xref: /freebsd/usr.sbin/rpc.lockd/lockd_lock.c (revision 0b3105a37d7adcadcb720112fed4dc4e8040be99)
1 /*	$NetBSD: lockd_lock.c,v 1.5 2000/11/21 03:47:41 enami Exp $	*/
2 
3 /*
4  * Copyright (c) 2001 Andrew P. Lentvorski, Jr.
5  * Copyright (c) 2000 Manuel Bouyer.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. All advertising materials mentioning features or use of this software
16  *    must display the following acknowledgement:
17  *	This product includes software developed by the University of
18  *	California, Berkeley and its contributors.
19  * 4. Neither the name of the University nor the names of its contributors
20  *    may be used to endorse or promote products derived from this software
21  *    without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  *
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #define LOCKD_DEBUG
41 
42 #include <stdio.h>
43 #ifdef LOCKD_DEBUG
44 #include <stdarg.h>
45 #endif
46 #include <stdlib.h>
47 #include <unistd.h>
48 #include <fcntl.h>
49 #include <syslog.h>
50 #include <errno.h>
51 #include <string.h>
52 #include <signal.h>
53 #include <rpc/rpc.h>
54 #include <sys/types.h>
55 #include <sys/stat.h>
56 #include <sys/socket.h>
57 #include <sys/param.h>
58 #include <sys/mount.h>
59 #include <sys/wait.h>
60 #include <rpcsvc/sm_inter.h>
61 #include <rpcsvc/nlm_prot.h>
62 #include "lockd_lock.h"
63 #include "lockd.h"
64 
65 #define MAXOBJECTSIZE 64
66 #define MAXBUFFERSIZE 1024
67 
68 /*
69  * A set of utilities for managing file locking
70  *
71  * XXX: All locks are in a linked list, a better structure should be used
72  * to improve search/access efficiency.
73  */
74 
75 /* struct describing a lock */
76 struct file_lock {
77 	LIST_ENTRY(file_lock) nfslocklist;
78 	fhandle_t filehandle; /* NFS filehandle */
79 	struct sockaddr *addr;
80 	struct nlm4_holder client; /* lock holder */
81 	/* XXX: client_cookie used *only* in send_granted */
82 	netobj client_cookie; /* cookie sent by the client */
83 	int nsm_status; /* status from the remote lock manager */
84 	int status; /* lock status, see below */
85 	int flags; /* lock flags, see lockd_lock.h */
86 	int blocking; /* blocking lock or not */
87 	char client_name[SM_MAXSTRLEN];	/* client_name is really variable
88 					   length and must be last! */
89 };
90 
91 LIST_HEAD(nfslocklist_head, file_lock);
92 struct nfslocklist_head nfslocklist_head = LIST_HEAD_INITIALIZER(nfslocklist_head);
93 
94 LIST_HEAD(blockedlocklist_head, file_lock);
95 struct blockedlocklist_head blockedlocklist_head = LIST_HEAD_INITIALIZER(blockedlocklist_head);
96 
97 /* lock status */
98 #define LKST_LOCKED	1 /* lock is locked */
99 /* XXX: Is this flag file specific or lock specific? */
100 #define LKST_WAITING	2 /* file is already locked by another host */
101 #define LKST_PROCESSING	3 /* child is trying to acquire the lock */
102 #define LKST_DYING	4 /* must dies when we get news from the child */
103 
104 /* struct describing a monitored host */
105 struct host {
106 	LIST_ENTRY(host) hostlst;
107 	int refcnt;
108 	char name[SM_MAXSTRLEN]; /* name is really variable length and
109                                     must be last! */
110 };
111 /* list of hosts we monitor */
112 LIST_HEAD(hostlst_head, host);
113 struct hostlst_head hostlst_head = LIST_HEAD_INITIALIZER(hostlst_head);
114 
115 /*
116  * File monitoring handlers
117  * XXX: These might be able to be removed when kevent support
118  * is placed into the hardware lock/unlock routines.  (ie.
119  * let the kernel do all the file monitoring)
120  */
121 
122 /* Struct describing a monitored file */
123 struct monfile {
124 	LIST_ENTRY(monfile) monfilelist;
125 	fhandle_t filehandle; /* Local access filehandle */
126 	int fd; /* file descriptor: remains open until unlock! */
127 	int refcount;
128 	int exclusive;
129 };
130 
131 /* List of files we monitor */
132 LIST_HEAD(monfilelist_head, monfile);
133 struct monfilelist_head monfilelist_head = LIST_HEAD_INITIALIZER(monfilelist_head);
134 
135 static int debugdelay = 0;
136 
137 enum nfslock_status { NFS_GRANTED = 0, NFS_GRANTED_DUPLICATE,
138 		      NFS_DENIED, NFS_DENIED_NOLOCK,
139 		      NFS_RESERR };
140 
141 enum hwlock_status { HW_GRANTED = 0, HW_GRANTED_DUPLICATE,
142 		     HW_DENIED, HW_DENIED_NOLOCK,
143 		     HW_STALEFH, HW_READONLY, HW_RESERR };
144 
145 enum partialfilelock_status { PFL_GRANTED=0, PFL_GRANTED_DUPLICATE, PFL_DENIED,
146 			      PFL_NFSDENIED, PFL_NFSBLOCKED, PFL_NFSDENIED_NOLOCK, PFL_NFSRESERR,
147 			      PFL_HWDENIED,  PFL_HWBLOCKED,  PFL_HWDENIED_NOLOCK, PFL_HWRESERR};
148 
149 enum LFLAGS {LEDGE_LEFT, LEDGE_LBOUNDARY, LEDGE_INSIDE, LEDGE_RBOUNDARY, LEDGE_RIGHT};
150 enum RFLAGS {REDGE_LEFT, REDGE_LBOUNDARY, REDGE_INSIDE, REDGE_RBOUNDARY, REDGE_RIGHT};
151 /* XXX: WARNING! I HAVE OVERLOADED THIS STATUS ENUM!  SPLIT IT APART INTO TWO */
152 enum split_status {SPL_DISJOINT=0, SPL_LOCK1=1, SPL_LOCK2=2, SPL_CONTAINED=4, SPL_RESERR=8};
153 
154 enum partialfilelock_status lock_partialfilelock(struct file_lock *fl);
155 
156 void send_granted(struct file_lock *fl, int opcode);
157 void siglock(void);
158 void sigunlock(void);
159 void monitor_lock_host(const char *hostname);
160 void unmonitor_lock_host(char *hostname);
161 
162 void	copy_nlm4_lock_to_nlm4_holder(const struct nlm4_lock *src,
163     const bool_t exclusive, struct nlm4_holder *dest);
164 struct file_lock *	allocate_file_lock(const netobj *lockowner,
165 					   const netobj *matchcookie,
166 					   const struct sockaddr *addr,
167 					   const char *caller_name);
168 void	deallocate_file_lock(struct file_lock *fl);
169 void	fill_file_lock(struct file_lock *fl, const fhandle_t *fh,
170 		       const bool_t exclusive, const int32_t svid,
171     const u_int64_t offset, const u_int64_t len,
172     const int state, const int status, const int flags, const int blocking);
173 int	regions_overlap(const u_int64_t start1, const u_int64_t len1,
174     const u_int64_t start2, const u_int64_t len2);
175 enum split_status  region_compare(const u_int64_t starte, const u_int64_t lene,
176     const u_int64_t startu, const u_int64_t lenu,
177     u_int64_t *start1, u_int64_t *len1, u_int64_t *start2, u_int64_t *len2);
178 int	same_netobj(const netobj *n0, const netobj *n1);
179 int	same_filelock_identity(const struct file_lock *fl0,
180     const struct file_lock *fl2);
181 
182 static void debuglog(char const *fmt, ...);
183 void dump_static_object(const unsigned char* object, const int sizeof_object,
184                         unsigned char* hbuff, const int sizeof_hbuff,
185                         unsigned char* cbuff, const int sizeof_cbuff);
186 void dump_netobj(const struct netobj *nobj);
187 void dump_filelock(const struct file_lock *fl);
188 struct file_lock *	get_lock_matching_unlock(const struct file_lock *fl);
189 enum nfslock_status	test_nfslock(const struct file_lock *fl,
190     struct file_lock **conflicting_fl);
191 enum nfslock_status	lock_nfslock(struct file_lock *fl);
192 enum nfslock_status	delete_nfslock(struct file_lock *fl);
193 enum nfslock_status	unlock_nfslock(const struct file_lock *fl,
194     struct file_lock **released_lock, struct file_lock **left_lock,
195     struct file_lock **right_lock);
196 enum hwlock_status lock_hwlock(struct file_lock *fl);
197 enum split_status split_nfslock(const struct file_lock *exist_lock,
198     const struct file_lock *unlock_lock, struct file_lock **left_lock,
199     struct file_lock **right_lock);
200 int	duplicate_block(struct file_lock *fl);
201 void	add_blockingfilelock(struct file_lock *fl);
202 enum hwlock_status	unlock_hwlock(const struct file_lock *fl);
203 enum hwlock_status	test_hwlock(const struct file_lock *fl,
204     struct file_lock **conflicting_fl);
205 void	remove_blockingfilelock(struct file_lock *fl);
206 void	clear_blockingfilelock(const char *hostname);
207 void	retry_blockingfilelocklist(void);
208 enum partialfilelock_status	unlock_partialfilelock(
209     const struct file_lock *fl);
210 void	clear_partialfilelock(const char *hostname);
211 enum partialfilelock_status	test_partialfilelock(
212     const struct file_lock *fl, struct file_lock **conflicting_fl);
213 enum nlm_stats	do_test(struct file_lock *fl,
214     struct file_lock **conflicting_fl);
215 enum nlm_stats	do_unlock(struct file_lock *fl);
216 enum nlm_stats	do_lock(struct file_lock *fl);
217 void	do_clear(const char *hostname);
218 size_t	strnlen(const char *, size_t);
219 
220 void
221 debuglog(char const *fmt, ...)
222 {
223 	va_list ap;
224 
225 	if (debug_level < 1) {
226 		return;
227 	}
228 
229 	sleep(debugdelay);
230 
231 	va_start(ap, fmt);
232 	vsyslog(LOG_DEBUG, fmt, ap);
233 	va_end(ap);
234 }
235 
236 void
237 dump_static_object(object, size_object, hbuff, size_hbuff, cbuff, size_cbuff)
238 	const unsigned char *object;
239 	const int size_object;
240 	unsigned char *hbuff;
241 	const int size_hbuff;
242 	unsigned char *cbuff;
243 	const int size_cbuff;
244 {
245 	int i, objectsize;
246 
247 	if (debug_level < 2) {
248 		return;
249 	}
250 
251 	objectsize = size_object;
252 
253 	if (objectsize == 0) {
254 		debuglog("object is size 0\n");
255 	} else {
256 		if (objectsize > MAXOBJECTSIZE) {
257 			debuglog("Object of size %d being clamped"
258 			    "to size %d\n", objectsize, MAXOBJECTSIZE);
259 			objectsize = MAXOBJECTSIZE;
260 		}
261 
262 		if (hbuff != NULL) {
263 			if (size_hbuff < objectsize*2+1) {
264 				debuglog("Hbuff not large enough."
265 				    "  Increase size\n");
266 			} else {
267 				for(i=0;i<objectsize;i++) {
268 					sprintf(hbuff+i*2,"%02x",*(object+i));
269 				}
270 				*(hbuff+i*2) = '\0';
271 			}
272 		}
273 
274 		if (cbuff != NULL) {
275 			if (size_cbuff < objectsize+1) {
276 				debuglog("Cbuff not large enough."
277 				    "  Increase Size\n");
278 			}
279 
280 			for(i=0;i<objectsize;i++) {
281 				if (*(object+i) >= 32 && *(object+i) <= 127) {
282 					*(cbuff+i) = *(object+i);
283 				} else {
284 					*(cbuff+i) = '.';
285 				}
286 			}
287 			*(cbuff+i) = '\0';
288 		}
289 	}
290 }
291 
292 void
293 dump_netobj(const struct netobj *nobj)
294 {
295 	char hbuff[MAXBUFFERSIZE*2];
296 	char cbuff[MAXBUFFERSIZE];
297 
298 	if (debug_level < 2) {
299 		return;
300 	}
301 
302 	if (nobj == NULL) {
303 		debuglog("Null netobj pointer\n");
304 	}
305 	else if (nobj->n_len == 0) {
306 		debuglog("Size zero netobj\n");
307 	} else {
308 		dump_static_object(nobj->n_bytes, nobj->n_len,
309 		    hbuff, sizeof(hbuff), cbuff, sizeof(cbuff));
310 		debuglog("netobj: len: %d  data: %s :::  %s\n",
311 		    nobj->n_len, hbuff, cbuff);
312 	}
313 }
314 
315 /* #define DUMP_FILELOCK_VERBOSE */
316 void
317 dump_filelock(const struct file_lock *fl)
318 {
319 #ifdef DUMP_FILELOCK_VERBOSE
320 	char hbuff[MAXBUFFERSIZE*2];
321 	char cbuff[MAXBUFFERSIZE];
322 #endif
323 
324 	if (debug_level < 2) {
325 		return;
326 	}
327 
328 	if (fl != NULL) {
329 		debuglog("Dumping file lock structure @ %p\n", fl);
330 
331 #ifdef DUMP_FILELOCK_VERBOSE
332 		dump_static_object((unsigned char *)&fl->filehandle,
333 		    sizeof(fl->filehandle), hbuff, sizeof(hbuff),
334 		    cbuff, sizeof(cbuff));
335 		debuglog("Filehandle: %8s  :::  %8s\n", hbuff, cbuff);
336 #endif
337 
338 		debuglog("Dumping nlm4_holder:\n"
339 		    "exc: %x  svid: %x  offset:len %llx:%llx\n",
340 		    fl->client.exclusive, fl->client.svid,
341 		    fl->client.l_offset, fl->client.l_len);
342 
343 #ifdef DUMP_FILELOCK_VERBOSE
344 		debuglog("Dumping client identity:\n");
345 		dump_netobj(&fl->client.oh);
346 
347 		debuglog("Dumping client cookie:\n");
348 		dump_netobj(&fl->client_cookie);
349 
350 		debuglog("nsm: %d  status: %d  flags: %d  svid: %x"
351 		    "  client_name: %s\n", fl->nsm_status, fl->status,
352 		    fl->flags, fl->client.svid, fl->client_name);
353 #endif
354 	} else {
355 		debuglog("NULL file lock structure\n");
356 	}
357 }
358 
359 void
360 copy_nlm4_lock_to_nlm4_holder(src, exclusive, dest)
361 	const struct nlm4_lock *src;
362 	const bool_t exclusive;
363 	struct nlm4_holder *dest;
364 {
365 
366 	dest->exclusive = exclusive;
367 	dest->oh.n_len = src->oh.n_len;
368 	dest->oh.n_bytes = src->oh.n_bytes;
369 	dest->svid = src->svid;
370 	dest->l_offset = src->l_offset;
371 	dest->l_len = src->l_len;
372 }
373 
374 
375 size_t
376 strnlen(const char *s, size_t len)
377 {
378     size_t n;
379 
380     for (n = 0;  s[n] != 0 && n < len; n++)
381         ;
382     return n;
383 }
384 
385 /*
386  * allocate_file_lock: Create a lock with the given parameters
387  */
388 
389 struct file_lock *
390 allocate_file_lock(const netobj *lockowner, const netobj *matchcookie,
391 		   const struct sockaddr *addr, const char *caller_name)
392 {
393 	struct file_lock *newfl;
394 	size_t n;
395 
396 	/* Beware of rubbish input! */
397 	n = strnlen(caller_name, SM_MAXSTRLEN);
398 	if (n == SM_MAXSTRLEN) {
399 		return NULL;
400 	}
401 
402 	newfl = malloc(sizeof(*newfl) - sizeof(newfl->client_name) + n + 1);
403 	if (newfl == NULL) {
404 		return NULL;
405 	}
406 	bzero(newfl, sizeof(*newfl) - sizeof(newfl->client_name));
407 	memcpy(newfl->client_name, caller_name, n);
408 	newfl->client_name[n] = 0;
409 
410 	newfl->client.oh.n_bytes = malloc(lockowner->n_len);
411 	if (newfl->client.oh.n_bytes == NULL) {
412 		free(newfl);
413 		return NULL;
414 	}
415 	newfl->client.oh.n_len = lockowner->n_len;
416 	bcopy(lockowner->n_bytes, newfl->client.oh.n_bytes, lockowner->n_len);
417 
418 	newfl->client_cookie.n_bytes = malloc(matchcookie->n_len);
419 	if (newfl->client_cookie.n_bytes == NULL) {
420 		free(newfl->client.oh.n_bytes);
421 		free(newfl);
422 		return NULL;
423 	}
424 	newfl->client_cookie.n_len = matchcookie->n_len;
425 	bcopy(matchcookie->n_bytes, newfl->client_cookie.n_bytes, matchcookie->n_len);
426 
427 	newfl->addr = malloc(addr->sa_len);
428 	if (newfl->addr == NULL) {
429 		free(newfl->client_cookie.n_bytes);
430 		free(newfl->client.oh.n_bytes);
431 		free(newfl);
432 		return NULL;
433 	}
434 	memcpy(newfl->addr, addr, addr->sa_len);
435 
436 	return newfl;
437 }
438 
439 /*
440  * file_file_lock: Force creation of a valid file lock
441  */
442 void
443 fill_file_lock(struct file_lock *fl, const fhandle_t *fh,
444     const bool_t exclusive, const int32_t svid,
445     const u_int64_t offset, const u_int64_t len,
446     const int state, const int status, const int flags, const int blocking)
447 {
448 	bcopy(fh, &fl->filehandle, sizeof(fhandle_t));
449 
450 	fl->client.exclusive = exclusive;
451 	fl->client.svid = svid;
452 	fl->client.l_offset = offset;
453 	fl->client.l_len = len;
454 
455 	fl->nsm_status = state;
456 	fl->status = status;
457 	fl->flags = flags;
458 	fl->blocking = blocking;
459 }
460 
461 /*
462  * deallocate_file_lock: Free all storage associated with a file lock
463  */
464 void
465 deallocate_file_lock(struct file_lock *fl)
466 {
467 	free(fl->addr);
468 	free(fl->client.oh.n_bytes);
469 	free(fl->client_cookie.n_bytes);
470 	free(fl);
471 }
472 
473 /*
474  * regions_overlap(): This function examines the two provided regions for
475  * overlap.
476  */
477 int
478 regions_overlap(start1, len1, start2, len2)
479 	const u_int64_t start1, len1, start2, len2;
480 {
481 	u_int64_t d1,d2,d3,d4;
482 	enum split_status result;
483 
484 	debuglog("Entering region overlap with vals: %llu:%llu--%llu:%llu\n",
485 		 start1, len1, start2, len2);
486 
487 	result = region_compare(start1, len1, start2, len2,
488 	    &d1, &d2, &d3, &d4);
489 
490 	debuglog("Exiting region overlap with val: %d\n",result);
491 
492 	if (result == SPL_DISJOINT) {
493 		return 0;
494 	} else {
495 		return 1;
496 	}
497 }
498 
499 /*
500  * region_compare(): Examine lock regions and split appropriately
501  *
502  * XXX: Fix 64 bit overflow problems
503  * XXX: Check to make sure I got *ALL* the cases.
504  * XXX: This DESPERATELY needs a regression test.
505  */
506 enum split_status
507 region_compare(starte, lene, startu, lenu,
508     start1, len1, start2, len2)
509 	const u_int64_t starte, lene, startu, lenu;
510 	u_int64_t *start1, *len1, *start2, *len2;
511 {
512 	/*
513 	 * Please pay attention to the sequential exclusions
514 	 * of the if statements!!!
515 	 */
516 	enum LFLAGS lflags;
517 	enum RFLAGS rflags;
518 	enum split_status retval;
519 
520 	retval = SPL_DISJOINT;
521 
522 	if (lene == 0 && lenu == 0) {
523 		/* Examine left edge of locker */
524 		lflags = LEDGE_INSIDE;
525 		if (startu < starte) {
526 			lflags = LEDGE_LEFT;
527 		} else if (startu == starte) {
528 			lflags = LEDGE_LBOUNDARY;
529 		}
530 
531 		rflags = REDGE_RBOUNDARY; /* Both are infiinite */
532 
533 		if (lflags == LEDGE_INSIDE) {
534 			*start1 = starte;
535 			*len1 = startu - starte;
536 		}
537 
538 		if (lflags == LEDGE_LEFT || lflags == LEDGE_LBOUNDARY) {
539 			retval = SPL_CONTAINED;
540 		} else {
541 			retval = SPL_LOCK1;
542 		}
543 	} else if (lene == 0 && lenu != 0) {
544 		/* Established lock is infinite */
545 		/* Examine left edge of unlocker */
546 		lflags = LEDGE_INSIDE;
547 		if (startu < starte) {
548 			lflags = LEDGE_LEFT;
549 		} else if (startu == starte) {
550 			lflags = LEDGE_LBOUNDARY;
551 		}
552 
553 		/* Examine right edge of unlocker */
554 		if (startu + lenu < starte) {
555 			/* Right edge of unlocker left of established lock */
556 			rflags = REDGE_LEFT;
557 			return SPL_DISJOINT;
558 		} else if (startu + lenu == starte) {
559 			/* Right edge of unlocker on start of established lock */
560 			rflags = REDGE_LBOUNDARY;
561 			return SPL_DISJOINT;
562 		} else { /* Infinifty is right of finity */
563 			/* Right edge of unlocker inside established lock */
564 			rflags = REDGE_INSIDE;
565 		}
566 
567 		if (lflags == LEDGE_INSIDE) {
568 			*start1 = starte;
569 			*len1 = startu - starte;
570 			retval |= SPL_LOCK1;
571 		}
572 
573 		if (rflags == REDGE_INSIDE) {
574 			/* Create right lock */
575 			*start2 = startu+lenu;
576 			*len2 = 0;
577 			retval |= SPL_LOCK2;
578 		}
579 	} else if (lene != 0 && lenu == 0) {
580 		/* Unlocker is infinite */
581 		/* Examine left edge of unlocker */
582 		lflags = LEDGE_RIGHT;
583 		if (startu < starte) {
584 			lflags = LEDGE_LEFT;
585 			retval = SPL_CONTAINED;
586 			return retval;
587 		} else if (startu == starte) {
588 			lflags = LEDGE_LBOUNDARY;
589 			retval = SPL_CONTAINED;
590 			return retval;
591 		} else if ((startu > starte) && (startu < starte + lene - 1)) {
592 			lflags = LEDGE_INSIDE;
593 		} else if (startu == starte + lene - 1) {
594 			lflags = LEDGE_RBOUNDARY;
595 		} else { /* startu > starte + lene -1 */
596 			lflags = LEDGE_RIGHT;
597 			return SPL_DISJOINT;
598 		}
599 
600 		rflags = REDGE_RIGHT; /* Infinity is right of finity */
601 
602 		if (lflags == LEDGE_INSIDE || lflags == LEDGE_RBOUNDARY) {
603 			*start1 = starte;
604 			*len1 = startu - starte;
605 			retval |= SPL_LOCK1;
606 			return retval;
607 		}
608 	} else {
609 		/* Both locks are finite */
610 
611 		/* Examine left edge of unlocker */
612 		lflags = LEDGE_RIGHT;
613 		if (startu < starte) {
614 			lflags = LEDGE_LEFT;
615 		} else if (startu == starte) {
616 			lflags = LEDGE_LBOUNDARY;
617 		} else if ((startu > starte) && (startu < starte + lene - 1)) {
618 			lflags = LEDGE_INSIDE;
619 		} else if (startu == starte + lene - 1) {
620 			lflags = LEDGE_RBOUNDARY;
621 		} else { /* startu > starte + lene -1 */
622 			lflags = LEDGE_RIGHT;
623 			return SPL_DISJOINT;
624 		}
625 
626 		/* Examine right edge of unlocker */
627 		if (startu + lenu < starte) {
628 			/* Right edge of unlocker left of established lock */
629 			rflags = REDGE_LEFT;
630 			return SPL_DISJOINT;
631 		} else if (startu + lenu == starte) {
632 			/* Right edge of unlocker on start of established lock */
633 			rflags = REDGE_LBOUNDARY;
634 			return SPL_DISJOINT;
635 		} else if (startu + lenu < starte + lene) {
636 			/* Right edge of unlocker inside established lock */
637 			rflags = REDGE_INSIDE;
638 		} else if (startu + lenu == starte + lene) {
639 			/* Right edge of unlocker on right edge of established lock */
640 			rflags = REDGE_RBOUNDARY;
641 		} else { /* startu + lenu > starte + lene */
642 			/* Right edge of unlocker is right of established lock */
643 			rflags = REDGE_RIGHT;
644 		}
645 
646 		if (lflags == LEDGE_INSIDE || lflags == LEDGE_RBOUNDARY) {
647 			/* Create left lock */
648 			*start1 = starte;
649 			*len1 = (startu - starte);
650 			retval |= SPL_LOCK1;
651 		}
652 
653 		if (rflags == REDGE_INSIDE) {
654 			/* Create right lock */
655 			*start2 = startu+lenu;
656 			*len2 = starte+lene-(startu+lenu);
657 			retval |= SPL_LOCK2;
658 		}
659 
660 		if ((lflags == LEDGE_LEFT || lflags == LEDGE_LBOUNDARY) &&
661 		    (rflags == REDGE_RBOUNDARY || rflags == REDGE_RIGHT)) {
662 			retval = SPL_CONTAINED;
663 		}
664 	}
665 	return retval;
666 }
667 
668 /*
669  * same_netobj: Compares the apprpriate bits of a netobj for identity
670  */
671 int
672 same_netobj(const netobj *n0, const netobj *n1)
673 {
674 	int retval;
675 
676 	retval = 0;
677 
678 	debuglog("Entering netobj identity check\n");
679 
680 	if (n0->n_len == n1->n_len) {
681 		debuglog("Preliminary length check passed\n");
682 		retval = !bcmp(n0->n_bytes, n1->n_bytes, n0->n_len);
683 		debuglog("netobj %smatch\n", retval ? "" : "mis");
684 	}
685 
686 	return (retval);
687 }
688 
689 /*
690  * same_filelock_identity: Compares the appropriate bits of a file_lock
691  */
692 int
693 same_filelock_identity(fl0, fl1)
694 	const struct file_lock *fl0, *fl1;
695 {
696 	int retval;
697 
698 	retval = 0;
699 
700 	debuglog("Checking filelock identity\n");
701 
702 	/*
703 	 * Check process ids and host information.
704 	 */
705 	retval = (fl0->client.svid == fl1->client.svid &&
706 	    same_netobj(&(fl0->client.oh), &(fl1->client.oh)));
707 
708 	debuglog("Exiting checking filelock identity: retval: %d\n",retval);
709 
710 	return (retval);
711 }
712 
713 /*
714  * Below here are routines associated with manipulating the NFS
715  * lock list.
716  */
717 
718 /*
719  * get_lock_matching_unlock: Return a lock which matches the given unlock lock
720  *                           or NULL otehrwise
721  * XXX: It is a shame that this duplicates so much code from test_nfslock.
722  */
723 struct file_lock *
724 get_lock_matching_unlock(const struct file_lock *fl)
725 {
726 	struct file_lock *ifl; /* Iterator */
727 
728 	debuglog("Entering get_lock_matching_unlock\n");
729 	debuglog("********Dump of fl*****************\n");
730 	dump_filelock(fl);
731 
732 	LIST_FOREACH(ifl, &nfslocklist_head, nfslocklist) {
733 		debuglog("Pointer to file lock: %p\n",ifl);
734 
735 		debuglog("****Dump of ifl****\n");
736 		dump_filelock(ifl);
737 		debuglog("*******************\n");
738 
739 		/*
740 		 * XXX: It is conceivable that someone could use the NLM RPC
741 		 * system to directly access filehandles.  This may be a
742 		 * security hazard as the filehandle code may bypass normal
743 		 * file access controls
744 		 */
745 		if (bcmp(&fl->filehandle, &ifl->filehandle, sizeof(fhandle_t)))
746 			continue;
747 
748 		debuglog("get_lock_matching_unlock: Filehandles match, "
749 		    "checking regions\n");
750 
751 		/* Filehandles match, check for region overlap */
752 		if (!regions_overlap(fl->client.l_offset, fl->client.l_len,
753 			ifl->client.l_offset, ifl->client.l_len))
754 			continue;
755 
756 		debuglog("get_lock_matching_unlock: Region overlap"
757 		    " found %llu : %llu -- %llu : %llu\n",
758 		    fl->client.l_offset,fl->client.l_len,
759 		    ifl->client.l_offset,ifl->client.l_len);
760 
761 		/* Regions overlap, check the identity */
762 		if (!same_filelock_identity(fl,ifl))
763 			continue;
764 
765 		debuglog("get_lock_matching_unlock: Duplicate lock id.  Granting\n");
766 		return (ifl);
767 	}
768 
769 	debuglog("Exiting bet_lock_matching_unlock\n");
770 
771 	return (NULL);
772 }
773 
774 /*
775  * test_nfslock: check for NFS lock in lock list
776  *
777  * This routine makes the following assumptions:
778  *    1) Nothing will adjust the lock list during a lookup
779  *
780  * This routine has an intersting quirk which bit me hard.
781  * The conflicting_fl is the pointer to the conflicting lock.
782  * However, to modify the "*pointer* to the conflicting lock" rather
783  * that the "conflicting lock itself" one must pass in a "pointer to
784  * the pointer of the conflicting lock".  Gross.
785  */
786 
787 enum nfslock_status
788 test_nfslock(const struct file_lock *fl, struct file_lock **conflicting_fl)
789 {
790 	struct file_lock *ifl; /* Iterator */
791 	enum nfslock_status retval;
792 
793 	debuglog("Entering test_nfslock\n");
794 
795 	retval = NFS_GRANTED;
796 	(*conflicting_fl) = NULL;
797 
798 	debuglog("Entering lock search loop\n");
799 
800 	debuglog("***********************************\n");
801 	debuglog("Dumping match filelock\n");
802 	debuglog("***********************************\n");
803 	dump_filelock(fl);
804 	debuglog("***********************************\n");
805 
806 	LIST_FOREACH(ifl, &nfslocklist_head, nfslocklist) {
807 		if (retval == NFS_DENIED)
808 			break;
809 
810 		debuglog("Top of lock loop\n");
811 		debuglog("Pointer to file lock: %p\n",ifl);
812 
813 		debuglog("***********************************\n");
814 		debuglog("Dumping test filelock\n");
815 		debuglog("***********************************\n");
816 		dump_filelock(ifl);
817 		debuglog("***********************************\n");
818 
819 		/*
820 		 * XXX: It is conceivable that someone could use the NLM RPC
821 		 * system to directly access filehandles.  This may be a
822 		 * security hazard as the filehandle code may bypass normal
823 		 * file access controls
824 		 */
825 		if (bcmp(&fl->filehandle, &ifl->filehandle, sizeof(fhandle_t)))
826 			continue;
827 
828 		debuglog("test_nfslock: filehandle match found\n");
829 
830 		/* Filehandles match, check for region overlap */
831 		if (!regions_overlap(fl->client.l_offset, fl->client.l_len,
832 			ifl->client.l_offset, ifl->client.l_len))
833 			continue;
834 
835 		debuglog("test_nfslock: Region overlap found"
836 		    " %llu : %llu -- %llu : %llu\n",
837 		    fl->client.l_offset,fl->client.l_len,
838 		    ifl->client.l_offset,ifl->client.l_len);
839 
840 		/* Regions overlap, check the exclusivity */
841 		if (!(fl->client.exclusive || ifl->client.exclusive))
842 			continue;
843 
844 		debuglog("test_nfslock: Exclusivity failure: %d %d\n",
845 		    fl->client.exclusive,
846 		    ifl->client.exclusive);
847 
848 		if (same_filelock_identity(fl,ifl)) {
849 			debuglog("test_nfslock: Duplicate id.  Granting\n");
850 			(*conflicting_fl) = ifl;
851 			retval = NFS_GRANTED_DUPLICATE;
852 		} else {
853 			/* locking attempt fails */
854 			debuglog("test_nfslock: Lock attempt failed\n");
855 			debuglog("Desired lock\n");
856 			dump_filelock(fl);
857 			debuglog("Conflicting lock\n");
858 			dump_filelock(ifl);
859 			(*conflicting_fl) = ifl;
860 			retval = NFS_DENIED;
861 		}
862 	}
863 
864 	debuglog("Dumping file locks\n");
865 	debuglog("Exiting test_nfslock\n");
866 
867 	return (retval);
868 }
869 
870 /*
871  * lock_nfslock: attempt to create a lock in the NFS lock list
872  *
873  * This routine tests whether the lock will be granted and then adds
874  * the entry to the lock list if so.
875  *
876  * Argument fl gets modified as its list housekeeping entries get modified
877  * upon insertion into the NFS lock list
878  *
879  * This routine makes several assumptions:
880  *    1) It is perfectly happy to grant a duplicate lock from the same pid.
881  *       While this seems to be intuitively wrong, it is required for proper
882  *       Posix semantics during unlock.  It is absolutely imperative to not
883  *       unlock the main lock before the two child locks are established. Thus,
884  *       one has be be able to create duplicate locks over an existing lock
885  *    2) It currently accepts duplicate locks from the same id,pid
886  */
887 
888 enum nfslock_status
889 lock_nfslock(struct file_lock *fl)
890 {
891 	enum nfslock_status retval;
892 	struct file_lock *dummy_fl;
893 
894 	dummy_fl = NULL;
895 
896 	debuglog("Entering lock_nfslock...\n");
897 
898 	retval = test_nfslock(fl,&dummy_fl);
899 
900 	if (retval == NFS_GRANTED || retval == NFS_GRANTED_DUPLICATE) {
901 		debuglog("Inserting lock...\n");
902 		dump_filelock(fl);
903 		LIST_INSERT_HEAD(&nfslocklist_head, fl, nfslocklist);
904 	}
905 
906 	debuglog("Exiting lock_nfslock...\n");
907 
908 	return (retval);
909 }
910 
911 /*
912  * delete_nfslock: delete an NFS lock list entry
913  *
914  * This routine is used to delete a lock out of the NFS lock list
915  * without regard to status, underlying locks, regions or anything else
916  *
917  * Note that this routine *does not deallocate memory* of the lock.
918  * It just disconnects it from the list.  The lock can then be used
919  * by other routines without fear of trashing the list.
920  */
921 
922 enum nfslock_status
923 delete_nfslock(struct file_lock *fl)
924 {
925 
926 	LIST_REMOVE(fl, nfslocklist);
927 
928 	return (NFS_GRANTED);
929 }
930 
931 enum split_status
932 split_nfslock(exist_lock, unlock_lock, left_lock, right_lock)
933 	const struct file_lock *exist_lock, *unlock_lock;
934 	struct file_lock **left_lock, **right_lock;
935 {
936 	u_int64_t start1, len1, start2, len2;
937 	enum split_status spstatus;
938 
939 	spstatus = region_compare(exist_lock->client.l_offset, exist_lock->client.l_len,
940 	    unlock_lock->client.l_offset, unlock_lock->client.l_len,
941 	    &start1, &len1, &start2, &len2);
942 
943 	if ((spstatus & SPL_LOCK1) != 0) {
944 		*left_lock = allocate_file_lock(&exist_lock->client.oh, &exist_lock->client_cookie, exist_lock->addr, exist_lock->client_name);
945 		if (*left_lock == NULL) {
946 			debuglog("Unable to allocate resource for split 1\n");
947 			return SPL_RESERR;
948 		}
949 
950 		fill_file_lock(*left_lock, &exist_lock->filehandle,
951 		    exist_lock->client.exclusive, exist_lock->client.svid,
952 		    start1, len1,
953 		    exist_lock->nsm_status,
954 		    exist_lock->status, exist_lock->flags, exist_lock->blocking);
955 	}
956 
957 	if ((spstatus & SPL_LOCK2) != 0) {
958 		*right_lock = allocate_file_lock(&exist_lock->client.oh, &exist_lock->client_cookie, exist_lock->addr, exist_lock->client_name);
959 		if (*right_lock == NULL) {
960 			debuglog("Unable to allocate resource for split 1\n");
961 			if (*left_lock != NULL) {
962 				deallocate_file_lock(*left_lock);
963 			}
964 			return SPL_RESERR;
965 		}
966 
967 		fill_file_lock(*right_lock, &exist_lock->filehandle,
968 		    exist_lock->client.exclusive, exist_lock->client.svid,
969 		    start2, len2,
970 		    exist_lock->nsm_status,
971 		    exist_lock->status, exist_lock->flags, exist_lock->blocking);
972 	}
973 
974 	return spstatus;
975 }
976 
977 enum nfslock_status
978 unlock_nfslock(fl, released_lock, left_lock, right_lock)
979 	const struct file_lock *fl;
980 	struct file_lock **released_lock;
981 	struct file_lock **left_lock;
982 	struct file_lock **right_lock;
983 {
984 	struct file_lock *mfl; /* Matching file lock */
985 	enum nfslock_status retval;
986 	enum split_status spstatus;
987 
988 	debuglog("Entering unlock_nfslock\n");
989 
990 	*released_lock = NULL;
991 	*left_lock = NULL;
992 	*right_lock = NULL;
993 
994 	retval = NFS_DENIED_NOLOCK;
995 
996 	debuglog("Attempting to match lock...\n");
997 	mfl = get_lock_matching_unlock(fl);
998 
999 	if (mfl != NULL) {
1000 		debuglog("Unlock matched.  Querying for split\n");
1001 
1002 		spstatus = split_nfslock(mfl, fl, left_lock, right_lock);
1003 
1004 		debuglog("Split returned %d %p %p %p %p\n",spstatus,mfl,fl,*left_lock,*right_lock);
1005 		debuglog("********Split dumps********");
1006 		dump_filelock(mfl);
1007 		dump_filelock(fl);
1008 		dump_filelock(*left_lock);
1009 		dump_filelock(*right_lock);
1010 		debuglog("********End Split dumps********");
1011 
1012 		if (spstatus == SPL_RESERR) {
1013 			if (*left_lock != NULL) {
1014 				deallocate_file_lock(*left_lock);
1015 				*left_lock = NULL;
1016 			}
1017 
1018 			if (*right_lock != NULL) {
1019 				deallocate_file_lock(*right_lock);
1020 				*right_lock = NULL;
1021 			}
1022 
1023 			return NFS_RESERR;
1024 		}
1025 
1026 		/* Insert new locks from split if required */
1027 		if (*left_lock != NULL) {
1028 			debuglog("Split left activated\n");
1029 			LIST_INSERT_HEAD(&nfslocklist_head, *left_lock, nfslocklist);
1030 		}
1031 
1032 		if (*right_lock != NULL) {
1033 			debuglog("Split right activated\n");
1034 			LIST_INSERT_HEAD(&nfslocklist_head, *right_lock, nfslocklist);
1035 		}
1036 
1037 		/* Unlock the lock since it matches identity */
1038 		LIST_REMOVE(mfl, nfslocklist);
1039 		*released_lock = mfl;
1040 		retval = NFS_GRANTED;
1041 	}
1042 
1043 	debuglog("Exiting unlock_nfslock\n");
1044 
1045 	return retval;
1046 }
1047 
1048 /*
1049  * Below here are the routines for manipulating the file lock directly
1050  * on the disk hardware itself
1051  */
1052 enum hwlock_status
1053 lock_hwlock(struct file_lock *fl)
1054 {
1055 	struct monfile *imf,*nmf;
1056 	int lflags, flerror;
1057 
1058 	/* Scan to see if filehandle already present */
1059 	LIST_FOREACH(imf, &monfilelist_head, monfilelist) {
1060 		if (bcmp(&fl->filehandle, &imf->filehandle,
1061 			sizeof(fl->filehandle)) == 0) {
1062 			/* imf is the correct filehandle */
1063 			break;
1064 		}
1065 	}
1066 
1067 	/*
1068 	 * Filehandle already exists (we control the file)
1069 	 * *AND* NFS has already cleared the lock for availability
1070 	 * Grant it and bump the refcount.
1071 	 */
1072 	if (imf != NULL) {
1073 		++(imf->refcount);
1074 		return (HW_GRANTED);
1075 	}
1076 
1077 	/* No filehandle found, create and go */
1078 	nmf = malloc(sizeof(struct monfile));
1079 	if (nmf == NULL) {
1080 		debuglog("hwlock resource allocation failure\n");
1081 		return (HW_RESERR);
1082 	}
1083 
1084 	/* XXX: Is O_RDWR always the correct mode? */
1085 	nmf->fd = fhopen(&fl->filehandle, O_RDWR);
1086 	if (nmf->fd < 0) {
1087 		debuglog("fhopen failed (from %16s): %32s\n",
1088 		    fl->client_name, strerror(errno));
1089 		free(nmf);
1090 		switch (errno) {
1091 		case ESTALE:
1092 			return (HW_STALEFH);
1093 		case EROFS:
1094 			return (HW_READONLY);
1095 		default:
1096 			return (HW_RESERR);
1097 		}
1098 	}
1099 
1100 	/* File opened correctly, fill the monitor struct */
1101 	bcopy(&fl->filehandle, &nmf->filehandle, sizeof(fl->filehandle));
1102 	nmf->refcount = 1;
1103 	nmf->exclusive = fl->client.exclusive;
1104 
1105 	lflags = (nmf->exclusive == 1) ?
1106 	    (LOCK_EX | LOCK_NB) : (LOCK_SH | LOCK_NB);
1107 
1108 	flerror = flock(nmf->fd, lflags);
1109 
1110 	if (flerror != 0) {
1111 		debuglog("flock failed (from %16s): %32s\n",
1112 		    fl->client_name, strerror(errno));
1113 		close(nmf->fd);
1114 		free(nmf);
1115 		switch (errno) {
1116 		case EAGAIN:
1117 			return (HW_DENIED);
1118 		case ESTALE:
1119 			return (HW_STALEFH);
1120 		case EROFS:
1121 			return (HW_READONLY);
1122 		default:
1123 			return (HW_RESERR);
1124 			break;
1125 		}
1126 	}
1127 
1128 	/* File opened and locked */
1129 	LIST_INSERT_HEAD(&monfilelist_head, nmf, monfilelist);
1130 
1131 	debuglog("flock succeeded (from %16s)\n", fl->client_name);
1132 	return (HW_GRANTED);
1133 }
1134 
1135 enum hwlock_status
1136 unlock_hwlock(const struct file_lock *fl)
1137 {
1138 	struct monfile *imf;
1139 
1140 	debuglog("Entering unlock_hwlock\n");
1141 	debuglog("Entering loop interation\n");
1142 
1143 	/* Scan to see if filehandle already present */
1144 	LIST_FOREACH(imf, &monfilelist_head, monfilelist) {
1145 		if (bcmp(&fl->filehandle, &imf->filehandle,
1146 			sizeof(fl->filehandle)) == 0) {
1147 			/* imf is the correct filehandle */
1148 			break;
1149 		}
1150 	}
1151 
1152 	debuglog("Completed iteration.  Proceeding\n");
1153 
1154 	if (imf == NULL) {
1155 		/* No lock found */
1156 		debuglog("Exiting unlock_hwlock (HW_DENIED_NOLOCK)\n");
1157 		return (HW_DENIED_NOLOCK);
1158 	}
1159 
1160 	/* Lock found */
1161 	--imf->refcount;
1162 
1163 	if (imf->refcount < 0) {
1164 		debuglog("Negative hardware reference count\n");
1165 	}
1166 
1167 	if (imf->refcount <= 0) {
1168 		close(imf->fd);
1169 		LIST_REMOVE(imf, monfilelist);
1170 		free(imf);
1171 	}
1172 	debuglog("Exiting unlock_hwlock (HW_GRANTED)\n");
1173 	return (HW_GRANTED);
1174 }
1175 
1176 enum hwlock_status
1177 test_hwlock(fl, conflicting_fl)
1178 	const struct file_lock *fl __unused;
1179 	struct file_lock **conflicting_fl __unused;
1180 {
1181 
1182 	/*
1183 	 * XXX: lock tests on hardware are not required until
1184 	 * true partial file testing is done on the underlying file
1185 	 */
1186 	return (HW_RESERR);
1187 }
1188 
1189 
1190 
1191 /*
1192  * Below here are routines for manipulating blocked lock requests
1193  * They should only be called from the XXX_partialfilelock routines
1194  * if at all possible
1195  */
1196 
1197 int
1198 duplicate_block(struct file_lock *fl)
1199 {
1200 	struct file_lock *ifl;
1201 	int retval = 0;
1202 
1203 	debuglog("Entering duplicate_block");
1204 
1205 	/*
1206 	 * Is this lock request already on the blocking list?
1207 	 * Consider it a dupe if the file handles, offset, length,
1208 	 * exclusivity and client match.
1209 	 */
1210 	LIST_FOREACH(ifl, &blockedlocklist_head, nfslocklist) {
1211 		if (!bcmp(&fl->filehandle, &ifl->filehandle,
1212 			sizeof(fhandle_t)) &&
1213 		    fl->client.exclusive == ifl->client.exclusive &&
1214 		    fl->client.l_offset == ifl->client.l_offset &&
1215 		    fl->client.l_len == ifl->client.l_len &&
1216 		    same_filelock_identity(fl, ifl)) {
1217 			retval = 1;
1218 			break;
1219 		}
1220 	}
1221 
1222 	debuglog("Exiting duplicate_block: %s\n", retval ? "already blocked"
1223 	    : "not already blocked");
1224 	return retval;
1225 }
1226 
1227 void
1228 add_blockingfilelock(struct file_lock *fl)
1229 {
1230 	debuglog("Entering add_blockingfilelock\n");
1231 
1232 	/*
1233 	 * A blocking lock request _should_ never be duplicated as a client
1234 	 * that is already blocked shouldn't be able to request another
1235 	 * lock. Alas, there are some buggy clients that do request the same
1236 	 * lock repeatedly. Make sure only unique locks are on the blocked
1237 	 * lock list.
1238 	 */
1239 	if (duplicate_block(fl)) {
1240 		debuglog("Exiting add_blockingfilelock: already blocked\n");
1241 		return;
1242 	}
1243 
1244 	/*
1245 	 * Clear the blocking flag so that it can be reused without
1246 	 * adding it to the blocking queue a second time
1247 	 */
1248 
1249 	fl->blocking = 0;
1250 	LIST_INSERT_HEAD(&blockedlocklist_head, fl, nfslocklist);
1251 
1252 	debuglog("Exiting add_blockingfilelock: added blocked lock\n");
1253 }
1254 
1255 void
1256 remove_blockingfilelock(struct file_lock *fl)
1257 {
1258 
1259 	debuglog("Entering remove_blockingfilelock\n");
1260 
1261 	LIST_REMOVE(fl, nfslocklist);
1262 
1263 	debuglog("Exiting remove_blockingfilelock\n");
1264 }
1265 
1266 void
1267 clear_blockingfilelock(const char *hostname)
1268 {
1269 	struct file_lock *ifl,*nfl;
1270 
1271 	/*
1272 	 * Normally, LIST_FOREACH is called for, but since
1273 	 * the current element *is* the iterator, deleting it
1274 	 * would mess up the iteration.  Thus, a next element
1275 	 * must be used explicitly
1276 	 */
1277 
1278 	ifl = LIST_FIRST(&blockedlocklist_head);
1279 
1280 	while (ifl != NULL) {
1281 		nfl = LIST_NEXT(ifl, nfslocklist);
1282 
1283 		if (strncmp(hostname, ifl->client_name, SM_MAXSTRLEN) == 0) {
1284 			remove_blockingfilelock(ifl);
1285 			deallocate_file_lock(ifl);
1286 		}
1287 
1288 		ifl = nfl;
1289 	}
1290 }
1291 
1292 void
1293 retry_blockingfilelocklist(void)
1294 {
1295 	/* Retry all locks in the blocked list */
1296 	struct file_lock *ifl, *nfl; /* Iterator */
1297 	enum partialfilelock_status pflstatus;
1298 
1299 	debuglog("Entering retry_blockingfilelocklist\n");
1300 
1301 	LIST_FOREACH_SAFE(ifl, &blockedlocklist_head, nfslocklist, nfl) {
1302 		debuglog("Iterator choice %p\n",ifl);
1303 		debuglog("Next iterator choice %p\n",nfl);
1304 
1305 		/*
1306 		 * SUBTLE BUG: The file_lock must be removed from the
1307 		 * old list so that it's list pointers get disconnected
1308 		 * before being allowed to participate in the new list
1309 		 * which will automatically add it in if necessary.
1310 		 */
1311 
1312 		LIST_REMOVE(ifl, nfslocklist);
1313 		pflstatus = lock_partialfilelock(ifl);
1314 
1315 		if (pflstatus == PFL_GRANTED || pflstatus == PFL_GRANTED_DUPLICATE) {
1316 			debuglog("Granted blocked lock\n");
1317 			/* lock granted and is now being used */
1318 			send_granted(ifl,0);
1319 		} else {
1320 			/* Reinsert lock back into blocked list */
1321 			debuglog("Replacing blocked lock\n");
1322 			LIST_INSERT_HEAD(&blockedlocklist_head, ifl, nfslocklist);
1323 		}
1324 	}
1325 
1326 	debuglog("Exiting retry_blockingfilelocklist\n");
1327 }
1328 
1329 /*
1330  * Below here are routines associated with manipulating all
1331  * aspects of the partial file locking system (list, hardware, etc.)
1332  */
1333 
1334 /*
1335  * Please note that lock monitoring must be done at this level which
1336  * keeps track of *individual* lock requests on lock and unlock
1337  *
1338  * XXX: Split unlocking is going to make the unlock code miserable
1339  */
1340 
1341 /*
1342  * lock_partialfilelock:
1343  *
1344  * Argument fl gets modified as its list housekeeping entries get modified
1345  * upon insertion into the NFS lock list
1346  *
1347  * This routine makes several assumptions:
1348  * 1) It (will) pass locks through to flock to lock the entire underlying file
1349  *     and then parcel out NFS locks if it gets control of the file.
1350  *         This matches the old rpc.lockd file semantics (except where it
1351  *         is now more correct).  It is the safe solution, but will cause
1352  *         overly restrictive blocking if someone is trying to use the
1353  *         underlying files without using NFS.  This appears to be an
1354  *         acceptable tradeoff since most people use standalone NFS servers.
1355  * XXX: The right solution is probably kevent combined with fcntl
1356  *
1357  *    2) Nothing modifies the lock lists between testing and granting
1358  *           I have no idea whether this is a useful assumption or not
1359  */
1360 
1361 enum partialfilelock_status
1362 lock_partialfilelock(struct file_lock *fl)
1363 {
1364 	enum partialfilelock_status retval;
1365 	enum nfslock_status lnlstatus;
1366 	enum hwlock_status hwstatus;
1367 
1368 	debuglog("Entering lock_partialfilelock\n");
1369 
1370 	retval = PFL_DENIED;
1371 
1372 	/*
1373 	 * Execute the NFS lock first, if possible, as it is significantly
1374 	 * easier and less expensive to undo than the filesystem lock
1375 	 */
1376 
1377 	lnlstatus = lock_nfslock(fl);
1378 
1379 	switch (lnlstatus) {
1380 	case NFS_GRANTED:
1381 	case NFS_GRANTED_DUPLICATE:
1382 		/*
1383 		 * At this point, the NFS lock is allocated and active.
1384 		 * Remember to clean it up if the hardware lock fails
1385 		 */
1386 		hwstatus = lock_hwlock(fl);
1387 
1388 		switch (hwstatus) {
1389 		case HW_GRANTED:
1390 		case HW_GRANTED_DUPLICATE:
1391 			debuglog("HW GRANTED\n");
1392 			/*
1393 			 * XXX: Fixme: Check hwstatus for duplicate when
1394 			 * true partial file locking and accounting is
1395 			 * done on the hardware.
1396 			 */
1397 			if (lnlstatus == NFS_GRANTED_DUPLICATE) {
1398 				retval = PFL_GRANTED_DUPLICATE;
1399 			} else {
1400 				retval = PFL_GRANTED;
1401 			}
1402 			monitor_lock_host(fl->client_name);
1403 			break;
1404 		case HW_RESERR:
1405 			debuglog("HW RESERR\n");
1406 			retval = PFL_HWRESERR;
1407 			break;
1408 		case HW_DENIED:
1409 			debuglog("HW DENIED\n");
1410 			retval = PFL_HWDENIED;
1411 			break;
1412 		default:
1413 			debuglog("Unmatched hwstatus %d\n",hwstatus);
1414 			break;
1415 		}
1416 
1417 		if (retval != PFL_GRANTED &&
1418 		    retval != PFL_GRANTED_DUPLICATE) {
1419 			/* Clean up the NFS lock */
1420 			debuglog("Deleting trial NFS lock\n");
1421 			delete_nfslock(fl);
1422 		}
1423 		break;
1424 	case NFS_DENIED:
1425 		retval = PFL_NFSDENIED;
1426 		break;
1427 	case NFS_RESERR:
1428 		retval = PFL_NFSRESERR;
1429 	default:
1430 		debuglog("Unmatched lnlstatus %d\n");
1431 		retval = PFL_NFSDENIED_NOLOCK;
1432 		break;
1433 	}
1434 
1435 	/*
1436 	 * By the time fl reaches here, it is completely free again on
1437 	 * failure.  The NFS lock done before attempting the
1438 	 * hardware lock has been backed out
1439 	 */
1440 
1441 	if (retval == PFL_NFSDENIED || retval == PFL_HWDENIED) {
1442 		/* Once last chance to check the lock */
1443 		if (fl->blocking == 1) {
1444 			if (retval == PFL_NFSDENIED) {
1445 				/* Queue the lock */
1446 				debuglog("BLOCKING LOCK RECEIVED\n");
1447 				retval = PFL_NFSBLOCKED;
1448 				add_blockingfilelock(fl);
1449 				dump_filelock(fl);
1450 			} else {
1451 				/* retval is okay as PFL_HWDENIED */
1452 				debuglog("BLOCKING LOCK DENIED IN HARDWARE\n");
1453 				dump_filelock(fl);
1454 			}
1455 		} else {
1456 			/* Leave retval alone, it's already correct */
1457 			debuglog("Lock denied.  Non-blocking failure\n");
1458 			dump_filelock(fl);
1459 		}
1460 	}
1461 
1462 	debuglog("Exiting lock_partialfilelock\n");
1463 
1464 	return retval;
1465 }
1466 
1467 /*
1468  * unlock_partialfilelock:
1469  *
1470  * Given a file_lock, unlock all locks which match.
1471  *
1472  * Note that a given lock might have to unlock ITSELF!  See
1473  * clear_partialfilelock for example.
1474  */
1475 
1476 enum partialfilelock_status
1477 unlock_partialfilelock(const struct file_lock *fl)
1478 {
1479 	struct file_lock *lfl,*rfl,*releasedfl,*selffl;
1480 	enum partialfilelock_status retval;
1481 	enum nfslock_status unlstatus;
1482 	enum hwlock_status unlhwstatus, lhwstatus;
1483 
1484 	debuglog("Entering unlock_partialfilelock\n");
1485 
1486 	selffl = NULL;
1487 	lfl = NULL;
1488 	rfl = NULL;
1489 	releasedfl = NULL;
1490 	retval = PFL_DENIED;
1491 
1492 	/*
1493 	 * There are significant overlap and atomicity issues
1494 	 * with partially releasing a lock.  For example, releasing
1495 	 * part of an NFS shared lock does *not* always release the
1496 	 * corresponding part of the file since there is only one
1497 	 * rpc.lockd UID but multiple users could be requesting it
1498 	 * from NFS.  Also, an unlock request should never allow
1499 	 * another process to gain a lock on the remaining parts.
1500 	 * ie. Always apply the new locks before releasing the
1501 	 * old one
1502 	 */
1503 
1504 	/*
1505 	 * Loop is required since multiple little locks
1506 	 * can be allocated and then deallocated with one
1507 	 * big unlock.
1508 	 *
1509 	 * The loop is required to be here so that the nfs &
1510 	 * hw subsystems do not need to communicate with one
1511 	 * one another
1512 	 */
1513 
1514 	do {
1515 		debuglog("Value of releasedfl: %p\n",releasedfl);
1516 		/* lfl&rfl are created *AND* placed into the NFS lock list if required */
1517 		unlstatus = unlock_nfslock(fl, &releasedfl, &lfl, &rfl);
1518 		debuglog("Value of releasedfl: %p\n",releasedfl);
1519 
1520 
1521 		/* XXX: This is grungy.  It should be refactored to be cleaner */
1522 		if (lfl != NULL) {
1523 			lhwstatus = lock_hwlock(lfl);
1524 			if (lhwstatus != HW_GRANTED &&
1525 			    lhwstatus != HW_GRANTED_DUPLICATE) {
1526 				debuglog("HW duplicate lock failure for left split\n");
1527 			}
1528 			monitor_lock_host(lfl->client_name);
1529 		}
1530 
1531 		if (rfl != NULL) {
1532 			lhwstatus = lock_hwlock(rfl);
1533 			if (lhwstatus != HW_GRANTED &&
1534 			    lhwstatus != HW_GRANTED_DUPLICATE) {
1535 				debuglog("HW duplicate lock failure for right split\n");
1536 			}
1537 			monitor_lock_host(rfl->client_name);
1538 		}
1539 
1540 		switch (unlstatus) {
1541 		case NFS_GRANTED:
1542 			/* Attempt to unlock on the hardware */
1543 			debuglog("NFS unlock granted.  Attempting hardware unlock\n");
1544 
1545 			/* This call *MUST NOT* unlock the two newly allocated locks */
1546 			unlhwstatus = unlock_hwlock(fl);
1547 			debuglog("HW unlock returned with code %d\n",unlhwstatus);
1548 
1549 			switch (unlhwstatus) {
1550 			case HW_GRANTED:
1551 				debuglog("HW unlock granted\n");
1552 				unmonitor_lock_host(releasedfl->client_name);
1553 				retval = PFL_GRANTED;
1554 				break;
1555 			case HW_DENIED_NOLOCK:
1556 				/* Huh?!?!  This shouldn't happen */
1557 				debuglog("HW unlock denied no lock\n");
1558 				retval = PFL_HWRESERR;
1559 				/* Break out of do-while */
1560 				unlstatus = NFS_RESERR;
1561 				break;
1562 			default:
1563 				debuglog("HW unlock failed\n");
1564 				retval = PFL_HWRESERR;
1565 				/* Break out of do-while */
1566 				unlstatus = NFS_RESERR;
1567 				break;
1568 			}
1569 
1570 			debuglog("Exiting with status retval: %d\n",retval);
1571 
1572 			retry_blockingfilelocklist();
1573 			break;
1574 		case NFS_DENIED_NOLOCK:
1575 			retval = PFL_GRANTED;
1576 			debuglog("All locks cleaned out\n");
1577 			break;
1578 		default:
1579 			retval = PFL_NFSRESERR;
1580 			debuglog("NFS unlock failure\n");
1581 			dump_filelock(fl);
1582 			break;
1583 		}
1584 
1585 		if (releasedfl != NULL) {
1586 			if (fl == releasedfl) {
1587 				/*
1588 				 * XXX: YECHHH!!! Attempt to unlock self succeeded
1589 				 * but we can't deallocate the space yet.  This is what
1590 				 * happens when you don't write malloc and free together
1591 				 */
1592 				debuglog("Attempt to unlock self\n");
1593 				selffl = releasedfl;
1594 			} else {
1595 				/*
1596 				 * XXX: this deallocation *still* needs to migrate closer
1597 				 * to the allocation code way up in get_lock or the allocation
1598 				 * code needs to migrate down (violation of "When you write
1599 				 * malloc you must write free")
1600 				 */
1601 
1602 				deallocate_file_lock(releasedfl);
1603 			}
1604 		}
1605 
1606 	} while (unlstatus == NFS_GRANTED);
1607 
1608 	if (selffl != NULL) {
1609 		/*
1610 		 * This statement wipes out the incoming file lock (fl)
1611 		 * in spite of the fact that it is declared const
1612 		 */
1613 		debuglog("WARNING!  Destroying incoming lock pointer\n");
1614 		deallocate_file_lock(selffl);
1615 	}
1616 
1617 	debuglog("Exiting unlock_partialfilelock\n");
1618 
1619 	return retval;
1620 }
1621 
1622 /*
1623  * clear_partialfilelock
1624  *
1625  * Normally called in response to statd state number change.
1626  * Wipe out all locks held by a host.  As a bonus, the act of
1627  * doing so should automatically clear their statd entries and
1628  * unmonitor the host.
1629  */
1630 
1631 void
1632 clear_partialfilelock(const char *hostname)
1633 {
1634 	struct file_lock *ifl, *nfl;
1635 
1636 	/* Clear blocking file lock list */
1637 	clear_blockingfilelock(hostname);
1638 
1639 	/* do all required unlocks */
1640 	/* Note that unlock can smash the current pointer to a lock */
1641 
1642 	/*
1643 	 * Normally, LIST_FOREACH is called for, but since
1644 	 * the current element *is* the iterator, deleting it
1645 	 * would mess up the iteration.  Thus, a next element
1646 	 * must be used explicitly
1647 	 */
1648 
1649 	ifl = LIST_FIRST(&nfslocklist_head);
1650 
1651 	while (ifl != NULL) {
1652 		nfl = LIST_NEXT(ifl, nfslocklist);
1653 
1654 		if (strncmp(hostname, ifl->client_name, SM_MAXSTRLEN) == 0) {
1655 			/* Unlock destroys ifl out from underneath */
1656 			unlock_partialfilelock(ifl);
1657 			/* ifl is NO LONGER VALID AT THIS POINT */
1658 		}
1659 		ifl = nfl;
1660 	}
1661 }
1662 
1663 /*
1664  * test_partialfilelock:
1665  */
1666 enum partialfilelock_status
1667 test_partialfilelock(const struct file_lock *fl,
1668     struct file_lock **conflicting_fl)
1669 {
1670 	enum partialfilelock_status retval;
1671 	enum nfslock_status teststatus;
1672 
1673 	debuglog("Entering testpartialfilelock...\n");
1674 
1675 	retval = PFL_DENIED;
1676 
1677 	teststatus = test_nfslock(fl, conflicting_fl);
1678 	debuglog("test_partialfilelock: teststatus %d\n",teststatus);
1679 
1680 	if (teststatus == NFS_GRANTED || teststatus == NFS_GRANTED_DUPLICATE) {
1681 		/* XXX: Add the underlying filesystem locking code */
1682 		retval = (teststatus == NFS_GRANTED) ?
1683 		    PFL_GRANTED : PFL_GRANTED_DUPLICATE;
1684 		debuglog("Dumping locks...\n");
1685 		dump_filelock(fl);
1686 		dump_filelock(*conflicting_fl);
1687 		debuglog("Done dumping locks...\n");
1688 	} else {
1689 		retval = PFL_NFSDENIED;
1690 		debuglog("NFS test denied.\n");
1691 		dump_filelock(fl);
1692 		debuglog("Conflicting.\n");
1693 		dump_filelock(*conflicting_fl);
1694 	}
1695 
1696 	debuglog("Exiting testpartialfilelock...\n");
1697 
1698 	return retval;
1699 }
1700 
1701 /*
1702  * Below here are routines associated with translating the partial file locking
1703  * codes into useful codes to send back to the NFS RPC messaging system
1704  */
1705 
1706 /*
1707  * These routines translate the (relatively) useful return codes back onto
1708  * the few return codes which the nlm subsystems wishes to trasmit
1709  */
1710 
1711 enum nlm_stats
1712 do_test(struct file_lock *fl, struct file_lock **conflicting_fl)
1713 {
1714 	enum partialfilelock_status pfsret;
1715 	enum nlm_stats retval;
1716 
1717 	debuglog("Entering do_test...\n");
1718 
1719 	pfsret = test_partialfilelock(fl,conflicting_fl);
1720 
1721 	switch (pfsret) {
1722 	case PFL_GRANTED:
1723 		debuglog("PFL test lock granted\n");
1724 		dump_filelock(fl);
1725 		dump_filelock(*conflicting_fl);
1726 		retval = (fl->flags & LOCK_V4) ? nlm4_granted : nlm_granted;
1727 		break;
1728 	case PFL_GRANTED_DUPLICATE:
1729 		debuglog("PFL test lock granted--duplicate id detected\n");
1730 		dump_filelock(fl);
1731 		dump_filelock(*conflicting_fl);
1732 		debuglog("Clearing conflicting_fl for call semantics\n");
1733 		*conflicting_fl = NULL;
1734 		retval = (fl->flags & LOCK_V4) ? nlm4_granted : nlm_granted;
1735 		break;
1736 	case PFL_NFSDENIED:
1737 	case PFL_HWDENIED:
1738 		debuglog("PFL test lock denied\n");
1739 		dump_filelock(fl);
1740 		dump_filelock(*conflicting_fl);
1741 		retval = (fl->flags & LOCK_V4) ? nlm4_denied : nlm_denied;
1742 		break;
1743 	case PFL_NFSRESERR:
1744 	case PFL_HWRESERR:
1745 		debuglog("PFL test lock resource fail\n");
1746 		dump_filelock(fl);
1747 		dump_filelock(*conflicting_fl);
1748 		retval = (fl->flags & LOCK_V4) ? nlm4_denied_nolocks : nlm_denied_nolocks;
1749 		break;
1750 	default:
1751 		debuglog("PFL test lock *FAILED*\n");
1752 		dump_filelock(fl);
1753 		dump_filelock(*conflicting_fl);
1754 		retval = (fl->flags & LOCK_V4) ? nlm4_failed : nlm_denied;
1755 		break;
1756 	}
1757 
1758 	debuglog("Exiting do_test...\n");
1759 
1760 	return retval;
1761 }
1762 
1763 /*
1764  * do_lock: Try to acquire a lock
1765  *
1766  * This routine makes a distinction between NLM versions.  I am pretty
1767  * convinced that this should be abstracted out and bounced up a level
1768  */
1769 
1770 enum nlm_stats
1771 do_lock(struct file_lock *fl)
1772 {
1773 	enum partialfilelock_status pfsret;
1774 	enum nlm_stats retval;
1775 
1776 	debuglog("Entering do_lock...\n");
1777 
1778 	pfsret = lock_partialfilelock(fl);
1779 
1780 	switch (pfsret) {
1781 	case PFL_GRANTED:
1782 		debuglog("PFL lock granted");
1783 		dump_filelock(fl);
1784 		retval = (fl->flags & LOCK_V4) ? nlm4_granted : nlm_granted;
1785 		break;
1786 	case PFL_GRANTED_DUPLICATE:
1787 		debuglog("PFL lock granted--duplicate id detected");
1788 		dump_filelock(fl);
1789 		retval = (fl->flags & LOCK_V4) ? nlm4_granted : nlm_granted;
1790 		break;
1791 	case PFL_NFSDENIED:
1792 	case PFL_HWDENIED:
1793 		debuglog("PFL_NFS lock denied");
1794 		dump_filelock(fl);
1795 		retval = (fl->flags & LOCK_V4) ? nlm4_denied : nlm_denied;
1796 		break;
1797 	case PFL_NFSBLOCKED:
1798 	case PFL_HWBLOCKED:
1799 		debuglog("PFL_NFS blocking lock denied.  Queued.\n");
1800 		dump_filelock(fl);
1801 		retval = (fl->flags & LOCK_V4) ? nlm4_blocked : nlm_blocked;
1802 		break;
1803 	case PFL_NFSRESERR:
1804 	case PFL_HWRESERR:
1805 		debuglog("PFL lock resource alocation fail\n");
1806 		dump_filelock(fl);
1807 		retval = (fl->flags & LOCK_V4) ? nlm4_denied_nolocks : nlm_denied_nolocks;
1808 		break;
1809 	default:
1810 		debuglog("PFL lock *FAILED*");
1811 		dump_filelock(fl);
1812 		retval = (fl->flags & LOCK_V4) ? nlm4_failed : nlm_denied;
1813 		break;
1814 	}
1815 
1816 	debuglog("Exiting do_lock...\n");
1817 
1818 	return retval;
1819 }
1820 
1821 enum nlm_stats
1822 do_unlock(struct file_lock *fl)
1823 {
1824 	enum partialfilelock_status pfsret;
1825 	enum nlm_stats retval;
1826 
1827 	debuglog("Entering do_unlock...\n");
1828 	pfsret = unlock_partialfilelock(fl);
1829 
1830 	switch (pfsret) {
1831 	case PFL_GRANTED:
1832 		debuglog("PFL unlock granted");
1833 		dump_filelock(fl);
1834 		retval = (fl->flags & LOCK_V4) ? nlm4_granted : nlm_granted;
1835 		break;
1836 	case PFL_NFSDENIED:
1837 	case PFL_HWDENIED:
1838 		debuglog("PFL_NFS unlock denied");
1839 		dump_filelock(fl);
1840 		retval = (fl->flags & LOCK_V4) ? nlm4_denied : nlm_denied;
1841 		break;
1842 	case PFL_NFSDENIED_NOLOCK:
1843 	case PFL_HWDENIED_NOLOCK:
1844 		debuglog("PFL_NFS no lock found\n");
1845 		retval = (fl->flags & LOCK_V4) ? nlm4_granted : nlm_granted;
1846 		break;
1847 	case PFL_NFSRESERR:
1848 	case PFL_HWRESERR:
1849 		debuglog("PFL unlock resource failure");
1850 		dump_filelock(fl);
1851 		retval = (fl->flags & LOCK_V4) ? nlm4_denied_nolocks : nlm_denied_nolocks;
1852 		break;
1853 	default:
1854 		debuglog("PFL unlock *FAILED*");
1855 		dump_filelock(fl);
1856 		retval = (fl->flags & LOCK_V4) ? nlm4_failed : nlm_denied;
1857 		break;
1858 	}
1859 
1860 	debuglog("Exiting do_unlock...\n");
1861 
1862 	return retval;
1863 }
1864 
1865 /*
1866  * do_clear
1867  *
1868  * This routine is non-existent because it doesn't have a return code.
1869  * It is here for completeness in case someone *does* need to do return
1870  * codes later.  A decent compiler should optimize this away.
1871  */
1872 
1873 void
1874 do_clear(const char *hostname)
1875 {
1876 
1877 	clear_partialfilelock(hostname);
1878 }
1879 
1880 /*
1881  * The following routines are all called from the code which the
1882  * RPC layer invokes
1883  */
1884 
1885 /*
1886  * testlock(): inform the caller if the requested lock would be granted
1887  *
1888  * returns NULL if lock would granted
1889  * returns pointer to a conflicting nlm4_holder if not
1890  */
1891 
1892 struct nlm4_holder *
1893 testlock(struct nlm4_lock *lock, bool_t exclusive, int flags __unused)
1894 {
1895 	struct file_lock test_fl, *conflicting_fl;
1896 
1897 	bzero(&test_fl, sizeof(test_fl));
1898 
1899 	bcopy(lock->fh.n_bytes, &(test_fl.filehandle), sizeof(fhandle_t));
1900 	copy_nlm4_lock_to_nlm4_holder(lock, exclusive, &test_fl.client);
1901 
1902 	siglock();
1903 	do_test(&test_fl, &conflicting_fl);
1904 
1905 	if (conflicting_fl == NULL) {
1906 		debuglog("No conflicting lock found\n");
1907 		sigunlock();
1908 		return NULL;
1909 	} else {
1910 		debuglog("Found conflicting lock\n");
1911 		dump_filelock(conflicting_fl);
1912 		sigunlock();
1913 		return (&conflicting_fl->client);
1914 	}
1915 }
1916 
1917 /*
1918  * getlock: try to acquire the lock.
1919  * If file is already locked and we can sleep, put the lock in the list with
1920  * status LKST_WAITING; it'll be processed later.
1921  * Otherwise try to lock. If we're allowed to block, fork a child which
1922  * will do the blocking lock.
1923  */
1924 
1925 enum nlm_stats
1926 getlock(nlm4_lockargs *lckarg, struct svc_req *rqstp, const int flags)
1927 {
1928 	struct file_lock *newfl;
1929 	enum nlm_stats retval;
1930 
1931 	debuglog("Entering getlock...\n");
1932 
1933 	if (grace_expired == 0 && lckarg->reclaim == 0)
1934 		return (flags & LOCK_V4) ?
1935 		    nlm4_denied_grace_period : nlm_denied_grace_period;
1936 
1937 	/* allocate new file_lock for this request */
1938 	newfl = allocate_file_lock(&lckarg->alock.oh, &lckarg->cookie,
1939 				   (struct sockaddr *)svc_getrpccaller(rqstp->rq_xprt)->buf, lckarg->alock.caller_name);
1940 	if (newfl == NULL) {
1941 		syslog(LOG_NOTICE, "lock allocate failed: %s", strerror(errno));
1942 		/* failed */
1943 		return (flags & LOCK_V4) ?
1944 		    nlm4_denied_nolocks : nlm_denied_nolocks;
1945 	}
1946 
1947 	if (lckarg->alock.fh.n_len != sizeof(fhandle_t)) {
1948 		debuglog("received fhandle size %d, local size %d",
1949 		    lckarg->alock.fh.n_len, (int)sizeof(fhandle_t));
1950 	}
1951 
1952 	fill_file_lock(newfl, (fhandle_t *)lckarg->alock.fh.n_bytes,
1953 	    lckarg->exclusive, lckarg->alock.svid, lckarg->alock.l_offset,
1954 	    lckarg->alock.l_len,
1955 	    lckarg->state, 0, flags, lckarg->block);
1956 
1957 	/*
1958 	 * newfl is now fully constructed and deallocate_file_lock
1959 	 * can now be used to delete it
1960 	 */
1961 
1962 	siglock();
1963 	debuglog("Pointer to new lock is %p\n",newfl);
1964 
1965 	retval = do_lock(newfl);
1966 
1967 	debuglog("Pointer to new lock is %p\n",newfl);
1968 	sigunlock();
1969 
1970 	switch (retval)
1971 		{
1972 		case nlm4_granted:
1973 			/* case nlm_granted: is the same as nlm4_granted */
1974 			/* do_mon(lckarg->alock.caller_name); */
1975 			break;
1976 		case nlm4_blocked:
1977 			/* case nlm_blocked: is the same as nlm4_blocked */
1978 			/* do_mon(lckarg->alock.caller_name); */
1979 			break;
1980 		default:
1981 			deallocate_file_lock(newfl);
1982 			break;
1983 		}
1984 
1985 	debuglog("Exiting getlock...\n");
1986 
1987 	return retval;
1988 }
1989 
1990 
1991 /* unlock a filehandle */
1992 enum nlm_stats
1993 unlock(nlm4_lock *lock, const int flags __unused)
1994 {
1995 	struct file_lock fl;
1996 	enum nlm_stats err;
1997 
1998 	siglock();
1999 
2000 	debuglog("Entering unlock...\n");
2001 
2002 	bzero(&fl,sizeof(struct file_lock));
2003 	bcopy(lock->fh.n_bytes, &fl.filehandle, sizeof(fhandle_t));
2004 
2005 	copy_nlm4_lock_to_nlm4_holder(lock, 0, &fl.client);
2006 
2007 	err = do_unlock(&fl);
2008 
2009 	sigunlock();
2010 
2011 	debuglog("Exiting unlock...\n");
2012 
2013 	return err;
2014 }
2015 
2016 /*
2017  * XXX: The following monitor/unmonitor routines
2018  * have not been extensively tested (ie. no regression
2019  * script exists like for the locking sections
2020  */
2021 
2022 /*
2023  * monitor_lock_host: monitor lock hosts locally with a ref count and
2024  * inform statd
2025  */
2026 void
2027 monitor_lock_host(const char *hostname)
2028 {
2029 	struct host *ihp, *nhp;
2030 	struct mon smon;
2031 	struct sm_stat_res sres;
2032 	int rpcret, statflag;
2033 	size_t n;
2034 
2035 	rpcret = 0;
2036 	statflag = 0;
2037 
2038 	LIST_FOREACH(ihp, &hostlst_head, hostlst) {
2039 		if (strncmp(hostname, ihp->name, SM_MAXSTRLEN) == 0) {
2040 			/* Host is already monitored, bump refcount */
2041 			++ihp->refcnt;
2042 			/* Host should only be in the monitor list once */
2043 			return;
2044 		}
2045 	}
2046 
2047 	/* Host is not yet monitored, add it */
2048 	n = strnlen(hostname, SM_MAXSTRLEN);
2049 	if (n == SM_MAXSTRLEN) {
2050 		return;
2051 	}
2052 	nhp = malloc(sizeof(*nhp) - sizeof(nhp->name) + n + 1);
2053 	if (nhp == NULL) {
2054 		debuglog("Unable to allocate entry for statd mon\n");
2055 		return;
2056 	}
2057 
2058 	/* Allocated new host entry, now fill the fields */
2059 	memcpy(nhp->name, hostname, n);
2060 	nhp->name[n] = 0;
2061 	nhp->refcnt = 1;
2062 	debuglog("Locally Monitoring host %16s\n",hostname);
2063 
2064 	debuglog("Attempting to tell statd\n");
2065 
2066 	bzero(&smon,sizeof(smon));
2067 
2068 	smon.mon_id.mon_name = nhp->name;
2069 	smon.mon_id.my_id.my_name = "localhost";
2070 	smon.mon_id.my_id.my_prog = NLM_PROG;
2071 	smon.mon_id.my_id.my_vers = NLM_SM;
2072 	smon.mon_id.my_id.my_proc = NLM_SM_NOTIFY;
2073 
2074 	rpcret = callrpc("localhost", SM_PROG, SM_VERS, SM_MON,
2075 	    (xdrproc_t)xdr_mon, &smon,
2076 	    (xdrproc_t)xdr_sm_stat_res, &sres);
2077 
2078 	if (rpcret == 0) {
2079 		if (sres.res_stat == stat_fail) {
2080 			debuglog("Statd call failed\n");
2081 			statflag = 0;
2082 		} else {
2083 			statflag = 1;
2084 		}
2085 	} else {
2086 		debuglog("Rpc call to statd failed with return value: %d\n",
2087 		    rpcret);
2088 		statflag = 0;
2089 	}
2090 
2091 	if (statflag == 1) {
2092 		LIST_INSERT_HEAD(&hostlst_head, nhp, hostlst);
2093 	} else {
2094 		free(nhp);
2095 	}
2096 
2097 }
2098 
2099 /*
2100  * unmonitor_lock_host: clear monitor ref counts and inform statd when gone
2101  */
2102 void
2103 unmonitor_lock_host(char *hostname)
2104 {
2105 	struct host *ihp;
2106 	struct mon_id smon_id;
2107 	struct sm_stat smstat;
2108 	int rpcret;
2109 
2110 	rpcret = 0;
2111 
2112 	for( ihp=LIST_FIRST(&hostlst_head); ihp != NULL;
2113 	     ihp=LIST_NEXT(ihp, hostlst)) {
2114 		if (strncmp(hostname, ihp->name, SM_MAXSTRLEN) == 0) {
2115 			/* Host is monitored, bump refcount */
2116 			--ihp->refcnt;
2117 			/* Host should only be in the monitor list once */
2118 			break;
2119 		}
2120 	}
2121 
2122 	if (ihp == NULL) {
2123 		debuglog("Could not find host %16s in mon list\n", hostname);
2124 		return;
2125 	}
2126 
2127 	if (ihp->refcnt > 0)
2128 		return;
2129 
2130 	if (ihp->refcnt < 0) {
2131 		debuglog("Negative refcount!: %d\n",
2132 		    ihp->refcnt);
2133 	}
2134 
2135 	debuglog("Attempting to unmonitor host %16s\n", hostname);
2136 
2137 	bzero(&smon_id,sizeof(smon_id));
2138 
2139 	smon_id.mon_name = hostname;
2140 	smon_id.my_id.my_name = "localhost";
2141 	smon_id.my_id.my_prog = NLM_PROG;
2142 	smon_id.my_id.my_vers = NLM_SM;
2143 	smon_id.my_id.my_proc = NLM_SM_NOTIFY;
2144 
2145 	rpcret = callrpc("localhost", SM_PROG, SM_VERS, SM_UNMON,
2146 	    (xdrproc_t)xdr_mon_id, &smon_id,
2147 	    (xdrproc_t)xdr_sm_stat, &smstat);
2148 
2149 	if (rpcret != 0) {
2150 		debuglog("Rpc call to unmonitor statd failed with "
2151 		   " return value: %d\n", rpcret);
2152 	}
2153 
2154 	LIST_REMOVE(ihp, hostlst);
2155 	free(ihp);
2156 }
2157 
2158 /*
2159  * notify: Clear all locks from a host if statd complains
2160  *
2161  * XXX: This routine has not been thoroughly tested.  However, neither
2162  * had the old one been.  It used to compare the statd crash state counter
2163  * to the current lock state.  The upshot of this was that it basically
2164  * cleared all locks from the specified host 99% of the time (with the
2165  * other 1% being a bug).  Consequently, the assumption is that clearing
2166  * all locks from a host when notified by statd is acceptable.
2167  *
2168  * Please note that this routine skips the usual level of redirection
2169  * through a do_* type routine.  This introduces a possible level of
2170  * error and might better be written as do_notify and take this one out.
2171 
2172  */
2173 
2174 void
2175 notify(const char *hostname, const int state)
2176 {
2177 	debuglog("notify from %s, new state %d", hostname, state);
2178 
2179 	siglock();
2180 	do_clear(hostname);
2181 	sigunlock();
2182 
2183 	debuglog("Leaving notify\n");
2184 }
2185 
2186 void
2187 send_granted(fl, opcode)
2188 	struct file_lock *fl;
2189 	int opcode __unused;
2190 {
2191 	CLIENT *cli;
2192 	static char dummy;
2193 	struct timeval timeo;
2194 	int success;
2195 	static struct nlm_res retval;
2196 	static struct nlm4_res retval4;
2197 
2198 	debuglog("About to send granted on blocked lock\n");
2199 
2200 	cli = get_client(fl->addr,
2201 	    (fl->flags & LOCK_V4) ? NLM_VERS4 : NLM_VERS);
2202 	if (cli == NULL) {
2203 		syslog(LOG_NOTICE, "failed to get CLIENT for %s",
2204 		    fl->client_name);
2205 		/*
2206 		 * We fail to notify remote that the lock has been granted.
2207 		 * The client will timeout and retry, the lock will be
2208 		 * granted at this time.
2209 		 */
2210 		return;
2211 	}
2212 	timeo.tv_sec = 0;
2213 	timeo.tv_usec = (fl->flags & LOCK_ASYNC) ? 0 : 500000; /* 0.5s */
2214 
2215 	if (fl->flags & LOCK_V4) {
2216 		static nlm4_testargs res;
2217 		res.cookie = fl->client_cookie;
2218 		res.exclusive = fl->client.exclusive;
2219 		res.alock.caller_name = fl->client_name;
2220 		res.alock.fh.n_len = sizeof(fhandle_t);
2221 		res.alock.fh.n_bytes = (char*)&fl->filehandle;
2222 		res.alock.oh = fl->client.oh;
2223 		res.alock.svid = fl->client.svid;
2224 		res.alock.l_offset = fl->client.l_offset;
2225 		res.alock.l_len = fl->client.l_len;
2226 		debuglog("sending v4 reply%s",
2227 			 (fl->flags & LOCK_ASYNC) ? " (async)":"");
2228 		if (fl->flags & LOCK_ASYNC) {
2229 			success = clnt_call(cli, NLM4_GRANTED_MSG,
2230 			    (xdrproc_t)xdr_nlm4_testargs, &res,
2231 			    (xdrproc_t)xdr_void, &dummy, timeo);
2232 		} else {
2233 			success = clnt_call(cli, NLM4_GRANTED,
2234 			    (xdrproc_t)xdr_nlm4_testargs, &res,
2235 			    (xdrproc_t)xdr_nlm4_res, &retval4, timeo);
2236 		}
2237 	} else {
2238 		static nlm_testargs res;
2239 
2240 		res.cookie = fl->client_cookie;
2241 		res.exclusive = fl->client.exclusive;
2242 		res.alock.caller_name = fl->client_name;
2243 		res.alock.fh.n_len = sizeof(fhandle_t);
2244 		res.alock.fh.n_bytes = (char*)&fl->filehandle;
2245 		res.alock.oh = fl->client.oh;
2246 		res.alock.svid = fl->client.svid;
2247 		res.alock.l_offset = fl->client.l_offset;
2248 		res.alock.l_len = fl->client.l_len;
2249 		debuglog("sending v1 reply%s",
2250 			 (fl->flags & LOCK_ASYNC) ? " (async)":"");
2251 		if (fl->flags & LOCK_ASYNC) {
2252 			success = clnt_call(cli, NLM_GRANTED_MSG,
2253 			    (xdrproc_t)xdr_nlm_testargs, &res,
2254 			    (xdrproc_t)xdr_void, &dummy, timeo);
2255 		} else {
2256 			success = clnt_call(cli, NLM_GRANTED,
2257 			    (xdrproc_t)xdr_nlm_testargs, &res,
2258 			    (xdrproc_t)xdr_nlm_res, &retval, timeo);
2259 		}
2260 	}
2261 	if (debug_level > 2)
2262 		debuglog("clnt_call returns %d(%s) for granted",
2263 			 success, clnt_sperrno(success));
2264 
2265 }
2266 
2267 /*
2268  * Routines below here have not been modified in the overhaul
2269  */
2270 
2271 /*
2272  * Are these two routines still required since lockd is not spawning off
2273  * children to service locks anymore?  Presumably they were originally
2274  * put in place to prevent a one child from changing the lock list out
2275  * from under another one.
2276  */
2277 
2278 void
2279 siglock(void)
2280 {
2281   sigset_t block;
2282 
2283   sigemptyset(&block);
2284   sigaddset(&block, SIGCHLD);
2285 
2286   if (sigprocmask(SIG_BLOCK, &block, NULL) < 0) {
2287     syslog(LOG_WARNING, "siglock failed: %s", strerror(errno));
2288   }
2289 }
2290 
2291 void
2292 sigunlock(void)
2293 {
2294   sigset_t block;
2295 
2296   sigemptyset(&block);
2297   sigaddset(&block, SIGCHLD);
2298 
2299   if (sigprocmask(SIG_UNBLOCK, &block, NULL) < 0) {
2300     syslog(LOG_WARNING, "sigunlock failed: %s", strerror(errno));
2301   }
2302 }
2303