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];
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);
551 clp = nfscl_findcl(nmp);
558 * Wait for recovery to complete.
560 while ((clp->nfsc_flags & NFSCLFLAGS_RECVRINPROG))
561 (void) nfsmsleep(&clp->nfsc_flags, NFSCLSTATEMUTEXPTR,
562 PZERO, "nfsrecvr", NULL);
565 * First, look for a delegation.
567 LIST_FOREACH(dp, NFSCLDELEGHASH(clp, nfhp, fhlen), nfsdl_hash) {
568 if (dp->nfsdl_fhlen == fhlen &&
569 !NFSBCMP(nfhp, dp->nfsdl_fh, fhlen)) {
570 if (!(mode & NFSV4OPEN_ACCESSWRITE) ||
571 (dp->nfsdl_flags & NFSCLDL_WRITE)) {
572 stateidp->seqid = dp->nfsdl_stateid.seqid;
573 stateidp->other[0] = dp->nfsdl_stateid.other[0];
574 stateidp->other[1] = dp->nfsdl_stateid.other[1];
575 stateidp->other[2] = dp->nfsdl_stateid.other[2];
576 if (!(np->n_flag & NDELEGRECALL)) {
577 TAILQ_REMOVE(&clp->nfsc_deleg, dp,
579 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp,
581 dp->nfsdl_timestamp = NFSD_MONOSEC +
583 dp->nfsdl_rwlock.nfslock_usecnt++;
584 *lckpp = (void *)&dp->nfsdl_rwlock;
595 * If p != NULL, we want to search the parentage tree
596 * for a matching OpenOwner and use that.
598 if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount)))
599 nfscl_filllockowner(NULL, own, F_POSIX);
601 nfscl_filllockowner(p->td_proc, own, F_POSIX);
603 error = nfscl_getopen(NULL, clp->nfsc_openhash, nfhp, fhlen,
604 own, own, mode, &lp, &op);
605 if (error == 0 && lp != NULL && fords == 0) {
606 /* Don't return a lock stateid for a DS. */
608 lp->nfsl_stateid.seqid;
610 lp->nfsl_stateid.other[0];
612 lp->nfsl_stateid.other[1];
614 lp->nfsl_stateid.other[2];
620 /* If not found, just look for any OpenOwner that will work. */
623 oph = NFSCLOPENHASH(clp, nfhp, fhlen);
624 LIST_FOREACH(op, oph, nfso_hash) {
625 if (op->nfso_fhlen == fhlen &&
626 !NFSBCMP(op->nfso_fh, nfhp, fhlen)) {
627 if (top == NULL && (op->nfso_mode &
628 NFSV4OPEN_ACCESSWRITE) != 0 &&
629 (mode & NFSV4OPEN_ACCESSREAD) != 0)
631 if ((mode & op->nfso_mode) == mode) {
632 /* LRU order the hash list. */
633 LIST_REMOVE(op, nfso_hash);
634 LIST_INSERT_HEAD(oph, op, nfso_hash);
641 NFSCL_DEBUG(2, "openmode top=%p\n", top);
642 if (top == NULL || NFSHASOPENMODE(nmp)) {
649 * For read aheads or write behinds, use the open cred.
650 * A read ahead or write behind is indicated by p == NULL.
653 newnfs_copycred(&op->nfso_cred, cred);
657 * No lock stateid, so return the open stateid.
659 stateidp->seqid = op->nfso_stateid.seqid;
660 stateidp->other[0] = op->nfso_stateid.other[0];
661 stateidp->other[1] = op->nfso_stateid.other[1];
662 stateidp->other[2] = op->nfso_stateid.other[2];
668 * Search for a matching file, mode and, optionally, lockowner.
671 nfscl_getopen(struct nfsclownerhead *ohp, struct nfsclopenhash *ohashp,
672 u_int8_t *nfhp, int fhlen, u_int8_t *openown, u_int8_t *lockown,
673 u_int32_t mode, struct nfscllockowner **lpp, struct nfsclopen **opp)
675 struct nfsclowner *owp;
676 struct nfsclopen *op, *rop, *rop2;
677 struct nfsclopenhash *oph;
680 KASSERT(ohp == NULL || ohashp == NULL, ("nfscl_getopen: "
681 "only one of ohp and ohashp can be set"));
685 * rop will be set to the open to be returned. There are three
686 * variants of this, all for an open of the correct file:
687 * 1 - A match of lockown.
688 * 2 - A match of the openown, when no lockown match exists.
689 * 3 - A match for any open, if no openown or lockown match exists.
690 * Looking for #2 over #3 probably isn't necessary, but since
691 * RFC3530 is vague w.r.t. the relationship between openowners and
692 * lockowners, I think this is the safer way to go.
697 /* Search the client list */
698 if (ohashp == NULL) {
699 /* Search the local opens on the delegation. */
700 LIST_FOREACH(owp, ohp, nfsow_list) {
701 /* and look for the correct open */
702 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
703 if (op->nfso_fhlen == fhlen &&
704 !NFSBCMP(op->nfso_fh, nfhp, fhlen)
705 && (op->nfso_mode & mode) == mode)
706 keep_looping = nfscl_checkown(owp, op, openown,
707 lockown, lpp, &rop, &rop2);
715 /* Search for matching opens on the hash list. */
716 oph = &ohashp[NFSCLOPENHASHFUNC(nfhp, fhlen)];
717 LIST_FOREACH(op, oph, nfso_hash) {
718 if (op->nfso_fhlen == fhlen &&
719 !NFSBCMP(op->nfso_fh, nfhp, fhlen)
720 && (op->nfso_mode & mode) == mode)
721 keep_looping = nfscl_checkown(op->nfso_own, op,
722 openown, lockown, lpp, &rop, &rop2);
724 /* LRU order the hash list. */
725 LIST_REMOVE(op, nfso_hash);
726 LIST_INSERT_HEAD(oph, op, nfso_hash);
739 /* Check for an owner match. */
741 nfscl_checkown(struct nfsclowner *owp, struct nfsclopen *op, uint8_t *openown,
742 uint8_t *lockown, struct nfscllockowner **lpp, struct nfsclopen **ropp,
743 struct nfsclopen **ropp2)
745 struct nfscllockowner *lp;
750 /* Now look for a matching lockowner. */
751 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
752 if (!NFSBCMP(lp->nfsl_owner, lockown,
753 NFSV4CL_LOCKNAMELEN)) {
760 if (*ropp == NULL && !NFSBCMP(owp->nfsow_owner, openown,
761 NFSV4CL_LOCKNAMELEN)) {
764 keep_looping = false;
768 return (keep_looping);
772 * Release use of an open owner. Called when open operations are done
773 * with the open owner.
776 nfscl_ownerrelease(struct nfsmount *nmp, struct nfsclowner *owp,
777 __unused int error, __unused int candelete, int unlocked)
784 if (NFSHASONEOPENOWN(nmp))
785 nfsv4_relref(&owp->nfsow_rwlock);
787 nfscl_lockunlock(&owp->nfsow_rwlock);
789 nfscl_clrelease(owp->nfsow_clp);
794 * Release use of an open structure under an open owner.
797 nfscl_openrelease(struct nfsmount *nmp, struct nfsclopen *op, int error,
800 struct nfsclclient *clp;
801 struct nfsclowner *owp;
807 if (NFSHASONEOPENOWN(nmp))
808 nfsv4_relref(&owp->nfsow_rwlock);
810 nfscl_lockunlock(&owp->nfsow_rwlock);
811 clp = owp->nfsow_clp;
812 if (error && candelete && op->nfso_opencnt == 0)
813 nfscl_freeopen(op, 0);
814 nfscl_clrelease(clp);
819 * Called to get a clientid structure. It will optionally lock the
820 * client data structures to do the SetClientId/SetClientId_confirm,
821 * but will release that lock and return the clientid with a reference
823 * If the "cred" argument is NULL, a new clientid should not be created.
824 * If the "p" argument is NULL, a SetClientID/SetClientIDConfirm cannot
826 * It always clpp with a reference count on it, unless returning an error.
829 nfscl_getcl(struct mount *mp, struct ucred *cred, NFSPROC_T *p,
830 bool tryminvers, struct nfsclclient **clpp)
832 struct nfsclclient *clp;
833 struct nfsclclient *newclp = NULL;
834 struct nfsmount *nmp;
835 char uuid[HOSTUUIDLEN];
836 int igotlock = 0, error, trystalecnt, clidinusedelay, i;
841 getcredhostuuid(cred, uuid, sizeof uuid);
842 idlen = strlen(uuid);
844 idlen += sizeof (u_int64_t);
846 idlen += sizeof (u_int64_t) + 16; /* 16 random bytes */
848 sizeof (struct nfsclclient) + idlen - 1, M_NFSCLCLIENT,
853 * If a forced dismount is already in progress, don't
854 * allocate a new clientid and get out now. For the case where
855 * clp != NULL, this is a harmless optimization.
857 if (NFSCL_FORCEDISM(mp)) {
860 free(newclp, M_NFSCLCLIENT);
865 if (newclp == NULL) {
870 clp->nfsc_idlen = idlen;
871 LIST_INIT(&clp->nfsc_owner);
872 TAILQ_INIT(&clp->nfsc_deleg);
873 TAILQ_INIT(&clp->nfsc_layout);
874 LIST_INIT(&clp->nfsc_devinfo);
875 for (i = 0; i < NFSCLDELEGHASHSIZE; i++)
876 LIST_INIT(&clp->nfsc_deleghash[i]);
877 for (i = 0; i < NFSCLOPENHASHSIZE; i++)
878 LIST_INIT(&clp->nfsc_openhash[i]);
879 for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++)
880 LIST_INIT(&clp->nfsc_layouthash[i]);
881 clp->nfsc_flags = NFSCLFLAGS_INITED;
882 clp->nfsc_clientidrev = 1;
883 clp->nfsc_cbident = nfscl_nextcbident();
884 nfscl_fillclid(nmp->nm_clval, uuid, clp->nfsc_id,
886 LIST_INSERT_HEAD(&nfsclhead, clp, nfsc_list);
891 free(newclp, M_NFSCLCLIENT);
893 while ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0 && !igotlock &&
894 !NFSCL_FORCEDISM(mp))
895 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
896 NFSCLSTATEMUTEXPTR, mp);
899 * Call nfsv4_lock() with "iwantlock == 0" so that it will
900 * wait for a pending exclusive lock request. This gives the
901 * exclusive lock request priority over this shared lock
903 * An exclusive lock on nfsc_lock is used mainly for server
906 nfsv4_lock(&clp->nfsc_lock, 0, NULL, NFSCLSTATEMUTEXPTR, mp);
907 nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
909 if (igotlock == 0 && NFSCL_FORCEDISM(mp)) {
911 * Both nfsv4_lock() and nfsv4_getref() know to check
912 * for NFSCL_FORCEDISM() and return without sleeping to
913 * wait for the exclusive lock to be released, since it
914 * might be held by nfscl_umount() and we need to get out
915 * now for that case and not wait until nfscl_umount()
924 * If it needs a clientid, do the setclientid now.
926 if ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0) {
928 panic("nfscl_clget");
929 if (p == NULL || cred == NULL) {
931 nfsv4_unlock(&clp->nfsc_lock, 0);
936 * If RFC3530 Sec. 14.2.33 is taken literally,
937 * NFSERR_CLIDINUSE will be returned persistently for the
938 * case where a new mount of the same file system is using
939 * a different principal. In practice, NFSERR_CLIDINUSE is
940 * only returned when there is outstanding unexpired state
941 * on the clientid. As such, try for twice the lease
942 * interval, if we know what that is. Otherwise, make a
944 * The case of returning NFSERR_STALECLIENTID is far less
945 * likely, but might occur if there is a significant delay
946 * between doing the SetClientID and SetClientIDConfirm Ops,
947 * such that the server throws away the clientid before
948 * receiving the SetClientIDConfirm.
950 if (clp->nfsc_renew > 0)
951 clidinusedelay = NFSCL_LEASE(clp->nfsc_renew) * 2;
953 clidinusedelay = 120;
956 error = nfsrpc_setclient(nmp, clp, 0, NULL, cred, p);
957 if (error == NFSERR_STALECLIENTID ||
958 error == NFSERR_STALEDONTRECOVER ||
959 error == NFSERR_BADSESSION ||
960 error == NFSERR_CLIDINUSE) {
961 (void) nfs_catnap(PZERO, error, "nfs_setcl");
963 } while (((error == NFSERR_STALECLIENTID ||
964 error == NFSERR_BADSESSION ||
965 error == NFSERR_STALEDONTRECOVER) && --trystalecnt > 0) ||
966 (error == NFSERR_CLIDINUSE && --clidinusedelay > 0));
969 nfsv4_unlock(&clp->nfsc_lock, 0);
973 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
977 nfsv4_unlock(&clp->nfsc_lock, 1);
986 * Get a reference to a clientid and return it, if valid.
989 nfscl_findcl(struct nfsmount *nmp)
991 struct nfsclclient *clp;
994 if (clp == NULL || !(clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID))
1000 * Release the clientid structure. It may be locked or reference counted.
1003 nfscl_clrelease(struct nfsclclient *clp)
1006 if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK)
1007 nfsv4_unlock(&clp->nfsc_lock, 0);
1009 nfsv4_relref(&clp->nfsc_lock);
1013 * External call for nfscl_clrelease.
1016 nfscl_clientrelease(struct nfsclclient *clp)
1020 if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK)
1021 nfsv4_unlock(&clp->nfsc_lock, 0);
1023 nfsv4_relref(&clp->nfsc_lock);
1028 * Called when wanting to lock a byte region.
1031 nfscl_getbytelock(vnode_t vp, u_int64_t off, u_int64_t len,
1032 short type, struct ucred *cred, NFSPROC_T *p, struct nfsclclient *rclp,
1033 int recovery, void *id, int flags, u_int8_t *rownp, u_int8_t *ropenownp,
1034 struct nfscllockowner **lpp, int *newonep, int *donelocallyp)
1036 struct nfscllockowner *lp;
1037 struct nfsclopen *op;
1038 struct nfsclclient *clp;
1039 struct nfscllockowner *nlp;
1040 struct nfscllock *nlop, *otherlop;
1041 struct nfscldeleg *dp = NULL, *ldp = NULL;
1042 struct nfscllockownerhead *lhp = NULL;
1044 u_int8_t own[NFSV4CL_LOCKNAMELEN], *ownp, openown[NFSV4CL_LOCKNAMELEN];
1046 int error = 0, ret, donelocally = 0;
1049 /* For Lock Ops, the open mode doesn't matter, so use 0 to match any. */
1058 * Might need these, so MALLOC them now, to
1059 * avoid a tsleep() in MALLOC later.
1062 sizeof (struct nfscllockowner), M_NFSCLLOCKOWNER, M_WAITOK);
1064 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1066 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1067 nlop->nfslo_type = type;
1068 nlop->nfslo_first = off;
1069 if (len == NFS64BITSSET) {
1070 nlop->nfslo_end = NFS64BITSSET;
1072 nlop->nfslo_end = off + len;
1073 if (nlop->nfslo_end <= nlop->nfslo_first)
1074 error = NFSERR_INVAL;
1081 error = nfscl_getcl(vp->v_mount, cred, p, false, &clp);
1084 free(nlp, M_NFSCLLOCKOWNER);
1085 free(otherlop, M_NFSCLLOCK);
1086 free(nlop, M_NFSCLLOCK);
1093 openownp = ropenownp;
1095 nfscl_filllockowner(id, own, flags);
1097 if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount)))
1098 nfscl_filllockowner(NULL, openown, F_POSIX);
1100 nfscl_filllockowner(p->td_proc, openown, F_POSIX);
1106 * First, search for a delegation. If one exists for this file,
1107 * the lock can be done locally against it, so long as there
1108 * isn't a local lock conflict.
1110 ldp = dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
1111 np->n_fhp->nfh_len);
1112 /* Just sanity check for correct type of delegation */
1113 if (dp != NULL && ((dp->nfsdl_flags &
1114 (NFSCLDL_RECALL | NFSCLDL_DELEGRET)) != 0 ||
1116 (dp->nfsdl_flags & NFSCLDL_WRITE) == 0)))
1120 /* Now, find an open and maybe a lockowner. */
1121 ret = nfscl_getopen(&dp->nfsdl_owner, NULL, np->n_fhp->nfh_fh,
1122 np->n_fhp->nfh_len, openownp, ownp, mode, NULL, &op);
1124 ret = nfscl_getopen(NULL, clp->nfsc_openhash,
1125 np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp,
1126 ownp, mode, NULL, &op);
1128 lhp = &dp->nfsdl_lock;
1129 TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
1130 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list);
1131 dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
1139 * Get the related Open and maybe lockowner.
1141 error = nfscl_getopen(NULL, clp->nfsc_openhash,
1142 np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp,
1143 ownp, mode, &lp, &op);
1145 lhp = &op->nfso_lock;
1147 if (!error && !recovery)
1148 error = nfscl_localconflict(clp, np->n_fhp->nfh_fh,
1149 np->n_fhp->nfh_len, nlop, ownp, ldp, NULL);
1152 nfscl_clrelease(clp);
1155 free(nlp, M_NFSCLLOCKOWNER);
1156 free(otherlop, M_NFSCLLOCK);
1157 free(nlop, M_NFSCLLOCK);
1162 * Ok, see if a lockowner exists and create one, as required.
1165 LIST_FOREACH(lp, lhp, nfsl_list) {
1166 if (!NFSBCMP(lp->nfsl_owner, ownp, NFSV4CL_LOCKNAMELEN))
1170 NFSBCOPY(ownp, nlp->nfsl_owner, NFSV4CL_LOCKNAMELEN);
1172 NFSBCOPY(ropenownp, nlp->nfsl_openowner,
1173 NFSV4CL_LOCKNAMELEN);
1175 NFSBCOPY(op->nfso_own->nfsow_owner, nlp->nfsl_openowner,
1176 NFSV4CL_LOCKNAMELEN);
1177 nlp->nfsl_seqid = 0;
1178 nlp->nfsl_lockflags = flags;
1179 nlp->nfsl_inprog = NULL;
1180 nfscl_lockinit(&nlp->nfsl_rwlock);
1181 LIST_INIT(&nlp->nfsl_lock);
1183 nlp->nfsl_open = NULL;
1184 nfsstatsv1.cllocallockowners++;
1186 nlp->nfsl_open = op;
1187 nfsstatsv1.cllockowners++;
1189 LIST_INSERT_HEAD(lhp, nlp, nfsl_list);
1196 * Now, update the byte ranges for locks.
1198 ret = nfscl_updatelock(lp, &nlop, &otherlop, donelocally);
1204 nfscl_clrelease(clp);
1207 * Serial modifications on the lock owner for multiple threads
1208 * for the same process using a read/write lock.
1211 nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR);
1217 free(nlp, M_NFSCLLOCKOWNER);
1219 free(nlop, M_NFSCLLOCK);
1221 free(otherlop, M_NFSCLLOCK);
1228 * Called to unlock a byte range, for LockU.
1231 nfscl_relbytelock(vnode_t vp, u_int64_t off, u_int64_t len,
1232 __unused struct ucred *cred, NFSPROC_T *p, int callcnt,
1233 struct nfsclclient *clp, void *id, int flags,
1234 struct nfscllockowner **lpp, int *dorpcp)
1236 struct nfscllockowner *lp;
1237 struct nfsclopen *op;
1238 struct nfscllock *nlop, *other_lop = NULL;
1239 struct nfscldeleg *dp;
1241 u_int8_t own[NFSV4CL_LOCKNAMELEN];
1249 * Might need these, so MALLOC them now, to
1250 * avoid a tsleep() in MALLOC later.
1253 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1254 nlop->nfslo_type = F_UNLCK;
1255 nlop->nfslo_first = off;
1256 if (len == NFS64BITSSET) {
1257 nlop->nfslo_end = NFS64BITSSET;
1259 nlop->nfslo_end = off + len;
1260 if (nlop->nfslo_end <= nlop->nfslo_first) {
1261 free(nlop, M_NFSCLLOCK);
1262 return (NFSERR_INVAL);
1267 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1270 nfscl_filllockowner(id, own, flags);
1274 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
1275 np->n_fhp->nfh_len);
1278 * First, unlock any local regions on a delegation.
1281 /* Look for this lockowner. */
1282 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
1283 if (!NFSBCMP(lp->nfsl_owner, own,
1284 NFSV4CL_LOCKNAMELEN))
1288 /* Use other_lop, so nlop is still available */
1289 (void)nfscl_updatelock(lp, &other_lop, NULL, 1);
1293 * Now, find a matching open/lockowner that hasn't already been done,
1294 * as marked by nfsl_inprog.
1298 LIST_FOREACH(op, NFSCLOPENHASH(clp, np->n_fhp->nfh_fh,
1299 np->n_fhp->nfh_len), nfso_hash) {
1300 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1301 !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1302 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1303 if (lp->nfsl_inprog == NULL &&
1304 !NFSBCMP(lp->nfsl_owner, own,
1305 NFSV4CL_LOCKNAMELEN)) {
1316 ret = nfscl_updatelock(lp, &nlop, NULL, 0);
1320 * Serial modifications on the lock owner for multiple
1321 * threads for the same process using a read/write lock.
1323 lp->nfsl_inprog = p;
1324 nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR);
1329 free(nlop, M_NFSCLLOCK);
1331 free(other_lop, M_NFSCLLOCK);
1336 * Release all lockowners marked in progess for this process and file.
1339 nfscl_releasealllocks(struct nfsclclient *clp, vnode_t vp, NFSPROC_T *p,
1340 void *id, int flags)
1342 struct nfsclopen *op;
1343 struct nfscllockowner *lp;
1345 u_int8_t own[NFSV4CL_LOCKNAMELEN];
1348 nfscl_filllockowner(id, own, flags);
1350 LIST_FOREACH(op, NFSCLOPENHASH(clp, np->n_fhp->nfh_fh,
1351 np->n_fhp->nfh_len), nfso_hash) {
1352 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1353 !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1354 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1355 if (lp->nfsl_inprog == p &&
1356 !NFSBCMP(lp->nfsl_owner, own,
1357 NFSV4CL_LOCKNAMELEN)) {
1358 lp->nfsl_inprog = NULL;
1359 nfscl_lockunlock(&lp->nfsl_rwlock);
1364 nfscl_clrelease(clp);
1369 * Called to find out if any bytes within the byte range specified are
1370 * write locked by the calling process. Used to determine if flushing
1371 * is required before a LockU.
1372 * If in doubt, return 1, so the flush will occur.
1375 nfscl_checkwritelocked(vnode_t vp, struct flock *fl,
1376 struct ucred *cred, NFSPROC_T *p, void *id, int flags)
1378 struct nfscllockowner *lp;
1379 struct nfsclopen *op;
1380 struct nfsclclient *clp;
1381 struct nfscllock *lop;
1382 struct nfscldeleg *dp;
1385 u_int8_t own[NFSV4CL_LOCKNAMELEN];
1389 switch (fl->l_whence) {
1393 * Caller is responsible for adding any necessary offset
1394 * when SEEK_CUR is used.
1399 off = np->n_size + fl->l_start;
1404 if (fl->l_len != 0) {
1405 end = off + fl->l_len;
1412 error = nfscl_getcl(vp->v_mount, cred, p, false, &clp);
1415 nfscl_filllockowner(id, own, flags);
1419 * First check the delegation locks.
1421 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
1423 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
1424 if (!NFSBCMP(lp->nfsl_owner, own,
1425 NFSV4CL_LOCKNAMELEN))
1429 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
1430 if (lop->nfslo_first >= end)
1432 if (lop->nfslo_end <= off)
1434 if (lop->nfslo_type == F_WRLCK) {
1435 nfscl_clrelease(clp);
1444 * Now, check state against the server.
1446 LIST_FOREACH(op, NFSCLOPENHASH(clp, np->n_fhp->nfh_fh,
1447 np->n_fhp->nfh_len), nfso_hash) {
1448 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1449 !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1450 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1451 if (!NFSBCMP(lp->nfsl_owner, own,
1452 NFSV4CL_LOCKNAMELEN))
1456 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
1457 if (lop->nfslo_first >= end)
1459 if (lop->nfslo_end <= off)
1461 if (lop->nfslo_type == F_WRLCK) {
1462 nfscl_clrelease(clp);
1470 nfscl_clrelease(clp);
1476 * Release a byte range lock owner structure.
1479 nfscl_lockrelease(struct nfscllockowner *lp, int error, int candelete)
1481 struct nfsclclient *clp;
1486 clp = lp->nfsl_open->nfso_own->nfsow_clp;
1487 if (error != 0 && candelete &&
1488 (lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED) == 0)
1489 nfscl_freelockowner(lp, 0);
1491 nfscl_lockunlock(&lp->nfsl_rwlock);
1492 nfscl_clrelease(clp);
1497 * Free up an open structure and any associated byte range lock structures.
1500 nfscl_freeopen(struct nfsclopen *op, int local)
1503 LIST_REMOVE(op, nfso_list);
1504 if (op->nfso_hash.le_prev != NULL)
1505 LIST_REMOVE(op, nfso_hash);
1506 nfscl_freealllocks(&op->nfso_lock, local);
1507 free(op, M_NFSCLOPEN);
1509 nfsstatsv1.cllocalopens--;
1511 nfsstatsv1.clopens--;
1515 * Free up all lock owners and associated locks.
1518 nfscl_freealllocks(struct nfscllockownerhead *lhp, int local)
1520 struct nfscllockowner *lp, *nlp;
1522 LIST_FOREACH_SAFE(lp, lhp, nfsl_list, nlp) {
1523 if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED))
1525 nfscl_freelockowner(lp, local);
1530 * Called for an Open when NFSERR_EXPIRED is received from the server.
1531 * If there are no byte range locks nor a Share Deny lost, try to do a
1532 * fresh Open. Otherwise, free the open.
1535 nfscl_expireopen(struct nfsclclient *clp, struct nfsclopen *op,
1536 struct nfsmount *nmp, struct ucred *cred, NFSPROC_T *p)
1538 struct nfscllockowner *lp;
1539 struct nfscldeleg *dp;
1540 int mustdelete = 0, error;
1543 * Look for any byte range lock(s).
1545 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1546 if (!LIST_EMPTY(&lp->nfsl_lock)) {
1553 * If no byte range lock(s) nor a Share deny, try to re-open.
1555 if (!mustdelete && (op->nfso_mode & NFSLCK_DENYBITS) == 0) {
1556 newnfs_copycred(&op->nfso_cred, cred);
1558 error = nfsrpc_reopen(nmp, op->nfso_fh,
1559 op->nfso_fhlen, op->nfso_mode, op, &dp, cred, p);
1563 free(dp, M_NFSCLDELEG);
1568 nfscl_deleg(nmp->nm_mountp, clp, op->nfso_fh,
1569 op->nfso_fhlen, cred, p, &dp);
1573 * If a byte range lock or Share deny or couldn't re-open, free it.
1576 nfscl_freeopen(op, 0);
1577 return (mustdelete);
1581 * Free up an open owner structure.
1584 nfscl_freeopenowner(struct nfsclowner *owp, int local)
1587 LIST_REMOVE(owp, nfsow_list);
1588 free(owp, M_NFSCLOWNER);
1590 nfsstatsv1.cllocalopenowners--;
1592 nfsstatsv1.clopenowners--;
1596 * Free up a byte range lock owner structure.
1599 nfscl_freelockowner(struct nfscllockowner *lp, int local)
1601 struct nfscllock *lop, *nlop;
1603 LIST_REMOVE(lp, nfsl_list);
1604 LIST_FOREACH_SAFE(lop, &lp->nfsl_lock, nfslo_list, nlop) {
1605 nfscl_freelock(lop, local);
1607 free(lp, M_NFSCLLOCKOWNER);
1609 nfsstatsv1.cllocallockowners--;
1611 nfsstatsv1.cllockowners--;
1615 * Free up a byte range lock structure.
1618 nfscl_freelock(struct nfscllock *lop, int local)
1621 LIST_REMOVE(lop, nfslo_list);
1622 free(lop, M_NFSCLLOCK);
1624 nfsstatsv1.cllocallocks--;
1626 nfsstatsv1.cllocks--;
1630 * Clean out the state related to a delegation.
1633 nfscl_cleandeleg(struct nfscldeleg *dp)
1635 struct nfsclowner *owp, *nowp;
1636 struct nfsclopen *op;
1638 LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) {
1639 op = LIST_FIRST(&owp->nfsow_open);
1641 if (LIST_NEXT(op, nfso_list) != NULL)
1642 panic("nfscleandel");
1643 nfscl_freeopen(op, 1);
1645 nfscl_freeopenowner(owp, 1);
1647 nfscl_freealllocks(&dp->nfsdl_lock, 1);
1651 * Free a delegation.
1654 nfscl_freedeleg(struct nfscldeleghead *hdp, struct nfscldeleg *dp, bool freeit)
1657 TAILQ_REMOVE(hdp, dp, nfsdl_list);
1658 LIST_REMOVE(dp, nfsdl_hash);
1660 free(dp, M_NFSCLDELEG);
1661 nfsstatsv1.cldelegates--;
1666 * Free up all state related to this client structure.
1669 nfscl_cleanclient(struct nfsclclient *clp)
1671 struct nfsclowner *owp, *nowp;
1672 struct nfsclopen *op, *nop;
1673 struct nfscllayout *lyp, *nlyp;
1674 struct nfscldevinfo *dip, *ndip;
1676 TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp)
1677 nfscl_freelayout(lyp);
1679 LIST_FOREACH_SAFE(dip, &clp->nfsc_devinfo, nfsdi_list, ndip)
1680 nfscl_freedevinfo(dip);
1682 /* Now, all the OpenOwners, etc. */
1683 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1684 LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
1685 nfscl_freeopen(op, 0);
1687 nfscl_freeopenowner(owp, 0);
1692 * Called when an NFSERR_EXPIRED is received from the server.
1695 nfscl_expireclient(struct nfsclclient *clp, struct nfsmount *nmp,
1696 struct ucred *cred, NFSPROC_T *p)
1698 struct nfsclowner *owp, *nowp, *towp;
1699 struct nfsclopen *op, *nop, *top;
1700 struct nfscldeleg *dp, *ndp;
1701 int ret, printed = 0;
1704 * First, merge locally issued Opens into the list for the server.
1706 dp = TAILQ_FIRST(&clp->nfsc_deleg);
1707 while (dp != NULL) {
1708 ndp = TAILQ_NEXT(dp, nfsdl_list);
1709 owp = LIST_FIRST(&dp->nfsdl_owner);
1710 while (owp != NULL) {
1711 nowp = LIST_NEXT(owp, nfsow_list);
1712 op = LIST_FIRST(&owp->nfsow_open);
1714 if (LIST_NEXT(op, nfso_list) != NULL)
1716 LIST_FOREACH(towp, &clp->nfsc_owner, nfsow_list) {
1717 if (!NFSBCMP(towp->nfsow_owner, owp->nfsow_owner,
1718 NFSV4CL_LOCKNAMELEN))
1722 /* Merge opens in */
1723 LIST_FOREACH(top, &towp->nfsow_open, nfso_list) {
1724 if (top->nfso_fhlen == op->nfso_fhlen &&
1725 !NFSBCMP(top->nfso_fh, op->nfso_fh,
1727 top->nfso_mode |= op->nfso_mode;
1728 top->nfso_opencnt += op->nfso_opencnt;
1733 /* Just add the open to the owner list */
1734 LIST_REMOVE(op, nfso_list);
1735 op->nfso_own = towp;
1736 LIST_INSERT_HEAD(&towp->nfsow_open, op, nfso_list);
1737 LIST_INSERT_HEAD(NFSCLOPENHASH(clp, op->nfso_fh,
1738 op->nfso_fhlen), op, nfso_hash);
1739 nfsstatsv1.cllocalopens--;
1740 nfsstatsv1.clopens++;
1743 /* Just add the openowner to the client list */
1744 LIST_REMOVE(owp, nfsow_list);
1745 owp->nfsow_clp = clp;
1746 LIST_INSERT_HEAD(&clp->nfsc_owner, owp, nfsow_list);
1747 LIST_INSERT_HEAD(NFSCLOPENHASH(clp, op->nfso_fh,
1748 op->nfso_fhlen), op, nfso_hash);
1749 nfsstatsv1.cllocalopenowners--;
1750 nfsstatsv1.clopenowners++;
1751 nfsstatsv1.cllocalopens--;
1752 nfsstatsv1.clopens++;
1757 if (!printed && !LIST_EMPTY(&dp->nfsdl_lock)) {
1759 printf("nfsv4 expired locks lost\n");
1761 nfscl_cleandeleg(dp);
1762 nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
1765 if (!TAILQ_EMPTY(&clp->nfsc_deleg))
1769 * Now, try and reopen against the server.
1771 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1772 owp->nfsow_seqid = 0;
1773 LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
1774 ret = nfscl_expireopen(clp, op, nmp, cred, p);
1775 if (ret && !printed) {
1777 printf("nfsv4 expired locks lost\n");
1780 if (LIST_EMPTY(&owp->nfsow_open))
1781 nfscl_freeopenowner(owp, 0);
1786 * This function must be called after the process represented by "own" has
1787 * exited. Must be called with CLSTATE lock held.
1790 nfscl_cleanup_common(struct nfsclclient *clp, u_int8_t *own)
1792 struct nfsclowner *owp, *nowp;
1793 struct nfscllockowner *lp, *nlp;
1794 struct nfscldeleg *dp;
1796 /* First, get rid of local locks on delegations. */
1797 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
1798 LIST_FOREACH_SAFE(lp, &dp->nfsdl_lock, nfsl_list, nlp) {
1799 if (!NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
1800 if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED))
1802 nfscl_freelockowner(lp, 1);
1806 owp = LIST_FIRST(&clp->nfsc_owner);
1807 while (owp != NULL) {
1808 nowp = LIST_NEXT(owp, nfsow_list);
1809 if (!NFSBCMP(owp->nfsow_owner, own,
1810 NFSV4CL_LOCKNAMELEN)) {
1812 * If there are children that haven't closed the
1813 * file descriptors yet, the opens will still be
1814 * here. For that case, let the renew thread clear
1815 * out the OpenOwner later.
1817 if (LIST_EMPTY(&owp->nfsow_open))
1818 nfscl_freeopenowner(owp, 0);
1820 owp->nfsow_defunct = 1;
1827 * Find open/lock owners for processes that have exited.
1830 nfscl_cleanupkext(struct nfsclclient *clp, struct nfscllockownerfhhead *lhp)
1832 struct nfsclowner *owp, *nowp;
1833 struct nfsclopen *op;
1834 struct nfscllockowner *lp, *nlp;
1835 struct nfscldeleg *dp;
1838 * All the pidhash locks must be acquired, since they are sx locks
1839 * and must be acquired before the mutexes. The pid(s) that will
1840 * be used aren't known yet, so all the locks need to be acquired.
1841 * Fortunately, this function is only performed once/sec.
1845 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1846 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1847 LIST_FOREACH_SAFE(lp, &op->nfso_lock, nfsl_list, nlp) {
1848 if (LIST_EMPTY(&lp->nfsl_lock))
1849 nfscl_emptylockowner(lp, lhp);
1852 if (nfscl_procdoesntexist(owp->nfsow_owner))
1853 nfscl_cleanup_common(clp, owp->nfsow_owner);
1857 * For the single open_owner case, these lock owners need to be
1858 * checked to see if they still exist separately.
1859 * This is because nfscl_procdoesntexist() never returns true for
1860 * the single open_owner so that the above doesn't ever call
1861 * nfscl_cleanup_common().
1863 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
1864 LIST_FOREACH_SAFE(lp, &dp->nfsdl_lock, nfsl_list, nlp) {
1865 if (nfscl_procdoesntexist(lp->nfsl_owner))
1866 nfscl_cleanup_common(clp, lp->nfsl_owner);
1870 pidhash_sunlockall();
1874 * Take the empty lock owner and move it to the local lhp list if the
1875 * associated process no longer exists.
1878 nfscl_emptylockowner(struct nfscllockowner *lp,
1879 struct nfscllockownerfhhead *lhp)
1881 struct nfscllockownerfh *lfhp, *mylfhp;
1882 struct nfscllockowner *nlp;
1885 /* If not a Posix lock owner, just return. */
1886 if ((lp->nfsl_lockflags & F_POSIX) == 0)
1892 * First, search to see if this lock owner is already in the list.
1893 * If it is, then the associated process no longer exists.
1895 SLIST_FOREACH(lfhp, lhp, nfslfh_list) {
1896 if (lfhp->nfslfh_len == lp->nfsl_open->nfso_fhlen &&
1897 !NFSBCMP(lfhp->nfslfh_fh, lp->nfsl_open->nfso_fh,
1900 LIST_FOREACH(nlp, &lfhp->nfslfh_lock, nfsl_list)
1901 if (!NFSBCMP(nlp->nfsl_owner, lp->nfsl_owner,
1902 NFSV4CL_LOCKNAMELEN))
1905 /* If not found, check if process still exists. */
1906 if (fnd_it == 0 && nfscl_procdoesntexist(lp->nfsl_owner) == 0)
1909 /* Move the lock owner over to the local list. */
1910 if (mylfhp == NULL) {
1911 mylfhp = malloc(sizeof(struct nfscllockownerfh), M_TEMP,
1915 mylfhp->nfslfh_len = lp->nfsl_open->nfso_fhlen;
1916 NFSBCOPY(lp->nfsl_open->nfso_fh, mylfhp->nfslfh_fh,
1917 mylfhp->nfslfh_len);
1918 LIST_INIT(&mylfhp->nfslfh_lock);
1919 SLIST_INSERT_HEAD(lhp, mylfhp, nfslfh_list);
1921 LIST_REMOVE(lp, nfsl_list);
1922 LIST_INSERT_HEAD(&mylfhp->nfslfh_lock, lp, nfsl_list);
1925 static int fake_global; /* Used to force visibility of MNTK_UNMOUNTF */
1927 * Called from nfs umount to free up the clientid.
1930 nfscl_umount(struct nfsmount *nmp, NFSPROC_T *p)
1932 struct nfsclclient *clp;
1937 * For the case that matters, this is the thread that set
1938 * MNTK_UNMOUNTF, so it will see it set. The code that follows is
1939 * done to ensure that any thread executing nfscl_getcl() after
1940 * this time, will see MNTK_UNMOUNTF set. nfscl_getcl() uses the
1941 * mutex for NFSLOCKCLSTATE(), so it is "m" for the following
1942 * explanation, courtesy of Alan Cox.
1943 * What follows is a snippet from Alan Cox's email at:
1944 * https://docs.FreeBSD.org/cgi/mid.cgi?BANLkTikR3d65zPHo9==08ZfJ2vmqZucEvw
1946 * 1. Set MNTK_UNMOUNTF
1947 * 2. Acquire a standard FreeBSD mutex "m".
1948 * 3. Update some data structures.
1949 * 4. Release mutex "m".
1951 * Then, other threads that acquire "m" after step 4 has occurred will
1952 * see MNTK_UNMOUNTF as set. But, other threads that beat thread X to
1953 * step 2 may or may not see MNTK_UNMOUNTF as set.
1956 if ((nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) {
1964 if ((clp->nfsc_flags & NFSCLFLAGS_INITED) == 0)
1965 panic("nfscl umount");
1968 * First, handshake with the nfscl renew thread, to terminate
1971 clp->nfsc_flags |= NFSCLFLAGS_UMOUNT;
1972 while (clp->nfsc_flags & NFSCLFLAGS_HASTHREAD)
1973 (void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT,
1977 * Now, get the exclusive lock on the client state, so
1978 * that no uses of the state are still in progress.
1981 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
1982 NFSCLSTATEMUTEXPTR, NULL);
1983 } while (!igotlock);
1987 * Free up all the state. It will expire on the server, but
1988 * maybe we should do a SetClientId/SetClientIdConfirm so
1989 * the server throws it away?
1991 LIST_REMOVE(clp, nfsc_list);
1992 nfscl_delegreturnall(clp, p);
1993 cred = newnfs_getcred();
1994 if (NFSHASNFSV4N(nmp)) {
1995 (void)nfsrpc_destroysession(nmp, clp, cred, p);
1996 (void)nfsrpc_destroyclient(nmp, clp, cred, p);
1998 (void)nfsrpc_setclient(nmp, clp, 0, NULL, cred, p);
1999 nfscl_cleanclient(clp);
2002 free(clp, M_NFSCLCLIENT);
2008 * This function is called when a server replies with NFSERR_STALECLIENTID
2009 * NFSERR_STALESTATEID or NFSERR_BADSESSION. It traverses the clientid lists,
2010 * doing Opens and Locks with reclaim. If these fail, it deletes the
2011 * corresponding state.
2014 nfscl_recover(struct nfsclclient *clp, bool *retokp, struct ucred *cred,
2017 struct nfsclowner *owp, *nowp;
2018 struct nfsclopen *op, *nop;
2019 struct nfscllockowner *lp, *nlp;
2020 struct nfscllock *lop, *nlop;
2021 struct nfscldeleg *dp, *ndp, *tdp;
2022 struct nfsmount *nmp;
2023 struct ucred *tcred;
2024 struct nfsclopenhead extra_open;
2025 struct nfscldeleghead extra_deleg;
2028 u_int32_t delegtype = NFSV4OPEN_DELEGATEWRITE, mode;
2029 int i, igotlock = 0, error, trycnt, firstlock;
2030 struct nfscllayout *lyp, *nlyp;
2034 * First, lock the client structure, so everyone else will
2035 * block when trying to use state.
2038 clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG;
2040 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
2041 NFSCLSTATEMUTEXPTR, NULL);
2042 } while (!igotlock);
2045 nmp = clp->nfsc_nmp;
2047 panic("nfscl recover");
2050 * For now, just get rid of all layouts. There may be a need
2051 * to do LayoutCommit Ops with reclaim == true later.
2053 TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp)
2054 nfscl_freelayout(lyp);
2055 TAILQ_INIT(&clp->nfsc_layout);
2056 for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++)
2057 LIST_INIT(&clp->nfsc_layouthash[i]);
2062 error = nfsrpc_setclient(nmp, clp, 1, retokp, cred, p);
2063 } while ((error == NFSERR_STALECLIENTID ||
2064 error == NFSERR_BADSESSION ||
2065 error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
2068 clp->nfsc_flags &= ~(NFSCLFLAGS_RECOVER |
2069 NFSCLFLAGS_RECVRINPROG);
2070 wakeup(&clp->nfsc_flags);
2071 nfsv4_unlock(&clp->nfsc_lock, 0);
2075 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
2076 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2079 * Mark requests already queued on the server, so that they don't
2080 * initiate another recovery cycle. Any requests already in the
2081 * queue that handle state information will have the old stale
2082 * clientid/stateid and will get a NFSERR_STALESTATEID,
2083 * NFSERR_STALECLIENTID or NFSERR_BADSESSION reply from the server.
2084 * This will be translated to NFSERR_STALEDONTRECOVER when
2085 * R_DONTRECOVER is set.
2088 TAILQ_FOREACH(rep, &nfsd_reqq, r_chain) {
2089 if (rep->r_nmp == nmp)
2090 rep->r_flags |= R_DONTRECOVER;
2095 * If nfsrpc_setclient() returns *retokp == true,
2096 * no more recovery is needed.
2102 * Now, mark all delegations "need reclaim".
2104 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list)
2105 dp->nfsdl_flags |= NFSCLDL_NEEDRECLAIM;
2107 TAILQ_INIT(&extra_deleg);
2108 LIST_INIT(&extra_open);
2110 * Now traverse the state lists, doing Open and Lock Reclaims.
2112 tcred = newnfs_getcred();
2113 recovered_one = false;
2114 owp = LIST_FIRST(&clp->nfsc_owner);
2115 while (owp != NULL) {
2116 nowp = LIST_NEXT(owp, nfsow_list);
2117 owp->nfsow_seqid = 0;
2118 op = LIST_FIRST(&owp->nfsow_open);
2119 while (op != NULL) {
2120 nop = LIST_NEXT(op, nfso_list);
2121 if (error != NFSERR_NOGRACE && error != NFSERR_BADSESSION) {
2122 /* Search for a delegation to reclaim with the open */
2123 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2124 if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
2126 if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
2127 mode = NFSV4OPEN_ACCESSWRITE;
2128 delegtype = NFSV4OPEN_DELEGATEWRITE;
2130 mode = NFSV4OPEN_ACCESSREAD;
2131 delegtype = NFSV4OPEN_DELEGATEREAD;
2133 if ((op->nfso_mode & mode) == mode &&
2134 op->nfso_fhlen == dp->nfsdl_fhlen &&
2135 !NFSBCMP(op->nfso_fh, dp->nfsdl_fh, op->nfso_fhlen))
2140 delegtype = NFSV4OPEN_DELEGATENONE;
2141 newnfs_copycred(&op->nfso_cred, tcred);
2142 error = nfscl_tryopen(nmp, NULL, op->nfso_fh,
2143 op->nfso_fhlen, op->nfso_fh, op->nfso_fhlen,
2144 op->nfso_mode, op, NULL, 0, &ndp, 1, delegtype,
2147 recovered_one = true;
2148 /* Handle any replied delegation */
2149 if (ndp != NULL && ((ndp->nfsdl_flags & NFSCLDL_WRITE)
2150 || NFSMNT_RDONLY(nmp->nm_mountp))) {
2151 if ((ndp->nfsdl_flags & NFSCLDL_WRITE))
2152 mode = NFSV4OPEN_ACCESSWRITE;
2154 mode = NFSV4OPEN_ACCESSREAD;
2155 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2156 if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
2158 if ((op->nfso_mode & mode) == mode &&
2159 op->nfso_fhlen == dp->nfsdl_fhlen &&
2160 !NFSBCMP(op->nfso_fh, dp->nfsdl_fh,
2162 dp->nfsdl_stateid = ndp->nfsdl_stateid;
2163 dp->nfsdl_sizelimit = ndp->nfsdl_sizelimit;
2164 dp->nfsdl_ace = ndp->nfsdl_ace;
2165 dp->nfsdl_change = ndp->nfsdl_change;
2166 dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
2167 if ((ndp->nfsdl_flags & NFSCLDL_RECALL))
2168 dp->nfsdl_flags |= NFSCLDL_RECALL;
2169 free(ndp, M_NFSCLDELEG);
2176 TAILQ_INSERT_HEAD(&extra_deleg, ndp, nfsdl_list);
2178 /* and reclaim all byte range locks */
2179 lp = LIST_FIRST(&op->nfso_lock);
2180 while (lp != NULL) {
2181 nlp = LIST_NEXT(lp, nfsl_list);
2184 lop = LIST_FIRST(&lp->nfsl_lock);
2185 while (lop != NULL) {
2186 nlop = LIST_NEXT(lop, nfslo_list);
2187 if (lop->nfslo_end == NFS64BITSSET)
2190 len = lop->nfslo_end - lop->nfslo_first;
2191 error = nfscl_trylock(nmp, NULL,
2192 op->nfso_fh, op->nfso_fhlen, lp,
2193 firstlock, 1, lop->nfslo_first, len,
2194 lop->nfslo_type, tcred, p);
2196 nfscl_freelock(lop, 0);
2201 /* If no locks, but a lockowner, just delete it. */
2202 if (LIST_EMPTY(&lp->nfsl_lock))
2203 nfscl_freelockowner(lp, 0);
2206 } else if (error == NFSERR_NOGRACE && !recovered_one &&
2207 NFSHASNFSV4N(nmp)) {
2209 * For NFSv4.1/4.2, the NFSERR_EXPIRED case will
2210 * actually end up here, since the client will do
2211 * a recovery for NFSERR_BADSESSION, but will get
2212 * an NFSERR_NOGRACE reply for the first "reclaim"
2214 * So, call nfscl_expireclient() to recover the
2215 * opens as best we can and then do a reclaim
2216 * complete and return.
2218 nfsrpc_reclaimcomplete(nmp, cred, p);
2219 nfscl_expireclient(clp, nmp, tcred, p);
2223 if (error != 0 && error != NFSERR_BADSESSION)
2224 nfscl_freeopen(op, 0);
2231 * Now, try and get any delegations not yet reclaimed by cobbling
2232 * to-gether an appropriate open.
2235 dp = TAILQ_FIRST(&clp->nfsc_deleg);
2236 while (dp != NULL) {
2237 ndp = TAILQ_NEXT(dp, nfsdl_list);
2238 if ((dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM)) {
2241 sizeof (struct nfsclowner), M_NFSCLOWNER, M_WAITOK);
2243 * Name must be as long an largest possible
2244 * NFSV4CL_LOCKNAMELEN. 12 for now.
2246 NFSBCOPY("RECLAIMDELEG", nowp->nfsow_owner,
2247 NFSV4CL_LOCKNAMELEN);
2248 LIST_INIT(&nowp->nfsow_open);
2249 nowp->nfsow_clp = clp;
2250 nowp->nfsow_seqid = 0;
2251 nowp->nfsow_defunct = 0;
2252 nfscl_lockinit(&nowp->nfsow_rwlock);
2255 if (error != NFSERR_NOGRACE && error != NFSERR_BADSESSION) {
2256 nop = malloc(sizeof (struct nfsclopen) +
2257 dp->nfsdl_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
2258 nop->nfso_own = nowp;
2259 if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
2260 nop->nfso_mode = NFSV4OPEN_ACCESSWRITE;
2261 delegtype = NFSV4OPEN_DELEGATEWRITE;
2263 nop->nfso_mode = NFSV4OPEN_ACCESSREAD;
2264 delegtype = NFSV4OPEN_DELEGATEREAD;
2266 nop->nfso_opencnt = 0;
2267 nop->nfso_posixlock = 1;
2268 nop->nfso_fhlen = dp->nfsdl_fhlen;
2269 NFSBCOPY(dp->nfsdl_fh, nop->nfso_fh, dp->nfsdl_fhlen);
2270 LIST_INIT(&nop->nfso_lock);
2271 nop->nfso_stateid.seqid = 0;
2272 nop->nfso_stateid.other[0] = 0;
2273 nop->nfso_stateid.other[1] = 0;
2274 nop->nfso_stateid.other[2] = 0;
2275 newnfs_copycred(&dp->nfsdl_cred, tcred);
2276 newnfs_copyincred(tcred, &nop->nfso_cred);
2278 error = nfscl_tryopen(nmp, NULL, nop->nfso_fh,
2279 nop->nfso_fhlen, nop->nfso_fh, nop->nfso_fhlen,
2280 nop->nfso_mode, nop, NULL, 0, &tdp, 1,
2281 delegtype, tcred, p);
2283 if ((tdp->nfsdl_flags & NFSCLDL_WRITE))
2284 mode = NFSV4OPEN_ACCESSWRITE;
2286 mode = NFSV4OPEN_ACCESSREAD;
2287 if ((nop->nfso_mode & mode) == mode &&
2288 nop->nfso_fhlen == tdp->nfsdl_fhlen &&
2289 !NFSBCMP(nop->nfso_fh, tdp->nfsdl_fh,
2291 dp->nfsdl_stateid = tdp->nfsdl_stateid;
2292 dp->nfsdl_sizelimit = tdp->nfsdl_sizelimit;
2293 dp->nfsdl_ace = tdp->nfsdl_ace;
2294 dp->nfsdl_change = tdp->nfsdl_change;
2295 dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
2296 if ((tdp->nfsdl_flags & NFSCLDL_RECALL))
2297 dp->nfsdl_flags |= NFSCLDL_RECALL;
2298 free(tdp, M_NFSCLDELEG);
2300 TAILQ_INSERT_HEAD(&extra_deleg, tdp, nfsdl_list);
2306 free(nop, M_NFSCLOPEN);
2307 if (error == NFSERR_NOGRACE && !recovered_one &&
2308 NFSHASNFSV4N(nmp)) {
2310 * For NFSv4.1/4.2, the NFSERR_EXPIRED case will
2311 * actually end up here, since the client will do
2312 * a recovery for NFSERR_BADSESSION, but will get
2313 * an NFSERR_NOGRACE reply for the first "reclaim"
2315 * So, call nfscl_expireclient() to recover the
2316 * opens as best we can and then do a reclaim
2317 * complete and return.
2319 nfsrpc_reclaimcomplete(nmp, cred, p);
2320 nfscl_expireclient(clp, nmp, tcred, p);
2321 free(nowp, M_NFSCLOWNER);
2325 * Couldn't reclaim it, so throw the state
2328 nfscl_cleandeleg(dp);
2329 nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
2331 recovered_one = true;
2332 LIST_INSERT_HEAD(&extra_open, nop, nfso_list);
2339 * Now, get rid of extra Opens and Delegations.
2341 LIST_FOREACH_SAFE(op, &extra_open, nfso_list, nop) {
2343 newnfs_copycred(&op->nfso_cred, tcred);
2344 error = nfscl_tryclose(op, tcred, nmp, p);
2345 if (error == NFSERR_GRACE)
2346 (void) nfs_catnap(PZERO, error, "nfsexcls");
2347 } while (error == NFSERR_GRACE);
2348 LIST_REMOVE(op, nfso_list);
2349 free(op, M_NFSCLOPEN);
2352 free(nowp, M_NFSCLOWNER);
2354 TAILQ_FOREACH_SAFE(dp, &extra_deleg, nfsdl_list, ndp) {
2356 newnfs_copycred(&dp->nfsdl_cred, tcred);
2357 error = nfscl_trydelegreturn(dp, tcred, nmp, p);
2358 if (error == NFSERR_GRACE)
2359 (void) nfs_catnap(PZERO, error, "nfsexdlg");
2360 } while (error == NFSERR_GRACE);
2361 TAILQ_REMOVE(&extra_deleg, dp, nfsdl_list);
2362 free(dp, M_NFSCLDELEG);
2365 /* For NFSv4.1 or later, do a RECLAIM_COMPLETE. */
2366 if (NFSHASNFSV4N(nmp))
2367 (void)nfsrpc_reclaimcomplete(nmp, cred, p);
2371 clp->nfsc_flags &= ~NFSCLFLAGS_RECVRINPROG;
2372 wakeup(&clp->nfsc_flags);
2373 nfsv4_unlock(&clp->nfsc_lock, 0);
2380 * This function is called when a server replies with NFSERR_EXPIRED.
2381 * It deletes all state for the client and does a fresh SetClientId/confirm.
2382 * XXX Someday it should post a signal to the process(es) that hold the
2383 * state, so they know that lock state has been lost.
2386 nfscl_hasexpired(struct nfsclclient *clp, u_int32_t clidrev, NFSPROC_T *p)
2388 struct nfsmount *nmp;
2390 int igotlock = 0, error, trycnt;
2393 * If the clientid has gone away or a new SetClientid has already
2394 * been done, just return ok.
2396 if (clp == NULL || clidrev != clp->nfsc_clientidrev)
2400 * First, lock the client structure, so everyone else will
2401 * block when trying to use state. Also, use NFSCLFLAGS_EXPIREIT so
2402 * that only one thread does the work.
2405 clp->nfsc_flags |= NFSCLFLAGS_EXPIREIT;
2407 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
2408 NFSCLSTATEMUTEXPTR, NULL);
2409 } while (!igotlock && (clp->nfsc_flags & NFSCLFLAGS_EXPIREIT));
2410 if ((clp->nfsc_flags & NFSCLFLAGS_EXPIREIT) == 0) {
2412 nfsv4_unlock(&clp->nfsc_lock, 0);
2416 clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG;
2419 nmp = clp->nfsc_nmp;
2421 panic("nfscl expired");
2422 cred = newnfs_getcred();
2425 error = nfsrpc_setclient(nmp, clp, 0, NULL, cred, p);
2426 } while ((error == NFSERR_STALECLIENTID ||
2427 error == NFSERR_BADSESSION ||
2428 error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
2431 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2434 * Expire the state for the client.
2436 nfscl_expireclient(clp, nmp, cred, p);
2438 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
2439 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2441 clp->nfsc_flags &= ~(NFSCLFLAGS_EXPIREIT | NFSCLFLAGS_RECVRINPROG);
2442 wakeup(&clp->nfsc_flags);
2443 nfsv4_unlock(&clp->nfsc_lock, 0);
2450 * This function inserts a lock in the list after insert_lop.
2453 nfscl_insertlock(struct nfscllockowner *lp, struct nfscllock *new_lop,
2454 struct nfscllock *insert_lop, int local)
2457 if ((struct nfscllockowner *)insert_lop == lp)
2458 LIST_INSERT_HEAD(&lp->nfsl_lock, new_lop, nfslo_list);
2460 LIST_INSERT_AFTER(insert_lop, new_lop, nfslo_list);
2462 nfsstatsv1.cllocallocks++;
2464 nfsstatsv1.cllocks++;
2468 * This function updates the locking for a lock owner and given file. It
2469 * maintains a list of lock ranges ordered on increasing file offset that
2470 * are NFSCLLOCK_READ or NFSCLLOCK_WRITE and non-overlapping (aka POSIX style).
2471 * It always adds new_lop to the list and sometimes uses the one pointed
2473 * Returns 1 if the locks were modified, 0 otherwise.
2476 nfscl_updatelock(struct nfscllockowner *lp, struct nfscllock **new_lopp,
2477 struct nfscllock **other_lopp, int local)
2479 struct nfscllock *new_lop = *new_lopp;
2480 struct nfscllock *lop, *tlop, *ilop;
2481 struct nfscllock *other_lop;
2482 int unlock = 0, modified = 0;
2486 * Work down the list until the lock is merged.
2488 if (new_lop->nfslo_type == F_UNLCK)
2490 ilop = (struct nfscllock *)lp;
2491 lop = LIST_FIRST(&lp->nfsl_lock);
2492 while (lop != NULL) {
2494 * Only check locks for this file that aren't before the start of
2497 if (lop->nfslo_end >= new_lop->nfslo_first) {
2498 if (new_lop->nfslo_end < lop->nfslo_first) {
2500 * If the new lock ends before the start of the
2501 * current lock's range, no merge, just insert
2506 if (new_lop->nfslo_type == lop->nfslo_type ||
2507 (new_lop->nfslo_first <= lop->nfslo_first &&
2508 new_lop->nfslo_end >= lop->nfslo_end)) {
2510 * This lock can be absorbed by the new lock/unlock.
2511 * This happens when it covers the entire range
2512 * of the old lock or is contiguous
2513 * with the old lock and is of the same type or an
2516 if (new_lop->nfslo_type != lop->nfslo_type ||
2517 new_lop->nfslo_first != lop->nfslo_first ||
2518 new_lop->nfslo_end != lop->nfslo_end)
2520 if (lop->nfslo_first < new_lop->nfslo_first)
2521 new_lop->nfslo_first = lop->nfslo_first;
2522 if (lop->nfslo_end > new_lop->nfslo_end)
2523 new_lop->nfslo_end = lop->nfslo_end;
2525 lop = LIST_NEXT(lop, nfslo_list);
2526 nfscl_freelock(tlop, local);
2531 * All these cases are for contiguous locks that are not the
2532 * same type, so they can't be merged.
2534 if (new_lop->nfslo_first <= lop->nfslo_first) {
2536 * This case is where the new lock overlaps with the
2537 * first part of the old lock. Move the start of the
2538 * old lock to just past the end of the new lock. The
2539 * new lock will be inserted in front of the old, since
2540 * ilop hasn't been updated. (We are done now.)
2542 if (lop->nfslo_first != new_lop->nfslo_end) {
2543 lop->nfslo_first = new_lop->nfslo_end;
2548 if (new_lop->nfslo_end >= lop->nfslo_end) {
2550 * This case is where the new lock overlaps with the
2551 * end of the old lock's range. Move the old lock's
2552 * end to just before the new lock's first and insert
2553 * the new lock after the old lock.
2554 * Might not be done yet, since the new lock could
2555 * overlap further locks with higher ranges.
2557 if (lop->nfslo_end != new_lop->nfslo_first) {
2558 lop->nfslo_end = new_lop->nfslo_first;
2562 lop = LIST_NEXT(lop, nfslo_list);
2566 * The final case is where the new lock's range is in the
2567 * middle of the current lock's and splits the current lock
2568 * up. Use *other_lopp to handle the second part of the
2569 * split old lock range. (We are done now.)
2570 * For unlock, we use new_lop as other_lop and tmp, since
2571 * other_lop and new_lop are the same for this case.
2572 * We noted the unlock case above, so we don't need
2573 * new_lop->nfslo_type any longer.
2575 tmp = new_lop->nfslo_first;
2577 other_lop = new_lop;
2580 other_lop = *other_lopp;
2583 other_lop->nfslo_first = new_lop->nfslo_end;
2584 other_lop->nfslo_end = lop->nfslo_end;
2585 other_lop->nfslo_type = lop->nfslo_type;
2586 lop->nfslo_end = tmp;
2587 nfscl_insertlock(lp, other_lop, lop, local);
2593 lop = LIST_NEXT(lop, nfslo_list);
2599 * Insert the new lock in the list at the appropriate place.
2602 nfscl_insertlock(lp, new_lop, ilop, local);
2610 * This function must be run as a kernel thread.
2611 * It does Renew Ops and recovery, when required.
2614 nfscl_renewthread(struct nfsclclient *clp, NFSPROC_T *p)
2616 struct nfsclowner *owp, *nowp;
2617 struct nfsclopen *op;
2618 struct nfscllockowner *lp, *nlp;
2619 struct nfscldeleghead dh;
2620 struct nfscldeleg *dp, *ndp;
2623 int error, cbpathdown, islept, igotlock, ret, clearok;
2624 uint32_t recover_done_time = 0;
2626 static time_t prevsec = 0;
2627 struct nfscllockownerfh *lfhp, *nlfhp;
2628 struct nfscllockownerfhhead lfh;
2629 struct nfscllayout *lyp, *nlyp;
2630 struct nfscldevinfo *dip, *ndip;
2631 struct nfscllayouthead rlh;
2632 struct nfsclrecalllayout *recallp;
2633 struct nfsclds *dsp;
2638 cred = newnfs_getcred();
2640 clp->nfsc_flags |= NFSCLFLAGS_HASTHREAD;
2641 mp = clp->nfsc_nmp->nm_mountp;
2644 newnfs_setroot(cred);
2646 if (clp->nfsc_flags & NFSCLFLAGS_RECOVER) {
2648 * Only allow one full recover within 1/2 of the lease
2649 * duration (nfsc_renew).
2650 * retok is value/result. If passed in set to true,
2651 * it indicates only a CreateSession operation should
2653 * If it is returned true, it indicates that the
2654 * recovery only required a CreateSession.
2657 if (recover_done_time < NFSD_MONOSEC) {
2658 recover_done_time = NFSD_MONOSEC +
2662 NFSCL_DEBUG(1, "Doing recovery, only "
2663 "createsession=%d\n", retok);
2664 nfscl_recover(clp, &retok, cred, p);
2666 if (clp->nfsc_expire <= NFSD_MONOSEC &&
2667 (clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID)) {
2668 clp->nfsc_expire = NFSD_MONOSEC + clp->nfsc_renew;
2669 clidrev = clp->nfsc_clientidrev;
2670 error = nfsrpc_renew(clp, NULL, cred, p);
2671 if (error == NFSERR_CBPATHDOWN)
2673 else if (error == NFSERR_STALECLIENTID ||
2674 error == NFSERR_BADSESSION) {
2676 clp->nfsc_flags |= NFSCLFLAGS_RECOVER;
2678 } else if (error == NFSERR_EXPIRED)
2679 (void) nfscl_hasexpired(clp, clidrev, p);
2683 if (NFSHASNFSV4N(clp->nfsc_nmp)) {
2684 /* Do renews for any DS sessions. */
2685 NFSLOCKMNT(clp->nfsc_nmp);
2686 /* Skip first entry, since the MDS is handled above. */
2687 dsp = TAILQ_FIRST(&clp->nfsc_nmp->nm_sess);
2689 dsp = TAILQ_NEXT(dsp, nfsclds_list);
2690 while (dsp != NULL) {
2691 if (dsp->nfsclds_expire <= NFSD_MONOSEC &&
2692 dsp->nfsclds_sess.nfsess_defunct == 0) {
2693 dsp->nfsclds_expire = NFSD_MONOSEC +
2695 NFSUNLOCKMNT(clp->nfsc_nmp);
2696 (void)nfsrpc_renew(clp, dsp, cred, p);
2699 dsp = TAILQ_NEXT(dsp, nfsclds_list);
2701 NFSUNLOCKMNT(clp->nfsc_nmp);
2707 /* It's a Total Recall! */
2708 nfscl_totalrecall(clp);
2711 * Now, handle defunct owners.
2713 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
2714 if (LIST_EMPTY(&owp->nfsow_open)) {
2715 if (owp->nfsow_defunct != 0)
2716 nfscl_freeopenowner(owp, 0);
2721 * Do the recall on any delegations. To avoid trouble, always
2722 * come back up here after having slept.
2726 dp = TAILQ_FIRST(&clp->nfsc_deleg);
2727 while (dp != NULL) {
2728 ndp = TAILQ_NEXT(dp, nfsdl_list);
2729 if ((dp->nfsdl_flags & NFSCLDL_RECALL)) {
2731 * Wait for outstanding I/O ops to be done.
2733 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
2735 nfsv4_unlock(&clp->nfsc_lock, 0);
2738 dp->nfsdl_rwlock.nfslock_lock |=
2740 msleep(&dp->nfsdl_rwlock,
2741 NFSCLSTATEMUTEXPTR, PVFS, "nfscld",
2743 if (NFSCL_FORCEDISM(mp))
2748 igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
2749 &islept, NFSCLSTATEMUTEXPTR, mp);
2750 if (igotlock == 0 && NFSCL_FORCEDISM(mp))
2756 newnfs_copycred(&dp->nfsdl_cred, cred);
2757 ret = nfscl_recalldeleg(clp, clp->nfsc_nmp, dp,
2758 NULL, cred, p, 1, &vp);
2760 nfscl_cleandeleg(dp);
2761 TAILQ_REMOVE(&clp->nfsc_deleg, dp,
2763 LIST_REMOVE(dp, nfsdl_hash);
2764 TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
2766 nfsstatsv1.cldelegates--;
2770 * The nfsc_lock must be released before doing
2771 * vrele(), since it might call nfs_inactive().
2772 * For the unlikely case where the vnode failed
2773 * to be acquired by nfscl_recalldeleg(), a
2774 * VOP_RECLAIM() should be in progress and it
2775 * will return the delegation.
2777 nfsv4_unlock(&clp->nfsc_lock, 0);
2790 * Clear out old delegations, if we are above the high water
2791 * mark. Only clear out ones with no state related to them.
2792 * The tailq list is in LRU order.
2794 dp = TAILQ_LAST(&clp->nfsc_deleg, nfscldeleghead);
2795 while (nfscl_delegcnt > nfscl_deleghighwater && dp != NULL) {
2796 ndp = TAILQ_PREV(dp, nfscldeleghead, nfsdl_list);
2797 if (dp->nfsdl_rwlock.nfslock_usecnt == 0 &&
2798 dp->nfsdl_rwlock.nfslock_lock == 0 &&
2799 dp->nfsdl_timestamp < NFSD_MONOSEC &&
2800 (dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_ZAPPED |
2801 NFSCLDL_NEEDRECLAIM | NFSCLDL_DELEGRET)) == 0) {
2803 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
2804 op = LIST_FIRST(&owp->nfsow_open);
2811 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
2812 if (!LIST_EMPTY(&lp->nfsl_lock)) {
2819 TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
2820 LIST_REMOVE(dp, nfsdl_hash);
2821 TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
2823 nfsstatsv1.cldelegates--;
2829 nfsv4_unlock(&clp->nfsc_lock, 0);
2832 * Do the recall on any layouts. To avoid trouble, always
2833 * come back up here after having slept.
2837 TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp) {
2838 if ((lyp->nfsly_flags & NFSLY_RECALL) != 0) {
2840 * Wait for outstanding I/O ops to be done.
2842 if (lyp->nfsly_lock.nfslock_usecnt > 0 ||
2843 (lyp->nfsly_lock.nfslock_lock &
2844 NFSV4LOCK_LOCK) != 0) {
2845 lyp->nfsly_lock.nfslock_lock |=
2847 msleep(&lyp->nfsly_lock.nfslock_lock,
2848 NFSCLSTATEMUTEXPTR, PVFS, "nfslyp",
2850 if (NFSCL_FORCEDISM(mp))
2854 /* Move the layout to the recall list. */
2855 TAILQ_REMOVE(&clp->nfsc_layout, lyp,
2857 LIST_REMOVE(lyp, nfsly_hash);
2858 TAILQ_INSERT_HEAD(&rlh, lyp, nfsly_list);
2860 /* Handle any layout commits. */
2861 if (!NFSHASNOLAYOUTCOMMIT(clp->nfsc_nmp) &&
2862 (lyp->nfsly_flags & NFSLY_WRITTEN) != 0) {
2863 lyp->nfsly_flags &= ~NFSLY_WRITTEN;
2865 NFSCL_DEBUG(3, "do layoutcommit\n");
2866 nfscl_dolayoutcommit(clp->nfsc_nmp, lyp,
2874 /* Now, look for stale layouts. */
2875 lyp = TAILQ_LAST(&clp->nfsc_layout, nfscllayouthead);
2876 while (lyp != NULL) {
2877 nlyp = TAILQ_PREV(lyp, nfscllayouthead, nfsly_list);
2878 if (lyp->nfsly_timestamp < NFSD_MONOSEC &&
2879 (lyp->nfsly_flags & NFSLY_RECALL) == 0 &&
2880 lyp->nfsly_lock.nfslock_usecnt == 0 &&
2881 lyp->nfsly_lock.nfslock_lock == 0) {
2882 NFSCL_DEBUG(4, "ret stale lay=%d\n",
2884 recallp = malloc(sizeof(*recallp),
2885 M_NFSLAYRECALL, M_NOWAIT);
2886 if (recallp == NULL)
2888 (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE,
2889 lyp, NFSLAYOUTIOMODE_ANY, 0, UINT64_MAX,
2890 lyp->nfsly_stateid.seqid, 0, 0, NULL,
2897 * Free up any unreferenced device info structures.
2899 LIST_FOREACH_SAFE(dip, &clp->nfsc_devinfo, nfsdi_list, ndip) {
2900 if (dip->nfsdi_layoutrefs == 0 &&
2901 dip->nfsdi_refcnt == 0) {
2902 NFSCL_DEBUG(4, "freeing devinfo\n");
2903 LIST_REMOVE(dip, nfsdi_list);
2904 nfscl_freedevinfo(dip);
2909 /* Do layout return(s), as required. */
2910 TAILQ_FOREACH_SAFE(lyp, &rlh, nfsly_list, nlyp) {
2911 TAILQ_REMOVE(&rlh, lyp, nfsly_list);
2912 NFSCL_DEBUG(4, "ret layout\n");
2913 nfscl_layoutreturn(clp->nfsc_nmp, lyp, cred, p);
2914 nfscl_freelayout(lyp);
2918 * Delegreturn any delegations cleaned out or recalled.
2920 TAILQ_FOREACH_SAFE(dp, &dh, nfsdl_list, ndp) {
2921 newnfs_copycred(&dp->nfsdl_cred, cred);
2922 (void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
2923 TAILQ_REMOVE(&dh, dp, nfsdl_list);
2924 free(dp, M_NFSCLDELEG);
2929 * Call nfscl_cleanupkext() once per second to check for
2930 * open/lock owners where the process has exited.
2932 mytime = NFSD_MONOSEC;
2933 if (prevsec != mytime) {
2935 nfscl_cleanupkext(clp, &lfh);
2939 * Do a ReleaseLockOwner for all lock owners where the
2940 * associated process no longer exists, as found by
2941 * nfscl_cleanupkext().
2943 newnfs_setroot(cred);
2944 SLIST_FOREACH_SAFE(lfhp, &lfh, nfslfh_list, nlfhp) {
2945 LIST_FOREACH_SAFE(lp, &lfhp->nfslfh_lock, nfsl_list,
2947 (void)nfsrpc_rellockown(clp->nfsc_nmp, lp,
2948 lfhp->nfslfh_fh, lfhp->nfslfh_len, cred,
2950 nfscl_freelockowner(lp, 0);
2957 if ((clp->nfsc_flags & NFSCLFLAGS_RECOVER) == 0)
2958 (void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT, "nfscl",
2961 if (clp->nfsc_flags & NFSCLFLAGS_UMOUNT) {
2962 clp->nfsc_flags &= ~NFSCLFLAGS_HASTHREAD;
2965 wakeup((caddr_t)clp);
2973 * Initiate state recovery. Called when NFSERR_STALECLIENTID,
2974 * NFSERR_STALESTATEID or NFSERR_BADSESSION is received.
2977 nfscl_initiate_recovery(struct nfsclclient *clp)
2983 clp->nfsc_flags |= NFSCLFLAGS_RECOVER;
2985 wakeup((caddr_t)clp);
2989 * Dump out the state stuff for debugging.
2992 nfscl_dumpstate(struct nfsmount *nmp, int openowner, int opens,
2993 int lockowner, int locks)
2995 struct nfsclclient *clp;
2996 struct nfsclowner *owp;
2997 struct nfsclopen *op;
2998 struct nfscllockowner *lp;
2999 struct nfscllock *lop;
3000 struct nfscldeleg *dp;
3004 printf("nfscl dumpstate NULL clp\n");
3008 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
3009 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
3010 if (openowner && !LIST_EMPTY(&owp->nfsow_open))
3011 printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
3012 owp->nfsow_owner[0], owp->nfsow_owner[1],
3013 owp->nfsow_owner[2], owp->nfsow_owner[3],
3015 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3017 printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
3018 op->nfso_stateid.other[0], op->nfso_stateid.other[1],
3019 op->nfso_stateid.other[2], op->nfso_opencnt,
3021 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
3023 printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
3024 lp->nfsl_owner[0], lp->nfsl_owner[1],
3025 lp->nfsl_owner[2], lp->nfsl_owner[3],
3027 lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
3028 lp->nfsl_stateid.other[2]);
3029 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
3032 printf("lck typ=%d fst=%ju end=%ju\n",
3033 lop->nfslo_type, (intmax_t)lop->nfslo_first,
3034 (intmax_t)lop->nfslo_end);
3036 printf("lck typ=%d fst=%qd end=%qd\n",
3037 lop->nfslo_type, lop->nfslo_first,
3045 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3046 if (openowner && !LIST_EMPTY(&owp->nfsow_open))
3047 printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
3048 owp->nfsow_owner[0], owp->nfsow_owner[1],
3049 owp->nfsow_owner[2], owp->nfsow_owner[3],
3051 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3053 printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
3054 op->nfso_stateid.other[0], op->nfso_stateid.other[1],
3055 op->nfso_stateid.other[2], op->nfso_opencnt,
3057 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
3059 printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
3060 lp->nfsl_owner[0], lp->nfsl_owner[1],
3061 lp->nfsl_owner[2], lp->nfsl_owner[3],
3063 lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
3064 lp->nfsl_stateid.other[2]);
3065 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
3068 printf("lck typ=%d fst=%ju end=%ju\n",
3069 lop->nfslo_type, (intmax_t)lop->nfslo_first,
3070 (intmax_t)lop->nfslo_end);
3072 printf("lck typ=%d fst=%qd end=%qd\n",
3073 lop->nfslo_type, lop->nfslo_first,
3084 * Check for duplicate open owners and opens.
3085 * (Only used as a diagnostic aid.)
3088 nfscl_dupopen(vnode_t vp, int dupopens)
3090 struct nfsclclient *clp;
3091 struct nfsclowner *owp, *owp2;
3092 struct nfsclopen *op, *op2;
3095 clp = VFSTONFS(vp->v_mount)->nm_clp;
3097 printf("nfscl dupopen NULL clp\n");
3100 nfhp = VTONFS(vp)->n_fhp;
3104 * First, search for duplicate owners.
3105 * These should never happen!
3107 LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3108 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3110 !NFSBCMP(owp->nfsow_owner, owp2->nfsow_owner,
3111 NFSV4CL_LOCKNAMELEN)) {
3113 printf("DUP OWNER\n");
3114 nfscl_dumpstate(VFSTONFS(vp->v_mount), 1, 1, 0, 0);
3121 * Now, search for duplicate stateids.
3122 * These shouldn't happen, either.
3124 LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3125 LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
3126 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3127 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3129 (op->nfso_stateid.other[0] != 0 ||
3130 op->nfso_stateid.other[1] != 0 ||
3131 op->nfso_stateid.other[2] != 0) &&
3132 op->nfso_stateid.other[0] == op2->nfso_stateid.other[0] &&
3133 op->nfso_stateid.other[1] == op2->nfso_stateid.other[1] &&
3134 op->nfso_stateid.other[2] == op2->nfso_stateid.other[2]) {
3136 printf("DUP STATEID\n");
3137 nfscl_dumpstate(VFSTONFS(vp->v_mount), 1, 1, 0, 0);
3146 * Now search for duplicate opens.
3147 * Duplicate opens for the same owner
3148 * should never occur. Other duplicates are
3149 * possible and are checked for if "dupopens"
3152 LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3153 LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
3154 if (nfhp->nfh_len == op2->nfso_fhlen &&
3155 !NFSBCMP(nfhp->nfh_fh, op2->nfso_fh, nfhp->nfh_len)) {
3156 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3157 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3158 if (op != op2 && nfhp->nfh_len == op->nfso_fhlen &&
3159 !NFSBCMP(nfhp->nfh_fh, op->nfso_fh, nfhp->nfh_len) &&
3160 (!NFSBCMP(op->nfso_own->nfsow_owner,
3161 op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN) ||
3163 if (!NFSBCMP(op->nfso_own->nfsow_owner,
3164 op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
3166 printf("BADDUP OPEN\n");
3169 printf("DUP OPEN\n");
3171 nfscl_dumpstate(VFSTONFS(vp->v_mount), 1, 1, 0,
3184 * During close, find an open that needs to be dereferenced and
3185 * dereference it. If there are no more opens for this file,
3186 * log a message to that effect.
3187 * Opens aren't actually Close'd until VOP_INACTIVE() is performed
3188 * on the file's vnode.
3189 * This is the safe way, since it is difficult to identify
3190 * which open the close is for and I/O can be performed after the
3191 * close(2) system call when a file is mmap'd.
3192 * If it returns 0 for success, there will be a referenced
3193 * clp returned via clpp.
3196 nfscl_getclose(vnode_t vp, struct nfsclclient **clpp)
3198 struct nfsclclient *clp;
3199 struct nfsclowner *owp;
3200 struct nfsclopen *op;
3201 struct nfscldeleg *dp;
3205 error = nfscl_getcl(vp->v_mount, NULL, NULL, false, &clp);
3210 nfhp = VTONFS(vp)->n_fhp;
3214 * First, look for one under a delegation that was locally issued
3215 * and just decrement the opencnt for it. Since all my Opens against
3216 * the server are DENY_NONE, I don't see a problem with hanging
3217 * onto them. (It is much easier to use one of the extant Opens
3218 * that I already have on the server when a Delegation is recalled
3219 * than to do fresh Opens.) Someday, I might need to rethink this, but.
3221 dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
3223 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
3224 op = LIST_FIRST(&owp->nfsow_open);
3227 * Since a delegation is for a file, there
3228 * should never be more than one open for
3231 if (LIST_NEXT(op, nfso_list) != NULL)
3232 panic("nfscdeleg opens");
3233 if (notdecr && op->nfso_opencnt > 0) {
3242 /* Now process the opens against the server. */
3243 LIST_FOREACH(op, NFSCLOPENHASH(clp, nfhp->nfh_fh, nfhp->nfh_len),
3245 if (op->nfso_fhlen == nfhp->nfh_len &&
3246 !NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
3248 /* Found an open, decrement cnt if possible */
3249 if (notdecr && op->nfso_opencnt > 0) {
3254 * There are more opens, so just return.
3256 if (op->nfso_opencnt > 0) {
3264 printf("nfscl: never fnd open\n");
3269 nfscl_doclose(vnode_t vp, struct nfsclclient **clpp, NFSPROC_T *p)
3271 struct nfsclclient *clp;
3272 struct nfsclowner *owp, *nowp;
3273 struct nfsclopen *op;
3274 struct nfscldeleg *dp;
3276 struct nfsclrecalllayout *recallp;
3279 error = nfscl_getcl(vp->v_mount, NULL, NULL, false, &clp);
3284 nfhp = VTONFS(vp)->n_fhp;
3285 recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_WAITOK);
3288 * First get rid of the local Open structures, which should be no
3291 dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
3293 LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) {
3294 op = LIST_FIRST(&owp->nfsow_open);
3296 KASSERT((op->nfso_opencnt == 0),
3297 ("nfscl: bad open cnt on deleg"));
3298 nfscl_freeopen(op, 1);
3300 nfscl_freeopenowner(owp, 1);
3304 /* Return any layouts marked return on close. */
3305 nfscl_retoncloselayout(vp, clp, nfhp->nfh_fh, nfhp->nfh_len, &recallp);
3307 /* Now process the opens against the server. */
3309 LIST_FOREACH(op, NFSCLOPENHASH(clp, nfhp->nfh_fh, nfhp->nfh_len),
3311 if (op->nfso_fhlen == nfhp->nfh_len &&
3312 !NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
3314 /* Found an open, close it. */
3316 KASSERT((op->nfso_opencnt == 0),
3317 ("nfscl: bad open cnt on server (%d)",
3321 nfsrpc_doclose(VFSTONFS(vp->v_mount), op, p);
3328 * recallp has been set NULL by nfscl_retoncloselayout() if it was
3329 * used by the function, but calling free() with a NULL pointer is ok.
3331 free(recallp, M_NFSLAYRECALL);
3336 * Return all delegations on this client.
3337 * (Must be called with client sleep lock.)
3340 nfscl_delegreturnall(struct nfsclclient *clp, NFSPROC_T *p)
3342 struct nfscldeleg *dp, *ndp;
3345 cred = newnfs_getcred();
3346 TAILQ_FOREACH_SAFE(dp, &clp->nfsc_deleg, nfsdl_list, ndp) {
3347 nfscl_cleandeleg(dp);
3348 (void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
3349 nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
3355 * Return any delegation for this vp.
3358 nfscl_delegreturnvp(vnode_t vp, NFSPROC_T *p)
3360 struct nfsclclient *clp;
3361 struct nfscldeleg *dp;
3364 struct nfsmount *nmp;
3366 nmp = VFSTONFS(vp->v_mount);
3368 if ((nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0) {
3374 cred = newnfs_getcred();
3379 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
3380 np->n_fhp->nfh_len);
3382 nfscl_cleandeleg(dp);
3383 nfscl_freedeleg(&clp->nfsc_deleg, dp, false);
3385 newnfs_copycred(&dp->nfsdl_cred, cred);
3386 nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
3387 free(dp, M_NFSCLDELEG);
3394 * Do a callback RPC.
3397 nfscl_docb(struct nfsrv_descript *nd, NFSPROC_T *p)
3399 int clist, gotseq_ok, i, j, k, op, rcalls;
3401 struct nfsclclient *clp;
3402 struct nfscldeleg *dp = NULL;
3403 int numops, taglen = -1, error = 0, trunc __unused;
3404 u_int32_t minorvers = 0, retops = 0, *retopsp = NULL, *repp, cbident;
3405 u_char tag[NFSV4_SMALLSTR + 1], *tagstr;
3411 nfsattrbit_t attrbits, rattrbits;
3412 nfsv4stateid_t stateid;
3413 uint32_t seqid, slotid = 0, highslot, cachethis __unused;
3414 uint8_t sessionid[NFSX_V4SESSIONID];
3416 struct nfscllayout *lyp;
3417 uint64_t filesid[2], len, off;
3418 int changed, gotone, laytype, recalltype;
3420 struct nfsclrecalllayout *recallp = NULL;
3421 struct nfsclsession *tsep;
3425 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3426 taglen = fxdr_unsigned(int, *tl);
3431 if (taglen <= NFSV4_SMALLSTR)
3434 tagstr = malloc(taglen + 1, M_TEMP, M_WAITOK);
3435 error = nfsrv_mtostr(nd, tagstr, taglen);
3437 if (taglen > NFSV4_SMALLSTR)
3438 free(tagstr, M_TEMP);
3442 (void) nfsm_strtom(nd, tag, taglen);
3443 if (taglen > NFSV4_SMALLSTR) {
3444 free(tagstr, M_TEMP);
3446 NFSM_BUILD(retopsp, u_int32_t *, NFSX_UNSIGNED);
3447 NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3448 minorvers = fxdr_unsigned(u_int32_t, *tl++);
3449 if (minorvers != NFSV4_MINORVERSION &&
3450 minorvers != NFSV41_MINORVERSION &&
3451 minorvers != NFSV42_MINORVERSION)
3452 nd->nd_repstat = NFSERR_MINORVERMISMATCH;
3453 cbident = fxdr_unsigned(u_int32_t, *tl++);
3457 numops = fxdr_unsigned(int, *tl);
3459 * Loop around doing the sub ops.
3461 for (i = 0; i < numops; i++) {
3462 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3463 NFSM_BUILD(repp, u_int32_t *, 2 * NFSX_UNSIGNED);
3465 op = fxdr_unsigned(int, *tl);
3466 if (op < NFSV4OP_CBGETATTR ||
3467 (op > NFSV4OP_CBRECALL && minorvers == NFSV4_MINORVERSION) ||
3468 (op > NFSV4OP_CBNOTIFYDEVID &&
3469 minorvers == NFSV41_MINORVERSION) ||
3470 (op > NFSV4OP_CBOFFLOAD &&
3471 minorvers == NFSV42_MINORVERSION)) {
3472 nd->nd_repstat = NFSERR_OPILLEGAL;
3473 *repp = nfscl_errmap(nd, minorvers);
3477 nd->nd_procnum = op;
3478 if (op < NFSV42_CBNOPS)
3479 nfsstatsv1.cbrpccnt[nd->nd_procnum]++;
3481 case NFSV4OP_CBGETATTR:
3482 NFSCL_DEBUG(4, "cbgetattr\n");
3485 error = nfsm_getfh(nd, &nfhp);
3487 error = nfsrv_getattrbits(nd, &attrbits,
3489 if (error == 0 && i == 0 &&
3490 minorvers != NFSV4_MINORVERSION)
3491 error = NFSERR_OPNOTINSESS;
3493 mp = nfscl_getmnt(minorvers, sessionid, cbident,
3496 error = NFSERR_SERVERFAULT;
3499 error = nfscl_ngetreopen(mp, nfhp->nfh_fh,
3500 nfhp->nfh_len, p, &np);
3505 NFSZERO_ATTRBIT(&rattrbits);
3507 dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
3510 if (NFSISSET_ATTRBIT(&attrbits,
3513 va.va_size = np->n_size;
3517 NFSSETBIT_ATTRBIT(&rattrbits,
3520 if (NFSISSET_ATTRBIT(&attrbits,
3521 NFSATTRBIT_CHANGE)) {
3525 (np->n_flag & NDELEGMOD))
3527 NFSSETBIT_ATTRBIT(&rattrbits,
3531 error = NFSERR_SERVERFAULT;
3539 free(nfhp, M_NFSFH);
3541 (void) nfsv4_fillattr(nd, NULL, NULL, NULL, &va,
3542 NULL, 0, &rattrbits, NULL, p, 0, 0, 0, 0,
3545 case NFSV4OP_CBRECALL:
3546 NFSCL_DEBUG(4, "cbrecall\n");
3547 NFSM_DISSECT(tl, u_int32_t *, NFSX_STATEID +
3549 stateid.seqid = *tl++;
3550 NFSBCOPY((caddr_t)tl, (caddr_t)stateid.other,
3552 tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
3553 trunc = fxdr_unsigned(int, *tl);
3554 error = nfsm_getfh(nd, &nfhp);
3555 if (error == 0 && i == 0 &&
3556 minorvers != NFSV4_MINORVERSION)
3557 error = NFSERR_OPNOTINSESS;
3560 if (minorvers == NFSV4_MINORVERSION)
3561 clp = nfscl_getclnt(cbident);
3563 clp = nfscl_getclntsess(sessionid);
3565 dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
3567 if (dp != NULL && (dp->nfsdl_flags &
3568 NFSCLDL_DELEGRET) == 0) {
3571 wakeup((caddr_t)clp);
3574 error = NFSERR_SERVERFAULT;
3579 free(nfhp, M_NFSFH);
3581 case NFSV4OP_CBLAYOUTRECALL:
3582 NFSCL_DEBUG(4, "cblayrec\n");
3584 NFSM_DISSECT(tl, uint32_t *, 4 * NFSX_UNSIGNED);
3585 laytype = fxdr_unsigned(int, *tl++);
3586 iomode = fxdr_unsigned(uint32_t, *tl++);
3587 if (newnfs_true == *tl++)
3591 recalltype = fxdr_unsigned(int, *tl);
3592 NFSCL_DEBUG(4, "layt=%d iom=%d ch=%d rectyp=%d\n",
3593 laytype, iomode, changed, recalltype);
3594 recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL,
3596 if (laytype != NFSLAYOUT_NFSV4_1_FILES &&
3597 laytype != NFSLAYOUT_FLEXFILE)
3598 error = NFSERR_NOMATCHLAYOUT;
3599 else if (recalltype == NFSLAYOUTRETURN_FILE) {
3600 error = nfsm_getfh(nd, &nfhp);
3601 NFSCL_DEBUG(4, "retfile getfh=%d\n", error);
3604 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_HYPER +
3606 off = fxdr_hyper(tl); tl += 2;
3607 len = fxdr_hyper(tl); tl += 2;
3608 stateid.seqid = fxdr_unsigned(uint32_t, *tl++);
3609 NFSBCOPY(tl, stateid.other, NFSX_STATEIDOTHER);
3610 if (minorvers == NFSV4_MINORVERSION)
3611 error = NFSERR_NOTSUPP;
3613 error = NFSERR_OPNOTINSESS;
3614 NFSCL_DEBUG(4, "off=%ju len=%ju sq=%u err=%d\n",
3615 (uintmax_t)off, (uintmax_t)len,
3616 stateid.seqid, error);
3619 clp = nfscl_getclntsess(sessionid);
3620 NFSCL_DEBUG(4, "cbly clp=%p\n", clp);
3622 lyp = nfscl_findlayout(clp,
3625 NFSCL_DEBUG(4, "cblyp=%p\n",
3630 NFSLY_FLEXFILE)) != 0 &&
3631 !NFSBCMP(stateid.other,
3632 lyp->nfsly_stateid.other,
3633 NFSX_STATEIDOTHER)) {
3643 lyp->nfsly_stateid.seqid)
3644 lyp->nfsly_stateid.seqid =
3652 lyp->nfsly_stateid.seqid);
3655 NFSERR_NOMATCHLAYOUT;
3657 error = NFSERR_NOMATCHLAYOUT;
3660 free(nfhp, M_NFSFH);
3661 } else if (recalltype == NFSLAYOUTRETURN_FSID) {
3662 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_HYPER);
3663 filesid[0] = fxdr_hyper(tl); tl += 2;
3664 filesid[1] = fxdr_hyper(tl); tl += 2;
3667 clp = nfscl_getclntsess(sessionid);
3669 TAILQ_FOREACH(lyp, &clp->nfsc_layout,
3671 if (lyp->nfsly_filesid[0] ==
3673 lyp->nfsly_filesid[1] ==
3680 lyp->nfsly_stateid.seqid,
3690 error = NFSERR_NOMATCHLAYOUT;
3692 error = NFSERR_NOMATCHLAYOUT;
3694 } else if (recalltype == NFSLAYOUTRETURN_ALL) {
3697 clp = nfscl_getclntsess(sessionid);
3699 TAILQ_FOREACH(lyp, &clp->nfsc_layout,
3701 error = nfscl_layoutrecall(
3702 recalltype, lyp, iomode, 0,
3704 lyp->nfsly_stateid.seqid,
3705 0, 0, NULL, recallp);
3712 error = NFSERR_NOMATCHLAYOUT;
3714 error = NFSERR_NOMATCHLAYOUT;
3717 error = NFSERR_NOMATCHLAYOUT;
3718 if (recallp != NULL) {
3719 free(recallp, M_NFSLAYRECALL);
3723 case NFSV4OP_CBSEQUENCE:
3724 NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID +
3726 bcopy(tl, sessionid, NFSX_V4SESSIONID);
3727 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3728 seqid = fxdr_unsigned(uint32_t, *tl++);
3729 slotid = fxdr_unsigned(uint32_t, *tl++);
3730 highslot = fxdr_unsigned(uint32_t, *tl++);
3732 /* Throw away the referring call stuff. */
3733 clist = fxdr_unsigned(int, *tl);
3734 for (j = 0; j < clist; j++) {
3735 NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID +
3737 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3738 rcalls = fxdr_unsigned(int, *tl);
3739 for (k = 0; k < rcalls; k++) {
3740 NFSM_DISSECT(tl, uint32_t *,
3746 clp = nfscl_getclntsess(sessionid);
3748 error = NFSERR_SERVERFAULT;
3750 error = NFSERR_SEQUENCEPOS;
3752 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3753 error = nfsv4_seqsession(seqid, slotid,
3754 highslot, tsep->nfsess_cbslots, &rep,
3755 tsep->nfsess_backslots);
3758 if (error == 0 || error == NFSERR_REPLYFROMCACHE) {
3762 * Handle a reply for a retried
3763 * callback. The reply will be
3764 * re-inserted in the session cache
3765 * by the nfsv4_seqsess_cacherep() call
3768 KASSERT(error == NFSERR_REPLYFROMCACHE,
3769 ("cbsequence: non-NULL rep"));
3770 NFSCL_DEBUG(4, "Got cbretry\n");
3771 m_freem(nd->nd_mreq);
3776 NFSM_BUILD(tl, uint32_t *,
3777 NFSX_V4SESSIONID + 4 * NFSX_UNSIGNED);
3778 bcopy(sessionid, tl, NFSX_V4SESSIONID);
3779 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3780 *tl++ = txdr_unsigned(seqid);
3781 *tl++ = txdr_unsigned(slotid);
3782 *tl++ = txdr_unsigned(NFSV4_CBSLOTS - 1);
3783 *tl = txdr_unsigned(NFSV4_CBSLOTS - 1);
3787 if (i == 0 && minorvers != NFSV4_MINORVERSION)
3788 error = NFSERR_OPNOTINSESS;
3790 NFSCL_DEBUG(1, "unsupp callback %d\n", op);
3791 error = NFSERR_NOTSUPP;
3796 if (error == EBADRPC || error == NFSERR_BADXDR) {
3797 nd->nd_repstat = NFSERR_BADXDR;
3799 nd->nd_repstat = error;
3804 if (nd->nd_repstat) {
3805 *repp = nfscl_errmap(nd, minorvers);
3808 *repp = 0; /* NFS4_OK */
3811 if (recallp != NULL)
3812 free(recallp, M_NFSLAYRECALL);
3814 if (error == EBADRPC || error == NFSERR_BADXDR)
3815 nd->nd_repstat = NFSERR_BADXDR;
3817 printf("nfsv4 comperr1=%d\n", error);
3820 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
3824 *retopsp = txdr_unsigned(retops);
3826 *nd->nd_errp = nfscl_errmap(nd, minorvers);
3828 if (gotseq_ok != 0) {
3829 rep = m_copym(nd->nd_mreq, 0, M_COPYALL, M_WAITOK);
3831 clp = nfscl_getclntsess(sessionid);
3833 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3834 nfsv4_seqsess_cacherep(slotid, tsep->nfsess_cbslots,
3845 * Generate the next cbident value. Basically just increment a static value
3846 * and then check that it isn't already in the list, if it has wrapped around.
3849 nfscl_nextcbident(void)
3851 struct nfsclclient *clp;
3853 static u_int32_t nextcbident = 0;
3854 static int haswrapped = 0;
3857 if (nextcbident == 0)
3861 * Search the clientid list for one already using this cbident.
3866 LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3867 if (clp->nfsc_cbident == nextcbident) {
3877 return (nextcbident);
3881 * Get the mount point related to a given cbident or session and busy it.
3884 nfscl_getmnt(int minorvers, uint8_t *sessionid, u_int32_t cbident,
3885 struct nfsclclient **clpp)
3887 struct nfsclclient *clp;
3890 struct nfsclsession *tsep;
3894 LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3895 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3896 if (minorvers == NFSV4_MINORVERSION) {
3897 if (clp->nfsc_cbident == cbident)
3899 } else if (!NFSBCMP(tsep->nfsess_sessionid, sessionid,
3907 mp = clp->nfsc_nmp->nm_mountp;
3910 error = vfs_busy(mp, 0);
3919 * Get the clientid pointer related to a given cbident.
3921 static struct nfsclclient *
3922 nfscl_getclnt(u_int32_t cbident)
3924 struct nfsclclient *clp;
3926 LIST_FOREACH(clp, &nfsclhead, nfsc_list)
3927 if (clp->nfsc_cbident == cbident)
3933 * Get the clientid pointer related to a given sessionid.
3935 static struct nfsclclient *
3936 nfscl_getclntsess(uint8_t *sessionid)
3938 struct nfsclclient *clp;
3939 struct nfsclsession *tsep;
3941 LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3942 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3943 if (!NFSBCMP(tsep->nfsess_sessionid, sessionid,
3951 * Search for a lock conflict locally on the client. A conflict occurs if
3952 * - not same owner and overlapping byte range and at least one of them is
3953 * a write lock or this is an unlock.
3956 nfscl_localconflict(struct nfsclclient *clp, u_int8_t *fhp, int fhlen,
3957 struct nfscllock *nlop, u_int8_t *own, struct nfscldeleg *dp,
3958 struct nfscllock **lopp)
3960 struct nfsclopen *op;
3964 ret = nfscl_checkconflict(&dp->nfsdl_lock, nlop, own, lopp);
3968 LIST_FOREACH(op, NFSCLOPENHASH(clp, fhp, fhlen), nfso_hash) {
3969 if (op->nfso_fhlen == fhlen &&
3970 !NFSBCMP(op->nfso_fh, fhp, fhlen)) {
3971 ret = nfscl_checkconflict(&op->nfso_lock, nlop,
3981 nfscl_checkconflict(struct nfscllockownerhead *lhp, struct nfscllock *nlop,
3982 u_int8_t *own, struct nfscllock **lopp)
3984 struct nfscllockowner *lp;
3985 struct nfscllock *lop;
3987 LIST_FOREACH(lp, lhp, nfsl_list) {
3988 if (NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
3989 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
3990 if (lop->nfslo_first >= nlop->nfslo_end)
3992 if (lop->nfslo_end <= nlop->nfslo_first)
3994 if (lop->nfslo_type == F_WRLCK ||
3995 nlop->nfslo_type == F_WRLCK ||
3996 nlop->nfslo_type == F_UNLCK) {
3999 return (NFSERR_DENIED);
4008 * Check for a local conflicting lock.
4011 nfscl_lockt(vnode_t vp, struct nfsclclient *clp, u_int64_t off,
4012 u_int64_t len, struct flock *fl, NFSPROC_T *p, void *id, int flags)
4014 struct nfscllock *lop, nlck;
4015 struct nfscldeleg *dp;
4017 u_int8_t own[NFSV4CL_LOCKNAMELEN];
4020 nlck.nfslo_type = fl->l_type;
4021 nlck.nfslo_first = off;
4022 if (len == NFS64BITSSET) {
4023 nlck.nfslo_end = NFS64BITSSET;
4025 nlck.nfslo_end = off + len;
4026 if (nlck.nfslo_end <= nlck.nfslo_first)
4027 return (NFSERR_INVAL);
4030 nfscl_filllockowner(id, own, flags);
4032 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4033 error = nfscl_localconflict(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len,
4034 &nlck, own, dp, &lop);
4036 fl->l_whence = SEEK_SET;
4037 fl->l_start = lop->nfslo_first;
4038 if (lop->nfslo_end == NFS64BITSSET)
4041 fl->l_len = lop->nfslo_end - lop->nfslo_first;
4042 fl->l_pid = (pid_t)0;
4043 fl->l_type = lop->nfslo_type;
4044 error = -1; /* no RPC required */
4045 } else if (dp != NULL && ((dp->nfsdl_flags & NFSCLDL_WRITE) ||
4046 fl->l_type == F_RDLCK)) {
4048 * The delegation ensures that there isn't a conflicting
4049 * lock on the server, so return -1 to indicate an RPC
4052 fl->l_type = F_UNLCK;
4060 * Handle Recall of a delegation.
4061 * The clp must be exclusive locked when this is called.
4064 nfscl_recalldeleg(struct nfsclclient *clp, struct nfsmount *nmp,
4065 struct nfscldeleg *dp, vnode_t vp, struct ucred *cred, NFSPROC_T *p,
4066 int called_from_renewthread, vnode_t *vpp)
4068 struct nfsclowner *owp, *lowp, *nowp;
4069 struct nfsclopen *op, *lop;
4070 struct nfscllockowner *lp;
4071 struct nfscllock *lckp;
4076 KASSERT(vpp != NULL, ("nfscl_recalldeleg: vpp NULL"));
4079 * First, get a vnode for the file. This is needed to do RPCs.
4081 ret = nfscl_ngetreopen(nmp->nm_mountp, dp->nfsdl_fh,
4082 dp->nfsdl_fhlen, p, &np);
4085 * File isn't open, so nothing to move over to the
4095 dp->nfsdl_flags &= ~NFSCLDL_MODTIMESET;
4098 * Ok, if it's a write delegation, flush data to the server, so
4099 * that close/open consistency is retained.
4103 if ((dp->nfsdl_flags & NFSCLDL_WRITE) && (np->n_flag & NMODIFIED)) {
4104 np->n_flag |= NDELEGRECALL;
4106 ret = ncl_flush(vp, MNT_WAIT, p, 1, called_from_renewthread);
4108 np->n_flag &= ~NDELEGRECALL;
4110 NFSINVALATTRCACHE(np);
4112 if (ret == EIO && called_from_renewthread != 0) {
4114 * If the flush failed with EIO for the renew thread,
4115 * return now, so that the dirty buffer will be flushed
4122 * Now, for each openowner with opens issued locally, move them
4123 * over to state against the server.
4125 LIST_FOREACH(lowp, &dp->nfsdl_owner, nfsow_list) {
4126 lop = LIST_FIRST(&lowp->nfsow_open);
4128 if (LIST_NEXT(lop, nfso_list) != NULL)
4129 panic("nfsdlg mult opens");
4131 * Look for the same openowner against the server.
4133 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
4134 if (!NFSBCMP(lowp->nfsow_owner,
4135 owp->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
4136 newnfs_copycred(&dp->nfsdl_cred, cred);
4137 ret = nfscl_moveopen(vp, clp, nmp, lop,
4139 if (ret == NFSERR_STALECLIENTID ||
4140 ret == NFSERR_STALEDONTRECOVER ||
4141 ret == NFSERR_BADSESSION)
4144 nfscl_freeopen(lop, 1);
4153 * If no openowner found, create one and get an open
4158 sizeof (struct nfsclowner), M_NFSCLOWNER,
4160 nfscl_newopen(clp, NULL, &owp, &nowp, &op,
4161 NULL, lowp->nfsow_owner, dp->nfsdl_fh,
4162 dp->nfsdl_fhlen, NULL, NULL);
4163 newnfs_copycred(&dp->nfsdl_cred, cred);
4164 ret = nfscl_moveopen(vp, clp, nmp, lop,
4167 nfscl_freeopenowner(owp, 0);
4168 if (ret == NFSERR_STALECLIENTID ||
4169 ret == NFSERR_STALEDONTRECOVER ||
4170 ret == NFSERR_BADSESSION)
4173 nfscl_freeopen(lop, 1);
4183 * Now, get byte range locks for any locks done locally.
4185 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4186 LIST_FOREACH(lckp, &lp->nfsl_lock, nfslo_list) {
4187 newnfs_copycred(&dp->nfsdl_cred, cred);
4188 ret = nfscl_relock(vp, clp, nmp, lp, lckp, cred, p);
4189 if (ret == NFSERR_STALESTATEID ||
4190 ret == NFSERR_STALEDONTRECOVER ||
4191 ret == NFSERR_STALECLIENTID ||
4192 ret == NFSERR_BADSESSION)
4202 * Move a locally issued open over to an owner on the state list.
4203 * SIDE EFFECT: If it needs to sleep (do an rpc), it unlocks clstate and
4204 * returns with it unlocked.
4207 nfscl_moveopen(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp,
4208 struct nfsclopen *lop, struct nfsclowner *owp, struct nfscldeleg *dp,
4209 struct ucred *cred, NFSPROC_T *p)
4211 struct nfsclopen *op, *nop;
4212 struct nfscldeleg *ndp;
4214 int error = 0, newone;
4217 * First, look for an appropriate open, If found, just increment the
4220 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
4221 if ((op->nfso_mode & lop->nfso_mode) == lop->nfso_mode &&
4222 op->nfso_fhlen == lop->nfso_fhlen &&
4223 !NFSBCMP(op->nfso_fh, lop->nfso_fh, op->nfso_fhlen)) {
4224 op->nfso_opencnt += lop->nfso_opencnt;
4225 nfscl_freeopen(lop, 1);
4230 /* No appropriate open, so we have to do one against the server. */
4232 nop = malloc(sizeof (struct nfsclopen) +
4233 lop->nfso_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
4234 nop->nfso_hash.le_prev = NULL;
4236 nfscl_newopen(clp, NULL, &owp, NULL, &op, &nop, owp->nfsow_owner,
4237 lop->nfso_fh, lop->nfso_fhlen, cred, &newone);
4239 error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data, np->n_v4->n4_fhlen,
4240 lop->nfso_fh, lop->nfso_fhlen, lop->nfso_mode, op,
4241 NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, &ndp, 0, 0, cred, p);
4244 nfscl_freeopen(op, 0);
4246 op->nfso_mode |= lop->nfso_mode;
4247 op->nfso_opencnt += lop->nfso_opencnt;
4248 nfscl_freeopen(lop, 1);
4251 free(nop, M_NFSCLOPEN);
4254 * What should I do with the returned delegation, since the
4255 * delegation is being recalled? For now, just printf and
4258 printf("Moveopen returned deleg\n");
4259 free(ndp, M_NFSCLDELEG);
4265 * Recall all delegations on this client.
4268 nfscl_totalrecall(struct nfsclclient *clp)
4270 struct nfscldeleg *dp;
4272 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
4273 if ((dp->nfsdl_flags & NFSCLDL_DELEGRET) == 0)
4274 dp->nfsdl_flags |= NFSCLDL_RECALL;
4279 * Relock byte ranges. Called for delegation recall and state expiry.
4282 nfscl_relock(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp,
4283 struct nfscllockowner *lp, struct nfscllock *lop, struct ucred *cred,
4286 struct nfscllockowner *nlp;
4289 int error, newone, donelocally;
4291 off = lop->nfslo_first;
4292 len = lop->nfslo_end - lop->nfslo_first;
4293 error = nfscl_getbytelock(vp, off, len, lop->nfslo_type, cred, p,
4294 clp, 1, NULL, lp->nfsl_lockflags, lp->nfsl_owner,
4295 lp->nfsl_openowner, &nlp, &newone, &donelocally);
4296 if (error || donelocally)
4298 nfhp = VTONFS(vp)->n_fhp;
4299 error = nfscl_trylock(nmp, vp, nfhp->nfh_fh,
4300 nfhp->nfh_len, nlp, newone, 0, off,
4301 len, lop->nfslo_type, cred, p);
4303 nfscl_freelockowner(nlp, 0);
4308 * Called to re-open a file. Basically get a vnode for the file handle
4309 * and then call nfsrpc_openrpc() to do the rest.
4312 nfsrpc_reopen(struct nfsmount *nmp, u_int8_t *fhp, int fhlen,
4313 u_int32_t mode, struct nfsclopen *op, struct nfscldeleg **dpp,
4314 struct ucred *cred, NFSPROC_T *p)
4320 error = nfscl_ngetreopen(nmp->nm_mountp, fhp, fhlen, p, &np);
4324 if (np->n_v4 != NULL) {
4325 error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data,
4326 np->n_v4->n4_fhlen, fhp, fhlen, mode, op,
4327 NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, dpp, 0, 0,
4337 * Try an open against the server. Just call nfsrpc_openrpc(), retrying while
4338 * NFSERR_DELAY. Also, try system credentials, if the passed in credentials
4342 nfscl_tryopen(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen,
4343 u_int8_t *newfhp, int newfhlen, u_int32_t mode, struct nfsclopen *op,
4344 u_int8_t *name, int namelen, struct nfscldeleg **ndpp,
4345 int reclaim, u_int32_t delegtype, struct ucred *cred, NFSPROC_T *p)
4350 error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp, newfhlen,
4351 mode, op, name, namelen, ndpp, reclaim, delegtype, cred, p,
4353 if (error == NFSERR_DELAY)
4354 (void) nfs_catnap(PZERO, error, "nfstryop");
4355 } while (error == NFSERR_DELAY);
4356 if (error == EAUTH || error == EACCES) {
4357 /* Try again using system credentials */
4358 newnfs_setroot(cred);
4360 error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp,
4361 newfhlen, mode, op, name, namelen, ndpp, reclaim,
4362 delegtype, cred, p, 1, 0);
4363 if (error == NFSERR_DELAY)
4364 (void) nfs_catnap(PZERO, error, "nfstryop");
4365 } while (error == NFSERR_DELAY);
4371 * Try a byte range lock. Just loop on nfsrpc_lock() while it returns
4372 * NFSERR_DELAY. Also, retry with system credentials, if the provided
4376 nfscl_trylock(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp,
4377 int fhlen, struct nfscllockowner *nlp, int newone, int reclaim,
4378 u_int64_t off, u_int64_t len, short type, struct ucred *cred, NFSPROC_T *p)
4380 struct nfsrv_descript nfsd, *nd = &nfsd;
4384 error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp, newone,
4385 reclaim, off, len, type, cred, p, 0);
4386 if (!error && nd->nd_repstat == NFSERR_DELAY)
4387 (void) nfs_catnap(PZERO, (int)nd->nd_repstat,
4389 } while (!error && nd->nd_repstat == NFSERR_DELAY);
4391 error = nd->nd_repstat;
4392 if (error == EAUTH || error == EACCES) {
4393 /* Try again using root credentials */
4394 newnfs_setroot(cred);
4396 error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp,
4397 newone, reclaim, off, len, type, cred, p, 1);
4398 if (!error && nd->nd_repstat == NFSERR_DELAY)
4399 (void) nfs_catnap(PZERO, (int)nd->nd_repstat,
4401 } while (!error && nd->nd_repstat == NFSERR_DELAY);
4403 error = nd->nd_repstat;
4409 * Try a delegreturn against the server. Just call nfsrpc_delegreturn(),
4410 * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
4414 nfscl_trydelegreturn(struct nfscldeleg *dp, struct ucred *cred,
4415 struct nfsmount *nmp, NFSPROC_T *p)
4420 error = nfsrpc_delegreturn(dp, cred, nmp, p, 0);
4421 if (error == NFSERR_DELAY)
4422 (void) nfs_catnap(PZERO, error, "nfstrydp");
4423 } while (error == NFSERR_DELAY);
4424 if (error == EAUTH || error == EACCES) {
4425 /* Try again using system credentials */
4426 newnfs_setroot(cred);
4428 error = nfsrpc_delegreturn(dp, cred, nmp, p, 1);
4429 if (error == NFSERR_DELAY)
4430 (void) nfs_catnap(PZERO, error, "nfstrydp");
4431 } while (error == NFSERR_DELAY);
4437 * Try a close against the server. Just call nfsrpc_closerpc(),
4438 * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
4442 nfscl_tryclose(struct nfsclopen *op, struct ucred *cred,
4443 struct nfsmount *nmp, NFSPROC_T *p)
4445 struct nfsrv_descript nfsd, *nd = &nfsd;
4449 error = nfsrpc_closerpc(nd, nmp, op, cred, p, 0);
4450 if (error == NFSERR_DELAY)
4451 (void) nfs_catnap(PZERO, error, "nfstrycl");
4452 } while (error == NFSERR_DELAY);
4453 if (error == EAUTH || error == EACCES) {
4454 /* Try again using system credentials */
4455 newnfs_setroot(cred);
4457 error = nfsrpc_closerpc(nd, nmp, op, cred, p, 1);
4458 if (error == NFSERR_DELAY)
4459 (void) nfs_catnap(PZERO, error, "nfstrycl");
4460 } while (error == NFSERR_DELAY);
4466 * Decide if a delegation on a file permits close without flushing writes
4467 * to the server. This might be a big performance win in some environments.
4468 * (Not useful until the client does caching on local stable storage.)
4471 nfscl_mustflush(vnode_t vp)
4473 struct nfsclclient *clp;
4474 struct nfscldeleg *dp;
4476 struct nfsmount *nmp;
4479 nmp = VFSTONFS(vp->v_mount);
4480 if (!NFSHASNFSV4(nmp))
4483 clp = nfscl_findcl(nmp);
4488 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4489 if (dp != NULL && (dp->nfsdl_flags &
4490 (NFSCLDL_WRITE | NFSCLDL_RECALL | NFSCLDL_DELEGRET)) ==
4492 (dp->nfsdl_sizelimit >= np->n_size ||
4493 !NFSHASSTRICT3530(nmp))) {
4502 * See if a (write) delegation exists for this file.
4505 nfscl_nodeleg(vnode_t vp, int writedeleg)
4507 struct nfsclclient *clp;
4508 struct nfscldeleg *dp;
4510 struct nfsmount *nmp;
4513 nmp = VFSTONFS(vp->v_mount);
4514 if (!NFSHASNFSV4(nmp))
4517 if ((nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0) {
4523 clp = nfscl_findcl(nmp);
4528 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4530 (dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_DELEGRET)) == 0 &&
4531 (writedeleg == 0 || (dp->nfsdl_flags & NFSCLDL_WRITE) ==
4541 * Look for an associated delegation that should be DelegReturned.
4544 nfscl_removedeleg(vnode_t vp, NFSPROC_T *p, nfsv4stateid_t *stp)
4546 struct nfsclclient *clp;
4547 struct nfscldeleg *dp;
4548 struct nfsclowner *owp;
4549 struct nfscllockowner *lp;
4550 struct nfsmount *nmp;
4553 int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
4555 nmp = VFSTONFS(vp->v_mount);
4559 * Loop around waiting for:
4560 * - outstanding I/O operations on delegations to complete
4561 * - for a delegation on vp that has state, lock the client and
4563 * - return delegation with no state
4566 clp = nfscl_findcl(nmp);
4571 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4572 np->n_fhp->nfh_len);
4575 * Wait for outstanding I/O ops to be done.
4577 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4579 nfsv4_unlock(&clp->nfsc_lock, 0);
4582 dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4583 (void) nfsmsleep(&dp->nfsdl_rwlock,
4584 NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4588 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4589 if (!LIST_EMPTY(&owp->nfsow_open)) {
4595 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4596 if (!LIST_EMPTY(&lp->nfsl_lock)) {
4602 if (needsrecall && !triedrecall) {
4603 dp->nfsdl_flags |= NFSCLDL_DELEGRET;
4606 igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
4607 &islept, NFSCLSTATEMUTEXPTR, NULL);
4614 cred = newnfs_getcred();
4615 newnfs_copycred(&dp->nfsdl_cred, cred);
4616 nfscl_recalldeleg(clp, nmp, dp, vp, cred, p, 0, NULL);
4620 nfsv4_unlock(&clp->nfsc_lock, 0);
4624 *stp = dp->nfsdl_stateid;
4626 nfscl_cleandeleg(dp);
4627 nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
4630 nfsv4_unlock(&clp->nfsc_lock, 0);
4637 * Look for associated delegation(s) that should be DelegReturned.
4640 nfscl_renamedeleg(vnode_t fvp, nfsv4stateid_t *fstp, int *gotfdp, vnode_t tvp,
4641 nfsv4stateid_t *tstp, int *gottdp, NFSPROC_T *p)
4643 struct nfsclclient *clp;
4644 struct nfscldeleg *dp;
4645 struct nfsclowner *owp;
4646 struct nfscllockowner *lp;
4647 struct nfsmount *nmp;
4650 int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
4652 nmp = VFSTONFS(fvp->v_mount);
4657 * Loop around waiting for:
4658 * - outstanding I/O operations on delegations to complete
4659 * - for a delegation on fvp that has state, lock the client and
4661 * - return delegation(s) with no state.
4664 clp = nfscl_findcl(nmp);
4670 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4671 np->n_fhp->nfh_len);
4672 if (dp != NULL && *gotfdp == 0) {
4674 * Wait for outstanding I/O ops to be done.
4676 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4678 nfsv4_unlock(&clp->nfsc_lock, 0);
4681 dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4682 (void) nfsmsleep(&dp->nfsdl_rwlock,
4683 NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4687 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4688 if (!LIST_EMPTY(&owp->nfsow_open)) {
4694 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4695 if (!LIST_EMPTY(&lp->nfsl_lock)) {
4701 if (needsrecall && !triedrecall) {
4702 dp->nfsdl_flags |= NFSCLDL_DELEGRET;
4705 igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
4706 &islept, NFSCLSTATEMUTEXPTR, NULL);
4713 cred = newnfs_getcred();
4714 newnfs_copycred(&dp->nfsdl_cred, cred);
4715 nfscl_recalldeleg(clp, nmp, dp, fvp, cred, p, 0, NULL);
4719 nfsv4_unlock(&clp->nfsc_lock, 0);
4723 *fstp = dp->nfsdl_stateid;
4726 nfscl_cleandeleg(dp);
4727 nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
4730 nfsv4_unlock(&clp->nfsc_lock, 0);
4735 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4736 np->n_fhp->nfh_len);
4737 if (dp != NULL && *gottdp == 0) {
4739 * Wait for outstanding I/O ops to be done.
4741 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4742 dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4743 (void) nfsmsleep(&dp->nfsdl_rwlock,
4744 NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4747 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4748 if (!LIST_EMPTY(&owp->nfsow_open)) {
4753 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4754 if (!LIST_EMPTY(&lp->nfsl_lock)) {
4759 *tstp = dp->nfsdl_stateid;
4762 nfscl_cleandeleg(dp);
4763 nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
4772 * Get a reference on the clientid associated with the mount point.
4773 * Return 1 if success, 0 otherwise.
4776 nfscl_getref(struct nfsmount *nmp)
4778 struct nfsclclient *clp;
4781 clp = nfscl_findcl(nmp);
4786 nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, NULL);
4792 * Release a reference on a clientid acquired with the above call.
4795 nfscl_relref(struct nfsmount *nmp)
4797 struct nfsclclient *clp;
4800 clp = nfscl_findcl(nmp);
4805 nfsv4_relref(&clp->nfsc_lock);
4810 * Save the size attribute in the delegation, since the nfsnode
4814 nfscl_reclaimnode(vnode_t vp)
4816 struct nfsclclient *clp;
4817 struct nfscldeleg *dp;
4818 struct nfsnode *np = VTONFS(vp);
4819 struct nfsmount *nmp;
4821 nmp = VFSTONFS(vp->v_mount);
4822 if (!NFSHASNFSV4(nmp))
4825 clp = nfscl_findcl(nmp);
4830 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4831 if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE))
4832 dp->nfsdl_size = np->n_size;
4837 * Get the saved size attribute in the delegation, since it is a
4838 * newly allocated nfsnode.
4841 nfscl_newnode(vnode_t vp)
4843 struct nfsclclient *clp;
4844 struct nfscldeleg *dp;
4845 struct nfsnode *np = VTONFS(vp);
4846 struct nfsmount *nmp;
4848 nmp = VFSTONFS(vp->v_mount);
4849 if (!NFSHASNFSV4(nmp))
4852 clp = nfscl_findcl(nmp);
4857 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4858 if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE))
4859 np->n_size = dp->nfsdl_size;
4864 * If there is a valid write delegation for this file, set the modtime
4865 * to the local clock time.
4868 nfscl_delegmodtime(vnode_t vp)
4870 struct nfsclclient *clp;
4871 struct nfscldeleg *dp;
4872 struct nfsnode *np = VTONFS(vp);
4873 struct nfsmount *nmp;
4875 nmp = VFSTONFS(vp->v_mount);
4876 if (!NFSHASNFSV4(nmp))
4879 if ((nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0) {
4885 clp = nfscl_findcl(nmp);
4890 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4891 if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE)) {
4892 nanotime(&dp->nfsdl_modtime);
4893 dp->nfsdl_flags |= NFSCLDL_MODTIMESET;
4899 * If there is a valid write delegation for this file with a modtime set,
4900 * put that modtime in mtime.
4903 nfscl_deleggetmodtime(vnode_t vp, struct timespec *mtime)
4905 struct nfsclclient *clp;
4906 struct nfscldeleg *dp;
4907 struct nfsnode *np = VTONFS(vp);
4908 struct nfsmount *nmp;
4910 nmp = VFSTONFS(vp->v_mount);
4911 if (!NFSHASNFSV4(nmp))
4914 if ((nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0) {
4920 clp = nfscl_findcl(nmp);
4925 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4927 (dp->nfsdl_flags & (NFSCLDL_WRITE | NFSCLDL_MODTIMESET)) ==
4928 (NFSCLDL_WRITE | NFSCLDL_MODTIMESET))
4929 *mtime = dp->nfsdl_modtime;
4934 nfscl_errmap(struct nfsrv_descript *nd, u_int32_t minorvers)
4936 short *defaulterrp, *errp;
4938 if (!nd->nd_repstat)
4940 if (nd->nd_procnum == NFSPROC_NOOP)
4941 return (txdr_unsigned(nd->nd_repstat & 0xffff));
4942 if (nd->nd_repstat == EBADRPC)
4943 return (txdr_unsigned(NFSERR_BADXDR));
4944 if (nd->nd_repstat == NFSERR_MINORVERMISMATCH ||
4945 nd->nd_repstat == NFSERR_OPILLEGAL)
4946 return (txdr_unsigned(nd->nd_repstat));
4947 if (nd->nd_repstat >= NFSERR_BADIOMODE && nd->nd_repstat < 20000 &&
4948 minorvers > NFSV4_MINORVERSION) {
4949 /* NFSv4.n error. */
4950 return (txdr_unsigned(nd->nd_repstat));
4952 if (nd->nd_procnum < NFSV4OP_CBNOPS)
4953 errp = defaulterrp = nfscl_cberrmap[nd->nd_procnum];
4955 return (txdr_unsigned(nd->nd_repstat));
4957 if (*errp == (short)nd->nd_repstat)
4958 return (txdr_unsigned(nd->nd_repstat));
4959 return (txdr_unsigned(*defaulterrp));
4963 * Called to find/add a layout to a client.
4964 * This function returns the layout with a refcnt (shared lock) upon
4965 * success (returns 0) or with no lock/refcnt on the layout when an
4966 * error is returned.
4967 * If a layout is passed in via lypp, it is locked (exclusively locked).
4970 nfscl_layout(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen,
4971 nfsv4stateid_t *stateidp, int layouttype, int retonclose,
4972 struct nfsclflayouthead *fhlp, struct nfscllayout **lypp,
4973 struct ucred *cred, NFSPROC_T *p)
4975 struct nfsclclient *clp;
4976 struct nfscllayout *lyp, *tlyp;
4977 struct nfsclflayout *flp;
4978 struct nfsnode *np = VTONFS(vp);
4980 int layout_passed_in;
4982 mp = nmp->nm_mountp;
4983 layout_passed_in = 1;
4987 layout_passed_in = 0;
4988 tlyp = malloc(sizeof(*tlyp) + fhlen - 1, M_NFSLAYOUT,
4995 if (layout_passed_in != 0)
4996 nfsv4_unlock(&lyp->nfsly_lock, 0);
4999 free(tlyp, M_NFSLAYOUT);
5004 * Although no lyp was passed in, another thread might have
5005 * allocated one. If one is found, just increment it's ref
5006 * count and return it.
5008 lyp = nfscl_findlayout(clp, fhp, fhlen);
5012 lyp->nfsly_stateid.seqid = stateidp->seqid;
5013 lyp->nfsly_stateid.other[0] = stateidp->other[0];
5014 lyp->nfsly_stateid.other[1] = stateidp->other[1];
5015 lyp->nfsly_stateid.other[2] = stateidp->other[2];
5016 lyp->nfsly_lastbyte = 0;
5017 LIST_INIT(&lyp->nfsly_flayread);
5018 LIST_INIT(&lyp->nfsly_flayrw);
5019 LIST_INIT(&lyp->nfsly_recall);
5020 lyp->nfsly_filesid[0] = np->n_vattr.na_filesid[0];
5021 lyp->nfsly_filesid[1] = np->n_vattr.na_filesid[1];
5022 lyp->nfsly_clp = clp;
5023 if (layouttype == NFSLAYOUT_FLEXFILE)
5024 lyp->nfsly_flags = NFSLY_FLEXFILE;
5026 lyp->nfsly_flags = NFSLY_FILES;
5027 if (retonclose != 0)
5028 lyp->nfsly_flags |= NFSLY_RETONCLOSE;
5029 lyp->nfsly_fhlen = fhlen;
5030 NFSBCOPY(fhp, lyp->nfsly_fh, fhlen);
5031 TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
5032 LIST_INSERT_HEAD(NFSCLLAYOUTHASH(clp, fhp, fhlen), lyp,
5034 lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
5037 if (retonclose != 0)
5038 lyp->nfsly_flags |= NFSLY_RETONCLOSE;
5039 if (stateidp->seqid > lyp->nfsly_stateid.seqid)
5040 lyp->nfsly_stateid.seqid = stateidp->seqid;
5041 TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list);
5042 TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
5043 lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
5045 nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
5046 if (NFSCL_FORCEDISM(mp)) {
5049 free(tlyp, M_NFSLAYOUT);
5053 } else if (stateidp->seqid > lyp->nfsly_stateid.seqid)
5054 lyp->nfsly_stateid.seqid = stateidp->seqid;
5056 /* Merge the new list of File Layouts into the list. */
5057 flp = LIST_FIRST(fhlp);
5059 if (flp->nfsfl_iomode == NFSLAYOUTIOMODE_READ)
5060 nfscl_mergeflayouts(&lyp->nfsly_flayread, fhlp);
5062 nfscl_mergeflayouts(&lyp->nfsly_flayrw, fhlp);
5064 if (layout_passed_in != 0)
5065 nfsv4_unlock(&lyp->nfsly_lock, 1);
5068 free(tlyp, M_NFSLAYOUT);
5073 * Search for a layout by MDS file handle.
5074 * If one is found, it is returned with a refcnt (shared lock) iff
5075 * retflpp returned non-NULL and locked (exclusive locked) iff retflpp is
5078 struct nfscllayout *
5079 nfscl_getlayout(struct nfsclclient *clp, uint8_t *fhp, int fhlen,
5080 uint64_t off, struct nfsclflayout **retflpp, int *recalledp)
5082 struct nfscllayout *lyp;
5084 int error, igotlock;
5086 mp = clp->nfsc_nmp->nm_mountp;
5090 lyp = nfscl_findlayout(clp, fhp, fhlen);
5092 if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
5093 TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list);
5094 TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
5095 lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
5096 error = nfscl_findlayoutforio(lyp, off,
5097 NFSV4OPEN_ACCESSREAD, retflpp);
5099 nfsv4_getref(&lyp->nfsly_lock, NULL,
5100 NFSCLSTATEMUTEXPTR, mp);
5103 igotlock = nfsv4_lock(&lyp->nfsly_lock,
5104 1, NULL, NFSCLSTATEMUTEXPTR, mp);
5105 } while (igotlock == 0 && !NFSCL_FORCEDISM(mp));
5108 if (NFSCL_FORCEDISM(mp)) {
5122 * Search for a layout by MDS file handle. If one is found, mark in to be
5123 * recalled, if it already marked "return on close".
5126 nfscl_retoncloselayout(vnode_t vp, struct nfsclclient *clp, uint8_t *fhp,
5127 int fhlen, struct nfsclrecalllayout **recallpp)
5129 struct nfscllayout *lyp;
5132 if (vp->v_type != VREG || !NFSHASPNFS(VFSTONFS(vp->v_mount)) ||
5133 nfscl_enablecallb == 0 || nfs_numnfscbd == 0 ||
5134 (VTONFS(vp)->n_flag & NNOLAYOUT) != 0)
5136 lyp = nfscl_findlayout(clp, fhp, fhlen);
5137 if (lyp != NULL && (lyp->nfsly_flags & (NFSLY_RETONCLOSE |
5138 NFSLY_RECALL)) == NFSLY_RETONCLOSE) {
5140 if (!LIST_EMPTY(&lyp->nfsly_flayread))
5141 iomode |= NFSLAYOUTIOMODE_READ;
5142 if (!LIST_EMPTY(&lyp->nfsly_flayrw))
5143 iomode |= NFSLAYOUTIOMODE_RW;
5144 (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE, lyp, iomode,
5145 0, UINT64_MAX, lyp->nfsly_stateid.seqid, 0, 0, NULL,
5147 NFSCL_DEBUG(4, "retoncls recall iomode=%d\n", iomode);
5153 * Mark the layout to be recalled and with an error.
5154 * Also, disable the dsp from further use.
5157 nfscl_dserr(uint32_t op, uint32_t stat, struct nfscldevinfo *dp,
5158 struct nfscllayout *lyp, struct nfsclds *dsp)
5160 struct nfsclrecalllayout *recallp;
5163 printf("DS being disabled, error=%d\n", stat);
5164 /* Set up the return of the layout. */
5165 recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_WAITOK);
5168 if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
5169 if (!LIST_EMPTY(&lyp->nfsly_flayread))
5170 iomode |= NFSLAYOUTIOMODE_READ;
5171 if (!LIST_EMPTY(&lyp->nfsly_flayrw))
5172 iomode |= NFSLAYOUTIOMODE_RW;
5173 (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE, lyp, iomode,
5174 0, UINT64_MAX, lyp->nfsly_stateid.seqid, stat, op,
5175 dp->nfsdi_deviceid, recallp);
5177 NFSCL_DEBUG(4, "nfscl_dserr recall iomode=%d\n", iomode);
5180 free(recallp, M_NFSLAYRECALL);
5183 /* And shut the TCP connection down. */
5184 nfscl_cancelreqs(dsp);
5188 * Cancel all RPCs for this "dsp" by closing the connection.
5189 * Also, mark the session as defunct.
5190 * If NFSCLDS_SAMECONN is set, the connection is shared with other DSs and
5191 * cannot be shut down.
5194 nfscl_cancelreqs(struct nfsclds *dsp)
5196 struct __rpc_client *cl;
5197 static int non_event;
5200 if ((dsp->nfsclds_flags & (NFSCLDS_CLOSED | NFSCLDS_SAMECONN)) == 0 &&
5201 dsp->nfsclds_sockp != NULL &&
5202 dsp->nfsclds_sockp->nr_client != NULL) {
5203 dsp->nfsclds_flags |= NFSCLDS_CLOSED;
5204 cl = dsp->nfsclds_sockp->nr_client;
5205 dsp->nfsclds_sess.nfsess_defunct = 1;
5209 * This 1sec sleep is done to reduce the number of reconnect
5210 * attempts made on the DS while it has failed.
5212 tsleep(&non_event, PVFS, "ndscls", hz);
5219 * Dereference a layout.
5222 nfscl_rellayout(struct nfscllayout *lyp, int exclocked)
5227 nfsv4_unlock(&lyp->nfsly_lock, 0);
5229 nfsv4_relref(&lyp->nfsly_lock);
5234 * Search for a devinfo by deviceid. If one is found, return it after
5235 * acquiring a reference count on it.
5237 struct nfscldevinfo *
5238 nfscl_getdevinfo(struct nfsclclient *clp, uint8_t *deviceid,
5239 struct nfscldevinfo *dip)
5244 dip = nfscl_finddevinfo(clp, deviceid);
5246 dip->nfsdi_refcnt++;
5252 * Dereference a devinfo structure.
5255 nfscl_reldevinfo_locked(struct nfscldevinfo *dip)
5258 dip->nfsdi_refcnt--;
5259 if (dip->nfsdi_refcnt == 0)
5260 wakeup(&dip->nfsdi_refcnt);
5264 * Dereference a devinfo structure.
5267 nfscl_reldevinfo(struct nfscldevinfo *dip)
5271 nfscl_reldevinfo_locked(dip);
5276 * Find a layout for this file handle. Return NULL upon failure.
5278 static struct nfscllayout *
5279 nfscl_findlayout(struct nfsclclient *clp, u_int8_t *fhp, int fhlen)
5281 struct nfscllayout *lyp;
5283 LIST_FOREACH(lyp, NFSCLLAYOUTHASH(clp, fhp, fhlen), nfsly_hash)
5284 if (lyp->nfsly_fhlen == fhlen &&
5285 !NFSBCMP(lyp->nfsly_fh, fhp, fhlen))
5291 * Find a devinfo for this deviceid. Return NULL upon failure.
5293 static struct nfscldevinfo *
5294 nfscl_finddevinfo(struct nfsclclient *clp, uint8_t *deviceid)
5296 struct nfscldevinfo *dip;
5298 LIST_FOREACH(dip, &clp->nfsc_devinfo, nfsdi_list)
5299 if (NFSBCMP(dip->nfsdi_deviceid, deviceid, NFSX_V4DEVICEID)
5306 * Merge the new file layout list into the main one, maintaining it in
5307 * increasing offset order.
5310 nfscl_mergeflayouts(struct nfsclflayouthead *fhlp,
5311 struct nfsclflayouthead *newfhlp)
5313 struct nfsclflayout *flp, *nflp, *prevflp, *tflp;
5315 flp = LIST_FIRST(fhlp);
5317 LIST_FOREACH_SAFE(nflp, newfhlp, nfsfl_list, tflp) {
5318 while (flp != NULL && flp->nfsfl_off < nflp->nfsfl_off) {
5320 flp = LIST_NEXT(flp, nfsfl_list);
5322 if (prevflp == NULL)
5323 LIST_INSERT_HEAD(fhlp, nflp, nfsfl_list);
5325 LIST_INSERT_AFTER(prevflp, nflp, nfsfl_list);
5331 * Add this nfscldevinfo to the client, if it doesn't already exist.
5332 * This function consumes the structure pointed at by dip, if not NULL.
5335 nfscl_adddevinfo(struct nfsmount *nmp, struct nfscldevinfo *dip, int ind,
5336 struct nfsclflayout *flp)
5338 struct nfsclclient *clp;
5339 struct nfscldevinfo *tdip;
5347 free(dip, M_NFSDEVINFO);
5350 if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5351 dev = flp->nfsfl_dev;
5353 dev = flp->nfsfl_ffm[ind].dev;
5354 tdip = nfscl_finddevinfo(clp, dev);
5356 tdip->nfsdi_layoutrefs++;
5357 if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5358 flp->nfsfl_devp = tdip;
5360 flp->nfsfl_ffm[ind].devp = tdip;
5361 nfscl_reldevinfo_locked(tdip);
5364 free(dip, M_NFSDEVINFO);
5368 LIST_INSERT_HEAD(&clp->nfsc_devinfo, dip, nfsdi_list);
5369 dip->nfsdi_layoutrefs = 1;
5370 if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5371 flp->nfsfl_devp = dip;
5373 flp->nfsfl_ffm[ind].devp = dip;
5382 * Free up a layout structure and associated file layout structure(s).
5385 nfscl_freelayout(struct nfscllayout *layp)
5387 struct nfsclflayout *flp, *nflp;
5388 struct nfsclrecalllayout *rp, *nrp;
5390 LIST_FOREACH_SAFE(flp, &layp->nfsly_flayread, nfsfl_list, nflp) {
5391 LIST_REMOVE(flp, nfsfl_list);
5392 nfscl_freeflayout(flp);
5394 LIST_FOREACH_SAFE(flp, &layp->nfsly_flayrw, nfsfl_list, nflp) {
5395 LIST_REMOVE(flp, nfsfl_list);
5396 nfscl_freeflayout(flp);
5398 LIST_FOREACH_SAFE(rp, &layp->nfsly_recall, nfsrecly_list, nrp) {
5399 LIST_REMOVE(rp, nfsrecly_list);
5400 free(rp, M_NFSLAYRECALL);
5403 free(layp, M_NFSLAYOUT);
5407 * Free up a file layout structure.
5410 nfscl_freeflayout(struct nfsclflayout *flp)
5414 if ((flp->nfsfl_flags & NFSFL_FILE) != 0) {
5415 for (i = 0; i < flp->nfsfl_fhcnt; i++)
5416 free(flp->nfsfl_fh[i], M_NFSFH);
5417 if (flp->nfsfl_devp != NULL)
5418 flp->nfsfl_devp->nfsdi_layoutrefs--;
5420 if ((flp->nfsfl_flags & NFSFL_FLEXFILE) != 0)
5421 for (i = 0; i < flp->nfsfl_mirrorcnt; i++) {
5422 for (j = 0; j < flp->nfsfl_ffm[i].fhcnt; j++)
5423 free(flp->nfsfl_ffm[i].fh[j], M_NFSFH);
5424 if (flp->nfsfl_ffm[i].devp != NULL)
5425 flp->nfsfl_ffm[i].devp->nfsdi_layoutrefs--;
5427 free(flp, M_NFSFLAYOUT);
5431 * Free up a file layout devinfo structure.
5434 nfscl_freedevinfo(struct nfscldevinfo *dip)
5437 free(dip, M_NFSDEVINFO);
5441 * Mark any layouts that match as recalled.
5444 nfscl_layoutrecall(int recalltype, struct nfscllayout *lyp, uint32_t iomode,
5445 uint64_t off, uint64_t len, uint32_t stateseqid, uint32_t stat, uint32_t op,
5446 char *devid, struct nfsclrecalllayout *recallp)
5448 struct nfsclrecalllayout *rp, *orp;
5450 recallp->nfsrecly_recalltype = recalltype;
5451 recallp->nfsrecly_iomode = iomode;
5452 recallp->nfsrecly_stateseqid = stateseqid;
5453 recallp->nfsrecly_off = off;
5454 recallp->nfsrecly_len = len;
5455 recallp->nfsrecly_stat = stat;
5456 recallp->nfsrecly_op = op;
5458 NFSBCOPY(devid, recallp->nfsrecly_devid, NFSX_V4DEVICEID);
5460 * Order the list as file returns first, followed by fsid and any
5461 * returns, both in increasing stateseqid order.
5462 * Note that the seqids wrap around, so 1 is after 0xffffffff.
5463 * (I'm not sure this is correct because I find RFC5661 confusing
5464 * on this, but hopefully it will work ok.)
5467 LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) {
5469 if ((recalltype == NFSLAYOUTRETURN_FILE &&
5470 (rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE ||
5471 nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) ||
5472 (recalltype != NFSLAYOUTRETURN_FILE &&
5473 rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE &&
5474 nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) {
5475 LIST_INSERT_BEFORE(rp, recallp, nfsrecly_list);
5480 * Put any error return on all the file returns that will
5483 if (rp->nfsrecly_recalltype == NFSLAYOUTRETURN_FILE &&
5484 stat != 0 && rp->nfsrecly_stat == 0) {
5485 rp->nfsrecly_stat = stat;
5486 rp->nfsrecly_op = op;
5488 NFSBCOPY(devid, rp->nfsrecly_devid,
5494 LIST_INSERT_HEAD(&lyp->nfsly_recall, recallp,
5497 LIST_INSERT_AFTER(orp, recallp, nfsrecly_list);
5499 lyp->nfsly_flags |= NFSLY_RECALL;
5500 wakeup(lyp->nfsly_clp);
5505 * Compare the two seqids for ordering. The trick is that the seqids can
5506 * wrap around from 0xffffffff->0, so check for the cases where one
5507 * has wrapped around.
5508 * Return 1 if seqid1 comes before seqid2, 0 otherwise.
5511 nfscl_seq(uint32_t seqid1, uint32_t seqid2)
5514 if (seqid2 > seqid1 && (seqid2 - seqid1) >= 0x7fffffff)
5515 /* seqid2 has wrapped around. */
5517 if (seqid1 > seqid2 && (seqid1 - seqid2) >= 0x7fffffff)
5518 /* seqid1 has wrapped around. */
5520 if (seqid1 <= seqid2)
5526 * Do a layout return for each of the recalls.
5529 nfscl_layoutreturn(struct nfsmount *nmp, struct nfscllayout *lyp,
5530 struct ucred *cred, NFSPROC_T *p)
5532 struct nfsclrecalllayout *rp;
5533 nfsv4stateid_t stateid;
5536 NFSBCOPY(lyp->nfsly_stateid.other, stateid.other, NFSX_STATEIDOTHER);
5537 stateid.seqid = lyp->nfsly_stateid.seqid;
5538 if ((lyp->nfsly_flags & NFSLY_FILES) != 0)
5539 layouttype = NFSLAYOUT_NFSV4_1_FILES;
5541 layouttype = NFSLAYOUT_FLEXFILE;
5542 LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) {
5543 (void)nfsrpc_layoutreturn(nmp, lyp->nfsly_fh,
5544 lyp->nfsly_fhlen, 0, layouttype,
5545 rp->nfsrecly_iomode, rp->nfsrecly_recalltype,
5546 rp->nfsrecly_off, rp->nfsrecly_len,
5547 &stateid, cred, p, rp->nfsrecly_stat, rp->nfsrecly_op,
5548 rp->nfsrecly_devid);
5553 * Do the layout commit for a file layout.
5556 nfscl_dolayoutcommit(struct nfsmount *nmp, struct nfscllayout *lyp,
5557 struct ucred *cred, NFSPROC_T *p)
5559 struct nfsclflayout *flp;
5561 int error, layouttype;
5563 if ((lyp->nfsly_flags & NFSLY_FILES) != 0)
5564 layouttype = NFSLAYOUT_NFSV4_1_FILES;
5566 layouttype = NFSLAYOUT_FLEXFILE;
5567 LIST_FOREACH(flp, &lyp->nfsly_flayrw, nfsfl_list) {
5568 if (layouttype == NFSLAYOUT_FLEXFILE &&
5569 (flp->nfsfl_fflags & NFSFLEXFLAG_NO_LAYOUTCOMMIT) != 0) {
5570 NFSCL_DEBUG(4, "Flex file: no layoutcommit\n");
5571 /* If not supported, don't bother doing it. */
5573 nmp->nm_state |= NFSSTA_NOLAYOUTCOMMIT;
5576 } else if (flp->nfsfl_off <= lyp->nfsly_lastbyte) {
5577 len = flp->nfsfl_end - flp->nfsfl_off;
5578 error = nfsrpc_layoutcommit(nmp, lyp->nfsly_fh,
5579 lyp->nfsly_fhlen, 0, flp->nfsfl_off, len,
5580 lyp->nfsly_lastbyte, &lyp->nfsly_stateid,
5581 layouttype, cred, p, NULL);
5582 NFSCL_DEBUG(4, "layoutcommit err=%d\n", error);
5583 if (error == NFSERR_NOTSUPP) {
5584 /* If not supported, don't bother doing it. */
5586 nmp->nm_state |= NFSSTA_NOLAYOUTCOMMIT;
5595 * Commit all layouts for a file (vnode).
5598 nfscl_layoutcommit(vnode_t vp, NFSPROC_T *p)
5600 struct nfsclclient *clp;
5601 struct nfscllayout *lyp;
5602 struct nfsnode *np = VTONFS(vp);
5604 struct nfsmount *nmp;
5608 if (NFSHASNOLAYOUTCOMMIT(nmp))
5616 lyp = nfscl_findlayout(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
5621 nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
5622 if (NFSCL_FORCEDISM(mp)) {
5627 if ((lyp->nfsly_flags & NFSLY_WRITTEN) != 0) {
5628 lyp->nfsly_flags &= ~NFSLY_WRITTEN;
5630 NFSCL_DEBUG(4, "do layoutcommit2\n");
5631 nfscl_dolayoutcommit(clp->nfsc_nmp, lyp, NFSPROCCRED(p), p);
5635 nfsv4_relref(&lyp->nfsly_lock);