2 * Copyright (c) 2009 Rick Macklem, University of Guelph
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
32 * These functions implement the client side state handling for NFSv4.
33 * NFSv4 state handling:
34 * - A lockowner is used to determine lock contention, so it
35 * corresponds directly to a Posix pid. (1 to 1 mapping)
36 * - The correct granularity of an OpenOwner is not nearly so
37 * obvious. An OpenOwner does the following:
38 * - provides a serial sequencing of Open/Close/Lock-with-new-lockowner
39 * - is used to check for Open/Share contention (not applicable to
40 * this client, since all Opens are Deny_None)
41 * As such, I considered both extreme.
42 * 1 OpenOwner per ClientID - Simple to manage, but fully serializes
43 * all Open, Close and Lock (with a new lockowner) Ops.
44 * 1 OpenOwner for each Open - This one results in an OpenConfirm for
45 * every Open, for most servers.
46 * So, I chose to use the same mapping as I did for LockOwnwers.
47 * The main concern here is that you can end up with multiple Opens
48 * for the same File Handle, but on different OpenOwners (opens
49 * inherited from parents, grandparents...) and you do not know
50 * which of these the vnodeop close applies to. This is handled by
51 * delaying the Close Op(s) until all of the Opens have been closed.
52 * (It is not yet obvious if this is the correct granularity.)
53 * - How the code handles serialization:
54 * - For the ClientId, it uses an exclusive lock while getting its
55 * SetClientId and during recovery. Otherwise, it uses a shared
56 * lock via a reference count.
57 * - For the rest of the data structures, it uses an SMP mutex
58 * (once the nfs client is SMP safe) and doesn't sleep while
59 * manipulating the linked lists.
60 * - The serialization of Open/Close/Lock/LockU falls out in the
61 * "wash", since OpenOwners and LockOwners are both mapped from
62 * Posix pid. In other words, there is only one Posix pid using
63 * any given owner, so that owner is serialized. (If you change
64 * the granularity of the OpenOwner, then code must be added to
65 * serialize Ops on the OpenOwner.)
66 * - When to get rid of OpenOwners and LockOwners.
67 * - The function nfscl_cleanup_common() is executed after a process exits.
68 * It goes through the client list looking for all Open and Lock Owners.
69 * When one is found, it is marked "defunct" or in the case of
70 * an OpenOwner without any Opens, freed.
71 * The renew thread scans for defunct Owners and gets rid of them,
72 * if it can. The LockOwners will also be deleted when the
73 * associated Open is closed.
74 * - If the LockU or Close Op(s) fail during close in a way
75 * that could be recovered upon retry, they are relinked to the
76 * ClientId's defunct open list and retried by the renew thread
77 * until they succeed or an unmount/recovery occurs.
78 * (Since we are done with them, they do not need to be recovered.)
82 #include <fs/nfs/nfsport.h>
87 extern struct nfsstats newnfsstats;
88 extern struct nfsreqhead nfsd_reqq;
89 extern u_int32_t newnfs_false, newnfs_true;
90 extern int nfscl_debuglevel;
94 struct nfsclhead nfsclhead; /* Head of clientid list */
95 int nfscl_deleghighwater = NFSCLDELEGHIGHWATER;
96 int nfscl_layouthighwater = NFSCLLAYOUTHIGHWATER;
97 #endif /* !APPLEKEXT */
99 static int nfscl_delegcnt = 0;
100 static int nfscl_layoutcnt = 0;
101 static int nfscl_getopen(struct nfsclownerhead *, u_int8_t *, int, u_int8_t *,
102 u_int8_t *, u_int32_t, struct nfscllockowner **, struct nfsclopen **);
103 static void nfscl_clrelease(struct nfsclclient *);
104 static void nfscl_cleanclient(struct nfsclclient *);
105 static void nfscl_expireclient(struct nfsclclient *, struct nfsmount *,
106 struct ucred *, NFSPROC_T *);
107 static int nfscl_expireopen(struct nfsclclient *, struct nfsclopen *,
108 struct nfsmount *, struct ucred *, NFSPROC_T *);
109 static void nfscl_recover(struct nfsclclient *, struct ucred *, NFSPROC_T *);
110 static void nfscl_insertlock(struct nfscllockowner *, struct nfscllock *,
111 struct nfscllock *, int);
112 static int nfscl_updatelock(struct nfscllockowner *, struct nfscllock **,
113 struct nfscllock **, int);
114 static void nfscl_delegreturnall(struct nfsclclient *, NFSPROC_T *);
115 static u_int32_t nfscl_nextcbident(void);
116 static mount_t nfscl_getmnt(int, uint8_t *, u_int32_t, struct nfsclclient **);
117 static struct nfsclclient *nfscl_getclnt(u_int32_t);
118 static struct nfsclclient *nfscl_getclntsess(uint8_t *);
119 static struct nfscldeleg *nfscl_finddeleg(struct nfsclclient *, u_int8_t *,
121 static void nfscl_retoncloselayout(struct nfsclclient *, uint8_t *, int);
122 static void nfscl_reldevinfo_locked(struct nfscldevinfo *);
123 static struct nfscllayout *nfscl_findlayout(struct nfsclclient *, u_int8_t *,
125 static struct nfscldevinfo *nfscl_finddevinfo(struct nfsclclient *, uint8_t *);
126 static int nfscl_checkconflict(struct nfscllockownerhead *, struct nfscllock *,
127 u_int8_t *, struct nfscllock **);
128 static void nfscl_freealllocks(struct nfscllockownerhead *, int);
129 static int nfscl_localconflict(struct nfsclclient *, u_int8_t *, int,
130 struct nfscllock *, u_int8_t *, struct nfscldeleg *, struct nfscllock **);
131 static void nfscl_newopen(struct nfsclclient *, struct nfscldeleg *,
132 struct nfsclowner **, struct nfsclowner **, struct nfsclopen **,
133 struct nfsclopen **, u_int8_t *, u_int8_t *, int, int *);
134 static int nfscl_moveopen(vnode_t , struct nfsclclient *,
135 struct nfsmount *, struct nfsclopen *, struct nfsclowner *,
136 struct nfscldeleg *, struct ucred *, NFSPROC_T *);
137 static void nfscl_totalrecall(struct nfsclclient *);
138 static int nfscl_relock(vnode_t , struct nfsclclient *, struct nfsmount *,
139 struct nfscllockowner *, struct nfscllock *, struct ucred *, NFSPROC_T *);
140 static int nfscl_tryopen(struct nfsmount *, vnode_t , u_int8_t *, int,
141 u_int8_t *, int, u_int32_t, struct nfsclopen *, u_int8_t *, int,
142 struct nfscldeleg **, int, u_int32_t, struct ucred *, NFSPROC_T *);
143 static int nfscl_trylock(struct nfsmount *, vnode_t , u_int8_t *,
144 int, struct nfscllockowner *, int, int, u_int64_t, u_int64_t, short,
145 struct ucred *, NFSPROC_T *);
146 static int nfsrpc_reopen(struct nfsmount *, u_int8_t *, int, u_int32_t,
147 struct nfsclopen *, struct nfscldeleg **, struct ucred *, NFSPROC_T *);
148 static void nfscl_freedeleg(struct nfscldeleghead *, struct nfscldeleg *);
149 static int nfscl_errmap(struct nfsrv_descript *, u_int32_t);
150 static void nfscl_cleanup_common(struct nfsclclient *, u_int8_t *);
151 static int nfscl_recalldeleg(struct nfsclclient *, struct nfsmount *,
152 struct nfscldeleg *, vnode_t, struct ucred *, NFSPROC_T *, int);
153 static void nfscl_freeopenowner(struct nfsclowner *, int);
154 static void nfscl_cleandeleg(struct nfscldeleg *);
155 static int nfscl_trydelegreturn(struct nfscldeleg *, struct ucred *,
156 struct nfsmount *, NFSPROC_T *);
157 static void nfscl_emptylockowner(struct nfscllockowner *,
158 struct nfscllockownerfhhead *);
159 static void nfscl_mergeflayouts(struct nfsclflayouthead *,
160 struct nfsclflayouthead *);
161 static int nfscl_layoutrecall(int, struct nfscllayout *, uint32_t, uint64_t,
162 uint64_t, uint32_t, struct nfsclrecalllayout *);
163 static int nfscl_seq(uint32_t, uint32_t);
164 static void nfscl_layoutreturn(struct nfsmount *, struct nfscllayout *,
165 struct ucred *, NFSPROC_T *);
166 static void nfscl_dolayoutcommit(struct nfsmount *, struct nfscllayout *,
167 struct ucred *, NFSPROC_T *);
169 static short nfscberr_null[] = {
174 static short nfscberr_getattr[] = {
183 static short nfscberr_recall[] = {
193 static short *nfscl_cberrmap[] = {
201 #define NETFAMILY(clp) \
202 (((clp)->nfsc_flags & NFSCLFLAGS_AFINET6) ? AF_INET6 : AF_INET)
205 * Called for an open operation.
206 * If the nfhp argument is NULL, just get an openowner.
209 nfscl_open(vnode_t vp, u_int8_t *nfhp, int fhlen, u_int32_t amode, int usedeleg,
210 struct ucred *cred, NFSPROC_T *p, struct nfsclowner **owpp,
211 struct nfsclopen **opp, int *newonep, int *retp, int lockit)
213 struct nfsclclient *clp;
214 struct nfsclowner *owp, *nowp;
215 struct nfsclopen *op = NULL, *nop = NULL;
216 struct nfscldeleg *dp;
217 struct nfsclownerhead *ohp;
218 u_int8_t own[NFSV4CL_LOCKNAMELEN];
229 * Might need one or both of these, so MALLOC them now, to
230 * avoid a tsleep() in MALLOC later.
232 MALLOC(nowp, struct nfsclowner *, sizeof (struct nfsclowner),
233 M_NFSCLOWNER, M_WAITOK);
235 MALLOC(nop, struct nfsclopen *, sizeof (struct nfsclopen) +
236 fhlen - 1, M_NFSCLOPEN, M_WAITOK);
237 ret = nfscl_getcl(vnode_mount(vp), cred, p, 1, &clp);
239 FREE((caddr_t)nowp, M_NFSCLOWNER);
241 FREE((caddr_t)nop, M_NFSCLOPEN);
246 * Get the Open iff it already exists.
247 * If none found, add the new one or return error, depending upon
250 nfscl_filllockowner(p->td_proc, own, F_POSIX);
253 /* First check the delegation list */
254 if (nfhp != NULL && usedeleg) {
255 LIST_FOREACH(dp, NFSCLDELEGHASH(clp, nfhp, fhlen), nfsdl_hash) {
256 if (dp->nfsdl_fhlen == fhlen &&
257 !NFSBCMP(nfhp, dp->nfsdl_fh, fhlen)) {
258 if (!(amode & NFSV4OPEN_ACCESSWRITE) ||
259 (dp->nfsdl_flags & NFSCLDL_WRITE))
268 ohp = &dp->nfsdl_owner;
270 ohp = &clp->nfsc_owner;
271 /* Now, search for an openowner */
272 LIST_FOREACH(owp, ohp, nfsow_list) {
273 if (!NFSBCMP(owp->nfsow_owner, own, NFSV4CL_LOCKNAMELEN))
278 * Create a new open, as required.
280 nfscl_newopen(clp, dp, &owp, &nowp, &op, &nop, own, nfhp, fhlen,
284 * Now, check the mode on the open and return the appropriate
288 if (nfhp != NULL && dp != NULL && nop == NULL)
289 /* new local open on delegation */
290 *retp = NFSCLOPEN_SETCRED;
292 *retp = NFSCLOPEN_OK;
294 if (op != NULL && (amode & ~(op->nfso_mode))) {
295 op->nfso_mode |= amode;
296 if (retp != NULL && dp == NULL)
297 *retp = NFSCLOPEN_DOOPEN;
301 * Serialize modifications to the open owner for multiple threads
302 * within the same process using a read/write sleep lock.
305 nfscl_lockexcl(&owp->nfsow_rwlock, NFSCLSTATEMUTEXPTR);
308 FREE((caddr_t)nowp, M_NFSCLOWNER);
310 FREE((caddr_t)nop, M_NFSCLOPEN);
319 * Create a new open, as required.
322 nfscl_newopen(struct nfsclclient *clp, struct nfscldeleg *dp,
323 struct nfsclowner **owpp, struct nfsclowner **nowpp, struct nfsclopen **opp,
324 struct nfsclopen **nopp, u_int8_t *own, u_int8_t *fhp, int fhlen,
327 struct nfsclowner *owp = *owpp, *nowp;
328 struct nfsclopen *op, *nop;
338 if (owp == NULL && nowp != NULL) {
339 NFSBCOPY(own, nowp->nfsow_owner, NFSV4CL_LOCKNAMELEN);
340 LIST_INIT(&nowp->nfsow_open);
341 nowp->nfsow_clp = clp;
342 nowp->nfsow_seqid = 0;
343 nowp->nfsow_defunct = 0;
344 nfscl_lockinit(&nowp->nfsow_rwlock);
346 newnfsstats.cllocalopenowners++;
347 LIST_INSERT_HEAD(&dp->nfsdl_owner, nowp, nfsow_list);
349 newnfsstats.clopenowners++;
350 LIST_INSERT_HEAD(&clp->nfsc_owner, nowp, nfsow_list);
358 /* If an fhp has been specified, create an Open as well. */
360 /* and look for the correct open, based upon FH */
361 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
362 if (op->nfso_fhlen == fhlen &&
363 !NFSBCMP(op->nfso_fh, fhp, fhlen))
366 if (op == NULL && nop != NULL) {
369 nop->nfso_opencnt = 0;
370 nop->nfso_posixlock = 1;
371 nop->nfso_fhlen = fhlen;
372 NFSBCOPY(fhp, nop->nfso_fh, fhlen);
373 LIST_INIT(&nop->nfso_lock);
374 nop->nfso_stateid.seqid = 0;
375 nop->nfso_stateid.other[0] = 0;
376 nop->nfso_stateid.other[1] = 0;
377 nop->nfso_stateid.other[2] = 0;
379 TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
380 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp,
382 dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
383 newnfsstats.cllocalopens++;
385 newnfsstats.clopens++;
387 LIST_INSERT_HEAD(&owp->nfsow_open, nop, nfso_list);
399 * Called to find/add a delegation to a client.
402 nfscl_deleg(mount_t mp, struct nfsclclient *clp, u_int8_t *nfhp,
403 int fhlen, struct ucred *cred, NFSPROC_T *p, struct nfscldeleg **dpp)
405 struct nfscldeleg *dp = *dpp, *tdp;
408 * First, if we have received a Read delegation for a file on a
409 * read/write file system, just return it, because they aren't
412 if (mp != NULL && dp != NULL && !NFSMNT_RDONLY(mp) &&
413 (dp->nfsdl_flags & NFSCLDL_READ)) {
414 (void) nfscl_trydelegreturn(dp, cred, VFSTONFS(mp), p);
415 FREE((caddr_t)dp, M_NFSCLDELEG);
420 /* Look for the correct deleg, based upon FH */
422 tdp = nfscl_finddeleg(clp, nfhp, fhlen);
426 return (NFSERR_BADSTATEID);
429 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list);
430 LIST_INSERT_HEAD(NFSCLDELEGHASH(clp, nfhp, fhlen), dp,
432 dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
433 newnfsstats.cldelegates++;
437 * Delegation already exists, what do we do if a new one??
440 printf("Deleg already exists!\n");
441 FREE((caddr_t)dp, M_NFSCLDELEG);
452 * Find a delegation for this file handle. Return NULL upon failure.
454 static struct nfscldeleg *
455 nfscl_finddeleg(struct nfsclclient *clp, u_int8_t *fhp, int fhlen)
457 struct nfscldeleg *dp;
459 LIST_FOREACH(dp, NFSCLDELEGHASH(clp, fhp, fhlen), nfsdl_hash) {
460 if (dp->nfsdl_fhlen == fhlen &&
461 !NFSBCMP(dp->nfsdl_fh, fhp, fhlen))
468 * Get a stateid for an I/O operation. First, look for an open and iff
469 * found, return either a lockowner stateid or the open stateid.
470 * If no Open is found, just return error and the special stateid of all zeros.
473 nfscl_getstateid(vnode_t vp, u_int8_t *nfhp, int fhlen, u_int32_t mode,
474 int fords, struct ucred *cred, NFSPROC_T *p, nfsv4stateid_t *stateidp,
477 struct nfsclclient *clp;
478 struct nfsclowner *owp;
479 struct nfsclopen *op = NULL;
480 struct nfscllockowner *lp;
481 struct nfscldeleg *dp;
483 u_int8_t own[NFSV4CL_LOCKNAMELEN];
488 * Initially, just set the special stateid of all zeros.
489 * (Don't do this for a DS, since the special stateid can't be used.)
493 stateidp->other[0] = 0;
494 stateidp->other[1] = 0;
495 stateidp->other[2] = 0;
497 if (vnode_vtype(vp) != VREG)
501 clp = nfscl_findcl(VFSTONFS(vnode_mount(vp)));
508 * Wait for recovery to complete.
510 while ((clp->nfsc_flags & NFSCLFLAGS_RECVRINPROG))
511 (void) nfsmsleep(&clp->nfsc_flags, NFSCLSTATEMUTEXPTR,
512 PZERO, "nfsrecvr", NULL);
515 * First, look for a delegation.
517 LIST_FOREACH(dp, NFSCLDELEGHASH(clp, nfhp, fhlen), nfsdl_hash) {
518 if (dp->nfsdl_fhlen == fhlen &&
519 !NFSBCMP(nfhp, dp->nfsdl_fh, fhlen)) {
520 if (!(mode & NFSV4OPEN_ACCESSWRITE) ||
521 (dp->nfsdl_flags & NFSCLDL_WRITE)) {
522 stateidp->seqid = dp->nfsdl_stateid.seqid;
523 stateidp->other[0] = dp->nfsdl_stateid.other[0];
524 stateidp->other[1] = dp->nfsdl_stateid.other[1];
525 stateidp->other[2] = dp->nfsdl_stateid.other[2];
526 if (!(np->n_flag & NDELEGRECALL)) {
527 TAILQ_REMOVE(&clp->nfsc_deleg, dp,
529 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp,
531 dp->nfsdl_timestamp = NFSD_MONOSEC +
533 dp->nfsdl_rwlock.nfslock_usecnt++;
534 *lckpp = (void *)&dp->nfsdl_rwlock;
545 * If p != NULL, we want to search the parentage tree
546 * for a matching OpenOwner and use that.
548 nfscl_filllockowner(p->td_proc, own, F_POSIX);
550 error = nfscl_getopen(&clp->nfsc_owner, nfhp, fhlen, own, own,
552 if (error == 0 && lp != NULL && fords == 0) {
553 /* Don't return a lock stateid for a DS. */
555 lp->nfsl_stateid.seqid;
557 lp->nfsl_stateid.other[0];
559 lp->nfsl_stateid.other[1];
561 lp->nfsl_stateid.other[2];
567 /* If not found, just look for any OpenOwner that will work. */
569 owp = LIST_FIRST(&clp->nfsc_owner);
570 while (!done && owp != NULL) {
571 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
572 if (op->nfso_fhlen == fhlen &&
573 !NFSBCMP(op->nfso_fh, nfhp, fhlen) &&
574 (mode & op->nfso_mode) == mode) {
580 owp = LIST_NEXT(owp, nfsow_list);
587 * For read aheads or write behinds, use the open cred.
588 * A read ahead or write behind is indicated by p == NULL.
591 newnfs_copycred(&op->nfso_cred, cred);
595 * No lock stateid, so return the open stateid.
597 stateidp->seqid = op->nfso_stateid.seqid;
598 stateidp->other[0] = op->nfso_stateid.other[0];
599 stateidp->other[1] = op->nfso_stateid.other[1];
600 stateidp->other[2] = op->nfso_stateid.other[2];
606 * Search for a matching file, mode and, optionally, lockowner.
609 nfscl_getopen(struct nfsclownerhead *ohp, u_int8_t *nfhp, int fhlen,
610 u_int8_t *openown, u_int8_t *lockown, u_int32_t mode,
611 struct nfscllockowner **lpp, struct nfsclopen **opp)
613 struct nfsclowner *owp;
614 struct nfsclopen *op, *rop, *rop2;
615 struct nfscllockowner *lp;
621 * rop will be set to the open to be returned. There are three
622 * variants of this, all for an open of the correct file:
623 * 1 - A match of lockown.
624 * 2 - A match of the openown, when no lockown match exists.
625 * 3 - A match for any open, if no openown or lockown match exists.
626 * Looking for #2 over #3 probably isn't necessary, but since
627 * RFC3530 is vague w.r.t. the relationship between openowners and
628 * lockowners, I think this is the safer way to go.
633 /* Search the client list */
634 owp = LIST_FIRST(ohp);
635 while (owp != NULL && keep_looping != 0) {
636 /* and look for the correct open */
637 op = LIST_FIRST(&owp->nfsow_open);
638 while (op != NULL && keep_looping != 0) {
639 if (op->nfso_fhlen == fhlen &&
640 !NFSBCMP(op->nfso_fh, nfhp, fhlen)
641 && (op->nfso_mode & mode) == mode) {
643 /* Now look for a matching lockowner. */
644 LIST_FOREACH(lp, &op->nfso_lock,
646 if (!NFSBCMP(lp->nfsl_owner,
648 NFSV4CL_LOCKNAMELEN)) {
656 if (rop == NULL && !NFSBCMP(owp->nfsow_owner,
657 openown, NFSV4CL_LOCKNAMELEN)) {
665 op = LIST_NEXT(op, nfso_list);
667 owp = LIST_NEXT(owp, nfsow_list);
678 * Release use of an open owner. Called when open operations are done
679 * with the open owner.
682 nfscl_ownerrelease(struct nfsclowner *owp, __unused int error,
683 __unused int candelete, int unlocked)
690 nfscl_lockunlock(&owp->nfsow_rwlock);
691 nfscl_clrelease(owp->nfsow_clp);
696 * Release use of an open structure under an open owner.
699 nfscl_openrelease(struct nfsclopen *op, int error, int candelete)
701 struct nfsclclient *clp;
702 struct nfsclowner *owp;
708 nfscl_lockunlock(&owp->nfsow_rwlock);
709 clp = owp->nfsow_clp;
710 if (error && candelete && op->nfso_opencnt == 0)
711 nfscl_freeopen(op, 0);
712 nfscl_clrelease(clp);
717 * Called to get a clientid structure. It will optionally lock the
718 * client data structures to do the SetClientId/SetClientId_confirm,
719 * but will release that lock and return the clientid with a refernce
721 * If the "cred" argument is NULL, a new clientid should not be created.
722 * If the "p" argument is NULL, a SetClientID/SetClientIDConfirm cannot
724 * The start_renewthread argument tells nfscl_getcl() to start a renew
725 * thread if this creates a new clp.
726 * It always clpp with a reference count on it, unless returning an error.
729 nfscl_getcl(struct mount *mp, struct ucred *cred, NFSPROC_T *p,
730 int start_renewthread, struct nfsclclient **clpp)
732 struct nfsclclient *clp;
733 struct nfsclclient *newclp = NULL;
734 struct nfsmount *nmp;
735 char uuid[HOSTUUIDLEN];
736 int igotlock = 0, error, trystalecnt, clidinusedelay, i;
741 getcredhostuuid(cred, uuid, sizeof uuid);
742 idlen = strlen(uuid);
744 idlen += sizeof (u_int64_t);
746 idlen += sizeof (u_int64_t) + 16; /* 16 random bytes */
747 MALLOC(newclp, struct nfsclclient *,
748 sizeof (struct nfsclclient) + idlen - 1, M_NFSCLCLIENT,
753 * If a forced dismount is already in progress, don't
754 * allocate a new clientid and get out now. For the case where
755 * clp != NULL, this is a harmless optimization.
757 if ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) {
760 free(newclp, M_NFSCLCLIENT);
765 if (newclp == NULL) {
770 clp->nfsc_idlen = idlen;
771 LIST_INIT(&clp->nfsc_owner);
772 TAILQ_INIT(&clp->nfsc_deleg);
773 TAILQ_INIT(&clp->nfsc_layout);
774 LIST_INIT(&clp->nfsc_devinfo);
775 for (i = 0; i < NFSCLDELEGHASHSIZE; i++)
776 LIST_INIT(&clp->nfsc_deleghash[i]);
777 for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++)
778 LIST_INIT(&clp->nfsc_layouthash[i]);
779 clp->nfsc_flags = NFSCLFLAGS_INITED;
780 clp->nfsc_clientidrev = 1;
781 clp->nfsc_cbident = nfscl_nextcbident();
782 nfscl_fillclid(nmp->nm_clval, uuid, clp->nfsc_id,
784 LIST_INSERT_HEAD(&nfsclhead, clp, nfsc_list);
788 if (start_renewthread != 0)
789 nfscl_start_renewthread(clp);
793 free(newclp, M_NFSCLCLIENT);
796 while ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0 && !igotlock &&
797 (mp->mnt_kern_flag & MNTK_UNMOUNTF) == 0)
798 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
799 NFSCLSTATEMUTEXPTR, mp);
801 nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
802 if (igotlock == 0 && (mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) {
804 * Both nfsv4_lock() and nfsv4_getref() know to check
805 * for MNTK_UNMOUNTF and return without sleeping to
806 * wait for the exclusive lock to be released, since it
807 * might be held by nfscl_umount() and we need to get out
808 * now for that case and not wait until nfscl_umount()
817 * If it needs a clientid, do the setclientid now.
819 if ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0) {
821 panic("nfscl_clget");
822 if (p == NULL || cred == NULL) {
824 nfsv4_unlock(&clp->nfsc_lock, 0);
829 * If RFC3530 Sec. 14.2.33 is taken literally,
830 * NFSERR_CLIDINUSE will be returned persistently for the
831 * case where a new mount of the same file system is using
832 * a different principal. In practice, NFSERR_CLIDINUSE is
833 * only returned when there is outstanding unexpired state
834 * on the clientid. As such, try for twice the lease
835 * interval, if we know what that is. Otherwise, make a
837 * The case of returning NFSERR_STALECLIENTID is far less
838 * likely, but might occur if there is a significant delay
839 * between doing the SetClientID and SetClientIDConfirm Ops,
840 * such that the server throws away the clientid before
841 * receiving the SetClientIDConfirm.
843 if (clp->nfsc_renew > 0)
844 clidinusedelay = NFSCL_LEASE(clp->nfsc_renew) * 2;
846 clidinusedelay = 120;
849 error = nfsrpc_setclient(nmp, clp, 0, cred, p);
850 if (error == NFSERR_STALECLIENTID ||
851 error == NFSERR_STALEDONTRECOVER ||
852 error == NFSERR_BADSESSION ||
853 error == NFSERR_CLIDINUSE) {
854 (void) nfs_catnap(PZERO, error, "nfs_setcl");
856 } while (((error == NFSERR_STALECLIENTID ||
857 error == NFSERR_BADSESSION ||
858 error == NFSERR_STALEDONTRECOVER) && --trystalecnt > 0) ||
859 (error == NFSERR_CLIDINUSE && --clidinusedelay > 0));
862 nfsv4_unlock(&clp->nfsc_lock, 0);
866 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
870 nfsv4_unlock(&clp->nfsc_lock, 1);
879 * Get a reference to a clientid and return it, if valid.
881 APPLESTATIC struct nfsclclient *
882 nfscl_findcl(struct nfsmount *nmp)
884 struct nfsclclient *clp;
887 if (clp == NULL || !(clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID))
893 * Release the clientid structure. It may be locked or reference counted.
896 nfscl_clrelease(struct nfsclclient *clp)
899 if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK)
900 nfsv4_unlock(&clp->nfsc_lock, 0);
902 nfsv4_relref(&clp->nfsc_lock);
906 * External call for nfscl_clrelease.
909 nfscl_clientrelease(struct nfsclclient *clp)
913 if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK)
914 nfsv4_unlock(&clp->nfsc_lock, 0);
916 nfsv4_relref(&clp->nfsc_lock);
921 * Called when wanting to lock a byte region.
924 nfscl_getbytelock(vnode_t vp, u_int64_t off, u_int64_t len,
925 short type, struct ucred *cred, NFSPROC_T *p, struct nfsclclient *rclp,
926 int recovery, void *id, int flags, u_int8_t *rownp, u_int8_t *ropenownp,
927 struct nfscllockowner **lpp, int *newonep, int *donelocallyp)
929 struct nfscllockowner *lp;
930 struct nfsclopen *op;
931 struct nfsclclient *clp;
932 struct nfscllockowner *nlp;
933 struct nfscllock *nlop, *otherlop;
934 struct nfscldeleg *dp = NULL, *ldp = NULL;
935 struct nfscllockownerhead *lhp = NULL;
937 u_int8_t own[NFSV4CL_LOCKNAMELEN], *ownp, openown[NFSV4CL_LOCKNAMELEN];
939 int error = 0, ret, donelocally = 0;
942 /* For Lock Ops, the open mode doesn't matter, so use 0 to match any. */
951 * Might need these, so MALLOC them now, to
952 * avoid a tsleep() in MALLOC later.
954 MALLOC(nlp, struct nfscllockowner *,
955 sizeof (struct nfscllockowner), M_NFSCLLOCKOWNER, M_WAITOK);
956 MALLOC(otherlop, struct nfscllock *,
957 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
958 MALLOC(nlop, struct nfscllock *,
959 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
960 nlop->nfslo_type = type;
961 nlop->nfslo_first = off;
962 if (len == NFS64BITSSET) {
963 nlop->nfslo_end = NFS64BITSSET;
965 nlop->nfslo_end = off + len;
966 if (nlop->nfslo_end <= nlop->nfslo_first)
967 error = NFSERR_INVAL;
974 error = nfscl_getcl(vnode_mount(vp), cred, p, 1, &clp);
977 FREE((caddr_t)nlp, M_NFSCLLOCKOWNER);
978 FREE((caddr_t)otherlop, M_NFSCLLOCK);
979 FREE((caddr_t)nlop, M_NFSCLLOCK);
986 openownp = ropenownp;
988 nfscl_filllockowner(id, own, flags);
990 nfscl_filllockowner(p->td_proc, openown, F_POSIX);
996 * First, search for a delegation. If one exists for this file,
997 * the lock can be done locally against it, so long as there
998 * isn't a local lock conflict.
1000 ldp = dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
1001 np->n_fhp->nfh_len);
1002 /* Just sanity check for correct type of delegation */
1003 if (dp != NULL && ((dp->nfsdl_flags &
1004 (NFSCLDL_RECALL | NFSCLDL_DELEGRET)) != 0 ||
1006 (dp->nfsdl_flags & NFSCLDL_WRITE) == 0)))
1010 /* Now, find an open and maybe a lockowner. */
1011 ret = nfscl_getopen(&dp->nfsdl_owner, np->n_fhp->nfh_fh,
1012 np->n_fhp->nfh_len, openownp, ownp, mode, NULL, &op);
1014 ret = nfscl_getopen(&clp->nfsc_owner,
1015 np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp,
1016 ownp, mode, NULL, &op);
1018 lhp = &dp->nfsdl_lock;
1019 TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
1020 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list);
1021 dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
1029 * Get the related Open and maybe lockowner.
1031 error = nfscl_getopen(&clp->nfsc_owner,
1032 np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp,
1033 ownp, mode, &lp, &op);
1035 lhp = &op->nfso_lock;
1037 if (!error && !recovery)
1038 error = nfscl_localconflict(clp, np->n_fhp->nfh_fh,
1039 np->n_fhp->nfh_len, nlop, ownp, ldp, NULL);
1042 nfscl_clrelease(clp);
1045 FREE((caddr_t)nlp, M_NFSCLLOCKOWNER);
1046 FREE((caddr_t)otherlop, M_NFSCLLOCK);
1047 FREE((caddr_t)nlop, M_NFSCLLOCK);
1052 * Ok, see if a lockowner exists and create one, as required.
1055 LIST_FOREACH(lp, lhp, nfsl_list) {
1056 if (!NFSBCMP(lp->nfsl_owner, ownp, NFSV4CL_LOCKNAMELEN))
1060 NFSBCOPY(ownp, nlp->nfsl_owner, NFSV4CL_LOCKNAMELEN);
1062 NFSBCOPY(ropenownp, nlp->nfsl_openowner,
1063 NFSV4CL_LOCKNAMELEN);
1065 NFSBCOPY(op->nfso_own->nfsow_owner, nlp->nfsl_openowner,
1066 NFSV4CL_LOCKNAMELEN);
1067 nlp->nfsl_seqid = 0;
1068 nlp->nfsl_lockflags = flags;
1069 nlp->nfsl_inprog = NULL;
1070 nfscl_lockinit(&nlp->nfsl_rwlock);
1071 LIST_INIT(&nlp->nfsl_lock);
1073 nlp->nfsl_open = NULL;
1074 newnfsstats.cllocallockowners++;
1076 nlp->nfsl_open = op;
1077 newnfsstats.cllockowners++;
1079 LIST_INSERT_HEAD(lhp, nlp, nfsl_list);
1086 * Now, update the byte ranges for locks.
1088 ret = nfscl_updatelock(lp, &nlop, &otherlop, donelocally);
1094 nfscl_clrelease(clp);
1097 * Serial modifications on the lock owner for multiple threads
1098 * for the same process using a read/write lock.
1101 nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR);
1107 FREE((caddr_t)nlp, M_NFSCLLOCKOWNER);
1109 FREE((caddr_t)nlop, M_NFSCLLOCK);
1111 FREE((caddr_t)otherlop, M_NFSCLLOCK);
1118 * Called to unlock a byte range, for LockU.
1121 nfscl_relbytelock(vnode_t vp, u_int64_t off, u_int64_t len,
1122 __unused struct ucred *cred, NFSPROC_T *p, int callcnt,
1123 struct nfsclclient *clp, void *id, int flags,
1124 struct nfscllockowner **lpp, int *dorpcp)
1126 struct nfscllockowner *lp;
1127 struct nfsclowner *owp;
1128 struct nfsclopen *op;
1129 struct nfscllock *nlop, *other_lop = NULL;
1130 struct nfscldeleg *dp;
1132 u_int8_t own[NFSV4CL_LOCKNAMELEN];
1140 * Might need these, so MALLOC them now, to
1141 * avoid a tsleep() in MALLOC later.
1143 MALLOC(nlop, struct nfscllock *,
1144 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1145 nlop->nfslo_type = F_UNLCK;
1146 nlop->nfslo_first = off;
1147 if (len == NFS64BITSSET) {
1148 nlop->nfslo_end = NFS64BITSSET;
1150 nlop->nfslo_end = off + len;
1151 if (nlop->nfslo_end <= nlop->nfslo_first) {
1152 FREE((caddr_t)nlop, M_NFSCLLOCK);
1153 return (NFSERR_INVAL);
1157 MALLOC(other_lop, struct nfscllock *,
1158 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1161 nfscl_filllockowner(id, own, flags);
1165 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
1166 np->n_fhp->nfh_len);
1169 * First, unlock any local regions on a delegation.
1172 /* Look for this lockowner. */
1173 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
1174 if (!NFSBCMP(lp->nfsl_owner, own,
1175 NFSV4CL_LOCKNAMELEN))
1179 /* Use other_lop, so nlop is still available */
1180 (void)nfscl_updatelock(lp, &other_lop, NULL, 1);
1184 * Now, find a matching open/lockowner that hasn't already been done,
1185 * as marked by nfsl_inprog.
1189 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
1190 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1191 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1192 !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1193 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1194 if (lp->nfsl_inprog == NULL &&
1195 !NFSBCMP(lp->nfsl_owner, own,
1196 NFSV4CL_LOCKNAMELEN)) {
1210 ret = nfscl_updatelock(lp, &nlop, NULL, 0);
1214 * Serial modifications on the lock owner for multiple
1215 * threads for the same process using a read/write lock.
1217 lp->nfsl_inprog = p;
1218 nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR);
1223 FREE((caddr_t)nlop, M_NFSCLLOCK);
1225 FREE((caddr_t)other_lop, M_NFSCLLOCK);
1230 * Release all lockowners marked in progess for this process and file.
1233 nfscl_releasealllocks(struct nfsclclient *clp, vnode_t vp, NFSPROC_T *p,
1234 void *id, int flags)
1236 struct nfsclowner *owp;
1237 struct nfsclopen *op;
1238 struct nfscllockowner *lp;
1240 u_int8_t own[NFSV4CL_LOCKNAMELEN];
1243 nfscl_filllockowner(id, own, flags);
1245 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
1246 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1247 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1248 !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1249 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1250 if (lp->nfsl_inprog == p &&
1251 !NFSBCMP(lp->nfsl_owner, own,
1252 NFSV4CL_LOCKNAMELEN)) {
1253 lp->nfsl_inprog = NULL;
1254 nfscl_lockunlock(&lp->nfsl_rwlock);
1260 nfscl_clrelease(clp);
1265 * Called to find out if any bytes within the byte range specified are
1266 * write locked by the calling process. Used to determine if flushing
1267 * is required before a LockU.
1268 * If in doubt, return 1, so the flush will occur.
1271 nfscl_checkwritelocked(vnode_t vp, struct flock *fl,
1272 struct ucred *cred, NFSPROC_T *p, void *id, int flags)
1274 struct nfsclowner *owp;
1275 struct nfscllockowner *lp;
1276 struct nfsclopen *op;
1277 struct nfsclclient *clp;
1278 struct nfscllock *lop;
1279 struct nfscldeleg *dp;
1282 u_int8_t own[NFSV4CL_LOCKNAMELEN];
1286 switch (fl->l_whence) {
1290 * Caller is responsible for adding any necessary offset
1291 * when SEEK_CUR is used.
1296 off = np->n_size + fl->l_start;
1301 if (fl->l_len != 0) {
1302 end = off + fl->l_len;
1309 error = nfscl_getcl(vnode_mount(vp), cred, p, 1, &clp);
1312 nfscl_filllockowner(id, own, flags);
1316 * First check the delegation locks.
1318 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
1320 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
1321 if (!NFSBCMP(lp->nfsl_owner, own,
1322 NFSV4CL_LOCKNAMELEN))
1326 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
1327 if (lop->nfslo_first >= end)
1329 if (lop->nfslo_end <= off)
1331 if (lop->nfslo_type == F_WRLCK) {
1332 nfscl_clrelease(clp);
1341 * Now, check state against the server.
1343 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
1344 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1345 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1346 !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1347 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1348 if (!NFSBCMP(lp->nfsl_owner, own,
1349 NFSV4CL_LOCKNAMELEN))
1353 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
1354 if (lop->nfslo_first >= end)
1356 if (lop->nfslo_end <= off)
1358 if (lop->nfslo_type == F_WRLCK) {
1359 nfscl_clrelease(clp);
1368 nfscl_clrelease(clp);
1374 * Release a byte range lock owner structure.
1377 nfscl_lockrelease(struct nfscllockowner *lp, int error, int candelete)
1379 struct nfsclclient *clp;
1384 clp = lp->nfsl_open->nfso_own->nfsow_clp;
1385 if (error != 0 && candelete &&
1386 (lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED) == 0)
1387 nfscl_freelockowner(lp, 0);
1389 nfscl_lockunlock(&lp->nfsl_rwlock);
1390 nfscl_clrelease(clp);
1395 * Free up an open structure and any associated byte range lock structures.
1398 nfscl_freeopen(struct nfsclopen *op, int local)
1401 LIST_REMOVE(op, nfso_list);
1402 nfscl_freealllocks(&op->nfso_lock, local);
1403 FREE((caddr_t)op, M_NFSCLOPEN);
1405 newnfsstats.cllocalopens--;
1407 newnfsstats.clopens--;
1411 * Free up all lock owners and associated locks.
1414 nfscl_freealllocks(struct nfscllockownerhead *lhp, int local)
1416 struct nfscllockowner *lp, *nlp;
1418 LIST_FOREACH_SAFE(lp, lhp, nfsl_list, nlp) {
1419 if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED))
1421 nfscl_freelockowner(lp, local);
1426 * Called for an Open when NFSERR_EXPIRED is received from the server.
1427 * If there are no byte range locks nor a Share Deny lost, try to do a
1428 * fresh Open. Otherwise, free the open.
1431 nfscl_expireopen(struct nfsclclient *clp, struct nfsclopen *op,
1432 struct nfsmount *nmp, struct ucred *cred, NFSPROC_T *p)
1434 struct nfscllockowner *lp;
1435 struct nfscldeleg *dp;
1436 int mustdelete = 0, error;
1439 * Look for any byte range lock(s).
1441 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1442 if (!LIST_EMPTY(&lp->nfsl_lock)) {
1449 * If no byte range lock(s) nor a Share deny, try to re-open.
1451 if (!mustdelete && (op->nfso_mode & NFSLCK_DENYBITS) == 0) {
1452 newnfs_copycred(&op->nfso_cred, cred);
1454 error = nfsrpc_reopen(nmp, op->nfso_fh,
1455 op->nfso_fhlen, op->nfso_mode, op, &dp, cred, p);
1459 FREE((caddr_t)dp, M_NFSCLDELEG);
1464 nfscl_deleg(nmp->nm_mountp, clp, op->nfso_fh,
1465 op->nfso_fhlen, cred, p, &dp);
1469 * If a byte range lock or Share deny or couldn't re-open, free it.
1472 nfscl_freeopen(op, 0);
1473 return (mustdelete);
1477 * Free up an open owner structure.
1480 nfscl_freeopenowner(struct nfsclowner *owp, int local)
1483 LIST_REMOVE(owp, nfsow_list);
1484 FREE((caddr_t)owp, M_NFSCLOWNER);
1486 newnfsstats.cllocalopenowners--;
1488 newnfsstats.clopenowners--;
1492 * Free up a byte range lock owner structure.
1495 nfscl_freelockowner(struct nfscllockowner *lp, int local)
1497 struct nfscllock *lop, *nlop;
1499 LIST_REMOVE(lp, nfsl_list);
1500 LIST_FOREACH_SAFE(lop, &lp->nfsl_lock, nfslo_list, nlop) {
1501 nfscl_freelock(lop, local);
1503 FREE((caddr_t)lp, M_NFSCLLOCKOWNER);
1505 newnfsstats.cllocallockowners--;
1507 newnfsstats.cllockowners--;
1511 * Free up a byte range lock structure.
1514 nfscl_freelock(struct nfscllock *lop, int local)
1517 LIST_REMOVE(lop, nfslo_list);
1518 FREE((caddr_t)lop, M_NFSCLLOCK);
1520 newnfsstats.cllocallocks--;
1522 newnfsstats.cllocks--;
1526 * Clean out the state related to a delegation.
1529 nfscl_cleandeleg(struct nfscldeleg *dp)
1531 struct nfsclowner *owp, *nowp;
1532 struct nfsclopen *op;
1534 LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) {
1535 op = LIST_FIRST(&owp->nfsow_open);
1537 if (LIST_NEXT(op, nfso_list) != NULL)
1538 panic("nfscleandel");
1539 nfscl_freeopen(op, 1);
1541 nfscl_freeopenowner(owp, 1);
1543 nfscl_freealllocks(&dp->nfsdl_lock, 1);
1547 * Free a delegation.
1550 nfscl_freedeleg(struct nfscldeleghead *hdp, struct nfscldeleg *dp)
1553 TAILQ_REMOVE(hdp, dp, nfsdl_list);
1554 LIST_REMOVE(dp, nfsdl_hash);
1555 FREE((caddr_t)dp, M_NFSCLDELEG);
1556 newnfsstats.cldelegates--;
1561 * Free up all state related to this client structure.
1564 nfscl_cleanclient(struct nfsclclient *clp)
1566 struct nfsclowner *owp, *nowp;
1567 struct nfsclopen *op, *nop;
1569 /* Now, all the OpenOwners, etc. */
1570 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1571 LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
1572 nfscl_freeopen(op, 0);
1574 nfscl_freeopenowner(owp, 0);
1579 * Called when an NFSERR_EXPIRED is received from the server.
1582 nfscl_expireclient(struct nfsclclient *clp, struct nfsmount *nmp,
1583 struct ucred *cred, NFSPROC_T *p)
1585 struct nfsclowner *owp, *nowp, *towp;
1586 struct nfsclopen *op, *nop, *top;
1587 struct nfscldeleg *dp, *ndp;
1588 int ret, printed = 0;
1591 * First, merge locally issued Opens into the list for the server.
1593 dp = TAILQ_FIRST(&clp->nfsc_deleg);
1594 while (dp != NULL) {
1595 ndp = TAILQ_NEXT(dp, nfsdl_list);
1596 owp = LIST_FIRST(&dp->nfsdl_owner);
1597 while (owp != NULL) {
1598 nowp = LIST_NEXT(owp, nfsow_list);
1599 op = LIST_FIRST(&owp->nfsow_open);
1601 if (LIST_NEXT(op, nfso_list) != NULL)
1603 LIST_FOREACH(towp, &clp->nfsc_owner, nfsow_list) {
1604 if (!NFSBCMP(towp->nfsow_owner, owp->nfsow_owner,
1605 NFSV4CL_LOCKNAMELEN))
1609 /* Merge opens in */
1610 LIST_FOREACH(top, &towp->nfsow_open, nfso_list) {
1611 if (top->nfso_fhlen == op->nfso_fhlen &&
1612 !NFSBCMP(top->nfso_fh, op->nfso_fh,
1614 top->nfso_mode |= op->nfso_mode;
1615 top->nfso_opencnt += op->nfso_opencnt;
1620 /* Just add the open to the owner list */
1621 LIST_REMOVE(op, nfso_list);
1622 op->nfso_own = towp;
1623 LIST_INSERT_HEAD(&towp->nfsow_open, op, nfso_list);
1624 newnfsstats.cllocalopens--;
1625 newnfsstats.clopens++;
1628 /* Just add the openowner to the client list */
1629 LIST_REMOVE(owp, nfsow_list);
1630 owp->nfsow_clp = clp;
1631 LIST_INSERT_HEAD(&clp->nfsc_owner, owp, nfsow_list);
1632 newnfsstats.cllocalopenowners--;
1633 newnfsstats.clopenowners++;
1634 newnfsstats.cllocalopens--;
1635 newnfsstats.clopens++;
1640 if (!printed && !LIST_EMPTY(&dp->nfsdl_lock)) {
1642 printf("nfsv4 expired locks lost\n");
1644 nfscl_cleandeleg(dp);
1645 nfscl_freedeleg(&clp->nfsc_deleg, dp);
1648 if (!TAILQ_EMPTY(&clp->nfsc_deleg))
1652 * Now, try and reopen against the server.
1654 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1655 owp->nfsow_seqid = 0;
1656 LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
1657 ret = nfscl_expireopen(clp, op, nmp, cred, p);
1658 if (ret && !printed) {
1660 printf("nfsv4 expired locks lost\n");
1663 if (LIST_EMPTY(&owp->nfsow_open))
1664 nfscl_freeopenowner(owp, 0);
1669 * This function must be called after the process represented by "own" has
1670 * exited. Must be called with CLSTATE lock held.
1673 nfscl_cleanup_common(struct nfsclclient *clp, u_int8_t *own)
1675 struct nfsclowner *owp, *nowp;
1676 struct nfscllockowner *lp, *nlp;
1677 struct nfscldeleg *dp;
1679 /* First, get rid of local locks on delegations. */
1680 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
1681 LIST_FOREACH_SAFE(lp, &dp->nfsdl_lock, nfsl_list, nlp) {
1682 if (!NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
1683 if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED))
1685 nfscl_freelockowner(lp, 1);
1689 owp = LIST_FIRST(&clp->nfsc_owner);
1690 while (owp != NULL) {
1691 nowp = LIST_NEXT(owp, nfsow_list);
1692 if (!NFSBCMP(owp->nfsow_owner, own,
1693 NFSV4CL_LOCKNAMELEN)) {
1695 * If there are children that haven't closed the
1696 * file descriptors yet, the opens will still be
1697 * here. For that case, let the renew thread clear
1698 * out the OpenOwner later.
1700 if (LIST_EMPTY(&owp->nfsow_open))
1701 nfscl_freeopenowner(owp, 0);
1703 owp->nfsow_defunct = 1;
1710 * Find open/lock owners for processes that have exited.
1713 nfscl_cleanupkext(struct nfsclclient *clp, struct nfscllockownerfhhead *lhp)
1715 struct nfsclowner *owp, *nowp;
1716 struct nfsclopen *op;
1717 struct nfscllockowner *lp, *nlp;
1721 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1722 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1723 LIST_FOREACH_SAFE(lp, &op->nfso_lock, nfsl_list, nlp) {
1724 if (LIST_EMPTY(&lp->nfsl_lock))
1725 nfscl_emptylockowner(lp, lhp);
1728 if (nfscl_procdoesntexist(owp->nfsow_owner))
1729 nfscl_cleanup_common(clp, owp->nfsow_owner);
1732 NFSPROCLISTUNLOCK();
1736 * Take the empty lock owner and move it to the local lhp list if the
1737 * associated process no longer exists.
1740 nfscl_emptylockowner(struct nfscllockowner *lp,
1741 struct nfscllockownerfhhead *lhp)
1743 struct nfscllockownerfh *lfhp, *mylfhp;
1744 struct nfscllockowner *nlp;
1747 /* If not a Posix lock owner, just return. */
1748 if ((lp->nfsl_lockflags & F_POSIX) == 0)
1754 * First, search to see if this lock owner is already in the list.
1755 * If it is, then the associated process no longer exists.
1757 SLIST_FOREACH(lfhp, lhp, nfslfh_list) {
1758 if (lfhp->nfslfh_len == lp->nfsl_open->nfso_fhlen &&
1759 !NFSBCMP(lfhp->nfslfh_fh, lp->nfsl_open->nfso_fh,
1762 LIST_FOREACH(nlp, &lfhp->nfslfh_lock, nfsl_list)
1763 if (!NFSBCMP(nlp->nfsl_owner, lp->nfsl_owner,
1764 NFSV4CL_LOCKNAMELEN))
1767 /* If not found, check if process still exists. */
1768 if (fnd_it == 0 && nfscl_procdoesntexist(lp->nfsl_owner) == 0)
1771 /* Move the lock owner over to the local list. */
1772 if (mylfhp == NULL) {
1773 mylfhp = malloc(sizeof(struct nfscllockownerfh), M_TEMP,
1777 mylfhp->nfslfh_len = lp->nfsl_open->nfso_fhlen;
1778 NFSBCOPY(lp->nfsl_open->nfso_fh, mylfhp->nfslfh_fh,
1779 mylfhp->nfslfh_len);
1780 LIST_INIT(&mylfhp->nfslfh_lock);
1781 SLIST_INSERT_HEAD(lhp, mylfhp, nfslfh_list);
1783 LIST_REMOVE(lp, nfsl_list);
1784 LIST_INSERT_HEAD(&mylfhp->nfslfh_lock, lp, nfsl_list);
1787 static int fake_global; /* Used to force visibility of MNTK_UNMOUNTF */
1789 * Called from nfs umount to free up the clientid.
1792 nfscl_umount(struct nfsmount *nmp, NFSPROC_T *p)
1794 struct nfsclclient *clp;
1799 * For the case that matters, this is the thread that set
1800 * MNTK_UNMOUNTF, so it will see it set. The code that follows is
1801 * done to ensure that any thread executing nfscl_getcl() after
1802 * this time, will see MNTK_UNMOUNTF set. nfscl_getcl() uses the
1803 * mutex for NFSLOCKCLSTATE(), so it is "m" for the following
1804 * explanation, courtesy of Alan Cox.
1805 * What follows is a snippet from Alan Cox's email at:
1806 * http://docs.FreeBSD.org/cgi/
1807 * mid.cgi?BANLkTikR3d65zPHo9==08ZfJ2vmqZucEvw
1809 * 1. Set MNTK_UNMOUNTF
1810 * 2. Acquire a standard FreeBSD mutex "m".
1811 * 3. Update some data structures.
1812 * 4. Release mutex "m".
1814 * Then, other threads that acquire "m" after step 4 has occurred will
1815 * see MNTK_UNMOUNTF as set. But, other threads that beat thread X to
1816 * step 2 may or may not see MNTK_UNMOUNTF as set.
1819 if ((nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) {
1827 if ((clp->nfsc_flags & NFSCLFLAGS_INITED) == 0)
1828 panic("nfscl umount");
1831 * First, handshake with the nfscl renew thread, to terminate
1834 clp->nfsc_flags |= NFSCLFLAGS_UMOUNT;
1835 while (clp->nfsc_flags & NFSCLFLAGS_HASTHREAD)
1836 (void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT,
1840 * Now, get the exclusive lock on the client state, so
1841 * that no uses of the state are still in progress.
1844 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
1845 NFSCLSTATEMUTEXPTR, NULL);
1846 } while (!igotlock);
1850 * Free up all the state. It will expire on the server, but
1851 * maybe we should do a SetClientId/SetClientIdConfirm so
1852 * the server throws it away?
1854 LIST_REMOVE(clp, nfsc_list);
1855 nfscl_delegreturnall(clp, p);
1856 cred = newnfs_getcred();
1857 if (NFSHASNFSV4N(nmp)) {
1858 (void)nfsrpc_destroysession(nmp, clp, cred, p);
1859 (void)nfsrpc_destroyclient(nmp, clp, cred, p);
1861 (void)nfsrpc_setclient(nmp, clp, 0, cred, p);
1862 nfscl_cleanclient(clp);
1865 free(clp, M_NFSCLCLIENT);
1871 * This function is called when a server replies with NFSERR_STALECLIENTID
1872 * NFSERR_STALESTATEID or NFSERR_BADSESSION. It traverses the clientid lists,
1873 * doing Opens and Locks with reclaim. If these fail, it deletes the
1874 * corresponding state.
1877 nfscl_recover(struct nfsclclient *clp, struct ucred *cred, NFSPROC_T *p)
1879 struct nfsclowner *owp, *nowp;
1880 struct nfsclopen *op, *nop;
1881 struct nfscllockowner *lp, *nlp;
1882 struct nfscllock *lop, *nlop;
1883 struct nfscldeleg *dp, *ndp, *tdp;
1884 struct nfsmount *nmp;
1885 struct ucred *tcred;
1886 struct nfsclopenhead extra_open;
1887 struct nfscldeleghead extra_deleg;
1890 u_int32_t delegtype = NFSV4OPEN_DELEGATEWRITE, mode;
1891 int i, igotlock = 0, error, trycnt, firstlock;
1892 struct nfscllayout *lyp, *nlyp;
1895 * First, lock the client structure, so everyone else will
1896 * block when trying to use state.
1899 clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG;
1901 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
1902 NFSCLSTATEMUTEXPTR, NULL);
1903 } while (!igotlock);
1906 nmp = clp->nfsc_nmp;
1908 panic("nfscl recover");
1911 * For now, just get rid of all layouts. There may be a need
1912 * to do LayoutCommit Ops with reclaim == true later.
1914 TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp)
1915 nfscl_freelayout(lyp);
1916 TAILQ_INIT(&clp->nfsc_layout);
1917 for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++)
1918 LIST_INIT(&clp->nfsc_layouthash[i]);
1922 error = nfsrpc_setclient(nmp, clp, 1, cred, p);
1923 } while ((error == NFSERR_STALECLIENTID ||
1924 error == NFSERR_BADSESSION ||
1925 error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
1927 nfscl_cleanclient(clp);
1929 clp->nfsc_flags &= ~(NFSCLFLAGS_HASCLIENTID |
1930 NFSCLFLAGS_RECOVER | NFSCLFLAGS_RECVRINPROG);
1931 wakeup(&clp->nfsc_flags);
1932 nfsv4_unlock(&clp->nfsc_lock, 0);
1936 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
1937 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
1940 * Mark requests already queued on the server, so that they don't
1941 * initiate another recovery cycle. Any requests already in the
1942 * queue that handle state information will have the old stale
1943 * clientid/stateid and will get a NFSERR_STALESTATEID,
1944 * NFSERR_STALECLIENTID or NFSERR_BADSESSION reply from the server.
1945 * This will be translated to NFSERR_STALEDONTRECOVER when
1946 * R_DONTRECOVER is set.
1949 TAILQ_FOREACH(rep, &nfsd_reqq, r_chain) {
1950 if (rep->r_nmp == nmp)
1951 rep->r_flags |= R_DONTRECOVER;
1956 * Now, mark all delegations "need reclaim".
1958 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list)
1959 dp->nfsdl_flags |= NFSCLDL_NEEDRECLAIM;
1961 TAILQ_INIT(&extra_deleg);
1962 LIST_INIT(&extra_open);
1964 * Now traverse the state lists, doing Open and Lock Reclaims.
1966 tcred = newnfs_getcred();
1967 owp = LIST_FIRST(&clp->nfsc_owner);
1968 while (owp != NULL) {
1969 nowp = LIST_NEXT(owp, nfsow_list);
1970 owp->nfsow_seqid = 0;
1971 op = LIST_FIRST(&owp->nfsow_open);
1972 while (op != NULL) {
1973 nop = LIST_NEXT(op, nfso_list);
1974 if (error != NFSERR_NOGRACE) {
1975 /* Search for a delegation to reclaim with the open */
1976 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
1977 if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
1979 if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
1980 mode = NFSV4OPEN_ACCESSWRITE;
1981 delegtype = NFSV4OPEN_DELEGATEWRITE;
1983 mode = NFSV4OPEN_ACCESSREAD;
1984 delegtype = NFSV4OPEN_DELEGATEREAD;
1986 if ((op->nfso_mode & mode) == mode &&
1987 op->nfso_fhlen == dp->nfsdl_fhlen &&
1988 !NFSBCMP(op->nfso_fh, dp->nfsdl_fh, op->nfso_fhlen))
1993 delegtype = NFSV4OPEN_DELEGATENONE;
1994 newnfs_copycred(&op->nfso_cred, tcred);
1995 error = nfscl_tryopen(nmp, NULL, op->nfso_fh,
1996 op->nfso_fhlen, op->nfso_fh, op->nfso_fhlen,
1997 op->nfso_mode, op, NULL, 0, &ndp, 1, delegtype,
2000 /* Handle any replied delegation */
2001 if (ndp != NULL && ((ndp->nfsdl_flags & NFSCLDL_WRITE)
2002 || NFSMNT_RDONLY(nmp->nm_mountp))) {
2003 if ((ndp->nfsdl_flags & NFSCLDL_WRITE))
2004 mode = NFSV4OPEN_ACCESSWRITE;
2006 mode = NFSV4OPEN_ACCESSREAD;
2007 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2008 if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
2010 if ((op->nfso_mode & mode) == mode &&
2011 op->nfso_fhlen == dp->nfsdl_fhlen &&
2012 !NFSBCMP(op->nfso_fh, dp->nfsdl_fh,
2014 dp->nfsdl_stateid = ndp->nfsdl_stateid;
2015 dp->nfsdl_sizelimit = ndp->nfsdl_sizelimit;
2016 dp->nfsdl_ace = ndp->nfsdl_ace;
2017 dp->nfsdl_change = ndp->nfsdl_change;
2018 dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
2019 if ((ndp->nfsdl_flags & NFSCLDL_RECALL))
2020 dp->nfsdl_flags |= NFSCLDL_RECALL;
2021 FREE((caddr_t)ndp, M_NFSCLDELEG);
2028 TAILQ_INSERT_HEAD(&extra_deleg, ndp, nfsdl_list);
2030 /* and reclaim all byte range locks */
2031 lp = LIST_FIRST(&op->nfso_lock);
2032 while (lp != NULL) {
2033 nlp = LIST_NEXT(lp, nfsl_list);
2036 lop = LIST_FIRST(&lp->nfsl_lock);
2037 while (lop != NULL) {
2038 nlop = LIST_NEXT(lop, nfslo_list);
2039 if (lop->nfslo_end == NFS64BITSSET)
2042 len = lop->nfslo_end - lop->nfslo_first;
2043 if (error != NFSERR_NOGRACE)
2044 error = nfscl_trylock(nmp, NULL,
2045 op->nfso_fh, op->nfso_fhlen, lp,
2046 firstlock, 1, lop->nfslo_first, len,
2047 lop->nfslo_type, tcred, p);
2049 nfscl_freelock(lop, 0);
2054 /* If no locks, but a lockowner, just delete it. */
2055 if (LIST_EMPTY(&lp->nfsl_lock))
2056 nfscl_freelockowner(lp, 0);
2060 nfscl_freeopen(op, 0);
2069 * Now, try and get any delegations not yet reclaimed by cobbling
2070 * to-gether an appropriate open.
2073 dp = TAILQ_FIRST(&clp->nfsc_deleg);
2074 while (dp != NULL) {
2075 ndp = TAILQ_NEXT(dp, nfsdl_list);
2076 if ((dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM)) {
2078 MALLOC(nowp, struct nfsclowner *,
2079 sizeof (struct nfsclowner), M_NFSCLOWNER, M_WAITOK);
2081 * Name must be as long an largest possible
2082 * NFSV4CL_LOCKNAMELEN. 12 for now.
2084 NFSBCOPY("RECLAIMDELEG", nowp->nfsow_owner,
2085 NFSV4CL_LOCKNAMELEN);
2086 LIST_INIT(&nowp->nfsow_open);
2087 nowp->nfsow_clp = clp;
2088 nowp->nfsow_seqid = 0;
2089 nowp->nfsow_defunct = 0;
2090 nfscl_lockinit(&nowp->nfsow_rwlock);
2093 if (error != NFSERR_NOGRACE) {
2094 MALLOC(nop, struct nfsclopen *, sizeof (struct nfsclopen) +
2095 dp->nfsdl_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
2096 nop->nfso_own = nowp;
2097 if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
2098 nop->nfso_mode = NFSV4OPEN_ACCESSWRITE;
2099 delegtype = NFSV4OPEN_DELEGATEWRITE;
2101 nop->nfso_mode = NFSV4OPEN_ACCESSREAD;
2102 delegtype = NFSV4OPEN_DELEGATEREAD;
2104 nop->nfso_opencnt = 0;
2105 nop->nfso_posixlock = 1;
2106 nop->nfso_fhlen = dp->nfsdl_fhlen;
2107 NFSBCOPY(dp->nfsdl_fh, nop->nfso_fh, dp->nfsdl_fhlen);
2108 LIST_INIT(&nop->nfso_lock);
2109 nop->nfso_stateid.seqid = 0;
2110 nop->nfso_stateid.other[0] = 0;
2111 nop->nfso_stateid.other[1] = 0;
2112 nop->nfso_stateid.other[2] = 0;
2113 newnfs_copycred(&dp->nfsdl_cred, tcred);
2114 newnfs_copyincred(tcred, &nop->nfso_cred);
2116 error = nfscl_tryopen(nmp, NULL, nop->nfso_fh,
2117 nop->nfso_fhlen, nop->nfso_fh, nop->nfso_fhlen,
2118 nop->nfso_mode, nop, NULL, 0, &tdp, 1,
2119 delegtype, tcred, p);
2121 if ((tdp->nfsdl_flags & NFSCLDL_WRITE))
2122 mode = NFSV4OPEN_ACCESSWRITE;
2124 mode = NFSV4OPEN_ACCESSREAD;
2125 if ((nop->nfso_mode & mode) == mode &&
2126 nop->nfso_fhlen == tdp->nfsdl_fhlen &&
2127 !NFSBCMP(nop->nfso_fh, tdp->nfsdl_fh,
2129 dp->nfsdl_stateid = tdp->nfsdl_stateid;
2130 dp->nfsdl_sizelimit = tdp->nfsdl_sizelimit;
2131 dp->nfsdl_ace = tdp->nfsdl_ace;
2132 dp->nfsdl_change = tdp->nfsdl_change;
2133 dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
2134 if ((tdp->nfsdl_flags & NFSCLDL_RECALL))
2135 dp->nfsdl_flags |= NFSCLDL_RECALL;
2136 FREE((caddr_t)tdp, M_NFSCLDELEG);
2138 TAILQ_INSERT_HEAD(&extra_deleg, tdp, nfsdl_list);
2144 FREE((caddr_t)nop, M_NFSCLOPEN);
2146 * Couldn't reclaim it, so throw the state
2149 nfscl_cleandeleg(dp);
2150 nfscl_freedeleg(&clp->nfsc_deleg, dp);
2152 LIST_INSERT_HEAD(&extra_open, nop, nfso_list);
2159 * Now, get rid of extra Opens and Delegations.
2161 LIST_FOREACH_SAFE(op, &extra_open, nfso_list, nop) {
2163 newnfs_copycred(&op->nfso_cred, tcred);
2164 error = nfscl_tryclose(op, tcred, nmp, p);
2165 if (error == NFSERR_GRACE)
2166 (void) nfs_catnap(PZERO, error, "nfsexcls");
2167 } while (error == NFSERR_GRACE);
2168 LIST_REMOVE(op, nfso_list);
2169 FREE((caddr_t)op, M_NFSCLOPEN);
2172 FREE((caddr_t)nowp, M_NFSCLOWNER);
2174 TAILQ_FOREACH_SAFE(dp, &extra_deleg, nfsdl_list, ndp) {
2176 newnfs_copycred(&dp->nfsdl_cred, tcred);
2177 error = nfscl_trydelegreturn(dp, tcred, nmp, p);
2178 if (error == NFSERR_GRACE)
2179 (void) nfs_catnap(PZERO, error, "nfsexdlg");
2180 } while (error == NFSERR_GRACE);
2181 TAILQ_REMOVE(&extra_deleg, dp, nfsdl_list);
2182 FREE((caddr_t)dp, M_NFSCLDELEG);
2185 /* For NFSv4.1 or later, do a RECLAIM_COMPLETE. */
2186 if (NFSHASNFSV4N(nmp))
2187 (void)nfsrpc_reclaimcomplete(nmp, cred, p);
2190 clp->nfsc_flags &= ~NFSCLFLAGS_RECVRINPROG;
2191 wakeup(&clp->nfsc_flags);
2192 nfsv4_unlock(&clp->nfsc_lock, 0);
2198 * This function is called when a server replies with NFSERR_EXPIRED.
2199 * It deletes all state for the client and does a fresh SetClientId/confirm.
2200 * XXX Someday it should post a signal to the process(es) that hold the
2201 * state, so they know that lock state has been lost.
2204 nfscl_hasexpired(struct nfsclclient *clp, u_int32_t clidrev, NFSPROC_T *p)
2206 struct nfsmount *nmp;
2208 int igotlock = 0, error, trycnt;
2211 * If the clientid has gone away or a new SetClientid has already
2212 * been done, just return ok.
2214 if (clp == NULL || clidrev != clp->nfsc_clientidrev)
2218 * First, lock the client structure, so everyone else will
2219 * block when trying to use state. Also, use NFSCLFLAGS_EXPIREIT so
2220 * that only one thread does the work.
2223 clp->nfsc_flags |= NFSCLFLAGS_EXPIREIT;
2225 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
2226 NFSCLSTATEMUTEXPTR, NULL);
2227 } while (!igotlock && (clp->nfsc_flags & NFSCLFLAGS_EXPIREIT));
2228 if ((clp->nfsc_flags & NFSCLFLAGS_EXPIREIT) == 0) {
2230 nfsv4_unlock(&clp->nfsc_lock, 0);
2234 clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG;
2237 nmp = clp->nfsc_nmp;
2239 panic("nfscl expired");
2240 cred = newnfs_getcred();
2243 error = nfsrpc_setclient(nmp, clp, 0, cred, p);
2244 } while ((error == NFSERR_STALECLIENTID ||
2245 error == NFSERR_BADSESSION ||
2246 error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
2249 * Clear out any state.
2251 nfscl_cleanclient(clp);
2253 clp->nfsc_flags &= ~(NFSCLFLAGS_HASCLIENTID |
2254 NFSCLFLAGS_RECOVER);
2257 * Expire the state for the client.
2259 nfscl_expireclient(clp, nmp, cred, p);
2261 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
2262 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2264 clp->nfsc_flags &= ~(NFSCLFLAGS_EXPIREIT | NFSCLFLAGS_RECVRINPROG);
2265 wakeup(&clp->nfsc_flags);
2266 nfsv4_unlock(&clp->nfsc_lock, 0);
2273 * This function inserts a lock in the list after insert_lop.
2276 nfscl_insertlock(struct nfscllockowner *lp, struct nfscllock *new_lop,
2277 struct nfscllock *insert_lop, int local)
2280 if ((struct nfscllockowner *)insert_lop == lp)
2281 LIST_INSERT_HEAD(&lp->nfsl_lock, new_lop, nfslo_list);
2283 LIST_INSERT_AFTER(insert_lop, new_lop, nfslo_list);
2285 newnfsstats.cllocallocks++;
2287 newnfsstats.cllocks++;
2291 * This function updates the locking for a lock owner and given file. It
2292 * maintains a list of lock ranges ordered on increasing file offset that
2293 * are NFSCLLOCK_READ or NFSCLLOCK_WRITE and non-overlapping (aka POSIX style).
2294 * It always adds new_lop to the list and sometimes uses the one pointed
2296 * Returns 1 if the locks were modified, 0 otherwise.
2299 nfscl_updatelock(struct nfscllockowner *lp, struct nfscllock **new_lopp,
2300 struct nfscllock **other_lopp, int local)
2302 struct nfscllock *new_lop = *new_lopp;
2303 struct nfscllock *lop, *tlop, *ilop;
2304 struct nfscllock *other_lop;
2305 int unlock = 0, modified = 0;
2309 * Work down the list until the lock is merged.
2311 if (new_lop->nfslo_type == F_UNLCK)
2313 ilop = (struct nfscllock *)lp;
2314 lop = LIST_FIRST(&lp->nfsl_lock);
2315 while (lop != NULL) {
2317 * Only check locks for this file that aren't before the start of
2320 if (lop->nfslo_end >= new_lop->nfslo_first) {
2321 if (new_lop->nfslo_end < lop->nfslo_first) {
2323 * If the new lock ends before the start of the
2324 * current lock's range, no merge, just insert
2329 if (new_lop->nfslo_type == lop->nfslo_type ||
2330 (new_lop->nfslo_first <= lop->nfslo_first &&
2331 new_lop->nfslo_end >= lop->nfslo_end)) {
2333 * This lock can be absorbed by the new lock/unlock.
2334 * This happens when it covers the entire range
2335 * of the old lock or is contiguous
2336 * with the old lock and is of the same type or an
2339 if (new_lop->nfslo_type != lop->nfslo_type ||
2340 new_lop->nfslo_first != lop->nfslo_first ||
2341 new_lop->nfslo_end != lop->nfslo_end)
2343 if (lop->nfslo_first < new_lop->nfslo_first)
2344 new_lop->nfslo_first = lop->nfslo_first;
2345 if (lop->nfslo_end > new_lop->nfslo_end)
2346 new_lop->nfslo_end = lop->nfslo_end;
2348 lop = LIST_NEXT(lop, nfslo_list);
2349 nfscl_freelock(tlop, local);
2354 * All these cases are for contiguous locks that are not the
2355 * same type, so they can't be merged.
2357 if (new_lop->nfslo_first <= lop->nfslo_first) {
2359 * This case is where the new lock overlaps with the
2360 * first part of the old lock. Move the start of the
2361 * old lock to just past the end of the new lock. The
2362 * new lock will be inserted in front of the old, since
2363 * ilop hasn't been updated. (We are done now.)
2365 if (lop->nfslo_first != new_lop->nfslo_end) {
2366 lop->nfslo_first = new_lop->nfslo_end;
2371 if (new_lop->nfslo_end >= lop->nfslo_end) {
2373 * This case is where the new lock overlaps with the
2374 * end of the old lock's range. Move the old lock's
2375 * end to just before the new lock's first and insert
2376 * the new lock after the old lock.
2377 * Might not be done yet, since the new lock could
2378 * overlap further locks with higher ranges.
2380 if (lop->nfslo_end != new_lop->nfslo_first) {
2381 lop->nfslo_end = new_lop->nfslo_first;
2385 lop = LIST_NEXT(lop, nfslo_list);
2389 * The final case is where the new lock's range is in the
2390 * middle of the current lock's and splits the current lock
2391 * up. Use *other_lopp to handle the second part of the
2392 * split old lock range. (We are done now.)
2393 * For unlock, we use new_lop as other_lop and tmp, since
2394 * other_lop and new_lop are the same for this case.
2395 * We noted the unlock case above, so we don't need
2396 * new_lop->nfslo_type any longer.
2398 tmp = new_lop->nfslo_first;
2400 other_lop = new_lop;
2403 other_lop = *other_lopp;
2406 other_lop->nfslo_first = new_lop->nfslo_end;
2407 other_lop->nfslo_end = lop->nfslo_end;
2408 other_lop->nfslo_type = lop->nfslo_type;
2409 lop->nfslo_end = tmp;
2410 nfscl_insertlock(lp, other_lop, lop, local);
2416 lop = LIST_NEXT(lop, nfslo_list);
2422 * Insert the new lock in the list at the appropriate place.
2425 nfscl_insertlock(lp, new_lop, ilop, local);
2433 * This function must be run as a kernel thread.
2434 * It does Renew Ops and recovery, when required.
2437 nfscl_renewthread(struct nfsclclient *clp, NFSPROC_T *p)
2439 struct nfsclowner *owp, *nowp;
2440 struct nfsclopen *op;
2441 struct nfscllockowner *lp, *nlp;
2442 struct nfscldeleghead dh;
2443 struct nfscldeleg *dp, *ndp;
2446 int error, cbpathdown, islept, igotlock, ret, clearok;
2447 uint32_t recover_done_time = 0;
2449 static time_t prevsec = 0;
2450 struct nfscllockownerfh *lfhp, *nlfhp;
2451 struct nfscllockownerfhhead lfh;
2452 struct nfscllayout *lyp, *nlyp;
2453 struct nfscldevinfo *dip, *ndip;
2454 struct nfscllayouthead rlh;
2455 struct nfsclrecalllayout *recallp;
2456 struct nfsclds *dsp;
2458 cred = newnfs_getcred();
2460 clp->nfsc_flags |= NFSCLFLAGS_HASTHREAD;
2463 newnfs_setroot(cred);
2465 if (clp->nfsc_flags & NFSCLFLAGS_RECOVER) {
2467 * Only allow one recover within 1/2 of the lease
2468 * duration (nfsc_renew).
2470 if (recover_done_time < NFSD_MONOSEC) {
2471 recover_done_time = NFSD_MONOSEC +
2473 nfscl_recover(clp, cred, p);
2476 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2480 if (clp->nfsc_expire <= NFSD_MONOSEC &&
2481 (clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID)) {
2482 clp->nfsc_expire = NFSD_MONOSEC + clp->nfsc_renew;
2483 clidrev = clp->nfsc_clientidrev;
2484 error = nfsrpc_renew(clp,
2485 TAILQ_FIRST(&clp->nfsc_nmp->nm_sess), cred, p);
2486 if (error == NFSERR_CBPATHDOWN)
2488 else if (error == NFSERR_STALECLIENTID ||
2489 error == NFSERR_BADSESSION) {
2491 clp->nfsc_flags |= NFSCLFLAGS_RECOVER;
2493 } else if (error == NFSERR_EXPIRED)
2494 (void) nfscl_hasexpired(clp, clidrev, p);
2497 /* Do renews for any DS sessions. */
2499 NFSLOCKMNT(clp->nfsc_nmp);
2500 /* Skip first entry, since the MDS is handled above. */
2501 dsp = TAILQ_FIRST(&clp->nfsc_nmp->nm_sess);
2503 dsp = TAILQ_NEXT(dsp, nfsclds_list);
2504 while (dsp != NULL) {
2505 if (dsp->nfsclds_expire <= NFSD_MONOSEC) {
2506 dsp->nfsclds_expire = NFSD_MONOSEC +
2508 NFSUNLOCKMNT(clp->nfsc_nmp);
2509 (void)nfsrpc_renew(clp, dsp, cred, p);
2512 dsp = TAILQ_NEXT(dsp, nfsclds_list);
2514 NFSUNLOCKMNT(clp->nfsc_nmp);
2519 /* It's a Total Recall! */
2520 nfscl_totalrecall(clp);
2523 * Now, handle defunct owners.
2525 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
2526 if (LIST_EMPTY(&owp->nfsow_open)) {
2527 if (owp->nfsow_defunct != 0)
2528 nfscl_freeopenowner(owp, 0);
2533 * Do the recall on any delegations. To avoid trouble, always
2534 * come back up here after having slept.
2538 dp = TAILQ_FIRST(&clp->nfsc_deleg);
2539 while (dp != NULL) {
2540 ndp = TAILQ_NEXT(dp, nfsdl_list);
2541 if ((dp->nfsdl_flags & NFSCLDL_RECALL)) {
2543 * Wait for outstanding I/O ops to be done.
2545 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
2547 nfsv4_unlock(&clp->nfsc_lock, 0);
2550 dp->nfsdl_rwlock.nfslock_lock |=
2552 (void) nfsmsleep(&dp->nfsdl_rwlock,
2553 NFSCLSTATEMUTEXPTR, PZERO, "nfscld",
2558 igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
2559 &islept, NFSCLSTATEMUTEXPTR, NULL);
2564 newnfs_copycred(&dp->nfsdl_cred, cred);
2565 ret = nfscl_recalldeleg(clp, clp->nfsc_nmp, dp,
2568 nfscl_cleandeleg(dp);
2569 TAILQ_REMOVE(&clp->nfsc_deleg, dp,
2571 LIST_REMOVE(dp, nfsdl_hash);
2572 TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
2574 newnfsstats.cldelegates--;
2582 * Clear out old delegations, if we are above the high water
2583 * mark. Only clear out ones with no state related to them.
2584 * The tailq list is in LRU order.
2586 dp = TAILQ_LAST(&clp->nfsc_deleg, nfscldeleghead);
2587 while (nfscl_delegcnt > nfscl_deleghighwater && dp != NULL) {
2588 ndp = TAILQ_PREV(dp, nfscldeleghead, nfsdl_list);
2589 if (dp->nfsdl_rwlock.nfslock_usecnt == 0 &&
2590 dp->nfsdl_rwlock.nfslock_lock == 0 &&
2591 dp->nfsdl_timestamp < NFSD_MONOSEC &&
2592 (dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_ZAPPED |
2593 NFSCLDL_NEEDRECLAIM | NFSCLDL_DELEGRET)) == 0) {
2595 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
2596 op = LIST_FIRST(&owp->nfsow_open);
2603 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
2604 if (!LIST_EMPTY(&lp->nfsl_lock)) {
2611 TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
2612 LIST_REMOVE(dp, nfsdl_hash);
2613 TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
2615 newnfsstats.cldelegates--;
2621 nfsv4_unlock(&clp->nfsc_lock, 0);
2624 * Do the recall on any layouts. To avoid trouble, always
2625 * come back up here after having slept.
2629 TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp) {
2630 if ((lyp->nfsly_flags & NFSLY_RECALL) != 0) {
2632 * Wait for outstanding I/O ops to be done.
2634 if (lyp->nfsly_lock.nfslock_usecnt > 0 ||
2635 (lyp->nfsly_lock.nfslock_lock &
2636 NFSV4LOCK_LOCK) != 0) {
2637 lyp->nfsly_lock.nfslock_lock |=
2639 (void)nfsmsleep(&lyp->nfsly_lock,
2640 NFSCLSTATEMUTEXPTR, PZERO, "nfslyp",
2644 /* Move the layout to the recall list. */
2645 TAILQ_REMOVE(&clp->nfsc_layout, lyp,
2647 LIST_REMOVE(lyp, nfsly_hash);
2648 TAILQ_INSERT_HEAD(&rlh, lyp, nfsly_list);
2650 /* Handle any layout commits. */
2651 if (!NFSHASNOLAYOUTCOMMIT(clp->nfsc_nmp) &&
2652 (lyp->nfsly_flags & NFSLY_WRITTEN) != 0) {
2653 lyp->nfsly_flags &= ~NFSLY_WRITTEN;
2655 NFSCL_DEBUG(3, "do layoutcommit\n");
2656 nfscl_dolayoutcommit(clp->nfsc_nmp, lyp,
2664 /* Now, look for stale layouts. */
2665 lyp = TAILQ_LAST(&clp->nfsc_layout, nfscllayouthead);
2666 while (lyp != NULL) {
2667 nlyp = TAILQ_PREV(lyp, nfscllayouthead, nfsly_list);
2668 if (lyp->nfsly_timestamp < NFSD_MONOSEC &&
2669 (lyp->nfsly_flags & NFSLY_RECALL) == 0 &&
2670 lyp->nfsly_lock.nfslock_usecnt == 0 &&
2671 lyp->nfsly_lock.nfslock_lock == 0) {
2672 NFSCL_DEBUG(4, "ret stale lay=%d\n",
2674 recallp = malloc(sizeof(*recallp),
2675 M_NFSLAYRECALL, M_NOWAIT);
2676 if (recallp == NULL)
2678 (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE,
2679 lyp, NFSLAYOUTIOMODE_ANY, 0, UINT64_MAX,
2680 lyp->nfsly_stateid.seqid, recallp);
2686 * Free up any unreferenced device info structures.
2688 LIST_FOREACH_SAFE(dip, &clp->nfsc_devinfo, nfsdi_list, ndip) {
2689 if (dip->nfsdi_layoutrefs == 0 &&
2690 dip->nfsdi_refcnt == 0) {
2691 NFSCL_DEBUG(4, "freeing devinfo\n");
2692 LIST_REMOVE(dip, nfsdi_list);
2693 nfscl_freedevinfo(dip);
2698 /* Do layout return(s), as required. */
2699 TAILQ_FOREACH_SAFE(lyp, &rlh, nfsly_list, nlyp) {
2700 TAILQ_REMOVE(&rlh, lyp, nfsly_list);
2701 NFSCL_DEBUG(4, "ret layout\n");
2702 nfscl_layoutreturn(clp->nfsc_nmp, lyp, cred, p);
2703 nfscl_freelayout(lyp);
2707 * Delegreturn any delegations cleaned out or recalled.
2709 TAILQ_FOREACH_SAFE(dp, &dh, nfsdl_list, ndp) {
2710 newnfs_copycred(&dp->nfsdl_cred, cred);
2711 (void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
2712 TAILQ_REMOVE(&dh, dp, nfsdl_list);
2713 FREE((caddr_t)dp, M_NFSCLDELEG);
2718 * Call nfscl_cleanupkext() once per second to check for
2719 * open/lock owners where the process has exited.
2721 mytime = NFSD_MONOSEC;
2722 if (prevsec != mytime) {
2724 nfscl_cleanupkext(clp, &lfh);
2728 * Do a ReleaseLockOwner for all lock owners where the
2729 * associated process no longer exists, as found by
2730 * nfscl_cleanupkext().
2732 newnfs_setroot(cred);
2733 SLIST_FOREACH_SAFE(lfhp, &lfh, nfslfh_list, nlfhp) {
2734 LIST_FOREACH_SAFE(lp, &lfhp->nfslfh_lock, nfsl_list,
2736 (void)nfsrpc_rellockown(clp->nfsc_nmp, lp,
2737 lfhp->nfslfh_fh, lfhp->nfslfh_len, cred,
2739 nfscl_freelockowner(lp, 0);
2746 if ((clp->nfsc_flags & NFSCLFLAGS_RECOVER) == 0)
2747 (void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT, "nfscl",
2749 if (clp->nfsc_flags & NFSCLFLAGS_UMOUNT) {
2750 clp->nfsc_flags &= ~NFSCLFLAGS_HASTHREAD;
2753 wakeup((caddr_t)clp);
2761 * Initiate state recovery. Called when NFSERR_STALECLIENTID,
2762 * NFSERR_STALESTATEID or NFSERR_BADSESSION is received.
2765 nfscl_initiate_recovery(struct nfsclclient *clp)
2771 clp->nfsc_flags |= NFSCLFLAGS_RECOVER;
2773 wakeup((caddr_t)clp);
2777 * Dump out the state stuff for debugging.
2780 nfscl_dumpstate(struct nfsmount *nmp, int openowner, int opens,
2781 int lockowner, int locks)
2783 struct nfsclclient *clp;
2784 struct nfsclowner *owp;
2785 struct nfsclopen *op;
2786 struct nfscllockowner *lp;
2787 struct nfscllock *lop;
2788 struct nfscldeleg *dp;
2792 printf("nfscl dumpstate NULL clp\n");
2796 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2797 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
2798 if (openowner && !LIST_EMPTY(&owp->nfsow_open))
2799 printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
2800 owp->nfsow_owner[0], owp->nfsow_owner[1],
2801 owp->nfsow_owner[2], owp->nfsow_owner[3],
2803 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
2805 printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
2806 op->nfso_stateid.other[0], op->nfso_stateid.other[1],
2807 op->nfso_stateid.other[2], op->nfso_opencnt,
2809 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
2811 printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
2812 lp->nfsl_owner[0], lp->nfsl_owner[1],
2813 lp->nfsl_owner[2], lp->nfsl_owner[3],
2815 lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
2816 lp->nfsl_stateid.other[2]);
2817 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
2820 printf("lck typ=%d fst=%ju end=%ju\n",
2821 lop->nfslo_type, (intmax_t)lop->nfslo_first,
2822 (intmax_t)lop->nfslo_end);
2824 printf("lck typ=%d fst=%qd end=%qd\n",
2825 lop->nfslo_type, lop->nfslo_first,
2833 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
2834 if (openowner && !LIST_EMPTY(&owp->nfsow_open))
2835 printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
2836 owp->nfsow_owner[0], owp->nfsow_owner[1],
2837 owp->nfsow_owner[2], owp->nfsow_owner[3],
2839 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
2841 printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
2842 op->nfso_stateid.other[0], op->nfso_stateid.other[1],
2843 op->nfso_stateid.other[2], op->nfso_opencnt,
2845 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
2847 printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
2848 lp->nfsl_owner[0], lp->nfsl_owner[1],
2849 lp->nfsl_owner[2], lp->nfsl_owner[3],
2851 lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
2852 lp->nfsl_stateid.other[2]);
2853 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
2856 printf("lck typ=%d fst=%ju end=%ju\n",
2857 lop->nfslo_type, (intmax_t)lop->nfslo_first,
2858 (intmax_t)lop->nfslo_end);
2860 printf("lck typ=%d fst=%qd end=%qd\n",
2861 lop->nfslo_type, lop->nfslo_first,
2872 * Check for duplicate open owners and opens.
2873 * (Only used as a diagnostic aid.)
2876 nfscl_dupopen(vnode_t vp, int dupopens)
2878 struct nfsclclient *clp;
2879 struct nfsclowner *owp, *owp2;
2880 struct nfsclopen *op, *op2;
2883 clp = VFSTONFS(vnode_mount(vp))->nm_clp;
2885 printf("nfscl dupopen NULL clp\n");
2888 nfhp = VTONFS(vp)->n_fhp;
2892 * First, search for duplicate owners.
2893 * These should never happen!
2895 LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
2896 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
2898 !NFSBCMP(owp->nfsow_owner, owp2->nfsow_owner,
2899 NFSV4CL_LOCKNAMELEN)) {
2901 printf("DUP OWNER\n");
2902 nfscl_dumpstate(VFSTONFS(vnode_mount(vp)), 1, 1, 0, 0);
2909 * Now, search for duplicate stateids.
2910 * These shouldn't happen, either.
2912 LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
2913 LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
2914 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
2915 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
2917 (op->nfso_stateid.other[0] != 0 ||
2918 op->nfso_stateid.other[1] != 0 ||
2919 op->nfso_stateid.other[2] != 0) &&
2920 op->nfso_stateid.other[0] == op2->nfso_stateid.other[0] &&
2921 op->nfso_stateid.other[1] == op2->nfso_stateid.other[1] &&
2922 op->nfso_stateid.other[2] == op2->nfso_stateid.other[2]) {
2924 printf("DUP STATEID\n");
2925 nfscl_dumpstate(VFSTONFS(vnode_mount(vp)), 1, 1, 0,
2935 * Now search for duplicate opens.
2936 * Duplicate opens for the same owner
2937 * should never occur. Other duplicates are
2938 * possible and are checked for if "dupopens"
2941 LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
2942 LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
2943 if (nfhp->nfh_len == op2->nfso_fhlen &&
2944 !NFSBCMP(nfhp->nfh_fh, op2->nfso_fh, nfhp->nfh_len)) {
2945 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
2946 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
2947 if (op != op2 && nfhp->nfh_len == op->nfso_fhlen &&
2948 !NFSBCMP(nfhp->nfh_fh, op->nfso_fh, nfhp->nfh_len) &&
2949 (!NFSBCMP(op->nfso_own->nfsow_owner,
2950 op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN) ||
2952 if (!NFSBCMP(op->nfso_own->nfsow_owner,
2953 op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
2955 printf("BADDUP OPEN\n");
2958 printf("DUP OPEN\n");
2960 nfscl_dumpstate(VFSTONFS(vnode_mount(vp)), 1, 1,
2973 * During close, find an open that needs to be dereferenced and
2974 * dereference it. If there are no more opens for this file,
2975 * log a message to that effect.
2976 * Opens aren't actually Close'd until VOP_INACTIVE() is performed
2977 * on the file's vnode.
2978 * This is the safe way, since it is difficult to identify
2979 * which open the close is for and I/O can be performed after the
2980 * close(2) system call when a file is mmap'd.
2981 * If it returns 0 for success, there will be a referenced
2982 * clp returned via clpp.
2985 nfscl_getclose(vnode_t vp, struct nfsclclient **clpp)
2987 struct nfsclclient *clp;
2988 struct nfsclowner *owp;
2989 struct nfsclopen *op;
2990 struct nfscldeleg *dp;
2994 error = nfscl_getcl(vnode_mount(vp), NULL, NULL, 1, &clp);
2999 nfhp = VTONFS(vp)->n_fhp;
3003 * First, look for one under a delegation that was locally issued
3004 * and just decrement the opencnt for it. Since all my Opens against
3005 * the server are DENY_NONE, I don't see a problem with hanging
3006 * onto them. (It is much easier to use one of the extant Opens
3007 * that I already have on the server when a Delegation is recalled
3008 * than to do fresh Opens.) Someday, I might need to rethink this, but.
3010 dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
3012 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
3013 op = LIST_FIRST(&owp->nfsow_open);
3016 * Since a delegation is for a file, there
3017 * should never be more than one open for
3020 if (LIST_NEXT(op, nfso_list) != NULL)
3021 panic("nfscdeleg opens");
3022 if (notdecr && op->nfso_opencnt > 0) {
3031 /* Now process the opens against the server. */
3032 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3033 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3034 if (op->nfso_fhlen == nfhp->nfh_len &&
3035 !NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
3037 /* Found an open, decrement cnt if possible */
3038 if (notdecr && op->nfso_opencnt > 0) {
3043 * There are more opens, so just return.
3045 if (op->nfso_opencnt > 0) {
3054 printf("nfscl: never fnd open\n");
3059 nfscl_doclose(vnode_t vp, struct nfsclclient **clpp, NFSPROC_T *p)
3061 struct nfsclclient *clp;
3062 struct nfsclowner *owp, *nowp;
3063 struct nfsclopen *op;
3064 struct nfscldeleg *dp;
3068 error = nfscl_getcl(vnode_mount(vp), NULL, NULL, 1, &clp);
3073 nfhp = VTONFS(vp)->n_fhp;
3076 * First get rid of the local Open structures, which should be no
3079 dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
3081 LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) {
3082 op = LIST_FIRST(&owp->nfsow_open);
3084 KASSERT((op->nfso_opencnt == 0),
3085 ("nfscl: bad open cnt on deleg"));
3086 nfscl_freeopen(op, 1);
3088 nfscl_freeopenowner(owp, 1);
3092 /* Return any layouts marked return on close. */
3093 nfscl_retoncloselayout(clp, nfhp->nfh_fh, nfhp->nfh_len);
3095 /* Now process the opens against the server. */
3097 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3098 op = LIST_FIRST(&owp->nfsow_open);
3099 while (op != NULL) {
3100 if (op->nfso_fhlen == nfhp->nfh_len &&
3101 !NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
3103 /* Found an open, close it. */
3104 KASSERT((op->nfso_opencnt == 0),
3105 ("nfscl: bad open cnt on server"));
3107 nfsrpc_doclose(VFSTONFS(vnode_mount(vp)), op,
3112 op = LIST_NEXT(op, nfso_list);
3120 * Return all delegations on this client.
3121 * (Must be called with client sleep lock.)
3124 nfscl_delegreturnall(struct nfsclclient *clp, NFSPROC_T *p)
3126 struct nfscldeleg *dp, *ndp;
3129 cred = newnfs_getcred();
3130 TAILQ_FOREACH_SAFE(dp, &clp->nfsc_deleg, nfsdl_list, ndp) {
3131 nfscl_cleandeleg(dp);
3132 (void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
3133 nfscl_freedeleg(&clp->nfsc_deleg, dp);
3139 * Do a callback RPC.
3142 nfscl_docb(struct nfsrv_descript *nd, NFSPROC_T *p)
3144 int clist, gotseq_ok, i, j, k, op, rcalls;
3146 struct nfsclclient *clp;
3147 struct nfscldeleg *dp = NULL;
3148 int numops, taglen = -1, error = 0, trunc;
3149 u_int32_t minorvers = 0, retops = 0, *retopsp = NULL, *repp, cbident;
3150 u_char tag[NFSV4_SMALLSTR + 1], *tagstr;
3156 nfsattrbit_t attrbits, rattrbits;
3157 nfsv4stateid_t stateid;
3158 uint32_t seqid, slotid = 0, highslot, cachethis;
3159 uint8_t sessionid[NFSX_V4SESSIONID];
3161 struct nfscllayout *lyp;
3162 uint64_t filesid[2], len, off;
3163 int changed, gotone, laytype, recalltype;
3165 struct nfsclrecalllayout *recallp = NULL;
3169 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3170 taglen = fxdr_unsigned(int, *tl);
3175 if (taglen <= NFSV4_SMALLSTR)
3178 tagstr = malloc(taglen + 1, M_TEMP, M_WAITOK);
3179 error = nfsrv_mtostr(nd, tagstr, taglen);
3181 if (taglen > NFSV4_SMALLSTR)
3182 free(tagstr, M_TEMP);
3186 (void) nfsm_strtom(nd, tag, taglen);
3187 if (taglen > NFSV4_SMALLSTR) {
3188 free(tagstr, M_TEMP);
3190 NFSM_BUILD(retopsp, u_int32_t *, NFSX_UNSIGNED);
3191 NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3192 minorvers = fxdr_unsigned(u_int32_t, *tl++);
3193 if (minorvers != NFSV4_MINORVERSION && minorvers != NFSV41_MINORVERSION)
3194 nd->nd_repstat = NFSERR_MINORVERMISMATCH;
3195 cbident = fxdr_unsigned(u_int32_t, *tl++);
3199 numops = fxdr_unsigned(int, *tl);
3201 * Loop around doing the sub ops.
3203 for (i = 0; i < numops; i++) {
3204 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3205 NFSM_BUILD(repp, u_int32_t *, 2 * NFSX_UNSIGNED);
3207 op = fxdr_unsigned(int, *tl);
3208 if (op < NFSV4OP_CBGETATTR ||
3209 (op > NFSV4OP_CBRECALL && minorvers == NFSV4_MINORVERSION) ||
3210 (op > NFSV4OP_CBNOTIFYDEVID &&
3211 minorvers == NFSV41_MINORVERSION)) {
3212 nd->nd_repstat = NFSERR_OPILLEGAL;
3213 *repp = nfscl_errmap(nd, minorvers);
3217 nd->nd_procnum = op;
3218 if (op < NFSV4OP_CBNOPS)
3219 newnfsstats.cbrpccnt[nd->nd_procnum]++;
3221 case NFSV4OP_CBGETATTR:
3222 NFSCL_DEBUG(4, "cbgetattr\n");
3225 error = nfsm_getfh(nd, &nfhp);
3227 error = nfsrv_getattrbits(nd, &attrbits,
3229 if (error == 0 && i == 0 &&
3230 minorvers != NFSV4_MINORVERSION)
3231 error = NFSERR_OPNOTINSESS;
3233 mp = nfscl_getmnt(minorvers, sessionid, cbident,
3236 error = NFSERR_SERVERFAULT;
3239 error = nfscl_ngetreopen(mp, nfhp->nfh_fh,
3240 nfhp->nfh_len, p, &np);
3245 NFSZERO_ATTRBIT(&rattrbits);
3247 dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
3250 if (NFSISSET_ATTRBIT(&attrbits,
3253 va.va_size = np->n_size;
3257 NFSSETBIT_ATTRBIT(&rattrbits,
3260 if (NFSISSET_ATTRBIT(&attrbits,
3261 NFSATTRBIT_CHANGE)) {
3265 (np->n_flag & NDELEGMOD))
3267 NFSSETBIT_ATTRBIT(&rattrbits,
3271 error = NFSERR_SERVERFAULT;
3279 FREE((caddr_t)nfhp, M_NFSFH);
3281 (void) nfsv4_fillattr(nd, NULL, NULL, NULL, &va,
3282 NULL, 0, &rattrbits, NULL, p, 0, 0, 0, 0,
3285 case NFSV4OP_CBRECALL:
3286 NFSCL_DEBUG(4, "cbrecall\n");
3287 NFSM_DISSECT(tl, u_int32_t *, NFSX_STATEID +
3289 stateid.seqid = *tl++;
3290 NFSBCOPY((caddr_t)tl, (caddr_t)stateid.other,
3292 tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
3293 trunc = fxdr_unsigned(int, *tl);
3294 error = nfsm_getfh(nd, &nfhp);
3295 if (error == 0 && i == 0 &&
3296 minorvers != NFSV4_MINORVERSION)
3297 error = NFSERR_OPNOTINSESS;
3300 if (minorvers == NFSV4_MINORVERSION)
3301 clp = nfscl_getclnt(cbident);
3303 clp = nfscl_getclntsess(sessionid);
3305 dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
3307 if (dp != NULL && (dp->nfsdl_flags &
3308 NFSCLDL_DELEGRET) == 0) {
3311 wakeup((caddr_t)clp);
3314 error = NFSERR_SERVERFAULT;
3319 FREE((caddr_t)nfhp, M_NFSFH);
3321 case NFSV4OP_CBLAYOUTRECALL:
3322 NFSCL_DEBUG(4, "cblayrec\n");
3324 NFSM_DISSECT(tl, uint32_t *, 4 * NFSX_UNSIGNED);
3325 laytype = fxdr_unsigned(int, *tl++);
3326 iomode = fxdr_unsigned(uint32_t, *tl++);
3327 if (newnfs_true == *tl++)
3331 recalltype = fxdr_unsigned(int, *tl);
3332 recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL,
3334 if (laytype != NFSLAYOUT_NFSV4_1_FILES)
3335 error = NFSERR_NOMATCHLAYOUT;
3336 else if (recalltype == NFSLAYOUTRETURN_FILE) {
3337 error = nfsm_getfh(nd, &nfhp);
3338 NFSCL_DEBUG(4, "retfile getfh=%d\n", error);
3341 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_HYPER +
3343 off = fxdr_hyper(tl); tl += 2;
3344 len = fxdr_hyper(tl); tl += 2;
3345 stateid.seqid = fxdr_unsigned(uint32_t, *tl++);
3346 NFSBCOPY(tl, stateid.other, NFSX_STATEIDOTHER);
3347 if (minorvers == NFSV4_MINORVERSION)
3348 error = NFSERR_NOTSUPP;
3350 error = NFSERR_OPNOTINSESS;
3353 clp = nfscl_getclntsess(sessionid);
3354 NFSCL_DEBUG(4, "cbly clp=%p\n", clp);
3356 lyp = nfscl_findlayout(clp,
3359 NFSCL_DEBUG(4, "cblyp=%p\n",
3363 NFSLY_FILES) != 0 &&
3364 !NFSBCMP(stateid.other,
3365 lyp->nfsly_stateid.other,
3366 NFSX_STATEIDOTHER)) {
3380 NFSERR_NOMATCHLAYOUT;
3382 error = NFSERR_NOMATCHLAYOUT;
3385 free(nfhp, M_NFSFH);
3386 } else if (recalltype == NFSLAYOUTRETURN_FSID) {
3387 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_HYPER);
3388 filesid[0] = fxdr_hyper(tl); tl += 2;
3389 filesid[1] = fxdr_hyper(tl); tl += 2;
3392 clp = nfscl_getclntsess(sessionid);
3394 TAILQ_FOREACH(lyp, &clp->nfsc_layout,
3396 if (lyp->nfsly_filesid[0] ==
3398 lyp->nfsly_filesid[1] ==
3405 lyp->nfsly_stateid.seqid,
3414 error = NFSERR_NOMATCHLAYOUT;
3416 error = NFSERR_NOMATCHLAYOUT;
3418 } else if (recalltype == NFSLAYOUTRETURN_ALL) {
3421 clp = nfscl_getclntsess(sessionid);
3423 TAILQ_FOREACH(lyp, &clp->nfsc_layout,
3425 error = nfscl_layoutrecall(
3426 recalltype, lyp, iomode, 0,
3428 lyp->nfsly_stateid.seqid,
3436 error = NFSERR_NOMATCHLAYOUT;
3438 error = NFSERR_NOMATCHLAYOUT;
3441 error = NFSERR_NOMATCHLAYOUT;
3442 if (recallp != NULL) {
3443 free(recallp, M_NFSLAYRECALL);
3447 case NFSV4OP_CBSEQUENCE:
3448 NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID +
3450 bcopy(tl, sessionid, NFSX_V4SESSIONID);
3451 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3452 seqid = fxdr_unsigned(uint32_t, *tl++);
3453 slotid = fxdr_unsigned(uint32_t, *tl++);
3454 highslot = fxdr_unsigned(uint32_t, *tl++);
3456 /* Throw away the referring call stuff. */
3457 clist = fxdr_unsigned(int, *tl);
3458 for (j = 0; j < clist; j++) {
3459 NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID +
3461 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3462 rcalls = fxdr_unsigned(int, *tl);
3463 for (k = 0; k < rcalls; k++) {
3464 NFSM_DISSECT(tl, uint32_t *,
3470 clp = nfscl_getclntsess(sessionid);
3472 error = NFSERR_SERVERFAULT;
3474 error = NFSERR_SEQUENCEPOS;
3476 error = nfsv4_seqsession(seqid, slotid,
3478 NFSMNT_MDSSESSION(clp->nfsc_nmp)->
3479 nfsess_cbslots, &rep,
3480 NFSMNT_MDSSESSION(clp->nfsc_nmp)->
3486 NFSCL_DEBUG(4, "Got cbretry\n");
3487 m_freem(nd->nd_mreq);
3492 NFSM_BUILD(tl, uint32_t *,
3493 NFSX_V4SESSIONID + 4 * NFSX_UNSIGNED);
3494 bcopy(sessionid, tl, NFSX_V4SESSIONID);
3495 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3496 *tl++ = txdr_unsigned(seqid);
3497 *tl++ = txdr_unsigned(slotid);
3498 *tl++ = txdr_unsigned(NFSV4_CBSLOTS - 1);
3499 *tl = txdr_unsigned(NFSV4_CBSLOTS - 1);
3503 if (i == 0 && minorvers == NFSV41_MINORVERSION)
3504 error = NFSERR_OPNOTINSESS;
3506 NFSCL_DEBUG(1, "unsupp callback %d\n", op);
3507 error = NFSERR_NOTSUPP;
3512 if (error == EBADRPC || error == NFSERR_BADXDR) {
3513 nd->nd_repstat = NFSERR_BADXDR;
3515 nd->nd_repstat = error;
3520 if (nd->nd_repstat) {
3521 *repp = nfscl_errmap(nd, minorvers);
3524 *repp = 0; /* NFS4_OK */
3527 if (recallp != NULL)
3528 free(recallp, M_NFSLAYRECALL);
3530 if (error == EBADRPC || error == NFSERR_BADXDR)
3531 nd->nd_repstat = NFSERR_BADXDR;
3533 printf("nfsv4 comperr1=%d\n", error);
3536 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
3540 *retopsp = txdr_unsigned(retops);
3542 *nd->nd_errp = nfscl_errmap(nd, minorvers);
3544 if (gotseq_ok != 0) {
3545 rep = m_copym(nd->nd_mreq, 0, M_COPYALL, M_WAITOK);
3547 clp = nfscl_getclntsess(sessionid);
3549 nfsv4_seqsess_cacherep(slotid,
3550 NFSMNT_MDSSESSION(clp->nfsc_nmp)->nfsess_cbslots,
3561 * Generate the next cbident value. Basically just increment a static value
3562 * and then check that it isn't already in the list, if it has wrapped around.
3565 nfscl_nextcbident(void)
3567 struct nfsclclient *clp;
3569 static u_int32_t nextcbident = 0;
3570 static int haswrapped = 0;
3573 if (nextcbident == 0)
3577 * Search the clientid list for one already using this cbident.
3582 LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3583 if (clp->nfsc_cbident == nextcbident) {
3593 return (nextcbident);
3597 * Get the mount point related to a given cbident or session and busy it.
3600 nfscl_getmnt(int minorvers, uint8_t *sessionid, u_int32_t cbident,
3601 struct nfsclclient **clpp)
3603 struct nfsclclient *clp;
3609 LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3610 if (minorvers == NFSV4_MINORVERSION) {
3611 if (clp->nfsc_cbident == cbident)
3613 } else if (!NFSBCMP(NFSMNT_MDSSESSION(clp->nfsc_nmp)->
3614 nfsess_sessionid, sessionid, NFSX_V4SESSIONID))
3621 mp = clp->nfsc_nmp->nm_mountp;
3624 error = vfs_busy(mp, 0);
3633 * Get the clientid pointer related to a given cbident.
3635 static struct nfsclclient *
3636 nfscl_getclnt(u_int32_t cbident)
3638 struct nfsclclient *clp;
3640 LIST_FOREACH(clp, &nfsclhead, nfsc_list)
3641 if (clp->nfsc_cbident == cbident)
3647 * Get the clientid pointer related to a given sessionid.
3649 static struct nfsclclient *
3650 nfscl_getclntsess(uint8_t *sessionid)
3652 struct nfsclclient *clp;
3654 LIST_FOREACH(clp, &nfsclhead, nfsc_list)
3655 if (!NFSBCMP(NFSMNT_MDSSESSION(clp->nfsc_nmp)->nfsess_sessionid,
3656 sessionid, NFSX_V4SESSIONID))
3662 * Search for a lock conflict locally on the client. A conflict occurs if
3663 * - not same owner and overlapping byte range and at least one of them is
3664 * a write lock or this is an unlock.
3667 nfscl_localconflict(struct nfsclclient *clp, u_int8_t *fhp, int fhlen,
3668 struct nfscllock *nlop, u_int8_t *own, struct nfscldeleg *dp,
3669 struct nfscllock **lopp)
3671 struct nfsclowner *owp;
3672 struct nfsclopen *op;
3676 ret = nfscl_checkconflict(&dp->nfsdl_lock, nlop, own, lopp);
3680 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3681 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3682 if (op->nfso_fhlen == fhlen &&
3683 !NFSBCMP(op->nfso_fh, fhp, fhlen)) {
3684 ret = nfscl_checkconflict(&op->nfso_lock, nlop,
3695 nfscl_checkconflict(struct nfscllockownerhead *lhp, struct nfscllock *nlop,
3696 u_int8_t *own, struct nfscllock **lopp)
3698 struct nfscllockowner *lp;
3699 struct nfscllock *lop;
3701 LIST_FOREACH(lp, lhp, nfsl_list) {
3702 if (NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
3703 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
3704 if (lop->nfslo_first >= nlop->nfslo_end)
3706 if (lop->nfslo_end <= nlop->nfslo_first)
3708 if (lop->nfslo_type == F_WRLCK ||
3709 nlop->nfslo_type == F_WRLCK ||
3710 nlop->nfslo_type == F_UNLCK) {
3713 return (NFSERR_DENIED);
3722 * Check for a local conflicting lock.
3725 nfscl_lockt(vnode_t vp, struct nfsclclient *clp, u_int64_t off,
3726 u_int64_t len, struct flock *fl, NFSPROC_T *p, void *id, int flags)
3728 struct nfscllock *lop, nlck;
3729 struct nfscldeleg *dp;
3731 u_int8_t own[NFSV4CL_LOCKNAMELEN];
3734 nlck.nfslo_type = fl->l_type;
3735 nlck.nfslo_first = off;
3736 if (len == NFS64BITSSET) {
3737 nlck.nfslo_end = NFS64BITSSET;
3739 nlck.nfslo_end = off + len;
3740 if (nlck.nfslo_end <= nlck.nfslo_first)
3741 return (NFSERR_INVAL);
3744 nfscl_filllockowner(id, own, flags);
3746 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
3747 error = nfscl_localconflict(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len,
3748 &nlck, own, dp, &lop);
3750 fl->l_whence = SEEK_SET;
3751 fl->l_start = lop->nfslo_first;
3752 if (lop->nfslo_end == NFS64BITSSET)
3755 fl->l_len = lop->nfslo_end - lop->nfslo_first;
3756 fl->l_pid = (pid_t)0;
3757 fl->l_type = lop->nfslo_type;
3758 error = -1; /* no RPC required */
3759 } else if (dp != NULL && ((dp->nfsdl_flags & NFSCLDL_WRITE) ||
3760 fl->l_type == F_RDLCK)) {
3762 * The delegation ensures that there isn't a conflicting
3763 * lock on the server, so return -1 to indicate an RPC
3766 fl->l_type = F_UNLCK;
3774 * Handle Recall of a delegation.
3775 * The clp must be exclusive locked when this is called.
3778 nfscl_recalldeleg(struct nfsclclient *clp, struct nfsmount *nmp,
3779 struct nfscldeleg *dp, vnode_t vp, struct ucred *cred, NFSPROC_T *p,
3780 int called_from_renewthread)
3782 struct nfsclowner *owp, *lowp, *nowp;
3783 struct nfsclopen *op, *lop;
3784 struct nfscllockowner *lp;
3785 struct nfscllock *lckp;
3787 int error = 0, ret, gotvp = 0;
3791 * First, get a vnode for the file. This is needed to do RPCs.
3793 ret = nfscl_ngetreopen(nmp->nm_mountp, dp->nfsdl_fh,
3794 dp->nfsdl_fhlen, p, &np);
3797 * File isn't open, so nothing to move over to the
3807 dp->nfsdl_flags &= ~NFSCLDL_MODTIMESET;
3810 * Ok, if it's a write delegation, flush data to the server, so
3811 * that close/open consistency is retained.
3815 if ((dp->nfsdl_flags & NFSCLDL_WRITE) && (np->n_flag & NMODIFIED)) {
3816 np->n_flag |= NDELEGRECALL;
3818 ret = ncl_flush(vp, MNT_WAIT, cred, p, 1,
3819 called_from_renewthread);
3821 np->n_flag &= ~NDELEGRECALL;
3823 NFSINVALATTRCACHE(np);
3825 if (ret == EIO && called_from_renewthread != 0) {
3827 * If the flush failed with EIO for the renew thread,
3828 * return now, so that the dirty buffer will be flushed
3837 * Now, for each openowner with opens issued locally, move them
3838 * over to state against the server.
3840 LIST_FOREACH(lowp, &dp->nfsdl_owner, nfsow_list) {
3841 lop = LIST_FIRST(&lowp->nfsow_open);
3843 if (LIST_NEXT(lop, nfso_list) != NULL)
3844 panic("nfsdlg mult opens");
3846 * Look for the same openowner against the server.
3848 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3849 if (!NFSBCMP(lowp->nfsow_owner,
3850 owp->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
3851 newnfs_copycred(&dp->nfsdl_cred, cred);
3852 ret = nfscl_moveopen(vp, clp, nmp, lop,
3854 if (ret == NFSERR_STALECLIENTID ||
3855 ret == NFSERR_STALEDONTRECOVER ||
3856 ret == NFSERR_BADSESSION) {
3862 nfscl_freeopen(lop, 1);
3871 * If no openowner found, create one and get an open
3875 MALLOC(nowp, struct nfsclowner *,
3876 sizeof (struct nfsclowner), M_NFSCLOWNER,
3878 nfscl_newopen(clp, NULL, &owp, &nowp, &op,
3879 NULL, lowp->nfsow_owner, dp->nfsdl_fh,
3880 dp->nfsdl_fhlen, NULL);
3881 newnfs_copycred(&dp->nfsdl_cred, cred);
3882 ret = nfscl_moveopen(vp, clp, nmp, lop,
3885 nfscl_freeopenowner(owp, 0);
3886 if (ret == NFSERR_STALECLIENTID ||
3887 ret == NFSERR_STALEDONTRECOVER ||
3888 ret == NFSERR_BADSESSION) {
3894 nfscl_freeopen(lop, 1);
3904 * Now, get byte range locks for any locks done locally.
3906 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
3907 LIST_FOREACH(lckp, &lp->nfsl_lock, nfslo_list) {
3908 newnfs_copycred(&dp->nfsdl_cred, cred);
3909 ret = nfscl_relock(vp, clp, nmp, lp, lckp, cred, p);
3910 if (ret == NFSERR_STALESTATEID ||
3911 ret == NFSERR_STALEDONTRECOVER ||
3912 ret == NFSERR_STALECLIENTID ||
3913 ret == NFSERR_BADSESSION) {
3928 * Move a locally issued open over to an owner on the state list.
3929 * SIDE EFFECT: If it needs to sleep (do an rpc), it unlocks clstate and
3930 * returns with it unlocked.
3933 nfscl_moveopen(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp,
3934 struct nfsclopen *lop, struct nfsclowner *owp, struct nfscldeleg *dp,
3935 struct ucred *cred, NFSPROC_T *p)
3937 struct nfsclopen *op, *nop;
3938 struct nfscldeleg *ndp;
3940 int error = 0, newone;
3943 * First, look for an appropriate open, If found, just increment the
3946 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3947 if ((op->nfso_mode & lop->nfso_mode) == lop->nfso_mode &&
3948 op->nfso_fhlen == lop->nfso_fhlen &&
3949 !NFSBCMP(op->nfso_fh, lop->nfso_fh, op->nfso_fhlen)) {
3950 op->nfso_opencnt += lop->nfso_opencnt;
3951 nfscl_freeopen(lop, 1);
3956 /* No appropriate open, so we have to do one against the server. */
3958 MALLOC(nop, struct nfsclopen *, sizeof (struct nfsclopen) +
3959 lop->nfso_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
3961 nfscl_newopen(clp, NULL, &owp, NULL, &op, &nop, owp->nfsow_owner,
3962 lop->nfso_fh, lop->nfso_fhlen, &newone);
3964 error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data, np->n_v4->n4_fhlen,
3965 lop->nfso_fh, lop->nfso_fhlen, lop->nfso_mode, op,
3966 NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, &ndp, 0, 0, cred, p);
3969 nfscl_freeopen(op, 0);
3972 newnfs_copyincred(cred, &op->nfso_cred);
3973 op->nfso_mode |= lop->nfso_mode;
3974 op->nfso_opencnt += lop->nfso_opencnt;
3975 nfscl_freeopen(lop, 1);
3978 FREE((caddr_t)nop, M_NFSCLOPEN);
3981 * What should I do with the returned delegation, since the
3982 * delegation is being recalled? For now, just printf and
3985 printf("Moveopen returned deleg\n");
3986 FREE((caddr_t)ndp, M_NFSCLDELEG);
3992 * Recall all delegations on this client.
3995 nfscl_totalrecall(struct nfsclclient *clp)
3997 struct nfscldeleg *dp;
3999 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
4000 if ((dp->nfsdl_flags & NFSCLDL_DELEGRET) == 0)
4001 dp->nfsdl_flags |= NFSCLDL_RECALL;
4006 * Relock byte ranges. Called for delegation recall and state expiry.
4009 nfscl_relock(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp,
4010 struct nfscllockowner *lp, struct nfscllock *lop, struct ucred *cred,
4013 struct nfscllockowner *nlp;
4016 u_int32_t clidrev = 0;
4017 int error, newone, donelocally;
4019 off = lop->nfslo_first;
4020 len = lop->nfslo_end - lop->nfslo_first;
4021 error = nfscl_getbytelock(vp, off, len, lop->nfslo_type, cred, p,
4022 clp, 1, NULL, lp->nfsl_lockflags, lp->nfsl_owner,
4023 lp->nfsl_openowner, &nlp, &newone, &donelocally);
4024 if (error || donelocally)
4026 if (nmp->nm_clp != NULL)
4027 clidrev = nmp->nm_clp->nfsc_clientidrev;
4030 nfhp = VTONFS(vp)->n_fhp;
4031 error = nfscl_trylock(nmp, vp, nfhp->nfh_fh,
4032 nfhp->nfh_len, nlp, newone, 0, off,
4033 len, lop->nfslo_type, cred, p);
4035 nfscl_freelockowner(nlp, 0);
4040 * Called to re-open a file. Basically get a vnode for the file handle
4041 * and then call nfsrpc_openrpc() to do the rest.
4044 nfsrpc_reopen(struct nfsmount *nmp, u_int8_t *fhp, int fhlen,
4045 u_int32_t mode, struct nfsclopen *op, struct nfscldeleg **dpp,
4046 struct ucred *cred, NFSPROC_T *p)
4052 error = nfscl_ngetreopen(nmp->nm_mountp, fhp, fhlen, p, &np);
4056 if (np->n_v4 != NULL) {
4057 error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data,
4058 np->n_v4->n4_fhlen, fhp, fhlen, mode, op,
4059 NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, dpp, 0, 0,
4069 * Try an open against the server. Just call nfsrpc_openrpc(), retrying while
4070 * NFSERR_DELAY. Also, try system credentials, if the passed in credentials
4074 nfscl_tryopen(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen,
4075 u_int8_t *newfhp, int newfhlen, u_int32_t mode, struct nfsclopen *op,
4076 u_int8_t *name, int namelen, struct nfscldeleg **ndpp,
4077 int reclaim, u_int32_t delegtype, struct ucred *cred, NFSPROC_T *p)
4082 error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp, newfhlen,
4083 mode, op, name, namelen, ndpp, reclaim, delegtype, cred, p,
4085 if (error == NFSERR_DELAY)
4086 (void) nfs_catnap(PZERO, error, "nfstryop");
4087 } while (error == NFSERR_DELAY);
4088 if (error == EAUTH || error == EACCES) {
4089 /* Try again using system credentials */
4090 newnfs_setroot(cred);
4092 error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp,
4093 newfhlen, mode, op, name, namelen, ndpp, reclaim,
4094 delegtype, cred, p, 1, 0);
4095 if (error == NFSERR_DELAY)
4096 (void) nfs_catnap(PZERO, error, "nfstryop");
4097 } while (error == NFSERR_DELAY);
4103 * Try a byte range lock. Just loop on nfsrpc_lock() while it returns
4104 * NFSERR_DELAY. Also, retry with system credentials, if the provided
4108 nfscl_trylock(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp,
4109 int fhlen, struct nfscllockowner *nlp, int newone, int reclaim,
4110 u_int64_t off, u_int64_t len, short type, struct ucred *cred, NFSPROC_T *p)
4112 struct nfsrv_descript nfsd, *nd = &nfsd;
4116 error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp, newone,
4117 reclaim, off, len, type, cred, p, 0);
4118 if (!error && nd->nd_repstat == NFSERR_DELAY)
4119 (void) nfs_catnap(PZERO, (int)nd->nd_repstat,
4121 } while (!error && nd->nd_repstat == NFSERR_DELAY);
4123 error = nd->nd_repstat;
4124 if (error == EAUTH || error == EACCES) {
4125 /* Try again using root credentials */
4126 newnfs_setroot(cred);
4128 error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp,
4129 newone, reclaim, off, len, type, cred, p, 1);
4130 if (!error && nd->nd_repstat == NFSERR_DELAY)
4131 (void) nfs_catnap(PZERO, (int)nd->nd_repstat,
4133 } while (!error && nd->nd_repstat == NFSERR_DELAY);
4135 error = nd->nd_repstat;
4141 * Try a delegreturn against the server. Just call nfsrpc_delegreturn(),
4142 * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
4146 nfscl_trydelegreturn(struct nfscldeleg *dp, struct ucred *cred,
4147 struct nfsmount *nmp, NFSPROC_T *p)
4152 error = nfsrpc_delegreturn(dp, cred, nmp, p, 0);
4153 if (error == NFSERR_DELAY)
4154 (void) nfs_catnap(PZERO, error, "nfstrydp");
4155 } while (error == NFSERR_DELAY);
4156 if (error == EAUTH || error == EACCES) {
4157 /* Try again using system credentials */
4158 newnfs_setroot(cred);
4160 error = nfsrpc_delegreturn(dp, cred, nmp, p, 1);
4161 if (error == NFSERR_DELAY)
4162 (void) nfs_catnap(PZERO, error, "nfstrydp");
4163 } while (error == NFSERR_DELAY);
4169 * Try a close against the server. Just call nfsrpc_closerpc(),
4170 * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
4174 nfscl_tryclose(struct nfsclopen *op, struct ucred *cred,
4175 struct nfsmount *nmp, NFSPROC_T *p)
4177 struct nfsrv_descript nfsd, *nd = &nfsd;
4181 error = nfsrpc_closerpc(nd, nmp, op, cred, p, 0);
4182 if (error == NFSERR_DELAY)
4183 (void) nfs_catnap(PZERO, error, "nfstrycl");
4184 } while (error == NFSERR_DELAY);
4185 if (error == EAUTH || error == EACCES) {
4186 /* Try again using system credentials */
4187 newnfs_setroot(cred);
4189 error = nfsrpc_closerpc(nd, nmp, op, cred, p, 1);
4190 if (error == NFSERR_DELAY)
4191 (void) nfs_catnap(PZERO, error, "nfstrycl");
4192 } while (error == NFSERR_DELAY);
4198 * Decide if a delegation on a file permits close without flushing writes
4199 * to the server. This might be a big performance win in some environments.
4200 * (Not useful until the client does caching on local stable storage.)
4203 nfscl_mustflush(vnode_t vp)
4205 struct nfsclclient *clp;
4206 struct nfscldeleg *dp;
4208 struct nfsmount *nmp;
4211 nmp = VFSTONFS(vnode_mount(vp));
4212 if (!NFSHASNFSV4(nmp))
4215 clp = nfscl_findcl(nmp);
4220 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4221 if (dp != NULL && (dp->nfsdl_flags &
4222 (NFSCLDL_WRITE | NFSCLDL_RECALL | NFSCLDL_DELEGRET)) ==
4224 (dp->nfsdl_sizelimit >= np->n_size ||
4225 !NFSHASSTRICT3530(nmp))) {
4234 * See if a (write) delegation exists for this file.
4237 nfscl_nodeleg(vnode_t vp, int writedeleg)
4239 struct nfsclclient *clp;
4240 struct nfscldeleg *dp;
4242 struct nfsmount *nmp;
4245 nmp = VFSTONFS(vnode_mount(vp));
4246 if (!NFSHASNFSV4(nmp))
4249 clp = nfscl_findcl(nmp);
4254 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4256 (dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_DELEGRET)) == 0 &&
4257 (writedeleg == 0 || (dp->nfsdl_flags & NFSCLDL_WRITE) ==
4267 * Look for an associated delegation that should be DelegReturned.
4270 nfscl_removedeleg(vnode_t vp, NFSPROC_T *p, nfsv4stateid_t *stp)
4272 struct nfsclclient *clp;
4273 struct nfscldeleg *dp;
4274 struct nfsclowner *owp;
4275 struct nfscllockowner *lp;
4276 struct nfsmount *nmp;
4279 int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
4281 nmp = VFSTONFS(vnode_mount(vp));
4285 * Loop around waiting for:
4286 * - outstanding I/O operations on delegations to complete
4287 * - for a delegation on vp that has state, lock the client and
4289 * - return delegation with no state
4292 clp = nfscl_findcl(nmp);
4297 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4298 np->n_fhp->nfh_len);
4301 * Wait for outstanding I/O ops to be done.
4303 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4305 nfsv4_unlock(&clp->nfsc_lock, 0);
4308 dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4309 (void) nfsmsleep(&dp->nfsdl_rwlock,
4310 NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4314 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4315 if (!LIST_EMPTY(&owp->nfsow_open)) {
4321 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4322 if (!LIST_EMPTY(&lp->nfsl_lock)) {
4328 if (needsrecall && !triedrecall) {
4329 dp->nfsdl_flags |= NFSCLDL_DELEGRET;
4332 igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
4333 &islept, NFSCLSTATEMUTEXPTR, NULL);
4340 cred = newnfs_getcred();
4341 newnfs_copycred(&dp->nfsdl_cred, cred);
4342 (void) nfscl_recalldeleg(clp, nmp, dp, vp, cred, p, 0);
4346 nfsv4_unlock(&clp->nfsc_lock, 0);
4350 *stp = dp->nfsdl_stateid;
4352 nfscl_cleandeleg(dp);
4353 nfscl_freedeleg(&clp->nfsc_deleg, dp);
4356 nfsv4_unlock(&clp->nfsc_lock, 0);
4363 * Look for associated delegation(s) that should be DelegReturned.
4366 nfscl_renamedeleg(vnode_t fvp, nfsv4stateid_t *fstp, int *gotfdp, vnode_t tvp,
4367 nfsv4stateid_t *tstp, int *gottdp, NFSPROC_T *p)
4369 struct nfsclclient *clp;
4370 struct nfscldeleg *dp;
4371 struct nfsclowner *owp;
4372 struct nfscllockowner *lp;
4373 struct nfsmount *nmp;
4376 int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
4378 nmp = VFSTONFS(vnode_mount(fvp));
4383 * Loop around waiting for:
4384 * - outstanding I/O operations on delegations to complete
4385 * - for a delegation on fvp that has state, lock the client and
4387 * - return delegation(s) with no state.
4390 clp = nfscl_findcl(nmp);
4396 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4397 np->n_fhp->nfh_len);
4398 if (dp != NULL && *gotfdp == 0) {
4400 * Wait for outstanding I/O ops to be done.
4402 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4404 nfsv4_unlock(&clp->nfsc_lock, 0);
4407 dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4408 (void) nfsmsleep(&dp->nfsdl_rwlock,
4409 NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4413 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4414 if (!LIST_EMPTY(&owp->nfsow_open)) {
4420 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4421 if (!LIST_EMPTY(&lp->nfsl_lock)) {
4427 if (needsrecall && !triedrecall) {
4428 dp->nfsdl_flags |= NFSCLDL_DELEGRET;
4431 igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
4432 &islept, NFSCLSTATEMUTEXPTR, NULL);
4439 cred = newnfs_getcred();
4440 newnfs_copycred(&dp->nfsdl_cred, cred);
4441 (void) nfscl_recalldeleg(clp, nmp, dp, fvp, cred, p, 0);
4445 nfsv4_unlock(&clp->nfsc_lock, 0);
4449 *fstp = dp->nfsdl_stateid;
4452 nfscl_cleandeleg(dp);
4453 nfscl_freedeleg(&clp->nfsc_deleg, dp);
4456 nfsv4_unlock(&clp->nfsc_lock, 0);
4461 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4462 np->n_fhp->nfh_len);
4463 if (dp != NULL && *gottdp == 0) {
4465 * Wait for outstanding I/O ops to be done.
4467 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4468 dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4469 (void) nfsmsleep(&dp->nfsdl_rwlock,
4470 NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4473 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4474 if (!LIST_EMPTY(&owp->nfsow_open)) {
4479 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4480 if (!LIST_EMPTY(&lp->nfsl_lock)) {
4485 *tstp = dp->nfsdl_stateid;
4488 nfscl_cleandeleg(dp);
4489 nfscl_freedeleg(&clp->nfsc_deleg, dp);
4498 * Get a reference on the clientid associated with the mount point.
4499 * Return 1 if success, 0 otherwise.
4502 nfscl_getref(struct nfsmount *nmp)
4504 struct nfsclclient *clp;
4507 clp = nfscl_findcl(nmp);
4512 nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, NULL);
4518 * Release a reference on a clientid acquired with the above call.
4521 nfscl_relref(struct nfsmount *nmp)
4523 struct nfsclclient *clp;
4526 clp = nfscl_findcl(nmp);
4531 nfsv4_relref(&clp->nfsc_lock);
4536 * Save the size attribute in the delegation, since the nfsnode
4540 nfscl_reclaimnode(vnode_t vp)
4542 struct nfsclclient *clp;
4543 struct nfscldeleg *dp;
4544 struct nfsnode *np = VTONFS(vp);
4545 struct nfsmount *nmp;
4547 nmp = VFSTONFS(vnode_mount(vp));
4548 if (!NFSHASNFSV4(nmp))
4551 clp = nfscl_findcl(nmp);
4556 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4557 if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE))
4558 dp->nfsdl_size = np->n_size;
4563 * Get the saved size attribute in the delegation, since it is a
4564 * newly allocated nfsnode.
4567 nfscl_newnode(vnode_t vp)
4569 struct nfsclclient *clp;
4570 struct nfscldeleg *dp;
4571 struct nfsnode *np = VTONFS(vp);
4572 struct nfsmount *nmp;
4574 nmp = VFSTONFS(vnode_mount(vp));
4575 if (!NFSHASNFSV4(nmp))
4578 clp = nfscl_findcl(nmp);
4583 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4584 if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE))
4585 np->n_size = dp->nfsdl_size;
4590 * If there is a valid write delegation for this file, set the modtime
4591 * to the local clock time.
4594 nfscl_delegmodtime(vnode_t vp)
4596 struct nfsclclient *clp;
4597 struct nfscldeleg *dp;
4598 struct nfsnode *np = VTONFS(vp);
4599 struct nfsmount *nmp;
4601 nmp = VFSTONFS(vnode_mount(vp));
4602 if (!NFSHASNFSV4(nmp))
4605 clp = nfscl_findcl(nmp);
4610 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4611 if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE)) {
4612 nanotime(&dp->nfsdl_modtime);
4613 dp->nfsdl_flags |= NFSCLDL_MODTIMESET;
4619 * If there is a valid write delegation for this file with a modtime set,
4620 * put that modtime in mtime.
4623 nfscl_deleggetmodtime(vnode_t vp, struct timespec *mtime)
4625 struct nfsclclient *clp;
4626 struct nfscldeleg *dp;
4627 struct nfsnode *np = VTONFS(vp);
4628 struct nfsmount *nmp;
4630 nmp = VFSTONFS(vnode_mount(vp));
4631 if (!NFSHASNFSV4(nmp))
4634 clp = nfscl_findcl(nmp);
4639 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4641 (dp->nfsdl_flags & (NFSCLDL_WRITE | NFSCLDL_MODTIMESET)) ==
4642 (NFSCLDL_WRITE | NFSCLDL_MODTIMESET))
4643 *mtime = dp->nfsdl_modtime;
4648 nfscl_errmap(struct nfsrv_descript *nd, u_int32_t minorvers)
4650 short *defaulterrp, *errp;
4652 if (!nd->nd_repstat)
4654 if (nd->nd_procnum == NFSPROC_NOOP)
4655 return (txdr_unsigned(nd->nd_repstat & 0xffff));
4656 if (nd->nd_repstat == EBADRPC)
4657 return (txdr_unsigned(NFSERR_BADXDR));
4658 if (nd->nd_repstat == NFSERR_MINORVERMISMATCH ||
4659 nd->nd_repstat == NFSERR_OPILLEGAL)
4660 return (txdr_unsigned(nd->nd_repstat));
4661 if (nd->nd_repstat >= NFSERR_BADIOMODE && nd->nd_repstat < 20000 &&
4662 minorvers > NFSV4_MINORVERSION) {
4663 /* NFSv4.n error. */
4664 return (txdr_unsigned(nd->nd_repstat));
4666 if (nd->nd_procnum < NFSV4OP_CBNOPS)
4667 errp = defaulterrp = nfscl_cberrmap[nd->nd_procnum];
4669 return (txdr_unsigned(nd->nd_repstat));
4671 if (*errp == (short)nd->nd_repstat)
4672 return (txdr_unsigned(nd->nd_repstat));
4673 return (txdr_unsigned(*defaulterrp));
4677 * Called to find/add a layout to a client.
4678 * This function returns the layout with a refcnt (shared lock) upon
4679 * success (returns 0) or with no lock/refcnt on the layout when an
4680 * error is returned.
4681 * If a layout is passed in via lypp, it is locked (exclusively locked).
4684 nfscl_layout(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen,
4685 nfsv4stateid_t *stateidp, int retonclose,
4686 struct nfsclflayouthead *fhlp, struct nfscllayout **lypp,
4687 struct ucred *cred, NFSPROC_T *p)
4689 struct nfsclclient *clp;
4690 struct nfscllayout *lyp, *tlyp;
4691 struct nfsclflayout *flp;
4692 struct nfsnode *np = VTONFS(vp);
4694 int layout_passed_in;
4696 mp = nmp->nm_mountp;
4697 layout_passed_in = 1;
4701 layout_passed_in = 0;
4702 tlyp = malloc(sizeof(*tlyp) + fhlen - 1, M_NFSLAYOUT,
4709 if (layout_passed_in != 0)
4710 nfsv4_unlock(&lyp->nfsly_lock, 0);
4713 free(tlyp, M_NFSLAYOUT);
4718 * Although no lyp was passed in, another thread might have
4719 * allocated one. If one is found, just increment it's ref
4720 * count and return it.
4722 lyp = nfscl_findlayout(clp, fhp, fhlen);
4726 lyp->nfsly_stateid.seqid = stateidp->seqid;
4727 lyp->nfsly_stateid.other[0] = stateidp->other[0];
4728 lyp->nfsly_stateid.other[1] = stateidp->other[1];
4729 lyp->nfsly_stateid.other[2] = stateidp->other[2];
4730 lyp->nfsly_lastbyte = 0;
4731 LIST_INIT(&lyp->nfsly_flayread);
4732 LIST_INIT(&lyp->nfsly_flayrw);
4733 LIST_INIT(&lyp->nfsly_recall);
4734 lyp->nfsly_filesid[0] = np->n_vattr.na_filesid[0];
4735 lyp->nfsly_filesid[1] = np->n_vattr.na_filesid[1];
4736 lyp->nfsly_clp = clp;
4737 lyp->nfsly_flags = (retonclose != 0) ?
4738 (NFSLY_FILES | NFSLY_RETONCLOSE) : NFSLY_FILES;
4739 lyp->nfsly_fhlen = fhlen;
4740 NFSBCOPY(fhp, lyp->nfsly_fh, fhlen);
4741 TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
4742 LIST_INSERT_HEAD(NFSCLLAYOUTHASH(clp, fhp, fhlen), lyp,
4744 lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
4747 if (retonclose != 0)
4748 lyp->nfsly_flags |= NFSLY_RETONCLOSE;
4749 TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list);
4750 TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
4751 lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
4753 nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
4754 if ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) {
4757 free(tlyp, M_NFSLAYOUT);
4762 lyp->nfsly_stateid.seqid = stateidp->seqid;
4764 /* Merge the new list of File Layouts into the list. */
4765 flp = LIST_FIRST(fhlp);
4767 if (flp->nfsfl_iomode == NFSLAYOUTIOMODE_READ)
4768 nfscl_mergeflayouts(&lyp->nfsly_flayread, fhlp);
4770 nfscl_mergeflayouts(&lyp->nfsly_flayrw, fhlp);
4772 if (layout_passed_in != 0)
4773 nfsv4_unlock(&lyp->nfsly_lock, 1);
4776 free(tlyp, M_NFSLAYOUT);
4781 * Search for a layout by MDS file handle.
4782 * If one is found, it is returned with a refcnt (shared lock) iff
4783 * retflpp returned non-NULL and locked (exclusive locked) iff retflpp is
4786 struct nfscllayout *
4787 nfscl_getlayout(struct nfsclclient *clp, uint8_t *fhp, int fhlen,
4788 uint64_t off, struct nfsclflayout **retflpp, int *recalledp)
4790 struct nfscllayout *lyp;
4792 int error, igotlock;
4794 mp = clp->nfsc_nmp->nm_mountp;
4798 lyp = nfscl_findlayout(clp, fhp, fhlen);
4800 if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
4801 TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list);
4802 TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
4803 lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
4804 error = nfscl_findlayoutforio(lyp, off,
4805 NFSV4OPEN_ACCESSREAD, retflpp);
4807 nfsv4_getref(&lyp->nfsly_lock, NULL,
4808 NFSCLSTATEMUTEXPTR, mp);
4811 igotlock = nfsv4_lock(&lyp->nfsly_lock,
4812 1, NULL, NFSCLSTATEMUTEXPTR, mp);
4813 } while (igotlock == 0 &&
4814 (mp->mnt_kern_flag & MNTK_UNMOUNTF) == 0);
4817 if ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) {
4831 * Search for a layout by MDS file handle. If one is found that is marked
4832 * "return on close", delete it, since it should now be forgotten.
4835 nfscl_retoncloselayout(struct nfsclclient *clp, uint8_t *fhp, int fhlen)
4837 struct nfscllayout *lyp;
4840 lyp = nfscl_findlayout(clp, fhp, fhlen);
4841 if (lyp != NULL && (lyp->nfsly_flags & NFSLY_RETONCLOSE) != 0) {
4843 * Wait for outstanding I/O ops to be done.
4845 if (lyp->nfsly_lock.nfslock_usecnt != 0 ||
4846 lyp->nfsly_lock.nfslock_lock != 0) {
4847 lyp->nfsly_lock.nfslock_lock |= NFSV4LOCK_WANTED;
4848 (void)mtx_sleep(&lyp->nfsly_lock,
4849 NFSCLSTATEMUTEXPTR, PZERO, "nfslyc", 0);
4852 nfscl_freelayout(lyp);
4857 * Dereference a layout.
4860 nfscl_rellayout(struct nfscllayout *lyp, int exclocked)
4865 nfsv4_unlock(&lyp->nfsly_lock, 0);
4867 nfsv4_relref(&lyp->nfsly_lock);
4872 * Search for a devinfo by deviceid. If one is found, return it after
4873 * acquiring a reference count on it.
4875 struct nfscldevinfo *
4876 nfscl_getdevinfo(struct nfsclclient *clp, uint8_t *deviceid,
4877 struct nfscldevinfo *dip)
4882 dip = nfscl_finddevinfo(clp, deviceid);
4884 dip->nfsdi_refcnt++;
4890 * Dereference a devinfo structure.
4893 nfscl_reldevinfo_locked(struct nfscldevinfo *dip)
4896 dip->nfsdi_refcnt--;
4897 if (dip->nfsdi_refcnt == 0)
4898 wakeup(&dip->nfsdi_refcnt);
4902 * Dereference a devinfo structure.
4905 nfscl_reldevinfo(struct nfscldevinfo *dip)
4909 nfscl_reldevinfo_locked(dip);
4914 * Find a layout for this file handle. Return NULL upon failure.
4916 static struct nfscllayout *
4917 nfscl_findlayout(struct nfsclclient *clp, u_int8_t *fhp, int fhlen)
4919 struct nfscllayout *lyp;
4921 LIST_FOREACH(lyp, NFSCLLAYOUTHASH(clp, fhp, fhlen), nfsly_hash)
4922 if (lyp->nfsly_fhlen == fhlen &&
4923 !NFSBCMP(lyp->nfsly_fh, fhp, fhlen))
4929 * Find a devinfo for this deviceid. Return NULL upon failure.
4931 static struct nfscldevinfo *
4932 nfscl_finddevinfo(struct nfsclclient *clp, uint8_t *deviceid)
4934 struct nfscldevinfo *dip;
4936 LIST_FOREACH(dip, &clp->nfsc_devinfo, nfsdi_list)
4937 if (NFSBCMP(dip->nfsdi_deviceid, deviceid, NFSX_V4DEVICEID)
4944 * Merge the new file layout list into the main one, maintaining it in
4945 * increasing offset order.
4948 nfscl_mergeflayouts(struct nfsclflayouthead *fhlp,
4949 struct nfsclflayouthead *newfhlp)
4951 struct nfsclflayout *flp, *nflp, *prevflp, *tflp;
4953 flp = LIST_FIRST(fhlp);
4955 LIST_FOREACH_SAFE(nflp, newfhlp, nfsfl_list, tflp) {
4956 while (flp != NULL && flp->nfsfl_off < nflp->nfsfl_off) {
4958 flp = LIST_NEXT(flp, nfsfl_list);
4960 if (prevflp == NULL)
4961 LIST_INSERT_HEAD(fhlp, nflp, nfsfl_list);
4963 LIST_INSERT_AFTER(prevflp, nflp, nfsfl_list);
4969 * Add this nfscldevinfo to the client, if it doesn't already exist.
4970 * This function consumes the structure pointed at by dip, if not NULL.
4973 nfscl_adddevinfo(struct nfsmount *nmp, struct nfscldevinfo *dip,
4974 struct nfsclflayout *flp)
4976 struct nfsclclient *clp;
4977 struct nfscldevinfo *tdip;
4984 free(dip, M_NFSDEVINFO);
4987 tdip = nfscl_finddevinfo(clp, flp->nfsfl_dev);
4989 tdip->nfsdi_layoutrefs++;
4990 flp->nfsfl_devp = tdip;
4991 nfscl_reldevinfo_locked(tdip);
4994 free(dip, M_NFSDEVINFO);
4998 LIST_INSERT_HEAD(&clp->nfsc_devinfo, dip, nfsdi_list);
4999 dip->nfsdi_layoutrefs = 1;
5000 flp->nfsfl_devp = dip;
5009 * Free up a layout structure and associated file layout structure(s).
5012 nfscl_freelayout(struct nfscllayout *layp)
5014 struct nfsclflayout *flp, *nflp;
5015 struct nfsclrecalllayout *rp, *nrp;
5017 LIST_FOREACH_SAFE(flp, &layp->nfsly_flayread, nfsfl_list, nflp) {
5018 LIST_REMOVE(flp, nfsfl_list);
5019 nfscl_freeflayout(flp);
5021 LIST_FOREACH_SAFE(flp, &layp->nfsly_flayrw, nfsfl_list, nflp) {
5022 LIST_REMOVE(flp, nfsfl_list);
5023 nfscl_freeflayout(flp);
5025 LIST_FOREACH_SAFE(rp, &layp->nfsly_recall, nfsrecly_list, nrp) {
5026 LIST_REMOVE(rp, nfsrecly_list);
5027 free(rp, M_NFSLAYRECALL);
5030 free(layp, M_NFSLAYOUT);
5034 * Free up a file layout structure.
5037 nfscl_freeflayout(struct nfsclflayout *flp)
5041 for (i = 0; i < flp->nfsfl_fhcnt; i++)
5042 free(flp->nfsfl_fh[i], M_NFSFH);
5043 if (flp->nfsfl_devp != NULL)
5044 flp->nfsfl_devp->nfsdi_layoutrefs--;
5045 free(flp, M_NFSFLAYOUT);
5049 * Free up a file layout devinfo structure.
5052 nfscl_freedevinfo(struct nfscldevinfo *dip)
5055 free(dip, M_NFSDEVINFO);
5059 * Mark any layouts that match as recalled.
5062 nfscl_layoutrecall(int recalltype, struct nfscllayout *lyp, uint32_t iomode,
5063 uint64_t off, uint64_t len, uint32_t stateseqid,
5064 struct nfsclrecalllayout *recallp)
5066 struct nfsclrecalllayout *rp, *orp;
5068 recallp->nfsrecly_recalltype = recalltype;
5069 recallp->nfsrecly_iomode = iomode;
5070 recallp->nfsrecly_stateseqid = stateseqid;
5071 recallp->nfsrecly_off = off;
5072 recallp->nfsrecly_len = len;
5074 * Order the list as file returns first, followed by fsid and any
5075 * returns, both in increasing stateseqid order.
5076 * Note that the seqids wrap around, so 1 is after 0xffffffff.
5077 * (I'm not sure this is correct because I find RFC5661 confusing
5078 * on this, but hopefully it will work ok.)
5081 LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) {
5083 if ((recalltype == NFSLAYOUTRETURN_FILE &&
5084 (rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE ||
5085 nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) ||
5086 (recalltype != NFSLAYOUTRETURN_FILE &&
5087 rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE &&
5088 nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) {
5089 LIST_INSERT_BEFORE(rp, recallp, nfsrecly_list);
5095 LIST_INSERT_HEAD(&lyp->nfsly_recall, recallp,
5098 LIST_INSERT_AFTER(orp, recallp, nfsrecly_list);
5100 lyp->nfsly_flags |= NFSLY_RECALL;
5105 * Compare the two seqids for ordering. The trick is that the seqids can
5106 * wrap around from 0xffffffff->0, so check for the cases where one
5107 * has wrapped around.
5108 * Return 1 if seqid1 comes before seqid2, 0 otherwise.
5111 nfscl_seq(uint32_t seqid1, uint32_t seqid2)
5114 if (seqid2 > seqid1 && (seqid2 - seqid1) >= 0x7fffffff)
5115 /* seqid2 has wrapped around. */
5117 if (seqid1 > seqid2 && (seqid1 - seqid2) >= 0x7fffffff)
5118 /* seqid1 has wrapped around. */
5120 if (seqid1 <= seqid2)
5126 * Do a layout return for each of the recalls.
5129 nfscl_layoutreturn(struct nfsmount *nmp, struct nfscllayout *lyp,
5130 struct ucred *cred, NFSPROC_T *p)
5132 struct nfsclrecalllayout *rp;
5133 nfsv4stateid_t stateid;
5135 NFSBCOPY(lyp->nfsly_stateid.other, stateid.other, NFSX_STATEIDOTHER);
5136 LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) {
5137 stateid.seqid = rp->nfsrecly_stateseqid;
5138 (void)nfsrpc_layoutreturn(nmp, lyp->nfsly_fh,
5139 lyp->nfsly_fhlen, 0, NFSLAYOUT_NFSV4_1_FILES,
5140 rp->nfsrecly_iomode, rp->nfsrecly_recalltype,
5141 rp->nfsrecly_off, rp->nfsrecly_len,
5142 &stateid, 0, NULL, cred, p, NULL);
5147 * Do the layout commit for a file layout.
5150 nfscl_dolayoutcommit(struct nfsmount *nmp, struct nfscllayout *lyp,
5151 struct ucred *cred, NFSPROC_T *p)
5153 struct nfsclflayout *flp;
5157 LIST_FOREACH(flp, &lyp->nfsly_flayrw, nfsfl_list) {
5158 if (flp->nfsfl_off <= lyp->nfsly_lastbyte) {
5159 len = flp->nfsfl_end - flp->nfsfl_off;
5160 error = nfsrpc_layoutcommit(nmp, lyp->nfsly_fh,
5161 lyp->nfsly_fhlen, 0, flp->nfsfl_off, len,
5162 lyp->nfsly_lastbyte, &lyp->nfsly_stateid,
5163 NFSLAYOUT_NFSV4_1_FILES, 0, NULL, cred, p, NULL);
5164 NFSCL_DEBUG(4, "layoutcommit err=%d\n", error);
5165 if (error == NFSERR_NOTSUPP) {
5166 /* If not supported, don't bother doing it. */
5168 nmp->nm_state |= NFSSTA_NOLAYOUTCOMMIT;
5177 * Commit all layouts for a file (vnode).
5180 nfscl_layoutcommit(vnode_t vp, NFSPROC_T *p)
5182 struct nfsclclient *clp;
5183 struct nfscllayout *lyp;
5184 struct nfsnode *np = VTONFS(vp);
5186 struct nfsmount *nmp;
5188 mp = vnode_mount(vp);
5190 if (NFSHASNOLAYOUTCOMMIT(nmp))
5198 lyp = nfscl_findlayout(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
5203 nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
5204 if ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) {
5209 if ((lyp->nfsly_flags & NFSLY_WRITTEN) != 0) {
5210 lyp->nfsly_flags &= ~NFSLY_WRITTEN;
5212 NFSCL_DEBUG(4, "do layoutcommit2\n");
5213 nfscl_dolayoutcommit(clp->nfsc_nmp, lyp, NFSPROCCRED(p), p);
5217 nfsv4_relref(&lyp->nfsly_lock);