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;
187 "disk_probe_slice (begin): %s (%s)\n",
188 dev->si_name, dp->d_cdev->si_name);
190 sno = slice ? slice - 1 : 0;
192 ops = &disklabel32_ops;
193 msg = ops->op_readdisklabel(dev, sp, &sp->ds_label, info);
194 if (msg && !strcmp(msg, "no disk label")) {
195 ops = &disklabel64_ops;
196 msg = ops->op_readdisklabel(dev, sp, &sp->ds_label, info);
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;
241 udev_dict_set_cstr(ndev, "subsystem", "disk");
242 /* Inherit parent's disk type */
243 if (dp->d_disktype) {
244 udev_dict_set_cstr(ndev, "disk-type",
245 __DECONST(char *, dp->d_disktype));
248 /* Create serno alias */
249 if (dp->d_info.d_serialno) {
252 dp->d_info.d_serialno,
256 /* Create UUID alias */
257 if (!kuuid_is_nil(&part.storage_uuid)) {
258 snprintf_uuid(uuid_buf,
264 udev_dict_set_cstr(ndev, "uuid", uuid_buf);
266 ndev->si_flags |= SI_REPROBE_TEST;
270 } else if (info->d_dsflags & DSO_COMPATLABEL) {
272 if (sp->ds_size >= 0x100000000ULL)
273 ops = &disklabel64_ops;
275 ops = &disklabel32_ops;
276 sp->ds_label = ops->op_clone_label(info, sp);
278 if (sp->ds_type == DOSPTYP_386BSD || /* XXX */
279 sp->ds_type == DOSPTYP_NETBSD ||
280 sp->ds_type == DOSPTYP_OPENBSD) {
281 log(LOG_WARNING, "%s: cannot find label (%s)\n",
287 sp->ds_wlabel = FALSE;
290 return (msg ? EINVAL : 0);
294 * This routine is only called for newly minted drives or to reprobe
295 * a drive with no open slices. disk_probe_slice() is called directly
296 * when reprobing partition changes within slices.
299 disk_probe(struct disk *dp, int reprobe)
301 struct disk_info *info = &dp->d_info;
302 cdev_t dev = dp->d_cdev;
305 struct diskslices *osp;
306 struct diskslice *sp;
309 KKASSERT (info->d_media_blksize != 0);
312 dp->d_slice = dsmakeslicestruct(BASE_SLICE, info);
313 disk_debug(1, "disk_probe (begin): %s\n", dp->d_cdev->si_name);
315 error = mbrinit(dev, info, &(dp->d_slice));
321 for (i = 0; i < dp->d_slice->dss_nslices; i++) {
323 * Ignore the whole-disk slice, it has already been created.
325 if (i == WHOLE_DISK_SLICE)
330 * Ignore the compatibility slice s0 if it's a device mapper
333 if ((i == COMPATIBILITY_SLICE) &&
334 (info->d_dsflags & DSO_DEVICEMAPPER))
338 sp = &dp->d_slice->dss_slices[i];
341 * Handle s0. s0 is a compatibility slice if there are no
342 * other slices and it has not otherwise been set up, else
345 if (i == COMPATIBILITY_SLICE) {
347 if (sp->ds_type == 0 &&
348 dp->d_slice->dss_nslices == BASE_SLICE) {
349 sp->ds_size = info->d_media_blocks;
358 * Ignore 0-length slices
360 if (sp->ds_size == 0)
364 (ndev = devfs_find_device_by_name("%ss%d",
365 dev->si_name, sno))) {
367 * Device already exists and is still valid
369 ndev->si_flags |= SI_REPROBE_TEST;
372 * Destroy old UUID alias
374 destroy_dev_alias(ndev, "slice-by-uuid/*");
376 /* Create UUID alias */
377 if (!kuuid_is_nil(&sp->ds_stor_uuid)) {
378 snprintf_uuid(uuid_buf, sizeof(uuid_buf),
380 make_dev_alias(ndev, "slice-by-uuid/%s",
385 * Else create new device
387 ndev = make_dev_covering(&disk_ops, dp->d_rawdev->si_ops,
388 dkmakewholeslice(dkunit(dev), i),
389 UID_ROOT, GID_OPERATOR, 0640,
390 (info->d_dsflags & DSO_DEVICEMAPPER)?
391 "%s.s%d" : "%ss%d", dev->si_name, sno);
392 ndev->si_parent = dev;
393 udev_dict_set_cstr(ndev, "subsystem", "disk");
394 /* Inherit parent's disk type */
395 if (dp->d_disktype) {
396 udev_dict_set_cstr(ndev, "disk-type",
397 __DECONST(char *, dp->d_disktype));
400 /* Create serno alias */
401 if (dp->d_info.d_serialno) {
402 make_dev_alias(ndev, "serno/%s.s%d",
403 dp->d_info.d_serialno, sno);
406 /* Create UUID alias */
407 if (!kuuid_is_nil(&sp->ds_stor_uuid)) {
408 snprintf_uuid(uuid_buf, sizeof(uuid_buf),
410 make_dev_alias(ndev, "slice-by-uuid/%s",
415 ndev->si_flags |= SI_REPROBE_TEST;
420 * Probe appropriate slices for a disklabel
422 * XXX slice type 1 used by our gpt probe code.
423 * XXX slice type 0 used by mbr compat slice.
425 if (sp->ds_type == DOSPTYP_386BSD ||
426 sp->ds_type == DOSPTYP_NETBSD ||
427 sp->ds_type == DOSPTYP_OPENBSD ||
430 if (dp->d_slice->dss_first_bsd_slice == 0)
431 dp->d_slice->dss_first_bsd_slice = i;
432 disk_probe_slice(dp, ndev, i, reprobe);
436 disk_debug(1, "disk_probe (end): %s\n", dp->d_cdev->si_name);
441 disk_msg_core(void *arg)
444 struct diskslice *sp;
448 lwkt_gettoken(&disklist_token);
449 lwkt_initport_thread(&disk_msg_port, curthread);
450 wakeup(curthread); /* synchronous startup */
451 lwkt_reltoken(&disklist_token);
453 get_mplock(); /* not mpsafe yet? */
457 msg = (disk_msg_t)lwkt_waitport(&disk_msg_port, 0);
459 switch (msg->hdr.u.ms_result) {
460 case DISK_DISK_PROBE:
461 dp = (struct disk *)msg->load;
463 "DISK_DISK_PROBE: %s\n",
464 dp->d_cdev->si_name);
467 case DISK_DISK_DESTROY:
468 dp = (struct disk *)msg->load;
470 "DISK_DISK_DESTROY: %s\n",
471 dp->d_cdev->si_name);
472 devfs_destroy_related(dp->d_cdev);
473 destroy_dev(dp->d_cdev);
474 destroy_only_dev(dp->d_rawdev);
475 lwkt_gettoken(&disklist_token);
476 LIST_REMOVE(dp, d_list);
477 lwkt_reltoken(&disklist_token);
478 if (dp->d_info.d_serialno) {
479 kfree(dp->d_info.d_serialno, M_TEMP);
480 dp->d_info.d_serialno = NULL;
484 dp = (struct disk *)msg->load;
486 "DISK_DISK_UNPROBE: %s\n",
487 dp->d_cdev->si_name);
488 devfs_destroy_related(dp->d_cdev);
490 case DISK_SLICE_REPROBE:
491 dp = (struct disk *)msg->load;
492 sp = (struct diskslice *)msg->load2;
493 devfs_clr_related_flag(sp->ds_dev,
496 "DISK_SLICE_REPROBE: %s\n",
497 sp->ds_dev->si_name);
498 disk_probe_slice(dp, sp->ds_dev,
499 dkslice(sp->ds_dev), 1);
500 devfs_destroy_related_without_flag(
501 sp->ds_dev, SI_REPROBE_TEST);
503 case DISK_DISK_REPROBE:
504 dp = (struct disk *)msg->load;
505 devfs_clr_related_flag(dp->d_cdev, SI_REPROBE_TEST);
507 "DISK_DISK_REPROBE: %s\n",
508 dp->d_cdev->si_name);
510 devfs_destroy_related_without_flag(
511 dp->d_cdev, SI_REPROBE_TEST);
514 disk_debug(1, "DISK_SYNC\n");
517 devfs_debug(DEVFS_DEBUG_WARNING,
518 "disk_msg_core: unknown message "
519 "received at core\n");
522 lwkt_replymsg(&msg->hdr, 0);
529 * Acts as a message drain. Any message that is replied to here gets
530 * destroyed and the memory freed.
533 disk_msg_autofree_reply(lwkt_port_t port, lwkt_msg_t msg)
535 objcache_put(disk_msg_cache, msg);
540 disk_msg_send(uint32_t cmd, void *load, void *load2)
543 lwkt_port_t port = &disk_msg_port;
545 disk_msg = objcache_get(disk_msg_cache, M_WAITOK);
547 lwkt_initmsg(&disk_msg->hdr, &disk_dispose_port, 0);
549 disk_msg->hdr.u.ms_result = cmd;
550 disk_msg->load = load;
551 disk_msg->load2 = load2;
553 lwkt_sendmsg(port, &disk_msg->hdr);
557 disk_msg_send_sync(uint32_t cmd, void *load, void *load2)
559 struct lwkt_port rep_port;
563 disk_msg = objcache_get(disk_msg_cache, M_WAITOK);
564 port = &disk_msg_port;
566 /* XXX could probably use curthread's built-in msgport */
567 lwkt_initport_thread(&rep_port, curthread);
568 lwkt_initmsg(&disk_msg->hdr, &rep_port, 0);
570 disk_msg->hdr.u.ms_result = cmd;
571 disk_msg->load = load;
572 disk_msg->load2 = load2;
574 lwkt_sendmsg(port, &disk_msg->hdr);
575 lwkt_waitmsg(&disk_msg->hdr, 0);
576 objcache_put(disk_msg_cache, disk_msg);
580 * Create a raw device for the dev_ops template (which is returned). Also
581 * create a slice and unit managed disk and overload the user visible
582 * device space with it.
584 * NOTE: The returned raw device is NOT a slice and unit managed device.
585 * It is an actual raw device representing the raw disk as specified by
586 * the passed dev_ops. The disk layer not only returns such a raw device,
587 * it also uses it internally when passing (modified) commands through.
590 disk_create(int unit, struct disk *dp, struct dev_ops *raw_ops)
592 return _disk_create_named(NULL, unit, dp, raw_ops, 0);
596 disk_create_clone(int unit, struct disk *dp, struct dev_ops *raw_ops)
598 return _disk_create_named(NULL, unit, dp, raw_ops, 1);
602 disk_create_named(const char *name, int unit, struct disk *dp, struct dev_ops *raw_ops)
604 return _disk_create_named(name, unit, dp, raw_ops, 0);
608 disk_create_named_clone(const char *name, int unit, struct disk *dp, struct dev_ops *raw_ops)
610 return _disk_create_named(name, unit, dp, raw_ops, 1);
614 _disk_create_named(const char *name, int unit, struct disk *dp, struct dev_ops *raw_ops, int clone)
618 disk_debug(1, "disk_create (begin): %s%d\n", name, unit);
621 rawdev = make_only_dev(raw_ops, dkmakewholedisk(unit),
622 UID_ROOT, GID_OPERATOR, 0640, "%s", name);
624 rawdev = make_only_dev(raw_ops, dkmakewholedisk(unit),
625 UID_ROOT, GID_OPERATOR, 0640,
626 "%s%d", raw_ops->head.name, unit);
629 bzero(dp, sizeof(*dp));
631 dp->d_rawdev = rawdev;
632 dp->d_raw_ops = raw_ops;
633 dp->d_dev_ops = &disk_ops;
637 dp->d_cdev = make_only_dev_covering(&disk_ops, dp->d_rawdev->si_ops,
638 dkmakewholedisk(unit), UID_ROOT, GID_OPERATOR, 0640,
641 dp->d_cdev = make_dev_covering(&disk_ops, dp->d_rawdev->si_ops,
642 dkmakewholedisk(unit), UID_ROOT, GID_OPERATOR, 0640,
647 dp->d_cdev = make_only_dev_covering(&disk_ops, dp->d_rawdev->si_ops,
648 dkmakewholedisk(unit),
649 UID_ROOT, GID_OPERATOR, 0640,
650 "%s%d", raw_ops->head.name, unit);
652 dp->d_cdev = make_dev_covering(&disk_ops, dp->d_rawdev->si_ops,
653 dkmakewholedisk(unit),
654 UID_ROOT, GID_OPERATOR, 0640,
655 "%s%d", raw_ops->head.name, unit);
659 udev_dict_set_cstr(dp->d_cdev, "subsystem", "disk");
660 dp->d_cdev->si_disk = dp;
663 dsched_disk_create_callback(dp, name, unit);
665 dsched_disk_create_callback(dp, raw_ops->head.name, unit);
667 lwkt_gettoken(&disklist_token);
668 LIST_INSERT_HEAD(&disklist, dp, d_list);
669 lwkt_reltoken(&disklist_token);
671 disk_debug(1, "disk_create (end): %s%d\n",
672 (name != NULL)?(name):(raw_ops->head.name), unit);
674 return (dp->d_rawdev);
678 disk_setdisktype(struct disk *disk, const char *type)
680 KKASSERT(disk != NULL);
682 disk->d_disktype = type;
683 return udev_dict_set_cstr(disk->d_cdev, "disk-type", __DECONST(char *, type));
687 disk_getopencount(struct disk *disk)
689 return disk->d_opencount;
693 _setdiskinfo(struct disk *disk, struct disk_info *info)
697 oldserialno = disk->d_info.d_serialno;
698 bcopy(info, &disk->d_info, sizeof(disk->d_info));
699 info = &disk->d_info;
702 "_setdiskinfo: %s\n",
703 disk->d_cdev->si_name);
706 * The serial number is duplicated so the caller can throw
709 if (info->d_serialno && info->d_serialno[0] &&
710 (info->d_serialno[0] != ' ' || strlen(info->d_serialno) > 1)) {
711 info->d_serialno = kstrdup(info->d_serialno, M_TEMP);
712 disk_cleanserial(info->d_serialno);
714 make_dev_alias(disk->d_cdev, "serno/%s",
718 info->d_serialno = NULL;
721 kfree(oldserialno, M_TEMP);
723 dsched_disk_update_callback(disk, info);
726 * The caller may set d_media_size or d_media_blocks and we
727 * calculate the other.
729 KKASSERT(info->d_media_size == 0 || info->d_media_blocks == 0);
730 if (info->d_media_size == 0 && info->d_media_blocks) {
731 info->d_media_size = (u_int64_t)info->d_media_blocks *
732 info->d_media_blksize;
733 } else if (info->d_media_size && info->d_media_blocks == 0 &&
734 info->d_media_blksize) {
735 info->d_media_blocks = info->d_media_size /
736 info->d_media_blksize;
740 * The si_* fields for rawdev are not set until after the
741 * disk_create() call, so someone using the cooked version
742 * of the raw device (i.e. da0s0) will not get the right
743 * si_iosize_max unless we fix it up here.
745 if (disk->d_cdev && disk->d_rawdev &&
746 disk->d_cdev->si_iosize_max == 0) {
747 disk->d_cdev->si_iosize_max = disk->d_rawdev->si_iosize_max;
748 disk->d_cdev->si_bsize_phys = disk->d_rawdev->si_bsize_phys;
749 disk->d_cdev->si_bsize_best = disk->d_rawdev->si_bsize_best;
752 /* Add the serial number to the udev_dictionary */
753 if (info->d_serialno)
754 udev_dict_set_cstr(disk->d_cdev, "serno", info->d_serialno);
758 * Disk drivers must call this routine when media parameters are available
762 disk_setdiskinfo(struct disk *disk, struct disk_info *info)
764 _setdiskinfo(disk, info);
765 disk_msg_send(DISK_DISK_PROBE, disk, NULL);
767 "disk_setdiskinfo: sent probe for %s\n",
768 disk->d_cdev->si_name);
772 disk_setdiskinfo_sync(struct disk *disk, struct disk_info *info)
774 _setdiskinfo(disk, info);
775 disk_msg_send_sync(DISK_DISK_PROBE, disk, NULL);
777 "disk_setdiskinfo_sync: sent probe for %s\n",
778 disk->d_cdev->si_name);
782 * This routine is called when an adapter detaches. The higher level
783 * managed disk device is destroyed while the lower level raw device is
787 disk_destroy(struct disk *disk)
789 dsched_disk_destroy_callback(disk);
790 disk_msg_send_sync(DISK_DISK_DESTROY, disk, NULL);
795 disk_dumpcheck(cdev_t dev, u_int64_t *size, u_int64_t *blkno, u_int32_t *secsize)
797 struct partinfo pinfo;
800 bzero(&pinfo, sizeof(pinfo));
801 error = dev_dioctl(dev, DIOCGPART, (void *)&pinfo, 0,
802 proc0.p_ucred, NULL);
806 if (pinfo.media_blksize == 0)
809 if (blkno) /* XXX: make sure this reserved stuff is right */
810 *blkno = pinfo.reserved_blocks +
811 pinfo.media_offset / pinfo.media_blksize;
813 *secsize = pinfo.media_blksize;
815 *size = (pinfo.media_blocks - pinfo.reserved_blocks);
821 disk_dumpconf(cdev_t dev, u_int onoff)
823 struct dumperinfo di;
824 u_int64_t size, blkno;
829 return set_dumper(NULL);
831 error = disk_dumpcheck(dev, &size, &blkno, &secsize);
836 bzero(&di, sizeof(struct dumperinfo));
837 di.dumper = diskdump;
839 di.blocksize = secsize;
840 di.mediaoffset = blkno * DEV_BSIZE;
841 di.mediasize = size * DEV_BSIZE;
843 return set_dumper(&di);
847 disk_unprobe(struct disk *disk)
852 disk_msg_send_sync(DISK_UNPROBE, disk, NULL);
856 disk_invalidate (struct disk *disk)
858 dsgone(&disk->d_slice);
862 disk_enumerate(struct disk *disk)
866 lwkt_gettoken(&disklist_token);
868 dp = (LIST_FIRST(&disklist));
870 dp = (LIST_NEXT(disk, d_list));
871 lwkt_reltoken(&disklist_token);
878 sysctl_disks(SYSCTL_HANDLER_ARGS)
886 while ((disk = disk_enumerate(disk))) {
888 error = SYSCTL_OUT(req, " ", 1);
894 error = SYSCTL_OUT(req, disk->d_rawdev->si_name,
895 strlen(disk->d_rawdev->si_name));
899 error = SYSCTL_OUT(req, "", 1);
903 SYSCTL_PROC(_kern, OID_AUTO, disks, CTLTYPE_STRING | CTLFLAG_RD, NULL, 0,
904 sysctl_disks, "A", "names of available disks");
907 * Open a disk device or partition.
911 diskopen(struct dev_open_args *ap)
913 cdev_t dev = ap->a_head.a_dev;
918 * dp can't be NULL here XXX.
920 * d_slice will be NULL if setdiskinfo() has not been called yet.
921 * setdiskinfo() is typically called whether the disk is present
922 * or not (e.g. CD), but the base disk device is created first
923 * and there may be a race.
926 if (dp == NULL || dp->d_slice == NULL)
931 * Deal with open races
934 while (dp->d_flags & DISKFLAG_LOCK) {
935 dp->d_flags |= DISKFLAG_WANTED;
936 error = tsleep(dp, PCATCH, "diskopen", hz);
942 dp->d_flags |= DISKFLAG_LOCK;
945 * Open the underlying raw device.
947 if (!dsisopen(dp->d_slice)) {
949 if (!pdev->si_iosize_max)
950 pdev->si_iosize_max = dev->si_iosize_max;
952 error = dev_dopen(dp->d_rawdev, ap->a_oflags,
953 ap->a_devtype, ap->a_cred);
958 error = dsopen(dev, ap->a_devtype, dp->d_info.d_dsflags,
959 &dp->d_slice, &dp->d_info);
960 if (!dsisopen(dp->d_slice)) {
961 dev_dclose(dp->d_rawdev, ap->a_oflags, ap->a_devtype);
964 dp->d_flags &= ~DISKFLAG_LOCK;
965 if (dp->d_flags & DISKFLAG_WANTED) {
966 dp->d_flags &= ~DISKFLAG_WANTED;
971 KKASSERT(dp->d_opencount >= 0);
972 /* If the open was successful, bump open count */
974 atomic_add_int(&dp->d_opencount, 1);
980 * Close a disk device or partition
984 diskclose(struct dev_close_args *ap)
986 cdev_t dev = ap->a_head.a_dev;
995 * The cdev_t represents the disk/slice/part. The shared
996 * dp structure governs all cdevs associated with the disk.
998 * As a safety only close the underlying raw device on the last
999 * close the disk device if our tracking of the slices/partitions
1000 * also indicates nothing is open.
1002 KKASSERT(dp->d_opencount >= 1);
1003 lcount = atomic_fetchadd_int(&dp->d_opencount, -1);
1006 dsclose(dev, ap->a_devtype, dp->d_slice);
1007 if (lcount <= 1 && !dsisopen(dp->d_slice)) {
1008 error = dev_dclose(dp->d_rawdev, ap->a_fflag, ap->a_devtype);
1015 * First execute the ioctl on the disk device, and if it isn't supported
1016 * try running it on the backing device.
1020 diskioctl(struct dev_ioctl_args *ap)
1022 cdev_t dev = ap->a_head.a_dev;
1031 devfs_debug(DEVFS_DEBUG_DEBUG,
1032 "diskioctl: cmd is: %lx (name: %s)\n",
1033 ap->a_cmd, dev->si_name);
1034 devfs_debug(DEVFS_DEBUG_DEBUG,
1035 "diskioctl: &dp->d_slice is: %p, %p\n",
1036 &dp->d_slice, dp->d_slice);
1038 if (ap->a_cmd == DIOCGKERNELDUMP) {
1039 u = *(u_int *)ap->a_data;
1040 return disk_dumpconf(dev, u);
1043 if (&dp->d_slice == NULL || dp->d_slice == NULL ||
1044 ((dp->d_info.d_dsflags & DSO_DEVICEMAPPER) &&
1045 dkslice(dev) == WHOLE_DISK_SLICE)) {
1049 error = dsioctl(dev, ap->a_cmd, ap->a_data, ap->a_fflag,
1050 &dp->d_slice, &dp->d_info);
1054 if (error == ENOIOCTL) {
1055 error = dev_dioctl(dp->d_rawdev, ap->a_cmd, ap->a_data,
1056 ap->a_fflag, ap->a_cred, NULL);
1062 * Execute strategy routine
1066 diskstrategy(struct dev_strategy_args *ap)
1068 cdev_t dev = ap->a_head.a_dev;
1069 struct bio *bio = ap->a_bio;
1076 bio->bio_buf->b_error = ENXIO;
1077 bio->bio_buf->b_flags |= B_ERROR;
1081 KKASSERT(dev->si_disk == dp);
1084 * The dscheck() function will also transform the slice relative
1085 * block number i.e. bio->bio_offset into a block number that can be
1086 * passed directly to the underlying raw device. If dscheck()
1087 * returns NULL it will have handled the bio for us (e.g. EOF
1088 * or error due to being beyond the device size).
1090 if ((nbio = dscheck(dev, bio, dp->d_slice)) != NULL) {
1091 dsched_queue(dp, nbio);
1099 * Return the partition size in ?blocks?
1103 diskpsize(struct dev_psize_args *ap)
1105 cdev_t dev = ap->a_head.a_dev;
1112 ap->a_result = dssize(dev, &dp->d_slice);
1114 if ((ap->a_result == -1) &&
1115 (dp->d_info.d_dsflags & DSO_DEVICEMAPPER)) {
1116 ap->a_head.a_dev = dp->d_rawdev;
1117 return dev_doperate(&ap->a_head);
1123 diskdump(struct dev_dump_args *ap)
1125 cdev_t dev = ap->a_head.a_dev;
1126 struct disk *dp = dev->si_disk;
1127 u_int64_t size, offset;
1130 error = disk_dumpcheck(dev, &size, &ap->a_blkno, &ap->a_secsize);
1131 /* XXX: this should probably go in disk_dumpcheck somehow */
1132 if (ap->a_length != 0) {
1134 offset = ap->a_blkno * DEV_BSIZE;
1135 if ((ap->a_offset < offset) ||
1136 (ap->a_offset + ap->a_length - offset > size)) {
1137 kprintf("Attempt to write outside dump device boundaries.\n");
1143 ap->a_head.a_dev = dp->d_rawdev;
1144 error = dev_doperate(&ap->a_head);
1151 SYSCTL_INT(_debug_sizeof, OID_AUTO, diskslices, CTLFLAG_RD,
1152 0, sizeof(struct diskslices), "sizeof(struct diskslices)");
1154 SYSCTL_INT(_debug_sizeof, OID_AUTO, disk, CTLFLAG_RD,
1155 0, sizeof(struct disk), "sizeof(struct disk)");
1158 * Reorder interval for burst write allowance and minor write
1161 * We always want to trickle some writes in to make use of the
1162 * disk's zone cache. Bursting occurs on a longer interval and only
1163 * runningbufspace is well over the hirunningspace limit.
1165 int bioq_reorder_burst_interval = 60; /* should be multiple of minor */
1166 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_burst_interval,
1167 CTLFLAG_RW, &bioq_reorder_burst_interval, 0, "");
1168 int bioq_reorder_minor_interval = 5;
1169 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_minor_interval,
1170 CTLFLAG_RW, &bioq_reorder_minor_interval, 0, "");
1172 int bioq_reorder_burst_bytes = 3000000;
1173 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_burst_bytes,
1174 CTLFLAG_RW, &bioq_reorder_burst_bytes, 0, "");
1175 int bioq_reorder_minor_bytes = 262144;
1176 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_minor_bytes,
1177 CTLFLAG_RW, &bioq_reorder_minor_bytes, 0, "");
1181 * Order I/Os. Generally speaking this code is designed to make better
1182 * use of drive zone caches. A drive zone cache can typically track linear
1183 * reads or writes for around 16 zones simultaniously.
1185 * Read prioritization issues: It is possible for hundreds of megabytes worth
1186 * of writes to be queued asynchronously. This creates a huge bottleneck
1187 * for reads which reduce read bandwidth to a trickle.
1189 * To solve this problem we generally reorder reads before writes.
1191 * However, a large number of random reads can also starve writes and
1192 * make poor use of the drive zone cache so we allow writes to trickle
1196 bioqdisksort(struct bio_queue_head *bioq, struct bio *bio)
1199 * The BIO wants to be ordered. Adding to the tail also
1200 * causes transition to be set to NULL, forcing the ordering
1201 * of all prior I/O's.
1203 if (bio->bio_buf->b_flags & B_ORDERED) {
1204 bioq_insert_tail(bioq, bio);
1208 switch(bio->bio_buf->b_cmd) {
1210 if (bioq->transition) {
1212 * Insert before the first write. Bleedover writes
1213 * based on reorder intervals to prevent starvation.
1215 TAILQ_INSERT_BEFORE(bioq->transition, bio, bio_act);
1217 if (bioq->reorder % bioq_reorder_minor_interval == 0) {
1218 bioqwritereorder(bioq);
1219 if (bioq->reorder >=
1220 bioq_reorder_burst_interval) {
1226 * No writes queued (or ordering was forced),
1229 TAILQ_INSERT_TAIL(&bioq->queue, bio, bio_act);
1234 * Writes are always appended. If no writes were previously
1235 * queued or an ordered tail insertion occured the transition
1236 * field will be NULL.
1238 TAILQ_INSERT_TAIL(&bioq->queue, bio, bio_act);
1239 if (bioq->transition == NULL)
1240 bioq->transition = bio;
1244 * All other request types are forced to be ordered.
1246 bioq_insert_tail(bioq, bio);
1252 * Move the read-write transition point to prevent reads from
1253 * completely starving our writes. This brings a number of writes into
1254 * the fold every N reads.
1256 * We bring a few linear writes into the fold on a minor interval
1257 * and we bring a non-linear burst of writes into the fold on a major
1258 * interval. Bursting only occurs if runningbufspace is really high
1259 * (typically from syncs, fsyncs, or HAMMER flushes).
1263 bioqwritereorder(struct bio_queue_head *bioq)
1271 if (bioq->reorder < bioq_reorder_burst_interval ||
1272 !buf_runningbufspace_severe()) {
1273 left = (size_t)bioq_reorder_minor_bytes;
1276 left = (size_t)bioq_reorder_burst_bytes;
1280 next_offset = bioq->transition->bio_offset;
1281 while ((bio = bioq->transition) != NULL &&
1282 (check_off == 0 || next_offset == bio->bio_offset)
1284 n = bio->bio_buf->b_bcount;
1285 next_offset = bio->bio_offset + n;
1286 bioq->transition = TAILQ_NEXT(bio, bio_act);
1294 * Bounds checking against the media size, used for the raw partition.
1295 * secsize, mediasize and b_blkno must all be the same units.
1296 * Possibly this has to be DEV_BSIZE (512).
1299 bounds_check_with_mediasize(struct bio *bio, int secsize, uint64_t mediasize)
1301 struct buf *bp = bio->bio_buf;
1304 sz = howmany(bp->b_bcount, secsize);
1306 if (bio->bio_offset/DEV_BSIZE + sz > mediasize) {
1307 sz = mediasize - bio->bio_offset/DEV_BSIZE;
1309 /* If exactly at end of disk, return EOF. */
1310 bp->b_resid = bp->b_bcount;
1314 /* If past end of disk, return EINVAL. */
1315 bp->b_error = EINVAL;
1318 /* Otherwise, truncate request. */
1319 bp->b_bcount = sz * secsize;
1326 * Disk error is the preface to plaintive error messages
1327 * about failing disk transfers. It prints messages of the form
1329 hp0g: hard error reading fsbn 12345 of 12344-12347 (hp0 bn %d cn %d tn %d sn %d)
1331 * if the offset of the error in the transfer and a disk label
1332 * are both available. blkdone should be -1 if the position of the error
1333 * is unknown; the disklabel pointer may be null from drivers that have not
1334 * been converted to use them. The message is printed with kprintf
1335 * if pri is LOG_PRINTF, otherwise it uses log at the specified priority.
1336 * The message should be completed (with at least a newline) with kprintf
1337 * or log(-1, ...), respectively. There is no trailing space.
1340 diskerr(struct bio *bio, cdev_t dev, const char *what, int pri, int donecnt)
1342 struct buf *bp = bio->bio_buf;
1356 kprintf("%s: %s %sing ", dev->si_name, what, term);
1357 kprintf("offset %012llx for %d",
1358 (long long)bio->bio_offset,
1362 kprintf(" (%d bytes completed)", donecnt);
1366 * Locate a disk device
1369 disk_locate(const char *devname)
1371 return devfs_find_device_by_name(devname);
1375 disk_config(void *arg)
1377 disk_msg_send_sync(DISK_SYNC, NULL, NULL);
1383 struct thread* td_core;
1385 disk_msg_cache = objcache_create("disk-msg-cache", 0, 0,
1387 objcache_malloc_alloc,
1388 objcache_malloc_free,
1389 &disk_msg_malloc_args);
1391 lwkt_token_init(&disklist_token, "disks");
1394 * Initialize the reply-only port which acts as a message drain
1396 lwkt_initport_replyonly(&disk_dispose_port, disk_msg_autofree_reply);
1398 lwkt_gettoken(&disklist_token);
1399 lwkt_create(disk_msg_core, /*args*/NULL, &td_core, NULL,
1400 0, -1, "disk_msg_core");
1401 tsleep(td_core, 0, "diskcore", 0);
1402 lwkt_reltoken(&disklist_token);
1408 objcache_destroy(disk_msg_cache);
1412 * Clean out illegal characters in serial numbers.
1415 disk_cleanserial(char *serno)
1419 while ((c = *serno) != 0) {
1420 if (c >= 'a' && c <= 'z')
1422 else if (c >= 'A' && c <= 'Z')
1424 else if (c >= '0' && c <= '9')
1426 else if (c == '-' || c == '@' || c == '+' || c == '.')
1434 TUNABLE_INT("kern.disk_debug", &disk_debug_enable);
1435 SYSCTL_INT(_kern, OID_AUTO, disk_debug, CTLFLAG_RW, &disk_debug_enable,
1436 0, "Enable subr_disk debugging");
1438 SYSINIT(disk_register, SI_SUB_PRE_DRIVERS, SI_ORDER_FIRST, disk_init, NULL);
1439 SYSUNINIT(disk_register, SI_SUB_PRE_DRIVERS, SI_ORDER_ANY, disk_uninit, NULL);