1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 2002 Poul-Henning Kamp
5 * Copyright (c) 2002 Networks Associates Technology, Inc.
6 * All rights reserved.
7 *
8 * This software was developed for the FreeBSD Project by Poul-Henning Kamp
9 * and NAI Labs, the Security Research Division of Network Associates, Inc.
10 * under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the
11 * DARPA CHATS research program.
12 *
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
15 * are met:
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 * 3. The names of the authors may not be used to endorse or promote
22 * products derived from this software without specific prior written
23 * permission.
24 *
25 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * SUCH DAMAGE.
36 */
37
38 #include <sys/cdefs.h>
39 #include "opt_geom.h"
40
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/kernel.h>
44 #include <sys/sysctl.h>
45 #include <sys/bio.h>
46 #include <sys/ctype.h>
47 #include <sys/devctl.h>
48 #include <sys/fcntl.h>
49 #include <sys/malloc.h>
50 #include <sys/msan.h>
51 #include <sys/sbuf.h>
52 #include <sys/devicestat.h>
53
54 #include <sys/lock.h>
55 #include <sys/mutex.h>
56 #include <geom/geom.h>
57 #include <geom/geom_disk.h>
58 #include <geom/geom_int.h>
59
60 #include <dev/led/led.h>
61
62 #include <machine/bus.h>
63
64 struct g_disk_softc {
65 struct disk *dp;
66 struct devstat *d_devstat;
67 struct sysctl_ctx_list sysctl_ctx;
68 struct sysctl_oid *sysctl_tree;
69 char led[64];
70 uint32_t state;
71 struct mtx done_mtx;
72 bool flush_notsup_succeed;
73 };
74
75 static g_access_t g_disk_access;
76 static g_start_t g_disk_start;
77 static g_ioctl_t g_disk_ioctl;
78 static g_dumpconf_t g_disk_dumpconf;
79 static g_provgone_t g_disk_providergone;
80
81 static int g_disk_sysctl_flags(SYSCTL_HANDLER_ARGS);
82
83 static struct g_class g_disk_class = {
84 .name = G_DISK_CLASS_NAME,
85 .version = G_VERSION,
86 .start = g_disk_start,
87 .access = g_disk_access,
88 .ioctl = g_disk_ioctl,
89 .providergone = g_disk_providergone,
90 .dumpconf = g_disk_dumpconf,
91 };
92
93 SYSCTL_DECL(_kern_geom);
94 static SYSCTL_NODE(_kern_geom, OID_AUTO, disk, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
95 "GEOM_DISK stuff");
96
97 DECLARE_GEOM_CLASS(g_disk_class, g_disk);
98
99 static int
g_disk_access(struct g_provider * pp,int r,int w,int e)100 g_disk_access(struct g_provider *pp, int r, int w, int e)
101 {
102 struct disk *dp;
103 struct g_disk_softc *sc;
104 int error;
105
106 g_trace(G_T_ACCESS, "g_disk_access(%s, %d, %d, %d)",
107 pp->name, r, w, e);
108 g_topology_assert();
109 sc = pp->private;
110 if ((dp = sc->dp) == NULL || dp->d_destroyed) {
111 /*
112 * Allow decreasing access count even if disk is not
113 * available anymore.
114 */
115 if (r <= 0 && w <= 0 && e <= 0)
116 return (0);
117 return (ENXIO);
118 }
119 r += pp->acr;
120 w += pp->acw;
121 e += pp->ace;
122 error = 0;
123 if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) {
124 /*
125 * It would be better to defer this decision to d_open if
126 * it was able to take flags.
127 */
128 if (w > 0 && (dp->d_flags & DISKFLAG_WRITE_PROTECT) != 0)
129 error = EROFS;
130 if (error == 0 && dp->d_open != NULL)
131 error = dp->d_open(dp);
132 if (bootverbose && error != 0)
133 printf("Opened disk %s -> %d\n", pp->name, error);
134 if (error != 0)
135 return (error);
136 pp->sectorsize = dp->d_sectorsize;
137 if (dp->d_maxsize == 0) {
138 printf("WARNING: Disk drive %s%d has no d_maxsize\n",
139 dp->d_name, dp->d_unit);
140 dp->d_maxsize = DFLTPHYS;
141 }
142 if (dp->d_delmaxsize == 0) {
143 if (bootverbose && dp->d_flags & DISKFLAG_CANDELETE) {
144 printf("WARNING: Disk drive %s%d has no "
145 "d_delmaxsize\n", dp->d_name, dp->d_unit);
146 }
147 dp->d_delmaxsize = dp->d_maxsize;
148 }
149 pp->stripeoffset = dp->d_stripeoffset;
150 pp->stripesize = dp->d_stripesize;
151 dp->d_flags |= DISKFLAG_OPEN;
152 /*
153 * Do not invoke resize event when initial size was zero.
154 * Some disks report its size only after first opening.
155 */
156 if (pp->mediasize == 0)
157 pp->mediasize = dp->d_mediasize;
158 else
159 g_resize_provider(pp, dp->d_mediasize);
160 } else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) {
161 if (dp->d_close != NULL) {
162 error = dp->d_close(dp);
163 if (error != 0)
164 printf("Closed disk %s -> %d\n",
165 pp->name, error);
166 }
167 sc->state = G_STATE_ACTIVE;
168 if (sc->led[0] != 0)
169 led_set(sc->led, "0");
170 dp->d_flags &= ~DISKFLAG_OPEN;
171 }
172 return (error);
173 }
174
175 static void
g_disk_kerneldump(struct bio * bp,struct disk * dp)176 g_disk_kerneldump(struct bio *bp, struct disk *dp)
177 {
178 struct g_kerneldump *gkd;
179 struct g_geom *gp;
180
181 gkd = (struct g_kerneldump*)bp->bio_data;
182 gp = bp->bio_to->geom;
183 g_trace(G_T_TOPOLOGY, "g_disk_kerneldump(%s, %jd, %jd)",
184 gp->name, (intmax_t)gkd->offset, (intmax_t)gkd->length);
185 if (dp->d_dump == NULL) {
186 g_io_deliver(bp, ENODEV);
187 return;
188 }
189 gkd->di.dumper = dp->d_dump;
190 gkd->di.priv = dp;
191 gkd->di.blocksize = dp->d_sectorsize;
192 gkd->di.maxiosize = dp->d_maxsize;
193 gkd->di.mediaoffset = gkd->offset;
194 if ((gkd->offset + gkd->length) > dp->d_mediasize)
195 gkd->length = dp->d_mediasize - gkd->offset;
196 gkd->di.mediasize = gkd->length;
197 g_io_deliver(bp, 0);
198 }
199
200 static void
g_disk_setstate(struct bio * bp,struct g_disk_softc * sc)201 g_disk_setstate(struct bio *bp, struct g_disk_softc *sc)
202 {
203 const char *cmd;
204
205 memcpy(&sc->state, bp->bio_data, sizeof(sc->state));
206 if (sc->led[0] != 0) {
207 switch (sc->state) {
208 case G_STATE_FAILED:
209 cmd = "1";
210 break;
211 case G_STATE_REBUILD:
212 cmd = "f5";
213 break;
214 case G_STATE_RESYNC:
215 cmd = "f1";
216 break;
217 default:
218 cmd = "0";
219 break;
220 }
221 led_set(sc->led, cmd);
222 }
223 g_io_deliver(bp, 0);
224 }
225
226 static void
g_disk_done(struct bio * bp)227 g_disk_done(struct bio *bp)
228 {
229 struct bintime now;
230 struct bio *bp2;
231 struct g_disk_softc *sc;
232
233 /* See "notes" for why we need a mutex here */
234 sc = bp->bio_caller1;
235 bp2 = bp->bio_parent;
236 binuptime(&now);
237 mtx_lock(&sc->done_mtx);
238 if (bp2->bio_error == 0) {
239 if ((bp->bio_flags & BIO_EXTERR) != 0) {
240 bp2->bio_flags |= BIO_EXTERR;
241 bp2->bio_exterr = bp->bio_exterr;
242 } else {
243 bp2->bio_error = bp->bio_error;
244 }
245 }
246 bp2->bio_completed += bp->bio_length - bp->bio_resid;
247
248 if (bp->bio_cmd == BIO_READ)
249 kmsan_check(bp2->bio_data, bp2->bio_completed, "g_disk_done");
250
251 switch (bp->bio_cmd) {
252 case BIO_ZONE:
253 bcopy(&bp->bio_zone, &bp2->bio_zone, sizeof(bp->bio_zone));
254 /*FALLTHROUGH*/
255 case BIO_READ:
256 case BIO_WRITE:
257 case BIO_DELETE:
258 case BIO_FLUSH:
259 devstat_end_transaction_bio_bt(sc->d_devstat, bp, &now);
260 break;
261 default:
262 break;
263 }
264 bp2->bio_inbed++;
265 if (bp2->bio_children == bp2->bio_inbed) {
266 mtx_unlock(&sc->done_mtx);
267 bp2->bio_resid = bp2->bio_bcount - bp2->bio_completed;
268 g_io_deliver(bp2, bp2->bio_error);
269 } else
270 mtx_unlock(&sc->done_mtx);
271 g_destroy_bio(bp);
272 }
273
274 static int
g_disk_ioctl(struct g_provider * pp,u_long cmd,void * data,int fflag,struct thread * td)275 g_disk_ioctl(struct g_provider *pp, u_long cmd, void * data, int fflag, struct thread *td)
276 {
277 struct disk *dp;
278 struct g_disk_softc *sc;
279
280 sc = pp->private;
281 dp = sc->dp;
282 KASSERT(dp != NULL && !dp->d_destroyed,
283 ("g_disk_ioctl(%lx) on destroyed disk %s", cmd, pp->name));
284
285 if (dp->d_ioctl == NULL)
286 return (ENOIOCTL);
287 return (dp->d_ioctl(dp, cmd, data, fflag, td));
288 }
289
290 static off_t
g_disk_maxsize(struct disk * dp,struct bio * bp)291 g_disk_maxsize(struct disk *dp, struct bio *bp)
292 {
293 if (bp->bio_cmd == BIO_DELETE)
294 return (dp->d_delmaxsize);
295 return (dp->d_maxsize);
296 }
297
298 static int
g_disk_maxsegs(struct disk * dp,struct bio * bp)299 g_disk_maxsegs(struct disk *dp, struct bio *bp)
300 {
301 return ((g_disk_maxsize(dp, bp) / PAGE_SIZE) + 1);
302 }
303
304 static void
g_disk_advance(struct disk * dp,struct bio * bp,off_t off)305 g_disk_advance(struct disk *dp, struct bio *bp, off_t off)
306 {
307
308 bp->bio_offset += off;
309 bp->bio_length -= off;
310
311 if ((bp->bio_flags & BIO_VLIST) != 0) {
312 bus_dma_segment_t *seg, *end;
313
314 seg = (bus_dma_segment_t *)bp->bio_data;
315 end = (bus_dma_segment_t *)bp->bio_data + bp->bio_ma_n;
316 off += bp->bio_ma_offset;
317 while (off >= seg->ds_len) {
318 KASSERT((seg != end),
319 ("vlist request runs off the end"));
320 off -= seg->ds_len;
321 seg++;
322 }
323 bp->bio_ma_offset = off;
324 bp->bio_ma_n = end - seg;
325 bp->bio_data = (void *)seg;
326 } else if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
327 bp->bio_ma += off / PAGE_SIZE;
328 bp->bio_ma_offset += off;
329 bp->bio_ma_offset %= PAGE_SIZE;
330 bp->bio_ma_n -= off / PAGE_SIZE;
331 } else {
332 bp->bio_data += off;
333 }
334 }
335
336 static void
g_disk_seg_limit(bus_dma_segment_t * seg,off_t * poffset,off_t * plength,int * ppages)337 g_disk_seg_limit(bus_dma_segment_t *seg, off_t *poffset,
338 off_t *plength, int *ppages)
339 {
340 uintptr_t seg_page_base;
341 uintptr_t seg_page_end;
342 off_t offset;
343 off_t length;
344 int seg_pages;
345
346 offset = *poffset;
347 length = *plength;
348
349 if (length > seg->ds_len - offset)
350 length = seg->ds_len - offset;
351
352 seg_page_base = trunc_page(seg->ds_addr + offset);
353 seg_page_end = round_page(seg->ds_addr + offset + length);
354 seg_pages = (seg_page_end - seg_page_base) >> PAGE_SHIFT;
355
356 if (seg_pages > *ppages) {
357 seg_pages = *ppages;
358 length = (seg_page_base + (seg_pages << PAGE_SHIFT)) -
359 (seg->ds_addr + offset);
360 }
361
362 *poffset = 0;
363 *plength -= length;
364 *ppages -= seg_pages;
365 }
366
367 static off_t
g_disk_vlist_limit(struct disk * dp,struct bio * bp,bus_dma_segment_t ** pendseg)368 g_disk_vlist_limit(struct disk *dp, struct bio *bp, bus_dma_segment_t **pendseg)
369 {
370 bus_dma_segment_t *seg, *end __diagused;
371 off_t residual;
372 off_t offset;
373 int pages;
374
375 seg = (bus_dma_segment_t *)bp->bio_data;
376 end = (bus_dma_segment_t *)bp->bio_data + bp->bio_ma_n;
377 residual = bp->bio_length;
378 offset = bp->bio_ma_offset;
379 pages = g_disk_maxsegs(dp, bp);
380 while (residual != 0 && pages != 0) {
381 KASSERT((seg != end),
382 ("vlist limit runs off the end"));
383 g_disk_seg_limit(seg, &offset, &residual, &pages);
384 seg++;
385 }
386 if (pendseg != NULL)
387 *pendseg = seg;
388 return (residual);
389 }
390
391 static bool
g_disk_limit(struct disk * dp,struct bio * bp)392 g_disk_limit(struct disk *dp, struct bio *bp)
393 {
394 bool limited = false;
395 off_t maxsz;
396
397 maxsz = g_disk_maxsize(dp, bp);
398
399 /*
400 * XXX: If we have a stripesize we should really use it here.
401 * Care should be taken in the delete case if this is done
402 * as deletes can be very sensitive to size given how they
403 * are processed.
404 */
405 if (bp->bio_length > maxsz) {
406 bp->bio_length = maxsz;
407 limited = true;
408 }
409
410 if ((bp->bio_flags & BIO_VLIST) != 0) {
411 bus_dma_segment_t *firstseg, *endseg;
412 off_t residual;
413
414 firstseg = (bus_dma_segment_t*)bp->bio_data;
415 residual = g_disk_vlist_limit(dp, bp, &endseg);
416 if (residual != 0) {
417 bp->bio_ma_n = endseg - firstseg;
418 bp->bio_length -= residual;
419 limited = true;
420 }
421 } else if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
422 bp->bio_ma_n =
423 howmany(bp->bio_ma_offset + bp->bio_length, PAGE_SIZE);
424 }
425
426 return (limited);
427 }
428
429 static void
g_disk_start(struct bio * bp)430 g_disk_start(struct bio *bp)
431 {
432 struct bio *bp2, *bp3;
433 struct disk *dp;
434 struct g_disk_softc *sc;
435 int error;
436 off_t off;
437
438 biotrack(bp, __func__);
439
440 sc = bp->bio_to->private;
441 dp = sc->dp;
442 KASSERT(dp != NULL && !dp->d_destroyed,
443 ("g_disk_start(%p) on destroyed disk %s", bp, bp->bio_to->name));
444 error = EJUSTRETURN;
445 switch(bp->bio_cmd) {
446 case BIO_DELETE:
447 if (!(dp->d_flags & DISKFLAG_CANDELETE)) {
448 error = EOPNOTSUPP;
449 break;
450 }
451 /* fall-through */
452 case BIO_READ:
453 case BIO_WRITE:
454 KASSERT((dp->d_flags & DISKFLAG_UNMAPPED_BIO) != 0 ||
455 (bp->bio_flags & BIO_UNMAPPED) == 0,
456 ("unmapped bio not supported by disk %s", dp->d_name));
457
458 if (bp->bio_cmd == BIO_WRITE)
459 kmsan_check_bio(bp, "g_disk_start");
460
461 off = 0;
462 bp3 = NULL;
463 bp2 = g_clone_bio(bp);
464 if (bp2 == NULL) {
465 error = ENOMEM;
466 break;
467 }
468 for (;;) {
469 if (g_disk_limit(dp, bp2)) {
470 off += bp2->bio_length;
471
472 /*
473 * To avoid a race, we need to grab the next bio
474 * before we schedule this one. See "notes".
475 */
476 bp3 = g_clone_bio(bp);
477 if (bp3 == NULL)
478 bp->bio_error = ENOMEM;
479 }
480 bp2->bio_done = g_disk_done;
481 bp2->bio_caller1 = sc;
482 bp2->bio_pblkno = bp2->bio_offset / dp->d_sectorsize;
483 bp2->bio_bcount = bp2->bio_length;
484 bp2->bio_disk = dp;
485 devstat_start_transaction_bio(dp->d_devstat, bp2);
486 dp->d_strategy(bp2);
487
488 if (bp3 == NULL)
489 break;
490
491 bp2 = bp3;
492 bp3 = NULL;
493 g_disk_advance(dp, bp2, off);
494 }
495 break;
496 case BIO_GETATTR:
497 /* Give the driver a chance to override */
498 if (dp->d_getattr != NULL) {
499 if (bp->bio_disk == NULL)
500 bp->bio_disk = dp;
501 error = dp->d_getattr(bp);
502 if (error != -1)
503 break;
504 error = EJUSTRETURN;
505 }
506 if (g_handleattr_int(bp, "GEOM::candelete",
507 (dp->d_flags & DISKFLAG_CANDELETE) != 0))
508 break;
509 else if (g_handleattr_int(bp, "GEOM::fwsectors",
510 dp->d_fwsectors))
511 break;
512 else if (g_handleattr_int(bp, "GEOM::fwheads", dp->d_fwheads))
513 break;
514 else if (g_handleattr_str(bp, "GEOM::ident", dp->d_ident))
515 break;
516 else if (g_handleattr_str(bp, "GEOM::descr", dp->d_descr))
517 break;
518 else if (g_handleattr_uint16_t(bp, "GEOM::hba_vendor",
519 dp->d_hba_vendor))
520 break;
521 else if (g_handleattr_uint16_t(bp, "GEOM::hba_device",
522 dp->d_hba_device))
523 break;
524 else if (g_handleattr_uint16_t(bp, "GEOM::hba_subvendor",
525 dp->d_hba_subvendor))
526 break;
527 else if (g_handleattr_uint16_t(bp, "GEOM::hba_subdevice",
528 dp->d_hba_subdevice))
529 break;
530 else if (!strcmp(bp->bio_attribute, "GEOM::kerneldump"))
531 g_disk_kerneldump(bp, dp);
532 else if (!strcmp(bp->bio_attribute, "GEOM::setstate"))
533 g_disk_setstate(bp, sc);
534 else if (g_handleattr_uint16_t(bp, "GEOM::rotation_rate",
535 dp->d_rotation_rate))
536 break;
537 else if (g_handleattr_str(bp, "GEOM::attachment",
538 dp->d_attachment))
539 break;
540 else
541 error = ENOIOCTL;
542 break;
543 case BIO_FLUSH:
544 g_trace(G_T_BIO, "g_disk_flushcache(%s)",
545 bp->bio_to->name);
546 if (!(dp->d_flags & DISKFLAG_CANFLUSHCACHE)) {
547 error = (sc->flush_notsup_succeed) ? 0 : EOPNOTSUPP;
548 break;
549 }
550 /*FALLTHROUGH*/
551 case BIO_ZONE:
552 if (bp->bio_cmd == BIO_ZONE) {
553 if (!(dp->d_flags & DISKFLAG_CANZONE)) {
554 error = EOPNOTSUPP;
555 break;
556 }
557 g_trace(G_T_BIO, "g_disk_zone(%s)",
558 bp->bio_to->name);
559 }
560 bp2 = g_clone_bio(bp);
561 if (bp2 == NULL) {
562 g_io_deliver(bp, ENOMEM);
563 return;
564 }
565 bp2->bio_done = g_disk_done;
566 bp2->bio_caller1 = sc;
567 bp2->bio_disk = dp;
568 devstat_start_transaction_bio(dp->d_devstat, bp2);
569 dp->d_strategy(bp2);
570 break;
571 case BIO_SPEEDUP:
572 bp2 = g_clone_bio(bp);
573 if (bp2 == NULL) {
574 g_io_deliver(bp, ENOMEM);
575 return;
576 }
577 bp2->bio_done = g_disk_done;
578 bp2->bio_caller1 = sc;
579 bp2->bio_disk = dp;
580 dp->d_strategy(bp2);
581 break;
582 default:
583 error = EOPNOTSUPP;
584 break;
585 }
586 if (error != EJUSTRETURN)
587 g_io_deliver(bp, error);
588 return;
589 }
590
591 static void
g_disk_dumpconf(struct sbuf * sb,const char * indent,struct g_geom * gp,struct g_consumer * cp,struct g_provider * pp)592 g_disk_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp)
593 {
594 struct bio *bp;
595 struct disk *dp;
596 struct g_disk_softc *sc;
597 char *buf;
598 int res = 0;
599
600 sc = gp->softc;
601 if (sc == NULL || (dp = sc->dp) == NULL)
602 return;
603 if (indent == NULL) {
604 sbuf_printf(sb, " hd %u", dp->d_fwheads);
605 sbuf_printf(sb, " sc %u", dp->d_fwsectors);
606 return;
607 }
608 if (pp != NULL) {
609 sbuf_printf(sb, "%s<fwheads>%u</fwheads>\n",
610 indent, dp->d_fwheads);
611 sbuf_printf(sb, "%s<fwsectors>%u</fwsectors>\n",
612 indent, dp->d_fwsectors);
613
614 /*
615 * "rotationrate" is a little complicated, because the value
616 * returned by the drive might not be the RPM; 0 and 1 are
617 * special cases, and there's also a valid range.
618 */
619 sbuf_printf(sb, "%s<rotationrate>", indent);
620 if (dp->d_rotation_rate == DISK_RR_UNKNOWN) /* Old drives */
621 sbuf_cat(sb, "unknown"); /* don't report RPM. */
622 else if (dp->d_rotation_rate == DISK_RR_NON_ROTATING)
623 sbuf_cat(sb, "0");
624 else if ((dp->d_rotation_rate >= DISK_RR_MIN) &&
625 (dp->d_rotation_rate <= DISK_RR_MAX))
626 sbuf_printf(sb, "%u", dp->d_rotation_rate);
627 else
628 sbuf_cat(sb, "invalid");
629 sbuf_cat(sb, "</rotationrate>\n");
630 if (dp->d_getattr != NULL) {
631 buf = g_malloc(DISK_IDENT_SIZE, M_WAITOK);
632 bp = g_alloc_bio();
633 bp->bio_disk = dp;
634 bp->bio_attribute = "GEOM::ident";
635 bp->bio_length = DISK_IDENT_SIZE;
636 bp->bio_data = buf;
637 res = dp->d_getattr(bp);
638 sbuf_printf(sb, "%s<ident>", indent);
639 g_conf_cat_escaped(sb, res == 0 ? buf : dp->d_ident);
640 sbuf_cat(sb, "</ident>\n");
641 bp->bio_attribute = "GEOM::lunid";
642 bp->bio_length = DISK_IDENT_SIZE;
643 bp->bio_data = buf;
644 if (dp->d_getattr(bp) == 0) {
645 sbuf_printf(sb, "%s<lunid>", indent);
646 g_conf_cat_escaped(sb, buf);
647 sbuf_cat(sb, "</lunid>\n");
648 }
649 bp->bio_attribute = "GEOM::lunname";
650 bp->bio_length = DISK_IDENT_SIZE;
651 bp->bio_data = buf;
652 if (dp->d_getattr(bp) == 0) {
653 sbuf_printf(sb, "%s<lunname>", indent);
654 g_conf_cat_escaped(sb, buf);
655 sbuf_cat(sb, "</lunname>\n");
656 }
657 g_destroy_bio(bp);
658 g_free(buf);
659 } else {
660 sbuf_printf(sb, "%s<ident>", indent);
661 g_conf_cat_escaped(sb, dp->d_ident);
662 sbuf_cat(sb, "</ident>\n");
663 }
664 sbuf_printf(sb, "%s<descr>", indent);
665 g_conf_cat_escaped(sb, dp->d_descr);
666 sbuf_cat(sb, "</descr>\n");
667 }
668 }
669
670 static void
g_disk_resize(void * ptr,int flag)671 g_disk_resize(void *ptr, int flag)
672 {
673 struct disk *dp;
674 struct g_geom *gp;
675 struct g_provider *pp;
676
677 if (flag == EV_CANCEL)
678 return;
679 g_topology_assert();
680
681 dp = ptr;
682 gp = dp->d_geom;
683
684 if (dp->d_destroyed || gp == NULL)
685 return;
686
687 LIST_FOREACH(pp, &gp->provider, provider) {
688 if (pp->sectorsize != 0 &&
689 pp->sectorsize != dp->d_sectorsize)
690 g_wither_provider(pp, ENXIO);
691 else
692 g_resize_provider(pp, dp->d_mediasize);
693 }
694 }
695
696 static void
g_disk_create(void * arg,int flag)697 g_disk_create(void *arg, int flag)
698 {
699 struct g_geom *gp;
700 struct g_provider *pp;
701 struct disk *dp;
702 struct g_disk_softc *sc;
703 struct disk_alias *dap;
704 char tmpstr[80];
705
706 if (flag == EV_CANCEL)
707 return;
708 g_topology_assert();
709 dp = arg;
710
711 mtx_pool_lock(mtxpool_sleep, dp);
712 dp->d_init_level = DISK_INIT_START;
713
714 /*
715 * If the disk has already gone away, we can just stop here and
716 * call the user's callback to tell him we've cleaned things up.
717 */
718 if (dp->d_goneflag != 0) {
719 mtx_pool_unlock(mtxpool_sleep, dp);
720 if (dp->d_gone != NULL)
721 dp->d_gone(dp);
722 return;
723 }
724 mtx_pool_unlock(mtxpool_sleep, dp);
725
726 sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO);
727 mtx_init(&sc->done_mtx, "g_disk_done", NULL, MTX_DEF);
728 sc->dp = dp;
729 if (dp->d_devstat == NULL) {
730 dp->d_devstat = devstat_new_entry(dp->d_name, dp->d_unit,
731 dp->d_sectorsize, DEVSTAT_ALL_SUPPORTED,
732 DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX);
733 }
734 sc->d_devstat = dp->d_devstat;
735 gp = g_new_geomf(&g_disk_class, "%s%d", dp->d_name, dp->d_unit);
736 gp->softc = sc;
737 pp = g_new_providerf(gp, "%s", gp->name);
738 LIST_FOREACH(dap, &dp->d_aliases, da_next)
739 g_provider_add_alias(pp, "%s%d", dap->da_alias, dp->d_unit);
740 devstat_remove_entry(pp->stat);
741 pp->stat = NULL;
742 dp->d_devstat->id = pp;
743 pp->mediasize = dp->d_mediasize;
744 pp->sectorsize = dp->d_sectorsize;
745 pp->stripeoffset = dp->d_stripeoffset;
746 pp->stripesize = dp->d_stripesize;
747 if ((dp->d_flags & DISKFLAG_UNMAPPED_BIO) != 0)
748 pp->flags |= G_PF_ACCEPT_UNMAPPED;
749 if ((dp->d_flags & DISKFLAG_DIRECT_COMPLETION) != 0)
750 pp->flags |= G_PF_DIRECT_SEND;
751 pp->flags |= G_PF_DIRECT_RECEIVE;
752 if (bootverbose)
753 printf("GEOM: new disk %s\n", gp->name);
754 sysctl_ctx_init(&sc->sysctl_ctx);
755 snprintf(tmpstr, sizeof(tmpstr), "GEOM disk %s", gp->name);
756 sc->sysctl_tree = SYSCTL_ADD_NODE(&sc->sysctl_ctx,
757 SYSCTL_STATIC_CHILDREN(_kern_geom_disk), OID_AUTO, gp->name,
758 CTLFLAG_RD | CTLFLAG_MPSAFE, 0, tmpstr);
759 if (sc->sysctl_tree != NULL) {
760 SYSCTL_ADD_STRING(&sc->sysctl_ctx,
761 SYSCTL_CHILDREN(sc->sysctl_tree), OID_AUTO, "led",
762 CTLFLAG_RWTUN, sc->led, sizeof(sc->led),
763 "LED name");
764 SYSCTL_ADD_PROC(&sc->sysctl_ctx,
765 SYSCTL_CHILDREN(sc->sysctl_tree), OID_AUTO, "flags",
766 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, dp, 0,
767 g_disk_sysctl_flags, "A", "Report disk flags");
768 SYSCTL_ADD_BOOL(&sc->sysctl_ctx,
769 SYSCTL_CHILDREN(sc->sysctl_tree), OID_AUTO, "flush_notsup_succeed",
770 CTLFLAG_RWTUN, &sc->flush_notsup_succeed, sizeof(sc->flush_notsup_succeed),
771 "Do not return EOPNOTSUPP if there is no cache to flush");
772 }
773 pp->private = sc;
774 dp->d_geom = gp;
775 g_error_provider(pp, 0);
776
777 mtx_pool_lock(mtxpool_sleep, dp);
778 dp->d_init_level = DISK_INIT_DONE;
779
780 /*
781 * If the disk has gone away at this stage, start the withering
782 * process for it.
783 */
784 if (dp->d_goneflag != 0) {
785 mtx_pool_unlock(mtxpool_sleep, dp);
786 g_wither_provider(pp, ENXIO);
787 return;
788 }
789 mtx_pool_unlock(mtxpool_sleep, dp);
790
791 }
792
793 /*
794 * We get this callback after all of the consumers have gone away, and just
795 * before the provider is freed. If the disk driver provided a d_gone
796 * callback, let them know that it is okay to free resources -- they won't
797 * be getting any more accesses from GEOM.
798 */
799 static void
g_disk_providergone(struct g_provider * pp)800 g_disk_providergone(struct g_provider *pp)
801 {
802 struct disk *dp;
803 struct g_disk_softc *sc;
804
805 sc = (struct g_disk_softc *)pp->private;
806 dp = sc->dp;
807 if (dp != NULL && dp->d_gone != NULL)
808 dp->d_gone(dp);
809 if (sc->sysctl_tree != NULL) {
810 sysctl_ctx_free(&sc->sysctl_ctx);
811 sc->sysctl_tree = NULL;
812 }
813 if (sc->led[0] != 0) {
814 led_set(sc->led, "0");
815 sc->led[0] = 0;
816 }
817 pp->private = NULL;
818 pp->geom->softc = NULL;
819 mtx_destroy(&sc->done_mtx);
820 g_free(sc);
821 }
822
823 static void
g_disk_destroy(void * ptr,int flag)824 g_disk_destroy(void *ptr, int flag)
825 {
826 struct disk *dp;
827 struct g_geom *gp;
828 struct g_disk_softc *sc;
829 struct disk_alias *dap, *daptmp;
830
831 g_topology_assert();
832 dp = ptr;
833 gp = dp->d_geom;
834 if (gp != NULL) {
835 sc = gp->softc;
836 if (sc != NULL)
837 sc->dp = NULL;
838 dp->d_geom = NULL;
839 g_wither_geom(gp, ENXIO);
840 }
841 LIST_FOREACH_SAFE(dap, &dp->d_aliases, da_next, daptmp)
842 g_free(dap);
843
844 g_free(dp);
845 }
846
847 /*
848 * We only allow printable characters in disk ident,
849 * the rest is converted to 'x<HH>'.
850 */
851 static void
g_disk_ident_adjust(char * ident,size_t size)852 g_disk_ident_adjust(char *ident, size_t size)
853 {
854 char *p, tmp[4], newid[DISK_IDENT_SIZE];
855
856 newid[0] = '\0';
857 for (p = ident; *p != '\0'; p++) {
858 if (isprint(*p)) {
859 tmp[0] = *p;
860 tmp[1] = '\0';
861 } else {
862 snprintf(tmp, sizeof(tmp), "x%02hhx",
863 *(unsigned char *)p);
864 }
865 if (strlcat(newid, tmp, sizeof(newid)) >= sizeof(newid))
866 break;
867 }
868 bzero(ident, size);
869 strlcpy(ident, newid, size);
870 }
871
872 struct disk *
disk_alloc(void)873 disk_alloc(void)
874 {
875 struct disk *dp;
876
877 dp = g_malloc(sizeof(struct disk), M_WAITOK | M_ZERO);
878 LIST_INIT(&dp->d_aliases);
879 dp->d_init_level = DISK_INIT_NONE;
880 dp->d_cevent = g_alloc_event(M_WAITOK);
881 dp->d_devent = g_alloc_event(M_WAITOK);
882 return (dp);
883 }
884
885 void
disk_create(struct disk * dp,int version)886 disk_create(struct disk *dp, int version)
887 {
888
889 if (version != DISK_VERSION) {
890 printf("WARNING: Attempt to add disk %s%d %s",
891 dp->d_name, dp->d_unit,
892 " using incompatible ABI version of disk(9)\n");
893 printf("WARNING: Ignoring disk %s%d\n",
894 dp->d_name, dp->d_unit);
895 return;
896 }
897 if (dp->d_flags & DISKFLAG_RESERVED) {
898 printf("WARNING: Attempt to add non-MPSAFE disk %s%d\n",
899 dp->d_name, dp->d_unit);
900 printf("WARNING: Ignoring disk %s%d\n",
901 dp->d_name, dp->d_unit);
902 return;
903 }
904 KASSERT(dp->d_strategy != NULL, ("disk_create need d_strategy"));
905 KASSERT(dp->d_name != NULL, ("disk_create need d_name"));
906 KASSERT(*dp->d_name != 0, ("disk_create need d_name"));
907 KASSERT(strlen(dp->d_name) < SPECNAMELEN - 4, ("disk name too long"));
908 g_disk_ident_adjust(dp->d_ident, sizeof(dp->d_ident));
909
910 dp->d_init_level = DISK_INIT_CREATE;
911
912 KASSERT(dp->d_cevent != NULL,
913 ("Disk create for %p with event NULL", dp));
914 g_post_event_ep(g_disk_create, dp, dp->d_cevent, dp, NULL);
915 }
916
917 void
disk_destroy(struct disk * dp)918 disk_destroy(struct disk *dp)
919 {
920 struct disk_alias *dap, *daptmp;
921
922 /* If disk_create() was never called, just free the resources. */
923 if (dp->d_init_level < DISK_INIT_CREATE) {
924 if (dp->d_devstat != NULL)
925 devstat_remove_entry(dp->d_devstat);
926 LIST_FOREACH_SAFE(dap, &dp->d_aliases, da_next, daptmp)
927 g_free(dap);
928 g_free(dp->d_cevent);
929 g_free(dp->d_devent);
930 g_free(dp);
931 return;
932 }
933
934 KASSERT(dp->d_devent != NULL,
935 ("Disk destroy for %p with event NULL", dp));
936 disk_gone(dp);
937 dp->d_destroyed = 1;
938 g_cancel_event(dp);
939 if (dp->d_devstat != NULL)
940 devstat_remove_entry(dp->d_devstat);
941 g_post_event_ep(g_disk_destroy, dp, dp->d_devent, NULL);
942 }
943
944 void
disk_add_alias(struct disk * dp,const char * name)945 disk_add_alias(struct disk *dp, const char *name)
946 {
947 struct disk_alias *dap;
948
949 dap = (struct disk_alias *)g_malloc(
950 sizeof(struct disk_alias) + strlen(name) + 1, M_WAITOK);
951 strcpy((char *)(dap + 1), name);
952 dap->da_alias = (const char *)(dap + 1);
953 LIST_INSERT_HEAD(&dp->d_aliases, dap, da_next);
954 }
955
956 void
disk_gone(struct disk * dp)957 disk_gone(struct disk *dp)
958 {
959 struct g_geom *gp;
960 struct g_provider *pp;
961
962 mtx_pool_lock(mtxpool_sleep, dp);
963
964 /*
965 * Second wither call makes no sense, plus we can not access the list
966 * of providers without topology lock after calling wither once.
967 */
968 if (dp->d_goneflag != 0) {
969 mtx_pool_unlock(mtxpool_sleep, dp);
970 return;
971 }
972
973 dp->d_goneflag = 1;
974
975 /*
976 * If we're still in the process of creating this disk (the
977 * g_disk_create() function is still queued, or is in
978 * progress), the init level will not yet be DISK_INIT_DONE.
979 *
980 * If that is the case, g_disk_create() will see d_goneflag
981 * and take care of cleaning things up.
982 *
983 * If the disk has already been created, we default to
984 * withering the provider as usual below.
985 *
986 * If the caller has not set a d_gone() callback, he will
987 * not be any worse off by returning here, because the geom
988 * has not been fully setup in any case.
989 */
990 if (dp->d_init_level < DISK_INIT_DONE) {
991 mtx_pool_unlock(mtxpool_sleep, dp);
992 return;
993 }
994 mtx_pool_unlock(mtxpool_sleep, dp);
995
996 gp = dp->d_geom;
997 pp = LIST_FIRST(&gp->provider);
998 if (pp != NULL) {
999 KASSERT(LIST_NEXT(pp, provider) == NULL,
1000 ("geom %p has more than one provider", gp));
1001 g_wither_provider(pp, ENXIO);
1002 }
1003 }
1004
1005 void
disk_attr_changed(struct disk * dp,const char * attr,int flag)1006 disk_attr_changed(struct disk *dp, const char *attr, int flag)
1007 {
1008 struct g_geom *gp = dp->d_geom;
1009 struct g_provider *pp;
1010 char devnamebuf[128];
1011
1012 if (gp == NULL)
1013 return;
1014 LIST_FOREACH(pp, &gp->provider, provider)
1015 (void)g_attr_changed(pp, attr, flag);
1016 snprintf(devnamebuf, sizeof(devnamebuf), "devname=%s%d", dp->d_name,
1017 dp->d_unit);
1018 devctl_notify("GEOM", "disk", attr, devnamebuf);
1019 }
1020
1021 void
disk_media_changed(struct disk * dp,int flag)1022 disk_media_changed(struct disk *dp, int flag)
1023 {
1024 struct g_geom *gp = dp->d_geom;
1025 struct g_provider *pp;
1026
1027 if (gp == NULL)
1028 return;
1029 pp = LIST_FIRST(&gp->provider);
1030 if (pp != NULL) {
1031 KASSERT(LIST_NEXT(pp, provider) == NULL,
1032 ("geom %p has more than one provider", gp));
1033 g_media_changed(pp, flag);
1034 }
1035 }
1036
1037 void
disk_media_gone(struct disk * dp,int flag)1038 disk_media_gone(struct disk *dp, int flag)
1039 {
1040 struct g_geom *gp = dp->d_geom;
1041 struct g_provider *pp;
1042
1043 if (gp == NULL)
1044 return;
1045 pp = LIST_FIRST(&gp->provider);
1046 if (pp != NULL) {
1047 KASSERT(LIST_NEXT(pp, provider) == NULL,
1048 ("geom %p has more than one provider", gp));
1049 g_media_gone(pp, flag);
1050 }
1051 }
1052
1053 int
disk_resize(struct disk * dp,int flag)1054 disk_resize(struct disk *dp, int flag)
1055 {
1056
1057 if (dp->d_destroyed || dp->d_geom == NULL)
1058 return (0);
1059
1060 return (g_post_event(g_disk_resize, dp, flag, NULL));
1061 }
1062
1063 static void
g_kern_disks(void * p,int flag __unused)1064 g_kern_disks(void *p, int flag __unused)
1065 {
1066 struct sbuf *sb;
1067 struct g_geom *gp;
1068 char *sp;
1069
1070 sb = p;
1071 sp = "";
1072 g_topology_assert();
1073 LIST_FOREACH(gp, &g_disk_class.geom, geom) {
1074 sbuf_printf(sb, "%s%s", sp, gp->name);
1075 sp = " ";
1076 }
1077 sbuf_finish(sb);
1078 }
1079
1080 static int
g_disk_sysctl_flags(SYSCTL_HANDLER_ARGS)1081 g_disk_sysctl_flags(SYSCTL_HANDLER_ARGS)
1082 {
1083 struct disk *dp;
1084 struct sbuf *sb;
1085 int error;
1086
1087 sb = sbuf_new_auto();
1088 dp = (struct disk *)arg1;
1089 sbuf_printf(sb, "%b", dp->d_flags,
1090 "\20"
1091 "\2OPEN"
1092 "\3CANDELETE"
1093 "\4CANFLUSHCACHE"
1094 "\5UNMAPPEDBIO"
1095 "\6DIRECTCOMPLETION"
1096 "\10CANZONE"
1097 "\11WRITEPROTECT");
1098
1099 sbuf_finish(sb);
1100 error = SYSCTL_OUT(req, sbuf_data(sb), sbuf_len(sb) + 1);
1101 sbuf_delete(sb);
1102 return (error);
1103 }
1104
1105 static int
sysctl_disks(SYSCTL_HANDLER_ARGS)1106 sysctl_disks(SYSCTL_HANDLER_ARGS)
1107 {
1108 int error;
1109 struct sbuf *sb;
1110
1111 sb = sbuf_new_auto();
1112 g_waitfor_event(g_kern_disks, sb, M_WAITOK, NULL);
1113 error = SYSCTL_OUT(req, sbuf_data(sb), sbuf_len(sb) + 1);
1114 sbuf_delete(sb);
1115 return error;
1116 }
1117
1118 SYSCTL_PROC(_kern, OID_AUTO, disks,
1119 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
1120 sysctl_disks, "A", "names of available disks");
1121