xref: /netbsd-src/sys/dev/dkwedge/dk.c (revision b1c86f5f087524e68db12794ee9c3e3da1ab17a0)
1 /*	$NetBSD: dk.c,v 1.57 2010/08/04 12:34:00 bouyer Exp $	*/
2 
3 /*-
4  * Copyright (c) 2004, 2005, 2006, 2007 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 #include <sys/cdefs.h>
33 __KERNEL_RCSID(0, "$NetBSD: dk.c,v 1.57 2010/08/04 12:34:00 bouyer Exp $");
34 
35 #ifdef _KERNEL_OPT
36 #include "opt_dkwedge.h"
37 #endif
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/proc.h>
42 #include <sys/errno.h>
43 #include <sys/pool.h>
44 #include <sys/ioctl.h>
45 #include <sys/disklabel.h>
46 #include <sys/disk.h>
47 #include <sys/fcntl.h>
48 #include <sys/buf.h>
49 #include <sys/bufq.h>
50 #include <sys/vnode.h>
51 #include <sys/stat.h>
52 #include <sys/conf.h>
53 #include <sys/callout.h>
54 #include <sys/kernel.h>
55 #include <sys/malloc.h>
56 #include <sys/device.h>
57 #include <sys/kauth.h>
58 
59 #include <miscfs/specfs/specdev.h>
60 
61 MALLOC_DEFINE(M_DKWEDGE, "dkwedge", "Disk wedge structures");
62 
63 typedef enum {
64 	DKW_STATE_LARVAL	= 0,
65 	DKW_STATE_RUNNING	= 1,
66 	DKW_STATE_DYING		= 2,
67 	DKW_STATE_DEAD		= 666
68 } dkwedge_state_t;
69 
70 struct dkwedge_softc {
71 	struct device	*sc_dev;	/* pointer to our pseudo-device */
72 	struct cfdata	sc_cfdata;	/* our cfdata structure */
73 	uint8_t		sc_wname[128];	/* wedge name (Unicode, UTF-8) */
74 
75 	dkwedge_state_t sc_state;	/* state this wedge is in */
76 
77 	struct disk	*sc_parent;	/* parent disk */
78 	daddr_t		sc_offset;	/* LBA offset of wedge in parent */
79 	uint64_t	sc_size;	/* size of wedge in blocks */
80 	char		sc_ptype[32];	/* partition type */
81 	dev_t		sc_pdev;	/* cached parent's dev_t */
82 					/* link on parent's wedge list */
83 	LIST_ENTRY(dkwedge_softc) sc_plink;
84 
85 	struct disk	sc_dk;		/* our own disk structure */
86 	struct bufq_state *sc_bufq;	/* buffer queue */
87 	struct callout	sc_restart_ch;	/* callout to restart I/O */
88 
89 	u_int		sc_iopend;	/* I/Os pending */
90 	int		sc_flags;	/* flags (splbio) */
91 };
92 
93 #define	DK_F_WAIT_DRAIN		0x0001	/* waiting for I/O to drain */
94 
95 static void	dkstart(struct dkwedge_softc *);
96 static void	dkiodone(struct buf *);
97 static void	dkrestart(void *);
98 static void	dkminphys(struct buf *);
99 
100 static int	dklastclose(struct dkwedge_softc *);
101 static int	dkwedge_detach(device_t, int);
102 
103 static dev_type_open(dkopen);
104 static dev_type_close(dkclose);
105 static dev_type_read(dkread);
106 static dev_type_write(dkwrite);
107 static dev_type_ioctl(dkioctl);
108 static dev_type_strategy(dkstrategy);
109 static dev_type_dump(dkdump);
110 static dev_type_size(dksize);
111 
112 const struct bdevsw dk_bdevsw = {
113 	dkopen, dkclose, dkstrategy, dkioctl, dkdump, dksize, D_DISK
114 };
115 
116 const struct cdevsw dk_cdevsw = {
117 	dkopen, dkclose, dkread, dkwrite, dkioctl,
118 	    nostop, notty, nopoll, nommap, nokqfilter, D_DISK
119 };
120 
121 static struct dkwedge_softc **dkwedges;
122 static u_int ndkwedges;
123 static krwlock_t dkwedges_lock;
124 
125 static LIST_HEAD(, dkwedge_discovery_method) dkwedge_discovery_methods;
126 static krwlock_t dkwedge_discovery_methods_lock;
127 
128 /*
129  * dkwedge_match:
130  *
131  *	Autoconfiguration match function for pseudo-device glue.
132  */
133 static int
134 dkwedge_match(device_t parent, cfdata_t match,
135     void *aux)
136 {
137 
138 	/* Pseudo-device; always present. */
139 	return (1);
140 }
141 
142 /*
143  * dkwedge_attach:
144  *
145  *	Autoconfiguration attach function for pseudo-device glue.
146  */
147 static void
148 dkwedge_attach(device_t parent, device_t self,
149     void *aux)
150 {
151 
152 	if (!pmf_device_register(self, NULL, NULL))
153 		aprint_error_dev(self, "couldn't establish power handler\n");
154 }
155 
156 CFDRIVER_DECL(dk, DV_DISK, NULL);
157 CFATTACH_DECL3_NEW(dk, 0,
158     dkwedge_match, dkwedge_attach, dkwedge_detach, NULL, NULL, NULL,
159     DVF_DETACH_SHUTDOWN);
160 
161 /*
162  * dkwedge_wait_drain:
163  *
164  *	Wait for I/O on the wedge to drain.
165  *	NOTE: Must be called at splbio()!
166  */
167 static void
168 dkwedge_wait_drain(struct dkwedge_softc *sc)
169 {
170 
171 	while (sc->sc_iopend != 0) {
172 		sc->sc_flags |= DK_F_WAIT_DRAIN;
173 		(void) tsleep(&sc->sc_iopend, PRIBIO, "dkdrn", 0);
174 	}
175 }
176 
177 /*
178  * dkwedge_compute_pdev:
179  *
180  *	Compute the parent disk's dev_t.
181  */
182 static int
183 dkwedge_compute_pdev(const char *pname, dev_t *pdevp)
184 {
185 	const char *name, *cp;
186 	int punit, pmaj;
187 	char devname[16];
188 
189 	name = pname;
190 	if ((pmaj = devsw_name2blk(name, devname, sizeof(devname))) == -1)
191 		return (ENODEV);
192 
193 	name += strlen(devname);
194 	for (cp = name, punit = 0; *cp >= '0' && *cp <= '9'; cp++)
195 		punit = (punit * 10) + (*cp - '0');
196 	if (cp == name) {
197 		/* Invalid parent disk name. */
198 		return (ENODEV);
199 	}
200 
201 	*pdevp = MAKEDISKDEV(pmaj, punit, RAW_PART);
202 
203 	return (0);
204 }
205 
206 /*
207  * dkwedge_array_expand:
208  *
209  *	Expand the dkwedges array.
210  */
211 static void
212 dkwedge_array_expand(void)
213 {
214 	int newcnt = ndkwedges + 16;
215 	struct dkwedge_softc **newarray, **oldarray;
216 
217 	newarray = malloc(newcnt * sizeof(*newarray), M_DKWEDGE,
218 	    M_WAITOK|M_ZERO);
219 	if ((oldarray = dkwedges) != NULL)
220 		memcpy(newarray, dkwedges, ndkwedges * sizeof(*newarray));
221 	dkwedges = newarray;
222 	ndkwedges = newcnt;
223 	if (oldarray != NULL)
224 		free(oldarray, M_DKWEDGE);
225 }
226 
227 static void
228 dkgetproperties(struct disk *disk, struct dkwedge_info *dkw)
229 {
230 	prop_dictionary_t disk_info, odisk_info, geom;
231 
232 	disk_info = prop_dictionary_create();
233 
234 	prop_dictionary_set_cstring_nocopy(disk_info, "type", "ESDI");
235 
236 	geom = prop_dictionary_create();
237 
238 	prop_dictionary_set_uint64(geom, "sectors-per-unit", dkw->dkw_size);
239 
240 	prop_dictionary_set_uint32(geom, "sector-size",
241 	    DEV_BSIZE /* XXX 512? */);
242 
243 	prop_dictionary_set_uint32(geom, "sectors-per-track", 32);
244 
245 	prop_dictionary_set_uint32(geom, "tracks-per-cylinder", 64);
246 
247 	prop_dictionary_set_uint32(geom, "cylinders-per-unit", dkw->dkw_size / 2048);
248 
249 	prop_dictionary_set(disk_info, "geometry", geom);
250 	prop_object_release(geom);
251 
252 	odisk_info = disk->dk_info;
253 
254 	disk->dk_info = disk_info;
255 
256 	if (odisk_info != NULL)
257 		prop_object_release(odisk_info);
258 }
259 
260 /*
261  * dkwedge_add:		[exported function]
262  *
263  *	Add a disk wedge based on the provided information.
264  *
265  *	The incoming dkw_devname[] is ignored, instead being
266  *	filled in and returned to the caller.
267  */
268 int
269 dkwedge_add(struct dkwedge_info *dkw)
270 {
271 	struct dkwedge_softc *sc, *lsc;
272 	struct disk *pdk;
273 	u_int unit;
274 	int error;
275 	dev_t pdev;
276 
277 	dkw->dkw_parent[sizeof(dkw->dkw_parent) - 1] = '\0';
278 	pdk = disk_find(dkw->dkw_parent);
279 	if (pdk == NULL)
280 		return (ENODEV);
281 
282 	error = dkwedge_compute_pdev(pdk->dk_name, &pdev);
283 	if (error)
284 		return (error);
285 
286 	if (dkw->dkw_offset < 0)
287 		return (EINVAL);
288 
289 	sc = malloc(sizeof(*sc), M_DKWEDGE, M_WAITOK|M_ZERO);
290 	sc->sc_state = DKW_STATE_LARVAL;
291 	sc->sc_parent = pdk;
292 	sc->sc_pdev = pdev;
293 	sc->sc_offset = dkw->dkw_offset;
294 	sc->sc_size = dkw->dkw_size;
295 
296 	memcpy(sc->sc_wname, dkw->dkw_wname, sizeof(sc->sc_wname));
297 	sc->sc_wname[sizeof(sc->sc_wname) - 1] = '\0';
298 
299 	memcpy(sc->sc_ptype, dkw->dkw_ptype, sizeof(sc->sc_ptype));
300 	sc->sc_ptype[sizeof(sc->sc_ptype) - 1] = '\0';
301 
302 	bufq_alloc(&sc->sc_bufq, "fcfs", 0);
303 
304 	callout_init(&sc->sc_restart_ch, 0);
305 	callout_setfunc(&sc->sc_restart_ch, dkrestart, sc);
306 
307 	/*
308 	 * Wedge will be added; increment the wedge count for the parent.
309 	 * Only allow this to happend if RAW_PART is the only thing open.
310 	 */
311 	mutex_enter(&pdk->dk_openlock);
312 	if (pdk->dk_openmask & ~(1 << RAW_PART))
313 		error = EBUSY;
314 	else {
315 		/* Check for wedge overlap. */
316 		LIST_FOREACH(lsc, &pdk->dk_wedges, sc_plink) {
317 			daddr_t lastblk = sc->sc_offset + sc->sc_size - 1;
318 			daddr_t llastblk = lsc->sc_offset + lsc->sc_size - 1;
319 
320 			if (sc->sc_offset >= lsc->sc_offset &&
321 			    sc->sc_offset <= llastblk) {
322 				/* Overlaps the tail of the exsiting wedge. */
323 				break;
324 			}
325 			if (lastblk >= lsc->sc_offset &&
326 			    lastblk <= llastblk) {
327 				/* Overlaps the head of the existing wedge. */
328 			    	break;
329 			}
330 		}
331 		if (lsc != NULL)
332 			error = EINVAL;
333 		else {
334 			pdk->dk_nwedges++;
335 			LIST_INSERT_HEAD(&pdk->dk_wedges, sc, sc_plink);
336 		}
337 	}
338 	mutex_exit(&pdk->dk_openlock);
339 	if (error) {
340 		bufq_free(sc->sc_bufq);
341 		free(sc, M_DKWEDGE);
342 		return (error);
343 	}
344 
345 	/* Fill in our cfdata for the pseudo-device glue. */
346 	sc->sc_cfdata.cf_name = dk_cd.cd_name;
347 	sc->sc_cfdata.cf_atname = dk_ca.ca_name;
348 	/* sc->sc_cfdata.cf_unit set below */
349 	sc->sc_cfdata.cf_fstate = FSTATE_STAR;
350 
351 	/* Insert the larval wedge into the array. */
352 	rw_enter(&dkwedges_lock, RW_WRITER);
353 	for (error = 0;;) {
354 		struct dkwedge_softc **scpp;
355 
356 		/*
357 		 * Check for a duplicate wname while searching for
358 		 * a slot.
359 		 */
360 		for (scpp = NULL, unit = 0; unit < ndkwedges; unit++) {
361 			if (dkwedges[unit] == NULL) {
362 				if (scpp == NULL) {
363 					scpp = &dkwedges[unit];
364 					sc->sc_cfdata.cf_unit = unit;
365 				}
366 			} else {
367 				/* XXX Unicode. */
368 				if (strcmp(dkwedges[unit]->sc_wname,
369 					   sc->sc_wname) == 0) {
370 					error = EEXIST;
371 					break;
372 				}
373 			}
374 		}
375 		if (error)
376 			break;
377 		KASSERT(unit == ndkwedges);
378 		if (scpp == NULL)
379 			dkwedge_array_expand();
380 		else {
381 			KASSERT(scpp == &dkwedges[sc->sc_cfdata.cf_unit]);
382 			*scpp = sc;
383 			break;
384 		}
385 	}
386 	rw_exit(&dkwedges_lock);
387 	if (error) {
388 		mutex_enter(&pdk->dk_openlock);
389 		pdk->dk_nwedges--;
390 		LIST_REMOVE(sc, sc_plink);
391 		mutex_exit(&pdk->dk_openlock);
392 
393 		bufq_free(sc->sc_bufq);
394 		free(sc, M_DKWEDGE);
395 		return (error);
396 	}
397 
398 	/*
399 	 * Now that we know the unit #, attach a pseudo-device for
400 	 * this wedge instance.  This will provide us with the
401 	 * "struct device" necessary for glue to other parts of the
402 	 * system.
403 	 *
404 	 * This should never fail, unless we're almost totally out of
405 	 * memory.
406 	 */
407 	if ((sc->sc_dev = config_attach_pseudo(&sc->sc_cfdata)) == NULL) {
408 		aprint_error("%s%u: unable to attach pseudo-device\n",
409 		    sc->sc_cfdata.cf_name, sc->sc_cfdata.cf_unit);
410 
411 		rw_enter(&dkwedges_lock, RW_WRITER);
412 		dkwedges[sc->sc_cfdata.cf_unit] = NULL;
413 		rw_exit(&dkwedges_lock);
414 
415 		mutex_enter(&pdk->dk_openlock);
416 		pdk->dk_nwedges--;
417 		LIST_REMOVE(sc, sc_plink);
418 		mutex_exit(&pdk->dk_openlock);
419 
420 		bufq_free(sc->sc_bufq);
421 		free(sc, M_DKWEDGE);
422 		return (ENOMEM);
423 	}
424 
425 	/* Return the devname to the caller. */
426 	strlcpy(dkw->dkw_devname, device_xname(sc->sc_dev),
427 		sizeof(dkw->dkw_devname));
428 
429 	/*
430 	 * XXX Really ought to make the disk_attach() and the changing
431 	 * of state to RUNNING atomic.
432 	 */
433 
434 	disk_init(&sc->sc_dk, device_xname(sc->sc_dev), NULL);
435 	dkgetproperties(&sc->sc_dk, dkw);
436 	disk_attach(&sc->sc_dk);
437 
438 	/* Disk wedge is ready for use! */
439 	sc->sc_state = DKW_STATE_RUNNING;
440 
441 	/* Announce our arrival. */
442 	aprint_normal("%s at %s: %s\n", device_xname(sc->sc_dev), pdk->dk_name,
443 	    sc->sc_wname);	/* XXX Unicode */
444 	aprint_normal("%s: %"PRIu64" blocks at %"PRId64", type: %s\n",
445 	    device_xname(sc->sc_dev), sc->sc_size, sc->sc_offset, sc->sc_ptype);
446 
447 	return (0);
448 }
449 
450 /*
451  * dkwedge_find:
452  *
453  *	Lookup a disk wedge based on the provided information.
454  *	NOTE: We look up the wedge based on the wedge devname,
455  *	not wname.
456  *
457  *	Return NULL if the wedge is not found, otherwise return
458  *	the wedge's softc.  Assign the wedge's unit number to unitp
459  *	if unitp is not NULL.
460  */
461 static struct dkwedge_softc *
462 dkwedge_find(struct dkwedge_info *dkw, u_int *unitp)
463 {
464 	struct dkwedge_softc *sc = NULL;
465 	u_int unit;
466 
467 	/* Find our softc. */
468 	dkw->dkw_devname[sizeof(dkw->dkw_devname) - 1] = '\0';
469 	rw_enter(&dkwedges_lock, RW_READER);
470 	for (unit = 0; unit < ndkwedges; unit++) {
471 		if ((sc = dkwedges[unit]) != NULL &&
472 		    strcmp(device_xname(sc->sc_dev), dkw->dkw_devname) == 0 &&
473 		    strcmp(sc->sc_parent->dk_name, dkw->dkw_parent) == 0) {
474 			break;
475 		}
476 	}
477 	rw_exit(&dkwedges_lock);
478 	if (unit == ndkwedges)
479 		return NULL;
480 
481 	if (unitp != NULL)
482 		*unitp = unit;
483 
484 	return sc;
485 }
486 
487 /*
488  * dkwedge_del:		[exported function]
489  *
490  *	Delete a disk wedge based on the provided information.
491  *	NOTE: We look up the wedge based on the wedge devname,
492  *	not wname.
493  */
494 int
495 dkwedge_del(struct dkwedge_info *dkw)
496 {
497 	struct dkwedge_softc *sc = NULL;
498 
499 	/* Find our softc. */
500 	if ((sc = dkwedge_find(dkw, NULL)) == NULL)
501 		return (ESRCH);
502 
503 	return config_detach(sc->sc_dev, DETACH_FORCE | DETACH_QUIET);
504 }
505 
506 static int
507 dkwedge_begindetach(struct dkwedge_softc *sc, int flags)
508 {
509 	struct disk *dk = &sc->sc_dk;
510 	int rc;
511 
512 	rc = 0;
513 	mutex_enter(&dk->dk_openlock);
514 	if (dk->dk_openmask == 0)
515 		;	/* nothing to do */
516 	else if ((flags & DETACH_FORCE) == 0)
517 		rc = EBUSY;
518 	else {
519 		mutex_enter(&sc->sc_parent->dk_rawlock);
520 		rc = dklastclose(sc); /* releases dk_rawlock */
521 	}
522 	mutex_exit(&dk->dk_openlock);
523 
524 	return rc;
525 }
526 
527 /*
528  * dkwedge_detach:
529  *
530  *	Autoconfiguration detach function for pseudo-device glue.
531  */
532 static int
533 dkwedge_detach(device_t self, int flags)
534 {
535 	struct dkwedge_softc *sc = NULL;
536 	u_int unit;
537 	int bmaj, cmaj, rc, s;
538 
539 	rw_enter(&dkwedges_lock, RW_WRITER);
540 	for (unit = 0; unit < ndkwedges; unit++) {
541 		if ((sc = dkwedges[unit]) != NULL && sc->sc_dev == self)
542 			break;
543 	}
544 	if (unit == ndkwedges)
545 		rc = ENXIO;
546 	else if ((rc = dkwedge_begindetach(sc, flags)) == 0) {
547 		/* Mark the wedge as dying. */
548 		sc->sc_state = DKW_STATE_DYING;
549 	}
550 	rw_exit(&dkwedges_lock);
551 
552 	if (rc != 0)
553 		return rc;
554 
555 	pmf_device_deregister(self);
556 
557 	/* Locate the wedge major numbers. */
558 	bmaj = bdevsw_lookup_major(&dk_bdevsw);
559 	cmaj = cdevsw_lookup_major(&dk_cdevsw);
560 
561 	/* Kill any pending restart. */
562 	callout_stop(&sc->sc_restart_ch);
563 
564 	/*
565 	 * dkstart() will kill any queued buffers now that the
566 	 * state of the wedge is not RUNNING.  Once we've done
567 	 * that, wait for any other pending I/O to complete.
568 	 */
569 	s = splbio();
570 	dkstart(sc);
571 	dkwedge_wait_drain(sc);
572 	splx(s);
573 
574 	/* Nuke the vnodes for any open instances. */
575 	vdevgone(bmaj, unit, unit, VBLK);
576 	vdevgone(cmaj, unit, unit, VCHR);
577 
578 	/* Clean up the parent. */
579 	mutex_enter(&sc->sc_dk.dk_openlock);
580 	if (sc->sc_dk.dk_openmask) {
581 		mutex_enter(&sc->sc_parent->dk_rawlock);
582 		if (sc->sc_parent->dk_rawopens-- == 1) {
583 			KASSERT(sc->sc_parent->dk_rawvp != NULL);
584 			mutex_exit(&sc->sc_parent->dk_rawlock);
585 			(void) vn_close(sc->sc_parent->dk_rawvp, FREAD | FWRITE,
586 			    NOCRED);
587 			sc->sc_parent->dk_rawvp = NULL;
588 		} else
589 			mutex_exit(&sc->sc_parent->dk_rawlock);
590 		sc->sc_dk.dk_openmask = 0;
591 	}
592 	mutex_exit(&sc->sc_dk.dk_openlock);
593 
594 	/* Announce our departure. */
595 	aprint_normal("%s at %s (%s) deleted\n", device_xname(sc->sc_dev),
596 	    sc->sc_parent->dk_name,
597 	    sc->sc_wname);	/* XXX Unicode */
598 
599 	mutex_enter(&sc->sc_parent->dk_openlock);
600 	sc->sc_parent->dk_nwedges--;
601 	LIST_REMOVE(sc, sc_plink);
602 	mutex_exit(&sc->sc_parent->dk_openlock);
603 
604 	/* Delete our buffer queue. */
605 	bufq_free(sc->sc_bufq);
606 
607 	/* Detach from the disk list. */
608 	disk_detach(&sc->sc_dk);
609 	disk_destroy(&sc->sc_dk);
610 
611 	/* Poof. */
612 	rw_enter(&dkwedges_lock, RW_WRITER);
613 	dkwedges[unit] = NULL;
614 	sc->sc_state = DKW_STATE_DEAD;
615 	rw_exit(&dkwedges_lock);
616 
617 	free(sc, M_DKWEDGE);
618 
619 	return 0;
620 }
621 
622 /*
623  * dkwedge_delall:	[exported function]
624  *
625  *	Delete all of the wedges on the specified disk.  Used when
626  *	a disk is being detached.
627  */
628 void
629 dkwedge_delall(struct disk *pdk)
630 {
631 	struct dkwedge_info dkw;
632 	struct dkwedge_softc *sc;
633 
634 	for (;;) {
635 		mutex_enter(&pdk->dk_openlock);
636 		if ((sc = LIST_FIRST(&pdk->dk_wedges)) == NULL) {
637 			KASSERT(pdk->dk_nwedges == 0);
638 			mutex_exit(&pdk->dk_openlock);
639 			return;
640 		}
641 		strcpy(dkw.dkw_parent, pdk->dk_name);
642 		strlcpy(dkw.dkw_devname, device_xname(sc->sc_dev),
643 			sizeof(dkw.dkw_devname));
644 		mutex_exit(&pdk->dk_openlock);
645 		(void) dkwedge_del(&dkw);
646 	}
647 }
648 
649 /*
650  * dkwedge_list:	[exported function]
651  *
652  *	List all of the wedges on a particular disk.
653  *	If p == NULL, the buffer is in kernel space.  Otherwise, it is
654  *	in user space of the specified process.
655  */
656 int
657 dkwedge_list(struct disk *pdk, struct dkwedge_list *dkwl, struct lwp *l)
658 {
659 	struct uio uio;
660 	struct iovec iov;
661 	struct dkwedge_softc *sc;
662 	struct dkwedge_info dkw;
663 	int error = 0;
664 
665 	iov.iov_base = dkwl->dkwl_buf;
666 	iov.iov_len = dkwl->dkwl_bufsize;
667 
668 	uio.uio_iov = &iov;
669 	uio.uio_iovcnt = 1;
670 	uio.uio_offset = 0;
671 	uio.uio_resid = dkwl->dkwl_bufsize;
672 	uio.uio_rw = UIO_READ;
673 	KASSERT(l == curlwp);
674 	uio.uio_vmspace = l->l_proc->p_vmspace;
675 
676 	dkwl->dkwl_ncopied = 0;
677 
678 	mutex_enter(&pdk->dk_openlock);
679 	LIST_FOREACH(sc, &pdk->dk_wedges, sc_plink) {
680 		if (uio.uio_resid < sizeof(dkw))
681 			break;
682 
683 		if (sc->sc_state != DKW_STATE_RUNNING)
684 			continue;
685 
686 		strlcpy(dkw.dkw_devname, device_xname(sc->sc_dev),
687 			sizeof(dkw.dkw_devname));
688 		memcpy(dkw.dkw_wname, sc->sc_wname, sizeof(dkw.dkw_wname));
689 		dkw.dkw_wname[sizeof(dkw.dkw_wname) - 1] = '\0';
690 		strcpy(dkw.dkw_parent, sc->sc_parent->dk_name);
691 		dkw.dkw_offset = sc->sc_offset;
692 		dkw.dkw_size = sc->sc_size;
693 		strcpy(dkw.dkw_ptype, sc->sc_ptype);
694 
695 		error = uiomove(&dkw, sizeof(dkw), &uio);
696 		if (error)
697 			break;
698 		dkwl->dkwl_ncopied++;
699 	}
700 	dkwl->dkwl_nwedges = pdk->dk_nwedges;
701 	mutex_exit(&pdk->dk_openlock);
702 
703 	return (error);
704 }
705 
706 device_t
707 dkwedge_find_by_wname(const char *wname)
708 {
709 	device_t dv = NULL;
710 	struct dkwedge_softc *sc;
711 	int i;
712 
713 	rw_enter(&dkwedges_lock, RW_WRITER);
714 	for (i = 0; i < ndkwedges; i++) {
715 		if ((sc = dkwedges[i]) == NULL)
716 			continue;
717 		if (strcmp(sc->sc_wname, wname) == 0) {
718 			if (dv != NULL) {
719 				printf(
720 				    "WARNING: double match for wedge name %s "
721 				    "(%s, %s)\n", wname, device_xname(dv),
722 				    device_xname(sc->sc_dev));
723 				continue;
724 			}
725 			dv = sc->sc_dev;
726 		}
727 	}
728 	rw_exit(&dkwedges_lock);
729 	return dv;
730 }
731 
732 void
733 dkwedge_print_wnames(void)
734 {
735 	struct dkwedge_softc *sc;
736 	int i;
737 
738 	rw_enter(&dkwedges_lock, RW_WRITER);
739 	for (i = 0; i < ndkwedges; i++) {
740 		if ((sc = dkwedges[i]) == NULL)
741 			continue;
742 		printf(" wedge:%s", sc->sc_wname);
743 	}
744 	rw_exit(&dkwedges_lock);
745 }
746 
747 /*
748  * dkwedge_set_bootwedge
749  *
750  *	Set the booted_wedge global based on the specified parent name
751  *	and offset/length.
752  */
753 void
754 dkwedge_set_bootwedge(device_t parent, daddr_t startblk, uint64_t nblks)
755 {
756 	struct dkwedge_softc *sc;
757 	int i;
758 
759 	rw_enter(&dkwedges_lock, RW_WRITER);
760 	for (i = 0; i < ndkwedges; i++) {
761 		if ((sc = dkwedges[i]) == NULL)
762 			continue;
763 		if (strcmp(sc->sc_parent->dk_name, device_xname(parent)) == 0 &&
764 		    sc->sc_offset == startblk &&
765 		    sc->sc_size == nblks) {
766 			if (booted_wedge) {
767 				printf("WARNING: double match for boot wedge "
768 				    "(%s, %s)\n",
769 				    device_xname(booted_wedge),
770 				    device_xname(sc->sc_dev));
771 				continue;
772 			}
773 			booted_device = parent;
774 			booted_wedge = sc->sc_dev;
775 			booted_partition = 0;
776 		}
777 	}
778 	/*
779 	 * XXX What if we don't find one?  Should we create a special
780 	 * XXX root wedge?
781 	 */
782 	rw_exit(&dkwedges_lock);
783 }
784 
785 /*
786  * We need a dummy object to stuff into the dkwedge discovery method link
787  * set to ensure that there is always at least one object in the set.
788  */
789 static struct dkwedge_discovery_method dummy_discovery_method;
790 __link_set_add_bss(dkwedge_methods, dummy_discovery_method);
791 
792 /*
793  * dkwedge_init:
794  *
795  *	Initialize the disk wedge subsystem.
796  */
797 void
798 dkwedge_init(void)
799 {
800 	__link_set_decl(dkwedge_methods, struct dkwedge_discovery_method);
801 	struct dkwedge_discovery_method * const *ddmp;
802 	struct dkwedge_discovery_method *lddm, *ddm;
803 
804 	rw_init(&dkwedges_lock);
805 	rw_init(&dkwedge_discovery_methods_lock);
806 
807 	if (config_cfdriver_attach(&dk_cd) != 0)
808 		panic("dkwedge: unable to attach cfdriver");
809 	if (config_cfattach_attach(dk_cd.cd_name, &dk_ca) != 0)
810 		panic("dkwedge: unable to attach cfattach");
811 
812 	rw_enter(&dkwedge_discovery_methods_lock, RW_WRITER);
813 
814 	LIST_INIT(&dkwedge_discovery_methods);
815 
816 	__link_set_foreach(ddmp, dkwedge_methods) {
817 		ddm = *ddmp;
818 		if (ddm == &dummy_discovery_method)
819 			continue;
820 		if (LIST_EMPTY(&dkwedge_discovery_methods)) {
821 			LIST_INSERT_HEAD(&dkwedge_discovery_methods,
822 					 ddm, ddm_list);
823 			continue;
824 		}
825 		LIST_FOREACH(lddm, &dkwedge_discovery_methods, ddm_list) {
826 			if (ddm->ddm_priority == lddm->ddm_priority) {
827 				aprint_error("dk-method-%s: method \"%s\" "
828 				    "already exists at priority %d\n",
829 				    ddm->ddm_name, lddm->ddm_name,
830 				    lddm->ddm_priority);
831 				/* Not inserted. */
832 				break;
833 			}
834 			if (ddm->ddm_priority < lddm->ddm_priority) {
835 				/* Higher priority; insert before. */
836 				LIST_INSERT_BEFORE(lddm, ddm, ddm_list);
837 				break;
838 			}
839 			if (LIST_NEXT(lddm, ddm_list) == NULL) {
840 				/* Last one; insert after. */
841 				KASSERT(lddm->ddm_priority < ddm->ddm_priority);
842 				LIST_INSERT_AFTER(lddm, ddm, ddm_list);
843 				break;
844 			}
845 		}
846 	}
847 
848 	rw_exit(&dkwedge_discovery_methods_lock);
849 }
850 
851 #ifdef DKWEDGE_AUTODISCOVER
852 int	dkwedge_autodiscover = 1;
853 #else
854 int	dkwedge_autodiscover = 0;
855 #endif
856 
857 /*
858  * dkwedge_discover:	[exported function]
859  *
860  *	Discover the wedges on a newly attached disk.
861  */
862 void
863 dkwedge_discover(struct disk *pdk)
864 {
865 	struct dkwedge_discovery_method *ddm;
866 	struct vnode *vp;
867 	int error;
868 	dev_t pdev;
869 
870 	/*
871 	 * Require people playing with wedges to enable this explicitly.
872 	 */
873 	if (dkwedge_autodiscover == 0)
874 		return;
875 
876 	rw_enter(&dkwedge_discovery_methods_lock, RW_READER);
877 
878 	error = dkwedge_compute_pdev(pdk->dk_name, &pdev);
879 	if (error) {
880 		aprint_error("%s: unable to compute pdev, error = %d\n",
881 		    pdk->dk_name, error);
882 		goto out;
883 	}
884 
885 	error = bdevvp(pdev, &vp);
886 	if (error) {
887 		aprint_error("%s: unable to find vnode for pdev, error = %d\n",
888 		    pdk->dk_name, error);
889 		goto out;
890 	}
891 
892 	error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
893 	if (error) {
894 		aprint_error("%s: unable to lock vnode for pdev, error = %d\n",
895 		    pdk->dk_name, error);
896 		vrele(vp);
897 		goto out;
898 	}
899 
900 	error = VOP_OPEN(vp, FREAD, NOCRED);
901 	if (error) {
902 		aprint_error("%s: unable to open device, error = %d\n",
903 		    pdk->dk_name, error);
904 		vput(vp);
905 		goto out;
906 	}
907 	VOP_UNLOCK(vp);
908 
909 	/*
910 	 * For each supported partition map type, look to see if
911 	 * this map type exists.  If so, parse it and add the
912 	 * corresponding wedges.
913 	 */
914 	LIST_FOREACH(ddm, &dkwedge_discovery_methods, ddm_list) {
915 		error = (*ddm->ddm_discover)(pdk, vp);
916 		if (error == 0) {
917 			/* Successfully created wedges; we're done. */
918 			break;
919 		}
920 	}
921 
922 	error = vn_close(vp, FREAD, NOCRED);
923 	if (error) {
924 		aprint_error("%s: unable to close device, error = %d\n",
925 		    pdk->dk_name, error);
926 		/* We'll just assume the vnode has been cleaned up. */
927 	}
928  out:
929 	rw_exit(&dkwedge_discovery_methods_lock);
930 }
931 
932 /*
933  * dkwedge_read:
934  *
935  *	Read some data from the specified disk, used for
936  *	partition discovery.
937  */
938 int
939 dkwedge_read(struct disk *pdk, struct vnode *vp, daddr_t blkno,
940     void *tbuf, size_t len)
941 {
942 	struct buf *bp;
943 	int result;
944 
945 	bp = getiobuf(vp, true);
946 
947 	bp->b_dev = vp->v_rdev;
948 	bp->b_blkno = blkno;
949 	bp->b_bcount = len;
950 	bp->b_resid = len;
951 	bp->b_flags = B_READ;
952 	bp->b_data = tbuf;
953 	SET(bp->b_cflags, BC_BUSY);	/* mark buffer busy */
954 
955 	VOP_STRATEGY(vp, bp);
956 	result = biowait(bp);
957 	putiobuf(bp);
958 
959 	return result;
960 }
961 
962 /*
963  * dkwedge_lookup:
964  *
965  *	Look up a dkwedge_softc based on the provided dev_t.
966  */
967 static struct dkwedge_softc *
968 dkwedge_lookup(dev_t dev)
969 {
970 	int unit = minor(dev);
971 
972 	if (unit >= ndkwedges)
973 		return (NULL);
974 
975 	KASSERT(dkwedges != NULL);
976 
977 	return (dkwedges[unit]);
978 }
979 
980 /*
981  * dkopen:		[devsw entry point]
982  *
983  *	Open a wedge.
984  */
985 static int
986 dkopen(dev_t dev, int flags, int fmt, struct lwp *l)
987 {
988 	struct dkwedge_softc *sc = dkwedge_lookup(dev);
989 	struct vnode *vp;
990 	int error = 0;
991 
992 	if (sc == NULL)
993 		return (ENODEV);
994 
995 	if (sc->sc_state != DKW_STATE_RUNNING)
996 		return (ENXIO);
997 
998 	/*
999 	 * We go through a complicated little dance to only open the parent
1000 	 * vnode once per wedge, no matter how many times the wedge is
1001 	 * opened.  The reason?  We see one dkopen() per open call, but
1002 	 * only dkclose() on the last close.
1003 	 */
1004 	mutex_enter(&sc->sc_dk.dk_openlock);
1005 	mutex_enter(&sc->sc_parent->dk_rawlock);
1006 	if (sc->sc_dk.dk_openmask == 0) {
1007 		if (sc->sc_parent->dk_rawopens == 0) {
1008 			KASSERT(sc->sc_parent->dk_rawvp == NULL);
1009 			error = bdevvp(sc->sc_pdev, &vp);
1010 			if (error)
1011 				goto popen_fail;
1012 			error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1013 			if (error) {
1014 				vrele(vp);
1015 				goto popen_fail;
1016 			}
1017 			error = VOP_OPEN(vp, FREAD | FWRITE, NOCRED);
1018 			if (error) {
1019 				vput(vp);
1020 				goto popen_fail;
1021 			}
1022 			/* VOP_OPEN() doesn't do this for us. */
1023 			mutex_enter(&vp->v_interlock);
1024 			vp->v_writecount++;
1025 			mutex_exit(&vp->v_interlock);
1026 			VOP_UNLOCK(vp);
1027 			sc->sc_parent->dk_rawvp = vp;
1028 		}
1029 		sc->sc_parent->dk_rawopens++;
1030 	}
1031 	if (fmt == S_IFCHR)
1032 		sc->sc_dk.dk_copenmask |= 1;
1033 	else
1034 		sc->sc_dk.dk_bopenmask |= 1;
1035 	sc->sc_dk.dk_openmask =
1036 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
1037 
1038  popen_fail:
1039 	mutex_exit(&sc->sc_parent->dk_rawlock);
1040 	mutex_exit(&sc->sc_dk.dk_openlock);
1041 	return (error);
1042 }
1043 
1044 /*
1045  * Caller must hold sc->sc_dk.dk_openlock and sc->sc_parent->dk_rawlock.
1046  */
1047 static int
1048 dklastclose(struct dkwedge_softc *sc)
1049 {
1050 	int error = 0;
1051 
1052 	if (sc->sc_parent->dk_rawopens-- == 1) {
1053 		KASSERT(sc->sc_parent->dk_rawvp != NULL);
1054 		mutex_exit(&sc->sc_parent->dk_rawlock);
1055 		error = vn_close(sc->sc_parent->dk_rawvp,
1056 		    FREAD | FWRITE, NOCRED);
1057 		sc->sc_parent->dk_rawvp = NULL;
1058 	} else
1059 		mutex_exit(&sc->sc_parent->dk_rawlock);
1060 	return error;
1061 }
1062 
1063 /*
1064  * dkclose:		[devsw entry point]
1065  *
1066  *	Close a wedge.
1067  */
1068 static int
1069 dkclose(dev_t dev, int flags, int fmt, struct lwp *l)
1070 {
1071 	struct dkwedge_softc *sc = dkwedge_lookup(dev);
1072 	int error = 0;
1073 
1074 	KASSERT(sc->sc_dk.dk_openmask != 0);
1075 
1076 	mutex_enter(&sc->sc_dk.dk_openlock);
1077 	mutex_enter(&sc->sc_parent->dk_rawlock);
1078 
1079 	if (fmt == S_IFCHR)
1080 		sc->sc_dk.dk_copenmask &= ~1;
1081 	else
1082 		sc->sc_dk.dk_bopenmask &= ~1;
1083 	sc->sc_dk.dk_openmask =
1084 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
1085 
1086 	if (sc->sc_dk.dk_openmask == 0)
1087 		error = dklastclose(sc); /* releases dk_rawlock */
1088 	else
1089 		mutex_exit(&sc->sc_parent->dk_rawlock);
1090 
1091 	mutex_exit(&sc->sc_dk.dk_openlock);
1092 
1093 	return (error);
1094 }
1095 
1096 /*
1097  * dkstragegy:		[devsw entry point]
1098  *
1099  *	Perform I/O based on the wedge I/O strategy.
1100  */
1101 static void
1102 dkstrategy(struct buf *bp)
1103 {
1104 	struct dkwedge_softc *sc = dkwedge_lookup(bp->b_dev);
1105 	uint64_t p_size, p_offset;
1106 	int s;
1107 
1108 	if (sc->sc_state != DKW_STATE_RUNNING) {
1109 		bp->b_error = ENXIO;
1110 		goto done;
1111 	}
1112 
1113 	/* If it's an empty transfer, wake up the top half now. */
1114 	if (bp->b_bcount == 0)
1115 		goto done;
1116 
1117 	p_offset = sc->sc_offset << sc->sc_parent->dk_blkshift;
1118 	p_size   = sc->sc_size << sc->sc_parent->dk_blkshift;
1119 
1120 	/* Make sure it's in-range. */
1121 	if (bounds_check_with_mediasize(bp, DEV_BSIZE, p_size) <= 0)
1122 		goto done;
1123 
1124 	/* Translate it to the parent's raw LBA. */
1125 	bp->b_rawblkno = bp->b_blkno + p_offset;
1126 
1127 	/* Place it in the queue and start I/O on the unit. */
1128 	s = splbio();
1129 	sc->sc_iopend++;
1130 	bufq_put(sc->sc_bufq, bp);
1131 	dkstart(sc);
1132 	splx(s);
1133 	return;
1134 
1135  done:
1136 	bp->b_resid = bp->b_bcount;
1137 	biodone(bp);
1138 }
1139 
1140 /*
1141  * dkstart:
1142  *
1143  *	Start I/O that has been enqueued on the wedge.
1144  *	NOTE: Must be called at splbio()!
1145  */
1146 static void
1147 dkstart(struct dkwedge_softc *sc)
1148 {
1149 	struct vnode *vp;
1150 	struct buf *bp, *nbp;
1151 
1152 	/* Do as much work as has been enqueued. */
1153 	while ((bp = bufq_peek(sc->sc_bufq)) != NULL) {
1154 		if (sc->sc_state != DKW_STATE_RUNNING) {
1155 			(void) bufq_get(sc->sc_bufq);
1156 			if (sc->sc_iopend-- == 1 &&
1157 			    (sc->sc_flags & DK_F_WAIT_DRAIN) != 0) {
1158 				sc->sc_flags &= ~DK_F_WAIT_DRAIN;
1159 				wakeup(&sc->sc_iopend);
1160 			}
1161 			bp->b_error = ENXIO;
1162 			bp->b_resid = bp->b_bcount;
1163 			biodone(bp);
1164 		}
1165 
1166 		/* Instrumentation. */
1167 		disk_busy(&sc->sc_dk);
1168 
1169 		nbp = getiobuf(sc->sc_parent->dk_rawvp, false);
1170 		if (nbp == NULL) {
1171 			/*
1172 			 * No resources to run this request; leave the
1173 			 * buffer queued up, and schedule a timer to
1174 			 * restart the queue in 1/2 a second.
1175 			 */
1176 			disk_unbusy(&sc->sc_dk, 0, bp->b_flags & B_READ);
1177 			callout_schedule(&sc->sc_restart_ch, hz / 2);
1178 			return;
1179 		}
1180 
1181 		(void) bufq_get(sc->sc_bufq);
1182 
1183 		nbp->b_data = bp->b_data;
1184 		nbp->b_flags = bp->b_flags;
1185 		nbp->b_oflags = bp->b_oflags;
1186 		nbp->b_cflags = bp->b_cflags;
1187 		nbp->b_iodone = dkiodone;
1188 		nbp->b_proc = bp->b_proc;
1189 		nbp->b_blkno = bp->b_rawblkno;
1190 		nbp->b_dev = sc->sc_parent->dk_rawvp->v_rdev;
1191 		nbp->b_bcount = bp->b_bcount;
1192 		nbp->b_private = bp;
1193 		BIO_COPYPRIO(nbp, bp);
1194 
1195 		vp = nbp->b_vp;
1196 		if ((nbp->b_flags & B_READ) == 0) {
1197 			mutex_enter(&vp->v_interlock);
1198 			vp->v_numoutput++;
1199 			mutex_exit(&vp->v_interlock);
1200 		}
1201 		VOP_STRATEGY(vp, nbp);
1202 	}
1203 }
1204 
1205 /*
1206  * dkiodone:
1207  *
1208  *	I/O to a wedge has completed; alert the top half.
1209  */
1210 static void
1211 dkiodone(struct buf *bp)
1212 {
1213 	struct buf *obp = bp->b_private;
1214 	struct dkwedge_softc *sc = dkwedge_lookup(obp->b_dev);
1215 
1216 	int s = splbio();
1217 
1218 	if (bp->b_error != 0)
1219 		obp->b_error = bp->b_error;
1220 	obp->b_resid = bp->b_resid;
1221 	putiobuf(bp);
1222 
1223 	if (sc->sc_iopend-- == 1 && (sc->sc_flags & DK_F_WAIT_DRAIN) != 0) {
1224 		sc->sc_flags &= ~DK_F_WAIT_DRAIN;
1225 		wakeup(&sc->sc_iopend);
1226 	}
1227 
1228 	disk_unbusy(&sc->sc_dk, obp->b_bcount - obp->b_resid,
1229 	    obp->b_flags & B_READ);
1230 
1231 	biodone(obp);
1232 
1233 	/* Kick the queue in case there is more work we can do. */
1234 	dkstart(sc);
1235 	splx(s);
1236 }
1237 
1238 /*
1239  * dkrestart:
1240  *
1241  *	Restart the work queue after it was stalled due to
1242  *	a resource shortage.  Invoked via a callout.
1243  */
1244 static void
1245 dkrestart(void *v)
1246 {
1247 	struct dkwedge_softc *sc = v;
1248 	int s;
1249 
1250 	s = splbio();
1251 	dkstart(sc);
1252 	splx(s);
1253 }
1254 
1255 /*
1256  * dkminphys:
1257  *
1258  *	Call parent's minphys function.
1259  */
1260 static void
1261 dkminphys(struct buf *bp)
1262 {
1263 	struct dkwedge_softc *sc = dkwedge_lookup(bp->b_dev);
1264 	dev_t dev;
1265 
1266 	dev = bp->b_dev;
1267 	bp->b_dev = sc->sc_pdev;
1268 	(*sc->sc_parent->dk_driver->d_minphys)(bp);
1269 	bp->b_dev = dev;
1270 }
1271 
1272 /*
1273  * dkread:		[devsw entry point]
1274  *
1275  *	Read from a wedge.
1276  */
1277 static int
1278 dkread(dev_t dev, struct uio *uio, int flags)
1279 {
1280 	struct dkwedge_softc *sc = dkwedge_lookup(dev);
1281 
1282 	if (sc->sc_state != DKW_STATE_RUNNING)
1283 		return (ENXIO);
1284 
1285 	return (physio(dkstrategy, NULL, dev, B_READ, dkminphys, uio));
1286 }
1287 
1288 /*
1289  * dkwrite:		[devsw entry point]
1290  *
1291  *	Write to a wedge.
1292  */
1293 static int
1294 dkwrite(dev_t dev, struct uio *uio, int flags)
1295 {
1296 	struct dkwedge_softc *sc = dkwedge_lookup(dev);
1297 
1298 	if (sc->sc_state != DKW_STATE_RUNNING)
1299 		return (ENXIO);
1300 
1301 	return (physio(dkstrategy, NULL, dev, B_WRITE, dkminphys, uio));
1302 }
1303 
1304 /*
1305  * dkioctl:		[devsw entry point]
1306  *
1307  *	Perform an ioctl request on a wedge.
1308  */
1309 static int
1310 dkioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
1311 {
1312 	struct dkwedge_softc *sc = dkwedge_lookup(dev);
1313 	int error = 0;
1314 
1315 	if (sc->sc_state != DKW_STATE_RUNNING)
1316 		return (ENXIO);
1317 
1318 	error = disk_ioctl(&sc->sc_dk, cmd, data, flag, l);
1319 	if (error != EPASSTHROUGH)
1320 		return (error);
1321 
1322 	error = 0;
1323 
1324 	switch (cmd) {
1325 	case DIOCCACHESYNC:
1326 		/*
1327 		 * XXX Do we really need to care about having a writable
1328 		 * file descriptor here?
1329 		 */
1330 		if ((flag & FWRITE) == 0)
1331 			error = EBADF;
1332 		else
1333 			error = VOP_IOCTL(sc->sc_parent->dk_rawvp,
1334 					  cmd, data, flag,
1335 					  l != NULL ? l->l_cred : NOCRED);
1336 		break;
1337 	case DIOCGWEDGEINFO:
1338 	    {
1339 		struct dkwedge_info *dkw = (void *) data;
1340 
1341 		strlcpy(dkw->dkw_devname, device_xname(sc->sc_dev),
1342 			sizeof(dkw->dkw_devname));
1343 	    	memcpy(dkw->dkw_wname, sc->sc_wname, sizeof(dkw->dkw_wname));
1344 		dkw->dkw_wname[sizeof(dkw->dkw_wname) - 1] = '\0';
1345 		strcpy(dkw->dkw_parent, sc->sc_parent->dk_name);
1346 		dkw->dkw_offset = sc->sc_offset;
1347 		dkw->dkw_size = sc->sc_size;
1348 		strcpy(dkw->dkw_ptype, sc->sc_ptype);
1349 
1350 		break;
1351 	    }
1352 
1353 	default:
1354 		error = ENOTTY;
1355 	}
1356 
1357 	return (error);
1358 }
1359 
1360 /*
1361  * dksize:		[devsw entry point]
1362  *
1363  *	Query the size of a wedge for the purpose of performing a dump
1364  *	or for swapping to.
1365  */
1366 static int
1367 dksize(dev_t dev)
1368 {
1369 	struct dkwedge_softc *sc = dkwedge_lookup(dev);
1370 	int rv = -1;
1371 
1372 	if (sc == NULL)
1373 		return (-1);
1374 
1375 	if (sc->sc_state != DKW_STATE_RUNNING)
1376 		return (-1);
1377 
1378 	mutex_enter(&sc->sc_dk.dk_openlock);
1379 	mutex_enter(&sc->sc_parent->dk_rawlock);
1380 
1381 	/* Our content type is static, no need to open the device. */
1382 
1383 	if (strcmp(sc->sc_ptype, DKW_PTYPE_SWAP) == 0) {
1384 		/* Saturate if we are larger than INT_MAX. */
1385 		if (sc->sc_size > INT_MAX)
1386 			rv = INT_MAX;
1387 		else
1388 			rv = (int) sc->sc_size;
1389 	}
1390 
1391 	mutex_exit(&sc->sc_parent->dk_rawlock);
1392 	mutex_exit(&sc->sc_dk.dk_openlock);
1393 
1394 	return (rv);
1395 }
1396 
1397 /*
1398  * dkdump:		[devsw entry point]
1399  *
1400  *	Perform a crash dump to a wedge.
1401  */
1402 static int
1403 dkdump(dev_t dev, daddr_t blkno, void *va, size_t size)
1404 {
1405 	struct dkwedge_softc *sc = dkwedge_lookup(dev);
1406 	const struct bdevsw *bdev;
1407 	int rv = 0;
1408 
1409 	if (sc == NULL)
1410 		return (ENXIO);
1411 
1412 	if (sc->sc_state != DKW_STATE_RUNNING)
1413 		return (ENXIO);
1414 
1415 	mutex_enter(&sc->sc_dk.dk_openlock);
1416 	mutex_enter(&sc->sc_parent->dk_rawlock);
1417 
1418 	/* Our content type is static, no need to open the device. */
1419 
1420 	if (strcmp(sc->sc_ptype, DKW_PTYPE_SWAP) != 0) {
1421 		rv = ENXIO;
1422 		goto out;
1423 	}
1424 	if (size % DEV_BSIZE != 0) {
1425 		rv = EINVAL;
1426 		goto out;
1427 	}
1428 	if (blkno + size / DEV_BSIZE > sc->sc_size) {
1429 		printf("%s: blkno (%" PRIu64 ") + size / DEV_BSIZE (%zu) > "
1430 		    "sc->sc_size (%" PRIu64 ")\n", __func__, blkno,
1431 		    size / DEV_BSIZE, sc->sc_size);
1432 		rv = EINVAL;
1433 		goto out;
1434 	}
1435 
1436 	bdev = bdevsw_lookup(sc->sc_pdev);
1437 	rv = (*bdev->d_dump)(sc->sc_pdev, blkno + sc->sc_offset, va, size);
1438 
1439 out:
1440 	mutex_exit(&sc->sc_parent->dk_rawlock);
1441 	mutex_exit(&sc->sc_dk.dk_openlock);
1442 
1443 	return rv;
1444 }
1445 
1446 /*
1447  * config glue
1448  */
1449 
1450 int
1451 config_handle_wedges(struct device *dv, int par)
1452 {
1453 	struct dkwedge_list wl;
1454 	struct dkwedge_info *wi;
1455 	struct vnode *vn;
1456 	char diskname[16];
1457 	int i, error;
1458 
1459 	if ((vn = opendisk(dv)) == NULL)
1460 		return -1;
1461 
1462 	wl.dkwl_bufsize = sizeof(*wi) * 16;
1463 	wl.dkwl_buf = wi = malloc(wl.dkwl_bufsize, M_TEMP, M_WAITOK);
1464 
1465 	error = VOP_IOCTL(vn, DIOCLWEDGES, &wl, FREAD, NOCRED);
1466 	VOP_CLOSE(vn, FREAD, NOCRED);
1467 	vput(vn);
1468 	if (error) {
1469 #ifdef DEBUG_WEDGE
1470 		printf("%s: List wedges returned %d\n",
1471 		    device_xname(dv), error);
1472 #endif
1473 		free(wi, M_TEMP);
1474 		return -1;
1475 	}
1476 
1477 #ifdef DEBUG_WEDGE
1478 	printf("%s: Returned %u(%u) wedges\n", device_xname(dv),
1479 	    wl.dkwl_nwedges, wl.dkwl_ncopied);
1480 #endif
1481 	snprintf(diskname, sizeof(diskname), "%s%c", device_xname(dv),
1482 	    par + 'a');
1483 
1484 	for (i = 0; i < wl.dkwl_ncopied; i++) {
1485 #ifdef DEBUG_WEDGE
1486 		printf("%s: Looking for %s in %s\n",
1487 		    device_xname(dv), diskname, wi[i].dkw_wname);
1488 #endif
1489 		if (strcmp(wi[i].dkw_wname, diskname) == 0)
1490 			break;
1491 	}
1492 
1493 	if (i == wl.dkwl_ncopied) {
1494 #ifdef DEBUG_WEDGE
1495 		printf("%s: Cannot find wedge with parent %s\n",
1496 		    device_xname(dv), diskname);
1497 #endif
1498 		free(wi, M_TEMP);
1499 		return -1;
1500 	}
1501 
1502 #ifdef DEBUG_WEDGE
1503 	printf("%s: Setting boot wedge %s (%s) at %llu %llu\n",
1504 		device_xname(dv), wi[i].dkw_devname, wi[i].dkw_wname,
1505 		(unsigned long long)wi[i].dkw_offset,
1506 		(unsigned long long)wi[i].dkw_size);
1507 #endif
1508 	dkwedge_set_bootwedge(dv, wi[i].dkw_offset, wi[i].dkw_size);
1509 	free(wi, M_TEMP);
1510 	return 0;
1511 }
1512