xref: /netbsd-src/sys/kern/subr_disk.c (revision 27578b9aac214cc7796ead81dcc5427e79d5f2a0)
1 /*	$NetBSD: subr_disk.c,v 1.30 2001/07/09 10:54:12 simonb Exp $	*/
2 
3 /*-
4  * Copyright (c) 1996, 1997, 1999, 2000 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
9  * NASA Ames Research Center.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. All advertising materials mentioning features or use of this software
20  *    must display the following acknowledgement:
21  *	This product includes software developed by the NetBSD
22  *	Foundation, Inc. and its contributors.
23  * 4. Neither the name of The NetBSD Foundation nor the names of its
24  *    contributors may be used to endorse or promote products derived
25  *    from this software without specific prior written permission.
26  *
27  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
28  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
29  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
30  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
31  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
32  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
33  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
34  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
35  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
36  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
37  * POSSIBILITY OF SUCH DAMAGE.
38  */
39 
40 /*
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.
48  *
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.
64  *
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.
76  *
77  *	@(#)ufs_disksubr.c	8.5 (Berkeley) 1/21/94
78  */
79 
80 #include <sys/param.h>
81 #include <sys/systm.h>
82 #include <sys/kernel.h>
83 #include <sys/malloc.h>
84 #include <sys/buf.h>
85 #include <sys/syslog.h>
86 #include <sys/time.h>
87 #include <sys/disklabel.h>
88 #include <sys/disk.h>
89 
90 /*
91  * A global list of all disks attached to the system.  May grow or
92  * shrink over time.
93  */
94 struct	disklist_head disklist;	/* TAILQ_HEAD */
95 int	disk_count;		/* number of drives in global disklist */
96 
97 /*
98  * Seek sort for disks.  We depend on the driver which calls us using b_resid
99  * as the current cylinder number.
100  *
101  * The argument bufq is an I/O queue for the device, on which there are
102  * actually two queues, sorted in ascending cylinder order.  The first
103  * queue holds those requests which are positioned after the current
104  * cylinder (in the first request); the second holds requests which came
105  * in after their cylinder number was passed.  Thus we implement a one-way
106  * scan, retracting after reaching the end of the drive to the first request
107  * on the second queue, at which time it becomes the first queue.
108  *
109  * A one-way scan is natural because of the way UNIX read-ahead blocks are
110  * allocated.
111  *
112  * This is further adjusted by any `barriers' which may exist in the queue.
113  * The bufq points to the last such ordered request.
114  */
115 void
116 disksort_cylinder(struct buf_queue *bufq, struct buf *bp)
117 {
118 	struct buf *bq, *nbq;
119 
120 	/*
121 	 * If there are ordered requests on the queue, we must start
122 	 * the elevator sort after the last of these.
123 	 */
124 	if ((bq = bufq->bq_barrier) == NULL)
125 		bq = BUFQ_FIRST(bufq);
126 
127 	/*
128 	 * If the queue is empty, of if it's an ordered request,
129 	 * it's easy; we just go on the end.
130 	 */
131 	if (bq == NULL || (bp->b_flags & B_ORDERED) != 0) {
132 		BUFQ_INSERT_TAIL(bufq, bp);
133 		return;
134 	}
135 
136 	/*
137 	 * If we lie after the first (currently active) request, then we
138 	 * must locate the second request list and add ourselves to it.
139 	 */
140 	if (bp->b_cylinder < bq->b_cylinder ||
141 	    (bp->b_cylinder == bq->b_cylinder &&
142 	     bp->b_rawblkno < bq->b_rawblkno)) {
143 		while ((nbq = BUFQ_NEXT(bq)) != NULL) {
144 			/*
145 			 * Check for an ``inversion'' in the normally ascending
146 			 * cylinder numbers, indicating the start of the second
147 			 * request list.
148 			 */
149 			if (nbq->b_cylinder < bq->b_cylinder) {
150 				/*
151 				 * Search the second request list for the first
152 				 * request at a larger cylinder number.  We go
153 				 * before that; if there is no such request, we
154 				 * go at end.
155 				 */
156 				do {
157 					if (bp->b_cylinder < nbq->b_cylinder)
158 						goto insert;
159 					if (bp->b_cylinder == nbq->b_cylinder &&
160 					    bp->b_rawblkno < nbq->b_rawblkno)
161 						goto insert;
162 					bq = nbq;
163 				} while ((nbq = BUFQ_NEXT(bq)) != NULL);
164 				goto insert;		/* after last */
165 			}
166 			bq = BUFQ_NEXT(bq);
167 		}
168 		/*
169 		 * No inversions... we will go after the last, and
170 		 * be the first request in the second request list.
171 		 */
172 		goto insert;
173 	}
174 	/*
175 	 * Request is at/after the current request...
176 	 * sort in the first request list.
177 	 */
178 	while ((nbq = BUFQ_NEXT(bq)) != NULL) {
179 		/*
180 		 * We want to go after the current request if there is an
181 		 * inversion after it (i.e. it is the end of the first
182 		 * request list), or if the next request is a larger cylinder
183 		 * than our request.
184 		 */
185 		if (nbq->b_cylinder < bq->b_cylinder ||
186 		    bp->b_cylinder < nbq->b_cylinder ||
187 		    (bp->b_cylinder == nbq->b_cylinder &&
188 		     bp->b_rawblkno < nbq->b_rawblkno))
189 			goto insert;
190 		bq = nbq;
191 	}
192 	/*
193 	 * Neither a second list nor a larger request... we go at the end of
194 	 * the first list, which is the same as the end of the whole schebang.
195 	 */
196 insert:	BUFQ_INSERT_AFTER(bufq, bq, bp);
197 }
198 
199 /*
200  * Seek sort for disks.  This version sorts based on b_rawblkno, which
201  * indicates the block number.
202  *
203  * As before, there are actually two queues, sorted in ascendening block
204  * order.  The first queue holds those requests which are positioned after
205  * the current block (in the first request); the second holds requests which
206  * came in after their block number was passed.  Thus we implement a one-way
207  * scan, retracting after reaching the end of the driver to the first request
208  * on the second queue, at which time it becomes the first queue.
209  *
210  * A one-way scan is natural because of the way UNIX read-ahead blocks are
211  * allocated.
212  *
213  * This is further adjusted by any `barriers' which may exist in the queue.
214  * The bufq points to the last such ordered request.
215  */
216 void
217 disksort_blkno(struct buf_queue *bufq, struct buf *bp)
218 {
219 	struct buf *bq, *nbq;
220 
221 	/*
222 	 * If there are ordered requests on the queue, we must start
223 	 * the elevator sort after the last of these.
224 	 */
225 	if ((bq = bufq->bq_barrier) == NULL)
226 		bq = BUFQ_FIRST(bufq);
227 
228 	/*
229 	 * If the queue is empty, or if it's an ordered request,
230 	 * it's easy; we just go on the end.
231 	 */
232 	if (bq == NULL || (bp->b_flags & B_ORDERED) != 0) {
233 		BUFQ_INSERT_TAIL(bufq, bp);
234 		return;
235 	}
236 
237 	/*
238 	 * If we lie after the first (currently active) request, then we
239 	 * must locate the second request list and add ourselves to it.
240 	 */
241 	if (bp->b_rawblkno < bq->b_rawblkno) {
242 		while ((nbq = BUFQ_NEXT(bq)) != NULL) {
243 			/*
244 			 * Check for an ``inversion'' in the normally ascending
245 			 * block numbers, indicating the start of the second
246 			 * request list.
247 			 */
248 			if (nbq->b_rawblkno < bq->b_rawblkno) {
249 				/*
250 				 * Search the second request list for the first
251 				 * request at a larger block number.  We go
252 				 * after that; if there is no such request, we
253 				 * go at the end.
254 				 */
255 				do {
256 					if (bp->b_rawblkno < nbq->b_rawblkno)
257 						goto insert;
258 					bq = nbq;
259 				} while ((nbq = BUFQ_NEXT(bq)) != NULL);
260 				goto insert;		/* after last */
261 			}
262 			bq = BUFQ_NEXT(bq);
263 		}
264 		/*
265 		 * No inversions... we will go after the last, and
266 		 * be the first request in the second request list.
267 		 */
268 		goto insert;
269 	}
270 	/*
271 	 * Request is at/after the current request...
272 	 * sort in the first request list.
273 	 */
274 	while ((nbq = BUFQ_NEXT(bq)) != NULL) {
275 		/*
276 		 * We want to go after the current request if there is an
277 		 * inversion after it (i.e. it is the end of the first
278 		 * request list), or if the next request is a larger cylinder
279 		 * than our request.
280 		 */
281 		if (nbq->b_rawblkno < bq->b_rawblkno ||
282 		    bp->b_rawblkno < nbq->b_rawblkno)
283 			goto insert;
284 		bq = nbq;
285 	}
286 	/*
287 	 * Neither a second list nor a larger request... we go at the end of
288 	 * the first list, which is the same as the end of the whole schebang.
289 	 */
290 insert:	BUFQ_INSERT_AFTER(bufq, bq, bp);
291 }
292 
293 /*
294  * Seek non-sort for disks.  This version simply inserts requests at
295  * the tail of the queue.
296  */
297 void
298 disksort_tail(struct buf_queue *bufq, struct buf *bp)
299 {
300 
301 	BUFQ_INSERT_TAIL(bufq, bp);
302 }
303 
304 /*
305  * Compute checksum for disk label.
306  */
307 u_int
308 dkcksum(struct disklabel *lp)
309 {
310 	u_short *start, *end;
311 	u_short sum = 0;
312 
313 	start = (u_short *)lp;
314 	end = (u_short *)&lp->d_partitions[lp->d_npartitions];
315 	while (start < end)
316 		sum ^= *start++;
317 	return (sum);
318 }
319 
320 /*
321  * Disk error is the preface to plaintive error messages
322  * about failing disk transfers.  It prints messages of the form
323 
324 hp0g: hard error reading fsbn 12345 of 12344-12347 (hp0 bn %d cn %d tn %d sn %d)
325 
326  * if the offset of the error in the transfer and a disk label
327  * are both available.  blkdone should be -1 if the position of the error
328  * is unknown; the disklabel pointer may be null from drivers that have not
329  * been converted to use them.  The message is printed with printf
330  * if pri is LOG_PRINTF, otherwise it uses log at the specified priority.
331  * The message should be completed (with at least a newline) with printf
332  * or addlog, respectively.  There is no trailing space.
333  */
334 void
335 diskerr(struct buf *bp, char *dname, char *what, int pri, int blkdone,
336     struct disklabel *lp)
337 {
338 	int unit = DISKUNIT(bp->b_dev), part = DISKPART(bp->b_dev);
339 	void (*pr)(const char *, ...);
340 	char partname = 'a' + part;
341 	int sn;
342 
343 	if (pri != LOG_PRINTF) {
344 		static const char fmt[] = "";
345 		log(pri, fmt);
346 		pr = addlog;
347 	} else
348 		pr = printf;
349 	(*pr)("%s%d%c: %s %sing fsbn ", dname, unit, partname, what,
350 	    bp->b_flags & B_READ ? "read" : "writ");
351 	sn = bp->b_blkno;
352 	if (bp->b_bcount <= DEV_BSIZE)
353 		(*pr)("%d", sn);
354 	else {
355 		if (blkdone >= 0) {
356 			sn += blkdone;
357 			(*pr)("%d of ", sn);
358 		}
359 		(*pr)("%d-%d", bp->b_blkno,
360 		    bp->b_blkno + (bp->b_bcount - 1) / DEV_BSIZE);
361 	}
362 	if (lp && (blkdone >= 0 || bp->b_bcount <= lp->d_secsize)) {
363 		sn += lp->d_partitions[part].p_offset;
364 		(*pr)(" (%s%d bn %d; cn %d", dname, unit, sn,
365 		    sn / lp->d_secpercyl);
366 		sn %= lp->d_secpercyl;
367 		(*pr)(" tn %d sn %d)", sn / lp->d_nsectors, sn % lp->d_nsectors);
368 	}
369 }
370 
371 /*
372  * Initialize the disklist.  Called by main() before autoconfiguration.
373  */
374 void
375 disk_init(void)
376 {
377 
378 	TAILQ_INIT(&disklist);
379 	disk_count = 0;
380 }
381 
382 /*
383  * Searches the disklist for the disk corresponding to the
384  * name provided.
385  */
386 struct disk *
387 disk_find(char *name)
388 {
389 	struct disk *diskp;
390 
391 	if ((name == NULL) || (disk_count <= 0))
392 		return (NULL);
393 
394 	for (diskp = disklist.tqh_first; diskp != NULL;
395 	    diskp = diskp->dk_link.tqe_next)
396 		if (strcmp(diskp->dk_name, name) == 0)
397 			return (diskp);
398 
399 	return (NULL);
400 }
401 
402 /*
403  * Attach a disk.
404  */
405 void
406 disk_attach(struct disk *diskp)
407 {
408 	int s;
409 
410 	/*
411 	 * Allocate and initialize the disklabel structures.  Note that
412 	 * it's not safe to sleep here, since we're probably going to be
413 	 * called during autoconfiguration.
414 	 */
415 	diskp->dk_label = malloc(sizeof(struct disklabel), M_DEVBUF, M_NOWAIT);
416 	diskp->dk_cpulabel = malloc(sizeof(struct cpu_disklabel), M_DEVBUF,
417 	    M_NOWAIT);
418 	if ((diskp->dk_label == NULL) || (diskp->dk_cpulabel == NULL))
419 		panic("disk_attach: can't allocate storage for disklabel");
420 
421 	memset(diskp->dk_label, 0, sizeof(struct disklabel));
422 	memset(diskp->dk_cpulabel, 0, sizeof(struct cpu_disklabel));
423 
424 	/*
425 	 * Set the attached timestamp.
426 	 */
427 	s = splclock();
428 	diskp->dk_attachtime = mono_time;
429 	splx(s);
430 
431 	/*
432 	 * Link into the disklist.
433 	 */
434 	TAILQ_INSERT_TAIL(&disklist, diskp, dk_link);
435 	++disk_count;
436 }
437 
438 /*
439  * Detach a disk.
440  */
441 void
442 disk_detach(struct disk *diskp)
443 {
444 
445 	/*
446 	 * Remove from the disklist.
447 	 */
448 	if (--disk_count < 0)
449 		panic("disk_detach: disk_count < 0");
450 	TAILQ_REMOVE(&disklist, diskp, dk_link);
451 
452 	/*
453 	 * Free the space used by the disklabel structures.
454 	 */
455 	free(diskp->dk_label, M_DEVBUF);
456 	free(diskp->dk_cpulabel, M_DEVBUF);
457 }
458 
459 /*
460  * Increment a disk's busy counter.  If the counter is going from
461  * 0 to 1, set the timestamp.
462  */
463 void
464 disk_busy(struct disk *diskp)
465 {
466 	int s;
467 
468 	/*
469 	 * XXX We'd like to use something as accurate as microtime(),
470 	 * but that doesn't depend on the system TOD clock.
471 	 */
472 	if (diskp->dk_busy++ == 0) {
473 		s = splclock();
474 		diskp->dk_timestamp = mono_time;
475 		splx(s);
476 	}
477 }
478 
479 /*
480  * Decrement a disk's busy counter, increment the byte count, total busy
481  * time, and reset the timestamp.
482  */
483 void
484 disk_unbusy(struct disk *diskp, long bcount)
485 {
486 	int s;
487 	struct timeval dv_time, diff_time;
488 
489 	if (diskp->dk_busy-- == 0) {
490 		printf("%s: dk_busy < 0\n", diskp->dk_name);
491 		panic("disk_unbusy");
492 	}
493 
494 	s = splclock();
495 	dv_time = mono_time;
496 	splx(s);
497 
498 	timersub(&dv_time, &diskp->dk_timestamp, &diff_time);
499 	timeradd(&diskp->dk_time, &diff_time, &diskp->dk_time);
500 
501 	diskp->dk_timestamp = dv_time;
502 	if (bcount > 0) {
503 		diskp->dk_bytes += bcount;
504 		diskp->dk_xfer++;
505 	}
506 }
507 
508 /*
509  * Reset the metrics counters on the given disk.  Note that we cannot
510  * reset the busy counter, as it may case a panic in disk_unbusy().
511  * We also must avoid playing with the timestamp information, as it
512  * may skew any pending transfer results.
513  */
514 void
515 disk_resetstat(struct disk *diskp)
516 {
517 	int s = splbio(), t;
518 
519 	diskp->dk_xfer = 0;
520 	diskp->dk_bytes = 0;
521 
522 	t = splclock();
523 	diskp->dk_attachtime = mono_time;
524 	splx(t);
525 
526 	timerclear(&diskp->dk_time);
527 
528 	splx(s);
529 }
530