsly_stateid.seqid)
4919 lyp->nfsly_stateid.seqid = stateidp->seqid;
4920 TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list);
4921 TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
4922 lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
4924 nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
4925 if (NFSCL_FORCEDISM(mp)) {
4928 free(tlyp, M_NFSLAYOUT);
4932 } else if (stateidp->seqid > lyp->nfsly_stateid.seqid)
4933 lyp->nfsly_stateid.seqid = stateidp->seqid;
4935 /* Merge the new list of File Layouts into the list. */
4936 flp = LIST_FIRST(fhlp);
4938 if (flp->nfsfl_iomode == NFSLAYOUTIOMODE_READ)
4939 nfscl_mergeflayouts(&lyp->nfsly_flayread, fhlp);
4941 nfscl_mergeflayouts(&lyp->nfsly_flayrw, fhlp);
4943 if (layout_passed_in != 0)
4944 nfsv4_unlock(&lyp->nfsly_lock, 1);
4947 free(tlyp, M_NFSLAYOUT);
4952 * Search for a layout by MDS file handle.
4953 * If one is found, it is returned with a refcnt (shared lock) iff
4954 * retflpp returned non-NULL and locked (exclusive locked) iff retflpp is
4957 struct nfscllayout *
4958 nfscl_getlayout(struct nfsclclient *clp, uint8_t *fhp, int fhlen,
4959 uint64_t off, struct nfsclflayout **retflpp, int *recalledp)
4961 struct nfscllayout *lyp;
4963 int error, igotlock;
4965 mp = clp->nfsc_nmp->nm_mountp;
4969 lyp = nfscl_findlayout(clp, fhp, fhlen);
4971 if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
4972 TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list);
4973 TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
4974 lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
4975 error = nfscl_findlayoutforio(lyp, off,
4976 NFSV4OPEN_ACCESSREAD, retflpp);
4978 nfsv4_getref(&lyp->nfsly_lock, NULL,
4979 NFSCLSTATEMUTEXPTR, mp);
4982 igotlock = nfsv4_lock(&lyp->nfsly_lock,
4983 1, NULL, NFSCLSTATEMUTEXPTR, mp);
4984 } while (igotlock == 0 && !NFSCL_FORCEDISM(mp));
4987 if (NFSCL_FORCEDISM(mp)) {
5001 * Search for a layout by MDS file handle. If one is found, mark in to be
5002 * recalled, if it already marked "return on close".
5005 nfscl_retoncloselayout(vnode_t vp, struct nfsclclient *clp, uint8_t *fhp,
5006 int fhlen, struct nfsclrecalllayout **recallpp)
5008 struct nfscllayout *lyp;
5011 if (vp->v_type != VREG || !NFSHASPNFS(VFSTONFS(vp->v_mount)) ||
5012 nfscl_enablecallb == 0 || nfs_numnfscbd == 0 ||
5013 (VTONFS(vp)->n_flag & NNOLAYOUT) != 0)
5015 lyp = nfscl_findlayout(clp, fhp, fhlen);
5016 if (lyp != NULL && (lyp->nfsly_flags & (NFSLY_RETONCLOSE |
5017 NFSLY_RECALL)) == NFSLY_RETONCLOSE) {
5019 if (!LIST_EMPTY(&lyp->nfsly_flayread))
5020 iomode |= NFSLAYOUTIOMODE_READ;
5021 if (!LIST_EMPTY(&lyp->nfsly_flayrw))
5022 iomode |= NFSLAYOUTIOMODE_RW;
5023 (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE, lyp, iomode,
5024 0, UINT64_MAX, lyp->nfsly_stateid.seqid, 0, 0, NULL,
5026 NFSCL_DEBUG(4, "retoncls recall iomode=%d\n", iomode);
5032 * Mark the layout to be recalled and with an error.
5033 * Also, disable the dsp from further use.
5036 nfscl_dserr(uint32_t op, uint32_t stat, struct nfscldevinfo *dp,
5037 struct nfscllayout *lyp, struct nfsclds *dsp)
5039 struct nfsclrecalllayout *recallp;
5042 printf("DS being disabled, error=%d\n", stat);
5043 /* Set up the return of the layout. */
5044 recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_WAITOK);
5047 if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
5048 if (!LIST_EMPTY(&lyp->nfsly_flayread))
5049 iomode |= NFSLAYOUTIOMODE_READ;
5050 if (!LIST_EMPTY(&lyp->nfsly_flayrw))
5051 iomode |= NFSLAYOUTIOMODE_RW;
5052 (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE, lyp, iomode,
5053 0, UINT64_MAX, lyp->nfsly_stateid.seqid, stat, op,
5054 dp->nfsdi_deviceid, recallp);
5056 NFSCL_DEBUG(4, "nfscl_dserr recall iomode=%d\n", iomode);
5059 free(recallp, M_NFSLAYRECALL);
5062 /* And shut the TCP connection down. */
5063 nfscl_cancelreqs(dsp);
5067 * Cancel all RPCs for this "dsp" by closing the connection.
5068 * Also, mark the session as defunct.
5069 * If NFSCLDS_SAMECONN is set, the connection is shared with other DSs and
5070 * cannot be shut down.
5073 nfscl_cancelreqs(struct nfsclds *dsp)
5075 struct __rpc_client *cl;
5076 static int non_event;
5079 if ((dsp->nfsclds_flags & (NFSCLDS_CLOSED | NFSCLDS_SAMECONN)) == 0 &&
5080 dsp->nfsclds_sockp != NULL &&
5081 dsp->nfsclds_sockp->nr_client != NULL) {
5082 dsp->nfsclds_flags |= NFSCLDS_CLOSED;
5083 cl = dsp->nfsclds_sockp->nr_client;
5084 dsp->nfsclds_sess.nfsess_defunct = 1;
5088 * This 1sec sleep is done to reduce the number of reconnect
5089 * attempts made on the DS while it has failed.
5091 tsleep(&non_event, PVFS, "ndscls", hz);
5098 * Dereference a layout.
5101 nfscl_rellayout(struct nfscllayout *lyp, int exclocked)
5106 nfsv4_unlock(&lyp->nfsly_lock, 0);
5108 nfsv4_relref(&lyp->nfsly_lock);
5113 * Search for a devinfo by deviceid. If one is found, return it after
5114 * acquiring a reference count on it.
5116 struct nfscldevinfo *
5117 nfscl_getdevinfo(struct nfsclclient *clp, uint8_t *deviceid,
5118 struct nfscldevinfo *dip)
5123 dip = nfscl_finddevinfo(clp, deviceid);
5125 dip->nfsdi_refcnt++;
5131 * Dereference a devinfo structure.
5134 nfscl_reldevinfo_locked(struct nfscldevinfo *dip)
5137 dip->nfsdi_refcnt--;
5138 if (dip->nfsdi_refcnt == 0)
5139 wakeup(&dip->nfsdi_refcnt);
5143 * Dereference a devinfo structure.
5146 nfscl_reldevinfo(struct nfscldevinfo *dip)
5150 nfscl_reldevinfo_locked(dip);
5155 * Find a layout for this file handle. Return NULL upon failure.
5157 static struct nfscllayout *
5158 nfscl_findlayout(struct nfsclclient *clp, u_int8_t *fhp, int fhlen)
5160 struct nfscllayout *lyp;
5162 LIST_FOREACH(lyp, NFSCLLAYOUTHASH(clp, fhp, fhlen), nfsly_hash)
5163 if (lyp->nfsly_fhlen == fhlen &&
5164 !NFSBCMP(lyp->nfsly_fh, fhp, fhlen))
5170 * Find a devinfo for this deviceid. Return NULL upon failure.
5172 static struct nfscldevinfo *
5173 nfscl_finddevinfo(struct nfsclclient *clp, uint8_t *deviceid)
5175 struct nfscldevinfo *dip;
5177 LIST_FOREACH(dip, &clp->nfsc_devinfo, nfsdi_list)
5178 if (NFSBCMP(dip->nfsdi_deviceid, deviceid, NFSX_V4DEVICEID)
5185 * Merge the new file layout list into the main one, maintaining it in
5186 * increasing offset order.
5189 nfscl_mergeflayouts(struct nfsclflayouthead *fhlp,
5190 struct nfsclflayouthead *newfhlp)
5192 struct nfsclflayout *flp, *nflp, *prevflp, *tflp;
5194 flp = LIST_FIRST(fhlp);
5196 LIST_FOREACH_SAFE(nflp, newfhlp, nfsfl_list, tflp) {
5197 while (flp != NULL && flp->nfsfl_off < nflp->nfsfl_off) {
5199 flp = LIST_NEXT(flp, nfsfl_list);
5201 if (prevflp == NULL)
5202 LIST_INSERT_HEAD(fhlp, nflp, nfsfl_list);
5204 LIST_INSERT_AFTER(prevflp, nflp, nfsfl_list);
5210 * Add this nfscldevinfo to the client, if it doesn't already exist.
5211 * This function consumes the structure pointed at by dip, if not NULL.
5214 nfscl_adddevinfo(struct nfsmount *nmp, struct nfscldevinfo *dip, int ind,
5215 struct nfsclflayout *flp)
5217 struct nfsclclient *clp;
5218 struct nfscldevinfo *tdip;
5226 free(dip, M_NFSDEVINFO);
5229 if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5230 dev = flp->nfsfl_dev;
5232 dev = flp->nfsfl_ffm[ind].dev;
5233 tdip = nfscl_finddevinfo(clp, dev);
5235 tdip->nfsdi_layoutrefs++;
5236 if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5237 flp->nfsfl_devp = tdip;
5239 flp->nfsfl_ffm[ind].devp = tdip;
5240 nfscl_reldevinfo_locked(tdip);
5243 free(dip, M_NFSDEVINFO);
5247 LIST_INSERT_HEAD(&clp->nfsc_devinfo, dip, nfsdi_list);
5248 dip->nfsdi_layoutrefs = 1;
5249 if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5250 flp->nfsfl_devp = dip;
5252 flp->nfsfl_ffm[ind].devp = dip;
5261 * Free up a layout structure and associated file layout structure(s).
5264 nfscl_freelayout(struct nfscllayout *layp)
5266 struct nfsclflayout *flp, *nflp;
5267 struct nfsclrecalllayout *rp, *nrp;
5269 LIST_FOREACH_SAFE(flp, &layp->nfsly_flayread, nfsfl_list, nflp) {
5270 LIST_REMOVE(flp, nfsfl_list);
5271 nfscl_freeflayout(flp);
5273 LIST_FOREACH_SAFE(flp, &layp->nfsly_flayrw, nfsfl_list, nflp) {
5274 LIST_REMOVE(flp, nfsfl_list);
5275 nfscl_freeflayout(flp);
5277 LIST_FOREACH_SAFE(rp, &layp->nfsly_recall, nfsrecly_list, nrp) {
5278 LIST_REMOVE(rp, nfsrecly_list);
5279 free(rp, M_NFSLAYRECALL);
5282 free(layp, M_NFSLAYOUT);
5286 * Free up a file layout structure.
5289 nfscl_freeflayout(struct nfsclflayout *flp)
5293 if ((flp->nfsfl_flags & NFSFL_FILE) != 0) {
5294 for (i = 0; i < flp->nfsfl_fhcnt; i++)
5295 free(flp->nfsfl_fh[i], M_NFSFH);
5296 if (flp->nfsfl_devp != NULL)
5297 flp->nfsfl_devp->nfsdi_layoutrefs--;
5299 if ((flp->nfsfl_flags & NFSFL_FLEXFILE) != 0)
5300 for (i = 0; i < flp->nfsfl_mirrorcnt; i++) {
5301 for (j = 0; j < flp->nfsfl_ffm[i].fhcnt; j++)
5302 free(flp->nfsfl_ffm[i].fh[j], M_NFSFH);
5303 if (flp->nfsfl_ffm[i].devp != NULL)
5304 flp->nfsfl_ffm[i].devp->nfsdi_layoutrefs--;
5306 free(flp, M_NFSFLAYOUT);
5310 * Free up a file layout devinfo structure.
5313 nfscl_freedevinfo(struct nfscldevinfo *dip)
5316 free(dip, M_NFSDEVINFO);
5320 * Mark any layouts that match as recalled.
5323 nfscl_layoutrecall(int recalltype, struct nfscllayout *lyp, uint32_t iomode,
5324 uint64_t off, uint64_t len, uint32_t stateseqid, uint32_t stat, uint32_t op,
5325 char *devid, struct nfsclrecalllayout *recallp)
5327 struct nfsclrecalllayout *rp, *orp;
5329 recallp->nfsrecly_recalltype = recalltype;
5330 recallp->nfsrecly_iomode = iomode;
5331 recallp->nfsrecly_stateseqid = stateseqid;
5332 recallp->nfsrecly_off = off;
5333 recallp->nfsrecly_len = len;
5334 recallp->nfsrecly_stat = stat;
5335 recallp->nfsrecly_op = op;
5337 NFSBCOPY(devid, recallp->nfsrecly_devid, NFSX_V4DEVICEID);
5339 * Order the list as file returns first, followed by fsid and any
5340 * returns, both in increasing stateseqid order.
5341 * Note that the seqids wrap around, so 1 is after 0xffffffff.
5342 * (I'm not sure this is correct because I find RFC5661 confusing
5343 * on this, but hopefully it will work ok.)
5346 LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) {
5348 if ((recalltype == NFSLAYOUTRETURN_FILE &&
5349 (rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE ||
5350 nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) ||
5351 (recalltype != NFSLAYOUTRETURN_FILE &&
5352 rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE &&
5353 nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) {
5354 LIST_INSERT_BEFORE(rp, recallp, nfsrecly_list);
5359 * Put any error return on all the file returns that will
5362 if (rp->nfsrecly_recalltype == NFSLAYOUTRETURN_FILE &&
5363 stat != 0 && rp->nfsrecly_stat == 0) {
5364 rp->nfsrecly_stat = stat;
5365 rp->nfsrecly_op = op;
5367 NFSBCOPY(devid, rp->nfsrecly_devid,
5373 LIST_INSERT_HEAD(&lyp->nfsly_recall, recallp,
5376 LIST_INSERT_AFTER(orp, recallp, nfsrecly_list);
5378 lyp->nfsly_flags |= NFSLY_RECALL;
5379 wakeup(lyp->nfsly_clp);
5384 * Compare the two seqids for ordering. The trick is that the seqids can
5385 * wrap around from 0xffffffff->0, so check for the cases where one
5386 * has wrapped around.
5387 * Return 1 if seqid1 comes before seqid2, 0 otherwise.
5390 nfscl_seq(uint32_t seqid1, uint32_t seqid2)
5393 if (seqid2 > seqid1 && (seqid2 - seqid1) >= 0x7fffffff)
5394 /* seqid2 has wrapped around. */
5396 if (seqid1 > seqid2 && (seqid1 - seqid2) >= 0x7fffffff)
5397 /* seqid1 has wrapped around. */
5399 if (seqid1 <= seqid2)
5405 * Do a layout return for each of the recalls.
5408 nfscl_layoutreturn(struct nfsmount *nmp, struct nfscllayout *lyp,
5409 struct ucred *cred, NFSPROC_T *p)
5411 struct nfsclrecalllayout *rp;
5412 nfsv4stateid_t stateid;
5415 NFSBCOPY(lyp->nfsly_stateid.other, stateid.other, NFSX_STATEIDOTHER);
5416 stateid.seqid = lyp->nfsly_stateid.seqid;
5417 if ((lyp->nfsly_flags & NFSLY_FILES) != 0)
5418 layouttype = NFSLAYOUT_NFSV4_1_FILES;
5420 layouttype = NFSLAYOUT_FLEXFILE;
5421 LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) {
5422 (void)nfsrpc_layoutreturn(nmp, lyp->nfsly_fh,
5423 lyp->nfsly_fhlen, 0, layouttype,
5424 rp->nfsrecly_iomode, rp->nfsrecly_recalltype,
5425 rp->nfsrecly_off, rp->nfsrecly_len,
5426 &stateid, cred, p, rp->nfsrecly_stat, rp->nfsrecly_op,
5427 rp->nfsrecly_devid);
5432 * Do the layout commit for a file layout.
5435 nfscl_dolayoutcommit(struct nfsmount *nmp, struct nfscllayout *lyp,
5436 struct ucred *cred, NFSPROC_T *p)
5438 struct nfsclflayout *flp;
5440 int error, layouttype;
5442 if ((lyp->nfsly_flags & NFSLY_FILES) != 0)
5443 layouttype = NFSLAYOUT_NFSV4_1_FILES;
5445 layouttype = NFSLAYOUT_FLEXFILE;
5446 LIST_FOREACH(flp, &lyp->nfsly_flayrw, nfsfl_list) {
5447 if (layouttype == NFSLAYOUT_FLEXFILE &&
5448 (flp->nfsfl_fflags & NFSFLEXFLAG_NO_LAYOUTCOMMIT) != 0) {
5449 NFSCL_DEBUG(4, "Flex file: no layoutcommit\n");
5450 /* If not supported, don't bother doing it. */
5452 nmp->nm_state |= NFSSTA_NOLAYOUTCOMMIT;
5455 } else if (flp->nfsfl_off <= lyp->nfsly_lastbyte) {
5456 len = flp->nfsfl_end - flp->nfsfl_off;
5457 error = nfsrpc_layoutcommit(nmp, lyp->nfsly_fh,
5458 lyp->nfsly_fhlen, 0, flp->nfsfl_off, len,
5459 lyp->nfsly_lastbyte, &lyp->nfsly_stateid,
5460 layouttype, cred, p, NULL);
5461 NFSCL_DEBUG(4, "layoutcommit err=%d\n", error);
5462 if (error == NFSERR_NOTSUPP) {
5463 /* If not supported, don't bother doing it. */
5465 nmp->nm_state |= NFSSTA_NOLAYOUTCOMMIT;
5474 * Commit all layouts for a file (vnode).
5477 nfscl_layoutcommit(vnode_t vp, NFSPROC_T *p)
5479 struct nfsclclient *clp;
5480 struct nfscllayout *lyp;
5481 struct nfsnode *np = VTONFS(vp);
5483 struct nfsmount *nmp;
5487 if (NFSHASNOLAYOUTCOMMIT(nmp))
5495 lyp = nfscl_findlayout(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
5500 nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
5501 if (NFSCL_FORCEDISM(mp)) {
5506 if ((lyp->nfsly_flags & NFSLY_WRITTEN) != 0) {
5507 lyp->nfsly_flags &= ~NFSLY_WRITTEN;
5509 NFSCL_DEBUG(4, "do layoutcommit2\n");
5510 nfscl_dolayoutcommit(clp->nfsc_nmp, lyp, NFSPROCCRED(p), p);
5514 nfsv4_relref(&lyp->nfsly_lock);