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>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * ----------------------------------------------------------------------------
35 * "THE BEER-WARE LICENSE" (Revision 42):
36 * <phk@FreeBSD.ORG> wrote this file. As long as you retain this notice you
37 * can do whatever you want with this stuff. If we meet some day, and you think
38 * this stuff is worth it, you can buy me a beer in return. Poul-Henning Kamp
39 * ----------------------------------------------------------------------------
41 * Copyright (c) 1982, 1986, 1988, 1993
42 * The Regents of the University of California. All rights reserved.
43 * (c) UNIX System Laboratories, Inc.
44 * All or some portions of this file are derived from material licensed
45 * to the University of California by American Telephone and Telegraph
46 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
47 * the permission of UNIX System Laboratories, Inc.
49 * Redistribution and use in source and binary forms, with or without
50 * modification, are permitted provided that the following conditions
52 * 1. Redistributions of source code must retain the above copyright
53 * notice, this list of conditions and the following disclaimer.
54 * 2. Redistributions in binary form must reproduce the above copyright
55 * notice, this list of conditions and the following disclaimer in the
56 * documentation and/or other materials provided with the distribution.
57 * 3. All advertising materials mentioning features or use of this software
58 * must display the following acknowledgement:
59 * This product includes software developed by the University of
60 * California, Berkeley and its contributors.
61 * 4. Neither the name of the University nor the names of its contributors
62 * may be used to endorse or promote products derived from this software
63 * without specific prior written permission.
65 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
66 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
67 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
68 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
69 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
70 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
71 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
72 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
73 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
74 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
77 * @(#)ufs_disksubr.c 8.5 (Berkeley) 1/21/94
78 * $FreeBSD: src/sys/kern/subr_disk.c,v 1.20.2.6 2001/10/05 07:14:57 peter Exp $
79 * $FreeBSD: src/sys/ufs/ufs/ufs_disksubr.c,v 1.44.2.3 2001/03/05 05:42:19 obrien Exp $
80 * $DragonFly: src/sys/kern/subr_disk.c,v 1.40 2008/06/05 18:06:32 swildner Exp $
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 <sys/sysctl.h>
99 #include <machine/md_var.h>
100 #include <sys/ctype.h>
101 #include <sys/syslog.h>
102 #include <sys/device.h>
103 #include <sys/msgport.h>
104 #include <sys/msgport2.h>
105 #include <sys/buf2.h>
106 #include <sys/devfs.h>
107 #include <sys/thread.h>
108 #include <sys/thread2.h>
109 #include <sys/dsched.h>
110 #include <sys/queue.h>
111 #include <sys/lock.h>
112 #include <sys/udev.h>
114 static MALLOC_DEFINE(M_DISK, "disk", "disk data");
115 static int disk_debug_enable = 0;
117 static void disk_msg_autofree_reply(lwkt_port_t, lwkt_msg_t);
118 static void disk_msg_core(void *);
119 static int disk_probe_slice(struct disk *dp, cdev_t dev, int slice, int reprobe);
120 static void disk_probe(struct disk *dp, int reprobe);
121 static void _setdiskinfo(struct disk *disk, struct disk_info *info);
122 static void bioqwritereorder(struct bio_queue_head *bioq);
123 static void disk_cleanserial(char *serno);
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_clone_t diskclone;
131 static d_dump_t diskdump;
133 static LIST_HEAD(, disk) disklist = LIST_HEAD_INITIALIZER(&disklist);
134 static struct lwkt_token disklist_token;
136 static struct dev_ops disk_ops = {
137 { "disk", 0, D_DISK },
139 .d_close = diskclose,
141 .d_write = physwrite,
142 .d_ioctl = diskioctl,
143 .d_strategy = diskstrategy,
145 .d_psize = diskpsize,
149 static struct objcache *disk_msg_cache;
151 struct objcache_malloc_args disk_msg_malloc_args = {
152 sizeof(struct disk_msg), M_DISK };
154 static struct lwkt_port disk_dispose_port;
155 static struct lwkt_port disk_msg_port;
158 disk_debug(int level, char *fmt, ...)
163 if (level <= disk_debug_enable)
171 disk_probe_slice(struct disk *dp, cdev_t dev, int slice, int reprobe)
173 struct disk_info *info = &dp->d_info;
174 struct diskslice *sp = &dp->d_slice->dss_slices[slice];
176 struct partinfo part;
183 "disk_probe_slice (begin): %s (%s)\n",
184 dev->si_name, dp->d_cdev->si_name);
186 sno = slice ? slice - 1 : 0;
188 ops = &disklabel32_ops;
189 msg = ops->op_readdisklabel(dev, sp, &sp->ds_label, info);
190 if (msg && !strcmp(msg, "no disk label")) {
191 ops = &disklabel64_ops;
192 msg = ops->op_readdisklabel(dev, sp, &sp->ds_label, info);
195 if (slice != WHOLE_DISK_SLICE)
196 ops->op_adjust_label_reserved(dp->d_slice, slice, sp);
201 for (i = 0; i < ops->op_getnumparts(sp->ds_label); i++) {
202 ops->op_loadpartinfo(sp->ds_label, i, &part);
205 (ndev = devfs_find_device_by_name("%s%c",
206 dev->si_name, 'a' + i))
209 * Device already exists and
212 ndev->si_flags |= SI_REPROBE_TEST;
214 ndev = make_dev_covering(&disk_ops, dp->d_rawdev->si_ops,
215 dkmakeminor(dkunit(dp->d_cdev),
217 UID_ROOT, GID_OPERATOR, 0640,
218 "%s%c", dev->si_name, 'a'+ i);
220 udev_dict_set_cstr(ndev, "subsystem", "disk");
221 /* Inherit parent's disk type */
222 if (dp->d_disktype) {
223 udev_dict_set_cstr(ndev, "disk-type",
224 __DECONST(char *, dp->d_disktype));
226 if (dp->d_info.d_serialno) {
229 dp->d_info.d_serialno,
232 ndev->si_flags |= SI_REPROBE_TEST;
236 } else if (info->d_dsflags & DSO_COMPATLABEL) {
238 if (sp->ds_size >= 0x100000000ULL)
239 ops = &disklabel64_ops;
241 ops = &disklabel32_ops;
242 sp->ds_label = ops->op_clone_label(info, sp);
244 if (sp->ds_type == DOSPTYP_386BSD || /* XXX */
245 sp->ds_type == DOSPTYP_NETBSD ||
246 sp->ds_type == DOSPTYP_OPENBSD) {
247 log(LOG_WARNING, "%s: cannot find label (%s)\n",
253 sp->ds_wlabel = FALSE;
256 return (msg ? EINVAL : 0);
260 * This routine is only called for newly minted drives or to reprobe
261 * a drive with no open slices. disk_probe_slice() is called directly
262 * when reprobing partition changes within slices.
265 disk_probe(struct disk *dp, int reprobe)
267 struct disk_info *info = &dp->d_info;
268 cdev_t dev = dp->d_cdev;
271 struct diskslices *osp;
272 struct diskslice *sp;
274 KKASSERT (info->d_media_blksize != 0);
277 dp->d_slice = dsmakeslicestruct(BASE_SLICE, info);
278 disk_debug(1, "disk_probe (begin): %s\n", dp->d_cdev->si_name);
280 error = mbrinit(dev, info, &(dp->d_slice));
286 for (i = 0; i < dp->d_slice->dss_nslices; i++) {
288 * Ignore the whole-disk slice, it has already been created.
290 if (i == WHOLE_DISK_SLICE)
292 sp = &dp->d_slice->dss_slices[i];
295 * Handle s0. s0 is a compatibility slice if there are no
296 * other slices and it has not otherwise been set up, else
299 if (i == COMPATIBILITY_SLICE) {
301 if (sp->ds_type == 0 &&
302 dp->d_slice->dss_nslices == BASE_SLICE) {
303 sp->ds_size = info->d_media_blocks;
312 * Ignore 0-length slices
314 if (sp->ds_size == 0)
318 (ndev = devfs_find_device_by_name("%ss%d",
319 dev->si_name, sno))) {
321 * Device already exists and is still valid
323 ndev->si_flags |= SI_REPROBE_TEST;
326 * Else create new device
328 ndev = make_dev_covering(&disk_ops, dp->d_rawdev->si_ops,
329 dkmakewholeslice(dkunit(dev), i),
330 UID_ROOT, GID_OPERATOR, 0640,
331 "%ss%d", dev->si_name, sno);
332 udev_dict_set_cstr(ndev, "subsystem", "disk");
333 /* Inherit parent's disk type */
334 if (dp->d_disktype) {
335 udev_dict_set_cstr(ndev, "disk-type",
336 __DECONST(char *, dp->d_disktype));
338 if (dp->d_info.d_serialno) {
339 make_dev_alias(ndev, "serno/%s.s%d",
340 dp->d_info.d_serialno, sno);
343 ndev->si_flags |= SI_REPROBE_TEST;
348 * Probe appropriate slices for a disklabel
350 * XXX slice type 1 used by our gpt probe code.
351 * XXX slice type 0 used by mbr compat slice.
353 if (sp->ds_type == DOSPTYP_386BSD ||
354 sp->ds_type == DOSPTYP_NETBSD ||
355 sp->ds_type == DOSPTYP_OPENBSD ||
358 if (dp->d_slice->dss_first_bsd_slice == 0)
359 dp->d_slice->dss_first_bsd_slice = i;
360 disk_probe_slice(dp, ndev, i, reprobe);
364 disk_debug(1, "disk_probe (end): %s\n", dp->d_cdev->si_name);
369 disk_msg_core(void *arg)
372 struct diskslice *sp;
376 lwkt_initport_thread(&disk_msg_port, curthread);
381 msg = (disk_msg_t)lwkt_waitport(&disk_msg_port, 0);
383 switch (msg->hdr.u.ms_result) {
384 case DISK_DISK_PROBE:
385 dp = (struct disk *)msg->load;
387 "DISK_DISK_PROBE: %s\n",
388 dp->d_cdev->si_name);
391 case DISK_DISK_DESTROY:
392 dp = (struct disk *)msg->load;
394 "DISK_DISK_DESTROY: %s\n",
395 dp->d_cdev->si_name);
396 devfs_destroy_subnames(dp->d_cdev->si_name);
397 devfs_destroy_dev(dp->d_cdev);
398 lwkt_gettoken(&disklist_token);
399 LIST_REMOVE(dp, d_list);
400 lwkt_reltoken(&disklist_token);
401 if (dp->d_info.d_serialno) {
402 kfree(dp->d_info.d_serialno, M_TEMP);
403 dp->d_info.d_serialno = NULL;
407 dp = (struct disk *)msg->load;
409 "DISK_DISK_UNPROBE: %s\n",
410 dp->d_cdev->si_name);
411 devfs_destroy_subnames(dp->d_cdev->si_name);
413 case DISK_SLICE_REPROBE:
414 dp = (struct disk *)msg->load;
415 sp = (struct diskslice *)msg->load2;
416 devfs_clr_subnames_flag(sp->ds_dev->si_name,
419 "DISK_SLICE_REPROBE: %s\n",
420 sp->ds_dev->si_name);
421 disk_probe_slice(dp, sp->ds_dev,
422 dkslice(sp->ds_dev), 1);
423 devfs_destroy_subnames_without_flag(
424 sp->ds_dev->si_name, SI_REPROBE_TEST);
426 case DISK_DISK_REPROBE:
427 dp = (struct disk *)msg->load;
428 devfs_clr_subnames_flag(dp->d_cdev->si_name, SI_REPROBE_TEST);
430 "DISK_DISK_REPROBE: %s\n",
431 dp->d_cdev->si_name);
433 devfs_destroy_subnames_without_flag(
434 dp->d_cdev->si_name, SI_REPROBE_TEST);
437 disk_debug(1, "DISK_SYNC\n");
440 devfs_debug(DEVFS_DEBUG_WARNING,
441 "disk_msg_core: unknown message "
442 "received at core\n");
445 lwkt_replymsg(&msg->hdr, 0);
452 * Acts as a message drain. Any message that is replied to here gets
453 * destroyed and the memory freed.
456 disk_msg_autofree_reply(lwkt_port_t port, lwkt_msg_t msg)
458 objcache_put(disk_msg_cache, msg);
463 disk_msg_send(uint32_t cmd, void *load, void *load2)
466 lwkt_port_t port = &disk_msg_port;
468 disk_msg = objcache_get(disk_msg_cache, M_WAITOK);
470 lwkt_initmsg(&disk_msg->hdr, &disk_dispose_port, 0);
472 disk_msg->hdr.u.ms_result = cmd;
473 disk_msg->load = load;
474 disk_msg->load2 = load2;
476 lwkt_sendmsg(port, &disk_msg->hdr);
480 disk_msg_send_sync(uint32_t cmd, void *load, void *load2)
482 struct lwkt_port rep_port;
486 disk_msg = objcache_get(disk_msg_cache, M_WAITOK);
487 port = &disk_msg_port;
489 /* XXX could probably use curthread's built-in msgport */
490 lwkt_initport_thread(&rep_port, curthread);
491 lwkt_initmsg(&disk_msg->hdr, &rep_port, 0);
493 disk_msg->hdr.u.ms_result = cmd;
494 disk_msg->load = load;
495 disk_msg->load2 = load2;
497 lwkt_sendmsg(port, &disk_msg->hdr);
498 lwkt_waitmsg(&disk_msg->hdr, 0);
499 objcache_put(disk_msg_cache, disk_msg);
503 * Create a raw device for the dev_ops template (which is returned). Also
504 * create a slice and unit managed disk and overload the user visible
505 * device space with it.
507 * NOTE: The returned raw device is NOT a slice and unit managed device.
508 * It is an actual raw device representing the raw disk as specified by
509 * the passed dev_ops. The disk layer not only returns such a raw device,
510 * it also uses it internally when passing (modified) commands through.
513 disk_create(int unit, struct disk *dp, struct dev_ops *raw_ops)
515 return disk_create_named(NULL, unit, dp, raw_ops);
519 disk_create_named(const char *name, int unit, struct disk *dp, struct dev_ops *raw_ops)
524 name = raw_ops->head.name;
526 disk_debug(1, "disk_create (begin): %s%d\n", name, unit);
528 rawdev = make_only_dev(raw_ops, dkmakewholedisk(unit),
529 UID_ROOT, GID_OPERATOR, 0640,
532 bzero(dp, sizeof(*dp));
534 dp->d_rawdev = rawdev;
535 dp->d_raw_ops = raw_ops;
536 dp->d_dev_ops = &disk_ops;
537 dp->d_cdev = make_dev_covering(&disk_ops, dp->d_rawdev->si_ops,
538 dkmakewholedisk(unit),
539 UID_ROOT, GID_OPERATOR, 0640,
541 udev_dict_set_cstr(dp->d_cdev, "subsystem", "disk");
542 dp->d_cdev->si_disk = dp;
544 dsched_disk_create_callback(dp, name, unit);
546 lwkt_gettoken(&disklist_token);
547 LIST_INSERT_HEAD(&disklist, dp, d_list);
548 lwkt_reltoken(&disklist_token);
550 disk_debug(1, "disk_create (end): %s%d\n", name, unit);
552 return (dp->d_rawdev);
556 disk_setdisktype(struct disk *disk, const char *type)
558 KKASSERT(disk != NULL);
560 disk->d_disktype = type;
561 return udev_dict_set_cstr(disk->d_cdev, "disk-type", __DECONST(char *, type));
565 _setdiskinfo(struct disk *disk, struct disk_info *info)
569 oldserialno = disk->d_info.d_serialno;
570 bcopy(info, &disk->d_info, sizeof(disk->d_info));
571 info = &disk->d_info;
574 "_setdiskinfo: %s\n",
575 disk->d_cdev->si_name);
578 * The serial number is duplicated so the caller can throw
581 if (info->d_serialno && info->d_serialno[0]) {
582 info->d_serialno = kstrdup(info->d_serialno, M_TEMP);
583 disk_cleanserial(info->d_serialno);
585 make_dev_alias(disk->d_cdev, "serno/%s",
589 info->d_serialno = NULL;
592 kfree(oldserialno, M_TEMP);
594 dsched_disk_update_callback(disk, info);
597 * The caller may set d_media_size or d_media_blocks and we
598 * calculate the other.
600 KKASSERT(info->d_media_size == 0 || info->d_media_blksize == 0);
601 if (info->d_media_size == 0 && info->d_media_blocks) {
602 info->d_media_size = (u_int64_t)info->d_media_blocks *
603 info->d_media_blksize;
604 } else if (info->d_media_size && info->d_media_blocks == 0 &&
605 info->d_media_blksize) {
606 info->d_media_blocks = info->d_media_size /
607 info->d_media_blksize;
611 * The si_* fields for rawdev are not set until after the
612 * disk_create() call, so someone using the cooked version
613 * of the raw device (i.e. da0s0) will not get the right
614 * si_iosize_max unless we fix it up here.
616 if (disk->d_cdev && disk->d_rawdev &&
617 disk->d_cdev->si_iosize_max == 0) {
618 disk->d_cdev->si_iosize_max = disk->d_rawdev->si_iosize_max;
619 disk->d_cdev->si_bsize_phys = disk->d_rawdev->si_bsize_phys;
620 disk->d_cdev->si_bsize_best = disk->d_rawdev->si_bsize_best;
623 /* Add the serial number to the udev_dictionary */
624 if (info->d_serialno)
625 udev_dict_set_cstr(disk->d_cdev, "serno", info->d_serialno);
629 * Disk drivers must call this routine when media parameters are available
633 disk_setdiskinfo(struct disk *disk, struct disk_info *info)
635 _setdiskinfo(disk, info);
636 disk_msg_send(DISK_DISK_PROBE, disk, NULL);
638 "disk_setdiskinfo: sent probe for %s\n",
639 disk->d_cdev->si_name);
643 disk_setdiskinfo_sync(struct disk *disk, struct disk_info *info)
645 _setdiskinfo(disk, info);
646 disk_msg_send_sync(DISK_DISK_PROBE, disk, NULL);
648 "disk_setdiskinfo_sync: sent probe for %s\n",
649 disk->d_cdev->si_name);
653 * This routine is called when an adapter detaches. The higher level
654 * managed disk device is destroyed while the lower level raw device is
658 disk_destroy(struct disk *disk)
660 dsched_disk_destroy_callback(disk);
661 disk_msg_send_sync(DISK_DISK_DESTROY, disk, NULL);
666 disk_dumpcheck(cdev_t dev, u_int64_t *size, u_int64_t *blkno, u_int32_t *secsize)
668 struct partinfo pinfo;
671 bzero(&pinfo, sizeof(pinfo));
672 error = dev_dioctl(dev, DIOCGPART, (void *)&pinfo, 0,
673 proc0.p_ucred, NULL);
677 if (pinfo.media_blksize == 0)
680 if (blkno) /* XXX: make sure this reserved stuff is right */
681 *blkno = pinfo.reserved_blocks +
682 pinfo.media_offset / pinfo.media_blksize;
684 *secsize = pinfo.media_blksize;
686 *size = (pinfo.media_blocks - pinfo.reserved_blocks);
692 disk_dumpconf(cdev_t dev, u_int onoff)
694 struct dumperinfo di;
695 u_int64_t size, blkno;
700 return set_dumper(NULL);
702 error = disk_dumpcheck(dev, &size, &blkno, &secsize);
707 bzero(&di, sizeof(struct dumperinfo));
708 di.dumper = diskdump;
710 di.blocksize = secsize;
711 di.mediaoffset = blkno * DEV_BSIZE;
712 di.mediasize = size * DEV_BSIZE;
714 return set_dumper(&di);
718 disk_unprobe(struct disk *disk)
723 disk_msg_send_sync(DISK_UNPROBE, disk, NULL);
727 disk_invalidate (struct disk *disk)
729 dsgone(&disk->d_slice);
733 disk_enumerate(struct disk *disk)
737 lwkt_gettoken(&disklist_token);
739 dp = (LIST_FIRST(&disklist));
741 dp = (LIST_NEXT(disk, d_list));
742 lwkt_reltoken(&disklist_token);
749 sysctl_disks(SYSCTL_HANDLER_ARGS)
757 while ((disk = disk_enumerate(disk))) {
759 error = SYSCTL_OUT(req, " ", 1);
765 error = SYSCTL_OUT(req, disk->d_rawdev->si_name,
766 strlen(disk->d_rawdev->si_name));
770 error = SYSCTL_OUT(req, "", 1);
774 SYSCTL_PROC(_kern, OID_AUTO, disks, CTLTYPE_STRING | CTLFLAG_RD, NULL, 0,
775 sysctl_disks, "A", "names of available disks");
778 * Open a disk device or partition.
782 diskopen(struct dev_open_args *ap)
784 cdev_t dev = ap->a_head.a_dev;
789 * dp can't be NULL here XXX.
791 * d_slice will be NULL if setdiskinfo() has not been called yet.
792 * setdiskinfo() is typically called whether the disk is present
793 * or not (e.g. CD), but the base disk device is created first
794 * and there may be a race.
797 if (dp == NULL || dp->d_slice == NULL)
802 * Deal with open races
804 while (dp->d_flags & DISKFLAG_LOCK) {
805 dp->d_flags |= DISKFLAG_WANTED;
806 error = tsleep(dp, PCATCH, "diskopen", hz);
810 dp->d_flags |= DISKFLAG_LOCK;
813 * Open the underlying raw device.
815 if (!dsisopen(dp->d_slice)) {
817 if (!pdev->si_iosize_max)
818 pdev->si_iosize_max = dev->si_iosize_max;
820 error = dev_dopen(dp->d_rawdev, ap->a_oflags,
821 ap->a_devtype, ap->a_cred);
825 * Inherit properties from the underlying device now that it is
833 error = dsopen(dev, ap->a_devtype, dp->d_info.d_dsflags,
834 &dp->d_slice, &dp->d_info);
835 if (!dsisopen(dp->d_slice)) {
836 dev_dclose(dp->d_rawdev, ap->a_oflags, ap->a_devtype);
839 dp->d_flags &= ~DISKFLAG_LOCK;
840 if (dp->d_flags & DISKFLAG_WANTED) {
841 dp->d_flags &= ~DISKFLAG_WANTED;
849 * Close a disk device or partition
853 diskclose(struct dev_close_args *ap)
855 cdev_t dev = ap->a_head.a_dev;
862 dsclose(dev, ap->a_devtype, dp->d_slice);
863 if (!dsisopen(dp->d_slice)) {
864 error = dev_dclose(dp->d_rawdev, ap->a_fflag, ap->a_devtype);
870 * First execute the ioctl on the disk device, and if it isn't supported
871 * try running it on the backing device.
875 diskioctl(struct dev_ioctl_args *ap)
877 cdev_t dev = ap->a_head.a_dev;
886 devfs_debug(DEVFS_DEBUG_DEBUG,
887 "diskioctl: cmd is: %lx (name: %s)\n",
888 ap->a_cmd, dev->si_name);
889 devfs_debug(DEVFS_DEBUG_DEBUG,
890 "diskioctl: &dp->d_slice is: %p, %p\n",
891 &dp->d_slice, dp->d_slice);
893 if (ap->a_cmd == DIOCGKERNELDUMP) {
894 u = *(u_int *)ap->a_data;
895 return disk_dumpconf(dev, u);
898 if (&dp->d_slice == NULL || dp->d_slice == NULL) {
901 error = dsioctl(dev, ap->a_cmd, ap->a_data, ap->a_fflag,
902 &dp->d_slice, &dp->d_info);
905 if (error == ENOIOCTL) {
906 error = dev_dioctl(dp->d_rawdev, ap->a_cmd, ap->a_data,
907 ap->a_fflag, ap->a_cred, NULL);
913 * Execute strategy routine
917 diskstrategy(struct dev_strategy_args *ap)
919 cdev_t dev = ap->a_head.a_dev;
920 struct bio *bio = ap->a_bio;
927 bio->bio_buf->b_error = ENXIO;
928 bio->bio_buf->b_flags |= B_ERROR;
932 KKASSERT(dev->si_disk == dp);
935 * The dscheck() function will also transform the slice relative
936 * block number i.e. bio->bio_offset into a block number that can be
937 * passed directly to the underlying raw device. If dscheck()
938 * returns NULL it will have handled the bio for us (e.g. EOF
939 * or error due to being beyond the device size).
941 if ((nbio = dscheck(dev, bio, dp->d_slice)) != NULL) {
942 dsched_queue(dp, nbio);
950 * Return the partition size in ?blocks?
954 diskpsize(struct dev_psize_args *ap)
956 cdev_t dev = ap->a_head.a_dev;
962 ap->a_result = dssize(dev, &dp->d_slice);
967 * When new device entries are instantiated, make sure they inherit our
968 * si_disk structure and block and iosize limits from the raw device.
970 * This routine is always called synchronously in the context of the
973 * XXX The various io and block size constraints are not always initialized
974 * properly by devices.
978 diskclone(struct dev_clone_args *ap)
980 cdev_t dev = ap->a_head.a_dev;
984 KKASSERT(dp != NULL);
986 dev->si_iosize_max = dp->d_rawdev->si_iosize_max;
987 dev->si_bsize_phys = dp->d_rawdev->si_bsize_phys;
988 dev->si_bsize_best = dp->d_rawdev->si_bsize_best;
993 diskdump(struct dev_dump_args *ap)
995 cdev_t dev = ap->a_head.a_dev;
996 struct disk *dp = dev->si_disk;
997 u_int64_t size, offset;
1000 error = disk_dumpcheck(dev, &size, &ap->a_blkno, &ap->a_secsize);
1001 /* XXX: this should probably go in disk_dumpcheck somehow */
1002 if (ap->a_length != 0) {
1004 offset = ap->a_blkno * DEV_BSIZE;
1005 if ((ap->a_offset < offset) ||
1006 (ap->a_offset + ap->a_length - offset > size)) {
1007 kprintf("Attempt to write outside dump device boundaries.\n");
1013 ap->a_head.a_dev = dp->d_rawdev;
1014 error = dev_doperate(&ap->a_head);
1021 SYSCTL_INT(_debug_sizeof, OID_AUTO, diskslices, CTLFLAG_RD,
1022 0, sizeof(struct diskslices), "sizeof(struct diskslices)");
1024 SYSCTL_INT(_debug_sizeof, OID_AUTO, disk, CTLFLAG_RD,
1025 0, sizeof(struct disk), "sizeof(struct disk)");
1028 * Reorder interval for burst write allowance and minor write
1031 * We always want to trickle some writes in to make use of the
1032 * disk's zone cache. Bursting occurs on a longer interval and only
1033 * runningbufspace is well over the hirunningspace limit.
1035 int bioq_reorder_burst_interval = 60; /* should be multiple of minor */
1036 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_burst_interval,
1037 CTLFLAG_RW, &bioq_reorder_burst_interval, 0, "");
1038 int bioq_reorder_minor_interval = 5;
1039 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_minor_interval,
1040 CTLFLAG_RW, &bioq_reorder_minor_interval, 0, "");
1042 int bioq_reorder_burst_bytes = 3000000;
1043 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_burst_bytes,
1044 CTLFLAG_RW, &bioq_reorder_burst_bytes, 0, "");
1045 int bioq_reorder_minor_bytes = 262144;
1046 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_minor_bytes,
1047 CTLFLAG_RW, &bioq_reorder_minor_bytes, 0, "");
1051 * Order I/Os. Generally speaking this code is designed to make better
1052 * use of drive zone caches. A drive zone cache can typically track linear
1053 * reads or writes for around 16 zones simultaniously.
1055 * Read prioritization issues: It is possible for hundreds of megabytes worth
1056 * of writes to be queued asynchronously. This creates a huge bottleneck
1057 * for reads which reduce read bandwidth to a trickle.
1059 * To solve this problem we generally reorder reads before writes.
1061 * However, a large number of random reads can also starve writes and
1062 * make poor use of the drive zone cache so we allow writes to trickle
1066 bioqdisksort(struct bio_queue_head *bioq, struct bio *bio)
1069 * The BIO wants to be ordered. Adding to the tail also
1070 * causes transition to be set to NULL, forcing the ordering
1071 * of all prior I/O's.
1073 if (bio->bio_buf->b_flags & B_ORDERED) {
1074 bioq_insert_tail(bioq, bio);
1078 switch(bio->bio_buf->b_cmd) {
1080 if (bioq->transition) {
1082 * Insert before the first write. Bleedover writes
1083 * based on reorder intervals to prevent starvation.
1085 TAILQ_INSERT_BEFORE(bioq->transition, bio, bio_act);
1087 if (bioq->reorder % bioq_reorder_minor_interval == 0) {
1088 bioqwritereorder(bioq);
1089 if (bioq->reorder >=
1090 bioq_reorder_burst_interval) {
1096 * No writes queued (or ordering was forced),
1099 TAILQ_INSERT_TAIL(&bioq->queue, bio, bio_act);
1104 * Writes are always appended. If no writes were previously
1105 * queued or an ordered tail insertion occured the transition
1106 * field will be NULL.
1108 TAILQ_INSERT_TAIL(&bioq->queue, bio, bio_act);
1109 if (bioq->transition == NULL)
1110 bioq->transition = bio;
1114 * All other request types are forced to be ordered.
1116 bioq_insert_tail(bioq, bio);
1122 * Move the read-write transition point to prevent reads from
1123 * completely starving our writes. This brings a number of writes into
1124 * the fold every N reads.
1126 * We bring a few linear writes into the fold on a minor interval
1127 * and we bring a non-linear burst of writes into the fold on a major
1128 * interval. Bursting only occurs if runningbufspace is really high
1129 * (typically from syncs, fsyncs, or HAMMER flushes).
1133 bioqwritereorder(struct bio_queue_head *bioq)
1141 if (bioq->reorder < bioq_reorder_burst_interval ||
1142 !buf_runningbufspace_severe()) {
1143 left = (size_t)bioq_reorder_minor_bytes;
1146 left = (size_t)bioq_reorder_burst_bytes;
1150 next_offset = bioq->transition->bio_offset;
1151 while ((bio = bioq->transition) != NULL &&
1152 (check_off == 0 || next_offset == bio->bio_offset)
1154 n = bio->bio_buf->b_bcount;
1155 next_offset = bio->bio_offset + n;
1156 bioq->transition = TAILQ_NEXT(bio, bio_act);
1164 * Bounds checking against the media size, used for the raw partition.
1165 * secsize, mediasize and b_blkno must all be the same units.
1166 * Possibly this has to be DEV_BSIZE (512).
1169 bounds_check_with_mediasize(struct bio *bio, int secsize, uint64_t mediasize)
1171 struct buf *bp = bio->bio_buf;
1174 sz = howmany(bp->b_bcount, secsize);
1176 if (bio->bio_offset/DEV_BSIZE + sz > mediasize) {
1177 sz = mediasize - bio->bio_offset/DEV_BSIZE;
1179 /* If exactly at end of disk, return EOF. */
1180 bp->b_resid = bp->b_bcount;
1184 /* If past end of disk, return EINVAL. */
1185 bp->b_error = EINVAL;
1188 /* Otherwise, truncate request. */
1189 bp->b_bcount = sz * secsize;
1196 * Disk error is the preface to plaintive error messages
1197 * about failing disk transfers. It prints messages of the form
1199 hp0g: hard error reading fsbn 12345 of 12344-12347 (hp0 bn %d cn %d tn %d sn %d)
1201 * if the offset of the error in the transfer and a disk label
1202 * are both available. blkdone should be -1 if the position of the error
1203 * is unknown; the disklabel pointer may be null from drivers that have not
1204 * been converted to use them. The message is printed with kprintf
1205 * if pri is LOG_PRINTF, otherwise it uses log at the specified priority.
1206 * The message should be completed (with at least a newline) with kprintf
1207 * or log(-1, ...), respectively. There is no trailing space.
1210 diskerr(struct bio *bio, cdev_t dev, const char *what, int pri, int donecnt)
1212 struct buf *bp = bio->bio_buf;
1226 kprintf("%s: %s %sing ", dev->si_name, what, term);
1227 kprintf("offset %012llx for %d",
1228 (long long)bio->bio_offset,
1232 kprintf(" (%d bytes completed)", donecnt);
1236 * Locate a disk device
1239 disk_locate(const char *devname)
1241 return devfs_find_device_by_name(devname);
1245 disk_config(void *arg)
1247 disk_msg_send_sync(DISK_SYNC, NULL, NULL);
1253 struct thread* td_core;
1255 disk_msg_cache = objcache_create("disk-msg-cache", 0, 0,
1257 objcache_malloc_alloc,
1258 objcache_malloc_free,
1259 &disk_msg_malloc_args);
1261 lwkt_token_init(&disklist_token, 1, "disks");
1264 * Initialize the reply-only port which acts as a message drain
1266 lwkt_initport_replyonly(&disk_dispose_port, disk_msg_autofree_reply);
1268 lwkt_create(disk_msg_core, /*args*/NULL, &td_core, NULL,
1269 0, 0, "disk_msg_core");
1271 tsleep(td_core, 0, "diskcore", 0);
1277 objcache_destroy(disk_msg_cache);
1281 * Clean out illegal characters in serial numbers.
1284 disk_cleanserial(char *serno)
1288 while ((c = *serno) != 0) {
1289 if (c >= 'a' && c <= 'z')
1291 else if (c >= 'A' && c <= 'Z')
1293 else if (c >= '0' && c <= '9')
1295 else if (c == '-' || c == '@' || c == '+' || c == '.')
1303 TUNABLE_INT("kern.disk_debug", &disk_debug_enable);
1304 SYSCTL_INT(_kern, OID_AUTO, disk_debug, CTLFLAG_RW, &disk_debug_enable,
1305 0, "Enable subr_disk debugging");
1307 SYSINIT(disk_register, SI_SUB_PRE_DRIVERS, SI_ORDER_FIRST, disk_init, NULL);
1308 SYSUNINIT(disk_register, SI_SUB_PRE_DRIVERS, SI_ORDER_ANY, disk_uninit, NULL);