xref: /freebsd/sys/geom/cache/g_cache.c (revision edf8578117e8844e02c0121147f45e4609b30680)
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 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/kernel.h>
33 #include <sys/module.h>
34 #include <sys/lock.h>
35 #include <sys/mutex.h>
36 #include <sys/bio.h>
37 #include <sys/sysctl.h>
38 #include <sys/malloc.h>
39 #include <sys/queue.h>
40 #include <sys/sbuf.h>
41 #include <sys/time.h>
42 #include <vm/uma.h>
43 #include <geom/geom.h>
44 #include <geom/geom_dbg.h>
45 #include <geom/cache/g_cache.h>
46 
47 FEATURE(geom_cache, "GEOM cache module");
48 
49 static MALLOC_DEFINE(M_GCACHE, "gcache_data", "GEOM_CACHE Data");
50 
51 SYSCTL_DECL(_kern_geom);
52 static SYSCTL_NODE(_kern_geom, OID_AUTO, cache, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
53     "GEOM_CACHE stuff");
54 static u_int g_cache_debug = 0;
55 SYSCTL_UINT(_kern_geom_cache, OID_AUTO, debug, CTLFLAG_RW, &g_cache_debug, 0,
56     "Debug level");
57 static u_int g_cache_enable = 1;
58 SYSCTL_UINT(_kern_geom_cache, OID_AUTO, enable, CTLFLAG_RW, &g_cache_enable, 0,
59     "");
60 static u_int g_cache_timeout = 10;
61 SYSCTL_UINT(_kern_geom_cache, OID_AUTO, timeout, CTLFLAG_RW, &g_cache_timeout,
62     0, "");
63 static u_int g_cache_idletime = 5;
64 SYSCTL_UINT(_kern_geom_cache, OID_AUTO, idletime, CTLFLAG_RW, &g_cache_idletime,
65     0, "");
66 static u_int g_cache_used_lo = 5;
67 static u_int g_cache_used_hi = 20;
68 static int
69 sysctl_handle_pct(SYSCTL_HANDLER_ARGS)
70 {
71 	u_int val = *(u_int *)arg1;
72 	int error;
73 
74 	error = sysctl_handle_int(oidp, &val, 0, req);
75 	if (error || !req->newptr)
76 		return (error);
77 	if (val > 100)
78 		return (EINVAL);
79 	if ((arg1 == &g_cache_used_lo && val > g_cache_used_hi) ||
80 	    (arg1 == &g_cache_used_hi && g_cache_used_lo > val))
81 		return (EINVAL);
82 	*(u_int *)arg1 = val;
83 	return (0);
84 }
85 SYSCTL_PROC(_kern_geom_cache, OID_AUTO, used_lo,
86     CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE, &g_cache_used_lo, 0,
87     sysctl_handle_pct, "IU",
88     "");
89 SYSCTL_PROC(_kern_geom_cache, OID_AUTO, used_hi,
90     CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE, &g_cache_used_hi, 0,
91     sysctl_handle_pct, "IU",
92     "");
93 
94 static int g_cache_destroy(struct g_cache_softc *sc, boolean_t force);
95 static g_ctl_destroy_geom_t g_cache_destroy_geom;
96 
97 static g_taste_t g_cache_taste;
98 static g_ctl_req_t g_cache_config;
99 static g_dumpconf_t g_cache_dumpconf;
100 
101 struct g_class g_cache_class = {
102 	.name = G_CACHE_CLASS_NAME,
103 	.version = G_VERSION,
104 	.ctlreq = g_cache_config,
105 	.taste = g_cache_taste,
106 	.destroy_geom = g_cache_destroy_geom
107 };
108 
109 #define	OFF2BNO(off, sc)	((off) >> (sc)->sc_bshift)
110 #define	BNO2OFF(bno, sc)	((bno) << (sc)->sc_bshift)
111 
112 static struct g_cache_desc *
113 g_cache_alloc(struct g_cache_softc *sc)
114 {
115 	struct g_cache_desc *dp;
116 
117 	mtx_assert(&sc->sc_mtx, MA_OWNED);
118 
119 	if (!TAILQ_EMPTY(&sc->sc_usedlist)) {
120 		dp = TAILQ_FIRST(&sc->sc_usedlist);
121 		TAILQ_REMOVE(&sc->sc_usedlist, dp, d_used);
122 		sc->sc_nused--;
123 		dp->d_flags = 0;
124 		LIST_REMOVE(dp, d_next);
125 		return (dp);
126 	}
127 	if (sc->sc_nent > sc->sc_maxent) {
128 		sc->sc_cachefull++;
129 		return (NULL);
130 	}
131 	dp = malloc(sizeof(*dp), M_GCACHE, M_NOWAIT | M_ZERO);
132 	if (dp == NULL)
133 		return (NULL);
134 	dp->d_data = uma_zalloc(sc->sc_zone, M_NOWAIT);
135 	if (dp->d_data == NULL) {
136 		free(dp, M_GCACHE);
137 		return (NULL);
138 	}
139 	sc->sc_nent++;
140 	return (dp);
141 }
142 
143 static void
144 g_cache_free(struct g_cache_softc *sc, struct g_cache_desc *dp)
145 {
146 
147 	mtx_assert(&sc->sc_mtx, MA_OWNED);
148 
149 	uma_zfree(sc->sc_zone, dp->d_data);
150 	free(dp, M_GCACHE);
151 	sc->sc_nent--;
152 }
153 
154 static void
155 g_cache_free_used(struct g_cache_softc *sc)
156 {
157 	struct g_cache_desc *dp;
158 	u_int n;
159 
160 	mtx_assert(&sc->sc_mtx, MA_OWNED);
161 
162 	n = g_cache_used_lo * sc->sc_maxent / 100;
163 	while (sc->sc_nused > n) {
164 		KASSERT(!TAILQ_EMPTY(&sc->sc_usedlist), ("used list empty"));
165 		dp = TAILQ_FIRST(&sc->sc_usedlist);
166 		TAILQ_REMOVE(&sc->sc_usedlist, dp, d_used);
167 		sc->sc_nused--;
168 		LIST_REMOVE(dp, d_next);
169 		g_cache_free(sc, dp);
170 	}
171 }
172 
173 static void
174 g_cache_deliver(struct g_cache_softc *sc, struct bio *bp,
175     struct g_cache_desc *dp, int error)
176 {
177 	off_t off1, off, len;
178 
179 	mtx_assert(&sc->sc_mtx, MA_OWNED);
180 	KASSERT(OFF2BNO(bp->bio_offset, sc) <= dp->d_bno, ("wrong entry"));
181 	KASSERT(OFF2BNO(bp->bio_offset + bp->bio_length - 1, sc) >=
182 	    dp->d_bno, ("wrong entry"));
183 
184 	off1 = BNO2OFF(dp->d_bno, sc);
185 	off = MAX(bp->bio_offset, off1);
186 	len = MIN(bp->bio_offset + bp->bio_length, off1 + sc->sc_bsize) - off;
187 
188 	if (bp->bio_error == 0)
189 		bp->bio_error = error;
190 	if (bp->bio_error == 0) {
191 		bcopy(dp->d_data + (off - off1),
192 		    bp->bio_data + (off - bp->bio_offset), len);
193 	}
194 	bp->bio_completed += len;
195 	KASSERT(bp->bio_completed <= bp->bio_length, ("extra data"));
196 	if (bp->bio_completed == bp->bio_length) {
197 		if (bp->bio_error != 0)
198 			bp->bio_completed = 0;
199 		g_io_deliver(bp, bp->bio_error);
200 	}
201 
202 	if (dp->d_flags & D_FLAG_USED) {
203 		TAILQ_REMOVE(&sc->sc_usedlist, dp, d_used);
204 		TAILQ_INSERT_TAIL(&sc->sc_usedlist, dp, d_used);
205 	} else if (OFF2BNO(off + len, sc) > dp->d_bno) {
206 		TAILQ_INSERT_TAIL(&sc->sc_usedlist, dp, d_used);
207 		sc->sc_nused++;
208 		dp->d_flags |= D_FLAG_USED;
209 	}
210 	dp->d_atime = time_uptime;
211 }
212 
213 static void
214 g_cache_done(struct bio *bp)
215 {
216 	struct g_cache_softc *sc;
217 	struct g_cache_desc *dp;
218 	struct bio *bp2, *tmpbp;
219 
220 	sc = bp->bio_from->geom->softc;
221 	KASSERT(G_CACHE_DESC1(bp) == sc, ("corrupt bio_caller in g_cache_done()"));
222 	dp = G_CACHE_DESC2(bp);
223 	mtx_lock(&sc->sc_mtx);
224 	bp2 = dp->d_biolist;
225 	while (bp2 != NULL) {
226 		KASSERT(G_CACHE_NEXT_BIO1(bp2) == sc, ("corrupt bio_driver in g_cache_done()"));
227 		tmpbp = G_CACHE_NEXT_BIO2(bp2);
228 		g_cache_deliver(sc, bp2, dp, bp->bio_error);
229 		bp2 = tmpbp;
230 	}
231 	dp->d_biolist = NULL;
232 	if (dp->d_flags & D_FLAG_INVALID) {
233 		sc->sc_invalid--;
234 		g_cache_free(sc, dp);
235 	} else if (bp->bio_error) {
236 		LIST_REMOVE(dp, d_next);
237 		if (dp->d_flags & D_FLAG_USED) {
238 			TAILQ_REMOVE(&sc->sc_usedlist, dp, d_used);
239 			sc->sc_nused--;
240 		}
241 		g_cache_free(sc, dp);
242 	}
243 	mtx_unlock(&sc->sc_mtx);
244 	g_destroy_bio(bp);
245 }
246 
247 static struct g_cache_desc *
248 g_cache_lookup(struct g_cache_softc *sc, off_t bno)
249 {
250 	struct g_cache_desc *dp;
251 
252 	mtx_assert(&sc->sc_mtx, MA_OWNED);
253 
254 	LIST_FOREACH(dp, &sc->sc_desclist[G_CACHE_BUCKET(bno)], d_next)
255 		if (dp->d_bno == bno)
256 			return (dp);
257 	return (NULL);
258 }
259 
260 static int
261 g_cache_read(struct g_cache_softc *sc, struct bio *bp)
262 {
263 	struct bio *cbp;
264 	struct g_cache_desc *dp;
265 
266 	mtx_lock(&sc->sc_mtx);
267 	dp = g_cache_lookup(sc,
268 	    OFF2BNO(bp->bio_offset + bp->bio_completed, sc));
269 	if (dp != NULL) {
270 		/* Add to waiters list or deliver. */
271 		sc->sc_cachehits++;
272 		if (dp->d_biolist != NULL) {
273 			G_CACHE_NEXT_BIO1(bp) = sc;
274 			G_CACHE_NEXT_BIO2(bp) = dp->d_biolist;
275 			dp->d_biolist = bp;
276 		} else
277 			g_cache_deliver(sc, bp, dp, 0);
278 		mtx_unlock(&sc->sc_mtx);
279 		return (0);
280 	}
281 
282 	/* Cache miss.  Allocate entry and schedule bio.  */
283 	sc->sc_cachemisses++;
284 	dp = g_cache_alloc(sc);
285 	if (dp == NULL) {
286 		mtx_unlock(&sc->sc_mtx);
287 		return (ENOMEM);
288 	}
289 	cbp = g_clone_bio(bp);
290 	if (cbp == NULL) {
291 		g_cache_free(sc, dp);
292 		mtx_unlock(&sc->sc_mtx);
293 		return (ENOMEM);
294 	}
295 
296 	dp->d_bno = OFF2BNO(bp->bio_offset + bp->bio_completed, sc);
297 	G_CACHE_NEXT_BIO1(bp) = sc;
298 	G_CACHE_NEXT_BIO2(bp) = NULL;
299 	dp->d_biolist = bp;
300 	LIST_INSERT_HEAD(&sc->sc_desclist[G_CACHE_BUCKET(dp->d_bno)],
301 	    dp, d_next);
302 	mtx_unlock(&sc->sc_mtx);
303 
304 	G_CACHE_DESC1(cbp) = sc;
305 	G_CACHE_DESC2(cbp) = dp;
306 	cbp->bio_done = g_cache_done;
307 	cbp->bio_offset = BNO2OFF(dp->d_bno, sc);
308 	cbp->bio_data = dp->d_data;
309 	cbp->bio_length = sc->sc_bsize;
310 	g_io_request(cbp, LIST_FIRST(&bp->bio_to->geom->consumer));
311 	return (0);
312 }
313 
314 static void
315 g_cache_invalidate(struct g_cache_softc *sc, struct bio *bp)
316 {
317 	struct g_cache_desc *dp;
318 	off_t bno, lim;
319 
320 	mtx_lock(&sc->sc_mtx);
321 	bno = OFF2BNO(bp->bio_offset, sc);
322 	lim = OFF2BNO(bp->bio_offset + bp->bio_length - 1, sc);
323 	do {
324 		if ((dp = g_cache_lookup(sc, bno)) != NULL) {
325 			LIST_REMOVE(dp, d_next);
326 			if (dp->d_flags & D_FLAG_USED) {
327 				TAILQ_REMOVE(&sc->sc_usedlist, dp, d_used);
328 				sc->sc_nused--;
329 			}
330 			if (dp->d_biolist == NULL)
331 				g_cache_free(sc, dp);
332 			else {
333 				dp->d_flags = D_FLAG_INVALID;
334 				sc->sc_invalid++;
335 			}
336 		}
337 		bno++;
338 	} while (bno <= lim);
339 	mtx_unlock(&sc->sc_mtx);
340 }
341 
342 static void
343 g_cache_start(struct bio *bp)
344 {
345 	struct g_cache_softc *sc;
346 	struct g_geom *gp;
347 	struct g_cache_desc *dp;
348 	struct bio *cbp;
349 
350 	gp = bp->bio_to->geom;
351 	sc = gp->softc;
352 	G_CACHE_LOGREQ(bp, "Request received.");
353 	switch (bp->bio_cmd) {
354 	case BIO_READ:
355 		sc->sc_reads++;
356 		sc->sc_readbytes += bp->bio_length;
357 		if (!g_cache_enable)
358 			break;
359 		if (bp->bio_offset + bp->bio_length > sc->sc_tail)
360 			break;
361 		if (OFF2BNO(bp->bio_offset, sc) ==
362 		    OFF2BNO(bp->bio_offset + bp->bio_length - 1, sc)) {
363 			sc->sc_cachereads++;
364 			sc->sc_cachereadbytes += bp->bio_length;
365 			if (g_cache_read(sc, bp) == 0)
366 				return;
367 			sc->sc_cachereads--;
368 			sc->sc_cachereadbytes -= bp->bio_length;
369 			break;
370 		} else if (OFF2BNO(bp->bio_offset, sc) + 1 ==
371 		    OFF2BNO(bp->bio_offset + bp->bio_length - 1, sc)) {
372 			mtx_lock(&sc->sc_mtx);
373 			dp = g_cache_lookup(sc, OFF2BNO(bp->bio_offset, sc));
374 			if (dp == NULL || dp->d_biolist != NULL) {
375 				mtx_unlock(&sc->sc_mtx);
376 				break;
377 			}
378 			sc->sc_cachereads++;
379 			sc->sc_cachereadbytes += bp->bio_length;
380 			g_cache_deliver(sc, bp, dp, 0);
381 			mtx_unlock(&sc->sc_mtx);
382 			if (g_cache_read(sc, bp) == 0)
383 				return;
384 			sc->sc_cachereads--;
385 			sc->sc_cachereadbytes -= bp->bio_length;
386 			break;
387 		}
388 		break;
389 	case BIO_WRITE:
390 		sc->sc_writes++;
391 		sc->sc_wrotebytes += bp->bio_length;
392 		g_cache_invalidate(sc, bp);
393 		break;
394 	}
395 	cbp = g_clone_bio(bp);
396 	if (cbp == NULL) {
397 		g_io_deliver(bp, ENOMEM);
398 		return;
399 	}
400 	cbp->bio_done = g_std_done;
401 	G_CACHE_LOGREQ(cbp, "Sending request.");
402 	g_io_request(cbp, LIST_FIRST(&gp->consumer));
403 }
404 
405 static void
406 g_cache_go(void *arg)
407 {
408 	struct g_cache_softc *sc = arg;
409 	struct g_cache_desc *dp;
410 	int i;
411 
412 	mtx_assert(&sc->sc_mtx, MA_OWNED);
413 
414 	/* Forcibly mark idle ready entries as used. */
415 	for (i = 0; i < G_CACHE_BUCKETS; i++) {
416 		LIST_FOREACH(dp, &sc->sc_desclist[i], d_next) {
417 			if (dp->d_flags & D_FLAG_USED ||
418 			    dp->d_biolist != NULL ||
419 			    time_uptime - dp->d_atime < g_cache_idletime)
420 				continue;
421 			TAILQ_INSERT_TAIL(&sc->sc_usedlist, dp, d_used);
422 			sc->sc_nused++;
423 			dp->d_flags |= D_FLAG_USED;
424 		}
425 	}
426 
427 	/* Keep the number of used entries low. */
428 	if (sc->sc_nused > g_cache_used_hi * sc->sc_maxent / 100)
429 		g_cache_free_used(sc);
430 
431 	callout_reset(&sc->sc_callout, g_cache_timeout * hz, g_cache_go, sc);
432 }
433 
434 static int
435 g_cache_access(struct g_provider *pp, int dr, int dw, int de)
436 {
437 	struct g_geom *gp;
438 	struct g_consumer *cp;
439 	int error;
440 
441 	gp = pp->geom;
442 	cp = LIST_FIRST(&gp->consumer);
443 	error = g_access(cp, dr, dw, de);
444 
445 	return (error);
446 }
447 
448 static void
449 g_cache_orphan(struct g_consumer *cp)
450 {
451 
452 	g_topology_assert();
453 	g_cache_destroy(cp->geom->softc, 1);
454 }
455 
456 static struct g_cache_softc *
457 g_cache_find_device(struct g_class *mp, const char *name)
458 {
459 	struct g_geom *gp;
460 
461 	LIST_FOREACH(gp, &mp->geom, geom) {
462 		if (strcmp(gp->name, name) == 0)
463 			return (gp->softc);
464 	}
465 	return (NULL);
466 }
467 
468 static struct g_geom *
469 g_cache_create(struct g_class *mp, struct g_provider *pp,
470     const struct g_cache_metadata *md, u_int type)
471 {
472 	struct g_cache_softc *sc;
473 	struct g_geom *gp;
474 	struct g_provider *newpp;
475 	struct g_consumer *cp;
476 	u_int bshift;
477 	int i;
478 
479 	g_topology_assert();
480 
481 	gp = NULL;
482 	newpp = NULL;
483 	cp = NULL;
484 
485 	G_CACHE_DEBUG(1, "Creating device %s.", md->md_name);
486 
487 	/* Cache size is minimum 100. */
488 	if (md->md_size < 100) {
489 		G_CACHE_DEBUG(0, "Invalid size for device %s.", md->md_name);
490 		return (NULL);
491 	}
492 
493 	/* Block size restrictions. */
494 	bshift = ffs(md->md_bsize) - 1;
495 	if (md->md_bsize == 0 || md->md_bsize > maxphys ||
496 	    md->md_bsize != 1 << bshift ||
497 	    (md->md_bsize % pp->sectorsize) != 0) {
498 		G_CACHE_DEBUG(0, "Invalid blocksize for provider %s.", pp->name);
499 		return (NULL);
500 	}
501 
502 	/* Check for duplicate unit. */
503 	if (g_cache_find_device(mp, (const char *)&md->md_name) != NULL) {
504 		G_CACHE_DEBUG(0, "Provider %s already exists.", md->md_name);
505 		return (NULL);
506 	}
507 
508 	gp = g_new_geomf(mp, "%s", md->md_name);
509 	sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO);
510 	sc->sc_type = type;
511 	sc->sc_bshift = bshift;
512 	sc->sc_bsize = 1 << bshift;
513 	sc->sc_zone = uma_zcreate("gcache", sc->sc_bsize, NULL, NULL, NULL, NULL,
514 	    UMA_ALIGN_PTR, 0);
515 	mtx_init(&sc->sc_mtx, "GEOM CACHE mutex", NULL, MTX_DEF);
516 	for (i = 0; i < G_CACHE_BUCKETS; i++)
517 		LIST_INIT(&sc->sc_desclist[i]);
518 	TAILQ_INIT(&sc->sc_usedlist);
519 	sc->sc_maxent = md->md_size;
520 	callout_init_mtx(&sc->sc_callout, &sc->sc_mtx, 0);
521 	gp->softc = sc;
522 	sc->sc_geom = gp;
523 	gp->start = g_cache_start;
524 	gp->orphan = g_cache_orphan;
525 	gp->access = g_cache_access;
526 	gp->dumpconf = g_cache_dumpconf;
527 
528 	newpp = g_new_providerf(gp, "cache/%s", gp->name);
529 	newpp->sectorsize = pp->sectorsize;
530 	newpp->mediasize = pp->mediasize;
531 	if (type == G_CACHE_TYPE_AUTOMATIC)
532 		newpp->mediasize -= pp->sectorsize;
533 	sc->sc_tail = BNO2OFF(OFF2BNO(newpp->mediasize, sc), sc);
534 
535 	cp = g_new_consumer(gp);
536 	if (g_attach(cp, pp) != 0) {
537 		G_CACHE_DEBUG(0, "Cannot attach to provider %s.", pp->name);
538 		g_destroy_consumer(cp);
539 		g_destroy_provider(newpp);
540 		mtx_destroy(&sc->sc_mtx);
541 		g_free(sc);
542 		g_destroy_geom(gp);
543 		return (NULL);
544 	}
545 
546 	g_error_provider(newpp, 0);
547 	G_CACHE_DEBUG(0, "Device %s created.", gp->name);
548 	callout_reset(&sc->sc_callout, g_cache_timeout * hz, g_cache_go, sc);
549 	return (gp);
550 }
551 
552 static int
553 g_cache_destroy(struct g_cache_softc *sc, boolean_t force)
554 {
555 	struct g_geom *gp;
556 	struct g_provider *pp;
557 	struct g_cache_desc *dp, *dp2;
558 	int i;
559 
560 	g_topology_assert();
561 	if (sc == NULL)
562 		return (ENXIO);
563 	gp = sc->sc_geom;
564 	pp = LIST_FIRST(&gp->provider);
565 	if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) {
566 		if (force) {
567 			G_CACHE_DEBUG(0, "Device %s is still open, so it "
568 			    "can't be definitely removed.", pp->name);
569 		} else {
570 			G_CACHE_DEBUG(1, "Device %s is still open (r%dw%de%d).",
571 			    pp->name, pp->acr, pp->acw, pp->ace);
572 			return (EBUSY);
573 		}
574 	} else {
575 		G_CACHE_DEBUG(0, "Device %s removed.", gp->name);
576 	}
577 	callout_drain(&sc->sc_callout);
578 	mtx_lock(&sc->sc_mtx);
579 	for (i = 0; i < G_CACHE_BUCKETS; i++) {
580 		dp = LIST_FIRST(&sc->sc_desclist[i]);
581 		while (dp != NULL) {
582 			dp2 = LIST_NEXT(dp, d_next);
583 			g_cache_free(sc, dp);
584 			dp = dp2;
585 		}
586 	}
587 	mtx_unlock(&sc->sc_mtx);
588 	mtx_destroy(&sc->sc_mtx);
589 	uma_zdestroy(sc->sc_zone);
590 	g_free(sc);
591 	gp->softc = NULL;
592 	g_wither_geom(gp, ENXIO);
593 
594 	return (0);
595 }
596 
597 static int
598 g_cache_destroy_geom(struct gctl_req *req, struct g_class *mp, struct g_geom *gp)
599 {
600 
601 	return (g_cache_destroy(gp->softc, 0));
602 }
603 
604 static int
605 g_cache_read_metadata(struct g_consumer *cp, struct g_cache_metadata *md)
606 {
607 	struct g_provider *pp;
608 	u_char *buf;
609 	int error;
610 
611 	g_topology_assert();
612 
613 	error = g_access(cp, 1, 0, 0);
614 	if (error != 0)
615 		return (error);
616 	pp = cp->provider;
617 	g_topology_unlock();
618 	buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize,
619 	    &error);
620 	g_topology_lock();
621 	g_access(cp, -1, 0, 0);
622 	if (buf == NULL)
623 		return (error);
624 
625 	/* Decode metadata. */
626 	cache_metadata_decode(buf, md);
627 	g_free(buf);
628 
629 	return (0);
630 }
631 
632 static int
633 g_cache_write_metadata(struct g_consumer *cp, struct g_cache_metadata *md)
634 {
635 	struct g_provider *pp;
636 	u_char *buf;
637 	int error;
638 
639 	g_topology_assert();
640 
641 	error = g_access(cp, 0, 1, 0);
642 	if (error != 0)
643 		return (error);
644 	pp = cp->provider;
645 	buf = malloc((size_t)pp->sectorsize, M_GCACHE, M_WAITOK | M_ZERO);
646 	cache_metadata_encode(md, buf);
647 	g_topology_unlock();
648 	error = g_write_data(cp, pp->mediasize - pp->sectorsize, buf, pp->sectorsize);
649 	g_topology_lock();
650 	g_access(cp, 0, -1, 0);
651 	free(buf, M_GCACHE);
652 
653 	return (error);
654 }
655 
656 static struct g_geom *
657 g_cache_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
658 {
659 	struct g_cache_metadata md;
660 	struct g_consumer *cp;
661 	struct g_geom *gp;
662 	int error;
663 
664 	g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name);
665 	g_topology_assert();
666 
667 	G_CACHE_DEBUG(3, "Tasting %s.", pp->name);
668 
669 	gp = g_new_geomf(mp, "cache:taste");
670 	gp->start = g_cache_start;
671 	gp->orphan = g_cache_orphan;
672 	gp->access = g_cache_access;
673 	cp = g_new_consumer(gp);
674 	cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
675 	error = g_attach(cp, pp);
676 	if (error == 0) {
677 		error = g_cache_read_metadata(cp, &md);
678 		g_detach(cp);
679 	}
680 	g_destroy_consumer(cp);
681 	g_destroy_geom(gp);
682 	if (error != 0)
683 		return (NULL);
684 
685 	if (strcmp(md.md_magic, G_CACHE_MAGIC) != 0)
686 		return (NULL);
687 	if (md.md_version > G_CACHE_VERSION) {
688 		printf("geom_cache.ko module is too old to handle %s.\n",
689 		    pp->name);
690 		return (NULL);
691 	}
692 	if (md.md_provsize != pp->mediasize)
693 		return (NULL);
694 
695 	gp = g_cache_create(mp, pp, &md, G_CACHE_TYPE_AUTOMATIC);
696 	if (gp == NULL) {
697 		G_CACHE_DEBUG(0, "Can't create %s.", md.md_name);
698 		return (NULL);
699 	}
700 	return (gp);
701 }
702 
703 static void
704 g_cache_ctl_create(struct gctl_req *req, struct g_class *mp)
705 {
706 	struct g_cache_metadata md;
707 	struct g_provider *pp;
708 	struct g_geom *gp;
709 	intmax_t *bsize, *size;
710 	const char *name;
711 	int *nargs;
712 
713 	g_topology_assert();
714 
715 	nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
716 	if (nargs == NULL) {
717 		gctl_error(req, "No '%s' argument", "nargs");
718 		return;
719 	}
720 	if (*nargs != 2) {
721 		gctl_error(req, "Invalid number of arguments.");
722 		return;
723 	}
724 
725 	strlcpy(md.md_magic, G_CACHE_MAGIC, sizeof(md.md_magic));
726 	md.md_version = G_CACHE_VERSION;
727 	name = gctl_get_asciiparam(req, "arg0");
728 	if (name == NULL) {
729 		gctl_error(req, "No 'arg0' argument");
730 		return;
731 	}
732 	strlcpy(md.md_name, name, sizeof(md.md_name));
733 
734 	size = gctl_get_paraml(req, "size", sizeof(*size));
735 	if (size == NULL) {
736 		gctl_error(req, "No '%s' argument", "size");
737 		return;
738 	}
739 	if ((u_int)*size < 100) {
740 		gctl_error(req, "Invalid '%s' argument", "size");
741 		return;
742 	}
743 	md.md_size = (u_int)*size;
744 
745 	bsize = gctl_get_paraml(req, "blocksize", sizeof(*bsize));
746 	if (bsize == NULL) {
747 		gctl_error(req, "No '%s' argument", "blocksize");
748 		return;
749 	}
750 	if (*bsize < 0) {
751 		gctl_error(req, "Invalid '%s' argument", "blocksize");
752 		return;
753 	}
754 	md.md_bsize = (u_int)*bsize;
755 
756 	/* This field is not important here. */
757 	md.md_provsize = 0;
758 
759 	pp = gctl_get_provider(req, "arg1");
760 	if (pp == NULL)
761 		return;
762 	gp = g_cache_create(mp, pp, &md, G_CACHE_TYPE_MANUAL);
763 	if (gp == NULL) {
764 		gctl_error(req, "Can't create %s.", md.md_name);
765 		return;
766 	}
767 }
768 
769 static void
770 g_cache_ctl_configure(struct gctl_req *req, struct g_class *mp)
771 {
772 	struct g_cache_metadata md;
773 	struct g_cache_softc *sc;
774 	struct g_consumer *cp;
775 	intmax_t *bsize, *size;
776 	const char *name;
777 	int error, *nargs;
778 
779 	g_topology_assert();
780 
781 	nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
782 	if (nargs == NULL) {
783 		gctl_error(req, "No '%s' argument", "nargs");
784 		return;
785 	}
786 	if (*nargs != 1) {
787 		gctl_error(req, "Missing device.");
788 		return;
789 	}
790 
791 	name = gctl_get_asciiparam(req, "arg0");
792 	if (name == NULL) {
793 		gctl_error(req, "No 'arg0' argument");
794 		return;
795 	}
796 	sc = g_cache_find_device(mp, name);
797 	if (sc == NULL) {
798 		G_CACHE_DEBUG(1, "Device %s is invalid.", name);
799 		gctl_error(req, "Device %s is invalid.", name);
800 		return;
801 	}
802 
803 	size = gctl_get_paraml(req, "size", sizeof(*size));
804 	if (size == NULL) {
805 		gctl_error(req, "No '%s' argument", "size");
806 		return;
807 	}
808 	if ((u_int)*size != 0 && (u_int)*size < 100) {
809 		gctl_error(req, "Invalid '%s' argument", "size");
810 		return;
811 	}
812 	if ((u_int)*size != 0)
813 		sc->sc_maxent = (u_int)*size;
814 
815 	bsize = gctl_get_paraml(req, "blocksize", sizeof(*bsize));
816 	if (bsize == NULL) {
817 		gctl_error(req, "No '%s' argument", "blocksize");
818 		return;
819 	}
820 	if (*bsize < 0) {
821 		gctl_error(req, "Invalid '%s' argument", "blocksize");
822 		return;
823 	}
824 
825 	if (sc->sc_type != G_CACHE_TYPE_AUTOMATIC)
826 		return;
827 
828 	strlcpy(md.md_name, name, sizeof(md.md_name));
829 	strlcpy(md.md_magic, G_CACHE_MAGIC, sizeof(md.md_magic));
830 	md.md_version = G_CACHE_VERSION;
831 	if ((u_int)*size != 0)
832 		md.md_size = (u_int)*size;
833 	else
834 		md.md_size = sc->sc_maxent;
835 	if ((u_int)*bsize != 0)
836 		md.md_bsize = (u_int)*bsize;
837 	else
838 		md.md_bsize = sc->sc_bsize;
839 	cp = LIST_FIRST(&sc->sc_geom->consumer);
840 	md.md_provsize = cp->provider->mediasize;
841 	error = g_cache_write_metadata(cp, &md);
842 	if (error == 0)
843 		G_CACHE_DEBUG(2, "Metadata on %s updated.", cp->provider->name);
844 	else
845 		G_CACHE_DEBUG(0, "Cannot update metadata on %s (error=%d).",
846 		    cp->provider->name, error);
847 }
848 
849 static void
850 g_cache_ctl_destroy(struct gctl_req *req, struct g_class *mp)
851 {
852 	int *nargs, *force, error, i;
853 	struct g_cache_softc *sc;
854 	const char *name;
855 	char param[16];
856 
857 	g_topology_assert();
858 
859 	nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
860 	if (nargs == NULL) {
861 		gctl_error(req, "No '%s' argument", "nargs");
862 		return;
863 	}
864 	if (*nargs <= 0) {
865 		gctl_error(req, "Missing device(s).");
866 		return;
867 	}
868 	force = gctl_get_paraml(req, "force", sizeof(*force));
869 	if (force == NULL) {
870 		gctl_error(req, "No 'force' argument");
871 		return;
872 	}
873 
874 	for (i = 0; i < *nargs; i++) {
875 		snprintf(param, sizeof(param), "arg%d", i);
876 		name = gctl_get_asciiparam(req, param);
877 		if (name == NULL) {
878 			gctl_error(req, "No 'arg%d' argument", i);
879 			return;
880 		}
881 		sc = g_cache_find_device(mp, name);
882 		if (sc == NULL) {
883 			G_CACHE_DEBUG(1, "Device %s is invalid.", name);
884 			gctl_error(req, "Device %s is invalid.", name);
885 			return;
886 		}
887 		error = g_cache_destroy(sc, *force);
888 		if (error != 0) {
889 			gctl_error(req, "Cannot destroy device %s (error=%d).",
890 			    sc->sc_name, error);
891 			return;
892 		}
893 	}
894 }
895 
896 static void
897 g_cache_ctl_reset(struct gctl_req *req, struct g_class *mp)
898 {
899 	struct g_cache_softc *sc;
900 	const char *name;
901 	char param[16];
902 	int i, *nargs;
903 
904 	g_topology_assert();
905 
906 	nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
907 	if (nargs == NULL) {
908 		gctl_error(req, "No '%s' argument", "nargs");
909 		return;
910 	}
911 	if (*nargs <= 0) {
912 		gctl_error(req, "Missing device(s).");
913 		return;
914 	}
915 
916 	for (i = 0; i < *nargs; i++) {
917 		snprintf(param, sizeof(param), "arg%d", i);
918 		name = gctl_get_asciiparam(req, param);
919 		if (name == NULL) {
920 			gctl_error(req, "No 'arg%d' argument", i);
921 			return;
922 		}
923 		sc = g_cache_find_device(mp, name);
924 		if (sc == NULL) {
925 			G_CACHE_DEBUG(1, "Device %s is invalid.", name);
926 			gctl_error(req, "Device %s is invalid.", name);
927 			return;
928 		}
929 		sc->sc_reads = 0;
930 		sc->sc_readbytes = 0;
931 		sc->sc_cachereads = 0;
932 		sc->sc_cachereadbytes = 0;
933 		sc->sc_cachehits = 0;
934 		sc->sc_cachemisses = 0;
935 		sc->sc_cachefull = 0;
936 		sc->sc_writes = 0;
937 		sc->sc_wrotebytes = 0;
938 	}
939 }
940 
941 static void
942 g_cache_config(struct gctl_req *req, struct g_class *mp, const char *verb)
943 {
944 	uint32_t *version;
945 
946 	g_topology_assert();
947 
948 	version = gctl_get_paraml(req, "version", sizeof(*version));
949 	if (version == NULL) {
950 		gctl_error(req, "No '%s' argument.", "version");
951 		return;
952 	}
953 	if (*version != G_CACHE_VERSION) {
954 		gctl_error(req, "Userland and kernel parts are out of sync.");
955 		return;
956 	}
957 
958 	if (strcmp(verb, "create") == 0) {
959 		g_cache_ctl_create(req, mp);
960 		return;
961 	} else if (strcmp(verb, "configure") == 0) {
962 		g_cache_ctl_configure(req, mp);
963 		return;
964 	} else if (strcmp(verb, "destroy") == 0 ||
965 	    strcmp(verb, "stop") == 0) {
966 		g_cache_ctl_destroy(req, mp);
967 		return;
968 	} else if (strcmp(verb, "reset") == 0) {
969 		g_cache_ctl_reset(req, mp);
970 		return;
971 	}
972 
973 	gctl_error(req, "Unknown verb.");
974 }
975 
976 static void
977 g_cache_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
978     struct g_consumer *cp, struct g_provider *pp)
979 {
980 	struct g_cache_softc *sc;
981 
982 	if (pp != NULL || cp != NULL)
983 		return;
984 	sc = gp->softc;
985 	sbuf_printf(sb, "%s<Size>%u</Size>\n", indent, sc->sc_maxent);
986 	sbuf_printf(sb, "%s<BlockSize>%u</BlockSize>\n", indent, sc->sc_bsize);
987 	sbuf_printf(sb, "%s<TailOffset>%ju</TailOffset>\n", indent,
988 	    (uintmax_t)sc->sc_tail);
989 	sbuf_printf(sb, "%s<Entries>%u</Entries>\n", indent, sc->sc_nent);
990 	sbuf_printf(sb, "%s<UsedEntries>%u</UsedEntries>\n", indent,
991 	    sc->sc_nused);
992 	sbuf_printf(sb, "%s<InvalidEntries>%u</InvalidEntries>\n", indent,
993 	    sc->sc_invalid);
994 	sbuf_printf(sb, "%s<Reads>%ju</Reads>\n", indent, sc->sc_reads);
995 	sbuf_printf(sb, "%s<ReadBytes>%ju</ReadBytes>\n", indent,
996 	    sc->sc_readbytes);
997 	sbuf_printf(sb, "%s<CacheReads>%ju</CacheReads>\n", indent,
998 	    sc->sc_cachereads);
999 	sbuf_printf(sb, "%s<CacheReadBytes>%ju</CacheReadBytes>\n", indent,
1000 	    sc->sc_cachereadbytes);
1001 	sbuf_printf(sb, "%s<CacheHits>%ju</CacheHits>\n", indent,
1002 	    sc->sc_cachehits);
1003 	sbuf_printf(sb, "%s<CacheMisses>%ju</CacheMisses>\n", indent,
1004 	    sc->sc_cachemisses);
1005 	sbuf_printf(sb, "%s<CacheFull>%ju</CacheFull>\n", indent,
1006 	    sc->sc_cachefull);
1007 	sbuf_printf(sb, "%s<Writes>%ju</Writes>\n", indent, sc->sc_writes);
1008 	sbuf_printf(sb, "%s<WroteBytes>%ju</WroteBytes>\n", indent,
1009 	    sc->sc_wrotebytes);
1010 }
1011 
1012 DECLARE_GEOM_CLASS(g_cache_class, g_cache);
1013 MODULE_VERSION(geom_cache, 0);
1014