vfs_lockf.c revision 1.21 1 1.21 thorpej /* $NetBSD: vfs_lockf.c,v 1.21 2003/01/18 10:06:37 thorpej Exp $ */
2 1.5 cgd
3 1.1 ws /*
4 1.4 mycroft * Copyright (c) 1982, 1986, 1989, 1993
5 1.4 mycroft * The Regents of the University of California. All rights reserved.
6 1.1 ws *
7 1.1 ws * This code is derived from software contributed to Berkeley by
8 1.1 ws * Scooter Morris at Genentech Inc.
9 1.1 ws *
10 1.1 ws * Redistribution and use in source and binary forms, with or without
11 1.1 ws * modification, are permitted provided that the following conditions
12 1.1 ws * are met:
13 1.1 ws * 1. Redistributions of source code must retain the above copyright
14 1.1 ws * notice, this list of conditions and the following disclaimer.
15 1.1 ws * 2. Redistributions in binary form must reproduce the above copyright
16 1.1 ws * notice, this list of conditions and the following disclaimer in the
17 1.1 ws * documentation and/or other materials provided with the distribution.
18 1.1 ws * 3. All advertising materials mentioning features or use of this software
19 1.1 ws * must display the following acknowledgement:
20 1.1 ws * This product includes software developed by the University of
21 1.1 ws * California, Berkeley and its contributors.
22 1.1 ws * 4. Neither the name of the University nor the names of its contributors
23 1.1 ws * may be used to endorse or promote products derived from this software
24 1.1 ws * without specific prior written permission.
25 1.1 ws *
26 1.1 ws * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 1.1 ws * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 1.1 ws * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 1.1 ws * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 1.1 ws * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 1.1 ws * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 1.1 ws * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 1.1 ws * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 1.1 ws * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 1.1 ws * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 1.1 ws * SUCH DAMAGE.
37 1.1 ws *
38 1.12 fvdl * @(#)ufs_lockf.c 8.4 (Berkeley) 10/26/94
39 1.1 ws */
40 1.18 lukem
41 1.18 lukem #include <sys/cdefs.h>
42 1.21 thorpej __KERNEL_RCSID(0, "$NetBSD: vfs_lockf.c,v 1.21 2003/01/18 10:06:37 thorpej Exp $");
43 1.1 ws
44 1.1 ws #include <sys/param.h>
45 1.1 ws #include <sys/systm.h>
46 1.1 ws #include <sys/kernel.h>
47 1.1 ws #include <sys/file.h>
48 1.1 ws #include <sys/proc.h>
49 1.1 ws #include <sys/vnode.h>
50 1.1 ws #include <sys/malloc.h>
51 1.1 ws #include <sys/fcntl.h>
52 1.1 ws #include <sys/lockf.h>
53 1.1 ws
54 1.1 ws /*
55 1.6 mycroft * This variable controls the maximum number of processes that will
56 1.6 mycroft * be checked in doing deadlock detection.
57 1.6 mycroft */
58 1.6 mycroft int maxlockdepth = MAXDEPTH;
59 1.6 mycroft
60 1.6 mycroft #ifdef LOCKF_DEBUG
61 1.6 mycroft int lockf_debug = 0;
62 1.6 mycroft #endif
63 1.6 mycroft
64 1.6 mycroft #define NOLOCKF (struct lockf *)0
65 1.6 mycroft #define SELF 0x1
66 1.6 mycroft #define OTHERS 0x2
67 1.6 mycroft
68 1.6 mycroft /*
69 1.16 sommerfe * XXX TODO
70 1.16 sommerfe * Misc cleanups: "caddr_t id" should be visible in the API as a
71 1.16 sommerfe * "struct proc *".
72 1.16 sommerfe * (This requires rototilling all VFS's which support advisory locking).
73 1.16 sommerfe *
74 1.16 sommerfe * Use pools for lock allocation.
75 1.16 sommerfe */
76 1.16 sommerfe
77 1.16 sommerfe /*
78 1.16 sommerfe * XXXSMP TODO: Using either (a) a global lock, or (b) the vnode's
79 1.16 sommerfe * interlock should be sufficient; (b) requires a change to the API
80 1.16 sommerfe * because the vnode isn't visible here.
81 1.16 sommerfe *
82 1.16 sommerfe * If there's a lot of lock contention on a single vnode, locking
83 1.16 sommerfe * schemes which allow for more paralleism would be needed. Given how
84 1.16 sommerfe * infrequently byte-range locks are actually used in typical BSD
85 1.16 sommerfe * code, a more complex approach probably isn't worth it.
86 1.16 sommerfe */
87 1.16 sommerfe
88 1.16 sommerfe /*
89 1.4 mycroft * Do an advisory lock operation.
90 1.1 ws */
91 1.4 mycroft int
92 1.17 jdolecek lf_advlock(ap, head, size)
93 1.17 jdolecek struct vop_advlock_args *ap;
94 1.1 ws struct lockf **head;
95 1.3 cgd off_t size;
96 1.1 ws {
97 1.17 jdolecek struct flock *fl = ap->a_fl;
98 1.15 augustss struct lockf *lock;
99 1.1 ws off_t start, end;
100 1.1 ws int error;
101 1.1 ws
102 1.1 ws /*
103 1.1 ws * Convert the flock structure into a start and end.
104 1.1 ws */
105 1.1 ws switch (fl->l_whence) {
106 1.1 ws case SEEK_SET:
107 1.1 ws case SEEK_CUR:
108 1.1 ws /*
109 1.1 ws * Caller is responsible for adding any necessary offset
110 1.1 ws * when SEEK_CUR is used.
111 1.1 ws */
112 1.1 ws start = fl->l_start;
113 1.1 ws break;
114 1.1 ws
115 1.1 ws case SEEK_END:
116 1.1 ws start = size + fl->l_start;
117 1.1 ws break;
118 1.1 ws
119 1.1 ws default:
120 1.1 ws return (EINVAL);
121 1.1 ws }
122 1.1 ws if (start < 0)
123 1.1 ws return (EINVAL);
124 1.10 kleink
125 1.10 kleink /*
126 1.10 kleink * Avoid the common case of unlocking when inode has no locks.
127 1.10 kleink */
128 1.10 kleink if (*head == (struct lockf *)0) {
129 1.17 jdolecek if (ap->a_op != F_SETLK) {
130 1.10 kleink fl->l_type = F_UNLCK;
131 1.10 kleink return (0);
132 1.10 kleink }
133 1.10 kleink }
134 1.10 kleink
135 1.1 ws if (fl->l_len == 0)
136 1.1 ws end = -1;
137 1.1 ws else
138 1.1 ws end = start + fl->l_len - 1;
139 1.1 ws /*
140 1.4 mycroft * Create the lockf structure.
141 1.1 ws */
142 1.13 perry MALLOC(lock, struct lockf *, sizeof(*lock), M_LOCKF, M_WAITOK);
143 1.1 ws lock->lf_start = start;
144 1.1 ws lock->lf_end = end;
145 1.21 thorpej /* XXX NJWLWP
146 1.21 thorpej * I don't want to make the entire VFS universe use LWPs, because
147 1.21 thorpej * they don't need them, for the most part. This is an exception,
148 1.21 thorpej * and a kluge.
149 1.21 thorpej */
150 1.21 thorpej
151 1.1 ws lock->lf_head = head;
152 1.1 ws lock->lf_type = fl->l_type;
153 1.1 ws lock->lf_next = (struct lockf *)0;
154 1.12 fvdl TAILQ_INIT(&lock->lf_blkhd);
155 1.17 jdolecek lock->lf_flags = ap->a_flags;
156 1.21 thorpej if (lock->lf_flags & F_POSIX) {
157 1.21 thorpej KASSERT(curproc == (struct proc *)ap->a_id);
158 1.21 thorpej lock->lf_id = (caddr_t) curlwp;
159 1.21 thorpej } else {
160 1.21 thorpej lock->lf_id = ap->a_id; /* Not a proc at all, but a file * */
161 1.21 thorpej }
162 1.21 thorpej
163 1.21 thorpej
164 1.1 ws /*
165 1.1 ws * Do the requested operation.
166 1.1 ws */
167 1.17 jdolecek switch (ap->a_op) {
168 1.4 mycroft
169 1.1 ws case F_SETLK:
170 1.1 ws return (lf_setlock(lock));
171 1.1 ws
172 1.1 ws case F_UNLCK:
173 1.1 ws error = lf_clearlock(lock);
174 1.1 ws FREE(lock, M_LOCKF);
175 1.1 ws return (error);
176 1.1 ws
177 1.1 ws case F_GETLK:
178 1.1 ws error = lf_getlock(lock, fl);
179 1.1 ws FREE(lock, M_LOCKF);
180 1.1 ws return (error);
181 1.4 mycroft
182 1.1 ws default:
183 1.4 mycroft FREE(lock, M_LOCKF);
184 1.1 ws return (EINVAL);
185 1.1 ws }
186 1.1 ws /* NOTREACHED */
187 1.1 ws }
188 1.1 ws
189 1.1 ws /*
190 1.1 ws * Set a byte-range lock.
191 1.1 ws */
192 1.4 mycroft int
193 1.1 ws lf_setlock(lock)
194 1.15 augustss struct lockf *lock;
195 1.1 ws {
196 1.15 augustss struct lockf *block;
197 1.1 ws struct lockf **head = lock->lf_head;
198 1.1 ws struct lockf **prev, *overlap, *ltmp;
199 1.1 ws static char lockstr[] = "lockf";
200 1.1 ws int ovcase, priority, needtolink, error;
201 1.1 ws
202 1.1 ws #ifdef LOCKF_DEBUG
203 1.1 ws if (lockf_debug & 1)
204 1.1 ws lf_print("lf_setlock", lock);
205 1.1 ws #endif /* LOCKF_DEBUG */
206 1.1 ws
207 1.1 ws /*
208 1.1 ws * Set the priority
209 1.1 ws */
210 1.1 ws priority = PLOCK;
211 1.1 ws if (lock->lf_type == F_WRLCK)
212 1.1 ws priority += 4;
213 1.1 ws priority |= PCATCH;
214 1.1 ws /*
215 1.1 ws * Scan lock list for this file looking for locks that would block us.
216 1.1 ws */
217 1.7 christos while ((block = lf_getblock(lock)) != NULL) {
218 1.1 ws /*
219 1.1 ws * Free the structure and return if nonblocking.
220 1.1 ws */
221 1.1 ws if ((lock->lf_flags & F_WAIT) == 0) {
222 1.1 ws FREE(lock, M_LOCKF);
223 1.1 ws return (EAGAIN);
224 1.1 ws }
225 1.1 ws /*
226 1.1 ws * We are blocked. Since flock style locks cover
227 1.1 ws * the whole file, there is no chance for deadlock.
228 1.1 ws * For byte-range locks we must check for deadlock.
229 1.1 ws *
230 1.1 ws * Deadlock detection is done by looking through the
231 1.1 ws * wait channels to see if there are any cycles that
232 1.1 ws * involve us. MAXDEPTH is set just to make sure we
233 1.16 sommerfe * do not go off into neverneverland.
234 1.1 ws */
235 1.1 ws if ((lock->lf_flags & F_POSIX) &&
236 1.1 ws (block->lf_flags & F_POSIX)) {
237 1.21 thorpej struct lwp *wlwp;
238 1.15 augustss struct lockf *waitblock;
239 1.1 ws int i = 0;
240 1.1 ws
241 1.1 ws /* The block is waiting on something */
242 1.21 thorpej wlwp = (struct lwp *)block->lf_id;
243 1.21 thorpej while (wlwp->l_wchan &&
244 1.21 thorpej (wlwp->l_wmesg == lockstr) &&
245 1.1 ws (i++ < maxlockdepth)) {
246 1.21 thorpej waitblock = (struct lockf *)wlwp->l_wchan;
247 1.1 ws /* Get the owner of the blocking lock */
248 1.1 ws waitblock = waitblock->lf_next;
249 1.1 ws if ((waitblock->lf_flags & F_POSIX) == 0)
250 1.1 ws break;
251 1.21 thorpej wlwp = (struct lwp *)waitblock->lf_id;
252 1.21 thorpej if (wlwp == (struct lwp *)lock->lf_id) {
253 1.1 ws free(lock, M_LOCKF);
254 1.1 ws return (EDEADLK);
255 1.1 ws }
256 1.1 ws }
257 1.16 sommerfe /*
258 1.16 sommerfe * If we're still following a dependancy chain
259 1.16 sommerfe * after maxlockdepth iterations, assume we're in
260 1.16 sommerfe * a cycle to be safe.
261 1.16 sommerfe */
262 1.16 sommerfe if (i >= maxlockdepth) {
263 1.16 sommerfe free(lock, M_LOCKF);
264 1.16 sommerfe return (EDEADLK);
265 1.16 sommerfe }
266 1.1 ws }
267 1.1 ws /*
268 1.1 ws * For flock type locks, we must first remove
269 1.1 ws * any shared locks that we hold before we sleep
270 1.1 ws * waiting for an exclusive lock.
271 1.1 ws */
272 1.1 ws if ((lock->lf_flags & F_FLOCK) &&
273 1.1 ws lock->lf_type == F_WRLCK) {
274 1.1 ws lock->lf_type = F_UNLCK;
275 1.1 ws (void) lf_clearlock(lock);
276 1.1 ws lock->lf_type = F_WRLCK;
277 1.1 ws }
278 1.1 ws /*
279 1.1 ws * Add our lock to the blocked list and sleep until we're free.
280 1.1 ws * Remember who blocked us (for deadlock detection).
281 1.1 ws */
282 1.1 ws lock->lf_next = block;
283 1.12 fvdl TAILQ_INSERT_TAIL(&block->lf_blkhd, lock, lf_block);
284 1.1 ws #ifdef LOCKF_DEBUG
285 1.1 ws if (lockf_debug & 1) {
286 1.1 ws lf_print("lf_setlock: blocking on", block);
287 1.1 ws lf_printlist("lf_setlock", block);
288 1.1 ws }
289 1.1 ws #endif /* LOCKF_DEBUG */
290 1.7 christos error = tsleep((caddr_t)lock, priority, lockstr, 0);
291 1.16 sommerfe
292 1.16 sommerfe /*
293 1.16 sommerfe * We may have been awakened by a signal (in
294 1.16 sommerfe * which case we must remove ourselves from the
295 1.16 sommerfe * blocked list) and/or by another process
296 1.16 sommerfe * releasing a lock (in which case we have already
297 1.16 sommerfe * been removed from the blocked list and our
298 1.16 sommerfe * lf_next field set to NOLOCKF).
299 1.16 sommerfe */
300 1.16 sommerfe if (lock->lf_next != NOLOCKF) {
301 1.16 sommerfe TAILQ_REMOVE(&lock->lf_next->lf_blkhd, lock, lf_block);
302 1.16 sommerfe lock->lf_next = NOLOCKF;
303 1.16 sommerfe }
304 1.7 christos if (error) {
305 1.4 mycroft free(lock, M_LOCKF);
306 1.4 mycroft return (error);
307 1.1 ws }
308 1.1 ws }
309 1.1 ws /*
310 1.1 ws * No blocks!! Add the lock. Note that we will
311 1.1 ws * downgrade or upgrade any overlapping locks this
312 1.1 ws * process already owns.
313 1.1 ws *
314 1.1 ws * Skip over locks owned by other processes.
315 1.1 ws * Handle any locks that overlap and are owned by ourselves.
316 1.1 ws */
317 1.1 ws prev = head;
318 1.1 ws block = *head;
319 1.1 ws needtolink = 1;
320 1.1 ws for (;;) {
321 1.7 christos ovcase = lf_findoverlap(block, lock, SELF, &prev, &overlap);
322 1.7 christos if (ovcase)
323 1.1 ws block = overlap->lf_next;
324 1.1 ws /*
325 1.1 ws * Six cases:
326 1.1 ws * 0) no overlap
327 1.1 ws * 1) overlap == lock
328 1.1 ws * 2) overlap contains lock
329 1.1 ws * 3) lock contains overlap
330 1.1 ws * 4) overlap starts before lock
331 1.1 ws * 5) overlap ends after lock
332 1.1 ws */
333 1.1 ws switch (ovcase) {
334 1.1 ws case 0: /* no overlap */
335 1.1 ws if (needtolink) {
336 1.1 ws *prev = lock;
337 1.1 ws lock->lf_next = overlap;
338 1.1 ws }
339 1.1 ws break;
340 1.1 ws
341 1.1 ws case 1: /* overlap == lock */
342 1.1 ws /*
343 1.1 ws * If downgrading lock, others may be
344 1.1 ws * able to acquire it.
345 1.1 ws */
346 1.1 ws if (lock->lf_type == F_RDLCK &&
347 1.1 ws overlap->lf_type == F_WRLCK)
348 1.1 ws lf_wakelock(overlap);
349 1.1 ws overlap->lf_type = lock->lf_type;
350 1.1 ws FREE(lock, M_LOCKF);
351 1.1 ws lock = overlap; /* for debug output below */
352 1.1 ws break;
353 1.1 ws
354 1.1 ws case 2: /* overlap contains lock */
355 1.1 ws /*
356 1.1 ws * Check for common starting point and different types.
357 1.1 ws */
358 1.1 ws if (overlap->lf_type == lock->lf_type) {
359 1.1 ws free(lock, M_LOCKF);
360 1.1 ws lock = overlap; /* for debug output below */
361 1.1 ws break;
362 1.1 ws }
363 1.1 ws if (overlap->lf_start == lock->lf_start) {
364 1.1 ws *prev = lock;
365 1.1 ws lock->lf_next = overlap;
366 1.1 ws overlap->lf_start = lock->lf_end + 1;
367 1.1 ws } else
368 1.1 ws lf_split(overlap, lock);
369 1.1 ws lf_wakelock(overlap);
370 1.1 ws break;
371 1.1 ws
372 1.1 ws case 3: /* lock contains overlap */
373 1.1 ws /*
374 1.1 ws * If downgrading lock, others may be able to
375 1.1 ws * acquire it, otherwise take the list.
376 1.1 ws */
377 1.1 ws if (lock->lf_type == F_RDLCK &&
378 1.1 ws overlap->lf_type == F_WRLCK) {
379 1.1 ws lf_wakelock(overlap);
380 1.1 ws } else {
381 1.19 matt while ((ltmp = TAILQ_FIRST(&overlap->lf_blkhd))) {
382 1.16 sommerfe KASSERT(ltmp->lf_next == overlap);
383 1.12 fvdl TAILQ_REMOVE(&overlap->lf_blkhd, ltmp,
384 1.12 fvdl lf_block);
385 1.16 sommerfe ltmp->lf_next = lock;
386 1.12 fvdl TAILQ_INSERT_TAIL(&lock->lf_blkhd,
387 1.12 fvdl ltmp, lf_block);
388 1.12 fvdl }
389 1.1 ws }
390 1.1 ws /*
391 1.1 ws * Add the new lock if necessary and delete the overlap.
392 1.1 ws */
393 1.1 ws if (needtolink) {
394 1.1 ws *prev = lock;
395 1.1 ws lock->lf_next = overlap->lf_next;
396 1.1 ws prev = &lock->lf_next;
397 1.1 ws needtolink = 0;
398 1.1 ws } else
399 1.1 ws *prev = overlap->lf_next;
400 1.1 ws free(overlap, M_LOCKF);
401 1.1 ws continue;
402 1.1 ws
403 1.1 ws case 4: /* overlap starts before lock */
404 1.1 ws /*
405 1.1 ws * Add lock after overlap on the list.
406 1.1 ws */
407 1.1 ws lock->lf_next = overlap->lf_next;
408 1.1 ws overlap->lf_next = lock;
409 1.1 ws overlap->lf_end = lock->lf_start - 1;
410 1.1 ws prev = &lock->lf_next;
411 1.1 ws lf_wakelock(overlap);
412 1.1 ws needtolink = 0;
413 1.1 ws continue;
414 1.1 ws
415 1.1 ws case 5: /* overlap ends after lock */
416 1.1 ws /*
417 1.1 ws * Add the new lock before overlap.
418 1.1 ws */
419 1.1 ws if (needtolink) {
420 1.1 ws *prev = lock;
421 1.1 ws lock->lf_next = overlap;
422 1.1 ws }
423 1.1 ws overlap->lf_start = lock->lf_end + 1;
424 1.1 ws lf_wakelock(overlap);
425 1.1 ws break;
426 1.1 ws }
427 1.1 ws break;
428 1.1 ws }
429 1.1 ws #ifdef LOCKF_DEBUG
430 1.1 ws if (lockf_debug & 1) {
431 1.1 ws lf_print("lf_setlock: got the lock", lock);
432 1.1 ws lf_printlist("lf_setlock", lock);
433 1.1 ws }
434 1.1 ws #endif /* LOCKF_DEBUG */
435 1.1 ws return (0);
436 1.1 ws }
437 1.1 ws
438 1.1 ws /*
439 1.1 ws * Remove a byte-range lock on an inode.
440 1.1 ws *
441 1.1 ws * Generally, find the lock (or an overlap to that lock)
442 1.1 ws * and remove it (or shrink it), then wakeup anyone we can.
443 1.1 ws */
444 1.4 mycroft int
445 1.1 ws lf_clearlock(unlock)
446 1.15 augustss struct lockf *unlock;
447 1.1 ws {
448 1.1 ws struct lockf **head = unlock->lf_head;
449 1.15 augustss struct lockf *lf = *head;
450 1.1 ws struct lockf *overlap, **prev;
451 1.1 ws int ovcase;
452 1.1 ws
453 1.1 ws if (lf == NOLOCKF)
454 1.1 ws return (0);
455 1.1 ws #ifdef LOCKF_DEBUG
456 1.1 ws if (unlock->lf_type != F_UNLCK)
457 1.1 ws panic("lf_clearlock: bad type");
458 1.1 ws if (lockf_debug & 1)
459 1.1 ws lf_print("lf_clearlock", unlock);
460 1.1 ws #endif /* LOCKF_DEBUG */
461 1.1 ws prev = head;
462 1.7 christos while ((ovcase = lf_findoverlap(lf, unlock, SELF,
463 1.7 christos &prev, &overlap)) != 0) {
464 1.1 ws /*
465 1.1 ws * Wakeup the list of locks to be retried.
466 1.1 ws */
467 1.1 ws lf_wakelock(overlap);
468 1.1 ws
469 1.1 ws switch (ovcase) {
470 1.1 ws
471 1.1 ws case 1: /* overlap == lock */
472 1.1 ws *prev = overlap->lf_next;
473 1.1 ws FREE(overlap, M_LOCKF);
474 1.1 ws break;
475 1.1 ws
476 1.1 ws case 2: /* overlap contains lock: split it */
477 1.1 ws if (overlap->lf_start == unlock->lf_start) {
478 1.1 ws overlap->lf_start = unlock->lf_end + 1;
479 1.1 ws break;
480 1.1 ws }
481 1.1 ws lf_split(overlap, unlock);
482 1.1 ws overlap->lf_next = unlock->lf_next;
483 1.1 ws break;
484 1.1 ws
485 1.1 ws case 3: /* lock contains overlap */
486 1.1 ws *prev = overlap->lf_next;
487 1.1 ws lf = overlap->lf_next;
488 1.1 ws free(overlap, M_LOCKF);
489 1.1 ws continue;
490 1.1 ws
491 1.1 ws case 4: /* overlap starts before lock */
492 1.1 ws overlap->lf_end = unlock->lf_start - 1;
493 1.1 ws prev = &overlap->lf_next;
494 1.1 ws lf = overlap->lf_next;
495 1.1 ws continue;
496 1.1 ws
497 1.1 ws case 5: /* overlap ends after lock */
498 1.1 ws overlap->lf_start = unlock->lf_end + 1;
499 1.1 ws break;
500 1.1 ws }
501 1.1 ws break;
502 1.1 ws }
503 1.1 ws #ifdef LOCKF_DEBUG
504 1.1 ws if (lockf_debug & 1)
505 1.1 ws lf_printlist("lf_clearlock", unlock);
506 1.1 ws #endif /* LOCKF_DEBUG */
507 1.1 ws return (0);
508 1.1 ws }
509 1.1 ws
510 1.1 ws /*
511 1.1 ws * Check whether there is a blocking lock,
512 1.1 ws * and if so return its process identifier.
513 1.1 ws */
514 1.4 mycroft int
515 1.1 ws lf_getlock(lock, fl)
516 1.15 augustss struct lockf *lock;
517 1.15 augustss struct flock *fl;
518 1.1 ws {
519 1.15 augustss struct lockf *block;
520 1.1 ws
521 1.1 ws #ifdef LOCKF_DEBUG
522 1.1 ws if (lockf_debug & 1)
523 1.1 ws lf_print("lf_getlock", lock);
524 1.1 ws #endif /* LOCKF_DEBUG */
525 1.1 ws
526 1.7 christos if ((block = lf_getblock(lock)) != NULL) {
527 1.1 ws fl->l_type = block->lf_type;
528 1.1 ws fl->l_whence = SEEK_SET;
529 1.1 ws fl->l_start = block->lf_start;
530 1.1 ws if (block->lf_end == -1)
531 1.1 ws fl->l_len = 0;
532 1.1 ws else
533 1.1 ws fl->l_len = block->lf_end - block->lf_start + 1;
534 1.1 ws if (block->lf_flags & F_POSIX)
535 1.21 thorpej fl->l_pid = ((struct lwp *)(block->lf_id))->l_proc->p_pid;
536 1.1 ws else
537 1.1 ws fl->l_pid = -1;
538 1.1 ws } else {
539 1.1 ws fl->l_type = F_UNLCK;
540 1.1 ws }
541 1.1 ws return (0);
542 1.1 ws }
543 1.1 ws
544 1.1 ws /*
545 1.1 ws * Walk the list of locks for an inode and
546 1.1 ws * return the first blocking lock.
547 1.1 ws */
548 1.1 ws struct lockf *
549 1.1 ws lf_getblock(lock)
550 1.15 augustss struct lockf *lock;
551 1.1 ws {
552 1.1 ws struct lockf **prev, *overlap, *lf = *(lock->lf_head);
553 1.1 ws
554 1.1 ws prev = lock->lf_head;
555 1.20 simonb while (lf_findoverlap(lf, lock, OTHERS, &prev, &overlap) != 0) {
556 1.1 ws /*
557 1.1 ws * We've found an overlap, see if it blocks us
558 1.1 ws */
559 1.1 ws if ((lock->lf_type == F_WRLCK || overlap->lf_type == F_WRLCK))
560 1.1 ws return (overlap);
561 1.1 ws /*
562 1.1 ws * Nope, point to the next one on the list and
563 1.1 ws * see if it blocks us
564 1.1 ws */
565 1.1 ws lf = overlap->lf_next;
566 1.1 ws }
567 1.1 ws return (NOLOCKF);
568 1.1 ws }
569 1.1 ws
570 1.1 ws /*
571 1.1 ws * Walk the list of locks for an inode to
572 1.1 ws * find an overlapping lock (if any).
573 1.1 ws *
574 1.1 ws * NOTE: this returns only the FIRST overlapping lock. There
575 1.1 ws * may be more than one.
576 1.1 ws */
577 1.4 mycroft int
578 1.1 ws lf_findoverlap(lf, lock, type, prev, overlap)
579 1.15 augustss struct lockf *lf;
580 1.1 ws struct lockf *lock;
581 1.1 ws int type;
582 1.1 ws struct lockf ***prev;
583 1.1 ws struct lockf **overlap;
584 1.1 ws {
585 1.1 ws off_t start, end;
586 1.1 ws
587 1.1 ws *overlap = lf;
588 1.1 ws if (lf == NOLOCKF)
589 1.1 ws return (0);
590 1.1 ws #ifdef LOCKF_DEBUG
591 1.1 ws if (lockf_debug & 2)
592 1.1 ws lf_print("lf_findoverlap: looking for overlap in", lock);
593 1.1 ws #endif /* LOCKF_DEBUG */
594 1.1 ws start = lock->lf_start;
595 1.1 ws end = lock->lf_end;
596 1.1 ws while (lf != NOLOCKF) {
597 1.1 ws if (((type & SELF) && lf->lf_id != lock->lf_id) ||
598 1.1 ws ((type & OTHERS) && lf->lf_id == lock->lf_id)) {
599 1.1 ws *prev = &lf->lf_next;
600 1.1 ws *overlap = lf = lf->lf_next;
601 1.1 ws continue;
602 1.1 ws }
603 1.1 ws #ifdef LOCKF_DEBUG
604 1.1 ws if (lockf_debug & 2)
605 1.1 ws lf_print("\tchecking", lf);
606 1.1 ws #endif /* LOCKF_DEBUG */
607 1.1 ws /*
608 1.1 ws * OK, check for overlap
609 1.1 ws *
610 1.1 ws * Six cases:
611 1.1 ws * 0) no overlap
612 1.1 ws * 1) overlap == lock
613 1.1 ws * 2) overlap contains lock
614 1.1 ws * 3) lock contains overlap
615 1.1 ws * 4) overlap starts before lock
616 1.1 ws * 5) overlap ends after lock
617 1.1 ws */
618 1.1 ws if ((lf->lf_end != -1 && start > lf->lf_end) ||
619 1.1 ws (end != -1 && lf->lf_start > end)) {
620 1.1 ws /* Case 0 */
621 1.1 ws #ifdef LOCKF_DEBUG
622 1.1 ws if (lockf_debug & 2)
623 1.9 christos printf("no overlap\n");
624 1.1 ws #endif /* LOCKF_DEBUG */
625 1.1 ws if ((type & SELF) && end != -1 && lf->lf_start > end)
626 1.1 ws return (0);
627 1.1 ws *prev = &lf->lf_next;
628 1.1 ws *overlap = lf = lf->lf_next;
629 1.1 ws continue;
630 1.1 ws }
631 1.1 ws if ((lf->lf_start == start) && (lf->lf_end == end)) {
632 1.1 ws /* Case 1 */
633 1.1 ws #ifdef LOCKF_DEBUG
634 1.1 ws if (lockf_debug & 2)
635 1.9 christos printf("overlap == lock\n");
636 1.1 ws #endif /* LOCKF_DEBUG */
637 1.1 ws return (1);
638 1.1 ws }
639 1.1 ws if ((lf->lf_start <= start) &&
640 1.1 ws (end != -1) &&
641 1.1 ws ((lf->lf_end >= end) || (lf->lf_end == -1))) {
642 1.1 ws /* Case 2 */
643 1.1 ws #ifdef LOCKF_DEBUG
644 1.1 ws if (lockf_debug & 2)
645 1.9 christos printf("overlap contains lock\n");
646 1.1 ws #endif /* LOCKF_DEBUG */
647 1.1 ws return (2);
648 1.1 ws }
649 1.1 ws if (start <= lf->lf_start &&
650 1.4 mycroft (end == -1 ||
651 1.1 ws (lf->lf_end != -1 && end >= lf->lf_end))) {
652 1.1 ws /* Case 3 */
653 1.1 ws #ifdef LOCKF_DEBUG
654 1.1 ws if (lockf_debug & 2)
655 1.9 christos printf("lock contains overlap\n");
656 1.1 ws #endif /* LOCKF_DEBUG */
657 1.1 ws return (3);
658 1.1 ws }
659 1.1 ws if ((lf->lf_start < start) &&
660 1.1 ws ((lf->lf_end >= start) || (lf->lf_end == -1))) {
661 1.1 ws /* Case 4 */
662 1.1 ws #ifdef LOCKF_DEBUG
663 1.1 ws if (lockf_debug & 2)
664 1.9 christos printf("overlap starts before lock\n");
665 1.1 ws #endif /* LOCKF_DEBUG */
666 1.1 ws return (4);
667 1.1 ws }
668 1.1 ws if ((lf->lf_start > start) &&
669 1.1 ws (end != -1) &&
670 1.1 ws ((lf->lf_end > end) || (lf->lf_end == -1))) {
671 1.1 ws /* Case 5 */
672 1.1 ws #ifdef LOCKF_DEBUG
673 1.1 ws if (lockf_debug & 2)
674 1.9 christos printf("overlap ends after lock\n");
675 1.1 ws #endif /* LOCKF_DEBUG */
676 1.1 ws return (5);
677 1.1 ws }
678 1.1 ws panic("lf_findoverlap: default");
679 1.1 ws }
680 1.1 ws return (0);
681 1.1 ws }
682 1.1 ws
683 1.1 ws /*
684 1.1 ws * Split a lock and a contained region into
685 1.1 ws * two or three locks as necessary.
686 1.1 ws */
687 1.2 cgd void
688 1.1 ws lf_split(lock1, lock2)
689 1.15 augustss struct lockf *lock1;
690 1.15 augustss struct lockf *lock2;
691 1.1 ws {
692 1.15 augustss struct lockf *splitlock;
693 1.1 ws
694 1.1 ws #ifdef LOCKF_DEBUG
695 1.1 ws if (lockf_debug & 2) {
696 1.1 ws lf_print("lf_split", lock1);
697 1.1 ws lf_print("splitting from", lock2);
698 1.1 ws }
699 1.1 ws #endif /* LOCKF_DEBUG */
700 1.1 ws /*
701 1.1 ws * Check to see if spliting into only two pieces.
702 1.1 ws */
703 1.1 ws if (lock1->lf_start == lock2->lf_start) {
704 1.1 ws lock1->lf_start = lock2->lf_end + 1;
705 1.1 ws lock2->lf_next = lock1;
706 1.1 ws return;
707 1.1 ws }
708 1.1 ws if (lock1->lf_end == lock2->lf_end) {
709 1.1 ws lock1->lf_end = lock2->lf_start - 1;
710 1.1 ws lock2->lf_next = lock1->lf_next;
711 1.1 ws lock1->lf_next = lock2;
712 1.1 ws return;
713 1.1 ws }
714 1.1 ws /*
715 1.1 ws * Make a new lock consisting of the last part of
716 1.1 ws * the encompassing lock
717 1.1 ws */
718 1.13 perry MALLOC(splitlock, struct lockf *, sizeof(*splitlock), M_LOCKF, M_WAITOK);
719 1.14 perry memcpy((caddr_t)splitlock, (caddr_t)lock1, sizeof(*splitlock));
720 1.1 ws splitlock->lf_start = lock2->lf_end + 1;
721 1.12 fvdl TAILQ_INIT(&splitlock->lf_blkhd);
722 1.1 ws lock1->lf_end = lock2->lf_start - 1;
723 1.1 ws /*
724 1.1 ws * OK, now link it in
725 1.1 ws */
726 1.1 ws splitlock->lf_next = lock1->lf_next;
727 1.1 ws lock2->lf_next = splitlock;
728 1.1 ws lock1->lf_next = lock2;
729 1.1 ws }
730 1.1 ws
731 1.1 ws /*
732 1.1 ws * Wakeup a blocklist
733 1.1 ws */
734 1.2 cgd void
735 1.1 ws lf_wakelock(listhead)
736 1.1 ws struct lockf *listhead;
737 1.1 ws {
738 1.15 augustss struct lockf *wakelock;
739 1.1 ws
740 1.19 matt while ((wakelock = TAILQ_FIRST(&listhead->lf_blkhd))) {
741 1.16 sommerfe KASSERT(wakelock->lf_next == listhead);
742 1.12 fvdl TAILQ_REMOVE(&listhead->lf_blkhd, wakelock, lf_block);
743 1.1 ws wakelock->lf_next = NOLOCKF;
744 1.1 ws #ifdef LOCKF_DEBUG
745 1.1 ws if (lockf_debug & 2)
746 1.1 ws lf_print("lf_wakelock: awakening", wakelock);
747 1.12 fvdl #endif
748 1.1 ws wakeup((caddr_t)wakelock);
749 1.1 ws }
750 1.1 ws }
751 1.1 ws
752 1.1 ws #ifdef LOCKF_DEBUG
753 1.1 ws /*
754 1.1 ws * Print out a lock.
755 1.1 ws */
756 1.4 mycroft void
757 1.1 ws lf_print(tag, lock)
758 1.1 ws char *tag;
759 1.15 augustss struct lockf *lock;
760 1.1 ws {
761 1.1 ws
762 1.9 christos printf("%s: lock %p for ", tag, lock);
763 1.1 ws if (lock->lf_flags & F_POSIX)
764 1.9 christos printf("proc %d", ((struct proc *)(lock->lf_id))->p_pid);
765 1.1 ws else
766 1.11 jtk printf("id 0x%p", lock->lf_id);
767 1.11 jtk printf(" %s, start %qx, end %qx",
768 1.1 ws lock->lf_type == F_RDLCK ? "shared" :
769 1.1 ws lock->lf_type == F_WRLCK ? "exclusive" :
770 1.1 ws lock->lf_type == F_UNLCK ? "unlock" :
771 1.1 ws "unknown", lock->lf_start, lock->lf_end);
772 1.19 matt if (TAILQ_FIRST(&lock->lf_blkhd))
773 1.19 matt printf(" block %p\n", TAILQ_FIRST(&lock->lf_blkhd));
774 1.1 ws else
775 1.9 christos printf("\n");
776 1.1 ws }
777 1.1 ws
778 1.4 mycroft void
779 1.1 ws lf_printlist(tag, lock)
780 1.1 ws char *tag;
781 1.1 ws struct lockf *lock;
782 1.1 ws {
783 1.15 augustss struct lockf *lf, *blk;
784 1.1 ws
785 1.11 jtk printf("%s: Lock list:\n", tag);
786 1.12 fvdl for (lf = *lock->lf_head; lf; lf = lf->lf_next) {
787 1.9 christos printf("\tlock %p for ", lf);
788 1.1 ws if (lf->lf_flags & F_POSIX)
789 1.9 christos printf("proc %d", ((struct proc *)(lf->lf_id))->p_pid);
790 1.1 ws else
791 1.11 jtk printf("id 0x%p", lf->lf_id);
792 1.11 jtk printf(", %s, start %qx, end %qx",
793 1.1 ws lf->lf_type == F_RDLCK ? "shared" :
794 1.1 ws lf->lf_type == F_WRLCK ? "exclusive" :
795 1.1 ws lf->lf_type == F_UNLCK ? "unlock" :
796 1.1 ws "unknown", lf->lf_start, lf->lf_end);
797 1.19 matt TAILQ_FOREACH(blk, &lf->lf_blkhd, lf_block) {
798 1.12 fvdl if (blk->lf_flags & F_POSIX)
799 1.12 fvdl printf("proc %d",
800 1.12 fvdl ((struct proc *)(blk->lf_id))->p_pid);
801 1.12 fvdl else
802 1.12 fvdl printf("id 0x%p", blk->lf_id);
803 1.12 fvdl printf(", %s, start %qx, end %qx",
804 1.12 fvdl blk->lf_type == F_RDLCK ? "shared" :
805 1.12 fvdl blk->lf_type == F_WRLCK ? "exclusive" :
806 1.12 fvdl blk->lf_type == F_UNLCK ? "unlock" :
807 1.12 fvdl "unknown", blk->lf_start, blk->lf_end);
808 1.19 matt if (TAILQ_FIRST(&blk->lf_blkhd))
809 1.12 fvdl panic("lf_printlist: bad list");
810 1.12 fvdl }
811 1.12 fvdl printf("\n");
812 1.1 ws }
813 1.1 ws }
814 1.1 ws #endif /* LOCKF_DEBUG */
815