xref: /freebsd/sys/nlm/nlm_advlock.c (revision f4b37ed0f8b307b1f3f0f630ca725d68f1dff30d)
1 /*-
2  * Copyright (c) 2008 Isilon Inc http://www.isilon.com/
3  * Authors: Doug Rabson <dfr@rabson.org>
4  * Developed with Red Inc: Alfred Perlstein <alfred@freebsd.org>
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  */
27 
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30 
31 #include <sys/param.h>
32 #include <sys/fcntl.h>
33 #include <sys/jail.h>
34 #include <sys/kernel.h>
35 #include <sys/limits.h>
36 #include <sys/lock.h>
37 #include <sys/lockf.h>
38 #include <sys/malloc.h>
39 #include <sys/mbuf.h>
40 #include <sys/mount.h>
41 #include <sys/mutex.h>
42 #include <sys/proc.h>
43 #include <sys/socket.h>
44 #include <sys/syslog.h>
45 #include <sys/systm.h>
46 #include <sys/unistd.h>
47 #include <sys/vnode.h>
48 
49 #include <nfs/nfsproto.h>
50 #include <nfsclient/nfs.h>
51 #include <nfsclient/nfsmount.h>
52 
53 #include <nlm/nlm_prot.h>
54 #include <nlm/nlm.h>
55 
56 /*
57  * We need to keep track of the svid values used for F_FLOCK locks.
58  */
59 struct nlm_file_svid {
60 	int		ns_refs;	/* thread count + 1 if active */
61 	int		ns_svid;	/* on-the-wire SVID for this file */
62 	struct ucred	*ns_ucred;	/* creds to use for lock recovery */
63 	void		*ns_id;		/* local struct file pointer */
64 	bool_t		ns_active;	/* TRUE if we own a lock */
65 	LIST_ENTRY(nlm_file_svid) ns_link;
66 };
67 LIST_HEAD(nlm_file_svid_list, nlm_file_svid);
68 
69 #define NLM_SVID_HASH_SIZE	256
70 struct nlm_file_svid_list nlm_file_svids[NLM_SVID_HASH_SIZE];
71 
72 struct mtx nlm_svid_lock;
73 static struct unrhdr *nlm_svid_allocator;
74 static volatile u_int nlm_xid = 1;
75 
76 static int nlm_setlock(struct nlm_host *host, struct rpc_callextra *ext,
77     rpcvers_t vers, struct timeval *timo, int retries,
78     struct vnode *vp, int op, struct flock *fl, int flags,
79     int svid, size_t fhlen, void *fh, off_t size, bool_t reclaim);
80 static int nlm_clearlock(struct nlm_host *host,  struct rpc_callextra *ext,
81     rpcvers_t vers, struct timeval *timo, int retries,
82     struct vnode *vp, int op, struct flock *fl, int flags,
83     int svid, size_t fhlen, void *fh, off_t size);
84 static int nlm_getlock(struct nlm_host *host, struct rpc_callextra *ext,
85     rpcvers_t vers, struct timeval *timo, int retries,
86     struct vnode *vp, int op, struct flock *fl, int flags,
87     int svid, size_t fhlen, void *fh, off_t size);
88 static int nlm_map_status(nlm4_stats stat);
89 static struct nlm_file_svid *nlm_find_svid(void *id);
90 static void nlm_free_svid(struct nlm_file_svid *nf);
91 static int nlm_init_lock(struct flock *fl, int flags, int svid,
92     rpcvers_t vers, size_t fhlen, void *fh, off_t size,
93     struct nlm4_lock *lock, char oh_space[32]);
94 
95 static void
96 nlm_client_init(void *dummy)
97 {
98 	int i;
99 
100 	mtx_init(&nlm_svid_lock, "NLM svid lock", NULL, MTX_DEF);
101 	/* pid_max cannot be greater than PID_MAX */
102 	nlm_svid_allocator = new_unrhdr(PID_MAX + 2, INT_MAX, &nlm_svid_lock);
103 	for (i = 0; i < NLM_SVID_HASH_SIZE; i++)
104 		LIST_INIT(&nlm_file_svids[i]);
105 }
106 SYSINIT(nlm_client_init, SI_SUB_LOCK, SI_ORDER_FIRST, nlm_client_init, NULL);
107 
108 static int
109 nlm_msg(struct thread *td, const char *server, const char *msg, int error)
110 {
111 	struct proc *p;
112 
113 	p = td ? td->td_proc : NULL;
114 	if (error) {
115 		tprintf(p, LOG_INFO, "nfs server %s: %s, error %d\n", server,
116 		    msg, error);
117 	} else {
118 		tprintf(p, LOG_INFO, "nfs server %s: %s\n", server, msg);
119 	}
120 	return (0);
121 }
122 
123 struct nlm_feedback_arg {
124 	bool_t	nf_printed;
125 	struct nfsmount *nf_nmp;
126 };
127 
128 static void
129 nlm_down(struct nlm_feedback_arg *nf, struct thread *td,
130     const char *msg, int error)
131 {
132 	struct nfsmount *nmp = nf->nf_nmp;
133 
134 	if (nmp == NULL)
135 		return;
136 	mtx_lock(&nmp->nm_mtx);
137 	if (!(nmp->nm_state & NFSSTA_LOCKTIMEO)) {
138 		nmp->nm_state |= NFSSTA_LOCKTIMEO;
139 		mtx_unlock(&nmp->nm_mtx);
140 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
141 		    VQ_NOTRESPLOCK, 0);
142 	} else {
143 		mtx_unlock(&nmp->nm_mtx);
144 	}
145 
146 	nf->nf_printed = TRUE;
147 	nlm_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error);
148 }
149 
150 static void
151 nlm_up(struct nlm_feedback_arg *nf, struct thread *td,
152     const char *msg)
153 {
154 	struct nfsmount *nmp = nf->nf_nmp;
155 
156 	if (!nf->nf_printed)
157 		return;
158 
159 	nlm_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0);
160 
161 	mtx_lock(&nmp->nm_mtx);
162 	if (nmp->nm_state & NFSSTA_LOCKTIMEO) {
163 		nmp->nm_state &= ~NFSSTA_LOCKTIMEO;
164 		mtx_unlock(&nmp->nm_mtx);
165 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
166 		    VQ_NOTRESPLOCK, 1);
167 	} else {
168 		mtx_unlock(&nmp->nm_mtx);
169 	}
170 }
171 
172 static void
173 nlm_feedback(int type, int proc, void *arg)
174 {
175 	struct thread *td = curthread;
176 	struct nlm_feedback_arg *nf = (struct nlm_feedback_arg *) arg;
177 
178 	switch (type) {
179 	case FEEDBACK_REXMIT2:
180 	case FEEDBACK_RECONNECT:
181 		nlm_down(nf, td, "lockd not responding", 0);
182 		break;
183 
184 	case FEEDBACK_OK:
185 		nlm_up(nf, td, "lockd is alive again");
186 		break;
187 	}
188 }
189 
190 /*
191  * nlm_advlock --
192  *      NFS advisory byte-level locks.
193  */
194 static int
195 nlm_advlock_internal(struct vnode *vp, void *id, int op, struct flock *fl,
196     int flags, bool_t reclaim, bool_t unlock_vp)
197 {
198 	struct thread *td = curthread;
199 	struct nfsmount *nmp;
200 	off_t size;
201 	size_t fhlen;
202 	union nfsfh fh;
203 	struct sockaddr *sa;
204 	struct sockaddr_storage ss;
205 	char servername[MNAMELEN];
206 	struct timeval timo;
207 	int retries;
208 	rpcvers_t vers;
209 	struct nlm_host *host;
210 	struct rpc_callextra ext;
211 	struct nlm_feedback_arg nf;
212 	AUTH *auth;
213 	struct ucred *cred;
214 	struct nlm_file_svid *ns;
215 	int svid;
216 	int error;
217 	int is_v3;
218 
219 	ASSERT_VOP_LOCKED(vp, "nlm_advlock_1");
220 
221 	nmp = VFSTONFS(vp->v_mount);
222 	/*
223 	 * Push any pending writes to the server and flush our cache
224 	 * so that if we are contending with another machine for a
225 	 * file, we get whatever they wrote and vice-versa.
226 	 */
227 	if (op == F_SETLK || op == F_UNLCK)
228 		nmp->nm_vinvalbuf(vp, V_SAVE, td, 1);
229 
230 	strcpy(servername, nmp->nm_hostname);
231 	nmp->nm_getinfo(vp, fh.fh_bytes, &fhlen, &ss, &is_v3, &size, &timo);
232 	sa = (struct sockaddr *) &ss;
233 	if (is_v3 != 0)
234 		vers = NLM_VERS4;
235 	else
236 		vers = NLM_VERS;
237 
238 	if (nmp->nm_flag & NFSMNT_SOFT)
239 		retries = nmp->nm_retry;
240 	else
241 		retries = INT_MAX;
242 
243 	if (unlock_vp)
244 		VOP_UNLOCK(vp, 0);
245 
246 	/*
247 	 * We need to switch to mount-point creds so that we can send
248 	 * packets from a privileged port.
249 	 */
250 	cred = td->td_ucred;
251 	td->td_ucred = vp->v_mount->mnt_cred;
252 
253 	host = nlm_find_host_by_name(servername, sa, vers);
254 	auth = authunix_create(cred);
255 	memset(&ext, 0, sizeof(ext));
256 
257 	nf.nf_printed = FALSE;
258 	nf.nf_nmp = nmp;
259 	ext.rc_auth = auth;
260 
261 	ext.rc_feedback = nlm_feedback;
262 	ext.rc_feedback_arg = &nf;
263 	ext.rc_timers = NULL;
264 
265 	ns = NULL;
266 	if (flags & F_FLOCK) {
267 		ns = nlm_find_svid(id);
268 		KASSERT(fl->l_start == 0 && fl->l_len == 0,
269 		    ("F_FLOCK lock requests must be whole-file locks"));
270 		if (!ns->ns_ucred) {
271 			/*
272 			 * Remember the creds used for locking in case
273 			 * we need to recover the lock later.
274 			 */
275 			ns->ns_ucred = crdup(cred);
276 		}
277 		svid = ns->ns_svid;
278 	} else if (flags & F_REMOTE) {
279 		/*
280 		 * If we are recovering after a server restart or
281 		 * trashing locks on a force unmount, use the same
282 		 * svid as last time.
283 		 */
284 		svid = fl->l_pid;
285 	} else {
286 		svid = ((struct proc *) id)->p_pid;
287 	}
288 
289 	switch(op) {
290 	case F_SETLK:
291 		if ((flags & (F_FLOCK|F_WAIT)) == (F_FLOCK|F_WAIT)
292 		    && fl->l_type == F_WRLCK) {
293 			/*
294 			 * The semantics for flock(2) require that any
295 			 * shared lock on the file must be released
296 			 * before an exclusive lock is granted. The
297 			 * local locking code interprets this by
298 			 * unlocking the file before sleeping on a
299 			 * blocked exclusive lock request. We
300 			 * approximate this by first attempting
301 			 * non-blocking and if that fails, we unlock
302 			 * the file and block.
303 			 */
304 			error = nlm_setlock(host, &ext, vers, &timo, retries,
305 			    vp, F_SETLK, fl, flags & ~F_WAIT,
306 			    svid, fhlen, &fh.fh_bytes, size, reclaim);
307 			if (error == EAGAIN) {
308 				fl->l_type = F_UNLCK;
309 				error = nlm_clearlock(host, &ext, vers, &timo,
310 				    retries, vp, F_UNLCK, fl, flags,
311 				    svid, fhlen, &fh.fh_bytes, size);
312 				fl->l_type = F_WRLCK;
313 				if (!error) {
314 					mtx_lock(&nlm_svid_lock);
315 					if (ns->ns_active) {
316 						ns->ns_refs--;
317 						ns->ns_active = FALSE;
318 					}
319 					mtx_unlock(&nlm_svid_lock);
320 					flags |= F_WAIT;
321 					error = nlm_setlock(host, &ext, vers,
322 					    &timo, retries, vp, F_SETLK, fl,
323 					    flags, svid, fhlen, &fh.fh_bytes,
324 					    size, reclaim);
325 				}
326 			}
327 		} else {
328 			error = nlm_setlock(host, &ext, vers, &timo, retries,
329 			    vp, op, fl, flags, svid, fhlen, &fh.fh_bytes,
330 			    size, reclaim);
331 		}
332 		if (!error && ns) {
333 			mtx_lock(&nlm_svid_lock);
334 			if (!ns->ns_active) {
335 				/*
336 				 * Add one to the reference count to
337 				 * hold onto the SVID for the lifetime
338 				 * of the lock. Note that since
339 				 * F_FLOCK only supports whole-file
340 				 * locks, there can only be one active
341 				 * lock for this SVID.
342 				 */
343 				ns->ns_refs++;
344 				ns->ns_active = TRUE;
345 			}
346 			mtx_unlock(&nlm_svid_lock);
347 		}
348 		break;
349 
350 	case F_UNLCK:
351 		error = nlm_clearlock(host, &ext, vers, &timo, retries,
352 		    vp, op, fl, flags, svid, fhlen, &fh.fh_bytes, size);
353 		if (!error && ns) {
354 			mtx_lock(&nlm_svid_lock);
355 			if (ns->ns_active) {
356 				ns->ns_refs--;
357 				ns->ns_active = FALSE;
358 			}
359 			mtx_unlock(&nlm_svid_lock);
360 		}
361 		break;
362 
363 	case F_GETLK:
364 		error = nlm_getlock(host, &ext, vers, &timo, retries,
365 		    vp, op, fl, flags, svid, fhlen, &fh.fh_bytes, size);
366 		break;
367 
368 	default:
369 		error = EINVAL;
370 		break;
371 	}
372 
373 	if (ns)
374 		nlm_free_svid(ns);
375 
376 	td->td_ucred = cred;
377 	AUTH_DESTROY(auth);
378 
379 	nlm_host_release(host);
380 
381 	return (error);
382 }
383 
384 int
385 nlm_advlock(struct vop_advlock_args *ap)
386 {
387 
388 	return (nlm_advlock_internal(ap->a_vp, ap->a_id, ap->a_op, ap->a_fl,
389 		ap->a_flags, FALSE, TRUE));
390 }
391 
392 /*
393  * Set the creds of td to the creds of the given lock's owner. The new
394  * creds reference count will be incremented via crhold. The caller is
395  * responsible for calling crfree and restoring td's original creds.
396  */
397 static void
398 nlm_set_creds_for_lock(struct thread *td, struct flock *fl)
399 {
400 	int i;
401 	struct nlm_file_svid *ns;
402 	struct proc *p;
403 	struct ucred *cred;
404 
405 	cred = NULL;
406 	if (fl->l_pid > PID_MAX) {
407 		/*
408 		 * If this was originally a F_FLOCK-style lock, we
409 		 * recorded the creds used when it was originally
410 		 * locked in the nlm_file_svid structure.
411 		 */
412 		mtx_lock(&nlm_svid_lock);
413 		for (i = 0; i < NLM_SVID_HASH_SIZE; i++) {
414 			for (ns = LIST_FIRST(&nlm_file_svids[i]); ns;
415 			     ns = LIST_NEXT(ns, ns_link)) {
416 				if (ns->ns_svid == fl->l_pid) {
417 					cred = crhold(ns->ns_ucred);
418 					break;
419 				}
420 			}
421 		}
422 		mtx_unlock(&nlm_svid_lock);
423 	} else {
424 		/*
425 		 * This lock is owned by a process. Get a reference to
426 		 * the process creds.
427 		 */
428 		p = pfind(fl->l_pid);
429 		if (p) {
430 			cred = crhold(p->p_ucred);
431 			PROC_UNLOCK(p);
432 		}
433 	}
434 
435 	/*
436 	 * If we can't find a cred, fall back on the recovery
437 	 * thread's cred.
438 	 */
439 	if (!cred) {
440 		cred = crhold(td->td_ucred);
441 	}
442 
443 	td->td_ucred = cred;
444 }
445 
446 static int
447 nlm_reclaim_free_lock(struct vnode *vp, struct flock *fl, void *arg)
448 {
449 	struct flock newfl;
450 	struct thread *td = curthread;
451 	struct ucred *oldcred;
452 	int error;
453 
454 	newfl = *fl;
455 	newfl.l_type = F_UNLCK;
456 
457 	oldcred = td->td_ucred;
458 	nlm_set_creds_for_lock(td, &newfl);
459 
460 	error = nlm_advlock_internal(vp, NULL, F_UNLCK, &newfl, F_REMOTE,
461 	    FALSE, FALSE);
462 
463 	crfree(td->td_ucred);
464 	td->td_ucred = oldcred;
465 
466 	return (error);
467 }
468 
469 int
470 nlm_reclaim(struct vop_reclaim_args *ap)
471 {
472 
473 	nlm_cancel_wait(ap->a_vp);
474 	lf_iteratelocks_vnode(ap->a_vp, nlm_reclaim_free_lock, NULL);
475 	return (0);
476 }
477 
478 struct nlm_recovery_context {
479 	struct nlm_host	*nr_host;	/* host we are recovering */
480 	int		nr_state;	/* remote NSM state for recovery */
481 };
482 
483 static int
484 nlm_client_recover_lock(struct vnode *vp, struct flock *fl, void *arg)
485 {
486 	struct nlm_recovery_context *nr = (struct nlm_recovery_context *) arg;
487 	struct thread *td = curthread;
488 	struct ucred *oldcred;
489 	int state, error;
490 
491 	/*
492 	 * If the remote NSM state changes during recovery, the host
493 	 * must have rebooted a second time. In that case, we must
494 	 * restart the recovery.
495 	 */
496 	state = nlm_host_get_state(nr->nr_host);
497 	if (nr->nr_state != state)
498 		return (ERESTART);
499 
500 	error = vn_lock(vp, LK_SHARED);
501 	if (error)
502 		return (error);
503 
504 	oldcred = td->td_ucred;
505 	nlm_set_creds_for_lock(td, fl);
506 
507 	error = nlm_advlock_internal(vp, NULL, F_SETLK, fl, F_REMOTE,
508 	    TRUE, TRUE);
509 
510 	crfree(td->td_ucred);
511 	td->td_ucred = oldcred;
512 
513 	return (error);
514 }
515 
516 void
517 nlm_client_recovery(struct nlm_host *host)
518 {
519 	struct nlm_recovery_context nr;
520 	int sysid, error;
521 
522 	sysid = NLM_SYSID_CLIENT | nlm_host_get_sysid(host);
523 	do {
524 		nr.nr_host = host;
525 		nr.nr_state = nlm_host_get_state(host);
526 		error = lf_iteratelocks_sysid(sysid,
527 		    nlm_client_recover_lock, &nr);
528 	} while (error == ERESTART);
529 }
530 
531 static void
532 nlm_convert_to_nlm_lock(struct nlm_lock *dst, struct nlm4_lock *src)
533 {
534 
535 	dst->caller_name = src->caller_name;
536 	dst->fh = src->fh;
537 	dst->oh = src->oh;
538 	dst->svid = src->svid;
539 	dst->l_offset = src->l_offset;
540 	dst->l_len = src->l_len;
541 }
542 
543 static void
544 nlm_convert_to_nlm4_holder(struct nlm4_holder *dst, struct nlm_holder *src)
545 {
546 
547 	dst->exclusive = src->exclusive;
548 	dst->svid = src->svid;
549 	dst->oh = src->oh;
550 	dst->l_offset = src->l_offset;
551 	dst->l_len = src->l_len;
552 }
553 
554 static void
555 nlm_convert_to_nlm4_res(struct nlm4_res *dst, struct nlm_res *src)
556 {
557 	dst->cookie = src->cookie;
558 	dst->stat.stat = (enum nlm4_stats) src->stat.stat;
559 }
560 
561 static enum clnt_stat
562 nlm_test_rpc(rpcvers_t vers, nlm4_testargs *args, nlm4_testres *res, CLIENT *client,
563     struct rpc_callextra *ext, struct timeval timo)
564 {
565 	if (vers == NLM_VERS4) {
566 		return nlm4_test_4(args, res, client, ext, timo);
567 	} else {
568 		nlm_testargs args1;
569 		nlm_testres res1;
570 		enum clnt_stat stat;
571 
572 		args1.cookie = args->cookie;
573 		args1.exclusive = args->exclusive;
574 		nlm_convert_to_nlm_lock(&args1.alock, &args->alock);
575 		memset(&res1, 0, sizeof(res1));
576 
577 		stat = nlm_test_1(&args1, &res1, client, ext, timo);
578 
579 		if (stat == RPC_SUCCESS) {
580 			res->cookie = res1.cookie;
581 			res->stat.stat = (enum nlm4_stats) res1.stat.stat;
582 			if (res1.stat.stat == nlm_denied)
583 				nlm_convert_to_nlm4_holder(
584 					&res->stat.nlm4_testrply_u.holder,
585 					&res1.stat.nlm_testrply_u.holder);
586 		}
587 
588 		return (stat);
589 	}
590 }
591 
592 static enum clnt_stat
593 nlm_lock_rpc(rpcvers_t vers, nlm4_lockargs *args, nlm4_res *res, CLIENT *client,
594     struct rpc_callextra *ext, struct timeval timo)
595 {
596 	if (vers == NLM_VERS4) {
597 		return nlm4_lock_4(args, res, client, ext, timo);
598 	} else {
599 		nlm_lockargs args1;
600 		nlm_res res1;
601 		enum clnt_stat stat;
602 
603 		args1.cookie = args->cookie;
604 		args1.block = args->block;
605 		args1.exclusive = args->exclusive;
606 		nlm_convert_to_nlm_lock(&args1.alock, &args->alock);
607 		args1.reclaim = args->reclaim;
608 		args1.state = args->state;
609 		memset(&res1, 0, sizeof(res1));
610 
611 		stat = nlm_lock_1(&args1, &res1, client, ext, timo);
612 
613 		if (stat == RPC_SUCCESS) {
614 			nlm_convert_to_nlm4_res(res, &res1);
615 		}
616 
617 		return (stat);
618 	}
619 }
620 
621 static enum clnt_stat
622 nlm_cancel_rpc(rpcvers_t vers, nlm4_cancargs *args, nlm4_res *res, CLIENT *client,
623     struct rpc_callextra *ext, struct timeval timo)
624 {
625 	if (vers == NLM_VERS4) {
626 		return nlm4_cancel_4(args, res, client, ext, timo);
627 	} else {
628 		nlm_cancargs args1;
629 		nlm_res res1;
630 		enum clnt_stat stat;
631 
632 		args1.cookie = args->cookie;
633 		args1.block = args->block;
634 		args1.exclusive = args->exclusive;
635 		nlm_convert_to_nlm_lock(&args1.alock, &args->alock);
636 		memset(&res1, 0, sizeof(res1));
637 
638 		stat = nlm_cancel_1(&args1, &res1, client, ext, timo);
639 
640 		if (stat == RPC_SUCCESS) {
641 			nlm_convert_to_nlm4_res(res, &res1);
642 		}
643 
644 		return (stat);
645 	}
646 }
647 
648 static enum clnt_stat
649 nlm_unlock_rpc(rpcvers_t vers, nlm4_unlockargs *args, nlm4_res *res, CLIENT *client,
650     struct rpc_callextra *ext, struct timeval timo)
651 {
652 	if (vers == NLM_VERS4) {
653 		return nlm4_unlock_4(args, res, client, ext, timo);
654 	} else {
655 		nlm_unlockargs args1;
656 		nlm_res res1;
657 		enum clnt_stat stat;
658 
659 		args1.cookie = args->cookie;
660 		nlm_convert_to_nlm_lock(&args1.alock, &args->alock);
661 		memset(&res1, 0, sizeof(res1));
662 
663 		stat = nlm_unlock_1(&args1, &res1, client, ext, timo);
664 
665 		if (stat == RPC_SUCCESS) {
666 			nlm_convert_to_nlm4_res(res, &res1);
667 		}
668 
669 		return (stat);
670 	}
671 }
672 
673 /*
674  * Called after a lock request (set or clear) succeeded. We record the
675  * details in the local lock manager. Note that since the remote
676  * server has granted the lock, we can be sure that it doesn't
677  * conflict with any other locks we have in the local lock manager.
678  *
679  * Since it is possible that host may also make NLM client requests to
680  * our NLM server, we use a different sysid value to record our own
681  * client locks.
682  *
683  * Note that since it is possible for us to receive replies from the
684  * server in a different order than the locks were granted (e.g. if
685  * many local threads are contending for the same lock), we must use a
686  * blocking operation when registering with the local lock manager.
687  * We expect that any actual wait will be rare and short hence we
688  * ignore signals for this.
689  */
690 static void
691 nlm_record_lock(struct vnode *vp, int op, struct flock *fl,
692     int svid, int sysid, off_t size)
693 {
694 	struct vop_advlockasync_args a;
695 	struct flock newfl;
696 	int error;
697 
698 	a.a_vp = vp;
699 	a.a_id = NULL;
700 	a.a_op = op;
701 	a.a_fl = &newfl;
702 	a.a_flags = F_REMOTE|F_WAIT|F_NOINTR;
703 	a.a_task = NULL;
704 	a.a_cookiep = NULL;
705 	newfl.l_start = fl->l_start;
706 	newfl.l_len = fl->l_len;
707 	newfl.l_type = fl->l_type;
708 	newfl.l_whence = fl->l_whence;
709 	newfl.l_pid = svid;
710 	newfl.l_sysid = NLM_SYSID_CLIENT | sysid;
711 
712 	error = lf_advlockasync(&a, &vp->v_lockf, size);
713 	KASSERT(error == 0 || error == ENOENT,
714 	    ("Failed to register NFS lock locally - error=%d", error));
715 }
716 
717 static int
718 nlm_setlock(struct nlm_host *host, struct rpc_callextra *ext,
719     rpcvers_t vers, struct timeval *timo, int retries,
720     struct vnode *vp, int op, struct flock *fl, int flags,
721     int svid, size_t fhlen, void *fh, off_t size, bool_t reclaim)
722 {
723 	struct nlm4_lockargs args;
724 	char oh_space[32];
725 	struct nlm4_res res;
726 	u_int xid;
727 	CLIENT *client;
728 	enum clnt_stat stat;
729 	int retry, block, exclusive;
730 	void *wait_handle = NULL;
731 	int error;
732 
733 	memset(&args, 0, sizeof(args));
734 	memset(&res, 0, sizeof(res));
735 
736 	block = (flags & F_WAIT) ? TRUE : FALSE;
737 	exclusive = (fl->l_type == F_WRLCK);
738 
739 	error = nlm_init_lock(fl, flags, svid, vers, fhlen, fh, size,
740 	    &args.alock, oh_space);
741 	if (error)
742 		return (error);
743 	args.block = block;
744 	args.exclusive = exclusive;
745 	args.reclaim = reclaim;
746 	args.state = nlm_nsm_state;
747 
748 	retry = 5*hz;
749 	for (;;) {
750 		client = nlm_host_get_rpc(host, FALSE);
751 		if (!client)
752 			return (ENOLCK); /* XXX retry? */
753 
754 		if (block)
755 			wait_handle = nlm_register_wait_lock(&args.alock, vp);
756 
757 		xid = atomic_fetchadd_int(&nlm_xid, 1);
758 		args.cookie.n_len = sizeof(xid);
759 		args.cookie.n_bytes = (char*) &xid;
760 
761 		stat = nlm_lock_rpc(vers, &args, &res, client, ext, *timo);
762 
763 		CLNT_RELEASE(client);
764 
765 		if (stat != RPC_SUCCESS) {
766 			if (block)
767 				nlm_deregister_wait_lock(wait_handle);
768 			if (retries) {
769 				retries--;
770 				continue;
771 			}
772 			return (EINVAL);
773 		}
774 
775 		/*
776 		 * Free res.cookie.
777 		 */
778 		xdr_free((xdrproc_t) xdr_nlm4_res, &res);
779 
780 		if (block && res.stat.stat != nlm4_blocked)
781 			nlm_deregister_wait_lock(wait_handle);
782 
783 		if (res.stat.stat == nlm4_denied_grace_period) {
784 			/*
785 			 * The server has recently rebooted and is
786 			 * giving old clients a change to reclaim
787 			 * their locks. Wait for a few seconds and try
788 			 * again.
789 			 */
790 			error = tsleep(&args, PCATCH, "nlmgrace", retry);
791 			if (error && error != EWOULDBLOCK)
792 				return (error);
793 			retry = 2*retry;
794 			if (retry > 30*hz)
795 				retry = 30*hz;
796 			continue;
797 		}
798 
799 		if (block && res.stat.stat == nlm4_blocked) {
800 			/*
801 			 * The server should call us back with a
802 			 * granted message when the lock succeeds. In
803 			 * order to deal with broken servers, lost
804 			 * granted messages and server reboots, we
805 			 * will also re-try every few seconds.
806 			 */
807 			error = nlm_wait_lock(wait_handle, retry);
808 			if (error == EWOULDBLOCK) {
809 				retry = 2*retry;
810 				if (retry > 30*hz)
811 					retry = 30*hz;
812 				continue;
813 			}
814 			if (error) {
815 				/*
816 				 * We need to call the server to
817 				 * cancel our lock request.
818 				 */
819 				nlm4_cancargs cancel;
820 
821 				memset(&cancel, 0, sizeof(cancel));
822 
823 				xid = atomic_fetchadd_int(&nlm_xid, 1);
824 				cancel.cookie.n_len = sizeof(xid);
825 				cancel.cookie.n_bytes = (char*) &xid;
826 				cancel.block = block;
827 				cancel.exclusive = exclusive;
828 				cancel.alock = args.alock;
829 
830 				do {
831 					client = nlm_host_get_rpc(host, FALSE);
832 					if (!client)
833 						/* XXX retry? */
834 						return (ENOLCK);
835 
836 					stat = nlm_cancel_rpc(vers, &cancel,
837 					    &res, client, ext, *timo);
838 
839 					CLNT_RELEASE(client);
840 
841 					if (stat != RPC_SUCCESS) {
842 						/*
843 						 * We need to cope
844 						 * with temporary
845 						 * network partitions
846 						 * as well as server
847 						 * reboots. This means
848 						 * we have to keep
849 						 * trying to cancel
850 						 * until the server
851 						 * wakes up again.
852 						 */
853 						pause("nlmcancel", 10*hz);
854 					}
855 				} while (stat != RPC_SUCCESS);
856 
857 				/*
858 				 * Free res.cookie.
859 				 */
860 				xdr_free((xdrproc_t) xdr_nlm4_res, &res);
861 
862 				switch (res.stat.stat) {
863 				case nlm_denied:
864 					/*
865 					 * There was nothing
866 					 * to cancel. We are
867 					 * going to go ahead
868 					 * and assume we got
869 					 * the lock.
870 					 */
871 					error = 0;
872 					break;
873 
874 				case nlm4_denied_grace_period:
875 					/*
876 					 * The server has
877 					 * recently rebooted -
878 					 * treat this as a
879 					 * successful
880 					 * cancellation.
881 					 */
882 					break;
883 
884 				case nlm4_granted:
885 					/*
886 					 * We managed to
887 					 * cancel.
888 					 */
889 					break;
890 
891 				default:
892 					/*
893 					 * Broken server
894 					 * implementation -
895 					 * can't really do
896 					 * anything here.
897 					 */
898 					break;
899 				}
900 
901 			}
902 		} else {
903 			error = nlm_map_status(res.stat.stat);
904 		}
905 
906 		if (!error && !reclaim) {
907 			nlm_record_lock(vp, op, fl, args.alock.svid,
908 			    nlm_host_get_sysid(host), size);
909 			nlm_host_monitor(host, 0);
910 		}
911 
912 		return (error);
913 	}
914 }
915 
916 static int
917 nlm_clearlock(struct nlm_host *host, struct rpc_callextra *ext,
918     rpcvers_t vers, struct timeval *timo, int retries,
919     struct vnode *vp, int op, struct flock *fl, int flags,
920     int svid, size_t fhlen, void *fh, off_t size)
921 {
922 	struct nlm4_unlockargs args;
923 	char oh_space[32];
924 	struct nlm4_res res;
925 	u_int xid;
926 	CLIENT *client;
927 	enum clnt_stat stat;
928 	int error;
929 
930 	memset(&args, 0, sizeof(args));
931 	memset(&res, 0, sizeof(res));
932 
933 	error = nlm_init_lock(fl, flags, svid, vers, fhlen, fh, size,
934 	    &args.alock, oh_space);
935 	if (error)
936 		return (error);
937 
938 	for (;;) {
939 		client = nlm_host_get_rpc(host, FALSE);
940 		if (!client)
941 			return (ENOLCK); /* XXX retry? */
942 
943 		xid = atomic_fetchadd_int(&nlm_xid, 1);
944 		args.cookie.n_len = sizeof(xid);
945 		args.cookie.n_bytes = (char*) &xid;
946 
947 		stat = nlm_unlock_rpc(vers, &args, &res, client, ext, *timo);
948 
949 		CLNT_RELEASE(client);
950 
951 		if (stat != RPC_SUCCESS) {
952 			if (retries) {
953 				retries--;
954 				continue;
955 			}
956 			return (EINVAL);
957 		}
958 
959 		/*
960 		 * Free res.cookie.
961 		 */
962 		xdr_free((xdrproc_t) xdr_nlm4_res, &res);
963 
964 		if (res.stat.stat == nlm4_denied_grace_period) {
965 			/*
966 			 * The server has recently rebooted and is
967 			 * giving old clients a change to reclaim
968 			 * their locks. Wait for a few seconds and try
969 			 * again.
970 			 */
971 			error = tsleep(&args, PCATCH, "nlmgrace", 5*hz);
972 			if (error && error != EWOULDBLOCK)
973 				return (error);
974 			continue;
975 		}
976 
977 		/*
978 		 * If we are being called via nlm_reclaim (which will
979 		 * use the F_REMOTE flag), don't record the lock
980 		 * operation in the local lock manager since the vnode
981 		 * is going away.
982 		 */
983 		if (!(flags & F_REMOTE))
984 			nlm_record_lock(vp, op, fl, args.alock.svid,
985 			    nlm_host_get_sysid(host), size);
986 
987 		return (0);
988 	}
989 }
990 
991 static int
992 nlm_getlock(struct nlm_host *host, struct rpc_callextra *ext,
993     rpcvers_t vers, struct timeval *timo, int retries,
994     struct vnode *vp, int op, struct flock *fl, int flags,
995     int svid, size_t fhlen, void *fh, off_t size)
996 {
997 	struct nlm4_testargs args;
998 	char oh_space[32];
999 	struct nlm4_testres res;
1000 	u_int xid;
1001 	CLIENT *client;
1002 	enum clnt_stat stat;
1003 	int exclusive;
1004 	int error;
1005 
1006 	KASSERT(!(flags & F_FLOCK), ("unexpected F_FLOCK for F_GETLK"));
1007 
1008 	memset(&args, 0, sizeof(args));
1009 	memset(&res, 0, sizeof(res));
1010 
1011 	exclusive = (fl->l_type == F_WRLCK);
1012 
1013 	error = nlm_init_lock(fl, flags, svid, vers, fhlen, fh, size,
1014 	    &args.alock, oh_space);
1015 	if (error)
1016 		return (error);
1017 	args.exclusive = exclusive;
1018 
1019 	for (;;) {
1020 		client = nlm_host_get_rpc(host, FALSE);
1021 		if (!client)
1022 			return (ENOLCK); /* XXX retry? */
1023 
1024 		xid = atomic_fetchadd_int(&nlm_xid, 1);
1025 		args.cookie.n_len = sizeof(xid);
1026 		args.cookie.n_bytes = (char*) &xid;
1027 
1028 		stat = nlm_test_rpc(vers, &args, &res, client, ext, *timo);
1029 
1030 		CLNT_RELEASE(client);
1031 
1032 		if (stat != RPC_SUCCESS) {
1033 			if (retries) {
1034 				retries--;
1035 				continue;
1036 			}
1037 			return (EINVAL);
1038 		}
1039 
1040 		if (res.stat.stat == nlm4_denied_grace_period) {
1041 			/*
1042 			 * The server has recently rebooted and is
1043 			 * giving old clients a change to reclaim
1044 			 * their locks. Wait for a few seconds and try
1045 			 * again.
1046 			 */
1047 			xdr_free((xdrproc_t) xdr_nlm4_testres, &res);
1048 			error = tsleep(&args, PCATCH, "nlmgrace", 5*hz);
1049 			if (error && error != EWOULDBLOCK)
1050 				return (error);
1051 			continue;
1052 		}
1053 
1054 		if (res.stat.stat == nlm4_denied) {
1055 			struct nlm4_holder *h =
1056 				&res.stat.nlm4_testrply_u.holder;
1057 			fl->l_start = h->l_offset;
1058 			fl->l_len = h->l_len;
1059 			fl->l_pid = h->svid;
1060 			if (h->exclusive)
1061 				fl->l_type = F_WRLCK;
1062 			else
1063 				fl->l_type = F_RDLCK;
1064 			fl->l_whence = SEEK_SET;
1065 			fl->l_sysid = 0;
1066 		} else {
1067 			fl->l_type = F_UNLCK;
1068 		}
1069 
1070 		xdr_free((xdrproc_t) xdr_nlm4_testres, &res);
1071 
1072 		return (0);
1073 	}
1074 }
1075 
1076 static int
1077 nlm_map_status(nlm4_stats stat)
1078 {
1079 	switch (stat) {
1080 	case nlm4_granted:
1081 		return (0);
1082 
1083 	case nlm4_denied:
1084 		return (EAGAIN);
1085 
1086 	case nlm4_denied_nolocks:
1087 		return (ENOLCK);
1088 
1089 	case nlm4_deadlck:
1090 		return (EDEADLK);
1091 
1092 	case nlm4_rofs:
1093 		return (EROFS);
1094 
1095 	case nlm4_stale_fh:
1096 		return (ESTALE);
1097 
1098 	case nlm4_fbig:
1099 		return (EFBIG);
1100 
1101 	case nlm4_failed:
1102 		return (EACCES);
1103 
1104 	default:
1105 		return (EINVAL);
1106 	}
1107 }
1108 
1109 static struct nlm_file_svid *
1110 nlm_find_svid(void *id)
1111 {
1112 	struct nlm_file_svid *ns, *newns;
1113 	int h;
1114 
1115 	h = (((uintptr_t) id) >> 7) % NLM_SVID_HASH_SIZE;
1116 
1117 	mtx_lock(&nlm_svid_lock);
1118 	LIST_FOREACH(ns, &nlm_file_svids[h], ns_link) {
1119 		if (ns->ns_id == id) {
1120 			ns->ns_refs++;
1121 			break;
1122 		}
1123 	}
1124 	mtx_unlock(&nlm_svid_lock);
1125 	if (!ns) {
1126 		int svid = alloc_unr(nlm_svid_allocator);
1127 		newns = malloc(sizeof(struct nlm_file_svid), M_NLM,
1128 		    M_WAITOK);
1129 		newns->ns_refs = 1;
1130 		newns->ns_id = id;
1131 		newns->ns_svid = svid;
1132 		newns->ns_ucred = NULL;
1133 		newns->ns_active = FALSE;
1134 
1135 		/*
1136 		 * We need to check for a race with some other
1137 		 * thread allocating a svid for this file.
1138 		 */
1139 		mtx_lock(&nlm_svid_lock);
1140 		LIST_FOREACH(ns, &nlm_file_svids[h], ns_link) {
1141 			if (ns->ns_id == id) {
1142 				ns->ns_refs++;
1143 				break;
1144 			}
1145 		}
1146 		if (ns) {
1147 			mtx_unlock(&nlm_svid_lock);
1148 			free_unr(nlm_svid_allocator, newns->ns_svid);
1149 			free(newns, M_NLM);
1150 		} else {
1151 			LIST_INSERT_HEAD(&nlm_file_svids[h], newns,
1152 			    ns_link);
1153 			ns = newns;
1154 			mtx_unlock(&nlm_svid_lock);
1155 		}
1156 	}
1157 
1158 	return (ns);
1159 }
1160 
1161 static void
1162 nlm_free_svid(struct nlm_file_svid *ns)
1163 {
1164 
1165 	mtx_lock(&nlm_svid_lock);
1166 	ns->ns_refs--;
1167 	if (!ns->ns_refs) {
1168 		KASSERT(!ns->ns_active, ("Freeing active SVID"));
1169 		LIST_REMOVE(ns, ns_link);
1170 		mtx_unlock(&nlm_svid_lock);
1171 		free_unr(nlm_svid_allocator, ns->ns_svid);
1172 		if (ns->ns_ucred)
1173 			crfree(ns->ns_ucred);
1174 		free(ns, M_NLM);
1175 	} else {
1176 		mtx_unlock(&nlm_svid_lock);
1177 	}
1178 }
1179 
1180 static int
1181 nlm_init_lock(struct flock *fl, int flags, int svid,
1182     rpcvers_t vers, size_t fhlen, void *fh, off_t size,
1183     struct nlm4_lock *lock, char oh_space[32])
1184 {
1185 	size_t oh_len;
1186 	off_t start, len;
1187 
1188 	if (fl->l_whence == SEEK_END) {
1189 		if (size > OFF_MAX
1190 		    || (fl->l_start > 0 && size > OFF_MAX - fl->l_start))
1191 			return (EOVERFLOW);
1192 		start = size + fl->l_start;
1193 	} else if (fl->l_whence == SEEK_SET || fl->l_whence == SEEK_CUR) {
1194 		start = fl->l_start;
1195 	} else {
1196 		return (EINVAL);
1197 	}
1198 	if (start < 0)
1199 		return (EINVAL);
1200 	if (fl->l_len < 0) {
1201 		len = -fl->l_len;
1202 		start -= len;
1203 		if (start < 0)
1204 			return (EINVAL);
1205 	} else {
1206 		len = fl->l_len;
1207 	}
1208 
1209 	if (vers == NLM_VERS) {
1210 		/*
1211 		 * Enforce range limits on V1 locks
1212 		 */
1213 		if (start > 0xffffffffLL || len > 0xffffffffLL)
1214 			return (EOVERFLOW);
1215 	}
1216 
1217 	snprintf(oh_space, 32, "%d@", svid);
1218 	oh_len = strlen(oh_space);
1219 	getcredhostname(NULL, oh_space + oh_len, 32 - oh_len);
1220 	oh_len = strlen(oh_space);
1221 
1222 	memset(lock, 0, sizeof(*lock));
1223 	lock->caller_name = prison0.pr_hostname;
1224 	lock->fh.n_len = fhlen;
1225 	lock->fh.n_bytes = fh;
1226 	lock->oh.n_len = oh_len;
1227 	lock->oh.n_bytes = oh_space;
1228 	lock->svid = svid;
1229 	lock->l_offset = start;
1230 	lock->l_len = len;
1231 
1232 	return (0);
1233 }
1234