xref: /openbsd-src/sys/kern/subr_disk.c (revision 1fc27e414118cd8922c6b93fbaeb7a5246bfd593)
1 /*	$OpenBSD: subr_disk.c,v 1.17 1999/11/17 04:31:22 d Exp $	*/
2 /*	$NetBSD: subr_disk.c,v 1.17 1996/03/16 23:17:08 christos Exp $	*/
3 
4 /*
5  * Copyright (c) 1995 Jason R. Thorpe.  All rights reserved.
6  * Copyright (c) 1982, 1986, 1988, 1993
7  *	The Regents of the University of California.  All rights reserved.
8  * (c) UNIX System Laboratories, Inc.
9  * All or some portions of this file are derived from material licensed
10  * to the University of California by American Telephone and Telegraph
11  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
12  * the permission of UNIX System Laboratories, Inc.
13  *
14  * Redistribution and use in source and binary forms, with or without
15  * modification, are permitted provided that the following conditions
16  * are met:
17  * 1. Redistributions of source code must retain the above copyright
18  *    notice, this list of conditions and the following disclaimer.
19  * 2. Redistributions in binary form must reproduce the above copyright
20  *    notice, this list of conditions and the following disclaimer in the
21  *    documentation and/or other materials provided with the distribution.
22  * 3. All advertising materials mentioning features or use of this software
23  *    must display the following acknowledgement:
24  *	This product includes software developed by the University of
25  *	California, Berkeley and its contributors.
26  * 4. Neither the name of the University nor the names of its contributors
27  *    may be used to endorse or promote products derived from this software
28  *    without specific prior written permission.
29  *
30  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
31  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
32  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
33  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
34  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
35  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
36  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
37  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
38  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
39  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
40  * SUCH DAMAGE.
41  *
42  *	@(#)ufs_disksubr.c	8.5 (Berkeley) 1/21/94
43  */
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/kernel.h>
48 #include <sys/malloc.h>
49 #include <sys/fcntl.h>
50 #include <sys/buf.h>
51 #include <sys/stat.h>
52 #include <sys/syslog.h>
53 #include <sys/time.h>
54 #include <sys/disklabel.h>
55 #include <sys/conf.h>
56 #include <sys/disk.h>
57 #include <sys/dkio.h>
58 #include <sys/dkstat.h>		/* XXX */
59 
60 #include <dev/rndvar.h>
61 
62 /*
63  * A global list of all disks attached to the system.  May grow or
64  * shrink over time.
65  */
66 struct	disklist_head disklist;	/* TAILQ_HEAD */
67 int	disk_count;		/* number of drives in global disklist */
68 
69 /*
70  * Seek sort for disks.  We depend on the driver which calls us using b_resid
71  * as the current cylinder number.
72  *
73  * The argument ap structure holds a b_actf activity chain pointer on which we
74  * keep two queues, sorted in ascending cylinder order.  The first queue holds
75  * those requests which are positioned after the current cylinder (in the first
76  * request); the second holds requests which came in after their cylinder number
77  * was passed.  Thus we implement a one way scan, retracting after reaching the
78  * end of the drive to the first request on the second queue, at which time it
79  * becomes the first queue.
80  *
81  * A one-way scan is natural because of the way UNIX read-ahead blocks are
82  * allocated.
83  */
84 
85 void
86 disksort(ap, bp)
87 	register struct buf *ap, *bp;
88 {
89 	register struct buf *bq;
90 
91 	/* If the queue is empty, then it's easy. */
92 	if (ap->b_actf == NULL) {
93 		bp->b_actf = NULL;
94 		ap->b_actf = bp;
95 		return;
96 	}
97 
98 	/*
99 	 * If we lie after the first (currently active) request, then we
100 	 * must locate the second request list and add ourselves to it.
101 	 */
102 	bq = ap->b_actf;
103 	if (bp->b_cylinder < bq->b_cylinder) {
104 		while (bq->b_actf) {
105 			/*
106 			 * Check for an ``inversion'' in the normally ascending
107 			 * cylinder numbers, indicating the start of the second
108 			 * request list.
109 			 */
110 			if (bq->b_actf->b_cylinder < bq->b_cylinder) {
111 				/*
112 				 * Search the second request list for the first
113 				 * request at a larger cylinder number.  We go
114 				 * before that; if there is no such request, we
115 				 * go at end.
116 				 */
117 				do {
118 					if (bp->b_cylinder <
119 					    bq->b_actf->b_cylinder)
120 						goto insert;
121 					if (bp->b_cylinder ==
122 					    bq->b_actf->b_cylinder &&
123 					    bp->b_blkno < bq->b_actf->b_blkno)
124 						goto insert;
125 					bq = bq->b_actf;
126 				} while (bq->b_actf);
127 				goto insert;		/* after last */
128 			}
129 			bq = bq->b_actf;
130 		}
131 		/*
132 		 * No inversions... we will go after the last, and
133 		 * be the first request in the second request list.
134 		 */
135 		goto insert;
136 	}
137 	/*
138 	 * Request is at/after the current request...
139 	 * sort in the first request list.
140 	 */
141 	while (bq->b_actf) {
142 		/*
143 		 * We want to go after the current request if there is an
144 		 * inversion after it (i.e. it is the end of the first
145 		 * request list), or if the next request is a larger cylinder
146 		 * than our request.
147 		 */
148 		if (bq->b_actf->b_cylinder < bq->b_cylinder ||
149 		    bp->b_cylinder < bq->b_actf->b_cylinder ||
150 		    (bp->b_cylinder == bq->b_actf->b_cylinder &&
151 		    bp->b_blkno < bq->b_actf->b_blkno))
152 			goto insert;
153 		bq = bq->b_actf;
154 	}
155 	/*
156 	 * Neither a second list nor a larger request... we go at the end of
157 	 * the first list, which is the same as the end of the whole schebang.
158 	 */
159 insert:	bp->b_actf = bq->b_actf;
160 	bq->b_actf = bp;
161 }
162 
163 /*
164  * Compute checksum for disk label.
165  */
166 u_int
167 dkcksum(lp)
168 	register struct disklabel *lp;
169 {
170 	register u_int16_t *start, *end;
171 	register u_int16_t sum = 0;
172 
173 	start = (u_int16_t *)lp;
174 	end = (u_int16_t *)&lp->d_partitions[lp->d_npartitions];
175 	while (start < end)
176 		sum ^= *start++;
177 	return (sum);
178 }
179 
180 /*
181  * Disk error is the preface to plaintive error messages
182  * about failing disk transfers.  It prints messages of the form
183 
184 hp0g: hard error reading fsbn 12345 of 12344-12347 (hp0 bn %d cn %d tn %d sn %d)
185 
186  * if the offset of the error in the transfer and a disk label
187  * are both available.  blkdone should be -1 if the position of the error
188  * is unknown; the disklabel pointer may be null from drivers that have not
189  * been converted to use them.  The message is printed with printf
190  * if pri is LOG_PRINTF, otherwise it uses log at the specified priority.
191  * The message should be completed (with at least a newline) with printf
192  * or addlog, respectively.  There is no trailing space.
193  */
194 void
195 diskerr(bp, dname, what, pri, blkdone, lp)
196 	register struct buf *bp;
197 	char *dname, *what;
198 	int pri, blkdone;
199 	register struct disklabel *lp;
200 {
201 	int unit = DISKUNIT(bp->b_dev), part = DISKPART(bp->b_dev);
202 	register int (*pr) __P((const char *, ...));
203 	char partname = 'a' + part;
204 	int sn;
205 
206 	if (pri != LOG_PRINTF) {
207 		static const char fmt[] = "";
208 		log(pri, fmt);
209 		pr = addlog;
210 	} else
211 		pr = printf;
212 	(*pr)("%s%d%c: %s %sing fsbn ", dname, unit, partname, what,
213 	    bp->b_flags & B_READ ? "read" : "writ");
214 	sn = bp->b_blkno;
215 	if (bp->b_bcount <= DEV_BSIZE)
216 		(*pr)("%d", sn);
217 	else {
218 		if (blkdone >= 0) {
219 			sn += blkdone;
220 			(*pr)("%d of ", sn);
221 		}
222 		(*pr)("%d-%d", bp->b_blkno,
223 		    bp->b_blkno + (bp->b_bcount - 1) / DEV_BSIZE);
224 	}
225 	if (lp && (blkdone >= 0 || bp->b_bcount <= lp->d_secsize)) {
226 #ifdef tahoe
227 		sn *= DEV_BSIZE / lp->d_secsize;		/* XXX */
228 #endif
229 		sn += lp->d_partitions[part].p_offset;
230 		(*pr)(" (%s%d bn %d; cn %d", dname, unit, sn,
231 		    sn / lp->d_secpercyl);
232 		sn %= lp->d_secpercyl;
233 		(*pr)(" tn %d sn %d)", sn / lp->d_nsectors, sn % lp->d_nsectors);
234 	}
235 }
236 
237 /*
238  * Initialize the disklist.  Called by main() before autoconfiguration.
239  */
240 void
241 disk_init()
242 {
243 
244 	TAILQ_INIT(&disklist);
245 	disk_count = 0;
246 }
247 
248 /*
249  * Searches the disklist for the disk corresponding to the
250  * name provided.
251  */
252 struct disk *
253 disk_find(name)
254 	char *name;
255 {
256 	struct disk *diskp;
257 
258 	if ((name == NULL) || (disk_count <= 0))
259 		return (NULL);
260 
261 	for (diskp = disklist.tqh_first; diskp != NULL;
262 	    diskp = diskp->dk_link.tqe_next)
263 		if (strcmp(diskp->dk_name, name) == 0)
264 			return (diskp);
265 
266 	return (NULL);
267 }
268 
269 /*
270  * Attach a disk.
271  */
272 void
273 disk_attach(diskp)
274 	struct disk *diskp;
275 {
276 	int s;
277 
278 	/*
279 	 * Allocate and initialize the disklabel structures.  Note that
280 	 * it's not safe to sleep here, since we're probably going to be
281 	 * called during autoconfiguration.
282 	 */
283 	diskp->dk_label = malloc(sizeof(struct disklabel), M_DEVBUF, M_NOWAIT);
284 	diskp->dk_cpulabel = malloc(sizeof(struct cpu_disklabel), M_DEVBUF,
285 	    M_NOWAIT);
286 	if ((diskp->dk_label == NULL) || (diskp->dk_cpulabel == NULL))
287 		panic("disk_attach: can't allocate storage for disklabel");
288 
289 	bzero(diskp->dk_label, sizeof(struct disklabel));
290 	bzero(diskp->dk_cpulabel, sizeof(struct cpu_disklabel));
291 
292 	/*
293 	 * Set the attached timestamp.
294 	 */
295 	s = splclock();
296 	diskp->dk_attachtime = mono_time;
297 	splx(s);
298 
299 	/*
300 	 * Link into the disklist.
301 	 */
302 	TAILQ_INSERT_TAIL(&disklist, diskp, dk_link);
303 	++disk_count;
304 }
305 
306 /*
307  * Detach a disk.
308  */
309 void
310 disk_detach(diskp)
311 	struct disk *diskp;
312 {
313 
314 	/*
315 	 * Free the space used by the disklabel structures.
316 	 */
317 	free(diskp->dk_label, M_DEVBUF);
318 	free(diskp->dk_cpulabel, M_DEVBUF);
319 
320 	/*
321 	 * Remove from the disklist.
322 	 */
323 	TAILQ_REMOVE(&disklist, diskp, dk_link);
324 	if (--disk_count < 0)
325 		panic("disk_detach: disk_count < 0");
326 }
327 
328 /*
329  * Increment a disk's busy counter.  If the counter is going from
330  * 0 to 1, set the timestamp.
331  */
332 void
333 disk_busy(diskp)
334 	struct disk *diskp;
335 {
336 	int s;
337 
338 	/*
339 	 * XXX We'd like to use something as accurate as microtime(),
340 	 * but that doesn't depend on the system TOD clock.
341 	 */
342 	if (diskp->dk_busy++ == 0) {
343 		s = splclock();
344 		diskp->dk_timestamp = mono_time;
345 		splx(s);
346 	}
347 }
348 
349 /*
350  * Decrement a disk's busy counter, increment the byte count, total busy
351  * time, and reset the timestamp.
352  */
353 void
354 disk_unbusy(diskp, bcount)
355 	struct disk *diskp;
356 	long bcount;
357 {
358 	int s;
359 	struct timeval dv_time, diff_time;
360 
361 	if (diskp->dk_busy-- == 0)
362 		printf("disk_unbusy: %s: dk_busy < 0\n", diskp->dk_name);
363 
364 	s = splclock();
365 	dv_time = mono_time;
366 	splx(s);
367 
368 	timersub(&dv_time, &diskp->dk_timestamp, &diff_time);
369 	timeradd(&diskp->dk_time, &diff_time, &diskp->dk_time);
370 
371 	diskp->dk_timestamp = dv_time;
372 	if (bcount > 0) {
373 		diskp->dk_bytes += bcount;
374 		diskp->dk_xfer++;
375 	}
376 	diskp->dk_seek++;
377 
378 	add_disk_randomness(bcount ^ diff_time.tv_usec);
379 }
380 
381 /*
382  * Reset the metrics counters on the given disk.  Note that we cannot
383  * reset the busy counter, as it may case a panic in disk_unbusy().
384  * We also must avoid playing with the timestamp information, as it
385  * may skew any pending transfer results.
386  */
387 void
388 disk_resetstat(diskp)
389 	struct disk *diskp;
390 {
391 	int s = splbio(), t;
392 
393 	diskp->dk_xfer = 0;
394 	diskp->dk_bytes = 0;
395 	diskp->dk_seek = 0;
396 
397 	t = splclock();
398 	diskp->dk_attachtime = mono_time;
399 	splx(t);
400 
401 	timerclear(&diskp->dk_time);
402 
403 	splx(s);
404 }
405 
406 
407 int
408 dk_mountroot()
409 {
410 	dev_t rawdev, rrootdev;
411 	int part = DISKPART(rootdev);
412 	int (*mountrootfn) __P((void));
413 	extern struct proc *curproc;
414 	struct disklabel dl;
415 	int error;
416 
417 	rrootdev = blktochr(rootdev);
418 	rawdev = MAKEDISKDEV(major(rrootdev), DISKUNIT(rootdev), RAW_PART);
419 	printf("rootdev=0x%x rrootdev=0x%x rawdev=0x%x\n", rootdev,
420 	    rrootdev, rawdev);
421 
422 	/*
423 	 * open device, ioctl for the disklabel, and close it.
424 	 */
425 	error = (cdevsw[major(rrootdev)].d_open)(rawdev, FREAD,
426 	    S_IFCHR, curproc);
427 	if (error)
428 		panic("cannot open disk, 0x%x/0x%x, error %d",
429 		    rootdev, rrootdev, error);
430 	error = (cdevsw[major(rrootdev)].d_ioctl)(rawdev, DIOCGDINFO,
431 	    (caddr_t)&dl, FREAD, curproc);
432 	if (error)
433 		panic("cannot read disk label, 0x%x/0x%x, error %d",
434 		    rootdev, rrootdev, error);
435 	(void) (cdevsw[major(rrootdev)].d_close)(rawdev, FREAD,
436 	    S_IFCHR, curproc);
437 
438 	if (dl.d_partitions[part].p_size == 0)
439 		panic("root filesystem has size 0");
440 	switch (dl.d_partitions[part].p_fstype) {
441 #ifdef EXT2FS
442 	case FS_EXT2FS:
443 		{
444 		extern int ext2fs_mountroot __P((void));
445 		mountrootfn = ext2fs_mountroot;
446 		}
447 		break;
448 #endif
449 #ifdef FFS
450 	case FS_BSDFFS:
451 		{
452 		extern int ffs_mountroot __P((void));
453 		mountrootfn = ffs_mountroot;
454 		}
455 		break;
456 #endif
457 #ifdef LFS
458 	case FS_BSDLFS:
459 		{
460 		extern int lfs_mountroot __P((void));
461 		mountrootfn = lfs_mountroot;
462 		}
463 		break;
464 #endif
465 #ifdef CD9660
466 	case FS_ISO9660:
467 		{
468 		extern int cd9660_mountroot __P((void));
469 		mountrootfn = cd9660_mountroot;
470 		}
471 		break;
472 #endif
473 	default:
474 #ifdef FFS
475 		{
476 		extern int ffs_mountroot __P((void));
477 
478 		printf("filesystem type %d not known.. assuming ffs\n",
479 		    dl.d_partitions[part].p_fstype);
480 		mountrootfn = ffs_mountroot;
481 		}
482 #else
483 		panic("disk 0x%x/0x%x filesystem type %d not known",
484 		    rootdev, rrootdev, dl.d_partitions[part].p_fstype);
485 #endif
486 	}
487 	return (*mountrootfn)();
488 }
489