1 /* $NetBSD: dksubr.c,v 1.114 2023/07/11 23:26:41 christos Exp $ */
2
3 /*-
4 * Copyright (c) 1996, 1997, 1998, 1999, 2002, 2008 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 and Roland C. Dowdeswell.
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 *
19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29 * POSSIBILITY OF SUCH DAMAGE.
30 */
31
32 #include <sys/cdefs.h>
33 __KERNEL_RCSID(0, "$NetBSD: dksubr.c,v 1.114 2023/07/11 23:26:41 christos Exp $");
34
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/stat.h>
38 #include <sys/proc.h>
39 #include <sys/ioctl.h>
40 #include <sys/device.h>
41 #include <sys/disk.h>
42 #include <sys/disklabel.h>
43 #include <sys/buf.h>
44 #include <sys/bufq.h>
45 #include <sys/vnode.h>
46 #include <sys/fcntl.h>
47 #include <sys/namei.h>
48 #include <sys/module.h>
49 #include <sys/syslog.h>
50
51 #include <dev/dkvar.h>
52 #include <miscfs/specfs/specdev.h> /* for v_rdev */
53
54 int dkdebug = 0;
55
56 #ifdef DEBUG
57 #define DKDB_FOLLOW 0x1
58 #define DKDB_INIT 0x2
59 #define DKDB_VNODE 0x4
60 #define DKDB_DUMP 0x8
61
62 #define IFDEBUG(x,y) if (dkdebug & (x)) y
63 #define DPRINTF(x,y) IFDEBUG(x, printf y)
64 #define DPRINTF_FOLLOW(y) DPRINTF(DKDB_FOLLOW, y)
65 #else
66 #define IFDEBUG(x,y)
67 #define DPRINTF(x,y)
68 #define DPRINTF_FOLLOW(y)
69 #endif
70
71 #define DKF_READYFORDUMP (DKF_INITED|DKF_TAKEDUMP)
72
73 static int dk_subr_modcmd(modcmd_t, void *);
74
75 #define DKLABELDEV(dev) \
76 (MAKEDISKDEV(major((dev)), DISKUNIT((dev)), RAW_PART))
77
78 static void dk_makedisklabel(struct dk_softc *);
79 static int dk_translate(struct dk_softc *, struct buf *);
80
81 void
dk_init(struct dk_softc * dksc,device_t dev,int dtype)82 dk_init(struct dk_softc *dksc, device_t dev, int dtype)
83 {
84
85 memset(dksc, 0x0, sizeof(*dksc));
86 dksc->sc_dtype = dtype;
87 dksc->sc_dev = dev;
88
89 strlcpy(dksc->sc_xname, device_xname(dev), DK_XNAME_SIZE);
90 dksc->sc_dkdev.dk_name = dksc->sc_xname;
91 }
92
93 void
dk_attach(struct dk_softc * dksc)94 dk_attach(struct dk_softc *dksc)
95 {
96 KASSERT(dksc->sc_dev != NULL);
97
98 mutex_init(&dksc->sc_iolock, MUTEX_DEFAULT, IPL_VM);
99 dksc->sc_flags |= DKF_READYFORDUMP;
100 #ifdef DIAGNOSTIC
101 dksc->sc_flags |= DKF_WARNLABEL | DKF_LABELSANITY;
102 #endif
103
104 if ((dksc->sc_flags & DKF_NO_RND) == 0) {
105 /* Attach the device into the rnd source list. */
106 rnd_attach_source(&dksc->sc_rnd_source, dksc->sc_xname,
107 RND_TYPE_DISK, RND_FLAG_DEFAULT);
108 }
109 }
110
111 void
dk_detach(struct dk_softc * dksc)112 dk_detach(struct dk_softc *dksc)
113 {
114 if ((dksc->sc_flags & DKF_NO_RND) == 0) {
115 /* Unhook the entropy source. */
116 rnd_detach_source(&dksc->sc_rnd_source);
117 }
118
119 dksc->sc_flags &= ~DKF_READYFORDUMP;
120 mutex_destroy(&dksc->sc_iolock);
121 }
122
123 /* ARGSUSED */
124 int
dk_open(struct dk_softc * dksc,dev_t dev,int flags,int fmt,struct lwp * l)125 dk_open(struct dk_softc *dksc, dev_t dev,
126 int flags, int fmt, struct lwp *l)
127 {
128 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver;
129 struct disklabel *lp = dksc->sc_dkdev.dk_label;
130 int part = DISKPART(dev);
131 int pmask = 1 << part;
132 int ret = 0;
133 struct disk *dk = &dksc->sc_dkdev;
134
135 DPRINTF_FOLLOW(("%s(%s, %p, 0x%"PRIx64", 0x%x)\n", __func__,
136 dksc->sc_xname, dksc, dev, flags));
137
138 mutex_enter(&dk->dk_openlock);
139
140 /*
141 * If there are wedges, and this is not RAW_PART, then we
142 * need to fail.
143 */
144 if (dk->dk_nwedges != 0 && part != RAW_PART) {
145 ret = EBUSY;
146 goto done;
147 }
148
149 /* If no dkdriver attached, bail */
150 if (dkd == NULL) {
151 ret = ENXIO;
152 goto done;
153 }
154
155 /*
156 * initialize driver for the first opener
157 */
158 if (dk->dk_openmask == 0 && dkd->d_firstopen != NULL) {
159 ret = (*dkd->d_firstopen)(dksc->sc_dev, dev, flags, fmt);
160 if (ret)
161 goto done;
162 }
163
164 /*
165 * If we're init'ed and there are no other open partitions then
166 * update the in-core disklabel.
167 */
168 if ((dksc->sc_flags & DKF_INITED)) {
169 if ((dksc->sc_flags & DKF_VLABEL) == 0) {
170 dksc->sc_flags |= DKF_VLABEL;
171 dk_getdisklabel(dksc, dev);
172 }
173 }
174
175 /* Fail if we can't find the partition. */
176 if (part != RAW_PART &&
177 ((dksc->sc_flags & DKF_VLABEL) == 0 ||
178 part >= lp->d_npartitions ||
179 lp->d_partitions[part].p_fstype == FS_UNUSED)) {
180 ret = ENXIO;
181 goto done;
182 }
183
184 /* Mark our unit as open. */
185 switch (fmt) {
186 case S_IFCHR:
187 dk->dk_copenmask |= pmask;
188 break;
189 case S_IFBLK:
190 dk->dk_bopenmask |= pmask;
191 break;
192 }
193
194 dk->dk_openmask = dk->dk_copenmask | dk->dk_bopenmask;
195
196 done:
197 mutex_exit(&dk->dk_openlock);
198 return ret;
199 }
200
201 /* ARGSUSED */
202 int
dk_close(struct dk_softc * dksc,dev_t dev,int flags,int fmt,struct lwp * l)203 dk_close(struct dk_softc *dksc, dev_t dev,
204 int flags, int fmt, struct lwp *l)
205 {
206 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver;
207 int part = DISKPART(dev);
208 int pmask = 1 << part;
209 struct disk *dk = &dksc->sc_dkdev;
210
211 DPRINTF_FOLLOW(("%s(%s, %p, 0x%"PRIx64", 0x%x)\n", __func__,
212 dksc->sc_xname, dksc, dev, flags));
213
214 mutex_enter(&dk->dk_openlock);
215
216 switch (fmt) {
217 case S_IFCHR:
218 dk->dk_copenmask &= ~pmask;
219 break;
220 case S_IFBLK:
221 dk->dk_bopenmask &= ~pmask;
222 break;
223 }
224 dk->dk_openmask = dk->dk_copenmask | dk->dk_bopenmask;
225
226 if (dk->dk_openmask == 0) {
227 if (dkd->d_lastclose != NULL)
228 (*dkd->d_lastclose)(dksc->sc_dev);
229 if ((dksc->sc_flags & DKF_KLABEL) == 0)
230 dksc->sc_flags &= ~DKF_VLABEL;
231 }
232
233 mutex_exit(&dk->dk_openlock);
234 return 0;
235 }
236
237 static int
dk_translate(struct dk_softc * dksc,struct buf * bp)238 dk_translate(struct dk_softc *dksc, struct buf *bp)
239 {
240 int part;
241 int wlabel;
242 daddr_t blkno;
243 struct disklabel *lp;
244 struct disk *dk;
245 uint64_t numsecs;
246 unsigned secsize;
247
248 lp = dksc->sc_dkdev.dk_label;
249 dk = &dksc->sc_dkdev;
250
251 part = DISKPART(bp->b_dev);
252 numsecs = dk->dk_geom.dg_secperunit;
253 secsize = dk->dk_geom.dg_secsize;
254
255 /*
256 * The transfer must be a whole number of blocks and the offset must
257 * not be negative.
258 */
259 if ((bp->b_bcount % secsize) != 0 || bp->b_blkno < 0) {
260 bp->b_error = EINVAL;
261 goto done;
262 }
263
264 /* If there is nothing to do, then we are done */
265 if (bp->b_bcount == 0)
266 goto done;
267
268 wlabel = dksc->sc_flags & (DKF_WLABEL|DKF_LABELLING);
269 if (part == RAW_PART) {
270 uint64_t numblocks = btodb(numsecs * secsize);
271 if (bounds_check_with_mediasize(bp, DEV_BSIZE, numblocks) <= 0)
272 goto done;
273 } else {
274 if (bounds_check_with_label(&dksc->sc_dkdev, bp, wlabel) <= 0)
275 goto done;
276 }
277
278 /*
279 * Convert the block number to absolute and put it in terms
280 * of the device's logical block size.
281 */
282 if (secsize >= DEV_BSIZE)
283 blkno = bp->b_blkno / (secsize / DEV_BSIZE);
284 else
285 blkno = bp->b_blkno * (DEV_BSIZE / secsize);
286
287 if (part != RAW_PART)
288 blkno += lp->d_partitions[DISKPART(bp->b_dev)].p_offset;
289 bp->b_rawblkno = blkno;
290
291 return -1;
292
293 done:
294 bp->b_resid = bp->b_bcount;
295 return bp->b_error;
296 }
297
298 static int
dk_strategy1(struct dk_softc * dksc,struct buf * bp)299 dk_strategy1(struct dk_softc *dksc, struct buf *bp)
300 {
301 int error;
302
303 DPRINTF_FOLLOW(("%s(%s, %p, %p)\n", __func__,
304 dksc->sc_xname, dksc, bp));
305
306 if (!(dksc->sc_flags & DKF_INITED)) {
307 DPRINTF_FOLLOW(("%s: not inited\n", __func__));
308 bp->b_error = ENXIO;
309 bp->b_resid = bp->b_bcount;
310 biodone(bp);
311 return 1;
312 }
313
314 error = dk_translate(dksc, bp);
315 if (error >= 0) {
316 biodone(bp);
317 return 1;
318 }
319
320 return 0;
321 }
322
323 void
dk_strategy(struct dk_softc * dksc,struct buf * bp)324 dk_strategy(struct dk_softc *dksc, struct buf *bp)
325 {
326 int error;
327
328 error = dk_strategy1(dksc, bp);
329 if (error)
330 return;
331
332 /*
333 * Queue buffer and start unit
334 */
335 dk_start(dksc, bp);
336 }
337
338 int
dk_strategy_defer(struct dk_softc * dksc,struct buf * bp)339 dk_strategy_defer(struct dk_softc *dksc, struct buf *bp)
340 {
341 int error;
342
343 error = dk_strategy1(dksc, bp);
344 if (error)
345 return error;
346
347 /*
348 * Queue buffer only
349 */
350 mutex_enter(&dksc->sc_iolock);
351 disk_wait(&dksc->sc_dkdev);
352 bufq_put(dksc->sc_bufq, bp);
353 mutex_exit(&dksc->sc_iolock);
354
355 return 0;
356 }
357
358 int
dk_strategy_pending(struct dk_softc * dksc)359 dk_strategy_pending(struct dk_softc *dksc)
360 {
361 struct buf *bp;
362
363 if (!(dksc->sc_flags & DKF_INITED)) {
364 DPRINTF_FOLLOW(("%s: not inited\n", __func__));
365 return 0;
366 }
367
368 mutex_enter(&dksc->sc_iolock);
369 bp = bufq_peek(dksc->sc_bufq);
370 mutex_exit(&dksc->sc_iolock);
371
372 return bp != NULL;
373 }
374
375 void
dk_start(struct dk_softc * dksc,struct buf * bp)376 dk_start(struct dk_softc *dksc, struct buf *bp)
377 {
378 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver;
379 int error;
380
381 if (!(dksc->sc_flags & DKF_INITED)) {
382 DPRINTF_FOLLOW(("%s: not inited\n", __func__));
383 return;
384 }
385
386 mutex_enter(&dksc->sc_iolock);
387
388 if (bp != NULL) {
389 bp->b_ci = curcpu();
390 disk_wait(&dksc->sc_dkdev);
391 bufq_put(dksc->sc_bufq, bp);
392 }
393
394 /*
395 * If another thread is running the queue, increment
396 * busy counter to 2 so that the queue is retried,
397 * because the driver may now accept additional
398 * requests.
399 */
400 if (dksc->sc_busy < 2)
401 dksc->sc_busy++;
402 if (dksc->sc_busy > 1)
403 goto done;
404
405 /*
406 * Peeking at the buffer queue and committing the operation
407 * only after success isn't atomic.
408 *
409 * So when a diskstart fails, the buffer is saved
410 * and tried again before the next buffer is fetched.
411 * dk_drain() handles flushing of a saved buffer.
412 *
413 * This keeps order of I/O operations, unlike bufq_put.
414 */
415
416 while (dksc->sc_busy > 0) {
417
418 bp = dksc->sc_deferred;
419 dksc->sc_deferred = NULL;
420
421 if (bp == NULL)
422 bp = bufq_get(dksc->sc_bufq);
423
424 while (bp != NULL) {
425
426 disk_busy(&dksc->sc_dkdev);
427 mutex_exit(&dksc->sc_iolock);
428 error = dkd->d_diskstart(dksc->sc_dev, bp);
429 mutex_enter(&dksc->sc_iolock);
430 if (error == EAGAIN || error == ENOMEM) {
431 /*
432 * Not a disk error. Retry later.
433 */
434 KASSERT(dksc->sc_deferred == NULL);
435 dksc->sc_deferred = bp;
436 disk_unbusy(&dksc->sc_dkdev, 0, (bp->b_flags & B_READ));
437 disk_wait(&dksc->sc_dkdev);
438 break;
439 }
440
441 if (error != 0) {
442 bp->b_error = error;
443 bp->b_resid = bp->b_bcount;
444 mutex_exit(&dksc->sc_iolock);
445 dk_done(dksc, bp);
446 mutex_enter(&dksc->sc_iolock);
447 }
448
449 bp = bufq_get(dksc->sc_bufq);
450 }
451
452 dksc->sc_busy--;
453 }
454 done:
455 mutex_exit(&dksc->sc_iolock);
456 }
457
458 void
dk_done(struct dk_softc * dksc,struct buf * bp)459 dk_done(struct dk_softc *dksc, struct buf *bp)
460 {
461 struct disk *dk = &dksc->sc_dkdev;
462
463 if (bp->b_error != 0) {
464 struct cfdriver *cd = device_cfdriver(dksc->sc_dev);
465
466 diskerr(bp, cd->cd_name, "error", LOG_PRINTF, 0,
467 dk->dk_label);
468 printf("\n");
469 }
470
471 mutex_enter(&dksc->sc_iolock);
472 disk_unbusy(dk, bp->b_bcount - bp->b_resid, (bp->b_flags & B_READ));
473 mutex_exit(&dksc->sc_iolock);
474
475 if ((dksc->sc_flags & DKF_NO_RND) == 0)
476 rnd_add_uint32(&dksc->sc_rnd_source, bp->b_rawblkno);
477
478 biodone(bp);
479 }
480
481 void
dk_drain(struct dk_softc * dksc)482 dk_drain(struct dk_softc *dksc)
483 {
484 struct buf *bp;
485
486 mutex_enter(&dksc->sc_iolock);
487 bp = dksc->sc_deferred;
488 dksc->sc_deferred = NULL;
489 if (bp != NULL) {
490 bp->b_error = EIO;
491 bp->b_resid = bp->b_bcount;
492 biodone(bp);
493 }
494 bufq_drain(dksc->sc_bufq);
495 mutex_exit(&dksc->sc_iolock);
496 }
497
498 int
dk_discard(struct dk_softc * dksc,dev_t dev,off_t pos,off_t len)499 dk_discard(struct dk_softc *dksc, dev_t dev, off_t pos, off_t len)
500 {
501 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver;
502 unsigned secsize = dksc->sc_dkdev.dk_geom.dg_secsize;
503 struct buf tmp, *bp = &tmp;
504 int maxsz;
505 int error = 0;
506
507 KASSERT(len >= 0);
508
509 DPRINTF_FOLLOW(("%s(%s, %p, 0x"PRIx64", %jd, %jd)\n", __func__,
510 dksc->sc_xname, dksc, (intmax_t)pos, (intmax_t)len));
511
512 if (!(dksc->sc_flags & DKF_INITED)) {
513 DPRINTF_FOLLOW(("%s: not inited\n", __func__));
514 return ENXIO;
515 }
516
517 if (secsize == 0 || (pos % secsize) != 0 || (len % secsize) != 0)
518 return EINVAL;
519
520 /* largest value that b_bcount can store */
521 maxsz = rounddown(INT_MAX, secsize);
522
523 while (len > 0) {
524 /* enough data to please the bounds checking code */
525 bp->b_dev = dev;
526 bp->b_blkno = (daddr_t)(pos / secsize);
527 bp->b_bcount = uimin(len, maxsz);
528 bp->b_flags = B_WRITE;
529
530 error = dk_translate(dksc, bp);
531 if (error >= 0)
532 break;
533
534 error = dkd->d_discard(dksc->sc_dev,
535 (off_t)bp->b_rawblkno * secsize,
536 (off_t)bp->b_bcount);
537 if (error)
538 break;
539
540 pos += bp->b_bcount;
541 len -= bp->b_bcount;
542 }
543
544 return error;
545 }
546
547 int
dk_size(struct dk_softc * dksc,dev_t dev)548 dk_size(struct dk_softc *dksc, dev_t dev)
549 {
550 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver;
551 struct disklabel *lp;
552 int is_open;
553 int part;
554 int size;
555
556 if ((dksc->sc_flags & DKF_INITED) == 0)
557 return -1;
558
559 part = DISKPART(dev);
560 is_open = dksc->sc_dkdev.dk_openmask & (1 << part);
561
562 if (!is_open && dkd->d_open(dev, 0, S_IFBLK, curlwp))
563 return -1;
564
565 lp = dksc->sc_dkdev.dk_label;
566 if (lp->d_partitions[part].p_fstype != FS_SWAP)
567 size = -1;
568 else
569 size = lp->d_partitions[part].p_size *
570 (lp->d_secsize / DEV_BSIZE);
571
572 if (!is_open && dkd->d_close(dev, 0, S_IFBLK, curlwp))
573 return -1;
574
575 return size;
576 }
577
578 int
dk_ioctl(struct dk_softc * dksc,dev_t dev,u_long cmd,void * data,int flag,struct lwp * l)579 dk_ioctl(struct dk_softc *dksc, dev_t dev,
580 u_long cmd, void *data, int flag, struct lwp *l)
581 {
582 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver;
583 struct disklabel *lp;
584 struct disk *dk = &dksc->sc_dkdev;
585 #ifdef __HAVE_OLD_DISKLABEL
586 struct disklabel newlabel;
587 #endif
588 int error;
589
590 DPRINTF_FOLLOW(("%s(%s, %p, 0x%"PRIx64", 0x%lx)\n", __func__,
591 dksc->sc_xname, dksc, dev, cmd));
592
593 /* ensure that the pseudo disk is open for writes for these commands */
594 switch (cmd) {
595 case DIOCSDINFO:
596 case DIOCWDINFO:
597 #ifdef __HAVE_OLD_DISKLABEL
598 case ODIOCSDINFO:
599 case ODIOCWDINFO:
600 #endif
601 case DIOCKLABEL:
602 case DIOCWLABEL:
603 case DIOCAWEDGE:
604 case DIOCDWEDGE:
605 case DIOCSSTRATEGY:
606 if ((flag & FWRITE) == 0)
607 return EBADF;
608 }
609
610 /* ensure that the pseudo-disk is initialized for these */
611 switch (cmd) {
612 case DIOCGDINFO:
613 case DIOCSDINFO:
614 case DIOCWDINFO:
615 case DIOCGPARTINFO:
616 case DIOCKLABEL:
617 case DIOCWLABEL:
618 case DIOCGDEFLABEL:
619 case DIOCAWEDGE:
620 case DIOCDWEDGE:
621 case DIOCLWEDGES:
622 case DIOCMWEDGES:
623 case DIOCRMWEDGES:
624 case DIOCCACHESYNC:
625 #ifdef __HAVE_OLD_DISKLABEL
626 case ODIOCGDINFO:
627 case ODIOCSDINFO:
628 case ODIOCWDINFO:
629 case ODIOCGDEFLABEL:
630 #endif
631 if ((dksc->sc_flags & DKF_INITED) == 0)
632 return ENXIO;
633 }
634
635 error = disk_ioctl(dk, dev, cmd, data, flag, l);
636 if (error != EPASSTHROUGH)
637 return error;
638 else
639 error = 0;
640
641 switch (cmd) {
642 case DIOCWDINFO:
643 case DIOCSDINFO:
644 #ifdef __HAVE_OLD_DISKLABEL
645 case ODIOCWDINFO:
646 case ODIOCSDINFO:
647 #endif
648 #ifdef __HAVE_OLD_DISKLABEL
649 if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
650 memset(&newlabel, 0, sizeof newlabel);
651 memcpy(&newlabel, data, sizeof (struct olddisklabel));
652 lp = &newlabel;
653 } else
654 #endif
655 lp = (struct disklabel *)data;
656
657 mutex_enter(&dk->dk_openlock);
658 dksc->sc_flags |= DKF_LABELLING;
659
660 error = setdisklabel(dksc->sc_dkdev.dk_label,
661 lp, 0, dksc->sc_dkdev.dk_cpulabel);
662 if (error == 0) {
663 if (cmd == DIOCWDINFO
664 #ifdef __HAVE_OLD_DISKLABEL
665 || cmd == ODIOCWDINFO
666 #endif
667 )
668 error = writedisklabel(DKLABELDEV(dev),
669 dkd->d_strategy, dksc->sc_dkdev.dk_label,
670 dksc->sc_dkdev.dk_cpulabel);
671 }
672
673 dksc->sc_flags &= ~DKF_LABELLING;
674 mutex_exit(&dk->dk_openlock);
675 break;
676
677 case DIOCKLABEL:
678 if (*(int *)data != 0)
679 dksc->sc_flags |= DKF_KLABEL;
680 else
681 dksc->sc_flags &= ~DKF_KLABEL;
682 break;
683
684 case DIOCWLABEL:
685 if (*(int *)data != 0)
686 dksc->sc_flags |= DKF_WLABEL;
687 else
688 dksc->sc_flags &= ~DKF_WLABEL;
689 break;
690
691 case DIOCGDEFLABEL:
692 dk_getdefaultlabel(dksc, (struct disklabel *)data);
693 break;
694
695 #ifdef __HAVE_OLD_DISKLABEL
696 case ODIOCGDEFLABEL:
697 dk_getdefaultlabel(dksc, &newlabel);
698 if (newlabel.d_npartitions > OLDMAXPARTITIONS)
699 return ENOTTY;
700 memcpy(data, &newlabel, sizeof (struct olddisklabel));
701 break;
702 #endif
703
704 case DIOCGSTRATEGY:
705 {
706 struct disk_strategy *dks = (void *)data;
707
708 mutex_enter(&dksc->sc_iolock);
709 if (dksc->sc_bufq != NULL)
710 strlcpy(dks->dks_name,
711 bufq_getstrategyname(dksc->sc_bufq),
712 sizeof(dks->dks_name));
713 else
714 error = EINVAL;
715 mutex_exit(&dksc->sc_iolock);
716 dks->dks_paramlen = 0;
717 break;
718 }
719
720 case DIOCSSTRATEGY:
721 {
722 struct disk_strategy *dks = (void *)data;
723 struct bufq_state *new;
724 struct bufq_state *old;
725
726 if (dks->dks_param != NULL) {
727 return EINVAL;
728 }
729 dks->dks_name[sizeof(dks->dks_name) - 1] = 0; /* ensure term */
730 error = bufq_alloc(&new, dks->dks_name,
731 BUFQ_EXACT|BUFQ_SORT_RAWBLOCK);
732 if (error) {
733 return error;
734 }
735 mutex_enter(&dksc->sc_iolock);
736 old = dksc->sc_bufq;
737 if (old)
738 bufq_move(new, old);
739 dksc->sc_bufq = new;
740 mutex_exit(&dksc->sc_iolock);
741 if (old)
742 bufq_free(old);
743 break;
744 }
745
746 default:
747 error = ENOTTY;
748 }
749
750 return error;
751 }
752
753 /*
754 * dk_dump dumps all of physical memory into the partition specified.
755 * This requires substantially more framework than {s,w}ddump, and hence
756 * is probably much more fragile.
757 *
758 */
759
760 #define DKFF_READYFORDUMP(x) (((x) & DKF_READYFORDUMP) == DKF_READYFORDUMP)
761 static volatile int dk_dumping = 0;
762
763 /* ARGSUSED */
764 int
dk_dump(struct dk_softc * dksc,dev_t dev,daddr_t blkno,void * vav,size_t size,int flags)765 dk_dump(struct dk_softc *dksc, dev_t dev,
766 daddr_t blkno, void *vav, size_t size, int flags)
767 {
768 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver;
769 struct disk_geom *dg = &dksc->sc_dkdev.dk_geom;
770 char *va = vav;
771 struct disklabel *lp;
772 struct partition *p;
773 int part, towrt, maxblkcnt, nblk;
774 int maxxfer, rv = 0;
775
776 /*
777 * ensure that we consider this device to be safe for dumping,
778 * and that the device is configured.
779 */
780 if (!DKFF_READYFORDUMP(dksc->sc_flags)) {
781 DPRINTF(DKDB_DUMP, ("%s: bad dump flags 0x%x\n", __func__,
782 dksc->sc_flags));
783 return ENXIO;
784 }
785
786 /* ensure that we are not already dumping */
787 if (dk_dumping)
788 return EFAULT;
789 if ((flags & DK_DUMP_RECURSIVE) == 0)
790 dk_dumping = 1;
791
792 if (dkd->d_dumpblocks == NULL) {
793 DPRINTF(DKDB_DUMP, ("%s: no dumpblocks\n", __func__));
794 return ENXIO;
795 }
796
797 /* device specific max transfer size */
798 maxxfer = MAXPHYS;
799 if (dkd->d_iosize != NULL)
800 (*dkd->d_iosize)(dksc->sc_dev, &maxxfer);
801
802 /* Convert to disk sectors. Request must be a multiple of size. */
803 part = DISKPART(dev);
804 lp = dksc->sc_dkdev.dk_label;
805 if ((size % lp->d_secsize) != 0) {
806 DPRINTF(DKDB_DUMP, ("%s: odd size %zu\n", __func__, size));
807 return EFAULT;
808 }
809 towrt = size / lp->d_secsize;
810 blkno = dbtob(blkno) / lp->d_secsize; /* blkno in secsize units */
811
812 p = &lp->d_partitions[part];
813 if (part == RAW_PART) {
814 if (p->p_fstype != FS_UNUSED) {
815 DPRINTF(DKDB_DUMP, ("%s: bad fstype %d\n", __func__,
816 p->p_fstype));
817 return ENXIO;
818 }
819 /* Check whether dump goes to a wedge */
820 if (dksc->sc_dkdev.dk_nwedges == 0) {
821 DPRINTF(DKDB_DUMP, ("%s: dump to raw\n", __func__));
822 return ENXIO;
823 }
824 /* Check transfer bounds against media size */
825 if (blkno < 0 || (blkno + towrt) > dg->dg_secperunit) {
826 DPRINTF(DKDB_DUMP, ("%s: out of bounds blkno=%jd, towrt=%d, "
827 "nsects=%jd\n", __func__, (intmax_t)blkno, towrt, dg->dg_secperunit));
828 return EINVAL;
829 }
830 } else {
831 int nsects, sectoff;
832
833 if (p->p_fstype != FS_SWAP) {
834 DPRINTF(DKDB_DUMP, ("%s: bad fstype %d\n", __func__,
835 p->p_fstype));
836 return ENXIO;
837 }
838 nsects = p->p_size;
839 sectoff = p->p_offset;
840
841 /* Check transfer bounds against partition size. */
842 if ((blkno < 0) || ((blkno + towrt) > nsects)) {
843 DPRINTF(DKDB_DUMP, ("%s: out of bounds blkno=%jd, towrt=%d, "
844 "nsects=%d\n", __func__, (intmax_t)blkno, towrt, nsects));
845 return EINVAL;
846 }
847
848 /* Offset block number to start of partition. */
849 blkno += sectoff;
850 }
851
852 /* Start dumping and return when done. */
853 maxblkcnt = howmany(maxxfer, lp->d_secsize);
854 while (towrt > 0) {
855 nblk = uimin(maxblkcnt, towrt);
856
857 if ((rv = (*dkd->d_dumpblocks)(dksc->sc_dev, va, blkno, nblk))
858 != 0) {
859 DPRINTF(DKDB_DUMP, ("%s: dumpblocks %d\n", __func__,
860 rv));
861 return rv;
862 }
863
864 towrt -= nblk;
865 blkno += nblk;
866 va += nblk * lp->d_secsize;
867 }
868
869 if ((flags & DK_DUMP_RECURSIVE) == 0)
870 dk_dumping = 0;
871
872 return 0;
873 }
874
875 /* ARGSUSED */
876 void
dk_getdefaultlabel(struct dk_softc * dksc,struct disklabel * lp)877 dk_getdefaultlabel(struct dk_softc *dksc, struct disklabel *lp)
878 {
879 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver;
880 struct disk_geom *dg = &dksc->sc_dkdev.dk_geom;
881
882 memset(lp, 0, sizeof(*lp));
883
884 if (dg->dg_secperunit > UINT32_MAX)
885 lp->d_secperunit = UINT32_MAX;
886 else
887 lp->d_secperunit = dg->dg_secperunit;
888 lp->d_secsize = dg->dg_secsize;
889 lp->d_nsectors = dg->dg_nsectors;
890 lp->d_ntracks = dg->dg_ntracks;
891 lp->d_ncylinders = dg->dg_ncylinders;
892 lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
893
894 strlcpy(lp->d_typename, dksc->sc_xname, sizeof(lp->d_typename));
895 lp->d_type = dksc->sc_dtype;
896 strlcpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
897 lp->d_rpm = 3600;
898 lp->d_interleave = 1;
899 lp->d_flags = 0;
900
901 lp->d_partitions[RAW_PART].p_offset = 0;
902 lp->d_partitions[RAW_PART].p_size = lp->d_secperunit;
903 lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
904 lp->d_npartitions = RAW_PART + 1;
905
906 lp->d_magic = DISKMAGIC;
907 lp->d_magic2 = DISKMAGIC;
908
909 if (dkd->d_label)
910 dkd->d_label(dksc->sc_dev, lp);
911
912 lp->d_checksum = dkcksum(lp);
913 }
914
915 /* ARGSUSED */
916 void
dk_getdisklabel(struct dk_softc * dksc,dev_t dev)917 dk_getdisklabel(struct dk_softc *dksc, dev_t dev)
918 {
919 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver;
920 struct disklabel *lp = dksc->sc_dkdev.dk_label;
921 struct cpu_disklabel *clp = dksc->sc_dkdev.dk_cpulabel;
922 struct disk_geom *dg = &dksc->sc_dkdev.dk_geom;
923 struct partition *pp;
924 int i, lpratio, dgratio;
925 const char *errstring;
926
927 memset(clp, 0x0, sizeof(*clp));
928 dk_getdefaultlabel(dksc, lp);
929 errstring = readdisklabel(DKLABELDEV(dev), dkd->d_strategy,
930 dksc->sc_dkdev.dk_label, dksc->sc_dkdev.dk_cpulabel);
931 if (errstring) {
932 dk_makedisklabel(dksc);
933 if (dksc->sc_flags & DKF_WARNLABEL)
934 printf("%s: %s\n", dksc->sc_xname, errstring);
935 return;
936 }
937
938 if ((dksc->sc_flags & DKF_LABELSANITY) == 0)
939 return;
940
941 /* Convert sector counts to multiple of DEV_BSIZE for comparison */
942 lpratio = dgratio = 1;
943 if (lp->d_secsize > DEV_BSIZE)
944 lpratio = lp->d_secsize / DEV_BSIZE;
945 if (dg->dg_secsize > DEV_BSIZE)
946 dgratio = dg->dg_secsize / DEV_BSIZE;
947
948 /* Sanity check */
949 if ((uint64_t)lp->d_secperunit * lpratio > dg->dg_secperunit * dgratio)
950 printf("WARNING: %s: "
951 "total unit size in disklabel (%" PRIu64 ") "
952 "!= the size of %s (%" PRIu64 ")\n", dksc->sc_xname,
953 (uint64_t)lp->d_secperunit * lpratio, dksc->sc_xname,
954 dg->dg_secperunit * dgratio);
955 else if (lp->d_secperunit < UINT32_MAX &&
956 (uint64_t)lp->d_secperunit * lpratio < dg->dg_secperunit * dgratio)
957 printf("%s: %" PRIu64 " trailing sectors not covered"
958 " by disklabel\n", dksc->sc_xname,
959 (dg->dg_secperunit * dgratio)
960 - (lp->d_secperunit * lpratio));
961
962 for (i=0; i < lp->d_npartitions; i++) {
963 uint64_t pend;
964
965 pp = &lp->d_partitions[i];
966 pend = pp->p_offset + pp->p_size;
967 if (pend * lpratio > dg->dg_secperunit * dgratio)
968 printf("WARNING: %s: end of partition `%c' exceeds "
969 "the size of %s (%" PRIu64 ")\n", dksc->sc_xname,
970 'a' + i, dksc->sc_xname,
971 dg->dg_secperunit * dgratio);
972 }
973 }
974
975 /*
976 * Heuristic to conjure a disklabel if reading a disklabel failed.
977 *
978 * This is to allow the raw partition to be used for a filesystem
979 * without caring about the write protected label sector.
980 *
981 * If the driver provides it's own callback, use that instead.
982 */
983 /* ARGSUSED */
984 static void
dk_makedisklabel(struct dk_softc * dksc)985 dk_makedisklabel(struct dk_softc *dksc)
986 {
987 const struct dkdriver *dkd = dksc->sc_dkdev.dk_driver;
988 struct disklabel *lp = dksc->sc_dkdev.dk_label;
989
990 strlcpy(lp->d_packname, "default label", sizeof(lp->d_packname));
991
992 if (dkd->d_label)
993 dkd->d_label(dksc->sc_dev, lp);
994 else
995 lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS;
996
997 lp->d_checksum = dkcksum(lp);
998 }
999
1000 MODULE(MODULE_CLASS_MISC, dk_subr, NULL);
1001
1002 static int
dk_subr_modcmd(modcmd_t cmd,void * arg)1003 dk_subr_modcmd(modcmd_t cmd, void *arg)
1004 {
1005 switch (cmd) {
1006 case MODULE_CMD_INIT:
1007 case MODULE_CMD_FINI:
1008 return 0;
1009 case MODULE_CMD_STAT:
1010 case MODULE_CMD_AUTOUNLOAD:
1011 default:
1012 return ENOTTY;
1013 }
1014 }
1015