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/malloc.h>
97 #include <sys/sysctl.h>
98 #include <machine/md_var.h>
99 #include <sys/ctype.h>
100 #include <sys/syslog.h>
101 #include <sys/device.h>
102 #include <sys/msgport.h>
103 #include <sys/msgport2.h>
104 #include <sys/buf2.h>
105 #include <sys/devfs.h>
106 #include <sys/thread.h>
107 #include <sys/thread2.h>
109 #include <sys/queue.h>
110 #include <sys/lock.h>
112 static MALLOC_DEFINE(M_DISK, "disk", "disk data");
114 static void disk_msg_autofree_reply(lwkt_port_t, lwkt_msg_t);
115 static void disk_msg_core(void *);
116 static int disk_probe_slice(struct disk *dp, cdev_t dev, int slice, int reprobe);
117 static void disk_probe(struct disk *dp, int reprobe);
118 static void _setdiskinfo(struct disk *disk, struct disk_info *info);
119 static void bioqwritereorder(struct bio_queue_head *bioq);
121 static d_open_t diskopen;
122 static d_close_t diskclose;
123 static d_ioctl_t diskioctl;
124 static d_strategy_t diskstrategy;
125 static d_psize_t diskpsize;
126 static d_clone_t diskclone;
127 static d_dump_t diskdump;
129 static LIST_HEAD(, disk) disklist = LIST_HEAD_INITIALIZER(&disklist);
130 static struct lwkt_token disklist_token;
132 static struct dev_ops disk_ops = {
133 { "disk", 0, D_DISK },
135 .d_close = diskclose,
137 .d_write = physwrite,
138 .d_ioctl = diskioctl,
139 .d_strategy = diskstrategy,
141 .d_psize = diskpsize,
145 static struct objcache *disk_msg_cache;
147 struct objcache_malloc_args disk_msg_malloc_args = {
148 sizeof(struct disk_msg), M_DISK };
150 static struct lwkt_port disk_dispose_port;
151 static struct lwkt_port disk_msg_port;
155 disk_probe_slice(struct disk *dp, cdev_t dev, int slice, int reprobe)
157 struct disk_info *info = &dp->d_info;
158 struct diskslice *sp = &dp->d_slice->dss_slices[slice];
160 struct partinfo part;
166 sno = slice ? slice - 1 : 0;
168 ops = &disklabel32_ops;
169 msg = ops->op_readdisklabel(dev, sp, &sp->ds_label, info);
170 if (msg && !strcmp(msg, "no disk label")) {
171 ops = &disklabel64_ops;
172 msg = ops->op_readdisklabel(dev, sp, &sp->ds_label, info);
175 if (slice != WHOLE_DISK_SLICE)
176 ops->op_adjust_label_reserved(dp->d_slice, slice, sp);
181 for (i = 0; i < ops->op_getnumparts(sp->ds_label); i++) {
182 ops->op_loadpartinfo(sp->ds_label, i, &part);
185 (ndev = devfs_find_device_by_name("%s%c",
186 dev->si_name, 'a' + i))
189 * Device already exists and
192 ndev->si_flags |= SI_REPROBE_TEST;
194 ndev = make_dev(&disk_ops,
195 dkmakeminor(dkunit(dp->d_cdev),
197 UID_ROOT, GID_OPERATOR, 0640,
198 "%s%c", dev->si_name, 'a'+ i);
200 if (dp->d_info.d_serialno) {
203 dp->d_info.d_serialno,
206 ndev->si_flags |= SI_REPROBE_TEST;
210 } else if (info->d_dsflags & DSO_COMPATLABEL) {
212 if (sp->ds_size >= 0x100000000ULL)
213 ops = &disklabel64_ops;
215 ops = &disklabel32_ops;
216 sp->ds_label = ops->op_clone_label(info, sp);
218 if (sp->ds_type == DOSPTYP_386BSD /* XXX */) {
219 log(LOG_WARNING, "%s: cannot find label (%s)\n",
225 sp->ds_wlabel = FALSE;
228 return (msg ? EINVAL : 0);
233 disk_probe(struct disk *dp, int reprobe)
235 struct disk_info *info = &dp->d_info;
236 cdev_t dev = dp->d_cdev;
239 struct diskslice *sp;
241 KKASSERT (info->d_media_blksize != 0);
243 dp->d_slice = dsmakeslicestruct(BASE_SLICE, info);
245 error = mbrinit(dev, info, &(dp->d_slice));
249 for (i = 0; i < dp->d_slice->dss_nslices; i++) {
251 * Ignore the whole-disk slice, it has already been created.
253 if (i == WHOLE_DISK_SLICE)
255 sp = &dp->d_slice->dss_slices[i];
258 * Handle s0. s0 is a compatibility slice if there are no
259 * other slices and it has not otherwise been set up, else
262 if (i == COMPATIBILITY_SLICE) {
264 if (sp->ds_type == 0 &&
265 dp->d_slice->dss_nslices == BASE_SLICE) {
266 sp->ds_size = info->d_media_blocks;
275 * Ignore 0-length slices
277 if (sp->ds_size == 0)
281 (ndev = devfs_find_device_by_name("%ss%d",
282 dev->si_name, sno))) {
284 * Device already exists and is still valid
286 ndev->si_flags |= SI_REPROBE_TEST;
289 * Else create new device
291 ndev = make_dev(&disk_ops,
292 dkmakewholeslice(dkunit(dev), i),
293 UID_ROOT, GID_OPERATOR, 0640,
294 "%ss%d", dev->si_name, sno);
295 if (dp->d_info.d_serialno) {
296 make_dev_alias(ndev, "serno/%s.s%d",
297 dp->d_info.d_serialno, sno);
300 ndev->si_flags |= SI_REPROBE_TEST;
305 * Probe appropriate slices for a disklabel
307 * XXX slice type 1 used by our gpt probe code.
308 * XXX slice type 0 used by mbr compat slice.
310 if (sp->ds_type == DOSPTYP_386BSD || sp->ds_type == 0 ||
312 if (dp->d_slice->dss_first_bsd_slice == 0)
313 dp->d_slice->dss_first_bsd_slice = i;
314 disk_probe_slice(dp, ndev, i, reprobe);
321 disk_msg_core(void *arg)
324 struct diskslice *sp;
329 lwkt_initport_thread(&disk_msg_port, curthread);
334 msg = (disk_msg_t)lwkt_waitport(&disk_msg_port, 0);
336 switch (msg->hdr.u.ms_result) {
337 case DISK_DISK_PROBE:
338 dp = (struct disk *)msg->load;
341 case DISK_DISK_DESTROY:
342 dp = (struct disk *)msg->load;
343 devfs_destroy_subnames(dp->d_cdev->si_name);
344 devfs_destroy_dev(dp->d_cdev);
345 lwkt_gettoken(&ilock, &disklist_token);
346 LIST_REMOVE(dp, d_list);
347 lwkt_reltoken(&ilock);
348 if (dp->d_info.d_serialno) {
349 kfree(dp->d_info.d_serialno, M_TEMP);
350 dp->d_info.d_serialno = NULL;
354 dp = (struct disk *)msg->load;
355 devfs_destroy_subnames(dp->d_cdev->si_name);
357 case DISK_SLICE_REPROBE:
358 dp = (struct disk *)msg->load;
359 sp = (struct diskslice *)msg->load2;
360 devfs_clr_subnames_flag(sp->ds_dev->si_name,
362 devfs_debug(DEVFS_DEBUG_DEBUG,
363 "DISK_SLICE_REPROBE: %s\n",
364 sp->ds_dev->si_name);
365 disk_probe_slice(dp, sp->ds_dev,
366 dkslice(sp->ds_dev), 1);
367 devfs_destroy_subnames_without_flag(
368 sp->ds_dev->si_name, SI_REPROBE_TEST);
370 case DISK_DISK_REPROBE:
371 dp = (struct disk *)msg->load;
372 devfs_clr_subnames_flag(dp->d_cdev->si_name, SI_REPROBE_TEST);
373 devfs_debug(DEVFS_DEBUG_DEBUG,
374 "DISK_DISK_REPROBE: %s\n",
375 dp->d_cdev->si_name);
377 devfs_destroy_subnames_without_flag(
378 dp->d_cdev->si_name, SI_REPROBE_TEST);
383 devfs_debug(DEVFS_DEBUG_WARNING,
384 "disk_msg_core: unknown message "
385 "received at core\n");
388 lwkt_replymsg((lwkt_msg_t)msg, 0);
395 * Acts as a message drain. Any message that is replied to here gets
396 * destroyed and the memory freed.
399 disk_msg_autofree_reply(lwkt_port_t port, lwkt_msg_t msg)
401 objcache_put(disk_msg_cache, msg);
406 disk_msg_send(uint32_t cmd, void *load, void *load2)
409 lwkt_port_t port = &disk_msg_port;
411 disk_msg = objcache_get(disk_msg_cache, M_WAITOK);
413 lwkt_initmsg(&disk_msg->hdr, &disk_dispose_port, 0);
415 disk_msg->hdr.u.ms_result = cmd;
416 disk_msg->load = load;
417 disk_msg->load2 = load2;
419 lwkt_sendmsg(port, (lwkt_msg_t)disk_msg);
423 disk_msg_send_sync(uint32_t cmd, void *load, void *load2)
425 struct lwkt_port rep_port;
426 disk_msg_t disk_msg = objcache_get(disk_msg_cache, M_WAITOK);
427 disk_msg_t msg_incoming;
428 lwkt_port_t port = &disk_msg_port;
430 lwkt_initport_thread(&rep_port, curthread);
431 lwkt_initmsg(&disk_msg->hdr, &rep_port, 0);
433 disk_msg->hdr.u.ms_result = cmd;
434 disk_msg->load = load;
435 disk_msg->load2 = load2;
438 lwkt_sendmsg(port, (lwkt_msg_t)disk_msg);
439 msg_incoming = lwkt_waitport(&rep_port, 0);
443 * Create a raw device for the dev_ops template (which is returned). Also
444 * create a slice and unit managed disk and overload the user visible
445 * device space with it.
447 * NOTE: The returned raw device is NOT a slice and unit managed device.
448 * It is an actual raw device representing the raw disk as specified by
449 * the passed dev_ops. The disk layer not only returns such a raw device,
450 * it also uses it internally when passing (modified) commands through.
453 disk_create(int unit, struct disk *dp, struct dev_ops *raw_ops)
458 rawdev = make_only_dev(raw_ops, dkmakewholedisk(unit),
459 UID_ROOT, GID_OPERATOR, 0640,
460 "%s%d", raw_ops->head.name, unit);
462 bzero(dp, sizeof(*dp));
464 dp->d_rawdev = rawdev;
465 dp->d_raw_ops = raw_ops;
466 dp->d_dev_ops = &disk_ops;
467 dp->d_cdev = make_dev(&disk_ops,
468 dkmakewholedisk(unit),
469 UID_ROOT, GID_OPERATOR, 0640,
470 "%s%d", raw_ops->head.name, unit);
472 dp->d_cdev->si_disk = dp;
474 lwkt_gettoken(&ilock, &disklist_token);
475 LIST_INSERT_HEAD(&disklist, dp, d_list);
476 lwkt_reltoken(&ilock);
477 return (dp->d_rawdev);
482 _setdiskinfo(struct disk *disk, struct disk_info *info)
486 oldserialno = disk->d_info.d_serialno;
487 bcopy(info, &disk->d_info, sizeof(disk->d_info));
488 info = &disk->d_info;
491 * The serial number is duplicated so the caller can throw
494 if (info->d_serialno && info->d_serialno[0]) {
495 info->d_serialno = kstrdup(info->d_serialno, M_TEMP);
497 make_dev_alias(disk->d_cdev, "serno/%s",
501 info->d_serialno = NULL;
504 kfree(oldserialno, M_TEMP);
507 * The caller may set d_media_size or d_media_blocks and we
508 * calculate the other.
510 KKASSERT(info->d_media_size == 0 || info->d_media_blksize == 0);
511 if (info->d_media_size == 0 && info->d_media_blocks) {
512 info->d_media_size = (u_int64_t)info->d_media_blocks *
513 info->d_media_blksize;
514 } else if (info->d_media_size && info->d_media_blocks == 0 &&
515 info->d_media_blksize) {
516 info->d_media_blocks = info->d_media_size /
517 info->d_media_blksize;
521 * The si_* fields for rawdev are not set until after the
522 * disk_create() call, so someone using the cooked version
523 * of the raw device (i.e. da0s0) will not get the right
524 * si_iosize_max unless we fix it up here.
526 if (disk->d_cdev && disk->d_rawdev &&
527 disk->d_cdev->si_iosize_max == 0) {
528 disk->d_cdev->si_iosize_max = disk->d_rawdev->si_iosize_max;
529 disk->d_cdev->si_bsize_phys = disk->d_rawdev->si_bsize_phys;
530 disk->d_cdev->si_bsize_best = disk->d_rawdev->si_bsize_best;
535 * Disk drivers must call this routine when media parameters are available
539 disk_setdiskinfo(struct disk *disk, struct disk_info *info)
541 _setdiskinfo(disk, info);
542 disk_msg_send(DISK_DISK_PROBE, disk, NULL);
546 disk_setdiskinfo_sync(struct disk *disk, struct disk_info *info)
548 _setdiskinfo(disk, info);
549 disk_msg_send_sync(DISK_DISK_PROBE, disk, NULL);
553 * This routine is called when an adapter detaches. The higher level
554 * managed disk device is destroyed while the lower level raw device is
558 disk_destroy(struct disk *disk)
560 disk_msg_send_sync(DISK_DISK_DESTROY, disk, NULL);
565 disk_dumpcheck(cdev_t dev, u_int64_t *count, u_int64_t *blkno, u_int *secsize)
567 struct partinfo pinfo;
570 bzero(&pinfo, sizeof(pinfo));
571 error = dev_dioctl(dev, DIOCGPART, (void *)&pinfo, 0,
572 proc0.p_ucred, NULL);
575 if (pinfo.media_blksize == 0)
577 *count = (u_int64_t)Maxmem * PAGE_SIZE / pinfo.media_blksize;
578 if (dumplo64 < pinfo.reserved_blocks ||
579 dumplo64 + *count > pinfo.media_blocks) {
582 *blkno = dumplo64 + pinfo.media_offset / pinfo.media_blksize;
583 *secsize = pinfo.media_blksize;
588 disk_unprobe(struct disk *disk)
593 disk_msg_send_sync(DISK_UNPROBE, disk, NULL);
597 disk_invalidate (struct disk *disk)
600 dsgone(&disk->d_slice);
604 disk_enumerate(struct disk *disk)
609 lwkt_gettoken(&ilock, &disklist_token);
611 dp = (LIST_FIRST(&disklist));
613 dp = (LIST_NEXT(disk, d_list));
614 lwkt_reltoken(&ilock);
621 sysctl_disks(SYSCTL_HANDLER_ARGS)
629 while ((disk = disk_enumerate(disk))) {
631 error = SYSCTL_OUT(req, " ", 1);
637 error = SYSCTL_OUT(req, disk->d_rawdev->si_name,
638 strlen(disk->d_rawdev->si_name));
642 error = SYSCTL_OUT(req, "", 1);
646 SYSCTL_PROC(_kern, OID_AUTO, disks, CTLTYPE_STRING | CTLFLAG_RD, NULL, 0,
647 sysctl_disks, "A", "names of available disks");
650 * Open a disk device or partition.
654 diskopen(struct dev_open_args *ap)
656 cdev_t dev = ap->a_head.a_dev;
661 * dp can't be NULL here XXX.
663 * d_slice will be NULL if setdiskinfo() has not been called yet.
664 * setdiskinfo() is typically called whether the disk is present
665 * or not (e.g. CD), but the base disk device is created first
666 * and there may be a race.
669 if (dp == NULL || dp->d_slice == NULL)
674 * Deal with open races
676 while (dp->d_flags & DISKFLAG_LOCK) {
677 dp->d_flags |= DISKFLAG_WANTED;
678 error = tsleep(dp, PCATCH, "diskopen", hz);
682 dp->d_flags |= DISKFLAG_LOCK;
685 * Open the underlying raw device.
687 if (!dsisopen(dp->d_slice)) {
689 if (!pdev->si_iosize_max)
690 pdev->si_iosize_max = dev->si_iosize_max;
692 error = dev_dopen(dp->d_rawdev, ap->a_oflags,
693 ap->a_devtype, ap->a_cred);
697 * Inherit properties from the underlying device now that it is
705 error = dsopen(dev, ap->a_devtype, dp->d_info.d_dsflags,
706 &dp->d_slice, &dp->d_info);
707 if (!dsisopen(dp->d_slice)) {
708 dev_dclose(dp->d_rawdev, ap->a_oflags, ap->a_devtype);
711 dp->d_flags &= ~DISKFLAG_LOCK;
712 if (dp->d_flags & DISKFLAG_WANTED) {
713 dp->d_flags &= ~DISKFLAG_WANTED;
721 * Close a disk device or partition
725 diskclose(struct dev_close_args *ap)
727 cdev_t dev = ap->a_head.a_dev;
734 dsclose(dev, ap->a_devtype, dp->d_slice);
735 if (!dsisopen(dp->d_slice)) {
736 error = dev_dclose(dp->d_rawdev, ap->a_fflag, ap->a_devtype);
742 * First execute the ioctl on the disk device, and if it isn't supported
743 * try running it on the backing device.
747 diskioctl(struct dev_ioctl_args *ap)
749 cdev_t dev = ap->a_head.a_dev;
757 devfs_debug(DEVFS_DEBUG_DEBUG,
758 "diskioctl: cmd is: %x (name: %s)\n",
759 ap->a_cmd, dev->si_name);
760 devfs_debug(DEVFS_DEBUG_DEBUG,
761 "diskioctl: &dp->d_slice is: %x, %x\n",
762 &dp->d_slice, dp->d_slice);
764 error = dsioctl(dev, ap->a_cmd, ap->a_data, ap->a_fflag,
765 &dp->d_slice, &dp->d_info);
767 if (error == ENOIOCTL) {
768 error = dev_dioctl(dp->d_rawdev, ap->a_cmd, ap->a_data,
769 ap->a_fflag, ap->a_cred, NULL);
775 * Execute strategy routine
779 diskstrategy(struct dev_strategy_args *ap)
781 cdev_t dev = ap->a_head.a_dev;
782 struct bio *bio = ap->a_bio;
789 bio->bio_buf->b_error = ENXIO;
790 bio->bio_buf->b_flags |= B_ERROR;
794 KKASSERT(dev->si_disk == dp);
797 * The dscheck() function will also transform the slice relative
798 * block number i.e. bio->bio_offset into a block number that can be
799 * passed directly to the underlying raw device. If dscheck()
800 * returns NULL it will have handled the bio for us (e.g. EOF
801 * or error due to being beyond the device size).
803 if ((nbio = dscheck(dev, bio, dp->d_slice)) != NULL) {
804 dev_dstrategy(dp->d_rawdev, nbio);
812 * Return the partition size in ?blocks?
816 diskpsize(struct dev_psize_args *ap)
818 cdev_t dev = ap->a_head.a_dev;
824 ap->a_result = dssize(dev, &dp->d_slice);
829 * When new device entries are instantiated, make sure they inherit our
830 * si_disk structure and block and iosize limits from the raw device.
832 * This routine is always called synchronously in the context of the
835 * XXX The various io and block size constraints are not always initialized
836 * properly by devices.
840 diskclone(struct dev_clone_args *ap)
842 cdev_t dev = ap->a_head.a_dev;
846 KKASSERT(dp != NULL);
848 dev->si_iosize_max = dp->d_rawdev->si_iosize_max;
849 dev->si_bsize_phys = dp->d_rawdev->si_bsize_phys;
850 dev->si_bsize_best = dp->d_rawdev->si_bsize_best;
855 diskdump(struct dev_dump_args *ap)
857 cdev_t dev = ap->a_head.a_dev;
858 struct disk *dp = dev->si_disk;
861 error = disk_dumpcheck(dev, &ap->a_count, &ap->a_blkno, &ap->a_secsize);
863 ap->a_head.a_dev = dp->d_rawdev;
864 error = dev_doperate(&ap->a_head);
871 SYSCTL_INT(_debug_sizeof, OID_AUTO, diskslices, CTLFLAG_RD,
872 0, sizeof(struct diskslices), "sizeof(struct diskslices)");
874 SYSCTL_INT(_debug_sizeof, OID_AUTO, disk, CTLFLAG_RD,
875 0, sizeof(struct disk), "sizeof(struct disk)");
878 * Reorder interval for burst write allowance and minor write
881 * We always want to trickle some writes in to make use of the
882 * disk's zone cache. Bursting occurs on a longer interval and only
883 * runningbufspace is well over the hirunningspace limit.
885 int bioq_reorder_burst_interval = 60; /* should be multiple of minor */
886 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_burst_interval,
887 CTLFLAG_RW, &bioq_reorder_burst_interval, 0, "");
888 int bioq_reorder_minor_interval = 5;
889 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_minor_interval,
890 CTLFLAG_RW, &bioq_reorder_minor_interval, 0, "");
892 int bioq_reorder_burst_bytes = 3000000;
893 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_burst_bytes,
894 CTLFLAG_RW, &bioq_reorder_burst_bytes, 0, "");
895 int bioq_reorder_minor_bytes = 262144;
896 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_minor_bytes,
897 CTLFLAG_RW, &bioq_reorder_minor_bytes, 0, "");
901 * Order I/Os. Generally speaking this code is designed to make better
902 * use of drive zone caches. A drive zone cache can typically track linear
903 * reads or writes for around 16 zones simultaniously.
905 * Read prioritization issues: It is possible for hundreds of megabytes worth
906 * of writes to be queued asynchronously. This creates a huge bottleneck
907 * for reads which reduce read bandwidth to a trickle.
909 * To solve this problem we generally reorder reads before writes.
911 * However, a large number of random reads can also starve writes and
912 * make poor use of the drive zone cache so we allow writes to trickle
916 bioqdisksort(struct bio_queue_head *bioq, struct bio *bio)
919 * The BIO wants to be ordered. Adding to the tail also
920 * causes transition to be set to NULL, forcing the ordering
921 * of all prior I/O's.
923 if (bio->bio_buf->b_flags & B_ORDERED) {
924 bioq_insert_tail(bioq, bio);
928 switch(bio->bio_buf->b_cmd) {
930 if (bioq->transition) {
932 * Insert before the first write. Bleedover writes
933 * based on reorder intervals to prevent starvation.
935 TAILQ_INSERT_BEFORE(bioq->transition, bio, bio_act);
937 if (bioq->reorder % bioq_reorder_minor_interval == 0) {
938 bioqwritereorder(bioq);
940 bioq_reorder_burst_interval) {
946 * No writes queued (or ordering was forced),
949 TAILQ_INSERT_TAIL(&bioq->queue, bio, bio_act);
954 * Writes are always appended. If no writes were previously
955 * queued or an ordered tail insertion occured the transition
956 * field will be NULL.
958 TAILQ_INSERT_TAIL(&bioq->queue, bio, bio_act);
959 if (bioq->transition == NULL)
960 bioq->transition = bio;
964 * All other request types are forced to be ordered.
966 bioq_insert_tail(bioq, bio);
972 * Move the read-write transition point to prevent reads from
973 * completely starving our writes. This brings a number of writes into
974 * the fold every N reads.
976 * We bring a few linear writes into the fold on a minor interval
977 * and we bring a non-linear burst of writes into the fold on a major
978 * interval. Bursting only occurs if runningbufspace is really high
979 * (typically from syncs, fsyncs, or HAMMER flushes).
983 bioqwritereorder(struct bio_queue_head *bioq)
991 if (bioq->reorder < bioq_reorder_burst_interval ||
992 !buf_runningbufspace_severe()) {
993 left = (size_t)bioq_reorder_minor_bytes;
996 left = (size_t)bioq_reorder_burst_bytes;
1000 next_offset = bioq->transition->bio_offset;
1001 while ((bio = bioq->transition) != NULL &&
1002 (check_off == 0 || next_offset == bio->bio_offset)
1004 n = bio->bio_buf->b_bcount;
1005 next_offset = bio->bio_offset + n;
1006 bioq->transition = TAILQ_NEXT(bio, bio_act);
1014 * Disk error is the preface to plaintive error messages
1015 * about failing disk transfers. It prints messages of the form
1017 hp0g: hard error reading fsbn 12345 of 12344-12347 (hp0 bn %d cn %d tn %d sn %d)
1019 * if the offset of the error in the transfer and a disk label
1020 * are both available. blkdone should be -1 if the position of the error
1021 * is unknown; the disklabel pointer may be null from drivers that have not
1022 * been converted to use them. The message is printed with kprintf
1023 * if pri is LOG_PRINTF, otherwise it uses log at the specified priority.
1024 * The message should be completed (with at least a newline) with kprintf
1025 * or log(-1, ...), respectively. There is no trailing space.
1028 diskerr(struct bio *bio, cdev_t dev, const char *what, int pri, int donecnt)
1030 struct buf *bp = bio->bio_buf;
1044 kprintf("%s: %s %sing ", dev->si_name, what, term);
1045 kprintf("offset %012llx for %d",
1046 (long long)bio->bio_offset,
1050 kprintf(" (%d bytes completed)", donecnt);
1054 * Locate a disk device
1057 disk_locate(const char *devname)
1059 return devfs_find_device_by_name(devname);
1063 disk_config(void *arg)
1065 disk_msg_send_sync(DISK_SYNC, NULL, NULL);
1071 struct thread* td_core;
1073 disk_msg_cache = objcache_create("disk-msg-cache", 0, 0,
1075 objcache_malloc_alloc,
1076 objcache_malloc_free,
1077 &disk_msg_malloc_args);
1079 lwkt_token_init(&disklist_token);
1082 * Initialize the reply-only port which acts as a message drain
1084 lwkt_initport_replyonly(&disk_dispose_port, disk_msg_autofree_reply);
1086 lwkt_create(disk_msg_core, /*args*/NULL, &td_core, NULL,
1087 0, 0, "disk_msg_core");
1089 tsleep(td_core, 0, "diskcore", 0);
1095 objcache_destroy(disk_msg_cache);
1098 SYSINIT(disk_register, SI_SUB_PRE_DRIVERS, SI_ORDER_FIRST, disk_init, NULL);
1099 SYSUNINIT(disk_register, SI_SUB_PRE_DRIVERS, SI_ORDER_ANY, disk_uninit, NULL);