xref: /dflybsd-src/sys/kern/subr_disk.c (revision 90ea502b8c5d21f908cedff6680ee2bc9e74ce74)
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  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  *
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in
15  *    the documentation and/or other materials provided with the
16  *    distribution.
17  * 3. Neither the name of The DragonFly Project nor the names of its
18  *    contributors may be used to endorse or promote products derived
19  *    from this software without specific, prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  * ----------------------------------------------------------------------------
35  * "THE BEER-WARE LICENSE" (Revision 42):
36  * <phk@FreeBSD.ORG> wrote this file.  As long as you retain this notice you
37  * can do whatever you want with this stuff. If we meet some day, and you think
38  * this stuff is worth it, you can buy me a beer in return.   Poul-Henning Kamp
39  * ----------------------------------------------------------------------------
40  *
41  * Copyright (c) 1982, 1986, 1988, 1993
42  *	The Regents of the University of California.  All rights reserved.
43  * (c) UNIX System Laboratories, Inc.
44  * All or some portions of this file are derived from material licensed
45  * to the University of California by American Telephone and Telegraph
46  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
47  * the permission of UNIX System Laboratories, Inc.
48  *
49  * Redistribution and use in source and binary forms, with or without
50  * modification, are permitted provided that the following conditions
51  * are met:
52  * 1. Redistributions of source code must retain the above copyright
53  *    notice, this list of conditions and the following disclaimer.
54  * 2. Redistributions in binary form must reproduce the above copyright
55  *    notice, this list of conditions and the following disclaimer in the
56  *    documentation and/or other materials provided with the distribution.
57  * 3. All advertising materials mentioning features or use of this software
58  *    must display the following acknowledgement:
59  *	This product includes software developed by the University of
60  *	California, Berkeley and its contributors.
61  * 4. Neither the name of the University nor the names of its contributors
62  *    may be used to endorse or promote products derived from this software
63  *    without specific prior written permission.
64  *
65  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
66  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
67  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
68  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
69  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
70  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
71  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
72  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
73  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
74  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
75  * SUCH DAMAGE.
76  *
77  *	@(#)ufs_disksubr.c	8.5 (Berkeley) 1/21/94
78  * $FreeBSD: src/sys/kern/subr_disk.c,v 1.20.2.6 2001/10/05 07:14:57 peter Exp $
79  * $FreeBSD: src/sys/ufs/ufs/ufs_disksubr.c,v 1.44.2.3 2001/03/05 05:42:19 obrien Exp $
80  * $DragonFly: src/sys/kern/subr_disk.c,v 1.40 2008/06/05 18:06:32 swildner 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 <sys/sysctl.h>
99 #include <machine/md_var.h>
100 #include <sys/ctype.h>
101 #include <sys/syslog.h>
102 #include <sys/device.h>
103 #include <sys/msgport.h>
104 #include <sys/msgport2.h>
105 #include <sys/buf2.h>
106 #include <sys/devfs.h>
107 #include <sys/thread.h>
108 #include <sys/thread2.h>
109 
110 #include <sys/queue.h>
111 #include <sys/lock.h>
112 
113 static MALLOC_DEFINE(M_DISK, "disk", "disk data");
114 static int disk_debug_enable = 0;
115 
116 static void disk_msg_autofree_reply(lwkt_port_t, lwkt_msg_t);
117 static void disk_msg_core(void *);
118 static int disk_probe_slice(struct disk *dp, cdev_t dev, int slice, int reprobe);
119 static void disk_probe(struct disk *dp, int reprobe);
120 static void _setdiskinfo(struct disk *disk, struct disk_info *info);
121 static void bioqwritereorder(struct bio_queue_head *bioq);
122 static void disk_cleanserial(char *serno);
123 
124 static d_open_t diskopen;
125 static d_close_t diskclose;
126 static d_ioctl_t diskioctl;
127 static d_strategy_t diskstrategy;
128 static d_psize_t diskpsize;
129 static d_clone_t diskclone;
130 static d_dump_t diskdump;
131 
132 static LIST_HEAD(, disk) disklist = LIST_HEAD_INITIALIZER(&disklist);
133 static struct lwkt_token disklist_token;
134 
135 static struct dev_ops disk_ops = {
136 	{ "disk", 0, D_DISK },
137 	.d_open = diskopen,
138 	.d_close = diskclose,
139 	.d_read = physread,
140 	.d_write = physwrite,
141 	.d_ioctl = diskioctl,
142 	.d_strategy = diskstrategy,
143 	.d_dump = diskdump,
144 	.d_psize = diskpsize,
145 	.d_clone = diskclone
146 };
147 
148 static struct objcache 	*disk_msg_cache;
149 
150 struct objcache_malloc_args disk_msg_malloc_args = {
151 	sizeof(struct disk_msg), M_DISK };
152 
153 static struct lwkt_port disk_dispose_port;
154 static struct lwkt_port disk_msg_port;
155 
156 static int
157 disk_debug(int level, char *fmt, ...)
158 {
159 	__va_list ap;
160 
161 	__va_start(ap, fmt);
162 	if (level <= disk_debug_enable)
163 		kvprintf(fmt, ap);
164 	__va_end(ap);
165 
166 	return 0;
167 }
168 
169 static int
170 disk_probe_slice(struct disk *dp, cdev_t dev, int slice, int reprobe)
171 {
172 	struct disk_info *info = &dp->d_info;
173 	struct diskslice *sp = &dp->d_slice->dss_slices[slice];
174 	disklabel_ops_t ops;
175 	struct partinfo part;
176 	const char *msg;
177 	cdev_t ndev;
178 	int sno;
179 	u_int i;
180 
181 	disk_debug(2,
182 		    "disk_probe_slice (begin): %s (%s)\n",
183 			dev->si_name, dp->d_cdev->si_name);
184 
185 	sno = slice ? slice - 1 : 0;
186 
187 	ops = &disklabel32_ops;
188 	msg = ops->op_readdisklabel(dev, sp, &sp->ds_label, info);
189 	if (msg && !strcmp(msg, "no disk label")) {
190 		ops = &disklabel64_ops;
191 		msg = ops->op_readdisklabel(dev, sp, &sp->ds_label, info);
192 	}
193 	if (msg == NULL) {
194 		if (slice != WHOLE_DISK_SLICE)
195 			ops->op_adjust_label_reserved(dp->d_slice, slice, sp);
196 		else
197 			sp->ds_reserved = 0;
198 
199 		sp->ds_ops = ops;
200 		for (i = 0; i < ops->op_getnumparts(sp->ds_label); i++) {
201 			ops->op_loadpartinfo(sp->ds_label, i, &part);
202 			if (part.fstype) {
203 				if (reprobe &&
204 				    (ndev = devfs_find_device_by_name("%s%c",
205 						dev->si_name, 'a' + i))
206 				) {
207 					/*
208 					 * Device already exists and
209 					 * is still valid.
210 					 */
211 					ndev->si_flags |= SI_REPROBE_TEST;
212 				} else {
213 					ndev = make_dev_covering(&disk_ops, dp->d_rawdev->si_ops,
214 						dkmakeminor(dkunit(dp->d_cdev),
215 							    slice, i),
216 						UID_ROOT, GID_OPERATOR, 0640,
217 						"%s%c", dev->si_name, 'a'+ i);
218 					ndev->si_disk = dp;
219 					if (dp->d_info.d_serialno) {
220 						make_dev_alias(ndev,
221 						    "serno/%s.s%d%c",
222 						    dp->d_info.d_serialno,
223 						    sno, 'a' + i);
224 					}
225 					ndev->si_flags |= SI_REPROBE_TEST;
226 				}
227 			}
228 		}
229 	} else if (info->d_dsflags & DSO_COMPATLABEL) {
230 		msg = NULL;
231 		if (sp->ds_size >= 0x100000000ULL)
232 			ops = &disklabel64_ops;
233 		else
234 			ops = &disklabel32_ops;
235 		sp->ds_label = ops->op_clone_label(info, sp);
236 	} else {
237 		if (sp->ds_type == DOSPTYP_386BSD || /* XXX */
238 		    sp->ds_type == DOSPTYP_NETBSD ||
239 		    sp->ds_type == DOSPTYP_OPENBSD) {
240 			log(LOG_WARNING, "%s: cannot find label (%s)\n",
241 			    dev->si_name, msg);
242 		}
243 	}
244 
245 	if (msg == NULL) {
246 		sp->ds_wlabel = FALSE;
247 	}
248 
249 	return (msg ? EINVAL : 0);
250 }
251 
252 
253 static void
254 disk_probe(struct disk *dp, int reprobe)
255 {
256 	struct disk_info *info = &dp->d_info;
257 	cdev_t dev = dp->d_cdev;
258 	cdev_t ndev;
259 	int error, i, sno;
260 	struct diskslice *sp;
261 
262 	KKASSERT (info->d_media_blksize != 0);
263 
264 	dp->d_slice = dsmakeslicestruct(BASE_SLICE, info);
265 	disk_debug(1,
266 		    "disk_probe (begin): %s\n",
267 			dp->d_cdev->si_name);
268 
269 	error = mbrinit(dev, info, &(dp->d_slice));
270 	if (error)
271 		return;
272 
273 	for (i = 0; i < dp->d_slice->dss_nslices; i++) {
274 		/*
275 		 * Ignore the whole-disk slice, it has already been created.
276 		 */
277 		if (i == WHOLE_DISK_SLICE)
278 			continue;
279 		sp = &dp->d_slice->dss_slices[i];
280 
281 		/*
282 		 * Handle s0.  s0 is a compatibility slice if there are no
283 		 * other slices and it has not otherwise been set up, else
284 		 * we ignore it.
285 		 */
286 		if (i == COMPATIBILITY_SLICE) {
287 			sno = 0;
288 			if (sp->ds_type == 0 &&
289 			    dp->d_slice->dss_nslices == BASE_SLICE) {
290 				sp->ds_size = info->d_media_blocks;
291 				sp->ds_reserved = 0;
292 			}
293 		} else {
294 			sno = i - 1;
295 			sp->ds_reserved = 0;
296 		}
297 
298 		/*
299 		 * Ignore 0-length slices
300 		 */
301 		if (sp->ds_size == 0)
302 			continue;
303 
304 		if (reprobe &&
305 		    (ndev = devfs_find_device_by_name("%ss%d",
306 						      dev->si_name, sno))) {
307 			/*
308 			 * Device already exists and is still valid
309 			 */
310 			ndev->si_flags |= SI_REPROBE_TEST;
311 		} else {
312 			/*
313 			 * Else create new device
314 			 */
315 			ndev = make_dev_covering(&disk_ops, dp->d_rawdev->si_ops,
316 					dkmakewholeslice(dkunit(dev), i),
317 					UID_ROOT, GID_OPERATOR, 0640,
318 					"%ss%d", dev->si_name, sno);
319 			if (dp->d_info.d_serialno) {
320 				make_dev_alias(ndev, "serno/%s.s%d",
321 					       dp->d_info.d_serialno, sno);
322 			}
323 			ndev->si_disk = dp;
324 			ndev->si_flags |= SI_REPROBE_TEST;
325 		}
326 		sp->ds_dev = ndev;
327 
328 		/*
329 		 * Probe appropriate slices for a disklabel
330 		 *
331 		 * XXX slice type 1 used by our gpt probe code.
332 		 * XXX slice type 0 used by mbr compat slice.
333 		 */
334 		if (sp->ds_type == DOSPTYP_386BSD ||
335 		    sp->ds_type == DOSPTYP_NETBSD ||
336 		    sp->ds_type == DOSPTYP_OPENBSD ||
337 		    sp->ds_type == 0 ||
338 		    sp->ds_type == 1) {
339 			if (dp->d_slice->dss_first_bsd_slice == 0)
340 				dp->d_slice->dss_first_bsd_slice = i;
341 			disk_probe_slice(dp, ndev, i, reprobe);
342 		}
343 	}
344 	disk_debug(1,
345 		    "disk_probe (end): %s\n",
346 			dp->d_cdev->si_name);
347 }
348 
349 
350 static void
351 disk_msg_core(void *arg)
352 {
353 	struct disk	*dp;
354 	struct diskslice *sp;
355 	lwkt_tokref ilock;
356 	disk_msg_t msg;
357 	int run;
358 
359 	lwkt_initport_thread(&disk_msg_port, curthread);
360 	wakeup(curthread);
361 	run = 1;
362 
363 	while (run) {
364 		msg = (disk_msg_t)lwkt_waitport(&disk_msg_port, 0);
365 
366 		switch (msg->hdr.u.ms_result) {
367 		case DISK_DISK_PROBE:
368 			dp = (struct disk *)msg->load;
369 			disk_debug(1,
370 				    "DISK_DISK_PROBE: %s\n",
371 					dp->d_cdev->si_name);
372 			disk_probe(dp, 0);
373 			break;
374 		case DISK_DISK_DESTROY:
375 			dp = (struct disk *)msg->load;
376 			disk_debug(1,
377 				    "DISK_DISK_DESTROY: %s\n",
378 					dp->d_cdev->si_name);
379 			devfs_destroy_subnames(dp->d_cdev->si_name);
380 			devfs_destroy_dev(dp->d_cdev);
381 			lwkt_gettoken(&ilock, &disklist_token);
382 			LIST_REMOVE(dp, d_list);
383 			lwkt_reltoken(&ilock);
384 			if (dp->d_info.d_serialno) {
385 				kfree(dp->d_info.d_serialno, M_TEMP);
386 				dp->d_info.d_serialno = NULL;
387 			}
388 			break;
389 		case DISK_UNPROBE:
390 			dp = (struct disk *)msg->load;
391 			disk_debug(1,
392 				    "DISK_DISK_UNPROBE: %s\n",
393 					dp->d_cdev->si_name);
394 			devfs_destroy_subnames(dp->d_cdev->si_name);
395 			break;
396 		case DISK_SLICE_REPROBE:
397 			dp = (struct disk *)msg->load;
398 			sp = (struct diskslice *)msg->load2;
399 			devfs_clr_subnames_flag(sp->ds_dev->si_name,
400 						SI_REPROBE_TEST);
401 			disk_debug(1,
402 				    "DISK_SLICE_REPROBE: %s\n",
403 				    sp->ds_dev->si_name);
404 			disk_probe_slice(dp, sp->ds_dev,
405 					 dkslice(sp->ds_dev), 1);
406 			devfs_destroy_subnames_without_flag(
407 					sp->ds_dev->si_name, SI_REPROBE_TEST);
408 			break;
409 		case DISK_DISK_REPROBE:
410 			dp = (struct disk *)msg->load;
411 			devfs_clr_subnames_flag(dp->d_cdev->si_name, SI_REPROBE_TEST);
412 			disk_debug(1,
413 				    "DISK_DISK_REPROBE: %s\n",
414 				    dp->d_cdev->si_name);
415 			disk_probe(dp, 1);
416 			devfs_destroy_subnames_without_flag(
417 					dp->d_cdev->si_name, SI_REPROBE_TEST);
418 			break;
419 		case DISK_SYNC:
420 			disk_debug(1, "DISK_SYNC\n");
421 			break;
422 		default:
423 			devfs_debug(DEVFS_DEBUG_WARNING,
424 				    "disk_msg_core: unknown message "
425 				    "received at core\n");
426 			break;
427 		}
428 		lwkt_replymsg((lwkt_msg_t)msg, 0);
429 	}
430 	lwkt_exit();
431 }
432 
433 
434 /*
435  * Acts as a message drain. Any message that is replied to here gets
436  * destroyed and the memory freed.
437  */
438 static void
439 disk_msg_autofree_reply(lwkt_port_t port, lwkt_msg_t msg)
440 {
441 	objcache_put(disk_msg_cache, msg);
442 }
443 
444 
445 void
446 disk_msg_send(uint32_t cmd, void *load, void *load2)
447 {
448 	disk_msg_t disk_msg;
449 	lwkt_port_t port = &disk_msg_port;
450 
451 	disk_msg = objcache_get(disk_msg_cache, M_WAITOK);
452 
453 	lwkt_initmsg(&disk_msg->hdr, &disk_dispose_port, 0);
454 
455 	disk_msg->hdr.u.ms_result = cmd;
456 	disk_msg->load = load;
457 	disk_msg->load2 = load2;
458 	KKASSERT(port);
459 	lwkt_sendmsg(port, (lwkt_msg_t)disk_msg);
460 }
461 
462 void
463 disk_msg_send_sync(uint32_t cmd, void *load, void *load2)
464 {
465 	struct lwkt_port rep_port;
466 	disk_msg_t disk_msg = objcache_get(disk_msg_cache, M_WAITOK);
467 	disk_msg_t	msg_incoming;
468 	lwkt_port_t port = &disk_msg_port;
469 
470 	lwkt_initport_thread(&rep_port, curthread);
471 	lwkt_initmsg(&disk_msg->hdr, &rep_port, 0);
472 
473 	disk_msg->hdr.u.ms_result = cmd;
474 	disk_msg->load = load;
475 	disk_msg->load2 = load2;
476 
477 	KKASSERT(port);
478 	lwkt_sendmsg(port, (lwkt_msg_t)disk_msg);
479 	msg_incoming = lwkt_waitport(&rep_port, 0);
480 }
481 
482 /*
483  * Create a raw device for the dev_ops template (which is returned).  Also
484  * create a slice and unit managed disk and overload the user visible
485  * device space with it.
486  *
487  * NOTE: The returned raw device is NOT a slice and unit managed device.
488  * It is an actual raw device representing the raw disk as specified by
489  * the passed dev_ops.  The disk layer not only returns such a raw device,
490  * it also uses it internally when passing (modified) commands through.
491  */
492 cdev_t
493 disk_create(int unit, struct disk *dp, struct dev_ops *raw_ops)
494 {
495 	lwkt_tokref ilock;
496 	cdev_t rawdev;
497 
498 	disk_debug(1,
499 		    "disk_create (begin): %s%d\n",
500 			raw_ops->head.name, unit);
501 
502 	rawdev = make_only_dev(raw_ops, dkmakewholedisk(unit),
503 			    UID_ROOT, GID_OPERATOR, 0640,
504 			    "%s%d", raw_ops->head.name, unit);
505 
506 	bzero(dp, sizeof(*dp));
507 
508 	dp->d_rawdev = rawdev;
509 	dp->d_raw_ops = raw_ops;
510 	dp->d_dev_ops = &disk_ops;
511 	dp->d_cdev = make_dev_covering(&disk_ops, dp->d_rawdev->si_ops,
512 			    dkmakewholedisk(unit),
513 			    UID_ROOT, GID_OPERATOR, 0640,
514 			    "%s%d", raw_ops->head.name, unit);
515 
516 	dp->d_cdev->si_disk = dp;
517 
518 	lwkt_gettoken(&ilock, &disklist_token);
519 	LIST_INSERT_HEAD(&disklist, dp, d_list);
520 	lwkt_reltoken(&ilock);
521 
522 	disk_debug(1,
523 		    "disk_create (end): %s%d\n",
524 			raw_ops->head.name, unit);
525 
526 	return (dp->d_rawdev);
527 }
528 
529 
530 static void
531 _setdiskinfo(struct disk *disk, struct disk_info *info)
532 {
533 	char *oldserialno;
534 
535 	oldserialno = disk->d_info.d_serialno;
536 	bcopy(info, &disk->d_info, sizeof(disk->d_info));
537 	info = &disk->d_info;
538 
539 	disk_debug(1,
540 		    "_setdiskinfo: %s\n",
541 			disk->d_cdev->si_name);
542 
543 	/*
544 	 * The serial number is duplicated so the caller can throw
545 	 * their copy away.
546 	 */
547 	if (info->d_serialno && info->d_serialno[0]) {
548 		info->d_serialno = kstrdup(info->d_serialno, M_TEMP);
549 		disk_cleanserial(info->d_serialno);
550 		if (disk->d_cdev) {
551 			make_dev_alias(disk->d_cdev, "serno/%s",
552 					info->d_serialno);
553 		}
554 	} else {
555 		info->d_serialno = NULL;
556 	}
557 	if (oldserialno)
558 		kfree(oldserialno, M_TEMP);
559 
560 	/*
561 	 * The caller may set d_media_size or d_media_blocks and we
562 	 * calculate the other.
563 	 */
564 	KKASSERT(info->d_media_size == 0 || info->d_media_blksize == 0);
565 	if (info->d_media_size == 0 && info->d_media_blocks) {
566 		info->d_media_size = (u_int64_t)info->d_media_blocks *
567 				     info->d_media_blksize;
568 	} else if (info->d_media_size && info->d_media_blocks == 0 &&
569 		   info->d_media_blksize) {
570 		info->d_media_blocks = info->d_media_size /
571 				       info->d_media_blksize;
572 	}
573 
574 	/*
575 	 * The si_* fields for rawdev are not set until after the
576 	 * disk_create() call, so someone using the cooked version
577 	 * of the raw device (i.e. da0s0) will not get the right
578 	 * si_iosize_max unless we fix it up here.
579 	 */
580 	if (disk->d_cdev && disk->d_rawdev &&
581 	    disk->d_cdev->si_iosize_max == 0) {
582 		disk->d_cdev->si_iosize_max = disk->d_rawdev->si_iosize_max;
583 		disk->d_cdev->si_bsize_phys = disk->d_rawdev->si_bsize_phys;
584 		disk->d_cdev->si_bsize_best = disk->d_rawdev->si_bsize_best;
585 	}
586 }
587 
588 /*
589  * Disk drivers must call this routine when media parameters are available
590  * or have changed.
591  */
592 void
593 disk_setdiskinfo(struct disk *disk, struct disk_info *info)
594 {
595 	_setdiskinfo(disk, info);
596 	disk_msg_send(DISK_DISK_PROBE, disk, NULL);
597 	disk_debug(1,
598 		    "disk_setdiskinfo: sent probe for %s\n",
599 			disk->d_cdev->si_name);
600 }
601 
602 void
603 disk_setdiskinfo_sync(struct disk *disk, struct disk_info *info)
604 {
605 	_setdiskinfo(disk, info);
606 	disk_msg_send_sync(DISK_DISK_PROBE, disk, NULL);
607 	disk_debug(1,
608 		    "disk_setdiskinfo_sync: sent probe for %s\n",
609 			disk->d_cdev->si_name);
610 }
611 
612 /*
613  * This routine is called when an adapter detaches.  The higher level
614  * managed disk device is destroyed while the lower level raw device is
615  * released.
616  */
617 void
618 disk_destroy(struct disk *disk)
619 {
620 	disk_msg_send_sync(DISK_DISK_DESTROY, disk, NULL);
621 	return;
622 }
623 
624 int
625 disk_dumpcheck(cdev_t dev, u_int64_t *size, u_int64_t *blkno, u_int32_t *secsize)
626 {
627 	struct partinfo pinfo;
628 	int error;
629 
630 	bzero(&pinfo, sizeof(pinfo));
631 	error = dev_dioctl(dev, DIOCGPART, (void *)&pinfo, 0,
632 			   proc0.p_ucred, NULL);
633 	if (error)
634 		return (error);
635 
636 	if (pinfo.media_blksize == 0)
637 		return (ENXIO);
638 
639 	if (blkno) /* XXX: make sure this reserved stuff is right */
640 		*blkno = pinfo.reserved_blocks +
641 			pinfo.media_offset / pinfo.media_blksize;
642 	if (secsize)
643 		*secsize = pinfo.media_blksize;
644 	if (size)
645 		*size = (pinfo.media_blocks - pinfo.reserved_blocks);
646 
647 	return (0);
648 }
649 
650 int
651 disk_dumpconf(cdev_t dev, u_int onoff)
652 {
653 	struct dumperinfo di;
654 	u_int64_t	size, blkno;
655 	u_int32_t	secsize;
656 	int error;
657 
658 	if (!onoff)
659 		return set_dumper(NULL);
660 
661 	error = disk_dumpcheck(dev, &size, &blkno, &secsize);
662 
663 	if (error)
664 		return ENXIO;
665 
666 	bzero(&di, sizeof(struct dumperinfo));
667 	di.dumper = diskdump;
668 	di.priv = dev;
669 	di.blocksize = secsize;
670 	di.mediaoffset = blkno * DEV_BSIZE;
671 	di.mediasize = size * DEV_BSIZE;
672 
673 	return set_dumper(&di);
674 }
675 
676 void
677 disk_unprobe(struct disk *disk)
678 {
679 	if (disk == NULL)
680 		return;
681 
682 	disk_msg_send_sync(DISK_UNPROBE, disk, NULL);
683 }
684 
685 void
686 disk_invalidate (struct disk *disk)
687 {
688 	if (disk->d_slice)
689 		dsgone(&disk->d_slice);
690 }
691 
692 struct disk *
693 disk_enumerate(struct disk *disk)
694 {
695 	struct disk *dp;
696 	lwkt_tokref ilock;
697 
698 	lwkt_gettoken(&ilock, &disklist_token);
699 	if (!disk)
700 		dp = (LIST_FIRST(&disklist));
701 	else
702 		dp = (LIST_NEXT(disk, d_list));
703 	lwkt_reltoken(&ilock);
704 
705 	return dp;
706 }
707 
708 static
709 int
710 sysctl_disks(SYSCTL_HANDLER_ARGS)
711 {
712 	struct disk *disk;
713 	int error, first;
714 
715 	disk = NULL;
716 	first = 1;
717 
718 	while ((disk = disk_enumerate(disk))) {
719 		if (!first) {
720 			error = SYSCTL_OUT(req, " ", 1);
721 			if (error)
722 				return error;
723 		} else {
724 			first = 0;
725 		}
726 		error = SYSCTL_OUT(req, disk->d_rawdev->si_name,
727 				   strlen(disk->d_rawdev->si_name));
728 		if (error)
729 			return error;
730 	}
731 	error = SYSCTL_OUT(req, "", 1);
732 	return error;
733 }
734 
735 SYSCTL_PROC(_kern, OID_AUTO, disks, CTLTYPE_STRING | CTLFLAG_RD, NULL, 0,
736     sysctl_disks, "A", "names of available disks");
737 
738 /*
739  * Open a disk device or partition.
740  */
741 static
742 int
743 diskopen(struct dev_open_args *ap)
744 {
745 	cdev_t dev = ap->a_head.a_dev;
746 	struct disk *dp;
747 	int error;
748 
749 	/*
750 	 * dp can't be NULL here XXX.
751 	 *
752 	 * d_slice will be NULL if setdiskinfo() has not been called yet.
753 	 * setdiskinfo() is typically called whether the disk is present
754 	 * or not (e.g. CD), but the base disk device is created first
755 	 * and there may be a race.
756 	 */
757 	dp = dev->si_disk;
758 	if (dp == NULL || dp->d_slice == NULL)
759 		return (ENXIO);
760 	error = 0;
761 
762 	/*
763 	 * Deal with open races
764 	 */
765 	while (dp->d_flags & DISKFLAG_LOCK) {
766 		dp->d_flags |= DISKFLAG_WANTED;
767 		error = tsleep(dp, PCATCH, "diskopen", hz);
768 		if (error)
769 			return (error);
770 	}
771 	dp->d_flags |= DISKFLAG_LOCK;
772 
773 	/*
774 	 * Open the underlying raw device.
775 	 */
776 	if (!dsisopen(dp->d_slice)) {
777 #if 0
778 		if (!pdev->si_iosize_max)
779 			pdev->si_iosize_max = dev->si_iosize_max;
780 #endif
781 		error = dev_dopen(dp->d_rawdev, ap->a_oflags,
782 				  ap->a_devtype, ap->a_cred);
783 	}
784 #if 0
785 	/*
786 	 * Inherit properties from the underlying device now that it is
787 	 * open.
788 	 */
789 	dev_dclone(dev);
790 #endif
791 
792 	if (error)
793 		goto out;
794 	error = dsopen(dev, ap->a_devtype, dp->d_info.d_dsflags,
795 		       &dp->d_slice, &dp->d_info);
796 	if (!dsisopen(dp->d_slice)) {
797 		dev_dclose(dp->d_rawdev, ap->a_oflags, ap->a_devtype);
798 	}
799 out:
800 	dp->d_flags &= ~DISKFLAG_LOCK;
801 	if (dp->d_flags & DISKFLAG_WANTED) {
802 		dp->d_flags &= ~DISKFLAG_WANTED;
803 		wakeup(dp);
804 	}
805 
806 	return(error);
807 }
808 
809 /*
810  * Close a disk device or partition
811  */
812 static
813 int
814 diskclose(struct dev_close_args *ap)
815 {
816 	cdev_t dev = ap->a_head.a_dev;
817 	struct disk *dp;
818 	int error;
819 
820 	error = 0;
821 	dp = dev->si_disk;
822 
823 	dsclose(dev, ap->a_devtype, dp->d_slice);
824 	if (!dsisopen(dp->d_slice)) {
825 		error = dev_dclose(dp->d_rawdev, ap->a_fflag, ap->a_devtype);
826 	}
827 	return (error);
828 }
829 
830 /*
831  * First execute the ioctl on the disk device, and if it isn't supported
832  * try running it on the backing device.
833  */
834 static
835 int
836 diskioctl(struct dev_ioctl_args *ap)
837 {
838 	cdev_t dev = ap->a_head.a_dev;
839 	struct disk *dp;
840 	int error;
841 	u_int u;
842 
843 	dp = dev->si_disk;
844 	if (dp == NULL)
845 		return (ENXIO);
846 
847 	devfs_debug(DEVFS_DEBUG_DEBUG,
848 		    "diskioctl: cmd is: %x (name: %s)\n",
849 		    ap->a_cmd, dev->si_name);
850 	devfs_debug(DEVFS_DEBUG_DEBUG,
851 		    "diskioctl: &dp->d_slice is: %x, %x\n",
852 		    &dp->d_slice, dp->d_slice);
853 
854 	if (ap->a_cmd == DIOCGKERNELDUMP) {
855 		u = *(u_int *)ap->a_data;
856 		return disk_dumpconf(dev, u);
857 	}
858 
859 	error = dsioctl(dev, ap->a_cmd, ap->a_data, ap->a_fflag,
860 			&dp->d_slice, &dp->d_info);
861 
862 	if (error == ENOIOCTL) {
863 		error = dev_dioctl(dp->d_rawdev, ap->a_cmd, ap->a_data,
864 				   ap->a_fflag, ap->a_cred, NULL);
865 	}
866 	return (error);
867 }
868 
869 /*
870  * Execute strategy routine
871  */
872 static
873 int
874 diskstrategy(struct dev_strategy_args *ap)
875 {
876 	cdev_t dev = ap->a_head.a_dev;
877 	struct bio *bio = ap->a_bio;
878 	struct bio *nbio;
879 	struct disk *dp;
880 
881 	dp = dev->si_disk;
882 
883 	if (dp == NULL) {
884 		bio->bio_buf->b_error = ENXIO;
885 		bio->bio_buf->b_flags |= B_ERROR;
886 		biodone(bio);
887 		return(0);
888 	}
889 	KKASSERT(dev->si_disk == dp);
890 
891 	/*
892 	 * The dscheck() function will also transform the slice relative
893 	 * block number i.e. bio->bio_offset into a block number that can be
894 	 * passed directly to the underlying raw device.  If dscheck()
895 	 * returns NULL it will have handled the bio for us (e.g. EOF
896 	 * or error due to being beyond the device size).
897 	 */
898 	if ((nbio = dscheck(dev, bio, dp->d_slice)) != NULL) {
899 		dev_dstrategy(dp->d_rawdev, nbio);
900 	} else {
901 		biodone(bio);
902 	}
903 	return(0);
904 }
905 
906 /*
907  * Return the partition size in ?blocks?
908  */
909 static
910 int
911 diskpsize(struct dev_psize_args *ap)
912 {
913 	cdev_t dev = ap->a_head.a_dev;
914 	struct disk *dp;
915 
916 	dp = dev->si_disk;
917 	if (dp == NULL)
918 		return(ENODEV);
919 	ap->a_result = dssize(dev, &dp->d_slice);
920 	return(0);
921 }
922 
923 /*
924  * When new device entries are instantiated, make sure they inherit our
925  * si_disk structure and block and iosize limits from the raw device.
926  *
927  * This routine is always called synchronously in the context of the
928  * client.
929  *
930  * XXX The various io and block size constraints are not always initialized
931  * properly by devices.
932  */
933 static
934 int
935 diskclone(struct dev_clone_args *ap)
936 {
937 	cdev_t dev = ap->a_head.a_dev;
938 	struct disk *dp;
939 	dp = dev->si_disk;
940 
941 	KKASSERT(dp != NULL);
942 	dev->si_disk = dp;
943 	dev->si_iosize_max = dp->d_rawdev->si_iosize_max;
944 	dev->si_bsize_phys = dp->d_rawdev->si_bsize_phys;
945 	dev->si_bsize_best = dp->d_rawdev->si_bsize_best;
946 	return(0);
947 }
948 
949 int
950 diskdump(struct dev_dump_args *ap)
951 {
952 	cdev_t dev = ap->a_head.a_dev;
953 	struct disk *dp = dev->si_disk;
954 	u_int64_t size, offset;
955 	int error;
956 
957 	error = disk_dumpcheck(dev, &size, &ap->a_blkno, &ap->a_secsize);
958 	/* XXX: this should probably go in disk_dumpcheck somehow */
959 	if (ap->a_length != 0) {
960 		size *= DEV_BSIZE;
961 		offset = ap->a_blkno * DEV_BSIZE;
962 		if ((ap->a_offset < offset) ||
963 		    (ap->a_offset + ap->a_length - offset > size)) {
964 			kprintf("Attempt to write outside dump device boundaries.\n");
965 			error = ENOSPC;
966 		}
967 	}
968 
969 	if (error == 0) {
970 		ap->a_head.a_dev = dp->d_rawdev;
971 		error = dev_doperate(&ap->a_head);
972 	}
973 
974 	return(error);
975 }
976 
977 
978 SYSCTL_INT(_debug_sizeof, OID_AUTO, diskslices, CTLFLAG_RD,
979     0, sizeof(struct diskslices), "sizeof(struct diskslices)");
980 
981 SYSCTL_INT(_debug_sizeof, OID_AUTO, disk, CTLFLAG_RD,
982     0, sizeof(struct disk), "sizeof(struct disk)");
983 
984 /*
985  * Reorder interval for burst write allowance and minor write
986  * allowance.
987  *
988  * We always want to trickle some writes in to make use of the
989  * disk's zone cache.  Bursting occurs on a longer interval and only
990  * runningbufspace is well over the hirunningspace limit.
991  */
992 int bioq_reorder_burst_interval = 60;	/* should be multiple of minor */
993 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_burst_interval,
994 	   CTLFLAG_RW, &bioq_reorder_burst_interval, 0, "");
995 int bioq_reorder_minor_interval = 5;
996 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_minor_interval,
997 	   CTLFLAG_RW, &bioq_reorder_minor_interval, 0, "");
998 
999 int bioq_reorder_burst_bytes = 3000000;
1000 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_burst_bytes,
1001 	   CTLFLAG_RW, &bioq_reorder_burst_bytes, 0, "");
1002 int bioq_reorder_minor_bytes = 262144;
1003 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_minor_bytes,
1004 	   CTLFLAG_RW, &bioq_reorder_minor_bytes, 0, "");
1005 
1006 
1007 /*
1008  * Order I/Os.  Generally speaking this code is designed to make better
1009  * use of drive zone caches.  A drive zone cache can typically track linear
1010  * reads or writes for around 16 zones simultaniously.
1011  *
1012  * Read prioritization issues:  It is possible for hundreds of megabytes worth
1013  * of writes to be queued asynchronously.  This creates a huge bottleneck
1014  * for reads which reduce read bandwidth to a trickle.
1015  *
1016  * To solve this problem we generally reorder reads before writes.
1017  *
1018  * However, a large number of random reads can also starve writes and
1019  * make poor use of the drive zone cache so we allow writes to trickle
1020  * in every N reads.
1021  */
1022 void
1023 bioqdisksort(struct bio_queue_head *bioq, struct bio *bio)
1024 {
1025 	/*
1026 	 * The BIO wants to be ordered.  Adding to the tail also
1027 	 * causes transition to be set to NULL, forcing the ordering
1028 	 * of all prior I/O's.
1029 	 */
1030 	if (bio->bio_buf->b_flags & B_ORDERED) {
1031 		bioq_insert_tail(bioq, bio);
1032 		return;
1033 	}
1034 
1035 	switch(bio->bio_buf->b_cmd) {
1036 	case BUF_CMD_READ:
1037 		if (bioq->transition) {
1038 			/*
1039 			 * Insert before the first write.  Bleedover writes
1040 			 * based on reorder intervals to prevent starvation.
1041 			 */
1042 			TAILQ_INSERT_BEFORE(bioq->transition, bio, bio_act);
1043 			++bioq->reorder;
1044 			if (bioq->reorder % bioq_reorder_minor_interval == 0) {
1045 				bioqwritereorder(bioq);
1046 				if (bioq->reorder >=
1047 				    bioq_reorder_burst_interval) {
1048 					bioq->reorder = 0;
1049 				}
1050 			}
1051 		} else {
1052 			/*
1053 			 * No writes queued (or ordering was forced),
1054 			 * insert at tail.
1055 			 */
1056 			TAILQ_INSERT_TAIL(&bioq->queue, bio, bio_act);
1057 		}
1058 		break;
1059 	case BUF_CMD_WRITE:
1060 		/*
1061 		 * Writes are always appended.  If no writes were previously
1062 		 * queued or an ordered tail insertion occured the transition
1063 		 * field will be NULL.
1064 		 */
1065 		TAILQ_INSERT_TAIL(&bioq->queue, bio, bio_act);
1066 		if (bioq->transition == NULL)
1067 			bioq->transition = bio;
1068 		break;
1069 	default:
1070 		/*
1071 		 * All other request types are forced to be ordered.
1072 		 */
1073 		bioq_insert_tail(bioq, bio);
1074 		break;
1075 	}
1076 }
1077 
1078 /*
1079  * Move the read-write transition point to prevent reads from
1080  * completely starving our writes.  This brings a number of writes into
1081  * the fold every N reads.
1082  *
1083  * We bring a few linear writes into the fold on a minor interval
1084  * and we bring a non-linear burst of writes into the fold on a major
1085  * interval.  Bursting only occurs if runningbufspace is really high
1086  * (typically from syncs, fsyncs, or HAMMER flushes).
1087  */
1088 static
1089 void
1090 bioqwritereorder(struct bio_queue_head *bioq)
1091 {
1092 	struct bio *bio;
1093 	off_t next_offset;
1094 	size_t left;
1095 	size_t n;
1096 	int check_off;
1097 
1098 	if (bioq->reorder < bioq_reorder_burst_interval ||
1099 	    !buf_runningbufspace_severe()) {
1100 		left = (size_t)bioq_reorder_minor_bytes;
1101 		check_off = 1;
1102 	} else {
1103 		left = (size_t)bioq_reorder_burst_bytes;
1104 		check_off = 0;
1105 	}
1106 
1107 	next_offset = bioq->transition->bio_offset;
1108 	while ((bio = bioq->transition) != NULL &&
1109 	       (check_off == 0 || next_offset == bio->bio_offset)
1110 	) {
1111 		n = bio->bio_buf->b_bcount;
1112 		next_offset = bio->bio_offset + n;
1113 		bioq->transition = TAILQ_NEXT(bio, bio_act);
1114 		if (left < n)
1115 			break;
1116 		left -= n;
1117 	}
1118 }
1119 
1120 /*
1121  * Disk error is the preface to plaintive error messages
1122  * about failing disk transfers.  It prints messages of the form
1123 
1124 hp0g: hard error reading fsbn 12345 of 12344-12347 (hp0 bn %d cn %d tn %d sn %d)
1125 
1126  * if the offset of the error in the transfer and a disk label
1127  * are both available.  blkdone should be -1 if the position of the error
1128  * is unknown; the disklabel pointer may be null from drivers that have not
1129  * been converted to use them.  The message is printed with kprintf
1130  * if pri is LOG_PRINTF, otherwise it uses log at the specified priority.
1131  * The message should be completed (with at least a newline) with kprintf
1132  * or log(-1, ...), respectively.  There is no trailing space.
1133  */
1134 void
1135 diskerr(struct bio *bio, cdev_t dev, const char *what, int pri, int donecnt)
1136 {
1137 	struct buf *bp = bio->bio_buf;
1138 	const char *term;
1139 
1140 	switch(bp->b_cmd) {
1141 	case BUF_CMD_READ:
1142 		term = "read";
1143 		break;
1144 	case BUF_CMD_WRITE:
1145 		term = "write";
1146 		break;
1147 	default:
1148 		term = "access";
1149 		break;
1150 	}
1151 	kprintf("%s: %s %sing ", dev->si_name, what, term);
1152 	kprintf("offset %012llx for %d",
1153 		(long long)bio->bio_offset,
1154 		bp->b_bcount);
1155 
1156 	if (donecnt)
1157 		kprintf(" (%d bytes completed)", donecnt);
1158 }
1159 
1160 /*
1161  * Locate a disk device
1162  */
1163 cdev_t
1164 disk_locate(const char *devname)
1165 {
1166 	return devfs_find_device_by_name(devname);
1167 }
1168 
1169 void
1170 disk_config(void *arg)
1171 {
1172 	disk_msg_send_sync(DISK_SYNC, NULL, NULL);
1173 }
1174 
1175 static void
1176 disk_init(void)
1177 {
1178 	struct thread* td_core;
1179 
1180 	disk_msg_cache = objcache_create("disk-msg-cache", 0, 0,
1181 					 NULL, NULL, NULL,
1182 					 objcache_malloc_alloc,
1183 					 objcache_malloc_free,
1184 					 &disk_msg_malloc_args);
1185 
1186 	lwkt_token_init(&disklist_token);
1187 
1188 	/*
1189 	 * Initialize the reply-only port which acts as a message drain
1190 	 */
1191 	lwkt_initport_replyonly(&disk_dispose_port, disk_msg_autofree_reply);
1192 
1193 	lwkt_create(disk_msg_core, /*args*/NULL, &td_core, NULL,
1194 		    0, 0, "disk_msg_core");
1195 
1196 	tsleep(td_core, 0, "diskcore", 0);
1197 }
1198 
1199 static void
1200 disk_uninit(void)
1201 {
1202 	objcache_destroy(disk_msg_cache);
1203 }
1204 
1205 /*
1206  * Clean out illegal characters in serial numbers.
1207  */
1208 static void
1209 disk_cleanserial(char *serno)
1210 {
1211 	char c;
1212 
1213 	while ((c = *serno) != 0) {
1214 		if (c >= 'a' && c <= 'z')
1215 			;
1216 		else if (c >= 'A' && c <= 'Z')
1217 			;
1218 		else if (c >= '0' && c <= '9')
1219 			;
1220 		else if (c == '-' || c == '@' || c == '+' || c == '.')
1221 			;
1222 		else
1223 			c = '_';
1224 		*serno++= c;
1225 	}
1226 }
1227 
1228 TUNABLE_INT("kern.disk_debug", &disk_debug_enable);
1229 SYSCTL_INT(_kern, OID_AUTO, disk_debug, CTLFLAG_RW, &disk_debug_enable,
1230 		0, "Enable subr_disk debugging");
1231 
1232 SYSINIT(disk_register, SI_SUB_PRE_DRIVERS, SI_ORDER_FIRST, disk_init, NULL);
1233 SYSUNINIT(disk_register, SI_SUB_PRE_DRIVERS, SI_ORDER_ANY, disk_uninit, NULL);
1234