FreeBSD/Linux Kernel Cross Reference
sys/geom/geom_disk.c
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 __FBSDID("$FreeBSD$");
40
41 #include "opt_geom.h"
42
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/kernel.h>
46 #include <sys/sysctl.h>
47 #include <sys/bio.h>
48 #include <sys/bus.h>
49 #include <sys/ctype.h>
50 #include <sys/fcntl.h>
51 #include <sys/malloc.h>
52 #include <sys/sbuf.h>
53 #include <sys/devicestat.h>
54 #include <machine/md_var.h>
55
56 #include <sys/lock.h>
57 #include <sys/mutex.h>
58 #include <geom/geom.h>
59 #include <geom/geom_disk.h>
60 #include <geom/geom_int.h>
61
62 #include <dev/led/led.h>
63
64 #include <machine/bus.h>
65
66 struct g_disk_softc {
67 struct mtx done_mtx;
68 struct disk *dp;
69 struct devstat *d_devstat;
70 struct sysctl_ctx_list sysctl_ctx;
71 struct sysctl_oid *sysctl_tree;
72 char led[64];
73 uint32_t state;
74 struct mtx start_mtx;
75 };
76
77 static g_access_t g_disk_access;
78 static g_start_t g_disk_start;
79 static g_ioctl_t g_disk_ioctl;
80 static g_dumpconf_t g_disk_dumpconf;
81 static g_provgone_t g_disk_providergone;
82
83 static int g_disk_sysctl_flags(SYSCTL_HANDLER_ARGS);
84
85 static struct g_class g_disk_class = {
86 .name = G_DISK_CLASS_NAME,
87 .version = G_VERSION,
88 .start = g_disk_start,
89 .access = g_disk_access,
90 .ioctl = g_disk_ioctl,
91 .providergone = g_disk_providergone,
92 .dumpconf = g_disk_dumpconf,
93 };
94
95 SYSCTL_DECL(_kern_geom);
96 static SYSCTL_NODE(_kern_geom, OID_AUTO, disk, CTLFLAG_RW, 0,
97 "GEOM_DISK stuff");
98
99 DECLARE_GEOM_CLASS(g_disk_class, g_disk);
100
101 static int
102 g_disk_access(struct g_provider *pp, int r, int w, int e)
103 {
104 struct disk *dp;
105 struct g_disk_softc *sc;
106 int error;
107
108 g_trace(G_T_ACCESS, "g_disk_access(%s, %d, %d, %d)",
109 pp->name, r, w, e);
110 g_topology_assert();
111 sc = pp->private;
112 if ((dp = sc->dp) == NULL || dp->d_destroyed) {
113 /*
114 * Allow decreasing access count even if disk is not
115 * available anymore.
116 */
117 if (r <= 0 && w <= 0 && e <= 0)
118 return (0);
119 return (ENXIO);
120 }
121 r += pp->acr;
122 w += pp->acw;
123 e += pp->ace;
124 error = 0;
125 if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) {
126 /*
127 * It would be better to defer this decision to d_open if
128 * it was able to take flags.
129 */
130 if (w > 0 && (dp->d_flags & DISKFLAG_WRITE_PROTECT) != 0)
131 error = EROFS;
132 if (error == 0 && dp->d_open != NULL)
133 error = dp->d_open(dp);
134 if (bootverbose && error != 0)
135 printf("Opened disk %s -> %d\n", pp->name, error);
136 if (error != 0)
137 return (error);
138 pp->sectorsize = dp->d_sectorsize;
139 if (dp->d_maxsize == 0) {
140 printf("WARNING: Disk drive %s%d has no d_maxsize\n",
141 dp->d_name, dp->d_unit);
142 dp->d_maxsize = DFLTPHYS;
143 }
144 if (dp->d_delmaxsize == 0) {
145 if (bootverbose && dp->d_flags & DISKFLAG_CANDELETE) {
146 printf("WARNING: Disk drive %s%d has no "
147 "d_delmaxsize\n", dp->d_name, dp->d_unit);
148 }
149 dp->d_delmaxsize = dp->d_maxsize;
150 }
151 pp->stripeoffset = dp->d_stripeoffset;
152 pp->stripesize = dp->d_stripesize;
153 dp->d_flags |= DISKFLAG_OPEN;
154 /*
155 * Do not invoke resize event when initial size was zero.
156 * Some disks report its size only after first opening.
157 */
158 if (pp->mediasize == 0)
159 pp->mediasize = dp->d_mediasize;
160 else
161 g_resize_provider(pp, dp->d_mediasize);
162 } else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) {
163 if (dp->d_close != NULL) {
164 error = dp->d_close(dp);
165 if (error != 0)
166 printf("Closed disk %s -> %d\n",
167 pp->name, error);
168 }
169 sc->state = G_STATE_ACTIVE;
170 if (sc->led[0] != 0)
171 led_set(sc->led, "");
172 dp->d_flags &= ~DISKFLAG_OPEN;
173 }
174 return (error);
175 }
176
177 static void
178 g_disk_kerneldump(struct bio *bp, struct disk *dp)
179 {
180 struct g_kerneldump *gkd;
181 struct g_geom *gp;
182
183 gkd = (struct g_kerneldump*)bp->bio_data;
184 gp = bp->bio_to->geom;
185 g_trace(G_T_TOPOLOGY, "g_disk_kerneldump(%s, %jd, %jd)",
186 gp->name, (intmax_t)gkd->offset, (intmax_t)gkd->length);
187 if (dp->d_dump == NULL) {
188 g_io_deliver(bp, ENODEV);
189 return;
190 }
191 gkd->di.dumper = dp->d_dump;
192 gkd->di.priv = dp;
193 gkd->di.blocksize = dp->d_sectorsize;
194 gkd->di.maxiosize = dp->d_maxsize;
195 gkd->di.mediaoffset = gkd->offset;
196 if ((gkd->offset + gkd->length) > dp->d_mediasize)
197 gkd->length = dp->d_mediasize - gkd->offset;
198 gkd->di.mediasize = gkd->length;
199 g_io_deliver(bp, 0);
200 }
201
202 static void
203 g_disk_setstate(struct bio *bp, struct g_disk_softc *sc)
204 {
205 const char *cmd;
206
207 memcpy(&sc->state, bp->bio_data, sizeof(sc->state));
208 if (sc->led[0] != 0) {
209 switch (sc->state) {
210 case G_STATE_FAILED:
211 cmd = "1";
212 break;
213 case G_STATE_REBUILD:
214 cmd = "f5";
215 break;
216 case G_STATE_RESYNC:
217 cmd = "f1";
218 break;
219 default:
220 cmd = "";
221 break;
222 }
223 led_set(sc->led, cmd);
224 }
225 g_io_deliver(bp, 0);
226 }
227
228 static void
229 g_disk_done(struct bio *bp)
230 {
231 struct bintime now;
232 struct bio *bp2;
233 struct g_disk_softc *sc;
234
235 /* See "notes" for why we need a mutex here */
236 sc = bp->bio_caller1;
237 bp2 = bp->bio_parent;
238 binuptime(&now);
239 mtx_lock(&sc->done_mtx);
240 if (bp2->bio_error == 0)
241 bp2->bio_error = bp->bio_error;
242 bp2->bio_completed += bp->bio_length - bp->bio_resid;
243
244 switch (bp->bio_cmd) {
245 case BIO_ZONE:
246 bcopy(&bp->bio_zone, &bp2->bio_zone, sizeof(bp->bio_zone));
247 /*FALLTHROUGH*/
248 case BIO_READ:
249 case BIO_WRITE:
250 case BIO_DELETE:
251 case BIO_FLUSH:
252 devstat_end_transaction_bio_bt(sc->d_devstat, bp, &now);
253 break;
254 default:
255 break;
256 }
257 bp2->bio_inbed++;
258 if (bp2->bio_children == bp2->bio_inbed) {
259 mtx_unlock(&sc->done_mtx);
260 bp2->bio_resid = bp2->bio_bcount - bp2->bio_completed;
261 g_io_deliver(bp2, bp2->bio_error);
262 } else
263 mtx_unlock(&sc->done_mtx);
264 g_destroy_bio(bp);
265 }
266
267 static int
268 g_disk_ioctl(struct g_provider *pp, u_long cmd, void * data, int fflag, struct thread *td)
269 {
270 struct disk *dp;
271 struct g_disk_softc *sc;
272 int error;
273
274 sc = pp->private;
275 dp = sc->dp;
276 KASSERT(dp != NULL && !dp->d_destroyed,
277 ("g_disk_ioctl(%lx) on destroyed disk %s", cmd, pp->name));
278
279 if (dp->d_ioctl == NULL)
280 return (ENOIOCTL);
281 error = dp->d_ioctl(dp, cmd, data, fflag, td);
282 return (error);
283 }
284
285 static off_t
286 g_disk_maxsize(struct disk *dp, struct bio *bp)
287 {
288 if (bp->bio_cmd == BIO_DELETE)
289 return (dp->d_delmaxsize);
290 return (dp->d_maxsize);
291 }
292
293 static int
294 g_disk_maxsegs(struct disk *dp, struct bio *bp)
295 {
296 return ((g_disk_maxsize(dp, bp) / PAGE_SIZE) + 1);
297 }
298
299 static void
300 g_disk_advance(struct disk *dp, struct bio *bp, off_t off)
301 {
302
303 bp->bio_offset += off;
304 bp->bio_length -= off;
305
306 if ((bp->bio_flags & BIO_VLIST) != 0) {
307 bus_dma_segment_t *seg, *end;
308
309 seg = (bus_dma_segment_t *)bp->bio_data;
310 end = (bus_dma_segment_t *)bp->bio_data + bp->bio_ma_n;
311 off += bp->bio_ma_offset;
312 while (off >= seg->ds_len) {
313 KASSERT((seg != end),
314 ("vlist request runs off the end"));
315 off -= seg->ds_len;
316 seg++;
317 }
318 bp->bio_ma_offset = off;
319 bp->bio_ma_n = end - seg;
320 bp->bio_data = (void *)seg;
321 } else if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
322 bp->bio_ma += off / PAGE_SIZE;
323 bp->bio_ma_offset += off;
324 bp->bio_ma_offset %= PAGE_SIZE;
325 bp->bio_ma_n -= off / PAGE_SIZE;
326 } else {
327 bp->bio_data += off;
328 }
329 }
330
331 static void
332 g_disk_seg_limit(bus_dma_segment_t *seg, off_t *poffset,
333 off_t *plength, int *ppages)
334 {
335 uintptr_t seg_page_base;
336 uintptr_t seg_page_end;
337 off_t offset;
338 off_t length;
339 int seg_pages;
340
341 offset = *poffset;
342 length = *plength;
343
344 if (length > seg->ds_len - offset)
345 length = seg->ds_len - offset;
346
347 seg_page_base = trunc_page(seg->ds_addr + offset);
348 seg_page_end = round_page(seg->ds_addr + offset + length);
349 seg_pages = (seg_page_end - seg_page_base) >> PAGE_SHIFT;
350
351 if (seg_pages > *ppages) {
352 seg_pages = *ppages;
353 length = (seg_page_base + (seg_pages << PAGE_SHIFT)) -
354 (seg->ds_addr + offset);
355 }
356
357 *poffset = 0;
358 *plength -= length;
359 *ppages -= seg_pages;
360 }
361
362 static off_t
363 g_disk_vlist_limit(struct disk *dp, struct bio *bp, bus_dma_segment_t **pendseg)
364 {
365 bus_dma_segment_t *seg, *end;
366 off_t residual;
367 off_t offset;
368 int pages;
369
370 seg = (bus_dma_segment_t *)bp->bio_data;
371 end = (bus_dma_segment_t *)bp->bio_data + bp->bio_ma_n;
372 residual = bp->bio_length;
373 offset = bp->bio_ma_offset;
374 pages = g_disk_maxsegs(dp, bp);
375 while (residual != 0 && pages != 0) {
376 KASSERT((seg != end),
377 ("vlist limit runs off the end"));
378 g_disk_seg_limit(seg, &offset, &residual, &pages);
379 seg++;
380 }
381 if (pendseg != NULL)
382 *pendseg = seg;
383 return (residual);
384 }
385
386 static bool
387 g_disk_limit(struct disk *dp, struct bio *bp)
388 {
389 bool limited = false;
390 off_t maxsz;
391
392 maxsz = g_disk_maxsize(dp, bp);
393
394 /*
395 * XXX: If we have a stripesize we should really use it here.
396 * Care should be taken in the delete case if this is done
397 * as deletes can be very sensitive to size given how they
398 * are processed.
399 */
400 if (bp->bio_length > maxsz) {
401 bp->bio_length = maxsz;
402 limited = true;
403 }
404
405 if ((bp->bio_flags & BIO_VLIST) != 0) {
406 bus_dma_segment_t *firstseg, *endseg;
407 off_t residual;
408
409 firstseg = (bus_dma_segment_t*)bp->bio_data;
410 residual = g_disk_vlist_limit(dp, bp, &endseg);
411 if (residual != 0) {
412 bp->bio_ma_n = endseg - firstseg;
413 bp->bio_length -= residual;
414 limited = true;
415 }
416 } else if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
417 bp->bio_ma_n =
418 howmany(bp->bio_ma_offset + bp->bio_length, PAGE_SIZE);
419 }
420
421 return (limited);
422 }
423
424 static void
425 g_disk_start(struct bio *bp)
426 {
427 struct bio *bp2, *bp3;
428 struct disk *dp;
429 struct g_disk_softc *sc;
430 int error;
431 off_t off;
432
433 biotrack(bp, __func__);
434
435 sc = bp->bio_to->private;
436 dp = sc->dp;
437 KASSERT(dp != NULL && !dp->d_destroyed,
438 ("g_disk_start(%p) on destroyed disk %s", bp, bp->bio_to->name));
439 error = EJUSTRETURN;
440 switch(bp->bio_cmd) {
441 case BIO_DELETE:
442 if (!(dp->d_flags & DISKFLAG_CANDELETE)) {
443 error = EOPNOTSUPP;
444 break;
445 }
446 /* fall-through */
447 case BIO_READ:
448 case BIO_WRITE:
449 KASSERT((dp->d_flags & DISKFLAG_UNMAPPED_BIO) != 0 ||
450 (bp->bio_flags & BIO_UNMAPPED) == 0,
451 ("unmapped bio not supported by disk %s", dp->d_name));
452 off = 0;
453 bp3 = NULL;
454 bp2 = g_clone_bio(bp);
455 if (bp2 == NULL) {
456 error = ENOMEM;
457 break;
458 }
459 for (;;) {
460 if (g_disk_limit(dp, bp2)) {
461 off += bp2->bio_length;
462
463 /*
464 * To avoid a race, we need to grab the next bio
465 * before we schedule this one. See "notes".
466 */
467 bp3 = g_clone_bio(bp);
468 if (bp3 == NULL)
469 bp->bio_error = ENOMEM;
470 }
471 bp2->bio_done = g_disk_done;
472 bp2->bio_caller1 = sc;
473 bp2->bio_pblkno = bp2->bio_offset / dp->d_sectorsize;
474 bp2->bio_bcount = bp2->bio_length;
475 bp2->bio_disk = dp;
476 mtx_lock(&sc->start_mtx);
477 devstat_start_transaction_bio(dp->d_devstat, bp2);
478 mtx_unlock(&sc->start_mtx);
479 dp->d_strategy(bp2);
480
481 if (bp3 == NULL)
482 break;
483
484 bp2 = bp3;
485 bp3 = NULL;
486 g_disk_advance(dp, bp2, off);
487 }
488 break;
489 case BIO_GETATTR:
490 /* Give the driver a chance to override */
491 if (dp->d_getattr != NULL) {
492 if (bp->bio_disk == NULL)
493 bp->bio_disk = dp;
494 error = dp->d_getattr(bp);
495 if (error != -1)
496 break;
497 error = EJUSTRETURN;
498 }
499 if (g_handleattr_int(bp, "GEOM::candelete",
500 (dp->d_flags & DISKFLAG_CANDELETE) != 0))
501 break;
502 else if (g_handleattr_int(bp, "GEOM::fwsectors",
503 dp->d_fwsectors))
504 break;
505 else if (g_handleattr_int(bp, "GEOM::fwheads", dp->d_fwheads))
506 break;
507 else if (g_handleattr_off_t(bp, "GEOM::frontstuff", 0))
508 break;
509 else if (g_handleattr_str(bp, "GEOM::ident", dp->d_ident))
510 break;
511 else if (g_handleattr_str(bp, "GEOM::descr", dp->d_descr))
512 break;
513 else if (g_handleattr_uint16_t(bp, "GEOM::hba_vendor",
514 dp->d_hba_vendor))
515 break;
516 else if (g_handleattr_uint16_t(bp, "GEOM::hba_device",
517 dp->d_hba_device))
518 break;
519 else if (g_handleattr_uint16_t(bp, "GEOM::hba_subvendor",
520 dp->d_hba_subvendor))
521 break;
522 else if (g_handleattr_uint16_t(bp, "GEOM::hba_subdevice",
523 dp->d_hba_subdevice))
524 break;
525 else if (!strcmp(bp->bio_attribute, "GEOM::kerneldump"))
526 g_disk_kerneldump(bp, dp);
527 else if (!strcmp(bp->bio_attribute, "GEOM::setstate"))
528 g_disk_setstate(bp, sc);
529 else if (g_handleattr_uint16_t(bp, "GEOM::rotation_rate",
530 dp->d_rotation_rate))
531 break;
532 else
533 error = ENOIOCTL;
534 break;
535 case BIO_FLUSH:
536 g_trace(G_T_BIO, "g_disk_flushcache(%s)",
537 bp->bio_to->name);
538 if (!(dp->d_flags & DISKFLAG_CANFLUSHCACHE)) {
539 error = EOPNOTSUPP;
540 break;
541 }
542 /*FALLTHROUGH*/
543 case BIO_ZONE:
544 if (bp->bio_cmd == BIO_ZONE) {
545 if (!(dp->d_flags & DISKFLAG_CANZONE)) {
546 error = EOPNOTSUPP;
547 break;
548 }
549 g_trace(G_T_BIO, "g_disk_zone(%s)",
550 bp->bio_to->name);
551 }
552 bp2 = g_clone_bio(bp);
553 if (bp2 == NULL) {
554 g_io_deliver(bp, ENOMEM);
555 return;
556 }
557 bp2->bio_done = g_disk_done;
558 bp2->bio_caller1 = sc;
559 bp2->bio_disk = dp;
560 mtx_lock(&sc->start_mtx);
561 devstat_start_transaction_bio(dp->d_devstat, bp2);
562 mtx_unlock(&sc->start_mtx);
563 dp->d_strategy(bp2);
564 break;
565 default:
566 error = EOPNOTSUPP;
567 break;
568 }
569 if (error != EJUSTRETURN)
570 g_io_deliver(bp, error);
571 return;
572 }
573
574 static void
575 g_disk_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp)
576 {
577 struct bio *bp;
578 struct disk *dp;
579 struct g_disk_softc *sc;
580 char *buf;
581 int res = 0;
582
583 sc = gp->softc;
584 if (sc == NULL || (dp = sc->dp) == NULL)
585 return;
586 if (indent == NULL) {
587 sbuf_printf(sb, " hd %u", dp->d_fwheads);
588 sbuf_printf(sb, " sc %u", dp->d_fwsectors);
589 return;
590 }
591 if (pp != NULL) {
592 sbuf_printf(sb, "%s<fwheads>%u</fwheads>\n",
593 indent, dp->d_fwheads);
594 sbuf_printf(sb, "%s<fwsectors>%u</fwsectors>\n",
595 indent, dp->d_fwsectors);
596
597 /*
598 * "rotationrate" is a little complicated, because the value
599 * returned by the drive might not be the RPM; 0 and 1 are
600 * special cases, and there's also a valid range.
601 */
602 sbuf_printf(sb, "%s<rotationrate>", indent);
603 if (dp->d_rotation_rate == DISK_RR_UNKNOWN) /* Old drives */
604 sbuf_cat(sb, "unknown"); /* don't report RPM. */
605 else if (dp->d_rotation_rate == DISK_RR_NON_ROTATING)
606 sbuf_cat(sb, "");
607 else if ((dp->d_rotation_rate >= DISK_RR_MIN) &&
608 (dp->d_rotation_rate <= DISK_RR_MAX))
609 sbuf_printf(sb, "%u", dp->d_rotation_rate);
610 else
611 sbuf_cat(sb, "invalid");
612 sbuf_cat(sb, "</rotationrate>\n");
613 if (dp->d_getattr != NULL) {
614 buf = g_malloc(DISK_IDENT_SIZE, M_WAITOK);
615 bp = g_alloc_bio();
616 bp->bio_disk = dp;
617 bp->bio_attribute = "GEOM::ident";
618 bp->bio_length = DISK_IDENT_SIZE;
619 bp->bio_data = buf;
620 res = dp->d_getattr(bp);
621 sbuf_printf(sb, "%s<ident>", indent);
622 g_conf_cat_escaped(sb, res == 0 ? buf : dp->d_ident);
623 sbuf_cat(sb, "</ident>\n");
624 bp->bio_attribute = "GEOM::lunid";
625 bp->bio_length = DISK_IDENT_SIZE;
626 bp->bio_data = buf;
627 if (dp->d_getattr(bp) == 0) {
628 sbuf_printf(sb, "%s<lunid>", indent);
629 g_conf_cat_escaped(sb, buf);
630 sbuf_cat(sb, "</lunid>\n");
631 }
632 bp->bio_attribute = "GEOM::lunname";
633 bp->bio_length = DISK_IDENT_SIZE;
634 bp->bio_data = buf;
635 if (dp->d_getattr(bp) == 0) {
636 sbuf_printf(sb, "%s<lunname>", indent);
637 g_conf_cat_escaped(sb, buf);
638 sbuf_cat(sb, "</lunname>\n");
639 }
640 g_destroy_bio(bp);
641 g_free(buf);
642 } else {
643 sbuf_printf(sb, "%s<ident>", indent);
644 g_conf_cat_escaped(sb, dp->d_ident);
645 sbuf_cat(sb, "</ident>\n");
646 }
647 sbuf_printf(sb, "%s<descr>", indent);
648 g_conf_cat_escaped(sb, dp->d_descr);
649 sbuf_cat(sb, "</descr>\n");
650 }
651 }
652
653 static void
654 g_disk_resize(void *ptr, int flag)
655 {
656 struct disk *dp;
657 struct g_geom *gp;
658 struct g_provider *pp;
659
660 if (flag == EV_CANCEL)
661 return;
662 g_topology_assert();
663
664 dp = ptr;
665 gp = dp->d_geom;
666
667 if (dp->d_destroyed || gp == NULL)
668 return;
669
670 LIST_FOREACH(pp, &gp->provider, provider) {
671 if (pp->sectorsize != 0 &&
672 pp->sectorsize != dp->d_sectorsize)
673 g_wither_provider(pp, ENXIO);
674 else
675 g_resize_provider(pp, dp->d_mediasize);
676 }
677 }
678
679 static void
680 g_disk_create(void *arg, int flag)
681 {
682 struct g_geom *gp;
683 struct g_provider *pp;
684 struct disk *dp;
685 struct g_disk_softc *sc;
686 struct disk_alias *dap;
687 char tmpstr[80];
688
689 if (flag == EV_CANCEL)
690 return;
691 g_topology_assert();
692 dp = arg;
693
694 mtx_pool_lock(mtxpool_sleep, dp);
695 dp->d_init_level = DISK_INIT_START;
696
697 /*
698 * If the disk has already gone away, we can just stop here and
699 * call the user's callback to tell him we've cleaned things up.
700 */
701 if (dp->d_goneflag != 0) {
702 mtx_pool_unlock(mtxpool_sleep, dp);
703 if (dp->d_gone != NULL)
704 dp->d_gone(dp);
705 return;
706 }
707 mtx_pool_unlock(mtxpool_sleep, dp);
708
709 sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO);
710 mtx_init(&sc->start_mtx, "g_disk_start", NULL, MTX_DEF);
711 mtx_init(&sc->done_mtx, "g_disk_done", NULL, MTX_DEF);
712 sc->dp = dp;
713 sc->d_devstat = dp->d_devstat;
714 gp = g_new_geomf(&g_disk_class, "%s%d", dp->d_name, dp->d_unit);
715 gp->softc = sc;
716 LIST_FOREACH(dap, &dp->d_aliases, da_next) {
717 snprintf(tmpstr, sizeof(tmpstr), "%s%d", dap->da_alias, dp->d_unit);
718 g_geom_add_alias(gp, tmpstr);
719 }
720 pp = g_new_providerf(gp, "%s", gp->name);
721 devstat_remove_entry(pp->stat);
722 pp->stat = NULL;
723 dp->d_devstat->id = pp;
724 pp->mediasize = dp->d_mediasize;
725 pp->sectorsize = dp->d_sectorsize;
726 pp->stripeoffset = dp->d_stripeoffset;
727 pp->stripesize = dp->d_stripesize;
728 if ((dp->d_flags & DISKFLAG_UNMAPPED_BIO) != 0)
729 pp->flags |= G_PF_ACCEPT_UNMAPPED;
730 if ((dp->d_flags & DISKFLAG_DIRECT_COMPLETION) != 0)
731 pp->flags |= G_PF_DIRECT_SEND;
732 pp->flags |= G_PF_DIRECT_RECEIVE;
733 if (bootverbose)
734 printf("GEOM: new disk %s\n", gp->name);
735 sysctl_ctx_init(&sc->sysctl_ctx);
736 snprintf(tmpstr, sizeof(tmpstr), "GEOM disk %s", gp->name);
737 sc->sysctl_tree = SYSCTL_ADD_NODE(&sc->sysctl_ctx,
738 SYSCTL_STATIC_CHILDREN(_kern_geom_disk), OID_AUTO, gp->name,
739 CTLFLAG_RD, 0, tmpstr);
740 if (sc->sysctl_tree != NULL) {
741 SYSCTL_ADD_STRING(&sc->sysctl_ctx,
742 SYSCTL_CHILDREN(sc->sysctl_tree), OID_AUTO, "led",
743 CTLFLAG_RWTUN, sc->led, sizeof(sc->led),
744 "LED name");
745 SYSCTL_ADD_PROC(&sc->sysctl_ctx,
746 SYSCTL_CHILDREN(sc->sysctl_tree), OID_AUTO, "flags",
747 CTLTYPE_STRING | CTLFLAG_RD, dp, 0, g_disk_sysctl_flags,
748 "A", "Report disk flags");
749 }
750 pp->private = sc;
751 dp->d_geom = gp;
752 g_error_provider(pp, 0);
753
754 mtx_pool_lock(mtxpool_sleep, dp);
755 dp->d_init_level = DISK_INIT_DONE;
756
757 /*
758 * If the disk has gone away at this stage, start the withering
759 * process for it.
760 */
761 if (dp->d_goneflag != 0) {
762 mtx_pool_unlock(mtxpool_sleep, dp);
763 g_wither_provider(pp, ENXIO);
764 return;
765 }
766 mtx_pool_unlock(mtxpool_sleep, dp);
767
768 }
769
770 /*
771 * We get this callback after all of the consumers have gone away, and just
772 * before the provider is freed. If the disk driver provided a d_gone
773 * callback, let them know that it is okay to free resources -- they won't
774 * be getting any more accesses from GEOM.
775 */
776 static void
777 g_disk_providergone(struct g_provider *pp)
778 {
779 struct disk *dp;
780 struct g_disk_softc *sc;
781
782 sc = (struct g_disk_softc *)pp->private;
783 dp = sc->dp;
784 if (dp != NULL && dp->d_gone != NULL)
785 dp->d_gone(dp);
786 if (sc->sysctl_tree != NULL) {
787 sysctl_ctx_free(&sc->sysctl_ctx);
788 sc->sysctl_tree = NULL;
789 }
790 if (sc->led[0] != 0) {
791 led_set(sc->led, "");
792 sc->led[0] = 0;
793 }
794 pp->private = NULL;
795 pp->geom->softc = NULL;
796 mtx_destroy(&sc->done_mtx);
797 mtx_destroy(&sc->start_mtx);
798 g_free(sc);
799 }
800
801 static void
802 g_disk_destroy(void *ptr, int flag)
803 {
804 struct disk *dp;
805 struct g_geom *gp;
806 struct g_disk_softc *sc;
807 struct disk_alias *dap, *daptmp;
808
809 g_topology_assert();
810 dp = ptr;
811 gp = dp->d_geom;
812 if (gp != NULL) {
813 sc = gp->softc;
814 if (sc != NULL)
815 sc->dp = NULL;
816 dp->d_geom = NULL;
817 g_wither_geom(gp, ENXIO);
818 }
819 LIST_FOREACH_SAFE(dap, &dp->d_aliases, da_next, daptmp)
820 g_free(dap);
821
822 g_free(dp);
823 }
824
825 /*
826 * We only allow printable characters in disk ident,
827 * the rest is converted to 'x<HH>'.
828 */
829 static void
830 g_disk_ident_adjust(char *ident, size_t size)
831 {
832 char *p, tmp[4], newid[DISK_IDENT_SIZE];
833
834 newid[0] = '\0';
835 for (p = ident; *p != '\0'; p++) {
836 if (isprint(*p)) {
837 tmp[0] = *p;
838 tmp[1] = '\0';
839 } else {
840 snprintf(tmp, sizeof(tmp), "x%02hhx",
841 *(unsigned char *)p);
842 }
843 if (strlcat(newid, tmp, sizeof(newid)) >= sizeof(newid))
844 break;
845 }
846 bzero(ident, size);
847 strlcpy(ident, newid, size);
848 }
849
850 struct disk *
851 disk_alloc(void)
852 {
853 struct disk *dp;
854
855 dp = g_malloc(sizeof(struct disk), M_WAITOK | M_ZERO);
856 LIST_INIT(&dp->d_aliases);
857 return (dp);
858 }
859
860 void
861 disk_create(struct disk *dp, int version)
862 {
863
864 if (version != DISK_VERSION) {
865 printf("WARNING: Attempt to add disk %s%d %s",
866 dp->d_name, dp->d_unit,
867 " using incompatible ABI version of disk(9)\n");
868 printf("WARNING: Ignoring disk %s%d\n",
869 dp->d_name, dp->d_unit);
870 return;
871 }
872 if (dp->d_flags & DISKFLAG_RESERVED) {
873 printf("WARNING: Attempt to add non-MPSAFE disk %s%d\n",
874 dp->d_name, dp->d_unit);
875 printf("WARNING: Ignoring disk %s%d\n",
876 dp->d_name, dp->d_unit);
877 return;
878 }
879 KASSERT(dp->d_strategy != NULL, ("disk_create need d_strategy"));
880 KASSERT(dp->d_name != NULL, ("disk_create need d_name"));
881 KASSERT(*dp->d_name != 0, ("disk_create need d_name"));
882 KASSERT(strlen(dp->d_name) < SPECNAMELEN - 4, ("disk name too long"));
883 if (dp->d_devstat == NULL)
884 dp->d_devstat = devstat_new_entry(dp->d_name, dp->d_unit,
885 dp->d_sectorsize, DEVSTAT_ALL_SUPPORTED,
886 DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX);
887 dp->d_geom = NULL;
888
889 dp->d_init_level = DISK_INIT_NONE;
890
891 g_disk_ident_adjust(dp->d_ident, sizeof(dp->d_ident));
892 g_post_event(g_disk_create, dp, M_WAITOK, dp, NULL);
893 }
894
895 void
896 disk_destroy(struct disk *dp)
897 {
898
899 disk_gone(dp);
900 dp->d_destroyed = 1;
901 g_cancel_event(dp);
902 if (dp->d_devstat != NULL)
903 devstat_remove_entry(dp->d_devstat);
904 g_post_event(g_disk_destroy, dp, M_WAITOK, NULL);
905 }
906
907 void
908 disk_add_alias(struct disk *dp, const char *name)
909 {
910 struct disk_alias *dap;
911
912 dap = (struct disk_alias *)g_malloc(
913 sizeof(struct disk_alias) + strlen(name) + 1, M_WAITOK);
914 strcpy((char *)(dap + 1), name);
915 dap->da_alias = (const char *)(dap + 1);
916 LIST_INSERT_HEAD(&dp->d_aliases, dap, da_next);
917 }
918
919 void
920 disk_gone(struct disk *dp)
921 {
922 struct g_geom *gp;
923 struct g_provider *pp;
924
925 mtx_pool_lock(mtxpool_sleep, dp);
926
927 /*
928 * Second wither call makes no sense, plus we can not access the list
929 * of providers without topology lock after calling wither once.
930 */
931 if (dp->d_goneflag != 0) {
932 mtx_pool_unlock(mtxpool_sleep, dp);
933 return;
934 }
935
936 dp->d_goneflag = 1;
937
938 /*
939 * If we're still in the process of creating this disk (the
940 * g_disk_create() function is still queued, or is in
941 * progress), the init level will not yet be DISK_INIT_DONE.
942 *
943 * If that is the case, g_disk_create() will see d_goneflag
944 * and take care of cleaning things up.
945 *
946 * If the disk has already been created, we default to
947 * withering the provider as usual below.
948 *
949 * If the caller has not set a d_gone() callback, he will
950 * not be any worse off by returning here, because the geom
951 * has not been fully setup in any case.
952 */
953 if (dp->d_init_level < DISK_INIT_DONE) {
954 mtx_pool_unlock(mtxpool_sleep, dp);
955 return;
956 }
957 mtx_pool_unlock(mtxpool_sleep, dp);
958
959 gp = dp->d_geom;
960 pp = LIST_FIRST(&gp->provider);
961 if (pp != NULL) {
962 KASSERT(LIST_NEXT(pp, provider) == NULL,
963 ("geom %p has more than one provider", gp));
964 g_wither_provider(pp, ENXIO);
965 }
966 }
967
968 void
969 disk_attr_changed(struct disk *dp, const char *attr, int flag)
970 {
971 struct g_geom *gp;
972 struct g_provider *pp;
973 char devnamebuf[128];
974
975 gp = dp->d_geom;
976 if (gp != NULL)
977 LIST_FOREACH(pp, &gp->provider, provider)
978 (void)g_attr_changed(pp, attr, flag);
979 snprintf(devnamebuf, sizeof(devnamebuf), "devname=%s%d", dp->d_name,
980 dp->d_unit);
981 devctl_notify("GEOM", "disk", attr, devnamebuf);
982 }
983
984 void
985 disk_media_changed(struct disk *dp, int flag)
986 {
987 struct g_geom *gp;
988 struct g_provider *pp;
989
990 gp = dp->d_geom;
991 if (gp != NULL) {
992 pp = LIST_FIRST(&gp->provider);
993 if (pp != NULL) {
994 KASSERT(LIST_NEXT(pp, provider) == NULL,
995 ("geom %p has more than one provider", gp));
996 g_media_changed(pp, flag);
997 }
998 }
999 }
1000
1001 void
1002 disk_media_gone(struct disk *dp, int flag)
1003 {
1004 struct g_geom *gp;
1005 struct g_provider *pp;
1006
1007 gp = dp->d_geom;
1008 if (gp != NULL) {
1009 pp = LIST_FIRST(&gp->provider);
1010 if (pp != NULL) {
1011 KASSERT(LIST_NEXT(pp, provider) == NULL,
1012 ("geom %p has more than one provider", gp));
1013 g_media_gone(pp, flag);
1014 }
1015 }
1016 }
1017
1018 int
1019 disk_resize(struct disk *dp, int flag)
1020 {
1021
1022 if (dp->d_destroyed || dp->d_geom == NULL)
1023 return (0);
1024
1025 return (g_post_event(g_disk_resize, dp, flag, NULL));
1026 }
1027
1028 static void
1029 g_kern_disks(void *p, int flag __unused)
1030 {
1031 struct sbuf *sb;
1032 struct g_geom *gp;
1033 char *sp;
1034
1035 sb = p;
1036 sp = "";
1037 g_topology_assert();
1038 LIST_FOREACH(gp, &g_disk_class.geom, geom) {
1039 sbuf_printf(sb, "%s%s", sp, gp->name);
1040 sp = " ";
1041 }
1042 sbuf_finish(sb);
1043 }
1044
1045 static int
1046 g_disk_sysctl_flags(SYSCTL_HANDLER_ARGS)
1047 {
1048 struct disk *dp;
1049 struct sbuf *sb;
1050 int error;
1051
1052 sb = sbuf_new_auto();
1053 dp = (struct disk *)arg1;
1054 sbuf_printf(sb, "%b", dp->d_flags,
1055 "\2"
1056 "\2OPEN"
1057 "\3CANDELETE"
1058 "\4CANFLUSHCACHE"
1059 "\5UNMAPPEDBIO"
1060 "\6DIRECTCOMPLETION"
1061 "\10CANZONE"
1062 "\11WRITEPROTECT");
1063
1064 sbuf_finish(sb);
1065 error = SYSCTL_OUT(req, sbuf_data(sb), sbuf_len(sb) + 1);
1066 sbuf_delete(sb);
1067 return (error);
1068 }
1069
1070 static int
1071 sysctl_disks(SYSCTL_HANDLER_ARGS)
1072 {
1073 int error;
1074 struct sbuf *sb;
1075
1076 sb = sbuf_new_auto();
1077 g_waitfor_event(g_kern_disks, sb, M_WAITOK, NULL);
1078 error = SYSCTL_OUT(req, sbuf_data(sb), sbuf_len(sb) + 1);
1079 sbuf_delete(sb);
1080 return error;
1081 }
1082
1083 SYSCTL_PROC(_kern, OID_AUTO, disks,
1084 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
1085 sysctl_disks, "A", "names of available disks");
Cache object: d7b16bd028c849b82bceafee2ce42cb9
|