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.)
84 #include <fs/nfs/nfsport.h>
89 extern struct nfsstatsv1 nfsstatsv1;
90 extern struct nfsreqhead nfsd_reqq;
91 extern u_int32_t newnfs_false, newnfs_true;
92 extern int nfscl_debuglevel;
93 extern int nfscl_enablecallb;
94 extern int nfs_numnfscbd;
98 struct nfsclhead nfsclhead; /* Head of clientid list */
99 int nfscl_deleghighwater = NFSCLDELEGHIGHWATER;
100 int nfscl_layouthighwater = NFSCLLAYOUTHIGHWATER;
101 #endif /* !APPLEKEXT */
103 static int nfscl_delegcnt = 0;
104 static int nfscl_layoutcnt = 0;
105 static int nfscl_getopen(struct nfsclownerhead *, u_int8_t *, int, u_int8_t *,
106 u_int8_t *, u_int32_t, struct nfscllockowner **, struct nfsclopen **);
107 static void nfscl_clrelease(struct nfsclclient *);
108 static void nfscl_cleanclient(struct nfsclclient *);
109 static void nfscl_expireclient(struct nfsclclient *, struct nfsmount *,
110 struct ucred *, NFSPROC_T *);
111 static int nfscl_expireopen(struct nfsclclient *, struct nfsclopen *,
112 struct nfsmount *, struct ucred *, NFSPROC_T *);
113 static void nfscl_recover(struct nfsclclient *, struct ucred *, NFSPROC_T *);
114 static void nfscl_insertlock(struct nfscllockowner *, struct nfscllock *,
115 struct nfscllock *, int);
116 static int nfscl_updatelock(struct nfscllockowner *, struct nfscllock **,
117 struct nfscllock **, int);
118 static void nfscl_delegreturnall(struct nfsclclient *, NFSPROC_T *);
119 static u_int32_t nfscl_nextcbident(void);
120 static mount_t nfscl_getmnt(int, uint8_t *, u_int32_t, struct nfsclclient **);
121 static struct nfsclclient *nfscl_getclnt(u_int32_t);
122 static struct nfsclclient *nfscl_getclntsess(uint8_t *);
123 static struct nfscldeleg *nfscl_finddeleg(struct nfsclclient *, u_int8_t *,
125 static void nfscl_retoncloselayout(vnode_t, struct nfsclclient *, uint8_t *,
126 int, struct nfsclrecalllayout **);
127 static void nfscl_reldevinfo_locked(struct nfscldevinfo *);
128 static struct nfscllayout *nfscl_findlayout(struct nfsclclient *, u_int8_t *,
130 static struct nfscldevinfo *nfscl_finddevinfo(struct nfsclclient *, uint8_t *);
131 static int nfscl_checkconflict(struct nfscllockownerhead *, struct nfscllock *,
132 u_int8_t *, struct nfscllock **);
133 static void nfscl_freealllocks(struct nfscllockownerhead *, int);
134 static int nfscl_localconflict(struct nfsclclient *, u_int8_t *, int,
135 struct nfscllock *, u_int8_t *, struct nfscldeleg *, struct nfscllock **);
136 static void nfscl_newopen(struct nfsclclient *, struct nfscldeleg *,
137 struct nfsclowner **, struct nfsclowner **, struct nfsclopen **,
138 struct nfsclopen **, u_int8_t *, u_int8_t *, int, struct ucred *, int *);
139 static int nfscl_moveopen(vnode_t , struct nfsclclient *,
140 struct nfsmount *, struct nfsclopen *, struct nfsclowner *,
141 struct nfscldeleg *, struct ucred *, NFSPROC_T *);
142 static void nfscl_totalrecall(struct nfsclclient *);
143 static int nfscl_relock(vnode_t , struct nfsclclient *, struct nfsmount *,
144 struct nfscllockowner *, struct nfscllock *, struct ucred *, NFSPROC_T *);
145 static int nfscl_tryopen(struct nfsmount *, vnode_t , u_int8_t *, int,
146 u_int8_t *, int, u_int32_t, struct nfsclopen *, u_int8_t *, int,
147 struct nfscldeleg **, int, u_int32_t, struct ucred *, NFSPROC_T *);
148 static int nfscl_trylock(struct nfsmount *, vnode_t , u_int8_t *,
149 int, struct nfscllockowner *, int, int, u_int64_t, u_int64_t, short,
150 struct ucred *, NFSPROC_T *);
151 static int nfsrpc_reopen(struct nfsmount *, u_int8_t *, int, u_int32_t,
152 struct nfsclopen *, struct nfscldeleg **, struct ucred *, NFSPROC_T *);
153 static void nfscl_freedeleg(struct nfscldeleghead *, struct nfscldeleg *);
154 static int nfscl_errmap(struct nfsrv_descript *, u_int32_t);
155 static void nfscl_cleanup_common(struct nfsclclient *, u_int8_t *);
156 static int nfscl_recalldeleg(struct nfsclclient *, struct nfsmount *,
157 struct nfscldeleg *, vnode_t, struct ucred *, NFSPROC_T *, int);
158 static void nfscl_freeopenowner(struct nfsclowner *, int);
159 static void nfscl_cleandeleg(struct nfscldeleg *);
160 static int nfscl_trydelegreturn(struct nfscldeleg *, struct ucred *,
161 struct nfsmount *, NFSPROC_T *);
162 static void nfscl_emptylockowner(struct nfscllockowner *,
163 struct nfscllockownerfhhead *);
164 static void nfscl_mergeflayouts(struct nfsclflayouthead *,
165 struct nfsclflayouthead *);
166 static int nfscl_layoutrecall(int, struct nfscllayout *, uint32_t, uint64_t,
167 uint64_t, uint32_t, uint32_t, uint32_t, char *, struct nfsclrecalllayout *);
168 static int nfscl_seq(uint32_t, uint32_t);
169 static void nfscl_layoutreturn(struct nfsmount *, struct nfscllayout *,
170 struct ucred *, NFSPROC_T *);
171 static void nfscl_dolayoutcommit(struct nfsmount *, struct nfscllayout *,
172 struct ucred *, NFSPROC_T *);
174 static short nfscberr_null[] = {
179 static short nfscberr_getattr[] = {
188 static short nfscberr_recall[] = {
198 static short *nfscl_cberrmap[] = {
206 #define NETFAMILY(clp) \
207 (((clp)->nfsc_flags & NFSCLFLAGS_AFINET6) ? AF_INET6 : AF_INET)
210 * Called for an open operation.
211 * If the nfhp argument is NULL, just get an openowner.
214 nfscl_open(vnode_t vp, u_int8_t *nfhp, int fhlen, u_int32_t amode, int usedeleg,
215 struct ucred *cred, NFSPROC_T *p, struct nfsclowner **owpp,
216 struct nfsclopen **opp, int *newonep, int *retp, int lockit)
218 struct nfsclclient *clp;
219 struct nfsclowner *owp, *nowp;
220 struct nfsclopen *op = NULL, *nop = NULL;
221 struct nfscldeleg *dp;
222 struct nfsclownerhead *ohp;
223 u_int8_t own[NFSV4CL_LOCKNAMELEN];
234 * Might need one or both of these, so MALLOC them now, to
235 * avoid a tsleep() in MALLOC later.
237 nowp = malloc(sizeof (struct nfsclowner),
238 M_NFSCLOWNER, M_WAITOK);
240 nop = malloc(sizeof (struct nfsclopen) +
241 fhlen - 1, M_NFSCLOPEN, M_WAITOK);
242 ret = nfscl_getcl(vnode_mount(vp), cred, p, 1, &clp);
244 free(nowp, M_NFSCLOWNER);
246 free(nop, M_NFSCLOPEN);
251 * Get the Open iff it already exists.
252 * If none found, add the new one or return error, depending upon
257 /* First check the delegation list */
258 if (nfhp != NULL && usedeleg) {
259 LIST_FOREACH(dp, NFSCLDELEGHASH(clp, nfhp, fhlen), nfsdl_hash) {
260 if (dp->nfsdl_fhlen == fhlen &&
261 !NFSBCMP(nfhp, dp->nfsdl_fh, fhlen)) {
262 if (!(amode & NFSV4OPEN_ACCESSWRITE) ||
263 (dp->nfsdl_flags & NFSCLDL_WRITE))
272 nfscl_filllockowner(p->td_proc, own, F_POSIX);
273 ohp = &dp->nfsdl_owner;
275 /* For NFSv4.1 and this option, use a single open_owner. */
276 if (NFSHASONEOPENOWN(VFSTONFS(vnode_mount(vp))))
277 nfscl_filllockowner(NULL, own, F_POSIX);
279 nfscl_filllockowner(p->td_proc, own, F_POSIX);
280 ohp = &clp->nfsc_owner;
282 /* Now, search for an openowner */
283 LIST_FOREACH(owp, ohp, nfsow_list) {
284 if (!NFSBCMP(owp->nfsow_owner, own, NFSV4CL_LOCKNAMELEN))
289 * Create a new open, as required.
291 nfscl_newopen(clp, dp, &owp, &nowp, &op, &nop, own, nfhp, fhlen,
295 * Now, check the mode on the open and return the appropriate
299 if (nfhp != NULL && dp != NULL && nop == NULL)
300 /* new local open on delegation */
301 *retp = NFSCLOPEN_SETCRED;
303 *retp = NFSCLOPEN_OK;
305 if (op != NULL && (amode & ~(op->nfso_mode))) {
306 op->nfso_mode |= amode;
307 if (retp != NULL && dp == NULL)
308 *retp = NFSCLOPEN_DOOPEN;
312 * Serialize modifications to the open owner for multiple threads
313 * within the same process using a read/write sleep lock.
314 * For NFSv4.1 and a single OpenOwner, allow concurrent open operations
315 * by acquiring a shared lock. The close operations still use an
316 * exclusive lock for this case.
319 if (NFSHASONEOPENOWN(VFSTONFS(vnode_mount(vp)))) {
321 * Get a shared lock on the OpenOwner, but first
322 * wait for any pending exclusive lock, so that the
323 * exclusive locker gets priority.
325 nfsv4_lock(&owp->nfsow_rwlock, 0, NULL,
326 NFSCLSTATEMUTEXPTR, NULL);
327 nfsv4_getref(&owp->nfsow_rwlock, NULL,
328 NFSCLSTATEMUTEXPTR, NULL);
330 nfscl_lockexcl(&owp->nfsow_rwlock, NFSCLSTATEMUTEXPTR);
334 free(nowp, M_NFSCLOWNER);
336 free(nop, M_NFSCLOPEN);
345 * Create a new open, as required.
348 nfscl_newopen(struct nfsclclient *clp, struct nfscldeleg *dp,
349 struct nfsclowner **owpp, struct nfsclowner **nowpp, struct nfsclopen **opp,
350 struct nfsclopen **nopp, u_int8_t *own, u_int8_t *fhp, int fhlen,
351 struct ucred *cred, int *newonep)
353 struct nfsclowner *owp = *owpp, *nowp;
354 struct nfsclopen *op, *nop;
364 if (owp == NULL && nowp != NULL) {
365 NFSBCOPY(own, nowp->nfsow_owner, NFSV4CL_LOCKNAMELEN);
366 LIST_INIT(&nowp->nfsow_open);
367 nowp->nfsow_clp = clp;
368 nowp->nfsow_seqid = 0;
369 nowp->nfsow_defunct = 0;
370 nfscl_lockinit(&nowp->nfsow_rwlock);
372 nfsstatsv1.cllocalopenowners++;
373 LIST_INSERT_HEAD(&dp->nfsdl_owner, nowp, nfsow_list);
375 nfsstatsv1.clopenowners++;
376 LIST_INSERT_HEAD(&clp->nfsc_owner, nowp, nfsow_list);
384 /* If an fhp has been specified, create an Open as well. */
386 /* and look for the correct open, based upon FH */
387 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
388 if (op->nfso_fhlen == fhlen &&
389 !NFSBCMP(op->nfso_fh, fhp, fhlen))
392 if (op == NULL && nop != NULL) {
395 nop->nfso_opencnt = 0;
396 nop->nfso_posixlock = 1;
397 nop->nfso_fhlen = fhlen;
398 NFSBCOPY(fhp, nop->nfso_fh, fhlen);
399 LIST_INIT(&nop->nfso_lock);
400 nop->nfso_stateid.seqid = 0;
401 nop->nfso_stateid.other[0] = 0;
402 nop->nfso_stateid.other[1] = 0;
403 nop->nfso_stateid.other[2] = 0;
404 KASSERT(cred != NULL, ("%s: cred NULL\n", __func__));
405 newnfs_copyincred(cred, &nop->nfso_cred);
407 TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
408 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp,
410 dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
411 nfsstatsv1.cllocalopens++;
413 nfsstatsv1.clopens++;
415 LIST_INSERT_HEAD(&owp->nfsow_open, nop, nfso_list);
427 * Called to find/add a delegation to a client.
430 nfscl_deleg(mount_t mp, struct nfsclclient *clp, u_int8_t *nfhp,
431 int fhlen, struct ucred *cred, NFSPROC_T *p, struct nfscldeleg **dpp)
433 struct nfscldeleg *dp = *dpp, *tdp;
436 * First, if we have received a Read delegation for a file on a
437 * read/write file system, just return it, because they aren't
440 if (mp != NULL && dp != NULL && !NFSMNT_RDONLY(mp) &&
441 (dp->nfsdl_flags & NFSCLDL_READ)) {
442 (void) nfscl_trydelegreturn(dp, cred, VFSTONFS(mp), p);
443 free(dp, M_NFSCLDELEG);
448 /* Look for the correct deleg, based upon FH */
450 tdp = nfscl_finddeleg(clp, nfhp, fhlen);
454 return (NFSERR_BADSTATEID);
457 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list);
458 LIST_INSERT_HEAD(NFSCLDELEGHASH(clp, nfhp, fhlen), dp,
460 dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
461 nfsstatsv1.cldelegates++;
465 * Delegation already exists, what do we do if a new one??
468 printf("Deleg already exists!\n");
469 free(dp, M_NFSCLDELEG);
480 * Find a delegation for this file handle. Return NULL upon failure.
482 static struct nfscldeleg *
483 nfscl_finddeleg(struct nfsclclient *clp, u_int8_t *fhp, int fhlen)
485 struct nfscldeleg *dp;
487 LIST_FOREACH(dp, NFSCLDELEGHASH(clp, fhp, fhlen), nfsdl_hash) {
488 if (dp->nfsdl_fhlen == fhlen &&
489 !NFSBCMP(dp->nfsdl_fh, fhp, fhlen))
496 * Get a stateid for an I/O operation. First, look for an open and iff
497 * found, return either a lockowner stateid or the open stateid.
498 * If no Open is found, just return error and the special stateid of all zeros.
501 nfscl_getstateid(vnode_t vp, u_int8_t *nfhp, int fhlen, u_int32_t mode,
502 int fords, struct ucred *cred, NFSPROC_T *p, nfsv4stateid_t *stateidp,
505 struct nfsclclient *clp;
506 struct nfsclowner *owp;
507 struct nfsclopen *op = NULL, *top;
508 struct nfscllockowner *lp;
509 struct nfscldeleg *dp;
511 struct nfsmount *nmp;
512 u_int8_t own[NFSV4CL_LOCKNAMELEN];
517 * Initially, just set the special stateid of all zeros.
518 * (Don't do this for a DS, since the special stateid can't be used.)
522 stateidp->other[0] = 0;
523 stateidp->other[1] = 0;
524 stateidp->other[2] = 0;
526 if (vnode_vtype(vp) != VREG)
529 nmp = VFSTONFS(vnode_mount(vp));
531 clp = nfscl_findcl(nmp);
538 * Wait for recovery to complete.
540 while ((clp->nfsc_flags & NFSCLFLAGS_RECVRINPROG))
541 (void) nfsmsleep(&clp->nfsc_flags, NFSCLSTATEMUTEXPTR,
542 PZERO, "nfsrecvr", NULL);
545 * First, look for a delegation.
547 LIST_FOREACH(dp, NFSCLDELEGHASH(clp, nfhp, fhlen), nfsdl_hash) {
548 if (dp->nfsdl_fhlen == fhlen &&
549 !NFSBCMP(nfhp, dp->nfsdl_fh, fhlen)) {
550 if (!(mode & NFSV4OPEN_ACCESSWRITE) ||
551 (dp->nfsdl_flags & NFSCLDL_WRITE)) {
552 stateidp->seqid = dp->nfsdl_stateid.seqid;
553 stateidp->other[0] = dp->nfsdl_stateid.other[0];
554 stateidp->other[1] = dp->nfsdl_stateid.other[1];
555 stateidp->other[2] = dp->nfsdl_stateid.other[2];
556 if (!(np->n_flag & NDELEGRECALL)) {
557 TAILQ_REMOVE(&clp->nfsc_deleg, dp,
559 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp,
561 dp->nfsdl_timestamp = NFSD_MONOSEC +
563 dp->nfsdl_rwlock.nfslock_usecnt++;
564 *lckpp = (void *)&dp->nfsdl_rwlock;
575 * If p != NULL, we want to search the parentage tree
576 * for a matching OpenOwner and use that.
578 if (NFSHASONEOPENOWN(VFSTONFS(vnode_mount(vp))))
579 nfscl_filllockowner(NULL, own, F_POSIX);
581 nfscl_filllockowner(p->td_proc, own, F_POSIX);
583 error = nfscl_getopen(&clp->nfsc_owner, nfhp, fhlen, own, own,
585 if (error == 0 && lp != NULL && fords == 0) {
586 /* Don't return a lock stateid for a DS. */
588 lp->nfsl_stateid.seqid;
590 lp->nfsl_stateid.other[0];
592 lp->nfsl_stateid.other[1];
594 lp->nfsl_stateid.other[2];
600 /* If not found, just look for any OpenOwner that will work. */
603 owp = LIST_FIRST(&clp->nfsc_owner);
604 while (!done && owp != NULL) {
605 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
606 if (op->nfso_fhlen == fhlen &&
607 !NFSBCMP(op->nfso_fh, nfhp, fhlen)) {
608 if (top == NULL && (op->nfso_mode &
609 NFSV4OPEN_ACCESSWRITE) != 0 &&
610 (mode & NFSV4OPEN_ACCESSREAD) != 0)
612 if ((mode & op->nfso_mode) == mode) {
619 owp = LIST_NEXT(owp, nfsow_list);
622 NFSCL_DEBUG(2, "openmode top=%p\n", top);
623 if (top == NULL || NFSHASOPENMODE(nmp)) {
630 * For read aheads or write behinds, use the open cred.
631 * A read ahead or write behind is indicated by p == NULL.
634 newnfs_copycred(&op->nfso_cred, cred);
638 * No lock stateid, so return the open stateid.
640 stateidp->seqid = op->nfso_stateid.seqid;
641 stateidp->other[0] = op->nfso_stateid.other[0];
642 stateidp->other[1] = op->nfso_stateid.other[1];
643 stateidp->other[2] = op->nfso_stateid.other[2];
649 * Search for a matching file, mode and, optionally, lockowner.
652 nfscl_getopen(struct nfsclownerhead *ohp, u_int8_t *nfhp, int fhlen,
653 u_int8_t *openown, u_int8_t *lockown, u_int32_t mode,
654 struct nfscllockowner **lpp, struct nfsclopen **opp)
656 struct nfsclowner *owp;
657 struct nfsclopen *op, *rop, *rop2;
658 struct nfscllockowner *lp;
664 * rop will be set to the open to be returned. There are three
665 * variants of this, all for an open of the correct file:
666 * 1 - A match of lockown.
667 * 2 - A match of the openown, when no lockown match exists.
668 * 3 - A match for any open, if no openown or lockown match exists.
669 * Looking for #2 over #3 probably isn't necessary, but since
670 * RFC3530 is vague w.r.t. the relationship between openowners and
671 * lockowners, I think this is the safer way to go.
676 /* Search the client list */
677 owp = LIST_FIRST(ohp);
678 while (owp != NULL && keep_looping != 0) {
679 /* and look for the correct open */
680 op = LIST_FIRST(&owp->nfsow_open);
681 while (op != NULL && keep_looping != 0) {
682 if (op->nfso_fhlen == fhlen &&
683 !NFSBCMP(op->nfso_fh, nfhp, fhlen)
684 && (op->nfso_mode & mode) == mode) {
686 /* Now look for a matching lockowner. */
687 LIST_FOREACH(lp, &op->nfso_lock,
689 if (!NFSBCMP(lp->nfsl_owner,
691 NFSV4CL_LOCKNAMELEN)) {
699 if (rop == NULL && !NFSBCMP(owp->nfsow_owner,
700 openown, NFSV4CL_LOCKNAMELEN)) {
708 op = LIST_NEXT(op, nfso_list);
710 owp = LIST_NEXT(owp, nfsow_list);
721 * Release use of an open owner. Called when open operations are done
722 * with the open owner.
725 nfscl_ownerrelease(struct nfsmount *nmp, struct nfsclowner *owp,
726 __unused int error, __unused int candelete, int unlocked)
733 if (NFSHASONEOPENOWN(nmp))
734 nfsv4_relref(&owp->nfsow_rwlock);
736 nfscl_lockunlock(&owp->nfsow_rwlock);
738 nfscl_clrelease(owp->nfsow_clp);
743 * Release use of an open structure under an open owner.
746 nfscl_openrelease(struct nfsmount *nmp, struct nfsclopen *op, int error,
749 struct nfsclclient *clp;
750 struct nfsclowner *owp;
756 if (NFSHASONEOPENOWN(nmp))
757 nfsv4_relref(&owp->nfsow_rwlock);
759 nfscl_lockunlock(&owp->nfsow_rwlock);
760 clp = owp->nfsow_clp;
761 if (error && candelete && op->nfso_opencnt == 0)
762 nfscl_freeopen(op, 0);
763 nfscl_clrelease(clp);
768 * Called to get a clientid structure. It will optionally lock the
769 * client data structures to do the SetClientId/SetClientId_confirm,
770 * but will release that lock and return the clientid with a reference
772 * If the "cred" argument is NULL, a new clientid should not be created.
773 * If the "p" argument is NULL, a SetClientID/SetClientIDConfirm cannot
775 * The start_renewthread argument tells nfscl_getcl() to start a renew
776 * thread if this creates a new clp.
777 * It always clpp with a reference count on it, unless returning an error.
780 nfscl_getcl(struct mount *mp, struct ucred *cred, NFSPROC_T *p,
781 int start_renewthread, struct nfsclclient **clpp)
783 struct nfsclclient *clp;
784 struct nfsclclient *newclp = NULL;
785 struct nfsmount *nmp;
786 char uuid[HOSTUUIDLEN];
787 int igotlock = 0, error, trystalecnt, clidinusedelay, i;
792 getcredhostuuid(cred, uuid, sizeof uuid);
793 idlen = strlen(uuid);
795 idlen += sizeof (u_int64_t);
797 idlen += sizeof (u_int64_t) + 16; /* 16 random bytes */
799 sizeof (struct nfsclclient) + idlen - 1, M_NFSCLCLIENT,
804 * If a forced dismount is already in progress, don't
805 * allocate a new clientid and get out now. For the case where
806 * clp != NULL, this is a harmless optimization.
808 if (NFSCL_FORCEDISM(mp)) {
811 free(newclp, M_NFSCLCLIENT);
816 if (newclp == NULL) {
821 clp->nfsc_idlen = idlen;
822 LIST_INIT(&clp->nfsc_owner);
823 TAILQ_INIT(&clp->nfsc_deleg);
824 TAILQ_INIT(&clp->nfsc_layout);
825 LIST_INIT(&clp->nfsc_devinfo);
826 for (i = 0; i < NFSCLDELEGHASHSIZE; i++)
827 LIST_INIT(&clp->nfsc_deleghash[i]);
828 for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++)
829 LIST_INIT(&clp->nfsc_layouthash[i]);
830 clp->nfsc_flags = NFSCLFLAGS_INITED;
831 clp->nfsc_clientidrev = 1;
832 clp->nfsc_cbident = nfscl_nextcbident();
833 nfscl_fillclid(nmp->nm_clval, uuid, clp->nfsc_id,
835 LIST_INSERT_HEAD(&nfsclhead, clp, nfsc_list);
839 if (start_renewthread != 0)
840 nfscl_start_renewthread(clp);
844 free(newclp, M_NFSCLCLIENT);
847 while ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0 && !igotlock &&
848 !NFSCL_FORCEDISM(mp))
849 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
850 NFSCLSTATEMUTEXPTR, mp);
853 * Call nfsv4_lock() with "iwantlock == 0" so that it will
854 * wait for a pending exclusive lock request. This gives the
855 * exclusive lock request priority over this shared lock
857 * An exclusive lock on nfsc_lock is used mainly for server
860 nfsv4_lock(&clp->nfsc_lock, 0, NULL, NFSCLSTATEMUTEXPTR, mp);
861 nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
863 if (igotlock == 0 && NFSCL_FORCEDISM(mp)) {
865 * Both nfsv4_lock() and nfsv4_getref() know to check
866 * for NFSCL_FORCEDISM() and return without sleeping to
867 * wait for the exclusive lock to be released, since it
868 * might be held by nfscl_umount() and we need to get out
869 * now for that case and not wait until nfscl_umount()
878 * If it needs a clientid, do the setclientid now.
880 if ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0) {
882 panic("nfscl_clget");
883 if (p == NULL || cred == NULL) {
885 nfsv4_unlock(&clp->nfsc_lock, 0);
890 * If RFC3530 Sec. 14.2.33 is taken literally,
891 * NFSERR_CLIDINUSE will be returned persistently for the
892 * case where a new mount of the same file system is using
893 * a different principal. In practice, NFSERR_CLIDINUSE is
894 * only returned when there is outstanding unexpired state
895 * on the clientid. As such, try for twice the lease
896 * interval, if we know what that is. Otherwise, make a
898 * The case of returning NFSERR_STALECLIENTID is far less
899 * likely, but might occur if there is a significant delay
900 * between doing the SetClientID and SetClientIDConfirm Ops,
901 * such that the server throws away the clientid before
902 * receiving the SetClientIDConfirm.
904 if (clp->nfsc_renew > 0)
905 clidinusedelay = NFSCL_LEASE(clp->nfsc_renew) * 2;
907 clidinusedelay = 120;
910 error = nfsrpc_setclient(nmp, clp, 0, cred, p);
911 if (error == NFSERR_STALECLIENTID ||
912 error == NFSERR_STALEDONTRECOVER ||
913 error == NFSERR_BADSESSION ||
914 error == NFSERR_CLIDINUSE) {
915 (void) nfs_catnap(PZERO, error, "nfs_setcl");
917 } while (((error == NFSERR_STALECLIENTID ||
918 error == NFSERR_BADSESSION ||
919 error == NFSERR_STALEDONTRECOVER) && --trystalecnt > 0) ||
920 (error == NFSERR_CLIDINUSE && --clidinusedelay > 0));
923 nfsv4_unlock(&clp->nfsc_lock, 0);
927 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
931 nfsv4_unlock(&clp->nfsc_lock, 1);
940 * Get a reference to a clientid and return it, if valid.
942 APPLESTATIC struct nfsclclient *
943 nfscl_findcl(struct nfsmount *nmp)
945 struct nfsclclient *clp;
948 if (clp == NULL || !(clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID))
954 * Release the clientid structure. It may be locked or reference counted.
957 nfscl_clrelease(struct nfsclclient *clp)
960 if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK)
961 nfsv4_unlock(&clp->nfsc_lock, 0);
963 nfsv4_relref(&clp->nfsc_lock);
967 * External call for nfscl_clrelease.
970 nfscl_clientrelease(struct nfsclclient *clp)
974 if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK)
975 nfsv4_unlock(&clp->nfsc_lock, 0);
977 nfsv4_relref(&clp->nfsc_lock);
982 * Called when wanting to lock a byte region.
985 nfscl_getbytelock(vnode_t vp, u_int64_t off, u_int64_t len,
986 short type, struct ucred *cred, NFSPROC_T *p, struct nfsclclient *rclp,
987 int recovery, void *id, int flags, u_int8_t *rownp, u_int8_t *ropenownp,
988 struct nfscllockowner **lpp, int *newonep, int *donelocallyp)
990 struct nfscllockowner *lp;
991 struct nfsclopen *op;
992 struct nfsclclient *clp;
993 struct nfscllockowner *nlp;
994 struct nfscllock *nlop, *otherlop;
995 struct nfscldeleg *dp = NULL, *ldp = NULL;
996 struct nfscllockownerhead *lhp = NULL;
998 u_int8_t own[NFSV4CL_LOCKNAMELEN], *ownp, openown[NFSV4CL_LOCKNAMELEN];
1000 int error = 0, ret, donelocally = 0;
1003 /* For Lock Ops, the open mode doesn't matter, so use 0 to match any. */
1012 * Might need these, so MALLOC them now, to
1013 * avoid a tsleep() in MALLOC later.
1016 sizeof (struct nfscllockowner), M_NFSCLLOCKOWNER, M_WAITOK);
1018 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1020 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1021 nlop->nfslo_type = type;
1022 nlop->nfslo_first = off;
1023 if (len == NFS64BITSSET) {
1024 nlop->nfslo_end = NFS64BITSSET;
1026 nlop->nfslo_end = off + len;
1027 if (nlop->nfslo_end <= nlop->nfslo_first)
1028 error = NFSERR_INVAL;
1035 error = nfscl_getcl(vnode_mount(vp), cred, p, 1, &clp);
1038 free(nlp, M_NFSCLLOCKOWNER);
1039 free(otherlop, M_NFSCLLOCK);
1040 free(nlop, M_NFSCLLOCK);
1047 openownp = ropenownp;
1049 nfscl_filllockowner(id, own, flags);
1051 if (NFSHASONEOPENOWN(VFSTONFS(vnode_mount(vp))))
1052 nfscl_filllockowner(NULL, openown, F_POSIX);
1054 nfscl_filllockowner(p->td_proc, openown, F_POSIX);
1060 * First, search for a delegation. If one exists for this file,
1061 * the lock can be done locally against it, so long as there
1062 * isn't a local lock conflict.
1064 ldp = dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
1065 np->n_fhp->nfh_len);
1066 /* Just sanity check for correct type of delegation */
1067 if (dp != NULL && ((dp->nfsdl_flags &
1068 (NFSCLDL_RECALL | NFSCLDL_DELEGRET)) != 0 ||
1070 (dp->nfsdl_flags & NFSCLDL_WRITE) == 0)))
1074 /* Now, find an open and maybe a lockowner. */
1075 ret = nfscl_getopen(&dp->nfsdl_owner, np->n_fhp->nfh_fh,
1076 np->n_fhp->nfh_len, openownp, ownp, mode, NULL, &op);
1078 ret = nfscl_getopen(&clp->nfsc_owner,
1079 np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp,
1080 ownp, mode, NULL, &op);
1082 lhp = &dp->nfsdl_lock;
1083 TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
1084 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list);
1085 dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
1093 * Get the related Open and maybe lockowner.
1095 error = nfscl_getopen(&clp->nfsc_owner,
1096 np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp,
1097 ownp, mode, &lp, &op);
1099 lhp = &op->nfso_lock;
1101 if (!error && !recovery)
1102 error = nfscl_localconflict(clp, np->n_fhp->nfh_fh,
1103 np->n_fhp->nfh_len, nlop, ownp, ldp, NULL);
1106 nfscl_clrelease(clp);
1109 free(nlp, M_NFSCLLOCKOWNER);
1110 free(otherlop, M_NFSCLLOCK);
1111 free(nlop, M_NFSCLLOCK);
1116 * Ok, see if a lockowner exists and create one, as required.
1119 LIST_FOREACH(lp, lhp, nfsl_list) {
1120 if (!NFSBCMP(lp->nfsl_owner, ownp, NFSV4CL_LOCKNAMELEN))
1124 NFSBCOPY(ownp, nlp->nfsl_owner, NFSV4CL_LOCKNAMELEN);
1126 NFSBCOPY(ropenownp, nlp->nfsl_openowner,
1127 NFSV4CL_LOCKNAMELEN);
1129 NFSBCOPY(op->nfso_own->nfsow_owner, nlp->nfsl_openowner,
1130 NFSV4CL_LOCKNAMELEN);
1131 nlp->nfsl_seqid = 0;
1132 nlp->nfsl_lockflags = flags;
1133 nlp->nfsl_inprog = NULL;
1134 nfscl_lockinit(&nlp->nfsl_rwlock);
1135 LIST_INIT(&nlp->nfsl_lock);
1137 nlp->nfsl_open = NULL;
1138 nfsstatsv1.cllocallockowners++;
1140 nlp->nfsl_open = op;
1141 nfsstatsv1.cllockowners++;
1143 LIST_INSERT_HEAD(lhp, nlp, nfsl_list);
1150 * Now, update the byte ranges for locks.
1152 ret = nfscl_updatelock(lp, &nlop, &otherlop, donelocally);
1158 nfscl_clrelease(clp);
1161 * Serial modifications on the lock owner for multiple threads
1162 * for the same process using a read/write lock.
1165 nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR);
1171 free(nlp, M_NFSCLLOCKOWNER);
1173 free(nlop, M_NFSCLLOCK);
1175 free(otherlop, M_NFSCLLOCK);
1182 * Called to unlock a byte range, for LockU.
1185 nfscl_relbytelock(vnode_t vp, u_int64_t off, u_int64_t len,
1186 __unused struct ucred *cred, NFSPROC_T *p, int callcnt,
1187 struct nfsclclient *clp, void *id, int flags,
1188 struct nfscllockowner **lpp, int *dorpcp)
1190 struct nfscllockowner *lp;
1191 struct nfsclowner *owp;
1192 struct nfsclopen *op;
1193 struct nfscllock *nlop, *other_lop = NULL;
1194 struct nfscldeleg *dp;
1196 u_int8_t own[NFSV4CL_LOCKNAMELEN];
1204 * Might need these, so MALLOC them now, to
1205 * avoid a tsleep() in MALLOC later.
1208 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1209 nlop->nfslo_type = F_UNLCK;
1210 nlop->nfslo_first = off;
1211 if (len == NFS64BITSSET) {
1212 nlop->nfslo_end = NFS64BITSSET;
1214 nlop->nfslo_end = off + len;
1215 if (nlop->nfslo_end <= nlop->nfslo_first) {
1216 free(nlop, M_NFSCLLOCK);
1217 return (NFSERR_INVAL);
1222 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1225 nfscl_filllockowner(id, own, flags);
1229 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
1230 np->n_fhp->nfh_len);
1233 * First, unlock any local regions on a delegation.
1236 /* Look for this lockowner. */
1237 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
1238 if (!NFSBCMP(lp->nfsl_owner, own,
1239 NFSV4CL_LOCKNAMELEN))
1243 /* Use other_lop, so nlop is still available */
1244 (void)nfscl_updatelock(lp, &other_lop, NULL, 1);
1248 * Now, find a matching open/lockowner that hasn't already been done,
1249 * as marked by nfsl_inprog.
1253 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
1254 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1255 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1256 !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1257 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1258 if (lp->nfsl_inprog == NULL &&
1259 !NFSBCMP(lp->nfsl_owner, own,
1260 NFSV4CL_LOCKNAMELEN)) {
1274 ret = nfscl_updatelock(lp, &nlop, NULL, 0);
1278 * Serial modifications on the lock owner for multiple
1279 * threads for the same process using a read/write lock.
1281 lp->nfsl_inprog = p;
1282 nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR);
1287 free(nlop, M_NFSCLLOCK);
1289 free(other_lop, M_NFSCLLOCK);
1294 * Release all lockowners marked in progess for this process and file.
1297 nfscl_releasealllocks(struct nfsclclient *clp, vnode_t vp, NFSPROC_T *p,
1298 void *id, int flags)
1300 struct nfsclowner *owp;
1301 struct nfsclopen *op;
1302 struct nfscllockowner *lp;
1304 u_int8_t own[NFSV4CL_LOCKNAMELEN];
1307 nfscl_filllockowner(id, own, flags);
1309 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
1310 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1311 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1312 !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1313 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1314 if (lp->nfsl_inprog == p &&
1315 !NFSBCMP(lp->nfsl_owner, own,
1316 NFSV4CL_LOCKNAMELEN)) {
1317 lp->nfsl_inprog = NULL;
1318 nfscl_lockunlock(&lp->nfsl_rwlock);
1324 nfscl_clrelease(clp);
1329 * Called to find out if any bytes within the byte range specified are
1330 * write locked by the calling process. Used to determine if flushing
1331 * is required before a LockU.
1332 * If in doubt, return 1, so the flush will occur.
1335 nfscl_checkwritelocked(vnode_t vp, struct flock *fl,
1336 struct ucred *cred, NFSPROC_T *p, void *id, int flags)
1338 struct nfsclowner *owp;
1339 struct nfscllockowner *lp;
1340 struct nfsclopen *op;
1341 struct nfsclclient *clp;
1342 struct nfscllock *lop;
1343 struct nfscldeleg *dp;
1346 u_int8_t own[NFSV4CL_LOCKNAMELEN];
1350 switch (fl->l_whence) {
1354 * Caller is responsible for adding any necessary offset
1355 * when SEEK_CUR is used.
1360 off = np->n_size + fl->l_start;
1365 if (fl->l_len != 0) {
1366 end = off + fl->l_len;
1373 error = nfscl_getcl(vnode_mount(vp), cred, p, 1, &clp);
1376 nfscl_filllockowner(id, own, flags);
1380 * First check the delegation locks.
1382 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
1384 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
1385 if (!NFSBCMP(lp->nfsl_owner, own,
1386 NFSV4CL_LOCKNAMELEN))
1390 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
1391 if (lop->nfslo_first >= end)
1393 if (lop->nfslo_end <= off)
1395 if (lop->nfslo_type == F_WRLCK) {
1396 nfscl_clrelease(clp);
1405 * Now, check state against the server.
1407 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
1408 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1409 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1410 !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1411 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1412 if (!NFSBCMP(lp->nfsl_owner, own,
1413 NFSV4CL_LOCKNAMELEN))
1417 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
1418 if (lop->nfslo_first >= end)
1420 if (lop->nfslo_end <= off)
1422 if (lop->nfslo_type == F_WRLCK) {
1423 nfscl_clrelease(clp);
1432 nfscl_clrelease(clp);
1438 * Release a byte range lock owner structure.
1441 nfscl_lockrelease(struct nfscllockowner *lp, int error, int candelete)
1443 struct nfsclclient *clp;
1448 clp = lp->nfsl_open->nfso_own->nfsow_clp;
1449 if (error != 0 && candelete &&
1450 (lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED) == 0)
1451 nfscl_freelockowner(lp, 0);
1453 nfscl_lockunlock(&lp->nfsl_rwlock);
1454 nfscl_clrelease(clp);
1459 * Free up an open structure and any associated byte range lock structures.
1462 nfscl_freeopen(struct nfsclopen *op, int local)
1465 LIST_REMOVE(op, nfso_list);
1466 nfscl_freealllocks(&op->nfso_lock, local);
1467 free(op, M_NFSCLOPEN);
1469 nfsstatsv1.cllocalopens--;
1471 nfsstatsv1.clopens--;
1475 * Free up all lock owners and associated locks.
1478 nfscl_freealllocks(struct nfscllockownerhead *lhp, int local)
1480 struct nfscllockowner *lp, *nlp;
1482 LIST_FOREACH_SAFE(lp, lhp, nfsl_list, nlp) {
1483 if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED))
1485 nfscl_freelockowner(lp, local);
1490 * Called for an Open when NFSERR_EXPIRED is received from the server.
1491 * If there are no byte range locks nor a Share Deny lost, try to do a
1492 * fresh Open. Otherwise, free the open.
1495 nfscl_expireopen(struct nfsclclient *clp, struct nfsclopen *op,
1496 struct nfsmount *nmp, struct ucred *cred, NFSPROC_T *p)
1498 struct nfscllockowner *lp;
1499 struct nfscldeleg *dp;
1500 int mustdelete = 0, error;
1503 * Look for any byte range lock(s).
1505 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1506 if (!LIST_EMPTY(&lp->nfsl_lock)) {
1513 * If no byte range lock(s) nor a Share deny, try to re-open.
1515 if (!mustdelete && (op->nfso_mode & NFSLCK_DENYBITS) == 0) {
1516 newnfs_copycred(&op->nfso_cred, cred);
1518 error = nfsrpc_reopen(nmp, op->nfso_fh,
1519 op->nfso_fhlen, op->nfso_mode, op, &dp, cred, p);
1523 free(dp, M_NFSCLDELEG);
1528 nfscl_deleg(nmp->nm_mountp, clp, op->nfso_fh,
1529 op->nfso_fhlen, cred, p, &dp);
1533 * If a byte range lock or Share deny or couldn't re-open, free it.
1536 nfscl_freeopen(op, 0);
1537 return (mustdelete);
1541 * Free up an open owner structure.
1544 nfscl_freeopenowner(struct nfsclowner *owp, int local)
1547 LIST_REMOVE(owp, nfsow_list);
1548 free(owp, M_NFSCLOWNER);
1550 nfsstatsv1.cllocalopenowners--;
1552 nfsstatsv1.clopenowners--;
1556 * Free up a byte range lock owner structure.
1559 nfscl_freelockowner(struct nfscllockowner *lp, int local)
1561 struct nfscllock *lop, *nlop;
1563 LIST_REMOVE(lp, nfsl_list);
1564 LIST_FOREACH_SAFE(lop, &lp->nfsl_lock, nfslo_list, nlop) {
1565 nfscl_freelock(lop, local);
1567 free(lp, M_NFSCLLOCKOWNER);
1569 nfsstatsv1.cllocallockowners--;
1571 nfsstatsv1.cllockowners--;
1575 * Free up a byte range lock structure.
1578 nfscl_freelock(struct nfscllock *lop, int local)
1581 LIST_REMOVE(lop, nfslo_list);
1582 free(lop, M_NFSCLLOCK);
1584 nfsstatsv1.cllocallocks--;
1586 nfsstatsv1.cllocks--;
1590 * Clean out the state related to a delegation.
1593 nfscl_cleandeleg(struct nfscldeleg *dp)
1595 struct nfsclowner *owp, *nowp;
1596 struct nfsclopen *op;
1598 LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) {
1599 op = LIST_FIRST(&owp->nfsow_open);
1601 if (LIST_NEXT(op, nfso_list) != NULL)
1602 panic("nfscleandel");
1603 nfscl_freeopen(op, 1);
1605 nfscl_freeopenowner(owp, 1);
1607 nfscl_freealllocks(&dp->nfsdl_lock, 1);
1611 * Free a delegation.
1614 nfscl_freedeleg(struct nfscldeleghead *hdp, struct nfscldeleg *dp)
1617 TAILQ_REMOVE(hdp, dp, nfsdl_list);
1618 LIST_REMOVE(dp, nfsdl_hash);
1619 free(dp, M_NFSCLDELEG);
1620 nfsstatsv1.cldelegates--;
1625 * Free up all state related to this client structure.
1628 nfscl_cleanclient(struct nfsclclient *clp)
1630 struct nfsclowner *owp, *nowp;
1631 struct nfsclopen *op, *nop;
1632 struct nfscllayout *lyp, *nlyp;
1633 struct nfscldevinfo *dip, *ndip;
1635 TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp)
1636 nfscl_freelayout(lyp);
1638 LIST_FOREACH_SAFE(dip, &clp->nfsc_devinfo, nfsdi_list, ndip)
1639 nfscl_freedevinfo(dip);
1641 /* Now, all the OpenOwners, etc. */
1642 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1643 LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
1644 nfscl_freeopen(op, 0);
1646 nfscl_freeopenowner(owp, 0);
1651 * Called when an NFSERR_EXPIRED is received from the server.
1654 nfscl_expireclient(struct nfsclclient *clp, struct nfsmount *nmp,
1655 struct ucred *cred, NFSPROC_T *p)
1657 struct nfsclowner *owp, *nowp, *towp;
1658 struct nfsclopen *op, *nop, *top;
1659 struct nfscldeleg *dp, *ndp;
1660 int ret, printed = 0;
1663 * First, merge locally issued Opens into the list for the server.
1665 dp = TAILQ_FIRST(&clp->nfsc_deleg);
1666 while (dp != NULL) {
1667 ndp = TAILQ_NEXT(dp, nfsdl_list);
1668 owp = LIST_FIRST(&dp->nfsdl_owner);
1669 while (owp != NULL) {
1670 nowp = LIST_NEXT(owp, nfsow_list);
1671 op = LIST_FIRST(&owp->nfsow_open);
1673 if (LIST_NEXT(op, nfso_list) != NULL)
1675 LIST_FOREACH(towp, &clp->nfsc_owner, nfsow_list) {
1676 if (!NFSBCMP(towp->nfsow_owner, owp->nfsow_owner,
1677 NFSV4CL_LOCKNAMELEN))
1681 /* Merge opens in */
1682 LIST_FOREACH(top, &towp->nfsow_open, nfso_list) {
1683 if (top->nfso_fhlen == op->nfso_fhlen &&
1684 !NFSBCMP(top->nfso_fh, op->nfso_fh,
1686 top->nfso_mode |= op->nfso_mode;
1687 top->nfso_opencnt += op->nfso_opencnt;
1692 /* Just add the open to the owner list */
1693 LIST_REMOVE(op, nfso_list);
1694 op->nfso_own = towp;
1695 LIST_INSERT_HEAD(&towp->nfsow_open, op, nfso_list);
1696 nfsstatsv1.cllocalopens--;
1697 nfsstatsv1.clopens++;
1700 /* Just add the openowner to the client list */
1701 LIST_REMOVE(owp, nfsow_list);
1702 owp->nfsow_clp = clp;
1703 LIST_INSERT_HEAD(&clp->nfsc_owner, owp, nfsow_list);
1704 nfsstatsv1.cllocalopenowners--;
1705 nfsstatsv1.clopenowners++;
1706 nfsstatsv1.cllocalopens--;
1707 nfsstatsv1.clopens++;
1712 if (!printed && !LIST_EMPTY(&dp->nfsdl_lock)) {
1714 printf("nfsv4 expired locks lost\n");
1716 nfscl_cleandeleg(dp);
1717 nfscl_freedeleg(&clp->nfsc_deleg, dp);
1720 if (!TAILQ_EMPTY(&clp->nfsc_deleg))
1724 * Now, try and reopen against the server.
1726 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1727 owp->nfsow_seqid = 0;
1728 LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
1729 ret = nfscl_expireopen(clp, op, nmp, cred, p);
1730 if (ret && !printed) {
1732 printf("nfsv4 expired locks lost\n");
1735 if (LIST_EMPTY(&owp->nfsow_open))
1736 nfscl_freeopenowner(owp, 0);
1741 * This function must be called after the process represented by "own" has
1742 * exited. Must be called with CLSTATE lock held.
1745 nfscl_cleanup_common(struct nfsclclient *clp, u_int8_t *own)
1747 struct nfsclowner *owp, *nowp;
1748 struct nfscllockowner *lp, *nlp;
1749 struct nfscldeleg *dp;
1751 /* First, get rid of local locks on delegations. */
1752 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
1753 LIST_FOREACH_SAFE(lp, &dp->nfsdl_lock, nfsl_list, nlp) {
1754 if (!NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
1755 if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED))
1757 nfscl_freelockowner(lp, 1);
1761 owp = LIST_FIRST(&clp->nfsc_owner);
1762 while (owp != NULL) {
1763 nowp = LIST_NEXT(owp, nfsow_list);
1764 if (!NFSBCMP(owp->nfsow_owner, own,
1765 NFSV4CL_LOCKNAMELEN)) {
1767 * If there are children that haven't closed the
1768 * file descriptors yet, the opens will still be
1769 * here. For that case, let the renew thread clear
1770 * out the OpenOwner later.
1772 if (LIST_EMPTY(&owp->nfsow_open))
1773 nfscl_freeopenowner(owp, 0);
1775 owp->nfsow_defunct = 1;
1782 * Find open/lock owners for processes that have exited.
1785 nfscl_cleanupkext(struct nfsclclient *clp, struct nfscllockownerfhhead *lhp)
1787 struct nfsclowner *owp, *nowp;
1788 struct nfsclopen *op;
1789 struct nfscllockowner *lp, *nlp;
1790 struct nfscldeleg *dp;
1793 * All the pidhash locks must be acquired, since they are sx locks
1794 * and must be acquired before the mutexes. The pid(s) that will
1795 * be used aren't known yet, so all the locks need to be acquired.
1796 * Fortunately, this function is only performed once/sec.
1800 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1801 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1802 LIST_FOREACH_SAFE(lp, &op->nfso_lock, nfsl_list, nlp) {
1803 if (LIST_EMPTY(&lp->nfsl_lock))
1804 nfscl_emptylockowner(lp, lhp);
1807 if (nfscl_procdoesntexist(owp->nfsow_owner))
1808 nfscl_cleanup_common(clp, owp->nfsow_owner);
1812 * For the single open_owner case, these lock owners need to be
1813 * checked to see if they still exist separately.
1814 * This is because nfscl_procdoesntexist() never returns true for
1815 * the single open_owner so that the above doesn't ever call
1816 * nfscl_cleanup_common().
1818 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
1819 LIST_FOREACH_SAFE(lp, &dp->nfsdl_lock, nfsl_list, nlp) {
1820 if (nfscl_procdoesntexist(lp->nfsl_owner))
1821 nfscl_cleanup_common(clp, lp->nfsl_owner);
1825 pidhash_sunlockall();
1829 * Take the empty lock owner and move it to the local lhp list if the
1830 * associated process no longer exists.
1833 nfscl_emptylockowner(struct nfscllockowner *lp,
1834 struct nfscllockownerfhhead *lhp)
1836 struct nfscllockownerfh *lfhp, *mylfhp;
1837 struct nfscllockowner *nlp;
1840 /* If not a Posix lock owner, just return. */
1841 if ((lp->nfsl_lockflags & F_POSIX) == 0)
1847 * First, search to see if this lock owner is already in the list.
1848 * If it is, then the associated process no longer exists.
1850 SLIST_FOREACH(lfhp, lhp, nfslfh_list) {
1851 if (lfhp->nfslfh_len == lp->nfsl_open->nfso_fhlen &&
1852 !NFSBCMP(lfhp->nfslfh_fh, lp->nfsl_open->nfso_fh,
1855 LIST_FOREACH(nlp, &lfhp->nfslfh_lock, nfsl_list)
1856 if (!NFSBCMP(nlp->nfsl_owner, lp->nfsl_owner,
1857 NFSV4CL_LOCKNAMELEN))
1860 /* If not found, check if process still exists. */
1861 if (fnd_it == 0 && nfscl_procdoesntexist(lp->nfsl_owner) == 0)
1864 /* Move the lock owner over to the local list. */
1865 if (mylfhp == NULL) {
1866 mylfhp = malloc(sizeof(struct nfscllockownerfh), M_TEMP,
1870 mylfhp->nfslfh_len = lp->nfsl_open->nfso_fhlen;
1871 NFSBCOPY(lp->nfsl_open->nfso_fh, mylfhp->nfslfh_fh,
1872 mylfhp->nfslfh_len);
1873 LIST_INIT(&mylfhp->nfslfh_lock);
1874 SLIST_INSERT_HEAD(lhp, mylfhp, nfslfh_list);
1876 LIST_REMOVE(lp, nfsl_list);
1877 LIST_INSERT_HEAD(&mylfhp->nfslfh_lock, lp, nfsl_list);
1880 static int fake_global; /* Used to force visibility of MNTK_UNMOUNTF */
1882 * Called from nfs umount to free up the clientid.
1885 nfscl_umount(struct nfsmount *nmp, NFSPROC_T *p)
1887 struct nfsclclient *clp;
1892 * For the case that matters, this is the thread that set
1893 * MNTK_UNMOUNTF, so it will see it set. The code that follows is
1894 * done to ensure that any thread executing nfscl_getcl() after
1895 * this time, will see MNTK_UNMOUNTF set. nfscl_getcl() uses the
1896 * mutex for NFSLOCKCLSTATE(), so it is "m" for the following
1897 * explanation, courtesy of Alan Cox.
1898 * What follows is a snippet from Alan Cox's email at:
1899 * https://docs.FreeBSD.org/cgi/mid.cgi?BANLkTikR3d65zPHo9==08ZfJ2vmqZucEvw
1901 * 1. Set MNTK_UNMOUNTF
1902 * 2. Acquire a standard FreeBSD mutex "m".
1903 * 3. Update some data structures.
1904 * 4. Release mutex "m".
1906 * Then, other threads that acquire "m" after step 4 has occurred will
1907 * see MNTK_UNMOUNTF as set. But, other threads that beat thread X to
1908 * step 2 may or may not see MNTK_UNMOUNTF as set.
1911 if ((nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) {
1919 if ((clp->nfsc_flags & NFSCLFLAGS_INITED) == 0)
1920 panic("nfscl umount");
1923 * First, handshake with the nfscl renew thread, to terminate
1926 clp->nfsc_flags |= NFSCLFLAGS_UMOUNT;
1927 while (clp->nfsc_flags & NFSCLFLAGS_HASTHREAD)
1928 (void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT,
1932 * Now, get the exclusive lock on the client state, so
1933 * that no uses of the state are still in progress.
1936 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
1937 NFSCLSTATEMUTEXPTR, NULL);
1938 } while (!igotlock);
1942 * Free up all the state. It will expire on the server, but
1943 * maybe we should do a SetClientId/SetClientIdConfirm so
1944 * the server throws it away?
1946 LIST_REMOVE(clp, nfsc_list);
1947 nfscl_delegreturnall(clp, p);
1948 cred = newnfs_getcred();
1949 if (NFSHASNFSV4N(nmp)) {
1950 (void)nfsrpc_destroysession(nmp, clp, cred, p);
1951 (void)nfsrpc_destroyclient(nmp, clp, cred, p);
1953 (void)nfsrpc_setclient(nmp, clp, 0, cred, p);
1954 nfscl_cleanclient(clp);
1957 free(clp, M_NFSCLCLIENT);
1963 * This function is called when a server replies with NFSERR_STALECLIENTID
1964 * NFSERR_STALESTATEID or NFSERR_BADSESSION. It traverses the clientid lists,
1965 * doing Opens and Locks with reclaim. If these fail, it deletes the
1966 * corresponding state.
1969 nfscl_recover(struct nfsclclient *clp, struct ucred *cred, NFSPROC_T *p)
1971 struct nfsclowner *owp, *nowp;
1972 struct nfsclopen *op, *nop;
1973 struct nfscllockowner *lp, *nlp;
1974 struct nfscllock *lop, *nlop;
1975 struct nfscldeleg *dp, *ndp, *tdp;
1976 struct nfsmount *nmp;
1977 struct ucred *tcred;
1978 struct nfsclopenhead extra_open;
1979 struct nfscldeleghead extra_deleg;
1982 u_int32_t delegtype = NFSV4OPEN_DELEGATEWRITE, mode;
1983 int i, igotlock = 0, error, trycnt, firstlock;
1984 struct nfscllayout *lyp, *nlyp;
1987 * First, lock the client structure, so everyone else will
1988 * block when trying to use state.
1991 clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG;
1993 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
1994 NFSCLSTATEMUTEXPTR, NULL);
1995 } while (!igotlock);
1998 nmp = clp->nfsc_nmp;
2000 panic("nfscl recover");
2003 * For now, just get rid of all layouts. There may be a need
2004 * to do LayoutCommit Ops with reclaim == true later.
2006 TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp)
2007 nfscl_freelayout(lyp);
2008 TAILQ_INIT(&clp->nfsc_layout);
2009 for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++)
2010 LIST_INIT(&clp->nfsc_layouthash[i]);
2014 error = nfsrpc_setclient(nmp, clp, 1, cred, p);
2015 } while ((error == NFSERR_STALECLIENTID ||
2016 error == NFSERR_BADSESSION ||
2017 error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
2020 clp->nfsc_flags &= ~(NFSCLFLAGS_RECOVER |
2021 NFSCLFLAGS_RECVRINPROG);
2022 wakeup(&clp->nfsc_flags);
2023 nfsv4_unlock(&clp->nfsc_lock, 0);
2027 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
2028 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2031 * Mark requests already queued on the server, so that they don't
2032 * initiate another recovery cycle. Any requests already in the
2033 * queue that handle state information will have the old stale
2034 * clientid/stateid and will get a NFSERR_STALESTATEID,
2035 * NFSERR_STALECLIENTID or NFSERR_BADSESSION reply from the server.
2036 * This will be translated to NFSERR_STALEDONTRECOVER when
2037 * R_DONTRECOVER is set.
2040 TAILQ_FOREACH(rep, &nfsd_reqq, r_chain) {
2041 if (rep->r_nmp == nmp)
2042 rep->r_flags |= R_DONTRECOVER;
2047 * Now, mark all delegations "need reclaim".
2049 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list)
2050 dp->nfsdl_flags |= NFSCLDL_NEEDRECLAIM;
2052 TAILQ_INIT(&extra_deleg);
2053 LIST_INIT(&extra_open);
2055 * Now traverse the state lists, doing Open and Lock Reclaims.
2057 tcred = newnfs_getcred();
2058 owp = LIST_FIRST(&clp->nfsc_owner);
2059 while (owp != NULL) {
2060 nowp = LIST_NEXT(owp, nfsow_list);
2061 owp->nfsow_seqid = 0;
2062 op = LIST_FIRST(&owp->nfsow_open);
2063 while (op != NULL) {
2064 nop = LIST_NEXT(op, nfso_list);
2065 if (error != NFSERR_NOGRACE && error != NFSERR_BADSESSION) {
2066 /* Search for a delegation to reclaim with the open */
2067 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2068 if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
2070 if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
2071 mode = NFSV4OPEN_ACCESSWRITE;
2072 delegtype = NFSV4OPEN_DELEGATEWRITE;
2074 mode = NFSV4OPEN_ACCESSREAD;
2075 delegtype = NFSV4OPEN_DELEGATEREAD;
2077 if ((op->nfso_mode & mode) == mode &&
2078 op->nfso_fhlen == dp->nfsdl_fhlen &&
2079 !NFSBCMP(op->nfso_fh, dp->nfsdl_fh, op->nfso_fhlen))
2084 delegtype = NFSV4OPEN_DELEGATENONE;
2085 newnfs_copycred(&op->nfso_cred, tcred);
2086 error = nfscl_tryopen(nmp, NULL, op->nfso_fh,
2087 op->nfso_fhlen, op->nfso_fh, op->nfso_fhlen,
2088 op->nfso_mode, op, NULL, 0, &ndp, 1, delegtype,
2091 /* Handle any replied delegation */
2092 if (ndp != NULL && ((ndp->nfsdl_flags & NFSCLDL_WRITE)
2093 || NFSMNT_RDONLY(nmp->nm_mountp))) {
2094 if ((ndp->nfsdl_flags & NFSCLDL_WRITE))
2095 mode = NFSV4OPEN_ACCESSWRITE;
2097 mode = NFSV4OPEN_ACCESSREAD;
2098 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2099 if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
2101 if ((op->nfso_mode & mode) == mode &&
2102 op->nfso_fhlen == dp->nfsdl_fhlen &&
2103 !NFSBCMP(op->nfso_fh, dp->nfsdl_fh,
2105 dp->nfsdl_stateid = ndp->nfsdl_stateid;
2106 dp->nfsdl_sizelimit = ndp->nfsdl_sizelimit;
2107 dp->nfsdl_ace = ndp->nfsdl_ace;
2108 dp->nfsdl_change = ndp->nfsdl_change;
2109 dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
2110 if ((ndp->nfsdl_flags & NFSCLDL_RECALL))
2111 dp->nfsdl_flags |= NFSCLDL_RECALL;
2112 free(ndp, M_NFSCLDELEG);
2119 TAILQ_INSERT_HEAD(&extra_deleg, ndp, nfsdl_list);
2121 /* and reclaim all byte range locks */
2122 lp = LIST_FIRST(&op->nfso_lock);
2123 while (lp != NULL) {
2124 nlp = LIST_NEXT(lp, nfsl_list);
2127 lop = LIST_FIRST(&lp->nfsl_lock);
2128 while (lop != NULL) {
2129 nlop = LIST_NEXT(lop, nfslo_list);
2130 if (lop->nfslo_end == NFS64BITSSET)
2133 len = lop->nfslo_end - lop->nfslo_first;
2134 error = nfscl_trylock(nmp, NULL,
2135 op->nfso_fh, op->nfso_fhlen, lp,
2136 firstlock, 1, lop->nfslo_first, len,
2137 lop->nfslo_type, tcred, p);
2139 nfscl_freelock(lop, 0);
2144 /* If no locks, but a lockowner, just delete it. */
2145 if (LIST_EMPTY(&lp->nfsl_lock))
2146 nfscl_freelockowner(lp, 0);
2151 if (error != 0 && error != NFSERR_BADSESSION)
2152 nfscl_freeopen(op, 0);
2159 * Now, try and get any delegations not yet reclaimed by cobbling
2160 * to-gether an appropriate open.
2163 dp = TAILQ_FIRST(&clp->nfsc_deleg);
2164 while (dp != NULL) {
2165 ndp = TAILQ_NEXT(dp, nfsdl_list);
2166 if ((dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM)) {
2169 sizeof (struct nfsclowner), M_NFSCLOWNER, M_WAITOK);
2171 * Name must be as long an largest possible
2172 * NFSV4CL_LOCKNAMELEN. 12 for now.
2174 NFSBCOPY("RECLAIMDELEG", nowp->nfsow_owner,
2175 NFSV4CL_LOCKNAMELEN);
2176 LIST_INIT(&nowp->nfsow_open);
2177 nowp->nfsow_clp = clp;
2178 nowp->nfsow_seqid = 0;
2179 nowp->nfsow_defunct = 0;
2180 nfscl_lockinit(&nowp->nfsow_rwlock);
2183 if (error != NFSERR_NOGRACE && error != NFSERR_BADSESSION) {
2184 nop = malloc(sizeof (struct nfsclopen) +
2185 dp->nfsdl_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
2186 nop->nfso_own = nowp;
2187 if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
2188 nop->nfso_mode = NFSV4OPEN_ACCESSWRITE;
2189 delegtype = NFSV4OPEN_DELEGATEWRITE;
2191 nop->nfso_mode = NFSV4OPEN_ACCESSREAD;
2192 delegtype = NFSV4OPEN_DELEGATEREAD;
2194 nop->nfso_opencnt = 0;
2195 nop->nfso_posixlock = 1;
2196 nop->nfso_fhlen = dp->nfsdl_fhlen;
2197 NFSBCOPY(dp->nfsdl_fh, nop->nfso_fh, dp->nfsdl_fhlen);
2198 LIST_INIT(&nop->nfso_lock);
2199 nop->nfso_stateid.seqid = 0;
2200 nop->nfso_stateid.other[0] = 0;
2201 nop->nfso_stateid.other[1] = 0;
2202 nop->nfso_stateid.other[2] = 0;
2203 newnfs_copycred(&dp->nfsdl_cred, tcred);
2204 newnfs_copyincred(tcred, &nop->nfso_cred);
2206 error = nfscl_tryopen(nmp, NULL, nop->nfso_fh,
2207 nop->nfso_fhlen, nop->nfso_fh, nop->nfso_fhlen,
2208 nop->nfso_mode, nop, NULL, 0, &tdp, 1,
2209 delegtype, tcred, p);
2211 if ((tdp->nfsdl_flags & NFSCLDL_WRITE))
2212 mode = NFSV4OPEN_ACCESSWRITE;
2214 mode = NFSV4OPEN_ACCESSREAD;
2215 if ((nop->nfso_mode & mode) == mode &&
2216 nop->nfso_fhlen == tdp->nfsdl_fhlen &&
2217 !NFSBCMP(nop->nfso_fh, tdp->nfsdl_fh,
2219 dp->nfsdl_stateid = tdp->nfsdl_stateid;
2220 dp->nfsdl_sizelimit = tdp->nfsdl_sizelimit;
2221 dp->nfsdl_ace = tdp->nfsdl_ace;
2222 dp->nfsdl_change = tdp->nfsdl_change;
2223 dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
2224 if ((tdp->nfsdl_flags & NFSCLDL_RECALL))
2225 dp->nfsdl_flags |= NFSCLDL_RECALL;
2226 free(tdp, M_NFSCLDELEG);
2228 TAILQ_INSERT_HEAD(&extra_deleg, tdp, nfsdl_list);
2234 free(nop, M_NFSCLOPEN);
2236 * Couldn't reclaim it, so throw the state
2239 nfscl_cleandeleg(dp);
2240 nfscl_freedeleg(&clp->nfsc_deleg, dp);
2242 LIST_INSERT_HEAD(&extra_open, nop, nfso_list);
2249 * Now, get rid of extra Opens and Delegations.
2251 LIST_FOREACH_SAFE(op, &extra_open, nfso_list, nop) {
2253 newnfs_copycred(&op->nfso_cred, tcred);
2254 error = nfscl_tryclose(op, tcred, nmp, p);
2255 if (error == NFSERR_GRACE)
2256 (void) nfs_catnap(PZERO, error, "nfsexcls");
2257 } while (error == NFSERR_GRACE);
2258 LIST_REMOVE(op, nfso_list);
2259 free(op, M_NFSCLOPEN);
2262 free(nowp, M_NFSCLOWNER);
2264 TAILQ_FOREACH_SAFE(dp, &extra_deleg, nfsdl_list, ndp) {
2266 newnfs_copycred(&dp->nfsdl_cred, tcred);
2267 error = nfscl_trydelegreturn(dp, tcred, nmp, p);
2268 if (error == NFSERR_GRACE)
2269 (void) nfs_catnap(PZERO, error, "nfsexdlg");
2270 } while (error == NFSERR_GRACE);
2271 TAILQ_REMOVE(&extra_deleg, dp, nfsdl_list);
2272 free(dp, M_NFSCLDELEG);
2275 /* For NFSv4.1 or later, do a RECLAIM_COMPLETE. */
2276 if (NFSHASNFSV4N(nmp))
2277 (void)nfsrpc_reclaimcomplete(nmp, cred, p);
2280 clp->nfsc_flags &= ~NFSCLFLAGS_RECVRINPROG;
2281 wakeup(&clp->nfsc_flags);
2282 nfsv4_unlock(&clp->nfsc_lock, 0);
2288 * This function is called when a server replies with NFSERR_EXPIRED.
2289 * It deletes all state for the client and does a fresh SetClientId/confirm.
2290 * XXX Someday it should post a signal to the process(es) that hold the
2291 * state, so they know that lock state has been lost.
2294 nfscl_hasexpired(struct nfsclclient *clp, u_int32_t clidrev, NFSPROC_T *p)
2296 struct nfsmount *nmp;
2298 int igotlock = 0, error, trycnt;
2301 * If the clientid has gone away or a new SetClientid has already
2302 * been done, just return ok.
2304 if (clp == NULL || clidrev != clp->nfsc_clientidrev)
2308 * First, lock the client structure, so everyone else will
2309 * block when trying to use state. Also, use NFSCLFLAGS_EXPIREIT so
2310 * that only one thread does the work.
2313 clp->nfsc_flags |= NFSCLFLAGS_EXPIREIT;
2315 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
2316 NFSCLSTATEMUTEXPTR, NULL);
2317 } while (!igotlock && (clp->nfsc_flags & NFSCLFLAGS_EXPIREIT));
2318 if ((clp->nfsc_flags & NFSCLFLAGS_EXPIREIT) == 0) {
2320 nfsv4_unlock(&clp->nfsc_lock, 0);
2324 clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG;
2327 nmp = clp->nfsc_nmp;
2329 panic("nfscl expired");
2330 cred = newnfs_getcred();
2333 error = nfsrpc_setclient(nmp, clp, 0, cred, p);
2334 } while ((error == NFSERR_STALECLIENTID ||
2335 error == NFSERR_BADSESSION ||
2336 error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
2339 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2342 * Expire the state for the client.
2344 nfscl_expireclient(clp, nmp, cred, p);
2346 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
2347 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2349 clp->nfsc_flags &= ~(NFSCLFLAGS_EXPIREIT | NFSCLFLAGS_RECVRINPROG);
2350 wakeup(&clp->nfsc_flags);
2351 nfsv4_unlock(&clp->nfsc_lock, 0);
2358 * This function inserts a lock in the list after insert_lop.
2361 nfscl_insertlock(struct nfscllockowner *lp, struct nfscllock *new_lop,
2362 struct nfscllock *insert_lop, int local)
2365 if ((struct nfscllockowner *)insert_lop == lp)
2366 LIST_INSERT_HEAD(&lp->nfsl_lock, new_lop, nfslo_list);
2368 LIST_INSERT_AFTER(insert_lop, new_lop, nfslo_list);
2370 nfsstatsv1.cllocallocks++;
2372 nfsstatsv1.cllocks++;
2376 * This function updates the locking for a lock owner and given file. It
2377 * maintains a list of lock ranges ordered on increasing file offset that
2378 * are NFSCLLOCK_READ or NFSCLLOCK_WRITE and non-overlapping (aka POSIX style).
2379 * It always adds new_lop to the list and sometimes uses the one pointed
2381 * Returns 1 if the locks were modified, 0 otherwise.
2384 nfscl_updatelock(struct nfscllockowner *lp, struct nfscllock **new_lopp,
2385 struct nfscllock **other_lopp, int local)
2387 struct nfscllock *new_lop = *new_lopp;
2388 struct nfscllock *lop, *tlop, *ilop;
2389 struct nfscllock *other_lop;
2390 int unlock = 0, modified = 0;
2394 * Work down the list until the lock is merged.
2396 if (new_lop->nfslo_type == F_UNLCK)
2398 ilop = (struct nfscllock *)lp;
2399 lop = LIST_FIRST(&lp->nfsl_lock);
2400 while (lop != NULL) {
2402 * Only check locks for this file that aren't before the start of
2405 if (lop->nfslo_end >= new_lop->nfslo_first) {
2406 if (new_lop->nfslo_end < lop->nfslo_first) {
2408 * If the new lock ends before the start of the
2409 * current lock's range, no merge, just insert
2414 if (new_lop->nfslo_type == lop->nfslo_type ||
2415 (new_lop->nfslo_first <= lop->nfslo_first &&
2416 new_lop->nfslo_end >= lop->nfslo_end)) {
2418 * This lock can be absorbed by the new lock/unlock.
2419 * This happens when it covers the entire range
2420 * of the old lock or is contiguous
2421 * with the old lock and is of the same type or an
2424 if (new_lop->nfslo_type != lop->nfslo_type ||
2425 new_lop->nfslo_first != lop->nfslo_first ||
2426 new_lop->nfslo_end != lop->nfslo_end)
2428 if (lop->nfslo_first < new_lop->nfslo_first)
2429 new_lop->nfslo_first = lop->nfslo_first;
2430 if (lop->nfslo_end > new_lop->nfslo_end)
2431 new_lop->nfslo_end = lop->nfslo_end;
2433 lop = LIST_NEXT(lop, nfslo_list);
2434 nfscl_freelock(tlop, local);
2439 * All these cases are for contiguous locks that are not the
2440 * same type, so they can't be merged.
2442 if (new_lop->nfslo_first <= lop->nfslo_first) {
2444 * This case is where the new lock overlaps with the
2445 * first part of the old lock. Move the start of the
2446 * old lock to just past the end of the new lock. The
2447 * new lock will be inserted in front of the old, since
2448 * ilop hasn't been updated. (We are done now.)
2450 if (lop->nfslo_first != new_lop->nfslo_end) {
2451 lop->nfslo_first = new_lop->nfslo_end;
2456 if (new_lop->nfslo_end >= lop->nfslo_end) {
2458 * This case is where the new lock overlaps with the
2459 * end of the old lock's range. Move the old lock's
2460 * end to just before the new lock's first and insert
2461 * the new lock after the old lock.
2462 * Might not be done yet, since the new lock could
2463 * overlap further locks with higher ranges.
2465 if (lop->nfslo_end != new_lop->nfslo_first) {
2466 lop->nfslo_end = new_lop->nfslo_first;
2470 lop = LIST_NEXT(lop, nfslo_list);
2474 * The final case is where the new lock's range is in the
2475 * middle of the current lock's and splits the current lock
2476 * up. Use *other_lopp to handle the second part of the
2477 * split old lock range. (We are done now.)
2478 * For unlock, we use new_lop as other_lop and tmp, since
2479 * other_lop and new_lop are the same for this case.
2480 * We noted the unlock case above, so we don't need
2481 * new_lop->nfslo_type any longer.
2483 tmp = new_lop->nfslo_first;
2485 other_lop = new_lop;
2488 other_lop = *other_lopp;
2491 other_lop->nfslo_first = new_lop->nfslo_end;
2492 other_lop->nfslo_end = lop->nfslo_end;
2493 other_lop->nfslo_type = lop->nfslo_type;
2494 lop->nfslo_end = tmp;
2495 nfscl_insertlock(lp, other_lop, lop, local);
2501 lop = LIST_NEXT(lop, nfslo_list);
2507 * Insert the new lock in the list at the appropriate place.
2510 nfscl_insertlock(lp, new_lop, ilop, local);
2518 * This function must be run as a kernel thread.
2519 * It does Renew Ops and recovery, when required.
2522 nfscl_renewthread(struct nfsclclient *clp, NFSPROC_T *p)
2524 struct nfsclowner *owp, *nowp;
2525 struct nfsclopen *op;
2526 struct nfscllockowner *lp, *nlp;
2527 struct nfscldeleghead dh;
2528 struct nfscldeleg *dp, *ndp;
2531 int error, cbpathdown, islept, igotlock, ret, clearok;
2532 uint32_t recover_done_time = 0;
2534 static time_t prevsec = 0;
2535 struct nfscllockownerfh *lfhp, *nlfhp;
2536 struct nfscllockownerfhhead lfh;
2537 struct nfscllayout *lyp, *nlyp;
2538 struct nfscldevinfo *dip, *ndip;
2539 struct nfscllayouthead rlh;
2540 struct nfsclrecalllayout *recallp;
2541 struct nfsclds *dsp;
2543 cred = newnfs_getcred();
2545 clp->nfsc_flags |= NFSCLFLAGS_HASTHREAD;
2548 newnfs_setroot(cred);
2550 if (clp->nfsc_flags & NFSCLFLAGS_RECOVER) {
2552 * Only allow one recover within 1/2 of the lease
2553 * duration (nfsc_renew).
2555 if (recover_done_time < NFSD_MONOSEC) {
2556 recover_done_time = NFSD_MONOSEC +
2558 NFSCL_DEBUG(1, "Doing recovery..\n");
2559 nfscl_recover(clp, cred, p);
2561 NFSCL_DEBUG(1, "Clear Recovery dt=%u ms=%jd\n",
2562 recover_done_time, (intmax_t)NFSD_MONOSEC);
2564 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2568 if (clp->nfsc_expire <= NFSD_MONOSEC &&
2569 (clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID)) {
2570 clp->nfsc_expire = NFSD_MONOSEC + clp->nfsc_renew;
2571 clidrev = clp->nfsc_clientidrev;
2572 error = nfsrpc_renew(clp, NULL, cred, p);
2573 if (error == NFSERR_CBPATHDOWN)
2575 else if (error == NFSERR_STALECLIENTID ||
2576 error == NFSERR_BADSESSION) {
2578 clp->nfsc_flags |= NFSCLFLAGS_RECOVER;
2580 } else if (error == NFSERR_EXPIRED)
2581 (void) nfscl_hasexpired(clp, clidrev, p);
2585 if (NFSHASNFSV4N(clp->nfsc_nmp)) {
2586 /* Do renews for any DS sessions. */
2587 NFSLOCKMNT(clp->nfsc_nmp);
2588 /* Skip first entry, since the MDS is handled above. */
2589 dsp = TAILQ_FIRST(&clp->nfsc_nmp->nm_sess);
2591 dsp = TAILQ_NEXT(dsp, nfsclds_list);
2592 while (dsp != NULL) {
2593 if (dsp->nfsclds_expire <= NFSD_MONOSEC &&
2594 dsp->nfsclds_sess.nfsess_defunct == 0) {
2595 dsp->nfsclds_expire = NFSD_MONOSEC +
2597 NFSUNLOCKMNT(clp->nfsc_nmp);
2598 (void)nfsrpc_renew(clp, dsp, cred, p);
2601 dsp = TAILQ_NEXT(dsp, nfsclds_list);
2603 NFSUNLOCKMNT(clp->nfsc_nmp);
2609 /* It's a Total Recall! */
2610 nfscl_totalrecall(clp);
2613 * Now, handle defunct owners.
2615 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
2616 if (LIST_EMPTY(&owp->nfsow_open)) {
2617 if (owp->nfsow_defunct != 0)
2618 nfscl_freeopenowner(owp, 0);
2623 * Do the recall on any delegations. To avoid trouble, always
2624 * come back up here after having slept.
2628 dp = TAILQ_FIRST(&clp->nfsc_deleg);
2629 while (dp != NULL) {
2630 ndp = TAILQ_NEXT(dp, nfsdl_list);
2631 if ((dp->nfsdl_flags & NFSCLDL_RECALL)) {
2633 * Wait for outstanding I/O ops to be done.
2635 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
2637 nfsv4_unlock(&clp->nfsc_lock, 0);
2640 dp->nfsdl_rwlock.nfslock_lock |=
2642 (void) nfsmsleep(&dp->nfsdl_rwlock,
2643 NFSCLSTATEMUTEXPTR, PZERO, "nfscld",
2648 igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
2649 &islept, NFSCLSTATEMUTEXPTR, NULL);
2654 newnfs_copycred(&dp->nfsdl_cred, cred);
2655 ret = nfscl_recalldeleg(clp, clp->nfsc_nmp, dp,
2658 nfscl_cleandeleg(dp);
2659 TAILQ_REMOVE(&clp->nfsc_deleg, dp,
2661 LIST_REMOVE(dp, nfsdl_hash);
2662 TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
2664 nfsstatsv1.cldelegates--;
2672 * Clear out old delegations, if we are above the high water
2673 * mark. Only clear out ones with no state related to them.
2674 * The tailq list is in LRU order.
2676 dp = TAILQ_LAST(&clp->nfsc_deleg, nfscldeleghead);
2677 while (nfscl_delegcnt > nfscl_deleghighwater && dp != NULL) {
2678 ndp = TAILQ_PREV(dp, nfscldeleghead, nfsdl_list);
2679 if (dp->nfsdl_rwlock.nfslock_usecnt == 0 &&
2680 dp->nfsdl_rwlock.nfslock_lock == 0 &&
2681 dp->nfsdl_timestamp < NFSD_MONOSEC &&
2682 (dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_ZAPPED |
2683 NFSCLDL_NEEDRECLAIM | NFSCLDL_DELEGRET)) == 0) {
2685 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
2686 op = LIST_FIRST(&owp->nfsow_open);
2693 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
2694 if (!LIST_EMPTY(&lp->nfsl_lock)) {
2701 TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
2702 LIST_REMOVE(dp, nfsdl_hash);
2703 TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
2705 nfsstatsv1.cldelegates--;
2711 nfsv4_unlock(&clp->nfsc_lock, 0);
2714 * Do the recall on any layouts. To avoid trouble, always
2715 * come back up here after having slept.
2719 TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp) {
2720 if ((lyp->nfsly_flags & NFSLY_RECALL) != 0) {
2722 * Wait for outstanding I/O ops to be done.
2724 if (lyp->nfsly_lock.nfslock_usecnt > 0 ||
2725 (lyp->nfsly_lock.nfslock_lock &
2726 NFSV4LOCK_LOCK) != 0) {
2727 lyp->nfsly_lock.nfslock_lock |=
2729 nfsmsleep(&lyp->nfsly_lock.nfslock_lock,
2730 NFSCLSTATEMUTEXPTR, PZERO, "nfslyp",
2734 /* Move the layout to the recall list. */
2735 TAILQ_REMOVE(&clp->nfsc_layout, lyp,
2737 LIST_REMOVE(lyp, nfsly_hash);
2738 TAILQ_INSERT_HEAD(&rlh, lyp, nfsly_list);
2740 /* Handle any layout commits. */
2741 if (!NFSHASNOLAYOUTCOMMIT(clp->nfsc_nmp) &&
2742 (lyp->nfsly_flags & NFSLY_WRITTEN) != 0) {
2743 lyp->nfsly_flags &= ~NFSLY_WRITTEN;
2745 NFSCL_DEBUG(3, "do layoutcommit\n");
2746 nfscl_dolayoutcommit(clp->nfsc_nmp, lyp,
2754 /* Now, look for stale layouts. */
2755 lyp = TAILQ_LAST(&clp->nfsc_layout, nfscllayouthead);
2756 while (lyp != NULL) {
2757 nlyp = TAILQ_PREV(lyp, nfscllayouthead, nfsly_list);
2758 if (lyp->nfsly_timestamp < NFSD_MONOSEC &&
2759 (lyp->nfsly_flags & NFSLY_RECALL) == 0 &&
2760 lyp->nfsly_lock.nfslock_usecnt == 0 &&
2761 lyp->nfsly_lock.nfslock_lock == 0) {
2762 NFSCL_DEBUG(4, "ret stale lay=%d\n",
2764 recallp = malloc(sizeof(*recallp),
2765 M_NFSLAYRECALL, M_NOWAIT);
2766 if (recallp == NULL)
2768 (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE,
2769 lyp, NFSLAYOUTIOMODE_ANY, 0, UINT64_MAX,
2770 lyp->nfsly_stateid.seqid, 0, 0, NULL,
2777 * Free up any unreferenced device info structures.
2779 LIST_FOREACH_SAFE(dip, &clp->nfsc_devinfo, nfsdi_list, ndip) {
2780 if (dip->nfsdi_layoutrefs == 0 &&
2781 dip->nfsdi_refcnt == 0) {
2782 NFSCL_DEBUG(4, "freeing devinfo\n");
2783 LIST_REMOVE(dip, nfsdi_list);
2784 nfscl_freedevinfo(dip);
2789 /* Do layout return(s), as required. */
2790 TAILQ_FOREACH_SAFE(lyp, &rlh, nfsly_list, nlyp) {
2791 TAILQ_REMOVE(&rlh, lyp, nfsly_list);
2792 NFSCL_DEBUG(4, "ret layout\n");
2793 nfscl_layoutreturn(clp->nfsc_nmp, lyp, cred, p);
2794 nfscl_freelayout(lyp);
2798 * Delegreturn any delegations cleaned out or recalled.
2800 TAILQ_FOREACH_SAFE(dp, &dh, nfsdl_list, ndp) {
2801 newnfs_copycred(&dp->nfsdl_cred, cred);
2802 (void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
2803 TAILQ_REMOVE(&dh, dp, nfsdl_list);
2804 free(dp, M_NFSCLDELEG);
2809 * Call nfscl_cleanupkext() once per second to check for
2810 * open/lock owners where the process has exited.
2812 mytime = NFSD_MONOSEC;
2813 if (prevsec != mytime) {
2815 nfscl_cleanupkext(clp, &lfh);
2819 * Do a ReleaseLockOwner for all lock owners where the
2820 * associated process no longer exists, as found by
2821 * nfscl_cleanupkext().
2823 newnfs_setroot(cred);
2824 SLIST_FOREACH_SAFE(lfhp, &lfh, nfslfh_list, nlfhp) {
2825 LIST_FOREACH_SAFE(lp, &lfhp->nfslfh_lock, nfsl_list,
2827 (void)nfsrpc_rellockown(clp->nfsc_nmp, lp,
2828 lfhp->nfslfh_fh, lfhp->nfslfh_len, cred,
2830 nfscl_freelockowner(lp, 0);
2837 if ((clp->nfsc_flags & NFSCLFLAGS_RECOVER) == 0)
2838 (void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT, "nfscl",
2840 if (clp->nfsc_flags & NFSCLFLAGS_UMOUNT) {
2841 clp->nfsc_flags &= ~NFSCLFLAGS_HASTHREAD;
2844 wakeup((caddr_t)clp);
2852 * Initiate state recovery. Called when NFSERR_STALECLIENTID,
2853 * NFSERR_STALESTATEID or NFSERR_BADSESSION is received.
2856 nfscl_initiate_recovery(struct nfsclclient *clp)
2862 clp->nfsc_flags |= NFSCLFLAGS_RECOVER;
2864 wakeup((caddr_t)clp);
2868 * Dump out the state stuff for debugging.
2871 nfscl_dumpstate(struct nfsmount *nmp, int openowner, int opens,
2872 int lockowner, int locks)
2874 struct nfsclclient *clp;
2875 struct nfsclowner *owp;
2876 struct nfsclopen *op;
2877 struct nfscllockowner *lp;
2878 struct nfscllock *lop;
2879 struct nfscldeleg *dp;
2883 printf("nfscl dumpstate NULL clp\n");
2887 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2888 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
2889 if (openowner && !LIST_EMPTY(&owp->nfsow_open))
2890 printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
2891 owp->nfsow_owner[0], owp->nfsow_owner[1],
2892 owp->nfsow_owner[2], owp->nfsow_owner[3],
2894 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
2896 printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
2897 op->nfso_stateid.other[0], op->nfso_stateid.other[1],
2898 op->nfso_stateid.other[2], op->nfso_opencnt,
2900 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
2902 printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
2903 lp->nfsl_owner[0], lp->nfsl_owner[1],
2904 lp->nfsl_owner[2], lp->nfsl_owner[3],
2906 lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
2907 lp->nfsl_stateid.other[2]);
2908 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
2911 printf("lck typ=%d fst=%ju end=%ju\n",
2912 lop->nfslo_type, (intmax_t)lop->nfslo_first,
2913 (intmax_t)lop->nfslo_end);
2915 printf("lck typ=%d fst=%qd end=%qd\n",
2916 lop->nfslo_type, lop->nfslo_first,
2924 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
2925 if (openowner && !LIST_EMPTY(&owp->nfsow_open))
2926 printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
2927 owp->nfsow_owner[0], owp->nfsow_owner[1],
2928 owp->nfsow_owner[2], owp->nfsow_owner[3],
2930 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
2932 printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
2933 op->nfso_stateid.other[0], op->nfso_stateid.other[1],
2934 op->nfso_stateid.other[2], op->nfso_opencnt,
2936 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
2938 printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
2939 lp->nfsl_owner[0], lp->nfsl_owner[1],
2940 lp->nfsl_owner[2], lp->nfsl_owner[3],
2942 lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
2943 lp->nfsl_stateid.other[2]);
2944 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
2947 printf("lck typ=%d fst=%ju end=%ju\n",
2948 lop->nfslo_type, (intmax_t)lop->nfslo_first,
2949 (intmax_t)lop->nfslo_end);
2951 printf("lck typ=%d fst=%qd end=%qd\n",
2952 lop->nfslo_type, lop->nfslo_first,
2963 * Check for duplicate open owners and opens.
2964 * (Only used as a diagnostic aid.)
2967 nfscl_dupopen(vnode_t vp, int dupopens)
2969 struct nfsclclient *clp;
2970 struct nfsclowner *owp, *owp2;
2971 struct nfsclopen *op, *op2;
2974 clp = VFSTONFS(vnode_mount(vp))->nm_clp;
2976 printf("nfscl dupopen NULL clp\n");
2979 nfhp = VTONFS(vp)->n_fhp;
2983 * First, search for duplicate owners.
2984 * These should never happen!
2986 LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
2987 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
2989 !NFSBCMP(owp->nfsow_owner, owp2->nfsow_owner,
2990 NFSV4CL_LOCKNAMELEN)) {
2992 printf("DUP OWNER\n");
2993 nfscl_dumpstate(VFSTONFS(vnode_mount(vp)), 1, 1, 0, 0);
3000 * Now, search for duplicate stateids.
3001 * These shouldn't happen, either.
3003 LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3004 LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
3005 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3006 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3008 (op->nfso_stateid.other[0] != 0 ||
3009 op->nfso_stateid.other[1] != 0 ||
3010 op->nfso_stateid.other[2] != 0) &&
3011 op->nfso_stateid.other[0] == op2->nfso_stateid.other[0] &&
3012 op->nfso_stateid.other[1] == op2->nfso_stateid.other[1] &&
3013 op->nfso_stateid.other[2] == op2->nfso_stateid.other[2]) {
3015 printf("DUP STATEID\n");
3016 nfscl_dumpstate(VFSTONFS(vnode_mount(vp)), 1, 1, 0,
3026 * Now search for duplicate opens.
3027 * Duplicate opens for the same owner
3028 * should never occur. Other duplicates are
3029 * possible and are checked for if "dupopens"
3032 LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3033 LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
3034 if (nfhp->nfh_len == op2->nfso_fhlen &&
3035 !NFSBCMP(nfhp->nfh_fh, op2->nfso_fh, nfhp->nfh_len)) {
3036 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3037 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3038 if (op != op2 && nfhp->nfh_len == op->nfso_fhlen &&
3039 !NFSBCMP(nfhp->nfh_fh, op->nfso_fh, nfhp->nfh_len) &&
3040 (!NFSBCMP(op->nfso_own->nfsow_owner,
3041 op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN) ||
3043 if (!NFSBCMP(op->nfso_own->nfsow_owner,
3044 op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
3046 printf("BADDUP OPEN\n");
3049 printf("DUP OPEN\n");
3051 nfscl_dumpstate(VFSTONFS(vnode_mount(vp)), 1, 1,
3064 * During close, find an open that needs to be dereferenced and
3065 * dereference it. If there are no more opens for this file,
3066 * log a message to that effect.
3067 * Opens aren't actually Close'd until VOP_INACTIVE() is performed
3068 * on the file's vnode.
3069 * This is the safe way, since it is difficult to identify
3070 * which open the close is for and I/O can be performed after the
3071 * close(2) system call when a file is mmap'd.
3072 * If it returns 0 for success, there will be a referenced
3073 * clp returned via clpp.
3076 nfscl_getclose(vnode_t vp, struct nfsclclient **clpp)
3078 struct nfsclclient *clp;
3079 struct nfsclowner *owp;
3080 struct nfsclopen *op;
3081 struct nfscldeleg *dp;
3085 error = nfscl_getcl(vnode_mount(vp), NULL, NULL, 1, &clp);
3090 nfhp = VTONFS(vp)->n_fhp;
3094 * First, look for one under a delegation that was locally issued
3095 * and just decrement the opencnt for it. Since all my Opens against
3096 * the server are DENY_NONE, I don't see a problem with hanging
3097 * onto them. (It is much easier to use one of the extant Opens
3098 * that I already have on the server when a Delegation is recalled
3099 * than to do fresh Opens.) Someday, I might need to rethink this, but.
3101 dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
3103 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
3104 op = LIST_FIRST(&owp->nfsow_open);
3107 * Since a delegation is for a file, there
3108 * should never be more than one open for
3111 if (LIST_NEXT(op, nfso_list) != NULL)
3112 panic("nfscdeleg opens");
3113 if (notdecr && op->nfso_opencnt > 0) {
3122 /* Now process the opens against the server. */
3123 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3124 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3125 if (op->nfso_fhlen == nfhp->nfh_len &&
3126 !NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
3128 /* Found an open, decrement cnt if possible */
3129 if (notdecr && op->nfso_opencnt > 0) {
3134 * There are more opens, so just return.
3136 if (op->nfso_opencnt > 0) {
3145 printf("nfscl: never fnd open\n");
3150 nfscl_doclose(vnode_t vp, struct nfsclclient **clpp, NFSPROC_T *p)
3152 struct nfsclclient *clp;
3153 struct nfsclowner *owp, *nowp;
3154 struct nfsclopen *op;
3155 struct nfscldeleg *dp;
3157 struct nfsclrecalllayout *recallp;
3160 error = nfscl_getcl(vnode_mount(vp), NULL, NULL, 1, &clp);
3165 nfhp = VTONFS(vp)->n_fhp;
3166 recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_WAITOK);
3169 * First get rid of the local Open structures, which should be no
3172 dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
3174 LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) {
3175 op = LIST_FIRST(&owp->nfsow_open);
3177 KASSERT((op->nfso_opencnt == 0),
3178 ("nfscl: bad open cnt on deleg"));
3179 nfscl_freeopen(op, 1);
3181 nfscl_freeopenowner(owp, 1);
3185 /* Return any layouts marked return on close. */
3186 nfscl_retoncloselayout(vp, clp, nfhp->nfh_fh, nfhp->nfh_len, &recallp);
3188 /* Now process the opens against the server. */
3190 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3191 op = LIST_FIRST(&owp->nfsow_open);
3192 while (op != NULL) {
3193 if (op->nfso_fhlen == nfhp->nfh_len &&
3194 !NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
3196 /* Found an open, close it. */
3198 KASSERT((op->nfso_opencnt == 0),
3199 ("nfscl: bad open cnt on server (%d)",
3203 nfsrpc_doclose(VFSTONFS(vnode_mount(vp)), op,
3208 op = LIST_NEXT(op, nfso_list);
3213 * recallp has been set NULL by nfscl_retoncloselayout() if it was
3214 * used by the function, but calling free() with a NULL pointer is ok.
3216 free(recallp, M_NFSLAYRECALL);
3221 * Return all delegations on this client.
3222 * (Must be called with client sleep lock.)
3225 nfscl_delegreturnall(struct nfsclclient *clp, NFSPROC_T *p)
3227 struct nfscldeleg *dp, *ndp;
3230 cred = newnfs_getcred();
3231 TAILQ_FOREACH_SAFE(dp, &clp->nfsc_deleg, nfsdl_list, ndp) {
3232 nfscl_cleandeleg(dp);
3233 (void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
3234 nfscl_freedeleg(&clp->nfsc_deleg, dp);
3240 * Do a callback RPC.
3243 nfscl_docb(struct nfsrv_descript *nd, NFSPROC_T *p)
3245 int clist, gotseq_ok, i, j, k, op, rcalls;
3247 struct nfsclclient *clp;
3248 struct nfscldeleg *dp = NULL;
3249 int numops, taglen = -1, error = 0, trunc __unused;
3250 u_int32_t minorvers = 0, retops = 0, *retopsp = NULL, *repp, cbident;
3251 u_char tag[NFSV4_SMALLSTR + 1], *tagstr;
3257 nfsattrbit_t attrbits, rattrbits;
3258 nfsv4stateid_t stateid;
3259 uint32_t seqid, slotid = 0, highslot, cachethis __unused;
3260 uint8_t sessionid[NFSX_V4SESSIONID];
3262 struct nfscllayout *lyp;
3263 uint64_t filesid[2], len, off;
3264 int changed, gotone, laytype, recalltype;
3266 struct nfsclrecalllayout *recallp = NULL;
3267 struct nfsclsession *tsep;
3271 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3272 taglen = fxdr_unsigned(int, *tl);
3277 if (taglen <= NFSV4_SMALLSTR)
3280 tagstr = malloc(taglen + 1, M_TEMP, M_WAITOK);
3281 error = nfsrv_mtostr(nd, tagstr, taglen);
3283 if (taglen > NFSV4_SMALLSTR)
3284 free(tagstr, M_TEMP);
3288 (void) nfsm_strtom(nd, tag, taglen);
3289 if (taglen > NFSV4_SMALLSTR) {
3290 free(tagstr, M_TEMP);
3292 NFSM_BUILD(retopsp, u_int32_t *, NFSX_UNSIGNED);
3293 NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3294 minorvers = fxdr_unsigned(u_int32_t, *tl++);
3295 if (minorvers != NFSV4_MINORVERSION && minorvers != NFSV41_MINORVERSION)
3296 nd->nd_repstat = NFSERR_MINORVERMISMATCH;
3297 cbident = fxdr_unsigned(u_int32_t, *tl++);
3301 numops = fxdr_unsigned(int, *tl);
3303 * Loop around doing the sub ops.
3305 for (i = 0; i < numops; i++) {
3306 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3307 NFSM_BUILD(repp, u_int32_t *, 2 * NFSX_UNSIGNED);
3309 op = fxdr_unsigned(int, *tl);
3310 if (op < NFSV4OP_CBGETATTR ||
3311 (op > NFSV4OP_CBRECALL && minorvers == NFSV4_MINORVERSION) ||
3312 (op > NFSV4OP_CBNOTIFYDEVID &&
3313 minorvers == NFSV41_MINORVERSION)) {
3314 nd->nd_repstat = NFSERR_OPILLEGAL;
3315 *repp = nfscl_errmap(nd, minorvers);
3319 nd->nd_procnum = op;
3320 if (op < NFSV41_CBNOPS)
3321 nfsstatsv1.cbrpccnt[nd->nd_procnum]++;
3323 case NFSV4OP_CBGETATTR:
3324 NFSCL_DEBUG(4, "cbgetattr\n");
3327 error = nfsm_getfh(nd, &nfhp);
3329 error = nfsrv_getattrbits(nd, &attrbits,
3331 if (error == 0 && i == 0 &&
3332 minorvers != NFSV4_MINORVERSION)
3333 error = NFSERR_OPNOTINSESS;
3335 mp = nfscl_getmnt(minorvers, sessionid, cbident,
3338 error = NFSERR_SERVERFAULT;
3341 error = nfscl_ngetreopen(mp, nfhp->nfh_fh,
3342 nfhp->nfh_len, p, &np);
3347 NFSZERO_ATTRBIT(&rattrbits);
3349 dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
3352 if (NFSISSET_ATTRBIT(&attrbits,
3355 va.va_size = np->n_size;
3359 NFSSETBIT_ATTRBIT(&rattrbits,
3362 if (NFSISSET_ATTRBIT(&attrbits,
3363 NFSATTRBIT_CHANGE)) {
3367 (np->n_flag & NDELEGMOD))
3369 NFSSETBIT_ATTRBIT(&rattrbits,
3373 error = NFSERR_SERVERFAULT;
3381 free(nfhp, M_NFSFH);
3383 (void) nfsv4_fillattr(nd, NULL, NULL, NULL, &va,
3384 NULL, 0, &rattrbits, NULL, p, 0, 0, 0, 0,
3387 case NFSV4OP_CBRECALL:
3388 NFSCL_DEBUG(4, "cbrecall\n");
3389 NFSM_DISSECT(tl, u_int32_t *, NFSX_STATEID +
3391 stateid.seqid = *tl++;
3392 NFSBCOPY((caddr_t)tl, (caddr_t)stateid.other,
3394 tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
3395 trunc = fxdr_unsigned(int, *tl);
3396 error = nfsm_getfh(nd, &nfhp);
3397 if (error == 0 && i == 0 &&
3398 minorvers != NFSV4_MINORVERSION)
3399 error = NFSERR_OPNOTINSESS;
3402 if (minorvers == NFSV4_MINORVERSION)
3403 clp = nfscl_getclnt(cbident);
3405 clp = nfscl_getclntsess(sessionid);
3407 dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
3409 if (dp != NULL && (dp->nfsdl_flags &
3410 NFSCLDL_DELEGRET) == 0) {
3413 wakeup((caddr_t)clp);
3416 error = NFSERR_SERVERFAULT;
3421 free(nfhp, M_NFSFH);
3423 case NFSV4OP_CBLAYOUTRECALL:
3424 NFSCL_DEBUG(4, "cblayrec\n");
3426 NFSM_DISSECT(tl, uint32_t *, 4 * NFSX_UNSIGNED);
3427 laytype = fxdr_unsigned(int, *tl++);
3428 iomode = fxdr_unsigned(uint32_t, *tl++);
3429 if (newnfs_true == *tl++)
3433 recalltype = fxdr_unsigned(int, *tl);
3434 NFSCL_DEBUG(4, "layt=%d iom=%d ch=%d rectyp=%d\n",
3435 laytype, iomode, changed, recalltype);
3436 recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL,
3438 if (laytype != NFSLAYOUT_NFSV4_1_FILES &&
3439 laytype != NFSLAYOUT_FLEXFILE)
3440 error = NFSERR_NOMATCHLAYOUT;
3441 else if (recalltype == NFSLAYOUTRETURN_FILE) {
3442 error = nfsm_getfh(nd, &nfhp);
3443 NFSCL_DEBUG(4, "retfile getfh=%d\n", error);
3446 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_HYPER +
3448 off = fxdr_hyper(tl); tl += 2;
3449 len = fxdr_hyper(tl); tl += 2;
3450 stateid.seqid = fxdr_unsigned(uint32_t, *tl++);
3451 NFSBCOPY(tl, stateid.other, NFSX_STATEIDOTHER);
3452 if (minorvers == NFSV4_MINORVERSION)
3453 error = NFSERR_NOTSUPP;
3455 error = NFSERR_OPNOTINSESS;
3456 NFSCL_DEBUG(4, "off=%ju len=%ju sq=%u err=%d\n",
3457 (uintmax_t)off, (uintmax_t)len,
3458 stateid.seqid, error);
3461 clp = nfscl_getclntsess(sessionid);
3462 NFSCL_DEBUG(4, "cbly clp=%p\n", clp);
3464 lyp = nfscl_findlayout(clp,
3467 NFSCL_DEBUG(4, "cblyp=%p\n",
3472 NFSLY_FLEXFILE)) != 0 &&
3473 !NFSBCMP(stateid.other,
3474 lyp->nfsly_stateid.other,
3475 NFSX_STATEIDOTHER)) {
3490 NFSERR_NOMATCHLAYOUT;
3492 error = NFSERR_NOMATCHLAYOUT;
3495 free(nfhp, M_NFSFH);
3496 } else if (recalltype == NFSLAYOUTRETURN_FSID) {
3497 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_HYPER);
3498 filesid[0] = fxdr_hyper(tl); tl += 2;
3499 filesid[1] = fxdr_hyper(tl); tl += 2;
3502 clp = nfscl_getclntsess(sessionid);
3504 TAILQ_FOREACH(lyp, &clp->nfsc_layout,
3506 if (lyp->nfsly_filesid[0] ==
3508 lyp->nfsly_filesid[1] ==
3515 lyp->nfsly_stateid.seqid,
3525 error = NFSERR_NOMATCHLAYOUT;
3527 error = NFSERR_NOMATCHLAYOUT;
3529 } else if (recalltype == NFSLAYOUTRETURN_ALL) {
3532 clp = nfscl_getclntsess(sessionid);
3534 TAILQ_FOREACH(lyp, &clp->nfsc_layout,
3536 error = nfscl_layoutrecall(
3537 recalltype, lyp, iomode, 0,
3539 lyp->nfsly_stateid.seqid,
3540 0, 0, NULL, recallp);
3547 error = NFSERR_NOMATCHLAYOUT;
3549 error = NFSERR_NOMATCHLAYOUT;
3552 error = NFSERR_NOMATCHLAYOUT;
3553 if (recallp != NULL) {
3554 free(recallp, M_NFSLAYRECALL);
3558 case NFSV4OP_CBSEQUENCE:
3559 NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID +
3561 bcopy(tl, sessionid, NFSX_V4SESSIONID);
3562 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3563 seqid = fxdr_unsigned(uint32_t, *tl++);
3564 slotid = fxdr_unsigned(uint32_t, *tl++);
3565 highslot = fxdr_unsigned(uint32_t, *tl++);
3567 /* Throw away the referring call stuff. */
3568 clist = fxdr_unsigned(int, *tl);
3569 for (j = 0; j < clist; j++) {
3570 NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID +
3572 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3573 rcalls = fxdr_unsigned(int, *tl);
3574 for (k = 0; k < rcalls; k++) {
3575 NFSM_DISSECT(tl, uint32_t *,
3581 clp = nfscl_getclntsess(sessionid);
3583 error = NFSERR_SERVERFAULT;
3585 error = NFSERR_SEQUENCEPOS;
3587 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3588 error = nfsv4_seqsession(seqid, slotid,
3589 highslot, tsep->nfsess_cbslots, &rep,
3590 tsep->nfsess_backslots);
3593 if (error == 0 || error == NFSERR_REPLYFROMCACHE) {
3597 * Handle a reply for a retried
3598 * callback. The reply will be
3599 * re-inserted in the session cache
3600 * by the nfsv4_seqsess_cacherep() call
3603 KASSERT(error == NFSERR_REPLYFROMCACHE,
3604 ("cbsequence: non-NULL rep"));
3605 NFSCL_DEBUG(4, "Got cbretry\n");
3606 m_freem(nd->nd_mreq);
3611 NFSM_BUILD(tl, uint32_t *,
3612 NFSX_V4SESSIONID + 4 * NFSX_UNSIGNED);
3613 bcopy(sessionid, tl, NFSX_V4SESSIONID);
3614 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3615 *tl++ = txdr_unsigned(seqid);
3616 *tl++ = txdr_unsigned(slotid);
3617 *tl++ = txdr_unsigned(NFSV4_CBSLOTS - 1);
3618 *tl = txdr_unsigned(NFSV4_CBSLOTS - 1);
3622 if (i == 0 && minorvers == NFSV41_MINORVERSION)
3623 error = NFSERR_OPNOTINSESS;
3625 NFSCL_DEBUG(1, "unsupp callback %d\n", op);
3626 error = NFSERR_NOTSUPP;
3631 if (error == EBADRPC || error == NFSERR_BADXDR) {
3632 nd->nd_repstat = NFSERR_BADXDR;
3634 nd->nd_repstat = error;
3639 if (nd->nd_repstat) {
3640 *repp = nfscl_errmap(nd, minorvers);
3643 *repp = 0; /* NFS4_OK */
3646 if (recallp != NULL)
3647 free(recallp, M_NFSLAYRECALL);
3649 if (error == EBADRPC || error == NFSERR_BADXDR)
3650 nd->nd_repstat = NFSERR_BADXDR;
3652 printf("nfsv4 comperr1=%d\n", error);
3655 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
3659 *retopsp = txdr_unsigned(retops);
3661 *nd->nd_errp = nfscl_errmap(nd, minorvers);
3663 if (gotseq_ok != 0) {
3664 rep = m_copym(nd->nd_mreq, 0, M_COPYALL, M_WAITOK);
3666 clp = nfscl_getclntsess(sessionid);
3668 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3669 nfsv4_seqsess_cacherep(slotid, tsep->nfsess_cbslots,
3680 * Generate the next cbident value. Basically just increment a static value
3681 * and then check that it isn't already in the list, if it has wrapped around.
3684 nfscl_nextcbident(void)
3686 struct nfsclclient *clp;
3688 static u_int32_t nextcbident = 0;
3689 static int haswrapped = 0;
3692 if (nextcbident == 0)
3696 * Search the clientid list for one already using this cbident.
3701 LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3702 if (clp->nfsc_cbident == nextcbident) {
3712 return (nextcbident);
3716 * Get the mount point related to a given cbident or session and busy it.
3719 nfscl_getmnt(int minorvers, uint8_t *sessionid, u_int32_t cbident,
3720 struct nfsclclient **clpp)
3722 struct nfsclclient *clp;
3725 struct nfsclsession *tsep;
3729 LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3730 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3731 if (minorvers == NFSV4_MINORVERSION) {
3732 if (clp->nfsc_cbident == cbident)
3734 } else if (!NFSBCMP(tsep->nfsess_sessionid, sessionid,
3742 mp = clp->nfsc_nmp->nm_mountp;
3745 error = vfs_busy(mp, 0);
3754 * Get the clientid pointer related to a given cbident.
3756 static struct nfsclclient *
3757 nfscl_getclnt(u_int32_t cbident)
3759 struct nfsclclient *clp;
3761 LIST_FOREACH(clp, &nfsclhead, nfsc_list)
3762 if (clp->nfsc_cbident == cbident)
3768 * Get the clientid pointer related to a given sessionid.
3770 static struct nfsclclient *
3771 nfscl_getclntsess(uint8_t *sessionid)
3773 struct nfsclclient *clp;
3774 struct nfsclsession *tsep;
3776 LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3777 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3778 if (!NFSBCMP(tsep->nfsess_sessionid, sessionid,
3786 * Search for a lock conflict locally on the client. A conflict occurs if
3787 * - not same owner and overlapping byte range and at least one of them is
3788 * a write lock or this is an unlock.
3791 nfscl_localconflict(struct nfsclclient *clp, u_int8_t *fhp, int fhlen,
3792 struct nfscllock *nlop, u_int8_t *own, struct nfscldeleg *dp,
3793 struct nfscllock **lopp)
3795 struct nfsclowner *owp;
3796 struct nfsclopen *op;
3800 ret = nfscl_checkconflict(&dp->nfsdl_lock, nlop, own, lopp);
3804 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3805 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3806 if (op->nfso_fhlen == fhlen &&
3807 !NFSBCMP(op->nfso_fh, fhp, fhlen)) {
3808 ret = nfscl_checkconflict(&op->nfso_lock, nlop,
3819 nfscl_checkconflict(struct nfscllockownerhead *lhp, struct nfscllock *nlop,
3820 u_int8_t *own, struct nfscllock **lopp)
3822 struct nfscllockowner *lp;
3823 struct nfscllock *lop;
3825 LIST_FOREACH(lp, lhp, nfsl_list) {
3826 if (NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
3827 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
3828 if (lop->nfslo_first >= nlop->nfslo_end)
3830 if (lop->nfslo_end <= nlop->nfslo_first)
3832 if (lop->nfslo_type == F_WRLCK ||
3833 nlop->nfslo_type == F_WRLCK ||
3834 nlop->nfslo_type == F_UNLCK) {
3837 return (NFSERR_DENIED);
3846 * Check for a local conflicting lock.
3849 nfscl_lockt(vnode_t vp, struct nfsclclient *clp, u_int64_t off,
3850 u_int64_t len, struct flock *fl, NFSPROC_T *p, void *id, int flags)
3852 struct nfscllock *lop, nlck;
3853 struct nfscldeleg *dp;
3855 u_int8_t own[NFSV4CL_LOCKNAMELEN];
3858 nlck.nfslo_type = fl->l_type;
3859 nlck.nfslo_first = off;
3860 if (len == NFS64BITSSET) {
3861 nlck.nfslo_end = NFS64BITSSET;
3863 nlck.nfslo_end = off + len;
3864 if (nlck.nfslo_end <= nlck.nfslo_first)
3865 return (NFSERR_INVAL);
3868 nfscl_filllockowner(id, own, flags);
3870 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
3871 error = nfscl_localconflict(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len,
3872 &nlck, own, dp, &lop);
3874 fl->l_whence = SEEK_SET;
3875 fl->l_start = lop->nfslo_first;
3876 if (lop->nfslo_end == NFS64BITSSET)
3879 fl->l_len = lop->nfslo_end - lop->nfslo_first;
3880 fl->l_pid = (pid_t)0;
3881 fl->l_type = lop->nfslo_type;
3882 error = -1; /* no RPC required */
3883 } else if (dp != NULL && ((dp->nfsdl_flags & NFSCLDL_WRITE) ||
3884 fl->l_type == F_RDLCK)) {
3886 * The delegation ensures that there isn't a conflicting
3887 * lock on the server, so return -1 to indicate an RPC
3890 fl->l_type = F_UNLCK;
3898 * Handle Recall of a delegation.
3899 * The clp must be exclusive locked when this is called.
3902 nfscl_recalldeleg(struct nfsclclient *clp, struct nfsmount *nmp,
3903 struct nfscldeleg *dp, vnode_t vp, struct ucred *cred, NFSPROC_T *p,
3904 int called_from_renewthread)
3906 struct nfsclowner *owp, *lowp, *nowp;
3907 struct nfsclopen *op, *lop;
3908 struct nfscllockowner *lp;
3909 struct nfscllock *lckp;
3911 int error = 0, ret, gotvp = 0;
3915 * First, get a vnode for the file. This is needed to do RPCs.
3917 ret = nfscl_ngetreopen(nmp->nm_mountp, dp->nfsdl_fh,
3918 dp->nfsdl_fhlen, p, &np);
3921 * File isn't open, so nothing to move over to the
3931 dp->nfsdl_flags &= ~NFSCLDL_MODTIMESET;
3934 * Ok, if it's a write delegation, flush data to the server, so
3935 * that close/open consistency is retained.
3939 if ((dp->nfsdl_flags & NFSCLDL_WRITE) && (np->n_flag & NMODIFIED)) {
3940 np->n_flag |= NDELEGRECALL;
3942 ret = ncl_flush(vp, MNT_WAIT, p, 1, called_from_renewthread);
3944 np->n_flag &= ~NDELEGRECALL;
3946 NFSINVALATTRCACHE(np);
3948 if (ret == EIO && called_from_renewthread != 0) {
3950 * If the flush failed with EIO for the renew thread,
3951 * return now, so that the dirty buffer will be flushed
3960 * Now, for each openowner with opens issued locally, move them
3961 * over to state against the server.
3963 LIST_FOREACH(lowp, &dp->nfsdl_owner, nfsow_list) {
3964 lop = LIST_FIRST(&lowp->nfsow_open);
3966 if (LIST_NEXT(lop, nfso_list) != NULL)
3967 panic("nfsdlg mult opens");
3969 * Look for the same openowner against the server.
3971 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3972 if (!NFSBCMP(lowp->nfsow_owner,
3973 owp->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
3974 newnfs_copycred(&dp->nfsdl_cred, cred);
3975 ret = nfscl_moveopen(vp, clp, nmp, lop,
3977 if (ret == NFSERR_STALECLIENTID ||
3978 ret == NFSERR_STALEDONTRECOVER ||
3979 ret == NFSERR_BADSESSION) {
3985 nfscl_freeopen(lop, 1);
3994 * If no openowner found, create one and get an open
3999 sizeof (struct nfsclowner), M_NFSCLOWNER,
4001 nfscl_newopen(clp, NULL, &owp, &nowp, &op,
4002 NULL, lowp->nfsow_owner, dp->nfsdl_fh,
4003 dp->nfsdl_fhlen, NULL, NULL);
4004 newnfs_copycred(&dp->nfsdl_cred, cred);
4005 ret = nfscl_moveopen(vp, clp, nmp, lop,
4008 nfscl_freeopenowner(owp, 0);
4009 if (ret == NFSERR_STALECLIENTID ||
4010 ret == NFSERR_STALEDONTRECOVER ||
4011 ret == NFSERR_BADSESSION) {
4017 nfscl_freeopen(lop, 1);
4027 * Now, get byte range locks for any locks done locally.
4029 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4030 LIST_FOREACH(lckp, &lp->nfsl_lock, nfslo_list) {
4031 newnfs_copycred(&dp->nfsdl_cred, cred);
4032 ret = nfscl_relock(vp, clp, nmp, lp, lckp, cred, p);
4033 if (ret == NFSERR_STALESTATEID ||
4034 ret == NFSERR_STALEDONTRECOVER ||
4035 ret == NFSERR_STALECLIENTID ||
4036 ret == NFSERR_BADSESSION) {
4051 * Move a locally issued open over to an owner on the state list.
4052 * SIDE EFFECT: If it needs to sleep (do an rpc), it unlocks clstate and
4053 * returns with it unlocked.
4056 nfscl_moveopen(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp,
4057 struct nfsclopen *lop, struct nfsclowner *owp, struct nfscldeleg *dp,
4058 struct ucred *cred, NFSPROC_T *p)
4060 struct nfsclopen *op, *nop;
4061 struct nfscldeleg *ndp;
4063 int error = 0, newone;
4066 * First, look for an appropriate open, If found, just increment the
4069 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
4070 if ((op->nfso_mode & lop->nfso_mode) == lop->nfso_mode &&
4071 op->nfso_fhlen == lop->nfso_fhlen &&
4072 !NFSBCMP(op->nfso_fh, lop->nfso_fh, op->nfso_fhlen)) {
4073 op->nfso_opencnt += lop->nfso_opencnt;
4074 nfscl_freeopen(lop, 1);
4079 /* No appropriate open, so we have to do one against the server. */
4081 nop = malloc(sizeof (struct nfsclopen) +
4082 lop->nfso_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
4084 nfscl_newopen(clp, NULL, &owp, NULL, &op, &nop, owp->nfsow_owner,
4085 lop->nfso_fh, lop->nfso_fhlen, cred, &newone);
4087 error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data, np->n_v4->n4_fhlen,
4088 lop->nfso_fh, lop->nfso_fhlen, lop->nfso_mode, op,
4089 NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, &ndp, 0, 0, cred, p);
4092 nfscl_freeopen(op, 0);
4094 op->nfso_mode |= lop->nfso_mode;
4095 op->nfso_opencnt += lop->nfso_opencnt;
4096 nfscl_freeopen(lop, 1);
4099 free(nop, M_NFSCLOPEN);
4102 * What should I do with the returned delegation, since the
4103 * delegation is being recalled? For now, just printf and
4106 printf("Moveopen returned deleg\n");
4107 free(ndp, M_NFSCLDELEG);
4113 * Recall all delegations on this client.
4116 nfscl_totalrecall(struct nfsclclient *clp)
4118 struct nfscldeleg *dp;
4120 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
4121 if ((dp->nfsdl_flags & NFSCLDL_DELEGRET) == 0)
4122 dp->nfsdl_flags |= NFSCLDL_RECALL;
4127 * Relock byte ranges. Called for delegation recall and state expiry.
4130 nfscl_relock(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp,
4131 struct nfscllockowner *lp, struct nfscllock *lop, struct ucred *cred,
4134 struct nfscllockowner *nlp;
4137 int error, newone, donelocally;
4139 off = lop->nfslo_first;
4140 len = lop->nfslo_end - lop->nfslo_first;
4141 error = nfscl_getbytelock(vp, off, len, lop->nfslo_type, cred, p,
4142 clp, 1, NULL, lp->nfsl_lockflags, lp->nfsl_owner,
4143 lp->nfsl_openowner, &nlp, &newone, &donelocally);
4144 if (error || donelocally)
4146 nfhp = VTONFS(vp)->n_fhp;
4147 error = nfscl_trylock(nmp, vp, nfhp->nfh_fh,
4148 nfhp->nfh_len, nlp, newone, 0, off,
4149 len, lop->nfslo_type, cred, p);
4151 nfscl_freelockowner(nlp, 0);
4156 * Called to re-open a file. Basically get a vnode for the file handle
4157 * and then call nfsrpc_openrpc() to do the rest.
4160 nfsrpc_reopen(struct nfsmount *nmp, u_int8_t *fhp, int fhlen,
4161 u_int32_t mode, struct nfsclopen *op, struct nfscldeleg **dpp,
4162 struct ucred *cred, NFSPROC_T *p)
4168 error = nfscl_ngetreopen(nmp->nm_mountp, fhp, fhlen, p, &np);
4172 if (np->n_v4 != NULL) {
4173 error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data,
4174 np->n_v4->n4_fhlen, fhp, fhlen, mode, op,
4175 NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, dpp, 0, 0,
4185 * Try an open against the server. Just call nfsrpc_openrpc(), retrying while
4186 * NFSERR_DELAY. Also, try system credentials, if the passed in credentials
4190 nfscl_tryopen(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen,
4191 u_int8_t *newfhp, int newfhlen, u_int32_t mode, struct nfsclopen *op,
4192 u_int8_t *name, int namelen, struct nfscldeleg **ndpp,
4193 int reclaim, u_int32_t delegtype, struct ucred *cred, NFSPROC_T *p)
4198 error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp, newfhlen,
4199 mode, op, name, namelen, ndpp, reclaim, delegtype, cred, p,
4201 if (error == NFSERR_DELAY)
4202 (void) nfs_catnap(PZERO, error, "nfstryop");
4203 } while (error == NFSERR_DELAY);
4204 if (error == EAUTH || error == EACCES) {
4205 /* Try again using system credentials */
4206 newnfs_setroot(cred);
4208 error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp,
4209 newfhlen, mode, op, name, namelen, ndpp, reclaim,
4210 delegtype, cred, p, 1, 0);
4211 if (error == NFSERR_DELAY)
4212 (void) nfs_catnap(PZERO, error, "nfstryop");
4213 } while (error == NFSERR_DELAY);
4219 * Try a byte range lock. Just loop on nfsrpc_lock() while it returns
4220 * NFSERR_DELAY. Also, retry with system credentials, if the provided
4224 nfscl_trylock(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp,
4225 int fhlen, struct nfscllockowner *nlp, int newone, int reclaim,
4226 u_int64_t off, u_int64_t len, short type, struct ucred *cred, NFSPROC_T *p)
4228 struct nfsrv_descript nfsd, *nd = &nfsd;
4232 error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp, newone,
4233 reclaim, off, len, type, cred, p, 0);
4234 if (!error && nd->nd_repstat == NFSERR_DELAY)
4235 (void) nfs_catnap(PZERO, (int)nd->nd_repstat,
4237 } while (!error && nd->nd_repstat == NFSERR_DELAY);
4239 error = nd->nd_repstat;
4240 if (error == EAUTH || error == EACCES) {
4241 /* Try again using root credentials */
4242 newnfs_setroot(cred);
4244 error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp,
4245 newone, reclaim, off, len, type, cred, p, 1);
4246 if (!error && nd->nd_repstat == NFSERR_DELAY)
4247 (void) nfs_catnap(PZERO, (int)nd->nd_repstat,
4249 } while (!error && nd->nd_repstat == NFSERR_DELAY);
4251 error = nd->nd_repstat;
4257 * Try a delegreturn against the server. Just call nfsrpc_delegreturn(),
4258 * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
4262 nfscl_trydelegreturn(struct nfscldeleg *dp, struct ucred *cred,
4263 struct nfsmount *nmp, NFSPROC_T *p)
4268 error = nfsrpc_delegreturn(dp, cred, nmp, p, 0);
4269 if (error == NFSERR_DELAY)
4270 (void) nfs_catnap(PZERO, error, "nfstrydp");
4271 } while (error == NFSERR_DELAY);
4272 if (error == EAUTH || error == EACCES) {
4273 /* Try again using system credentials */
4274 newnfs_setroot(cred);
4276 error = nfsrpc_delegreturn(dp, cred, nmp, p, 1);
4277 if (error == NFSERR_DELAY)
4278 (void) nfs_catnap(PZERO, error, "nfstrydp");
4279 } while (error == NFSERR_DELAY);
4285 * Try a close against the server. Just call nfsrpc_closerpc(),
4286 * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
4290 nfscl_tryclose(struct nfsclopen *op, struct ucred *cred,
4291 struct nfsmount *nmp, NFSPROC_T *p)
4293 struct nfsrv_descript nfsd, *nd = &nfsd;
4297 error = nfsrpc_closerpc(nd, nmp, op, cred, p, 0);
4298 if (error == NFSERR_DELAY)
4299 (void) nfs_catnap(PZERO, error, "nfstrycl");
4300 } while (error == NFSERR_DELAY);
4301 if (error == EAUTH || error == EACCES) {
4302 /* Try again using system credentials */
4303 newnfs_setroot(cred);
4305 error = nfsrpc_closerpc(nd, nmp, op, cred, p, 1);
4306 if (error == NFSERR_DELAY)
4307 (void) nfs_catnap(PZERO, error, "nfstrycl");
4308 } while (error == NFSERR_DELAY);
4314 * Decide if a delegation on a file permits close without flushing writes
4315 * to the server. This might be a big performance win in some environments.
4316 * (Not useful until the client does caching on local stable storage.)
4319 nfscl_mustflush(vnode_t vp)
4321 struct nfsclclient *clp;
4322 struct nfscldeleg *dp;
4324 struct nfsmount *nmp;
4327 nmp = VFSTONFS(vnode_mount(vp));
4328 if (!NFSHASNFSV4(nmp))
4331 clp = nfscl_findcl(nmp);
4336 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4337 if (dp != NULL && (dp->nfsdl_flags &
4338 (NFSCLDL_WRITE | NFSCLDL_RECALL | NFSCLDL_DELEGRET)) ==
4340 (dp->nfsdl_sizelimit >= np->n_size ||
4341 !NFSHASSTRICT3530(nmp))) {
4350 * See if a (write) delegation exists for this file.
4353 nfscl_nodeleg(vnode_t vp, int writedeleg)
4355 struct nfsclclient *clp;
4356 struct nfscldeleg *dp;
4358 struct nfsmount *nmp;
4361 nmp = VFSTONFS(vnode_mount(vp));
4362 if (!NFSHASNFSV4(nmp))
4365 clp = nfscl_findcl(nmp);
4370 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4372 (dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_DELEGRET)) == 0 &&
4373 (writedeleg == 0 || (dp->nfsdl_flags & NFSCLDL_WRITE) ==
4383 * Look for an associated delegation that should be DelegReturned.
4386 nfscl_removedeleg(vnode_t vp, NFSPROC_T *p, nfsv4stateid_t *stp)
4388 struct nfsclclient *clp;
4389 struct nfscldeleg *dp;
4390 struct nfsclowner *owp;
4391 struct nfscllockowner *lp;
4392 struct nfsmount *nmp;
4395 int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
4397 nmp = VFSTONFS(vnode_mount(vp));
4401 * Loop around waiting for:
4402 * - outstanding I/O operations on delegations to complete
4403 * - for a delegation on vp that has state, lock the client and
4405 * - return delegation with no state
4408 clp = nfscl_findcl(nmp);
4413 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4414 np->n_fhp->nfh_len);
4417 * Wait for outstanding I/O ops to be done.
4419 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4421 nfsv4_unlock(&clp->nfsc_lock, 0);
4424 dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4425 (void) nfsmsleep(&dp->nfsdl_rwlock,
4426 NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4430 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4431 if (!LIST_EMPTY(&owp->nfsow_open)) {
4437 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4438 if (!LIST_EMPTY(&lp->nfsl_lock)) {
4444 if (needsrecall && !triedrecall) {
4445 dp->nfsdl_flags |= NFSCLDL_DELEGRET;
4448 igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
4449 &islept, NFSCLSTATEMUTEXPTR, NULL);
4456 cred = newnfs_getcred();
4457 newnfs_copycred(&dp->nfsdl_cred, cred);
4458 (void) nfscl_recalldeleg(clp, nmp, dp, vp, cred, p, 0);
4462 nfsv4_unlock(&clp->nfsc_lock, 0);
4466 *stp = dp->nfsdl_stateid;
4468 nfscl_cleandeleg(dp);
4469 nfscl_freedeleg(&clp->nfsc_deleg, dp);
4472 nfsv4_unlock(&clp->nfsc_lock, 0);
4479 * Look for associated delegation(s) that should be DelegReturned.
4482 nfscl_renamedeleg(vnode_t fvp, nfsv4stateid_t *fstp, int *gotfdp, vnode_t tvp,
4483 nfsv4stateid_t *tstp, int *gottdp, NFSPROC_T *p)
4485 struct nfsclclient *clp;
4486 struct nfscldeleg *dp;
4487 struct nfsclowner *owp;
4488 struct nfscllockowner *lp;
4489 struct nfsmount *nmp;
4492 int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
4494 nmp = VFSTONFS(vnode_mount(fvp));
4499 * Loop around waiting for:
4500 * - outstanding I/O operations on delegations to complete
4501 * - for a delegation on fvp that has state, lock the client and
4503 * - return delegation(s) with no state.
4506 clp = nfscl_findcl(nmp);
4512 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4513 np->n_fhp->nfh_len);
4514 if (dp != NULL && *gotfdp == 0) {
4516 * Wait for outstanding I/O ops to be done.
4518 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4520 nfsv4_unlock(&clp->nfsc_lock, 0);
4523 dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4524 (void) nfsmsleep(&dp->nfsdl_rwlock,
4525 NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4529 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4530 if (!LIST_EMPTY(&owp->nfsow_open)) {
4536 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4537 if (!LIST_EMPTY(&lp->nfsl_lock)) {
4543 if (needsrecall && !triedrecall) {
4544 dp->nfsdl_flags |= NFSCLDL_DELEGRET;
4547 igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
4548 &islept, NFSCLSTATEMUTEXPTR, NULL);
4555 cred = newnfs_getcred();
4556 newnfs_copycred(&dp->nfsdl_cred, cred);
4557 (void) nfscl_recalldeleg(clp, nmp, dp, fvp, cred, p, 0);
4561 nfsv4_unlock(&clp->nfsc_lock, 0);
4565 *fstp = dp->nfsdl_stateid;
4568 nfscl_cleandeleg(dp);
4569 nfscl_freedeleg(&clp->nfsc_deleg, dp);
4572 nfsv4_unlock(&clp->nfsc_lock, 0);
4577 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4578 np->n_fhp->nfh_len);
4579 if (dp != NULL && *gottdp == 0) {
4581 * Wait for outstanding I/O ops to be done.
4583 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4584 dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4585 (void) nfsmsleep(&dp->nfsdl_rwlock,
4586 NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4589 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4590 if (!LIST_EMPTY(&owp->nfsow_open)) {
4595 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4596 if (!LIST_EMPTY(&lp->nfsl_lock)) {
4601 *tstp = dp->nfsdl_stateid;
4604 nfscl_cleandeleg(dp);
4605 nfscl_freedeleg(&clp->nfsc_deleg, dp);
4614 * Get a reference on the clientid associated with the mount point.
4615 * Return 1 if success, 0 otherwise.
4618 nfscl_getref(struct nfsmount *nmp)
4620 struct nfsclclient *clp;
4623 clp = nfscl_findcl(nmp);
4628 nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, NULL);
4634 * Release a reference on a clientid acquired with the above call.
4637 nfscl_relref(struct nfsmount *nmp)
4639 struct nfsclclient *clp;
4642 clp = nfscl_findcl(nmp);
4647 nfsv4_relref(&clp->nfsc_lock);
4652 * Save the size attribute in the delegation, since the nfsnode
4656 nfscl_reclaimnode(vnode_t vp)
4658 struct nfsclclient *clp;
4659 struct nfscldeleg *dp;
4660 struct nfsnode *np = VTONFS(vp);
4661 struct nfsmount *nmp;
4663 nmp = VFSTONFS(vnode_mount(vp));
4664 if (!NFSHASNFSV4(nmp))
4667 clp = nfscl_findcl(nmp);
4672 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4673 if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE))
4674 dp->nfsdl_size = np->n_size;
4679 * Get the saved size attribute in the delegation, since it is a
4680 * newly allocated nfsnode.
4683 nfscl_newnode(vnode_t vp)
4685 struct nfsclclient *clp;
4686 struct nfscldeleg *dp;
4687 struct nfsnode *np = VTONFS(vp);
4688 struct nfsmount *nmp;
4690 nmp = VFSTONFS(vnode_mount(vp));
4691 if (!NFSHASNFSV4(nmp))
4694 clp = nfscl_findcl(nmp);
4699 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4700 if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE))
4701 np->n_size = dp->nfsdl_size;
4706 * If there is a valid write delegation for this file, set the modtime
4707 * to the local clock time.
4710 nfscl_delegmodtime(vnode_t vp)
4712 struct nfsclclient *clp;
4713 struct nfscldeleg *dp;
4714 struct nfsnode *np = VTONFS(vp);
4715 struct nfsmount *nmp;
4717 nmp = VFSTONFS(vnode_mount(vp));
4718 if (!NFSHASNFSV4(nmp))
4721 clp = nfscl_findcl(nmp);
4726 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4727 if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE)) {
4728 nanotime(&dp->nfsdl_modtime);
4729 dp->nfsdl_flags |= NFSCLDL_MODTIMESET;
4735 * If there is a valid write delegation for this file with a modtime set,
4736 * put that modtime in mtime.
4739 nfscl_deleggetmodtime(vnode_t vp, struct timespec *mtime)
4741 struct nfsclclient *clp;
4742 struct nfscldeleg *dp;
4743 struct nfsnode *np = VTONFS(vp);
4744 struct nfsmount *nmp;
4746 nmp = VFSTONFS(vnode_mount(vp));
4747 if (!NFSHASNFSV4(nmp))
4750 clp = nfscl_findcl(nmp);
4755 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4757 (dp->nfsdl_flags & (NFSCLDL_WRITE | NFSCLDL_MODTIMESET)) ==
4758 (NFSCLDL_WRITE | NFSCLDL_MODTIMESET))
4759 *mtime = dp->nfsdl_modtime;
4764 nfscl_errmap(struct nfsrv_descript *nd, u_int32_t minorvers)
4766 short *defaulterrp, *errp;
4768 if (!nd->nd_repstat)
4770 if (nd->nd_procnum == NFSPROC_NOOP)
4771 return (txdr_unsigned(nd->nd_repstat & 0xffff));
4772 if (nd->nd_repstat == EBADRPC)
4773 return (txdr_unsigned(NFSERR_BADXDR));
4774 if (nd->nd_repstat == NFSERR_MINORVERMISMATCH ||
4775 nd->nd_repstat == NFSERR_OPILLEGAL)
4776 return (txdr_unsigned(nd->nd_repstat));
4777 if (nd->nd_repstat >= NFSERR_BADIOMODE && nd->nd_repstat < 20000 &&
4778 minorvers > NFSV4_MINORVERSION) {
4779 /* NFSv4.n error. */
4780 return (txdr_unsigned(nd->nd_repstat));
4782 if (nd->nd_procnum < NFSV4OP_CBNOPS)
4783 errp = defaulterrp = nfscl_cberrmap[nd->nd_procnum];
4785 return (txdr_unsigned(nd->nd_repstat));
4787 if (*errp == (short)nd->nd_repstat)
4788 return (txdr_unsigned(nd->nd_repstat));
4789 return (txdr_unsigned(*defaulterrp));
4793 * Called to find/add a layout to a client.
4794 * This function returns the layout with a refcnt (shared lock) upon
4795 * success (returns 0) or with no lock/refcnt on the layout when an
4796 * error is returned.
4797 * If a layout is passed in via lypp, it is locked (exclusively locked).
4800 nfscl_layout(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen,
4801 nfsv4stateid_t *stateidp, int layouttype, int retonclose,
4802 struct nfsclflayouthead *fhlp, struct nfscllayout **lypp,
4803 struct ucred *cred, NFSPROC_T *p)
4805 struct nfsclclient *clp;
4806 struct nfscllayout *lyp, *tlyp;
4807 struct nfsclflayout *flp;
4808 struct nfsnode *np = VTONFS(vp);
4810 int layout_passed_in;
4812 mp = nmp->nm_mountp;
4813 layout_passed_in = 1;
4817 layout_passed_in = 0;
4818 tlyp = malloc(sizeof(*tlyp) + fhlen - 1, M_NFSLAYOUT,
4825 if (layout_passed_in != 0)
4826 nfsv4_unlock(&lyp->nfsly_lock, 0);
4829 free(tlyp, M_NFSLAYOUT);
4834 * Although no lyp was passed in, another thread might have
4835 * allocated one. If one is found, just increment it's ref
4836 * count and return it.
4838 lyp = nfscl_findlayout(clp, fhp, fhlen);
4842 lyp->nfsly_stateid.seqid = stateidp->seqid;
4843 lyp->nfsly_stateid.other[0] = stateidp->other[0];
4844 lyp->nfsly_stateid.other[1] = stateidp->other[1];
4845 lyp->nfsly_stateid.other[2] = stateidp->other[2];
4846 lyp->nfsly_lastbyte = 0;
4847 LIST_INIT(&lyp->nfsly_flayread);
4848 LIST_INIT(&lyp->nfsly_flayrw);
4849 LIST_INIT(&lyp->nfsly_recall);
4850 lyp->nfsly_filesid[0] = np->n_vattr.na_filesid[0];
4851 lyp->nfsly_filesid[1] = np->n_vattr.na_filesid[1];
4852 lyp->nfsly_clp = clp;
4853 if (layouttype == NFSLAYOUT_FLEXFILE)
4854 lyp->nfsly_flags = NFSLY_FLEXFILE;
4856 lyp->nfsly_flags = NFSLY_FILES;
4857 if (retonclose != 0)
4858 lyp->nfsly_flags |= NFSLY_RETONCLOSE;
4859 lyp->nfsly_fhlen = fhlen;
4860 NFSBCOPY(fhp, lyp->nfsly_fh, fhlen);
4861 TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
4862 LIST_INSERT_HEAD(NFSCLLAYOUTHASH(clp, fhp, fhlen), lyp,
4864 lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
4867 if (retonclose != 0)
4868 lyp->nfsly_flags |= NFSLY_RETONCLOSE;
4869 TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list);
4870 TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
4871 lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
4873 nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
4874 if (NFSCL_FORCEDISM(mp)) {
4877 free(tlyp, M_NFSLAYOUT);
4882 lyp->nfsly_stateid.seqid = stateidp->seqid;
4884 /* Merge the new list of File Layouts into the list. */
4885 flp = LIST_FIRST(fhlp);
4887 if (flp->nfsfl_iomode == NFSLAYOUTIOMODE_READ)
4888 nfscl_mergeflayouts(&lyp->nfsly_flayread, fhlp);
4890 nfscl_mergeflayouts(&lyp->nfsly_flayrw, fhlp);
4892 if (layout_passed_in != 0)
4893 nfsv4_unlock(&lyp->nfsly_lock, 1);
4896 free(tlyp, M_NFSLAYOUT);
4901 * Search for a layout by MDS file handle.
4902 * If one is found, it is returned with a refcnt (shared lock) iff
4903 * retflpp returned non-NULL and locked (exclusive locked) iff retflpp is
4906 struct nfscllayout *
4907 nfscl_getlayout(struct nfsclclient *clp, uint8_t *fhp, int fhlen,
4908 uint64_t off, struct nfsclflayout **retflpp, int *recalledp)
4910 struct nfscllayout *lyp;
4912 int error, igotlock;
4914 mp = clp->nfsc_nmp->nm_mountp;
4918 lyp = nfscl_findlayout(clp, fhp, fhlen);
4920 if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
4921 TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list);
4922 TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
4923 lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
4924 error = nfscl_findlayoutforio(lyp, off,
4925 NFSV4OPEN_ACCESSREAD, retflpp);
4927 nfsv4_getref(&lyp->nfsly_lock, NULL,
4928 NFSCLSTATEMUTEXPTR, mp);
4931 igotlock = nfsv4_lock(&lyp->nfsly_lock,
4932 1, NULL, NFSCLSTATEMUTEXPTR, mp);
4933 } while (igotlock == 0 && !NFSCL_FORCEDISM(mp));
4936 if (NFSCL_FORCEDISM(mp)) {
4950 * Search for a layout by MDS file handle. If one is found, mark in to be
4951 * recalled, if it already marked "return on close".
4954 nfscl_retoncloselayout(vnode_t vp, struct nfsclclient *clp, uint8_t *fhp,
4955 int fhlen, struct nfsclrecalllayout **recallpp)
4957 struct nfscllayout *lyp;
4960 if (vp->v_type != VREG || !NFSHASPNFS(VFSTONFS(vnode_mount(vp))) ||
4961 nfscl_enablecallb == 0 || nfs_numnfscbd == 0 ||
4962 (VTONFS(vp)->n_flag & NNOLAYOUT) != 0)
4964 lyp = nfscl_findlayout(clp, fhp, fhlen);
4965 if (lyp != NULL && (lyp->nfsly_flags & (NFSLY_RETONCLOSE |
4966 NFSLY_RECALL)) == NFSLY_RETONCLOSE) {
4968 if (!LIST_EMPTY(&lyp->nfsly_flayread))
4969 iomode |= NFSLAYOUTIOMODE_READ;
4970 if (!LIST_EMPTY(&lyp->nfsly_flayrw))
4971 iomode |= NFSLAYOUTIOMODE_RW;
4972 (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE, lyp, iomode,
4973 0, UINT64_MAX, lyp->nfsly_stateid.seqid, 0, 0, NULL,
4975 NFSCL_DEBUG(4, "retoncls recall iomode=%d\n", iomode);
4981 * Mark the layout to be recalled and with an error.
4982 * Also, disable the dsp from further use.
4985 nfscl_dserr(uint32_t op, uint32_t stat, struct nfscldevinfo *dp,
4986 struct nfscllayout *lyp, struct nfsclds *dsp)
4988 struct nfsclrecalllayout *recallp;
4991 printf("DS being disabled, error=%d\n", stat);
4992 /* Set up the return of the layout. */
4993 recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_WAITOK);
4996 if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
4997 if (!LIST_EMPTY(&lyp->nfsly_flayread))
4998 iomode |= NFSLAYOUTIOMODE_READ;
4999 if (!LIST_EMPTY(&lyp->nfsly_flayrw))
5000 iomode |= NFSLAYOUTIOMODE_RW;
5001 (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE, lyp, iomode,
5002 0, UINT64_MAX, lyp->nfsly_stateid.seqid, stat, op,
5003 dp->nfsdi_deviceid, recallp);
5005 NFSCL_DEBUG(4, "nfscl_dserr recall iomode=%d\n", iomode);
5008 free(recallp, M_NFSLAYRECALL);
5011 /* And shut the TCP connection down. */
5012 nfscl_cancelreqs(dsp);
5016 * Cancel all RPCs for this "dsp" by closing the connection.
5017 * Also, mark the session as defunct.
5018 * If NFSCLDS_SAMECONN is set, the connection is shared with other DSs and
5019 * cannot be shut down.
5022 nfscl_cancelreqs(struct nfsclds *dsp)
5024 struct __rpc_client *cl;
5025 static int non_event;
5028 if ((dsp->nfsclds_flags & (NFSCLDS_CLOSED | NFSCLDS_SAMECONN)) == 0 &&
5029 dsp->nfsclds_sockp != NULL &&
5030 dsp->nfsclds_sockp->nr_client != NULL) {
5031 dsp->nfsclds_flags |= NFSCLDS_CLOSED;
5032 cl = dsp->nfsclds_sockp->nr_client;
5033 dsp->nfsclds_sess.nfsess_defunct = 1;
5037 * This 1sec sleep is done to reduce the number of reconnect
5038 * attempts made on the DS while it has failed.
5040 tsleep(&non_event, PVFS, "ndscls", hz);
5047 * Dereference a layout.
5050 nfscl_rellayout(struct nfscllayout *lyp, int exclocked)
5055 nfsv4_unlock(&lyp->nfsly_lock, 0);
5057 nfsv4_relref(&lyp->nfsly_lock);
5062 * Search for a devinfo by deviceid. If one is found, return it after
5063 * acquiring a reference count on it.
5065 struct nfscldevinfo *
5066 nfscl_getdevinfo(struct nfsclclient *clp, uint8_t *deviceid,
5067 struct nfscldevinfo *dip)
5072 dip = nfscl_finddevinfo(clp, deviceid);
5074 dip->nfsdi_refcnt++;
5080 * Dereference a devinfo structure.
5083 nfscl_reldevinfo_locked(struct nfscldevinfo *dip)
5086 dip->nfsdi_refcnt--;
5087 if (dip->nfsdi_refcnt == 0)
5088 wakeup(&dip->nfsdi_refcnt);
5092 * Dereference a devinfo structure.
5095 nfscl_reldevinfo(struct nfscldevinfo *dip)
5099 nfscl_reldevinfo_locked(dip);
5104 * Find a layout for this file handle. Return NULL upon failure.
5106 static struct nfscllayout *
5107 nfscl_findlayout(struct nfsclclient *clp, u_int8_t *fhp, int fhlen)
5109 struct nfscllayout *lyp;
5111 LIST_FOREACH(lyp, NFSCLLAYOUTHASH(clp, fhp, fhlen), nfsly_hash)
5112 if (lyp->nfsly_fhlen == fhlen &&
5113 !NFSBCMP(lyp->nfsly_fh, fhp, fhlen))
5119 * Find a devinfo for this deviceid. Return NULL upon failure.
5121 static struct nfscldevinfo *
5122 nfscl_finddevinfo(struct nfsclclient *clp, uint8_t *deviceid)
5124 struct nfscldevinfo *dip;
5126 LIST_FOREACH(dip, &clp->nfsc_devinfo, nfsdi_list)
5127 if (NFSBCMP(dip->nfsdi_deviceid, deviceid, NFSX_V4DEVICEID)
5134 * Merge the new file layout list into the main one, maintaining it in
5135 * increasing offset order.
5138 nfscl_mergeflayouts(struct nfsclflayouthead *fhlp,
5139 struct nfsclflayouthead *newfhlp)
5141 struct nfsclflayout *flp, *nflp, *prevflp, *tflp;
5143 flp = LIST_FIRST(fhlp);
5145 LIST_FOREACH_SAFE(nflp, newfhlp, nfsfl_list, tflp) {
5146 while (flp != NULL && flp->nfsfl_off < nflp->nfsfl_off) {
5148 flp = LIST_NEXT(flp, nfsfl_list);
5150 if (prevflp == NULL)
5151 LIST_INSERT_HEAD(fhlp, nflp, nfsfl_list);
5153 LIST_INSERT_AFTER(prevflp, nflp, nfsfl_list);
5159 * Add this nfscldevinfo to the client, if it doesn't already exist.
5160 * This function consumes the structure pointed at by dip, if not NULL.
5163 nfscl_adddevinfo(struct nfsmount *nmp, struct nfscldevinfo *dip, int ind,
5164 struct nfsclflayout *flp)
5166 struct nfsclclient *clp;
5167 struct nfscldevinfo *tdip;
5175 free(dip, M_NFSDEVINFO);
5178 if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5179 dev = flp->nfsfl_dev;
5181 dev = flp->nfsfl_ffm[ind].dev;
5182 tdip = nfscl_finddevinfo(clp, dev);
5184 tdip->nfsdi_layoutrefs++;
5185 if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5186 flp->nfsfl_devp = tdip;
5188 flp->nfsfl_ffm[ind].devp = tdip;
5189 nfscl_reldevinfo_locked(tdip);
5192 free(dip, M_NFSDEVINFO);
5196 LIST_INSERT_HEAD(&clp->nfsc_devinfo, dip, nfsdi_list);
5197 dip->nfsdi_layoutrefs = 1;
5198 if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5199 flp->nfsfl_devp = dip;
5201 flp->nfsfl_ffm[ind].devp = dip;
5210 * Free up a layout structure and associated file layout structure(s).
5213 nfscl_freelayout(struct nfscllayout *layp)
5215 struct nfsclflayout *flp, *nflp;
5216 struct nfsclrecalllayout *rp, *nrp;
5218 LIST_FOREACH_SAFE(flp, &layp->nfsly_flayread, nfsfl_list, nflp) {
5219 LIST_REMOVE(flp, nfsfl_list);
5220 nfscl_freeflayout(flp);
5222 LIST_FOREACH_SAFE(flp, &layp->nfsly_flayrw, nfsfl_list, nflp) {
5223 LIST_REMOVE(flp, nfsfl_list);
5224 nfscl_freeflayout(flp);
5226 LIST_FOREACH_SAFE(rp, &layp->nfsly_recall, nfsrecly_list, nrp) {
5227 LIST_REMOVE(rp, nfsrecly_list);
5228 free(rp, M_NFSLAYRECALL);
5231 free(layp, M_NFSLAYOUT);
5235 * Free up a file layout structure.
5238 nfscl_freeflayout(struct nfsclflayout *flp)
5242 if ((flp->nfsfl_flags & NFSFL_FILE) != 0) {
5243 for (i = 0; i < flp->nfsfl_fhcnt; i++)
5244 free(flp->nfsfl_fh[i], M_NFSFH);
5245 if (flp->nfsfl_devp != NULL)
5246 flp->nfsfl_devp->nfsdi_layoutrefs--;
5248 if ((flp->nfsfl_flags & NFSFL_FLEXFILE) != 0)
5249 for (i = 0; i < flp->nfsfl_mirrorcnt; i++) {
5250 for (j = 0; j < flp->nfsfl_ffm[i].fhcnt; j++)
5251 free(flp->nfsfl_ffm[i].fh[j], M_NFSFH);
5252 if (flp->nfsfl_ffm[i].devp != NULL)
5253 flp->nfsfl_ffm[i].devp->nfsdi_layoutrefs--;
5255 free(flp, M_NFSFLAYOUT);
5259 * Free up a file layout devinfo structure.
5262 nfscl_freedevinfo(struct nfscldevinfo *dip)
5265 free(dip, M_NFSDEVINFO);
5269 * Mark any layouts that match as recalled.
5272 nfscl_layoutrecall(int recalltype, struct nfscllayout *lyp, uint32_t iomode,
5273 uint64_t off, uint64_t len, uint32_t stateseqid, uint32_t stat, uint32_t op,
5274 char *devid, struct nfsclrecalllayout *recallp)
5276 struct nfsclrecalllayout *rp, *orp;
5278 recallp->nfsrecly_recalltype = recalltype;
5279 recallp->nfsrecly_iomode = iomode;
5280 recallp->nfsrecly_stateseqid = stateseqid;
5281 recallp->nfsrecly_off = off;
5282 recallp->nfsrecly_len = len;
5283 recallp->nfsrecly_stat = stat;
5284 recallp->nfsrecly_op = op;
5286 NFSBCOPY(devid, recallp->nfsrecly_devid, NFSX_V4DEVICEID);
5288 * Order the list as file returns first, followed by fsid and any
5289 * returns, both in increasing stateseqid order.
5290 * Note that the seqids wrap around, so 1 is after 0xffffffff.
5291 * (I'm not sure this is correct because I find RFC5661 confusing
5292 * on this, but hopefully it will work ok.)
5295 LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) {
5297 if ((recalltype == NFSLAYOUTRETURN_FILE &&
5298 (rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE ||
5299 nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) ||
5300 (recalltype != NFSLAYOUTRETURN_FILE &&
5301 rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE &&
5302 nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) {
5303 LIST_INSERT_BEFORE(rp, recallp, nfsrecly_list);
5308 * Put any error return on all the file returns that will
5311 if (rp->nfsrecly_recalltype == NFSLAYOUTRETURN_FILE &&
5312 stat != 0 && rp->nfsrecly_stat == 0) {
5313 rp->nfsrecly_stat = stat;
5314 rp->nfsrecly_op = op;
5316 NFSBCOPY(devid, rp->nfsrecly_devid,
5322 LIST_INSERT_HEAD(&lyp->nfsly_recall, recallp,
5325 LIST_INSERT_AFTER(orp, recallp, nfsrecly_list);
5327 lyp->nfsly_flags |= NFSLY_RECALL;
5328 wakeup(lyp->nfsly_clp);
5333 * Compare the two seqids for ordering. The trick is that the seqids can
5334 * wrap around from 0xffffffff->0, so check for the cases where one
5335 * has wrapped around.
5336 * Return 1 if seqid1 comes before seqid2, 0 otherwise.
5339 nfscl_seq(uint32_t seqid1, uint32_t seqid2)
5342 if (seqid2 > seqid1 && (seqid2 - seqid1) >= 0x7fffffff)
5343 /* seqid2 has wrapped around. */
5345 if (seqid1 > seqid2 && (seqid1 - seqid2) >= 0x7fffffff)
5346 /* seqid1 has wrapped around. */
5348 if (seqid1 <= seqid2)
5354 * Do a layout return for each of the recalls.
5357 nfscl_layoutreturn(struct nfsmount *nmp, struct nfscllayout *lyp,
5358 struct ucred *cred, NFSPROC_T *p)
5360 struct nfsclrecalllayout *rp;
5361 nfsv4stateid_t stateid;
5364 NFSBCOPY(lyp->nfsly_stateid.other, stateid.other, NFSX_STATEIDOTHER);
5365 stateid.seqid = lyp->nfsly_stateid.seqid;
5366 if ((lyp->nfsly_flags & NFSLY_FILES) != 0)
5367 layouttype = NFSLAYOUT_NFSV4_1_FILES;
5369 layouttype = NFSLAYOUT_FLEXFILE;
5370 LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) {
5371 (void)nfsrpc_layoutreturn(nmp, lyp->nfsly_fh,
5372 lyp->nfsly_fhlen, 0, layouttype,
5373 rp->nfsrecly_iomode, rp->nfsrecly_recalltype,
5374 rp->nfsrecly_off, rp->nfsrecly_len,
5375 &stateid, cred, p, rp->nfsrecly_stat, rp->nfsrecly_op,
5376 rp->nfsrecly_devid);
5381 * Do the layout commit for a file layout.
5384 nfscl_dolayoutcommit(struct nfsmount *nmp, struct nfscllayout *lyp,
5385 struct ucred *cred, NFSPROC_T *p)
5387 struct nfsclflayout *flp;
5389 int error, layouttype;
5391 if ((lyp->nfsly_flags & NFSLY_FILES) != 0)
5392 layouttype = NFSLAYOUT_NFSV4_1_FILES;
5394 layouttype = NFSLAYOUT_FLEXFILE;
5395 LIST_FOREACH(flp, &lyp->nfsly_flayrw, nfsfl_list) {
5396 if (layouttype == NFSLAYOUT_FLEXFILE &&
5397 (flp->nfsfl_fflags & NFSFLEXFLAG_NO_LAYOUTCOMMIT) != 0) {
5398 NFSCL_DEBUG(4, "Flex file: no layoutcommit\n");
5399 /* If not supported, don't bother doing it. */
5401 nmp->nm_state |= NFSSTA_NOLAYOUTCOMMIT;
5404 } else if (flp->nfsfl_off <= lyp->nfsly_lastbyte) {
5405 len = flp->nfsfl_end - flp->nfsfl_off;
5406 error = nfsrpc_layoutcommit(nmp, lyp->nfsly_fh,
5407 lyp->nfsly_fhlen, 0, flp->nfsfl_off, len,
5408 lyp->nfsly_lastbyte, &lyp->nfsly_stateid,
5409 layouttype, cred, p, NULL);
5410 NFSCL_DEBUG(4, "layoutcommit err=%d\n", error);
5411 if (error == NFSERR_NOTSUPP) {
5412 /* If not supported, don't bother doing it. */
5414 nmp->nm_state |= NFSSTA_NOLAYOUTCOMMIT;
5423 * Commit all layouts for a file (vnode).
5426 nfscl_layoutcommit(vnode_t vp, NFSPROC_T *p)
5428 struct nfsclclient *clp;
5429 struct nfscllayout *lyp;
5430 struct nfsnode *np = VTONFS(vp);
5432 struct nfsmount *nmp;
5434 mp = vnode_mount(vp);
5436 if (NFSHASNOLAYOUTCOMMIT(nmp))
5444 lyp = nfscl_findlayout(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
5449 nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
5450 if (NFSCL_FORCEDISM(mp)) {
5455 if ((lyp->nfsly_flags & NFSLY_WRITTEN) != 0) {
5456 lyp->nfsly_flags &= ~NFSLY_WRITTEN;
5458 NFSCL_DEBUG(4, "do layoutcommit2\n");
5459 nfscl_dolayoutcommit(clp->nfsc_nmp, lyp, NFSPROCCRED(p), p);
5463 nfsv4_relref(&lyp->nfsly_lock);