ld.c revision 1.6 1 /* $NetBSD: ld.c,v 1.6 2001/01/08 06:57:21 itojun Exp $ */
2
3 /*-
4 * Copyright (c) 1998, 2000 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Andrew Doran and Charles M. Hannum.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the NetBSD
21 * Foundation, Inc. and its contributors.
22 * 4. Neither the name of The NetBSD Foundation nor the names of its
23 * contributors may be used to endorse or promote products derived
24 * from this software without specific prior written permission.
25 *
26 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36 * POSSIBILITY OF SUCH DAMAGE.
37 */
38
39 /*
40 * Disk driver for use by RAID controllers.
41 */
42
43 #include "rnd.h"
44
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/kernel.h>
48 #include <sys/device.h>
49 #include <sys/queue.h>
50 #include <sys/proc.h>
51 #include <sys/buf.h>
52 #include <sys/endian.h>
53 #include <sys/disklabel.h>
54 #include <sys/disk.h>
55 #include <sys/dkio.h>
56 #include <sys/stat.h>
57 #include <sys/lock.h>
58 #include <sys/conf.h>
59 #include <sys/fcntl.h>
60 #include <sys/vnode.h>
61 #include <sys/syslog.h>
62 #if NRND > 0
63 #include <sys/rnd.h>
64 #endif
65
66 #include <dev/ldvar.h>
67
68 static void ldgetdefaultlabel(struct ld_softc *, struct disklabel *);
69 static void ldgetdisklabel(struct ld_softc *);
70 static int ldlock(struct ld_softc *);
71 static void ldminphys(struct buf *bp);
72 static void ldshutdown(void *);
73 static int ldstart(struct ld_softc *, struct buf *);
74 static void ldunlock(struct ld_softc *);
75
76 extern struct cfdriver ld_cd;
77
78 static struct dkdriver lddkdriver = { ldstrategy };
79 static void *ld_sdh;
80
81 void
82 ldattach(struct ld_softc *sc)
83 {
84 char buf[9];
85
86 /* Initialise and attach the disk structure. */
87 sc->sc_dk.dk_driver = &lddkdriver;
88 sc->sc_dk.dk_name = sc->sc_dv.dv_xname;
89 disk_attach(&sc->sc_dk);
90
91 if ((sc->sc_flags & LDF_ENABLED) == 0) {
92 printf("%s: disabled\n", sc->sc_dv.dv_xname);
93 return;
94 }
95 if (sc->sc_maxxfer > MAXPHYS)
96 sc->sc_maxxfer = MAXPHYS;
97
98 format_bytes(buf, sizeof(buf), (u_int64_t)sc->sc_secperunit *
99 sc->sc_secsize);
100 printf("%s: %s, %d cyl, %d head, %d sec, %d bytes/sect x %d sectors\n",
101 sc->sc_dv.dv_xname, buf, sc->sc_ncylinders, sc->sc_nheads,
102 sc->sc_nsectors, sc->sc_secsize, sc->sc_secperunit);
103
104 #if NRND > 0
105 /* Attach the device into the rnd source list. */
106 rnd_attach_source(&sc->sc_rnd_source, sc->sc_dv.dv_xname,
107 RND_TYPE_DISK, 0);
108 #endif
109
110 /* Set the `shutdownhook'. */
111 if (ld_sdh == NULL)
112 ld_sdh = shutdownhook_establish(ldshutdown, NULL);
113 BUFQ_INIT(&sc->sc_bufq);
114 }
115
116 int
117 lddrain(struct ld_softc *sc, int flags)
118 {
119 int s;
120
121 if ((flags & DETACH_FORCE) == 0 && sc->sc_dk.dk_openmask != 0)
122 return (EBUSY);
123
124 s = splbio();
125 sc->sc_flags |= LDF_DRAIN;
126 splx(s);
127 return (0);
128 }
129
130 void
131 lddetach(struct ld_softc *sc)
132 {
133 struct buf *bp;
134 int s, bmaj, cmaj, mn;
135
136 /* Wait for commands queued with the hardware to complete. */
137 if (sc->sc_queuecnt != 0)
138 tsleep(&sc->sc_queuecnt, PRIBIO, "lddrn", 30 * hz);
139
140 /* Locate the major numbers. */
141 for (bmaj = 0; bmaj <= nblkdev; bmaj++)
142 if (bdevsw[bmaj].d_open == sdopen)
143 break;
144 for (cmaj = 0; cmaj <= nchrdev; cmaj++)
145 if (cdevsw[cmaj].d_open == sdopen)
146 break;
147
148 /* Kill off any queued buffers. */
149 s = splbio();
150 while ((bp = BUFQ_FIRST(&sc->sc_bufq)) != NULL) {
151 BUFQ_REMOVE(&sc->sc_bufq, bp);
152 bp->b_error = EIO;
153 bp->b_flags |= B_ERROR;
154 bp->b_resid = bp->b_bcount;
155 biodone(bp);
156 }
157 splx(s);
158
159 /* Nuke the vnodes for any open instances. */
160 mn = DISKUNIT(sc->sc_dv.dv_unit);
161 vdevgone(bmaj, mn, mn + (MAXPARTITIONS - 1), VBLK);
162 vdevgone(cmaj, mn, mn + (MAXPARTITIONS - 1), VCHR);
163
164 /* Detach from the disk list. */
165 disk_detach(&sc->sc_dk);
166
167 #if NRND > 0
168 /* Unhook the entropy source. */
169 rnd_detach_source(&sc->sc_rnd_source);
170 #endif
171
172 /* Flush the device's cache. */
173 if (sc->sc_flush != NULL)
174 if ((*sc->sc_flush)(sc) != 0)
175 printf("%s: unable to flush cache\n",
176 sc->sc_dv.dv_xname);
177 }
178
179 static void
180 ldshutdown(void *cookie)
181 {
182 struct ld_softc *sc;
183 int i;
184
185 for (i = 0; i < ld_cd.cd_ndevs; i++) {
186 if ((sc = device_lookup(&ld_cd, i)) == NULL)
187 continue;
188 if (sc->sc_flush != NULL && (*sc->sc_flush)(sc) != 0)
189 printf("%s: unable to flush cache\n",
190 sc->sc_dv.dv_xname);
191 }
192 }
193
194 int
195 ldopen(dev_t dev, int flags, int fmt, struct proc *p)
196 {
197 struct ld_softc *sc;
198 int unit, part;
199
200 unit = DISKUNIT(dev);
201 if ((sc = device_lookup(&ld_cd, unit))== NULL)
202 return (ENXIO);
203 if ((sc->sc_flags & LDF_ENABLED) == 0)
204 return (ENODEV);
205 part = DISKPART(dev);
206 ldlock(sc);
207
208 if (sc->sc_dk.dk_openmask == 0)
209 ldgetdisklabel(sc);
210
211 /* Check that the partition exists. */
212 if (part != RAW_PART && (part >= sc->sc_dk.dk_label->d_npartitions ||
213 sc->sc_dk.dk_label->d_partitions[part].p_fstype == FS_UNUSED)) {
214 ldunlock(sc);
215 return (ENXIO);
216 }
217
218 /* Ensure only one open at a time. */
219 switch (fmt) {
220 case S_IFCHR:
221 sc->sc_dk.dk_copenmask |= (1 << part);
222 break;
223 case S_IFBLK:
224 sc->sc_dk.dk_bopenmask |= (1 << part);
225 break;
226 }
227 sc->sc_dk.dk_openmask =
228 sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
229
230 ldunlock(sc);
231 return (0);
232 }
233
234 int
235 ldclose(dev_t dev, int flags, int fmt, struct proc *p)
236 {
237 struct ld_softc *sc;
238 int part, unit;
239
240 unit = DISKUNIT(dev);
241 part = DISKPART(dev);
242 sc = device_lookup(&ld_cd, unit);
243 ldlock(sc);
244
245 switch (fmt) {
246 case S_IFCHR:
247 sc->sc_dk.dk_copenmask &= ~(1 << part);
248 break;
249 case S_IFBLK:
250 sc->sc_dk.dk_bopenmask &= ~(1 << part);
251 break;
252 }
253 sc->sc_dk.dk_openmask =
254 sc->sc_dk.dk_copenmask | sc->sc_dk.dk_bopenmask;
255
256 if (sc->sc_dk.dk_openmask == 0 && sc->sc_flush != NULL)
257 if ((*sc->sc_flush)(sc) != 0)
258 printf("%s: unable to flush cache\n",
259 sc->sc_dv.dv_xname);
260
261 ldunlock(sc);
262 return (0);
263 }
264
265 int
266 ldread(dev_t dev, struct uio *uio, int ioflag)
267 {
268
269 return (physio(ldstrategy, NULL, dev, B_READ, ldminphys, uio));
270 }
271
272 int
273 ldwrite(dev_t dev, struct uio *uio, int ioflag)
274 {
275
276 return (physio(ldstrategy, NULL, dev, B_WRITE, ldminphys, uio));
277 }
278
279 int
280 ldioctl(dev_t dev, u_long cmd, caddr_t addr, int32_t flag, struct proc *p)
281 {
282 struct ld_softc *sc;
283 int part, unit, error;
284 #ifdef __HAVE_OLD_DISKLABEL
285 struct disklabel newlabel;
286 #endif
287 struct disklabel *lp;
288
289 unit = DISKUNIT(dev);
290 part = DISKPART(dev);
291 sc = device_lookup(&ld_cd, unit);
292 error = 0;
293
294 switch (cmd) {
295 case DIOCGDINFO:
296 memcpy(addr, sc->sc_dk.dk_label, sizeof(struct disklabel));
297 return (0);
298
299 #ifdef __HAVE_OLD_DISKLABEL
300 case ODIOCGDINFO:
301 newlabel = *(sc->sc_dk.dk_label);
302 if (newlabel.d_npartitions > OLDMAXPARTITIONS)
303 return ENOTTY;
304 memcpy(addr, &newlabel, sizeof(struct olddisklabel));
305 return (0);
306 #endif
307
308 case DIOCGPART:
309 ((struct partinfo *)addr)->disklab = sc->sc_dk.dk_label;
310 ((struct partinfo *)addr)->part =
311 &sc->sc_dk.dk_label->d_partitions[part];
312 break;
313
314 case DIOCWDINFO:
315 case DIOCSDINFO:
316 #ifdef __HAVE_OLD_DISKLABEL
317 case ODIOCWDINFO:
318 case ODIOCSDINFO:
319
320 if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
321 memset(&newlabel, 0, sizeof newlabel);
322 memcpy(&newlabel, addr, sizeof (struct olddisklabel));
323 lp = &newlabel;
324 } else
325 #endif
326 lp = (struct disklabel *)addr;
327
328 if ((flag & FWRITE) == 0)
329 return (EBADF);
330
331 if ((error = ldlock(sc)) != 0)
332 return (error);
333 sc->sc_flags |= LDF_LABELLING;
334
335 error = setdisklabel(sc->sc_dk.dk_label,
336 lp, /*sc->sc_dk.dk_openmask : */0,
337 sc->sc_dk.dk_cpulabel);
338 if (error == 0 && (cmd == DIOCWDINFO
339 #ifdef __HAVE_OLD_DISKLABEL
340 || cmd == ODIOCWDINFO
341 #endif
342 ))
343 error = writedisklabel(
344 MAKEDISKDEV(major(dev), DISKUNIT(dev), RAW_PART),
345 ldstrategy, sc->sc_dk.dk_label,
346 sc->sc_dk.dk_cpulabel);
347
348 sc->sc_flags &= ~LDF_LABELLING;
349 ldunlock(sc);
350 break;
351
352 case DIOCWLABEL:
353 if ((flag & FWRITE) == 0)
354 return (EBADF);
355 if (*(int *)addr)
356 sc->sc_flags |= LDF_WLABEL;
357 else
358 sc->sc_flags &= ~LDF_WLABEL;
359 break;
360
361 case DIOCGDEFLABEL:
362 ldgetdefaultlabel(sc, (struct disklabel *)addr);
363 break;
364
365 #ifdef __HAVE_OLD_DISKLABEL
366 case ODIOCGDEFLABEL:
367 ldgetdefaultlabel(sc, &newlabel);
368 if (newlabel.d_npartitions > OLDMAXPARTITIONS)
369 return ENOTTY;
370 memcpy(addr, &newlabel, sizeof (struct olddisklabel));
371 break;
372 #endif
373
374 default:
375 error = ENOTTY;
376 break;
377 }
378
379 return (error);
380 }
381
382 void
383 ldstrategy(struct buf *bp)
384 {
385 struct ld_softc *sc;
386 int s;
387
388 sc = device_lookup(&ld_cd, DISKUNIT(bp->b_dev));
389
390 s = splbio();
391 if (sc->sc_queuecnt == sc->sc_maxqueuecnt) {
392 BUFQ_INSERT_TAIL(&sc->sc_bufq, bp);
393 splx(s);
394 return;
395 }
396 splx(s);
397 ldstart(sc, bp);
398 }
399
400 static int
401 ldstart(struct ld_softc *sc, struct buf *bp)
402 {
403 struct disklabel *lp;
404 int part, s, rv;
405
406 if ((sc->sc_flags & LDF_DRAIN) != 0) {
407 bp->b_error = EIO;
408 bp->b_flags |= B_ERROR;
409 bp->b_resid = bp->b_bcount;
410 biodone(bp);
411 return (-1);
412 }
413
414 part = DISKPART(bp->b_dev);
415 lp = sc->sc_dk.dk_label;
416
417 /*
418 * The transfer must be a whole number of blocks and the offset must
419 * not be negative.
420 */
421 if ((bp->b_bcount % lp->d_secsize) != 0 || bp->b_blkno < 0) {
422 bp->b_flags |= B_ERROR;
423 biodone(bp);
424 return (-1);
425 }
426
427 /*
428 * If it's a null transfer, return.
429 */
430 if (bp->b_bcount == 0) {
431 bp->b_resid = bp->b_bcount;
432 biodone(bp);
433 return (-1);
434 }
435
436 /*
437 * Do bounds checking and adjust the transfer. If error, process.
438 * If past the end of partition, just return.
439 */
440 if (part != RAW_PART &&
441 bounds_check_with_label(bp, lp,
442 (sc->sc_flags & (LDF_WLABEL | LDF_LABELLING)) != 0) <= 0) {
443 bp->b_resid = bp->b_bcount;
444 biodone(bp);
445 return (-1);
446 }
447
448 /*
449 * Convert the logical block number to a physical one and put it in
450 * terms of the device's logical block size.
451 */
452 if (lp->d_secsize >= DEV_BSIZE)
453 bp->b_rawblkno = bp->b_blkno / (lp->d_secsize / DEV_BSIZE);
454 else
455 bp->b_rawblkno = bp->b_blkno * (DEV_BSIZE / lp->d_secsize);
456
457 if (bp->b_dev != RAW_PART)
458 bp->b_rawblkno += lp->d_partitions[part].p_offset;
459
460 s = splbio();
461 disk_busy(&sc->sc_dk);
462 sc->sc_queuecnt++;
463 splx(s);
464
465 if ((rv = (*sc->sc_start)(sc, bp)) != 0) {
466 bp->b_error = rv;
467 bp->b_flags |= B_ERROR;
468 bp->b_resid = bp->b_bcount;
469 s = splbio();
470 lddone(sc, bp);
471 splx(s);
472 }
473
474 return (0);
475 }
476
477 void
478 lddone(struct ld_softc *sc, struct buf *bp)
479 {
480
481 if ((bp->b_flags & B_ERROR) != 0) {
482 diskerr(bp, "ld", "error", LOG_PRINTF, 0, sc->sc_dk.dk_label);
483 printf("\n");
484 }
485
486 disk_unbusy(&sc->sc_dk, bp->b_bcount - bp->b_resid);
487 #if NRND > 0
488 rnd_add_uint32(&sc->sc_rnd_source, bp->b_rawblkno);
489 #endif
490 biodone(bp);
491 if (--sc->sc_queuecnt == 0 && (sc->sc_flags & LDF_DRAIN) != 0)
492 wakeup(&sc->sc_queuecnt);
493
494 while ((bp = BUFQ_FIRST(&sc->sc_bufq)) != NULL) {
495 BUFQ_REMOVE(&sc->sc_bufq, bp);
496 if (!ldstart(sc, bp))
497 break;
498 }
499 }
500
501 int
502 ldsize(dev_t dev)
503 {
504 struct ld_softc *sc;
505 int part, unit, omask, size;
506
507 unit = DISKUNIT(dev);
508 if ((sc = device_lookup(&ld_cd, unit)) == NULL)
509 return (ENODEV);
510 if ((sc->sc_flags & LDF_ENABLED) == 0)
511 return (ENODEV);
512 part = DISKPART(dev);
513
514 omask = sc->sc_dk.dk_openmask & (1 << part);
515
516 if (omask == 0 && ldopen(dev, 0, S_IFBLK, NULL) != 0)
517 return (-1);
518 else if (sc->sc_dk.dk_label->d_partitions[part].p_fstype != FS_SWAP)
519 size = -1;
520 else
521 size = sc->sc_dk.dk_label->d_partitions[part].p_size *
522 (sc->sc_dk.dk_label->d_secsize / DEV_BSIZE);
523 if (omask == 0 && ldclose(dev, 0, S_IFBLK, NULL) != 0)
524 return (-1);
525
526 return (size);
527 }
528
529 /*
530 * Load the label information from the specified device.
531 */
532 static void
533 ldgetdisklabel(struct ld_softc *sc)
534 {
535 const char *errstring;
536
537 ldgetdefaultlabel(sc, sc->sc_dk.dk_label);
538
539 /* Call the generic disklabel extraction routine. */
540 errstring = readdisklabel(MAKEDISKDEV(0, sc->sc_dv.dv_unit, RAW_PART),
541 ldstrategy, sc->sc_dk.dk_label, sc->sc_dk.dk_cpulabel);
542 if (errstring != NULL)
543 printf("%s: %s\n", sc->sc_dv.dv_xname, errstring);
544 }
545
546 /*
547 * Construct a ficticious label.
548 */
549 static void
550 ldgetdefaultlabel(struct ld_softc *sc, struct disklabel *lp)
551 {
552
553 memset(lp, 0, sizeof(struct disklabel));
554
555 lp->d_secsize = sc->sc_secsize;
556 lp->d_ntracks = sc->sc_nheads;
557 lp->d_nsectors = sc->sc_nsectors;
558 lp->d_ncylinders = sc->sc_ncylinders;
559 lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
560 lp->d_type = DTYPE_LD;
561 strcpy(lp->d_typename, "unknown");
562 strcpy(lp->d_packname, "fictitious");
563 lp->d_secperunit = sc->sc_secperunit;
564 lp->d_rpm = 7200;
565 lp->d_interleave = 1;
566 lp->d_flags = 0;
567
568 lp->d_partitions[RAW_PART].p_offset = 0;
569 lp->d_partitions[RAW_PART].p_size =
570 lp->d_secperunit * (lp->d_secsize / DEV_BSIZE);
571 lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
572 lp->d_npartitions = RAW_PART + 1;
573
574 lp->d_magic = DISKMAGIC;
575 lp->d_magic2 = DISKMAGIC;
576 lp->d_checksum = dkcksum(lp);
577 }
578
579 /*
580 * Wait interruptibly for an exclusive lock.
581 *
582 * XXX Several drivers do this; it should be abstracted and made MP-safe.
583 */
584 static int
585 ldlock(struct ld_softc *sc)
586 {
587 int error;
588
589 while ((sc->sc_flags & LDF_LKHELD) != 0) {
590 sc->sc_flags |= LDF_LKWANTED;
591 if ((error = tsleep(sc, PRIBIO | PCATCH, "ldlck", 0)) != 0)
592 return (error);
593 }
594 sc->sc_flags |= LDF_LKHELD;
595 return (0);
596 }
597
598 /*
599 * Unlock and wake up any waiters.
600 */
601 static void
602 ldunlock(struct ld_softc *sc)
603 {
604
605 sc->sc_flags &= ~LDF_LKHELD;
606 if ((sc->sc_flags & LDF_LKWANTED) != 0) {
607 sc->sc_flags &= ~LDF_LKWANTED;
608 wakeup(sc);
609 }
610 }
611
612 /*
613 * Take a dump.
614 */
615 int
616 lddump(dev_t dev, daddr_t blkno, caddr_t va, size_t size)
617 {
618 struct ld_softc *sc;
619 struct disklabel *lp;
620 int unit, part, nsects, sectoff, towrt, nblk, maxblkcnt, rv;
621 static int dumping;
622
623 unit = DISKUNIT(dev);
624 if ((sc = device_lookup(&ld_cd, unit)) == NULL)
625 return (ENXIO);
626 if ((sc->sc_flags & LDF_ENABLED) == 0)
627 return (ENODEV);
628 if (sc->sc_dump == NULL)
629 return (ENXIO);
630
631 /* Check if recursive dump; if so, punt. */
632 if (dumping)
633 return (EFAULT);
634 dumping = 1;
635
636 /* Convert to disk sectors. Request must be a multiple of size. */
637 part = DISKPART(dev);
638 lp = sc->sc_dk.dk_label;
639 if ((size % lp->d_secsize) != 0)
640 return (EFAULT);
641 towrt = size / lp->d_secsize;
642 blkno = dbtob(blkno) / lp->d_secsize; /* blkno in DEV_BSIZE units */
643
644 nsects = lp->d_partitions[part].p_size;
645 sectoff = lp->d_partitions[part].p_offset;
646
647 /* Check transfer bounds against partition size. */
648 if ((blkno < 0) || ((blkno + towrt) > nsects))
649 return (EINVAL);
650
651 /* Offset block number to start of partition. */
652 blkno += sectoff;
653
654 /* Start dumping and return when done. */
655 maxblkcnt = sc->sc_maxxfer / sc->sc_secsize - 1;
656 while (towrt > 0) {
657 nblk = min(maxblkcnt, towrt);
658
659 if ((rv = (*sc->sc_dump)(sc, va, blkno, nblk)) != 0)
660 return (rv);
661
662 towrt -= nblk;
663 blkno += nblk;
664 va += nblk * sc->sc_secsize;
665 }
666
667 dumping = 0;
668 return (0);
669 }
670
671 /*
672 * Adjust the size of a transfer.
673 */
674 static void
675 ldminphys(struct buf *bp)
676 {
677 struct ld_softc *sc;
678
679 sc = device_lookup(&ld_cd, DISKUNIT(bp->b_dev));
680
681 if (bp->b_bcount > sc->sc_maxxfer)
682 bp->b_bcount = sc->sc_maxxfer;
683 minphys(bp);
684 }
685