vfs_lockf.c revision 1.17.2.3 1 1.17.2.3 nathanw /* $NetBSD: vfs_lockf.c,v 1.17.2.3 2002/06/24 22:11:07 nathanw 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 1.17.2.2 nathanw
41 1.17.2.2 nathanw #include <sys/cdefs.h>
42 1.17.2.3 nathanw __KERNEL_RCSID(0, "$NetBSD: vfs_lockf.c,v 1.17.2.3 2002/06/24 22:11:07 nathanw 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 1.17.2.1 nathanw #include <sys/lwp.h>
49 1.1 ws #include <sys/proc.h>
50 1.1 ws #include <sys/vnode.h>
51 1.1 ws #include <sys/malloc.h>
52 1.1 ws #include <sys/fcntl.h>
53 1.1 ws #include <sys/lockf.h>
54 1.1 ws
55 1.1 ws /*
56 1.6 mycroft * This variable controls the maximum number of processes that will
57 1.6 mycroft * be checked in doing deadlock detection.
58 1.6 mycroft */
59 1.6 mycroft int maxlockdepth = MAXDEPTH;
60 1.6 mycroft
61 1.6 mycroft #ifdef LOCKF_DEBUG
62 1.6 mycroft int lockf_debug = 0;
63 1.6 mycroft #endif
64 1.6 mycroft
65 1.6 mycroft #define NOLOCKF (struct lockf *)0
66 1.6 mycroft #define SELF 0x1
67 1.6 mycroft #define OTHERS 0x2
68 1.6 mycroft
69 1.6 mycroft /*
70 1.16 sommerfe * XXX TODO
71 1.16 sommerfe * Misc cleanups: "caddr_t id" should be visible in the API as a
72 1.16 sommerfe * "struct proc *".
73 1.16 sommerfe * (This requires rototilling all VFS's which support advisory locking).
74 1.16 sommerfe *
75 1.16 sommerfe * Use pools for lock allocation.
76 1.16 sommerfe */
77 1.16 sommerfe
78 1.16 sommerfe /*
79 1.16 sommerfe * XXXSMP TODO: Using either (a) a global lock, or (b) the vnode's
80 1.16 sommerfe * interlock should be sufficient; (b) requires a change to the API
81 1.16 sommerfe * because the vnode isn't visible here.
82 1.16 sommerfe *
83 1.16 sommerfe * If there's a lot of lock contention on a single vnode, locking
84 1.16 sommerfe * schemes which allow for more paralleism would be needed. Given how
85 1.16 sommerfe * infrequently byte-range locks are actually used in typical BSD
86 1.16 sommerfe * code, a more complex approach probably isn't worth it.
87 1.16 sommerfe */
88 1.16 sommerfe
89 1.16 sommerfe /*
90 1.4 mycroft * Do an advisory lock operation.
91 1.1 ws */
92 1.4 mycroft int
93 1.17 jdolecek lf_advlock(ap, head, size)
94 1.17 jdolecek struct vop_advlock_args *ap;
95 1.1 ws struct lockf **head;
96 1.3 cgd off_t size;
97 1.1 ws {
98 1.17 jdolecek struct flock *fl = ap->a_fl;
99 1.15 augustss struct lockf *lock;
100 1.1 ws off_t start, end;
101 1.1 ws int error;
102 1.1 ws
103 1.1 ws /*
104 1.1 ws * Convert the flock structure into a start and end.
105 1.1 ws */
106 1.1 ws switch (fl->l_whence) {
107 1.1 ws case SEEK_SET:
108 1.1 ws case SEEK_CUR:
109 1.1 ws /*
110 1.1 ws * Caller is responsible for adding any necessary offset
111 1.1 ws * when SEEK_CUR is used.
112 1.1 ws */
113 1.1 ws start = fl->l_start;
114 1.1 ws break;
115 1.1 ws
116 1.1 ws case SEEK_END:
117 1.1 ws start = size + fl->l_start;
118 1.1 ws break;
119 1.1 ws
120 1.1 ws default:
121 1.1 ws return (EINVAL);
122 1.1 ws }
123 1.1 ws if (start < 0)
124 1.1 ws return (EINVAL);
125 1.10 kleink
126 1.10 kleink /*
127 1.10 kleink * Avoid the common case of unlocking when inode has no locks.
128 1.10 kleink */
129 1.10 kleink if (*head == (struct lockf *)0) {
130 1.17 jdolecek if (ap->a_op != F_SETLK) {
131 1.10 kleink fl->l_type = F_UNLCK;
132 1.10 kleink return (0);
133 1.10 kleink }
134 1.10 kleink }
135 1.10 kleink
136 1.1 ws if (fl->l_len == 0)
137 1.1 ws end = -1;
138 1.1 ws else
139 1.1 ws end = start + fl->l_len - 1;
140 1.1 ws /*
141 1.4 mycroft * Create the lockf structure.
142 1.1 ws */
143 1.13 perry MALLOC(lock, struct lockf *, sizeof(*lock), M_LOCKF, M_WAITOK);
144 1.1 ws lock->lf_start = start;
145 1.1 ws lock->lf_end = end;
146 1.17.2.1 nathanw /* XXX NJWLWP
147 1.17.2.1 nathanw * I don't want to make the entire VFS universe use LWPs, because
148 1.17.2.1 nathanw * they don't need them, for the most part. This is an exception,
149 1.17.2.1 nathanw * and a kluge.
150 1.17.2.1 nathanw */
151 1.17.2.1 nathanw
152 1.1 ws lock->lf_head = head;
153 1.1 ws lock->lf_type = fl->l_type;
154 1.1 ws lock->lf_next = (struct lockf *)0;
155 1.12 fvdl TAILQ_INIT(&lock->lf_blkhd);
156 1.17 jdolecek lock->lf_flags = ap->a_flags;
157 1.17.2.1 nathanw if (lock->lf_flags & F_POSIX) {
158 1.17.2.3 nathanw KASSERT(curproc == (struct proc *)ap->a_id);
159 1.17.2.3 nathanw lock->lf_id = (caddr_t) curlwp;
160 1.17.2.1 nathanw } else {
161 1.17.2.1 nathanw lock->lf_id = ap->a_id; /* Not a proc at all, but a file * */
162 1.17.2.1 nathanw }
163 1.17.2.1 nathanw
164 1.17.2.1 nathanw
165 1.1 ws /*
166 1.1 ws * Do the requested operation.
167 1.1 ws */
168 1.17 jdolecek switch (ap->a_op) {
169 1.4 mycroft
170 1.1 ws case F_SETLK:
171 1.1 ws return (lf_setlock(lock));
172 1.1 ws
173 1.1 ws case F_UNLCK:
174 1.1 ws error = lf_clearlock(lock);
175 1.1 ws FREE(lock, M_LOCKF);
176 1.1 ws return (error);
177 1.1 ws
178 1.1 ws case F_GETLK:
179 1.1 ws error = lf_getlock(lock, fl);
180 1.1 ws FREE(lock, M_LOCKF);
181 1.1 ws return (error);
182 1.4 mycroft
183 1.1 ws default:
184 1.4 mycroft FREE(lock, M_LOCKF);
185 1.1 ws return (EINVAL);
186 1.1 ws }
187 1.1 ws /* NOTREACHED */
188 1.1 ws }
189 1.1 ws
190 1.1 ws /*
191 1.1 ws * Set a byte-range lock.
192 1.1 ws */
193 1.4 mycroft int
194 1.1 ws lf_setlock(lock)
195 1.15 augustss struct lockf *lock;
196 1.1 ws {
197 1.15 augustss struct lockf *block;
198 1.1 ws struct lockf **head = lock->lf_head;
199 1.1 ws struct lockf **prev, *overlap, *ltmp;
200 1.1 ws static char lockstr[] = "lockf";
201 1.1 ws int ovcase, priority, needtolink, error;
202 1.1 ws
203 1.1 ws #ifdef LOCKF_DEBUG
204 1.1 ws if (lockf_debug & 1)
205 1.1 ws lf_print("lf_setlock", lock);
206 1.1 ws #endif /* LOCKF_DEBUG */
207 1.1 ws
208 1.1 ws /*
209 1.1 ws * Set the priority
210 1.1 ws */
211 1.1 ws priority = PLOCK;
212 1.1 ws if (lock->lf_type == F_WRLCK)
213 1.1 ws priority += 4;
214 1.1 ws priority |= PCATCH;
215 1.1 ws /*
216 1.1 ws * Scan lock list for this file looking for locks that would block us.
217 1.1 ws */
218 1.7 christos while ((block = lf_getblock(lock)) != NULL) {
219 1.1 ws /*
220 1.1 ws * Free the structure and return if nonblocking.
221 1.1 ws */
222 1.1 ws if ((lock->lf_flags & F_WAIT) == 0) {
223 1.1 ws FREE(lock, M_LOCKF);
224 1.1 ws return (EAGAIN);
225 1.1 ws }
226 1.1 ws /*
227 1.1 ws * We are blocked. Since flock style locks cover
228 1.1 ws * the whole file, there is no chance for deadlock.
229 1.1 ws * For byte-range locks we must check for deadlock.
230 1.1 ws *
231 1.1 ws * Deadlock detection is done by looking through the
232 1.1 ws * wait channels to see if there are any cycles that
233 1.1 ws * involve us. MAXDEPTH is set just to make sure we
234 1.16 sommerfe * do not go off into neverneverland.
235 1.1 ws */
236 1.1 ws if ((lock->lf_flags & F_POSIX) &&
237 1.1 ws (block->lf_flags & F_POSIX)) {
238 1.17.2.1 nathanw struct lwp *wlwp;
239 1.15 augustss struct lockf *waitblock;
240 1.1 ws int i = 0;
241 1.1 ws
242 1.1 ws /* The block is waiting on something */
243 1.17.2.1 nathanw wlwp = (struct lwp *)block->lf_id;
244 1.17.2.1 nathanw while (wlwp->l_wchan &&
245 1.17.2.1 nathanw (wlwp->l_wmesg == lockstr) &&
246 1.1 ws (i++ < maxlockdepth)) {
247 1.17.2.1 nathanw waitblock = (struct lockf *)wlwp->l_wchan;
248 1.1 ws /* Get the owner of the blocking lock */
249 1.1 ws waitblock = waitblock->lf_next;
250 1.1 ws if ((waitblock->lf_flags & F_POSIX) == 0)
251 1.1 ws break;
252 1.17.2.1 nathanw wlwp = (struct lwp *)waitblock->lf_id;
253 1.17.2.1 nathanw if (wlwp == (struct lwp *)lock->lf_id) {
254 1.1 ws free(lock, M_LOCKF);
255 1.1 ws return (EDEADLK);
256 1.1 ws }
257 1.1 ws }
258 1.16 sommerfe /*
259 1.16 sommerfe * If we're still following a dependancy chain
260 1.16 sommerfe * after maxlockdepth iterations, assume we're in
261 1.16 sommerfe * a cycle to be safe.
262 1.16 sommerfe */
263 1.16 sommerfe if (i >= maxlockdepth) {
264 1.16 sommerfe free(lock, M_LOCKF);
265 1.16 sommerfe return (EDEADLK);
266 1.16 sommerfe }
267 1.1 ws }
268 1.1 ws /*
269 1.1 ws * For flock type locks, we must first remove
270 1.1 ws * any shared locks that we hold before we sleep
271 1.1 ws * waiting for an exclusive lock.
272 1.1 ws */
273 1.1 ws if ((lock->lf_flags & F_FLOCK) &&
274 1.1 ws lock->lf_type == F_WRLCK) {
275 1.1 ws lock->lf_type = F_UNLCK;
276 1.1 ws (void) lf_clearlock(lock);
277 1.1 ws lock->lf_type = F_WRLCK;
278 1.1 ws }
279 1.1 ws /*
280 1.1 ws * Add our lock to the blocked list and sleep until we're free.
281 1.1 ws * Remember who blocked us (for deadlock detection).
282 1.1 ws */
283 1.1 ws lock->lf_next = block;
284 1.12 fvdl TAILQ_INSERT_TAIL(&block->lf_blkhd, lock, lf_block);
285 1.1 ws #ifdef LOCKF_DEBUG
286 1.1 ws if (lockf_debug & 1) {
287 1.1 ws lf_print("lf_setlock: blocking on", block);
288 1.1 ws lf_printlist("lf_setlock", block);
289 1.1 ws }
290 1.1 ws #endif /* LOCKF_DEBUG */
291 1.7 christos error = tsleep((caddr_t)lock, priority, lockstr, 0);
292 1.16 sommerfe
293 1.16 sommerfe /*
294 1.16 sommerfe * We may have been awakened by a signal (in
295 1.16 sommerfe * which case we must remove ourselves from the
296 1.16 sommerfe * blocked list) and/or by another process
297 1.16 sommerfe * releasing a lock (in which case we have already
298 1.16 sommerfe * been removed from the blocked list and our
299 1.16 sommerfe * lf_next field set to NOLOCKF).
300 1.16 sommerfe */
301 1.16 sommerfe if (lock->lf_next != NOLOCKF) {
302 1.16 sommerfe TAILQ_REMOVE(&lock->lf_next->lf_blkhd, lock, lf_block);
303 1.16 sommerfe lock->lf_next = NOLOCKF;
304 1.16 sommerfe }
305 1.7 christos if (error) {
306 1.4 mycroft free(lock, M_LOCKF);
307 1.4 mycroft return (error);
308 1.1 ws }
309 1.1 ws }
310 1.1 ws /*
311 1.1 ws * No blocks!! Add the lock. Note that we will
312 1.1 ws * downgrade or upgrade any overlapping locks this
313 1.1 ws * process already owns.
314 1.1 ws *
315 1.1 ws * Skip over locks owned by other processes.
316 1.1 ws * Handle any locks that overlap and are owned by ourselves.
317 1.1 ws */
318 1.1 ws prev = head;
319 1.1 ws block = *head;
320 1.1 ws needtolink = 1;
321 1.1 ws for (;;) {
322 1.7 christos ovcase = lf_findoverlap(block, lock, SELF, &prev, &overlap);
323 1.7 christos if (ovcase)
324 1.1 ws block = overlap->lf_next;
325 1.1 ws /*
326 1.1 ws * Six cases:
327 1.1 ws * 0) no overlap
328 1.1 ws * 1) overlap == lock
329 1.1 ws * 2) overlap contains lock
330 1.1 ws * 3) lock contains overlap
331 1.1 ws * 4) overlap starts before lock
332 1.1 ws * 5) overlap ends after lock
333 1.1 ws */
334 1.1 ws switch (ovcase) {
335 1.1 ws case 0: /* no overlap */
336 1.1 ws if (needtolink) {
337 1.1 ws *prev = lock;
338 1.1 ws lock->lf_next = overlap;
339 1.1 ws }
340 1.1 ws break;
341 1.1 ws
342 1.1 ws case 1: /* overlap == lock */
343 1.1 ws /*
344 1.1 ws * If downgrading lock, others may be
345 1.1 ws * able to acquire it.
346 1.1 ws */
347 1.1 ws if (lock->lf_type == F_RDLCK &&
348 1.1 ws overlap->lf_type == F_WRLCK)
349 1.1 ws lf_wakelock(overlap);
350 1.1 ws overlap->lf_type = lock->lf_type;
351 1.1 ws FREE(lock, M_LOCKF);
352 1.1 ws lock = overlap; /* for debug output below */
353 1.1 ws break;
354 1.1 ws
355 1.1 ws case 2: /* overlap contains lock */
356 1.1 ws /*
357 1.1 ws * Check for common starting point and different types.
358 1.1 ws */
359 1.1 ws if (overlap->lf_type == lock->lf_type) {
360 1.1 ws free(lock, M_LOCKF);
361 1.1 ws lock = overlap; /* for debug output below */
362 1.1 ws break;
363 1.1 ws }
364 1.1 ws if (overlap->lf_start == lock->lf_start) {
365 1.1 ws *prev = lock;
366 1.1 ws lock->lf_next = overlap;
367 1.1 ws overlap->lf_start = lock->lf_end + 1;
368 1.1 ws } else
369 1.1 ws lf_split(overlap, lock);
370 1.1 ws lf_wakelock(overlap);
371 1.1 ws break;
372 1.1 ws
373 1.1 ws case 3: /* lock contains overlap */
374 1.1 ws /*
375 1.1 ws * If downgrading lock, others may be able to
376 1.1 ws * acquire it, otherwise take the list.
377 1.1 ws */
378 1.1 ws if (lock->lf_type == F_RDLCK &&
379 1.1 ws overlap->lf_type == F_WRLCK) {
380 1.1 ws lf_wakelock(overlap);
381 1.1 ws } else {
382 1.12 fvdl while ((ltmp = overlap->lf_blkhd.tqh_first)) {
383 1.16 sommerfe KASSERT(ltmp->lf_next == overlap);
384 1.12 fvdl TAILQ_REMOVE(&overlap->lf_blkhd, ltmp,
385 1.12 fvdl lf_block);
386 1.16 sommerfe ltmp->lf_next = lock;
387 1.12 fvdl TAILQ_INSERT_TAIL(&lock->lf_blkhd,
388 1.12 fvdl ltmp, lf_block);
389 1.12 fvdl }
390 1.1 ws }
391 1.1 ws /*
392 1.1 ws * Add the new lock if necessary and delete the overlap.
393 1.1 ws */
394 1.1 ws if (needtolink) {
395 1.1 ws *prev = lock;
396 1.1 ws lock->lf_next = overlap->lf_next;
397 1.1 ws prev = &lock->lf_next;
398 1.1 ws needtolink = 0;
399 1.1 ws } else
400 1.1 ws *prev = overlap->lf_next;
401 1.1 ws free(overlap, M_LOCKF);
402 1.1 ws continue;
403 1.1 ws
404 1.1 ws case 4: /* overlap starts before lock */
405 1.1 ws /*
406 1.1 ws * Add lock after overlap on the list.
407 1.1 ws */
408 1.1 ws lock->lf_next = overlap->lf_next;
409 1.1 ws overlap->lf_next = lock;
410 1.1 ws overlap->lf_end = lock->lf_start - 1;
411 1.1 ws prev = &lock->lf_next;
412 1.1 ws lf_wakelock(overlap);
413 1.1 ws needtolink = 0;
414 1.1 ws continue;
415 1.1 ws
416 1.1 ws case 5: /* overlap ends after lock */
417 1.1 ws /*
418 1.1 ws * Add the new lock before overlap.
419 1.1 ws */
420 1.1 ws if (needtolink) {
421 1.1 ws *prev = lock;
422 1.1 ws lock->lf_next = overlap;
423 1.1 ws }
424 1.1 ws overlap->lf_start = lock->lf_end + 1;
425 1.1 ws lf_wakelock(overlap);
426 1.1 ws break;
427 1.1 ws }
428 1.1 ws break;
429 1.1 ws }
430 1.1 ws #ifdef LOCKF_DEBUG
431 1.1 ws if (lockf_debug & 1) {
432 1.1 ws lf_print("lf_setlock: got the lock", lock);
433 1.1 ws lf_printlist("lf_setlock", lock);
434 1.1 ws }
435 1.1 ws #endif /* LOCKF_DEBUG */
436 1.1 ws return (0);
437 1.1 ws }
438 1.1 ws
439 1.1 ws /*
440 1.1 ws * Remove a byte-range lock on an inode.
441 1.1 ws *
442 1.1 ws * Generally, find the lock (or an overlap to that lock)
443 1.1 ws * and remove it (or shrink it), then wakeup anyone we can.
444 1.1 ws */
445 1.4 mycroft int
446 1.1 ws lf_clearlock(unlock)
447 1.15 augustss struct lockf *unlock;
448 1.1 ws {
449 1.1 ws struct lockf **head = unlock->lf_head;
450 1.15 augustss struct lockf *lf = *head;
451 1.1 ws struct lockf *overlap, **prev;
452 1.1 ws int ovcase;
453 1.1 ws
454 1.1 ws if (lf == NOLOCKF)
455 1.1 ws return (0);
456 1.1 ws #ifdef LOCKF_DEBUG
457 1.1 ws if (unlock->lf_type != F_UNLCK)
458 1.1 ws panic("lf_clearlock: bad type");
459 1.1 ws if (lockf_debug & 1)
460 1.1 ws lf_print("lf_clearlock", unlock);
461 1.1 ws #endif /* LOCKF_DEBUG */
462 1.1 ws prev = head;
463 1.7 christos while ((ovcase = lf_findoverlap(lf, unlock, SELF,
464 1.7 christos &prev, &overlap)) != 0) {
465 1.1 ws /*
466 1.1 ws * Wakeup the list of locks to be retried.
467 1.1 ws */
468 1.1 ws lf_wakelock(overlap);
469 1.1 ws
470 1.1 ws switch (ovcase) {
471 1.1 ws
472 1.1 ws case 1: /* overlap == lock */
473 1.1 ws *prev = overlap->lf_next;
474 1.1 ws FREE(overlap, M_LOCKF);
475 1.1 ws break;
476 1.1 ws
477 1.1 ws case 2: /* overlap contains lock: split it */
478 1.1 ws if (overlap->lf_start == unlock->lf_start) {
479 1.1 ws overlap->lf_start = unlock->lf_end + 1;
480 1.1 ws break;
481 1.1 ws }
482 1.1 ws lf_split(overlap, unlock);
483 1.1 ws overlap->lf_next = unlock->lf_next;
484 1.1 ws break;
485 1.1 ws
486 1.1 ws case 3: /* lock contains overlap */
487 1.1 ws *prev = overlap->lf_next;
488 1.1 ws lf = overlap->lf_next;
489 1.1 ws free(overlap, M_LOCKF);
490 1.1 ws continue;
491 1.1 ws
492 1.1 ws case 4: /* overlap starts before lock */
493 1.1 ws overlap->lf_end = unlock->lf_start - 1;
494 1.1 ws prev = &overlap->lf_next;
495 1.1 ws lf = overlap->lf_next;
496 1.1 ws continue;
497 1.1 ws
498 1.1 ws case 5: /* overlap ends after lock */
499 1.1 ws overlap->lf_start = unlock->lf_end + 1;
500 1.1 ws break;
501 1.1 ws }
502 1.1 ws break;
503 1.1 ws }
504 1.1 ws #ifdef LOCKF_DEBUG
505 1.1 ws if (lockf_debug & 1)
506 1.1 ws lf_printlist("lf_clearlock", unlock);
507 1.1 ws #endif /* LOCKF_DEBUG */
508 1.1 ws return (0);
509 1.1 ws }
510 1.1 ws
511 1.1 ws /*
512 1.1 ws * Check whether there is a blocking lock,
513 1.1 ws * and if so return its process identifier.
514 1.1 ws */
515 1.4 mycroft int
516 1.1 ws lf_getlock(lock, fl)
517 1.15 augustss struct lockf *lock;
518 1.15 augustss struct flock *fl;
519 1.1 ws {
520 1.15 augustss struct lockf *block;
521 1.1 ws
522 1.1 ws #ifdef LOCKF_DEBUG
523 1.1 ws if (lockf_debug & 1)
524 1.1 ws lf_print("lf_getlock", lock);
525 1.1 ws #endif /* LOCKF_DEBUG */
526 1.1 ws
527 1.7 christos if ((block = lf_getblock(lock)) != NULL) {
528 1.1 ws fl->l_type = block->lf_type;
529 1.1 ws fl->l_whence = SEEK_SET;
530 1.1 ws fl->l_start = block->lf_start;
531 1.1 ws if (block->lf_end == -1)
532 1.1 ws fl->l_len = 0;
533 1.1 ws else
534 1.1 ws fl->l_len = block->lf_end - block->lf_start + 1;
535 1.1 ws if (block->lf_flags & F_POSIX)
536 1.17.2.1 nathanw fl->l_pid = ((struct lwp *)(block->lf_id))->l_proc->p_pid;
537 1.1 ws else
538 1.1 ws fl->l_pid = -1;
539 1.1 ws } else {
540 1.1 ws fl->l_type = F_UNLCK;
541 1.1 ws }
542 1.1 ws return (0);
543 1.1 ws }
544 1.1 ws
545 1.1 ws /*
546 1.1 ws * Walk the list of locks for an inode and
547 1.1 ws * return the first blocking lock.
548 1.1 ws */
549 1.1 ws struct lockf *
550 1.1 ws lf_getblock(lock)
551 1.15 augustss struct lockf *lock;
552 1.1 ws {
553 1.1 ws struct lockf **prev, *overlap, *lf = *(lock->lf_head);
554 1.1 ws int ovcase;
555 1.1 ws
556 1.1 ws prev = lock->lf_head;
557 1.7 christos while ((ovcase = lf_findoverlap(lf, lock, OTHERS,
558 1.7 christos &prev, &overlap)) != 0) {
559 1.1 ws /*
560 1.1 ws * We've found an overlap, see if it blocks us
561 1.1 ws */
562 1.1 ws if ((lock->lf_type == F_WRLCK || overlap->lf_type == F_WRLCK))
563 1.1 ws return (overlap);
564 1.1 ws /*
565 1.1 ws * Nope, point to the next one on the list and
566 1.1 ws * see if it blocks us
567 1.1 ws */
568 1.1 ws lf = overlap->lf_next;
569 1.1 ws }
570 1.1 ws return (NOLOCKF);
571 1.1 ws }
572 1.1 ws
573 1.1 ws /*
574 1.1 ws * Walk the list of locks for an inode to
575 1.1 ws * find an overlapping lock (if any).
576 1.1 ws *
577 1.1 ws * NOTE: this returns only the FIRST overlapping lock. There
578 1.1 ws * may be more than one.
579 1.1 ws */
580 1.4 mycroft int
581 1.1 ws lf_findoverlap(lf, lock, type, prev, overlap)
582 1.15 augustss struct lockf *lf;
583 1.1 ws struct lockf *lock;
584 1.1 ws int type;
585 1.1 ws struct lockf ***prev;
586 1.1 ws struct lockf **overlap;
587 1.1 ws {
588 1.1 ws off_t start, end;
589 1.1 ws
590 1.1 ws *overlap = lf;
591 1.1 ws if (lf == NOLOCKF)
592 1.1 ws return (0);
593 1.1 ws #ifdef LOCKF_DEBUG
594 1.1 ws if (lockf_debug & 2)
595 1.1 ws lf_print("lf_findoverlap: looking for overlap in", lock);
596 1.1 ws #endif /* LOCKF_DEBUG */
597 1.1 ws start = lock->lf_start;
598 1.1 ws end = lock->lf_end;
599 1.1 ws while (lf != NOLOCKF) {
600 1.1 ws if (((type & SELF) && lf->lf_id != lock->lf_id) ||
601 1.1 ws ((type & OTHERS) && lf->lf_id == lock->lf_id)) {
602 1.1 ws *prev = &lf->lf_next;
603 1.1 ws *overlap = lf = lf->lf_next;
604 1.1 ws continue;
605 1.1 ws }
606 1.1 ws #ifdef LOCKF_DEBUG
607 1.1 ws if (lockf_debug & 2)
608 1.1 ws lf_print("\tchecking", lf);
609 1.1 ws #endif /* LOCKF_DEBUG */
610 1.1 ws /*
611 1.1 ws * OK, check for overlap
612 1.1 ws *
613 1.1 ws * Six cases:
614 1.1 ws * 0) no overlap
615 1.1 ws * 1) overlap == lock
616 1.1 ws * 2) overlap contains lock
617 1.1 ws * 3) lock contains overlap
618 1.1 ws * 4) overlap starts before lock
619 1.1 ws * 5) overlap ends after lock
620 1.1 ws */
621 1.1 ws if ((lf->lf_end != -1 && start > lf->lf_end) ||
622 1.1 ws (end != -1 && lf->lf_start > end)) {
623 1.1 ws /* Case 0 */
624 1.1 ws #ifdef LOCKF_DEBUG
625 1.1 ws if (lockf_debug & 2)
626 1.9 christos printf("no overlap\n");
627 1.1 ws #endif /* LOCKF_DEBUG */
628 1.1 ws if ((type & SELF) && end != -1 && lf->lf_start > end)
629 1.1 ws return (0);
630 1.1 ws *prev = &lf->lf_next;
631 1.1 ws *overlap = lf = lf->lf_next;
632 1.1 ws continue;
633 1.1 ws }
634 1.1 ws if ((lf->lf_start == start) && (lf->lf_end == end)) {
635 1.1 ws /* Case 1 */
636 1.1 ws #ifdef LOCKF_DEBUG
637 1.1 ws if (lockf_debug & 2)
638 1.9 christos printf("overlap == lock\n");
639 1.1 ws #endif /* LOCKF_DEBUG */
640 1.1 ws return (1);
641 1.1 ws }
642 1.1 ws if ((lf->lf_start <= start) &&
643 1.1 ws (end != -1) &&
644 1.1 ws ((lf->lf_end >= end) || (lf->lf_end == -1))) {
645 1.1 ws /* Case 2 */
646 1.1 ws #ifdef LOCKF_DEBUG
647 1.1 ws if (lockf_debug & 2)
648 1.9 christos printf("overlap contains lock\n");
649 1.1 ws #endif /* LOCKF_DEBUG */
650 1.1 ws return (2);
651 1.1 ws }
652 1.1 ws if (start <= lf->lf_start &&
653 1.4 mycroft (end == -1 ||
654 1.1 ws (lf->lf_end != -1 && end >= lf->lf_end))) {
655 1.1 ws /* Case 3 */
656 1.1 ws #ifdef LOCKF_DEBUG
657 1.1 ws if (lockf_debug & 2)
658 1.9 christos printf("lock contains overlap\n");
659 1.1 ws #endif /* LOCKF_DEBUG */
660 1.1 ws return (3);
661 1.1 ws }
662 1.1 ws if ((lf->lf_start < start) &&
663 1.1 ws ((lf->lf_end >= start) || (lf->lf_end == -1))) {
664 1.1 ws /* Case 4 */
665 1.1 ws #ifdef LOCKF_DEBUG
666 1.1 ws if (lockf_debug & 2)
667 1.9 christos printf("overlap starts before lock\n");
668 1.1 ws #endif /* LOCKF_DEBUG */
669 1.1 ws return (4);
670 1.1 ws }
671 1.1 ws if ((lf->lf_start > start) &&
672 1.1 ws (end != -1) &&
673 1.1 ws ((lf->lf_end > end) || (lf->lf_end == -1))) {
674 1.1 ws /* Case 5 */
675 1.1 ws #ifdef LOCKF_DEBUG
676 1.1 ws if (lockf_debug & 2)
677 1.9 christos printf("overlap ends after lock\n");
678 1.1 ws #endif /* LOCKF_DEBUG */
679 1.1 ws return (5);
680 1.1 ws }
681 1.1 ws panic("lf_findoverlap: default");
682 1.1 ws }
683 1.1 ws return (0);
684 1.1 ws }
685 1.1 ws
686 1.1 ws /*
687 1.1 ws * Split a lock and a contained region into
688 1.1 ws * two or three locks as necessary.
689 1.1 ws */
690 1.2 cgd void
691 1.1 ws lf_split(lock1, lock2)
692 1.15 augustss struct lockf *lock1;
693 1.15 augustss struct lockf *lock2;
694 1.1 ws {
695 1.15 augustss struct lockf *splitlock;
696 1.1 ws
697 1.1 ws #ifdef LOCKF_DEBUG
698 1.1 ws if (lockf_debug & 2) {
699 1.1 ws lf_print("lf_split", lock1);
700 1.1 ws lf_print("splitting from", lock2);
701 1.1 ws }
702 1.1 ws #endif /* LOCKF_DEBUG */
703 1.1 ws /*
704 1.1 ws * Check to see if spliting into only two pieces.
705 1.1 ws */
706 1.1 ws if (lock1->lf_start == lock2->lf_start) {
707 1.1 ws lock1->lf_start = lock2->lf_end + 1;
708 1.1 ws lock2->lf_next = lock1;
709 1.1 ws return;
710 1.1 ws }
711 1.1 ws if (lock1->lf_end == lock2->lf_end) {
712 1.1 ws lock1->lf_end = lock2->lf_start - 1;
713 1.1 ws lock2->lf_next = lock1->lf_next;
714 1.1 ws lock1->lf_next = lock2;
715 1.1 ws return;
716 1.1 ws }
717 1.1 ws /*
718 1.1 ws * Make a new lock consisting of the last part of
719 1.1 ws * the encompassing lock
720 1.1 ws */
721 1.13 perry MALLOC(splitlock, struct lockf *, sizeof(*splitlock), M_LOCKF, M_WAITOK);
722 1.14 perry memcpy((caddr_t)splitlock, (caddr_t)lock1, sizeof(*splitlock));
723 1.1 ws splitlock->lf_start = lock2->lf_end + 1;
724 1.12 fvdl TAILQ_INIT(&splitlock->lf_blkhd);
725 1.1 ws lock1->lf_end = lock2->lf_start - 1;
726 1.1 ws /*
727 1.1 ws * OK, now link it in
728 1.1 ws */
729 1.1 ws splitlock->lf_next = lock1->lf_next;
730 1.1 ws lock2->lf_next = splitlock;
731 1.1 ws lock1->lf_next = lock2;
732 1.1 ws }
733 1.1 ws
734 1.1 ws /*
735 1.1 ws * Wakeup a blocklist
736 1.1 ws */
737 1.2 cgd void
738 1.1 ws lf_wakelock(listhead)
739 1.1 ws struct lockf *listhead;
740 1.1 ws {
741 1.15 augustss struct lockf *wakelock;
742 1.1 ws
743 1.12 fvdl while ((wakelock = listhead->lf_blkhd.tqh_first)) {
744 1.16 sommerfe KASSERT(wakelock->lf_next == listhead);
745 1.12 fvdl TAILQ_REMOVE(&listhead->lf_blkhd, wakelock, lf_block);
746 1.1 ws wakelock->lf_next = NOLOCKF;
747 1.1 ws #ifdef LOCKF_DEBUG
748 1.1 ws if (lockf_debug & 2)
749 1.1 ws lf_print("lf_wakelock: awakening", wakelock);
750 1.12 fvdl #endif
751 1.1 ws wakeup((caddr_t)wakelock);
752 1.1 ws }
753 1.1 ws }
754 1.1 ws
755 1.1 ws #ifdef LOCKF_DEBUG
756 1.1 ws /*
757 1.1 ws * Print out a lock.
758 1.1 ws */
759 1.4 mycroft void
760 1.1 ws lf_print(tag, lock)
761 1.1 ws char *tag;
762 1.15 augustss struct lockf *lock;
763 1.1 ws {
764 1.1 ws
765 1.9 christos printf("%s: lock %p for ", tag, lock);
766 1.1 ws if (lock->lf_flags & F_POSIX)
767 1.9 christos printf("proc %d", ((struct proc *)(lock->lf_id))->p_pid);
768 1.1 ws else
769 1.11 jtk printf("id 0x%p", lock->lf_id);
770 1.11 jtk printf(" %s, start %qx, end %qx",
771 1.1 ws lock->lf_type == F_RDLCK ? "shared" :
772 1.1 ws lock->lf_type == F_WRLCK ? "exclusive" :
773 1.1 ws lock->lf_type == F_UNLCK ? "unlock" :
774 1.1 ws "unknown", lock->lf_start, lock->lf_end);
775 1.12 fvdl if (lock->lf_blkhd.tqh_first)
776 1.12 fvdl printf(" block %p\n", lock->lf_blkhd.tqh_first);
777 1.1 ws else
778 1.9 christos printf("\n");
779 1.1 ws }
780 1.1 ws
781 1.4 mycroft void
782 1.1 ws lf_printlist(tag, lock)
783 1.1 ws char *tag;
784 1.1 ws struct lockf *lock;
785 1.1 ws {
786 1.15 augustss struct lockf *lf, *blk;
787 1.1 ws
788 1.11 jtk printf("%s: Lock list:\n", tag);
789 1.12 fvdl for (lf = *lock->lf_head; lf; lf = lf->lf_next) {
790 1.9 christos printf("\tlock %p for ", lf);
791 1.1 ws if (lf->lf_flags & F_POSIX)
792 1.9 christos printf("proc %d", ((struct proc *)(lf->lf_id))->p_pid);
793 1.1 ws else
794 1.11 jtk printf("id 0x%p", lf->lf_id);
795 1.11 jtk printf(", %s, start %qx, end %qx",
796 1.1 ws lf->lf_type == F_RDLCK ? "shared" :
797 1.1 ws lf->lf_type == F_WRLCK ? "exclusive" :
798 1.1 ws lf->lf_type == F_UNLCK ? "unlock" :
799 1.1 ws "unknown", lf->lf_start, lf->lf_end);
800 1.12 fvdl for (blk = lf->lf_blkhd.tqh_first; blk;
801 1.12 fvdl blk = blk->lf_block.tqe_next) {
802 1.12 fvdl if (blk->lf_flags & F_POSIX)
803 1.12 fvdl printf("proc %d",
804 1.12 fvdl ((struct proc *)(blk->lf_id))->p_pid);
805 1.12 fvdl else
806 1.12 fvdl printf("id 0x%p", blk->lf_id);
807 1.12 fvdl printf(", %s, start %qx, end %qx",
808 1.12 fvdl blk->lf_type == F_RDLCK ? "shared" :
809 1.12 fvdl blk->lf_type == F_WRLCK ? "exclusive" :
810 1.12 fvdl blk->lf_type == F_UNLCK ? "unlock" :
811 1.12 fvdl "unknown", blk->lf_start, blk->lf_end);
812 1.12 fvdl if (blk->lf_blkhd.tqh_first)
813 1.12 fvdl panic("lf_printlist: bad list");
814 1.12 fvdl }
815 1.12 fvdl printf("\n");
816 1.1 ws }
817 1.1 ws }
818 1.1 ws #endif /* LOCKF_DEBUG */
819