xref: /freebsd/usr.sbin/rpc.lockd/lockd_lock.c (revision 1d386b48a555f61cb7325543adbbb5c3f3407a66)
1 /*	$NetBSD: lockd_lock.c,v 1.5 2000/11/21 03:47:41 enami Exp $	*/
2 
3 /*-
4  * SPDX-License-Identifier: BSD-4-Clause
5  *
6  * Copyright (c) 2001 Andrew P. Lentvorski, Jr.
7  * Copyright (c) 2000 Manuel Bouyer.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 3. All advertising materials mentioning features or use of this software
18  *    must display the following acknowledgement:
19  *	This product includes software developed by the University of
20  *	California, Berkeley and its contributors.
21  * 4. Neither the name of the University nor the names of its contributors
22  *    may be used to endorse or promote products derived from this software
23  *    without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  *
37  */
38 
39 #include <sys/cdefs.h>
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 
219 void
debuglog(char const * fmt,...)220 debuglog(char const *fmt, ...)
221 {
222 	va_list ap;
223 
224 	if (debug_level < 1) {
225 		return;
226 	}
227 
228 	sleep(debugdelay);
229 
230 	va_start(ap, fmt);
231 	vsyslog(LOG_DEBUG, fmt, ap);
232 	va_end(ap);
233 }
234 
235 void
dump_static_object(const unsigned char * object,const int size_object,unsigned char * hbuff,const int size_hbuff,unsigned char * cbuff,const int size_cbuff)236 dump_static_object(const unsigned char *object, const int size_object,
237     unsigned char *hbuff, const int size_hbuff, unsigned char *cbuff,
238     const int size_cbuff)
239 {
240 	int i, objectsize;
241 
242 	if (debug_level < 2) {
243 		return;
244 	}
245 
246 	objectsize = size_object;
247 
248 	if (objectsize == 0) {
249 		debuglog("object is size 0\n");
250 	} else {
251 		if (objectsize > MAXOBJECTSIZE) {
252 			debuglog("Object of size %d being clamped"
253 			    "to size %d\n", objectsize, MAXOBJECTSIZE);
254 			objectsize = MAXOBJECTSIZE;
255 		}
256 
257 		if (hbuff != NULL) {
258 			if (size_hbuff < objectsize*2+1) {
259 				debuglog("Hbuff not large enough."
260 				    "  Increase size\n");
261 			} else {
262 				for(i=0;i<objectsize;i++) {
263 					sprintf(hbuff+i*2,"%02x",*(object+i));
264 				}
265 				*(hbuff+i*2) = '\0';
266 			}
267 		}
268 
269 		if (cbuff != NULL) {
270 			if (size_cbuff < objectsize+1) {
271 				debuglog("Cbuff not large enough."
272 				    "  Increase Size\n");
273 			}
274 
275 			for(i=0;i<objectsize;i++) {
276 				if (*(object+i) >= 32 && *(object+i) <= 127) {
277 					*(cbuff+i) = *(object+i);
278 				} else {
279 					*(cbuff+i) = '.';
280 				}
281 			}
282 			*(cbuff+i) = '\0';
283 		}
284 	}
285 }
286 
287 void
dump_netobj(const struct netobj * nobj)288 dump_netobj(const struct netobj *nobj)
289 {
290 	char hbuff[MAXBUFFERSIZE*2];
291 	char cbuff[MAXBUFFERSIZE];
292 
293 	if (debug_level < 2) {
294 		return;
295 	}
296 
297 	if (nobj == NULL) {
298 		debuglog("Null netobj pointer\n");
299 	}
300 	else if (nobj->n_len == 0) {
301 		debuglog("Size zero netobj\n");
302 	} else {
303 		dump_static_object(nobj->n_bytes, nobj->n_len,
304 		    hbuff, sizeof(hbuff), cbuff, sizeof(cbuff));
305 		debuglog("netobj: len: %d  data: %s :::  %s\n",
306 		    nobj->n_len, hbuff, cbuff);
307 	}
308 }
309 
310 /* #define DUMP_FILELOCK_VERBOSE */
311 void
dump_filelock(const struct file_lock * fl)312 dump_filelock(const struct file_lock *fl)
313 {
314 #ifdef DUMP_FILELOCK_VERBOSE
315 	char hbuff[MAXBUFFERSIZE*2];
316 	char cbuff[MAXBUFFERSIZE];
317 #endif
318 
319 	if (debug_level < 2) {
320 		return;
321 	}
322 
323 	if (fl != NULL) {
324 		debuglog("Dumping file lock structure @ %p\n", fl);
325 
326 #ifdef DUMP_FILELOCK_VERBOSE
327 		dump_static_object((unsigned char *)&fl->filehandle,
328 		    sizeof(fl->filehandle), hbuff, sizeof(hbuff),
329 		    cbuff, sizeof(cbuff));
330 		debuglog("Filehandle: %8s  :::  %8s\n", hbuff, cbuff);
331 #endif
332 
333 		debuglog("Dumping nlm4_holder:\n"
334 		    "exc: %x  svid: %x  offset:len %llx:%llx\n",
335 		    fl->client.exclusive, fl->client.svid,
336 		    fl->client.l_offset, fl->client.l_len);
337 
338 #ifdef DUMP_FILELOCK_VERBOSE
339 		debuglog("Dumping client identity:\n");
340 		dump_netobj(&fl->client.oh);
341 
342 		debuglog("Dumping client cookie:\n");
343 		dump_netobj(&fl->client_cookie);
344 
345 		debuglog("nsm: %d  status: %d  flags: %d  svid: %x"
346 		    "  client_name: %s\n", fl->nsm_status, fl->status,
347 		    fl->flags, fl->client.svid, fl->client_name);
348 #endif
349 	} else {
350 		debuglog("NULL file lock structure\n");
351 	}
352 }
353 
354 void
copy_nlm4_lock_to_nlm4_holder(const struct nlm4_lock * src,const bool_t exclusive,struct nlm4_holder * dest)355 copy_nlm4_lock_to_nlm4_holder(const struct nlm4_lock *src,
356     const bool_t exclusive, struct nlm4_holder *dest)
357 {
358 
359 	dest->exclusive = exclusive;
360 	dest->oh.n_len = src->oh.n_len;
361 	dest->oh.n_bytes = src->oh.n_bytes;
362 	dest->svid = src->svid;
363 	dest->l_offset = src->l_offset;
364 	dest->l_len = src->l_len;
365 }
366 
367 /*
368  * allocate_file_lock: Create a lock with the given parameters
369  */
370 
371 struct file_lock *
allocate_file_lock(const netobj * lockowner,const netobj * matchcookie,const struct sockaddr * addr,const char * caller_name)372 allocate_file_lock(const netobj *lockowner, const netobj *matchcookie,
373 		   const struct sockaddr *addr, const char *caller_name)
374 {
375 	struct file_lock *newfl;
376 	size_t n;
377 
378 	/* Beware of rubbish input! */
379 	n = strnlen(caller_name, SM_MAXSTRLEN);
380 	if (n == SM_MAXSTRLEN) {
381 		return NULL;
382 	}
383 
384 	newfl = malloc(sizeof(*newfl) - sizeof(newfl->client_name) + n + 1);
385 	if (newfl == NULL) {
386 		return NULL;
387 	}
388 	bzero(newfl, sizeof(*newfl) - sizeof(newfl->client_name));
389 	memcpy(newfl->client_name, caller_name, n);
390 	newfl->client_name[n] = 0;
391 
392 	newfl->client.oh.n_bytes = malloc(lockowner->n_len);
393 	if (newfl->client.oh.n_bytes == NULL) {
394 		free(newfl);
395 		return NULL;
396 	}
397 	newfl->client.oh.n_len = lockowner->n_len;
398 	bcopy(lockowner->n_bytes, newfl->client.oh.n_bytes, lockowner->n_len);
399 
400 	newfl->client_cookie.n_bytes = malloc(matchcookie->n_len);
401 	if (newfl->client_cookie.n_bytes == NULL) {
402 		free(newfl->client.oh.n_bytes);
403 		free(newfl);
404 		return NULL;
405 	}
406 	newfl->client_cookie.n_len = matchcookie->n_len;
407 	bcopy(matchcookie->n_bytes, newfl->client_cookie.n_bytes, matchcookie->n_len);
408 
409 	newfl->addr = malloc(addr->sa_len);
410 	if (newfl->addr == NULL) {
411 		free(newfl->client_cookie.n_bytes);
412 		free(newfl->client.oh.n_bytes);
413 		free(newfl);
414 		return NULL;
415 	}
416 	memcpy(newfl->addr, addr, addr->sa_len);
417 
418 	return newfl;
419 }
420 
421 /*
422  * file_file_lock: Force creation of a valid file lock
423  */
424 void
fill_file_lock(struct file_lock * fl,const fhandle_t * fh,const bool_t exclusive,const int32_t svid,const u_int64_t offset,const u_int64_t len,const int state,const int status,const int flags,const int blocking)425 fill_file_lock(struct file_lock *fl, const fhandle_t *fh,
426     const bool_t exclusive, const int32_t svid,
427     const u_int64_t offset, const u_int64_t len,
428     const int state, const int status, const int flags, const int blocking)
429 {
430 	bcopy(fh, &fl->filehandle, sizeof(fhandle_t));
431 
432 	fl->client.exclusive = exclusive;
433 	fl->client.svid = svid;
434 	fl->client.l_offset = offset;
435 	fl->client.l_len = len;
436 
437 	fl->nsm_status = state;
438 	fl->status = status;
439 	fl->flags = flags;
440 	fl->blocking = blocking;
441 }
442 
443 /*
444  * deallocate_file_lock: Free all storage associated with a file lock
445  */
446 void
deallocate_file_lock(struct file_lock * fl)447 deallocate_file_lock(struct file_lock *fl)
448 {
449 	free(fl->addr);
450 	free(fl->client.oh.n_bytes);
451 	free(fl->client_cookie.n_bytes);
452 	free(fl);
453 }
454 
455 /*
456  * regions_overlap(): This function examines the two provided regions for
457  * overlap.
458  */
459 int
regions_overlap(const u_int64_t start1,const u_int64_t len1,const u_int64_t start2,const u_int64_t len2)460 regions_overlap(const u_int64_t start1, const u_int64_t len1,
461     const u_int64_t start2, const u_int64_t len2)
462 {
463 	u_int64_t d1,d2,d3,d4;
464 	enum split_status result;
465 
466 	debuglog("Entering region overlap with vals: %llu:%llu--%llu:%llu\n",
467 		 start1, len1, start2, len2);
468 
469 	result = region_compare(start1, len1, start2, len2,
470 	    &d1, &d2, &d3, &d4);
471 
472 	debuglog("Exiting region overlap with val: %d\n",result);
473 
474 	if (result == SPL_DISJOINT) {
475 		return 0;
476 	} else {
477 		return 1;
478 	}
479 }
480 
481 /*
482  * region_compare(): Examine lock regions and split appropriately
483  *
484  * XXX: Fix 64 bit overflow problems
485  * XXX: Check to make sure I got *ALL* the cases.
486  * XXX: This DESPERATELY needs a regression test.
487  */
488 enum split_status
region_compare(const u_int64_t starte,const u_int64_t lene,const u_int64_t startu,const u_int64_t lenu,u_int64_t * start1,u_int64_t * len1,u_int64_t * start2,u_int64_t * len2)489 region_compare(const u_int64_t starte, const u_int64_t lene,
490     const u_int64_t startu, const u_int64_t lenu, u_int64_t *start1,
491     u_int64_t *len1, u_int64_t *start2, u_int64_t *len2)
492 {
493 	/*
494 	 * Please pay attention to the sequential exclusions
495 	 * of the if statements!!!
496 	 */
497 	enum LFLAGS lflags;
498 	enum RFLAGS rflags;
499 	enum split_status retval;
500 
501 	retval = SPL_DISJOINT;
502 
503 	if (lene == 0 && lenu == 0) {
504 		/* Examine left edge of locker */
505 		lflags = LEDGE_INSIDE;
506 		if (startu < starte) {
507 			lflags = LEDGE_LEFT;
508 		} else if (startu == starte) {
509 			lflags = LEDGE_LBOUNDARY;
510 		}
511 
512 		rflags = REDGE_RBOUNDARY; /* Both are infiinite */
513 
514 		if (lflags == LEDGE_INSIDE) {
515 			*start1 = starte;
516 			*len1 = startu - starte;
517 		}
518 
519 		if (lflags == LEDGE_LEFT || lflags == LEDGE_LBOUNDARY) {
520 			retval = SPL_CONTAINED;
521 		} else {
522 			retval = SPL_LOCK1;
523 		}
524 	} else if (lene == 0 && lenu != 0) {
525 		/* Established lock is infinite */
526 		/* Examine left edge of unlocker */
527 		lflags = LEDGE_INSIDE;
528 		if (startu < starte) {
529 			lflags = LEDGE_LEFT;
530 		} else if (startu == starte) {
531 			lflags = LEDGE_LBOUNDARY;
532 		}
533 
534 		/* Examine right edge of unlocker */
535 		if (startu + lenu < starte) {
536 			/* Right edge of unlocker left of established lock */
537 			rflags = REDGE_LEFT;
538 			return SPL_DISJOINT;
539 		} else if (startu + lenu == starte) {
540 			/* Right edge of unlocker on start of established lock */
541 			rflags = REDGE_LBOUNDARY;
542 			return SPL_DISJOINT;
543 		} else { /* Infinifty is right of finity */
544 			/* Right edge of unlocker inside established lock */
545 			rflags = REDGE_INSIDE;
546 		}
547 
548 		if (lflags == LEDGE_INSIDE) {
549 			*start1 = starte;
550 			*len1 = startu - starte;
551 			retval |= SPL_LOCK1;
552 		}
553 
554 		if (rflags == REDGE_INSIDE) {
555 			/* Create right lock */
556 			*start2 = startu+lenu;
557 			*len2 = 0;
558 			retval |= SPL_LOCK2;
559 		}
560 	} else if (lene != 0 && lenu == 0) {
561 		/* Unlocker is infinite */
562 		/* Examine left edge of unlocker */
563 		lflags = LEDGE_RIGHT;
564 		if (startu < starte) {
565 			lflags = LEDGE_LEFT;
566 			retval = SPL_CONTAINED;
567 			return retval;
568 		} else if (startu == starte) {
569 			lflags = LEDGE_LBOUNDARY;
570 			retval = SPL_CONTAINED;
571 			return retval;
572 		} else if ((startu > starte) && (startu < starte + lene - 1)) {
573 			lflags = LEDGE_INSIDE;
574 		} else if (startu == starte + lene - 1) {
575 			lflags = LEDGE_RBOUNDARY;
576 		} else { /* startu > starte + lene -1 */
577 			lflags = LEDGE_RIGHT;
578 			return SPL_DISJOINT;
579 		}
580 
581 		rflags = REDGE_RIGHT; /* Infinity is right of finity */
582 
583 		if (lflags == LEDGE_INSIDE || lflags == LEDGE_RBOUNDARY) {
584 			*start1 = starte;
585 			*len1 = startu - starte;
586 			retval |= SPL_LOCK1;
587 			return retval;
588 		}
589 	} else {
590 		/* Both locks are finite */
591 
592 		/* Examine left edge of unlocker */
593 		lflags = LEDGE_RIGHT;
594 		if (startu < starte) {
595 			lflags = LEDGE_LEFT;
596 		} else if (startu == starte) {
597 			lflags = LEDGE_LBOUNDARY;
598 		} else if ((startu > starte) && (startu < starte + lene - 1)) {
599 			lflags = LEDGE_INSIDE;
600 		} else if (startu == starte + lene - 1) {
601 			lflags = LEDGE_RBOUNDARY;
602 		} else { /* startu > starte + lene -1 */
603 			lflags = LEDGE_RIGHT;
604 			return SPL_DISJOINT;
605 		}
606 
607 		/* Examine right edge of unlocker */
608 		if (startu + lenu < starte) {
609 			/* Right edge of unlocker left of established lock */
610 			rflags = REDGE_LEFT;
611 			return SPL_DISJOINT;
612 		} else if (startu + lenu == starte) {
613 			/* Right edge of unlocker on start of established lock */
614 			rflags = REDGE_LBOUNDARY;
615 			return SPL_DISJOINT;
616 		} else if (startu + lenu < starte + lene) {
617 			/* Right edge of unlocker inside established lock */
618 			rflags = REDGE_INSIDE;
619 		} else if (startu + lenu == starte + lene) {
620 			/* Right edge of unlocker on right edge of established lock */
621 			rflags = REDGE_RBOUNDARY;
622 		} else { /* startu + lenu > starte + lene */
623 			/* Right edge of unlocker is right of established lock */
624 			rflags = REDGE_RIGHT;
625 		}
626 
627 		if (lflags == LEDGE_INSIDE || lflags == LEDGE_RBOUNDARY) {
628 			/* Create left lock */
629 			*start1 = starte;
630 			*len1 = (startu - starte);
631 			retval |= SPL_LOCK1;
632 		}
633 
634 		if (rflags == REDGE_INSIDE) {
635 			/* Create right lock */
636 			*start2 = startu+lenu;
637 			*len2 = starte+lene-(startu+lenu);
638 			retval |= SPL_LOCK2;
639 		}
640 
641 		if ((lflags == LEDGE_LEFT || lflags == LEDGE_LBOUNDARY) &&
642 		    (rflags == REDGE_RBOUNDARY || rflags == REDGE_RIGHT)) {
643 			retval = SPL_CONTAINED;
644 		}
645 	}
646 	return retval;
647 }
648 
649 /*
650  * same_netobj: Compares the apprpriate bits of a netobj for identity
651  */
652 int
same_netobj(const netobj * n0,const netobj * n1)653 same_netobj(const netobj *n0, const netobj *n1)
654 {
655 	int retval;
656 
657 	retval = 0;
658 
659 	debuglog("Entering netobj identity check\n");
660 
661 	if (n0->n_len == n1->n_len) {
662 		debuglog("Preliminary length check passed\n");
663 		retval = !bcmp(n0->n_bytes, n1->n_bytes, n0->n_len);
664 		debuglog("netobj %smatch\n", retval ? "" : "mis");
665 	}
666 
667 	return (retval);
668 }
669 
670 /*
671  * same_filelock_identity: Compares the appropriate bits of a file_lock
672  */
673 int
same_filelock_identity(const struct file_lock * fl0,const struct file_lock * fl1)674 same_filelock_identity(const struct file_lock *fl0, const struct file_lock *fl1)
675 {
676 	int retval;
677 
678 	retval = 0;
679 
680 	debuglog("Checking filelock identity\n");
681 
682 	/*
683 	 * Check process ids and host information.
684 	 */
685 	retval = (fl0->client.svid == fl1->client.svid &&
686 	    same_netobj(&(fl0->client.oh), &(fl1->client.oh)));
687 
688 	debuglog("Exiting checking filelock identity: retval: %d\n",retval);
689 
690 	return (retval);
691 }
692 
693 /*
694  * Below here are routines associated with manipulating the NFS
695  * lock list.
696  */
697 
698 /*
699  * get_lock_matching_unlock: Return a lock which matches the given unlock lock
700  *                           or NULL otehrwise
701  * XXX: It is a shame that this duplicates so much code from test_nfslock.
702  */
703 struct file_lock *
get_lock_matching_unlock(const struct file_lock * fl)704 get_lock_matching_unlock(const struct file_lock *fl)
705 {
706 	struct file_lock *ifl; /* Iterator */
707 
708 	debuglog("Entering get_lock_matching_unlock\n");
709 	debuglog("********Dump of fl*****************\n");
710 	dump_filelock(fl);
711 
712 	LIST_FOREACH(ifl, &nfslocklist_head, nfslocklist) {
713 		debuglog("Pointer to file lock: %p\n",ifl);
714 
715 		debuglog("****Dump of ifl****\n");
716 		dump_filelock(ifl);
717 		debuglog("*******************\n");
718 
719 		/*
720 		 * XXX: It is conceivable that someone could use the NLM RPC
721 		 * system to directly access filehandles.  This may be a
722 		 * security hazard as the filehandle code may bypass normal
723 		 * file access controls
724 		 */
725 		if (bcmp(&fl->filehandle, &ifl->filehandle, sizeof(fhandle_t)))
726 			continue;
727 
728 		debuglog("get_lock_matching_unlock: Filehandles match, "
729 		    "checking regions\n");
730 
731 		/* Filehandles match, check for region overlap */
732 		if (!regions_overlap(fl->client.l_offset, fl->client.l_len,
733 			ifl->client.l_offset, ifl->client.l_len))
734 			continue;
735 
736 		debuglog("get_lock_matching_unlock: Region overlap"
737 		    " found %llu : %llu -- %llu : %llu\n",
738 		    fl->client.l_offset,fl->client.l_len,
739 		    ifl->client.l_offset,ifl->client.l_len);
740 
741 		/* Regions overlap, check the identity */
742 		if (!same_filelock_identity(fl,ifl))
743 			continue;
744 
745 		debuglog("get_lock_matching_unlock: Duplicate lock id.  Granting\n");
746 		return (ifl);
747 	}
748 
749 	debuglog("Exiting bet_lock_matching_unlock\n");
750 
751 	return (NULL);
752 }
753 
754 /*
755  * test_nfslock: check for NFS lock in lock list
756  *
757  * This routine makes the following assumptions:
758  *    1) Nothing will adjust the lock list during a lookup
759  *
760  * This routine has an interesting quirk which bit me hard.
761  * The conflicting_fl is the pointer to the conflicting lock.
762  * However, to modify the "*pointer* to the conflicting lock" rather
763  * that the "conflicting lock itself" one must pass in a "pointer to
764  * the pointer of the conflicting lock".  Gross.
765  */
766 
767 enum nfslock_status
test_nfslock(const struct file_lock * fl,struct file_lock ** conflicting_fl)768 test_nfslock(const struct file_lock *fl, struct file_lock **conflicting_fl)
769 {
770 	struct file_lock *ifl; /* Iterator */
771 	enum nfslock_status retval;
772 
773 	debuglog("Entering test_nfslock\n");
774 
775 	retval = NFS_GRANTED;
776 	(*conflicting_fl) = NULL;
777 
778 	debuglog("Entering lock search loop\n");
779 
780 	debuglog("***********************************\n");
781 	debuglog("Dumping match filelock\n");
782 	debuglog("***********************************\n");
783 	dump_filelock(fl);
784 	debuglog("***********************************\n");
785 
786 	LIST_FOREACH(ifl, &nfslocklist_head, nfslocklist) {
787 		if (retval == NFS_DENIED)
788 			break;
789 
790 		debuglog("Top of lock loop\n");
791 		debuglog("Pointer to file lock: %p\n",ifl);
792 
793 		debuglog("***********************************\n");
794 		debuglog("Dumping test filelock\n");
795 		debuglog("***********************************\n");
796 		dump_filelock(ifl);
797 		debuglog("***********************************\n");
798 
799 		/*
800 		 * XXX: It is conceivable that someone could use the NLM RPC
801 		 * system to directly access filehandles.  This may be a
802 		 * security hazard as the filehandle code may bypass normal
803 		 * file access controls
804 		 */
805 		if (bcmp(&fl->filehandle, &ifl->filehandle, sizeof(fhandle_t)))
806 			continue;
807 
808 		debuglog("test_nfslock: filehandle match found\n");
809 
810 		/* Filehandles match, check for region overlap */
811 		if (!regions_overlap(fl->client.l_offset, fl->client.l_len,
812 			ifl->client.l_offset, ifl->client.l_len))
813 			continue;
814 
815 		debuglog("test_nfslock: Region overlap found"
816 		    " %llu : %llu -- %llu : %llu\n",
817 		    fl->client.l_offset,fl->client.l_len,
818 		    ifl->client.l_offset,ifl->client.l_len);
819 
820 		/* Regions overlap, check the exclusivity */
821 		if (!(fl->client.exclusive || ifl->client.exclusive))
822 			continue;
823 
824 		debuglog("test_nfslock: Exclusivity failure: %d %d\n",
825 		    fl->client.exclusive,
826 		    ifl->client.exclusive);
827 
828 		if (same_filelock_identity(fl,ifl)) {
829 			debuglog("test_nfslock: Duplicate id.  Granting\n");
830 			(*conflicting_fl) = ifl;
831 			retval = NFS_GRANTED_DUPLICATE;
832 		} else {
833 			/* locking attempt fails */
834 			debuglog("test_nfslock: Lock attempt failed\n");
835 			debuglog("Desired lock\n");
836 			dump_filelock(fl);
837 			debuglog("Conflicting lock\n");
838 			dump_filelock(ifl);
839 			(*conflicting_fl) = ifl;
840 			retval = NFS_DENIED;
841 		}
842 	}
843 
844 	debuglog("Dumping file locks\n");
845 	debuglog("Exiting test_nfslock\n");
846 
847 	return (retval);
848 }
849 
850 /*
851  * lock_nfslock: attempt to create a lock in the NFS lock list
852  *
853  * This routine tests whether the lock will be granted and then adds
854  * the entry to the lock list if so.
855  *
856  * Argument fl gets modified as its list housekeeping entries get modified
857  * upon insertion into the NFS lock list
858  *
859  * This routine makes several assumptions:
860  *    1) It is perfectly happy to grant a duplicate lock from the same pid.
861  *       While this seems to be intuitively wrong, it is required for proper
862  *       Posix semantics during unlock.  It is absolutely imperative to not
863  *       unlock the main lock before the two child locks are established. Thus,
864  *       one has to be able to create duplicate locks over an existing lock
865  *    2) It currently accepts duplicate locks from the same id,pid
866  */
867 
868 enum nfslock_status
lock_nfslock(struct file_lock * fl)869 lock_nfslock(struct file_lock *fl)
870 {
871 	enum nfslock_status retval;
872 	struct file_lock *dummy_fl;
873 
874 	dummy_fl = NULL;
875 
876 	debuglog("Entering lock_nfslock...\n");
877 
878 	retval = test_nfslock(fl,&dummy_fl);
879 
880 	if (retval == NFS_GRANTED || retval == NFS_GRANTED_DUPLICATE) {
881 		debuglog("Inserting lock...\n");
882 		dump_filelock(fl);
883 		LIST_INSERT_HEAD(&nfslocklist_head, fl, nfslocklist);
884 	}
885 
886 	debuglog("Exiting lock_nfslock...\n");
887 
888 	return (retval);
889 }
890 
891 /*
892  * delete_nfslock: delete an NFS lock list entry
893  *
894  * This routine is used to delete a lock out of the NFS lock list
895  * without regard to status, underlying locks, regions or anything else
896  *
897  * Note that this routine *does not deallocate memory* of the lock.
898  * It just disconnects it from the list.  The lock can then be used
899  * by other routines without fear of trashing the list.
900  */
901 
902 enum nfslock_status
delete_nfslock(struct file_lock * fl)903 delete_nfslock(struct file_lock *fl)
904 {
905 
906 	LIST_REMOVE(fl, nfslocklist);
907 
908 	return (NFS_GRANTED);
909 }
910 
911 enum split_status
split_nfslock(const struct file_lock * exist_lock,const struct file_lock * unlock_lock,struct file_lock ** left_lock,struct file_lock ** right_lock)912 split_nfslock(const struct file_lock *exist_lock,
913     const struct file_lock *unlock_lock, struct file_lock **left_lock,
914     struct file_lock **right_lock)
915 {
916 	u_int64_t start1, len1, start2, len2;
917 	enum split_status spstatus;
918 
919 	spstatus = region_compare(exist_lock->client.l_offset, exist_lock->client.l_len,
920 	    unlock_lock->client.l_offset, unlock_lock->client.l_len,
921 	    &start1, &len1, &start2, &len2);
922 
923 	if ((spstatus & SPL_LOCK1) != 0) {
924 		*left_lock = allocate_file_lock(&exist_lock->client.oh, &exist_lock->client_cookie, exist_lock->addr, exist_lock->client_name);
925 		if (*left_lock == NULL) {
926 			debuglog("Unable to allocate resource for split 1\n");
927 			return SPL_RESERR;
928 		}
929 
930 		fill_file_lock(*left_lock, &exist_lock->filehandle,
931 		    exist_lock->client.exclusive, exist_lock->client.svid,
932 		    start1, len1,
933 		    exist_lock->nsm_status,
934 		    exist_lock->status, exist_lock->flags, exist_lock->blocking);
935 	}
936 
937 	if ((spstatus & SPL_LOCK2) != 0) {
938 		*right_lock = allocate_file_lock(&exist_lock->client.oh, &exist_lock->client_cookie, exist_lock->addr, exist_lock->client_name);
939 		if (*right_lock == NULL) {
940 			debuglog("Unable to allocate resource for split 1\n");
941 			if (*left_lock != NULL) {
942 				deallocate_file_lock(*left_lock);
943 			}
944 			return SPL_RESERR;
945 		}
946 
947 		fill_file_lock(*right_lock, &exist_lock->filehandle,
948 		    exist_lock->client.exclusive, exist_lock->client.svid,
949 		    start2, len2,
950 		    exist_lock->nsm_status,
951 		    exist_lock->status, exist_lock->flags, exist_lock->blocking);
952 	}
953 
954 	return spstatus;
955 }
956 
957 enum nfslock_status
unlock_nfslock(const struct file_lock * fl,struct file_lock ** released_lock,struct file_lock ** left_lock,struct file_lock ** right_lock)958 unlock_nfslock(const struct file_lock *fl, struct file_lock **released_lock,
959     struct file_lock **left_lock, struct file_lock **right_lock)
960 {
961 	struct file_lock *mfl; /* Matching file lock */
962 	enum nfslock_status retval;
963 	enum split_status spstatus;
964 
965 	debuglog("Entering unlock_nfslock\n");
966 
967 	*released_lock = NULL;
968 	*left_lock = NULL;
969 	*right_lock = NULL;
970 
971 	retval = NFS_DENIED_NOLOCK;
972 
973 	debuglog("Attempting to match lock...\n");
974 	mfl = get_lock_matching_unlock(fl);
975 
976 	if (mfl != NULL) {
977 		debuglog("Unlock matched.  Querying for split\n");
978 
979 		spstatus = split_nfslock(mfl, fl, left_lock, right_lock);
980 
981 		debuglog("Split returned %d %p %p %p %p\n",spstatus,mfl,fl,*left_lock,*right_lock);
982 		debuglog("********Split dumps********");
983 		dump_filelock(mfl);
984 		dump_filelock(fl);
985 		dump_filelock(*left_lock);
986 		dump_filelock(*right_lock);
987 		debuglog("********End Split dumps********");
988 
989 		if (spstatus == SPL_RESERR) {
990 			if (*left_lock != NULL) {
991 				deallocate_file_lock(*left_lock);
992 				*left_lock = NULL;
993 			}
994 
995 			if (*right_lock != NULL) {
996 				deallocate_file_lock(*right_lock);
997 				*right_lock = NULL;
998 			}
999 
1000 			return NFS_RESERR;
1001 		}
1002 
1003 		/* Insert new locks from split if required */
1004 		if (*left_lock != NULL) {
1005 			debuglog("Split left activated\n");
1006 			LIST_INSERT_HEAD(&nfslocklist_head, *left_lock, nfslocklist);
1007 		}
1008 
1009 		if (*right_lock != NULL) {
1010 			debuglog("Split right activated\n");
1011 			LIST_INSERT_HEAD(&nfslocklist_head, *right_lock, nfslocklist);
1012 		}
1013 
1014 		/* Unlock the lock since it matches identity */
1015 		LIST_REMOVE(mfl, nfslocklist);
1016 		*released_lock = mfl;
1017 		retval = NFS_GRANTED;
1018 	}
1019 
1020 	debuglog("Exiting unlock_nfslock\n");
1021 
1022 	return retval;
1023 }
1024 
1025 /*
1026  * Below here are the routines for manipulating the file lock directly
1027  * on the disk hardware itself
1028  */
1029 enum hwlock_status
lock_hwlock(struct file_lock * fl)1030 lock_hwlock(struct file_lock *fl)
1031 {
1032 	struct monfile *imf,*nmf;
1033 	int lflags, flerror;
1034 
1035 	/* Scan to see if filehandle already present */
1036 	LIST_FOREACH(imf, &monfilelist_head, monfilelist) {
1037 		if (bcmp(&fl->filehandle, &imf->filehandle,
1038 			sizeof(fl->filehandle)) == 0) {
1039 			/* imf is the correct filehandle */
1040 			break;
1041 		}
1042 	}
1043 
1044 	/*
1045 	 * Filehandle already exists (we control the file)
1046 	 * *AND* NFS has already cleared the lock for availability
1047 	 * Grant it and bump the refcount.
1048 	 */
1049 	if (imf != NULL) {
1050 		++(imf->refcount);
1051 		return (HW_GRANTED);
1052 	}
1053 
1054 	/* No filehandle found, create and go */
1055 	nmf = malloc(sizeof(struct monfile));
1056 	if (nmf == NULL) {
1057 		debuglog("hwlock resource allocation failure\n");
1058 		return (HW_RESERR);
1059 	}
1060 
1061 	/* XXX: Is O_RDWR always the correct mode? */
1062 	nmf->fd = fhopen(&fl->filehandle, O_RDWR);
1063 	if (nmf->fd < 0) {
1064 		debuglog("fhopen failed (from %16s): %32s\n",
1065 		    fl->client_name, strerror(errno));
1066 		free(nmf);
1067 		switch (errno) {
1068 		case ESTALE:
1069 			return (HW_STALEFH);
1070 		case EROFS:
1071 			return (HW_READONLY);
1072 		default:
1073 			return (HW_RESERR);
1074 		}
1075 	}
1076 
1077 	/* File opened correctly, fill the monitor struct */
1078 	bcopy(&fl->filehandle, &nmf->filehandle, sizeof(fl->filehandle));
1079 	nmf->refcount = 1;
1080 	nmf->exclusive = fl->client.exclusive;
1081 
1082 	lflags = (nmf->exclusive == 1) ?
1083 	    (LOCK_EX | LOCK_NB) : (LOCK_SH | LOCK_NB);
1084 
1085 	flerror = flock(nmf->fd, lflags);
1086 
1087 	if (flerror != 0) {
1088 		debuglog("flock failed (from %16s): %32s\n",
1089 		    fl->client_name, strerror(errno));
1090 		close(nmf->fd);
1091 		free(nmf);
1092 		switch (errno) {
1093 		case EAGAIN:
1094 			return (HW_DENIED);
1095 		case ESTALE:
1096 			return (HW_STALEFH);
1097 		case EROFS:
1098 			return (HW_READONLY);
1099 		default:
1100 			return (HW_RESERR);
1101 			break;
1102 		}
1103 	}
1104 
1105 	/* File opened and locked */
1106 	LIST_INSERT_HEAD(&monfilelist_head, nmf, monfilelist);
1107 
1108 	debuglog("flock succeeded (from %16s)\n", fl->client_name);
1109 	return (HW_GRANTED);
1110 }
1111 
1112 enum hwlock_status
unlock_hwlock(const struct file_lock * fl)1113 unlock_hwlock(const struct file_lock *fl)
1114 {
1115 	struct monfile *imf;
1116 
1117 	debuglog("Entering unlock_hwlock\n");
1118 	debuglog("Entering loop interation\n");
1119 
1120 	/* Scan to see if filehandle already present */
1121 	LIST_FOREACH(imf, &monfilelist_head, monfilelist) {
1122 		if (bcmp(&fl->filehandle, &imf->filehandle,
1123 			sizeof(fl->filehandle)) == 0) {
1124 			/* imf is the correct filehandle */
1125 			break;
1126 		}
1127 	}
1128 
1129 	debuglog("Completed iteration.  Proceeding\n");
1130 
1131 	if (imf == NULL) {
1132 		/* No lock found */
1133 		debuglog("Exiting unlock_hwlock (HW_DENIED_NOLOCK)\n");
1134 		return (HW_DENIED_NOLOCK);
1135 	}
1136 
1137 	/* Lock found */
1138 	--imf->refcount;
1139 
1140 	if (imf->refcount < 0) {
1141 		debuglog("Negative hardware reference count\n");
1142 	}
1143 
1144 	if (imf->refcount <= 0) {
1145 		close(imf->fd);
1146 		LIST_REMOVE(imf, monfilelist);
1147 		free(imf);
1148 	}
1149 	debuglog("Exiting unlock_hwlock (HW_GRANTED)\n");
1150 	return (HW_GRANTED);
1151 }
1152 
1153 enum hwlock_status
test_hwlock(const struct file_lock * fl __unused,struct file_lock ** conflicting_fl __unused)1154 test_hwlock(const struct file_lock *fl __unused,
1155     struct file_lock **conflicting_fl __unused)
1156 {
1157 
1158 	/*
1159 	 * XXX: lock tests on hardware are not required until
1160 	 * true partial file testing is done on the underlying file
1161 	 */
1162 	return (HW_RESERR);
1163 }
1164 
1165 
1166 
1167 /*
1168  * Below here are routines for manipulating blocked lock requests
1169  * They should only be called from the XXX_partialfilelock routines
1170  * if at all possible
1171  */
1172 
1173 int
duplicate_block(struct file_lock * fl)1174 duplicate_block(struct file_lock *fl)
1175 {
1176 	struct file_lock *ifl;
1177 	int retval = 0;
1178 
1179 	debuglog("Entering duplicate_block");
1180 
1181 	/*
1182 	 * Is this lock request already on the blocking list?
1183 	 * Consider it a dupe if the file handles, offset, length,
1184 	 * exclusivity and client match.
1185 	 */
1186 	LIST_FOREACH(ifl, &blockedlocklist_head, nfslocklist) {
1187 		if (!bcmp(&fl->filehandle, &ifl->filehandle,
1188 			sizeof(fhandle_t)) &&
1189 		    fl->client.exclusive == ifl->client.exclusive &&
1190 		    fl->client.l_offset == ifl->client.l_offset &&
1191 		    fl->client.l_len == ifl->client.l_len &&
1192 		    same_filelock_identity(fl, ifl)) {
1193 			retval = 1;
1194 			break;
1195 		}
1196 	}
1197 
1198 	debuglog("Exiting duplicate_block: %s\n", retval ? "already blocked"
1199 	    : "not already blocked");
1200 	return retval;
1201 }
1202 
1203 void
add_blockingfilelock(struct file_lock * fl)1204 add_blockingfilelock(struct file_lock *fl)
1205 {
1206 	debuglog("Entering add_blockingfilelock\n");
1207 
1208 	/*
1209 	 * A blocking lock request _should_ never be duplicated as a client
1210 	 * that is already blocked shouldn't be able to request another
1211 	 * lock. Alas, there are some buggy clients that do request the same
1212 	 * lock repeatedly. Make sure only unique locks are on the blocked
1213 	 * lock list.
1214 	 */
1215 	if (duplicate_block(fl)) {
1216 		debuglog("Exiting add_blockingfilelock: already blocked\n");
1217 		return;
1218 	}
1219 
1220 	/*
1221 	 * Clear the blocking flag so that it can be reused without
1222 	 * adding it to the blocking queue a second time
1223 	 */
1224 
1225 	fl->blocking = 0;
1226 	LIST_INSERT_HEAD(&blockedlocklist_head, fl, nfslocklist);
1227 
1228 	debuglog("Exiting add_blockingfilelock: added blocked lock\n");
1229 }
1230 
1231 void
remove_blockingfilelock(struct file_lock * fl)1232 remove_blockingfilelock(struct file_lock *fl)
1233 {
1234 
1235 	debuglog("Entering remove_blockingfilelock\n");
1236 
1237 	LIST_REMOVE(fl, nfslocklist);
1238 
1239 	debuglog("Exiting remove_blockingfilelock\n");
1240 }
1241 
1242 void
clear_blockingfilelock(const char * hostname)1243 clear_blockingfilelock(const char *hostname)
1244 {
1245 	struct file_lock *ifl,*nfl;
1246 
1247 	/*
1248 	 * Normally, LIST_FOREACH is called for, but since
1249 	 * the current element *is* the iterator, deleting it
1250 	 * would mess up the iteration.  Thus, a next element
1251 	 * must be used explicitly
1252 	 */
1253 
1254 	ifl = LIST_FIRST(&blockedlocklist_head);
1255 
1256 	while (ifl != NULL) {
1257 		nfl = LIST_NEXT(ifl, nfslocklist);
1258 
1259 		if (strncmp(hostname, ifl->client_name, SM_MAXSTRLEN) == 0) {
1260 			remove_blockingfilelock(ifl);
1261 			deallocate_file_lock(ifl);
1262 		}
1263 
1264 		ifl = nfl;
1265 	}
1266 }
1267 
1268 void
retry_blockingfilelocklist(void)1269 retry_blockingfilelocklist(void)
1270 {
1271 	/* Retry all locks in the blocked list */
1272 	struct file_lock *ifl, *nfl; /* Iterator */
1273 	enum partialfilelock_status pflstatus;
1274 
1275 	debuglog("Entering retry_blockingfilelocklist\n");
1276 
1277 	LIST_FOREACH_SAFE(ifl, &blockedlocklist_head, nfslocklist, nfl) {
1278 		debuglog("Iterator choice %p\n",ifl);
1279 		debuglog("Next iterator choice %p\n",nfl);
1280 
1281 		/*
1282 		 * SUBTLE BUG: The file_lock must be removed from the
1283 		 * old list so that it's list pointers get disconnected
1284 		 * before being allowed to participate in the new list
1285 		 * which will automatically add it in if necessary.
1286 		 */
1287 
1288 		LIST_REMOVE(ifl, nfslocklist);
1289 		pflstatus = lock_partialfilelock(ifl);
1290 
1291 		if (pflstatus == PFL_GRANTED || pflstatus == PFL_GRANTED_DUPLICATE) {
1292 			debuglog("Granted blocked lock\n");
1293 			/* lock granted and is now being used */
1294 			send_granted(ifl,0);
1295 		} else {
1296 			/* Reinsert lock back into blocked list */
1297 			debuglog("Replacing blocked lock\n");
1298 			LIST_INSERT_HEAD(&blockedlocklist_head, ifl, nfslocklist);
1299 		}
1300 	}
1301 
1302 	debuglog("Exiting retry_blockingfilelocklist\n");
1303 }
1304 
1305 /*
1306  * Below here are routines associated with manipulating all
1307  * aspects of the partial file locking system (list, hardware, etc.)
1308  */
1309 
1310 /*
1311  * Please note that lock monitoring must be done at this level which
1312  * keeps track of *individual* lock requests on lock and unlock
1313  *
1314  * XXX: Split unlocking is going to make the unlock code miserable
1315  */
1316 
1317 /*
1318  * lock_partialfilelock:
1319  *
1320  * Argument fl gets modified as its list housekeeping entries get modified
1321  * upon insertion into the NFS lock list
1322  *
1323  * This routine makes several assumptions:
1324  * 1) It (will) pass locks through to flock to lock the entire underlying file
1325  *     and then parcel out NFS locks if it gets control of the file.
1326  *         This matches the old rpc.lockd file semantics (except where it
1327  *         is now more correct).  It is the safe solution, but will cause
1328  *         overly restrictive blocking if someone is trying to use the
1329  *         underlying files without using NFS.  This appears to be an
1330  *         acceptable tradeoff since most people use standalone NFS servers.
1331  * XXX: The right solution is probably kevent combined with fcntl
1332  *
1333  *    2) Nothing modifies the lock lists between testing and granting
1334  *           I have no idea whether this is a useful assumption or not
1335  */
1336 
1337 enum partialfilelock_status
lock_partialfilelock(struct file_lock * fl)1338 lock_partialfilelock(struct file_lock *fl)
1339 {
1340 	enum partialfilelock_status retval;
1341 	enum nfslock_status lnlstatus;
1342 	enum hwlock_status hwstatus;
1343 
1344 	debuglog("Entering lock_partialfilelock\n");
1345 
1346 	retval = PFL_DENIED;
1347 
1348 	/*
1349 	 * Execute the NFS lock first, if possible, as it is significantly
1350 	 * easier and less expensive to undo than the filesystem lock
1351 	 */
1352 
1353 	lnlstatus = lock_nfslock(fl);
1354 
1355 	switch (lnlstatus) {
1356 	case NFS_GRANTED:
1357 	case NFS_GRANTED_DUPLICATE:
1358 		/*
1359 		 * At this point, the NFS lock is allocated and active.
1360 		 * Remember to clean it up if the hardware lock fails
1361 		 */
1362 		hwstatus = lock_hwlock(fl);
1363 
1364 		switch (hwstatus) {
1365 		case HW_GRANTED:
1366 		case HW_GRANTED_DUPLICATE:
1367 			debuglog("HW GRANTED\n");
1368 			/*
1369 			 * XXX: Fixme: Check hwstatus for duplicate when
1370 			 * true partial file locking and accounting is
1371 			 * done on the hardware.
1372 			 */
1373 			if (lnlstatus == NFS_GRANTED_DUPLICATE) {
1374 				retval = PFL_GRANTED_DUPLICATE;
1375 			} else {
1376 				retval = PFL_GRANTED;
1377 			}
1378 			monitor_lock_host(fl->client_name);
1379 			break;
1380 		case HW_RESERR:
1381 			debuglog("HW RESERR\n");
1382 			retval = PFL_HWRESERR;
1383 			break;
1384 		case HW_DENIED:
1385 			debuglog("HW DENIED\n");
1386 			retval = PFL_HWDENIED;
1387 			break;
1388 		default:
1389 			debuglog("Unmatched hwstatus %d\n",hwstatus);
1390 			break;
1391 		}
1392 
1393 		if (retval != PFL_GRANTED &&
1394 		    retval != PFL_GRANTED_DUPLICATE) {
1395 			/* Clean up the NFS lock */
1396 			debuglog("Deleting trial NFS lock\n");
1397 			delete_nfslock(fl);
1398 		}
1399 		break;
1400 	case NFS_DENIED:
1401 		retval = PFL_NFSDENIED;
1402 		break;
1403 	case NFS_RESERR:
1404 		retval = PFL_NFSRESERR;
1405 		break;
1406 	default:
1407 		debuglog("Unmatched lnlstatus %d\n");
1408 		retval = PFL_NFSDENIED_NOLOCK;
1409 		break;
1410 	}
1411 
1412 	/*
1413 	 * By the time fl reaches here, it is completely free again on
1414 	 * failure.  The NFS lock done before attempting the
1415 	 * hardware lock has been backed out
1416 	 */
1417 
1418 	if (retval == PFL_NFSDENIED || retval == PFL_HWDENIED) {
1419 		/* Once last chance to check the lock */
1420 		if (fl->blocking == 1) {
1421 			if (retval == PFL_NFSDENIED) {
1422 				/* Queue the lock */
1423 				debuglog("BLOCKING LOCK RECEIVED\n");
1424 				retval = PFL_NFSBLOCKED;
1425 				add_blockingfilelock(fl);
1426 				dump_filelock(fl);
1427 			} else {
1428 				/* retval is okay as PFL_HWDENIED */
1429 				debuglog("BLOCKING LOCK DENIED IN HARDWARE\n");
1430 				dump_filelock(fl);
1431 			}
1432 		} else {
1433 			/* Leave retval alone, it's already correct */
1434 			debuglog("Lock denied.  Non-blocking failure\n");
1435 			dump_filelock(fl);
1436 		}
1437 	}
1438 
1439 	debuglog("Exiting lock_partialfilelock\n");
1440 
1441 	return retval;
1442 }
1443 
1444 /*
1445  * unlock_partialfilelock:
1446  *
1447  * Given a file_lock, unlock all locks which match.
1448  *
1449  * Note that a given lock might have to unlock ITSELF!  See
1450  * clear_partialfilelock for example.
1451  */
1452 
1453 enum partialfilelock_status
unlock_partialfilelock(const struct file_lock * fl)1454 unlock_partialfilelock(const struct file_lock *fl)
1455 {
1456 	struct file_lock *lfl,*rfl,*releasedfl,*selffl;
1457 	enum partialfilelock_status retval;
1458 	enum nfslock_status unlstatus;
1459 	enum hwlock_status unlhwstatus, lhwstatus;
1460 
1461 	debuglog("Entering unlock_partialfilelock\n");
1462 
1463 	selffl = NULL;
1464 	lfl = NULL;
1465 	rfl = NULL;
1466 	releasedfl = NULL;
1467 	retval = PFL_DENIED;
1468 
1469 	/*
1470 	 * There are significant overlap and atomicity issues
1471 	 * with partially releasing a lock.  For example, releasing
1472 	 * part of an NFS shared lock does *not* always release the
1473 	 * corresponding part of the file since there is only one
1474 	 * rpc.lockd UID but multiple users could be requesting it
1475 	 * from NFS.  Also, an unlock request should never allow
1476 	 * another process to gain a lock on the remaining parts.
1477 	 * ie. Always apply the new locks before releasing the
1478 	 * old one
1479 	 */
1480 
1481 	/*
1482 	 * Loop is required since multiple little locks
1483 	 * can be allocated and then deallocated with one
1484 	 * big unlock.
1485 	 *
1486 	 * The loop is required to be here so that the nfs &
1487 	 * hw subsystems do not need to communicate with one
1488 	 * one another
1489 	 */
1490 
1491 	do {
1492 		debuglog("Value of releasedfl: %p\n",releasedfl);
1493 		/* lfl&rfl are created *AND* placed into the NFS lock list if required */
1494 		unlstatus = unlock_nfslock(fl, &releasedfl, &lfl, &rfl);
1495 		debuglog("Value of releasedfl: %p\n",releasedfl);
1496 
1497 
1498 		/* XXX: This is grungy.  It should be refactored to be cleaner */
1499 		if (lfl != NULL) {
1500 			lhwstatus = lock_hwlock(lfl);
1501 			if (lhwstatus != HW_GRANTED &&
1502 			    lhwstatus != HW_GRANTED_DUPLICATE) {
1503 				debuglog("HW duplicate lock failure for left split\n");
1504 			}
1505 			monitor_lock_host(lfl->client_name);
1506 		}
1507 
1508 		if (rfl != NULL) {
1509 			lhwstatus = lock_hwlock(rfl);
1510 			if (lhwstatus != HW_GRANTED &&
1511 			    lhwstatus != HW_GRANTED_DUPLICATE) {
1512 				debuglog("HW duplicate lock failure for right split\n");
1513 			}
1514 			monitor_lock_host(rfl->client_name);
1515 		}
1516 
1517 		switch (unlstatus) {
1518 		case NFS_GRANTED:
1519 			/* Attempt to unlock on the hardware */
1520 			debuglog("NFS unlock granted.  Attempting hardware unlock\n");
1521 
1522 			/* This call *MUST NOT* unlock the two newly allocated locks */
1523 			unlhwstatus = unlock_hwlock(fl);
1524 			debuglog("HW unlock returned with code %d\n",unlhwstatus);
1525 
1526 			switch (unlhwstatus) {
1527 			case HW_GRANTED:
1528 				debuglog("HW unlock granted\n");
1529 				unmonitor_lock_host(releasedfl->client_name);
1530 				retval = PFL_GRANTED;
1531 				break;
1532 			case HW_DENIED_NOLOCK:
1533 				/* Huh?!?!  This shouldn't happen */
1534 				debuglog("HW unlock denied no lock\n");
1535 				retval = PFL_HWRESERR;
1536 				/* Break out of do-while */
1537 				unlstatus = NFS_RESERR;
1538 				break;
1539 			default:
1540 				debuglog("HW unlock failed\n");
1541 				retval = PFL_HWRESERR;
1542 				/* Break out of do-while */
1543 				unlstatus = NFS_RESERR;
1544 				break;
1545 			}
1546 
1547 			debuglog("Exiting with status retval: %d\n",retval);
1548 
1549 			retry_blockingfilelocklist();
1550 			break;
1551 		case NFS_DENIED_NOLOCK:
1552 			retval = PFL_GRANTED;
1553 			debuglog("All locks cleaned out\n");
1554 			break;
1555 		default:
1556 			retval = PFL_NFSRESERR;
1557 			debuglog("NFS unlock failure\n");
1558 			dump_filelock(fl);
1559 			break;
1560 		}
1561 
1562 		if (releasedfl != NULL) {
1563 			if (fl == releasedfl) {
1564 				/*
1565 				 * XXX: YECHHH!!! Attempt to unlock self succeeded
1566 				 * but we can't deallocate the space yet.  This is what
1567 				 * happens when you don't write malloc and free together
1568 				 */
1569 				debuglog("Attempt to unlock self\n");
1570 				selffl = releasedfl;
1571 			} else {
1572 				/*
1573 				 * XXX: this deallocation *still* needs to migrate closer
1574 				 * to the allocation code way up in get_lock or the allocation
1575 				 * code needs to migrate down (violation of "When you write
1576 				 * malloc you must write free")
1577 				 */
1578 
1579 				deallocate_file_lock(releasedfl);
1580 				releasedfl = NULL;
1581 			}
1582 		}
1583 
1584 	} while (unlstatus == NFS_GRANTED);
1585 
1586 	if (selffl != NULL) {
1587 		/*
1588 		 * This statement wipes out the incoming file lock (fl)
1589 		 * in spite of the fact that it is declared const
1590 		 */
1591 		debuglog("WARNING!  Destroying incoming lock pointer\n");
1592 		deallocate_file_lock(selffl);
1593 	}
1594 
1595 	debuglog("Exiting unlock_partialfilelock\n");
1596 
1597 	return retval;
1598 }
1599 
1600 /*
1601  * clear_partialfilelock
1602  *
1603  * Normally called in response to statd state number change.
1604  * Wipe out all locks held by a host.  As a bonus, the act of
1605  * doing so should automatically clear their statd entries and
1606  * unmonitor the host.
1607  */
1608 
1609 void
clear_partialfilelock(const char * hostname)1610 clear_partialfilelock(const char *hostname)
1611 {
1612 	struct file_lock *ifl, *nfl;
1613 
1614 	/* Clear blocking file lock list */
1615 	clear_blockingfilelock(hostname);
1616 
1617 	/* do all required unlocks */
1618 	/* Note that unlock can smash the current pointer to a lock */
1619 
1620 	/*
1621 	 * Normally, LIST_FOREACH is called for, but since
1622 	 * the current element *is* the iterator, deleting it
1623 	 * would mess up the iteration.  Thus, a next element
1624 	 * must be used explicitly
1625 	 */
1626 
1627 	ifl = LIST_FIRST(&nfslocklist_head);
1628 
1629 	while (ifl != NULL) {
1630 		nfl = LIST_NEXT(ifl, nfslocklist);
1631 
1632 		if (strncmp(hostname, ifl->client_name, SM_MAXSTRLEN) == 0) {
1633 			/* Unlock destroys ifl out from underneath */
1634 			unlock_partialfilelock(ifl);
1635 			/* ifl is NO LONGER VALID AT THIS POINT */
1636 		}
1637 		ifl = nfl;
1638 	}
1639 }
1640 
1641 /*
1642  * test_partialfilelock:
1643  */
1644 enum partialfilelock_status
test_partialfilelock(const struct file_lock * fl,struct file_lock ** conflicting_fl)1645 test_partialfilelock(const struct file_lock *fl,
1646     struct file_lock **conflicting_fl)
1647 {
1648 	enum partialfilelock_status retval;
1649 	enum nfslock_status teststatus;
1650 
1651 	debuglog("Entering testpartialfilelock...\n");
1652 
1653 	retval = PFL_DENIED;
1654 
1655 	teststatus = test_nfslock(fl, conflicting_fl);
1656 	debuglog("test_partialfilelock: teststatus %d\n",teststatus);
1657 
1658 	if (teststatus == NFS_GRANTED || teststatus == NFS_GRANTED_DUPLICATE) {
1659 		/* XXX: Add the underlying filesystem locking code */
1660 		retval = (teststatus == NFS_GRANTED) ?
1661 		    PFL_GRANTED : PFL_GRANTED_DUPLICATE;
1662 		debuglog("Dumping locks...\n");
1663 		dump_filelock(fl);
1664 		dump_filelock(*conflicting_fl);
1665 		debuglog("Done dumping locks...\n");
1666 	} else {
1667 		retval = PFL_NFSDENIED;
1668 		debuglog("NFS test denied.\n");
1669 		dump_filelock(fl);
1670 		debuglog("Conflicting.\n");
1671 		dump_filelock(*conflicting_fl);
1672 	}
1673 
1674 	debuglog("Exiting testpartialfilelock...\n");
1675 
1676 	return retval;
1677 }
1678 
1679 /*
1680  * Below here are routines associated with translating the partial file locking
1681  * codes into useful codes to send back to the NFS RPC messaging system
1682  */
1683 
1684 /*
1685  * These routines translate the (relatively) useful return codes back onto
1686  * the few return codes which the nlm subsystems wishes to trasmit
1687  */
1688 
1689 enum nlm_stats
do_test(struct file_lock * fl,struct file_lock ** conflicting_fl)1690 do_test(struct file_lock *fl, struct file_lock **conflicting_fl)
1691 {
1692 	enum partialfilelock_status pfsret;
1693 	enum nlm_stats retval;
1694 
1695 	debuglog("Entering do_test...\n");
1696 
1697 	pfsret = test_partialfilelock(fl,conflicting_fl);
1698 
1699 	switch (pfsret) {
1700 	case PFL_GRANTED:
1701 		debuglog("PFL test lock granted\n");
1702 		dump_filelock(fl);
1703 		dump_filelock(*conflicting_fl);
1704 		retval = (fl->flags & LOCK_V4) ? nlm4_granted : nlm_granted;
1705 		break;
1706 	case PFL_GRANTED_DUPLICATE:
1707 		debuglog("PFL test lock granted--duplicate id detected\n");
1708 		dump_filelock(fl);
1709 		dump_filelock(*conflicting_fl);
1710 		debuglog("Clearing conflicting_fl for call semantics\n");
1711 		*conflicting_fl = NULL;
1712 		retval = (fl->flags & LOCK_V4) ? nlm4_granted : nlm_granted;
1713 		break;
1714 	case PFL_NFSDENIED:
1715 	case PFL_HWDENIED:
1716 		debuglog("PFL test lock denied\n");
1717 		dump_filelock(fl);
1718 		dump_filelock(*conflicting_fl);
1719 		retval = (fl->flags & LOCK_V4) ? nlm4_denied : nlm_denied;
1720 		break;
1721 	case PFL_NFSRESERR:
1722 	case PFL_HWRESERR:
1723 		debuglog("PFL test lock resource fail\n");
1724 		dump_filelock(fl);
1725 		dump_filelock(*conflicting_fl);
1726 		retval = (fl->flags & LOCK_V4) ? nlm4_denied_nolocks : nlm_denied_nolocks;
1727 		break;
1728 	default:
1729 		debuglog("PFL test lock *FAILED*\n");
1730 		dump_filelock(fl);
1731 		dump_filelock(*conflicting_fl);
1732 		retval = (fl->flags & LOCK_V4) ? nlm4_failed : nlm_denied;
1733 		break;
1734 	}
1735 
1736 	debuglog("Exiting do_test...\n");
1737 
1738 	return retval;
1739 }
1740 
1741 /*
1742  * do_lock: Try to acquire a lock
1743  *
1744  * This routine makes a distinction between NLM versions.  I am pretty
1745  * convinced that this should be abstracted out and bounced up a level
1746  */
1747 
1748 enum nlm_stats
do_lock(struct file_lock * fl)1749 do_lock(struct file_lock *fl)
1750 {
1751 	enum partialfilelock_status pfsret;
1752 	enum nlm_stats retval;
1753 
1754 	debuglog("Entering do_lock...\n");
1755 
1756 	pfsret = lock_partialfilelock(fl);
1757 
1758 	switch (pfsret) {
1759 	case PFL_GRANTED:
1760 		debuglog("PFL lock granted");
1761 		dump_filelock(fl);
1762 		retval = (fl->flags & LOCK_V4) ? nlm4_granted : nlm_granted;
1763 		break;
1764 	case PFL_GRANTED_DUPLICATE:
1765 		debuglog("PFL lock granted--duplicate id detected");
1766 		dump_filelock(fl);
1767 		retval = (fl->flags & LOCK_V4) ? nlm4_granted : nlm_granted;
1768 		break;
1769 	case PFL_NFSDENIED:
1770 	case PFL_HWDENIED:
1771 		debuglog("PFL_NFS lock denied");
1772 		dump_filelock(fl);
1773 		retval = (fl->flags & LOCK_V4) ? nlm4_denied : nlm_denied;
1774 		break;
1775 	case PFL_NFSBLOCKED:
1776 	case PFL_HWBLOCKED:
1777 		debuglog("PFL_NFS blocking lock denied.  Queued.\n");
1778 		dump_filelock(fl);
1779 		retval = (fl->flags & LOCK_V4) ? nlm4_blocked : nlm_blocked;
1780 		break;
1781 	case PFL_NFSRESERR:
1782 	case PFL_HWRESERR:
1783 		debuglog("PFL lock resource alocation fail\n");
1784 		dump_filelock(fl);
1785 		retval = (fl->flags & LOCK_V4) ? nlm4_denied_nolocks : nlm_denied_nolocks;
1786 		break;
1787 	default:
1788 		debuglog("PFL lock *FAILED*");
1789 		dump_filelock(fl);
1790 		retval = (fl->flags & LOCK_V4) ? nlm4_failed : nlm_denied;
1791 		break;
1792 	}
1793 
1794 	debuglog("Exiting do_lock...\n");
1795 
1796 	return retval;
1797 }
1798 
1799 enum nlm_stats
do_unlock(struct file_lock * fl)1800 do_unlock(struct file_lock *fl)
1801 {
1802 	enum partialfilelock_status pfsret;
1803 	enum nlm_stats retval;
1804 
1805 	debuglog("Entering do_unlock...\n");
1806 	pfsret = unlock_partialfilelock(fl);
1807 
1808 	switch (pfsret) {
1809 	case PFL_GRANTED:
1810 		debuglog("PFL unlock granted");
1811 		dump_filelock(fl);
1812 		retval = (fl->flags & LOCK_V4) ? nlm4_granted : nlm_granted;
1813 		break;
1814 	case PFL_NFSDENIED:
1815 	case PFL_HWDENIED:
1816 		debuglog("PFL_NFS unlock denied");
1817 		dump_filelock(fl);
1818 		retval = (fl->flags & LOCK_V4) ? nlm4_denied : nlm_denied;
1819 		break;
1820 	case PFL_NFSDENIED_NOLOCK:
1821 	case PFL_HWDENIED_NOLOCK:
1822 		debuglog("PFL_NFS no lock found\n");
1823 		retval = (fl->flags & LOCK_V4) ? nlm4_granted : nlm_granted;
1824 		break;
1825 	case PFL_NFSRESERR:
1826 	case PFL_HWRESERR:
1827 		debuglog("PFL unlock resource failure");
1828 		dump_filelock(fl);
1829 		retval = (fl->flags & LOCK_V4) ? nlm4_denied_nolocks : nlm_denied_nolocks;
1830 		break;
1831 	default:
1832 		debuglog("PFL unlock *FAILED*");
1833 		dump_filelock(fl);
1834 		retval = (fl->flags & LOCK_V4) ? nlm4_failed : nlm_denied;
1835 		break;
1836 	}
1837 
1838 	debuglog("Exiting do_unlock...\n");
1839 
1840 	return retval;
1841 }
1842 
1843 /*
1844  * do_clear
1845  *
1846  * This routine is non-existent because it doesn't have a return code.
1847  * It is here for completeness in case someone *does* need to do return
1848  * codes later.  A decent compiler should optimize this away.
1849  */
1850 
1851 void
do_clear(const char * hostname)1852 do_clear(const char *hostname)
1853 {
1854 
1855 	clear_partialfilelock(hostname);
1856 }
1857 
1858 /*
1859  * The following routines are all called from the code which the
1860  * RPC layer invokes
1861  */
1862 
1863 /*
1864  * testlock(): inform the caller if the requested lock would be granted
1865  *
1866  * returns NULL if lock would granted
1867  * returns pointer to a conflicting nlm4_holder if not
1868  */
1869 
1870 struct nlm4_holder *
testlock(struct nlm4_lock * lock,bool_t exclusive,int flags __unused)1871 testlock(struct nlm4_lock *lock, bool_t exclusive, int flags __unused)
1872 {
1873 	struct file_lock test_fl, *conflicting_fl;
1874 
1875 	bzero(&test_fl, sizeof(test_fl));
1876 
1877 	bcopy(lock->fh.n_bytes, &(test_fl.filehandle), sizeof(fhandle_t));
1878 	copy_nlm4_lock_to_nlm4_holder(lock, exclusive, &test_fl.client);
1879 
1880 	siglock();
1881 	do_test(&test_fl, &conflicting_fl);
1882 
1883 	if (conflicting_fl == NULL) {
1884 		debuglog("No conflicting lock found\n");
1885 		sigunlock();
1886 		return NULL;
1887 	} else {
1888 		debuglog("Found conflicting lock\n");
1889 		dump_filelock(conflicting_fl);
1890 		sigunlock();
1891 		return (&conflicting_fl->client);
1892 	}
1893 }
1894 
1895 /*
1896  * getlock: try to acquire the lock.
1897  * If file is already locked and we can sleep, put the lock in the list with
1898  * status LKST_WAITING; it'll be processed later.
1899  * Otherwise try to lock. If we're allowed to block, fork a child which
1900  * will do the blocking lock.
1901  */
1902 
1903 enum nlm_stats
getlock(nlm4_lockargs * lckarg,struct svc_req * rqstp,const int flags)1904 getlock(nlm4_lockargs *lckarg, struct svc_req *rqstp, const int flags)
1905 {
1906 	struct file_lock *newfl;
1907 	enum nlm_stats retval;
1908 
1909 	debuglog("Entering getlock...\n");
1910 
1911 	if (grace_expired == 0 && lckarg->reclaim == 0)
1912 		return (flags & LOCK_V4) ?
1913 		    nlm4_denied_grace_period : nlm_denied_grace_period;
1914 
1915 	/* allocate new file_lock for this request */
1916 	newfl = allocate_file_lock(&lckarg->alock.oh, &lckarg->cookie,
1917 				   (struct sockaddr *)svc_getrpccaller(rqstp->rq_xprt)->buf, lckarg->alock.caller_name);
1918 	if (newfl == NULL) {
1919 		syslog(LOG_NOTICE, "lock allocate failed: %s", strerror(errno));
1920 		/* failed */
1921 		return (flags & LOCK_V4) ?
1922 		    nlm4_denied_nolocks : nlm_denied_nolocks;
1923 	}
1924 
1925 	if (lckarg->alock.fh.n_len != sizeof(fhandle_t)) {
1926 		debuglog("received fhandle size %d, local size %d",
1927 		    lckarg->alock.fh.n_len, (int)sizeof(fhandle_t));
1928 	}
1929 
1930 	fill_file_lock(newfl, (fhandle_t *)lckarg->alock.fh.n_bytes,
1931 	    lckarg->exclusive, lckarg->alock.svid, lckarg->alock.l_offset,
1932 	    lckarg->alock.l_len,
1933 	    lckarg->state, 0, flags, lckarg->block);
1934 
1935 	/*
1936 	 * newfl is now fully constructed and deallocate_file_lock
1937 	 * can now be used to delete it
1938 	 */
1939 
1940 	siglock();
1941 	debuglog("Pointer to new lock is %p\n",newfl);
1942 
1943 	retval = do_lock(newfl);
1944 
1945 	debuglog("Pointer to new lock is %p\n",newfl);
1946 	sigunlock();
1947 
1948 	switch (retval)
1949 		{
1950 		case nlm4_granted:
1951 			/* case nlm_granted: is the same as nlm4_granted */
1952 			/* do_mon(lckarg->alock.caller_name); */
1953 			break;
1954 		case nlm4_blocked:
1955 			/* case nlm_blocked: is the same as nlm4_blocked */
1956 			/* do_mon(lckarg->alock.caller_name); */
1957 			break;
1958 		default:
1959 			deallocate_file_lock(newfl);
1960 			break;
1961 		}
1962 
1963 	debuglog("Exiting getlock...\n");
1964 
1965 	return retval;
1966 }
1967 
1968 
1969 /* unlock a filehandle */
1970 enum nlm_stats
unlock(nlm4_lock * lock,const int flags __unused)1971 unlock(nlm4_lock *lock, const int flags __unused)
1972 {
1973 	struct file_lock fl;
1974 	enum nlm_stats err;
1975 
1976 	siglock();
1977 
1978 	debuglog("Entering unlock...\n");
1979 
1980 	bzero(&fl,sizeof(struct file_lock));
1981 	bcopy(lock->fh.n_bytes, &fl.filehandle, sizeof(fhandle_t));
1982 
1983 	copy_nlm4_lock_to_nlm4_holder(lock, 0, &fl.client);
1984 
1985 	err = do_unlock(&fl);
1986 
1987 	sigunlock();
1988 
1989 	debuglog("Exiting unlock...\n");
1990 
1991 	return err;
1992 }
1993 
1994 /*
1995  * XXX: The following monitor/unmonitor routines
1996  * have not been extensively tested (ie. no regression
1997  * script exists like for the locking sections
1998  */
1999 
2000 /*
2001  * monitor_lock_host: monitor lock hosts locally with a ref count and
2002  * inform statd
2003  */
2004 void
monitor_lock_host(const char * hostname)2005 monitor_lock_host(const char *hostname)
2006 {
2007 	struct host *ihp, *nhp;
2008 	struct mon smon;
2009 	struct sm_stat_res sres;
2010 	int rpcret, statflag;
2011 	size_t n;
2012 
2013 	rpcret = 0;
2014 	statflag = 0;
2015 
2016 	LIST_FOREACH(ihp, &hostlst_head, hostlst) {
2017 		if (strncmp(hostname, ihp->name, SM_MAXSTRLEN) == 0) {
2018 			/* Host is already monitored, bump refcount */
2019 			++ihp->refcnt;
2020 			/* Host should only be in the monitor list once */
2021 			return;
2022 		}
2023 	}
2024 
2025 	/* Host is not yet monitored, add it */
2026 	n = strnlen(hostname, SM_MAXSTRLEN);
2027 	if (n == SM_MAXSTRLEN) {
2028 		return;
2029 	}
2030 	nhp = malloc(sizeof(*nhp) - sizeof(nhp->name) + n + 1);
2031 	if (nhp == NULL) {
2032 		debuglog("Unable to allocate entry for statd mon\n");
2033 		return;
2034 	}
2035 
2036 	/* Allocated new host entry, now fill the fields */
2037 	memcpy(nhp->name, hostname, n);
2038 	nhp->name[n] = 0;
2039 	nhp->refcnt = 1;
2040 	debuglog("Locally Monitoring host %16s\n",hostname);
2041 
2042 	debuglog("Attempting to tell statd\n");
2043 
2044 	bzero(&smon,sizeof(smon));
2045 
2046 	smon.mon_id.mon_name = nhp->name;
2047 	smon.mon_id.my_id.my_name = "localhost";
2048 	smon.mon_id.my_id.my_prog = NLM_PROG;
2049 	smon.mon_id.my_id.my_vers = NLM_SM;
2050 	smon.mon_id.my_id.my_proc = NLM_SM_NOTIFY;
2051 
2052 	rpcret = callrpc("localhost", SM_PROG, SM_VERS, SM_MON,
2053 	    (xdrproc_t)xdr_mon, &smon,
2054 	    (xdrproc_t)xdr_sm_stat_res, &sres);
2055 
2056 	if (rpcret == 0) {
2057 		if (sres.res_stat == stat_fail) {
2058 			debuglog("Statd call failed\n");
2059 			statflag = 0;
2060 		} else {
2061 			statflag = 1;
2062 		}
2063 	} else {
2064 		debuglog("Rpc call to statd failed with return value: %d\n",
2065 		    rpcret);
2066 		statflag = 0;
2067 	}
2068 
2069 	if (statflag == 1) {
2070 		LIST_INSERT_HEAD(&hostlst_head, nhp, hostlst);
2071 	} else {
2072 		free(nhp);
2073 	}
2074 
2075 }
2076 
2077 /*
2078  * unmonitor_lock_host: clear monitor ref counts and inform statd when gone
2079  */
2080 void
unmonitor_lock_host(char * hostname)2081 unmonitor_lock_host(char *hostname)
2082 {
2083 	struct host *ihp;
2084 	struct mon_id smon_id;
2085 	struct sm_stat smstat;
2086 	int rpcret;
2087 
2088 	rpcret = 0;
2089 
2090 	for( ihp=LIST_FIRST(&hostlst_head); ihp != NULL;
2091 	     ihp=LIST_NEXT(ihp, hostlst)) {
2092 		if (strncmp(hostname, ihp->name, SM_MAXSTRLEN) == 0) {
2093 			/* Host is monitored, bump refcount */
2094 			--ihp->refcnt;
2095 			/* Host should only be in the monitor list once */
2096 			break;
2097 		}
2098 	}
2099 
2100 	if (ihp == NULL) {
2101 		debuglog("Could not find host %16s in mon list\n", hostname);
2102 		return;
2103 	}
2104 
2105 	if (ihp->refcnt > 0)
2106 		return;
2107 
2108 	if (ihp->refcnt < 0) {
2109 		debuglog("Negative refcount!: %d\n",
2110 		    ihp->refcnt);
2111 	}
2112 
2113 	debuglog("Attempting to unmonitor host %16s\n", hostname);
2114 
2115 	bzero(&smon_id,sizeof(smon_id));
2116 
2117 	smon_id.mon_name = hostname;
2118 	smon_id.my_id.my_name = "localhost";
2119 	smon_id.my_id.my_prog = NLM_PROG;
2120 	smon_id.my_id.my_vers = NLM_SM;
2121 	smon_id.my_id.my_proc = NLM_SM_NOTIFY;
2122 
2123 	rpcret = callrpc("localhost", SM_PROG, SM_VERS, SM_UNMON,
2124 	    (xdrproc_t)xdr_mon_id, &smon_id,
2125 	    (xdrproc_t)xdr_sm_stat, &smstat);
2126 
2127 	if (rpcret != 0) {
2128 		debuglog("Rpc call to unmonitor statd failed with "
2129 		   " return value: %d\n", rpcret);
2130 	}
2131 
2132 	LIST_REMOVE(ihp, hostlst);
2133 	free(ihp);
2134 }
2135 
2136 /*
2137  * notify: Clear all locks from a host if statd complains
2138  *
2139  * XXX: This routine has not been thoroughly tested.  However, neither
2140  * had the old one been.  It used to compare the statd crash state counter
2141  * to the current lock state.  The upshot of this was that it basically
2142  * cleared all locks from the specified host 99% of the time (with the
2143  * other 1% being a bug).  Consequently, the assumption is that clearing
2144  * all locks from a host when notified by statd is acceptable.
2145  *
2146  * Please note that this routine skips the usual level of redirection
2147  * through a do_* type routine.  This introduces a possible level of
2148  * error and might better be written as do_notify and take this one out.
2149 
2150  */
2151 
2152 void
notify(const char * hostname,const int state)2153 notify(const char *hostname, const int state)
2154 {
2155 	debuglog("notify from %s, new state %d", hostname, state);
2156 
2157 	siglock();
2158 	do_clear(hostname);
2159 	sigunlock();
2160 
2161 	debuglog("Leaving notify\n");
2162 }
2163 
2164 void
send_granted(struct file_lock * fl,int opcode __unused)2165 send_granted(struct file_lock *fl, int opcode __unused)
2166 {
2167 	CLIENT *cli;
2168 	static char dummy;
2169 	struct timeval timeo;
2170 	int success;
2171 	static struct nlm_res retval;
2172 	static struct nlm4_res retval4;
2173 
2174 	debuglog("About to send granted on blocked lock\n");
2175 
2176 	cli = get_client(fl->addr,
2177 	    (fl->flags & LOCK_V4) ? NLM_VERS4 : NLM_VERS);
2178 	if (cli == NULL) {
2179 		syslog(LOG_NOTICE, "failed to get CLIENT for %s",
2180 		    fl->client_name);
2181 		/*
2182 		 * We fail to notify remote that the lock has been granted.
2183 		 * The client will timeout and retry, the lock will be
2184 		 * granted at this time.
2185 		 */
2186 		return;
2187 	}
2188 	timeo.tv_sec = 0;
2189 	timeo.tv_usec = (fl->flags & LOCK_ASYNC) ? 0 : 500000; /* 0.5s */
2190 
2191 	if (fl->flags & LOCK_V4) {
2192 		static nlm4_testargs res;
2193 		res.cookie = fl->client_cookie;
2194 		res.exclusive = fl->client.exclusive;
2195 		res.alock.caller_name = fl->client_name;
2196 		res.alock.fh.n_len = sizeof(fhandle_t);
2197 		res.alock.fh.n_bytes = (char*)&fl->filehandle;
2198 		res.alock.oh = fl->client.oh;
2199 		res.alock.svid = fl->client.svid;
2200 		res.alock.l_offset = fl->client.l_offset;
2201 		res.alock.l_len = fl->client.l_len;
2202 		debuglog("sending v4 reply%s",
2203 			 (fl->flags & LOCK_ASYNC) ? " (async)":"");
2204 		if (fl->flags & LOCK_ASYNC) {
2205 			success = clnt_call(cli, NLM4_GRANTED_MSG,
2206 			    (xdrproc_t)xdr_nlm4_testargs, &res,
2207 			    (xdrproc_t)xdr_void, &dummy, timeo);
2208 		} else {
2209 			success = clnt_call(cli, NLM4_GRANTED,
2210 			    (xdrproc_t)xdr_nlm4_testargs, &res,
2211 			    (xdrproc_t)xdr_nlm4_res, &retval4, timeo);
2212 		}
2213 	} else {
2214 		static nlm_testargs res;
2215 
2216 		res.cookie = fl->client_cookie;
2217 		res.exclusive = fl->client.exclusive;
2218 		res.alock.caller_name = fl->client_name;
2219 		res.alock.fh.n_len = sizeof(fhandle_t);
2220 		res.alock.fh.n_bytes = (char*)&fl->filehandle;
2221 		res.alock.oh = fl->client.oh;
2222 		res.alock.svid = fl->client.svid;
2223 		res.alock.l_offset = fl->client.l_offset;
2224 		res.alock.l_len = fl->client.l_len;
2225 		debuglog("sending v1 reply%s",
2226 			 (fl->flags & LOCK_ASYNC) ? " (async)":"");
2227 		if (fl->flags & LOCK_ASYNC) {
2228 			success = clnt_call(cli, NLM_GRANTED_MSG,
2229 			    (xdrproc_t)xdr_nlm_testargs, &res,
2230 			    (xdrproc_t)xdr_void, &dummy, timeo);
2231 		} else {
2232 			success = clnt_call(cli, NLM_GRANTED,
2233 			    (xdrproc_t)xdr_nlm_testargs, &res,
2234 			    (xdrproc_t)xdr_nlm_res, &retval, timeo);
2235 		}
2236 	}
2237 	if (debug_level > 2)
2238 		debuglog("clnt_call returns %d(%s) for granted",
2239 			 success, clnt_sperrno(success));
2240 
2241 }
2242 
2243 /*
2244  * Routines below here have not been modified in the overhaul
2245  */
2246 
2247 /*
2248  * Are these two routines still required since lockd is not spawning off
2249  * children to service locks anymore?  Presumably they were originally
2250  * put in place to prevent a one child from changing the lock list out
2251  * from under another one.
2252  */
2253 
2254 void
siglock(void)2255 siglock(void)
2256 {
2257   sigset_t block;
2258 
2259   sigemptyset(&block);
2260   sigaddset(&block, SIGCHLD);
2261 
2262   if (sigprocmask(SIG_BLOCK, &block, NULL) < 0) {
2263     syslog(LOG_WARNING, "siglock failed: %s", strerror(errno));
2264   }
2265 }
2266 
2267 void
sigunlock(void)2268 sigunlock(void)
2269 {
2270   sigset_t block;
2271 
2272   sigemptyset(&block);
2273   sigaddset(&block, SIGCHLD);
2274 
2275   if (sigprocmask(SIG_UNBLOCK, &block, NULL) < 0) {
2276     syslog(LOG_WARNING, "sigunlock failed: %s", strerror(errno));
2277   }
2278 }
2279