xref: /dflybsd-src/sys/kern/subr_disk.c (revision e54488bbec5c9f80e95cedd395b0e3d31fde253d)
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/malloc.h>
97 #include <sys/sysctl.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/msgport2.h>
104 #include <sys/buf2.h>
105 #include <sys/devfs.h>
106 #include <sys/thread.h>
107 #include <sys/thread2.h>
108 
109 #include <sys/queue.h>
110 #include <sys/lock.h>
111 
112 static MALLOC_DEFINE(M_DISK, "disk", "disk data");
113 
114 static void disk_msg_autofree_reply(lwkt_port_t, lwkt_msg_t);
115 static void disk_msg_core(void *);
116 static int disk_probe_slice(struct disk *dp, cdev_t dev, int slice, int reprobe);
117 static void disk_probe(struct disk *dp, int reprobe);
118 static void _setdiskinfo(struct disk *disk, struct disk_info *info);
119 
120 static d_open_t diskopen;
121 static d_close_t diskclose;
122 static d_ioctl_t diskioctl;
123 static d_strategy_t diskstrategy;
124 static d_psize_t diskpsize;
125 static d_clone_t diskclone;
126 static d_dump_t diskdump;
127 
128 static LIST_HEAD(, disk) disklist = LIST_HEAD_INITIALIZER(&disklist);
129 static struct lwkt_token disklist_token;
130 
131 static struct dev_ops disk_ops = {
132 	{ "disk", 0, D_DISK },
133 	.d_open = diskopen,
134 	.d_close = diskclose,
135 	.d_read = physread,
136 	.d_write = physwrite,
137 	.d_ioctl = diskioctl,
138 	.d_strategy = diskstrategy,
139 	.d_dump = diskdump,
140 	.d_psize = diskpsize,
141 	.d_clone = diskclone
142 };
143 
144 static struct objcache 	*disk_msg_cache;
145 
146 struct objcache_malloc_args disk_msg_malloc_args = {
147 	sizeof(struct disk_msg), M_DISK };
148 
149 static struct lwkt_port disk_dispose_port;
150 static struct lwkt_port disk_msg_port;
151 
152 
153 static int
154 disk_probe_slice(struct disk *dp, cdev_t dev, int slice, int reprobe)
155 {
156 	struct disk_info *info = &dp->d_info;
157 	struct diskslice *sp = &dp->d_slice->dss_slices[slice];
158 	disklabel_ops_t ops;
159 	struct partinfo part;
160 	const char *msg;
161 	cdev_t ndev;
162 	int sno;
163 	u_int i;
164 
165 	sno = slice ? slice - 1 : 0;
166 
167 	ops = &disklabel32_ops;
168 	msg = ops->op_readdisklabel(dev, sp, &sp->ds_label, info);
169 	if (msg && !strcmp(msg, "no disk label")) {
170 		ops = &disklabel64_ops;
171 		msg = ops->op_readdisklabel(dev, sp, &sp->ds_label, info);
172 	}
173 	if (msg == NULL) {
174 		if (slice != WHOLE_DISK_SLICE)
175 			ops->op_adjust_label_reserved(dp->d_slice, slice, sp);
176 		else
177 			sp->ds_reserved = 0;
178 
179 		sp->ds_ops = ops;
180 		for (i = 0; i < ops->op_getnumparts(sp->ds_label); i++) {
181 			ops->op_loadpartinfo(sp->ds_label, i, &part);
182 			if (part.fstype) {
183 				if (reprobe &&
184 				    (ndev = devfs_find_device_by_name("%s%c",
185 						dev->si_name, 'a' + i))
186 				) {
187 					/*
188 					 * Device already exists and
189 					 * is still valid.
190 					 */
191 					ndev->si_flags |= SI_REPROBE_TEST;
192 				} else {
193 					ndev = make_dev(&disk_ops,
194 						dkmakeminor(dkunit(dp->d_cdev),
195 							    slice, i),
196 						UID_ROOT, GID_OPERATOR, 0640,
197 						"%s%c", dev->si_name, 'a'+ i);
198 					ndev->si_disk = dp;
199 					if (dp->d_info.d_serialno) {
200 						make_dev_alias(ndev,
201 						    "serno/%s.s%d%c",
202 						    dp->d_info.d_serialno,
203 						    sno, 'a' + i);
204 					}
205 					ndev->si_flags |= SI_REPROBE_TEST;
206 				}
207 			}
208 		}
209 	} else if (info->d_dsflags & DSO_COMPATLABEL) {
210 		msg = NULL;
211 		if (sp->ds_size >= 0x100000000ULL)
212 			ops = &disklabel64_ops;
213 		else
214 			ops = &disklabel32_ops;
215 		sp->ds_label = ops->op_clone_label(info, sp);
216 	} else {
217 		if (sp->ds_type == DOSPTYP_386BSD /* XXX */) {
218 			log(LOG_WARNING, "%s: cannot find label (%s)\n",
219 			    dev->si_name, msg);
220 		}
221 	}
222 
223 	if (msg == NULL) {
224 		sp->ds_wlabel = FALSE;
225 	}
226 
227 	return (msg ? EINVAL : 0);
228 }
229 
230 
231 static void
232 disk_probe(struct disk *dp, int reprobe)
233 {
234 	struct disk_info *info = &dp->d_info;
235 	cdev_t dev = dp->d_cdev;
236 	cdev_t ndev;
237 	int error, i, sno;
238 	struct diskslice *sp;
239 
240 	KKASSERT (info->d_media_blksize != 0);
241 
242 	dp->d_slice = dsmakeslicestruct(BASE_SLICE, info);
243 
244 	error = mbrinit(dev, info, &(dp->d_slice));
245 	if (error)
246 		return;
247 
248 	for (i = 0; i < dp->d_slice->dss_nslices; i++) {
249 		/*
250 		 * Ignore the whole-disk slice, it has already been created.
251 		 */
252 		if (i == WHOLE_DISK_SLICE)
253 			continue;
254 		sp = &dp->d_slice->dss_slices[i];
255 
256 		/*
257 		 * Handle s0.  s0 is a compatibility slice if there are no
258 		 * other slices and it has not otherwise been set up, else
259 		 * we ignore it.
260 		 */
261 		if (i == COMPATIBILITY_SLICE) {
262 			sno = 0;
263 			if (sp->ds_type == 0 &&
264 			    dp->d_slice->dss_nslices == BASE_SLICE) {
265 				sp->ds_size = info->d_media_blocks;
266 				sp->ds_reserved = 0;
267 			}
268 		} else {
269 			sno = i - 1;
270 			sp->ds_reserved = 0;
271 		}
272 
273 		/*
274 		 * Ignore 0-length slices
275 		 */
276 		if (sp->ds_size == 0)
277 			continue;
278 
279 		if (reprobe &&
280 		    (ndev = devfs_find_device_by_name("%ss%d",
281 						      dev->si_name, sno))) {
282 			/*
283 			 * Device already exists and is still valid
284 			 */
285 			ndev->si_flags |= SI_REPROBE_TEST;
286 		} else {
287 			/*
288 			 * Else create new device
289 			 */
290 			ndev = make_dev(&disk_ops,
291 					dkmakewholeslice(dkunit(dev), i),
292 					UID_ROOT, GID_OPERATOR, 0640,
293 					"%ss%d", dev->si_name, sno);
294 			if (dp->d_info.d_serialno) {
295 				make_dev_alias(ndev, "serno/%s.s%d",
296 					       dp->d_info.d_serialno, sno);
297 			}
298 			ndev->si_disk = dp;
299 			ndev->si_flags |= SI_REPROBE_TEST;
300 		}
301 		sp->ds_dev = ndev;
302 
303 		/*
304 		 * Probe appropriate slices for a disklabel
305 		 *
306 		 * XXX slice type 1 used by our gpt probe code.
307 		 * XXX slice type 0 used by mbr compat slice.
308 		 */
309 		if (sp->ds_type == DOSPTYP_386BSD || sp->ds_type == 0 ||
310 			sp->ds_type == 1) {
311 			if (dp->d_slice->dss_first_bsd_slice == 0)
312 				dp->d_slice->dss_first_bsd_slice = i;
313 			disk_probe_slice(dp, ndev, i, reprobe);
314 		}
315 	}
316 }
317 
318 
319 static void
320 disk_msg_core(void *arg)
321 {
322 	struct disk	*dp;
323 	struct diskslice *sp;
324 	lwkt_tokref ilock;
325 	disk_msg_t msg;
326 	int run;
327 
328 	lwkt_initport_thread(&disk_msg_port, curthread);
329 	wakeup(curthread);
330 	run = 1;
331 
332 	while (run) {
333 		msg = (disk_msg_t)lwkt_waitport(&disk_msg_port, 0);
334 
335 		switch (msg->hdr.u.ms_result) {
336 		case DISK_DISK_PROBE:
337 			dp = (struct disk *)msg->load;
338 			disk_probe(dp, 0);
339 			break;
340 		case DISK_DISK_DESTROY:
341 			dp = (struct disk *)msg->load;
342 			devfs_destroy_subnames(dp->d_cdev->si_name);
343 			devfs_destroy_dev(dp->d_cdev);
344 			lwkt_gettoken(&ilock, &disklist_token);
345 			LIST_REMOVE(dp, d_list);
346 			lwkt_reltoken(&ilock);
347 			if (dp->d_info.d_serialno) {
348 				kfree(dp->d_info.d_serialno, M_TEMP);
349 				dp->d_info.d_serialno = NULL;
350 			}
351 			break;
352 		case DISK_UNPROBE:
353 			dp = (struct disk *)msg->load;
354 			devfs_destroy_subnames(dp->d_cdev->si_name);
355 			break;
356 		case DISK_SLICE_REPROBE:
357 			dp = (struct disk *)msg->load;
358 			sp = (struct diskslice *)msg->load2;
359 			devfs_clr_subnames_flag(sp->ds_dev->si_name,
360 						SI_REPROBE_TEST);
361 			devfs_debug(DEVFS_DEBUG_DEBUG,
362 				    "DISK_SLICE_REPROBE: %s\n",
363 				    sp->ds_dev->si_name);
364 			disk_probe_slice(dp, sp->ds_dev,
365 					 dkslice(sp->ds_dev), 1);
366 			devfs_destroy_subnames_without_flag(
367 					sp->ds_dev->si_name, SI_REPROBE_TEST);
368 			break;
369 		case DISK_DISK_REPROBE:
370 			dp = (struct disk *)msg->load;
371 			devfs_clr_subnames_flag(dp->d_cdev->si_name, SI_REPROBE_TEST);
372 			devfs_debug(DEVFS_DEBUG_DEBUG,
373 				    "DISK_DISK_REPROBE: %s\n",
374 				    dp->d_cdev->si_name);
375 			disk_probe(dp, 1);
376 			devfs_destroy_subnames_without_flag(
377 					dp->d_cdev->si_name, SI_REPROBE_TEST);
378 			break;
379 		case DISK_SYNC:
380 			break;
381 		default:
382 			devfs_debug(DEVFS_DEBUG_WARNING,
383 				    "disk_msg_core: unknown message "
384 				    "received at core\n");
385 			break;
386 		}
387 		lwkt_replymsg((lwkt_msg_t)msg, 0);
388 	}
389 	lwkt_exit();
390 }
391 
392 
393 /*
394  * Acts as a message drain. Any message that is replied to here gets
395  * destroyed and the memory freed.
396  */
397 static void
398 disk_msg_autofree_reply(lwkt_port_t port, lwkt_msg_t msg)
399 {
400 	objcache_put(disk_msg_cache, msg);
401 }
402 
403 
404 void
405 disk_msg_send(uint32_t cmd, void *load, void *load2)
406 {
407 	disk_msg_t disk_msg;
408 	lwkt_port_t port = &disk_msg_port;
409 
410 	disk_msg = objcache_get(disk_msg_cache, M_WAITOK);
411 
412 	lwkt_initmsg(&disk_msg->hdr, &disk_dispose_port, 0);
413 
414 	disk_msg->hdr.u.ms_result = cmd;
415 	disk_msg->load = load;
416 	disk_msg->load2 = load2;
417 	KKASSERT(port);
418 	lwkt_sendmsg(port, (lwkt_msg_t)disk_msg);
419 }
420 
421 void
422 disk_msg_send_sync(uint32_t cmd, void *load, void *load2)
423 {
424 	struct lwkt_port rep_port;
425 	disk_msg_t disk_msg = objcache_get(disk_msg_cache, M_WAITOK);
426 	disk_msg_t	msg_incoming;
427 	lwkt_port_t port = &disk_msg_port;
428 
429 	lwkt_initport_thread(&rep_port, curthread);
430 	lwkt_initmsg(&disk_msg->hdr, &rep_port, 0);
431 
432 	disk_msg->hdr.u.ms_result = cmd;
433 	disk_msg->load = load;
434 	disk_msg->load2 = load2;
435 
436 	KKASSERT(port);
437 	lwkt_sendmsg(port, (lwkt_msg_t)disk_msg);
438 	msg_incoming = lwkt_waitport(&rep_port, 0);
439 }
440 
441 /*
442  * Create a raw device for the dev_ops template (which is returned).  Also
443  * create a slice and unit managed disk and overload the user visible
444  * device space with it.
445  *
446  * NOTE: The returned raw device is NOT a slice and unit managed device.
447  * It is an actual raw device representing the raw disk as specified by
448  * the passed dev_ops.  The disk layer not only returns such a raw device,
449  * it also uses it internally when passing (modified) commands through.
450  */
451 cdev_t
452 disk_create(int unit, struct disk *dp, struct dev_ops *raw_ops)
453 {
454 	lwkt_tokref ilock;
455 	cdev_t rawdev;
456 
457 	rawdev = make_only_dev(raw_ops, dkmakewholedisk(unit),
458 			    UID_ROOT, GID_OPERATOR, 0640,
459 			    "%s%d", raw_ops->head.name, unit);
460 
461 	bzero(dp, sizeof(*dp));
462 
463 	dp->d_rawdev = rawdev;
464 	dp->d_raw_ops = raw_ops;
465 	dp->d_dev_ops = &disk_ops;
466 	dp->d_cdev = make_dev(&disk_ops,
467 			    dkmakewholedisk(unit),
468 			    UID_ROOT, GID_OPERATOR, 0640,
469 			    "%s%d", raw_ops->head.name, unit);
470 
471 	dp->d_cdev->si_disk = dp;
472 
473 	lwkt_gettoken(&ilock, &disklist_token);
474 	LIST_INSERT_HEAD(&disklist, dp, d_list);
475 	lwkt_reltoken(&ilock);
476 	return (dp->d_rawdev);
477 }
478 
479 
480 static void
481 _setdiskinfo(struct disk *disk, struct disk_info *info)
482 {
483 	char *oldserialno;
484 
485 	oldserialno = disk->d_info.d_serialno;
486 	bcopy(info, &disk->d_info, sizeof(disk->d_info));
487 	info = &disk->d_info;
488 
489 	/*
490 	 * The serial number is duplicated so the caller can throw
491 	 * their copy away.
492 	 */
493 	if (info->d_serialno && info->d_serialno[0]) {
494 		info->d_serialno = kstrdup(info->d_serialno, M_TEMP);
495 		if (disk->d_cdev) {
496 			make_dev_alias(disk->d_cdev, "serno/%s",
497 					info->d_serialno);
498 		}
499 	} else {
500 		info->d_serialno = NULL;
501 	}
502 	if (oldserialno)
503 		kfree(oldserialno, M_TEMP);
504 
505 	/*
506 	 * The caller may set d_media_size or d_media_blocks and we
507 	 * calculate the other.
508 	 */
509 	KKASSERT(info->d_media_size == 0 || info->d_media_blksize == 0);
510 	if (info->d_media_size == 0 && info->d_media_blocks) {
511 		info->d_media_size = (u_int64_t)info->d_media_blocks *
512 				     info->d_media_blksize;
513 	} else if (info->d_media_size && info->d_media_blocks == 0 &&
514 		   info->d_media_blksize) {
515 		info->d_media_blocks = info->d_media_size /
516 				       info->d_media_blksize;
517 	}
518 
519 	/*
520 	 * The si_* fields for rawdev are not set until after the
521 	 * disk_create() call, so someone using the cooked version
522 	 * of the raw device (i.e. da0s0) will not get the right
523 	 * si_iosize_max unless we fix it up here.
524 	 */
525 	if (disk->d_cdev && disk->d_rawdev &&
526 	    disk->d_cdev->si_iosize_max == 0) {
527 		disk->d_cdev->si_iosize_max = disk->d_rawdev->si_iosize_max;
528 		disk->d_cdev->si_bsize_phys = disk->d_rawdev->si_bsize_phys;
529 		disk->d_cdev->si_bsize_best = disk->d_rawdev->si_bsize_best;
530 	}
531 }
532 
533 /*
534  * Disk drivers must call this routine when media parameters are available
535  * or have changed.
536  */
537 void
538 disk_setdiskinfo(struct disk *disk, struct disk_info *info)
539 {
540 	_setdiskinfo(disk, info);
541 	disk_msg_send(DISK_DISK_PROBE, disk, NULL);
542 }
543 
544 void
545 disk_setdiskinfo_sync(struct disk *disk, struct disk_info *info)
546 {
547 	_setdiskinfo(disk, info);
548 	disk_msg_send_sync(DISK_DISK_PROBE, disk, NULL);
549 }
550 
551 /*
552  * This routine is called when an adapter detaches.  The higher level
553  * managed disk device is destroyed while the lower level raw device is
554  * released.
555  */
556 void
557 disk_destroy(struct disk *disk)
558 {
559 	disk_msg_send_sync(DISK_DISK_DESTROY, disk, NULL);
560 	return;
561 }
562 
563 int
564 disk_dumpcheck(cdev_t dev, u_int64_t *count, u_int64_t *blkno, u_int *secsize)
565 {
566 	struct partinfo pinfo;
567 	int error;
568 
569 	bzero(&pinfo, sizeof(pinfo));
570 	error = dev_dioctl(dev, DIOCGPART, (void *)&pinfo, 0, proc0.p_ucred);
571 	if (error)
572 		return (error);
573 	if (pinfo.media_blksize == 0)
574 		return (ENXIO);
575 	*count = (u_int64_t)Maxmem * PAGE_SIZE / pinfo.media_blksize;
576 	if (dumplo64 < pinfo.reserved_blocks ||
577 	    dumplo64 + *count > pinfo.media_blocks) {
578 		return (ENOSPC);
579 	}
580 	*blkno = dumplo64 + pinfo.media_offset / pinfo.media_blksize;
581 	*secsize = pinfo.media_blksize;
582 	return (0);
583 }
584 
585 void
586 disk_unprobe(struct disk *disk)
587 {
588 	if (disk == NULL)
589 		return;
590 
591 	disk_msg_send_sync(DISK_UNPROBE, disk, NULL);
592 }
593 
594 void
595 disk_invalidate (struct disk *disk)
596 {
597 	if (disk->d_slice)
598 		dsgone(&disk->d_slice);
599 }
600 
601 struct disk *
602 disk_enumerate(struct disk *disk)
603 {
604 	struct disk *dp;
605 	lwkt_tokref ilock;
606 
607 	lwkt_gettoken(&ilock, &disklist_token);
608 	if (!disk)
609 		dp = (LIST_FIRST(&disklist));
610 	else
611 		dp = (LIST_NEXT(disk, d_list));
612 	lwkt_reltoken(&ilock);
613 
614 	return dp;
615 }
616 
617 static
618 int
619 sysctl_disks(SYSCTL_HANDLER_ARGS)
620 {
621 	struct disk *disk;
622 	int error, first;
623 
624 	disk = NULL;
625 	first = 1;
626 
627 	while ((disk = disk_enumerate(disk))) {
628 		if (!first) {
629 			error = SYSCTL_OUT(req, " ", 1);
630 			if (error)
631 				return error;
632 		} else {
633 			first = 0;
634 		}
635 		error = SYSCTL_OUT(req, disk->d_rawdev->si_name,
636 				   strlen(disk->d_rawdev->si_name));
637 		if (error)
638 			return error;
639 	}
640 	error = SYSCTL_OUT(req, "", 1);
641 	return error;
642 }
643 
644 SYSCTL_PROC(_kern, OID_AUTO, disks, CTLTYPE_STRING | CTLFLAG_RD, NULL, 0,
645     sysctl_disks, "A", "names of available disks");
646 
647 /*
648  * Open a disk device or partition.
649  */
650 static
651 int
652 diskopen(struct dev_open_args *ap)
653 {
654 	cdev_t dev = ap->a_head.a_dev;
655 	struct disk *dp;
656 	int error;
657 
658 	/*
659 	 * dp can't be NULL here XXX.
660 	 *
661 	 * d_slice will be NULL if setdiskinfo() has not been called yet.
662 	 * setdiskinfo() is typically called whether the disk is present
663 	 * or not (e.g. CD), but the base disk device is created first
664 	 * and there may be a race.
665 	 */
666 	dp = dev->si_disk;
667 	if (dp == NULL || dp->d_slice == NULL)
668 		return (ENXIO);
669 	error = 0;
670 
671 	/*
672 	 * Deal with open races
673 	 */
674 	while (dp->d_flags & DISKFLAG_LOCK) {
675 		dp->d_flags |= DISKFLAG_WANTED;
676 		error = tsleep(dp, PCATCH, "diskopen", hz);
677 		if (error)
678 			return (error);
679 	}
680 	dp->d_flags |= DISKFLAG_LOCK;
681 
682 	/*
683 	 * Open the underlying raw device.
684 	 */
685 	if (!dsisopen(dp->d_slice)) {
686 #if 0
687 		if (!pdev->si_iosize_max)
688 			pdev->si_iosize_max = dev->si_iosize_max;
689 #endif
690 		error = dev_dopen(dp->d_rawdev, ap->a_oflags,
691 				  ap->a_devtype, ap->a_cred);
692 	}
693 #if 0
694 	/*
695 	 * Inherit properties from the underlying device now that it is
696 	 * open.
697 	 */
698 	dev_dclone(dev);
699 #endif
700 
701 	if (error)
702 		goto out;
703 	error = dsopen(dev, ap->a_devtype, dp->d_info.d_dsflags,
704 		       &dp->d_slice, &dp->d_info);
705 	if (!dsisopen(dp->d_slice)) {
706 		dev_dclose(dp->d_rawdev, ap->a_oflags, ap->a_devtype);
707 	}
708 out:
709 	dp->d_flags &= ~DISKFLAG_LOCK;
710 	if (dp->d_flags & DISKFLAG_WANTED) {
711 		dp->d_flags &= ~DISKFLAG_WANTED;
712 		wakeup(dp);
713 	}
714 
715 	return(error);
716 }
717 
718 /*
719  * Close a disk device or partition
720  */
721 static
722 int
723 diskclose(struct dev_close_args *ap)
724 {
725 	cdev_t dev = ap->a_head.a_dev;
726 	struct disk *dp;
727 	int error;
728 
729 	error = 0;
730 	dp = dev->si_disk;
731 
732 	dsclose(dev, ap->a_devtype, dp->d_slice);
733 	if (!dsisopen(dp->d_slice)) {
734 		error = dev_dclose(dp->d_rawdev, ap->a_fflag, ap->a_devtype);
735 	}
736 	return (error);
737 }
738 
739 /*
740  * First execute the ioctl on the disk device, and if it isn't supported
741  * try running it on the backing device.
742  */
743 static
744 int
745 diskioctl(struct dev_ioctl_args *ap)
746 {
747 	cdev_t dev = ap->a_head.a_dev;
748 	struct disk *dp;
749 	int error;
750 
751 	dp = dev->si_disk;
752 	if (dp == NULL)
753 		return (ENXIO);
754 
755 	devfs_debug(DEVFS_DEBUG_DEBUG,
756 		    "diskioctl: cmd is: %x (name: %s)\n",
757 		    ap->a_cmd, dev->si_name);
758 	devfs_debug(DEVFS_DEBUG_DEBUG,
759 		    "diskioctl: &dp->d_slice is: %x, %x\n",
760 		    &dp->d_slice, dp->d_slice);
761 
762 	error = dsioctl(dev, ap->a_cmd, ap->a_data, ap->a_fflag,
763 			&dp->d_slice, &dp->d_info);
764 
765 	if (error == ENOIOCTL) {
766 		error = dev_dioctl(dp->d_rawdev, ap->a_cmd, ap->a_data,
767 				   ap->a_fflag, ap->a_cred);
768 	}
769 	return (error);
770 }
771 
772 /*
773  * Execute strategy routine
774  */
775 static
776 int
777 diskstrategy(struct dev_strategy_args *ap)
778 {
779 	cdev_t dev = ap->a_head.a_dev;
780 	struct bio *bio = ap->a_bio;
781 	struct bio *nbio;
782 	struct disk *dp;
783 
784 	dp = dev->si_disk;
785 
786 	if (dp == NULL) {
787 		bio->bio_buf->b_error = ENXIO;
788 		bio->bio_buf->b_flags |= B_ERROR;
789 		biodone(bio);
790 		return(0);
791 	}
792 	KKASSERT(dev->si_disk == dp);
793 
794 	/*
795 	 * The dscheck() function will also transform the slice relative
796 	 * block number i.e. bio->bio_offset into a block number that can be
797 	 * passed directly to the underlying raw device.  If dscheck()
798 	 * returns NULL it will have handled the bio for us (e.g. EOF
799 	 * or error due to being beyond the device size).
800 	 */
801 	if ((nbio = dscheck(dev, bio, dp->d_slice)) != NULL) {
802 		dev_dstrategy(dp->d_rawdev, nbio);
803 	} else {
804 		biodone(bio);
805 	}
806 	return(0);
807 }
808 
809 /*
810  * Return the partition size in ?blocks?
811  */
812 static
813 int
814 diskpsize(struct dev_psize_args *ap)
815 {
816 	cdev_t dev = ap->a_head.a_dev;
817 	struct disk *dp;
818 
819 	dp = dev->si_disk;
820 	if (dp == NULL)
821 		return(ENODEV);
822 	ap->a_result = dssize(dev, &dp->d_slice);
823 	return(0);
824 }
825 
826 /*
827  * When new device entries are instantiated, make sure they inherit our
828  * si_disk structure and block and iosize limits from the raw device.
829  *
830  * This routine is always called synchronously in the context of the
831  * client.
832  *
833  * XXX The various io and block size constraints are not always initialized
834  * properly by devices.
835  */
836 static
837 int
838 diskclone(struct dev_clone_args *ap)
839 {
840 	cdev_t dev = ap->a_head.a_dev;
841 	struct disk *dp;
842 	dp = dev->si_disk;
843 
844 	KKASSERT(dp != NULL);
845 	dev->si_disk = dp;
846 	dev->si_iosize_max = dp->d_rawdev->si_iosize_max;
847 	dev->si_bsize_phys = dp->d_rawdev->si_bsize_phys;
848 	dev->si_bsize_best = dp->d_rawdev->si_bsize_best;
849 	return(0);
850 }
851 
852 int
853 diskdump(struct dev_dump_args *ap)
854 {
855 	cdev_t dev = ap->a_head.a_dev;
856 	struct disk *dp = dev->si_disk;
857 	int error;
858 
859 	error = disk_dumpcheck(dev, &ap->a_count, &ap->a_blkno, &ap->a_secsize);
860 	if (error == 0) {
861 		ap->a_head.a_dev = dp->d_rawdev;
862 		error = dev_doperate(&ap->a_head);
863 	}
864 
865 	return(error);
866 }
867 
868 
869 SYSCTL_INT(_debug_sizeof, OID_AUTO, diskslices, CTLFLAG_RD,
870     0, sizeof(struct diskslices), "sizeof(struct diskslices)");
871 
872 SYSCTL_INT(_debug_sizeof, OID_AUTO, disk, CTLFLAG_RD,
873     0, sizeof(struct disk), "sizeof(struct disk)");
874 
875 /*
876  * How sorted do we want to be?  The higher the number the harder we try
877  * to sort, but also the higher the risk of bio's getting starved do
878  * to insertions in front of them.
879  */
880 static int bioq_barrier = 16;
881 SYSCTL_INT(_kern, OID_AUTO, bioq_barrier, CTLFLAG_RW, &bioq_barrier, 0, "");
882 
883 
884 /*
885  * Seek sort for disks.
886  *
887  * The bio_queue keep two queues, sorted in ascending block order.  The first
888  * queue holds those requests which are positioned after the current block
889  * (in the first request); the second, which starts at queue->switch_point,
890  * holds requests which came in after their block number was passed.  Thus
891  * we implement a one way scan, retracting after reaching the end of the drive
892  * to the first request on the second queue, at which time it becomes the
893  * first queue.
894  *
895  * A one-way scan is natural because of the way UNIX read-ahead blocks are
896  * allocated.
897  */
898 void
899 bioqdisksort(struct bio_queue_head *bioq, struct bio *bio)
900 {
901 	struct bio *bq;
902 	struct bio *bn;
903 	struct bio *be;
904 
905 	be = TAILQ_LAST(&bioq->queue, bio_queue);
906 
907 	/*
908 	 * If the queue is empty or we are an
909 	 * ordered transaction, then it's easy.
910 	 */
911 	if ((bq = bioq_first(bioq)) == NULL ||
912 	    (bio->bio_buf->b_flags & B_ORDERED) != 0) {
913 		bioq_insert_tail(bioq, bio);
914 		return;
915 	}
916 
917 	/*
918 	 * Avoid permanent request starvation by forcing the request to
919 	 * be ordered every 16 requests.  Without this long sequential
920 	 * write pipelines can prevent requests later in the queue from
921 	 * getting serviced for many seconds.
922 	 */
923 	if (++bioq->order_count >= bioq_barrier) {
924 		bioq_insert_tail_order(bioq, bio, 1);
925 		return;
926 	}
927 
928 	if (bioq->insert_point != NULL) {
929 		/*
930 		 * A certain portion of the list is
931 		 * "locked" to preserve ordering, so
932 		 * we can only insert after the insert
933 		 * point.
934 		 */
935 		bq = bioq->insert_point;
936 	} else {
937 		/*
938 		 * If we lie before the last removed (currently active)
939 		 * request, and are not inserting ourselves into the
940 		 * "locked" portion of the list, then we must add ourselves
941 		 * to the second request list.
942 		 */
943 		if (bio->bio_offset < bioq->last_offset) {
944 			bq = bioq->switch_point;
945 
946 			/*
947 			 * If we are starting a new secondary list,
948 			 * then it's easy.
949 			 */
950 			if (bq == NULL) {
951 				bioq->switch_point = bio;
952 				bioq_insert_tail(bioq, bio);
953 				return;
954 			}
955 
956 			/*
957 			 * If we lie ahead of the current switch point,
958 			 * insert us before the switch point and move
959 			 * the switch point.
960 			 */
961 			if (bio->bio_offset < bq->bio_offset) {
962 				bioq->switch_point = bio;
963 				TAILQ_INSERT_BEFORE(bq, bio, bio_act);
964 				return;
965 			}
966 		} else {
967 			if (bioq->switch_point != NULL)
968 				be = TAILQ_PREV(bioq->switch_point,
969 						bio_queue, bio_act);
970 			/*
971 			 * If we lie between last_offset and bq,
972 			 * insert before bq.
973 			 */
974 			if (bio->bio_offset < bq->bio_offset) {
975 				TAILQ_INSERT_BEFORE(bq, bio, bio_act);
976 				return;
977 			}
978 		}
979 	}
980 
981 	/*
982 	 * Request is at/after our current position in the list.
983 	 * Optimize for sequential I/O by seeing if we go at the tail.
984 	 */
985 	if (bio->bio_offset > be->bio_offset) {
986 		TAILQ_INSERT_AFTER(&bioq->queue, be, bio, bio_act);
987 		return;
988 	}
989 
990 	/* Otherwise, insertion sort */
991 	while ((bn = TAILQ_NEXT(bq, bio_act)) != NULL) {
992 		/*
993 		 * We want to go after the current request if it is the end
994 		 * of the first request list, or if the next request is a
995 		 * larger cylinder than our request.
996 		 */
997 		if (bn == bioq->switch_point ||
998 		    bio->bio_offset < bn->bio_offset) {
999 			break;
1000 		}
1001 		bq = bn;
1002 	}
1003 	TAILQ_INSERT_AFTER(&bioq->queue, bq, bio, bio_act);
1004 }
1005 
1006 /*
1007  * Disk error is the preface to plaintive error messages
1008  * about failing disk transfers.  It prints messages of the form
1009 
1010 hp0g: hard error reading fsbn 12345 of 12344-12347 (hp0 bn %d cn %d tn %d sn %d)
1011 
1012  * if the offset of the error in the transfer and a disk label
1013  * are both available.  blkdone should be -1 if the position of the error
1014  * is unknown; the disklabel pointer may be null from drivers that have not
1015  * been converted to use them.  The message is printed with kprintf
1016  * if pri is LOG_PRINTF, otherwise it uses log at the specified priority.
1017  * The message should be completed (with at least a newline) with kprintf
1018  * or log(-1, ...), respectively.  There is no trailing space.
1019  */
1020 void
1021 diskerr(struct bio *bio, cdev_t dev, const char *what, int pri, int donecnt)
1022 {
1023 	struct buf *bp = bio->bio_buf;
1024 	const char *term;
1025 
1026 	switch(bp->b_cmd) {
1027 	case BUF_CMD_READ:
1028 		term = "read";
1029 		break;
1030 	case BUF_CMD_WRITE:
1031 		term = "write";
1032 		break;
1033 	default:
1034 		term = "access";
1035 		break;
1036 	}
1037 	kprintf("%s: %s %sing ", dev->si_name, what, term);
1038 	kprintf("offset %012llx for %d",
1039 		(long long)bio->bio_offset,
1040 		bp->b_bcount);
1041 
1042 	if (donecnt)
1043 		kprintf(" (%d bytes completed)", donecnt);
1044 }
1045 
1046 /*
1047  * Locate a disk device
1048  */
1049 cdev_t
1050 disk_locate(const char *devname)
1051 {
1052 	return devfs_find_device_by_name(devname);
1053 }
1054 
1055 void
1056 disk_config(void *arg)
1057 {
1058 	disk_msg_send_sync(DISK_SYNC, NULL, NULL);
1059 }
1060 
1061 static void
1062 disk_init(void)
1063 {
1064 	struct thread* td_core;
1065 
1066 	disk_msg_cache = objcache_create("disk-msg-cache", 0, 0,
1067 					 NULL, NULL, NULL,
1068 					 objcache_malloc_alloc,
1069 					 objcache_malloc_free,
1070 					 &disk_msg_malloc_args);
1071 
1072 	lwkt_token_init(&disklist_token);
1073 
1074 	/*
1075 	 * Initialize the reply-only port which acts as a message drain
1076 	 */
1077 	lwkt_initport_replyonly(&disk_dispose_port, disk_msg_autofree_reply);
1078 
1079 	lwkt_create(disk_msg_core, /*args*/NULL, &td_core, NULL,
1080 		    0, 0, "disk_msg_core");
1081 
1082 	tsleep(td_core, 0, "diskcore", 0);
1083 }
1084 
1085 static void
1086 disk_uninit(void)
1087 {
1088 	objcache_destroy(disk_msg_cache);
1089 }
1090 
1091 SYSINIT(disk_register, SI_SUB_PRE_DRIVERS, SI_ORDER_FIRST, disk_init, NULL);
1092 SYSUNINIT(disk_register, SI_SUB_PRE_DRIVERS, SI_ORDER_ANY, disk_uninit, NULL);
1093