2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2009 Rick Macklem, University of Guelph
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
34 * These functions implement the client side state handling for NFSv4.
35 * NFSv4 state handling:
36 * - A lockowner is used to determine lock contention, so it
37 * corresponds directly to a Posix pid. (1 to 1 mapping)
38 * - The correct granularity of an OpenOwner is not nearly so
39 * obvious. An OpenOwner does the following:
40 * - provides a serial sequencing of Open/Close/Lock-with-new-lockowner
41 * - is used to check for Open/Share contention (not applicable to
42 * this client, since all Opens are Deny_None)
43 * As such, I considered both extreme.
44 * 1 OpenOwner per ClientID - Simple to manage, but fully serializes
45 * all Open, Close and Lock (with a new lockowner) Ops.
46 * 1 OpenOwner for each Open - This one results in an OpenConfirm for
47 * every Open, for most servers.
48 * So, I chose to use the same mapping as I did for LockOwnwers.
49 * The main concern here is that you can end up with multiple Opens
50 * for the same File Handle, but on different OpenOwners (opens
51 * inherited from parents, grandparents...) and you do not know
52 * which of these the vnodeop close applies to. This is handled by
53 * delaying the Close Op(s) until all of the Opens have been closed.
54 * (It is not yet obvious if this is the correct granularity.)
55 * - How the code handles serialization:
56 * - For the ClientId, it uses an exclusive lock while getting its
57 * SetClientId and during recovery. Otherwise, it uses a shared
58 * lock via a reference count.
59 * - For the rest of the data structures, it uses an SMP mutex
60 * (once the nfs client is SMP safe) and doesn't sleep while
61 * manipulating the linked lists.
62 * - The serialization of Open/Close/Lock/LockU falls out in the
63 * "wash", since OpenOwners and LockOwners are both mapped from
64 * Posix pid. In other words, there is only one Posix pid using
65 * any given owner, so that owner is serialized. (If you change
66 * the granularity of the OpenOwner, then code must be added to
67 * serialize Ops on the OpenOwner.)
68 * - When to get rid of OpenOwners and LockOwners.
69 * - The function nfscl_cleanup_common() is executed after a process exits.
70 * It goes through the client list looking for all Open and Lock Owners.
71 * When one is found, it is marked "defunct" or in the case of
72 * an OpenOwner without any Opens, freed.
73 * The renew thread scans for defunct Owners and gets rid of them,
74 * if it can. The LockOwners will also be deleted when the
75 * associated Open is closed.
76 * - If the LockU or Close Op(s) fail during close in a way
77 * that could be recovered upon retry, they are relinked to the
78 * ClientId's defunct open list and retried by the renew thread
79 * until they succeed or an unmount/recovery occurs.
80 * (Since we are done with them, they do not need to be recovered.)
83 #include <fs/nfs/nfsport.h>
88 extern struct nfsstatsv1 nfsstatsv1;
89 extern struct nfsreqhead nfsd_reqq;
90 extern u_int32_t newnfs_false, newnfs_true;
91 extern int nfscl_debuglevel;
92 extern int nfscl_enablecallb;
93 extern int nfs_numnfscbd;
97 struct nfsclhead nfsclhead; /* Head of clientid list */
98 int nfscl_deleghighwater = NFSCLDELEGHIGHWATER;
99 int nfscl_layouthighwater = NFSCLLAYOUTHIGHWATER;
101 static int nfscl_delegcnt = 0;
102 static int nfscl_layoutcnt = 0;
103 static int nfscl_getopen(struct nfsclownerhead *, u_int8_t *, int, u_int8_t *,
104 u_int8_t *, u_int32_t, struct nfscllockowner **, struct nfsclopen **);
105 static void nfscl_clrelease(struct nfsclclient *);
106 static void nfscl_cleanclient(struct nfsclclient *);
107 static void nfscl_expireclient(struct nfsclclient *, struct nfsmount *,
108 struct ucred *, NFSPROC_T *);
109 static int nfscl_expireopen(struct nfsclclient *, struct nfsclopen *,
110 struct nfsmount *, struct ucred *, NFSPROC_T *);
111 static void nfscl_recover(struct nfsclclient *, struct ucred *, NFSPROC_T *);
112 static void nfscl_insertlock(struct nfscllockowner *, struct nfscllock *,
113 struct nfscllock *, int);
114 static int nfscl_updatelock(struct nfscllockowner *, struct nfscllock **,
115 struct nfscllock **, int);
116 static void nfscl_delegreturnall(struct nfsclclient *, NFSPROC_T *);
117 static u_int32_t nfscl_nextcbident(void);
118 static mount_t nfscl_getmnt(int, uint8_t *, u_int32_t, struct nfsclclient **);
119 static struct nfsclclient *nfscl_getclnt(u_int32_t);
120 static struct nfsclclient *nfscl_getclntsess(uint8_t *);
121 static struct nfscldeleg *nfscl_finddeleg(struct nfsclclient *, u_int8_t *,
123 static void nfscl_retoncloselayout(vnode_t, struct nfsclclient *, uint8_t *,
124 int, struct nfsclrecalllayout **);
125 static void nfscl_reldevinfo_locked(struct nfscldevinfo *);
126 static struct nfscllayout *nfscl_findlayout(struct nfsclclient *, u_int8_t *,
128 static struct nfscldevinfo *nfscl_finddevinfo(struct nfsclclient *, uint8_t *);
129 static int nfscl_checkconflict(struct nfscllockownerhead *, struct nfscllock *,
130 u_int8_t *, struct nfscllock **);
131 static void nfscl_freealllocks(struct nfscllockownerhead *, int);
132 static int nfscl_localconflict(struct nfsclclient *, u_int8_t *, int,
133 struct nfscllock *, u_int8_t *, struct nfscldeleg *, struct nfscllock **);
134 static void nfscl_newopen(struct nfsclclient *, struct nfscldeleg *,
135 struct nfsclowner **, struct nfsclowner **, struct nfsclopen **,
136 struct nfsclopen **, u_int8_t *, u_int8_t *, int, struct ucred *, int *);
137 static int nfscl_moveopen(vnode_t , struct nfsclclient *,
138 struct nfsmount *, struct nfsclopen *, struct nfsclowner *,
139 struct nfscldeleg *, struct ucred *, NFSPROC_T *);
140 static void nfscl_totalrecall(struct nfsclclient *);
141 static int nfscl_relock(vnode_t , struct nfsclclient *, struct nfsmount *,
142 struct nfscllockowner *, struct nfscllock *, struct ucred *, NFSPROC_T *);
143 static int nfscl_tryopen(struct nfsmount *, vnode_t , u_int8_t *, int,
144 u_int8_t *, int, u_int32_t, struct nfsclopen *, u_int8_t *, int,
145 struct nfscldeleg **, int, u_int32_t, struct ucred *, NFSPROC_T *);
146 static int nfscl_trylock(struct nfsmount *, vnode_t , u_int8_t *,
147 int, struct nfscllockowner *, int, int, u_int64_t, u_int64_t, short,
148 struct ucred *, NFSPROC_T *);
149 static int nfsrpc_reopen(struct nfsmount *, u_int8_t *, int, u_int32_t,
150 struct nfsclopen *, struct nfscldeleg **, struct ucred *, NFSPROC_T *);
151 static void nfscl_freedeleg(struct nfscldeleghead *, struct nfscldeleg *,
153 static int nfscl_errmap(struct nfsrv_descript *, u_int32_t);
154 static void nfscl_cleanup_common(struct nfsclclient *, u_int8_t *);
155 static int nfscl_recalldeleg(struct nfsclclient *, struct nfsmount *,
156 struct nfscldeleg *, vnode_t, struct ucred *, NFSPROC_T *, int,
158 static void nfscl_freeopenowner(struct nfsclowner *, int);
159 static void nfscl_cleandeleg(struct nfscldeleg *);
160 static int nfscl_trydelegreturn(struct nfscldeleg *, struct ucred *,
161 struct nfsmount *, NFSPROC_T *);
162 static void nfscl_emptylockowner(struct nfscllockowner *,
163 struct nfscllockownerfhhead *);
164 static void nfscl_mergeflayouts(struct nfsclflayouthead *,
165 struct nfsclflayouthead *);
166 static int nfscl_layoutrecall(int, struct nfscllayout *, uint32_t, uint64_t,
167 uint64_t, uint32_t, uint32_t, uint32_t, char *, struct nfsclrecalllayout *);
168 static int nfscl_seq(uint32_t, uint32_t);
169 static void nfscl_layoutreturn(struct nfsmount *, struct nfscllayout *,
170 struct ucred *, NFSPROC_T *);
171 static void nfscl_dolayoutcommit(struct nfsmount *, struct nfscllayout *,
172 struct ucred *, NFSPROC_T *);
174 static short nfscberr_null[] = {
179 static short nfscberr_getattr[] = {
188 static short nfscberr_recall[] = {
198 static short *nfscl_cberrmap[] = {
206 #define NETFAMILY(clp) \
207 (((clp)->nfsc_flags & NFSCLFLAGS_AFINET6) ? AF_INET6 : AF_INET)
210 * Called for an open operation.
211 * If the nfhp argument is NULL, just get an openowner.
214 nfscl_open(vnode_t vp, u_int8_t *nfhp, int fhlen, u_int32_t amode, int usedeleg,
215 struct ucred *cred, NFSPROC_T *p, struct nfsclowner **owpp,
216 struct nfsclopen **opp, int *newonep, int *retp, int lockit, bool firstref)
218 struct nfsclclient *clp;
219 struct nfsclowner *owp, *nowp;
220 struct nfsclopen *op = NULL, *nop = NULL;
221 struct nfscldeleg *dp;
222 struct nfsclownerhead *ohp;
223 u_int8_t own[NFSV4CL_LOCKNAMELEN];
234 * Might need one or both of these, so MALLOC them now, to
235 * avoid a tsleep() in MALLOC later.
237 nowp = malloc(sizeof (struct nfsclowner),
238 M_NFSCLOWNER, M_WAITOK);
240 nop = malloc(sizeof (struct nfsclopen) +
241 fhlen - 1, M_NFSCLOPEN, M_WAITOK);
242 ret = nfscl_getcl(vnode_mount(vp), cred, p, 1, firstref, &clp);
244 free(nowp, M_NFSCLOWNER);
246 free(nop, M_NFSCLOPEN);
251 * Get the Open iff it already exists.
252 * If none found, add the new one or return error, depending upon
257 /* First check the delegation list */
258 if (nfhp != NULL && usedeleg) {
259 LIST_FOREACH(dp, NFSCLDELEGHASH(clp, nfhp, fhlen), nfsdl_hash) {
260 if (dp->nfsdl_fhlen == fhlen &&
261 !NFSBCMP(nfhp, dp->nfsdl_fh, fhlen)) {
262 if (!(amode & NFSV4OPEN_ACCESSWRITE) ||
263 (dp->nfsdl_flags & NFSCLDL_WRITE))
272 nfscl_filllockowner(p->td_proc, own, F_POSIX);
273 ohp = &dp->nfsdl_owner;
275 /* For NFSv4.1 and this option, use a single open_owner. */
276 if (NFSHASONEOPENOWN(VFSTONFS(vnode_mount(vp))))
277 nfscl_filllockowner(NULL, own, F_POSIX);
279 nfscl_filllockowner(p->td_proc, own, F_POSIX);
280 ohp = &clp->nfsc_owner;
282 /* Now, search for an openowner */
283 LIST_FOREACH(owp, ohp, nfsow_list) {
284 if (!NFSBCMP(owp->nfsow_owner, own, NFSV4CL_LOCKNAMELEN))
289 * Create a new open, as required.
291 nfscl_newopen(clp, dp, &owp, &nowp, &op, &nop, own, nfhp, fhlen,
295 * Now, check the mode on the open and return the appropriate
299 if (nfhp != NULL && dp != NULL && nop == NULL)
300 /* new local open on delegation */
301 *retp = NFSCLOPEN_SETCRED;
303 *retp = NFSCLOPEN_OK;
305 if (op != NULL && (amode & ~(op->nfso_mode))) {
306 op->nfso_mode |= amode;
307 if (retp != NULL && dp == NULL)
308 *retp = NFSCLOPEN_DOOPEN;
312 * Serialize modifications to the open owner for multiple threads
313 * within the same process using a read/write sleep lock.
314 * For NFSv4.1 and a single OpenOwner, allow concurrent open operations
315 * by acquiring a shared lock. The close operations still use an
316 * exclusive lock for this case.
319 if (NFSHASONEOPENOWN(VFSTONFS(vnode_mount(vp)))) {
321 * Get a shared lock on the OpenOwner, but first
322 * wait for any pending exclusive lock, so that the
323 * exclusive locker gets priority.
325 nfsv4_lock(&owp->nfsow_rwlock, 0, NULL,
326 NFSCLSTATEMUTEXPTR, NULL);
327 nfsv4_getref(&owp->nfsow_rwlock, NULL,
328 NFSCLSTATEMUTEXPTR, NULL);
330 nfscl_lockexcl(&owp->nfsow_rwlock, NFSCLSTATEMUTEXPTR);
334 free(nowp, M_NFSCLOWNER);
336 free(nop, M_NFSCLOPEN);
345 * Create a new open, as required.
348 nfscl_newopen(struct nfsclclient *clp, struct nfscldeleg *dp,
349 struct nfsclowner **owpp, struct nfsclowner **nowpp, struct nfsclopen **opp,
350 struct nfsclopen **nopp, u_int8_t *own, u_int8_t *fhp, int fhlen,
351 struct ucred *cred, int *newonep)
353 struct nfsclowner *owp = *owpp, *nowp;
354 struct nfsclopen *op, *nop;
364 if (owp == NULL && nowp != NULL) {
365 NFSBCOPY(own, nowp->nfsow_owner, NFSV4CL_LOCKNAMELEN);
366 LIST_INIT(&nowp->nfsow_open);
367 nowp->nfsow_clp = clp;
368 nowp->nfsow_seqid = 0;
369 nowp->nfsow_defunct = 0;
370 nfscl_lockinit(&nowp->nfsow_rwlock);
372 nfsstatsv1.cllocalopenowners++;
373 LIST_INSERT_HEAD(&dp->nfsdl_owner, nowp, nfsow_list);
375 nfsstatsv1.clopenowners++;
376 LIST_INSERT_HEAD(&clp->nfsc_owner, nowp, nfsow_list);
384 /* If an fhp has been specified, create an Open as well. */
386 /* and look for the correct open, based upon FH */
387 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
388 if (op->nfso_fhlen == fhlen &&
389 !NFSBCMP(op->nfso_fh, fhp, fhlen))
392 if (op == NULL && nop != NULL) {
395 nop->nfso_opencnt = 0;
396 nop->nfso_posixlock = 1;
397 nop->nfso_fhlen = fhlen;
398 NFSBCOPY(fhp, nop->nfso_fh, fhlen);
399 LIST_INIT(&nop->nfso_lock);
400 nop->nfso_stateid.seqid = 0;
401 nop->nfso_stateid.other[0] = 0;
402 nop->nfso_stateid.other[1] = 0;
403 nop->nfso_stateid.other[2] = 0;
404 KASSERT(cred != NULL, ("%s: cred NULL\n", __func__));
405 newnfs_copyincred(cred, &nop->nfso_cred);
407 TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
408 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp,
410 dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
411 nfsstatsv1.cllocalopens++;
413 nfsstatsv1.clopens++;
415 LIST_INSERT_HEAD(&owp->nfsow_open, nop, nfso_list);
427 * Called to find/add a delegation to a client.
430 nfscl_deleg(mount_t mp, struct nfsclclient *clp, u_int8_t *nfhp,
431 int fhlen, struct ucred *cred, NFSPROC_T *p, struct nfscldeleg **dpp)
433 struct nfscldeleg *dp = *dpp, *tdp;
436 * First, if we have received a Read delegation for a file on a
437 * read/write file system, just return it, because they aren't
440 if (mp != NULL && dp != NULL && !NFSMNT_RDONLY(mp) &&
441 (dp->nfsdl_flags & NFSCLDL_READ)) {
442 (void) nfscl_trydelegreturn(dp, cred, VFSTONFS(mp), p);
443 free(dp, M_NFSCLDELEG);
448 /* Look for the correct deleg, based upon FH */
450 tdp = nfscl_finddeleg(clp, nfhp, fhlen);
454 return (NFSERR_BADSTATEID);
457 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list);
458 LIST_INSERT_HEAD(NFSCLDELEGHASH(clp, nfhp, fhlen), dp,
460 dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
461 nfsstatsv1.cldelegates++;
465 * Delegation already exists, what do we do if a new one??
468 printf("Deleg already exists!\n");
469 free(dp, M_NFSCLDELEG);
480 * Find a delegation for this file handle. Return NULL upon failure.
482 static struct nfscldeleg *
483 nfscl_finddeleg(struct nfsclclient *clp, u_int8_t *fhp, int fhlen)
485 struct nfscldeleg *dp;
487 LIST_FOREACH(dp, NFSCLDELEGHASH(clp, fhp, fhlen), nfsdl_hash) {
488 if (dp->nfsdl_fhlen == fhlen &&
489 !NFSBCMP(dp->nfsdl_fh, fhp, fhlen))
496 * Get a stateid for an I/O operation. First, look for an open and iff
497 * found, return either a lockowner stateid or the open stateid.
498 * If no Open is found, just return error and the special stateid of all zeros.
501 nfscl_getstateid(vnode_t vp, u_int8_t *nfhp, int fhlen, u_int32_t mode,
502 int fords, struct ucred *cred, NFSPROC_T *p, nfsv4stateid_t *stateidp,
505 struct nfsclclient *clp;
506 struct nfsclowner *owp;
507 struct nfsclopen *op = NULL, *top;
508 struct nfscllockowner *lp;
509 struct nfscldeleg *dp;
511 struct nfsmount *nmp;
512 u_int8_t own[NFSV4CL_LOCKNAMELEN];
517 * Initially, just set the special stateid of all zeros.
518 * (Don't do this for a DS, since the special stateid can't be used.)
522 stateidp->other[0] = 0;
523 stateidp->other[1] = 0;
524 stateidp->other[2] = 0;
526 if (vnode_vtype(vp) != VREG)
529 nmp = VFSTONFS(vnode_mount(vp));
531 clp = nfscl_findcl(nmp);
538 * Wait for recovery to complete.
540 while ((clp->nfsc_flags & NFSCLFLAGS_RECVRINPROG))
541 (void) nfsmsleep(&clp->nfsc_flags, NFSCLSTATEMUTEXPTR,
542 PZERO, "nfsrecvr", NULL);
545 * First, look for a delegation.
547 LIST_FOREACH(dp, NFSCLDELEGHASH(clp, nfhp, fhlen), nfsdl_hash) {
548 if (dp->nfsdl_fhlen == fhlen &&
549 !NFSBCMP(nfhp, dp->nfsdl_fh, fhlen)) {
550 if (!(mode & NFSV4OPEN_ACCESSWRITE) ||
551 (dp->nfsdl_flags & NFSCLDL_WRITE)) {
552 stateidp->seqid = dp->nfsdl_stateid.seqid;
553 stateidp->other[0] = dp->nfsdl_stateid.other[0];
554 stateidp->other[1] = dp->nfsdl_stateid.other[1];
555 stateidp->other[2] = dp->nfsdl_stateid.other[2];
556 if (!(np->n_flag & NDELEGRECALL)) {
557 TAILQ_REMOVE(&clp->nfsc_deleg, dp,
559 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp,
561 dp->nfsdl_timestamp = NFSD_MONOSEC +
563 dp->nfsdl_rwlock.nfslock_usecnt++;
564 *lckpp = (void *)&dp->nfsdl_rwlock;
575 * If p != NULL, we want to search the parentage tree
576 * for a matching OpenOwner and use that.
578 if (NFSHASONEOPENOWN(VFSTONFS(vnode_mount(vp))))
579 nfscl_filllockowner(NULL, own, F_POSIX);
581 nfscl_filllockowner(p->td_proc, own, F_POSIX);
583 error = nfscl_getopen(&clp->nfsc_owner, nfhp, fhlen, own, own,
585 if (error == 0 && lp != NULL && fords == 0) {
586 /* Don't return a lock stateid for a DS. */
588 lp->nfsl_stateid.seqid;
590 lp->nfsl_stateid.other[0];
592 lp->nfsl_stateid.other[1];
594 lp->nfsl_stateid.other[2];
600 /* If not found, just look for any OpenOwner that will work. */
603 owp = LIST_FIRST(&clp->nfsc_owner);
604 while (!done && owp != NULL) {
605 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
606 if (op->nfso_fhlen == fhlen &&
607 !NFSBCMP(op->nfso_fh, nfhp, fhlen)) {
608 if (top == NULL && (op->nfso_mode &
609 NFSV4OPEN_ACCESSWRITE) != 0 &&
610 (mode & NFSV4OPEN_ACCESSREAD) != 0)
612 if ((mode & op->nfso_mode) == mode) {
619 owp = LIST_NEXT(owp, nfsow_list);
622 NFSCL_DEBUG(2, "openmode top=%p\n", top);
623 if (top == NULL || NFSHASOPENMODE(nmp)) {
630 * For read aheads or write behinds, use the open cred.
631 * A read ahead or write behind is indicated by p == NULL.
634 newnfs_copycred(&op->nfso_cred, cred);
638 * No lock stateid, so return the open stateid.
640 stateidp->seqid = op->nfso_stateid.seqid;
641 stateidp->other[0] = op->nfso_stateid.other[0];
642 stateidp->other[1] = op->nfso_stateid.other[1];
643 stateidp->other[2] = op->nfso_stateid.other[2];
649 * Search for a matching file, mode and, optionally, lockowner.
652 nfscl_getopen(struct nfsclownerhead *ohp, u_int8_t *nfhp, int fhlen,
653 u_int8_t *openown, u_int8_t *lockown, u_int32_t mode,
654 struct nfscllockowner **lpp, struct nfsclopen **opp)
656 struct nfsclowner *owp;
657 struct nfsclopen *op, *rop, *rop2;
658 struct nfscllockowner *lp;
664 * rop will be set to the open to be returned. There are three
665 * variants of this, all for an open of the correct file:
666 * 1 - A match of lockown.
667 * 2 - A match of the openown, when no lockown match exists.
668 * 3 - A match for any open, if no openown or lockown match exists.
669 * Looking for #2 over #3 probably isn't necessary, but since
670 * RFC3530 is vague w.r.t. the relationship between openowners and
671 * lockowners, I think this is the safer way to go.
676 /* Search the client list */
677 owp = LIST_FIRST(ohp);
678 while (owp != NULL && keep_looping != 0) {
679 /* and look for the correct open */
680 op = LIST_FIRST(&owp->nfsow_open);
681 while (op != NULL && keep_looping != 0) {
682 if (op->nfso_fhlen == fhlen &&
683 !NFSBCMP(op->nfso_fh, nfhp, fhlen)
684 && (op->nfso_mode & mode) == mode) {
686 /* Now look for a matching lockowner. */
687 LIST_FOREACH(lp, &op->nfso_lock,
689 if (!NFSBCMP(lp->nfsl_owner,
691 NFSV4CL_LOCKNAMELEN)) {
699 if (rop == NULL && !NFSBCMP(owp->nfsow_owner,
700 openown, NFSV4CL_LOCKNAMELEN)) {
708 op = LIST_NEXT(op, nfso_list);
710 owp = LIST_NEXT(owp, nfsow_list);
721 * Release use of an open owner. Called when open operations are done
722 * with the open owner.
725 nfscl_ownerrelease(struct nfsmount *nmp, struct nfsclowner *owp,
726 __unused int error, __unused int candelete, int unlocked)
733 if (NFSHASONEOPENOWN(nmp))
734 nfsv4_relref(&owp->nfsow_rwlock);
736 nfscl_lockunlock(&owp->nfsow_rwlock);
738 nfscl_clrelease(owp->nfsow_clp);
743 * Release use of an open structure under an open owner.
746 nfscl_openrelease(struct nfsmount *nmp, struct nfsclopen *op, int error,
749 struct nfsclclient *clp;
750 struct nfsclowner *owp;
756 if (NFSHASONEOPENOWN(nmp))
757 nfsv4_relref(&owp->nfsow_rwlock);
759 nfscl_lockunlock(&owp->nfsow_rwlock);
760 clp = owp->nfsow_clp;
761 if (error && candelete && op->nfso_opencnt == 0)
762 nfscl_freeopen(op, 0);
763 nfscl_clrelease(clp);
768 * Called to get a clientid structure. It will optionally lock the
769 * client data structures to do the SetClientId/SetClientId_confirm,
770 * but will release that lock and return the clientid with a reference
772 * If the "cred" argument is NULL, a new clientid should not be created.
773 * If the "p" argument is NULL, a SetClientID/SetClientIDConfirm cannot
775 * The start_renewthread argument tells nfscl_getcl() to start a renew
776 * thread if this creates a new clp.
777 * It always clpp with a reference count on it, unless returning an error.
780 nfscl_getcl(struct mount *mp, struct ucred *cred, NFSPROC_T *p,
781 int start_renewthread, bool firstref, struct nfsclclient **clpp)
783 struct nfsclclient *clp;
784 struct nfsclclient *newclp = NULL;
785 struct nfsmount *nmp;
786 char uuid[HOSTUUIDLEN];
787 int igotlock = 0, error, trystalecnt, clidinusedelay, i;
792 getcredhostuuid(cred, uuid, sizeof uuid);
793 idlen = strlen(uuid);
795 idlen += sizeof (u_int64_t);
797 idlen += sizeof (u_int64_t) + 16; /* 16 random bytes */
799 sizeof (struct nfsclclient) + idlen - 1, M_NFSCLCLIENT,
804 * If a forced dismount is already in progress, don't
805 * allocate a new clientid and get out now. For the case where
806 * clp != NULL, this is a harmless optimization.
808 if (NFSCL_FORCEDISM(mp)) {
811 free(newclp, M_NFSCLCLIENT);
816 if (newclp == NULL) {
821 clp->nfsc_idlen = idlen;
822 LIST_INIT(&clp->nfsc_owner);
823 TAILQ_INIT(&clp->nfsc_deleg);
824 TAILQ_INIT(&clp->nfsc_layout);
825 LIST_INIT(&clp->nfsc_devinfo);
826 for (i = 0; i < NFSCLDELEGHASHSIZE; i++)
827 LIST_INIT(&clp->nfsc_deleghash[i]);
828 for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++)
829 LIST_INIT(&clp->nfsc_layouthash[i]);
830 clp->nfsc_flags = NFSCLFLAGS_INITED;
831 clp->nfsc_clientidrev = 1;
832 clp->nfsc_cbident = nfscl_nextcbident();
833 nfscl_fillclid(nmp->nm_clval, uuid, clp->nfsc_id,
835 LIST_INSERT_HEAD(&nfsclhead, clp, nfsc_list);
839 if (start_renewthread != 0)
840 nfscl_start_renewthread(clp);
844 free(newclp, M_NFSCLCLIENT);
847 while ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0 && !igotlock &&
848 !NFSCL_FORCEDISM(mp))
849 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
850 NFSCLSTATEMUTEXPTR, mp);
853 * Call nfsv4_lock() with "iwantlock == 0" on the firstref so
854 * that it will wait for a pending exclusive lock request.
855 * This gives the exclusive lock request priority over this
856 * shared lock request.
857 * An exclusive lock on nfsc_lock is used mainly for server
858 * crash recoveries and delegation recalls.
861 nfsv4_lock(&clp->nfsc_lock, 0, NULL, NFSCLSTATEMUTEXPTR,
863 nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
865 if (igotlock == 0 && NFSCL_FORCEDISM(mp)) {
867 * Both nfsv4_lock() and nfsv4_getref() know to check
868 * for NFSCL_FORCEDISM() and return without sleeping to
869 * wait for the exclusive lock to be released, since it
870 * might be held by nfscl_umount() and we need to get out
871 * now for that case and not wait until nfscl_umount()
880 * If it needs a clientid, do the setclientid now.
882 if ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0) {
884 panic("nfscl_clget");
885 if (p == NULL || cred == NULL) {
887 nfsv4_unlock(&clp->nfsc_lock, 0);
892 * If RFC3530 Sec. 14.2.33 is taken literally,
893 * NFSERR_CLIDINUSE will be returned persistently for the
894 * case where a new mount of the same file system is using
895 * a different principal. In practice, NFSERR_CLIDINUSE is
896 * only returned when there is outstanding unexpired state
897 * on the clientid. As such, try for twice the lease
898 * interval, if we know what that is. Otherwise, make a
900 * The case of returning NFSERR_STALECLIENTID is far less
901 * likely, but might occur if there is a significant delay
902 * between doing the SetClientID and SetClientIDConfirm Ops,
903 * such that the server throws away the clientid before
904 * receiving the SetClientIDConfirm.
906 if (clp->nfsc_renew > 0)
907 clidinusedelay = NFSCL_LEASE(clp->nfsc_renew) * 2;
909 clidinusedelay = 120;
912 error = nfsrpc_setclient(nmp, clp, 0, cred, p);
913 if (error == NFSERR_STALECLIENTID ||
914 error == NFSERR_STALEDONTRECOVER ||
915 error == NFSERR_BADSESSION ||
916 error == NFSERR_CLIDINUSE) {
917 (void) nfs_catnap(PZERO, error, "nfs_setcl");
919 } while (((error == NFSERR_STALECLIENTID ||
920 error == NFSERR_BADSESSION ||
921 error == NFSERR_STALEDONTRECOVER) && --trystalecnt > 0) ||
922 (error == NFSERR_CLIDINUSE && --clidinusedelay > 0));
925 nfsv4_unlock(&clp->nfsc_lock, 0);
929 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
933 nfsv4_unlock(&clp->nfsc_lock, 1);
942 * Get a reference to a clientid and return it, if valid.
945 nfscl_findcl(struct nfsmount *nmp)
947 struct nfsclclient *clp;
950 if (clp == NULL || !(clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID))
956 * Release the clientid structure. It may be locked or reference counted.
959 nfscl_clrelease(struct nfsclclient *clp)
962 if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK)
963 nfsv4_unlock(&clp->nfsc_lock, 0);
965 nfsv4_relref(&clp->nfsc_lock);
969 * External call for nfscl_clrelease.
972 nfscl_clientrelease(struct nfsclclient *clp)
976 if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK)
977 nfsv4_unlock(&clp->nfsc_lock, 0);
979 nfsv4_relref(&clp->nfsc_lock);
984 * Called when wanting to lock a byte region.
987 nfscl_getbytelock(vnode_t vp, u_int64_t off, u_int64_t len,
988 short type, struct ucred *cred, NFSPROC_T *p, struct nfsclclient *rclp,
989 int recovery, void *id, int flags, u_int8_t *rownp, u_int8_t *ropenownp,
990 struct nfscllockowner **lpp, int *newonep, int *donelocallyp)
992 struct nfscllockowner *lp;
993 struct nfsclopen *op;
994 struct nfsclclient *clp;
995 struct nfscllockowner *nlp;
996 struct nfscllock *nlop, *otherlop;
997 struct nfscldeleg *dp = NULL, *ldp = NULL;
998 struct nfscllockownerhead *lhp = NULL;
1000 u_int8_t own[NFSV4CL_LOCKNAMELEN], *ownp, openown[NFSV4CL_LOCKNAMELEN];
1002 int error = 0, ret, donelocally = 0;
1005 /* For Lock Ops, the open mode doesn't matter, so use 0 to match any. */
1014 * Might need these, so MALLOC them now, to
1015 * avoid a tsleep() in MALLOC later.
1018 sizeof (struct nfscllockowner), M_NFSCLLOCKOWNER, M_WAITOK);
1020 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1022 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1023 nlop->nfslo_type = type;
1024 nlop->nfslo_first = off;
1025 if (len == NFS64BITSSET) {
1026 nlop->nfslo_end = NFS64BITSSET;
1028 nlop->nfslo_end = off + len;
1029 if (nlop->nfslo_end <= nlop->nfslo_first)
1030 error = NFSERR_INVAL;
1037 error = nfscl_getcl(vnode_mount(vp), cred, p, 1, true,
1041 free(nlp, M_NFSCLLOCKOWNER);
1042 free(otherlop, M_NFSCLLOCK);
1043 free(nlop, M_NFSCLLOCK);
1050 openownp = ropenownp;
1052 nfscl_filllockowner(id, own, flags);
1054 if (NFSHASONEOPENOWN(VFSTONFS(vnode_mount(vp))))
1055 nfscl_filllockowner(NULL, openown, F_POSIX);
1057 nfscl_filllockowner(p->td_proc, openown, F_POSIX);
1063 * First, search for a delegation. If one exists for this file,
1064 * the lock can be done locally against it, so long as there
1065 * isn't a local lock conflict.
1067 ldp = dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
1068 np->n_fhp->nfh_len);
1069 /* Just sanity check for correct type of delegation */
1070 if (dp != NULL && ((dp->nfsdl_flags &
1071 (NFSCLDL_RECALL | NFSCLDL_DELEGRET)) != 0 ||
1073 (dp->nfsdl_flags & NFSCLDL_WRITE) == 0)))
1077 /* Now, find an open and maybe a lockowner. */
1078 ret = nfscl_getopen(&dp->nfsdl_owner, np->n_fhp->nfh_fh,
1079 np->n_fhp->nfh_len, openownp, ownp, mode, NULL, &op);
1081 ret = nfscl_getopen(&clp->nfsc_owner,
1082 np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp,
1083 ownp, mode, NULL, &op);
1085 lhp = &dp->nfsdl_lock;
1086 TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
1087 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list);
1088 dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
1096 * Get the related Open and maybe lockowner.
1098 error = nfscl_getopen(&clp->nfsc_owner,
1099 np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp,
1100 ownp, mode, &lp, &op);
1102 lhp = &op->nfso_lock;
1104 if (!error && !recovery)
1105 error = nfscl_localconflict(clp, np->n_fhp->nfh_fh,
1106 np->n_fhp->nfh_len, nlop, ownp, ldp, NULL);
1109 nfscl_clrelease(clp);
1112 free(nlp, M_NFSCLLOCKOWNER);
1113 free(otherlop, M_NFSCLLOCK);
1114 free(nlop, M_NFSCLLOCK);
1119 * Ok, see if a lockowner exists and create one, as required.
1122 LIST_FOREACH(lp, lhp, nfsl_list) {
1123 if (!NFSBCMP(lp->nfsl_owner, ownp, NFSV4CL_LOCKNAMELEN))
1127 NFSBCOPY(ownp, nlp->nfsl_owner, NFSV4CL_LOCKNAMELEN);
1129 NFSBCOPY(ropenownp, nlp->nfsl_openowner,
1130 NFSV4CL_LOCKNAMELEN);
1132 NFSBCOPY(op->nfso_own->nfsow_owner, nlp->nfsl_openowner,
1133 NFSV4CL_LOCKNAMELEN);
1134 nlp->nfsl_seqid = 0;
1135 nlp->nfsl_lockflags = flags;
1136 nlp->nfsl_inprog = NULL;
1137 nfscl_lockinit(&nlp->nfsl_rwlock);
1138 LIST_INIT(&nlp->nfsl_lock);
1140 nlp->nfsl_open = NULL;
1141 nfsstatsv1.cllocallockowners++;
1143 nlp->nfsl_open = op;
1144 nfsstatsv1.cllockowners++;
1146 LIST_INSERT_HEAD(lhp, nlp, nfsl_list);
1153 * Now, update the byte ranges for locks.
1155 ret = nfscl_updatelock(lp, &nlop, &otherlop, donelocally);
1161 nfscl_clrelease(clp);
1164 * Serial modifications on the lock owner for multiple threads
1165 * for the same process using a read/write lock.
1168 nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR);
1174 free(nlp, M_NFSCLLOCKOWNER);
1176 free(nlop, M_NFSCLLOCK);
1178 free(otherlop, M_NFSCLLOCK);
1185 * Called to unlock a byte range, for LockU.
1188 nfscl_relbytelock(vnode_t vp, u_int64_t off, u_int64_t len,
1189 __unused struct ucred *cred, NFSPROC_T *p, int callcnt,
1190 struct nfsclclient *clp, void *id, int flags,
1191 struct nfscllockowner **lpp, int *dorpcp)
1193 struct nfscllockowner *lp;
1194 struct nfsclowner *owp;
1195 struct nfsclopen *op;
1196 struct nfscllock *nlop, *other_lop = NULL;
1197 struct nfscldeleg *dp;
1199 u_int8_t own[NFSV4CL_LOCKNAMELEN];
1207 * Might need these, so MALLOC them now, to
1208 * avoid a tsleep() in MALLOC later.
1211 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1212 nlop->nfslo_type = F_UNLCK;
1213 nlop->nfslo_first = off;
1214 if (len == NFS64BITSSET) {
1215 nlop->nfslo_end = NFS64BITSSET;
1217 nlop->nfslo_end = off + len;
1218 if (nlop->nfslo_end <= nlop->nfslo_first) {
1219 free(nlop, M_NFSCLLOCK);
1220 return (NFSERR_INVAL);
1225 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1228 nfscl_filllockowner(id, own, flags);
1232 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
1233 np->n_fhp->nfh_len);
1236 * First, unlock any local regions on a delegation.
1239 /* Look for this lockowner. */
1240 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
1241 if (!NFSBCMP(lp->nfsl_owner, own,
1242 NFSV4CL_LOCKNAMELEN))
1246 /* Use other_lop, so nlop is still available */
1247 (void)nfscl_updatelock(lp, &other_lop, NULL, 1);
1251 * Now, find a matching open/lockowner that hasn't already been done,
1252 * as marked by nfsl_inprog.
1256 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
1257 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1258 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1259 !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1260 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1261 if (lp->nfsl_inprog == NULL &&
1262 !NFSBCMP(lp->nfsl_owner, own,
1263 NFSV4CL_LOCKNAMELEN)) {
1277 ret = nfscl_updatelock(lp, &nlop, NULL, 0);
1281 * Serial modifications on the lock owner for multiple
1282 * threads for the same process using a read/write lock.
1284 lp->nfsl_inprog = p;
1285 nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR);
1290 free(nlop, M_NFSCLLOCK);
1292 free(other_lop, M_NFSCLLOCK);
1297 * Release all lockowners marked in progess for this process and file.
1300 nfscl_releasealllocks(struct nfsclclient *clp, vnode_t vp, NFSPROC_T *p,
1301 void *id, int flags)
1303 struct nfsclowner *owp;
1304 struct nfsclopen *op;
1305 struct nfscllockowner *lp;
1307 u_int8_t own[NFSV4CL_LOCKNAMELEN];
1310 nfscl_filllockowner(id, own, flags);
1312 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
1313 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1314 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1315 !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1316 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1317 if (lp->nfsl_inprog == p &&
1318 !NFSBCMP(lp->nfsl_owner, own,
1319 NFSV4CL_LOCKNAMELEN)) {
1320 lp->nfsl_inprog = NULL;
1321 nfscl_lockunlock(&lp->nfsl_rwlock);
1327 nfscl_clrelease(clp);
1332 * Called to find out if any bytes within the byte range specified are
1333 * write locked by the calling process. Used to determine if flushing
1334 * is required before a LockU.
1335 * If in doubt, return 1, so the flush will occur.
1338 nfscl_checkwritelocked(vnode_t vp, struct flock *fl,
1339 struct ucred *cred, NFSPROC_T *p, void *id, int flags)
1341 struct nfsclowner *owp;
1342 struct nfscllockowner *lp;
1343 struct nfsclopen *op;
1344 struct nfsclclient *clp;
1345 struct nfscllock *lop;
1346 struct nfscldeleg *dp;
1349 u_int8_t own[NFSV4CL_LOCKNAMELEN];
1353 switch (fl->l_whence) {
1357 * Caller is responsible for adding any necessary offset
1358 * when SEEK_CUR is used.
1363 off = np->n_size + fl->l_start;
1368 if (fl->l_len != 0) {
1369 end = off + fl->l_len;
1376 error = nfscl_getcl(vnode_mount(vp), cred, p, 1, true, &clp);
1379 nfscl_filllockowner(id, own, flags);
1383 * First check the delegation locks.
1385 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
1387 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
1388 if (!NFSBCMP(lp->nfsl_owner, own,
1389 NFSV4CL_LOCKNAMELEN))
1393 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
1394 if (lop->nfslo_first >= end)
1396 if (lop->nfslo_end <= off)
1398 if (lop->nfslo_type == F_WRLCK) {
1399 nfscl_clrelease(clp);
1408 * Now, check state against the server.
1410 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
1411 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1412 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1413 !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1414 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1415 if (!NFSBCMP(lp->nfsl_owner, own,
1416 NFSV4CL_LOCKNAMELEN))
1420 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
1421 if (lop->nfslo_first >= end)
1423 if (lop->nfslo_end <= off)
1425 if (lop->nfslo_type == F_WRLCK) {
1426 nfscl_clrelease(clp);
1435 nfscl_clrelease(clp);
1441 * Release a byte range lock owner structure.
1444 nfscl_lockrelease(struct nfscllockowner *lp, int error, int candelete)
1446 struct nfsclclient *clp;
1451 clp = lp->nfsl_open->nfso_own->nfsow_clp;
1452 if (error != 0 && candelete &&
1453 (lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED) == 0)
1454 nfscl_freelockowner(lp, 0);
1456 nfscl_lockunlock(&lp->nfsl_rwlock);
1457 nfscl_clrelease(clp);
1462 * Free up an open structure and any associated byte range lock structures.
1465 nfscl_freeopen(struct nfsclopen *op, int local)
1468 LIST_REMOVE(op, nfso_list);
1469 nfscl_freealllocks(&op->nfso_lock, local);
1470 free(op, M_NFSCLOPEN);
1472 nfsstatsv1.cllocalopens--;
1474 nfsstatsv1.clopens--;
1478 * Free up all lock owners and associated locks.
1481 nfscl_freealllocks(struct nfscllockownerhead *lhp, int local)
1483 struct nfscllockowner *lp, *nlp;
1485 LIST_FOREACH_SAFE(lp, lhp, nfsl_list, nlp) {
1486 if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED))
1488 nfscl_freelockowner(lp, local);
1493 * Called for an Open when NFSERR_EXPIRED is received from the server.
1494 * If there are no byte range locks nor a Share Deny lost, try to do a
1495 * fresh Open. Otherwise, free the open.
1498 nfscl_expireopen(struct nfsclclient *clp, struct nfsclopen *op,
1499 struct nfsmount *nmp, struct ucred *cred, NFSPROC_T *p)
1501 struct nfscllockowner *lp;
1502 struct nfscldeleg *dp;
1503 int mustdelete = 0, error;
1506 * Look for any byte range lock(s).
1508 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1509 if (!LIST_EMPTY(&lp->nfsl_lock)) {
1516 * If no byte range lock(s) nor a Share deny, try to re-open.
1518 if (!mustdelete && (op->nfso_mode & NFSLCK_DENYBITS) == 0) {
1519 newnfs_copycred(&op->nfso_cred, cred);
1521 error = nfsrpc_reopen(nmp, op->nfso_fh,
1522 op->nfso_fhlen, op->nfso_mode, op, &dp, cred, p);
1526 free(dp, M_NFSCLDELEG);
1531 nfscl_deleg(nmp->nm_mountp, clp, op->nfso_fh,
1532 op->nfso_fhlen, cred, p, &dp);
1536 * If a byte range lock or Share deny or couldn't re-open, free it.
1539 nfscl_freeopen(op, 0);
1540 return (mustdelete);
1544 * Free up an open owner structure.
1547 nfscl_freeopenowner(struct nfsclowner *owp, int local)
1550 LIST_REMOVE(owp, nfsow_list);
1551 free(owp, M_NFSCLOWNER);
1553 nfsstatsv1.cllocalopenowners--;
1555 nfsstatsv1.clopenowners--;
1559 * Free up a byte range lock owner structure.
1562 nfscl_freelockowner(struct nfscllockowner *lp, int local)
1564 struct nfscllock *lop, *nlop;
1566 LIST_REMOVE(lp, nfsl_list);
1567 LIST_FOREACH_SAFE(lop, &lp->nfsl_lock, nfslo_list, nlop) {
1568 nfscl_freelock(lop, local);
1570 free(lp, M_NFSCLLOCKOWNER);
1572 nfsstatsv1.cllocallockowners--;
1574 nfsstatsv1.cllockowners--;
1578 * Free up a byte range lock structure.
1581 nfscl_freelock(struct nfscllock *lop, int local)
1584 LIST_REMOVE(lop, nfslo_list);
1585 free(lop, M_NFSCLLOCK);
1587 nfsstatsv1.cllocallocks--;
1589 nfsstatsv1.cllocks--;
1593 * Clean out the state related to a delegation.
1596 nfscl_cleandeleg(struct nfscldeleg *dp)
1598 struct nfsclowner *owp, *nowp;
1599 struct nfsclopen *op;
1601 LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) {
1602 op = LIST_FIRST(&owp->nfsow_open);
1604 if (LIST_NEXT(op, nfso_list) != NULL)
1605 panic("nfscleandel");
1606 nfscl_freeopen(op, 1);
1608 nfscl_freeopenowner(owp, 1);
1610 nfscl_freealllocks(&dp->nfsdl_lock, 1);
1614 * Free a delegation.
1617 nfscl_freedeleg(struct nfscldeleghead *hdp, struct nfscldeleg *dp, bool freeit)
1620 TAILQ_REMOVE(hdp, dp, nfsdl_list);
1621 LIST_REMOVE(dp, nfsdl_hash);
1623 free(dp, M_NFSCLDELEG);
1624 nfsstatsv1.cldelegates--;
1629 * Free up all state related to this client structure.
1632 nfscl_cleanclient(struct nfsclclient *clp)
1634 struct nfsclowner *owp, *nowp;
1635 struct nfsclopen *op, *nop;
1636 struct nfscllayout *lyp, *nlyp;
1637 struct nfscldevinfo *dip, *ndip;
1639 TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp)
1640 nfscl_freelayout(lyp);
1642 LIST_FOREACH_SAFE(dip, &clp->nfsc_devinfo, nfsdi_list, ndip)
1643 nfscl_freedevinfo(dip);
1645 /* Now, all the OpenOwners, etc. */
1646 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1647 LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
1648 nfscl_freeopen(op, 0);
1650 nfscl_freeopenowner(owp, 0);
1655 * Called when an NFSERR_EXPIRED is received from the server.
1658 nfscl_expireclient(struct nfsclclient *clp, struct nfsmount *nmp,
1659 struct ucred *cred, NFSPROC_T *p)
1661 struct nfsclowner *owp, *nowp, *towp;
1662 struct nfsclopen *op, *nop, *top;
1663 struct nfscldeleg *dp, *ndp;
1664 int ret, printed = 0;
1667 * First, merge locally issued Opens into the list for the server.
1669 dp = TAILQ_FIRST(&clp->nfsc_deleg);
1670 while (dp != NULL) {
1671 ndp = TAILQ_NEXT(dp, nfsdl_list);
1672 owp = LIST_FIRST(&dp->nfsdl_owner);
1673 while (owp != NULL) {
1674 nowp = LIST_NEXT(owp, nfsow_list);
1675 op = LIST_FIRST(&owp->nfsow_open);
1677 if (LIST_NEXT(op, nfso_list) != NULL)
1679 LIST_FOREACH(towp, &clp->nfsc_owner, nfsow_list) {
1680 if (!NFSBCMP(towp->nfsow_owner, owp->nfsow_owner,
1681 NFSV4CL_LOCKNAMELEN))
1685 /* Merge opens in */
1686 LIST_FOREACH(top, &towp->nfsow_open, nfso_list) {
1687 if (top->nfso_fhlen == op->nfso_fhlen &&
1688 !NFSBCMP(top->nfso_fh, op->nfso_fh,
1690 top->nfso_mode |= op->nfso_mode;
1691 top->nfso_opencnt += op->nfso_opencnt;
1696 /* Just add the open to the owner list */
1697 LIST_REMOVE(op, nfso_list);
1698 op->nfso_own = towp;
1699 LIST_INSERT_HEAD(&towp->nfsow_open, op, nfso_list);
1700 nfsstatsv1.cllocalopens--;
1701 nfsstatsv1.clopens++;
1704 /* Just add the openowner to the client list */
1705 LIST_REMOVE(owp, nfsow_list);
1706 owp->nfsow_clp = clp;
1707 LIST_INSERT_HEAD(&clp->nfsc_owner, owp, nfsow_list);
1708 nfsstatsv1.cllocalopenowners--;
1709 nfsstatsv1.clopenowners++;
1710 nfsstatsv1.cllocalopens--;
1711 nfsstatsv1.clopens++;
1716 if (!printed && !LIST_EMPTY(&dp->nfsdl_lock)) {
1718 printf("nfsv4 expired locks lost\n");
1720 nfscl_cleandeleg(dp);
1721 nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
1724 if (!TAILQ_EMPTY(&clp->nfsc_deleg))
1728 * Now, try and reopen against the server.
1730 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1731 owp->nfsow_seqid = 0;
1732 LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
1733 ret = nfscl_expireopen(clp, op, nmp, cred, p);
1734 if (ret && !printed) {
1736 printf("nfsv4 expired locks lost\n");
1739 if (LIST_EMPTY(&owp->nfsow_open))
1740 nfscl_freeopenowner(owp, 0);
1745 * This function must be called after the process represented by "own" has
1746 * exited. Must be called with CLSTATE lock held.
1749 nfscl_cleanup_common(struct nfsclclient *clp, u_int8_t *own)
1751 struct nfsclowner *owp, *nowp;
1752 struct nfscllockowner *lp, *nlp;
1753 struct nfscldeleg *dp;
1755 /* First, get rid of local locks on delegations. */
1756 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
1757 LIST_FOREACH_SAFE(lp, &dp->nfsdl_lock, nfsl_list, nlp) {
1758 if (!NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
1759 if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED))
1761 nfscl_freelockowner(lp, 1);
1765 owp = LIST_FIRST(&clp->nfsc_owner);
1766 while (owp != NULL) {
1767 nowp = LIST_NEXT(owp, nfsow_list);
1768 if (!NFSBCMP(owp->nfsow_owner, own,
1769 NFSV4CL_LOCKNAMELEN)) {
1771 * If there are children that haven't closed the
1772 * file descriptors yet, the opens will still be
1773 * here. For that case, let the renew thread clear
1774 * out the OpenOwner later.
1776 if (LIST_EMPTY(&owp->nfsow_open))
1777 nfscl_freeopenowner(owp, 0);
1779 owp->nfsow_defunct = 1;
1786 * Find open/lock owners for processes that have exited.
1789 nfscl_cleanupkext(struct nfsclclient *clp, struct nfscllockownerfhhead *lhp)
1791 struct nfsclowner *owp, *nowp;
1792 struct nfsclopen *op;
1793 struct nfscllockowner *lp, *nlp;
1794 struct nfscldeleg *dp;
1798 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1799 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1800 LIST_FOREACH_SAFE(lp, &op->nfso_lock, nfsl_list, nlp) {
1801 if (LIST_EMPTY(&lp->nfsl_lock))
1802 nfscl_emptylockowner(lp, lhp);
1805 if (nfscl_procdoesntexist(owp->nfsow_owner))
1806 nfscl_cleanup_common(clp, owp->nfsow_owner);
1810 * For the single open_owner case, these lock owners need to be
1811 * checked to see if they still exist separately.
1812 * This is because nfscl_procdoesntexist() never returns true for
1813 * the single open_owner so that the above doesn't ever call
1814 * nfscl_cleanup_common().
1816 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
1817 LIST_FOREACH_SAFE(lp, &dp->nfsdl_lock, nfsl_list, nlp) {
1818 if (nfscl_procdoesntexist(lp->nfsl_owner))
1819 nfscl_cleanup_common(clp, lp->nfsl_owner);
1823 NFSPROCLISTUNLOCK();
1827 * Take the empty lock owner and move it to the local lhp list if the
1828 * associated process no longer exists.
1831 nfscl_emptylockowner(struct nfscllockowner *lp,
1832 struct nfscllockownerfhhead *lhp)
1834 struct nfscllockownerfh *lfhp, *mylfhp;
1835 struct nfscllockowner *nlp;
1838 /* If not a Posix lock owner, just return. */
1839 if ((lp->nfsl_lockflags & F_POSIX) == 0)
1845 * First, search to see if this lock owner is already in the list.
1846 * If it is, then the associated process no longer exists.
1848 SLIST_FOREACH(lfhp, lhp, nfslfh_list) {
1849 if (lfhp->nfslfh_len == lp->nfsl_open->nfso_fhlen &&
1850 !NFSBCMP(lfhp->nfslfh_fh, lp->nfsl_open->nfso_fh,
1853 LIST_FOREACH(nlp, &lfhp->nfslfh_lock, nfsl_list)
1854 if (!NFSBCMP(nlp->nfsl_owner, lp->nfsl_owner,
1855 NFSV4CL_LOCKNAMELEN))
1858 /* If not found, check if process still exists. */
1859 if (fnd_it == 0 && nfscl_procdoesntexist(lp->nfsl_owner) == 0)
1862 /* Move the lock owner over to the local list. */
1863 if (mylfhp == NULL) {
1864 mylfhp = malloc(sizeof(struct nfscllockownerfh), M_TEMP,
1868 mylfhp->nfslfh_len = lp->nfsl_open->nfso_fhlen;
1869 NFSBCOPY(lp->nfsl_open->nfso_fh, mylfhp->nfslfh_fh,
1870 mylfhp->nfslfh_len);
1871 LIST_INIT(&mylfhp->nfslfh_lock);
1872 SLIST_INSERT_HEAD(lhp, mylfhp, nfslfh_list);
1874 LIST_REMOVE(lp, nfsl_list);
1875 LIST_INSERT_HEAD(&mylfhp->nfslfh_lock, lp, nfsl_list);
1878 static int fake_global; /* Used to force visibility of MNTK_UNMOUNTF */
1880 * Called from nfs umount to free up the clientid.
1883 nfscl_umount(struct nfsmount *nmp, NFSPROC_T *p)
1885 struct nfsclclient *clp;
1890 * For the case that matters, this is the thread that set
1891 * MNTK_UNMOUNTF, so it will see it set. The code that follows is
1892 * done to ensure that any thread executing nfscl_getcl() after
1893 * this time, will see MNTK_UNMOUNTF set. nfscl_getcl() uses the
1894 * mutex for NFSLOCKCLSTATE(), so it is "m" for the following
1895 * explanation, courtesy of Alan Cox.
1896 * What follows is a snippet from Alan Cox's email at:
1897 * https://docs.FreeBSD.org/cgi/mid.cgi?BANLkTikR3d65zPHo9==08ZfJ2vmqZucEvw
1899 * 1. Set MNTK_UNMOUNTF
1900 * 2. Acquire a standard FreeBSD mutex "m".
1901 * 3. Update some data structures.
1902 * 4. Release mutex "m".
1904 * Then, other threads that acquire "m" after step 4 has occurred will
1905 * see MNTK_UNMOUNTF as set. But, other threads that beat thread X to
1906 * step 2 may or may not see MNTK_UNMOUNTF as set.
1909 if ((nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) {
1917 if ((clp->nfsc_flags & NFSCLFLAGS_INITED) == 0)
1918 panic("nfscl umount");
1921 * First, handshake with the nfscl renew thread, to terminate
1924 clp->nfsc_flags |= NFSCLFLAGS_UMOUNT;
1925 while (clp->nfsc_flags & NFSCLFLAGS_HASTHREAD)
1926 (void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT,
1930 * Now, get the exclusive lock on the client state, so
1931 * that no uses of the state are still in progress.
1934 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
1935 NFSCLSTATEMUTEXPTR, NULL);
1936 } while (!igotlock);
1940 * Free up all the state. It will expire on the server, but
1941 * maybe we should do a SetClientId/SetClientIdConfirm so
1942 * the server throws it away?
1944 LIST_REMOVE(clp, nfsc_list);
1945 nfscl_delegreturnall(clp, p);
1946 cred = newnfs_getcred();
1947 if (NFSHASNFSV4N(nmp)) {
1948 (void)nfsrpc_destroysession(nmp, clp, cred, p);
1949 (void)nfsrpc_destroyclient(nmp, clp, cred, p);
1951 (void)nfsrpc_setclient(nmp, clp, 0, cred, p);
1952 nfscl_cleanclient(clp);
1955 free(clp, M_NFSCLCLIENT);
1961 * This function is called when a server replies with NFSERR_STALECLIENTID
1962 * NFSERR_STALESTATEID or NFSERR_BADSESSION. It traverses the clientid lists,
1963 * doing Opens and Locks with reclaim. If these fail, it deletes the
1964 * corresponding state.
1967 nfscl_recover(struct nfsclclient *clp, struct ucred *cred, NFSPROC_T *p)
1969 struct nfsclowner *owp, *nowp;
1970 struct nfsclopen *op, *nop;
1971 struct nfscllockowner *lp, *nlp;
1972 struct nfscllock *lop, *nlop;
1973 struct nfscldeleg *dp, *ndp, *tdp;
1974 struct nfsmount *nmp;
1975 struct ucred *tcred;
1976 struct nfsclopenhead extra_open;
1977 struct nfscldeleghead extra_deleg;
1980 u_int32_t delegtype = NFSV4OPEN_DELEGATEWRITE, mode;
1981 int i, igotlock = 0, error, trycnt, firstlock;
1982 struct nfscllayout *lyp, *nlyp;
1986 * First, lock the client structure, so everyone else will
1987 * block when trying to use state.
1990 clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG;
1992 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
1993 NFSCLSTATEMUTEXPTR, NULL);
1994 } while (!igotlock);
1997 nmp = clp->nfsc_nmp;
1999 panic("nfscl recover");
2002 * For now, just get rid of all layouts. There may be a need
2003 * to do LayoutCommit Ops with reclaim == true later.
2005 TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp)
2006 nfscl_freelayout(lyp);
2007 TAILQ_INIT(&clp->nfsc_layout);
2008 for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++)
2009 LIST_INIT(&clp->nfsc_layouthash[i]);
2013 error = nfsrpc_setclient(nmp, clp, 1, cred, p);
2014 } while ((error == NFSERR_STALECLIENTID ||
2015 error == NFSERR_BADSESSION ||
2016 error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
2019 clp->nfsc_flags &= ~(NFSCLFLAGS_RECOVER |
2020 NFSCLFLAGS_RECVRINPROG);
2021 wakeup(&clp->nfsc_flags);
2022 nfsv4_unlock(&clp->nfsc_lock, 0);
2026 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
2027 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2030 * Mark requests already queued on the server, so that they don't
2031 * initiate another recovery cycle. Any requests already in the
2032 * queue that handle state information will have the old stale
2033 * clientid/stateid and will get a NFSERR_STALESTATEID,
2034 * NFSERR_STALECLIENTID or NFSERR_BADSESSION reply from the server.
2035 * This will be translated to NFSERR_STALEDONTRECOVER when
2036 * R_DONTRECOVER is set.
2039 TAILQ_FOREACH(rep, &nfsd_reqq, r_chain) {
2040 if (rep->r_nmp == nmp)
2041 rep->r_flags |= R_DONTRECOVER;
2046 * Now, mark all delegations "need reclaim".
2048 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list)
2049 dp->nfsdl_flags |= NFSCLDL_NEEDRECLAIM;
2051 TAILQ_INIT(&extra_deleg);
2052 LIST_INIT(&extra_open);
2054 * Now traverse the state lists, doing Open and Lock Reclaims.
2056 tcred = newnfs_getcred();
2057 recovered_one = false;
2058 owp = LIST_FIRST(&clp->nfsc_owner);
2059 while (owp != NULL) {
2060 nowp = LIST_NEXT(owp, nfsow_list);
2061 owp->nfsow_seqid = 0;
2062 op = LIST_FIRST(&owp->nfsow_open);
2063 while (op != NULL) {
2064 nop = LIST_NEXT(op, nfso_list);
2065 if (error != NFSERR_NOGRACE && error != NFSERR_BADSESSION) {
2066 /* Search for a delegation to reclaim with the open */
2067 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2068 if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
2070 if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
2071 mode = NFSV4OPEN_ACCESSWRITE;
2072 delegtype = NFSV4OPEN_DELEGATEWRITE;
2074 mode = NFSV4OPEN_ACCESSREAD;
2075 delegtype = NFSV4OPEN_DELEGATEREAD;
2077 if ((op->nfso_mode & mode) == mode &&
2078 op->nfso_fhlen == dp->nfsdl_fhlen &&
2079 !NFSBCMP(op->nfso_fh, dp->nfsdl_fh, op->nfso_fhlen))
2084 delegtype = NFSV4OPEN_DELEGATENONE;
2085 newnfs_copycred(&op->nfso_cred, tcred);
2086 error = nfscl_tryopen(nmp, NULL, op->nfso_fh,
2087 op->nfso_fhlen, op->nfso_fh, op->nfso_fhlen,
2088 op->nfso_mode, op, NULL, 0, &ndp, 1, delegtype,
2091 recovered_one = true;
2092 /* Handle any replied delegation */
2093 if (ndp != NULL && ((ndp->nfsdl_flags & NFSCLDL_WRITE)
2094 || NFSMNT_RDONLY(nmp->nm_mountp))) {
2095 if ((ndp->nfsdl_flags & NFSCLDL_WRITE))
2096 mode = NFSV4OPEN_ACCESSWRITE;
2098 mode = NFSV4OPEN_ACCESSREAD;
2099 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2100 if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
2102 if ((op->nfso_mode & mode) == mode &&
2103 op->nfso_fhlen == dp->nfsdl_fhlen &&
2104 !NFSBCMP(op->nfso_fh, dp->nfsdl_fh,
2106 dp->nfsdl_stateid = ndp->nfsdl_stateid;
2107 dp->nfsdl_sizelimit = ndp->nfsdl_sizelimit;
2108 dp->nfsdl_ace = ndp->nfsdl_ace;
2109 dp->nfsdl_change = ndp->nfsdl_change;
2110 dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
2111 if ((ndp->nfsdl_flags & NFSCLDL_RECALL))
2112 dp->nfsdl_flags |= NFSCLDL_RECALL;
2113 free(ndp, M_NFSCLDELEG);
2120 TAILQ_INSERT_HEAD(&extra_deleg, ndp, nfsdl_list);
2122 /* and reclaim all byte range locks */
2123 lp = LIST_FIRST(&op->nfso_lock);
2124 while (lp != NULL) {
2125 nlp = LIST_NEXT(lp, nfsl_list);
2128 lop = LIST_FIRST(&lp->nfsl_lock);
2129 while (lop != NULL) {
2130 nlop = LIST_NEXT(lop, nfslo_list);
2131 if (lop->nfslo_end == NFS64BITSSET)
2134 len = lop->nfslo_end - lop->nfslo_first;
2135 error = nfscl_trylock(nmp, NULL,
2136 op->nfso_fh, op->nfso_fhlen, lp,
2137 firstlock, 1, lop->nfslo_first, len,
2138 lop->nfslo_type, tcred, p);
2140 nfscl_freelock(lop, 0);
2145 /* If no locks, but a lockowner, just delete it. */
2146 if (LIST_EMPTY(&lp->nfsl_lock))
2147 nfscl_freelockowner(lp, 0);
2150 } else if (error == NFSERR_NOGRACE && !recovered_one &&
2151 NFSHASNFSV4N(nmp)) {
2153 * For NFSv4.1/4.2, the NFSERR_EXPIRED case will
2154 * actually end up here, since the client will do
2155 * a recovery for NFSERR_BADSESSION, but will get
2156 * an NFSERR_NOGRACE reply for the first "reclaim"
2158 * So, call nfscl_expireclient() to recover the
2159 * opens as best we can and then do a reclaim
2160 * complete and return.
2162 nfsrpc_reclaimcomplete(nmp, cred, p);
2163 nfscl_expireclient(clp, nmp, tcred, p);
2167 if (error != 0 && error != NFSERR_BADSESSION)
2168 nfscl_freeopen(op, 0);
2175 * Now, try and get any delegations not yet reclaimed by cobbling
2176 * to-gether an appropriate open.
2179 dp = TAILQ_FIRST(&clp->nfsc_deleg);
2180 while (dp != NULL) {
2181 ndp = TAILQ_NEXT(dp, nfsdl_list);
2182 if ((dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM)) {
2185 sizeof (struct nfsclowner), M_NFSCLOWNER, M_WAITOK);
2187 * Name must be as long an largest possible
2188 * NFSV4CL_LOCKNAMELEN. 12 for now.
2190 NFSBCOPY("RECLAIMDELEG", nowp->nfsow_owner,
2191 NFSV4CL_LOCKNAMELEN);
2192 LIST_INIT(&nowp->nfsow_open);
2193 nowp->nfsow_clp = clp;
2194 nowp->nfsow_seqid = 0;
2195 nowp->nfsow_defunct = 0;
2196 nfscl_lockinit(&nowp->nfsow_rwlock);
2199 if (error != NFSERR_NOGRACE && error != NFSERR_BADSESSION) {
2200 nop = malloc(sizeof (struct nfsclopen) +
2201 dp->nfsdl_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
2202 nop->nfso_own = nowp;
2203 if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
2204 nop->nfso_mode = NFSV4OPEN_ACCESSWRITE;
2205 delegtype = NFSV4OPEN_DELEGATEWRITE;
2207 nop->nfso_mode = NFSV4OPEN_ACCESSREAD;
2208 delegtype = NFSV4OPEN_DELEGATEREAD;
2210 nop->nfso_opencnt = 0;
2211 nop->nfso_posixlock = 1;
2212 nop->nfso_fhlen = dp->nfsdl_fhlen;
2213 NFSBCOPY(dp->nfsdl_fh, nop->nfso_fh, dp->nfsdl_fhlen);
2214 LIST_INIT(&nop->nfso_lock);
2215 nop->nfso_stateid.seqid = 0;
2216 nop->nfso_stateid.other[0] = 0;
2217 nop->nfso_stateid.other[1] = 0;
2218 nop->nfso_stateid.other[2] = 0;
2219 newnfs_copycred(&dp->nfsdl_cred, tcred);
2220 newnfs_copyincred(tcred, &nop->nfso_cred);
2222 error = nfscl_tryopen(nmp, NULL, nop->nfso_fh,
2223 nop->nfso_fhlen, nop->nfso_fh, nop->nfso_fhlen,
2224 nop->nfso_mode, nop, NULL, 0, &tdp, 1,
2225 delegtype, tcred, p);
2227 if ((tdp->nfsdl_flags & NFSCLDL_WRITE))
2228 mode = NFSV4OPEN_ACCESSWRITE;
2230 mode = NFSV4OPEN_ACCESSREAD;
2231 if ((nop->nfso_mode & mode) == mode &&
2232 nop->nfso_fhlen == tdp->nfsdl_fhlen &&
2233 !NFSBCMP(nop->nfso_fh, tdp->nfsdl_fh,
2235 dp->nfsdl_stateid = tdp->nfsdl_stateid;
2236 dp->nfsdl_sizelimit = tdp->nfsdl_sizelimit;
2237 dp->nfsdl_ace = tdp->nfsdl_ace;
2238 dp->nfsdl_change = tdp->nfsdl_change;
2239 dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
2240 if ((tdp->nfsdl_flags & NFSCLDL_RECALL))
2241 dp->nfsdl_flags |= NFSCLDL_RECALL;
2242 free(tdp, M_NFSCLDELEG);
2244 TAILQ_INSERT_HEAD(&extra_deleg, tdp, nfsdl_list);
2250 free(nop, M_NFSCLOPEN);
2251 if (error == NFSERR_NOGRACE && !recovered_one &&
2252 NFSHASNFSV4N(nmp)) {
2254 * For NFSv4.1/4.2, the NFSERR_EXPIRED case will
2255 * actually end up here, since the client will do
2256 * a recovery for NFSERR_BADSESSION, but will get
2257 * an NFSERR_NOGRACE reply for the first "reclaim"
2259 * So, call nfscl_expireclient() to recover the
2260 * opens as best we can and then do a reclaim
2261 * complete and return.
2263 nfsrpc_reclaimcomplete(nmp, cred, p);
2264 nfscl_expireclient(clp, nmp, tcred, p);
2265 free(nowp, M_NFSCLOWNER);
2269 * Couldn't reclaim it, so throw the state
2272 nfscl_cleandeleg(dp);
2273 nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
2275 recovered_one = true;
2276 LIST_INSERT_HEAD(&extra_open, nop, nfso_list);
2283 * Now, get rid of extra Opens and Delegations.
2285 LIST_FOREACH_SAFE(op, &extra_open, nfso_list, nop) {
2287 newnfs_copycred(&op->nfso_cred, tcred);
2288 error = nfscl_tryclose(op, tcred, nmp, p);
2289 if (error == NFSERR_GRACE)
2290 (void) nfs_catnap(PZERO, error, "nfsexcls");
2291 } while (error == NFSERR_GRACE);
2292 LIST_REMOVE(op, nfso_list);
2293 free(op, M_NFSCLOPEN);
2296 free(nowp, M_NFSCLOWNER);
2298 TAILQ_FOREACH_SAFE(dp, &extra_deleg, nfsdl_list, ndp) {
2300 newnfs_copycred(&dp->nfsdl_cred, tcred);
2301 error = nfscl_trydelegreturn(dp, tcred, nmp, p);
2302 if (error == NFSERR_GRACE)
2303 (void) nfs_catnap(PZERO, error, "nfsexdlg");
2304 } while (error == NFSERR_GRACE);
2305 TAILQ_REMOVE(&extra_deleg, dp, nfsdl_list);
2306 free(dp, M_NFSCLDELEG);
2309 /* For NFSv4.1 or later, do a RECLAIM_COMPLETE. */
2310 if (NFSHASNFSV4N(nmp))
2311 (void)nfsrpc_reclaimcomplete(nmp, cred, p);
2315 clp->nfsc_flags &= ~NFSCLFLAGS_RECVRINPROG;
2316 wakeup(&clp->nfsc_flags);
2317 nfsv4_unlock(&clp->nfsc_lock, 0);
2323 * This function is called when a server replies with NFSERR_EXPIRED.
2324 * It deletes all state for the client and does a fresh SetClientId/confirm.
2325 * XXX Someday it should post a signal to the process(es) that hold the
2326 * state, so they know that lock state has been lost.
2329 nfscl_hasexpired(struct nfsclclient *clp, u_int32_t clidrev, NFSPROC_T *p)
2331 struct nfsmount *nmp;
2333 int igotlock = 0, error, trycnt;
2336 * If the clientid has gone away or a new SetClientid has already
2337 * been done, just return ok.
2339 if (clp == NULL || clidrev != clp->nfsc_clientidrev)
2343 * First, lock the client structure, so everyone else will
2344 * block when trying to use state. Also, use NFSCLFLAGS_EXPIREIT so
2345 * that only one thread does the work.
2348 clp->nfsc_flags |= NFSCLFLAGS_EXPIREIT;
2350 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
2351 NFSCLSTATEMUTEXPTR, NULL);
2352 } while (!igotlock && (clp->nfsc_flags & NFSCLFLAGS_EXPIREIT));
2353 if ((clp->nfsc_flags & NFSCLFLAGS_EXPIREIT) == 0) {
2355 nfsv4_unlock(&clp->nfsc_lock, 0);
2359 clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG;
2362 nmp = clp->nfsc_nmp;
2364 panic("nfscl expired");
2365 cred = newnfs_getcred();
2368 error = nfsrpc_setclient(nmp, clp, 0, cred, p);
2369 } while ((error == NFSERR_STALECLIENTID ||
2370 error == NFSERR_BADSESSION ||
2371 error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
2374 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2377 * Expire the state for the client.
2379 nfscl_expireclient(clp, nmp, cred, p);
2381 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
2382 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2384 clp->nfsc_flags &= ~(NFSCLFLAGS_EXPIREIT | NFSCLFLAGS_RECVRINPROG);
2385 wakeup(&clp->nfsc_flags);
2386 nfsv4_unlock(&clp->nfsc_lock, 0);
2393 * This function inserts a lock in the list after insert_lop.
2396 nfscl_insertlock(struct nfscllockowner *lp, struct nfscllock *new_lop,
2397 struct nfscllock *insert_lop, int local)
2400 if ((struct nfscllockowner *)insert_lop == lp)
2401 LIST_INSERT_HEAD(&lp->nfsl_lock, new_lop, nfslo_list);
2403 LIST_INSERT_AFTER(insert_lop, new_lop, nfslo_list);
2405 nfsstatsv1.cllocallocks++;
2407 nfsstatsv1.cllocks++;
2411 * This function updates the locking for a lock owner and given file. It
2412 * maintains a list of lock ranges ordered on increasing file offset that
2413 * are NFSCLLOCK_READ or NFSCLLOCK_WRITE and non-overlapping (aka POSIX style).
2414 * It always adds new_lop to the list and sometimes uses the one pointed
2416 * Returns 1 if the locks were modified, 0 otherwise.
2419 nfscl_updatelock(struct nfscllockowner *lp, struct nfscllock **new_lopp,
2420 struct nfscllock **other_lopp, int local)
2422 struct nfscllock *new_lop = *new_lopp;
2423 struct nfscllock *lop, *tlop, *ilop;
2424 struct nfscllock *other_lop;
2425 int unlock = 0, modified = 0;
2429 * Work down the list until the lock is merged.
2431 if (new_lop->nfslo_type == F_UNLCK)
2433 ilop = (struct nfscllock *)lp;
2434 lop = LIST_FIRST(&lp->nfsl_lock);
2435 while (lop != NULL) {
2437 * Only check locks for this file that aren't before the start of
2440 if (lop->nfslo_end >= new_lop->nfslo_first) {
2441 if (new_lop->nfslo_end < lop->nfslo_first) {
2443 * If the new lock ends before the start of the
2444 * current lock's range, no merge, just insert
2449 if (new_lop->nfslo_type == lop->nfslo_type ||
2450 (new_lop->nfslo_first <= lop->nfslo_first &&
2451 new_lop->nfslo_end >= lop->nfslo_end)) {
2453 * This lock can be absorbed by the new lock/unlock.
2454 * This happens when it covers the entire range
2455 * of the old lock or is contiguous
2456 * with the old lock and is of the same type or an
2459 if (new_lop->nfslo_type != lop->nfslo_type ||
2460 new_lop->nfslo_first != lop->nfslo_first ||
2461 new_lop->nfslo_end != lop->nfslo_end)
2463 if (lop->nfslo_first < new_lop->nfslo_first)
2464 new_lop->nfslo_first = lop->nfslo_first;
2465 if (lop->nfslo_end > new_lop->nfslo_end)
2466 new_lop->nfslo_end = lop->nfslo_end;
2468 lop = LIST_NEXT(lop, nfslo_list);
2469 nfscl_freelock(tlop, local);
2474 * All these cases are for contiguous locks that are not the
2475 * same type, so they can't be merged.
2477 if (new_lop->nfslo_first <= lop->nfslo_first) {
2479 * This case is where the new lock overlaps with the
2480 * first part of the old lock. Move the start of the
2481 * old lock to just past the end of the new lock. The
2482 * new lock will be inserted in front of the old, since
2483 * ilop hasn't been updated. (We are done now.)
2485 if (lop->nfslo_first != new_lop->nfslo_end) {
2486 lop->nfslo_first = new_lop->nfslo_end;
2491 if (new_lop->nfslo_end >= lop->nfslo_end) {
2493 * This case is where the new lock overlaps with the
2494 * end of the old lock's range. Move the old lock's
2495 * end to just before the new lock's first and insert
2496 * the new lock after the old lock.
2497 * Might not be done yet, since the new lock could
2498 * overlap further locks with higher ranges.
2500 if (lop->nfslo_end != new_lop->nfslo_first) {
2501 lop->nfslo_end = new_lop->nfslo_first;
2505 lop = LIST_NEXT(lop, nfslo_list);
2509 * The final case is where the new lock's range is in the
2510 * middle of the current lock's and splits the current lock
2511 * up. Use *other_lopp to handle the second part of the
2512 * split old lock range. (We are done now.)
2513 * For unlock, we use new_lop as other_lop and tmp, since
2514 * other_lop and new_lop are the same for this case.
2515 * We noted the unlock case above, so we don't need
2516 * new_lop->nfslo_type any longer.
2518 tmp = new_lop->nfslo_first;
2520 other_lop = new_lop;
2523 other_lop = *other_lopp;
2526 other_lop->nfslo_first = new_lop->nfslo_end;
2527 other_lop->nfslo_end = lop->nfslo_end;
2528 other_lop->nfslo_type = lop->nfslo_type;
2529 lop->nfslo_end = tmp;
2530 nfscl_insertlock(lp, other_lop, lop, local);
2536 lop = LIST_NEXT(lop, nfslo_list);
2542 * Insert the new lock in the list at the appropriate place.
2545 nfscl_insertlock(lp, new_lop, ilop, local);
2553 * This function must be run as a kernel thread.
2554 * It does Renew Ops and recovery, when required.
2557 nfscl_renewthread(struct nfsclclient *clp, NFSPROC_T *p)
2559 struct nfsclowner *owp, *nowp;
2560 struct nfsclopen *op;
2561 struct nfscllockowner *lp, *nlp;
2562 struct nfscldeleghead dh;
2563 struct nfscldeleg *dp, *ndp;
2566 int error, cbpathdown, islept, igotlock, ret, clearok;
2567 uint32_t recover_done_time = 0;
2569 static time_t prevsec = 0;
2570 struct nfscllockownerfh *lfhp, *nlfhp;
2571 struct nfscllockownerfhhead lfh;
2572 struct nfscllayout *lyp, *nlyp;
2573 struct nfscldevinfo *dip, *ndip;
2574 struct nfscllayouthead rlh;
2575 struct nfsclrecalllayout *recallp;
2576 struct nfsclds *dsp;
2580 cred = newnfs_getcred();
2582 clp->nfsc_flags |= NFSCLFLAGS_HASTHREAD;
2583 mp = clp->nfsc_nmp->nm_mountp;
2586 newnfs_setroot(cred);
2588 if (clp->nfsc_flags & NFSCLFLAGS_RECOVER) {
2590 * Only allow one recover within 1/2 of the lease
2591 * duration (nfsc_renew).
2593 if (recover_done_time < NFSD_MONOSEC) {
2594 recover_done_time = NFSD_MONOSEC +
2596 NFSCL_DEBUG(1, "Doing recovery..\n");
2597 nfscl_recover(clp, cred, p);
2599 NFSCL_DEBUG(1, "Clear Recovery dt=%u ms=%jd\n",
2600 recover_done_time, (intmax_t)NFSD_MONOSEC);
2602 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2606 if (clp->nfsc_expire <= NFSD_MONOSEC &&
2607 (clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID)) {
2608 clp->nfsc_expire = NFSD_MONOSEC + clp->nfsc_renew;
2609 clidrev = clp->nfsc_clientidrev;
2610 error = nfsrpc_renew(clp, NULL, cred, p);
2611 if (error == NFSERR_CBPATHDOWN)
2613 else if (error == NFSERR_STALECLIENTID ||
2614 error == NFSERR_BADSESSION) {
2616 clp->nfsc_flags |= NFSCLFLAGS_RECOVER;
2618 } else if (error == NFSERR_EXPIRED)
2619 (void) nfscl_hasexpired(clp, clidrev, p);
2623 if (NFSHASNFSV4N(clp->nfsc_nmp)) {
2624 /* Do renews for any DS sessions. */
2625 NFSLOCKMNT(clp->nfsc_nmp);
2626 /* Skip first entry, since the MDS is handled above. */
2627 dsp = TAILQ_FIRST(&clp->nfsc_nmp->nm_sess);
2629 dsp = TAILQ_NEXT(dsp, nfsclds_list);
2630 while (dsp != NULL) {
2631 if (dsp->nfsclds_expire <= NFSD_MONOSEC &&
2632 dsp->nfsclds_sess.nfsess_defunct == 0) {
2633 dsp->nfsclds_expire = NFSD_MONOSEC +
2635 NFSUNLOCKMNT(clp->nfsc_nmp);
2636 (void)nfsrpc_renew(clp, dsp, cred, p);
2639 dsp = TAILQ_NEXT(dsp, nfsclds_list);
2641 NFSUNLOCKMNT(clp->nfsc_nmp);
2647 /* It's a Total Recall! */
2648 nfscl_totalrecall(clp);
2651 * Now, handle defunct owners.
2653 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
2654 if (LIST_EMPTY(&owp->nfsow_open)) {
2655 if (owp->nfsow_defunct != 0)
2656 nfscl_freeopenowner(owp, 0);
2661 * Do the recall on any delegations. To avoid trouble, always
2662 * come back up here after having slept.
2666 dp = TAILQ_FIRST(&clp->nfsc_deleg);
2667 while (dp != NULL) {
2668 ndp = TAILQ_NEXT(dp, nfsdl_list);
2669 if ((dp->nfsdl_flags & NFSCLDL_RECALL)) {
2671 * Wait for outstanding I/O ops to be done.
2673 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
2675 nfsv4_unlock(&clp->nfsc_lock, 0);
2678 dp->nfsdl_rwlock.nfslock_lock |=
2680 msleep(&dp->nfsdl_rwlock,
2681 NFSCLSTATEMUTEXPTR, PVFS, "nfscld",
2683 if (NFSCL_FORCEDISM(mp))
2688 igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
2689 &islept, NFSCLSTATEMUTEXPTR, mp);
2690 if (igotlock == 0 && NFSCL_FORCEDISM(mp))
2696 newnfs_copycred(&dp->nfsdl_cred, cred);
2697 ret = nfscl_recalldeleg(clp, clp->nfsc_nmp, dp,
2698 NULL, cred, p, 1, &vp);
2700 nfscl_cleandeleg(dp);
2701 TAILQ_REMOVE(&clp->nfsc_deleg, dp,
2703 LIST_REMOVE(dp, nfsdl_hash);
2704 TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
2706 nfsstatsv1.cldelegates--;
2710 * The nfsc_lock must be released before doing
2711 * vrele(), since it might call nfs_inactive().
2712 * For the unlikely case where the vnode failed
2713 * to be acquired by nfscl_recalldeleg(), a
2714 * VOP_RECLAIM() should be in progress and it
2715 * will return the delegation.
2717 nfsv4_unlock(&clp->nfsc_lock, 0);
2730 * Clear out old delegations, if we are above the high water
2731 * mark. Only clear out ones with no state related to them.
2732 * The tailq list is in LRU order.
2734 dp = TAILQ_LAST(&clp->nfsc_deleg, nfscldeleghead);
2735 while (nfscl_delegcnt > nfscl_deleghighwater && dp != NULL) {
2736 ndp = TAILQ_PREV(dp, nfscldeleghead, nfsdl_list);
2737 if (dp->nfsdl_rwlock.nfslock_usecnt == 0 &&
2738 dp->nfsdl_rwlock.nfslock_lock == 0 &&
2739 dp->nfsdl_timestamp < NFSD_MONOSEC &&
2740 (dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_ZAPPED |
2741 NFSCLDL_NEEDRECLAIM | NFSCLDL_DELEGRET)) == 0) {
2743 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
2744 op = LIST_FIRST(&owp->nfsow_open);
2751 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
2752 if (!LIST_EMPTY(&lp->nfsl_lock)) {
2759 TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
2760 LIST_REMOVE(dp, nfsdl_hash);
2761 TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
2763 nfsstatsv1.cldelegates--;
2769 nfsv4_unlock(&clp->nfsc_lock, 0);
2772 * Do the recall on any layouts. To avoid trouble, always
2773 * come back up here after having slept.
2777 TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp) {
2778 if ((lyp->nfsly_flags & NFSLY_RECALL) != 0) {
2780 * Wait for outstanding I/O ops to be done.
2782 if (lyp->nfsly_lock.nfslock_usecnt > 0 ||
2783 (lyp->nfsly_lock.nfslock_lock &
2784 NFSV4LOCK_LOCK) != 0) {
2785 lyp->nfsly_lock.nfslock_lock |=
2787 msleep(&lyp->nfsly_lock.nfslock_lock,
2788 NFSCLSTATEMUTEXPTR, PVFS, "nfslyp",
2790 if (NFSCL_FORCEDISM(mp))
2794 /* Move the layout to the recall list. */
2795 TAILQ_REMOVE(&clp->nfsc_layout, lyp,
2797 LIST_REMOVE(lyp, nfsly_hash);
2798 TAILQ_INSERT_HEAD(&rlh, lyp, nfsly_list);
2800 /* Handle any layout commits. */
2801 if (!NFSHASNOLAYOUTCOMMIT(clp->nfsc_nmp) &&
2802 (lyp->nfsly_flags & NFSLY_WRITTEN) != 0) {
2803 lyp->nfsly_flags &= ~NFSLY_WRITTEN;
2805 NFSCL_DEBUG(3, "do layoutcommit\n");
2806 nfscl_dolayoutcommit(clp->nfsc_nmp, lyp,
2814 /* Now, look for stale layouts. */
2815 lyp = TAILQ_LAST(&clp->nfsc_layout, nfscllayouthead);
2816 while (lyp != NULL) {
2817 nlyp = TAILQ_PREV(lyp, nfscllayouthead, nfsly_list);
2818 if (lyp->nfsly_timestamp < NFSD_MONOSEC &&
2819 (lyp->nfsly_flags & NFSLY_RECALL) == 0 &&
2820 lyp->nfsly_lock.nfslock_usecnt == 0 &&
2821 lyp->nfsly_lock.nfslock_lock == 0) {
2822 NFSCL_DEBUG(4, "ret stale lay=%d\n",
2824 recallp = malloc(sizeof(*recallp),
2825 M_NFSLAYRECALL, M_NOWAIT);
2826 if (recallp == NULL)
2828 (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE,
2829 lyp, NFSLAYOUTIOMODE_ANY, 0, UINT64_MAX,
2830 lyp->nfsly_stateid.seqid, 0, 0, NULL,
2837 * Free up any unreferenced device info structures.
2839 LIST_FOREACH_SAFE(dip, &clp->nfsc_devinfo, nfsdi_list, ndip) {
2840 if (dip->nfsdi_layoutrefs == 0 &&
2841 dip->nfsdi_refcnt == 0) {
2842 NFSCL_DEBUG(4, "freeing devinfo\n");
2843 LIST_REMOVE(dip, nfsdi_list);
2844 nfscl_freedevinfo(dip);
2849 /* Do layout return(s), as required. */
2850 TAILQ_FOREACH_SAFE(lyp, &rlh, nfsly_list, nlyp) {
2851 TAILQ_REMOVE(&rlh, lyp, nfsly_list);
2852 NFSCL_DEBUG(4, "ret layout\n");
2853 nfscl_layoutreturn(clp->nfsc_nmp, lyp, cred, p);
2854 nfscl_freelayout(lyp);
2858 * Delegreturn any delegations cleaned out or recalled.
2860 TAILQ_FOREACH_SAFE(dp, &dh, nfsdl_list, ndp) {
2861 newnfs_copycred(&dp->nfsdl_cred, cred);
2862 (void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
2863 TAILQ_REMOVE(&dh, dp, nfsdl_list);
2864 free(dp, M_NFSCLDELEG);
2869 * Call nfscl_cleanupkext() once per second to check for
2870 * open/lock owners where the process has exited.
2872 mytime = NFSD_MONOSEC;
2873 if (prevsec != mytime) {
2875 nfscl_cleanupkext(clp, &lfh);
2879 * Do a ReleaseLockOwner for all lock owners where the
2880 * associated process no longer exists, as found by
2881 * nfscl_cleanupkext().
2883 newnfs_setroot(cred);
2884 SLIST_FOREACH_SAFE(lfhp, &lfh, nfslfh_list, nlfhp) {
2885 LIST_FOREACH_SAFE(lp, &lfhp->nfslfh_lock, nfsl_list,
2887 (void)nfsrpc_rellockown(clp->nfsc_nmp, lp,
2888 lfhp->nfslfh_fh, lfhp->nfslfh_len, cred,
2890 nfscl_freelockowner(lp, 0);
2897 if ((clp->nfsc_flags & NFSCLFLAGS_RECOVER) == 0)
2898 (void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT, "nfscl",
2901 if (clp->nfsc_flags & NFSCLFLAGS_UMOUNT) {
2902 clp->nfsc_flags &= ~NFSCLFLAGS_HASTHREAD;
2905 wakeup((caddr_t)clp);
2913 * Initiate state recovery. Called when NFSERR_STALECLIENTID,
2914 * NFSERR_STALESTATEID or NFSERR_BADSESSION is received.
2917 nfscl_initiate_recovery(struct nfsclclient *clp)
2923 clp->nfsc_flags |= NFSCLFLAGS_RECOVER;
2925 wakeup((caddr_t)clp);
2929 * Dump out the state stuff for debugging.
2932 nfscl_dumpstate(struct nfsmount *nmp, int openowner, int opens,
2933 int lockowner, int locks)
2935 struct nfsclclient *clp;
2936 struct nfsclowner *owp;
2937 struct nfsclopen *op;
2938 struct nfscllockowner *lp;
2939 struct nfscllock *lop;
2940 struct nfscldeleg *dp;
2944 printf("nfscl dumpstate NULL clp\n");
2948 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2949 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
2950 if (openowner && !LIST_EMPTY(&owp->nfsow_open))
2951 printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
2952 owp->nfsow_owner[0], owp->nfsow_owner[1],
2953 owp->nfsow_owner[2], owp->nfsow_owner[3],
2955 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
2957 printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
2958 op->nfso_stateid.other[0], op->nfso_stateid.other[1],
2959 op->nfso_stateid.other[2], op->nfso_opencnt,
2961 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
2963 printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
2964 lp->nfsl_owner[0], lp->nfsl_owner[1],
2965 lp->nfsl_owner[2], lp->nfsl_owner[3],
2967 lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
2968 lp->nfsl_stateid.other[2]);
2969 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
2972 printf("lck typ=%d fst=%ju end=%ju\n",
2973 lop->nfslo_type, (intmax_t)lop->nfslo_first,
2974 (intmax_t)lop->nfslo_end);
2976 printf("lck typ=%d fst=%qd end=%qd\n",
2977 lop->nfslo_type, lop->nfslo_first,
2985 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
2986 if (openowner && !LIST_EMPTY(&owp->nfsow_open))
2987 printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
2988 owp->nfsow_owner[0], owp->nfsow_owner[1],
2989 owp->nfsow_owner[2], owp->nfsow_owner[3],
2991 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
2993 printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
2994 op->nfso_stateid.other[0], op->nfso_stateid.other[1],
2995 op->nfso_stateid.other[2], op->nfso_opencnt,
2997 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
2999 printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
3000 lp->nfsl_owner[0], lp->nfsl_owner[1],
3001 lp->nfsl_owner[2], lp->nfsl_owner[3],
3003 lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
3004 lp->nfsl_stateid.other[2]);
3005 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
3008 printf("lck typ=%d fst=%ju end=%ju\n",
3009 lop->nfslo_type, (intmax_t)lop->nfslo_first,
3010 (intmax_t)lop->nfslo_end);
3012 printf("lck typ=%d fst=%qd end=%qd\n",
3013 lop->nfslo_type, lop->nfslo_first,
3024 * Check for duplicate open owners and opens.
3025 * (Only used as a diagnostic aid.)
3028 nfscl_dupopen(vnode_t vp, int dupopens)
3030 struct nfsclclient *clp;
3031 struct nfsclowner *owp, *owp2;
3032 struct nfsclopen *op, *op2;
3035 clp = VFSTONFS(vnode_mount(vp))->nm_clp;
3037 printf("nfscl dupopen NULL clp\n");
3040 nfhp = VTONFS(vp)->n_fhp;
3044 * First, search for duplicate owners.
3045 * These should never happen!
3047 LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3048 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3050 !NFSBCMP(owp->nfsow_owner, owp2->nfsow_owner,
3051 NFSV4CL_LOCKNAMELEN)) {
3053 printf("DUP OWNER\n");
3054 nfscl_dumpstate(VFSTONFS(vnode_mount(vp)), 1, 1, 0, 0);
3061 * Now, search for duplicate stateids.
3062 * These shouldn't happen, either.
3064 LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3065 LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
3066 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3067 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3069 (op->nfso_stateid.other[0] != 0 ||
3070 op->nfso_stateid.other[1] != 0 ||
3071 op->nfso_stateid.other[2] != 0) &&
3072 op->nfso_stateid.other[0] == op2->nfso_stateid.other[0] &&
3073 op->nfso_stateid.other[1] == op2->nfso_stateid.other[1] &&
3074 op->nfso_stateid.other[2] == op2->nfso_stateid.other[2]) {
3076 printf("DUP STATEID\n");
3077 nfscl_dumpstate(VFSTONFS(vnode_mount(vp)), 1, 1, 0,
3087 * Now search for duplicate opens.
3088 * Duplicate opens for the same owner
3089 * should never occur. Other duplicates are
3090 * possible and are checked for if "dupopens"
3093 LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3094 LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
3095 if (nfhp->nfh_len == op2->nfso_fhlen &&
3096 !NFSBCMP(nfhp->nfh_fh, op2->nfso_fh, nfhp->nfh_len)) {
3097 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3098 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3099 if (op != op2 && nfhp->nfh_len == op->nfso_fhlen &&
3100 !NFSBCMP(nfhp->nfh_fh, op->nfso_fh, nfhp->nfh_len) &&
3101 (!NFSBCMP(op->nfso_own->nfsow_owner,
3102 op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN) ||
3104 if (!NFSBCMP(op->nfso_own->nfsow_owner,
3105 op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
3107 printf("BADDUP OPEN\n");
3110 printf("DUP OPEN\n");
3112 nfscl_dumpstate(VFSTONFS(vnode_mount(vp)), 1, 1,
3125 * During close, find an open that needs to be dereferenced and
3126 * dereference it. If there are no more opens for this file,
3127 * log a message to that effect.
3128 * Opens aren't actually Close'd until VOP_INACTIVE() is performed
3129 * on the file's vnode.
3130 * This is the safe way, since it is difficult to identify
3131 * which open the close is for and I/O can be performed after the
3132 * close(2) system call when a file is mmap'd.
3133 * If it returns 0 for success, there will be a referenced
3134 * clp returned via clpp.
3137 nfscl_getclose(vnode_t vp, struct nfsclclient **clpp)
3139 struct nfsclclient *clp;
3140 struct nfsclowner *owp;
3141 struct nfsclopen *op;
3142 struct nfscldeleg *dp;
3146 error = nfscl_getcl(vnode_mount(vp), NULL, NULL, 1, true, &clp);
3151 nfhp = VTONFS(vp)->n_fhp;
3155 * First, look for one under a delegation that was locally issued
3156 * and just decrement the opencnt for it. Since all my Opens against
3157 * the server are DENY_NONE, I don't see a problem with hanging
3158 * onto them. (It is much easier to use one of the extant Opens
3159 * that I already have on the server when a Delegation is recalled
3160 * than to do fresh Opens.) Someday, I might need to rethink this, but.
3162 dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
3164 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
3165 op = LIST_FIRST(&owp->nfsow_open);
3168 * Since a delegation is for a file, there
3169 * should never be more than one open for
3172 if (LIST_NEXT(op, nfso_list) != NULL)
3173 panic("nfscdeleg opens");
3174 if (notdecr && op->nfso_opencnt > 0) {
3183 /* Now process the opens against the server. */
3184 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3185 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3186 if (op->nfso_fhlen == nfhp->nfh_len &&
3187 !NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
3189 /* Found an open, decrement cnt if possible */
3190 if (notdecr && op->nfso_opencnt > 0) {
3195 * There are more opens, so just return.
3197 if (op->nfso_opencnt > 0) {
3206 printf("nfscl: never fnd open\n");
3211 nfscl_doclose(vnode_t vp, struct nfsclclient **clpp, NFSPROC_T *p)
3213 struct nfsclclient *clp;
3214 struct nfsclowner *owp, *nowp;
3215 struct nfsclopen *op;
3216 struct nfscldeleg *dp;
3218 struct nfsclrecalllayout *recallp;
3221 error = nfscl_getcl(vnode_mount(vp), NULL, NULL, 1, true, &clp);
3226 nfhp = VTONFS(vp)->n_fhp;
3227 recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_WAITOK);
3230 * First get rid of the local Open structures, which should be no
3233 dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
3235 LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) {
3236 op = LIST_FIRST(&owp->nfsow_open);
3238 KASSERT((op->nfso_opencnt == 0),
3239 ("nfscl: bad open cnt on deleg"));
3240 nfscl_freeopen(op, 1);
3242 nfscl_freeopenowner(owp, 1);
3246 /* Return any layouts marked return on close. */
3247 nfscl_retoncloselayout(vp, clp, nfhp->nfh_fh, nfhp->nfh_len, &recallp);
3249 /* Now process the opens against the server. */
3251 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3252 op = LIST_FIRST(&owp->nfsow_open);
3253 while (op != NULL) {
3254 if (op->nfso_fhlen == nfhp->nfh_len &&
3255 !NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
3257 /* Found an open, close it. */
3258 KASSERT((op->nfso_opencnt == 0),
3259 ("nfscl: bad open cnt on server"));
3261 nfsrpc_doclose(VFSTONFS(vnode_mount(vp)), op,
3266 op = LIST_NEXT(op, nfso_list);
3271 * recallp has been set NULL by nfscl_retoncloselayout() if it was
3272 * used by the function, but calling free() with a NULL pointer is ok.
3274 free(recallp, M_NFSLAYRECALL);
3279 * Return all delegations on this client.
3280 * (Must be called with client sleep lock.)
3283 nfscl_delegreturnall(struct nfsclclient *clp, NFSPROC_T *p)
3285 struct nfscldeleg *dp, *ndp;
3288 cred = newnfs_getcred();
3289 TAILQ_FOREACH_SAFE(dp, &clp->nfsc_deleg, nfsdl_list, ndp) {
3290 nfscl_cleandeleg(dp);
3291 (void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
3292 nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
3298 * Return any delegation for this vp.
3301 nfscl_delegreturnvp(vnode_t vp, NFSPROC_T *p)
3303 struct nfsclclient *clp;
3304 struct nfscldeleg *dp;
3309 cred = newnfs_getcred();
3312 clp = VFSTONFS(vp->v_mount)->nm_clp;
3314 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
3315 np->n_fhp->nfh_len);
3317 nfscl_cleandeleg(dp);
3318 nfscl_freedeleg(&clp->nfsc_deleg, dp, false);
3320 newnfs_copycred(&dp->nfsdl_cred, cred);
3321 nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
3322 free(dp, M_NFSCLDELEG);
3329 * Do a callback RPC.
3332 nfscl_docb(struct nfsrv_descript *nd, NFSPROC_T *p)
3334 int clist, gotseq_ok, i, j, k, op, rcalls;
3336 struct nfsclclient *clp;
3337 struct nfscldeleg *dp = NULL;
3338 int numops, taglen = -1, error = 0, trunc __unused;
3339 u_int32_t minorvers = 0, retops = 0, *retopsp = NULL, *repp, cbident;
3340 u_char tag[NFSV4_SMALLSTR + 1], *tagstr;
3346 nfsattrbit_t attrbits, rattrbits;
3347 nfsv4stateid_t stateid;
3348 uint32_t seqid, slotid = 0, highslot, cachethis __unused;
3349 uint8_t sessionid[NFSX_V4SESSIONID];
3351 struct nfscllayout *lyp;
3352 uint64_t filesid[2], len, off;
3353 int changed, gotone, laytype, recalltype;
3355 struct nfsclrecalllayout *recallp = NULL;
3356 struct nfsclsession *tsep;
3360 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3361 taglen = fxdr_unsigned(int, *tl);
3366 if (taglen <= NFSV4_SMALLSTR)
3369 tagstr = malloc(taglen + 1, M_TEMP, M_WAITOK);
3370 error = nfsrv_mtostr(nd, tagstr, taglen);
3372 if (taglen > NFSV4_SMALLSTR)
3373 free(tagstr, M_TEMP);
3377 (void) nfsm_strtom(nd, tag, taglen);
3378 if (taglen > NFSV4_SMALLSTR) {
3379 free(tagstr, M_TEMP);
3381 NFSM_BUILD(retopsp, u_int32_t *, NFSX_UNSIGNED);
3382 NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3383 minorvers = fxdr_unsigned(u_int32_t, *tl++);
3384 if (minorvers != NFSV4_MINORVERSION && minorvers != NFSV41_MINORVERSION)
3385 nd->nd_repstat = NFSERR_MINORVERMISMATCH;
3386 cbident = fxdr_unsigned(u_int32_t, *tl++);
3390 numops = fxdr_unsigned(int, *tl);
3392 * Loop around doing the sub ops.
3394 for (i = 0; i < numops; i++) {
3395 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3396 NFSM_BUILD(repp, u_int32_t *, 2 * NFSX_UNSIGNED);
3398 op = fxdr_unsigned(int, *tl);
3399 if (op < NFSV4OP_CBGETATTR ||
3400 (op > NFSV4OP_CBRECALL && minorvers == NFSV4_MINORVERSION) ||
3401 (op > NFSV4OP_CBNOTIFYDEVID &&
3402 minorvers == NFSV41_MINORVERSION)) {
3403 nd->nd_repstat = NFSERR_OPILLEGAL;
3404 *repp = nfscl_errmap(nd, minorvers);
3408 nd->nd_procnum = op;
3409 if (op < NFSV41_CBNOPS)
3410 nfsstatsv1.cbrpccnt[nd->nd_procnum]++;
3412 case NFSV4OP_CBGETATTR:
3413 NFSCL_DEBUG(4, "cbgetattr\n");
3416 error = nfsm_getfh(nd, &nfhp);
3418 error = nfsrv_getattrbits(nd, &attrbits,
3420 if (error == 0 && i == 0 &&
3421 minorvers != NFSV4_MINORVERSION)
3422 error = NFSERR_OPNOTINSESS;
3424 mp = nfscl_getmnt(minorvers, sessionid, cbident,
3427 error = NFSERR_SERVERFAULT;
3430 error = nfscl_ngetreopen(mp, nfhp->nfh_fh,
3431 nfhp->nfh_len, p, &np);
3436 NFSZERO_ATTRBIT(&rattrbits);
3438 dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
3441 if (NFSISSET_ATTRBIT(&attrbits,
3444 va.va_size = np->n_size;
3448 NFSSETBIT_ATTRBIT(&rattrbits,
3451 if (NFSISSET_ATTRBIT(&attrbits,
3452 NFSATTRBIT_CHANGE)) {
3456 (np->n_flag & NDELEGMOD))
3458 NFSSETBIT_ATTRBIT(&rattrbits,
3462 error = NFSERR_SERVERFAULT;
3470 free(nfhp, M_NFSFH);
3472 (void) nfsv4_fillattr(nd, NULL, NULL, NULL, &va,
3473 NULL, 0, &rattrbits, NULL, p, 0, 0, 0, 0,
3476 case NFSV4OP_CBRECALL:
3477 NFSCL_DEBUG(4, "cbrecall\n");
3478 NFSM_DISSECT(tl, u_int32_t *, NFSX_STATEID +
3480 stateid.seqid = *tl++;
3481 NFSBCOPY((caddr_t)tl, (caddr_t)stateid.other,
3483 tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
3484 trunc = fxdr_unsigned(int, *tl);
3485 error = nfsm_getfh(nd, &nfhp);
3486 if (error == 0 && i == 0 &&
3487 minorvers != NFSV4_MINORVERSION)
3488 error = NFSERR_OPNOTINSESS;
3491 if (minorvers == NFSV4_MINORVERSION)
3492 clp = nfscl_getclnt(cbident);
3494 clp = nfscl_getclntsess(sessionid);
3496 dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
3498 if (dp != NULL && (dp->nfsdl_flags &
3499 NFSCLDL_DELEGRET) == 0) {
3502 wakeup((caddr_t)clp);
3505 error = NFSERR_SERVERFAULT;
3510 free(nfhp, M_NFSFH);
3512 case NFSV4OP_CBLAYOUTRECALL:
3513 NFSCL_DEBUG(4, "cblayrec\n");
3515 NFSM_DISSECT(tl, uint32_t *, 4 * NFSX_UNSIGNED);
3516 laytype = fxdr_unsigned(int, *tl++);
3517 iomode = fxdr_unsigned(uint32_t, *tl++);
3518 if (newnfs_true == *tl++)
3522 recalltype = fxdr_unsigned(int, *tl);
3523 NFSCL_DEBUG(4, "layt=%d iom=%d ch=%d rectyp=%d\n",
3524 laytype, iomode, changed, recalltype);
3525 recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL,
3527 if (laytype != NFSLAYOUT_NFSV4_1_FILES &&
3528 laytype != NFSLAYOUT_FLEXFILE)
3529 error = NFSERR_NOMATCHLAYOUT;
3530 else if (recalltype == NFSLAYOUTRETURN_FILE) {
3531 error = nfsm_getfh(nd, &nfhp);
3532 NFSCL_DEBUG(4, "retfile getfh=%d\n", error);
3535 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_HYPER +
3537 off = fxdr_hyper(tl); tl += 2;
3538 len = fxdr_hyper(tl); tl += 2;
3539 stateid.seqid = fxdr_unsigned(uint32_t, *tl++);
3540 NFSBCOPY(tl, stateid.other, NFSX_STATEIDOTHER);
3541 if (minorvers == NFSV4_MINORVERSION)
3542 error = NFSERR_NOTSUPP;
3544 error = NFSERR_OPNOTINSESS;
3545 NFSCL_DEBUG(4, "off=%ju len=%ju sq=%u err=%d\n",
3546 (uintmax_t)off, (uintmax_t)len,
3547 stateid.seqid, error);
3550 clp = nfscl_getclntsess(sessionid);
3551 NFSCL_DEBUG(4, "cbly clp=%p\n", clp);
3553 lyp = nfscl_findlayout(clp,
3556 NFSCL_DEBUG(4, "cblyp=%p\n",
3561 NFSLY_FLEXFILE)) != 0 &&
3562 !NFSBCMP(stateid.other,
3563 lyp->nfsly_stateid.other,
3564 NFSX_STATEIDOTHER)) {
3574 lyp->nfsly_stateid.seqid)
3575 lyp->nfsly_stateid.seqid =
3583 lyp->nfsly_stateid.seqid);
3586 NFSERR_NOMATCHLAYOUT;
3588 error = NFSERR_NOMATCHLAYOUT;
3591 free(nfhp, M_NFSFH);
3592 } else if (recalltype == NFSLAYOUTRETURN_FSID) {
3593 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_HYPER);
3594 filesid[0] = fxdr_hyper(tl); tl += 2;
3595 filesid[1] = fxdr_hyper(tl); tl += 2;
3598 clp = nfscl_getclntsess(sessionid);
3600 TAILQ_FOREACH(lyp, &clp->nfsc_layout,
3602 if (lyp->nfsly_filesid[0] ==
3604 lyp->nfsly_filesid[1] ==
3611 lyp->nfsly_stateid.seqid,
3621 error = NFSERR_NOMATCHLAYOUT;
3623 error = NFSERR_NOMATCHLAYOUT;
3625 } else if (recalltype == NFSLAYOUTRETURN_ALL) {
3628 clp = nfscl_getclntsess(sessionid);
3630 TAILQ_FOREACH(lyp, &clp->nfsc_layout,
3632 error = nfscl_layoutrecall(
3633 recalltype, lyp, iomode, 0,
3635 lyp->nfsly_stateid.seqid,
3636 0, 0, NULL, recallp);
3643 error = NFSERR_NOMATCHLAYOUT;
3645 error = NFSERR_NOMATCHLAYOUT;
3648 error = NFSERR_NOMATCHLAYOUT;
3649 if (recallp != NULL) {
3650 free(recallp, M_NFSLAYRECALL);
3654 case NFSV4OP_CBSEQUENCE:
3655 NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID +
3657 bcopy(tl, sessionid, NFSX_V4SESSIONID);
3658 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3659 seqid = fxdr_unsigned(uint32_t, *tl++);
3660 slotid = fxdr_unsigned(uint32_t, *tl++);
3661 highslot = fxdr_unsigned(uint32_t, *tl++);
3663 /* Throw away the referring call stuff. */
3664 clist = fxdr_unsigned(int, *tl);
3665 for (j = 0; j < clist; j++) {
3666 NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID +
3668 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3669 rcalls = fxdr_unsigned(int, *tl);
3670 for (k = 0; k < rcalls; k++) {
3671 NFSM_DISSECT(tl, uint32_t *,
3677 clp = nfscl_getclntsess(sessionid);
3679 error = NFSERR_SERVERFAULT;
3681 error = NFSERR_SEQUENCEPOS;
3683 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3684 error = nfsv4_seqsession(seqid, slotid,
3685 highslot, tsep->nfsess_cbslots, &rep,
3686 tsep->nfsess_backslots);
3689 if (error == 0 || error == NFSERR_REPLYFROMCACHE) {
3693 * Handle a reply for a retried
3694 * callback. The reply will be
3695 * re-inserted in the session cache
3696 * by the nfsv4_seqsess_cacherep() call
3699 KASSERT(error == NFSERR_REPLYFROMCACHE,
3700 ("cbsequence: non-NULL rep"));
3701 NFSCL_DEBUG(4, "Got cbretry\n");
3702 m_freem(nd->nd_mreq);
3707 NFSM_BUILD(tl, uint32_t *,
3708 NFSX_V4SESSIONID + 4 * NFSX_UNSIGNED);
3709 bcopy(sessionid, tl, NFSX_V4SESSIONID);
3710 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3711 *tl++ = txdr_unsigned(seqid);
3712 *tl++ = txdr_unsigned(slotid);
3713 *tl++ = txdr_unsigned(NFSV4_CBSLOTS - 1);
3714 *tl = txdr_unsigned(NFSV4_CBSLOTS - 1);
3718 if (i == 0 && minorvers == NFSV41_MINORVERSION)
3719 error = NFSERR_OPNOTINSESS;
3721 NFSCL_DEBUG(1, "unsupp callback %d\n", op);
3722 error = NFSERR_NOTSUPP;
3727 if (error == EBADRPC || error == NFSERR_BADXDR) {
3728 nd->nd_repstat = NFSERR_BADXDR;
3730 nd->nd_repstat = error;
3735 if (nd->nd_repstat) {
3736 *repp = nfscl_errmap(nd, minorvers);
3739 *repp = 0; /* NFS4_OK */
3742 if (recallp != NULL)
3743 free(recallp, M_NFSLAYRECALL);
3745 if (error == EBADRPC || error == NFSERR_BADXDR)
3746 nd->nd_repstat = NFSERR_BADXDR;
3748 printf("nfsv4 comperr1=%d\n", error);
3751 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
3755 *retopsp = txdr_unsigned(retops);
3757 *nd->nd_errp = nfscl_errmap(nd, minorvers);
3759 if (gotseq_ok != 0) {
3760 rep = m_copym(nd->nd_mreq, 0, M_COPYALL, M_WAITOK);
3762 clp = nfscl_getclntsess(sessionid);
3764 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3765 nfsv4_seqsess_cacherep(slotid, tsep->nfsess_cbslots,
3776 * Generate the next cbident value. Basically just increment a static value
3777 * and then check that it isn't already in the list, if it has wrapped around.
3780 nfscl_nextcbident(void)
3782 struct nfsclclient *clp;
3784 static u_int32_t nextcbident = 0;
3785 static int haswrapped = 0;
3788 if (nextcbident == 0)
3792 * Search the clientid list for one already using this cbident.
3797 LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3798 if (clp->nfsc_cbident == nextcbident) {
3808 return (nextcbident);
3812 * Get the mount point related to a given cbident or session and busy it.
3815 nfscl_getmnt(int minorvers, uint8_t *sessionid, u_int32_t cbident,
3816 struct nfsclclient **clpp)
3818 struct nfsclclient *clp;
3821 struct nfsclsession *tsep;
3825 LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3826 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3827 if (minorvers == NFSV4_MINORVERSION) {
3828 if (clp->nfsc_cbident == cbident)
3830 } else if (!NFSBCMP(tsep->nfsess_sessionid, sessionid,
3838 mp = clp->nfsc_nmp->nm_mountp;
3841 error = vfs_busy(mp, 0);
3850 * Get the clientid pointer related to a given cbident.
3852 static struct nfsclclient *
3853 nfscl_getclnt(u_int32_t cbident)
3855 struct nfsclclient *clp;
3857 LIST_FOREACH(clp, &nfsclhead, nfsc_list)
3858 if (clp->nfsc_cbident == cbident)
3864 * Get the clientid pointer related to a given sessionid.
3866 static struct nfsclclient *
3867 nfscl_getclntsess(uint8_t *sessionid)
3869 struct nfsclclient *clp;
3870 struct nfsclsession *tsep;
3872 LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3873 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3874 if (!NFSBCMP(tsep->nfsess_sessionid, sessionid,
3882 * Search for a lock conflict locally on the client. A conflict occurs if
3883 * - not same owner and overlapping byte range and at least one of them is
3884 * a write lock or this is an unlock.
3887 nfscl_localconflict(struct nfsclclient *clp, u_int8_t *fhp, int fhlen,
3888 struct nfscllock *nlop, u_int8_t *own, struct nfscldeleg *dp,
3889 struct nfscllock **lopp)
3891 struct nfsclowner *owp;
3892 struct nfsclopen *op;
3896 ret = nfscl_checkconflict(&dp->nfsdl_lock, nlop, own, lopp);
3900 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3901 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3902 if (op->nfso_fhlen == fhlen &&
3903 !NFSBCMP(op->nfso_fh, fhp, fhlen)) {
3904 ret = nfscl_checkconflict(&op->nfso_lock, nlop,
3915 nfscl_checkconflict(struct nfscllockownerhead *lhp, struct nfscllock *nlop,
3916 u_int8_t *own, struct nfscllock **lopp)
3918 struct nfscllockowner *lp;
3919 struct nfscllock *lop;
3921 LIST_FOREACH(lp, lhp, nfsl_list) {
3922 if (NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
3923 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
3924 if (lop->nfslo_first >= nlop->nfslo_end)
3926 if (lop->nfslo_end <= nlop->nfslo_first)
3928 if (lop->nfslo_type == F_WRLCK ||
3929 nlop->nfslo_type == F_WRLCK ||
3930 nlop->nfslo_type == F_UNLCK) {
3933 return (NFSERR_DENIED);
3942 * Check for a local conflicting lock.
3945 nfscl_lockt(vnode_t vp, struct nfsclclient *clp, u_int64_t off,
3946 u_int64_t len, struct flock *fl, NFSPROC_T *p, void *id, int flags)
3948 struct nfscllock *lop, nlck;
3949 struct nfscldeleg *dp;
3951 u_int8_t own[NFSV4CL_LOCKNAMELEN];
3954 nlck.nfslo_type = fl->l_type;
3955 nlck.nfslo_first = off;
3956 if (len == NFS64BITSSET) {
3957 nlck.nfslo_end = NFS64BITSSET;
3959 nlck.nfslo_end = off + len;
3960 if (nlck.nfslo_end <= nlck.nfslo_first)
3961 return (NFSERR_INVAL);
3964 nfscl_filllockowner(id, own, flags);
3966 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
3967 error = nfscl_localconflict(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len,
3968 &nlck, own, dp, &lop);
3970 fl->l_whence = SEEK_SET;
3971 fl->l_start = lop->nfslo_first;
3972 if (lop->nfslo_end == NFS64BITSSET)
3975 fl->l_len = lop->nfslo_end - lop->nfslo_first;
3976 fl->l_pid = (pid_t)0;
3977 fl->l_type = lop->nfslo_type;
3978 error = -1; /* no RPC required */
3979 } else if (dp != NULL && ((dp->nfsdl_flags & NFSCLDL_WRITE) ||
3980 fl->l_type == F_RDLCK)) {
3982 * The delegation ensures that there isn't a conflicting
3983 * lock on the server, so return -1 to indicate an RPC
3986 fl->l_type = F_UNLCK;
3994 * Handle Recall of a delegation.
3995 * The clp must be exclusive locked when this is called.
3998 nfscl_recalldeleg(struct nfsclclient *clp, struct nfsmount *nmp,
3999 struct nfscldeleg *dp, vnode_t vp, struct ucred *cred, NFSPROC_T *p,
4000 int called_from_renewthread, vnode_t *vpp)
4002 struct nfsclowner *owp, *lowp, *nowp;
4003 struct nfsclopen *op, *lop;
4004 struct nfscllockowner *lp;
4005 struct nfscllock *lckp;
4010 KASSERT(vpp != NULL, ("nfscl_recalldeleg: vpp NULL"));
4013 * First, get a vnode for the file. This is needed to do RPCs.
4015 ret = nfscl_ngetreopen(nmp->nm_mountp, dp->nfsdl_fh,
4016 dp->nfsdl_fhlen, p, &np);
4019 * File isn't open, so nothing to move over to the
4029 dp->nfsdl_flags &= ~NFSCLDL_MODTIMESET;
4032 * Ok, if it's a write delegation, flush data to the server, so
4033 * that close/open consistency is retained.
4037 if ((dp->nfsdl_flags & NFSCLDL_WRITE) && (np->n_flag & NMODIFIED)) {
4038 np->n_flag |= NDELEGRECALL;
4040 ret = ncl_flush(vp, MNT_WAIT, p, 1, called_from_renewthread);
4042 np->n_flag &= ~NDELEGRECALL;
4044 NFSINVALATTRCACHE(np);
4046 if (ret == EIO && called_from_renewthread != 0) {
4048 * If the flush failed with EIO for the renew thread,
4049 * return now, so that the dirty buffer will be flushed
4056 * Now, for each openowner with opens issued locally, move them
4057 * over to state against the server.
4059 LIST_FOREACH(lowp, &dp->nfsdl_owner, nfsow_list) {
4060 lop = LIST_FIRST(&lowp->nfsow_open);
4062 if (LIST_NEXT(lop, nfso_list) != NULL)
4063 panic("nfsdlg mult opens");
4065 * Look for the same openowner against the server.
4067 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
4068 if (!NFSBCMP(lowp->nfsow_owner,
4069 owp->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
4070 newnfs_copycred(&dp->nfsdl_cred, cred);
4071 ret = nfscl_moveopen(vp, clp, nmp, lop,
4073 if (ret == NFSERR_STALECLIENTID ||
4074 ret == NFSERR_STALEDONTRECOVER ||
4075 ret == NFSERR_BADSESSION)
4078 nfscl_freeopen(lop, 1);
4087 * If no openowner found, create one and get an open
4092 sizeof (struct nfsclowner), M_NFSCLOWNER,
4094 nfscl_newopen(clp, NULL, &owp, &nowp, &op,
4095 NULL, lowp->nfsow_owner, dp->nfsdl_fh,
4096 dp->nfsdl_fhlen, NULL, NULL);
4097 newnfs_copycred(&dp->nfsdl_cred, cred);
4098 ret = nfscl_moveopen(vp, clp, nmp, lop,
4101 nfscl_freeopenowner(owp, 0);
4102 if (ret == NFSERR_STALECLIENTID ||
4103 ret == NFSERR_STALEDONTRECOVER ||
4104 ret == NFSERR_BADSESSION)
4107 nfscl_freeopen(lop, 1);
4117 * Now, get byte range locks for any locks done locally.
4119 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4120 LIST_FOREACH(lckp, &lp->nfsl_lock, nfslo_list) {
4121 newnfs_copycred(&dp->nfsdl_cred, cred);
4122 ret = nfscl_relock(vp, clp, nmp, lp, lckp, cred, p);
4123 if (ret == NFSERR_STALESTATEID ||
4124 ret == NFSERR_STALEDONTRECOVER ||
4125 ret == NFSERR_STALECLIENTID ||
4126 ret == NFSERR_BADSESSION)
4136 * Move a locally issued open over to an owner on the state list.
4137 * SIDE EFFECT: If it needs to sleep (do an rpc), it unlocks clstate and
4138 * returns with it unlocked.
4141 nfscl_moveopen(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp,
4142 struct nfsclopen *lop, struct nfsclowner *owp, struct nfscldeleg *dp,
4143 struct ucred *cred, NFSPROC_T *p)
4145 struct nfsclopen *op, *nop;
4146 struct nfscldeleg *ndp;
4148 int error = 0, newone;
4151 * First, look for an appropriate open, If found, just increment the
4154 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
4155 if ((op->nfso_mode & lop->nfso_mode) == lop->nfso_mode &&
4156 op->nfso_fhlen == lop->nfso_fhlen &&
4157 !NFSBCMP(op->nfso_fh, lop->nfso_fh, op->nfso_fhlen)) {
4158 op->nfso_opencnt += lop->nfso_opencnt;
4159 nfscl_freeopen(lop, 1);
4164 /* No appropriate open, so we have to do one against the server. */
4166 nop = malloc(sizeof (struct nfsclopen) +
4167 lop->nfso_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
4169 nfscl_newopen(clp, NULL, &owp, NULL, &op, &nop, owp->nfsow_owner,
4170 lop->nfso_fh, lop->nfso_fhlen, cred, &newone);
4172 error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data, np->n_v4->n4_fhlen,
4173 lop->nfso_fh, lop->nfso_fhlen, lop->nfso_mode, op,
4174 NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, &ndp, 0, 0, cred, p);
4177 nfscl_freeopen(op, 0);
4179 op->nfso_mode |= lop->nfso_mode;
4180 op->nfso_opencnt += lop->nfso_opencnt;
4181 nfscl_freeopen(lop, 1);
4184 free(nop, M_NFSCLOPEN);
4187 * What should I do with the returned delegation, since the
4188 * delegation is being recalled? For now, just printf and
4191 printf("Moveopen returned deleg\n");
4192 free(ndp, M_NFSCLDELEG);
4198 * Recall all delegations on this client.
4201 nfscl_totalrecall(struct nfsclclient *clp)
4203 struct nfscldeleg *dp;
4205 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
4206 if ((dp->nfsdl_flags & NFSCLDL_DELEGRET) == 0)
4207 dp->nfsdl_flags |= NFSCLDL_RECALL;
4212 * Relock byte ranges. Called for delegation recall and state expiry.
4215 nfscl_relock(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp,
4216 struct nfscllockowner *lp, struct nfscllock *lop, struct ucred *cred,
4219 struct nfscllockowner *nlp;
4222 int error, newone, donelocally;
4224 off = lop->nfslo_first;
4225 len = lop->nfslo_end - lop->nfslo_first;
4226 error = nfscl_getbytelock(vp, off, len, lop->nfslo_type, cred, p,
4227 clp, 1, NULL, lp->nfsl_lockflags, lp->nfsl_owner,
4228 lp->nfsl_openowner, &nlp, &newone, &donelocally);
4229 if (error || donelocally)
4231 nfhp = VTONFS(vp)->n_fhp;
4232 error = nfscl_trylock(nmp, vp, nfhp->nfh_fh,
4233 nfhp->nfh_len, nlp, newone, 0, off,
4234 len, lop->nfslo_type, cred, p);
4236 nfscl_freelockowner(nlp, 0);
4241 * Called to re-open a file. Basically get a vnode for the file handle
4242 * and then call nfsrpc_openrpc() to do the rest.
4245 nfsrpc_reopen(struct nfsmount *nmp, u_int8_t *fhp, int fhlen,
4246 u_int32_t mode, struct nfsclopen *op, struct nfscldeleg **dpp,
4247 struct ucred *cred, NFSPROC_T *p)
4253 error = nfscl_ngetreopen(nmp->nm_mountp, fhp, fhlen, p, &np);
4257 if (np->n_v4 != NULL) {
4258 error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data,
4259 np->n_v4->n4_fhlen, fhp, fhlen, mode, op,
4260 NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, dpp, 0, 0,
4270 * Try an open against the server. Just call nfsrpc_openrpc(), retrying while
4271 * NFSERR_DELAY. Also, try system credentials, if the passed in credentials
4275 nfscl_tryopen(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen,
4276 u_int8_t *newfhp, int newfhlen, u_int32_t mode, struct nfsclopen *op,
4277 u_int8_t *name, int namelen, struct nfscldeleg **ndpp,
4278 int reclaim, u_int32_t delegtype, struct ucred *cred, NFSPROC_T *p)
4283 error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp, newfhlen,
4284 mode, op, name, namelen, ndpp, reclaim, delegtype, cred, p,
4286 if (error == NFSERR_DELAY)
4287 (void) nfs_catnap(PZERO, error, "nfstryop");
4288 } while (error == NFSERR_DELAY);
4289 if (error == EAUTH || error == EACCES) {
4290 /* Try again using system credentials */
4291 newnfs_setroot(cred);
4293 error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp,
4294 newfhlen, mode, op, name, namelen, ndpp, reclaim,
4295 delegtype, cred, p, 1, 0);
4296 if (error == NFSERR_DELAY)
4297 (void) nfs_catnap(PZERO, error, "nfstryop");
4298 } while (error == NFSERR_DELAY);
4304 * Try a byte range lock. Just loop on nfsrpc_lock() while it returns
4305 * NFSERR_DELAY. Also, retry with system credentials, if the provided
4309 nfscl_trylock(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp,
4310 int fhlen, struct nfscllockowner *nlp, int newone, int reclaim,
4311 u_int64_t off, u_int64_t len, short type, struct ucred *cred, NFSPROC_T *p)
4313 struct nfsrv_descript nfsd, *nd = &nfsd;
4317 error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp, newone,
4318 reclaim, off, len, type, cred, p, 0);
4319 if (!error && nd->nd_repstat == NFSERR_DELAY)
4320 (void) nfs_catnap(PZERO, (int)nd->nd_repstat,
4322 } while (!error && nd->nd_repstat == NFSERR_DELAY);
4324 error = nd->nd_repstat;
4325 if (error == EAUTH || error == EACCES) {
4326 /* Try again using root credentials */
4327 newnfs_setroot(cred);
4329 error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp,
4330 newone, reclaim, off, len, type, cred, p, 1);
4331 if (!error && nd->nd_repstat == NFSERR_DELAY)
4332 (void) nfs_catnap(PZERO, (int)nd->nd_repstat,
4334 } while (!error && nd->nd_repstat == NFSERR_DELAY);
4336 error = nd->nd_repstat;
4342 * Try a delegreturn against the server. Just call nfsrpc_delegreturn(),
4343 * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
4347 nfscl_trydelegreturn(struct nfscldeleg *dp, struct ucred *cred,
4348 struct nfsmount *nmp, NFSPROC_T *p)
4353 error = nfsrpc_delegreturn(dp, cred, nmp, p, 0);
4354 if (error == NFSERR_DELAY)
4355 (void) nfs_catnap(PZERO, error, "nfstrydp");
4356 } while (error == NFSERR_DELAY);
4357 if (error == EAUTH || error == EACCES) {
4358 /* Try again using system credentials */
4359 newnfs_setroot(cred);
4361 error = nfsrpc_delegreturn(dp, cred, nmp, p, 1);
4362 if (error == NFSERR_DELAY)
4363 (void) nfs_catnap(PZERO, error, "nfstrydp");
4364 } while (error == NFSERR_DELAY);
4370 * Try a close against the server. Just call nfsrpc_closerpc(),
4371 * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
4375 nfscl_tryclose(struct nfsclopen *op, struct ucred *cred,
4376 struct nfsmount *nmp, NFSPROC_T *p)
4378 struct nfsrv_descript nfsd, *nd = &nfsd;
4382 error = nfsrpc_closerpc(nd, nmp, op, cred, p, 0);
4383 if (error == NFSERR_DELAY)
4384 (void) nfs_catnap(PZERO, error, "nfstrycl");
4385 } while (error == NFSERR_DELAY);
4386 if (error == EAUTH || error == EACCES) {
4387 /* Try again using system credentials */
4388 newnfs_setroot(cred);
4390 error = nfsrpc_closerpc(nd, nmp, op, cred, p, 1);
4391 if (error == NFSERR_DELAY)
4392 (void) nfs_catnap(PZERO, error, "nfstrycl");
4393 } while (error == NFSERR_DELAY);
4399 * Decide if a delegation on a file permits close without flushing writes
4400 * to the server. This might be a big performance win in some environments.
4401 * (Not useful until the client does caching on local stable storage.)
4404 nfscl_mustflush(vnode_t vp)
4406 struct nfsclclient *clp;
4407 struct nfscldeleg *dp;
4409 struct nfsmount *nmp;
4412 nmp = VFSTONFS(vnode_mount(vp));
4413 if (!NFSHASNFSV4(nmp))
4416 clp = nfscl_findcl(nmp);
4421 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4422 if (dp != NULL && (dp->nfsdl_flags &
4423 (NFSCLDL_WRITE | NFSCLDL_RECALL | NFSCLDL_DELEGRET)) ==
4425 (dp->nfsdl_sizelimit >= np->n_size ||
4426 !NFSHASSTRICT3530(nmp))) {
4435 * See if a (write) delegation exists for this file.
4438 nfscl_nodeleg(vnode_t vp, int writedeleg)
4440 struct nfsclclient *clp;
4441 struct nfscldeleg *dp;
4443 struct nfsmount *nmp;
4446 nmp = VFSTONFS(vnode_mount(vp));
4447 if (!NFSHASNFSV4(nmp))
4450 clp = nfscl_findcl(nmp);
4455 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4457 (dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_DELEGRET)) == 0 &&
4458 (writedeleg == 0 || (dp->nfsdl_flags & NFSCLDL_WRITE) ==
4468 * Look for an associated delegation that should be DelegReturned.
4471 nfscl_removedeleg(vnode_t vp, NFSPROC_T *p, nfsv4stateid_t *stp)
4473 struct nfsclclient *clp;
4474 struct nfscldeleg *dp;
4475 struct nfsclowner *owp;
4476 struct nfscllockowner *lp;
4477 struct nfsmount *nmp;
4480 int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
4482 nmp = VFSTONFS(vnode_mount(vp));
4486 * Loop around waiting for:
4487 * - outstanding I/O operations on delegations to complete
4488 * - for a delegation on vp that has state, lock the client and
4490 * - return delegation with no state
4493 clp = nfscl_findcl(nmp);
4498 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4499 np->n_fhp->nfh_len);
4502 * Wait for outstanding I/O ops to be done.
4504 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4506 nfsv4_unlock(&clp->nfsc_lock, 0);
4509 dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4510 (void) nfsmsleep(&dp->nfsdl_rwlock,
4511 NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4515 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4516 if (!LIST_EMPTY(&owp->nfsow_open)) {
4522 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4523 if (!LIST_EMPTY(&lp->nfsl_lock)) {
4529 if (needsrecall && !triedrecall) {
4530 dp->nfsdl_flags |= NFSCLDL_DELEGRET;
4533 igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
4534 &islept, NFSCLSTATEMUTEXPTR, NULL);
4541 cred = newnfs_getcred();
4542 newnfs_copycred(&dp->nfsdl_cred, cred);
4543 nfscl_recalldeleg(clp, nmp, dp, vp, cred, p, 0, NULL);
4547 nfsv4_unlock(&clp->nfsc_lock, 0);
4551 *stp = dp->nfsdl_stateid;
4553 nfscl_cleandeleg(dp);
4554 nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
4557 nfsv4_unlock(&clp->nfsc_lock, 0);
4564 * Look for associated delegation(s) that should be DelegReturned.
4567 nfscl_renamedeleg(vnode_t fvp, nfsv4stateid_t *fstp, int *gotfdp, vnode_t tvp,
4568 nfsv4stateid_t *tstp, int *gottdp, NFSPROC_T *p)
4570 struct nfsclclient *clp;
4571 struct nfscldeleg *dp;
4572 struct nfsclowner *owp;
4573 struct nfscllockowner *lp;
4574 struct nfsmount *nmp;
4577 int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
4579 nmp = VFSTONFS(vnode_mount(fvp));
4584 * Loop around waiting for:
4585 * - outstanding I/O operations on delegations to complete
4586 * - for a delegation on fvp that has state, lock the client and
4588 * - return delegation(s) with no state.
4591 clp = nfscl_findcl(nmp);
4597 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4598 np->n_fhp->nfh_len);
4599 if (dp != NULL && *gotfdp == 0) {
4601 * Wait for outstanding I/O ops to be done.
4603 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4605 nfsv4_unlock(&clp->nfsc_lock, 0);
4608 dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4609 (void) nfsmsleep(&dp->nfsdl_rwlock,
4610 NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4614 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4615 if (!LIST_EMPTY(&owp->nfsow_open)) {
4621 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4622 if (!LIST_EMPTY(&lp->nfsl_lock)) {
4628 if (needsrecall && !triedrecall) {
4629 dp->nfsdl_flags |= NFSCLDL_DELEGRET;
4632 igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
4633 &islept, NFSCLSTATEMUTEXPTR, NULL);
4640 cred = newnfs_getcred();
4641 newnfs_copycred(&dp->nfsdl_cred, cred);
4642 nfscl_recalldeleg(clp, nmp, dp, fvp, cred, p, 0, NULL);
4646 nfsv4_unlock(&clp->nfsc_lock, 0);
4650 *fstp = dp->nfsdl_stateid;
4653 nfscl_cleandeleg(dp);
4654 nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
4657 nfsv4_unlock(&clp->nfsc_lock, 0);
4662 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4663 np->n_fhp->nfh_len);
4664 if (dp != NULL && *gottdp == 0) {
4666 * Wait for outstanding I/O ops to be done.
4668 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4669 dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4670 (void) nfsmsleep(&dp->nfsdl_rwlock,
4671 NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4674 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4675 if (!LIST_EMPTY(&owp->nfsow_open)) {
4680 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4681 if (!LIST_EMPTY(&lp->nfsl_lock)) {
4686 *tstp = dp->nfsdl_stateid;
4689 nfscl_cleandeleg(dp);
4690 nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
4699 * Get a reference on the clientid associated with the mount point.
4700 * Return 1 if success, 0 otherwise.
4703 nfscl_getref(struct nfsmount *nmp)
4705 struct nfsclclient *clp;
4708 clp = nfscl_findcl(nmp);
4713 nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, NULL);
4719 * Release a reference on a clientid acquired with the above call.
4722 nfscl_relref(struct nfsmount *nmp)
4724 struct nfsclclient *clp;
4727 clp = nfscl_findcl(nmp);
4732 nfsv4_relref(&clp->nfsc_lock);
4737 * Save the size attribute in the delegation, since the nfsnode
4741 nfscl_reclaimnode(vnode_t vp)
4743 struct nfsclclient *clp;
4744 struct nfscldeleg *dp;
4745 struct nfsnode *np = VTONFS(vp);
4746 struct nfsmount *nmp;
4748 nmp = VFSTONFS(vnode_mount(vp));
4749 if (!NFSHASNFSV4(nmp))
4752 clp = nfscl_findcl(nmp);
4757 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4758 if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE))
4759 dp->nfsdl_size = np->n_size;
4764 * Get the saved size attribute in the delegation, since it is a
4765 * newly allocated nfsnode.
4768 nfscl_newnode(vnode_t vp)
4770 struct nfsclclient *clp;
4771 struct nfscldeleg *dp;
4772 struct nfsnode *np = VTONFS(vp);
4773 struct nfsmount *nmp;
4775 nmp = VFSTONFS(vnode_mount(vp));
4776 if (!NFSHASNFSV4(nmp))
4779 clp = nfscl_findcl(nmp);
4784 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4785 if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE))
4786 np->n_size = dp->nfsdl_size;
4791 * If there is a valid write delegation for this file, set the modtime
4792 * to the local clock time.
4795 nfscl_delegmodtime(vnode_t vp)
4797 struct nfsclclient *clp;
4798 struct nfscldeleg *dp;
4799 struct nfsnode *np = VTONFS(vp);
4800 struct nfsmount *nmp;
4802 nmp = VFSTONFS(vnode_mount(vp));
4803 if (!NFSHASNFSV4(nmp))
4806 clp = nfscl_findcl(nmp);
4811 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4812 if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE)) {
4813 nanotime(&dp->nfsdl_modtime);
4814 dp->nfsdl_flags |= NFSCLDL_MODTIMESET;
4820 * If there is a valid write delegation for this file with a modtime set,
4821 * put that modtime in mtime.
4824 nfscl_deleggetmodtime(vnode_t vp, struct timespec *mtime)
4826 struct nfsclclient *clp;
4827 struct nfscldeleg *dp;
4828 struct nfsnode *np = VTONFS(vp);
4829 struct nfsmount *nmp;
4831 nmp = VFSTONFS(vnode_mount(vp));
4832 if (!NFSHASNFSV4(nmp))
4835 clp = nfscl_findcl(nmp);
4840 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4842 (dp->nfsdl_flags & (NFSCLDL_WRITE | NFSCLDL_MODTIMESET)) ==
4843 (NFSCLDL_WRITE | NFSCLDL_MODTIMESET))
4844 *mtime = dp->nfsdl_modtime;
4849 nfscl_errmap(struct nfsrv_descript *nd, u_int32_t minorvers)
4851 short *defaulterrp, *errp;
4853 if (!nd->nd_repstat)
4855 if (nd->nd_procnum == NFSPROC_NOOP)
4856 return (txdr_unsigned(nd->nd_repstat & 0xffff));
4857 if (nd->nd_repstat == EBADRPC)
4858 return (txdr_unsigned(NFSERR_BADXDR));
4859 if (nd->nd_repstat == NFSERR_MINORVERMISMATCH ||
4860 nd->nd_repstat == NFSERR_OPILLEGAL)
4861 return (txdr_unsigned(nd->nd_repstat));
4862 if (nd->nd_repstat >= NFSERR_BADIOMODE && nd->nd_repstat < 20000 &&
4863 minorvers > NFSV4_MINORVERSION) {
4864 /* NFSv4.n error. */
4865 return (txdr_unsigned(nd->nd_repstat));
4867 if (nd->nd_procnum < NFSV4OP_CBNOPS)
4868 errp = defaulterrp = nfscl_cberrmap[nd->nd_procnum];
4870 return (txdr_unsigned(nd->nd_repstat));
4872 if (*errp == (short)nd->nd_repstat)
4873 return (txdr_unsigned(nd->nd_repstat));
4874 return (txdr_unsigned(*defaulterrp));
4878 * Called to find/add a layout to a client.
4879 * This function returns the layout with a refcnt (shared lock) upon
4880 * success (returns 0) or with no lock/refcnt on the layout when an
4881 * error is returned.
4882 * If a layout is passed in via lypp, it is locked (exclusively locked).
4885 nfscl_layout(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen,
4886 nfsv4stateid_t *stateidp, int layouttype, int retonclose,
4887 struct nfsclflayouthead *fhlp, struct nfscllayout **lypp,
4888 struct ucred *cred, NFSPROC_T *p)
4890 struct nfsclclient *clp;
4891 struct nfscllayout *lyp, *tlyp;
4892 struct nfsclflayout *flp;
4893 struct nfsnode *np = VTONFS(vp);
4895 int layout_passed_in;
4897 mp = nmp->nm_mountp;
4898 layout_passed_in = 1;
4902 layout_passed_in = 0;
4903 tlyp = malloc(sizeof(*tlyp) + fhlen - 1, M_NFSLAYOUT,
4910 if (layout_passed_in != 0)
4911 nfsv4_unlock(&lyp->nfsly_lock, 0);
4914 free(tlyp, M_NFSLAYOUT);
4919 * Although no lyp was passed in, another thread might have
4920 * allocated one. If one is found, just increment it's ref
4921 * count and return it.
4923 lyp = nfscl_findlayout(clp, fhp, fhlen);
4927 lyp->nfsly_stateid.seqid = stateidp->seqid;
4928 lyp->nfsly_stateid.other[0] = stateidp->other[0];
4929 lyp->nfsly_stateid.other[1] = stateidp->other[1];
4930 lyp->nfsly_stateid.other[2] = stateidp->other[2];
4931 lyp->nfsly_lastbyte = 0;
4932 LIST_INIT(&lyp->nfsly_flayread);
4933 LIST_INIT(&lyp->nfsly_flayrw);
4934 LIST_INIT(&lyp->nfsly_recall);
4935 lyp->nfsly_filesid[0] = np->n_vattr.na_filesid[0];
4936 lyp->nfsly_filesid[1] = np->n_vattr.na_filesid[1];
4937 lyp->nfsly_clp = clp;
4938 if (layouttype == NFSLAYOUT_FLEXFILE)
4939 lyp->nfsly_flags = NFSLY_FLEXFILE;
4941 lyp->nfsly_flags = NFSLY_FILES;
4942 if (retonclose != 0)
4943 lyp->nfsly_flags |= NFSLY_RETONCLOSE;
4944 lyp->nfsly_fhlen = fhlen;
4945 NFSBCOPY(fhp, lyp->nfsly_fh, fhlen);
4946 TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
4947 LIST_INSERT_HEAD(NFSCLLAYOUTHASH(clp, fhp, fhlen), lyp,
4949 lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
4952 if (retonclose != 0)
4953 lyp->nfsly_flags |= NFSLY_RETONCLOSE;
4954 if (stateidp->seqid > lyp->nfsly_stateid.seqid)
4955 lyp->nfsly_stateid.seqid = stateidp->seqid;
4956 TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list);
4957 TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
4958 lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
4960 nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
4961 if (NFSCL_FORCEDISM(mp)) {
4964 free(tlyp, M_NFSLAYOUT);
4968 } else if (stateidp->seqid > lyp->nfsly_stateid.seqid)
4969 lyp->nfsly_stateid.seqid = stateidp->seqid;
4971 /* Merge the new list of File Layouts into the list. */
4972 flp = LIST_FIRST(fhlp);
4974 if (flp->nfsfl_iomode == NFSLAYOUTIOMODE_READ)
4975 nfscl_mergeflayouts(&lyp->nfsly_flayread, fhlp);
4977 nfscl_mergeflayouts(&lyp->nfsly_flayrw, fhlp);
4979 if (layout_passed_in != 0)
4980 nfsv4_unlock(&lyp->nfsly_lock, 1);
4983 free(tlyp, M_NFSLAYOUT);
4988 * Search for a layout by MDS file handle.
4989 * If one is found, it is returned with a refcnt (shared lock) iff
4990 * retflpp returned non-NULL and locked (exclusive locked) iff retflpp is
4993 struct nfscllayout *
4994 nfscl_getlayout(struct nfsclclient *clp, uint8_t *fhp, int fhlen,
4995 uint64_t off, struct nfsclflayout **retflpp, int *recalledp)
4997 struct nfscllayout *lyp;
4999 int error, igotlock;
5001 mp = clp->nfsc_nmp->nm_mountp;
5005 lyp = nfscl_findlayout(clp, fhp, fhlen);
5007 if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
5008 TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list);
5009 TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
5010 lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
5011 error = nfscl_findlayoutforio(lyp, off,
5012 NFSV4OPEN_ACCESSREAD, retflpp);
5014 nfsv4_getref(&lyp->nfsly_lock, NULL,
5015 NFSCLSTATEMUTEXPTR, mp);
5018 igotlock = nfsv4_lock(&lyp->nfsly_lock,
5019 1, NULL, NFSCLSTATEMUTEXPTR, mp);
5020 } while (igotlock == 0 && !NFSCL_FORCEDISM(mp));
5023 if (NFSCL_FORCEDISM(mp)) {
5037 * Search for a layout by MDS file handle. If one is found, mark in to be
5038 * recalled, if it already marked "return on close".
5041 nfscl_retoncloselayout(vnode_t vp, struct nfsclclient *clp, uint8_t *fhp,
5042 int fhlen, struct nfsclrecalllayout **recallpp)
5044 struct nfscllayout *lyp;
5047 if (vp->v_type != VREG || !NFSHASPNFS(VFSTONFS(vnode_mount(vp))) ||
5048 nfscl_enablecallb == 0 || nfs_numnfscbd == 0 ||
5049 (VTONFS(vp)->n_flag & NNOLAYOUT) != 0)
5051 lyp = nfscl_findlayout(clp, fhp, fhlen);
5052 if (lyp != NULL && (lyp->nfsly_flags & (NFSLY_RETONCLOSE |
5053 NFSLY_RECALL)) == NFSLY_RETONCLOSE) {
5055 if (!LIST_EMPTY(&lyp->nfsly_flayread))
5056 iomode |= NFSLAYOUTIOMODE_READ;
5057 if (!LIST_EMPTY(&lyp->nfsly_flayrw))
5058 iomode |= NFSLAYOUTIOMODE_RW;
5059 (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE, lyp, iomode,
5060 0, UINT64_MAX, lyp->nfsly_stateid.seqid, 0, 0, NULL,
5062 NFSCL_DEBUG(4, "retoncls recall iomode=%d\n", iomode);
5068 * Mark the layout to be recalled and with an error.
5069 * Also, disable the dsp from further use.
5072 nfscl_dserr(uint32_t op, uint32_t stat, struct nfscldevinfo *dp,
5073 struct nfscllayout *lyp, struct nfsclds *dsp)
5075 struct nfsclrecalllayout *recallp;
5078 printf("DS being disabled, error=%d\n", stat);
5079 /* Set up the return of the layout. */
5080 recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_WAITOK);
5083 if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
5084 if (!LIST_EMPTY(&lyp->nfsly_flayread))
5085 iomode |= NFSLAYOUTIOMODE_READ;
5086 if (!LIST_EMPTY(&lyp->nfsly_flayrw))
5087 iomode |= NFSLAYOUTIOMODE_RW;
5088 (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE, lyp, iomode,
5089 0, UINT64_MAX, lyp->nfsly_stateid.seqid, stat, op,
5090 dp->nfsdi_deviceid, recallp);
5092 NFSCL_DEBUG(4, "nfscl_dserr recall iomode=%d\n", iomode);
5095 free(recallp, M_NFSLAYRECALL);
5098 /* And shut the TCP connection down. */
5099 nfscl_cancelreqs(dsp);
5103 * Cancel all RPCs for this "dsp" by closing the connection.
5104 * Also, mark the session as defunct.
5105 * If NFSCLDS_SAMECONN is set, the connection is shared with other DSs and
5106 * cannot be shut down.
5109 nfscl_cancelreqs(struct nfsclds *dsp)
5111 struct __rpc_client *cl;
5112 static int non_event;
5115 if ((dsp->nfsclds_flags & (NFSCLDS_CLOSED | NFSCLDS_SAMECONN)) == 0 &&
5116 dsp->nfsclds_sockp != NULL &&
5117 dsp->nfsclds_sockp->nr_client != NULL) {
5118 dsp->nfsclds_flags |= NFSCLDS_CLOSED;
5119 cl = dsp->nfsclds_sockp->nr_client;
5120 dsp->nfsclds_sess.nfsess_defunct = 1;
5124 * This 1sec sleep is done to reduce the number of reconnect
5125 * attempts made on the DS while it has failed.
5127 tsleep(&non_event, PVFS, "ndscls", hz);
5134 * Dereference a layout.
5137 nfscl_rellayout(struct nfscllayout *lyp, int exclocked)
5142 nfsv4_unlock(&lyp->nfsly_lock, 0);
5144 nfsv4_relref(&lyp->nfsly_lock);
5149 * Search for a devinfo by deviceid. If one is found, return it after
5150 * acquiring a reference count on it.
5152 struct nfscldevinfo *
5153 nfscl_getdevinfo(struct nfsclclient *clp, uint8_t *deviceid,
5154 struct nfscldevinfo *dip)
5159 dip = nfscl_finddevinfo(clp, deviceid);
5161 dip->nfsdi_refcnt++;
5167 * Dereference a devinfo structure.
5170 nfscl_reldevinfo_locked(struct nfscldevinfo *dip)
5173 dip->nfsdi_refcnt--;
5174 if (dip->nfsdi_refcnt == 0)
5175 wakeup(&dip->nfsdi_refcnt);
5179 * Dereference a devinfo structure.
5182 nfscl_reldevinfo(struct nfscldevinfo *dip)
5186 nfscl_reldevinfo_locked(dip);
5191 * Find a layout for this file handle. Return NULL upon failure.
5193 static struct nfscllayout *
5194 nfscl_findlayout(struct nfsclclient *clp, u_int8_t *fhp, int fhlen)
5196 struct nfscllayout *lyp;
5198 LIST_FOREACH(lyp, NFSCLLAYOUTHASH(clp, fhp, fhlen), nfsly_hash)
5199 if (lyp->nfsly_fhlen == fhlen &&
5200 !NFSBCMP(lyp->nfsly_fh, fhp, fhlen))
5206 * Find a devinfo for this deviceid. Return NULL upon failure.
5208 static struct nfscldevinfo *
5209 nfscl_finddevinfo(struct nfsclclient *clp, uint8_t *deviceid)
5211 struct nfscldevinfo *dip;
5213 LIST_FOREACH(dip, &clp->nfsc_devinfo, nfsdi_list)
5214 if (NFSBCMP(dip->nfsdi_deviceid, deviceid, NFSX_V4DEVICEID)
5221 * Merge the new file layout list into the main one, maintaining it in
5222 * increasing offset order.
5225 nfscl_mergeflayouts(struct nfsclflayouthead *fhlp,
5226 struct nfsclflayouthead *newfhlp)
5228 struct nfsclflayout *flp, *nflp, *prevflp, *tflp;
5230 flp = LIST_FIRST(fhlp);
5232 LIST_FOREACH_SAFE(nflp, newfhlp, nfsfl_list, tflp) {
5233 while (flp != NULL && flp->nfsfl_off < nflp->nfsfl_off) {
5235 flp = LIST_NEXT(flp, nfsfl_list);
5237 if (prevflp == NULL)
5238 LIST_INSERT_HEAD(fhlp, nflp, nfsfl_list);
5240 LIST_INSERT_AFTER(prevflp, nflp, nfsfl_list);
5246 * Add this nfscldevinfo to the client, if it doesn't already exist.
5247 * This function consumes the structure pointed at by dip, if not NULL.
5250 nfscl_adddevinfo(struct nfsmount *nmp, struct nfscldevinfo *dip, int ind,
5251 struct nfsclflayout *flp)
5253 struct nfsclclient *clp;
5254 struct nfscldevinfo *tdip;
5262 free(dip, M_NFSDEVINFO);
5265 if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5266 dev = flp->nfsfl_dev;
5268 dev = flp->nfsfl_ffm[ind].dev;
5269 tdip = nfscl_finddevinfo(clp, dev);
5271 tdip->nfsdi_layoutrefs++;
5272 if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5273 flp->nfsfl_devp = tdip;
5275 flp->nfsfl_ffm[ind].devp = tdip;
5276 nfscl_reldevinfo_locked(tdip);
5279 free(dip, M_NFSDEVINFO);
5283 LIST_INSERT_HEAD(&clp->nfsc_devinfo, dip, nfsdi_list);
5284 dip->nfsdi_layoutrefs = 1;
5285 if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5286 flp->nfsfl_devp = dip;
5288 flp->nfsfl_ffm[ind].devp = dip;
5297 * Free up a layout structure and associated file layout structure(s).
5300 nfscl_freelayout(struct nfscllayout *layp)
5302 struct nfsclflayout *flp, *nflp;
5303 struct nfsclrecalllayout *rp, *nrp;
5305 LIST_FOREACH_SAFE(flp, &layp->nfsly_flayread, nfsfl_list, nflp) {
5306 LIST_REMOVE(flp, nfsfl_list);
5307 nfscl_freeflayout(flp);
5309 LIST_FOREACH_SAFE(flp, &layp->nfsly_flayrw, nfsfl_list, nflp) {
5310 LIST_REMOVE(flp, nfsfl_list);
5311 nfscl_freeflayout(flp);
5313 LIST_FOREACH_SAFE(rp, &layp->nfsly_recall, nfsrecly_list, nrp) {
5314 LIST_REMOVE(rp, nfsrecly_list);
5315 free(rp, M_NFSLAYRECALL);
5318 free(layp, M_NFSLAYOUT);
5322 * Free up a file layout structure.
5325 nfscl_freeflayout(struct nfsclflayout *flp)
5329 if ((flp->nfsfl_flags & NFSFL_FILE) != 0) {
5330 for (i = 0; i < flp->nfsfl_fhcnt; i++)
5331 free(flp->nfsfl_fh[i], M_NFSFH);
5332 if (flp->nfsfl_devp != NULL)
5333 flp->nfsfl_devp->nfsdi_layoutrefs--;
5335 if ((flp->nfsfl_flags & NFSFL_FLEXFILE) != 0)
5336 for (i = 0; i < flp->nfsfl_mirrorcnt; i++) {
5337 for (j = 0; j < flp->nfsfl_ffm[i].fhcnt; j++)
5338 free(flp->nfsfl_ffm[i].fh[j], M_NFSFH);
5339 if (flp->nfsfl_ffm[i].devp != NULL)
5340 flp->nfsfl_ffm[i].devp->nfsdi_layoutrefs--;
5342 free(flp, M_NFSFLAYOUT);
5346 * Free up a file layout devinfo structure.
5349 nfscl_freedevinfo(struct nfscldevinfo *dip)
5352 free(dip, M_NFSDEVINFO);
5356 * Mark any layouts that match as recalled.
5359 nfscl_layoutrecall(int recalltype, struct nfscllayout *lyp, uint32_t iomode,
5360 uint64_t off, uint64_t len, uint32_t stateseqid, uint32_t stat, uint32_t op,
5361 char *devid, struct nfsclrecalllayout *recallp)
5363 struct nfsclrecalllayout *rp, *orp;
5365 recallp->nfsrecly_recalltype = recalltype;
5366 recallp->nfsrecly_iomode = iomode;
5367 recallp->nfsrecly_stateseqid = stateseqid;
5368 recallp->nfsrecly_off = off;
5369 recallp->nfsrecly_len = len;
5370 recallp->nfsrecly_stat = stat;
5371 recallp->nfsrecly_op = op;
5373 NFSBCOPY(devid, recallp->nfsrecly_devid, NFSX_V4DEVICEID);
5375 * Order the list as file returns first, followed by fsid and any
5376 * returns, both in increasing stateseqid order.
5377 * Note that the seqids wrap around, so 1 is after 0xffffffff.
5378 * (I'm not sure this is correct because I find RFC5661 confusing
5379 * on this, but hopefully it will work ok.)
5382 LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) {
5384 if ((recalltype == NFSLAYOUTRETURN_FILE &&
5385 (rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE ||
5386 nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) ||
5387 (recalltype != NFSLAYOUTRETURN_FILE &&
5388 rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE &&
5389 nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) {
5390 LIST_INSERT_BEFORE(rp, recallp, nfsrecly_list);
5395 * Put any error return on all the file returns that will
5398 if (rp->nfsrecly_recalltype == NFSLAYOUTRETURN_FILE &&
5399 stat != 0 && rp->nfsrecly_stat == 0) {
5400 rp->nfsrecly_stat = stat;
5401 rp->nfsrecly_op = op;
5403 NFSBCOPY(devid, rp->nfsrecly_devid,
5409 LIST_INSERT_HEAD(&lyp->nfsly_recall, recallp,
5412 LIST_INSERT_AFTER(orp, recallp, nfsrecly_list);
5414 lyp->nfsly_flags |= NFSLY_RECALL;
5415 wakeup(lyp->nfsly_clp);
5420 * Compare the two seqids for ordering. The trick is that the seqids can
5421 * wrap around from 0xffffffff->0, so check for the cases where one
5422 * has wrapped around.
5423 * Return 1 if seqid1 comes before seqid2, 0 otherwise.
5426 nfscl_seq(uint32_t seqid1, uint32_t seqid2)
5429 if (seqid2 > seqid1 && (seqid2 - seqid1) >= 0x7fffffff)
5430 /* seqid2 has wrapped around. */
5432 if (seqid1 > seqid2 && (seqid1 - seqid2) >= 0x7fffffff)
5433 /* seqid1 has wrapped around. */
5435 if (seqid1 <= seqid2)
5441 * Do a layout return for each of the recalls.
5444 nfscl_layoutreturn(struct nfsmount *nmp, struct nfscllayout *lyp,
5445 struct ucred *cred, NFSPROC_T *p)
5447 struct nfsclrecalllayout *rp;
5448 nfsv4stateid_t stateid;
5451 NFSBCOPY(lyp->nfsly_stateid.other, stateid.other, NFSX_STATEIDOTHER);
5452 stateid.seqid = lyp->nfsly_stateid.seqid;
5453 if ((lyp->nfsly_flags & NFSLY_FILES) != 0)
5454 layouttype = NFSLAYOUT_NFSV4_1_FILES;
5456 layouttype = NFSLAYOUT_FLEXFILE;
5457 LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) {
5458 (void)nfsrpc_layoutreturn(nmp, lyp->nfsly_fh,
5459 lyp->nfsly_fhlen, 0, layouttype,
5460 rp->nfsrecly_iomode, rp->nfsrecly_recalltype,
5461 rp->nfsrecly_off, rp->nfsrecly_len,
5462 &stateid, cred, p, rp->nfsrecly_stat, rp->nfsrecly_op,
5463 rp->nfsrecly_devid);
5468 * Do the layout commit for a file layout.
5471 nfscl_dolayoutcommit(struct nfsmount *nmp, struct nfscllayout *lyp,
5472 struct ucred *cred, NFSPROC_T *p)
5474 struct nfsclflayout *flp;
5476 int error, layouttype;
5478 if ((lyp->nfsly_flags & NFSLY_FILES) != 0)
5479 layouttype = NFSLAYOUT_NFSV4_1_FILES;
5481 layouttype = NFSLAYOUT_FLEXFILE;
5482 LIST_FOREACH(flp, &lyp->nfsly_flayrw, nfsfl_list) {
5483 if (layouttype == NFSLAYOUT_FLEXFILE &&
5484 (flp->nfsfl_fflags & NFSFLEXFLAG_NO_LAYOUTCOMMIT) != 0) {
5485 NFSCL_DEBUG(4, "Flex file: no layoutcommit\n");
5486 /* If not supported, don't bother doing it. */
5488 nmp->nm_state |= NFSSTA_NOLAYOUTCOMMIT;
5491 } else if (flp->nfsfl_off <= lyp->nfsly_lastbyte) {
5492 len = flp->nfsfl_end - flp->nfsfl_off;
5493 error = nfsrpc_layoutcommit(nmp, lyp->nfsly_fh,
5494 lyp->nfsly_fhlen, 0, flp->nfsfl_off, len,
5495 lyp->nfsly_lastbyte, &lyp->nfsly_stateid,
5496 layouttype, cred, p, NULL);
5497 NFSCL_DEBUG(4, "layoutcommit err=%d\n", error);
5498 if (error == NFSERR_NOTSUPP) {
5499 /* If not supported, don't bother doing it. */
5501 nmp->nm_state |= NFSSTA_NOLAYOUTCOMMIT;
5510 * Commit all layouts for a file (vnode).
5513 nfscl_layoutcommit(vnode_t vp, NFSPROC_T *p)
5515 struct nfsclclient *clp;
5516 struct nfscllayout *lyp;
5517 struct nfsnode *np = VTONFS(vp);
5519 struct nfsmount *nmp;
5521 mp = vnode_mount(vp);
5523 if (NFSHASNOLAYOUTCOMMIT(nmp))
5531 lyp = nfscl_findlayout(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
5536 nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
5537 if (NFSCL_FORCEDISM(mp)) {
5542 if ((lyp->nfsly_flags & NFSLY_WRITTEN) != 0) {
5543 lyp->nfsly_flags &= ~NFSLY_WRITTEN;
5545 NFSCL_DEBUG(4, "do layoutcommit2\n");
5546 nfscl_dolayoutcommit(clp->nfsc_nmp, lyp, NFSPROCCRED(p), p);
5550 nfsv4_relref(&lyp->nfsly_lock);