2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2009 Rick Macklem, University of Guelph
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
34 * These functions implement the client side state handling for NFSv4.
35 * NFSv4 state handling:
36 * - A lockowner is used to determine lock contention, so it
37 * corresponds directly to a Posix pid. (1 to 1 mapping)
38 * - The correct granularity of an OpenOwner is not nearly so
39 * obvious. An OpenOwner does the following:
40 * - provides a serial sequencing of Open/Close/Lock-with-new-lockowner
41 * - is used to check for Open/Share contention (not applicable to
42 * this client, since all Opens are Deny_None)
43 * As such, I considered both extreme.
44 * 1 OpenOwner per ClientID - Simple to manage, but fully serializes
45 * all Open, Close and Lock (with a new lockowner) Ops.
46 * 1 OpenOwner for each Open - This one results in an OpenConfirm for
47 * every Open, for most servers.
48 * So, I chose to use the same mapping as I did for LockOwnwers.
49 * The main concern here is that you can end up with multiple Opens
50 * for the same File Handle, but on different OpenOwners (opens
51 * inherited from parents, grandparents...) and you do not know
52 * which of these the vnodeop close applies to. This is handled by
53 * delaying the Close Op(s) until all of the Opens have been closed.
54 * (It is not yet obvious if this is the correct granularity.)
55 * - How the code handles serialization:
56 * - For the ClientId, it uses an exclusive lock while getting its
57 * SetClientId and during recovery. Otherwise, it uses a shared
58 * lock via a reference count.
59 * - For the rest of the data structures, it uses an SMP mutex
60 * (once the nfs client is SMP safe) and doesn't sleep while
61 * manipulating the linked lists.
62 * - The serialization of Open/Close/Lock/LockU falls out in the
63 * "wash", since OpenOwners and LockOwners are both mapped from
64 * Posix pid. In other words, there is only one Posix pid using
65 * any given owner, so that owner is serialized. (If you change
66 * the granularity of the OpenOwner, then code must be added to
67 * serialize Ops on the OpenOwner.)
68 * - When to get rid of OpenOwners and LockOwners.
69 * - The function nfscl_cleanup_common() is executed after a process exits.
70 * It goes through the client list looking for all Open and Lock Owners.
71 * When one is found, it is marked "defunct" or in the case of
72 * an OpenOwner without any Opens, freed.
73 * The renew thread scans for defunct Owners and gets rid of them,
74 * if it can. The LockOwners will also be deleted when the
75 * associated Open is closed.
76 * - If the LockU or Close Op(s) fail during close in a way
77 * that could be recovered upon retry, they are relinked to the
78 * ClientId's defunct open list and retried by the renew thread
79 * until they succeed or an unmount/recovery occurs.
80 * (Since we are done with them, they do not need to be recovered.)
83 #include <fs/nfs/nfsport.h>
88 extern struct nfsstatsv1 nfsstatsv1;
89 extern struct nfsreqhead nfsd_reqq;
90 extern u_int32_t newnfs_false, newnfs_true;
91 extern int nfscl_debuglevel;
92 extern int nfscl_enablecallb;
93 extern int nfs_numnfscbd;
97 struct nfsclhead nfsclhead; /* Head of clientid list */
98 int nfscl_deleghighwater = NFSCLDELEGHIGHWATER;
99 int nfscl_layouthighwater = NFSCLLAYOUTHIGHWATER;
101 static int nfscl_delegcnt = 0;
102 static int nfscl_layoutcnt = 0;
103 static int nfscl_getopen(struct nfsclownerhead *, u_int8_t *, int, u_int8_t *,
104 u_int8_t *, u_int32_t, struct nfscllockowner **, struct nfsclopen **);
105 static bool nfscl_checkown(struct nfsclowner *, struct nfsclopen *, uint8_t *,
106 uint8_t *, struct nfscllockowner **, struct nfsclopen **,
107 struct nfsclopen **);
108 static void nfscl_clrelease(struct nfsclclient *);
109 static void nfscl_cleanclient(struct nfsclclient *);
110 static void nfscl_expireclient(struct nfsclclient *, struct nfsmount *,
111 struct ucred *, NFSPROC_T *);
112 static int nfscl_expireopen(struct nfsclclient *, struct nfsclopen *,
113 struct nfsmount *, struct ucred *, NFSPROC_T *);
114 static void nfscl_recover(struct nfsclclient *, bool *, struct ucred *,
116 static void nfscl_insertlock(struct nfscllockowner *, struct nfscllock *,
117 struct nfscllock *, int);
118 static int nfscl_updatelock(struct nfscllockowner *, struct nfscllock **,
119 struct nfscllock **, int);
120 static void nfscl_delegreturnall(struct nfsclclient *, NFSPROC_T *);
121 static u_int32_t nfscl_nextcbident(void);
122 static mount_t nfscl_getmnt(int, uint8_t *, u_int32_t, struct nfsclclient **);
123 static struct nfsclclient *nfscl_getclnt(u_int32_t);
124 static struct nfsclclient *nfscl_getclntsess(uint8_t *);
125 static struct nfscldeleg *nfscl_finddeleg(struct nfsclclient *, u_int8_t *,
127 static void nfscl_retoncloselayout(vnode_t, struct nfsclclient *, uint8_t *,
128 int, struct nfsclrecalllayout **);
129 static void nfscl_reldevinfo_locked(struct nfscldevinfo *);
130 static struct nfscllayout *nfscl_findlayout(struct nfsclclient *, u_int8_t *,
132 static struct nfscldevinfo *nfscl_finddevinfo(struct nfsclclient *, uint8_t *);
133 static int nfscl_checkconflict(struct nfscllockownerhead *, struct nfscllock *,
134 u_int8_t *, struct nfscllock **);
135 static void nfscl_freealllocks(struct nfscllockownerhead *, int);
136 static int nfscl_localconflict(struct nfsclclient *, u_int8_t *, int,
137 struct nfscllock *, u_int8_t *, struct nfscldeleg *, struct nfscllock **);
138 static void nfscl_newopen(struct nfsclclient *, struct nfscldeleg *,
139 struct nfsclowner **, struct nfsclowner **, struct nfsclopen **,
140 struct nfsclopen **, u_int8_t *, u_int8_t *, int, struct ucred *, int *);
141 static int nfscl_moveopen(vnode_t , struct nfsclclient *,
142 struct nfsmount *, struct nfsclopen *, struct nfsclowner *,
143 struct nfscldeleg *, struct ucred *, NFSPROC_T *);
144 static void nfscl_totalrecall(struct nfsclclient *);
145 static int nfscl_relock(vnode_t , struct nfsclclient *, struct nfsmount *,
146 struct nfscllockowner *, struct nfscllock *, struct ucred *, NFSPROC_T *);
147 static int nfscl_tryopen(struct nfsmount *, vnode_t , u_int8_t *, int,
148 u_int8_t *, int, u_int32_t, struct nfsclopen *, u_int8_t *, int,
149 struct nfscldeleg **, int, u_int32_t, struct ucred *, NFSPROC_T *);
150 static int nfscl_trylock(struct nfsmount *, vnode_t , u_int8_t *,
151 int, struct nfscllockowner *, int, int, u_int64_t, u_int64_t, short,
152 struct ucred *, NFSPROC_T *);
153 static int nfsrpc_reopen(struct nfsmount *, u_int8_t *, int, u_int32_t,
154 struct nfsclopen *, struct nfscldeleg **, struct ucred *, NFSPROC_T *);
155 static void nfscl_freedeleg(struct nfscldeleghead *, struct nfscldeleg *);
156 static int nfscl_errmap(struct nfsrv_descript *, u_int32_t);
157 static void nfscl_cleanup_common(struct nfsclclient *, u_int8_t *);
158 static int nfscl_recalldeleg(struct nfsclclient *, struct nfsmount *,
159 struct nfscldeleg *, vnode_t, struct ucred *, NFSPROC_T *, int);
160 static void nfscl_freeopenowner(struct nfsclowner *, int);
161 static void nfscl_cleandeleg(struct nfscldeleg *);
162 static int nfscl_trydelegreturn(struct nfscldeleg *, struct ucred *,
163 struct nfsmount *, NFSPROC_T *);
164 static void nfscl_emptylockowner(struct nfscllockowner *,
165 struct nfscllockownerfhhead *);
166 static void nfscl_mergeflayouts(struct nfsclflayouthead *,
167 struct nfsclflayouthead *);
168 static int nfscl_layoutrecall(int, struct nfscllayout *, uint32_t, uint64_t,
169 uint64_t, uint32_t, uint32_t, uint32_t, char *, struct nfsclrecalllayout *);
170 static int nfscl_seq(uint32_t, uint32_t);
171 static void nfscl_layoutreturn(struct nfsmount *, struct nfscllayout *,
172 struct ucred *, NFSPROC_T *);
173 static void nfscl_dolayoutcommit(struct nfsmount *, struct nfscllayout *,
174 struct ucred *, NFSPROC_T *);
176 static short nfscberr_null[] = {
181 static short nfscberr_getattr[] = {
190 static short nfscberr_recall[] = {
200 static short *nfscl_cberrmap[] = {
208 #define NETFAMILY(clp) \
209 (((clp)->nfsc_flags & NFSCLFLAGS_AFINET6) ? AF_INET6 : AF_INET)
212 * Called for an open operation.
213 * If the nfhp argument is NULL, just get an openowner.
216 nfscl_open(vnode_t vp, u_int8_t *nfhp, int fhlen, u_int32_t amode, int usedeleg,
217 struct ucred *cred, NFSPROC_T *p, struct nfsclowner **owpp,
218 struct nfsclopen **opp, int *newonep, int *retp, int lockit)
220 struct nfsclclient *clp;
221 struct nfsclowner *owp, *nowp;
222 struct nfsclopen *op = NULL, *nop = NULL;
223 struct nfscldeleg *dp;
224 struct nfsclownerhead *ohp;
225 u_int8_t own[NFSV4CL_LOCKNAMELEN];
236 * Might need one or both of these, so MALLOC them now, to
237 * avoid a tsleep() in MALLOC later.
239 nowp = malloc(sizeof (struct nfsclowner),
240 M_NFSCLOWNER, M_WAITOK);
242 nop = malloc(sizeof (struct nfsclopen) +
243 fhlen - 1, M_NFSCLOPEN, M_WAITOK);
244 ret = nfscl_getcl(vp->v_mount, cred, p, 1, &clp);
246 free(nowp, M_NFSCLOWNER);
248 free(nop, M_NFSCLOPEN);
253 * Get the Open iff it already exists.
254 * If none found, add the new one or return error, depending upon
259 /* First check the delegation list */
260 if (nfhp != NULL && usedeleg) {
261 LIST_FOREACH(dp, NFSCLDELEGHASH(clp, nfhp, fhlen), nfsdl_hash) {
262 if (dp->nfsdl_fhlen == fhlen &&
263 !NFSBCMP(nfhp, dp->nfsdl_fh, fhlen)) {
264 if (!(amode & NFSV4OPEN_ACCESSWRITE) ||
265 (dp->nfsdl_flags & NFSCLDL_WRITE))
273 /* For NFSv4.1/4.2 and this option, use a single open_owner. */
274 if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount)))
275 nfscl_filllockowner(NULL, own, F_POSIX);
277 nfscl_filllockowner(p->td_proc, own, F_POSIX);
279 ohp = &dp->nfsdl_owner;
281 ohp = &clp->nfsc_owner;
282 /* Now, search for an openowner */
283 LIST_FOREACH(owp, ohp, nfsow_list) {
284 if (!NFSBCMP(owp->nfsow_owner, own, NFSV4CL_LOCKNAMELEN))
289 * Create a new open, as required.
291 nfscl_newopen(clp, dp, &owp, &nowp, &op, &nop, own, nfhp, fhlen,
295 * Now, check the mode on the open and return the appropriate
299 if (nfhp != NULL && dp != NULL && nop == NULL)
300 /* new local open on delegation */
301 *retp = NFSCLOPEN_SETCRED;
303 *retp = NFSCLOPEN_OK;
305 if (op != NULL && (amode & ~(op->nfso_mode))) {
306 op->nfso_mode |= amode;
307 if (retp != NULL && dp == NULL)
308 *retp = NFSCLOPEN_DOOPEN;
312 * Serialize modifications to the open owner for multiple threads
313 * within the same process using a read/write sleep lock.
314 * For NFSv4.1 and a single OpenOwner, allow concurrent open operations
315 * by acquiring a shared lock. The close operations still use an
316 * exclusive lock for this case.
319 if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount))) {
321 * Get a shared lock on the OpenOwner, but first
322 * wait for any pending exclusive lock, so that the
323 * exclusive locker gets priority.
325 nfsv4_lock(&owp->nfsow_rwlock, 0, NULL,
326 NFSCLSTATEMUTEXPTR, NULL);
327 nfsv4_getref(&owp->nfsow_rwlock, NULL,
328 NFSCLSTATEMUTEXPTR, NULL);
330 nfscl_lockexcl(&owp->nfsow_rwlock, NFSCLSTATEMUTEXPTR);
334 free(nowp, M_NFSCLOWNER);
336 free(nop, M_NFSCLOPEN);
345 * Create a new open, as required.
348 nfscl_newopen(struct nfsclclient *clp, struct nfscldeleg *dp,
349 struct nfsclowner **owpp, struct nfsclowner **nowpp, struct nfsclopen **opp,
350 struct nfsclopen **nopp, u_int8_t *own, u_int8_t *fhp, int fhlen,
351 struct ucred *cred, int *newonep)
353 struct nfsclowner *owp = *owpp, *nowp;
354 struct nfsclopen *op, *nop;
364 if (owp == NULL && nowp != NULL) {
365 NFSBCOPY(own, nowp->nfsow_owner, NFSV4CL_LOCKNAMELEN);
366 LIST_INIT(&nowp->nfsow_open);
367 nowp->nfsow_clp = clp;
368 nowp->nfsow_seqid = 0;
369 nowp->nfsow_defunct = 0;
370 nfscl_lockinit(&nowp->nfsow_rwlock);
372 nfsstatsv1.cllocalopenowners++;
373 LIST_INSERT_HEAD(&dp->nfsdl_owner, nowp, nfsow_list);
375 nfsstatsv1.clopenowners++;
376 LIST_INSERT_HEAD(&clp->nfsc_owner, nowp, nfsow_list);
384 /* If an fhp has been specified, create an Open as well. */
386 /* and look for the correct open, based upon FH */
387 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
388 if (op->nfso_fhlen == fhlen &&
389 !NFSBCMP(op->nfso_fh, fhp, fhlen))
392 if (op == NULL && nop != NULL) {
395 nop->nfso_opencnt = 0;
396 nop->nfso_posixlock = 1;
397 nop->nfso_fhlen = fhlen;
398 NFSBCOPY(fhp, nop->nfso_fh, fhlen);
399 LIST_INIT(&nop->nfso_lock);
400 nop->nfso_stateid.seqid = 0;
401 nop->nfso_stateid.other[0] = 0;
402 nop->nfso_stateid.other[1] = 0;
403 nop->nfso_stateid.other[2] = 0;
404 KASSERT(cred != NULL, ("%s: cred NULL\n", __func__));
405 newnfs_copyincred(cred, &nop->nfso_cred);
407 TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
408 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp,
410 dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
411 nfsstatsv1.cllocalopens++;
413 nfsstatsv1.clopens++;
415 LIST_INSERT_HEAD(&owp->nfsow_open, nop, nfso_list);
427 * Called to find/add a delegation to a client.
430 nfscl_deleg(mount_t mp, struct nfsclclient *clp, u_int8_t *nfhp,
431 int fhlen, struct ucred *cred, NFSPROC_T *p, struct nfscldeleg **dpp)
433 struct nfscldeleg *dp = *dpp, *tdp;
436 * First, if we have received a Read delegation for a file on a
437 * read/write file system, just return it, because they aren't
440 if (mp != NULL && dp != NULL && !NFSMNT_RDONLY(mp) &&
441 (dp->nfsdl_flags & NFSCLDL_READ)) {
442 (void) nfscl_trydelegreturn(dp, cred, VFSTONFS(mp), p);
443 free(dp, M_NFSCLDELEG);
448 /* Look for the correct deleg, based upon FH */
450 tdp = nfscl_finddeleg(clp, nfhp, fhlen);
454 return (NFSERR_BADSTATEID);
457 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list);
458 LIST_INSERT_HEAD(NFSCLDELEGHASH(clp, nfhp, fhlen), dp,
460 dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
461 nfsstatsv1.cldelegates++;
465 * Delegation already exists, what do we do if a new one??
468 printf("Deleg already exists!\n");
469 free(dp, M_NFSCLDELEG);
480 * Find a delegation for this file handle. Return NULL upon failure.
482 static struct nfscldeleg *
483 nfscl_finddeleg(struct nfsclclient *clp, u_int8_t *fhp, int fhlen)
485 struct nfscldeleg *dp;
487 LIST_FOREACH(dp, NFSCLDELEGHASH(clp, fhp, fhlen), nfsdl_hash) {
488 if (dp->nfsdl_fhlen == fhlen &&
489 !NFSBCMP(dp->nfsdl_fh, fhp, fhlen))
496 * Get a stateid for an I/O operation. First, look for an open and iff
497 * found, return either a lockowner stateid or the open stateid.
498 * If no Open is found, just return error and the special stateid of all zeros.
501 nfscl_getstateid(vnode_t vp, u_int8_t *nfhp, int fhlen, u_int32_t mode,
502 int fords, struct ucred *cred, NFSPROC_T *p, nfsv4stateid_t *stateidp,
505 struct nfsclclient *clp;
506 struct nfsclowner *owp;
507 struct nfsclopen *op = NULL, *top;
508 struct nfscllockowner *lp;
509 struct nfscldeleg *dp;
511 struct nfsmount *nmp;
512 u_int8_t own[NFSV4CL_LOCKNAMELEN];
518 * Initially, just set the special stateid of all zeros.
519 * (Don't do this for a DS, since the special stateid can't be used.)
523 stateidp->other[0] = 0;
524 stateidp->other[1] = 0;
525 stateidp->other[2] = 0;
527 if (vnode_vtype(vp) != VREG)
530 nmp = VFSTONFS(vp->v_mount);
532 clp = nfscl_findcl(nmp);
539 * Wait for recovery to complete.
541 while ((clp->nfsc_flags & NFSCLFLAGS_RECVRINPROG))
542 (void) nfsmsleep(&clp->nfsc_flags, NFSCLSTATEMUTEXPTR,
543 PZERO, "nfsrecvr", NULL);
546 * First, look for a delegation.
548 LIST_FOREACH(dp, NFSCLDELEGHASH(clp, nfhp, fhlen), nfsdl_hash) {
549 if (dp->nfsdl_fhlen == fhlen &&
550 !NFSBCMP(nfhp, dp->nfsdl_fh, fhlen)) {
551 if (!(mode & NFSV4OPEN_ACCESSWRITE) ||
552 (dp->nfsdl_flags & NFSCLDL_WRITE)) {
553 stateidp->seqid = dp->nfsdl_stateid.seqid;
554 stateidp->other[0] = dp->nfsdl_stateid.other[0];
555 stateidp->other[1] = dp->nfsdl_stateid.other[1];
556 stateidp->other[2] = dp->nfsdl_stateid.other[2];
557 if (!(np->n_flag & NDELEGRECALL)) {
558 TAILQ_REMOVE(&clp->nfsc_deleg, dp,
560 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp,
562 dp->nfsdl_timestamp = NFSD_MONOSEC +
564 dp->nfsdl_rwlock.nfslock_usecnt++;
565 *lckpp = (void *)&dp->nfsdl_rwlock;
576 * If p != NULL, we want to search the parentage tree
577 * for a matching OpenOwner and use that.
579 if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount)))
580 nfscl_filllockowner(NULL, own, F_POSIX);
582 nfscl_filllockowner(p->td_proc, own, F_POSIX);
584 error = nfscl_getopen(&clp->nfsc_owner, nfhp, fhlen, own, own,
586 if (error == 0 && lp != NULL && fords == 0) {
587 /* Don't return a lock stateid for a DS. */
589 lp->nfsl_stateid.seqid;
591 lp->nfsl_stateid.other[0];
593 lp->nfsl_stateid.other[1];
595 lp->nfsl_stateid.other[2];
601 /* If not found, just look for any OpenOwner that will work. */
604 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
605 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
606 if (op->nfso_fhlen == fhlen &&
607 !NFSBCMP(op->nfso_fh, nfhp, fhlen)) {
608 if (top == NULL && (op->nfso_mode &
609 NFSV4OPEN_ACCESSWRITE) != 0 &&
610 (mode & NFSV4OPEN_ACCESSREAD) != 0)
612 if ((mode & op->nfso_mode) == mode) {
622 NFSCL_DEBUG(2, "openmode top=%p\n", top);
623 if (top == NULL || NFSHASOPENMODE(nmp)) {
630 * For read aheads or write behinds, use the open cred.
631 * A read ahead or write behind is indicated by p == NULL.
634 newnfs_copycred(&op->nfso_cred, cred);
638 * No lock stateid, so return the open stateid.
640 stateidp->seqid = op->nfso_stateid.seqid;
641 stateidp->other[0] = op->nfso_stateid.other[0];
642 stateidp->other[1] = op->nfso_stateid.other[1];
643 stateidp->other[2] = op->nfso_stateid.other[2];
649 * Search for a matching file, mode and, optionally, lockowner.
652 nfscl_getopen(struct nfsclownerhead *ohp, u_int8_t *nfhp, int fhlen,
653 u_int8_t *openown, u_int8_t *lockown, u_int32_t mode,
654 struct nfscllockowner **lpp, struct nfsclopen **opp)
656 struct nfsclowner *owp;
657 struct nfsclopen *op, *rop, *rop2;
663 * rop will be set to the open to be returned. There are three
664 * variants of this, all for an open of the correct file:
665 * 1 - A match of lockown.
666 * 2 - A match of the openown, when no lockown match exists.
667 * 3 - A match for any open, if no openown or lockown match exists.
668 * Looking for #2 over #3 probably isn't necessary, but since
669 * RFC3530 is vague w.r.t. the relationship between openowners and
670 * lockowners, I think this is the safer way to go.
675 /* Search the client list */
676 LIST_FOREACH(owp, ohp, nfsow_list) {
677 /* and look for the correct open */
678 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
679 if (op->nfso_fhlen == fhlen &&
680 !NFSBCMP(op->nfso_fh, nfhp, fhlen)
681 && (op->nfso_mode & mode) == mode)
682 keep_looping = nfscl_checkown(owp, op, openown,
683 lockown, lpp, &rop, &rop2);
698 /* Check for an owner match. */
700 nfscl_checkown(struct nfsclowner *owp, struct nfsclopen *op, uint8_t *openown,
701 uint8_t *lockown, struct nfscllockowner **lpp, struct nfsclopen **ropp,
702 struct nfsclopen **ropp2)
704 struct nfscllockowner *lp;
709 /* Now look for a matching lockowner. */
710 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
711 if (!NFSBCMP(lp->nfsl_owner, lockown,
712 NFSV4CL_LOCKNAMELEN)) {
719 if (*ropp == NULL && !NFSBCMP(owp->nfsow_owner, openown,
720 NFSV4CL_LOCKNAMELEN)) {
723 keep_looping = false;
727 return (keep_looping);
731 * Release use of an open owner. Called when open operations are done
732 * with the open owner.
735 nfscl_ownerrelease(struct nfsmount *nmp, struct nfsclowner *owp,
736 __unused int error, __unused int candelete, int unlocked)
743 if (NFSHASONEOPENOWN(nmp))
744 nfsv4_relref(&owp->nfsow_rwlock);
746 nfscl_lockunlock(&owp->nfsow_rwlock);
748 nfscl_clrelease(owp->nfsow_clp);
753 * Release use of an open structure under an open owner.
756 nfscl_openrelease(struct nfsmount *nmp, struct nfsclopen *op, int error,
759 struct nfsclclient *clp;
760 struct nfsclowner *owp;
766 if (NFSHASONEOPENOWN(nmp))
767 nfsv4_relref(&owp->nfsow_rwlock);
769 nfscl_lockunlock(&owp->nfsow_rwlock);
770 clp = owp->nfsow_clp;
771 if (error && candelete && op->nfso_opencnt == 0)
772 nfscl_freeopen(op, 0);
773 nfscl_clrelease(clp);
778 * Called to get a clientid structure. It will optionally lock the
779 * client data structures to do the SetClientId/SetClientId_confirm,
780 * but will release that lock and return the clientid with a reference
782 * If the "cred" argument is NULL, a new clientid should not be created.
783 * If the "p" argument is NULL, a SetClientID/SetClientIDConfirm cannot
785 * The start_renewthread argument tells nfscl_getcl() to start a renew
786 * thread if this creates a new clp.
787 * It always clpp with a reference count on it, unless returning an error.
790 nfscl_getcl(struct mount *mp, struct ucred *cred, NFSPROC_T *p,
791 int start_renewthread, struct nfsclclient **clpp)
793 struct nfsclclient *clp;
794 struct nfsclclient *newclp = NULL;
795 struct nfsmount *nmp;
796 char uuid[HOSTUUIDLEN];
797 int igotlock = 0, error, trystalecnt, clidinusedelay, i;
802 getcredhostuuid(cred, uuid, sizeof uuid);
803 idlen = strlen(uuid);
805 idlen += sizeof (u_int64_t);
807 idlen += sizeof (u_int64_t) + 16; /* 16 random bytes */
809 sizeof (struct nfsclclient) + idlen - 1, M_NFSCLCLIENT,
814 * If a forced dismount is already in progress, don't
815 * allocate a new clientid and get out now. For the case where
816 * clp != NULL, this is a harmless optimization.
818 if (NFSCL_FORCEDISM(mp)) {
821 free(newclp, M_NFSCLCLIENT);
826 if (newclp == NULL) {
831 clp->nfsc_idlen = idlen;
832 LIST_INIT(&clp->nfsc_owner);
833 TAILQ_INIT(&clp->nfsc_deleg);
834 TAILQ_INIT(&clp->nfsc_layout);
835 LIST_INIT(&clp->nfsc_devinfo);
836 for (i = 0; i < NFSCLDELEGHASHSIZE; i++)
837 LIST_INIT(&clp->nfsc_deleghash[i]);
838 for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++)
839 LIST_INIT(&clp->nfsc_layouthash[i]);
840 clp->nfsc_flags = NFSCLFLAGS_INITED;
841 clp->nfsc_clientidrev = 1;
842 clp->nfsc_cbident = nfscl_nextcbident();
843 nfscl_fillclid(nmp->nm_clval, uuid, clp->nfsc_id,
845 LIST_INSERT_HEAD(&nfsclhead, clp, nfsc_list);
849 if (start_renewthread != 0)
850 nfscl_start_renewthread(clp);
854 free(newclp, M_NFSCLCLIENT);
857 while ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0 && !igotlock &&
858 !NFSCL_FORCEDISM(mp))
859 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
860 NFSCLSTATEMUTEXPTR, mp);
863 * Call nfsv4_lock() with "iwantlock == 0" so that it will
864 * wait for a pending exclusive lock request. This gives the
865 * exclusive lock request priority over this shared lock
867 * An exclusive lock on nfsc_lock is used mainly for server
870 nfsv4_lock(&clp->nfsc_lock, 0, NULL, NFSCLSTATEMUTEXPTR, mp);
871 nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
873 if (igotlock == 0 && NFSCL_FORCEDISM(mp)) {
875 * Both nfsv4_lock() and nfsv4_getref() know to check
876 * for NFSCL_FORCEDISM() and return without sleeping to
877 * wait for the exclusive lock to be released, since it
878 * might be held by nfscl_umount() and we need to get out
879 * now for that case and not wait until nfscl_umount()
888 * If it needs a clientid, do the setclientid now.
890 if ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0) {
892 panic("nfscl_clget");
893 if (p == NULL || cred == NULL) {
895 nfsv4_unlock(&clp->nfsc_lock, 0);
900 * If RFC3530 Sec. 14.2.33 is taken literally,
901 * NFSERR_CLIDINUSE will be returned persistently for the
902 * case where a new mount of the same file system is using
903 * a different principal. In practice, NFSERR_CLIDINUSE is
904 * only returned when there is outstanding unexpired state
905 * on the clientid. As such, try for twice the lease
906 * interval, if we know what that is. Otherwise, make a
908 * The case of returning NFSERR_STALECLIENTID is far less
909 * likely, but might occur if there is a significant delay
910 * between doing the SetClientID and SetClientIDConfirm Ops,
911 * such that the server throws away the clientid before
912 * receiving the SetClientIDConfirm.
914 if (clp->nfsc_renew > 0)
915 clidinusedelay = NFSCL_LEASE(clp->nfsc_renew) * 2;
917 clidinusedelay = 120;
920 error = nfsrpc_setclient(nmp, clp, 0, NULL, cred, p);
921 if (error == NFSERR_STALECLIENTID ||
922 error == NFSERR_STALEDONTRECOVER ||
923 error == NFSERR_BADSESSION ||
924 error == NFSERR_CLIDINUSE) {
925 (void) nfs_catnap(PZERO, error, "nfs_setcl");
927 } while (((error == NFSERR_STALECLIENTID ||
928 error == NFSERR_BADSESSION ||
929 error == NFSERR_STALEDONTRECOVER) && --trystalecnt > 0) ||
930 (error == NFSERR_CLIDINUSE && --clidinusedelay > 0));
933 nfsv4_unlock(&clp->nfsc_lock, 0);
937 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
941 nfsv4_unlock(&clp->nfsc_lock, 1);
950 * Get a reference to a clientid and return it, if valid.
953 nfscl_findcl(struct nfsmount *nmp)
955 struct nfsclclient *clp;
958 if (clp == NULL || !(clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID))
964 * Release the clientid structure. It may be locked or reference counted.
967 nfscl_clrelease(struct nfsclclient *clp)
970 if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK)
971 nfsv4_unlock(&clp->nfsc_lock, 0);
973 nfsv4_relref(&clp->nfsc_lock);
977 * External call for nfscl_clrelease.
980 nfscl_clientrelease(struct nfsclclient *clp)
984 if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK)
985 nfsv4_unlock(&clp->nfsc_lock, 0);
987 nfsv4_relref(&clp->nfsc_lock);
992 * Called when wanting to lock a byte region.
995 nfscl_getbytelock(vnode_t vp, u_int64_t off, u_int64_t len,
996 short type, struct ucred *cred, NFSPROC_T *p, struct nfsclclient *rclp,
997 int recovery, void *id, int flags, u_int8_t *rownp, u_int8_t *ropenownp,
998 struct nfscllockowner **lpp, int *newonep, int *donelocallyp)
1000 struct nfscllockowner *lp;
1001 struct nfsclopen *op;
1002 struct nfsclclient *clp;
1003 struct nfscllockowner *nlp;
1004 struct nfscllock *nlop, *otherlop;
1005 struct nfscldeleg *dp = NULL, *ldp = NULL;
1006 struct nfscllockownerhead *lhp = NULL;
1008 u_int8_t own[NFSV4CL_LOCKNAMELEN], *ownp, openown[NFSV4CL_LOCKNAMELEN];
1010 int error = 0, ret, donelocally = 0;
1013 /* For Lock Ops, the open mode doesn't matter, so use 0 to match any. */
1022 * Might need these, so MALLOC them now, to
1023 * avoid a tsleep() in MALLOC later.
1026 sizeof (struct nfscllockowner), M_NFSCLLOCKOWNER, M_WAITOK);
1028 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1030 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1031 nlop->nfslo_type = type;
1032 nlop->nfslo_first = off;
1033 if (len == NFS64BITSSET) {
1034 nlop->nfslo_end = NFS64BITSSET;
1036 nlop->nfslo_end = off + len;
1037 if (nlop->nfslo_end <= nlop->nfslo_first)
1038 error = NFSERR_INVAL;
1045 error = nfscl_getcl(vp->v_mount, cred, p, 1, &clp);
1048 free(nlp, M_NFSCLLOCKOWNER);
1049 free(otherlop, M_NFSCLLOCK);
1050 free(nlop, M_NFSCLLOCK);
1057 openownp = ropenownp;
1059 nfscl_filllockowner(id, own, flags);
1061 if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount)))
1062 nfscl_filllockowner(NULL, openown, F_POSIX);
1064 nfscl_filllockowner(p->td_proc, openown, F_POSIX);
1070 * First, search for a delegation. If one exists for this file,
1071 * the lock can be done locally against it, so long as there
1072 * isn't a local lock conflict.
1074 ldp = dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
1075 np->n_fhp->nfh_len);
1076 /* Just sanity check for correct type of delegation */
1077 if (dp != NULL && ((dp->nfsdl_flags &
1078 (NFSCLDL_RECALL | NFSCLDL_DELEGRET)) != 0 ||
1080 (dp->nfsdl_flags & NFSCLDL_WRITE) == 0)))
1084 /* Now, find an open and maybe a lockowner. */
1085 ret = nfscl_getopen(&dp->nfsdl_owner, np->n_fhp->nfh_fh,
1086 np->n_fhp->nfh_len, openownp, ownp, mode, NULL, &op);
1088 ret = nfscl_getopen(&clp->nfsc_owner,
1089 np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp,
1090 ownp, mode, NULL, &op);
1092 lhp = &dp->nfsdl_lock;
1093 TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
1094 TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list);
1095 dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
1103 * Get the related Open and maybe lockowner.
1105 error = nfscl_getopen(&clp->nfsc_owner,
1106 np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp,
1107 ownp, mode, &lp, &op);
1109 lhp = &op->nfso_lock;
1111 if (!error && !recovery)
1112 error = nfscl_localconflict(clp, np->n_fhp->nfh_fh,
1113 np->n_fhp->nfh_len, nlop, ownp, ldp, NULL);
1116 nfscl_clrelease(clp);
1119 free(nlp, M_NFSCLLOCKOWNER);
1120 free(otherlop, M_NFSCLLOCK);
1121 free(nlop, M_NFSCLLOCK);
1126 * Ok, see if a lockowner exists and create one, as required.
1129 LIST_FOREACH(lp, lhp, nfsl_list) {
1130 if (!NFSBCMP(lp->nfsl_owner, ownp, NFSV4CL_LOCKNAMELEN))
1134 NFSBCOPY(ownp, nlp->nfsl_owner, NFSV4CL_LOCKNAMELEN);
1136 NFSBCOPY(ropenownp, nlp->nfsl_openowner,
1137 NFSV4CL_LOCKNAMELEN);
1139 NFSBCOPY(op->nfso_own->nfsow_owner, nlp->nfsl_openowner,
1140 NFSV4CL_LOCKNAMELEN);
1141 nlp->nfsl_seqid = 0;
1142 nlp->nfsl_lockflags = flags;
1143 nlp->nfsl_inprog = NULL;
1144 nfscl_lockinit(&nlp->nfsl_rwlock);
1145 LIST_INIT(&nlp->nfsl_lock);
1147 nlp->nfsl_open = NULL;
1148 nfsstatsv1.cllocallockowners++;
1150 nlp->nfsl_open = op;
1151 nfsstatsv1.cllockowners++;
1153 LIST_INSERT_HEAD(lhp, nlp, nfsl_list);
1160 * Now, update the byte ranges for locks.
1162 ret = nfscl_updatelock(lp, &nlop, &otherlop, donelocally);
1168 nfscl_clrelease(clp);
1171 * Serial modifications on the lock owner for multiple threads
1172 * for the same process using a read/write lock.
1175 nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR);
1181 free(nlp, M_NFSCLLOCKOWNER);
1183 free(nlop, M_NFSCLLOCK);
1185 free(otherlop, M_NFSCLLOCK);
1192 * Called to unlock a byte range, for LockU.
1195 nfscl_relbytelock(vnode_t vp, u_int64_t off, u_int64_t len,
1196 __unused struct ucred *cred, NFSPROC_T *p, int callcnt,
1197 struct nfsclclient *clp, void *id, int flags,
1198 struct nfscllockowner **lpp, int *dorpcp)
1200 struct nfscllockowner *lp;
1201 struct nfsclowner *owp;
1202 struct nfsclopen *op;
1203 struct nfscllock *nlop, *other_lop = NULL;
1204 struct nfscldeleg *dp;
1206 u_int8_t own[NFSV4CL_LOCKNAMELEN];
1214 * Might need these, so MALLOC them now, to
1215 * avoid a tsleep() in MALLOC later.
1218 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1219 nlop->nfslo_type = F_UNLCK;
1220 nlop->nfslo_first = off;
1221 if (len == NFS64BITSSET) {
1222 nlop->nfslo_end = NFS64BITSSET;
1224 nlop->nfslo_end = off + len;
1225 if (nlop->nfslo_end <= nlop->nfslo_first) {
1226 free(nlop, M_NFSCLLOCK);
1227 return (NFSERR_INVAL);
1232 sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1235 nfscl_filllockowner(id, own, flags);
1239 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
1240 np->n_fhp->nfh_len);
1243 * First, unlock any local regions on a delegation.
1246 /* Look for this lockowner. */
1247 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
1248 if (!NFSBCMP(lp->nfsl_owner, own,
1249 NFSV4CL_LOCKNAMELEN))
1253 /* Use other_lop, so nlop is still available */
1254 (void)nfscl_updatelock(lp, &other_lop, NULL, 1);
1258 * Now, find a matching open/lockowner that hasn't already been done,
1259 * as marked by nfsl_inprog.
1263 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
1264 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1265 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1266 !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1267 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1268 if (lp->nfsl_inprog == NULL &&
1269 !NFSBCMP(lp->nfsl_owner, own,
1270 NFSV4CL_LOCKNAMELEN)) {
1284 ret = nfscl_updatelock(lp, &nlop, NULL, 0);
1288 * Serial modifications on the lock owner for multiple
1289 * threads for the same process using a read/write lock.
1291 lp->nfsl_inprog = p;
1292 nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR);
1297 free(nlop, M_NFSCLLOCK);
1299 free(other_lop, M_NFSCLLOCK);
1304 * Release all lockowners marked in progess for this process and file.
1307 nfscl_releasealllocks(struct nfsclclient *clp, vnode_t vp, NFSPROC_T *p,
1308 void *id, int flags)
1310 struct nfsclowner *owp;
1311 struct nfsclopen *op;
1312 struct nfscllockowner *lp;
1314 u_int8_t own[NFSV4CL_LOCKNAMELEN];
1317 nfscl_filllockowner(id, own, flags);
1319 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
1320 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1321 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1322 !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1323 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1324 if (lp->nfsl_inprog == p &&
1325 !NFSBCMP(lp->nfsl_owner, own,
1326 NFSV4CL_LOCKNAMELEN)) {
1327 lp->nfsl_inprog = NULL;
1328 nfscl_lockunlock(&lp->nfsl_rwlock);
1334 nfscl_clrelease(clp);
1339 * Called to find out if any bytes within the byte range specified are
1340 * write locked by the calling process. Used to determine if flushing
1341 * is required before a LockU.
1342 * If in doubt, return 1, so the flush will occur.
1345 nfscl_checkwritelocked(vnode_t vp, struct flock *fl,
1346 struct ucred *cred, NFSPROC_T *p, void *id, int flags)
1348 struct nfsclowner *owp;
1349 struct nfscllockowner *lp;
1350 struct nfsclopen *op;
1351 struct nfsclclient *clp;
1352 struct nfscllock *lop;
1353 struct nfscldeleg *dp;
1356 u_int8_t own[NFSV4CL_LOCKNAMELEN];
1360 switch (fl->l_whence) {
1364 * Caller is responsible for adding any necessary offset
1365 * when SEEK_CUR is used.
1370 off = np->n_size + fl->l_start;
1375 if (fl->l_len != 0) {
1376 end = off + fl->l_len;
1383 error = nfscl_getcl(vp->v_mount, cred, p, 1, &clp);
1386 nfscl_filllockowner(id, own, flags);
1390 * First check the delegation locks.
1392 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
1394 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
1395 if (!NFSBCMP(lp->nfsl_owner, own,
1396 NFSV4CL_LOCKNAMELEN))
1400 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
1401 if (lop->nfslo_first >= end)
1403 if (lop->nfslo_end <= off)
1405 if (lop->nfslo_type == F_WRLCK) {
1406 nfscl_clrelease(clp);
1415 * Now, check state against the server.
1417 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
1418 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1419 if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1420 !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1421 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1422 if (!NFSBCMP(lp->nfsl_owner, own,
1423 NFSV4CL_LOCKNAMELEN))
1427 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
1428 if (lop->nfslo_first >= end)
1430 if (lop->nfslo_end <= off)
1432 if (lop->nfslo_type == F_WRLCK) {
1433 nfscl_clrelease(clp);
1442 nfscl_clrelease(clp);
1448 * Release a byte range lock owner structure.
1451 nfscl_lockrelease(struct nfscllockowner *lp, int error, int candelete)
1453 struct nfsclclient *clp;
1458 clp = lp->nfsl_open->nfso_own->nfsow_clp;
1459 if (error != 0 && candelete &&
1460 (lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED) == 0)
1461 nfscl_freelockowner(lp, 0);
1463 nfscl_lockunlock(&lp->nfsl_rwlock);
1464 nfscl_clrelease(clp);
1469 * Free up an open structure and any associated byte range lock structures.
1472 nfscl_freeopen(struct nfsclopen *op, int local)
1475 LIST_REMOVE(op, nfso_list);
1476 nfscl_freealllocks(&op->nfso_lock, local);
1477 free(op, M_NFSCLOPEN);
1479 nfsstatsv1.cllocalopens--;
1481 nfsstatsv1.clopens--;
1485 * Free up all lock owners and associated locks.
1488 nfscl_freealllocks(struct nfscllockownerhead *lhp, int local)
1490 struct nfscllockowner *lp, *nlp;
1492 LIST_FOREACH_SAFE(lp, lhp, nfsl_list, nlp) {
1493 if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED))
1495 nfscl_freelockowner(lp, local);
1500 * Called for an Open when NFSERR_EXPIRED is received from the server.
1501 * If there are no byte range locks nor a Share Deny lost, try to do a
1502 * fresh Open. Otherwise, free the open.
1505 nfscl_expireopen(struct nfsclclient *clp, struct nfsclopen *op,
1506 struct nfsmount *nmp, struct ucred *cred, NFSPROC_T *p)
1508 struct nfscllockowner *lp;
1509 struct nfscldeleg *dp;
1510 int mustdelete = 0, error;
1513 * Look for any byte range lock(s).
1515 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1516 if (!LIST_EMPTY(&lp->nfsl_lock)) {
1523 * If no byte range lock(s) nor a Share deny, try to re-open.
1525 if (!mustdelete && (op->nfso_mode & NFSLCK_DENYBITS) == 0) {
1526 newnfs_copycred(&op->nfso_cred, cred);
1528 error = nfsrpc_reopen(nmp, op->nfso_fh,
1529 op->nfso_fhlen, op->nfso_mode, op, &dp, cred, p);
1533 free(dp, M_NFSCLDELEG);
1538 nfscl_deleg(nmp->nm_mountp, clp, op->nfso_fh,
1539 op->nfso_fhlen, cred, p, &dp);
1543 * If a byte range lock or Share deny or couldn't re-open, free it.
1546 nfscl_freeopen(op, 0);
1547 return (mustdelete);
1551 * Free up an open owner structure.
1554 nfscl_freeopenowner(struct nfsclowner *owp, int local)
1557 LIST_REMOVE(owp, nfsow_list);
1558 free(owp, M_NFSCLOWNER);
1560 nfsstatsv1.cllocalopenowners--;
1562 nfsstatsv1.clopenowners--;
1566 * Free up a byte range lock owner structure.
1569 nfscl_freelockowner(struct nfscllockowner *lp, int local)
1571 struct nfscllock *lop, *nlop;
1573 LIST_REMOVE(lp, nfsl_list);
1574 LIST_FOREACH_SAFE(lop, &lp->nfsl_lock, nfslo_list, nlop) {
1575 nfscl_freelock(lop, local);
1577 free(lp, M_NFSCLLOCKOWNER);
1579 nfsstatsv1.cllocallockowners--;
1581 nfsstatsv1.cllockowners--;
1585 * Free up a byte range lock structure.
1588 nfscl_freelock(struct nfscllock *lop, int local)
1591 LIST_REMOVE(lop, nfslo_list);
1592 free(lop, M_NFSCLLOCK);
1594 nfsstatsv1.cllocallocks--;
1596 nfsstatsv1.cllocks--;
1600 * Clean out the state related to a delegation.
1603 nfscl_cleandeleg(struct nfscldeleg *dp)
1605 struct nfsclowner *owp, *nowp;
1606 struct nfsclopen *op;
1608 LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) {
1609 op = LIST_FIRST(&owp->nfsow_open);
1611 if (LIST_NEXT(op, nfso_list) != NULL)
1612 panic("nfscleandel");
1613 nfscl_freeopen(op, 1);
1615 nfscl_freeopenowner(owp, 1);
1617 nfscl_freealllocks(&dp->nfsdl_lock, 1);
1621 * Free a delegation.
1624 nfscl_freedeleg(struct nfscldeleghead *hdp, struct nfscldeleg *dp)
1627 TAILQ_REMOVE(hdp, dp, nfsdl_list);
1628 LIST_REMOVE(dp, nfsdl_hash);
1629 free(dp, M_NFSCLDELEG);
1630 nfsstatsv1.cldelegates--;
1635 * Free up all state related to this client structure.
1638 nfscl_cleanclient(struct nfsclclient *clp)
1640 struct nfsclowner *owp, *nowp;
1641 struct nfsclopen *op, *nop;
1642 struct nfscllayout *lyp, *nlyp;
1643 struct nfscldevinfo *dip, *ndip;
1645 TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp)
1646 nfscl_freelayout(lyp);
1648 LIST_FOREACH_SAFE(dip, &clp->nfsc_devinfo, nfsdi_list, ndip)
1649 nfscl_freedevinfo(dip);
1651 /* Now, all the OpenOwners, etc. */
1652 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1653 LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
1654 nfscl_freeopen(op, 0);
1656 nfscl_freeopenowner(owp, 0);
1661 * Called when an NFSERR_EXPIRED is received from the server.
1664 nfscl_expireclient(struct nfsclclient *clp, struct nfsmount *nmp,
1665 struct ucred *cred, NFSPROC_T *p)
1667 struct nfsclowner *owp, *nowp, *towp;
1668 struct nfsclopen *op, *nop, *top;
1669 struct nfscldeleg *dp, *ndp;
1670 int ret, printed = 0;
1673 * First, merge locally issued Opens into the list for the server.
1675 dp = TAILQ_FIRST(&clp->nfsc_deleg);
1676 while (dp != NULL) {
1677 ndp = TAILQ_NEXT(dp, nfsdl_list);
1678 owp = LIST_FIRST(&dp->nfsdl_owner);
1679 while (owp != NULL) {
1680 nowp = LIST_NEXT(owp, nfsow_list);
1681 op = LIST_FIRST(&owp->nfsow_open);
1683 if (LIST_NEXT(op, nfso_list) != NULL)
1685 LIST_FOREACH(towp, &clp->nfsc_owner, nfsow_list) {
1686 if (!NFSBCMP(towp->nfsow_owner, owp->nfsow_owner,
1687 NFSV4CL_LOCKNAMELEN))
1691 /* Merge opens in */
1692 LIST_FOREACH(top, &towp->nfsow_open, nfso_list) {
1693 if (top->nfso_fhlen == op->nfso_fhlen &&
1694 !NFSBCMP(top->nfso_fh, op->nfso_fh,
1696 top->nfso_mode |= op->nfso_mode;
1697 top->nfso_opencnt += op->nfso_opencnt;
1702 /* Just add the open to the owner list */
1703 LIST_REMOVE(op, nfso_list);
1704 op->nfso_own = towp;
1705 LIST_INSERT_HEAD(&towp->nfsow_open, op, nfso_list);
1706 nfsstatsv1.cllocalopens--;
1707 nfsstatsv1.clopens++;
1710 /* Just add the openowner to the client list */
1711 LIST_REMOVE(owp, nfsow_list);
1712 owp->nfsow_clp = clp;
1713 LIST_INSERT_HEAD(&clp->nfsc_owner, owp, nfsow_list);
1714 nfsstatsv1.cllocalopenowners--;
1715 nfsstatsv1.clopenowners++;
1716 nfsstatsv1.cllocalopens--;
1717 nfsstatsv1.clopens++;
1722 if (!printed && !LIST_EMPTY(&dp->nfsdl_lock)) {
1724 printf("nfsv4 expired locks lost\n");
1726 nfscl_cleandeleg(dp);
1727 nfscl_freedeleg(&clp->nfsc_deleg, dp);
1730 if (!TAILQ_EMPTY(&clp->nfsc_deleg))
1734 * Now, try and reopen against the server.
1736 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1737 owp->nfsow_seqid = 0;
1738 LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
1739 ret = nfscl_expireopen(clp, op, nmp, cred, p);
1740 if (ret && !printed) {
1742 printf("nfsv4 expired locks lost\n");
1745 if (LIST_EMPTY(&owp->nfsow_open))
1746 nfscl_freeopenowner(owp, 0);
1751 * This function must be called after the process represented by "own" has
1752 * exited. Must be called with CLSTATE lock held.
1755 nfscl_cleanup_common(struct nfsclclient *clp, u_int8_t *own)
1757 struct nfsclowner *owp, *nowp;
1758 struct nfscllockowner *lp, *nlp;
1759 struct nfscldeleg *dp;
1761 /* First, get rid of local locks on delegations. */
1762 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
1763 LIST_FOREACH_SAFE(lp, &dp->nfsdl_lock, nfsl_list, nlp) {
1764 if (!NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
1765 if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED))
1767 nfscl_freelockowner(lp, 1);
1771 owp = LIST_FIRST(&clp->nfsc_owner);
1772 while (owp != NULL) {
1773 nowp = LIST_NEXT(owp, nfsow_list);
1774 if (!NFSBCMP(owp->nfsow_owner, own,
1775 NFSV4CL_LOCKNAMELEN)) {
1777 * If there are children that haven't closed the
1778 * file descriptors yet, the opens will still be
1779 * here. For that case, let the renew thread clear
1780 * out the OpenOwner later.
1782 if (LIST_EMPTY(&owp->nfsow_open))
1783 nfscl_freeopenowner(owp, 0);
1785 owp->nfsow_defunct = 1;
1792 * Find open/lock owners for processes that have exited.
1795 nfscl_cleanupkext(struct nfsclclient *clp, struct nfscllockownerfhhead *lhp)
1797 struct nfsclowner *owp, *nowp;
1798 struct nfsclopen *op;
1799 struct nfscllockowner *lp, *nlp;
1800 struct nfscldeleg *dp;
1803 * All the pidhash locks must be acquired, since they are sx locks
1804 * and must be acquired before the mutexes. The pid(s) that will
1805 * be used aren't known yet, so all the locks need to be acquired.
1806 * Fortunately, this function is only performed once/sec.
1810 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1811 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1812 LIST_FOREACH_SAFE(lp, &op->nfso_lock, nfsl_list, nlp) {
1813 if (LIST_EMPTY(&lp->nfsl_lock))
1814 nfscl_emptylockowner(lp, lhp);
1817 if (nfscl_procdoesntexist(owp->nfsow_owner))
1818 nfscl_cleanup_common(clp, owp->nfsow_owner);
1822 * For the single open_owner case, these lock owners need to be
1823 * checked to see if they still exist separately.
1824 * This is because nfscl_procdoesntexist() never returns true for
1825 * the single open_owner so that the above doesn't ever call
1826 * nfscl_cleanup_common().
1828 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
1829 LIST_FOREACH_SAFE(lp, &dp->nfsdl_lock, nfsl_list, nlp) {
1830 if (nfscl_procdoesntexist(lp->nfsl_owner))
1831 nfscl_cleanup_common(clp, lp->nfsl_owner);
1835 pidhash_sunlockall();
1839 * Take the empty lock owner and move it to the local lhp list if the
1840 * associated process no longer exists.
1843 nfscl_emptylockowner(struct nfscllockowner *lp,
1844 struct nfscllockownerfhhead *lhp)
1846 struct nfscllockownerfh *lfhp, *mylfhp;
1847 struct nfscllockowner *nlp;
1850 /* If not a Posix lock owner, just return. */
1851 if ((lp->nfsl_lockflags & F_POSIX) == 0)
1857 * First, search to see if this lock owner is already in the list.
1858 * If it is, then the associated process no longer exists.
1860 SLIST_FOREACH(lfhp, lhp, nfslfh_list) {
1861 if (lfhp->nfslfh_len == lp->nfsl_open->nfso_fhlen &&
1862 !NFSBCMP(lfhp->nfslfh_fh, lp->nfsl_open->nfso_fh,
1865 LIST_FOREACH(nlp, &lfhp->nfslfh_lock, nfsl_list)
1866 if (!NFSBCMP(nlp->nfsl_owner, lp->nfsl_owner,
1867 NFSV4CL_LOCKNAMELEN))
1870 /* If not found, check if process still exists. */
1871 if (fnd_it == 0 && nfscl_procdoesntexist(lp->nfsl_owner) == 0)
1874 /* Move the lock owner over to the local list. */
1875 if (mylfhp == NULL) {
1876 mylfhp = malloc(sizeof(struct nfscllockownerfh), M_TEMP,
1880 mylfhp->nfslfh_len = lp->nfsl_open->nfso_fhlen;
1881 NFSBCOPY(lp->nfsl_open->nfso_fh, mylfhp->nfslfh_fh,
1882 mylfhp->nfslfh_len);
1883 LIST_INIT(&mylfhp->nfslfh_lock);
1884 SLIST_INSERT_HEAD(lhp, mylfhp, nfslfh_list);
1886 LIST_REMOVE(lp, nfsl_list);
1887 LIST_INSERT_HEAD(&mylfhp->nfslfh_lock, lp, nfsl_list);
1890 static int fake_global; /* Used to force visibility of MNTK_UNMOUNTF */
1892 * Called from nfs umount to free up the clientid.
1895 nfscl_umount(struct nfsmount *nmp, NFSPROC_T *p)
1897 struct nfsclclient *clp;
1902 * For the case that matters, this is the thread that set
1903 * MNTK_UNMOUNTF, so it will see it set. The code that follows is
1904 * done to ensure that any thread executing nfscl_getcl() after
1905 * this time, will see MNTK_UNMOUNTF set. nfscl_getcl() uses the
1906 * mutex for NFSLOCKCLSTATE(), so it is "m" for the following
1907 * explanation, courtesy of Alan Cox.
1908 * What follows is a snippet from Alan Cox's email at:
1909 * https://docs.FreeBSD.org/cgi/mid.cgi?BANLkTikR3d65zPHo9==08ZfJ2vmqZucEvw
1911 * 1. Set MNTK_UNMOUNTF
1912 * 2. Acquire a standard FreeBSD mutex "m".
1913 * 3. Update some data structures.
1914 * 4. Release mutex "m".
1916 * Then, other threads that acquire "m" after step 4 has occurred will
1917 * see MNTK_UNMOUNTF as set. But, other threads that beat thread X to
1918 * step 2 may or may not see MNTK_UNMOUNTF as set.
1921 if ((nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) {
1929 if ((clp->nfsc_flags & NFSCLFLAGS_INITED) == 0)
1930 panic("nfscl umount");
1933 * First, handshake with the nfscl renew thread, to terminate
1936 clp->nfsc_flags |= NFSCLFLAGS_UMOUNT;
1937 while (clp->nfsc_flags & NFSCLFLAGS_HASTHREAD)
1938 (void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT,
1942 * Now, get the exclusive lock on the client state, so
1943 * that no uses of the state are still in progress.
1946 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
1947 NFSCLSTATEMUTEXPTR, NULL);
1948 } while (!igotlock);
1952 * Free up all the state. It will expire on the server, but
1953 * maybe we should do a SetClientId/SetClientIdConfirm so
1954 * the server throws it away?
1956 LIST_REMOVE(clp, nfsc_list);
1957 nfscl_delegreturnall(clp, p);
1958 cred = newnfs_getcred();
1959 if (NFSHASNFSV4N(nmp)) {
1960 (void)nfsrpc_destroysession(nmp, clp, cred, p);
1961 (void)nfsrpc_destroyclient(nmp, clp, cred, p);
1963 (void)nfsrpc_setclient(nmp, clp, 0, NULL, cred, p);
1964 nfscl_cleanclient(clp);
1967 free(clp, M_NFSCLCLIENT);
1973 * This function is called when a server replies with NFSERR_STALECLIENTID
1974 * NFSERR_STALESTATEID or NFSERR_BADSESSION. It traverses the clientid lists,
1975 * doing Opens and Locks with reclaim. If these fail, it deletes the
1976 * corresponding state.
1979 nfscl_recover(struct nfsclclient *clp, bool *retokp, struct ucred *cred,
1982 struct nfsclowner *owp, *nowp;
1983 struct nfsclopen *op, *nop;
1984 struct nfscllockowner *lp, *nlp;
1985 struct nfscllock *lop, *nlop;
1986 struct nfscldeleg *dp, *ndp, *tdp;
1987 struct nfsmount *nmp;
1988 struct ucred *tcred;
1989 struct nfsclopenhead extra_open;
1990 struct nfscldeleghead extra_deleg;
1993 u_int32_t delegtype = NFSV4OPEN_DELEGATEWRITE, mode;
1994 int i, igotlock = 0, error, trycnt, firstlock;
1995 struct nfscllayout *lyp, *nlyp;
1998 * First, lock the client structure, so everyone else will
1999 * block when trying to use state.
2002 clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG;
2004 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
2005 NFSCLSTATEMUTEXPTR, NULL);
2006 } while (!igotlock);
2009 nmp = clp->nfsc_nmp;
2011 panic("nfscl recover");
2014 * For now, just get rid of all layouts. There may be a need
2015 * to do LayoutCommit Ops with reclaim == true later.
2017 TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp)
2018 nfscl_freelayout(lyp);
2019 TAILQ_INIT(&clp->nfsc_layout);
2020 for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++)
2021 LIST_INIT(&clp->nfsc_layouthash[i]);
2026 error = nfsrpc_setclient(nmp, clp, 1, retokp, cred, p);
2027 } while ((error == NFSERR_STALECLIENTID ||
2028 error == NFSERR_BADSESSION ||
2029 error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
2032 clp->nfsc_flags &= ~(NFSCLFLAGS_RECOVER |
2033 NFSCLFLAGS_RECVRINPROG);
2034 wakeup(&clp->nfsc_flags);
2035 nfsv4_unlock(&clp->nfsc_lock, 0);
2039 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
2040 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2043 * Mark requests already queued on the server, so that they don't
2044 * initiate another recovery cycle. Any requests already in the
2045 * queue that handle state information will have the old stale
2046 * clientid/stateid and will get a NFSERR_STALESTATEID,
2047 * NFSERR_STALECLIENTID or NFSERR_BADSESSION reply from the server.
2048 * This will be translated to NFSERR_STALEDONTRECOVER when
2049 * R_DONTRECOVER is set.
2052 TAILQ_FOREACH(rep, &nfsd_reqq, r_chain) {
2053 if (rep->r_nmp == nmp)
2054 rep->r_flags |= R_DONTRECOVER;
2059 * If nfsrpc_setclient() returns *retokp == true,
2060 * no more recovery is needed.
2066 * Now, mark all delegations "need reclaim".
2068 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list)
2069 dp->nfsdl_flags |= NFSCLDL_NEEDRECLAIM;
2071 TAILQ_INIT(&extra_deleg);
2072 LIST_INIT(&extra_open);
2074 * Now traverse the state lists, doing Open and Lock Reclaims.
2076 tcred = newnfs_getcred();
2077 owp = LIST_FIRST(&clp->nfsc_owner);
2078 while (owp != NULL) {
2079 nowp = LIST_NEXT(owp, nfsow_list);
2080 owp->nfsow_seqid = 0;
2081 op = LIST_FIRST(&owp->nfsow_open);
2082 while (op != NULL) {
2083 nop = LIST_NEXT(op, nfso_list);
2084 if (error != NFSERR_NOGRACE && error != NFSERR_BADSESSION) {
2085 /* Search for a delegation to reclaim with the open */
2086 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2087 if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
2089 if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
2090 mode = NFSV4OPEN_ACCESSWRITE;
2091 delegtype = NFSV4OPEN_DELEGATEWRITE;
2093 mode = NFSV4OPEN_ACCESSREAD;
2094 delegtype = NFSV4OPEN_DELEGATEREAD;
2096 if ((op->nfso_mode & mode) == mode &&
2097 op->nfso_fhlen == dp->nfsdl_fhlen &&
2098 !NFSBCMP(op->nfso_fh, dp->nfsdl_fh, op->nfso_fhlen))
2103 delegtype = NFSV4OPEN_DELEGATENONE;
2104 newnfs_copycred(&op->nfso_cred, tcred);
2105 error = nfscl_tryopen(nmp, NULL, op->nfso_fh,
2106 op->nfso_fhlen, op->nfso_fh, op->nfso_fhlen,
2107 op->nfso_mode, op, NULL, 0, &ndp, 1, delegtype,
2110 /* Handle any replied delegation */
2111 if (ndp != NULL && ((ndp->nfsdl_flags & NFSCLDL_WRITE)
2112 || NFSMNT_RDONLY(nmp->nm_mountp))) {
2113 if ((ndp->nfsdl_flags & NFSCLDL_WRITE))
2114 mode = NFSV4OPEN_ACCESSWRITE;
2116 mode = NFSV4OPEN_ACCESSREAD;
2117 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2118 if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
2120 if ((op->nfso_mode & mode) == mode &&
2121 op->nfso_fhlen == dp->nfsdl_fhlen &&
2122 !NFSBCMP(op->nfso_fh, dp->nfsdl_fh,
2124 dp->nfsdl_stateid = ndp->nfsdl_stateid;
2125 dp->nfsdl_sizelimit = ndp->nfsdl_sizelimit;
2126 dp->nfsdl_ace = ndp->nfsdl_ace;
2127 dp->nfsdl_change = ndp->nfsdl_change;
2128 dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
2129 if ((ndp->nfsdl_flags & NFSCLDL_RECALL))
2130 dp->nfsdl_flags |= NFSCLDL_RECALL;
2131 free(ndp, M_NFSCLDELEG);
2138 TAILQ_INSERT_HEAD(&extra_deleg, ndp, nfsdl_list);
2140 /* and reclaim all byte range locks */
2141 lp = LIST_FIRST(&op->nfso_lock);
2142 while (lp != NULL) {
2143 nlp = LIST_NEXT(lp, nfsl_list);
2146 lop = LIST_FIRST(&lp->nfsl_lock);
2147 while (lop != NULL) {
2148 nlop = LIST_NEXT(lop, nfslo_list);
2149 if (lop->nfslo_end == NFS64BITSSET)
2152 len = lop->nfslo_end - lop->nfslo_first;
2153 error = nfscl_trylock(nmp, NULL,
2154 op->nfso_fh, op->nfso_fhlen, lp,
2155 firstlock, 1, lop->nfslo_first, len,
2156 lop->nfslo_type, tcred, p);
2158 nfscl_freelock(lop, 0);
2163 /* If no locks, but a lockowner, just delete it. */
2164 if (LIST_EMPTY(&lp->nfsl_lock))
2165 nfscl_freelockowner(lp, 0);
2170 if (error != 0 && error != NFSERR_BADSESSION)
2171 nfscl_freeopen(op, 0);
2178 * Now, try and get any delegations not yet reclaimed by cobbling
2179 * to-gether an appropriate open.
2182 dp = TAILQ_FIRST(&clp->nfsc_deleg);
2183 while (dp != NULL) {
2184 ndp = TAILQ_NEXT(dp, nfsdl_list);
2185 if ((dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM)) {
2188 sizeof (struct nfsclowner), M_NFSCLOWNER, M_WAITOK);
2190 * Name must be as long an largest possible
2191 * NFSV4CL_LOCKNAMELEN. 12 for now.
2193 NFSBCOPY("RECLAIMDELEG", nowp->nfsow_owner,
2194 NFSV4CL_LOCKNAMELEN);
2195 LIST_INIT(&nowp->nfsow_open);
2196 nowp->nfsow_clp = clp;
2197 nowp->nfsow_seqid = 0;
2198 nowp->nfsow_defunct = 0;
2199 nfscl_lockinit(&nowp->nfsow_rwlock);
2202 if (error != NFSERR_NOGRACE && error != NFSERR_BADSESSION) {
2203 nop = malloc(sizeof (struct nfsclopen) +
2204 dp->nfsdl_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
2205 nop->nfso_own = nowp;
2206 if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
2207 nop->nfso_mode = NFSV4OPEN_ACCESSWRITE;
2208 delegtype = NFSV4OPEN_DELEGATEWRITE;
2210 nop->nfso_mode = NFSV4OPEN_ACCESSREAD;
2211 delegtype = NFSV4OPEN_DELEGATEREAD;
2213 nop->nfso_opencnt = 0;
2214 nop->nfso_posixlock = 1;
2215 nop->nfso_fhlen = dp->nfsdl_fhlen;
2216 NFSBCOPY(dp->nfsdl_fh, nop->nfso_fh, dp->nfsdl_fhlen);
2217 LIST_INIT(&nop->nfso_lock);
2218 nop->nfso_stateid.seqid = 0;
2219 nop->nfso_stateid.other[0] = 0;
2220 nop->nfso_stateid.other[1] = 0;
2221 nop->nfso_stateid.other[2] = 0;
2222 newnfs_copycred(&dp->nfsdl_cred, tcred);
2223 newnfs_copyincred(tcred, &nop->nfso_cred);
2225 error = nfscl_tryopen(nmp, NULL, nop->nfso_fh,
2226 nop->nfso_fhlen, nop->nfso_fh, nop->nfso_fhlen,
2227 nop->nfso_mode, nop, NULL, 0, &tdp, 1,
2228 delegtype, tcred, p);
2230 if ((tdp->nfsdl_flags & NFSCLDL_WRITE))
2231 mode = NFSV4OPEN_ACCESSWRITE;
2233 mode = NFSV4OPEN_ACCESSREAD;
2234 if ((nop->nfso_mode & mode) == mode &&
2235 nop->nfso_fhlen == tdp->nfsdl_fhlen &&
2236 !NFSBCMP(nop->nfso_fh, tdp->nfsdl_fh,
2238 dp->nfsdl_stateid = tdp->nfsdl_stateid;
2239 dp->nfsdl_sizelimit = tdp->nfsdl_sizelimit;
2240 dp->nfsdl_ace = tdp->nfsdl_ace;
2241 dp->nfsdl_change = tdp->nfsdl_change;
2242 dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
2243 if ((tdp->nfsdl_flags & NFSCLDL_RECALL))
2244 dp->nfsdl_flags |= NFSCLDL_RECALL;
2245 free(tdp, M_NFSCLDELEG);
2247 TAILQ_INSERT_HEAD(&extra_deleg, tdp, nfsdl_list);
2253 free(nop, M_NFSCLOPEN);
2255 * Couldn't reclaim it, so throw the state
2258 nfscl_cleandeleg(dp);
2259 nfscl_freedeleg(&clp->nfsc_deleg, dp);
2261 LIST_INSERT_HEAD(&extra_open, nop, nfso_list);
2268 * Now, get rid of extra Opens and Delegations.
2270 LIST_FOREACH_SAFE(op, &extra_open, nfso_list, nop) {
2272 newnfs_copycred(&op->nfso_cred, tcred);
2273 error = nfscl_tryclose(op, tcred, nmp, p);
2274 if (error == NFSERR_GRACE)
2275 (void) nfs_catnap(PZERO, error, "nfsexcls");
2276 } while (error == NFSERR_GRACE);
2277 LIST_REMOVE(op, nfso_list);
2278 free(op, M_NFSCLOPEN);
2281 free(nowp, M_NFSCLOWNER);
2283 TAILQ_FOREACH_SAFE(dp, &extra_deleg, nfsdl_list, ndp) {
2285 newnfs_copycred(&dp->nfsdl_cred, tcred);
2286 error = nfscl_trydelegreturn(dp, tcred, nmp, p);
2287 if (error == NFSERR_GRACE)
2288 (void) nfs_catnap(PZERO, error, "nfsexdlg");
2289 } while (error == NFSERR_GRACE);
2290 TAILQ_REMOVE(&extra_deleg, dp, nfsdl_list);
2291 free(dp, M_NFSCLDELEG);
2294 /* For NFSv4.1 or later, do a RECLAIM_COMPLETE. */
2295 if (NFSHASNFSV4N(nmp))
2296 (void)nfsrpc_reclaimcomplete(nmp, cred, p);
2300 clp->nfsc_flags &= ~NFSCLFLAGS_RECVRINPROG;
2301 wakeup(&clp->nfsc_flags);
2302 nfsv4_unlock(&clp->nfsc_lock, 0);
2309 * This function is called when a server replies with NFSERR_EXPIRED.
2310 * It deletes all state for the client and does a fresh SetClientId/confirm.
2311 * XXX Someday it should post a signal to the process(es) that hold the
2312 * state, so they know that lock state has been lost.
2315 nfscl_hasexpired(struct nfsclclient *clp, u_int32_t clidrev, NFSPROC_T *p)
2317 struct nfsmount *nmp;
2319 int igotlock = 0, error, trycnt;
2322 * If the clientid has gone away or a new SetClientid has already
2323 * been done, just return ok.
2325 if (clp == NULL || clidrev != clp->nfsc_clientidrev)
2329 * First, lock the client structure, so everyone else will
2330 * block when trying to use state. Also, use NFSCLFLAGS_EXPIREIT so
2331 * that only one thread does the work.
2334 clp->nfsc_flags |= NFSCLFLAGS_EXPIREIT;
2336 igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
2337 NFSCLSTATEMUTEXPTR, NULL);
2338 } while (!igotlock && (clp->nfsc_flags & NFSCLFLAGS_EXPIREIT));
2339 if ((clp->nfsc_flags & NFSCLFLAGS_EXPIREIT) == 0) {
2341 nfsv4_unlock(&clp->nfsc_lock, 0);
2345 clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG;
2348 nmp = clp->nfsc_nmp;
2350 panic("nfscl expired");
2351 cred = newnfs_getcred();
2354 error = nfsrpc_setclient(nmp, clp, 0, NULL, cred, p);
2355 } while ((error == NFSERR_STALECLIENTID ||
2356 error == NFSERR_BADSESSION ||
2357 error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
2360 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2363 * Expire the state for the client.
2365 nfscl_expireclient(clp, nmp, cred, p);
2367 clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
2368 clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2370 clp->nfsc_flags &= ~(NFSCLFLAGS_EXPIREIT | NFSCLFLAGS_RECVRINPROG);
2371 wakeup(&clp->nfsc_flags);
2372 nfsv4_unlock(&clp->nfsc_lock, 0);
2379 * This function inserts a lock in the list after insert_lop.
2382 nfscl_insertlock(struct nfscllockowner *lp, struct nfscllock *new_lop,
2383 struct nfscllock *insert_lop, int local)
2386 if ((struct nfscllockowner *)insert_lop == lp)
2387 LIST_INSERT_HEAD(&lp->nfsl_lock, new_lop, nfslo_list);
2389 LIST_INSERT_AFTER(insert_lop, new_lop, nfslo_list);
2391 nfsstatsv1.cllocallocks++;
2393 nfsstatsv1.cllocks++;
2397 * This function updates the locking for a lock owner and given file. It
2398 * maintains a list of lock ranges ordered on increasing file offset that
2399 * are NFSCLLOCK_READ or NFSCLLOCK_WRITE and non-overlapping (aka POSIX style).
2400 * It always adds new_lop to the list and sometimes uses the one pointed
2402 * Returns 1 if the locks were modified, 0 otherwise.
2405 nfscl_updatelock(struct nfscllockowner *lp, struct nfscllock **new_lopp,
2406 struct nfscllock **other_lopp, int local)
2408 struct nfscllock *new_lop = *new_lopp;
2409 struct nfscllock *lop, *tlop, *ilop;
2410 struct nfscllock *other_lop;
2411 int unlock = 0, modified = 0;
2415 * Work down the list until the lock is merged.
2417 if (new_lop->nfslo_type == F_UNLCK)
2419 ilop = (struct nfscllock *)lp;
2420 lop = LIST_FIRST(&lp->nfsl_lock);
2421 while (lop != NULL) {
2423 * Only check locks for this file that aren't before the start of
2426 if (lop->nfslo_end >= new_lop->nfslo_first) {
2427 if (new_lop->nfslo_end < lop->nfslo_first) {
2429 * If the new lock ends before the start of the
2430 * current lock's range, no merge, just insert
2435 if (new_lop->nfslo_type == lop->nfslo_type ||
2436 (new_lop->nfslo_first <= lop->nfslo_first &&
2437 new_lop->nfslo_end >= lop->nfslo_end)) {
2439 * This lock can be absorbed by the new lock/unlock.
2440 * This happens when it covers the entire range
2441 * of the old lock or is contiguous
2442 * with the old lock and is of the same type or an
2445 if (new_lop->nfslo_type != lop->nfslo_type ||
2446 new_lop->nfslo_first != lop->nfslo_first ||
2447 new_lop->nfslo_end != lop->nfslo_end)
2449 if (lop->nfslo_first < new_lop->nfslo_first)
2450 new_lop->nfslo_first = lop->nfslo_first;
2451 if (lop->nfslo_end > new_lop->nfslo_end)
2452 new_lop->nfslo_end = lop->nfslo_end;
2454 lop = LIST_NEXT(lop, nfslo_list);
2455 nfscl_freelock(tlop, local);
2460 * All these cases are for contiguous locks that are not the
2461 * same type, so they can't be merged.
2463 if (new_lop->nfslo_first <= lop->nfslo_first) {
2465 * This case is where the new lock overlaps with the
2466 * first part of the old lock. Move the start of the
2467 * old lock to just past the end of the new lock. The
2468 * new lock will be inserted in front of the old, since
2469 * ilop hasn't been updated. (We are done now.)
2471 if (lop->nfslo_first != new_lop->nfslo_end) {
2472 lop->nfslo_first = new_lop->nfslo_end;
2477 if (new_lop->nfslo_end >= lop->nfslo_end) {
2479 * This case is where the new lock overlaps with the
2480 * end of the old lock's range. Move the old lock's
2481 * end to just before the new lock's first and insert
2482 * the new lock after the old lock.
2483 * Might not be done yet, since the new lock could
2484 * overlap further locks with higher ranges.
2486 if (lop->nfslo_end != new_lop->nfslo_first) {
2487 lop->nfslo_end = new_lop->nfslo_first;
2491 lop = LIST_NEXT(lop, nfslo_list);
2495 * The final case is where the new lock's range is in the
2496 * middle of the current lock's and splits the current lock
2497 * up. Use *other_lopp to handle the second part of the
2498 * split old lock range. (We are done now.)
2499 * For unlock, we use new_lop as other_lop and tmp, since
2500 * other_lop and new_lop are the same for this case.
2501 * We noted the unlock case above, so we don't need
2502 * new_lop->nfslo_type any longer.
2504 tmp = new_lop->nfslo_first;
2506 other_lop = new_lop;
2509 other_lop = *other_lopp;
2512 other_lop->nfslo_first = new_lop->nfslo_end;
2513 other_lop->nfslo_end = lop->nfslo_end;
2514 other_lop->nfslo_type = lop->nfslo_type;
2515 lop->nfslo_end = tmp;
2516 nfscl_insertlock(lp, other_lop, lop, local);
2522 lop = LIST_NEXT(lop, nfslo_list);
2528 * Insert the new lock in the list at the appropriate place.
2531 nfscl_insertlock(lp, new_lop, ilop, local);
2539 * This function must be run as a kernel thread.
2540 * It does Renew Ops and recovery, when required.
2543 nfscl_renewthread(struct nfsclclient *clp, NFSPROC_T *p)
2545 struct nfsclowner *owp, *nowp;
2546 struct nfsclopen *op;
2547 struct nfscllockowner *lp, *nlp;
2548 struct nfscldeleghead dh;
2549 struct nfscldeleg *dp, *ndp;
2552 int error, cbpathdown, islept, igotlock, ret, clearok;
2553 uint32_t recover_done_time = 0;
2555 static time_t prevsec = 0;
2556 struct nfscllockownerfh *lfhp, *nlfhp;
2557 struct nfscllockownerfhhead lfh;
2558 struct nfscllayout *lyp, *nlyp;
2559 struct nfscldevinfo *dip, *ndip;
2560 struct nfscllayouthead rlh;
2561 struct nfsclrecalllayout *recallp;
2562 struct nfsclds *dsp;
2566 cred = newnfs_getcred();
2568 clp->nfsc_flags |= NFSCLFLAGS_HASTHREAD;
2569 mp = clp->nfsc_nmp->nm_mountp;
2572 newnfs_setroot(cred);
2574 if (clp->nfsc_flags & NFSCLFLAGS_RECOVER) {
2576 * Only allow one full recover within 1/2 of the lease
2577 * duration (nfsc_renew).
2578 * retok is value/result. If passed in set to true,
2579 * it indicates only a CreateSession operation should
2581 * If it is returned true, it indicates that the
2582 * recovery only required a CreateSession.
2585 if (recover_done_time < NFSD_MONOSEC) {
2586 recover_done_time = NFSD_MONOSEC +
2590 NFSCL_DEBUG(1, "Doing recovery, only "
2591 "createsession=%d\n", retok);
2592 nfscl_recover(clp, &retok, cred, p);
2594 if (clp->nfsc_expire <= NFSD_MONOSEC &&
2595 (clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID)) {
2596 clp->nfsc_expire = NFSD_MONOSEC + clp->nfsc_renew;
2597 clidrev = clp->nfsc_clientidrev;
2598 error = nfsrpc_renew(clp, NULL, cred, p);
2599 if (error == NFSERR_CBPATHDOWN)
2601 else if (error == NFSERR_STALECLIENTID ||
2602 error == NFSERR_BADSESSION) {
2604 clp->nfsc_flags |= NFSCLFLAGS_RECOVER;
2606 } else if (error == NFSERR_EXPIRED)
2607 (void) nfscl_hasexpired(clp, clidrev, p);
2611 if (NFSHASNFSV4N(clp->nfsc_nmp)) {
2612 /* Do renews for any DS sessions. */
2613 NFSLOCKMNT(clp->nfsc_nmp);
2614 /* Skip first entry, since the MDS is handled above. */
2615 dsp = TAILQ_FIRST(&clp->nfsc_nmp->nm_sess);
2617 dsp = TAILQ_NEXT(dsp, nfsclds_list);
2618 while (dsp != NULL) {
2619 if (dsp->nfsclds_expire <= NFSD_MONOSEC &&
2620 dsp->nfsclds_sess.nfsess_defunct == 0) {
2621 dsp->nfsclds_expire = NFSD_MONOSEC +
2623 NFSUNLOCKMNT(clp->nfsc_nmp);
2624 (void)nfsrpc_renew(clp, dsp, cred, p);
2627 dsp = TAILQ_NEXT(dsp, nfsclds_list);
2629 NFSUNLOCKMNT(clp->nfsc_nmp);
2635 /* It's a Total Recall! */
2636 nfscl_totalrecall(clp);
2639 * Now, handle defunct owners.
2641 LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
2642 if (LIST_EMPTY(&owp->nfsow_open)) {
2643 if (owp->nfsow_defunct != 0)
2644 nfscl_freeopenowner(owp, 0);
2649 * Do the recall on any delegations. To avoid trouble, always
2650 * come back up here after having slept.
2654 dp = TAILQ_FIRST(&clp->nfsc_deleg);
2655 while (dp != NULL) {
2656 ndp = TAILQ_NEXT(dp, nfsdl_list);
2657 if ((dp->nfsdl_flags & NFSCLDL_RECALL)) {
2659 * Wait for outstanding I/O ops to be done.
2661 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
2663 nfsv4_unlock(&clp->nfsc_lock, 0);
2666 dp->nfsdl_rwlock.nfslock_lock |=
2668 msleep(&dp->nfsdl_rwlock,
2669 NFSCLSTATEMUTEXPTR, PVFS, "nfscld",
2671 if (NFSCL_FORCEDISM(mp))
2676 igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
2677 &islept, NFSCLSTATEMUTEXPTR, mp);
2678 if (igotlock == 0 && NFSCL_FORCEDISM(mp))
2684 newnfs_copycred(&dp->nfsdl_cred, cred);
2685 ret = nfscl_recalldeleg(clp, clp->nfsc_nmp, dp,
2688 nfscl_cleandeleg(dp);
2689 TAILQ_REMOVE(&clp->nfsc_deleg, dp,
2691 LIST_REMOVE(dp, nfsdl_hash);
2692 TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
2694 nfsstatsv1.cldelegates--;
2702 * Clear out old delegations, if we are above the high water
2703 * mark. Only clear out ones with no state related to them.
2704 * The tailq list is in LRU order.
2706 dp = TAILQ_LAST(&clp->nfsc_deleg, nfscldeleghead);
2707 while (nfscl_delegcnt > nfscl_deleghighwater && dp != NULL) {
2708 ndp = TAILQ_PREV(dp, nfscldeleghead, nfsdl_list);
2709 if (dp->nfsdl_rwlock.nfslock_usecnt == 0 &&
2710 dp->nfsdl_rwlock.nfslock_lock == 0 &&
2711 dp->nfsdl_timestamp < NFSD_MONOSEC &&
2712 (dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_ZAPPED |
2713 NFSCLDL_NEEDRECLAIM | NFSCLDL_DELEGRET)) == 0) {
2715 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
2716 op = LIST_FIRST(&owp->nfsow_open);
2723 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
2724 if (!LIST_EMPTY(&lp->nfsl_lock)) {
2731 TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
2732 LIST_REMOVE(dp, nfsdl_hash);
2733 TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
2735 nfsstatsv1.cldelegates--;
2741 nfsv4_unlock(&clp->nfsc_lock, 0);
2744 * Do the recall on any layouts. To avoid trouble, always
2745 * come back up here after having slept.
2749 TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp) {
2750 if ((lyp->nfsly_flags & NFSLY_RECALL) != 0) {
2752 * Wait for outstanding I/O ops to be done.
2754 if (lyp->nfsly_lock.nfslock_usecnt > 0 ||
2755 (lyp->nfsly_lock.nfslock_lock &
2756 NFSV4LOCK_LOCK) != 0) {
2757 lyp->nfsly_lock.nfslock_lock |=
2759 msleep(&lyp->nfsly_lock.nfslock_lock,
2760 NFSCLSTATEMUTEXPTR, PVFS, "nfslyp",
2762 if (NFSCL_FORCEDISM(mp))
2766 /* Move the layout to the recall list. */
2767 TAILQ_REMOVE(&clp->nfsc_layout, lyp,
2769 LIST_REMOVE(lyp, nfsly_hash);
2770 TAILQ_INSERT_HEAD(&rlh, lyp, nfsly_list);
2772 /* Handle any layout commits. */
2773 if (!NFSHASNOLAYOUTCOMMIT(clp->nfsc_nmp) &&
2774 (lyp->nfsly_flags & NFSLY_WRITTEN) != 0) {
2775 lyp->nfsly_flags &= ~NFSLY_WRITTEN;
2777 NFSCL_DEBUG(3, "do layoutcommit\n");
2778 nfscl_dolayoutcommit(clp->nfsc_nmp, lyp,
2786 /* Now, look for stale layouts. */
2787 lyp = TAILQ_LAST(&clp->nfsc_layout, nfscllayouthead);
2788 while (lyp != NULL) {
2789 nlyp = TAILQ_PREV(lyp, nfscllayouthead, nfsly_list);
2790 if (lyp->nfsly_timestamp < NFSD_MONOSEC &&
2791 (lyp->nfsly_flags & NFSLY_RECALL) == 0 &&
2792 lyp->nfsly_lock.nfslock_usecnt == 0 &&
2793 lyp->nfsly_lock.nfslock_lock == 0) {
2794 NFSCL_DEBUG(4, "ret stale lay=%d\n",
2796 recallp = malloc(sizeof(*recallp),
2797 M_NFSLAYRECALL, M_NOWAIT);
2798 if (recallp == NULL)
2800 (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE,
2801 lyp, NFSLAYOUTIOMODE_ANY, 0, UINT64_MAX,
2802 lyp->nfsly_stateid.seqid, 0, 0, NULL,
2809 * Free up any unreferenced device info structures.
2811 LIST_FOREACH_SAFE(dip, &clp->nfsc_devinfo, nfsdi_list, ndip) {
2812 if (dip->nfsdi_layoutrefs == 0 &&
2813 dip->nfsdi_refcnt == 0) {
2814 NFSCL_DEBUG(4, "freeing devinfo\n");
2815 LIST_REMOVE(dip, nfsdi_list);
2816 nfscl_freedevinfo(dip);
2821 /* Do layout return(s), as required. */
2822 TAILQ_FOREACH_SAFE(lyp, &rlh, nfsly_list, nlyp) {
2823 TAILQ_REMOVE(&rlh, lyp, nfsly_list);
2824 NFSCL_DEBUG(4, "ret layout\n");
2825 nfscl_layoutreturn(clp->nfsc_nmp, lyp, cred, p);
2826 nfscl_freelayout(lyp);
2830 * Delegreturn any delegations cleaned out or recalled.
2832 TAILQ_FOREACH_SAFE(dp, &dh, nfsdl_list, ndp) {
2833 newnfs_copycred(&dp->nfsdl_cred, cred);
2834 (void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
2835 TAILQ_REMOVE(&dh, dp, nfsdl_list);
2836 free(dp, M_NFSCLDELEG);
2841 * Call nfscl_cleanupkext() once per second to check for
2842 * open/lock owners where the process has exited.
2844 mytime = NFSD_MONOSEC;
2845 if (prevsec != mytime) {
2847 nfscl_cleanupkext(clp, &lfh);
2851 * Do a ReleaseLockOwner for all lock owners where the
2852 * associated process no longer exists, as found by
2853 * nfscl_cleanupkext().
2855 newnfs_setroot(cred);
2856 SLIST_FOREACH_SAFE(lfhp, &lfh, nfslfh_list, nlfhp) {
2857 LIST_FOREACH_SAFE(lp, &lfhp->nfslfh_lock, nfsl_list,
2859 (void)nfsrpc_rellockown(clp->nfsc_nmp, lp,
2860 lfhp->nfslfh_fh, lfhp->nfslfh_len, cred,
2862 nfscl_freelockowner(lp, 0);
2869 if ((clp->nfsc_flags & NFSCLFLAGS_RECOVER) == 0)
2870 (void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT, "nfscl",
2873 if (clp->nfsc_flags & NFSCLFLAGS_UMOUNT) {
2874 clp->nfsc_flags &= ~NFSCLFLAGS_HASTHREAD;
2877 wakeup((caddr_t)clp);
2885 * Initiate state recovery. Called when NFSERR_STALECLIENTID,
2886 * NFSERR_STALESTATEID or NFSERR_BADSESSION is received.
2889 nfscl_initiate_recovery(struct nfsclclient *clp)
2895 clp->nfsc_flags |= NFSCLFLAGS_RECOVER;
2897 wakeup((caddr_t)clp);
2901 * Dump out the state stuff for debugging.
2904 nfscl_dumpstate(struct nfsmount *nmp, int openowner, int opens,
2905 int lockowner, int locks)
2907 struct nfsclclient *clp;
2908 struct nfsclowner *owp;
2909 struct nfsclopen *op;
2910 struct nfscllockowner *lp;
2911 struct nfscllock *lop;
2912 struct nfscldeleg *dp;
2916 printf("nfscl dumpstate NULL clp\n");
2920 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2921 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
2922 if (openowner && !LIST_EMPTY(&owp->nfsow_open))
2923 printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
2924 owp->nfsow_owner[0], owp->nfsow_owner[1],
2925 owp->nfsow_owner[2], owp->nfsow_owner[3],
2927 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
2929 printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
2930 op->nfso_stateid.other[0], op->nfso_stateid.other[1],
2931 op->nfso_stateid.other[2], op->nfso_opencnt,
2933 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
2935 printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
2936 lp->nfsl_owner[0], lp->nfsl_owner[1],
2937 lp->nfsl_owner[2], lp->nfsl_owner[3],
2939 lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
2940 lp->nfsl_stateid.other[2]);
2941 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
2944 printf("lck typ=%d fst=%ju end=%ju\n",
2945 lop->nfslo_type, (intmax_t)lop->nfslo_first,
2946 (intmax_t)lop->nfslo_end);
2948 printf("lck typ=%d fst=%qd end=%qd\n",
2949 lop->nfslo_type, lop->nfslo_first,
2957 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
2958 if (openowner && !LIST_EMPTY(&owp->nfsow_open))
2959 printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
2960 owp->nfsow_owner[0], owp->nfsow_owner[1],
2961 owp->nfsow_owner[2], owp->nfsow_owner[3],
2963 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
2965 printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
2966 op->nfso_stateid.other[0], op->nfso_stateid.other[1],
2967 op->nfso_stateid.other[2], op->nfso_opencnt,
2969 LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
2971 printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
2972 lp->nfsl_owner[0], lp->nfsl_owner[1],
2973 lp->nfsl_owner[2], lp->nfsl_owner[3],
2975 lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
2976 lp->nfsl_stateid.other[2]);
2977 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
2980 printf("lck typ=%d fst=%ju end=%ju\n",
2981 lop->nfslo_type, (intmax_t)lop->nfslo_first,
2982 (intmax_t)lop->nfslo_end);
2984 printf("lck typ=%d fst=%qd end=%qd\n",
2985 lop->nfslo_type, lop->nfslo_first,
2996 * Check for duplicate open owners and opens.
2997 * (Only used as a diagnostic aid.)
3000 nfscl_dupopen(vnode_t vp, int dupopens)
3002 struct nfsclclient *clp;
3003 struct nfsclowner *owp, *owp2;
3004 struct nfsclopen *op, *op2;
3007 clp = VFSTONFS(vp->v_mount)->nm_clp;
3009 printf("nfscl dupopen NULL clp\n");
3012 nfhp = VTONFS(vp)->n_fhp;
3016 * First, search for duplicate owners.
3017 * These should never happen!
3019 LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3020 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3022 !NFSBCMP(owp->nfsow_owner, owp2->nfsow_owner,
3023 NFSV4CL_LOCKNAMELEN)) {
3025 printf("DUP OWNER\n");
3026 nfscl_dumpstate(VFSTONFS(vp->v_mount), 1, 1, 0, 0);
3033 * Now, search for duplicate stateids.
3034 * These shouldn't happen, either.
3036 LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3037 LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
3038 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3039 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3041 (op->nfso_stateid.other[0] != 0 ||
3042 op->nfso_stateid.other[1] != 0 ||
3043 op->nfso_stateid.other[2] != 0) &&
3044 op->nfso_stateid.other[0] == op2->nfso_stateid.other[0] &&
3045 op->nfso_stateid.other[1] == op2->nfso_stateid.other[1] &&
3046 op->nfso_stateid.other[2] == op2->nfso_stateid.other[2]) {
3048 printf("DUP STATEID\n");
3049 nfscl_dumpstate(VFSTONFS(vp->v_mount), 1, 1, 0, 0);
3058 * Now search for duplicate opens.
3059 * Duplicate opens for the same owner
3060 * should never occur. Other duplicates are
3061 * possible and are checked for if "dupopens"
3064 LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3065 LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
3066 if (nfhp->nfh_len == op2->nfso_fhlen &&
3067 !NFSBCMP(nfhp->nfh_fh, op2->nfso_fh, nfhp->nfh_len)) {
3068 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3069 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3070 if (op != op2 && nfhp->nfh_len == op->nfso_fhlen &&
3071 !NFSBCMP(nfhp->nfh_fh, op->nfso_fh, nfhp->nfh_len) &&
3072 (!NFSBCMP(op->nfso_own->nfsow_owner,
3073 op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN) ||
3075 if (!NFSBCMP(op->nfso_own->nfsow_owner,
3076 op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
3078 printf("BADDUP OPEN\n");
3081 printf("DUP OPEN\n");
3083 nfscl_dumpstate(VFSTONFS(vp->v_mount), 1, 1, 0,
3096 * During close, find an open that needs to be dereferenced and
3097 * dereference it. If there are no more opens for this file,
3098 * log a message to that effect.
3099 * Opens aren't actually Close'd until VOP_INACTIVE() is performed
3100 * on the file's vnode.
3101 * This is the safe way, since it is difficult to identify
3102 * which open the close is for and I/O can be performed after the
3103 * close(2) system call when a file is mmap'd.
3104 * If it returns 0 for success, there will be a referenced
3105 * clp returned via clpp.
3108 nfscl_getclose(vnode_t vp, struct nfsclclient **clpp)
3110 struct nfsclclient *clp;
3111 struct nfsclowner *owp;
3112 struct nfsclopen *op;
3113 struct nfscldeleg *dp;
3117 error = nfscl_getcl(vp->v_mount, NULL, NULL, 1, &clp);
3122 nfhp = VTONFS(vp)->n_fhp;
3126 * First, look for one under a delegation that was locally issued
3127 * and just decrement the opencnt for it. Since all my Opens against
3128 * the server are DENY_NONE, I don't see a problem with hanging
3129 * onto them. (It is much easier to use one of the extant Opens
3130 * that I already have on the server when a Delegation is recalled
3131 * than to do fresh Opens.) Someday, I might need to rethink this, but.
3133 dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
3135 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
3136 op = LIST_FIRST(&owp->nfsow_open);
3139 * Since a delegation is for a file, there
3140 * should never be more than one open for
3143 if (LIST_NEXT(op, nfso_list) != NULL)
3144 panic("nfscdeleg opens");
3145 if (notdecr && op->nfso_opencnt > 0) {
3154 /* Now process the opens against the server. */
3155 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3156 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3157 if (op->nfso_fhlen == nfhp->nfh_len &&
3158 !NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
3160 /* Found an open, decrement cnt if possible */
3161 if (notdecr && op->nfso_opencnt > 0) {
3166 * There are more opens, so just return.
3168 if (op->nfso_opencnt > 0) {
3177 printf("nfscl: never fnd open\n");
3182 nfscl_doclose(vnode_t vp, struct nfsclclient **clpp, NFSPROC_T *p)
3184 struct nfsclclient *clp;
3185 struct nfsclowner *owp, *nowp;
3186 struct nfsclopen *op;
3187 struct nfscldeleg *dp;
3189 struct nfsclrecalllayout *recallp;
3192 error = nfscl_getcl(vp->v_mount, NULL, NULL, 1, &clp);
3197 nfhp = VTONFS(vp)->n_fhp;
3198 recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_WAITOK);
3201 * First get rid of the local Open structures, which should be no
3204 dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
3206 LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) {
3207 op = LIST_FIRST(&owp->nfsow_open);
3209 KASSERT((op->nfso_opencnt == 0),
3210 ("nfscl: bad open cnt on deleg"));
3211 nfscl_freeopen(op, 1);
3213 nfscl_freeopenowner(owp, 1);
3217 /* Return any layouts marked return on close. */
3218 nfscl_retoncloselayout(vp, clp, nfhp->nfh_fh, nfhp->nfh_len, &recallp);
3220 /* Now process the opens against the server. */
3222 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3223 op = LIST_FIRST(&owp->nfsow_open);
3224 while (op != NULL) {
3225 if (op->nfso_fhlen == nfhp->nfh_len &&
3226 !NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
3228 /* Found an open, close it. */
3230 KASSERT((op->nfso_opencnt == 0),
3231 ("nfscl: bad open cnt on server (%d)",
3235 nfsrpc_doclose(VFSTONFS(vp->v_mount), op, p);
3239 op = LIST_NEXT(op, nfso_list);
3244 * recallp has been set NULL by nfscl_retoncloselayout() if it was
3245 * used by the function, but calling free() with a NULL pointer is ok.
3247 free(recallp, M_NFSLAYRECALL);
3252 * Return all delegations on this client.
3253 * (Must be called with client sleep lock.)
3256 nfscl_delegreturnall(struct nfsclclient *clp, NFSPROC_T *p)
3258 struct nfscldeleg *dp, *ndp;
3261 cred = newnfs_getcred();
3262 TAILQ_FOREACH_SAFE(dp, &clp->nfsc_deleg, nfsdl_list, ndp) {
3263 nfscl_cleandeleg(dp);
3264 (void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
3265 nfscl_freedeleg(&clp->nfsc_deleg, dp);
3271 * Do a callback RPC.
3274 nfscl_docb(struct nfsrv_descript *nd, NFSPROC_T *p)
3276 int clist, gotseq_ok, i, j, k, op, rcalls;
3278 struct nfsclclient *clp;
3279 struct nfscldeleg *dp = NULL;
3280 int numops, taglen = -1, error = 0, trunc __unused;
3281 u_int32_t minorvers = 0, retops = 0, *retopsp = NULL, *repp, cbident;
3282 u_char tag[NFSV4_SMALLSTR + 1], *tagstr;
3288 nfsattrbit_t attrbits, rattrbits;
3289 nfsv4stateid_t stateid;
3290 uint32_t seqid, slotid = 0, highslot, cachethis __unused;
3291 uint8_t sessionid[NFSX_V4SESSIONID];
3293 struct nfscllayout *lyp;
3294 uint64_t filesid[2], len, off;
3295 int changed, gotone, laytype, recalltype;
3297 struct nfsclrecalllayout *recallp = NULL;
3298 struct nfsclsession *tsep;
3302 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3303 taglen = fxdr_unsigned(int, *tl);
3308 if (taglen <= NFSV4_SMALLSTR)
3311 tagstr = malloc(taglen + 1, M_TEMP, M_WAITOK);
3312 error = nfsrv_mtostr(nd, tagstr, taglen);
3314 if (taglen > NFSV4_SMALLSTR)
3315 free(tagstr, M_TEMP);
3319 (void) nfsm_strtom(nd, tag, taglen);
3320 if (taglen > NFSV4_SMALLSTR) {
3321 free(tagstr, M_TEMP);
3323 NFSM_BUILD(retopsp, u_int32_t *, NFSX_UNSIGNED);
3324 NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3325 minorvers = fxdr_unsigned(u_int32_t, *tl++);
3326 if (minorvers != NFSV4_MINORVERSION &&
3327 minorvers != NFSV41_MINORVERSION &&
3328 minorvers != NFSV42_MINORVERSION)
3329 nd->nd_repstat = NFSERR_MINORVERMISMATCH;
3330 cbident = fxdr_unsigned(u_int32_t, *tl++);
3334 numops = fxdr_unsigned(int, *tl);
3336 * Loop around doing the sub ops.
3338 for (i = 0; i < numops; i++) {
3339 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3340 NFSM_BUILD(repp, u_int32_t *, 2 * NFSX_UNSIGNED);
3342 op = fxdr_unsigned(int, *tl);
3343 if (op < NFSV4OP_CBGETATTR ||
3344 (op > NFSV4OP_CBRECALL && minorvers == NFSV4_MINORVERSION) ||
3345 (op > NFSV4OP_CBNOTIFYDEVID &&
3346 minorvers == NFSV41_MINORVERSION) ||
3347 (op > NFSV4OP_CBOFFLOAD &&
3348 minorvers == NFSV42_MINORVERSION)) {
3349 nd->nd_repstat = NFSERR_OPILLEGAL;
3350 *repp = nfscl_errmap(nd, minorvers);
3354 nd->nd_procnum = op;
3355 if (op < NFSV42_CBNOPS)
3356 nfsstatsv1.cbrpccnt[nd->nd_procnum]++;
3358 case NFSV4OP_CBGETATTR:
3359 NFSCL_DEBUG(4, "cbgetattr\n");
3362 error = nfsm_getfh(nd, &nfhp);
3364 error = nfsrv_getattrbits(nd, &attrbits,
3366 if (error == 0 && i == 0 &&
3367 minorvers != NFSV4_MINORVERSION)
3368 error = NFSERR_OPNOTINSESS;
3370 mp = nfscl_getmnt(minorvers, sessionid, cbident,
3373 error = NFSERR_SERVERFAULT;
3376 error = nfscl_ngetreopen(mp, nfhp->nfh_fh,
3377 nfhp->nfh_len, p, &np);
3382 NFSZERO_ATTRBIT(&rattrbits);
3384 dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
3387 if (NFSISSET_ATTRBIT(&attrbits,
3390 va.va_size = np->n_size;
3394 NFSSETBIT_ATTRBIT(&rattrbits,
3397 if (NFSISSET_ATTRBIT(&attrbits,
3398 NFSATTRBIT_CHANGE)) {
3402 (np->n_flag & NDELEGMOD))
3404 NFSSETBIT_ATTRBIT(&rattrbits,
3408 error = NFSERR_SERVERFAULT;
3416 free(nfhp, M_NFSFH);
3418 (void) nfsv4_fillattr(nd, NULL, NULL, NULL, &va,
3419 NULL, 0, &rattrbits, NULL, p, 0, 0, 0, 0,
3422 case NFSV4OP_CBRECALL:
3423 NFSCL_DEBUG(4, "cbrecall\n");
3424 NFSM_DISSECT(tl, u_int32_t *, NFSX_STATEID +
3426 stateid.seqid = *tl++;
3427 NFSBCOPY((caddr_t)tl, (caddr_t)stateid.other,
3429 tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
3430 trunc = fxdr_unsigned(int, *tl);
3431 error = nfsm_getfh(nd, &nfhp);
3432 if (error == 0 && i == 0 &&
3433 minorvers != NFSV4_MINORVERSION)
3434 error = NFSERR_OPNOTINSESS;
3437 if (minorvers == NFSV4_MINORVERSION)
3438 clp = nfscl_getclnt(cbident);
3440 clp = nfscl_getclntsess(sessionid);
3442 dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
3444 if (dp != NULL && (dp->nfsdl_flags &
3445 NFSCLDL_DELEGRET) == 0) {
3448 wakeup((caddr_t)clp);
3451 error = NFSERR_SERVERFAULT;
3456 free(nfhp, M_NFSFH);
3458 case NFSV4OP_CBLAYOUTRECALL:
3459 NFSCL_DEBUG(4, "cblayrec\n");
3461 NFSM_DISSECT(tl, uint32_t *, 4 * NFSX_UNSIGNED);
3462 laytype = fxdr_unsigned(int, *tl++);
3463 iomode = fxdr_unsigned(uint32_t, *tl++);
3464 if (newnfs_true == *tl++)
3468 recalltype = fxdr_unsigned(int, *tl);
3469 NFSCL_DEBUG(4, "layt=%d iom=%d ch=%d rectyp=%d\n",
3470 laytype, iomode, changed, recalltype);
3471 recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL,
3473 if (laytype != NFSLAYOUT_NFSV4_1_FILES &&
3474 laytype != NFSLAYOUT_FLEXFILE)
3475 error = NFSERR_NOMATCHLAYOUT;
3476 else if (recalltype == NFSLAYOUTRETURN_FILE) {
3477 error = nfsm_getfh(nd, &nfhp);
3478 NFSCL_DEBUG(4, "retfile getfh=%d\n", error);
3481 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_HYPER +
3483 off = fxdr_hyper(tl); tl += 2;
3484 len = fxdr_hyper(tl); tl += 2;
3485 stateid.seqid = fxdr_unsigned(uint32_t, *tl++);
3486 NFSBCOPY(tl, stateid.other, NFSX_STATEIDOTHER);
3487 if (minorvers == NFSV4_MINORVERSION)
3488 error = NFSERR_NOTSUPP;
3490 error = NFSERR_OPNOTINSESS;
3491 NFSCL_DEBUG(4, "off=%ju len=%ju sq=%u err=%d\n",
3492 (uintmax_t)off, (uintmax_t)len,
3493 stateid.seqid, error);
3496 clp = nfscl_getclntsess(sessionid);
3497 NFSCL_DEBUG(4, "cbly clp=%p\n", clp);
3499 lyp = nfscl_findlayout(clp,
3502 NFSCL_DEBUG(4, "cblyp=%p\n",
3507 NFSLY_FLEXFILE)) != 0 &&
3508 !NFSBCMP(stateid.other,
3509 lyp->nfsly_stateid.other,
3510 NFSX_STATEIDOTHER)) {
3520 lyp->nfsly_stateid.seqid)
3521 lyp->nfsly_stateid.seqid =
3529 lyp->nfsly_stateid.seqid);
3532 NFSERR_NOMATCHLAYOUT;
3534 error = NFSERR_NOMATCHLAYOUT;
3537 free(nfhp, M_NFSFH);
3538 } else if (recalltype == NFSLAYOUTRETURN_FSID) {
3539 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_HYPER);
3540 filesid[0] = fxdr_hyper(tl); tl += 2;
3541 filesid[1] = fxdr_hyper(tl); tl += 2;
3544 clp = nfscl_getclntsess(sessionid);
3546 TAILQ_FOREACH(lyp, &clp->nfsc_layout,
3548 if (lyp->nfsly_filesid[0] ==
3550 lyp->nfsly_filesid[1] ==
3557 lyp->nfsly_stateid.seqid,
3567 error = NFSERR_NOMATCHLAYOUT;
3569 error = NFSERR_NOMATCHLAYOUT;
3571 } else if (recalltype == NFSLAYOUTRETURN_ALL) {
3574 clp = nfscl_getclntsess(sessionid);
3576 TAILQ_FOREACH(lyp, &clp->nfsc_layout,
3578 error = nfscl_layoutrecall(
3579 recalltype, lyp, iomode, 0,
3581 lyp->nfsly_stateid.seqid,
3582 0, 0, NULL, recallp);
3589 error = NFSERR_NOMATCHLAYOUT;
3591 error = NFSERR_NOMATCHLAYOUT;
3594 error = NFSERR_NOMATCHLAYOUT;
3595 if (recallp != NULL) {
3596 free(recallp, M_NFSLAYRECALL);
3600 case NFSV4OP_CBSEQUENCE:
3601 NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID +
3603 bcopy(tl, sessionid, NFSX_V4SESSIONID);
3604 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3605 seqid = fxdr_unsigned(uint32_t, *tl++);
3606 slotid = fxdr_unsigned(uint32_t, *tl++);
3607 highslot = fxdr_unsigned(uint32_t, *tl++);
3609 /* Throw away the referring call stuff. */
3610 clist = fxdr_unsigned(int, *tl);
3611 for (j = 0; j < clist; j++) {
3612 NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID +
3614 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3615 rcalls = fxdr_unsigned(int, *tl);
3616 for (k = 0; k < rcalls; k++) {
3617 NFSM_DISSECT(tl, uint32_t *,
3623 clp = nfscl_getclntsess(sessionid);
3625 error = NFSERR_SERVERFAULT;
3627 error = NFSERR_SEQUENCEPOS;
3629 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3630 error = nfsv4_seqsession(seqid, slotid,
3631 highslot, tsep->nfsess_cbslots, &rep,
3632 tsep->nfsess_backslots);
3635 if (error == 0 || error == NFSERR_REPLYFROMCACHE) {
3639 * Handle a reply for a retried
3640 * callback. The reply will be
3641 * re-inserted in the session cache
3642 * by the nfsv4_seqsess_cacherep() call
3645 KASSERT(error == NFSERR_REPLYFROMCACHE,
3646 ("cbsequence: non-NULL rep"));
3647 NFSCL_DEBUG(4, "Got cbretry\n");
3648 m_freem(nd->nd_mreq);
3653 NFSM_BUILD(tl, uint32_t *,
3654 NFSX_V4SESSIONID + 4 * NFSX_UNSIGNED);
3655 bcopy(sessionid, tl, NFSX_V4SESSIONID);
3656 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3657 *tl++ = txdr_unsigned(seqid);
3658 *tl++ = txdr_unsigned(slotid);
3659 *tl++ = txdr_unsigned(NFSV4_CBSLOTS - 1);
3660 *tl = txdr_unsigned(NFSV4_CBSLOTS - 1);
3664 if (i == 0 && minorvers != NFSV4_MINORVERSION)
3665 error = NFSERR_OPNOTINSESS;
3667 NFSCL_DEBUG(1, "unsupp callback %d\n", op);
3668 error = NFSERR_NOTSUPP;
3673 if (error == EBADRPC || error == NFSERR_BADXDR) {
3674 nd->nd_repstat = NFSERR_BADXDR;
3676 nd->nd_repstat = error;
3681 if (nd->nd_repstat) {
3682 *repp = nfscl_errmap(nd, minorvers);
3685 *repp = 0; /* NFS4_OK */
3688 if (recallp != NULL)
3689 free(recallp, M_NFSLAYRECALL);
3691 if (error == EBADRPC || error == NFSERR_BADXDR)
3692 nd->nd_repstat = NFSERR_BADXDR;
3694 printf("nfsv4 comperr1=%d\n", error);
3697 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
3701 *retopsp = txdr_unsigned(retops);
3703 *nd->nd_errp = nfscl_errmap(nd, minorvers);
3705 if (gotseq_ok != 0) {
3706 rep = m_copym(nd->nd_mreq, 0, M_COPYALL, M_WAITOK);
3708 clp = nfscl_getclntsess(sessionid);
3710 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3711 nfsv4_seqsess_cacherep(slotid, tsep->nfsess_cbslots,
3722 * Generate the next cbident value. Basically just increment a static value
3723 * and then check that it isn't already in the list, if it has wrapped around.
3726 nfscl_nextcbident(void)
3728 struct nfsclclient *clp;
3730 static u_int32_t nextcbident = 0;
3731 static int haswrapped = 0;
3734 if (nextcbident == 0)
3738 * Search the clientid list for one already using this cbident.
3743 LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3744 if (clp->nfsc_cbident == nextcbident) {
3754 return (nextcbident);
3758 * Get the mount point related to a given cbident or session and busy it.
3761 nfscl_getmnt(int minorvers, uint8_t *sessionid, u_int32_t cbident,
3762 struct nfsclclient **clpp)
3764 struct nfsclclient *clp;
3767 struct nfsclsession *tsep;
3771 LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3772 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3773 if (minorvers == NFSV4_MINORVERSION) {
3774 if (clp->nfsc_cbident == cbident)
3776 } else if (!NFSBCMP(tsep->nfsess_sessionid, sessionid,
3784 mp = clp->nfsc_nmp->nm_mountp;
3787 error = vfs_busy(mp, 0);
3796 * Get the clientid pointer related to a given cbident.
3798 static struct nfsclclient *
3799 nfscl_getclnt(u_int32_t cbident)
3801 struct nfsclclient *clp;
3803 LIST_FOREACH(clp, &nfsclhead, nfsc_list)
3804 if (clp->nfsc_cbident == cbident)
3810 * Get the clientid pointer related to a given sessionid.
3812 static struct nfsclclient *
3813 nfscl_getclntsess(uint8_t *sessionid)
3815 struct nfsclclient *clp;
3816 struct nfsclsession *tsep;
3818 LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3819 tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3820 if (!NFSBCMP(tsep->nfsess_sessionid, sessionid,
3828 * Search for a lock conflict locally on the client. A conflict occurs if
3829 * - not same owner and overlapping byte range and at least one of them is
3830 * a write lock or this is an unlock.
3833 nfscl_localconflict(struct nfsclclient *clp, u_int8_t *fhp, int fhlen,
3834 struct nfscllock *nlop, u_int8_t *own, struct nfscldeleg *dp,
3835 struct nfscllock **lopp)
3837 struct nfsclowner *owp;
3838 struct nfsclopen *op;
3842 ret = nfscl_checkconflict(&dp->nfsdl_lock, nlop, own, lopp);
3846 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3847 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3848 if (op->nfso_fhlen == fhlen &&
3849 !NFSBCMP(op->nfso_fh, fhp, fhlen)) {
3850 ret = nfscl_checkconflict(&op->nfso_lock, nlop,
3861 nfscl_checkconflict(struct nfscllockownerhead *lhp, struct nfscllock *nlop,
3862 u_int8_t *own, struct nfscllock **lopp)
3864 struct nfscllockowner *lp;
3865 struct nfscllock *lop;
3867 LIST_FOREACH(lp, lhp, nfsl_list) {
3868 if (NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
3869 LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
3870 if (lop->nfslo_first >= nlop->nfslo_end)
3872 if (lop->nfslo_end <= nlop->nfslo_first)
3874 if (lop->nfslo_type == F_WRLCK ||
3875 nlop->nfslo_type == F_WRLCK ||
3876 nlop->nfslo_type == F_UNLCK) {
3879 return (NFSERR_DENIED);
3888 * Check for a local conflicting lock.
3891 nfscl_lockt(vnode_t vp, struct nfsclclient *clp, u_int64_t off,
3892 u_int64_t len, struct flock *fl, NFSPROC_T *p, void *id, int flags)
3894 struct nfscllock *lop, nlck;
3895 struct nfscldeleg *dp;
3897 u_int8_t own[NFSV4CL_LOCKNAMELEN];
3900 nlck.nfslo_type = fl->l_type;
3901 nlck.nfslo_first = off;
3902 if (len == NFS64BITSSET) {
3903 nlck.nfslo_end = NFS64BITSSET;
3905 nlck.nfslo_end = off + len;
3906 if (nlck.nfslo_end <= nlck.nfslo_first)
3907 return (NFSERR_INVAL);
3910 nfscl_filllockowner(id, own, flags);
3912 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
3913 error = nfscl_localconflict(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len,
3914 &nlck, own, dp, &lop);
3916 fl->l_whence = SEEK_SET;
3917 fl->l_start = lop->nfslo_first;
3918 if (lop->nfslo_end == NFS64BITSSET)
3921 fl->l_len = lop->nfslo_end - lop->nfslo_first;
3922 fl->l_pid = (pid_t)0;
3923 fl->l_type = lop->nfslo_type;
3924 error = -1; /* no RPC required */
3925 } else if (dp != NULL && ((dp->nfsdl_flags & NFSCLDL_WRITE) ||
3926 fl->l_type == F_RDLCK)) {
3928 * The delegation ensures that there isn't a conflicting
3929 * lock on the server, so return -1 to indicate an RPC
3932 fl->l_type = F_UNLCK;
3940 * Handle Recall of a delegation.
3941 * The clp must be exclusive locked when this is called.
3944 nfscl_recalldeleg(struct nfsclclient *clp, struct nfsmount *nmp,
3945 struct nfscldeleg *dp, vnode_t vp, struct ucred *cred, NFSPROC_T *p,
3946 int called_from_renewthread)
3948 struct nfsclowner *owp, *lowp, *nowp;
3949 struct nfsclopen *op, *lop;
3950 struct nfscllockowner *lp;
3951 struct nfscllock *lckp;
3953 int error = 0, ret, gotvp = 0;
3957 * First, get a vnode for the file. This is needed to do RPCs.
3959 ret = nfscl_ngetreopen(nmp->nm_mountp, dp->nfsdl_fh,
3960 dp->nfsdl_fhlen, p, &np);
3963 * File isn't open, so nothing to move over to the
3973 dp->nfsdl_flags &= ~NFSCLDL_MODTIMESET;
3976 * Ok, if it's a write delegation, flush data to the server, so
3977 * that close/open consistency is retained.
3981 if ((dp->nfsdl_flags & NFSCLDL_WRITE) && (np->n_flag & NMODIFIED)) {
3982 np->n_flag |= NDELEGRECALL;
3984 ret = ncl_flush(vp, MNT_WAIT, p, 1, called_from_renewthread);
3986 np->n_flag &= ~NDELEGRECALL;
3988 NFSINVALATTRCACHE(np);
3990 if (ret == EIO && called_from_renewthread != 0) {
3992 * If the flush failed with EIO for the renew thread,
3993 * return now, so that the dirty buffer will be flushed
4002 * Now, for each openowner with opens issued locally, move them
4003 * over to state against the server.
4005 LIST_FOREACH(lowp, &dp->nfsdl_owner, nfsow_list) {
4006 lop = LIST_FIRST(&lowp->nfsow_open);
4008 if (LIST_NEXT(lop, nfso_list) != NULL)
4009 panic("nfsdlg mult opens");
4011 * Look for the same openowner against the server.
4013 LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
4014 if (!NFSBCMP(lowp->nfsow_owner,
4015 owp->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
4016 newnfs_copycred(&dp->nfsdl_cred, cred);
4017 ret = nfscl_moveopen(vp, clp, nmp, lop,
4019 if (ret == NFSERR_STALECLIENTID ||
4020 ret == NFSERR_STALEDONTRECOVER ||
4021 ret == NFSERR_BADSESSION) {
4027 nfscl_freeopen(lop, 1);
4036 * If no openowner found, create one and get an open
4041 sizeof (struct nfsclowner), M_NFSCLOWNER,
4043 nfscl_newopen(clp, NULL, &owp, &nowp, &op,
4044 NULL, lowp->nfsow_owner, dp->nfsdl_fh,
4045 dp->nfsdl_fhlen, NULL, NULL);
4046 newnfs_copycred(&dp->nfsdl_cred, cred);
4047 ret = nfscl_moveopen(vp, clp, nmp, lop,
4050 nfscl_freeopenowner(owp, 0);
4051 if (ret == NFSERR_STALECLIENTID ||
4052 ret == NFSERR_STALEDONTRECOVER ||
4053 ret == NFSERR_BADSESSION) {
4059 nfscl_freeopen(lop, 1);
4069 * Now, get byte range locks for any locks done locally.
4071 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4072 LIST_FOREACH(lckp, &lp->nfsl_lock, nfslo_list) {
4073 newnfs_copycred(&dp->nfsdl_cred, cred);
4074 ret = nfscl_relock(vp, clp, nmp, lp, lckp, cred, p);
4075 if (ret == NFSERR_STALESTATEID ||
4076 ret == NFSERR_STALEDONTRECOVER ||
4077 ret == NFSERR_STALECLIENTID ||
4078 ret == NFSERR_BADSESSION) {
4093 * Move a locally issued open over to an owner on the state list.
4094 * SIDE EFFECT: If it needs to sleep (do an rpc), it unlocks clstate and
4095 * returns with it unlocked.
4098 nfscl_moveopen(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp,
4099 struct nfsclopen *lop, struct nfsclowner *owp, struct nfscldeleg *dp,
4100 struct ucred *cred, NFSPROC_T *p)
4102 struct nfsclopen *op, *nop;
4103 struct nfscldeleg *ndp;
4105 int error = 0, newone;
4108 * First, look for an appropriate open, If found, just increment the
4111 LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
4112 if ((op->nfso_mode & lop->nfso_mode) == lop->nfso_mode &&
4113 op->nfso_fhlen == lop->nfso_fhlen &&
4114 !NFSBCMP(op->nfso_fh, lop->nfso_fh, op->nfso_fhlen)) {
4115 op->nfso_opencnt += lop->nfso_opencnt;
4116 nfscl_freeopen(lop, 1);
4121 /* No appropriate open, so we have to do one against the server. */
4123 nop = malloc(sizeof (struct nfsclopen) +
4124 lop->nfso_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
4126 nfscl_newopen(clp, NULL, &owp, NULL, &op, &nop, owp->nfsow_owner,
4127 lop->nfso_fh, lop->nfso_fhlen, cred, &newone);
4129 error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data, np->n_v4->n4_fhlen,
4130 lop->nfso_fh, lop->nfso_fhlen, lop->nfso_mode, op,
4131 NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, &ndp, 0, 0, cred, p);
4134 nfscl_freeopen(op, 0);
4136 op->nfso_mode |= lop->nfso_mode;
4137 op->nfso_opencnt += lop->nfso_opencnt;
4138 nfscl_freeopen(lop, 1);
4141 free(nop, M_NFSCLOPEN);
4144 * What should I do with the returned delegation, since the
4145 * delegation is being recalled? For now, just printf and
4148 printf("Moveopen returned deleg\n");
4149 free(ndp, M_NFSCLDELEG);
4155 * Recall all delegations on this client.
4158 nfscl_totalrecall(struct nfsclclient *clp)
4160 struct nfscldeleg *dp;
4162 TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
4163 if ((dp->nfsdl_flags & NFSCLDL_DELEGRET) == 0)
4164 dp->nfsdl_flags |= NFSCLDL_RECALL;
4169 * Relock byte ranges. Called for delegation recall and state expiry.
4172 nfscl_relock(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp,
4173 struct nfscllockowner *lp, struct nfscllock *lop, struct ucred *cred,
4176 struct nfscllockowner *nlp;
4179 int error, newone, donelocally;
4181 off = lop->nfslo_first;
4182 len = lop->nfslo_end - lop->nfslo_first;
4183 error = nfscl_getbytelock(vp, off, len, lop->nfslo_type, cred, p,
4184 clp, 1, NULL, lp->nfsl_lockflags, lp->nfsl_owner,
4185 lp->nfsl_openowner, &nlp, &newone, &donelocally);
4186 if (error || donelocally)
4188 nfhp = VTONFS(vp)->n_fhp;
4189 error = nfscl_trylock(nmp, vp, nfhp->nfh_fh,
4190 nfhp->nfh_len, nlp, newone, 0, off,
4191 len, lop->nfslo_type, cred, p);
4193 nfscl_freelockowner(nlp, 0);
4198 * Called to re-open a file. Basically get a vnode for the file handle
4199 * and then call nfsrpc_openrpc() to do the rest.
4202 nfsrpc_reopen(struct nfsmount *nmp, u_int8_t *fhp, int fhlen,
4203 u_int32_t mode, struct nfsclopen *op, struct nfscldeleg **dpp,
4204 struct ucred *cred, NFSPROC_T *p)
4210 error = nfscl_ngetreopen(nmp->nm_mountp, fhp, fhlen, p, &np);
4214 if (np->n_v4 != NULL) {
4215 error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data,
4216 np->n_v4->n4_fhlen, fhp, fhlen, mode, op,
4217 NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, dpp, 0, 0,
4227 * Try an open against the server. Just call nfsrpc_openrpc(), retrying while
4228 * NFSERR_DELAY. Also, try system credentials, if the passed in credentials
4232 nfscl_tryopen(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen,
4233 u_int8_t *newfhp, int newfhlen, u_int32_t mode, struct nfsclopen *op,
4234 u_int8_t *name, int namelen, struct nfscldeleg **ndpp,
4235 int reclaim, u_int32_t delegtype, struct ucred *cred, NFSPROC_T *p)
4240 error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp, newfhlen,
4241 mode, op, name, namelen, ndpp, reclaim, delegtype, cred, p,
4243 if (error == NFSERR_DELAY)
4244 (void) nfs_catnap(PZERO, error, "nfstryop");
4245 } while (error == NFSERR_DELAY);
4246 if (error == EAUTH || error == EACCES) {
4247 /* Try again using system credentials */
4248 newnfs_setroot(cred);
4250 error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp,
4251 newfhlen, mode, op, name, namelen, ndpp, reclaim,
4252 delegtype, cred, p, 1, 0);
4253 if (error == NFSERR_DELAY)
4254 (void) nfs_catnap(PZERO, error, "nfstryop");
4255 } while (error == NFSERR_DELAY);
4261 * Try a byte range lock. Just loop on nfsrpc_lock() while it returns
4262 * NFSERR_DELAY. Also, retry with system credentials, if the provided
4266 nfscl_trylock(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp,
4267 int fhlen, struct nfscllockowner *nlp, int newone, int reclaim,
4268 u_int64_t off, u_int64_t len, short type, struct ucred *cred, NFSPROC_T *p)
4270 struct nfsrv_descript nfsd, *nd = &nfsd;
4274 error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp, newone,
4275 reclaim, off, len, type, cred, p, 0);
4276 if (!error && nd->nd_repstat == NFSERR_DELAY)
4277 (void) nfs_catnap(PZERO, (int)nd->nd_repstat,
4279 } while (!error && nd->nd_repstat == NFSERR_DELAY);
4281 error = nd->nd_repstat;
4282 if (error == EAUTH || error == EACCES) {
4283 /* Try again using root credentials */
4284 newnfs_setroot(cred);
4286 error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp,
4287 newone, reclaim, off, len, type, cred, p, 1);
4288 if (!error && nd->nd_repstat == NFSERR_DELAY)
4289 (void) nfs_catnap(PZERO, (int)nd->nd_repstat,
4291 } while (!error && nd->nd_repstat == NFSERR_DELAY);
4293 error = nd->nd_repstat;
4299 * Try a delegreturn against the server. Just call nfsrpc_delegreturn(),
4300 * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
4304 nfscl_trydelegreturn(struct nfscldeleg *dp, struct ucred *cred,
4305 struct nfsmount *nmp, NFSPROC_T *p)
4310 error = nfsrpc_delegreturn(dp, cred, nmp, p, 0);
4311 if (error == NFSERR_DELAY)
4312 (void) nfs_catnap(PZERO, error, "nfstrydp");
4313 } while (error == NFSERR_DELAY);
4314 if (error == EAUTH || error == EACCES) {
4315 /* Try again using system credentials */
4316 newnfs_setroot(cred);
4318 error = nfsrpc_delegreturn(dp, cred, nmp, p, 1);
4319 if (error == NFSERR_DELAY)
4320 (void) nfs_catnap(PZERO, error, "nfstrydp");
4321 } while (error == NFSERR_DELAY);
4327 * Try a close against the server. Just call nfsrpc_closerpc(),
4328 * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
4332 nfscl_tryclose(struct nfsclopen *op, struct ucred *cred,
4333 struct nfsmount *nmp, NFSPROC_T *p)
4335 struct nfsrv_descript nfsd, *nd = &nfsd;
4339 error = nfsrpc_closerpc(nd, nmp, op, cred, p, 0);
4340 if (error == NFSERR_DELAY)
4341 (void) nfs_catnap(PZERO, error, "nfstrycl");
4342 } while (error == NFSERR_DELAY);
4343 if (error == EAUTH || error == EACCES) {
4344 /* Try again using system credentials */
4345 newnfs_setroot(cred);
4347 error = nfsrpc_closerpc(nd, nmp, op, cred, p, 1);
4348 if (error == NFSERR_DELAY)
4349 (void) nfs_catnap(PZERO, error, "nfstrycl");
4350 } while (error == NFSERR_DELAY);
4356 * Decide if a delegation on a file permits close without flushing writes
4357 * to the server. This might be a big performance win in some environments.
4358 * (Not useful until the client does caching on local stable storage.)
4361 nfscl_mustflush(vnode_t vp)
4363 struct nfsclclient *clp;
4364 struct nfscldeleg *dp;
4366 struct nfsmount *nmp;
4369 nmp = VFSTONFS(vp->v_mount);
4370 if (!NFSHASNFSV4(nmp))
4373 clp = nfscl_findcl(nmp);
4378 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4379 if (dp != NULL && (dp->nfsdl_flags &
4380 (NFSCLDL_WRITE | NFSCLDL_RECALL | NFSCLDL_DELEGRET)) ==
4382 (dp->nfsdl_sizelimit >= np->n_size ||
4383 !NFSHASSTRICT3530(nmp))) {
4392 * See if a (write) delegation exists for this file.
4395 nfscl_nodeleg(vnode_t vp, int writedeleg)
4397 struct nfsclclient *clp;
4398 struct nfscldeleg *dp;
4400 struct nfsmount *nmp;
4403 nmp = VFSTONFS(vp->v_mount);
4404 if (!NFSHASNFSV4(nmp))
4407 clp = nfscl_findcl(nmp);
4412 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4414 (dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_DELEGRET)) == 0 &&
4415 (writedeleg == 0 || (dp->nfsdl_flags & NFSCLDL_WRITE) ==
4425 * Look for an associated delegation that should be DelegReturned.
4428 nfscl_removedeleg(vnode_t vp, NFSPROC_T *p, nfsv4stateid_t *stp)
4430 struct nfsclclient *clp;
4431 struct nfscldeleg *dp;
4432 struct nfsclowner *owp;
4433 struct nfscllockowner *lp;
4434 struct nfsmount *nmp;
4437 int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
4439 nmp = VFSTONFS(vp->v_mount);
4443 * Loop around waiting for:
4444 * - outstanding I/O operations on delegations to complete
4445 * - for a delegation on vp that has state, lock the client and
4447 * - return delegation with no state
4450 clp = nfscl_findcl(nmp);
4455 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4456 np->n_fhp->nfh_len);
4459 * Wait for outstanding I/O ops to be done.
4461 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4463 nfsv4_unlock(&clp->nfsc_lock, 0);
4466 dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4467 (void) nfsmsleep(&dp->nfsdl_rwlock,
4468 NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4472 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4473 if (!LIST_EMPTY(&owp->nfsow_open)) {
4479 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4480 if (!LIST_EMPTY(&lp->nfsl_lock)) {
4486 if (needsrecall && !triedrecall) {
4487 dp->nfsdl_flags |= NFSCLDL_DELEGRET;
4490 igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
4491 &islept, NFSCLSTATEMUTEXPTR, NULL);
4498 cred = newnfs_getcred();
4499 newnfs_copycred(&dp->nfsdl_cred, cred);
4500 (void) nfscl_recalldeleg(clp, nmp, dp, vp, cred, p, 0);
4504 nfsv4_unlock(&clp->nfsc_lock, 0);
4508 *stp = dp->nfsdl_stateid;
4510 nfscl_cleandeleg(dp);
4511 nfscl_freedeleg(&clp->nfsc_deleg, dp);
4514 nfsv4_unlock(&clp->nfsc_lock, 0);
4521 * Look for associated delegation(s) that should be DelegReturned.
4524 nfscl_renamedeleg(vnode_t fvp, nfsv4stateid_t *fstp, int *gotfdp, vnode_t tvp,
4525 nfsv4stateid_t *tstp, int *gottdp, NFSPROC_T *p)
4527 struct nfsclclient *clp;
4528 struct nfscldeleg *dp;
4529 struct nfsclowner *owp;
4530 struct nfscllockowner *lp;
4531 struct nfsmount *nmp;
4534 int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
4536 nmp = VFSTONFS(fvp->v_mount);
4541 * Loop around waiting for:
4542 * - outstanding I/O operations on delegations to complete
4543 * - for a delegation on fvp that has state, lock the client and
4545 * - return delegation(s) with no state.
4548 clp = nfscl_findcl(nmp);
4554 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4555 np->n_fhp->nfh_len);
4556 if (dp != NULL && *gotfdp == 0) {
4558 * Wait for outstanding I/O ops to be done.
4560 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4562 nfsv4_unlock(&clp->nfsc_lock, 0);
4565 dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4566 (void) nfsmsleep(&dp->nfsdl_rwlock,
4567 NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4571 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4572 if (!LIST_EMPTY(&owp->nfsow_open)) {
4578 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4579 if (!LIST_EMPTY(&lp->nfsl_lock)) {
4585 if (needsrecall && !triedrecall) {
4586 dp->nfsdl_flags |= NFSCLDL_DELEGRET;
4589 igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
4590 &islept, NFSCLSTATEMUTEXPTR, NULL);
4597 cred = newnfs_getcred();
4598 newnfs_copycred(&dp->nfsdl_cred, cred);
4599 (void) nfscl_recalldeleg(clp, nmp, dp, fvp, cred, p, 0);
4603 nfsv4_unlock(&clp->nfsc_lock, 0);
4607 *fstp = dp->nfsdl_stateid;
4610 nfscl_cleandeleg(dp);
4611 nfscl_freedeleg(&clp->nfsc_deleg, dp);
4614 nfsv4_unlock(&clp->nfsc_lock, 0);
4619 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4620 np->n_fhp->nfh_len);
4621 if (dp != NULL && *gottdp == 0) {
4623 * Wait for outstanding I/O ops to be done.
4625 if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4626 dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4627 (void) nfsmsleep(&dp->nfsdl_rwlock,
4628 NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4631 LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4632 if (!LIST_EMPTY(&owp->nfsow_open)) {
4637 LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4638 if (!LIST_EMPTY(&lp->nfsl_lock)) {
4643 *tstp = dp->nfsdl_stateid;
4646 nfscl_cleandeleg(dp);
4647 nfscl_freedeleg(&clp->nfsc_deleg, dp);
4656 * Get a reference on the clientid associated with the mount point.
4657 * Return 1 if success, 0 otherwise.
4660 nfscl_getref(struct nfsmount *nmp)
4662 struct nfsclclient *clp;
4665 clp = nfscl_findcl(nmp);
4670 nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, NULL);
4676 * Release a reference on a clientid acquired with the above call.
4679 nfscl_relref(struct nfsmount *nmp)
4681 struct nfsclclient *clp;
4684 clp = nfscl_findcl(nmp);
4689 nfsv4_relref(&clp->nfsc_lock);
4694 * Save the size attribute in the delegation, since the nfsnode
4698 nfscl_reclaimnode(vnode_t vp)
4700 struct nfsclclient *clp;
4701 struct nfscldeleg *dp;
4702 struct nfsnode *np = VTONFS(vp);
4703 struct nfsmount *nmp;
4705 nmp = VFSTONFS(vp->v_mount);
4706 if (!NFSHASNFSV4(nmp))
4709 clp = nfscl_findcl(nmp);
4714 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4715 if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE))
4716 dp->nfsdl_size = np->n_size;
4721 * Get the saved size attribute in the delegation, since it is a
4722 * newly allocated nfsnode.
4725 nfscl_newnode(vnode_t vp)
4727 struct nfsclclient *clp;
4728 struct nfscldeleg *dp;
4729 struct nfsnode *np = VTONFS(vp);
4730 struct nfsmount *nmp;
4732 nmp = VFSTONFS(vp->v_mount);
4733 if (!NFSHASNFSV4(nmp))
4736 clp = nfscl_findcl(nmp);
4741 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4742 if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE))
4743 np->n_size = dp->nfsdl_size;
4748 * If there is a valid write delegation for this file, set the modtime
4749 * to the local clock time.
4752 nfscl_delegmodtime(vnode_t vp)
4754 struct nfsclclient *clp;
4755 struct nfscldeleg *dp;
4756 struct nfsnode *np = VTONFS(vp);
4757 struct nfsmount *nmp;
4759 nmp = VFSTONFS(vp->v_mount);
4760 if (!NFSHASNFSV4(nmp))
4763 clp = nfscl_findcl(nmp);
4768 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4769 if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE)) {
4770 nanotime(&dp->nfsdl_modtime);
4771 dp->nfsdl_flags |= NFSCLDL_MODTIMESET;
4777 * If there is a valid write delegation for this file with a modtime set,
4778 * put that modtime in mtime.
4781 nfscl_deleggetmodtime(vnode_t vp, struct timespec *mtime)
4783 struct nfsclclient *clp;
4784 struct nfscldeleg *dp;
4785 struct nfsnode *np = VTONFS(vp);
4786 struct nfsmount *nmp;
4788 nmp = VFSTONFS(vp->v_mount);
4789 if (!NFSHASNFSV4(nmp))
4792 clp = nfscl_findcl(nmp);
4797 dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4799 (dp->nfsdl_flags & (NFSCLDL_WRITE | NFSCLDL_MODTIMESET)) ==
4800 (NFSCLDL_WRITE | NFSCLDL_MODTIMESET))
4801 *mtime = dp->nfsdl_modtime;
4806 nfscl_errmap(struct nfsrv_descript *nd, u_int32_t minorvers)
4808 short *defaulterrp, *errp;
4810 if (!nd->nd_repstat)
4812 if (nd->nd_procnum == NFSPROC_NOOP)
4813 return (txdr_unsigned(nd->nd_repstat & 0xffff));
4814 if (nd->nd_repstat == EBADRPC)
4815 return (txdr_unsigned(NFSERR_BADXDR));
4816 if (nd->nd_repstat == NFSERR_MINORVERMISMATCH ||
4817 nd->nd_repstat == NFSERR_OPILLEGAL)
4818 return (txdr_unsigned(nd->nd_repstat));
4819 if (nd->nd_repstat >= NFSERR_BADIOMODE && nd->nd_repstat < 20000 &&
4820 minorvers > NFSV4_MINORVERSION) {
4821 /* NFSv4.n error. */
4822 return (txdr_unsigned(nd->nd_repstat));
4824 if (nd->nd_procnum < NFSV4OP_CBNOPS)
4825 errp = defaulterrp = nfscl_cberrmap[nd->nd_procnum];
4827 return (txdr_unsigned(nd->nd_repstat));
4829 if (*errp == (short)nd->nd_repstat)
4830 return (txdr_unsigned(nd->nd_repstat));
4831 return (txdr_unsigned(*defaulterrp));
4835 * Called to find/add a layout to a client.
4836 * This function returns the layout with a refcnt (shared lock) upon
4837 * success (returns 0) or with no lock/refcnt on the layout when an
4838 * error is returned.
4839 * If a layout is passed in via lypp, it is locked (exclusively locked).
4842 nfscl_layout(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen,
4843 nfsv4stateid_t *stateidp, int layouttype, int retonclose,
4844 struct nfsclflayouthead *fhlp, struct nfscllayout **lypp,
4845 struct ucred *cred, NFSPROC_T *p)
4847 struct nfsclclient *clp;
4848 struct nfscllayout *lyp, *tlyp;
4849 struct nfsclflayout *flp;
4850 struct nfsnode *np = VTONFS(vp);
4852 int layout_passed_in;
4854 mp = nmp->nm_mountp;
4855 layout_passed_in = 1;
4859 layout_passed_in = 0;
4860 tlyp = malloc(sizeof(*tlyp) + fhlen - 1, M_NFSLAYOUT,
4867 if (layout_passed_in != 0)
4868 nfsv4_unlock(&lyp->nfsly_lock, 0);
4871 free(tlyp, M_NFSLAYOUT);
4876 * Although no lyp was passed in, another thread might have
4877 * allocated one. If one is found, just increment it's ref
4878 * count and return it.
4880 lyp = nfscl_findlayout(clp, fhp, fhlen);
4884 lyp->nfsly_stateid.seqid = stateidp->seqid;
4885 lyp->nfsly_stateid.other[0] = stateidp->other[0];
4886 lyp->nfsly_stateid.other[1] = stateidp->other[1];
4887 lyp->nfsly_stateid.other[2] = stateidp->other[2];
4888 lyp->nfsly_lastbyte = 0;
4889 LIST_INIT(&lyp->nfsly_flayread);
4890 LIST_INIT(&lyp->nfsly_flayrw);
4891 LIST_INIT(&lyp->nfsly_recall);
4892 lyp->nfsly_filesid[0] = np->n_vattr.na_filesid[0];
4893 lyp->nfsly_filesid[1] = np->n_vattr.na_filesid[1];
4894 lyp->nfsly_clp = clp;
4895 if (layouttype == NFSLAYOUT_FLEXFILE)
4896 lyp->nfsly_flags = NFSLY_FLEXFILE;
4898 lyp->nfsly_flags = NFSLY_FILES;
4899 if (retonclose != 0)
4900 lyp->nfsly_flags |= NFSLY_RETONCLOSE;
4901 lyp->nfsly_fhlen = fhlen;
4902 NFSBCOPY(fhp, lyp->nfsly_fh, fhlen);
4903 TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
4904 LIST_INSERT_HEAD(NFSCLLAYOUTHASH(clp, fhp, fhlen), lyp,
4906 lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
4909 if (retonclose != 0)
4910 lyp->nfsly_flags |= NFSLY_RETONCLOSE;
4911 if (stateidp->seqid > lyp->nfsly_stateid.seqid)
4912 lyp->nfsly_stateid.seqid = stateidp->seqid;
4913 TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list);
4914 TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
4915 lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
4917 nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
4918 if (NFSCL_FORCEDISM(mp)) {
4921 free(tlyp, M_NFSLAYOUT);
4925 } else if (stateidp->seqid > lyp->nfsly_stateid.seqid)
4926 lyp->nfsly_stateid.seqid = stateidp->seqid;
4928 /* Merge the new list of File Layouts into the list. */
4929 flp = LIST_FIRST(fhlp);
4931 if (flp->nfsfl_iomode == NFSLAYOUTIOMODE_READ)
4932 nfscl_mergeflayouts(&lyp->nfsly_flayread, fhlp);
4934 nfscl_mergeflayouts(&lyp->nfsly_flayrw, fhlp);
4936 if (layout_passed_in != 0)
4937 nfsv4_unlock(&lyp->nfsly_lock, 1);
4940 free(tlyp, M_NFSLAYOUT);
4945 * Search for a layout by MDS file handle.
4946 * If one is found, it is returned with a refcnt (shared lock) iff
4947 * retflpp returned non-NULL and locked (exclusive locked) iff retflpp is
4950 struct nfscllayout *
4951 nfscl_getlayout(struct nfsclclient *clp, uint8_t *fhp, int fhlen,
4952 uint64_t off, struct nfsclflayout **retflpp, int *recalledp)
4954 struct nfscllayout *lyp;
4956 int error, igotlock;
4958 mp = clp->nfsc_nmp->nm_mountp;
4962 lyp = nfscl_findlayout(clp, fhp, fhlen);
4964 if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
4965 TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list);
4966 TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
4967 lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
4968 error = nfscl_findlayoutforio(lyp, off,
4969 NFSV4OPEN_ACCESSREAD, retflpp);
4971 nfsv4_getref(&lyp->nfsly_lock, NULL,
4972 NFSCLSTATEMUTEXPTR, mp);
4975 igotlock = nfsv4_lock(&lyp->nfsly_lock,
4976 1, NULL, NFSCLSTATEMUTEXPTR, mp);
4977 } while (igotlock == 0 && !NFSCL_FORCEDISM(mp));
4980 if (NFSCL_FORCEDISM(mp)) {
4994 * Search for a layout by MDS file handle. If one is found, mark in to be
4995 * recalled, if it already marked "return on close".
4998 nfscl_retoncloselayout(vnode_t vp, struct nfsclclient *clp, uint8_t *fhp,
4999 int fhlen, struct nfsclrecalllayout **recallpp)
5001 struct nfscllayout *lyp;
5004 if (vp->v_type != VREG || !NFSHASPNFS(VFSTONFS(vp->v_mount)) ||
5005 nfscl_enablecallb == 0 || nfs_numnfscbd == 0 ||
5006 (VTONFS(vp)->n_flag & NNOLAYOUT) != 0)
5008 lyp = nfscl_findlayout(clp, fhp, fhlen);
5009 if (lyp != NULL && (lyp->nfsly_flags & (NFSLY_RETONCLOSE |
5010 NFSLY_RECALL)) == NFSLY_RETONCLOSE) {
5012 if (!LIST_EMPTY(&lyp->nfsly_flayread))
5013 iomode |= NFSLAYOUTIOMODE_READ;
5014 if (!LIST_EMPTY(&lyp->nfsly_flayrw))
5015 iomode |= NFSLAYOUTIOMODE_RW;
5016 (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE, lyp, iomode,
5017 0, UINT64_MAX, lyp->nfsly_stateid.seqid, 0, 0, NULL,
5019 NFSCL_DEBUG(4, "retoncls recall iomode=%d\n", iomode);
5025 * Mark the layout to be recalled and with an error.
5026 * Also, disable the dsp from further use.
5029 nfscl_dserr(uint32_t op, uint32_t stat, struct nfscldevinfo *dp,
5030 struct nfscllayout *lyp, struct nfsclds *dsp)
5032 struct nfsclrecalllayout *recallp;
5035 printf("DS being disabled, error=%d\n", stat);
5036 /* Set up the return of the layout. */
5037 recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_WAITOK);
5040 if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
5041 if (!LIST_EMPTY(&lyp->nfsly_flayread))
5042 iomode |= NFSLAYOUTIOMODE_READ;
5043 if (!LIST_EMPTY(&lyp->nfsly_flayrw))
5044 iomode |= NFSLAYOUTIOMODE_RW;
5045 (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE, lyp, iomode,
5046 0, UINT64_MAX, lyp->nfsly_stateid.seqid, stat, op,
5047 dp->nfsdi_deviceid, recallp);
5049 NFSCL_DEBUG(4, "nfscl_dserr recall iomode=%d\n", iomode);
5052 free(recallp, M_NFSLAYRECALL);
5055 /* And shut the TCP connection down. */
5056 nfscl_cancelreqs(dsp);
5060 * Cancel all RPCs for this "dsp" by closing the connection.
5061 * Also, mark the session as defunct.
5062 * If NFSCLDS_SAMECONN is set, the connection is shared with other DSs and
5063 * cannot be shut down.
5066 nfscl_cancelreqs(struct nfsclds *dsp)
5068 struct __rpc_client *cl;
5069 static int non_event;
5072 if ((dsp->nfsclds_flags & (NFSCLDS_CLOSED | NFSCLDS_SAMECONN)) == 0 &&
5073 dsp->nfsclds_sockp != NULL &&
5074 dsp->nfsclds_sockp->nr_client != NULL) {
5075 dsp->nfsclds_flags |= NFSCLDS_CLOSED;
5076 cl = dsp->nfsclds_sockp->nr_client;
5077 dsp->nfsclds_sess.nfsess_defunct = 1;
5081 * This 1sec sleep is done to reduce the number of reconnect
5082 * attempts made on the DS while it has failed.
5084 tsleep(&non_event, PVFS, "ndscls", hz);
5091 * Dereference a layout.
5094 nfscl_rellayout(struct nfscllayout *lyp, int exclocked)
5099 nfsv4_unlock(&lyp->nfsly_lock, 0);
5101 nfsv4_relref(&lyp->nfsly_lock);
5106 * Search for a devinfo by deviceid. If one is found, return it after
5107 * acquiring a reference count on it.
5109 struct nfscldevinfo *
5110 nfscl_getdevinfo(struct nfsclclient *clp, uint8_t *deviceid,
5111 struct nfscldevinfo *dip)
5116 dip = nfscl_finddevinfo(clp, deviceid);
5118 dip->nfsdi_refcnt++;
5124 * Dereference a devinfo structure.
5127 nfscl_reldevinfo_locked(struct nfscldevinfo *dip)
5130 dip->nfsdi_refcnt--;
5131 if (dip->nfsdi_refcnt == 0)
5132 wakeup(&dip->nfsdi_refcnt);
5136 * Dereference a devinfo structure.
5139 nfscl_reldevinfo(struct nfscldevinfo *dip)
5143 nfscl_reldevinfo_locked(dip);
5148 * Find a layout for this file handle. Return NULL upon failure.
5150 static struct nfscllayout *
5151 nfscl_findlayout(struct nfsclclient *clp, u_int8_t *fhp, int fhlen)
5153 struct nfscllayout *lyp;
5155 LIST_FOREACH(lyp, NFSCLLAYOUTHASH(clp, fhp, fhlen), nfsly_hash)
5156 if (lyp->nfsly_fhlen == fhlen &&
5157 !NFSBCMP(lyp->nfsly_fh, fhp, fhlen))
5163 * Find a devinfo for this deviceid. Return NULL upon failure.
5165 static struct nfscldevinfo *
5166 nfscl_finddevinfo(struct nfsclclient *clp, uint8_t *deviceid)
5168 struct nfscldevinfo *dip;
5170 LIST_FOREACH(dip, &clp->nfsc_devinfo, nfsdi_list)
5171 if (NFSBCMP(dip->nfsdi_deviceid, deviceid, NFSX_V4DEVICEID)
5178 * Merge the new file layout list into the main one, maintaining it in
5179 * increasing offset order.
5182 nfscl_mergeflayouts(struct nfsclflayouthead *fhlp,
5183 struct nfsclflayouthead *newfhlp)
5185 struct nfsclflayout *flp, *nflp, *prevflp, *tflp;
5187 flp = LIST_FIRST(fhlp);
5189 LIST_FOREACH_SAFE(nflp, newfhlp, nfsfl_list, tflp) {
5190 while (flp != NULL && flp->nfsfl_off < nflp->nfsfl_off) {
5192 flp = LIST_NEXT(flp, nfsfl_list);
5194 if (prevflp == NULL)
5195 LIST_INSERT_HEAD(fhlp, nflp, nfsfl_list);
5197 LIST_INSERT_AFTER(prevflp, nflp, nfsfl_list);
5203 * Add this nfscldevinfo to the client, if it doesn't already exist.
5204 * This function consumes the structure pointed at by dip, if not NULL.
5207 nfscl_adddevinfo(struct nfsmount *nmp, struct nfscldevinfo *dip, int ind,
5208 struct nfsclflayout *flp)
5210 struct nfsclclient *clp;
5211 struct nfscldevinfo *tdip;
5219 free(dip, M_NFSDEVINFO);
5222 if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5223 dev = flp->nfsfl_dev;
5225 dev = flp->nfsfl_ffm[ind].dev;
5226 tdip = nfscl_finddevinfo(clp, dev);
5228 tdip->nfsdi_layoutrefs++;
5229 if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5230 flp->nfsfl_devp = tdip;
5232 flp->nfsfl_ffm[ind].devp = tdip;
5233 nfscl_reldevinfo_locked(tdip);
5236 free(dip, M_NFSDEVINFO);
5240 LIST_INSERT_HEAD(&clp->nfsc_devinfo, dip, nfsdi_list);
5241 dip->nfsdi_layoutrefs = 1;
5242 if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5243 flp->nfsfl_devp = dip;
5245 flp->nfsfl_ffm[ind].devp = dip;
5254 * Free up a layout structure and associated file layout structure(s).
5257 nfscl_freelayout(struct nfscllayout *layp)
5259 struct nfsclflayout *flp, *nflp;
5260 struct nfsclrecalllayout *rp, *nrp;
5262 LIST_FOREACH_SAFE(flp, &layp->nfsly_flayread, nfsfl_list, nflp) {
5263 LIST_REMOVE(flp, nfsfl_list);
5264 nfscl_freeflayout(flp);
5266 LIST_FOREACH_SAFE(flp, &layp->nfsly_flayrw, nfsfl_list, nflp) {
5267 LIST_REMOVE(flp, nfsfl_list);
5268 nfscl_freeflayout(flp);
5270 LIST_FOREACH_SAFE(rp, &layp->nfsly_recall, nfsrecly_list, nrp) {
5271 LIST_REMOVE(rp, nfsrecly_list);
5272 free(rp, M_NFSLAYRECALL);
5275 free(layp, M_NFSLAYOUT);
5279 * Free up a file layout structure.
5282 nfscl_freeflayout(struct nfsclflayout *flp)
5286 if ((flp->nfsfl_flags & NFSFL_FILE) != 0) {
5287 for (i = 0; i < flp->nfsfl_fhcnt; i++)
5288 free(flp->nfsfl_fh[i], M_NFSFH);
5289 if (flp->nfsfl_devp != NULL)
5290 flp->nfsfl_devp->nfsdi_layoutrefs--;
5292 if ((flp->nfsfl_flags & NFSFL_FLEXFILE) != 0)
5293 for (i = 0; i < flp->nfsfl_mirrorcnt; i++) {
5294 for (j = 0; j < flp->nfsfl_ffm[i].fhcnt; j++)
5295 free(flp->nfsfl_ffm[i].fh[j], M_NFSFH);
5296 if (flp->nfsfl_ffm[i].devp != NULL)
5297 flp->nfsfl_ffm[i].devp->nfsdi_layoutrefs--;
5299 free(flp, M_NFSFLAYOUT);
5303 * Free up a file layout devinfo structure.
5306 nfscl_freedevinfo(struct nfscldevinfo *dip)
5309 free(dip, M_NFSDEVINFO);
5313 * Mark any layouts that match as recalled.
5316 nfscl_layoutrecall(int recalltype, struct nfscllayout *lyp, uint32_t iomode,
5317 uint64_t off, uint64_t len, uint32_t stateseqid, uint32_t stat, uint32_t op,
5318 char *devid, struct nfsclrecalllayout *recallp)
5320 struct nfsclrecalllayout *rp, *orp;
5322 recallp->nfsrecly_recalltype = recalltype;
5323 recallp->nfsrecly_iomode = iomode;
5324 recallp->nfsrecly_stateseqid = stateseqid;
5325 recallp->nfsrecly_off = off;
5326 recallp->nfsrecly_len = len;
5327 recallp->nfsrecly_stat = stat;
5328 recallp->nfsrecly_op = op;
5330 NFSBCOPY(devid, recallp->nfsrecly_devid, NFSX_V4DEVICEID);
5332 * Order the list as file returns first, followed by fsid and any
5333 * returns, both in increasing stateseqid order.
5334 * Note that the seqids wrap around, so 1 is after 0xffffffff.
5335 * (I'm not sure this is correct because I find RFC5661 confusing
5336 * on this, but hopefully it will work ok.)
5339 LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) {
5341 if ((recalltype == NFSLAYOUTRETURN_FILE &&
5342 (rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE ||
5343 nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) ||
5344 (recalltype != NFSLAYOUTRETURN_FILE &&
5345 rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE &&
5346 nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) {
5347 LIST_INSERT_BEFORE(rp, recallp, nfsrecly_list);
5352 * Put any error return on all the file returns that will
5355 if (rp->nfsrecly_recalltype == NFSLAYOUTRETURN_FILE &&
5356 stat != 0 && rp->nfsrecly_stat == 0) {
5357 rp->nfsrecly_stat = stat;
5358 rp->nfsrecly_op = op;
5360 NFSBCOPY(devid, rp->nfsrecly_devid,
5366 LIST_INSERT_HEAD(&lyp->nfsly_recall, recallp,
5369 LIST_INSERT_AFTER(orp, recallp, nfsrecly_list);
5371 lyp->nfsly_flags |= NFSLY_RECALL;
5372 wakeup(lyp->nfsly_clp);
5377 * Compare the two seqids for ordering. The trick is that the seqids can
5378 * wrap around from 0xffffffff->0, so check for the cases where one
5379 * has wrapped around.
5380 * Return 1 if seqid1 comes before seqid2, 0 otherwise.
5383 nfscl_seq(uint32_t seqid1, uint32_t seqid2)
5386 if (seqid2 > seqid1 && (seqid2 - seqid1) >= 0x7fffffff)
5387 /* seqid2 has wrapped around. */
5389 if (seqid1 > seqid2 && (seqid1 - seqid2) >= 0x7fffffff)
5390 /* seqid1 has wrapped around. */
5392 if (seqid1 <= seqid2)
5398 * Do a layout return for each of the recalls.
5401 nfscl_layoutreturn(struct nfsmount *nmp, struct nfscllayout *lyp,
5402 struct ucred *cred, NFSPROC_T *p)
5404 struct nfsclrecalllayout *rp;
5405 nfsv4stateid_t stateid;
5408 NFSBCOPY(lyp->nfsly_stateid.other, stateid.other, NFSX_STATEIDOTHER);
5409 stateid.seqid = lyp->nfsly_stateid.seqid;
5410 if ((lyp->nfsly_flags & NFSLY_FILES) != 0)
5411 layouttype = NFSLAYOUT_NFSV4_1_FILES;
5413 layouttype = NFSLAYOUT_FLEXFILE;
5414 LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) {
5415 (void)nfsrpc_layoutreturn(nmp, lyp->nfsly_fh,
5416 lyp->nfsly_fhlen, 0, layouttype,
5417 rp->nfsrecly_iomode, rp->nfsrecly_recalltype,
5418 rp->nfsrecly_off, rp->nfsrecly_len,
5419 &stateid, cred, p, rp->nfsrecly_stat, rp->nfsrecly_op,
5420 rp->nfsrecly_devid);
5425 * Do the layout commit for a file layout.
5428 nfscl_dolayoutcommit(struct nfsmount *nmp, struct nfscllayout *lyp,
5429 struct ucred *cred, NFSPROC_T *p)
5431 struct nfsclflayout *flp;
5433 int error, layouttype;
5435 if ((lyp->nfsly_flags & NFSLY_FILES) != 0)
5436 layouttype = NFSLAYOUT_NFSV4_1_FILES;
5438 layouttype = NFSLAYOUT_FLEXFILE;
5439 LIST_FOREACH(flp, &lyp->nfsly_flayrw, nfsfl_list) {
5440 if (layouttype == NFSLAYOUT_FLEXFILE &&
5441 (flp->nfsfl_fflags & NFSFLEXFLAG_NO_LAYOUTCOMMIT) != 0) {
5442 NFSCL_DEBUG(4, "Flex file: no layoutcommit\n");
5443 /* If not supported, don't bother doing it. */
5445 nmp->nm_state |= NFSSTA_NOLAYOUTCOMMIT;
5448 } else if (flp->nfsfl_off <= lyp->nfsly_lastbyte) {
5449 len = flp->nfsfl_end - flp->nfsfl_off;
5450 error = nfsrpc_layoutcommit(nmp, lyp->nfsly_fh,
5451 lyp->nfsly_fhlen, 0, flp->nfsfl_off, len,
5452 lyp->nfsly_lastbyte, &lyp->nfsly_stateid,
5453 layouttype, cred, p, NULL);
5454 NFSCL_DEBUG(4, "layoutcommit err=%d\n", error);
5455 if (error == NFSERR_NOTSUPP) {
5456 /* If not supported, don't bother doing it. */
5458 nmp->nm_state |= NFSSTA_NOLAYOUTCOMMIT;
5467 * Commit all layouts for a file (vnode).
5470 nfscl_layoutcommit(vnode_t vp, NFSPROC_T *p)
5472 struct nfsclclient *clp;
5473 struct nfscllayout *lyp;
5474 struct nfsnode *np = VTONFS(vp);
5476 struct nfsmount *nmp;
5480 if (NFSHASNOLAYOUTCOMMIT(nmp))
5488 lyp = nfscl_findlayout(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
5493 nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
5494 if (NFSCL_FORCEDISM(mp)) {
5499 if ((lyp->nfsly_flags & NFSLY_WRITTEN) != 0) {
5500 lyp->nfsly_flags &= ~NFSLY_WRITTEN;
5502 NFSCL_DEBUG(4, "do layoutcommit2\n");
5503 nfscl_dolayoutcommit(clp->nfsc_nmp, lyp, NFSPROCCRED(p), p);
5507 nfsv4_relref(&lyp->nfsly_lock);