xref: /freebsd/sys/fs/nfsclient/nfs_clstate.c (revision b616d997cb48eaafe13069eecd95f0495b2358eb)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2009 Rick Macklem, University of Guelph
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
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.
15  *
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
26  * SUCH DAMAGE.
27  *
28  */
29 
30 #include <sys/cdefs.h>
31 /*
32  * These functions implement the client side state handling for NFSv4.
33  * NFSv4 state handling:
34  * - A lockowner is used to determine lock contention, so it
35  *   corresponds directly to a Posix pid. (1 to 1 mapping)
36  * - The correct granularity of an OpenOwner is not nearly so
37  *   obvious. An OpenOwner does the following:
38  *   - provides a serial sequencing of Open/Close/Lock-with-new-lockowner
39  *   - is used to check for Open/Share contention (not applicable to
40  *     this client, since all Opens are Deny_None)
41  *   As such, I considered both extreme.
42  *   1 OpenOwner per ClientID - Simple to manage, but fully serializes
43  *   all Open, Close and Lock (with a new lockowner) Ops.
44  *   1 OpenOwner for each Open - This one results in an OpenConfirm for
45  *   every Open, for most servers.
46  *   So, I chose to use the same mapping as I did for LockOwnwers.
47  *   The main concern here is that you can end up with multiple Opens
48  *   for the same File Handle, but on different OpenOwners (opens
49  *   inherited from parents, grandparents...) and you do not know
50  *   which of these the vnodeop close applies to. This is handled by
51  *   delaying the Close Op(s) until all of the Opens have been closed.
52  *   (It is not yet obvious if this is the correct granularity.)
53  * - How the code handles serialization:
54  *   - For the ClientId, it uses an exclusive lock while getting its
55  *     SetClientId and during recovery. Otherwise, it uses a shared
56  *     lock via a reference count.
57  *   - For the rest of the data structures, it uses an SMP mutex
58  *     (once the nfs client is SMP safe) and doesn't sleep while
59  *     manipulating the linked lists.
60  *   - The serialization of Open/Close/Lock/LockU falls out in the
61  *     "wash", since OpenOwners and LockOwners are both mapped from
62  *     Posix pid. In other words, there is only one Posix pid using
63  *     any given owner, so that owner is serialized. (If you change
64  *     the granularity of the OpenOwner, then code must be added to
65  *     serialize Ops on the OpenOwner.)
66  * - When to get rid of OpenOwners and LockOwners.
67  *   - The function nfscl_cleanup_common() is executed after a process exits.
68  *     It goes through the client list looking for all Open and Lock Owners.
69  *     When one is found, it is marked "defunct" or in the case of
70  *     an OpenOwner without any Opens, freed.
71  *     The renew thread scans for defunct Owners and gets rid of them,
72  *     if it can. The LockOwners will also be deleted when the
73  *     associated Open is closed.
74  *   - If the LockU or Close Op(s) fail during close in a way
75  *     that could be recovered upon retry, they are relinked to the
76  *     ClientId's defunct open list and retried by the renew thread
77  *     until they succeed or an unmount/recovery occurs.
78  *     (Since we are done with them, they do not need to be recovered.)
79  */
80 
81 #include <fs/nfs/nfsport.h>
82 
83 /*
84  * Global variables
85  */
86 extern struct nfsstatsv1 nfsstatsv1;
87 extern struct nfsreqhead nfsd_reqq;
88 extern u_int32_t newnfs_false, newnfs_true;
89 extern int nfscl_debuglevel;
90 extern int nfscl_enablecallb;
91 extern int nfs_numnfscbd;
92 NFSREQSPINLOCK;
93 NFSCLSTATEMUTEX;
94 int nfscl_inited = 0;
95 struct nfsclhead nfsclhead;	/* Head of clientid list */
96 
97 static int nfscl_getopen(struct nfsclownerhead *, struct nfsclopenhash *,
98     u_int8_t *, int, u_int8_t *, u_int8_t *, u_int32_t,
99     struct nfscllockowner **, struct nfsclopen **);
100 static bool nfscl_checkown(struct nfsclowner *, struct nfsclopen *, uint8_t *,
101     uint8_t *, struct nfscllockowner **, struct nfsclopen **,
102     struct nfsclopen **);
103 static void nfscl_clrelease(struct nfsclclient *);
104 static void nfscl_unlinkopen(struct nfsclopen *);
105 static void nfscl_cleanclient(struct nfsclclient *);
106 static void nfscl_expireclient(struct nfsclclient *, struct nfsmount *,
107     struct ucred *, NFSPROC_T *);
108 static int nfscl_expireopen(struct nfsclclient *, struct nfsclopen *,
109     struct nfsmount *, struct ucred *, NFSPROC_T *);
110 static void nfscl_recover(struct nfsclclient *, bool *, struct ucred *,
111     NFSPROC_T *);
112 static void nfscl_insertlock(struct nfscllockowner *, struct nfscllock *,
113     struct nfscllock *, int);
114 static int nfscl_updatelock(struct nfscllockowner *, struct nfscllock **,
115     struct nfscllock **, int);
116 static void nfscl_delegreturnall(struct nfsclclient *, NFSPROC_T *,
117     struct nfscldeleghead *);
118 static u_int32_t nfscl_nextcbident(void);
119 static mount_t nfscl_getmnt(int, uint8_t *, u_int32_t, struct nfsclclient **);
120 static struct nfsclclient *nfscl_getclnt(u_int32_t);
121 static struct nfsclclient *nfscl_getclntsess(uint8_t *);
122 static struct nfscldeleg *nfscl_finddeleg(struct nfsclclient *, u_int8_t *,
123     int);
124 static void nfscl_retoncloselayout(vnode_t, struct nfsclclient *, uint8_t *,
125     int, struct nfsclrecalllayout **, struct nfscllayout **);
126 static void nfscl_reldevinfo_locked(struct nfscldevinfo *);
127 static struct nfscllayout *nfscl_findlayout(struct nfsclclient *, u_int8_t *,
128     int);
129 static struct nfscldevinfo *nfscl_finddevinfo(struct nfsclclient *, uint8_t *);
130 static int nfscl_checkconflict(struct nfscllockownerhead *, struct nfscllock *,
131     u_int8_t *, struct nfscllock **);
132 static void nfscl_freealllocks(struct nfscllockownerhead *, int);
133 static int nfscl_localconflict(struct nfsclclient *, u_int8_t *, int,
134     struct nfscllock *, u_int8_t *, struct nfscldeleg *, struct nfscllock **);
135 static void nfscl_newopen(struct nfsclclient *, struct nfscldeleg *,
136     struct nfsclowner **, struct nfsclowner **, struct nfsclopen **,
137     struct nfsclopen **, u_int8_t *, u_int8_t *, int, struct ucred *, int *);
138 static int nfscl_moveopen(vnode_t , struct nfsclclient *,
139     struct nfsmount *, struct nfsclopen *, struct nfsclowner *,
140     struct nfscldeleg *, struct ucred *, NFSPROC_T *);
141 static void nfscl_totalrecall(struct nfsclclient *);
142 static int nfscl_relock(vnode_t , struct nfsclclient *, struct nfsmount *,
143     struct nfscllockowner *, struct nfscllock *, struct ucred *, NFSPROC_T *);
144 static int nfscl_tryopen(struct nfsmount *, vnode_t , u_int8_t *, int,
145     u_int8_t *, int, u_int32_t, struct nfsclopen *, u_int8_t *, int,
146     struct nfscldeleg **, int, u_int32_t, struct ucred *, NFSPROC_T *);
147 static int nfscl_trylock(struct nfsmount *, vnode_t , u_int8_t *,
148     int, struct nfscllockowner *, int, int, u_int64_t, u_int64_t, short,
149     struct ucred *, NFSPROC_T *);
150 static int nfsrpc_reopen(struct nfsmount *, u_int8_t *, int, u_int32_t,
151     struct nfsclopen *, struct nfscldeleg **, struct ucred *, NFSPROC_T *);
152 static void nfscl_freedeleg(struct nfscldeleghead *, struct nfscldeleg *,
153     bool);
154 static int nfscl_errmap(struct nfsrv_descript *, u_int32_t);
155 static void nfscl_cleanup_common(struct nfsclclient *, u_int8_t *);
156 static int nfscl_recalldeleg(struct nfsclclient *, struct nfsmount *,
157     struct nfscldeleg *, vnode_t, struct ucred *, NFSPROC_T *, int,
158     vnode_t *);
159 static void nfscl_freeopenowner(struct nfsclowner *, int);
160 static void nfscl_cleandeleg(struct nfscldeleg *);
161 static void nfscl_emptylockowner(struct nfscllockowner *,
162     struct nfscllockownerfhhead *);
163 static void nfscl_mergeflayouts(struct nfsclflayouthead *,
164     struct nfsclflayouthead *);
165 static int nfscl_layoutrecall(int, struct nfscllayout *, uint32_t, uint64_t,
166     uint64_t, uint32_t, uint32_t, uint32_t, char *, struct nfsclrecalllayout *);
167 static int nfscl_seq(uint32_t, uint32_t);
168 static void nfscl_layoutreturn(struct nfsmount *, struct nfscllayout *,
169     struct ucred *, NFSPROC_T *);
170 static void nfscl_dolayoutcommit(struct nfsmount *, struct nfscllayout *,
171     struct ucred *, NFSPROC_T *);
172 
173 static short nfscberr_null[] = {
174 	0,
175 	0,
176 };
177 
178 static short nfscberr_getattr[] = {
179 	NFSERR_RESOURCE,
180 	NFSERR_BADHANDLE,
181 	NFSERR_BADXDR,
182 	NFSERR_RESOURCE,
183 	NFSERR_SERVERFAULT,
184 	0,
185 };
186 
187 static short nfscberr_recall[] = {
188 	NFSERR_RESOURCE,
189 	NFSERR_BADHANDLE,
190 	NFSERR_BADSTATEID,
191 	NFSERR_BADXDR,
192 	NFSERR_RESOURCE,
193 	NFSERR_SERVERFAULT,
194 	0,
195 };
196 
197 static short *nfscl_cberrmap[] = {
198 	nfscberr_null,
199 	nfscberr_null,
200 	nfscberr_null,
201 	nfscberr_getattr,
202 	nfscberr_recall
203 };
204 
205 #define	NETFAMILY(clp) \
206 		(((clp)->nfsc_flags & NFSCLFLAGS_AFINET6) ? AF_INET6 : AF_INET)
207 
208 /*
209  * Called for an open operation.
210  * If the nfhp argument is NULL, just get an openowner.
211  */
212 int
nfscl_open(vnode_t vp,u_int8_t * nfhp,int fhlen,u_int32_t amode,int usedeleg,struct ucred * cred,NFSPROC_T * p,struct nfsclowner ** owpp,struct nfsclopen ** opp,int * newonep,int * retp,int lockit,bool firstref)213 nfscl_open(vnode_t vp, u_int8_t *nfhp, int fhlen, u_int32_t amode, int usedeleg,
214     struct ucred *cred, NFSPROC_T *p, struct nfsclowner **owpp,
215     struct nfsclopen **opp, int *newonep, int *retp, int lockit, bool firstref)
216 {
217 	struct nfsclclient *clp;
218 	struct nfsclowner *owp, *nowp;
219 	struct nfsclopen *op = NULL, *nop = NULL;
220 	struct nfscldeleg *dp;
221 	struct nfsclownerhead *ohp;
222 	u_int8_t own[NFSV4CL_LOCKNAMELEN];
223 	int ret;
224 
225 	if (newonep != NULL)
226 		*newonep = 0;
227 	if (opp != NULL)
228 		*opp = NULL;
229 	if (owpp != NULL)
230 		*owpp = NULL;
231 
232 	/*
233 	 * Might need one or both of these, so MALLOC them now, to
234 	 * avoid a tsleep() in MALLOC later.
235 	 */
236 	nowp = malloc(sizeof (struct nfsclowner),
237 	    M_NFSCLOWNER, M_WAITOK);
238 	if (nfhp != NULL) {
239 	    nop = malloc(sizeof (struct nfsclopen) +
240 		fhlen - 1, M_NFSCLOPEN, M_WAITOK);
241 	    nop->nfso_hash.le_prev = NULL;
242 	}
243 	ret = nfscl_getcl(vp->v_mount, cred, p, false, firstref, &clp);
244 	if (ret != 0) {
245 		free(nowp, M_NFSCLOWNER);
246 		if (nop != NULL)
247 			free(nop, M_NFSCLOPEN);
248 		return (ret);
249 	}
250 
251 	/*
252 	 * Get the Open iff it already exists.
253 	 * If none found, add the new one or return error, depending upon
254 	 * "create".
255 	 */
256 	NFSLOCKCLSTATE();
257 	dp = NULL;
258 	/* First check the delegation list */
259 	if (nfhp != NULL && usedeleg) {
260 		LIST_FOREACH(dp, NFSCLDELEGHASH(clp, nfhp, fhlen), nfsdl_hash) {
261 			if (dp->nfsdl_fhlen == fhlen &&
262 			    !NFSBCMP(nfhp, dp->nfsdl_fh, fhlen)) {
263 				if (!(amode & NFSV4OPEN_ACCESSWRITE) ||
264 				    (dp->nfsdl_flags & NFSCLDL_WRITE))
265 					break;
266 				dp = NULL;
267 				break;
268 			}
269 		}
270 	}
271 
272 	/* For NFSv4.1/4.2 and this option, use a single open_owner. */
273 	if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount)))
274 		nfscl_filllockowner(NULL, own, F_POSIX);
275 	else
276 		nfscl_filllockowner(p->td_proc, own, F_POSIX);
277 	if (dp != NULL)
278 		ohp = &dp->nfsdl_owner;
279 	else
280 		ohp = &clp->nfsc_owner;
281 	/* Now, search for an openowner */
282 	LIST_FOREACH(owp, ohp, nfsow_list) {
283 		if (!NFSBCMP(owp->nfsow_owner, own, NFSV4CL_LOCKNAMELEN))
284 			break;
285 	}
286 
287 	/*
288 	 * Create a new open, as required.
289 	 */
290 	nfscl_newopen(clp, dp, &owp, &nowp, &op, &nop, own, nfhp, fhlen,
291 	    cred, newonep);
292 
293 	/*
294 	 * Now, check the mode on the open and return the appropriate
295 	 * value.
296 	 */
297 	if (retp != NULL) {
298 		if (nfhp != NULL && dp != NULL && nop == NULL)
299 			/* new local open on delegation */
300 			*retp = NFSCLOPEN_SETCRED;
301 		else
302 			*retp = NFSCLOPEN_OK;
303 	}
304 	if (op != NULL && (amode & ~(op->nfso_mode))) {
305 		op->nfso_mode |= amode;
306 		if (retp != NULL && dp == NULL)
307 			*retp = NFSCLOPEN_DOOPEN;
308 	}
309 
310 	/*
311 	 * Serialize modifications to the open owner for multiple threads
312 	 * within the same process using a read/write sleep lock.
313 	 * For NFSv4.1 and a single OpenOwner, allow concurrent open operations
314 	 * by acquiring a shared lock.  The close operations still use an
315 	 * exclusive lock for this case.
316 	 */
317 	if (lockit != 0) {
318 		if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount))) {
319 			/*
320 			 * Get a shared lock on the OpenOwner, but first
321 			 * wait for any pending exclusive lock, so that the
322 			 * exclusive locker gets priority.
323 			 */
324 			nfsv4_lock(&owp->nfsow_rwlock, 0, NULL,
325 			    NFSCLSTATEMUTEXPTR, NULL);
326 			nfsv4_getref(&owp->nfsow_rwlock, NULL,
327 			    NFSCLSTATEMUTEXPTR, NULL);
328 		} else
329 			nfscl_lockexcl(&owp->nfsow_rwlock, NFSCLSTATEMUTEXPTR);
330 	}
331 	NFSUNLOCKCLSTATE();
332 	if (nowp != NULL)
333 		free(nowp, M_NFSCLOWNER);
334 	if (nop != NULL)
335 		free(nop, M_NFSCLOPEN);
336 	if (owpp != NULL)
337 		*owpp = owp;
338 	if (opp != NULL)
339 		*opp = op;
340 	return (0);
341 }
342 
343 /*
344  * Create a new open, as required.
345  */
346 static void
nfscl_newopen(struct nfsclclient * clp,struct nfscldeleg * dp,struct nfsclowner ** owpp,struct nfsclowner ** nowpp,struct nfsclopen ** opp,struct nfsclopen ** nopp,u_int8_t * own,u_int8_t * fhp,int fhlen,struct ucred * cred,int * newonep)347 nfscl_newopen(struct nfsclclient *clp, struct nfscldeleg *dp,
348     struct nfsclowner **owpp, struct nfsclowner **nowpp, struct nfsclopen **opp,
349     struct nfsclopen **nopp, u_int8_t *own, u_int8_t *fhp, int fhlen,
350     struct ucred *cred, int *newonep)
351 {
352 	struct nfsclowner *owp = *owpp, *nowp;
353 	struct nfsclopen *op, *nop;
354 
355 	if (nowpp != NULL)
356 		nowp = *nowpp;
357 	else
358 		nowp = NULL;
359 	if (nopp != NULL)
360 		nop = *nopp;
361 	else
362 		nop = NULL;
363 	if (owp == NULL && nowp != NULL) {
364 		NFSBCOPY(own, nowp->nfsow_owner, NFSV4CL_LOCKNAMELEN);
365 		LIST_INIT(&nowp->nfsow_open);
366 		nowp->nfsow_clp = clp;
367 		nowp->nfsow_seqid = 0;
368 		nowp->nfsow_defunct = 0;
369 		nfscl_lockinit(&nowp->nfsow_rwlock);
370 		if (dp != NULL) {
371 			nfsstatsv1.cllocalopenowners++;
372 			LIST_INSERT_HEAD(&dp->nfsdl_owner, nowp, nfsow_list);
373 		} else {
374 			nfsstatsv1.clopenowners++;
375 			LIST_INSERT_HEAD(&clp->nfsc_owner, nowp, nfsow_list);
376 		}
377 		owp = *owpp = nowp;
378 		*nowpp = NULL;
379 		if (newonep != NULL)
380 			*newonep = 1;
381 	}
382 
383 	 /* If an fhp has been specified, create an Open as well. */
384 	if (fhp != NULL) {
385 		/* and look for the correct open, based upon FH */
386 		LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
387 			if (op->nfso_fhlen == fhlen &&
388 			    !NFSBCMP(op->nfso_fh, fhp, fhlen))
389 				break;
390 		}
391 		if (op == NULL && nop != NULL) {
392 			nop->nfso_own = owp;
393 			nop->nfso_mode = 0;
394 			nop->nfso_opencnt = 0;
395 			nop->nfso_posixlock = 1;
396 			nop->nfso_fhlen = fhlen;
397 			NFSBCOPY(fhp, nop->nfso_fh, fhlen);
398 			LIST_INIT(&nop->nfso_lock);
399 			nop->nfso_stateid.seqid = 0;
400 			nop->nfso_stateid.other[0] = 0;
401 			nop->nfso_stateid.other[1] = 0;
402 			nop->nfso_stateid.other[2] = 0;
403 			KASSERT(cred != NULL, ("%s: cred NULL\n", __func__));
404 			newnfs_copyincred(cred, &nop->nfso_cred);
405 			if (dp != NULL) {
406 				TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
407 				TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp,
408 				    nfsdl_list);
409 				dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
410 				nfsstatsv1.cllocalopens++;
411 			} else {
412 				LIST_INSERT_HEAD(NFSCLOPENHASH(clp, fhp, fhlen),
413 				    nop, nfso_hash);
414 				nfsstatsv1.clopens++;
415 			}
416 			LIST_INSERT_HEAD(&owp->nfsow_open, nop, nfso_list);
417 			*opp = nop;
418 			*nopp = NULL;
419 			if (newonep != NULL)
420 				*newonep = 1;
421 		} else {
422 			*opp = op;
423 		}
424 	}
425 }
426 
427 /*
428  * Called to find/add a delegation to a client.
429  */
430 int
nfscl_deleg(mount_t mp,struct nfsclclient * clp,u_int8_t * nfhp,int fhlen,struct ucred * cred,NFSPROC_T * p,struct nfscldeleg * dp)431 nfscl_deleg(mount_t mp, struct nfsclclient *clp, u_int8_t *nfhp,
432     int fhlen, struct ucred *cred, NFSPROC_T *p, struct nfscldeleg *dp)
433 {
434 	struct nfscldeleg *tdp;
435 	struct nfsmount *nmp;
436 
437 	KASSERT(mp != NULL, ("nfscl_deleg: mp NULL"));
438 	nmp = VFSTONFS(mp);
439 
440 	/*
441 	 * Since a delegation might be added to the mount,
442 	 * set NFSMNTP_DELEGISSUED now.  If a delegation already
443 	 * exagain ists, setting this flag is harmless.
444 	 */
445 	NFSLOCKMNT(nmp);
446 	nmp->nm_privflag |= NFSMNTP_DELEGISSUED;
447 	NFSUNLOCKMNT(nmp);
448 
449 	/* Look for the correct deleg, based upon FH */
450 	NFSLOCKCLSTATE();
451 	tdp = nfscl_finddeleg(clp, nfhp, fhlen);
452 	if (tdp == NULL) {
453 		if (dp == NULL) {
454 			NFSUNLOCKCLSTATE();
455 			return (NFSERR_BADSTATEID);
456 		}
457 		TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list);
458 		LIST_INSERT_HEAD(NFSCLDELEGHASH(clp, nfhp, fhlen), dp,
459 		    nfsdl_hash);
460 		dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
461 		nfsstatsv1.cldelegates++;
462 		clp->nfsc_delegcnt++;
463 	} else {
464 		/*
465 		 * A delegation already exists.  If the new one is a Write
466 		 * delegation and the old one a Read delegation, return the
467 		 * Read delegation.  Otherwise, return the new delegation.
468 		 */
469 		if (dp != NULL) {
470 			if ((dp->nfsdl_flags & NFSCLDL_WRITE) != 0 &&
471 			    (tdp->nfsdl_flags & NFSCLDL_READ) != 0) {
472 				TAILQ_REMOVE(&clp->nfsc_deleg, tdp, nfsdl_list);
473 				LIST_REMOVE(tdp, nfsdl_hash);
474 				TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp,
475 				    nfsdl_list);
476 				LIST_INSERT_HEAD(NFSCLDELEGHASH(clp, nfhp,
477 				    fhlen), dp, nfsdl_hash);
478 				dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
479 			} else {
480 				tdp = dp;	/* Return this one. */
481 			}
482 		} else {
483 			tdp = NULL;
484 		}
485 	}
486 	NFSUNLOCKCLSTATE();
487 	if (tdp != NULL) {
488 		nfscl_trydelegreturn(tdp, cred, nmp, p);
489 		free(tdp, M_NFSCLDELEG);
490 	}
491 	return (0);
492 }
493 
494 /*
495  * Find a delegation for this file handle. Return NULL upon failure.
496  */
497 static struct nfscldeleg *
nfscl_finddeleg(struct nfsclclient * clp,u_int8_t * fhp,int fhlen)498 nfscl_finddeleg(struct nfsclclient *clp, u_int8_t *fhp, int fhlen)
499 {
500 	struct nfscldeleg *dp;
501 
502 	LIST_FOREACH(dp, NFSCLDELEGHASH(clp, fhp, fhlen), nfsdl_hash) {
503 	    if (dp->nfsdl_fhlen == fhlen &&
504 		!NFSBCMP(dp->nfsdl_fh, fhp, fhlen))
505 		break;
506 	}
507 	return (dp);
508 }
509 
510 /*
511  * Get a stateid for an I/O operation. First, look for an open and iff
512  * found, return either a lockowner stateid or the open stateid.
513  * If no Open is found, just return error and the special stateid of all zeros.
514  */
515 int
nfscl_getstateid(vnode_t vp,u_int8_t * nfhp,int fhlen,u_int32_t mode,int fords,struct ucred * cred,NFSPROC_T * p,nfsv4stateid_t * stateidp,void ** lckpp)516 nfscl_getstateid(vnode_t vp, u_int8_t *nfhp, int fhlen, u_int32_t mode,
517     int fords, struct ucred *cred, NFSPROC_T *p, nfsv4stateid_t *stateidp,
518     void **lckpp)
519 {
520 	struct nfsclclient *clp;
521 	struct nfsclopen *op = NULL, *top;
522 	struct nfsclopenhash *oph;
523 	struct nfscllockowner *lp;
524 	struct nfscldeleg *dp;
525 	struct nfsnode *np;
526 	struct nfsmount *nmp;
527 	struct nfscred ncr;
528 	u_int8_t own[NFSV4CL_LOCKNAMELEN], lockown[NFSV4CL_LOCKNAMELEN];
529 	int error;
530 	bool done;
531 
532 	*lckpp = NULL;
533 	/*
534 	 * Initially, just set the special stateid of all zeros.
535 	 * (Don't do this for a DS, since the special stateid can't be used.)
536 	 */
537 	if (fords == 0) {
538 		stateidp->seqid = 0;
539 		stateidp->other[0] = 0;
540 		stateidp->other[1] = 0;
541 		stateidp->other[2] = 0;
542 	}
543 	if (vp->v_type != VREG)
544 		return (EISDIR);
545 	np = VTONFS(vp);
546 	nmp = VFSTONFS(vp->v_mount);
547 
548 	/*
549 	 * For "oneopenown" mounts, first check for a cached open in the
550 	 * NFS vnode, that can be used as a stateid.  This can only be
551 	 * done if no delegations have been issued to the mount and no
552 	 * byte range file locking has been done for the file.
553 	 */
554 	if (NFSHASNFSV4N(nmp) && NFSHASONEOPENOWN(nmp) && fords == 0) {
555 		NFSLOCKMNT(nmp);
556 		NFSLOCKNODE(np);
557 		if ((nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0 &&
558 		    (np->n_flag & NMIGHTBELOCKED) == 0 &&
559 		    np->n_openstateid != NULL) {
560 			stateidp->seqid = 0;
561 			stateidp->other[0] =
562 			    np->n_openstateid->nfso_stateid.other[0];
563 			stateidp->other[1] =
564 			    np->n_openstateid->nfso_stateid.other[1];
565 			stateidp->other[2] =
566 			    np->n_openstateid->nfso_stateid.other[2];
567 			NFSUNLOCKNODE(np);
568 			NFSUNLOCKMNT(nmp);
569 			return (0);
570 		}
571 		NFSUNLOCKNODE(np);
572 		NFSUNLOCKMNT(nmp);
573 	}
574 
575 	NFSLOCKCLSTATE();
576 	clp = nfscl_findcl(nmp);
577 	if (clp == NULL) {
578 		NFSUNLOCKCLSTATE();
579 		return (EACCES);
580 	}
581 
582 	/*
583 	 * Wait for recovery to complete.
584 	 */
585 	while ((clp->nfsc_flags & NFSCLFLAGS_RECVRINPROG))
586 		(void) nfsmsleep(&clp->nfsc_flags, NFSCLSTATEMUTEXPTR,
587 		    PZERO, "nfsrecvr", NULL);
588 
589 	/*
590 	 * First, look for a delegation.
591 	 */
592 	LIST_FOREACH(dp, NFSCLDELEGHASH(clp, nfhp, fhlen), nfsdl_hash) {
593 		if (dp->nfsdl_fhlen == fhlen &&
594 		    !NFSBCMP(nfhp, dp->nfsdl_fh, fhlen)) {
595 			if (!(mode & NFSV4OPEN_ACCESSWRITE) ||
596 			    (dp->nfsdl_flags & NFSCLDL_WRITE)) {
597 				if (NFSHASNFSV4N(nmp))
598 					stateidp->seqid = 0;
599 				else
600 					stateidp->seqid =
601 					    dp->nfsdl_stateid.seqid;
602 				stateidp->other[0] = dp->nfsdl_stateid.other[0];
603 				stateidp->other[1] = dp->nfsdl_stateid.other[1];
604 				stateidp->other[2] = dp->nfsdl_stateid.other[2];
605 				if (!(np->n_flag & NDELEGRECALL)) {
606 					TAILQ_REMOVE(&clp->nfsc_deleg, dp,
607 					    nfsdl_list);
608 					TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp,
609 					    nfsdl_list);
610 					dp->nfsdl_timestamp = NFSD_MONOSEC +
611 					    120;
612 					dp->nfsdl_rwlock.nfslock_usecnt++;
613 					*lckpp = (void *)&dp->nfsdl_rwlock;
614 				}
615 				NFSUNLOCKCLSTATE();
616 				return (0);
617 			}
618 			break;
619 		}
620 	}
621 
622 	if (p != NULL) {
623 		/*
624 		 * If p != NULL, we want to search the parentage tree
625 		 * for a matching OpenOwner and use that.
626 		 */
627 		if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount)))
628 			nfscl_filllockowner(NULL, own, F_POSIX);
629 		else
630 			nfscl_filllockowner(p->td_proc, own, F_POSIX);
631 		nfscl_filllockowner(p->td_proc, lockown, F_POSIX);
632 		lp = NULL;
633 		error = nfscl_getopen(NULL, clp->nfsc_openhash, nfhp, fhlen,
634 		    own, lockown, mode, &lp, &op);
635 		if (error == 0 && lp != NULL && fords == 0) {
636 			/* Don't return a lock stateid for a DS. */
637 			if (NFSHASNFSV4N(nmp))
638 				stateidp->seqid = 0;
639 			else
640 				stateidp->seqid = lp->nfsl_stateid.seqid;
641 			stateidp->other[0] =
642 			    lp->nfsl_stateid.other[0];
643 			stateidp->other[1] =
644 			    lp->nfsl_stateid.other[1];
645 			stateidp->other[2] =
646 			    lp->nfsl_stateid.other[2];
647 			NFSUNLOCKCLSTATE();
648 			return (0);
649 		}
650 	}
651 	if (op == NULL) {
652 		/* If not found, just look for any OpenOwner that will work. */
653 		top = NULL;
654 		done = false;
655 		oph = NFSCLOPENHASH(clp, nfhp, fhlen);
656 		LIST_FOREACH(op, oph, nfso_hash) {
657 			if (op->nfso_fhlen == fhlen &&
658 			    !NFSBCMP(op->nfso_fh, nfhp, fhlen)) {
659 				if (top == NULL && (op->nfso_mode &
660 				    NFSV4OPEN_ACCESSWRITE) != 0 &&
661 				    (mode & NFSV4OPEN_ACCESSREAD) != 0)
662 					top = op;
663 				if ((mode & op->nfso_mode) == mode) {
664 					/* LRU order the hash list. */
665 					LIST_REMOVE(op, nfso_hash);
666 					LIST_INSERT_HEAD(oph, op, nfso_hash);
667 					done = true;
668 					break;
669 				}
670 			}
671 		}
672 		if (!done) {
673 			NFSCL_DEBUG(2, "openmode top=%p\n", top);
674 			if (top == NULL || NFSHASOPENMODE(nmp)) {
675 				NFSUNLOCKCLSTATE();
676 				return (ENOENT);
677 			} else
678 				op = top;
679 		}
680 		/*
681 		 * For read aheads or write behinds, use the open cred.
682 		 * A read ahead or write behind is indicated by p == NULL.
683 		 */
684 		if (p == NULL)
685 			memcpy(&ncr, &op->nfso_cred, sizeof(ncr));
686 	}
687 
688 	/*
689 	 * No lock stateid, so return the open stateid.
690 	 */
691 	if (NFSHASNFSV4N(nmp))
692 		stateidp->seqid = 0;
693 	else
694 		stateidp->seqid = op->nfso_stateid.seqid;
695 	stateidp->other[0] = op->nfso_stateid.other[0];
696 	stateidp->other[1] = op->nfso_stateid.other[1];
697 	stateidp->other[2] = op->nfso_stateid.other[2];
698 	NFSUNLOCKCLSTATE();
699 	if (p == NULL)
700 		newnfs_copycred(&ncr, cred);
701 	return (0);
702 }
703 
704 /*
705  * Search for a matching file, mode and, optionally, lockowner.
706  */
707 static int
nfscl_getopen(struct nfsclownerhead * ohp,struct nfsclopenhash * ohashp,u_int8_t * nfhp,int fhlen,u_int8_t * openown,u_int8_t * lockown,u_int32_t mode,struct nfscllockowner ** lpp,struct nfsclopen ** opp)708 nfscl_getopen(struct nfsclownerhead *ohp, struct nfsclopenhash *ohashp,
709     u_int8_t *nfhp, int fhlen, u_int8_t *openown, u_int8_t *lockown,
710     u_int32_t mode, struct nfscllockowner **lpp, struct nfsclopen **opp)
711 {
712 	struct nfsclowner *owp;
713 	struct nfsclopen *op, *rop, *rop2;
714 	struct nfsclopenhash *oph;
715 	bool keep_looping;
716 
717 	KASSERT(ohp == NULL || ohashp == NULL, ("nfscl_getopen: "
718 	    "only one of ohp and ohashp can be set"));
719 	if (lpp != NULL)
720 		*lpp = NULL;
721 	/*
722 	 * rop will be set to the open to be returned. There are three
723 	 * variants of this, all for an open of the correct file:
724 	 * 1 - A match of lockown.
725 	 * 2 - A match of the openown, when no lockown match exists.
726 	 * 3 - A match for any open, if no openown or lockown match exists.
727 	 * Looking for #2 over #3 probably isn't necessary, but since
728 	 * RFC3530 is vague w.r.t. the relationship between openowners and
729 	 * lockowners, I think this is the safer way to go.
730 	 */
731 	rop = NULL;
732 	rop2 = NULL;
733 	keep_looping = true;
734 	/* Search the client list */
735 	if (ohashp == NULL) {
736 		/* Search the local opens on the delegation. */
737 		LIST_FOREACH(owp, ohp, nfsow_list) {
738 			/* and look for the correct open */
739 			LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
740 				if (op->nfso_fhlen == fhlen &&
741 				    !NFSBCMP(op->nfso_fh, nfhp, fhlen)
742 				    && (op->nfso_mode & mode) == mode)
743 					keep_looping = nfscl_checkown(owp, op, openown,
744 					    lockown, lpp, &rop, &rop2);
745 				if (!keep_looping)
746 					break;
747 			}
748 			if (!keep_looping)
749 				break;
750 		}
751 	} else {
752 		/* Search for matching opens on the hash list. */
753 		oph = &ohashp[NFSCLOPENHASHFUNC(nfhp, fhlen)];
754 		LIST_FOREACH(op, oph, nfso_hash) {
755 			if (op->nfso_fhlen == fhlen &&
756 			    !NFSBCMP(op->nfso_fh, nfhp, fhlen)
757 			    && (op->nfso_mode & mode) == mode)
758 				keep_looping = nfscl_checkown(op->nfso_own, op,
759 				    openown, lockown, lpp, &rop, &rop2);
760 			if (!keep_looping) {
761 				/* LRU order the hash list. */
762 				LIST_REMOVE(op, nfso_hash);
763 				LIST_INSERT_HEAD(oph, op, nfso_hash);
764 				break;
765 			}
766 		}
767 	}
768 	if (rop == NULL)
769 		rop = rop2;
770 	if (rop == NULL)
771 		return (EBADF);
772 	*opp = rop;
773 	return (0);
774 }
775 
776 /* Check for an owner match. */
777 static bool
nfscl_checkown(struct nfsclowner * owp,struct nfsclopen * op,uint8_t * openown,uint8_t * lockown,struct nfscllockowner ** lpp,struct nfsclopen ** ropp,struct nfsclopen ** ropp2)778 nfscl_checkown(struct nfsclowner *owp, struct nfsclopen *op, uint8_t *openown,
779     uint8_t *lockown, struct nfscllockowner **lpp, struct nfsclopen **ropp,
780     struct nfsclopen **ropp2)
781 {
782 	struct nfscllockowner *lp;
783 	bool keep_looping;
784 
785 	keep_looping = true;
786 	if (lpp != NULL) {
787 		/* Now look for a matching lockowner. */
788 		LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
789 			if (!NFSBCMP(lp->nfsl_owner, lockown,
790 			    NFSV4CL_LOCKNAMELEN)) {
791 				*lpp = lp;
792 				*ropp = op;
793 				return (false);
794 			}
795 		}
796 	}
797 	if (*ropp == NULL && !NFSBCMP(owp->nfsow_owner, openown,
798 	    NFSV4CL_LOCKNAMELEN)) {
799 		*ropp = op;
800 		if (lpp == NULL)
801 			keep_looping = false;
802 	}
803 	if (*ropp2 == NULL)
804 		*ropp2 = op;
805 	return (keep_looping);
806 }
807 
808 /*
809  * Release use of an open owner. Called when open operations are done
810  * with the open owner.
811  */
812 void
nfscl_ownerrelease(struct nfsmount * nmp,struct nfsclowner * owp,__unused int error,__unused int candelete,int unlocked)813 nfscl_ownerrelease(struct nfsmount *nmp, struct nfsclowner *owp,
814     __unused int error, __unused int candelete, int unlocked)
815 {
816 
817 	if (owp == NULL)
818 		return;
819 	NFSLOCKCLSTATE();
820 	if (unlocked == 0) {
821 		if (NFSHASONEOPENOWN(nmp))
822 			nfsv4_relref(&owp->nfsow_rwlock);
823 		else
824 			nfscl_lockunlock(&owp->nfsow_rwlock);
825 	}
826 	nfscl_clrelease(owp->nfsow_clp);
827 	NFSUNLOCKCLSTATE();
828 }
829 
830 /*
831  * Release use of an open structure under an open owner.
832  */
833 void
nfscl_openrelease(struct nfsmount * nmp,struct nfsclopen * op,int error,int candelete)834 nfscl_openrelease(struct nfsmount *nmp, struct nfsclopen *op, int error,
835     int candelete)
836 {
837 	struct nfsclclient *clp;
838 	struct nfsclowner *owp;
839 
840 	if (op == NULL)
841 		return;
842 	NFSLOCKCLSTATE();
843 	owp = op->nfso_own;
844 	if (NFSHASONEOPENOWN(nmp))
845 		nfsv4_relref(&owp->nfsow_rwlock);
846 	else
847 		nfscl_lockunlock(&owp->nfsow_rwlock);
848 	clp = owp->nfsow_clp;
849 	if (error && candelete && op->nfso_opencnt == 0)
850 		nfscl_freeopen(op, 0, true);
851 	nfscl_clrelease(clp);
852 	NFSUNLOCKCLSTATE();
853 }
854 
855 /*
856  * Called to get a clientid structure. It will optionally lock the
857  * client data structures to do the SetClientId/SetClientId_confirm,
858  * but will release that lock and return the clientid with a reference
859  * count on it.
860  * If the "cred" argument is NULL, a new clientid should not be created.
861  * If the "p" argument is NULL, a SetClientID/SetClientIDConfirm cannot
862  * be done.
863  * It always clpp with a reference count on it, unless returning an error.
864  */
865 int
nfscl_getcl(struct mount * mp,struct ucred * cred,NFSPROC_T * p,bool tryminvers,bool firstref,struct nfsclclient ** clpp)866 nfscl_getcl(struct mount *mp, struct ucred *cred, NFSPROC_T *p,
867     bool tryminvers, bool firstref, struct nfsclclient **clpp)
868 {
869 	struct nfsclclient *clp;
870 	struct nfsclclient *newclp = NULL;
871 	struct nfsmount *nmp;
872 	char uuid[HOSTUUIDLEN];
873 	int igotlock = 0, error, trystalecnt, clidinusedelay, i;
874 	u_int16_t idlen = 0;
875 
876 	nmp = VFSTONFS(mp);
877 	if (cred != NULL) {
878 		getcredhostuuid(cred, uuid, sizeof uuid);
879 		idlen = strlen(uuid);
880 		if (idlen > 0)
881 			idlen += sizeof (u_int64_t);
882 		else
883 			idlen += sizeof (u_int64_t) + 16; /* 16 random bytes */
884 		newclp = malloc(
885 		    sizeof (struct nfsclclient) + idlen - 1, M_NFSCLCLIENT,
886 		    M_WAITOK | M_ZERO);
887 	}
888 	NFSLOCKCLSTATE();
889 	/*
890 	 * If a forced dismount is already in progress, don't
891 	 * allocate a new clientid and get out now. For the case where
892 	 * clp != NULL, this is a harmless optimization.
893 	 */
894 	if (NFSCL_FORCEDISM(mp)) {
895 		NFSUNLOCKCLSTATE();
896 		if (newclp != NULL)
897 			free(newclp, M_NFSCLCLIENT);
898 		return (EBADF);
899 	}
900 	clp = nmp->nm_clp;
901 	if (clp == NULL) {
902 		if (newclp == NULL) {
903 			NFSUNLOCKCLSTATE();
904 			return (EACCES);
905 		}
906 		clp = newclp;
907 		clp->nfsc_idlen = idlen;
908 		LIST_INIT(&clp->nfsc_owner);
909 		TAILQ_INIT(&clp->nfsc_deleg);
910 		TAILQ_INIT(&clp->nfsc_layout);
911 		LIST_INIT(&clp->nfsc_devinfo);
912 		for (i = 0; i < NFSCLDELEGHASHSIZE; i++)
913 			LIST_INIT(&clp->nfsc_deleghash[i]);
914 		for (i = 0; i < NFSCLOPENHASHSIZE; i++)
915 			LIST_INIT(&clp->nfsc_openhash[i]);
916 		for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++)
917 			LIST_INIT(&clp->nfsc_layouthash[i]);
918 		clp->nfsc_flags = NFSCLFLAGS_INITED;
919 		clp->nfsc_delegcnt = 0;
920 		clp->nfsc_deleghighwater = NFSCLDELEGHIGHWATER;
921 		clp->nfsc_layoutcnt = 0;
922 		clp->nfsc_layouthighwater = NFSCLLAYOUTHIGHWATER;
923 		clp->nfsc_clientidrev = 1;
924 		clp->nfsc_cbident = nfscl_nextcbident();
925 		nfscl_fillclid(nmp->nm_clval, uuid, clp->nfsc_id,
926 		    clp->nfsc_idlen);
927 		LIST_INSERT_HEAD(&nfsclhead, clp, nfsc_list);
928 		nmp->nm_clp = clp;
929 		clp->nfsc_nmp = nmp;
930 	} else {
931 		if (newclp != NULL)
932 			free(newclp, M_NFSCLCLIENT);
933 	}
934 	while ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0 && !igotlock &&
935 	    !NFSCL_FORCEDISM(mp))
936 		igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
937 		    NFSCLSTATEMUTEXPTR, mp);
938 	if (igotlock == 0) {
939 		/*
940 		 * Call nfsv4_lock() with "iwantlock == 0" on the firstref so
941 		 * that it will wait for a pending exclusive lock request.
942 		 * This gives the exclusive lock request priority over this
943 		 * shared lock request.
944 		 * An exclusive lock on nfsc_lock is used mainly for server
945 		 * crash recoveries and delegation recalls.
946 		 */
947 		if (firstref)
948 			nfsv4_lock(&clp->nfsc_lock, 0, NULL, NFSCLSTATEMUTEXPTR,
949 			    mp);
950 		nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
951 	}
952 	if (igotlock == 0 && NFSCL_FORCEDISM(mp)) {
953 		/*
954 		 * Both nfsv4_lock() and nfsv4_getref() know to check
955 		 * for NFSCL_FORCEDISM() and return without sleeping to
956 		 * wait for the exclusive lock to be released, since it
957 		 * might be held by nfscl_umount() and we need to get out
958 		 * now for that case and not wait until nfscl_umount()
959 		 * releases it.
960 		 */
961 		NFSUNLOCKCLSTATE();
962 		return (EBADF);
963 	}
964 	NFSUNLOCKCLSTATE();
965 
966 	/*
967 	 * If it needs a clientid, do the setclientid now.
968 	 */
969 	if ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0) {
970 		if (!igotlock)
971 			panic("nfscl_clget");
972 		if (p == NULL || cred == NULL) {
973 			NFSLOCKCLSTATE();
974 			nfsv4_unlock(&clp->nfsc_lock, 0);
975 			NFSUNLOCKCLSTATE();
976 			return (EACCES);
977 		}
978 		/*
979 		 * If RFC3530 Sec. 14.2.33 is taken literally,
980 		 * NFSERR_CLIDINUSE will be returned persistently for the
981 		 * case where a new mount of the same file system is using
982 		 * a different principal. In practice, NFSERR_CLIDINUSE is
983 		 * only returned when there is outstanding unexpired state
984 		 * on the clientid. As such, try for twice the lease
985 		 * interval, if we know what that is. Otherwise, make a
986 		 * wild ass guess.
987 		 * The case of returning NFSERR_STALECLIENTID is far less
988 		 * likely, but might occur if there is a significant delay
989 		 * between doing the SetClientID and SetClientIDConfirm Ops,
990 		 * such that the server throws away the clientid before
991 		 * receiving the SetClientIDConfirm.
992 		 */
993 		if (clp->nfsc_renew > 0)
994 			clidinusedelay = NFSCL_LEASE(clp->nfsc_renew) * 2;
995 		else
996 			clidinusedelay = 120;
997 		trystalecnt = 3;
998 		do {
999 			error = nfsrpc_setclient(nmp, clp, 0, NULL, cred, p);
1000 			if (error == NFSERR_STALECLIENTID ||
1001 			    error == NFSERR_STALEDONTRECOVER ||
1002 			    error == NFSERR_BADSESSION ||
1003 			    error == NFSERR_CLIDINUSE) {
1004 				(void) nfs_catnap(PZERO, error, "nfs_setcl");
1005 			} else if (error == NFSERR_MINORVERMISMATCH &&
1006 			    tryminvers) {
1007 				if (nmp->nm_minorvers > 0)
1008 					nmp->nm_minorvers--;
1009 				else
1010 					tryminvers = false;
1011 			}
1012 		} while (((error == NFSERR_STALECLIENTID ||
1013 		     error == NFSERR_BADSESSION ||
1014 		     error == NFSERR_STALEDONTRECOVER) && --trystalecnt > 0) ||
1015 		    (error == NFSERR_CLIDINUSE && --clidinusedelay > 0) ||
1016 		    (error == NFSERR_MINORVERMISMATCH && tryminvers));
1017 		if (error) {
1018 			NFSLOCKCLSTATE();
1019 			nfsv4_unlock(&clp->nfsc_lock, 0);
1020 			NFSUNLOCKCLSTATE();
1021 			return (error);
1022 		}
1023 		clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
1024 	}
1025 	if (igotlock) {
1026 		NFSLOCKCLSTATE();
1027 		nfsv4_unlock(&clp->nfsc_lock, 1);
1028 		NFSUNLOCKCLSTATE();
1029 	}
1030 
1031 	*clpp = clp;
1032 	return (0);
1033 }
1034 
1035 /*
1036  * Get a reference to a clientid and return it, if valid.
1037  */
1038 struct nfsclclient *
nfscl_findcl(struct nfsmount * nmp)1039 nfscl_findcl(struct nfsmount *nmp)
1040 {
1041 	struct nfsclclient *clp;
1042 
1043 	clp = nmp->nm_clp;
1044 	if (clp == NULL || !(clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID))
1045 		return (NULL);
1046 	return (clp);
1047 }
1048 
1049 /*
1050  * Release the clientid structure. It may be locked or reference counted.
1051  */
1052 static void
nfscl_clrelease(struct nfsclclient * clp)1053 nfscl_clrelease(struct nfsclclient *clp)
1054 {
1055 
1056 	if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK)
1057 		nfsv4_unlock(&clp->nfsc_lock, 0);
1058 	else
1059 		nfsv4_relref(&clp->nfsc_lock);
1060 }
1061 
1062 /*
1063  * External call for nfscl_clrelease.
1064  */
1065 void
nfscl_clientrelease(struct nfsclclient * clp)1066 nfscl_clientrelease(struct nfsclclient *clp)
1067 {
1068 
1069 	NFSLOCKCLSTATE();
1070 	if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK)
1071 		nfsv4_unlock(&clp->nfsc_lock, 0);
1072 	else
1073 		nfsv4_relref(&clp->nfsc_lock);
1074 	NFSUNLOCKCLSTATE();
1075 }
1076 
1077 /*
1078  * Called when wanting to lock a byte region.
1079  */
1080 int
nfscl_getbytelock(vnode_t vp,u_int64_t off,u_int64_t len,short type,struct ucred * cred,NFSPROC_T * p,struct nfsclclient * rclp,int recovery,void * id,int flags,u_int8_t * rownp,u_int8_t * ropenownp,struct nfscllockowner ** lpp,int * newonep,int * donelocallyp)1081 nfscl_getbytelock(vnode_t vp, u_int64_t off, u_int64_t len,
1082     short type, struct ucred *cred, NFSPROC_T *p, struct nfsclclient *rclp,
1083     int recovery, void *id, int flags, u_int8_t *rownp, u_int8_t *ropenownp,
1084     struct nfscllockowner **lpp, int *newonep, int *donelocallyp)
1085 {
1086 	struct nfscllockowner *lp;
1087 	struct nfsclopen *op;
1088 	struct nfsclclient *clp;
1089 	struct nfscllockowner *nlp;
1090 	struct nfscllock *nlop, *otherlop;
1091 	struct nfscldeleg *dp = NULL, *ldp = NULL;
1092 	struct nfscllockownerhead *lhp = NULL;
1093 	struct nfsnode *np;
1094 	u_int8_t own[NFSV4CL_LOCKNAMELEN], *ownp, openown[NFSV4CL_LOCKNAMELEN];
1095 	u_int8_t *openownp;
1096 	int error = 0, ret, donelocally = 0;
1097 	u_int32_t mode;
1098 
1099 	/* For Lock Ops, the open mode doesn't matter, so use 0 to match any. */
1100 	mode = 0;
1101 	np = VTONFS(vp);
1102 	*lpp = NULL;
1103 	lp = NULL;
1104 	*newonep = 0;
1105 	*donelocallyp = 0;
1106 
1107 	/*
1108 	 * Might need these, so MALLOC them now, to
1109 	 * avoid a tsleep() in MALLOC later.
1110 	 */
1111 	nlp = malloc(
1112 	    sizeof (struct nfscllockowner), M_NFSCLLOCKOWNER, M_WAITOK);
1113 	otherlop = malloc(
1114 	    sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1115 	nlop = malloc(
1116 	    sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1117 	nlop->nfslo_type = type;
1118 	nlop->nfslo_first = off;
1119 	if (len == NFS64BITSSET) {
1120 		nlop->nfslo_end = NFS64BITSSET;
1121 	} else {
1122 		nlop->nfslo_end = off + len;
1123 		if (nlop->nfslo_end <= nlop->nfslo_first)
1124 			error = NFSERR_INVAL;
1125 	}
1126 
1127 	if (!error) {
1128 		if (recovery)
1129 			clp = rclp;
1130 		else
1131 			error = nfscl_getcl(vp->v_mount, cred, p, false, true,
1132 			    &clp);
1133 	}
1134 	if (error) {
1135 		free(nlp, M_NFSCLLOCKOWNER);
1136 		free(otherlop, M_NFSCLLOCK);
1137 		free(nlop, M_NFSCLLOCK);
1138 		return (error);
1139 	}
1140 
1141 	op = NULL;
1142 	if (recovery) {
1143 		ownp = rownp;
1144 		openownp = ropenownp;
1145 	} else {
1146 		nfscl_filllockowner(id, own, flags);
1147 		ownp = own;
1148 		if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount)))
1149 			nfscl_filllockowner(NULL, openown, F_POSIX);
1150 		else
1151 			nfscl_filllockowner(p->td_proc, openown, F_POSIX);
1152 		openownp = openown;
1153 	}
1154 	if (!recovery) {
1155 		NFSLOCKCLSTATE();
1156 		/*
1157 		 * First, search for a delegation. If one exists for this file,
1158 		 * the lock can be done locally against it, so long as there
1159 		 * isn't a local lock conflict.
1160 		 */
1161 		ldp = dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
1162 		    np->n_fhp->nfh_len);
1163 		/* Just sanity check for correct type of delegation */
1164 		if (dp != NULL && ((dp->nfsdl_flags &
1165 		    (NFSCLDL_RECALL | NFSCLDL_DELEGRET)) != 0 ||
1166 		     (type == F_WRLCK &&
1167 		      (dp->nfsdl_flags & NFSCLDL_WRITE) == 0)))
1168 			dp = NULL;
1169 	}
1170 	if (dp != NULL) {
1171 		/* Now, find an open and maybe a lockowner. */
1172 		ret = nfscl_getopen(&dp->nfsdl_owner, NULL, np->n_fhp->nfh_fh,
1173 		    np->n_fhp->nfh_len, openownp, ownp, mode, NULL, &op);
1174 		if (ret)
1175 			ret = nfscl_getopen(NULL, clp->nfsc_openhash,
1176 			    np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp,
1177 			    ownp, mode, NULL, &op);
1178 		if (!ret) {
1179 			lhp = &dp->nfsdl_lock;
1180 			TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
1181 			TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list);
1182 			dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
1183 			donelocally = 1;
1184 		} else {
1185 			dp = NULL;
1186 		}
1187 	}
1188 	if (!donelocally) {
1189 		/*
1190 		 * Get the related Open and maybe lockowner.
1191 		 */
1192 		error = nfscl_getopen(NULL, clp->nfsc_openhash,
1193 		    np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp,
1194 		    ownp, mode, &lp, &op);
1195 		if (!error)
1196 			lhp = &op->nfso_lock;
1197 	}
1198 	if (!error && !recovery)
1199 		error = nfscl_localconflict(clp, np->n_fhp->nfh_fh,
1200 		    np->n_fhp->nfh_len, nlop, ownp, ldp, NULL);
1201 	if (error) {
1202 		if (!recovery) {
1203 			nfscl_clrelease(clp);
1204 			NFSUNLOCKCLSTATE();
1205 		}
1206 		free(nlp, M_NFSCLLOCKOWNER);
1207 		free(otherlop, M_NFSCLLOCK);
1208 		free(nlop, M_NFSCLLOCK);
1209 		return (error);
1210 	}
1211 
1212 	/*
1213 	 * Ok, see if a lockowner exists and create one, as required.
1214 	 */
1215 	if (lp == NULL)
1216 		LIST_FOREACH(lp, lhp, nfsl_list) {
1217 			if (!NFSBCMP(lp->nfsl_owner, ownp, NFSV4CL_LOCKNAMELEN))
1218 				break;
1219 		}
1220 	if (lp == NULL) {
1221 		NFSBCOPY(ownp, nlp->nfsl_owner, NFSV4CL_LOCKNAMELEN);
1222 		if (recovery)
1223 			NFSBCOPY(ropenownp, nlp->nfsl_openowner,
1224 			    NFSV4CL_LOCKNAMELEN);
1225 		else
1226 			NFSBCOPY(op->nfso_own->nfsow_owner, nlp->nfsl_openowner,
1227 			    NFSV4CL_LOCKNAMELEN);
1228 		nlp->nfsl_seqid = 0;
1229 		nlp->nfsl_lockflags = flags;
1230 		nlp->nfsl_inprog = NULL;
1231 		nfscl_lockinit(&nlp->nfsl_rwlock);
1232 		LIST_INIT(&nlp->nfsl_lock);
1233 		if (donelocally) {
1234 			nlp->nfsl_open = NULL;
1235 			nfsstatsv1.cllocallockowners++;
1236 		} else {
1237 			nlp->nfsl_open = op;
1238 			nfsstatsv1.cllockowners++;
1239 		}
1240 		LIST_INSERT_HEAD(lhp, nlp, nfsl_list);
1241 		lp = nlp;
1242 		nlp = NULL;
1243 		*newonep = 1;
1244 	}
1245 
1246 	/*
1247 	 * Now, update the byte ranges for locks.
1248 	 */
1249 	ret = nfscl_updatelock(lp, &nlop, &otherlop, donelocally);
1250 	if (!ret)
1251 		donelocally = 1;
1252 	if (donelocally) {
1253 		*donelocallyp = 1;
1254 		if (!recovery)
1255 			nfscl_clrelease(clp);
1256 	} else {
1257 		/*
1258 		 * Serial modifications on the lock owner for multiple threads
1259 		 * for the same process using a read/write lock.
1260 		 */
1261 		if (!recovery)
1262 			nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR);
1263 	}
1264 	if (!recovery)
1265 		NFSUNLOCKCLSTATE();
1266 
1267 	if (nlp)
1268 		free(nlp, M_NFSCLLOCKOWNER);
1269 	if (nlop)
1270 		free(nlop, M_NFSCLLOCK);
1271 	if (otherlop)
1272 		free(otherlop, M_NFSCLLOCK);
1273 
1274 	*lpp = lp;
1275 	return (0);
1276 }
1277 
1278 /*
1279  * Called to unlock a byte range, for LockU.
1280  */
1281 int
nfscl_relbytelock(vnode_t vp,u_int64_t off,u_int64_t len,__unused struct ucred * cred,NFSPROC_T * p,int callcnt,struct nfsclclient * clp,void * id,int flags,struct nfscllockowner ** lpp,int * dorpcp)1282 nfscl_relbytelock(vnode_t vp, u_int64_t off, u_int64_t len,
1283     __unused struct ucred *cred, NFSPROC_T *p, int callcnt,
1284     struct nfsclclient *clp, void *id, int flags,
1285     struct nfscllockowner **lpp, int *dorpcp)
1286 {
1287 	struct nfscllockowner *lp;
1288 	struct nfsclopen *op;
1289 	struct nfscllock *nlop, *other_lop = NULL;
1290 	struct nfscldeleg *dp;
1291 	struct nfsnode *np;
1292 	u_int8_t own[NFSV4CL_LOCKNAMELEN];
1293 	int ret = 0, fnd;
1294 
1295 	np = VTONFS(vp);
1296 	*lpp = NULL;
1297 	*dorpcp = 0;
1298 
1299 	/*
1300 	 * Might need these, so MALLOC them now, to
1301 	 * avoid a tsleep() in MALLOC later.
1302 	 */
1303 	nlop = malloc(
1304 	    sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1305 	nlop->nfslo_type = F_UNLCK;
1306 	nlop->nfslo_first = off;
1307 	if (len == NFS64BITSSET) {
1308 		nlop->nfslo_end = NFS64BITSSET;
1309 	} else {
1310 		nlop->nfslo_end = off + len;
1311 		if (nlop->nfslo_end <= nlop->nfslo_first) {
1312 			free(nlop, M_NFSCLLOCK);
1313 			return (NFSERR_INVAL);
1314 		}
1315 	}
1316 	if (callcnt == 0) {
1317 		other_lop = malloc(
1318 		    sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1319 		*other_lop = *nlop;
1320 	}
1321 	nfscl_filllockowner(id, own, flags);
1322 	dp = NULL;
1323 	NFSLOCKCLSTATE();
1324 	if (callcnt == 0)
1325 		dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
1326 		    np->n_fhp->nfh_len);
1327 
1328 	/*
1329 	 * First, unlock any local regions on a delegation.
1330 	 */
1331 	if (dp != NULL) {
1332 		/* Look for this lockowner. */
1333 		LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
1334 			if (!NFSBCMP(lp->nfsl_owner, own,
1335 			    NFSV4CL_LOCKNAMELEN))
1336 				break;
1337 		}
1338 		if (lp != NULL)
1339 			/* Use other_lop, so nlop is still available */
1340 			(void)nfscl_updatelock(lp, &other_lop, NULL, 1);
1341 	}
1342 
1343 	/*
1344 	 * Now, find a matching open/lockowner that hasn't already been done,
1345 	 * as marked by nfsl_inprog.
1346 	 */
1347 	lp = NULL;
1348 	fnd = 0;
1349 	LIST_FOREACH(op, NFSCLOPENHASH(clp, np->n_fhp->nfh_fh,
1350 	    np->n_fhp->nfh_len), nfso_hash) {
1351 		if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1352 		    !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1353 			LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1354 				if (lp->nfsl_inprog == NULL &&
1355 				    !NFSBCMP(lp->nfsl_owner, own,
1356 				     NFSV4CL_LOCKNAMELEN)) {
1357 					fnd = 1;
1358 					break;
1359 				}
1360 			}
1361 		}
1362 		if (fnd)
1363 			break;
1364 	}
1365 
1366 	if (lp != NULL) {
1367 		ret = nfscl_updatelock(lp, &nlop, NULL, 0);
1368 		if (ret)
1369 			*dorpcp = 1;
1370 		/*
1371 		 * Serial modifications on the lock owner for multiple
1372 		 * threads for the same process using a read/write lock.
1373 		 */
1374 		lp->nfsl_inprog = p;
1375 		nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR);
1376 		*lpp = lp;
1377 	}
1378 	NFSUNLOCKCLSTATE();
1379 	if (nlop)
1380 		free(nlop, M_NFSCLLOCK);
1381 	if (other_lop)
1382 		free(other_lop, M_NFSCLLOCK);
1383 	return (0);
1384 }
1385 
1386 /*
1387  * Release all lockowners marked in progess for this process and file.
1388  */
1389 void
nfscl_releasealllocks(struct nfsclclient * clp,vnode_t vp,NFSPROC_T * p,void * id,int flags)1390 nfscl_releasealllocks(struct nfsclclient *clp, vnode_t vp, NFSPROC_T *p,
1391     void *id, int flags)
1392 {
1393 	struct nfsclopen *op;
1394 	struct nfscllockowner *lp;
1395 	struct nfsnode *np;
1396 	u_int8_t own[NFSV4CL_LOCKNAMELEN];
1397 
1398 	np = VTONFS(vp);
1399 	nfscl_filllockowner(id, own, flags);
1400 	NFSLOCKCLSTATE();
1401 	LIST_FOREACH(op, NFSCLOPENHASH(clp, np->n_fhp->nfh_fh,
1402 	    np->n_fhp->nfh_len), nfso_hash) {
1403 		if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1404 		    !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1405 			LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1406 				if (lp->nfsl_inprog == p &&
1407 				    !NFSBCMP(lp->nfsl_owner, own,
1408 				    NFSV4CL_LOCKNAMELEN)) {
1409 					lp->nfsl_inprog = NULL;
1410 					nfscl_lockunlock(&lp->nfsl_rwlock);
1411 				}
1412 			}
1413 		}
1414 	}
1415 	nfscl_clrelease(clp);
1416 	NFSUNLOCKCLSTATE();
1417 }
1418 
1419 /*
1420  * Called to find out if any bytes within the byte range specified are
1421  * write locked by the calling process. Used to determine if flushing
1422  * is required before a LockU.
1423  * If in doubt, return 1, so the flush will occur.
1424  */
1425 int
nfscl_checkwritelocked(vnode_t vp,struct flock * fl,struct ucred * cred,NFSPROC_T * p,void * id,int flags)1426 nfscl_checkwritelocked(vnode_t vp, struct flock *fl,
1427     struct ucred *cred, NFSPROC_T *p, void *id, int flags)
1428 {
1429 	struct nfscllockowner *lp;
1430 	struct nfsclopen *op;
1431 	struct nfsclclient *clp;
1432 	struct nfscllock *lop;
1433 	struct nfscldeleg *dp;
1434 	struct nfsnode *np;
1435 	u_int64_t off, end;
1436 	u_int8_t own[NFSV4CL_LOCKNAMELEN];
1437 	int error = 0;
1438 
1439 	np = VTONFS(vp);
1440 	switch (fl->l_whence) {
1441 	case SEEK_SET:
1442 	case SEEK_CUR:
1443 		/*
1444 		 * Caller is responsible for adding any necessary offset
1445 		 * when SEEK_CUR is used.
1446 		 */
1447 		off = fl->l_start;
1448 		break;
1449 	case SEEK_END:
1450 		off = np->n_size + fl->l_start;
1451 		break;
1452 	default:
1453 		return (1);
1454 	}
1455 	if (fl->l_len != 0) {
1456 		end = off + fl->l_len;
1457 		if (end < off)
1458 			return (1);
1459 	} else {
1460 		end = NFS64BITSSET;
1461 	}
1462 
1463 	error = nfscl_getcl(vp->v_mount, cred, p, false, true, &clp);
1464 	if (error)
1465 		return (1);
1466 	nfscl_filllockowner(id, own, flags);
1467 	NFSLOCKCLSTATE();
1468 
1469 	/*
1470 	 * First check the delegation locks.
1471 	 */
1472 	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
1473 	if (dp != NULL) {
1474 		/* No need to flush if it is a write delegation. */
1475 		if ((dp->nfsdl_flags & NFSCLDL_WRITE) != 0) {
1476 			nfscl_clrelease(clp);
1477 			NFSUNLOCKCLSTATE();
1478 			return (0);
1479 		}
1480 		LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
1481 			if (!NFSBCMP(lp->nfsl_owner, own,
1482 			    NFSV4CL_LOCKNAMELEN))
1483 				break;
1484 		}
1485 		if (lp != NULL) {
1486 			LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
1487 				if (lop->nfslo_first >= end)
1488 					break;
1489 				if (lop->nfslo_end <= off)
1490 					continue;
1491 				if (lop->nfslo_type == F_WRLCK) {
1492 					nfscl_clrelease(clp);
1493 					NFSUNLOCKCLSTATE();
1494 					return (1);
1495 				}
1496 			}
1497 		}
1498 	}
1499 
1500 	/*
1501 	 * Now, check state against the server.
1502 	 */
1503 	LIST_FOREACH(op, NFSCLOPENHASH(clp, np->n_fhp->nfh_fh,
1504 	    np->n_fhp->nfh_len), nfso_hash) {
1505 		if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1506 		    !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1507 			LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1508 				if (!NFSBCMP(lp->nfsl_owner, own,
1509 				    NFSV4CL_LOCKNAMELEN))
1510 					break;
1511 			}
1512 			if (lp != NULL) {
1513 				LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
1514 					if (lop->nfslo_first >= end)
1515 						break;
1516 					if (lop->nfslo_end <= off)
1517 						continue;
1518 					if (lop->nfslo_type == F_WRLCK) {
1519 						nfscl_clrelease(clp);
1520 						NFSUNLOCKCLSTATE();
1521 						return (1);
1522 					}
1523 				}
1524 			}
1525 		}
1526 	}
1527 	nfscl_clrelease(clp);
1528 	NFSUNLOCKCLSTATE();
1529 	return (0);
1530 }
1531 
1532 /*
1533  * Release a byte range lock owner structure.
1534  */
1535 void
nfscl_lockrelease(struct nfscllockowner * lp,int error,int candelete)1536 nfscl_lockrelease(struct nfscllockowner *lp, int error, int candelete)
1537 {
1538 	struct nfsclclient *clp;
1539 
1540 	if (lp == NULL)
1541 		return;
1542 	NFSLOCKCLSTATE();
1543 	clp = lp->nfsl_open->nfso_own->nfsow_clp;
1544 	if (error != 0 && candelete &&
1545 	    (lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED) == 0)
1546 		nfscl_freelockowner(lp, 0);
1547 	else
1548 		nfscl_lockunlock(&lp->nfsl_rwlock);
1549 	nfscl_clrelease(clp);
1550 	NFSUNLOCKCLSTATE();
1551 }
1552 
1553 /*
1554  * Unlink the open structure.
1555  */
1556 static void
nfscl_unlinkopen(struct nfsclopen * op)1557 nfscl_unlinkopen(struct nfsclopen *op)
1558 {
1559 
1560 	LIST_REMOVE(op, nfso_list);
1561 	if (op->nfso_hash.le_prev != NULL)
1562 		LIST_REMOVE(op, nfso_hash);
1563 }
1564 
1565 /*
1566  * Free up an open structure and any associated byte range lock structures.
1567  */
1568 void
nfscl_freeopen(struct nfsclopen * op,int local,bool unlink)1569 nfscl_freeopen(struct nfsclopen *op, int local, bool unlink)
1570 {
1571 
1572 	if (unlink)
1573 		nfscl_unlinkopen(op);
1574 	nfscl_freealllocks(&op->nfso_lock, local);
1575 	free(op, M_NFSCLOPEN);
1576 	if (local)
1577 		nfsstatsv1.cllocalopens--;
1578 	else
1579 		nfsstatsv1.clopens--;
1580 }
1581 
1582 /*
1583  * Free up all lock owners and associated locks.
1584  */
1585 static void
nfscl_freealllocks(struct nfscllockownerhead * lhp,int local)1586 nfscl_freealllocks(struct nfscllockownerhead *lhp, int local)
1587 {
1588 	struct nfscllockowner *lp, *nlp;
1589 
1590 	LIST_FOREACH_SAFE(lp, lhp, nfsl_list, nlp) {
1591 		if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED))
1592 			panic("nfscllckw");
1593 		nfscl_freelockowner(lp, local);
1594 	}
1595 }
1596 
1597 /*
1598  * Called for an Open when NFSERR_EXPIRED is received from the server.
1599  * If there are no byte range locks nor a Share Deny lost, try to do a
1600  * fresh Open. Otherwise, free the open.
1601  */
1602 static int
nfscl_expireopen(struct nfsclclient * clp,struct nfsclopen * op,struct nfsmount * nmp,struct ucred * cred,NFSPROC_T * p)1603 nfscl_expireopen(struct nfsclclient *clp, struct nfsclopen *op,
1604     struct nfsmount *nmp, struct ucred *cred, NFSPROC_T *p)
1605 {
1606 	struct nfscllockowner *lp;
1607 	struct nfscldeleg *dp;
1608 	int mustdelete = 0, error;
1609 
1610 	/*
1611 	 * Look for any byte range lock(s).
1612 	 */
1613 	LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1614 		if (!LIST_EMPTY(&lp->nfsl_lock)) {
1615 			mustdelete = 1;
1616 			break;
1617 		}
1618 	}
1619 
1620 	/*
1621 	 * If no byte range lock(s) nor a Share deny, try to re-open.
1622 	 */
1623 	if (!mustdelete && (op->nfso_mode & NFSLCK_DENYBITS) == 0) {
1624 		newnfs_copycred(&op->nfso_cred, cred);
1625 		dp = NULL;
1626 		error = nfsrpc_reopen(nmp, op->nfso_fh,
1627 		    op->nfso_fhlen, op->nfso_mode, op, &dp, cred, p);
1628 		if (error) {
1629 			mustdelete = 1;
1630 			if (dp != NULL) {
1631 				free(dp, M_NFSCLDELEG);
1632 				dp = NULL;
1633 			}
1634 		}
1635 		if (dp != NULL)
1636 			nfscl_deleg(nmp->nm_mountp, clp, op->nfso_fh,
1637 			    op->nfso_fhlen, cred, p, dp);
1638 	}
1639 
1640 	/*
1641 	 * If a byte range lock or Share deny or couldn't re-open, free it.
1642 	 */
1643 	if (mustdelete)
1644 		nfscl_freeopen(op, 0, true);
1645 	return (mustdelete);
1646 }
1647 
1648 /*
1649  * Free up an open owner structure.
1650  */
1651 static void
nfscl_freeopenowner(struct nfsclowner * owp,int local)1652 nfscl_freeopenowner(struct nfsclowner *owp, int local)
1653 {
1654 	int owned;
1655 
1656 	/*
1657 	 * Make sure the NFSCLSTATE mutex is held, to avoid races with
1658 	 * calls in nfscl_renewthread() that do not hold a reference
1659 	 * count on the nfsclclient and just the mutex.
1660 	 * The mutex will not be held for calls done with the exclusive
1661 	 * nfsclclient lock held, in particular, nfscl_hasexpired()
1662 	 * and nfscl_recalldeleg() might do this.
1663 	 */
1664 	owned = mtx_owned(NFSCLSTATEMUTEXPTR);
1665 	if (owned == 0)
1666 		NFSLOCKCLSTATE();
1667 	LIST_REMOVE(owp, nfsow_list);
1668 	if (owned == 0)
1669 		NFSUNLOCKCLSTATE();
1670 	free(owp, M_NFSCLOWNER);
1671 	if (local)
1672 		nfsstatsv1.cllocalopenowners--;
1673 	else
1674 		nfsstatsv1.clopenowners--;
1675 }
1676 
1677 /*
1678  * Free up a byte range lock owner structure.
1679  */
1680 void
nfscl_freelockowner(struct nfscllockowner * lp,int local)1681 nfscl_freelockowner(struct nfscllockowner *lp, int local)
1682 {
1683 	struct nfscllock *lop, *nlop;
1684 	int owned;
1685 
1686 	/*
1687 	 * Make sure the NFSCLSTATE mutex is held, to avoid races with
1688 	 * calls in nfscl_renewthread() that do not hold a reference
1689 	 * count on the nfsclclient and just the mutex.
1690 	 * The mutex will not be held for calls done with the exclusive
1691 	 * nfsclclient lock held, in particular, nfscl_hasexpired()
1692 	 * and nfscl_recalldeleg() might do this.
1693 	 */
1694 	owned = mtx_owned(NFSCLSTATEMUTEXPTR);
1695 	if (owned == 0)
1696 		NFSLOCKCLSTATE();
1697 	LIST_REMOVE(lp, nfsl_list);
1698 	if (owned == 0)
1699 		NFSUNLOCKCLSTATE();
1700 	LIST_FOREACH_SAFE(lop, &lp->nfsl_lock, nfslo_list, nlop) {
1701 		nfscl_freelock(lop, local);
1702 	}
1703 	free(lp, M_NFSCLLOCKOWNER);
1704 	if (local)
1705 		nfsstatsv1.cllocallockowners--;
1706 	else
1707 		nfsstatsv1.cllockowners--;
1708 }
1709 
1710 /*
1711  * Free up a byte range lock structure.
1712  */
1713 void
nfscl_freelock(struct nfscllock * lop,int local)1714 nfscl_freelock(struct nfscllock *lop, int local)
1715 {
1716 
1717 	LIST_REMOVE(lop, nfslo_list);
1718 	free(lop, M_NFSCLLOCK);
1719 	if (local)
1720 		nfsstatsv1.cllocallocks--;
1721 	else
1722 		nfsstatsv1.cllocks--;
1723 }
1724 
1725 /*
1726  * Clean out the state related to a delegation.
1727  */
1728 static void
nfscl_cleandeleg(struct nfscldeleg * dp)1729 nfscl_cleandeleg(struct nfscldeleg *dp)
1730 {
1731 	struct nfsclowner *owp, *nowp;
1732 	struct nfsclopen *op;
1733 
1734 	LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) {
1735 		op = LIST_FIRST(&owp->nfsow_open);
1736 		if (op != NULL) {
1737 			if (LIST_NEXT(op, nfso_list) != NULL)
1738 				panic("nfscleandel");
1739 			nfscl_freeopen(op, 1, true);
1740 		}
1741 		nfscl_freeopenowner(owp, 1);
1742 	}
1743 	nfscl_freealllocks(&dp->nfsdl_lock, 1);
1744 }
1745 
1746 /*
1747  * Free a delegation.
1748  */
1749 static void
nfscl_freedeleg(struct nfscldeleghead * hdp,struct nfscldeleg * dp,bool freeit)1750 nfscl_freedeleg(struct nfscldeleghead *hdp, struct nfscldeleg *dp, bool freeit)
1751 {
1752 
1753 	TAILQ_REMOVE(hdp, dp, nfsdl_list);
1754 	LIST_REMOVE(dp, nfsdl_hash);
1755 	dp->nfsdl_clp->nfsc_delegcnt--;
1756 	if (freeit)
1757 		free(dp, M_NFSCLDELEG);
1758 	nfsstatsv1.cldelegates--;
1759 }
1760 
1761 /*
1762  * Free up all state related to this client structure.
1763  */
1764 static void
nfscl_cleanclient(struct nfsclclient * clp)1765 nfscl_cleanclient(struct nfsclclient *clp)
1766 {
1767 	struct nfsclowner *owp, *nowp;
1768 	struct nfsclopen *op, *nop;
1769 	struct nfscllayout *lyp, *nlyp;
1770 	struct nfscldevinfo *dip, *ndip;
1771 
1772 	TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp)
1773 		nfscl_freelayout(lyp);
1774 
1775 	LIST_FOREACH_SAFE(dip, &clp->nfsc_devinfo, nfsdi_list, ndip)
1776 		nfscl_freedevinfo(dip);
1777 
1778 	/* Now, all the OpenOwners, etc. */
1779 	LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1780 		LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
1781 			nfscl_freeopen(op, 0, true);
1782 		}
1783 		nfscl_freeopenowner(owp, 0);
1784 	}
1785 }
1786 
1787 /*
1788  * Called when an NFSERR_EXPIRED is received from the server.
1789  */
1790 static void
nfscl_expireclient(struct nfsclclient * clp,struct nfsmount * nmp,struct ucred * cred,NFSPROC_T * p)1791 nfscl_expireclient(struct nfsclclient *clp, struct nfsmount *nmp,
1792     struct ucred *cred, NFSPROC_T *p)
1793 {
1794 	struct nfsclowner *owp, *nowp, *towp;
1795 	struct nfsclopen *op, *nop, *top;
1796 	struct nfscldeleg *dp, *ndp;
1797 	int ret, printed = 0;
1798 
1799 	/*
1800 	 * First, merge locally issued Opens into the list for the server.
1801 	 */
1802 	dp = TAILQ_FIRST(&clp->nfsc_deleg);
1803 	while (dp != NULL) {
1804 	    ndp = TAILQ_NEXT(dp, nfsdl_list);
1805 	    owp = LIST_FIRST(&dp->nfsdl_owner);
1806 	    while (owp != NULL) {
1807 		nowp = LIST_NEXT(owp, nfsow_list);
1808 		op = LIST_FIRST(&owp->nfsow_open);
1809 		if (op != NULL) {
1810 		    if (LIST_NEXT(op, nfso_list) != NULL)
1811 			panic("nfsclexp");
1812 		    LIST_FOREACH(towp, &clp->nfsc_owner, nfsow_list) {
1813 			if (!NFSBCMP(towp->nfsow_owner, owp->nfsow_owner,
1814 			    NFSV4CL_LOCKNAMELEN))
1815 			    break;
1816 		    }
1817 		    if (towp != NULL) {
1818 			/* Merge opens in */
1819 			LIST_FOREACH(top, &towp->nfsow_open, nfso_list) {
1820 			    if (top->nfso_fhlen == op->nfso_fhlen &&
1821 				!NFSBCMP(top->nfso_fh, op->nfso_fh,
1822 				 op->nfso_fhlen)) {
1823 				top->nfso_mode |= op->nfso_mode;
1824 				top->nfso_opencnt += op->nfso_opencnt;
1825 				break;
1826 			    }
1827 			}
1828 			if (top == NULL) {
1829 			    /* Just add the open to the owner list */
1830 			    LIST_REMOVE(op, nfso_list);
1831 			    op->nfso_own = towp;
1832 			    LIST_INSERT_HEAD(&towp->nfsow_open, op, nfso_list);
1833 			    LIST_INSERT_HEAD(NFSCLOPENHASH(clp, op->nfso_fh,
1834 				op->nfso_fhlen), op, nfso_hash);
1835 			    nfsstatsv1.cllocalopens--;
1836 			    nfsstatsv1.clopens++;
1837 			}
1838 		    } else {
1839 			/* Just add the openowner to the client list */
1840 			LIST_REMOVE(owp, nfsow_list);
1841 			owp->nfsow_clp = clp;
1842 			LIST_INSERT_HEAD(&clp->nfsc_owner, owp, nfsow_list);
1843 			LIST_INSERT_HEAD(NFSCLOPENHASH(clp, op->nfso_fh,
1844 			    op->nfso_fhlen), op, nfso_hash);
1845 			nfsstatsv1.cllocalopenowners--;
1846 			nfsstatsv1.clopenowners++;
1847 			nfsstatsv1.cllocalopens--;
1848 			nfsstatsv1.clopens++;
1849 		    }
1850 		}
1851 		owp = nowp;
1852 	    }
1853 	    if (!printed && !LIST_EMPTY(&dp->nfsdl_lock)) {
1854 		printed = 1;
1855 		printf("nfsv4 expired locks lost\n");
1856 	    }
1857 	    nfscl_cleandeleg(dp);
1858 	    nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
1859 	    dp = ndp;
1860 	}
1861 	if (!TAILQ_EMPTY(&clp->nfsc_deleg))
1862 	    panic("nfsclexp");
1863 
1864 	/*
1865 	 * Now, try and reopen against the server.
1866 	 */
1867 	LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1868 		owp->nfsow_seqid = 0;
1869 		LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
1870 			ret = nfscl_expireopen(clp, op, nmp, cred, p);
1871 			if (ret && !printed) {
1872 				printed = 1;
1873 				printf("nfsv4 expired locks lost\n");
1874 			}
1875 		}
1876 		if (LIST_EMPTY(&owp->nfsow_open))
1877 			nfscl_freeopenowner(owp, 0);
1878 	}
1879 }
1880 
1881 /*
1882  * This function must be called after the process represented by "own" has
1883  * exited. Must be called with CLSTATE lock held.
1884  */
1885 static void
nfscl_cleanup_common(struct nfsclclient * clp,u_int8_t * own)1886 nfscl_cleanup_common(struct nfsclclient *clp, u_int8_t *own)
1887 {
1888 	struct nfsclowner *owp, *nowp;
1889 	struct nfscllockowner *lp;
1890 	struct nfscldeleg *dp;
1891 
1892 	/* First, get rid of local locks on delegations. */
1893 	TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
1894 		LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
1895 		    if (!NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
1896 			if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED))
1897 			    panic("nfscllckw");
1898 			nfscl_freelockowner(lp, 1);
1899 			break;
1900 		    }
1901 		}
1902 	}
1903 	owp = LIST_FIRST(&clp->nfsc_owner);
1904 	while (owp != NULL) {
1905 		nowp = LIST_NEXT(owp, nfsow_list);
1906 		if (!NFSBCMP(owp->nfsow_owner, own,
1907 		    NFSV4CL_LOCKNAMELEN)) {
1908 			/*
1909 			 * If there are children that haven't closed the
1910 			 * file descriptors yet, the opens will still be
1911 			 * here. For that case, let the renew thread clear
1912 			 * out the OpenOwner later.
1913 			 */
1914 			if (LIST_EMPTY(&owp->nfsow_open))
1915 				nfscl_freeopenowner(owp, 0);
1916 			else
1917 				owp->nfsow_defunct = 1;
1918 			break;
1919 		}
1920 		owp = nowp;
1921 	}
1922 }
1923 
1924 /*
1925  * Find open/lock owners for processes that have exited.
1926  */
1927 static void
nfscl_cleanupkext(struct nfsclclient * clp,struct nfscllockownerfhhead * lhp)1928 nfscl_cleanupkext(struct nfsclclient *clp, struct nfscllockownerfhhead *lhp)
1929 {
1930 	struct nfsclowner *owp, *nowp;
1931 	struct nfsclopen *op;
1932 	struct nfscllockowner *lp, *nlp;
1933 	struct nfscldeleg *dp;
1934 	uint8_t own[NFSV4CL_LOCKNAMELEN];
1935 
1936 	/*
1937 	 * All the pidhash locks must be acquired, since they are sx locks
1938 	 * and must be acquired before the mutexes.  The pid(s) that will
1939 	 * be used aren't known yet, so all the locks need to be acquired.
1940 	 * Fortunately, this function is only performed once/sec.
1941 	 */
1942 	pidhash_slockall();
1943 	NFSLOCKCLSTATE();
1944 	LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1945 		LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1946 			LIST_FOREACH_SAFE(lp, &op->nfso_lock, nfsl_list, nlp) {
1947 				if (LIST_EMPTY(&lp->nfsl_lock))
1948 					nfscl_emptylockowner(lp, lhp);
1949 			}
1950 		}
1951 		if (nfscl_procdoesntexist(owp->nfsow_owner)) {
1952 			memcpy(own, owp->nfsow_owner, NFSV4CL_LOCKNAMELEN);
1953 			nfscl_cleanup_common(clp, own);
1954 		}
1955 	}
1956 
1957 	/*
1958 	 * For the single open_owner case, these lock owners need to be
1959 	 * checked to see if they still exist separately.
1960 	 * This is because nfscl_procdoesntexist() never returns true for
1961 	 * the single open_owner so that the above doesn't ever call
1962 	 * nfscl_cleanup_common().
1963 	 */
1964 	TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
1965 		LIST_FOREACH_SAFE(lp, &dp->nfsdl_lock, nfsl_list, nlp) {
1966 			if (nfscl_procdoesntexist(lp->nfsl_owner)) {
1967 				memcpy(own, lp->nfsl_owner,
1968 				    NFSV4CL_LOCKNAMELEN);
1969 				nfscl_cleanup_common(clp, own);
1970 			}
1971 		}
1972 	}
1973 	NFSUNLOCKCLSTATE();
1974 	pidhash_sunlockall();
1975 }
1976 
1977 /*
1978  * Take the empty lock owner and move it to the local lhp list if the
1979  * associated process no longer exists.
1980  */
1981 static void
nfscl_emptylockowner(struct nfscllockowner * lp,struct nfscllockownerfhhead * lhp)1982 nfscl_emptylockowner(struct nfscllockowner *lp,
1983     struct nfscllockownerfhhead *lhp)
1984 {
1985 	struct nfscllockownerfh *lfhp, *mylfhp;
1986 	struct nfscllockowner *nlp;
1987 	int fnd_it;
1988 
1989 	/* If not a Posix lock owner, just return. */
1990 	if ((lp->nfsl_lockflags & F_POSIX) == 0)
1991 		return;
1992 
1993 	fnd_it = 0;
1994 	mylfhp = NULL;
1995 	/*
1996 	 * First, search to see if this lock owner is already in the list.
1997 	 * If it is, then the associated process no longer exists.
1998 	 */
1999 	SLIST_FOREACH(lfhp, lhp, nfslfh_list) {
2000 		if (lfhp->nfslfh_len == lp->nfsl_open->nfso_fhlen &&
2001 		    !NFSBCMP(lfhp->nfslfh_fh, lp->nfsl_open->nfso_fh,
2002 		    lfhp->nfslfh_len))
2003 			mylfhp = lfhp;
2004 		LIST_FOREACH(nlp, &lfhp->nfslfh_lock, nfsl_list)
2005 			if (!NFSBCMP(nlp->nfsl_owner, lp->nfsl_owner,
2006 			    NFSV4CL_LOCKNAMELEN))
2007 				fnd_it = 1;
2008 	}
2009 	/* If not found, check if process still exists. */
2010 	if (fnd_it == 0 && nfscl_procdoesntexist(lp->nfsl_owner) == 0)
2011 		return;
2012 
2013 	/* Move the lock owner over to the local list. */
2014 	if (mylfhp == NULL) {
2015 		mylfhp = malloc(sizeof(struct nfscllockownerfh), M_TEMP,
2016 		    M_NOWAIT);
2017 		if (mylfhp == NULL)
2018 			return;
2019 		mylfhp->nfslfh_len = lp->nfsl_open->nfso_fhlen;
2020 		NFSBCOPY(lp->nfsl_open->nfso_fh, mylfhp->nfslfh_fh,
2021 		    mylfhp->nfslfh_len);
2022 		LIST_INIT(&mylfhp->nfslfh_lock);
2023 		SLIST_INSERT_HEAD(lhp, mylfhp, nfslfh_list);
2024 	}
2025 	LIST_REMOVE(lp, nfsl_list);
2026 	LIST_INSERT_HEAD(&mylfhp->nfslfh_lock, lp, nfsl_list);
2027 }
2028 
2029 static int	fake_global;	/* Used to force visibility of MNTK_UNMOUNTF */
2030 /*
2031  * Called from nfs umount to free up the clientid.
2032  */
2033 void
nfscl_umount(struct nfsmount * nmp,NFSPROC_T * p,struct nfscldeleghead * dhp)2034 nfscl_umount(struct nfsmount *nmp, NFSPROC_T *p, struct nfscldeleghead *dhp)
2035 {
2036 	struct nfsclclient *clp;
2037 	struct ucred *cred;
2038 	int igotlock;
2039 
2040 	/*
2041 	 * For the case that matters, this is the thread that set
2042 	 * MNTK_UNMOUNTF, so it will see it set. The code that follows is
2043 	 * done to ensure that any thread executing nfscl_getcl() after
2044 	 * this time, will see MNTK_UNMOUNTF set. nfscl_getcl() uses the
2045 	 * mutex for NFSLOCKCLSTATE(), so it is "m" for the following
2046 	 * explanation, courtesy of Alan Cox.
2047 	 * What follows is a snippet from Alan Cox's email at:
2048 	 * https://docs.FreeBSD.org/cgi/mid.cgi?BANLkTikR3d65zPHo9==08ZfJ2vmqZucEvw
2049 	 *
2050 	 * 1. Set MNTK_UNMOUNTF
2051 	 * 2. Acquire a standard FreeBSD mutex "m".
2052 	 * 3. Update some data structures.
2053 	 * 4. Release mutex "m".
2054 	 *
2055 	 * Then, other threads that acquire "m" after step 4 has occurred will
2056 	 * see MNTK_UNMOUNTF as set.  But, other threads that beat thread X to
2057 	 * step 2 may or may not see MNTK_UNMOUNTF as set.
2058 	 */
2059 	NFSLOCKCLSTATE();
2060 	if ((nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) {
2061 		fake_global++;
2062 		NFSUNLOCKCLSTATE();
2063 		NFSLOCKCLSTATE();
2064 	}
2065 
2066 	clp = nmp->nm_clp;
2067 	if (clp != NULL) {
2068 		if ((clp->nfsc_flags & NFSCLFLAGS_INITED) == 0)
2069 			panic("nfscl umount");
2070 
2071 		/*
2072 		 * First, handshake with the nfscl renew thread, to terminate
2073 		 * it.
2074 		 */
2075 		clp->nfsc_flags |= NFSCLFLAGS_UMOUNT;
2076 		while (clp->nfsc_flags & NFSCLFLAGS_HASTHREAD)
2077 			(void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT,
2078 			    "nfsclumnt", hz);
2079 
2080 		/*
2081 		 * Now, get the exclusive lock on the client state, so
2082 		 * that no uses of the state are still in progress.
2083 		 */
2084 		do {
2085 			igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
2086 			    NFSCLSTATEMUTEXPTR, NULL);
2087 		} while (!igotlock);
2088 		NFSUNLOCKCLSTATE();
2089 
2090 		/*
2091 		 * Free up all the state. It will expire on the server, but
2092 		 * maybe we should do a SetClientId/SetClientIdConfirm so
2093 		 * the server throws it away?
2094 		 */
2095 		LIST_REMOVE(clp, nfsc_list);
2096 		nfscl_delegreturnall(clp, p, dhp);
2097 		cred = newnfs_getcred();
2098 		if (NFSHASNFSV4N(nmp)) {
2099 			nfsrpc_destroysession(nmp, NULL, cred, p);
2100 			nfsrpc_destroyclient(nmp, clp, cred, p);
2101 		} else
2102 			nfsrpc_setclient(nmp, clp, 0, NULL, cred, p);
2103 		nfscl_cleanclient(clp);
2104 		nmp->nm_clp = NULL;
2105 		NFSFREECRED(cred);
2106 		free(clp, M_NFSCLCLIENT);
2107 	} else
2108 		NFSUNLOCKCLSTATE();
2109 }
2110 
2111 /*
2112  * This function is called when a server replies with NFSERR_STALECLIENTID
2113  * NFSERR_STALESTATEID or NFSERR_BADSESSION. It traverses the clientid lists,
2114  * doing Opens and Locks with reclaim. If these fail, it deletes the
2115  * corresponding state.
2116  */
2117 static void
nfscl_recover(struct nfsclclient * clp,bool * retokp,struct ucred * cred,NFSPROC_T * p)2118 nfscl_recover(struct nfsclclient *clp, bool *retokp, struct ucred *cred,
2119     NFSPROC_T *p)
2120 {
2121 	struct nfsclowner *owp, *nowp;
2122 	struct nfsclopen *op, *nop;
2123 	struct nfscllockowner *lp, *nlp;
2124 	struct nfscllock *lop, *nlop;
2125 	struct nfscldeleg *dp, *ndp, *tdp;
2126 	struct nfsmount *nmp;
2127 	struct ucred *tcred;
2128 	struct nfsclopenhead extra_open;
2129 	struct nfscldeleghead extra_deleg;
2130 	struct nfsreq *rep;
2131 	u_int64_t len;
2132 	u_int32_t delegtype = NFSV4OPEN_DELEGATEWRITE, mode;
2133 	int i, igotlock = 0, error, trycnt, firstlock;
2134 	struct nfscllayout *lyp, *nlyp;
2135 	bool recovered_one;
2136 
2137 	/*
2138 	 * First, lock the client structure, so everyone else will
2139 	 * block when trying to use state.
2140 	 */
2141 	NFSLOCKCLSTATE();
2142 	clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG;
2143 	do {
2144 		igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
2145 		    NFSCLSTATEMUTEXPTR, NULL);
2146 	} while (!igotlock);
2147 	NFSUNLOCKCLSTATE();
2148 
2149 	nmp = clp->nfsc_nmp;
2150 	if (nmp == NULL)
2151 		panic("nfscl recover");
2152 
2153 	/*
2154 	 * For now, just get rid of all layouts. There may be a need
2155 	 * to do LayoutCommit Ops with reclaim == true later.
2156 	 */
2157 	TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp)
2158 		nfscl_freelayout(lyp);
2159 	TAILQ_INIT(&clp->nfsc_layout);
2160 	for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++)
2161 		LIST_INIT(&clp->nfsc_layouthash[i]);
2162 
2163 	trycnt = 5;
2164 	tcred = NULL;
2165 	do {
2166 		error = nfsrpc_setclient(nmp, clp, 1, retokp, cred, p);
2167 	} while ((error == NFSERR_STALECLIENTID ||
2168 	     error == NFSERR_BADSESSION ||
2169 	     error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
2170 	if (error) {
2171 		NFSLOCKCLSTATE();
2172 		clp->nfsc_flags &= ~(NFSCLFLAGS_RECOVER |
2173 		    NFSCLFLAGS_RECVRINPROG);
2174 		wakeup(&clp->nfsc_flags);
2175 		nfsv4_unlock(&clp->nfsc_lock, 0);
2176 		NFSUNLOCKCLSTATE();
2177 		return;
2178 	}
2179 	clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
2180 	clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2181 
2182 	/*
2183 	 * Mark requests already queued on the server, so that they don't
2184 	 * initiate another recovery cycle. Any requests already in the
2185 	 * queue that handle state information will have the old stale
2186 	 * clientid/stateid and will get a NFSERR_STALESTATEID,
2187 	 * NFSERR_STALECLIENTID or NFSERR_BADSESSION reply from the server.
2188 	 * This will be translated to NFSERR_STALEDONTRECOVER when
2189 	 * R_DONTRECOVER is set.
2190 	 */
2191 	NFSLOCKREQ();
2192 	TAILQ_FOREACH(rep, &nfsd_reqq, r_chain) {
2193 		if (rep->r_nmp == nmp)
2194 			rep->r_flags |= R_DONTRECOVER;
2195 	}
2196 	NFSUNLOCKREQ();
2197 
2198 	/*
2199 	 * If nfsrpc_setclient() returns *retokp == true,
2200 	 * no more recovery is needed.
2201 	 */
2202 	if (*retokp)
2203 		goto out;
2204 
2205 	/*
2206 	 * Now, mark all delegations "need reclaim".
2207 	 */
2208 	TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list)
2209 		dp->nfsdl_flags |= NFSCLDL_NEEDRECLAIM;
2210 
2211 	TAILQ_INIT(&extra_deleg);
2212 	LIST_INIT(&extra_open);
2213 	/*
2214 	 * Now traverse the state lists, doing Open and Lock Reclaims.
2215 	 */
2216 	tcred = newnfs_getcred();
2217 	recovered_one = false;
2218 	owp = LIST_FIRST(&clp->nfsc_owner);
2219 	while (owp != NULL) {
2220 	    nowp = LIST_NEXT(owp, nfsow_list);
2221 	    owp->nfsow_seqid = 0;
2222 	    op = LIST_FIRST(&owp->nfsow_open);
2223 	    while (op != NULL) {
2224 		nop = LIST_NEXT(op, nfso_list);
2225 		if (error != NFSERR_NOGRACE && error != NFSERR_BADSESSION) {
2226 		    /* Search for a delegation to reclaim with the open */
2227 		    TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2228 			if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
2229 			    continue;
2230 			if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
2231 			    mode = NFSV4OPEN_ACCESSWRITE;
2232 			    delegtype = NFSV4OPEN_DELEGATEWRITE;
2233 			} else {
2234 			    mode = NFSV4OPEN_ACCESSREAD;
2235 			    delegtype = NFSV4OPEN_DELEGATEREAD;
2236 			}
2237 			if ((op->nfso_mode & mode) == mode &&
2238 			    op->nfso_fhlen == dp->nfsdl_fhlen &&
2239 			    !NFSBCMP(op->nfso_fh, dp->nfsdl_fh, op->nfso_fhlen))
2240 			    break;
2241 		    }
2242 		    ndp = dp;
2243 		    if (dp == NULL)
2244 			delegtype = NFSV4OPEN_DELEGATENONE;
2245 		    newnfs_copycred(&op->nfso_cred, tcred);
2246 		    error = nfscl_tryopen(nmp, NULL, op->nfso_fh,
2247 			op->nfso_fhlen, op->nfso_fh, op->nfso_fhlen,
2248 			op->nfso_mode, op, NULL, 0, &ndp, 1, delegtype,
2249 			tcred, p);
2250 		    if (!error) {
2251 			recovered_one = true;
2252 			/* Handle any replied delegation */
2253 			if (ndp != NULL && ((ndp->nfsdl_flags & NFSCLDL_WRITE)
2254 			    || NFSMNT_RDONLY(nmp->nm_mountp))) {
2255 			    if ((ndp->nfsdl_flags & NFSCLDL_WRITE))
2256 				mode = NFSV4OPEN_ACCESSWRITE;
2257 			    else
2258 				mode = NFSV4OPEN_ACCESSREAD;
2259 			    TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2260 				if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
2261 				    continue;
2262 				if ((op->nfso_mode & mode) == mode &&
2263 				    op->nfso_fhlen == dp->nfsdl_fhlen &&
2264 				    !NFSBCMP(op->nfso_fh, dp->nfsdl_fh,
2265 				    op->nfso_fhlen)) {
2266 				    dp->nfsdl_stateid = ndp->nfsdl_stateid;
2267 				    dp->nfsdl_sizelimit = ndp->nfsdl_sizelimit;
2268 				    dp->nfsdl_ace = ndp->nfsdl_ace;
2269 				    dp->nfsdl_change = ndp->nfsdl_change;
2270 				    dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
2271 				    if ((ndp->nfsdl_flags & NFSCLDL_RECALL))
2272 					dp->nfsdl_flags |= NFSCLDL_RECALL;
2273 				    free(ndp, M_NFSCLDELEG);
2274 				    ndp = NULL;
2275 				    break;
2276 				}
2277 			    }
2278 			}
2279 			if (ndp != NULL)
2280 			    TAILQ_INSERT_HEAD(&extra_deleg, ndp, nfsdl_list);
2281 
2282 			/* and reclaim all byte range locks */
2283 			lp = LIST_FIRST(&op->nfso_lock);
2284 			while (lp != NULL) {
2285 			    nlp = LIST_NEXT(lp, nfsl_list);
2286 			    lp->nfsl_seqid = 0;
2287 			    firstlock = 1;
2288 			    lop = LIST_FIRST(&lp->nfsl_lock);
2289 			    while (lop != NULL) {
2290 				nlop = LIST_NEXT(lop, nfslo_list);
2291 				if (lop->nfslo_end == NFS64BITSSET)
2292 				    len = NFS64BITSSET;
2293 				else
2294 				    len = lop->nfslo_end - lop->nfslo_first;
2295 				error = nfscl_trylock(nmp, NULL,
2296 				    op->nfso_fh, op->nfso_fhlen, lp,
2297 				    firstlock, 1, lop->nfslo_first, len,
2298 				    lop->nfslo_type, tcred, p);
2299 				if (error != 0)
2300 				    nfscl_freelock(lop, 0);
2301 				else
2302 				    firstlock = 0;
2303 				lop = nlop;
2304 			    }
2305 			    /* If no locks, but a lockowner, just delete it. */
2306 			    if (LIST_EMPTY(&lp->nfsl_lock))
2307 				nfscl_freelockowner(lp, 0);
2308 			    lp = nlp;
2309 			}
2310 		    } else if (error == NFSERR_NOGRACE && !recovered_one &&
2311 			NFSHASNFSV4N(nmp)) {
2312 			/*
2313 			 * For NFSv4.1/4.2, the NFSERR_EXPIRED case will
2314 			 * actually end up here, since the client will do
2315 			 * a recovery for NFSERR_BADSESSION, but will get
2316 			 * an NFSERR_NOGRACE reply for the first "reclaim"
2317 			 * attempt.
2318 			 * So, call nfscl_expireclient() to recover the
2319 			 * opens as best we can and then do a reclaim
2320 			 * complete and return.
2321 			 */
2322 			nfsrpc_reclaimcomplete(nmp, cred, p);
2323 			nfscl_expireclient(clp, nmp, tcred, p);
2324 			goto out;
2325 		    }
2326 		}
2327 		if (error != 0 && error != NFSERR_BADSESSION)
2328 		    nfscl_freeopen(op, 0, true);
2329 		op = nop;
2330 	    }
2331 	    owp = nowp;
2332 	}
2333 
2334 	/*
2335 	 * Now, try and get any delegations not yet reclaimed by cobbling
2336 	 * to-gether an appropriate open.
2337 	 */
2338 	nowp = NULL;
2339 	dp = TAILQ_FIRST(&clp->nfsc_deleg);
2340 	while (dp != NULL) {
2341 	    ndp = TAILQ_NEXT(dp, nfsdl_list);
2342 	    if ((dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM)) {
2343 		if (nowp == NULL) {
2344 		    nowp = malloc(
2345 			sizeof (struct nfsclowner), M_NFSCLOWNER, M_WAITOK);
2346 		    /*
2347 		     * Name must be as long an largest possible
2348 		     * NFSV4CL_LOCKNAMELEN. 12 for now.
2349 		     */
2350 		    NFSBCOPY("RECLAIMDELEG", nowp->nfsow_owner,
2351 			NFSV4CL_LOCKNAMELEN);
2352 		    LIST_INIT(&nowp->nfsow_open);
2353 		    nowp->nfsow_clp = clp;
2354 		    nowp->nfsow_seqid = 0;
2355 		    nowp->nfsow_defunct = 0;
2356 		    nfscl_lockinit(&nowp->nfsow_rwlock);
2357 		}
2358 		nop = NULL;
2359 		if (error != NFSERR_NOGRACE && error != NFSERR_BADSESSION) {
2360 		    nop = malloc(sizeof (struct nfsclopen) +
2361 			dp->nfsdl_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
2362 		    nop->nfso_own = nowp;
2363 		    if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
2364 			nop->nfso_mode = NFSV4OPEN_ACCESSWRITE;
2365 			delegtype = NFSV4OPEN_DELEGATEWRITE;
2366 		    } else {
2367 			nop->nfso_mode = NFSV4OPEN_ACCESSREAD;
2368 			delegtype = NFSV4OPEN_DELEGATEREAD;
2369 		    }
2370 		    nop->nfso_opencnt = 0;
2371 		    nop->nfso_posixlock = 1;
2372 		    nop->nfso_fhlen = dp->nfsdl_fhlen;
2373 		    NFSBCOPY(dp->nfsdl_fh, nop->nfso_fh, dp->nfsdl_fhlen);
2374 		    LIST_INIT(&nop->nfso_lock);
2375 		    nop->nfso_stateid.seqid = 0;
2376 		    nop->nfso_stateid.other[0] = 0;
2377 		    nop->nfso_stateid.other[1] = 0;
2378 		    nop->nfso_stateid.other[2] = 0;
2379 		    newnfs_copycred(&dp->nfsdl_cred, tcred);
2380 		    newnfs_copyincred(tcred, &nop->nfso_cred);
2381 		    tdp = NULL;
2382 		    error = nfscl_tryopen(nmp, NULL, nop->nfso_fh,
2383 			nop->nfso_fhlen, nop->nfso_fh, nop->nfso_fhlen,
2384 			nop->nfso_mode, nop, NULL, 0, &tdp, 1,
2385 			delegtype, tcred, p);
2386 		    if (tdp != NULL) {
2387 			if ((tdp->nfsdl_flags & NFSCLDL_WRITE))
2388 			    mode = NFSV4OPEN_ACCESSWRITE;
2389 			else
2390 			    mode = NFSV4OPEN_ACCESSREAD;
2391 			if ((nop->nfso_mode & mode) == mode &&
2392 			    nop->nfso_fhlen == tdp->nfsdl_fhlen &&
2393 			    !NFSBCMP(nop->nfso_fh, tdp->nfsdl_fh,
2394 			    nop->nfso_fhlen)) {
2395 			    dp->nfsdl_stateid = tdp->nfsdl_stateid;
2396 			    dp->nfsdl_sizelimit = tdp->nfsdl_sizelimit;
2397 			    dp->nfsdl_ace = tdp->nfsdl_ace;
2398 			    dp->nfsdl_change = tdp->nfsdl_change;
2399 			    dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
2400 			    if ((tdp->nfsdl_flags & NFSCLDL_RECALL))
2401 				dp->nfsdl_flags |= NFSCLDL_RECALL;
2402 			    free(tdp, M_NFSCLDELEG);
2403 			} else {
2404 			    TAILQ_INSERT_HEAD(&extra_deleg, tdp, nfsdl_list);
2405 			}
2406 		    }
2407 		}
2408 		if (error) {
2409 		    if (nop != NULL)
2410 			free(nop, M_NFSCLOPEN);
2411 		    if (error == NFSERR_NOGRACE && !recovered_one &&
2412 			NFSHASNFSV4N(nmp)) {
2413 			/*
2414 			 * For NFSv4.1/4.2, the NFSERR_EXPIRED case will
2415 			 * actually end up here, since the client will do
2416 			 * a recovery for NFSERR_BADSESSION, but will get
2417 			 * an NFSERR_NOGRACE reply for the first "reclaim"
2418 			 * attempt.
2419 			 * So, call nfscl_expireclient() to recover the
2420 			 * opens as best we can and then do a reclaim
2421 			 * complete and return.
2422 			 */
2423 			nfsrpc_reclaimcomplete(nmp, cred, p);
2424 			nfscl_expireclient(clp, nmp, tcred, p);
2425 			free(nowp, M_NFSCLOWNER);
2426 			goto out;
2427 		    }
2428 		    /*
2429 		     * Couldn't reclaim it, so throw the state
2430 		     * away. Ouch!!
2431 		     */
2432 		    nfscl_cleandeleg(dp);
2433 		    nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
2434 		} else {
2435 		    recovered_one = true;
2436 		    LIST_INSERT_HEAD(&extra_open, nop, nfso_list);
2437 		}
2438 	    }
2439 	    dp = ndp;
2440 	}
2441 
2442 	/*
2443 	 * Now, get rid of extra Opens and Delegations.
2444 	 */
2445 	LIST_FOREACH_SAFE(op, &extra_open, nfso_list, nop) {
2446 		do {
2447 			newnfs_copycred(&op->nfso_cred, tcred);
2448 			error = nfscl_tryclose(op, tcred, nmp, p, true);
2449 			if (error == NFSERR_GRACE)
2450 				(void) nfs_catnap(PZERO, error, "nfsexcls");
2451 		} while (error == NFSERR_GRACE);
2452 		LIST_REMOVE(op, nfso_list);
2453 		free(op, M_NFSCLOPEN);
2454 	}
2455 	if (nowp != NULL)
2456 		free(nowp, M_NFSCLOWNER);
2457 
2458 	TAILQ_FOREACH_SAFE(dp, &extra_deleg, nfsdl_list, ndp) {
2459 		do {
2460 			newnfs_copycred(&dp->nfsdl_cred, tcred);
2461 			error = nfscl_trydelegreturn(dp, tcred, nmp, p);
2462 			if (error == NFSERR_GRACE)
2463 				(void) nfs_catnap(PZERO, error, "nfsexdlg");
2464 		} while (error == NFSERR_GRACE);
2465 		TAILQ_REMOVE(&extra_deleg, dp, nfsdl_list);
2466 		free(dp, M_NFSCLDELEG);
2467 	}
2468 
2469 	/* For NFSv4.1 or later, do a RECLAIM_COMPLETE. */
2470 	if (NFSHASNFSV4N(nmp))
2471 		(void)nfsrpc_reclaimcomplete(nmp, cred, p);
2472 
2473 out:
2474 	NFSLOCKCLSTATE();
2475 	clp->nfsc_flags &= ~NFSCLFLAGS_RECVRINPROG;
2476 	wakeup(&clp->nfsc_flags);
2477 	nfsv4_unlock(&clp->nfsc_lock, 0);
2478 	NFSUNLOCKCLSTATE();
2479 	if (tcred != NULL)
2480 		NFSFREECRED(tcred);
2481 }
2482 
2483 /*
2484  * This function is called when a server replies with NFSERR_EXPIRED.
2485  * It deletes all state for the client and does a fresh SetClientId/confirm.
2486  * XXX Someday it should post a signal to the process(es) that hold the
2487  * state, so they know that lock state has been lost.
2488  */
2489 int
nfscl_hasexpired(struct nfsclclient * clp,u_int32_t clidrev,NFSPROC_T * p)2490 nfscl_hasexpired(struct nfsclclient *clp, u_int32_t clidrev, NFSPROC_T *p)
2491 {
2492 	struct nfsmount *nmp;
2493 	struct ucred *cred;
2494 	int igotlock = 0, error, trycnt;
2495 
2496 	/*
2497 	 * If the clientid has gone away or a new SetClientid has already
2498 	 * been done, just return ok.
2499 	 */
2500 	if (clp == NULL || clidrev != clp->nfsc_clientidrev)
2501 		return (0);
2502 
2503 	/*
2504 	 * First, lock the client structure, so everyone else will
2505 	 * block when trying to use state. Also, use NFSCLFLAGS_EXPIREIT so
2506 	 * that only one thread does the work.
2507 	 */
2508 	NFSLOCKCLSTATE();
2509 	clp->nfsc_flags |= NFSCLFLAGS_EXPIREIT;
2510 	do {
2511 		igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
2512 		    NFSCLSTATEMUTEXPTR, NULL);
2513 	} while (!igotlock && (clp->nfsc_flags & NFSCLFLAGS_EXPIREIT));
2514 	if ((clp->nfsc_flags & NFSCLFLAGS_EXPIREIT) == 0) {
2515 		if (igotlock)
2516 			nfsv4_unlock(&clp->nfsc_lock, 0);
2517 		NFSUNLOCKCLSTATE();
2518 		return (0);
2519 	}
2520 	clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG;
2521 	NFSUNLOCKCLSTATE();
2522 
2523 	nmp = clp->nfsc_nmp;
2524 	if (nmp == NULL)
2525 		panic("nfscl expired");
2526 	cred = newnfs_getcred();
2527 	trycnt = 5;
2528 	do {
2529 		error = nfsrpc_setclient(nmp, clp, 0, NULL, cred, p);
2530 	} while ((error == NFSERR_STALECLIENTID ||
2531 	     error == NFSERR_BADSESSION ||
2532 	     error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
2533 	if (error) {
2534 		NFSLOCKCLSTATE();
2535 		clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2536 	} else {
2537 		/*
2538 		 * Expire the state for the client.
2539 		 */
2540 		nfscl_expireclient(clp, nmp, cred, p);
2541 		NFSLOCKCLSTATE();
2542 		clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
2543 		clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2544 	}
2545 	clp->nfsc_flags &= ~(NFSCLFLAGS_EXPIREIT | NFSCLFLAGS_RECVRINPROG);
2546 	wakeup(&clp->nfsc_flags);
2547 	nfsv4_unlock(&clp->nfsc_lock, 0);
2548 	NFSUNLOCKCLSTATE();
2549 	NFSFREECRED(cred);
2550 	return (error);
2551 }
2552 
2553 /*
2554  * This function inserts a lock in the list after insert_lop.
2555  */
2556 static void
nfscl_insertlock(struct nfscllockowner * lp,struct nfscllock * new_lop,struct nfscllock * insert_lop,int local)2557 nfscl_insertlock(struct nfscllockowner *lp, struct nfscllock *new_lop,
2558     struct nfscllock *insert_lop, int local)
2559 {
2560 
2561 	if ((struct nfscllockowner *)insert_lop == lp)
2562 		LIST_INSERT_HEAD(&lp->nfsl_lock, new_lop, nfslo_list);
2563 	else
2564 		LIST_INSERT_AFTER(insert_lop, new_lop, nfslo_list);
2565 	if (local)
2566 		nfsstatsv1.cllocallocks++;
2567 	else
2568 		nfsstatsv1.cllocks++;
2569 }
2570 
2571 /*
2572  * This function updates the locking for a lock owner and given file. It
2573  * maintains a list of lock ranges ordered on increasing file offset that
2574  * are NFSCLLOCK_READ or NFSCLLOCK_WRITE and non-overlapping (aka POSIX style).
2575  * It always adds new_lop to the list and sometimes uses the one pointed
2576  * at by other_lopp.
2577  * Returns 1 if the locks were modified, 0 otherwise.
2578  */
2579 static int
nfscl_updatelock(struct nfscllockowner * lp,struct nfscllock ** new_lopp,struct nfscllock ** other_lopp,int local)2580 nfscl_updatelock(struct nfscllockowner *lp, struct nfscllock **new_lopp,
2581     struct nfscllock **other_lopp, int local)
2582 {
2583 	struct nfscllock *new_lop = *new_lopp;
2584 	struct nfscllock *lop, *tlop, *ilop;
2585 	struct nfscllock *other_lop;
2586 	int unlock = 0, modified = 0;
2587 	u_int64_t tmp;
2588 
2589 	/*
2590 	 * Work down the list until the lock is merged.
2591 	 */
2592 	if (new_lop->nfslo_type == F_UNLCK)
2593 		unlock = 1;
2594 	ilop = (struct nfscllock *)lp;
2595 	lop = LIST_FIRST(&lp->nfsl_lock);
2596 	while (lop != NULL) {
2597 	    /*
2598 	     * Only check locks for this file that aren't before the start of
2599 	     * new lock's range.
2600 	     */
2601 	    if (lop->nfslo_end >= new_lop->nfslo_first) {
2602 		if (new_lop->nfslo_end < lop->nfslo_first) {
2603 		    /*
2604 		     * If the new lock ends before the start of the
2605 		     * current lock's range, no merge, just insert
2606 		     * the new lock.
2607 		     */
2608 		    break;
2609 		}
2610 		if (new_lop->nfslo_type == lop->nfslo_type ||
2611 		    (new_lop->nfslo_first <= lop->nfslo_first &&
2612 		     new_lop->nfslo_end >= lop->nfslo_end)) {
2613 		    /*
2614 		     * This lock can be absorbed by the new lock/unlock.
2615 		     * This happens when it covers the entire range
2616 		     * of the old lock or is contiguous
2617 		     * with the old lock and is of the same type or an
2618 		     * unlock.
2619 		     */
2620 		    if (new_lop->nfslo_type != lop->nfslo_type ||
2621 			new_lop->nfslo_first != lop->nfslo_first ||
2622 			new_lop->nfslo_end != lop->nfslo_end)
2623 			modified = 1;
2624 		    if (lop->nfslo_first < new_lop->nfslo_first)
2625 			new_lop->nfslo_first = lop->nfslo_first;
2626 		    if (lop->nfslo_end > new_lop->nfslo_end)
2627 			new_lop->nfslo_end = lop->nfslo_end;
2628 		    tlop = lop;
2629 		    lop = LIST_NEXT(lop, nfslo_list);
2630 		    nfscl_freelock(tlop, local);
2631 		    continue;
2632 		}
2633 
2634 		/*
2635 		 * All these cases are for contiguous locks that are not the
2636 		 * same type, so they can't be merged.
2637 		 */
2638 		if (new_lop->nfslo_first <= lop->nfslo_first) {
2639 		    /*
2640 		     * This case is where the new lock overlaps with the
2641 		     * first part of the old lock. Move the start of the
2642 		     * old lock to just past the end of the new lock. The
2643 		     * new lock will be inserted in front of the old, since
2644 		     * ilop hasn't been updated. (We are done now.)
2645 		     */
2646 		    if (lop->nfslo_first != new_lop->nfslo_end) {
2647 			lop->nfslo_first = new_lop->nfslo_end;
2648 			modified = 1;
2649 		    }
2650 		    break;
2651 		}
2652 		if (new_lop->nfslo_end >= lop->nfslo_end) {
2653 		    /*
2654 		     * This case is where the new lock overlaps with the
2655 		     * end of the old lock's range. Move the old lock's
2656 		     * end to just before the new lock's first and insert
2657 		     * the new lock after the old lock.
2658 		     * Might not be done yet, since the new lock could
2659 		     * overlap further locks with higher ranges.
2660 		     */
2661 		    if (lop->nfslo_end != new_lop->nfslo_first) {
2662 			lop->nfslo_end = new_lop->nfslo_first;
2663 			modified = 1;
2664 		    }
2665 		    ilop = lop;
2666 		    lop = LIST_NEXT(lop, nfslo_list);
2667 		    continue;
2668 		}
2669 		/*
2670 		 * The final case is where the new lock's range is in the
2671 		 * middle of the current lock's and splits the current lock
2672 		 * up. Use *other_lopp to handle the second part of the
2673 		 * split old lock range. (We are done now.)
2674 		 * For unlock, we use new_lop as other_lop and tmp, since
2675 		 * other_lop and new_lop are the same for this case.
2676 		 * We noted the unlock case above, so we don't need
2677 		 * new_lop->nfslo_type any longer.
2678 		 */
2679 		tmp = new_lop->nfslo_first;
2680 		if (unlock) {
2681 		    other_lop = new_lop;
2682 		    *new_lopp = NULL;
2683 		} else {
2684 		    other_lop = *other_lopp;
2685 		    *other_lopp = NULL;
2686 		}
2687 		other_lop->nfslo_first = new_lop->nfslo_end;
2688 		other_lop->nfslo_end = lop->nfslo_end;
2689 		other_lop->nfslo_type = lop->nfslo_type;
2690 		lop->nfslo_end = tmp;
2691 		nfscl_insertlock(lp, other_lop, lop, local);
2692 		ilop = lop;
2693 		modified = 1;
2694 		break;
2695 	    }
2696 	    ilop = lop;
2697 	    lop = LIST_NEXT(lop, nfslo_list);
2698 	    if (lop == NULL)
2699 		break;
2700 	}
2701 
2702 	/*
2703 	 * Insert the new lock in the list at the appropriate place.
2704 	 */
2705 	if (!unlock) {
2706 		nfscl_insertlock(lp, new_lop, ilop, local);
2707 		*new_lopp = NULL;
2708 		modified = 1;
2709 	}
2710 	return (modified);
2711 }
2712 
2713 /*
2714  * This function must be run as a kernel thread.
2715  * It does Renew Ops and recovery, when required.
2716  */
2717 void
nfscl_renewthread(struct nfsclclient * clp,NFSPROC_T * p)2718 nfscl_renewthread(struct nfsclclient *clp, NFSPROC_T *p)
2719 {
2720 	struct nfsclowner *owp, *nowp;
2721 	struct nfsclopen *op;
2722 	struct nfscllockowner *lp, *nlp;
2723 	struct nfscldeleghead dh;
2724 	struct nfscldeleg *dp, *ndp;
2725 	struct ucred *cred;
2726 	u_int32_t clidrev;
2727 	int error, cbpathdown, islept, igotlock, ret, clearok;
2728 	uint32_t recover_done_time = 0;
2729 	time_t mytime;
2730 	static time_t prevsec = 0;
2731 	struct nfscllockownerfh *lfhp, *nlfhp;
2732 	struct nfscllockownerfhhead lfh;
2733 	struct nfscllayout *lyp, *nlyp;
2734 	struct nfscldevinfo *dip, *ndip;
2735 	struct nfscllayouthead rlh;
2736 	struct nfsclrecalllayout *recallp;
2737 	struct nfsclds *dsp;
2738 	bool retok;
2739 	struct mount *mp;
2740 	vnode_t vp;
2741 
2742 	cred = newnfs_getcred();
2743 	NFSLOCKCLSTATE();
2744 	clp->nfsc_flags |= NFSCLFLAGS_HASTHREAD;
2745 	mp = clp->nfsc_nmp->nm_mountp;
2746 	NFSUNLOCKCLSTATE();
2747 	for(;;) {
2748 		newnfs_setroot(cred);
2749 		cbpathdown = 0;
2750 		if (clp->nfsc_flags & NFSCLFLAGS_RECOVER) {
2751 			/*
2752 			 * Only allow one full recover within 1/2 of the lease
2753 			 * duration (nfsc_renew).
2754 			 * retok is value/result.  If passed in set to true,
2755 			 * it indicates only a CreateSession operation should
2756 			 * be attempted.
2757 			 * If it is returned true, it indicates that the
2758 			 * recovery only required a CreateSession.
2759 			 */
2760 			retok = true;
2761 			if (recover_done_time < NFSD_MONOSEC) {
2762 				recover_done_time = NFSD_MONOSEC +
2763 				    clp->nfsc_renew;
2764 				retok = false;
2765 			}
2766 			NFSCL_DEBUG(1, "Doing recovery, only "
2767 			    "createsession=%d\n", retok);
2768 			nfscl_recover(clp, &retok, cred, p);
2769 		}
2770 		if (clp->nfsc_expire <= NFSD_MONOSEC &&
2771 		    (clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID)) {
2772 			clp->nfsc_expire = NFSD_MONOSEC + clp->nfsc_renew;
2773 			clidrev = clp->nfsc_clientidrev;
2774 			error = nfsrpc_renew(clp, NULL, cred, p);
2775 			if (error == NFSERR_CBPATHDOWN)
2776 			    cbpathdown = 1;
2777 			else if (error == NFSERR_STALECLIENTID) {
2778 			    NFSLOCKCLSTATE();
2779 			    clp->nfsc_flags |= NFSCLFLAGS_RECOVER;
2780 			    NFSUNLOCKCLSTATE();
2781 			} else if (error == NFSERR_EXPIRED)
2782 			    (void) nfscl_hasexpired(clp, clidrev, p);
2783 		}
2784 
2785 checkdsrenew:
2786 		if (NFSHASNFSV4N(clp->nfsc_nmp)) {
2787 			/* Do renews for any DS sessions. */
2788 			NFSLOCKMNT(clp->nfsc_nmp);
2789 			/* Skip first entry, since the MDS is handled above. */
2790 			dsp = TAILQ_FIRST(&clp->nfsc_nmp->nm_sess);
2791 			if (dsp != NULL)
2792 				dsp = TAILQ_NEXT(dsp, nfsclds_list);
2793 			while (dsp != NULL) {
2794 				if (dsp->nfsclds_expire <= NFSD_MONOSEC &&
2795 				    dsp->nfsclds_sess.nfsess_defunct == 0) {
2796 					dsp->nfsclds_expire = NFSD_MONOSEC +
2797 					    clp->nfsc_renew;
2798 					NFSUNLOCKMNT(clp->nfsc_nmp);
2799 					(void)nfsrpc_renew(clp, dsp, cred, p);
2800 					goto checkdsrenew;
2801 				}
2802 				dsp = TAILQ_NEXT(dsp, nfsclds_list);
2803 			}
2804 			NFSUNLOCKMNT(clp->nfsc_nmp);
2805 		}
2806 
2807 		TAILQ_INIT(&dh);
2808 		NFSLOCKCLSTATE();
2809 		if (cbpathdown)
2810 			/* It's a Total Recall! */
2811 			nfscl_totalrecall(clp);
2812 
2813 		/*
2814 		 * Now, handle defunct owners.
2815 		 */
2816 		LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
2817 			if (LIST_EMPTY(&owp->nfsow_open)) {
2818 				if (owp->nfsow_defunct != 0)
2819 					nfscl_freeopenowner(owp, 0);
2820 			}
2821 		}
2822 
2823 		/*
2824 		 * Do the recall on any delegations. To avoid trouble, always
2825 		 * come back up here after having slept.
2826 		 */
2827 		igotlock = 0;
2828 tryagain:
2829 		dp = TAILQ_FIRST(&clp->nfsc_deleg);
2830 		while (dp != NULL) {
2831 			ndp = TAILQ_NEXT(dp, nfsdl_list);
2832 			if ((dp->nfsdl_flags & NFSCLDL_RECALL)) {
2833 				/*
2834 				 * Wait for outstanding I/O ops to be done.
2835 				 */
2836 				if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
2837 				    if (igotlock) {
2838 					nfsv4_unlock(&clp->nfsc_lock, 0);
2839 					igotlock = 0;
2840 				    }
2841 				    dp->nfsdl_rwlock.nfslock_lock |=
2842 					NFSV4LOCK_WANTED;
2843 				    msleep(&dp->nfsdl_rwlock,
2844 					NFSCLSTATEMUTEXPTR, PVFS, "nfscld",
2845 					5 * hz);
2846 				    if (NFSCL_FORCEDISM(mp))
2847 					goto terminate;
2848 				    goto tryagain;
2849 				}
2850 				while (!igotlock) {
2851 				    igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
2852 					&islept, NFSCLSTATEMUTEXPTR, mp);
2853 				    if (igotlock == 0 && NFSCL_FORCEDISM(mp))
2854 					goto terminate;
2855 				    if (islept)
2856 					goto tryagain;
2857 				}
2858 				NFSUNLOCKCLSTATE();
2859 				newnfs_copycred(&dp->nfsdl_cred, cred);
2860 				ret = nfscl_recalldeleg(clp, clp->nfsc_nmp, dp,
2861 				    NULL, cred, p, 1, &vp);
2862 				if (!ret) {
2863 				    nfscl_cleandeleg(dp);
2864 				    TAILQ_REMOVE(&clp->nfsc_deleg, dp,
2865 					nfsdl_list);
2866 				    LIST_REMOVE(dp, nfsdl_hash);
2867 				    TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
2868 				    clp->nfsc_delegcnt--;
2869 				    nfsstatsv1.cldelegates--;
2870 				}
2871 				NFSLOCKCLSTATE();
2872 				/*
2873 				 * The nfsc_lock must be released before doing
2874 				 * vrele(), since it might call nfs_inactive().
2875 				 * For the unlikely case where the vnode failed
2876 				 * to be acquired by nfscl_recalldeleg(), a
2877 				 * VOP_RECLAIM() should be in progress and it
2878 				 * will return the delegation.
2879 				 */
2880 				nfsv4_unlock(&clp->nfsc_lock, 0);
2881 				igotlock = 0;
2882 				if (vp != NULL) {
2883 					NFSUNLOCKCLSTATE();
2884 					vrele(vp);
2885 					NFSLOCKCLSTATE();
2886 				}
2887 				goto tryagain;
2888 			}
2889 			dp = ndp;
2890 		}
2891 
2892 		/*
2893 		 * Clear out old delegations, if we are above the high water
2894 		 * mark. Only clear out ones with no state related to them.
2895 		 * The tailq list is in LRU order.
2896 		 */
2897 		dp = TAILQ_LAST(&clp->nfsc_deleg, nfscldeleghead);
2898 		while (clp->nfsc_delegcnt > clp->nfsc_deleghighwater &&
2899 		    dp != NULL) {
2900 		    ndp = TAILQ_PREV(dp, nfscldeleghead, nfsdl_list);
2901 		    if (dp->nfsdl_rwlock.nfslock_usecnt == 0 &&
2902 			dp->nfsdl_rwlock.nfslock_lock == 0 &&
2903 			dp->nfsdl_timestamp < NFSD_MONOSEC &&
2904 			(dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_ZAPPED |
2905 			  NFSCLDL_NEEDRECLAIM | NFSCLDL_DELEGRET)) == 0) {
2906 			clearok = 1;
2907 			LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
2908 			    op = LIST_FIRST(&owp->nfsow_open);
2909 			    if (op != NULL) {
2910 				clearok = 0;
2911 				break;
2912 			    }
2913 			}
2914 			if (clearok) {
2915 			    LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
2916 				if (!LIST_EMPTY(&lp->nfsl_lock)) {
2917 				    clearok = 0;
2918 				    break;
2919 				}
2920 			    }
2921 			}
2922 			if (clearok) {
2923 			    TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
2924 			    LIST_REMOVE(dp, nfsdl_hash);
2925 			    TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
2926 			    clp->nfsc_delegcnt--;
2927 			    nfsstatsv1.cldelegates--;
2928 			}
2929 		    }
2930 		    dp = ndp;
2931 		}
2932 		if (igotlock)
2933 			nfsv4_unlock(&clp->nfsc_lock, 0);
2934 
2935 		/*
2936 		 * Do the recall on any layouts. To avoid trouble, always
2937 		 * come back up here after having slept.
2938 		 */
2939 		TAILQ_INIT(&rlh);
2940 tryagain2:
2941 		TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp) {
2942 			if ((lyp->nfsly_flags & NFSLY_RECALL) != 0) {
2943 				/*
2944 				 * Wait for outstanding I/O ops to be done.
2945 				 */
2946 				if (lyp->nfsly_lock.nfslock_usecnt > 0 ||
2947 				    (lyp->nfsly_lock.nfslock_lock &
2948 				     NFSV4LOCK_LOCK) != 0) {
2949 					lyp->nfsly_lock.nfslock_lock |=
2950 					    NFSV4LOCK_WANTED;
2951 					msleep(&lyp->nfsly_lock.nfslock_lock,
2952 					    NFSCLSTATEMUTEXPTR, PVFS, "nfslyp",
2953 					    5 * hz);
2954 					if (NFSCL_FORCEDISM(mp))
2955 					    goto terminate;
2956 					goto tryagain2;
2957 				}
2958 				/* Move the layout to the recall list. */
2959 				TAILQ_REMOVE(&clp->nfsc_layout, lyp,
2960 				    nfsly_list);
2961 				LIST_REMOVE(lyp, nfsly_hash);
2962 				TAILQ_INSERT_HEAD(&rlh, lyp, nfsly_list);
2963 
2964 				/* Handle any layout commits. */
2965 				if (!NFSHASNOLAYOUTCOMMIT(clp->nfsc_nmp) &&
2966 				    (lyp->nfsly_flags & NFSLY_WRITTEN) != 0) {
2967 					lyp->nfsly_flags &= ~NFSLY_WRITTEN;
2968 					NFSUNLOCKCLSTATE();
2969 					NFSCL_DEBUG(3, "do layoutcommit\n");
2970 					nfscl_dolayoutcommit(clp->nfsc_nmp, lyp,
2971 					    cred, p);
2972 					NFSLOCKCLSTATE();
2973 					goto tryagain2;
2974 				}
2975 			}
2976 		}
2977 
2978 		/* Now, look for stale layouts. */
2979 		lyp = TAILQ_LAST(&clp->nfsc_layout, nfscllayouthead);
2980 		while (lyp != NULL) {
2981 			nlyp = TAILQ_PREV(lyp, nfscllayouthead, nfsly_list);
2982 			if ((lyp->nfsly_timestamp < NFSD_MONOSEC ||
2983 			     clp->nfsc_layoutcnt > clp->nfsc_layouthighwater) &&
2984 			    (lyp->nfsly_flags & (NFSLY_RECALL |
2985 			     NFSLY_RETONCLOSE)) == 0 &&
2986 			    lyp->nfsly_lock.nfslock_usecnt == 0 &&
2987 			    lyp->nfsly_lock.nfslock_lock == 0) {
2988 				NFSCL_DEBUG(4, "ret stale lay=%d\n",
2989 				    clp->nfsc_layoutcnt);
2990 				recallp = malloc(sizeof(*recallp),
2991 				    M_NFSLAYRECALL, M_NOWAIT);
2992 				if (recallp == NULL)
2993 					break;
2994 				(void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE,
2995 				    lyp, NFSLAYOUTIOMODE_ANY, 0, UINT64_MAX,
2996 				    lyp->nfsly_stateid.seqid, 0, 0, NULL,
2997 				    recallp);
2998 			}
2999 			lyp = nlyp;
3000 		}
3001 
3002 		/*
3003 		 * Free up any unreferenced device info structures.
3004 		 */
3005 		LIST_FOREACH_SAFE(dip, &clp->nfsc_devinfo, nfsdi_list, ndip) {
3006 			if (dip->nfsdi_layoutrefs == 0 &&
3007 			    dip->nfsdi_refcnt == 0) {
3008 				NFSCL_DEBUG(4, "freeing devinfo\n");
3009 				LIST_REMOVE(dip, nfsdi_list);
3010 				nfscl_freedevinfo(dip);
3011 			}
3012 		}
3013 		NFSUNLOCKCLSTATE();
3014 
3015 		/* Do layout return(s), as required. */
3016 		TAILQ_FOREACH_SAFE(lyp, &rlh, nfsly_list, nlyp) {
3017 			TAILQ_REMOVE(&rlh, lyp, nfsly_list);
3018 			NFSCL_DEBUG(4, "ret layout\n");
3019 			nfscl_layoutreturn(clp->nfsc_nmp, lyp, cred, p);
3020 			if ((lyp->nfsly_flags & NFSLY_RETONCLOSE) != 0) {
3021 				NFSLOCKCLSTATE();
3022 				lyp->nfsly_flags |= NFSLY_RETURNED;
3023 				wakeup(lyp);
3024 				NFSUNLOCKCLSTATE();
3025 			} else
3026 				nfscl_freelayout(lyp);
3027 		}
3028 
3029 		/*
3030 		 * Delegreturn any delegations cleaned out or recalled.
3031 		 */
3032 		TAILQ_FOREACH_SAFE(dp, &dh, nfsdl_list, ndp) {
3033 			newnfs_copycred(&dp->nfsdl_cred, cred);
3034 			(void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
3035 			TAILQ_REMOVE(&dh, dp, nfsdl_list);
3036 			free(dp, M_NFSCLDELEG);
3037 		}
3038 
3039 		SLIST_INIT(&lfh);
3040 		/*
3041 		 * Call nfscl_cleanupkext() once per second to check for
3042 		 * open/lock owners where the process has exited.
3043 		 */
3044 		mytime = NFSD_MONOSEC;
3045 		if (prevsec != mytime) {
3046 			prevsec = mytime;
3047 			nfscl_cleanupkext(clp, &lfh);
3048 		}
3049 
3050 		/*
3051 		 * Do a ReleaseLockOwner for all lock owners where the
3052 		 * associated process no longer exists, as found by
3053 		 * nfscl_cleanupkext().
3054 		 */
3055 		newnfs_setroot(cred);
3056 		SLIST_FOREACH_SAFE(lfhp, &lfh, nfslfh_list, nlfhp) {
3057 			LIST_FOREACH_SAFE(lp, &lfhp->nfslfh_lock, nfsl_list,
3058 			    nlp) {
3059 				(void)nfsrpc_rellockown(clp->nfsc_nmp, lp,
3060 				    lfhp->nfslfh_fh, lfhp->nfslfh_len, cred,
3061 				    p);
3062 				nfscl_freelockowner(lp, 0);
3063 			}
3064 			free(lfhp, M_TEMP);
3065 		}
3066 		SLIST_INIT(&lfh);
3067 
3068 		NFSLOCKCLSTATE();
3069 		if ((clp->nfsc_flags & NFSCLFLAGS_RECOVER) == 0)
3070 			(void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT, "nfscl",
3071 			    hz);
3072 terminate:
3073 		if (clp->nfsc_flags & NFSCLFLAGS_UMOUNT) {
3074 			clp->nfsc_flags &= ~NFSCLFLAGS_HASTHREAD;
3075 			NFSUNLOCKCLSTATE();
3076 			NFSFREECRED(cred);
3077 			wakeup((caddr_t)clp);
3078 			return;
3079 		}
3080 		NFSUNLOCKCLSTATE();
3081 	}
3082 }
3083 
3084 /*
3085  * Initiate state recovery. Called when NFSERR_STALECLIENTID,
3086  * NFSERR_STALESTATEID or NFSERR_BADSESSION is received.
3087  */
3088 void
nfscl_initiate_recovery(struct nfsclclient * clp)3089 nfscl_initiate_recovery(struct nfsclclient *clp)
3090 {
3091 
3092 	if (clp == NULL)
3093 		return;
3094 	NFSLOCKCLSTATE();
3095 	clp->nfsc_flags |= NFSCLFLAGS_RECOVER;
3096 	NFSUNLOCKCLSTATE();
3097 	wakeup((caddr_t)clp);
3098 }
3099 
3100 /*
3101  * Dump out the state stuff for debugging.
3102  */
3103 void
nfscl_dumpstate(struct nfsmount * nmp,int openowner,int opens,int lockowner,int locks)3104 nfscl_dumpstate(struct nfsmount *nmp, int openowner, int opens,
3105     int lockowner, int locks)
3106 {
3107 	struct nfsclclient *clp;
3108 	struct nfsclowner *owp;
3109 	struct nfsclopen *op;
3110 	struct nfscllockowner *lp;
3111 	struct nfscllock *lop;
3112 	struct nfscldeleg *dp;
3113 
3114 	clp = nmp->nm_clp;
3115 	if (clp == NULL) {
3116 		printf("nfscl dumpstate NULL clp\n");
3117 		return;
3118 	}
3119 	NFSLOCKCLSTATE();
3120 	TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
3121 	  LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
3122 	    if (openowner && !LIST_EMPTY(&owp->nfsow_open))
3123 		printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
3124 		    owp->nfsow_owner[0], owp->nfsow_owner[1],
3125 		    owp->nfsow_owner[2], owp->nfsow_owner[3],
3126 		    owp->nfsow_seqid);
3127 	    LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3128 		if (opens)
3129 		    printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
3130 			op->nfso_stateid.other[0], op->nfso_stateid.other[1],
3131 			op->nfso_stateid.other[2], op->nfso_opencnt,
3132 			op->nfso_fh[12]);
3133 		LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
3134 		    if (lockowner)
3135 			printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
3136 			    lp->nfsl_owner[0], lp->nfsl_owner[1],
3137 			    lp->nfsl_owner[2], lp->nfsl_owner[3],
3138 			    lp->nfsl_seqid,
3139 			    lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
3140 			    lp->nfsl_stateid.other[2]);
3141 		    LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
3142 			if (locks)
3143 #ifdef __FreeBSD__
3144 			    printf("lck typ=%d fst=%ju end=%ju\n",
3145 				lop->nfslo_type, (intmax_t)lop->nfslo_first,
3146 				(intmax_t)lop->nfslo_end);
3147 #else
3148 			    printf("lck typ=%d fst=%qd end=%qd\n",
3149 				lop->nfslo_type, lop->nfslo_first,
3150 				lop->nfslo_end);
3151 #endif
3152 		    }
3153 		}
3154 	    }
3155 	  }
3156 	}
3157 	LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3158 	    if (openowner && !LIST_EMPTY(&owp->nfsow_open))
3159 		printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
3160 		    owp->nfsow_owner[0], owp->nfsow_owner[1],
3161 		    owp->nfsow_owner[2], owp->nfsow_owner[3],
3162 		    owp->nfsow_seqid);
3163 	    LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3164 		if (opens)
3165 		    printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
3166 			op->nfso_stateid.other[0], op->nfso_stateid.other[1],
3167 			op->nfso_stateid.other[2], op->nfso_opencnt,
3168 			op->nfso_fh[12]);
3169 		LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
3170 		    if (lockowner)
3171 			printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
3172 			    lp->nfsl_owner[0], lp->nfsl_owner[1],
3173 			    lp->nfsl_owner[2], lp->nfsl_owner[3],
3174 			    lp->nfsl_seqid,
3175 			    lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
3176 			    lp->nfsl_stateid.other[2]);
3177 		    LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
3178 			if (locks)
3179 #ifdef __FreeBSD__
3180 			    printf("lck typ=%d fst=%ju end=%ju\n",
3181 				lop->nfslo_type, (intmax_t)lop->nfslo_first,
3182 				(intmax_t)lop->nfslo_end);
3183 #else
3184 			    printf("lck typ=%d fst=%qd end=%qd\n",
3185 				lop->nfslo_type, lop->nfslo_first,
3186 				lop->nfslo_end);
3187 #endif
3188 		    }
3189 		}
3190 	    }
3191 	}
3192 	NFSUNLOCKCLSTATE();
3193 }
3194 
3195 /*
3196  * Check for duplicate open owners and opens.
3197  * (Only used as a diagnostic aid.)
3198  */
3199 void
nfscl_dupopen(vnode_t vp,int dupopens)3200 nfscl_dupopen(vnode_t vp, int dupopens)
3201 {
3202 	struct nfsclclient *clp;
3203 	struct nfsclowner *owp, *owp2;
3204 	struct nfsclopen *op, *op2;
3205 	struct nfsfh *nfhp;
3206 
3207 	clp = VFSTONFS(vp->v_mount)->nm_clp;
3208 	if (clp == NULL) {
3209 		printf("nfscl dupopen NULL clp\n");
3210 		return;
3211 	}
3212 	nfhp = VTONFS(vp)->n_fhp;
3213 	NFSLOCKCLSTATE();
3214 
3215 	/*
3216 	 * First, search for duplicate owners.
3217 	 * These should never happen!
3218 	 */
3219 	LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3220 	    LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3221 		if (owp != owp2 &&
3222 		    !NFSBCMP(owp->nfsow_owner, owp2->nfsow_owner,
3223 		    NFSV4CL_LOCKNAMELEN)) {
3224 			NFSUNLOCKCLSTATE();
3225 			printf("DUP OWNER\n");
3226 			nfscl_dumpstate(VFSTONFS(vp->v_mount), 1, 1, 0, 0);
3227 			return;
3228 		}
3229 	    }
3230 	}
3231 
3232 	/*
3233 	 * Now, search for duplicate stateids.
3234 	 * These shouldn't happen, either.
3235 	 */
3236 	LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3237 	    LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
3238 		LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3239 		    LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3240 			if (op != op2 &&
3241 			    (op->nfso_stateid.other[0] != 0 ||
3242 			     op->nfso_stateid.other[1] != 0 ||
3243 			     op->nfso_stateid.other[2] != 0) &&
3244 			    op->nfso_stateid.other[0] == op2->nfso_stateid.other[0] &&
3245 			    op->nfso_stateid.other[1] == op2->nfso_stateid.other[1] &&
3246 			    op->nfso_stateid.other[2] == op2->nfso_stateid.other[2]) {
3247 			    NFSUNLOCKCLSTATE();
3248 			    printf("DUP STATEID\n");
3249 			    nfscl_dumpstate(VFSTONFS(vp->v_mount), 1, 1, 0, 0);
3250 			    return;
3251 			}
3252 		    }
3253 		}
3254 	    }
3255 	}
3256 
3257 	/*
3258 	 * Now search for duplicate opens.
3259 	 * Duplicate opens for the same owner
3260 	 * should never occur. Other duplicates are
3261 	 * possible and are checked for if "dupopens"
3262 	 * is true.
3263 	 */
3264 	LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3265 	    LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
3266 		if (nfhp->nfh_len == op2->nfso_fhlen &&
3267 		    !NFSBCMP(nfhp->nfh_fh, op2->nfso_fh, nfhp->nfh_len)) {
3268 		    LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3269 			LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3270 			    if (op != op2 && nfhp->nfh_len == op->nfso_fhlen &&
3271 				!NFSBCMP(nfhp->nfh_fh, op->nfso_fh, nfhp->nfh_len) &&
3272 				(!NFSBCMP(op->nfso_own->nfsow_owner,
3273 				 op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN) ||
3274 				 dupopens)) {
3275 				if (!NFSBCMP(op->nfso_own->nfsow_owner,
3276 				    op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
3277 				    NFSUNLOCKCLSTATE();
3278 				    printf("BADDUP OPEN\n");
3279 				} else {
3280 				    NFSUNLOCKCLSTATE();
3281 				    printf("DUP OPEN\n");
3282 				}
3283 				nfscl_dumpstate(VFSTONFS(vp->v_mount), 1, 1, 0,
3284 				    0);
3285 				return;
3286 			    }
3287 			}
3288 		    }
3289 		}
3290 	    }
3291 	}
3292 	NFSUNLOCKCLSTATE();
3293 }
3294 
3295 /*
3296  * During close, find an open that needs to be dereferenced and
3297  * dereference it. If there are no more opens for this file,
3298  * log a message to that effect.
3299  * Opens aren't actually Close'd until VOP_INACTIVE() is performed
3300  * on the file's vnode.
3301  * This is the safe way, since it is difficult to identify
3302  * which open the close is for and I/O can be performed after the
3303  * close(2) system call when a file is mmap'd.
3304  * If it returns 0 for success, there will be a referenced
3305  * clp returned via clpp.
3306  */
3307 int
nfscl_getclose(vnode_t vp,struct nfsclclient ** clpp)3308 nfscl_getclose(vnode_t vp, struct nfsclclient **clpp)
3309 {
3310 	struct nfsclclient *clp;
3311 	struct nfsclowner *owp;
3312 	struct nfsclopen *op;
3313 	struct nfscldeleg *dp;
3314 	struct nfsfh *nfhp;
3315 	int error, notdecr;
3316 
3317 	error = nfscl_getcl(vp->v_mount, NULL, NULL, false, true, &clp);
3318 	if (error)
3319 		return (error);
3320 	*clpp = clp;
3321 
3322 	nfhp = VTONFS(vp)->n_fhp;
3323 	notdecr = 1;
3324 	NFSLOCKCLSTATE();
3325 	/*
3326 	 * First, look for one under a delegation that was locally issued
3327 	 * and just decrement the opencnt for it. Since all my Opens against
3328 	 * the server are DENY_NONE, I don't see a problem with hanging
3329 	 * onto them. (It is much easier to use one of the extant Opens
3330 	 * that I already have on the server when a Delegation is recalled
3331 	 * than to do fresh Opens.) Someday, I might need to rethink this, but.
3332 	 */
3333 	dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
3334 	if (dp != NULL) {
3335 		LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
3336 			op = LIST_FIRST(&owp->nfsow_open);
3337 			if (op != NULL) {
3338 				/*
3339 				 * Since a delegation is for a file, there
3340 				 * should never be more than one open for
3341 				 * each openowner.
3342 				 */
3343 				if (LIST_NEXT(op, nfso_list) != NULL)
3344 					panic("nfscdeleg opens");
3345 				if (notdecr && op->nfso_opencnt > 0) {
3346 					notdecr = 0;
3347 					op->nfso_opencnt--;
3348 					break;
3349 				}
3350 			}
3351 		}
3352 	}
3353 
3354 	/* Now process the opens against the server. */
3355 	LIST_FOREACH(op, NFSCLOPENHASH(clp, nfhp->nfh_fh, nfhp->nfh_len),
3356 	    nfso_hash) {
3357 		if (op->nfso_fhlen == nfhp->nfh_len &&
3358 		    !NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
3359 		    nfhp->nfh_len)) {
3360 			/* Found an open, decrement cnt if possible */
3361 			if (notdecr && op->nfso_opencnt > 0) {
3362 				notdecr = 0;
3363 				op->nfso_opencnt--;
3364 			}
3365 			/*
3366 			 * There are more opens, so just return.
3367 			 */
3368 			if (op->nfso_opencnt > 0) {
3369 				NFSUNLOCKCLSTATE();
3370 				return (0);
3371 			}
3372 		}
3373 	}
3374 	NFSUNLOCKCLSTATE();
3375 	if (notdecr)
3376 		printf("nfscl: never fnd open\n");
3377 	return (0);
3378 }
3379 
3380 int
nfscl_doclose(vnode_t vp,struct nfsclclient ** clpp,NFSPROC_T * p)3381 nfscl_doclose(vnode_t vp, struct nfsclclient **clpp, NFSPROC_T *p)
3382 {
3383 	struct nfsclclient *clp;
3384 	struct nfsmount *nmp;
3385 	struct nfsclowner *owp, *nowp;
3386 	struct nfsclopen *op, *nop;
3387 	struct nfsclopenhead delayed;
3388 	struct nfscldeleg *dp;
3389 	struct nfsfh *nfhp;
3390 	struct nfsclrecalllayout *recallp;
3391 	struct nfscllayout *lyp;
3392 	int error;
3393 
3394 	error = nfscl_getcl(vp->v_mount, NULL, NULL, false, true, &clp);
3395 	if (error)
3396 		return (error);
3397 	*clpp = clp;
3398 
3399 	nmp = VFSTONFS(vp->v_mount);
3400 	nfhp = VTONFS(vp)->n_fhp;
3401 	recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_WAITOK);
3402 	NFSLOCKCLSTATE();
3403 	/*
3404 	 * First get rid of the local Open structures, which should be no
3405 	 * longer in use.
3406 	 */
3407 	dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
3408 	if (dp != NULL) {
3409 		LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) {
3410 			op = LIST_FIRST(&owp->nfsow_open);
3411 			if (op != NULL) {
3412 				KASSERT((op->nfso_opencnt == 0),
3413 				    ("nfscl: bad open cnt on deleg"));
3414 				nfscl_freeopen(op, 1, true);
3415 			}
3416 			nfscl_freeopenowner(owp, 1);
3417 		}
3418 	}
3419 
3420 	/* Return any layouts marked return on close. */
3421 	nfscl_retoncloselayout(vp, clp, nfhp->nfh_fh, nfhp->nfh_len, &recallp,
3422 	    &lyp);
3423 
3424 	/* Now process the opens against the server. */
3425 	LIST_INIT(&delayed);
3426 lookformore:
3427 	LIST_FOREACH(op, NFSCLOPENHASH(clp, nfhp->nfh_fh, nfhp->nfh_len),
3428 	    nfso_hash) {
3429 		if (op->nfso_fhlen == nfhp->nfh_len &&
3430 		    !NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
3431 		    nfhp->nfh_len)) {
3432 			/* Found an open, close it. */
3433 #ifdef DIAGNOSTIC
3434 			KASSERT((op->nfso_opencnt == 0),
3435 			    ("nfscl: bad open cnt on server (%d)",
3436 			     op->nfso_opencnt));
3437 #endif
3438 			NFSUNLOCKCLSTATE();
3439 			if (NFSHASNFSV4N(nmp))
3440 				error = nfsrpc_doclose(nmp, op, p, false, true);
3441 			else
3442 				error = nfsrpc_doclose(nmp, op, p, true, true);
3443 			NFSLOCKCLSTATE();
3444 			if (error == NFSERR_DELAY) {
3445 				nfscl_unlinkopen(op);
3446 				op->nfso_own = NULL;
3447 				LIST_INSERT_HEAD(&delayed, op, nfso_list);
3448 			}
3449 			goto lookformore;
3450 		}
3451 	}
3452 	nfscl_clrelease(clp);
3453 
3454 	/* Now, wait for any layout that is returned upon close. */
3455 	if (lyp != NULL) {
3456 		while ((lyp->nfsly_flags & NFSLY_RETURNED) == 0) {
3457 			if (NFSCL_FORCEDISM(nmp->nm_mountp)) {
3458 				lyp = NULL;
3459 				break;
3460 			}
3461 			msleep(lyp, NFSCLSTATEMUTEXPTR, PZERO, "nfslroc", hz);
3462 		}
3463 		if (lyp != NULL)
3464 			nfscl_freelayout(lyp);
3465 	}
3466 
3467 	NFSUNLOCKCLSTATE();
3468 	/*
3469 	 * recallp has been set NULL by nfscl_retoncloselayout() if it was
3470 	 * used by the function, but calling free() with a NULL pointer is ok.
3471 	 */
3472 	free(recallp, M_NFSLAYRECALL);
3473 
3474 	/* Now, loop retrying the delayed closes. */
3475 	LIST_FOREACH_SAFE(op, &delayed, nfso_list, nop) {
3476 		nfsrpc_doclose(nmp, op, p, true, false);
3477 		LIST_REMOVE(op, nfso_list);
3478 		nfscl_freeopen(op, 0, false);
3479 	}
3480 	return (0);
3481 }
3482 
3483 /*
3484  * Return all delegations on this client.
3485  * (Must be called with client sleep lock.)
3486  */
3487 static void
nfscl_delegreturnall(struct nfsclclient * clp,NFSPROC_T * p,struct nfscldeleghead * dhp)3488 nfscl_delegreturnall(struct nfsclclient *clp, NFSPROC_T *p,
3489     struct nfscldeleghead *dhp)
3490 {
3491 	struct nfscldeleg *dp, *ndp;
3492 	struct ucred *cred;
3493 
3494 	cred = newnfs_getcred();
3495 	TAILQ_FOREACH_SAFE(dp, &clp->nfsc_deleg, nfsdl_list, ndp) {
3496 		nfscl_cleandeleg(dp);
3497 		(void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
3498 		if (dhp != NULL) {
3499 			nfscl_freedeleg(&clp->nfsc_deleg, dp, false);
3500 			TAILQ_INSERT_HEAD(dhp, dp, nfsdl_list);
3501 		} else
3502 			nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
3503 	}
3504 	NFSFREECRED(cred);
3505 }
3506 
3507 /*
3508  * Return any delegation for this vp.
3509  */
3510 void
nfscl_delegreturnvp(vnode_t vp,NFSPROC_T * p)3511 nfscl_delegreturnvp(vnode_t vp, NFSPROC_T *p)
3512 {
3513 	struct nfsclclient *clp;
3514 	struct nfscldeleg *dp;
3515 	struct ucred *cred;
3516 	struct nfsnode *np;
3517 	struct nfsmount *nmp;
3518 
3519 	nmp = VFSTONFS(vp->v_mount);
3520 	NFSLOCKMNT(nmp);
3521 	if ((nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0) {
3522 		NFSUNLOCKMNT(nmp);
3523 		return;
3524 	}
3525 	NFSUNLOCKMNT(nmp);
3526 	np = VTONFS(vp);
3527 	cred = newnfs_getcred();
3528 	dp = NULL;
3529 	NFSLOCKCLSTATE();
3530 	clp = nmp->nm_clp;
3531 	if (clp != NULL)
3532 		dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
3533 		    np->n_fhp->nfh_len);
3534 	if (dp != NULL) {
3535 		nfscl_cleandeleg(dp);
3536 		nfscl_freedeleg(&clp->nfsc_deleg, dp, false);
3537 		NFSUNLOCKCLSTATE();
3538 		newnfs_copycred(&dp->nfsdl_cred, cred);
3539 		nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
3540 		free(dp, M_NFSCLDELEG);
3541 	} else
3542 		NFSUNLOCKCLSTATE();
3543 	NFSFREECRED(cred);
3544 }
3545 
3546 /*
3547  * Do a callback RPC.
3548  */
3549 void
nfscl_docb(struct nfsrv_descript * nd,NFSPROC_T * p)3550 nfscl_docb(struct nfsrv_descript *nd, NFSPROC_T *p)
3551 {
3552 	int clist, gotseq_ok, i, j, k, op, rcalls;
3553 	u_int32_t *tl;
3554 	struct nfsclclient *clp;
3555 	struct nfscldeleg *dp = NULL;
3556 	int numops, taglen = -1, error = 0, trunc __unused;
3557 	u_int32_t minorvers = 0, retops = 0, *retopsp = NULL, *repp, cbident;
3558 	u_char tag[NFSV4_SMALLSTR + 1], *tagstr;
3559 	vnode_t vp = NULL;
3560 	struct nfsnode *np;
3561 	struct vattr va;
3562 	struct nfsfh *nfhp;
3563 	mount_t mp;
3564 	nfsattrbit_t attrbits, rattrbits;
3565 	nfsv4stateid_t stateid;
3566 	uint32_t seqid, slotid = 0, highslot, cachethis __unused;
3567 	uint8_t sessionid[NFSX_V4SESSIONID];
3568 	struct mbuf *rep;
3569 	struct nfscllayout *lyp;
3570 	uint64_t filesid[2], len, off;
3571 	int changed, gotone, laytype, recalltype;
3572 	uint32_t iomode;
3573 	struct nfsclrecalllayout *recallp = NULL;
3574 	struct nfsclsession *tsep;
3575 
3576 	gotseq_ok = 0;
3577 	nfsrvd_rephead(nd);
3578 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3579 	taglen = fxdr_unsigned(int, *tl);
3580 	if (taglen < 0 || taglen > NFSV4_OPAQUELIMIT) {
3581 		error = EBADRPC;
3582 		taglen = -1;
3583 		goto nfsmout;
3584 	}
3585 	if (taglen <= NFSV4_SMALLSTR)
3586 		tagstr = tag;
3587 	else
3588 		tagstr = malloc(taglen + 1, M_TEMP, M_WAITOK);
3589 	error = nfsrv_mtostr(nd, tagstr, taglen);
3590 	if (error) {
3591 		if (taglen > NFSV4_SMALLSTR)
3592 			free(tagstr, M_TEMP);
3593 		taglen = -1;
3594 		goto nfsmout;
3595 	}
3596 	(void) nfsm_strtom(nd, tag, taglen);
3597 	if (taglen > NFSV4_SMALLSTR) {
3598 		free(tagstr, M_TEMP);
3599 	}
3600 	NFSM_BUILD(retopsp, u_int32_t *, NFSX_UNSIGNED);
3601 	NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3602 	minorvers = fxdr_unsigned(u_int32_t, *tl++);
3603 	if (minorvers != NFSV4_MINORVERSION &&
3604 	    minorvers != NFSV41_MINORVERSION &&
3605 	    minorvers != NFSV42_MINORVERSION)
3606 		nd->nd_repstat = NFSERR_MINORVERMISMATCH;
3607 	cbident = fxdr_unsigned(u_int32_t, *tl++);
3608 	if (nd->nd_repstat)
3609 		numops = 0;
3610 	else
3611 		numops = fxdr_unsigned(int, *tl);
3612 	/*
3613 	 * Loop around doing the sub ops.
3614 	 */
3615 	for (i = 0; i < numops; i++) {
3616 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3617 		NFSM_BUILD(repp, u_int32_t *, 2 * NFSX_UNSIGNED);
3618 		*repp++ = *tl;
3619 		op = fxdr_unsigned(int, *tl);
3620 		nd->nd_procnum = op;
3621 		if (i == 0 && op != NFSV4OP_CBSEQUENCE && minorvers !=
3622 		    NFSV4_MINORVERSION) {
3623 		    nd->nd_repstat = NFSERR_OPNOTINSESS;
3624 		    *repp = nfscl_errmap(nd, minorvers);
3625 		    retops++;
3626 		    break;
3627 		}
3628 		if (op < NFSV4OP_CBGETATTR ||
3629 		   (op > NFSV4OP_CBRECALL && minorvers == NFSV4_MINORVERSION) ||
3630 		   (op > NFSV4OP_CBNOTIFYDEVID &&
3631 		    minorvers == NFSV41_MINORVERSION) ||
3632 		   (op > NFSV4OP_CBOFFLOAD &&
3633 		    minorvers == NFSV42_MINORVERSION)) {
3634 		    nd->nd_repstat = NFSERR_OPILLEGAL;
3635 		    *repp = nfscl_errmap(nd, minorvers);
3636 		    retops++;
3637 		    break;
3638 		}
3639 		if (op < NFSV42_CBNOPS)
3640 			nfsstatsv1.cbrpccnt[nd->nd_procnum]++;
3641 		switch (op) {
3642 		case NFSV4OP_CBGETATTR:
3643 			NFSCL_DEBUG(4, "cbgetattr\n");
3644 			mp = NULL;
3645 			vp = NULL;
3646 			error = nfsm_getfh(nd, &nfhp);
3647 			if (!error)
3648 				error = nfsrv_getattrbits(nd, &attrbits,
3649 				    NULL, NULL);
3650 			if (!error) {
3651 				mp = nfscl_getmnt(minorvers, sessionid, cbident,
3652 				    &clp);
3653 				if (mp == NULL)
3654 					error = NFSERR_SERVERFAULT;
3655 			}
3656 			if (!error) {
3657 				error = nfscl_ngetreopen(mp, nfhp->nfh_fh,
3658 				    nfhp->nfh_len, p, &np);
3659 				if (!error)
3660 					vp = NFSTOV(np);
3661 			}
3662 			if (!error) {
3663 				NFSZERO_ATTRBIT(&rattrbits);
3664 				NFSLOCKCLSTATE();
3665 				dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
3666 				    nfhp->nfh_len);
3667 				if (dp != NULL) {
3668 					if (NFSISSET_ATTRBIT(&attrbits,
3669 					    NFSATTRBIT_SIZE)) {
3670 						if (vp != NULL)
3671 							va.va_size = np->n_size;
3672 						else
3673 							va.va_size =
3674 							    dp->nfsdl_size;
3675 						NFSSETBIT_ATTRBIT(&rattrbits,
3676 						    NFSATTRBIT_SIZE);
3677 					}
3678 					if (NFSISSET_ATTRBIT(&attrbits,
3679 					    NFSATTRBIT_CHANGE)) {
3680 						va.va_filerev =
3681 						    dp->nfsdl_change;
3682 						if (vp == NULL ||
3683 						    (np->n_flag & NDELEGMOD))
3684 							va.va_filerev++;
3685 						NFSSETBIT_ATTRBIT(&rattrbits,
3686 						    NFSATTRBIT_CHANGE);
3687 					}
3688 				} else
3689 					error = NFSERR_SERVERFAULT;
3690 				NFSUNLOCKCLSTATE();
3691 			}
3692 			if (vp != NULL)
3693 				vrele(vp);
3694 			if (mp != NULL)
3695 				vfs_unbusy(mp);
3696 			if (nfhp != NULL)
3697 				free(nfhp, M_NFSFH);
3698 			if (!error)
3699 				(void) nfsv4_fillattr(nd, NULL, NULL, NULL, &va,
3700 				    NULL, 0, &rattrbits, NULL, p, 0, 0, 0, 0,
3701 				    (uint64_t)0, NULL);
3702 			break;
3703 		case NFSV4OP_CBRECALL:
3704 			NFSCL_DEBUG(4, "cbrecall\n");
3705 			NFSM_DISSECT(tl, u_int32_t *, NFSX_STATEID +
3706 			    NFSX_UNSIGNED);
3707 			stateid.seqid = *tl++;
3708 			NFSBCOPY((caddr_t)tl, (caddr_t)stateid.other,
3709 			    NFSX_STATEIDOTHER);
3710 			tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
3711 			trunc = fxdr_unsigned(int, *tl);
3712 			error = nfsm_getfh(nd, &nfhp);
3713 			if (!error) {
3714 				NFSLOCKCLSTATE();
3715 				if (minorvers == NFSV4_MINORVERSION)
3716 					clp = nfscl_getclnt(cbident);
3717 				else
3718 					clp = nfscl_getclntsess(sessionid);
3719 				if (clp != NULL) {
3720 					dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
3721 					    nfhp->nfh_len);
3722 					if (dp != NULL && (dp->nfsdl_flags &
3723 					    NFSCLDL_DELEGRET) == 0) {
3724 						dp->nfsdl_flags |=
3725 						    NFSCLDL_RECALL;
3726 						wakeup((caddr_t)clp);
3727 					}
3728 				} else {
3729 					error = NFSERR_SERVERFAULT;
3730 				}
3731 				NFSUNLOCKCLSTATE();
3732 			}
3733 			if (nfhp != NULL)
3734 				free(nfhp, M_NFSFH);
3735 			break;
3736 		case NFSV4OP_CBLAYOUTRECALL:
3737 			NFSCL_DEBUG(4, "cblayrec\n");
3738 			nfhp = NULL;
3739 			NFSM_DISSECT(tl, uint32_t *, 4 * NFSX_UNSIGNED);
3740 			laytype = fxdr_unsigned(int, *tl++);
3741 			iomode = fxdr_unsigned(uint32_t, *tl++);
3742 			if (newnfs_true == *tl++)
3743 				changed = 1;
3744 			else
3745 				changed = 0;
3746 			recalltype = fxdr_unsigned(int, *tl);
3747 			NFSCL_DEBUG(4, "layt=%d iom=%d ch=%d rectyp=%d\n",
3748 			    laytype, iomode, changed, recalltype);
3749 			recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL,
3750 			    M_WAITOK);
3751 			if (laytype != NFSLAYOUT_NFSV4_1_FILES &&
3752 			    laytype != NFSLAYOUT_FLEXFILE)
3753 				error = NFSERR_NOMATCHLAYOUT;
3754 			else if (recalltype == NFSLAYOUTRETURN_FILE) {
3755 				error = nfsm_getfh(nd, &nfhp);
3756 				NFSCL_DEBUG(4, "retfile getfh=%d\n", error);
3757 				if (error != 0)
3758 					goto nfsmout;
3759 				NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_HYPER +
3760 				    NFSX_STATEID);
3761 				off = fxdr_hyper(tl); tl += 2;
3762 				len = fxdr_hyper(tl); tl += 2;
3763 				stateid.seqid = fxdr_unsigned(uint32_t, *tl++);
3764 				NFSBCOPY(tl, stateid.other, NFSX_STATEIDOTHER);
3765 				if (minorvers == NFSV4_MINORVERSION)
3766 					error = NFSERR_NOTSUPP;
3767 				NFSCL_DEBUG(4, "off=%ju len=%ju sq=%u err=%d\n",
3768 				    (uintmax_t)off, (uintmax_t)len,
3769 				    stateid.seqid, error);
3770 				if (error == 0) {
3771 					NFSLOCKCLSTATE();
3772 					clp = nfscl_getclntsess(sessionid);
3773 					NFSCL_DEBUG(4, "cbly clp=%p\n", clp);
3774 					if (clp != NULL) {
3775 						lyp = nfscl_findlayout(clp,
3776 						    nfhp->nfh_fh,
3777 						    nfhp->nfh_len);
3778 						NFSCL_DEBUG(4, "cblyp=%p\n",
3779 						    lyp);
3780 						if (lyp != NULL &&
3781 						    (lyp->nfsly_flags &
3782 						     (NFSLY_FILES |
3783 						      NFSLY_FLEXFILE)) != 0 &&
3784 						    !NFSBCMP(stateid.other,
3785 						    lyp->nfsly_stateid.other,
3786 						    NFSX_STATEIDOTHER)) {
3787 							error =
3788 							    nfscl_layoutrecall(
3789 							    recalltype,
3790 							    lyp, iomode, off,
3791 							    len, stateid.seqid,
3792 							    0, 0, NULL,
3793 							    recallp);
3794 							if (error == 0 &&
3795 							    stateid.seqid >
3796 							    lyp->nfsly_stateid.seqid)
3797 								lyp->nfsly_stateid.seqid =
3798 								    stateid.seqid;
3799 							recallp = NULL;
3800 							wakeup(clp);
3801 							NFSCL_DEBUG(4,
3802 							    "aft layrcal=%d "
3803 							    "layseqid=%d\n",
3804 							    error,
3805 							    lyp->nfsly_stateid.seqid);
3806 						} else
3807 							error =
3808 							  NFSERR_NOMATCHLAYOUT;
3809 					} else
3810 						error = NFSERR_NOMATCHLAYOUT;
3811 					NFSUNLOCKCLSTATE();
3812 				}
3813 				free(nfhp, M_NFSFH);
3814 			} else if (recalltype == NFSLAYOUTRETURN_FSID) {
3815 				NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_HYPER);
3816 				filesid[0] = fxdr_hyper(tl); tl += 2;
3817 				filesid[1] = fxdr_hyper(tl); tl += 2;
3818 				gotone = 0;
3819 				NFSLOCKCLSTATE();
3820 				clp = nfscl_getclntsess(sessionid);
3821 				if (clp != NULL) {
3822 					TAILQ_FOREACH(lyp, &clp->nfsc_layout,
3823 					    nfsly_list) {
3824 						if (lyp->nfsly_filesid[0] ==
3825 						    filesid[0] &&
3826 						    lyp->nfsly_filesid[1] ==
3827 						    filesid[1]) {
3828 							error =
3829 							    nfscl_layoutrecall(
3830 							    recalltype,
3831 							    lyp, iomode, 0,
3832 							    UINT64_MAX,
3833 							    lyp->nfsly_stateid.seqid,
3834 							    0, 0, NULL,
3835 							    recallp);
3836 							recallp = NULL;
3837 							gotone = 1;
3838 						}
3839 					}
3840 					if (gotone != 0)
3841 						wakeup(clp);
3842 					else
3843 						error = NFSERR_NOMATCHLAYOUT;
3844 				} else
3845 					error = NFSERR_NOMATCHLAYOUT;
3846 				NFSUNLOCKCLSTATE();
3847 			} else if (recalltype == NFSLAYOUTRETURN_ALL) {
3848 				gotone = 0;
3849 				NFSLOCKCLSTATE();
3850 				clp = nfscl_getclntsess(sessionid);
3851 				if (clp != NULL) {
3852 					TAILQ_FOREACH(lyp, &clp->nfsc_layout,
3853 					    nfsly_list) {
3854 						error = nfscl_layoutrecall(
3855 						    recalltype, lyp, iomode, 0,
3856 						    UINT64_MAX,
3857 						    lyp->nfsly_stateid.seqid,
3858 						    0, 0, NULL, recallp);
3859 						recallp = NULL;
3860 						gotone = 1;
3861 					}
3862 					if (gotone != 0)
3863 						wakeup(clp);
3864 					else
3865 						error = NFSERR_NOMATCHLAYOUT;
3866 				} else
3867 					error = NFSERR_NOMATCHLAYOUT;
3868 				NFSUNLOCKCLSTATE();
3869 			} else
3870 				error = NFSERR_NOMATCHLAYOUT;
3871 			if (recallp != NULL) {
3872 				free(recallp, M_NFSLAYRECALL);
3873 				recallp = NULL;
3874 			}
3875 			break;
3876 		case NFSV4OP_CBSEQUENCE:
3877 			if (i != 0) {
3878 			    error = NFSERR_SEQUENCEPOS;
3879 			    break;
3880 			}
3881 			NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID +
3882 			    5 * NFSX_UNSIGNED);
3883 			bcopy(tl, sessionid, NFSX_V4SESSIONID);
3884 			tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3885 			seqid = fxdr_unsigned(uint32_t, *tl++);
3886 			slotid = fxdr_unsigned(uint32_t, *tl++);
3887 			highslot = fxdr_unsigned(uint32_t, *tl++);
3888 			cachethis = *tl++;
3889 			/* Throw away the referring call stuff. */
3890 			clist = fxdr_unsigned(int, *tl);
3891 			for (j = 0; j < clist; j++) {
3892 				NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID +
3893 				    NFSX_UNSIGNED);
3894 				tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3895 				rcalls = fxdr_unsigned(int, *tl);
3896 				for (k = 0; k < rcalls; k++) {
3897 					NFSM_DISSECT(tl, uint32_t *,
3898 					    2 * NFSX_UNSIGNED);
3899 				}
3900 			}
3901 			NFSLOCKCLSTATE();
3902 			clp = nfscl_getclntsess(sessionid);
3903 			if (clp == NULL)
3904 				error = NFSERR_SERVERFAULT;
3905 			if (error == 0) {
3906 				tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3907 				error = nfsv4_seqsession(seqid, slotid,
3908 				    highslot, tsep->nfsess_cbslots, &rep,
3909 				    tsep->nfsess_backslots);
3910 			}
3911 			NFSUNLOCKCLSTATE();
3912 			if (error == 0 || error == NFSERR_REPLYFROMCACHE) {
3913 				gotseq_ok = 1;
3914 				if (rep != NULL) {
3915 					/*
3916 					 * Handle a reply for a retried
3917 					 * callback.  The reply will be
3918 					 * re-inserted in the session cache
3919 					 * by the nfsv4_seqsess_cacherep() call
3920 					 * after out:
3921 					 */
3922 					KASSERT(error == NFSERR_REPLYFROMCACHE,
3923 					    ("cbsequence: non-NULL rep"));
3924 					NFSCL_DEBUG(4, "Got cbretry\n");
3925 					m_freem(nd->nd_mreq);
3926 					nd->nd_mreq = rep;
3927 					rep = NULL;
3928 					goto out;
3929 				}
3930 				NFSM_BUILD(tl, uint32_t *,
3931 				    NFSX_V4SESSIONID + 4 * NFSX_UNSIGNED);
3932 				bcopy(sessionid, tl, NFSX_V4SESSIONID);
3933 				tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3934 				*tl++ = txdr_unsigned(seqid);
3935 				*tl++ = txdr_unsigned(slotid);
3936 				*tl++ = txdr_unsigned(NFSV4_CBSLOTS - 1);
3937 				*tl = txdr_unsigned(NFSV4_CBSLOTS - 1);
3938 			}
3939 			break;
3940 		case NFSV4OP_CBRECALLSLOT:
3941 			NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED);
3942 			highslot = fxdr_unsigned(uint32_t, *tl);
3943 			NFSLOCKCLSTATE();
3944 			clp = nfscl_getclntsess(sessionid);
3945 			if (clp == NULL)
3946 				error = NFSERR_SERVERFAULT;
3947 			if (error == 0) {
3948 				tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3949 				mtx_lock(&tsep->nfsess_mtx);
3950 				if ((highslot + 1) < tsep->nfsess_foreslots) {
3951 					tsep->nfsess_foreslots = (highslot + 1);
3952 					nfs_resetslots(tsep);
3953 				}
3954 				mtx_unlock(&tsep->nfsess_mtx);
3955 			}
3956 			NFSUNLOCKCLSTATE();
3957 			break;
3958 		case NFSV4OP_CBRECALLANY:
3959 			NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
3960 			i = fxdr_unsigned(int, *tl++);
3961 			j = fxdr_unsigned(int, *tl);
3962 			if (i < 0 || j != 1)
3963 				error = NFSERR_BADXDR;
3964 			if (error == 0) {
3965 				NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED);
3966 				j = fxdr_unsigned(int, *tl);
3967 				if (i < 100)
3968 					i = 100;
3969 				else if (i > 100000)
3970 					i = 100000;
3971 				NFSLOCKCLSTATE();
3972 				clp = nfscl_getclntsess(sessionid);
3973 				if (clp == NULL)
3974 					error = NFSERR_SERVERFAULT;
3975 				if (((j & NFSRCA4_RDATA_DLG) != 0 ||
3976 				    (j & NFSRCA4_WDATA_DLG) != 0) &&
3977 				    error == 0 && i <
3978 				    clp->nfsc_deleghighwater)
3979 					clp->nfsc_deleghighwater = i;
3980 				if (error == 0 &&
3981 				    ((!NFSHASFLEXFILE(clp->nfsc_nmp) &&
3982 				     (j & NFSRCA4_FILE_LAYOUT) != 0 &&
3983 				     i < clp->nfsc_layouthighwater) ||
3984 				     (NFSHASFLEXFILE(clp->nfsc_nmp) &&
3985 				     (j & (NFSRCA4_FF_LAYOUT_READ |
3986 				     NFSRCA4_FF_LAYOUT_RW)) != 0 &&
3987 				     i < clp->nfsc_layouthighwater)))
3988 					clp->nfsc_layouthighwater = i;
3989 				NFSUNLOCKCLSTATE();
3990 			}
3991 			break;
3992 		case NFSV4OP_CBNOTIFY:
3993 		case NFSV4OP_CBRECALLOBJAVAIL:
3994 		case NFSV4OP_CBNOTIFYLOCK:
3995 			/*
3996 			 * These callbacks are not necessarily optional,
3997 			 * so I think it is better to reply NFS_OK than
3998 			 * NFSERR_NOTSUPP.
3999 			 * All provide information for which the FreeBSD client
4000 			 * does not currently have a use.
4001 			 * I am not sure if any of these could be generated
4002 			 * by a NFSv4.1/4.2 server for this client?
4003 			 */
4004 			error = 0;
4005 			NFSCL_DEBUG(1, "unsupp callback %d\n", op);
4006 			break;
4007 		case NFSV4OP_CBPUSHDELEG:
4008 			error = NFSERR_REJECTDELEG;
4009 			NFSCL_DEBUG(1, "unsupp callback %d\n", op);
4010 			break;
4011 		default:
4012 			if (i == 0 && minorvers != NFSV4_MINORVERSION)
4013 				error = NFSERR_OPNOTINSESS;
4014 			else {
4015 				NFSCL_DEBUG(1, "unsupp callback %d\n", op);
4016 				error = NFSERR_NOTSUPP;
4017 			}
4018 			break;
4019 		}
4020 		if (error) {
4021 			if (error == EBADRPC || error == NFSERR_BADXDR) {
4022 				nd->nd_repstat = NFSERR_BADXDR;
4023 			} else {
4024 				nd->nd_repstat = error;
4025 			}
4026 			error = 0;
4027 		}
4028 		retops++;
4029 		if (nd->nd_repstat) {
4030 			*repp = nfscl_errmap(nd, minorvers);
4031 			break;
4032 		} else
4033 			*repp = 0;	/* NFS4_OK */
4034 	}
4035 nfsmout:
4036 	if (recallp != NULL)
4037 		free(recallp, M_NFSLAYRECALL);
4038 	if (error) {
4039 		if (error == EBADRPC || error == NFSERR_BADXDR)
4040 			nd->nd_repstat = NFSERR_BADXDR;
4041 		else
4042 			printf("nfsv4 comperr1=%d\n", error);
4043 	}
4044 	if (taglen == -1) {
4045 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
4046 		*tl++ = 0;
4047 		*tl = 0;
4048 	} else {
4049 		*retopsp = txdr_unsigned(retops);
4050 	}
4051 	*nd->nd_errp = nfscl_errmap(nd, minorvers);
4052 out:
4053 	if (gotseq_ok != 0) {
4054 		rep = m_copym(nd->nd_mreq, 0, M_COPYALL, M_WAITOK);
4055 		NFSLOCKCLSTATE();
4056 		clp = nfscl_getclntsess(sessionid);
4057 		if (clp != NULL) {
4058 			tsep = nfsmnt_mdssession(clp->nfsc_nmp);
4059 			nfsv4_seqsess_cacherep(slotid, tsep->nfsess_cbslots,
4060 			    NFSERR_OK, &rep);
4061 			NFSUNLOCKCLSTATE();
4062 		} else {
4063 			NFSUNLOCKCLSTATE();
4064 			m_freem(rep);
4065 		}
4066 	}
4067 }
4068 
4069 /*
4070  * Generate the next cbident value. Basically just increment a static value
4071  * and then check that it isn't already in the list, if it has wrapped around.
4072  */
4073 static u_int32_t
nfscl_nextcbident(void)4074 nfscl_nextcbident(void)
4075 {
4076 	struct nfsclclient *clp;
4077 	int matched;
4078 	static u_int32_t nextcbident = 0;
4079 	static int haswrapped = 0;
4080 
4081 	nextcbident++;
4082 	if (nextcbident == 0)
4083 		haswrapped = 1;
4084 	if (haswrapped) {
4085 		/*
4086 		 * Search the clientid list for one already using this cbident.
4087 		 */
4088 		do {
4089 			matched = 0;
4090 			NFSLOCKCLSTATE();
4091 			LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
4092 				if (clp->nfsc_cbident == nextcbident) {
4093 					matched = 1;
4094 					break;
4095 				}
4096 			}
4097 			NFSUNLOCKCLSTATE();
4098 			if (matched == 1)
4099 				nextcbident++;
4100 		} while (matched);
4101 	}
4102 	return (nextcbident);
4103 }
4104 
4105 /*
4106  * Get the mount point related to a given cbident or session and busy it.
4107  */
4108 static mount_t
nfscl_getmnt(int minorvers,uint8_t * sessionid,u_int32_t cbident,struct nfsclclient ** clpp)4109 nfscl_getmnt(int minorvers, uint8_t *sessionid, u_int32_t cbident,
4110     struct nfsclclient **clpp)
4111 {
4112 	struct nfsclclient *clp;
4113 	mount_t mp;
4114 	int error;
4115 	struct nfsclsession *tsep;
4116 
4117 	*clpp = NULL;
4118 	NFSLOCKCLSTATE();
4119 	LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
4120 		tsep = nfsmnt_mdssession(clp->nfsc_nmp);
4121 		if (minorvers == NFSV4_MINORVERSION) {
4122 			if (clp->nfsc_cbident == cbident)
4123 				break;
4124 		} else if (!NFSBCMP(tsep->nfsess_sessionid, sessionid,
4125 		    NFSX_V4SESSIONID))
4126 			break;
4127 	}
4128 	if (clp == NULL) {
4129 		NFSUNLOCKCLSTATE();
4130 		return (NULL);
4131 	}
4132 	mp = clp->nfsc_nmp->nm_mountp;
4133 	vfs_ref(mp);
4134 	NFSUNLOCKCLSTATE();
4135 	error = vfs_busy(mp, 0);
4136 	vfs_rel(mp);
4137 	if (error != 0)
4138 		return (NULL);
4139 	*clpp = clp;
4140 	return (mp);
4141 }
4142 
4143 /*
4144  * Get the clientid pointer related to a given cbident.
4145  */
4146 static struct nfsclclient *
nfscl_getclnt(u_int32_t cbident)4147 nfscl_getclnt(u_int32_t cbident)
4148 {
4149 	struct nfsclclient *clp;
4150 
4151 	LIST_FOREACH(clp, &nfsclhead, nfsc_list)
4152 		if (clp->nfsc_cbident == cbident)
4153 			break;
4154 	return (clp);
4155 }
4156 
4157 /*
4158  * Get the clientid pointer related to a given sessionid.
4159  */
4160 static struct nfsclclient *
nfscl_getclntsess(uint8_t * sessionid)4161 nfscl_getclntsess(uint8_t *sessionid)
4162 {
4163 	struct nfsclclient *clp;
4164 	struct nfsclsession *tsep;
4165 
4166 	LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
4167 		tsep = nfsmnt_mdssession(clp->nfsc_nmp);
4168 		if (!NFSBCMP(tsep->nfsess_sessionid, sessionid,
4169 		    NFSX_V4SESSIONID))
4170 			break;
4171 	}
4172 	return (clp);
4173 }
4174 
4175 /*
4176  * Search for a lock conflict locally on the client. A conflict occurs if
4177  * - not same owner and overlapping byte range and at least one of them is
4178  *   a write lock or this is an unlock.
4179  */
4180 static int
nfscl_localconflict(struct nfsclclient * clp,u_int8_t * fhp,int fhlen,struct nfscllock * nlop,u_int8_t * own,struct nfscldeleg * dp,struct nfscllock ** lopp)4181 nfscl_localconflict(struct nfsclclient *clp, u_int8_t *fhp, int fhlen,
4182     struct nfscllock *nlop, u_int8_t *own, struct nfscldeleg *dp,
4183     struct nfscllock **lopp)
4184 {
4185 	struct nfsclopen *op;
4186 	int ret;
4187 
4188 	if (dp != NULL) {
4189 		ret = nfscl_checkconflict(&dp->nfsdl_lock, nlop, own, lopp);
4190 		if (ret)
4191 			return (ret);
4192 	}
4193 	LIST_FOREACH(op, NFSCLOPENHASH(clp, fhp, fhlen), nfso_hash) {
4194 		if (op->nfso_fhlen == fhlen &&
4195 		    !NFSBCMP(op->nfso_fh, fhp, fhlen)) {
4196 			ret = nfscl_checkconflict(&op->nfso_lock, nlop,
4197 			    own, lopp);
4198 			if (ret)
4199 				return (ret);
4200 		}
4201 	}
4202 	return (0);
4203 }
4204 
4205 static int
nfscl_checkconflict(struct nfscllockownerhead * lhp,struct nfscllock * nlop,u_int8_t * own,struct nfscllock ** lopp)4206 nfscl_checkconflict(struct nfscllockownerhead *lhp, struct nfscllock *nlop,
4207     u_int8_t *own, struct nfscllock **lopp)
4208 {
4209 	struct nfscllockowner *lp;
4210 	struct nfscllock *lop;
4211 
4212 	LIST_FOREACH(lp, lhp, nfsl_list) {
4213 		if (NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
4214 			LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
4215 				if (lop->nfslo_first >= nlop->nfslo_end)
4216 					break;
4217 				if (lop->nfslo_end <= nlop->nfslo_first)
4218 					continue;
4219 				if (lop->nfslo_type == F_WRLCK ||
4220 				    nlop->nfslo_type == F_WRLCK ||
4221 				    nlop->nfslo_type == F_UNLCK) {
4222 					if (lopp != NULL)
4223 						*lopp = lop;
4224 					return (NFSERR_DENIED);
4225 				}
4226 			}
4227 		}
4228 	}
4229 	return (0);
4230 }
4231 
4232 /*
4233  * Check for a local conflicting lock.
4234  */
4235 int
nfscl_lockt(vnode_t vp,struct nfsclclient * clp,u_int64_t off,u_int64_t len,struct flock * fl,NFSPROC_T * p,void * id,int flags)4236 nfscl_lockt(vnode_t vp, struct nfsclclient *clp, u_int64_t off,
4237     u_int64_t len, struct flock *fl, NFSPROC_T *p, void *id, int flags)
4238 {
4239 	struct nfscllock *lop, nlck;
4240 	struct nfscldeleg *dp;
4241 	struct nfsnode *np;
4242 	u_int8_t own[NFSV4CL_LOCKNAMELEN];
4243 	int error;
4244 
4245 	nlck.nfslo_type = fl->l_type;
4246 	nlck.nfslo_first = off;
4247 	if (len == NFS64BITSSET) {
4248 		nlck.nfslo_end = NFS64BITSSET;
4249 	} else {
4250 		nlck.nfslo_end = off + len;
4251 		if (nlck.nfslo_end <= nlck.nfslo_first)
4252 			return (NFSERR_INVAL);
4253 	}
4254 	np = VTONFS(vp);
4255 	nfscl_filllockowner(id, own, flags);
4256 	NFSLOCKCLSTATE();
4257 	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4258 	error = nfscl_localconflict(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len,
4259 	    &nlck, own, dp, &lop);
4260 	if (error != 0) {
4261 		fl->l_whence = SEEK_SET;
4262 		fl->l_start = lop->nfslo_first;
4263 		if (lop->nfslo_end == NFS64BITSSET)
4264 			fl->l_len = 0;
4265 		else
4266 			fl->l_len = lop->nfslo_end - lop->nfslo_first;
4267 		fl->l_pid = (pid_t)0;
4268 		fl->l_type = lop->nfslo_type;
4269 		error = -1;			/* no RPC required */
4270 	} else if (dp != NULL && ((dp->nfsdl_flags & NFSCLDL_WRITE) ||
4271 	    fl->l_type == F_RDLCK)) {
4272 		/*
4273 		 * The delegation ensures that there isn't a conflicting
4274 		 * lock on the server, so return -1 to indicate an RPC
4275 		 * isn't required.
4276 		 */
4277 		fl->l_type = F_UNLCK;
4278 		error = -1;
4279 	}
4280 	NFSUNLOCKCLSTATE();
4281 	return (error);
4282 }
4283 
4284 /*
4285  * Handle Recall of a delegation.
4286  * The clp must be exclusive locked when this is called.
4287  */
4288 static int
nfscl_recalldeleg(struct nfsclclient * clp,struct nfsmount * nmp,struct nfscldeleg * dp,vnode_t vp,struct ucred * cred,NFSPROC_T * p,int called_from_renewthread,vnode_t * vpp)4289 nfscl_recalldeleg(struct nfsclclient *clp, struct nfsmount *nmp,
4290     struct nfscldeleg *dp, vnode_t vp, struct ucred *cred, NFSPROC_T *p,
4291     int called_from_renewthread, vnode_t *vpp)
4292 {
4293 	struct nfsclowner *owp, *lowp, *nowp;
4294 	struct nfsclopen *op, *lop;
4295 	struct nfscllockowner *lp;
4296 	struct nfscllock *lckp;
4297 	struct nfsnode *np;
4298 	int error = 0, ret;
4299 
4300 	if (vp == NULL) {
4301 		KASSERT(vpp != NULL, ("nfscl_recalldeleg: vpp NULL"));
4302 		*vpp = NULL;
4303 		/*
4304 		 * First, get a vnode for the file. This is needed to do RPCs.
4305 		 */
4306 		ret = nfscl_ngetreopen(nmp->nm_mountp, dp->nfsdl_fh,
4307 		    dp->nfsdl_fhlen, p, &np);
4308 		if (ret) {
4309 			/*
4310 			 * File isn't open, so nothing to move over to the
4311 			 * server.
4312 			 */
4313 			return (0);
4314 		}
4315 		vp = NFSTOV(np);
4316 		*vpp = vp;
4317 	} else {
4318 		np = VTONFS(vp);
4319 	}
4320 	dp->nfsdl_flags &= ~NFSCLDL_MODTIMESET;
4321 
4322 	/*
4323 	 * Ok, if it's a write delegation, flush data to the server, so
4324 	 * that close/open consistency is retained.
4325 	 */
4326 	ret = 0;
4327 	NFSLOCKNODE(np);
4328 	if ((dp->nfsdl_flags & NFSCLDL_WRITE) && (np->n_flag & NMODIFIED)) {
4329 		np->n_flag |= NDELEGRECALL;
4330 		NFSUNLOCKNODE(np);
4331 		ret = ncl_flush(vp, MNT_WAIT, p, 1, called_from_renewthread);
4332 		NFSLOCKNODE(np);
4333 		np->n_flag &= ~NDELEGRECALL;
4334 	}
4335 	NFSINVALATTRCACHE(np);
4336 	NFSUNLOCKNODE(np);
4337 	if (ret == EIO && called_from_renewthread != 0) {
4338 		/*
4339 		 * If the flush failed with EIO for the renew thread,
4340 		 * return now, so that the dirty buffer will be flushed
4341 		 * later.
4342 		 */
4343 		return (ret);
4344 	}
4345 
4346 	/*
4347 	 * Now, for each openowner with opens issued locally, move them
4348 	 * over to state against the server.
4349 	 */
4350 	LIST_FOREACH(lowp, &dp->nfsdl_owner, nfsow_list) {
4351 		lop = LIST_FIRST(&lowp->nfsow_open);
4352 		if (lop != NULL) {
4353 			if (LIST_NEXT(lop, nfso_list) != NULL)
4354 				panic("nfsdlg mult opens");
4355 			/*
4356 			 * Look for the same openowner against the server.
4357 			 */
4358 			LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
4359 				if (!NFSBCMP(lowp->nfsow_owner,
4360 				    owp->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
4361 					newnfs_copycred(&dp->nfsdl_cred, cred);
4362 					ret = nfscl_moveopen(vp, clp, nmp, lop,
4363 					    owp, dp, cred, p);
4364 					if (ret == NFSERR_STALECLIENTID ||
4365 					    ret == NFSERR_STALEDONTRECOVER ||
4366 					    ret == NFSERR_BADSESSION)
4367 						return (ret);
4368 					if (ret) {
4369 						nfscl_freeopen(lop, 1, true);
4370 						if (!error)
4371 							error = ret;
4372 					}
4373 					break;
4374 				}
4375 			}
4376 
4377 			/*
4378 			 * If no openowner found, create one and get an open
4379 			 * for it.
4380 			 */
4381 			if (owp == NULL) {
4382 				nowp = malloc(
4383 				    sizeof (struct nfsclowner), M_NFSCLOWNER,
4384 				    M_WAITOK);
4385 				nfscl_newopen(clp, NULL, &owp, &nowp, &op,
4386 				    NULL, lowp->nfsow_owner, dp->nfsdl_fh,
4387 				    dp->nfsdl_fhlen, NULL, NULL);
4388 				newnfs_copycred(&dp->nfsdl_cred, cred);
4389 				ret = nfscl_moveopen(vp, clp, nmp, lop,
4390 				    owp, dp, cred, p);
4391 				if (ret) {
4392 					nfscl_freeopenowner(owp, 0);
4393 					if (ret == NFSERR_STALECLIENTID ||
4394 					    ret == NFSERR_STALEDONTRECOVER ||
4395 					    ret == NFSERR_BADSESSION)
4396 						return (ret);
4397 					if (ret) {
4398 						nfscl_freeopen(lop, 1, true);
4399 						if (!error)
4400 							error = ret;
4401 					}
4402 				}
4403 			}
4404 		}
4405 	}
4406 
4407 	/*
4408 	 * Now, get byte range locks for any locks done locally.
4409 	 */
4410 	LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4411 		LIST_FOREACH(lckp, &lp->nfsl_lock, nfslo_list) {
4412 			newnfs_copycred(&dp->nfsdl_cred, cred);
4413 			ret = nfscl_relock(vp, clp, nmp, lp, lckp, cred, p);
4414 			if (ret == NFSERR_STALESTATEID ||
4415 			    ret == NFSERR_STALEDONTRECOVER ||
4416 			    ret == NFSERR_STALECLIENTID ||
4417 			    ret == NFSERR_BADSESSION)
4418 				return (ret);
4419 			if (ret && !error)
4420 				error = ret;
4421 		}
4422 	}
4423 	return (error);
4424 }
4425 
4426 /*
4427  * Move a locally issued open over to an owner on the state list.
4428  * SIDE EFFECT: If it needs to sleep (do an rpc), it unlocks clstate and
4429  * returns with it unlocked.
4430  */
4431 static int
nfscl_moveopen(vnode_t vp,struct nfsclclient * clp,struct nfsmount * nmp,struct nfsclopen * lop,struct nfsclowner * owp,struct nfscldeleg * dp,struct ucred * cred,NFSPROC_T * p)4432 nfscl_moveopen(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp,
4433     struct nfsclopen *lop, struct nfsclowner *owp, struct nfscldeleg *dp,
4434     struct ucred *cred, NFSPROC_T *p)
4435 {
4436 	struct nfsclopen *op, *nop;
4437 	struct nfscldeleg *ndp;
4438 	struct nfsnode *np;
4439 	int error = 0, newone;
4440 
4441 	/*
4442 	 * First, look for an appropriate open, If found, just increment the
4443 	 * opencnt in it.
4444 	 */
4445 	LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
4446 		if ((op->nfso_mode & lop->nfso_mode) == lop->nfso_mode &&
4447 		    op->nfso_fhlen == lop->nfso_fhlen &&
4448 		    !NFSBCMP(op->nfso_fh, lop->nfso_fh, op->nfso_fhlen)) {
4449 			op->nfso_opencnt += lop->nfso_opencnt;
4450 			nfscl_freeopen(lop, 1, true);
4451 			return (0);
4452 		}
4453 	}
4454 
4455 	/* No appropriate open, so we have to do one against the server. */
4456 	np = VTONFS(vp);
4457 	nop = malloc(sizeof (struct nfsclopen) +
4458 	    lop->nfso_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
4459 	nop->nfso_hash.le_prev = NULL;
4460 	newone = 0;
4461 	nfscl_newopen(clp, NULL, &owp, NULL, &op, &nop, owp->nfsow_owner,
4462 	    lop->nfso_fh, lop->nfso_fhlen, cred, &newone);
4463 	ndp = dp;
4464 	if (NFSHASNFSV4N(nmp))
4465 		error = nfscl_tryopen(nmp, vp, lop->nfso_fh, lop->nfso_fhlen,
4466 		    lop->nfso_fh, lop->nfso_fhlen, lop->nfso_mode, op,
4467 		    NULL, 0, &ndp, 0, 0, cred, p);
4468 	else
4469 		error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data,
4470 		    np->n_v4->n4_fhlen, lop->nfso_fh, lop->nfso_fhlen,
4471 		    lop->nfso_mode, op, NFS4NODENAME(np->n_v4),
4472 		    np->n_v4->n4_namelen, &ndp, 0, 0, cred, p);
4473 	if (error) {
4474 		if (newone)
4475 			nfscl_freeopen(op, 0, true);
4476 	} else {
4477 		op->nfso_mode |= lop->nfso_mode;
4478 		op->nfso_opencnt += lop->nfso_opencnt;
4479 		nfscl_freeopen(lop, 1, true);
4480 	}
4481 	if (nop != NULL)
4482 		free(nop, M_NFSCLOPEN);
4483 	if (ndp != NULL) {
4484 		/*
4485 		 * What should I do with the returned delegation, since the
4486 		 * delegation is being recalled? For now, just printf and
4487 		 * through it away.
4488 		 */
4489 		printf("Moveopen returned deleg\n");
4490 		free(ndp, M_NFSCLDELEG);
4491 	}
4492 	return (error);
4493 }
4494 
4495 /*
4496  * Recall all delegations on this client.
4497  */
4498 static void
nfscl_totalrecall(struct nfsclclient * clp)4499 nfscl_totalrecall(struct nfsclclient *clp)
4500 {
4501 	struct nfscldeleg *dp;
4502 
4503 	TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
4504 		if ((dp->nfsdl_flags & NFSCLDL_DELEGRET) == 0)
4505 			dp->nfsdl_flags |= NFSCLDL_RECALL;
4506 	}
4507 }
4508 
4509 /*
4510  * Relock byte ranges. Called for delegation recall and state expiry.
4511  */
4512 static int
nfscl_relock(vnode_t vp,struct nfsclclient * clp,struct nfsmount * nmp,struct nfscllockowner * lp,struct nfscllock * lop,struct ucred * cred,NFSPROC_T * p)4513 nfscl_relock(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp,
4514     struct nfscllockowner *lp, struct nfscllock *lop, struct ucred *cred,
4515     NFSPROC_T *p)
4516 {
4517 	struct nfscllockowner *nlp;
4518 	struct nfsfh *nfhp;
4519 	struct nfsnode *np;
4520 	u_int64_t off, len;
4521 	int error, newone, donelocally;
4522 
4523 	if (NFSHASNFSV4N(nmp) && NFSHASONEOPENOWN(nmp)) {
4524 		np = VTONFS(vp);
4525 		NFSLOCKNODE(np);
4526 		np->n_flag |= NMIGHTBELOCKED;
4527 		NFSUNLOCKNODE(np);
4528 	}
4529 
4530 	off = lop->nfslo_first;
4531 	len = lop->nfslo_end - lop->nfslo_first;
4532 	error = nfscl_getbytelock(vp, off, len, lop->nfslo_type, cred, p,
4533 	    clp, 1, NULL, lp->nfsl_lockflags, lp->nfsl_owner,
4534 	    lp->nfsl_openowner, &nlp, &newone, &donelocally);
4535 	if (error || donelocally)
4536 		return (error);
4537 	nfhp = VTONFS(vp)->n_fhp;
4538 	error = nfscl_trylock(nmp, vp, nfhp->nfh_fh,
4539 	    nfhp->nfh_len, nlp, newone, 0, off,
4540 	    len, lop->nfslo_type, cred, p);
4541 	if (error)
4542 		nfscl_freelockowner(nlp, 0);
4543 	return (error);
4544 }
4545 
4546 /*
4547  * Called to re-open a file. Basically get a vnode for the file handle
4548  * and then call nfsrpc_openrpc() to do the rest.
4549  */
4550 static int
nfsrpc_reopen(struct nfsmount * nmp,u_int8_t * fhp,int fhlen,u_int32_t mode,struct nfsclopen * op,struct nfscldeleg ** dpp,struct ucred * cred,NFSPROC_T * p)4551 nfsrpc_reopen(struct nfsmount *nmp, u_int8_t *fhp, int fhlen,
4552     u_int32_t mode, struct nfsclopen *op, struct nfscldeleg **dpp,
4553     struct ucred *cred, NFSPROC_T *p)
4554 {
4555 	struct nfsnode *np;
4556 	vnode_t vp;
4557 	int error;
4558 
4559 	error = nfscl_ngetreopen(nmp->nm_mountp, fhp, fhlen, p, &np);
4560 	if (error)
4561 		return (error);
4562 	vp = NFSTOV(np);
4563 	if (NFSHASNFSV4N(nmp))
4564 		error = nfscl_tryopen(nmp, vp, fhp, fhlen, fhp, fhlen, mode, op,
4565 		    NULL, 0, dpp, 0, 0, cred, p);
4566 	else if (np->n_v4 != NULL)
4567 		error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data,
4568 		    np->n_v4->n4_fhlen, fhp, fhlen, mode, op,
4569 		    NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, dpp, 0, 0,
4570 		    cred, p);
4571 	else
4572 		error = EINVAL;
4573 	vrele(vp);
4574 	return (error);
4575 }
4576 
4577 /*
4578  * Try an open against the server. Just call nfsrpc_openrpc(), retrying while
4579  * NFSERR_DELAY. Also, try system credentials, if the passed in credentials
4580  * fail.
4581  */
4582 static int
nfscl_tryopen(struct nfsmount * nmp,vnode_t vp,u_int8_t * fhp,int fhlen,u_int8_t * newfhp,int newfhlen,u_int32_t mode,struct nfsclopen * op,u_int8_t * name,int namelen,struct nfscldeleg ** ndpp,int reclaim,u_int32_t delegtype,struct ucred * cred,NFSPROC_T * p)4583 nfscl_tryopen(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen,
4584     u_int8_t *newfhp, int newfhlen, u_int32_t mode, struct nfsclopen *op,
4585     u_int8_t *name, int namelen, struct nfscldeleg **ndpp,
4586     int reclaim, u_int32_t delegtype, struct ucred *cred, NFSPROC_T *p)
4587 {
4588 	int error;
4589 	struct nfscldeleg *dp;
4590 
4591 	dp = *ndpp;
4592 	do {
4593 		*ndpp = dp;	/* *ndpp needs to be set for retries. */
4594 		error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp, newfhlen,
4595 		    mode, op, name, namelen, ndpp, reclaim, delegtype, cred, p,
4596 		    0, 0);
4597 		if (error == NFSERR_DELAY)
4598 			(void) nfs_catnap(PZERO, error, "nfstryop");
4599 	} while (error == NFSERR_DELAY);
4600 	if (error == EAUTH || error == EACCES) {
4601 		/* Try again using system credentials */
4602 		newnfs_setroot(cred);
4603 		do {
4604 		    *ndpp = dp;	/* *ndpp needs to be set for retries. */
4605 		    error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp,
4606 			newfhlen, mode, op, name, namelen, ndpp, reclaim,
4607 			delegtype, cred, p, 1, 0);
4608 		    if (error == NFSERR_DELAY)
4609 			(void) nfs_catnap(PZERO, error, "nfstryop");
4610 		} while (error == NFSERR_DELAY);
4611 	}
4612 	return (error);
4613 }
4614 
4615 /*
4616  * Try a byte range lock. Just loop on nfsrpc_lock() while it returns
4617  * NFSERR_DELAY. Also, retry with system credentials, if the provided
4618  * cred don't work.
4619  */
4620 static int
nfscl_trylock(struct nfsmount * nmp,vnode_t vp,u_int8_t * fhp,int fhlen,struct nfscllockowner * nlp,int newone,int reclaim,u_int64_t off,u_int64_t len,short type,struct ucred * cred,NFSPROC_T * p)4621 nfscl_trylock(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp,
4622     int fhlen, struct nfscllockowner *nlp, int newone, int reclaim,
4623     u_int64_t off, u_int64_t len, short type, struct ucred *cred, NFSPROC_T *p)
4624 {
4625 	struct nfsrv_descript nfsd, *nd = &nfsd;
4626 	int error;
4627 
4628 	do {
4629 		error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp, newone,
4630 		    reclaim, off, len, type, cred, p, 0);
4631 		if (!error && nd->nd_repstat == NFSERR_DELAY)
4632 			(void) nfs_catnap(PZERO, (int)nd->nd_repstat,
4633 			    "nfstrylck");
4634 	} while (!error && nd->nd_repstat == NFSERR_DELAY);
4635 	if (!error)
4636 		error = nd->nd_repstat;
4637 	if (error == EAUTH || error == EACCES) {
4638 		/* Try again using root credentials */
4639 		newnfs_setroot(cred);
4640 		do {
4641 			error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp,
4642 			    newone, reclaim, off, len, type, cred, p, 1);
4643 			if (!error && nd->nd_repstat == NFSERR_DELAY)
4644 				(void) nfs_catnap(PZERO, (int)nd->nd_repstat,
4645 				    "nfstrylck");
4646 		} while (!error && nd->nd_repstat == NFSERR_DELAY);
4647 		if (!error)
4648 			error = nd->nd_repstat;
4649 	}
4650 	return (error);
4651 }
4652 
4653 /*
4654  * Try a delegreturn against the server. Just call nfsrpc_delegreturn(),
4655  * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
4656  * credentials fail.
4657  */
4658 int
nfscl_trydelegreturn(struct nfscldeleg * dp,struct ucred * cred,struct nfsmount * nmp,NFSPROC_T * p)4659 nfscl_trydelegreturn(struct nfscldeleg *dp, struct ucred *cred,
4660     struct nfsmount *nmp, NFSPROC_T *p)
4661 {
4662 	int error;
4663 
4664 	do {
4665 		error = nfsrpc_delegreturn(dp, cred, nmp, p, 0);
4666 		if (error == NFSERR_DELAY)
4667 			(void) nfs_catnap(PZERO, error, "nfstrydp");
4668 	} while (error == NFSERR_DELAY);
4669 	if (error == EAUTH || error == EACCES) {
4670 		/* Try again using system credentials */
4671 		newnfs_setroot(cred);
4672 		do {
4673 			error = nfsrpc_delegreturn(dp, cred, nmp, p, 1);
4674 			if (error == NFSERR_DELAY)
4675 				(void) nfs_catnap(PZERO, error, "nfstrydp");
4676 		} while (error == NFSERR_DELAY);
4677 	}
4678 	return (error);
4679 }
4680 
4681 /*
4682  * Try a close against the server. Just call nfsrpc_closerpc(),
4683  * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
4684  * credentials fail.
4685  */
4686 int
nfscl_tryclose(struct nfsclopen * op,struct ucred * cred,struct nfsmount * nmp,NFSPROC_T * p,bool loop_on_delayed)4687 nfscl_tryclose(struct nfsclopen *op, struct ucred *cred,
4688     struct nfsmount *nmp, NFSPROC_T *p, bool loop_on_delayed)
4689 {
4690 	struct nfsrv_descript nfsd, *nd = &nfsd;
4691 	int error;
4692 
4693 	do {
4694 		error = nfsrpc_closerpc(nd, nmp, op, cred, p, 0);
4695 		if (loop_on_delayed && error == NFSERR_DELAY)
4696 			(void) nfs_catnap(PZERO, error, "nfstrycl");
4697 	} while (loop_on_delayed && error == NFSERR_DELAY);
4698 	if (error == EAUTH || error == EACCES) {
4699 		/* Try again using system credentials */
4700 		newnfs_setroot(cred);
4701 		do {
4702 			error = nfsrpc_closerpc(nd, nmp, op, cred, p, 1);
4703 			if (loop_on_delayed && error == NFSERR_DELAY)
4704 				(void) nfs_catnap(PZERO, error, "nfstrycl");
4705 		} while (loop_on_delayed && error == NFSERR_DELAY);
4706 	}
4707 	return (error);
4708 }
4709 
4710 /*
4711  * Decide if a delegation on a file permits close without flushing writes
4712  * to the server. This might be a big performance win in some environments.
4713  * (Not useful until the client does caching on local stable storage.)
4714  */
4715 int
nfscl_mustflush(vnode_t vp)4716 nfscl_mustflush(vnode_t vp)
4717 {
4718 	struct nfsclclient *clp;
4719 	struct nfscldeleg *dp;
4720 	struct nfsnode *np;
4721 	struct nfsmount *nmp;
4722 
4723 	np = VTONFS(vp);
4724 	nmp = VFSTONFS(vp->v_mount);
4725 	if (!NFSHASNFSV4(nmp) || vp->v_type != VREG)
4726 		return (1);
4727 	NFSLOCKMNT(nmp);
4728 	if ((nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0) {
4729 		NFSUNLOCKMNT(nmp);
4730 		return (1);
4731 	}
4732 	NFSUNLOCKMNT(nmp);
4733 	NFSLOCKCLSTATE();
4734 	clp = nfscl_findcl(nmp);
4735 	if (clp == NULL) {
4736 		NFSUNLOCKCLSTATE();
4737 		return (1);
4738 	}
4739 	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4740 	if (dp != NULL && (dp->nfsdl_flags &
4741 	    (NFSCLDL_WRITE | NFSCLDL_RECALL | NFSCLDL_DELEGRET)) ==
4742 	     NFSCLDL_WRITE &&
4743 	    (dp->nfsdl_sizelimit >= np->n_size ||
4744 	     !NFSHASSTRICT3530(nmp))) {
4745 		NFSUNLOCKCLSTATE();
4746 		return (0);
4747 	}
4748 	NFSUNLOCKCLSTATE();
4749 	return (1);
4750 }
4751 
4752 /*
4753  * See if a (write) delegation exists for this file.
4754  */
4755 int
nfscl_nodeleg(vnode_t vp,int writedeleg)4756 nfscl_nodeleg(vnode_t vp, int writedeleg)
4757 {
4758 	struct nfsclclient *clp;
4759 	struct nfscldeleg *dp;
4760 	struct nfsnode *np;
4761 	struct nfsmount *nmp;
4762 
4763 	np = VTONFS(vp);
4764 	nmp = VFSTONFS(vp->v_mount);
4765 	if (!NFSHASNFSV4(nmp) || vp->v_type != VREG)
4766 		return (1);
4767 	NFSLOCKMNT(nmp);
4768 	if ((nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0) {
4769 		NFSUNLOCKMNT(nmp);
4770 		return (1);
4771 	}
4772 	NFSUNLOCKMNT(nmp);
4773 	NFSLOCKCLSTATE();
4774 	clp = nfscl_findcl(nmp);
4775 	if (clp == NULL) {
4776 		NFSUNLOCKCLSTATE();
4777 		return (1);
4778 	}
4779 	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4780 	if (dp != NULL &&
4781 	    (dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_DELEGRET)) == 0 &&
4782 	    (writedeleg == 0 || (dp->nfsdl_flags & NFSCLDL_WRITE) ==
4783 	     NFSCLDL_WRITE)) {
4784 		NFSUNLOCKCLSTATE();
4785 		return (0);
4786 	}
4787 	NFSUNLOCKCLSTATE();
4788 	return (1);
4789 }
4790 
4791 /*
4792  * Look for an associated delegation that should be DelegReturned.
4793  */
4794 int
nfscl_removedeleg(vnode_t vp,NFSPROC_T * p,nfsv4stateid_t * stp)4795 nfscl_removedeleg(vnode_t vp, NFSPROC_T *p, nfsv4stateid_t *stp)
4796 {
4797 	struct nfsclclient *clp;
4798 	struct nfscldeleg *dp;
4799 	struct nfsclowner *owp;
4800 	struct nfscllockowner *lp;
4801 	struct nfsmount *nmp;
4802 	struct mount *mp;
4803 	struct ucred *cred;
4804 	struct nfsnode *np;
4805 	int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
4806 
4807 	nmp = VFSTONFS(vp->v_mount);
4808 	if (NFSHASPNFS(nmp))
4809 		return (retcnt);
4810 	NFSLOCKMNT(nmp);
4811 	if ((nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0) {
4812 		NFSUNLOCKMNT(nmp);
4813 		return (retcnt);
4814 	}
4815 	NFSUNLOCKMNT(nmp);
4816 	np = VTONFS(vp);
4817 	mp = nmp->nm_mountp;
4818 	NFSLOCKCLSTATE();
4819 	/*
4820 	 * Loop around waiting for:
4821 	 * - outstanding I/O operations on delegations to complete
4822 	 * - for a delegation on vp that has state, lock the client and
4823 	 *   do a recall
4824 	 * - return delegation with no state
4825 	 */
4826 	while (1) {
4827 		clp = nfscl_findcl(nmp);
4828 		if (clp == NULL) {
4829 			NFSUNLOCKCLSTATE();
4830 			return (retcnt);
4831 		}
4832 		dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4833 		    np->n_fhp->nfh_len);
4834 		if (dp != NULL) {
4835 		    /*
4836 		     * Wait for outstanding I/O ops to be done.
4837 		     */
4838 		    if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4839 			if (igotlock) {
4840 			    nfsv4_unlock(&clp->nfsc_lock, 0);
4841 			    igotlock = 0;
4842 			}
4843 			dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4844 			msleep(&dp->nfsdl_rwlock, NFSCLSTATEMUTEXPTR, PZERO,
4845 			    "nfscld", hz);
4846 			if (NFSCL_FORCEDISM(mp)) {
4847 			    dp->nfsdl_flags &= ~NFSCLDL_DELEGRET;
4848 			    NFSUNLOCKCLSTATE();
4849 			    return (0);
4850 			}
4851 			continue;
4852 		    }
4853 		    needsrecall = 0;
4854 		    LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4855 			if (!LIST_EMPTY(&owp->nfsow_open)) {
4856 			    needsrecall = 1;
4857 			    break;
4858 			}
4859 		    }
4860 		    if (!needsrecall) {
4861 			LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4862 			    if (!LIST_EMPTY(&lp->nfsl_lock)) {
4863 				needsrecall = 1;
4864 				break;
4865 			    }
4866 			}
4867 		    }
4868 		    if (needsrecall && !triedrecall) {
4869 			dp->nfsdl_flags |= NFSCLDL_DELEGRET;
4870 			islept = 0;
4871 			while (!igotlock) {
4872 			    igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
4873 				&islept, NFSCLSTATEMUTEXPTR, mp);
4874 			    if (NFSCL_FORCEDISM(mp)) {
4875 				dp->nfsdl_flags &= ~NFSCLDL_DELEGRET;
4876 				if (igotlock)
4877 				    nfsv4_unlock(&clp->nfsc_lock, 0);
4878 				NFSUNLOCKCLSTATE();
4879 				return (0);
4880 			    }
4881 			    if (islept)
4882 				break;
4883 			}
4884 			if (islept)
4885 			    continue;
4886 			NFSUNLOCKCLSTATE();
4887 			cred = newnfs_getcred();
4888 			newnfs_copycred(&dp->nfsdl_cred, cred);
4889 			nfscl_recalldeleg(clp, nmp, dp, vp, cred, p, 0, NULL);
4890 			NFSFREECRED(cred);
4891 			triedrecall = 1;
4892 			NFSLOCKCLSTATE();
4893 			nfsv4_unlock(&clp->nfsc_lock, 0);
4894 			igotlock = 0;
4895 			continue;
4896 		    }
4897 		    *stp = dp->nfsdl_stateid;
4898 		    retcnt = 1;
4899 		    nfscl_cleandeleg(dp);
4900 		    nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
4901 		}
4902 		if (igotlock)
4903 		    nfsv4_unlock(&clp->nfsc_lock, 0);
4904 		NFSUNLOCKCLSTATE();
4905 		return (retcnt);
4906 	}
4907 }
4908 
4909 /*
4910  * Look for associated delegation(s) that should be DelegReturned.
4911  */
4912 int
nfscl_renamedeleg(vnode_t fvp,nfsv4stateid_t * fstp,int * gotfdp,vnode_t tvp,nfsv4stateid_t * tstp,int * gottdp,NFSPROC_T * p)4913 nfscl_renamedeleg(vnode_t fvp, nfsv4stateid_t *fstp, int *gotfdp, vnode_t tvp,
4914     nfsv4stateid_t *tstp, int *gottdp, NFSPROC_T *p)
4915 {
4916 	struct nfsclclient *clp;
4917 	struct nfscldeleg *dp;
4918 	struct nfsclowner *owp;
4919 	struct nfscllockowner *lp;
4920 	struct nfsmount *nmp;
4921 	struct mount *mp;
4922 	struct ucred *cred;
4923 	struct nfsnode *np;
4924 	int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
4925 
4926 	nmp = VFSTONFS(fvp->v_mount);
4927 	*gotfdp = 0;
4928 	*gottdp = 0;
4929 	if (NFSHASPNFS(nmp))
4930 		return (retcnt);
4931 	NFSLOCKMNT(nmp);
4932 	if ((nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0) {
4933 		NFSUNLOCKMNT(nmp);
4934 		return (retcnt);
4935 	}
4936 	NFSUNLOCKMNT(nmp);
4937 	mp = nmp->nm_mountp;
4938 	NFSLOCKCLSTATE();
4939 	/*
4940 	 * Loop around waiting for:
4941 	 * - outstanding I/O operations on delegations to complete
4942 	 * - for a delegation on fvp that has state, lock the client and
4943 	 *   do a recall
4944 	 * - return delegation(s) with no state.
4945 	 */
4946 	while (1) {
4947 		clp = nfscl_findcl(nmp);
4948 		if (clp == NULL) {
4949 			NFSUNLOCKCLSTATE();
4950 			return (retcnt);
4951 		}
4952 		np = VTONFS(fvp);
4953 		dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4954 		    np->n_fhp->nfh_len);
4955 		if (dp != NULL && *gotfdp == 0) {
4956 		    /*
4957 		     * Wait for outstanding I/O ops to be done.
4958 		     */
4959 		    if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4960 			if (igotlock) {
4961 			    nfsv4_unlock(&clp->nfsc_lock, 0);
4962 			    igotlock = 0;
4963 			}
4964 			dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4965 			msleep(&dp->nfsdl_rwlock, NFSCLSTATEMUTEXPTR, PZERO,
4966 			    "nfscld", hz);
4967 			if (NFSCL_FORCEDISM(mp)) {
4968 			    dp->nfsdl_flags &= ~NFSCLDL_DELEGRET;
4969 			    NFSUNLOCKCLSTATE();
4970 			    *gotfdp = 0;
4971 			    *gottdp = 0;
4972 			    return (0);
4973 			}
4974 			continue;
4975 		    }
4976 		    needsrecall = 0;
4977 		    LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4978 			if (!LIST_EMPTY(&owp->nfsow_open)) {
4979 			    needsrecall = 1;
4980 			    break;
4981 			}
4982 		    }
4983 		    if (!needsrecall) {
4984 			LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4985 			    if (!LIST_EMPTY(&lp->nfsl_lock)) {
4986 				needsrecall = 1;
4987 				break;
4988 			    }
4989 			}
4990 		    }
4991 		    if (needsrecall && !triedrecall) {
4992 			dp->nfsdl_flags |= NFSCLDL_DELEGRET;
4993 			islept = 0;
4994 			while (!igotlock) {
4995 			    igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
4996 				&islept, NFSCLSTATEMUTEXPTR, mp);
4997 			    if (NFSCL_FORCEDISM(mp)) {
4998 				dp->nfsdl_flags &= ~NFSCLDL_DELEGRET;
4999 				if (igotlock)
5000 				    nfsv4_unlock(&clp->nfsc_lock, 0);
5001 				NFSUNLOCKCLSTATE();
5002 				*gotfdp = 0;
5003 				*gottdp = 0;
5004 				return (0);
5005 			    }
5006 			    if (islept)
5007 				break;
5008 			}
5009 			if (islept)
5010 			    continue;
5011 			NFSUNLOCKCLSTATE();
5012 			cred = newnfs_getcred();
5013 			newnfs_copycred(&dp->nfsdl_cred, cred);
5014 			nfscl_recalldeleg(clp, nmp, dp, fvp, cred, p, 0, NULL);
5015 			NFSFREECRED(cred);
5016 			triedrecall = 1;
5017 			NFSLOCKCLSTATE();
5018 			nfsv4_unlock(&clp->nfsc_lock, 0);
5019 			igotlock = 0;
5020 			continue;
5021 		    }
5022 		    *fstp = dp->nfsdl_stateid;
5023 		    retcnt++;
5024 		    *gotfdp = 1;
5025 		    nfscl_cleandeleg(dp);
5026 		    nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
5027 		}
5028 		if (igotlock) {
5029 		    nfsv4_unlock(&clp->nfsc_lock, 0);
5030 		    igotlock = 0;
5031 		}
5032 		if (tvp != NULL) {
5033 		    np = VTONFS(tvp);
5034 		    dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
5035 			np->n_fhp->nfh_len);
5036 		    if (dp != NULL && *gottdp == 0) {
5037 			/*
5038 			 * Wait for outstanding I/O ops to be done.
5039 			 */
5040 			if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
5041 			    dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
5042 			    msleep(&dp->nfsdl_rwlock, NFSCLSTATEMUTEXPTR, PZERO,
5043 				"nfscld", hz);
5044 			    if (NFSCL_FORCEDISM(mp)) {
5045 				NFSUNLOCKCLSTATE();
5046 				*gotfdp = 0;
5047 				*gottdp = 0;
5048 				return (0);
5049 			    }
5050 			    continue;
5051 			}
5052 			LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
5053 			    if (!LIST_EMPTY(&owp->nfsow_open)) {
5054 				NFSUNLOCKCLSTATE();
5055 				return (retcnt);
5056 			    }
5057 			}
5058 			LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
5059 			    if (!LIST_EMPTY(&lp->nfsl_lock)) {
5060 				NFSUNLOCKCLSTATE();
5061 				return (retcnt);
5062 			    }
5063 			}
5064 			*tstp = dp->nfsdl_stateid;
5065 			retcnt++;
5066 			*gottdp = 1;
5067 			nfscl_cleandeleg(dp);
5068 			nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
5069 		    }
5070 		}
5071 		NFSUNLOCKCLSTATE();
5072 		return (retcnt);
5073 	}
5074 }
5075 
5076 /*
5077  * Get a reference on the clientid associated with the mount point.
5078  * Return 1 if success, 0 otherwise.
5079  */
5080 int
nfscl_getref(struct nfsmount * nmp)5081 nfscl_getref(struct nfsmount *nmp)
5082 {
5083 	struct nfsclclient *clp;
5084 	int ret;
5085 
5086 	NFSLOCKCLSTATE();
5087 	clp = nfscl_findcl(nmp);
5088 	if (clp == NULL) {
5089 		NFSUNLOCKCLSTATE();
5090 		return (0);
5091 	}
5092 	nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, nmp->nm_mountp);
5093 	ret = 1;
5094 	if (NFSCL_FORCEDISM(nmp->nm_mountp))
5095 		ret = 0;
5096 	NFSUNLOCKCLSTATE();
5097 	return (ret);
5098 }
5099 
5100 /*
5101  * Release a reference on a clientid acquired with the above call.
5102  */
5103 void
nfscl_relref(struct nfsmount * nmp)5104 nfscl_relref(struct nfsmount *nmp)
5105 {
5106 	struct nfsclclient *clp;
5107 
5108 	NFSLOCKCLSTATE();
5109 	clp = nfscl_findcl(nmp);
5110 	if (clp == NULL) {
5111 		NFSUNLOCKCLSTATE();
5112 		return;
5113 	}
5114 	nfsv4_relref(&clp->nfsc_lock);
5115 	NFSUNLOCKCLSTATE();
5116 }
5117 
5118 /*
5119  * Save the size attribute in the delegation, since the nfsnode
5120  * is going away.
5121  */
5122 void
nfscl_reclaimnode(vnode_t vp)5123 nfscl_reclaimnode(vnode_t vp)
5124 {
5125 	struct nfsclclient *clp;
5126 	struct nfscldeleg *dp;
5127 	struct nfsnode *np = VTONFS(vp);
5128 	struct nfsmount *nmp;
5129 
5130 	nmp = VFSTONFS(vp->v_mount);
5131 	if (!NFSHASNFSV4(nmp))
5132 		return;
5133 	NFSLOCKCLSTATE();
5134 	clp = nfscl_findcl(nmp);
5135 	if (clp == NULL) {
5136 		NFSUNLOCKCLSTATE();
5137 		return;
5138 	}
5139 	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
5140 	if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE))
5141 		dp->nfsdl_size = np->n_size;
5142 	NFSUNLOCKCLSTATE();
5143 }
5144 
5145 /*
5146  * Get the saved size attribute in the delegation, since it is a
5147  * newly allocated nfsnode.
5148  */
5149 void
nfscl_newnode(vnode_t vp)5150 nfscl_newnode(vnode_t vp)
5151 {
5152 	struct nfsclclient *clp;
5153 	struct nfscldeleg *dp;
5154 	struct nfsnode *np = VTONFS(vp);
5155 	struct nfsmount *nmp;
5156 
5157 	nmp = VFSTONFS(vp->v_mount);
5158 	if (!NFSHASNFSV4(nmp))
5159 		return;
5160 	NFSLOCKCLSTATE();
5161 	clp = nfscl_findcl(nmp);
5162 	if (clp == NULL) {
5163 		NFSUNLOCKCLSTATE();
5164 		return;
5165 	}
5166 	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
5167 	if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE))
5168 		np->n_size = dp->nfsdl_size;
5169 	NFSUNLOCKCLSTATE();
5170 }
5171 
5172 /*
5173  * If there is a valid write delegation for this file, set the modtime
5174  * to the local clock time.
5175  */
5176 void
nfscl_delegmodtime(struct vnode * vp,struct timespec * mtime)5177 nfscl_delegmodtime(struct vnode *vp, struct timespec *mtime)
5178 {
5179 	struct nfsclclient *clp;
5180 	struct nfscldeleg *dp;
5181 	struct nfsnode *np = VTONFS(vp);
5182 	struct nfsmount *nmp;
5183 
5184 	nmp = VFSTONFS(vp->v_mount);
5185 	if (!NFSHASNFSV4(nmp))
5186 		return;
5187 	NFSLOCKMNT(nmp);
5188 	if ((nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0) {
5189 		NFSUNLOCKMNT(nmp);
5190 		return;
5191 	}
5192 	NFSUNLOCKMNT(nmp);
5193 	NFSLOCKCLSTATE();
5194 	clp = nfscl_findcl(nmp);
5195 	if (clp == NULL) {
5196 		NFSUNLOCKCLSTATE();
5197 		return;
5198 	}
5199 	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
5200 	if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE)) {
5201 		if (mtime != NULL)
5202 			dp->nfsdl_modtime = *mtime;
5203 		else
5204 			nanotime(&dp->nfsdl_modtime);
5205 		dp->nfsdl_flags |= NFSCLDL_MODTIMESET;
5206 	}
5207 	NFSUNLOCKCLSTATE();
5208 }
5209 
5210 /*
5211  * If there is a valid write delegation for this file with a modtime set,
5212  * put that modtime in mtime.
5213  */
5214 void
nfscl_deleggetmodtime(vnode_t vp,struct timespec * mtime)5215 nfscl_deleggetmodtime(vnode_t vp, struct timespec *mtime)
5216 {
5217 	struct nfsclclient *clp;
5218 	struct nfscldeleg *dp;
5219 	struct nfsnode *np = VTONFS(vp);
5220 	struct nfsmount *nmp;
5221 
5222 	nmp = VFSTONFS(vp->v_mount);
5223 	if (!NFSHASNFSV4(nmp))
5224 		return;
5225 	NFSLOCKMNT(nmp);
5226 	if ((nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0) {
5227 		NFSUNLOCKMNT(nmp);
5228 		return;
5229 	}
5230 	NFSUNLOCKMNT(nmp);
5231 	NFSLOCKCLSTATE();
5232 	clp = nfscl_findcl(nmp);
5233 	if (clp == NULL) {
5234 		NFSUNLOCKCLSTATE();
5235 		return;
5236 	}
5237 	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
5238 	if (dp != NULL &&
5239 	    (dp->nfsdl_flags & (NFSCLDL_WRITE | NFSCLDL_MODTIMESET)) ==
5240 	    (NFSCLDL_WRITE | NFSCLDL_MODTIMESET))
5241 		*mtime = dp->nfsdl_modtime;
5242 	NFSUNLOCKCLSTATE();
5243 }
5244 
5245 static int
nfscl_errmap(struct nfsrv_descript * nd,u_int32_t minorvers)5246 nfscl_errmap(struct nfsrv_descript *nd, u_int32_t minorvers)
5247 {
5248 	short *defaulterrp, *errp;
5249 
5250 	if (!nd->nd_repstat)
5251 		return (0);
5252 	if (nd->nd_procnum == NFSPROC_NOOP)
5253 		return (txdr_unsigned(nd->nd_repstat & 0xffff));
5254 	if (nd->nd_repstat == EBADRPC)
5255 		return (txdr_unsigned(NFSERR_BADXDR));
5256 	if (nd->nd_repstat == NFSERR_MINORVERMISMATCH ||
5257 	    nd->nd_repstat == NFSERR_OPILLEGAL)
5258 		return (txdr_unsigned(nd->nd_repstat));
5259 	if (nd->nd_repstat >= NFSERR_BADIOMODE && nd->nd_repstat < 20000 &&
5260 	    minorvers > NFSV4_MINORVERSION) {
5261 		/* NFSv4.n error. */
5262 		return (txdr_unsigned(nd->nd_repstat));
5263 	}
5264 	if (nd->nd_procnum < NFSV4OP_CBNOPS)
5265 		errp = defaulterrp = nfscl_cberrmap[nd->nd_procnum];
5266 	else
5267 		return (txdr_unsigned(nd->nd_repstat));
5268 	while (*++errp)
5269 		if (*errp == (short)nd->nd_repstat)
5270 			return (txdr_unsigned(nd->nd_repstat));
5271 	return (txdr_unsigned(*defaulterrp));
5272 }
5273 
5274 /*
5275  * Called to find/add a layout to a client.
5276  * This function returns the layout with a refcnt (shared lock) upon
5277  * success (returns 0) or with no lock/refcnt on the layout when an
5278  * error is returned.
5279  * If a layout is passed in via lypp, it is locked (exclusively locked).
5280  */
5281 int
nfscl_layout(struct nfsmount * nmp,vnode_t vp,u_int8_t * fhp,int fhlen,nfsv4stateid_t * stateidp,int layouttype,int retonclose,struct nfsclflayouthead * fhlp,struct nfscllayout ** lypp,struct ucred * cred,NFSPROC_T * p)5282 nfscl_layout(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen,
5283     nfsv4stateid_t *stateidp, int layouttype, int retonclose,
5284     struct nfsclflayouthead *fhlp, struct nfscllayout **lypp,
5285     struct ucred *cred, NFSPROC_T *p)
5286 {
5287 	struct nfsclclient *clp;
5288 	struct nfscllayout *lyp, *tlyp;
5289 	struct nfsclflayout *flp;
5290 	struct nfsnode *np = VTONFS(vp);
5291 	mount_t mp;
5292 	int layout_passed_in;
5293 
5294 	mp = nmp->nm_mountp;
5295 	layout_passed_in = 1;
5296 	tlyp = NULL;
5297 	lyp = *lypp;
5298 	if (lyp == NULL) {
5299 		layout_passed_in = 0;
5300 		tlyp = malloc(sizeof(*tlyp) + fhlen - 1, M_NFSLAYOUT,
5301 		    M_WAITOK | M_ZERO);
5302 	}
5303 
5304 	NFSLOCKCLSTATE();
5305 	clp = nmp->nm_clp;
5306 	if (clp == NULL) {
5307 		if (layout_passed_in != 0)
5308 			nfsv4_unlock(&lyp->nfsly_lock, 0);
5309 		NFSUNLOCKCLSTATE();
5310 		if (tlyp != NULL)
5311 			free(tlyp, M_NFSLAYOUT);
5312 		return (EPERM);
5313 	}
5314 	if (lyp == NULL) {
5315 		/*
5316 		 * Although no lyp was passed in, another thread might have
5317 		 * allocated one. If one is found, just increment it's ref
5318 		 * count and return it.
5319 		 */
5320 		lyp = nfscl_findlayout(clp, fhp, fhlen);
5321 		if (lyp == NULL) {
5322 			lyp = tlyp;
5323 			tlyp = NULL;
5324 			lyp->nfsly_stateid.seqid = stateidp->seqid;
5325 			lyp->nfsly_stateid.other[0] = stateidp->other[0];
5326 			lyp->nfsly_stateid.other[1] = stateidp->other[1];
5327 			lyp->nfsly_stateid.other[2] = stateidp->other[2];
5328 			lyp->nfsly_lastbyte = 0;
5329 			LIST_INIT(&lyp->nfsly_flayread);
5330 			LIST_INIT(&lyp->nfsly_flayrw);
5331 			LIST_INIT(&lyp->nfsly_recall);
5332 			lyp->nfsly_filesid[0] = np->n_vattr.na_filesid[0];
5333 			lyp->nfsly_filesid[1] = np->n_vattr.na_filesid[1];
5334 			lyp->nfsly_clp = clp;
5335 			if (layouttype == NFSLAYOUT_FLEXFILE)
5336 				lyp->nfsly_flags = NFSLY_FLEXFILE;
5337 			else
5338 				lyp->nfsly_flags = NFSLY_FILES;
5339 			if (retonclose != 0)
5340 				lyp->nfsly_flags |= NFSLY_RETONCLOSE;
5341 			lyp->nfsly_fhlen = fhlen;
5342 			NFSBCOPY(fhp, lyp->nfsly_fh, fhlen);
5343 			TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
5344 			LIST_INSERT_HEAD(NFSCLLAYOUTHASH(clp, fhp, fhlen), lyp,
5345 			    nfsly_hash);
5346 			lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
5347 			clp->nfsc_layoutcnt++;
5348 			nfsstatsv1.cllayouts++;
5349 		} else {
5350 			if (retonclose != 0)
5351 				lyp->nfsly_flags |= NFSLY_RETONCLOSE;
5352 			if (stateidp->seqid > lyp->nfsly_stateid.seqid)
5353 				lyp->nfsly_stateid.seqid = stateidp->seqid;
5354 			TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list);
5355 			TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
5356 			lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
5357 		}
5358 		nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
5359 		if (NFSCL_FORCEDISM(mp)) {
5360 			NFSUNLOCKCLSTATE();
5361 			if (tlyp != NULL)
5362 				free(tlyp, M_NFSLAYOUT);
5363 			return (EPERM);
5364 		}
5365 		*lypp = lyp;
5366 	} else if (stateidp->seqid > lyp->nfsly_stateid.seqid)
5367 		lyp->nfsly_stateid.seqid = stateidp->seqid;
5368 
5369 	/* Merge the new list of File Layouts into the list. */
5370 	flp = LIST_FIRST(fhlp);
5371 	if (flp != NULL) {
5372 		if (flp->nfsfl_iomode == NFSLAYOUTIOMODE_READ)
5373 			nfscl_mergeflayouts(&lyp->nfsly_flayread, fhlp);
5374 		else
5375 			nfscl_mergeflayouts(&lyp->nfsly_flayrw, fhlp);
5376 	}
5377 	if (layout_passed_in != 0)
5378 		nfsv4_unlock(&lyp->nfsly_lock, 1);
5379 	NFSUNLOCKCLSTATE();
5380 	if (tlyp != NULL)
5381 		free(tlyp, M_NFSLAYOUT);
5382 	return (0);
5383 }
5384 
5385 /*
5386  * Search for a layout by MDS file handle.
5387  * If one is found, it is returned with a refcnt (shared lock) iff
5388  * retflpp returned non-NULL and locked (exclusive locked) iff retflpp is
5389  * returned NULL.
5390  */
5391 struct nfscllayout *
nfscl_getlayout(struct nfsclclient * clp,uint8_t * fhp,int fhlen,uint64_t off,uint32_t rwaccess,struct nfsclflayout ** retflpp,int * recalledp)5392 nfscl_getlayout(struct nfsclclient *clp, uint8_t *fhp, int fhlen,
5393     uint64_t off, uint32_t rwaccess, struct nfsclflayout **retflpp,
5394     int *recalledp)
5395 {
5396 	struct nfscllayout *lyp;
5397 	mount_t mp;
5398 	int error, igotlock;
5399 
5400 	mp = clp->nfsc_nmp->nm_mountp;
5401 	*recalledp = 0;
5402 	*retflpp = NULL;
5403 	NFSLOCKCLSTATE();
5404 	lyp = nfscl_findlayout(clp, fhp, fhlen);
5405 	if (lyp != NULL) {
5406 		if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
5407 			TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list);
5408 			TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
5409 			lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
5410 			error = nfscl_findlayoutforio(lyp, off, rwaccess,
5411 			    retflpp);
5412 			if (error == 0)
5413 				nfsv4_getref(&lyp->nfsly_lock, NULL,
5414 				    NFSCLSTATEMUTEXPTR, mp);
5415 			else {
5416 				do {
5417 					igotlock = nfsv4_lock(&lyp->nfsly_lock,
5418 					    1, NULL, NFSCLSTATEMUTEXPTR, mp);
5419 				} while (igotlock == 0 && !NFSCL_FORCEDISM(mp));
5420 				*retflpp = NULL;
5421 			}
5422 			if (NFSCL_FORCEDISM(mp)) {
5423 				lyp = NULL;
5424 				*recalledp = 1;
5425 			}
5426 		} else {
5427 			lyp = NULL;
5428 			*recalledp = 1;
5429 		}
5430 	}
5431 	NFSUNLOCKCLSTATE();
5432 	return (lyp);
5433 }
5434 
5435 /*
5436  * Search for a layout by MDS file handle. If one is found, mark in to be
5437  * recalled, if it already marked "return on close".
5438  */
5439 static void
nfscl_retoncloselayout(vnode_t vp,struct nfsclclient * clp,uint8_t * fhp,int fhlen,struct nfsclrecalllayout ** recallpp,struct nfscllayout ** lypp)5440 nfscl_retoncloselayout(vnode_t vp, struct nfsclclient *clp, uint8_t *fhp,
5441     int fhlen, struct nfsclrecalllayout **recallpp, struct nfscllayout **lypp)
5442 {
5443 	struct nfscllayout *lyp;
5444 	uint32_t iomode;
5445 
5446 	*lypp = NULL;
5447 	if (vp->v_type != VREG || !NFSHASPNFS(VFSTONFS(vp->v_mount)) ||
5448 	    nfscl_enablecallb == 0 || nfs_numnfscbd == 0 ||
5449 	    (VTONFS(vp)->n_flag & NNOLAYOUT) != 0)
5450 		return;
5451 	lyp = nfscl_findlayout(clp, fhp, fhlen);
5452 	if (lyp != NULL && (lyp->nfsly_flags & NFSLY_RETONCLOSE) != 0) {
5453 		if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
5454 			iomode = 0;
5455 			if (!LIST_EMPTY(&lyp->nfsly_flayread))
5456 				iomode |= NFSLAYOUTIOMODE_READ;
5457 			if (!LIST_EMPTY(&lyp->nfsly_flayrw))
5458 				iomode |= NFSLAYOUTIOMODE_RW;
5459 			nfscl_layoutrecall(NFSLAYOUTRETURN_FILE, lyp, iomode,
5460 			    0, UINT64_MAX, lyp->nfsly_stateid.seqid, 0, 0, NULL,
5461 			    *recallpp);
5462 			NFSCL_DEBUG(4, "retoncls recall iomode=%d\n", iomode);
5463 			*recallpp = NULL;
5464 		}
5465 
5466 		/* Now, wake up renew thread to do LayoutReturn. */
5467 		wakeup(clp);
5468 		*lypp = lyp;
5469 	}
5470 }
5471 
5472 /*
5473  * Mark the layout to be recalled and with an error.
5474  * Also, disable the dsp from further use.
5475  */
5476 void
nfscl_dserr(uint32_t op,uint32_t stat,struct nfscldevinfo * dp,struct nfscllayout * lyp,struct nfsclds * dsp)5477 nfscl_dserr(uint32_t op, uint32_t stat, struct nfscldevinfo *dp,
5478     struct nfscllayout *lyp, struct nfsclds *dsp)
5479 {
5480 	struct nfsclrecalllayout *recallp;
5481 	uint32_t iomode;
5482 
5483 	printf("DS being disabled, error=%d\n", stat);
5484 	/* Set up the return of the layout. */
5485 	recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_WAITOK);
5486 	iomode = 0;
5487 	NFSLOCKCLSTATE();
5488 	if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
5489 		if (!LIST_EMPTY(&lyp->nfsly_flayread))
5490 			iomode |= NFSLAYOUTIOMODE_READ;
5491 		if (!LIST_EMPTY(&lyp->nfsly_flayrw))
5492 			iomode |= NFSLAYOUTIOMODE_RW;
5493 		(void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE, lyp, iomode,
5494 		    0, UINT64_MAX, lyp->nfsly_stateid.seqid, stat, op,
5495 		    dp->nfsdi_deviceid, recallp);
5496 		NFSUNLOCKCLSTATE();
5497 		NFSCL_DEBUG(4, "nfscl_dserr recall iomode=%d\n", iomode);
5498 	} else {
5499 		NFSUNLOCKCLSTATE();
5500 		free(recallp, M_NFSLAYRECALL);
5501 	}
5502 
5503 	/* And shut the TCP connection down. */
5504 	nfscl_cancelreqs(dsp);
5505 }
5506 
5507 /*
5508  * Cancel all RPCs for this "dsp" by closing the connection.
5509  * Also, mark the session as defunct.
5510  * If NFSCLDS_SAMECONN is set, the connection is shared with other DSs and
5511  * cannot be shut down.
5512  */
5513 void
nfscl_cancelreqs(struct nfsclds * dsp)5514 nfscl_cancelreqs(struct nfsclds *dsp)
5515 {
5516 	struct __rpc_client *cl;
5517 	static int non_event;
5518 
5519 	NFSLOCKDS(dsp);
5520 	if ((dsp->nfsclds_flags & (NFSCLDS_CLOSED | NFSCLDS_SAMECONN)) == 0 &&
5521 	    dsp->nfsclds_sockp != NULL &&
5522 	    dsp->nfsclds_sockp->nr_client != NULL) {
5523 		dsp->nfsclds_flags |= NFSCLDS_CLOSED;
5524 		cl = dsp->nfsclds_sockp->nr_client;
5525 		dsp->nfsclds_sess.nfsess_defunct = 1;
5526 		NFSUNLOCKDS(dsp);
5527 		CLNT_CLOSE(cl);
5528 		/*
5529 		 * This 1sec sleep is done to reduce the number of reconnect
5530 		 * attempts made on the DS while it has failed.
5531 		 */
5532 		tsleep(&non_event, PVFS, "ndscls", hz);
5533 		return;
5534 	}
5535 	NFSUNLOCKDS(dsp);
5536 }
5537 
5538 /*
5539  * Dereference a layout.
5540  */
5541 void
nfscl_rellayout(struct nfscllayout * lyp,int exclocked)5542 nfscl_rellayout(struct nfscllayout *lyp, int exclocked)
5543 {
5544 
5545 	NFSLOCKCLSTATE();
5546 	if (exclocked != 0)
5547 		nfsv4_unlock(&lyp->nfsly_lock, 0);
5548 	else
5549 		nfsv4_relref(&lyp->nfsly_lock);
5550 	NFSUNLOCKCLSTATE();
5551 }
5552 
5553 /*
5554  * Search for a devinfo by deviceid. If one is found, return it after
5555  * acquiring a reference count on it.
5556  */
5557 struct nfscldevinfo *
nfscl_getdevinfo(struct nfsclclient * clp,uint8_t * deviceid,struct nfscldevinfo * dip)5558 nfscl_getdevinfo(struct nfsclclient *clp, uint8_t *deviceid,
5559     struct nfscldevinfo *dip)
5560 {
5561 
5562 	NFSLOCKCLSTATE();
5563 	if (dip == NULL)
5564 		dip = nfscl_finddevinfo(clp, deviceid);
5565 	if (dip != NULL)
5566 		dip->nfsdi_refcnt++;
5567 	NFSUNLOCKCLSTATE();
5568 	return (dip);
5569 }
5570 
5571 /*
5572  * Dereference a devinfo structure.
5573  */
5574 static void
nfscl_reldevinfo_locked(struct nfscldevinfo * dip)5575 nfscl_reldevinfo_locked(struct nfscldevinfo *dip)
5576 {
5577 
5578 	dip->nfsdi_refcnt--;
5579 	if (dip->nfsdi_refcnt == 0)
5580 		wakeup(&dip->nfsdi_refcnt);
5581 }
5582 
5583 /*
5584  * Dereference a devinfo structure.
5585  */
5586 void
nfscl_reldevinfo(struct nfscldevinfo * dip)5587 nfscl_reldevinfo(struct nfscldevinfo *dip)
5588 {
5589 
5590 	NFSLOCKCLSTATE();
5591 	nfscl_reldevinfo_locked(dip);
5592 	NFSUNLOCKCLSTATE();
5593 }
5594 
5595 /*
5596  * Find a layout for this file handle. Return NULL upon failure.
5597  */
5598 static struct nfscllayout *
nfscl_findlayout(struct nfsclclient * clp,u_int8_t * fhp,int fhlen)5599 nfscl_findlayout(struct nfsclclient *clp, u_int8_t *fhp, int fhlen)
5600 {
5601 	struct nfscllayout *lyp;
5602 
5603 	LIST_FOREACH(lyp, NFSCLLAYOUTHASH(clp, fhp, fhlen), nfsly_hash)
5604 		if (lyp->nfsly_fhlen == fhlen &&
5605 		    !NFSBCMP(lyp->nfsly_fh, fhp, fhlen))
5606 			break;
5607 	return (lyp);
5608 }
5609 
5610 /*
5611  * Find a devinfo for this deviceid. Return NULL upon failure.
5612  */
5613 static struct nfscldevinfo *
nfscl_finddevinfo(struct nfsclclient * clp,uint8_t * deviceid)5614 nfscl_finddevinfo(struct nfsclclient *clp, uint8_t *deviceid)
5615 {
5616 	struct nfscldevinfo *dip;
5617 
5618 	LIST_FOREACH(dip, &clp->nfsc_devinfo, nfsdi_list)
5619 		if (NFSBCMP(dip->nfsdi_deviceid, deviceid, NFSX_V4DEVICEID)
5620 		    == 0)
5621 			break;
5622 	return (dip);
5623 }
5624 
5625 /*
5626  * Merge the new file layout list into the main one, maintaining it in
5627  * increasing offset order.
5628  */
5629 static void
nfscl_mergeflayouts(struct nfsclflayouthead * fhlp,struct nfsclflayouthead * newfhlp)5630 nfscl_mergeflayouts(struct nfsclflayouthead *fhlp,
5631     struct nfsclflayouthead *newfhlp)
5632 {
5633 	struct nfsclflayout *flp, *nflp, *prevflp, *tflp;
5634 
5635 	flp = LIST_FIRST(fhlp);
5636 	prevflp = NULL;
5637 	LIST_FOREACH_SAFE(nflp, newfhlp, nfsfl_list, tflp) {
5638 		while (flp != NULL && flp->nfsfl_off < nflp->nfsfl_off) {
5639 			prevflp = flp;
5640 			flp = LIST_NEXT(flp, nfsfl_list);
5641 		}
5642 		if (prevflp == NULL)
5643 			LIST_INSERT_HEAD(fhlp, nflp, nfsfl_list);
5644 		else
5645 			LIST_INSERT_AFTER(prevflp, nflp, nfsfl_list);
5646 		prevflp = nflp;
5647 	}
5648 }
5649 
5650 /*
5651  * Add this nfscldevinfo to the client, if it doesn't already exist.
5652  * This function consumes the structure pointed at by dip, if not NULL.
5653  */
5654 int
nfscl_adddevinfo(struct nfsmount * nmp,struct nfscldevinfo * dip,int ind,struct nfsclflayout * flp)5655 nfscl_adddevinfo(struct nfsmount *nmp, struct nfscldevinfo *dip, int ind,
5656     struct nfsclflayout *flp)
5657 {
5658 	struct nfsclclient *clp;
5659 	struct nfscldevinfo *tdip;
5660 	uint8_t *dev;
5661 
5662 	NFSLOCKCLSTATE();
5663 	clp = nmp->nm_clp;
5664 	if (clp == NULL) {
5665 		NFSUNLOCKCLSTATE();
5666 		if (dip != NULL)
5667 			free(dip, M_NFSDEVINFO);
5668 		return (ENODEV);
5669 	}
5670 	if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5671 		dev = flp->nfsfl_dev;
5672 	else
5673 		dev = flp->nfsfl_ffm[ind].dev;
5674 	tdip = nfscl_finddevinfo(clp, dev);
5675 	if (tdip != NULL) {
5676 		tdip->nfsdi_layoutrefs++;
5677 		if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5678 			flp->nfsfl_devp = tdip;
5679 		else
5680 			flp->nfsfl_ffm[ind].devp = tdip;
5681 		nfscl_reldevinfo_locked(tdip);
5682 		NFSUNLOCKCLSTATE();
5683 		if (dip != NULL)
5684 			free(dip, M_NFSDEVINFO);
5685 		return (0);
5686 	}
5687 	if (dip != NULL) {
5688 		LIST_INSERT_HEAD(&clp->nfsc_devinfo, dip, nfsdi_list);
5689 		dip->nfsdi_layoutrefs = 1;
5690 		if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5691 			flp->nfsfl_devp = dip;
5692 		else
5693 			flp->nfsfl_ffm[ind].devp = dip;
5694 	}
5695 	NFSUNLOCKCLSTATE();
5696 	if (dip == NULL)
5697 		return (ENODEV);
5698 	return (0);
5699 }
5700 
5701 /*
5702  * Free up a layout structure and associated file layout structure(s).
5703  */
5704 void
nfscl_freelayout(struct nfscllayout * layp)5705 nfscl_freelayout(struct nfscllayout *layp)
5706 {
5707 	struct nfsclflayout *flp, *nflp;
5708 	struct nfsclrecalllayout *rp, *nrp;
5709 
5710 	LIST_FOREACH_SAFE(flp, &layp->nfsly_flayread, nfsfl_list, nflp) {
5711 		LIST_REMOVE(flp, nfsfl_list);
5712 		nfscl_freeflayout(flp);
5713 	}
5714 	LIST_FOREACH_SAFE(flp, &layp->nfsly_flayrw, nfsfl_list, nflp) {
5715 		LIST_REMOVE(flp, nfsfl_list);
5716 		nfscl_freeflayout(flp);
5717 	}
5718 	LIST_FOREACH_SAFE(rp, &layp->nfsly_recall, nfsrecly_list, nrp) {
5719 		LIST_REMOVE(rp, nfsrecly_list);
5720 		free(rp, M_NFSLAYRECALL);
5721 	}
5722 	layp->nfsly_clp->nfsc_layoutcnt--;
5723 	nfsstatsv1.cllayouts--;
5724 	free(layp, M_NFSLAYOUT);
5725 }
5726 
5727 /*
5728  * Free up a file layout structure.
5729  */
5730 void
nfscl_freeflayout(struct nfsclflayout * flp)5731 nfscl_freeflayout(struct nfsclflayout *flp)
5732 {
5733 	int i, j;
5734 
5735 	if ((flp->nfsfl_flags & NFSFL_FILE) != 0) {
5736 		for (i = 0; i < flp->nfsfl_fhcnt; i++)
5737 			free(flp->nfsfl_fh[i], M_NFSFH);
5738 		if (flp->nfsfl_devp != NULL)
5739 			flp->nfsfl_devp->nfsdi_layoutrefs--;
5740 	}
5741 	if ((flp->nfsfl_flags & NFSFL_FLEXFILE) != 0)
5742 		for (i = 0; i < flp->nfsfl_mirrorcnt; i++) {
5743 			for (j = 0; j < flp->nfsfl_ffm[i].fhcnt; j++)
5744 				free(flp->nfsfl_ffm[i].fh[j], M_NFSFH);
5745 			if (flp->nfsfl_ffm[i].devp != NULL)
5746 				flp->nfsfl_ffm[i].devp->nfsdi_layoutrefs--;
5747 		}
5748 	free(flp, M_NFSFLAYOUT);
5749 }
5750 
5751 /*
5752  * Free up a file layout devinfo structure.
5753  */
5754 void
nfscl_freedevinfo(struct nfscldevinfo * dip)5755 nfscl_freedevinfo(struct nfscldevinfo *dip)
5756 {
5757 
5758 	free(dip, M_NFSDEVINFO);
5759 }
5760 
5761 /*
5762  * Mark any layouts that match as recalled.
5763  */
5764 static int
nfscl_layoutrecall(int recalltype,struct nfscllayout * lyp,uint32_t iomode,uint64_t off,uint64_t len,uint32_t stateseqid,uint32_t stat,uint32_t op,char * devid,struct nfsclrecalllayout * recallp)5765 nfscl_layoutrecall(int recalltype, struct nfscllayout *lyp, uint32_t iomode,
5766     uint64_t off, uint64_t len, uint32_t stateseqid, uint32_t stat, uint32_t op,
5767     char *devid, struct nfsclrecalllayout *recallp)
5768 {
5769 	struct nfsclrecalllayout *rp, *orp;
5770 
5771 	recallp->nfsrecly_recalltype = recalltype;
5772 	recallp->nfsrecly_iomode = iomode;
5773 	recallp->nfsrecly_stateseqid = stateseqid;
5774 	recallp->nfsrecly_off = off;
5775 	recallp->nfsrecly_len = len;
5776 	recallp->nfsrecly_stat = stat;
5777 	recallp->nfsrecly_op = op;
5778 	if (devid != NULL)
5779 		NFSBCOPY(devid, recallp->nfsrecly_devid, NFSX_V4DEVICEID);
5780 	/*
5781 	 * Order the list as file returns first, followed by fsid and any
5782 	 * returns, both in increasing stateseqid order.
5783 	 * Note that the seqids wrap around, so 1 is after 0xffffffff.
5784 	 * (I'm not sure this is correct because I find RFC5661 confusing
5785 	 *  on this, but hopefully it will work ok.)
5786 	 */
5787 	orp = NULL;
5788 	LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) {
5789 		orp = rp;
5790 		if ((recalltype == NFSLAYOUTRETURN_FILE &&
5791 		     (rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE ||
5792 		      nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) ||
5793 		    (recalltype != NFSLAYOUTRETURN_FILE &&
5794 		     rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE &&
5795 		     nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) {
5796 			LIST_INSERT_BEFORE(rp, recallp, nfsrecly_list);
5797 			break;
5798 		}
5799 
5800 		/*
5801 		 * Put any error return on all the file returns that will
5802 		 * preceed this one.
5803 		 */
5804 		if (rp->nfsrecly_recalltype == NFSLAYOUTRETURN_FILE &&
5805 		   stat != 0 && rp->nfsrecly_stat == 0) {
5806 			rp->nfsrecly_stat = stat;
5807 			rp->nfsrecly_op = op;
5808 			if (devid != NULL)
5809 				NFSBCOPY(devid, rp->nfsrecly_devid,
5810 				    NFSX_V4DEVICEID);
5811 		}
5812 	}
5813 	if (rp == NULL) {
5814 		if (orp == NULL)
5815 			LIST_INSERT_HEAD(&lyp->nfsly_recall, recallp,
5816 			    nfsrecly_list);
5817 		else
5818 			LIST_INSERT_AFTER(orp, recallp, nfsrecly_list);
5819 	}
5820 	lyp->nfsly_flags |= NFSLY_RECALL;
5821 	wakeup(lyp->nfsly_clp);
5822 	return (0);
5823 }
5824 
5825 /*
5826  * Compare the two seqids for ordering. The trick is that the seqids can
5827  * wrap around from 0xffffffff->0, so check for the cases where one
5828  * has wrapped around.
5829  * Return 1 if seqid1 comes before seqid2, 0 otherwise.
5830  */
5831 static int
nfscl_seq(uint32_t seqid1,uint32_t seqid2)5832 nfscl_seq(uint32_t seqid1, uint32_t seqid2)
5833 {
5834 
5835 	if (seqid2 > seqid1 && (seqid2 - seqid1) >= 0x7fffffff)
5836 		/* seqid2 has wrapped around. */
5837 		return (0);
5838 	if (seqid1 > seqid2 && (seqid1 - seqid2) >= 0x7fffffff)
5839 		/* seqid1 has wrapped around. */
5840 		return (1);
5841 	if (seqid1 <= seqid2)
5842 		return (1);
5843 	return (0);
5844 }
5845 
5846 /*
5847  * Do a layout return for each of the recalls.
5848  */
5849 static void
nfscl_layoutreturn(struct nfsmount * nmp,struct nfscllayout * lyp,struct ucred * cred,NFSPROC_T * p)5850 nfscl_layoutreturn(struct nfsmount *nmp, struct nfscllayout *lyp,
5851     struct ucred *cred, NFSPROC_T *p)
5852 {
5853 	struct nfsclrecalllayout *rp;
5854 	nfsv4stateid_t stateid;
5855 	int layouttype;
5856 
5857 	NFSBCOPY(lyp->nfsly_stateid.other, stateid.other, NFSX_STATEIDOTHER);
5858 	stateid.seqid = lyp->nfsly_stateid.seqid;
5859 	if ((lyp->nfsly_flags & NFSLY_FILES) != 0)
5860 		layouttype = NFSLAYOUT_NFSV4_1_FILES;
5861 	else
5862 		layouttype = NFSLAYOUT_FLEXFILE;
5863 	LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) {
5864 		(void)nfsrpc_layoutreturn(nmp, lyp->nfsly_fh,
5865 		    lyp->nfsly_fhlen, 0, layouttype,
5866 		    rp->nfsrecly_iomode, rp->nfsrecly_recalltype,
5867 		    rp->nfsrecly_off, rp->nfsrecly_len,
5868 		    &stateid, cred, p, rp->nfsrecly_stat, rp->nfsrecly_op,
5869 		    rp->nfsrecly_devid);
5870 	}
5871 }
5872 
5873 /*
5874  * Do the layout commit for a file layout.
5875  */
5876 static void
nfscl_dolayoutcommit(struct nfsmount * nmp,struct nfscllayout * lyp,struct ucred * cred,NFSPROC_T * p)5877 nfscl_dolayoutcommit(struct nfsmount *nmp, struct nfscllayout *lyp,
5878     struct ucred *cred, NFSPROC_T *p)
5879 {
5880 	struct nfsclflayout *flp;
5881 	uint64_t len;
5882 	int error, layouttype;
5883 
5884 	if ((lyp->nfsly_flags & NFSLY_FILES) != 0)
5885 		layouttype = NFSLAYOUT_NFSV4_1_FILES;
5886 	else
5887 		layouttype = NFSLAYOUT_FLEXFILE;
5888 	LIST_FOREACH(flp, &lyp->nfsly_flayrw, nfsfl_list) {
5889 		if (layouttype == NFSLAYOUT_FLEXFILE &&
5890 		    (flp->nfsfl_fflags & NFSFLEXFLAG_NO_LAYOUTCOMMIT) != 0) {
5891 			NFSCL_DEBUG(4, "Flex file: no layoutcommit\n");
5892 			/* If not supported, don't bother doing it. */
5893 			NFSLOCKMNT(nmp);
5894 			nmp->nm_state |= NFSSTA_NOLAYOUTCOMMIT;
5895 			NFSUNLOCKMNT(nmp);
5896 			break;
5897 		} else if (flp->nfsfl_off <= lyp->nfsly_lastbyte) {
5898 			len = flp->nfsfl_end - flp->nfsfl_off;
5899 			error = nfsrpc_layoutcommit(nmp, lyp->nfsly_fh,
5900 			    lyp->nfsly_fhlen, 0, flp->nfsfl_off, len,
5901 			    lyp->nfsly_lastbyte, &lyp->nfsly_stateid,
5902 			    layouttype, cred, p);
5903 			NFSCL_DEBUG(4, "layoutcommit err=%d\n", error);
5904 			if (error == NFSERR_NOTSUPP) {
5905 				/* If not supported, don't bother doing it. */
5906 				NFSLOCKMNT(nmp);
5907 				nmp->nm_state |= NFSSTA_NOLAYOUTCOMMIT;
5908 				NFSUNLOCKMNT(nmp);
5909 				break;
5910 			}
5911 		}
5912 	}
5913 }
5914 
5915 /*
5916  * Commit all layouts for a file (vnode).
5917  */
5918 int
nfscl_layoutcommit(vnode_t vp,NFSPROC_T * p)5919 nfscl_layoutcommit(vnode_t vp, NFSPROC_T *p)
5920 {
5921 	struct nfsclclient *clp;
5922 	struct nfscllayout *lyp;
5923 	struct nfsnode *np = VTONFS(vp);
5924 	mount_t mp;
5925 	struct nfsmount *nmp;
5926 
5927 	mp = vp->v_mount;
5928 	nmp = VFSTONFS(mp);
5929 	if (NFSHASNOLAYOUTCOMMIT(nmp))
5930 		return (0);
5931 	NFSLOCKCLSTATE();
5932 	clp = nmp->nm_clp;
5933 	if (clp == NULL) {
5934 		NFSUNLOCKCLSTATE();
5935 		return (EPERM);
5936 	}
5937 	lyp = nfscl_findlayout(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
5938 	if (lyp == NULL) {
5939 		NFSUNLOCKCLSTATE();
5940 		return (EPERM);
5941 	}
5942 	nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
5943 	if (NFSCL_FORCEDISM(mp)) {
5944 		NFSUNLOCKCLSTATE();
5945 		return (EPERM);
5946 	}
5947 tryagain:
5948 	if ((lyp->nfsly_flags & NFSLY_WRITTEN) != 0) {
5949 		lyp->nfsly_flags &= ~NFSLY_WRITTEN;
5950 		NFSUNLOCKCLSTATE();
5951 		NFSCL_DEBUG(4, "do layoutcommit2\n");
5952 		nfscl_dolayoutcommit(clp->nfsc_nmp, lyp, NFSPROCCRED(p), p);
5953 		NFSLOCKCLSTATE();
5954 		goto tryagain;
5955 	}
5956 	nfsv4_relref(&lyp->nfsly_lock);
5957 	NFSUNLOCKCLSTATE();
5958 	return (0);
5959 }
5960