subr_disk.c revision 1.58 1 /* $NetBSD: subr_disk.c,v 1.58 2004/01/10 14:49:44 yamt 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. Neither the name of the University nor the names of its contributors
58 * may be used to endorse or promote products derived from this software
59 * without specific prior written permission.
60 *
61 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
62 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
63 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
64 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
65 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
66 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
67 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
68 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
69 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
70 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
71 * SUCH DAMAGE.
72 *
73 * @(#)ufs_disksubr.c 8.5 (Berkeley) 1/21/94
74 */
75
76 #include <sys/cdefs.h>
77 __KERNEL_RCSID(0, "$NetBSD: subr_disk.c,v 1.58 2004/01/10 14:49:44 yamt Exp $");
78
79 #include "opt_compat_netbsd.h"
80
81 #include <sys/param.h>
82 #include <sys/kernel.h>
83 #include <sys/malloc.h>
84 #include <sys/buf.h>
85 #include <sys/syslog.h>
86 #include <sys/disklabel.h>
87 #include <sys/disk.h>
88 #include <sys/sysctl.h>
89 #include <lib/libkern/libkern.h>
90
91 /*
92 * A global list of all disks attached to the system. May grow or
93 * shrink over time.
94 */
95 struct disklist_head disklist; /* TAILQ_HEAD */
96 int disk_count; /* number of drives in global disklist */
97 struct simplelock disklist_slock = SIMPLELOCK_INITIALIZER;
98
99 /*
100 * Compute checksum for disk label.
101 */
102 u_int
103 dkcksum(struct disklabel *lp)
104 {
105 u_short *start, *end;
106 u_short sum = 0;
107
108 start = (u_short *)lp;
109 end = (u_short *)&lp->d_partitions[lp->d_npartitions];
110 while (start < end)
111 sum ^= *start++;
112 return (sum);
113 }
114
115 /*
116 * Disk error is the preface to plaintive error messages
117 * about failing disk transfers. It prints messages of the form
118
119 hp0g: hard error reading fsbn 12345 of 12344-12347 (hp0 bn %d cn %d tn %d sn %d)
120
121 * if the offset of the error in the transfer and a disk label
122 * are both available. blkdone should be -1 if the position of the error
123 * is unknown; the disklabel pointer may be null from drivers that have not
124 * been converted to use them. The message is printed with printf
125 * if pri is LOG_PRINTF, otherwise it uses log at the specified priority.
126 * The message should be completed (with at least a newline) with printf
127 * or addlog, respectively. There is no trailing space.
128 */
129 #ifndef PRIdaddr
130 #define PRIdaddr PRId64
131 #endif
132 void
133 diskerr(const struct buf *bp, const char *dname, const char *what, int pri,
134 int blkdone, const struct disklabel *lp)
135 {
136 int unit = DISKUNIT(bp->b_dev), part = DISKPART(bp->b_dev);
137 void (*pr)(const char *, ...);
138 char partname = 'a' + part;
139 daddr_t sn;
140
141 if (/*CONSTCOND*/0)
142 /* Compiler will error this is the format is wrong... */
143 printf("%" PRIdaddr, bp->b_blkno);
144
145 if (pri != LOG_PRINTF) {
146 static const char fmt[] = "";
147 log(pri, fmt);
148 pr = addlog;
149 } else
150 pr = printf;
151 (*pr)("%s%d%c: %s %sing fsbn ", dname, unit, partname, what,
152 bp->b_flags & B_READ ? "read" : "writ");
153 sn = bp->b_blkno;
154 if (bp->b_bcount <= DEV_BSIZE)
155 (*pr)("%" PRIdaddr, sn);
156 else {
157 if (blkdone >= 0) {
158 sn += blkdone;
159 (*pr)("%" PRIdaddr " of ", sn);
160 }
161 (*pr)("%" PRIdaddr "-%" PRIdaddr "", bp->b_blkno,
162 bp->b_blkno + (bp->b_bcount - 1) / DEV_BSIZE);
163 }
164 if (lp && (blkdone >= 0 || bp->b_bcount <= lp->d_secsize)) {
165 sn += lp->d_partitions[part].p_offset;
166 (*pr)(" (%s%d bn %" PRIdaddr "; cn %" PRIdaddr "",
167 dname, unit, sn, sn / lp->d_secpercyl);
168 sn %= lp->d_secpercyl;
169 (*pr)(" tn %" PRIdaddr " sn %" PRIdaddr ")",
170 sn / lp->d_nsectors, sn % lp->d_nsectors);
171 }
172 }
173
174 /*
175 * Initialize the disklist. Called by main() before autoconfiguration.
176 */
177 void
178 disk_init(void)
179 {
180
181 TAILQ_INIT(&disklist);
182 disk_count = 0;
183 }
184
185 /*
186 * Searches the disklist for the disk corresponding to the
187 * name provided.
188 */
189 struct disk *
190 disk_find(char *name)
191 {
192 struct disk *diskp;
193
194 if ((name == NULL) || (disk_count <= 0))
195 return (NULL);
196
197 simple_lock(&disklist_slock);
198 for (diskp = TAILQ_FIRST(&disklist); diskp != NULL;
199 diskp = TAILQ_NEXT(diskp, dk_link))
200 if (strcmp(diskp->dk_name, name) == 0) {
201 simple_unlock(&disklist_slock);
202 return (diskp);
203 }
204 simple_unlock(&disklist_slock);
205
206 return (NULL);
207 }
208
209 /*
210 * Attach a disk.
211 */
212 void
213 disk_attach(struct disk *diskp)
214 {
215 int s;
216
217 /*
218 * Allocate and initialize the disklabel structures. Note that
219 * it's not safe to sleep here, since we're probably going to be
220 * called during autoconfiguration.
221 */
222 diskp->dk_label = malloc(sizeof(struct disklabel), M_DEVBUF, M_NOWAIT);
223 diskp->dk_cpulabel = malloc(sizeof(struct cpu_disklabel), M_DEVBUF,
224 M_NOWAIT);
225 if ((diskp->dk_label == NULL) || (diskp->dk_cpulabel == NULL))
226 panic("disk_attach: can't allocate storage for disklabel");
227
228 memset(diskp->dk_label, 0, sizeof(struct disklabel));
229 memset(diskp->dk_cpulabel, 0, sizeof(struct cpu_disklabel));
230
231 /*
232 * Set the attached timestamp.
233 */
234 s = splclock();
235 diskp->dk_attachtime = mono_time;
236 splx(s);
237
238 /*
239 * Link into the disklist.
240 */
241 simple_lock(&disklist_slock);
242 TAILQ_INSERT_TAIL(&disklist, diskp, dk_link);
243 simple_unlock(&disklist_slock);
244 ++disk_count;
245 }
246
247 /*
248 * Detach a disk.
249 */
250 void
251 disk_detach(struct disk *diskp)
252 {
253
254 /*
255 * Remove from the disklist.
256 */
257 if (--disk_count < 0)
258 panic("disk_detach: disk_count < 0");
259 simple_lock(&disklist_slock);
260 TAILQ_REMOVE(&disklist, diskp, dk_link);
261 simple_unlock(&disklist_slock);
262
263 /*
264 * Free the space used by the disklabel structures.
265 */
266 free(diskp->dk_label, M_DEVBUF);
267 free(diskp->dk_cpulabel, M_DEVBUF);
268 }
269
270 /*
271 * Increment a disk's busy counter. If the counter is going from
272 * 0 to 1, set the timestamp.
273 */
274 void
275 disk_busy(struct disk *diskp)
276 {
277 int s;
278
279 /*
280 * XXX We'd like to use something as accurate as microtime(),
281 * but that doesn't depend on the system TOD clock.
282 */
283 if (diskp->dk_busy++ == 0) {
284 s = splclock();
285 diskp->dk_timestamp = mono_time;
286 splx(s);
287 }
288 }
289
290 /*
291 * Decrement a disk's busy counter, increment the byte count, total busy
292 * time, and reset the timestamp.
293 */
294 void
295 disk_unbusy(struct disk *diskp, long bcount, int read)
296 {
297 int s;
298 struct timeval dv_time, diff_time;
299
300 if (diskp->dk_busy-- == 0) {
301 printf("%s: dk_busy < 0\n", diskp->dk_name);
302 panic("disk_unbusy");
303 }
304
305 s = splclock();
306 dv_time = mono_time;
307 splx(s);
308
309 timersub(&dv_time, &diskp->dk_timestamp, &diff_time);
310 timeradd(&diskp->dk_time, &diff_time, &diskp->dk_time);
311
312 diskp->dk_timestamp = dv_time;
313 if (bcount > 0) {
314 if (read) {
315 diskp->dk_rbytes += bcount;
316 diskp->dk_rxfer++;
317 } else {
318 diskp->dk_wbytes += bcount;
319 diskp->dk_wxfer++;
320 }
321 }
322 }
323
324 /*
325 * Reset the metrics counters on the given disk. Note that we cannot
326 * reset the busy counter, as it may case a panic in disk_unbusy().
327 * We also must avoid playing with the timestamp information, as it
328 * may skew any pending transfer results.
329 */
330 void
331 disk_resetstat(struct disk *diskp)
332 {
333 int s = splbio(), t;
334
335 diskp->dk_rxfer = 0;
336 diskp->dk_rbytes = 0;
337 diskp->dk_wxfer = 0;
338 diskp->dk_wbytes = 0;
339
340 t = splclock();
341 diskp->dk_attachtime = mono_time;
342 splx(t);
343
344 timerclear(&diskp->dk_time);
345
346 splx(s);
347 }
348
349 int
350 sysctl_hw_disknames(SYSCTLFN_ARGS)
351 {
352 char buf[DK_DISKNAMELEN + 1];
353 char *where = oldp;
354 struct disk *diskp;
355 size_t needed, left, slen;
356 int error, first;
357
358 if (newp != NULL)
359 return (EPERM);
360 if (namelen != 0)
361 return (EINVAL);
362
363 first = 1;
364 error = 0;
365 needed = 0;
366 left = *oldlenp;
367
368 simple_lock(&disklist_slock);
369 for (diskp = TAILQ_FIRST(&disklist); diskp != NULL;
370 diskp = TAILQ_NEXT(diskp, dk_link)) {
371 if (where == NULL)
372 needed += strlen(diskp->dk_name) + 1;
373 else {
374 memset(buf, 0, sizeof(buf));
375 if (first) {
376 strncpy(buf, diskp->dk_name, sizeof(buf));
377 first = 0;
378 } else {
379 buf[0] = ' ';
380 strncpy(buf + 1, diskp->dk_name,
381 sizeof(buf) - 1);
382 }
383 buf[DK_DISKNAMELEN] = '\0';
384 slen = strlen(buf);
385 if (left < slen + 1)
386 break;
387 /* +1 to copy out the trailing NUL byte */
388 error = copyout(buf, where, slen + 1);
389 if (error)
390 break;
391 where += slen;
392 needed += slen;
393 left -= slen;
394 }
395 }
396 simple_unlock(&disklist_slock);
397 *oldlenp = needed;
398 return (error);
399 }
400
401 int
402 sysctl_hw_diskstats(SYSCTLFN_ARGS)
403 {
404 struct disk_sysctl sdisk;
405 struct disk *diskp;
406 char *where = oldp;
407 size_t tocopy, left;
408 int error;
409
410 if (newp != NULL)
411 return (EPERM);
412
413 /*
414 * The original hw.diskstats call was broken and did not require
415 * the userland to pass in it's size of struct disk_sysctl. This
416 * was fixed after NetBSD 1.6 was released, and any applications
417 * that do not pass in the size are given an error only, unless
418 * we care about 1.6 compatibility.
419 */
420 if (namelen == 0)
421 #ifdef COMPAT_16
422 tocopy = offsetof(struct disk_sysctl, dk_rxfer);
423 #else
424 return (EINVAL);
425 #endif
426 else
427 tocopy = name[0];
428
429 if (where == NULL) {
430 *oldlenp = disk_count * tocopy;
431 return (0);
432 }
433
434 error = 0;
435 left = *oldlenp;
436 memset(&sdisk, 0, sizeof(sdisk));
437 *oldlenp = 0;
438
439 simple_lock(&disklist_slock);
440 TAILQ_FOREACH(diskp, &disklist, dk_link) {
441 if (left < tocopy)
442 break;
443 strncpy(sdisk.dk_name, diskp->dk_name, sizeof(sdisk.dk_name));
444 sdisk.dk_xfer = diskp->dk_rxfer + diskp->dk_wxfer;
445 sdisk.dk_rxfer = diskp->dk_rxfer;
446 sdisk.dk_wxfer = diskp->dk_wxfer;
447 sdisk.dk_seek = diskp->dk_seek;
448 sdisk.dk_bytes = diskp->dk_rbytes + diskp->dk_wbytes;
449 sdisk.dk_rbytes = diskp->dk_rbytes;
450 sdisk.dk_wbytes = diskp->dk_wbytes;
451 sdisk.dk_attachtime_sec = diskp->dk_attachtime.tv_sec;
452 sdisk.dk_attachtime_usec = diskp->dk_attachtime.tv_usec;
453 sdisk.dk_timestamp_sec = diskp->dk_timestamp.tv_sec;
454 sdisk.dk_timestamp_usec = diskp->dk_timestamp.tv_usec;
455 sdisk.dk_time_sec = diskp->dk_time.tv_sec;
456 sdisk.dk_time_usec = diskp->dk_time.tv_usec;
457 sdisk.dk_busy = diskp->dk_busy;
458
459 error = copyout(&sdisk, where, min(tocopy, sizeof(sdisk)));
460 if (error)
461 break;
462 where += tocopy;
463 *oldlenp += tocopy;
464 left -= tocopy;
465 }
466 simple_unlock(&disklist_slock);
467 return (error);
468 }
469
470 struct bufq_fcfs {
471 TAILQ_HEAD(, buf) bq_head; /* actual list of buffers */
472 };
473
474 struct bufq_disksort {
475 TAILQ_HEAD(, buf) bq_head; /* actual list of buffers */
476 };
477
478 #define PRIO_READ_BURST 48
479 #define PRIO_WRITE_REQ 16
480
481 struct bufq_prio {
482 TAILQ_HEAD(, buf) bq_read, bq_write; /* actual list of buffers */
483 struct buf *bq_write_next; /* next request in bq_write */
484 struct buf *bq_next; /* current request */
485 int bq_read_burst; /* # of consecutive reads */
486 };
487
488
489 /*
490 * Check if two buf's are in ascending order.
491 */
492 static __inline int
493 buf_inorder(struct buf *bp, struct buf *bq, int sortby)
494 {
495
496 if (bp == NULL || bq == NULL)
497 return (bq == NULL);
498
499 if (sortby == BUFQ_SORT_CYLINDER) {
500 if (bp->b_cylinder != bq->b_cylinder)
501 return bp->b_cylinder < bq->b_cylinder;
502 else
503 return bp->b_rawblkno < bq->b_rawblkno;
504 } else
505 return bp->b_rawblkno < bq->b_rawblkno;
506 }
507
508
509 /*
510 * First-come first-served sort for disks.
511 *
512 * Requests are appended to the queue without any reordering.
513 */
514 static void
515 bufq_fcfs_put(struct bufq_state *bufq, struct buf *bp)
516 {
517 struct bufq_fcfs *fcfs = bufq->bq_private;
518
519 TAILQ_INSERT_TAIL(&fcfs->bq_head, bp, b_actq);
520 }
521
522 static struct buf *
523 bufq_fcfs_get(struct bufq_state *bufq, int remove)
524 {
525 struct bufq_fcfs *fcfs = bufq->bq_private;
526 struct buf *bp;
527
528 bp = TAILQ_FIRST(&fcfs->bq_head);
529
530 if (bp != NULL && remove)
531 TAILQ_REMOVE(&fcfs->bq_head, bp, b_actq);
532
533 return (bp);
534 }
535
536
537 /*
538 * Seek sort for disks.
539 *
540 * There are actually two queues, sorted in ascendening order. The first
541 * queue holds those requests which are positioned after the current block;
542 * the second holds requests which came in after their position was passed.
543 * Thus we implement a one-way scan, retracting after reaching the end of
544 * the drive to the first request on the second queue, at which time it
545 * becomes the first queue.
546 *
547 * A one-way scan is natural because of the way UNIX read-ahead blocks are
548 * allocated.
549 */
550 static void
551 bufq_disksort_put(struct bufq_state *bufq, struct buf *bp)
552 {
553 struct bufq_disksort *disksort = bufq->bq_private;
554 struct buf *bq, *nbq;
555 int sortby;
556
557 sortby = bufq->bq_flags & BUFQ_SORT_MASK;
558
559 bq = TAILQ_FIRST(&disksort->bq_head);
560
561 /*
562 * If the queue is empty it's easy; we just go on the end.
563 */
564 if (bq == NULL) {
565 TAILQ_INSERT_TAIL(&disksort->bq_head, bp, b_actq);
566 return;
567 }
568
569 /*
570 * If we lie before the currently active request, then we
571 * must locate the second request list and add ourselves to it.
572 */
573 if (buf_inorder(bp, bq, sortby)) {
574 while ((nbq = TAILQ_NEXT(bq, b_actq)) != NULL) {
575 /*
576 * Check for an ``inversion'' in the normally ascending
577 * block numbers, indicating the start of the second
578 * request list.
579 */
580 if (buf_inorder(nbq, bq, sortby)) {
581 /*
582 * Search the second request list for the first
583 * request at a larger block number. We go
584 * after that; if there is no such request, we
585 * go at the end.
586 */
587 do {
588 if (buf_inorder(bp, nbq, sortby))
589 goto insert;
590 bq = nbq;
591 } while ((nbq =
592 TAILQ_NEXT(bq, b_actq)) != NULL);
593 goto insert; /* after last */
594 }
595 bq = nbq;
596 }
597 /*
598 * No inversions... we will go after the last, and
599 * be the first request in the second request list.
600 */
601 goto insert;
602 }
603 /*
604 * Request is at/after the current request...
605 * sort in the first request list.
606 */
607 while ((nbq = TAILQ_NEXT(bq, b_actq)) != NULL) {
608 /*
609 * We want to go after the current request if there is an
610 * inversion after it (i.e. it is the end of the first
611 * request list), or if the next request is a larger cylinder
612 * than our request.
613 */
614 if (buf_inorder(nbq, bq, sortby) ||
615 buf_inorder(bp, nbq, sortby))
616 goto insert;
617 bq = nbq;
618 }
619 /*
620 * Neither a second list nor a larger request... we go at the end of
621 * the first list, which is the same as the end of the whole schebang.
622 */
623 insert: TAILQ_INSERT_AFTER(&disksort->bq_head, bq, bp, b_actq);
624 }
625
626 static struct buf *
627 bufq_disksort_get(struct bufq_state *bufq, int remove)
628 {
629 struct bufq_disksort *disksort = bufq->bq_private;
630 struct buf *bp;
631
632 bp = TAILQ_FIRST(&disksort->bq_head);
633
634 if (bp != NULL && remove)
635 TAILQ_REMOVE(&disksort->bq_head, bp, b_actq);
636
637 return (bp);
638 }
639
640
641 /*
642 * Seek sort for disks.
643 *
644 * There are two queues. The first queue holds read requests; the second
645 * holds write requests. The read queue is first-come first-served; the
646 * write queue is sorted in ascendening block order.
647 * The read queue is processed first. After PRIO_READ_BURST consecutive
648 * read requests with non-empty write queue PRIO_WRITE_REQ requests from
649 * the write queue will be processed.
650 */
651 static void
652 bufq_prio_put(struct bufq_state *bufq, struct buf *bp)
653 {
654 struct bufq_prio *prio = bufq->bq_private;
655 struct buf *bq;
656 int sortby;
657
658 sortby = bufq->bq_flags & BUFQ_SORT_MASK;
659
660 /*
661 * If it's a read request append it to the list.
662 */
663 if ((bp->b_flags & B_READ) == B_READ) {
664 TAILQ_INSERT_TAIL(&prio->bq_read, bp, b_actq);
665 return;
666 }
667
668 bq = TAILQ_FIRST(&prio->bq_write);
669
670 /*
671 * If the write list is empty, simply append it to the list.
672 */
673 if (bq == NULL) {
674 TAILQ_INSERT_TAIL(&prio->bq_write, bp, b_actq);
675 prio->bq_write_next = bp;
676 return;
677 }
678
679 /*
680 * If we lie after the next request, insert after this request.
681 */
682 if (buf_inorder(prio->bq_write_next, bp, sortby))
683 bq = prio->bq_write_next;
684
685 /*
686 * Search for the first request at a larger block number.
687 * We go before this request if it exists.
688 */
689 while (bq != NULL && buf_inorder(bq, bp, sortby))
690 bq = TAILQ_NEXT(bq, b_actq);
691
692 if (bq != NULL)
693 TAILQ_INSERT_BEFORE(bq, bp, b_actq);
694 else
695 TAILQ_INSERT_TAIL(&prio->bq_write, bp, b_actq);
696 }
697
698 static struct buf *
699 bufq_prio_get(struct bufq_state *bufq, int remove)
700 {
701 struct bufq_prio *prio = bufq->bq_private;
702 struct buf *bp;
703
704 /*
705 * If no current request, get next from the lists.
706 */
707 if (prio->bq_next == NULL) {
708 /*
709 * If at least one list is empty, select the other.
710 */
711 if (TAILQ_FIRST(&prio->bq_read) == NULL) {
712 prio->bq_next = prio->bq_write_next;
713 prio->bq_read_burst = 0;
714 } else if (prio->bq_write_next == NULL) {
715 prio->bq_next = TAILQ_FIRST(&prio->bq_read);
716 prio->bq_read_burst = 0;
717 } else {
718 /*
719 * Both list have requests. Select the read list up
720 * to PRIO_READ_BURST times, then select the write
721 * list PRIO_WRITE_REQ times.
722 */
723 if (prio->bq_read_burst++ < PRIO_READ_BURST)
724 prio->bq_next = TAILQ_FIRST(&prio->bq_read);
725 else if (prio->bq_read_burst <
726 PRIO_READ_BURST + PRIO_WRITE_REQ)
727 prio->bq_next = prio->bq_write_next;
728 else {
729 prio->bq_next = TAILQ_FIRST(&prio->bq_read);
730 prio->bq_read_burst = 0;
731 }
732 }
733 }
734
735 bp = prio->bq_next;
736
737 if (bp != NULL && remove) {
738 if ((bp->b_flags & B_READ) == B_READ)
739 TAILQ_REMOVE(&prio->bq_read, bp, b_actq);
740 else {
741 /*
742 * Advance the write pointer before removing
743 * bp since it is actually prio->bq_write_next.
744 */
745 prio->bq_write_next =
746 TAILQ_NEXT(prio->bq_write_next, b_actq);
747 TAILQ_REMOVE(&prio->bq_write, bp, b_actq);
748 if (prio->bq_write_next == NULL)
749 prio->bq_write_next =
750 TAILQ_FIRST(&prio->bq_write);
751 }
752
753 prio->bq_next = NULL;
754 }
755
756 return (bp);
757 }
758
759
760 /*
761 * Cyclical scan (CSCAN)
762 */
763 TAILQ_HEAD(bqhead, buf);
764 struct cscan_queue {
765 struct bqhead cq_head[2]; /* actual lists of buffers */
766 int cq_idx; /* current list index */
767 int cq_lastcylinder; /* b_cylinder of the last request */
768 daddr_t cq_lastrawblkno; /* b_rawblkno of the last request */
769 };
770
771 static int __inline cscan_empty(const struct cscan_queue *);
772 static void cscan_put(struct cscan_queue *, struct buf *, int);
773 static struct buf *cscan_get(struct cscan_queue *, int);
774 static void cscan_init(struct cscan_queue *);
775
776 static __inline int
777 cscan_empty(const struct cscan_queue *q)
778 {
779
780 return TAILQ_EMPTY(&q->cq_head[0]) && TAILQ_EMPTY(&q->cq_head[1]);
781 }
782
783 static void
784 cscan_put(struct cscan_queue *q, struct buf *bp, int sortby)
785 {
786 struct buf tmp;
787 struct buf *it;
788 struct bqhead *bqh;
789 int idx;
790
791 tmp.b_cylinder = q->cq_lastcylinder;
792 tmp.b_rawblkno = q->cq_lastrawblkno;
793
794 if (buf_inorder(bp, &tmp, sortby))
795 idx = 1 - q->cq_idx;
796 else
797 idx = q->cq_idx;
798
799 bqh = &q->cq_head[idx];
800
801 TAILQ_FOREACH(it, bqh, b_actq)
802 if (buf_inorder(bp, it, sortby))
803 break;
804
805 if (it != NULL)
806 TAILQ_INSERT_BEFORE(it, bp, b_actq);
807 else
808 TAILQ_INSERT_TAIL(bqh, bp, b_actq);
809 }
810
811 static struct buf *
812 cscan_get(struct cscan_queue *q, int remove)
813 {
814 int idx = q->cq_idx;
815 struct bqhead *bqh;
816 struct buf *bp;
817
818 bqh = &q->cq_head[idx];
819 bp = TAILQ_FIRST(bqh);
820
821 if (bp == NULL) {
822 /* switch queue */
823 idx = 1 - idx;
824 bqh = &q->cq_head[idx];
825 bp = TAILQ_FIRST(bqh);
826 }
827
828 KDASSERT((bp != NULL && !cscan_empty(q)) ||
829 (bp == NULL && cscan_empty(q)));
830
831 if (bp != NULL && remove) {
832 q->cq_idx = idx;
833 TAILQ_REMOVE(bqh, bp, b_actq);
834
835 q->cq_lastcylinder = bp->b_cylinder;
836 q->cq_lastrawblkno =
837 bp->b_rawblkno + (bp->b_bcount >> DEV_BSHIFT);
838 }
839
840 return (bp);
841 }
842
843 static void
844 cscan_init(struct cscan_queue *q)
845 {
846
847 TAILQ_INIT(&q->cq_head[0]);
848 TAILQ_INIT(&q->cq_head[1]);
849 }
850
851
852 /*
853 * Per-prioritiy CSCAN.
854 *
855 * XXX probably we should have a way to raise
856 * priority of the on-queue requests.
857 */
858 #define PRIOCSCAN_NQUEUE 3
859
860 struct priocscan_queue {
861 struct cscan_queue q_queue;
862 int q_burst;
863 };
864
865 struct bufq_priocscan {
866 struct priocscan_queue bq_queue[PRIOCSCAN_NQUEUE];
867
868 #if 0
869 /*
870 * XXX using "global" head position can reduce positioning time
871 * when switching between queues.
872 * although it might affect against fairness.
873 */
874 daddr_t bq_lastrawblkno;
875 int bq_lastcylinder;
876 #endif
877 };
878
879 /*
880 * how many requests to serve when having pending requests on other queues.
881 *
882 * XXX tune
883 */
884 const int priocscan_burst[] = {
885 64, 16, 4
886 };
887
888 static void bufq_priocscan_put(struct bufq_state *, struct buf *);
889 static struct buf *bufq_priocscan_get(struct bufq_state *, int);
890 static void bufq_priocscan_init(struct bufq_state *);
891 static __inline struct cscan_queue *bufq_priocscan_selectqueue(
892 struct bufq_priocscan *, const struct buf *);
893
894 static __inline struct cscan_queue *
895 bufq_priocscan_selectqueue(struct bufq_priocscan *q, const struct buf *bp)
896 {
897 static const int priocscan_priomap[] = {
898 [BPRIO_TIMENONCRITICAL] = 2,
899 [BPRIO_TIMELIMITED] = 1,
900 [BPRIO_TIMECRITICAL] = 0
901 };
902
903 return &q->bq_queue[priocscan_priomap[BIO_GETPRIO(bp)]].q_queue;
904 }
905
906 static void
907 bufq_priocscan_put(struct bufq_state *bufq, struct buf *bp)
908 {
909 struct bufq_priocscan *q = bufq->bq_private;
910 struct cscan_queue *cq;
911 const int sortby = bufq->bq_flags & BUFQ_SORT_MASK;
912
913 cq = bufq_priocscan_selectqueue(q, bp);
914 cscan_put(cq, bp, sortby);
915 }
916
917 static struct buf *
918 bufq_priocscan_get(struct bufq_state *bufq, int remove)
919 {
920 struct bufq_priocscan *q = bufq->bq_private;
921 struct priocscan_queue *pq, *npq;
922 struct priocscan_queue *first; /* first non-empty queue */
923 const struct priocscan_queue *epq;
924 const struct cscan_queue *cq;
925 struct buf *bp;
926 boolean_t single; /* true if there's only one non-empty queue */
927
928 pq = &q->bq_queue[0];
929 epq = pq + PRIOCSCAN_NQUEUE;
930 for (; pq < epq; pq++) {
931 cq = &pq->q_queue;
932 if (!cscan_empty(cq))
933 break;
934 }
935 if (pq == epq) {
936 /* there's no requests */
937 return NULL;
938 }
939
940 first = pq;
941 single = TRUE;
942 for (npq = first + 1; npq < epq; npq++) {
943 cq = &npq->q_queue;
944 if (!cscan_empty(cq)) {
945 single = FALSE;
946 if (pq->q_burst > 0)
947 break;
948 pq = npq;
949 }
950 }
951 if (single) {
952 /*
953 * there's only a non-empty queue. just serve it.
954 */
955 pq = first;
956 } else if (pq->q_burst > 0) {
957 /*
958 * XXX account only by number of requests. is it good enough?
959 */
960 pq->q_burst--;
961 } else {
962 /*
963 * no queue was selected due to burst counts
964 */
965 int i;
966 #ifdef DEBUG
967 for (i = 0; i < PRIOCSCAN_NQUEUE; i++) {
968 pq = &q->bq_queue[i];
969 cq = &pq->q_queue;
970 if (!cscan_empty(cq) && pq->q_burst)
971 panic("%s: inconsist", __func__);
972 }
973 #endif /* DEBUG */
974
975 /*
976 * reset burst counts
977 */
978 for (i = 0; i < PRIOCSCAN_NQUEUE; i++) {
979 pq = &q->bq_queue[i];
980 pq->q_burst = priocscan_burst[i];
981 }
982
983 /*
984 * serve first non-empty queue.
985 */
986 pq = first;
987 }
988
989 KDASSERT(!cscan_empty(&pq->q_queue));
990 bp = cscan_get(&pq->q_queue, remove);
991 KDASSERT(bp != NULL);
992 KDASSERT(&pq->q_queue == bufq_priocscan_selectqueue(q, bp));
993
994 return bp;
995 }
996
997 static void
998 bufq_priocscan_init(struct bufq_state *bufq)
999 {
1000 struct bufq_priocscan *q;
1001 int i;
1002
1003 bufq->bq_get = bufq_priocscan_get;
1004 bufq->bq_put = bufq_priocscan_put;
1005 bufq->bq_private = malloc(sizeof(struct bufq_priocscan),
1006 M_DEVBUF, M_ZERO);
1007
1008 q = bufq->bq_private;
1009 for (i = 0; i < PRIOCSCAN_NQUEUE; i++) {
1010 struct cscan_queue *cq = &q->bq_queue[i].q_queue;
1011
1012 cscan_init(cq);
1013 }
1014 }
1015
1016
1017 /*
1018 * Create a device buffer queue.
1019 */
1020 void
1021 bufq_alloc(struct bufq_state *bufq, int flags)
1022 {
1023 struct bufq_fcfs *fcfs;
1024 struct bufq_disksort *disksort;
1025 struct bufq_prio *prio;
1026
1027 bufq->bq_flags = flags;
1028
1029 switch (flags & BUFQ_SORT_MASK) {
1030 case BUFQ_SORT_RAWBLOCK:
1031 case BUFQ_SORT_CYLINDER:
1032 break;
1033 case 0:
1034 if ((flags & BUFQ_METHOD_MASK) == BUFQ_FCFS)
1035 break;
1036 /* FALLTHROUGH */
1037 default:
1038 panic("bufq_alloc: sort out of range");
1039 }
1040
1041 switch (flags & BUFQ_METHOD_MASK) {
1042 case BUFQ_FCFS:
1043 bufq->bq_get = bufq_fcfs_get;
1044 bufq->bq_put = bufq_fcfs_put;
1045 MALLOC(bufq->bq_private, struct bufq_fcfs *,
1046 sizeof(struct bufq_fcfs), M_DEVBUF, M_ZERO);
1047 fcfs = (struct bufq_fcfs *)bufq->bq_private;
1048 TAILQ_INIT(&fcfs->bq_head);
1049 break;
1050 case BUFQ_DISKSORT:
1051 bufq->bq_get = bufq_disksort_get;
1052 bufq->bq_put = bufq_disksort_put;
1053 MALLOC(bufq->bq_private, struct bufq_disksort *,
1054 sizeof(struct bufq_disksort), M_DEVBUF, M_ZERO);
1055 disksort = (struct bufq_disksort *)bufq->bq_private;
1056 TAILQ_INIT(&disksort->bq_head);
1057 break;
1058 case BUFQ_READ_PRIO:
1059 bufq->bq_get = bufq_prio_get;
1060 bufq->bq_put = bufq_prio_put;
1061 MALLOC(bufq->bq_private, struct bufq_prio *,
1062 sizeof(struct bufq_prio), M_DEVBUF, M_ZERO);
1063 prio = (struct bufq_prio *)bufq->bq_private;
1064 TAILQ_INIT(&prio->bq_read);
1065 TAILQ_INIT(&prio->bq_write);
1066 break;
1067 case BUFQ_PRIOCSCAN:
1068 bufq_priocscan_init(bufq);
1069 break;
1070 default:
1071 panic("bufq_alloc: method out of range");
1072 }
1073 }
1074
1075 /*
1076 * Destroy a device buffer queue.
1077 */
1078 void
1079 bufq_free(struct bufq_state *bufq)
1080 {
1081
1082 KASSERT(bufq->bq_private != NULL);
1083 KASSERT(BUFQ_PEEK(bufq) == NULL);
1084
1085 FREE(bufq->bq_private, M_DEVBUF);
1086 bufq->bq_get = NULL;
1087 bufq->bq_put = NULL;
1088 }
1089
1090 /*
1091 * Bounds checking against the media size, used for the raw partition.
1092 * The sector size passed in should currently always be DEV_BSIZE,
1093 * and the media size the size of the device in DEV_BSIZE sectors.
1094 */
1095 int
1096 bounds_check_with_mediasize(struct buf *bp, int secsize, u_int64_t mediasize)
1097 {
1098 int sz;
1099
1100 sz = howmany(bp->b_bcount, secsize);
1101
1102 if (bp->b_blkno + sz > mediasize) {
1103 sz = mediasize - bp->b_blkno;
1104 if (sz == 0) {
1105 /* If exactly at end of disk, return EOF. */
1106 bp->b_resid = bp->b_bcount;
1107 goto done;
1108 }
1109 if (sz < 0) {
1110 /* If past end of disk, return EINVAL. */
1111 bp->b_error = EINVAL;
1112 goto bad;
1113 }
1114 /* Otherwise, truncate request. */
1115 bp->b_bcount = sz << DEV_BSHIFT;
1116 }
1117
1118 return 1;
1119
1120 bad:
1121 bp->b_flags |= B_ERROR;
1122 done:
1123 return 0;
1124 }
1125