2 * Copyright (c) 2003,2004,2009 The DragonFly Project. All rights reserved.
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>
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
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
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.
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
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 * ----------------------------------------------------------------------------
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.
50 * Redistribution and use in source and binary forms, with or without
51 * modification, are permitted provided that the following conditions
53 * 1. Redistributions of source code must retain the above copyright
54 * notice, this list of conditions and the following disclaimer.
55 * 2. Redistributions in binary form must reproduce the above copyright
56 * notice, this list of conditions and the following disclaimer in the
57 * documentation and/or other materials provided with the distribution.
58 * 3. All advertising materials mentioning features or use of this software
59 * must display the following acknowledgement:
60 * This product includes software developed by the University of
61 * California, Berkeley and its contributors.
62 * 4. Neither the name of the University nor the names of its contributors
63 * may be used to endorse or promote products derived from this software
64 * without specific prior written permission.
66 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
67 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
68 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
69 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
70 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
71 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
72 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
73 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
74 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
75 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
78 * @(#)ufs_disksubr.c 8.5 (Berkeley) 1/21/94
79 * $FreeBSD: src/sys/kern/subr_disk.c,v 1.20.2.6 2001/10/05 07:14:57 peter Exp $
80 * $FreeBSD: src/sys/ufs/ufs/ufs_disksubr.c,v 1.44.2.3 2001/03/05 05:42:19 obrien Exp $
83 #include <sys/param.h>
84 #include <sys/systm.h>
85 #include <sys/kernel.h>
87 #include <sys/sysctl.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>
96 #include <sys/kerneldump.h>
97 #include <sys/malloc.h>
98 #include <machine/md_var.h>
99 #include <sys/ctype.h>
100 #include <sys/syslog.h>
101 #include <sys/device.h>
102 #include <sys/msgport.h>
103 #include <sys/devfs.h>
104 #include <sys/thread.h>
105 #include <sys/dsched.h>
106 #include <sys/queue.h>
107 #include <sys/lock.h>
108 #include <sys/udev.h>
109 #include <sys/uuid.h>
111 #include <sys/buf2.h>
112 #include <sys/mplock2.h>
113 #include <sys/msgport2.h>
114 #include <sys/thread2.h>
116 static MALLOC_DEFINE(M_DISK, "disk", "disk data");
117 static int disk_debug_enable = 0;
119 static void disk_msg_autofree_reply(lwkt_port_t, lwkt_msg_t);
120 static void disk_msg_core(void *);
121 static int disk_probe_slice(struct disk *dp, cdev_t dev, int slice, int reprobe);
122 static void disk_probe(struct disk *dp, int reprobe);
123 static void _setdiskinfo(struct disk *disk, struct disk_info *info);
124 static void bioqwritereorder(struct bio_queue_head *bioq);
125 static void disk_cleanserial(char *serno);
126 static int disk_debug(int, char *, ...) __printflike(2, 3);
127 static cdev_t _disk_create_named(const char *name, int unit, struct disk *dp,
128 struct dev_ops *raw_ops, int clone);
130 static d_open_t diskopen;
131 static d_close_t diskclose;
132 static d_ioctl_t diskioctl;
133 static d_strategy_t diskstrategy;
134 static d_psize_t diskpsize;
135 static d_dump_t diskdump;
137 static LIST_HEAD(, disk) disklist = LIST_HEAD_INITIALIZER(&disklist);
138 static struct lwkt_token disklist_token;
140 static struct dev_ops disk_ops = {
141 { "disk", 0, D_DISK | D_MPSAFE | D_TRACKCLOSE },
143 .d_close = diskclose,
145 .d_write = physwrite,
146 .d_ioctl = diskioctl,
147 .d_strategy = diskstrategy,
149 .d_psize = diskpsize,
152 static struct objcache *disk_msg_cache;
154 struct objcache_malloc_args disk_msg_malloc_args = {
155 sizeof(struct disk_msg), M_DISK };
157 static struct lwkt_port disk_dispose_port;
158 static struct lwkt_port disk_msg_port;
161 disk_debug(int level, char *fmt, ...)
166 if (level <= disk_debug_enable)
174 disk_probe_slice(struct disk *dp, cdev_t dev, int slice, int reprobe)
176 struct disk_info *info = &dp->d_info;
177 struct diskslice *sp = &dp->d_slice->dss_slices[slice];
179 struct partinfo part;
186 disk_debug(2, "disk_probe_slice (begin): %s (%s)\n",
187 dev->si_name, dp->d_cdev->si_name);
189 sno = slice ? slice - 1 : 0;
191 ops = &disklabel32_ops;
192 msg = ops->op_readdisklabel(dev, sp, &sp->ds_label, info);
193 if (msg && !strcmp(msg, "no disk label")) {
194 ops = &disklabel64_ops;
195 msg = ops->op_readdisklabel(dev, sp, &sp->ds_label, info);
199 if (slice != WHOLE_DISK_SLICE)
200 ops->op_adjust_label_reserved(dp->d_slice, slice, sp);
205 for (i = 0; i < ops->op_getnumparts(sp->ds_label); i++) {
206 ops->op_loadpartinfo(sp->ds_label, i, &part);
209 (ndev = devfs_find_device_by_name("%s%c",
210 dev->si_name, 'a' + i))
213 * Device already exists and
216 ndev->si_flags |= SI_REPROBE_TEST;
219 * Destroy old UUID alias
221 destroy_dev_alias(ndev, "part-by-uuid/*");
223 /* Create UUID alias */
224 if (!kuuid_is_nil(&part.storage_uuid)) {
225 snprintf_uuid(uuid_buf,
231 udev_dict_set_cstr(ndev, "uuid", uuid_buf);
234 ndev = make_dev_covering(&disk_ops, dp->d_rawdev->si_ops,
235 dkmakeminor(dkunit(dp->d_cdev),
237 UID_ROOT, GID_OPERATOR, 0640,
238 "%s%c", dev->si_name, 'a'+ i);
239 ndev->si_parent = dev;
240 ndev->si_iosize_max = dev->si_iosize_max;
242 udev_dict_set_cstr(ndev, "subsystem", "disk");
243 /* Inherit parent's disk type */
244 if (dp->d_disktype) {
245 udev_dict_set_cstr(ndev, "disk-type",
246 __DECONST(char *, dp->d_disktype));
249 /* Create serno alias */
250 if (dp->d_info.d_serialno) {
253 dp->d_info.d_serialno,
257 /* Create UUID alias */
258 if (!kuuid_is_nil(&part.storage_uuid)) {
259 snprintf_uuid(uuid_buf,
265 udev_dict_set_cstr(ndev, "uuid", uuid_buf);
267 ndev->si_flags |= SI_REPROBE_TEST;
271 } else if (info->d_dsflags & DSO_COMPATLABEL) {
273 if (sp->ds_size >= 0x100000000ULL)
274 ops = &disklabel64_ops;
276 ops = &disklabel32_ops;
277 sp->ds_label = ops->op_clone_label(info, sp);
279 if (sp->ds_type == DOSPTYP_386BSD || /* XXX */
280 sp->ds_type == DOSPTYP_NETBSD ||
281 sp->ds_type == DOSPTYP_OPENBSD) {
282 log(LOG_WARNING, "%s: cannot find label (%s)\n",
286 if (sp->ds_label.opaque != NULL && sp->ds_ops != NULL) {
287 /* Clear out old label - it's not around anymore */
289 "disk_probe_slice: clear out old diskabel on %s\n",
292 sp->ds_ops->op_freedisklabel(&sp->ds_label);
298 sp->ds_wlabel = FALSE;
301 return (msg ? EINVAL : 0);
305 * This routine is only called for newly minted drives or to reprobe
306 * a drive with no open slices. disk_probe_slice() is called directly
307 * when reprobing partition changes within slices.
310 disk_probe(struct disk *dp, int reprobe)
312 struct disk_info *info = &dp->d_info;
313 cdev_t dev = dp->d_cdev;
316 struct diskslices *osp;
317 struct diskslice *sp;
320 KKASSERT (info->d_media_blksize != 0);
323 dp->d_slice = dsmakeslicestruct(BASE_SLICE, info);
324 disk_debug(1, "disk_probe (begin): %s\n", dp->d_cdev->si_name);
326 error = mbrinit(dev, info, &(dp->d_slice));
332 for (i = 0; i < dp->d_slice->dss_nslices; i++) {
334 * Ignore the whole-disk slice, it has already been created.
336 if (i == WHOLE_DISK_SLICE)
341 * Ignore the compatibility slice s0 if it's a device mapper
344 if ((i == COMPATIBILITY_SLICE) &&
345 (info->d_dsflags & DSO_DEVICEMAPPER))
349 sp = &dp->d_slice->dss_slices[i];
352 * Handle s0. s0 is a compatibility slice if there are no
353 * other slices and it has not otherwise been set up, else
356 if (i == COMPATIBILITY_SLICE) {
358 if (sp->ds_type == 0 &&
359 dp->d_slice->dss_nslices == BASE_SLICE) {
360 sp->ds_size = info->d_media_blocks;
369 * Ignore 0-length slices
371 if (sp->ds_size == 0)
375 (ndev = devfs_find_device_by_name("%ss%d",
376 dev->si_name, sno))) {
378 * Device already exists and is still valid
380 ndev->si_flags |= SI_REPROBE_TEST;
383 * Destroy old UUID alias
385 destroy_dev_alias(ndev, "slice-by-uuid/*");
387 /* Create UUID alias */
388 if (!kuuid_is_nil(&sp->ds_stor_uuid)) {
389 snprintf_uuid(uuid_buf, sizeof(uuid_buf),
391 make_dev_alias(ndev, "slice-by-uuid/%s",
396 * Else create new device
398 ndev = make_dev_covering(&disk_ops, dp->d_rawdev->si_ops,
399 dkmakewholeslice(dkunit(dev), i),
400 UID_ROOT, GID_OPERATOR, 0640,
401 (info->d_dsflags & DSO_DEVICEMAPPER)?
402 "%s.s%d" : "%ss%d", dev->si_name, sno);
403 ndev->si_parent = dev;
404 ndev->si_iosize_max = dev->si_iosize_max;
405 udev_dict_set_cstr(ndev, "subsystem", "disk");
406 /* Inherit parent's disk type */
407 if (dp->d_disktype) {
408 udev_dict_set_cstr(ndev, "disk-type",
409 __DECONST(char *, dp->d_disktype));
412 /* Create serno alias */
413 if (dp->d_info.d_serialno) {
414 make_dev_alias(ndev, "serno/%s.s%d",
415 dp->d_info.d_serialno, sno);
418 /* Create UUID alias */
419 if (!kuuid_is_nil(&sp->ds_stor_uuid)) {
420 snprintf_uuid(uuid_buf, sizeof(uuid_buf),
422 make_dev_alias(ndev, "slice-by-uuid/%s",
427 ndev->si_flags |= SI_REPROBE_TEST;
432 * Probe appropriate slices for a disklabel
434 * XXX slice type 1 used by our gpt probe code.
435 * XXX slice type 0 used by mbr compat slice.
437 if (sp->ds_type == DOSPTYP_386BSD ||
438 sp->ds_type == DOSPTYP_NETBSD ||
439 sp->ds_type == DOSPTYP_OPENBSD ||
442 if (dp->d_slice->dss_first_bsd_slice == 0)
443 dp->d_slice->dss_first_bsd_slice = i;
444 disk_probe_slice(dp, ndev, i, reprobe);
448 disk_debug(1, "disk_probe (end): %s\n", dp->d_cdev->si_name);
453 disk_msg_core(void *arg)
456 struct diskslice *sp;
460 lwkt_gettoken(&disklist_token);
461 lwkt_initport_thread(&disk_msg_port, curthread);
462 wakeup(curthread); /* synchronous startup */
463 lwkt_reltoken(&disklist_token);
465 get_mplock(); /* not mpsafe yet? */
469 msg = (disk_msg_t)lwkt_waitport(&disk_msg_port, 0);
471 switch (msg->hdr.u.ms_result) {
472 case DISK_DISK_PROBE:
473 dp = (struct disk *)msg->load;
475 "DISK_DISK_PROBE: %s\n",
476 dp->d_cdev->si_name);
477 disk_iocom_update(dp);
480 case DISK_DISK_DESTROY:
481 dp = (struct disk *)msg->load;
483 "DISK_DISK_DESTROY: %s\n",
484 dp->d_cdev->si_name);
485 disk_iocom_uninit(dp);
486 devfs_destroy_related(dp->d_cdev);
487 destroy_dev(dp->d_cdev);
488 destroy_only_dev(dp->d_rawdev);
489 lwkt_gettoken(&disklist_token);
490 LIST_REMOVE(dp, d_list);
491 lwkt_reltoken(&disklist_token);
492 if (dp->d_info.d_serialno) {
493 kfree(dp->d_info.d_serialno, M_TEMP);
494 dp->d_info.d_serialno = NULL;
498 dp = (struct disk *)msg->load;
500 "DISK_DISK_UNPROBE: %s\n",
501 dp->d_cdev->si_name);
502 devfs_destroy_related(dp->d_cdev);
504 case DISK_SLICE_REPROBE:
505 dp = (struct disk *)msg->load;
506 sp = (struct diskslice *)msg->load2;
507 devfs_clr_related_flag(sp->ds_dev,
510 "DISK_SLICE_REPROBE: %s\n",
511 sp->ds_dev->si_name);
512 disk_probe_slice(dp, sp->ds_dev,
513 dkslice(sp->ds_dev), 1);
514 devfs_destroy_related_without_flag(
515 sp->ds_dev, SI_REPROBE_TEST);
517 case DISK_DISK_REPROBE:
518 dp = (struct disk *)msg->load;
519 devfs_clr_related_flag(dp->d_cdev, SI_REPROBE_TEST);
521 "DISK_DISK_REPROBE: %s\n",
522 dp->d_cdev->si_name);
524 devfs_destroy_related_without_flag(
525 dp->d_cdev, SI_REPROBE_TEST);
528 disk_debug(1, "DISK_SYNC\n");
531 devfs_debug(DEVFS_DEBUG_WARNING,
532 "disk_msg_core: unknown message "
533 "received at core\n");
536 lwkt_replymsg(&msg->hdr, 0);
543 * Acts as a message drain. Any message that is replied to here gets
544 * destroyed and the memory freed.
547 disk_msg_autofree_reply(lwkt_port_t port, lwkt_msg_t msg)
549 objcache_put(disk_msg_cache, msg);
554 disk_msg_send(uint32_t cmd, void *load, void *load2)
557 lwkt_port_t port = &disk_msg_port;
559 disk_msg = objcache_get(disk_msg_cache, M_WAITOK);
561 lwkt_initmsg(&disk_msg->hdr, &disk_dispose_port, 0);
563 disk_msg->hdr.u.ms_result = cmd;
564 disk_msg->load = load;
565 disk_msg->load2 = load2;
567 lwkt_sendmsg(port, &disk_msg->hdr);
571 disk_msg_send_sync(uint32_t cmd, void *load, void *load2)
573 struct lwkt_port rep_port;
577 disk_msg = objcache_get(disk_msg_cache, M_WAITOK);
578 port = &disk_msg_port;
580 /* XXX could probably use curthread's built-in msgport */
581 lwkt_initport_thread(&rep_port, curthread);
582 lwkt_initmsg(&disk_msg->hdr, &rep_port, 0);
584 disk_msg->hdr.u.ms_result = cmd;
585 disk_msg->load = load;
586 disk_msg->load2 = load2;
588 lwkt_sendmsg(port, &disk_msg->hdr);
589 lwkt_waitmsg(&disk_msg->hdr, 0);
590 objcache_put(disk_msg_cache, disk_msg);
594 * Create a raw device for the dev_ops template (which is returned). Also
595 * create a slice and unit managed disk and overload the user visible
596 * device space with it.
598 * NOTE: The returned raw device is NOT a slice and unit managed device.
599 * It is an actual raw device representing the raw disk as specified by
600 * the passed dev_ops. The disk layer not only returns such a raw device,
601 * it also uses it internally when passing (modified) commands through.
604 disk_create(int unit, struct disk *dp, struct dev_ops *raw_ops)
606 return _disk_create_named(NULL, unit, dp, raw_ops, 0);
610 disk_create_clone(int unit, struct disk *dp,
611 struct dev_ops *raw_ops)
613 return _disk_create_named(NULL, unit, dp, raw_ops, 1);
617 disk_create_named(const char *name, int unit, struct disk *dp,
618 struct dev_ops *raw_ops)
620 return _disk_create_named(name, unit, dp, raw_ops, 0);
624 disk_create_named_clone(const char *name, int unit, struct disk *dp,
625 struct dev_ops *raw_ops)
627 return _disk_create_named(name, unit, dp, raw_ops, 1);
631 _disk_create_named(const char *name, int unit, struct disk *dp,
632 struct dev_ops *raw_ops, int clone)
636 disk_debug(1, "disk_create (begin): %s%d\n", name, unit);
639 rawdev = make_only_dev(raw_ops, dkmakewholedisk(unit),
640 UID_ROOT, GID_OPERATOR, 0640, "%s", name);
642 rawdev = make_only_dev(raw_ops, dkmakewholedisk(unit),
643 UID_ROOT, GID_OPERATOR, 0640,
644 "%s%d", raw_ops->head.name, unit);
647 bzero(dp, sizeof(*dp));
649 dp->d_rawdev = rawdev;
650 dp->d_raw_ops = raw_ops;
651 dp->d_dev_ops = &disk_ops;
655 dp->d_cdev = make_only_dev_covering(
656 &disk_ops, dp->d_rawdev->si_ops,
657 dkmakewholedisk(unit),
658 UID_ROOT, GID_OPERATOR, 0640,
661 dp->d_cdev = make_dev_covering(
662 &disk_ops, dp->d_rawdev->si_ops,
663 dkmakewholedisk(unit),
664 UID_ROOT, GID_OPERATOR, 0640,
669 dp->d_cdev = make_only_dev_covering(
670 &disk_ops, dp->d_rawdev->si_ops,
671 dkmakewholedisk(unit),
672 UID_ROOT, GID_OPERATOR, 0640,
673 "%s%d", raw_ops->head.name, unit);
675 dp->d_cdev = make_dev_covering(
676 &disk_ops, dp->d_rawdev->si_ops,
677 dkmakewholedisk(unit),
678 UID_ROOT, GID_OPERATOR, 0640,
679 "%s%d", raw_ops->head.name, unit);
683 udev_dict_set_cstr(dp->d_cdev, "subsystem", "disk");
684 dp->d_cdev->si_disk = dp;
687 dsched_disk_create_callback(dp, name, unit);
689 dsched_disk_create_callback(dp, raw_ops->head.name, unit);
691 lwkt_gettoken(&disklist_token);
692 LIST_INSERT_HEAD(&disklist, dp, d_list);
693 lwkt_reltoken(&disklist_token);
697 disk_debug(1, "disk_create (end): %s%d\n",
698 (name != NULL)?(name):(raw_ops->head.name), unit);
700 return (dp->d_rawdev);
704 disk_setdisktype(struct disk *disk, const char *type)
708 KKASSERT(disk != NULL);
710 disk->d_disktype = type;
711 error = udev_dict_set_cstr(disk->d_cdev, "disk-type",
712 __DECONST(char *, type));
717 disk_getopencount(struct disk *disk)
719 return disk->d_opencount;
723 _setdiskinfo(struct disk *disk, struct disk_info *info)
727 oldserialno = disk->d_info.d_serialno;
728 bcopy(info, &disk->d_info, sizeof(disk->d_info));
729 info = &disk->d_info;
731 disk_debug(1, "_setdiskinfo: %s\n", disk->d_cdev->si_name);
734 * The serial number is duplicated so the caller can throw
737 if (info->d_serialno && info->d_serialno[0] &&
738 (info->d_serialno[0] != ' ' || strlen(info->d_serialno) > 1)) {
739 info->d_serialno = kstrdup(info->d_serialno, M_TEMP);
740 disk_cleanserial(info->d_serialno);
742 make_dev_alias(disk->d_cdev, "serno/%s",
746 info->d_serialno = NULL;
749 kfree(oldserialno, M_TEMP);
751 dsched_disk_update_callback(disk, info);
754 * The caller may set d_media_size or d_media_blocks and we
755 * calculate the other.
757 KKASSERT(info->d_media_size == 0 || info->d_media_blocks == 0);
758 if (info->d_media_size == 0 && info->d_media_blocks) {
759 info->d_media_size = (u_int64_t)info->d_media_blocks *
760 info->d_media_blksize;
761 } else if (info->d_media_size && info->d_media_blocks == 0 &&
762 info->d_media_blksize) {
763 info->d_media_blocks = info->d_media_size /
764 info->d_media_blksize;
768 * The si_* fields for rawdev are not set until after the
769 * disk_create() call, so someone using the cooked version
770 * of the raw device (i.e. da0s0) will not get the right
771 * si_iosize_max unless we fix it up here.
773 if (disk->d_cdev && disk->d_rawdev &&
774 disk->d_cdev->si_iosize_max == 0) {
775 disk->d_cdev->si_iosize_max = disk->d_rawdev->si_iosize_max;
776 disk->d_cdev->si_bsize_phys = disk->d_rawdev->si_bsize_phys;
777 disk->d_cdev->si_bsize_best = disk->d_rawdev->si_bsize_best;
780 /* Add the serial number to the udev_dictionary */
781 if (info->d_serialno)
782 udev_dict_set_cstr(disk->d_cdev, "serno", info->d_serialno);
786 * Disk drivers must call this routine when media parameters are available
790 disk_setdiskinfo(struct disk *disk, struct disk_info *info)
792 _setdiskinfo(disk, info);
793 disk_msg_send(DISK_DISK_PROBE, disk, NULL);
794 disk_debug(1, "disk_setdiskinfo: sent probe for %s\n",
795 disk->d_cdev->si_name);
799 disk_setdiskinfo_sync(struct disk *disk, struct disk_info *info)
801 _setdiskinfo(disk, info);
802 disk_msg_send_sync(DISK_DISK_PROBE, disk, NULL);
803 disk_debug(1, "disk_setdiskinfo_sync: sent probe for %s\n",
804 disk->d_cdev->si_name);
808 * This routine is called when an adapter detaches. The higher level
809 * managed disk device is destroyed while the lower level raw device is
813 disk_destroy(struct disk *disk)
815 dsched_disk_destroy_callback(disk);
816 disk_msg_send_sync(DISK_DISK_DESTROY, disk, NULL);
821 disk_dumpcheck(cdev_t dev, u_int64_t *size,
822 u_int64_t *blkno, u_int32_t *secsize)
824 struct partinfo pinfo;
827 bzero(&pinfo, sizeof(pinfo));
828 error = dev_dioctl(dev, DIOCGPART, (void *)&pinfo, 0,
829 proc0.p_ucred, NULL);
833 if (pinfo.media_blksize == 0)
836 if (blkno) /* XXX: make sure this reserved stuff is right */
837 *blkno = pinfo.reserved_blocks +
838 pinfo.media_offset / pinfo.media_blksize;
840 *secsize = pinfo.media_blksize;
842 *size = (pinfo.media_blocks - pinfo.reserved_blocks);
848 disk_dumpconf(cdev_t dev, u_int onoff)
850 struct dumperinfo di;
851 u_int64_t size, blkno;
856 return set_dumper(NULL);
858 error = disk_dumpcheck(dev, &size, &blkno, &secsize);
863 bzero(&di, sizeof(struct dumperinfo));
864 di.dumper = diskdump;
866 di.blocksize = secsize;
867 di.maxiosize = dev->si_iosize_max;
868 di.mediaoffset = blkno * DEV_BSIZE;
869 di.mediasize = size * DEV_BSIZE;
871 return set_dumper(&di);
875 disk_unprobe(struct disk *disk)
880 disk_msg_send_sync(DISK_UNPROBE, disk, NULL);
884 disk_invalidate (struct disk *disk)
886 dsgone(&disk->d_slice);
890 disk_enumerate(struct disk *disk)
894 lwkt_gettoken(&disklist_token);
896 dp = (LIST_FIRST(&disklist));
898 dp = (LIST_NEXT(disk, d_list));
899 lwkt_reltoken(&disklist_token);
906 sysctl_disks(SYSCTL_HANDLER_ARGS)
914 while ((disk = disk_enumerate(disk))) {
916 error = SYSCTL_OUT(req, " ", 1);
922 error = SYSCTL_OUT(req, disk->d_rawdev->si_name,
923 strlen(disk->d_rawdev->si_name));
927 error = SYSCTL_OUT(req, "", 1);
931 SYSCTL_PROC(_kern, OID_AUTO, disks, CTLTYPE_STRING | CTLFLAG_RD, NULL, 0,
932 sysctl_disks, "A", "names of available disks");
935 * Open a disk device or partition.
939 diskopen(struct dev_open_args *ap)
941 cdev_t dev = ap->a_head.a_dev;
946 * dp can't be NULL here XXX.
948 * d_slice will be NULL if setdiskinfo() has not been called yet.
949 * setdiskinfo() is typically called whether the disk is present
950 * or not (e.g. CD), but the base disk device is created first
951 * and there may be a race.
954 if (dp == NULL || dp->d_slice == NULL)
959 * Deal with open races
962 while (dp->d_flags & DISKFLAG_LOCK) {
963 dp->d_flags |= DISKFLAG_WANTED;
964 error = tsleep(dp, PCATCH, "diskopen", hz);
970 dp->d_flags |= DISKFLAG_LOCK;
973 * Open the underlying raw device.
975 if (!dsisopen(dp->d_slice)) {
977 if (!pdev->si_iosize_max)
978 pdev->si_iosize_max = dev->si_iosize_max;
980 error = dev_dopen(dp->d_rawdev, ap->a_oflags,
981 ap->a_devtype, ap->a_cred);
986 error = dsopen(dev, ap->a_devtype, dp->d_info.d_dsflags,
987 &dp->d_slice, &dp->d_info);
988 if (!dsisopen(dp->d_slice)) {
989 dev_dclose(dp->d_rawdev, ap->a_oflags, ap->a_devtype);
992 dp->d_flags &= ~DISKFLAG_LOCK;
993 if (dp->d_flags & DISKFLAG_WANTED) {
994 dp->d_flags &= ~DISKFLAG_WANTED;
999 KKASSERT(dp->d_opencount >= 0);
1000 /* If the open was successful, bump open count */
1002 atomic_add_int(&dp->d_opencount, 1);
1008 * Close a disk device or partition
1012 diskclose(struct dev_close_args *ap)
1014 cdev_t dev = ap->a_head.a_dev;
1023 * The cdev_t represents the disk/slice/part. The shared
1024 * dp structure governs all cdevs associated with the disk.
1026 * As a safety only close the underlying raw device on the last
1027 * close the disk device if our tracking of the slices/partitions
1028 * also indicates nothing is open.
1030 KKASSERT(dp->d_opencount >= 1);
1031 lcount = atomic_fetchadd_int(&dp->d_opencount, -1);
1034 dsclose(dev, ap->a_devtype, dp->d_slice);
1035 if (lcount <= 1 && !dsisopen(dp->d_slice)) {
1036 error = dev_dclose(dp->d_rawdev, ap->a_fflag, ap->a_devtype);
1043 * First execute the ioctl on the disk device, and if it isn't supported
1044 * try running it on the backing device.
1048 diskioctl(struct dev_ioctl_args *ap)
1050 cdev_t dev = ap->a_head.a_dev;
1059 devfs_debug(DEVFS_DEBUG_DEBUG,
1060 "diskioctl: cmd is: %lx (name: %s)\n",
1061 ap->a_cmd, dev->si_name);
1062 devfs_debug(DEVFS_DEBUG_DEBUG,
1063 "diskioctl: &dp->d_slice is: %p, %p\n",
1064 &dp->d_slice, dp->d_slice);
1066 if (ap->a_cmd == DIOCGKERNELDUMP) {
1067 u = *(u_int *)ap->a_data;
1068 return disk_dumpconf(dev, u);
1071 if (ap->a_cmd == DIOCRECLUSTER && dev == dp->d_cdev) {
1072 kprintf("RECLUSTER\n");
1073 error = disk_iocom_ioctl(dp, ap->a_cmd, ap->a_data);
1077 if (&dp->d_slice == NULL || dp->d_slice == NULL ||
1078 ((dp->d_info.d_dsflags & DSO_DEVICEMAPPER) &&
1079 dkslice(dev) == WHOLE_DISK_SLICE)) {
1083 error = dsioctl(dev, ap->a_cmd, ap->a_data, ap->a_fflag,
1084 &dp->d_slice, &dp->d_info);
1088 if (error == ENOIOCTL) {
1089 error = dev_dioctl(dp->d_rawdev, ap->a_cmd, ap->a_data,
1090 ap->a_fflag, ap->a_cred, NULL);
1096 * Execute strategy routine
1100 diskstrategy(struct dev_strategy_args *ap)
1102 cdev_t dev = ap->a_head.a_dev;
1103 struct bio *bio = ap->a_bio;
1110 bio->bio_buf->b_error = ENXIO;
1111 bio->bio_buf->b_flags |= B_ERROR;
1115 KKASSERT(dev->si_disk == dp);
1118 * The dscheck() function will also transform the slice relative
1119 * block number i.e. bio->bio_offset into a block number that can be
1120 * passed directly to the underlying raw device. If dscheck()
1121 * returns NULL it will have handled the bio for us (e.g. EOF
1122 * or error due to being beyond the device size).
1124 if ((nbio = dscheck(dev, bio, dp->d_slice)) != NULL) {
1125 dsched_queue(dp, nbio);
1133 * Return the partition size in ?blocks?
1137 diskpsize(struct dev_psize_args *ap)
1139 cdev_t dev = ap->a_head.a_dev;
1146 ap->a_result = dssize(dev, &dp->d_slice);
1148 if ((ap->a_result == -1) &&
1149 (dp->d_info.d_dsflags & DSO_RAWPSIZE)) {
1150 ap->a_head.a_dev = dp->d_rawdev;
1151 return dev_doperate(&ap->a_head);
1157 diskdump(struct dev_dump_args *ap)
1159 cdev_t dev = ap->a_head.a_dev;
1160 struct disk *dp = dev->si_disk;
1161 u_int64_t size, offset;
1164 error = disk_dumpcheck(dev, &size, &ap->a_blkno, &ap->a_secsize);
1165 /* XXX: this should probably go in disk_dumpcheck somehow */
1166 if (ap->a_length != 0) {
1168 offset = ap->a_blkno * DEV_BSIZE;
1169 if ((ap->a_offset < offset) ||
1170 (ap->a_offset + ap->a_length - offset > size)) {
1171 kprintf("Attempt to write outside dump "
1172 "device boundaries.\n");
1178 ap->a_head.a_dev = dp->d_rawdev;
1179 error = dev_doperate(&ap->a_head);
1186 SYSCTL_INT(_debug_sizeof, OID_AUTO, diskslices, CTLFLAG_RD,
1187 0, sizeof(struct diskslices), "sizeof(struct diskslices)");
1189 SYSCTL_INT(_debug_sizeof, OID_AUTO, disk, CTLFLAG_RD,
1190 0, sizeof(struct disk), "sizeof(struct disk)");
1193 * Reorder interval for burst write allowance and minor write
1196 * We always want to trickle some writes in to make use of the
1197 * disk's zone cache. Bursting occurs on a longer interval and only
1198 * runningbufspace is well over the hirunningspace limit.
1200 int bioq_reorder_burst_interval = 60; /* should be multiple of minor */
1201 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_burst_interval,
1202 CTLFLAG_RW, &bioq_reorder_burst_interval, 0, "");
1203 int bioq_reorder_minor_interval = 5;
1204 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_minor_interval,
1205 CTLFLAG_RW, &bioq_reorder_minor_interval, 0, "");
1207 int bioq_reorder_burst_bytes = 3000000;
1208 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_burst_bytes,
1209 CTLFLAG_RW, &bioq_reorder_burst_bytes, 0, "");
1210 int bioq_reorder_minor_bytes = 262144;
1211 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_minor_bytes,
1212 CTLFLAG_RW, &bioq_reorder_minor_bytes, 0, "");
1216 * Order I/Os. Generally speaking this code is designed to make better
1217 * use of drive zone caches. A drive zone cache can typically track linear
1218 * reads or writes for around 16 zones simultaniously.
1220 * Read prioritization issues: It is possible for hundreds of megabytes worth
1221 * of writes to be queued asynchronously. This creates a huge bottleneck
1222 * for reads which reduce read bandwidth to a trickle.
1224 * To solve this problem we generally reorder reads before writes.
1226 * However, a large number of random reads can also starve writes and
1227 * make poor use of the drive zone cache so we allow writes to trickle
1231 bioqdisksort(struct bio_queue_head *bioq, struct bio *bio)
1234 * The BIO wants to be ordered. Adding to the tail also
1235 * causes transition to be set to NULL, forcing the ordering
1236 * of all prior I/O's.
1238 if (bio->bio_buf->b_flags & B_ORDERED) {
1239 bioq_insert_tail(bioq, bio);
1243 switch(bio->bio_buf->b_cmd) {
1245 if (bioq->transition) {
1247 * Insert before the first write. Bleedover writes
1248 * based on reorder intervals to prevent starvation.
1250 TAILQ_INSERT_BEFORE(bioq->transition, bio, bio_act);
1252 if (bioq->reorder % bioq_reorder_minor_interval == 0) {
1253 bioqwritereorder(bioq);
1254 if (bioq->reorder >=
1255 bioq_reorder_burst_interval) {
1261 * No writes queued (or ordering was forced),
1264 TAILQ_INSERT_TAIL(&bioq->queue, bio, bio_act);
1269 * Writes are always appended. If no writes were previously
1270 * queued or an ordered tail insertion occured the transition
1271 * field will be NULL.
1273 TAILQ_INSERT_TAIL(&bioq->queue, bio, bio_act);
1274 if (bioq->transition == NULL)
1275 bioq->transition = bio;
1279 * All other request types are forced to be ordered.
1281 bioq_insert_tail(bioq, bio);
1287 * Move the read-write transition point to prevent reads from
1288 * completely starving our writes. This brings a number of writes into
1289 * the fold every N reads.
1291 * We bring a few linear writes into the fold on a minor interval
1292 * and we bring a non-linear burst of writes into the fold on a major
1293 * interval. Bursting only occurs if runningbufspace is really high
1294 * (typically from syncs, fsyncs, or HAMMER flushes).
1298 bioqwritereorder(struct bio_queue_head *bioq)
1306 if (bioq->reorder < bioq_reorder_burst_interval ||
1307 !buf_runningbufspace_severe()) {
1308 left = (size_t)bioq_reorder_minor_bytes;
1311 left = (size_t)bioq_reorder_burst_bytes;
1315 next_offset = bioq->transition->bio_offset;
1316 while ((bio = bioq->transition) != NULL &&
1317 (check_off == 0 || next_offset == bio->bio_offset)
1319 n = bio->bio_buf->b_bcount;
1320 next_offset = bio->bio_offset + n;
1321 bioq->transition = TAILQ_NEXT(bio, bio_act);
1329 * Bounds checking against the media size, used for the raw partition.
1330 * secsize, mediasize and b_blkno must all be the same units.
1331 * Possibly this has to be DEV_BSIZE (512).
1334 bounds_check_with_mediasize(struct bio *bio, int secsize, uint64_t mediasize)
1336 struct buf *bp = bio->bio_buf;
1339 sz = howmany(bp->b_bcount, secsize);
1341 if (bio->bio_offset/DEV_BSIZE + sz > mediasize) {
1342 sz = mediasize - bio->bio_offset/DEV_BSIZE;
1344 /* If exactly at end of disk, return EOF. */
1345 bp->b_resid = bp->b_bcount;
1349 /* If past end of disk, return EINVAL. */
1350 bp->b_error = EINVAL;
1353 /* Otherwise, truncate request. */
1354 bp->b_bcount = sz * secsize;
1361 * Disk error is the preface to plaintive error messages
1362 * about failing disk transfers. It prints messages of the form
1364 hp0g: hard error reading fsbn 12345 of 12344-12347 (hp0 bn %d cn %d tn %d sn %d)
1366 * if the offset of the error in the transfer and a disk label
1367 * are both available. blkdone should be -1 if the position of the error
1368 * is unknown; the disklabel pointer may be null from drivers that have not
1369 * been converted to use them. The message is printed with kprintf
1370 * if pri is LOG_PRINTF, otherwise it uses log at the specified priority.
1371 * The message should be completed (with at least a newline) with kprintf
1372 * or log(-1, ...), respectively. There is no trailing space.
1375 diskerr(struct bio *bio, cdev_t dev, const char *what, int pri, int donecnt)
1377 struct buf *bp = bio->bio_buf;
1391 kprintf("%s: %s %sing ", dev->si_name, what, term);
1392 kprintf("offset %012llx for %d",
1393 (long long)bio->bio_offset,
1397 kprintf(" (%d bytes completed)", donecnt);
1401 * Locate a disk device
1404 disk_locate(const char *devname)
1406 return devfs_find_device_by_name("%s", devname);
1410 disk_config(void *arg)
1412 disk_msg_send_sync(DISK_SYNC, NULL, NULL);
1418 struct thread* td_core;
1420 disk_msg_cache = objcache_create("disk-msg-cache", 0, 0,
1422 objcache_malloc_alloc,
1423 objcache_malloc_free,
1424 &disk_msg_malloc_args);
1426 lwkt_token_init(&disklist_token, "disks");
1429 * Initialize the reply-only port which acts as a message drain
1431 lwkt_initport_replyonly(&disk_dispose_port, disk_msg_autofree_reply);
1433 lwkt_gettoken(&disklist_token);
1434 lwkt_create(disk_msg_core, /*args*/NULL, &td_core, NULL,
1435 0, -1, "disk_msg_core");
1436 tsleep(td_core, 0, "diskcore", 0);
1437 lwkt_reltoken(&disklist_token);
1443 objcache_destroy(disk_msg_cache);
1447 * Clean out illegal characters in serial numbers.
1450 disk_cleanserial(char *serno)
1454 while ((c = *serno) != 0) {
1455 if (c >= 'a' && c <= 'z')
1457 else if (c >= 'A' && c <= 'Z')
1459 else if (c >= '0' && c <= '9')
1461 else if (c == '-' || c == '@' || c == '+' || c == '.')
1469 TUNABLE_INT("kern.disk_debug", &disk_debug_enable);
1470 SYSCTL_INT(_kern, OID_AUTO, disk_debug, CTLFLAG_RW, &disk_debug_enable,
1471 0, "Enable subr_disk debugging");
1473 SYSINIT(disk_register, SI_SUB_PRE_DRIVERS, SI_ORDER_FIRST, disk_init, NULL);
1474 SYSUNINIT(disk_register, SI_SUB_PRE_DRIVERS, SI_ORDER_ANY, disk_uninit, NULL);