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 void nfscl_clrelease(struct nfsclclient *);
106 static void nfscl_cleanclient(struct nfsclclient *);
107 static void nfscl_expireclient(struct nfsclclient *, struct nfsmount *,
108 struct ucred *, NFSPROC_T *);
109 static int nfscl_expireopen(struct nfsclclient *, struct nfsclopen *,
110 struct nfsmount *, struct ucred *, NFSPROC_T *);
111 static void nfscl_recover(struct nfsclclient *, struct ucred *, NFSPROC_T *);
112 static void nfscl_insertlock(struct nfscllockowner *, struct nfscllock *,
113 struct nfscllock *, int);
114 static int nfscl_updatelock(struct nfscllockowner *, struct nfscllock **,
115 struct nfscllock **, int);
116 static void nfscl_delegreturnall(struct nfsclclient *, NFSPROC_T *);
117 static u_int32_t nfscl_nextcbident(void);
118 static mount_t nfscl_getmnt(int, uint8_t *, u_int32_t, struct nfsclclient **);
119 static struct nfsclclient *nfscl_getclnt(u_int32_t);
120 static struct nfsclclient *nfscl_getclntsess(uint8_t *);
121 static struct nfscldeleg *nfscl_finddeleg(struct nfsclclient *, u_int8_t *,
123 static void nfscl_retoncloselayout(vnode_t, struct nfsclclient *, uint8_t *,
124 int, struct nfsclrecalllayout **);
125 static void nfscl_reldevinfo_locked(struct nfscldevinfo *);
126 static struct nfscllayout *nfscl_findlayout(struct nfsclclient *, u_int8_t *,
128 static struct nfscldevinfo *nfscl_finddevinfo(struct nfsclclient *, uint8_t *);
129 static int nfscl_checkconflict(struct nfscllockownerhead *, struct nfscllock *,
130 u_int8_t *, struct nfscllock **);
131 static void nfscl_freealllocks(struct nfscllockownerhead *, int);
132 static int nfscl_localconflict(struct nfsclclient *, u_int8_t *, int,
133 struct nfscllock *, u_int8_t *, struct nfscldeleg *, struct nfscllock **);
134 static void nfscl_newopen(struct nfsclclient *, struct nfscldeleg *,
135 struct nfsclowner **, struct nfsclowner **, struct nfsclopen **,
136 struct nfsclopen **, u_int8_t *, u_int8_t *, int, struct ucred *, int *);
137 static int nfscl_moveopen(vnode_t , struct nfsclclient *,
138 struct nfsmount *, struct nfsclopen *, struct nfsclowner *,
139 struct nfscldeleg *, struct ucred *, NFSPROC_T *);
140 static void nfscl_totalrecall(struct nfsclclient *);
141 static int nfscl_relock(vnode_t , struct nfsclclient *, struct nfsmount *,
142 struct nfscllockowner *, struct nfscllock *, struct ucred *, NFSPROC_T *);
143 static int nfscl_tryopen(struct nfsmount *, vnode_t , u_int8_t *, int,
144 u_int8_t *, int, u_int32_t, struct nfsclopen *, u_int8_t *, int,
145 struct nfscldeleg **, int, u_int32_t, struct ucred *, NFSPROC_T *);
146 static int nfscl_trylock(struct nfsmount *, vnode_t , u_int8_t *,
147 int, struct nfscllockowner *, int, int, u_int64_t, u_int64_t, short,
148 struct ucred *, NFSPROC_T *);
149 static int nfsrpc_reopen(struct nfsmount *, u_int8_t *, int, u_int32_t,
150 struct nfsclopen *, struct nfscldeleg **, struct ucred *, NFSPROC_T *);
151 static void nfscl_freedeleg(struct nfscldeleghead *, struct nfscldeleg *);
152 static int nfscl_errmap(struct nfsrv_descript *, u_int32_t);
153 static void nfscl_cleanup_common(struct nfsclclient *, u_int8_t *);
154 static int nfscl_recalldeleg(struct nfsclclient *, struct nfsmount *,
155 struct nfscldeleg *, vnode_t, struct ucred *, NFSPROC_T *, int);
156 static void nfscl_freeopenowner(struct nfsclowner *, int);
157 static void nfscl_cleandeleg(struct nfscldeleg *);
158 static int nfscl_trydelegreturn(struct nfscldeleg *, struct ucred *,
159 struct nfsmount *, NFSPROC_T *);
160 static void nfscl_emptylockowner(struct nfscllockowner *,
161 struct nfscllockownerfhhead *);
162 static void nfscl_mergeflayouts(struct nfsclflayouthead *,
163 struct nfsclflayouthead *);
164 static int nfscl_layoutrecall(int, struct nfscllayout *, uint32_t, uint64_t,
165 uint64_t, uint32_t, uint32_t, uint32_t, char *, struct nfsclrecalllayout *);
166 static int nfscl_seq(uint32_t, uint32_t);
167 static void nfscl_layoutreturn(struct nfsmount *, struct nfscllayout *,
168 struct ucred *, NFSPROC_T *);
169 static void nfscl_dolayoutcommit(struct nfsmount *, struct nfscllayout *,
170 struct ucred *, NFSPROC_T *);
172 static short nfscberr_null[] = {
177 static short nfscberr_getattr[] = {
186 static short nfscberr_recall[] = {
196 static short *nfscl_cberrmap[] = {
204 #define NETFAMILY(clp) \
205 (((clp)->nfsc_flags & NFSCLFLAGS_AFINET6) ? AF_INET6 : AF_INET)
208 * Called for an open operation.
209 * If the nfhp argument is NULL, just get an openowner.
212 nfscl_open(vnode_t vp, u_int8_t *nfhp, int fhlen, u_int32_t amode, int usedeleg,
213 struct ucred *cred, NFSPROC_T *p, struct nfsclowner **owpp,
214 struct nfsclopen **opp, int *newonep, int *retp, int lockit)
216 struct nfsclclient *clp;
217 struct nfsclowner *owp, *nowp;
218 struct nfsclopen *op = NULL, *nop = NULL;
219 struct nfscldeleg *dp;
220 struct nfsclownerhead *ohp;
221 u_int8_t own[NFSV4CL_LOCKNAMELEN];
232 * Might need one or both of these, so MALLOC them now, to
233 * avoid a tsleep() in MALLOC later.
235 nowp = malloc(sizeof (struct nfsclowner),
236 M_NFSCLOWNER, M_WAITOK);
238 nop = malloc(sizeof (struct nfsclopen) +
239 fhlen - 1, M_NFSCLOPEN, M_WAITOK);
240 ret = nfscl_getcl(vnode_mount(vp), cred, p, 1, &clp);
242 free(nowp, M_NFSCLOWNER);
244 free(nop, M_NFSCLOPEN);
249 * Get the Open iff it already exists.
250 * If none found, add the new one or return error, depending upon
255 /* First check the delegation list */
256 if (nfhp != NULL && usedeleg) {
257 LIST_FOREACH(dp, NFSCLDELEGHASH(clp, nfhp, fhlen), nfsdl_hash) {
258 if (dp->nfsdl_fhlen == fhlen &&
259 !NFSBCMP(nfhp, dp->nfsdl_fh, fhlen)) {
260 if (!(amode & NFSV4OPEN_ACCESSWRITE) ||
261 (dp->nfsdl_flags & NFSCLDL_WRITE))
270 nfscl_filllockowner(p->td_proc, own, F_POSIX);
271 ohp = &dp->nfsdl_owner;
273 /* For NFSv4.1 and this option, use a single open_owner. */
274 if (NFSHASONEOPENOWN(VFSTONFS(vnode_mount(vp))))
275 nfscl_filllockowner(NULL, own, F_POSIX);
277 nfscl_filllockowner(p->td_proc, own, F_POSIX);
278 ohp = &clp->nfsc_owner;
280 /* Now, search for an openowner */
281 LIST_FOREACH(owp, ohp, nfsow_list) {
282 if (!NFSBCMP(owp->nfsow_owner, own, NFSV4CL_LOCKNAMELEN))
287 * Create a new open, as required.
289 nfscl_newopen(clp, dp, &owp, &nowp, &op, &nop, own, nfhp, fhlen,
293 * Now, check the mode on the open and return the appropriate
297 if (nfhp != NULL && dp != NULL && nop == NULL)
298 /* new local open on delegation */
299 *retp = NFSCLOPEN_SETCRED;
301 *retp = NFSCLOPEN_OK;
303 if (op != NULL && (amode & ~(op->nfso_mode))) {
304 op->nfso_mode |= amode;
305 if (retp != NULL && dp == NULL)
306 *retp = NFSCLOPEN_DOOPEN;
310 * Serialize modifications to the open owner for multiple threads
311 * within the same process using a read/write sleep lock.
312 * For NFSv4.1 and a single OpenOwner, allow concurrent open operations
313 * by acquiring a shared lock. The close operations still use an
314 * exclusive lock for this case.
317 if (NFSHASONEOPENOWN(VFSTONFS(vnode_mount(vp)))) {
319 * Get a shared lock on the OpenOwner, but first
320 * wait for any pending exclusive lock, so that the
321 * exclusive locker gets priority.
323 nfsv4_lock(&owp->nfsow_rwlock, 0, NULL,
324 NFSCLSTATEMUTEXPTR, NULL);
325 nfsv4_getref(&owp->nfsow_rwlock, NULL,
326 NFSCLSTATEMUTEXPTR, NULL);
328 nfscl_lockexcl(&owp->nfsow_rwlock, NFSCLSTATEMUTEXPTR);
332 free(nowp, M_NFSCLOWNER);
334 free(nop, M_NFSCLOPEN);
343 * Create a new open, as required.
346 nfscl_newopen(struct nfsclclient *clp, struct nfscldeleg *dp,
347 struct nfsclowner **owpp, struct nfsclowner **nowpp, struct nfsclopen **opp,
348 struct nfsclopen **nopp, u_int8_t *own, u_int8_t *fhp, int fhlen,
349 struct ucred *cred, int *newonep)
351 struct nfsclowner *owp = *owpp, *nowp;
352 struct nfsclopen *op, *nop;
362 if (owp == NULL && nowp != NULL) {
363 NFSBCOPY(own, nowp->nfsow_owner, NFSV4CL_LOCKNAMELEN);
364 LIST_INIT(&nowp->nfsow_open);
365 nowp->nfsow_clp = clp;
366 nowp->nfsow_seqid = 0;
367 nowp->nfsow_defunct = 0;
368 nfscl_lockinit(&nowp->nfsow_rwlock);
370 nfsstatsv1.cllocalopenowners++;
371 LIST_INSERT_HEAD(&dp->nfsdl_owner, nowp, nfsow_list);
373 nfsstatsv1.clopenowners++;
374 LIST_INSERT_HEAD(&clp->nfsc_owner, nowp, nfsow_list);
382 /* If an fhp has been specified, create an Open as well. */
384 /* and look for the correct open, based upon FH */
385 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
386 if (op->nfso_fhlen == fhlen &&
387 !NFSBCMP(op->nfso_fh, fhp, fhlen))
390 if (op == NULL && nop != NULL) {
393 nop->nfso_opencnt = 0;
394 nop->nfso_posixlock = 1;
395 nop->nfso_fhlen = fhlen;
396 NFSBCOPY(fhp, nop->nfso_fh, fhlen);
397 LIST_INIT(&nop->nfso_lock);
398 nop->nfso_stateid.seqid = 0;
399 nop->nfso_stateid.other[0] = 0;
400 nop->nfso_stateid.other[1] = 0;
401 nop->nfso_stateid.other[2] = 0;
402 KASSERT(cred != NULL, ("%s: cred NULL\n", __func__));
403 newnfs_copyincred(cred, &nop->nfso_cred);
405 TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
406 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp,
408 dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
409 nfsstatsv1.cllocalopens++;
411 nfsstatsv1.clopens++;
413 LIST_INSERT_HEAD(&owp->nfsow_open, nop, nfso_list);
425 * Called to find/add a delegation to a client.
428 nfscl_deleg(mount_t mp, struct nfsclclient *clp, u_int8_t *nfhp,
429 int fhlen, struct ucred *cred, NFSPROC_T *p, struct nfscldeleg **dpp)
431 struct nfscldeleg *dp = *dpp, *tdp;
434 * First, if we have received a Read delegation for a file on a
435 * read/write file system, just return it, because they aren't
438 if (mp != NULL && dp != NULL && !NFSMNT_RDONLY(mp) &&
439 (dp->nfsdl_flags & NFSCLDL_READ)) {
440 (void) nfscl_trydelegreturn(dp, cred, VFSTONFS(mp), p);
441 free(dp, M_NFSCLDELEG);
446 /* Look for the correct deleg, based upon FH */
448 tdp = nfscl_finddeleg(clp, nfhp, fhlen);
452 return (NFSERR_BADSTATEID);
455 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list);
456 LIST_INSERT_HEAD(NFSCLDELEGHASH(clp, nfhp, fhlen), dp,
458 dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
459 nfsstatsv1.cldelegates++;
463 * Delegation already exists, what do we do if a new one??
466 printf("Deleg already exists!\n");
467 free(dp, M_NFSCLDELEG);
478 * Find a delegation for this file handle. Return NULL upon failure.
480 static struct nfscldeleg *
481 nfscl_finddeleg(struct nfsclclient *clp, u_int8_t *fhp, int fhlen)
483 struct nfscldeleg *dp;
485 LIST_FOREACH(dp, NFSCLDELEGHASH(clp, fhp, fhlen), nfsdl_hash) {
486 if (dp->nfsdl_fhlen == fhlen &&
487 !NFSBCMP(dp->nfsdl_fh, fhp, fhlen))
494 * Get a stateid for an I/O operation. First, look for an open and iff
495 * found, return either a lockowner stateid or the open stateid.
496 * If no Open is found, just return error and the special stateid of all zeros.
499 nfscl_getstateid(vnode_t vp, u_int8_t *nfhp, int fhlen, u_int32_t mode,
500 int fords, struct ucred *cred, NFSPROC_T *p, nfsv4stateid_t *stateidp,
503 struct nfsclclient *clp;
504 struct nfsclowner *owp;
505 struct nfsclopen *op = NULL, *top;
506 struct nfscllockowner *lp;
507 struct nfscldeleg *dp;
509 struct nfsmount *nmp;
510 u_int8_t own[NFSV4CL_LOCKNAMELEN];
515 * Initially, just set the special stateid of all zeros.
516 * (Don't do this for a DS, since the special stateid can't be used.)
520 stateidp->other[0] = 0;
521 stateidp->other[1] = 0;
522 stateidp->other[2] = 0;
524 if (vnode_vtype(vp) != VREG)
527 nmp = VFSTONFS(vnode_mount(vp));
529 clp = nfscl_findcl(nmp);
536 * Wait for recovery to complete.
538 while ((clp->nfsc_flags & NFSCLFLAGS_RECVRINPROG))
539 (void) nfsmsleep(&clp->nfsc_flags, NFSCLSTATEMUTEXPTR,
540 PZERO, "nfsrecvr", NULL);
543 * First, look for a delegation.
545 LIST_FOREACH(dp, NFSCLDELEGHASH(clp, nfhp, fhlen), nfsdl_hash) {
546 if (dp->nfsdl_fhlen == fhlen &&
547 !NFSBCMP(nfhp, dp->nfsdl_fh, fhlen)) {
548 if (!(mode & NFSV4OPEN_ACCESSWRITE) ||
549 (dp->nfsdl_flags & NFSCLDL_WRITE)) {
550 stateidp->seqid = dp->nfsdl_stateid.seqid;
551 stateidp->other[0] = dp->nfsdl_stateid.other[0];
552 stateidp->other[1] = dp->nfsdl_stateid.other[1];
553 stateidp->other[2] = dp->nfsdl_stateid.other[2];
554 if (!(np->n_flag & NDELEGRECALL)) {
555 TAILQ_REMOVE(&clp->nfsc_deleg, dp,
557 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp,
559 dp->nfsdl_timestamp = NFSD_MONOSEC +
561 dp->nfsdl_rwlock.nfslock_usecnt++;
562 *lckpp = (void *)&dp->nfsdl_rwlock;
573 * If p != NULL, we want to search the parentage tree
574 * for a matching OpenOwner and use that.
576 if (NFSHASONEOPENOWN(VFSTONFS(vnode_mount(vp))))
577 nfscl_filllockowner(NULL, own, F_POSIX);
579 nfscl_filllockowner(p->td_proc, own, F_POSIX);
581 error = nfscl_getopen(&clp->nfsc_owner, nfhp, fhlen, own, own,
583 if (error == 0 && lp != NULL && fords == 0) {
584 /* Don't return a lock stateid for a DS. */
586 lp->nfsl_stateid.seqid;
588 lp->nfsl_stateid.other[0];
590 lp->nfsl_stateid.other[1];
592 lp->nfsl_stateid.other[2];
598 /* If not found, just look for any OpenOwner that will work. */
601 owp = LIST_FIRST(&clp->nfsc_owner);
602 while (!done && owp != NULL) {
603 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
604 if (op->nfso_fhlen == fhlen &&
605 !NFSBCMP(op->nfso_fh, nfhp, fhlen)) {
606 if (top == NULL && (op->nfso_mode &
607 NFSV4OPEN_ACCESSWRITE) != 0 &&
608 (mode & NFSV4OPEN_ACCESSREAD) != 0)
610 if ((mode & op->nfso_mode) == mode) {
617 owp = LIST_NEXT(owp, nfsow_list);
620 NFSCL_DEBUG(2, "openmode top=%p\n", top);
621 if (top == NULL || NFSHASOPENMODE(nmp)) {
628 * For read aheads or write behinds, use the open cred.
629 * A read ahead or write behind is indicated by p == NULL.
632 newnfs_copycred(&op->nfso_cred, cred);
636 * No lock stateid, so return the open stateid.
638 stateidp->seqid = op->nfso_stateid.seqid;
639 stateidp->other[0] = op->nfso_stateid.other[0];
640 stateidp->other[1] = op->nfso_stateid.other[1];
641 stateidp->other[2] = op->nfso_stateid.other[2];
647 * Search for a matching file, mode and, optionally, lockowner.
650 nfscl_getopen(struct nfsclownerhead *ohp, u_int8_t *nfhp, int fhlen,
651 u_int8_t *openown, u_int8_t *lockown, u_int32_t mode,
652 struct nfscllockowner **lpp, struct nfsclopen **opp)
654 struct nfsclowner *owp;
655 struct nfsclopen *op, *rop, *rop2;
656 struct nfscllockowner *lp;
662 * rop will be set to the open to be returned. There are three
663 * variants of this, all for an open of the correct file:
664 * 1 - A match of lockown.
665 * 2 - A match of the openown, when no lockown match exists.
666 * 3 - A match for any open, if no openown or lockown match exists.
667 * Looking for #2 over #3 probably isn't necessary, but since
668 * RFC3530 is vague w.r.t. the relationship between openowners and
669 * lockowners, I think this is the safer way to go.
674 /* Search the client list */
675 owp = LIST_FIRST(ohp);
676 while (owp != NULL && keep_looping != 0) {
677 /* and look for the correct open */
678 op = LIST_FIRST(&owp->nfsow_open);
679 while (op != NULL && keep_looping != 0) {
680 if (op->nfso_fhlen == fhlen &&
681 !NFSBCMP(op->nfso_fh, nfhp, fhlen)
682 && (op->nfso_mode & mode) == mode) {
684 /* Now look for a matching lockowner. */
685 LIST_FOREACH(lp, &op->nfso_lock,
687 if (!NFSBCMP(lp->nfsl_owner,
689 NFSV4CL_LOCKNAMELEN)) {
697 if (rop == NULL && !NFSBCMP(owp->nfsow_owner,
698 openown, NFSV4CL_LOCKNAMELEN)) {
706 op = LIST_NEXT(op, nfso_list);
708 owp = LIST_NEXT(owp, nfsow_list);
719 * Release use of an open owner. Called when open operations are done
720 * with the open owner.
723 nfscl_ownerrelease(struct nfsmount *nmp, struct nfsclowner *owp,
724 __unused int error, __unused int candelete, int unlocked)
731 if (NFSHASONEOPENOWN(nmp))
732 nfsv4_relref(&owp->nfsow_rwlock);
734 nfscl_lockunlock(&owp->nfsow_rwlock);
736 nfscl_clrelease(owp->nfsow_clp);
741 * Release use of an open structure under an open owner.
744 nfscl_openrelease(struct nfsmount *nmp, struct nfsclopen *op, int error,
747 struct nfsclclient *clp;
748 struct nfsclowner *owp;
754 if (NFSHASONEOPENOWN(nmp))
755 nfsv4_relref(&owp->nfsow_rwlock);
757 nfscl_lockunlock(&owp->nfsow_rwlock);
758 clp = owp->nfsow_clp;
759 if (error && candelete && op->nfso_opencnt == 0)
760 nfscl_freeopen(op, 0);
761 nfscl_clrelease(clp);
766 * Called to get a clientid structure. It will optionally lock the
767 * client data structures to do the SetClientId/SetClientId_confirm,
768 * but will release that lock and return the clientid with a reference
770 * If the "cred" argument is NULL, a new clientid should not be created.
771 * If the "p" argument is NULL, a SetClientID/SetClientIDConfirm cannot
773 * The start_renewthread argument tells nfscl_getcl() to start a renew
774 * thread if this creates a new clp.
775 * It always clpp with a reference count on it, unless returning an error.
778 nfscl_getcl(struct mount *mp, struct ucred *cred, NFSPROC_T *p,
779 int start_renewthread, struct nfsclclient **clpp)
781 struct nfsclclient *clp;
782 struct nfsclclient *newclp = NULL;
783 struct nfsmount *nmp;
784 char uuid[HOSTUUIDLEN];
785 int igotlock = 0, error, trystalecnt, clidinusedelay, i;
790 getcredhostuuid(cred, uuid, sizeof uuid);
791 idlen = strlen(uuid);
793 idlen += sizeof (u_int64_t);
795 idlen += sizeof (u_int64_t) + 16; /* 16 random bytes */
797 sizeof (struct nfsclclient) + idlen - 1, M_NFSCLCLIENT,
802 * If a forced dismount is already in progress, don't
803 * allocate a new clientid and get out now. For the case where
804 * clp != NULL, this is a harmless optimization.
806 if (NFSCL_FORCEDISM(mp)) {
809 free(newclp, M_NFSCLCLIENT);
814 if (newclp == NULL) {
819 clp->nfsc_idlen = idlen;
820 LIST_INIT(&clp->nfsc_owner);
821 TAILQ_INIT(&clp->nfsc_deleg);
822 TAILQ_INIT(&clp->nfsc_layout);
823 LIST_INIT(&clp->nfsc_devinfo);
824 for (i = 0; i < NFSCLDELEGHASHSIZE; i++)
825 LIST_INIT(&clp->nfsc_deleghash[i]);
826 for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++)
827 LIST_INIT(&clp->nfsc_layouthash[i]);
828 clp->nfsc_flags = NFSCLFLAGS_INITED;
829 clp->nfsc_clientidrev = 1;
830 clp->nfsc_cbident = nfscl_nextcbident();
831 nfscl_fillclid(nmp->nm_clval, uuid, clp->nfsc_id,
833 LIST_INSERT_HEAD(&nfsclhead, clp, nfsc_list);
837 if (start_renewthread != 0)
838 nfscl_start_renewthread(clp);
842 free(newclp, M_NFSCLCLIENT);
845 while ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0 && !igotlock &&
846 !NFSCL_FORCEDISM(mp))
847 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
848 NFSCLSTATEMUTEXPTR, mp);
851 * Call nfsv4_lock() with "iwantlock == 0" so that it will
852 * wait for a pending exclusive lock request. This gives the
853 * exclusive lock request priority over this shared lock
855 * An exclusive lock on nfsc_lock is used mainly for server
858 nfsv4_lock(&clp->nfsc_lock, 0, NULL, NFSCLSTATEMUTEXPTR, mp);
859 nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
861 if (igotlock == 0 && NFSCL_FORCEDISM(mp)) {
863 * Both nfsv4_lock() and nfsv4_getref() know to check
864 * for NFSCL_FORCEDISM() and return without sleeping to
865 * wait for the exclusive lock to be released, since it
866 * might be held by nfscl_umount() and we need to get out
867 * now for that case and not wait until nfscl_umount()
876 * If it needs a clientid, do the setclientid now.
878 if ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0) {
880 panic("nfscl_clget");
881 if (p == NULL || cred == NULL) {
883 nfsv4_unlock(&clp->nfsc_lock, 0);
888 * If RFC3530 Sec. 14.2.33 is taken literally,
889 * NFSERR_CLIDINUSE will be returned persistently for the
890 * case where a new mount of the same file system is using
891 * a different principal. In practice, NFSERR_CLIDINUSE is
892 * only returned when there is outstanding unexpired state
893 * on the clientid. As such, try for twice the lease
894 * interval, if we know what that is. Otherwise, make a
896 * The case of returning NFSERR_STALECLIENTID is far less
897 * likely, but might occur if there is a significant delay
898 * between doing the SetClientID and SetClientIDConfirm Ops,
899 * such that the server throws away the clientid before
900 * receiving the SetClientIDConfirm.
902 if (clp->nfsc_renew > 0)
903 clidinusedelay = NFSCL_LEASE(clp->nfsc_renew) * 2;
905 clidinusedelay = 120;
908 error = nfsrpc_setclient(nmp, clp, 0, cred, p);
909 if (error == NFSERR_STALECLIENTID ||
910 error == NFSERR_STALEDONTRECOVER ||
911 error == NFSERR_BADSESSION ||
912 error == NFSERR_CLIDINUSE) {
913 (void) nfs_catnap(PZERO, error, "nfs_setcl");
915 } while (((error == NFSERR_STALECLIENTID ||
916 error == NFSERR_BADSESSION ||
917 error == NFSERR_STALEDONTRECOVER) && --trystalecnt > 0) ||
918 (error == NFSERR_CLIDINUSE && --clidinusedelay > 0));
921 nfsv4_unlock(&clp->nfsc_lock, 0);
925 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
929 nfsv4_unlock(&clp->nfsc_lock, 1);
938 * Get a reference to a clientid and return it, if valid.
941 nfscl_findcl(struct nfsmount *nmp)
943 struct nfsclclient *clp;
946 if (clp == NULL || !(clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID))
952 * Release the clientid structure. It may be locked or reference counted.
955 nfscl_clrelease(struct nfsclclient *clp)
958 if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK)
959 nfsv4_unlock(&clp->nfsc_lock, 0);
961 nfsv4_relref(&clp->nfsc_lock);
965 * External call for nfscl_clrelease.
968 nfscl_clientrelease(struct nfsclclient *clp)
972 if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK)
973 nfsv4_unlock(&clp->nfsc_lock, 0);
975 nfsv4_relref(&clp->nfsc_lock);
980 * Called when wanting to lock a byte region.
983 nfscl_getbytelock(vnode_t vp, u_int64_t off, u_int64_t len,
984 short type, struct ucred *cred, NFSPROC_T *p, struct nfsclclient *rclp,
985 int recovery, void *id, int flags, u_int8_t *rownp, u_int8_t *ropenownp,
986 struct nfscllockowner **lpp, int *newonep, int *donelocallyp)
988 struct nfscllockowner *lp;
989 struct nfsclopen *op;
990 struct nfsclclient *clp;
991 struct nfscllockowner *nlp;
992 struct nfscllock *nlop, *otherlop;
993 struct nfscldeleg *dp = NULL, *ldp = NULL;
994 struct nfscllockownerhead *lhp = NULL;
996 u_int8_t own[NFSV4CL_LOCKNAMELEN], *ownp, openown[NFSV4CL_LOCKNAMELEN];
998 int error = 0, ret, donelocally = 0;
1001 /* For Lock Ops, the open mode doesn't matter, so use 0 to match any. */
1010 * Might need these, so MALLOC them now, to
1011 * avoid a tsleep() in MALLOC later.
1014 sizeof (struct nfscllockowner), M_NFSCLLOCKOWNER, M_WAITOK);
1016 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1018 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1019 nlop->nfslo_type = type;
1020 nlop->nfslo_first = off;
1021 if (len == NFS64BITSSET) {
1022 nlop->nfslo_end = NFS64BITSSET;
1024 nlop->nfslo_end = off + len;
1025 if (nlop->nfslo_end <= nlop->nfslo_first)
1026 error = NFSERR_INVAL;
1033 error = nfscl_getcl(vnode_mount(vp), cred, p, 1, &clp);
1036 free(nlp, M_NFSCLLOCKOWNER);
1037 free(otherlop, M_NFSCLLOCK);
1038 free(nlop, M_NFSCLLOCK);
1045 openownp = ropenownp;
1047 nfscl_filllockowner(id, own, flags);
1049 if (NFSHASONEOPENOWN(VFSTONFS(vnode_mount(vp))))
1050 nfscl_filllockowner(NULL, openown, F_POSIX);
1052 nfscl_filllockowner(p->td_proc, openown, F_POSIX);
1058 * First, search for a delegation. If one exists for this file,
1059 * the lock can be done locally against it, so long as there
1060 * isn't a local lock conflict.
1062 ldp = dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
1063 np->n_fhp->nfh_len);
1064 /* Just sanity check for correct type of delegation */
1065 if (dp != NULL && ((dp->nfsdl_flags &
1066 (NFSCLDL_RECALL | NFSCLDL_DELEGRET)) != 0 ||
1068 (dp->nfsdl_flags & NFSCLDL_WRITE) == 0)))
1072 /* Now, find an open and maybe a lockowner. */
1073 ret = nfscl_getopen(&dp->nfsdl_owner, np->n_fhp->nfh_fh,
1074 np->n_fhp->nfh_len, openownp, ownp, mode, NULL, &op);
1076 ret = nfscl_getopen(&clp->nfsc_owner,
1077 np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp,
1078 ownp, mode, NULL, &op);
1080 lhp = &dp->nfsdl_lock;
1081 TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
1082 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list);
1083 dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
1091 * Get the related Open and maybe lockowner.
1093 error = nfscl_getopen(&clp->nfsc_owner,
1094 np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp,
1095 ownp, mode, &lp, &op);
1097 lhp = &op->nfso_lock;
1099 if (!error && !recovery)
1100 error = nfscl_localconflict(clp, np->n_fhp->nfh_fh,
1101 np->n_fhp->nfh_len, nlop, ownp, ldp, NULL);
1104 nfscl_clrelease(clp);
1107 free(nlp, M_NFSCLLOCKOWNER);
1108 free(otherlop, M_NFSCLLOCK);
1109 free(nlop, M_NFSCLLOCK);
1114 * Ok, see if a lockowner exists and create one, as required.
1117 LIST_FOREACH(lp, lhp, nfsl_list) {
1118 if (!NFSBCMP(lp->nfsl_owner, ownp, NFSV4CL_LOCKNAMELEN))
1122 NFSBCOPY(ownp, nlp->nfsl_owner, NFSV4CL_LOCKNAMELEN);
1124 NFSBCOPY(ropenownp, nlp->nfsl_openowner,
1125 NFSV4CL_LOCKNAMELEN);
1127 NFSBCOPY(op->nfso_own->nfsow_owner, nlp->nfsl_openowner,
1128 NFSV4CL_LOCKNAMELEN);
1129 nlp->nfsl_seqid = 0;
1130 nlp->nfsl_lockflags = flags;
1131 nlp->nfsl_inprog = NULL;
1132 nfscl_lockinit(&nlp->nfsl_rwlock);
1133 LIST_INIT(&nlp->nfsl_lock);
1135 nlp->nfsl_open = NULL;
1136 nfsstatsv1.cllocallockowners++;
1138 nlp->nfsl_open = op;
1139 nfsstatsv1.cllockowners++;
1141 LIST_INSERT_HEAD(lhp, nlp, nfsl_list);
1148 * Now, update the byte ranges for locks.
1150 ret = nfscl_updatelock(lp, &nlop, &otherlop, donelocally);
1156 nfscl_clrelease(clp);
1159 * Serial modifications on the lock owner for multiple threads
1160 * for the same process using a read/write lock.
1163 nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR);
1169 free(nlp, M_NFSCLLOCKOWNER);
1171 free(nlop, M_NFSCLLOCK);
1173 free(otherlop, M_NFSCLLOCK);
1180 * Called to unlock a byte range, for LockU.
1183 nfscl_relbytelock(vnode_t vp, u_int64_t off, u_int64_t len,
1184 __unused struct ucred *cred, NFSPROC_T *p, int callcnt,
1185 struct nfsclclient *clp, void *id, int flags,
1186 struct nfscllockowner **lpp, int *dorpcp)
1188 struct nfscllockowner *lp;
1189 struct nfsclowner *owp;
1190 struct nfsclopen *op;
1191 struct nfscllock *nlop, *other_lop = NULL;
1192 struct nfscldeleg *dp;
1194 u_int8_t own[NFSV4CL_LOCKNAMELEN];
1202 * Might need these, so MALLOC them now, to
1203 * avoid a tsleep() in MALLOC later.
1206 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1207 nlop->nfslo_type = F_UNLCK;
1208 nlop->nfslo_first = off;
1209 if (len == NFS64BITSSET) {
1210 nlop->nfslo_end = NFS64BITSSET;
1212 nlop->nfslo_end = off + len;
1213 if (nlop->nfslo_end <= nlop->nfslo_first) {
1214 free(nlop, M_NFSCLLOCK);
1215 return (NFSERR_INVAL);
1220 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1223 nfscl_filllockowner(id, own, flags);
1227 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
1228 np->n_fhp->nfh_len);
1231 * First, unlock any local regions on a delegation.
1234 /* Look for this lockowner. */
1235 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
1236 if (!NFSBCMP(lp->nfsl_owner, own,
1237 NFSV4CL_LOCKNAMELEN))
1241 /* Use other_lop, so nlop is still available */
1242 (void)nfscl_updatelock(lp, &other_lop, NULL, 1);
1246 * Now, find a matching open/lockowner that hasn't already been done,
1247 * as marked by nfsl_inprog.
1251 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
1252 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1253 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1254 !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1255 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1256 if (lp->nfsl_inprog == NULL &&
1257 !NFSBCMP(lp->nfsl_owner, own,
1258 NFSV4CL_LOCKNAMELEN)) {
1272 ret = nfscl_updatelock(lp, &nlop, NULL, 0);
1276 * Serial modifications on the lock owner for multiple
1277 * threads for the same process using a read/write lock.
1279 lp->nfsl_inprog = p;
1280 nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR);
1285 free(nlop, M_NFSCLLOCK);
1287 free(other_lop, M_NFSCLLOCK);
1292 * Release all lockowners marked in progess for this process and file.
1295 nfscl_releasealllocks(struct nfsclclient *clp, vnode_t vp, NFSPROC_T *p,
1296 void *id, int flags)
1298 struct nfsclowner *owp;
1299 struct nfsclopen *op;
1300 struct nfscllockowner *lp;
1302 u_int8_t own[NFSV4CL_LOCKNAMELEN];
1305 nfscl_filllockowner(id, own, flags);
1307 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
1308 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1309 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1310 !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1311 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1312 if (lp->nfsl_inprog == p &&
1313 !NFSBCMP(lp->nfsl_owner, own,
1314 NFSV4CL_LOCKNAMELEN)) {
1315 lp->nfsl_inprog = NULL;
1316 nfscl_lockunlock(&lp->nfsl_rwlock);
1322 nfscl_clrelease(clp);
1327 * Called to find out if any bytes within the byte range specified are
1328 * write locked by the calling process. Used to determine if flushing
1329 * is required before a LockU.
1330 * If in doubt, return 1, so the flush will occur.
1333 nfscl_checkwritelocked(vnode_t vp, struct flock *fl,
1334 struct ucred *cred, NFSPROC_T *p, void *id, int flags)
1336 struct nfsclowner *owp;
1337 struct nfscllockowner *lp;
1338 struct nfsclopen *op;
1339 struct nfsclclient *clp;
1340 struct nfscllock *lop;
1341 struct nfscldeleg *dp;
1344 u_int8_t own[NFSV4CL_LOCKNAMELEN];
1348 switch (fl->l_whence) {
1352 * Caller is responsible for adding any necessary offset
1353 * when SEEK_CUR is used.
1358 off = np->n_size + fl->l_start;
1363 if (fl->l_len != 0) {
1364 end = off + fl->l_len;
1371 error = nfscl_getcl(vnode_mount(vp), cred, p, 1, &clp);
1374 nfscl_filllockowner(id, own, flags);
1378 * First check the delegation locks.
1380 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
1382 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
1383 if (!NFSBCMP(lp->nfsl_owner, own,
1384 NFSV4CL_LOCKNAMELEN))
1388 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
1389 if (lop->nfslo_first >= end)
1391 if (lop->nfslo_end <= off)
1393 if (lop->nfslo_type == F_WRLCK) {
1394 nfscl_clrelease(clp);
1403 * Now, check state against the server.
1405 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
1406 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1407 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1408 !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1409 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1410 if (!NFSBCMP(lp->nfsl_owner, own,
1411 NFSV4CL_LOCKNAMELEN))
1415 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
1416 if (lop->nfslo_first >= end)
1418 if (lop->nfslo_end <= off)
1420 if (lop->nfslo_type == F_WRLCK) {
1421 nfscl_clrelease(clp);
1430 nfscl_clrelease(clp);
1436 * Release a byte range lock owner structure.
1439 nfscl_lockrelease(struct nfscllockowner *lp, int error, int candelete)
1441 struct nfsclclient *clp;
1446 clp = lp->nfsl_open->nfso_own->nfsow_clp;
1447 if (error != 0 && candelete &&
1448 (lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED) == 0)
1449 nfscl_freelockowner(lp, 0);
1451 nfscl_lockunlock(&lp->nfsl_rwlock);
1452 nfscl_clrelease(clp);
1457 * Free up an open structure and any associated byte range lock structures.
1460 nfscl_freeopen(struct nfsclopen *op, int local)
1463 LIST_REMOVE(op, nfso_list);
1464 nfscl_freealllocks(&op->nfso_lock, local);
1465 free(op, M_NFSCLOPEN);
1467 nfsstatsv1.cllocalopens--;
1469 nfsstatsv1.clopens--;
1473 * Free up all lock owners and associated locks.
1476 nfscl_freealllocks(struct nfscllockownerhead *lhp, int local)
1478 struct nfscllockowner *lp, *nlp;
1480 LIST_FOREACH_SAFE(lp, lhp, nfsl_list, nlp) {
1481 if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED))
1483 nfscl_freelockowner(lp, local);
1488 * Called for an Open when NFSERR_EXPIRED is received from the server.
1489 * If there are no byte range locks nor a Share Deny lost, try to do a
1490 * fresh Open. Otherwise, free the open.
1493 nfscl_expireopen(struct nfsclclient *clp, struct nfsclopen *op,
1494 struct nfsmount *nmp, struct ucred *cred, NFSPROC_T *p)
1496 struct nfscllockowner *lp;
1497 struct nfscldeleg *dp;
1498 int mustdelete = 0, error;
1501 * Look for any byte range lock(s).
1503 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1504 if (!LIST_EMPTY(&lp->nfsl_lock)) {
1511 * If no byte range lock(s) nor a Share deny, try to re-open.
1513 if (!mustdelete && (op->nfso_mode & NFSLCK_DENYBITS) == 0) {
1514 newnfs_copycred(&op->nfso_cred, cred);
1516 error = nfsrpc_reopen(nmp, op->nfso_fh,
1517 op->nfso_fhlen, op->nfso_mode, op, &dp, cred, p);
1521 free(dp, M_NFSCLDELEG);
1526 nfscl_deleg(nmp->nm_mountp, clp, op->nfso_fh,
1527 op->nfso_fhlen, cred, p, &dp);
1531 * If a byte range lock or Share deny or couldn't re-open, free it.
1534 nfscl_freeopen(op, 0);
1535 return (mustdelete);
1539 * Free up an open owner structure.
1542 nfscl_freeopenowner(struct nfsclowner *owp, int local)
1545 LIST_REMOVE(owp, nfsow_list);
1546 free(owp, M_NFSCLOWNER);
1548 nfsstatsv1.cllocalopenowners--;
1550 nfsstatsv1.clopenowners--;
1554 * Free up a byte range lock owner structure.
1557 nfscl_freelockowner(struct nfscllockowner *lp, int local)
1559 struct nfscllock *lop, *nlop;
1561 LIST_REMOVE(lp, nfsl_list);
1562 LIST_FOREACH_SAFE(lop, &lp->nfsl_lock, nfslo_list, nlop) {
1563 nfscl_freelock(lop, local);
1565 free(lp, M_NFSCLLOCKOWNER);
1567 nfsstatsv1.cllocallockowners--;
1569 nfsstatsv1.cllockowners--;
1573 * Free up a byte range lock structure.
1576 nfscl_freelock(struct nfscllock *lop, int local)
1579 LIST_REMOVE(lop, nfslo_list);
1580 free(lop, M_NFSCLLOCK);
1582 nfsstatsv1.cllocallocks--;
1584 nfsstatsv1.cllocks--;
1588 * Clean out the state related to a delegation.
1591 nfscl_cleandeleg(struct nfscldeleg *dp)
1593 struct nfsclowner *owp, *nowp;
1594 struct nfsclopen *op;
1596 LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) {
1597 op = LIST_FIRST(&owp->nfsow_open);
1599 if (LIST_NEXT(op, nfso_list) != NULL)
1600 panic("nfscleandel");
1601 nfscl_freeopen(op, 1);
1603 nfscl_freeopenowner(owp, 1);
1605 nfscl_freealllocks(&dp->nfsdl_lock, 1);
1609 * Free a delegation.
1612 nfscl_freedeleg(struct nfscldeleghead *hdp, struct nfscldeleg *dp)
1615 TAILQ_REMOVE(hdp, dp, nfsdl_list);
1616 LIST_REMOVE(dp, nfsdl_hash);
1617 free(dp, M_NFSCLDELEG);
1618 nfsstatsv1.cldelegates--;
1623 * Free up all state related to this client structure.
1626 nfscl_cleanclient(struct nfsclclient *clp)
1628 struct nfsclowner *owp, *nowp;
1629 struct nfsclopen *op, *nop;
1630 struct nfscllayout *lyp, *nlyp;
1631 struct nfscldevinfo *dip, *ndip;
1633 TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp)
1634 nfscl_freelayout(lyp);
1636 LIST_FOREACH_SAFE(dip, &clp->nfsc_devinfo, nfsdi_list, ndip)
1637 nfscl_freedevinfo(dip);
1639 /* Now, all the OpenOwners, etc. */
1640 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1641 LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
1642 nfscl_freeopen(op, 0);
1644 nfscl_freeopenowner(owp, 0);
1649 * Called when an NFSERR_EXPIRED is received from the server.
1652 nfscl_expireclient(struct nfsclclient *clp, struct nfsmount *nmp,
1653 struct ucred *cred, NFSPROC_T *p)
1655 struct nfsclowner *owp, *nowp, *towp;
1656 struct nfsclopen *op, *nop, *top;
1657 struct nfscldeleg *dp, *ndp;
1658 int ret, printed = 0;
1661 * First, merge locally issued Opens into the list for the server.
1663 dp = TAILQ_FIRST(&clp->nfsc_deleg);
1664 while (dp != NULL) {
1665 ndp = TAILQ_NEXT(dp, nfsdl_list);
1666 owp = LIST_FIRST(&dp->nfsdl_owner);
1667 while (owp != NULL) {
1668 nowp = LIST_NEXT(owp, nfsow_list);
1669 op = LIST_FIRST(&owp->nfsow_open);
1671 if (LIST_NEXT(op, nfso_list) != NULL)
1673 LIST_FOREACH(towp, &clp->nfsc_owner, nfsow_list) {
1674 if (!NFSBCMP(towp->nfsow_owner, owp->nfsow_owner,
1675 NFSV4CL_LOCKNAMELEN))
1679 /* Merge opens in */
1680 LIST_FOREACH(top, &towp->nfsow_open, nfso_list) {
1681 if (top->nfso_fhlen == op->nfso_fhlen &&
1682 !NFSBCMP(top->nfso_fh, op->nfso_fh,
1684 top->nfso_mode |= op->nfso_mode;
1685 top->nfso_opencnt += op->nfso_opencnt;
1690 /* Just add the open to the owner list */
1691 LIST_REMOVE(op, nfso_list);
1692 op->nfso_own = towp;
1693 LIST_INSERT_HEAD(&towp->nfsow_open, op, nfso_list);
1694 nfsstatsv1.cllocalopens--;
1695 nfsstatsv1.clopens++;
1698 /* Just add the openowner to the client list */
1699 LIST_REMOVE(owp, nfsow_list);
1700 owp->nfsow_clp = clp;
1701 LIST_INSERT_HEAD(&clp->nfsc_owner, owp, nfsow_list);
1702 nfsstatsv1.cllocalopenowners--;
1703 nfsstatsv1.clopenowners++;
1704 nfsstatsv1.cllocalopens--;
1705 nfsstatsv1.clopens++;
1710 if (!printed && !LIST_EMPTY(&dp->nfsdl_lock)) {
1712 printf("nfsv4 expired locks lost\n");
1714 nfscl_cleandeleg(dp);
1715 nfscl_freedeleg(&clp->nfsc_deleg, dp);
1718 if (!TAILQ_EMPTY(&clp->nfsc_deleg))
1722 * Now, try and reopen against the server.
1724 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1725 owp->nfsow_seqid = 0;
1726 LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
1727 ret = nfscl_expireopen(clp, op, nmp, cred, p);
1728 if (ret && !printed) {
1730 printf("nfsv4 expired locks lost\n");
1733 if (LIST_EMPTY(&owp->nfsow_open))
1734 nfscl_freeopenowner(owp, 0);
1739 * This function must be called after the process represented by "own" has
1740 * exited. Must be called with CLSTATE lock held.
1743 nfscl_cleanup_common(struct nfsclclient *clp, u_int8_t *own)
1745 struct nfsclowner *owp, *nowp;
1746 struct nfscllockowner *lp, *nlp;
1747 struct nfscldeleg *dp;
1749 /* First, get rid of local locks on delegations. */
1750 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
1751 LIST_FOREACH_SAFE(lp, &dp->nfsdl_lock, nfsl_list, nlp) {
1752 if (!NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
1753 if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED))
1755 nfscl_freelockowner(lp, 1);
1759 owp = LIST_FIRST(&clp->nfsc_owner);
1760 while (owp != NULL) {
1761 nowp = LIST_NEXT(owp, nfsow_list);
1762 if (!NFSBCMP(owp->nfsow_owner, own,
1763 NFSV4CL_LOCKNAMELEN)) {
1765 * If there are children that haven't closed the
1766 * file descriptors yet, the opens will still be
1767 * here. For that case, let the renew thread clear
1768 * out the OpenOwner later.
1770 if (LIST_EMPTY(&owp->nfsow_open))
1771 nfscl_freeopenowner(owp, 0);
1773 owp->nfsow_defunct = 1;
1780 * Find open/lock owners for processes that have exited.
1783 nfscl_cleanupkext(struct nfsclclient *clp, struct nfscllockownerfhhead *lhp)
1785 struct nfsclowner *owp, *nowp;
1786 struct nfsclopen *op;
1787 struct nfscllockowner *lp, *nlp;
1788 struct nfscldeleg *dp;
1792 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1793 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1794 LIST_FOREACH_SAFE(lp, &op->nfso_lock, nfsl_list, nlp) {
1795 if (LIST_EMPTY(&lp->nfsl_lock))
1796 nfscl_emptylockowner(lp, lhp);
1799 if (nfscl_procdoesntexist(owp->nfsow_owner))
1800 nfscl_cleanup_common(clp, owp->nfsow_owner);
1804 * For the single open_owner case, these lock owners need to be
1805 * checked to see if they still exist separately.
1806 * This is because nfscl_procdoesntexist() never returns true for
1807 * the single open_owner so that the above doesn't ever call
1808 * nfscl_cleanup_common().
1810 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
1811 LIST_FOREACH_SAFE(lp, &dp->nfsdl_lock, nfsl_list, nlp) {
1812 if (nfscl_procdoesntexist(lp->nfsl_owner))
1813 nfscl_cleanup_common(clp, lp->nfsl_owner);
1817 NFSPROCLISTUNLOCK();
1821 * Take the empty lock owner and move it to the local lhp list if the
1822 * associated process no longer exists.
1825 nfscl_emptylockowner(struct nfscllockowner *lp,
1826 struct nfscllockownerfhhead *lhp)
1828 struct nfscllockownerfh *lfhp, *mylfhp;
1829 struct nfscllockowner *nlp;
1832 /* If not a Posix lock owner, just return. */
1833 if ((lp->nfsl_lockflags & F_POSIX) == 0)
1839 * First, search to see if this lock owner is already in the list.
1840 * If it is, then the associated process no longer exists.
1842 SLIST_FOREACH(lfhp, lhp, nfslfh_list) {
1843 if (lfhp->nfslfh_len == lp->nfsl_open->nfso_fhlen &&
1844 !NFSBCMP(lfhp->nfslfh_fh, lp->nfsl_open->nfso_fh,
1847 LIST_FOREACH(nlp, &lfhp->nfslfh_lock, nfsl_list)
1848 if (!NFSBCMP(nlp->nfsl_owner, lp->nfsl_owner,
1849 NFSV4CL_LOCKNAMELEN))
1852 /* If not found, check if process still exists. */
1853 if (fnd_it == 0 && nfscl_procdoesntexist(lp->nfsl_owner) == 0)
1856 /* Move the lock owner over to the local list. */
1857 if (mylfhp == NULL) {
1858 mylfhp = malloc(sizeof(struct nfscllockownerfh), M_TEMP,
1862 mylfhp->nfslfh_len = lp->nfsl_open->nfso_fhlen;
1863 NFSBCOPY(lp->nfsl_open->nfso_fh, mylfhp->nfslfh_fh,
1864 mylfhp->nfslfh_len);
1865 LIST_INIT(&mylfhp->nfslfh_lock);
1866 SLIST_INSERT_HEAD(lhp, mylfhp, nfslfh_list);
1868 LIST_REMOVE(lp, nfsl_list);
1869 LIST_INSERT_HEAD(&mylfhp->nfslfh_lock, lp, nfsl_list);
1872 static int fake_global; /* Used to force visibility of MNTK_UNMOUNTF */
1874 * Called from nfs umount to free up the clientid.
1877 nfscl_umount(struct nfsmount *nmp, NFSPROC_T *p)
1879 struct nfsclclient *clp;
1884 * For the case that matters, this is the thread that set
1885 * MNTK_UNMOUNTF, so it will see it set. The code that follows is
1886 * done to ensure that any thread executing nfscl_getcl() after
1887 * this time, will see MNTK_UNMOUNTF set. nfscl_getcl() uses the
1888 * mutex for NFSLOCKCLSTATE(), so it is "m" for the following
1889 * explanation, courtesy of Alan Cox.
1890 * What follows is a snippet from Alan Cox's email at:
1891 * https://docs.FreeBSD.org/cgi/mid.cgi?BANLkTikR3d65zPHo9==08ZfJ2vmqZucEvw
1893 * 1. Set MNTK_UNMOUNTF
1894 * 2. Acquire a standard FreeBSD mutex "m".
1895 * 3. Update some data structures.
1896 * 4. Release mutex "m".
1898 * Then, other threads that acquire "m" after step 4 has occurred will
1899 * see MNTK_UNMOUNTF as set. But, other threads that beat thread X to
1900 * step 2 may or may not see MNTK_UNMOUNTF as set.
1903 if ((nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) {
1911 if ((clp->nfsc_flags & NFSCLFLAGS_INITED) == 0)
1912 panic("nfscl umount");
1915 * First, handshake with the nfscl renew thread, to terminate
1918 clp->nfsc_flags |= NFSCLFLAGS_UMOUNT;
1919 while (clp->nfsc_flags & NFSCLFLAGS_HASTHREAD)
1920 (void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT,
1924 * Now, get the exclusive lock on the client state, so
1925 * that no uses of the state are still in progress.
1928 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
1929 NFSCLSTATEMUTEXPTR, NULL);
1930 } while (!igotlock);
1934 * Free up all the state. It will expire on the server, but
1935 * maybe we should do a SetClientId/SetClientIdConfirm so
1936 * the server throws it away?
1938 LIST_REMOVE(clp, nfsc_list);
1939 nfscl_delegreturnall(clp, p);
1940 cred = newnfs_getcred();
1941 if (NFSHASNFSV4N(nmp)) {
1942 (void)nfsrpc_destroysession(nmp, clp, cred, p);
1943 (void)nfsrpc_destroyclient(nmp, clp, cred, p);
1945 (void)nfsrpc_setclient(nmp, clp, 0, cred, p);
1946 nfscl_cleanclient(clp);
1949 free(clp, M_NFSCLCLIENT);
1955 * This function is called when a server replies with NFSERR_STALECLIENTID
1956 * NFSERR_STALESTATEID or NFSERR_BADSESSION. It traverses the clientid lists,
1957 * doing Opens and Locks with reclaim. If these fail, it deletes the
1958 * corresponding state.
1961 nfscl_recover(struct nfsclclient *clp, struct ucred *cred, NFSPROC_T *p)
1963 struct nfsclowner *owp, *nowp;
1964 struct nfsclopen *op, *nop;
1965 struct nfscllockowner *lp, *nlp;
1966 struct nfscllock *lop, *nlop;
1967 struct nfscldeleg *dp, *ndp, *tdp;
1968 struct nfsmount *nmp;
1969 struct ucred *tcred;
1970 struct nfsclopenhead extra_open;
1971 struct nfscldeleghead extra_deleg;
1974 u_int32_t delegtype = NFSV4OPEN_DELEGATEWRITE, mode;
1975 int i, igotlock = 0, error, trycnt, firstlock;
1976 struct nfscllayout *lyp, *nlyp;
1979 * First, lock the client structure, so everyone else will
1980 * block when trying to use state.
1983 clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG;
1985 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
1986 NFSCLSTATEMUTEXPTR, NULL);
1987 } while (!igotlock);
1990 nmp = clp->nfsc_nmp;
1992 panic("nfscl recover");
1995 * For now, just get rid of all layouts. There may be a need
1996 * to do LayoutCommit Ops with reclaim == true later.
1998 TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp)
1999 nfscl_freelayout(lyp);
2000 TAILQ_INIT(&clp->nfsc_layout);
2001 for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++)
2002 LIST_INIT(&clp->nfsc_layouthash[i]);
2006 error = nfsrpc_setclient(nmp, clp, 1, cred, p);
2007 } while ((error == NFSERR_STALECLIENTID ||
2008 error == NFSERR_BADSESSION ||
2009 error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
2012 clp->nfsc_flags &= ~(NFSCLFLAGS_RECOVER |
2013 NFSCLFLAGS_RECVRINPROG);
2014 wakeup(&clp->nfsc_flags);
2015 nfsv4_unlock(&clp->nfsc_lock, 0);
2019 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
2020 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2023 * Mark requests already queued on the server, so that they don't
2024 * initiate another recovery cycle. Any requests already in the
2025 * queue that handle state information will have the old stale
2026 * clientid/stateid and will get a NFSERR_STALESTATEID,
2027 * NFSERR_STALECLIENTID or NFSERR_BADSESSION reply from the server.
2028 * This will be translated to NFSERR_STALEDONTRECOVER when
2029 * R_DONTRECOVER is set.
2032 TAILQ_FOREACH(rep, &nfsd_reqq, r_chain) {
2033 if (rep->r_nmp == nmp)
2034 rep->r_flags |= R_DONTRECOVER;
2039 * Now, mark all delegations "need reclaim".
2041 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list)
2042 dp->nfsdl_flags |= NFSCLDL_NEEDRECLAIM;
2044 TAILQ_INIT(&extra_deleg);
2045 LIST_INIT(&extra_open);
2047 * Now traverse the state lists, doing Open and Lock Reclaims.
2049 tcred = newnfs_getcred();
2050 owp = LIST_FIRST(&clp->nfsc_owner);
2051 while (owp != NULL) {
2052 nowp = LIST_NEXT(owp, nfsow_list);
2053 owp->nfsow_seqid = 0;
2054 op = LIST_FIRST(&owp->nfsow_open);
2055 while (op != NULL) {
2056 nop = LIST_NEXT(op, nfso_list);
2057 if (error != NFSERR_NOGRACE && error != NFSERR_BADSESSION) {
2058 /* Search for a delegation to reclaim with the open */
2059 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2060 if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
2062 if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
2063 mode = NFSV4OPEN_ACCESSWRITE;
2064 delegtype = NFSV4OPEN_DELEGATEWRITE;
2066 mode = NFSV4OPEN_ACCESSREAD;
2067 delegtype = NFSV4OPEN_DELEGATEREAD;
2069 if ((op->nfso_mode & mode) == mode &&
2070 op->nfso_fhlen == dp->nfsdl_fhlen &&
2071 !NFSBCMP(op->nfso_fh, dp->nfsdl_fh, op->nfso_fhlen))
2076 delegtype = NFSV4OPEN_DELEGATENONE;
2077 newnfs_copycred(&op->nfso_cred, tcred);
2078 error = nfscl_tryopen(nmp, NULL, op->nfso_fh,
2079 op->nfso_fhlen, op->nfso_fh, op->nfso_fhlen,
2080 op->nfso_mode, op, NULL, 0, &ndp, 1, delegtype,
2083 /* Handle any replied delegation */
2084 if (ndp != NULL && ((ndp->nfsdl_flags & NFSCLDL_WRITE)
2085 || NFSMNT_RDONLY(nmp->nm_mountp))) {
2086 if ((ndp->nfsdl_flags & NFSCLDL_WRITE))
2087 mode = NFSV4OPEN_ACCESSWRITE;
2089 mode = NFSV4OPEN_ACCESSREAD;
2090 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2091 if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
2093 if ((op->nfso_mode & mode) == mode &&
2094 op->nfso_fhlen == dp->nfsdl_fhlen &&
2095 !NFSBCMP(op->nfso_fh, dp->nfsdl_fh,
2097 dp->nfsdl_stateid = ndp->nfsdl_stateid;
2098 dp->nfsdl_sizelimit = ndp->nfsdl_sizelimit;
2099 dp->nfsdl_ace = ndp->nfsdl_ace;
2100 dp->nfsdl_change = ndp->nfsdl_change;
2101 dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
2102 if ((ndp->nfsdl_flags & NFSCLDL_RECALL))
2103 dp->nfsdl_flags |= NFSCLDL_RECALL;
2104 free(ndp, M_NFSCLDELEG);
2111 TAILQ_INSERT_HEAD(&extra_deleg, ndp, nfsdl_list);
2113 /* and reclaim all byte range locks */
2114 lp = LIST_FIRST(&op->nfso_lock);
2115 while (lp != NULL) {
2116 nlp = LIST_NEXT(lp, nfsl_list);
2119 lop = LIST_FIRST(&lp->nfsl_lock);
2120 while (lop != NULL) {
2121 nlop = LIST_NEXT(lop, nfslo_list);
2122 if (lop->nfslo_end == NFS64BITSSET)
2125 len = lop->nfslo_end - lop->nfslo_first;
2126 error = nfscl_trylock(nmp, NULL,
2127 op->nfso_fh, op->nfso_fhlen, lp,
2128 firstlock, 1, lop->nfslo_first, len,
2129 lop->nfslo_type, tcred, p);
2131 nfscl_freelock(lop, 0);
2136 /* If no locks, but a lockowner, just delete it. */
2137 if (LIST_EMPTY(&lp->nfsl_lock))
2138 nfscl_freelockowner(lp, 0);
2143 if (error != 0 && error != NFSERR_BADSESSION)
2144 nfscl_freeopen(op, 0);
2151 * Now, try and get any delegations not yet reclaimed by cobbling
2152 * to-gether an appropriate open.
2155 dp = TAILQ_FIRST(&clp->nfsc_deleg);
2156 while (dp != NULL) {
2157 ndp = TAILQ_NEXT(dp, nfsdl_list);
2158 if ((dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM)) {
2161 sizeof (struct nfsclowner), M_NFSCLOWNER, M_WAITOK);
2163 * Name must be as long an largest possible
2164 * NFSV4CL_LOCKNAMELEN. 12 for now.
2166 NFSBCOPY("RECLAIMDELEG", nowp->nfsow_owner,
2167 NFSV4CL_LOCKNAMELEN);
2168 LIST_INIT(&nowp->nfsow_open);
2169 nowp->nfsow_clp = clp;
2170 nowp->nfsow_seqid = 0;
2171 nowp->nfsow_defunct = 0;
2172 nfscl_lockinit(&nowp->nfsow_rwlock);
2175 if (error != NFSERR_NOGRACE && error != NFSERR_BADSESSION) {
2176 nop = malloc(sizeof (struct nfsclopen) +
2177 dp->nfsdl_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
2178 nop->nfso_own = nowp;
2179 if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
2180 nop->nfso_mode = NFSV4OPEN_ACCESSWRITE;
2181 delegtype = NFSV4OPEN_DELEGATEWRITE;
2183 nop->nfso_mode = NFSV4OPEN_ACCESSREAD;
2184 delegtype = NFSV4OPEN_DELEGATEREAD;
2186 nop->nfso_opencnt = 0;
2187 nop->nfso_posixlock = 1;
2188 nop->nfso_fhlen = dp->nfsdl_fhlen;
2189 NFSBCOPY(dp->nfsdl_fh, nop->nfso_fh, dp->nfsdl_fhlen);
2190 LIST_INIT(&nop->nfso_lock);
2191 nop->nfso_stateid.seqid = 0;
2192 nop->nfso_stateid.other[0] = 0;
2193 nop->nfso_stateid.other[1] = 0;
2194 nop->nfso_stateid.other[2] = 0;
2195 newnfs_copycred(&dp->nfsdl_cred, tcred);
2196 newnfs_copyincred(tcred, &nop->nfso_cred);
2198 error = nfscl_tryopen(nmp, NULL, nop->nfso_fh,
2199 nop->nfso_fhlen, nop->nfso_fh, nop->nfso_fhlen,
2200 nop->nfso_mode, nop, NULL, 0, &tdp, 1,
2201 delegtype, tcred, p);
2203 if ((tdp->nfsdl_flags & NFSCLDL_WRITE))
2204 mode = NFSV4OPEN_ACCESSWRITE;
2206 mode = NFSV4OPEN_ACCESSREAD;
2207 if ((nop->nfso_mode & mode) == mode &&
2208 nop->nfso_fhlen == tdp->nfsdl_fhlen &&
2209 !NFSBCMP(nop->nfso_fh, tdp->nfsdl_fh,
2211 dp->nfsdl_stateid = tdp->nfsdl_stateid;
2212 dp->nfsdl_sizelimit = tdp->nfsdl_sizelimit;
2213 dp->nfsdl_ace = tdp->nfsdl_ace;
2214 dp->nfsdl_change = tdp->nfsdl_change;
2215 dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
2216 if ((tdp->nfsdl_flags & NFSCLDL_RECALL))
2217 dp->nfsdl_flags |= NFSCLDL_RECALL;
2218 free(tdp, M_NFSCLDELEG);
2220 TAILQ_INSERT_HEAD(&extra_deleg, tdp, nfsdl_list);
2226 free(nop, M_NFSCLOPEN);
2228 * Couldn't reclaim it, so throw the state
2231 nfscl_cleandeleg(dp);
2232 nfscl_freedeleg(&clp->nfsc_deleg, dp);
2234 LIST_INSERT_HEAD(&extra_open, nop, nfso_list);
2241 * Now, get rid of extra Opens and Delegations.
2243 LIST_FOREACH_SAFE(op, &extra_open, nfso_list, nop) {
2245 newnfs_copycred(&op->nfso_cred, tcred);
2246 error = nfscl_tryclose(op, tcred, nmp, p);
2247 if (error == NFSERR_GRACE)
2248 (void) nfs_catnap(PZERO, error, "nfsexcls");
2249 } while (error == NFSERR_GRACE);
2250 LIST_REMOVE(op, nfso_list);
2251 free(op, M_NFSCLOPEN);
2254 free(nowp, M_NFSCLOWNER);
2256 TAILQ_FOREACH_SAFE(dp, &extra_deleg, nfsdl_list, ndp) {
2258 newnfs_copycred(&dp->nfsdl_cred, tcred);
2259 error = nfscl_trydelegreturn(dp, tcred, nmp, p);
2260 if (error == NFSERR_GRACE)
2261 (void) nfs_catnap(PZERO, error, "nfsexdlg");
2262 } while (error == NFSERR_GRACE);
2263 TAILQ_REMOVE(&extra_deleg, dp, nfsdl_list);
2264 free(dp, M_NFSCLDELEG);
2267 /* For NFSv4.1 or later, do a RECLAIM_COMPLETE. */
2268 if (NFSHASNFSV4N(nmp))
2269 (void)nfsrpc_reclaimcomplete(nmp, cred, p);
2272 clp->nfsc_flags &= ~NFSCLFLAGS_RECVRINPROG;
2273 wakeup(&clp->nfsc_flags);
2274 nfsv4_unlock(&clp->nfsc_lock, 0);
2280 * This function is called when a server replies with NFSERR_EXPIRED.
2281 * It deletes all state for the client and does a fresh SetClientId/confirm.
2282 * XXX Someday it should post a signal to the process(es) that hold the
2283 * state, so they know that lock state has been lost.
2286 nfscl_hasexpired(struct nfsclclient *clp, u_int32_t clidrev, NFSPROC_T *p)
2288 struct nfsmount *nmp;
2290 int igotlock = 0, error, trycnt;
2293 * If the clientid has gone away or a new SetClientid has already
2294 * been done, just return ok.
2296 if (clp == NULL || clidrev != clp->nfsc_clientidrev)
2300 * First, lock the client structure, so everyone else will
2301 * block when trying to use state. Also, use NFSCLFLAGS_EXPIREIT so
2302 * that only one thread does the work.
2305 clp->nfsc_flags |= NFSCLFLAGS_EXPIREIT;
2307 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
2308 NFSCLSTATEMUTEXPTR, NULL);
2309 } while (!igotlock && (clp->nfsc_flags & NFSCLFLAGS_EXPIREIT));
2310 if ((clp->nfsc_flags & NFSCLFLAGS_EXPIREIT) == 0) {
2312 nfsv4_unlock(&clp->nfsc_lock, 0);
2316 clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG;
2319 nmp = clp->nfsc_nmp;
2321 panic("nfscl expired");
2322 cred = newnfs_getcred();
2325 error = nfsrpc_setclient(nmp, clp, 0, cred, p);
2326 } while ((error == NFSERR_STALECLIENTID ||
2327 error == NFSERR_BADSESSION ||
2328 error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
2331 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2334 * Expire the state for the client.
2336 nfscl_expireclient(clp, nmp, cred, p);
2338 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
2339 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2341 clp->nfsc_flags &= ~(NFSCLFLAGS_EXPIREIT | NFSCLFLAGS_RECVRINPROG);
2342 wakeup(&clp->nfsc_flags);
2343 nfsv4_unlock(&clp->nfsc_lock, 0);
2350 * This function inserts a lock in the list after insert_lop.
2353 nfscl_insertlock(struct nfscllockowner *lp, struct nfscllock *new_lop,
2354 struct nfscllock *insert_lop, int local)
2357 if ((struct nfscllockowner *)insert_lop == lp)
2358 LIST_INSERT_HEAD(&lp->nfsl_lock, new_lop, nfslo_list);
2360 LIST_INSERT_AFTER(insert_lop, new_lop, nfslo_list);
2362 nfsstatsv1.cllocallocks++;
2364 nfsstatsv1.cllocks++;
2368 * This function updates the locking for a lock owner and given file. It
2369 * maintains a list of lock ranges ordered on increasing file offset that
2370 * are NFSCLLOCK_READ or NFSCLLOCK_WRITE and non-overlapping (aka POSIX style).
2371 * It always adds new_lop to the list and sometimes uses the one pointed
2373 * Returns 1 if the locks were modified, 0 otherwise.
2376 nfscl_updatelock(struct nfscllockowner *lp, struct nfscllock **new_lopp,
2377 struct nfscllock **other_lopp, int local)
2379 struct nfscllock *new_lop = *new_lopp;
2380 struct nfscllock *lop, *tlop, *ilop;
2381 struct nfscllock *other_lop;
2382 int unlock = 0, modified = 0;
2386 * Work down the list until the lock is merged.
2388 if (new_lop->nfslo_type == F_UNLCK)
2390 ilop = (struct nfscllock *)lp;
2391 lop = LIST_FIRST(&lp->nfsl_lock);
2392 while (lop != NULL) {
2394 * Only check locks for this file that aren't before the start of
2397 if (lop->nfslo_end >= new_lop->nfslo_first) {
2398 if (new_lop->nfslo_end < lop->nfslo_first) {
2400 * If the new lock ends before the start of the
2401 * current lock's range, no merge, just insert
2406 if (new_lop->nfslo_type == lop->nfslo_type ||
2407 (new_lop->nfslo_first <= lop->nfslo_first &&
2408 new_lop->nfslo_end >= lop->nfslo_end)) {
2410 * This lock can be absorbed by the new lock/unlock.
2411 * This happens when it covers the entire range
2412 * of the old lock or is contiguous
2413 * with the old lock and is of the same type or an
2416 if (new_lop->nfslo_type != lop->nfslo_type ||
2417 new_lop->nfslo_first != lop->nfslo_first ||
2418 new_lop->nfslo_end != lop->nfslo_end)
2420 if (lop->nfslo_first < new_lop->nfslo_first)
2421 new_lop->nfslo_first = lop->nfslo_first;
2422 if (lop->nfslo_end > new_lop->nfslo_end)
2423 new_lop->nfslo_end = lop->nfslo_end;
2425 lop = LIST_NEXT(lop, nfslo_list);
2426 nfscl_freelock(tlop, local);
2431 * All these cases are for contiguous locks that are not the
2432 * same type, so they can't be merged.
2434 if (new_lop->nfslo_first <= lop->nfslo_first) {
2436 * This case is where the new lock overlaps with the
2437 * first part of the old lock. Move the start of the
2438 * old lock to just past the end of the new lock. The
2439 * new lock will be inserted in front of the old, since
2440 * ilop hasn't been updated. (We are done now.)
2442 if (lop->nfslo_first != new_lop->nfslo_end) {
2443 lop->nfslo_first = new_lop->nfslo_end;
2448 if (new_lop->nfslo_end >= lop->nfslo_end) {
2450 * This case is where the new lock overlaps with the
2451 * end of the old lock's range. Move the old lock's
2452 * end to just before the new lock's first and insert
2453 * the new lock after the old lock.
2454 * Might not be done yet, since the new lock could
2455 * overlap further locks with higher ranges.
2457 if (lop->nfslo_end != new_lop->nfslo_first) {
2458 lop->nfslo_end = new_lop->nfslo_first;
2462 lop = LIST_NEXT(lop, nfslo_list);
2466 * The final case is where the new lock's range is in the
2467 * middle of the current lock's and splits the current lock
2468 * up. Use *other_lopp to handle the second part of the
2469 * split old lock range. (We are done now.)
2470 * For unlock, we use new_lop as other_lop and tmp, since
2471 * other_lop and new_lop are the same for this case.
2472 * We noted the unlock case above, so we don't need
2473 * new_lop->nfslo_type any longer.
2475 tmp = new_lop->nfslo_first;
2477 other_lop = new_lop;
2480 other_lop = *other_lopp;
2483 other_lop->nfslo_first = new_lop->nfslo_end;
2484 other_lop->nfslo_end = lop->nfslo_end;
2485 other_lop->nfslo_type = lop->nfslo_type;
2486 lop->nfslo_end = tmp;
2487 nfscl_insertlock(lp, other_lop, lop, local);
2493 lop = LIST_NEXT(lop, nfslo_list);
2499 * Insert the new lock in the list at the appropriate place.
2502 nfscl_insertlock(lp, new_lop, ilop, local);
2510 * This function must be run as a kernel thread.
2511 * It does Renew Ops and recovery, when required.
2514 nfscl_renewthread(struct nfsclclient *clp, NFSPROC_T *p)
2516 struct nfsclowner *owp, *nowp;
2517 struct nfsclopen *op;
2518 struct nfscllockowner *lp, *nlp;
2519 struct nfscldeleghead dh;
2520 struct nfscldeleg *dp, *ndp;
2523 int error, cbpathdown, islept, igotlock, ret, clearok;
2524 uint32_t recover_done_time = 0;
2526 static time_t prevsec = 0;
2527 struct nfscllockownerfh *lfhp, *nlfhp;
2528 struct nfscllockownerfhhead lfh;
2529 struct nfscllayout *lyp, *nlyp;
2530 struct nfscldevinfo *dip, *ndip;
2531 struct nfscllayouthead rlh;
2532 struct nfsclrecalllayout *recallp;
2533 struct nfsclds *dsp;
2535 cred = newnfs_getcred();
2537 clp->nfsc_flags |= NFSCLFLAGS_HASTHREAD;
2540 newnfs_setroot(cred);
2542 if (clp->nfsc_flags & NFSCLFLAGS_RECOVER) {
2544 * Only allow one recover within 1/2 of the lease
2545 * duration (nfsc_renew).
2547 if (recover_done_time < NFSD_MONOSEC) {
2548 recover_done_time = NFSD_MONOSEC +
2550 NFSCL_DEBUG(1, "Doing recovery..\n");
2551 nfscl_recover(clp, cred, p);
2553 NFSCL_DEBUG(1, "Clear Recovery dt=%u ms=%jd\n",
2554 recover_done_time, (intmax_t)NFSD_MONOSEC);
2556 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2560 if (clp->nfsc_expire <= NFSD_MONOSEC &&
2561 (clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID)) {
2562 clp->nfsc_expire = NFSD_MONOSEC + clp->nfsc_renew;
2563 clidrev = clp->nfsc_clientidrev;
2564 error = nfsrpc_renew(clp, NULL, cred, p);
2565 if (error == NFSERR_CBPATHDOWN)
2567 else if (error == NFSERR_STALECLIENTID ||
2568 error == NFSERR_BADSESSION) {
2570 clp->nfsc_flags |= NFSCLFLAGS_RECOVER;
2572 } else if (error == NFSERR_EXPIRED)
2573 (void) nfscl_hasexpired(clp, clidrev, p);
2577 if (NFSHASNFSV4N(clp->nfsc_nmp)) {
2578 /* Do renews for any DS sessions. */
2579 NFSLOCKMNT(clp->nfsc_nmp);
2580 /* Skip first entry, since the MDS is handled above. */
2581 dsp = TAILQ_FIRST(&clp->nfsc_nmp->nm_sess);
2583 dsp = TAILQ_NEXT(dsp, nfsclds_list);
2584 while (dsp != NULL) {
2585 if (dsp->nfsclds_expire <= NFSD_MONOSEC &&
2586 dsp->nfsclds_sess.nfsess_defunct == 0) {
2587 dsp->nfsclds_expire = NFSD_MONOSEC +
2589 NFSUNLOCKMNT(clp->nfsc_nmp);
2590 (void)nfsrpc_renew(clp, dsp, cred, p);
2593 dsp = TAILQ_NEXT(dsp, nfsclds_list);
2595 NFSUNLOCKMNT(clp->nfsc_nmp);
2601 /* It's a Total Recall! */
2602 nfscl_totalrecall(clp);
2605 * Now, handle defunct owners.
2607 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
2608 if (LIST_EMPTY(&owp->nfsow_open)) {
2609 if (owp->nfsow_defunct != 0)
2610 nfscl_freeopenowner(owp, 0);
2615 * Do the recall on any delegations. To avoid trouble, always
2616 * come back up here after having slept.
2620 dp = TAILQ_FIRST(&clp->nfsc_deleg);
2621 while (dp != NULL) {
2622 ndp = TAILQ_NEXT(dp, nfsdl_list);
2623 if ((dp->nfsdl_flags & NFSCLDL_RECALL)) {
2625 * Wait for outstanding I/O ops to be done.
2627 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
2629 nfsv4_unlock(&clp->nfsc_lock, 0);
2632 dp->nfsdl_rwlock.nfslock_lock |=
2634 (void) nfsmsleep(&dp->nfsdl_rwlock,
2635 NFSCLSTATEMUTEXPTR, PZERO, "nfscld",
2640 igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
2641 &islept, NFSCLSTATEMUTEXPTR, NULL);
2646 newnfs_copycred(&dp->nfsdl_cred, cred);
2647 ret = nfscl_recalldeleg(clp, clp->nfsc_nmp, dp,
2650 nfscl_cleandeleg(dp);
2651 TAILQ_REMOVE(&clp->nfsc_deleg, dp,
2653 LIST_REMOVE(dp, nfsdl_hash);
2654 TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
2656 nfsstatsv1.cldelegates--;
2664 * Clear out old delegations, if we are above the high water
2665 * mark. Only clear out ones with no state related to them.
2666 * The tailq list is in LRU order.
2668 dp = TAILQ_LAST(&clp->nfsc_deleg, nfscldeleghead);
2669 while (nfscl_delegcnt > nfscl_deleghighwater && dp != NULL) {
2670 ndp = TAILQ_PREV(dp, nfscldeleghead, nfsdl_list);
2671 if (dp->nfsdl_rwlock.nfslock_usecnt == 0 &&
2672 dp->nfsdl_rwlock.nfslock_lock == 0 &&
2673 dp->nfsdl_timestamp < NFSD_MONOSEC &&
2674 (dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_ZAPPED |
2675 NFSCLDL_NEEDRECLAIM | NFSCLDL_DELEGRET)) == 0) {
2677 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
2678 op = LIST_FIRST(&owp->nfsow_open);
2685 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
2686 if (!LIST_EMPTY(&lp->nfsl_lock)) {
2693 TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
2694 LIST_REMOVE(dp, nfsdl_hash);
2695 TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
2697 nfsstatsv1.cldelegates--;
2703 nfsv4_unlock(&clp->nfsc_lock, 0);
2706 * Do the recall on any layouts. To avoid trouble, always
2707 * come back up here after having slept.
2711 TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp) {
2712 if ((lyp->nfsly_flags & NFSLY_RECALL) != 0) {
2714 * Wait for outstanding I/O ops to be done.
2716 if (lyp->nfsly_lock.nfslock_usecnt > 0 ||
2717 (lyp->nfsly_lock.nfslock_lock &
2718 NFSV4LOCK_LOCK) != 0) {
2719 lyp->nfsly_lock.nfslock_lock |=
2721 nfsmsleep(&lyp->nfsly_lock.nfslock_lock,
2722 NFSCLSTATEMUTEXPTR, PZERO, "nfslyp",
2726 /* Move the layout to the recall list. */
2727 TAILQ_REMOVE(&clp->nfsc_layout, lyp,
2729 LIST_REMOVE(lyp, nfsly_hash);
2730 TAILQ_INSERT_HEAD(&rlh, lyp, nfsly_list);
2732 /* Handle any layout commits. */
2733 if (!NFSHASNOLAYOUTCOMMIT(clp->nfsc_nmp) &&
2734 (lyp->nfsly_flags & NFSLY_WRITTEN) != 0) {
2735 lyp->nfsly_flags &= ~NFSLY_WRITTEN;
2737 NFSCL_DEBUG(3, "do layoutcommit\n");
2738 nfscl_dolayoutcommit(clp->nfsc_nmp, lyp,
2746 /* Now, look for stale layouts. */
2747 lyp = TAILQ_LAST(&clp->nfsc_layout, nfscllayouthead);
2748 while (lyp != NULL) {
2749 nlyp = TAILQ_PREV(lyp, nfscllayouthead, nfsly_list);
2750 if (lyp->nfsly_timestamp < NFSD_MONOSEC &&
2751 (lyp->nfsly_flags & NFSLY_RECALL) == 0 &&
2752 lyp->nfsly_lock.nfslock_usecnt == 0 &&
2753 lyp->nfsly_lock.nfslock_lock == 0) {
2754 NFSCL_DEBUG(4, "ret stale lay=%d\n",
2756 recallp = malloc(sizeof(*recallp),
2757 M_NFSLAYRECALL, M_NOWAIT);
2758 if (recallp == NULL)
2760 (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE,
2761 lyp, NFSLAYOUTIOMODE_ANY, 0, UINT64_MAX,
2762 lyp->nfsly_stateid.seqid, 0, 0, NULL,
2769 * Free up any unreferenced device info structures.
2771 LIST_FOREACH_SAFE(dip, &clp->nfsc_devinfo, nfsdi_list, ndip) {
2772 if (dip->nfsdi_layoutrefs == 0 &&
2773 dip->nfsdi_refcnt == 0) {
2774 NFSCL_DEBUG(4, "freeing devinfo\n");
2775 LIST_REMOVE(dip, nfsdi_list);
2776 nfscl_freedevinfo(dip);
2781 /* Do layout return(s), as required. */
2782 TAILQ_FOREACH_SAFE(lyp, &rlh, nfsly_list, nlyp) {
2783 TAILQ_REMOVE(&rlh, lyp, nfsly_list);
2784 NFSCL_DEBUG(4, "ret layout\n");
2785 nfscl_layoutreturn(clp->nfsc_nmp, lyp, cred, p);
2786 nfscl_freelayout(lyp);
2790 * Delegreturn any delegations cleaned out or recalled.
2792 TAILQ_FOREACH_SAFE(dp, &dh, nfsdl_list, ndp) {
2793 newnfs_copycred(&dp->nfsdl_cred, cred);
2794 (void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
2795 TAILQ_REMOVE(&dh, dp, nfsdl_list);
2796 free(dp, M_NFSCLDELEG);
2801 * Call nfscl_cleanupkext() once per second to check for
2802 * open/lock owners where the process has exited.
2804 mytime = NFSD_MONOSEC;
2805 if (prevsec != mytime) {
2807 nfscl_cleanupkext(clp, &lfh);
2811 * Do a ReleaseLockOwner for all lock owners where the
2812 * associated process no longer exists, as found by
2813 * nfscl_cleanupkext().
2815 newnfs_setroot(cred);
2816 SLIST_FOREACH_SAFE(lfhp, &lfh, nfslfh_list, nlfhp) {
2817 LIST_FOREACH_SAFE(lp, &lfhp->nfslfh_lock, nfsl_list,
2819 (void)nfsrpc_rellockown(clp->nfsc_nmp, lp,
2820 lfhp->nfslfh_fh, lfhp->nfslfh_len, cred,
2822 nfscl_freelockowner(lp, 0);
2829 if ((clp->nfsc_flags & NFSCLFLAGS_RECOVER) == 0)
2830 (void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT, "nfscl",
2832 if (clp->nfsc_flags & NFSCLFLAGS_UMOUNT) {
2833 clp->nfsc_flags &= ~NFSCLFLAGS_HASTHREAD;
2836 wakeup((caddr_t)clp);
2844 * Initiate state recovery. Called when NFSERR_STALECLIENTID,
2845 * NFSERR_STALESTATEID or NFSERR_BADSESSION is received.
2848 nfscl_initiate_recovery(struct nfsclclient *clp)
2854 clp->nfsc_flags |= NFSCLFLAGS_RECOVER;
2856 wakeup((caddr_t)clp);
2860 * Dump out the state stuff for debugging.
2863 nfscl_dumpstate(struct nfsmount *nmp, int openowner, int opens,
2864 int lockowner, int locks)
2866 struct nfsclclient *clp;
2867 struct nfsclowner *owp;
2868 struct nfsclopen *op;
2869 struct nfscllockowner *lp;
2870 struct nfscllock *lop;
2871 struct nfscldeleg *dp;
2875 printf("nfscl dumpstate NULL clp\n");
2879 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2880 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
2881 if (openowner && !LIST_EMPTY(&owp->nfsow_open))
2882 printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
2883 owp->nfsow_owner[0], owp->nfsow_owner[1],
2884 owp->nfsow_owner[2], owp->nfsow_owner[3],
2886 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
2888 printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
2889 op->nfso_stateid.other[0], op->nfso_stateid.other[1],
2890 op->nfso_stateid.other[2], op->nfso_opencnt,
2892 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
2894 printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
2895 lp->nfsl_owner[0], lp->nfsl_owner[1],
2896 lp->nfsl_owner[2], lp->nfsl_owner[3],
2898 lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
2899 lp->nfsl_stateid.other[2]);
2900 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
2903 printf("lck typ=%d fst=%ju end=%ju\n",
2904 lop->nfslo_type, (intmax_t)lop->nfslo_first,
2905 (intmax_t)lop->nfslo_end);
2907 printf("lck typ=%d fst=%qd end=%qd\n",
2908 lop->nfslo_type, lop->nfslo_first,
2916 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
2917 if (openowner && !LIST_EMPTY(&owp->nfsow_open))
2918 printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
2919 owp->nfsow_owner[0], owp->nfsow_owner[1],
2920 owp->nfsow_owner[2], owp->nfsow_owner[3],
2922 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
2924 printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
2925 op->nfso_stateid.other[0], op->nfso_stateid.other[1],
2926 op->nfso_stateid.other[2], op->nfso_opencnt,
2928 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
2930 printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
2931 lp->nfsl_owner[0], lp->nfsl_owner[1],
2932 lp->nfsl_owner[2], lp->nfsl_owner[3],
2934 lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
2935 lp->nfsl_stateid.other[2]);
2936 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
2939 printf("lck typ=%d fst=%ju end=%ju\n",
2940 lop->nfslo_type, (intmax_t)lop->nfslo_first,
2941 (intmax_t)lop->nfslo_end);
2943 printf("lck typ=%d fst=%qd end=%qd\n",
2944 lop->nfslo_type, lop->nfslo_first,
2955 * Check for duplicate open owners and opens.
2956 * (Only used as a diagnostic aid.)
2959 nfscl_dupopen(vnode_t vp, int dupopens)
2961 struct nfsclclient *clp;
2962 struct nfsclowner *owp, *owp2;
2963 struct nfsclopen *op, *op2;
2966 clp = VFSTONFS(vnode_mount(vp))->nm_clp;
2968 printf("nfscl dupopen NULL clp\n");
2971 nfhp = VTONFS(vp)->n_fhp;
2975 * First, search for duplicate owners.
2976 * These should never happen!
2978 LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
2979 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
2981 !NFSBCMP(owp->nfsow_owner, owp2->nfsow_owner,
2982 NFSV4CL_LOCKNAMELEN)) {
2984 printf("DUP OWNER\n");
2985 nfscl_dumpstate(VFSTONFS(vnode_mount(vp)), 1, 1, 0, 0);
2992 * Now, search for duplicate stateids.
2993 * These shouldn't happen, either.
2995 LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
2996 LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
2997 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
2998 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3000 (op->nfso_stateid.other[0] != 0 ||
3001 op->nfso_stateid.other[1] != 0 ||
3002 op->nfso_stateid.other[2] != 0) &&
3003 op->nfso_stateid.other[0] == op2->nfso_stateid.other[0] &&
3004 op->nfso_stateid.other[1] == op2->nfso_stateid.other[1] &&
3005 op->nfso_stateid.other[2] == op2->nfso_stateid.other[2]) {
3007 printf("DUP STATEID\n");
3008 nfscl_dumpstate(VFSTONFS(vnode_mount(vp)), 1, 1, 0,
3018 * Now search for duplicate opens.
3019 * Duplicate opens for the same owner
3020 * should never occur. Other duplicates are
3021 * possible and are checked for if "dupopens"
3024 LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3025 LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
3026 if (nfhp->nfh_len == op2->nfso_fhlen &&
3027 !NFSBCMP(nfhp->nfh_fh, op2->nfso_fh, nfhp->nfh_len)) {
3028 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3029 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3030 if (op != op2 && nfhp->nfh_len == op->nfso_fhlen &&
3031 !NFSBCMP(nfhp->nfh_fh, op->nfso_fh, nfhp->nfh_len) &&
3032 (!NFSBCMP(op->nfso_own->nfsow_owner,
3033 op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN) ||
3035 if (!NFSBCMP(op->nfso_own->nfsow_owner,
3036 op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
3038 printf("BADDUP OPEN\n");
3041 printf("DUP OPEN\n");
3043 nfscl_dumpstate(VFSTONFS(vnode_mount(vp)), 1, 1,
3056 * During close, find an open that needs to be dereferenced and
3057 * dereference it. If there are no more opens for this file,
3058 * log a message to that effect.
3059 * Opens aren't actually Close'd until VOP_INACTIVE() is performed
3060 * on the file's vnode.
3061 * This is the safe way, since it is difficult to identify
3062 * which open the close is for and I/O can be performed after the
3063 * close(2) system call when a file is mmap'd.
3064 * If it returns 0 for success, there will be a referenced
3065 * clp returned via clpp.
3068 nfscl_getclose(vnode_t vp, struct nfsclclient **clpp)
3070 struct nfsclclient *clp;
3071 struct nfsclowner *owp;
3072 struct nfsclopen *op;
3073 struct nfscldeleg *dp;
3077 error = nfscl_getcl(vnode_mount(vp), NULL, NULL, 1, &clp);
3082 nfhp = VTONFS(vp)->n_fhp;
3086 * First, look for one under a delegation that was locally issued
3087 * and just decrement the opencnt for it. Since all my Opens against
3088 * the server are DENY_NONE, I don't see a problem with hanging
3089 * onto them. (It is much easier to use one of the extant Opens
3090 * that I already have on the server when a Delegation is recalled
3091 * than to do fresh Opens.) Someday, I might need to rethink this, but.
3093 dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
3095 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
3096 op = LIST_FIRST(&owp->nfsow_open);
3099 * Since a delegation is for a file, there
3100 * should never be more than one open for
3103 if (LIST_NEXT(op, nfso_list) != NULL)
3104 panic("nfscdeleg opens");
3105 if (notdecr && op->nfso_opencnt > 0) {
3114 /* Now process the opens against the server. */
3115 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3116 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3117 if (op->nfso_fhlen == nfhp->nfh_len &&
3118 !NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
3120 /* Found an open, decrement cnt if possible */
3121 if (notdecr && op->nfso_opencnt > 0) {
3126 * There are more opens, so just return.
3128 if (op->nfso_opencnt > 0) {
3137 printf("nfscl: never fnd open\n");
3142 nfscl_doclose(vnode_t vp, struct nfsclclient **clpp, NFSPROC_T *p)
3144 struct nfsclclient *clp;
3145 struct nfsclowner *owp, *nowp;
3146 struct nfsclopen *op;
3147 struct nfscldeleg *dp;
3149 struct nfsclrecalllayout *recallp;
3152 error = nfscl_getcl(vnode_mount(vp), NULL, NULL, 1, &clp);
3157 nfhp = VTONFS(vp)->n_fhp;
3158 recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_WAITOK);
3161 * First get rid of the local Open structures, which should be no
3164 dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
3166 LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) {
3167 op = LIST_FIRST(&owp->nfsow_open);
3169 KASSERT((op->nfso_opencnt == 0),
3170 ("nfscl: bad open cnt on deleg"));
3171 nfscl_freeopen(op, 1);
3173 nfscl_freeopenowner(owp, 1);
3177 /* Return any layouts marked return on close. */
3178 nfscl_retoncloselayout(vp, clp, nfhp->nfh_fh, nfhp->nfh_len, &recallp);
3180 /* Now process the opens against the server. */
3182 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3183 op = LIST_FIRST(&owp->nfsow_open);
3184 while (op != NULL) {
3185 if (op->nfso_fhlen == nfhp->nfh_len &&
3186 !NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
3188 /* Found an open, close it. */
3189 KASSERT((op->nfso_opencnt == 0),
3190 ("nfscl: bad open cnt on server"));
3192 nfsrpc_doclose(VFSTONFS(vnode_mount(vp)), op,
3197 op = LIST_NEXT(op, nfso_list);
3202 * recallp has been set NULL by nfscl_retoncloselayout() if it was
3203 * used by the function, but calling free() with a NULL pointer is ok.
3205 free(recallp, M_NFSLAYRECALL);
3210 * Return all delegations on this client.
3211 * (Must be called with client sleep lock.)
3214 nfscl_delegreturnall(struct nfsclclient *clp, NFSPROC_T *p)
3216 struct nfscldeleg *dp, *ndp;
3219 cred = newnfs_getcred();
3220 TAILQ_FOREACH_SAFE(dp, &clp->nfsc_deleg, nfsdl_list, ndp) {
3221 nfscl_cleandeleg(dp);
3222 (void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
3223 nfscl_freedeleg(&clp->nfsc_deleg, dp);
3229 * Do a callback RPC.
3232 nfscl_docb(struct nfsrv_descript *nd, NFSPROC_T *p)
3234 int clist, gotseq_ok, i, j, k, op, rcalls;
3236 struct nfsclclient *clp;
3237 struct nfscldeleg *dp = NULL;
3238 int numops, taglen = -1, error = 0, trunc __unused;
3239 u_int32_t minorvers = 0, retops = 0, *retopsp = NULL, *repp, cbident;
3240 u_char tag[NFSV4_SMALLSTR + 1], *tagstr;
3246 nfsattrbit_t attrbits, rattrbits;
3247 nfsv4stateid_t stateid;
3248 uint32_t seqid, slotid = 0, highslot, cachethis __unused;
3249 uint8_t sessionid[NFSX_V4SESSIONID];
3251 struct nfscllayout *lyp;
3252 uint64_t filesid[2], len, off;
3253 int changed, gotone, laytype, recalltype;
3255 struct nfsclrecalllayout *recallp = NULL;
3256 struct nfsclsession *tsep;
3260 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3261 taglen = fxdr_unsigned(int, *tl);
3266 if (taglen <= NFSV4_SMALLSTR)
3269 tagstr = malloc(taglen + 1, M_TEMP, M_WAITOK);
3270 error = nfsrv_mtostr(nd, tagstr, taglen);
3272 if (taglen > NFSV4_SMALLSTR)
3273 free(tagstr, M_TEMP);
3277 (void) nfsm_strtom(nd, tag, taglen);
3278 if (taglen > NFSV4_SMALLSTR) {
3279 free(tagstr, M_TEMP);
3281 NFSM_BUILD(retopsp, u_int32_t *, NFSX_UNSIGNED);
3282 NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3283 minorvers = fxdr_unsigned(u_int32_t, *tl++);
3284 if (minorvers != NFSV4_MINORVERSION && minorvers != NFSV41_MINORVERSION)
3285 nd->nd_repstat = NFSERR_MINORVERMISMATCH;
3286 cbident = fxdr_unsigned(u_int32_t, *tl++);
3290 numops = fxdr_unsigned(int, *tl);
3292 * Loop around doing the sub ops.
3294 for (i = 0; i < numops; i++) {
3295 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3296 NFSM_BUILD(repp, u_int32_t *, 2 * NFSX_UNSIGNED);
3298 op = fxdr_unsigned(int, *tl);
3299 if (op < NFSV4OP_CBGETATTR ||
3300 (op > NFSV4OP_CBRECALL && minorvers == NFSV4_MINORVERSION) ||
3301 (op > NFSV4OP_CBNOTIFYDEVID &&
3302 minorvers == NFSV41_MINORVERSION)) {
3303 nd->nd_repstat = NFSERR_OPILLEGAL;
3304 *repp = nfscl_errmap(nd, minorvers);
3308 nd->nd_procnum = op;
3309 if (op < NFSV41_CBNOPS)
3310 nfsstatsv1.cbrpccnt[nd->nd_procnum]++;
3312 case NFSV4OP_CBGETATTR:
3313 NFSCL_DEBUG(4, "cbgetattr\n");
3316 error = nfsm_getfh(nd, &nfhp);
3318 error = nfsrv_getattrbits(nd, &attrbits,
3320 if (error == 0 && i == 0 &&
3321 minorvers != NFSV4_MINORVERSION)
3322 error = NFSERR_OPNOTINSESS;
3324 mp = nfscl_getmnt(minorvers, sessionid, cbident,
3327 error = NFSERR_SERVERFAULT;
3330 error = nfscl_ngetreopen(mp, nfhp->nfh_fh,
3331 nfhp->nfh_len, p, &np);
3336 NFSZERO_ATTRBIT(&rattrbits);
3338 dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
3341 if (NFSISSET_ATTRBIT(&attrbits,
3344 va.va_size = np->n_size;
3348 NFSSETBIT_ATTRBIT(&rattrbits,
3351 if (NFSISSET_ATTRBIT(&attrbits,
3352 NFSATTRBIT_CHANGE)) {
3356 (np->n_flag & NDELEGMOD))
3358 NFSSETBIT_ATTRBIT(&rattrbits,
3362 error = NFSERR_SERVERFAULT;
3370 free(nfhp, M_NFSFH);
3372 (void) nfsv4_fillattr(nd, NULL, NULL, NULL, &va,
3373 NULL, 0, &rattrbits, NULL, p, 0, 0, 0, 0,
3376 case NFSV4OP_CBRECALL:
3377 NFSCL_DEBUG(4, "cbrecall\n");
3378 NFSM_DISSECT(tl, u_int32_t *, NFSX_STATEID +
3380 stateid.seqid = *tl++;
3381 NFSBCOPY((caddr_t)tl, (caddr_t)stateid.other,
3383 tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
3384 trunc = fxdr_unsigned(int, *tl);
3385 error = nfsm_getfh(nd, &nfhp);
3386 if (error == 0 && i == 0 &&
3387 minorvers != NFSV4_MINORVERSION)
3388 error = NFSERR_OPNOTINSESS;
3391 if (minorvers == NFSV4_MINORVERSION)
3392 clp = nfscl_getclnt(cbident);
3394 clp = nfscl_getclntsess(sessionid);
3396 dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
3398 if (dp != NULL && (dp->nfsdl_flags &
3399 NFSCLDL_DELEGRET) == 0) {
3402 wakeup((caddr_t)clp);
3405 error = NFSERR_SERVERFAULT;
3410 free(nfhp, M_NFSFH);
3412 case NFSV4OP_CBLAYOUTRECALL:
3413 NFSCL_DEBUG(4, "cblayrec\n");
3415 NFSM_DISSECT(tl, uint32_t *, 4 * NFSX_UNSIGNED);
3416 laytype = fxdr_unsigned(int, *tl++);
3417 iomode = fxdr_unsigned(uint32_t, *tl++);
3418 if (newnfs_true == *tl++)
3422 recalltype = fxdr_unsigned(int, *tl);
3423 NFSCL_DEBUG(4, "layt=%d iom=%d ch=%d rectyp=%d\n",
3424 laytype, iomode, changed, recalltype);
3425 recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL,
3427 if (laytype != NFSLAYOUT_NFSV4_1_FILES &&
3428 laytype != NFSLAYOUT_FLEXFILE)
3429 error = NFSERR_NOMATCHLAYOUT;
3430 else if (recalltype == NFSLAYOUTRETURN_FILE) {
3431 error = nfsm_getfh(nd, &nfhp);
3432 NFSCL_DEBUG(4, "retfile getfh=%d\n", error);
3435 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_HYPER +
3437 off = fxdr_hyper(tl); tl += 2;
3438 len = fxdr_hyper(tl); tl += 2;
3439 stateid.seqid = fxdr_unsigned(uint32_t, *tl++);
3440 NFSBCOPY(tl, stateid.other, NFSX_STATEIDOTHER);
3441 if (minorvers == NFSV4_MINORVERSION)
3442 error = NFSERR_NOTSUPP;
3444 error = NFSERR_OPNOTINSESS;
3445 NFSCL_DEBUG(4, "off=%ju len=%ju sq=%u err=%d\n",
3446 (uintmax_t)off, (uintmax_t)len,
3447 stateid.seqid, error);
3450 clp = nfscl_getclntsess(sessionid);
3451 NFSCL_DEBUG(4, "cbly clp=%p\n", clp);
3453 lyp = nfscl_findlayout(clp,
3456 NFSCL_DEBUG(4, "cblyp=%p\n",
3461 NFSLY_FLEXFILE)) != 0 &&
3462 !NFSBCMP(stateid.other,
3463 lyp->nfsly_stateid.other,
3464 NFSX_STATEIDOTHER)) {
3474 lyp->nfsly_stateid.seqid)
3475 lyp->nfsly_stateid.seqid =
3483 lyp->nfsly_stateid.seqid);
3486 NFSERR_NOMATCHLAYOUT;
3488 error = NFSERR_NOMATCHLAYOUT;
3491 free(nfhp, M_NFSFH);
3492 } else if (recalltype == NFSLAYOUTRETURN_FSID) {
3493 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_HYPER);
3494 filesid[0] = fxdr_hyper(tl); tl += 2;
3495 filesid[1] = fxdr_hyper(tl); tl += 2;
3498 clp = nfscl_getclntsess(sessionid);
3500 TAILQ_FOREACH(lyp, &clp->nfsc_layout,
3502 if (lyp->nfsly_filesid[0] ==
3504 lyp->nfsly_filesid[1] ==
3511 lyp->nfsly_stateid.seqid,
3521 error = NFSERR_NOMATCHLAYOUT;
3523 error = NFSERR_NOMATCHLAYOUT;
3525 } else if (recalltype == NFSLAYOUTRETURN_ALL) {
3528 clp = nfscl_getclntsess(sessionid);
3530 TAILQ_FOREACH(lyp, &clp->nfsc_layout,
3532 error = nfscl_layoutrecall(
3533 recalltype, lyp, iomode, 0,
3535 lyp->nfsly_stateid.seqid,
3536 0, 0, NULL, recallp);
3543 error = NFSERR_NOMATCHLAYOUT;
3545 error = NFSERR_NOMATCHLAYOUT;
3548 error = NFSERR_NOMATCHLAYOUT;
3549 if (recallp != NULL) {
3550 free(recallp, M_NFSLAYRECALL);
3554 case NFSV4OP_CBSEQUENCE:
3555 NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID +
3557 bcopy(tl, sessionid, NFSX_V4SESSIONID);
3558 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3559 seqid = fxdr_unsigned(uint32_t, *tl++);
3560 slotid = fxdr_unsigned(uint32_t, *tl++);
3561 highslot = fxdr_unsigned(uint32_t, *tl++);
3563 /* Throw away the referring call stuff. */
3564 clist = fxdr_unsigned(int, *tl);
3565 for (j = 0; j < clist; j++) {
3566 NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID +
3568 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3569 rcalls = fxdr_unsigned(int, *tl);
3570 for (k = 0; k < rcalls; k++) {
3571 NFSM_DISSECT(tl, uint32_t *,
3577 clp = nfscl_getclntsess(sessionid);
3579 error = NFSERR_SERVERFAULT;
3581 error = NFSERR_SEQUENCEPOS;
3583 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3584 error = nfsv4_seqsession(seqid, slotid,
3585 highslot, tsep->nfsess_cbslots, &rep,
3586 tsep->nfsess_backslots);
3589 if (error == 0 || error == NFSERR_REPLYFROMCACHE) {
3593 * Handle a reply for a retried
3594 * callback. The reply will be
3595 * re-inserted in the session cache
3596 * by the nfsv4_seqsess_cacherep() call
3599 KASSERT(error == NFSERR_REPLYFROMCACHE,
3600 ("cbsequence: non-NULL rep"));
3601 NFSCL_DEBUG(4, "Got cbretry\n");
3602 m_freem(nd->nd_mreq);
3607 NFSM_BUILD(tl, uint32_t *,
3608 NFSX_V4SESSIONID + 4 * NFSX_UNSIGNED);
3609 bcopy(sessionid, tl, NFSX_V4SESSIONID);
3610 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3611 *tl++ = txdr_unsigned(seqid);
3612 *tl++ = txdr_unsigned(slotid);
3613 *tl++ = txdr_unsigned(NFSV4_CBSLOTS - 1);
3614 *tl = txdr_unsigned(NFSV4_CBSLOTS - 1);
3618 if (i == 0 && minorvers == NFSV41_MINORVERSION)
3619 error = NFSERR_OPNOTINSESS;
3621 NFSCL_DEBUG(1, "unsupp callback %d\n", op);
3622 error = NFSERR_NOTSUPP;
3627 if (error == EBADRPC || error == NFSERR_BADXDR) {
3628 nd->nd_repstat = NFSERR_BADXDR;
3630 nd->nd_repstat = error;
3635 if (nd->nd_repstat) {
3636 *repp = nfscl_errmap(nd, minorvers);
3639 *repp = 0; /* NFS4_OK */
3642 if (recallp != NULL)
3643 free(recallp, M_NFSLAYRECALL);
3645 if (error == EBADRPC || error == NFSERR_BADXDR)
3646 nd->nd_repstat = NFSERR_BADXDR;
3648 printf("nfsv4 comperr1=%d\n", error);
3651 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
3655 *retopsp = txdr_unsigned(retops);
3657 *nd->nd_errp = nfscl_errmap(nd, minorvers);
3659 if (gotseq_ok != 0) {
3660 rep = m_copym(nd->nd_mreq, 0, M_COPYALL, M_WAITOK);
3662 clp = nfscl_getclntsess(sessionid);
3664 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3665 nfsv4_seqsess_cacherep(slotid, tsep->nfsess_cbslots,
3676 * Generate the next cbident value. Basically just increment a static value
3677 * and then check that it isn't already in the list, if it has wrapped around.
3680 nfscl_nextcbident(void)
3682 struct nfsclclient *clp;
3684 static u_int32_t nextcbident = 0;
3685 static int haswrapped = 0;
3688 if (nextcbident == 0)
3692 * Search the clientid list for one already using this cbident.
3697 LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3698 if (clp->nfsc_cbident == nextcbident) {
3708 return (nextcbident);
3712 * Get the mount point related to a given cbident or session and busy it.
3715 nfscl_getmnt(int minorvers, uint8_t *sessionid, u_int32_t cbident,
3716 struct nfsclclient **clpp)
3718 struct nfsclclient *clp;
3721 struct nfsclsession *tsep;
3725 LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3726 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3727 if (minorvers == NFSV4_MINORVERSION) {
3728 if (clp->nfsc_cbident == cbident)
3730 } else if (!NFSBCMP(tsep->nfsess_sessionid, sessionid,
3738 mp = clp->nfsc_nmp->nm_mountp;
3741 error = vfs_busy(mp, 0);
3750 * Get the clientid pointer related to a given cbident.
3752 static struct nfsclclient *
3753 nfscl_getclnt(u_int32_t cbident)
3755 struct nfsclclient *clp;
3757 LIST_FOREACH(clp, &nfsclhead, nfsc_list)
3758 if (clp->nfsc_cbident == cbident)
3764 * Get the clientid pointer related to a given sessionid.
3766 static struct nfsclclient *
3767 nfscl_getclntsess(uint8_t *sessionid)
3769 struct nfsclclient *clp;
3770 struct nfsclsession *tsep;
3772 LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3773 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3774 if (!NFSBCMP(tsep->nfsess_sessionid, sessionid,
3782 * Search for a lock conflict locally on the client. A conflict occurs if
3783 * - not same owner and overlapping byte range and at least one of them is
3784 * a write lock or this is an unlock.
3787 nfscl_localconflict(struct nfsclclient *clp, u_int8_t *fhp, int fhlen,
3788 struct nfscllock *nlop, u_int8_t *own, struct nfscldeleg *dp,
3789 struct nfscllock **lopp)
3791 struct nfsclowner *owp;
3792 struct nfsclopen *op;
3796 ret = nfscl_checkconflict(&dp->nfsdl_lock, nlop, own, lopp);
3800 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3801 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3802 if (op->nfso_fhlen == fhlen &&
3803 !NFSBCMP(op->nfso_fh, fhp, fhlen)) {
3804 ret = nfscl_checkconflict(&op->nfso_lock, nlop,
3815 nfscl_checkconflict(struct nfscllockownerhead *lhp, struct nfscllock *nlop,
3816 u_int8_t *own, struct nfscllock **lopp)
3818 struct nfscllockowner *lp;
3819 struct nfscllock *lop;
3821 LIST_FOREACH(lp, lhp, nfsl_list) {
3822 if (NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
3823 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
3824 if (lop->nfslo_first >= nlop->nfslo_end)
3826 if (lop->nfslo_end <= nlop->nfslo_first)
3828 if (lop->nfslo_type == F_WRLCK ||
3829 nlop->nfslo_type == F_WRLCK ||
3830 nlop->nfslo_type == F_UNLCK) {
3833 return (NFSERR_DENIED);
3842 * Check for a local conflicting lock.
3845 nfscl_lockt(vnode_t vp, struct nfsclclient *clp, u_int64_t off,
3846 u_int64_t len, struct flock *fl, NFSPROC_T *p, void *id, int flags)
3848 struct nfscllock *lop, nlck;
3849 struct nfscldeleg *dp;
3851 u_int8_t own[NFSV4CL_LOCKNAMELEN];
3854 nlck.nfslo_type = fl->l_type;
3855 nlck.nfslo_first = off;
3856 if (len == NFS64BITSSET) {
3857 nlck.nfslo_end = NFS64BITSSET;
3859 nlck.nfslo_end = off + len;
3860 if (nlck.nfslo_end <= nlck.nfslo_first)
3861 return (NFSERR_INVAL);
3864 nfscl_filllockowner(id, own, flags);
3866 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
3867 error = nfscl_localconflict(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len,
3868 &nlck, own, dp, &lop);
3870 fl->l_whence = SEEK_SET;
3871 fl->l_start = lop->nfslo_first;
3872 if (lop->nfslo_end == NFS64BITSSET)
3875 fl->l_len = lop->nfslo_end - lop->nfslo_first;
3876 fl->l_pid = (pid_t)0;
3877 fl->l_type = lop->nfslo_type;
3878 error = -1; /* no RPC required */
3879 } else if (dp != NULL && ((dp->nfsdl_flags & NFSCLDL_WRITE) ||
3880 fl->l_type == F_RDLCK)) {
3882 * The delegation ensures that there isn't a conflicting
3883 * lock on the server, so return -1 to indicate an RPC
3886 fl->l_type = F_UNLCK;
3894 * Handle Recall of a delegation.
3895 * The clp must be exclusive locked when this is called.
3898 nfscl_recalldeleg(struct nfsclclient *clp, struct nfsmount *nmp,
3899 struct nfscldeleg *dp, vnode_t vp, struct ucred *cred, NFSPROC_T *p,
3900 int called_from_renewthread)
3902 struct nfsclowner *owp, *lowp, *nowp;
3903 struct nfsclopen *op, *lop;
3904 struct nfscllockowner *lp;
3905 struct nfscllock *lckp;
3907 int error = 0, ret, gotvp = 0;
3911 * First, get a vnode for the file. This is needed to do RPCs.
3913 ret = nfscl_ngetreopen(nmp->nm_mountp, dp->nfsdl_fh,
3914 dp->nfsdl_fhlen, p, &np);
3917 * File isn't open, so nothing to move over to the
3927 dp->nfsdl_flags &= ~NFSCLDL_MODTIMESET;
3930 * Ok, if it's a write delegation, flush data to the server, so
3931 * that close/open consistency is retained.
3935 if ((dp->nfsdl_flags & NFSCLDL_WRITE) && (np->n_flag & NMODIFIED)) {
3936 np->n_flag |= NDELEGRECALL;
3938 ret = ncl_flush(vp, MNT_WAIT, p, 1, called_from_renewthread);
3940 np->n_flag &= ~NDELEGRECALL;
3942 NFSINVALATTRCACHE(np);
3944 if (ret == EIO && called_from_renewthread != 0) {
3946 * If the flush failed with EIO for the renew thread,
3947 * return now, so that the dirty buffer will be flushed
3956 * Now, for each openowner with opens issued locally, move them
3957 * over to state against the server.
3959 LIST_FOREACH(lowp, &dp->nfsdl_owner, nfsow_list) {
3960 lop = LIST_FIRST(&lowp->nfsow_open);
3962 if (LIST_NEXT(lop, nfso_list) != NULL)
3963 panic("nfsdlg mult opens");
3965 * Look for the same openowner against the server.
3967 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3968 if (!NFSBCMP(lowp->nfsow_owner,
3969 owp->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
3970 newnfs_copycred(&dp->nfsdl_cred, cred);
3971 ret = nfscl_moveopen(vp, clp, nmp, lop,
3973 if (ret == NFSERR_STALECLIENTID ||
3974 ret == NFSERR_STALEDONTRECOVER ||
3975 ret == NFSERR_BADSESSION) {
3981 nfscl_freeopen(lop, 1);
3990 * If no openowner found, create one and get an open
3995 sizeof (struct nfsclowner), M_NFSCLOWNER,
3997 nfscl_newopen(clp, NULL, &owp, &nowp, &op,
3998 NULL, lowp->nfsow_owner, dp->nfsdl_fh,
3999 dp->nfsdl_fhlen, NULL, NULL);
4000 newnfs_copycred(&dp->nfsdl_cred, cred);
4001 ret = nfscl_moveopen(vp, clp, nmp, lop,
4004 nfscl_freeopenowner(owp, 0);
4005 if (ret == NFSERR_STALECLIENTID ||
4006 ret == NFSERR_STALEDONTRECOVER ||
4007 ret == NFSERR_BADSESSION) {
4013 nfscl_freeopen(lop, 1);
4023 * Now, get byte range locks for any locks done locally.
4025 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4026 LIST_FOREACH(lckp, &lp->nfsl_lock, nfslo_list) {
4027 newnfs_copycred(&dp->nfsdl_cred, cred);
4028 ret = nfscl_relock(vp, clp, nmp, lp, lckp, cred, p);
4029 if (ret == NFSERR_STALESTATEID ||
4030 ret == NFSERR_STALEDONTRECOVER ||
4031 ret == NFSERR_STALECLIENTID ||
4032 ret == NFSERR_BADSESSION) {
4047 * Move a locally issued open over to an owner on the state list.
4048 * SIDE EFFECT: If it needs to sleep (do an rpc), it unlocks clstate and
4049 * returns with it unlocked.
4052 nfscl_moveopen(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp,
4053 struct nfsclopen *lop, struct nfsclowner *owp, struct nfscldeleg *dp,
4054 struct ucred *cred, NFSPROC_T *p)
4056 struct nfsclopen *op, *nop;
4057 struct nfscldeleg *ndp;
4059 int error = 0, newone;
4062 * First, look for an appropriate open, If found, just increment the
4065 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
4066 if ((op->nfso_mode & lop->nfso_mode) == lop->nfso_mode &&
4067 op->nfso_fhlen == lop->nfso_fhlen &&
4068 !NFSBCMP(op->nfso_fh, lop->nfso_fh, op->nfso_fhlen)) {
4069 op->nfso_opencnt += lop->nfso_opencnt;
4070 nfscl_freeopen(lop, 1);
4075 /* No appropriate open, so we have to do one against the server. */
4077 nop = malloc(sizeof (struct nfsclopen) +
4078 lop->nfso_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
4080 nfscl_newopen(clp, NULL, &owp, NULL, &op, &nop, owp->nfsow_owner,
4081 lop->nfso_fh, lop->nfso_fhlen, cred, &newone);
4083 error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data, np->n_v4->n4_fhlen,
4084 lop->nfso_fh, lop->nfso_fhlen, lop->nfso_mode, op,
4085 NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, &ndp, 0, 0, cred, p);
4088 nfscl_freeopen(op, 0);
4090 op->nfso_mode |= lop->nfso_mode;
4091 op->nfso_opencnt += lop->nfso_opencnt;
4092 nfscl_freeopen(lop, 1);
4095 free(nop, M_NFSCLOPEN);
4098 * What should I do with the returned delegation, since the
4099 * delegation is being recalled? For now, just printf and
4102 printf("Moveopen returned deleg\n");
4103 free(ndp, M_NFSCLDELEG);
4109 * Recall all delegations on this client.
4112 nfscl_totalrecall(struct nfsclclient *clp)
4114 struct nfscldeleg *dp;
4116 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
4117 if ((dp->nfsdl_flags & NFSCLDL_DELEGRET) == 0)
4118 dp->nfsdl_flags |= NFSCLDL_RECALL;
4123 * Relock byte ranges. Called for delegation recall and state expiry.
4126 nfscl_relock(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp,
4127 struct nfscllockowner *lp, struct nfscllock *lop, struct ucred *cred,
4130 struct nfscllockowner *nlp;
4133 int error, newone, donelocally;
4135 off = lop->nfslo_first;
4136 len = lop->nfslo_end - lop->nfslo_first;
4137 error = nfscl_getbytelock(vp, off, len, lop->nfslo_type, cred, p,
4138 clp, 1, NULL, lp->nfsl_lockflags, lp->nfsl_owner,
4139 lp->nfsl_openowner, &nlp, &newone, &donelocally);
4140 if (error || donelocally)
4142 nfhp = VTONFS(vp)->n_fhp;
4143 error = nfscl_trylock(nmp, vp, nfhp->nfh_fh,
4144 nfhp->nfh_len, nlp, newone, 0, off,
4145 len, lop->nfslo_type, cred, p);
4147 nfscl_freelockowner(nlp, 0);
4152 * Called to re-open a file. Basically get a vnode for the file handle
4153 * and then call nfsrpc_openrpc() to do the rest.
4156 nfsrpc_reopen(struct nfsmount *nmp, u_int8_t *fhp, int fhlen,
4157 u_int32_t mode, struct nfsclopen *op, struct nfscldeleg **dpp,
4158 struct ucred *cred, NFSPROC_T *p)
4164 error = nfscl_ngetreopen(nmp->nm_mountp, fhp, fhlen, p, &np);
4168 if (np->n_v4 != NULL) {
4169 error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data,
4170 np->n_v4->n4_fhlen, fhp, fhlen, mode, op,
4171 NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, dpp, 0, 0,
4181 * Try an open against the server. Just call nfsrpc_openrpc(), retrying while
4182 * NFSERR_DELAY. Also, try system credentials, if the passed in credentials
4186 nfscl_tryopen(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen,
4187 u_int8_t *newfhp, int newfhlen, u_int32_t mode, struct nfsclopen *op,
4188 u_int8_t *name, int namelen, struct nfscldeleg **ndpp,
4189 int reclaim, u_int32_t delegtype, struct ucred *cred, NFSPROC_T *p)
4194 error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp, newfhlen,
4195 mode, op, name, namelen, ndpp, reclaim, delegtype, cred, p,
4197 if (error == NFSERR_DELAY)
4198 (void) nfs_catnap(PZERO, error, "nfstryop");
4199 } while (error == NFSERR_DELAY);
4200 if (error == EAUTH || error == EACCES) {
4201 /* Try again using system credentials */
4202 newnfs_setroot(cred);
4204 error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp,
4205 newfhlen, mode, op, name, namelen, ndpp, reclaim,
4206 delegtype, cred, p, 1, 0);
4207 if (error == NFSERR_DELAY)
4208 (void) nfs_catnap(PZERO, error, "nfstryop");
4209 } while (error == NFSERR_DELAY);
4215 * Try a byte range lock. Just loop on nfsrpc_lock() while it returns
4216 * NFSERR_DELAY. Also, retry with system credentials, if the provided
4220 nfscl_trylock(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp,
4221 int fhlen, struct nfscllockowner *nlp, int newone, int reclaim,
4222 u_int64_t off, u_int64_t len, short type, struct ucred *cred, NFSPROC_T *p)
4224 struct nfsrv_descript nfsd, *nd = &nfsd;
4228 error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp, newone,
4229 reclaim, off, len, type, cred, p, 0);
4230 if (!error && nd->nd_repstat == NFSERR_DELAY)
4231 (void) nfs_catnap(PZERO, (int)nd->nd_repstat,
4233 } while (!error && nd->nd_repstat == NFSERR_DELAY);
4235 error = nd->nd_repstat;
4236 if (error == EAUTH || error == EACCES) {
4237 /* Try again using root credentials */
4238 newnfs_setroot(cred);
4240 error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp,
4241 newone, reclaim, off, len, type, cred, p, 1);
4242 if (!error && nd->nd_repstat == NFSERR_DELAY)
4243 (void) nfs_catnap(PZERO, (int)nd->nd_repstat,
4245 } while (!error && nd->nd_repstat == NFSERR_DELAY);
4247 error = nd->nd_repstat;
4253 * Try a delegreturn against the server. Just call nfsrpc_delegreturn(),
4254 * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
4258 nfscl_trydelegreturn(struct nfscldeleg *dp, struct ucred *cred,
4259 struct nfsmount *nmp, NFSPROC_T *p)
4264 error = nfsrpc_delegreturn(dp, cred, nmp, p, 0);
4265 if (error == NFSERR_DELAY)
4266 (void) nfs_catnap(PZERO, error, "nfstrydp");
4267 } while (error == NFSERR_DELAY);
4268 if (error == EAUTH || error == EACCES) {
4269 /* Try again using system credentials */
4270 newnfs_setroot(cred);
4272 error = nfsrpc_delegreturn(dp, cred, nmp, p, 1);
4273 if (error == NFSERR_DELAY)
4274 (void) nfs_catnap(PZERO, error, "nfstrydp");
4275 } while (error == NFSERR_DELAY);
4281 * Try a close against the server. Just call nfsrpc_closerpc(),
4282 * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
4286 nfscl_tryclose(struct nfsclopen *op, struct ucred *cred,
4287 struct nfsmount *nmp, NFSPROC_T *p)
4289 struct nfsrv_descript nfsd, *nd = &nfsd;
4293 error = nfsrpc_closerpc(nd, nmp, op, cred, p, 0);
4294 if (error == NFSERR_DELAY)
4295 (void) nfs_catnap(PZERO, error, "nfstrycl");
4296 } while (error == NFSERR_DELAY);
4297 if (error == EAUTH || error == EACCES) {
4298 /* Try again using system credentials */
4299 newnfs_setroot(cred);
4301 error = nfsrpc_closerpc(nd, nmp, op, cred, p, 1);
4302 if (error == NFSERR_DELAY)
4303 (void) nfs_catnap(PZERO, error, "nfstrycl");
4304 } while (error == NFSERR_DELAY);
4310 * Decide if a delegation on a file permits close without flushing writes
4311 * to the server. This might be a big performance win in some environments.
4312 * (Not useful until the client does caching on local stable storage.)
4315 nfscl_mustflush(vnode_t vp)
4317 struct nfsclclient *clp;
4318 struct nfscldeleg *dp;
4320 struct nfsmount *nmp;
4323 nmp = VFSTONFS(vnode_mount(vp));
4324 if (!NFSHASNFSV4(nmp))
4327 clp = nfscl_findcl(nmp);
4332 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4333 if (dp != NULL && (dp->nfsdl_flags &
4334 (NFSCLDL_WRITE | NFSCLDL_RECALL | NFSCLDL_DELEGRET)) ==
4336 (dp->nfsdl_sizelimit >= np->n_size ||
4337 !NFSHASSTRICT3530(nmp))) {
4346 * See if a (write) delegation exists for this file.
4349 nfscl_nodeleg(vnode_t vp, int writedeleg)
4351 struct nfsclclient *clp;
4352 struct nfscldeleg *dp;
4354 struct nfsmount *nmp;
4357 nmp = VFSTONFS(vnode_mount(vp));
4358 if (!NFSHASNFSV4(nmp))
4361 clp = nfscl_findcl(nmp);
4366 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4368 (dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_DELEGRET)) == 0 &&
4369 (writedeleg == 0 || (dp->nfsdl_flags & NFSCLDL_WRITE) ==
4379 * Look for an associated delegation that should be DelegReturned.
4382 nfscl_removedeleg(vnode_t vp, NFSPROC_T *p, nfsv4stateid_t *stp)
4384 struct nfsclclient *clp;
4385 struct nfscldeleg *dp;
4386 struct nfsclowner *owp;
4387 struct nfscllockowner *lp;
4388 struct nfsmount *nmp;
4391 int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
4393 nmp = VFSTONFS(vnode_mount(vp));
4397 * Loop around waiting for:
4398 * - outstanding I/O operations on delegations to complete
4399 * - for a delegation on vp that has state, lock the client and
4401 * - return delegation with no state
4404 clp = nfscl_findcl(nmp);
4409 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4410 np->n_fhp->nfh_len);
4413 * Wait for outstanding I/O ops to be done.
4415 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4417 nfsv4_unlock(&clp->nfsc_lock, 0);
4420 dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4421 (void) nfsmsleep(&dp->nfsdl_rwlock,
4422 NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4426 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4427 if (!LIST_EMPTY(&owp->nfsow_open)) {
4433 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4434 if (!LIST_EMPTY(&lp->nfsl_lock)) {
4440 if (needsrecall && !triedrecall) {
4441 dp->nfsdl_flags |= NFSCLDL_DELEGRET;
4444 igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
4445 &islept, NFSCLSTATEMUTEXPTR, NULL);
4452 cred = newnfs_getcred();
4453 newnfs_copycred(&dp->nfsdl_cred, cred);
4454 (void) nfscl_recalldeleg(clp, nmp, dp, vp, cred, p, 0);
4458 nfsv4_unlock(&clp->nfsc_lock, 0);
4462 *stp = dp->nfsdl_stateid;
4464 nfscl_cleandeleg(dp);
4465 nfscl_freedeleg(&clp->nfsc_deleg, dp);
4468 nfsv4_unlock(&clp->nfsc_lock, 0);
4475 * Look for associated delegation(s) that should be DelegReturned.
4478 nfscl_renamedeleg(vnode_t fvp, nfsv4stateid_t *fstp, int *gotfdp, vnode_t tvp,
4479 nfsv4stateid_t *tstp, int *gottdp, NFSPROC_T *p)
4481 struct nfsclclient *clp;
4482 struct nfscldeleg *dp;
4483 struct nfsclowner *owp;
4484 struct nfscllockowner *lp;
4485 struct nfsmount *nmp;
4488 int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
4490 nmp = VFSTONFS(vnode_mount(fvp));
4495 * Loop around waiting for:
4496 * - outstanding I/O operations on delegations to complete
4497 * - for a delegation on fvp that has state, lock the client and
4499 * - return delegation(s) with no state.
4502 clp = nfscl_findcl(nmp);
4508 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4509 np->n_fhp->nfh_len);
4510 if (dp != NULL && *gotfdp == 0) {
4512 * Wait for outstanding I/O ops to be done.
4514 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4516 nfsv4_unlock(&clp->nfsc_lock, 0);
4519 dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4520 (void) nfsmsleep(&dp->nfsdl_rwlock,
4521 NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4525 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4526 if (!LIST_EMPTY(&owp->nfsow_open)) {
4532 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4533 if (!LIST_EMPTY(&lp->nfsl_lock)) {
4539 if (needsrecall && !triedrecall) {
4540 dp->nfsdl_flags |= NFSCLDL_DELEGRET;
4543 igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
4544 &islept, NFSCLSTATEMUTEXPTR, NULL);
4551 cred = newnfs_getcred();
4552 newnfs_copycred(&dp->nfsdl_cred, cred);
4553 (void) nfscl_recalldeleg(clp, nmp, dp, fvp, cred, p, 0);
4557 nfsv4_unlock(&clp->nfsc_lock, 0);
4561 *fstp = dp->nfsdl_stateid;
4564 nfscl_cleandeleg(dp);
4565 nfscl_freedeleg(&clp->nfsc_deleg, dp);
4568 nfsv4_unlock(&clp->nfsc_lock, 0);
4573 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4574 np->n_fhp->nfh_len);
4575 if (dp != NULL && *gottdp == 0) {
4577 * Wait for outstanding I/O ops to be done.
4579 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4580 dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4581 (void) nfsmsleep(&dp->nfsdl_rwlock,
4582 NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4585 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4586 if (!LIST_EMPTY(&owp->nfsow_open)) {
4591 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4592 if (!LIST_EMPTY(&lp->nfsl_lock)) {
4597 *tstp = dp->nfsdl_stateid;
4600 nfscl_cleandeleg(dp);
4601 nfscl_freedeleg(&clp->nfsc_deleg, dp);
4610 * Get a reference on the clientid associated with the mount point.
4611 * Return 1 if success, 0 otherwise.
4614 nfscl_getref(struct nfsmount *nmp)
4616 struct nfsclclient *clp;
4619 clp = nfscl_findcl(nmp);
4624 nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, NULL);
4630 * Release a reference on a clientid acquired with the above call.
4633 nfscl_relref(struct nfsmount *nmp)
4635 struct nfsclclient *clp;
4638 clp = nfscl_findcl(nmp);
4643 nfsv4_relref(&clp->nfsc_lock);
4648 * Save the size attribute in the delegation, since the nfsnode
4652 nfscl_reclaimnode(vnode_t vp)
4654 struct nfsclclient *clp;
4655 struct nfscldeleg *dp;
4656 struct nfsnode *np = VTONFS(vp);
4657 struct nfsmount *nmp;
4659 nmp = VFSTONFS(vnode_mount(vp));
4660 if (!NFSHASNFSV4(nmp))
4663 clp = nfscl_findcl(nmp);
4668 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4669 if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE))
4670 dp->nfsdl_size = np->n_size;
4675 * Get the saved size attribute in the delegation, since it is a
4676 * newly allocated nfsnode.
4679 nfscl_newnode(vnode_t vp)
4681 struct nfsclclient *clp;
4682 struct nfscldeleg *dp;
4683 struct nfsnode *np = VTONFS(vp);
4684 struct nfsmount *nmp;
4686 nmp = VFSTONFS(vnode_mount(vp));
4687 if (!NFSHASNFSV4(nmp))
4690 clp = nfscl_findcl(nmp);
4695 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4696 if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE))
4697 np->n_size = dp->nfsdl_size;
4702 * If there is a valid write delegation for this file, set the modtime
4703 * to the local clock time.
4706 nfscl_delegmodtime(vnode_t vp)
4708 struct nfsclclient *clp;
4709 struct nfscldeleg *dp;
4710 struct nfsnode *np = VTONFS(vp);
4711 struct nfsmount *nmp;
4713 nmp = VFSTONFS(vnode_mount(vp));
4714 if (!NFSHASNFSV4(nmp))
4717 clp = nfscl_findcl(nmp);
4722 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4723 if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE)) {
4724 nanotime(&dp->nfsdl_modtime);
4725 dp->nfsdl_flags |= NFSCLDL_MODTIMESET;
4731 * If there is a valid write delegation for this file with a modtime set,
4732 * put that modtime in mtime.
4735 nfscl_deleggetmodtime(vnode_t vp, struct timespec *mtime)
4737 struct nfsclclient *clp;
4738 struct nfscldeleg *dp;
4739 struct nfsnode *np = VTONFS(vp);
4740 struct nfsmount *nmp;
4742 nmp = VFSTONFS(vnode_mount(vp));
4743 if (!NFSHASNFSV4(nmp))
4746 clp = nfscl_findcl(nmp);
4751 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4753 (dp->nfsdl_flags & (NFSCLDL_WRITE | NFSCLDL_MODTIMESET)) ==
4754 (NFSCLDL_WRITE | NFSCLDL_MODTIMESET))
4755 *mtime = dp->nfsdl_modtime;
4760 nfscl_errmap(struct nfsrv_descript *nd, u_int32_t minorvers)
4762 short *defaulterrp, *errp;
4764 if (!nd->nd_repstat)
4766 if (nd->nd_procnum == NFSPROC_NOOP)
4767 return (txdr_unsigned(nd->nd_repstat & 0xffff));
4768 if (nd->nd_repstat == EBADRPC)
4769 return (txdr_unsigned(NFSERR_BADXDR));
4770 if (nd->nd_repstat == NFSERR_MINORVERMISMATCH ||
4771 nd->nd_repstat == NFSERR_OPILLEGAL)
4772 return (txdr_unsigned(nd->nd_repstat));
4773 if (nd->nd_repstat >= NFSERR_BADIOMODE && nd->nd_repstat < 20000 &&
4774 minorvers > NFSV4_MINORVERSION) {
4775 /* NFSv4.n error. */
4776 return (txdr_unsigned(nd->nd_repstat));
4778 if (nd->nd_procnum < NFSV4OP_CBNOPS)
4779 errp = defaulterrp = nfscl_cberrmap[nd->nd_procnum];
4781 return (txdr_unsigned(nd->nd_repstat));
4783 if (*errp == (short)nd->nd_repstat)
4784 return (txdr_unsigned(nd->nd_repstat));
4785 return (txdr_unsigned(*defaulterrp));
4789 * Called to find/add a layout to a client.
4790 * This function returns the layout with a refcnt (shared lock) upon
4791 * success (returns 0) or with no lock/refcnt on the layout when an
4792 * error is returned.
4793 * If a layout is passed in via lypp, it is locked (exclusively locked).
4796 nfscl_layout(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen,
4797 nfsv4stateid_t *stateidp, int layouttype, int retonclose,
4798 struct nfsclflayouthead *fhlp, struct nfscllayout **lypp,
4799 struct ucred *cred, NFSPROC_T *p)
4801 struct nfsclclient *clp;
4802 struct nfscllayout *lyp, *tlyp;
4803 struct nfsclflayout *flp;
4804 struct nfsnode *np = VTONFS(vp);
4806 int layout_passed_in;
4808 mp = nmp->nm_mountp;
4809 layout_passed_in = 1;
4813 layout_passed_in = 0;
4814 tlyp = malloc(sizeof(*tlyp) + fhlen - 1, M_NFSLAYOUT,
4821 if (layout_passed_in != 0)
4822 nfsv4_unlock(&lyp->nfsly_lock, 0);
4825 free(tlyp, M_NFSLAYOUT);
4830 * Although no lyp was passed in, another thread might have
4831 * allocated one. If one is found, just increment it's ref
4832 * count and return it.
4834 lyp = nfscl_findlayout(clp, fhp, fhlen);
4838 lyp->nfsly_stateid.seqid = stateidp->seqid;
4839 lyp->nfsly_stateid.other[0] = stateidp->other[0];
4840 lyp->nfsly_stateid.other[1] = stateidp->other[1];
4841 lyp->nfsly_stateid.other[2] = stateidp->other[2];
4842 lyp->nfsly_lastbyte = 0;
4843 LIST_INIT(&lyp->nfsly_flayread);
4844 LIST_INIT(&lyp->nfsly_flayrw);
4845 LIST_INIT(&lyp->nfsly_recall);
4846 lyp->nfsly_filesid[0] = np->n_vattr.na_filesid[0];
4847 lyp->nfsly_filesid[1] = np->n_vattr.na_filesid[1];
4848 lyp->nfsly_clp = clp;
4849 if (layouttype == NFSLAYOUT_FLEXFILE)
4850 lyp->nfsly_flags = NFSLY_FLEXFILE;
4852 lyp->nfsly_flags = NFSLY_FILES;
4853 if (retonclose != 0)
4854 lyp->nfsly_flags |= NFSLY_RETONCLOSE;
4855 lyp->nfsly_fhlen = fhlen;
4856 NFSBCOPY(fhp, lyp->nfsly_fh, fhlen);
4857 TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
4858 LIST_INSERT_HEAD(NFSCLLAYOUTHASH(clp, fhp, fhlen), lyp,
4860 lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
4863 if (retonclose != 0)
4864 lyp->nfsly_flags |= NFSLY_RETONCLOSE;
4865 if (stateidp->seqid > lyp->nfsly_stateid.seqid)
4866 lyp->nfsly_stateid.seqid = stateidp->seqid;
4867 TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list);
4868 TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
4869 lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
4871 nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
4872 if (NFSCL_FORCEDISM(mp)) {
4875 free(tlyp, M_NFSLAYOUT);
4879 } else if (stateidp->seqid > lyp->nfsly_stateid.seqid)
4880 lyp->nfsly_stateid.seqid = stateidp->seqid;
4882 /* Merge the new list of File Layouts into the list. */
4883 flp = LIST_FIRST(fhlp);
4885 if (flp->nfsfl_iomode == NFSLAYOUTIOMODE_READ)
4886 nfscl_mergeflayouts(&lyp->nfsly_flayread, fhlp);
4888 nfscl_mergeflayouts(&lyp->nfsly_flayrw, fhlp);
4890 if (layout_passed_in != 0)
4891 nfsv4_unlock(&lyp->nfsly_lock, 1);
4894 free(tlyp, M_NFSLAYOUT);
4899 * Search for a layout by MDS file handle.
4900 * If one is found, it is returned with a refcnt (shared lock) iff
4901 * retflpp returned non-NULL and locked (exclusive locked) iff retflpp is
4904 struct nfscllayout *
4905 nfscl_getlayout(struct nfsclclient *clp, uint8_t *fhp, int fhlen,
4906 uint64_t off, struct nfsclflayout **retflpp, int *recalledp)
4908 struct nfscllayout *lyp;
4910 int error, igotlock;
4912 mp = clp->nfsc_nmp->nm_mountp;
4916 lyp = nfscl_findlayout(clp, fhp, fhlen);
4918 if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
4919 TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list);
4920 TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
4921 lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
4922 error = nfscl_findlayoutforio(lyp, off,
4923 NFSV4OPEN_ACCESSREAD, retflpp);
4925 nfsv4_getref(&lyp->nfsly_lock, NULL,
4926 NFSCLSTATEMUTEXPTR, mp);
4929 igotlock = nfsv4_lock(&lyp->nfsly_lock,
4930 1, NULL, NFSCLSTATEMUTEXPTR, mp);
4931 } while (igotlock == 0 && !NFSCL_FORCEDISM(mp));
4934 if (NFSCL_FORCEDISM(mp)) {
4948 * Search for a layout by MDS file handle. If one is found, mark in to be
4949 * recalled, if it already marked "return on close".
4952 nfscl_retoncloselayout(vnode_t vp, struct nfsclclient *clp, uint8_t *fhp,
4953 int fhlen, struct nfsclrecalllayout **recallpp)
4955 struct nfscllayout *lyp;
4958 if (vp->v_type != VREG || !NFSHASPNFS(VFSTONFS(vnode_mount(vp))) ||
4959 nfscl_enablecallb == 0 || nfs_numnfscbd == 0 ||
4960 (VTONFS(vp)->n_flag & NNOLAYOUT) != 0)
4962 lyp = nfscl_findlayout(clp, fhp, fhlen);
4963 if (lyp != NULL && (lyp->nfsly_flags & (NFSLY_RETONCLOSE |
4964 NFSLY_RECALL)) == NFSLY_RETONCLOSE) {
4966 if (!LIST_EMPTY(&lyp->nfsly_flayread))
4967 iomode |= NFSLAYOUTIOMODE_READ;
4968 if (!LIST_EMPTY(&lyp->nfsly_flayrw))
4969 iomode |= NFSLAYOUTIOMODE_RW;
4970 (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE, lyp, iomode,
4971 0, UINT64_MAX, lyp->nfsly_stateid.seqid, 0, 0, NULL,
4973 NFSCL_DEBUG(4, "retoncls recall iomode=%d\n", iomode);
4979 * Mark the layout to be recalled and with an error.
4980 * Also, disable the dsp from further use.
4983 nfscl_dserr(uint32_t op, uint32_t stat, struct nfscldevinfo *dp,
4984 struct nfscllayout *lyp, struct nfsclds *dsp)
4986 struct nfsclrecalllayout *recallp;
4989 printf("DS being disabled, error=%d\n", stat);
4990 /* Set up the return of the layout. */
4991 recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_WAITOK);
4994 if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
4995 if (!LIST_EMPTY(&lyp->nfsly_flayread))
4996 iomode |= NFSLAYOUTIOMODE_READ;
4997 if (!LIST_EMPTY(&lyp->nfsly_flayrw))
4998 iomode |= NFSLAYOUTIOMODE_RW;
4999 (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE, lyp, iomode,
5000 0, UINT64_MAX, lyp->nfsly_stateid.seqid, stat, op,
5001 dp->nfsdi_deviceid, recallp);
5003 NFSCL_DEBUG(4, "nfscl_dserr recall iomode=%d\n", iomode);
5006 free(recallp, M_NFSLAYRECALL);
5009 /* And shut the TCP connection down. */
5010 nfscl_cancelreqs(dsp);
5014 * Cancel all RPCs for this "dsp" by closing the connection.
5015 * Also, mark the session as defunct.
5016 * If NFSCLDS_SAMECONN is set, the connection is shared with other DSs and
5017 * cannot be shut down.
5020 nfscl_cancelreqs(struct nfsclds *dsp)
5022 struct __rpc_client *cl;
5023 static int non_event;
5026 if ((dsp->nfsclds_flags & (NFSCLDS_CLOSED | NFSCLDS_SAMECONN)) == 0 &&
5027 dsp->nfsclds_sockp != NULL &&
5028 dsp->nfsclds_sockp->nr_client != NULL) {
5029 dsp->nfsclds_flags |= NFSCLDS_CLOSED;
5030 cl = dsp->nfsclds_sockp->nr_client;
5031 dsp->nfsclds_sess.nfsess_defunct = 1;
5035 * This 1sec sleep is done to reduce the number of reconnect
5036 * attempts made on the DS while it has failed.
5038 tsleep(&non_event, PVFS, "ndscls", hz);
5045 * Dereference a layout.
5048 nfscl_rellayout(struct nfscllayout *lyp, int exclocked)
5053 nfsv4_unlock(&lyp->nfsly_lock, 0);
5055 nfsv4_relref(&lyp->nfsly_lock);
5060 * Search for a devinfo by deviceid. If one is found, return it after
5061 * acquiring a reference count on it.
5063 struct nfscldevinfo *
5064 nfscl_getdevinfo(struct nfsclclient *clp, uint8_t *deviceid,
5065 struct nfscldevinfo *dip)
5070 dip = nfscl_finddevinfo(clp, deviceid);
5072 dip->nfsdi_refcnt++;
5078 * Dereference a devinfo structure.
5081 nfscl_reldevinfo_locked(struct nfscldevinfo *dip)
5084 dip->nfsdi_refcnt--;
5085 if (dip->nfsdi_refcnt == 0)
5086 wakeup(&dip->nfsdi_refcnt);
5090 * Dereference a devinfo structure.
5093 nfscl_reldevinfo(struct nfscldevinfo *dip)
5097 nfscl_reldevinfo_locked(dip);
5102 * Find a layout for this file handle. Return NULL upon failure.
5104 static struct nfscllayout *
5105 nfscl_findlayout(struct nfsclclient *clp, u_int8_t *fhp, int fhlen)
5107 struct nfscllayout *lyp;
5109 LIST_FOREACH(lyp, NFSCLLAYOUTHASH(clp, fhp, fhlen), nfsly_hash)
5110 if (lyp->nfsly_fhlen == fhlen &&
5111 !NFSBCMP(lyp->nfsly_fh, fhp, fhlen))
5117 * Find a devinfo for this deviceid. Return NULL upon failure.
5119 static struct nfscldevinfo *
5120 nfscl_finddevinfo(struct nfsclclient *clp, uint8_t *deviceid)
5122 struct nfscldevinfo *dip;
5124 LIST_FOREACH(dip, &clp->nfsc_devinfo, nfsdi_list)
5125 if (NFSBCMP(dip->nfsdi_deviceid, deviceid, NFSX_V4DEVICEID)
5132 * Merge the new file layout list into the main one, maintaining it in
5133 * increasing offset order.
5136 nfscl_mergeflayouts(struct nfsclflayouthead *fhlp,
5137 struct nfsclflayouthead *newfhlp)
5139 struct nfsclflayout *flp, *nflp, *prevflp, *tflp;
5141 flp = LIST_FIRST(fhlp);
5143 LIST_FOREACH_SAFE(nflp, newfhlp, nfsfl_list, tflp) {
5144 while (flp != NULL && flp->nfsfl_off < nflp->nfsfl_off) {
5146 flp = LIST_NEXT(flp, nfsfl_list);
5148 if (prevflp == NULL)
5149 LIST_INSERT_HEAD(fhlp, nflp, nfsfl_list);
5151 LIST_INSERT_AFTER(prevflp, nflp, nfsfl_list);
5157 * Add this nfscldevinfo to the client, if it doesn't already exist.
5158 * This function consumes the structure pointed at by dip, if not NULL.
5161 nfscl_adddevinfo(struct nfsmount *nmp, struct nfscldevinfo *dip, int ind,
5162 struct nfsclflayout *flp)
5164 struct nfsclclient *clp;
5165 struct nfscldevinfo *tdip;
5173 free(dip, M_NFSDEVINFO);
5176 if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5177 dev = flp->nfsfl_dev;
5179 dev = flp->nfsfl_ffm[ind].dev;
5180 tdip = nfscl_finddevinfo(clp, dev);
5182 tdip->nfsdi_layoutrefs++;
5183 if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5184 flp->nfsfl_devp = tdip;
5186 flp->nfsfl_ffm[ind].devp = tdip;
5187 nfscl_reldevinfo_locked(tdip);
5190 free(dip, M_NFSDEVINFO);
5194 LIST_INSERT_HEAD(&clp->nfsc_devinfo, dip, nfsdi_list);
5195 dip->nfsdi_layoutrefs = 1;
5196 if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5197 flp->nfsfl_devp = dip;
5199 flp->nfsfl_ffm[ind].devp = dip;
5208 * Free up a layout structure and associated file layout structure(s).
5211 nfscl_freelayout(struct nfscllayout *layp)
5213 struct nfsclflayout *flp, *nflp;
5214 struct nfsclrecalllayout *rp, *nrp;
5216 LIST_FOREACH_SAFE(flp, &layp->nfsly_flayread, nfsfl_list, nflp) {
5217 LIST_REMOVE(flp, nfsfl_list);
5218 nfscl_freeflayout(flp);
5220 LIST_FOREACH_SAFE(flp, &layp->nfsly_flayrw, nfsfl_list, nflp) {
5221 LIST_REMOVE(flp, nfsfl_list);
5222 nfscl_freeflayout(flp);
5224 LIST_FOREACH_SAFE(rp, &layp->nfsly_recall, nfsrecly_list, nrp) {
5225 LIST_REMOVE(rp, nfsrecly_list);
5226 free(rp, M_NFSLAYRECALL);
5229 free(layp, M_NFSLAYOUT);
5233 * Free up a file layout structure.
5236 nfscl_freeflayout(struct nfsclflayout *flp)
5240 if ((flp->nfsfl_flags & NFSFL_FILE) != 0) {
5241 for (i = 0; i < flp->nfsfl_fhcnt; i++)
5242 free(flp->nfsfl_fh[i], M_NFSFH);
5243 if (flp->nfsfl_devp != NULL)
5244 flp->nfsfl_devp->nfsdi_layoutrefs--;
5246 if ((flp->nfsfl_flags & NFSFL_FLEXFILE) != 0)
5247 for (i = 0; i < flp->nfsfl_mirrorcnt; i++) {
5248 for (j = 0; j < flp->nfsfl_ffm[i].fhcnt; j++)
5249 free(flp->nfsfl_ffm[i].fh[j], M_NFSFH);
5250 if (flp->nfsfl_ffm[i].devp != NULL)
5251 flp->nfsfl_ffm[i].devp->nfsdi_layoutrefs--;
5253 free(flp, M_NFSFLAYOUT);
5257 * Free up a file layout devinfo structure.
5260 nfscl_freedevinfo(struct nfscldevinfo *dip)
5263 free(dip, M_NFSDEVINFO);
5267 * Mark any layouts that match as recalled.
5270 nfscl_layoutrecall(int recalltype, struct nfscllayout *lyp, uint32_t iomode,
5271 uint64_t off, uint64_t len, uint32_t stateseqid, uint32_t stat, uint32_t op,
5272 char *devid, struct nfsclrecalllayout *recallp)
5274 struct nfsclrecalllayout *rp, *orp;
5276 recallp->nfsrecly_recalltype = recalltype;
5277 recallp->nfsrecly_iomode = iomode;
5278 recallp->nfsrecly_stateseqid = stateseqid;
5279 recallp->nfsrecly_off = off;
5280 recallp->nfsrecly_len = len;
5281 recallp->nfsrecly_stat = stat;
5282 recallp->nfsrecly_op = op;
5284 NFSBCOPY(devid, recallp->nfsrecly_devid, NFSX_V4DEVICEID);
5286 * Order the list as file returns first, followed by fsid and any
5287 * returns, both in increasing stateseqid order.
5288 * Note that the seqids wrap around, so 1 is after 0xffffffff.
5289 * (I'm not sure this is correct because I find RFC5661 confusing
5290 * on this, but hopefully it will work ok.)
5293 LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) {
5295 if ((recalltype == NFSLAYOUTRETURN_FILE &&
5296 (rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE ||
5297 nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) ||
5298 (recalltype != NFSLAYOUTRETURN_FILE &&
5299 rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE &&
5300 nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) {
5301 LIST_INSERT_BEFORE(rp, recallp, nfsrecly_list);
5306 * Put any error return on all the file returns that will
5309 if (rp->nfsrecly_recalltype == NFSLAYOUTRETURN_FILE &&
5310 stat != 0 && rp->nfsrecly_stat == 0) {
5311 rp->nfsrecly_stat = stat;
5312 rp->nfsrecly_op = op;
5314 NFSBCOPY(devid, rp->nfsrecly_devid,
5320 LIST_INSERT_HEAD(&lyp->nfsly_recall, recallp,
5323 LIST_INSERT_AFTER(orp, recallp, nfsrecly_list);
5325 lyp->nfsly_flags |= NFSLY_RECALL;
5326 wakeup(lyp->nfsly_clp);
5331 * Compare the two seqids for ordering. The trick is that the seqids can
5332 * wrap around from 0xffffffff->0, so check for the cases where one
5333 * has wrapped around.
5334 * Return 1 if seqid1 comes before seqid2, 0 otherwise.
5337 nfscl_seq(uint32_t seqid1, uint32_t seqid2)
5340 if (seqid2 > seqid1 && (seqid2 - seqid1) >= 0x7fffffff)
5341 /* seqid2 has wrapped around. */
5343 if (seqid1 > seqid2 && (seqid1 - seqid2) >= 0x7fffffff)
5344 /* seqid1 has wrapped around. */
5346 if (seqid1 <= seqid2)
5352 * Do a layout return for each of the recalls.
5355 nfscl_layoutreturn(struct nfsmount *nmp, struct nfscllayout *lyp,
5356 struct ucred *cred, NFSPROC_T *p)
5358 struct nfsclrecalllayout *rp;
5359 nfsv4stateid_t stateid;
5362 NFSBCOPY(lyp->nfsly_stateid.other, stateid.other, NFSX_STATEIDOTHER);
5363 stateid.seqid = lyp->nfsly_stateid.seqid;
5364 if ((lyp->nfsly_flags & NFSLY_FILES) != 0)
5365 layouttype = NFSLAYOUT_NFSV4_1_FILES;
5367 layouttype = NFSLAYOUT_FLEXFILE;
5368 LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) {
5369 (void)nfsrpc_layoutreturn(nmp, lyp->nfsly_fh,
5370 lyp->nfsly_fhlen, 0, layouttype,
5371 rp->nfsrecly_iomode, rp->nfsrecly_recalltype,
5372 rp->nfsrecly_off, rp->nfsrecly_len,
5373 &stateid, cred, p, rp->nfsrecly_stat, rp->nfsrecly_op,
5374 rp->nfsrecly_devid);
5379 * Do the layout commit for a file layout.
5382 nfscl_dolayoutcommit(struct nfsmount *nmp, struct nfscllayout *lyp,
5383 struct ucred *cred, NFSPROC_T *p)
5385 struct nfsclflayout *flp;
5387 int error, layouttype;
5389 if ((lyp->nfsly_flags & NFSLY_FILES) != 0)
5390 layouttype = NFSLAYOUT_NFSV4_1_FILES;
5392 layouttype = NFSLAYOUT_FLEXFILE;
5393 LIST_FOREACH(flp, &lyp->nfsly_flayrw, nfsfl_list) {
5394 if (layouttype == NFSLAYOUT_FLEXFILE &&
5395 (flp->nfsfl_fflags & NFSFLEXFLAG_NO_LAYOUTCOMMIT) != 0) {
5396 NFSCL_DEBUG(4, "Flex file: no layoutcommit\n");
5397 /* If not supported, don't bother doing it. */
5399 nmp->nm_state |= NFSSTA_NOLAYOUTCOMMIT;
5402 } else if (flp->nfsfl_off <= lyp->nfsly_lastbyte) {
5403 len = flp->nfsfl_end - flp->nfsfl_off;
5404 error = nfsrpc_layoutcommit(nmp, lyp->nfsly_fh,
5405 lyp->nfsly_fhlen, 0, flp->nfsfl_off, len,
5406 lyp->nfsly_lastbyte, &lyp->nfsly_stateid,
5407 layouttype, cred, p, NULL);
5408 NFSCL_DEBUG(4, "layoutcommit err=%d\n", error);
5409 if (error == NFSERR_NOTSUPP) {
5410 /* If not supported, don't bother doing it. */
5412 nmp->nm_state |= NFSSTA_NOLAYOUTCOMMIT;
5421 * Commit all layouts for a file (vnode).
5424 nfscl_layoutcommit(vnode_t vp, NFSPROC_T *p)
5426 struct nfsclclient *clp;
5427 struct nfscllayout *lyp;
5428 struct nfsnode *np = VTONFS(vp);
5430 struct nfsmount *nmp;
5432 mp = vnode_mount(vp);
5434 if (NFSHASNOLAYOUTCOMMIT(nmp))
5442 lyp = nfscl_findlayout(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
5447 nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
5448 if (NFSCL_FORCEDISM(mp)) {
5453 if ((lyp->nfsly_flags & NFSLY_WRITTEN) != 0) {
5454 lyp->nfsly_flags &= ~NFSLY_WRITTEN;
5456 NFSCL_DEBUG(4, "do layoutcommit2\n");
5457 nfscl_dolayoutcommit(clp->nfsc_nmp, lyp, NFSPROCCRED(p), p);
5461 nfsv4_relref(&lyp->nfsly_lock);