2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2009 Rick Macklem, University of Guelph
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
34 * These functions implement the client side state handling for NFSv4.
35 * NFSv4 state handling:
36 * - A lockowner is used to determine lock contention, so it
37 * corresponds directly to a Posix pid. (1 to 1 mapping)
38 * - The correct granularity of an OpenOwner is not nearly so
39 * obvious. An OpenOwner does the following:
40 * - provides a serial sequencing of Open/Close/Lock-with-new-lockowner
41 * - is used to check for Open/Share contention (not applicable to
42 * this client, since all Opens are Deny_None)
43 * As such, I considered both extreme.
44 * 1 OpenOwner per ClientID - Simple to manage, but fully serializes
45 * all Open, Close and Lock (with a new lockowner) Ops.
46 * 1 OpenOwner for each Open - This one results in an OpenConfirm for
47 * every Open, for most servers.
48 * So, I chose to use the same mapping as I did for LockOwnwers.
49 * The main concern here is that you can end up with multiple Opens
50 * for the same File Handle, but on different OpenOwners (opens
51 * inherited from parents, grandparents...) and you do not know
52 * which of these the vnodeop close applies to. This is handled by
53 * delaying the Close Op(s) until all of the Opens have been closed.
54 * (It is not yet obvious if this is the correct granularity.)
55 * - How the code handles serialization:
56 * - For the ClientId, it uses an exclusive lock while getting its
57 * SetClientId and during recovery. Otherwise, it uses a shared
58 * lock via a reference count.
59 * - For the rest of the data structures, it uses an SMP mutex
60 * (once the nfs client is SMP safe) and doesn't sleep while
61 * manipulating the linked lists.
62 * - The serialization of Open/Close/Lock/LockU falls out in the
63 * "wash", since OpenOwners and LockOwners are both mapped from
64 * Posix pid. In other words, there is only one Posix pid using
65 * any given owner, so that owner is serialized. (If you change
66 * the granularity of the OpenOwner, then code must be added to
67 * serialize Ops on the OpenOwner.)
68 * - When to get rid of OpenOwners and LockOwners.
69 * - The function nfscl_cleanup_common() is executed after a process exits.
70 * It goes through the client list looking for all Open and Lock Owners.
71 * When one is found, it is marked "defunct" or in the case of
72 * an OpenOwner without any Opens, freed.
73 * The renew thread scans for defunct Owners and gets rid of them,
74 * if it can. The LockOwners will also be deleted when the
75 * associated Open is closed.
76 * - If the LockU or Close Op(s) fail during close in a way
77 * that could be recovered upon retry, they are relinked to the
78 * ClientId's defunct open list and retried by the renew thread
79 * until they succeed or an unmount/recovery occurs.
80 * (Since we are done with them, they do not need to be recovered.)
83 #include <fs/nfs/nfsport.h>
88 extern struct nfsstatsv1 nfsstatsv1;
89 extern struct nfsreqhead nfsd_reqq;
90 extern u_int32_t newnfs_false, newnfs_true;
91 extern int nfscl_debuglevel;
92 extern int nfscl_enablecallb;
93 extern int nfs_numnfscbd;
97 struct nfsclhead nfsclhead; /* Head of clientid list */
98 int nfscl_deleghighwater = NFSCLDELEGHIGHWATER;
99 int nfscl_layouthighwater = NFSCLLAYOUTHIGHWATER;
101 static int nfscl_delegcnt = 0;
102 static int nfscl_layoutcnt = 0;
103 static int nfscl_getopen(struct nfsclownerhead *, struct nfsclopenhash *,
104 u_int8_t *, int, u_int8_t *, u_int8_t *, u_int32_t,
105 struct nfscllockowner **, struct nfsclopen **);
106 static bool nfscl_checkown(struct nfsclowner *, struct nfsclopen *, uint8_t *,
107 uint8_t *, struct nfscllockowner **, struct nfsclopen **,
108 struct nfsclopen **);
109 static void nfscl_clrelease(struct nfsclclient *);
110 static void nfscl_cleanclient(struct nfsclclient *);
111 static void nfscl_expireclient(struct nfsclclient *, struct nfsmount *,
112 struct ucred *, NFSPROC_T *);
113 static int nfscl_expireopen(struct nfsclclient *, struct nfsclopen *,
114 struct nfsmount *, struct ucred *, NFSPROC_T *);
115 static void nfscl_recover(struct nfsclclient *, bool *, struct ucred *,
117 static void nfscl_insertlock(struct nfscllockowner *, struct nfscllock *,
118 struct nfscllock *, int);
119 static int nfscl_updatelock(struct nfscllockowner *, struct nfscllock **,
120 struct nfscllock **, int);
121 static void nfscl_delegreturnall(struct nfsclclient *, NFSPROC_T *);
122 static u_int32_t nfscl_nextcbident(void);
123 static mount_t nfscl_getmnt(int, uint8_t *, u_int32_t, struct nfsclclient **);
124 static struct nfsclclient *nfscl_getclnt(u_int32_t);
125 static struct nfsclclient *nfscl_getclntsess(uint8_t *);
126 static struct nfscldeleg *nfscl_finddeleg(struct nfsclclient *, u_int8_t *,
128 static void nfscl_retoncloselayout(vnode_t, struct nfsclclient *, uint8_t *,
129 int, struct nfsclrecalllayout **);
130 static void nfscl_reldevinfo_locked(struct nfscldevinfo *);
131 static struct nfscllayout *nfscl_findlayout(struct nfsclclient *, u_int8_t *,
133 static struct nfscldevinfo *nfscl_finddevinfo(struct nfsclclient *, uint8_t *);
134 static int nfscl_checkconflict(struct nfscllockownerhead *, struct nfscllock *,
135 u_int8_t *, struct nfscllock **);
136 static void nfscl_freealllocks(struct nfscllockownerhead *, int);
137 static int nfscl_localconflict(struct nfsclclient *, u_int8_t *, int,
138 struct nfscllock *, u_int8_t *, struct nfscldeleg *, struct nfscllock **);
139 static void nfscl_newopen(struct nfsclclient *, struct nfscldeleg *,
140 struct nfsclowner **, struct nfsclowner **, struct nfsclopen **,
141 struct nfsclopen **, u_int8_t *, u_int8_t *, int, struct ucred *, int *);
142 static int nfscl_moveopen(vnode_t , struct nfsclclient *,
143 struct nfsmount *, struct nfsclopen *, struct nfsclowner *,
144 struct nfscldeleg *, struct ucred *, NFSPROC_T *);
145 static void nfscl_totalrecall(struct nfsclclient *);
146 static int nfscl_relock(vnode_t , struct nfsclclient *, struct nfsmount *,
147 struct nfscllockowner *, struct nfscllock *, struct ucred *, NFSPROC_T *);
148 static int nfscl_tryopen(struct nfsmount *, vnode_t , u_int8_t *, int,
149 u_int8_t *, int, u_int32_t, struct nfsclopen *, u_int8_t *, int,
150 struct nfscldeleg **, int, u_int32_t, struct ucred *, NFSPROC_T *);
151 static int nfscl_trylock(struct nfsmount *, vnode_t , u_int8_t *,
152 int, struct nfscllockowner *, int, int, u_int64_t, u_int64_t, short,
153 struct ucred *, NFSPROC_T *);
154 static int nfsrpc_reopen(struct nfsmount *, u_int8_t *, int, u_int32_t,
155 struct nfsclopen *, struct nfscldeleg **, struct ucred *, NFSPROC_T *);
156 static void nfscl_freedeleg(struct nfscldeleghead *, struct nfscldeleg *,
158 static int nfscl_errmap(struct nfsrv_descript *, u_int32_t);
159 static void nfscl_cleanup_common(struct nfsclclient *, u_int8_t *);
160 static int nfscl_recalldeleg(struct nfsclclient *, struct nfsmount *,
161 struct nfscldeleg *, vnode_t, struct ucred *, NFSPROC_T *, int,
163 static void nfscl_freeopenowner(struct nfsclowner *, int);
164 static void nfscl_cleandeleg(struct nfscldeleg *);
165 static int nfscl_trydelegreturn(struct nfscldeleg *, struct ucred *,
166 struct nfsmount *, NFSPROC_T *);
167 static void nfscl_emptylockowner(struct nfscllockowner *,
168 struct nfscllockownerfhhead *);
169 static void nfscl_mergeflayouts(struct nfsclflayouthead *,
170 struct nfsclflayouthead *);
171 static int nfscl_layoutrecall(int, struct nfscllayout *, uint32_t, uint64_t,
172 uint64_t, uint32_t, uint32_t, uint32_t, char *, struct nfsclrecalllayout *);
173 static int nfscl_seq(uint32_t, uint32_t);
174 static void nfscl_layoutreturn(struct nfsmount *, struct nfscllayout *,
175 struct ucred *, NFSPROC_T *);
176 static void nfscl_dolayoutcommit(struct nfsmount *, struct nfscllayout *,
177 struct ucred *, NFSPROC_T *);
179 static short nfscberr_null[] = {
184 static short nfscberr_getattr[] = {
193 static short nfscberr_recall[] = {
203 static short *nfscl_cberrmap[] = {
211 #define NETFAMILY(clp) \
212 (((clp)->nfsc_flags & NFSCLFLAGS_AFINET6) ? AF_INET6 : AF_INET)
215 * Called for an open operation.
216 * If the nfhp argument is NULL, just get an openowner.
219 nfscl_open(vnode_t vp, u_int8_t *nfhp, int fhlen, u_int32_t amode, int usedeleg,
220 struct ucred *cred, NFSPROC_T *p, struct nfsclowner **owpp,
221 struct nfsclopen **opp, int *newonep, int *retp, int lockit)
223 struct nfsclclient *clp;
224 struct nfsclowner *owp, *nowp;
225 struct nfsclopen *op = NULL, *nop = NULL;
226 struct nfscldeleg *dp;
227 struct nfsclownerhead *ohp;
228 u_int8_t own[NFSV4CL_LOCKNAMELEN];
239 * Might need one or both of these, so MALLOC them now, to
240 * avoid a tsleep() in MALLOC later.
242 nowp = malloc(sizeof (struct nfsclowner),
243 M_NFSCLOWNER, M_WAITOK);
245 nop = malloc(sizeof (struct nfsclopen) +
246 fhlen - 1, M_NFSCLOPEN, M_WAITOK);
247 nop->nfso_hash.le_prev = NULL;
249 ret = nfscl_getcl(vp->v_mount, cred, p, false, &clp);
251 free(nowp, M_NFSCLOWNER);
253 free(nop, M_NFSCLOPEN);
258 * Get the Open iff it already exists.
259 * If none found, add the new one or return error, depending upon
264 /* First check the delegation list */
265 if (nfhp != NULL && usedeleg) {
266 LIST_FOREACH(dp, NFSCLDELEGHASH(clp, nfhp, fhlen), nfsdl_hash) {
267 if (dp->nfsdl_fhlen == fhlen &&
268 !NFSBCMP(nfhp, dp->nfsdl_fh, fhlen)) {
269 if (!(amode & NFSV4OPEN_ACCESSWRITE) ||
270 (dp->nfsdl_flags & NFSCLDL_WRITE))
278 /* For NFSv4.1/4.2 and this option, use a single open_owner. */
279 if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount)))
280 nfscl_filllockowner(NULL, own, F_POSIX);
282 nfscl_filllockowner(p->td_proc, own, F_POSIX);
284 ohp = &dp->nfsdl_owner;
286 ohp = &clp->nfsc_owner;
287 /* Now, search for an openowner */
288 LIST_FOREACH(owp, ohp, nfsow_list) {
289 if (!NFSBCMP(owp->nfsow_owner, own, NFSV4CL_LOCKNAMELEN))
294 * Create a new open, as required.
296 nfscl_newopen(clp, dp, &owp, &nowp, &op, &nop, own, nfhp, fhlen,
300 * Now, check the mode on the open and return the appropriate
304 if (nfhp != NULL && dp != NULL && nop == NULL)
305 /* new local open on delegation */
306 *retp = NFSCLOPEN_SETCRED;
308 *retp = NFSCLOPEN_OK;
310 if (op != NULL && (amode & ~(op->nfso_mode))) {
311 op->nfso_mode |= amode;
312 if (retp != NULL && dp == NULL)
313 *retp = NFSCLOPEN_DOOPEN;
317 * Serialize modifications to the open owner for multiple threads
318 * within the same process using a read/write sleep lock.
319 * For NFSv4.1 and a single OpenOwner, allow concurrent open operations
320 * by acquiring a shared lock. The close operations still use an
321 * exclusive lock for this case.
324 if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount))) {
326 * Get a shared lock on the OpenOwner, but first
327 * wait for any pending exclusive lock, so that the
328 * exclusive locker gets priority.
330 nfsv4_lock(&owp->nfsow_rwlock, 0, NULL,
331 NFSCLSTATEMUTEXPTR, NULL);
332 nfsv4_getref(&owp->nfsow_rwlock, NULL,
333 NFSCLSTATEMUTEXPTR, NULL);
335 nfscl_lockexcl(&owp->nfsow_rwlock, NFSCLSTATEMUTEXPTR);
339 free(nowp, M_NFSCLOWNER);
341 free(nop, M_NFSCLOPEN);
350 * Create a new open, as required.
353 nfscl_newopen(struct nfsclclient *clp, struct nfscldeleg *dp,
354 struct nfsclowner **owpp, struct nfsclowner **nowpp, struct nfsclopen **opp,
355 struct nfsclopen **nopp, u_int8_t *own, u_int8_t *fhp, int fhlen,
356 struct ucred *cred, int *newonep)
358 struct nfsclowner *owp = *owpp, *nowp;
359 struct nfsclopen *op, *nop;
369 if (owp == NULL && nowp != NULL) {
370 NFSBCOPY(own, nowp->nfsow_owner, NFSV4CL_LOCKNAMELEN);
371 LIST_INIT(&nowp->nfsow_open);
372 nowp->nfsow_clp = clp;
373 nowp->nfsow_seqid = 0;
374 nowp->nfsow_defunct = 0;
375 nfscl_lockinit(&nowp->nfsow_rwlock);
377 nfsstatsv1.cllocalopenowners++;
378 LIST_INSERT_HEAD(&dp->nfsdl_owner, nowp, nfsow_list);
380 nfsstatsv1.clopenowners++;
381 LIST_INSERT_HEAD(&clp->nfsc_owner, nowp, nfsow_list);
389 /* If an fhp has been specified, create an Open as well. */
391 /* and look for the correct open, based upon FH */
392 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
393 if (op->nfso_fhlen == fhlen &&
394 !NFSBCMP(op->nfso_fh, fhp, fhlen))
397 if (op == NULL && nop != NULL) {
400 nop->nfso_opencnt = 0;
401 nop->nfso_posixlock = 1;
402 nop->nfso_fhlen = fhlen;
403 NFSBCOPY(fhp, nop->nfso_fh, fhlen);
404 LIST_INIT(&nop->nfso_lock);
405 nop->nfso_stateid.seqid = 0;
406 nop->nfso_stateid.other[0] = 0;
407 nop->nfso_stateid.other[1] = 0;
408 nop->nfso_stateid.other[2] = 0;
409 KASSERT(cred != NULL, ("%s: cred NULL\n", __func__));
410 newnfs_copyincred(cred, &nop->nfso_cred);
412 TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
413 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp,
415 dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
416 nfsstatsv1.cllocalopens++;
418 LIST_INSERT_HEAD(NFSCLOPENHASH(clp, fhp, fhlen),
420 nfsstatsv1.clopens++;
422 LIST_INSERT_HEAD(&owp->nfsow_open, nop, nfso_list);
434 * Called to find/add a delegation to a client.
437 nfscl_deleg(mount_t mp, struct nfsclclient *clp, u_int8_t *nfhp,
438 int fhlen, struct ucred *cred, NFSPROC_T *p, struct nfscldeleg **dpp)
440 struct nfscldeleg *dp = *dpp, *tdp;
441 struct nfsmount *nmp;
443 KASSERT(mp != NULL, ("nfscl_deleg: mp NULL"));
446 * First, if we have received a Read delegation for a file on a
447 * read/write file system, just return it, because they aren't
450 if (dp != NULL && !NFSMNT_RDONLY(mp) &&
451 (dp->nfsdl_flags & NFSCLDL_READ)) {
452 nfscl_trydelegreturn(dp, cred, nmp, p);
453 free(dp, M_NFSCLDELEG);
459 * Since a delegation might be added to the mount,
460 * set NFSMNTP_DELEGISSUED now. If a delegation already
461 * exagain ists, setting this flag is harmless.
464 nmp->nm_privflag |= NFSMNTP_DELEGISSUED;
467 /* Look for the correct deleg, based upon FH */
469 tdp = nfscl_finddeleg(clp, nfhp, fhlen);
473 return (NFSERR_BADSTATEID);
476 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list);
477 LIST_INSERT_HEAD(NFSCLDELEGHASH(clp, nfhp, fhlen), dp,
479 dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
480 nfsstatsv1.cldelegates++;
484 * Delegation already exists, what do we do if a new one??
487 printf("Deleg already exists!\n");
488 free(dp, M_NFSCLDELEG);
499 * Find a delegation for this file handle. Return NULL upon failure.
501 static struct nfscldeleg *
502 nfscl_finddeleg(struct nfsclclient *clp, u_int8_t *fhp, int fhlen)
504 struct nfscldeleg *dp;
506 LIST_FOREACH(dp, NFSCLDELEGHASH(clp, fhp, fhlen), nfsdl_hash) {
507 if (dp->nfsdl_fhlen == fhlen &&
508 !NFSBCMP(dp->nfsdl_fh, fhp, fhlen))
515 * Get a stateid for an I/O operation. First, look for an open and iff
516 * found, return either a lockowner stateid or the open stateid.
517 * If no Open is found, just return error and the special stateid of all zeros.
520 nfscl_getstateid(vnode_t vp, u_int8_t *nfhp, int fhlen, u_int32_t mode,
521 int fords, struct ucred *cred, NFSPROC_T *p, nfsv4stateid_t *stateidp,
524 struct nfsclclient *clp;
525 struct nfsclopen *op = NULL, *top;
526 struct nfsclopenhash *oph;
527 struct nfscllockowner *lp;
528 struct nfscldeleg *dp;
530 struct nfsmount *nmp;
531 u_int8_t own[NFSV4CL_LOCKNAMELEN], lockown[NFSV4CL_LOCKNAMELEN];
537 * Initially, just set the special stateid of all zeros.
538 * (Don't do this for a DS, since the special stateid can't be used.)
542 stateidp->other[0] = 0;
543 stateidp->other[1] = 0;
544 stateidp->other[2] = 0;
546 if (vnode_vtype(vp) != VREG)
549 nmp = VFSTONFS(vp->v_mount);
552 * For "oneopenown" mounts, first check for a cached open in the
553 * NFS vnode, that can be used as a stateid. This can only be
554 * done if no delegations have been issued to the mount and no
555 * byte range file locking has been done for the file.
557 if (NFSHASNFSV4N(nmp) && NFSHASONEOPENOWN(nmp) && fords == 0) {
560 if ((nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0 &&
561 (np->n_flag & NMIGHTBELOCKED) == 0 &&
562 np->n_openstateid != NULL) {
565 np->n_openstateid->nfso_stateid.other[0];
567 np->n_openstateid->nfso_stateid.other[1];
569 np->n_openstateid->nfso_stateid.other[2];
579 clp = nfscl_findcl(nmp);
586 * Wait for recovery to complete.
588 while ((clp->nfsc_flags & NFSCLFLAGS_RECVRINPROG))
589 (void) nfsmsleep(&clp->nfsc_flags, NFSCLSTATEMUTEXPTR,
590 PZERO, "nfsrecvr", NULL);
593 * First, look for a delegation.
595 LIST_FOREACH(dp, NFSCLDELEGHASH(clp, nfhp, fhlen), nfsdl_hash) {
596 if (dp->nfsdl_fhlen == fhlen &&
597 !NFSBCMP(nfhp, dp->nfsdl_fh, fhlen)) {
598 if (!(mode & NFSV4OPEN_ACCESSWRITE) ||
599 (dp->nfsdl_flags & NFSCLDL_WRITE)) {
600 if (NFSHASNFSV4N(nmp))
604 dp->nfsdl_stateid.seqid;
605 stateidp->other[0] = dp->nfsdl_stateid.other[0];
606 stateidp->other[1] = dp->nfsdl_stateid.other[1];
607 stateidp->other[2] = dp->nfsdl_stateid.other[2];
608 if (!(np->n_flag & NDELEGRECALL)) {
609 TAILQ_REMOVE(&clp->nfsc_deleg, dp,
611 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp,
613 dp->nfsdl_timestamp = NFSD_MONOSEC +
615 dp->nfsdl_rwlock.nfslock_usecnt++;
616 *lckpp = (void *)&dp->nfsdl_rwlock;
627 * If p != NULL, we want to search the parentage tree
628 * for a matching OpenOwner and use that.
630 if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount)))
631 nfscl_filllockowner(NULL, own, F_POSIX);
633 nfscl_filllockowner(p->td_proc, own, F_POSIX);
634 nfscl_filllockowner(p->td_proc, lockown, F_POSIX);
636 error = nfscl_getopen(NULL, clp->nfsc_openhash, nfhp, fhlen,
637 own, lockown, mode, &lp, &op);
638 if (error == 0 && lp != NULL && fords == 0) {
639 /* Don't return a lock stateid for a DS. */
640 if (NFSHASNFSV4N(nmp))
643 stateidp->seqid = lp->nfsl_stateid.seqid;
645 lp->nfsl_stateid.other[0];
647 lp->nfsl_stateid.other[1];
649 lp->nfsl_stateid.other[2];
655 /* If not found, just look for any OpenOwner that will work. */
658 oph = NFSCLOPENHASH(clp, nfhp, fhlen);
659 LIST_FOREACH(op, oph, nfso_hash) {
660 if (op->nfso_fhlen == fhlen &&
661 !NFSBCMP(op->nfso_fh, nfhp, fhlen)) {
662 if (top == NULL && (op->nfso_mode &
663 NFSV4OPEN_ACCESSWRITE) != 0 &&
664 (mode & NFSV4OPEN_ACCESSREAD) != 0)
666 if ((mode & op->nfso_mode) == mode) {
667 /* LRU order the hash list. */
668 LIST_REMOVE(op, nfso_hash);
669 LIST_INSERT_HEAD(oph, op, nfso_hash);
676 NFSCL_DEBUG(2, "openmode top=%p\n", top);
677 if (top == NULL || NFSHASOPENMODE(nmp)) {
684 * For read aheads or write behinds, use the open cred.
685 * A read ahead or write behind is indicated by p == NULL.
688 newnfs_copycred(&op->nfso_cred, cred);
692 * No lock stateid, so return the open stateid.
694 if (NFSHASNFSV4N(nmp))
697 stateidp->seqid = op->nfso_stateid.seqid;
698 stateidp->other[0] = op->nfso_stateid.other[0];
699 stateidp->other[1] = op->nfso_stateid.other[1];
700 stateidp->other[2] = op->nfso_stateid.other[2];
706 * Search for a matching file, mode and, optionally, lockowner.
709 nfscl_getopen(struct nfsclownerhead *ohp, struct nfsclopenhash *ohashp,
710 u_int8_t *nfhp, int fhlen, u_int8_t *openown, u_int8_t *lockown,
711 u_int32_t mode, struct nfscllockowner **lpp, struct nfsclopen **opp)
713 struct nfsclowner *owp;
714 struct nfsclopen *op, *rop, *rop2;
715 struct nfsclopenhash *oph;
718 KASSERT(ohp == NULL || ohashp == NULL, ("nfscl_getopen: "
719 "only one of ohp and ohashp can be set"));
723 * rop will be set to the open to be returned. There are three
724 * variants of this, all for an open of the correct file:
725 * 1 - A match of lockown.
726 * 2 - A match of the openown, when no lockown match exists.
727 * 3 - A match for any open, if no openown or lockown match exists.
728 * Looking for #2 over #3 probably isn't necessary, but since
729 * RFC3530 is vague w.r.t. the relationship between openowners and
730 * lockowners, I think this is the safer way to go.
735 /* Search the client list */
736 if (ohashp == NULL) {
737 /* Search the local opens on the delegation. */
738 LIST_FOREACH(owp, ohp, nfsow_list) {
739 /* and look for the correct open */
740 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
741 if (op->nfso_fhlen == fhlen &&
742 !NFSBCMP(op->nfso_fh, nfhp, fhlen)
743 && (op->nfso_mode & mode) == mode)
744 keep_looping = nfscl_checkown(owp, op, openown,
745 lockown, lpp, &rop, &rop2);
753 /* Search for matching opens on the hash list. */
754 oph = &ohashp[NFSCLOPENHASHFUNC(nfhp, fhlen)];
755 LIST_FOREACH(op, oph, nfso_hash) {
756 if (op->nfso_fhlen == fhlen &&
757 !NFSBCMP(op->nfso_fh, nfhp, fhlen)
758 && (op->nfso_mode & mode) == mode)
759 keep_looping = nfscl_checkown(op->nfso_own, op,
760 openown, lockown, lpp, &rop, &rop2);
762 /* LRU order the hash list. */
763 LIST_REMOVE(op, nfso_hash);
764 LIST_INSERT_HEAD(oph, op, nfso_hash);
777 /* Check for an owner match. */
779 nfscl_checkown(struct nfsclowner *owp, struct nfsclopen *op, uint8_t *openown,
780 uint8_t *lockown, struct nfscllockowner **lpp, struct nfsclopen **ropp,
781 struct nfsclopen **ropp2)
783 struct nfscllockowner *lp;
788 /* Now look for a matching lockowner. */
789 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
790 if (!NFSBCMP(lp->nfsl_owner, lockown,
791 NFSV4CL_LOCKNAMELEN)) {
798 if (*ropp == NULL && !NFSBCMP(owp->nfsow_owner, openown,
799 NFSV4CL_LOCKNAMELEN)) {
802 keep_looping = false;
806 return (keep_looping);
810 * Release use of an open owner. Called when open operations are done
811 * with the open owner.
814 nfscl_ownerrelease(struct nfsmount *nmp, struct nfsclowner *owp,
815 __unused int error, __unused int candelete, int unlocked)
822 if (NFSHASONEOPENOWN(nmp))
823 nfsv4_relref(&owp->nfsow_rwlock);
825 nfscl_lockunlock(&owp->nfsow_rwlock);
827 nfscl_clrelease(owp->nfsow_clp);
832 * Release use of an open structure under an open owner.
835 nfscl_openrelease(struct nfsmount *nmp, struct nfsclopen *op, int error,
838 struct nfsclclient *clp;
839 struct nfsclowner *owp;
845 if (NFSHASONEOPENOWN(nmp))
846 nfsv4_relref(&owp->nfsow_rwlock);
848 nfscl_lockunlock(&owp->nfsow_rwlock);
849 clp = owp->nfsow_clp;
850 if (error && candelete && op->nfso_opencnt == 0)
851 nfscl_freeopen(op, 0);
852 nfscl_clrelease(clp);
857 * Called to get a clientid structure. It will optionally lock the
858 * client data structures to do the SetClientId/SetClientId_confirm,
859 * but will release that lock and return the clientid with a reference
861 * If the "cred" argument is NULL, a new clientid should not be created.
862 * If the "p" argument is NULL, a SetClientID/SetClientIDConfirm cannot
864 * It always clpp with a reference count on it, unless returning an error.
867 nfscl_getcl(struct mount *mp, struct ucred *cred, NFSPROC_T *p,
868 bool tryminvers, struct nfsclclient **clpp)
870 struct nfsclclient *clp;
871 struct nfsclclient *newclp = NULL;
872 struct nfsmount *nmp;
873 char uuid[HOSTUUIDLEN];
874 int igotlock = 0, error, trystalecnt, clidinusedelay, i;
879 getcredhostuuid(cred, uuid, sizeof uuid);
880 idlen = strlen(uuid);
882 idlen += sizeof (u_int64_t);
884 idlen += sizeof (u_int64_t) + 16; /* 16 random bytes */
886 sizeof (struct nfsclclient) + idlen - 1, M_NFSCLCLIENT,
891 * If a forced dismount is already in progress, don't
892 * allocate a new clientid and get out now. For the case where
893 * clp != NULL, this is a harmless optimization.
895 if (NFSCL_FORCEDISM(mp)) {
898 free(newclp, M_NFSCLCLIENT);
903 if (newclp == NULL) {
908 clp->nfsc_idlen = idlen;
909 LIST_INIT(&clp->nfsc_owner);
910 TAILQ_INIT(&clp->nfsc_deleg);
911 TAILQ_INIT(&clp->nfsc_layout);
912 LIST_INIT(&clp->nfsc_devinfo);
913 for (i = 0; i < NFSCLDELEGHASHSIZE; i++)
914 LIST_INIT(&clp->nfsc_deleghash[i]);
915 for (i = 0; i < NFSCLOPENHASHSIZE; i++)
916 LIST_INIT(&clp->nfsc_openhash[i]);
917 for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++)
918 LIST_INIT(&clp->nfsc_layouthash[i]);
919 clp->nfsc_flags = NFSCLFLAGS_INITED;
920 clp->nfsc_clientidrev = 1;
921 clp->nfsc_cbident = nfscl_nextcbident();
922 nfscl_fillclid(nmp->nm_clval, uuid, clp->nfsc_id,
924 LIST_INSERT_HEAD(&nfsclhead, clp, nfsc_list);
929 free(newclp, M_NFSCLCLIENT);
931 while ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0 && !igotlock &&
932 !NFSCL_FORCEDISM(mp))
933 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
934 NFSCLSTATEMUTEXPTR, mp);
937 * Call nfsv4_lock() with "iwantlock == 0" so that it will
938 * wait for a pending exclusive lock request. This gives the
939 * exclusive lock request priority over this shared lock
941 * An exclusive lock on nfsc_lock is used mainly for server
944 nfsv4_lock(&clp->nfsc_lock, 0, NULL, NFSCLSTATEMUTEXPTR, mp);
945 nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
947 if (igotlock == 0 && NFSCL_FORCEDISM(mp)) {
949 * Both nfsv4_lock() and nfsv4_getref() know to check
950 * for NFSCL_FORCEDISM() and return without sleeping to
951 * wait for the exclusive lock to be released, since it
952 * might be held by nfscl_umount() and we need to get out
953 * now for that case and not wait until nfscl_umount()
962 * If it needs a clientid, do the setclientid now.
964 if ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0) {
966 panic("nfscl_clget");
967 if (p == NULL || cred == NULL) {
969 nfsv4_unlock(&clp->nfsc_lock, 0);
974 * If RFC3530 Sec. 14.2.33 is taken literally,
975 * NFSERR_CLIDINUSE will be returned persistently for the
976 * case where a new mount of the same file system is using
977 * a different principal. In practice, NFSERR_CLIDINUSE is
978 * only returned when there is outstanding unexpired state
979 * on the clientid. As such, try for twice the lease
980 * interval, if we know what that is. Otherwise, make a
982 * The case of returning NFSERR_STALECLIENTID is far less
983 * likely, but might occur if there is a significant delay
984 * between doing the SetClientID and SetClientIDConfirm Ops,
985 * such that the server throws away the clientid before
986 * receiving the SetClientIDConfirm.
988 if (clp->nfsc_renew > 0)
989 clidinusedelay = NFSCL_LEASE(clp->nfsc_renew) * 2;
991 clidinusedelay = 120;
994 error = nfsrpc_setclient(nmp, clp, 0, NULL, cred, p);
995 if (error == NFSERR_STALECLIENTID ||
996 error == NFSERR_STALEDONTRECOVER ||
997 error == NFSERR_BADSESSION ||
998 error == NFSERR_CLIDINUSE) {
999 (void) nfs_catnap(PZERO, error, "nfs_setcl");
1000 } else if (error == NFSERR_MINORVERMISMATCH &&
1002 if (nmp->nm_minorvers > 0)
1003 nmp->nm_minorvers--;
1007 } while (((error == NFSERR_STALECLIENTID ||
1008 error == NFSERR_BADSESSION ||
1009 error == NFSERR_STALEDONTRECOVER) && --trystalecnt > 0) ||
1010 (error == NFSERR_CLIDINUSE && --clidinusedelay > 0) ||
1011 (error == NFSERR_MINORVERMISMATCH && tryminvers));
1014 nfsv4_unlock(&clp->nfsc_lock, 0);
1018 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
1022 nfsv4_unlock(&clp->nfsc_lock, 1);
1031 * Get a reference to a clientid and return it, if valid.
1033 struct nfsclclient *
1034 nfscl_findcl(struct nfsmount *nmp)
1036 struct nfsclclient *clp;
1039 if (clp == NULL || !(clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID))
1045 * Release the clientid structure. It may be locked or reference counted.
1048 nfscl_clrelease(struct nfsclclient *clp)
1051 if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK)
1052 nfsv4_unlock(&clp->nfsc_lock, 0);
1054 nfsv4_relref(&clp->nfsc_lock);
1058 * External call for nfscl_clrelease.
1061 nfscl_clientrelease(struct nfsclclient *clp)
1065 if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK)
1066 nfsv4_unlock(&clp->nfsc_lock, 0);
1068 nfsv4_relref(&clp->nfsc_lock);
1073 * Called when wanting to lock a byte region.
1076 nfscl_getbytelock(vnode_t vp, u_int64_t off, u_int64_t len,
1077 short type, struct ucred *cred, NFSPROC_T *p, struct nfsclclient *rclp,
1078 int recovery, void *id, int flags, u_int8_t *rownp, u_int8_t *ropenownp,
1079 struct nfscllockowner **lpp, int *newonep, int *donelocallyp)
1081 struct nfscllockowner *lp;
1082 struct nfsclopen *op;
1083 struct nfsclclient *clp;
1084 struct nfscllockowner *nlp;
1085 struct nfscllock *nlop, *otherlop;
1086 struct nfscldeleg *dp = NULL, *ldp = NULL;
1087 struct nfscllockownerhead *lhp = NULL;
1089 u_int8_t own[NFSV4CL_LOCKNAMELEN], *ownp, openown[NFSV4CL_LOCKNAMELEN];
1091 int error = 0, ret, donelocally = 0;
1094 /* For Lock Ops, the open mode doesn't matter, so use 0 to match any. */
1103 * Might need these, so MALLOC them now, to
1104 * avoid a tsleep() in MALLOC later.
1107 sizeof (struct nfscllockowner), M_NFSCLLOCKOWNER, M_WAITOK);
1109 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1111 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1112 nlop->nfslo_type = type;
1113 nlop->nfslo_first = off;
1114 if (len == NFS64BITSSET) {
1115 nlop->nfslo_end = NFS64BITSSET;
1117 nlop->nfslo_end = off + len;
1118 if (nlop->nfslo_end <= nlop->nfslo_first)
1119 error = NFSERR_INVAL;
1126 error = nfscl_getcl(vp->v_mount, cred, p, false, &clp);
1129 free(nlp, M_NFSCLLOCKOWNER);
1130 free(otherlop, M_NFSCLLOCK);
1131 free(nlop, M_NFSCLLOCK);
1138 openownp = ropenownp;
1140 nfscl_filllockowner(id, own, flags);
1142 if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount)))
1143 nfscl_filllockowner(NULL, openown, F_POSIX);
1145 nfscl_filllockowner(p->td_proc, openown, F_POSIX);
1151 * First, search for a delegation. If one exists for this file,
1152 * the lock can be done locally against it, so long as there
1153 * isn't a local lock conflict.
1155 ldp = dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
1156 np->n_fhp->nfh_len);
1157 /* Just sanity check for correct type of delegation */
1158 if (dp != NULL && ((dp->nfsdl_flags &
1159 (NFSCLDL_RECALL | NFSCLDL_DELEGRET)) != 0 ||
1161 (dp->nfsdl_flags & NFSCLDL_WRITE) == 0)))
1165 /* Now, find an open and maybe a lockowner. */
1166 ret = nfscl_getopen(&dp->nfsdl_owner, NULL, np->n_fhp->nfh_fh,
1167 np->n_fhp->nfh_len, openownp, ownp, mode, NULL, &op);
1169 ret = nfscl_getopen(NULL, clp->nfsc_openhash,
1170 np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp,
1171 ownp, mode, NULL, &op);
1173 lhp = &dp->nfsdl_lock;
1174 TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
1175 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list);
1176 dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
1184 * Get the related Open and maybe lockowner.
1186 error = nfscl_getopen(NULL, clp->nfsc_openhash,
1187 np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp,
1188 ownp, mode, &lp, &op);
1190 lhp = &op->nfso_lock;
1192 if (!error && !recovery)
1193 error = nfscl_localconflict(clp, np->n_fhp->nfh_fh,
1194 np->n_fhp->nfh_len, nlop, ownp, ldp, NULL);
1197 nfscl_clrelease(clp);
1200 free(nlp, M_NFSCLLOCKOWNER);
1201 free(otherlop, M_NFSCLLOCK);
1202 free(nlop, M_NFSCLLOCK);
1207 * Ok, see if a lockowner exists and create one, as required.
1210 LIST_FOREACH(lp, lhp, nfsl_list) {
1211 if (!NFSBCMP(lp->nfsl_owner, ownp, NFSV4CL_LOCKNAMELEN))
1215 NFSBCOPY(ownp, nlp->nfsl_owner, NFSV4CL_LOCKNAMELEN);
1217 NFSBCOPY(ropenownp, nlp->nfsl_openowner,
1218 NFSV4CL_LOCKNAMELEN);
1220 NFSBCOPY(op->nfso_own->nfsow_owner, nlp->nfsl_openowner,
1221 NFSV4CL_LOCKNAMELEN);
1222 nlp->nfsl_seqid = 0;
1223 nlp->nfsl_lockflags = flags;
1224 nlp->nfsl_inprog = NULL;
1225 nfscl_lockinit(&nlp->nfsl_rwlock);
1226 LIST_INIT(&nlp->nfsl_lock);
1228 nlp->nfsl_open = NULL;
1229 nfsstatsv1.cllocallockowners++;
1231 nlp->nfsl_open = op;
1232 nfsstatsv1.cllockowners++;
1234 LIST_INSERT_HEAD(lhp, nlp, nfsl_list);
1241 * Now, update the byte ranges for locks.
1243 ret = nfscl_updatelock(lp, &nlop, &otherlop, donelocally);
1249 nfscl_clrelease(clp);
1252 * Serial modifications on the lock owner for multiple threads
1253 * for the same process using a read/write lock.
1256 nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR);
1262 free(nlp, M_NFSCLLOCKOWNER);
1264 free(nlop, M_NFSCLLOCK);
1266 free(otherlop, M_NFSCLLOCK);
1273 * Called to unlock a byte range, for LockU.
1276 nfscl_relbytelock(vnode_t vp, u_int64_t off, u_int64_t len,
1277 __unused struct ucred *cred, NFSPROC_T *p, int callcnt,
1278 struct nfsclclient *clp, void *id, int flags,
1279 struct nfscllockowner **lpp, int *dorpcp)
1281 struct nfscllockowner *lp;
1282 struct nfsclopen *op;
1283 struct nfscllock *nlop, *other_lop = NULL;
1284 struct nfscldeleg *dp;
1286 u_int8_t own[NFSV4CL_LOCKNAMELEN];
1294 * Might need these, so MALLOC them now, to
1295 * avoid a tsleep() in MALLOC later.
1298 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1299 nlop->nfslo_type = F_UNLCK;
1300 nlop->nfslo_first = off;
1301 if (len == NFS64BITSSET) {
1302 nlop->nfslo_end = NFS64BITSSET;
1304 nlop->nfslo_end = off + len;
1305 if (nlop->nfslo_end <= nlop->nfslo_first) {
1306 free(nlop, M_NFSCLLOCK);
1307 return (NFSERR_INVAL);
1312 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1315 nfscl_filllockowner(id, own, flags);
1319 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
1320 np->n_fhp->nfh_len);
1323 * First, unlock any local regions on a delegation.
1326 /* Look for this lockowner. */
1327 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
1328 if (!NFSBCMP(lp->nfsl_owner, own,
1329 NFSV4CL_LOCKNAMELEN))
1333 /* Use other_lop, so nlop is still available */
1334 (void)nfscl_updatelock(lp, &other_lop, NULL, 1);
1338 * Now, find a matching open/lockowner that hasn't already been done,
1339 * as marked by nfsl_inprog.
1343 LIST_FOREACH(op, NFSCLOPENHASH(clp, np->n_fhp->nfh_fh,
1344 np->n_fhp->nfh_len), nfso_hash) {
1345 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1346 !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1347 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1348 if (lp->nfsl_inprog == NULL &&
1349 !NFSBCMP(lp->nfsl_owner, own,
1350 NFSV4CL_LOCKNAMELEN)) {
1361 ret = nfscl_updatelock(lp, &nlop, NULL, 0);
1365 * Serial modifications on the lock owner for multiple
1366 * threads for the same process using a read/write lock.
1368 lp->nfsl_inprog = p;
1369 nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR);
1374 free(nlop, M_NFSCLLOCK);
1376 free(other_lop, M_NFSCLLOCK);
1381 * Release all lockowners marked in progess for this process and file.
1384 nfscl_releasealllocks(struct nfsclclient *clp, vnode_t vp, NFSPROC_T *p,
1385 void *id, int flags)
1387 struct nfsclopen *op;
1388 struct nfscllockowner *lp;
1390 u_int8_t own[NFSV4CL_LOCKNAMELEN];
1393 nfscl_filllockowner(id, own, flags);
1395 LIST_FOREACH(op, NFSCLOPENHASH(clp, np->n_fhp->nfh_fh,
1396 np->n_fhp->nfh_len), nfso_hash) {
1397 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1398 !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1399 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1400 if (lp->nfsl_inprog == p &&
1401 !NFSBCMP(lp->nfsl_owner, own,
1402 NFSV4CL_LOCKNAMELEN)) {
1403 lp->nfsl_inprog = NULL;
1404 nfscl_lockunlock(&lp->nfsl_rwlock);
1409 nfscl_clrelease(clp);
1414 * Called to find out if any bytes within the byte range specified are
1415 * write locked by the calling process. Used to determine if flushing
1416 * is required before a LockU.
1417 * If in doubt, return 1, so the flush will occur.
1420 nfscl_checkwritelocked(vnode_t vp, struct flock *fl,
1421 struct ucred *cred, NFSPROC_T *p, void *id, int flags)
1423 struct nfscllockowner *lp;
1424 struct nfsclopen *op;
1425 struct nfsclclient *clp;
1426 struct nfscllock *lop;
1427 struct nfscldeleg *dp;
1430 u_int8_t own[NFSV4CL_LOCKNAMELEN];
1434 switch (fl->l_whence) {
1438 * Caller is responsible for adding any necessary offset
1439 * when SEEK_CUR is used.
1444 off = np->n_size + fl->l_start;
1449 if (fl->l_len != 0) {
1450 end = off + fl->l_len;
1457 error = nfscl_getcl(vp->v_mount, cred, p, false, &clp);
1460 nfscl_filllockowner(id, own, flags);
1464 * First check the delegation locks.
1466 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
1468 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
1469 if (!NFSBCMP(lp->nfsl_owner, own,
1470 NFSV4CL_LOCKNAMELEN))
1474 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
1475 if (lop->nfslo_first >= end)
1477 if (lop->nfslo_end <= off)
1479 if (lop->nfslo_type == F_WRLCK) {
1480 nfscl_clrelease(clp);
1489 * Now, check state against the server.
1491 LIST_FOREACH(op, NFSCLOPENHASH(clp, np->n_fhp->nfh_fh,
1492 np->n_fhp->nfh_len), nfso_hash) {
1493 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1494 !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1495 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1496 if (!NFSBCMP(lp->nfsl_owner, own,
1497 NFSV4CL_LOCKNAMELEN))
1501 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
1502 if (lop->nfslo_first >= end)
1504 if (lop->nfslo_end <= off)
1506 if (lop->nfslo_type == F_WRLCK) {
1507 nfscl_clrelease(clp);
1515 nfscl_clrelease(clp);
1521 * Release a byte range lock owner structure.
1524 nfscl_lockrelease(struct nfscllockowner *lp, int error, int candelete)
1526 struct nfsclclient *clp;
1531 clp = lp->nfsl_open->nfso_own->nfsow_clp;
1532 if (error != 0 && candelete &&
1533 (lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED) == 0)
1534 nfscl_freelockowner(lp, 0);
1536 nfscl_lockunlock(&lp->nfsl_rwlock);
1537 nfscl_clrelease(clp);
1542 * Free up an open structure and any associated byte range lock structures.
1545 nfscl_freeopen(struct nfsclopen *op, int local)
1548 LIST_REMOVE(op, nfso_list);
1549 if (op->nfso_hash.le_prev != NULL)
1550 LIST_REMOVE(op, nfso_hash);
1551 nfscl_freealllocks(&op->nfso_lock, local);
1552 free(op, M_NFSCLOPEN);
1554 nfsstatsv1.cllocalopens--;
1556 nfsstatsv1.clopens--;
1560 * Free up all lock owners and associated locks.
1563 nfscl_freealllocks(struct nfscllockownerhead *lhp, int local)
1565 struct nfscllockowner *lp, *nlp;
1567 LIST_FOREACH_SAFE(lp, lhp, nfsl_list, nlp) {
1568 if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED))
1570 nfscl_freelockowner(lp, local);
1575 * Called for an Open when NFSERR_EXPIRED is received from the server.
1576 * If there are no byte range locks nor a Share Deny lost, try to do a
1577 * fresh Open. Otherwise, free the open.
1580 nfscl_expireopen(struct nfsclclient *clp, struct nfsclopen *op,
1581 struct nfsmount *nmp, struct ucred *cred, NFSPROC_T *p)
1583 struct nfscllockowner *lp;
1584 struct nfscldeleg *dp;
1585 int mustdelete = 0, error;
1588 * Look for any byte range lock(s).
1590 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1591 if (!LIST_EMPTY(&lp->nfsl_lock)) {
1598 * If no byte range lock(s) nor a Share deny, try to re-open.
1600 if (!mustdelete && (op->nfso_mode & NFSLCK_DENYBITS) == 0) {
1601 newnfs_copycred(&op->nfso_cred, cred);
1603 error = nfsrpc_reopen(nmp, op->nfso_fh,
1604 op->nfso_fhlen, op->nfso_mode, op, &dp, cred, p);
1608 free(dp, M_NFSCLDELEG);
1613 nfscl_deleg(nmp->nm_mountp, clp, op->nfso_fh,
1614 op->nfso_fhlen, cred, p, &dp);
1618 * If a byte range lock or Share deny or couldn't re-open, free it.
1621 nfscl_freeopen(op, 0);
1622 return (mustdelete);
1626 * Free up an open owner structure.
1629 nfscl_freeopenowner(struct nfsclowner *owp, int local)
1632 LIST_REMOVE(owp, nfsow_list);
1633 free(owp, M_NFSCLOWNER);
1635 nfsstatsv1.cllocalopenowners--;
1637 nfsstatsv1.clopenowners--;
1641 * Free up a byte range lock owner structure.
1644 nfscl_freelockowner(struct nfscllockowner *lp, int local)
1646 struct nfscllock *lop, *nlop;
1648 LIST_REMOVE(lp, nfsl_list);
1649 LIST_FOREACH_SAFE(lop, &lp->nfsl_lock, nfslo_list, nlop) {
1650 nfscl_freelock(lop, local);
1652 free(lp, M_NFSCLLOCKOWNER);
1654 nfsstatsv1.cllocallockowners--;
1656 nfsstatsv1.cllockowners--;
1660 * Free up a byte range lock structure.
1663 nfscl_freelock(struct nfscllock *lop, int local)
1666 LIST_REMOVE(lop, nfslo_list);
1667 free(lop, M_NFSCLLOCK);
1669 nfsstatsv1.cllocallocks--;
1671 nfsstatsv1.cllocks--;
1675 * Clean out the state related to a delegation.
1678 nfscl_cleandeleg(struct nfscldeleg *dp)
1680 struct nfsclowner *owp, *nowp;
1681 struct nfsclopen *op;
1683 LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) {
1684 op = LIST_FIRST(&owp->nfsow_open);
1686 if (LIST_NEXT(op, nfso_list) != NULL)
1687 panic("nfscleandel");
1688 nfscl_freeopen(op, 1);
1690 nfscl_freeopenowner(owp, 1);
1692 nfscl_freealllocks(&dp->nfsdl_lock, 1);
1696 * Free a delegation.
1699 nfscl_freedeleg(struct nfscldeleghead *hdp, struct nfscldeleg *dp, bool freeit)
1702 TAILQ_REMOVE(hdp, dp, nfsdl_list);
1703 LIST_REMOVE(dp, nfsdl_hash);
1705 free(dp, M_NFSCLDELEG);
1706 nfsstatsv1.cldelegates--;
1711 * Free up all state related to this client structure.
1714 nfscl_cleanclient(struct nfsclclient *clp)
1716 struct nfsclowner *owp, *nowp;
1717 struct nfsclopen *op, *nop;
1718 struct nfscllayout *lyp, *nlyp;
1719 struct nfscldevinfo *dip, *ndip;
1721 TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp)
1722 nfscl_freelayout(lyp);
1724 LIST_FOREACH_SAFE(dip, &clp->nfsc_devinfo, nfsdi_list, ndip)
1725 nfscl_freedevinfo(dip);
1727 /* Now, all the OpenOwners, etc. */
1728 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1729 LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
1730 nfscl_freeopen(op, 0);
1732 nfscl_freeopenowner(owp, 0);
1737 * Called when an NFSERR_EXPIRED is received from the server.
1740 nfscl_expireclient(struct nfsclclient *clp, struct nfsmount *nmp,
1741 struct ucred *cred, NFSPROC_T *p)
1743 struct nfsclowner *owp, *nowp, *towp;
1744 struct nfsclopen *op, *nop, *top;
1745 struct nfscldeleg *dp, *ndp;
1746 int ret, printed = 0;
1749 * First, merge locally issued Opens into the list for the server.
1751 dp = TAILQ_FIRST(&clp->nfsc_deleg);
1752 while (dp != NULL) {
1753 ndp = TAILQ_NEXT(dp, nfsdl_list);
1754 owp = LIST_FIRST(&dp->nfsdl_owner);
1755 while (owp != NULL) {
1756 nowp = LIST_NEXT(owp, nfsow_list);
1757 op = LIST_FIRST(&owp->nfsow_open);
1759 if (LIST_NEXT(op, nfso_list) != NULL)
1761 LIST_FOREACH(towp, &clp->nfsc_owner, nfsow_list) {
1762 if (!NFSBCMP(towp->nfsow_owner, owp->nfsow_owner,
1763 NFSV4CL_LOCKNAMELEN))
1767 /* Merge opens in */
1768 LIST_FOREACH(top, &towp->nfsow_open, nfso_list) {
1769 if (top->nfso_fhlen == op->nfso_fhlen &&
1770 !NFSBCMP(top->nfso_fh, op->nfso_fh,
1772 top->nfso_mode |= op->nfso_mode;
1773 top->nfso_opencnt += op->nfso_opencnt;
1778 /* Just add the open to the owner list */
1779 LIST_REMOVE(op, nfso_list);
1780 op->nfso_own = towp;
1781 LIST_INSERT_HEAD(&towp->nfsow_open, op, nfso_list);
1782 LIST_INSERT_HEAD(NFSCLOPENHASH(clp, op->nfso_fh,
1783 op->nfso_fhlen), op, nfso_hash);
1784 nfsstatsv1.cllocalopens--;
1785 nfsstatsv1.clopens++;
1788 /* Just add the openowner to the client list */
1789 LIST_REMOVE(owp, nfsow_list);
1790 owp->nfsow_clp = clp;
1791 LIST_INSERT_HEAD(&clp->nfsc_owner, owp, nfsow_list);
1792 LIST_INSERT_HEAD(NFSCLOPENHASH(clp, op->nfso_fh,
1793 op->nfso_fhlen), op, nfso_hash);
1794 nfsstatsv1.cllocalopenowners--;
1795 nfsstatsv1.clopenowners++;
1796 nfsstatsv1.cllocalopens--;
1797 nfsstatsv1.clopens++;
1802 if (!printed && !LIST_EMPTY(&dp->nfsdl_lock)) {
1804 printf("nfsv4 expired locks lost\n");
1806 nfscl_cleandeleg(dp);
1807 nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
1810 if (!TAILQ_EMPTY(&clp->nfsc_deleg))
1814 * Now, try and reopen against the server.
1816 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1817 owp->nfsow_seqid = 0;
1818 LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
1819 ret = nfscl_expireopen(clp, op, nmp, cred, p);
1820 if (ret && !printed) {
1822 printf("nfsv4 expired locks lost\n");
1825 if (LIST_EMPTY(&owp->nfsow_open))
1826 nfscl_freeopenowner(owp, 0);
1831 * This function must be called after the process represented by "own" has
1832 * exited. Must be called with CLSTATE lock held.
1835 nfscl_cleanup_common(struct nfsclclient *clp, u_int8_t *own)
1837 struct nfsclowner *owp, *nowp;
1838 struct nfscllockowner *lp, *nlp;
1839 struct nfscldeleg *dp;
1841 /* First, get rid of local locks on delegations. */
1842 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
1843 LIST_FOREACH_SAFE(lp, &dp->nfsdl_lock, nfsl_list, nlp) {
1844 if (!NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
1845 if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED))
1847 nfscl_freelockowner(lp, 1);
1851 owp = LIST_FIRST(&clp->nfsc_owner);
1852 while (owp != NULL) {
1853 nowp = LIST_NEXT(owp, nfsow_list);
1854 if (!NFSBCMP(owp->nfsow_owner, own,
1855 NFSV4CL_LOCKNAMELEN)) {
1857 * If there are children that haven't closed the
1858 * file descriptors yet, the opens will still be
1859 * here. For that case, let the renew thread clear
1860 * out the OpenOwner later.
1862 if (LIST_EMPTY(&owp->nfsow_open))
1863 nfscl_freeopenowner(owp, 0);
1865 owp->nfsow_defunct = 1;
1872 * Find open/lock owners for processes that have exited.
1875 nfscl_cleanupkext(struct nfsclclient *clp, struct nfscllockownerfhhead *lhp)
1877 struct nfsclowner *owp, *nowp;
1878 struct nfsclopen *op;
1879 struct nfscllockowner *lp, *nlp;
1880 struct nfscldeleg *dp;
1883 * All the pidhash locks must be acquired, since they are sx locks
1884 * and must be acquired before the mutexes. The pid(s) that will
1885 * be used aren't known yet, so all the locks need to be acquired.
1886 * Fortunately, this function is only performed once/sec.
1890 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1891 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1892 LIST_FOREACH_SAFE(lp, &op->nfso_lock, nfsl_list, nlp) {
1893 if (LIST_EMPTY(&lp->nfsl_lock))
1894 nfscl_emptylockowner(lp, lhp);
1897 if (nfscl_procdoesntexist(owp->nfsow_owner))
1898 nfscl_cleanup_common(clp, owp->nfsow_owner);
1902 * For the single open_owner case, these lock owners need to be
1903 * checked to see if they still exist separately.
1904 * This is because nfscl_procdoesntexist() never returns true for
1905 * the single open_owner so that the above doesn't ever call
1906 * nfscl_cleanup_common().
1908 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
1909 LIST_FOREACH_SAFE(lp, &dp->nfsdl_lock, nfsl_list, nlp) {
1910 if (nfscl_procdoesntexist(lp->nfsl_owner))
1911 nfscl_cleanup_common(clp, lp->nfsl_owner);
1915 pidhash_sunlockall();
1919 * Take the empty lock owner and move it to the local lhp list if the
1920 * associated process no longer exists.
1923 nfscl_emptylockowner(struct nfscllockowner *lp,
1924 struct nfscllockownerfhhead *lhp)
1926 struct nfscllockownerfh *lfhp, *mylfhp;
1927 struct nfscllockowner *nlp;
1930 /* If not a Posix lock owner, just return. */
1931 if ((lp->nfsl_lockflags & F_POSIX) == 0)
1937 * First, search to see if this lock owner is already in the list.
1938 * If it is, then the associated process no longer exists.
1940 SLIST_FOREACH(lfhp, lhp, nfslfh_list) {
1941 if (lfhp->nfslfh_len == lp->nfsl_open->nfso_fhlen &&
1942 !NFSBCMP(lfhp->nfslfh_fh, lp->nfsl_open->nfso_fh,
1945 LIST_FOREACH(nlp, &lfhp->nfslfh_lock, nfsl_list)
1946 if (!NFSBCMP(nlp->nfsl_owner, lp->nfsl_owner,
1947 NFSV4CL_LOCKNAMELEN))
1950 /* If not found, check if process still exists. */
1951 if (fnd_it == 0 && nfscl_procdoesntexist(lp->nfsl_owner) == 0)
1954 /* Move the lock owner over to the local list. */
1955 if (mylfhp == NULL) {
1956 mylfhp = malloc(sizeof(struct nfscllockownerfh), M_TEMP,
1960 mylfhp->nfslfh_len = lp->nfsl_open->nfso_fhlen;
1961 NFSBCOPY(lp->nfsl_open->nfso_fh, mylfhp->nfslfh_fh,
1962 mylfhp->nfslfh_len);
1963 LIST_INIT(&mylfhp->nfslfh_lock);
1964 SLIST_INSERT_HEAD(lhp, mylfhp, nfslfh_list);
1966 LIST_REMOVE(lp, nfsl_list);
1967 LIST_INSERT_HEAD(&mylfhp->nfslfh_lock, lp, nfsl_list);
1970 static int fake_global; /* Used to force visibility of MNTK_UNMOUNTF */
1972 * Called from nfs umount to free up the clientid.
1975 nfscl_umount(struct nfsmount *nmp, NFSPROC_T *p)
1977 struct nfsclclient *clp;
1982 * For the case that matters, this is the thread that set
1983 * MNTK_UNMOUNTF, so it will see it set. The code that follows is
1984 * done to ensure that any thread executing nfscl_getcl() after
1985 * this time, will see MNTK_UNMOUNTF set. nfscl_getcl() uses the
1986 * mutex for NFSLOCKCLSTATE(), so it is "m" for the following
1987 * explanation, courtesy of Alan Cox.
1988 * What follows is a snippet from Alan Cox's email at:
1989 * https://docs.FreeBSD.org/cgi/mid.cgi?BANLkTikR3d65zPHo9==08ZfJ2vmqZucEvw
1991 * 1. Set MNTK_UNMOUNTF
1992 * 2. Acquire a standard FreeBSD mutex "m".
1993 * 3. Update some data structures.
1994 * 4. Release mutex "m".
1996 * Then, other threads that acquire "m" after step 4 has occurred will
1997 * see MNTK_UNMOUNTF as set. But, other threads that beat thread X to
1998 * step 2 may or may not see MNTK_UNMOUNTF as set.
2001 if ((nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) {
2009 if ((clp->nfsc_flags & NFSCLFLAGS_INITED) == 0)
2010 panic("nfscl umount");
2013 * First, handshake with the nfscl renew thread, to terminate
2016 clp->nfsc_flags |= NFSCLFLAGS_UMOUNT;
2017 while (clp->nfsc_flags & NFSCLFLAGS_HASTHREAD)
2018 (void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT,
2022 * Now, get the exclusive lock on the client state, so
2023 * that no uses of the state are still in progress.
2026 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
2027 NFSCLSTATEMUTEXPTR, NULL);
2028 } while (!igotlock);
2032 * Free up all the state. It will expire on the server, but
2033 * maybe we should do a SetClientId/SetClientIdConfirm so
2034 * the server throws it away?
2036 LIST_REMOVE(clp, nfsc_list);
2037 nfscl_delegreturnall(clp, p);
2038 cred = newnfs_getcred();
2039 if (NFSHASNFSV4N(nmp)) {
2040 (void)nfsrpc_destroysession(nmp, clp, cred, p);
2041 (void)nfsrpc_destroyclient(nmp, clp, cred, p);
2043 (void)nfsrpc_setclient(nmp, clp, 0, NULL, cred, p);
2044 nfscl_cleanclient(clp);
2047 free(clp, M_NFSCLCLIENT);
2053 * This function is called when a server replies with NFSERR_STALECLIENTID
2054 * NFSERR_STALESTATEID or NFSERR_BADSESSION. It traverses the clientid lists,
2055 * doing Opens and Locks with reclaim. If these fail, it deletes the
2056 * corresponding state.
2059 nfscl_recover(struct nfsclclient *clp, bool *retokp, struct ucred *cred,
2062 struct nfsclowner *owp, *nowp;
2063 struct nfsclopen *op, *nop;
2064 struct nfscllockowner *lp, *nlp;
2065 struct nfscllock *lop, *nlop;
2066 struct nfscldeleg *dp, *ndp, *tdp;
2067 struct nfsmount *nmp;
2068 struct ucred *tcred;
2069 struct nfsclopenhead extra_open;
2070 struct nfscldeleghead extra_deleg;
2073 u_int32_t delegtype = NFSV4OPEN_DELEGATEWRITE, mode;
2074 int i, igotlock = 0, error, trycnt, firstlock;
2075 struct nfscllayout *lyp, *nlyp;
2079 * First, lock the client structure, so everyone else will
2080 * block when trying to use state.
2083 clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG;
2085 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
2086 NFSCLSTATEMUTEXPTR, NULL);
2087 } while (!igotlock);
2090 nmp = clp->nfsc_nmp;
2092 panic("nfscl recover");
2095 * For now, just get rid of all layouts. There may be a need
2096 * to do LayoutCommit Ops with reclaim == true later.
2098 TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp)
2099 nfscl_freelayout(lyp);
2100 TAILQ_INIT(&clp->nfsc_layout);
2101 for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++)
2102 LIST_INIT(&clp->nfsc_layouthash[i]);
2107 error = nfsrpc_setclient(nmp, clp, 1, retokp, cred, p);
2108 } while ((error == NFSERR_STALECLIENTID ||
2109 error == NFSERR_BADSESSION ||
2110 error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
2113 clp->nfsc_flags &= ~(NFSCLFLAGS_RECOVER |
2114 NFSCLFLAGS_RECVRINPROG);
2115 wakeup(&clp->nfsc_flags);
2116 nfsv4_unlock(&clp->nfsc_lock, 0);
2120 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
2121 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2124 * Mark requests already queued on the server, so that they don't
2125 * initiate another recovery cycle. Any requests already in the
2126 * queue that handle state information will have the old stale
2127 * clientid/stateid and will get a NFSERR_STALESTATEID,
2128 * NFSERR_STALECLIENTID or NFSERR_BADSESSION reply from the server.
2129 * This will be translated to NFSERR_STALEDONTRECOVER when
2130 * R_DONTRECOVER is set.
2133 TAILQ_FOREACH(rep, &nfsd_reqq, r_chain) {
2134 if (rep->r_nmp == nmp)
2135 rep->r_flags |= R_DONTRECOVER;
2140 * If nfsrpc_setclient() returns *retokp == true,
2141 * no more recovery is needed.
2147 * Now, mark all delegations "need reclaim".
2149 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list)
2150 dp->nfsdl_flags |= NFSCLDL_NEEDRECLAIM;
2152 TAILQ_INIT(&extra_deleg);
2153 LIST_INIT(&extra_open);
2155 * Now traverse the state lists, doing Open and Lock Reclaims.
2157 tcred = newnfs_getcred();
2158 recovered_one = false;
2159 owp = LIST_FIRST(&clp->nfsc_owner);
2160 while (owp != NULL) {
2161 nowp = LIST_NEXT(owp, nfsow_list);
2162 owp->nfsow_seqid = 0;
2163 op = LIST_FIRST(&owp->nfsow_open);
2164 while (op != NULL) {
2165 nop = LIST_NEXT(op, nfso_list);
2166 if (error != NFSERR_NOGRACE && error != NFSERR_BADSESSION) {
2167 /* Search for a delegation to reclaim with the open */
2168 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2169 if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
2171 if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
2172 mode = NFSV4OPEN_ACCESSWRITE;
2173 delegtype = NFSV4OPEN_DELEGATEWRITE;
2175 mode = NFSV4OPEN_ACCESSREAD;
2176 delegtype = NFSV4OPEN_DELEGATEREAD;
2178 if ((op->nfso_mode & mode) == mode &&
2179 op->nfso_fhlen == dp->nfsdl_fhlen &&
2180 !NFSBCMP(op->nfso_fh, dp->nfsdl_fh, op->nfso_fhlen))
2185 delegtype = NFSV4OPEN_DELEGATENONE;
2186 newnfs_copycred(&op->nfso_cred, tcred);
2187 error = nfscl_tryopen(nmp, NULL, op->nfso_fh,
2188 op->nfso_fhlen, op->nfso_fh, op->nfso_fhlen,
2189 op->nfso_mode, op, NULL, 0, &ndp, 1, delegtype,
2192 recovered_one = true;
2193 /* Handle any replied delegation */
2194 if (ndp != NULL && ((ndp->nfsdl_flags & NFSCLDL_WRITE)
2195 || NFSMNT_RDONLY(nmp->nm_mountp))) {
2196 if ((ndp->nfsdl_flags & NFSCLDL_WRITE))
2197 mode = NFSV4OPEN_ACCESSWRITE;
2199 mode = NFSV4OPEN_ACCESSREAD;
2200 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2201 if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
2203 if ((op->nfso_mode & mode) == mode &&
2204 op->nfso_fhlen == dp->nfsdl_fhlen &&
2205 !NFSBCMP(op->nfso_fh, dp->nfsdl_fh,
2207 dp->nfsdl_stateid = ndp->nfsdl_stateid;
2208 dp->nfsdl_sizelimit = ndp->nfsdl_sizelimit;
2209 dp->nfsdl_ace = ndp->nfsdl_ace;
2210 dp->nfsdl_change = ndp->nfsdl_change;
2211 dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
2212 if ((ndp->nfsdl_flags & NFSCLDL_RECALL))
2213 dp->nfsdl_flags |= NFSCLDL_RECALL;
2214 free(ndp, M_NFSCLDELEG);
2221 TAILQ_INSERT_HEAD(&extra_deleg, ndp, nfsdl_list);
2223 /* and reclaim all byte range locks */
2224 lp = LIST_FIRST(&op->nfso_lock);
2225 while (lp != NULL) {
2226 nlp = LIST_NEXT(lp, nfsl_list);
2229 lop = LIST_FIRST(&lp->nfsl_lock);
2230 while (lop != NULL) {
2231 nlop = LIST_NEXT(lop, nfslo_list);
2232 if (lop->nfslo_end == NFS64BITSSET)
2235 len = lop->nfslo_end - lop->nfslo_first;
2236 error = nfscl_trylock(nmp, NULL,
2237 op->nfso_fh, op->nfso_fhlen, lp,
2238 firstlock, 1, lop->nfslo_first, len,
2239 lop->nfslo_type, tcred, p);
2241 nfscl_freelock(lop, 0);
2246 /* If no locks, but a lockowner, just delete it. */
2247 if (LIST_EMPTY(&lp->nfsl_lock))
2248 nfscl_freelockowner(lp, 0);
2251 } else if (error == NFSERR_NOGRACE && !recovered_one &&
2252 NFSHASNFSV4N(nmp)) {
2254 * For NFSv4.1/4.2, the NFSERR_EXPIRED case will
2255 * actually end up here, since the client will do
2256 * a recovery for NFSERR_BADSESSION, but will get
2257 * an NFSERR_NOGRACE reply for the first "reclaim"
2259 * So, call nfscl_expireclient() to recover the
2260 * opens as best we can and then do a reclaim
2261 * complete and return.
2263 nfsrpc_reclaimcomplete(nmp, cred, p);
2264 nfscl_expireclient(clp, nmp, tcred, p);
2268 if (error != 0 && error != NFSERR_BADSESSION)
2269 nfscl_freeopen(op, 0);
2276 * Now, try and get any delegations not yet reclaimed by cobbling
2277 * to-gether an appropriate open.
2280 dp = TAILQ_FIRST(&clp->nfsc_deleg);
2281 while (dp != NULL) {
2282 ndp = TAILQ_NEXT(dp, nfsdl_list);
2283 if ((dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM)) {
2286 sizeof (struct nfsclowner), M_NFSCLOWNER, M_WAITOK);
2288 * Name must be as long an largest possible
2289 * NFSV4CL_LOCKNAMELEN. 12 for now.
2291 NFSBCOPY("RECLAIMDELEG", nowp->nfsow_owner,
2292 NFSV4CL_LOCKNAMELEN);
2293 LIST_INIT(&nowp->nfsow_open);
2294 nowp->nfsow_clp = clp;
2295 nowp->nfsow_seqid = 0;
2296 nowp->nfsow_defunct = 0;
2297 nfscl_lockinit(&nowp->nfsow_rwlock);
2300 if (error != NFSERR_NOGRACE && error != NFSERR_BADSESSION) {
2301 nop = malloc(sizeof (struct nfsclopen) +
2302 dp->nfsdl_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
2303 nop->nfso_own = nowp;
2304 if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
2305 nop->nfso_mode = NFSV4OPEN_ACCESSWRITE;
2306 delegtype = NFSV4OPEN_DELEGATEWRITE;
2308 nop->nfso_mode = NFSV4OPEN_ACCESSREAD;
2309 delegtype = NFSV4OPEN_DELEGATEREAD;
2311 nop->nfso_opencnt = 0;
2312 nop->nfso_posixlock = 1;
2313 nop->nfso_fhlen = dp->nfsdl_fhlen;
2314 NFSBCOPY(dp->nfsdl_fh, nop->nfso_fh, dp->nfsdl_fhlen);
2315 LIST_INIT(&nop->nfso_lock);
2316 nop->nfso_stateid.seqid = 0;
2317 nop->nfso_stateid.other[0] = 0;
2318 nop->nfso_stateid.other[1] = 0;
2319 nop->nfso_stateid.other[2] = 0;
2320 newnfs_copycred(&dp->nfsdl_cred, tcred);
2321 newnfs_copyincred(tcred, &nop->nfso_cred);
2323 error = nfscl_tryopen(nmp, NULL, nop->nfso_fh,
2324 nop->nfso_fhlen, nop->nfso_fh, nop->nfso_fhlen,
2325 nop->nfso_mode, nop, NULL, 0, &tdp, 1,
2326 delegtype, tcred, p);
2328 if ((tdp->nfsdl_flags & NFSCLDL_WRITE))
2329 mode = NFSV4OPEN_ACCESSWRITE;
2331 mode = NFSV4OPEN_ACCESSREAD;
2332 if ((nop->nfso_mode & mode) == mode &&
2333 nop->nfso_fhlen == tdp->nfsdl_fhlen &&
2334 !NFSBCMP(nop->nfso_fh, tdp->nfsdl_fh,
2336 dp->nfsdl_stateid = tdp->nfsdl_stateid;
2337 dp->nfsdl_sizelimit = tdp->nfsdl_sizelimit;
2338 dp->nfsdl_ace = tdp->nfsdl_ace;
2339 dp->nfsdl_change = tdp->nfsdl_change;
2340 dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
2341 if ((tdp->nfsdl_flags & NFSCLDL_RECALL))
2342 dp->nfsdl_flags |= NFSCLDL_RECALL;
2343 free(tdp, M_NFSCLDELEG);
2345 TAILQ_INSERT_HEAD(&extra_deleg, tdp, nfsdl_list);
2351 free(nop, M_NFSCLOPEN);
2352 if (error == NFSERR_NOGRACE && !recovered_one &&
2353 NFSHASNFSV4N(nmp)) {
2355 * For NFSv4.1/4.2, the NFSERR_EXPIRED case will
2356 * actually end up here, since the client will do
2357 * a recovery for NFSERR_BADSESSION, but will get
2358 * an NFSERR_NOGRACE reply for the first "reclaim"
2360 * So, call nfscl_expireclient() to recover the
2361 * opens as best we can and then do a reclaim
2362 * complete and return.
2364 nfsrpc_reclaimcomplete(nmp, cred, p);
2365 nfscl_expireclient(clp, nmp, tcred, p);
2366 free(nowp, M_NFSCLOWNER);
2370 * Couldn't reclaim it, so throw the state
2373 nfscl_cleandeleg(dp);
2374 nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
2376 recovered_one = true;
2377 LIST_INSERT_HEAD(&extra_open, nop, nfso_list);
2384 * Now, get rid of extra Opens and Delegations.
2386 LIST_FOREACH_SAFE(op, &extra_open, nfso_list, nop) {
2388 newnfs_copycred(&op->nfso_cred, tcred);
2389 error = nfscl_tryclose(op, tcred, nmp, p);
2390 if (error == NFSERR_GRACE)
2391 (void) nfs_catnap(PZERO, error, "nfsexcls");
2392 } while (error == NFSERR_GRACE);
2393 LIST_REMOVE(op, nfso_list);
2394 free(op, M_NFSCLOPEN);
2397 free(nowp, M_NFSCLOWNER);
2399 TAILQ_FOREACH_SAFE(dp, &extra_deleg, nfsdl_list, ndp) {
2401 newnfs_copycred(&dp->nfsdl_cred, tcred);
2402 error = nfscl_trydelegreturn(dp, tcred, nmp, p);
2403 if (error == NFSERR_GRACE)
2404 (void) nfs_catnap(PZERO, error, "nfsexdlg");
2405 } while (error == NFSERR_GRACE);
2406 TAILQ_REMOVE(&extra_deleg, dp, nfsdl_list);
2407 free(dp, M_NFSCLDELEG);
2410 /* For NFSv4.1 or later, do a RECLAIM_COMPLETE. */
2411 if (NFSHASNFSV4N(nmp))
2412 (void)nfsrpc_reclaimcomplete(nmp, cred, p);
2416 clp->nfsc_flags &= ~NFSCLFLAGS_RECVRINPROG;
2417 wakeup(&clp->nfsc_flags);
2418 nfsv4_unlock(&clp->nfsc_lock, 0);
2425 * This function is called when a server replies with NFSERR_EXPIRED.
2426 * It deletes all state for the client and does a fresh SetClientId/confirm.
2427 * XXX Someday it should post a signal to the process(es) that hold the
2428 * state, so they know that lock state has been lost.
2431 nfscl_hasexpired(struct nfsclclient *clp, u_int32_t clidrev, NFSPROC_T *p)
2433 struct nfsmount *nmp;
2435 int igotlock = 0, error, trycnt;
2438 * If the clientid has gone away or a new SetClientid has already
2439 * been done, just return ok.
2441 if (clp == NULL || clidrev != clp->nfsc_clientidrev)
2445 * First, lock the client structure, so everyone else will
2446 * block when trying to use state. Also, use NFSCLFLAGS_EXPIREIT so
2447 * that only one thread does the work.
2450 clp->nfsc_flags |= NFSCLFLAGS_EXPIREIT;
2452 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
2453 NFSCLSTATEMUTEXPTR, NULL);
2454 } while (!igotlock && (clp->nfsc_flags & NFSCLFLAGS_EXPIREIT));
2455 if ((clp->nfsc_flags & NFSCLFLAGS_EXPIREIT) == 0) {
2457 nfsv4_unlock(&clp->nfsc_lock, 0);
2461 clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG;
2464 nmp = clp->nfsc_nmp;
2466 panic("nfscl expired");
2467 cred = newnfs_getcred();
2470 error = nfsrpc_setclient(nmp, clp, 0, NULL, cred, p);
2471 } while ((error == NFSERR_STALECLIENTID ||
2472 error == NFSERR_BADSESSION ||
2473 error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
2476 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2479 * Expire the state for the client.
2481 nfscl_expireclient(clp, nmp, cred, p);
2483 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
2484 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2486 clp->nfsc_flags &= ~(NFSCLFLAGS_EXPIREIT | NFSCLFLAGS_RECVRINPROG);
2487 wakeup(&clp->nfsc_flags);
2488 nfsv4_unlock(&clp->nfsc_lock, 0);
2495 * This function inserts a lock in the list after insert_lop.
2498 nfscl_insertlock(struct nfscllockowner *lp, struct nfscllock *new_lop,
2499 struct nfscllock *insert_lop, int local)
2502 if ((struct nfscllockowner *)insert_lop == lp)
2503 LIST_INSERT_HEAD(&lp->nfsl_lock, new_lop, nfslo_list);
2505 LIST_INSERT_AFTER(insert_lop, new_lop, nfslo_list);
2507 nfsstatsv1.cllocallocks++;
2509 nfsstatsv1.cllocks++;
2513 * This function updates the locking for a lock owner and given file. It
2514 * maintains a list of lock ranges ordered on increasing file offset that
2515 * are NFSCLLOCK_READ or NFSCLLOCK_WRITE and non-overlapping (aka POSIX style).
2516 * It always adds new_lop to the list and sometimes uses the one pointed
2518 * Returns 1 if the locks were modified, 0 otherwise.
2521 nfscl_updatelock(struct nfscllockowner *lp, struct nfscllock **new_lopp,
2522 struct nfscllock **other_lopp, int local)
2524 struct nfscllock *new_lop = *new_lopp;
2525 struct nfscllock *lop, *tlop, *ilop;
2526 struct nfscllock *other_lop;
2527 int unlock = 0, modified = 0;
2531 * Work down the list until the lock is merged.
2533 if (new_lop->nfslo_type == F_UNLCK)
2535 ilop = (struct nfscllock *)lp;
2536 lop = LIST_FIRST(&lp->nfsl_lock);
2537 while (lop != NULL) {
2539 * Only check locks for this file that aren't before the start of
2542 if (lop->nfslo_end >= new_lop->nfslo_first) {
2543 if (new_lop->nfslo_end < lop->nfslo_first) {
2545 * If the new lock ends before the start of the
2546 * current lock's range, no merge, just insert
2551 if (new_lop->nfslo_type == lop->nfslo_type ||
2552 (new_lop->nfslo_first <= lop->nfslo_first &&
2553 new_lop->nfslo_end >= lop->nfslo_end)) {
2555 * This lock can be absorbed by the new lock/unlock.
2556 * This happens when it covers the entire range
2557 * of the old lock or is contiguous
2558 * with the old lock and is of the same type or an
2561 if (new_lop->nfslo_type != lop->nfslo_type ||
2562 new_lop->nfslo_first != lop->nfslo_first ||
2563 new_lop->nfslo_end != lop->nfslo_end)
2565 if (lop->nfslo_first < new_lop->nfslo_first)
2566 new_lop->nfslo_first = lop->nfslo_first;
2567 if (lop->nfslo_end > new_lop->nfslo_end)
2568 new_lop->nfslo_end = lop->nfslo_end;
2570 lop = LIST_NEXT(lop, nfslo_list);
2571 nfscl_freelock(tlop, local);
2576 * All these cases are for contiguous locks that are not the
2577 * same type, so they can't be merged.
2579 if (new_lop->nfslo_first <= lop->nfslo_first) {
2581 * This case is where the new lock overlaps with the
2582 * first part of the old lock. Move the start of the
2583 * old lock to just past the end of the new lock. The
2584 * new lock will be inserted in front of the old, since
2585 * ilop hasn't been updated. (We are done now.)
2587 if (lop->nfslo_first != new_lop->nfslo_end) {
2588 lop->nfslo_first = new_lop->nfslo_end;
2593 if (new_lop->nfslo_end >= lop->nfslo_end) {
2595 * This case is where the new lock overlaps with the
2596 * end of the old lock's range. Move the old lock's
2597 * end to just before the new lock's first and insert
2598 * the new lock after the old lock.
2599 * Might not be done yet, since the new lock could
2600 * overlap further locks with higher ranges.
2602 if (lop->nfslo_end != new_lop->nfslo_first) {
2603 lop->nfslo_end = new_lop->nfslo_first;
2607 lop = LIST_NEXT(lop, nfslo_list);
2611 * The final case is where the new lock's range is in the
2612 * middle of the current lock's and splits the current lock
2613 * up. Use *other_lopp to handle the second part of the
2614 * split old lock range. (We are done now.)
2615 * For unlock, we use new_lop as other_lop and tmp, since
2616 * other_lop and new_lop are the same for this case.
2617 * We noted the unlock case above, so we don't need
2618 * new_lop->nfslo_type any longer.
2620 tmp = new_lop->nfslo_first;
2622 other_lop = new_lop;
2625 other_lop = *other_lopp;
2628 other_lop->nfslo_first = new_lop->nfslo_end;
2629 other_lop->nfslo_end = lop->nfslo_end;
2630 other_lop->nfslo_type = lop->nfslo_type;
2631 lop->nfslo_end = tmp;
2632 nfscl_insertlock(lp, other_lop, lop, local);
2638 lop = LIST_NEXT(lop, nfslo_list);
2644 * Insert the new lock in the list at the appropriate place.
2647 nfscl_insertlock(lp, new_lop, ilop, local);
2655 * This function must be run as a kernel thread.
2656 * It does Renew Ops and recovery, when required.
2659 nfscl_renewthread(struct nfsclclient *clp, NFSPROC_T *p)
2661 struct nfsclowner *owp, *nowp;
2662 struct nfsclopen *op;
2663 struct nfscllockowner *lp, *nlp;
2664 struct nfscldeleghead dh;
2665 struct nfscldeleg *dp, *ndp;
2668 int error, cbpathdown, islept, igotlock, ret, clearok;
2669 uint32_t recover_done_time = 0;
2671 static time_t prevsec = 0;
2672 struct nfscllockownerfh *lfhp, *nlfhp;
2673 struct nfscllockownerfhhead lfh;
2674 struct nfscllayout *lyp, *nlyp;
2675 struct nfscldevinfo *dip, *ndip;
2676 struct nfscllayouthead rlh;
2677 struct nfsclrecalllayout *recallp;
2678 struct nfsclds *dsp;
2683 cred = newnfs_getcred();
2685 clp->nfsc_flags |= NFSCLFLAGS_HASTHREAD;
2686 mp = clp->nfsc_nmp->nm_mountp;
2689 newnfs_setroot(cred);
2691 if (clp->nfsc_flags & NFSCLFLAGS_RECOVER) {
2693 * Only allow one full recover within 1/2 of the lease
2694 * duration (nfsc_renew).
2695 * retok is value/result. If passed in set to true,
2696 * it indicates only a CreateSession operation should
2698 * If it is returned true, it indicates that the
2699 * recovery only required a CreateSession.
2702 if (recover_done_time < NFSD_MONOSEC) {
2703 recover_done_time = NFSD_MONOSEC +
2707 NFSCL_DEBUG(1, "Doing recovery, only "
2708 "createsession=%d\n", retok);
2709 nfscl_recover(clp, &retok, cred, p);
2711 if (clp->nfsc_expire <= NFSD_MONOSEC &&
2712 (clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID)) {
2713 clp->nfsc_expire = NFSD_MONOSEC + clp->nfsc_renew;
2714 clidrev = clp->nfsc_clientidrev;
2715 error = nfsrpc_renew(clp, NULL, cred, p);
2716 if (error == NFSERR_CBPATHDOWN)
2718 else if (error == NFSERR_STALECLIENTID ||
2719 error == NFSERR_BADSESSION) {
2721 clp->nfsc_flags |= NFSCLFLAGS_RECOVER;
2723 } else if (error == NFSERR_EXPIRED)
2724 (void) nfscl_hasexpired(clp, clidrev, p);
2728 if (NFSHASNFSV4N(clp->nfsc_nmp)) {
2729 /* Do renews for any DS sessions. */
2730 NFSLOCKMNT(clp->nfsc_nmp);
2731 /* Skip first entry, since the MDS is handled above. */
2732 dsp = TAILQ_FIRST(&clp->nfsc_nmp->nm_sess);
2734 dsp = TAILQ_NEXT(dsp, nfsclds_list);
2735 while (dsp != NULL) {
2736 if (dsp->nfsclds_expire <= NFSD_MONOSEC &&
2737 dsp->nfsclds_sess.nfsess_defunct == 0) {
2738 dsp->nfsclds_expire = NFSD_MONOSEC +
2740 NFSUNLOCKMNT(clp->nfsc_nmp);
2741 (void)nfsrpc_renew(clp, dsp, cred, p);
2744 dsp = TAILQ_NEXT(dsp, nfsclds_list);
2746 NFSUNLOCKMNT(clp->nfsc_nmp);
2752 /* It's a Total Recall! */
2753 nfscl_totalrecall(clp);
2756 * Now, handle defunct owners.
2758 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
2759 if (LIST_EMPTY(&owp->nfsow_open)) {
2760 if (owp->nfsow_defunct != 0)
2761 nfscl_freeopenowner(owp, 0);
2766 * Do the recall on any delegations. To avoid trouble, always
2767 * come back up here after having slept.
2771 dp = TAILQ_FIRST(&clp->nfsc_deleg);
2772 while (dp != NULL) {
2773 ndp = TAILQ_NEXT(dp, nfsdl_list);
2774 if ((dp->nfsdl_flags & NFSCLDL_RECALL)) {
2776 * Wait for outstanding I/O ops to be done.
2778 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
2780 nfsv4_unlock(&clp->nfsc_lock, 0);
2783 dp->nfsdl_rwlock.nfslock_lock |=
2785 msleep(&dp->nfsdl_rwlock,
2786 NFSCLSTATEMUTEXPTR, PVFS, "nfscld",
2788 if (NFSCL_FORCEDISM(mp))
2793 igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
2794 &islept, NFSCLSTATEMUTEXPTR, mp);
2795 if (igotlock == 0 && NFSCL_FORCEDISM(mp))
2801 newnfs_copycred(&dp->nfsdl_cred, cred);
2802 ret = nfscl_recalldeleg(clp, clp->nfsc_nmp, dp,
2803 NULL, cred, p, 1, &vp);
2805 nfscl_cleandeleg(dp);
2806 TAILQ_REMOVE(&clp->nfsc_deleg, dp,
2808 LIST_REMOVE(dp, nfsdl_hash);
2809 TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
2811 nfsstatsv1.cldelegates--;
2815 * The nfsc_lock must be released before doing
2816 * vrele(), since it might call nfs_inactive().
2817 * For the unlikely case where the vnode failed
2818 * to be acquired by nfscl_recalldeleg(), a
2819 * VOP_RECLAIM() should be in progress and it
2820 * will return the delegation.
2822 nfsv4_unlock(&clp->nfsc_lock, 0);
2835 * Clear out old delegations, if we are above the high water
2836 * mark. Only clear out ones with no state related to them.
2837 * The tailq list is in LRU order.
2839 dp = TAILQ_LAST(&clp->nfsc_deleg, nfscldeleghead);
2840 while (nfscl_delegcnt > nfscl_deleghighwater && dp != NULL) {
2841 ndp = TAILQ_PREV(dp, nfscldeleghead, nfsdl_list);
2842 if (dp->nfsdl_rwlock.nfslock_usecnt == 0 &&
2843 dp->nfsdl_rwlock.nfslock_lock == 0 &&
2844 dp->nfsdl_timestamp < NFSD_MONOSEC &&
2845 (dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_ZAPPED |
2846 NFSCLDL_NEEDRECLAIM | NFSCLDL_DELEGRET)) == 0) {
2848 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
2849 op = LIST_FIRST(&owp->nfsow_open);
2856 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
2857 if (!LIST_EMPTY(&lp->nfsl_lock)) {
2864 TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
2865 LIST_REMOVE(dp, nfsdl_hash);
2866 TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
2868 nfsstatsv1.cldelegates--;
2874 nfsv4_unlock(&clp->nfsc_lock, 0);
2877 * Do the recall on any layouts. To avoid trouble, always
2878 * come back up here after having slept.
2882 TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp) {
2883 if ((lyp->nfsly_flags & NFSLY_RECALL) != 0) {
2885 * Wait for outstanding I/O ops to be done.
2887 if (lyp->nfsly_lock.nfslock_usecnt > 0 ||
2888 (lyp->nfsly_lock.nfslock_lock &
2889 NFSV4LOCK_LOCK) != 0) {
2890 lyp->nfsly_lock.nfslock_lock |=
2892 msleep(&lyp->nfsly_lock.nfslock_lock,
2893 NFSCLSTATEMUTEXPTR, PVFS, "nfslyp",
2895 if (NFSCL_FORCEDISM(mp))
2899 /* Move the layout to the recall list. */
2900 TAILQ_REMOVE(&clp->nfsc_layout, lyp,
2902 LIST_REMOVE(lyp, nfsly_hash);
2903 TAILQ_INSERT_HEAD(&rlh, lyp, nfsly_list);
2905 /* Handle any layout commits. */
2906 if (!NFSHASNOLAYOUTCOMMIT(clp->nfsc_nmp) &&
2907 (lyp->nfsly_flags & NFSLY_WRITTEN) != 0) {
2908 lyp->nfsly_flags &= ~NFSLY_WRITTEN;
2910 NFSCL_DEBUG(3, "do layoutcommit\n");
2911 nfscl_dolayoutcommit(clp->nfsc_nmp, lyp,
2919 /* Now, look for stale layouts. */
2920 lyp = TAILQ_LAST(&clp->nfsc_layout, nfscllayouthead);
2921 while (lyp != NULL) {
2922 nlyp = TAILQ_PREV(lyp, nfscllayouthead, nfsly_list);
2923 if (lyp->nfsly_timestamp < NFSD_MONOSEC &&
2924 (lyp->nfsly_flags & NFSLY_RECALL) == 0 &&
2925 lyp->nfsly_lock.nfslock_usecnt == 0 &&
2926 lyp->nfsly_lock.nfslock_lock == 0) {
2927 NFSCL_DEBUG(4, "ret stale lay=%d\n",
2929 recallp = malloc(sizeof(*recallp),
2930 M_NFSLAYRECALL, M_NOWAIT);
2931 if (recallp == NULL)
2933 (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE,
2934 lyp, NFSLAYOUTIOMODE_ANY, 0, UINT64_MAX,
2935 lyp->nfsly_stateid.seqid, 0, 0, NULL,
2942 * Free up any unreferenced device info structures.
2944 LIST_FOREACH_SAFE(dip, &clp->nfsc_devinfo, nfsdi_list, ndip) {
2945 if (dip->nfsdi_layoutrefs == 0 &&
2946 dip->nfsdi_refcnt == 0) {
2947 NFSCL_DEBUG(4, "freeing devinfo\n");
2948 LIST_REMOVE(dip, nfsdi_list);
2949 nfscl_freedevinfo(dip);
2954 /* Do layout return(s), as required. */
2955 TAILQ_FOREACH_SAFE(lyp, &rlh, nfsly_list, nlyp) {
2956 TAILQ_REMOVE(&rlh, lyp, nfsly_list);
2957 NFSCL_DEBUG(4, "ret layout\n");
2958 nfscl_layoutreturn(clp->nfsc_nmp, lyp, cred, p);
2959 nfscl_freelayout(lyp);
2963 * Delegreturn any delegations cleaned out or recalled.
2965 TAILQ_FOREACH_SAFE(dp, &dh, nfsdl_list, ndp) {
2966 newnfs_copycred(&dp->nfsdl_cred, cred);
2967 (void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
2968 TAILQ_REMOVE(&dh, dp, nfsdl_list);
2969 free(dp, M_NFSCLDELEG);
2974 * Call nfscl_cleanupkext() once per second to check for
2975 * open/lock owners where the process has exited.
2977 mytime = NFSD_MONOSEC;
2978 if (prevsec != mytime) {
2980 nfscl_cleanupkext(clp, &lfh);
2984 * Do a ReleaseLockOwner for all lock owners where the
2985 * associated process no longer exists, as found by
2986 * nfscl_cleanupkext().
2988 newnfs_setroot(cred);
2989 SLIST_FOREACH_SAFE(lfhp, &lfh, nfslfh_list, nlfhp) {
2990 LIST_FOREACH_SAFE(lp, &lfhp->nfslfh_lock, nfsl_list,
2992 (void)nfsrpc_rellockown(clp->nfsc_nmp, lp,
2993 lfhp->nfslfh_fh, lfhp->nfslfh_len, cred,
2995 nfscl_freelockowner(lp, 0);
3002 if ((clp->nfsc_flags & NFSCLFLAGS_RECOVER) == 0)
3003 (void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT, "nfscl",
3006 if (clp->nfsc_flags & NFSCLFLAGS_UMOUNT) {
3007 clp->nfsc_flags &= ~NFSCLFLAGS_HASTHREAD;
3010 wakeup((caddr_t)clp);
3018 * Initiate state recovery. Called when NFSERR_STALECLIENTID,
3019 * NFSERR_STALESTATEID or NFSERR_BADSESSION is received.
3022 nfscl_initiate_recovery(struct nfsclclient *clp)
3028 clp->nfsc_flags |= NFSCLFLAGS_RECOVER;
3030 wakeup((caddr_t)clp);
3034 * Dump out the state stuff for debugging.
3037 nfscl_dumpstate(struct nfsmount *nmp, int openowner, int opens,
3038 int lockowner, int locks)
3040 struct nfsclclient *clp;
3041 struct nfsclowner *owp;
3042 struct nfsclopen *op;
3043 struct nfscllockowner *lp;
3044 struct nfscllock *lop;
3045 struct nfscldeleg *dp;
3049 printf("nfscl dumpstate NULL clp\n");
3053 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
3054 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
3055 if (openowner && !LIST_EMPTY(&owp->nfsow_open))
3056 printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
3057 owp->nfsow_owner[0], owp->nfsow_owner[1],
3058 owp->nfsow_owner[2], owp->nfsow_owner[3],
3060 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3062 printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
3063 op->nfso_stateid.other[0], op->nfso_stateid.other[1],
3064 op->nfso_stateid.other[2], op->nfso_opencnt,
3066 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
3068 printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
3069 lp->nfsl_owner[0], lp->nfsl_owner[1],
3070 lp->nfsl_owner[2], lp->nfsl_owner[3],
3072 lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
3073 lp->nfsl_stateid.other[2]);
3074 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
3077 printf("lck typ=%d fst=%ju end=%ju\n",
3078 lop->nfslo_type, (intmax_t)lop->nfslo_first,
3079 (intmax_t)lop->nfslo_end);
3081 printf("lck typ=%d fst=%qd end=%qd\n",
3082 lop->nfslo_type, lop->nfslo_first,
3090 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3091 if (openowner && !LIST_EMPTY(&owp->nfsow_open))
3092 printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
3093 owp->nfsow_owner[0], owp->nfsow_owner[1],
3094 owp->nfsow_owner[2], owp->nfsow_owner[3],
3096 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3098 printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
3099 op->nfso_stateid.other[0], op->nfso_stateid.other[1],
3100 op->nfso_stateid.other[2], op->nfso_opencnt,
3102 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
3104 printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
3105 lp->nfsl_owner[0], lp->nfsl_owner[1],
3106 lp->nfsl_owner[2], lp->nfsl_owner[3],
3108 lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
3109 lp->nfsl_stateid.other[2]);
3110 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
3113 printf("lck typ=%d fst=%ju end=%ju\n",
3114 lop->nfslo_type, (intmax_t)lop->nfslo_first,
3115 (intmax_t)lop->nfslo_end);
3117 printf("lck typ=%d fst=%qd end=%qd\n",
3118 lop->nfslo_type, lop->nfslo_first,
3129 * Check for duplicate open owners and opens.
3130 * (Only used as a diagnostic aid.)
3133 nfscl_dupopen(vnode_t vp, int dupopens)
3135 struct nfsclclient *clp;
3136 struct nfsclowner *owp, *owp2;
3137 struct nfsclopen *op, *op2;
3140 clp = VFSTONFS(vp->v_mount)->nm_clp;
3142 printf("nfscl dupopen NULL clp\n");
3145 nfhp = VTONFS(vp)->n_fhp;
3149 * First, search for duplicate owners.
3150 * These should never happen!
3152 LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3153 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3155 !NFSBCMP(owp->nfsow_owner, owp2->nfsow_owner,
3156 NFSV4CL_LOCKNAMELEN)) {
3158 printf("DUP OWNER\n");
3159 nfscl_dumpstate(VFSTONFS(vp->v_mount), 1, 1, 0, 0);
3166 * Now, search for duplicate stateids.
3167 * These shouldn't happen, either.
3169 LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3170 LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
3171 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3172 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3174 (op->nfso_stateid.other[0] != 0 ||
3175 op->nfso_stateid.other[1] != 0 ||
3176 op->nfso_stateid.other[2] != 0) &&
3177 op->nfso_stateid.other[0] == op2->nfso_stateid.other[0] &&
3178 op->nfso_stateid.other[1] == op2->nfso_stateid.other[1] &&
3179 op->nfso_stateid.other[2] == op2->nfso_stateid.other[2]) {
3181 printf("DUP STATEID\n");
3182 nfscl_dumpstate(VFSTONFS(vp->v_mount), 1, 1, 0, 0);
3191 * Now search for duplicate opens.
3192 * Duplicate opens for the same owner
3193 * should never occur. Other duplicates are
3194 * possible and are checked for if "dupopens"
3197 LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3198 LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
3199 if (nfhp->nfh_len == op2->nfso_fhlen &&
3200 !NFSBCMP(nfhp->nfh_fh, op2->nfso_fh, nfhp->nfh_len)) {
3201 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3202 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3203 if (op != op2 && nfhp->nfh_len == op->nfso_fhlen &&
3204 !NFSBCMP(nfhp->nfh_fh, op->nfso_fh, nfhp->nfh_len) &&
3205 (!NFSBCMP(op->nfso_own->nfsow_owner,
3206 op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN) ||
3208 if (!NFSBCMP(op->nfso_own->nfsow_owner,
3209 op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
3211 printf("BADDUP OPEN\n");
3214 printf("DUP OPEN\n");
3216 nfscl_dumpstate(VFSTONFS(vp->v_mount), 1, 1, 0,
3229 * During close, find an open that needs to be dereferenced and
3230 * dereference it. If there are no more opens for this file,
3231 * log a message to that effect.
3232 * Opens aren't actually Close'd until VOP_INACTIVE() is performed
3233 * on the file's vnode.
3234 * This is the safe way, since it is difficult to identify
3235 * which open the close is for and I/O can be performed after the
3236 * close(2) system call when a file is mmap'd.
3237 * If it returns 0 for success, there will be a referenced
3238 * clp returned via clpp.
3241 nfscl_getclose(vnode_t vp, struct nfsclclient **clpp)
3243 struct nfsclclient *clp;
3244 struct nfsclowner *owp;
3245 struct nfsclopen *op;
3246 struct nfscldeleg *dp;
3250 error = nfscl_getcl(vp->v_mount, NULL, NULL, false, &clp);
3255 nfhp = VTONFS(vp)->n_fhp;
3259 * First, look for one under a delegation that was locally issued
3260 * and just decrement the opencnt for it. Since all my Opens against
3261 * the server are DENY_NONE, I don't see a problem with hanging
3262 * onto them. (It is much easier to use one of the extant Opens
3263 * that I already have on the server when a Delegation is recalled
3264 * than to do fresh Opens.) Someday, I might need to rethink this, but.
3266 dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
3268 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
3269 op = LIST_FIRST(&owp->nfsow_open);
3272 * Since a delegation is for a file, there
3273 * should never be more than one open for
3276 if (LIST_NEXT(op, nfso_list) != NULL)
3277 panic("nfscdeleg opens");
3278 if (notdecr && op->nfso_opencnt > 0) {
3287 /* Now process the opens against the server. */
3288 LIST_FOREACH(op, NFSCLOPENHASH(clp, nfhp->nfh_fh, nfhp->nfh_len),
3290 if (op->nfso_fhlen == nfhp->nfh_len &&
3291 !NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
3293 /* Found an open, decrement cnt if possible */
3294 if (notdecr && op->nfso_opencnt > 0) {
3299 * There are more opens, so just return.
3301 if (op->nfso_opencnt > 0) {
3309 printf("nfscl: never fnd open\n");
3314 nfscl_doclose(vnode_t vp, struct nfsclclient **clpp, NFSPROC_T *p)
3316 struct nfsclclient *clp;
3317 struct nfsclowner *owp, *nowp;
3318 struct nfsclopen *op;
3319 struct nfscldeleg *dp;
3321 struct nfsclrecalllayout *recallp;
3324 error = nfscl_getcl(vp->v_mount, NULL, NULL, false, &clp);
3329 nfhp = VTONFS(vp)->n_fhp;
3330 recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_WAITOK);
3333 * First get rid of the local Open structures, which should be no
3336 dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
3338 LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) {
3339 op = LIST_FIRST(&owp->nfsow_open);
3341 KASSERT((op->nfso_opencnt == 0),
3342 ("nfscl: bad open cnt on deleg"));
3343 nfscl_freeopen(op, 1);
3345 nfscl_freeopenowner(owp, 1);
3349 /* Return any layouts marked return on close. */
3350 nfscl_retoncloselayout(vp, clp, nfhp->nfh_fh, nfhp->nfh_len, &recallp);
3352 /* Now process the opens against the server. */
3354 LIST_FOREACH(op, NFSCLOPENHASH(clp, nfhp->nfh_fh, nfhp->nfh_len),
3356 if (op->nfso_fhlen == nfhp->nfh_len &&
3357 !NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
3359 /* Found an open, close it. */
3361 KASSERT((op->nfso_opencnt == 0),
3362 ("nfscl: bad open cnt on server (%d)",
3366 nfsrpc_doclose(VFSTONFS(vp->v_mount), op, p);
3373 * recallp has been set NULL by nfscl_retoncloselayout() if it was
3374 * used by the function, but calling free() with a NULL pointer is ok.
3376 free(recallp, M_NFSLAYRECALL);
3381 * Return all delegations on this client.
3382 * (Must be called with client sleep lock.)
3385 nfscl_delegreturnall(struct nfsclclient *clp, NFSPROC_T *p)
3387 struct nfscldeleg *dp, *ndp;
3390 cred = newnfs_getcred();
3391 TAILQ_FOREACH_SAFE(dp, &clp->nfsc_deleg, nfsdl_list, ndp) {
3392 nfscl_cleandeleg(dp);
3393 (void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
3394 nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
3400 * Return any delegation for this vp.
3403 nfscl_delegreturnvp(vnode_t vp, NFSPROC_T *p)
3405 struct nfsclclient *clp;
3406 struct nfscldeleg *dp;
3409 struct nfsmount *nmp;
3411 nmp = VFSTONFS(vp->v_mount);
3413 if ((nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0) {
3419 cred = newnfs_getcred();
3424 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
3425 np->n_fhp->nfh_len);
3427 nfscl_cleandeleg(dp);
3428 nfscl_freedeleg(&clp->nfsc_deleg, dp, false);
3430 newnfs_copycred(&dp->nfsdl_cred, cred);
3431 nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
3432 free(dp, M_NFSCLDELEG);
3439 * Do a callback RPC.
3442 nfscl_docb(struct nfsrv_descript *nd, NFSPROC_T *p)
3444 int clist, gotseq_ok, i, j, k, op, rcalls;
3446 struct nfsclclient *clp;
3447 struct nfscldeleg *dp = NULL;
3448 int numops, taglen = -1, error = 0, trunc __unused;
3449 u_int32_t minorvers = 0, retops = 0, *retopsp = NULL, *repp, cbident;
3450 u_char tag[NFSV4_SMALLSTR + 1], *tagstr;
3456 nfsattrbit_t attrbits, rattrbits;
3457 nfsv4stateid_t stateid;
3458 uint32_t seqid, slotid = 0, highslot, cachethis __unused;
3459 uint8_t sessionid[NFSX_V4SESSIONID];
3461 struct nfscllayout *lyp;
3462 uint64_t filesid[2], len, off;
3463 int changed, gotone, laytype, recalltype;
3465 struct nfsclrecalllayout *recallp = NULL;
3466 struct nfsclsession *tsep;
3470 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3471 taglen = fxdr_unsigned(int, *tl);
3476 if (taglen <= NFSV4_SMALLSTR)
3479 tagstr = malloc(taglen + 1, M_TEMP, M_WAITOK);
3480 error = nfsrv_mtostr(nd, tagstr, taglen);
3482 if (taglen > NFSV4_SMALLSTR)
3483 free(tagstr, M_TEMP);
3487 (void) nfsm_strtom(nd, tag, taglen);
3488 if (taglen > NFSV4_SMALLSTR) {
3489 free(tagstr, M_TEMP);
3491 NFSM_BUILD(retopsp, u_int32_t *, NFSX_UNSIGNED);
3492 NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3493 minorvers = fxdr_unsigned(u_int32_t, *tl++);
3494 if (minorvers != NFSV4_MINORVERSION &&
3495 minorvers != NFSV41_MINORVERSION &&
3496 minorvers != NFSV42_MINORVERSION)
3497 nd->nd_repstat = NFSERR_MINORVERMISMATCH;
3498 cbident = fxdr_unsigned(u_int32_t, *tl++);
3502 numops = fxdr_unsigned(int, *tl);
3504 * Loop around doing the sub ops.
3506 for (i = 0; i < numops; i++) {
3507 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3508 NFSM_BUILD(repp, u_int32_t *, 2 * NFSX_UNSIGNED);
3510 op = fxdr_unsigned(int, *tl);
3511 if (op < NFSV4OP_CBGETATTR ||
3512 (op > NFSV4OP_CBRECALL && minorvers == NFSV4_MINORVERSION) ||
3513 (op > NFSV4OP_CBNOTIFYDEVID &&
3514 minorvers == NFSV41_MINORVERSION) ||
3515 (op > NFSV4OP_CBOFFLOAD &&
3516 minorvers == NFSV42_MINORVERSION)) {
3517 nd->nd_repstat = NFSERR_OPILLEGAL;
3518 *repp = nfscl_errmap(nd, minorvers);
3522 nd->nd_procnum = op;
3523 if (op < NFSV42_CBNOPS)
3524 nfsstatsv1.cbrpccnt[nd->nd_procnum]++;
3526 case NFSV4OP_CBGETATTR:
3527 NFSCL_DEBUG(4, "cbgetattr\n");
3530 error = nfsm_getfh(nd, &nfhp);
3532 error = nfsrv_getattrbits(nd, &attrbits,
3534 if (error == 0 && i == 0 &&
3535 minorvers != NFSV4_MINORVERSION)
3536 error = NFSERR_OPNOTINSESS;
3538 mp = nfscl_getmnt(minorvers, sessionid, cbident,
3541 error = NFSERR_SERVERFAULT;
3544 error = nfscl_ngetreopen(mp, nfhp->nfh_fh,
3545 nfhp->nfh_len, p, &np);
3550 NFSZERO_ATTRBIT(&rattrbits);
3552 dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
3555 if (NFSISSET_ATTRBIT(&attrbits,
3558 va.va_size = np->n_size;
3562 NFSSETBIT_ATTRBIT(&rattrbits,
3565 if (NFSISSET_ATTRBIT(&attrbits,
3566 NFSATTRBIT_CHANGE)) {
3570 (np->n_flag & NDELEGMOD))
3572 NFSSETBIT_ATTRBIT(&rattrbits,
3576 error = NFSERR_SERVERFAULT;
3584 free(nfhp, M_NFSFH);
3586 (void) nfsv4_fillattr(nd, NULL, NULL, NULL, &va,
3587 NULL, 0, &rattrbits, NULL, p, 0, 0, 0, 0,
3590 case NFSV4OP_CBRECALL:
3591 NFSCL_DEBUG(4, "cbrecall\n");
3592 NFSM_DISSECT(tl, u_int32_t *, NFSX_STATEID +
3594 stateid.seqid = *tl++;
3595 NFSBCOPY((caddr_t)tl, (caddr_t)stateid.other,
3597 tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
3598 trunc = fxdr_unsigned(int, *tl);
3599 error = nfsm_getfh(nd, &nfhp);
3600 if (error == 0 && i == 0 &&
3601 minorvers != NFSV4_MINORVERSION)
3602 error = NFSERR_OPNOTINSESS;
3605 if (minorvers == NFSV4_MINORVERSION)
3606 clp = nfscl_getclnt(cbident);
3608 clp = nfscl_getclntsess(sessionid);
3610 dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
3612 if (dp != NULL && (dp->nfsdl_flags &
3613 NFSCLDL_DELEGRET) == 0) {
3616 wakeup((caddr_t)clp);
3619 error = NFSERR_SERVERFAULT;
3624 free(nfhp, M_NFSFH);
3626 case NFSV4OP_CBLAYOUTRECALL:
3627 NFSCL_DEBUG(4, "cblayrec\n");
3629 NFSM_DISSECT(tl, uint32_t *, 4 * NFSX_UNSIGNED);
3630 laytype = fxdr_unsigned(int, *tl++);
3631 iomode = fxdr_unsigned(uint32_t, *tl++);
3632 if (newnfs_true == *tl++)
3636 recalltype = fxdr_unsigned(int, *tl);
3637 NFSCL_DEBUG(4, "layt=%d iom=%d ch=%d rectyp=%d\n",
3638 laytype, iomode, changed, recalltype);
3639 recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL,
3641 if (laytype != NFSLAYOUT_NFSV4_1_FILES &&
3642 laytype != NFSLAYOUT_FLEXFILE)
3643 error = NFSERR_NOMATCHLAYOUT;
3644 else if (recalltype == NFSLAYOUTRETURN_FILE) {
3645 error = nfsm_getfh(nd, &nfhp);
3646 NFSCL_DEBUG(4, "retfile getfh=%d\n", error);
3649 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_HYPER +
3651 off = fxdr_hyper(tl); tl += 2;
3652 len = fxdr_hyper(tl); tl += 2;
3653 stateid.seqid = fxdr_unsigned(uint32_t, *tl++);
3654 NFSBCOPY(tl, stateid.other, NFSX_STATEIDOTHER);
3655 if (minorvers == NFSV4_MINORVERSION)
3656 error = NFSERR_NOTSUPP;
3658 error = NFSERR_OPNOTINSESS;
3659 NFSCL_DEBUG(4, "off=%ju len=%ju sq=%u err=%d\n",
3660 (uintmax_t)off, (uintmax_t)len,
3661 stateid.seqid, error);
3664 clp = nfscl_getclntsess(sessionid);
3665 NFSCL_DEBUG(4, "cbly clp=%p\n", clp);
3667 lyp = nfscl_findlayout(clp,
3670 NFSCL_DEBUG(4, "cblyp=%p\n",
3675 NFSLY_FLEXFILE)) != 0 &&
3676 !NFSBCMP(stateid.other,
3677 lyp->nfsly_stateid.other,
3678 NFSX_STATEIDOTHER)) {
3688 lyp->nfsly_stateid.seqid)
3689 lyp->nfsly_stateid.seqid =
3697 lyp->nfsly_stateid.seqid);
3700 NFSERR_NOMATCHLAYOUT;
3702 error = NFSERR_NOMATCHLAYOUT;
3705 free(nfhp, M_NFSFH);
3706 } else if (recalltype == NFSLAYOUTRETURN_FSID) {
3707 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_HYPER);
3708 filesid[0] = fxdr_hyper(tl); tl += 2;
3709 filesid[1] = fxdr_hyper(tl); tl += 2;
3712 clp = nfscl_getclntsess(sessionid);
3714 TAILQ_FOREACH(lyp, &clp->nfsc_layout,
3716 if (lyp->nfsly_filesid[0] ==
3718 lyp->nfsly_filesid[1] ==
3725 lyp->nfsly_stateid.seqid,
3735 error = NFSERR_NOMATCHLAYOUT;
3737 error = NFSERR_NOMATCHLAYOUT;
3739 } else if (recalltype == NFSLAYOUTRETURN_ALL) {
3742 clp = nfscl_getclntsess(sessionid);
3744 TAILQ_FOREACH(lyp, &clp->nfsc_layout,
3746 error = nfscl_layoutrecall(
3747 recalltype, lyp, iomode, 0,
3749 lyp->nfsly_stateid.seqid,
3750 0, 0, NULL, recallp);
3757 error = NFSERR_NOMATCHLAYOUT;
3759 error = NFSERR_NOMATCHLAYOUT;
3762 error = NFSERR_NOMATCHLAYOUT;
3763 if (recallp != NULL) {
3764 free(recallp, M_NFSLAYRECALL);
3768 case NFSV4OP_CBSEQUENCE:
3769 NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID +
3771 bcopy(tl, sessionid, NFSX_V4SESSIONID);
3772 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3773 seqid = fxdr_unsigned(uint32_t, *tl++);
3774 slotid = fxdr_unsigned(uint32_t, *tl++);
3775 highslot = fxdr_unsigned(uint32_t, *tl++);
3777 /* Throw away the referring call stuff. */
3778 clist = fxdr_unsigned(int, *tl);
3779 for (j = 0; j < clist; j++) {
3780 NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID +
3782 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3783 rcalls = fxdr_unsigned(int, *tl);
3784 for (k = 0; k < rcalls; k++) {
3785 NFSM_DISSECT(tl, uint32_t *,
3791 clp = nfscl_getclntsess(sessionid);
3793 error = NFSERR_SERVERFAULT;
3795 error = NFSERR_SEQUENCEPOS;
3797 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3798 error = nfsv4_seqsession(seqid, slotid,
3799 highslot, tsep->nfsess_cbslots, &rep,
3800 tsep->nfsess_backslots);
3803 if (error == 0 || error == NFSERR_REPLYFROMCACHE) {
3807 * Handle a reply for a retried
3808 * callback. The reply will be
3809 * re-inserted in the session cache
3810 * by the nfsv4_seqsess_cacherep() call
3813 KASSERT(error == NFSERR_REPLYFROMCACHE,
3814 ("cbsequence: non-NULL rep"));
3815 NFSCL_DEBUG(4, "Got cbretry\n");
3816 m_freem(nd->nd_mreq);
3821 NFSM_BUILD(tl, uint32_t *,
3822 NFSX_V4SESSIONID + 4 * NFSX_UNSIGNED);
3823 bcopy(sessionid, tl, NFSX_V4SESSIONID);
3824 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3825 *tl++ = txdr_unsigned(seqid);
3826 *tl++ = txdr_unsigned(slotid);
3827 *tl++ = txdr_unsigned(NFSV4_CBSLOTS - 1);
3828 *tl = txdr_unsigned(NFSV4_CBSLOTS - 1);
3832 if (i == 0 && minorvers != NFSV4_MINORVERSION)
3833 error = NFSERR_OPNOTINSESS;
3835 NFSCL_DEBUG(1, "unsupp callback %d\n", op);
3836 error = NFSERR_NOTSUPP;
3841 if (error == EBADRPC || error == NFSERR_BADXDR) {
3842 nd->nd_repstat = NFSERR_BADXDR;
3844 nd->nd_repstat = error;
3849 if (nd->nd_repstat) {
3850 *repp = nfscl_errmap(nd, minorvers);
3853 *repp = 0; /* NFS4_OK */
3856 if (recallp != NULL)
3857 free(recallp, M_NFSLAYRECALL);
3859 if (error == EBADRPC || error == NFSERR_BADXDR)
3860 nd->nd_repstat = NFSERR_BADXDR;
3862 printf("nfsv4 comperr1=%d\n", error);
3865 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
3869 *retopsp = txdr_unsigned(retops);
3871 *nd->nd_errp = nfscl_errmap(nd, minorvers);
3873 if (gotseq_ok != 0) {
3874 rep = m_copym(nd->nd_mreq, 0, M_COPYALL, M_WAITOK);
3876 clp = nfscl_getclntsess(sessionid);
3878 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3879 nfsv4_seqsess_cacherep(slotid, tsep->nfsess_cbslots,
3890 * Generate the next cbident value. Basically just increment a static value
3891 * and then check that it isn't already in the list, if it has wrapped around.
3894 nfscl_nextcbident(void)
3896 struct nfsclclient *clp;
3898 static u_int32_t nextcbident = 0;
3899 static int haswrapped = 0;
3902 if (nextcbident == 0)
3906 * Search the clientid list for one already using this cbident.
3911 LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3912 if (clp->nfsc_cbident == nextcbident) {
3922 return (nextcbident);
3926 * Get the mount point related to a given cbident or session and busy it.
3929 nfscl_getmnt(int minorvers, uint8_t *sessionid, u_int32_t cbident,
3930 struct nfsclclient **clpp)
3932 struct nfsclclient *clp;
3935 struct nfsclsession *tsep;
3939 LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3940 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3941 if (minorvers == NFSV4_MINORVERSION) {
3942 if (clp->nfsc_cbident == cbident)
3944 } else if (!NFSBCMP(tsep->nfsess_sessionid, sessionid,
3952 mp = clp->nfsc_nmp->nm_mountp;
3955 error = vfs_busy(mp, 0);
3964 * Get the clientid pointer related to a given cbident.
3966 static struct nfsclclient *
3967 nfscl_getclnt(u_int32_t cbident)
3969 struct nfsclclient *clp;
3971 LIST_FOREACH(clp, &nfsclhead, nfsc_list)
3972 if (clp->nfsc_cbident == cbident)
3978 * Get the clientid pointer related to a given sessionid.
3980 static struct nfsclclient *
3981 nfscl_getclntsess(uint8_t *sessionid)
3983 struct nfsclclient *clp;
3984 struct nfsclsession *tsep;
3986 LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3987 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3988 if (!NFSBCMP(tsep->nfsess_sessionid, sessionid,
3996 * Search for a lock conflict locally on the client. A conflict occurs if
3997 * - not same owner and overlapping byte range and at least one of them is
3998 * a write lock or this is an unlock.
4001 nfscl_localconflict(struct nfsclclient *clp, u_int8_t *fhp, int fhlen,
4002 struct nfscllock *nlop, u_int8_t *own, struct nfscldeleg *dp,
4003 struct nfscllock **lopp)
4005 struct nfsclopen *op;
4009 ret = nfscl_checkconflict(&dp->nfsdl_lock, nlop, own, lopp);
4013 LIST_FOREACH(op, NFSCLOPENHASH(clp, fhp, fhlen), nfso_hash) {
4014 if (op->nfso_fhlen == fhlen &&
4015 !NFSBCMP(op->nfso_fh, fhp, fhlen)) {
4016 ret = nfscl_checkconflict(&op->nfso_lock, nlop,
4026 nfscl_checkconflict(struct nfscllockownerhead *lhp, struct nfscllock *nlop,
4027 u_int8_t *own, struct nfscllock **lopp)
4029 struct nfscllockowner *lp;
4030 struct nfscllock *lop;
4032 LIST_FOREACH(lp, lhp, nfsl_list) {
4033 if (NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
4034 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
4035 if (lop->nfslo_first >= nlop->nfslo_end)
4037 if (lop->nfslo_end <= nlop->nfslo_first)
4039 if (lop->nfslo_type == F_WRLCK ||
4040 nlop->nfslo_type == F_WRLCK ||
4041 nlop->nfslo_type == F_UNLCK) {
4044 return (NFSERR_DENIED);
4053 * Check for a local conflicting lock.
4056 nfscl_lockt(vnode_t vp, struct nfsclclient *clp, u_int64_t off,
4057 u_int64_t len, struct flock *fl, NFSPROC_T *p, void *id, int flags)
4059 struct nfscllock *lop, nlck;
4060 struct nfscldeleg *dp;
4062 u_int8_t own[NFSV4CL_LOCKNAMELEN];
4065 nlck.nfslo_type = fl->l_type;
4066 nlck.nfslo_first = off;
4067 if (len == NFS64BITSSET) {
4068 nlck.nfslo_end = NFS64BITSSET;
4070 nlck.nfslo_end = off + len;
4071 if (nlck.nfslo_end <= nlck.nfslo_first)
4072 return (NFSERR_INVAL);
4075 nfscl_filllockowner(id, own, flags);
4077 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4078 error = nfscl_localconflict(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len,
4079 &nlck, own, dp, &lop);
4081 fl->l_whence = SEEK_SET;
4082 fl->l_start = lop->nfslo_first;
4083 if (lop->nfslo_end == NFS64BITSSET)
4086 fl->l_len = lop->nfslo_end - lop->nfslo_first;
4087 fl->l_pid = (pid_t)0;
4088 fl->l_type = lop->nfslo_type;
4089 error = -1; /* no RPC required */
4090 } else if (dp != NULL && ((dp->nfsdl_flags & NFSCLDL_WRITE) ||
4091 fl->l_type == F_RDLCK)) {
4093 * The delegation ensures that there isn't a conflicting
4094 * lock on the server, so return -1 to indicate an RPC
4097 fl->l_type = F_UNLCK;
4105 * Handle Recall of a delegation.
4106 * The clp must be exclusive locked when this is called.
4109 nfscl_recalldeleg(struct nfsclclient *clp, struct nfsmount *nmp,
4110 struct nfscldeleg *dp, vnode_t vp, struct ucred *cred, NFSPROC_T *p,
4111 int called_from_renewthread, vnode_t *vpp)
4113 struct nfsclowner *owp, *lowp, *nowp;
4114 struct nfsclopen *op, *lop;
4115 struct nfscllockowner *lp;
4116 struct nfscllock *lckp;
4121 KASSERT(vpp != NULL, ("nfscl_recalldeleg: vpp NULL"));
4124 * First, get a vnode for the file. This is needed to do RPCs.
4126 ret = nfscl_ngetreopen(nmp->nm_mountp, dp->nfsdl_fh,
4127 dp->nfsdl_fhlen, p, &np);
4130 * File isn't open, so nothing to move over to the
4140 dp->nfsdl_flags &= ~NFSCLDL_MODTIMESET;
4143 * Ok, if it's a write delegation, flush data to the server, so
4144 * that close/open consistency is retained.
4148 if ((dp->nfsdl_flags & NFSCLDL_WRITE) && (np->n_flag & NMODIFIED)) {
4149 np->n_flag |= NDELEGRECALL;
4151 ret = ncl_flush(vp, MNT_WAIT, p, 1, called_from_renewthread);
4153 np->n_flag &= ~NDELEGRECALL;
4155 NFSINVALATTRCACHE(np);
4157 if (ret == EIO && called_from_renewthread != 0) {
4159 * If the flush failed with EIO for the renew thread,
4160 * return now, so that the dirty buffer will be flushed
4167 * Now, for each openowner with opens issued locally, move them
4168 * over to state against the server.
4170 LIST_FOREACH(lowp, &dp->nfsdl_owner, nfsow_list) {
4171 lop = LIST_FIRST(&lowp->nfsow_open);
4173 if (LIST_NEXT(lop, nfso_list) != NULL)
4174 panic("nfsdlg mult opens");
4176 * Look for the same openowner against the server.
4178 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
4179 if (!NFSBCMP(lowp->nfsow_owner,
4180 owp->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
4181 newnfs_copycred(&dp->nfsdl_cred, cred);
4182 ret = nfscl_moveopen(vp, clp, nmp, lop,
4184 if (ret == NFSERR_STALECLIENTID ||
4185 ret == NFSERR_STALEDONTRECOVER ||
4186 ret == NFSERR_BADSESSION)
4189 nfscl_freeopen(lop, 1);
4198 * If no openowner found, create one and get an open
4203 sizeof (struct nfsclowner), M_NFSCLOWNER,
4205 nfscl_newopen(clp, NULL, &owp, &nowp, &op,
4206 NULL, lowp->nfsow_owner, dp->nfsdl_fh,
4207 dp->nfsdl_fhlen, NULL, NULL);
4208 newnfs_copycred(&dp->nfsdl_cred, cred);
4209 ret = nfscl_moveopen(vp, clp, nmp, lop,
4212 nfscl_freeopenowner(owp, 0);
4213 if (ret == NFSERR_STALECLIENTID ||
4214 ret == NFSERR_STALEDONTRECOVER ||
4215 ret == NFSERR_BADSESSION)
4218 nfscl_freeopen(lop, 1);
4228 * Now, get byte range locks for any locks done locally.
4230 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4231 LIST_FOREACH(lckp, &lp->nfsl_lock, nfslo_list) {
4232 newnfs_copycred(&dp->nfsdl_cred, cred);
4233 ret = nfscl_relock(vp, clp, nmp, lp, lckp, cred, p);
4234 if (ret == NFSERR_STALESTATEID ||
4235 ret == NFSERR_STALEDONTRECOVER ||
4236 ret == NFSERR_STALECLIENTID ||
4237 ret == NFSERR_BADSESSION)
4247 * Move a locally issued open over to an owner on the state list.
4248 * SIDE EFFECT: If it needs to sleep (do an rpc), it unlocks clstate and
4249 * returns with it unlocked.
4252 nfscl_moveopen(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp,
4253 struct nfsclopen *lop, struct nfsclowner *owp, struct nfscldeleg *dp,
4254 struct ucred *cred, NFSPROC_T *p)
4256 struct nfsclopen *op, *nop;
4257 struct nfscldeleg *ndp;
4259 int error = 0, newone;
4262 * First, look for an appropriate open, If found, just increment the
4265 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
4266 if ((op->nfso_mode & lop->nfso_mode) == lop->nfso_mode &&
4267 op->nfso_fhlen == lop->nfso_fhlen &&
4268 !NFSBCMP(op->nfso_fh, lop->nfso_fh, op->nfso_fhlen)) {
4269 op->nfso_opencnt += lop->nfso_opencnt;
4270 nfscl_freeopen(lop, 1);
4275 /* No appropriate open, so we have to do one against the server. */
4277 nop = malloc(sizeof (struct nfsclopen) +
4278 lop->nfso_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
4279 nop->nfso_hash.le_prev = NULL;
4281 nfscl_newopen(clp, NULL, &owp, NULL, &op, &nop, owp->nfsow_owner,
4282 lop->nfso_fh, lop->nfso_fhlen, cred, &newone);
4284 error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data, np->n_v4->n4_fhlen,
4285 lop->nfso_fh, lop->nfso_fhlen, lop->nfso_mode, op,
4286 NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, &ndp, 0, 0, cred, p);
4289 nfscl_freeopen(op, 0);
4291 op->nfso_mode |= lop->nfso_mode;
4292 op->nfso_opencnt += lop->nfso_opencnt;
4293 nfscl_freeopen(lop, 1);
4296 free(nop, M_NFSCLOPEN);
4299 * What should I do with the returned delegation, since the
4300 * delegation is being recalled? For now, just printf and
4303 printf("Moveopen returned deleg\n");
4304 free(ndp, M_NFSCLDELEG);
4310 * Recall all delegations on this client.
4313 nfscl_totalrecall(struct nfsclclient *clp)
4315 struct nfscldeleg *dp;
4317 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
4318 if ((dp->nfsdl_flags & NFSCLDL_DELEGRET) == 0)
4319 dp->nfsdl_flags |= NFSCLDL_RECALL;
4324 * Relock byte ranges. Called for delegation recall and state expiry.
4327 nfscl_relock(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp,
4328 struct nfscllockowner *lp, struct nfscllock *lop, struct ucred *cred,
4331 struct nfscllockowner *nlp;
4335 int error, newone, donelocally;
4337 if (NFSHASNFSV4N(nmp) && NFSHASONEOPENOWN(nmp)) {
4340 np->n_flag |= NMIGHTBELOCKED;
4344 off = lop->nfslo_first;
4345 len = lop->nfslo_end - lop->nfslo_first;
4346 error = nfscl_getbytelock(vp, off, len, lop->nfslo_type, cred, p,
4347 clp, 1, NULL, lp->nfsl_lockflags, lp->nfsl_owner,
4348 lp->nfsl_openowner, &nlp, &newone, &donelocally);
4349 if (error || donelocally)
4351 nfhp = VTONFS(vp)->n_fhp;
4352 error = nfscl_trylock(nmp, vp, nfhp->nfh_fh,
4353 nfhp->nfh_len, nlp, newone, 0, off,
4354 len, lop->nfslo_type, cred, p);
4356 nfscl_freelockowner(nlp, 0);
4361 * Called to re-open a file. Basically get a vnode for the file handle
4362 * and then call nfsrpc_openrpc() to do the rest.
4365 nfsrpc_reopen(struct nfsmount *nmp, u_int8_t *fhp, int fhlen,
4366 u_int32_t mode, struct nfsclopen *op, struct nfscldeleg **dpp,
4367 struct ucred *cred, NFSPROC_T *p)
4373 error = nfscl_ngetreopen(nmp->nm_mountp, fhp, fhlen, p, &np);
4377 if (np->n_v4 != NULL) {
4378 error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data,
4379 np->n_v4->n4_fhlen, fhp, fhlen, mode, op,
4380 NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, dpp, 0, 0,
4390 * Try an open against the server. Just call nfsrpc_openrpc(), retrying while
4391 * NFSERR_DELAY. Also, try system credentials, if the passed in credentials
4395 nfscl_tryopen(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen,
4396 u_int8_t *newfhp, int newfhlen, u_int32_t mode, struct nfsclopen *op,
4397 u_int8_t *name, int namelen, struct nfscldeleg **ndpp,
4398 int reclaim, u_int32_t delegtype, struct ucred *cred, NFSPROC_T *p)
4403 error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp, newfhlen,
4404 mode, op, name, namelen, ndpp, reclaim, delegtype, cred, p,
4406 if (error == NFSERR_DELAY)
4407 (void) nfs_catnap(PZERO, error, "nfstryop");
4408 } while (error == NFSERR_DELAY);
4409 if (error == EAUTH || error == EACCES) {
4410 /* Try again using system credentials */
4411 newnfs_setroot(cred);
4413 error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp,
4414 newfhlen, mode, op, name, namelen, ndpp, reclaim,
4415 delegtype, cred, p, 1, 0);
4416 if (error == NFSERR_DELAY)
4417 (void) nfs_catnap(PZERO, error, "nfstryop");
4418 } while (error == NFSERR_DELAY);
4424 * Try a byte range lock. Just loop on nfsrpc_lock() while it returns
4425 * NFSERR_DELAY. Also, retry with system credentials, if the provided
4429 nfscl_trylock(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp,
4430 int fhlen, struct nfscllockowner *nlp, int newone, int reclaim,
4431 u_int64_t off, u_int64_t len, short type, struct ucred *cred, NFSPROC_T *p)
4433 struct nfsrv_descript nfsd, *nd = &nfsd;
4437 error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp, newone,
4438 reclaim, off, len, type, cred, p, 0);
4439 if (!error && nd->nd_repstat == NFSERR_DELAY)
4440 (void) nfs_catnap(PZERO, (int)nd->nd_repstat,
4442 } while (!error && nd->nd_repstat == NFSERR_DELAY);
4444 error = nd->nd_repstat;
4445 if (error == EAUTH || error == EACCES) {
4446 /* Try again using root credentials */
4447 newnfs_setroot(cred);
4449 error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp,
4450 newone, reclaim, off, len, type, cred, p, 1);
4451 if (!error && nd->nd_repstat == NFSERR_DELAY)
4452 (void) nfs_catnap(PZERO, (int)nd->nd_repstat,
4454 } while (!error && nd->nd_repstat == NFSERR_DELAY);
4456 error = nd->nd_repstat;
4462 * Try a delegreturn against the server. Just call nfsrpc_delegreturn(),
4463 * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
4467 nfscl_trydelegreturn(struct nfscldeleg *dp, struct ucred *cred,
4468 struct nfsmount *nmp, NFSPROC_T *p)
4473 error = nfsrpc_delegreturn(dp, cred, nmp, p, 0);
4474 if (error == NFSERR_DELAY)
4475 (void) nfs_catnap(PZERO, error, "nfstrydp");
4476 } while (error == NFSERR_DELAY);
4477 if (error == EAUTH || error == EACCES) {
4478 /* Try again using system credentials */
4479 newnfs_setroot(cred);
4481 error = nfsrpc_delegreturn(dp, cred, nmp, p, 1);
4482 if (error == NFSERR_DELAY)
4483 (void) nfs_catnap(PZERO, error, "nfstrydp");
4484 } while (error == NFSERR_DELAY);
4490 * Try a close against the server. Just call nfsrpc_closerpc(),
4491 * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
4495 nfscl_tryclose(struct nfsclopen *op, struct ucred *cred,
4496 struct nfsmount *nmp, NFSPROC_T *p)
4498 struct nfsrv_descript nfsd, *nd = &nfsd;
4502 error = nfsrpc_closerpc(nd, nmp, op, cred, p, 0);
4503 if (error == NFSERR_DELAY)
4504 (void) nfs_catnap(PZERO, error, "nfstrycl");
4505 } while (error == NFSERR_DELAY);
4506 if (error == EAUTH || error == EACCES) {
4507 /* Try again using system credentials */
4508 newnfs_setroot(cred);
4510 error = nfsrpc_closerpc(nd, nmp, op, cred, p, 1);
4511 if (error == NFSERR_DELAY)
4512 (void) nfs_catnap(PZERO, error, "nfstrycl");
4513 } while (error == NFSERR_DELAY);
4519 * Decide if a delegation on a file permits close without flushing writes
4520 * to the server. This might be a big performance win in some environments.
4521 * (Not useful until the client does caching on local stable storage.)
4524 nfscl_mustflush(vnode_t vp)
4526 struct nfsclclient *clp;
4527 struct nfscldeleg *dp;
4529 struct nfsmount *nmp;
4532 nmp = VFSTONFS(vp->v_mount);
4533 if (!NFSHASNFSV4(nmp))
4536 clp = nfscl_findcl(nmp);
4541 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4542 if (dp != NULL && (dp->nfsdl_flags &
4543 (NFSCLDL_WRITE | NFSCLDL_RECALL | NFSCLDL_DELEGRET)) ==
4545 (dp->nfsdl_sizelimit >= np->n_size ||
4546 !NFSHASSTRICT3530(nmp))) {
4555 * See if a (write) delegation exists for this file.
4558 nfscl_nodeleg(vnode_t vp, int writedeleg)
4560 struct nfsclclient *clp;
4561 struct nfscldeleg *dp;
4563 struct nfsmount *nmp;
4566 nmp = VFSTONFS(vp->v_mount);
4567 if (!NFSHASNFSV4(nmp))
4570 if ((nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0) {
4576 clp = nfscl_findcl(nmp);
4581 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4583 (dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_DELEGRET)) == 0 &&
4584 (writedeleg == 0 || (dp->nfsdl_flags & NFSCLDL_WRITE) ==
4594 * Look for an associated delegation that should be DelegReturned.
4597 nfscl_removedeleg(vnode_t vp, NFSPROC_T *p, nfsv4stateid_t *stp)
4599 struct nfsclclient *clp;
4600 struct nfscldeleg *dp;
4601 struct nfsclowner *owp;
4602 struct nfscllockowner *lp;
4603 struct nfsmount *nmp;
4606 int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
4608 nmp = VFSTONFS(vp->v_mount);
4612 * Loop around waiting for:
4613 * - outstanding I/O operations on delegations to complete
4614 * - for a delegation on vp that has state, lock the client and
4616 * - return delegation with no state
4619 clp = nfscl_findcl(nmp);
4624 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4625 np->n_fhp->nfh_len);
4628 * Wait for outstanding I/O ops to be done.
4630 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4632 nfsv4_unlock(&clp->nfsc_lock, 0);
4635 dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4636 (void) nfsmsleep(&dp->nfsdl_rwlock,
4637 NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4641 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4642 if (!LIST_EMPTY(&owp->nfsow_open)) {
4648 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4649 if (!LIST_EMPTY(&lp->nfsl_lock)) {
4655 if (needsrecall && !triedrecall) {
4656 dp->nfsdl_flags |= NFSCLDL_DELEGRET;
4659 igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
4660 &islept, NFSCLSTATEMUTEXPTR, NULL);
4667 cred = newnfs_getcred();
4668 newnfs_copycred(&dp->nfsdl_cred, cred);
4669 nfscl_recalldeleg(clp, nmp, dp, vp, cred, p, 0, NULL);
4673 nfsv4_unlock(&clp->nfsc_lock, 0);
4677 *stp = dp->nfsdl_stateid;
4679 nfscl_cleandeleg(dp);
4680 nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
4683 nfsv4_unlock(&clp->nfsc_lock, 0);
4690 * Look for associated delegation(s) that should be DelegReturned.
4693 nfscl_renamedeleg(vnode_t fvp, nfsv4stateid_t *fstp, int *gotfdp, vnode_t tvp,
4694 nfsv4stateid_t *tstp, int *gottdp, NFSPROC_T *p)
4696 struct nfsclclient *clp;
4697 struct nfscldeleg *dp;
4698 struct nfsclowner *owp;
4699 struct nfscllockowner *lp;
4700 struct nfsmount *nmp;
4703 int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
4705 nmp = VFSTONFS(fvp->v_mount);
4710 * Loop around waiting for:
4711 * - outstanding I/O operations on delegations to complete
4712 * - for a delegation on fvp that has state, lock the client and
4714 * - return delegation(s) with no state.
4717 clp = nfscl_findcl(nmp);
4723 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4724 np->n_fhp->nfh_len);
4725 if (dp != NULL && *gotfdp == 0) {
4727 * Wait for outstanding I/O ops to be done.
4729 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4731 nfsv4_unlock(&clp->nfsc_lock, 0);
4734 dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4735 (void) nfsmsleep(&dp->nfsdl_rwlock,
4736 NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4740 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4741 if (!LIST_EMPTY(&owp->nfsow_open)) {
4747 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4748 if (!LIST_EMPTY(&lp->nfsl_lock)) {
4754 if (needsrecall && !triedrecall) {
4755 dp->nfsdl_flags |= NFSCLDL_DELEGRET;
4758 igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
4759 &islept, NFSCLSTATEMUTEXPTR, NULL);
4766 cred = newnfs_getcred();
4767 newnfs_copycred(&dp->nfsdl_cred, cred);
4768 nfscl_recalldeleg(clp, nmp, dp, fvp, cred, p, 0, NULL);
4772 nfsv4_unlock(&clp->nfsc_lock, 0);
4776 *fstp = dp->nfsdl_stateid;
4779 nfscl_cleandeleg(dp);
4780 nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
4783 nfsv4_unlock(&clp->nfsc_lock, 0);
4788 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4789 np->n_fhp->nfh_len);
4790 if (dp != NULL && *gottdp == 0) {
4792 * Wait for outstanding I/O ops to be done.
4794 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4795 dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4796 (void) nfsmsleep(&dp->nfsdl_rwlock,
4797 NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4800 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4801 if (!LIST_EMPTY(&owp->nfsow_open)) {
4806 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4807 if (!LIST_EMPTY(&lp->nfsl_lock)) {
4812 *tstp = dp->nfsdl_stateid;
4815 nfscl_cleandeleg(dp);
4816 nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
4825 * Get a reference on the clientid associated with the mount point.
4826 * Return 1 if success, 0 otherwise.
4829 nfscl_getref(struct nfsmount *nmp)
4831 struct nfsclclient *clp;
4834 clp = nfscl_findcl(nmp);
4839 nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, NULL);
4845 * Release a reference on a clientid acquired with the above call.
4848 nfscl_relref(struct nfsmount *nmp)
4850 struct nfsclclient *clp;
4853 clp = nfscl_findcl(nmp);
4858 nfsv4_relref(&clp->nfsc_lock);
4863 * Save the size attribute in the delegation, since the nfsnode
4867 nfscl_reclaimnode(vnode_t vp)
4869 struct nfsclclient *clp;
4870 struct nfscldeleg *dp;
4871 struct nfsnode *np = VTONFS(vp);
4872 struct nfsmount *nmp;
4874 nmp = VFSTONFS(vp->v_mount);
4875 if (!NFSHASNFSV4(nmp))
4878 clp = nfscl_findcl(nmp);
4883 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4884 if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE))
4885 dp->nfsdl_size = np->n_size;
4890 * Get the saved size attribute in the delegation, since it is a
4891 * newly allocated nfsnode.
4894 nfscl_newnode(vnode_t vp)
4896 struct nfsclclient *clp;
4897 struct nfscldeleg *dp;
4898 struct nfsnode *np = VTONFS(vp);
4899 struct nfsmount *nmp;
4901 nmp = VFSTONFS(vp->v_mount);
4902 if (!NFSHASNFSV4(nmp))
4905 clp = nfscl_findcl(nmp);
4910 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4911 if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE))
4912 np->n_size = dp->nfsdl_size;
4917 * If there is a valid write delegation for this file, set the modtime
4918 * to the local clock time.
4921 nfscl_delegmodtime(vnode_t vp)
4923 struct nfsclclient *clp;
4924 struct nfscldeleg *dp;
4925 struct nfsnode *np = VTONFS(vp);
4926 struct nfsmount *nmp;
4928 nmp = VFSTONFS(vp->v_mount);
4929 if (!NFSHASNFSV4(nmp))
4932 if ((nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0) {
4938 clp = nfscl_findcl(nmp);
4943 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4944 if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE)) {
4945 nanotime(&dp->nfsdl_modtime);
4946 dp->nfsdl_flags |= NFSCLDL_MODTIMESET;
4952 * If there is a valid write delegation for this file with a modtime set,
4953 * put that modtime in mtime.
4956 nfscl_deleggetmodtime(vnode_t vp, struct timespec *mtime)
4958 struct nfsclclient *clp;
4959 struct nfscldeleg *dp;
4960 struct nfsnode *np = VTONFS(vp);
4961 struct nfsmount *nmp;
4963 nmp = VFSTONFS(vp->v_mount);
4964 if (!NFSHASNFSV4(nmp))
4967 if ((nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0) {
4973 clp = nfscl_findcl(nmp);
4978 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4980 (dp->nfsdl_flags & (NFSCLDL_WRITE | NFSCLDL_MODTIMESET)) ==
4981 (NFSCLDL_WRITE | NFSCLDL_MODTIMESET))
4982 *mtime = dp->nfsdl_modtime;
4987 nfscl_errmap(struct nfsrv_descript *nd, u_int32_t minorvers)
4989 short *defaulterrp, *errp;
4991 if (!nd->nd_repstat)
4993 if (nd->nd_procnum == NFSPROC_NOOP)
4994 return (txdr_unsigned(nd->nd_repstat & 0xffff));
4995 if (nd->nd_repstat == EBADRPC)
4996 return (txdr_unsigned(NFSERR_BADXDR));
4997 if (nd->nd_repstat == NFSERR_MINORVERMISMATCH ||
4998 nd->nd_repstat == NFSERR_OPILLEGAL)
4999 return (txdr_unsigned(nd->nd_repstat));
5000 if (nd->nd_repstat >= NFSERR_BADIOMODE && nd->nd_repstat < 20000 &&
5001 minorvers > NFSV4_MINORVERSION) {
5002 /* NFSv4.n error. */
5003 return (txdr_unsigned(nd->nd_repstat));
5005 if (nd->nd_procnum < NFSV4OP_CBNOPS)
5006 errp = defaulterrp = nfscl_cberrmap[nd->nd_procnum];
5008 return (txdr_unsigned(nd->nd_repstat));
5010 if (*errp == (short)nd->nd_repstat)
5011 return (txdr_unsigned(nd->nd_repstat));
5012 return (txdr_unsigned(*defaulterrp));
5016 * Called to find/add a layout to a client.
5017 * This function returns the layout with a refcnt (shared lock) upon
5018 * success (returns 0) or with no lock/refcnt on the layout when an
5019 * error is returned.
5020 * If a layout is passed in via lypp, it is locked (exclusively locked).
5023 nfscl_layout(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen,
5024 nfsv4stateid_t *stateidp, int layouttype, int retonclose,
5025 struct nfsclflayouthead *fhlp, struct nfscllayout **lypp,
5026 struct ucred *cred, NFSPROC_T *p)
5028 struct nfsclclient *clp;
5029 struct nfscllayout *lyp, *tlyp;
5030 struct nfsclflayout *flp;
5031 struct nfsnode *np = VTONFS(vp);
5033 int layout_passed_in;
5035 mp = nmp->nm_mountp;
5036 layout_passed_in = 1;
5040 layout_passed_in = 0;
5041 tlyp = malloc(sizeof(*tlyp) + fhlen - 1, M_NFSLAYOUT,
5048 if (layout_passed_in != 0)
5049 nfsv4_unlock(&lyp->nfsly_lock, 0);
5052 free(tlyp, M_NFSLAYOUT);
5057 * Although no lyp was passed in, another thread might have
5058 * allocated one. If one is found, just increment it's ref
5059 * count and return it.
5061 lyp = nfscl_findlayout(clp, fhp, fhlen);
5065 lyp->nfsly_stateid.seqid = stateidp->seqid;
5066 lyp->nfsly_stateid.other[0] = stateidp->other[0];
5067 lyp->nfsly_stateid.other[1] = stateidp->other[1];
5068 lyp->nfsly_stateid.other[2] = stateidp->other[2];
5069 lyp->nfsly_lastbyte = 0;
5070 LIST_INIT(&lyp->nfsly_flayread);
5071 LIST_INIT(&lyp->nfsly_flayrw);
5072 LIST_INIT(&lyp->nfsly_recall);
5073 lyp->nfsly_filesid[0] = np->n_vattr.na_filesid[0];
5074 lyp->nfsly_filesid[1] = np->n_vattr.na_filesid[1];
5075 lyp->nfsly_clp = clp;
5076 if (layouttype == NFSLAYOUT_FLEXFILE)
5077 lyp->nfsly_flags = NFSLY_FLEXFILE;
5079 lyp->nfsly_flags = NFSLY_FILES;
5080 if (retonclose != 0)
5081 lyp->nfsly_flags |= NFSLY_RETONCLOSE;
5082 lyp->nfsly_fhlen = fhlen;
5083 NFSBCOPY(fhp, lyp->nfsly_fh, fhlen);
5084 TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
5085 LIST_INSERT_HEAD(NFSCLLAYOUTHASH(clp, fhp, fhlen), lyp,
5087 lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
5090 if (retonclose != 0)
5091 lyp->nfsly_flags |= NFSLY_RETONCLOSE;
5092 if (stateidp->seqid > lyp->nfsly_stateid.seqid)
5093 lyp->nfsly_stateid.seqid = stateidp->seqid;
5094 TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list);
5095 TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
5096 lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
5098 nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
5099 if (NFSCL_FORCEDISM(mp)) {
5102 free(tlyp, M_NFSLAYOUT);
5106 } else if (stateidp->seqid > lyp->nfsly_stateid.seqid)
5107 lyp->nfsly_stateid.seqid = stateidp->seqid;
5109 /* Merge the new list of File Layouts into the list. */
5110 flp = LIST_FIRST(fhlp);
5112 if (flp->nfsfl_iomode == NFSLAYOUTIOMODE_READ)
5113 nfscl_mergeflayouts(&lyp->nfsly_flayread, fhlp);
5115 nfscl_mergeflayouts(&lyp->nfsly_flayrw, fhlp);
5117 if (layout_passed_in != 0)
5118 nfsv4_unlock(&lyp->nfsly_lock, 1);
5121 free(tlyp, M_NFSLAYOUT);
5126 * Search for a layout by MDS file handle.
5127 * If one is found, it is returned with a refcnt (shared lock) iff
5128 * retflpp returned non-NULL and locked (exclusive locked) iff retflpp is
5131 struct nfscllayout *
5132 nfscl_getlayout(struct nfsclclient *clp, uint8_t *fhp, int fhlen,
5133 uint64_t off, struct nfsclflayout **retflpp, int *recalledp)
5135 struct nfscllayout *lyp;
5137 int error, igotlock;
5139 mp = clp->nfsc_nmp->nm_mountp;
5143 lyp = nfscl_findlayout(clp, fhp, fhlen);
5145 if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
5146 TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list);
5147 TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
5148 lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
5149 error = nfscl_findlayoutforio(lyp, off,
5150 NFSV4OPEN_ACCESSREAD, retflpp);
5152 nfsv4_getref(&lyp->nfsly_lock, NULL,
5153 NFSCLSTATEMUTEXPTR, mp);
5156 igotlock = nfsv4_lock(&lyp->nfsly_lock,
5157 1, NULL, NFSCLSTATEMUTEXPTR, mp);
5158 } while (igotlock == 0 && !NFSCL_FORCEDISM(mp));
5161 if (NFSCL_FORCEDISM(mp)) {
5175 * Search for a layout by MDS file handle. If one is found, mark in to be
5176 * recalled, if it already marked "return on close".
5179 nfscl_retoncloselayout(vnode_t vp, struct nfsclclient *clp, uint8_t *fhp,
5180 int fhlen, struct nfsclrecalllayout **recallpp)
5182 struct nfscllayout *lyp;
5185 if (vp->v_type != VREG || !NFSHASPNFS(VFSTONFS(vp->v_mount)) ||
5186 nfscl_enablecallb == 0 || nfs_numnfscbd == 0 ||
5187 (VTONFS(vp)->n_flag & NNOLAYOUT) != 0)
5189 lyp = nfscl_findlayout(clp, fhp, fhlen);
5190 if (lyp != NULL && (lyp->nfsly_flags & (NFSLY_RETONCLOSE |
5191 NFSLY_RECALL)) == NFSLY_RETONCLOSE) {
5193 if (!LIST_EMPTY(&lyp->nfsly_flayread))
5194 iomode |= NFSLAYOUTIOMODE_READ;
5195 if (!LIST_EMPTY(&lyp->nfsly_flayrw))
5196 iomode |= NFSLAYOUTIOMODE_RW;
5197 (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE, lyp, iomode,
5198 0, UINT64_MAX, lyp->nfsly_stateid.seqid, 0, 0, NULL,
5200 NFSCL_DEBUG(4, "retoncls recall iomode=%d\n", iomode);
5206 * Mark the layout to be recalled and with an error.
5207 * Also, disable the dsp from further use.
5210 nfscl_dserr(uint32_t op, uint32_t stat, struct nfscldevinfo *dp,
5211 struct nfscllayout *lyp, struct nfsclds *dsp)
5213 struct nfsclrecalllayout *recallp;
5216 printf("DS being disabled, error=%d\n", stat);
5217 /* Set up the return of the layout. */
5218 recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_WAITOK);
5221 if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
5222 if (!LIST_EMPTY(&lyp->nfsly_flayread))
5223 iomode |= NFSLAYOUTIOMODE_READ;
5224 if (!LIST_EMPTY(&lyp->nfsly_flayrw))
5225 iomode |= NFSLAYOUTIOMODE_RW;
5226 (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE, lyp, iomode,
5227 0, UINT64_MAX, lyp->nfsly_stateid.seqid, stat, op,
5228 dp->nfsdi_deviceid, recallp);
5230 NFSCL_DEBUG(4, "nfscl_dserr recall iomode=%d\n", iomode);
5233 free(recallp, M_NFSLAYRECALL);
5236 /* And shut the TCP connection down. */
5237 nfscl_cancelreqs(dsp);
5241 * Cancel all RPCs for this "dsp" by closing the connection.
5242 * Also, mark the session as defunct.
5243 * If NFSCLDS_SAMECONN is set, the connection is shared with other DSs and
5244 * cannot be shut down.
5247 nfscl_cancelreqs(struct nfsclds *dsp)
5249 struct __rpc_client *cl;
5250 static int non_event;
5253 if ((dsp->nfsclds_flags & (NFSCLDS_CLOSED | NFSCLDS_SAMECONN)) == 0 &&
5254 dsp->nfsclds_sockp != NULL &&
5255 dsp->nfsclds_sockp->nr_client != NULL) {
5256 dsp->nfsclds_flags |= NFSCLDS_CLOSED;
5257 cl = dsp->nfsclds_sockp->nr_client;
5258 dsp->nfsclds_sess.nfsess_defunct = 1;
5262 * This 1sec sleep is done to reduce the number of reconnect
5263 * attempts made on the DS while it has failed.
5265 tsleep(&non_event, PVFS, "ndscls", hz);
5272 * Dereference a layout.
5275 nfscl_rellayout(struct nfscllayout *lyp, int exclocked)
5280 nfsv4_unlock(&lyp->nfsly_lock, 0);
5282 nfsv4_relref(&lyp->nfsly_lock);
5287 * Search for a devinfo by deviceid. If one is found, return it after
5288 * acquiring a reference count on it.
5290 struct nfscldevinfo *
5291 nfscl_getdevinfo(struct nfsclclient *clp, uint8_t *deviceid,
5292 struct nfscldevinfo *dip)
5297 dip = nfscl_finddevinfo(clp, deviceid);
5299 dip->nfsdi_refcnt++;
5305 * Dereference a devinfo structure.
5308 nfscl_reldevinfo_locked(struct nfscldevinfo *dip)
5311 dip->nfsdi_refcnt--;
5312 if (dip->nfsdi_refcnt == 0)
5313 wakeup(&dip->nfsdi_refcnt);
5317 * Dereference a devinfo structure.
5320 nfscl_reldevinfo(struct nfscldevinfo *dip)
5324 nfscl_reldevinfo_locked(dip);
5329 * Find a layout for this file handle. Return NULL upon failure.
5331 static struct nfscllayout *
5332 nfscl_findlayout(struct nfsclclient *clp, u_int8_t *fhp, int fhlen)
5334 struct nfscllayout *lyp;
5336 LIST_FOREACH(lyp, NFSCLLAYOUTHASH(clp, fhp, fhlen), nfsly_hash)
5337 if (lyp->nfsly_fhlen == fhlen &&
5338 !NFSBCMP(lyp->nfsly_fh, fhp, fhlen))
5344 * Find a devinfo for this deviceid. Return NULL upon failure.
5346 static struct nfscldevinfo *
5347 nfscl_finddevinfo(struct nfsclclient *clp, uint8_t *deviceid)
5349 struct nfscldevinfo *dip;
5351 LIST_FOREACH(dip, &clp->nfsc_devinfo, nfsdi_list)
5352 if (NFSBCMP(dip->nfsdi_deviceid, deviceid, NFSX_V4DEVICEID)
5359 * Merge the new file layout list into the main one, maintaining it in
5360 * increasing offset order.
5363 nfscl_mergeflayouts(struct nfsclflayouthead *fhlp,
5364 struct nfsclflayouthead *newfhlp)
5366 struct nfsclflayout *flp, *nflp, *prevflp, *tflp;
5368 flp = LIST_FIRST(fhlp);
5370 LIST_FOREACH_SAFE(nflp, newfhlp, nfsfl_list, tflp) {
5371 while (flp != NULL && flp->nfsfl_off < nflp->nfsfl_off) {
5373 flp = LIST_NEXT(flp, nfsfl_list);
5375 if (prevflp == NULL)
5376 LIST_INSERT_HEAD(fhlp, nflp, nfsfl_list);
5378 LIST_INSERT_AFTER(prevflp, nflp, nfsfl_list);
5384 * Add this nfscldevinfo to the client, if it doesn't already exist.
5385 * This function consumes the structure pointed at by dip, if not NULL.
5388 nfscl_adddevinfo(struct nfsmount *nmp, struct nfscldevinfo *dip, int ind,
5389 struct nfsclflayout *flp)
5391 struct nfsclclient *clp;
5392 struct nfscldevinfo *tdip;
5400 free(dip, M_NFSDEVINFO);
5403 if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5404 dev = flp->nfsfl_dev;
5406 dev = flp->nfsfl_ffm[ind].dev;
5407 tdip = nfscl_finddevinfo(clp, dev);
5409 tdip->nfsdi_layoutrefs++;
5410 if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5411 flp->nfsfl_devp = tdip;
5413 flp->nfsfl_ffm[ind].devp = tdip;
5414 nfscl_reldevinfo_locked(tdip);
5417 free(dip, M_NFSDEVINFO);
5421 LIST_INSERT_HEAD(&clp->nfsc_devinfo, dip, nfsdi_list);
5422 dip->nfsdi_layoutrefs = 1;
5423 if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5424 flp->nfsfl_devp = dip;
5426 flp->nfsfl_ffm[ind].devp = dip;
5435 * Free up a layout structure and associated file layout structure(s).
5438 nfscl_freelayout(struct nfscllayout *layp)
5440 struct nfsclflayout *flp, *nflp;
5441 struct nfsclrecalllayout *rp, *nrp;
5443 LIST_FOREACH_SAFE(flp, &layp->nfsly_flayread, nfsfl_list, nflp) {
5444 LIST_REMOVE(flp, nfsfl_list);
5445 nfscl_freeflayout(flp);
5447 LIST_FOREACH_SAFE(flp, &layp->nfsly_flayrw, nfsfl_list, nflp) {
5448 LIST_REMOVE(flp, nfsfl_list);
5449 nfscl_freeflayout(flp);
5451 LIST_FOREACH_SAFE(rp, &layp->nfsly_recall, nfsrecly_list, nrp) {
5452 LIST_REMOVE(rp, nfsrecly_list);
5453 free(rp, M_NFSLAYRECALL);
5456 free(layp, M_NFSLAYOUT);
5460 * Free up a file layout structure.
5463 nfscl_freeflayout(struct nfsclflayout *flp)
5467 if ((flp->nfsfl_flags & NFSFL_FILE) != 0) {
5468 for (i = 0; i < flp->nfsfl_fhcnt; i++)
5469 free(flp->nfsfl_fh[i], M_NFSFH);
5470 if (flp->nfsfl_devp != NULL)
5471 flp->nfsfl_devp->nfsdi_layoutrefs--;
5473 if ((flp->nfsfl_flags & NFSFL_FLEXFILE) != 0)
5474 for (i = 0; i < flp->nfsfl_mirrorcnt; i++) {
5475 for (j = 0; j < flp->nfsfl_ffm[i].fhcnt; j++)
5476 free(flp->nfsfl_ffm[i].fh[j], M_NFSFH);
5477 if (flp->nfsfl_ffm[i].devp != NULL)
5478 flp->nfsfl_ffm[i].devp->nfsdi_layoutrefs--;
5480 free(flp, M_NFSFLAYOUT);
5484 * Free up a file layout devinfo structure.
5487 nfscl_freedevinfo(struct nfscldevinfo *dip)
5490 free(dip, M_NFSDEVINFO);
5494 * Mark any layouts that match as recalled.
5497 nfscl_layoutrecall(int recalltype, struct nfscllayout *lyp, uint32_t iomode,
5498 uint64_t off, uint64_t len, uint32_t stateseqid, uint32_t stat, uint32_t op,
5499 char *devid, struct nfsclrecalllayout *recallp)
5501 struct nfsclrecalllayout *rp, *orp;
5503 recallp->nfsrecly_recalltype = recalltype;
5504 recallp->nfsrecly_iomode = iomode;
5505 recallp->nfsrecly_stateseqid = stateseqid;
5506 recallp->nfsrecly_off = off;
5507 recallp->nfsrecly_len = len;
5508 recallp->nfsrecly_stat = stat;
5509 recallp->nfsrecly_op = op;
5511 NFSBCOPY(devid, recallp->nfsrecly_devid, NFSX_V4DEVICEID);
5513 * Order the list as file returns first, followed by fsid and any
5514 * returns, both in increasing stateseqid order.
5515 * Note that the seqids wrap around, so 1 is after 0xffffffff.
5516 * (I'm not sure this is correct because I find RFC5661 confusing
5517 * on this, but hopefully it will work ok.)
5520 LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) {
5522 if ((recalltype == NFSLAYOUTRETURN_FILE &&
5523 (rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE ||
5524 nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) ||
5525 (recalltype != NFSLAYOUTRETURN_FILE &&
5526 rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE &&
5527 nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) {
5528 LIST_INSERT_BEFORE(rp, recallp, nfsrecly_list);
5533 * Put any error return on all the file returns that will
5536 if (rp->nfsrecly_recalltype == NFSLAYOUTRETURN_FILE &&
5537 stat != 0 && rp->nfsrecly_stat == 0) {
5538 rp->nfsrecly_stat = stat;
5539 rp->nfsrecly_op = op;
5541 NFSBCOPY(devid, rp->nfsrecly_devid,
5547 LIST_INSERT_HEAD(&lyp->nfsly_recall, recallp,
5550 LIST_INSERT_AFTER(orp, recallp, nfsrecly_list);
5552 lyp->nfsly_flags |= NFSLY_RECALL;
5553 wakeup(lyp->nfsly_clp);
5558 * Compare the two seqids for ordering. The trick is that the seqids can
5559 * wrap around from 0xffffffff->0, so check for the cases where one
5560 * has wrapped around.
5561 * Return 1 if seqid1 comes before seqid2, 0 otherwise.
5564 nfscl_seq(uint32_t seqid1, uint32_t seqid2)
5567 if (seqid2 > seqid1 && (seqid2 - seqid1) >= 0x7fffffff)
5568 /* seqid2 has wrapped around. */
5570 if (seqid1 > seqid2 && (seqid1 - seqid2) >= 0x7fffffff)
5571 /* seqid1 has wrapped around. */
5573 if (seqid1 <= seqid2)
5579 * Do a layout return for each of the recalls.
5582 nfscl_layoutreturn(struct nfsmount *nmp, struct nfscllayout *lyp,
5583 struct ucred *cred, NFSPROC_T *p)
5585 struct nfsclrecalllayout *rp;
5586 nfsv4stateid_t stateid;
5589 NFSBCOPY(lyp->nfsly_stateid.other, stateid.other, NFSX_STATEIDOTHER);
5590 stateid.seqid = lyp->nfsly_stateid.seqid;
5591 if ((lyp->nfsly_flags & NFSLY_FILES) != 0)
5592 layouttype = NFSLAYOUT_NFSV4_1_FILES;
5594 layouttype = NFSLAYOUT_FLEXFILE;
5595 LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) {
5596 (void)nfsrpc_layoutreturn(nmp, lyp->nfsly_fh,
5597 lyp->nfsly_fhlen, 0, layouttype,
5598 rp->nfsrecly_iomode, rp->nfsrecly_recalltype,
5599 rp->nfsrecly_off, rp->nfsrecly_len,
5600 &stateid, cred, p, rp->nfsrecly_stat, rp->nfsrecly_op,
5601 rp->nfsrecly_devid);
5606 * Do the layout commit for a file layout.
5609 nfscl_dolayoutcommit(struct nfsmount *nmp, struct nfscllayout *lyp,
5610 struct ucred *cred, NFSPROC_T *p)
5612 struct nfsclflayout *flp;
5614 int error, layouttype;
5616 if ((lyp->nfsly_flags & NFSLY_FILES) != 0)
5617 layouttype = NFSLAYOUT_NFSV4_1_FILES;
5619 layouttype = NFSLAYOUT_FLEXFILE;
5620 LIST_FOREACH(flp, &lyp->nfsly_flayrw, nfsfl_list) {
5621 if (layouttype == NFSLAYOUT_FLEXFILE &&
5622 (flp->nfsfl_fflags & NFSFLEXFLAG_NO_LAYOUTCOMMIT) != 0) {
5623 NFSCL_DEBUG(4, "Flex file: no layoutcommit\n");
5624 /* If not supported, don't bother doing it. */
5626 nmp->nm_state |= NFSSTA_NOLAYOUTCOMMIT;
5629 } else if (flp->nfsfl_off <= lyp->nfsly_lastbyte) {
5630 len = flp->nfsfl_end - flp->nfsfl_off;
5631 error = nfsrpc_layoutcommit(nmp, lyp->nfsly_fh,
5632 lyp->nfsly_fhlen, 0, flp->nfsfl_off, len,
5633 lyp->nfsly_lastbyte, &lyp->nfsly_stateid,
5634 layouttype, cred, p, NULL);
5635 NFSCL_DEBUG(4, "layoutcommit err=%d\n", error);
5636 if (error == NFSERR_NOTSUPP) {
5637 /* If not supported, don't bother doing it. */
5639 nmp->nm_state |= NFSSTA_NOLAYOUTCOMMIT;
5648 * Commit all layouts for a file (vnode).
5651 nfscl_layoutcommit(vnode_t vp, NFSPROC_T *p)
5653 struct nfsclclient *clp;
5654 struct nfscllayout *lyp;
5655 struct nfsnode *np = VTONFS(vp);
5657 struct nfsmount *nmp;
5661 if (NFSHASNOLAYOUTCOMMIT(nmp))
5669 lyp = nfscl_findlayout(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
5674 nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
5675 if (NFSCL_FORCEDISM(mp)) {
5680 if ((lyp->nfsly_flags & NFSLY_WRITTEN) != 0) {
5681 lyp->nfsly_flags &= ~NFSLY_WRITTEN;
5683 NFSCL_DEBUG(4, "do layoutcommit2\n");
5684 nfscl_dolayoutcommit(clp->nfsc_nmp, lyp, NFSPROCCRED(p), p);
5688 nfsv4_relref(&lyp->nfsly_lock);