xref: /netbsd-src/sys/kern/subr_disk.c (revision 001c68bd94f75ce9270b69227c4199fbf34ee396)
1 /*	$NetBSD: subr_disk.c,v 1.52 2003/04/13 09:08:04 dsl 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/cdefs.h>
81 __KERNEL_RCSID(0, "$NetBSD: subr_disk.c,v 1.52 2003/04/13 09:08:04 dsl Exp $");
82 
83 #include "opt_compat_netbsd.h"
84 
85 #include <sys/param.h>
86 #include <sys/kernel.h>
87 #include <sys/malloc.h>
88 #include <sys/buf.h>
89 #include <sys/syslog.h>
90 #include <sys/disklabel.h>
91 #include <sys/disk.h>
92 #include <sys/sysctl.h>
93 #include <lib/libkern/libkern.h>
94 
95 /*
96  * A global list of all disks attached to the system.  May grow or
97  * shrink over time.
98  */
99 struct	disklist_head disklist;	/* TAILQ_HEAD */
100 int	disk_count;		/* number of drives in global disklist */
101 struct simplelock disklist_slock = SIMPLELOCK_INITIALIZER;
102 
103 /*
104  * Compute checksum for disk label.
105  */
106 u_int
107 dkcksum(struct disklabel *lp)
108 {
109 	u_short *start, *end;
110 	u_short sum = 0;
111 
112 	start = (u_short *)lp;
113 	end = (u_short *)&lp->d_partitions[lp->d_npartitions];
114 	while (start < end)
115 		sum ^= *start++;
116 	return (sum);
117 }
118 
119 /*
120  * Disk error is the preface to plaintive error messages
121  * about failing disk transfers.  It prints messages of the form
122 
123 hp0g: hard error reading fsbn 12345 of 12344-12347 (hp0 bn %d cn %d tn %d sn %d)
124 
125  * if the offset of the error in the transfer and a disk label
126  * are both available.  blkdone should be -1 if the position of the error
127  * is unknown; the disklabel pointer may be null from drivers that have not
128  * been converted to use them.  The message is printed with printf
129  * if pri is LOG_PRINTF, otherwise it uses log at the specified priority.
130  * The message should be completed (with at least a newline) with printf
131  * or addlog, respectively.  There is no trailing space.
132  */
133 #ifndef PRIdaddr
134 #define PRIdaddr PRId64
135 #endif
136 void
137 diskerr(const struct buf *bp, const char *dname, const char *what, int pri,
138     int blkdone, const struct disklabel *lp)
139 {
140 	int unit = DISKUNIT(bp->b_dev), part = DISKPART(bp->b_dev);
141 	void (*pr)(const char *, ...);
142 	char partname = 'a' + part;
143 	daddr_t sn;
144 
145 	if (/*CONSTCOND*/0)
146 		/* Compiler will error this is the format is wrong... */
147 		printf("%" PRIdaddr, bp->b_blkno);
148 
149 	if (pri != LOG_PRINTF) {
150 		static const char fmt[] = "";
151 		log(pri, fmt);
152 		pr = addlog;
153 	} else
154 		pr = printf;
155 	(*pr)("%s%d%c: %s %sing fsbn ", dname, unit, partname, what,
156 	    bp->b_flags & B_READ ? "read" : "writ");
157 	sn = bp->b_blkno;
158 	if (bp->b_bcount <= DEV_BSIZE)
159 		(*pr)("%" PRIdaddr, sn);
160 	else {
161 		if (blkdone >= 0) {
162 			sn += blkdone;
163 			(*pr)("%" PRIdaddr " of ", sn);
164 		}
165 		(*pr)("%" PRIdaddr "-%" PRIdaddr "", bp->b_blkno,
166 		    bp->b_blkno + (bp->b_bcount - 1) / DEV_BSIZE);
167 	}
168 	if (lp && (blkdone >= 0 || bp->b_bcount <= lp->d_secsize)) {
169 		sn += lp->d_partitions[part].p_offset;
170 		(*pr)(" (%s%d bn %" PRIdaddr "; cn %" PRIdaddr "",
171 		    dname, unit, sn, sn / lp->d_secpercyl);
172 		sn %= lp->d_secpercyl;
173 		(*pr)(" tn %" PRIdaddr " sn %" PRIdaddr ")",
174 		    sn / lp->d_nsectors, sn % lp->d_nsectors);
175 	}
176 }
177 
178 /*
179  * Initialize the disklist.  Called by main() before autoconfiguration.
180  */
181 void
182 disk_init(void)
183 {
184 
185 	TAILQ_INIT(&disklist);
186 	disk_count = 0;
187 }
188 
189 /*
190  * Searches the disklist for the disk corresponding to the
191  * name provided.
192  */
193 struct disk *
194 disk_find(char *name)
195 {
196 	struct disk *diskp;
197 
198 	if ((name == NULL) || (disk_count <= 0))
199 		return (NULL);
200 
201 	simple_lock(&disklist_slock);
202 	for (diskp = TAILQ_FIRST(&disklist); diskp != NULL;
203 	    diskp = TAILQ_NEXT(diskp, dk_link))
204 		if (strcmp(diskp->dk_name, name) == 0) {
205 			simple_unlock(&disklist_slock);
206 			return (diskp);
207 		}
208 	simple_unlock(&disklist_slock);
209 
210 	return (NULL);
211 }
212 
213 /*
214  * Attach a disk.
215  */
216 void
217 disk_attach(struct disk *diskp)
218 {
219 	int s;
220 
221 	/*
222 	 * Allocate and initialize the disklabel structures.  Note that
223 	 * it's not safe to sleep here, since we're probably going to be
224 	 * called during autoconfiguration.
225 	 */
226 	diskp->dk_label = malloc(sizeof(struct disklabel), M_DEVBUF, M_NOWAIT);
227 	diskp->dk_cpulabel = malloc(sizeof(struct cpu_disklabel), M_DEVBUF,
228 	    M_NOWAIT);
229 	if ((diskp->dk_label == NULL) || (diskp->dk_cpulabel == NULL))
230 		panic("disk_attach: can't allocate storage for disklabel");
231 
232 	memset(diskp->dk_label, 0, sizeof(struct disklabel));
233 	memset(diskp->dk_cpulabel, 0, sizeof(struct cpu_disklabel));
234 
235 	/*
236 	 * Set the attached timestamp.
237 	 */
238 	s = splclock();
239 	diskp->dk_attachtime = mono_time;
240 	splx(s);
241 
242 	/*
243 	 * Link into the disklist.
244 	 */
245 	simple_lock(&disklist_slock);
246 	TAILQ_INSERT_TAIL(&disklist, diskp, dk_link);
247 	simple_unlock(&disklist_slock);
248 	++disk_count;
249 }
250 
251 /*
252  * Detach a disk.
253  */
254 void
255 disk_detach(struct disk *diskp)
256 {
257 
258 	/*
259 	 * Remove from the disklist.
260 	 */
261 	if (--disk_count < 0)
262 		panic("disk_detach: disk_count < 0");
263 	simple_lock(&disklist_slock);
264 	TAILQ_REMOVE(&disklist, diskp, dk_link);
265 	simple_unlock(&disklist_slock);
266 
267 	/*
268 	 * Free the space used by the disklabel structures.
269 	 */
270 	free(diskp->dk_label, M_DEVBUF);
271 	free(diskp->dk_cpulabel, M_DEVBUF);
272 }
273 
274 /*
275  * Increment a disk's busy counter.  If the counter is going from
276  * 0 to 1, set the timestamp.
277  */
278 void
279 disk_busy(struct disk *diskp)
280 {
281 	int s;
282 
283 	/*
284 	 * XXX We'd like to use something as accurate as microtime(),
285 	 * but that doesn't depend on the system TOD clock.
286 	 */
287 	if (diskp->dk_busy++ == 0) {
288 		s = splclock();
289 		diskp->dk_timestamp = mono_time;
290 		splx(s);
291 	}
292 }
293 
294 /*
295  * Decrement a disk's busy counter, increment the byte count, total busy
296  * time, and reset the timestamp.
297  */
298 void
299 disk_unbusy(struct disk *diskp, long bcount, int read)
300 {
301 	int s;
302 	struct timeval dv_time, diff_time;
303 
304 	if (diskp->dk_busy-- == 0) {
305 		printf("%s: dk_busy < 0\n", diskp->dk_name);
306 		panic("disk_unbusy");
307 	}
308 
309 	s = splclock();
310 	dv_time = mono_time;
311 	splx(s);
312 
313 	timersub(&dv_time, &diskp->dk_timestamp, &diff_time);
314 	timeradd(&diskp->dk_time, &diff_time, &diskp->dk_time);
315 
316 	diskp->dk_timestamp = dv_time;
317 	if (bcount > 0) {
318 		if (read) {
319 			diskp->dk_rbytes += bcount;
320 			diskp->dk_rxfer++;
321 		} else {
322 			diskp->dk_wbytes += bcount;
323 			diskp->dk_wxfer++;
324 		}
325 	}
326 }
327 
328 /*
329  * Reset the metrics counters on the given disk.  Note that we cannot
330  * reset the busy counter, as it may case a panic in disk_unbusy().
331  * We also must avoid playing with the timestamp information, as it
332  * may skew any pending transfer results.
333  */
334 void
335 disk_resetstat(struct disk *diskp)
336 {
337 	int s = splbio(), t;
338 
339 	diskp->dk_rxfer = 0;
340 	diskp->dk_rbytes = 0;
341 	diskp->dk_wxfer = 0;
342 	diskp->dk_wbytes = 0;
343 
344 	t = splclock();
345 	diskp->dk_attachtime = mono_time;
346 	splx(t);
347 
348 	timerclear(&diskp->dk_time);
349 
350 	splx(s);
351 }
352 
353 int
354 sysctl_disknames(void *vwhere, size_t *sizep)
355 {
356 	char buf[DK_DISKNAMELEN + 1];
357 	char *where = vwhere;
358 	struct disk *diskp;
359 	size_t needed, left, slen;
360 	int error, first;
361 
362 	first = 1;
363 	error = 0;
364 	needed = 0;
365 	left = *sizep;
366 
367 	simple_lock(&disklist_slock);
368 	for (diskp = TAILQ_FIRST(&disklist); diskp != NULL;
369 	    diskp = TAILQ_NEXT(diskp, dk_link)) {
370 		if (where == NULL)
371 			needed += strlen(diskp->dk_name) + 1;
372 		else {
373 			memset(buf, 0, sizeof(buf));
374 			if (first) {
375 				strncpy(buf, diskp->dk_name, sizeof(buf));
376 				first = 0;
377 			} else {
378 				buf[0] = ' ';
379 				strncpy(buf + 1, diskp->dk_name,
380 				    sizeof(buf) - 1);
381 			}
382 			buf[DK_DISKNAMELEN] = '\0';
383 			slen = strlen(buf);
384 			if (left < slen + 1)
385 				break;
386 			/* +1 to copy out the trailing NUL byte */
387 			error = copyout(buf, where, slen + 1);
388 			if (error)
389 				break;
390 			where += slen;
391 			needed += slen;
392 			left -= slen;
393 		}
394 	}
395 	simple_unlock(&disklist_slock);
396 	*sizep = needed;
397 	return (error);
398 }
399 
400 int
401 sysctl_diskstats(int *name, u_int namelen, void *vwhere, size_t *sizep)
402 {
403 	struct disk_sysctl sdisk;
404 	struct disk *diskp;
405 	char *where = vwhere;
406 	size_t tocopy, left;
407 	int error;
408 
409 	/*
410 	 * The original hw.diskstats call was broken and did not require
411 	 * the userland to pass in it's size of struct disk_sysctl.  This
412 	 * was fixed after NetBSD 1.6 was released, and any applications
413 	 * that do not pass in the size are given an error only, unless
414 	 * we care about 1.6 compatibility.
415 	 */
416 	if (namelen == 0)
417 #ifdef COMPAT_16
418 		tocopy = offsetof(struct disk_sysctl, dk_rxfer);
419 #else
420 		return (EINVAL);
421 #endif
422 	else
423 		tocopy = name[0];
424 
425 	if (where == NULL) {
426 		*sizep = disk_count * tocopy;
427 		return (0);
428 	}
429 
430 	error = 0;
431 	left = *sizep;
432 	memset(&sdisk, 0, sizeof(sdisk));
433 	*sizep = 0;
434 
435 	simple_lock(&disklist_slock);
436 	TAILQ_FOREACH(diskp, &disklist, dk_link) {
437 		if (left < tocopy)
438 			break;
439 		strncpy(sdisk.dk_name, diskp->dk_name, sizeof(sdisk.dk_name));
440 		sdisk.dk_xfer = diskp->dk_rxfer + diskp->dk_wxfer;
441 		sdisk.dk_rxfer = diskp->dk_rxfer;
442 		sdisk.dk_wxfer = diskp->dk_wxfer;
443 		sdisk.dk_seek = diskp->dk_seek;
444 		sdisk.dk_bytes = diskp->dk_rbytes + diskp->dk_wbytes;
445 		sdisk.dk_rbytes = diskp->dk_rbytes;
446 		sdisk.dk_wbytes = diskp->dk_wbytes;
447 		sdisk.dk_attachtime_sec = diskp->dk_attachtime.tv_sec;
448 		sdisk.dk_attachtime_usec = diskp->dk_attachtime.tv_usec;
449 		sdisk.dk_timestamp_sec = diskp->dk_timestamp.tv_sec;
450 		sdisk.dk_timestamp_usec = diskp->dk_timestamp.tv_usec;
451 		sdisk.dk_time_sec = diskp->dk_time.tv_sec;
452 		sdisk.dk_time_usec = diskp->dk_time.tv_usec;
453 		sdisk.dk_busy = diskp->dk_busy;
454 
455 		error = copyout(&sdisk, where, min(tocopy, sizeof(sdisk)));
456 		if (error)
457 			break;
458 		where += tocopy;
459 		*sizep += tocopy;
460 		left -= tocopy;
461 	}
462 	simple_unlock(&disklist_slock);
463 	return (error);
464 }
465 
466 struct bufq_fcfs {
467 	TAILQ_HEAD(, buf) bq_head;	/* actual list of buffers */
468 };
469 
470 struct bufq_disksort {
471 	TAILQ_HEAD(, buf) bq_head;	/* actual list of buffers */
472 };
473 
474 #define PRIO_READ_BURST		48
475 #define PRIO_WRITE_REQ		16
476 
477 struct bufq_prio {
478 	TAILQ_HEAD(, buf) bq_read, bq_write; /* actual list of buffers */
479 	struct buf *bq_write_next;	/* next request in bq_write */
480 	struct buf *bq_next;		/* current request */
481 	int bq_read_burst;		/* # of consecutive reads */
482 };
483 
484 
485 /*
486  * Check if two buf's are in ascending order.
487  */
488 static __inline int
489 buf_inorder(struct buf *bp, struct buf *bq, int sortby)
490 {
491 	int r;
492 
493 	if (bp == NULL || bq == NULL)
494 		return (bq == NULL);
495 
496 	if (sortby == BUFQ_SORT_CYLINDER)
497 		r = bp->b_cylinder - bq->b_cylinder;
498 	else
499 		r = 0;
500 
501 	if (r == 0)
502 		r = bp->b_rawblkno - bq->b_rawblkno;
503 
504 	return (r <= 0);
505 }
506 
507 
508 /*
509  * First-come first-served sort for disks.
510  *
511  * Requests are appended to the queue without any reordering.
512  */
513 static void
514 bufq_fcfs_put(struct bufq_state *bufq, struct buf *bp)
515 {
516 	struct bufq_fcfs *fcfs = bufq->bq_private;
517 
518 	TAILQ_INSERT_TAIL(&fcfs->bq_head, bp, b_actq);
519 }
520 
521 static struct buf *
522 bufq_fcfs_get(struct bufq_state *bufq, int remove)
523 {
524 	struct bufq_fcfs *fcfs = bufq->bq_private;
525 	struct buf *bp;
526 
527 	bp = TAILQ_FIRST(&fcfs->bq_head);
528 
529 	if (bp != NULL && remove)
530 		TAILQ_REMOVE(&fcfs->bq_head, bp, b_actq);
531 
532 	return (bp);
533 }
534 
535 
536 /*
537  * Seek sort for disks.
538  *
539  * There are actually two queues, sorted in ascendening order.  The first
540  * queue holds those requests which are positioned after the current block;
541  * the second holds requests which came in after their position was passed.
542  * Thus we implement a one-way scan, retracting after reaching the end of
543  * the drive to the first request on the second queue, at which time it
544  * becomes the first queue.
545  *
546  * A one-way scan is natural because of the way UNIX read-ahead blocks are
547  * allocated.
548  */
549 static void
550 bufq_disksort_put(struct bufq_state *bufq, struct buf *bp)
551 {
552 	struct bufq_disksort *disksort = bufq->bq_private;
553 	struct buf *bq, *nbq;
554 	int sortby;
555 
556 	sortby = bufq->bq_flags & BUFQ_SORT_MASK;
557 
558 	bq = TAILQ_FIRST(&disksort->bq_head);
559 
560 	/*
561 	 * If the queue is empty it's easy; we just go on the end.
562 	 */
563 	if (bq == NULL) {
564 		TAILQ_INSERT_TAIL(&disksort->bq_head, bp, b_actq);
565 		return;
566 	}
567 
568 	/*
569 	 * If we lie before the currently active request, then we
570 	 * must locate the second request list and add ourselves to it.
571 	 */
572 	if (buf_inorder(bp, bq, sortby)) {
573 		while ((nbq = TAILQ_NEXT(bq, b_actq)) != NULL) {
574 			/*
575 			 * Check for an ``inversion'' in the normally ascending
576 			 * block numbers, indicating the start of the second
577 			 * request list.
578 			 */
579 			if (buf_inorder(nbq, bq, sortby)) {
580 				/*
581 				 * Search the second request list for the first
582 				 * request at a larger block number.  We go
583 				 * after that; if there is no such request, we
584 				 * go at the end.
585 				 */
586 				do {
587 					if (buf_inorder(bp, nbq, sortby))
588 						goto insert;
589 					bq = nbq;
590 				} while ((nbq =
591 				    TAILQ_NEXT(bq, b_actq)) != NULL);
592 				goto insert;		/* after last */
593 			}
594 			bq = nbq;
595 		}
596 		/*
597 		 * No inversions... we will go after the last, and
598 		 * be the first request in the second request list.
599 		 */
600 		goto insert;
601 	}
602 	/*
603 	 * Request is at/after the current request...
604 	 * sort in the first request list.
605 	 */
606 	while ((nbq = TAILQ_NEXT(bq, b_actq)) != NULL) {
607 		/*
608 		 * We want to go after the current request if there is an
609 		 * inversion after it (i.e. it is the end of the first
610 		 * request list), or if the next request is a larger cylinder
611 		 * than our request.
612 		 */
613 		if (buf_inorder(nbq, bq, sortby) ||
614 		    buf_inorder(bp, nbq, sortby))
615 			goto insert;
616 		bq = nbq;
617 	}
618 	/*
619 	 * Neither a second list nor a larger request... we go at the end of
620 	 * the first list, which is the same as the end of the whole schebang.
621 	 */
622 insert:	TAILQ_INSERT_AFTER(&disksort->bq_head, bq, bp, b_actq);
623 }
624 
625 static struct buf *
626 bufq_disksort_get(struct bufq_state *bufq, int remove)
627 {
628 	struct bufq_disksort *disksort = bufq->bq_private;
629 	struct buf *bp;
630 
631 	bp = TAILQ_FIRST(&disksort->bq_head);
632 
633 	if (bp != NULL && remove)
634 		TAILQ_REMOVE(&disksort->bq_head, bp, b_actq);
635 
636 	return (bp);
637 }
638 
639 
640 /*
641  * Seek sort for disks.
642  *
643  * There are two queues.  The first queue holds read requests; the second
644  * holds write requests.  The read queue is first-come first-served; the
645  * write queue is sorted in ascendening block order.
646  * The read queue is processed first.  After PRIO_READ_BURST consecutive
647  * read requests with non-empty write queue PRIO_WRITE_REQ requests from
648  * the write queue will be processed.
649  */
650 static void
651 bufq_prio_put(struct bufq_state *bufq, struct buf *bp)
652 {
653 	struct bufq_prio *prio = bufq->bq_private;
654 	struct buf *bq;
655 	int sortby;
656 
657 	sortby = bufq->bq_flags & BUFQ_SORT_MASK;
658 
659 	/*
660 	 * If it's a read request append it to the list.
661 	 */
662 	if ((bp->b_flags & B_READ) == B_READ) {
663 		TAILQ_INSERT_TAIL(&prio->bq_read, bp, b_actq);
664 		return;
665 	}
666 
667 	bq = TAILQ_FIRST(&prio->bq_write);
668 
669 	/*
670 	 * If the write list is empty, simply append it to the list.
671 	 */
672 	if (bq == NULL) {
673 		TAILQ_INSERT_TAIL(&prio->bq_write, bp, b_actq);
674 		prio->bq_write_next = bp;
675 		return;
676 	}
677 
678 	/*
679 	 * If we lie after the next request, insert after this request.
680 	 */
681 	if (buf_inorder(prio->bq_write_next, bp, sortby))
682 		bq = prio->bq_write_next;
683 
684 	/*
685 	 * Search for the first request at a larger block number.
686 	 * We go before this request if it exists.
687 	 */
688 	while (bq != NULL && buf_inorder(bq, bp, sortby))
689 		bq = TAILQ_NEXT(bq, b_actq);
690 
691 	if (bq != NULL)
692 		TAILQ_INSERT_BEFORE(bq, bp, b_actq);
693 	else
694 		TAILQ_INSERT_TAIL(&prio->bq_write, bp, b_actq);
695 }
696 
697 static struct buf *
698 bufq_prio_get(struct bufq_state *bufq, int remove)
699 {
700 	struct bufq_prio *prio = bufq->bq_private;
701 	struct buf *bp;
702 
703 	/*
704 	 * If no current request, get next from the lists.
705 	 */
706 	if (prio->bq_next == NULL) {
707 		/*
708 		 * If at least one list is empty, select the other.
709 		 */
710 		if (TAILQ_FIRST(&prio->bq_read) == NULL) {
711 			prio->bq_next = prio->bq_write_next;
712 			prio->bq_read_burst = 0;
713 		} else if (prio->bq_write_next == NULL) {
714 			prio->bq_next = TAILQ_FIRST(&prio->bq_read);
715 			prio->bq_read_burst = 0;
716 		} else {
717 			/*
718 			 * Both list have requests.  Select the read list up
719 			 * to PRIO_READ_BURST times, then select the write
720 			 * list PRIO_WRITE_REQ times.
721 			 */
722 			if (prio->bq_read_burst++ < PRIO_READ_BURST)
723 				prio->bq_next = TAILQ_FIRST(&prio->bq_read);
724 			else if (prio->bq_read_burst <
725 			    PRIO_READ_BURST + PRIO_WRITE_REQ)
726 				prio->bq_next = prio->bq_write_next;
727 			else {
728 				prio->bq_next = TAILQ_FIRST(&prio->bq_read);
729 				prio->bq_read_burst = 0;
730 			}
731 		}
732 	}
733 
734 	bp = prio->bq_next;
735 
736 	if (bp != NULL && remove) {
737 		if ((bp->b_flags & B_READ) == B_READ)
738 			TAILQ_REMOVE(&prio->bq_read, bp, b_actq);
739 		else {
740 			/*
741 			 * Advance the write pointer before removing
742 			 * bp since it is actually prio->bq_write_next.
743 			 */
744 			prio->bq_write_next =
745 			    TAILQ_NEXT(prio->bq_write_next, b_actq);
746 			TAILQ_REMOVE(&prio->bq_write, bp, b_actq);
747 			if (prio->bq_write_next == NULL)
748 				prio->bq_write_next =
749 				    TAILQ_FIRST(&prio->bq_write);
750 		}
751 
752 		prio->bq_next = NULL;
753 	}
754 
755 	return (bp);
756 }
757 
758 /*
759  * Create a device buffer queue.
760  */
761 void
762 bufq_alloc(struct bufq_state *bufq, int flags)
763 {
764 	struct bufq_fcfs *fcfs;
765 	struct bufq_disksort *disksort;
766 	struct bufq_prio *prio;
767 
768 	bufq->bq_flags = flags;
769 
770 	switch (flags & BUFQ_SORT_MASK) {
771 	case BUFQ_SORT_RAWBLOCK:
772 	case BUFQ_SORT_CYLINDER:
773 		break;
774 	case 0:
775 		if ((flags & BUFQ_METHOD_MASK) == BUFQ_FCFS)
776 			break;
777 		/* FALLTHROUGH */
778 	default:
779 		panic("bufq_alloc: sort out of range");
780 	}
781 
782 	switch (flags & BUFQ_METHOD_MASK) {
783 	case BUFQ_FCFS:
784 		bufq->bq_get = bufq_fcfs_get;
785 		bufq->bq_put = bufq_fcfs_put;
786 		MALLOC(bufq->bq_private, struct bufq_fcfs *,
787 		    sizeof(struct bufq_fcfs), M_DEVBUF, M_ZERO);
788 		fcfs = (struct bufq_fcfs *)bufq->bq_private;
789 		TAILQ_INIT(&fcfs->bq_head);
790 		break;
791 	case BUFQ_DISKSORT:
792 		bufq->bq_get = bufq_disksort_get;
793 		bufq->bq_put = bufq_disksort_put;
794 		MALLOC(bufq->bq_private, struct bufq_disksort *,
795 		    sizeof(struct bufq_disksort), M_DEVBUF, M_ZERO);
796 		disksort = (struct bufq_disksort *)bufq->bq_private;
797 		TAILQ_INIT(&disksort->bq_head);
798 		break;
799 	case BUFQ_READ_PRIO:
800 		bufq->bq_get = bufq_prio_get;
801 		bufq->bq_put = bufq_prio_put;
802 		MALLOC(bufq->bq_private, struct bufq_prio *,
803 		    sizeof(struct bufq_prio), M_DEVBUF, M_ZERO);
804 		prio = (struct bufq_prio *)bufq->bq_private;
805 		TAILQ_INIT(&prio->bq_read);
806 		TAILQ_INIT(&prio->bq_write);
807 		break;
808 	default:
809 		panic("bufq_alloc: method out of range");
810 	}
811 }
812 
813 /*
814  * Destroy a device buffer queue.
815  */
816 void
817 bufq_free(struct bufq_state *bufq)
818 {
819 
820 	KASSERT(bufq->bq_private != NULL);
821 	KASSERT(BUFQ_PEEK(bufq) == NULL);
822 
823 	FREE(bufq->bq_private, M_DEVBUF);
824 	bufq->bq_get = NULL;
825 	bufq->bq_put = NULL;
826 }
827 
828 /*
829  * Bounds checking against the media size, used for the raw partition.
830  * The sector size passed in should currently always be DEV_BSIZE,
831  * and the media size the size of the device in DEV_BSIZE sectors.
832  */
833 int
834 bounds_check_with_mediasize(struct buf *bp, int secsize, u_int64_t mediasize)
835 {
836 	int sz;
837 
838 	sz = howmany(bp->b_bcount, secsize);
839 
840 	if (bp->b_blkno + sz > mediasize) {
841 		sz = mediasize - bp->b_blkno;
842 		if (sz == 0) {
843 			/* If exactly at end of disk, return EOF. */
844 			bp->b_resid = bp->b_bcount;
845 			goto done;
846 		}
847 		if (sz < 0) {
848 			/* If past end of disk, return EINVAL. */
849 			bp->b_error = EINVAL;
850 			goto bad;
851 		}
852 		/* Otherwise, truncate request. */
853 		bp->b_bcount = sz << DEV_BSHIFT;
854 	}
855 
856 	return 1;
857 
858 bad:
859 	bp->b_flags |= B_ERROR;
860 done:
861 	return 0;
862 }
863