xref: /netbsd-src/sys/fs/nfs/nlm/nlm_advlock.c (revision 413d532bcc3f62d122e56d92e13ac64825a40baf)
1 /*	$NetBSD: nlm_advlock.c,v 1.1.1.1 2013/09/30 07:19:43 dholland Exp $	*/
2 /*-
3  * Copyright (c) 2008 Isilon Inc http://www.isilon.com/
4  * Authors: Doug Rabson <dfr@rabson.org>
5  * Developed with Red Inc: Alfred Perlstein <alfred@freebsd.org>
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 #include <sys/cdefs.h>
30 /* __FBSDID("FreeBSD: head/sys/nlm/nlm_advlock.c 239328 2012-08-16 13:01:56Z kib "); */
31 __RCSID("$NetBSD: nlm_advlock.c,v 1.1.1.1 2013/09/30 07:19:43 dholland Exp $");
32 
33 #include <sys/param.h>
34 #include <sys/fcntl.h>
35 #include <sys/jail.h>
36 #include <sys/kernel.h>
37 #include <sys/limits.h>
38 #include <sys/lock.h>
39 #include <sys/lockf.h>
40 #include <sys/malloc.h>
41 #include <sys/mbuf.h>
42 #include <sys/mount.h>
43 #include <sys/mutex.h>
44 #include <sys/proc.h>
45 #include <sys/socket.h>
46 #include <sys/syslog.h>
47 #include <sys/systm.h>
48 #include <sys/unistd.h>
49 #include <sys/vnode.h>
50 
51 #include <nfs/nfsproto.h>
52 #include <nfsclient/nfs.h>
53 #include <nfsclient/nfsmount.h>
54 
55 #include <nlm/nlm_prot.h>
56 #include <nlm/nlm.h>
57 
58 /*
59  * We need to keep track of the svid values used for F_FLOCK locks.
60  */
61 struct nlm_file_svid {
62 	int		ns_refs;	/* thread count + 1 if active */
63 	int		ns_svid;	/* on-the-wire SVID for this file */
64 	struct ucred	*ns_ucred;	/* creds to use for lock recovery */
65 	void		*ns_id;		/* local struct file pointer */
66 	bool_t		ns_active;	/* TRUE if we own a lock */
67 	LIST_ENTRY(nlm_file_svid) ns_link;
68 };
69 LIST_HEAD(nlm_file_svid_list, nlm_file_svid);
70 
71 #define NLM_SVID_HASH_SIZE	256
72 struct nlm_file_svid_list nlm_file_svids[NLM_SVID_HASH_SIZE];
73 
74 struct mtx nlm_svid_lock;
75 static struct unrhdr *nlm_svid_allocator;
76 static volatile u_int nlm_xid = 1;
77 
78 static int nlm_setlock(struct nlm_host *host, struct rpc_callextra *ext,
79     rpcvers_t vers, struct timeval *timo, int retries,
80     struct vnode *vp, int op, struct flock *fl, int flags,
81     int svid, size_t fhlen, void *fh, off_t size, bool_t reclaim);
82 static int nlm_clearlock(struct nlm_host *host,  struct rpc_callextra *ext,
83     rpcvers_t vers, struct timeval *timo, int retries,
84     struct vnode *vp, int op, struct flock *fl, int flags,
85     int svid, size_t fhlen, void *fh, off_t size);
86 static int nlm_getlock(struct nlm_host *host, struct rpc_callextra *ext,
87     rpcvers_t vers, struct timeval *timo, int retries,
88     struct vnode *vp, int op, struct flock *fl, int flags,
89     int svid, size_t fhlen, void *fh, off_t size);
90 static int nlm_map_status(nlm4_stats stat);
91 static struct nlm_file_svid *nlm_find_svid(void *id);
92 static void nlm_free_svid(struct nlm_file_svid *nf);
93 static int nlm_init_lock(struct flock *fl, int flags, int svid,
94     rpcvers_t vers, size_t fhlen, void *fh, off_t size,
95     struct nlm4_lock *lock, char oh_space[32]);
96 
97 static void
98 nlm_client_init(void *dummy)
99 {
100 	int i;
101 
102 	mtx_init(&nlm_svid_lock, "NLM svid lock", NULL, MTX_DEF);
103 	/* pid_max cannot be greater than PID_MAX */
104 	nlm_svid_allocator = new_unrhdr(PID_MAX + 2, INT_MAX, &nlm_svid_lock);
105 	for (i = 0; i < NLM_SVID_HASH_SIZE; i++)
106 		LIST_INIT(&nlm_file_svids[i]);
107 }
108 SYSINIT(nlm_client_init, SI_SUB_LOCK, SI_ORDER_FIRST, nlm_client_init, NULL);
109 
110 static int
111 nlm_msg(struct thread *td, const char *server, const char *msg, int error)
112 {
113 	struct proc *p;
114 
115 	p = td ? td->td_proc : NULL;
116 	if (error) {
117 		tprintf(p, LOG_INFO, "nfs server %s: %s, error %d\n", server,
118 		    msg, error);
119 	} else {
120 		tprintf(p, LOG_INFO, "nfs server %s: %s\n", server, msg);
121 	}
122 	return (0);
123 }
124 
125 struct nlm_feedback_arg {
126 	bool_t	nf_printed;
127 	struct nfsmount *nf_nmp;
128 };
129 
130 static void
131 nlm_down(struct nlm_feedback_arg *nf, struct thread *td,
132     const char *msg, int error)
133 {
134 	struct nfsmount *nmp = nf->nf_nmp;
135 
136 	if (nmp == NULL)
137 		return;
138 	mtx_lock(&nmp->nm_mtx);
139 	if (!(nmp->nm_state & NFSSTA_LOCKTIMEO)) {
140 		nmp->nm_state |= NFSSTA_LOCKTIMEO;
141 		mtx_unlock(&nmp->nm_mtx);
142 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
143 		    VQ_NOTRESPLOCK, 0);
144 	} else {
145 		mtx_unlock(&nmp->nm_mtx);
146 	}
147 
148 	nf->nf_printed = TRUE;
149 	nlm_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error);
150 }
151 
152 static void
153 nlm_up(struct nlm_feedback_arg *nf, struct thread *td,
154     const char *msg)
155 {
156 	struct nfsmount *nmp = nf->nf_nmp;
157 
158 	if (!nf->nf_printed)
159 		return;
160 
161 	nlm_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0);
162 
163 	mtx_lock(&nmp->nm_mtx);
164 	if (nmp->nm_state & NFSSTA_LOCKTIMEO) {
165 		nmp->nm_state &= ~NFSSTA_LOCKTIMEO;
166 		mtx_unlock(&nmp->nm_mtx);
167 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
168 		    VQ_NOTRESPLOCK, 1);
169 	} else {
170 		mtx_unlock(&nmp->nm_mtx);
171 	}
172 }
173 
174 static void
175 nlm_feedback(int type, int proc, void *arg)
176 {
177 	struct thread *td = curthread;
178 	struct nlm_feedback_arg *nf = (struct nlm_feedback_arg *) arg;
179 
180 	switch (type) {
181 	case FEEDBACK_REXMIT2:
182 	case FEEDBACK_RECONNECT:
183 		nlm_down(nf, td, "lockd not responding", 0);
184 		break;
185 
186 	case FEEDBACK_OK:
187 		nlm_up(nf, td, "lockd is alive again");
188 		break;
189 	}
190 }
191 
192 /*
193  * nlm_advlock --
194  *      NFS advisory byte-level locks.
195  */
196 static int
197 nlm_advlock_internal(struct vnode *vp, void *id, int op, struct flock *fl,
198     int flags, bool_t reclaim, bool_t unlock_vp)
199 {
200 	struct thread *td = curthread;
201 	struct nfsmount *nmp;
202 	off_t size;
203 	size_t fhlen;
204 	union nfsfh fh;
205 	struct sockaddr *sa;
206 	struct sockaddr_storage ss;
207 	char servername[MNAMELEN];
208 	struct timeval timo;
209 	int retries;
210 	rpcvers_t vers;
211 	struct nlm_host *host;
212 	struct rpc_callextra ext;
213 	struct nlm_feedback_arg nf;
214 	AUTH *auth;
215 	struct ucred *cred;
216 	struct nlm_file_svid *ns;
217 	int svid;
218 	int error;
219 	int is_v3;
220 
221 	ASSERT_VOP_LOCKED(vp, "nlm_advlock_1");
222 
223 	nmp = VFSTONFS(vp->v_mount);
224 	/*
225 	 * Push any pending writes to the server and flush our cache
226 	 * so that if we are contending with another machine for a
227 	 * file, we get whatever they wrote and vice-versa.
228 	 */
229 	if (op == F_SETLK || op == F_UNLCK)
230 		nmp->nm_vinvalbuf(vp, V_SAVE, td, 1);
231 
232 	strcpy(servername, nmp->nm_hostname);
233 	nmp->nm_getinfo(vp, fh.fh_bytes, &fhlen, &ss, &is_v3, &size, &timo);
234 	sa = (struct sockaddr *) &ss;
235 	if (is_v3 != 0)
236 		vers = NLM_VERS4;
237 	else
238 		vers = NLM_VERS;
239 
240 	if (nmp->nm_flag & NFSMNT_SOFT)
241 		retries = nmp->nm_retry;
242 	else
243 		retries = INT_MAX;
244 
245 	if (unlock_vp)
246 		VOP_UNLOCK(vp, 0);
247 
248 	/*
249 	 * We need to switch to mount-point creds so that we can send
250 	 * packets from a privileged port.
251 	 */
252 	cred = td->td_ucred;
253 	td->td_ucred = vp->v_mount->mnt_cred;
254 
255 	host = nlm_find_host_by_name(servername, sa, vers);
256 	auth = authunix_create(cred);
257 	memset(&ext, 0, sizeof(ext));
258 
259 	nf.nf_printed = FALSE;
260 	nf.nf_nmp = nmp;
261 	ext.rc_auth = auth;
262 
263 	ext.rc_feedback = nlm_feedback;
264 	ext.rc_feedback_arg = &nf;
265 	ext.rc_timers = NULL;
266 
267 	ns = NULL;
268 	if (flags & F_FLOCK) {
269 		ns = nlm_find_svid(id);
270 		KASSERT(fl->l_start == 0 && fl->l_len == 0,
271 		    ("F_FLOCK lock requests must be whole-file locks"));
272 		if (!ns->ns_ucred) {
273 			/*
274 			 * Remember the creds used for locking in case
275 			 * we need to recover the lock later.
276 			 */
277 			ns->ns_ucred = crdup(cred);
278 		}
279 		svid = ns->ns_svid;
280 	} else if (flags & F_REMOTE) {
281 		/*
282 		 * If we are recovering after a server restart or
283 		 * trashing locks on a force unmount, use the same
284 		 * svid as last time.
285 		 */
286 		svid = fl->l_pid;
287 	} else {
288 		svid = ((struct proc *) id)->p_pid;
289 	}
290 
291 	switch(op) {
292 	case F_SETLK:
293 		if ((flags & (F_FLOCK|F_WAIT)) == (F_FLOCK|F_WAIT)
294 		    && fl->l_type == F_WRLCK) {
295 			/*
296 			 * The semantics for flock(2) require that any
297 			 * shared lock on the file must be released
298 			 * before an exclusive lock is granted. The
299 			 * local locking code interprets this by
300 			 * unlocking the file before sleeping on a
301 			 * blocked exclusive lock request. We
302 			 * approximate this by first attempting
303 			 * non-blocking and if that fails, we unlock
304 			 * the file and block.
305 			 */
306 			error = nlm_setlock(host, &ext, vers, &timo, retries,
307 			    vp, F_SETLK, fl, flags & ~F_WAIT,
308 			    svid, fhlen, &fh.fh_bytes, size, reclaim);
309 			if (error == EAGAIN) {
310 				fl->l_type = F_UNLCK;
311 				error = nlm_clearlock(host, &ext, vers, &timo,
312 				    retries, vp, F_UNLCK, fl, flags,
313 				    svid, fhlen, &fh.fh_bytes, size);
314 				fl->l_type = F_WRLCK;
315 				if (!error) {
316 					mtx_lock(&nlm_svid_lock);
317 					if (ns->ns_active) {
318 						ns->ns_refs--;
319 						ns->ns_active = FALSE;
320 					}
321 					mtx_unlock(&nlm_svid_lock);
322 					flags |= F_WAIT;
323 					error = nlm_setlock(host, &ext, vers,
324 					    &timo, retries, vp, F_SETLK, fl,
325 					    flags, svid, fhlen, &fh.fh_bytes,
326 					    size, reclaim);
327 				}
328 			}
329 		} else {
330 			error = nlm_setlock(host, &ext, vers, &timo, retries,
331 			    vp, op, fl, flags, svid, fhlen, &fh.fh_bytes,
332 			    size, reclaim);
333 		}
334 		if (!error && ns) {
335 			mtx_lock(&nlm_svid_lock);
336 			if (!ns->ns_active) {
337 				/*
338 				 * Add one to the reference count to
339 				 * hold onto the SVID for the lifetime
340 				 * of the lock. Note that since
341 				 * F_FLOCK only supports whole-file
342 				 * locks, there can only be one active
343 				 * lock for this SVID.
344 				 */
345 				ns->ns_refs++;
346 				ns->ns_active = TRUE;
347 			}
348 			mtx_unlock(&nlm_svid_lock);
349 		}
350 		break;
351 
352 	case F_UNLCK:
353 		error = nlm_clearlock(host, &ext, vers, &timo, retries,
354 		    vp, op, fl, flags, svid, fhlen, &fh.fh_bytes, size);
355 		if (!error && ns) {
356 			mtx_lock(&nlm_svid_lock);
357 			if (ns->ns_active) {
358 				ns->ns_refs--;
359 				ns->ns_active = FALSE;
360 			}
361 			mtx_unlock(&nlm_svid_lock);
362 		}
363 		break;
364 
365 	case F_GETLK:
366 		error = nlm_getlock(host, &ext, vers, &timo, retries,
367 		    vp, op, fl, flags, svid, fhlen, &fh.fh_bytes, size);
368 		break;
369 
370 	default:
371 		error = EINVAL;
372 		break;
373 	}
374 
375 	if (ns)
376 		nlm_free_svid(ns);
377 
378 	td->td_ucred = cred;
379 	AUTH_DESTROY(auth);
380 
381 	nlm_host_release(host);
382 
383 	return (error);
384 }
385 
386 int
387 nlm_advlock(struct vop_advlock_args *ap)
388 {
389 
390 	return (nlm_advlock_internal(ap->a_vp, ap->a_id, ap->a_op, ap->a_fl,
391 		ap->a_flags, FALSE, TRUE));
392 }
393 
394 /*
395  * Set the creds of td to the creds of the given lock's owner. The new
396  * creds reference count will be incremented via crhold. The caller is
397  * responsible for calling crfree and restoring td's original creds.
398  */
399 static void
400 nlm_set_creds_for_lock(struct thread *td, struct flock *fl)
401 {
402 	int i;
403 	struct nlm_file_svid *ns;
404 	struct proc *p;
405 	struct ucred *cred;
406 
407 	cred = NULL;
408 	if (fl->l_pid > PID_MAX) {
409 		/*
410 		 * If this was originally a F_FLOCK-style lock, we
411 		 * recorded the creds used when it was originally
412 		 * locked in the nlm_file_svid structure.
413 		 */
414 		mtx_lock(&nlm_svid_lock);
415 		for (i = 0; i < NLM_SVID_HASH_SIZE; i++) {
416 			for (ns = LIST_FIRST(&nlm_file_svids[i]); ns;
417 			     ns = LIST_NEXT(ns, ns_link)) {
418 				if (ns->ns_svid == fl->l_pid) {
419 					cred = crhold(ns->ns_ucred);
420 					break;
421 				}
422 			}
423 		}
424 		mtx_unlock(&nlm_svid_lock);
425 	} else {
426 		/*
427 		 * This lock is owned by a process. Get a reference to
428 		 * the process creds.
429 		 */
430 		p = pfind(fl->l_pid);
431 		if (p) {
432 			cred = crhold(p->p_ucred);
433 			PROC_UNLOCK(p);
434 		}
435 	}
436 
437 	/*
438 	 * If we can't find a cred, fall back on the recovery
439 	 * thread's cred.
440 	 */
441 	if (!cred) {
442 		cred = crhold(td->td_ucred);
443 	}
444 
445 	td->td_ucred = cred;
446 }
447 
448 static int
449 nlm_reclaim_free_lock(struct vnode *vp, struct flock *fl, void *arg)
450 {
451 	struct flock newfl;
452 	struct thread *td = curthread;
453 	struct ucred *oldcred;
454 	int error;
455 
456 	newfl = *fl;
457 	newfl.l_type = F_UNLCK;
458 
459 	oldcred = td->td_ucred;
460 	nlm_set_creds_for_lock(td, &newfl);
461 
462 	error = nlm_advlock_internal(vp, NULL, F_UNLCK, &newfl, F_REMOTE,
463 	    FALSE, FALSE);
464 
465 	crfree(td->td_ucred);
466 	td->td_ucred = oldcred;
467 
468 	return (error);
469 }
470 
471 int
472 nlm_reclaim(struct vop_reclaim_args *ap)
473 {
474 
475 	nlm_cancel_wait(ap->a_vp);
476 	lf_iteratelocks_vnode(ap->a_vp, nlm_reclaim_free_lock, NULL);
477 	return (0);
478 }
479 
480 struct nlm_recovery_context {
481 	struct nlm_host	*nr_host;	/* host we are recovering */
482 	int		nr_state;	/* remote NSM state for recovery */
483 };
484 
485 static int
486 nlm_client_recover_lock(struct vnode *vp, struct flock *fl, void *arg)
487 {
488 	struct nlm_recovery_context *nr = (struct nlm_recovery_context *) arg;
489 	struct thread *td = curthread;
490 	struct ucred *oldcred;
491 	int state, error;
492 
493 	/*
494 	 * If the remote NSM state changes during recovery, the host
495 	 * must have rebooted a second time. In that case, we must
496 	 * restart the recovery.
497 	 */
498 	state = nlm_host_get_state(nr->nr_host);
499 	if (nr->nr_state != state)
500 		return (ERESTART);
501 
502 	error = vn_lock(vp, LK_SHARED);
503 	if (error)
504 		return (error);
505 
506 	oldcred = td->td_ucred;
507 	nlm_set_creds_for_lock(td, fl);
508 
509 	error = nlm_advlock_internal(vp, NULL, F_SETLK, fl, F_REMOTE,
510 	    TRUE, TRUE);
511 
512 	crfree(td->td_ucred);
513 	td->td_ucred = oldcred;
514 
515 	return (error);
516 }
517 
518 void
519 nlm_client_recovery(struct nlm_host *host)
520 {
521 	struct nlm_recovery_context nr;
522 	int sysid, error;
523 
524 	sysid = NLM_SYSID_CLIENT | nlm_host_get_sysid(host);
525 	do {
526 		nr.nr_host = host;
527 		nr.nr_state = nlm_host_get_state(host);
528 		error = lf_iteratelocks_sysid(sysid,
529 		    nlm_client_recover_lock, &nr);
530 	} while (error == ERESTART);
531 }
532 
533 static void
534 nlm_convert_to_nlm_lock(struct nlm_lock *dst, struct nlm4_lock *src)
535 {
536 
537 	dst->caller_name = src->caller_name;
538 	dst->fh = src->fh;
539 	dst->oh = src->oh;
540 	dst->svid = src->svid;
541 	dst->l_offset = src->l_offset;
542 	dst->l_len = src->l_len;
543 }
544 
545 static void
546 nlm_convert_to_nlm4_holder(struct nlm4_holder *dst, struct nlm_holder *src)
547 {
548 
549 	dst->exclusive = src->exclusive;
550 	dst->svid = src->svid;
551 	dst->oh = src->oh;
552 	dst->l_offset = src->l_offset;
553 	dst->l_len = src->l_len;
554 }
555 
556 static void
557 nlm_convert_to_nlm4_res(struct nlm4_res *dst, struct nlm_res *src)
558 {
559 	dst->cookie = src->cookie;
560 	dst->stat.stat = (enum nlm4_stats) src->stat.stat;
561 }
562 
563 static enum clnt_stat
564 nlm_test_rpc(rpcvers_t vers, nlm4_testargs *args, nlm4_testres *res, CLIENT *client,
565     struct rpc_callextra *ext, struct timeval timo)
566 {
567 	if (vers == NLM_VERS4) {
568 		return nlm4_test_4(args, res, client, ext, timo);
569 	} else {
570 		nlm_testargs args1;
571 		nlm_testres res1;
572 		enum clnt_stat stat;
573 
574 		args1.cookie = args->cookie;
575 		args1.exclusive = args->exclusive;
576 		nlm_convert_to_nlm_lock(&args1.alock, &args->alock);
577 		memset(&res1, 0, sizeof(res1));
578 
579 		stat = nlm_test_1(&args1, &res1, client, ext, timo);
580 
581 		if (stat == RPC_SUCCESS) {
582 			res->cookie = res1.cookie;
583 			res->stat.stat = (enum nlm4_stats) res1.stat.stat;
584 			if (res1.stat.stat == nlm_denied)
585 				nlm_convert_to_nlm4_holder(
586 					&res->stat.nlm4_testrply_u.holder,
587 					&res1.stat.nlm_testrply_u.holder);
588 		}
589 
590 		return (stat);
591 	}
592 }
593 
594 static enum clnt_stat
595 nlm_lock_rpc(rpcvers_t vers, nlm4_lockargs *args, nlm4_res *res, CLIENT *client,
596     struct rpc_callextra *ext, struct timeval timo)
597 {
598 	if (vers == NLM_VERS4) {
599 		return nlm4_lock_4(args, res, client, ext, timo);
600 	} else {
601 		nlm_lockargs args1;
602 		nlm_res res1;
603 		enum clnt_stat stat;
604 
605 		args1.cookie = args->cookie;
606 		args1.block = args->block;
607 		args1.exclusive = args->exclusive;
608 		nlm_convert_to_nlm_lock(&args1.alock, &args->alock);
609 		args1.reclaim = args->reclaim;
610 		args1.state = args->state;
611 		memset(&res1, 0, sizeof(res1));
612 
613 		stat = nlm_lock_1(&args1, &res1, client, ext, timo);
614 
615 		if (stat == RPC_SUCCESS) {
616 			nlm_convert_to_nlm4_res(res, &res1);
617 		}
618 
619 		return (stat);
620 	}
621 }
622 
623 static enum clnt_stat
624 nlm_cancel_rpc(rpcvers_t vers, nlm4_cancargs *args, nlm4_res *res, CLIENT *client,
625     struct rpc_callextra *ext, struct timeval timo)
626 {
627 	if (vers == NLM_VERS4) {
628 		return nlm4_cancel_4(args, res, client, ext, timo);
629 	} else {
630 		nlm_cancargs args1;
631 		nlm_res res1;
632 		enum clnt_stat stat;
633 
634 		args1.cookie = args->cookie;
635 		args1.block = args->block;
636 		args1.exclusive = args->exclusive;
637 		nlm_convert_to_nlm_lock(&args1.alock, &args->alock);
638 		memset(&res1, 0, sizeof(res1));
639 
640 		stat = nlm_cancel_1(&args1, &res1, client, ext, timo);
641 
642 		if (stat == RPC_SUCCESS) {
643 			nlm_convert_to_nlm4_res(res, &res1);
644 		}
645 
646 		return (stat);
647 	}
648 }
649 
650 static enum clnt_stat
651 nlm_unlock_rpc(rpcvers_t vers, nlm4_unlockargs *args, nlm4_res *res, CLIENT *client,
652     struct rpc_callextra *ext, struct timeval timo)
653 {
654 	if (vers == NLM_VERS4) {
655 		return nlm4_unlock_4(args, res, client, ext, timo);
656 	} else {
657 		nlm_unlockargs args1;
658 		nlm_res res1;
659 		enum clnt_stat stat;
660 
661 		args1.cookie = args->cookie;
662 		nlm_convert_to_nlm_lock(&args1.alock, &args->alock);
663 		memset(&res1, 0, sizeof(res1));
664 
665 		stat = nlm_unlock_1(&args1, &res1, client, ext, timo);
666 
667 		if (stat == RPC_SUCCESS) {
668 			nlm_convert_to_nlm4_res(res, &res1);
669 		}
670 
671 		return (stat);
672 	}
673 }
674 
675 /*
676  * Called after a lock request (set or clear) succeeded. We record the
677  * details in the local lock manager. Note that since the remote
678  * server has granted the lock, we can be sure that it doesn't
679  * conflict with any other locks we have in the local lock manager.
680  *
681  * Since it is possible that host may also make NLM client requests to
682  * our NLM server, we use a different sysid value to record our own
683  * client locks.
684  *
685  * Note that since it is possible for us to receive replies from the
686  * server in a different order than the locks were granted (e.g. if
687  * many local threads are contending for the same lock), we must use a
688  * blocking operation when registering with the local lock manager.
689  * We expect that any actual wait will be rare and short hence we
690  * ignore signals for this.
691  */
692 static void
693 nlm_record_lock(struct vnode *vp, int op, struct flock *fl,
694     int svid, int sysid, off_t size)
695 {
696 	struct vop_advlockasync_args a;
697 	struct flock newfl;
698 	int error;
699 
700 	a.a_vp = vp;
701 	a.a_id = NULL;
702 	a.a_op = op;
703 	a.a_fl = &newfl;
704 	a.a_flags = F_REMOTE|F_WAIT|F_NOINTR;
705 	a.a_task = NULL;
706 	a.a_cookiep = NULL;
707 	newfl.l_start = fl->l_start;
708 	newfl.l_len = fl->l_len;
709 	newfl.l_type = fl->l_type;
710 	newfl.l_whence = fl->l_whence;
711 	newfl.l_pid = svid;
712 	newfl.l_sysid = NLM_SYSID_CLIENT | sysid;
713 
714 	error = lf_advlockasync(&a, &vp->v_lockf, size);
715 	KASSERT(error == 0 || error == ENOENT,
716 	    ("Failed to register NFS lock locally - error=%d", error));
717 }
718 
719 static int
720 nlm_setlock(struct nlm_host *host, struct rpc_callextra *ext,
721     rpcvers_t vers, struct timeval *timo, int retries,
722     struct vnode *vp, int op, struct flock *fl, int flags,
723     int svid, size_t fhlen, void *fh, off_t size, bool_t reclaim)
724 {
725 	struct nlm4_lockargs args;
726 	char oh_space[32];
727 	struct nlm4_res res;
728 	u_int xid;
729 	CLIENT *client;
730 	enum clnt_stat stat;
731 	int retry, block, exclusive;
732 	void *wait_handle = NULL;
733 	int error;
734 
735 	memset(&args, 0, sizeof(args));
736 	memset(&res, 0, sizeof(res));
737 
738 	block = (flags & F_WAIT) ? TRUE : FALSE;
739 	exclusive = (fl->l_type == F_WRLCK);
740 
741 	error = nlm_init_lock(fl, flags, svid, vers, fhlen, fh, size,
742 	    &args.alock, oh_space);
743 	if (error)
744 		return (error);
745 	args.block = block;
746 	args.exclusive = exclusive;
747 	args.reclaim = reclaim;
748 	args.state = nlm_nsm_state;
749 
750 	retry = 5*hz;
751 	for (;;) {
752 		client = nlm_host_get_rpc(host, FALSE);
753 		if (!client)
754 			return (ENOLCK); /* XXX retry? */
755 
756 		if (block)
757 			wait_handle = nlm_register_wait_lock(&args.alock, vp);
758 
759 		xid = atomic_fetchadd_int(&nlm_xid, 1);
760 		args.cookie.n_len = sizeof(xid);
761 		args.cookie.n_bytes = (char*) &xid;
762 
763 		stat = nlm_lock_rpc(vers, &args, &res, client, ext, *timo);
764 
765 		CLNT_RELEASE(client);
766 
767 		if (stat != RPC_SUCCESS) {
768 			if (block)
769 				nlm_deregister_wait_lock(wait_handle);
770 			if (retries) {
771 				retries--;
772 				continue;
773 			}
774 			return (EINVAL);
775 		}
776 
777 		/*
778 		 * Free res.cookie.
779 		 */
780 		xdr_free((xdrproc_t) xdr_nlm4_res, &res);
781 
782 		if (block && res.stat.stat != nlm4_blocked)
783 			nlm_deregister_wait_lock(wait_handle);
784 
785 		if (res.stat.stat == nlm4_denied_grace_period) {
786 			/*
787 			 * The server has recently rebooted and is
788 			 * giving old clients a change to reclaim
789 			 * their locks. Wait for a few seconds and try
790 			 * again.
791 			 */
792 			error = tsleep(&args, PCATCH, "nlmgrace", retry);
793 			if (error && error != EWOULDBLOCK)
794 				return (error);
795 			retry = 2*retry;
796 			if (retry > 30*hz)
797 				retry = 30*hz;
798 			continue;
799 		}
800 
801 		if (block && res.stat.stat == nlm4_blocked) {
802 			/*
803 			 * The server should call us back with a
804 			 * granted message when the lock succeeds. In
805 			 * order to deal with broken servers, lost
806 			 * granted messages and server reboots, we
807 			 * will also re-try every few seconds.
808 			 */
809 			error = nlm_wait_lock(wait_handle, retry);
810 			if (error == EWOULDBLOCK) {
811 				retry = 2*retry;
812 				if (retry > 30*hz)
813 					retry = 30*hz;
814 				continue;
815 			}
816 			if (error) {
817 				/*
818 				 * We need to call the server to
819 				 * cancel our lock request.
820 				 */
821 				nlm4_cancargs cancel;
822 
823 				memset(&cancel, 0, sizeof(cancel));
824 
825 				xid = atomic_fetchadd_int(&nlm_xid, 1);
826 				cancel.cookie.n_len = sizeof(xid);
827 				cancel.cookie.n_bytes = (char*) &xid;
828 				cancel.block = block;
829 				cancel.exclusive = exclusive;
830 				cancel.alock = args.alock;
831 
832 				do {
833 					client = nlm_host_get_rpc(host, FALSE);
834 					if (!client)
835 						/* XXX retry? */
836 						return (ENOLCK);
837 
838 					stat = nlm_cancel_rpc(vers, &cancel,
839 					    &res, client, ext, *timo);
840 
841 					CLNT_RELEASE(client);
842 
843 					if (stat != RPC_SUCCESS) {
844 						/*
845 						 * We need to cope
846 						 * with temporary
847 						 * network partitions
848 						 * as well as server
849 						 * reboots. This means
850 						 * we have to keep
851 						 * trying to cancel
852 						 * until the server
853 						 * wakes up again.
854 						 */
855 						pause("nlmcancel", 10*hz);
856 					}
857 				} while (stat != RPC_SUCCESS);
858 
859 				/*
860 				 * Free res.cookie.
861 				 */
862 				xdr_free((xdrproc_t) xdr_nlm4_res, &res);
863 
864 				switch (res.stat.stat) {
865 				case nlm_denied:
866 					/*
867 					 * There was nothing
868 					 * to cancel. We are
869 					 * going to go ahead
870 					 * and assume we got
871 					 * the lock.
872 					 */
873 					error = 0;
874 					break;
875 
876 				case nlm4_denied_grace_period:
877 					/*
878 					 * The server has
879 					 * recently rebooted -
880 					 * treat this as a
881 					 * successful
882 					 * cancellation.
883 					 */
884 					break;
885 
886 				case nlm4_granted:
887 					/*
888 					 * We managed to
889 					 * cancel.
890 					 */
891 					break;
892 
893 				default:
894 					/*
895 					 * Broken server
896 					 * implementation -
897 					 * can't really do
898 					 * anything here.
899 					 */
900 					break;
901 				}
902 
903 			}
904 		} else {
905 			error = nlm_map_status(res.stat.stat);
906 		}
907 
908 		if (!error && !reclaim) {
909 			nlm_record_lock(vp, op, fl, args.alock.svid,
910 			    nlm_host_get_sysid(host), size);
911 			nlm_host_monitor(host, 0);
912 		}
913 
914 		return (error);
915 	}
916 }
917 
918 static int
919 nlm_clearlock(struct nlm_host *host, struct rpc_callextra *ext,
920     rpcvers_t vers, struct timeval *timo, int retries,
921     struct vnode *vp, int op, struct flock *fl, int flags,
922     int svid, size_t fhlen, void *fh, off_t size)
923 {
924 	struct nlm4_unlockargs args;
925 	char oh_space[32];
926 	struct nlm4_res res;
927 	u_int xid;
928 	CLIENT *client;
929 	enum clnt_stat stat;
930 	int error;
931 
932 	memset(&args, 0, sizeof(args));
933 	memset(&res, 0, sizeof(res));
934 
935 	error = nlm_init_lock(fl, flags, svid, vers, fhlen, fh, size,
936 	    &args.alock, oh_space);
937 	if (error)
938 		return (error);
939 
940 	for (;;) {
941 		client = nlm_host_get_rpc(host, FALSE);
942 		if (!client)
943 			return (ENOLCK); /* XXX retry? */
944 
945 		xid = atomic_fetchadd_int(&nlm_xid, 1);
946 		args.cookie.n_len = sizeof(xid);
947 		args.cookie.n_bytes = (char*) &xid;
948 
949 		stat = nlm_unlock_rpc(vers, &args, &res, client, ext, *timo);
950 
951 		CLNT_RELEASE(client);
952 
953 		if (stat != RPC_SUCCESS) {
954 			if (retries) {
955 				retries--;
956 				continue;
957 			}
958 			return (EINVAL);
959 		}
960 
961 		/*
962 		 * Free res.cookie.
963 		 */
964 		xdr_free((xdrproc_t) xdr_nlm4_res, &res);
965 
966 		if (res.stat.stat == nlm4_denied_grace_period) {
967 			/*
968 			 * The server has recently rebooted and is
969 			 * giving old clients a change to reclaim
970 			 * their locks. Wait for a few seconds and try
971 			 * again.
972 			 */
973 			error = tsleep(&args, PCATCH, "nlmgrace", 5*hz);
974 			if (error && error != EWOULDBLOCK)
975 				return (error);
976 			continue;
977 		}
978 
979 		/*
980 		 * If we are being called via nlm_reclaim (which will
981 		 * use the F_REMOTE flag), don't record the lock
982 		 * operation in the local lock manager since the vnode
983 		 * is going away.
984 		 */
985 		if (!(flags & F_REMOTE))
986 			nlm_record_lock(vp, op, fl, args.alock.svid,
987 			    nlm_host_get_sysid(host), size);
988 
989 		return (0);
990 	}
991 }
992 
993 static int
994 nlm_getlock(struct nlm_host *host, struct rpc_callextra *ext,
995     rpcvers_t vers, struct timeval *timo, int retries,
996     struct vnode *vp, int op, struct flock *fl, int flags,
997     int svid, size_t fhlen, void *fh, off_t size)
998 {
999 	struct nlm4_testargs args;
1000 	char oh_space[32];
1001 	struct nlm4_testres res;
1002 	u_int xid;
1003 	CLIENT *client;
1004 	enum clnt_stat stat;
1005 	int exclusive;
1006 	int error;
1007 
1008 	KASSERT(!(flags & F_FLOCK), ("unexpected F_FLOCK for F_GETLK"));
1009 
1010 	memset(&args, 0, sizeof(args));
1011 	memset(&res, 0, sizeof(res));
1012 
1013 	exclusive = (fl->l_type == F_WRLCK);
1014 
1015 	error = nlm_init_lock(fl, flags, svid, vers, fhlen, fh, size,
1016 	    &args.alock, oh_space);
1017 	if (error)
1018 		return (error);
1019 	args.exclusive = exclusive;
1020 
1021 	for (;;) {
1022 		client = nlm_host_get_rpc(host, FALSE);
1023 		if (!client)
1024 			return (ENOLCK); /* XXX retry? */
1025 
1026 		xid = atomic_fetchadd_int(&nlm_xid, 1);
1027 		args.cookie.n_len = sizeof(xid);
1028 		args.cookie.n_bytes = (char*) &xid;
1029 
1030 		stat = nlm_test_rpc(vers, &args, &res, client, ext, *timo);
1031 
1032 		CLNT_RELEASE(client);
1033 
1034 		if (stat != RPC_SUCCESS) {
1035 			if (retries) {
1036 				retries--;
1037 				continue;
1038 			}
1039 			return (EINVAL);
1040 		}
1041 
1042 		if (res.stat.stat == nlm4_denied_grace_period) {
1043 			/*
1044 			 * The server has recently rebooted and is
1045 			 * giving old clients a change to reclaim
1046 			 * their locks. Wait for a few seconds and try
1047 			 * again.
1048 			 */
1049 			xdr_free((xdrproc_t) xdr_nlm4_testres, &res);
1050 			error = tsleep(&args, PCATCH, "nlmgrace", 5*hz);
1051 			if (error && error != EWOULDBLOCK)
1052 				return (error);
1053 			continue;
1054 		}
1055 
1056 		if (res.stat.stat == nlm4_denied) {
1057 			struct nlm4_holder *h =
1058 				&res.stat.nlm4_testrply_u.holder;
1059 			fl->l_start = h->l_offset;
1060 			fl->l_len = h->l_len;
1061 			fl->l_pid = h->svid;
1062 			if (h->exclusive)
1063 				fl->l_type = F_WRLCK;
1064 			else
1065 				fl->l_type = F_RDLCK;
1066 			fl->l_whence = SEEK_SET;
1067 			fl->l_sysid = 0;
1068 		} else {
1069 			fl->l_type = F_UNLCK;
1070 		}
1071 
1072 		xdr_free((xdrproc_t) xdr_nlm4_testres, &res);
1073 
1074 		return (0);
1075 	}
1076 }
1077 
1078 static int
1079 nlm_map_status(nlm4_stats stat)
1080 {
1081 	switch (stat) {
1082 	case nlm4_granted:
1083 		return (0);
1084 
1085 	case nlm4_denied:
1086 		return (EAGAIN);
1087 
1088 	case nlm4_denied_nolocks:
1089 		return (ENOLCK);
1090 
1091 	case nlm4_deadlck:
1092 		return (EDEADLK);
1093 
1094 	case nlm4_rofs:
1095 		return (EROFS);
1096 
1097 	case nlm4_stale_fh:
1098 		return (ESTALE);
1099 
1100 	case nlm4_fbig:
1101 		return (EFBIG);
1102 
1103 	case nlm4_failed:
1104 		return (EACCES);
1105 
1106 	default:
1107 		return (EINVAL);
1108 	}
1109 }
1110 
1111 static struct nlm_file_svid *
1112 nlm_find_svid(void *id)
1113 {
1114 	struct nlm_file_svid *ns, *newns;
1115 	int h;
1116 
1117 	h = (((uintptr_t) id) >> 7) % NLM_SVID_HASH_SIZE;
1118 
1119 	mtx_lock(&nlm_svid_lock);
1120 	LIST_FOREACH(ns, &nlm_file_svids[h], ns_link) {
1121 		if (ns->ns_id == id) {
1122 			ns->ns_refs++;
1123 			break;
1124 		}
1125 	}
1126 	mtx_unlock(&nlm_svid_lock);
1127 	if (!ns) {
1128 		int svid = alloc_unr(nlm_svid_allocator);
1129 		newns = malloc(sizeof(struct nlm_file_svid), M_NLM,
1130 		    M_WAITOK);
1131 		newns->ns_refs = 1;
1132 		newns->ns_id = id;
1133 		newns->ns_svid = svid;
1134 		newns->ns_ucred = NULL;
1135 		newns->ns_active = FALSE;
1136 
1137 		/*
1138 		 * We need to check for a race with some other
1139 		 * thread allocating a svid for this file.
1140 		 */
1141 		mtx_lock(&nlm_svid_lock);
1142 		LIST_FOREACH(ns, &nlm_file_svids[h], ns_link) {
1143 			if (ns->ns_id == id) {
1144 				ns->ns_refs++;
1145 				break;
1146 			}
1147 		}
1148 		if (ns) {
1149 			mtx_unlock(&nlm_svid_lock);
1150 			free_unr(nlm_svid_allocator, newns->ns_svid);
1151 			free(newns, M_NLM);
1152 		} else {
1153 			LIST_INSERT_HEAD(&nlm_file_svids[h], newns,
1154 			    ns_link);
1155 			ns = newns;
1156 			mtx_unlock(&nlm_svid_lock);
1157 		}
1158 	}
1159 
1160 	return (ns);
1161 }
1162 
1163 static void
1164 nlm_free_svid(struct nlm_file_svid *ns)
1165 {
1166 
1167 	mtx_lock(&nlm_svid_lock);
1168 	ns->ns_refs--;
1169 	if (!ns->ns_refs) {
1170 		KASSERT(!ns->ns_active, ("Freeing active SVID"));
1171 		LIST_REMOVE(ns, ns_link);
1172 		mtx_unlock(&nlm_svid_lock);
1173 		free_unr(nlm_svid_allocator, ns->ns_svid);
1174 		if (ns->ns_ucred)
1175 			crfree(ns->ns_ucred);
1176 		free(ns, M_NLM);
1177 	} else {
1178 		mtx_unlock(&nlm_svid_lock);
1179 	}
1180 }
1181 
1182 static int
1183 nlm_init_lock(struct flock *fl, int flags, int svid,
1184     rpcvers_t vers, size_t fhlen, void *fh, off_t size,
1185     struct nlm4_lock *lock, char oh_space[32])
1186 {
1187 	size_t oh_len;
1188 	off_t start, len;
1189 
1190 	if (fl->l_whence == SEEK_END) {
1191 		if (size > OFF_MAX
1192 		    || (fl->l_start > 0 && size > OFF_MAX - fl->l_start))
1193 			return (EOVERFLOW);
1194 		start = size + fl->l_start;
1195 	} else if (fl->l_whence == SEEK_SET || fl->l_whence == SEEK_CUR) {
1196 		start = fl->l_start;
1197 	} else {
1198 		return (EINVAL);
1199 	}
1200 	if (start < 0)
1201 		return (EINVAL);
1202 	if (fl->l_len < 0) {
1203 		len = -fl->l_len;
1204 		start -= len;
1205 		if (start < 0)
1206 			return (EINVAL);
1207 	} else {
1208 		len = fl->l_len;
1209 	}
1210 
1211 	if (vers == NLM_VERS) {
1212 		/*
1213 		 * Enforce range limits on V1 locks
1214 		 */
1215 		if (start > 0xffffffffLL || len > 0xffffffffLL)
1216 			return (EOVERFLOW);
1217 	}
1218 
1219 	snprintf(oh_space, 32, "%d@", svid);
1220 	oh_len = strlen(oh_space);
1221 	getcredhostname(NULL, oh_space + oh_len, 32 - oh_len);
1222 	oh_len = strlen(oh_space);
1223 
1224 	memset(lock, 0, sizeof(*lock));
1225 	lock->caller_name = prison0.pr_hostname;
1226 	lock->fh.n_len = fhlen;
1227 	lock->fh.n_bytes = fh;
1228 	lock->oh.n_len = oh_len;
1229 	lock->oh.n_bytes = oh_space;
1230 	lock->svid = svid;
1231 	lock->l_offset = start;
1232 	lock->l_len = len;
1233 
1234 	return (0);
1235 }
1236