xref: /netbsd-src/sys/dev/dkwedge/dk.c (revision 9fb66d812c00ebfb445c0b47dea128f32aa6fe96)
1 /*	$NetBSD: dk.c,v 1.102 2020/10/06 15:05:54 mlelstv 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 1.102 2020/10/06 15:05:54 mlelstv 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 	device_t	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 	kmutex_t	sc_iolock;
90 	kcondvar_t	sc_dkdrn;
91 	u_int		sc_iopend;	/* I/Os pending */
92 	int		sc_flags;	/* flags (sc_iolock) */
93 };
94 
95 #define	DK_F_WAIT_DRAIN		0x0001	/* waiting for I/O to drain */
96 
97 static void	dkstart(struct dkwedge_softc *);
98 static void	dkiodone(struct buf *);
99 static void	dkrestart(void *);
100 static void	dkminphys(struct buf *);
101 
102 static int	dklastclose(struct dkwedge_softc *);
103 static int	dkwedge_cleanup_parent(struct dkwedge_softc *, int);
104 static int	dkwedge_detach(device_t, int);
105 static void	dkwedge_delall1(struct disk *, bool);
106 static int	dkwedge_del1(struct dkwedge_info *, int);
107 static int	dk_open_parent(dev_t, int, struct vnode **);
108 static int	dk_close_parent(struct vnode *, int);
109 
110 static dev_type_open(dkopen);
111 static dev_type_close(dkclose);
112 static dev_type_read(dkread);
113 static dev_type_write(dkwrite);
114 static dev_type_ioctl(dkioctl);
115 static dev_type_strategy(dkstrategy);
116 static dev_type_dump(dkdump);
117 static dev_type_size(dksize);
118 static dev_type_discard(dkdiscard);
119 
120 const struct bdevsw dk_bdevsw = {
121 	.d_open = dkopen,
122 	.d_close = dkclose,
123 	.d_strategy = dkstrategy,
124 	.d_ioctl = dkioctl,
125 	.d_dump = dkdump,
126 	.d_psize = dksize,
127 	.d_discard = dkdiscard,
128 	.d_flag = D_DISK | D_MPSAFE
129 };
130 
131 const struct cdevsw dk_cdevsw = {
132 	.d_open = dkopen,
133 	.d_close = dkclose,
134 	.d_read = dkread,
135 	.d_write = dkwrite,
136 	.d_ioctl = dkioctl,
137 	.d_stop = nostop,
138 	.d_tty = notty,
139 	.d_poll = nopoll,
140 	.d_mmap = nommap,
141 	.d_kqfilter = nokqfilter,
142 	.d_discard = dkdiscard,
143 	.d_flag = D_DISK | D_MPSAFE
144 };
145 
146 static struct dkwedge_softc **dkwedges;
147 static u_int ndkwedges;
148 static krwlock_t dkwedges_lock;
149 
150 static LIST_HEAD(, dkwedge_discovery_method) dkwedge_discovery_methods;
151 static krwlock_t dkwedge_discovery_methods_lock;
152 
153 /*
154  * dkwedge_match:
155  *
156  *	Autoconfiguration match function for pseudo-device glue.
157  */
158 static int
159 dkwedge_match(device_t parent, cfdata_t match,
160     void *aux)
161 {
162 
163 	/* Pseudo-device; always present. */
164 	return (1);
165 }
166 
167 /*
168  * dkwedge_attach:
169  *
170  *	Autoconfiguration attach function for pseudo-device glue.
171  */
172 static void
173 dkwedge_attach(device_t parent, device_t self,
174     void *aux)
175 {
176 
177 	if (!pmf_device_register(self, NULL, NULL))
178 		aprint_error_dev(self, "couldn't establish power handler\n");
179 }
180 
181 CFDRIVER_DECL(dk, DV_DISK, NULL);
182 CFATTACH_DECL3_NEW(dk, 0,
183     dkwedge_match, dkwedge_attach, dkwedge_detach, NULL, NULL, NULL,
184     DVF_DETACH_SHUTDOWN);
185 
186 /*
187  * dkwedge_wait_drain:
188  *
189  *	Wait for I/O on the wedge to drain.
190  */
191 static void
192 dkwedge_wait_drain(struct dkwedge_softc *sc)
193 {
194 
195 	mutex_enter(&sc->sc_iolock);
196 	while (sc->sc_iopend != 0) {
197 		sc->sc_flags |= DK_F_WAIT_DRAIN;
198 		cv_wait(&sc->sc_dkdrn, &sc->sc_iolock);
199 	}
200 	mutex_exit(&sc->sc_iolock);
201 }
202 
203 /*
204  * dkwedge_compute_pdev:
205  *
206  *	Compute the parent disk's dev_t.
207  */
208 static int
209 dkwedge_compute_pdev(const char *pname, dev_t *pdevp, enum vtype type)
210 {
211 	const char *name, *cp;
212 	devmajor_t pmaj;
213 	int punit;
214 	char devname[16];
215 
216 	name = pname;
217 	switch (type) {
218 	case VBLK:
219 		pmaj = devsw_name2blk(name, devname, sizeof(devname));
220 		break;
221 	case VCHR:
222 		pmaj = devsw_name2chr(name, devname, sizeof(devname));
223 		break;
224 	default:
225 		pmaj = NODEVMAJOR;
226 		break;
227 	}
228 	if (pmaj == NODEVMAJOR)
229 		return (ENODEV);
230 
231 	name += strlen(devname);
232 	for (cp = name, punit = 0; *cp >= '0' && *cp <= '9'; cp++)
233 		punit = (punit * 10) + (*cp - '0');
234 	if (cp == name) {
235 		/* Invalid parent disk name. */
236 		return (ENODEV);
237 	}
238 
239 	*pdevp = MAKEDISKDEV(pmaj, punit, RAW_PART);
240 
241 	return (0);
242 }
243 
244 /*
245  * dkwedge_array_expand:
246  *
247  *	Expand the dkwedges array.
248  */
249 static void
250 dkwedge_array_expand(void)
251 {
252 	int newcnt = ndkwedges + 16;
253 	struct dkwedge_softc **newarray, **oldarray;
254 
255 	newarray = malloc(newcnt * sizeof(*newarray), M_DKWEDGE,
256 	    M_WAITOK|M_ZERO);
257 	if ((oldarray = dkwedges) != NULL)
258 		memcpy(newarray, dkwedges, ndkwedges * sizeof(*newarray));
259 	dkwedges = newarray;
260 	ndkwedges = newcnt;
261 	if (oldarray != NULL)
262 		free(oldarray, M_DKWEDGE);
263 }
264 
265 static void
266 dk_set_geometry(struct dkwedge_softc *sc, struct disk *pdk)
267 {
268 	struct disk *dk = &sc->sc_dk;
269 	struct disk_geom *dg = &dk->dk_geom;
270 
271 	memset(dg, 0, sizeof(*dg));
272 
273 	dg->dg_secperunit = sc->sc_size;
274 	dg->dg_secsize = DEV_BSIZE << pdk->dk_blkshift;
275 
276 	/* fake numbers, 1 cylinder is 1 MB with default sector size */
277 	dg->dg_nsectors = 32;
278 	dg->dg_ntracks = 64;
279 	dg->dg_ncylinders = dg->dg_secperunit / (dg->dg_nsectors * dg->dg_ntracks);
280 
281 	disk_set_info(sc->sc_dev, dk, NULL);
282 }
283 
284 /*
285  * dkwedge_add:		[exported function]
286  *
287  *	Add a disk wedge based on the provided information.
288  *
289  *	The incoming dkw_devname[] is ignored, instead being
290  *	filled in and returned to the caller.
291  */
292 int
293 dkwedge_add(struct dkwedge_info *dkw)
294 {
295 	struct dkwedge_softc *sc, *lsc;
296 	struct disk *pdk;
297 	u_int unit;
298 	int error;
299 	dev_t pdev;
300 
301 	dkw->dkw_parent[sizeof(dkw->dkw_parent) - 1] = '\0';
302 	pdk = disk_find(dkw->dkw_parent);
303 	if (pdk == NULL)
304 		return (ENODEV);
305 
306 	error = dkwedge_compute_pdev(pdk->dk_name, &pdev, VBLK);
307 	if (error)
308 		return (error);
309 
310 	if (dkw->dkw_offset < 0)
311 		return (EINVAL);
312 
313 	/*
314 	 * Check for an existing wedge at the same disk offset. Allow
315 	 * updating a wedge if the only change is the size, and the new
316 	 * size is larger than the old.
317 	 */
318 	sc = NULL;
319 	mutex_enter(&pdk->dk_openlock);
320 	LIST_FOREACH(lsc, &pdk->dk_wedges, sc_plink) {
321 		if (lsc->sc_offset != dkw->dkw_offset)
322 			continue;
323 		if (strcmp(lsc->sc_wname, dkw->dkw_wname) != 0)
324 			break;
325 		if (strcmp(lsc->sc_ptype, dkw->dkw_ptype) != 0)
326 			break;
327 		if (lsc->sc_size > dkw->dkw_size)
328 			break;
329 
330 		sc = lsc;
331 		sc->sc_size = dkw->dkw_size;
332 		dk_set_geometry(sc, pdk);
333 
334 		break;
335 	}
336 	mutex_exit(&pdk->dk_openlock);
337 
338 	if (sc != NULL)
339 		goto announce;
340 
341 	sc = malloc(sizeof(*sc), M_DKWEDGE, M_WAITOK|M_ZERO);
342 	sc->sc_state = DKW_STATE_LARVAL;
343 	sc->sc_parent = pdk;
344 	sc->sc_pdev = pdev;
345 	sc->sc_offset = dkw->dkw_offset;
346 	sc->sc_size = dkw->dkw_size;
347 
348 	memcpy(sc->sc_wname, dkw->dkw_wname, sizeof(sc->sc_wname));
349 	sc->sc_wname[sizeof(sc->sc_wname) - 1] = '\0';
350 
351 	memcpy(sc->sc_ptype, dkw->dkw_ptype, sizeof(sc->sc_ptype));
352 	sc->sc_ptype[sizeof(sc->sc_ptype) - 1] = '\0';
353 
354 	bufq_alloc(&sc->sc_bufq, "fcfs", 0);
355 
356 	callout_init(&sc->sc_restart_ch, 0);
357 	callout_setfunc(&sc->sc_restart_ch, dkrestart, sc);
358 
359 	mutex_init(&sc->sc_iolock, MUTEX_DEFAULT, IPL_BIO);
360 	cv_init(&sc->sc_dkdrn, "dkdrn");
361 
362 	/*
363 	 * Wedge will be added; increment the wedge count for the parent.
364 	 * Only allow this to happend if RAW_PART is the only thing open.
365 	 */
366 	mutex_enter(&pdk->dk_openlock);
367 	if (pdk->dk_openmask & ~(1 << RAW_PART))
368 		error = EBUSY;
369 	else {
370 		/* Check for wedge overlap. */
371 		LIST_FOREACH(lsc, &pdk->dk_wedges, sc_plink) {
372 			daddr_t lastblk = sc->sc_offset + sc->sc_size - 1;
373 			daddr_t llastblk = lsc->sc_offset + lsc->sc_size - 1;
374 
375 			if (sc->sc_offset >= lsc->sc_offset &&
376 			    sc->sc_offset <= llastblk) {
377 				/* Overlaps the tail of the existing wedge. */
378 				break;
379 			}
380 			if (lastblk >= lsc->sc_offset &&
381 			    lastblk <= llastblk) {
382 				/* Overlaps the head of the existing wedge. */
383 			    	break;
384 			}
385 		}
386 		if (lsc != NULL) {
387 			if (sc->sc_offset == lsc->sc_offset &&
388 			    sc->sc_size == lsc->sc_size &&
389 			    strcmp(sc->sc_wname, lsc->sc_wname) == 0)
390 				error = EEXIST;
391 			else
392 				error = EINVAL;
393 		} else {
394 			pdk->dk_nwedges++;
395 			LIST_INSERT_HEAD(&pdk->dk_wedges, sc, sc_plink);
396 		}
397 	}
398 	mutex_exit(&pdk->dk_openlock);
399 	if (error) {
400 		cv_destroy(&sc->sc_dkdrn);
401 		mutex_destroy(&sc->sc_iolock);
402 		bufq_free(sc->sc_bufq);
403 		free(sc, M_DKWEDGE);
404 		return (error);
405 	}
406 
407 	/* Fill in our cfdata for the pseudo-device glue. */
408 	sc->sc_cfdata.cf_name = dk_cd.cd_name;
409 	sc->sc_cfdata.cf_atname = dk_ca.ca_name;
410 	/* sc->sc_cfdata.cf_unit set below */
411 	sc->sc_cfdata.cf_fstate = FSTATE_STAR;
412 
413 	/* Insert the larval wedge into the array. */
414 	rw_enter(&dkwedges_lock, RW_WRITER);
415 	for (error = 0;;) {
416 		struct dkwedge_softc **scpp;
417 
418 		/*
419 		 * Check for a duplicate wname while searching for
420 		 * a slot.
421 		 */
422 		for (scpp = NULL, unit = 0; unit < ndkwedges; unit++) {
423 			if (dkwedges[unit] == NULL) {
424 				if (scpp == NULL) {
425 					scpp = &dkwedges[unit];
426 					sc->sc_cfdata.cf_unit = unit;
427 				}
428 			} else {
429 				/* XXX Unicode. */
430 				if (strcmp(dkwedges[unit]->sc_wname,
431 					   sc->sc_wname) == 0) {
432 					error = EEXIST;
433 					break;
434 				}
435 			}
436 		}
437 		if (error)
438 			break;
439 		KASSERT(unit == ndkwedges);
440 		if (scpp == NULL)
441 			dkwedge_array_expand();
442 		else {
443 			KASSERT(scpp == &dkwedges[sc->sc_cfdata.cf_unit]);
444 			*scpp = sc;
445 			break;
446 		}
447 	}
448 	rw_exit(&dkwedges_lock);
449 	if (error) {
450 		mutex_enter(&pdk->dk_openlock);
451 		pdk->dk_nwedges--;
452 		LIST_REMOVE(sc, sc_plink);
453 		mutex_exit(&pdk->dk_openlock);
454 
455 		cv_destroy(&sc->sc_dkdrn);
456 		mutex_destroy(&sc->sc_iolock);
457 		bufq_free(sc->sc_bufq);
458 		free(sc, M_DKWEDGE);
459 		return (error);
460 	}
461 
462 	/*
463 	 * Now that we know the unit #, attach a pseudo-device for
464 	 * this wedge instance.  This will provide us with the
465 	 * device_t necessary for glue to other parts of the system.
466 	 *
467 	 * This should never fail, unless we're almost totally out of
468 	 * memory.
469 	 */
470 	if ((sc->sc_dev = config_attach_pseudo(&sc->sc_cfdata)) == NULL) {
471 		aprint_error("%s%u: unable to attach pseudo-device\n",
472 		    sc->sc_cfdata.cf_name, sc->sc_cfdata.cf_unit);
473 
474 		rw_enter(&dkwedges_lock, RW_WRITER);
475 		dkwedges[sc->sc_cfdata.cf_unit] = NULL;
476 		rw_exit(&dkwedges_lock);
477 
478 		mutex_enter(&pdk->dk_openlock);
479 		pdk->dk_nwedges--;
480 		LIST_REMOVE(sc, sc_plink);
481 		mutex_exit(&pdk->dk_openlock);
482 
483 		cv_destroy(&sc->sc_dkdrn);
484 		mutex_destroy(&sc->sc_iolock);
485 		bufq_free(sc->sc_bufq);
486 		free(sc, M_DKWEDGE);
487 		return (ENOMEM);
488 	}
489 
490 	/* Return the devname to the caller. */
491 	strlcpy(dkw->dkw_devname, device_xname(sc->sc_dev),
492 		sizeof(dkw->dkw_devname));
493 
494 	/*
495 	 * XXX Really ought to make the disk_attach() and the changing
496 	 * of state to RUNNING atomic.
497 	 */
498 
499 	disk_init(&sc->sc_dk, device_xname(sc->sc_dev), NULL);
500 	dk_set_geometry(sc, pdk);
501 	disk_attach(&sc->sc_dk);
502 
503 	/* Disk wedge is ready for use! */
504 	sc->sc_state = DKW_STATE_RUNNING;
505 
506 announce:
507 	/* Announce our arrival. */
508 	aprint_normal(
509 	    "%s at %s: \"%s\", %"PRIu64" blocks at %"PRId64", type: %s\n",
510 	    device_xname(sc->sc_dev), pdk->dk_name,
511 	    sc->sc_wname,	/* XXX Unicode */
512 	    sc->sc_size, sc->sc_offset,
513 	    sc->sc_ptype[0] == '\0' ? "<unknown>" : sc->sc_ptype);
514 
515 	return (0);
516 }
517 
518 /*
519  * dkwedge_find:
520  *
521  *	Lookup a disk wedge based on the provided information.
522  *	NOTE: We look up the wedge based on the wedge devname,
523  *	not wname.
524  *
525  *	Return NULL if the wedge is not found, otherwise return
526  *	the wedge's softc.  Assign the wedge's unit number to unitp
527  *	if unitp is not NULL.
528  */
529 static struct dkwedge_softc *
530 dkwedge_find(struct dkwedge_info *dkw, u_int *unitp)
531 {
532 	struct dkwedge_softc *sc = NULL;
533 	u_int unit;
534 
535 	/* Find our softc. */
536 	dkw->dkw_devname[sizeof(dkw->dkw_devname) - 1] = '\0';
537 	rw_enter(&dkwedges_lock, RW_READER);
538 	for (unit = 0; unit < ndkwedges; unit++) {
539 		if ((sc = dkwedges[unit]) != NULL &&
540 		    strcmp(device_xname(sc->sc_dev), dkw->dkw_devname) == 0 &&
541 		    strcmp(sc->sc_parent->dk_name, dkw->dkw_parent) == 0) {
542 			break;
543 		}
544 	}
545 	rw_exit(&dkwedges_lock);
546 	if (unit == ndkwedges)
547 		return NULL;
548 
549 	if (unitp != NULL)
550 		*unitp = unit;
551 
552 	return sc;
553 }
554 
555 /*
556  * dkwedge_del:		[exported function]
557  *
558  *	Delete a disk wedge based on the provided information.
559  *	NOTE: We look up the wedge based on the wedge devname,
560  *	not wname.
561  */
562 int
563 dkwedge_del(struct dkwedge_info *dkw)
564 {
565 	return dkwedge_del1(dkw, 0);
566 }
567 
568 int
569 dkwedge_del1(struct dkwedge_info *dkw, int flags)
570 {
571 	struct dkwedge_softc *sc = NULL;
572 
573 	/* Find our softc. */
574 	if ((sc = dkwedge_find(dkw, NULL)) == NULL)
575 		return (ESRCH);
576 
577 	return config_detach(sc->sc_dev, flags);
578 }
579 
580 static int
581 dkwedge_cleanup_parent(struct dkwedge_softc *sc, int flags)
582 {
583 	struct disk *dk = &sc->sc_dk;
584 	int rc;
585 
586 	rc = 0;
587 	mutex_enter(&dk->dk_openlock);
588 	if (dk->dk_openmask == 0)
589 		/* nothing to do */
590 		mutex_exit(&dk->dk_openlock);
591 	else if ((flags & DETACH_FORCE) == 0) {
592 		rc = EBUSY;
593 		mutex_exit(&dk->dk_openlock);
594 	}  else {
595 		mutex_enter(&sc->sc_parent->dk_rawlock);
596 		rc = dklastclose(sc); /* releases locks */
597 	}
598 
599 	return rc;
600 }
601 
602 /*
603  * dkwedge_detach:
604  *
605  *	Autoconfiguration detach function for pseudo-device glue.
606  */
607 static int
608 dkwedge_detach(device_t self, int flags)
609 {
610 	struct dkwedge_softc *sc = NULL;
611 	u_int unit;
612 	int bmaj, cmaj, rc;
613 
614 	rw_enter(&dkwedges_lock, RW_WRITER);
615 	for (unit = 0; unit < ndkwedges; unit++) {
616 		if ((sc = dkwedges[unit]) != NULL && sc->sc_dev == self)
617 			break;
618 	}
619 	if (unit == ndkwedges)
620 		rc = ENXIO;
621 	else if ((rc = dkwedge_cleanup_parent(sc, flags)) == 0) {
622 		/* Mark the wedge as dying. */
623 		sc->sc_state = DKW_STATE_DYING;
624 	}
625 	rw_exit(&dkwedges_lock);
626 
627 	if (rc != 0)
628 		return rc;
629 
630 	pmf_device_deregister(self);
631 
632 	/* Locate the wedge major numbers. */
633 	bmaj = bdevsw_lookup_major(&dk_bdevsw);
634 	cmaj = cdevsw_lookup_major(&dk_cdevsw);
635 
636 	/* Kill any pending restart. */
637 	callout_stop(&sc->sc_restart_ch);
638 
639 	/*
640 	 * dkstart() will kill any queued buffers now that the
641 	 * state of the wedge is not RUNNING.  Once we've done
642 	 * that, wait for any other pending I/O to complete.
643 	 */
644 	dkstart(sc);
645 	dkwedge_wait_drain(sc);
646 
647 	/* Nuke the vnodes for any open instances. */
648 	vdevgone(bmaj, unit, unit, VBLK);
649 	vdevgone(cmaj, unit, unit, VCHR);
650 
651 	/* Clean up the parent. */
652 	dkwedge_cleanup_parent(sc, flags | DETACH_FORCE);
653 
654 	/* Announce our departure. */
655 	aprint_normal("%s at %s (%s) deleted\n", device_xname(sc->sc_dev),
656 	    sc->sc_parent->dk_name,
657 	    sc->sc_wname);	/* XXX Unicode */
658 
659 	mutex_enter(&sc->sc_parent->dk_openlock);
660 	sc->sc_parent->dk_nwedges--;
661 	LIST_REMOVE(sc, sc_plink);
662 	mutex_exit(&sc->sc_parent->dk_openlock);
663 
664 	/* Delete our buffer queue. */
665 	bufq_free(sc->sc_bufq);
666 
667 	/* Detach from the disk list. */
668 	disk_detach(&sc->sc_dk);
669 	disk_destroy(&sc->sc_dk);
670 
671 	/* Poof. */
672 	rw_enter(&dkwedges_lock, RW_WRITER);
673 	dkwedges[unit] = NULL;
674 	sc->sc_state = DKW_STATE_DEAD;
675 	rw_exit(&dkwedges_lock);
676 
677 	mutex_destroy(&sc->sc_iolock);
678 	cv_destroy(&sc->sc_dkdrn);
679 
680 	free(sc, M_DKWEDGE);
681 
682 	return 0;
683 }
684 
685 /*
686  * dkwedge_delall:	[exported function]
687  *
688  *	Delete all of the wedges on the specified disk.  Used when
689  *	a disk is being detached.
690  */
691 void
692 dkwedge_delall(struct disk *pdk)
693 {
694 	dkwedge_delall1(pdk, false);
695 }
696 
697 static void
698 dkwedge_delall1(struct disk *pdk, bool idleonly)
699 {
700 	struct dkwedge_info dkw;
701 	struct dkwedge_softc *sc;
702 	int flags;
703 
704 	flags = DETACH_QUIET;
705 	if (!idleonly) flags |= DETACH_FORCE;
706 
707 	for (;;) {
708 		mutex_enter(&pdk->dk_openlock);
709 		LIST_FOREACH(sc, &pdk->dk_wedges, sc_plink) {
710 			if (!idleonly || sc->sc_dk.dk_openmask == 0)
711 				break;
712 		}
713 		if (sc == NULL) {
714 			KASSERT(idleonly || pdk->dk_nwedges == 0);
715 			mutex_exit(&pdk->dk_openlock);
716 			return;
717 		}
718 		strlcpy(dkw.dkw_parent, pdk->dk_name, sizeof(dkw.dkw_parent));
719 		strlcpy(dkw.dkw_devname, device_xname(sc->sc_dev),
720 			sizeof(dkw.dkw_devname));
721 		mutex_exit(&pdk->dk_openlock);
722 		(void) dkwedge_del1(&dkw, flags);
723 	}
724 }
725 
726 /*
727  * dkwedge_list:	[exported function]
728  *
729  *	List all of the wedges on a particular disk.
730  */
731 int
732 dkwedge_list(struct disk *pdk, struct dkwedge_list *dkwl, struct lwp *l)
733 {
734 	struct uio uio;
735 	struct iovec iov;
736 	struct dkwedge_softc *sc;
737 	struct dkwedge_info dkw;
738 	int error = 0;
739 
740 	iov.iov_base = dkwl->dkwl_buf;
741 	iov.iov_len = dkwl->dkwl_bufsize;
742 
743 	uio.uio_iov = &iov;
744 	uio.uio_iovcnt = 1;
745 	uio.uio_offset = 0;
746 	uio.uio_resid = dkwl->dkwl_bufsize;
747 	uio.uio_rw = UIO_READ;
748 	KASSERT(l == curlwp);
749 	uio.uio_vmspace = l->l_proc->p_vmspace;
750 
751 	dkwl->dkwl_ncopied = 0;
752 
753 	mutex_enter(&pdk->dk_openlock);
754 	LIST_FOREACH(sc, &pdk->dk_wedges, sc_plink) {
755 		if (uio.uio_resid < sizeof(dkw))
756 			break;
757 
758 		if (sc->sc_state != DKW_STATE_RUNNING)
759 			continue;
760 
761 		strlcpy(dkw.dkw_devname, device_xname(sc->sc_dev),
762 			sizeof(dkw.dkw_devname));
763 		memcpy(dkw.dkw_wname, sc->sc_wname, sizeof(dkw.dkw_wname));
764 		dkw.dkw_wname[sizeof(dkw.dkw_wname) - 1] = '\0';
765 		strlcpy(dkw.dkw_parent, sc->sc_parent->dk_name,
766 		    sizeof(dkw.dkw_parent));
767 		dkw.dkw_offset = sc->sc_offset;
768 		dkw.dkw_size = sc->sc_size;
769 		strlcpy(dkw.dkw_ptype, sc->sc_ptype, sizeof(dkw.dkw_ptype));
770 
771 		error = uiomove(&dkw, sizeof(dkw), &uio);
772 		if (error)
773 			break;
774 		dkwl->dkwl_ncopied++;
775 	}
776 	dkwl->dkwl_nwedges = pdk->dk_nwedges;
777 	mutex_exit(&pdk->dk_openlock);
778 
779 	return (error);
780 }
781 
782 device_t
783 dkwedge_find_by_wname(const char *wname)
784 {
785 	device_t dv = NULL;
786 	struct dkwedge_softc *sc;
787 	int i;
788 
789 	rw_enter(&dkwedges_lock, RW_WRITER);
790 	for (i = 0; i < ndkwedges; i++) {
791 		if ((sc = dkwedges[i]) == NULL)
792 			continue;
793 		if (strcmp(sc->sc_wname, wname) == 0) {
794 			if (dv != NULL) {
795 				printf(
796 				    "WARNING: double match for wedge name %s "
797 				    "(%s, %s)\n", wname, device_xname(dv),
798 				    device_xname(sc->sc_dev));
799 				continue;
800 			}
801 			dv = sc->sc_dev;
802 		}
803 	}
804 	rw_exit(&dkwedges_lock);
805 	return dv;
806 }
807 
808 device_t
809 dkwedge_find_by_parent(const char *name, size_t *i)
810 {
811 	rw_enter(&dkwedges_lock, RW_WRITER);
812 	for (; *i < (size_t)ndkwedges; (*i)++) {
813 		struct dkwedge_softc *sc;
814 		if ((sc = dkwedges[*i]) == NULL)
815 			continue;
816 		if (strcmp(sc->sc_parent->dk_name, name) != 0)
817 			continue;
818 		rw_exit(&dkwedges_lock);
819 		return sc->sc_dev;
820 	}
821 	rw_exit(&dkwedges_lock);
822 	return NULL;
823 }
824 
825 void
826 dkwedge_print_wnames(void)
827 {
828 	struct dkwedge_softc *sc;
829 	int i;
830 
831 	rw_enter(&dkwedges_lock, RW_WRITER);
832 	for (i = 0; i < ndkwedges; i++) {
833 		if ((sc = dkwedges[i]) == NULL)
834 			continue;
835 		printf(" wedge:%s", sc->sc_wname);
836 	}
837 	rw_exit(&dkwedges_lock);
838 }
839 
840 /*
841  * We need a dummy object to stuff into the dkwedge discovery method link
842  * set to ensure that there is always at least one object in the set.
843  */
844 static struct dkwedge_discovery_method dummy_discovery_method;
845 __link_set_add_bss(dkwedge_methods, dummy_discovery_method);
846 
847 /*
848  * dkwedge_init:
849  *
850  *	Initialize the disk wedge subsystem.
851  */
852 void
853 dkwedge_init(void)
854 {
855 	__link_set_decl(dkwedge_methods, struct dkwedge_discovery_method);
856 	struct dkwedge_discovery_method * const *ddmp;
857 	struct dkwedge_discovery_method *lddm, *ddm;
858 
859 	rw_init(&dkwedges_lock);
860 	rw_init(&dkwedge_discovery_methods_lock);
861 
862 	if (config_cfdriver_attach(&dk_cd) != 0)
863 		panic("dkwedge: unable to attach cfdriver");
864 	if (config_cfattach_attach(dk_cd.cd_name, &dk_ca) != 0)
865 		panic("dkwedge: unable to attach cfattach");
866 
867 	rw_enter(&dkwedge_discovery_methods_lock, RW_WRITER);
868 
869 	LIST_INIT(&dkwedge_discovery_methods);
870 
871 	__link_set_foreach(ddmp, dkwedge_methods) {
872 		ddm = *ddmp;
873 		if (ddm == &dummy_discovery_method)
874 			continue;
875 		if (LIST_EMPTY(&dkwedge_discovery_methods)) {
876 			LIST_INSERT_HEAD(&dkwedge_discovery_methods,
877 					 ddm, ddm_list);
878 			continue;
879 		}
880 		LIST_FOREACH(lddm, &dkwedge_discovery_methods, ddm_list) {
881 			if (ddm->ddm_priority == lddm->ddm_priority) {
882 				aprint_error("dk-method-%s: method \"%s\" "
883 				    "already exists at priority %d\n",
884 				    ddm->ddm_name, lddm->ddm_name,
885 				    lddm->ddm_priority);
886 				/* Not inserted. */
887 				break;
888 			}
889 			if (ddm->ddm_priority < lddm->ddm_priority) {
890 				/* Higher priority; insert before. */
891 				LIST_INSERT_BEFORE(lddm, ddm, ddm_list);
892 				break;
893 			}
894 			if (LIST_NEXT(lddm, ddm_list) == NULL) {
895 				/* Last one; insert after. */
896 				KASSERT(lddm->ddm_priority < ddm->ddm_priority);
897 				LIST_INSERT_AFTER(lddm, ddm, ddm_list);
898 				break;
899 			}
900 		}
901 	}
902 
903 	rw_exit(&dkwedge_discovery_methods_lock);
904 }
905 
906 #ifdef DKWEDGE_AUTODISCOVER
907 int	dkwedge_autodiscover = 1;
908 #else
909 int	dkwedge_autodiscover = 0;
910 #endif
911 
912 /*
913  * dkwedge_discover:	[exported function]
914  *
915  *	Discover the wedges on a newly attached disk.
916  *	Remove all unused wedges on the disk first.
917  */
918 void
919 dkwedge_discover(struct disk *pdk)
920 {
921 	struct dkwedge_discovery_method *ddm;
922 	struct vnode *vp;
923 	int error;
924 	dev_t pdev;
925 
926 	/*
927 	 * Require people playing with wedges to enable this explicitly.
928 	 */
929 	if (dkwedge_autodiscover == 0)
930 		return;
931 
932 	rw_enter(&dkwedge_discovery_methods_lock, RW_READER);
933 
934 	/*
935 	 * Use the character device for scanning, the block device
936 	 * is busy if there are already wedges attached.
937 	 */
938 	error = dkwedge_compute_pdev(pdk->dk_name, &pdev, VCHR);
939 	if (error) {
940 		aprint_error("%s: unable to compute pdev, error = %d\n",
941 		    pdk->dk_name, error);
942 		goto out;
943 	}
944 
945 	error = cdevvp(pdev, &vp);
946 	if (error) {
947 		aprint_error("%s: unable to find vnode for pdev, error = %d\n",
948 		    pdk->dk_name, error);
949 		goto out;
950 	}
951 
952 	error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
953 	if (error) {
954 		aprint_error("%s: unable to lock vnode for pdev, error = %d\n",
955 		    pdk->dk_name, error);
956 		vrele(vp);
957 		goto out;
958 	}
959 
960 	error = VOP_OPEN(vp, FREAD | FSILENT, NOCRED);
961 	if (error) {
962 		if (error != ENODEV)
963 			aprint_error("%s: unable to open device, error = %d\n",
964 			    pdk->dk_name, error);
965 		vput(vp);
966 		goto out;
967 	}
968 	VOP_UNLOCK(vp);
969 
970 	/*
971 	 * Remove unused wedges
972 	 */
973 	dkwedge_delall1(pdk, true);
974 
975 	/*
976 	 * For each supported partition map type, look to see if
977 	 * this map type exists.  If so, parse it and add the
978 	 * corresponding wedges.
979 	 */
980 	LIST_FOREACH(ddm, &dkwedge_discovery_methods, ddm_list) {
981 		error = (*ddm->ddm_discover)(pdk, vp);
982 		if (error == 0) {
983 			/* Successfully created wedges; we're done. */
984 			break;
985 		}
986 	}
987 
988 	error = vn_close(vp, FREAD, NOCRED);
989 	if (error) {
990 		aprint_error("%s: unable to close device, error = %d\n",
991 		    pdk->dk_name, error);
992 		/* We'll just assume the vnode has been cleaned up. */
993 	}
994 
995  out:
996 	rw_exit(&dkwedge_discovery_methods_lock);
997 }
998 
999 /*
1000  * dkwedge_read:
1001  *
1002  *	Read some data from the specified disk, used for
1003  *	partition discovery.
1004  */
1005 int
1006 dkwedge_read(struct disk *pdk, struct vnode *vp, daddr_t blkno,
1007     void *tbuf, size_t len)
1008 {
1009 	buf_t *bp;
1010 	int error;
1011 	bool isopen;
1012 	dev_t bdev;
1013 	struct vnode *bdvp;
1014 
1015 	/*
1016 	 * The kernel cannot read from a character device vnode
1017 	 * as physio() only handles user memory.
1018 	 *
1019 	 * If the block device has already been opened by a wedge
1020 	 * use that vnode and temporarily bump the open counter.
1021 	 *
1022 	 * Otherwise try to open the block device.
1023 	 */
1024 
1025 	bdev = devsw_chr2blk(vp->v_rdev);
1026 
1027 	mutex_enter(&pdk->dk_rawlock);
1028 	if (pdk->dk_rawopens != 0) {
1029 		KASSERT(pdk->dk_rawvp != NULL);
1030 		isopen = true;
1031 		++pdk->dk_rawopens;
1032 		bdvp = pdk->dk_rawvp;
1033 		error = 0;
1034 	} else {
1035 		isopen = false;
1036 		error = dk_open_parent(bdev, FREAD, &bdvp);
1037 	}
1038 	mutex_exit(&pdk->dk_rawlock);
1039 
1040 	if (error)
1041 		return error;
1042 
1043 	bp = getiobuf(bdvp, true);
1044 	bp->b_flags = B_READ;
1045 	bp->b_cflags = BC_BUSY;
1046 	bp->b_dev = bdev;
1047 	bp->b_data = tbuf;
1048 	bp->b_bufsize = bp->b_bcount = len;
1049 	bp->b_blkno = blkno;
1050 	bp->b_cylinder = 0;
1051 	bp->b_error = 0;
1052 
1053 	VOP_STRATEGY(bdvp, bp);
1054 	error = biowait(bp);
1055 	putiobuf(bp);
1056 
1057 	mutex_enter(&pdk->dk_rawlock);
1058 	if (isopen) {
1059 		--pdk->dk_rawopens;
1060 	} else {
1061 		dk_close_parent(bdvp, FREAD);
1062 	}
1063 	mutex_exit(&pdk->dk_rawlock);
1064 
1065 	return error;
1066 }
1067 
1068 /*
1069  * dkwedge_lookup:
1070  *
1071  *	Look up a dkwedge_softc based on the provided dev_t.
1072  */
1073 static struct dkwedge_softc *
1074 dkwedge_lookup(dev_t dev)
1075 {
1076 	int unit = minor(dev);
1077 
1078 	if (unit >= ndkwedges)
1079 		return (NULL);
1080 
1081 	KASSERT(dkwedges != NULL);
1082 
1083 	return (dkwedges[unit]);
1084 }
1085 
1086 static int
1087 dk_open_parent(dev_t dev, int mode, struct vnode **vpp)
1088 {
1089 	struct vnode *vp;
1090 	int error;
1091 
1092 	error = bdevvp(dev, &vp);
1093 	if (error)
1094 		return error;
1095 
1096 	error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1097 	if (error) {
1098 		vrele(vp);
1099 		return error;
1100 	}
1101 	error = VOP_OPEN(vp, mode, NOCRED);
1102 	if (error) {
1103 		vput(vp);
1104 		return error;
1105 	}
1106 
1107 	/* VOP_OPEN() doesn't do this for us. */
1108 	if (mode & FWRITE) {
1109 		mutex_enter(vp->v_interlock);
1110 		vp->v_writecount++;
1111 		mutex_exit(vp->v_interlock);
1112 	}
1113 
1114 	VOP_UNLOCK(vp);
1115 
1116 	*vpp = vp;
1117 
1118 	return 0;
1119 }
1120 
1121 static int
1122 dk_close_parent(struct vnode *vp, int mode)
1123 {
1124 	int error;
1125 
1126 	error = vn_close(vp, mode, NOCRED);
1127 	return error;
1128 }
1129 
1130 /*
1131  * dkopen:		[devsw entry point]
1132  *
1133  *	Open a wedge.
1134  */
1135 static int
1136 dkopen(dev_t dev, int flags, int fmt, struct lwp *l)
1137 {
1138 	struct dkwedge_softc *sc = dkwedge_lookup(dev);
1139 	struct vnode *vp;
1140 	int error = 0;
1141 
1142 	if (sc == NULL)
1143 		return (ENODEV);
1144 	if (sc->sc_state != DKW_STATE_RUNNING)
1145 		return (ENXIO);
1146 
1147 	/*
1148 	 * We go through a complicated little dance to only open the parent
1149 	 * vnode once per wedge, no matter how many times the wedge is
1150 	 * opened.  The reason?  We see one dkopen() per open call, but
1151 	 * only dkclose() on the last close.
1152 	 */
1153 	mutex_enter(&sc->sc_dk.dk_openlock);
1154 	mutex_enter(&sc->sc_parent->dk_rawlock);
1155 	if (sc->sc_dk.dk_openmask == 0) {
1156 		if (sc->sc_parent->dk_rawopens == 0) {
1157 			KASSERT(sc->sc_parent->dk_rawvp == NULL);
1158 			error = dk_open_parent(sc->sc_pdev, FREAD | FWRITE, &vp);
1159 			if (error)
1160 				goto popen_fail;
1161 			sc->sc_parent->dk_rawvp = vp;
1162 		}
1163 		sc->sc_parent->dk_rawopens++;
1164 	}
1165 	if (fmt == S_IFCHR)
1166 		sc->sc_dk.dk_copenmask |= 1;
1167 	else
1168 		sc->sc_dk.dk_bopenmask |= 1;
1169 	sc->sc_dk.dk_openmask =
1170 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
1171 
1172  popen_fail:
1173 	mutex_exit(&sc->sc_parent->dk_rawlock);
1174 	mutex_exit(&sc->sc_dk.dk_openlock);
1175 	return (error);
1176 }
1177 
1178 /*
1179  * Caller must hold sc->sc_dk.dk_openlock and sc->sc_parent->dk_rawlock.
1180  */
1181 static int
1182 dklastclose(struct dkwedge_softc *sc)
1183 {
1184 	struct vnode *vp;
1185 	int error = 0;
1186 
1187 	vp = NULL;
1188 	if (sc->sc_parent->dk_rawopens > 0) {
1189 		if (--sc->sc_parent->dk_rawopens == 0) {
1190 			KASSERT(sc->sc_parent->dk_rawvp != NULL);
1191 			vp = sc->sc_parent->dk_rawvp;
1192 			sc->sc_parent->dk_rawvp = NULL;
1193 		}
1194 	}
1195 
1196 	mutex_exit(&sc->sc_parent->dk_rawlock);
1197 	mutex_exit(&sc->sc_dk.dk_openlock);
1198 
1199 	if (vp) {
1200 		dk_close_parent(vp, FREAD | FWRITE);
1201 	}
1202 
1203 	return error;
1204 }
1205 
1206 /*
1207  * dkclose:		[devsw entry point]
1208  *
1209  *	Close a wedge.
1210  */
1211 static int
1212 dkclose(dev_t dev, int flags, int fmt, struct lwp *l)
1213 {
1214 	struct dkwedge_softc *sc = dkwedge_lookup(dev);
1215 	int error = 0;
1216 
1217 	if (sc == NULL)
1218 		return (ENODEV);
1219 	if (sc->sc_state != DKW_STATE_RUNNING)
1220 		return (ENXIO);
1221 
1222 	KASSERT(sc->sc_dk.dk_openmask != 0);
1223 
1224 	mutex_enter(&sc->sc_dk.dk_openlock);
1225 	mutex_enter(&sc->sc_parent->dk_rawlock);
1226 
1227 	if (fmt == S_IFCHR)
1228 		sc->sc_dk.dk_copenmask &= ~1;
1229 	else
1230 		sc->sc_dk.dk_bopenmask &= ~1;
1231 	sc->sc_dk.dk_openmask =
1232 	    sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
1233 
1234 	if (sc->sc_dk.dk_openmask == 0)
1235 		error = dklastclose(sc); /* releases locks */
1236 	else {
1237 		mutex_exit(&sc->sc_parent->dk_rawlock);
1238 		mutex_exit(&sc->sc_dk.dk_openlock);
1239 	}
1240 
1241 	return (error);
1242 }
1243 
1244 /*
1245  * dkstragegy:		[devsw entry point]
1246  *
1247  *	Perform I/O based on the wedge I/O strategy.
1248  */
1249 static void
1250 dkstrategy(struct buf *bp)
1251 {
1252 	struct dkwedge_softc *sc = dkwedge_lookup(bp->b_dev);
1253 	uint64_t p_size, p_offset;
1254 
1255 	if (sc == NULL) {
1256 		bp->b_error = ENODEV;
1257 		goto done;
1258 	}
1259 
1260 	if (sc->sc_state != DKW_STATE_RUNNING ||
1261 	    sc->sc_parent->dk_rawvp == NULL) {
1262 		bp->b_error = ENXIO;
1263 		goto done;
1264 	}
1265 
1266 	/* If it's an empty transfer, wake up the top half now. */
1267 	if (bp->b_bcount == 0)
1268 		goto done;
1269 
1270 	p_offset = sc->sc_offset << sc->sc_parent->dk_blkshift;
1271 	p_size   = sc->sc_size << sc->sc_parent->dk_blkshift;
1272 
1273 	/* Make sure it's in-range. */
1274 	if (bounds_check_with_mediasize(bp, DEV_BSIZE, p_size) <= 0)
1275 		goto done;
1276 
1277 	/* Translate it to the parent's raw LBA. */
1278 	bp->b_rawblkno = bp->b_blkno + p_offset;
1279 
1280 	/* Place it in the queue and start I/O on the unit. */
1281 	mutex_enter(&sc->sc_iolock);
1282 	sc->sc_iopend++;
1283 	disk_wait(&sc->sc_dk);
1284 	bufq_put(sc->sc_bufq, bp);
1285 	mutex_exit(&sc->sc_iolock);
1286 
1287 	dkstart(sc);
1288 	return;
1289 
1290  done:
1291 	bp->b_resid = bp->b_bcount;
1292 	biodone(bp);
1293 }
1294 
1295 /*
1296  * dkstart:
1297  *
1298  *	Start I/O that has been enqueued on the wedge.
1299  */
1300 static void
1301 dkstart(struct dkwedge_softc *sc)
1302 {
1303 	struct vnode *vp;
1304 	struct buf *bp, *nbp;
1305 
1306 	mutex_enter(&sc->sc_iolock);
1307 
1308 	/* Do as much work as has been enqueued. */
1309 	while ((bp = bufq_peek(sc->sc_bufq)) != NULL) {
1310 
1311 		if (sc->sc_state != DKW_STATE_RUNNING) {
1312 			(void) bufq_get(sc->sc_bufq);
1313 			if (sc->sc_iopend-- == 1 &&
1314 			    (sc->sc_flags & DK_F_WAIT_DRAIN) != 0) {
1315 				sc->sc_flags &= ~DK_F_WAIT_DRAIN;
1316 				cv_broadcast(&sc->sc_dkdrn);
1317 			}
1318 			mutex_exit(&sc->sc_iolock);
1319 			bp->b_error = ENXIO;
1320 			bp->b_resid = bp->b_bcount;
1321 			biodone(bp);
1322 			mutex_enter(&sc->sc_iolock);
1323 			continue;
1324 		}
1325 
1326 		/* fetch an I/O buf with sc_iolock dropped */
1327 		mutex_exit(&sc->sc_iolock);
1328 		nbp = getiobuf(sc->sc_parent->dk_rawvp, false);
1329 		mutex_enter(&sc->sc_iolock);
1330 		if (nbp == NULL) {
1331 			/*
1332 			 * No resources to run this request; leave the
1333 			 * buffer queued up, and schedule a timer to
1334 			 * restart the queue in 1/2 a second.
1335 			 */
1336 			callout_schedule(&sc->sc_restart_ch, hz / 2);
1337 			break;
1338 		}
1339 
1340 		/*
1341 		 * fetch buf, this can fail if another thread
1342 		 * has already processed the queue, it can also
1343 		 * return a completely different buf.
1344 		 */
1345 		bp = bufq_get(sc->sc_bufq);
1346 		if (bp == NULL) {
1347 			mutex_exit(&sc->sc_iolock);
1348 			putiobuf(nbp);
1349 			mutex_enter(&sc->sc_iolock);
1350 			continue;
1351 		}
1352 
1353 		/* Instrumentation. */
1354 		disk_busy(&sc->sc_dk);
1355 
1356 		/* release lock for VOP_STRATEGY */
1357 		mutex_exit(&sc->sc_iolock);
1358 
1359 		nbp->b_data = bp->b_data;
1360 		nbp->b_flags = bp->b_flags;
1361 		nbp->b_oflags = bp->b_oflags;
1362 		nbp->b_cflags = bp->b_cflags;
1363 		nbp->b_iodone = dkiodone;
1364 		nbp->b_proc = bp->b_proc;
1365 		nbp->b_blkno = bp->b_rawblkno;
1366 		nbp->b_dev = sc->sc_parent->dk_rawvp->v_rdev;
1367 		nbp->b_bcount = bp->b_bcount;
1368 		nbp->b_private = bp;
1369 		BIO_COPYPRIO(nbp, bp);
1370 
1371 		vp = nbp->b_vp;
1372 		if ((nbp->b_flags & B_READ) == 0) {
1373 			mutex_enter(vp->v_interlock);
1374 			vp->v_numoutput++;
1375 			mutex_exit(vp->v_interlock);
1376 		}
1377 		VOP_STRATEGY(vp, nbp);
1378 
1379 		mutex_enter(&sc->sc_iolock);
1380 	}
1381 
1382 	mutex_exit(&sc->sc_iolock);
1383 }
1384 
1385 /*
1386  * dkiodone:
1387  *
1388  *	I/O to a wedge has completed; alert the top half.
1389  */
1390 static void
1391 dkiodone(struct buf *bp)
1392 {
1393 	struct buf *obp = bp->b_private;
1394 	struct dkwedge_softc *sc = dkwedge_lookup(obp->b_dev);
1395 
1396 	if (bp->b_error != 0)
1397 		obp->b_error = bp->b_error;
1398 	obp->b_resid = bp->b_resid;
1399 	putiobuf(bp);
1400 
1401 	mutex_enter(&sc->sc_iolock);
1402 	if (sc->sc_iopend-- == 1 && (sc->sc_flags & DK_F_WAIT_DRAIN) != 0) {
1403 		sc->sc_flags &= ~DK_F_WAIT_DRAIN;
1404 		cv_broadcast(&sc->sc_dkdrn);
1405 	}
1406 
1407 	disk_unbusy(&sc->sc_dk, obp->b_bcount - obp->b_resid,
1408 	    obp->b_flags & B_READ);
1409 	mutex_exit(&sc->sc_iolock);
1410 
1411 	biodone(obp);
1412 
1413 	/* Kick the queue in case there is more work we can do. */
1414 	dkstart(sc);
1415 }
1416 
1417 /*
1418  * dkrestart:
1419  *
1420  *	Restart the work queue after it was stalled due to
1421  *	a resource shortage.  Invoked via a callout.
1422  */
1423 static void
1424 dkrestart(void *v)
1425 {
1426 	struct dkwedge_softc *sc = v;
1427 
1428 	dkstart(sc);
1429 }
1430 
1431 /*
1432  * dkminphys:
1433  *
1434  *	Call parent's minphys function.
1435  */
1436 static void
1437 dkminphys(struct buf *bp)
1438 {
1439 	struct dkwedge_softc *sc = dkwedge_lookup(bp->b_dev);
1440 	dev_t dev;
1441 
1442 	dev = bp->b_dev;
1443 	bp->b_dev = sc->sc_pdev;
1444 	if (sc->sc_parent->dk_driver && sc->sc_parent->dk_driver->d_minphys)
1445 		(*sc->sc_parent->dk_driver->d_minphys)(bp);
1446 	else
1447 		minphys(bp);
1448 	bp->b_dev = dev;
1449 }
1450 
1451 /*
1452  * dkread:		[devsw entry point]
1453  *
1454  *	Read from a wedge.
1455  */
1456 static int
1457 dkread(dev_t dev, struct uio *uio, int flags)
1458 {
1459 	struct dkwedge_softc *sc = dkwedge_lookup(dev);
1460 
1461 	if (sc == NULL)
1462 		return (ENODEV);
1463 	if (sc->sc_state != DKW_STATE_RUNNING)
1464 		return (ENXIO);
1465 
1466 	return (physio(dkstrategy, NULL, dev, B_READ, dkminphys, uio));
1467 }
1468 
1469 /*
1470  * dkwrite:		[devsw entry point]
1471  *
1472  *	Write to a wedge.
1473  */
1474 static int
1475 dkwrite(dev_t dev, struct uio *uio, int flags)
1476 {
1477 	struct dkwedge_softc *sc = dkwedge_lookup(dev);
1478 
1479 	if (sc == NULL)
1480 		return (ENODEV);
1481 	if (sc->sc_state != DKW_STATE_RUNNING)
1482 		return (ENXIO);
1483 
1484 	return (physio(dkstrategy, NULL, dev, B_WRITE, dkminphys, uio));
1485 }
1486 
1487 /*
1488  * dkioctl:		[devsw entry point]
1489  *
1490  *	Perform an ioctl request on a wedge.
1491  */
1492 static int
1493 dkioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
1494 {
1495 	struct dkwedge_softc *sc = dkwedge_lookup(dev);
1496 	int error = 0;
1497 
1498 	if (sc == NULL)
1499 		return (ENODEV);
1500 	if (sc->sc_state != DKW_STATE_RUNNING)
1501 		return (ENXIO);
1502 	if (sc->sc_parent->dk_rawvp == NULL)
1503 		return (ENXIO);
1504 
1505 	/*
1506 	 * We pass NODEV instead of our device to indicate we don't
1507 	 * want to handle disklabel ioctls
1508 	 */
1509 	error = disk_ioctl(&sc->sc_dk, NODEV, cmd, data, flag, l);
1510 	if (error != EPASSTHROUGH)
1511 		return (error);
1512 
1513 	error = 0;
1514 
1515 	switch (cmd) {
1516 	case DIOCGSTRATEGY:
1517 	case DIOCGCACHE:
1518 	case DIOCCACHESYNC:
1519 		error = VOP_IOCTL(sc->sc_parent->dk_rawvp, cmd, data, flag,
1520 					  l != NULL ? l->l_cred : NOCRED);
1521 		break;
1522 	case DIOCGWEDGEINFO:
1523 	    {
1524 		struct dkwedge_info *dkw = (void *) data;
1525 
1526 		strlcpy(dkw->dkw_devname, device_xname(sc->sc_dev),
1527 			sizeof(dkw->dkw_devname));
1528 	    	memcpy(dkw->dkw_wname, sc->sc_wname, sizeof(dkw->dkw_wname));
1529 		dkw->dkw_wname[sizeof(dkw->dkw_wname) - 1] = '\0';
1530 		strlcpy(dkw->dkw_parent, sc->sc_parent->dk_name,
1531 		    sizeof(dkw->dkw_parent));
1532 		dkw->dkw_offset = sc->sc_offset;
1533 		dkw->dkw_size = sc->sc_size;
1534 		strlcpy(dkw->dkw_ptype, sc->sc_ptype, sizeof(dkw->dkw_ptype));
1535 
1536 		break;
1537 	    }
1538 	case DIOCGSECTORALIGN:
1539 	    {
1540 		struct disk_sectoralign *dsa = data;
1541 		uint32_t r;
1542 
1543 		error = VOP_IOCTL(sc->sc_parent->dk_rawvp, cmd, dsa, flag,
1544 		    l != NULL ? l->l_cred : NOCRED);
1545 		if (error)
1546 			break;
1547 
1548 		r = sc->sc_offset % dsa->dsa_alignment;
1549 		if (r < dsa->dsa_firstaligned)
1550 			dsa->dsa_firstaligned = dsa->dsa_firstaligned - r;
1551 		else
1552 			dsa->dsa_firstaligned = (dsa->dsa_firstaligned +
1553 			    dsa->dsa_alignment) - r;
1554 		break;
1555 	    }
1556 	default:
1557 		error = ENOTTY;
1558 	}
1559 
1560 	return (error);
1561 }
1562 
1563 /*
1564  * dkdiscard:		[devsw entry point]
1565  *
1566  *	Perform a discard-range request on a wedge.
1567  */
1568 static int
1569 dkdiscard(dev_t dev, off_t pos, off_t len)
1570 {
1571 	struct dkwedge_softc *sc = dkwedge_lookup(dev);
1572 	unsigned shift;
1573 	off_t offset, maxlen;
1574 
1575 	if (sc == NULL)
1576 		return (ENODEV);
1577 	if (sc->sc_state != DKW_STATE_RUNNING)
1578 		return (ENXIO);
1579 	if (sc->sc_parent->dk_rawvp == NULL)
1580 		return (ENXIO);
1581 
1582 	shift = (sc->sc_parent->dk_blkshift + DEV_BSHIFT);
1583 	KASSERT(__type_fit(off_t, sc->sc_size));
1584 	KASSERT(__type_fit(off_t, sc->sc_offset));
1585 	KASSERT(0 <= sc->sc_offset);
1586 	KASSERT(sc->sc_size <= (__type_max(off_t) >> shift));
1587 	KASSERT(sc->sc_offset <= ((__type_max(off_t) >> shift) - sc->sc_size));
1588 	offset = ((off_t)sc->sc_offset << shift);
1589 	maxlen = ((off_t)sc->sc_size << shift);
1590 
1591 	if (len > maxlen)
1592 		return (EINVAL);
1593 	if (pos > (maxlen - len))
1594 		return (EINVAL);
1595 
1596 	pos += offset;
1597 	return VOP_FDISCARD(sc->sc_parent->dk_rawvp, pos, len);
1598 }
1599 
1600 /*
1601  * dksize:		[devsw entry point]
1602  *
1603  *	Query the size of a wedge for the purpose of performing a dump
1604  *	or for swapping to.
1605  */
1606 static int
1607 dksize(dev_t dev)
1608 {
1609 	struct dkwedge_softc *sc = dkwedge_lookup(dev);
1610 	int rv = -1;
1611 
1612 	if (sc == NULL)
1613 		return (-1);
1614 	if (sc->sc_state != DKW_STATE_RUNNING)
1615 		return (-1);
1616 
1617 	mutex_enter(&sc->sc_dk.dk_openlock);
1618 	mutex_enter(&sc->sc_parent->dk_rawlock);
1619 
1620 	/* Our content type is static, no need to open the device. */
1621 
1622 	if (strcmp(sc->sc_ptype, DKW_PTYPE_SWAP) == 0) {
1623 		/* Saturate if we are larger than INT_MAX. */
1624 		if (sc->sc_size > INT_MAX)
1625 			rv = INT_MAX;
1626 		else
1627 			rv = (int) sc->sc_size;
1628 	}
1629 
1630 	mutex_exit(&sc->sc_parent->dk_rawlock);
1631 	mutex_exit(&sc->sc_dk.dk_openlock);
1632 
1633 	return (rv);
1634 }
1635 
1636 /*
1637  * dkdump:		[devsw entry point]
1638  *
1639  *	Perform a crash dump to a wedge.
1640  */
1641 static int
1642 dkdump(dev_t dev, daddr_t blkno, void *va, size_t size)
1643 {
1644 	struct dkwedge_softc *sc = dkwedge_lookup(dev);
1645 	const struct bdevsw *bdev;
1646 	int rv = 0;
1647 
1648 	if (sc == NULL)
1649 		return (ENODEV);
1650 	if (sc->sc_state != DKW_STATE_RUNNING)
1651 		return (ENXIO);
1652 
1653 	mutex_enter(&sc->sc_dk.dk_openlock);
1654 	mutex_enter(&sc->sc_parent->dk_rawlock);
1655 
1656 	/* Our content type is static, no need to open the device. */
1657 
1658 	if (strcmp(sc->sc_ptype, DKW_PTYPE_SWAP) != 0 &&
1659 	    strcmp(sc->sc_ptype, DKW_PTYPE_RAID) != 0 &&
1660 	    strcmp(sc->sc_ptype, DKW_PTYPE_CGD) != 0) {
1661 		rv = ENXIO;
1662 		goto out;
1663 	}
1664 	if (size % DEV_BSIZE != 0) {
1665 		rv = EINVAL;
1666 		goto out;
1667 	}
1668 	if (blkno < 0 || blkno + size / DEV_BSIZE > sc->sc_size) {
1669 		printf("%s: blkno (%" PRIu64 ") + size / DEV_BSIZE (%zu) > "
1670 		    "sc->sc_size (%" PRIu64 ")\n", __func__, blkno,
1671 		    size / DEV_BSIZE, sc->sc_size);
1672 		rv = EINVAL;
1673 		goto out;
1674 	}
1675 
1676 	bdev = bdevsw_lookup(sc->sc_pdev);
1677 	rv = (*bdev->d_dump)(sc->sc_pdev, blkno + sc->sc_offset, va, size);
1678 
1679 out:
1680 	mutex_exit(&sc->sc_parent->dk_rawlock);
1681 	mutex_exit(&sc->sc_dk.dk_openlock);
1682 
1683 	return rv;
1684 }
1685 
1686 /*
1687  * config glue
1688  */
1689 
1690 /*
1691  * dkwedge_find_partition
1692  *
1693  *	Find wedge corresponding to the specified parent name
1694  *	and offset/length.
1695  */
1696 device_t
1697 dkwedge_find_partition(device_t parent, daddr_t startblk, uint64_t nblks)
1698 {
1699 	struct dkwedge_softc *sc;
1700 	int i;
1701 	device_t wedge = NULL;
1702 
1703 	rw_enter(&dkwedges_lock, RW_READER);
1704 	for (i = 0; i < ndkwedges; i++) {
1705 		if ((sc = dkwedges[i]) == NULL)
1706 			continue;
1707 		if (strcmp(sc->sc_parent->dk_name, device_xname(parent)) == 0 &&
1708 		    sc->sc_offset == startblk &&
1709 		    sc->sc_size == nblks) {
1710 			if (wedge) {
1711 				printf("WARNING: double match for boot wedge "
1712 				    "(%s, %s)\n",
1713 				    device_xname(wedge),
1714 				    device_xname(sc->sc_dev));
1715 				continue;
1716 			}
1717 			wedge = sc->sc_dev;
1718 		}
1719 	}
1720 	rw_exit(&dkwedges_lock);
1721 
1722 	return wedge;
1723 }
1724 
1725 const char *
1726 dkwedge_get_parent_name(dev_t dev)
1727 {
1728 	/* XXX: perhaps do this in lookup? */
1729 	int bmaj = bdevsw_lookup_major(&dk_bdevsw);
1730 	int cmaj = cdevsw_lookup_major(&dk_cdevsw);
1731 	if (major(dev) != bmaj && major(dev) != cmaj)
1732 		return NULL;
1733 	struct dkwedge_softc *sc = dkwedge_lookup(dev);
1734 	if (sc == NULL)
1735 		return NULL;
1736 	return sc->sc_parent->dk_name;
1737 }
1738 
1739