devfs - Change make_dev_covering to use dev_ops
[dragonfly.git] / sys / kern / subr_disk.c
blob459f5fe9142b3b27ec1932080cd00a90752c9433
1 /*
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
9 * are met:
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
16 * distribution.
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
32 * SUCH DAMAGE.
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
51 * are met:
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
75 * SUCH DAMAGE.
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>
86 #include <sys/proc.h>
87 #include <sys/sysctl.h>
88 #include <sys/buf.h>
89 #include <sys/conf.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>
95 #include <sys/disk.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");
113 static int disk_debug_enable = 0;
115 static void disk_msg_autofree_reply(lwkt_port_t, lwkt_msg_t);
116 static void disk_msg_core(void *);
117 static int disk_probe_slice(struct disk *dp, cdev_t dev, int slice, int reprobe);
118 static void disk_probe(struct disk *dp, int reprobe);
119 static void _setdiskinfo(struct disk *disk, struct disk_info *info);
120 static void bioqwritereorder(struct bio_queue_head *bioq);
121 static void disk_cleanserial(char *serno);
123 static d_open_t diskopen;
124 static d_close_t diskclose;
125 static d_ioctl_t diskioctl;
126 static d_strategy_t diskstrategy;
127 static d_psize_t diskpsize;
128 static d_clone_t diskclone;
129 static d_dump_t diskdump;
131 static LIST_HEAD(, disk) disklist = LIST_HEAD_INITIALIZER(&disklist);
132 static struct lwkt_token disklist_token;
134 static struct dev_ops disk_ops = {
135 { "disk", 0, D_DISK },
136 .d_open = diskopen,
137 .d_close = diskclose,
138 .d_read = physread,
139 .d_write = physwrite,
140 .d_ioctl = diskioctl,
141 .d_strategy = diskstrategy,
142 .d_dump = diskdump,
143 .d_psize = diskpsize,
144 .d_clone = diskclone
147 static struct objcache *disk_msg_cache;
149 struct objcache_malloc_args disk_msg_malloc_args = {
150 sizeof(struct disk_msg), M_DISK };
152 static struct lwkt_port disk_dispose_port;
153 static struct lwkt_port disk_msg_port;
155 static int
156 disk_debug(int level, char *fmt, ...)
158 __va_list ap;
160 __va_start(ap, fmt);
161 if (level <= disk_debug_enable)
162 kvprintf(fmt, ap);
163 __va_end(ap);
165 return 0;
168 static int
169 disk_probe_slice(struct disk *dp, cdev_t dev, int slice, int reprobe)
171 struct disk_info *info = &dp->d_info;
172 struct diskslice *sp = &dp->d_slice->dss_slices[slice];
173 disklabel_ops_t ops;
174 struct partinfo part;
175 const char *msg;
176 cdev_t ndev;
177 int sno;
178 u_int i;
180 disk_debug(2,
181 "disk_probe_slice (begin): %s (%s)\n",
182 dev->si_name, dp->d_cdev->si_name);
184 sno = slice ? slice - 1 : 0;
186 ops = &disklabel32_ops;
187 msg = ops->op_readdisklabel(dev, sp, &sp->ds_label, info);
188 if (msg && !strcmp(msg, "no disk label")) {
189 ops = &disklabel64_ops;
190 msg = ops->op_readdisklabel(dev, sp, &sp->ds_label, info);
192 if (msg == NULL) {
193 if (slice != WHOLE_DISK_SLICE)
194 ops->op_adjust_label_reserved(dp->d_slice, slice, sp);
195 else
196 sp->ds_reserved = 0;
198 sp->ds_ops = ops;
199 for (i = 0; i < ops->op_getnumparts(sp->ds_label); i++) {
200 ops->op_loadpartinfo(sp->ds_label, i, &part);
201 if (part.fstype) {
202 if (reprobe &&
203 (ndev = devfs_find_device_by_name("%s%c",
204 dev->si_name, 'a' + i))
207 * Device already exists and
208 * is still valid.
210 ndev->si_flags |= SI_REPROBE_TEST;
211 } else {
212 ndev = make_dev_covering(&disk_ops, dp->d_rawdev->si_ops,
213 dkmakeminor(dkunit(dp->d_cdev),
214 slice, i),
215 UID_ROOT, GID_OPERATOR, 0640,
216 "%s%c", dev->si_name, 'a'+ i);
217 ndev->si_disk = dp;
218 if (dp->d_info.d_serialno) {
219 make_dev_alias(ndev,
220 "serno/%s.s%d%c",
221 dp->d_info.d_serialno,
222 sno, 'a' + i);
224 ndev->si_flags |= SI_REPROBE_TEST;
228 } else if (info->d_dsflags & DSO_COMPATLABEL) {
229 msg = NULL;
230 if (sp->ds_size >= 0x100000000ULL)
231 ops = &disklabel64_ops;
232 else
233 ops = &disklabel32_ops;
234 sp->ds_label = ops->op_clone_label(info, sp);
235 } else {
236 if (sp->ds_type == DOSPTYP_386BSD /* XXX */) {
237 log(LOG_WARNING, "%s: cannot find label (%s)\n",
238 dev->si_name, msg);
242 if (msg == NULL) {
243 sp->ds_wlabel = FALSE;
246 return (msg ? EINVAL : 0);
250 static void
251 disk_probe(struct disk *dp, int reprobe)
253 struct disk_info *info = &dp->d_info;
254 cdev_t dev = dp->d_cdev;
255 cdev_t ndev;
256 int error, i, sno;
257 struct diskslice *sp;
259 KKASSERT (info->d_media_blksize != 0);
261 dp->d_slice = dsmakeslicestruct(BASE_SLICE, info);
262 disk_debug(1,
263 "disk_probe (begin): %s\n",
264 dp->d_cdev->si_name);
266 error = mbrinit(dev, info, &(dp->d_slice));
267 if (error)
268 return;
270 for (i = 0; i < dp->d_slice->dss_nslices; i++) {
272 * Ignore the whole-disk slice, it has already been created.
274 if (i == WHOLE_DISK_SLICE)
275 continue;
276 sp = &dp->d_slice->dss_slices[i];
279 * Handle s0. s0 is a compatibility slice if there are no
280 * other slices and it has not otherwise been set up, else
281 * we ignore it.
283 if (i == COMPATIBILITY_SLICE) {
284 sno = 0;
285 if (sp->ds_type == 0 &&
286 dp->d_slice->dss_nslices == BASE_SLICE) {
287 sp->ds_size = info->d_media_blocks;
288 sp->ds_reserved = 0;
290 } else {
291 sno = i - 1;
292 sp->ds_reserved = 0;
296 * Ignore 0-length slices
298 if (sp->ds_size == 0)
299 continue;
301 if (reprobe &&
302 (ndev = devfs_find_device_by_name("%ss%d",
303 dev->si_name, sno))) {
305 * Device already exists and is still valid
307 ndev->si_flags |= SI_REPROBE_TEST;
308 } else {
310 * Else create new device
312 ndev = make_dev_covering(&disk_ops, dp->d_rawdev->si_ops,
313 dkmakewholeslice(dkunit(dev), i),
314 UID_ROOT, GID_OPERATOR, 0640,
315 "%ss%d", dev->si_name, sno);
316 if (dp->d_info.d_serialno) {
317 make_dev_alias(ndev, "serno/%s.s%d",
318 dp->d_info.d_serialno, sno);
320 ndev->si_disk = dp;
321 ndev->si_flags |= SI_REPROBE_TEST;
323 sp->ds_dev = ndev;
326 * Probe appropriate slices for a disklabel
328 * XXX slice type 1 used by our gpt probe code.
329 * XXX slice type 0 used by mbr compat slice.
331 if (sp->ds_type == DOSPTYP_386BSD || sp->ds_type == 0 ||
332 sp->ds_type == 1 || sp->ds_type == DOSPTYP_NBSD) {
333 if (dp->d_slice->dss_first_bsd_slice == 0)
334 dp->d_slice->dss_first_bsd_slice = i;
335 disk_probe_slice(dp, ndev, i, reprobe);
338 disk_debug(1,
339 "disk_probe (end): %s\n",
340 dp->d_cdev->si_name);
344 static void
345 disk_msg_core(void *arg)
347 struct disk *dp;
348 struct diskslice *sp;
349 lwkt_tokref ilock;
350 disk_msg_t msg;
351 int run;
353 lwkt_initport_thread(&disk_msg_port, curthread);
354 wakeup(curthread);
355 run = 1;
357 while (run) {
358 msg = (disk_msg_t)lwkt_waitport(&disk_msg_port, 0);
360 switch (msg->hdr.u.ms_result) {
361 case DISK_DISK_PROBE:
362 dp = (struct disk *)msg->load;
363 disk_debug(1,
364 "DISK_DISK_PROBE: %s\n",
365 dp->d_cdev->si_name);
366 disk_probe(dp, 0);
367 break;
368 case DISK_DISK_DESTROY:
369 dp = (struct disk *)msg->load;
370 disk_debug(1,
371 "DISK_DISK_DESTROY: %s\n",
372 dp->d_cdev->si_name);
373 devfs_destroy_subnames(dp->d_cdev->si_name);
374 devfs_destroy_dev(dp->d_cdev);
375 lwkt_gettoken(&ilock, &disklist_token);
376 LIST_REMOVE(dp, d_list);
377 lwkt_reltoken(&ilock);
378 if (dp->d_info.d_serialno) {
379 kfree(dp->d_info.d_serialno, M_TEMP);
380 dp->d_info.d_serialno = NULL;
382 break;
383 case DISK_UNPROBE:
384 dp = (struct disk *)msg->load;
385 disk_debug(1,
386 "DISK_DISK_UNPROBE: %s\n",
387 dp->d_cdev->si_name);
388 devfs_destroy_subnames(dp->d_cdev->si_name);
389 break;
390 case DISK_SLICE_REPROBE:
391 dp = (struct disk *)msg->load;
392 sp = (struct diskslice *)msg->load2;
393 devfs_clr_subnames_flag(sp->ds_dev->si_name,
394 SI_REPROBE_TEST);
395 disk_debug(1,
396 "DISK_SLICE_REPROBE: %s\n",
397 sp->ds_dev->si_name);
398 disk_probe_slice(dp, sp->ds_dev,
399 dkslice(sp->ds_dev), 1);
400 devfs_destroy_subnames_without_flag(
401 sp->ds_dev->si_name, SI_REPROBE_TEST);
402 break;
403 case DISK_DISK_REPROBE:
404 dp = (struct disk *)msg->load;
405 devfs_clr_subnames_flag(dp->d_cdev->si_name, SI_REPROBE_TEST);
406 disk_debug(1,
407 "DISK_DISK_REPROBE: %s\n",
408 dp->d_cdev->si_name);
409 disk_probe(dp, 1);
410 devfs_destroy_subnames_without_flag(
411 dp->d_cdev->si_name, SI_REPROBE_TEST);
412 break;
413 case DISK_SYNC:
414 disk_debug(1, "DISK_SYNC\n");
415 break;
416 default:
417 devfs_debug(DEVFS_DEBUG_WARNING,
418 "disk_msg_core: unknown message "
419 "received at core\n");
420 break;
422 lwkt_replymsg((lwkt_msg_t)msg, 0);
424 lwkt_exit();
429 * Acts as a message drain. Any message that is replied to here gets
430 * destroyed and the memory freed.
432 static void
433 disk_msg_autofree_reply(lwkt_port_t port, lwkt_msg_t msg)
435 objcache_put(disk_msg_cache, msg);
439 void
440 disk_msg_send(uint32_t cmd, void *load, void *load2)
442 disk_msg_t disk_msg;
443 lwkt_port_t port = &disk_msg_port;
445 disk_msg = objcache_get(disk_msg_cache, M_WAITOK);
447 lwkt_initmsg(&disk_msg->hdr, &disk_dispose_port, 0);
449 disk_msg->hdr.u.ms_result = cmd;
450 disk_msg->load = load;
451 disk_msg->load2 = load2;
452 KKASSERT(port);
453 lwkt_sendmsg(port, (lwkt_msg_t)disk_msg);
456 void
457 disk_msg_send_sync(uint32_t cmd, void *load, void *load2)
459 struct lwkt_port rep_port;
460 disk_msg_t disk_msg = objcache_get(disk_msg_cache, M_WAITOK);
461 disk_msg_t msg_incoming;
462 lwkt_port_t port = &disk_msg_port;
464 lwkt_initport_thread(&rep_port, curthread);
465 lwkt_initmsg(&disk_msg->hdr, &rep_port, 0);
467 disk_msg->hdr.u.ms_result = cmd;
468 disk_msg->load = load;
469 disk_msg->load2 = load2;
471 KKASSERT(port);
472 lwkt_sendmsg(port, (lwkt_msg_t)disk_msg);
473 msg_incoming = lwkt_waitport(&rep_port, 0);
477 * Create a raw device for the dev_ops template (which is returned). Also
478 * create a slice and unit managed disk and overload the user visible
479 * device space with it.
481 * NOTE: The returned raw device is NOT a slice and unit managed device.
482 * It is an actual raw device representing the raw disk as specified by
483 * the passed dev_ops. The disk layer not only returns such a raw device,
484 * it also uses it internally when passing (modified) commands through.
486 cdev_t
487 disk_create(int unit, struct disk *dp, struct dev_ops *raw_ops)
489 lwkt_tokref ilock;
490 cdev_t rawdev;
492 disk_debug(1,
493 "disk_create (begin): %s%d\n",
494 raw_ops->head.name, unit);
496 rawdev = make_only_dev(raw_ops, dkmakewholedisk(unit),
497 UID_ROOT, GID_OPERATOR, 0640,
498 "%s%d", raw_ops->head.name, unit);
500 bzero(dp, sizeof(*dp));
502 dp->d_rawdev = rawdev;
503 dp->d_raw_ops = raw_ops;
504 dp->d_dev_ops = &disk_ops;
505 dp->d_cdev = make_dev_covering(&disk_ops, dp->d_rawdev->si_ops,
506 dkmakewholedisk(unit),
507 UID_ROOT, GID_OPERATOR, 0640,
508 "%s%d", raw_ops->head.name, unit);
510 dp->d_cdev->si_disk = dp;
512 lwkt_gettoken(&ilock, &disklist_token);
513 LIST_INSERT_HEAD(&disklist, dp, d_list);
514 lwkt_reltoken(&ilock);
516 disk_debug(1,
517 "disk_create (end): %s%d\n",
518 raw_ops->head.name, unit);
520 return (dp->d_rawdev);
524 static void
525 _setdiskinfo(struct disk *disk, struct disk_info *info)
527 char *oldserialno;
529 oldserialno = disk->d_info.d_serialno;
530 bcopy(info, &disk->d_info, sizeof(disk->d_info));
531 info = &disk->d_info;
533 disk_debug(1,
534 "_setdiskinfo: %s\n",
535 disk->d_cdev->si_name);
538 * The serial number is duplicated so the caller can throw
539 * their copy away.
541 if (info->d_serialno && info->d_serialno[0]) {
542 info->d_serialno = kstrdup(info->d_serialno, M_TEMP);
543 disk_cleanserial(info->d_serialno);
544 if (disk->d_cdev) {
545 make_dev_alias(disk->d_cdev, "serno/%s",
546 info->d_serialno);
548 } else {
549 info->d_serialno = NULL;
551 if (oldserialno)
552 kfree(oldserialno, M_TEMP);
555 * The caller may set d_media_size or d_media_blocks and we
556 * calculate the other.
558 KKASSERT(info->d_media_size == 0 || info->d_media_blksize == 0);
559 if (info->d_media_size == 0 && info->d_media_blocks) {
560 info->d_media_size = (u_int64_t)info->d_media_blocks *
561 info->d_media_blksize;
562 } else if (info->d_media_size && info->d_media_blocks == 0 &&
563 info->d_media_blksize) {
564 info->d_media_blocks = info->d_media_size /
565 info->d_media_blksize;
569 * The si_* fields for rawdev are not set until after the
570 * disk_create() call, so someone using the cooked version
571 * of the raw device (i.e. da0s0) will not get the right
572 * si_iosize_max unless we fix it up here.
574 if (disk->d_cdev && disk->d_rawdev &&
575 disk->d_cdev->si_iosize_max == 0) {
576 disk->d_cdev->si_iosize_max = disk->d_rawdev->si_iosize_max;
577 disk->d_cdev->si_bsize_phys = disk->d_rawdev->si_bsize_phys;
578 disk->d_cdev->si_bsize_best = disk->d_rawdev->si_bsize_best;
583 * Disk drivers must call this routine when media parameters are available
584 * or have changed.
586 void
587 disk_setdiskinfo(struct disk *disk, struct disk_info *info)
589 _setdiskinfo(disk, info);
590 disk_msg_send(DISK_DISK_PROBE, disk, NULL);
591 disk_debug(1,
592 "disk_setdiskinfo: sent probe for %s\n",
593 disk->d_cdev->si_name);
596 void
597 disk_setdiskinfo_sync(struct disk *disk, struct disk_info *info)
599 _setdiskinfo(disk, info);
600 disk_msg_send_sync(DISK_DISK_PROBE, disk, NULL);
601 disk_debug(1,
602 "disk_setdiskinfo_sync: sent probe for %s\n",
603 disk->d_cdev->si_name);
607 * This routine is called when an adapter detaches. The higher level
608 * managed disk device is destroyed while the lower level raw device is
609 * released.
611 void
612 disk_destroy(struct disk *disk)
614 disk_msg_send_sync(DISK_DISK_DESTROY, disk, NULL);
615 return;
619 disk_dumpcheck(cdev_t dev, u_int64_t *count, u_int64_t *blkno, u_int *secsize)
621 struct partinfo pinfo;
622 int error;
624 bzero(&pinfo, sizeof(pinfo));
625 error = dev_dioctl(dev, DIOCGPART, (void *)&pinfo, 0,
626 proc0.p_ucred, NULL);
627 if (error)
628 return (error);
629 if (pinfo.media_blksize == 0)
630 return (ENXIO);
631 *count = (u_int64_t)Maxmem * PAGE_SIZE / pinfo.media_blksize;
632 if (dumplo64 < pinfo.reserved_blocks ||
633 dumplo64 + *count > pinfo.media_blocks) {
634 return (ENOSPC);
636 *blkno = dumplo64 + pinfo.media_offset / pinfo.media_blksize;
637 *secsize = pinfo.media_blksize;
638 return (0);
641 void
642 disk_unprobe(struct disk *disk)
644 if (disk == NULL)
645 return;
647 disk_msg_send_sync(DISK_UNPROBE, disk, NULL);
650 void
651 disk_invalidate (struct disk *disk)
653 if (disk->d_slice)
654 dsgone(&disk->d_slice);
657 struct disk *
658 disk_enumerate(struct disk *disk)
660 struct disk *dp;
661 lwkt_tokref ilock;
663 lwkt_gettoken(&ilock, &disklist_token);
664 if (!disk)
665 dp = (LIST_FIRST(&disklist));
666 else
667 dp = (LIST_NEXT(disk, d_list));
668 lwkt_reltoken(&ilock);
670 return dp;
673 static
675 sysctl_disks(SYSCTL_HANDLER_ARGS)
677 struct disk *disk;
678 int error, first;
680 disk = NULL;
681 first = 1;
683 while ((disk = disk_enumerate(disk))) {
684 if (!first) {
685 error = SYSCTL_OUT(req, " ", 1);
686 if (error)
687 return error;
688 } else {
689 first = 0;
691 error = SYSCTL_OUT(req, disk->d_rawdev->si_name,
692 strlen(disk->d_rawdev->si_name));
693 if (error)
694 return error;
696 error = SYSCTL_OUT(req, "", 1);
697 return error;
700 SYSCTL_PROC(_kern, OID_AUTO, disks, CTLTYPE_STRING | CTLFLAG_RD, NULL, 0,
701 sysctl_disks, "A", "names of available disks");
704 * Open a disk device or partition.
706 static
708 diskopen(struct dev_open_args *ap)
710 cdev_t dev = ap->a_head.a_dev;
711 struct disk *dp;
712 int error;
715 * dp can't be NULL here XXX.
717 * d_slice will be NULL if setdiskinfo() has not been called yet.
718 * setdiskinfo() is typically called whether the disk is present
719 * or not (e.g. CD), but the base disk device is created first
720 * and there may be a race.
722 dp = dev->si_disk;
723 if (dp == NULL || dp->d_slice == NULL)
724 return (ENXIO);
725 error = 0;
728 * Deal with open races
730 while (dp->d_flags & DISKFLAG_LOCK) {
731 dp->d_flags |= DISKFLAG_WANTED;
732 error = tsleep(dp, PCATCH, "diskopen", hz);
733 if (error)
734 return (error);
736 dp->d_flags |= DISKFLAG_LOCK;
739 * Open the underlying raw device.
741 if (!dsisopen(dp->d_slice)) {
742 #if 0
743 if (!pdev->si_iosize_max)
744 pdev->si_iosize_max = dev->si_iosize_max;
745 #endif
746 error = dev_dopen(dp->d_rawdev, ap->a_oflags,
747 ap->a_devtype, ap->a_cred);
749 #if 0
751 * Inherit properties from the underlying device now that it is
752 * open.
754 dev_dclone(dev);
755 #endif
757 if (error)
758 goto out;
759 error = dsopen(dev, ap->a_devtype, dp->d_info.d_dsflags,
760 &dp->d_slice, &dp->d_info);
761 if (!dsisopen(dp->d_slice)) {
762 dev_dclose(dp->d_rawdev, ap->a_oflags, ap->a_devtype);
764 out:
765 dp->d_flags &= ~DISKFLAG_LOCK;
766 if (dp->d_flags & DISKFLAG_WANTED) {
767 dp->d_flags &= ~DISKFLAG_WANTED;
768 wakeup(dp);
771 return(error);
775 * Close a disk device or partition
777 static
779 diskclose(struct dev_close_args *ap)
781 cdev_t dev = ap->a_head.a_dev;
782 struct disk *dp;
783 int error;
785 error = 0;
786 dp = dev->si_disk;
788 dsclose(dev, ap->a_devtype, dp->d_slice);
789 if (!dsisopen(dp->d_slice)) {
790 error = dev_dclose(dp->d_rawdev, ap->a_fflag, ap->a_devtype);
792 return (error);
796 * First execute the ioctl on the disk device, and if it isn't supported
797 * try running it on the backing device.
799 static
801 diskioctl(struct dev_ioctl_args *ap)
803 cdev_t dev = ap->a_head.a_dev;
804 struct disk *dp;
805 int error;
807 dp = dev->si_disk;
808 if (dp == NULL)
809 return (ENXIO);
811 devfs_debug(DEVFS_DEBUG_DEBUG,
812 "diskioctl: cmd is: %x (name: %s)\n",
813 ap->a_cmd, dev->si_name);
814 devfs_debug(DEVFS_DEBUG_DEBUG,
815 "diskioctl: &dp->d_slice is: %x, %x\n",
816 &dp->d_slice, dp->d_slice);
818 error = dsioctl(dev, ap->a_cmd, ap->a_data, ap->a_fflag,
819 &dp->d_slice, &dp->d_info);
821 if (error == ENOIOCTL) {
822 error = dev_dioctl(dp->d_rawdev, ap->a_cmd, ap->a_data,
823 ap->a_fflag, ap->a_cred, NULL);
825 return (error);
829 * Execute strategy routine
831 static
833 diskstrategy(struct dev_strategy_args *ap)
835 cdev_t dev = ap->a_head.a_dev;
836 struct bio *bio = ap->a_bio;
837 struct bio *nbio;
838 struct disk *dp;
840 dp = dev->si_disk;
842 if (dp == NULL) {
843 bio->bio_buf->b_error = ENXIO;
844 bio->bio_buf->b_flags |= B_ERROR;
845 biodone(bio);
846 return(0);
848 KKASSERT(dev->si_disk == dp);
851 * The dscheck() function will also transform the slice relative
852 * block number i.e. bio->bio_offset into a block number that can be
853 * passed directly to the underlying raw device. If dscheck()
854 * returns NULL it will have handled the bio for us (e.g. EOF
855 * or error due to being beyond the device size).
857 if ((nbio = dscheck(dev, bio, dp->d_slice)) != NULL) {
858 dev_dstrategy(dp->d_rawdev, nbio);
859 } else {
860 biodone(bio);
862 return(0);
866 * Return the partition size in ?blocks?
868 static
870 diskpsize(struct dev_psize_args *ap)
872 cdev_t dev = ap->a_head.a_dev;
873 struct disk *dp;
875 dp = dev->si_disk;
876 if (dp == NULL)
877 return(ENODEV);
878 ap->a_result = dssize(dev, &dp->d_slice);
879 return(0);
883 * When new device entries are instantiated, make sure they inherit our
884 * si_disk structure and block and iosize limits from the raw device.
886 * This routine is always called synchronously in the context of the
887 * client.
889 * XXX The various io and block size constraints are not always initialized
890 * properly by devices.
892 static
894 diskclone(struct dev_clone_args *ap)
896 cdev_t dev = ap->a_head.a_dev;
897 struct disk *dp;
898 dp = dev->si_disk;
900 KKASSERT(dp != NULL);
901 dev->si_disk = dp;
902 dev->si_iosize_max = dp->d_rawdev->si_iosize_max;
903 dev->si_bsize_phys = dp->d_rawdev->si_bsize_phys;
904 dev->si_bsize_best = dp->d_rawdev->si_bsize_best;
905 return(0);
909 diskdump(struct dev_dump_args *ap)
911 cdev_t dev = ap->a_head.a_dev;
912 struct disk *dp = dev->si_disk;
913 int error;
915 error = disk_dumpcheck(dev, &ap->a_count, &ap->a_blkno, &ap->a_secsize);
916 if (error == 0) {
917 ap->a_head.a_dev = dp->d_rawdev;
918 error = dev_doperate(&ap->a_head);
921 return(error);
925 SYSCTL_INT(_debug_sizeof, OID_AUTO, diskslices, CTLFLAG_RD,
926 0, sizeof(struct diskslices), "sizeof(struct diskslices)");
928 SYSCTL_INT(_debug_sizeof, OID_AUTO, disk, CTLFLAG_RD,
929 0, sizeof(struct disk), "sizeof(struct disk)");
932 * Reorder interval for burst write allowance and minor write
933 * allowance.
935 * We always want to trickle some writes in to make use of the
936 * disk's zone cache. Bursting occurs on a longer interval and only
937 * runningbufspace is well over the hirunningspace limit.
939 int bioq_reorder_burst_interval = 60; /* should be multiple of minor */
940 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_burst_interval,
941 CTLFLAG_RW, &bioq_reorder_burst_interval, 0, "");
942 int bioq_reorder_minor_interval = 5;
943 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_minor_interval,
944 CTLFLAG_RW, &bioq_reorder_minor_interval, 0, "");
946 int bioq_reorder_burst_bytes = 3000000;
947 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_burst_bytes,
948 CTLFLAG_RW, &bioq_reorder_burst_bytes, 0, "");
949 int bioq_reorder_minor_bytes = 262144;
950 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_minor_bytes,
951 CTLFLAG_RW, &bioq_reorder_minor_bytes, 0, "");
955 * Order I/Os. Generally speaking this code is designed to make better
956 * use of drive zone caches. A drive zone cache can typically track linear
957 * reads or writes for around 16 zones simultaniously.
959 * Read prioritization issues: It is possible for hundreds of megabytes worth
960 * of writes to be queued asynchronously. This creates a huge bottleneck
961 * for reads which reduce read bandwidth to a trickle.
963 * To solve this problem we generally reorder reads before writes.
965 * However, a large number of random reads can also starve writes and
966 * make poor use of the drive zone cache so we allow writes to trickle
967 * in every N reads.
969 void
970 bioqdisksort(struct bio_queue_head *bioq, struct bio *bio)
973 * The BIO wants to be ordered. Adding to the tail also
974 * causes transition to be set to NULL, forcing the ordering
975 * of all prior I/O's.
977 if (bio->bio_buf->b_flags & B_ORDERED) {
978 bioq_insert_tail(bioq, bio);
979 return;
982 switch(bio->bio_buf->b_cmd) {
983 case BUF_CMD_READ:
984 if (bioq->transition) {
986 * Insert before the first write. Bleedover writes
987 * based on reorder intervals to prevent starvation.
989 TAILQ_INSERT_BEFORE(bioq->transition, bio, bio_act);
990 ++bioq->reorder;
991 if (bioq->reorder % bioq_reorder_minor_interval == 0) {
992 bioqwritereorder(bioq);
993 if (bioq->reorder >=
994 bioq_reorder_burst_interval) {
995 bioq->reorder = 0;
998 } else {
1000 * No writes queued (or ordering was forced),
1001 * insert at tail.
1003 TAILQ_INSERT_TAIL(&bioq->queue, bio, bio_act);
1005 break;
1006 case BUF_CMD_WRITE:
1008 * Writes are always appended. If no writes were previously
1009 * queued or an ordered tail insertion occured the transition
1010 * field will be NULL.
1012 TAILQ_INSERT_TAIL(&bioq->queue, bio, bio_act);
1013 if (bioq->transition == NULL)
1014 bioq->transition = bio;
1015 break;
1016 default:
1018 * All other request types are forced to be ordered.
1020 bioq_insert_tail(bioq, bio);
1021 break;
1026 * Move the read-write transition point to prevent reads from
1027 * completely starving our writes. This brings a number of writes into
1028 * the fold every N reads.
1030 * We bring a few linear writes into the fold on a minor interval
1031 * and we bring a non-linear burst of writes into the fold on a major
1032 * interval. Bursting only occurs if runningbufspace is really high
1033 * (typically from syncs, fsyncs, or HAMMER flushes).
1035 static
1036 void
1037 bioqwritereorder(struct bio_queue_head *bioq)
1039 struct bio *bio;
1040 off_t next_offset;
1041 size_t left;
1042 size_t n;
1043 int check_off;
1045 if (bioq->reorder < bioq_reorder_burst_interval ||
1046 !buf_runningbufspace_severe()) {
1047 left = (size_t)bioq_reorder_minor_bytes;
1048 check_off = 1;
1049 } else {
1050 left = (size_t)bioq_reorder_burst_bytes;
1051 check_off = 0;
1054 next_offset = bioq->transition->bio_offset;
1055 while ((bio = bioq->transition) != NULL &&
1056 (check_off == 0 || next_offset == bio->bio_offset)
1058 n = bio->bio_buf->b_bcount;
1059 next_offset = bio->bio_offset + n;
1060 bioq->transition = TAILQ_NEXT(bio, bio_act);
1061 if (left < n)
1062 break;
1063 left -= n;
1068 * Disk error is the preface to plaintive error messages
1069 * about failing disk transfers. It prints messages of the form
1071 hp0g: hard error reading fsbn 12345 of 12344-12347 (hp0 bn %d cn %d tn %d sn %d)
1073 * if the offset of the error in the transfer and a disk label
1074 * are both available. blkdone should be -1 if the position of the error
1075 * is unknown; the disklabel pointer may be null from drivers that have not
1076 * been converted to use them. The message is printed with kprintf
1077 * if pri is LOG_PRINTF, otherwise it uses log at the specified priority.
1078 * The message should be completed (with at least a newline) with kprintf
1079 * or log(-1, ...), respectively. There is no trailing space.
1081 void
1082 diskerr(struct bio *bio, cdev_t dev, const char *what, int pri, int donecnt)
1084 struct buf *bp = bio->bio_buf;
1085 const char *term;
1087 switch(bp->b_cmd) {
1088 case BUF_CMD_READ:
1089 term = "read";
1090 break;
1091 case BUF_CMD_WRITE:
1092 term = "write";
1093 break;
1094 default:
1095 term = "access";
1096 break;
1098 kprintf("%s: %s %sing ", dev->si_name, what, term);
1099 kprintf("offset %012llx for %d",
1100 (long long)bio->bio_offset,
1101 bp->b_bcount);
1103 if (donecnt)
1104 kprintf(" (%d bytes completed)", donecnt);
1108 * Locate a disk device
1110 cdev_t
1111 disk_locate(const char *devname)
1113 return devfs_find_device_by_name(devname);
1116 void
1117 disk_config(void *arg)
1119 disk_msg_send_sync(DISK_SYNC, NULL, NULL);
1122 static void
1123 disk_init(void)
1125 struct thread* td_core;
1127 disk_msg_cache = objcache_create("disk-msg-cache", 0, 0,
1128 NULL, NULL, NULL,
1129 objcache_malloc_alloc,
1130 objcache_malloc_free,
1131 &disk_msg_malloc_args);
1133 lwkt_token_init(&disklist_token);
1136 * Initialize the reply-only port which acts as a message drain
1138 lwkt_initport_replyonly(&disk_dispose_port, disk_msg_autofree_reply);
1140 lwkt_create(disk_msg_core, /*args*/NULL, &td_core, NULL,
1141 0, 0, "disk_msg_core");
1143 tsleep(td_core, 0, "diskcore", 0);
1146 static void
1147 disk_uninit(void)
1149 objcache_destroy(disk_msg_cache);
1153 * Clean out illegal characters in serial numbers.
1155 static void
1156 disk_cleanserial(char *serno)
1158 char c;
1160 while ((c = *serno) != 0) {
1161 if (c >= 'a' && c <= 'z')
1163 else if (c >= 'A' && c <= 'Z')
1165 else if (c >= '0' && c <= '9')
1167 else if (c == '-' || c == '@' || c == '+' || c == '.')
1169 else
1170 c = '_';
1171 *serno++= c;
1175 TUNABLE_INT("kern.disk_debug", &disk_debug_enable);
1176 SYSCTL_INT(_kern, OID_AUTO, disk_debug, CTLFLAG_RW, &disk_debug_enable,
1177 0, "Enable subr_disk debugging");
1179 SYSINIT(disk_register, SI_SUB_PRE_DRIVERS, SI_ORDER_FIRST, disk_init, NULL);
1180 SYSUNINIT(disk_register, SI_SUB_PRE_DRIVERS, SI_ORDER_ANY, disk_uninit, NULL);