1/*
2 * Copyright (c) 2002-2010 Apple Inc.  All rights reserved.
3 *
4 * @APPLE_LICENSE_HEADER_START@
5 *
6 * This file contains Original Code and/or Modifications of Original Code
7 * as defined in and that are subject to the Apple Public Source License
8 * Version 2.0 (the 'License'). You may not use this file except in
9 * compliance with the License. Please obtain a copy of the License at
10 * http://www.opensource.apple.com/apsl/ and read it before using this
11 * file.
12 *
13 * The Original Code and all software distributed under the License are
14 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
15 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
16 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
18 * Please see the License for the specific language governing rights and
19 * limitations under the License.
20 *
21 * @APPLE_LICENSE_HEADER_END@
22 */
23/*	$NetBSD: lockd_lock.c,v 1.5 2000/11/21 03:47:41 enami Exp $	*/
24/*	$FreeBSD: src/usr.sbin/rpc.lockd/lockd_lock.c,v 1.10 2002/03/22 19:57:09 alfred Exp $ */
25
26/*
27 * Copyright (c) 2001 Andrew P. Lentvorski, Jr.
28 * Copyright (c) 2000 Manuel Bouyer.
29 *
30 * Redistribution and use in source and binary forms, with or without
31 * modification, are permitted provided that the following conditions
32 * are met:
33 * 1. Redistributions of source code must retain the above copyright
34 *    notice, this list of conditions and the following disclaimer.
35 * 2. Redistributions in binary form must reproduce the above copyright
36 *    notice, this list of conditions and the following disclaimer in the
37 *    documentation and/or other materials provided with the distribution.
38 * 3. All advertising materials mentioning features or use of this software
39 *    must display the following acknowledgement:
40 *	This product includes software developed by the University of
41 *	California, Berkeley and its contributors.
42 * 4. Neither the name of the University nor the names of its contributors
43 *    may be used to endorse or promote products derived from this software
44 *    without specific prior written permission.
45 *
46 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
47 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
48 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
49 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
50 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
51 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
52 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
53 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
54 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
55 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
56 * SUCH DAMAGE.
57 *
58 */
59
60#define LOCKD_DEBUG
61
62#include <stdio.h>
63#ifdef LOCKD_DEBUG
64#include <stdarg.h>
65#endif
66#include <stdlib.h>
67#include <unistd.h>
68#include <fcntl.h>
69#include <syslog.h>
70#include <errno.h>
71#include <string.h>
72#include <signal.h>
73#include <oncrpc/rpc.h>
74#include <sys/types.h>
75#include <sys/stat.h>
76#include <sys/socket.h>
77#include <sys/param.h>
78#include <sys/mount.h>
79#include <sys/wait.h>
80#include <arpa/inet.h>
81#include "sm_inter.h"
82#include "nlm_prot.h"
83
84#include <nfs/rpcv2.h>
85#include <nfs/nfs_lock.h>
86#include <nfs/nfs.h>
87
88#include "lockd.h"
89#include "lockd_lock.h"
90
91#define MAXOBJECTSIZE 64
92#define MAXBUFFERSIZE 1024
93
94#define AOK	(void *)	// assert alignment is OK
95
96/*
97 * A set of utilities for managing file locking
98 *
99 * XXX: All locks are in a linked list, a better structure should be used
100 * to improve search/access effeciency.
101 */
102
103/* struct describing a lock */
104struct file_lock {
105	LIST_ENTRY(file_lock) nfslocklist;
106	netobj filehandle; /* NFS filehandle */
107	struct sockaddr *addr;
108	struct nlm4_holder client; /* lock holder */
109	u_int64_t granted_cookie;
110	int nsm_status; /* status from the remote lock manager */
111	int status; /* lock status, see below */
112	int flags; /* lock flags, see lockd_lock.h */
113	int blocking; /* blocking lock or not */
114	char client_name[SM_MAXSTRLEN];	/* client_name is really variable length and must be last! */
115};
116
117LIST_HEAD(nfslocklist_head, file_lock);
118struct nfslocklist_head nfslocklist_head = LIST_HEAD_INITIALIZER(nfslocklist_head);
119
120LIST_HEAD(blockedlocklist_head, file_lock);
121struct blockedlocklist_head blockedlocklist_head = LIST_HEAD_INITIALIZER(blockedlocklist_head);
122
123/* struct describing a share reservation */
124struct file_share {
125	LIST_ENTRY(file_share) nfssharelist;
126	netobj oh; /* share holder */
127	short mode;
128	short access;
129	char client_name[SM_MAXSTRLEN]; /* name is really variable length and must be last! */
130};
131LIST_HEAD(nfssharelist_head, file_share);
132
133/* Struct describing a file with share reservations */
134struct sharefile {
135	LIST_ENTRY(sharefile) sharefilelist;
136	netobj filehandle; /* Local access filehandle */
137	int fd; /* file descriptor: remains open until no more shares */
138	int refcount;
139	struct nfssharelist_head sharelist_head;
140};
141LIST_HEAD(nfssharefilelist_head, sharefile);
142struct nfssharefilelist_head nfssharefilelist_head = LIST_HEAD_INITIALIZER(nfssharefilelist_head);
143
144/* lock status */
145#define LKST_LOCKED	1 /* lock is locked */
146/* XXX: Is this flag file specific or lock specific? */
147#define LKST_WAITING	2 /* file is already locked by another host */
148#define LKST_PROCESSING	3 /* child is trying to aquire the lock */
149#define LKST_DYING	4 /* must dies when we get news from the child */
150
151/* struct describing a monitored host */
152struct host {
153	TAILQ_ENTRY(host) hostlst;
154	int refcnt;
155	time_t lastuse;
156	struct sockaddr_storage addr;
157	char *name;	/* host name provided by client via caller_name */
158	char *revname;	/* host name mapped from addr */
159};
160/* list of hosts we monitor */
161TAILQ_HEAD(hostlst_head, host);
162struct hostlst_head hostlst_head = TAILQ_HEAD_INITIALIZER(hostlst_head);
163struct hostlst_head hostlst_unref = TAILQ_HEAD_INITIALIZER(hostlst_unref);
164
165u_int64_t send_granted_cookie = 0;
166
167/*
168 * File monitoring handlers
169 * XXX: These might be able to be removed when kevent support
170 * is placed into the hardware lock/unlock routines.  (ie.
171 * let the kernel do all the file monitoring)
172 */
173
174/* Struct describing a monitored file */
175struct monfile {
176	LIST_ENTRY(monfile) monfilelist;
177	netobj filehandle; /* Local access filehandle */
178	int fd; /* file descriptor: remains open until unlock! */
179	int refcount;
180	int exclusive;
181};
182
183/* List of files we monitor */
184LIST_HEAD(monfilelist_head, monfile);
185struct monfilelist_head monfilelist_head = LIST_HEAD_INITIALIZER(monfilelist_head);
186
187static int debugdelay = 0;
188
189enum nfslock_status { NFS_GRANTED = 0, NFS_GRANTED_DUPLICATE,
190		      NFS_DENIED, NFS_DENIED_NOLOCK,
191		      NFS_RESERR };
192
193enum hwlock_status { HW_GRANTED = 0, HW_GRANTED_DUPLICATE,
194		     HW_DENIED, HW_DENIED_NOLOCK,
195		     HW_STALEFH, HW_READONLY, HW_RESERR };
196
197enum partialfilelock_status { PFL_GRANTED=0, PFL_GRANTED_DUPLICATE, PFL_DENIED,
198			      PFL_NFSDENIED, PFL_NFSBLOCKED, PFL_NFSDENIED_NOLOCK, PFL_NFSRESERR,
199			      PFL_HWDENIED,  PFL_HWBLOCKED,  PFL_HWDENIED_NOLOCK, PFL_HWRESERR,
200			      PFL_HWDENIED_STALEFH, PFL_HWDENIED_READONLY };
201
202enum LFLAGS {LEDGE_LEFT, LEDGE_LBOUNDARY, LEDGE_INSIDE, LEDGE_RBOUNDARY, LEDGE_RIGHT};
203enum RFLAGS {REDGE_LEFT, REDGE_LBOUNDARY, REDGE_INSIDE, REDGE_RBOUNDARY, REDGE_RIGHT};
204/* XXX: WARNING! I HAVE OVERLOADED THIS STATUS ENUM!  SPLIT IT APART INTO TWO */
205enum split_status {SPL_DISJOINT=0, SPL_LOCK1=1, SPL_LOCK2=2, SPL_CONTAINED=4, SPL_RESERR=8};
206
207enum partialfilelock_status lock_partialfilelock(struct file_lock *fl);
208
209int send_granted(struct file_lock *fl, int opcode);
210void siglock(void);
211void sigunlock(void);
212void destroy_lock_host(struct host *ihp);
213static void monitor_lock_host(const char *hostname, const struct sockaddr *addr);
214
215void	copy_nlm4_lock_to_nlm4_holder(const struct nlm4_lock *src,
216    const bool_t exclusive, struct nlm4_holder *dest);
217struct file_lock *	allocate_file_lock(const netobj *lockowner,
218    const netobj *filehandle, const struct sockaddr *addr,
219    const char *caller_name);
220void	deallocate_file_lock(struct file_lock *fl);
221void	fill_file_lock(struct file_lock *fl,
222    const bool_t exclusive, const int32_t svid,
223    const u_int64_t offset, const u_int64_t len,
224    const int state, const int status, const int flags, const int blocking);
225int	regions_overlap(const u_int64_t start1, const u_int64_t len1,
226    const u_int64_t start2, const u_int64_t len2);;
227enum split_status  region_compare(const u_int64_t starte, const u_int64_t lene,
228    const u_int64_t startu, const u_int64_t lenu,
229    u_int64_t *start1, u_int64_t *len1, u_int64_t *start2, u_int64_t *len2);
230int	same_netobj(const netobj *n0, const netobj *n1);
231int	same_filelock_identity(const struct file_lock *fl0,
232    const struct file_lock *fl2);
233
234static void debuglog(char const *fmt, ...);
235void dump_static_object(const unsigned char* object, const int sizeof_object,
236                        unsigned char* hbuff, const int sizeof_hbuff,
237                        unsigned char* cbuff, const int sizeof_cbuff);
238void dump_netobj(const struct netobj *nobj);
239void dump_filelock(const struct file_lock *fl);
240struct file_lock *	get_lock_matching_unlock(const struct file_lock *fl, int cleanup);
241enum nfslock_status	test_nfslock(const struct file_lock *fl,
242    struct file_lock **conflicting_fl);
243enum nfslock_status	lock_nfslock(struct file_lock *fl);
244enum nfslock_status	delete_nfslock(struct file_lock *fl);
245enum nfslock_status	unlock_nfslock(const struct file_lock *fl,
246    struct file_lock **released_lock, struct file_lock **left_lock,
247    struct file_lock **right_lock, int cleanup);
248enum hwlock_status lock_hwlock(struct file_lock *fl);
249enum split_status split_nfslock(const struct file_lock *exist_lock,
250    const struct file_lock *unlock_lock, struct file_lock **left_lock,
251    struct file_lock **right_lock);
252void	add_blockingfilelock(struct file_lock *fl);
253enum hwlock_status	unlock_hwlock(const struct file_lock *fl);
254enum hwlock_status	test_hwlock(const struct file_lock *fl,
255    struct file_lock **conflicting_fl);
256void	remove_blockingfilelock(struct file_lock *fl);
257void	clear_blockingfilelock(const char *hostname);
258void	retry_blockingfilelocklist(netobj *fh);
259enum partialfilelock_status	unlock_partialfilelock(
260    const struct file_lock *fl, int cleanup);
261void	clear_partialfilelock(const char *hostname);
262enum partialfilelock_status	test_partialfilelock(
263    const struct file_lock *fl, struct file_lock **conflicting_fl);
264enum nlm4_stats	do_test(struct file_lock *fl, struct file_lock **conflicting_fl);
265enum nlm4_stats	do_unlock(struct file_lock *fl);
266enum nlm4_stats	do_lock(struct file_lock *fl);
267void	do_clear(const char *hostname);
268
269
270#pragma clang diagnostic push
271#pragma clang diagnostic ignored "-Wformat-nonliteral"
272
273void
274debuglog(char const *fmt, ...)
275{
276	va_list ap;
277
278	if (config.verbose < 1) {
279		return;
280	}
281
282	sleep(debugdelay);
283
284	va_start(ap, fmt);
285	vsyslog(LOG_DEBUG, fmt, ap);
286	va_end(ap);
287}
288#pragma clang diagnostic pop
289
290void
291dump_static_object(object, size_object, hbuff, size_hbuff, cbuff, size_cbuff)
292	const unsigned char *object;
293	const int size_object;
294	unsigned char *hbuff;
295	const int size_hbuff;
296	unsigned char *cbuff;
297	const int size_cbuff;
298{
299	int i, objectsize;
300	char *tmp = (char*)hbuff;
301
302	if (config.verbose < 2) {
303		return;
304	}
305
306	objectsize = size_object;
307
308	if (objectsize == 0) {
309		debuglog("object is size 0\n");
310	} else {
311		if (objectsize > MAXOBJECTSIZE) {
312			debuglog("Object of size %d being clamped"
313			    "to size %d\n", objectsize, MAXOBJECTSIZE);
314			objectsize = MAXOBJECTSIZE;
315		}
316
317		if (hbuff != NULL) {
318			if (size_hbuff < objectsize*2+1) {
319				debuglog("Hbuff not large enough."
320				    "  Increase size\n");
321			} else {
322				for(i=0;i<objectsize;i++,tmp+=2) {
323					snprintf(tmp, size_hbuff - (2*i), "%02X", *(object+i));
324				}
325				*tmp = '\0';
326			}
327		}
328
329		if (cbuff != NULL) {
330			if (size_cbuff < objectsize+1) {
331				debuglog("Cbuff not large enough."
332				    "  Increase Size\n");
333			}
334
335			for(i=0;i<objectsize;i++) {
336				if (*(object+i) >= 32 && *(object+i) <= 127) {
337					*(cbuff+i) = *(object+i);
338				} else {
339					*(cbuff+i) = '.';
340				}
341			}
342			*(cbuff+i) = '\0';
343		}
344	}
345}
346
347void
348dump_netobj(const struct netobj *nobj)
349{
350	char hbuff[MAXBUFFERSIZE*2];
351	char cbuff[MAXBUFFERSIZE];
352
353	if (config.verbose < 2) {
354		return;
355	}
356
357	if (nobj == NULL) {
358		debuglog("Null netobj pointer\n");
359	}
360	else if (nobj->n_len == 0) {
361		debuglog("Size zero netobj\n");
362	} else {
363		dump_static_object((const unsigned char *)nobj->n_bytes, nobj->n_len,
364		    (unsigned char *)hbuff, sizeof(hbuff), (unsigned char *)cbuff, sizeof(cbuff));
365		debuglog("netobj: len: %d  data: %s :::  %s\n",
366		    nobj->n_len, hbuff, cbuff);
367	}
368}
369
370/* #define DUMP_FILELOCK_VERBOSE */
371void
372dump_filelock(const struct file_lock *fl)
373{
374#ifdef DUMP_FILELOCK_VERBOSE
375	char hbuff[MAXBUFFERSIZE*2];
376	char cbuff[MAXBUFFERSIZE];
377#endif
378
379	if (config.verbose < 2) {
380		return;
381	}
382
383	if (fl != NULL) {
384		debuglog("Dumping file lock structure @ %p\n", fl);
385
386#ifdef DUMP_FILELOCK_VERBOSE
387		dump_static_object((unsigned char *)&fl->filehandle.n_bytes,
388		    fl->filehandle.n_len, hbuff, sizeof(hbuff),
389		    cbuff, sizeof(cbuff));
390		debuglog("Filehandle: %8s  :::  %8s\n", hbuff, cbuff);
391#endif
392
393		debuglog("Dumping nlm4_holder:\n"
394		    "exc: %x  svid: %x  offset:len %llx:%llx\n",
395		    fl->client.exclusive, fl->client.svid,
396		    fl->client.l_offset, fl->client.l_len);
397
398#ifdef DUMP_FILELOCK_VERBOSE
399		debuglog("Dumping client identity:\n");
400		dump_netobj(&fl->client.oh);
401
402		debuglog("nsm: %d  status: %d  flags: %d  locker: %d"
403		    "  fd:  %d\n", fl->nsm_status, fl->status,
404		    fl->flags, fl->locker, fl->fd);
405#endif
406	} else {
407		debuglog("NULL file lock structure\n");
408	}
409}
410
411void
412copy_nlm4_lock_to_nlm4_holder(src, exclusive, dest)
413	const struct nlm4_lock *src;
414	const bool_t exclusive;
415	struct nlm4_holder *dest;
416{
417
418	dest->exclusive = exclusive;
419	dest->oh.n_len = src->oh.n_len;
420	dest->oh.n_bytes = src->oh.n_bytes;
421	dest->svid = src->svid;
422	dest->l_offset = src->l_offset;
423	dest->l_len = src->l_len;
424}
425
426
427/*
428 * allocate_file_lock: Create a lock with the given parameters
429 */
430
431struct file_lock *
432allocate_file_lock(const netobj *lockowner, const netobj *filehandle,
433		   const struct sockaddr *addr, const char *caller_name)
434{
435	struct file_lock *newfl;
436	size_t n;
437
438	/* Beware of rubbish input! */
439	n = strnlen(caller_name, SM_MAXSTRLEN);
440	if (n == SM_MAXSTRLEN) {
441		return NULL;
442	}
443
444	newfl = malloc(sizeof(*newfl) - sizeof(newfl->client_name) + n + 1);
445	if (newfl == NULL) {
446		return NULL;
447	}
448	bzero(newfl, sizeof(*newfl) - sizeof(newfl->client_name));
449	memcpy(newfl->client_name, caller_name, n);
450	newfl->client_name[n] = 0;
451
452	newfl->client.oh.n_bytes = malloc(lockowner->n_len);
453	if (newfl->client.oh.n_bytes == NULL) {
454		free(newfl);
455		return NULL;
456	}
457	newfl->client.oh.n_len = lockowner->n_len;
458	bcopy(lockowner->n_bytes, newfl->client.oh.n_bytes, lockowner->n_len);
459
460	newfl->filehandle.n_bytes = malloc(filehandle->n_len);
461	if (newfl->filehandle.n_bytes == NULL) {
462		free(newfl->client.oh.n_bytes);
463		free(newfl);
464		return NULL;
465	}
466	newfl->filehandle.n_len = filehandle->n_len;
467	bcopy(filehandle->n_bytes, newfl->filehandle.n_bytes, filehandle->n_len);
468
469	newfl->addr = malloc(addr->sa_len);
470	if (newfl->addr == NULL) {
471		free(newfl->client.oh.n_bytes);
472		free(newfl);
473		return NULL;
474	}
475	memcpy(newfl->addr, addr, addr->sa_len);
476
477	return newfl;
478}
479
480/*
481 * file_file_lock: Force creation of a valid file lock
482 */
483void
484fill_file_lock(struct file_lock *fl,
485    const bool_t exclusive, const int32_t svid,
486    const u_int64_t offset, const u_int64_t len,
487    const int state, const int status, const int flags, const int blocking)
488{
489	fl->client.exclusive = exclusive;
490	fl->client.svid = svid;
491	fl->client.l_offset = offset;
492	fl->client.l_len = len;
493
494	fl->nsm_status = state;
495	fl->status = status;
496	fl->flags = flags;
497	fl->blocking = blocking;
498}
499
500/*
501 * deallocate_file_lock: Free all storage associated with a file lock
502 */
503void
504deallocate_file_lock(struct file_lock *fl)
505{
506	free(fl->addr);
507	free(fl->client.oh.n_bytes);
508	free(fl->filehandle.n_bytes);
509	free(fl);
510}
511
512/*
513 * regions_overlap(): This function examines the two provided regions for
514 * overlap.
515 */
516int
517regions_overlap(start1, len1, start2, len2)
518	const u_int64_t start1, len1, start2, len2;
519{
520	u_int64_t d1,d2,d3,d4;
521	enum split_status result;
522
523	debuglog("Entering region overlap with vals: %llu:%llu--%llu:%llu\n",
524		 start1, len1, start2, len2);
525
526	result = region_compare(start1, len1, start2, len2,
527	    &d1, &d2, &d3, &d4);
528
529	debuglog("Exiting region overlap with val: %d\n",result);
530
531	if (result == SPL_DISJOINT) {
532		return 0;
533	} else {
534		return 1;
535	}
536
537	return (result);
538}
539
540/*
541 * region_compare(): Examine lock regions and split appropriately
542 *
543 * XXX: Fix 64 bit overflow problems
544 * XXX: Check to make sure I got *ALL* the cases.
545 * XXX: This DESPERATELY needs a regression test.
546 */
547enum split_status
548region_compare(starte, lene, startu, lenu,
549    start1, len1, start2, len2)
550	const u_int64_t starte, lene, startu, lenu;
551	u_int64_t *start1, *len1, *start2, *len2;
552{
553	/*
554	 * Please pay attention to the sequential exclusions
555	 * of the if statements!!!
556	 */
557	enum LFLAGS lflags = LEDGE_LEFT;
558	enum RFLAGS rflags;
559	enum split_status retval;
560
561	retval = SPL_DISJOINT;
562
563	if (lene == 0 && lenu == 0) {
564		/* Examine left edge of locker */
565		if (startu < starte) {
566			lflags = LEDGE_LEFT;
567		} else if (startu == starte) {
568			lflags = LEDGE_LBOUNDARY;
569		} else {
570			lflags = LEDGE_INSIDE;
571		}
572
573		// rflags = REDGE_RBOUNDARY; /* Both are infiinite */
574
575		if (lflags == LEDGE_INSIDE) {
576			*start1 = starte;
577			*len1 = startu - starte;
578		}
579
580		if (lflags == LEDGE_LEFT || lflags == LEDGE_LBOUNDARY) {
581			retval = SPL_CONTAINED;
582		} else {
583			retval = SPL_LOCK1;
584		}
585	} else if (lene == 0 && lenu != 0) {
586		/* Established lock is infinite */
587		/* Examine left edge of unlocker */
588		if (startu < starte) {
589			lflags = LEDGE_LEFT;
590		} else if (startu == starte) {
591			lflags = LEDGE_LBOUNDARY;
592		} else if (startu > starte) {
593			lflags = LEDGE_INSIDE;
594		}
595
596		/* Examine right edge of unlocker */
597		if (startu + lenu < starte) {
598			/* Right edge of unlocker left of established lock */
599			// rflags = REDGE_LEFT;
600			return SPL_DISJOINT;
601		} else if (startu + lenu == starte) {
602			/* Right edge of unlocker on start of established lock */
603			// rflags = REDGE_LBOUNDARY;
604			return SPL_DISJOINT;
605		} else { /* Infinifty is right of finity */
606			/* Right edge of unlocker inside established lock */
607			rflags = REDGE_INSIDE;
608		}
609
610		if (lflags == LEDGE_INSIDE) {
611			*start1 = starte;
612			*len1 = startu - starte;
613			retval |= SPL_LOCK1;
614		}
615
616		if (rflags == REDGE_INSIDE) {
617			/* Create right lock */
618			*start2 = startu+lenu;
619			*len2 = 0;
620			retval |= SPL_LOCK2;
621		}
622	} else if (lene != 0 && lenu == 0) {
623		/* Unlocker is infinite */
624		/* Examine left edge of unlocker */
625		if (startu < starte) {
626			// lflags = LEDGE_LEFT;
627			retval = SPL_CONTAINED;
628			return retval;
629		} else if (startu == starte) {
630			// lflags = LEDGE_LBOUNDARY;
631			retval = SPL_CONTAINED;
632			return retval;
633		} else if ((startu > starte) && (startu < starte + lene - 1)) {
634			lflags = LEDGE_INSIDE;
635		} else if (startu == starte + lene - 1) {
636			lflags = LEDGE_RBOUNDARY;
637		} else { /* startu > starte + lene -1 */
638			// lflags = LEDGE_RIGHT;
639			return SPL_DISJOINT;
640		}
641
642		// rflags = REDGE_RIGHT; /* Infinity is right of finity */
643
644		if (lflags == LEDGE_INSIDE || lflags == LEDGE_RBOUNDARY) {
645			*start1 = starte;
646			*len1 = startu - starte;
647			retval |= SPL_LOCK1;
648			return retval;
649		}
650
651	} else {
652		/* Both locks are finite */
653
654		/* Examine left edge of unlocker */
655		if (startu < starte) {
656			lflags = LEDGE_LEFT;
657		} else if (startu == starte) {
658			lflags = LEDGE_LBOUNDARY;
659		} else if ((startu > starte) && (startu < starte + lene - 1)) {
660			lflags = LEDGE_INSIDE;
661		} else if (startu == starte + lene - 1) {
662			lflags = LEDGE_RBOUNDARY;
663		} else { /* startu > starte + lene -1 */
664			// lflags = LEDGE_RIGHT;
665			return SPL_DISJOINT;
666		}
667
668		/* Examine right edge of unlocker */
669		if (startu + lenu < starte) {
670			/* Right edge of unlocker left of established lock */
671			// rflags = REDGE_LEFT;
672			return SPL_DISJOINT;
673		} else if (startu + lenu == starte) {
674			/* Right edge of unlocker on start of established lock */
675			// rflags = REDGE_LBOUNDARY;
676			return SPL_DISJOINT;
677		} else if (startu + lenu < starte + lene) {
678			/* Right edge of unlocker inside established lock */
679			rflags = REDGE_INSIDE;
680		} else if (startu + lenu == starte + lene) {
681			/* Right edge of unlocker on right edge of established lock */
682			rflags = REDGE_RBOUNDARY;
683		} else { /* startu + lenu > starte + lene */
684			/* Right edge of unlocker is right of established lock */
685			rflags = REDGE_RIGHT;
686		}
687
688		if (lflags == LEDGE_INSIDE || lflags == LEDGE_RBOUNDARY) {
689			/* Create left lock */
690			*start1 = starte;
691			*len1 = (startu - starte);
692			retval |= SPL_LOCK1;
693		}
694
695		if (rflags == REDGE_INSIDE) {
696			/* Create right lock */
697			*start2 = startu+lenu;
698			*len2 = starte+lene-(startu+lenu);
699			retval |= SPL_LOCK2;
700		}
701
702		if ((lflags == LEDGE_LEFT || lflags == LEDGE_LBOUNDARY) &&
703		    (rflags == REDGE_RBOUNDARY || rflags == REDGE_RIGHT)) {
704			retval = SPL_CONTAINED;
705		}
706	}
707
708	return retval;
709}
710
711/*
712 * same_netobj: Compares the apprpriate bits of a netobj for identity
713 */
714int
715same_netobj(const netobj *n0, const netobj *n1)
716{
717	int retval;
718
719	retval = 0;
720
721	debuglog("Entering netobj identity check\n");
722
723	if (n0->n_len == n1->n_len) {
724		debuglog("Preliminary length check passed\n");
725		retval = !bcmp(n0->n_bytes, n1->n_bytes, n0->n_len);
726		debuglog("netobj %smatch\n", retval ? "" : "mis");
727	}
728
729	return (retval);
730}
731
732/*
733 * same_filelock_identity: Compares the appropriate bits of a file_lock
734 */
735int
736same_filelock_identity(fl0, fl1)
737	const struct file_lock *fl0, *fl1;
738{
739	int retval;
740
741	debuglog("Checking filelock identity\n");
742
743	/*
744	 * Check process ids and host information.
745	 */
746	retval = (fl0->client.svid == fl1->client.svid &&
747	    same_netobj(&(fl0->client.oh), &(fl1->client.oh)));
748
749	debuglog("Exiting checking filelock identity: retval: %d\n",retval);
750
751	return (retval);
752}
753
754/*
755 * Below here are routines associated with manipulating the NFS
756 * lock list.
757 */
758
759/*
760 * get_lock_matching_unlock: Return a lock which matches the given unlock lock
761 *                           or NULL otherwise
762 * XXX: It is a shame that this duplicates so much code from test_nfslock.
763 */
764struct file_lock *
765get_lock_matching_unlock(const struct file_lock *fl, int cleanup)
766{
767	struct file_lock *ifl; /* Iterator */
768
769	debuglog("Entering lock_matching_unlock\n");
770	debuglog("********Dump of fl*****************\n");
771	dump_filelock(fl);
772
773	LIST_FOREACH(ifl, &nfslocklist_head, nfslocklist) {
774		debuglog("Pointer to file lock: %p\n",ifl);
775
776		debuglog("****Dump of ifl****\n");
777		dump_filelock(ifl);
778		debuglog("*******************\n");
779
780		if (cleanup && (ifl == fl))  /* don't match the lock we're cleaning up under */
781			continue;
782
783		/*
784		 * XXX: It is conceivable that someone could use the NLM RPC
785		 * system to directly access filehandles.  This may be a
786		 * security hazard as the filehandle code may bypass normal
787		 * file access controls
788		 */
789		if (fl->filehandle.n_len != ifl->filehandle.n_len)
790			continue;
791		if (bcmp(fl->filehandle.n_bytes, ifl->filehandle.n_bytes,
792			fl->filehandle.n_len))
793			continue;
794
795		debuglog("matching_unlock: Filehandles match, "
796		    "checking regions\n");
797
798		/* Filehandles match, check for region overlap */
799		if (!regions_overlap(fl->client.l_offset, fl->client.l_len,
800			ifl->client.l_offset, ifl->client.l_len))
801			continue;
802
803		debuglog("matching_unlock: Region overlap"
804		    " found %llu : %llu -- %llu : %llu\n",
805		    fl->client.l_offset,fl->client.l_len,
806		    ifl->client.l_offset,ifl->client.l_len);
807
808		/* Regions overlap, check the identity */
809		if (!same_filelock_identity(fl,ifl))
810			continue;
811
812		debuglog("matching_unlock: Duplicate lock id.  Granting\n");
813		return (ifl);
814	}
815
816	debuglog("Exiting lock_matching_unlock\n");
817
818	return (NULL);
819}
820
821/*
822 * test_nfslock: check for NFS lock in lock list
823 *
824 * This routine makes the following assumptions:
825 *    1) Nothing will adjust the lock list during a lookup
826 *
827 * This routine has an intersting quirk which bit me hard.
828 * The conflicting_fl is the pointer to the conflicting lock.
829 * However, to modify the "*pointer* to the conflicting lock" rather
830 * that the "conflicting lock itself" one must pass in a "pointer to
831 * the pointer of the conflicting lock".  Gross.
832 */
833
834enum nfslock_status
835test_nfslock(const struct file_lock *fl, struct file_lock **conflicting_fl)
836{
837	struct file_lock *ifl; /* Iterator */
838	enum nfslock_status retval;
839
840	debuglog("Entering test_nfslock\n");
841
842	retval = NFS_GRANTED;
843	(*conflicting_fl) = NULL;
844
845	debuglog("Entering lock search loop\n");
846
847	debuglog("***********************************\n");
848	debuglog("Dumping match filelock\n");
849	debuglog("***********************************\n");
850	dump_filelock(fl);
851	debuglog("***********************************\n");
852
853	LIST_FOREACH(ifl, &nfslocklist_head, nfslocklist) {
854		if (retval == NFS_DENIED)
855			break;
856
857		debuglog("Top of lock loop\n");
858		debuglog("Pointer to file lock: %p\n",ifl);
859
860		debuglog("***********************************\n");
861		debuglog("Dumping test filelock\n");
862		debuglog("***********************************\n");
863		dump_filelock(ifl);
864		debuglog("***********************************\n");
865
866		/*
867		 * XXX: It is conceivable that someone could use the NLM RPC
868		 * system to directly access filehandles.  This may be a
869		 * security hazard as the filehandle code may bypass normal
870		 * file access controls
871		 */
872		if (fl->filehandle.n_len != ifl->filehandle.n_len)
873			continue;
874		if (bcmp(fl->filehandle.n_bytes, ifl->filehandle.n_bytes,
875			fl->filehandle.n_len))
876			continue;
877
878		debuglog("test_nfslock: filehandle match found\n");
879
880		/* Filehandles match, check for region overlap */
881		if (!regions_overlap(fl->client.l_offset, fl->client.l_len,
882			ifl->client.l_offset, ifl->client.l_len))
883			continue;
884
885		debuglog("test_nfslock: Region overlap found"
886		    " %llu : %llu -- %llu : %llu\n",
887		    fl->client.l_offset,fl->client.l_len,
888		    ifl->client.l_offset,ifl->client.l_len);
889
890		/* Regions overlap, check the exclusivity */
891		if (!(fl->client.exclusive || ifl->client.exclusive))
892			continue;
893
894		debuglog("test_nfslock: Exclusivity failure: %d %d\n",
895		    fl->client.exclusive,
896		    ifl->client.exclusive);
897
898		if (same_filelock_identity(fl,ifl)) {
899			debuglog("test_nfslock: Duplicate id.  Granting\n");
900			(*conflicting_fl) = ifl;
901			retval = NFS_GRANTED_DUPLICATE;
902		} else {
903			/* locking attempt fails */
904			debuglog("test_nfslock: Lock attempt failed\n");
905			debuglog("Desired lock\n");
906			dump_filelock(fl);
907			debuglog("Conflicting lock\n");
908			dump_filelock(ifl);
909			(*conflicting_fl) = ifl;
910			retval = NFS_DENIED;
911		}
912	}
913
914	debuglog("Dumping file locks\n");
915	debuglog("Exiting test_nfslock\n");
916
917	return (retval);
918}
919
920/*
921 * lock_nfslock: attempt to create a lock in the NFS lock list
922 *
923 * This routine tests whether the lock will be granted and then adds
924 * the entry to the lock list if so.
925 *
926 * Argument fl gets modified as its list housekeeping entries get modified
927 * upon insertion into the NFS lock list
928 *
929 * This routine makes several assumptions:
930 *    1) It is perfectly happy to grant a duplicate lock from the same pid.
931 *       While this seems to be intuitively wrong, it is required for proper
932 *       Posix semantics during unlock.  It is absolutely imperative to not
933 *       unlock the main lock before the two child locks are established. Thus,
934 *       one has be be able to create duplicate locks over an existing lock
935 *    2) It currently accepts duplicate locks from the same id,pid
936 */
937
938enum nfslock_status
939lock_nfslock(struct file_lock *fl)
940{
941	enum nfslock_status retval;
942	struct file_lock *dummy_fl;
943
944	dummy_fl = NULL;
945
946	debuglog("Entering lock_nfslock...\n");
947
948	retval = test_nfslock(fl,&dummy_fl);
949
950	if (retval == NFS_GRANTED || retval == NFS_GRANTED_DUPLICATE) {
951		debuglog("Inserting lock...\n");
952		dump_filelock(fl);
953		LIST_INSERT_HEAD(&nfslocklist_head, fl, nfslocklist);
954	}
955
956	debuglog("Exiting lock_nfslock...\n");
957
958	return (retval);
959}
960
961/*
962 * delete_nfslock: delete an NFS lock list entry
963 *
964 * This routine is used to delete a lock out of the NFS lock list
965 * without regard to status, underlying locks, regions or anything else
966 *
967 * Note that this routine *does not deallocate memory* of the lock.
968 * It just disconnects it from the list.  The lock can then be used
969 * by other routines without fear of trashing the list.
970 */
971
972enum nfslock_status
973delete_nfslock(struct file_lock *fl)
974{
975
976	LIST_REMOVE(fl, nfslocklist);
977
978	return (NFS_GRANTED);
979}
980
981enum split_status
982split_nfslock(exist_lock, unlock_lock, left_lock, right_lock)
983	const struct file_lock *exist_lock, *unlock_lock;
984	struct file_lock **left_lock, **right_lock;
985{
986	u_int64_t start1, len1, start2, len2;
987	enum split_status spstatus;
988
989	spstatus = region_compare(exist_lock->client.l_offset, exist_lock->client.l_len,
990	    unlock_lock->client.l_offset, unlock_lock->client.l_len,
991	    &start1, &len1, &start2, &len2);
992
993	if ((spstatus & SPL_LOCK1) != 0) {
994		*left_lock = allocate_file_lock(&exist_lock->client.oh, &exist_lock->filehandle, exist_lock->addr, exist_lock->client_name);
995		if (*left_lock == NULL) {
996			debuglog("Unable to allocate resource for split 1\n");
997			return SPL_RESERR;
998		}
999
1000		fill_file_lock(*left_lock,
1001		    exist_lock->client.exclusive, exist_lock->client.svid,
1002		    start1, len1,
1003		    exist_lock->nsm_status,
1004		    exist_lock->status, exist_lock->flags, exist_lock->blocking);
1005	}
1006
1007	if ((spstatus & SPL_LOCK2) != 0) {
1008		*right_lock = allocate_file_lock(&exist_lock->client.oh, &exist_lock->filehandle, exist_lock->addr, exist_lock->client_name);
1009		if (*right_lock == NULL) {
1010			debuglog("Unable to allocate resource for split 1\n");
1011			if (*left_lock != NULL) {
1012				deallocate_file_lock(*left_lock);
1013			}
1014			return SPL_RESERR;
1015		}
1016
1017		fill_file_lock(*right_lock,
1018		    exist_lock->client.exclusive, exist_lock->client.svid,
1019		    start2, len2,
1020		    exist_lock->nsm_status,
1021		    exist_lock->status, exist_lock->flags, exist_lock->blocking);
1022	}
1023
1024	return spstatus;
1025}
1026
1027enum nfslock_status
1028unlock_nfslock(fl, released_lock, left_lock, right_lock, cleanup)
1029	const struct file_lock *fl;
1030	struct file_lock **released_lock;
1031	struct file_lock **left_lock;
1032	struct file_lock **right_lock;
1033	int cleanup;
1034{
1035	struct file_lock *mfl; /* Matching file lock */
1036	enum nfslock_status retval;
1037	enum split_status spstatus;
1038
1039	debuglog("Entering unlock_nfslock\n");
1040
1041	*released_lock = NULL;
1042	*left_lock = NULL;
1043	*right_lock = NULL;
1044
1045	retval = NFS_DENIED_NOLOCK;
1046
1047	debuglog("Attempting to match lock...\n");
1048	mfl = get_lock_matching_unlock(fl, cleanup);
1049
1050	if (mfl != NULL) {
1051		debuglog("Unlock matched.  Querying for split\n");
1052
1053		spstatus = split_nfslock(mfl, fl, left_lock, right_lock);
1054
1055		debuglog("Split returned %d %p %p %p %p\n",spstatus,mfl,fl,*left_lock,*right_lock);
1056		debuglog("********Split dumps********");
1057		dump_filelock(mfl);
1058		dump_filelock(fl);
1059		dump_filelock(*left_lock);
1060		dump_filelock(*right_lock);
1061		debuglog("********End Split dumps********");
1062
1063		if (spstatus == SPL_RESERR) {
1064			if (*left_lock != NULL) {
1065				deallocate_file_lock(*left_lock);
1066				*left_lock = NULL;
1067			}
1068
1069			if (*right_lock != NULL) {
1070				deallocate_file_lock(*right_lock);
1071				*right_lock = NULL;
1072			}
1073
1074			return NFS_RESERR;
1075		}
1076
1077		/* Insert new locks from split if required */
1078		if (*left_lock != NULL) {
1079			debuglog("Split left activated\n");
1080			LIST_INSERT_HEAD(&nfslocklist_head, *left_lock, nfslocklist);
1081		}
1082
1083		if (*right_lock != NULL) {
1084			debuglog("Split right activated\n");
1085			LIST_INSERT_HEAD(&nfslocklist_head, *right_lock, nfslocklist);
1086		}
1087
1088		/* Unlock the lock since it matches identity */
1089		LIST_REMOVE(mfl, nfslocklist);
1090		*released_lock = mfl;
1091		retval = NFS_GRANTED;
1092	}
1093
1094	debuglog("Exiting unlock_nfslock\n");
1095
1096	return retval;
1097}
1098
1099/*
1100 * Below here are the routines for manipulating the file lock directly
1101 * on the disk hardware itself
1102 */
1103enum hwlock_status
1104lock_hwlock(struct file_lock *fl)
1105{
1106	struct monfile *imf,*nmf;
1107	int lflags, flerror;
1108	fhandle_t fh;
1109
1110	/* Scan to see if filehandle already present */
1111	LIST_FOREACH(imf, &monfilelist_head, monfilelist) {
1112		if ((fl->filehandle.n_len == imf->filehandle.n_len) &&
1113		    (bcmp(fl->filehandle.n_bytes, imf->filehandle.n_bytes,
1114			fl->filehandle.n_len) == 0)) {
1115			/* imf is the correct filehandle */
1116			break;
1117		}
1118	}
1119
1120	/*
1121	 * Filehandle already exists (we control the file)
1122	 * *AND* NFS has already cleared the lock for availability
1123	 * Grant it and bump the refcount.
1124	 */
1125	if (imf != NULL) {
1126		++(imf->refcount);
1127		return (HW_GRANTED);
1128	}
1129
1130	/* No filehandle found, create and go */
1131	nmf = malloc(sizeof(struct monfile));
1132	if (nmf == NULL) {
1133		debuglog("hwlock resource allocation failure\n");
1134		return (HW_RESERR);
1135	}
1136	nmf->filehandle.n_bytes = malloc(fl->filehandle.n_len);
1137	if (nmf->filehandle.n_bytes == NULL) {
1138		debuglog("hwlock resource allocation failure\n");
1139		free(nmf);
1140		return (HW_RESERR);
1141	}
1142
1143	if (fl->filehandle.n_len > NFSV3_MAX_FH_SIZE) {
1144		debuglog("hwlock: bad fh length %d (from %16s): %32s\n",
1145		    fl->filehandle.n_len, fl->client_name, strerror(errno));
1146		free(nmf->filehandle.n_bytes);
1147		free(nmf);
1148		return (HW_STALEFH);
1149	}
1150	fh.fh_len = fl->filehandle.n_len;
1151	bcopy(fl->filehandle.n_bytes, fh.fh_data, fh.fh_len);
1152
1153	/* O_RDWR may not work if file system is read-only */
1154	nmf->fd = fhopen(&fh, O_RDWR);
1155	if ((nmf->fd < 0) && (errno == EROFS) && !fl->client.exclusive)
1156		nmf->fd = fhopen(&fh, O_RDONLY);
1157	if (nmf->fd < 0) {
1158		debuglog("fhopen failed (from %16s): %32s\n",
1159		    fl->client_name, strerror(errno));
1160		free(nmf->filehandle.n_bytes);
1161		free(nmf);
1162		switch (errno) {
1163		case ESTALE:
1164			return (HW_STALEFH);
1165		case EROFS:
1166			return (HW_READONLY);
1167		default:
1168			return (HW_RESERR);
1169		}
1170	}
1171
1172	/* File opened correctly, fill the monitor struct */
1173	nmf->filehandle.n_len = fl->filehandle.n_len;
1174	bcopy(fl->filehandle.n_bytes, nmf->filehandle.n_bytes, fl->filehandle.n_len);
1175	nmf->refcount = 1;
1176	nmf->exclusive = fl->client.exclusive;
1177
1178	lflags = (nmf->exclusive == 1) ?
1179	    (LOCK_EX | LOCK_NB) : (LOCK_SH | LOCK_NB);
1180
1181	flerror = flock(nmf->fd, lflags);
1182
1183	if (flerror != 0) {
1184		debuglog("flock failed (from %16s): %32s\n",
1185		    fl->client_name, strerror(errno));
1186		close(nmf->fd);
1187		free(nmf->filehandle.n_bytes);
1188		free(nmf);
1189		switch (errno) {
1190		case EAGAIN:
1191			return (HW_DENIED);
1192		case ESTALE:
1193			return (HW_STALEFH);
1194		case EROFS:
1195			return (HW_READONLY);
1196		default:
1197			return (HW_RESERR);
1198			break;
1199		}
1200	}
1201
1202	/* File opened and locked */
1203	LIST_INSERT_HEAD(&monfilelist_head, nmf, monfilelist);
1204
1205	debuglog("flock succeeded (from %16s)\n", fl->client_name);
1206	return (HW_GRANTED);
1207}
1208
1209enum hwlock_status
1210unlock_hwlock(const struct file_lock *fl)
1211{
1212	struct monfile *imf;
1213
1214	debuglog("Entering unlock_hwlock\n");
1215	debuglog("Entering loop interation\n");
1216
1217	/* Scan to see if filehandle already present */
1218	LIST_FOREACH(imf, &monfilelist_head, monfilelist) {
1219		if ((fl->filehandle.n_len == imf->filehandle.n_len) &&
1220		    (bcmp(fl->filehandle.n_bytes, imf->filehandle.n_bytes,
1221			fl->filehandle.n_len) == 0)) {
1222			/* imf is the correct filehandle */
1223			break;
1224		}
1225	}
1226
1227	debuglog("Completed iteration.  Proceeding\n");
1228
1229	if (imf == NULL) {
1230		/* No lock found */
1231		debuglog("Exiting unlock_hwlock (HW_DENIED_NOLOCK)\n");
1232		return (HW_DENIED_NOLOCK);
1233	}
1234
1235	/* Lock found */
1236	--imf->refcount;
1237
1238	if (imf->refcount < 0) {
1239		debuglog("Negative hardware reference count\n");
1240	}
1241
1242	if (imf->refcount <= 0) {
1243		close(imf->fd);
1244		LIST_REMOVE(imf, monfilelist);
1245		free(imf->filehandle.n_bytes);
1246		free(imf);
1247	}
1248	debuglog("Exiting unlock_hwlock (HW_GRANTED)\n");
1249	return (HW_GRANTED);
1250}
1251
1252enum hwlock_status
1253test_hwlock(fl, conflicting_fl)
1254	const struct file_lock *fl __unused;
1255	struct file_lock **conflicting_fl __unused;
1256{
1257
1258	/*
1259	 * XXX: lock tests on hardware are not required until
1260	 * true partial file testing is done on the underlying file
1261	 */
1262	return (HW_RESERR);
1263}
1264
1265
1266
1267/*
1268 * Below here are routines for manipulating blocked lock requests
1269 * They should only be called from the XXX_partialfilelock routines
1270 * if at all possible
1271 */
1272
1273void
1274add_blockingfilelock(struct file_lock *fl)
1275{
1276	struct file_lock *ifl, *nfl;
1277
1278	debuglog("Entering add_blockingfilelock\n");
1279
1280	/*
1281	 * Check for a duplicate lock request.
1282	 * If found, deallocate the older request.
1283	 */
1284	ifl = LIST_FIRST(&blockedlocklist_head);
1285	for (; ifl != NULL; ifl = nfl) {
1286		debuglog("Pointer to file lock: %p\n",ifl);
1287		debuglog("****Dump of ifl****\n");
1288		dump_filelock(ifl);
1289		debuglog("*******************\n");
1290
1291		nfl = LIST_NEXT(ifl, nfslocklist);
1292
1293		if (fl->filehandle.n_len != ifl->filehandle.n_len)
1294			continue;
1295		if (bcmp(fl->filehandle.n_bytes, ifl->filehandle.n_bytes,
1296			fl->filehandle.n_len))
1297			continue;
1298
1299		/* Filehandles match, check region */
1300		if ((fl->client.l_offset != ifl->client.l_offset) ||
1301		    (fl->client.l_len != ifl->client.l_len))
1302			continue;
1303
1304		/* Regions match, check the identity */
1305		if (!same_filelock_identity(fl,ifl))
1306			continue;
1307
1308		debuglog("add_blockingfilelock: removing duplicate lock request.\n");
1309		remove_blockingfilelock(ifl);
1310		deallocate_file_lock(ifl);
1311		break;
1312	}
1313
1314	/*
1315	 * Clear the blocking flag so that it can be reused without
1316	 * adding it to the blocking queue a second time
1317	 */
1318
1319	fl->blocking = 0;
1320	LIST_INSERT_HEAD(&blockedlocklist_head, fl, nfslocklist);
1321
1322	debuglog("Exiting add_blockingfilelock\n");
1323}
1324
1325void
1326remove_blockingfilelock(struct file_lock *fl)
1327{
1328
1329	debuglog("Entering remove_blockingfilelock\n");
1330
1331	LIST_REMOVE(fl, nfslocklist);
1332
1333	debuglog("Exiting remove_blockingfilelock\n");
1334}
1335
1336void
1337clear_blockingfilelock(const char *hostname)
1338{
1339	struct file_lock *ifl,*nfl;
1340
1341	/*
1342	 * Normally, LIST_FOREACH is called for, but since
1343	 * the current element *is* the iterator, deleting it
1344	 * would mess up the iteration.  Thus, a next element
1345	 * must be used explicitly
1346	 */
1347
1348	ifl = LIST_FIRST(&blockedlocklist_head);
1349
1350	while (ifl != NULL) {
1351		nfl = LIST_NEXT(ifl, nfslocklist);
1352
1353		if (strncmp(hostname, ifl->client_name, SM_MAXSTRLEN) == 0) {
1354			remove_blockingfilelock(ifl);
1355			deallocate_file_lock(ifl);
1356		}
1357
1358		ifl = nfl;
1359	}
1360}
1361
1362int need_retry_blocked_locks = 0; /* need to call retry_blockingfilelocklist() */
1363
1364void
1365retry_blockingfilelocklist(netobj *fh)
1366{
1367	/*
1368	 * If fh is given, then retry just the locks with the
1369	 * same filehandle in the blocked list.
1370	 * Otherwise, simply retry all locks in the blocked list.
1371	 */
1372	struct file_lock *ifl, *nfl, *pfl; /* Iterator */
1373	enum partialfilelock_status pflstatus;
1374	int rv;
1375
1376	debuglog("Entering retry_blockingfilelocklist\n");
1377
1378	need_retry_blocked_locks = 0;
1379
1380	pfl = NULL;
1381	ifl = LIST_FIRST(&blockedlocklist_head);
1382	debuglog("Iterator choice %p\n",ifl);
1383
1384	while (ifl != NULL) {
1385		/*
1386		 * SUBTLE BUG: The next element must be worked out before the
1387		 * current element has been moved
1388		 */
1389		nfl = LIST_NEXT(ifl, nfslocklist);
1390		debuglog("Iterator choice %p\n",ifl);
1391		debuglog("Prev iterator choice %p\n",pfl);
1392		debuglog("Next iterator choice %p\n",nfl);
1393
1394		/* if given a filehandle, only retry locks for the same filehandle */
1395		if (fh && !same_netobj(fh, &ifl->filehandle)) {
1396			ifl = nfl;
1397			continue;
1398		}
1399
1400		/*
1401		 * SUBTLE BUG: The file_lock must be removed from the
1402		 * old list so that it's list pointers get disconnected
1403		 * before being allowed to participate in the new list
1404		 * which will automatically add it in if necessary.
1405		 */
1406
1407		LIST_REMOVE(ifl, nfslocklist);
1408		pflstatus = lock_partialfilelock(ifl);
1409
1410		if (pflstatus == PFL_GRANTED || pflstatus == PFL_GRANTED_DUPLICATE) {
1411			debuglog("Granted blocked lock\n");
1412			/* lock granted and is now being used */
1413			rv = send_granted(ifl, 0);
1414			if (rv) {
1415				/*
1416				 * Uh oh... the NLM_GRANTED message failed.
1417				 * About the only thing we can do is drop the lock.
1418				 * Note: this could be bad if the error was only
1419				 * transient.  Hopefully, if the client is still
1420				 * waiting for the lock, they will resend the request.
1421				 */
1422				do_unlock(ifl);
1423				/* ifl is NO LONGER VALID AT THIS POINT */
1424			}
1425		} else if (pflstatus == PFL_HWDENIED_STALEFH) {
1426			/*
1427			 * Uh oh...
1428			 * It would be nice if we could inform the client of
1429			 * this error.  Unfortunately, there's no way to do
1430			 * that in the NLM protocol (can't send "granted"
1431			 * message with an error and there's no "never going
1432			 * to be granted" message).
1433			 *
1434			 * Since there's no chance of this blocked request ever
1435			 * succeeding, we drop the lock request rather than
1436			 * needlessly keeping it around just to rot forever in
1437			 * the blocked lock list.
1438			 *
1439			 * Hopefully, if the client is still waiting for the lock,
1440			 * they will resend the request (and get an error then).
1441			 *
1442			 * XXX Note: PFL_HWDENIED_READONLY could potentially
1443			 * be handled this way as well, although that would
1444			 * only be an issue if a file system changed from
1445			 * read-write to read-only out from under a blocked
1446			 * lock request, and that's far less likely than a
1447			 * file disappearing out from under such a request.
1448			 */
1449			deallocate_file_lock(ifl);
1450			/* ifl is NO LONGER VALID AT THIS POINT */
1451		} else {
1452			/* Reinsert lock back into same place in blocked list */
1453			debuglog("Replacing blocked lock\n");
1454			if (pfl != NULL)
1455				LIST_INSERT_AFTER(pfl, ifl, nfslocklist);
1456			else
1457				/* ifl is the only elem. in the list */
1458				LIST_INSERT_HEAD(&blockedlocklist_head, ifl, nfslocklist);
1459		}
1460
1461		if (pflstatus == PFL_GRANTED || pflstatus == PFL_GRANTED_DUPLICATE ||
1462		    pflstatus == PFL_HWDENIED_STALEFH) {
1463			/* If ifl was permanently removed from the list, (e.g it */
1464			/* was granted or dropped), pfl should remain where it's at. */
1465		} else {
1466			/* If ifl was left in the list, (e.g it was reinserted back */
1467			/* in place), pfl should simply be moved forward to be ifl */
1468			pfl = ifl;
1469		}
1470		/* Valid increment behavior regardless of state of ifl */
1471		ifl = nfl;
1472	}
1473
1474	debuglog("Exiting retry_blockingfilelocklist\n");
1475}
1476
1477/*
1478 * Below here are routines associated with manipulating all
1479 * aspects of the partial file locking system (list, hardware, etc.)
1480 */
1481
1482/*
1483 * Please note that lock monitoring must be done at this level which
1484 * keeps track of *individual* lock requests on lock and unlock
1485 *
1486 * XXX: Split unlocking is going to make the unlock code miserable
1487 */
1488
1489/*
1490 * lock_partialfilelock:
1491 *
1492 * Argument fl gets modified as its list housekeeping entries get modified
1493 * upon insertion into the NFS lock list
1494 *
1495 * This routine makes several assumptions:
1496 * 1) It (will) pass locks through to flock to lock the entire underlying file
1497 *     and then parcel out NFS locks if it gets control of the file.
1498 *         This matches the old rpc.lockd file semantics (except where it
1499 *         is now more correct).  It is the safe solution, but will cause
1500 *         overly restrictive blocking if someone is trying to use the
1501 *         underlying files without using NFS.  This appears to be an
1502 *         acceptable tradeoff since most people use standalone NFS servers.
1503 * XXX: The right solution is probably kevent combined with fcntl
1504 *
1505 *    2) Nothing modifies the lock lists between testing and granting
1506 *           I have no idea whether this is a useful assumption or not
1507 */
1508
1509enum partialfilelock_status
1510lock_partialfilelock(struct file_lock *fl)
1511{
1512	enum partialfilelock_status retval;
1513	enum nfslock_status lnlstatus;
1514	enum hwlock_status hwstatus;
1515
1516	debuglog("Entering lock_partialfilelock\n");
1517
1518	retval = PFL_DENIED;
1519
1520	/*
1521	 * Execute the NFS lock first, if possible, as it is significantly
1522	 * easier and less expensive to undo than the filesystem lock
1523	 */
1524
1525	lnlstatus = lock_nfslock(fl);
1526
1527	switch (lnlstatus) {
1528	case NFS_GRANTED:
1529	case NFS_GRANTED_DUPLICATE:
1530		/*
1531		 * At this point, the NFS lock is allocated and active.
1532		 * Remember to clean it up if the hardware lock fails
1533		 */
1534		hwstatus = lock_hwlock(fl);
1535
1536		switch (hwstatus) {
1537		case HW_GRANTED:
1538		case HW_GRANTED_DUPLICATE:
1539			debuglog("HW GRANTED\n");
1540			/*
1541			 * XXX: Fixme: Check hwstatus for duplicate when
1542			 * true partial file locking and accounting is
1543			 * done on the hardware
1544			 */
1545			if (lnlstatus == NFS_GRANTED_DUPLICATE) {
1546				retval = PFL_GRANTED_DUPLICATE;
1547			} else {
1548				retval = PFL_GRANTED;
1549			}
1550			if (fl->flags & LOCK_MON)
1551				monitor_lock_host_by_name(fl->client_name, fl->addr);
1552			break;
1553		case HW_RESERR:
1554			debuglog("HW RESERR\n");
1555			retval = PFL_HWRESERR;
1556			break;
1557		case HW_DENIED:
1558			debuglog("HW DENIED\n");
1559			retval = PFL_HWDENIED;
1560			break;
1561		case HW_DENIED_NOLOCK:
1562			debuglog("HW DENIED NOLOCK\n");
1563			retval = PFL_HWDENIED_NOLOCK;
1564			break;
1565		case HW_STALEFH:
1566			debuglog("HW STALE FH\n");
1567			retval = PFL_HWDENIED_STALEFH;
1568			break;
1569		case HW_READONLY:
1570			debuglog("HW READONLY\n");
1571			retval = PFL_HWDENIED_READONLY;
1572			break;
1573		default:
1574			debuglog("Unmatched hwstatus %d\n",hwstatus);
1575			break;
1576		}
1577
1578		if (retval != PFL_GRANTED &&
1579		    retval != PFL_GRANTED_DUPLICATE) {
1580			/* Clean up the NFS lock */
1581			debuglog("Deleting trial NFS lock\n");
1582			delete_nfslock(fl);
1583		}
1584		if (retval == PFL_GRANTED_DUPLICATE) {
1585			/* Clean up locks replaced by this lock */
1586			debuglog("Cleaning up replaced locks\n");
1587			unlock_partialfilelock(fl, 1);
1588		}
1589		break;
1590	case NFS_DENIED:
1591		retval = PFL_NFSDENIED;
1592		break;
1593	case NFS_RESERR:
1594		retval = PFL_NFSRESERR;
1595		break;
1596	default:
1597		debuglog("Unmatched lnlstatus %d\n");
1598		retval = PFL_NFSDENIED_NOLOCK;
1599		break;
1600	}
1601
1602	/*
1603	 * By the time fl reaches here, it is completely free again on
1604	 * failure.  The NFS lock done before attempting the
1605	 * hardware lock has been backed out
1606	 */
1607
1608	if (retval == PFL_NFSDENIED || retval == PFL_HWDENIED) {
1609		/* Once last chance to check the lock */
1610		if (fl->blocking == 1) {
1611			if (retval == PFL_NFSDENIED) {
1612				/* Queue the lock */
1613				debuglog("BLOCKING LOCK RECEIVED\n");
1614				retval = PFL_NFSBLOCKED;
1615				add_blockingfilelock(fl);
1616				dump_filelock(fl);
1617			} else {
1618				/* retval is okay as PFL_HWDENIED */
1619				debuglog("BLOCKING LOCK DENIED IN HARDWARE\n");
1620				dump_filelock(fl);
1621			}
1622		} else {
1623			/* Leave retval alone, it's already correct */
1624			debuglog("Lock denied.  Non-blocking failure\n");
1625			dump_filelock(fl);
1626		}
1627	}
1628
1629	debuglog("Exiting lock_partialfilelock\n");
1630
1631	return retval;
1632}
1633
1634/*
1635 * unlock_partialfilelock:
1636 *
1637 * Given a file_lock, unlock all locks which match.
1638 *
1639 * Note that a given lock might have to unlock ITSELF!  See
1640 * clear_partialfilelock for example.
1641 *
1642 * If cleanup is set, we will NOT unlock the lock passed in.
1643 * We will merely clean up all other locks that the given lock
1644 * has replaced.
1645 */
1646
1647enum partialfilelock_status
1648unlock_partialfilelock(const struct file_lock *fl, int cleanup)
1649{
1650	struct file_lock *lfl,*rfl,*releasedfl,*selffl;
1651	enum partialfilelock_status retval;
1652	enum nfslock_status unlstatus;
1653	enum hwlock_status unlhwstatus, lhwstatus;
1654
1655	debuglog("Entering unlock_partialfilelock\n");
1656
1657	selffl = NULL;
1658	lfl = NULL;
1659	rfl = NULL;
1660	releasedfl = NULL;
1661
1662	/*
1663	 * There are significant overlap and atomicity issues
1664	 * with partially releasing a lock.  For example, releasing
1665	 * part of an NFS shared lock does *not* always release the
1666	 * corresponding part of the file since there is only one
1667	 * rpc.lockd UID but multiple users could be requesting it
1668	 * from NFS.  Also, an unlock request should never allow
1669	 * another process to gain a lock on the remaining parts.
1670	 * ie. Always apply the new locks before releasing the
1671	 * old one
1672	 */
1673
1674	/*
1675	 * Loop is required since multiple little locks
1676	 * can be allocated and then deallocated with one
1677	 * big unlock.
1678	 *
1679	 * The loop is required to be here so that the nfs &
1680	 * hw subsystems do not need to communicate with one
1681	 * one another
1682	 */
1683
1684	do {
1685		debuglog("Value of releasedfl: %p\n",releasedfl);
1686		/* lfl&rfl are created *AND* placed into the NFS lock list if required */
1687		unlstatus = unlock_nfslock(fl, &releasedfl, &lfl, &rfl, cleanup);
1688		debuglog("Value of releasedfl: %p\n",releasedfl);
1689
1690
1691		/* XXX: This is grungy.  It should be refactored to be cleaner */
1692		if (lfl != NULL) {
1693			lhwstatus = lock_hwlock(lfl);
1694			if (lhwstatus != HW_GRANTED &&
1695			    lhwstatus != HW_GRANTED_DUPLICATE) {
1696				debuglog("HW duplicate lock failure for left split\n");
1697			}
1698			if (lfl->flags & LOCK_MON)
1699				monitor_lock_host_by_name(lfl->client_name, lfl->addr);
1700		}
1701
1702		if (rfl != NULL) {
1703			lhwstatus = lock_hwlock(rfl);
1704			if (lhwstatus != HW_GRANTED &&
1705			    lhwstatus != HW_GRANTED_DUPLICATE) {
1706				debuglog("HW duplicate lock failure for right split\n");
1707			}
1708			if (rfl->flags & LOCK_MON)
1709				monitor_lock_host_by_name(rfl->client_name, rfl->addr);
1710		}
1711
1712		switch (unlstatus) {
1713		case NFS_GRANTED:
1714			/* Attempt to unlock on the hardware */
1715			debuglog("NFS unlock granted.  Attempting hardware unlock\n");
1716
1717			/* This call *MUST NOT* unlock the two newly allocated locks */
1718			unlhwstatus = unlock_hwlock(fl);
1719			debuglog("HW unlock returned with code %d\n",unlhwstatus);
1720
1721			switch (unlhwstatus) {
1722			case HW_GRANTED:
1723				debuglog("HW unlock granted\n");
1724				if (releasedfl->flags & LOCK_MON)
1725					unmonitor_lock_host(releasedfl->client_name);
1726				retval = PFL_GRANTED;
1727				break;
1728			case HW_DENIED_NOLOCK:
1729				/* Huh?!?!  This shouldn't happen */
1730				debuglog("HW unlock denied no lock\n");
1731				retval = PFL_HWRESERR;
1732				/* Break out of do-while */
1733				unlstatus = NFS_RESERR;
1734				break;
1735			default:
1736				debuglog("HW unlock failed\n");
1737				retval = PFL_HWRESERR;
1738				/* Break out of do-while */
1739				unlstatus = NFS_RESERR;
1740				break;
1741			}
1742
1743			debuglog("Exiting with status retval: %d\n",retval);
1744
1745			// XXX sending granted messages before unlock response
1746			// XXX causes unlock response to be corrupted?
1747			// XXX Workaround is to move this to nlm_prot_svc.c
1748			// XXX after the unlock response is sent.
1749			// retry_blockingfilelocklist();
1750			need_retry_blocked_locks = 1;
1751			break;
1752		case NFS_DENIED_NOLOCK:
1753			retval = PFL_GRANTED;
1754			debuglog("All locks cleaned out\n");
1755			break;
1756		default:
1757			retval = PFL_NFSRESERR;
1758			debuglog("NFS unlock failure\n");
1759			dump_filelock(fl);
1760			break;
1761		}
1762
1763		if (releasedfl != NULL) {
1764			if (fl == releasedfl) {
1765				/*
1766				 * XXX: YECHHH!!! Attempt to unlock self succeeded
1767				 * but we can't deallocate the space yet.  This is what
1768				 * happens when you don't write malloc and free together
1769				 */
1770				if (!cleanup) {
1771					debuglog("Attempt to unlock self\n");
1772					selffl = releasedfl;
1773				}
1774			} else {
1775				/*
1776				 * XXX: this deallocation *still* needs to migrate closer
1777				 * to the allocation code way up in get_lock or the allocation
1778				 * code needs to migrate down (violation of "When you write
1779				 * malloc you must write free")
1780				 */
1781
1782				deallocate_file_lock(releasedfl);
1783			}
1784		}
1785
1786	} while (unlstatus == NFS_GRANTED);
1787
1788	if (!cleanup && (selffl != NULL)) {
1789		/*
1790		 * This statement wipes out the incoming file lock (fl)
1791		 * in spite of the fact that it is declared const
1792		 */
1793		debuglog("WARNING!  Destroying incoming lock pointer\n");
1794		deallocate_file_lock(selffl);
1795	}
1796
1797	debuglog("Exiting unlock_partialfilelock\n");
1798
1799	return retval;
1800}
1801
1802/*
1803 * clear_partialfilelock
1804 *
1805 * Normally called in response to statd state number change.
1806 * Wipe out all locks held by a host.  As a bonus, the act of
1807 * doing so should automatically clear their statd entries and
1808 * unmonitor the host.
1809 */
1810
1811void
1812clear_partialfilelock(const char *hostname)
1813{
1814	struct file_lock *ifl, *nfl;
1815	enum partialfilelock_status pfsret;
1816	struct host *ihp;
1817
1818	/*
1819	 * Check if the name we got from statd is
1820	 * actually one reverse-mapped from the client's
1821	 * address.  If so, use the name provided as
1822	 * the caller_name in lock requests instead so that
1823	 * we can correctly identify the client's locks.
1824	 */
1825	TAILQ_FOREACH(ihp, &hostlst_head, hostlst) {
1826		if (ihp->revname && strncmp(hostname, ihp->revname,
1827			SM_MAXSTRLEN) == 0) {
1828			hostname = ihp->name;
1829			debuglog("Clearing locks for %s (%s)\n",
1830				hostname, ihp->revname);
1831			break;
1832		}
1833	}
1834
1835	/* Clear blocking file lock list */
1836	clear_blockingfilelock(hostname);
1837
1838	/* do all required unlocks */
1839	/* Note that unlock can smash the current pointer to a lock */
1840
1841	/*
1842	 * Normally, LIST_FOREACH is called for, but since
1843	 * the current element *is* the iterator, deleting it
1844	 * would mess up the iteration.  Thus, a next element
1845	 * must be used explicitly
1846	 */
1847restart:
1848	ifl = LIST_FIRST(&nfslocklist_head);
1849
1850	while (ifl != NULL) {
1851		nfl = LIST_NEXT(ifl, nfslocklist);
1852
1853		if (strncmp(hostname, ifl->client_name, SM_MAXSTRLEN) == 0) {
1854			/* Unlock destroys ifl out from underneath */
1855			pfsret = unlock_partialfilelock(ifl, 0);
1856			if (pfsret != PFL_GRANTED) {
1857				/* Uh oh... there was some sort of problem. */
1858				/* If we restart the loop, we may get */
1859				/* stuck here forever getting errors. */
1860				/* So, let's just abort the whole scan. */
1861				syslog(LOG_WARNING, "lock clearing for %s failed: %d",
1862					hostname, pfsret);
1863				break;
1864			}
1865			/* ifl is NO LONGER VALID AT THIS POINT */
1866			/* Note: the unlock may deallocate several existing locks. */
1867			/* Therefore, we need to restart the scanning of the list, */
1868			/* because nfl could be pointing to a freed lock. */
1869			goto restart;
1870		}
1871		ifl = nfl;
1872	}
1873}
1874
1875/*
1876 * test_partialfilelock:
1877 */
1878enum partialfilelock_status
1879test_partialfilelock(const struct file_lock *fl,
1880    struct file_lock **conflicting_fl)
1881{
1882	enum partialfilelock_status retval;
1883	enum nfslock_status teststatus;
1884
1885	debuglog("Entering testpartialfilelock...\n");
1886
1887	teststatus = test_nfslock(fl, conflicting_fl);
1888	debuglog("test_partialfilelock: teststatus %d\n",teststatus);
1889
1890	if (teststatus == NFS_GRANTED || teststatus == NFS_GRANTED_DUPLICATE) {
1891		/* XXX: Add the underlying filesystem locking code */
1892		retval = (teststatus == NFS_GRANTED) ?
1893		    PFL_GRANTED : PFL_GRANTED_DUPLICATE;
1894		debuglog("Dumping locks...\n");
1895		dump_filelock(fl);
1896		dump_filelock(*conflicting_fl);
1897		debuglog("Done dumping locks...\n");
1898	} else {
1899		retval = PFL_NFSDENIED;
1900		debuglog("NFS test denied.\n");
1901		dump_filelock(fl);
1902		debuglog("Conflicting.\n");
1903		dump_filelock(*conflicting_fl);
1904	}
1905
1906	debuglog("Exiting testpartialfilelock...\n");
1907
1908	return retval;
1909}
1910
1911/*
1912 * Below here are routines associated with translating the partial file locking
1913 * codes into useful codes to send back to the NFS RPC messaging system
1914 */
1915
1916/*
1917 * These routines translate the (relatively) useful return codes back onto
1918 * the few return codes which the nlm subsystems wishes to trasmit
1919 */
1920
1921enum nlm4_stats
1922do_test(struct file_lock *fl, struct file_lock **conflicting_fl)
1923{
1924	enum partialfilelock_status pfsret;
1925	enum nlm4_stats retval;
1926
1927	debuglog("Entering do_test...\n");
1928
1929	pfsret = test_partialfilelock(fl,conflicting_fl);
1930
1931	switch (pfsret) {
1932	case PFL_GRANTED:
1933		debuglog("PFL test lock granted\n");
1934		dump_filelock(fl);
1935		dump_filelock(*conflicting_fl);
1936		retval = (fl->flags & LOCK_V4) ? nlm4_granted : (nlm4_stats) nlm_granted;
1937		break;
1938	case PFL_GRANTED_DUPLICATE:
1939		debuglog("PFL test lock granted--duplicate id detected\n");
1940		dump_filelock(fl);
1941		dump_filelock(*conflicting_fl);
1942		debuglog("Clearing conflicting_fl for call semantics\n");
1943		*conflicting_fl = NULL;
1944		retval = (fl->flags & LOCK_V4) ? nlm4_granted : (nlm4_stats) nlm_granted;
1945		break;
1946	case PFL_NFSDENIED:
1947	case PFL_HWDENIED:
1948		debuglog("PFL test lock denied\n");
1949		dump_filelock(fl);
1950		dump_filelock(*conflicting_fl);
1951		retval = (fl->flags & LOCK_V4) ? nlm4_denied : (nlm4_stats) nlm_denied;
1952		break;
1953	case PFL_NFSRESERR:
1954	case PFL_HWRESERR:
1955		debuglog("PFL test lock resource fail\n");
1956		dump_filelock(fl);
1957		dump_filelock(*conflicting_fl);
1958		retval = (fl->flags & LOCK_V4) ? nlm4_denied_nolocks : (nlm4_stats) nlm_denied_nolocks;
1959		break;
1960	default:
1961		debuglog("PFL test lock *FAILED*\n");
1962		dump_filelock(fl);
1963		dump_filelock(*conflicting_fl);
1964		retval = (fl->flags & LOCK_V4) ? nlm4_failed : (nlm4_stats) nlm_denied;
1965		break;
1966	}
1967
1968	debuglog("Exiting do_test...\n");
1969
1970	return retval;
1971}
1972
1973/*
1974 * do_lock: Try to acquire a lock
1975 *
1976 * This routine makes a distinction between NLM versions.  I am pretty
1977 * convinced that this should be abstracted out and bounced up a level
1978 */
1979
1980enum nlm4_stats
1981do_lock(struct file_lock *fl)
1982{
1983	enum partialfilelock_status pfsret;
1984	enum nlm4_stats retval;
1985
1986	debuglog("Entering do_lock...\n");
1987
1988	pfsret = lock_partialfilelock(fl);
1989
1990	switch (pfsret) {
1991	case PFL_GRANTED:
1992		debuglog("PFL lock granted");
1993		dump_filelock(fl);
1994		retval = (fl->flags & LOCK_V4) ? nlm4_granted : (nlm4_stats) nlm_granted;
1995		break;
1996	case PFL_GRANTED_DUPLICATE:
1997		debuglog("PFL lock granted--duplicate id detected");
1998		dump_filelock(fl);
1999		retval = (fl->flags & LOCK_V4) ? nlm4_granted : (nlm4_stats) nlm_granted;
2000		break;
2001	case PFL_NFSDENIED:
2002	case PFL_HWDENIED:
2003		debuglog("PFL_NFS lock denied");
2004		dump_filelock(fl);
2005		retval = (fl->flags & LOCK_V4) ? nlm4_denied : (nlm4_stats) nlm_denied;
2006		break;
2007	case PFL_NFSBLOCKED:
2008	case PFL_HWBLOCKED:
2009		debuglog("PFL_NFS blocking lock denied.  Queued.\n");
2010		dump_filelock(fl);
2011		retval = (fl->flags & LOCK_V4) ? nlm4_blocked : (nlm4_stats) nlm_blocked;
2012		break;
2013	case PFL_NFSRESERR:
2014	case PFL_HWRESERR:
2015	case PFL_NFSDENIED_NOLOCK:
2016	case PFL_HWDENIED_NOLOCK:
2017		debuglog("PFL lock resource alocation fail\n");
2018		dump_filelock(fl);
2019		retval = (fl->flags & LOCK_V4) ? nlm4_denied_nolocks : (nlm4_stats) nlm_denied_nolocks;
2020		break;
2021	case PFL_HWDENIED_STALEFH:
2022		debuglog("PFL_NFS lock denied STALEFH");
2023		dump_filelock(fl);
2024		retval = (fl->flags & LOCK_V4) ? nlm4_stale_fh : (nlm4_stats) nlm_denied;
2025		break;
2026	case PFL_HWDENIED_READONLY:
2027		debuglog("PFL_NFS lock denied READONLY");
2028		dump_filelock(fl);
2029		retval = (fl->flags & LOCK_V4) ? nlm4_rofs : (nlm4_stats) nlm_denied;
2030		break;
2031	default:
2032		debuglog("PFL lock *FAILED*");
2033		dump_filelock(fl);
2034		retval = (fl->flags & LOCK_V4) ? nlm4_failed : (nlm4_stats) nlm_denied;
2035		break;
2036	}
2037
2038	debuglog("Exiting do_lock...\n");
2039
2040	return retval;
2041}
2042
2043enum nlm4_stats
2044do_unlock(struct file_lock *fl)
2045{
2046	enum partialfilelock_status pfsret;
2047	enum nlm4_stats retval;
2048
2049	debuglog("Entering do_unlock...\n");
2050	pfsret = unlock_partialfilelock(fl, 0);
2051
2052	switch (pfsret) {
2053	case PFL_GRANTED:
2054		debuglog("PFL unlock granted");
2055		dump_filelock(fl);
2056		retval = (fl->flags & LOCK_V4) ? nlm4_granted : (nlm4_stats) nlm_granted;
2057		break;
2058	case PFL_NFSDENIED:
2059	case PFL_HWDENIED:
2060		debuglog("PFL_NFS unlock denied");
2061		dump_filelock(fl);
2062		retval = (fl->flags & LOCK_V4) ? nlm4_denied : (nlm4_stats) nlm_denied;
2063		break;
2064	case PFL_NFSDENIED_NOLOCK:
2065	case PFL_HWDENIED_NOLOCK:
2066		debuglog("PFL_NFS no lock found\n");
2067		retval = (fl->flags & LOCK_V4) ? nlm4_granted : (nlm4_stats) nlm_granted;
2068		break;
2069	case PFL_NFSRESERR:
2070	case PFL_HWRESERR:
2071		debuglog("PFL unlock resource failure");
2072		dump_filelock(fl);
2073		retval = (fl->flags & LOCK_V4) ? nlm4_denied_nolocks : (nlm4_stats) nlm_denied_nolocks;
2074		break;
2075	default:
2076		debuglog("PFL unlock *FAILED*");
2077		dump_filelock(fl);
2078		retval = (fl->flags & LOCK_V4) ? nlm4_failed : (nlm4_stats) nlm_denied;
2079		break;
2080	}
2081
2082	debuglog("Exiting do_unlock...\n");
2083
2084	return retval;
2085}
2086
2087/*
2088 * do_clear
2089 *
2090 * This routine is non-existent because it doesn't have a return code.
2091 * It is here for completeness in case someone *does* need to do return
2092 * codes later.  A decent compiler should optimize this away.
2093 */
2094
2095void
2096do_clear(const char *hostname)
2097{
2098
2099	clear_partialfilelock(hostname);
2100}
2101
2102/*
2103 * do_notify_mounts()
2104 *
2105 * Notify any NFS mounts to recover the locks they held on the given server.
2106 */
2107void
2108do_notify_mounts(const char *hostname)
2109{
2110	struct addrinfo *ailist = NULL, *ai, aihints;
2111	struct lockd_notify ln, *lnp = NULL;
2112	int addrcount, addrmax;
2113
2114	/* Get the list of addresses for this hostname. */
2115	/* Send a lockd-notify with the addresses into the kernel. */
2116	debuglog("do_notify_mounts for %s", hostname);
2117	bzero(&aihints, sizeof(aihints));
2118	aihints.ai_socktype = SOCK_STREAM;
2119	if (getaddrinfo(hostname, NULL, &aihints, &ailist))
2120		return;
2121	for (addrcount=0, ai = ailist; ai; ai = ai->ai_next) {
2122		if ((ai->ai_family != AF_INET) && (ai->ai_family != AF_INET6))
2123			continue;
2124		addrcount++;
2125	}
2126	debuglog("do_notify_mounts for %s, addrcount %d", hostname, addrcount);
2127	if (addrcount > 1)
2128		lnp = malloc(sizeof(*lnp) + ((addrcount-1) * sizeof(struct sockaddr_storage)));
2129	if (lnp) { /* send all the addresses in one notification */
2130		addrmax = addrcount;
2131	} else { /* just send single-address notification(s) */
2132		lnp = &ln;
2133		addrmax = 1;
2134	}
2135
2136	ai = ailist;
2137	while (ai && (addrcount > 0)) {
2138		lnp->ln_version = LOCKD_NOTIFY_VERSION;
2139		lnp->ln_flags = 0;
2140		lnp->ln_addrcount = 0;
2141		for (; ai && (lnp->ln_addrcount < addrmax); ai = ai->ai_next) {
2142			if ((ai->ai_family != AF_INET) && (ai->ai_family != AF_INET6))
2143				continue;
2144			if (config.verbose > 1) {
2145				char addrbuf[NI_MAXHOST], *addr = NULL;
2146				void *sinaddr = (ai->ai_family == AF_INET) ?
2147					(void*)&((struct sockaddr_in*)  AOK ai->ai_addr)->sin_addr :
2148					(void*)&((struct sockaddr_in6*) AOK ai->ai_addr)->sin6_addr;
2149				if (inet_ntop(ai->ai_family, sinaddr, addrbuf, sizeof(addrbuf)))
2150					addr = addrbuf;
2151				debuglog("do_notify_mounts for %s, addr %d ai fam %d len %d %s",
2152					hostname, lnp->ln_addrcount, ai->ai_family, ai->ai_addrlen,
2153					addr ? addr : "???");
2154			}
2155			bcopy(ai->ai_addr, &lnp->ln_addr[lnp->ln_addrcount],
2156				MIN(sizeof(lnp->ln_addr[lnp->ln_addrcount]), ai->ai_addr->sa_len));
2157			lnp->ln_addrcount++;
2158			addrcount--;
2159		}
2160		nfsclnt(NFSCLNT_LOCKDNOTIFY, lnp);
2161	}
2162	freeaddrinfo(ailist);
2163}
2164
2165/*
2166 * The following routines are all called from the code which the
2167 * RPC layer invokes
2168 */
2169
2170/*
2171 * testlock(): inform the caller if the requested lock would be granted
2172 *
2173 * returns NULL if lock would granted
2174 * returns pointer to a conflicting nlm4_holder if not
2175 */
2176
2177struct nlm4_holder *
2178testlock(struct nlm4_lock *lock, bool_t exclusive, int flags __unused)
2179{
2180	struct file_lock test_fl, *conflicting_fl;
2181
2182	if (lock->fh.n_len > NFSV3_MAX_FH_SIZE) {
2183		debuglog("received fhandle size %d, max size %d",
2184		    lock->fh.n_len, NFSV3_MAX_FH_SIZE);
2185		return NULL;
2186	}
2187
2188	bzero(&test_fl, sizeof(test_fl));
2189
2190	test_fl.filehandle.n_len = lock->fh.n_len;
2191	test_fl.filehandle.n_bytes = lock->fh.n_bytes;
2192	copy_nlm4_lock_to_nlm4_holder(lock, exclusive, &test_fl.client);
2193
2194	siglock();
2195	do_test(&test_fl, &conflicting_fl);
2196
2197	if (conflicting_fl == NULL) {
2198		debuglog("No conflicting lock found\n");
2199		sigunlock();
2200		return NULL;
2201	} else {
2202		debuglog("Found conflicting lock\n");
2203		dump_filelock(conflicting_fl);
2204		sigunlock();
2205		return (&conflicting_fl->client);
2206	}
2207}
2208
2209/*
2210 * getlock: try to aquire the lock.
2211 * If file is already locked and we can sleep, put the lock in the list with
2212 * status LKST_WAITING; it'll be processed later.
2213 * Otherwise try to lock. If we're allowed to block, fork a child which
2214 * will do the blocking lock.
2215 */
2216
2217enum nlm4_stats
2218getlock(nlm4_lockargs *lckarg, struct svc_req *rqstp, const int flags)
2219{
2220	struct file_lock *newfl;
2221	enum nlm4_stats retval;
2222
2223	debuglog("Entering getlock...\n");
2224
2225	if (grace_expired == 0 && lckarg->reclaim == 0)
2226		return (flags & LOCK_V4) ?
2227		    nlm4_denied_grace_period : (nlm4_stats) nlm_denied_grace_period;
2228
2229	if (lckarg->alock.fh.n_len > NFSV3_MAX_FH_SIZE) {
2230		debuglog("received fhandle size %d, max size %d",
2231		    lckarg->alock.fh.n_len, NFSV3_MAX_FH_SIZE);
2232		return (flags & LOCK_V4) ? nlm4_failed : (nlm4_stats) nlm_denied;
2233	}
2234
2235	/* allocate new file_lock for this request */
2236	newfl = allocate_file_lock(&lckarg->alock.oh, &lckarg->alock.fh,
2237				   svc_getcaller_sa(rqstp->rq_xprt),
2238				   lckarg->alock.caller_name);
2239	if (newfl == NULL) {
2240		syslog(LOG_NOTICE, "lock allocate failed: %s", strerror(errno));
2241		/* failed */
2242		return (flags & LOCK_V4) ?
2243		    nlm4_denied_nolocks : (nlm4_stats) nlm_denied_nolocks;
2244	}
2245
2246	fill_file_lock(newfl,
2247	    lckarg->exclusive, lckarg->alock.svid, lckarg->alock.l_offset,
2248	    lckarg->alock.l_len,
2249	    lckarg->state, 0, flags, lckarg->block);
2250
2251	/*
2252	 * newfl is now fully constructed and deallocate_file_lock
2253	 * can now be used to delete it
2254	 */
2255
2256	siglock();
2257	debuglog("Pointer to new lock is %p\n",newfl);
2258
2259	retval = do_lock(newfl);
2260
2261	debuglog("Pointer to new lock is %p\n",newfl);
2262	sigunlock();
2263
2264	switch (retval)
2265		{
2266		case nlm4_granted:
2267			/* case nlm_granted: is the same as nlm4_granted */
2268			/* do_mon(lckarg->alock.caller_name); */
2269			break;
2270		case nlm4_blocked:
2271			/* case nlm_blocked: is the same as nlm4_blocked */
2272			/* do_mon(lckarg->alock.caller_name); */
2273			break;
2274		default:
2275			deallocate_file_lock(newfl);
2276			break;
2277		}
2278
2279	debuglog("Exiting getlock...\n");
2280
2281	return retval;
2282}
2283
2284
2285/* unlock a filehandle */
2286enum nlm4_stats
2287unlock(nlm4_lock *lock, const int flags)
2288{
2289	struct file_lock fl;
2290	enum nlm4_stats err;
2291
2292	debuglog("Entering unlock...\n");
2293
2294	if (lock->fh.n_len > NFSV3_MAX_FH_SIZE) {
2295		debuglog("received fhandle size %d, max size %d",
2296		    lock->fh.n_len, NFSV3_MAX_FH_SIZE);
2297		return (flags & LOCK_V4) ? nlm4_failed : (nlm4_stats) nlm_denied;
2298	}
2299
2300	siglock();
2301
2302	bzero(&fl,sizeof(struct file_lock));
2303	fl.filehandle.n_len = lock->fh.n_len;
2304	fl.filehandle.n_bytes = lock->fh.n_bytes;
2305
2306	copy_nlm4_lock_to_nlm4_holder(lock, 0, &fl.client);
2307
2308	err = do_unlock(&fl);
2309
2310	sigunlock();
2311
2312	debuglog("Exiting unlock...\n");
2313
2314	return err;
2315}
2316
2317/* cancel a blocked lock request */
2318enum nlm4_stats
2319cancellock(nlm4_cancargs *args, const int flags)
2320{
2321	struct file_lock *ifl, *nfl;
2322	enum nlm4_stats err;
2323
2324	debuglog("Entering cancellock...\n");
2325
2326	if (args->alock.fh.n_len > NFSV3_MAX_FH_SIZE) {
2327		debuglog("received fhandle size %d, max size %d",
2328		    args->alock.fh.n_len, NFSV3_MAX_FH_SIZE);
2329		return (flags & LOCK_V4) ? nlm4_failed : (nlm4_stats) nlm_denied;
2330	}
2331
2332	siglock();
2333
2334	err = (flags & LOCK_V4) ? nlm4_denied : (nlm4_stats) nlm_denied;
2335
2336	/*
2337	 * scan blocked lock list for matching request and remove/destroy
2338	 */
2339	ifl = LIST_FIRST(&blockedlocklist_head);
2340	for ( ; ifl != NULL; ifl = nfl) {
2341		nfl = LIST_NEXT(ifl, nfslocklist);
2342
2343		/* compare lock fh - filehandle */
2344		if (!same_netobj(&args->alock.fh, &ifl->filehandle))
2345			continue;
2346
2347		/* compare lock caller_name - client_name */
2348		if (strncmp(args->alock.caller_name, ifl->client_name, SM_MAXSTRLEN))
2349			continue;
2350
2351		/* Note: done't compare cookie - client_cookie */
2352		/* The cookie may be specific to the cancel request */
2353		/* and not be the same as the one in the original lock request. */
2354
2355		/* compare lock oh - client.oh */
2356		if (!same_netobj(&args->alock.oh, &ifl->client.oh))
2357			continue;
2358
2359		/* compare lock svid - client.svid */
2360		if (args->alock.svid != ifl->client.svid)
2361			continue;
2362
2363		/* compare lock l_offset - client.l_offset */
2364		if (args->alock.l_offset != ifl->client.l_offset)
2365			continue;
2366
2367		/* compare lock l_len - client.l_len */
2368		if (args->alock.l_len != ifl->client.l_len)
2369			continue;
2370
2371		/* compare exclusive - client.exclusive */
2372		if (args->exclusive != ifl->client.exclusive)
2373			continue;
2374
2375		/* got it */
2376		remove_blockingfilelock(ifl);
2377		deallocate_file_lock(ifl);
2378		err = (flags & LOCK_V4) ? nlm4_granted : (nlm4_stats) nlm_granted;
2379		break;
2380	}
2381
2382	sigunlock();
2383
2384	debuglog("Exiting cancellock...\n");
2385
2386	return err;
2387}
2388
2389
2390/*
2391 * XXX: The following monitor/unmonitor routines
2392 * have not been extensively tested (ie. no regression
2393 * script exists like for the locking sections)
2394 */
2395
2396/*
2397 * Find a lock host on a queue.  If found:
2398 *    bump the ref,
2399 *    bump the access time,
2400 *    dequeue it from the queue it was found on,
2401 *    enqueue it at the front of the "in use" queue.
2402 */
2403static struct host *
2404get_lock_host(struct hostlst_head *hd, const char *hostname, const struct sockaddr *saddr)
2405{
2406	struct host *ihp;
2407
2408	if (!hostname && !saddr)
2409		return (NULL);
2410
2411	debuglog("get_lock_host %s\n", hostname ? hostname : "addr");
2412	TAILQ_FOREACH(ihp, hd, hostlst) {
2413		if (hostname && (strncmp(hostname, ihp->name, SM_MAXSTRLEN) != 0))
2414			continue;
2415		if (saddr && addrcmp(saddr, (struct sockaddr*)&ihp->addr))
2416			continue;
2417		TAILQ_REMOVE(hd, ihp, hostlst);
2418		/*
2419		 * Host is already monitored, so just bump the
2420		 * reference count.  But don't bump the reference
2421		 * count if we're adding additional client-side
2422		 * references.  Client-side monitors are done by
2423		 * address, are never unmonitored, and should only
2424		 * take one refcount.  Otherwise, repeated calls
2425		 * could cause the refcount to wrap.
2426		 */
2427		if (!saddr || !ihp->addr.ss_len)
2428			++ihp->refcnt;
2429		ihp->lastuse = currsec;
2430		/* Host should only be in the monitor list once */
2431		TAILQ_INSERT_HEAD(&hostlst_head, ihp, hostlst);
2432		break;
2433	}
2434	debuglog("get_lock_host %s %s\n",
2435	    ihp == NULL ? "did not find" : "found", hostname ? hostname : "addr");
2436	return (ihp);
2437}
2438
2439/*
2440 * monitor_lock_host: monitor lock hosts locally with a ref count and
2441 * inform statd
2442 */
2443void
2444monitor_lock_host_by_name(const char *hostname, const struct sockaddr *saddr)
2445{
2446	struct host *ihp;
2447
2448	debuglog("monitor_lock_host: %s\n", hostname);
2449	ihp = get_lock_host(&hostlst_head, hostname, NULL);
2450	if (ihp == NULL)
2451		ihp = get_lock_host(&hostlst_unref, hostname, NULL);
2452	if (ihp != NULL) {
2453		if (ihp->revname)
2454			debuglog("Monitor_lock_host: %s - %s (cached)\n",
2455				ihp->name, ihp->revname);
2456		else
2457			debuglog("Monitor_lock_host: %s (cached)\n", ihp->name);
2458		return;
2459	}
2460
2461	monitor_lock_host(hostname, saddr);
2462}
2463
2464void
2465monitor_lock_host_by_addr(const struct sockaddr *saddr)
2466{
2467	struct host *ihp;
2468	char hostaddr[NI_MAXHOST];
2469	char hostname[NI_MAXHOST];
2470
2471	if (getnameinfo(saddr, saddr->sa_len, hostaddr, sizeof(hostaddr),
2472			NULL, 0, NI_NUMERICHOST)) {
2473		debuglog("monitor_lock_host: bad address\n");
2474		return;
2475	}
2476	debuglog("monitor_lock_host: %s\n", hostaddr);
2477	ihp = get_lock_host(&hostlst_head, NULL, saddr);
2478	if (ihp == NULL)
2479		ihp = get_lock_host(&hostlst_unref, NULL, saddr);
2480	if (ihp != NULL) {
2481		debuglog("Monitor_lock_host: %s (cached)\n", ihp->name);
2482		return;
2483	}
2484
2485	if (!getnameinfo(saddr, saddr->sa_len, hostname, sizeof(hostname), NULL, 0, 0))
2486		monitor_lock_host(hostname, saddr);
2487	else
2488		monitor_lock_host(hostaddr, saddr);
2489}
2490
2491static void
2492monitor_lock_host(const char *hostname, const struct sockaddr *saddr)
2493{
2494	struct host *nhp;
2495	struct mon smon;
2496	struct sm_stat_res sres;
2497	int rpcret;
2498	int retrying = 0;
2499	size_t n;
2500	char hostbuf[NI_MAXHOST];
2501
2502	/* Host is not yet monitored, add it */
2503	debuglog("Monitor_lock_host: %s (creating)\n", hostname);
2504	n = strnlen(hostname, SM_MAXSTRLEN);
2505	if (n == SM_MAXSTRLEN) {
2506		debuglog("monitor_lock_host: hostname too long\n");
2507		return;
2508	}
2509	nhp = (struct host *) malloc(sizeof(struct host));
2510	if (nhp == NULL) {
2511		debuglog("Unable to allocate entry for statd mon\n");
2512		return;
2513	}
2514
2515	/* Allocated new host entry, now fill the fields */
2516	nhp->name = strdup(hostname);
2517	if (nhp->name == NULL) {
2518		debuglog("Unable to allocate entry name for statd mon\n");
2519		free(nhp);
2520		return;
2521	}
2522	nhp->revname = NULL;
2523	nhp->refcnt = 1;
2524	nhp->lastuse = currsec;
2525	if (saddr) {
2526		bcopy(saddr, &nhp->addr, saddr->sa_len);
2527	} else {
2528		nhp->addr.ss_len = 0;
2529	}
2530	debuglog("Locally Monitoring host '%s'\n", hostname);
2531
2532	debuglog("Attempting to tell statd\n");
2533
2534	bzero(&smon,sizeof(smon));
2535
2536	smon.mon_id.mon_name = nhp->name;
2537	smon.mon_id.my_id.my_name = (char *) "localhost\0";
2538
2539	smon.mon_id.my_id.my_prog = NLM_PROG;
2540	smon.mon_id.my_id.my_vers = NLM_SM;
2541	smon.mon_id.my_id.my_proc = NLM_SM_NOTIFY;
2542
2543retry:
2544	rpcret = callrpc((char *) "localhost", SM_PROG, SM_VERS, SM_MON, (xdrproc_t)xdr_mon,
2545	    &smon, (xdrproc_t)xdr_sm_stat_res, &sres);
2546
2547	if (rpcret == 0) {
2548		if (sres.res_stat == stat_fail && !retrying) {
2549			debuglog("Statd call failed\n");
2550			/*
2551			 * It's possible that the hostname provided
2552			 * by the client isn't valid. Retry with a
2553			 * a hostname reverse-mapped from the client's
2554			 * address (if provided) and stash this name
2555			 * in the host entry so that we can identify it
2556			 * with this name later when we ask statd to
2557			 * unmonitor it.
2558			 */
2559			if (saddr &&
2560			    !getnameinfo(saddr, saddr->sa_len, hostbuf, sizeof(hostbuf), NULL, 0, 0) &&
2561			    strcmp(nhp->name, hostbuf)) {
2562				debuglog("Statd retry with '%s'\n", hostbuf);
2563				smon.mon_id.mon_name = hostbuf;
2564				nhp->revname = strdup(hostbuf);
2565				if (nhp->revname == NULL) {
2566					debuglog("No memory for revname\n");
2567					free(nhp->name);
2568					free(nhp);
2569					return;
2570				}
2571				retrying = 1;
2572				goto retry;
2573			}
2574		}
2575	} else {
2576		debuglog("Rpc call to statd failed with return value: %d\n",
2577		    rpcret);
2578	}
2579
2580	/*
2581	 * Insert host in the monitor list, even if statd
2582	 * doesn't like it.  If the name is unacceptable
2583	 * to statd, at least we'll avoid subsequent rejection
2584	 * on every lock request.
2585	 */
2586	TAILQ_INSERT_HEAD(&hostlst_head, nhp, hostlst);
2587}
2588
2589/*
2590 * unmonitor_lock_host: clear monitor ref counts and inform statd when gone
2591 */
2592void
2593unmonitor_lock_host(const char *hostname)
2594{
2595	struct host *ihp;
2596
2597	TAILQ_FOREACH(ihp, &hostlst_head, hostlst) {
2598		if (strncmp(hostname, ihp->name, SM_MAXSTRLEN) == 0) {
2599			/* Host is unmonitored, drop refcount */
2600			--ihp->refcnt;
2601			/* Host should only be in the monitor list once */
2602			break;
2603		}
2604	}
2605
2606	if (ihp == NULL) {
2607		debuglog("Could not find host %16s in mon list\n", hostname);
2608		return;
2609	}
2610
2611	if (ihp->refcnt > 0)
2612		return;
2613
2614	if (ihp->refcnt < 0) {
2615		debuglog("Negative refcount!: %d\n", ihp->refcnt);
2616	}
2617
2618	TAILQ_REMOVE(&hostlst_head, ihp, hostlst);
2619	TAILQ_INSERT_HEAD(&hostlst_unref, ihp, hostlst);
2620	if (config.host_monitor_cache_timeout <= 0)
2621		destroy_lock_host(ihp);
2622}
2623
2624void
2625destroy_lock_host(struct host *ihp)
2626{
2627	struct mon_id smon_id;
2628	struct sm_stat smstat;
2629	int rpcret;
2630	char *name;
2631
2632	/*
2633	 * If the client was monitored with a hostname obtained from
2634	 * its address, then use that instead of the possibly invalid
2635	 * hostname provided in caller_name.
2636	 */
2637	name = ihp->revname ? ihp->revname : ihp->name;
2638	debuglog("Attempting to unmonitor host %16s\n", name);
2639
2640	bzero(&smon_id,sizeof(smon_id));
2641	smon_id.mon_name = name;
2642	smon_id.my_id.my_name = (char *) "localhost";
2643	smon_id.my_id.my_prog = NLM_PROG;
2644	smon_id.my_id.my_vers = NLM_SM;
2645	smon_id.my_id.my_proc = NLM_SM_NOTIFY;
2646
2647	rpcret = callrpc((char *) "localhost", SM_PROG, SM_VERS, SM_UNMON, (xdrproc_t)xdr_mon_id,
2648	    &smon_id, (xdrproc_t)xdr_sm_stat, &smstat);
2649
2650	if (rpcret != 0) {
2651		debuglog("Rpc call to unmonitor statd failed with "
2652		   " return value: %d: %s", rpcret, clnt_sperrno(rpcret));
2653	} else {
2654		debuglog("Succeeded unmonitoring %16s\n", ihp->name);
2655	}
2656
2657	TAILQ_REMOVE(&hostlst_unref, ihp, hostlst);
2658	if (ihp->name)
2659		free(ihp->name);
2660	if (ihp->revname)
2661		free(ihp->revname);
2662	free(ihp);
2663}
2664
2665/*
2666 * returns 1 if there are hosts to expire or 0 if there are none.
2667 */
2668int
2669expire_lock_hosts(void)
2670{
2671	struct host *ihp;
2672
2673	for ( ;; ) {
2674		ihp = TAILQ_LAST(&hostlst_unref, hostlst_head);
2675		if (ihp == NULL)
2676			break;
2677		if ((config.host_monitor_cache_timeout > 0) &&
2678		    (ihp->lastuse >= (currsec - config.host_monitor_cache_timeout)))
2679			break;
2680		debuglog("expire_lock_hosts: expiring %s %d %d %d\n",
2681		    ihp->name, (int)ihp->lastuse,
2682		    (int)currsec, (int)currsec - config.host_monitor_cache_timeout);
2683		destroy_lock_host(ihp);
2684	}
2685	return (TAILQ_LAST(&hostlst_unref, hostlst_head) != NULL);
2686}
2687
2688/*
2689 * notify: Clear all locks from a host if statd complains
2690 *
2691 * XXX: This routine has not been thoroughly tested.  However, neither
2692 * had the old one been.  It used to compare the statd crash state counter
2693 * to the current lock state.  The upshot of this was that it basically
2694 * cleared all locks from the specified host 99% of the time (with the
2695 * other 1% being a bug).  Consequently, the assumption is that clearing
2696 * all locks from a host when notified by statd is acceptable.
2697 *
2698 * Please note that this routine skips the usual level of redirection
2699 * through a do_* type routine.  This introduces a possible level of
2700 * error and might better be written as do_notify and take this one out.
2701
2702 */
2703
2704void
2705notify(const char *hostname, const int state)
2706{
2707	debuglog("notify from %s, new state %d", hostname, state);
2708
2709	siglock();
2710	do_clear(hostname);
2711	do_notify_mounts(hostname);
2712	sigunlock();
2713
2714	debuglog("Leaving notify\n");
2715}
2716
2717int
2718send_granted(fl, opcode)
2719	struct file_lock *fl;
2720	int opcode __unused;
2721{
2722	CLIENT *cli;
2723	static char dummy;
2724	struct timeval timeo;
2725	enum clnt_stat rv;
2726	static struct nlm_res retval;
2727	static struct nlm4_res retval4;
2728
2729	debuglog("About to send granted on blocked lock\n");
2730
2731	cli = get_client(fl->addr, (fl->flags & LOCK_V4) ? NLM_VERS4 : NLM_VERS, 0, 0);
2732	if (cli == NULL) {
2733		syslog(LOG_NOTICE, "failed to get CLIENT for %s",
2734		    fl->client_name);
2735		/*
2736		 * We fail to notify remote that the lock has been granted.
2737		 * The client will timeout and retry, the lock will be
2738		 * granted at this time.
2739		 */
2740		return -1;
2741	}
2742	timeo.tv_sec = 0;
2743	timeo.tv_usec = (fl->flags & LOCK_ASYNC) ? 0 : 500000; /* 0.5s */
2744
2745	fl->granted_cookie = ++send_granted_cookie;
2746	if (!send_granted_cookie)
2747		send_granted_cookie++;
2748
2749	if (fl->flags & LOCK_V4) {
2750		static nlm4_testargs res;
2751		res.cookie.n_len = sizeof(fl->granted_cookie);
2752		res.cookie.n_bytes = (uint8_t*)&fl->granted_cookie;
2753		res.exclusive = fl->client.exclusive;
2754		res.alock.caller_name = fl->client_name;
2755		res.alock.fh.n_len = fl->filehandle.n_len;
2756		res.alock.fh.n_bytes = fl->filehandle.n_bytes;
2757		res.alock.oh = fl->client.oh;
2758		res.alock.svid = fl->client.svid;
2759		res.alock.l_offset = fl->client.l_offset;
2760		res.alock.l_len = fl->client.l_len;
2761		debuglog("sending v4 reply%s",
2762			 (fl->flags & LOCK_ASYNC) ? " (async)":"");
2763		if (fl->flags & LOCK_ASYNC) {
2764			rv = clnt_call(cli, NLM4_GRANTED_MSG,
2765			    (xdrproc_t)xdr_nlm4_testargs, &res, (xdrproc_t)xdr_void, &dummy, timeo);
2766		} else {
2767			rv = clnt_call(cli, NLM4_GRANTED,
2768			    (xdrproc_t)xdr_nlm4_testargs, &res, (xdrproc_t)xdr_nlm4_res,
2769			    &retval4, timeo);
2770		}
2771	} else {
2772		static nlm_testargs res;
2773
2774		res.cookie.n_len = sizeof(fl->granted_cookie);
2775		res.cookie.n_bytes = (uint8_t*)&fl->granted_cookie;
2776		res.exclusive = fl->client.exclusive;
2777		res.alock.caller_name = fl->client_name;
2778		res.alock.fh.n_len = fl->filehandle.n_len;
2779		res.alock.fh.n_bytes = fl->filehandle.n_bytes;
2780		res.alock.oh = fl->client.oh;
2781		res.alock.svid = fl->client.svid;
2782		res.alock.l_offset = fl->client.l_offset;
2783		res.alock.l_len = fl->client.l_len;
2784		debuglog("sending v1 reply%s",
2785			 (fl->flags & LOCK_ASYNC) ? " (async)":"");
2786		if (fl->flags & LOCK_ASYNC) {
2787			rv = clnt_call(cli, NLM_GRANTED_MSG,
2788			    (xdrproc_t)xdr_nlm_testargs, &res, (xdrproc_t)xdr_void, &dummy, timeo);
2789		} else {
2790			rv = clnt_call(cli, NLM_GRANTED,
2791			    (xdrproc_t)xdr_nlm_testargs, &res, (xdrproc_t)xdr_nlm_res,
2792			    &retval, timeo);
2793		}
2794	}
2795	if (config.verbose > 2)
2796		debuglog("clnt_call returns %d(%s) for granted",
2797			 rv, clnt_sperrno(rv));
2798
2799	if ((rv != RPC_SUCCESS) &&
2800	    !((fl->flags & LOCK_ASYNC) && (rv == RPC_TIMEDOUT)))
2801		return -1;
2802	return 0;
2803}
2804
2805/*
2806 * granted_failed: remove a granted lock that wasn't successfully
2807 * accepted by the client
2808 */
2809void
2810granted_failed(nlm4_res *arg)
2811{
2812	u_int64_t cookie;
2813	struct file_lock *ifl;
2814
2815	debuglog("Entering granted_failed, status %d\n", arg->stat.stat);
2816
2817	if (arg->cookie.n_len != sizeof(cookie)) {
2818		debuglog("Exiting granted_failed: bogus cookie size %d\n",
2819			arg->cookie.n_len);
2820		return;
2821	}
2822	bcopy(arg->cookie.n_bytes, &cookie, sizeof(cookie));
2823	debuglog("granted_failed, cookie 0x%llx\n", cookie);
2824
2825	LIST_FOREACH(ifl, &nfslocklist_head, nfslocklist) {
2826		debuglog("Pointer to file lock: %p\n",ifl);
2827
2828		debuglog("****Dump of ifl****\n");
2829		dump_filelock(ifl);
2830
2831		if (ifl->granted_cookie != cookie)
2832			continue;
2833
2834		debuglog("granted_failed: cookie found\n");
2835		break;
2836	}
2837
2838	if (ifl) {
2839		do_unlock(ifl);
2840		/* ifl is NO LONGER VALID AT THIS POINT */
2841	} else {
2842		debuglog("granted_failed: cookie NOT FOUND\n");
2843	}
2844
2845	debuglog("Exiting granted_failed\n");
2846}
2847
2848/*
2849 * getshare: try to acquire a share reservation
2850 */
2851enum nlm4_stats
2852getshare(nlm_shareargs *shrarg, struct svc_req *rqstp __unused, const int flags)
2853{
2854	struct sharefile *shrfile;
2855	struct file_share *sh;
2856	size_t n;
2857
2858	debuglog("Entering getshare...\n");
2859
2860	if (grace_expired == 0 && shrarg->reclaim == 0) {
2861		debuglog("getshare denied - grace period\n");
2862		return (flags & LOCK_V4) ?
2863			nlm4_denied_grace_period :
2864			(nlm4_stats) nlm_denied_grace_period;
2865	}
2866
2867	if (shrarg->share.fh.n_len > NFSV3_MAX_FH_SIZE) {
2868		debuglog("received fhandle size %d, max size %d",
2869		    shrarg->share.fh.n_len, NFSV3_MAX_FH_SIZE);
2870		return (flags & LOCK_V4) ? nlm4_failed : (nlm4_stats) nlm_denied;
2871	}
2872
2873	/* find file in list of share files */
2874	LIST_FOREACH(shrfile, &nfssharefilelist_head, sharefilelist) {
2875		if ((shrarg->share.fh.n_len == shrfile->filehandle.n_len) &&
2876		    (bcmp(shrarg->share.fh.n_bytes, shrfile->filehandle.n_bytes,
2877			shrarg->share.fh.n_len) == 0)) {
2878			/* shrfile is the correct file */
2879			break;
2880		}
2881	}
2882
2883	/* if share file not found, create a new share file */
2884	if (!shrfile) {
2885		fhandle_t fh;
2886		int fd;
2887		fh.fh_len = shrarg->share.fh.n_len;
2888		bcopy(shrarg->share.fh.n_bytes, fh.fh_data, fh.fh_len);
2889		fd = fhopen(&fh, O_RDONLY);
2890		if (fd < 0) {
2891			debuglog("fhopen failed (from %16s): %32s\n",
2892			    shrarg->share.caller_name, strerror(errno));
2893			if ((flags & LOCK_V4) == 0)
2894				return (nlm4_stats) nlm_denied;
2895			switch (errno) {
2896			case ESTALE:
2897				return nlm4_stale_fh;
2898			default:
2899				return nlm4_failed;
2900			}
2901		}
2902		shrfile = malloc(sizeof(struct sharefile));
2903		if (!shrfile) {
2904			debuglog("getshare failed: can't allocate sharefile\n");
2905			close(fd);
2906			return (flags & LOCK_V4) ? nlm4_denied_nolocks : (nlm4_stats) nlm_denied_nolocks;
2907		}
2908		shrfile->filehandle.n_len = shrarg->share.fh.n_len;
2909		shrfile->filehandle.n_bytes = malloc(shrarg->share.fh.n_len);
2910		if (!shrfile->filehandle.n_bytes) {
2911			debuglog("getshare failed: can't allocate sharefile filehandle\n");
2912			free(shrfile);
2913			close(fd);
2914			return (flags & LOCK_V4) ? nlm4_denied_nolocks : (nlm4_stats) nlm_denied_nolocks;
2915		}
2916		bcopy(shrarg->share.fh.n_bytes, shrfile->filehandle.n_bytes,
2917			shrarg->share.fh.n_len);
2918		shrfile->fd = fd;
2919		shrfile->refcount = 0;
2920		shrfile->sharelist_head.lh_first = NULL;
2921		LIST_INSERT_HEAD(&nfssharefilelist_head, shrfile, sharefilelist);
2922	}
2923
2924	/* compare request mode/access to current shares */
2925	LIST_FOREACH(sh, &shrfile->sharelist_head, nfssharelist) {
2926		/* if request host/owner matches a current share... */
2927		if ((strncmp(shrarg->share.caller_name, sh->client_name, SM_MAXSTRLEN) == 0) &&
2928			same_netobj(&shrarg->share.oh, &sh->oh)) {
2929			/* ...then just update share mode/access */
2930			sh->mode = shrarg->share.mode;
2931			sh->access = shrarg->share.access;
2932			debuglog("getshare: updated existing share\n");
2933			return (flags & LOCK_V4) ? nlm4_granted : (nlm4_stats) nlm_granted;
2934		}
2935		if (((shrarg->share.mode & sh->access) != 0) ||
2936		    ((shrarg->share.access & sh->mode) != 0)) {
2937			/* share request conflicts with existing share */
2938			debuglog("getshare: conflicts with existing share\n");
2939			return (flags & LOCK_V4) ? nlm4_denied : (nlm4_stats) nlm_denied;
2940		}
2941	}
2942
2943	/* create/init new share */
2944	n = strnlen(shrarg->share.caller_name, SM_MAXSTRLEN);
2945	if (n < SM_MAXSTRLEN) {
2946		sh = malloc(sizeof(*sh) - sizeof(sh->client_name) + n + 1);
2947	} else {
2948		debuglog("getshare failed: hostname too long\n");
2949		sh = NULL;
2950	}
2951	if (!sh) {
2952		debuglog("getshare failed: can't allocate share\n");
2953		if (!shrfile->refcount) {
2954			LIST_REMOVE(shrfile, sharefilelist);
2955			close(shrfile->fd);
2956			free(shrfile->filehandle.n_bytes);
2957			free(shrfile);
2958		}
2959		return (flags & LOCK_V4) ? nlm4_denied_nolocks : (nlm4_stats) nlm_denied_nolocks;
2960	}
2961	bzero(sh, sizeof(*sh) - sizeof(sh->client_name));
2962	sh->oh.n_len = shrarg->share.oh.n_len;
2963	sh->oh.n_bytes = malloc(sh->oh.n_len);
2964	if (!sh->oh.n_bytes) {
2965		debuglog("getshare failed: can't allocate share owner handle\n");
2966		free(sh);
2967		if (!shrfile->refcount) {
2968			LIST_REMOVE(shrfile, sharefilelist);
2969			close(shrfile->fd);
2970			free(shrfile->filehandle.n_bytes);
2971			free(shrfile);
2972		}
2973		return (flags & LOCK_V4) ? nlm4_denied_nolocks : (nlm4_stats) nlm_denied_nolocks;
2974	}
2975	memcpy(sh->client_name, shrarg->share.caller_name, n);
2976	sh->client_name[n] = 0;
2977	sh->mode = shrarg->share.mode;
2978	sh->access = shrarg->share.access;
2979
2980	/* insert new share into file's share list */
2981	LIST_INSERT_HEAD(&shrfile->sharelist_head, sh, nfssharelist);
2982	shrfile->refcount++;
2983
2984	debuglog("Exiting getshare...\n");
2985
2986	return (flags & LOCK_V4) ? nlm4_granted : (nlm4_stats) nlm_granted;
2987}
2988
2989
2990/* remove a share reservation */
2991enum nlm4_stats
2992unshare(nlm_shareargs *shrarg, struct svc_req *rqstp __unused, const int flags)
2993{
2994	struct sharefile *shrfile;
2995	struct file_share *sh;
2996
2997	debuglog("Entering unshare...\n");
2998
2999	if (shrarg->share.fh.n_len > NFSV3_MAX_FH_SIZE) {
3000		debuglog("received fhandle size %d, max size %d",
3001		    shrarg->share.fh.n_len, NFSV3_MAX_FH_SIZE);
3002		return (flags & LOCK_V4) ? nlm4_failed : (nlm4_stats) nlm_denied;
3003	}
3004
3005	/* find file in list of share files */
3006	LIST_FOREACH(shrfile, &nfssharefilelist_head, sharefilelist) {
3007		if ((shrarg->share.fh.n_len == shrfile->filehandle.n_len) &&
3008		    (bcmp(shrarg->share.fh.n_bytes, shrfile->filehandle.n_bytes,
3009			shrarg->share.fh.n_len) == 0)) {
3010			/* shrfile is the correct file */
3011			break;
3012		}
3013	}
3014
3015	/* if share file not found, return success (per spec) */
3016	if (!shrfile) {
3017		debuglog("unshare: no such share file\n");
3018		return (flags & LOCK_V4) ? nlm4_granted : (nlm4_stats) nlm_granted;
3019	}
3020
3021	/* find share */
3022	LIST_FOREACH(sh, &shrfile->sharelist_head, nfssharelist) {
3023		/* if request host/owner matches a current share... */
3024		if ((strncmp(shrarg->share.caller_name, sh->client_name, SM_MAXSTRLEN) == 0) &&
3025			same_netobj(&shrarg->share.oh, &sh->oh))
3026			break;
3027	}
3028
3029	/* if share not found, return success (per spec) */
3030	if (!sh) {
3031		debuglog("unshare: no such share\n");
3032		return (flags & LOCK_V4) ? nlm4_granted : (nlm4_stats) nlm_granted;
3033	}
3034
3035	/* remove share from file and deallocate */
3036	shrfile->refcount--;
3037	LIST_REMOVE(sh, nfssharelist);
3038	free(sh->oh.n_bytes);
3039	free(sh);
3040
3041	/* if file has no more shares, deallocate share file */
3042	if (!shrfile->refcount) {
3043		debuglog("unshare: file has no more shares\n");
3044		LIST_REMOVE(shrfile, sharefilelist);
3045		close(shrfile->fd);
3046		free(shrfile->filehandle.n_bytes);
3047		free(shrfile);
3048	}
3049
3050	debuglog("Exiting unshare...\n");
3051
3052	return (flags & LOCK_V4) ? nlm4_granted : (nlm4_stats) nlm_granted;
3053}
3054
3055/*
3056 * do_free_all
3057 *
3058 * Wipe out all non-monitored locks and shares held by a host.
3059 */
3060
3061void
3062do_free_all(const char *hostname)
3063{
3064	struct file_lock *ifl, *nfl;
3065	struct sharefile *shrfile, *nshrfile;
3066	struct file_share *ifs, *nfs;
3067	enum partialfilelock_status pfsret;
3068
3069	/* clear non-monitored blocking file locks */
3070	ifl = LIST_FIRST(&blockedlocklist_head);
3071	while (ifl != NULL) {
3072		nfl = LIST_NEXT(ifl, nfslocklist);
3073
3074		if (((ifl->flags & LOCK_MON) == 0) &&
3075		    (strncmp(hostname, ifl->client_name, SM_MAXSTRLEN) == 0)) {
3076			remove_blockingfilelock(ifl);
3077			deallocate_file_lock(ifl);
3078		}
3079
3080		ifl = nfl;
3081	}
3082
3083	/* clear non-monitored file locks */
3084restart:
3085	ifl = LIST_FIRST(&nfslocklist_head);
3086	while (ifl != NULL) {
3087		nfl = LIST_NEXT(ifl, nfslocklist);
3088
3089		if (((ifl->flags & LOCK_MON) == 0) &&
3090		    (strncmp(hostname, ifl->client_name, SM_MAXSTRLEN) == 0)) {
3091			/* Unlock destroys ifl out from underneath */
3092			pfsret = unlock_partialfilelock(ifl, 0);
3093			if (pfsret != PFL_GRANTED) {
3094				/* Uh oh... there was some sort of problem. */
3095				/* If we restart the loop, we may get */
3096				/* stuck here forever getting errors. */
3097				/* So, let's just abort the whole scan. */
3098				syslog(LOG_WARNING, "unmonitored lock clearing for %s failed: %d",
3099					hostname, pfsret);
3100				break;
3101			}
3102			/* ifl is NO LONGER VALID AT THIS POINT */
3103			/* Note: the unlock may deallocate several existing locks. */
3104			/* Therefore, we need to restart the scanning of the list, */
3105			/* because nfl could be pointing to a freed lock. */
3106			goto restart;
3107		}
3108
3109		ifl = nfl;
3110	}
3111
3112	/* clear shares */
3113	shrfile = LIST_FIRST(&nfssharefilelist_head);
3114	while (shrfile != NULL) {
3115		nshrfile = LIST_NEXT(shrfile, sharefilelist);
3116
3117		ifs = LIST_FIRST(&shrfile->sharelist_head);
3118		while (ifs != NULL) {
3119			nfs = LIST_NEXT(ifs, nfssharelist);
3120
3121			if (strncmp(hostname, ifs->client_name, SM_MAXSTRLEN) == 0) {
3122				shrfile->refcount--;
3123				LIST_REMOVE(ifs, nfssharelist);
3124				free(ifs->oh.n_bytes);
3125				free(ifs);
3126			}
3127
3128			ifs = nfs;
3129		}
3130
3131		if (!shrfile->refcount) {
3132			LIST_REMOVE(shrfile, sharefilelist);
3133			close(shrfile->fd);
3134			free(shrfile->filehandle.n_bytes);
3135			free(shrfile);
3136		}
3137
3138		shrfile = nshrfile;
3139	}
3140
3141}
3142
3143
3144
3145/*
3146 * Routines below here have not been modified in the overhaul
3147 */
3148
3149/*
3150 * Are these two routines still required since lockd is not spawning off
3151 * children to service locks anymore?  Presumably they were originally
3152 * put in place to prevent a one child from changing the lock list out
3153 * from under another one.
3154 */
3155
3156void
3157siglock(void)
3158{
3159  sigset_t block;
3160
3161  sigemptyset(&block);
3162  sigaddset(&block, SIGCHLD);
3163
3164  if (sigprocmask(SIG_BLOCK, &block, NULL) < 0) {
3165    syslog(LOG_WARNING, "siglock failed: %s", strerror(errno));
3166  }
3167}
3168
3169void
3170sigunlock(void)
3171{
3172  sigset_t block;
3173
3174  sigemptyset(&block);
3175  sigaddset(&block, SIGCHLD);
3176
3177  if (sigprocmask(SIG_UNBLOCK, &block, NULL) < 0) {
3178    syslog(LOG_WARNING, "sigunlock failed: %s", strerror(errno));
3179  }
3180}
3181
3182
3183