xref: /dflybsd-src/sys/kern/subr_disk.c (revision 807be17b7eecdefa2ff6404857d013f19a0ee834)
1 /*
2  * Copyright (c) 2003,2004,2009 The DragonFly Project.  All rights reserved.
3  *
4  * This code is derived from software contributed to The DragonFly Project
5  * by Matthew Dillon <dillon@backplane.com>
6  * and Alex Hornung <ahornung@gmail.com>
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  *
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in
16  *    the documentation and/or other materials provided with the
17  *    distribution.
18  * 3. Neither the name of The DragonFly Project nor the names of its
19  *    contributors may be used to endorse or promote products derived
20  *    from this software without specific, prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
25  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
26  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
27  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
28  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
29  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
30  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
31  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
32  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  *
35  * ----------------------------------------------------------------------------
36  * "THE BEER-WARE LICENSE" (Revision 42):
37  * <phk@FreeBSD.ORG> wrote this file.  As long as you retain this notice you
38  * can do whatever you want with this stuff. If we meet some day, and you think
39  * this stuff is worth it, you can buy me a beer in return.   Poul-Henning Kamp
40  * ----------------------------------------------------------------------------
41  *
42  * Copyright (c) 1982, 1986, 1988, 1993
43  *	The Regents of the University of California.  All rights reserved.
44  * (c) UNIX System Laboratories, Inc.
45  * All or some portions of this file are derived from material licensed
46  * to the University of California by American Telephone and Telegraph
47  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
48  * the permission of UNIX System Laboratories, Inc.
49  *
50  * Redistribution and use in source and binary forms, with or without
51  * modification, are permitted provided that the following conditions
52  * are met:
53  * 1. Redistributions of source code must retain the above copyright
54  *    notice, this list of conditions and the following disclaimer.
55  * 2. Redistributions in binary form must reproduce the above copyright
56  *    notice, this list of conditions and the following disclaimer in the
57  *    documentation and/or other materials provided with the distribution.
58  * 3. All advertising materials mentioning features or use of this software
59  *    must display the following acknowledgement:
60  *	This product includes software developed by the University of
61  *	California, Berkeley and its contributors.
62  * 4. Neither the name of the University nor the names of its contributors
63  *    may be used to endorse or promote products derived from this software
64  *    without specific prior written permission.
65  *
66  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
67  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
68  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
69  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
70  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
71  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
72  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
73  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
74  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
75  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
76  * SUCH DAMAGE.
77  *
78  *	@(#)ufs_disksubr.c	8.5 (Berkeley) 1/21/94
79  * $FreeBSD: src/sys/kern/subr_disk.c,v 1.20.2.6 2001/10/05 07:14:57 peter Exp $
80  * $FreeBSD: src/sys/ufs/ufs/ufs_disksubr.c,v 1.44.2.3 2001/03/05 05:42:19 obrien Exp $
81  */
82 
83 #include <sys/param.h>
84 #include <sys/systm.h>
85 #include <sys/kernel.h>
86 #include <sys/proc.h>
87 #include <sys/sysctl.h>
88 #include <sys/buf.h>
89 #include <sys/conf.h>
90 #include <sys/disklabel.h>
91 #include <sys/disklabel32.h>
92 #include <sys/disklabel64.h>
93 #include <sys/diskslice.h>
94 #include <sys/diskmbr.h>
95 #include <sys/disk.h>
96 #include <sys/kerneldump.h>
97 #include <sys/malloc.h>
98 #include <machine/md_var.h>
99 #include <sys/ctype.h>
100 #include <sys/syslog.h>
101 #include <sys/device.h>
102 #include <sys/msgport.h>
103 #include <sys/devfs.h>
104 #include <sys/thread.h>
105 #include <sys/dsched.h>
106 #include <sys/queue.h>
107 #include <sys/lock.h>
108 #include <sys/udev.h>
109 #include <sys/uuid.h>
110 
111 #include <sys/buf2.h>
112 #include <sys/mplock2.h>
113 #include <sys/msgport2.h>
114 #include <sys/thread2.h>
115 
116 static MALLOC_DEFINE(M_DISK, "disk", "disk data");
117 static int disk_debug_enable = 0;
118 
119 static void disk_msg_autofree_reply(lwkt_port_t, lwkt_msg_t);
120 static void disk_msg_core(void *);
121 static int disk_probe_slice(struct disk *dp, cdev_t dev, int slice, int reprobe);
122 static void disk_probe(struct disk *dp, int reprobe);
123 static void _setdiskinfo(struct disk *disk, struct disk_info *info);
124 static void bioqwritereorder(struct bio_queue_head *bioq);
125 static void disk_cleanserial(char *serno);
126 static int disk_debug(int, char *, ...) __printflike(2, 3);
127 static cdev_t _disk_create_named(const char *name, int unit, struct disk *dp,
128     struct dev_ops *raw_ops, int clone);
129 
130 static d_open_t diskopen;
131 static d_close_t diskclose;
132 static d_ioctl_t diskioctl;
133 static d_strategy_t diskstrategy;
134 static d_psize_t diskpsize;
135 static d_dump_t diskdump;
136 
137 static LIST_HEAD(, disk) disklist = LIST_HEAD_INITIALIZER(&disklist);
138 static struct lwkt_token disklist_token;
139 
140 static struct dev_ops disk_ops = {
141 	{ "disk", 0, D_DISK | D_MPSAFE | D_TRACKCLOSE },
142 	.d_open = diskopen,
143 	.d_close = diskclose,
144 	.d_read = physread,
145 	.d_write = physwrite,
146 	.d_ioctl = diskioctl,
147 	.d_strategy = diskstrategy,
148 	.d_dump = diskdump,
149 	.d_psize = diskpsize,
150 };
151 
152 static struct objcache 	*disk_msg_cache;
153 
154 struct objcache_malloc_args disk_msg_malloc_args = {
155 	sizeof(struct disk_msg), M_DISK };
156 
157 static struct lwkt_port disk_dispose_port;
158 static struct lwkt_port disk_msg_port;
159 
160 static int
161 disk_debug(int level, char *fmt, ...)
162 {
163 	__va_list ap;
164 
165 	__va_start(ap, fmt);
166 	if (level <= disk_debug_enable)
167 		kvprintf(fmt, ap);
168 	__va_end(ap);
169 
170 	return 0;
171 }
172 
173 static int
174 disk_probe_slice(struct disk *dp, cdev_t dev, int slice, int reprobe)
175 {
176 	struct disk_info *info = &dp->d_info;
177 	struct diskslice *sp = &dp->d_slice->dss_slices[slice];
178 	disklabel_ops_t ops;
179 	struct partinfo part;
180 	const char *msg;
181 	char uuid_buf[128];
182 	cdev_t ndev;
183 	int sno;
184 	u_int i;
185 
186 	disk_debug(2,
187 		    "disk_probe_slice (begin): %s (%s)\n",
188 			dev->si_name, dp->d_cdev->si_name);
189 
190 	sno = slice ? slice - 1 : 0;
191 
192 	ops = &disklabel32_ops;
193 	msg = ops->op_readdisklabel(dev, sp, &sp->ds_label, info);
194 	if (msg && !strcmp(msg, "no disk label")) {
195 		ops = &disklabel64_ops;
196 		msg = ops->op_readdisklabel(dev, sp, &sp->ds_label, info);
197 	}
198 
199 	if (msg == NULL) {
200 		if (slice != WHOLE_DISK_SLICE)
201 			ops->op_adjust_label_reserved(dp->d_slice, slice, sp);
202 		else
203 			sp->ds_reserved = 0;
204 
205 		sp->ds_ops = ops;
206 		for (i = 0; i < ops->op_getnumparts(sp->ds_label); i++) {
207 			ops->op_loadpartinfo(sp->ds_label, i, &part);
208 			if (part.fstype) {
209 				if (reprobe &&
210 				    (ndev = devfs_find_device_by_name("%s%c",
211 						dev->si_name, 'a' + i))
212 				) {
213 					/*
214 					 * Device already exists and
215 					 * is still valid.
216 					 */
217 					ndev->si_flags |= SI_REPROBE_TEST;
218 
219 					/*
220 					 * Destroy old UUID alias
221 					 */
222 					destroy_dev_alias(ndev, "part-by-uuid/*");
223 
224 					/* Create UUID alias */
225 					if (!kuuid_is_nil(&part.storage_uuid)) {
226 						snprintf_uuid(uuid_buf,
227 						    sizeof(uuid_buf),
228 						    &part.storage_uuid);
229 						make_dev_alias(ndev,
230 						    "part-by-uuid/%s",
231 						    uuid_buf);
232 						udev_dict_set_cstr(ndev, "uuid", uuid_buf);
233 					}
234 				} else {
235 					ndev = make_dev_covering(&disk_ops, dp->d_rawdev->si_ops,
236 						dkmakeminor(dkunit(dp->d_cdev),
237 							    slice, i),
238 						UID_ROOT, GID_OPERATOR, 0640,
239 						"%s%c", dev->si_name, 'a'+ i);
240 					ndev->si_parent = dev;
241 					ndev->si_disk = dp;
242 					udev_dict_set_cstr(ndev, "subsystem", "disk");
243 					/* Inherit parent's disk type */
244 					if (dp->d_disktype) {
245 						udev_dict_set_cstr(ndev, "disk-type",
246 						    __DECONST(char *, dp->d_disktype));
247 					}
248 
249 					/* Create serno alias */
250 					if (dp->d_info.d_serialno) {
251 						make_dev_alias(ndev,
252 						    "serno/%s.s%d%c",
253 						    dp->d_info.d_serialno,
254 						    sno, 'a' + i);
255 					}
256 
257 					/* Create UUID alias */
258 					if (!kuuid_is_nil(&part.storage_uuid)) {
259 						snprintf_uuid(uuid_buf,
260 						    sizeof(uuid_buf),
261 						    &part.storage_uuid);
262 						make_dev_alias(ndev,
263 						    "part-by-uuid/%s",
264 						    uuid_buf);
265 						udev_dict_set_cstr(ndev, "uuid", uuid_buf);
266 					}
267 					ndev->si_flags |= SI_REPROBE_TEST;
268 				}
269 			}
270 		}
271 	} else if (info->d_dsflags & DSO_COMPATLABEL) {
272 		msg = NULL;
273 		if (sp->ds_size >= 0x100000000ULL)
274 			ops = &disklabel64_ops;
275 		else
276 			ops = &disklabel32_ops;
277 		sp->ds_label = ops->op_clone_label(info, sp);
278 	} else {
279 		if (sp->ds_type == DOSPTYP_386BSD || /* XXX */
280 		    sp->ds_type == DOSPTYP_NETBSD ||
281 		    sp->ds_type == DOSPTYP_OPENBSD) {
282 			log(LOG_WARNING, "%s: cannot find label (%s)\n",
283 			    dev->si_name, msg);
284 		}
285 
286 		if (sp->ds_label.opaque != NULL && sp->ds_ops != NULL) {
287 			/* Clear out old label - it's not around anymore */
288 			disk_debug(2,
289 			    "disk_probe_slice: clear out old diskabel on %s\n",
290 			    dev->si_name);
291 
292 			sp->ds_ops->op_freedisklabel(&sp->ds_label);
293 			sp->ds_ops = NULL;
294 		}
295 	}
296 
297 	if (msg == NULL) {
298 		sp->ds_wlabel = FALSE;
299 	}
300 
301 	return (msg ? EINVAL : 0);
302 }
303 
304 /*
305  * This routine is only called for newly minted drives or to reprobe
306  * a drive with no open slices.  disk_probe_slice() is called directly
307  * when reprobing partition changes within slices.
308  */
309 static void
310 disk_probe(struct disk *dp, int reprobe)
311 {
312 	struct disk_info *info = &dp->d_info;
313 	cdev_t dev = dp->d_cdev;
314 	cdev_t ndev;
315 	int error, i, sno;
316 	struct diskslices *osp;
317 	struct diskslice *sp;
318 	char uuid_buf[128];
319 
320 	KKASSERT (info->d_media_blksize != 0);
321 
322 	osp = dp->d_slice;
323 	dp->d_slice = dsmakeslicestruct(BASE_SLICE, info);
324 	disk_debug(1, "disk_probe (begin): %s\n", dp->d_cdev->si_name);
325 
326 	error = mbrinit(dev, info, &(dp->d_slice));
327 	if (error) {
328 		dsgone(&osp);
329 		return;
330 	}
331 
332 	for (i = 0; i < dp->d_slice->dss_nslices; i++) {
333 		/*
334 		 * Ignore the whole-disk slice, it has already been created.
335 		 */
336 		if (i == WHOLE_DISK_SLICE)
337 			continue;
338 
339 #if 1
340 		/*
341 		 * Ignore the compatibility slice s0 if it's a device mapper
342 		 * volume.
343 		 */
344 		if ((i == COMPATIBILITY_SLICE) &&
345 		    (info->d_dsflags & DSO_DEVICEMAPPER))
346 			continue;
347 #endif
348 
349 		sp = &dp->d_slice->dss_slices[i];
350 
351 		/*
352 		 * Handle s0.  s0 is a compatibility slice if there are no
353 		 * other slices and it has not otherwise been set up, else
354 		 * we ignore it.
355 		 */
356 		if (i == COMPATIBILITY_SLICE) {
357 			sno = 0;
358 			if (sp->ds_type == 0 &&
359 			    dp->d_slice->dss_nslices == BASE_SLICE) {
360 				sp->ds_size = info->d_media_blocks;
361 				sp->ds_reserved = 0;
362 			}
363 		} else {
364 			sno = i - 1;
365 			sp->ds_reserved = 0;
366 		}
367 
368 		/*
369 		 * Ignore 0-length slices
370 		 */
371 		if (sp->ds_size == 0)
372 			continue;
373 
374 		if (reprobe &&
375 		    (ndev = devfs_find_device_by_name("%ss%d",
376 						      dev->si_name, sno))) {
377 			/*
378 			 * Device already exists and is still valid
379 			 */
380 			ndev->si_flags |= SI_REPROBE_TEST;
381 
382 			/*
383 			 * Destroy old UUID alias
384 			 */
385 			destroy_dev_alias(ndev, "slice-by-uuid/*");
386 
387 			/* Create UUID alias */
388 			if (!kuuid_is_nil(&sp->ds_stor_uuid)) {
389 				snprintf_uuid(uuid_buf, sizeof(uuid_buf),
390 				    &sp->ds_stor_uuid);
391 				make_dev_alias(ndev, "slice-by-uuid/%s",
392 				    uuid_buf);
393 			}
394 		} else {
395 			/*
396 			 * Else create new device
397 			 */
398 			ndev = make_dev_covering(&disk_ops, dp->d_rawdev->si_ops,
399 					dkmakewholeslice(dkunit(dev), i),
400 					UID_ROOT, GID_OPERATOR, 0640,
401 					(info->d_dsflags & DSO_DEVICEMAPPER)?
402 					"%s.s%d" : "%ss%d", dev->si_name, sno);
403 			ndev->si_parent = dev;
404 			udev_dict_set_cstr(ndev, "subsystem", "disk");
405 			/* Inherit parent's disk type */
406 			if (dp->d_disktype) {
407 				udev_dict_set_cstr(ndev, "disk-type",
408 				    __DECONST(char *, dp->d_disktype));
409 			}
410 
411 			/* Create serno alias */
412 			if (dp->d_info.d_serialno) {
413 				make_dev_alias(ndev, "serno/%s.s%d",
414 					       dp->d_info.d_serialno, sno);
415 			}
416 
417 			/* Create UUID alias */
418 			if (!kuuid_is_nil(&sp->ds_stor_uuid)) {
419 				snprintf_uuid(uuid_buf, sizeof(uuid_buf),
420 				    &sp->ds_stor_uuid);
421 				make_dev_alias(ndev, "slice-by-uuid/%s",
422 				    uuid_buf);
423 			}
424 
425 			ndev->si_disk = dp;
426 			ndev->si_flags |= SI_REPROBE_TEST;
427 		}
428 		sp->ds_dev = ndev;
429 
430 		/*
431 		 * Probe appropriate slices for a disklabel
432 		 *
433 		 * XXX slice type 1 used by our gpt probe code.
434 		 * XXX slice type 0 used by mbr compat slice.
435 		 */
436 		if (sp->ds_type == DOSPTYP_386BSD ||
437 		    sp->ds_type == DOSPTYP_NETBSD ||
438 		    sp->ds_type == DOSPTYP_OPENBSD ||
439 		    sp->ds_type == 0 ||
440 		    sp->ds_type == 1) {
441 			if (dp->d_slice->dss_first_bsd_slice == 0)
442 				dp->d_slice->dss_first_bsd_slice = i;
443 			disk_probe_slice(dp, ndev, i, reprobe);
444 		}
445 	}
446 	dsgone(&osp);
447 	disk_debug(1, "disk_probe (end): %s\n", dp->d_cdev->si_name);
448 }
449 
450 
451 static void
452 disk_msg_core(void *arg)
453 {
454 	struct disk	*dp;
455 	struct diskslice *sp;
456 	disk_msg_t msg;
457 	int run;
458 
459 	lwkt_gettoken(&disklist_token);
460 	lwkt_initport_thread(&disk_msg_port, curthread);
461 	wakeup(curthread);	/* synchronous startup */
462 	lwkt_reltoken(&disklist_token);
463 
464 	get_mplock();	/* not mpsafe yet? */
465 	run = 1;
466 
467 	while (run) {
468 		msg = (disk_msg_t)lwkt_waitport(&disk_msg_port, 0);
469 
470 		switch (msg->hdr.u.ms_result) {
471 		case DISK_DISK_PROBE:
472 			dp = (struct disk *)msg->load;
473 			disk_debug(1,
474 				    "DISK_DISK_PROBE: %s\n",
475 					dp->d_cdev->si_name);
476 			disk_probe(dp, 0);
477 			break;
478 		case DISK_DISK_DESTROY:
479 			dp = (struct disk *)msg->load;
480 			disk_debug(1,
481 				    "DISK_DISK_DESTROY: %s\n",
482 					dp->d_cdev->si_name);
483 			devfs_destroy_related(dp->d_cdev);
484 			destroy_dev(dp->d_cdev);
485 			destroy_only_dev(dp->d_rawdev);
486 			lwkt_gettoken(&disklist_token);
487 			LIST_REMOVE(dp, d_list);
488 			lwkt_reltoken(&disklist_token);
489 			if (dp->d_info.d_serialno) {
490 				kfree(dp->d_info.d_serialno, M_TEMP);
491 				dp->d_info.d_serialno = NULL;
492 			}
493 			break;
494 		case DISK_UNPROBE:
495 			dp = (struct disk *)msg->load;
496 			disk_debug(1,
497 				    "DISK_DISK_UNPROBE: %s\n",
498 					dp->d_cdev->si_name);
499 			devfs_destroy_related(dp->d_cdev);
500 			break;
501 		case DISK_SLICE_REPROBE:
502 			dp = (struct disk *)msg->load;
503 			sp = (struct diskslice *)msg->load2;
504 			devfs_clr_related_flag(sp->ds_dev,
505 						SI_REPROBE_TEST);
506 			disk_debug(1,
507 				    "DISK_SLICE_REPROBE: %s\n",
508 				    sp->ds_dev->si_name);
509 			disk_probe_slice(dp, sp->ds_dev,
510 					 dkslice(sp->ds_dev), 1);
511 			devfs_destroy_related_without_flag(
512 					sp->ds_dev, SI_REPROBE_TEST);
513 			break;
514 		case DISK_DISK_REPROBE:
515 			dp = (struct disk *)msg->load;
516 			devfs_clr_related_flag(dp->d_cdev, SI_REPROBE_TEST);
517 			disk_debug(1,
518 				    "DISK_DISK_REPROBE: %s\n",
519 				    dp->d_cdev->si_name);
520 			disk_probe(dp, 1);
521 			devfs_destroy_related_without_flag(
522 					dp->d_cdev, SI_REPROBE_TEST);
523 			break;
524 		case DISK_SYNC:
525 			disk_debug(1, "DISK_SYNC\n");
526 			break;
527 		default:
528 			devfs_debug(DEVFS_DEBUG_WARNING,
529 				    "disk_msg_core: unknown message "
530 				    "received at core\n");
531 			break;
532 		}
533 		lwkt_replymsg(&msg->hdr, 0);
534 	}
535 	lwkt_exit();
536 }
537 
538 
539 /*
540  * Acts as a message drain. Any message that is replied to here gets
541  * destroyed and the memory freed.
542  */
543 static void
544 disk_msg_autofree_reply(lwkt_port_t port, lwkt_msg_t msg)
545 {
546 	objcache_put(disk_msg_cache, msg);
547 }
548 
549 
550 void
551 disk_msg_send(uint32_t cmd, void *load, void *load2)
552 {
553 	disk_msg_t disk_msg;
554 	lwkt_port_t port = &disk_msg_port;
555 
556 	disk_msg = objcache_get(disk_msg_cache, M_WAITOK);
557 
558 	lwkt_initmsg(&disk_msg->hdr, &disk_dispose_port, 0);
559 
560 	disk_msg->hdr.u.ms_result = cmd;
561 	disk_msg->load = load;
562 	disk_msg->load2 = load2;
563 	KKASSERT(port);
564 	lwkt_sendmsg(port, &disk_msg->hdr);
565 }
566 
567 void
568 disk_msg_send_sync(uint32_t cmd, void *load, void *load2)
569 {
570 	struct lwkt_port rep_port;
571 	disk_msg_t disk_msg;
572 	lwkt_port_t port;
573 
574 	disk_msg = objcache_get(disk_msg_cache, M_WAITOK);
575 	port = &disk_msg_port;
576 
577 	/* XXX could probably use curthread's built-in msgport */
578 	lwkt_initport_thread(&rep_port, curthread);
579 	lwkt_initmsg(&disk_msg->hdr, &rep_port, 0);
580 
581 	disk_msg->hdr.u.ms_result = cmd;
582 	disk_msg->load = load;
583 	disk_msg->load2 = load2;
584 
585 	lwkt_sendmsg(port, &disk_msg->hdr);
586 	lwkt_waitmsg(&disk_msg->hdr, 0);
587 	objcache_put(disk_msg_cache, disk_msg);
588 }
589 
590 /*
591  * Create a raw device for the dev_ops template (which is returned).  Also
592  * create a slice and unit managed disk and overload the user visible
593  * device space with it.
594  *
595  * NOTE: The returned raw device is NOT a slice and unit managed device.
596  * It is an actual raw device representing the raw disk as specified by
597  * the passed dev_ops.  The disk layer not only returns such a raw device,
598  * it also uses it internally when passing (modified) commands through.
599  */
600 cdev_t
601 disk_create(int unit, struct disk *dp, struct dev_ops *raw_ops)
602 {
603 	return _disk_create_named(NULL, unit, dp, raw_ops, 0);
604 }
605 
606 cdev_t
607 disk_create_clone(int unit, struct disk *dp, struct dev_ops *raw_ops)
608 {
609 	return _disk_create_named(NULL, unit, dp, raw_ops, 1);
610 }
611 
612 cdev_t
613 disk_create_named(const char *name, int unit, struct disk *dp, struct dev_ops *raw_ops)
614 {
615 	return _disk_create_named(name, unit, dp, raw_ops, 0);
616 }
617 
618 cdev_t
619 disk_create_named_clone(const char *name, int unit, struct disk *dp, struct dev_ops *raw_ops)
620 {
621 	return _disk_create_named(name, unit, dp, raw_ops, 1);
622 }
623 
624 static cdev_t
625 _disk_create_named(const char *name, int unit, struct disk *dp, struct dev_ops *raw_ops, int clone)
626 {
627 	cdev_t rawdev;
628 
629 	disk_debug(1, "disk_create (begin): %s%d\n", name, unit);
630 
631 	if (name) {
632 		rawdev = make_only_dev(raw_ops, dkmakewholedisk(unit),
633 		    UID_ROOT, GID_OPERATOR, 0640, "%s", name);
634 	} else {
635 		rawdev = make_only_dev(raw_ops, dkmakewholedisk(unit),
636 		    UID_ROOT, GID_OPERATOR, 0640,
637 		    "%s%d", raw_ops->head.name, unit);
638 	}
639 
640 	bzero(dp, sizeof(*dp));
641 
642 	dp->d_rawdev = rawdev;
643 	dp->d_raw_ops = raw_ops;
644 	dp->d_dev_ops = &disk_ops;
645 
646 	if (name) {
647 		if (clone) {
648 			dp->d_cdev = make_only_dev_covering(&disk_ops, dp->d_rawdev->si_ops,
649 			    dkmakewholedisk(unit), UID_ROOT, GID_OPERATOR, 0640,
650 			    "%s", name);
651 		} else {
652 			dp->d_cdev = make_dev_covering(&disk_ops, dp->d_rawdev->si_ops,
653 			    dkmakewholedisk(unit), UID_ROOT, GID_OPERATOR, 0640,
654 			    "%s", name);
655 		}
656 	} else {
657 		if (clone) {
658 			dp->d_cdev = make_only_dev_covering(&disk_ops, dp->d_rawdev->si_ops,
659 			    dkmakewholedisk(unit),
660 			    UID_ROOT, GID_OPERATOR, 0640,
661 			    "%s%d", raw_ops->head.name, unit);
662 		} else {
663 			dp->d_cdev = make_dev_covering(&disk_ops, dp->d_rawdev->si_ops,
664 			    dkmakewholedisk(unit),
665 			    UID_ROOT, GID_OPERATOR, 0640,
666 			    "%s%d", raw_ops->head.name, unit);
667 		}
668 	}
669 
670 	udev_dict_set_cstr(dp->d_cdev, "subsystem", "disk");
671 	dp->d_cdev->si_disk = dp;
672 
673 	if (name)
674 		dsched_disk_create_callback(dp, name, unit);
675 	else
676 		dsched_disk_create_callback(dp, raw_ops->head.name, unit);
677 
678 	lwkt_gettoken(&disklist_token);
679 	LIST_INSERT_HEAD(&disklist, dp, d_list);
680 	lwkt_reltoken(&disklist_token);
681 
682 	disk_debug(1, "disk_create (end): %s%d\n",
683 	    (name != NULL)?(name):(raw_ops->head.name), unit);
684 
685 	return (dp->d_rawdev);
686 }
687 
688 int
689 disk_setdisktype(struct disk *disk, const char *type)
690 {
691 	KKASSERT(disk != NULL);
692 
693 	disk->d_disktype = type;
694 	return udev_dict_set_cstr(disk->d_cdev, "disk-type", __DECONST(char *, type));
695 }
696 
697 int
698 disk_getopencount(struct disk *disk)
699 {
700 	return disk->d_opencount;
701 }
702 
703 static void
704 _setdiskinfo(struct disk *disk, struct disk_info *info)
705 {
706 	char *oldserialno;
707 
708 	oldserialno = disk->d_info.d_serialno;
709 	bcopy(info, &disk->d_info, sizeof(disk->d_info));
710 	info = &disk->d_info;
711 
712 	disk_debug(1,
713 		    "_setdiskinfo: %s\n",
714 			disk->d_cdev->si_name);
715 
716 	/*
717 	 * The serial number is duplicated so the caller can throw
718 	 * their copy away.
719 	 */
720 	if (info->d_serialno && info->d_serialno[0] &&
721 	    (info->d_serialno[0] != ' ' || strlen(info->d_serialno) > 1)) {
722 		info->d_serialno = kstrdup(info->d_serialno, M_TEMP);
723 		disk_cleanserial(info->d_serialno);
724 		if (disk->d_cdev) {
725 			make_dev_alias(disk->d_cdev, "serno/%s",
726 					info->d_serialno);
727 		}
728 	} else {
729 		info->d_serialno = NULL;
730 	}
731 	if (oldserialno)
732 		kfree(oldserialno, M_TEMP);
733 
734 	dsched_disk_update_callback(disk, info);
735 
736 	/*
737 	 * The caller may set d_media_size or d_media_blocks and we
738 	 * calculate the other.
739 	 */
740 	KKASSERT(info->d_media_size == 0 || info->d_media_blocks == 0);
741 	if (info->d_media_size == 0 && info->d_media_blocks) {
742 		info->d_media_size = (u_int64_t)info->d_media_blocks *
743 				     info->d_media_blksize;
744 	} else if (info->d_media_size && info->d_media_blocks == 0 &&
745 		   info->d_media_blksize) {
746 		info->d_media_blocks = info->d_media_size /
747 				       info->d_media_blksize;
748 	}
749 
750 	/*
751 	 * The si_* fields for rawdev are not set until after the
752 	 * disk_create() call, so someone using the cooked version
753 	 * of the raw device (i.e. da0s0) will not get the right
754 	 * si_iosize_max unless we fix it up here.
755 	 */
756 	if (disk->d_cdev && disk->d_rawdev &&
757 	    disk->d_cdev->si_iosize_max == 0) {
758 		disk->d_cdev->si_iosize_max = disk->d_rawdev->si_iosize_max;
759 		disk->d_cdev->si_bsize_phys = disk->d_rawdev->si_bsize_phys;
760 		disk->d_cdev->si_bsize_best = disk->d_rawdev->si_bsize_best;
761 	}
762 
763 	/* Add the serial number to the udev_dictionary */
764 	if (info->d_serialno)
765 		udev_dict_set_cstr(disk->d_cdev, "serno", info->d_serialno);
766 }
767 
768 /*
769  * Disk drivers must call this routine when media parameters are available
770  * or have changed.
771  */
772 void
773 disk_setdiskinfo(struct disk *disk, struct disk_info *info)
774 {
775 	_setdiskinfo(disk, info);
776 	disk_msg_send(DISK_DISK_PROBE, disk, NULL);
777 	disk_debug(1,
778 		    "disk_setdiskinfo: sent probe for %s\n",
779 			disk->d_cdev->si_name);
780 }
781 
782 void
783 disk_setdiskinfo_sync(struct disk *disk, struct disk_info *info)
784 {
785 	_setdiskinfo(disk, info);
786 	disk_msg_send_sync(DISK_DISK_PROBE, disk, NULL);
787 	disk_debug(1,
788 		    "disk_setdiskinfo_sync: sent probe for %s\n",
789 			disk->d_cdev->si_name);
790 }
791 
792 /*
793  * This routine is called when an adapter detaches.  The higher level
794  * managed disk device is destroyed while the lower level raw device is
795  * released.
796  */
797 void
798 disk_destroy(struct disk *disk)
799 {
800 	dsched_disk_destroy_callback(disk);
801 	disk_msg_send_sync(DISK_DISK_DESTROY, disk, NULL);
802 	return;
803 }
804 
805 int
806 disk_dumpcheck(cdev_t dev, u_int64_t *size, u_int64_t *blkno, u_int32_t *secsize)
807 {
808 	struct partinfo pinfo;
809 	int error;
810 
811 	bzero(&pinfo, sizeof(pinfo));
812 	error = dev_dioctl(dev, DIOCGPART, (void *)&pinfo, 0,
813 			   proc0.p_ucred, NULL);
814 	if (error)
815 		return (error);
816 
817 	if (pinfo.media_blksize == 0)
818 		return (ENXIO);
819 
820 	if (blkno) /* XXX: make sure this reserved stuff is right */
821 		*blkno = pinfo.reserved_blocks +
822 			pinfo.media_offset / pinfo.media_blksize;
823 	if (secsize)
824 		*secsize = pinfo.media_blksize;
825 	if (size)
826 		*size = (pinfo.media_blocks - pinfo.reserved_blocks);
827 
828 	return (0);
829 }
830 
831 int
832 disk_dumpconf(cdev_t dev, u_int onoff)
833 {
834 	struct dumperinfo di;
835 	u_int64_t	size, blkno;
836 	u_int32_t	secsize;
837 	int error;
838 
839 	if (!onoff)
840 		return set_dumper(NULL);
841 
842 	error = disk_dumpcheck(dev, &size, &blkno, &secsize);
843 
844 	if (error)
845 		return ENXIO;
846 
847 	bzero(&di, sizeof(struct dumperinfo));
848 	di.dumper = diskdump;
849 	di.priv = dev;
850 	di.blocksize = secsize;
851 	di.mediaoffset = blkno * DEV_BSIZE;
852 	di.mediasize = size * DEV_BSIZE;
853 
854 	return set_dumper(&di);
855 }
856 
857 void
858 disk_unprobe(struct disk *disk)
859 {
860 	if (disk == NULL)
861 		return;
862 
863 	disk_msg_send_sync(DISK_UNPROBE, disk, NULL);
864 }
865 
866 void
867 disk_invalidate (struct disk *disk)
868 {
869 	dsgone(&disk->d_slice);
870 }
871 
872 struct disk *
873 disk_enumerate(struct disk *disk)
874 {
875 	struct disk *dp;
876 
877 	lwkt_gettoken(&disklist_token);
878 	if (!disk)
879 		dp = (LIST_FIRST(&disklist));
880 	else
881 		dp = (LIST_NEXT(disk, d_list));
882 	lwkt_reltoken(&disklist_token);
883 
884 	return dp;
885 }
886 
887 static
888 int
889 sysctl_disks(SYSCTL_HANDLER_ARGS)
890 {
891 	struct disk *disk;
892 	int error, first;
893 
894 	disk = NULL;
895 	first = 1;
896 
897 	while ((disk = disk_enumerate(disk))) {
898 		if (!first) {
899 			error = SYSCTL_OUT(req, " ", 1);
900 			if (error)
901 				return error;
902 		} else {
903 			first = 0;
904 		}
905 		error = SYSCTL_OUT(req, disk->d_rawdev->si_name,
906 				   strlen(disk->d_rawdev->si_name));
907 		if (error)
908 			return error;
909 	}
910 	error = SYSCTL_OUT(req, "", 1);
911 	return error;
912 }
913 
914 SYSCTL_PROC(_kern, OID_AUTO, disks, CTLTYPE_STRING | CTLFLAG_RD, NULL, 0,
915     sysctl_disks, "A", "names of available disks");
916 
917 /*
918  * Open a disk device or partition.
919  */
920 static
921 int
922 diskopen(struct dev_open_args *ap)
923 {
924 	cdev_t dev = ap->a_head.a_dev;
925 	struct disk *dp;
926 	int error;
927 
928 	/*
929 	 * dp can't be NULL here XXX.
930 	 *
931 	 * d_slice will be NULL if setdiskinfo() has not been called yet.
932 	 * setdiskinfo() is typically called whether the disk is present
933 	 * or not (e.g. CD), but the base disk device is created first
934 	 * and there may be a race.
935 	 */
936 	dp = dev->si_disk;
937 	if (dp == NULL || dp->d_slice == NULL)
938 		return (ENXIO);
939 	error = 0;
940 
941 	/*
942 	 * Deal with open races
943 	 */
944 	get_mplock();
945 	while (dp->d_flags & DISKFLAG_LOCK) {
946 		dp->d_flags |= DISKFLAG_WANTED;
947 		error = tsleep(dp, PCATCH, "diskopen", hz);
948 		if (error) {
949 			rel_mplock();
950 			return (error);
951 		}
952 	}
953 	dp->d_flags |= DISKFLAG_LOCK;
954 
955 	/*
956 	 * Open the underlying raw device.
957 	 */
958 	if (!dsisopen(dp->d_slice)) {
959 #if 0
960 		if (!pdev->si_iosize_max)
961 			pdev->si_iosize_max = dev->si_iosize_max;
962 #endif
963 		error = dev_dopen(dp->d_rawdev, ap->a_oflags,
964 				  ap->a_devtype, ap->a_cred);
965 	}
966 
967 	if (error)
968 		goto out;
969 	error = dsopen(dev, ap->a_devtype, dp->d_info.d_dsflags,
970 		       &dp->d_slice, &dp->d_info);
971 	if (!dsisopen(dp->d_slice)) {
972 		dev_dclose(dp->d_rawdev, ap->a_oflags, ap->a_devtype);
973 	}
974 out:
975 	dp->d_flags &= ~DISKFLAG_LOCK;
976 	if (dp->d_flags & DISKFLAG_WANTED) {
977 		dp->d_flags &= ~DISKFLAG_WANTED;
978 		wakeup(dp);
979 	}
980 	rel_mplock();
981 
982 	KKASSERT(dp->d_opencount >= 0);
983 	/* If the open was successful, bump open count */
984 	if (error == 0)
985 		atomic_add_int(&dp->d_opencount, 1);
986 
987 	return(error);
988 }
989 
990 /*
991  * Close a disk device or partition
992  */
993 static
994 int
995 diskclose(struct dev_close_args *ap)
996 {
997 	cdev_t dev = ap->a_head.a_dev;
998 	struct disk *dp;
999 	int error;
1000 	int lcount;
1001 
1002 	error = 0;
1003 	dp = dev->si_disk;
1004 
1005 	/*
1006 	 * The cdev_t represents the disk/slice/part.  The shared
1007 	 * dp structure governs all cdevs associated with the disk.
1008 	 *
1009 	 * As a safety only close the underlying raw device on the last
1010 	 * close the disk device if our tracking of the slices/partitions
1011 	 * also indicates nothing is open.
1012 	 */
1013 	KKASSERT(dp->d_opencount >= 1);
1014 	lcount = atomic_fetchadd_int(&dp->d_opencount, -1);
1015 
1016 	get_mplock();
1017 	dsclose(dev, ap->a_devtype, dp->d_slice);
1018 	if (lcount <= 1 && !dsisopen(dp->d_slice)) {
1019 		error = dev_dclose(dp->d_rawdev, ap->a_fflag, ap->a_devtype);
1020 	}
1021 	rel_mplock();
1022 	return (error);
1023 }
1024 
1025 /*
1026  * First execute the ioctl on the disk device, and if it isn't supported
1027  * try running it on the backing device.
1028  */
1029 static
1030 int
1031 diskioctl(struct dev_ioctl_args *ap)
1032 {
1033 	cdev_t dev = ap->a_head.a_dev;
1034 	struct disk *dp;
1035 	int error;
1036 	u_int u;
1037 
1038 	dp = dev->si_disk;
1039 	if (dp == NULL)
1040 		return (ENXIO);
1041 
1042 	devfs_debug(DEVFS_DEBUG_DEBUG,
1043 		    "diskioctl: cmd is: %lx (name: %s)\n",
1044 		    ap->a_cmd, dev->si_name);
1045 	devfs_debug(DEVFS_DEBUG_DEBUG,
1046 		    "diskioctl: &dp->d_slice is: %p, %p\n",
1047 		    &dp->d_slice, dp->d_slice);
1048 
1049 	if (ap->a_cmd == DIOCGKERNELDUMP) {
1050 		u = *(u_int *)ap->a_data;
1051 		return disk_dumpconf(dev, u);
1052 	}
1053 
1054 	if (&dp->d_slice == NULL || dp->d_slice == NULL ||
1055 	    ((dp->d_info.d_dsflags & DSO_DEVICEMAPPER) &&
1056 	     dkslice(dev) == WHOLE_DISK_SLICE)) {
1057 		error = ENOIOCTL;
1058 	} else {
1059 		get_mplock();
1060 		error = dsioctl(dev, ap->a_cmd, ap->a_data, ap->a_fflag,
1061 				&dp->d_slice, &dp->d_info);
1062 		rel_mplock();
1063 	}
1064 
1065 	if (error == ENOIOCTL) {
1066 		error = dev_dioctl(dp->d_rawdev, ap->a_cmd, ap->a_data,
1067 				   ap->a_fflag, ap->a_cred, NULL);
1068 	}
1069 	return (error);
1070 }
1071 
1072 /*
1073  * Execute strategy routine
1074  */
1075 static
1076 int
1077 diskstrategy(struct dev_strategy_args *ap)
1078 {
1079 	cdev_t dev = ap->a_head.a_dev;
1080 	struct bio *bio = ap->a_bio;
1081 	struct bio *nbio;
1082 	struct disk *dp;
1083 
1084 	dp = dev->si_disk;
1085 
1086 	if (dp == NULL) {
1087 		bio->bio_buf->b_error = ENXIO;
1088 		bio->bio_buf->b_flags |= B_ERROR;
1089 		biodone(bio);
1090 		return(0);
1091 	}
1092 	KKASSERT(dev->si_disk == dp);
1093 
1094 	/*
1095 	 * The dscheck() function will also transform the slice relative
1096 	 * block number i.e. bio->bio_offset into a block number that can be
1097 	 * passed directly to the underlying raw device.  If dscheck()
1098 	 * returns NULL it will have handled the bio for us (e.g. EOF
1099 	 * or error due to being beyond the device size).
1100 	 */
1101 	if ((nbio = dscheck(dev, bio, dp->d_slice)) != NULL) {
1102 		dsched_queue(dp, nbio);
1103 	} else {
1104 		biodone(bio);
1105 	}
1106 	return(0);
1107 }
1108 
1109 /*
1110  * Return the partition size in ?blocks?
1111  */
1112 static
1113 int
1114 diskpsize(struct dev_psize_args *ap)
1115 {
1116 	cdev_t dev = ap->a_head.a_dev;
1117 	struct disk *dp;
1118 
1119 	dp = dev->si_disk;
1120 	if (dp == NULL)
1121 		return(ENODEV);
1122 
1123 	ap->a_result = dssize(dev, &dp->d_slice);
1124 
1125 	if ((ap->a_result == -1) &&
1126 	   (dp->d_info.d_dsflags & DSO_RAWPSIZE)) {
1127 		ap->a_head.a_dev = dp->d_rawdev;
1128 		return dev_doperate(&ap->a_head);
1129 	}
1130 	return(0);
1131 }
1132 
1133 int
1134 diskdump(struct dev_dump_args *ap)
1135 {
1136 	cdev_t dev = ap->a_head.a_dev;
1137 	struct disk *dp = dev->si_disk;
1138 	u_int64_t size, offset;
1139 	int error;
1140 
1141 	error = disk_dumpcheck(dev, &size, &ap->a_blkno, &ap->a_secsize);
1142 	/* XXX: this should probably go in disk_dumpcheck somehow */
1143 	if (ap->a_length != 0) {
1144 		size *= DEV_BSIZE;
1145 		offset = ap->a_blkno * DEV_BSIZE;
1146 		if ((ap->a_offset < offset) ||
1147 		    (ap->a_offset + ap->a_length - offset > size)) {
1148 			kprintf("Attempt to write outside dump device boundaries.\n");
1149 			error = ENOSPC;
1150 		}
1151 	}
1152 
1153 	if (error == 0) {
1154 		ap->a_head.a_dev = dp->d_rawdev;
1155 		error = dev_doperate(&ap->a_head);
1156 	}
1157 
1158 	return(error);
1159 }
1160 
1161 
1162 SYSCTL_INT(_debug_sizeof, OID_AUTO, diskslices, CTLFLAG_RD,
1163     0, sizeof(struct diskslices), "sizeof(struct diskslices)");
1164 
1165 SYSCTL_INT(_debug_sizeof, OID_AUTO, disk, CTLFLAG_RD,
1166     0, sizeof(struct disk), "sizeof(struct disk)");
1167 
1168 /*
1169  * Reorder interval for burst write allowance and minor write
1170  * allowance.
1171  *
1172  * We always want to trickle some writes in to make use of the
1173  * disk's zone cache.  Bursting occurs on a longer interval and only
1174  * runningbufspace is well over the hirunningspace limit.
1175  */
1176 int bioq_reorder_burst_interval = 60;	/* should be multiple of minor */
1177 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_burst_interval,
1178 	   CTLFLAG_RW, &bioq_reorder_burst_interval, 0, "");
1179 int bioq_reorder_minor_interval = 5;
1180 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_minor_interval,
1181 	   CTLFLAG_RW, &bioq_reorder_minor_interval, 0, "");
1182 
1183 int bioq_reorder_burst_bytes = 3000000;
1184 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_burst_bytes,
1185 	   CTLFLAG_RW, &bioq_reorder_burst_bytes, 0, "");
1186 int bioq_reorder_minor_bytes = 262144;
1187 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_minor_bytes,
1188 	   CTLFLAG_RW, &bioq_reorder_minor_bytes, 0, "");
1189 
1190 
1191 /*
1192  * Order I/Os.  Generally speaking this code is designed to make better
1193  * use of drive zone caches.  A drive zone cache can typically track linear
1194  * reads or writes for around 16 zones simultaniously.
1195  *
1196  * Read prioritization issues:  It is possible for hundreds of megabytes worth
1197  * of writes to be queued asynchronously.  This creates a huge bottleneck
1198  * for reads which reduce read bandwidth to a trickle.
1199  *
1200  * To solve this problem we generally reorder reads before writes.
1201  *
1202  * However, a large number of random reads can also starve writes and
1203  * make poor use of the drive zone cache so we allow writes to trickle
1204  * in every N reads.
1205  */
1206 void
1207 bioqdisksort(struct bio_queue_head *bioq, struct bio *bio)
1208 {
1209 	/*
1210 	 * The BIO wants to be ordered.  Adding to the tail also
1211 	 * causes transition to be set to NULL, forcing the ordering
1212 	 * of all prior I/O's.
1213 	 */
1214 	if (bio->bio_buf->b_flags & B_ORDERED) {
1215 		bioq_insert_tail(bioq, bio);
1216 		return;
1217 	}
1218 
1219 	switch(bio->bio_buf->b_cmd) {
1220 	case BUF_CMD_READ:
1221 		if (bioq->transition) {
1222 			/*
1223 			 * Insert before the first write.  Bleedover writes
1224 			 * based on reorder intervals to prevent starvation.
1225 			 */
1226 			TAILQ_INSERT_BEFORE(bioq->transition, bio, bio_act);
1227 			++bioq->reorder;
1228 			if (bioq->reorder % bioq_reorder_minor_interval == 0) {
1229 				bioqwritereorder(bioq);
1230 				if (bioq->reorder >=
1231 				    bioq_reorder_burst_interval) {
1232 					bioq->reorder = 0;
1233 				}
1234 			}
1235 		} else {
1236 			/*
1237 			 * No writes queued (or ordering was forced),
1238 			 * insert at tail.
1239 			 */
1240 			TAILQ_INSERT_TAIL(&bioq->queue, bio, bio_act);
1241 		}
1242 		break;
1243 	case BUF_CMD_WRITE:
1244 		/*
1245 		 * Writes are always appended.  If no writes were previously
1246 		 * queued or an ordered tail insertion occured the transition
1247 		 * field will be NULL.
1248 		 */
1249 		TAILQ_INSERT_TAIL(&bioq->queue, bio, bio_act);
1250 		if (bioq->transition == NULL)
1251 			bioq->transition = bio;
1252 		break;
1253 	default:
1254 		/*
1255 		 * All other request types are forced to be ordered.
1256 		 */
1257 		bioq_insert_tail(bioq, bio);
1258 		break;
1259 	}
1260 }
1261 
1262 /*
1263  * Move the read-write transition point to prevent reads from
1264  * completely starving our writes.  This brings a number of writes into
1265  * the fold every N reads.
1266  *
1267  * We bring a few linear writes into the fold on a minor interval
1268  * and we bring a non-linear burst of writes into the fold on a major
1269  * interval.  Bursting only occurs if runningbufspace is really high
1270  * (typically from syncs, fsyncs, or HAMMER flushes).
1271  */
1272 static
1273 void
1274 bioqwritereorder(struct bio_queue_head *bioq)
1275 {
1276 	struct bio *bio;
1277 	off_t next_offset;
1278 	size_t left;
1279 	size_t n;
1280 	int check_off;
1281 
1282 	if (bioq->reorder < bioq_reorder_burst_interval ||
1283 	    !buf_runningbufspace_severe()) {
1284 		left = (size_t)bioq_reorder_minor_bytes;
1285 		check_off = 1;
1286 	} else {
1287 		left = (size_t)bioq_reorder_burst_bytes;
1288 		check_off = 0;
1289 	}
1290 
1291 	next_offset = bioq->transition->bio_offset;
1292 	while ((bio = bioq->transition) != NULL &&
1293 	       (check_off == 0 || next_offset == bio->bio_offset)
1294 	) {
1295 		n = bio->bio_buf->b_bcount;
1296 		next_offset = bio->bio_offset + n;
1297 		bioq->transition = TAILQ_NEXT(bio, bio_act);
1298 		if (left < n)
1299 			break;
1300 		left -= n;
1301 	}
1302 }
1303 
1304 /*
1305  * Bounds checking against the media size, used for the raw partition.
1306  * secsize, mediasize and b_blkno must all be the same units.
1307  * Possibly this has to be DEV_BSIZE (512).
1308  */
1309 int
1310 bounds_check_with_mediasize(struct bio *bio, int secsize, uint64_t mediasize)
1311 {
1312 	struct buf *bp = bio->bio_buf;
1313 	int64_t sz;
1314 
1315 	sz = howmany(bp->b_bcount, secsize);
1316 
1317 	if (bio->bio_offset/DEV_BSIZE + sz > mediasize) {
1318 		sz = mediasize - bio->bio_offset/DEV_BSIZE;
1319 		if (sz == 0) {
1320 			/* If exactly at end of disk, return EOF. */
1321 			bp->b_resid = bp->b_bcount;
1322 			return 0;
1323 		}
1324 		if (sz < 0) {
1325 			/* If past end of disk, return EINVAL. */
1326 			bp->b_error = EINVAL;
1327 			return 0;
1328 		}
1329 		/* Otherwise, truncate request. */
1330 		bp->b_bcount = sz * secsize;
1331 	}
1332 
1333 	return 1;
1334 }
1335 
1336 /*
1337  * Disk error is the preface to plaintive error messages
1338  * about failing disk transfers.  It prints messages of the form
1339 
1340 hp0g: hard error reading fsbn 12345 of 12344-12347 (hp0 bn %d cn %d tn %d sn %d)
1341 
1342  * if the offset of the error in the transfer and a disk label
1343  * are both available.  blkdone should be -1 if the position of the error
1344  * is unknown; the disklabel pointer may be null from drivers that have not
1345  * been converted to use them.  The message is printed with kprintf
1346  * if pri is LOG_PRINTF, otherwise it uses log at the specified priority.
1347  * The message should be completed (with at least a newline) with kprintf
1348  * or log(-1, ...), respectively.  There is no trailing space.
1349  */
1350 void
1351 diskerr(struct bio *bio, cdev_t dev, const char *what, int pri, int donecnt)
1352 {
1353 	struct buf *bp = bio->bio_buf;
1354 	const char *term;
1355 
1356 	switch(bp->b_cmd) {
1357 	case BUF_CMD_READ:
1358 		term = "read";
1359 		break;
1360 	case BUF_CMD_WRITE:
1361 		term = "write";
1362 		break;
1363 	default:
1364 		term = "access";
1365 		break;
1366 	}
1367 	kprintf("%s: %s %sing ", dev->si_name, what, term);
1368 	kprintf("offset %012llx for %d",
1369 		(long long)bio->bio_offset,
1370 		bp->b_bcount);
1371 
1372 	if (donecnt)
1373 		kprintf(" (%d bytes completed)", donecnt);
1374 }
1375 
1376 /*
1377  * Locate a disk device
1378  */
1379 cdev_t
1380 disk_locate(const char *devname)
1381 {
1382 	return devfs_find_device_by_name(devname);
1383 }
1384 
1385 void
1386 disk_config(void *arg)
1387 {
1388 	disk_msg_send_sync(DISK_SYNC, NULL, NULL);
1389 }
1390 
1391 static void
1392 disk_init(void)
1393 {
1394 	struct thread* td_core;
1395 
1396 	disk_msg_cache = objcache_create("disk-msg-cache", 0, 0,
1397 					 NULL, NULL, NULL,
1398 					 objcache_malloc_alloc,
1399 					 objcache_malloc_free,
1400 					 &disk_msg_malloc_args);
1401 
1402 	lwkt_token_init(&disklist_token, "disks");
1403 
1404 	/*
1405 	 * Initialize the reply-only port which acts as a message drain
1406 	 */
1407 	lwkt_initport_replyonly(&disk_dispose_port, disk_msg_autofree_reply);
1408 
1409 	lwkt_gettoken(&disklist_token);
1410 	lwkt_create(disk_msg_core, /*args*/NULL, &td_core, NULL,
1411 		    0, -1, "disk_msg_core");
1412 	tsleep(td_core, 0, "diskcore", 0);
1413 	lwkt_reltoken(&disklist_token);
1414 }
1415 
1416 static void
1417 disk_uninit(void)
1418 {
1419 	objcache_destroy(disk_msg_cache);
1420 }
1421 
1422 /*
1423  * Clean out illegal characters in serial numbers.
1424  */
1425 static void
1426 disk_cleanserial(char *serno)
1427 {
1428 	char c;
1429 
1430 	while ((c = *serno) != 0) {
1431 		if (c >= 'a' && c <= 'z')
1432 			;
1433 		else if (c >= 'A' && c <= 'Z')
1434 			;
1435 		else if (c >= '0' && c <= '9')
1436 			;
1437 		else if (c == '-' || c == '@' || c == '+' || c == '.')
1438 			;
1439 		else
1440 			c = '_';
1441 		*serno++= c;
1442 	}
1443 }
1444 
1445 TUNABLE_INT("kern.disk_debug", &disk_debug_enable);
1446 SYSCTL_INT(_kern, OID_AUTO, disk_debug, CTLFLAG_RW, &disk_debug_enable,
1447 		0, "Enable subr_disk debugging");
1448 
1449 SYSINIT(disk_register, SI_SUB_PRE_DRIVERS, SI_ORDER_FIRST, disk_init, NULL);
1450 SYSUNINIT(disk_register, SI_SUB_PRE_DRIVERS, SI_ORDER_ANY, disk_uninit, NULL);
1451