vfs_lockf.c revision 1.33 1 1.33 agc /* $NetBSD: vfs_lockf.c,v 1.33 2003/08/07 16:32:02 agc 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 1.33 agc * 3. Neither the name of the University nor the names of its contributors
19 1.1 ws * may be used to endorse or promote products derived from this software
20 1.1 ws * without specific prior written permission.
21 1.1 ws *
22 1.1 ws * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 1.1 ws * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 1.1 ws * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 1.1 ws * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 1.1 ws * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 1.1 ws * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 1.1 ws * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 1.1 ws * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 1.1 ws * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 1.1 ws * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 1.1 ws * SUCH DAMAGE.
33 1.1 ws *
34 1.12 fvdl * @(#)ufs_lockf.c 8.4 (Berkeley) 10/26/94
35 1.1 ws */
36 1.18 lukem
37 1.18 lukem #include <sys/cdefs.h>
38 1.33 agc __KERNEL_RCSID(0, "$NetBSD: vfs_lockf.c,v 1.33 2003/08/07 16:32:02 agc Exp $");
39 1.1 ws
40 1.1 ws #include <sys/param.h>
41 1.1 ws #include <sys/systm.h>
42 1.1 ws #include <sys/kernel.h>
43 1.1 ws #include <sys/file.h>
44 1.1 ws #include <sys/proc.h>
45 1.1 ws #include <sys/vnode.h>
46 1.1 ws #include <sys/malloc.h>
47 1.1 ws #include <sys/fcntl.h>
48 1.1 ws #include <sys/lockf.h>
49 1.22 thorpej
50 1.22 thorpej MALLOC_DEFINE(M_LOCKF, "lockf", "Byte-range locking structures");
51 1.1 ws
52 1.1 ws /*
53 1.6 mycroft * This variable controls the maximum number of processes that will
54 1.6 mycroft * be checked in doing deadlock detection.
55 1.6 mycroft */
56 1.6 mycroft int maxlockdepth = MAXDEPTH;
57 1.6 mycroft
58 1.6 mycroft #ifdef LOCKF_DEBUG
59 1.6 mycroft int lockf_debug = 0;
60 1.6 mycroft #endif
61 1.6 mycroft
62 1.6 mycroft #define NOLOCKF (struct lockf *)0
63 1.6 mycroft #define SELF 0x1
64 1.6 mycroft #define OTHERS 0x2
65 1.6 mycroft
66 1.27 yamt static int lf_clearlock(struct lockf *, struct lockf **);
67 1.25 yamt static int lf_findoverlap(struct lockf *,
68 1.25 yamt struct lockf *, int, struct lockf ***, struct lockf **);
69 1.25 yamt static struct lockf *lf_getblock(struct lockf *);
70 1.25 yamt static int lf_getlock(struct lockf *, struct flock *);
71 1.27 yamt static int lf_setlock(struct lockf *, struct lockf **, struct simplelock *);
72 1.27 yamt static void lf_split(struct lockf *, struct lockf *, struct lockf **);
73 1.25 yamt static void lf_wakelock(struct lockf *);
74 1.24 yamt
75 1.24 yamt #ifdef LOCKF_DEBUG
76 1.25 yamt static void lf_print(char *, struct lockf *);
77 1.25 yamt static void lf_printlist(char *, struct lockf *);
78 1.24 yamt #endif
79 1.24 yamt
80 1.6 mycroft /*
81 1.16 sommerfe * XXX TODO
82 1.16 sommerfe * Misc cleanups: "caddr_t id" should be visible in the API as a
83 1.16 sommerfe * "struct proc *".
84 1.16 sommerfe * (This requires rototilling all VFS's which support advisory locking).
85 1.16 sommerfe *
86 1.16 sommerfe * Use pools for lock allocation.
87 1.16 sommerfe */
88 1.16 sommerfe
89 1.16 sommerfe /*
90 1.16 sommerfe * If there's a lot of lock contention on a single vnode, locking
91 1.16 sommerfe * schemes which allow for more paralleism would be needed. Given how
92 1.16 sommerfe * infrequently byte-range locks are actually used in typical BSD
93 1.16 sommerfe * code, a more complex approach probably isn't worth it.
94 1.16 sommerfe */
95 1.16 sommerfe
96 1.16 sommerfe /*
97 1.4 mycroft * Do an advisory lock operation.
98 1.1 ws */
99 1.4 mycroft int
100 1.25 yamt lf_advlock(struct vop_advlock_args *ap, struct lockf **head, off_t size)
101 1.1 ws {
102 1.17 jdolecek struct flock *fl = ap->a_fl;
103 1.27 yamt struct lockf *lock = NULL;
104 1.27 yamt struct lockf *sparelock;
105 1.30 yamt struct simplelock *interlock = &ap->a_vp->v_interlock;
106 1.1 ws off_t start, end;
107 1.1 ws int error;
108 1.1 ws
109 1.1 ws /*
110 1.1 ws * Convert the flock structure into a start and end.
111 1.1 ws */
112 1.1 ws switch (fl->l_whence) {
113 1.1 ws case SEEK_SET:
114 1.1 ws case SEEK_CUR:
115 1.1 ws /*
116 1.1 ws * Caller is responsible for adding any necessary offset
117 1.1 ws * when SEEK_CUR is used.
118 1.1 ws */
119 1.1 ws start = fl->l_start;
120 1.1 ws break;
121 1.1 ws
122 1.1 ws case SEEK_END:
123 1.1 ws start = size + fl->l_start;
124 1.1 ws break;
125 1.1 ws
126 1.1 ws default:
127 1.29 yamt return EINVAL;
128 1.1 ws }
129 1.1 ws if (start < 0)
130 1.29 yamt return EINVAL;
131 1.10 kleink
132 1.10 kleink /*
133 1.27 yamt * allocate locks before acquire simple lock.
134 1.27 yamt * we need two locks in the worst case.
135 1.27 yamt */
136 1.27 yamt switch (ap->a_op) {
137 1.27 yamt case F_SETLK:
138 1.27 yamt case F_UNLCK:
139 1.27 yamt /*
140 1.27 yamt * XXX for F_UNLCK case, we can re-use lock.
141 1.27 yamt */
142 1.27 yamt if ((fl->l_type & F_FLOCK) == 0) {
143 1.27 yamt /*
144 1.27 yamt * byte-range lock might need one more lock.
145 1.27 yamt */
146 1.27 yamt MALLOC(sparelock, struct lockf *, sizeof(*lock),
147 1.27 yamt M_LOCKF, M_WAITOK);
148 1.27 yamt if (sparelock == NULL) {
149 1.27 yamt error = ENOMEM;
150 1.27 yamt goto quit;
151 1.27 yamt }
152 1.27 yamt break;
153 1.27 yamt }
154 1.27 yamt /* FALLTHROUGH */
155 1.27 yamt
156 1.27 yamt case F_GETLK:
157 1.27 yamt sparelock = NULL;
158 1.27 yamt break;
159 1.27 yamt
160 1.27 yamt default:
161 1.29 yamt return EINVAL;
162 1.27 yamt }
163 1.27 yamt
164 1.27 yamt MALLOC(lock, struct lockf *, sizeof(*lock), M_LOCKF, M_WAITOK);
165 1.27 yamt if (lock == NULL) {
166 1.27 yamt error = ENOMEM;
167 1.27 yamt goto quit;
168 1.27 yamt }
169 1.27 yamt
170 1.30 yamt simple_lock(interlock);
171 1.27 yamt
172 1.27 yamt /*
173 1.10 kleink * Avoid the common case of unlocking when inode has no locks.
174 1.10 kleink */
175 1.10 kleink if (*head == (struct lockf *)0) {
176 1.17 jdolecek if (ap->a_op != F_SETLK) {
177 1.10 kleink fl->l_type = F_UNLCK;
178 1.27 yamt error = 0;
179 1.27 yamt goto quit_unlock;
180 1.10 kleink }
181 1.10 kleink }
182 1.10 kleink
183 1.1 ws if (fl->l_len == 0)
184 1.1 ws end = -1;
185 1.1 ws else
186 1.1 ws end = start + fl->l_len - 1;
187 1.1 ws /*
188 1.4 mycroft * Create the lockf structure.
189 1.1 ws */
190 1.1 ws lock->lf_start = start;
191 1.1 ws lock->lf_end = end;
192 1.21 thorpej /* XXX NJWLWP
193 1.21 thorpej * I don't want to make the entire VFS universe use LWPs, because
194 1.21 thorpej * they don't need them, for the most part. This is an exception,
195 1.21 thorpej * and a kluge.
196 1.21 thorpej */
197 1.21 thorpej
198 1.1 ws lock->lf_head = head;
199 1.1 ws lock->lf_type = fl->l_type;
200 1.1 ws lock->lf_next = (struct lockf *)0;
201 1.12 fvdl TAILQ_INIT(&lock->lf_blkhd);
202 1.17 jdolecek lock->lf_flags = ap->a_flags;
203 1.21 thorpej if (lock->lf_flags & F_POSIX) {
204 1.21 thorpej KASSERT(curproc == (struct proc *)ap->a_id);
205 1.21 thorpej }
206 1.23 mycroft lock->lf_id = (struct proc *)ap->a_id;
207 1.23 mycroft lock->lf_lwp = curlwp;
208 1.21 thorpej
209 1.1 ws /*
210 1.1 ws * Do the requested operation.
211 1.1 ws */
212 1.17 jdolecek switch (ap->a_op) {
213 1.4 mycroft
214 1.1 ws case F_SETLK:
215 1.30 yamt error = lf_setlock(lock, &sparelock, interlock);
216 1.27 yamt lock = NULL; /* lf_setlock freed it */
217 1.27 yamt break;
218 1.1 ws
219 1.1 ws case F_UNLCK:
220 1.27 yamt error = lf_clearlock(lock, &sparelock);
221 1.27 yamt break;
222 1.1 ws
223 1.1 ws case F_GETLK:
224 1.1 ws error = lf_getlock(lock, fl);
225 1.27 yamt break;
226 1.4 mycroft
227 1.1 ws default:
228 1.31 fvdl break;
229 1.27 yamt /* NOTREACHED */
230 1.27 yamt }
231 1.27 yamt
232 1.27 yamt quit_unlock:
233 1.30 yamt simple_unlock(interlock);
234 1.27 yamt quit:
235 1.27 yamt if (lock)
236 1.4 mycroft FREE(lock, M_LOCKF);
237 1.27 yamt if (sparelock)
238 1.27 yamt FREE(sparelock, M_LOCKF);
239 1.27 yamt
240 1.29 yamt return error;
241 1.1 ws }
242 1.1 ws
243 1.1 ws /*
244 1.1 ws * Set a byte-range lock.
245 1.1 ws */
246 1.24 yamt static int
247 1.27 yamt lf_setlock(struct lockf *lock, struct lockf **sparelock,
248 1.27 yamt struct simplelock *interlock)
249 1.1 ws {
250 1.15 augustss struct lockf *block;
251 1.1 ws struct lockf **head = lock->lf_head;
252 1.1 ws struct lockf **prev, *overlap, *ltmp;
253 1.1 ws static char lockstr[] = "lockf";
254 1.1 ws int ovcase, priority, needtolink, error;
255 1.1 ws
256 1.1 ws #ifdef LOCKF_DEBUG
257 1.1 ws if (lockf_debug & 1)
258 1.1 ws lf_print("lf_setlock", lock);
259 1.1 ws #endif /* LOCKF_DEBUG */
260 1.1 ws
261 1.1 ws /*
262 1.1 ws * Set the priority
263 1.1 ws */
264 1.1 ws priority = PLOCK;
265 1.1 ws if (lock->lf_type == F_WRLCK)
266 1.1 ws priority += 4;
267 1.1 ws priority |= PCATCH;
268 1.1 ws /*
269 1.1 ws * Scan lock list for this file looking for locks that would block us.
270 1.1 ws */
271 1.7 christos while ((block = lf_getblock(lock)) != NULL) {
272 1.1 ws /*
273 1.1 ws * Free the structure and return if nonblocking.
274 1.1 ws */
275 1.1 ws if ((lock->lf_flags & F_WAIT) == 0) {
276 1.1 ws FREE(lock, M_LOCKF);
277 1.29 yamt return EAGAIN;
278 1.1 ws }
279 1.1 ws /*
280 1.1 ws * We are blocked. Since flock style locks cover
281 1.1 ws * the whole file, there is no chance for deadlock.
282 1.1 ws * For byte-range locks we must check for deadlock.
283 1.1 ws *
284 1.1 ws * Deadlock detection is done by looking through the
285 1.1 ws * wait channels to see if there are any cycles that
286 1.1 ws * involve us. MAXDEPTH is set just to make sure we
287 1.16 sommerfe * do not go off into neverneverland.
288 1.1 ws */
289 1.1 ws if ((lock->lf_flags & F_POSIX) &&
290 1.1 ws (block->lf_flags & F_POSIX)) {
291 1.21 thorpej struct lwp *wlwp;
292 1.15 augustss struct lockf *waitblock;
293 1.1 ws int i = 0;
294 1.1 ws
295 1.23 mycroft /*
296 1.23 mycroft * The block is waiting on something. if_lwp will be
297 1.23 mycroft * 0 once the lock is granted, so we terminate the
298 1.23 mycroft * loop if we find this.
299 1.23 mycroft */
300 1.23 mycroft wlwp = block->lf_lwp;
301 1.23 mycroft while (wlwp && (i++ < maxlockdepth)) {
302 1.21 thorpej waitblock = (struct lockf *)wlwp->l_wchan;
303 1.1 ws /* Get the owner of the blocking lock */
304 1.1 ws waitblock = waitblock->lf_next;
305 1.1 ws if ((waitblock->lf_flags & F_POSIX) == 0)
306 1.1 ws break;
307 1.23 mycroft wlwp = waitblock->lf_lwp;
308 1.23 mycroft if (wlwp == lock->lf_lwp) {
309 1.32 yamt free(lock, M_LOCKF);
310 1.29 yamt return EDEADLK;
311 1.1 ws }
312 1.1 ws }
313 1.16 sommerfe /*
314 1.16 sommerfe * If we're still following a dependancy chain
315 1.16 sommerfe * after maxlockdepth iterations, assume we're in
316 1.16 sommerfe * a cycle to be safe.
317 1.16 sommerfe */
318 1.16 sommerfe if (i >= maxlockdepth) {
319 1.32 yamt free(lock, M_LOCKF);
320 1.29 yamt return EDEADLK;
321 1.16 sommerfe }
322 1.1 ws }
323 1.1 ws /*
324 1.1 ws * For flock type locks, we must first remove
325 1.1 ws * any shared locks that we hold before we sleep
326 1.1 ws * waiting for an exclusive lock.
327 1.1 ws */
328 1.1 ws if ((lock->lf_flags & F_FLOCK) &&
329 1.1 ws lock->lf_type == F_WRLCK) {
330 1.1 ws lock->lf_type = F_UNLCK;
331 1.27 yamt (void) lf_clearlock(lock, NULL);
332 1.1 ws lock->lf_type = F_WRLCK;
333 1.1 ws }
334 1.1 ws /*
335 1.1 ws * Add our lock to the blocked list and sleep until we're free.
336 1.1 ws * Remember who blocked us (for deadlock detection).
337 1.1 ws */
338 1.1 ws lock->lf_next = block;
339 1.12 fvdl TAILQ_INSERT_TAIL(&block->lf_blkhd, lock, lf_block);
340 1.1 ws #ifdef LOCKF_DEBUG
341 1.1 ws if (lockf_debug & 1) {
342 1.1 ws lf_print("lf_setlock: blocking on", block);
343 1.1 ws lf_printlist("lf_setlock", block);
344 1.1 ws }
345 1.1 ws #endif /* LOCKF_DEBUG */
346 1.27 yamt error = ltsleep(lock, priority, lockstr, 0, interlock);
347 1.16 sommerfe
348 1.16 sommerfe /*
349 1.16 sommerfe * We may have been awakened by a signal (in
350 1.16 sommerfe * which case we must remove ourselves from the
351 1.16 sommerfe * blocked list) and/or by another process
352 1.16 sommerfe * releasing a lock (in which case we have already
353 1.16 sommerfe * been removed from the blocked list and our
354 1.16 sommerfe * lf_next field set to NOLOCKF).
355 1.16 sommerfe */
356 1.16 sommerfe if (lock->lf_next != NOLOCKF) {
357 1.16 sommerfe TAILQ_REMOVE(&lock->lf_next->lf_blkhd, lock, lf_block);
358 1.16 sommerfe lock->lf_next = NOLOCKF;
359 1.16 sommerfe }
360 1.7 christos if (error) {
361 1.32 yamt free(lock, M_LOCKF);
362 1.29 yamt return error;
363 1.1 ws }
364 1.1 ws }
365 1.1 ws /*
366 1.1 ws * No blocks!! Add the lock. Note that we will
367 1.1 ws * downgrade or upgrade any overlapping locks this
368 1.1 ws * process already owns.
369 1.1 ws *
370 1.1 ws * Skip over locks owned by other processes.
371 1.1 ws * Handle any locks that overlap and are owned by ourselves.
372 1.1 ws */
373 1.23 mycroft lock->lf_lwp = 0;
374 1.1 ws prev = head;
375 1.1 ws block = *head;
376 1.1 ws needtolink = 1;
377 1.1 ws for (;;) {
378 1.7 christos ovcase = lf_findoverlap(block, lock, SELF, &prev, &overlap);
379 1.7 christos if (ovcase)
380 1.1 ws block = overlap->lf_next;
381 1.1 ws /*
382 1.1 ws * Six cases:
383 1.1 ws * 0) no overlap
384 1.1 ws * 1) overlap == lock
385 1.1 ws * 2) overlap contains lock
386 1.1 ws * 3) lock contains overlap
387 1.1 ws * 4) overlap starts before lock
388 1.1 ws * 5) overlap ends after lock
389 1.1 ws */
390 1.1 ws switch (ovcase) {
391 1.1 ws case 0: /* no overlap */
392 1.1 ws if (needtolink) {
393 1.1 ws *prev = lock;
394 1.1 ws lock->lf_next = overlap;
395 1.1 ws }
396 1.1 ws break;
397 1.1 ws
398 1.1 ws case 1: /* overlap == lock */
399 1.1 ws /*
400 1.1 ws * If downgrading lock, others may be
401 1.1 ws * able to acquire it.
402 1.1 ws */
403 1.1 ws if (lock->lf_type == F_RDLCK &&
404 1.1 ws overlap->lf_type == F_WRLCK)
405 1.1 ws lf_wakelock(overlap);
406 1.1 ws overlap->lf_type = lock->lf_type;
407 1.1 ws FREE(lock, M_LOCKF);
408 1.1 ws lock = overlap; /* for debug output below */
409 1.1 ws break;
410 1.1 ws
411 1.1 ws case 2: /* overlap contains lock */
412 1.1 ws /*
413 1.1 ws * Check for common starting point and different types.
414 1.1 ws */
415 1.1 ws if (overlap->lf_type == lock->lf_type) {
416 1.32 yamt free(lock, M_LOCKF);
417 1.1 ws lock = overlap; /* for debug output below */
418 1.1 ws break;
419 1.1 ws }
420 1.1 ws if (overlap->lf_start == lock->lf_start) {
421 1.1 ws *prev = lock;
422 1.1 ws lock->lf_next = overlap;
423 1.1 ws overlap->lf_start = lock->lf_end + 1;
424 1.1 ws } else
425 1.27 yamt lf_split(overlap, lock, sparelock);
426 1.1 ws lf_wakelock(overlap);
427 1.1 ws break;
428 1.1 ws
429 1.1 ws case 3: /* lock contains overlap */
430 1.1 ws /*
431 1.1 ws * If downgrading lock, others may be able to
432 1.1 ws * acquire it, otherwise take the list.
433 1.1 ws */
434 1.1 ws if (lock->lf_type == F_RDLCK &&
435 1.1 ws overlap->lf_type == F_WRLCK) {
436 1.1 ws lf_wakelock(overlap);
437 1.1 ws } else {
438 1.19 matt while ((ltmp = TAILQ_FIRST(&overlap->lf_blkhd))) {
439 1.16 sommerfe KASSERT(ltmp->lf_next == overlap);
440 1.12 fvdl TAILQ_REMOVE(&overlap->lf_blkhd, ltmp,
441 1.12 fvdl lf_block);
442 1.16 sommerfe ltmp->lf_next = lock;
443 1.12 fvdl TAILQ_INSERT_TAIL(&lock->lf_blkhd,
444 1.12 fvdl ltmp, lf_block);
445 1.12 fvdl }
446 1.1 ws }
447 1.1 ws /*
448 1.1 ws * Add the new lock if necessary and delete the overlap.
449 1.1 ws */
450 1.1 ws if (needtolink) {
451 1.1 ws *prev = lock;
452 1.1 ws lock->lf_next = overlap->lf_next;
453 1.1 ws prev = &lock->lf_next;
454 1.1 ws needtolink = 0;
455 1.1 ws } else
456 1.1 ws *prev = overlap->lf_next;
457 1.32 yamt free(overlap, M_LOCKF);
458 1.1 ws continue;
459 1.1 ws
460 1.1 ws case 4: /* overlap starts before lock */
461 1.1 ws /*
462 1.1 ws * Add lock after overlap on the list.
463 1.1 ws */
464 1.1 ws lock->lf_next = overlap->lf_next;
465 1.1 ws overlap->lf_next = lock;
466 1.1 ws overlap->lf_end = lock->lf_start - 1;
467 1.1 ws prev = &lock->lf_next;
468 1.1 ws lf_wakelock(overlap);
469 1.1 ws needtolink = 0;
470 1.1 ws continue;
471 1.1 ws
472 1.1 ws case 5: /* overlap ends after lock */
473 1.1 ws /*
474 1.1 ws * Add the new lock before overlap.
475 1.1 ws */
476 1.1 ws if (needtolink) {
477 1.1 ws *prev = lock;
478 1.1 ws lock->lf_next = overlap;
479 1.1 ws }
480 1.1 ws overlap->lf_start = lock->lf_end + 1;
481 1.1 ws lf_wakelock(overlap);
482 1.1 ws break;
483 1.1 ws }
484 1.1 ws break;
485 1.1 ws }
486 1.1 ws #ifdef LOCKF_DEBUG
487 1.1 ws if (lockf_debug & 1) {
488 1.1 ws lf_print("lf_setlock: got the lock", lock);
489 1.1 ws lf_printlist("lf_setlock", lock);
490 1.1 ws }
491 1.1 ws #endif /* LOCKF_DEBUG */
492 1.29 yamt return 0;
493 1.1 ws }
494 1.1 ws
495 1.1 ws /*
496 1.1 ws * Remove a byte-range lock on an inode.
497 1.1 ws *
498 1.1 ws * Generally, find the lock (or an overlap to that lock)
499 1.1 ws * and remove it (or shrink it), then wakeup anyone we can.
500 1.1 ws */
501 1.24 yamt static int
502 1.27 yamt lf_clearlock(struct lockf *unlock, struct lockf **sparelock)
503 1.1 ws {
504 1.1 ws struct lockf **head = unlock->lf_head;
505 1.15 augustss struct lockf *lf = *head;
506 1.1 ws struct lockf *overlap, **prev;
507 1.1 ws int ovcase;
508 1.1 ws
509 1.1 ws if (lf == NOLOCKF)
510 1.29 yamt return 0;
511 1.1 ws #ifdef LOCKF_DEBUG
512 1.1 ws if (unlock->lf_type != F_UNLCK)
513 1.1 ws panic("lf_clearlock: bad type");
514 1.1 ws if (lockf_debug & 1)
515 1.1 ws lf_print("lf_clearlock", unlock);
516 1.1 ws #endif /* LOCKF_DEBUG */
517 1.1 ws prev = head;
518 1.7 christos while ((ovcase = lf_findoverlap(lf, unlock, SELF,
519 1.7 christos &prev, &overlap)) != 0) {
520 1.1 ws /*
521 1.1 ws * Wakeup the list of locks to be retried.
522 1.1 ws */
523 1.1 ws lf_wakelock(overlap);
524 1.1 ws
525 1.1 ws switch (ovcase) {
526 1.1 ws
527 1.1 ws case 1: /* overlap == lock */
528 1.1 ws *prev = overlap->lf_next;
529 1.1 ws FREE(overlap, M_LOCKF);
530 1.1 ws break;
531 1.1 ws
532 1.1 ws case 2: /* overlap contains lock: split it */
533 1.1 ws if (overlap->lf_start == unlock->lf_start) {
534 1.1 ws overlap->lf_start = unlock->lf_end + 1;
535 1.1 ws break;
536 1.1 ws }
537 1.27 yamt lf_split(overlap, unlock, sparelock);
538 1.1 ws overlap->lf_next = unlock->lf_next;
539 1.1 ws break;
540 1.1 ws
541 1.1 ws case 3: /* lock contains overlap */
542 1.1 ws *prev = overlap->lf_next;
543 1.1 ws lf = overlap->lf_next;
544 1.32 yamt free(overlap, M_LOCKF);
545 1.1 ws continue;
546 1.1 ws
547 1.1 ws case 4: /* overlap starts before lock */
548 1.1 ws overlap->lf_end = unlock->lf_start - 1;
549 1.1 ws prev = &overlap->lf_next;
550 1.1 ws lf = overlap->lf_next;
551 1.1 ws continue;
552 1.1 ws
553 1.1 ws case 5: /* overlap ends after lock */
554 1.1 ws overlap->lf_start = unlock->lf_end + 1;
555 1.1 ws break;
556 1.1 ws }
557 1.1 ws break;
558 1.1 ws }
559 1.1 ws #ifdef LOCKF_DEBUG
560 1.1 ws if (lockf_debug & 1)
561 1.1 ws lf_printlist("lf_clearlock", unlock);
562 1.1 ws #endif /* LOCKF_DEBUG */
563 1.29 yamt return 0;
564 1.1 ws }
565 1.1 ws
566 1.1 ws /*
567 1.1 ws * Check whether there is a blocking lock,
568 1.1 ws * and if so return its process identifier.
569 1.1 ws */
570 1.24 yamt static int
571 1.25 yamt lf_getlock(struct lockf *lock, struct flock *fl)
572 1.1 ws {
573 1.15 augustss struct lockf *block;
574 1.1 ws
575 1.1 ws #ifdef LOCKF_DEBUG
576 1.1 ws if (lockf_debug & 1)
577 1.1 ws lf_print("lf_getlock", lock);
578 1.1 ws #endif /* LOCKF_DEBUG */
579 1.1 ws
580 1.7 christos if ((block = lf_getblock(lock)) != NULL) {
581 1.1 ws fl->l_type = block->lf_type;
582 1.1 ws fl->l_whence = SEEK_SET;
583 1.1 ws fl->l_start = block->lf_start;
584 1.1 ws if (block->lf_end == -1)
585 1.1 ws fl->l_len = 0;
586 1.1 ws else
587 1.1 ws fl->l_len = block->lf_end - block->lf_start + 1;
588 1.1 ws if (block->lf_flags & F_POSIX)
589 1.23 mycroft fl->l_pid = ((struct proc *)block->lf_id)->p_pid;
590 1.1 ws else
591 1.1 ws fl->l_pid = -1;
592 1.1 ws } else {
593 1.1 ws fl->l_type = F_UNLCK;
594 1.1 ws }
595 1.29 yamt return 0;
596 1.1 ws }
597 1.1 ws
598 1.1 ws /*
599 1.1 ws * Walk the list of locks for an inode and
600 1.1 ws * return the first blocking lock.
601 1.1 ws */
602 1.24 yamt static struct lockf *
603 1.25 yamt lf_getblock(struct lockf *lock)
604 1.1 ws {
605 1.1 ws struct lockf **prev, *overlap, *lf = *(lock->lf_head);
606 1.1 ws
607 1.1 ws prev = lock->lf_head;
608 1.20 simonb while (lf_findoverlap(lf, lock, OTHERS, &prev, &overlap) != 0) {
609 1.1 ws /*
610 1.1 ws * We've found an overlap, see if it blocks us
611 1.1 ws */
612 1.1 ws if ((lock->lf_type == F_WRLCK || overlap->lf_type == F_WRLCK))
613 1.29 yamt return overlap;
614 1.1 ws /*
615 1.1 ws * Nope, point to the next one on the list and
616 1.1 ws * see if it blocks us
617 1.1 ws */
618 1.1 ws lf = overlap->lf_next;
619 1.1 ws }
620 1.29 yamt return NOLOCKF;
621 1.1 ws }
622 1.1 ws
623 1.1 ws /*
624 1.1 ws * Walk the list of locks for an inode to
625 1.1 ws * find an overlapping lock (if any).
626 1.1 ws *
627 1.1 ws * NOTE: this returns only the FIRST overlapping lock. There
628 1.1 ws * may be more than one.
629 1.1 ws */
630 1.24 yamt static int
631 1.25 yamt lf_findoverlap(struct lockf *lf, struct lockf *lock, int type,
632 1.25 yamt struct lockf ***prev, struct lockf **overlap)
633 1.1 ws {
634 1.1 ws off_t start, end;
635 1.1 ws
636 1.1 ws *overlap = lf;
637 1.1 ws if (lf == NOLOCKF)
638 1.29 yamt return 0;
639 1.1 ws #ifdef LOCKF_DEBUG
640 1.1 ws if (lockf_debug & 2)
641 1.1 ws lf_print("lf_findoverlap: looking for overlap in", lock);
642 1.1 ws #endif /* LOCKF_DEBUG */
643 1.1 ws start = lock->lf_start;
644 1.1 ws end = lock->lf_end;
645 1.1 ws while (lf != NOLOCKF) {
646 1.23 mycroft if (((type == SELF) && lf->lf_id != lock->lf_id) ||
647 1.23 mycroft ((type == OTHERS) && lf->lf_id == lock->lf_id)) {
648 1.1 ws *prev = &lf->lf_next;
649 1.1 ws *overlap = lf = lf->lf_next;
650 1.1 ws continue;
651 1.1 ws }
652 1.1 ws #ifdef LOCKF_DEBUG
653 1.1 ws if (lockf_debug & 2)
654 1.1 ws lf_print("\tchecking", lf);
655 1.1 ws #endif /* LOCKF_DEBUG */
656 1.1 ws /*
657 1.1 ws * OK, check for overlap
658 1.1 ws *
659 1.1 ws * Six cases:
660 1.1 ws * 0) no overlap
661 1.1 ws * 1) overlap == lock
662 1.1 ws * 2) overlap contains lock
663 1.1 ws * 3) lock contains overlap
664 1.1 ws * 4) overlap starts before lock
665 1.1 ws * 5) overlap ends after lock
666 1.1 ws */
667 1.1 ws if ((lf->lf_end != -1 && start > lf->lf_end) ||
668 1.1 ws (end != -1 && lf->lf_start > end)) {
669 1.1 ws /* Case 0 */
670 1.1 ws #ifdef LOCKF_DEBUG
671 1.1 ws if (lockf_debug & 2)
672 1.9 christos printf("no overlap\n");
673 1.1 ws #endif /* LOCKF_DEBUG */
674 1.1 ws if ((type & SELF) && end != -1 && lf->lf_start > end)
675 1.29 yamt return 0;
676 1.1 ws *prev = &lf->lf_next;
677 1.1 ws *overlap = lf = lf->lf_next;
678 1.1 ws continue;
679 1.1 ws }
680 1.1 ws if ((lf->lf_start == start) && (lf->lf_end == end)) {
681 1.1 ws /* Case 1 */
682 1.1 ws #ifdef LOCKF_DEBUG
683 1.1 ws if (lockf_debug & 2)
684 1.9 christos printf("overlap == lock\n");
685 1.1 ws #endif /* LOCKF_DEBUG */
686 1.29 yamt return 1;
687 1.1 ws }
688 1.1 ws if ((lf->lf_start <= start) &&
689 1.1 ws (end != -1) &&
690 1.1 ws ((lf->lf_end >= end) || (lf->lf_end == -1))) {
691 1.1 ws /* Case 2 */
692 1.1 ws #ifdef LOCKF_DEBUG
693 1.1 ws if (lockf_debug & 2)
694 1.9 christos printf("overlap contains lock\n");
695 1.1 ws #endif /* LOCKF_DEBUG */
696 1.29 yamt return 2;
697 1.1 ws }
698 1.1 ws if (start <= lf->lf_start &&
699 1.4 mycroft (end == -1 ||
700 1.1 ws (lf->lf_end != -1 && end >= lf->lf_end))) {
701 1.1 ws /* Case 3 */
702 1.1 ws #ifdef LOCKF_DEBUG
703 1.1 ws if (lockf_debug & 2)
704 1.9 christos printf("lock contains overlap\n");
705 1.1 ws #endif /* LOCKF_DEBUG */
706 1.29 yamt return 3;
707 1.1 ws }
708 1.1 ws if ((lf->lf_start < start) &&
709 1.1 ws ((lf->lf_end >= start) || (lf->lf_end == -1))) {
710 1.1 ws /* Case 4 */
711 1.1 ws #ifdef LOCKF_DEBUG
712 1.1 ws if (lockf_debug & 2)
713 1.9 christos printf("overlap starts before lock\n");
714 1.1 ws #endif /* LOCKF_DEBUG */
715 1.29 yamt return 4;
716 1.1 ws }
717 1.1 ws if ((lf->lf_start > start) &&
718 1.1 ws (end != -1) &&
719 1.1 ws ((lf->lf_end > end) || (lf->lf_end == -1))) {
720 1.1 ws /* Case 5 */
721 1.1 ws #ifdef LOCKF_DEBUG
722 1.1 ws if (lockf_debug & 2)
723 1.9 christos printf("overlap ends after lock\n");
724 1.1 ws #endif /* LOCKF_DEBUG */
725 1.29 yamt return 5;
726 1.1 ws }
727 1.1 ws panic("lf_findoverlap: default");
728 1.1 ws }
729 1.29 yamt return 0;
730 1.1 ws }
731 1.1 ws
732 1.1 ws /*
733 1.1 ws * Split a lock and a contained region into
734 1.1 ws * two or three locks as necessary.
735 1.1 ws */
736 1.24 yamt static void
737 1.27 yamt lf_split(struct lockf *lock1, struct lockf *lock2, struct lockf **sparelock)
738 1.1 ws {
739 1.15 augustss struct lockf *splitlock;
740 1.1 ws
741 1.1 ws #ifdef LOCKF_DEBUG
742 1.1 ws if (lockf_debug & 2) {
743 1.1 ws lf_print("lf_split", lock1);
744 1.1 ws lf_print("splitting from", lock2);
745 1.1 ws }
746 1.1 ws #endif /* LOCKF_DEBUG */
747 1.1 ws /*
748 1.1 ws * Check to see if spliting into only two pieces.
749 1.1 ws */
750 1.1 ws if (lock1->lf_start == lock2->lf_start) {
751 1.1 ws lock1->lf_start = lock2->lf_end + 1;
752 1.1 ws lock2->lf_next = lock1;
753 1.1 ws return;
754 1.1 ws }
755 1.1 ws if (lock1->lf_end == lock2->lf_end) {
756 1.1 ws lock1->lf_end = lock2->lf_start - 1;
757 1.1 ws lock2->lf_next = lock1->lf_next;
758 1.1 ws lock1->lf_next = lock2;
759 1.1 ws return;
760 1.1 ws }
761 1.1 ws /*
762 1.1 ws * Make a new lock consisting of the last part of
763 1.1 ws * the encompassing lock
764 1.1 ws */
765 1.27 yamt splitlock = *sparelock;
766 1.27 yamt *sparelock = NULL;
767 1.29 yamt memcpy(splitlock, lock1, sizeof(*splitlock));
768 1.1 ws splitlock->lf_start = lock2->lf_end + 1;
769 1.12 fvdl TAILQ_INIT(&splitlock->lf_blkhd);
770 1.1 ws lock1->lf_end = lock2->lf_start - 1;
771 1.1 ws /*
772 1.1 ws * OK, now link it in
773 1.1 ws */
774 1.1 ws splitlock->lf_next = lock1->lf_next;
775 1.1 ws lock2->lf_next = splitlock;
776 1.1 ws lock1->lf_next = lock2;
777 1.1 ws }
778 1.1 ws
779 1.1 ws /*
780 1.1 ws * Wakeup a blocklist
781 1.1 ws */
782 1.24 yamt static void
783 1.25 yamt lf_wakelock(struct lockf *listhead)
784 1.1 ws {
785 1.15 augustss struct lockf *wakelock;
786 1.1 ws
787 1.19 matt while ((wakelock = TAILQ_FIRST(&listhead->lf_blkhd))) {
788 1.16 sommerfe KASSERT(wakelock->lf_next == listhead);
789 1.12 fvdl TAILQ_REMOVE(&listhead->lf_blkhd, wakelock, lf_block);
790 1.1 ws wakelock->lf_next = NOLOCKF;
791 1.1 ws #ifdef LOCKF_DEBUG
792 1.1 ws if (lockf_debug & 2)
793 1.1 ws lf_print("lf_wakelock: awakening", wakelock);
794 1.12 fvdl #endif
795 1.29 yamt wakeup(wakelock);
796 1.1 ws }
797 1.1 ws }
798 1.1 ws
799 1.1 ws #ifdef LOCKF_DEBUG
800 1.1 ws /*
801 1.1 ws * Print out a lock.
802 1.1 ws */
803 1.24 yamt static void
804 1.25 yamt lf_print(char *tag, struct lockf *lock)
805 1.1 ws {
806 1.1 ws
807 1.9 christos printf("%s: lock %p for ", tag, lock);
808 1.1 ws if (lock->lf_flags & F_POSIX)
809 1.23 mycroft printf("proc %d", ((struct proc *)lock->lf_id)->p_pid);
810 1.1 ws else
811 1.23 mycroft printf("file 0x%p", (struct file *)lock->lf_id);
812 1.11 jtk printf(" %s, start %qx, end %qx",
813 1.1 ws lock->lf_type == F_RDLCK ? "shared" :
814 1.1 ws lock->lf_type == F_WRLCK ? "exclusive" :
815 1.1 ws lock->lf_type == F_UNLCK ? "unlock" :
816 1.1 ws "unknown", lock->lf_start, lock->lf_end);
817 1.19 matt if (TAILQ_FIRST(&lock->lf_blkhd))
818 1.19 matt printf(" block %p\n", TAILQ_FIRST(&lock->lf_blkhd));
819 1.1 ws else
820 1.9 christos printf("\n");
821 1.1 ws }
822 1.1 ws
823 1.24 yamt static void
824 1.25 yamt lf_printlist(char *tag, struct lockf *lock)
825 1.1 ws {
826 1.15 augustss struct lockf *lf, *blk;
827 1.1 ws
828 1.11 jtk printf("%s: Lock list:\n", tag);
829 1.12 fvdl for (lf = *lock->lf_head; lf; lf = lf->lf_next) {
830 1.9 christos printf("\tlock %p for ", lf);
831 1.1 ws if (lf->lf_flags & F_POSIX)
832 1.23 mycroft printf("proc %d", ((struct proc *)lf->lf_id)->p_pid);
833 1.1 ws else
834 1.23 mycroft printf("file 0x%p", (struct file *)lf->lf_id);
835 1.11 jtk printf(", %s, start %qx, end %qx",
836 1.1 ws lf->lf_type == F_RDLCK ? "shared" :
837 1.1 ws lf->lf_type == F_WRLCK ? "exclusive" :
838 1.1 ws lf->lf_type == F_UNLCK ? "unlock" :
839 1.1 ws "unknown", lf->lf_start, lf->lf_end);
840 1.19 matt TAILQ_FOREACH(blk, &lf->lf_blkhd, lf_block) {
841 1.12 fvdl if (blk->lf_flags & F_POSIX)
842 1.12 fvdl printf("proc %d",
843 1.23 mycroft ((struct proc *)blk->lf_id)->p_pid);
844 1.12 fvdl else
845 1.23 mycroft printf("file 0x%p", (struct file *)blk->lf_id);
846 1.12 fvdl printf(", %s, start %qx, end %qx",
847 1.12 fvdl blk->lf_type == F_RDLCK ? "shared" :
848 1.12 fvdl blk->lf_type == F_WRLCK ? "exclusive" :
849 1.12 fvdl blk->lf_type == F_UNLCK ? "unlock" :
850 1.12 fvdl "unknown", blk->lf_start, blk->lf_end);
851 1.19 matt if (TAILQ_FIRST(&blk->lf_blkhd))
852 1.12 fvdl panic("lf_printlist: bad list");
853 1.12 fvdl }
854 1.12 fvdl printf("\n");
855 1.1 ws }
856 1.1 ws }
857 1.1 ws #endif /* LOCKF_DEBUG */
858