xref: /dragonfly/sys/kern/subr_disk.c (revision 110def69)
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. Neither the name of the University nor the names of its contributors
59  *    may be used to endorse or promote products derived from this software
60  *    without specific prior written permission.
61  *
62  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
63  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
64  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
65  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
66  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
67  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
68  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
69  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
70  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
71  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
72  * SUCH DAMAGE.
73  *
74  *	@(#)ufs_disksubr.c	8.5 (Berkeley) 1/21/94
75  * $FreeBSD: src/sys/kern/subr_disk.c,v 1.20.2.6 2001/10/05 07:14:57 peter Exp $
76  * $FreeBSD: src/sys/ufs/ufs/ufs_disksubr.c,v 1.44.2.3 2001/03/05 05:42:19 obrien Exp $
77  */
78 
79 #include <sys/param.h>
80 #include <sys/systm.h>
81 #include <sys/kernel.h>
82 #include <sys/proc.h>
83 #include <sys/sysctl.h>
84 #include <sys/buf.h>
85 #include <sys/conf.h>
86 #include <sys/disklabel.h>
87 #include <sys/disklabel32.h>
88 #include <sys/disklabel64.h>
89 #include <sys/diskslice.h>
90 #include <sys/diskmbr.h>
91 #include <sys/disk.h>
92 #include <sys/kerneldump.h>
93 #include <sys/malloc.h>
94 #include <machine/md_var.h>
95 #include <sys/ctype.h>
96 #include <sys/syslog.h>
97 #include <sys/device.h>
98 #include <sys/msgport.h>
99 #include <sys/devfs.h>
100 #include <sys/thread.h>
101 #include <sys/dsched.h>
102 #include <sys/queue.h>
103 #include <sys/lock.h>
104 #include <sys/udev.h>
105 #include <sys/uuid.h>
106 
107 #include <sys/buf2.h>
108 #include <sys/msgport2.h>
109 #include <sys/thread2.h>
110 
111 static MALLOC_DEFINE(M_DISK, "disk", "disk data");
112 static int disk_debug_enable = 0;
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 static void bioqwritereorder(struct bio_queue_head *bioq);
120 static void disk_cleanserial(char *serno);
121 static int disk_debug(int, char *, ...) __printflike(2, 3);
122 static cdev_t _disk_create_named(const char *name, int unit, struct disk *dp,
123     struct dev_ops *raw_ops, int clone);
124 
125 static d_open_t diskopen;
126 static d_close_t diskclose;
127 static d_ioctl_t diskioctl;
128 static d_strategy_t diskstrategy;
129 static d_psize_t diskpsize;
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 static struct lwkt_token ds_token;
135 
136 static struct dev_ops disk1_ops = {
137 	{ "disk", 0, D_DISK | D_MPSAFE | D_TRACKCLOSE | D_KVABIO },
138 	.d_open = diskopen,
139 	.d_close = diskclose,
140 	.d_read = physread,
141 	.d_write = physwrite,
142 	.d_ioctl = diskioctl,
143 	.d_strategy = diskstrategy,
144 	.d_dump = diskdump,
145 	.d_psize = diskpsize,
146 };
147 
148 static struct dev_ops disk2_ops = {
149 	{ "disk", 0, D_DISK | D_MPSAFE | D_TRACKCLOSE | D_KVABIO |
150 		     D_NOEMERGPGR },
151 	.d_open = diskopen,
152 	.d_close = diskclose,
153 	.d_read = physread,
154 	.d_write = physwrite,
155 	.d_ioctl = diskioctl,
156 	.d_strategy = diskstrategy,
157 	.d_dump = diskdump,
158 	.d_psize = diskpsize,
159 };
160 
161 static struct objcache 	*disk_msg_cache;
162 
163 struct objcache_malloc_args disk_msg_malloc_args = {
164 	sizeof(struct disk_msg), M_DISK };
165 
166 static struct lwkt_port disk_dispose_port;
167 static struct lwkt_port disk_msg_port;
168 
169 static int
170 disk_debug(int level, char *fmt, ...)
171 {
172 	__va_list ap;
173 
174 	__va_start(ap, fmt);
175 	if (level <= disk_debug_enable)
176 		kvprintf(fmt, ap);
177 	__va_end(ap);
178 
179 	return 0;
180 }
181 
182 static int
183 disk_probe_slice(struct disk *dp, cdev_t dev, int slice, int reprobe)
184 {
185 	struct disk_info *info = &dp->d_info;
186 	struct diskslice *sp = &dp->d_slice->dss_slices[slice];
187 	disklabel_ops_t ops;
188 	struct dev_ops *dops;
189 	struct partinfo part;
190 	const char *msg;
191 	char uuid_buf[128];
192 	cdev_t ndev;
193 	int sno;
194 	u_int i;
195 
196 	disk_debug(2, "disk_probe_slice (begin): %s (%s)\n",
197 		   dev->si_name, dp->d_cdev->si_name);
198 
199 	sno = slice ? slice - 1 : 0;
200 	dops = (dp->d_rawdev->si_ops->head.flags & D_NOEMERGPGR) ?
201 		&disk2_ops : &disk1_ops;
202 
203 	ops = &disklabel32_ops;
204 	msg = ops->op_readdisklabel(dev, sp, &sp->ds_label, info);
205 	if (msg && !strcmp(msg, "no disk label")) {
206 		ops = &disklabel64_ops;
207 		msg = ops->op_readdisklabel(dev, sp, &sp->ds_label, info);
208 	}
209 
210 	if (msg == NULL) {
211 		if (slice != WHOLE_DISK_SLICE)
212 			ops->op_adjust_label_reserved(dp->d_slice, slice, sp);
213 		else
214 			sp->ds_reserved = 0;
215 
216 		sp->ds_ops = ops;
217 		for (i = 0; i < ops->op_getnumparts(sp->ds_label); i++) {
218 			ops->op_loadpartinfo(sp->ds_label, i, &part);
219 			if (part.fstype) {
220 				if (reprobe &&
221 				    (ndev = devfs_find_device_by_name("%s%c",
222 						dev->si_name, 'a' + i))
223 				) {
224 					/*
225 					 * Device already exists and
226 					 * is still valid.
227 					 */
228 					ndev->si_flags |= SI_REPROBE_TEST;
229 
230 					/*
231 					 * Destroy old UUID alias
232 					 */
233 					destroy_dev_alias(ndev, "part-by-uuid/*");
234 
235 					/* Create UUID alias */
236 					if (!kuuid_is_nil(&part.storage_uuid)) {
237 						snprintf_uuid(uuid_buf,
238 						    sizeof(uuid_buf),
239 						    &part.storage_uuid);
240 						make_dev_alias(ndev,
241 						    "part-by-uuid/%s",
242 						    uuid_buf);
243 						udev_dict_set_cstr(ndev, "uuid", uuid_buf);
244 					}
245 				} else {
246 					ndev = make_dev_covering(dops,
247 						dp->d_rawdev->si_ops,
248 						dkmakeminor(dkunit(dp->d_cdev),
249 							    slice, i),
250 						UID_ROOT, GID_OPERATOR, 0640,
251 						"%s%c", dev->si_name, 'a'+ i);
252 					ndev->si_parent = dev;
253 					ndev->si_iosize_max = dev->si_iosize_max;
254 					ndev->si_disk = dp;
255 					udev_dict_set_cstr(ndev, "subsystem", "disk");
256 					/* Inherit parent's disk type */
257 					if (dp->d_disktype) {
258 						udev_dict_set_cstr(ndev, "disk-type",
259 						    __DECONST(char *, dp->d_disktype));
260 					}
261 
262 					/* Create serno alias */
263 					if (dp->d_info.d_serialno) {
264 						make_dev_alias(ndev,
265 						    "serno/%s.s%d%c",
266 						    dp->d_info.d_serialno,
267 						    sno, 'a' + i);
268 					}
269 
270 					/* Create UUID alias */
271 					if (!kuuid_is_nil(&part.storage_uuid)) {
272 						snprintf_uuid(uuid_buf,
273 						    sizeof(uuid_buf),
274 						    &part.storage_uuid);
275 						make_dev_alias(ndev,
276 						    "part-by-uuid/%s",
277 						    uuid_buf);
278 						udev_dict_set_cstr(ndev, "uuid", uuid_buf);
279 					}
280 					ndev->si_flags |= SI_REPROBE_TEST;
281 				}
282 			}
283 		}
284 	} else if (info->d_dsflags & DSO_COMPATLABEL) {
285 		msg = NULL;
286 		if (sp->ds_size >= 0x100000000ULL)
287 			ops = &disklabel64_ops;
288 		else
289 			ops = &disklabel32_ops;
290 		sp->ds_label = ops->op_clone_label(info, sp);
291 	} else {
292 		if (sp->ds_type == DOSPTYP_386BSD || /* XXX */
293 		    sp->ds_type == DOSPTYP_NETBSD ||
294 		    sp->ds_type == DOSPTYP_OPENBSD ||
295 		    sp->ds_type == DOSPTYP_DFLYBSD) {
296 			log(LOG_WARNING, "%s: cannot find label (%s)\n",
297 			    dev->si_name, msg);
298 		}
299 
300 		if (sp->ds_label.opaque != NULL && sp->ds_ops != NULL) {
301 			/* Clear out old label - it's not around anymore */
302 			disk_debug(2,
303 			    "disk_probe_slice: clear out old diskabel on %s\n",
304 			    dev->si_name);
305 
306 			sp->ds_ops->op_freedisklabel(&sp->ds_label);
307 			sp->ds_ops = NULL;
308 		}
309 	}
310 
311 	if (msg == NULL) {
312 		sp->ds_wlabel = FALSE;
313 	}
314 
315 	return (msg ? EINVAL : 0);
316 }
317 
318 /*
319  * This routine is only called for newly minted drives or to reprobe
320  * a drive with no open slices.  disk_probe_slice() is called directly
321  * when reprobing partition changes within slices.
322  */
323 static void
324 disk_probe(struct disk *dp, int reprobe)
325 {
326 	struct disk_info *info = &dp->d_info;
327 	cdev_t dev = dp->d_cdev;
328 	cdev_t ndev;
329 	int error, i, sno;
330 	struct diskslices *osp;
331 	struct diskslice *sp;
332 	struct dev_ops *dops;
333 	char uuid_buf[128];
334 
335 	/*
336 	 * d_media_blksize can be 0 for non-disk storage devices such
337 	 * as audio CDs.
338 	 */
339 	if (info->d_media_blksize == 0)
340 		return;
341 
342 	osp = dp->d_slice;
343 	dp->d_slice = dsmakeslicestruct(BASE_SLICE, info);
344 	disk_debug(1, "disk_probe (begin): %s\n", dp->d_cdev->si_name);
345 
346 	error = mbrinit(dev, info, &(dp->d_slice));
347 	if (error) {
348 		dsgone(&osp);
349 		return;
350 	}
351 
352 	dops = (dp->d_rawdev->si_ops->head.flags & D_NOEMERGPGR) ?
353 		&disk2_ops : &disk1_ops;
354 
355 	for (i = 0; i < dp->d_slice->dss_nslices; i++) {
356 		/*
357 		 * Ignore the whole-disk slice, it has already been created.
358 		 */
359 		if (i == WHOLE_DISK_SLICE)
360 			continue;
361 
362 #if 1
363 		/*
364 		 * Ignore the compatibility slice s0 if it's a device mapper
365 		 * volume.
366 		 */
367 		if ((i == COMPATIBILITY_SLICE) &&
368 		    (info->d_dsflags & DSO_DEVICEMAPPER))
369 			continue;
370 #endif
371 
372 		sp = &dp->d_slice->dss_slices[i];
373 
374 		/*
375 		 * Handle s0.  s0 is a compatibility slice if there are no
376 		 * other slices and it has not otherwise been set up, else
377 		 * we ignore it.
378 		 */
379 		if (i == COMPATIBILITY_SLICE) {
380 			sno = 0;
381 			if (sp->ds_type == 0 &&
382 			    dp->d_slice->dss_nslices == BASE_SLICE) {
383 				sp->ds_size = info->d_media_blocks;
384 				sp->ds_reserved = 0;
385 			}
386 		} else {
387 			sno = i - 1;
388 			sp->ds_reserved = 0;
389 		}
390 
391 		/*
392 		 * Ignore 0-length slices
393 		 */
394 		if (sp->ds_size == 0)
395 			continue;
396 
397 		if (reprobe &&
398 		    (ndev = devfs_find_device_by_name("%ss%d",
399 						      dev->si_name, sno))) {
400 			/*
401 			 * Device already exists and is still valid
402 			 */
403 			ndev->si_flags |= SI_REPROBE_TEST;
404 
405 			/*
406 			 * Destroy old UUID alias
407 			 */
408 			destroy_dev_alias(ndev, "slice-by-uuid/*");
409 
410 			/* Create UUID alias */
411 			if (!kuuid_is_nil(&sp->ds_stor_uuid)) {
412 				snprintf_uuid(uuid_buf, sizeof(uuid_buf),
413 				    &sp->ds_stor_uuid);
414 				make_dev_alias(ndev, "slice-by-uuid/%s",
415 				    uuid_buf);
416 			}
417 		} else {
418 			/*
419 			 * Else create new device
420 			 */
421 			ndev = make_dev_covering(dops, dp->d_rawdev->si_ops,
422 					dkmakewholeslice(dkunit(dev), i),
423 					UID_ROOT, GID_OPERATOR, 0640,
424 					(info->d_dsflags & DSO_DEVICEMAPPER)?
425 					"%s.s%d" : "%ss%d", dev->si_name, sno);
426 			ndev->si_parent = dev;
427 			ndev->si_iosize_max = dev->si_iosize_max;
428 			udev_dict_set_cstr(ndev, "subsystem", "disk");
429 			/* Inherit parent's disk type */
430 			if (dp->d_disktype) {
431 				udev_dict_set_cstr(ndev, "disk-type",
432 				    __DECONST(char *, dp->d_disktype));
433 			}
434 
435 			/* Create serno alias */
436 			if (dp->d_info.d_serialno) {
437 				make_dev_alias(ndev, "serno/%s.s%d",
438 					       dp->d_info.d_serialno, sno);
439 			}
440 
441 			/* Create UUID alias */
442 			if (!kuuid_is_nil(&sp->ds_stor_uuid)) {
443 				snprintf_uuid(uuid_buf, sizeof(uuid_buf),
444 				    &sp->ds_stor_uuid);
445 				make_dev_alias(ndev, "slice-by-uuid/%s",
446 				    uuid_buf);
447 			}
448 
449 			ndev->si_disk = dp;
450 			ndev->si_flags |= SI_REPROBE_TEST;
451 		}
452 		sp->ds_dev = ndev;
453 
454 		/*
455 		 * Probe appropriate slices for a disklabel
456 		 *
457 		 * XXX slice type 1 used by our gpt probe code.
458 		 * XXX slice type 0 used by mbr compat slice.
459 		 */
460 		if (sp->ds_type == DOSPTYP_386BSD ||
461 		    sp->ds_type == DOSPTYP_NETBSD ||
462 		    sp->ds_type == DOSPTYP_OPENBSD ||
463 		    sp->ds_type == DOSPTYP_DFLYBSD ||
464 		    sp->ds_type == 0 ||
465 		    sp->ds_type == 1) {
466 			if (dp->d_slice->dss_first_bsd_slice == 0)
467 				dp->d_slice->dss_first_bsd_slice = i;
468 			disk_probe_slice(dp, ndev, i, reprobe);
469 		}
470 	}
471 	dsgone(&osp);
472 	disk_debug(1, "disk_probe (end): %s\n", dp->d_cdev->si_name);
473 }
474 
475 
476 static void
477 disk_msg_core(void *arg)
478 {
479 	struct disk	*dp;
480 	struct diskslice *sp;
481 	disk_msg_t msg;
482 	int run;
483 
484 	lwkt_gettoken(&disklist_token);
485 	lwkt_initport_thread(&disk_msg_port, curthread);
486 	wakeup(curthread);	/* synchronous startup */
487 	lwkt_reltoken(&disklist_token);
488 
489 	lwkt_gettoken(&ds_token);
490 	run = 1;
491 
492 	while (run) {
493 		msg = (disk_msg_t)lwkt_waitport(&disk_msg_port, 0);
494 
495 		switch (msg->hdr.u.ms_result) {
496 		case DISK_DISK_PROBE:
497 			dp = (struct disk *)msg->load;
498 			disk_debug(1,
499 				    "DISK_DISK_PROBE: %s\n",
500 					dp->d_cdev->si_name);
501 			disk_iocom_update(dp);
502 			disk_probe(dp, 0);
503 			break;
504 		case DISK_DISK_DESTROY:
505 			dp = (struct disk *)msg->load;
506 			disk_debug(1,
507 				    "DISK_DISK_DESTROY: %s\n",
508 					dp->d_cdev->si_name);
509 			disk_iocom_uninit(dp);
510 
511 			/*
512 			 * Interlock against struct disk enumerations.
513 			 * Wait for enumerations to complete then remove
514 			 * the dp from the list before tearing it down.
515 			 * This avoids numerous races.
516 			 */
517 			lwkt_gettoken(&disklist_token);
518 			while (dp->d_refs)
519 				tsleep(&dp->d_refs, 0, "diskdel", hz / 10);
520 			LIST_REMOVE(dp, d_list);
521 
522 			dsched_disk_destroy(dp);
523 			devfs_destroy_related(dp->d_cdev);
524 			destroy_dev(dp->d_cdev);
525 			destroy_only_dev(dp->d_rawdev);
526 
527 			lwkt_reltoken(&disklist_token);
528 
529 			if (dp->d_info.d_serialno) {
530 				kfree(dp->d_info.d_serialno, M_TEMP);
531 				dp->d_info.d_serialno = NULL;
532 			}
533 			break;
534 		case DISK_UNPROBE:
535 			dp = (struct disk *)msg->load;
536 			disk_debug(1,
537 				    "DISK_DISK_UNPROBE: %s\n",
538 					dp->d_cdev->si_name);
539 			devfs_destroy_related(dp->d_cdev);
540 			break;
541 		case DISK_SLICE_REPROBE:
542 			dp = (struct disk *)msg->load;
543 			sp = (struct diskslice *)msg->load2;
544 			devfs_clr_related_flag(sp->ds_dev,
545 						SI_REPROBE_TEST);
546 			disk_debug(1,
547 				    "DISK_SLICE_REPROBE: %s\n",
548 				    sp->ds_dev->si_name);
549 			disk_probe_slice(dp, sp->ds_dev,
550 					 dkslice(sp->ds_dev), 1);
551 			devfs_destroy_related_without_flag(
552 					sp->ds_dev, SI_REPROBE_TEST);
553 			break;
554 		case DISK_DISK_REPROBE:
555 			dp = (struct disk *)msg->load;
556 			devfs_clr_related_flag(dp->d_cdev, SI_REPROBE_TEST);
557 			disk_debug(1,
558 				    "DISK_DISK_REPROBE: %s\n",
559 				    dp->d_cdev->si_name);
560 			disk_probe(dp, 1);
561 			devfs_destroy_related_without_flag(
562 					dp->d_cdev, SI_REPROBE_TEST);
563 			break;
564 		case DISK_SYNC:
565 			disk_debug(1, "DISK_SYNC\n");
566 			break;
567 		default:
568 			devfs_debug(DEVFS_DEBUG_WARNING,
569 				    "disk_msg_core: unknown message "
570 				    "received at core\n");
571 			break;
572 		}
573 		lwkt_replymsg(&msg->hdr, 0);
574 	}
575 	lwkt_reltoken(&ds_token);
576 	lwkt_exit();
577 }
578 
579 
580 /*
581  * Acts as a message drain. Any message that is replied to here gets
582  * destroyed and the memory freed.
583  */
584 static void
585 disk_msg_autofree_reply(lwkt_port_t port, lwkt_msg_t msg)
586 {
587 	objcache_put(disk_msg_cache, msg);
588 }
589 
590 
591 void
592 disk_msg_send(uint32_t cmd, void *load, void *load2)
593 {
594 	disk_msg_t disk_msg;
595 	lwkt_port_t port = &disk_msg_port;
596 
597 	disk_msg = objcache_get(disk_msg_cache, M_WAITOK);
598 
599 	lwkt_initmsg(&disk_msg->hdr, &disk_dispose_port, 0);
600 
601 	disk_msg->hdr.u.ms_result = cmd;
602 	disk_msg->load = load;
603 	disk_msg->load2 = load2;
604 	KKASSERT(port);
605 	lwkt_sendmsg(port, &disk_msg->hdr);
606 }
607 
608 void
609 disk_msg_send_sync(uint32_t cmd, void *load, void *load2)
610 {
611 	struct lwkt_port rep_port;
612 	disk_msg_t disk_msg;
613 	lwkt_port_t port;
614 
615 	disk_msg = objcache_get(disk_msg_cache, M_WAITOK);
616 	port = &disk_msg_port;
617 
618 	/* XXX could probably use curthread's built-in msgport */
619 	lwkt_initport_thread(&rep_port, curthread);
620 	lwkt_initmsg(&disk_msg->hdr, &rep_port, 0);
621 
622 	disk_msg->hdr.u.ms_result = cmd;
623 	disk_msg->load = load;
624 	disk_msg->load2 = load2;
625 
626 	lwkt_domsg(port, &disk_msg->hdr, 0);
627 	objcache_put(disk_msg_cache, disk_msg);
628 }
629 
630 /*
631  * Create a raw device for the dev_ops template (which is returned).  Also
632  * create a slice and unit managed disk and overload the user visible
633  * device space with it.
634  *
635  * NOTE: The returned raw device is NOT a slice and unit managed device.
636  * It is an actual raw device representing the raw disk as specified by
637  * the passed dev_ops.  The disk layer not only returns such a raw device,
638  * it also uses it internally when passing (modified) commands through.
639  */
640 cdev_t
641 disk_create(int unit, struct disk *dp, struct dev_ops *raw_ops)
642 {
643 	return _disk_create_named(NULL, unit, dp, raw_ops, 0);
644 }
645 
646 cdev_t
647 disk_create_clone(int unit, struct disk *dp,
648 		  struct dev_ops *raw_ops)
649 {
650 	return _disk_create_named(NULL, unit, dp, raw_ops, 1);
651 }
652 
653 cdev_t
654 disk_create_named(const char *name, int unit, struct disk *dp,
655 		  struct dev_ops *raw_ops)
656 {
657 	return _disk_create_named(name, unit, dp, raw_ops, 0);
658 }
659 
660 cdev_t
661 disk_create_named_clone(const char *name, int unit, struct disk *dp,
662 			struct dev_ops *raw_ops)
663 {
664 	return _disk_create_named(name, unit, dp, raw_ops, 1);
665 }
666 
667 static cdev_t
668 _disk_create_named(const char *name, int unit, struct disk *dp,
669 		   struct dev_ops *raw_ops, int clone)
670 {
671 	cdev_t rawdev;
672 	struct dev_ops *dops;
673 
674 	disk_debug(1, "disk_create (begin): %s%d\n", name, unit);
675 
676 	if (name) {
677 		rawdev = make_only_dev(raw_ops, dkmakewholedisk(unit),
678 		    UID_ROOT, GID_OPERATOR, 0640, "%s", name);
679 	} else {
680 		rawdev = make_only_dev(raw_ops, dkmakewholedisk(unit),
681 		    UID_ROOT, GID_OPERATOR, 0640,
682 		    "%s%d", raw_ops->head.name, unit);
683 	}
684 
685 	bzero(dp, sizeof(*dp));
686 
687 	dops = (raw_ops->head.flags & D_NOEMERGPGR) ? &disk2_ops : &disk1_ops;
688 
689 	dp->d_rawdev = rawdev;
690 	dp->d_raw_ops = raw_ops;
691 	dp->d_dev_ops = dops;
692 
693 	if (name) {
694 		if (clone) {
695 			dp->d_cdev = make_only_dev_covering(
696 					dops, dp->d_rawdev->si_ops,
697 					dkmakewholedisk(unit),
698 					UID_ROOT, GID_OPERATOR, 0640,
699 					"%s", name);
700 		} else {
701 			dp->d_cdev = make_dev_covering(
702 					dops, dp->d_rawdev->si_ops,
703 					dkmakewholedisk(unit),
704 					UID_ROOT, GID_OPERATOR, 0640,
705 					"%s", name);
706 		}
707 	} else {
708 		if (clone) {
709 			dp->d_cdev = make_only_dev_covering(
710 					dops, dp->d_rawdev->si_ops,
711 					dkmakewholedisk(unit),
712 					UID_ROOT, GID_OPERATOR, 0640,
713 					"%s%d", raw_ops->head.name, unit);
714 		} else {
715 			dp->d_cdev = make_dev_covering(
716 					dops, dp->d_rawdev->si_ops,
717 					dkmakewholedisk(unit),
718 					UID_ROOT, GID_OPERATOR, 0640,
719 					"%s%d", raw_ops->head.name, unit);
720 		}
721 	}
722 
723 	udev_dict_set_cstr(dp->d_cdev, "subsystem", "disk");
724 	dp->d_cdev->si_disk = dp;
725 
726 	if (name)
727 		dsched_disk_create(dp, name, unit);
728 	else
729 		dsched_disk_create(dp, raw_ops->head.name, unit);
730 
731 	lwkt_gettoken(&disklist_token);
732 	LIST_INSERT_HEAD(&disklist, dp, d_list);
733 	lwkt_reltoken(&disklist_token);
734 
735 	disk_iocom_init(dp);
736 
737 	disk_debug(1, "disk_create (end): %s%d\n",
738 		   (name != NULL)?(name):(raw_ops->head.name), unit);
739 
740 	return (dp->d_rawdev);
741 }
742 
743 int
744 disk_setdisktype(struct disk *disk, const char *type)
745 {
746 	int error;
747 
748 	KKASSERT(disk != NULL);
749 
750 	disk->d_disktype = type;
751 	error = udev_dict_set_cstr(disk->d_cdev, "disk-type",
752 				   __DECONST(char *, type));
753 	return error;
754 }
755 
756 int
757 disk_getopencount(struct disk *disk)
758 {
759 	return disk->d_opencount;
760 }
761 
762 static void
763 _setdiskinfo(struct disk *disk, struct disk_info *info)
764 {
765 	char *oldserialno;
766 
767 	oldserialno = disk->d_info.d_serialno;
768 	bcopy(info, &disk->d_info, sizeof(disk->d_info));
769 	info = &disk->d_info;
770 
771 	disk_debug(1, "_setdiskinfo: %s\n", disk->d_cdev->si_name);
772 
773 	/*
774 	 * The serial number is duplicated so the caller can throw
775 	 * their copy away.
776 	 */
777 	if (info->d_serialno && info->d_serialno[0] &&
778 	    (info->d_serialno[0] != ' ' || strlen(info->d_serialno) > 1)) {
779 		info->d_serialno = kstrdup(info->d_serialno, M_TEMP);
780 		disk_cleanserial(info->d_serialno);
781 		if (disk->d_cdev) {
782 			make_dev_alias(disk->d_cdev, "serno/%s",
783 				       info->d_serialno);
784 		}
785 	} else {
786 		info->d_serialno = NULL;
787 	}
788 	if (oldserialno)
789 		kfree(oldserialno, M_TEMP);
790 
791 	dsched_disk_update(disk, info);
792 
793 	/*
794 	 * The caller may set d_media_size or d_media_blocks and we
795 	 * calculate the other.
796 	 */
797 	KKASSERT(info->d_media_size == 0 || info->d_media_blocks == 0);
798 	if (info->d_media_size == 0 && info->d_media_blocks) {
799 		info->d_media_size = (u_int64_t)info->d_media_blocks *
800 				     info->d_media_blksize;
801 	} else if (info->d_media_size && info->d_media_blocks == 0 &&
802 		   info->d_media_blksize) {
803 		info->d_media_blocks = info->d_media_size /
804 				       info->d_media_blksize;
805 	}
806 
807 	/*
808 	 * The si_* fields for rawdev are not set until after the
809 	 * disk_create() call, so someone using the cooked version
810 	 * of the raw device (i.e. da0s0) will not get the right
811 	 * si_iosize_max unless we fix it up here.
812 	 */
813 	if (disk->d_cdev && disk->d_rawdev &&
814 	    disk->d_cdev->si_iosize_max == 0) {
815 		disk->d_cdev->si_iosize_max = disk->d_rawdev->si_iosize_max;
816 		disk->d_cdev->si_bsize_phys = disk->d_rawdev->si_bsize_phys;
817 		disk->d_cdev->si_bsize_best = disk->d_rawdev->si_bsize_best;
818 	}
819 
820 	/* Add the serial number to the udev_dictionary */
821 	if (info->d_serialno)
822 		udev_dict_set_cstr(disk->d_cdev, "serno", info->d_serialno);
823 }
824 
825 /*
826  * Disk drivers must call this routine when media parameters are available
827  * or have changed.
828  */
829 void
830 disk_setdiskinfo(struct disk *disk, struct disk_info *info)
831 {
832 	_setdiskinfo(disk, info);
833 	disk_msg_send(DISK_DISK_PROBE, disk, NULL);
834 	disk_debug(1, "disk_setdiskinfo: sent probe for %s\n",
835 		   disk->d_cdev->si_name);
836 }
837 
838 void
839 disk_setdiskinfo_sync(struct disk *disk, struct disk_info *info)
840 {
841 	_setdiskinfo(disk, info);
842 	disk_msg_send_sync(DISK_DISK_PROBE, disk, NULL);
843 	disk_debug(1, "disk_setdiskinfo_sync: sent probe for %s\n",
844 		   disk->d_cdev->si_name);
845 }
846 
847 /*
848  * This routine is called when an adapter detaches.  The higher level
849  * managed disk device is destroyed while the lower level raw device is
850  * released.
851  */
852 void
853 disk_destroy(struct disk *disk)
854 {
855 	disk_msg_send_sync(DISK_DISK_DESTROY, disk, NULL);
856 	return;
857 }
858 
859 int
860 disk_dumpcheck(cdev_t dev, u_int64_t *size,
861 	       u_int64_t *blkno, u_int32_t *secsize)
862 {
863 	struct partinfo pinfo;
864 	int error;
865 
866 	if (size)
867 		*size = 0;	/* avoid gcc warnings */
868 	if (secsize)
869 		*secsize = 512;	/* avoid gcc warnings */
870 	bzero(&pinfo, sizeof(pinfo));
871 
872 	error = dev_dioctl(dev, DIOCGPART, (void *)&pinfo, 0,
873 			   proc0.p_ucred, NULL, NULL);
874 	if (error)
875 		return (error);
876 
877 	if (pinfo.media_blksize == 0)
878 		return (ENXIO);
879 
880 	if (blkno) /* XXX: make sure this reserved stuff is right */
881 		*blkno = pinfo.reserved_blocks +
882 			pinfo.media_offset / pinfo.media_blksize;
883 	if (secsize)
884 		*secsize = pinfo.media_blksize;
885 	if (size)
886 		*size = (pinfo.media_blocks - pinfo.reserved_blocks);
887 
888 	return (0);
889 }
890 
891 int
892 disk_dumpconf(cdev_t dev, u_int onoff)
893 {
894 	struct dumperinfo di;
895 	u_int64_t	size, blkno;
896 	u_int32_t	secsize;
897 	int error;
898 
899 	if (!onoff)
900 		return set_dumper(NULL);
901 
902 	error = disk_dumpcheck(dev, &size, &blkno, &secsize);
903 
904 	if (error)
905 		return ENXIO;
906 
907 	bzero(&di, sizeof(struct dumperinfo));
908 	di.dumper = diskdump;
909 	di.priv = dev;
910 	di.blocksize = secsize;
911 	di.maxiosize = dev->si_iosize_max;
912 	di.mediaoffset = blkno * DEV_BSIZE;
913 	di.mediasize = size * DEV_BSIZE;
914 
915 	return set_dumper(&di);
916 }
917 
918 void
919 disk_unprobe(struct disk *disk)
920 {
921 	if (disk == NULL)
922 		return;
923 
924 	disk_msg_send_sync(DISK_UNPROBE, disk, NULL);
925 }
926 
927 void
928 disk_invalidate (struct disk *disk)
929 {
930 	dsgone(&disk->d_slice);
931 }
932 
933 /*
934  * Enumerate disks, pass a marker and an initial NULL dp to initialize,
935  * then loop with the previously returned dp.
936  *
937  * The returned dp will be referenced, preventing its destruction.  When
938  * you pass the returned dp back into the loop the ref is dropped.
939  *
940  * WARNING: If terminating your loop early you must call
941  *	    disk_enumerate_stop().
942  */
943 struct disk *
944 disk_enumerate(struct disk *marker, struct disk *dp)
945 {
946 	lwkt_gettoken(&disklist_token);
947 	if (dp) {
948 		--dp->d_refs;
949 		dp = LIST_NEXT(marker, d_list);
950 		LIST_REMOVE(marker, d_list);
951 	} else {
952 		bzero(marker, sizeof(*marker));
953 		marker->d_flags = DISKFLAG_MARKER;
954 		dp = LIST_FIRST(&disklist);
955 	}
956 	while (dp) {
957 		if ((dp->d_flags & DISKFLAG_MARKER) == 0)
958 			break;
959 		dp = LIST_NEXT(dp, d_list);
960 	}
961 	if (dp) {
962 		++dp->d_refs;
963 		LIST_INSERT_AFTER(dp, marker, d_list);
964 	}
965 	lwkt_reltoken(&disklist_token);
966 	return (dp);
967 }
968 
969 /*
970  * Terminate an enumeration early.  Do not call this function if the
971  * enumeration ended normally.  dp can be NULL, indicating that you
972  * wish to retain the ref count on dp.
973  *
974  * This function removes the marker.
975  */
976 void
977 disk_enumerate_stop(struct disk *marker, struct disk *dp)
978 {
979 	lwkt_gettoken(&disklist_token);
980 	LIST_REMOVE(marker, d_list);
981 	if (dp)
982 		--dp->d_refs;
983 	lwkt_reltoken(&disklist_token);
984 }
985 
986 static
987 int
988 sysctl_disks(SYSCTL_HANDLER_ARGS)
989 {
990 	struct disk marker;
991 	struct disk *dp;
992 	int error, first;
993 
994 	first = 1;
995 	error = 0;
996 	dp = NULL;
997 
998 	while ((dp = disk_enumerate(&marker, dp))) {
999 		if (!first) {
1000 			error = SYSCTL_OUT(req, " ", 1);
1001 			if (error) {
1002 				disk_enumerate_stop(&marker, dp);
1003 				break;
1004 			}
1005 		} else {
1006 			first = 0;
1007 		}
1008 		error = SYSCTL_OUT(req, dp->d_rawdev->si_name,
1009 				   strlen(dp->d_rawdev->si_name));
1010 		if (error) {
1011 			disk_enumerate_stop(&marker, dp);
1012 			break;
1013 		}
1014 	}
1015 	if (error == 0)
1016 		error = SYSCTL_OUT(req, "", 1);
1017 	return error;
1018 }
1019 
1020 SYSCTL_PROC(_kern, OID_AUTO, disks, CTLTYPE_STRING | CTLFLAG_RD, NULL, 0,
1021     sysctl_disks, "A", "names of available disks");
1022 
1023 /*
1024  * Open a disk device or partition.
1025  */
1026 static
1027 int
1028 diskopen(struct dev_open_args *ap)
1029 {
1030 	cdev_t dev = ap->a_head.a_dev;
1031 	struct disk *dp;
1032 	int error;
1033 
1034 	/*
1035 	 * dp can't be NULL here XXX.
1036 	 *
1037 	 * d_slice will be NULL if setdiskinfo() has not been called yet.
1038 	 * setdiskinfo() is typically called whether the disk is present
1039 	 * or not (e.g. CD), but the base disk device is created first
1040 	 * and there may be a race.
1041 	 */
1042 	dp = dev->si_disk;
1043 	if (dp == NULL || dp->d_slice == NULL)
1044 		return (ENXIO);
1045 	error = 0;
1046 
1047 	/*
1048 	 * Deal with open races
1049 	 */
1050 	lwkt_gettoken(&ds_token);
1051 	while (dp->d_flags & DISKFLAG_LOCK) {
1052 		dp->d_flags |= DISKFLAG_WANTED;
1053 		error = tsleep(dp, PCATCH, "diskopen", hz);
1054 		if (error) {
1055 			lwkt_reltoken(&ds_token);
1056 			return (error);
1057 		}
1058 	}
1059 	dp->d_flags |= DISKFLAG_LOCK;
1060 
1061 	/*
1062 	 * Open the underlying raw device.
1063 	 */
1064 	if (!dsisopen(dp->d_slice)) {
1065 #if 0
1066 		if (!pdev->si_iosize_max)
1067 			pdev->si_iosize_max = dev->si_iosize_max;
1068 #endif
1069 		error = dev_dopen(dp->d_rawdev, ap->a_oflags,
1070 				  ap->a_devtype, ap->a_cred, NULL);
1071 	}
1072 
1073 	if (error)
1074 		goto out;
1075 	error = dsopen(dev, ap->a_devtype, dp->d_info.d_dsflags,
1076 		       &dp->d_slice, &dp->d_info);
1077 	if (!dsisopen(dp->d_slice)) {
1078 		dev_dclose(dp->d_rawdev, ap->a_oflags, ap->a_devtype, NULL);
1079 	}
1080 out:
1081 	dp->d_flags &= ~DISKFLAG_LOCK;
1082 	if (dp->d_flags & DISKFLAG_WANTED) {
1083 		dp->d_flags &= ~DISKFLAG_WANTED;
1084 		wakeup(dp);
1085 	}
1086 	lwkt_reltoken(&ds_token);
1087 
1088 	KKASSERT(dp->d_opencount >= 0);
1089 	/* If the open was successful, bump open count */
1090 	if (error == 0)
1091 		atomic_add_int(&dp->d_opencount, 1);
1092 
1093 	return(error);
1094 }
1095 
1096 /*
1097  * Close a disk device or partition
1098  */
1099 static
1100 int
1101 diskclose(struct dev_close_args *ap)
1102 {
1103 	cdev_t dev = ap->a_head.a_dev;
1104 	struct disk *dp;
1105 	int error;
1106 	int lcount;
1107 
1108 	error = 0;
1109 	dp = dev->si_disk;
1110 
1111 	/*
1112 	 * The cdev_t represents the disk/slice/part.  The shared
1113 	 * dp structure governs all cdevs associated with the disk.
1114 	 *
1115 	 * As a safety only close the underlying raw device on the last
1116 	 * close the disk device if our tracking of the slices/partitions
1117 	 * also indicates nothing is open.
1118 	 */
1119 	KKASSERT(dp->d_opencount >= 1);
1120 	lcount = atomic_fetchadd_int(&dp->d_opencount, -1);
1121 
1122 	lwkt_gettoken(&ds_token);
1123 	dsclose(dev, ap->a_devtype, dp->d_slice);
1124 	if (lcount <= 1 && !dsisopen(dp->d_slice)) {
1125 		error = dev_dclose(dp->d_rawdev, ap->a_fflag, ap->a_devtype, NULL);
1126 	}
1127 	lwkt_reltoken(&ds_token);
1128 
1129 	return (error);
1130 }
1131 
1132 /*
1133  * First execute the ioctl on the disk device, and if it isn't supported
1134  * try running it on the backing device.
1135  */
1136 static
1137 int
1138 diskioctl(struct dev_ioctl_args *ap)
1139 {
1140 	cdev_t dev = ap->a_head.a_dev;
1141 	struct disk *dp;
1142 	int error;
1143 	u_int u;
1144 
1145 	dp = dev->si_disk;
1146 	if (dp == NULL)
1147 		return (ENXIO);
1148 
1149 	devfs_debug(DEVFS_DEBUG_DEBUG,
1150 		    "diskioctl: cmd is: %lx (name: %s)\n",
1151 		    ap->a_cmd, dev->si_name);
1152 	devfs_debug(DEVFS_DEBUG_DEBUG,
1153 		    "diskioctl: &dp->d_slice is: %p, %p\n",
1154 		    &dp->d_slice, dp->d_slice);
1155 
1156 	if (ap->a_cmd == DIOCGKERNELDUMP) {
1157 		u = *(u_int *)ap->a_data;
1158 		return disk_dumpconf(dev, u);
1159 	}
1160 
1161 	if (ap->a_cmd == DIOCRECLUSTER && dev == dp->d_cdev) {
1162 		error = disk_iocom_ioctl(dp, ap->a_cmd, ap->a_data);
1163 		return error;
1164 	}
1165 
1166 	if (&dp->d_slice == NULL || dp->d_slice == NULL ||
1167 	    ((dp->d_info.d_dsflags & DSO_DEVICEMAPPER) &&
1168 	     dkslice(dev) == WHOLE_DISK_SLICE)) {
1169 		error = ENOIOCTL;
1170 	} else {
1171 		lwkt_gettoken(&ds_token);
1172 		error = dsioctl(dev, ap->a_cmd, ap->a_data, ap->a_fflag,
1173 				&dp->d_slice, &dp->d_info);
1174 		lwkt_reltoken(&ds_token);
1175 	}
1176 
1177 	if (error == ENOIOCTL) {
1178 		error = dev_dioctl(dp->d_rawdev, ap->a_cmd, ap->a_data,
1179 				   ap->a_fflag, ap->a_cred, NULL, NULL);
1180 	}
1181 	return (error);
1182 }
1183 
1184 /*
1185  * Execute strategy routine
1186  *
1187  * WARNING! We are using the KVABIO API and must not access memory
1188  *         through bp->b_data without first calling bkvasync(bp).
1189  */
1190 static
1191 int
1192 diskstrategy(struct dev_strategy_args *ap)
1193 {
1194 	cdev_t dev = ap->a_head.a_dev;
1195 	struct bio *bio = ap->a_bio;
1196 	struct bio *nbio;
1197 	struct disk *dp;
1198 
1199 	dp = dev->si_disk;
1200 
1201 	if (dp == NULL) {
1202 		bio->bio_buf->b_error = ENXIO;
1203 		bio->bio_buf->b_flags |= B_ERROR;
1204 		biodone(bio);
1205 		return(0);
1206 	}
1207 	KKASSERT(dev->si_disk == dp);
1208 
1209 	/*
1210 	 * The dscheck() function will also transform the slice relative
1211 	 * block number i.e. bio->bio_offset into a block number that can be
1212 	 * passed directly to the underlying raw device.  If dscheck()
1213 	 * returns NULL it will have handled the bio for us (e.g. EOF
1214 	 * or error due to being beyond the device size).
1215 	 */
1216 	if ((nbio = dscheck(dev, bio, dp->d_slice)) != NULL) {
1217 		dev_dstrategy(dp->d_rawdev, nbio);
1218 	} else {
1219 		biodone(bio);
1220 	}
1221 	return(0);
1222 }
1223 
1224 /*
1225  * Return the partition size in ?blocks?
1226  */
1227 static
1228 int
1229 diskpsize(struct dev_psize_args *ap)
1230 {
1231 	cdev_t dev = ap->a_head.a_dev;
1232 	struct disk *dp;
1233 
1234 	dp = dev->si_disk;
1235 	if (dp == NULL)
1236 		return(ENODEV);
1237 
1238 	ap->a_result = dssize(dev, &dp->d_slice);
1239 
1240 	if ((ap->a_result == -1) &&
1241 	   (dp->d_info.d_dsflags & DSO_RAWPSIZE)) {
1242 		ap->a_head.a_dev = dp->d_rawdev;
1243 		return dev_doperate(&ap->a_head);
1244 	}
1245 	return(0);
1246 }
1247 
1248 static int
1249 diskdump(struct dev_dump_args *ap)
1250 {
1251 	cdev_t dev = ap->a_head.a_dev;
1252 	struct disk *dp = dev->si_disk;
1253 	u_int64_t size, offset;
1254 	int error;
1255 
1256 	error = disk_dumpcheck(dev, &size, &ap->a_blkno, &ap->a_secsize);
1257 	/* XXX: this should probably go in disk_dumpcheck somehow */
1258 	if (ap->a_length != 0) {
1259 		size *= DEV_BSIZE;
1260 		offset = ap->a_blkno * DEV_BSIZE;
1261 		if ((ap->a_offset < offset) ||
1262 		    (ap->a_offset + ap->a_length - offset > size)) {
1263 			kprintf("Attempt to write outside dump "
1264 				"device boundaries.\n");
1265 			error = ENOSPC;
1266 		}
1267 	}
1268 
1269 	if (error == 0) {
1270 		ap->a_head.a_dev = dp->d_rawdev;
1271 		error = dev_doperate(&ap->a_head);
1272 	}
1273 
1274 	return(error);
1275 }
1276 
1277 
1278 SYSCTL_INT(_debug_sizeof, OID_AUTO, diskslices, CTLFLAG_RD,
1279 	   0, sizeof(struct diskslices), "sizeof(struct diskslices)");
1280 
1281 SYSCTL_INT(_debug_sizeof, OID_AUTO, disk, CTLFLAG_RD,
1282 	   0, sizeof(struct disk), "sizeof(struct disk)");
1283 
1284 /*
1285  * Reorder interval for burst write allowance and minor write
1286  * allowance.
1287  *
1288  * We always want to trickle some writes in to make use of the
1289  * disk's zone cache.  Bursting occurs on a longer interval and only
1290  * runningbufspace is well over the hirunningspace limit.
1291  */
1292 int bioq_reorder_burst_interval = 60;	/* should be multiple of minor */
1293 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_burst_interval,
1294 	   CTLFLAG_RW, &bioq_reorder_burst_interval, 0, "");
1295 int bioq_reorder_minor_interval = 5;
1296 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_minor_interval,
1297 	   CTLFLAG_RW, &bioq_reorder_minor_interval, 0, "");
1298 
1299 int bioq_reorder_burst_bytes = 3000000;
1300 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_burst_bytes,
1301 	   CTLFLAG_RW, &bioq_reorder_burst_bytes, 0, "");
1302 int bioq_reorder_minor_bytes = 262144;
1303 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_minor_bytes,
1304 	   CTLFLAG_RW, &bioq_reorder_minor_bytes, 0, "");
1305 
1306 
1307 /*
1308  * Order I/Os.  Generally speaking this code is designed to make better
1309  * use of drive zone caches.  A drive zone cache can typically track linear
1310  * reads or writes for around 16 zones simultaniously.
1311  *
1312  * Read prioritization issues:  It is possible for hundreds of megabytes worth
1313  * of writes to be queued asynchronously.  This creates a huge bottleneck
1314  * for reads which reduce read bandwidth to a trickle.
1315  *
1316  * To solve this problem we generally reorder reads before writes.
1317  *
1318  * However, a large number of random reads can also starve writes and
1319  * make poor use of the drive zone cache so we allow writes to trickle
1320  * in every N reads.
1321  */
1322 void
1323 bioqdisksort(struct bio_queue_head *bioq, struct bio *bio)
1324 {
1325 #if 0
1326 	/*
1327 	 * The BIO wants to be ordered.  Adding to the tail also
1328 	 * causes transition to be set to NULL, forcing the ordering
1329 	 * of all prior I/O's.
1330 	 */
1331 	if (bio->bio_buf->b_flags & B_ORDERED) {
1332 		bioq_insert_tail(bioq, bio);
1333 		return;
1334 	}
1335 #endif
1336 
1337 	switch(bio->bio_buf->b_cmd) {
1338 	case BUF_CMD_READ:
1339 		if (bioq->transition) {
1340 			/*
1341 			 * Insert before the first write.  Bleedover writes
1342 			 * based on reorder intervals to prevent starvation.
1343 			 */
1344 			TAILQ_INSERT_BEFORE(bioq->transition, bio, bio_act);
1345 			++bioq->reorder;
1346 			if (bioq->reorder % bioq_reorder_minor_interval == 0) {
1347 				bioqwritereorder(bioq);
1348 				if (bioq->reorder >=
1349 				    bioq_reorder_burst_interval) {
1350 					bioq->reorder = 0;
1351 				}
1352 			}
1353 		} else {
1354 			/*
1355 			 * No writes queued (or ordering was forced),
1356 			 * insert at tail.
1357 			 */
1358 			TAILQ_INSERT_TAIL(&bioq->queue, bio, bio_act);
1359 		}
1360 		break;
1361 	case BUF_CMD_WRITE:
1362 		/*
1363 		 * Writes are always appended.  If no writes were previously
1364 		 * queued or an ordered tail insertion occured the transition
1365 		 * field will be NULL.
1366 		 */
1367 		TAILQ_INSERT_TAIL(&bioq->queue, bio, bio_act);
1368 		if (bioq->transition == NULL)
1369 			bioq->transition = bio;
1370 		break;
1371 	default:
1372 		/*
1373 		 * All other request types are forced to be ordered.
1374 		 */
1375 		bioq_insert_tail(bioq, bio);
1376 		break;
1377 	}
1378 }
1379 
1380 /*
1381  * Move the read-write transition point to prevent reads from
1382  * completely starving our writes.  This brings a number of writes into
1383  * the fold every N reads.
1384  *
1385  * We bring a few linear writes into the fold on a minor interval
1386  * and we bring a non-linear burst of writes into the fold on a major
1387  * interval.  Bursting only occurs if runningbufspace is really high
1388  * (typically from syncs, fsyncs, or HAMMER flushes).
1389  */
1390 static
1391 void
1392 bioqwritereorder(struct bio_queue_head *bioq)
1393 {
1394 	struct bio *bio;
1395 	off_t next_offset;
1396 	size_t left;
1397 	size_t n;
1398 	int check_off;
1399 
1400 	if (bioq->reorder < bioq_reorder_burst_interval ||
1401 	    !buf_runningbufspace_severe()) {
1402 		left = (size_t)bioq_reorder_minor_bytes;
1403 		check_off = 1;
1404 	} else {
1405 		left = (size_t)bioq_reorder_burst_bytes;
1406 		check_off = 0;
1407 	}
1408 
1409 	next_offset = bioq->transition->bio_offset;
1410 	while ((bio = bioq->transition) != NULL &&
1411 	       (check_off == 0 || next_offset == bio->bio_offset)
1412 	) {
1413 		n = bio->bio_buf->b_bcount;
1414 		next_offset = bio->bio_offset + n;
1415 		bioq->transition = TAILQ_NEXT(bio, bio_act);
1416 		if (left < n)
1417 			break;
1418 		left -= n;
1419 	}
1420 }
1421 
1422 /*
1423  * Bounds checking against the media size, used for the raw partition.
1424  * secsize, mediasize and b_blkno must all be the same units.
1425  * Possibly this has to be DEV_BSIZE (512).
1426  */
1427 int
1428 bounds_check_with_mediasize(struct bio *bio, int secsize, uint64_t mediasize)
1429 {
1430 	struct buf *bp = bio->bio_buf;
1431 	int64_t sz;
1432 
1433 	sz = howmany(bp->b_bcount, secsize);
1434 
1435 	if (bio->bio_offset/DEV_BSIZE + sz > mediasize) {
1436 		sz = mediasize - bio->bio_offset/DEV_BSIZE;
1437 		if (sz == 0) {
1438 			/* If exactly at end of disk, return EOF. */
1439 			bp->b_resid = bp->b_bcount;
1440 			return 0;
1441 		}
1442 		if (sz < 0) {
1443 			/* If past end of disk, return EINVAL. */
1444 			bp->b_error = EINVAL;
1445 			return 0;
1446 		}
1447 		/* Otherwise, truncate request. */
1448 		bp->b_bcount = sz * secsize;
1449 	}
1450 
1451 	return 1;
1452 }
1453 
1454 /*
1455  * Disk error is the preface to plaintive error messages
1456  * about failing disk transfers.  It prints messages of the form
1457 
1458 hp0g: hard error reading fsbn 12345 of 12344-12347 (hp0 bn %d cn %d tn %d sn %d)
1459 
1460  * if the offset of the error in the transfer and a disk label
1461  * are both available.  blkdone should be -1 if the position of the error
1462  * is unknown; the disklabel pointer may be null from drivers that have not
1463  * been converted to use them.  The message is printed with kprintf
1464  * if pri is LOG_PRINTF, otherwise it uses log at the specified priority.
1465  * The message should be completed (with at least a newline) with kprintf
1466  * or log(-1, ...), respectively.  There is no trailing space.
1467  */
1468 void
1469 diskerr(struct bio *bio, cdev_t dev, const char *what, int pri, int donecnt)
1470 {
1471 	struct buf *bp = bio->bio_buf;
1472 	const char *term;
1473 
1474 	switch(bp->b_cmd) {
1475 	case BUF_CMD_READ:
1476 		term = "read";
1477 		break;
1478 	case BUF_CMD_WRITE:
1479 		term = "write";
1480 		break;
1481 	default:
1482 		term = "access";
1483 		break;
1484 	}
1485 	kprintf("%s: %s %sing ", dev->si_name, what, term);
1486 	kprintf("offset %012llx for %d",
1487 		(long long)bio->bio_offset,
1488 		bp->b_bcount);
1489 
1490 	if (donecnt)
1491 		kprintf(" (%d bytes completed)", donecnt);
1492 }
1493 
1494 /*
1495  * Locate a disk device
1496  */
1497 cdev_t
1498 disk_locate(const char *devname)
1499 {
1500 	return devfs_find_device_by_name("%s", devname);
1501 }
1502 
1503 void
1504 disk_config(void *arg)
1505 {
1506 	disk_msg_send_sync(DISK_SYNC, NULL, NULL);
1507 }
1508 
1509 static void
1510 disk_init(void)
1511 {
1512 	struct thread* td_core;
1513 
1514 	disk_msg_cache = objcache_create("disk-msg-cache", 0, 0,
1515 					 NULL, NULL, NULL,
1516 					 objcache_malloc_alloc,
1517 					 objcache_malloc_free,
1518 					 &disk_msg_malloc_args);
1519 
1520 	lwkt_token_init(&disklist_token, "disks");
1521 	lwkt_token_init(&ds_token, "ds");
1522 
1523 	/*
1524 	 * Initialize the reply-only port which acts as a message drain
1525 	 */
1526 	lwkt_initport_replyonly(&disk_dispose_port, disk_msg_autofree_reply);
1527 
1528 	lwkt_gettoken(&disklist_token);
1529 	lwkt_create(disk_msg_core, /*args*/NULL, &td_core, NULL,
1530 		    0, -1, "disk_msg_core");
1531 	tsleep(td_core, 0, "diskcore", 0);
1532 	lwkt_reltoken(&disklist_token);
1533 }
1534 
1535 static void
1536 disk_uninit(void)
1537 {
1538 	objcache_destroy(disk_msg_cache);
1539 }
1540 
1541 /*
1542  * Clean out illegal characters in serial numbers.
1543  */
1544 static void
1545 disk_cleanserial(char *serno)
1546 {
1547 	char c;
1548 
1549 	while ((c = *serno) != 0) {
1550 		if (c >= 'a' && c <= 'z')
1551 			;
1552 		else if (c >= 'A' && c <= 'Z')
1553 			;
1554 		else if (c >= '0' && c <= '9')
1555 			;
1556 		else if (c == '-' || c == '@' || c == '+' || c == '.')
1557 			;
1558 		else
1559 			c = '_';
1560 		*serno++= c;
1561 	}
1562 }
1563 
1564 TUNABLE_INT("kern.disk_debug", &disk_debug_enable);
1565 SYSCTL_INT(_kern, OID_AUTO, disk_debug, CTLFLAG_RW, &disk_debug_enable,
1566 	   0, "Enable subr_disk debugging");
1567 
1568 SYSINIT(disk_register, SI_SUB_PRE_DRIVERS, SI_ORDER_FIRST, disk_init, NULL);
1569 SYSUNINIT(disk_register, SI_SUB_PRE_DRIVERS, SI_ORDER_ANY, disk_uninit, NULL);
1570