xref: /freebsd/sys/geom/cache/g_cache.c (revision 4d846d26)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2006 Ruslan Ermilov <ru@FreeBSD.org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/kernel.h>
35 #include <sys/module.h>
36 #include <sys/lock.h>
37 #include <sys/mutex.h>
38 #include <sys/bio.h>
39 #include <sys/sysctl.h>
40 #include <sys/malloc.h>
41 #include <sys/queue.h>
42 #include <sys/sbuf.h>
43 #include <sys/time.h>
44 #include <vm/uma.h>
45 #include <geom/geom.h>
46 #include <geom/geom_dbg.h>
47 #include <geom/cache/g_cache.h>
48 
49 FEATURE(geom_cache, "GEOM cache module");
50 
51 static MALLOC_DEFINE(M_GCACHE, "gcache_data", "GEOM_CACHE Data");
52 
53 SYSCTL_DECL(_kern_geom);
54 static SYSCTL_NODE(_kern_geom, OID_AUTO, cache, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
55     "GEOM_CACHE stuff");
56 static u_int g_cache_debug = 0;
57 SYSCTL_UINT(_kern_geom_cache, OID_AUTO, debug, CTLFLAG_RW, &g_cache_debug, 0,
58     "Debug level");
59 static u_int g_cache_enable = 1;
60 SYSCTL_UINT(_kern_geom_cache, OID_AUTO, enable, CTLFLAG_RW, &g_cache_enable, 0,
61     "");
62 static u_int g_cache_timeout = 10;
63 SYSCTL_UINT(_kern_geom_cache, OID_AUTO, timeout, CTLFLAG_RW, &g_cache_timeout,
64     0, "");
65 static u_int g_cache_idletime = 5;
66 SYSCTL_UINT(_kern_geom_cache, OID_AUTO, idletime, CTLFLAG_RW, &g_cache_idletime,
67     0, "");
68 static u_int g_cache_used_lo = 5;
69 static u_int g_cache_used_hi = 20;
70 static int
71 sysctl_handle_pct(SYSCTL_HANDLER_ARGS)
72 {
73 	u_int val = *(u_int *)arg1;
74 	int error;
75 
76 	error = sysctl_handle_int(oidp, &val, 0, req);
77 	if (error || !req->newptr)
78 		return (error);
79 	if (val > 100)
80 		return (EINVAL);
81 	if ((arg1 == &g_cache_used_lo && val > g_cache_used_hi) ||
82 	    (arg1 == &g_cache_used_hi && g_cache_used_lo > val))
83 		return (EINVAL);
84 	*(u_int *)arg1 = val;
85 	return (0);
86 }
87 SYSCTL_PROC(_kern_geom_cache, OID_AUTO, used_lo,
88     CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE, &g_cache_used_lo, 0,
89     sysctl_handle_pct, "IU",
90     "");
91 SYSCTL_PROC(_kern_geom_cache, OID_AUTO, used_hi,
92     CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE, &g_cache_used_hi, 0,
93     sysctl_handle_pct, "IU",
94     "");
95 
96 static int g_cache_destroy(struct g_cache_softc *sc, boolean_t force);
97 static g_ctl_destroy_geom_t g_cache_destroy_geom;
98 
99 static g_taste_t g_cache_taste;
100 static g_ctl_req_t g_cache_config;
101 static g_dumpconf_t g_cache_dumpconf;
102 
103 struct g_class g_cache_class = {
104 	.name = G_CACHE_CLASS_NAME,
105 	.version = G_VERSION,
106 	.ctlreq = g_cache_config,
107 	.taste = g_cache_taste,
108 	.destroy_geom = g_cache_destroy_geom
109 };
110 
111 #define	OFF2BNO(off, sc)	((off) >> (sc)->sc_bshift)
112 #define	BNO2OFF(bno, sc)	((bno) << (sc)->sc_bshift)
113 
114 static struct g_cache_desc *
115 g_cache_alloc(struct g_cache_softc *sc)
116 {
117 	struct g_cache_desc *dp;
118 
119 	mtx_assert(&sc->sc_mtx, MA_OWNED);
120 
121 	if (!TAILQ_EMPTY(&sc->sc_usedlist)) {
122 		dp = TAILQ_FIRST(&sc->sc_usedlist);
123 		TAILQ_REMOVE(&sc->sc_usedlist, dp, d_used);
124 		sc->sc_nused--;
125 		dp->d_flags = 0;
126 		LIST_REMOVE(dp, d_next);
127 		return (dp);
128 	}
129 	if (sc->sc_nent > sc->sc_maxent) {
130 		sc->sc_cachefull++;
131 		return (NULL);
132 	}
133 	dp = malloc(sizeof(*dp), M_GCACHE, M_NOWAIT | M_ZERO);
134 	if (dp == NULL)
135 		return (NULL);
136 	dp->d_data = uma_zalloc(sc->sc_zone, M_NOWAIT);
137 	if (dp->d_data == NULL) {
138 		free(dp, M_GCACHE);
139 		return (NULL);
140 	}
141 	sc->sc_nent++;
142 	return (dp);
143 }
144 
145 static void
146 g_cache_free(struct g_cache_softc *sc, struct g_cache_desc *dp)
147 {
148 
149 	mtx_assert(&sc->sc_mtx, MA_OWNED);
150 
151 	uma_zfree(sc->sc_zone, dp->d_data);
152 	free(dp, M_GCACHE);
153 	sc->sc_nent--;
154 }
155 
156 static void
157 g_cache_free_used(struct g_cache_softc *sc)
158 {
159 	struct g_cache_desc *dp;
160 	u_int n;
161 
162 	mtx_assert(&sc->sc_mtx, MA_OWNED);
163 
164 	n = g_cache_used_lo * sc->sc_maxent / 100;
165 	while (sc->sc_nused > n) {
166 		KASSERT(!TAILQ_EMPTY(&sc->sc_usedlist), ("used list empty"));
167 		dp = TAILQ_FIRST(&sc->sc_usedlist);
168 		TAILQ_REMOVE(&sc->sc_usedlist, dp, d_used);
169 		sc->sc_nused--;
170 		LIST_REMOVE(dp, d_next);
171 		g_cache_free(sc, dp);
172 	}
173 }
174 
175 static void
176 g_cache_deliver(struct g_cache_softc *sc, struct bio *bp,
177     struct g_cache_desc *dp, int error)
178 {
179 	off_t off1, off, len;
180 
181 	mtx_assert(&sc->sc_mtx, MA_OWNED);
182 	KASSERT(OFF2BNO(bp->bio_offset, sc) <= dp->d_bno, ("wrong entry"));
183 	KASSERT(OFF2BNO(bp->bio_offset + bp->bio_length - 1, sc) >=
184 	    dp->d_bno, ("wrong entry"));
185 
186 	off1 = BNO2OFF(dp->d_bno, sc);
187 	off = MAX(bp->bio_offset, off1);
188 	len = MIN(bp->bio_offset + bp->bio_length, off1 + sc->sc_bsize) - off;
189 
190 	if (bp->bio_error == 0)
191 		bp->bio_error = error;
192 	if (bp->bio_error == 0) {
193 		bcopy(dp->d_data + (off - off1),
194 		    bp->bio_data + (off - bp->bio_offset), len);
195 	}
196 	bp->bio_completed += len;
197 	KASSERT(bp->bio_completed <= bp->bio_length, ("extra data"));
198 	if (bp->bio_completed == bp->bio_length) {
199 		if (bp->bio_error != 0)
200 			bp->bio_completed = 0;
201 		g_io_deliver(bp, bp->bio_error);
202 	}
203 
204 	if (dp->d_flags & D_FLAG_USED) {
205 		TAILQ_REMOVE(&sc->sc_usedlist, dp, d_used);
206 		TAILQ_INSERT_TAIL(&sc->sc_usedlist, dp, d_used);
207 	} else if (OFF2BNO(off + len, sc) > dp->d_bno) {
208 		TAILQ_INSERT_TAIL(&sc->sc_usedlist, dp, d_used);
209 		sc->sc_nused++;
210 		dp->d_flags |= D_FLAG_USED;
211 	}
212 	dp->d_atime = time_uptime;
213 }
214 
215 static void
216 g_cache_done(struct bio *bp)
217 {
218 	struct g_cache_softc *sc;
219 	struct g_cache_desc *dp;
220 	struct bio *bp2, *tmpbp;
221 
222 	sc = bp->bio_from->geom->softc;
223 	KASSERT(G_CACHE_DESC1(bp) == sc, ("corrupt bio_caller in g_cache_done()"));
224 	dp = G_CACHE_DESC2(bp);
225 	mtx_lock(&sc->sc_mtx);
226 	bp2 = dp->d_biolist;
227 	while (bp2 != NULL) {
228 		KASSERT(G_CACHE_NEXT_BIO1(bp2) == sc, ("corrupt bio_driver in g_cache_done()"));
229 		tmpbp = G_CACHE_NEXT_BIO2(bp2);
230 		g_cache_deliver(sc, bp2, dp, bp->bio_error);
231 		bp2 = tmpbp;
232 	}
233 	dp->d_biolist = NULL;
234 	if (dp->d_flags & D_FLAG_INVALID) {
235 		sc->sc_invalid--;
236 		g_cache_free(sc, dp);
237 	} else if (bp->bio_error) {
238 		LIST_REMOVE(dp, d_next);
239 		if (dp->d_flags & D_FLAG_USED) {
240 			TAILQ_REMOVE(&sc->sc_usedlist, dp, d_used);
241 			sc->sc_nused--;
242 		}
243 		g_cache_free(sc, dp);
244 	}
245 	mtx_unlock(&sc->sc_mtx);
246 	g_destroy_bio(bp);
247 }
248 
249 static struct g_cache_desc *
250 g_cache_lookup(struct g_cache_softc *sc, off_t bno)
251 {
252 	struct g_cache_desc *dp;
253 
254 	mtx_assert(&sc->sc_mtx, MA_OWNED);
255 
256 	LIST_FOREACH(dp, &sc->sc_desclist[G_CACHE_BUCKET(bno)], d_next)
257 		if (dp->d_bno == bno)
258 			return (dp);
259 	return (NULL);
260 }
261 
262 static int
263 g_cache_read(struct g_cache_softc *sc, struct bio *bp)
264 {
265 	struct bio *cbp;
266 	struct g_cache_desc *dp;
267 
268 	mtx_lock(&sc->sc_mtx);
269 	dp = g_cache_lookup(sc,
270 	    OFF2BNO(bp->bio_offset + bp->bio_completed, sc));
271 	if (dp != NULL) {
272 		/* Add to waiters list or deliver. */
273 		sc->sc_cachehits++;
274 		if (dp->d_biolist != NULL) {
275 			G_CACHE_NEXT_BIO1(bp) = sc;
276 			G_CACHE_NEXT_BIO2(bp) = dp->d_biolist;
277 			dp->d_biolist = bp;
278 		} else
279 			g_cache_deliver(sc, bp, dp, 0);
280 		mtx_unlock(&sc->sc_mtx);
281 		return (0);
282 	}
283 
284 	/* Cache miss.  Allocate entry and schedule bio.  */
285 	sc->sc_cachemisses++;
286 	dp = g_cache_alloc(sc);
287 	if (dp == NULL) {
288 		mtx_unlock(&sc->sc_mtx);
289 		return (ENOMEM);
290 	}
291 	cbp = g_clone_bio(bp);
292 	if (cbp == NULL) {
293 		g_cache_free(sc, dp);
294 		mtx_unlock(&sc->sc_mtx);
295 		return (ENOMEM);
296 	}
297 
298 	dp->d_bno = OFF2BNO(bp->bio_offset + bp->bio_completed, sc);
299 	G_CACHE_NEXT_BIO1(bp) = sc;
300 	G_CACHE_NEXT_BIO2(bp) = NULL;
301 	dp->d_biolist = bp;
302 	LIST_INSERT_HEAD(&sc->sc_desclist[G_CACHE_BUCKET(dp->d_bno)],
303 	    dp, d_next);
304 	mtx_unlock(&sc->sc_mtx);
305 
306 	G_CACHE_DESC1(cbp) = sc;
307 	G_CACHE_DESC2(cbp) = dp;
308 	cbp->bio_done = g_cache_done;
309 	cbp->bio_offset = BNO2OFF(dp->d_bno, sc);
310 	cbp->bio_data = dp->d_data;
311 	cbp->bio_length = sc->sc_bsize;
312 	g_io_request(cbp, LIST_FIRST(&bp->bio_to->geom->consumer));
313 	return (0);
314 }
315 
316 static void
317 g_cache_invalidate(struct g_cache_softc *sc, struct bio *bp)
318 {
319 	struct g_cache_desc *dp;
320 	off_t bno, lim;
321 
322 	mtx_lock(&sc->sc_mtx);
323 	bno = OFF2BNO(bp->bio_offset, sc);
324 	lim = OFF2BNO(bp->bio_offset + bp->bio_length - 1, sc);
325 	do {
326 		if ((dp = g_cache_lookup(sc, bno)) != NULL) {
327 			LIST_REMOVE(dp, d_next);
328 			if (dp->d_flags & D_FLAG_USED) {
329 				TAILQ_REMOVE(&sc->sc_usedlist, dp, d_used);
330 				sc->sc_nused--;
331 			}
332 			if (dp->d_biolist == NULL)
333 				g_cache_free(sc, dp);
334 			else {
335 				dp->d_flags = D_FLAG_INVALID;
336 				sc->sc_invalid++;
337 			}
338 		}
339 		bno++;
340 	} while (bno <= lim);
341 	mtx_unlock(&sc->sc_mtx);
342 }
343 
344 static void
345 g_cache_start(struct bio *bp)
346 {
347 	struct g_cache_softc *sc;
348 	struct g_geom *gp;
349 	struct g_cache_desc *dp;
350 	struct bio *cbp;
351 
352 	gp = bp->bio_to->geom;
353 	sc = gp->softc;
354 	G_CACHE_LOGREQ(bp, "Request received.");
355 	switch (bp->bio_cmd) {
356 	case BIO_READ:
357 		sc->sc_reads++;
358 		sc->sc_readbytes += bp->bio_length;
359 		if (!g_cache_enable)
360 			break;
361 		if (bp->bio_offset + bp->bio_length > sc->sc_tail)
362 			break;
363 		if (OFF2BNO(bp->bio_offset, sc) ==
364 		    OFF2BNO(bp->bio_offset + bp->bio_length - 1, sc)) {
365 			sc->sc_cachereads++;
366 			sc->sc_cachereadbytes += bp->bio_length;
367 			if (g_cache_read(sc, bp) == 0)
368 				return;
369 			sc->sc_cachereads--;
370 			sc->sc_cachereadbytes -= bp->bio_length;
371 			break;
372 		} else if (OFF2BNO(bp->bio_offset, sc) + 1 ==
373 		    OFF2BNO(bp->bio_offset + bp->bio_length - 1, sc)) {
374 			mtx_lock(&sc->sc_mtx);
375 			dp = g_cache_lookup(sc, OFF2BNO(bp->bio_offset, sc));
376 			if (dp == NULL || dp->d_biolist != NULL) {
377 				mtx_unlock(&sc->sc_mtx);
378 				break;
379 			}
380 			sc->sc_cachereads++;
381 			sc->sc_cachereadbytes += bp->bio_length;
382 			g_cache_deliver(sc, bp, dp, 0);
383 			mtx_unlock(&sc->sc_mtx);
384 			if (g_cache_read(sc, bp) == 0)
385 				return;
386 			sc->sc_cachereads--;
387 			sc->sc_cachereadbytes -= bp->bio_length;
388 			break;
389 		}
390 		break;
391 	case BIO_WRITE:
392 		sc->sc_writes++;
393 		sc->sc_wrotebytes += bp->bio_length;
394 		g_cache_invalidate(sc, bp);
395 		break;
396 	}
397 	cbp = g_clone_bio(bp);
398 	if (cbp == NULL) {
399 		g_io_deliver(bp, ENOMEM);
400 		return;
401 	}
402 	cbp->bio_done = g_std_done;
403 	G_CACHE_LOGREQ(cbp, "Sending request.");
404 	g_io_request(cbp, LIST_FIRST(&gp->consumer));
405 }
406 
407 static void
408 g_cache_go(void *arg)
409 {
410 	struct g_cache_softc *sc = arg;
411 	struct g_cache_desc *dp;
412 	int i;
413 
414 	mtx_assert(&sc->sc_mtx, MA_OWNED);
415 
416 	/* Forcibly mark idle ready entries as used. */
417 	for (i = 0; i < G_CACHE_BUCKETS; i++) {
418 		LIST_FOREACH(dp, &sc->sc_desclist[i], d_next) {
419 			if (dp->d_flags & D_FLAG_USED ||
420 			    dp->d_biolist != NULL ||
421 			    time_uptime - dp->d_atime < g_cache_idletime)
422 				continue;
423 			TAILQ_INSERT_TAIL(&sc->sc_usedlist, dp, d_used);
424 			sc->sc_nused++;
425 			dp->d_flags |= D_FLAG_USED;
426 		}
427 	}
428 
429 	/* Keep the number of used entries low. */
430 	if (sc->sc_nused > g_cache_used_hi * sc->sc_maxent / 100)
431 		g_cache_free_used(sc);
432 
433 	callout_reset(&sc->sc_callout, g_cache_timeout * hz, g_cache_go, sc);
434 }
435 
436 static int
437 g_cache_access(struct g_provider *pp, int dr, int dw, int de)
438 {
439 	struct g_geom *gp;
440 	struct g_consumer *cp;
441 	int error;
442 
443 	gp = pp->geom;
444 	cp = LIST_FIRST(&gp->consumer);
445 	error = g_access(cp, dr, dw, de);
446 
447 	return (error);
448 }
449 
450 static void
451 g_cache_orphan(struct g_consumer *cp)
452 {
453 
454 	g_topology_assert();
455 	g_cache_destroy(cp->geom->softc, 1);
456 }
457 
458 static struct g_cache_softc *
459 g_cache_find_device(struct g_class *mp, const char *name)
460 {
461 	struct g_geom *gp;
462 
463 	LIST_FOREACH(gp, &mp->geom, geom) {
464 		if (strcmp(gp->name, name) == 0)
465 			return (gp->softc);
466 	}
467 	return (NULL);
468 }
469 
470 static struct g_geom *
471 g_cache_create(struct g_class *mp, struct g_provider *pp,
472     const struct g_cache_metadata *md, u_int type)
473 {
474 	struct g_cache_softc *sc;
475 	struct g_geom *gp;
476 	struct g_provider *newpp;
477 	struct g_consumer *cp;
478 	u_int bshift;
479 	int i;
480 
481 	g_topology_assert();
482 
483 	gp = NULL;
484 	newpp = NULL;
485 	cp = NULL;
486 
487 	G_CACHE_DEBUG(1, "Creating device %s.", md->md_name);
488 
489 	/* Cache size is minimum 100. */
490 	if (md->md_size < 100) {
491 		G_CACHE_DEBUG(0, "Invalid size for device %s.", md->md_name);
492 		return (NULL);
493 	}
494 
495 	/* Block size restrictions. */
496 	bshift = ffs(md->md_bsize) - 1;
497 	if (md->md_bsize == 0 || md->md_bsize > maxphys ||
498 	    md->md_bsize != 1 << bshift ||
499 	    (md->md_bsize % pp->sectorsize) != 0) {
500 		G_CACHE_DEBUG(0, "Invalid blocksize for provider %s.", pp->name);
501 		return (NULL);
502 	}
503 
504 	/* Check for duplicate unit. */
505 	if (g_cache_find_device(mp, (const char *)&md->md_name) != NULL) {
506 		G_CACHE_DEBUG(0, "Provider %s already exists.", md->md_name);
507 		return (NULL);
508 	}
509 
510 	gp = g_new_geomf(mp, "%s", md->md_name);
511 	sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO);
512 	sc->sc_type = type;
513 	sc->sc_bshift = bshift;
514 	sc->sc_bsize = 1 << bshift;
515 	sc->sc_zone = uma_zcreate("gcache", sc->sc_bsize, NULL, NULL, NULL, NULL,
516 	    UMA_ALIGN_PTR, 0);
517 	mtx_init(&sc->sc_mtx, "GEOM CACHE mutex", NULL, MTX_DEF);
518 	for (i = 0; i < G_CACHE_BUCKETS; i++)
519 		LIST_INIT(&sc->sc_desclist[i]);
520 	TAILQ_INIT(&sc->sc_usedlist);
521 	sc->sc_maxent = md->md_size;
522 	callout_init_mtx(&sc->sc_callout, &sc->sc_mtx, 0);
523 	gp->softc = sc;
524 	sc->sc_geom = gp;
525 	gp->start = g_cache_start;
526 	gp->orphan = g_cache_orphan;
527 	gp->access = g_cache_access;
528 	gp->dumpconf = g_cache_dumpconf;
529 
530 	newpp = g_new_providerf(gp, "cache/%s", gp->name);
531 	newpp->sectorsize = pp->sectorsize;
532 	newpp->mediasize = pp->mediasize;
533 	if (type == G_CACHE_TYPE_AUTOMATIC)
534 		newpp->mediasize -= pp->sectorsize;
535 	sc->sc_tail = BNO2OFF(OFF2BNO(newpp->mediasize, sc), sc);
536 
537 	cp = g_new_consumer(gp);
538 	if (g_attach(cp, pp) != 0) {
539 		G_CACHE_DEBUG(0, "Cannot attach to provider %s.", pp->name);
540 		g_destroy_consumer(cp);
541 		g_destroy_provider(newpp);
542 		mtx_destroy(&sc->sc_mtx);
543 		g_free(sc);
544 		g_destroy_geom(gp);
545 		return (NULL);
546 	}
547 
548 	g_error_provider(newpp, 0);
549 	G_CACHE_DEBUG(0, "Device %s created.", gp->name);
550 	callout_reset(&sc->sc_callout, g_cache_timeout * hz, g_cache_go, sc);
551 	return (gp);
552 }
553 
554 static int
555 g_cache_destroy(struct g_cache_softc *sc, boolean_t force)
556 {
557 	struct g_geom *gp;
558 	struct g_provider *pp;
559 	struct g_cache_desc *dp, *dp2;
560 	int i;
561 
562 	g_topology_assert();
563 	if (sc == NULL)
564 		return (ENXIO);
565 	gp = sc->sc_geom;
566 	pp = LIST_FIRST(&gp->provider);
567 	if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) {
568 		if (force) {
569 			G_CACHE_DEBUG(0, "Device %s is still open, so it "
570 			    "can't be definitely removed.", pp->name);
571 		} else {
572 			G_CACHE_DEBUG(1, "Device %s is still open (r%dw%de%d).",
573 			    pp->name, pp->acr, pp->acw, pp->ace);
574 			return (EBUSY);
575 		}
576 	} else {
577 		G_CACHE_DEBUG(0, "Device %s removed.", gp->name);
578 	}
579 	callout_drain(&sc->sc_callout);
580 	mtx_lock(&sc->sc_mtx);
581 	for (i = 0; i < G_CACHE_BUCKETS; i++) {
582 		dp = LIST_FIRST(&sc->sc_desclist[i]);
583 		while (dp != NULL) {
584 			dp2 = LIST_NEXT(dp, d_next);
585 			g_cache_free(sc, dp);
586 			dp = dp2;
587 		}
588 	}
589 	mtx_unlock(&sc->sc_mtx);
590 	mtx_destroy(&sc->sc_mtx);
591 	uma_zdestroy(sc->sc_zone);
592 	g_free(sc);
593 	gp->softc = NULL;
594 	g_wither_geom(gp, ENXIO);
595 
596 	return (0);
597 }
598 
599 static int
600 g_cache_destroy_geom(struct gctl_req *req, struct g_class *mp, struct g_geom *gp)
601 {
602 
603 	return (g_cache_destroy(gp->softc, 0));
604 }
605 
606 static int
607 g_cache_read_metadata(struct g_consumer *cp, struct g_cache_metadata *md)
608 {
609 	struct g_provider *pp;
610 	u_char *buf;
611 	int error;
612 
613 	g_topology_assert();
614 
615 	error = g_access(cp, 1, 0, 0);
616 	if (error != 0)
617 		return (error);
618 	pp = cp->provider;
619 	g_topology_unlock();
620 	buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize,
621 	    &error);
622 	g_topology_lock();
623 	g_access(cp, -1, 0, 0);
624 	if (buf == NULL)
625 		return (error);
626 
627 	/* Decode metadata. */
628 	cache_metadata_decode(buf, md);
629 	g_free(buf);
630 
631 	return (0);
632 }
633 
634 static int
635 g_cache_write_metadata(struct g_consumer *cp, struct g_cache_metadata *md)
636 {
637 	struct g_provider *pp;
638 	u_char *buf;
639 	int error;
640 
641 	g_topology_assert();
642 
643 	error = g_access(cp, 0, 1, 0);
644 	if (error != 0)
645 		return (error);
646 	pp = cp->provider;
647 	buf = malloc((size_t)pp->sectorsize, M_GCACHE, M_WAITOK | M_ZERO);
648 	cache_metadata_encode(md, buf);
649 	g_topology_unlock();
650 	error = g_write_data(cp, pp->mediasize - pp->sectorsize, buf, pp->sectorsize);
651 	g_topology_lock();
652 	g_access(cp, 0, -1, 0);
653 	free(buf, M_GCACHE);
654 
655 	return (error);
656 }
657 
658 static struct g_geom *
659 g_cache_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
660 {
661 	struct g_cache_metadata md;
662 	struct g_consumer *cp;
663 	struct g_geom *gp;
664 	int error;
665 
666 	g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name);
667 	g_topology_assert();
668 
669 	G_CACHE_DEBUG(3, "Tasting %s.", pp->name);
670 
671 	gp = g_new_geomf(mp, "cache:taste");
672 	gp->start = g_cache_start;
673 	gp->orphan = g_cache_orphan;
674 	gp->access = g_cache_access;
675 	cp = g_new_consumer(gp);
676 	cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
677 	error = g_attach(cp, pp);
678 	if (error == 0) {
679 		error = g_cache_read_metadata(cp, &md);
680 		g_detach(cp);
681 	}
682 	g_destroy_consumer(cp);
683 	g_destroy_geom(gp);
684 	if (error != 0)
685 		return (NULL);
686 
687 	if (strcmp(md.md_magic, G_CACHE_MAGIC) != 0)
688 		return (NULL);
689 	if (md.md_version > G_CACHE_VERSION) {
690 		printf("geom_cache.ko module is too old to handle %s.\n",
691 		    pp->name);
692 		return (NULL);
693 	}
694 	if (md.md_provsize != pp->mediasize)
695 		return (NULL);
696 
697 	gp = g_cache_create(mp, pp, &md, G_CACHE_TYPE_AUTOMATIC);
698 	if (gp == NULL) {
699 		G_CACHE_DEBUG(0, "Can't create %s.", md.md_name);
700 		return (NULL);
701 	}
702 	return (gp);
703 }
704 
705 static void
706 g_cache_ctl_create(struct gctl_req *req, struct g_class *mp)
707 {
708 	struct g_cache_metadata md;
709 	struct g_provider *pp;
710 	struct g_geom *gp;
711 	intmax_t *bsize, *size;
712 	const char *name;
713 	int *nargs;
714 
715 	g_topology_assert();
716 
717 	nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
718 	if (nargs == NULL) {
719 		gctl_error(req, "No '%s' argument", "nargs");
720 		return;
721 	}
722 	if (*nargs != 2) {
723 		gctl_error(req, "Invalid number of arguments.");
724 		return;
725 	}
726 
727 	strlcpy(md.md_magic, G_CACHE_MAGIC, sizeof(md.md_magic));
728 	md.md_version = G_CACHE_VERSION;
729 	name = gctl_get_asciiparam(req, "arg0");
730 	if (name == NULL) {
731 		gctl_error(req, "No 'arg0' argument");
732 		return;
733 	}
734 	strlcpy(md.md_name, name, sizeof(md.md_name));
735 
736 	size = gctl_get_paraml(req, "size", sizeof(*size));
737 	if (size == NULL) {
738 		gctl_error(req, "No '%s' argument", "size");
739 		return;
740 	}
741 	if ((u_int)*size < 100) {
742 		gctl_error(req, "Invalid '%s' argument", "size");
743 		return;
744 	}
745 	md.md_size = (u_int)*size;
746 
747 	bsize = gctl_get_paraml(req, "blocksize", sizeof(*bsize));
748 	if (bsize == NULL) {
749 		gctl_error(req, "No '%s' argument", "blocksize");
750 		return;
751 	}
752 	if (*bsize < 0) {
753 		gctl_error(req, "Invalid '%s' argument", "blocksize");
754 		return;
755 	}
756 	md.md_bsize = (u_int)*bsize;
757 
758 	/* This field is not important here. */
759 	md.md_provsize = 0;
760 
761 	pp = gctl_get_provider(req, "arg1");
762 	if (pp == NULL)
763 		return;
764 	gp = g_cache_create(mp, pp, &md, G_CACHE_TYPE_MANUAL);
765 	if (gp == NULL) {
766 		gctl_error(req, "Can't create %s.", md.md_name);
767 		return;
768 	}
769 }
770 
771 static void
772 g_cache_ctl_configure(struct gctl_req *req, struct g_class *mp)
773 {
774 	struct g_cache_metadata md;
775 	struct g_cache_softc *sc;
776 	struct g_consumer *cp;
777 	intmax_t *bsize, *size;
778 	const char *name;
779 	int error, *nargs;
780 
781 	g_topology_assert();
782 
783 	nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
784 	if (nargs == NULL) {
785 		gctl_error(req, "No '%s' argument", "nargs");
786 		return;
787 	}
788 	if (*nargs != 1) {
789 		gctl_error(req, "Missing device.");
790 		return;
791 	}
792 
793 	name = gctl_get_asciiparam(req, "arg0");
794 	if (name == NULL) {
795 		gctl_error(req, "No 'arg0' argument");
796 		return;
797 	}
798 	sc = g_cache_find_device(mp, name);
799 	if (sc == NULL) {
800 		G_CACHE_DEBUG(1, "Device %s is invalid.", name);
801 		gctl_error(req, "Device %s is invalid.", name);
802 		return;
803 	}
804 
805 	size = gctl_get_paraml(req, "size", sizeof(*size));
806 	if (size == NULL) {
807 		gctl_error(req, "No '%s' argument", "size");
808 		return;
809 	}
810 	if ((u_int)*size != 0 && (u_int)*size < 100) {
811 		gctl_error(req, "Invalid '%s' argument", "size");
812 		return;
813 	}
814 	if ((u_int)*size != 0)
815 		sc->sc_maxent = (u_int)*size;
816 
817 	bsize = gctl_get_paraml(req, "blocksize", sizeof(*bsize));
818 	if (bsize == NULL) {
819 		gctl_error(req, "No '%s' argument", "blocksize");
820 		return;
821 	}
822 	if (*bsize < 0) {
823 		gctl_error(req, "Invalid '%s' argument", "blocksize");
824 		return;
825 	}
826 
827 	if (sc->sc_type != G_CACHE_TYPE_AUTOMATIC)
828 		return;
829 
830 	strlcpy(md.md_name, name, sizeof(md.md_name));
831 	strlcpy(md.md_magic, G_CACHE_MAGIC, sizeof(md.md_magic));
832 	md.md_version = G_CACHE_VERSION;
833 	if ((u_int)*size != 0)
834 		md.md_size = (u_int)*size;
835 	else
836 		md.md_size = sc->sc_maxent;
837 	if ((u_int)*bsize != 0)
838 		md.md_bsize = (u_int)*bsize;
839 	else
840 		md.md_bsize = sc->sc_bsize;
841 	cp = LIST_FIRST(&sc->sc_geom->consumer);
842 	md.md_provsize = cp->provider->mediasize;
843 	error = g_cache_write_metadata(cp, &md);
844 	if (error == 0)
845 		G_CACHE_DEBUG(2, "Metadata on %s updated.", cp->provider->name);
846 	else
847 		G_CACHE_DEBUG(0, "Cannot update metadata on %s (error=%d).",
848 		    cp->provider->name, error);
849 }
850 
851 static void
852 g_cache_ctl_destroy(struct gctl_req *req, struct g_class *mp)
853 {
854 	int *nargs, *force, error, i;
855 	struct g_cache_softc *sc;
856 	const char *name;
857 	char param[16];
858 
859 	g_topology_assert();
860 
861 	nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
862 	if (nargs == NULL) {
863 		gctl_error(req, "No '%s' argument", "nargs");
864 		return;
865 	}
866 	if (*nargs <= 0) {
867 		gctl_error(req, "Missing device(s).");
868 		return;
869 	}
870 	force = gctl_get_paraml(req, "force", sizeof(*force));
871 	if (force == NULL) {
872 		gctl_error(req, "No 'force' argument");
873 		return;
874 	}
875 
876 	for (i = 0; i < *nargs; i++) {
877 		snprintf(param, sizeof(param), "arg%d", i);
878 		name = gctl_get_asciiparam(req, param);
879 		if (name == NULL) {
880 			gctl_error(req, "No 'arg%d' argument", i);
881 			return;
882 		}
883 		sc = g_cache_find_device(mp, name);
884 		if (sc == NULL) {
885 			G_CACHE_DEBUG(1, "Device %s is invalid.", name);
886 			gctl_error(req, "Device %s is invalid.", name);
887 			return;
888 		}
889 		error = g_cache_destroy(sc, *force);
890 		if (error != 0) {
891 			gctl_error(req, "Cannot destroy device %s (error=%d).",
892 			    sc->sc_name, error);
893 			return;
894 		}
895 	}
896 }
897 
898 static void
899 g_cache_ctl_reset(struct gctl_req *req, struct g_class *mp)
900 {
901 	struct g_cache_softc *sc;
902 	const char *name;
903 	char param[16];
904 	int i, *nargs;
905 
906 	g_topology_assert();
907 
908 	nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
909 	if (nargs == NULL) {
910 		gctl_error(req, "No '%s' argument", "nargs");
911 		return;
912 	}
913 	if (*nargs <= 0) {
914 		gctl_error(req, "Missing device(s).");
915 		return;
916 	}
917 
918 	for (i = 0; i < *nargs; i++) {
919 		snprintf(param, sizeof(param), "arg%d", i);
920 		name = gctl_get_asciiparam(req, param);
921 		if (name == NULL) {
922 			gctl_error(req, "No 'arg%d' argument", i);
923 			return;
924 		}
925 		sc = g_cache_find_device(mp, name);
926 		if (sc == NULL) {
927 			G_CACHE_DEBUG(1, "Device %s is invalid.", name);
928 			gctl_error(req, "Device %s is invalid.", name);
929 			return;
930 		}
931 		sc->sc_reads = 0;
932 		sc->sc_readbytes = 0;
933 		sc->sc_cachereads = 0;
934 		sc->sc_cachereadbytes = 0;
935 		sc->sc_cachehits = 0;
936 		sc->sc_cachemisses = 0;
937 		sc->sc_cachefull = 0;
938 		sc->sc_writes = 0;
939 		sc->sc_wrotebytes = 0;
940 	}
941 }
942 
943 static void
944 g_cache_config(struct gctl_req *req, struct g_class *mp, const char *verb)
945 {
946 	uint32_t *version;
947 
948 	g_topology_assert();
949 
950 	version = gctl_get_paraml(req, "version", sizeof(*version));
951 	if (version == NULL) {
952 		gctl_error(req, "No '%s' argument.", "version");
953 		return;
954 	}
955 	if (*version != G_CACHE_VERSION) {
956 		gctl_error(req, "Userland and kernel parts are out of sync.");
957 		return;
958 	}
959 
960 	if (strcmp(verb, "create") == 0) {
961 		g_cache_ctl_create(req, mp);
962 		return;
963 	} else if (strcmp(verb, "configure") == 0) {
964 		g_cache_ctl_configure(req, mp);
965 		return;
966 	} else if (strcmp(verb, "destroy") == 0 ||
967 	    strcmp(verb, "stop") == 0) {
968 		g_cache_ctl_destroy(req, mp);
969 		return;
970 	} else if (strcmp(verb, "reset") == 0) {
971 		g_cache_ctl_reset(req, mp);
972 		return;
973 	}
974 
975 	gctl_error(req, "Unknown verb.");
976 }
977 
978 static void
979 g_cache_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
980     struct g_consumer *cp, struct g_provider *pp)
981 {
982 	struct g_cache_softc *sc;
983 
984 	if (pp != NULL || cp != NULL)
985 		return;
986 	sc = gp->softc;
987 	sbuf_printf(sb, "%s<Size>%u</Size>\n", indent, sc->sc_maxent);
988 	sbuf_printf(sb, "%s<BlockSize>%u</BlockSize>\n", indent, sc->sc_bsize);
989 	sbuf_printf(sb, "%s<TailOffset>%ju</TailOffset>\n", indent,
990 	    (uintmax_t)sc->sc_tail);
991 	sbuf_printf(sb, "%s<Entries>%u</Entries>\n", indent, sc->sc_nent);
992 	sbuf_printf(sb, "%s<UsedEntries>%u</UsedEntries>\n", indent,
993 	    sc->sc_nused);
994 	sbuf_printf(sb, "%s<InvalidEntries>%u</InvalidEntries>\n", indent,
995 	    sc->sc_invalid);
996 	sbuf_printf(sb, "%s<Reads>%ju</Reads>\n", indent, sc->sc_reads);
997 	sbuf_printf(sb, "%s<ReadBytes>%ju</ReadBytes>\n", indent,
998 	    sc->sc_readbytes);
999 	sbuf_printf(sb, "%s<CacheReads>%ju</CacheReads>\n", indent,
1000 	    sc->sc_cachereads);
1001 	sbuf_printf(sb, "%s<CacheReadBytes>%ju</CacheReadBytes>\n", indent,
1002 	    sc->sc_cachereadbytes);
1003 	sbuf_printf(sb, "%s<CacheHits>%ju</CacheHits>\n", indent,
1004 	    sc->sc_cachehits);
1005 	sbuf_printf(sb, "%s<CacheMisses>%ju</CacheMisses>\n", indent,
1006 	    sc->sc_cachemisses);
1007 	sbuf_printf(sb, "%s<CacheFull>%ju</CacheFull>\n", indent,
1008 	    sc->sc_cachefull);
1009 	sbuf_printf(sb, "%s<Writes>%ju</Writes>\n", indent, sc->sc_writes);
1010 	sbuf_printf(sb, "%s<WroteBytes>%ju</WroteBytes>\n", indent,
1011 	    sc->sc_wrotebytes);
1012 }
1013 
1014 DECLARE_GEOM_CLASS(g_cache_class, g_cache);
1015 MODULE_VERSION(geom_cache, 0);
1016