vfs_lockf.c revision 1.31 1 1.31 fvdl /* $NetBSD: vfs_lockf.c,v 1.31 2003/05/03 15:02:54 fvdl 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 1.31 fvdl __KERNEL_RCSID(0, "$NetBSD: vfs_lockf.c,v 1.31 2003/05/03 15:02:54 fvdl 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 1.22 thorpej
54 1.22 thorpej MALLOC_DEFINE(M_LOCKF, "lockf", "Byte-range locking structures");
55 1.1 ws
56 1.1 ws /*
57 1.6 mycroft * This variable controls the maximum number of processes that will
58 1.6 mycroft * be checked in doing deadlock detection.
59 1.6 mycroft */
60 1.6 mycroft int maxlockdepth = MAXDEPTH;
61 1.6 mycroft
62 1.6 mycroft #ifdef LOCKF_DEBUG
63 1.6 mycroft int lockf_debug = 0;
64 1.6 mycroft #endif
65 1.6 mycroft
66 1.6 mycroft #define NOLOCKF (struct lockf *)0
67 1.6 mycroft #define SELF 0x1
68 1.6 mycroft #define OTHERS 0x2
69 1.6 mycroft
70 1.27 yamt static int lf_clearlock(struct lockf *, struct lockf **);
71 1.25 yamt static int lf_findoverlap(struct lockf *,
72 1.25 yamt struct lockf *, int, struct lockf ***, struct lockf **);
73 1.25 yamt static struct lockf *lf_getblock(struct lockf *);
74 1.25 yamt static int lf_getlock(struct lockf *, struct flock *);
75 1.27 yamt static int lf_setlock(struct lockf *, struct lockf **, struct simplelock *);
76 1.27 yamt static void lf_split(struct lockf *, struct lockf *, struct lockf **);
77 1.25 yamt static void lf_wakelock(struct lockf *);
78 1.24 yamt
79 1.24 yamt #ifdef LOCKF_DEBUG
80 1.25 yamt static void lf_print(char *, struct lockf *);
81 1.25 yamt static void lf_printlist(char *, struct lockf *);
82 1.24 yamt #endif
83 1.24 yamt
84 1.6 mycroft /*
85 1.16 sommerfe * XXX TODO
86 1.16 sommerfe * Misc cleanups: "caddr_t id" should be visible in the API as a
87 1.16 sommerfe * "struct proc *".
88 1.16 sommerfe * (This requires rototilling all VFS's which support advisory locking).
89 1.16 sommerfe *
90 1.16 sommerfe * Use pools for lock allocation.
91 1.16 sommerfe */
92 1.16 sommerfe
93 1.16 sommerfe /*
94 1.16 sommerfe * If there's a lot of lock contention on a single vnode, locking
95 1.16 sommerfe * schemes which allow for more paralleism would be needed. Given how
96 1.16 sommerfe * infrequently byte-range locks are actually used in typical BSD
97 1.16 sommerfe * code, a more complex approach probably isn't worth it.
98 1.16 sommerfe */
99 1.16 sommerfe
100 1.16 sommerfe /*
101 1.4 mycroft * Do an advisory lock operation.
102 1.1 ws */
103 1.4 mycroft int
104 1.25 yamt lf_advlock(struct vop_advlock_args *ap, struct lockf **head, off_t size)
105 1.1 ws {
106 1.17 jdolecek struct flock *fl = ap->a_fl;
107 1.27 yamt struct lockf *lock = NULL;
108 1.27 yamt struct lockf *sparelock;
109 1.30 yamt struct simplelock *interlock = &ap->a_vp->v_interlock;
110 1.1 ws off_t start, end;
111 1.1 ws int error;
112 1.1 ws
113 1.1 ws /*
114 1.1 ws * Convert the flock structure into a start and end.
115 1.1 ws */
116 1.1 ws switch (fl->l_whence) {
117 1.1 ws case SEEK_SET:
118 1.1 ws case SEEK_CUR:
119 1.1 ws /*
120 1.1 ws * Caller is responsible for adding any necessary offset
121 1.1 ws * when SEEK_CUR is used.
122 1.1 ws */
123 1.1 ws start = fl->l_start;
124 1.1 ws break;
125 1.1 ws
126 1.1 ws case SEEK_END:
127 1.1 ws start = size + fl->l_start;
128 1.1 ws break;
129 1.1 ws
130 1.1 ws default:
131 1.29 yamt return EINVAL;
132 1.1 ws }
133 1.1 ws if (start < 0)
134 1.29 yamt return EINVAL;
135 1.10 kleink
136 1.10 kleink /*
137 1.27 yamt * allocate locks before acquire simple lock.
138 1.27 yamt * we need two locks in the worst case.
139 1.27 yamt */
140 1.27 yamt switch (ap->a_op) {
141 1.27 yamt case F_SETLK:
142 1.27 yamt case F_UNLCK:
143 1.27 yamt /*
144 1.27 yamt * XXX for F_UNLCK case, we can re-use lock.
145 1.27 yamt */
146 1.27 yamt if ((fl->l_type & F_FLOCK) == 0) {
147 1.27 yamt /*
148 1.27 yamt * byte-range lock might need one more lock.
149 1.27 yamt */
150 1.27 yamt MALLOC(sparelock, struct lockf *, sizeof(*lock),
151 1.27 yamt M_LOCKF, M_WAITOK);
152 1.27 yamt if (sparelock == NULL) {
153 1.27 yamt error = ENOMEM;
154 1.27 yamt goto quit;
155 1.27 yamt }
156 1.27 yamt break;
157 1.27 yamt }
158 1.27 yamt /* FALLTHROUGH */
159 1.27 yamt
160 1.27 yamt case F_GETLK:
161 1.27 yamt sparelock = NULL;
162 1.27 yamt break;
163 1.27 yamt
164 1.27 yamt default:
165 1.29 yamt return EINVAL;
166 1.27 yamt }
167 1.27 yamt
168 1.27 yamt MALLOC(lock, struct lockf *, sizeof(*lock), M_LOCKF, M_WAITOK);
169 1.27 yamt if (lock == NULL) {
170 1.27 yamt error = ENOMEM;
171 1.27 yamt goto quit;
172 1.27 yamt }
173 1.27 yamt
174 1.30 yamt simple_lock(interlock);
175 1.27 yamt
176 1.27 yamt /*
177 1.10 kleink * Avoid the common case of unlocking when inode has no locks.
178 1.10 kleink */
179 1.10 kleink if (*head == (struct lockf *)0) {
180 1.17 jdolecek if (ap->a_op != F_SETLK) {
181 1.10 kleink fl->l_type = F_UNLCK;
182 1.27 yamt error = 0;
183 1.27 yamt goto quit_unlock;
184 1.10 kleink }
185 1.10 kleink }
186 1.10 kleink
187 1.1 ws if (fl->l_len == 0)
188 1.1 ws end = -1;
189 1.1 ws else
190 1.1 ws end = start + fl->l_len - 1;
191 1.1 ws /*
192 1.4 mycroft * Create the lockf structure.
193 1.1 ws */
194 1.1 ws lock->lf_start = start;
195 1.1 ws lock->lf_end = end;
196 1.21 thorpej /* XXX NJWLWP
197 1.21 thorpej * I don't want to make the entire VFS universe use LWPs, because
198 1.21 thorpej * they don't need them, for the most part. This is an exception,
199 1.21 thorpej * and a kluge.
200 1.21 thorpej */
201 1.21 thorpej
202 1.1 ws lock->lf_head = head;
203 1.1 ws lock->lf_type = fl->l_type;
204 1.1 ws lock->lf_next = (struct lockf *)0;
205 1.12 fvdl TAILQ_INIT(&lock->lf_blkhd);
206 1.17 jdolecek lock->lf_flags = ap->a_flags;
207 1.21 thorpej if (lock->lf_flags & F_POSIX) {
208 1.21 thorpej KASSERT(curproc == (struct proc *)ap->a_id);
209 1.21 thorpej }
210 1.23 mycroft lock->lf_id = (struct proc *)ap->a_id;
211 1.23 mycroft lock->lf_lwp = curlwp;
212 1.21 thorpej
213 1.1 ws /*
214 1.1 ws * Do the requested operation.
215 1.1 ws */
216 1.17 jdolecek switch (ap->a_op) {
217 1.4 mycroft
218 1.1 ws case F_SETLK:
219 1.30 yamt error = lf_setlock(lock, &sparelock, interlock);
220 1.27 yamt lock = NULL; /* lf_setlock freed it */
221 1.27 yamt break;
222 1.1 ws
223 1.1 ws case F_UNLCK:
224 1.27 yamt error = lf_clearlock(lock, &sparelock);
225 1.27 yamt break;
226 1.1 ws
227 1.1 ws case F_GETLK:
228 1.1 ws error = lf_getlock(lock, fl);
229 1.27 yamt break;
230 1.4 mycroft
231 1.1 ws default:
232 1.31 fvdl break;
233 1.27 yamt /* NOTREACHED */
234 1.27 yamt }
235 1.27 yamt
236 1.27 yamt quit_unlock:
237 1.30 yamt simple_unlock(interlock);
238 1.27 yamt quit:
239 1.27 yamt if (lock)
240 1.4 mycroft FREE(lock, M_LOCKF);
241 1.27 yamt if (sparelock)
242 1.27 yamt FREE(sparelock, M_LOCKF);
243 1.27 yamt
244 1.29 yamt return error;
245 1.1 ws }
246 1.1 ws
247 1.1 ws /*
248 1.1 ws * Set a byte-range lock.
249 1.1 ws */
250 1.24 yamt static int
251 1.27 yamt lf_setlock(struct lockf *lock, struct lockf **sparelock,
252 1.27 yamt struct simplelock *interlock)
253 1.1 ws {
254 1.15 augustss struct lockf *block;
255 1.1 ws struct lockf **head = lock->lf_head;
256 1.1 ws struct lockf **prev, *overlap, *ltmp;
257 1.1 ws static char lockstr[] = "lockf";
258 1.1 ws int ovcase, priority, needtolink, error;
259 1.1 ws
260 1.1 ws #ifdef LOCKF_DEBUG
261 1.1 ws if (lockf_debug & 1)
262 1.1 ws lf_print("lf_setlock", lock);
263 1.1 ws #endif /* LOCKF_DEBUG */
264 1.1 ws
265 1.1 ws /*
266 1.1 ws * Set the priority
267 1.1 ws */
268 1.1 ws priority = PLOCK;
269 1.1 ws if (lock->lf_type == F_WRLCK)
270 1.1 ws priority += 4;
271 1.1 ws priority |= PCATCH;
272 1.1 ws /*
273 1.1 ws * Scan lock list for this file looking for locks that would block us.
274 1.1 ws */
275 1.7 christos while ((block = lf_getblock(lock)) != NULL) {
276 1.1 ws /*
277 1.1 ws * Free the structure and return if nonblocking.
278 1.1 ws */
279 1.1 ws if ((lock->lf_flags & F_WAIT) == 0) {
280 1.1 ws FREE(lock, M_LOCKF);
281 1.29 yamt return EAGAIN;
282 1.1 ws }
283 1.1 ws /*
284 1.1 ws * We are blocked. Since flock style locks cover
285 1.1 ws * the whole file, there is no chance for deadlock.
286 1.1 ws * For byte-range locks we must check for deadlock.
287 1.1 ws *
288 1.1 ws * Deadlock detection is done by looking through the
289 1.1 ws * wait channels to see if there are any cycles that
290 1.1 ws * involve us. MAXDEPTH is set just to make sure we
291 1.16 sommerfe * do not go off into neverneverland.
292 1.1 ws */
293 1.1 ws if ((lock->lf_flags & F_POSIX) &&
294 1.1 ws (block->lf_flags & F_POSIX)) {
295 1.21 thorpej struct lwp *wlwp;
296 1.15 augustss struct lockf *waitblock;
297 1.1 ws int i = 0;
298 1.1 ws
299 1.23 mycroft /*
300 1.23 mycroft * The block is waiting on something. if_lwp will be
301 1.23 mycroft * 0 once the lock is granted, so we terminate the
302 1.23 mycroft * loop if we find this.
303 1.23 mycroft */
304 1.23 mycroft wlwp = block->lf_lwp;
305 1.23 mycroft while (wlwp && (i++ < maxlockdepth)) {
306 1.21 thorpej waitblock = (struct lockf *)wlwp->l_wchan;
307 1.1 ws /* Get the owner of the blocking lock */
308 1.1 ws waitblock = waitblock->lf_next;
309 1.1 ws if ((waitblock->lf_flags & F_POSIX) == 0)
310 1.1 ws break;
311 1.23 mycroft wlwp = waitblock->lf_lwp;
312 1.23 mycroft if (wlwp == lock->lf_lwp) {
313 1.28 yamt FREE(lock, M_LOCKF);
314 1.29 yamt return EDEADLK;
315 1.1 ws }
316 1.1 ws }
317 1.16 sommerfe /*
318 1.16 sommerfe * If we're still following a dependancy chain
319 1.16 sommerfe * after maxlockdepth iterations, assume we're in
320 1.16 sommerfe * a cycle to be safe.
321 1.16 sommerfe */
322 1.16 sommerfe if (i >= maxlockdepth) {
323 1.28 yamt FREE(lock, M_LOCKF);
324 1.29 yamt return EDEADLK;
325 1.16 sommerfe }
326 1.1 ws }
327 1.1 ws /*
328 1.1 ws * For flock type locks, we must first remove
329 1.1 ws * any shared locks that we hold before we sleep
330 1.1 ws * waiting for an exclusive lock.
331 1.1 ws */
332 1.1 ws if ((lock->lf_flags & F_FLOCK) &&
333 1.1 ws lock->lf_type == F_WRLCK) {
334 1.1 ws lock->lf_type = F_UNLCK;
335 1.27 yamt (void) lf_clearlock(lock, NULL);
336 1.1 ws lock->lf_type = F_WRLCK;
337 1.1 ws }
338 1.1 ws /*
339 1.1 ws * Add our lock to the blocked list and sleep until we're free.
340 1.1 ws * Remember who blocked us (for deadlock detection).
341 1.1 ws */
342 1.1 ws lock->lf_next = block;
343 1.12 fvdl TAILQ_INSERT_TAIL(&block->lf_blkhd, lock, lf_block);
344 1.1 ws #ifdef LOCKF_DEBUG
345 1.1 ws if (lockf_debug & 1) {
346 1.1 ws lf_print("lf_setlock: blocking on", block);
347 1.1 ws lf_printlist("lf_setlock", block);
348 1.1 ws }
349 1.1 ws #endif /* LOCKF_DEBUG */
350 1.27 yamt error = ltsleep(lock, priority, lockstr, 0, interlock);
351 1.16 sommerfe
352 1.16 sommerfe /*
353 1.16 sommerfe * We may have been awakened by a signal (in
354 1.16 sommerfe * which case we must remove ourselves from the
355 1.16 sommerfe * blocked list) and/or by another process
356 1.16 sommerfe * releasing a lock (in which case we have already
357 1.16 sommerfe * been removed from the blocked list and our
358 1.16 sommerfe * lf_next field set to NOLOCKF).
359 1.16 sommerfe */
360 1.16 sommerfe if (lock->lf_next != NOLOCKF) {
361 1.16 sommerfe TAILQ_REMOVE(&lock->lf_next->lf_blkhd, lock, lf_block);
362 1.16 sommerfe lock->lf_next = NOLOCKF;
363 1.16 sommerfe }
364 1.7 christos if (error) {
365 1.28 yamt FREE(lock, M_LOCKF);
366 1.29 yamt return error;
367 1.1 ws }
368 1.1 ws }
369 1.1 ws /*
370 1.1 ws * No blocks!! Add the lock. Note that we will
371 1.1 ws * downgrade or upgrade any overlapping locks this
372 1.1 ws * process already owns.
373 1.1 ws *
374 1.1 ws * Skip over locks owned by other processes.
375 1.1 ws * Handle any locks that overlap and are owned by ourselves.
376 1.1 ws */
377 1.23 mycroft lock->lf_lwp = 0;
378 1.1 ws prev = head;
379 1.1 ws block = *head;
380 1.1 ws needtolink = 1;
381 1.1 ws for (;;) {
382 1.7 christos ovcase = lf_findoverlap(block, lock, SELF, &prev, &overlap);
383 1.7 christos if (ovcase)
384 1.1 ws block = overlap->lf_next;
385 1.1 ws /*
386 1.1 ws * Six cases:
387 1.1 ws * 0) no overlap
388 1.1 ws * 1) overlap == lock
389 1.1 ws * 2) overlap contains lock
390 1.1 ws * 3) lock contains overlap
391 1.1 ws * 4) overlap starts before lock
392 1.1 ws * 5) overlap ends after lock
393 1.1 ws */
394 1.1 ws switch (ovcase) {
395 1.1 ws case 0: /* no overlap */
396 1.1 ws if (needtolink) {
397 1.1 ws *prev = lock;
398 1.1 ws lock->lf_next = overlap;
399 1.1 ws }
400 1.1 ws break;
401 1.1 ws
402 1.1 ws case 1: /* overlap == lock */
403 1.1 ws /*
404 1.1 ws * If downgrading lock, others may be
405 1.1 ws * able to acquire it.
406 1.1 ws */
407 1.1 ws if (lock->lf_type == F_RDLCK &&
408 1.1 ws overlap->lf_type == F_WRLCK)
409 1.1 ws lf_wakelock(overlap);
410 1.1 ws overlap->lf_type = lock->lf_type;
411 1.1 ws FREE(lock, M_LOCKF);
412 1.1 ws lock = overlap; /* for debug output below */
413 1.1 ws break;
414 1.1 ws
415 1.1 ws case 2: /* overlap contains lock */
416 1.1 ws /*
417 1.1 ws * Check for common starting point and different types.
418 1.1 ws */
419 1.1 ws if (overlap->lf_type == lock->lf_type) {
420 1.28 yamt FREE(lock, M_LOCKF);
421 1.1 ws lock = overlap; /* for debug output below */
422 1.1 ws break;
423 1.1 ws }
424 1.1 ws if (overlap->lf_start == lock->lf_start) {
425 1.1 ws *prev = lock;
426 1.1 ws lock->lf_next = overlap;
427 1.1 ws overlap->lf_start = lock->lf_end + 1;
428 1.1 ws } else
429 1.27 yamt lf_split(overlap, lock, sparelock);
430 1.1 ws lf_wakelock(overlap);
431 1.1 ws break;
432 1.1 ws
433 1.1 ws case 3: /* lock contains overlap */
434 1.1 ws /*
435 1.1 ws * If downgrading lock, others may be able to
436 1.1 ws * acquire it, otherwise take the list.
437 1.1 ws */
438 1.1 ws if (lock->lf_type == F_RDLCK &&
439 1.1 ws overlap->lf_type == F_WRLCK) {
440 1.1 ws lf_wakelock(overlap);
441 1.1 ws } else {
442 1.19 matt while ((ltmp = TAILQ_FIRST(&overlap->lf_blkhd))) {
443 1.16 sommerfe KASSERT(ltmp->lf_next == overlap);
444 1.12 fvdl TAILQ_REMOVE(&overlap->lf_blkhd, ltmp,
445 1.12 fvdl lf_block);
446 1.16 sommerfe ltmp->lf_next = lock;
447 1.12 fvdl TAILQ_INSERT_TAIL(&lock->lf_blkhd,
448 1.12 fvdl ltmp, lf_block);
449 1.12 fvdl }
450 1.1 ws }
451 1.1 ws /*
452 1.1 ws * Add the new lock if necessary and delete the overlap.
453 1.1 ws */
454 1.1 ws if (needtolink) {
455 1.1 ws *prev = lock;
456 1.1 ws lock->lf_next = overlap->lf_next;
457 1.1 ws prev = &lock->lf_next;
458 1.1 ws needtolink = 0;
459 1.1 ws } else
460 1.1 ws *prev = overlap->lf_next;
461 1.28 yamt FREE(overlap, M_LOCKF);
462 1.1 ws continue;
463 1.1 ws
464 1.1 ws case 4: /* overlap starts before lock */
465 1.1 ws /*
466 1.1 ws * Add lock after overlap on the list.
467 1.1 ws */
468 1.1 ws lock->lf_next = overlap->lf_next;
469 1.1 ws overlap->lf_next = lock;
470 1.1 ws overlap->lf_end = lock->lf_start - 1;
471 1.1 ws prev = &lock->lf_next;
472 1.1 ws lf_wakelock(overlap);
473 1.1 ws needtolink = 0;
474 1.1 ws continue;
475 1.1 ws
476 1.1 ws case 5: /* overlap ends after lock */
477 1.1 ws /*
478 1.1 ws * Add the new lock before overlap.
479 1.1 ws */
480 1.1 ws if (needtolink) {
481 1.1 ws *prev = lock;
482 1.1 ws lock->lf_next = overlap;
483 1.1 ws }
484 1.1 ws overlap->lf_start = lock->lf_end + 1;
485 1.1 ws lf_wakelock(overlap);
486 1.1 ws break;
487 1.1 ws }
488 1.1 ws break;
489 1.1 ws }
490 1.1 ws #ifdef LOCKF_DEBUG
491 1.1 ws if (lockf_debug & 1) {
492 1.1 ws lf_print("lf_setlock: got the lock", lock);
493 1.1 ws lf_printlist("lf_setlock", lock);
494 1.1 ws }
495 1.1 ws #endif /* LOCKF_DEBUG */
496 1.29 yamt return 0;
497 1.1 ws }
498 1.1 ws
499 1.1 ws /*
500 1.1 ws * Remove a byte-range lock on an inode.
501 1.1 ws *
502 1.1 ws * Generally, find the lock (or an overlap to that lock)
503 1.1 ws * and remove it (or shrink it), then wakeup anyone we can.
504 1.1 ws */
505 1.24 yamt static int
506 1.27 yamt lf_clearlock(struct lockf *unlock, struct lockf **sparelock)
507 1.1 ws {
508 1.1 ws struct lockf **head = unlock->lf_head;
509 1.15 augustss struct lockf *lf = *head;
510 1.1 ws struct lockf *overlap, **prev;
511 1.1 ws int ovcase;
512 1.1 ws
513 1.1 ws if (lf == NOLOCKF)
514 1.29 yamt return 0;
515 1.1 ws #ifdef LOCKF_DEBUG
516 1.1 ws if (unlock->lf_type != F_UNLCK)
517 1.1 ws panic("lf_clearlock: bad type");
518 1.1 ws if (lockf_debug & 1)
519 1.1 ws lf_print("lf_clearlock", unlock);
520 1.1 ws #endif /* LOCKF_DEBUG */
521 1.1 ws prev = head;
522 1.7 christos while ((ovcase = lf_findoverlap(lf, unlock, SELF,
523 1.7 christos &prev, &overlap)) != 0) {
524 1.1 ws /*
525 1.1 ws * Wakeup the list of locks to be retried.
526 1.1 ws */
527 1.1 ws lf_wakelock(overlap);
528 1.1 ws
529 1.1 ws switch (ovcase) {
530 1.1 ws
531 1.1 ws case 1: /* overlap == lock */
532 1.1 ws *prev = overlap->lf_next;
533 1.1 ws FREE(overlap, M_LOCKF);
534 1.1 ws break;
535 1.1 ws
536 1.1 ws case 2: /* overlap contains lock: split it */
537 1.1 ws if (overlap->lf_start == unlock->lf_start) {
538 1.1 ws overlap->lf_start = unlock->lf_end + 1;
539 1.1 ws break;
540 1.1 ws }
541 1.27 yamt lf_split(overlap, unlock, sparelock);
542 1.1 ws overlap->lf_next = unlock->lf_next;
543 1.1 ws break;
544 1.1 ws
545 1.1 ws case 3: /* lock contains overlap */
546 1.1 ws *prev = overlap->lf_next;
547 1.1 ws lf = overlap->lf_next;
548 1.28 yamt FREE(overlap, M_LOCKF);
549 1.1 ws continue;
550 1.1 ws
551 1.1 ws case 4: /* overlap starts before lock */
552 1.1 ws overlap->lf_end = unlock->lf_start - 1;
553 1.1 ws prev = &overlap->lf_next;
554 1.1 ws lf = overlap->lf_next;
555 1.1 ws continue;
556 1.1 ws
557 1.1 ws case 5: /* overlap ends after lock */
558 1.1 ws overlap->lf_start = unlock->lf_end + 1;
559 1.1 ws break;
560 1.1 ws }
561 1.1 ws break;
562 1.1 ws }
563 1.1 ws #ifdef LOCKF_DEBUG
564 1.1 ws if (lockf_debug & 1)
565 1.1 ws lf_printlist("lf_clearlock", unlock);
566 1.1 ws #endif /* LOCKF_DEBUG */
567 1.29 yamt return 0;
568 1.1 ws }
569 1.1 ws
570 1.1 ws /*
571 1.1 ws * Check whether there is a blocking lock,
572 1.1 ws * and if so return its process identifier.
573 1.1 ws */
574 1.24 yamt static int
575 1.25 yamt lf_getlock(struct lockf *lock, struct flock *fl)
576 1.1 ws {
577 1.15 augustss struct lockf *block;
578 1.1 ws
579 1.1 ws #ifdef LOCKF_DEBUG
580 1.1 ws if (lockf_debug & 1)
581 1.1 ws lf_print("lf_getlock", lock);
582 1.1 ws #endif /* LOCKF_DEBUG */
583 1.1 ws
584 1.7 christos if ((block = lf_getblock(lock)) != NULL) {
585 1.1 ws fl->l_type = block->lf_type;
586 1.1 ws fl->l_whence = SEEK_SET;
587 1.1 ws fl->l_start = block->lf_start;
588 1.1 ws if (block->lf_end == -1)
589 1.1 ws fl->l_len = 0;
590 1.1 ws else
591 1.1 ws fl->l_len = block->lf_end - block->lf_start + 1;
592 1.1 ws if (block->lf_flags & F_POSIX)
593 1.23 mycroft fl->l_pid = ((struct proc *)block->lf_id)->p_pid;
594 1.1 ws else
595 1.1 ws fl->l_pid = -1;
596 1.1 ws } else {
597 1.1 ws fl->l_type = F_UNLCK;
598 1.1 ws }
599 1.29 yamt return 0;
600 1.1 ws }
601 1.1 ws
602 1.1 ws /*
603 1.1 ws * Walk the list of locks for an inode and
604 1.1 ws * return the first blocking lock.
605 1.1 ws */
606 1.24 yamt static struct lockf *
607 1.25 yamt lf_getblock(struct lockf *lock)
608 1.1 ws {
609 1.1 ws struct lockf **prev, *overlap, *lf = *(lock->lf_head);
610 1.1 ws
611 1.1 ws prev = lock->lf_head;
612 1.20 simonb while (lf_findoverlap(lf, lock, OTHERS, &prev, &overlap) != 0) {
613 1.1 ws /*
614 1.1 ws * We've found an overlap, see if it blocks us
615 1.1 ws */
616 1.1 ws if ((lock->lf_type == F_WRLCK || overlap->lf_type == F_WRLCK))
617 1.29 yamt return overlap;
618 1.1 ws /*
619 1.1 ws * Nope, point to the next one on the list and
620 1.1 ws * see if it blocks us
621 1.1 ws */
622 1.1 ws lf = overlap->lf_next;
623 1.1 ws }
624 1.29 yamt return NOLOCKF;
625 1.1 ws }
626 1.1 ws
627 1.1 ws /*
628 1.1 ws * Walk the list of locks for an inode to
629 1.1 ws * find an overlapping lock (if any).
630 1.1 ws *
631 1.1 ws * NOTE: this returns only the FIRST overlapping lock. There
632 1.1 ws * may be more than one.
633 1.1 ws */
634 1.24 yamt static int
635 1.25 yamt lf_findoverlap(struct lockf *lf, struct lockf *lock, int type,
636 1.25 yamt struct lockf ***prev, struct lockf **overlap)
637 1.1 ws {
638 1.1 ws off_t start, end;
639 1.1 ws
640 1.1 ws *overlap = lf;
641 1.1 ws if (lf == NOLOCKF)
642 1.29 yamt return 0;
643 1.1 ws #ifdef LOCKF_DEBUG
644 1.1 ws if (lockf_debug & 2)
645 1.1 ws lf_print("lf_findoverlap: looking for overlap in", lock);
646 1.1 ws #endif /* LOCKF_DEBUG */
647 1.1 ws start = lock->lf_start;
648 1.1 ws end = lock->lf_end;
649 1.1 ws while (lf != NOLOCKF) {
650 1.23 mycroft if (((type == SELF) && lf->lf_id != lock->lf_id) ||
651 1.23 mycroft ((type == OTHERS) && lf->lf_id == lock->lf_id)) {
652 1.1 ws *prev = &lf->lf_next;
653 1.1 ws *overlap = lf = lf->lf_next;
654 1.1 ws continue;
655 1.1 ws }
656 1.1 ws #ifdef LOCKF_DEBUG
657 1.1 ws if (lockf_debug & 2)
658 1.1 ws lf_print("\tchecking", lf);
659 1.1 ws #endif /* LOCKF_DEBUG */
660 1.1 ws /*
661 1.1 ws * OK, check for overlap
662 1.1 ws *
663 1.1 ws * Six cases:
664 1.1 ws * 0) no overlap
665 1.1 ws * 1) overlap == lock
666 1.1 ws * 2) overlap contains lock
667 1.1 ws * 3) lock contains overlap
668 1.1 ws * 4) overlap starts before lock
669 1.1 ws * 5) overlap ends after lock
670 1.1 ws */
671 1.1 ws if ((lf->lf_end != -1 && start > lf->lf_end) ||
672 1.1 ws (end != -1 && lf->lf_start > end)) {
673 1.1 ws /* Case 0 */
674 1.1 ws #ifdef LOCKF_DEBUG
675 1.1 ws if (lockf_debug & 2)
676 1.9 christos printf("no overlap\n");
677 1.1 ws #endif /* LOCKF_DEBUG */
678 1.1 ws if ((type & SELF) && end != -1 && lf->lf_start > end)
679 1.29 yamt return 0;
680 1.1 ws *prev = &lf->lf_next;
681 1.1 ws *overlap = lf = lf->lf_next;
682 1.1 ws continue;
683 1.1 ws }
684 1.1 ws if ((lf->lf_start == start) && (lf->lf_end == end)) {
685 1.1 ws /* Case 1 */
686 1.1 ws #ifdef LOCKF_DEBUG
687 1.1 ws if (lockf_debug & 2)
688 1.9 christos printf("overlap == lock\n");
689 1.1 ws #endif /* LOCKF_DEBUG */
690 1.29 yamt return 1;
691 1.1 ws }
692 1.1 ws if ((lf->lf_start <= start) &&
693 1.1 ws (end != -1) &&
694 1.1 ws ((lf->lf_end >= end) || (lf->lf_end == -1))) {
695 1.1 ws /* Case 2 */
696 1.1 ws #ifdef LOCKF_DEBUG
697 1.1 ws if (lockf_debug & 2)
698 1.9 christos printf("overlap contains lock\n");
699 1.1 ws #endif /* LOCKF_DEBUG */
700 1.29 yamt return 2;
701 1.1 ws }
702 1.1 ws if (start <= lf->lf_start &&
703 1.4 mycroft (end == -1 ||
704 1.1 ws (lf->lf_end != -1 && end >= lf->lf_end))) {
705 1.1 ws /* Case 3 */
706 1.1 ws #ifdef LOCKF_DEBUG
707 1.1 ws if (lockf_debug & 2)
708 1.9 christos printf("lock contains overlap\n");
709 1.1 ws #endif /* LOCKF_DEBUG */
710 1.29 yamt return 3;
711 1.1 ws }
712 1.1 ws if ((lf->lf_start < start) &&
713 1.1 ws ((lf->lf_end >= start) || (lf->lf_end == -1))) {
714 1.1 ws /* Case 4 */
715 1.1 ws #ifdef LOCKF_DEBUG
716 1.1 ws if (lockf_debug & 2)
717 1.9 christos printf("overlap starts before lock\n");
718 1.1 ws #endif /* LOCKF_DEBUG */
719 1.29 yamt return 4;
720 1.1 ws }
721 1.1 ws if ((lf->lf_start > start) &&
722 1.1 ws (end != -1) &&
723 1.1 ws ((lf->lf_end > end) || (lf->lf_end == -1))) {
724 1.1 ws /* Case 5 */
725 1.1 ws #ifdef LOCKF_DEBUG
726 1.1 ws if (lockf_debug & 2)
727 1.9 christos printf("overlap ends after lock\n");
728 1.1 ws #endif /* LOCKF_DEBUG */
729 1.29 yamt return 5;
730 1.1 ws }
731 1.1 ws panic("lf_findoverlap: default");
732 1.1 ws }
733 1.29 yamt return 0;
734 1.1 ws }
735 1.1 ws
736 1.1 ws /*
737 1.1 ws * Split a lock and a contained region into
738 1.1 ws * two or three locks as necessary.
739 1.1 ws */
740 1.24 yamt static void
741 1.27 yamt lf_split(struct lockf *lock1, struct lockf *lock2, struct lockf **sparelock)
742 1.1 ws {
743 1.15 augustss struct lockf *splitlock;
744 1.1 ws
745 1.1 ws #ifdef LOCKF_DEBUG
746 1.1 ws if (lockf_debug & 2) {
747 1.1 ws lf_print("lf_split", lock1);
748 1.1 ws lf_print("splitting from", lock2);
749 1.1 ws }
750 1.1 ws #endif /* LOCKF_DEBUG */
751 1.1 ws /*
752 1.1 ws * Check to see if spliting into only two pieces.
753 1.1 ws */
754 1.1 ws if (lock1->lf_start == lock2->lf_start) {
755 1.1 ws lock1->lf_start = lock2->lf_end + 1;
756 1.1 ws lock2->lf_next = lock1;
757 1.1 ws return;
758 1.1 ws }
759 1.1 ws if (lock1->lf_end == lock2->lf_end) {
760 1.1 ws lock1->lf_end = lock2->lf_start - 1;
761 1.1 ws lock2->lf_next = lock1->lf_next;
762 1.1 ws lock1->lf_next = lock2;
763 1.1 ws return;
764 1.1 ws }
765 1.1 ws /*
766 1.1 ws * Make a new lock consisting of the last part of
767 1.1 ws * the encompassing lock
768 1.1 ws */
769 1.27 yamt splitlock = *sparelock;
770 1.27 yamt *sparelock = NULL;
771 1.29 yamt memcpy(splitlock, lock1, sizeof(*splitlock));
772 1.1 ws splitlock->lf_start = lock2->lf_end + 1;
773 1.12 fvdl TAILQ_INIT(&splitlock->lf_blkhd);
774 1.1 ws lock1->lf_end = lock2->lf_start - 1;
775 1.1 ws /*
776 1.1 ws * OK, now link it in
777 1.1 ws */
778 1.1 ws splitlock->lf_next = lock1->lf_next;
779 1.1 ws lock2->lf_next = splitlock;
780 1.1 ws lock1->lf_next = lock2;
781 1.1 ws }
782 1.1 ws
783 1.1 ws /*
784 1.1 ws * Wakeup a blocklist
785 1.1 ws */
786 1.24 yamt static void
787 1.25 yamt lf_wakelock(struct lockf *listhead)
788 1.1 ws {
789 1.15 augustss struct lockf *wakelock;
790 1.1 ws
791 1.19 matt while ((wakelock = TAILQ_FIRST(&listhead->lf_blkhd))) {
792 1.16 sommerfe KASSERT(wakelock->lf_next == listhead);
793 1.12 fvdl TAILQ_REMOVE(&listhead->lf_blkhd, wakelock, lf_block);
794 1.1 ws wakelock->lf_next = NOLOCKF;
795 1.1 ws #ifdef LOCKF_DEBUG
796 1.1 ws if (lockf_debug & 2)
797 1.1 ws lf_print("lf_wakelock: awakening", wakelock);
798 1.12 fvdl #endif
799 1.29 yamt wakeup(wakelock);
800 1.1 ws }
801 1.1 ws }
802 1.1 ws
803 1.1 ws #ifdef LOCKF_DEBUG
804 1.1 ws /*
805 1.1 ws * Print out a lock.
806 1.1 ws */
807 1.24 yamt static void
808 1.25 yamt lf_print(char *tag, struct lockf *lock)
809 1.1 ws {
810 1.1 ws
811 1.9 christos printf("%s: lock %p for ", tag, lock);
812 1.1 ws if (lock->lf_flags & F_POSIX)
813 1.23 mycroft printf("proc %d", ((struct proc *)lock->lf_id)->p_pid);
814 1.1 ws else
815 1.23 mycroft printf("file 0x%p", (struct file *)lock->lf_id);
816 1.11 jtk printf(" %s, start %qx, end %qx",
817 1.1 ws lock->lf_type == F_RDLCK ? "shared" :
818 1.1 ws lock->lf_type == F_WRLCK ? "exclusive" :
819 1.1 ws lock->lf_type == F_UNLCK ? "unlock" :
820 1.1 ws "unknown", lock->lf_start, lock->lf_end);
821 1.19 matt if (TAILQ_FIRST(&lock->lf_blkhd))
822 1.19 matt printf(" block %p\n", TAILQ_FIRST(&lock->lf_blkhd));
823 1.1 ws else
824 1.9 christos printf("\n");
825 1.1 ws }
826 1.1 ws
827 1.24 yamt static void
828 1.25 yamt lf_printlist(char *tag, struct lockf *lock)
829 1.1 ws {
830 1.15 augustss struct lockf *lf, *blk;
831 1.1 ws
832 1.11 jtk printf("%s: Lock list:\n", tag);
833 1.12 fvdl for (lf = *lock->lf_head; lf; lf = lf->lf_next) {
834 1.9 christos printf("\tlock %p for ", lf);
835 1.1 ws if (lf->lf_flags & F_POSIX)
836 1.23 mycroft printf("proc %d", ((struct proc *)lf->lf_id)->p_pid);
837 1.1 ws else
838 1.23 mycroft printf("file 0x%p", (struct file *)lf->lf_id);
839 1.11 jtk printf(", %s, start %qx, end %qx",
840 1.1 ws lf->lf_type == F_RDLCK ? "shared" :
841 1.1 ws lf->lf_type == F_WRLCK ? "exclusive" :
842 1.1 ws lf->lf_type == F_UNLCK ? "unlock" :
843 1.1 ws "unknown", lf->lf_start, lf->lf_end);
844 1.19 matt TAILQ_FOREACH(blk, &lf->lf_blkhd, lf_block) {
845 1.12 fvdl if (blk->lf_flags & F_POSIX)
846 1.12 fvdl printf("proc %d",
847 1.23 mycroft ((struct proc *)blk->lf_id)->p_pid);
848 1.12 fvdl else
849 1.23 mycroft printf("file 0x%p", (struct file *)blk->lf_id);
850 1.12 fvdl printf(", %s, start %qx, end %qx",
851 1.12 fvdl blk->lf_type == F_RDLCK ? "shared" :
852 1.12 fvdl blk->lf_type == F_WRLCK ? "exclusive" :
853 1.12 fvdl blk->lf_type == F_UNLCK ? "unlock" :
854 1.12 fvdl "unknown", blk->lf_start, blk->lf_end);
855 1.19 matt if (TAILQ_FIRST(&blk->lf_blkhd))
856 1.12 fvdl panic("lf_printlist: bad list");
857 1.12 fvdl }
858 1.12 fvdl printf("\n");
859 1.1 ws }
860 1.1 ws }
861 1.1 ws #endif /* LOCKF_DEBUG */
862