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