2 * Copyright (c) 2002 Poul-Henning Kamp
3 * Copyright (c) 2002 Networks Associates Technology, Inc.
6 * This software was developed for the FreeBSD Project by Poul-Henning Kamp
7 * and NAI Labs, the Security Research Division of Network Associates, Inc.
8 * under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the
9 * DARPA CHATS research program.
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. The names of the authors may not be used to endorse or promote
20 * products derived from this software without specific prior written
23 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 * This is the method for dealing with BSD disklabels. It has been
38 * extensively (by my standards at least) commented, in the vain hope that
39 * it will serve as the source in future copy&paste operations.
42 #include <sys/cdefs.h>
43 __FBSDID("$FreeBSD$");
45 #include <sys/param.h>
46 #include <sys/endian.h>
47 #include <sys/systm.h>
48 #include <sys/kernel.h>
49 #include <sys/fcntl.h>
52 #include <sys/malloc.h>
54 #include <sys/mutex.h>
56 #include <sys/errno.h>
57 #include <sys/disklabel.h>
60 #include <geom/geom.h>
61 #include <geom/geom_slice.h>
63 #define BSD_CLASS_NAME "BSD"
65 #define ALPHA_LABEL_OFFSET 64
66 #define HISTORIC_LABEL_OFFSET 512
68 #define LABELSIZE (148 + 16 * MAXPARTITIONS)
70 static void g_bsd_hotwrite(void *arg, int flag);
72 * Our private data about one instance. All the rest is handled by the
73 * slice code and stored in its softc, so this is just the stuff
74 * specific to BSD disklabels.
80 struct disklabel ondisk;
81 u_char label[LABELSIZE];
86 * Modify our slicer to match proposed disklabel, if possible.
87 * This is where we make sure we don't do something stupid.
90 g_bsd_modify(struct g_geom *gp, u_char *label)
93 struct partition *ppp;
95 struct g_consumer *cp;
96 struct g_bsd_softc *ms;
106 error = bsd_disklabel_le_dec(label, &dl, MAXPARTITIONS);
111 /* Get dimensions of our device. */
112 cp = LIST_FIRST(&gp->consumer);
113 secsize = cp->provider->sectorsize;
115 /* ... or a smaller sector size. */
116 if (dl.d_secsize < secsize) {
120 /* ... or a non-multiple sector size. */
121 if (dl.d_secsize % secsize != 0) {
125 /* Historical braindamage... */
126 rawoffset = (off_t)dl.d_partitions[RAW_PART].p_offset * dl.d_secsize;
128 for (i = 0; i < dl.d_npartitions; i++) {
129 ppp = &dl.d_partitions[i];
130 if (ppp->p_size == 0)
132 o = (off_t)ppp->p_offset * dl.d_secsize;
138 if (rawoffset != 0 && (off_t)rawoffset != ms->mbroffset)
139 printf("WARNING: %s expected rawoffset %jd, found %jd\n",
141 (intmax_t)ms->mbroffset/dl.d_secsize,
142 (intmax_t)rawoffset/dl.d_secsize);
144 /* Don't munge open partitions. */
145 for (i = 0; i < dl.d_npartitions; i++) {
146 ppp = &dl.d_partitions[i];
148 o = (off_t)ppp->p_offset * dl.d_secsize;
151 error = g_slice_config(gp, i, G_SLICE_CONFIG_CHECK,
153 (off_t)ppp->p_size * dl.d_secsize,
155 "%s%c", gp->name, 'a' + i);
160 /* Look good, go for it... */
161 for (u = 0; u < gsp->nslice; u++) {
162 ppp = &dl.d_partitions[u];
163 o = (off_t)ppp->p_offset * dl.d_secsize;
166 g_slice_config(gp, u, G_SLICE_CONFIG_SET,
168 (off_t)ppp->p_size * dl.d_secsize,
170 "%s%c", gp->name, 'a' + u);
173 /* Update our softc */
175 if (label != ms->label)
176 bcopy(label, ms->label, LABELSIZE);
177 ms->rawoffset = rawoffset;
180 * In order to avoid recursively attaching to the same
181 * on-disk label (it's usually visible through the 'c'
182 * partition) we calculate an MD5 and ask if other BSD's
183 * below us love that label. If they do, we don't.
186 MD5Update(&md5sum, ms->label, sizeof(ms->label));
187 MD5Final(ms->labelsum, &md5sum);
193 * This is an internal helper function, called multiple times from the taste
194 * function to try to locate a disklabel on the disk. More civilized formats
195 * will not need this, as there is only one possible place on disk to look
196 * for the magic spot.
200 g_bsd_try(struct g_geom *gp, struct g_slicer *gsp, struct g_consumer *cp, int secsize, struct g_bsd_softc *ms, off_t offset)
204 struct disklabel *dl;
208 * We need to read entire aligned sectors, and we assume that the
209 * disklabel does not span sectors, so one sector is enough.
211 secoff = offset % secsize;
212 buf = g_read_data(cp, offset - secoff, secsize, NULL);
216 /* Decode into our native format. */
218 error = bsd_disklabel_le_dec(buf + secoff, dl, MAXPARTITIONS);
220 bcopy(buf + secoff, ms->label, LABELSIZE);
222 /* Remember to free the buffer g_read_data() gave us. */
225 ms->labeloffset = offset;
230 * This function writes the current label to disk, possibly updating
231 * the alpha SRM checksum.
235 g_bsd_writelabel(struct g_geom *gp, u_char *bootcode)
239 struct g_consumer *cp;
240 struct g_slicer *gsp;
241 struct g_bsd_softc *ms;
248 cp = LIST_FIRST(&gp->consumer);
249 /* Get sector size, we need it to read data. */
250 secsize = cp->provider->sectorsize;
251 secoff = ms->labeloffset % secsize;
252 if (bootcode == NULL) {
253 buf = g_read_data(cp, ms->labeloffset - secoff, secsize, &error);
256 bcopy(ms->label, buf + secoff, sizeof(ms->label));
259 bcopy(ms->label, buf + ms->labeloffset, sizeof(ms->label));
261 if (ms->labeloffset == ALPHA_LABEL_OFFSET) {
263 for (i = 0; i < 63; i++)
264 sum += le64dec(buf + i * 8);
265 le64enc(buf + 504, sum);
267 if (bootcode == NULL) {
268 error = g_write_data(cp, ms->labeloffset - secoff, buf, secsize);
271 error = g_write_data(cp, 0, bootcode, BBSIZE);
277 * If the user tries to overwrite our disklabel through an open partition
278 * or via a magicwrite config call, we end up here and try to prevent
279 * footshooting as best we can.
282 g_bsd_hotwrite(void *arg, int flag)
286 struct g_slicer *gsp;
288 struct g_bsd_softc *ms;
294 * We should never get canceled, because that would amount to a removal
295 * of the geom while there was outstanding I/O requests.
297 KASSERT(flag != EV_CANCEL, ("g_bsd_hotwrite cancelled"));
299 gp = bp->bio_to->geom;
302 gsl = &gsp->slices[bp->bio_to->index];
303 p = (u_char*)bp->bio_data + ms->labeloffset
304 - (bp->bio_offset + gsl->offset);
305 error = g_bsd_modify(gp, p);
307 g_io_deliver(bp, EPERM);
310 g_slice_finish_hot(bp);
314 * This start routine is only called for non-trivial requests, all the
315 * trivial ones are handled autonomously by the slice code.
316 * For requests we handle here, we must call the g_io_deliver() on the
317 * bio, and return non-zero to indicate to the slice code that we did so.
318 * This code executes in the "DOWN" I/O path, this means:
320 * * Don't grab the topology lock.
321 * * Don't call biowait, g_getattr(), g_setattr() or g_read_data()
324 g_bsd_ioctl(struct g_provider *pp, u_long cmd, void *data, int fflag, struct thread *td)
327 struct g_bsd_softc *ms;
328 struct g_slicer *gsp;
338 /* Return a copy of the disklabel to userland. */
339 bsd_disklabel_le_dec(ms->label, data, MAXPARTITIONS);
342 struct g_consumer *cp;
348 if (!(fflag & FWRITE))
350 /* The disklabel to set is the ioctl argument. */
351 buf = g_malloc(BBSIZE, M_WAITOK);
353 error = copyin(p, buf, BBSIZE);
355 /* XXX: Rude, but supposedly safe */
358 /* Validate and modify our slice instance to match. */
359 error = g_bsd_modify(gp, buf + ms->labeloffset);
361 cp = LIST_FIRST(&gp->consumer);
362 if (ms->labeloffset == ALPHA_LABEL_OFFSET) {
364 for (i = 0; i < 63; i++)
365 sum += le64dec(buf + i * 8);
366 le64enc(buf + 504, sum);
368 error = g_write_data(cp, 0, buf, BBSIZE);
378 if (!(fflag & FWRITE))
380 label = g_malloc(LABELSIZE, M_WAITOK);
381 /* The disklabel to set is the ioctl argument. */
382 bsd_disklabel_le_enc(label, data);
386 /* Validate and modify our slice instance to match. */
387 error = g_bsd_modify(gp, label);
388 if (error == 0 && cmd == DIOCWDINFO)
389 error = g_bsd_writelabel(gp, NULL);
401 g_bsd_start(struct bio *bp)
404 struct g_bsd_softc *ms;
405 struct g_slicer *gsp;
407 gp = bp->bio_to->geom;
410 if (bp->bio_cmd == BIO_GETATTR) {
411 if (g_handleattr(bp, "BSD::labelsum", ms->labelsum,
412 sizeof(ms->labelsum)))
419 * Dump configuration information in XML format.
420 * Notice that the function is called once for the geom and once for each
421 * consumer and provider. We let g_slice_dumpconf() do most of the work.
424 g_bsd_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp)
426 struct g_bsd_softc *ms;
427 struct g_slicer *gsp;
431 g_slice_dumpconf(sb, indent, gp, cp, pp);
432 if (indent != NULL && pp == NULL && cp == NULL) {
433 sbuf_printf(sb, "%s<labeloffset>%jd</labeloffset>\n",
434 indent, (intmax_t)ms->labeloffset);
435 sbuf_printf(sb, "%s<rawoffset>%jd</rawoffset>\n",
436 indent, (intmax_t)ms->rawoffset);
437 sbuf_printf(sb, "%s<mbroffset>%jd</mbroffset>\n",
438 indent, (intmax_t)ms->mbroffset);
439 } else if (pp != NULL) {
441 sbuf_printf(sb, " ty %d",
442 ms->ondisk.d_partitions[pp->index].p_fstype);
444 sbuf_printf(sb, "%s<type>%d</type>\n", indent,
445 ms->ondisk.d_partitions[pp->index].p_fstype);
450 * The taste function is called from the event-handler, with the topology
451 * lock already held and a provider to examine. The flags are unused.
453 * If flags == G_TF_NORMAL, the idea is to take a bite of the provider and
454 * if we find valid, consistent magic on it, build a geom on it.
456 * There may be cases where the operator would like to put a BSD-geom on
457 * providers which do not meet all of the requirements. This can be done
458 * by instead passing the G_TF_INSIST flag, which will override these
461 * The final flags value is G_TF_TRANSPARENT, which instructs the method
462 * to put a geom on top of the provider and configure it to be as transparent
463 * as possible. This is not really relevant to the BSD method and therefore
464 * not implemented here.
467 static struct uuid freebsd_slice = GPT_ENT_TYPE_FREEBSD;
469 static struct g_geom *
470 g_bsd_taste(struct g_class *mp, struct g_provider *pp, int flags)
473 struct g_consumer *cp;
475 struct g_bsd_softc *ms;
477 struct g_slicer *gsp;
482 g_trace(G_T_TOPOLOGY, "bsd_taste(%s,%s)", mp->name, pp->name);
485 /* We don't implement transparent inserts. */
486 if (flags == G_TF_TRANSPARENT)
490 * BSD labels are a subclass of the general "slicing" topology so
491 * a lot of the work can be done by the common "slice" code.
492 * Create a geom with space for MAXPARTITIONS providers, one consumer
493 * and a softc structure for us. Specify the provider to attach
494 * the consumer to and our "start" routine for special requests.
495 * The provider is opened with mode (1,0,0) so we can do reads
498 gp = g_slice_new(mp, MAXPARTITIONS, pp, &cp, &ms,
499 sizeof(*ms), g_bsd_start);
503 /* Get the geom_slicer softc from the geom. */
507 * The do...while loop here allows us to have multiple escapes
508 * using a simple "break". This improves code clarity without
509 * ending up in deep nesting and without using goto or come from.
513 * If the provider is an MBR we will only auto attach
514 * to type 165 slices in the G_TF_NORMAL case. We will
515 * attach to any other type.
517 error = g_getattr("MBR::type", cp, &i);
519 if (i != 165 && flags == G_TF_NORMAL)
521 error = g_getattr("MBR::offset", cp, &ms->mbroffset);
526 /* Same thing if we are inside a PC98 */
527 error = g_getattr("PC98::type", cp, &i);
529 if (i != 0xc494 && flags == G_TF_NORMAL)
531 error = g_getattr("PC98::offset", cp, &ms->mbroffset);
536 /* Same thing if we are inside a GPT */
537 error = g_getattr("GPT::type", cp, &uuid);
539 if (memcmp(&uuid, &freebsd_slice, sizeof(uuid)) != 0 &&
540 flags == G_TF_NORMAL)
544 /* Get sector size, we need it to read data. */
545 secsize = cp->provider->sectorsize;
549 /* First look for a label at the start of the second sector. */
550 error = g_bsd_try(gp, gsp, cp, secsize, ms, secsize);
553 * If sector size is not 512 the label still can be at
554 * offset 512, not at the start of the second sector. At least
555 * it's true for labels created by the FreeBSD's bsdlabel(8).
557 if (error && secsize != HISTORIC_LABEL_OFFSET)
558 error = g_bsd_try(gp, gsp, cp, secsize, ms,
559 HISTORIC_LABEL_OFFSET);
561 /* Next, look for alpha labels */
563 error = g_bsd_try(gp, gsp, cp, secsize, ms,
566 /* If we didn't find a label, punt. */
571 * In order to avoid recursively attaching to the same
572 * on-disk label (it's usually visible through the 'c'
573 * partition) we calculate an MD5 and ask if other BSD's
574 * below us love that label. If they do, we don't.
577 MD5Update(&md5sum, ms->label, sizeof(ms->label));
578 MD5Final(ms->labelsum, &md5sum);
580 error = g_getattr("BSD::labelsum", cp, &hash);
581 if (!error && !bcmp(ms->labelsum, hash, sizeof(hash)))
585 * Process the found disklabel, and modify our "slice"
586 * instance to match it, if possible.
588 error = g_bsd_modify(gp, ms->label);
591 /* Success or failure, we can close our provider now. */
592 g_access(cp, -1, 0, 0);
594 /* If we have configured any providers, return the new geom. */
595 if (gsp->nprovider > 0) {
596 g_slice_conf_hot(gp, 0, ms->labeloffset, LABELSIZE,
597 G_SLICE_HOT_ALLOW, G_SLICE_HOT_DENY, G_SLICE_HOT_CALL);
598 gsp->hot = g_bsd_hotwrite;
602 * ...else push the "self-destruct" button, by spoiling our own
603 * consumer. This triggers a call to g_slice_spoiled which will
604 * dismantle what was setup.
612 struct g_bsd_softc *ms;
618 g_bsd_callconfig(void *arg, int flag)
623 hp->error = g_bsd_modify(hp->gp, hp->label);
625 hp->error = g_bsd_writelabel(hp->gp, NULL);
629 * NB! curthread is user process which GCTL'ed.
632 g_bsd_config(struct gctl_req *req, struct g_class *mp, char const *verb)
638 struct g_slicer *gsp;
639 struct g_consumer *cp;
640 struct g_bsd_softc *ms;
643 gp = gctl_get_geom(req, mp, "geom");
646 cp = LIST_FIRST(&gp->consumer);
649 if (!strcmp(verb, "read mbroffset")) {
650 gctl_set_param_err(req, "mbroffset", &ms->mbroffset,
651 sizeof(ms->mbroffset));
653 } else if (!strcmp(verb, "write label")) {
654 label = gctl_get_paraml(req, "label", LABELSIZE);
658 h0h0.ms = gsp->softc;
661 /* XXX: Does this reference register with our selfdestruct code ? */
662 error = g_access(cp, 1, 1, 1);
664 gctl_error(req, "could not access consumer");
667 g_bsd_callconfig(&h0h0, 0);
669 g_access(cp, -1, -1, -1);
670 } else if (!strcmp(verb, "write bootcode")) {
671 label = gctl_get_paraml(req, "bootcode", BBSIZE);
674 /* XXX: Does this reference register with our selfdestruct code ? */
675 error = g_access(cp, 1, 1, 1);
677 gctl_error(req, "could not access consumer");
680 error = g_bsd_writelabel(gp, label);
681 g_access(cp, -1, -1, -1);
683 gctl_error(req, "Unknown verb parameter");
689 /* Finally, register with GEOM infrastructure. */
690 static struct g_class g_bsd_class = {
691 .name = BSD_CLASS_NAME,
692 .version = G_VERSION,
693 .taste = g_bsd_taste,
694 .ctlreq = g_bsd_config,
695 .dumpconf = g_bsd_dumpconf,
696 .ioctl = g_bsd_ioctl,
699 DECLARE_GEOM_CLASS(g_bsd_class, g_bsd);